1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * Common eBPF ELF object loading operations.
5 *
6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
8 * Copyright (C) 2015 Huawei Inc.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
11 */
12
13 #ifndef _GNU_SOURCE
14 #define _GNU_SOURCE
15 #endif
16 #include <stdlib.h>
17 #include <stdio.h>
18 #include <stdarg.h>
19 #include <libgen.h>
20 #include <inttypes.h>
21 #include <limits.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <endian.h>
25 #include <fcntl.h>
26 #include <errno.h>
27 #include <ctype.h>
28 #include <asm/unistd.h>
29 #include <linux/err.h>
30 #include <linux/kernel.h>
31 #include <linux/bpf.h>
32 #include <linux/btf.h>
33 #include <linux/filter.h>
34 #include <linux/limits.h>
35 #include <linux/perf_event.h>
36 #include <linux/bpf_perf_event.h>
37 #include <linux/ring_buffer.h>
38 #include <sys/epoll.h>
39 #include <sys/ioctl.h>
40 #include <sys/mman.h>
41 #include <sys/stat.h>
42 #include <sys/types.h>
43 #include <sys/vfs.h>
44 #include <sys/utsname.h>
45 #include <sys/resource.h>
46 #include <libelf.h>
47 #include <gelf.h>
48 #include <zlib.h>
49
50 #include "libbpf.h"
51 #include "bpf.h"
52 #include "btf.h"
53 #include "str_error.h"
54 #include "libbpf_internal.h"
55 #include "hashmap.h"
56 #include "bpf_gen_internal.h"
57 #include "zip.h"
58
59 #ifndef BPF_FS_MAGIC
60 #define BPF_FS_MAGIC 0xcafe4a11
61 #endif
62
63 #define BPF_FS_DEFAULT_PATH "/sys/fs/bpf"
64
65 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
66
67 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
68 * compilation if user enables corresponding warning. Disable it explicitly.
69 */
70 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
71
72 #define __printf(a, b) __attribute__((format(printf, a, b)))
73
74 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
75 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
76 static int map_set_def_max_entries(struct bpf_map *map);
77
78 static const char * const attach_type_name[] = {
79 [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress",
80 [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress",
81 [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create",
82 [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release",
83 [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops",
84 [BPF_CGROUP_DEVICE] = "cgroup_device",
85 [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind",
86 [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind",
87 [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect",
88 [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect",
89 [BPF_CGROUP_UNIX_CONNECT] = "cgroup_unix_connect",
90 [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind",
91 [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind",
92 [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername",
93 [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername",
94 [BPF_CGROUP_UNIX_GETPEERNAME] = "cgroup_unix_getpeername",
95 [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname",
96 [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname",
97 [BPF_CGROUP_UNIX_GETSOCKNAME] = "cgroup_unix_getsockname",
98 [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg",
99 [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg",
100 [BPF_CGROUP_UNIX_SENDMSG] = "cgroup_unix_sendmsg",
101 [BPF_CGROUP_SYSCTL] = "cgroup_sysctl",
102 [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg",
103 [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg",
104 [BPF_CGROUP_UNIX_RECVMSG] = "cgroup_unix_recvmsg",
105 [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt",
106 [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt",
107 [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser",
108 [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict",
109 [BPF_SK_SKB_VERDICT] = "sk_skb_verdict",
110 [BPF_SK_MSG_VERDICT] = "sk_msg_verdict",
111 [BPF_LIRC_MODE2] = "lirc_mode2",
112 [BPF_FLOW_DISSECTOR] = "flow_dissector",
113 [BPF_TRACE_RAW_TP] = "trace_raw_tp",
114 [BPF_TRACE_FENTRY] = "trace_fentry",
115 [BPF_TRACE_FEXIT] = "trace_fexit",
116 [BPF_MODIFY_RETURN] = "modify_return",
117 [BPF_LSM_MAC] = "lsm_mac",
118 [BPF_LSM_CGROUP] = "lsm_cgroup",
119 [BPF_SK_LOOKUP] = "sk_lookup",
120 [BPF_TRACE_ITER] = "trace_iter",
121 [BPF_XDP_DEVMAP] = "xdp_devmap",
122 [BPF_XDP_CPUMAP] = "xdp_cpumap",
123 [BPF_XDP] = "xdp",
124 [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select",
125 [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate",
126 [BPF_PERF_EVENT] = "perf_event",
127 [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi",
128 [BPF_STRUCT_OPS] = "struct_ops",
129 [BPF_NETFILTER] = "netfilter",
130 [BPF_TCX_INGRESS] = "tcx_ingress",
131 [BPF_TCX_EGRESS] = "tcx_egress",
132 [BPF_TRACE_UPROBE_MULTI] = "trace_uprobe_multi",
133 [BPF_NETKIT_PRIMARY] = "netkit_primary",
134 [BPF_NETKIT_PEER] = "netkit_peer",
135 [BPF_TRACE_KPROBE_SESSION] = "trace_kprobe_session",
136 [BPF_TRACE_UPROBE_SESSION] = "trace_uprobe_session",
137 };
138
139 static const char * const link_type_name[] = {
140 [BPF_LINK_TYPE_UNSPEC] = "unspec",
141 [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
142 [BPF_LINK_TYPE_TRACING] = "tracing",
143 [BPF_LINK_TYPE_CGROUP] = "cgroup",
144 [BPF_LINK_TYPE_ITER] = "iter",
145 [BPF_LINK_TYPE_NETNS] = "netns",
146 [BPF_LINK_TYPE_XDP] = "xdp",
147 [BPF_LINK_TYPE_PERF_EVENT] = "perf_event",
148 [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi",
149 [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops",
150 [BPF_LINK_TYPE_NETFILTER] = "netfilter",
151 [BPF_LINK_TYPE_TCX] = "tcx",
152 [BPF_LINK_TYPE_UPROBE_MULTI] = "uprobe_multi",
153 [BPF_LINK_TYPE_NETKIT] = "netkit",
154 [BPF_LINK_TYPE_SOCKMAP] = "sockmap",
155 };
156
157 static const char * const map_type_name[] = {
158 [BPF_MAP_TYPE_UNSPEC] = "unspec",
159 [BPF_MAP_TYPE_HASH] = "hash",
160 [BPF_MAP_TYPE_ARRAY] = "array",
161 [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array",
162 [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array",
163 [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash",
164 [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array",
165 [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace",
166 [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array",
167 [BPF_MAP_TYPE_LRU_HASH] = "lru_hash",
168 [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash",
169 [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie",
170 [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps",
171 [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps",
172 [BPF_MAP_TYPE_DEVMAP] = "devmap",
173 [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash",
174 [BPF_MAP_TYPE_SOCKMAP] = "sockmap",
175 [BPF_MAP_TYPE_CPUMAP] = "cpumap",
176 [BPF_MAP_TYPE_XSKMAP] = "xskmap",
177 [BPF_MAP_TYPE_SOCKHASH] = "sockhash",
178 [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage",
179 [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray",
180 [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage",
181 [BPF_MAP_TYPE_QUEUE] = "queue",
182 [BPF_MAP_TYPE_STACK] = "stack",
183 [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage",
184 [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops",
185 [BPF_MAP_TYPE_RINGBUF] = "ringbuf",
186 [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage",
187 [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage",
188 [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter",
189 [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf",
190 [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage",
191 [BPF_MAP_TYPE_ARENA] = "arena",
192 };
193
194 static const char * const prog_type_name[] = {
195 [BPF_PROG_TYPE_UNSPEC] = "unspec",
196 [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter",
197 [BPF_PROG_TYPE_KPROBE] = "kprobe",
198 [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls",
199 [BPF_PROG_TYPE_SCHED_ACT] = "sched_act",
200 [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint",
201 [BPF_PROG_TYPE_XDP] = "xdp",
202 [BPF_PROG_TYPE_PERF_EVENT] = "perf_event",
203 [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb",
204 [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock",
205 [BPF_PROG_TYPE_LWT_IN] = "lwt_in",
206 [BPF_PROG_TYPE_LWT_OUT] = "lwt_out",
207 [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit",
208 [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops",
209 [BPF_PROG_TYPE_SK_SKB] = "sk_skb",
210 [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device",
211 [BPF_PROG_TYPE_SK_MSG] = "sk_msg",
212 [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
213 [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr",
214 [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local",
215 [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2",
216 [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport",
217 [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector",
218 [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl",
219 [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable",
220 [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt",
221 [BPF_PROG_TYPE_TRACING] = "tracing",
222 [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops",
223 [BPF_PROG_TYPE_EXT] = "ext",
224 [BPF_PROG_TYPE_LSM] = "lsm",
225 [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup",
226 [BPF_PROG_TYPE_SYSCALL] = "syscall",
227 [BPF_PROG_TYPE_NETFILTER] = "netfilter",
228 };
229
__base_pr(enum libbpf_print_level level,const char * format,va_list args)230 static int __base_pr(enum libbpf_print_level level, const char *format,
231 va_list args)
232 {
233 const char *env_var = "LIBBPF_LOG_LEVEL";
234 static enum libbpf_print_level min_level = LIBBPF_INFO;
235 static bool initialized;
236
237 if (!initialized) {
238 char *verbosity;
239
240 initialized = true;
241 verbosity = getenv(env_var);
242 if (verbosity) {
243 if (strcasecmp(verbosity, "warn") == 0)
244 min_level = LIBBPF_WARN;
245 else if (strcasecmp(verbosity, "debug") == 0)
246 min_level = LIBBPF_DEBUG;
247 else if (strcasecmp(verbosity, "info") == 0)
248 min_level = LIBBPF_INFO;
249 else
250 fprintf(stderr, "libbpf: unrecognized '%s' envvar value: '%s', should be one of 'warn', 'debug', or 'info'.\n",
251 env_var, verbosity);
252 }
253 }
254
255 /* if too verbose, skip logging */
256 if (level > min_level)
257 return 0;
258
259 return vfprintf(stderr, format, args);
260 }
261
262 static libbpf_print_fn_t __libbpf_pr = __base_pr;
263
libbpf_set_print(libbpf_print_fn_t fn)264 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
265 {
266 libbpf_print_fn_t old_print_fn;
267
268 old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
269
270 return old_print_fn;
271 }
272
273 __printf(2, 3)
libbpf_print(enum libbpf_print_level level,const char * format,...)274 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
275 {
276 va_list args;
277 int old_errno;
278 libbpf_print_fn_t print_fn;
279
280 print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
281 if (!print_fn)
282 return;
283
284 old_errno = errno;
285
286 va_start(args, format);
287 __libbpf_pr(level, format, args);
288 va_end(args);
289
290 errno = old_errno;
291 }
292
pr_perm_msg(int err)293 static void pr_perm_msg(int err)
294 {
295 struct rlimit limit;
296 char buf[100];
297
298 if (err != -EPERM || geteuid() != 0)
299 return;
300
301 err = getrlimit(RLIMIT_MEMLOCK, &limit);
302 if (err)
303 return;
304
305 if (limit.rlim_cur == RLIM_INFINITY)
306 return;
307
308 if (limit.rlim_cur < 1024)
309 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
310 else if (limit.rlim_cur < 1024*1024)
311 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
312 else
313 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
314
315 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
316 buf);
317 }
318
319 #define STRERR_BUFSIZE 128
320
321 /* Copied from tools/perf/util/util.h */
322 #ifndef zfree
323 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
324 #endif
325
326 #ifndef zclose
327 # define zclose(fd) ({ \
328 int ___err = 0; \
329 if ((fd) >= 0) \
330 ___err = close((fd)); \
331 fd = -1; \
332 ___err; })
333 #endif
334
ptr_to_u64(const void * ptr)335 static inline __u64 ptr_to_u64(const void *ptr)
336 {
337 return (__u64) (unsigned long) ptr;
338 }
339
libbpf_set_strict_mode(enum libbpf_strict_mode mode)340 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
341 {
342 /* as of v1.0 libbpf_set_strict_mode() is a no-op */
343 return 0;
344 }
345
libbpf_major_version(void)346 __u32 libbpf_major_version(void)
347 {
348 return LIBBPF_MAJOR_VERSION;
349 }
350
libbpf_minor_version(void)351 __u32 libbpf_minor_version(void)
352 {
353 return LIBBPF_MINOR_VERSION;
354 }
355
libbpf_version_string(void)356 const char *libbpf_version_string(void)
357 {
358 #define __S(X) #X
359 #define _S(X) __S(X)
360 return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
361 #undef _S
362 #undef __S
363 }
364
365 enum reloc_type {
366 RELO_LD64,
367 RELO_CALL,
368 RELO_DATA,
369 RELO_EXTERN_LD64,
370 RELO_EXTERN_CALL,
371 RELO_SUBPROG_ADDR,
372 RELO_CORE,
373 };
374
375 struct reloc_desc {
376 enum reloc_type type;
377 int insn_idx;
378 union {
379 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
380 struct {
381 int map_idx;
382 int sym_off;
383 int ext_idx;
384 };
385 };
386 };
387
388 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
389 enum sec_def_flags {
390 SEC_NONE = 0,
391 /* expected_attach_type is optional, if kernel doesn't support that */
392 SEC_EXP_ATTACH_OPT = 1,
393 /* legacy, only used by libbpf_get_type_names() and
394 * libbpf_attach_type_by_name(), not used by libbpf itself at all.
395 * This used to be associated with cgroup (and few other) BPF programs
396 * that were attachable through BPF_PROG_ATTACH command. Pretty
397 * meaningless nowadays, though.
398 */
399 SEC_ATTACHABLE = 2,
400 SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
401 /* attachment target is specified through BTF ID in either kernel or
402 * other BPF program's BTF object
403 */
404 SEC_ATTACH_BTF = 4,
405 /* BPF program type allows sleeping/blocking in kernel */
406 SEC_SLEEPABLE = 8,
407 /* BPF program support non-linear XDP buffer */
408 SEC_XDP_FRAGS = 16,
409 /* Setup proper attach type for usdt probes. */
410 SEC_USDT = 32,
411 };
412
413 struct bpf_sec_def {
414 char *sec;
415 enum bpf_prog_type prog_type;
416 enum bpf_attach_type expected_attach_type;
417 long cookie;
418 int handler_id;
419
420 libbpf_prog_setup_fn_t prog_setup_fn;
421 libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
422 libbpf_prog_attach_fn_t prog_attach_fn;
423 };
424
425 /*
426 * bpf_prog should be a better name but it has been used in
427 * linux/filter.h.
428 */
429 struct bpf_program {
430 char *name;
431 char *sec_name;
432 size_t sec_idx;
433 const struct bpf_sec_def *sec_def;
434 /* this program's instruction offset (in number of instructions)
435 * within its containing ELF section
436 */
437 size_t sec_insn_off;
438 /* number of original instructions in ELF section belonging to this
439 * program, not taking into account subprogram instructions possible
440 * appended later during relocation
441 */
442 size_t sec_insn_cnt;
443 /* Offset (in number of instructions) of the start of instruction
444 * belonging to this BPF program within its containing main BPF
445 * program. For the entry-point (main) BPF program, this is always
446 * zero. For a sub-program, this gets reset before each of main BPF
447 * programs are processed and relocated and is used to determined
448 * whether sub-program was already appended to the main program, and
449 * if yes, at which instruction offset.
450 */
451 size_t sub_insn_off;
452
453 /* instructions that belong to BPF program; insns[0] is located at
454 * sec_insn_off instruction within its ELF section in ELF file, so
455 * when mapping ELF file instruction index to the local instruction,
456 * one needs to subtract sec_insn_off; and vice versa.
457 */
458 struct bpf_insn *insns;
459 /* actual number of instruction in this BPF program's image; for
460 * entry-point BPF programs this includes the size of main program
461 * itself plus all the used sub-programs, appended at the end
462 */
463 size_t insns_cnt;
464
465 struct reloc_desc *reloc_desc;
466 int nr_reloc;
467
468 /* BPF verifier log settings */
469 char *log_buf;
470 size_t log_size;
471 __u32 log_level;
472
473 struct bpf_object *obj;
474
475 int fd;
476 bool autoload;
477 bool autoattach;
478 bool sym_global;
479 bool mark_btf_static;
480 enum bpf_prog_type type;
481 enum bpf_attach_type expected_attach_type;
482 int exception_cb_idx;
483
484 int prog_ifindex;
485 __u32 attach_btf_obj_fd;
486 __u32 attach_btf_id;
487 __u32 attach_prog_fd;
488
489 void *func_info;
490 __u32 func_info_rec_size;
491 __u32 func_info_cnt;
492
493 void *line_info;
494 __u32 line_info_rec_size;
495 __u32 line_info_cnt;
496 __u32 prog_flags;
497 };
498
499 struct bpf_struct_ops {
500 struct bpf_program **progs;
501 __u32 *kern_func_off;
502 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
503 void *data;
504 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
505 * btf_vmlinux's format.
506 * struct bpf_struct_ops_tcp_congestion_ops {
507 * [... some other kernel fields ...]
508 * struct tcp_congestion_ops data;
509 * }
510 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
511 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
512 * from "data".
513 */
514 void *kern_vdata;
515 __u32 type_id;
516 };
517
518 #define DATA_SEC ".data"
519 #define BSS_SEC ".bss"
520 #define RODATA_SEC ".rodata"
521 #define KCONFIG_SEC ".kconfig"
522 #define KSYMS_SEC ".ksyms"
523 #define STRUCT_OPS_SEC ".struct_ops"
524 #define STRUCT_OPS_LINK_SEC ".struct_ops.link"
525 #define ARENA_SEC ".addr_space.1"
526
527 enum libbpf_map_type {
528 LIBBPF_MAP_UNSPEC,
529 LIBBPF_MAP_DATA,
530 LIBBPF_MAP_BSS,
531 LIBBPF_MAP_RODATA,
532 LIBBPF_MAP_KCONFIG,
533 };
534
535 struct bpf_map_def {
536 unsigned int type;
537 unsigned int key_size;
538 unsigned int value_size;
539 unsigned int max_entries;
540 unsigned int map_flags;
541 };
542
543 struct bpf_map {
544 struct bpf_object *obj;
545 char *name;
546 /* real_name is defined for special internal maps (.rodata*,
547 * .data*, .bss, .kconfig) and preserves their original ELF section
548 * name. This is important to be able to find corresponding BTF
549 * DATASEC information.
550 */
551 char *real_name;
552 int fd;
553 int sec_idx;
554 size_t sec_offset;
555 int map_ifindex;
556 int inner_map_fd;
557 struct bpf_map_def def;
558 __u32 numa_node;
559 __u32 btf_var_idx;
560 int mod_btf_fd;
561 __u32 btf_key_type_id;
562 __u32 btf_value_type_id;
563 __u32 btf_vmlinux_value_type_id;
564 enum libbpf_map_type libbpf_type;
565 void *mmaped;
566 struct bpf_struct_ops *st_ops;
567 struct bpf_map *inner_map;
568 void **init_slots;
569 int init_slots_sz;
570 char *pin_path;
571 bool pinned;
572 bool reused;
573 bool autocreate;
574 bool autoattach;
575 __u64 map_extra;
576 };
577
578 enum extern_type {
579 EXT_UNKNOWN,
580 EXT_KCFG,
581 EXT_KSYM,
582 };
583
584 enum kcfg_type {
585 KCFG_UNKNOWN,
586 KCFG_CHAR,
587 KCFG_BOOL,
588 KCFG_INT,
589 KCFG_TRISTATE,
590 KCFG_CHAR_ARR,
591 };
592
593 struct extern_desc {
594 enum extern_type type;
595 int sym_idx;
596 int btf_id;
597 int sec_btf_id;
598 const char *name;
599 char *essent_name;
600 bool is_set;
601 bool is_weak;
602 union {
603 struct {
604 enum kcfg_type type;
605 int sz;
606 int align;
607 int data_off;
608 bool is_signed;
609 } kcfg;
610 struct {
611 unsigned long long addr;
612
613 /* target btf_id of the corresponding kernel var. */
614 int kernel_btf_obj_fd;
615 int kernel_btf_id;
616
617 /* local btf_id of the ksym extern's type. */
618 __u32 type_id;
619 /* BTF fd index to be patched in for insn->off, this is
620 * 0 for vmlinux BTF, index in obj->fd_array for module
621 * BTF
622 */
623 __s16 btf_fd_idx;
624 } ksym;
625 };
626 };
627
628 struct module_btf {
629 struct btf *btf;
630 char *name;
631 __u32 id;
632 int fd;
633 int fd_array_idx;
634 };
635
636 enum sec_type {
637 SEC_UNUSED = 0,
638 SEC_RELO,
639 SEC_BSS,
640 SEC_DATA,
641 SEC_RODATA,
642 SEC_ST_OPS,
643 };
644
645 struct elf_sec_desc {
646 enum sec_type sec_type;
647 Elf64_Shdr *shdr;
648 Elf_Data *data;
649 };
650
651 struct elf_state {
652 int fd;
653 const void *obj_buf;
654 size_t obj_buf_sz;
655 Elf *elf;
656 Elf64_Ehdr *ehdr;
657 Elf_Data *symbols;
658 Elf_Data *arena_data;
659 size_t shstrndx; /* section index for section name strings */
660 size_t strtabidx;
661 struct elf_sec_desc *secs;
662 size_t sec_cnt;
663 int btf_maps_shndx;
664 __u32 btf_maps_sec_btf_id;
665 int text_shndx;
666 int symbols_shndx;
667 bool has_st_ops;
668 int arena_data_shndx;
669 };
670
671 struct usdt_manager;
672
673 struct bpf_object {
674 char name[BPF_OBJ_NAME_LEN];
675 char license[64];
676 __u32 kern_version;
677
678 struct bpf_program *programs;
679 size_t nr_programs;
680 struct bpf_map *maps;
681 size_t nr_maps;
682 size_t maps_cap;
683
684 char *kconfig;
685 struct extern_desc *externs;
686 int nr_extern;
687 int kconfig_map_idx;
688
689 bool loaded;
690 bool has_subcalls;
691 bool has_rodata;
692
693 struct bpf_gen *gen_loader;
694
695 /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
696 struct elf_state efile;
697
698 unsigned char byteorder;
699
700 struct btf *btf;
701 struct btf_ext *btf_ext;
702
703 /* Parse and load BTF vmlinux if any of the programs in the object need
704 * it at load time.
705 */
706 struct btf *btf_vmlinux;
707 /* Path to the custom BTF to be used for BPF CO-RE relocations as an
708 * override for vmlinux BTF.
709 */
710 char *btf_custom_path;
711 /* vmlinux BTF override for CO-RE relocations */
712 struct btf *btf_vmlinux_override;
713 /* Lazily initialized kernel module BTFs */
714 struct module_btf *btf_modules;
715 bool btf_modules_loaded;
716 size_t btf_module_cnt;
717 size_t btf_module_cap;
718
719 /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
720 char *log_buf;
721 size_t log_size;
722 __u32 log_level;
723
724 int *fd_array;
725 size_t fd_array_cap;
726 size_t fd_array_cnt;
727
728 struct usdt_manager *usdt_man;
729
730 struct bpf_map *arena_map;
731 void *arena_data;
732 size_t arena_data_sz;
733
734 struct kern_feature_cache *feat_cache;
735 char *token_path;
736 int token_fd;
737
738 char path[];
739 };
740
741 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
742 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
743 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
744 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
745 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
746 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
747 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
748 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
749 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
750
bpf_program__unload(struct bpf_program * prog)751 void bpf_program__unload(struct bpf_program *prog)
752 {
753 if (!prog)
754 return;
755
756 zclose(prog->fd);
757
758 zfree(&prog->func_info);
759 zfree(&prog->line_info);
760 }
761
bpf_program__exit(struct bpf_program * prog)762 static void bpf_program__exit(struct bpf_program *prog)
763 {
764 if (!prog)
765 return;
766
767 bpf_program__unload(prog);
768 zfree(&prog->name);
769 zfree(&prog->sec_name);
770 zfree(&prog->insns);
771 zfree(&prog->reloc_desc);
772
773 prog->nr_reloc = 0;
774 prog->insns_cnt = 0;
775 prog->sec_idx = -1;
776 }
777
insn_is_subprog_call(const struct bpf_insn * insn)778 static bool insn_is_subprog_call(const struct bpf_insn *insn)
779 {
780 return BPF_CLASS(insn->code) == BPF_JMP &&
781 BPF_OP(insn->code) == BPF_CALL &&
782 BPF_SRC(insn->code) == BPF_K &&
783 insn->src_reg == BPF_PSEUDO_CALL &&
784 insn->dst_reg == 0 &&
785 insn->off == 0;
786 }
787
is_call_insn(const struct bpf_insn * insn)788 static bool is_call_insn(const struct bpf_insn *insn)
789 {
790 return insn->code == (BPF_JMP | BPF_CALL);
791 }
792
insn_is_pseudo_func(struct bpf_insn * insn)793 static bool insn_is_pseudo_func(struct bpf_insn *insn)
794 {
795 return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
796 }
797
798 static int
bpf_object__init_prog(struct bpf_object * obj,struct bpf_program * prog,const char * name,size_t sec_idx,const char * sec_name,size_t sec_off,void * insn_data,size_t insn_data_sz)799 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
800 const char *name, size_t sec_idx, const char *sec_name,
801 size_t sec_off, void *insn_data, size_t insn_data_sz)
802 {
803 if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
804 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
805 sec_name, name, sec_off, insn_data_sz);
806 return -EINVAL;
807 }
808
809 memset(prog, 0, sizeof(*prog));
810 prog->obj = obj;
811
812 prog->sec_idx = sec_idx;
813 prog->sec_insn_off = sec_off / BPF_INSN_SZ;
814 prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
815 /* insns_cnt can later be increased by appending used subprograms */
816 prog->insns_cnt = prog->sec_insn_cnt;
817
818 prog->type = BPF_PROG_TYPE_UNSPEC;
819 prog->fd = -1;
820 prog->exception_cb_idx = -1;
821
822 /* libbpf's convention for SEC("?abc...") is that it's just like
823 * SEC("abc...") but the corresponding bpf_program starts out with
824 * autoload set to false.
825 */
826 if (sec_name[0] == '?') {
827 prog->autoload = false;
828 /* from now on forget there was ? in section name */
829 sec_name++;
830 } else {
831 prog->autoload = true;
832 }
833
834 prog->autoattach = true;
835
836 /* inherit object's log_level */
837 prog->log_level = obj->log_level;
838
839 prog->sec_name = strdup(sec_name);
840 if (!prog->sec_name)
841 goto errout;
842
843 prog->name = strdup(name);
844 if (!prog->name)
845 goto errout;
846
847 prog->insns = malloc(insn_data_sz);
848 if (!prog->insns)
849 goto errout;
850 memcpy(prog->insns, insn_data, insn_data_sz);
851
852 return 0;
853 errout:
854 pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
855 bpf_program__exit(prog);
856 return -ENOMEM;
857 }
858
859 static int
bpf_object__add_programs(struct bpf_object * obj,Elf_Data * sec_data,const char * sec_name,int sec_idx)860 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
861 const char *sec_name, int sec_idx)
862 {
863 Elf_Data *symbols = obj->efile.symbols;
864 struct bpf_program *prog, *progs;
865 void *data = sec_data->d_buf;
866 size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
867 int nr_progs, err, i;
868 const char *name;
869 Elf64_Sym *sym;
870
871 progs = obj->programs;
872 nr_progs = obj->nr_programs;
873 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
874
875 for (i = 0; i < nr_syms; i++) {
876 sym = elf_sym_by_idx(obj, i);
877
878 if (sym->st_shndx != sec_idx)
879 continue;
880 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
881 continue;
882
883 prog_sz = sym->st_size;
884 sec_off = sym->st_value;
885
886 name = elf_sym_str(obj, sym->st_name);
887 if (!name) {
888 pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
889 sec_name, sec_off);
890 return -LIBBPF_ERRNO__FORMAT;
891 }
892
893 if (sec_off + prog_sz > sec_sz) {
894 pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
895 sec_name, sec_off);
896 return -LIBBPF_ERRNO__FORMAT;
897 }
898
899 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
900 pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
901 return -ENOTSUP;
902 }
903
904 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
905 sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
906
907 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
908 if (!progs) {
909 /*
910 * In this case the original obj->programs
911 * is still valid, so don't need special treat for
912 * bpf_close_object().
913 */
914 pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
915 sec_name, name);
916 return -ENOMEM;
917 }
918 obj->programs = progs;
919
920 prog = &progs[nr_progs];
921
922 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
923 sec_off, data + sec_off, prog_sz);
924 if (err)
925 return err;
926
927 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL)
928 prog->sym_global = true;
929
930 /* if function is a global/weak symbol, but has restricted
931 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
932 * as static to enable more permissive BPF verification mode
933 * with more outside context available to BPF verifier
934 */
935 if (prog->sym_global && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
936 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
937 prog->mark_btf_static = true;
938
939 nr_progs++;
940 obj->nr_programs = nr_progs;
941 }
942
943 return 0;
944 }
945
bpf_object_bswap_progs(struct bpf_object * obj)946 static void bpf_object_bswap_progs(struct bpf_object *obj)
947 {
948 struct bpf_program *prog = obj->programs;
949 struct bpf_insn *insn;
950 int p, i;
951
952 for (p = 0; p < obj->nr_programs; p++, prog++) {
953 insn = prog->insns;
954 for (i = 0; i < prog->insns_cnt; i++, insn++)
955 bpf_insn_bswap(insn);
956 }
957 pr_debug("converted %zu BPF programs to native byte order\n", obj->nr_programs);
958 }
959
960 static const struct btf_member *
find_member_by_offset(const struct btf_type * t,__u32 bit_offset)961 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
962 {
963 struct btf_member *m;
964 int i;
965
966 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
967 if (btf_member_bit_offset(t, i) == bit_offset)
968 return m;
969 }
970
971 return NULL;
972 }
973
974 static const struct btf_member *
find_member_by_name(const struct btf * btf,const struct btf_type * t,const char * name)975 find_member_by_name(const struct btf *btf, const struct btf_type *t,
976 const char *name)
977 {
978 struct btf_member *m;
979 int i;
980
981 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
982 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
983 return m;
984 }
985
986 return NULL;
987 }
988
989 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
990 __u16 kind, struct btf **res_btf,
991 struct module_btf **res_mod_btf);
992
993 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
994 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
995 const char *name, __u32 kind);
996
997 static int
find_struct_ops_kern_types(struct bpf_object * obj,const char * tname_raw,struct module_btf ** mod_btf,const struct btf_type ** type,__u32 * type_id,const struct btf_type ** vtype,__u32 * vtype_id,const struct btf_member ** data_member)998 find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw,
999 struct module_btf **mod_btf,
1000 const struct btf_type **type, __u32 *type_id,
1001 const struct btf_type **vtype, __u32 *vtype_id,
1002 const struct btf_member **data_member)
1003 {
1004 const struct btf_type *kern_type, *kern_vtype;
1005 const struct btf_member *kern_data_member;
1006 struct btf *btf = NULL;
1007 __s32 kern_vtype_id, kern_type_id;
1008 char tname[256];
1009 __u32 i;
1010
1011 snprintf(tname, sizeof(tname), "%.*s",
1012 (int)bpf_core_essential_name_len(tname_raw), tname_raw);
1013
1014 kern_type_id = find_ksym_btf_id(obj, tname, BTF_KIND_STRUCT,
1015 &btf, mod_btf);
1016 if (kern_type_id < 0) {
1017 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
1018 tname);
1019 return kern_type_id;
1020 }
1021 kern_type = btf__type_by_id(btf, kern_type_id);
1022
1023 /* Find the corresponding "map_value" type that will be used
1024 * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example,
1025 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
1026 * btf_vmlinux.
1027 */
1028 kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
1029 tname, BTF_KIND_STRUCT);
1030 if (kern_vtype_id < 0) {
1031 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
1032 STRUCT_OPS_VALUE_PREFIX, tname);
1033 return kern_vtype_id;
1034 }
1035 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
1036
1037 /* Find "struct tcp_congestion_ops" from
1038 * struct bpf_struct_ops_tcp_congestion_ops {
1039 * [ ... ]
1040 * struct tcp_congestion_ops data;
1041 * }
1042 */
1043 kern_data_member = btf_members(kern_vtype);
1044 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
1045 if (kern_data_member->type == kern_type_id)
1046 break;
1047 }
1048 if (i == btf_vlen(kern_vtype)) {
1049 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
1050 tname, STRUCT_OPS_VALUE_PREFIX, tname);
1051 return -EINVAL;
1052 }
1053
1054 *type = kern_type;
1055 *type_id = kern_type_id;
1056 *vtype = kern_vtype;
1057 *vtype_id = kern_vtype_id;
1058 *data_member = kern_data_member;
1059
1060 return 0;
1061 }
1062
bpf_map__is_struct_ops(const struct bpf_map * map)1063 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
1064 {
1065 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
1066 }
1067
is_valid_st_ops_program(struct bpf_object * obj,const struct bpf_program * prog)1068 static bool is_valid_st_ops_program(struct bpf_object *obj,
1069 const struct bpf_program *prog)
1070 {
1071 int i;
1072
1073 for (i = 0; i < obj->nr_programs; i++) {
1074 if (&obj->programs[i] == prog)
1075 return prog->type == BPF_PROG_TYPE_STRUCT_OPS;
1076 }
1077
1078 return false;
1079 }
1080
1081 /* For each struct_ops program P, referenced from some struct_ops map M,
1082 * enable P.autoload if there are Ms for which M.autocreate is true,
1083 * disable P.autoload if for all Ms M.autocreate is false.
1084 * Don't change P.autoload for programs that are not referenced from any maps.
1085 */
bpf_object_adjust_struct_ops_autoload(struct bpf_object * obj)1086 static int bpf_object_adjust_struct_ops_autoload(struct bpf_object *obj)
1087 {
1088 struct bpf_program *prog, *slot_prog;
1089 struct bpf_map *map;
1090 int i, j, k, vlen;
1091
1092 for (i = 0; i < obj->nr_programs; ++i) {
1093 int should_load = false;
1094 int use_cnt = 0;
1095
1096 prog = &obj->programs[i];
1097 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS)
1098 continue;
1099
1100 for (j = 0; j < obj->nr_maps; ++j) {
1101 const struct btf_type *type;
1102
1103 map = &obj->maps[j];
1104 if (!bpf_map__is_struct_ops(map))
1105 continue;
1106
1107 type = btf__type_by_id(obj->btf, map->st_ops->type_id);
1108 vlen = btf_vlen(type);
1109 for (k = 0; k < vlen; ++k) {
1110 slot_prog = map->st_ops->progs[k];
1111 if (prog != slot_prog)
1112 continue;
1113
1114 use_cnt++;
1115 if (map->autocreate)
1116 should_load = true;
1117 }
1118 }
1119 if (use_cnt)
1120 prog->autoload = should_load;
1121 }
1122
1123 return 0;
1124 }
1125
1126 /* Init the map's fields that depend on kern_btf */
bpf_map__init_kern_struct_ops(struct bpf_map * map)1127 static int bpf_map__init_kern_struct_ops(struct bpf_map *map)
1128 {
1129 const struct btf_member *member, *kern_member, *kern_data_member;
1130 const struct btf_type *type, *kern_type, *kern_vtype;
1131 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
1132 struct bpf_object *obj = map->obj;
1133 const struct btf *btf = obj->btf;
1134 struct bpf_struct_ops *st_ops;
1135 const struct btf *kern_btf;
1136 struct module_btf *mod_btf = NULL;
1137 void *data, *kern_data;
1138 const char *tname;
1139 int err;
1140
1141 st_ops = map->st_ops;
1142 type = btf__type_by_id(btf, st_ops->type_id);
1143 tname = btf__name_by_offset(btf, type->name_off);
1144 err = find_struct_ops_kern_types(obj, tname, &mod_btf,
1145 &kern_type, &kern_type_id,
1146 &kern_vtype, &kern_vtype_id,
1147 &kern_data_member);
1148 if (err)
1149 return err;
1150
1151 kern_btf = mod_btf ? mod_btf->btf : obj->btf_vmlinux;
1152
1153 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
1154 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
1155
1156 map->mod_btf_fd = mod_btf ? mod_btf->fd : -1;
1157 map->def.value_size = kern_vtype->size;
1158 map->btf_vmlinux_value_type_id = kern_vtype_id;
1159
1160 st_ops->kern_vdata = calloc(1, kern_vtype->size);
1161 if (!st_ops->kern_vdata)
1162 return -ENOMEM;
1163
1164 data = st_ops->data;
1165 kern_data_off = kern_data_member->offset / 8;
1166 kern_data = st_ops->kern_vdata + kern_data_off;
1167
1168 member = btf_members(type);
1169 for (i = 0; i < btf_vlen(type); i++, member++) {
1170 const struct btf_type *mtype, *kern_mtype;
1171 __u32 mtype_id, kern_mtype_id;
1172 void *mdata, *kern_mdata;
1173 struct bpf_program *prog;
1174 __s64 msize, kern_msize;
1175 __u32 moff, kern_moff;
1176 __u32 kern_member_idx;
1177 const char *mname;
1178
1179 mname = btf__name_by_offset(btf, member->name_off);
1180 moff = member->offset / 8;
1181 mdata = data + moff;
1182 msize = btf__resolve_size(btf, member->type);
1183 if (msize < 0) {
1184 pr_warn("struct_ops init_kern %s: failed to resolve the size of member %s\n",
1185 map->name, mname);
1186 return msize;
1187 }
1188
1189 kern_member = find_member_by_name(kern_btf, kern_type, mname);
1190 if (!kern_member) {
1191 if (!libbpf_is_mem_zeroed(mdata, msize)) {
1192 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
1193 map->name, mname);
1194 return -ENOTSUP;
1195 }
1196
1197 if (st_ops->progs[i]) {
1198 /* If we had declaratively set struct_ops callback, we need to
1199 * force its autoload to false, because it doesn't have
1200 * a chance of succeeding from POV of the current struct_ops map.
1201 * If this program is still referenced somewhere else, though,
1202 * then bpf_object_adjust_struct_ops_autoload() will update its
1203 * autoload accordingly.
1204 */
1205 st_ops->progs[i]->autoload = false;
1206 st_ops->progs[i] = NULL;
1207 }
1208
1209 /* Skip all-zero/NULL fields if they are not present in the kernel BTF */
1210 pr_info("struct_ops %s: member %s not found in kernel, skipping it as it's set to zero\n",
1211 map->name, mname);
1212 continue;
1213 }
1214
1215 kern_member_idx = kern_member - btf_members(kern_type);
1216 if (btf_member_bitfield_size(type, i) ||
1217 btf_member_bitfield_size(kern_type, kern_member_idx)) {
1218 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
1219 map->name, mname);
1220 return -ENOTSUP;
1221 }
1222
1223 kern_moff = kern_member->offset / 8;
1224 kern_mdata = kern_data + kern_moff;
1225
1226 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
1227 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
1228 &kern_mtype_id);
1229 if (BTF_INFO_KIND(mtype->info) !=
1230 BTF_INFO_KIND(kern_mtype->info)) {
1231 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1232 map->name, mname, BTF_INFO_KIND(mtype->info),
1233 BTF_INFO_KIND(kern_mtype->info));
1234 return -ENOTSUP;
1235 }
1236
1237 if (btf_is_ptr(mtype)) {
1238 prog = *(void **)mdata;
1239 /* just like for !kern_member case above, reset declaratively
1240 * set (at compile time) program's autload to false,
1241 * if user replaced it with another program or NULL
1242 */
1243 if (st_ops->progs[i] && st_ops->progs[i] != prog)
1244 st_ops->progs[i]->autoload = false;
1245
1246 /* Update the value from the shadow type */
1247 st_ops->progs[i] = prog;
1248 if (!prog)
1249 continue;
1250
1251 if (!is_valid_st_ops_program(obj, prog)) {
1252 pr_warn("struct_ops init_kern %s: member %s is not a struct_ops program\n",
1253 map->name, mname);
1254 return -ENOTSUP;
1255 }
1256
1257 kern_mtype = skip_mods_and_typedefs(kern_btf,
1258 kern_mtype->type,
1259 &kern_mtype_id);
1260
1261 /* mtype->type must be a func_proto which was
1262 * guaranteed in bpf_object__collect_st_ops_relos(),
1263 * so only check kern_mtype for func_proto here.
1264 */
1265 if (!btf_is_func_proto(kern_mtype)) {
1266 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1267 map->name, mname);
1268 return -ENOTSUP;
1269 }
1270
1271 if (mod_btf)
1272 prog->attach_btf_obj_fd = mod_btf->fd;
1273
1274 /* if we haven't yet processed this BPF program, record proper
1275 * attach_btf_id and member_idx
1276 */
1277 if (!prog->attach_btf_id) {
1278 prog->attach_btf_id = kern_type_id;
1279 prog->expected_attach_type = kern_member_idx;
1280 }
1281
1282 /* struct_ops BPF prog can be re-used between multiple
1283 * .struct_ops & .struct_ops.link as long as it's the
1284 * same struct_ops struct definition and the same
1285 * function pointer field
1286 */
1287 if (prog->attach_btf_id != kern_type_id) {
1288 pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: attach_btf_id %u != kern_type_id %u\n",
1289 map->name, mname, prog->name, prog->sec_name, prog->type,
1290 prog->attach_btf_id, kern_type_id);
1291 return -EINVAL;
1292 }
1293 if (prog->expected_attach_type != kern_member_idx) {
1294 pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: expected_attach_type %u != kern_member_idx %u\n",
1295 map->name, mname, prog->name, prog->sec_name, prog->type,
1296 prog->expected_attach_type, kern_member_idx);
1297 return -EINVAL;
1298 }
1299
1300 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1301
1302 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1303 map->name, mname, prog->name, moff,
1304 kern_moff);
1305
1306 continue;
1307 }
1308
1309 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1310 if (kern_msize < 0 || msize != kern_msize) {
1311 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1312 map->name, mname, (ssize_t)msize,
1313 (ssize_t)kern_msize);
1314 return -ENOTSUP;
1315 }
1316
1317 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1318 map->name, mname, (unsigned int)msize,
1319 moff, kern_moff);
1320 memcpy(kern_mdata, mdata, msize);
1321 }
1322
1323 return 0;
1324 }
1325
bpf_object__init_kern_struct_ops_maps(struct bpf_object * obj)1326 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1327 {
1328 struct bpf_map *map;
1329 size_t i;
1330 int err;
1331
1332 for (i = 0; i < obj->nr_maps; i++) {
1333 map = &obj->maps[i];
1334
1335 if (!bpf_map__is_struct_ops(map))
1336 continue;
1337
1338 if (!map->autocreate)
1339 continue;
1340
1341 err = bpf_map__init_kern_struct_ops(map);
1342 if (err)
1343 return err;
1344 }
1345
1346 return 0;
1347 }
1348
init_struct_ops_maps(struct bpf_object * obj,const char * sec_name,int shndx,Elf_Data * data)1349 static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name,
1350 int shndx, Elf_Data *data)
1351 {
1352 const struct btf_type *type, *datasec;
1353 const struct btf_var_secinfo *vsi;
1354 struct bpf_struct_ops *st_ops;
1355 const char *tname, *var_name;
1356 __s32 type_id, datasec_id;
1357 const struct btf *btf;
1358 struct bpf_map *map;
1359 __u32 i;
1360
1361 if (shndx == -1)
1362 return 0;
1363
1364 btf = obj->btf;
1365 datasec_id = btf__find_by_name_kind(btf, sec_name,
1366 BTF_KIND_DATASEC);
1367 if (datasec_id < 0) {
1368 pr_warn("struct_ops init: DATASEC %s not found\n",
1369 sec_name);
1370 return -EINVAL;
1371 }
1372
1373 datasec = btf__type_by_id(btf, datasec_id);
1374 vsi = btf_var_secinfos(datasec);
1375 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1376 type = btf__type_by_id(obj->btf, vsi->type);
1377 var_name = btf__name_by_offset(obj->btf, type->name_off);
1378
1379 type_id = btf__resolve_type(obj->btf, vsi->type);
1380 if (type_id < 0) {
1381 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1382 vsi->type, sec_name);
1383 return -EINVAL;
1384 }
1385
1386 type = btf__type_by_id(obj->btf, type_id);
1387 tname = btf__name_by_offset(obj->btf, type->name_off);
1388 if (!tname[0]) {
1389 pr_warn("struct_ops init: anonymous type is not supported\n");
1390 return -ENOTSUP;
1391 }
1392 if (!btf_is_struct(type)) {
1393 pr_warn("struct_ops init: %s is not a struct\n", tname);
1394 return -EINVAL;
1395 }
1396
1397 map = bpf_object__add_map(obj);
1398 if (IS_ERR(map))
1399 return PTR_ERR(map);
1400
1401 map->sec_idx = shndx;
1402 map->sec_offset = vsi->offset;
1403 map->name = strdup(var_name);
1404 if (!map->name)
1405 return -ENOMEM;
1406 map->btf_value_type_id = type_id;
1407
1408 /* Follow same convention as for programs autoload:
1409 * SEC("?.struct_ops") means map is not created by default.
1410 */
1411 if (sec_name[0] == '?') {
1412 map->autocreate = false;
1413 /* from now on forget there was ? in section name */
1414 sec_name++;
1415 }
1416
1417 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1418 map->def.key_size = sizeof(int);
1419 map->def.value_size = type->size;
1420 map->def.max_entries = 1;
1421 map->def.map_flags = strcmp(sec_name, STRUCT_OPS_LINK_SEC) == 0 ? BPF_F_LINK : 0;
1422 map->autoattach = true;
1423
1424 map->st_ops = calloc(1, sizeof(*map->st_ops));
1425 if (!map->st_ops)
1426 return -ENOMEM;
1427 st_ops = map->st_ops;
1428 st_ops->data = malloc(type->size);
1429 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1430 st_ops->kern_func_off = malloc(btf_vlen(type) *
1431 sizeof(*st_ops->kern_func_off));
1432 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1433 return -ENOMEM;
1434
1435 if (vsi->offset + type->size > data->d_size) {
1436 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1437 var_name, sec_name);
1438 return -EINVAL;
1439 }
1440
1441 memcpy(st_ops->data,
1442 data->d_buf + vsi->offset,
1443 type->size);
1444 st_ops->type_id = type_id;
1445
1446 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1447 tname, type_id, var_name, vsi->offset);
1448 }
1449
1450 return 0;
1451 }
1452
bpf_object_init_struct_ops(struct bpf_object * obj)1453 static int bpf_object_init_struct_ops(struct bpf_object *obj)
1454 {
1455 const char *sec_name;
1456 int sec_idx, err;
1457
1458 for (sec_idx = 0; sec_idx < obj->efile.sec_cnt; ++sec_idx) {
1459 struct elf_sec_desc *desc = &obj->efile.secs[sec_idx];
1460
1461 if (desc->sec_type != SEC_ST_OPS)
1462 continue;
1463
1464 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1465 if (!sec_name)
1466 return -LIBBPF_ERRNO__FORMAT;
1467
1468 err = init_struct_ops_maps(obj, sec_name, sec_idx, desc->data);
1469 if (err)
1470 return err;
1471 }
1472
1473 return 0;
1474 }
1475
bpf_object__new(const char * path,const void * obj_buf,size_t obj_buf_sz,const char * obj_name)1476 static struct bpf_object *bpf_object__new(const char *path,
1477 const void *obj_buf,
1478 size_t obj_buf_sz,
1479 const char *obj_name)
1480 {
1481 struct bpf_object *obj;
1482 char *end;
1483
1484 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1485 if (!obj) {
1486 pr_warn("alloc memory failed for %s\n", path);
1487 return ERR_PTR(-ENOMEM);
1488 }
1489
1490 strcpy(obj->path, path);
1491 if (obj_name) {
1492 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1493 } else {
1494 /* Using basename() GNU version which doesn't modify arg. */
1495 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1496 end = strchr(obj->name, '.');
1497 if (end)
1498 *end = 0;
1499 }
1500
1501 obj->efile.fd = -1;
1502 /*
1503 * Caller of this function should also call
1504 * bpf_object__elf_finish() after data collection to return
1505 * obj_buf to user. If not, we should duplicate the buffer to
1506 * avoid user freeing them before elf finish.
1507 */
1508 obj->efile.obj_buf = obj_buf;
1509 obj->efile.obj_buf_sz = obj_buf_sz;
1510 obj->efile.btf_maps_shndx = -1;
1511 obj->kconfig_map_idx = -1;
1512
1513 obj->kern_version = get_kernel_version();
1514 obj->loaded = false;
1515
1516 return obj;
1517 }
1518
bpf_object__elf_finish(struct bpf_object * obj)1519 static void bpf_object__elf_finish(struct bpf_object *obj)
1520 {
1521 if (!obj->efile.elf)
1522 return;
1523
1524 elf_end(obj->efile.elf);
1525 obj->efile.elf = NULL;
1526 obj->efile.ehdr = NULL;
1527 obj->efile.symbols = NULL;
1528 obj->efile.arena_data = NULL;
1529
1530 zfree(&obj->efile.secs);
1531 obj->efile.sec_cnt = 0;
1532 zclose(obj->efile.fd);
1533 obj->efile.obj_buf = NULL;
1534 obj->efile.obj_buf_sz = 0;
1535 }
1536
bpf_object__elf_init(struct bpf_object * obj)1537 static int bpf_object__elf_init(struct bpf_object *obj)
1538 {
1539 Elf64_Ehdr *ehdr;
1540 int err = 0;
1541 Elf *elf;
1542
1543 if (obj->efile.elf) {
1544 pr_warn("elf: init internal error\n");
1545 return -LIBBPF_ERRNO__LIBELF;
1546 }
1547
1548 if (obj->efile.obj_buf_sz > 0) {
1549 /* obj_buf should have been validated by bpf_object__open_mem(). */
1550 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1551 } else {
1552 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1553 if (obj->efile.fd < 0) {
1554 err = -errno;
1555 pr_warn("elf: failed to open %s: %s\n", obj->path, errstr(err));
1556 return err;
1557 }
1558
1559 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1560 }
1561
1562 if (!elf) {
1563 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1564 err = -LIBBPF_ERRNO__LIBELF;
1565 goto errout;
1566 }
1567
1568 obj->efile.elf = elf;
1569
1570 if (elf_kind(elf) != ELF_K_ELF) {
1571 err = -LIBBPF_ERRNO__FORMAT;
1572 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1573 goto errout;
1574 }
1575
1576 if (gelf_getclass(elf) != ELFCLASS64) {
1577 err = -LIBBPF_ERRNO__FORMAT;
1578 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1579 goto errout;
1580 }
1581
1582 obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1583 if (!obj->efile.ehdr) {
1584 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1585 err = -LIBBPF_ERRNO__FORMAT;
1586 goto errout;
1587 }
1588
1589 /* Validate ELF object endianness... */
1590 if (ehdr->e_ident[EI_DATA] != ELFDATA2LSB &&
1591 ehdr->e_ident[EI_DATA] != ELFDATA2MSB) {
1592 err = -LIBBPF_ERRNO__ENDIAN;
1593 pr_warn("elf: '%s' has unknown byte order\n", obj->path);
1594 goto errout;
1595 }
1596 /* and save after bpf_object_open() frees ELF data */
1597 obj->byteorder = ehdr->e_ident[EI_DATA];
1598
1599 if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1600 pr_warn("elf: failed to get section names section index for %s: %s\n",
1601 obj->path, elf_errmsg(-1));
1602 err = -LIBBPF_ERRNO__FORMAT;
1603 goto errout;
1604 }
1605
1606 /* ELF is corrupted/truncated, avoid calling elf_strptr. */
1607 if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1608 pr_warn("elf: failed to get section names strings from %s: %s\n",
1609 obj->path, elf_errmsg(-1));
1610 err = -LIBBPF_ERRNO__FORMAT;
1611 goto errout;
1612 }
1613
1614 /* Old LLVM set e_machine to EM_NONE */
1615 if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1616 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1617 err = -LIBBPF_ERRNO__FORMAT;
1618 goto errout;
1619 }
1620
1621 return 0;
1622 errout:
1623 bpf_object__elf_finish(obj);
1624 return err;
1625 }
1626
is_native_endianness(struct bpf_object * obj)1627 static bool is_native_endianness(struct bpf_object *obj)
1628 {
1629 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1630 return obj->byteorder == ELFDATA2LSB;
1631 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1632 return obj->byteorder == ELFDATA2MSB;
1633 #else
1634 # error "Unrecognized __BYTE_ORDER__"
1635 #endif
1636 }
1637
1638 static int
bpf_object__init_license(struct bpf_object * obj,void * data,size_t size)1639 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1640 {
1641 if (!data) {
1642 pr_warn("invalid license section in %s\n", obj->path);
1643 return -LIBBPF_ERRNO__FORMAT;
1644 }
1645 /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1646 * go over allowed ELF data section buffer
1647 */
1648 libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1649 pr_debug("license of %s is %s\n", obj->path, obj->license);
1650 return 0;
1651 }
1652
1653 static int
bpf_object__init_kversion(struct bpf_object * obj,void * data,size_t size)1654 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1655 {
1656 __u32 kver;
1657
1658 if (!data || size != sizeof(kver)) {
1659 pr_warn("invalid kver section in %s\n", obj->path);
1660 return -LIBBPF_ERRNO__FORMAT;
1661 }
1662 memcpy(&kver, data, sizeof(kver));
1663 obj->kern_version = kver;
1664 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1665 return 0;
1666 }
1667
bpf_map_type__is_map_in_map(enum bpf_map_type type)1668 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1669 {
1670 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1671 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1672 return true;
1673 return false;
1674 }
1675
find_elf_sec_sz(const struct bpf_object * obj,const char * name,__u32 * size)1676 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1677 {
1678 Elf_Data *data;
1679 Elf_Scn *scn;
1680
1681 if (!name)
1682 return -EINVAL;
1683
1684 scn = elf_sec_by_name(obj, name);
1685 data = elf_sec_data(obj, scn);
1686 if (data) {
1687 *size = data->d_size;
1688 return 0; /* found it */
1689 }
1690
1691 return -ENOENT;
1692 }
1693
find_elf_var_sym(const struct bpf_object * obj,const char * name)1694 static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name)
1695 {
1696 Elf_Data *symbols = obj->efile.symbols;
1697 const char *sname;
1698 size_t si;
1699
1700 for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1701 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1702
1703 if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1704 continue;
1705
1706 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
1707 ELF64_ST_BIND(sym->st_info) != STB_WEAK)
1708 continue;
1709
1710 sname = elf_sym_str(obj, sym->st_name);
1711 if (!sname) {
1712 pr_warn("failed to get sym name string for var %s\n", name);
1713 return ERR_PTR(-EIO);
1714 }
1715 if (strcmp(name, sname) == 0)
1716 return sym;
1717 }
1718
1719 return ERR_PTR(-ENOENT);
1720 }
1721
1722 /* Some versions of Android don't provide memfd_create() in their libc
1723 * implementation, so avoid complications and just go straight to Linux
1724 * syscall.
1725 */
sys_memfd_create(const char * name,unsigned flags)1726 static int sys_memfd_create(const char *name, unsigned flags)
1727 {
1728 return syscall(__NR_memfd_create, name, flags);
1729 }
1730
1731 #ifndef MFD_CLOEXEC
1732 #define MFD_CLOEXEC 0x0001U
1733 #endif
1734 #ifndef MFD_NOEXEC_SEAL
1735 #define MFD_NOEXEC_SEAL 0x0008U
1736 #endif
1737
create_placeholder_fd(void)1738 static int create_placeholder_fd(void)
1739 {
1740 unsigned int flags = MFD_CLOEXEC | MFD_NOEXEC_SEAL;
1741 const char *name = "libbpf-placeholder-fd";
1742 int fd;
1743
1744 fd = ensure_good_fd(sys_memfd_create(name, flags));
1745 if (fd >= 0)
1746 return fd;
1747 else if (errno != EINVAL)
1748 return -errno;
1749
1750 /* Possibly running on kernel without MFD_NOEXEC_SEAL */
1751 fd = ensure_good_fd(sys_memfd_create(name, flags & ~MFD_NOEXEC_SEAL));
1752 if (fd < 0)
1753 return -errno;
1754 return fd;
1755 }
1756
bpf_object__add_map(struct bpf_object * obj)1757 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1758 {
1759 struct bpf_map *map;
1760 int err;
1761
1762 err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap,
1763 sizeof(*obj->maps), obj->nr_maps + 1);
1764 if (err)
1765 return ERR_PTR(err);
1766
1767 map = &obj->maps[obj->nr_maps++];
1768 map->obj = obj;
1769 /* Preallocate map FD without actually creating BPF map just yet.
1770 * These map FD "placeholders" will be reused later without changing
1771 * FD value when map is actually created in the kernel.
1772 *
1773 * This is useful to be able to perform BPF program relocations
1774 * without having to create BPF maps before that step. This allows us
1775 * to finalize and load BTF very late in BPF object's loading phase,
1776 * right before BPF maps have to be created and BPF programs have to
1777 * be loaded. By having these map FD placeholders we can perform all
1778 * the sanitizations, relocations, and any other adjustments before we
1779 * start creating actual BPF kernel objects (BTF, maps, progs).
1780 */
1781 map->fd = create_placeholder_fd();
1782 if (map->fd < 0)
1783 return ERR_PTR(map->fd);
1784 map->inner_map_fd = -1;
1785 map->autocreate = true;
1786
1787 return map;
1788 }
1789
array_map_mmap_sz(unsigned int value_sz,unsigned int max_entries)1790 static size_t array_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
1791 {
1792 const long page_sz = sysconf(_SC_PAGE_SIZE);
1793 size_t map_sz;
1794
1795 map_sz = (size_t)roundup(value_sz, 8) * max_entries;
1796 map_sz = roundup(map_sz, page_sz);
1797 return map_sz;
1798 }
1799
bpf_map_mmap_sz(const struct bpf_map * map)1800 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1801 {
1802 const long page_sz = sysconf(_SC_PAGE_SIZE);
1803
1804 switch (map->def.type) {
1805 case BPF_MAP_TYPE_ARRAY:
1806 return array_map_mmap_sz(map->def.value_size, map->def.max_entries);
1807 case BPF_MAP_TYPE_ARENA:
1808 return page_sz * map->def.max_entries;
1809 default:
1810 return 0; /* not supported */
1811 }
1812 }
1813
bpf_map_mmap_resize(struct bpf_map * map,size_t old_sz,size_t new_sz)1814 static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz)
1815 {
1816 void *mmaped;
1817
1818 if (!map->mmaped)
1819 return -EINVAL;
1820
1821 if (old_sz == new_sz)
1822 return 0;
1823
1824 mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1825 if (mmaped == MAP_FAILED)
1826 return -errno;
1827
1828 memcpy(mmaped, map->mmaped, min(old_sz, new_sz));
1829 munmap(map->mmaped, old_sz);
1830 map->mmaped = mmaped;
1831 return 0;
1832 }
1833
internal_map_name(struct bpf_object * obj,const char * real_name)1834 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1835 {
1836 char map_name[BPF_OBJ_NAME_LEN], *p;
1837 int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1838
1839 /* This is one of the more confusing parts of libbpf for various
1840 * reasons, some of which are historical. The original idea for naming
1841 * internal names was to include as much of BPF object name prefix as
1842 * possible, so that it can be distinguished from similar internal
1843 * maps of a different BPF object.
1844 * As an example, let's say we have bpf_object named 'my_object_name'
1845 * and internal map corresponding to '.rodata' ELF section. The final
1846 * map name advertised to user and to the kernel will be
1847 * 'my_objec.rodata', taking first 8 characters of object name and
1848 * entire 7 characters of '.rodata'.
1849 * Somewhat confusingly, if internal map ELF section name is shorter
1850 * than 7 characters, e.g., '.bss', we still reserve 7 characters
1851 * for the suffix, even though we only have 4 actual characters, and
1852 * resulting map will be called 'my_objec.bss', not even using all 15
1853 * characters allowed by the kernel. Oh well, at least the truncated
1854 * object name is somewhat consistent in this case. But if the map
1855 * name is '.kconfig', we'll still have entirety of '.kconfig' added
1856 * (8 chars) and thus will be left with only first 7 characters of the
1857 * object name ('my_obje'). Happy guessing, user, that the final map
1858 * name will be "my_obje.kconfig".
1859 * Now, with libbpf starting to support arbitrarily named .rodata.*
1860 * and .data.* data sections, it's possible that ELF section name is
1861 * longer than allowed 15 chars, so we now need to be careful to take
1862 * only up to 15 first characters of ELF name, taking no BPF object
1863 * name characters at all. So '.rodata.abracadabra' will result in
1864 * '.rodata.abracad' kernel and user-visible name.
1865 * We need to keep this convoluted logic intact for .data, .bss and
1866 * .rodata maps, but for new custom .data.custom and .rodata.custom
1867 * maps we use their ELF names as is, not prepending bpf_object name
1868 * in front. We still need to truncate them to 15 characters for the
1869 * kernel. Full name can be recovered for such maps by using DATASEC
1870 * BTF type associated with such map's value type, though.
1871 */
1872 if (sfx_len >= BPF_OBJ_NAME_LEN)
1873 sfx_len = BPF_OBJ_NAME_LEN - 1;
1874
1875 /* if there are two or more dots in map name, it's a custom dot map */
1876 if (strchr(real_name + 1, '.') != NULL)
1877 pfx_len = 0;
1878 else
1879 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1880
1881 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1882 sfx_len, real_name);
1883
1884 /* sanities map name to characters allowed by kernel */
1885 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1886 if (!isalnum(*p) && *p != '_' && *p != '.')
1887 *p = '_';
1888
1889 return strdup(map_name);
1890 }
1891
1892 static int
1893 map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map);
1894
1895 /* Internal BPF map is mmap()'able only if at least one of corresponding
1896 * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL
1897 * variable and it's not marked as __hidden (which turns it into, effectively,
1898 * a STATIC variable).
1899 */
map_is_mmapable(struct bpf_object * obj,struct bpf_map * map)1900 static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map)
1901 {
1902 const struct btf_type *t, *vt;
1903 struct btf_var_secinfo *vsi;
1904 int i, n;
1905
1906 if (!map->btf_value_type_id)
1907 return false;
1908
1909 t = btf__type_by_id(obj->btf, map->btf_value_type_id);
1910 if (!btf_is_datasec(t))
1911 return false;
1912
1913 vsi = btf_var_secinfos(t);
1914 for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) {
1915 vt = btf__type_by_id(obj->btf, vsi->type);
1916 if (!btf_is_var(vt))
1917 continue;
1918
1919 if (btf_var(vt)->linkage != BTF_VAR_STATIC)
1920 return true;
1921 }
1922
1923 return false;
1924 }
1925
1926 static int
bpf_object__init_internal_map(struct bpf_object * obj,enum libbpf_map_type type,const char * real_name,int sec_idx,void * data,size_t data_sz)1927 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1928 const char *real_name, int sec_idx, void *data, size_t data_sz)
1929 {
1930 struct bpf_map_def *def;
1931 struct bpf_map *map;
1932 size_t mmap_sz;
1933 int err;
1934
1935 map = bpf_object__add_map(obj);
1936 if (IS_ERR(map))
1937 return PTR_ERR(map);
1938
1939 map->libbpf_type = type;
1940 map->sec_idx = sec_idx;
1941 map->sec_offset = 0;
1942 map->real_name = strdup(real_name);
1943 map->name = internal_map_name(obj, real_name);
1944 if (!map->real_name || !map->name) {
1945 zfree(&map->real_name);
1946 zfree(&map->name);
1947 return -ENOMEM;
1948 }
1949
1950 def = &map->def;
1951 def->type = BPF_MAP_TYPE_ARRAY;
1952 def->key_size = sizeof(int);
1953 def->value_size = data_sz;
1954 def->max_entries = 1;
1955 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1956 ? BPF_F_RDONLY_PROG : 0;
1957
1958 /* failures are fine because of maps like .rodata.str1.1 */
1959 (void) map_fill_btf_type_info(obj, map);
1960
1961 if (map_is_mmapable(obj, map))
1962 def->map_flags |= BPF_F_MMAPABLE;
1963
1964 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1965 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1966
1967 mmap_sz = bpf_map_mmap_sz(map);
1968 map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE,
1969 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1970 if (map->mmaped == MAP_FAILED) {
1971 err = -errno;
1972 map->mmaped = NULL;
1973 pr_warn("failed to alloc map '%s' content buffer: %s\n", map->name, errstr(err));
1974 zfree(&map->real_name);
1975 zfree(&map->name);
1976 return err;
1977 }
1978
1979 if (data)
1980 memcpy(map->mmaped, data, data_sz);
1981
1982 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1983 return 0;
1984 }
1985
bpf_object__init_global_data_maps(struct bpf_object * obj)1986 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1987 {
1988 struct elf_sec_desc *sec_desc;
1989 const char *sec_name;
1990 int err = 0, sec_idx;
1991
1992 /*
1993 * Populate obj->maps with libbpf internal maps.
1994 */
1995 for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1996 sec_desc = &obj->efile.secs[sec_idx];
1997
1998 /* Skip recognized sections with size 0. */
1999 if (!sec_desc->data || sec_desc->data->d_size == 0)
2000 continue;
2001
2002 switch (sec_desc->sec_type) {
2003 case SEC_DATA:
2004 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2005 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
2006 sec_name, sec_idx,
2007 sec_desc->data->d_buf,
2008 sec_desc->data->d_size);
2009 break;
2010 case SEC_RODATA:
2011 obj->has_rodata = true;
2012 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2013 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
2014 sec_name, sec_idx,
2015 sec_desc->data->d_buf,
2016 sec_desc->data->d_size);
2017 break;
2018 case SEC_BSS:
2019 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2020 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
2021 sec_name, sec_idx,
2022 NULL,
2023 sec_desc->data->d_size);
2024 break;
2025 default:
2026 /* skip */
2027 break;
2028 }
2029 if (err)
2030 return err;
2031 }
2032 return 0;
2033 }
2034
2035
find_extern_by_name(const struct bpf_object * obj,const void * name)2036 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
2037 const void *name)
2038 {
2039 int i;
2040
2041 for (i = 0; i < obj->nr_extern; i++) {
2042 if (strcmp(obj->externs[i].name, name) == 0)
2043 return &obj->externs[i];
2044 }
2045 return NULL;
2046 }
2047
find_extern_by_name_with_len(const struct bpf_object * obj,const void * name,int len)2048 static struct extern_desc *find_extern_by_name_with_len(const struct bpf_object *obj,
2049 const void *name, int len)
2050 {
2051 const char *ext_name;
2052 int i;
2053
2054 for (i = 0; i < obj->nr_extern; i++) {
2055 ext_name = obj->externs[i].name;
2056 if (strlen(ext_name) == len && strncmp(ext_name, name, len) == 0)
2057 return &obj->externs[i];
2058 }
2059 return NULL;
2060 }
2061
set_kcfg_value_tri(struct extern_desc * ext,void * ext_val,char value)2062 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
2063 char value)
2064 {
2065 switch (ext->kcfg.type) {
2066 case KCFG_BOOL:
2067 if (value == 'm') {
2068 pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n",
2069 ext->name, value);
2070 return -EINVAL;
2071 }
2072 *(bool *)ext_val = value == 'y' ? true : false;
2073 break;
2074 case KCFG_TRISTATE:
2075 if (value == 'y')
2076 *(enum libbpf_tristate *)ext_val = TRI_YES;
2077 else if (value == 'm')
2078 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
2079 else /* value == 'n' */
2080 *(enum libbpf_tristate *)ext_val = TRI_NO;
2081 break;
2082 case KCFG_CHAR:
2083 *(char *)ext_val = value;
2084 break;
2085 case KCFG_UNKNOWN:
2086 case KCFG_INT:
2087 case KCFG_CHAR_ARR:
2088 default:
2089 pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n",
2090 ext->name, value);
2091 return -EINVAL;
2092 }
2093 ext->is_set = true;
2094 return 0;
2095 }
2096
set_kcfg_value_str(struct extern_desc * ext,char * ext_val,const char * value)2097 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
2098 const char *value)
2099 {
2100 size_t len;
2101
2102 if (ext->kcfg.type != KCFG_CHAR_ARR) {
2103 pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n",
2104 ext->name, value);
2105 return -EINVAL;
2106 }
2107
2108 len = strlen(value);
2109 if (value[len - 1] != '"') {
2110 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
2111 ext->name, value);
2112 return -EINVAL;
2113 }
2114
2115 /* strip quotes */
2116 len -= 2;
2117 if (len >= ext->kcfg.sz) {
2118 pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n",
2119 ext->name, value, len, ext->kcfg.sz - 1);
2120 len = ext->kcfg.sz - 1;
2121 }
2122 memcpy(ext_val, value + 1, len);
2123 ext_val[len] = '\0';
2124 ext->is_set = true;
2125 return 0;
2126 }
2127
parse_u64(const char * value,__u64 * res)2128 static int parse_u64(const char *value, __u64 *res)
2129 {
2130 char *value_end;
2131 int err;
2132
2133 errno = 0;
2134 *res = strtoull(value, &value_end, 0);
2135 if (errno) {
2136 err = -errno;
2137 pr_warn("failed to parse '%s': %s\n", value, errstr(err));
2138 return err;
2139 }
2140 if (*value_end) {
2141 pr_warn("failed to parse '%s' as integer completely\n", value);
2142 return -EINVAL;
2143 }
2144 return 0;
2145 }
2146
is_kcfg_value_in_range(const struct extern_desc * ext,__u64 v)2147 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
2148 {
2149 int bit_sz = ext->kcfg.sz * 8;
2150
2151 if (ext->kcfg.sz == 8)
2152 return true;
2153
2154 /* Validate that value stored in u64 fits in integer of `ext->sz`
2155 * bytes size without any loss of information. If the target integer
2156 * is signed, we rely on the following limits of integer type of
2157 * Y bits and subsequent transformation:
2158 *
2159 * -2^(Y-1) <= X <= 2^(Y-1) - 1
2160 * 0 <= X + 2^(Y-1) <= 2^Y - 1
2161 * 0 <= X + 2^(Y-1) < 2^Y
2162 *
2163 * For unsigned target integer, check that all the (64 - Y) bits are
2164 * zero.
2165 */
2166 if (ext->kcfg.is_signed)
2167 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
2168 else
2169 return (v >> bit_sz) == 0;
2170 }
2171
set_kcfg_value_num(struct extern_desc * ext,void * ext_val,__u64 value)2172 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
2173 __u64 value)
2174 {
2175 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR &&
2176 ext->kcfg.type != KCFG_BOOL) {
2177 pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n",
2178 ext->name, (unsigned long long)value);
2179 return -EINVAL;
2180 }
2181 if (ext->kcfg.type == KCFG_BOOL && value > 1) {
2182 pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n",
2183 ext->name, (unsigned long long)value);
2184 return -EINVAL;
2185
2186 }
2187 if (!is_kcfg_value_in_range(ext, value)) {
2188 pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n",
2189 ext->name, (unsigned long long)value, ext->kcfg.sz);
2190 return -ERANGE;
2191 }
2192 switch (ext->kcfg.sz) {
2193 case 1:
2194 *(__u8 *)ext_val = value;
2195 break;
2196 case 2:
2197 *(__u16 *)ext_val = value;
2198 break;
2199 case 4:
2200 *(__u32 *)ext_val = value;
2201 break;
2202 case 8:
2203 *(__u64 *)ext_val = value;
2204 break;
2205 default:
2206 return -EINVAL;
2207 }
2208 ext->is_set = true;
2209 return 0;
2210 }
2211
bpf_object__process_kconfig_line(struct bpf_object * obj,char * buf,void * data)2212 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
2213 char *buf, void *data)
2214 {
2215 struct extern_desc *ext;
2216 char *sep, *value;
2217 int len, err = 0;
2218 void *ext_val;
2219 __u64 num;
2220
2221 if (!str_has_pfx(buf, "CONFIG_"))
2222 return 0;
2223
2224 sep = strchr(buf, '=');
2225 if (!sep) {
2226 pr_warn("failed to parse '%s': no separator\n", buf);
2227 return -EINVAL;
2228 }
2229
2230 /* Trim ending '\n' */
2231 len = strlen(buf);
2232 if (buf[len - 1] == '\n')
2233 buf[len - 1] = '\0';
2234 /* Split on '=' and ensure that a value is present. */
2235 *sep = '\0';
2236 if (!sep[1]) {
2237 *sep = '=';
2238 pr_warn("failed to parse '%s': no value\n", buf);
2239 return -EINVAL;
2240 }
2241
2242 ext = find_extern_by_name(obj, buf);
2243 if (!ext || ext->is_set)
2244 return 0;
2245
2246 ext_val = data + ext->kcfg.data_off;
2247 value = sep + 1;
2248
2249 switch (*value) {
2250 case 'y': case 'n': case 'm':
2251 err = set_kcfg_value_tri(ext, ext_val, *value);
2252 break;
2253 case '"':
2254 err = set_kcfg_value_str(ext, ext_val, value);
2255 break;
2256 default:
2257 /* assume integer */
2258 err = parse_u64(value, &num);
2259 if (err) {
2260 pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value);
2261 return err;
2262 }
2263 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
2264 pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value);
2265 return -EINVAL;
2266 }
2267 err = set_kcfg_value_num(ext, ext_val, num);
2268 break;
2269 }
2270 if (err)
2271 return err;
2272 pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value);
2273 return 0;
2274 }
2275
bpf_object__read_kconfig_file(struct bpf_object * obj,void * data)2276 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
2277 {
2278 char buf[PATH_MAX];
2279 struct utsname uts;
2280 int len, err = 0;
2281 gzFile file;
2282
2283 uname(&uts);
2284 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
2285 if (len < 0)
2286 return -EINVAL;
2287 else if (len >= PATH_MAX)
2288 return -ENAMETOOLONG;
2289
2290 /* gzopen also accepts uncompressed files. */
2291 file = gzopen(buf, "re");
2292 if (!file)
2293 file = gzopen("/proc/config.gz", "re");
2294
2295 if (!file) {
2296 pr_warn("failed to open system Kconfig\n");
2297 return -ENOENT;
2298 }
2299
2300 while (gzgets(file, buf, sizeof(buf))) {
2301 err = bpf_object__process_kconfig_line(obj, buf, data);
2302 if (err) {
2303 pr_warn("error parsing system Kconfig line '%s': %s\n",
2304 buf, errstr(err));
2305 goto out;
2306 }
2307 }
2308
2309 out:
2310 gzclose(file);
2311 return err;
2312 }
2313
bpf_object__read_kconfig_mem(struct bpf_object * obj,const char * config,void * data)2314 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
2315 const char *config, void *data)
2316 {
2317 char buf[PATH_MAX];
2318 int err = 0;
2319 FILE *file;
2320
2321 file = fmemopen((void *)config, strlen(config), "r");
2322 if (!file) {
2323 err = -errno;
2324 pr_warn("failed to open in-memory Kconfig: %s\n", errstr(err));
2325 return err;
2326 }
2327
2328 while (fgets(buf, sizeof(buf), file)) {
2329 err = bpf_object__process_kconfig_line(obj, buf, data);
2330 if (err) {
2331 pr_warn("error parsing in-memory Kconfig line '%s': %s\n",
2332 buf, errstr(err));
2333 break;
2334 }
2335 }
2336
2337 fclose(file);
2338 return err;
2339 }
2340
bpf_object__init_kconfig_map(struct bpf_object * obj)2341 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
2342 {
2343 struct extern_desc *last_ext = NULL, *ext;
2344 size_t map_sz;
2345 int i, err;
2346
2347 for (i = 0; i < obj->nr_extern; i++) {
2348 ext = &obj->externs[i];
2349 if (ext->type == EXT_KCFG)
2350 last_ext = ext;
2351 }
2352
2353 if (!last_ext)
2354 return 0;
2355
2356 map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
2357 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
2358 ".kconfig", obj->efile.symbols_shndx,
2359 NULL, map_sz);
2360 if (err)
2361 return err;
2362
2363 obj->kconfig_map_idx = obj->nr_maps - 1;
2364
2365 return 0;
2366 }
2367
2368 const struct btf_type *
skip_mods_and_typedefs(const struct btf * btf,__u32 id,__u32 * res_id)2369 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2370 {
2371 const struct btf_type *t = btf__type_by_id(btf, id);
2372
2373 if (res_id)
2374 *res_id = id;
2375
2376 while (btf_is_mod(t) || btf_is_typedef(t)) {
2377 if (res_id)
2378 *res_id = t->type;
2379 t = btf__type_by_id(btf, t->type);
2380 }
2381
2382 return t;
2383 }
2384
2385 static const struct btf_type *
resolve_func_ptr(const struct btf * btf,__u32 id,__u32 * res_id)2386 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2387 {
2388 const struct btf_type *t;
2389
2390 t = skip_mods_and_typedefs(btf, id, NULL);
2391 if (!btf_is_ptr(t))
2392 return NULL;
2393
2394 t = skip_mods_and_typedefs(btf, t->type, res_id);
2395
2396 return btf_is_func_proto(t) ? t : NULL;
2397 }
2398
__btf_kind_str(__u16 kind)2399 static const char *__btf_kind_str(__u16 kind)
2400 {
2401 switch (kind) {
2402 case BTF_KIND_UNKN: return "void";
2403 case BTF_KIND_INT: return "int";
2404 case BTF_KIND_PTR: return "ptr";
2405 case BTF_KIND_ARRAY: return "array";
2406 case BTF_KIND_STRUCT: return "struct";
2407 case BTF_KIND_UNION: return "union";
2408 case BTF_KIND_ENUM: return "enum";
2409 case BTF_KIND_FWD: return "fwd";
2410 case BTF_KIND_TYPEDEF: return "typedef";
2411 case BTF_KIND_VOLATILE: return "volatile";
2412 case BTF_KIND_CONST: return "const";
2413 case BTF_KIND_RESTRICT: return "restrict";
2414 case BTF_KIND_FUNC: return "func";
2415 case BTF_KIND_FUNC_PROTO: return "func_proto";
2416 case BTF_KIND_VAR: return "var";
2417 case BTF_KIND_DATASEC: return "datasec";
2418 case BTF_KIND_FLOAT: return "float";
2419 case BTF_KIND_DECL_TAG: return "decl_tag";
2420 case BTF_KIND_TYPE_TAG: return "type_tag";
2421 case BTF_KIND_ENUM64: return "enum64";
2422 default: return "unknown";
2423 }
2424 }
2425
btf_kind_str(const struct btf_type * t)2426 const char *btf_kind_str(const struct btf_type *t)
2427 {
2428 return __btf_kind_str(btf_kind(t));
2429 }
2430
2431 /*
2432 * Fetch integer attribute of BTF map definition. Such attributes are
2433 * represented using a pointer to an array, in which dimensionality of array
2434 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2435 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2436 * type definition, while using only sizeof(void *) space in ELF data section.
2437 */
get_map_field_int(const char * map_name,const struct btf * btf,const struct btf_member * m,__u32 * res)2438 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2439 const struct btf_member *m, __u32 *res)
2440 {
2441 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2442 const char *name = btf__name_by_offset(btf, m->name_off);
2443 const struct btf_array *arr_info;
2444 const struct btf_type *arr_t;
2445
2446 if (!btf_is_ptr(t)) {
2447 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2448 map_name, name, btf_kind_str(t));
2449 return false;
2450 }
2451
2452 arr_t = btf__type_by_id(btf, t->type);
2453 if (!arr_t) {
2454 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2455 map_name, name, t->type);
2456 return false;
2457 }
2458 if (!btf_is_array(arr_t)) {
2459 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2460 map_name, name, btf_kind_str(arr_t));
2461 return false;
2462 }
2463 arr_info = btf_array(arr_t);
2464 *res = arr_info->nelems;
2465 return true;
2466 }
2467
get_map_field_long(const char * map_name,const struct btf * btf,const struct btf_member * m,__u64 * res)2468 static bool get_map_field_long(const char *map_name, const struct btf *btf,
2469 const struct btf_member *m, __u64 *res)
2470 {
2471 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2472 const char *name = btf__name_by_offset(btf, m->name_off);
2473
2474 if (btf_is_ptr(t)) {
2475 __u32 res32;
2476 bool ret;
2477
2478 ret = get_map_field_int(map_name, btf, m, &res32);
2479 if (ret)
2480 *res = (__u64)res32;
2481 return ret;
2482 }
2483
2484 if (!btf_is_enum(t) && !btf_is_enum64(t)) {
2485 pr_warn("map '%s': attr '%s': expected ENUM or ENUM64, got %s.\n",
2486 map_name, name, btf_kind_str(t));
2487 return false;
2488 }
2489
2490 if (btf_vlen(t) != 1) {
2491 pr_warn("map '%s': attr '%s': invalid __ulong\n",
2492 map_name, name);
2493 return false;
2494 }
2495
2496 if (btf_is_enum(t)) {
2497 const struct btf_enum *e = btf_enum(t);
2498
2499 *res = e->val;
2500 } else {
2501 const struct btf_enum64 *e = btf_enum64(t);
2502
2503 *res = btf_enum64_value(e);
2504 }
2505 return true;
2506 }
2507
pathname_concat(char * buf,size_t buf_sz,const char * path,const char * name)2508 static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name)
2509 {
2510 int len;
2511
2512 len = snprintf(buf, buf_sz, "%s/%s", path, name);
2513 if (len < 0)
2514 return -EINVAL;
2515 if (len >= buf_sz)
2516 return -ENAMETOOLONG;
2517
2518 return 0;
2519 }
2520
build_map_pin_path(struct bpf_map * map,const char * path)2521 static int build_map_pin_path(struct bpf_map *map, const char *path)
2522 {
2523 char buf[PATH_MAX];
2524 int err;
2525
2526 if (!path)
2527 path = BPF_FS_DEFAULT_PATH;
2528
2529 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
2530 if (err)
2531 return err;
2532
2533 return bpf_map__set_pin_path(map, buf);
2534 }
2535
2536 /* should match definition in bpf_helpers.h */
2537 enum libbpf_pin_type {
2538 LIBBPF_PIN_NONE,
2539 /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
2540 LIBBPF_PIN_BY_NAME,
2541 };
2542
parse_btf_map_def(const char * map_name,struct btf * btf,const struct btf_type * def_t,bool strict,struct btf_map_def * map_def,struct btf_map_def * inner_def)2543 int parse_btf_map_def(const char *map_name, struct btf *btf,
2544 const struct btf_type *def_t, bool strict,
2545 struct btf_map_def *map_def, struct btf_map_def *inner_def)
2546 {
2547 const struct btf_type *t;
2548 const struct btf_member *m;
2549 bool is_inner = inner_def == NULL;
2550 int vlen, i;
2551
2552 vlen = btf_vlen(def_t);
2553 m = btf_members(def_t);
2554 for (i = 0; i < vlen; i++, m++) {
2555 const char *name = btf__name_by_offset(btf, m->name_off);
2556
2557 if (!name) {
2558 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2559 return -EINVAL;
2560 }
2561 if (strcmp(name, "type") == 0) {
2562 if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2563 return -EINVAL;
2564 map_def->parts |= MAP_DEF_MAP_TYPE;
2565 } else if (strcmp(name, "max_entries") == 0) {
2566 if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2567 return -EINVAL;
2568 map_def->parts |= MAP_DEF_MAX_ENTRIES;
2569 } else if (strcmp(name, "map_flags") == 0) {
2570 if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2571 return -EINVAL;
2572 map_def->parts |= MAP_DEF_MAP_FLAGS;
2573 } else if (strcmp(name, "numa_node") == 0) {
2574 if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2575 return -EINVAL;
2576 map_def->parts |= MAP_DEF_NUMA_NODE;
2577 } else if (strcmp(name, "key_size") == 0) {
2578 __u32 sz;
2579
2580 if (!get_map_field_int(map_name, btf, m, &sz))
2581 return -EINVAL;
2582 if (map_def->key_size && map_def->key_size != sz) {
2583 pr_warn("map '%s': conflicting key size %u != %u.\n",
2584 map_name, map_def->key_size, sz);
2585 return -EINVAL;
2586 }
2587 map_def->key_size = sz;
2588 map_def->parts |= MAP_DEF_KEY_SIZE;
2589 } else if (strcmp(name, "key") == 0) {
2590 __s64 sz;
2591
2592 t = btf__type_by_id(btf, m->type);
2593 if (!t) {
2594 pr_warn("map '%s': key type [%d] not found.\n",
2595 map_name, m->type);
2596 return -EINVAL;
2597 }
2598 if (!btf_is_ptr(t)) {
2599 pr_warn("map '%s': key spec is not PTR: %s.\n",
2600 map_name, btf_kind_str(t));
2601 return -EINVAL;
2602 }
2603 sz = btf__resolve_size(btf, t->type);
2604 if (sz < 0) {
2605 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2606 map_name, t->type, (ssize_t)sz);
2607 return sz;
2608 }
2609 if (map_def->key_size && map_def->key_size != sz) {
2610 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2611 map_name, map_def->key_size, (ssize_t)sz);
2612 return -EINVAL;
2613 }
2614 map_def->key_size = sz;
2615 map_def->key_type_id = t->type;
2616 map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2617 } else if (strcmp(name, "value_size") == 0) {
2618 __u32 sz;
2619
2620 if (!get_map_field_int(map_name, btf, m, &sz))
2621 return -EINVAL;
2622 if (map_def->value_size && map_def->value_size != sz) {
2623 pr_warn("map '%s': conflicting value size %u != %u.\n",
2624 map_name, map_def->value_size, sz);
2625 return -EINVAL;
2626 }
2627 map_def->value_size = sz;
2628 map_def->parts |= MAP_DEF_VALUE_SIZE;
2629 } else if (strcmp(name, "value") == 0) {
2630 __s64 sz;
2631
2632 t = btf__type_by_id(btf, m->type);
2633 if (!t) {
2634 pr_warn("map '%s': value type [%d] not found.\n",
2635 map_name, m->type);
2636 return -EINVAL;
2637 }
2638 if (!btf_is_ptr(t)) {
2639 pr_warn("map '%s': value spec is not PTR: %s.\n",
2640 map_name, btf_kind_str(t));
2641 return -EINVAL;
2642 }
2643 sz = btf__resolve_size(btf, t->type);
2644 if (sz < 0) {
2645 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2646 map_name, t->type, (ssize_t)sz);
2647 return sz;
2648 }
2649 if (map_def->value_size && map_def->value_size != sz) {
2650 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2651 map_name, map_def->value_size, (ssize_t)sz);
2652 return -EINVAL;
2653 }
2654 map_def->value_size = sz;
2655 map_def->value_type_id = t->type;
2656 map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2657 }
2658 else if (strcmp(name, "values") == 0) {
2659 bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2660 bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2661 const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2662 char inner_map_name[128];
2663 int err;
2664
2665 if (is_inner) {
2666 pr_warn("map '%s': multi-level inner maps not supported.\n",
2667 map_name);
2668 return -ENOTSUP;
2669 }
2670 if (i != vlen - 1) {
2671 pr_warn("map '%s': '%s' member should be last.\n",
2672 map_name, name);
2673 return -EINVAL;
2674 }
2675 if (!is_map_in_map && !is_prog_array) {
2676 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2677 map_name);
2678 return -ENOTSUP;
2679 }
2680 if (map_def->value_size && map_def->value_size != 4) {
2681 pr_warn("map '%s': conflicting value size %u != 4.\n",
2682 map_name, map_def->value_size);
2683 return -EINVAL;
2684 }
2685 map_def->value_size = 4;
2686 t = btf__type_by_id(btf, m->type);
2687 if (!t) {
2688 pr_warn("map '%s': %s type [%d] not found.\n",
2689 map_name, desc, m->type);
2690 return -EINVAL;
2691 }
2692 if (!btf_is_array(t) || btf_array(t)->nelems) {
2693 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2694 map_name, desc);
2695 return -EINVAL;
2696 }
2697 t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2698 if (!btf_is_ptr(t)) {
2699 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2700 map_name, desc, btf_kind_str(t));
2701 return -EINVAL;
2702 }
2703 t = skip_mods_and_typedefs(btf, t->type, NULL);
2704 if (is_prog_array) {
2705 if (!btf_is_func_proto(t)) {
2706 pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2707 map_name, btf_kind_str(t));
2708 return -EINVAL;
2709 }
2710 continue;
2711 }
2712 if (!btf_is_struct(t)) {
2713 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2714 map_name, btf_kind_str(t));
2715 return -EINVAL;
2716 }
2717
2718 snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2719 err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2720 if (err)
2721 return err;
2722
2723 map_def->parts |= MAP_DEF_INNER_MAP;
2724 } else if (strcmp(name, "pinning") == 0) {
2725 __u32 val;
2726
2727 if (is_inner) {
2728 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2729 return -EINVAL;
2730 }
2731 if (!get_map_field_int(map_name, btf, m, &val))
2732 return -EINVAL;
2733 if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2734 pr_warn("map '%s': invalid pinning value %u.\n",
2735 map_name, val);
2736 return -EINVAL;
2737 }
2738 map_def->pinning = val;
2739 map_def->parts |= MAP_DEF_PINNING;
2740 } else if (strcmp(name, "map_extra") == 0) {
2741 __u64 map_extra;
2742
2743 if (!get_map_field_long(map_name, btf, m, &map_extra))
2744 return -EINVAL;
2745 map_def->map_extra = map_extra;
2746 map_def->parts |= MAP_DEF_MAP_EXTRA;
2747 } else {
2748 if (strict) {
2749 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2750 return -ENOTSUP;
2751 }
2752 pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2753 }
2754 }
2755
2756 if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2757 pr_warn("map '%s': map type isn't specified.\n", map_name);
2758 return -EINVAL;
2759 }
2760
2761 return 0;
2762 }
2763
adjust_ringbuf_sz(size_t sz)2764 static size_t adjust_ringbuf_sz(size_t sz)
2765 {
2766 __u32 page_sz = sysconf(_SC_PAGE_SIZE);
2767 __u32 mul;
2768
2769 /* if user forgot to set any size, make sure they see error */
2770 if (sz == 0)
2771 return 0;
2772 /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be
2773 * a power-of-2 multiple of kernel's page size. If user diligently
2774 * satisified these conditions, pass the size through.
2775 */
2776 if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz))
2777 return sz;
2778
2779 /* Otherwise find closest (page_sz * power_of_2) product bigger than
2780 * user-set size to satisfy both user size request and kernel
2781 * requirements and substitute correct max_entries for map creation.
2782 */
2783 for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) {
2784 if (mul * page_sz > sz)
2785 return mul * page_sz;
2786 }
2787
2788 /* if it's impossible to satisfy the conditions (i.e., user size is
2789 * very close to UINT_MAX but is not a power-of-2 multiple of
2790 * page_size) then just return original size and let kernel reject it
2791 */
2792 return sz;
2793 }
2794
map_is_ringbuf(const struct bpf_map * map)2795 static bool map_is_ringbuf(const struct bpf_map *map)
2796 {
2797 return map->def.type == BPF_MAP_TYPE_RINGBUF ||
2798 map->def.type == BPF_MAP_TYPE_USER_RINGBUF;
2799 }
2800
fill_map_from_def(struct bpf_map * map,const struct btf_map_def * def)2801 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2802 {
2803 map->def.type = def->map_type;
2804 map->def.key_size = def->key_size;
2805 map->def.value_size = def->value_size;
2806 map->def.max_entries = def->max_entries;
2807 map->def.map_flags = def->map_flags;
2808 map->map_extra = def->map_extra;
2809
2810 map->numa_node = def->numa_node;
2811 map->btf_key_type_id = def->key_type_id;
2812 map->btf_value_type_id = def->value_type_id;
2813
2814 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
2815 if (map_is_ringbuf(map))
2816 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
2817
2818 if (def->parts & MAP_DEF_MAP_TYPE)
2819 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2820
2821 if (def->parts & MAP_DEF_KEY_TYPE)
2822 pr_debug("map '%s': found key [%u], sz = %u.\n",
2823 map->name, def->key_type_id, def->key_size);
2824 else if (def->parts & MAP_DEF_KEY_SIZE)
2825 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2826
2827 if (def->parts & MAP_DEF_VALUE_TYPE)
2828 pr_debug("map '%s': found value [%u], sz = %u.\n",
2829 map->name, def->value_type_id, def->value_size);
2830 else if (def->parts & MAP_DEF_VALUE_SIZE)
2831 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2832
2833 if (def->parts & MAP_DEF_MAX_ENTRIES)
2834 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2835 if (def->parts & MAP_DEF_MAP_FLAGS)
2836 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2837 if (def->parts & MAP_DEF_MAP_EXTRA)
2838 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2839 (unsigned long long)def->map_extra);
2840 if (def->parts & MAP_DEF_PINNING)
2841 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2842 if (def->parts & MAP_DEF_NUMA_NODE)
2843 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2844
2845 if (def->parts & MAP_DEF_INNER_MAP)
2846 pr_debug("map '%s': found inner map definition.\n", map->name);
2847 }
2848
btf_var_linkage_str(__u32 linkage)2849 static const char *btf_var_linkage_str(__u32 linkage)
2850 {
2851 switch (linkage) {
2852 case BTF_VAR_STATIC: return "static";
2853 case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2854 case BTF_VAR_GLOBAL_EXTERN: return "extern";
2855 default: return "unknown";
2856 }
2857 }
2858
bpf_object__init_user_btf_map(struct bpf_object * obj,const struct btf_type * sec,int var_idx,int sec_idx,const Elf_Data * data,bool strict,const char * pin_root_path)2859 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2860 const struct btf_type *sec,
2861 int var_idx, int sec_idx,
2862 const Elf_Data *data, bool strict,
2863 const char *pin_root_path)
2864 {
2865 struct btf_map_def map_def = {}, inner_def = {};
2866 const struct btf_type *var, *def;
2867 const struct btf_var_secinfo *vi;
2868 const struct btf_var *var_extra;
2869 const char *map_name;
2870 struct bpf_map *map;
2871 int err;
2872
2873 vi = btf_var_secinfos(sec) + var_idx;
2874 var = btf__type_by_id(obj->btf, vi->type);
2875 var_extra = btf_var(var);
2876 map_name = btf__name_by_offset(obj->btf, var->name_off);
2877
2878 if (map_name == NULL || map_name[0] == '\0') {
2879 pr_warn("map #%d: empty name.\n", var_idx);
2880 return -EINVAL;
2881 }
2882 if ((__u64)vi->offset + vi->size > data->d_size) {
2883 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2884 return -EINVAL;
2885 }
2886 if (!btf_is_var(var)) {
2887 pr_warn("map '%s': unexpected var kind %s.\n",
2888 map_name, btf_kind_str(var));
2889 return -EINVAL;
2890 }
2891 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2892 pr_warn("map '%s': unsupported map linkage %s.\n",
2893 map_name, btf_var_linkage_str(var_extra->linkage));
2894 return -EOPNOTSUPP;
2895 }
2896
2897 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2898 if (!btf_is_struct(def)) {
2899 pr_warn("map '%s': unexpected def kind %s.\n",
2900 map_name, btf_kind_str(var));
2901 return -EINVAL;
2902 }
2903 if (def->size > vi->size) {
2904 pr_warn("map '%s': invalid def size.\n", map_name);
2905 return -EINVAL;
2906 }
2907
2908 map = bpf_object__add_map(obj);
2909 if (IS_ERR(map))
2910 return PTR_ERR(map);
2911 map->name = strdup(map_name);
2912 if (!map->name) {
2913 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2914 return -ENOMEM;
2915 }
2916 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2917 map->def.type = BPF_MAP_TYPE_UNSPEC;
2918 map->sec_idx = sec_idx;
2919 map->sec_offset = vi->offset;
2920 map->btf_var_idx = var_idx;
2921 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2922 map_name, map->sec_idx, map->sec_offset);
2923
2924 err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2925 if (err)
2926 return err;
2927
2928 fill_map_from_def(map, &map_def);
2929
2930 if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2931 err = build_map_pin_path(map, pin_root_path);
2932 if (err) {
2933 pr_warn("map '%s': couldn't build pin path.\n", map->name);
2934 return err;
2935 }
2936 }
2937
2938 if (map_def.parts & MAP_DEF_INNER_MAP) {
2939 map->inner_map = calloc(1, sizeof(*map->inner_map));
2940 if (!map->inner_map)
2941 return -ENOMEM;
2942 map->inner_map->fd = create_placeholder_fd();
2943 if (map->inner_map->fd < 0)
2944 return map->inner_map->fd;
2945 map->inner_map->sec_idx = sec_idx;
2946 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2947 if (!map->inner_map->name)
2948 return -ENOMEM;
2949 sprintf(map->inner_map->name, "%s.inner", map_name);
2950
2951 fill_map_from_def(map->inner_map, &inner_def);
2952 }
2953
2954 err = map_fill_btf_type_info(obj, map);
2955 if (err)
2956 return err;
2957
2958 return 0;
2959 }
2960
init_arena_map_data(struct bpf_object * obj,struct bpf_map * map,const char * sec_name,int sec_idx,void * data,size_t data_sz)2961 static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
2962 const char *sec_name, int sec_idx,
2963 void *data, size_t data_sz)
2964 {
2965 const long page_sz = sysconf(_SC_PAGE_SIZE);
2966 size_t mmap_sz;
2967
2968 mmap_sz = bpf_map_mmap_sz(obj->arena_map);
2969 if (roundup(data_sz, page_sz) > mmap_sz) {
2970 pr_warn("elf: sec '%s': declared ARENA map size (%zu) is too small to hold global __arena variables of size %zu\n",
2971 sec_name, mmap_sz, data_sz);
2972 return -E2BIG;
2973 }
2974
2975 obj->arena_data = malloc(data_sz);
2976 if (!obj->arena_data)
2977 return -ENOMEM;
2978 memcpy(obj->arena_data, data, data_sz);
2979 obj->arena_data_sz = data_sz;
2980
2981 /* make bpf_map__init_value() work for ARENA maps */
2982 map->mmaped = obj->arena_data;
2983
2984 return 0;
2985 }
2986
bpf_object__init_user_btf_maps(struct bpf_object * obj,bool strict,const char * pin_root_path)2987 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2988 const char *pin_root_path)
2989 {
2990 const struct btf_type *sec = NULL;
2991 int nr_types, i, vlen, err;
2992 const struct btf_type *t;
2993 const char *name;
2994 Elf_Data *data;
2995 Elf_Scn *scn;
2996
2997 if (obj->efile.btf_maps_shndx < 0)
2998 return 0;
2999
3000 scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
3001 data = elf_sec_data(obj, scn);
3002 if (!scn || !data) {
3003 pr_warn("elf: failed to get %s map definitions for %s\n",
3004 MAPS_ELF_SEC, obj->path);
3005 return -EINVAL;
3006 }
3007
3008 nr_types = btf__type_cnt(obj->btf);
3009 for (i = 1; i < nr_types; i++) {
3010 t = btf__type_by_id(obj->btf, i);
3011 if (!btf_is_datasec(t))
3012 continue;
3013 name = btf__name_by_offset(obj->btf, t->name_off);
3014 if (strcmp(name, MAPS_ELF_SEC) == 0) {
3015 sec = t;
3016 obj->efile.btf_maps_sec_btf_id = i;
3017 break;
3018 }
3019 }
3020
3021 if (!sec) {
3022 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
3023 return -ENOENT;
3024 }
3025
3026 vlen = btf_vlen(sec);
3027 for (i = 0; i < vlen; i++) {
3028 err = bpf_object__init_user_btf_map(obj, sec, i,
3029 obj->efile.btf_maps_shndx,
3030 data, strict,
3031 pin_root_path);
3032 if (err)
3033 return err;
3034 }
3035
3036 for (i = 0; i < obj->nr_maps; i++) {
3037 struct bpf_map *map = &obj->maps[i];
3038
3039 if (map->def.type != BPF_MAP_TYPE_ARENA)
3040 continue;
3041
3042 if (obj->arena_map) {
3043 pr_warn("map '%s': only single ARENA map is supported (map '%s' is also ARENA)\n",
3044 map->name, obj->arena_map->name);
3045 return -EINVAL;
3046 }
3047 obj->arena_map = map;
3048
3049 if (obj->efile.arena_data) {
3050 err = init_arena_map_data(obj, map, ARENA_SEC, obj->efile.arena_data_shndx,
3051 obj->efile.arena_data->d_buf,
3052 obj->efile.arena_data->d_size);
3053 if (err)
3054 return err;
3055 }
3056 }
3057 if (obj->efile.arena_data && !obj->arena_map) {
3058 pr_warn("elf: sec '%s': to use global __arena variables the ARENA map should be explicitly declared in SEC(\".maps\")\n",
3059 ARENA_SEC);
3060 return -ENOENT;
3061 }
3062
3063 return 0;
3064 }
3065
bpf_object__init_maps(struct bpf_object * obj,const struct bpf_object_open_opts * opts)3066 static int bpf_object__init_maps(struct bpf_object *obj,
3067 const struct bpf_object_open_opts *opts)
3068 {
3069 const char *pin_root_path;
3070 bool strict;
3071 int err = 0;
3072
3073 strict = !OPTS_GET(opts, relaxed_maps, false);
3074 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
3075
3076 err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
3077 err = err ?: bpf_object__init_global_data_maps(obj);
3078 err = err ?: bpf_object__init_kconfig_map(obj);
3079 err = err ?: bpf_object_init_struct_ops(obj);
3080
3081 return err;
3082 }
3083
section_have_execinstr(struct bpf_object * obj,int idx)3084 static bool section_have_execinstr(struct bpf_object *obj, int idx)
3085 {
3086 Elf64_Shdr *sh;
3087
3088 sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
3089 if (!sh)
3090 return false;
3091
3092 return sh->sh_flags & SHF_EXECINSTR;
3093 }
3094
starts_with_qmark(const char * s)3095 static bool starts_with_qmark(const char *s)
3096 {
3097 return s && s[0] == '?';
3098 }
3099
btf_needs_sanitization(struct bpf_object * obj)3100 static bool btf_needs_sanitization(struct bpf_object *obj)
3101 {
3102 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
3103 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
3104 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
3105 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
3106 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
3107 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
3108 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
3109 bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC);
3110
3111 return !has_func || !has_datasec || !has_func_global || !has_float ||
3112 !has_decl_tag || !has_type_tag || !has_enum64 || !has_qmark_datasec;
3113 }
3114
bpf_object__sanitize_btf(struct bpf_object * obj,struct btf * btf)3115 static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
3116 {
3117 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
3118 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
3119 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
3120 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
3121 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
3122 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
3123 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
3124 bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC);
3125 int enum64_placeholder_id = 0;
3126 struct btf_type *t;
3127 int i, j, vlen;
3128
3129 for (i = 1; i < btf__type_cnt(btf); i++) {
3130 t = (struct btf_type *)btf__type_by_id(btf, i);
3131
3132 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
3133 /* replace VAR/DECL_TAG with INT */
3134 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
3135 /*
3136 * using size = 1 is the safest choice, 4 will be too
3137 * big and cause kernel BTF validation failure if
3138 * original variable took less than 4 bytes
3139 */
3140 t->size = 1;
3141 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
3142 } else if (!has_datasec && btf_is_datasec(t)) {
3143 /* replace DATASEC with STRUCT */
3144 const struct btf_var_secinfo *v = btf_var_secinfos(t);
3145 struct btf_member *m = btf_members(t);
3146 struct btf_type *vt;
3147 char *name;
3148
3149 name = (char *)btf__name_by_offset(btf, t->name_off);
3150 while (*name) {
3151 if (*name == '.' || *name == '?')
3152 *name = '_';
3153 name++;
3154 }
3155
3156 vlen = btf_vlen(t);
3157 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
3158 for (j = 0; j < vlen; j++, v++, m++) {
3159 /* order of field assignments is important */
3160 m->offset = v->offset * 8;
3161 m->type = v->type;
3162 /* preserve variable name as member name */
3163 vt = (void *)btf__type_by_id(btf, v->type);
3164 m->name_off = vt->name_off;
3165 }
3166 } else if (!has_qmark_datasec && btf_is_datasec(t) &&
3167 starts_with_qmark(btf__name_by_offset(btf, t->name_off))) {
3168 /* replace '?' prefix with '_' for DATASEC names */
3169 char *name;
3170
3171 name = (char *)btf__name_by_offset(btf, t->name_off);
3172 if (name[0] == '?')
3173 name[0] = '_';
3174 } else if (!has_func && btf_is_func_proto(t)) {
3175 /* replace FUNC_PROTO with ENUM */
3176 vlen = btf_vlen(t);
3177 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
3178 t->size = sizeof(__u32); /* kernel enforced */
3179 } else if (!has_func && btf_is_func(t)) {
3180 /* replace FUNC with TYPEDEF */
3181 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
3182 } else if (!has_func_global && btf_is_func(t)) {
3183 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
3184 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
3185 } else if (!has_float && btf_is_float(t)) {
3186 /* replace FLOAT with an equally-sized empty STRUCT;
3187 * since C compilers do not accept e.g. "float" as a
3188 * valid struct name, make it anonymous
3189 */
3190 t->name_off = 0;
3191 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
3192 } else if (!has_type_tag && btf_is_type_tag(t)) {
3193 /* replace TYPE_TAG with a CONST */
3194 t->name_off = 0;
3195 t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
3196 } else if (!has_enum64 && btf_is_enum(t)) {
3197 /* clear the kflag */
3198 t->info = btf_type_info(btf_kind(t), btf_vlen(t), false);
3199 } else if (!has_enum64 && btf_is_enum64(t)) {
3200 /* replace ENUM64 with a union */
3201 struct btf_member *m;
3202
3203 if (enum64_placeholder_id == 0) {
3204 enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0);
3205 if (enum64_placeholder_id < 0)
3206 return enum64_placeholder_id;
3207
3208 t = (struct btf_type *)btf__type_by_id(btf, i);
3209 }
3210
3211 m = btf_members(t);
3212 vlen = btf_vlen(t);
3213 t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen);
3214 for (j = 0; j < vlen; j++, m++) {
3215 m->type = enum64_placeholder_id;
3216 m->offset = 0;
3217 }
3218 }
3219 }
3220
3221 return 0;
3222 }
3223
libbpf_needs_btf(const struct bpf_object * obj)3224 static bool libbpf_needs_btf(const struct bpf_object *obj)
3225 {
3226 return obj->efile.btf_maps_shndx >= 0 ||
3227 obj->efile.has_st_ops ||
3228 obj->nr_extern > 0;
3229 }
3230
kernel_needs_btf(const struct bpf_object * obj)3231 static bool kernel_needs_btf(const struct bpf_object *obj)
3232 {
3233 return obj->efile.has_st_ops;
3234 }
3235
bpf_object__init_btf(struct bpf_object * obj,Elf_Data * btf_data,Elf_Data * btf_ext_data)3236 static int bpf_object__init_btf(struct bpf_object *obj,
3237 Elf_Data *btf_data,
3238 Elf_Data *btf_ext_data)
3239 {
3240 int err = -ENOENT;
3241
3242 if (btf_data) {
3243 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
3244 err = libbpf_get_error(obj->btf);
3245 if (err) {
3246 obj->btf = NULL;
3247 pr_warn("Error loading ELF section %s: %s.\n", BTF_ELF_SEC, errstr(err));
3248 goto out;
3249 }
3250 /* enforce 8-byte pointers for BPF-targeted BTFs */
3251 btf__set_pointer_size(obj->btf, 8);
3252 }
3253 if (btf_ext_data) {
3254 struct btf_ext_info *ext_segs[3];
3255 int seg_num, sec_num;
3256
3257 if (!obj->btf) {
3258 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
3259 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
3260 goto out;
3261 }
3262 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
3263 err = libbpf_get_error(obj->btf_ext);
3264 if (err) {
3265 pr_warn("Error loading ELF section %s: %s. Ignored and continue.\n",
3266 BTF_EXT_ELF_SEC, errstr(err));
3267 obj->btf_ext = NULL;
3268 goto out;
3269 }
3270
3271 /* setup .BTF.ext to ELF section mapping */
3272 ext_segs[0] = &obj->btf_ext->func_info;
3273 ext_segs[1] = &obj->btf_ext->line_info;
3274 ext_segs[2] = &obj->btf_ext->core_relo_info;
3275 for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) {
3276 struct btf_ext_info *seg = ext_segs[seg_num];
3277 const struct btf_ext_info_sec *sec;
3278 const char *sec_name;
3279 Elf_Scn *scn;
3280
3281 if (seg->sec_cnt == 0)
3282 continue;
3283
3284 seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs));
3285 if (!seg->sec_idxs) {
3286 err = -ENOMEM;
3287 goto out;
3288 }
3289
3290 sec_num = 0;
3291 for_each_btf_ext_sec(seg, sec) {
3292 /* preventively increment index to avoid doing
3293 * this before every continue below
3294 */
3295 sec_num++;
3296
3297 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3298 if (str_is_empty(sec_name))
3299 continue;
3300 scn = elf_sec_by_name(obj, sec_name);
3301 if (!scn)
3302 continue;
3303
3304 seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn);
3305 }
3306 }
3307 }
3308 out:
3309 if (err && libbpf_needs_btf(obj)) {
3310 pr_warn("BTF is required, but is missing or corrupted.\n");
3311 return err;
3312 }
3313 return 0;
3314 }
3315
compare_vsi_off(const void * _a,const void * _b)3316 static int compare_vsi_off(const void *_a, const void *_b)
3317 {
3318 const struct btf_var_secinfo *a = _a;
3319 const struct btf_var_secinfo *b = _b;
3320
3321 return a->offset - b->offset;
3322 }
3323
btf_fixup_datasec(struct bpf_object * obj,struct btf * btf,struct btf_type * t)3324 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
3325 struct btf_type *t)
3326 {
3327 __u32 size = 0, i, vars = btf_vlen(t);
3328 const char *sec_name = btf__name_by_offset(btf, t->name_off);
3329 struct btf_var_secinfo *vsi;
3330 bool fixup_offsets = false;
3331 int err;
3332
3333 if (!sec_name) {
3334 pr_debug("No name found in string section for DATASEC kind.\n");
3335 return -ENOENT;
3336 }
3337
3338 /* Extern-backing datasecs (.ksyms, .kconfig) have their size and
3339 * variable offsets set at the previous step. Further, not every
3340 * extern BTF VAR has corresponding ELF symbol preserved, so we skip
3341 * all fixups altogether for such sections and go straight to sorting
3342 * VARs within their DATASEC.
3343 */
3344 if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0)
3345 goto sort_vars;
3346
3347 /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to
3348 * fix this up. But BPF static linker already fixes this up and fills
3349 * all the sizes and offsets during static linking. So this step has
3350 * to be optional. But the STV_HIDDEN handling is non-optional for any
3351 * non-extern DATASEC, so the variable fixup loop below handles both
3352 * functions at the same time, paying the cost of BTF VAR <-> ELF
3353 * symbol matching just once.
3354 */
3355 if (t->size == 0) {
3356 err = find_elf_sec_sz(obj, sec_name, &size);
3357 if (err || !size) {
3358 pr_debug("sec '%s': failed to determine size from ELF: size %u, err %s\n",
3359 sec_name, size, errstr(err));
3360 return -ENOENT;
3361 }
3362
3363 t->size = size;
3364 fixup_offsets = true;
3365 }
3366
3367 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
3368 const struct btf_type *t_var;
3369 struct btf_var *var;
3370 const char *var_name;
3371 Elf64_Sym *sym;
3372
3373 t_var = btf__type_by_id(btf, vsi->type);
3374 if (!t_var || !btf_is_var(t_var)) {
3375 pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name);
3376 return -EINVAL;
3377 }
3378
3379 var = btf_var(t_var);
3380 if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN)
3381 continue;
3382
3383 var_name = btf__name_by_offset(btf, t_var->name_off);
3384 if (!var_name) {
3385 pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n",
3386 sec_name, i);
3387 return -ENOENT;
3388 }
3389
3390 sym = find_elf_var_sym(obj, var_name);
3391 if (IS_ERR(sym)) {
3392 pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n",
3393 sec_name, var_name);
3394 return -ENOENT;
3395 }
3396
3397 if (fixup_offsets)
3398 vsi->offset = sym->st_value;
3399
3400 /* if variable is a global/weak symbol, but has restricted
3401 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR
3402 * as static. This follows similar logic for functions (BPF
3403 * subprogs) and influences libbpf's further decisions about
3404 * whether to make global data BPF array maps as
3405 * BPF_F_MMAPABLE.
3406 */
3407 if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
3408 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)
3409 var->linkage = BTF_VAR_STATIC;
3410 }
3411
3412 sort_vars:
3413 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
3414 return 0;
3415 }
3416
bpf_object_fixup_btf(struct bpf_object * obj)3417 static int bpf_object_fixup_btf(struct bpf_object *obj)
3418 {
3419 int i, n, err = 0;
3420
3421 if (!obj->btf)
3422 return 0;
3423
3424 n = btf__type_cnt(obj->btf);
3425 for (i = 1; i < n; i++) {
3426 struct btf_type *t = btf_type_by_id(obj->btf, i);
3427
3428 /* Loader needs to fix up some of the things compiler
3429 * couldn't get its hands on while emitting BTF. This
3430 * is section size and global variable offset. We use
3431 * the info from the ELF itself for this purpose.
3432 */
3433 if (btf_is_datasec(t)) {
3434 err = btf_fixup_datasec(obj, obj->btf, t);
3435 if (err)
3436 return err;
3437 }
3438 }
3439
3440 return 0;
3441 }
3442
prog_needs_vmlinux_btf(struct bpf_program * prog)3443 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
3444 {
3445 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
3446 prog->type == BPF_PROG_TYPE_LSM)
3447 return true;
3448
3449 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
3450 * also need vmlinux BTF
3451 */
3452 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
3453 return true;
3454
3455 return false;
3456 }
3457
map_needs_vmlinux_btf(struct bpf_map * map)3458 static bool map_needs_vmlinux_btf(struct bpf_map *map)
3459 {
3460 return bpf_map__is_struct_ops(map);
3461 }
3462
obj_needs_vmlinux_btf(const struct bpf_object * obj)3463 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
3464 {
3465 struct bpf_program *prog;
3466 struct bpf_map *map;
3467 int i;
3468
3469 /* CO-RE relocations need kernel BTF, only when btf_custom_path
3470 * is not specified
3471 */
3472 if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
3473 return true;
3474
3475 /* Support for typed ksyms needs kernel BTF */
3476 for (i = 0; i < obj->nr_extern; i++) {
3477 const struct extern_desc *ext;
3478
3479 ext = &obj->externs[i];
3480 if (ext->type == EXT_KSYM && ext->ksym.type_id)
3481 return true;
3482 }
3483
3484 bpf_object__for_each_program(prog, obj) {
3485 if (!prog->autoload)
3486 continue;
3487 if (prog_needs_vmlinux_btf(prog))
3488 return true;
3489 }
3490
3491 bpf_object__for_each_map(map, obj) {
3492 if (map_needs_vmlinux_btf(map))
3493 return true;
3494 }
3495
3496 return false;
3497 }
3498
bpf_object__load_vmlinux_btf(struct bpf_object * obj,bool force)3499 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
3500 {
3501 int err;
3502
3503 /* btf_vmlinux could be loaded earlier */
3504 if (obj->btf_vmlinux || obj->gen_loader)
3505 return 0;
3506
3507 if (!force && !obj_needs_vmlinux_btf(obj))
3508 return 0;
3509
3510 obj->btf_vmlinux = btf__load_vmlinux_btf();
3511 err = libbpf_get_error(obj->btf_vmlinux);
3512 if (err) {
3513 pr_warn("Error loading vmlinux BTF: %s\n", errstr(err));
3514 obj->btf_vmlinux = NULL;
3515 return err;
3516 }
3517 return 0;
3518 }
3519
bpf_object__sanitize_and_load_btf(struct bpf_object * obj)3520 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
3521 {
3522 struct btf *kern_btf = obj->btf;
3523 bool btf_mandatory, sanitize;
3524 int i, err = 0;
3525
3526 if (!obj->btf)
3527 return 0;
3528
3529 if (!kernel_supports(obj, FEAT_BTF)) {
3530 if (kernel_needs_btf(obj)) {
3531 err = -EOPNOTSUPP;
3532 goto report;
3533 }
3534 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
3535 return 0;
3536 }
3537
3538 /* Even though some subprogs are global/weak, user might prefer more
3539 * permissive BPF verification process that BPF verifier performs for
3540 * static functions, taking into account more context from the caller
3541 * functions. In such case, they need to mark such subprogs with
3542 * __attribute__((visibility("hidden"))) and libbpf will adjust
3543 * corresponding FUNC BTF type to be marked as static and trigger more
3544 * involved BPF verification process.
3545 */
3546 for (i = 0; i < obj->nr_programs; i++) {
3547 struct bpf_program *prog = &obj->programs[i];
3548 struct btf_type *t;
3549 const char *name;
3550 int j, n;
3551
3552 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
3553 continue;
3554
3555 n = btf__type_cnt(obj->btf);
3556 for (j = 1; j < n; j++) {
3557 t = btf_type_by_id(obj->btf, j);
3558 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
3559 continue;
3560
3561 name = btf__str_by_offset(obj->btf, t->name_off);
3562 if (strcmp(name, prog->name) != 0)
3563 continue;
3564
3565 t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
3566 break;
3567 }
3568 }
3569
3570 sanitize = btf_needs_sanitization(obj);
3571 if (sanitize) {
3572 const void *raw_data;
3573 __u32 sz;
3574
3575 /* clone BTF to sanitize a copy and leave the original intact */
3576 raw_data = btf__raw_data(obj->btf, &sz);
3577 kern_btf = btf__new(raw_data, sz);
3578 err = libbpf_get_error(kern_btf);
3579 if (err)
3580 return err;
3581
3582 /* enforce 8-byte pointers for BPF-targeted BTFs */
3583 btf__set_pointer_size(obj->btf, 8);
3584 err = bpf_object__sanitize_btf(obj, kern_btf);
3585 if (err)
3586 return err;
3587 }
3588
3589 if (obj->gen_loader) {
3590 __u32 raw_size = 0;
3591 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3592
3593 if (!raw_data)
3594 return -ENOMEM;
3595 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3596 /* Pretend to have valid FD to pass various fd >= 0 checks.
3597 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3598 */
3599 btf__set_fd(kern_btf, 0);
3600 } else {
3601 /* currently BPF_BTF_LOAD only supports log_level 1 */
3602 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3603 obj->log_level ? 1 : 0, obj->token_fd);
3604 }
3605 if (sanitize) {
3606 if (!err) {
3607 /* move fd to libbpf's BTF */
3608 btf__set_fd(obj->btf, btf__fd(kern_btf));
3609 btf__set_fd(kern_btf, -1);
3610 }
3611 btf__free(kern_btf);
3612 }
3613 report:
3614 if (err) {
3615 btf_mandatory = kernel_needs_btf(obj);
3616 if (btf_mandatory) {
3617 pr_warn("Error loading .BTF into kernel: %s. BTF is mandatory, can't proceed.\n",
3618 errstr(err));
3619 } else {
3620 pr_info("Error loading .BTF into kernel: %s. BTF is optional, ignoring.\n",
3621 errstr(err));
3622 err = 0;
3623 }
3624 }
3625 return err;
3626 }
3627
elf_sym_str(const struct bpf_object * obj,size_t off)3628 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3629 {
3630 const char *name;
3631
3632 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3633 if (!name) {
3634 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3635 off, obj->path, elf_errmsg(-1));
3636 return NULL;
3637 }
3638
3639 return name;
3640 }
3641
elf_sec_str(const struct bpf_object * obj,size_t off)3642 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3643 {
3644 const char *name;
3645
3646 name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3647 if (!name) {
3648 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3649 off, obj->path, elf_errmsg(-1));
3650 return NULL;
3651 }
3652
3653 return name;
3654 }
3655
elf_sec_by_idx(const struct bpf_object * obj,size_t idx)3656 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3657 {
3658 Elf_Scn *scn;
3659
3660 scn = elf_getscn(obj->efile.elf, idx);
3661 if (!scn) {
3662 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3663 idx, obj->path, elf_errmsg(-1));
3664 return NULL;
3665 }
3666 return scn;
3667 }
3668
elf_sec_by_name(const struct bpf_object * obj,const char * name)3669 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3670 {
3671 Elf_Scn *scn = NULL;
3672 Elf *elf = obj->efile.elf;
3673 const char *sec_name;
3674
3675 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3676 sec_name = elf_sec_name(obj, scn);
3677 if (!sec_name)
3678 return NULL;
3679
3680 if (strcmp(sec_name, name) != 0)
3681 continue;
3682
3683 return scn;
3684 }
3685 return NULL;
3686 }
3687
elf_sec_hdr(const struct bpf_object * obj,Elf_Scn * scn)3688 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3689 {
3690 Elf64_Shdr *shdr;
3691
3692 if (!scn)
3693 return NULL;
3694
3695 shdr = elf64_getshdr(scn);
3696 if (!shdr) {
3697 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3698 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3699 return NULL;
3700 }
3701
3702 return shdr;
3703 }
3704
elf_sec_name(const struct bpf_object * obj,Elf_Scn * scn)3705 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3706 {
3707 const char *name;
3708 Elf64_Shdr *sh;
3709
3710 if (!scn)
3711 return NULL;
3712
3713 sh = elf_sec_hdr(obj, scn);
3714 if (!sh)
3715 return NULL;
3716
3717 name = elf_sec_str(obj, sh->sh_name);
3718 if (!name) {
3719 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3720 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3721 return NULL;
3722 }
3723
3724 return name;
3725 }
3726
elf_sec_data(const struct bpf_object * obj,Elf_Scn * scn)3727 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3728 {
3729 Elf_Data *data;
3730
3731 if (!scn)
3732 return NULL;
3733
3734 data = elf_getdata(scn, 0);
3735 if (!data) {
3736 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3737 elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3738 obj->path, elf_errmsg(-1));
3739 return NULL;
3740 }
3741
3742 return data;
3743 }
3744
elf_sym_by_idx(const struct bpf_object * obj,size_t idx)3745 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3746 {
3747 if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3748 return NULL;
3749
3750 return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3751 }
3752
elf_rel_by_idx(Elf_Data * data,size_t idx)3753 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3754 {
3755 if (idx >= data->d_size / sizeof(Elf64_Rel))
3756 return NULL;
3757
3758 return (Elf64_Rel *)data->d_buf + idx;
3759 }
3760
is_sec_name_dwarf(const char * name)3761 static bool is_sec_name_dwarf(const char *name)
3762 {
3763 /* approximation, but the actual list is too long */
3764 return str_has_pfx(name, ".debug_");
3765 }
3766
ignore_elf_section(Elf64_Shdr * hdr,const char * name)3767 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3768 {
3769 /* no special handling of .strtab */
3770 if (hdr->sh_type == SHT_STRTAB)
3771 return true;
3772
3773 /* ignore .llvm_addrsig section as well */
3774 if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3775 return true;
3776
3777 /* no subprograms will lead to an empty .text section, ignore it */
3778 if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3779 strcmp(name, ".text") == 0)
3780 return true;
3781
3782 /* DWARF sections */
3783 if (is_sec_name_dwarf(name))
3784 return true;
3785
3786 if (str_has_pfx(name, ".rel")) {
3787 name += sizeof(".rel") - 1;
3788 /* DWARF section relocations */
3789 if (is_sec_name_dwarf(name))
3790 return true;
3791
3792 /* .BTF and .BTF.ext don't need relocations */
3793 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3794 strcmp(name, BTF_EXT_ELF_SEC) == 0)
3795 return true;
3796 }
3797
3798 return false;
3799 }
3800
cmp_progs(const void * _a,const void * _b)3801 static int cmp_progs(const void *_a, const void *_b)
3802 {
3803 const struct bpf_program *a = _a;
3804 const struct bpf_program *b = _b;
3805
3806 if (a->sec_idx != b->sec_idx)
3807 return a->sec_idx < b->sec_idx ? -1 : 1;
3808
3809 /* sec_insn_off can't be the same within the section */
3810 return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3811 }
3812
bpf_object__elf_collect(struct bpf_object * obj)3813 static int bpf_object__elf_collect(struct bpf_object *obj)
3814 {
3815 struct elf_sec_desc *sec_desc;
3816 Elf *elf = obj->efile.elf;
3817 Elf_Data *btf_ext_data = NULL;
3818 Elf_Data *btf_data = NULL;
3819 int idx = 0, err = 0;
3820 const char *name;
3821 Elf_Data *data;
3822 Elf_Scn *scn;
3823 Elf64_Shdr *sh;
3824
3825 /* ELF section indices are 0-based, but sec #0 is special "invalid"
3826 * section. Since section count retrieved by elf_getshdrnum() does
3827 * include sec #0, it is already the necessary size of an array to keep
3828 * all the sections.
3829 */
3830 if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) {
3831 pr_warn("elf: failed to get the number of sections for %s: %s\n",
3832 obj->path, elf_errmsg(-1));
3833 return -LIBBPF_ERRNO__FORMAT;
3834 }
3835 obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3836 if (!obj->efile.secs)
3837 return -ENOMEM;
3838
3839 /* a bunch of ELF parsing functionality depends on processing symbols,
3840 * so do the first pass and find the symbol table
3841 */
3842 scn = NULL;
3843 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3844 sh = elf_sec_hdr(obj, scn);
3845 if (!sh)
3846 return -LIBBPF_ERRNO__FORMAT;
3847
3848 if (sh->sh_type == SHT_SYMTAB) {
3849 if (obj->efile.symbols) {
3850 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3851 return -LIBBPF_ERRNO__FORMAT;
3852 }
3853
3854 data = elf_sec_data(obj, scn);
3855 if (!data)
3856 return -LIBBPF_ERRNO__FORMAT;
3857
3858 idx = elf_ndxscn(scn);
3859
3860 obj->efile.symbols = data;
3861 obj->efile.symbols_shndx = idx;
3862 obj->efile.strtabidx = sh->sh_link;
3863 }
3864 }
3865
3866 if (!obj->efile.symbols) {
3867 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3868 obj->path);
3869 return -ENOENT;
3870 }
3871
3872 scn = NULL;
3873 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3874 idx = elf_ndxscn(scn);
3875 sec_desc = &obj->efile.secs[idx];
3876
3877 sh = elf_sec_hdr(obj, scn);
3878 if (!sh)
3879 return -LIBBPF_ERRNO__FORMAT;
3880
3881 name = elf_sec_str(obj, sh->sh_name);
3882 if (!name)
3883 return -LIBBPF_ERRNO__FORMAT;
3884
3885 if (ignore_elf_section(sh, name))
3886 continue;
3887
3888 data = elf_sec_data(obj, scn);
3889 if (!data)
3890 return -LIBBPF_ERRNO__FORMAT;
3891
3892 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3893 idx, name, (unsigned long)data->d_size,
3894 (int)sh->sh_link, (unsigned long)sh->sh_flags,
3895 (int)sh->sh_type);
3896
3897 if (strcmp(name, "license") == 0) {
3898 err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3899 if (err)
3900 return err;
3901 } else if (strcmp(name, "version") == 0) {
3902 err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3903 if (err)
3904 return err;
3905 } else if (strcmp(name, "maps") == 0) {
3906 pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n");
3907 return -ENOTSUP;
3908 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3909 obj->efile.btf_maps_shndx = idx;
3910 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3911 if (sh->sh_type != SHT_PROGBITS)
3912 return -LIBBPF_ERRNO__FORMAT;
3913 btf_data = data;
3914 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3915 if (sh->sh_type != SHT_PROGBITS)
3916 return -LIBBPF_ERRNO__FORMAT;
3917 btf_ext_data = data;
3918 } else if (sh->sh_type == SHT_SYMTAB) {
3919 /* already processed during the first pass above */
3920 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3921 if (sh->sh_flags & SHF_EXECINSTR) {
3922 if (strcmp(name, ".text") == 0)
3923 obj->efile.text_shndx = idx;
3924 err = bpf_object__add_programs(obj, data, name, idx);
3925 if (err)
3926 return err;
3927 } else if (strcmp(name, DATA_SEC) == 0 ||
3928 str_has_pfx(name, DATA_SEC ".")) {
3929 sec_desc->sec_type = SEC_DATA;
3930 sec_desc->shdr = sh;
3931 sec_desc->data = data;
3932 } else if (strcmp(name, RODATA_SEC) == 0 ||
3933 str_has_pfx(name, RODATA_SEC ".")) {
3934 sec_desc->sec_type = SEC_RODATA;
3935 sec_desc->shdr = sh;
3936 sec_desc->data = data;
3937 } else if (strcmp(name, STRUCT_OPS_SEC) == 0 ||
3938 strcmp(name, STRUCT_OPS_LINK_SEC) == 0 ||
3939 strcmp(name, "?" STRUCT_OPS_SEC) == 0 ||
3940 strcmp(name, "?" STRUCT_OPS_LINK_SEC) == 0) {
3941 sec_desc->sec_type = SEC_ST_OPS;
3942 sec_desc->shdr = sh;
3943 sec_desc->data = data;
3944 obj->efile.has_st_ops = true;
3945 } else if (strcmp(name, ARENA_SEC) == 0) {
3946 obj->efile.arena_data = data;
3947 obj->efile.arena_data_shndx = idx;
3948 } else {
3949 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3950 idx, name);
3951 }
3952 } else if (sh->sh_type == SHT_REL) {
3953 int targ_sec_idx = sh->sh_info; /* points to other section */
3954
3955 if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3956 targ_sec_idx >= obj->efile.sec_cnt)
3957 return -LIBBPF_ERRNO__FORMAT;
3958
3959 /* Only do relo for section with exec instructions */
3960 if (!section_have_execinstr(obj, targ_sec_idx) &&
3961 strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3962 strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) &&
3963 strcmp(name, ".rel?" STRUCT_OPS_SEC) &&
3964 strcmp(name, ".rel?" STRUCT_OPS_LINK_SEC) &&
3965 strcmp(name, ".rel" MAPS_ELF_SEC)) {
3966 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3967 idx, name, targ_sec_idx,
3968 elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3969 continue;
3970 }
3971
3972 sec_desc->sec_type = SEC_RELO;
3973 sec_desc->shdr = sh;
3974 sec_desc->data = data;
3975 } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 ||
3976 str_has_pfx(name, BSS_SEC "."))) {
3977 sec_desc->sec_type = SEC_BSS;
3978 sec_desc->shdr = sh;
3979 sec_desc->data = data;
3980 } else {
3981 pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3982 (size_t)sh->sh_size);
3983 }
3984 }
3985
3986 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3987 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3988 return -LIBBPF_ERRNO__FORMAT;
3989 }
3990
3991 /* change BPF program insns to native endianness for introspection */
3992 if (!is_native_endianness(obj))
3993 bpf_object_bswap_progs(obj);
3994
3995 /* sort BPF programs by section name and in-section instruction offset
3996 * for faster search
3997 */
3998 if (obj->nr_programs)
3999 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
4000
4001 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
4002 }
4003
sym_is_extern(const Elf64_Sym * sym)4004 static bool sym_is_extern(const Elf64_Sym *sym)
4005 {
4006 int bind = ELF64_ST_BIND(sym->st_info);
4007 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
4008 return sym->st_shndx == SHN_UNDEF &&
4009 (bind == STB_GLOBAL || bind == STB_WEAK) &&
4010 ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
4011 }
4012
sym_is_subprog(const Elf64_Sym * sym,int text_shndx)4013 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
4014 {
4015 int bind = ELF64_ST_BIND(sym->st_info);
4016 int type = ELF64_ST_TYPE(sym->st_info);
4017
4018 /* in .text section */
4019 if (sym->st_shndx != text_shndx)
4020 return false;
4021
4022 /* local function */
4023 if (bind == STB_LOCAL && type == STT_SECTION)
4024 return true;
4025
4026 /* global function */
4027 return (bind == STB_GLOBAL || bind == STB_WEAK) && type == STT_FUNC;
4028 }
4029
find_extern_btf_id(const struct btf * btf,const char * ext_name)4030 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
4031 {
4032 const struct btf_type *t;
4033 const char *tname;
4034 int i, n;
4035
4036 if (!btf)
4037 return -ESRCH;
4038
4039 n = btf__type_cnt(btf);
4040 for (i = 1; i < n; i++) {
4041 t = btf__type_by_id(btf, i);
4042
4043 if (!btf_is_var(t) && !btf_is_func(t))
4044 continue;
4045
4046 tname = btf__name_by_offset(btf, t->name_off);
4047 if (strcmp(tname, ext_name))
4048 continue;
4049
4050 if (btf_is_var(t) &&
4051 btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
4052 return -EINVAL;
4053
4054 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
4055 return -EINVAL;
4056
4057 return i;
4058 }
4059
4060 return -ENOENT;
4061 }
4062
find_extern_sec_btf_id(struct btf * btf,int ext_btf_id)4063 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
4064 const struct btf_var_secinfo *vs;
4065 const struct btf_type *t;
4066 int i, j, n;
4067
4068 if (!btf)
4069 return -ESRCH;
4070
4071 n = btf__type_cnt(btf);
4072 for (i = 1; i < n; i++) {
4073 t = btf__type_by_id(btf, i);
4074
4075 if (!btf_is_datasec(t))
4076 continue;
4077
4078 vs = btf_var_secinfos(t);
4079 for (j = 0; j < btf_vlen(t); j++, vs++) {
4080 if (vs->type == ext_btf_id)
4081 return i;
4082 }
4083 }
4084
4085 return -ENOENT;
4086 }
4087
find_kcfg_type(const struct btf * btf,int id,bool * is_signed)4088 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
4089 bool *is_signed)
4090 {
4091 const struct btf_type *t;
4092 const char *name;
4093
4094 t = skip_mods_and_typedefs(btf, id, NULL);
4095 name = btf__name_by_offset(btf, t->name_off);
4096
4097 if (is_signed)
4098 *is_signed = false;
4099 switch (btf_kind(t)) {
4100 case BTF_KIND_INT: {
4101 int enc = btf_int_encoding(t);
4102
4103 if (enc & BTF_INT_BOOL)
4104 return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
4105 if (is_signed)
4106 *is_signed = enc & BTF_INT_SIGNED;
4107 if (t->size == 1)
4108 return KCFG_CHAR;
4109 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
4110 return KCFG_UNKNOWN;
4111 return KCFG_INT;
4112 }
4113 case BTF_KIND_ENUM:
4114 if (t->size != 4)
4115 return KCFG_UNKNOWN;
4116 if (strcmp(name, "libbpf_tristate"))
4117 return KCFG_UNKNOWN;
4118 return KCFG_TRISTATE;
4119 case BTF_KIND_ENUM64:
4120 if (strcmp(name, "libbpf_tristate"))
4121 return KCFG_UNKNOWN;
4122 return KCFG_TRISTATE;
4123 case BTF_KIND_ARRAY:
4124 if (btf_array(t)->nelems == 0)
4125 return KCFG_UNKNOWN;
4126 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
4127 return KCFG_UNKNOWN;
4128 return KCFG_CHAR_ARR;
4129 default:
4130 return KCFG_UNKNOWN;
4131 }
4132 }
4133
cmp_externs(const void * _a,const void * _b)4134 static int cmp_externs(const void *_a, const void *_b)
4135 {
4136 const struct extern_desc *a = _a;
4137 const struct extern_desc *b = _b;
4138
4139 if (a->type != b->type)
4140 return a->type < b->type ? -1 : 1;
4141
4142 if (a->type == EXT_KCFG) {
4143 /* descending order by alignment requirements */
4144 if (a->kcfg.align != b->kcfg.align)
4145 return a->kcfg.align > b->kcfg.align ? -1 : 1;
4146 /* ascending order by size, within same alignment class */
4147 if (a->kcfg.sz != b->kcfg.sz)
4148 return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
4149 }
4150
4151 /* resolve ties by name */
4152 return strcmp(a->name, b->name);
4153 }
4154
find_int_btf_id(const struct btf * btf)4155 static int find_int_btf_id(const struct btf *btf)
4156 {
4157 const struct btf_type *t;
4158 int i, n;
4159
4160 n = btf__type_cnt(btf);
4161 for (i = 1; i < n; i++) {
4162 t = btf__type_by_id(btf, i);
4163
4164 if (btf_is_int(t) && btf_int_bits(t) == 32)
4165 return i;
4166 }
4167
4168 return 0;
4169 }
4170
add_dummy_ksym_var(struct btf * btf)4171 static int add_dummy_ksym_var(struct btf *btf)
4172 {
4173 int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
4174 const struct btf_var_secinfo *vs;
4175 const struct btf_type *sec;
4176
4177 if (!btf)
4178 return 0;
4179
4180 sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
4181 BTF_KIND_DATASEC);
4182 if (sec_btf_id < 0)
4183 return 0;
4184
4185 sec = btf__type_by_id(btf, sec_btf_id);
4186 vs = btf_var_secinfos(sec);
4187 for (i = 0; i < btf_vlen(sec); i++, vs++) {
4188 const struct btf_type *vt;
4189
4190 vt = btf__type_by_id(btf, vs->type);
4191 if (btf_is_func(vt))
4192 break;
4193 }
4194
4195 /* No func in ksyms sec. No need to add dummy var. */
4196 if (i == btf_vlen(sec))
4197 return 0;
4198
4199 int_btf_id = find_int_btf_id(btf);
4200 dummy_var_btf_id = btf__add_var(btf,
4201 "dummy_ksym",
4202 BTF_VAR_GLOBAL_ALLOCATED,
4203 int_btf_id);
4204 if (dummy_var_btf_id < 0)
4205 pr_warn("cannot create a dummy_ksym var\n");
4206
4207 return dummy_var_btf_id;
4208 }
4209
bpf_object__collect_externs(struct bpf_object * obj)4210 static int bpf_object__collect_externs(struct bpf_object *obj)
4211 {
4212 struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
4213 const struct btf_type *t;
4214 struct extern_desc *ext;
4215 int i, n, off, dummy_var_btf_id;
4216 const char *ext_name, *sec_name;
4217 size_t ext_essent_len;
4218 Elf_Scn *scn;
4219 Elf64_Shdr *sh;
4220
4221 if (!obj->efile.symbols)
4222 return 0;
4223
4224 scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
4225 sh = elf_sec_hdr(obj, scn);
4226 if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
4227 return -LIBBPF_ERRNO__FORMAT;
4228
4229 dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
4230 if (dummy_var_btf_id < 0)
4231 return dummy_var_btf_id;
4232
4233 n = sh->sh_size / sh->sh_entsize;
4234 pr_debug("looking for externs among %d symbols...\n", n);
4235
4236 for (i = 0; i < n; i++) {
4237 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
4238
4239 if (!sym)
4240 return -LIBBPF_ERRNO__FORMAT;
4241 if (!sym_is_extern(sym))
4242 continue;
4243 ext_name = elf_sym_str(obj, sym->st_name);
4244 if (!ext_name || !ext_name[0])
4245 continue;
4246
4247 ext = obj->externs;
4248 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
4249 if (!ext)
4250 return -ENOMEM;
4251 obj->externs = ext;
4252 ext = &ext[obj->nr_extern];
4253 memset(ext, 0, sizeof(*ext));
4254 obj->nr_extern++;
4255
4256 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
4257 if (ext->btf_id <= 0) {
4258 pr_warn("failed to find BTF for extern '%s': %d\n",
4259 ext_name, ext->btf_id);
4260 return ext->btf_id;
4261 }
4262 t = btf__type_by_id(obj->btf, ext->btf_id);
4263 ext->name = btf__name_by_offset(obj->btf, t->name_off);
4264 ext->sym_idx = i;
4265 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
4266
4267 ext_essent_len = bpf_core_essential_name_len(ext->name);
4268 ext->essent_name = NULL;
4269 if (ext_essent_len != strlen(ext->name)) {
4270 ext->essent_name = strndup(ext->name, ext_essent_len);
4271 if (!ext->essent_name)
4272 return -ENOMEM;
4273 }
4274
4275 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
4276 if (ext->sec_btf_id <= 0) {
4277 pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
4278 ext_name, ext->btf_id, ext->sec_btf_id);
4279 return ext->sec_btf_id;
4280 }
4281 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
4282 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
4283
4284 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
4285 if (btf_is_func(t)) {
4286 pr_warn("extern function %s is unsupported under %s section\n",
4287 ext->name, KCONFIG_SEC);
4288 return -ENOTSUP;
4289 }
4290 kcfg_sec = sec;
4291 ext->type = EXT_KCFG;
4292 ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
4293 if (ext->kcfg.sz <= 0) {
4294 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
4295 ext_name, ext->kcfg.sz);
4296 return ext->kcfg.sz;
4297 }
4298 ext->kcfg.align = btf__align_of(obj->btf, t->type);
4299 if (ext->kcfg.align <= 0) {
4300 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
4301 ext_name, ext->kcfg.align);
4302 return -EINVAL;
4303 }
4304 ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
4305 &ext->kcfg.is_signed);
4306 if (ext->kcfg.type == KCFG_UNKNOWN) {
4307 pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name);
4308 return -ENOTSUP;
4309 }
4310 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
4311 ksym_sec = sec;
4312 ext->type = EXT_KSYM;
4313 skip_mods_and_typedefs(obj->btf, t->type,
4314 &ext->ksym.type_id);
4315 } else {
4316 pr_warn("unrecognized extern section '%s'\n", sec_name);
4317 return -ENOTSUP;
4318 }
4319 }
4320 pr_debug("collected %d externs total\n", obj->nr_extern);
4321
4322 if (!obj->nr_extern)
4323 return 0;
4324
4325 /* sort externs by type, for kcfg ones also by (align, size, name) */
4326 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
4327
4328 /* for .ksyms section, we need to turn all externs into allocated
4329 * variables in BTF to pass kernel verification; we do this by
4330 * pretending that each extern is a 8-byte variable
4331 */
4332 if (ksym_sec) {
4333 /* find existing 4-byte integer type in BTF to use for fake
4334 * extern variables in DATASEC
4335 */
4336 int int_btf_id = find_int_btf_id(obj->btf);
4337 /* For extern function, a dummy_var added earlier
4338 * will be used to replace the vs->type and
4339 * its name string will be used to refill
4340 * the missing param's name.
4341 */
4342 const struct btf_type *dummy_var;
4343
4344 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
4345 for (i = 0; i < obj->nr_extern; i++) {
4346 ext = &obj->externs[i];
4347 if (ext->type != EXT_KSYM)
4348 continue;
4349 pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
4350 i, ext->sym_idx, ext->name);
4351 }
4352
4353 sec = ksym_sec;
4354 n = btf_vlen(sec);
4355 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
4356 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4357 struct btf_type *vt;
4358
4359 vt = (void *)btf__type_by_id(obj->btf, vs->type);
4360 ext_name = btf__name_by_offset(obj->btf, vt->name_off);
4361 ext = find_extern_by_name(obj, ext_name);
4362 if (!ext) {
4363 pr_warn("failed to find extern definition for BTF %s '%s'\n",
4364 btf_kind_str(vt), ext_name);
4365 return -ESRCH;
4366 }
4367 if (btf_is_func(vt)) {
4368 const struct btf_type *func_proto;
4369 struct btf_param *param;
4370 int j;
4371
4372 func_proto = btf__type_by_id(obj->btf,
4373 vt->type);
4374 param = btf_params(func_proto);
4375 /* Reuse the dummy_var string if the
4376 * func proto does not have param name.
4377 */
4378 for (j = 0; j < btf_vlen(func_proto); j++)
4379 if (param[j].type && !param[j].name_off)
4380 param[j].name_off =
4381 dummy_var->name_off;
4382 vs->type = dummy_var_btf_id;
4383 vt->info &= ~0xffff;
4384 vt->info |= BTF_FUNC_GLOBAL;
4385 } else {
4386 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4387 vt->type = int_btf_id;
4388 }
4389 vs->offset = off;
4390 vs->size = sizeof(int);
4391 }
4392 sec->size = off;
4393 }
4394
4395 if (kcfg_sec) {
4396 sec = kcfg_sec;
4397 /* for kcfg externs calculate their offsets within a .kconfig map */
4398 off = 0;
4399 for (i = 0; i < obj->nr_extern; i++) {
4400 ext = &obj->externs[i];
4401 if (ext->type != EXT_KCFG)
4402 continue;
4403
4404 ext->kcfg.data_off = roundup(off, ext->kcfg.align);
4405 off = ext->kcfg.data_off + ext->kcfg.sz;
4406 pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
4407 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
4408 }
4409 sec->size = off;
4410 n = btf_vlen(sec);
4411 for (i = 0; i < n; i++) {
4412 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4413
4414 t = btf__type_by_id(obj->btf, vs->type);
4415 ext_name = btf__name_by_offset(obj->btf, t->name_off);
4416 ext = find_extern_by_name(obj, ext_name);
4417 if (!ext) {
4418 pr_warn("failed to find extern definition for BTF var '%s'\n",
4419 ext_name);
4420 return -ESRCH;
4421 }
4422 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4423 vs->offset = ext->kcfg.data_off;
4424 }
4425 }
4426 return 0;
4427 }
4428
prog_is_subprog(const struct bpf_object * obj,const struct bpf_program * prog)4429 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog)
4430 {
4431 return prog->sec_idx == obj->efile.text_shndx;
4432 }
4433
4434 struct bpf_program *
bpf_object__find_program_by_name(const struct bpf_object * obj,const char * name)4435 bpf_object__find_program_by_name(const struct bpf_object *obj,
4436 const char *name)
4437 {
4438 struct bpf_program *prog;
4439
4440 bpf_object__for_each_program(prog, obj) {
4441 if (prog_is_subprog(obj, prog))
4442 continue;
4443 if (!strcmp(prog->name, name))
4444 return prog;
4445 }
4446 return errno = ENOENT, NULL;
4447 }
4448
bpf_object__shndx_is_data(const struct bpf_object * obj,int shndx)4449 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
4450 int shndx)
4451 {
4452 switch (obj->efile.secs[shndx].sec_type) {
4453 case SEC_BSS:
4454 case SEC_DATA:
4455 case SEC_RODATA:
4456 return true;
4457 default:
4458 return false;
4459 }
4460 }
4461
bpf_object__shndx_is_maps(const struct bpf_object * obj,int shndx)4462 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
4463 int shndx)
4464 {
4465 return shndx == obj->efile.btf_maps_shndx;
4466 }
4467
4468 static enum libbpf_map_type
bpf_object__section_to_libbpf_map_type(const struct bpf_object * obj,int shndx)4469 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
4470 {
4471 if (shndx == obj->efile.symbols_shndx)
4472 return LIBBPF_MAP_KCONFIG;
4473
4474 switch (obj->efile.secs[shndx].sec_type) {
4475 case SEC_BSS:
4476 return LIBBPF_MAP_BSS;
4477 case SEC_DATA:
4478 return LIBBPF_MAP_DATA;
4479 case SEC_RODATA:
4480 return LIBBPF_MAP_RODATA;
4481 default:
4482 return LIBBPF_MAP_UNSPEC;
4483 }
4484 }
4485
bpf_program__record_reloc(struct bpf_program * prog,struct reloc_desc * reloc_desc,__u32 insn_idx,const char * sym_name,const Elf64_Sym * sym,const Elf64_Rel * rel)4486 static int bpf_program__record_reloc(struct bpf_program *prog,
4487 struct reloc_desc *reloc_desc,
4488 __u32 insn_idx, const char *sym_name,
4489 const Elf64_Sym *sym, const Elf64_Rel *rel)
4490 {
4491 struct bpf_insn *insn = &prog->insns[insn_idx];
4492 size_t map_idx, nr_maps = prog->obj->nr_maps;
4493 struct bpf_object *obj = prog->obj;
4494 __u32 shdr_idx = sym->st_shndx;
4495 enum libbpf_map_type type;
4496 const char *sym_sec_name;
4497 struct bpf_map *map;
4498
4499 if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
4500 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
4501 prog->name, sym_name, insn_idx, insn->code);
4502 return -LIBBPF_ERRNO__RELOC;
4503 }
4504
4505 if (sym_is_extern(sym)) {
4506 int sym_idx = ELF64_R_SYM(rel->r_info);
4507 int i, n = obj->nr_extern;
4508 struct extern_desc *ext;
4509
4510 for (i = 0; i < n; i++) {
4511 ext = &obj->externs[i];
4512 if (ext->sym_idx == sym_idx)
4513 break;
4514 }
4515 if (i >= n) {
4516 pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
4517 prog->name, sym_name, sym_idx);
4518 return -LIBBPF_ERRNO__RELOC;
4519 }
4520 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
4521 prog->name, i, ext->name, ext->sym_idx, insn_idx);
4522 if (insn->code == (BPF_JMP | BPF_CALL))
4523 reloc_desc->type = RELO_EXTERN_CALL;
4524 else
4525 reloc_desc->type = RELO_EXTERN_LD64;
4526 reloc_desc->insn_idx = insn_idx;
4527 reloc_desc->ext_idx = i;
4528 return 0;
4529 }
4530
4531 /* sub-program call relocation */
4532 if (is_call_insn(insn)) {
4533 if (insn->src_reg != BPF_PSEUDO_CALL) {
4534 pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
4535 return -LIBBPF_ERRNO__RELOC;
4536 }
4537 /* text_shndx can be 0, if no default "main" program exists */
4538 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
4539 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4540 pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
4541 prog->name, sym_name, sym_sec_name);
4542 return -LIBBPF_ERRNO__RELOC;
4543 }
4544 if (sym->st_value % BPF_INSN_SZ) {
4545 pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
4546 prog->name, sym_name, (size_t)sym->st_value);
4547 return -LIBBPF_ERRNO__RELOC;
4548 }
4549 reloc_desc->type = RELO_CALL;
4550 reloc_desc->insn_idx = insn_idx;
4551 reloc_desc->sym_off = sym->st_value;
4552 return 0;
4553 }
4554
4555 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
4556 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
4557 prog->name, sym_name, shdr_idx);
4558 return -LIBBPF_ERRNO__RELOC;
4559 }
4560
4561 /* loading subprog addresses */
4562 if (sym_is_subprog(sym, obj->efile.text_shndx)) {
4563 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
4564 * local_func: sym->st_value = 0, insn->imm = offset in the section.
4565 */
4566 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
4567 pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
4568 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
4569 return -LIBBPF_ERRNO__RELOC;
4570 }
4571
4572 reloc_desc->type = RELO_SUBPROG_ADDR;
4573 reloc_desc->insn_idx = insn_idx;
4574 reloc_desc->sym_off = sym->st_value;
4575 return 0;
4576 }
4577
4578 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4579 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4580
4581 /* arena data relocation */
4582 if (shdr_idx == obj->efile.arena_data_shndx) {
4583 reloc_desc->type = RELO_DATA;
4584 reloc_desc->insn_idx = insn_idx;
4585 reloc_desc->map_idx = obj->arena_map - obj->maps;
4586 reloc_desc->sym_off = sym->st_value;
4587 return 0;
4588 }
4589
4590 /* generic map reference relocation */
4591 if (type == LIBBPF_MAP_UNSPEC) {
4592 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4593 pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4594 prog->name, sym_name, sym_sec_name);
4595 return -LIBBPF_ERRNO__RELOC;
4596 }
4597 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4598 map = &obj->maps[map_idx];
4599 if (map->libbpf_type != type ||
4600 map->sec_idx != sym->st_shndx ||
4601 map->sec_offset != sym->st_value)
4602 continue;
4603 pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4604 prog->name, map_idx, map->name, map->sec_idx,
4605 map->sec_offset, insn_idx);
4606 break;
4607 }
4608 if (map_idx >= nr_maps) {
4609 pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4610 prog->name, sym_sec_name, (size_t)sym->st_value);
4611 return -LIBBPF_ERRNO__RELOC;
4612 }
4613 reloc_desc->type = RELO_LD64;
4614 reloc_desc->insn_idx = insn_idx;
4615 reloc_desc->map_idx = map_idx;
4616 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4617 return 0;
4618 }
4619
4620 /* global data map relocation */
4621 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4622 pr_warn("prog '%s': bad data relo against section '%s'\n",
4623 prog->name, sym_sec_name);
4624 return -LIBBPF_ERRNO__RELOC;
4625 }
4626 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4627 map = &obj->maps[map_idx];
4628 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4629 continue;
4630 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4631 prog->name, map_idx, map->name, map->sec_idx,
4632 map->sec_offset, insn_idx);
4633 break;
4634 }
4635 if (map_idx >= nr_maps) {
4636 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4637 prog->name, sym_sec_name);
4638 return -LIBBPF_ERRNO__RELOC;
4639 }
4640
4641 reloc_desc->type = RELO_DATA;
4642 reloc_desc->insn_idx = insn_idx;
4643 reloc_desc->map_idx = map_idx;
4644 reloc_desc->sym_off = sym->st_value;
4645 return 0;
4646 }
4647
prog_contains_insn(const struct bpf_program * prog,size_t insn_idx)4648 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4649 {
4650 return insn_idx >= prog->sec_insn_off &&
4651 insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4652 }
4653
find_prog_by_sec_insn(const struct bpf_object * obj,size_t sec_idx,size_t insn_idx)4654 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4655 size_t sec_idx, size_t insn_idx)
4656 {
4657 int l = 0, r = obj->nr_programs - 1, m;
4658 struct bpf_program *prog;
4659
4660 if (!obj->nr_programs)
4661 return NULL;
4662
4663 while (l < r) {
4664 m = l + (r - l + 1) / 2;
4665 prog = &obj->programs[m];
4666
4667 if (prog->sec_idx < sec_idx ||
4668 (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4669 l = m;
4670 else
4671 r = m - 1;
4672 }
4673 /* matching program could be at index l, but it still might be the
4674 * wrong one, so we need to double check conditions for the last time
4675 */
4676 prog = &obj->programs[l];
4677 if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4678 return prog;
4679 return NULL;
4680 }
4681
4682 static int
bpf_object__collect_prog_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)4683 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4684 {
4685 const char *relo_sec_name, *sec_name;
4686 size_t sec_idx = shdr->sh_info, sym_idx;
4687 struct bpf_program *prog;
4688 struct reloc_desc *relos;
4689 int err, i, nrels;
4690 const char *sym_name;
4691 __u32 insn_idx;
4692 Elf_Scn *scn;
4693 Elf_Data *scn_data;
4694 Elf64_Sym *sym;
4695 Elf64_Rel *rel;
4696
4697 if (sec_idx >= obj->efile.sec_cnt)
4698 return -EINVAL;
4699
4700 scn = elf_sec_by_idx(obj, sec_idx);
4701 scn_data = elf_sec_data(obj, scn);
4702 if (!scn_data)
4703 return -LIBBPF_ERRNO__FORMAT;
4704
4705 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4706 sec_name = elf_sec_name(obj, scn);
4707 if (!relo_sec_name || !sec_name)
4708 return -EINVAL;
4709
4710 pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4711 relo_sec_name, sec_idx, sec_name);
4712 nrels = shdr->sh_size / shdr->sh_entsize;
4713
4714 for (i = 0; i < nrels; i++) {
4715 rel = elf_rel_by_idx(data, i);
4716 if (!rel) {
4717 pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4718 return -LIBBPF_ERRNO__FORMAT;
4719 }
4720
4721 sym_idx = ELF64_R_SYM(rel->r_info);
4722 sym = elf_sym_by_idx(obj, sym_idx);
4723 if (!sym) {
4724 pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4725 relo_sec_name, sym_idx, i);
4726 return -LIBBPF_ERRNO__FORMAT;
4727 }
4728
4729 if (sym->st_shndx >= obj->efile.sec_cnt) {
4730 pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4731 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4732 return -LIBBPF_ERRNO__FORMAT;
4733 }
4734
4735 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4736 pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4737 relo_sec_name, (size_t)rel->r_offset, i);
4738 return -LIBBPF_ERRNO__FORMAT;
4739 }
4740
4741 insn_idx = rel->r_offset / BPF_INSN_SZ;
4742 /* relocations against static functions are recorded as
4743 * relocations against the section that contains a function;
4744 * in such case, symbol will be STT_SECTION and sym.st_name
4745 * will point to empty string (0), so fetch section name
4746 * instead
4747 */
4748 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4749 sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4750 else
4751 sym_name = elf_sym_str(obj, sym->st_name);
4752 sym_name = sym_name ?: "<?";
4753
4754 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4755 relo_sec_name, i, insn_idx, sym_name);
4756
4757 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4758 if (!prog) {
4759 pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4760 relo_sec_name, i, sec_name, insn_idx);
4761 continue;
4762 }
4763
4764 relos = libbpf_reallocarray(prog->reloc_desc,
4765 prog->nr_reloc + 1, sizeof(*relos));
4766 if (!relos)
4767 return -ENOMEM;
4768 prog->reloc_desc = relos;
4769
4770 /* adjust insn_idx to local BPF program frame of reference */
4771 insn_idx -= prog->sec_insn_off;
4772 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4773 insn_idx, sym_name, sym, rel);
4774 if (err)
4775 return err;
4776
4777 prog->nr_reloc++;
4778 }
4779 return 0;
4780 }
4781
map_fill_btf_type_info(struct bpf_object * obj,struct bpf_map * map)4782 static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map)
4783 {
4784 int id;
4785
4786 if (!obj->btf)
4787 return -ENOENT;
4788
4789 /* if it's BTF-defined map, we don't need to search for type IDs.
4790 * For struct_ops map, it does not need btf_key_type_id and
4791 * btf_value_type_id.
4792 */
4793 if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map))
4794 return 0;
4795
4796 /*
4797 * LLVM annotates global data differently in BTF, that is,
4798 * only as '.data', '.bss' or '.rodata'.
4799 */
4800 if (!bpf_map__is_internal(map))
4801 return -ENOENT;
4802
4803 id = btf__find_by_name(obj->btf, map->real_name);
4804 if (id < 0)
4805 return id;
4806
4807 map->btf_key_type_id = 0;
4808 map->btf_value_type_id = id;
4809 return 0;
4810 }
4811
bpf_get_map_info_from_fdinfo(int fd,struct bpf_map_info * info)4812 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4813 {
4814 char file[PATH_MAX], buff[4096];
4815 FILE *fp;
4816 __u32 val;
4817 int err;
4818
4819 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4820 memset(info, 0, sizeof(*info));
4821
4822 fp = fopen(file, "re");
4823 if (!fp) {
4824 err = -errno;
4825 pr_warn("failed to open %s: %s. No procfs support?\n", file,
4826 errstr(err));
4827 return err;
4828 }
4829
4830 while (fgets(buff, sizeof(buff), fp)) {
4831 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4832 info->type = val;
4833 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4834 info->key_size = val;
4835 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4836 info->value_size = val;
4837 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4838 info->max_entries = val;
4839 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4840 info->map_flags = val;
4841 }
4842
4843 fclose(fp);
4844
4845 return 0;
4846 }
4847
bpf_map__autocreate(const struct bpf_map * map)4848 bool bpf_map__autocreate(const struct bpf_map *map)
4849 {
4850 return map->autocreate;
4851 }
4852
bpf_map__set_autocreate(struct bpf_map * map,bool autocreate)4853 int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
4854 {
4855 if (map->obj->loaded)
4856 return libbpf_err(-EBUSY);
4857
4858 map->autocreate = autocreate;
4859 return 0;
4860 }
4861
bpf_map__set_autoattach(struct bpf_map * map,bool autoattach)4862 int bpf_map__set_autoattach(struct bpf_map *map, bool autoattach)
4863 {
4864 if (!bpf_map__is_struct_ops(map))
4865 return libbpf_err(-EINVAL);
4866
4867 map->autoattach = autoattach;
4868 return 0;
4869 }
4870
bpf_map__autoattach(const struct bpf_map * map)4871 bool bpf_map__autoattach(const struct bpf_map *map)
4872 {
4873 return map->autoattach;
4874 }
4875
bpf_map__reuse_fd(struct bpf_map * map,int fd)4876 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4877 {
4878 struct bpf_map_info info;
4879 __u32 len = sizeof(info), name_len;
4880 int new_fd, err;
4881 char *new_name;
4882
4883 memset(&info, 0, len);
4884 err = bpf_map_get_info_by_fd(fd, &info, &len);
4885 if (err && errno == EINVAL)
4886 err = bpf_get_map_info_from_fdinfo(fd, &info);
4887 if (err)
4888 return libbpf_err(err);
4889
4890 name_len = strlen(info.name);
4891 if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0)
4892 new_name = strdup(map->name);
4893 else
4894 new_name = strdup(info.name);
4895
4896 if (!new_name)
4897 return libbpf_err(-errno);
4898
4899 /*
4900 * Like dup(), but make sure new FD is >= 3 and has O_CLOEXEC set.
4901 * This is similar to what we do in ensure_good_fd(), but without
4902 * closing original FD.
4903 */
4904 new_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3);
4905 if (new_fd < 0) {
4906 err = -errno;
4907 goto err_free_new_name;
4908 }
4909
4910 err = reuse_fd(map->fd, new_fd);
4911 if (err)
4912 goto err_free_new_name;
4913
4914 free(map->name);
4915
4916 map->name = new_name;
4917 map->def.type = info.type;
4918 map->def.key_size = info.key_size;
4919 map->def.value_size = info.value_size;
4920 map->def.max_entries = info.max_entries;
4921 map->def.map_flags = info.map_flags;
4922 map->btf_key_type_id = info.btf_key_type_id;
4923 map->btf_value_type_id = info.btf_value_type_id;
4924 map->reused = true;
4925 map->map_extra = info.map_extra;
4926
4927 return 0;
4928
4929 err_free_new_name:
4930 free(new_name);
4931 return libbpf_err(err);
4932 }
4933
bpf_map__max_entries(const struct bpf_map * map)4934 __u32 bpf_map__max_entries(const struct bpf_map *map)
4935 {
4936 return map->def.max_entries;
4937 }
4938
bpf_map__inner_map(struct bpf_map * map)4939 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4940 {
4941 if (!bpf_map_type__is_map_in_map(map->def.type))
4942 return errno = EINVAL, NULL;
4943
4944 return map->inner_map;
4945 }
4946
bpf_map__set_max_entries(struct bpf_map * map,__u32 max_entries)4947 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4948 {
4949 if (map->obj->loaded)
4950 return libbpf_err(-EBUSY);
4951
4952 map->def.max_entries = max_entries;
4953
4954 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
4955 if (map_is_ringbuf(map))
4956 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
4957
4958 return 0;
4959 }
4960
bpf_object_prepare_token(struct bpf_object * obj)4961 static int bpf_object_prepare_token(struct bpf_object *obj)
4962 {
4963 const char *bpffs_path;
4964 int bpffs_fd = -1, token_fd, err;
4965 bool mandatory;
4966 enum libbpf_print_level level;
4967
4968 /* token is explicitly prevented */
4969 if (obj->token_path && obj->token_path[0] == '\0') {
4970 pr_debug("object '%s': token is prevented, skipping...\n", obj->name);
4971 return 0;
4972 }
4973
4974 mandatory = obj->token_path != NULL;
4975 level = mandatory ? LIBBPF_WARN : LIBBPF_DEBUG;
4976
4977 bpffs_path = obj->token_path ?: BPF_FS_DEFAULT_PATH;
4978 bpffs_fd = open(bpffs_path, O_DIRECTORY, O_RDWR);
4979 if (bpffs_fd < 0) {
4980 err = -errno;
4981 __pr(level, "object '%s': failed (%s) to open BPF FS mount at '%s'%s\n",
4982 obj->name, errstr(err), bpffs_path,
4983 mandatory ? "" : ", skipping optional step...");
4984 return mandatory ? err : 0;
4985 }
4986
4987 token_fd = bpf_token_create(bpffs_fd, 0);
4988 close(bpffs_fd);
4989 if (token_fd < 0) {
4990 if (!mandatory && token_fd == -ENOENT) {
4991 pr_debug("object '%s': BPF FS at '%s' doesn't have BPF token delegation set up, skipping...\n",
4992 obj->name, bpffs_path);
4993 return 0;
4994 }
4995 __pr(level, "object '%s': failed (%d) to create BPF token from '%s'%s\n",
4996 obj->name, token_fd, bpffs_path,
4997 mandatory ? "" : ", skipping optional step...");
4998 return mandatory ? token_fd : 0;
4999 }
5000
5001 obj->feat_cache = calloc(1, sizeof(*obj->feat_cache));
5002 if (!obj->feat_cache) {
5003 close(token_fd);
5004 return -ENOMEM;
5005 }
5006
5007 obj->token_fd = token_fd;
5008 obj->feat_cache->token_fd = token_fd;
5009
5010 return 0;
5011 }
5012
5013 static int
bpf_object__probe_loading(struct bpf_object * obj)5014 bpf_object__probe_loading(struct bpf_object *obj)
5015 {
5016 struct bpf_insn insns[] = {
5017 BPF_MOV64_IMM(BPF_REG_0, 0),
5018 BPF_EXIT_INSN(),
5019 };
5020 int ret, insn_cnt = ARRAY_SIZE(insns);
5021 LIBBPF_OPTS(bpf_prog_load_opts, opts,
5022 .token_fd = obj->token_fd,
5023 .prog_flags = obj->token_fd ? BPF_F_TOKEN_FD : 0,
5024 );
5025
5026 if (obj->gen_loader)
5027 return 0;
5028
5029 ret = bump_rlimit_memlock();
5030 if (ret)
5031 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %s), you might need to do it explicitly!\n",
5032 errstr(ret));
5033
5034 /* make sure basic loading works */
5035 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, &opts);
5036 if (ret < 0)
5037 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, &opts);
5038 if (ret < 0) {
5039 ret = errno;
5040 pr_warn("Error in %s(): %s. Couldn't load trivial BPF program. Make sure your kernel supports BPF (CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is set to big enough value.\n",
5041 __func__, errstr(ret));
5042 return -ret;
5043 }
5044 close(ret);
5045
5046 return 0;
5047 }
5048
kernel_supports(const struct bpf_object * obj,enum kern_feature_id feat_id)5049 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
5050 {
5051 if (obj->gen_loader)
5052 /* To generate loader program assume the latest kernel
5053 * to avoid doing extra prog_load, map_create syscalls.
5054 */
5055 return true;
5056
5057 if (obj->token_fd)
5058 return feat_supported(obj->feat_cache, feat_id);
5059
5060 return feat_supported(NULL, feat_id);
5061 }
5062
map_is_reuse_compat(const struct bpf_map * map,int map_fd)5063 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
5064 {
5065 struct bpf_map_info map_info;
5066 __u32 map_info_len = sizeof(map_info);
5067 int err;
5068
5069 memset(&map_info, 0, map_info_len);
5070 err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len);
5071 if (err && errno == EINVAL)
5072 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
5073 if (err) {
5074 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
5075 errstr(err));
5076 return false;
5077 }
5078
5079 return (map_info.type == map->def.type &&
5080 map_info.key_size == map->def.key_size &&
5081 map_info.value_size == map->def.value_size &&
5082 map_info.max_entries == map->def.max_entries &&
5083 map_info.map_flags == map->def.map_flags &&
5084 map_info.map_extra == map->map_extra);
5085 }
5086
5087 static int
bpf_object__reuse_map(struct bpf_map * map)5088 bpf_object__reuse_map(struct bpf_map *map)
5089 {
5090 int err, pin_fd;
5091
5092 pin_fd = bpf_obj_get(map->pin_path);
5093 if (pin_fd < 0) {
5094 err = -errno;
5095 if (err == -ENOENT) {
5096 pr_debug("found no pinned map to reuse at '%s'\n",
5097 map->pin_path);
5098 return 0;
5099 }
5100
5101 pr_warn("couldn't retrieve pinned map '%s': %s\n",
5102 map->pin_path, errstr(err));
5103 return err;
5104 }
5105
5106 if (!map_is_reuse_compat(map, pin_fd)) {
5107 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
5108 map->pin_path);
5109 close(pin_fd);
5110 return -EINVAL;
5111 }
5112
5113 err = bpf_map__reuse_fd(map, pin_fd);
5114 close(pin_fd);
5115 if (err)
5116 return err;
5117
5118 map->pinned = true;
5119 pr_debug("reused pinned map at '%s'\n", map->pin_path);
5120
5121 return 0;
5122 }
5123
5124 static int
bpf_object__populate_internal_map(struct bpf_object * obj,struct bpf_map * map)5125 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
5126 {
5127 enum libbpf_map_type map_type = map->libbpf_type;
5128 int err, zero = 0;
5129 size_t mmap_sz;
5130
5131 if (obj->gen_loader) {
5132 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
5133 map->mmaped, map->def.value_size);
5134 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
5135 bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
5136 return 0;
5137 }
5138
5139 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
5140 if (err) {
5141 err = -errno;
5142 pr_warn("map '%s': failed to set initial contents: %s\n",
5143 bpf_map__name(map), errstr(err));
5144 return err;
5145 }
5146
5147 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
5148 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
5149 err = bpf_map_freeze(map->fd);
5150 if (err) {
5151 err = -errno;
5152 pr_warn("map '%s': failed to freeze as read-only: %s\n",
5153 bpf_map__name(map), errstr(err));
5154 return err;
5155 }
5156 }
5157
5158 /* Remap anonymous mmap()-ed "map initialization image" as
5159 * a BPF map-backed mmap()-ed memory, but preserving the same
5160 * memory address. This will cause kernel to change process'
5161 * page table to point to a different piece of kernel memory,
5162 * but from userspace point of view memory address (and its
5163 * contents, being identical at this point) will stay the
5164 * same. This mapping will be released by bpf_object__close()
5165 * as per normal clean up procedure.
5166 */
5167 mmap_sz = bpf_map_mmap_sz(map);
5168 if (map->def.map_flags & BPF_F_MMAPABLE) {
5169 void *mmaped;
5170 int prot;
5171
5172 if (map->def.map_flags & BPF_F_RDONLY_PROG)
5173 prot = PROT_READ;
5174 else
5175 prot = PROT_READ | PROT_WRITE;
5176 mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map->fd, 0);
5177 if (mmaped == MAP_FAILED) {
5178 err = -errno;
5179 pr_warn("map '%s': failed to re-mmap() contents: %s\n",
5180 bpf_map__name(map), errstr(err));
5181 return err;
5182 }
5183 map->mmaped = mmaped;
5184 } else if (map->mmaped) {
5185 munmap(map->mmaped, mmap_sz);
5186 map->mmaped = NULL;
5187 }
5188
5189 return 0;
5190 }
5191
5192 static void bpf_map__destroy(struct bpf_map *map);
5193
map_is_created(const struct bpf_map * map)5194 static bool map_is_created(const struct bpf_map *map)
5195 {
5196 return map->obj->loaded || map->reused;
5197 }
5198
bpf_object__create_map(struct bpf_object * obj,struct bpf_map * map,bool is_inner)5199 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
5200 {
5201 LIBBPF_OPTS(bpf_map_create_opts, create_attr);
5202 struct bpf_map_def *def = &map->def;
5203 const char *map_name = NULL;
5204 int err = 0, map_fd;
5205
5206 if (kernel_supports(obj, FEAT_PROG_NAME))
5207 map_name = map->name;
5208 create_attr.map_ifindex = map->map_ifindex;
5209 create_attr.map_flags = def->map_flags;
5210 create_attr.numa_node = map->numa_node;
5211 create_attr.map_extra = map->map_extra;
5212 create_attr.token_fd = obj->token_fd;
5213 if (obj->token_fd)
5214 create_attr.map_flags |= BPF_F_TOKEN_FD;
5215
5216 if (bpf_map__is_struct_ops(map)) {
5217 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5218 if (map->mod_btf_fd >= 0) {
5219 create_attr.value_type_btf_obj_fd = map->mod_btf_fd;
5220 create_attr.map_flags |= BPF_F_VTYPE_BTF_OBJ_FD;
5221 }
5222 }
5223
5224 if (obj->btf && btf__fd(obj->btf) >= 0) {
5225 create_attr.btf_fd = btf__fd(obj->btf);
5226 create_attr.btf_key_type_id = map->btf_key_type_id;
5227 create_attr.btf_value_type_id = map->btf_value_type_id;
5228 }
5229
5230 if (bpf_map_type__is_map_in_map(def->type)) {
5231 if (map->inner_map) {
5232 err = map_set_def_max_entries(map->inner_map);
5233 if (err)
5234 return err;
5235 err = bpf_object__create_map(obj, map->inner_map, true);
5236 if (err) {
5237 pr_warn("map '%s': failed to create inner map: %s\n",
5238 map->name, errstr(err));
5239 return err;
5240 }
5241 map->inner_map_fd = map->inner_map->fd;
5242 }
5243 if (map->inner_map_fd >= 0)
5244 create_attr.inner_map_fd = map->inner_map_fd;
5245 }
5246
5247 switch (def->type) {
5248 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5249 case BPF_MAP_TYPE_CGROUP_ARRAY:
5250 case BPF_MAP_TYPE_STACK_TRACE:
5251 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
5252 case BPF_MAP_TYPE_HASH_OF_MAPS:
5253 case BPF_MAP_TYPE_DEVMAP:
5254 case BPF_MAP_TYPE_DEVMAP_HASH:
5255 case BPF_MAP_TYPE_CPUMAP:
5256 case BPF_MAP_TYPE_XSKMAP:
5257 case BPF_MAP_TYPE_SOCKMAP:
5258 case BPF_MAP_TYPE_SOCKHASH:
5259 case BPF_MAP_TYPE_QUEUE:
5260 case BPF_MAP_TYPE_STACK:
5261 case BPF_MAP_TYPE_ARENA:
5262 create_attr.btf_fd = 0;
5263 create_attr.btf_key_type_id = 0;
5264 create_attr.btf_value_type_id = 0;
5265 map->btf_key_type_id = 0;
5266 map->btf_value_type_id = 0;
5267 break;
5268 case BPF_MAP_TYPE_STRUCT_OPS:
5269 create_attr.btf_value_type_id = 0;
5270 break;
5271 default:
5272 break;
5273 }
5274
5275 if (obj->gen_loader) {
5276 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
5277 def->key_size, def->value_size, def->max_entries,
5278 &create_attr, is_inner ? -1 : map - obj->maps);
5279 /* We keep pretenting we have valid FD to pass various fd >= 0
5280 * checks by just keeping original placeholder FDs in place.
5281 * See bpf_object__add_map() comment.
5282 * This placeholder fd will not be used with any syscall and
5283 * will be reset to -1 eventually.
5284 */
5285 map_fd = map->fd;
5286 } else {
5287 map_fd = bpf_map_create(def->type, map_name,
5288 def->key_size, def->value_size,
5289 def->max_entries, &create_attr);
5290 }
5291 if (map_fd < 0 && (create_attr.btf_key_type_id || create_attr.btf_value_type_id)) {
5292 err = -errno;
5293 pr_warn("Error in bpf_create_map_xattr(%s): %s. Retrying without BTF.\n",
5294 map->name, errstr(err));
5295 create_attr.btf_fd = 0;
5296 create_attr.btf_key_type_id = 0;
5297 create_attr.btf_value_type_id = 0;
5298 map->btf_key_type_id = 0;
5299 map->btf_value_type_id = 0;
5300 map_fd = bpf_map_create(def->type, map_name,
5301 def->key_size, def->value_size,
5302 def->max_entries, &create_attr);
5303 }
5304
5305 if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
5306 if (obj->gen_loader)
5307 map->inner_map->fd = -1;
5308 bpf_map__destroy(map->inner_map);
5309 zfree(&map->inner_map);
5310 }
5311
5312 if (map_fd < 0)
5313 return map_fd;
5314
5315 /* obj->gen_loader case, prevent reuse_fd() from closing map_fd */
5316 if (map->fd == map_fd)
5317 return 0;
5318
5319 /* Keep placeholder FD value but now point it to the BPF map object.
5320 * This way everything that relied on this map's FD (e.g., relocated
5321 * ldimm64 instructions) will stay valid and won't need adjustments.
5322 * map->fd stays valid but now point to what map_fd points to.
5323 */
5324 return reuse_fd(map->fd, map_fd);
5325 }
5326
init_map_in_map_slots(struct bpf_object * obj,struct bpf_map * map)5327 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
5328 {
5329 const struct bpf_map *targ_map;
5330 unsigned int i;
5331 int fd, err = 0;
5332
5333 for (i = 0; i < map->init_slots_sz; i++) {
5334 if (!map->init_slots[i])
5335 continue;
5336
5337 targ_map = map->init_slots[i];
5338 fd = targ_map->fd;
5339
5340 if (obj->gen_loader) {
5341 bpf_gen__populate_outer_map(obj->gen_loader,
5342 map - obj->maps, i,
5343 targ_map - obj->maps);
5344 } else {
5345 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5346 }
5347 if (err) {
5348 err = -errno;
5349 pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %s\n",
5350 map->name, i, targ_map->name, fd, errstr(err));
5351 return err;
5352 }
5353 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5354 map->name, i, targ_map->name, fd);
5355 }
5356
5357 zfree(&map->init_slots);
5358 map->init_slots_sz = 0;
5359
5360 return 0;
5361 }
5362
init_prog_array_slots(struct bpf_object * obj,struct bpf_map * map)5363 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5364 {
5365 const struct bpf_program *targ_prog;
5366 unsigned int i;
5367 int fd, err;
5368
5369 if (obj->gen_loader)
5370 return -ENOTSUP;
5371
5372 for (i = 0; i < map->init_slots_sz; i++) {
5373 if (!map->init_slots[i])
5374 continue;
5375
5376 targ_prog = map->init_slots[i];
5377 fd = bpf_program__fd(targ_prog);
5378
5379 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5380 if (err) {
5381 err = -errno;
5382 pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %s\n",
5383 map->name, i, targ_prog->name, fd, errstr(err));
5384 return err;
5385 }
5386 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5387 map->name, i, targ_prog->name, fd);
5388 }
5389
5390 zfree(&map->init_slots);
5391 map->init_slots_sz = 0;
5392
5393 return 0;
5394 }
5395
bpf_object_init_prog_arrays(struct bpf_object * obj)5396 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5397 {
5398 struct bpf_map *map;
5399 int i, err;
5400
5401 for (i = 0; i < obj->nr_maps; i++) {
5402 map = &obj->maps[i];
5403
5404 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5405 continue;
5406
5407 err = init_prog_array_slots(obj, map);
5408 if (err < 0)
5409 return err;
5410 }
5411 return 0;
5412 }
5413
map_set_def_max_entries(struct bpf_map * map)5414 static int map_set_def_max_entries(struct bpf_map *map)
5415 {
5416 if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5417 int nr_cpus;
5418
5419 nr_cpus = libbpf_num_possible_cpus();
5420 if (nr_cpus < 0) {
5421 pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5422 map->name, nr_cpus);
5423 return nr_cpus;
5424 }
5425 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5426 map->def.max_entries = nr_cpus;
5427 }
5428
5429 return 0;
5430 }
5431
5432 static int
bpf_object__create_maps(struct bpf_object * obj)5433 bpf_object__create_maps(struct bpf_object *obj)
5434 {
5435 struct bpf_map *map;
5436 unsigned int i, j;
5437 int err;
5438 bool retried;
5439
5440 for (i = 0; i < obj->nr_maps; i++) {
5441 map = &obj->maps[i];
5442
5443 /* To support old kernels, we skip creating global data maps
5444 * (.rodata, .data, .kconfig, etc); later on, during program
5445 * loading, if we detect that at least one of the to-be-loaded
5446 * programs is referencing any global data map, we'll error
5447 * out with program name and relocation index logged.
5448 * This approach allows to accommodate Clang emitting
5449 * unnecessary .rodata.str1.1 sections for string literals,
5450 * but also it allows to have CO-RE applications that use
5451 * global variables in some of BPF programs, but not others.
5452 * If those global variable-using programs are not loaded at
5453 * runtime due to bpf_program__set_autoload(prog, false),
5454 * bpf_object loading will succeed just fine even on old
5455 * kernels.
5456 */
5457 if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA))
5458 map->autocreate = false;
5459
5460 if (!map->autocreate) {
5461 pr_debug("map '%s': skipped auto-creating...\n", map->name);
5462 continue;
5463 }
5464
5465 err = map_set_def_max_entries(map);
5466 if (err)
5467 goto err_out;
5468
5469 retried = false;
5470 retry:
5471 if (map->pin_path) {
5472 err = bpf_object__reuse_map(map);
5473 if (err) {
5474 pr_warn("map '%s': error reusing pinned map\n",
5475 map->name);
5476 goto err_out;
5477 }
5478 if (retried && map->fd < 0) {
5479 pr_warn("map '%s': cannot find pinned map\n",
5480 map->name);
5481 err = -ENOENT;
5482 goto err_out;
5483 }
5484 }
5485
5486 if (map->reused) {
5487 pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5488 map->name, map->fd);
5489 } else {
5490 err = bpf_object__create_map(obj, map, false);
5491 if (err)
5492 goto err_out;
5493
5494 pr_debug("map '%s': created successfully, fd=%d\n",
5495 map->name, map->fd);
5496
5497 if (bpf_map__is_internal(map)) {
5498 err = bpf_object__populate_internal_map(obj, map);
5499 if (err < 0)
5500 goto err_out;
5501 } else if (map->def.type == BPF_MAP_TYPE_ARENA) {
5502 map->mmaped = mmap((void *)(long)map->map_extra,
5503 bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
5504 map->map_extra ? MAP_SHARED | MAP_FIXED : MAP_SHARED,
5505 map->fd, 0);
5506 if (map->mmaped == MAP_FAILED) {
5507 err = -errno;
5508 map->mmaped = NULL;
5509 pr_warn("map '%s': failed to mmap arena: %s\n",
5510 map->name, errstr(err));
5511 return err;
5512 }
5513 if (obj->arena_data) {
5514 memcpy(map->mmaped, obj->arena_data, obj->arena_data_sz);
5515 zfree(&obj->arena_data);
5516 }
5517 }
5518 if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5519 err = init_map_in_map_slots(obj, map);
5520 if (err < 0)
5521 goto err_out;
5522 }
5523 }
5524
5525 if (map->pin_path && !map->pinned) {
5526 err = bpf_map__pin(map, NULL);
5527 if (err) {
5528 if (!retried && err == -EEXIST) {
5529 retried = true;
5530 goto retry;
5531 }
5532 pr_warn("map '%s': failed to auto-pin at '%s': %s\n",
5533 map->name, map->pin_path, errstr(err));
5534 goto err_out;
5535 }
5536 }
5537 }
5538
5539 return 0;
5540
5541 err_out:
5542 pr_warn("map '%s': failed to create: %s\n", map->name, errstr(err));
5543 pr_perm_msg(err);
5544 for (j = 0; j < i; j++)
5545 zclose(obj->maps[j].fd);
5546 return err;
5547 }
5548
bpf_core_is_flavor_sep(const char * s)5549 static bool bpf_core_is_flavor_sep(const char *s)
5550 {
5551 /* check X___Y name pattern, where X and Y are not underscores */
5552 return s[0] != '_' && /* X */
5553 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
5554 s[4] != '_'; /* Y */
5555 }
5556
5557 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5558 * before last triple underscore. Struct name part after last triple
5559 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5560 */
bpf_core_essential_name_len(const char * name)5561 size_t bpf_core_essential_name_len(const char *name)
5562 {
5563 size_t n = strlen(name);
5564 int i;
5565
5566 for (i = n - 5; i >= 0; i--) {
5567 if (bpf_core_is_flavor_sep(name + i))
5568 return i + 1;
5569 }
5570 return n;
5571 }
5572
bpf_core_free_cands(struct bpf_core_cand_list * cands)5573 void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5574 {
5575 if (!cands)
5576 return;
5577
5578 free(cands->cands);
5579 free(cands);
5580 }
5581
bpf_core_add_cands(struct bpf_core_cand * local_cand,size_t local_essent_len,const struct btf * targ_btf,const char * targ_btf_name,int targ_start_id,struct bpf_core_cand_list * cands)5582 int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5583 size_t local_essent_len,
5584 const struct btf *targ_btf,
5585 const char *targ_btf_name,
5586 int targ_start_id,
5587 struct bpf_core_cand_list *cands)
5588 {
5589 struct bpf_core_cand *new_cands, *cand;
5590 const struct btf_type *t, *local_t;
5591 const char *targ_name, *local_name;
5592 size_t targ_essent_len;
5593 int n, i;
5594
5595 local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5596 local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5597
5598 n = btf__type_cnt(targ_btf);
5599 for (i = targ_start_id; i < n; i++) {
5600 t = btf__type_by_id(targ_btf, i);
5601 if (!btf_kind_core_compat(t, local_t))
5602 continue;
5603
5604 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5605 if (str_is_empty(targ_name))
5606 continue;
5607
5608 targ_essent_len = bpf_core_essential_name_len(targ_name);
5609 if (targ_essent_len != local_essent_len)
5610 continue;
5611
5612 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5613 continue;
5614
5615 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5616 local_cand->id, btf_kind_str(local_t),
5617 local_name, i, btf_kind_str(t), targ_name,
5618 targ_btf_name);
5619 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5620 sizeof(*cands->cands));
5621 if (!new_cands)
5622 return -ENOMEM;
5623
5624 cand = &new_cands[cands->len];
5625 cand->btf = targ_btf;
5626 cand->id = i;
5627
5628 cands->cands = new_cands;
5629 cands->len++;
5630 }
5631 return 0;
5632 }
5633
load_module_btfs(struct bpf_object * obj)5634 static int load_module_btfs(struct bpf_object *obj)
5635 {
5636 struct bpf_btf_info info;
5637 struct module_btf *mod_btf;
5638 struct btf *btf;
5639 char name[64];
5640 __u32 id = 0, len;
5641 int err, fd;
5642
5643 if (obj->btf_modules_loaded)
5644 return 0;
5645
5646 if (obj->gen_loader)
5647 return 0;
5648
5649 /* don't do this again, even if we find no module BTFs */
5650 obj->btf_modules_loaded = true;
5651
5652 /* kernel too old to support module BTFs */
5653 if (!kernel_supports(obj, FEAT_MODULE_BTF))
5654 return 0;
5655
5656 while (true) {
5657 err = bpf_btf_get_next_id(id, &id);
5658 if (err && errno == ENOENT)
5659 return 0;
5660 if (err && errno == EPERM) {
5661 pr_debug("skipping module BTFs loading, missing privileges\n");
5662 return 0;
5663 }
5664 if (err) {
5665 err = -errno;
5666 pr_warn("failed to iterate BTF objects: %s\n", errstr(err));
5667 return err;
5668 }
5669
5670 fd = bpf_btf_get_fd_by_id(id);
5671 if (fd < 0) {
5672 if (errno == ENOENT)
5673 continue; /* expected race: BTF was unloaded */
5674 err = -errno;
5675 pr_warn("failed to get BTF object #%d FD: %s\n", id, errstr(err));
5676 return err;
5677 }
5678
5679 len = sizeof(info);
5680 memset(&info, 0, sizeof(info));
5681 info.name = ptr_to_u64(name);
5682 info.name_len = sizeof(name);
5683
5684 err = bpf_btf_get_info_by_fd(fd, &info, &len);
5685 if (err) {
5686 err = -errno;
5687 pr_warn("failed to get BTF object #%d info: %s\n", id, errstr(err));
5688 goto err_out;
5689 }
5690
5691 /* ignore non-module BTFs */
5692 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5693 close(fd);
5694 continue;
5695 }
5696
5697 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5698 err = libbpf_get_error(btf);
5699 if (err) {
5700 pr_warn("failed to load module [%s]'s BTF object #%d: %s\n",
5701 name, id, errstr(err));
5702 goto err_out;
5703 }
5704
5705 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5706 sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5707 if (err)
5708 goto err_out;
5709
5710 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5711
5712 mod_btf->btf = btf;
5713 mod_btf->id = id;
5714 mod_btf->fd = fd;
5715 mod_btf->name = strdup(name);
5716 if (!mod_btf->name) {
5717 err = -ENOMEM;
5718 goto err_out;
5719 }
5720 continue;
5721
5722 err_out:
5723 close(fd);
5724 return err;
5725 }
5726
5727 return 0;
5728 }
5729
5730 static struct bpf_core_cand_list *
bpf_core_find_cands(struct bpf_object * obj,const struct btf * local_btf,__u32 local_type_id)5731 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5732 {
5733 struct bpf_core_cand local_cand = {};
5734 struct bpf_core_cand_list *cands;
5735 const struct btf *main_btf;
5736 const struct btf_type *local_t;
5737 const char *local_name;
5738 size_t local_essent_len;
5739 int err, i;
5740
5741 local_cand.btf = local_btf;
5742 local_cand.id = local_type_id;
5743 local_t = btf__type_by_id(local_btf, local_type_id);
5744 if (!local_t)
5745 return ERR_PTR(-EINVAL);
5746
5747 local_name = btf__name_by_offset(local_btf, local_t->name_off);
5748 if (str_is_empty(local_name))
5749 return ERR_PTR(-EINVAL);
5750 local_essent_len = bpf_core_essential_name_len(local_name);
5751
5752 cands = calloc(1, sizeof(*cands));
5753 if (!cands)
5754 return ERR_PTR(-ENOMEM);
5755
5756 /* Attempt to find target candidates in vmlinux BTF first */
5757 main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5758 err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5759 if (err)
5760 goto err_out;
5761
5762 /* if vmlinux BTF has any candidate, don't got for module BTFs */
5763 if (cands->len)
5764 return cands;
5765
5766 /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5767 if (obj->btf_vmlinux_override)
5768 return cands;
5769
5770 /* now look through module BTFs, trying to still find candidates */
5771 err = load_module_btfs(obj);
5772 if (err)
5773 goto err_out;
5774
5775 for (i = 0; i < obj->btf_module_cnt; i++) {
5776 err = bpf_core_add_cands(&local_cand, local_essent_len,
5777 obj->btf_modules[i].btf,
5778 obj->btf_modules[i].name,
5779 btf__type_cnt(obj->btf_vmlinux),
5780 cands);
5781 if (err)
5782 goto err_out;
5783 }
5784
5785 return cands;
5786 err_out:
5787 bpf_core_free_cands(cands);
5788 return ERR_PTR(err);
5789 }
5790
5791 /* Check local and target types for compatibility. This check is used for
5792 * type-based CO-RE relocations and follow slightly different rules than
5793 * field-based relocations. This function assumes that root types were already
5794 * checked for name match. Beyond that initial root-level name check, names
5795 * are completely ignored. Compatibility rules are as follows:
5796 * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5797 * kind should match for local and target types (i.e., STRUCT is not
5798 * compatible with UNION);
5799 * - for ENUMs, the size is ignored;
5800 * - for INT, size and signedness are ignored;
5801 * - for ARRAY, dimensionality is ignored, element types are checked for
5802 * compatibility recursively;
5803 * - CONST/VOLATILE/RESTRICT modifiers are ignored;
5804 * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5805 * - FUNC_PROTOs are compatible if they have compatible signature: same
5806 * number of input args and compatible return and argument types.
5807 * These rules are not set in stone and probably will be adjusted as we get
5808 * more experience with using BPF CO-RE relocations.
5809 */
bpf_core_types_are_compat(const struct btf * local_btf,__u32 local_id,const struct btf * targ_btf,__u32 targ_id)5810 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5811 const struct btf *targ_btf, __u32 targ_id)
5812 {
5813 return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32);
5814 }
5815
bpf_core_types_match(const struct btf * local_btf,__u32 local_id,const struct btf * targ_btf,__u32 targ_id)5816 int bpf_core_types_match(const struct btf *local_btf, __u32 local_id,
5817 const struct btf *targ_btf, __u32 targ_id)
5818 {
5819 return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32);
5820 }
5821
bpf_core_hash_fn(const long key,void * ctx)5822 static size_t bpf_core_hash_fn(const long key, void *ctx)
5823 {
5824 return key;
5825 }
5826
bpf_core_equal_fn(const long k1,const long k2,void * ctx)5827 static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx)
5828 {
5829 return k1 == k2;
5830 }
5831
record_relo_core(struct bpf_program * prog,const struct bpf_core_relo * core_relo,int insn_idx)5832 static int record_relo_core(struct bpf_program *prog,
5833 const struct bpf_core_relo *core_relo, int insn_idx)
5834 {
5835 struct reloc_desc *relos, *relo;
5836
5837 relos = libbpf_reallocarray(prog->reloc_desc,
5838 prog->nr_reloc + 1, sizeof(*relos));
5839 if (!relos)
5840 return -ENOMEM;
5841 relo = &relos[prog->nr_reloc];
5842 relo->type = RELO_CORE;
5843 relo->insn_idx = insn_idx;
5844 relo->core_relo = core_relo;
5845 prog->reloc_desc = relos;
5846 prog->nr_reloc++;
5847 return 0;
5848 }
5849
find_relo_core(struct bpf_program * prog,int insn_idx)5850 static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx)
5851 {
5852 struct reloc_desc *relo;
5853 int i;
5854
5855 for (i = 0; i < prog->nr_reloc; i++) {
5856 relo = &prog->reloc_desc[i];
5857 if (relo->type != RELO_CORE || relo->insn_idx != insn_idx)
5858 continue;
5859
5860 return relo->core_relo;
5861 }
5862
5863 return NULL;
5864 }
5865
bpf_core_resolve_relo(struct bpf_program * prog,const struct bpf_core_relo * relo,int relo_idx,const struct btf * local_btf,struct hashmap * cand_cache,struct bpf_core_relo_res * targ_res)5866 static int bpf_core_resolve_relo(struct bpf_program *prog,
5867 const struct bpf_core_relo *relo,
5868 int relo_idx,
5869 const struct btf *local_btf,
5870 struct hashmap *cand_cache,
5871 struct bpf_core_relo_res *targ_res)
5872 {
5873 struct bpf_core_spec specs_scratch[3] = {};
5874 struct bpf_core_cand_list *cands = NULL;
5875 const char *prog_name = prog->name;
5876 const struct btf_type *local_type;
5877 const char *local_name;
5878 __u32 local_id = relo->type_id;
5879 int err;
5880
5881 local_type = btf__type_by_id(local_btf, local_id);
5882 if (!local_type)
5883 return -EINVAL;
5884
5885 local_name = btf__name_by_offset(local_btf, local_type->name_off);
5886 if (!local_name)
5887 return -EINVAL;
5888
5889 if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
5890 !hashmap__find(cand_cache, local_id, &cands)) {
5891 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5892 if (IS_ERR(cands)) {
5893 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5894 prog_name, relo_idx, local_id, btf_kind_str(local_type),
5895 local_name, PTR_ERR(cands));
5896 return PTR_ERR(cands);
5897 }
5898 err = hashmap__set(cand_cache, local_id, cands, NULL, NULL);
5899 if (err) {
5900 bpf_core_free_cands(cands);
5901 return err;
5902 }
5903 }
5904
5905 return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
5906 targ_res);
5907 }
5908
5909 static int
bpf_object__relocate_core(struct bpf_object * obj,const char * targ_btf_path)5910 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5911 {
5912 const struct btf_ext_info_sec *sec;
5913 struct bpf_core_relo_res targ_res;
5914 const struct bpf_core_relo *rec;
5915 const struct btf_ext_info *seg;
5916 struct hashmap_entry *entry;
5917 struct hashmap *cand_cache = NULL;
5918 struct bpf_program *prog;
5919 struct bpf_insn *insn;
5920 const char *sec_name;
5921 int i, err = 0, insn_idx, sec_idx, sec_num;
5922
5923 if (obj->btf_ext->core_relo_info.len == 0)
5924 return 0;
5925
5926 if (targ_btf_path) {
5927 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5928 err = libbpf_get_error(obj->btf_vmlinux_override);
5929 if (err) {
5930 pr_warn("failed to parse target BTF: %s\n", errstr(err));
5931 return err;
5932 }
5933 }
5934
5935 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5936 if (IS_ERR(cand_cache)) {
5937 err = PTR_ERR(cand_cache);
5938 goto out;
5939 }
5940
5941 seg = &obj->btf_ext->core_relo_info;
5942 sec_num = 0;
5943 for_each_btf_ext_sec(seg, sec) {
5944 sec_idx = seg->sec_idxs[sec_num];
5945 sec_num++;
5946
5947 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5948 if (str_is_empty(sec_name)) {
5949 err = -EINVAL;
5950 goto out;
5951 }
5952
5953 pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info);
5954
5955 for_each_btf_ext_rec(seg, sec, i, rec) {
5956 if (rec->insn_off % BPF_INSN_SZ)
5957 return -EINVAL;
5958 insn_idx = rec->insn_off / BPF_INSN_SZ;
5959 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5960 if (!prog) {
5961 /* When __weak subprog is "overridden" by another instance
5962 * of the subprog from a different object file, linker still
5963 * appends all the .BTF.ext info that used to belong to that
5964 * eliminated subprogram.
5965 * This is similar to what x86-64 linker does for relocations.
5966 * So just ignore such relocations just like we ignore
5967 * subprog instructions when discovering subprograms.
5968 */
5969 pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n",
5970 sec_name, i, insn_idx);
5971 continue;
5972 }
5973 /* no need to apply CO-RE relocation if the program is
5974 * not going to be loaded
5975 */
5976 if (!prog->autoload)
5977 continue;
5978
5979 /* adjust insn_idx from section frame of reference to the local
5980 * program's frame of reference; (sub-)program code is not yet
5981 * relocated, so it's enough to just subtract in-section offset
5982 */
5983 insn_idx = insn_idx - prog->sec_insn_off;
5984 if (insn_idx >= prog->insns_cnt)
5985 return -EINVAL;
5986 insn = &prog->insns[insn_idx];
5987
5988 err = record_relo_core(prog, rec, insn_idx);
5989 if (err) {
5990 pr_warn("prog '%s': relo #%d: failed to record relocation: %s\n",
5991 prog->name, i, errstr(err));
5992 goto out;
5993 }
5994
5995 if (prog->obj->gen_loader)
5996 continue;
5997
5998 err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
5999 if (err) {
6000 pr_warn("prog '%s': relo #%d: failed to relocate: %s\n",
6001 prog->name, i, errstr(err));
6002 goto out;
6003 }
6004
6005 err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
6006 if (err) {
6007 pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %s\n",
6008 prog->name, i, insn_idx, errstr(err));
6009 goto out;
6010 }
6011 }
6012 }
6013
6014 out:
6015 /* obj->btf_vmlinux and module BTFs are freed after object load */
6016 btf__free(obj->btf_vmlinux_override);
6017 obj->btf_vmlinux_override = NULL;
6018
6019 if (!IS_ERR_OR_NULL(cand_cache)) {
6020 hashmap__for_each_entry(cand_cache, entry, i) {
6021 bpf_core_free_cands(entry->pvalue);
6022 }
6023 hashmap__free(cand_cache);
6024 }
6025 return err;
6026 }
6027
6028 /* base map load ldimm64 special constant, used also for log fixup logic */
6029 #define POISON_LDIMM64_MAP_BASE 2001000000
6030 #define POISON_LDIMM64_MAP_PFX "200100"
6031
poison_map_ldimm64(struct bpf_program * prog,int relo_idx,int insn_idx,struct bpf_insn * insn,int map_idx,const struct bpf_map * map)6032 static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx,
6033 int insn_idx, struct bpf_insn *insn,
6034 int map_idx, const struct bpf_map *map)
6035 {
6036 int i;
6037
6038 pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n",
6039 prog->name, relo_idx, insn_idx, map_idx, map->name);
6040
6041 /* we turn single ldimm64 into two identical invalid calls */
6042 for (i = 0; i < 2; i++) {
6043 insn->code = BPF_JMP | BPF_CALL;
6044 insn->dst_reg = 0;
6045 insn->src_reg = 0;
6046 insn->off = 0;
6047 /* if this instruction is reachable (not a dead code),
6048 * verifier will complain with something like:
6049 * invalid func unknown#2001000123
6050 * where lower 123 is map index into obj->maps[] array
6051 */
6052 insn->imm = POISON_LDIMM64_MAP_BASE + map_idx;
6053
6054 insn++;
6055 }
6056 }
6057
6058 /* unresolved kfunc call special constant, used also for log fixup logic */
6059 #define POISON_CALL_KFUNC_BASE 2002000000
6060 #define POISON_CALL_KFUNC_PFX "2002"
6061
poison_kfunc_call(struct bpf_program * prog,int relo_idx,int insn_idx,struct bpf_insn * insn,int ext_idx,const struct extern_desc * ext)6062 static void poison_kfunc_call(struct bpf_program *prog, int relo_idx,
6063 int insn_idx, struct bpf_insn *insn,
6064 int ext_idx, const struct extern_desc *ext)
6065 {
6066 pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n",
6067 prog->name, relo_idx, insn_idx, ext->name);
6068
6069 /* we turn kfunc call into invalid helper call with identifiable constant */
6070 insn->code = BPF_JMP | BPF_CALL;
6071 insn->dst_reg = 0;
6072 insn->src_reg = 0;
6073 insn->off = 0;
6074 /* if this instruction is reachable (not a dead code),
6075 * verifier will complain with something like:
6076 * invalid func unknown#2001000123
6077 * where lower 123 is extern index into obj->externs[] array
6078 */
6079 insn->imm = POISON_CALL_KFUNC_BASE + ext_idx;
6080 }
6081
6082 /* Relocate data references within program code:
6083 * - map references;
6084 * - global variable references;
6085 * - extern references.
6086 */
6087 static int
bpf_object__relocate_data(struct bpf_object * obj,struct bpf_program * prog)6088 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
6089 {
6090 int i;
6091
6092 for (i = 0; i < prog->nr_reloc; i++) {
6093 struct reloc_desc *relo = &prog->reloc_desc[i];
6094 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6095 const struct bpf_map *map;
6096 struct extern_desc *ext;
6097
6098 switch (relo->type) {
6099 case RELO_LD64:
6100 map = &obj->maps[relo->map_idx];
6101 if (obj->gen_loader) {
6102 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
6103 insn[0].imm = relo->map_idx;
6104 } else if (map->autocreate) {
6105 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
6106 insn[0].imm = map->fd;
6107 } else {
6108 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6109 relo->map_idx, map);
6110 }
6111 break;
6112 case RELO_DATA:
6113 map = &obj->maps[relo->map_idx];
6114 insn[1].imm = insn[0].imm + relo->sym_off;
6115 if (obj->gen_loader) {
6116 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6117 insn[0].imm = relo->map_idx;
6118 } else if (map->autocreate) {
6119 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6120 insn[0].imm = map->fd;
6121 } else {
6122 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6123 relo->map_idx, map);
6124 }
6125 break;
6126 case RELO_EXTERN_LD64:
6127 ext = &obj->externs[relo->ext_idx];
6128 if (ext->type == EXT_KCFG) {
6129 if (obj->gen_loader) {
6130 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6131 insn[0].imm = obj->kconfig_map_idx;
6132 } else {
6133 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6134 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
6135 }
6136 insn[1].imm = ext->kcfg.data_off;
6137 } else /* EXT_KSYM */ {
6138 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
6139 insn[0].src_reg = BPF_PSEUDO_BTF_ID;
6140 insn[0].imm = ext->ksym.kernel_btf_id;
6141 insn[1].imm = ext->ksym.kernel_btf_obj_fd;
6142 } else { /* typeless ksyms or unresolved typed ksyms */
6143 insn[0].imm = (__u32)ext->ksym.addr;
6144 insn[1].imm = ext->ksym.addr >> 32;
6145 }
6146 }
6147 break;
6148 case RELO_EXTERN_CALL:
6149 ext = &obj->externs[relo->ext_idx];
6150 insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
6151 if (ext->is_set) {
6152 insn[0].imm = ext->ksym.kernel_btf_id;
6153 insn[0].off = ext->ksym.btf_fd_idx;
6154 } else { /* unresolved weak kfunc call */
6155 poison_kfunc_call(prog, i, relo->insn_idx, insn,
6156 relo->ext_idx, ext);
6157 }
6158 break;
6159 case RELO_SUBPROG_ADDR:
6160 if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
6161 pr_warn("prog '%s': relo #%d: bad insn\n",
6162 prog->name, i);
6163 return -EINVAL;
6164 }
6165 /* handled already */
6166 break;
6167 case RELO_CALL:
6168 /* handled already */
6169 break;
6170 case RELO_CORE:
6171 /* will be handled by bpf_program_record_relos() */
6172 break;
6173 default:
6174 pr_warn("prog '%s': relo #%d: bad relo type %d\n",
6175 prog->name, i, relo->type);
6176 return -EINVAL;
6177 }
6178 }
6179
6180 return 0;
6181 }
6182
adjust_prog_btf_ext_info(const struct bpf_object * obj,const struct bpf_program * prog,const struct btf_ext_info * ext_info,void ** prog_info,__u32 * prog_rec_cnt,__u32 * prog_rec_sz)6183 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
6184 const struct bpf_program *prog,
6185 const struct btf_ext_info *ext_info,
6186 void **prog_info, __u32 *prog_rec_cnt,
6187 __u32 *prog_rec_sz)
6188 {
6189 void *copy_start = NULL, *copy_end = NULL;
6190 void *rec, *rec_end, *new_prog_info;
6191 const struct btf_ext_info_sec *sec;
6192 size_t old_sz, new_sz;
6193 int i, sec_num, sec_idx, off_adj;
6194
6195 sec_num = 0;
6196 for_each_btf_ext_sec(ext_info, sec) {
6197 sec_idx = ext_info->sec_idxs[sec_num];
6198 sec_num++;
6199 if (prog->sec_idx != sec_idx)
6200 continue;
6201
6202 for_each_btf_ext_rec(ext_info, sec, i, rec) {
6203 __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
6204
6205 if (insn_off < prog->sec_insn_off)
6206 continue;
6207 if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6208 break;
6209
6210 if (!copy_start)
6211 copy_start = rec;
6212 copy_end = rec + ext_info->rec_size;
6213 }
6214
6215 if (!copy_start)
6216 return -ENOENT;
6217
6218 /* append func/line info of a given (sub-)program to the main
6219 * program func/line info
6220 */
6221 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6222 new_sz = old_sz + (copy_end - copy_start);
6223 new_prog_info = realloc(*prog_info, new_sz);
6224 if (!new_prog_info)
6225 return -ENOMEM;
6226 *prog_info = new_prog_info;
6227 *prog_rec_cnt = new_sz / ext_info->rec_size;
6228 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6229
6230 /* Kernel instruction offsets are in units of 8-byte
6231 * instructions, while .BTF.ext instruction offsets generated
6232 * by Clang are in units of bytes. So convert Clang offsets
6233 * into kernel offsets and adjust offset according to program
6234 * relocated position.
6235 */
6236 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6237 rec = new_prog_info + old_sz;
6238 rec_end = new_prog_info + new_sz;
6239 for (; rec < rec_end; rec += ext_info->rec_size) {
6240 __u32 *insn_off = rec;
6241
6242 *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6243 }
6244 *prog_rec_sz = ext_info->rec_size;
6245 return 0;
6246 }
6247
6248 return -ENOENT;
6249 }
6250
6251 static int
reloc_prog_func_and_line_info(const struct bpf_object * obj,struct bpf_program * main_prog,const struct bpf_program * prog)6252 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6253 struct bpf_program *main_prog,
6254 const struct bpf_program *prog)
6255 {
6256 int err;
6257
6258 /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6259 * support func/line info
6260 */
6261 if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
6262 return 0;
6263
6264 /* only attempt func info relocation if main program's func_info
6265 * relocation was successful
6266 */
6267 if (main_prog != prog && !main_prog->func_info)
6268 goto line_info;
6269
6270 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6271 &main_prog->func_info,
6272 &main_prog->func_info_cnt,
6273 &main_prog->func_info_rec_size);
6274 if (err) {
6275 if (err != -ENOENT) {
6276 pr_warn("prog '%s': error relocating .BTF.ext function info: %s\n",
6277 prog->name, errstr(err));
6278 return err;
6279 }
6280 if (main_prog->func_info) {
6281 /*
6282 * Some info has already been found but has problem
6283 * in the last btf_ext reloc. Must have to error out.
6284 */
6285 pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6286 return err;
6287 }
6288 /* Have problem loading the very first info. Ignore the rest. */
6289 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6290 prog->name);
6291 }
6292
6293 line_info:
6294 /* don't relocate line info if main program's relocation failed */
6295 if (main_prog != prog && !main_prog->line_info)
6296 return 0;
6297
6298 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6299 &main_prog->line_info,
6300 &main_prog->line_info_cnt,
6301 &main_prog->line_info_rec_size);
6302 if (err) {
6303 if (err != -ENOENT) {
6304 pr_warn("prog '%s': error relocating .BTF.ext line info: %s\n",
6305 prog->name, errstr(err));
6306 return err;
6307 }
6308 if (main_prog->line_info) {
6309 /*
6310 * Some info has already been found but has problem
6311 * in the last btf_ext reloc. Must have to error out.
6312 */
6313 pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6314 return err;
6315 }
6316 /* Have problem loading the very first info. Ignore the rest. */
6317 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6318 prog->name);
6319 }
6320 return 0;
6321 }
6322
cmp_relo_by_insn_idx(const void * key,const void * elem)6323 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6324 {
6325 size_t insn_idx = *(const size_t *)key;
6326 const struct reloc_desc *relo = elem;
6327
6328 if (insn_idx == relo->insn_idx)
6329 return 0;
6330 return insn_idx < relo->insn_idx ? -1 : 1;
6331 }
6332
find_prog_insn_relo(const struct bpf_program * prog,size_t insn_idx)6333 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6334 {
6335 if (!prog->nr_reloc)
6336 return NULL;
6337 return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6338 sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6339 }
6340
append_subprog_relos(struct bpf_program * main_prog,struct bpf_program * subprog)6341 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
6342 {
6343 int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
6344 struct reloc_desc *relos;
6345 int i;
6346
6347 if (main_prog == subprog)
6348 return 0;
6349 relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
6350 /* if new count is zero, reallocarray can return a valid NULL result;
6351 * in this case the previous pointer will be freed, so we *have to*
6352 * reassign old pointer to the new value (even if it's NULL)
6353 */
6354 if (!relos && new_cnt)
6355 return -ENOMEM;
6356 if (subprog->nr_reloc)
6357 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
6358 sizeof(*relos) * subprog->nr_reloc);
6359
6360 for (i = main_prog->nr_reloc; i < new_cnt; i++)
6361 relos[i].insn_idx += subprog->sub_insn_off;
6362 /* After insn_idx adjustment the 'relos' array is still sorted
6363 * by insn_idx and doesn't break bsearch.
6364 */
6365 main_prog->reloc_desc = relos;
6366 main_prog->nr_reloc = new_cnt;
6367 return 0;
6368 }
6369
6370 static int
bpf_object__append_subprog_code(struct bpf_object * obj,struct bpf_program * main_prog,struct bpf_program * subprog)6371 bpf_object__append_subprog_code(struct bpf_object *obj, struct bpf_program *main_prog,
6372 struct bpf_program *subprog)
6373 {
6374 struct bpf_insn *insns;
6375 size_t new_cnt;
6376 int err;
6377
6378 subprog->sub_insn_off = main_prog->insns_cnt;
6379
6380 new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6381 insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6382 if (!insns) {
6383 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6384 return -ENOMEM;
6385 }
6386 main_prog->insns = insns;
6387 main_prog->insns_cnt = new_cnt;
6388
6389 memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6390 subprog->insns_cnt * sizeof(*insns));
6391
6392 pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6393 main_prog->name, subprog->insns_cnt, subprog->name);
6394
6395 /* The subprog insns are now appended. Append its relos too. */
6396 err = append_subprog_relos(main_prog, subprog);
6397 if (err)
6398 return err;
6399 return 0;
6400 }
6401
6402 static int
bpf_object__reloc_code(struct bpf_object * obj,struct bpf_program * main_prog,struct bpf_program * prog)6403 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6404 struct bpf_program *prog)
6405 {
6406 size_t sub_insn_idx, insn_idx;
6407 struct bpf_program *subprog;
6408 struct reloc_desc *relo;
6409 struct bpf_insn *insn;
6410 int err;
6411
6412 err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6413 if (err)
6414 return err;
6415
6416 for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6417 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6418 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6419 continue;
6420
6421 relo = find_prog_insn_relo(prog, insn_idx);
6422 if (relo && relo->type == RELO_EXTERN_CALL)
6423 /* kfunc relocations will be handled later
6424 * in bpf_object__relocate_data()
6425 */
6426 continue;
6427 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6428 pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6429 prog->name, insn_idx, relo->type);
6430 return -LIBBPF_ERRNO__RELOC;
6431 }
6432 if (relo) {
6433 /* sub-program instruction index is a combination of
6434 * an offset of a symbol pointed to by relocation and
6435 * call instruction's imm field; for global functions,
6436 * call always has imm = -1, but for static functions
6437 * relocation is against STT_SECTION and insn->imm
6438 * points to a start of a static function
6439 *
6440 * for subprog addr relocation, the relo->sym_off + insn->imm is
6441 * the byte offset in the corresponding section.
6442 */
6443 if (relo->type == RELO_CALL)
6444 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6445 else
6446 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6447 } else if (insn_is_pseudo_func(insn)) {
6448 /*
6449 * RELO_SUBPROG_ADDR relo is always emitted even if both
6450 * functions are in the same section, so it shouldn't reach here.
6451 */
6452 pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6453 prog->name, insn_idx);
6454 return -LIBBPF_ERRNO__RELOC;
6455 } else {
6456 /* if subprogram call is to a static function within
6457 * the same ELF section, there won't be any relocation
6458 * emitted, but it also means there is no additional
6459 * offset necessary, insns->imm is relative to
6460 * instruction's original position within the section
6461 */
6462 sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6463 }
6464
6465 /* we enforce that sub-programs should be in .text section */
6466 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6467 if (!subprog) {
6468 pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6469 prog->name);
6470 return -LIBBPF_ERRNO__RELOC;
6471 }
6472
6473 /* if it's the first call instruction calling into this
6474 * subprogram (meaning this subprog hasn't been processed
6475 * yet) within the context of current main program:
6476 * - append it at the end of main program's instructions blog;
6477 * - process is recursively, while current program is put on hold;
6478 * - if that subprogram calls some other not yet processes
6479 * subprogram, same thing will happen recursively until
6480 * there are no more unprocesses subprograms left to append
6481 * and relocate.
6482 */
6483 if (subprog->sub_insn_off == 0) {
6484 err = bpf_object__append_subprog_code(obj, main_prog, subprog);
6485 if (err)
6486 return err;
6487 err = bpf_object__reloc_code(obj, main_prog, subprog);
6488 if (err)
6489 return err;
6490 }
6491
6492 /* main_prog->insns memory could have been re-allocated, so
6493 * calculate pointer again
6494 */
6495 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6496 /* calculate correct instruction position within current main
6497 * prog; each main prog can have a different set of
6498 * subprograms appended (potentially in different order as
6499 * well), so position of any subprog can be different for
6500 * different main programs
6501 */
6502 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6503
6504 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6505 prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6506 }
6507
6508 return 0;
6509 }
6510
6511 /*
6512 * Relocate sub-program calls.
6513 *
6514 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6515 * main prog) is processed separately. For each subprog (non-entry functions,
6516 * that can be called from either entry progs or other subprogs) gets their
6517 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6518 * hasn't been yet appended and relocated within current main prog. Once its
6519 * relocated, sub_insn_off will point at the position within current main prog
6520 * where given subprog was appended. This will further be used to relocate all
6521 * the call instructions jumping into this subprog.
6522 *
6523 * We start with main program and process all call instructions. If the call
6524 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6525 * is zero), subprog instructions are appended at the end of main program's
6526 * instruction array. Then main program is "put on hold" while we recursively
6527 * process newly appended subprogram. If that subprogram calls into another
6528 * subprogram that hasn't been appended, new subprogram is appended again to
6529 * the *main* prog's instructions (subprog's instructions are always left
6530 * untouched, as they need to be in unmodified state for subsequent main progs
6531 * and subprog instructions are always sent only as part of a main prog) and
6532 * the process continues recursively. Once all the subprogs called from a main
6533 * prog or any of its subprogs are appended (and relocated), all their
6534 * positions within finalized instructions array are known, so it's easy to
6535 * rewrite call instructions with correct relative offsets, corresponding to
6536 * desired target subprog.
6537 *
6538 * Its important to realize that some subprogs might not be called from some
6539 * main prog and any of its called/used subprogs. Those will keep their
6540 * subprog->sub_insn_off as zero at all times and won't be appended to current
6541 * main prog and won't be relocated within the context of current main prog.
6542 * They might still be used from other main progs later.
6543 *
6544 * Visually this process can be shown as below. Suppose we have two main
6545 * programs mainA and mainB and BPF object contains three subprogs: subA,
6546 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6547 * subC both call subB:
6548 *
6549 * +--------+ +-------+
6550 * | v v |
6551 * +--+---+ +--+-+-+ +---+--+
6552 * | subA | | subB | | subC |
6553 * +--+---+ +------+ +---+--+
6554 * ^ ^
6555 * | |
6556 * +---+-------+ +------+----+
6557 * | mainA | | mainB |
6558 * +-----------+ +-----------+
6559 *
6560 * We'll start relocating mainA, will find subA, append it and start
6561 * processing sub A recursively:
6562 *
6563 * +-----------+------+
6564 * | mainA | subA |
6565 * +-----------+------+
6566 *
6567 * At this point we notice that subB is used from subA, so we append it and
6568 * relocate (there are no further subcalls from subB):
6569 *
6570 * +-----------+------+------+
6571 * | mainA | subA | subB |
6572 * +-----------+------+------+
6573 *
6574 * At this point, we relocate subA calls, then go one level up and finish with
6575 * relocatin mainA calls. mainA is done.
6576 *
6577 * For mainB process is similar but results in different order. We start with
6578 * mainB and skip subA and subB, as mainB never calls them (at least
6579 * directly), but we see subC is needed, so we append and start processing it:
6580 *
6581 * +-----------+------+
6582 * | mainB | subC |
6583 * +-----------+------+
6584 * Now we see subC needs subB, so we go back to it, append and relocate it:
6585 *
6586 * +-----------+------+------+
6587 * | mainB | subC | subB |
6588 * +-----------+------+------+
6589 *
6590 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6591 */
6592 static int
bpf_object__relocate_calls(struct bpf_object * obj,struct bpf_program * prog)6593 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6594 {
6595 struct bpf_program *subprog;
6596 int i, err;
6597
6598 /* mark all subprogs as not relocated (yet) within the context of
6599 * current main program
6600 */
6601 for (i = 0; i < obj->nr_programs; i++) {
6602 subprog = &obj->programs[i];
6603 if (!prog_is_subprog(obj, subprog))
6604 continue;
6605
6606 subprog->sub_insn_off = 0;
6607 }
6608
6609 err = bpf_object__reloc_code(obj, prog, prog);
6610 if (err)
6611 return err;
6612
6613 return 0;
6614 }
6615
6616 static void
bpf_object__free_relocs(struct bpf_object * obj)6617 bpf_object__free_relocs(struct bpf_object *obj)
6618 {
6619 struct bpf_program *prog;
6620 int i;
6621
6622 /* free up relocation descriptors */
6623 for (i = 0; i < obj->nr_programs; i++) {
6624 prog = &obj->programs[i];
6625 zfree(&prog->reloc_desc);
6626 prog->nr_reloc = 0;
6627 }
6628 }
6629
cmp_relocs(const void * _a,const void * _b)6630 static int cmp_relocs(const void *_a, const void *_b)
6631 {
6632 const struct reloc_desc *a = _a;
6633 const struct reloc_desc *b = _b;
6634
6635 if (a->insn_idx != b->insn_idx)
6636 return a->insn_idx < b->insn_idx ? -1 : 1;
6637
6638 /* no two relocations should have the same insn_idx, but ... */
6639 if (a->type != b->type)
6640 return a->type < b->type ? -1 : 1;
6641
6642 return 0;
6643 }
6644
bpf_object__sort_relos(struct bpf_object * obj)6645 static void bpf_object__sort_relos(struct bpf_object *obj)
6646 {
6647 int i;
6648
6649 for (i = 0; i < obj->nr_programs; i++) {
6650 struct bpf_program *p = &obj->programs[i];
6651
6652 if (!p->nr_reloc)
6653 continue;
6654
6655 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6656 }
6657 }
6658
bpf_prog_assign_exc_cb(struct bpf_object * obj,struct bpf_program * prog)6659 static int bpf_prog_assign_exc_cb(struct bpf_object *obj, struct bpf_program *prog)
6660 {
6661 const char *str = "exception_callback:";
6662 size_t pfx_len = strlen(str);
6663 int i, j, n;
6664
6665 if (!obj->btf || !kernel_supports(obj, FEAT_BTF_DECL_TAG))
6666 return 0;
6667
6668 n = btf__type_cnt(obj->btf);
6669 for (i = 1; i < n; i++) {
6670 const char *name;
6671 struct btf_type *t;
6672
6673 t = btf_type_by_id(obj->btf, i);
6674 if (!btf_is_decl_tag(t) || btf_decl_tag(t)->component_idx != -1)
6675 continue;
6676
6677 name = btf__str_by_offset(obj->btf, t->name_off);
6678 if (strncmp(name, str, pfx_len) != 0)
6679 continue;
6680
6681 t = btf_type_by_id(obj->btf, t->type);
6682 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) {
6683 pr_warn("prog '%s': exception_callback:<value> decl tag not applied to the main program\n",
6684 prog->name);
6685 return -EINVAL;
6686 }
6687 if (strcmp(prog->name, btf__str_by_offset(obj->btf, t->name_off)) != 0)
6688 continue;
6689 /* Multiple callbacks are specified for the same prog,
6690 * the verifier will eventually return an error for this
6691 * case, hence simply skip appending a subprog.
6692 */
6693 if (prog->exception_cb_idx >= 0) {
6694 prog->exception_cb_idx = -1;
6695 break;
6696 }
6697
6698 name += pfx_len;
6699 if (str_is_empty(name)) {
6700 pr_warn("prog '%s': exception_callback:<value> decl tag contains empty value\n",
6701 prog->name);
6702 return -EINVAL;
6703 }
6704
6705 for (j = 0; j < obj->nr_programs; j++) {
6706 struct bpf_program *subprog = &obj->programs[j];
6707
6708 if (!prog_is_subprog(obj, subprog))
6709 continue;
6710 if (strcmp(name, subprog->name) != 0)
6711 continue;
6712 /* Enforce non-hidden, as from verifier point of
6713 * view it expects global functions, whereas the
6714 * mark_btf_static fixes up linkage as static.
6715 */
6716 if (!subprog->sym_global || subprog->mark_btf_static) {
6717 pr_warn("prog '%s': exception callback %s must be a global non-hidden function\n",
6718 prog->name, subprog->name);
6719 return -EINVAL;
6720 }
6721 /* Let's see if we already saw a static exception callback with the same name */
6722 if (prog->exception_cb_idx >= 0) {
6723 pr_warn("prog '%s': multiple subprogs with same name as exception callback '%s'\n",
6724 prog->name, subprog->name);
6725 return -EINVAL;
6726 }
6727 prog->exception_cb_idx = j;
6728 break;
6729 }
6730
6731 if (prog->exception_cb_idx >= 0)
6732 continue;
6733
6734 pr_warn("prog '%s': cannot find exception callback '%s'\n", prog->name, name);
6735 return -ENOENT;
6736 }
6737
6738 return 0;
6739 }
6740
6741 static struct {
6742 enum bpf_prog_type prog_type;
6743 const char *ctx_name;
6744 } global_ctx_map[] = {
6745 { BPF_PROG_TYPE_CGROUP_DEVICE, "bpf_cgroup_dev_ctx" },
6746 { BPF_PROG_TYPE_CGROUP_SKB, "__sk_buff" },
6747 { BPF_PROG_TYPE_CGROUP_SOCK, "bpf_sock" },
6748 { BPF_PROG_TYPE_CGROUP_SOCK_ADDR, "bpf_sock_addr" },
6749 { BPF_PROG_TYPE_CGROUP_SOCKOPT, "bpf_sockopt" },
6750 { BPF_PROG_TYPE_CGROUP_SYSCTL, "bpf_sysctl" },
6751 { BPF_PROG_TYPE_FLOW_DISSECTOR, "__sk_buff" },
6752 { BPF_PROG_TYPE_KPROBE, "bpf_user_pt_regs_t" },
6753 { BPF_PROG_TYPE_LWT_IN, "__sk_buff" },
6754 { BPF_PROG_TYPE_LWT_OUT, "__sk_buff" },
6755 { BPF_PROG_TYPE_LWT_SEG6LOCAL, "__sk_buff" },
6756 { BPF_PROG_TYPE_LWT_XMIT, "__sk_buff" },
6757 { BPF_PROG_TYPE_NETFILTER, "bpf_nf_ctx" },
6758 { BPF_PROG_TYPE_PERF_EVENT, "bpf_perf_event_data" },
6759 { BPF_PROG_TYPE_RAW_TRACEPOINT, "bpf_raw_tracepoint_args" },
6760 { BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, "bpf_raw_tracepoint_args" },
6761 { BPF_PROG_TYPE_SCHED_ACT, "__sk_buff" },
6762 { BPF_PROG_TYPE_SCHED_CLS, "__sk_buff" },
6763 { BPF_PROG_TYPE_SK_LOOKUP, "bpf_sk_lookup" },
6764 { BPF_PROG_TYPE_SK_MSG, "sk_msg_md" },
6765 { BPF_PROG_TYPE_SK_REUSEPORT, "sk_reuseport_md" },
6766 { BPF_PROG_TYPE_SK_SKB, "__sk_buff" },
6767 { BPF_PROG_TYPE_SOCK_OPS, "bpf_sock_ops" },
6768 { BPF_PROG_TYPE_SOCKET_FILTER, "__sk_buff" },
6769 { BPF_PROG_TYPE_XDP, "xdp_md" },
6770 /* all other program types don't have "named" context structs */
6771 };
6772
6773 /* forward declarations for arch-specific underlying types of bpf_user_pt_regs_t typedef,
6774 * for below __builtin_types_compatible_p() checks;
6775 * with this approach we don't need any extra arch-specific #ifdef guards
6776 */
6777 struct pt_regs;
6778 struct user_pt_regs;
6779 struct user_regs_struct;
6780
need_func_arg_type_fixup(const struct btf * btf,const struct bpf_program * prog,const char * subprog_name,int arg_idx,int arg_type_id,const char * ctx_name)6781 static bool need_func_arg_type_fixup(const struct btf *btf, const struct bpf_program *prog,
6782 const char *subprog_name, int arg_idx,
6783 int arg_type_id, const char *ctx_name)
6784 {
6785 const struct btf_type *t;
6786 const char *tname;
6787
6788 /* check if existing parameter already matches verifier expectations */
6789 t = skip_mods_and_typedefs(btf, arg_type_id, NULL);
6790 if (!btf_is_ptr(t))
6791 goto out_warn;
6792
6793 /* typedef bpf_user_pt_regs_t is a special PITA case, valid for kprobe
6794 * and perf_event programs, so check this case early on and forget
6795 * about it for subsequent checks
6796 */
6797 while (btf_is_mod(t))
6798 t = btf__type_by_id(btf, t->type);
6799 if (btf_is_typedef(t) &&
6800 (prog->type == BPF_PROG_TYPE_KPROBE || prog->type == BPF_PROG_TYPE_PERF_EVENT)) {
6801 tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>";
6802 if (strcmp(tname, "bpf_user_pt_regs_t") == 0)
6803 return false; /* canonical type for kprobe/perf_event */
6804 }
6805
6806 /* now we can ignore typedefs moving forward */
6807 t = skip_mods_and_typedefs(btf, t->type, NULL);
6808
6809 /* if it's `void *`, definitely fix up BTF info */
6810 if (btf_is_void(t))
6811 return true;
6812
6813 /* if it's already proper canonical type, no need to fix up */
6814 tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>";
6815 if (btf_is_struct(t) && strcmp(tname, ctx_name) == 0)
6816 return false;
6817
6818 /* special cases */
6819 switch (prog->type) {
6820 case BPF_PROG_TYPE_KPROBE:
6821 /* `struct pt_regs *` is expected, but we need to fix up */
6822 if (btf_is_struct(t) && strcmp(tname, "pt_regs") == 0)
6823 return true;
6824 break;
6825 case BPF_PROG_TYPE_PERF_EVENT:
6826 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct pt_regs) &&
6827 btf_is_struct(t) && strcmp(tname, "pt_regs") == 0)
6828 return true;
6829 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_pt_regs) &&
6830 btf_is_struct(t) && strcmp(tname, "user_pt_regs") == 0)
6831 return true;
6832 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_regs_struct) &&
6833 btf_is_struct(t) && strcmp(tname, "user_regs_struct") == 0)
6834 return true;
6835 break;
6836 case BPF_PROG_TYPE_RAW_TRACEPOINT:
6837 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
6838 /* allow u64* as ctx */
6839 if (btf_is_int(t) && t->size == 8)
6840 return true;
6841 break;
6842 default:
6843 break;
6844 }
6845
6846 out_warn:
6847 pr_warn("prog '%s': subprog '%s' arg#%d is expected to be of `struct %s *` type\n",
6848 prog->name, subprog_name, arg_idx, ctx_name);
6849 return false;
6850 }
6851
clone_func_btf_info(struct btf * btf,int orig_fn_id,struct bpf_program * prog)6852 static int clone_func_btf_info(struct btf *btf, int orig_fn_id, struct bpf_program *prog)
6853 {
6854 int fn_id, fn_proto_id, ret_type_id, orig_proto_id;
6855 int i, err, arg_cnt, fn_name_off, linkage;
6856 struct btf_type *fn_t, *fn_proto_t, *t;
6857 struct btf_param *p;
6858
6859 /* caller already validated FUNC -> FUNC_PROTO validity */
6860 fn_t = btf_type_by_id(btf, orig_fn_id);
6861 fn_proto_t = btf_type_by_id(btf, fn_t->type);
6862
6863 /* Note that each btf__add_xxx() operation invalidates
6864 * all btf_type and string pointers, so we need to be
6865 * very careful when cloning BTF types. BTF type
6866 * pointers have to be always refetched. And to avoid
6867 * problems with invalidated string pointers, we
6868 * add empty strings initially, then just fix up
6869 * name_off offsets in place. Offsets are stable for
6870 * existing strings, so that works out.
6871 */
6872 fn_name_off = fn_t->name_off; /* we are about to invalidate fn_t */
6873 linkage = btf_func_linkage(fn_t);
6874 orig_proto_id = fn_t->type; /* original FUNC_PROTO ID */
6875 ret_type_id = fn_proto_t->type; /* fn_proto_t will be invalidated */
6876 arg_cnt = btf_vlen(fn_proto_t);
6877
6878 /* clone FUNC_PROTO and its params */
6879 fn_proto_id = btf__add_func_proto(btf, ret_type_id);
6880 if (fn_proto_id < 0)
6881 return -EINVAL;
6882
6883 for (i = 0; i < arg_cnt; i++) {
6884 int name_off;
6885
6886 /* copy original parameter data */
6887 t = btf_type_by_id(btf, orig_proto_id);
6888 p = &btf_params(t)[i];
6889 name_off = p->name_off;
6890
6891 err = btf__add_func_param(btf, "", p->type);
6892 if (err)
6893 return err;
6894
6895 fn_proto_t = btf_type_by_id(btf, fn_proto_id);
6896 p = &btf_params(fn_proto_t)[i];
6897 p->name_off = name_off; /* use remembered str offset */
6898 }
6899
6900 /* clone FUNC now, btf__add_func() enforces non-empty name, so use
6901 * entry program's name as a placeholder, which we replace immediately
6902 * with original name_off
6903 */
6904 fn_id = btf__add_func(btf, prog->name, linkage, fn_proto_id);
6905 if (fn_id < 0)
6906 return -EINVAL;
6907
6908 fn_t = btf_type_by_id(btf, fn_id);
6909 fn_t->name_off = fn_name_off; /* reuse original string */
6910
6911 return fn_id;
6912 }
6913
6914 /* Check if main program or global subprog's function prototype has `arg:ctx`
6915 * argument tags, and, if necessary, substitute correct type to match what BPF
6916 * verifier would expect, taking into account specific program type. This
6917 * allows to support __arg_ctx tag transparently on old kernels that don't yet
6918 * have a native support for it in the verifier, making user's life much
6919 * easier.
6920 */
bpf_program_fixup_func_info(struct bpf_object * obj,struct bpf_program * prog)6921 static int bpf_program_fixup_func_info(struct bpf_object *obj, struct bpf_program *prog)
6922 {
6923 const char *ctx_name = NULL, *ctx_tag = "arg:ctx", *fn_name;
6924 struct bpf_func_info_min *func_rec;
6925 struct btf_type *fn_t, *fn_proto_t;
6926 struct btf *btf = obj->btf;
6927 const struct btf_type *t;
6928 struct btf_param *p;
6929 int ptr_id = 0, struct_id, tag_id, orig_fn_id;
6930 int i, n, arg_idx, arg_cnt, err, rec_idx;
6931 int *orig_ids;
6932
6933 /* no .BTF.ext, no problem */
6934 if (!obj->btf_ext || !prog->func_info)
6935 return 0;
6936
6937 /* don't do any fix ups if kernel natively supports __arg_ctx */
6938 if (kernel_supports(obj, FEAT_ARG_CTX_TAG))
6939 return 0;
6940
6941 /* some BPF program types just don't have named context structs, so
6942 * this fallback mechanism doesn't work for them
6943 */
6944 for (i = 0; i < ARRAY_SIZE(global_ctx_map); i++) {
6945 if (global_ctx_map[i].prog_type != prog->type)
6946 continue;
6947 ctx_name = global_ctx_map[i].ctx_name;
6948 break;
6949 }
6950 if (!ctx_name)
6951 return 0;
6952
6953 /* remember original func BTF IDs to detect if we already cloned them */
6954 orig_ids = calloc(prog->func_info_cnt, sizeof(*orig_ids));
6955 if (!orig_ids)
6956 return -ENOMEM;
6957 for (i = 0; i < prog->func_info_cnt; i++) {
6958 func_rec = prog->func_info + prog->func_info_rec_size * i;
6959 orig_ids[i] = func_rec->type_id;
6960 }
6961
6962 /* go through each DECL_TAG with "arg:ctx" and see if it points to one
6963 * of our subprogs; if yes and subprog is global and needs adjustment,
6964 * clone and adjust FUNC -> FUNC_PROTO combo
6965 */
6966 for (i = 1, n = btf__type_cnt(btf); i < n; i++) {
6967 /* only DECL_TAG with "arg:ctx" value are interesting */
6968 t = btf__type_by_id(btf, i);
6969 if (!btf_is_decl_tag(t))
6970 continue;
6971 if (strcmp(btf__str_by_offset(btf, t->name_off), ctx_tag) != 0)
6972 continue;
6973
6974 /* only global funcs need adjustment, if at all */
6975 orig_fn_id = t->type;
6976 fn_t = btf_type_by_id(btf, orig_fn_id);
6977 if (!btf_is_func(fn_t) || btf_func_linkage(fn_t) != BTF_FUNC_GLOBAL)
6978 continue;
6979
6980 /* sanity check FUNC -> FUNC_PROTO chain, just in case */
6981 fn_proto_t = btf_type_by_id(btf, fn_t->type);
6982 if (!fn_proto_t || !btf_is_func_proto(fn_proto_t))
6983 continue;
6984
6985 /* find corresponding func_info record */
6986 func_rec = NULL;
6987 for (rec_idx = 0; rec_idx < prog->func_info_cnt; rec_idx++) {
6988 if (orig_ids[rec_idx] == t->type) {
6989 func_rec = prog->func_info + prog->func_info_rec_size * rec_idx;
6990 break;
6991 }
6992 }
6993 /* current main program doesn't call into this subprog */
6994 if (!func_rec)
6995 continue;
6996
6997 /* some more sanity checking of DECL_TAG */
6998 arg_cnt = btf_vlen(fn_proto_t);
6999 arg_idx = btf_decl_tag(t)->component_idx;
7000 if (arg_idx < 0 || arg_idx >= arg_cnt)
7001 continue;
7002
7003 /* check if we should fix up argument type */
7004 p = &btf_params(fn_proto_t)[arg_idx];
7005 fn_name = btf__str_by_offset(btf, fn_t->name_off) ?: "<anon>";
7006 if (!need_func_arg_type_fixup(btf, prog, fn_name, arg_idx, p->type, ctx_name))
7007 continue;
7008
7009 /* clone fn/fn_proto, unless we already did it for another arg */
7010 if (func_rec->type_id == orig_fn_id) {
7011 int fn_id;
7012
7013 fn_id = clone_func_btf_info(btf, orig_fn_id, prog);
7014 if (fn_id < 0) {
7015 err = fn_id;
7016 goto err_out;
7017 }
7018
7019 /* point func_info record to a cloned FUNC type */
7020 func_rec->type_id = fn_id;
7021 }
7022
7023 /* create PTR -> STRUCT type chain to mark PTR_TO_CTX argument;
7024 * we do it just once per main BPF program, as all global
7025 * funcs share the same program type, so need only PTR ->
7026 * STRUCT type chain
7027 */
7028 if (ptr_id == 0) {
7029 struct_id = btf__add_struct(btf, ctx_name, 0);
7030 ptr_id = btf__add_ptr(btf, struct_id);
7031 if (ptr_id < 0 || struct_id < 0) {
7032 err = -EINVAL;
7033 goto err_out;
7034 }
7035 }
7036
7037 /* for completeness, clone DECL_TAG and point it to cloned param */
7038 tag_id = btf__add_decl_tag(btf, ctx_tag, func_rec->type_id, arg_idx);
7039 if (tag_id < 0) {
7040 err = -EINVAL;
7041 goto err_out;
7042 }
7043
7044 /* all the BTF manipulations invalidated pointers, refetch them */
7045 fn_t = btf_type_by_id(btf, func_rec->type_id);
7046 fn_proto_t = btf_type_by_id(btf, fn_t->type);
7047
7048 /* fix up type ID pointed to by param */
7049 p = &btf_params(fn_proto_t)[arg_idx];
7050 p->type = ptr_id;
7051 }
7052
7053 free(orig_ids);
7054 return 0;
7055 err_out:
7056 free(orig_ids);
7057 return err;
7058 }
7059
bpf_object__relocate(struct bpf_object * obj,const char * targ_btf_path)7060 static int bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
7061 {
7062 struct bpf_program *prog;
7063 size_t i, j;
7064 int err;
7065
7066 if (obj->btf_ext) {
7067 err = bpf_object__relocate_core(obj, targ_btf_path);
7068 if (err) {
7069 pr_warn("failed to perform CO-RE relocations: %s\n",
7070 errstr(err));
7071 return err;
7072 }
7073 bpf_object__sort_relos(obj);
7074 }
7075
7076 /* Before relocating calls pre-process relocations and mark
7077 * few ld_imm64 instructions that points to subprogs.
7078 * Otherwise bpf_object__reloc_code() later would have to consider
7079 * all ld_imm64 insns as relocation candidates. That would
7080 * reduce relocation speed, since amount of find_prog_insn_relo()
7081 * would increase and most of them will fail to find a relo.
7082 */
7083 for (i = 0; i < obj->nr_programs; i++) {
7084 prog = &obj->programs[i];
7085 for (j = 0; j < prog->nr_reloc; j++) {
7086 struct reloc_desc *relo = &prog->reloc_desc[j];
7087 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
7088
7089 /* mark the insn, so it's recognized by insn_is_pseudo_func() */
7090 if (relo->type == RELO_SUBPROG_ADDR)
7091 insn[0].src_reg = BPF_PSEUDO_FUNC;
7092 }
7093 }
7094
7095 /* relocate subprogram calls and append used subprograms to main
7096 * programs; each copy of subprogram code needs to be relocated
7097 * differently for each main program, because its code location might
7098 * have changed.
7099 * Append subprog relos to main programs to allow data relos to be
7100 * processed after text is completely relocated.
7101 */
7102 for (i = 0; i < obj->nr_programs; i++) {
7103 prog = &obj->programs[i];
7104 /* sub-program's sub-calls are relocated within the context of
7105 * its main program only
7106 */
7107 if (prog_is_subprog(obj, prog))
7108 continue;
7109 if (!prog->autoload)
7110 continue;
7111
7112 err = bpf_object__relocate_calls(obj, prog);
7113 if (err) {
7114 pr_warn("prog '%s': failed to relocate calls: %s\n",
7115 prog->name, errstr(err));
7116 return err;
7117 }
7118
7119 err = bpf_prog_assign_exc_cb(obj, prog);
7120 if (err)
7121 return err;
7122 /* Now, also append exception callback if it has not been done already. */
7123 if (prog->exception_cb_idx >= 0) {
7124 struct bpf_program *subprog = &obj->programs[prog->exception_cb_idx];
7125
7126 /* Calling exception callback directly is disallowed, which the
7127 * verifier will reject later. In case it was processed already,
7128 * we can skip this step, otherwise for all other valid cases we
7129 * have to append exception callback now.
7130 */
7131 if (subprog->sub_insn_off == 0) {
7132 err = bpf_object__append_subprog_code(obj, prog, subprog);
7133 if (err)
7134 return err;
7135 err = bpf_object__reloc_code(obj, prog, subprog);
7136 if (err)
7137 return err;
7138 }
7139 }
7140 }
7141 for (i = 0; i < obj->nr_programs; i++) {
7142 prog = &obj->programs[i];
7143 if (prog_is_subprog(obj, prog))
7144 continue;
7145 if (!prog->autoload)
7146 continue;
7147
7148 /* Process data relos for main programs */
7149 err = bpf_object__relocate_data(obj, prog);
7150 if (err) {
7151 pr_warn("prog '%s': failed to relocate data references: %s\n",
7152 prog->name, errstr(err));
7153 return err;
7154 }
7155
7156 /* Fix up .BTF.ext information, if necessary */
7157 err = bpf_program_fixup_func_info(obj, prog);
7158 if (err) {
7159 pr_warn("prog '%s': failed to perform .BTF.ext fix ups: %s\n",
7160 prog->name, errstr(err));
7161 return err;
7162 }
7163 }
7164
7165 return 0;
7166 }
7167
7168 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
7169 Elf64_Shdr *shdr, Elf_Data *data);
7170
bpf_object__collect_map_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)7171 static int bpf_object__collect_map_relos(struct bpf_object *obj,
7172 Elf64_Shdr *shdr, Elf_Data *data)
7173 {
7174 const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
7175 int i, j, nrels, new_sz;
7176 const struct btf_var_secinfo *vi = NULL;
7177 const struct btf_type *sec, *var, *def;
7178 struct bpf_map *map = NULL, *targ_map = NULL;
7179 struct bpf_program *targ_prog = NULL;
7180 bool is_prog_array, is_map_in_map;
7181 const struct btf_member *member;
7182 const char *name, *mname, *type;
7183 unsigned int moff;
7184 Elf64_Sym *sym;
7185 Elf64_Rel *rel;
7186 void *tmp;
7187
7188 if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
7189 return -EINVAL;
7190 sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
7191 if (!sec)
7192 return -EINVAL;
7193
7194 nrels = shdr->sh_size / shdr->sh_entsize;
7195 for (i = 0; i < nrels; i++) {
7196 rel = elf_rel_by_idx(data, i);
7197 if (!rel) {
7198 pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
7199 return -LIBBPF_ERRNO__FORMAT;
7200 }
7201
7202 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
7203 if (!sym) {
7204 pr_warn(".maps relo #%d: symbol %zx not found\n",
7205 i, (size_t)ELF64_R_SYM(rel->r_info));
7206 return -LIBBPF_ERRNO__FORMAT;
7207 }
7208 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
7209
7210 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
7211 i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
7212 (size_t)rel->r_offset, sym->st_name, name);
7213
7214 for (j = 0; j < obj->nr_maps; j++) {
7215 map = &obj->maps[j];
7216 if (map->sec_idx != obj->efile.btf_maps_shndx)
7217 continue;
7218
7219 vi = btf_var_secinfos(sec) + map->btf_var_idx;
7220 if (vi->offset <= rel->r_offset &&
7221 rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
7222 break;
7223 }
7224 if (j == obj->nr_maps) {
7225 pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
7226 i, name, (size_t)rel->r_offset);
7227 return -EINVAL;
7228 }
7229
7230 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
7231 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
7232 type = is_map_in_map ? "map" : "prog";
7233 if (is_map_in_map) {
7234 if (sym->st_shndx != obj->efile.btf_maps_shndx) {
7235 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
7236 i, name);
7237 return -LIBBPF_ERRNO__RELOC;
7238 }
7239 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
7240 map->def.key_size != sizeof(int)) {
7241 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
7242 i, map->name, sizeof(int));
7243 return -EINVAL;
7244 }
7245 targ_map = bpf_object__find_map_by_name(obj, name);
7246 if (!targ_map) {
7247 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
7248 i, name);
7249 return -ESRCH;
7250 }
7251 } else if (is_prog_array) {
7252 targ_prog = bpf_object__find_program_by_name(obj, name);
7253 if (!targ_prog) {
7254 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
7255 i, name);
7256 return -ESRCH;
7257 }
7258 if (targ_prog->sec_idx != sym->st_shndx ||
7259 targ_prog->sec_insn_off * 8 != sym->st_value ||
7260 prog_is_subprog(obj, targ_prog)) {
7261 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
7262 i, name);
7263 return -LIBBPF_ERRNO__RELOC;
7264 }
7265 } else {
7266 return -EINVAL;
7267 }
7268
7269 var = btf__type_by_id(obj->btf, vi->type);
7270 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
7271 if (btf_vlen(def) == 0)
7272 return -EINVAL;
7273 member = btf_members(def) + btf_vlen(def) - 1;
7274 mname = btf__name_by_offset(obj->btf, member->name_off);
7275 if (strcmp(mname, "values"))
7276 return -EINVAL;
7277
7278 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
7279 if (rel->r_offset - vi->offset < moff)
7280 return -EINVAL;
7281
7282 moff = rel->r_offset - vi->offset - moff;
7283 /* here we use BPF pointer size, which is always 64 bit, as we
7284 * are parsing ELF that was built for BPF target
7285 */
7286 if (moff % bpf_ptr_sz)
7287 return -EINVAL;
7288 moff /= bpf_ptr_sz;
7289 if (moff >= map->init_slots_sz) {
7290 new_sz = moff + 1;
7291 tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
7292 if (!tmp)
7293 return -ENOMEM;
7294 map->init_slots = tmp;
7295 memset(map->init_slots + map->init_slots_sz, 0,
7296 (new_sz - map->init_slots_sz) * host_ptr_sz);
7297 map->init_slots_sz = new_sz;
7298 }
7299 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
7300
7301 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
7302 i, map->name, moff, type, name);
7303 }
7304
7305 return 0;
7306 }
7307
bpf_object__collect_relos(struct bpf_object * obj)7308 static int bpf_object__collect_relos(struct bpf_object *obj)
7309 {
7310 int i, err;
7311
7312 for (i = 0; i < obj->efile.sec_cnt; i++) {
7313 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
7314 Elf64_Shdr *shdr;
7315 Elf_Data *data;
7316 int idx;
7317
7318 if (sec_desc->sec_type != SEC_RELO)
7319 continue;
7320
7321 shdr = sec_desc->shdr;
7322 data = sec_desc->data;
7323 idx = shdr->sh_info;
7324
7325 if (shdr->sh_type != SHT_REL || idx < 0 || idx >= obj->efile.sec_cnt) {
7326 pr_warn("internal error at %d\n", __LINE__);
7327 return -LIBBPF_ERRNO__INTERNAL;
7328 }
7329
7330 if (obj->efile.secs[idx].sec_type == SEC_ST_OPS)
7331 err = bpf_object__collect_st_ops_relos(obj, shdr, data);
7332 else if (idx == obj->efile.btf_maps_shndx)
7333 err = bpf_object__collect_map_relos(obj, shdr, data);
7334 else
7335 err = bpf_object__collect_prog_relos(obj, shdr, data);
7336 if (err)
7337 return err;
7338 }
7339
7340 bpf_object__sort_relos(obj);
7341 return 0;
7342 }
7343
insn_is_helper_call(struct bpf_insn * insn,enum bpf_func_id * func_id)7344 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
7345 {
7346 if (BPF_CLASS(insn->code) == BPF_JMP &&
7347 BPF_OP(insn->code) == BPF_CALL &&
7348 BPF_SRC(insn->code) == BPF_K &&
7349 insn->src_reg == 0 &&
7350 insn->dst_reg == 0) {
7351 *func_id = insn->imm;
7352 return true;
7353 }
7354 return false;
7355 }
7356
bpf_object__sanitize_prog(struct bpf_object * obj,struct bpf_program * prog)7357 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
7358 {
7359 struct bpf_insn *insn = prog->insns;
7360 enum bpf_func_id func_id;
7361 int i;
7362
7363 if (obj->gen_loader)
7364 return 0;
7365
7366 for (i = 0; i < prog->insns_cnt; i++, insn++) {
7367 if (!insn_is_helper_call(insn, &func_id))
7368 continue;
7369
7370 /* on kernels that don't yet support
7371 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
7372 * to bpf_probe_read() which works well for old kernels
7373 */
7374 switch (func_id) {
7375 case BPF_FUNC_probe_read_kernel:
7376 case BPF_FUNC_probe_read_user:
7377 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7378 insn->imm = BPF_FUNC_probe_read;
7379 break;
7380 case BPF_FUNC_probe_read_kernel_str:
7381 case BPF_FUNC_probe_read_user_str:
7382 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7383 insn->imm = BPF_FUNC_probe_read_str;
7384 break;
7385 default:
7386 break;
7387 }
7388 }
7389 return 0;
7390 }
7391
7392 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
7393 int *btf_obj_fd, int *btf_type_id);
7394
7395 /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
libbpf_prepare_prog_load(struct bpf_program * prog,struct bpf_prog_load_opts * opts,long cookie)7396 static int libbpf_prepare_prog_load(struct bpf_program *prog,
7397 struct bpf_prog_load_opts *opts, long cookie)
7398 {
7399 enum sec_def_flags def = cookie;
7400
7401 /* old kernels might not support specifying expected_attach_type */
7402 if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
7403 opts->expected_attach_type = 0;
7404
7405 if (def & SEC_SLEEPABLE)
7406 opts->prog_flags |= BPF_F_SLEEPABLE;
7407
7408 if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
7409 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
7410
7411 /* special check for usdt to use uprobe_multi link */
7412 if ((def & SEC_USDT) && kernel_supports(prog->obj, FEAT_UPROBE_MULTI_LINK)) {
7413 /* for BPF_TRACE_UPROBE_MULTI, user might want to query expected_attach_type
7414 * in prog, and expected_attach_type we set in kernel is from opts, so we
7415 * update both.
7416 */
7417 prog->expected_attach_type = BPF_TRACE_UPROBE_MULTI;
7418 opts->expected_attach_type = BPF_TRACE_UPROBE_MULTI;
7419 }
7420
7421 if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) {
7422 int btf_obj_fd = 0, btf_type_id = 0, err;
7423 const char *attach_name;
7424
7425 attach_name = strchr(prog->sec_name, '/');
7426 if (!attach_name) {
7427 /* if BPF program is annotated with just SEC("fentry")
7428 * (or similar) without declaratively specifying
7429 * target, then it is expected that target will be
7430 * specified with bpf_program__set_attach_target() at
7431 * runtime before BPF object load step. If not, then
7432 * there is nothing to load into the kernel as BPF
7433 * verifier won't be able to validate BPF program
7434 * correctness anyways.
7435 */
7436 pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n",
7437 prog->name);
7438 return -EINVAL;
7439 }
7440 attach_name++; /* skip over / */
7441
7442 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
7443 if (err)
7444 return err;
7445
7446 /* cache resolved BTF FD and BTF type ID in the prog */
7447 prog->attach_btf_obj_fd = btf_obj_fd;
7448 prog->attach_btf_id = btf_type_id;
7449
7450 /* but by now libbpf common logic is not utilizing
7451 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
7452 * this callback is called after opts were populated by
7453 * libbpf, so this callback has to update opts explicitly here
7454 */
7455 opts->attach_btf_obj_fd = btf_obj_fd;
7456 opts->attach_btf_id = btf_type_id;
7457 }
7458 return 0;
7459 }
7460
7461 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz);
7462
bpf_object_load_prog(struct bpf_object * obj,struct bpf_program * prog,struct bpf_insn * insns,int insns_cnt,const char * license,__u32 kern_version,int * prog_fd)7463 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
7464 struct bpf_insn *insns, int insns_cnt,
7465 const char *license, __u32 kern_version, int *prog_fd)
7466 {
7467 LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
7468 const char *prog_name = NULL;
7469 size_t log_buf_size = 0;
7470 char *log_buf = NULL, *tmp;
7471 bool own_log_buf = true;
7472 __u32 log_level = prog->log_level;
7473 int ret, err;
7474
7475 /* Be more helpful by rejecting programs that can't be validated early
7476 * with more meaningful and actionable error message.
7477 */
7478 switch (prog->type) {
7479 case BPF_PROG_TYPE_UNSPEC:
7480 /*
7481 * The program type must be set. Most likely we couldn't find a proper
7482 * section definition at load time, and thus we didn't infer the type.
7483 */
7484 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
7485 prog->name, prog->sec_name);
7486 return -EINVAL;
7487 case BPF_PROG_TYPE_STRUCT_OPS:
7488 if (prog->attach_btf_id == 0) {
7489 pr_warn("prog '%s': SEC(\"struct_ops\") program isn't referenced anywhere, did you forget to use it?\n",
7490 prog->name);
7491 return -EINVAL;
7492 }
7493 break;
7494 default:
7495 break;
7496 }
7497
7498 if (!insns || !insns_cnt)
7499 return -EINVAL;
7500
7501 if (kernel_supports(obj, FEAT_PROG_NAME))
7502 prog_name = prog->name;
7503 load_attr.attach_prog_fd = prog->attach_prog_fd;
7504 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
7505 load_attr.attach_btf_id = prog->attach_btf_id;
7506 load_attr.kern_version = kern_version;
7507 load_attr.prog_ifindex = prog->prog_ifindex;
7508 load_attr.expected_attach_type = prog->expected_attach_type;
7509
7510 /* specify func_info/line_info only if kernel supports them */
7511 if (obj->btf && btf__fd(obj->btf) >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
7512 load_attr.prog_btf_fd = btf__fd(obj->btf);
7513 load_attr.func_info = prog->func_info;
7514 load_attr.func_info_rec_size = prog->func_info_rec_size;
7515 load_attr.func_info_cnt = prog->func_info_cnt;
7516 load_attr.line_info = prog->line_info;
7517 load_attr.line_info_rec_size = prog->line_info_rec_size;
7518 load_attr.line_info_cnt = prog->line_info_cnt;
7519 }
7520 load_attr.log_level = log_level;
7521 load_attr.prog_flags = prog->prog_flags;
7522 load_attr.fd_array = obj->fd_array;
7523
7524 load_attr.token_fd = obj->token_fd;
7525 if (obj->token_fd)
7526 load_attr.prog_flags |= BPF_F_TOKEN_FD;
7527
7528 /* adjust load_attr if sec_def provides custom preload callback */
7529 if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
7530 err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
7531 if (err < 0) {
7532 pr_warn("prog '%s': failed to prepare load attributes: %s\n",
7533 prog->name, errstr(err));
7534 return err;
7535 }
7536 insns = prog->insns;
7537 insns_cnt = prog->insns_cnt;
7538 }
7539
7540 if (obj->gen_loader) {
7541 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
7542 license, insns, insns_cnt, &load_attr,
7543 prog - obj->programs);
7544 *prog_fd = -1;
7545 return 0;
7546 }
7547
7548 retry_load:
7549 /* if log_level is zero, we don't request logs initially even if
7550 * custom log_buf is specified; if the program load fails, then we'll
7551 * bump log_level to 1 and use either custom log_buf or we'll allocate
7552 * our own and retry the load to get details on what failed
7553 */
7554 if (log_level) {
7555 if (prog->log_buf) {
7556 log_buf = prog->log_buf;
7557 log_buf_size = prog->log_size;
7558 own_log_buf = false;
7559 } else if (obj->log_buf) {
7560 log_buf = obj->log_buf;
7561 log_buf_size = obj->log_size;
7562 own_log_buf = false;
7563 } else {
7564 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
7565 tmp = realloc(log_buf, log_buf_size);
7566 if (!tmp) {
7567 ret = -ENOMEM;
7568 goto out;
7569 }
7570 log_buf = tmp;
7571 log_buf[0] = '\0';
7572 own_log_buf = true;
7573 }
7574 }
7575
7576 load_attr.log_buf = log_buf;
7577 load_attr.log_size = log_buf_size;
7578 load_attr.log_level = log_level;
7579
7580 ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
7581 if (ret >= 0) {
7582 if (log_level && own_log_buf) {
7583 pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7584 prog->name, log_buf);
7585 }
7586
7587 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
7588 struct bpf_map *map;
7589 int i;
7590
7591 for (i = 0; i < obj->nr_maps; i++) {
7592 map = &prog->obj->maps[i];
7593 if (map->libbpf_type != LIBBPF_MAP_RODATA)
7594 continue;
7595
7596 if (bpf_prog_bind_map(ret, map->fd, NULL)) {
7597 pr_warn("prog '%s': failed to bind map '%s': %s\n",
7598 prog->name, map->real_name, errstr(errno));
7599 /* Don't fail hard if can't bind rodata. */
7600 }
7601 }
7602 }
7603
7604 *prog_fd = ret;
7605 ret = 0;
7606 goto out;
7607 }
7608
7609 if (log_level == 0) {
7610 log_level = 1;
7611 goto retry_load;
7612 }
7613 /* On ENOSPC, increase log buffer size and retry, unless custom
7614 * log_buf is specified.
7615 * Be careful to not overflow u32, though. Kernel's log buf size limit
7616 * isn't part of UAPI so it can always be bumped to full 4GB. So don't
7617 * multiply by 2 unless we are sure we'll fit within 32 bits.
7618 * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
7619 */
7620 if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
7621 goto retry_load;
7622
7623 ret = -errno;
7624
7625 /* post-process verifier log to improve error descriptions */
7626 fixup_verifier_log(prog, log_buf, log_buf_size);
7627
7628 pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, errstr(errno));
7629 pr_perm_msg(ret);
7630
7631 if (own_log_buf && log_buf && log_buf[0] != '\0') {
7632 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7633 prog->name, log_buf);
7634 }
7635
7636 out:
7637 if (own_log_buf)
7638 free(log_buf);
7639 return ret;
7640 }
7641
find_prev_line(char * buf,char * cur)7642 static char *find_prev_line(char *buf, char *cur)
7643 {
7644 char *p;
7645
7646 if (cur == buf) /* end of a log buf */
7647 return NULL;
7648
7649 p = cur - 1;
7650 while (p - 1 >= buf && *(p - 1) != '\n')
7651 p--;
7652
7653 return p;
7654 }
7655
patch_log(char * buf,size_t buf_sz,size_t log_sz,char * orig,size_t orig_sz,const char * patch)7656 static void patch_log(char *buf, size_t buf_sz, size_t log_sz,
7657 char *orig, size_t orig_sz, const char *patch)
7658 {
7659 /* size of the remaining log content to the right from the to-be-replaced part */
7660 size_t rem_sz = (buf + log_sz) - (orig + orig_sz);
7661 size_t patch_sz = strlen(patch);
7662
7663 if (patch_sz != orig_sz) {
7664 /* If patch line(s) are longer than original piece of verifier log,
7665 * shift log contents by (patch_sz - orig_sz) bytes to the right
7666 * starting from after to-be-replaced part of the log.
7667 *
7668 * If patch line(s) are shorter than original piece of verifier log,
7669 * shift log contents by (orig_sz - patch_sz) bytes to the left
7670 * starting from after to-be-replaced part of the log
7671 *
7672 * We need to be careful about not overflowing available
7673 * buf_sz capacity. If that's the case, we'll truncate the end
7674 * of the original log, as necessary.
7675 */
7676 if (patch_sz > orig_sz) {
7677 if (orig + patch_sz >= buf + buf_sz) {
7678 /* patch is big enough to cover remaining space completely */
7679 patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1;
7680 rem_sz = 0;
7681 } else if (patch_sz - orig_sz > buf_sz - log_sz) {
7682 /* patch causes part of remaining log to be truncated */
7683 rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz);
7684 }
7685 }
7686 /* shift remaining log to the right by calculated amount */
7687 memmove(orig + patch_sz, orig + orig_sz, rem_sz);
7688 }
7689
7690 memcpy(orig, patch, patch_sz);
7691 }
7692
fixup_log_failed_core_relo(struct bpf_program * prog,char * buf,size_t buf_sz,size_t log_sz,char * line1,char * line2,char * line3)7693 static void fixup_log_failed_core_relo(struct bpf_program *prog,
7694 char *buf, size_t buf_sz, size_t log_sz,
7695 char *line1, char *line2, char *line3)
7696 {
7697 /* Expected log for failed and not properly guarded CO-RE relocation:
7698 * line1 -> 123: (85) call unknown#195896080
7699 * line2 -> invalid func unknown#195896080
7700 * line3 -> <anything else or end of buffer>
7701 *
7702 * "123" is the index of the instruction that was poisoned. We extract
7703 * instruction index to find corresponding CO-RE relocation and
7704 * replace this part of the log with more relevant information about
7705 * failed CO-RE relocation.
7706 */
7707 const struct bpf_core_relo *relo;
7708 struct bpf_core_spec spec;
7709 char patch[512], spec_buf[256];
7710 int insn_idx, err, spec_len;
7711
7712 if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1)
7713 return;
7714
7715 relo = find_relo_core(prog, insn_idx);
7716 if (!relo)
7717 return;
7718
7719 err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec);
7720 if (err)
7721 return;
7722
7723 spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec);
7724 snprintf(patch, sizeof(patch),
7725 "%d: <invalid CO-RE relocation>\n"
7726 "failed to resolve CO-RE relocation %s%s\n",
7727 insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : "");
7728
7729 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7730 }
7731
fixup_log_missing_map_load(struct bpf_program * prog,char * buf,size_t buf_sz,size_t log_sz,char * line1,char * line2,char * line3)7732 static void fixup_log_missing_map_load(struct bpf_program *prog,
7733 char *buf, size_t buf_sz, size_t log_sz,
7734 char *line1, char *line2, char *line3)
7735 {
7736 /* Expected log for failed and not properly guarded map reference:
7737 * line1 -> 123: (85) call unknown#2001000345
7738 * line2 -> invalid func unknown#2001000345
7739 * line3 -> <anything else or end of buffer>
7740 *
7741 * "123" is the index of the instruction that was poisoned.
7742 * "345" in "2001000345" is a map index in obj->maps to fetch map name.
7743 */
7744 struct bpf_object *obj = prog->obj;
7745 const struct bpf_map *map;
7746 int insn_idx, map_idx;
7747 char patch[128];
7748
7749 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2)
7750 return;
7751
7752 map_idx -= POISON_LDIMM64_MAP_BASE;
7753 if (map_idx < 0 || map_idx >= obj->nr_maps)
7754 return;
7755 map = &obj->maps[map_idx];
7756
7757 snprintf(patch, sizeof(patch),
7758 "%d: <invalid BPF map reference>\n"
7759 "BPF map '%s' is referenced but wasn't created\n",
7760 insn_idx, map->name);
7761
7762 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7763 }
7764
fixup_log_missing_kfunc_call(struct bpf_program * prog,char * buf,size_t buf_sz,size_t log_sz,char * line1,char * line2,char * line3)7765 static void fixup_log_missing_kfunc_call(struct bpf_program *prog,
7766 char *buf, size_t buf_sz, size_t log_sz,
7767 char *line1, char *line2, char *line3)
7768 {
7769 /* Expected log for failed and not properly guarded kfunc call:
7770 * line1 -> 123: (85) call unknown#2002000345
7771 * line2 -> invalid func unknown#2002000345
7772 * line3 -> <anything else or end of buffer>
7773 *
7774 * "123" is the index of the instruction that was poisoned.
7775 * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name.
7776 */
7777 struct bpf_object *obj = prog->obj;
7778 const struct extern_desc *ext;
7779 int insn_idx, ext_idx;
7780 char patch[128];
7781
7782 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2)
7783 return;
7784
7785 ext_idx -= POISON_CALL_KFUNC_BASE;
7786 if (ext_idx < 0 || ext_idx >= obj->nr_extern)
7787 return;
7788 ext = &obj->externs[ext_idx];
7789
7790 snprintf(patch, sizeof(patch),
7791 "%d: <invalid kfunc call>\n"
7792 "kfunc '%s' is referenced but wasn't resolved\n",
7793 insn_idx, ext->name);
7794
7795 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7796 }
7797
fixup_verifier_log(struct bpf_program * prog,char * buf,size_t buf_sz)7798 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz)
7799 {
7800 /* look for familiar error patterns in last N lines of the log */
7801 const size_t max_last_line_cnt = 10;
7802 char *prev_line, *cur_line, *next_line;
7803 size_t log_sz;
7804 int i;
7805
7806 if (!buf)
7807 return;
7808
7809 log_sz = strlen(buf) + 1;
7810 next_line = buf + log_sz - 1;
7811
7812 for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) {
7813 cur_line = find_prev_line(buf, next_line);
7814 if (!cur_line)
7815 return;
7816
7817 if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) {
7818 prev_line = find_prev_line(buf, cur_line);
7819 if (!prev_line)
7820 continue;
7821
7822 /* failed CO-RE relocation case */
7823 fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz,
7824 prev_line, cur_line, next_line);
7825 return;
7826 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) {
7827 prev_line = find_prev_line(buf, cur_line);
7828 if (!prev_line)
7829 continue;
7830
7831 /* reference to uncreated BPF map */
7832 fixup_log_missing_map_load(prog, buf, buf_sz, log_sz,
7833 prev_line, cur_line, next_line);
7834 return;
7835 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) {
7836 prev_line = find_prev_line(buf, cur_line);
7837 if (!prev_line)
7838 continue;
7839
7840 /* reference to unresolved kfunc */
7841 fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz,
7842 prev_line, cur_line, next_line);
7843 return;
7844 }
7845 }
7846 }
7847
bpf_program_record_relos(struct bpf_program * prog)7848 static int bpf_program_record_relos(struct bpf_program *prog)
7849 {
7850 struct bpf_object *obj = prog->obj;
7851 int i;
7852
7853 for (i = 0; i < prog->nr_reloc; i++) {
7854 struct reloc_desc *relo = &prog->reloc_desc[i];
7855 struct extern_desc *ext = &obj->externs[relo->ext_idx];
7856 int kind;
7857
7858 switch (relo->type) {
7859 case RELO_EXTERN_LD64:
7860 if (ext->type != EXT_KSYM)
7861 continue;
7862 kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ?
7863 BTF_KIND_VAR : BTF_KIND_FUNC;
7864 bpf_gen__record_extern(obj->gen_loader, ext->name,
7865 ext->is_weak, !ext->ksym.type_id,
7866 true, kind, relo->insn_idx);
7867 break;
7868 case RELO_EXTERN_CALL:
7869 bpf_gen__record_extern(obj->gen_loader, ext->name,
7870 ext->is_weak, false, false, BTF_KIND_FUNC,
7871 relo->insn_idx);
7872 break;
7873 case RELO_CORE: {
7874 struct bpf_core_relo cr = {
7875 .insn_off = relo->insn_idx * 8,
7876 .type_id = relo->core_relo->type_id,
7877 .access_str_off = relo->core_relo->access_str_off,
7878 .kind = relo->core_relo->kind,
7879 };
7880
7881 bpf_gen__record_relo_core(obj->gen_loader, &cr);
7882 break;
7883 }
7884 default:
7885 continue;
7886 }
7887 }
7888 return 0;
7889 }
7890
7891 static int
bpf_object__load_progs(struct bpf_object * obj,int log_level)7892 bpf_object__load_progs(struct bpf_object *obj, int log_level)
7893 {
7894 struct bpf_program *prog;
7895 size_t i;
7896 int err;
7897
7898 for (i = 0; i < obj->nr_programs; i++) {
7899 prog = &obj->programs[i];
7900 err = bpf_object__sanitize_prog(obj, prog);
7901 if (err)
7902 return err;
7903 }
7904
7905 for (i = 0; i < obj->nr_programs; i++) {
7906 prog = &obj->programs[i];
7907 if (prog_is_subprog(obj, prog))
7908 continue;
7909 if (!prog->autoload) {
7910 pr_debug("prog '%s': skipped loading\n", prog->name);
7911 continue;
7912 }
7913 prog->log_level |= log_level;
7914
7915 if (obj->gen_loader)
7916 bpf_program_record_relos(prog);
7917
7918 err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt,
7919 obj->license, obj->kern_version, &prog->fd);
7920 if (err) {
7921 pr_warn("prog '%s': failed to load: %s\n", prog->name, errstr(err));
7922 return err;
7923 }
7924 }
7925
7926 bpf_object__free_relocs(obj);
7927 return 0;
7928 }
7929
7930 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
7931
bpf_object_init_progs(struct bpf_object * obj,const struct bpf_object_open_opts * opts)7932 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
7933 {
7934 struct bpf_program *prog;
7935 int err;
7936
7937 bpf_object__for_each_program(prog, obj) {
7938 prog->sec_def = find_sec_def(prog->sec_name);
7939 if (!prog->sec_def) {
7940 /* couldn't guess, but user might manually specify */
7941 pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
7942 prog->name, prog->sec_name);
7943 continue;
7944 }
7945
7946 prog->type = prog->sec_def->prog_type;
7947 prog->expected_attach_type = prog->sec_def->expected_attach_type;
7948
7949 /* sec_def can have custom callback which should be called
7950 * after bpf_program is initialized to adjust its properties
7951 */
7952 if (prog->sec_def->prog_setup_fn) {
7953 err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
7954 if (err < 0) {
7955 pr_warn("prog '%s': failed to initialize: %s\n",
7956 prog->name, errstr(err));
7957 return err;
7958 }
7959 }
7960 }
7961
7962 return 0;
7963 }
7964
bpf_object_open(const char * path,const void * obj_buf,size_t obj_buf_sz,const char * obj_name,const struct bpf_object_open_opts * opts)7965 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7966 const char *obj_name,
7967 const struct bpf_object_open_opts *opts)
7968 {
7969 const char *kconfig, *btf_tmp_path, *token_path;
7970 struct bpf_object *obj;
7971 int err;
7972 char *log_buf;
7973 size_t log_size;
7974 __u32 log_level;
7975
7976 if (obj_buf && !obj_name)
7977 return ERR_PTR(-EINVAL);
7978
7979 if (elf_version(EV_CURRENT) == EV_NONE) {
7980 pr_warn("failed to init libelf for %s\n",
7981 path ? : "(mem buf)");
7982 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7983 }
7984
7985 if (!OPTS_VALID(opts, bpf_object_open_opts))
7986 return ERR_PTR(-EINVAL);
7987
7988 obj_name = OPTS_GET(opts, object_name, NULL) ?: obj_name;
7989 if (obj_buf) {
7990 path = obj_name;
7991 pr_debug("loading object '%s' from buffer\n", obj_name);
7992 } else {
7993 pr_debug("loading object from %s\n", path);
7994 }
7995
7996 log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7997 log_size = OPTS_GET(opts, kernel_log_size, 0);
7998 log_level = OPTS_GET(opts, kernel_log_level, 0);
7999 if (log_size > UINT_MAX)
8000 return ERR_PTR(-EINVAL);
8001 if (log_size && !log_buf)
8002 return ERR_PTR(-EINVAL);
8003
8004 token_path = OPTS_GET(opts, bpf_token_path, NULL);
8005 /* if user didn't specify bpf_token_path explicitly, check if
8006 * LIBBPF_BPF_TOKEN_PATH envvar was set and treat it as bpf_token_path
8007 * option
8008 */
8009 if (!token_path)
8010 token_path = getenv("LIBBPF_BPF_TOKEN_PATH");
8011 if (token_path && strlen(token_path) >= PATH_MAX)
8012 return ERR_PTR(-ENAMETOOLONG);
8013
8014 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
8015 if (IS_ERR(obj))
8016 return obj;
8017
8018 obj->log_buf = log_buf;
8019 obj->log_size = log_size;
8020 obj->log_level = log_level;
8021
8022 if (token_path) {
8023 obj->token_path = strdup(token_path);
8024 if (!obj->token_path) {
8025 err = -ENOMEM;
8026 goto out;
8027 }
8028 }
8029
8030 btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
8031 if (btf_tmp_path) {
8032 if (strlen(btf_tmp_path) >= PATH_MAX) {
8033 err = -ENAMETOOLONG;
8034 goto out;
8035 }
8036 obj->btf_custom_path = strdup(btf_tmp_path);
8037 if (!obj->btf_custom_path) {
8038 err = -ENOMEM;
8039 goto out;
8040 }
8041 }
8042
8043 kconfig = OPTS_GET(opts, kconfig, NULL);
8044 if (kconfig) {
8045 obj->kconfig = strdup(kconfig);
8046 if (!obj->kconfig) {
8047 err = -ENOMEM;
8048 goto out;
8049 }
8050 }
8051
8052 err = bpf_object__elf_init(obj);
8053 err = err ? : bpf_object__elf_collect(obj);
8054 err = err ? : bpf_object__collect_externs(obj);
8055 err = err ? : bpf_object_fixup_btf(obj);
8056 err = err ? : bpf_object__init_maps(obj, opts);
8057 err = err ? : bpf_object_init_progs(obj, opts);
8058 err = err ? : bpf_object__collect_relos(obj);
8059 if (err)
8060 goto out;
8061
8062 bpf_object__elf_finish(obj);
8063
8064 return obj;
8065 out:
8066 bpf_object__close(obj);
8067 return ERR_PTR(err);
8068 }
8069
8070 struct bpf_object *
bpf_object__open_file(const char * path,const struct bpf_object_open_opts * opts)8071 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
8072 {
8073 if (!path)
8074 return libbpf_err_ptr(-EINVAL);
8075
8076 return libbpf_ptr(bpf_object_open(path, NULL, 0, NULL, opts));
8077 }
8078
bpf_object__open(const char * path)8079 struct bpf_object *bpf_object__open(const char *path)
8080 {
8081 return bpf_object__open_file(path, NULL);
8082 }
8083
8084 struct bpf_object *
bpf_object__open_mem(const void * obj_buf,size_t obj_buf_sz,const struct bpf_object_open_opts * opts)8085 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
8086 const struct bpf_object_open_opts *opts)
8087 {
8088 char tmp_name[64];
8089
8090 if (!obj_buf || obj_buf_sz == 0)
8091 return libbpf_err_ptr(-EINVAL);
8092
8093 /* create a (quite useless) default "name" for this memory buffer object */
8094 snprintf(tmp_name, sizeof(tmp_name), "%lx-%zx", (unsigned long)obj_buf, obj_buf_sz);
8095
8096 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, tmp_name, opts));
8097 }
8098
bpf_object_unload(struct bpf_object * obj)8099 static int bpf_object_unload(struct bpf_object *obj)
8100 {
8101 size_t i;
8102
8103 if (!obj)
8104 return libbpf_err(-EINVAL);
8105
8106 for (i = 0; i < obj->nr_maps; i++) {
8107 zclose(obj->maps[i].fd);
8108 if (obj->maps[i].st_ops)
8109 zfree(&obj->maps[i].st_ops->kern_vdata);
8110 }
8111
8112 for (i = 0; i < obj->nr_programs; i++)
8113 bpf_program__unload(&obj->programs[i]);
8114
8115 return 0;
8116 }
8117
bpf_object__sanitize_maps(struct bpf_object * obj)8118 static int bpf_object__sanitize_maps(struct bpf_object *obj)
8119 {
8120 struct bpf_map *m;
8121
8122 bpf_object__for_each_map(m, obj) {
8123 if (!bpf_map__is_internal(m))
8124 continue;
8125 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
8126 m->def.map_flags &= ~BPF_F_MMAPABLE;
8127 }
8128
8129 return 0;
8130 }
8131
8132 typedef int (*kallsyms_cb_t)(unsigned long long sym_addr, char sym_type,
8133 const char *sym_name, void *ctx);
8134
libbpf_kallsyms_parse(kallsyms_cb_t cb,void * ctx)8135 static int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
8136 {
8137 char sym_type, sym_name[500];
8138 unsigned long long sym_addr;
8139 int ret, err = 0;
8140 FILE *f;
8141
8142 f = fopen("/proc/kallsyms", "re");
8143 if (!f) {
8144 err = -errno;
8145 pr_warn("failed to open /proc/kallsyms: %s\n", errstr(err));
8146 return err;
8147 }
8148
8149 while (true) {
8150 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
8151 &sym_addr, &sym_type, sym_name);
8152 if (ret == EOF && feof(f))
8153 break;
8154 if (ret != 3) {
8155 pr_warn("failed to read kallsyms entry: %d\n", ret);
8156 err = -EINVAL;
8157 break;
8158 }
8159
8160 err = cb(sym_addr, sym_type, sym_name, ctx);
8161 if (err)
8162 break;
8163 }
8164
8165 fclose(f);
8166 return err;
8167 }
8168
kallsyms_cb(unsigned long long sym_addr,char sym_type,const char * sym_name,void * ctx)8169 static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
8170 const char *sym_name, void *ctx)
8171 {
8172 struct bpf_object *obj = ctx;
8173 const struct btf_type *t;
8174 struct extern_desc *ext;
8175 char *res;
8176
8177 res = strstr(sym_name, ".llvm.");
8178 if (sym_type == 'd' && res)
8179 ext = find_extern_by_name_with_len(obj, sym_name, res - sym_name);
8180 else
8181 ext = find_extern_by_name(obj, sym_name);
8182 if (!ext || ext->type != EXT_KSYM)
8183 return 0;
8184
8185 t = btf__type_by_id(obj->btf, ext->btf_id);
8186 if (!btf_is_var(t))
8187 return 0;
8188
8189 if (ext->is_set && ext->ksym.addr != sym_addr) {
8190 pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n",
8191 sym_name, ext->ksym.addr, sym_addr);
8192 return -EINVAL;
8193 }
8194 if (!ext->is_set) {
8195 ext->is_set = true;
8196 ext->ksym.addr = sym_addr;
8197 pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr);
8198 }
8199 return 0;
8200 }
8201
bpf_object__read_kallsyms_file(struct bpf_object * obj)8202 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
8203 {
8204 return libbpf_kallsyms_parse(kallsyms_cb, obj);
8205 }
8206
find_ksym_btf_id(struct bpf_object * obj,const char * ksym_name,__u16 kind,struct btf ** res_btf,struct module_btf ** res_mod_btf)8207 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
8208 __u16 kind, struct btf **res_btf,
8209 struct module_btf **res_mod_btf)
8210 {
8211 struct module_btf *mod_btf;
8212 struct btf *btf;
8213 int i, id, err;
8214
8215 btf = obj->btf_vmlinux;
8216 mod_btf = NULL;
8217 id = btf__find_by_name_kind(btf, ksym_name, kind);
8218
8219 if (id == -ENOENT) {
8220 err = load_module_btfs(obj);
8221 if (err)
8222 return err;
8223
8224 for (i = 0; i < obj->btf_module_cnt; i++) {
8225 /* we assume module_btf's BTF FD is always >0 */
8226 mod_btf = &obj->btf_modules[i];
8227 btf = mod_btf->btf;
8228 id = btf__find_by_name_kind_own(btf, ksym_name, kind);
8229 if (id != -ENOENT)
8230 break;
8231 }
8232 }
8233 if (id <= 0)
8234 return -ESRCH;
8235
8236 *res_btf = btf;
8237 *res_mod_btf = mod_btf;
8238 return id;
8239 }
8240
bpf_object__resolve_ksym_var_btf_id(struct bpf_object * obj,struct extern_desc * ext)8241 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
8242 struct extern_desc *ext)
8243 {
8244 const struct btf_type *targ_var, *targ_type;
8245 __u32 targ_type_id, local_type_id;
8246 struct module_btf *mod_btf = NULL;
8247 const char *targ_var_name;
8248 struct btf *btf = NULL;
8249 int id, err;
8250
8251 id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
8252 if (id < 0) {
8253 if (id == -ESRCH && ext->is_weak)
8254 return 0;
8255 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
8256 ext->name);
8257 return id;
8258 }
8259
8260 /* find local type_id */
8261 local_type_id = ext->ksym.type_id;
8262
8263 /* find target type_id */
8264 targ_var = btf__type_by_id(btf, id);
8265 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
8266 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
8267
8268 err = bpf_core_types_are_compat(obj->btf, local_type_id,
8269 btf, targ_type_id);
8270 if (err <= 0) {
8271 const struct btf_type *local_type;
8272 const char *targ_name, *local_name;
8273
8274 local_type = btf__type_by_id(obj->btf, local_type_id);
8275 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
8276 targ_name = btf__name_by_offset(btf, targ_type->name_off);
8277
8278 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
8279 ext->name, local_type_id,
8280 btf_kind_str(local_type), local_name, targ_type_id,
8281 btf_kind_str(targ_type), targ_name);
8282 return -EINVAL;
8283 }
8284
8285 ext->is_set = true;
8286 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
8287 ext->ksym.kernel_btf_id = id;
8288 pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
8289 ext->name, id, btf_kind_str(targ_var), targ_var_name);
8290
8291 return 0;
8292 }
8293
bpf_object__resolve_ksym_func_btf_id(struct bpf_object * obj,struct extern_desc * ext)8294 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
8295 struct extern_desc *ext)
8296 {
8297 int local_func_proto_id, kfunc_proto_id, kfunc_id;
8298 struct module_btf *mod_btf = NULL;
8299 const struct btf_type *kern_func;
8300 struct btf *kern_btf = NULL;
8301 int ret;
8302
8303 local_func_proto_id = ext->ksym.type_id;
8304
8305 kfunc_id = find_ksym_btf_id(obj, ext->essent_name ?: ext->name, BTF_KIND_FUNC, &kern_btf,
8306 &mod_btf);
8307 if (kfunc_id < 0) {
8308 if (kfunc_id == -ESRCH && ext->is_weak)
8309 return 0;
8310 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
8311 ext->name);
8312 return kfunc_id;
8313 }
8314
8315 kern_func = btf__type_by_id(kern_btf, kfunc_id);
8316 kfunc_proto_id = kern_func->type;
8317
8318 ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
8319 kern_btf, kfunc_proto_id);
8320 if (ret <= 0) {
8321 if (ext->is_weak)
8322 return 0;
8323
8324 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n",
8325 ext->name, local_func_proto_id,
8326 mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id);
8327 return -EINVAL;
8328 }
8329
8330 /* set index for module BTF fd in fd_array, if unset */
8331 if (mod_btf && !mod_btf->fd_array_idx) {
8332 /* insn->off is s16 */
8333 if (obj->fd_array_cnt == INT16_MAX) {
8334 pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
8335 ext->name, mod_btf->fd_array_idx);
8336 return -E2BIG;
8337 }
8338 /* Cannot use index 0 for module BTF fd */
8339 if (!obj->fd_array_cnt)
8340 obj->fd_array_cnt = 1;
8341
8342 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
8343 obj->fd_array_cnt + 1);
8344 if (ret)
8345 return ret;
8346 mod_btf->fd_array_idx = obj->fd_array_cnt;
8347 /* we assume module BTF FD is always >0 */
8348 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
8349 }
8350
8351 ext->is_set = true;
8352 ext->ksym.kernel_btf_id = kfunc_id;
8353 ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
8354 /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data()
8355 * populates FD into ld_imm64 insn when it's used to point to kfunc.
8356 * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call.
8357 * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64.
8358 */
8359 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
8360 pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n",
8361 ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id);
8362
8363 return 0;
8364 }
8365
bpf_object__resolve_ksyms_btf_id(struct bpf_object * obj)8366 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
8367 {
8368 const struct btf_type *t;
8369 struct extern_desc *ext;
8370 int i, err;
8371
8372 for (i = 0; i < obj->nr_extern; i++) {
8373 ext = &obj->externs[i];
8374 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
8375 continue;
8376
8377 if (obj->gen_loader) {
8378 ext->is_set = true;
8379 ext->ksym.kernel_btf_obj_fd = 0;
8380 ext->ksym.kernel_btf_id = 0;
8381 continue;
8382 }
8383 t = btf__type_by_id(obj->btf, ext->btf_id);
8384 if (btf_is_var(t))
8385 err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
8386 else
8387 err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
8388 if (err)
8389 return err;
8390 }
8391 return 0;
8392 }
8393
bpf_object__resolve_externs(struct bpf_object * obj,const char * extra_kconfig)8394 static int bpf_object__resolve_externs(struct bpf_object *obj,
8395 const char *extra_kconfig)
8396 {
8397 bool need_config = false, need_kallsyms = false;
8398 bool need_vmlinux_btf = false;
8399 struct extern_desc *ext;
8400 void *kcfg_data = NULL;
8401 int err, i;
8402
8403 if (obj->nr_extern == 0)
8404 return 0;
8405
8406 if (obj->kconfig_map_idx >= 0)
8407 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
8408
8409 for (i = 0; i < obj->nr_extern; i++) {
8410 ext = &obj->externs[i];
8411
8412 if (ext->type == EXT_KSYM) {
8413 if (ext->ksym.type_id)
8414 need_vmlinux_btf = true;
8415 else
8416 need_kallsyms = true;
8417 continue;
8418 } else if (ext->type == EXT_KCFG) {
8419 void *ext_ptr = kcfg_data + ext->kcfg.data_off;
8420 __u64 value = 0;
8421
8422 /* Kconfig externs need actual /proc/config.gz */
8423 if (str_has_pfx(ext->name, "CONFIG_")) {
8424 need_config = true;
8425 continue;
8426 }
8427
8428 /* Virtual kcfg externs are customly handled by libbpf */
8429 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
8430 value = get_kernel_version();
8431 if (!value) {
8432 pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name);
8433 return -EINVAL;
8434 }
8435 } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) {
8436 value = kernel_supports(obj, FEAT_BPF_COOKIE);
8437 } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) {
8438 value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER);
8439 } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) {
8440 /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed
8441 * __kconfig externs, where LINUX_ ones are virtual and filled out
8442 * customly by libbpf (their values don't come from Kconfig).
8443 * If LINUX_xxx variable is not recognized by libbpf, but is marked
8444 * __weak, it defaults to zero value, just like for CONFIG_xxx
8445 * externs.
8446 */
8447 pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name);
8448 return -EINVAL;
8449 }
8450
8451 err = set_kcfg_value_num(ext, ext_ptr, value);
8452 if (err)
8453 return err;
8454 pr_debug("extern (kcfg) '%s': set to 0x%llx\n",
8455 ext->name, (long long)value);
8456 } else {
8457 pr_warn("extern '%s': unrecognized extern kind\n", ext->name);
8458 return -EINVAL;
8459 }
8460 }
8461 if (need_config && extra_kconfig) {
8462 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
8463 if (err)
8464 return -EINVAL;
8465 need_config = false;
8466 for (i = 0; i < obj->nr_extern; i++) {
8467 ext = &obj->externs[i];
8468 if (ext->type == EXT_KCFG && !ext->is_set) {
8469 need_config = true;
8470 break;
8471 }
8472 }
8473 }
8474 if (need_config) {
8475 err = bpf_object__read_kconfig_file(obj, kcfg_data);
8476 if (err)
8477 return -EINVAL;
8478 }
8479 if (need_kallsyms) {
8480 err = bpf_object__read_kallsyms_file(obj);
8481 if (err)
8482 return -EINVAL;
8483 }
8484 if (need_vmlinux_btf) {
8485 err = bpf_object__resolve_ksyms_btf_id(obj);
8486 if (err)
8487 return -EINVAL;
8488 }
8489 for (i = 0; i < obj->nr_extern; i++) {
8490 ext = &obj->externs[i];
8491
8492 if (!ext->is_set && !ext->is_weak) {
8493 pr_warn("extern '%s' (strong): not resolved\n", ext->name);
8494 return -ESRCH;
8495 } else if (!ext->is_set) {
8496 pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n",
8497 ext->name);
8498 }
8499 }
8500
8501 return 0;
8502 }
8503
bpf_map_prepare_vdata(const struct bpf_map * map)8504 static void bpf_map_prepare_vdata(const struct bpf_map *map)
8505 {
8506 const struct btf_type *type;
8507 struct bpf_struct_ops *st_ops;
8508 __u32 i;
8509
8510 st_ops = map->st_ops;
8511 type = btf__type_by_id(map->obj->btf, st_ops->type_id);
8512 for (i = 0; i < btf_vlen(type); i++) {
8513 struct bpf_program *prog = st_ops->progs[i];
8514 void *kern_data;
8515 int prog_fd;
8516
8517 if (!prog)
8518 continue;
8519
8520 prog_fd = bpf_program__fd(prog);
8521 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
8522 *(unsigned long *)kern_data = prog_fd;
8523 }
8524 }
8525
bpf_object_prepare_struct_ops(struct bpf_object * obj)8526 static int bpf_object_prepare_struct_ops(struct bpf_object *obj)
8527 {
8528 struct bpf_map *map;
8529 int i;
8530
8531 for (i = 0; i < obj->nr_maps; i++) {
8532 map = &obj->maps[i];
8533
8534 if (!bpf_map__is_struct_ops(map))
8535 continue;
8536
8537 if (!map->autocreate)
8538 continue;
8539
8540 bpf_map_prepare_vdata(map);
8541 }
8542
8543 return 0;
8544 }
8545
bpf_object_load(struct bpf_object * obj,int extra_log_level,const char * target_btf_path)8546 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
8547 {
8548 int err, i;
8549
8550 if (!obj)
8551 return libbpf_err(-EINVAL);
8552
8553 if (obj->loaded) {
8554 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
8555 return libbpf_err(-EINVAL);
8556 }
8557
8558 /* Disallow kernel loading programs of non-native endianness but
8559 * permit cross-endian creation of "light skeleton".
8560 */
8561 if (obj->gen_loader) {
8562 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
8563 } else if (!is_native_endianness(obj)) {
8564 pr_warn("object '%s': loading non-native endianness is unsupported\n", obj->name);
8565 return libbpf_err(-LIBBPF_ERRNO__ENDIAN);
8566 }
8567
8568 err = bpf_object_prepare_token(obj);
8569 err = err ? : bpf_object__probe_loading(obj);
8570 err = err ? : bpf_object__load_vmlinux_btf(obj, false);
8571 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
8572 err = err ? : bpf_object__sanitize_maps(obj);
8573 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
8574 err = err ? : bpf_object_adjust_struct_ops_autoload(obj);
8575 err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
8576 err = err ? : bpf_object__sanitize_and_load_btf(obj);
8577 err = err ? : bpf_object__create_maps(obj);
8578 err = err ? : bpf_object__load_progs(obj, extra_log_level);
8579 err = err ? : bpf_object_init_prog_arrays(obj);
8580 err = err ? : bpf_object_prepare_struct_ops(obj);
8581
8582 if (obj->gen_loader) {
8583 /* reset FDs */
8584 if (obj->btf)
8585 btf__set_fd(obj->btf, -1);
8586 if (!err)
8587 err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
8588 }
8589
8590 /* clean up fd_array */
8591 zfree(&obj->fd_array);
8592
8593 /* clean up module BTFs */
8594 for (i = 0; i < obj->btf_module_cnt; i++) {
8595 close(obj->btf_modules[i].fd);
8596 btf__free(obj->btf_modules[i].btf);
8597 free(obj->btf_modules[i].name);
8598 }
8599 free(obj->btf_modules);
8600
8601 /* clean up vmlinux BTF */
8602 btf__free(obj->btf_vmlinux);
8603 obj->btf_vmlinux = NULL;
8604
8605 obj->loaded = true; /* doesn't matter if successfully or not */
8606
8607 if (err)
8608 goto out;
8609
8610 return 0;
8611 out:
8612 /* unpin any maps that were auto-pinned during load */
8613 for (i = 0; i < obj->nr_maps; i++)
8614 if (obj->maps[i].pinned && !obj->maps[i].reused)
8615 bpf_map__unpin(&obj->maps[i], NULL);
8616
8617 bpf_object_unload(obj);
8618 pr_warn("failed to load object '%s'\n", obj->path);
8619 return libbpf_err(err);
8620 }
8621
bpf_object__load(struct bpf_object * obj)8622 int bpf_object__load(struct bpf_object *obj)
8623 {
8624 return bpf_object_load(obj, 0, NULL);
8625 }
8626
make_parent_dir(const char * path)8627 static int make_parent_dir(const char *path)
8628 {
8629 char *dname, *dir;
8630 int err = 0;
8631
8632 dname = strdup(path);
8633 if (dname == NULL)
8634 return -ENOMEM;
8635
8636 dir = dirname(dname);
8637 if (mkdir(dir, 0700) && errno != EEXIST)
8638 err = -errno;
8639
8640 free(dname);
8641 if (err) {
8642 pr_warn("failed to mkdir %s: %s\n", path, errstr(err));
8643 }
8644 return err;
8645 }
8646
check_path(const char * path)8647 static int check_path(const char *path)
8648 {
8649 struct statfs st_fs;
8650 char *dname, *dir;
8651 int err = 0;
8652
8653 if (path == NULL)
8654 return -EINVAL;
8655
8656 dname = strdup(path);
8657 if (dname == NULL)
8658 return -ENOMEM;
8659
8660 dir = dirname(dname);
8661 if (statfs(dir, &st_fs)) {
8662 pr_warn("failed to statfs %s: %s\n", dir, errstr(errno));
8663 err = -errno;
8664 }
8665 free(dname);
8666
8667 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
8668 pr_warn("specified path %s is not on BPF FS\n", path);
8669 err = -EINVAL;
8670 }
8671
8672 return err;
8673 }
8674
bpf_program__pin(struct bpf_program * prog,const char * path)8675 int bpf_program__pin(struct bpf_program *prog, const char *path)
8676 {
8677 int err;
8678
8679 if (prog->fd < 0) {
8680 pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name);
8681 return libbpf_err(-EINVAL);
8682 }
8683
8684 err = make_parent_dir(path);
8685 if (err)
8686 return libbpf_err(err);
8687
8688 err = check_path(path);
8689 if (err)
8690 return libbpf_err(err);
8691
8692 if (bpf_obj_pin(prog->fd, path)) {
8693 err = -errno;
8694 pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, errstr(err));
8695 return libbpf_err(err);
8696 }
8697
8698 pr_debug("prog '%s': pinned at '%s'\n", prog->name, path);
8699 return 0;
8700 }
8701
bpf_program__unpin(struct bpf_program * prog,const char * path)8702 int bpf_program__unpin(struct bpf_program *prog, const char *path)
8703 {
8704 int err;
8705
8706 if (prog->fd < 0) {
8707 pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name);
8708 return libbpf_err(-EINVAL);
8709 }
8710
8711 err = check_path(path);
8712 if (err)
8713 return libbpf_err(err);
8714
8715 err = unlink(path);
8716 if (err)
8717 return libbpf_err(-errno);
8718
8719 pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path);
8720 return 0;
8721 }
8722
bpf_map__pin(struct bpf_map * map,const char * path)8723 int bpf_map__pin(struct bpf_map *map, const char *path)
8724 {
8725 int err;
8726
8727 if (map == NULL) {
8728 pr_warn("invalid map pointer\n");
8729 return libbpf_err(-EINVAL);
8730 }
8731
8732 if (map->fd < 0) {
8733 pr_warn("map '%s': can't pin BPF map without FD (was it created?)\n", map->name);
8734 return libbpf_err(-EINVAL);
8735 }
8736
8737 if (map->pin_path) {
8738 if (path && strcmp(path, map->pin_path)) {
8739 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8740 bpf_map__name(map), map->pin_path, path);
8741 return libbpf_err(-EINVAL);
8742 } else if (map->pinned) {
8743 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
8744 bpf_map__name(map), map->pin_path);
8745 return 0;
8746 }
8747 } else {
8748 if (!path) {
8749 pr_warn("missing a path to pin map '%s' at\n",
8750 bpf_map__name(map));
8751 return libbpf_err(-EINVAL);
8752 } else if (map->pinned) {
8753 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
8754 return libbpf_err(-EEXIST);
8755 }
8756
8757 map->pin_path = strdup(path);
8758 if (!map->pin_path) {
8759 err = -errno;
8760 goto out_err;
8761 }
8762 }
8763
8764 err = make_parent_dir(map->pin_path);
8765 if (err)
8766 return libbpf_err(err);
8767
8768 err = check_path(map->pin_path);
8769 if (err)
8770 return libbpf_err(err);
8771
8772 if (bpf_obj_pin(map->fd, map->pin_path)) {
8773 err = -errno;
8774 goto out_err;
8775 }
8776
8777 map->pinned = true;
8778 pr_debug("pinned map '%s'\n", map->pin_path);
8779
8780 return 0;
8781
8782 out_err:
8783 pr_warn("failed to pin map: %s\n", errstr(err));
8784 return libbpf_err(err);
8785 }
8786
bpf_map__unpin(struct bpf_map * map,const char * path)8787 int bpf_map__unpin(struct bpf_map *map, const char *path)
8788 {
8789 int err;
8790
8791 if (map == NULL) {
8792 pr_warn("invalid map pointer\n");
8793 return libbpf_err(-EINVAL);
8794 }
8795
8796 if (map->pin_path) {
8797 if (path && strcmp(path, map->pin_path)) {
8798 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8799 bpf_map__name(map), map->pin_path, path);
8800 return libbpf_err(-EINVAL);
8801 }
8802 path = map->pin_path;
8803 } else if (!path) {
8804 pr_warn("no path to unpin map '%s' from\n",
8805 bpf_map__name(map));
8806 return libbpf_err(-EINVAL);
8807 }
8808
8809 err = check_path(path);
8810 if (err)
8811 return libbpf_err(err);
8812
8813 err = unlink(path);
8814 if (err != 0)
8815 return libbpf_err(-errno);
8816
8817 map->pinned = false;
8818 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
8819
8820 return 0;
8821 }
8822
bpf_map__set_pin_path(struct bpf_map * map,const char * path)8823 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
8824 {
8825 char *new = NULL;
8826
8827 if (path) {
8828 new = strdup(path);
8829 if (!new)
8830 return libbpf_err(-errno);
8831 }
8832
8833 free(map->pin_path);
8834 map->pin_path = new;
8835 return 0;
8836 }
8837
8838 __alias(bpf_map__pin_path)
8839 const char *bpf_map__get_pin_path(const struct bpf_map *map);
8840
bpf_map__pin_path(const struct bpf_map * map)8841 const char *bpf_map__pin_path(const struct bpf_map *map)
8842 {
8843 return map->pin_path;
8844 }
8845
bpf_map__is_pinned(const struct bpf_map * map)8846 bool bpf_map__is_pinned(const struct bpf_map *map)
8847 {
8848 return map->pinned;
8849 }
8850
sanitize_pin_path(char * s)8851 static void sanitize_pin_path(char *s)
8852 {
8853 /* bpffs disallows periods in path names */
8854 while (*s) {
8855 if (*s == '.')
8856 *s = '_';
8857 s++;
8858 }
8859 }
8860
bpf_object__pin_maps(struct bpf_object * obj,const char * path)8861 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
8862 {
8863 struct bpf_map *map;
8864 int err;
8865
8866 if (!obj)
8867 return libbpf_err(-ENOENT);
8868
8869 if (!obj->loaded) {
8870 pr_warn("object not yet loaded; load it first\n");
8871 return libbpf_err(-ENOENT);
8872 }
8873
8874 bpf_object__for_each_map(map, obj) {
8875 char *pin_path = NULL;
8876 char buf[PATH_MAX];
8877
8878 if (!map->autocreate)
8879 continue;
8880
8881 if (path) {
8882 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8883 if (err)
8884 goto err_unpin_maps;
8885 sanitize_pin_path(buf);
8886 pin_path = buf;
8887 } else if (!map->pin_path) {
8888 continue;
8889 }
8890
8891 err = bpf_map__pin(map, pin_path);
8892 if (err)
8893 goto err_unpin_maps;
8894 }
8895
8896 return 0;
8897
8898 err_unpin_maps:
8899 while ((map = bpf_object__prev_map(obj, map))) {
8900 if (!map->pin_path)
8901 continue;
8902
8903 bpf_map__unpin(map, NULL);
8904 }
8905
8906 return libbpf_err(err);
8907 }
8908
bpf_object__unpin_maps(struct bpf_object * obj,const char * path)8909 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
8910 {
8911 struct bpf_map *map;
8912 int err;
8913
8914 if (!obj)
8915 return libbpf_err(-ENOENT);
8916
8917 bpf_object__for_each_map(map, obj) {
8918 char *pin_path = NULL;
8919 char buf[PATH_MAX];
8920
8921 if (path) {
8922 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8923 if (err)
8924 return libbpf_err(err);
8925 sanitize_pin_path(buf);
8926 pin_path = buf;
8927 } else if (!map->pin_path) {
8928 continue;
8929 }
8930
8931 err = bpf_map__unpin(map, pin_path);
8932 if (err)
8933 return libbpf_err(err);
8934 }
8935
8936 return 0;
8937 }
8938
bpf_object__pin_programs(struct bpf_object * obj,const char * path)8939 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8940 {
8941 struct bpf_program *prog;
8942 char buf[PATH_MAX];
8943 int err;
8944
8945 if (!obj)
8946 return libbpf_err(-ENOENT);
8947
8948 if (!obj->loaded) {
8949 pr_warn("object not yet loaded; load it first\n");
8950 return libbpf_err(-ENOENT);
8951 }
8952
8953 bpf_object__for_each_program(prog, obj) {
8954 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8955 if (err)
8956 goto err_unpin_programs;
8957
8958 err = bpf_program__pin(prog, buf);
8959 if (err)
8960 goto err_unpin_programs;
8961 }
8962
8963 return 0;
8964
8965 err_unpin_programs:
8966 while ((prog = bpf_object__prev_program(obj, prog))) {
8967 if (pathname_concat(buf, sizeof(buf), path, prog->name))
8968 continue;
8969
8970 bpf_program__unpin(prog, buf);
8971 }
8972
8973 return libbpf_err(err);
8974 }
8975
bpf_object__unpin_programs(struct bpf_object * obj,const char * path)8976 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8977 {
8978 struct bpf_program *prog;
8979 int err;
8980
8981 if (!obj)
8982 return libbpf_err(-ENOENT);
8983
8984 bpf_object__for_each_program(prog, obj) {
8985 char buf[PATH_MAX];
8986
8987 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8988 if (err)
8989 return libbpf_err(err);
8990
8991 err = bpf_program__unpin(prog, buf);
8992 if (err)
8993 return libbpf_err(err);
8994 }
8995
8996 return 0;
8997 }
8998
bpf_object__pin(struct bpf_object * obj,const char * path)8999 int bpf_object__pin(struct bpf_object *obj, const char *path)
9000 {
9001 int err;
9002
9003 err = bpf_object__pin_maps(obj, path);
9004 if (err)
9005 return libbpf_err(err);
9006
9007 err = bpf_object__pin_programs(obj, path);
9008 if (err) {
9009 bpf_object__unpin_maps(obj, path);
9010 return libbpf_err(err);
9011 }
9012
9013 return 0;
9014 }
9015
bpf_object__unpin(struct bpf_object * obj,const char * path)9016 int bpf_object__unpin(struct bpf_object *obj, const char *path)
9017 {
9018 int err;
9019
9020 err = bpf_object__unpin_programs(obj, path);
9021 if (err)
9022 return libbpf_err(err);
9023
9024 err = bpf_object__unpin_maps(obj, path);
9025 if (err)
9026 return libbpf_err(err);
9027
9028 return 0;
9029 }
9030
bpf_map__destroy(struct bpf_map * map)9031 static void bpf_map__destroy(struct bpf_map *map)
9032 {
9033 if (map->inner_map) {
9034 bpf_map__destroy(map->inner_map);
9035 zfree(&map->inner_map);
9036 }
9037
9038 zfree(&map->init_slots);
9039 map->init_slots_sz = 0;
9040
9041 if (map->mmaped && map->mmaped != map->obj->arena_data)
9042 munmap(map->mmaped, bpf_map_mmap_sz(map));
9043 map->mmaped = NULL;
9044
9045 if (map->st_ops) {
9046 zfree(&map->st_ops->data);
9047 zfree(&map->st_ops->progs);
9048 zfree(&map->st_ops->kern_func_off);
9049 zfree(&map->st_ops);
9050 }
9051
9052 zfree(&map->name);
9053 zfree(&map->real_name);
9054 zfree(&map->pin_path);
9055
9056 if (map->fd >= 0)
9057 zclose(map->fd);
9058 }
9059
bpf_object__close(struct bpf_object * obj)9060 void bpf_object__close(struct bpf_object *obj)
9061 {
9062 size_t i;
9063
9064 if (IS_ERR_OR_NULL(obj))
9065 return;
9066
9067 usdt_manager_free(obj->usdt_man);
9068 obj->usdt_man = NULL;
9069
9070 bpf_gen__free(obj->gen_loader);
9071 bpf_object__elf_finish(obj);
9072 bpf_object_unload(obj);
9073 btf__free(obj->btf);
9074 btf__free(obj->btf_vmlinux);
9075 btf_ext__free(obj->btf_ext);
9076
9077 for (i = 0; i < obj->nr_maps; i++)
9078 bpf_map__destroy(&obj->maps[i]);
9079
9080 zfree(&obj->btf_custom_path);
9081 zfree(&obj->kconfig);
9082
9083 for (i = 0; i < obj->nr_extern; i++)
9084 zfree(&obj->externs[i].essent_name);
9085
9086 zfree(&obj->externs);
9087 obj->nr_extern = 0;
9088
9089 zfree(&obj->maps);
9090 obj->nr_maps = 0;
9091
9092 if (obj->programs && obj->nr_programs) {
9093 for (i = 0; i < obj->nr_programs; i++)
9094 bpf_program__exit(&obj->programs[i]);
9095 }
9096 zfree(&obj->programs);
9097
9098 zfree(&obj->feat_cache);
9099 zfree(&obj->token_path);
9100 if (obj->token_fd > 0)
9101 close(obj->token_fd);
9102
9103 zfree(&obj->arena_data);
9104
9105 free(obj);
9106 }
9107
bpf_object__name(const struct bpf_object * obj)9108 const char *bpf_object__name(const struct bpf_object *obj)
9109 {
9110 return obj ? obj->name : libbpf_err_ptr(-EINVAL);
9111 }
9112
bpf_object__kversion(const struct bpf_object * obj)9113 unsigned int bpf_object__kversion(const struct bpf_object *obj)
9114 {
9115 return obj ? obj->kern_version : 0;
9116 }
9117
bpf_object__token_fd(const struct bpf_object * obj)9118 int bpf_object__token_fd(const struct bpf_object *obj)
9119 {
9120 return obj->token_fd ?: -1;
9121 }
9122
bpf_object__btf(const struct bpf_object * obj)9123 struct btf *bpf_object__btf(const struct bpf_object *obj)
9124 {
9125 return obj ? obj->btf : NULL;
9126 }
9127
bpf_object__btf_fd(const struct bpf_object * obj)9128 int bpf_object__btf_fd(const struct bpf_object *obj)
9129 {
9130 return obj->btf ? btf__fd(obj->btf) : -1;
9131 }
9132
bpf_object__set_kversion(struct bpf_object * obj,__u32 kern_version)9133 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
9134 {
9135 if (obj->loaded)
9136 return libbpf_err(-EINVAL);
9137
9138 obj->kern_version = kern_version;
9139
9140 return 0;
9141 }
9142
bpf_object__gen_loader(struct bpf_object * obj,struct gen_loader_opts * opts)9143 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
9144 {
9145 struct bpf_gen *gen;
9146
9147 if (!opts)
9148 return -EFAULT;
9149 if (!OPTS_VALID(opts, gen_loader_opts))
9150 return -EINVAL;
9151 gen = calloc(sizeof(*gen), 1);
9152 if (!gen)
9153 return -ENOMEM;
9154 gen->opts = opts;
9155 gen->swapped_endian = !is_native_endianness(obj);
9156 obj->gen_loader = gen;
9157 return 0;
9158 }
9159
9160 static struct bpf_program *
__bpf_program__iter(const struct bpf_program * p,const struct bpf_object * obj,bool forward)9161 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
9162 bool forward)
9163 {
9164 size_t nr_programs = obj->nr_programs;
9165 ssize_t idx;
9166
9167 if (!nr_programs)
9168 return NULL;
9169
9170 if (!p)
9171 /* Iter from the beginning */
9172 return forward ? &obj->programs[0] :
9173 &obj->programs[nr_programs - 1];
9174
9175 if (p->obj != obj) {
9176 pr_warn("error: program handler doesn't match object\n");
9177 return errno = EINVAL, NULL;
9178 }
9179
9180 idx = (p - obj->programs) + (forward ? 1 : -1);
9181 if (idx >= obj->nr_programs || idx < 0)
9182 return NULL;
9183 return &obj->programs[idx];
9184 }
9185
9186 struct bpf_program *
bpf_object__next_program(const struct bpf_object * obj,struct bpf_program * prev)9187 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
9188 {
9189 struct bpf_program *prog = prev;
9190
9191 do {
9192 prog = __bpf_program__iter(prog, obj, true);
9193 } while (prog && prog_is_subprog(obj, prog));
9194
9195 return prog;
9196 }
9197
9198 struct bpf_program *
bpf_object__prev_program(const struct bpf_object * obj,struct bpf_program * next)9199 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
9200 {
9201 struct bpf_program *prog = next;
9202
9203 do {
9204 prog = __bpf_program__iter(prog, obj, false);
9205 } while (prog && prog_is_subprog(obj, prog));
9206
9207 return prog;
9208 }
9209
bpf_program__set_ifindex(struct bpf_program * prog,__u32 ifindex)9210 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
9211 {
9212 prog->prog_ifindex = ifindex;
9213 }
9214
bpf_program__name(const struct bpf_program * prog)9215 const char *bpf_program__name(const struct bpf_program *prog)
9216 {
9217 return prog->name;
9218 }
9219
bpf_program__section_name(const struct bpf_program * prog)9220 const char *bpf_program__section_name(const struct bpf_program *prog)
9221 {
9222 return prog->sec_name;
9223 }
9224
bpf_program__autoload(const struct bpf_program * prog)9225 bool bpf_program__autoload(const struct bpf_program *prog)
9226 {
9227 return prog->autoload;
9228 }
9229
bpf_program__set_autoload(struct bpf_program * prog,bool autoload)9230 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
9231 {
9232 if (prog->obj->loaded)
9233 return libbpf_err(-EINVAL);
9234
9235 prog->autoload = autoload;
9236 return 0;
9237 }
9238
bpf_program__autoattach(const struct bpf_program * prog)9239 bool bpf_program__autoattach(const struct bpf_program *prog)
9240 {
9241 return prog->autoattach;
9242 }
9243
bpf_program__set_autoattach(struct bpf_program * prog,bool autoattach)9244 void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach)
9245 {
9246 prog->autoattach = autoattach;
9247 }
9248
bpf_program__insns(const struct bpf_program * prog)9249 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
9250 {
9251 return prog->insns;
9252 }
9253
bpf_program__insn_cnt(const struct bpf_program * prog)9254 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
9255 {
9256 return prog->insns_cnt;
9257 }
9258
bpf_program__set_insns(struct bpf_program * prog,struct bpf_insn * new_insns,size_t new_insn_cnt)9259 int bpf_program__set_insns(struct bpf_program *prog,
9260 struct bpf_insn *new_insns, size_t new_insn_cnt)
9261 {
9262 struct bpf_insn *insns;
9263
9264 if (prog->obj->loaded)
9265 return -EBUSY;
9266
9267 insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns));
9268 /* NULL is a valid return from reallocarray if the new count is zero */
9269 if (!insns && new_insn_cnt) {
9270 pr_warn("prog '%s': failed to realloc prog code\n", prog->name);
9271 return -ENOMEM;
9272 }
9273 memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns));
9274
9275 prog->insns = insns;
9276 prog->insns_cnt = new_insn_cnt;
9277 return 0;
9278 }
9279
bpf_program__fd(const struct bpf_program * prog)9280 int bpf_program__fd(const struct bpf_program *prog)
9281 {
9282 if (!prog)
9283 return libbpf_err(-EINVAL);
9284
9285 if (prog->fd < 0)
9286 return libbpf_err(-ENOENT);
9287
9288 return prog->fd;
9289 }
9290
9291 __alias(bpf_program__type)
9292 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
9293
bpf_program__type(const struct bpf_program * prog)9294 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
9295 {
9296 return prog->type;
9297 }
9298
9299 static size_t custom_sec_def_cnt;
9300 static struct bpf_sec_def *custom_sec_defs;
9301 static struct bpf_sec_def custom_fallback_def;
9302 static bool has_custom_fallback_def;
9303 static int last_custom_sec_def_handler_id;
9304
bpf_program__set_type(struct bpf_program * prog,enum bpf_prog_type type)9305 int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
9306 {
9307 if (prog->obj->loaded)
9308 return libbpf_err(-EBUSY);
9309
9310 /* if type is not changed, do nothing */
9311 if (prog->type == type)
9312 return 0;
9313
9314 prog->type = type;
9315
9316 /* If a program type was changed, we need to reset associated SEC()
9317 * handler, as it will be invalid now. The only exception is a generic
9318 * fallback handler, which by definition is program type-agnostic and
9319 * is a catch-all custom handler, optionally set by the application,
9320 * so should be able to handle any type of BPF program.
9321 */
9322 if (prog->sec_def != &custom_fallback_def)
9323 prog->sec_def = NULL;
9324 return 0;
9325 }
9326
9327 __alias(bpf_program__expected_attach_type)
9328 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
9329
bpf_program__expected_attach_type(const struct bpf_program * prog)9330 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
9331 {
9332 return prog->expected_attach_type;
9333 }
9334
bpf_program__set_expected_attach_type(struct bpf_program * prog,enum bpf_attach_type type)9335 int bpf_program__set_expected_attach_type(struct bpf_program *prog,
9336 enum bpf_attach_type type)
9337 {
9338 if (prog->obj->loaded)
9339 return libbpf_err(-EBUSY);
9340
9341 prog->expected_attach_type = type;
9342 return 0;
9343 }
9344
bpf_program__flags(const struct bpf_program * prog)9345 __u32 bpf_program__flags(const struct bpf_program *prog)
9346 {
9347 return prog->prog_flags;
9348 }
9349
bpf_program__set_flags(struct bpf_program * prog,__u32 flags)9350 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
9351 {
9352 if (prog->obj->loaded)
9353 return libbpf_err(-EBUSY);
9354
9355 prog->prog_flags = flags;
9356 return 0;
9357 }
9358
bpf_program__log_level(const struct bpf_program * prog)9359 __u32 bpf_program__log_level(const struct bpf_program *prog)
9360 {
9361 return prog->log_level;
9362 }
9363
bpf_program__set_log_level(struct bpf_program * prog,__u32 log_level)9364 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
9365 {
9366 if (prog->obj->loaded)
9367 return libbpf_err(-EBUSY);
9368
9369 prog->log_level = log_level;
9370 return 0;
9371 }
9372
bpf_program__log_buf(const struct bpf_program * prog,size_t * log_size)9373 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
9374 {
9375 *log_size = prog->log_size;
9376 return prog->log_buf;
9377 }
9378
bpf_program__set_log_buf(struct bpf_program * prog,char * log_buf,size_t log_size)9379 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
9380 {
9381 if (log_size && !log_buf)
9382 return -EINVAL;
9383 if (prog->log_size > UINT_MAX)
9384 return -EINVAL;
9385 if (prog->obj->loaded)
9386 return -EBUSY;
9387
9388 prog->log_buf = log_buf;
9389 prog->log_size = log_size;
9390 return 0;
9391 }
9392
9393 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
9394 .sec = (char *)sec_pfx, \
9395 .prog_type = BPF_PROG_TYPE_##ptype, \
9396 .expected_attach_type = atype, \
9397 .cookie = (long)(flags), \
9398 .prog_prepare_load_fn = libbpf_prepare_prog_load, \
9399 __VA_ARGS__ \
9400 }
9401
9402 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9403 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9404 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9405 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9406 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9407 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9408 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9409 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9410 static int attach_kprobe_session(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9411 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9412 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9413 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9414
9415 static const struct bpf_sec_def section_defs[] = {
9416 SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE),
9417 SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE),
9418 SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE),
9419 SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
9420 SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
9421 SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
9422 SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
9423 SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
9424 SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
9425 SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
9426 SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
9427 SEC_DEF("kprobe.session+", KPROBE, BPF_TRACE_KPROBE_SESSION, SEC_NONE, attach_kprobe_session),
9428 SEC_DEF("uprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
9429 SEC_DEF("uretprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
9430 SEC_DEF("uprobe.session+", KPROBE, BPF_TRACE_UPROBE_SESSION, SEC_NONE, attach_uprobe_multi),
9431 SEC_DEF("uprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
9432 SEC_DEF("uretprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
9433 SEC_DEF("uprobe.session.s+", KPROBE, BPF_TRACE_UPROBE_SESSION, SEC_SLEEPABLE, attach_uprobe_multi),
9434 SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
9435 SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
9436 SEC_DEF("usdt+", KPROBE, 0, SEC_USDT, attach_usdt),
9437 SEC_DEF("usdt.s+", KPROBE, 0, SEC_USDT | SEC_SLEEPABLE, attach_usdt),
9438 SEC_DEF("tc/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), /* alias for tcx */
9439 SEC_DEF("tc/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), /* alias for tcx */
9440 SEC_DEF("tcx/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE),
9441 SEC_DEF("tcx/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE),
9442 SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9443 SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9444 SEC_DEF("action", SCHED_ACT, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9445 SEC_DEF("netkit/primary", SCHED_CLS, BPF_NETKIT_PRIMARY, SEC_NONE),
9446 SEC_DEF("netkit/peer", SCHED_CLS, BPF_NETKIT_PEER, SEC_NONE),
9447 SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp),
9448 SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp),
9449 SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
9450 SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
9451 SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
9452 SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
9453 SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
9454 SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
9455 SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
9456 SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
9457 SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9458 SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9459 SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9460 SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace),
9461 SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
9462 SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
9463 SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF),
9464 SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
9465 SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
9466 SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE),
9467 SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
9468 SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
9469 SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
9470 SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
9471 SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS),
9472 SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT),
9473 SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE),
9474 SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE),
9475 SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE),
9476 SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE),
9477 SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE),
9478 SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT),
9479 SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT),
9480 SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT),
9481 SEC_DEF("sk_skb/verdict", SK_SKB, BPF_SK_SKB_VERDICT, SEC_ATTACHABLE_OPT),
9482 SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE),
9483 SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT),
9484 SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT),
9485 SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT),
9486 SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT),
9487 SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT),
9488 SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE),
9489 SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE),
9490 SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE),
9491 SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT),
9492 SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE),
9493 SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE),
9494 SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE),
9495 SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE),
9496 SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE),
9497 SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE),
9498 SEC_DEF("cgroup/connect_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_CONNECT, SEC_ATTACHABLE),
9499 SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE),
9500 SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE),
9501 SEC_DEF("cgroup/sendmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_SENDMSG, SEC_ATTACHABLE),
9502 SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE),
9503 SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE),
9504 SEC_DEF("cgroup/recvmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_RECVMSG, SEC_ATTACHABLE),
9505 SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE),
9506 SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE),
9507 SEC_DEF("cgroup/getpeername_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETPEERNAME, SEC_ATTACHABLE),
9508 SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE),
9509 SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE),
9510 SEC_DEF("cgroup/getsockname_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETSOCKNAME, SEC_ATTACHABLE),
9511 SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE),
9512 SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE),
9513 SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE),
9514 SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT),
9515 SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE),
9516 SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE),
9517 SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE),
9518 SEC_DEF("netfilter", NETFILTER, BPF_NETFILTER, SEC_NONE),
9519 };
9520
libbpf_register_prog_handler(const char * sec,enum bpf_prog_type prog_type,enum bpf_attach_type exp_attach_type,const struct libbpf_prog_handler_opts * opts)9521 int libbpf_register_prog_handler(const char *sec,
9522 enum bpf_prog_type prog_type,
9523 enum bpf_attach_type exp_attach_type,
9524 const struct libbpf_prog_handler_opts *opts)
9525 {
9526 struct bpf_sec_def *sec_def;
9527
9528 if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
9529 return libbpf_err(-EINVAL);
9530
9531 if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
9532 return libbpf_err(-E2BIG);
9533
9534 if (sec) {
9535 sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
9536 sizeof(*sec_def));
9537 if (!sec_def)
9538 return libbpf_err(-ENOMEM);
9539
9540 custom_sec_defs = sec_def;
9541 sec_def = &custom_sec_defs[custom_sec_def_cnt];
9542 } else {
9543 if (has_custom_fallback_def)
9544 return libbpf_err(-EBUSY);
9545
9546 sec_def = &custom_fallback_def;
9547 }
9548
9549 sec_def->sec = sec ? strdup(sec) : NULL;
9550 if (sec && !sec_def->sec)
9551 return libbpf_err(-ENOMEM);
9552
9553 sec_def->prog_type = prog_type;
9554 sec_def->expected_attach_type = exp_attach_type;
9555 sec_def->cookie = OPTS_GET(opts, cookie, 0);
9556
9557 sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
9558 sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
9559 sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
9560
9561 sec_def->handler_id = ++last_custom_sec_def_handler_id;
9562
9563 if (sec)
9564 custom_sec_def_cnt++;
9565 else
9566 has_custom_fallback_def = true;
9567
9568 return sec_def->handler_id;
9569 }
9570
libbpf_unregister_prog_handler(int handler_id)9571 int libbpf_unregister_prog_handler(int handler_id)
9572 {
9573 struct bpf_sec_def *sec_defs;
9574 int i;
9575
9576 if (handler_id <= 0)
9577 return libbpf_err(-EINVAL);
9578
9579 if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
9580 memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
9581 has_custom_fallback_def = false;
9582 return 0;
9583 }
9584
9585 for (i = 0; i < custom_sec_def_cnt; i++) {
9586 if (custom_sec_defs[i].handler_id == handler_id)
9587 break;
9588 }
9589
9590 if (i == custom_sec_def_cnt)
9591 return libbpf_err(-ENOENT);
9592
9593 free(custom_sec_defs[i].sec);
9594 for (i = i + 1; i < custom_sec_def_cnt; i++)
9595 custom_sec_defs[i - 1] = custom_sec_defs[i];
9596 custom_sec_def_cnt--;
9597
9598 /* try to shrink the array, but it's ok if we couldn't */
9599 sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
9600 /* if new count is zero, reallocarray can return a valid NULL result;
9601 * in this case the previous pointer will be freed, so we *have to*
9602 * reassign old pointer to the new value (even if it's NULL)
9603 */
9604 if (sec_defs || custom_sec_def_cnt == 0)
9605 custom_sec_defs = sec_defs;
9606
9607 return 0;
9608 }
9609
sec_def_matches(const struct bpf_sec_def * sec_def,const char * sec_name)9610 static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name)
9611 {
9612 size_t len = strlen(sec_def->sec);
9613
9614 /* "type/" always has to have proper SEC("type/extras") form */
9615 if (sec_def->sec[len - 1] == '/') {
9616 if (str_has_pfx(sec_name, sec_def->sec))
9617 return true;
9618 return false;
9619 }
9620
9621 /* "type+" means it can be either exact SEC("type") or
9622 * well-formed SEC("type/extras") with proper '/' separator
9623 */
9624 if (sec_def->sec[len - 1] == '+') {
9625 len--;
9626 /* not even a prefix */
9627 if (strncmp(sec_name, sec_def->sec, len) != 0)
9628 return false;
9629 /* exact match or has '/' separator */
9630 if (sec_name[len] == '\0' || sec_name[len] == '/')
9631 return true;
9632 return false;
9633 }
9634
9635 return strcmp(sec_name, sec_def->sec) == 0;
9636 }
9637
find_sec_def(const char * sec_name)9638 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
9639 {
9640 const struct bpf_sec_def *sec_def;
9641 int i, n;
9642
9643 n = custom_sec_def_cnt;
9644 for (i = 0; i < n; i++) {
9645 sec_def = &custom_sec_defs[i];
9646 if (sec_def_matches(sec_def, sec_name))
9647 return sec_def;
9648 }
9649
9650 n = ARRAY_SIZE(section_defs);
9651 for (i = 0; i < n; i++) {
9652 sec_def = §ion_defs[i];
9653 if (sec_def_matches(sec_def, sec_name))
9654 return sec_def;
9655 }
9656
9657 if (has_custom_fallback_def)
9658 return &custom_fallback_def;
9659
9660 return NULL;
9661 }
9662
9663 #define MAX_TYPE_NAME_SIZE 32
9664
libbpf_get_type_names(bool attach_type)9665 static char *libbpf_get_type_names(bool attach_type)
9666 {
9667 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
9668 char *buf;
9669
9670 buf = malloc(len);
9671 if (!buf)
9672 return NULL;
9673
9674 buf[0] = '\0';
9675 /* Forge string buf with all available names */
9676 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
9677 const struct bpf_sec_def *sec_def = §ion_defs[i];
9678
9679 if (attach_type) {
9680 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9681 continue;
9682
9683 if (!(sec_def->cookie & SEC_ATTACHABLE))
9684 continue;
9685 }
9686
9687 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
9688 free(buf);
9689 return NULL;
9690 }
9691 strcat(buf, " ");
9692 strcat(buf, section_defs[i].sec);
9693 }
9694
9695 return buf;
9696 }
9697
libbpf_prog_type_by_name(const char * name,enum bpf_prog_type * prog_type,enum bpf_attach_type * expected_attach_type)9698 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
9699 enum bpf_attach_type *expected_attach_type)
9700 {
9701 const struct bpf_sec_def *sec_def;
9702 char *type_names;
9703
9704 if (!name)
9705 return libbpf_err(-EINVAL);
9706
9707 sec_def = find_sec_def(name);
9708 if (sec_def) {
9709 *prog_type = sec_def->prog_type;
9710 *expected_attach_type = sec_def->expected_attach_type;
9711 return 0;
9712 }
9713
9714 pr_debug("failed to guess program type from ELF section '%s'\n", name);
9715 type_names = libbpf_get_type_names(false);
9716 if (type_names != NULL) {
9717 pr_debug("supported section(type) names are:%s\n", type_names);
9718 free(type_names);
9719 }
9720
9721 return libbpf_err(-ESRCH);
9722 }
9723
libbpf_bpf_attach_type_str(enum bpf_attach_type t)9724 const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t)
9725 {
9726 if (t < 0 || t >= ARRAY_SIZE(attach_type_name))
9727 return NULL;
9728
9729 return attach_type_name[t];
9730 }
9731
libbpf_bpf_link_type_str(enum bpf_link_type t)9732 const char *libbpf_bpf_link_type_str(enum bpf_link_type t)
9733 {
9734 if (t < 0 || t >= ARRAY_SIZE(link_type_name))
9735 return NULL;
9736
9737 return link_type_name[t];
9738 }
9739
libbpf_bpf_map_type_str(enum bpf_map_type t)9740 const char *libbpf_bpf_map_type_str(enum bpf_map_type t)
9741 {
9742 if (t < 0 || t >= ARRAY_SIZE(map_type_name))
9743 return NULL;
9744
9745 return map_type_name[t];
9746 }
9747
libbpf_bpf_prog_type_str(enum bpf_prog_type t)9748 const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t)
9749 {
9750 if (t < 0 || t >= ARRAY_SIZE(prog_type_name))
9751 return NULL;
9752
9753 return prog_type_name[t];
9754 }
9755
find_struct_ops_map_by_offset(struct bpf_object * obj,int sec_idx,size_t offset)9756 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
9757 int sec_idx,
9758 size_t offset)
9759 {
9760 struct bpf_map *map;
9761 size_t i;
9762
9763 for (i = 0; i < obj->nr_maps; i++) {
9764 map = &obj->maps[i];
9765 if (!bpf_map__is_struct_ops(map))
9766 continue;
9767 if (map->sec_idx == sec_idx &&
9768 map->sec_offset <= offset &&
9769 offset - map->sec_offset < map->def.value_size)
9770 return map;
9771 }
9772
9773 return NULL;
9774 }
9775
9776 /* Collect the reloc from ELF, populate the st_ops->progs[], and update
9777 * st_ops->data for shadow type.
9778 */
bpf_object__collect_st_ops_relos(struct bpf_object * obj,Elf64_Shdr * shdr,Elf_Data * data)9779 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
9780 Elf64_Shdr *shdr, Elf_Data *data)
9781 {
9782 const struct btf_type *type;
9783 const struct btf_member *member;
9784 struct bpf_struct_ops *st_ops;
9785 struct bpf_program *prog;
9786 unsigned int shdr_idx;
9787 const struct btf *btf;
9788 struct bpf_map *map;
9789 unsigned int moff, insn_idx;
9790 const char *name;
9791 __u32 member_idx;
9792 Elf64_Sym *sym;
9793 Elf64_Rel *rel;
9794 int i, nrels;
9795
9796 btf = obj->btf;
9797 nrels = shdr->sh_size / shdr->sh_entsize;
9798 for (i = 0; i < nrels; i++) {
9799 rel = elf_rel_by_idx(data, i);
9800 if (!rel) {
9801 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
9802 return -LIBBPF_ERRNO__FORMAT;
9803 }
9804
9805 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
9806 if (!sym) {
9807 pr_warn("struct_ops reloc: symbol %zx not found\n",
9808 (size_t)ELF64_R_SYM(rel->r_info));
9809 return -LIBBPF_ERRNO__FORMAT;
9810 }
9811
9812 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
9813 map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset);
9814 if (!map) {
9815 pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
9816 (size_t)rel->r_offset);
9817 return -EINVAL;
9818 }
9819
9820 moff = rel->r_offset - map->sec_offset;
9821 shdr_idx = sym->st_shndx;
9822 st_ops = map->st_ops;
9823 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
9824 map->name,
9825 (long long)(rel->r_info >> 32),
9826 (long long)sym->st_value,
9827 shdr_idx, (size_t)rel->r_offset,
9828 map->sec_offset, sym->st_name, name);
9829
9830 if (shdr_idx >= SHN_LORESERVE) {
9831 pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
9832 map->name, (size_t)rel->r_offset, shdr_idx);
9833 return -LIBBPF_ERRNO__RELOC;
9834 }
9835 if (sym->st_value % BPF_INSN_SZ) {
9836 pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
9837 map->name, (unsigned long long)sym->st_value);
9838 return -LIBBPF_ERRNO__FORMAT;
9839 }
9840 insn_idx = sym->st_value / BPF_INSN_SZ;
9841
9842 type = btf__type_by_id(btf, st_ops->type_id);
9843 member = find_member_by_offset(type, moff * 8);
9844 if (!member) {
9845 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
9846 map->name, moff);
9847 return -EINVAL;
9848 }
9849 member_idx = member - btf_members(type);
9850 name = btf__name_by_offset(btf, member->name_off);
9851
9852 if (!resolve_func_ptr(btf, member->type, NULL)) {
9853 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
9854 map->name, name);
9855 return -EINVAL;
9856 }
9857
9858 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
9859 if (!prog) {
9860 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
9861 map->name, shdr_idx, name);
9862 return -EINVAL;
9863 }
9864
9865 /* prevent the use of BPF prog with invalid type */
9866 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
9867 pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
9868 map->name, prog->name);
9869 return -EINVAL;
9870 }
9871
9872 st_ops->progs[member_idx] = prog;
9873
9874 /* st_ops->data will be exposed to users, being returned by
9875 * bpf_map__initial_value() as a pointer to the shadow
9876 * type. All function pointers in the original struct type
9877 * should be converted to a pointer to struct bpf_program
9878 * in the shadow type.
9879 */
9880 *((struct bpf_program **)(st_ops->data + moff)) = prog;
9881 }
9882
9883 return 0;
9884 }
9885
9886 #define BTF_TRACE_PREFIX "btf_trace_"
9887 #define BTF_LSM_PREFIX "bpf_lsm_"
9888 #define BTF_ITER_PREFIX "bpf_iter_"
9889 #define BTF_MAX_NAME_SIZE 128
9890
btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,const char ** prefix,int * kind)9891 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
9892 const char **prefix, int *kind)
9893 {
9894 switch (attach_type) {
9895 case BPF_TRACE_RAW_TP:
9896 *prefix = BTF_TRACE_PREFIX;
9897 *kind = BTF_KIND_TYPEDEF;
9898 break;
9899 case BPF_LSM_MAC:
9900 case BPF_LSM_CGROUP:
9901 *prefix = BTF_LSM_PREFIX;
9902 *kind = BTF_KIND_FUNC;
9903 break;
9904 case BPF_TRACE_ITER:
9905 *prefix = BTF_ITER_PREFIX;
9906 *kind = BTF_KIND_FUNC;
9907 break;
9908 default:
9909 *prefix = "";
9910 *kind = BTF_KIND_FUNC;
9911 }
9912 }
9913
find_btf_by_prefix_kind(const struct btf * btf,const char * prefix,const char * name,__u32 kind)9914 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
9915 const char *name, __u32 kind)
9916 {
9917 char btf_type_name[BTF_MAX_NAME_SIZE];
9918 int ret;
9919
9920 ret = snprintf(btf_type_name, sizeof(btf_type_name),
9921 "%s%s", prefix, name);
9922 /* snprintf returns the number of characters written excluding the
9923 * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
9924 * indicates truncation.
9925 */
9926 if (ret < 0 || ret >= sizeof(btf_type_name))
9927 return -ENAMETOOLONG;
9928 return btf__find_by_name_kind(btf, btf_type_name, kind);
9929 }
9930
find_attach_btf_id(struct btf * btf,const char * name,enum bpf_attach_type attach_type)9931 static inline int find_attach_btf_id(struct btf *btf, const char *name,
9932 enum bpf_attach_type attach_type)
9933 {
9934 const char *prefix;
9935 int kind;
9936
9937 btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
9938 return find_btf_by_prefix_kind(btf, prefix, name, kind);
9939 }
9940
libbpf_find_vmlinux_btf_id(const char * name,enum bpf_attach_type attach_type)9941 int libbpf_find_vmlinux_btf_id(const char *name,
9942 enum bpf_attach_type attach_type)
9943 {
9944 struct btf *btf;
9945 int err;
9946
9947 btf = btf__load_vmlinux_btf();
9948 err = libbpf_get_error(btf);
9949 if (err) {
9950 pr_warn("vmlinux BTF is not found\n");
9951 return libbpf_err(err);
9952 }
9953
9954 err = find_attach_btf_id(btf, name, attach_type);
9955 if (err <= 0)
9956 pr_warn("%s is not found in vmlinux BTF\n", name);
9957
9958 btf__free(btf);
9959 return libbpf_err(err);
9960 }
9961
libbpf_find_prog_btf_id(const char * name,__u32 attach_prog_fd)9962 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
9963 {
9964 struct bpf_prog_info info;
9965 __u32 info_len = sizeof(info);
9966 struct btf *btf;
9967 int err;
9968
9969 memset(&info, 0, info_len);
9970 err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len);
9971 if (err) {
9972 pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %s\n",
9973 attach_prog_fd, errstr(err));
9974 return err;
9975 }
9976
9977 err = -EINVAL;
9978 if (!info.btf_id) {
9979 pr_warn("The target program doesn't have BTF\n");
9980 goto out;
9981 }
9982 btf = btf__load_from_kernel_by_id(info.btf_id);
9983 err = libbpf_get_error(btf);
9984 if (err) {
9985 pr_warn("Failed to get BTF %d of the program: %s\n", info.btf_id, errstr(err));
9986 goto out;
9987 }
9988 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9989 btf__free(btf);
9990 if (err <= 0) {
9991 pr_warn("%s is not found in prog's BTF\n", name);
9992 goto out;
9993 }
9994 out:
9995 return err;
9996 }
9997
find_kernel_btf_id(struct bpf_object * obj,const char * attach_name,enum bpf_attach_type attach_type,int * btf_obj_fd,int * btf_type_id)9998 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9999 enum bpf_attach_type attach_type,
10000 int *btf_obj_fd, int *btf_type_id)
10001 {
10002 int ret, i, mod_len;
10003 const char *fn_name, *mod_name = NULL;
10004
10005 fn_name = strchr(attach_name, ':');
10006 if (fn_name) {
10007 mod_name = attach_name;
10008 mod_len = fn_name - mod_name;
10009 fn_name++;
10010 }
10011
10012 if (!mod_name || strncmp(mod_name, "vmlinux", mod_len) == 0) {
10013 ret = find_attach_btf_id(obj->btf_vmlinux,
10014 mod_name ? fn_name : attach_name,
10015 attach_type);
10016 if (ret > 0) {
10017 *btf_obj_fd = 0; /* vmlinux BTF */
10018 *btf_type_id = ret;
10019 return 0;
10020 }
10021 if (ret != -ENOENT)
10022 return ret;
10023 }
10024
10025 ret = load_module_btfs(obj);
10026 if (ret)
10027 return ret;
10028
10029 for (i = 0; i < obj->btf_module_cnt; i++) {
10030 const struct module_btf *mod = &obj->btf_modules[i];
10031
10032 if (mod_name && strncmp(mod->name, mod_name, mod_len) != 0)
10033 continue;
10034
10035 ret = find_attach_btf_id(mod->btf,
10036 mod_name ? fn_name : attach_name,
10037 attach_type);
10038 if (ret > 0) {
10039 *btf_obj_fd = mod->fd;
10040 *btf_type_id = ret;
10041 return 0;
10042 }
10043 if (ret == -ENOENT)
10044 continue;
10045
10046 return ret;
10047 }
10048
10049 return -ESRCH;
10050 }
10051
libbpf_find_attach_btf_id(struct bpf_program * prog,const char * attach_name,int * btf_obj_fd,int * btf_type_id)10052 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
10053 int *btf_obj_fd, int *btf_type_id)
10054 {
10055 enum bpf_attach_type attach_type = prog->expected_attach_type;
10056 __u32 attach_prog_fd = prog->attach_prog_fd;
10057 int err = 0;
10058
10059 /* BPF program's BTF ID */
10060 if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) {
10061 if (!attach_prog_fd) {
10062 pr_warn("prog '%s': attach program FD is not set\n", prog->name);
10063 return -EINVAL;
10064 }
10065 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
10066 if (err < 0) {
10067 pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %s\n",
10068 prog->name, attach_prog_fd, attach_name, errstr(err));
10069 return err;
10070 }
10071 *btf_obj_fd = 0;
10072 *btf_type_id = err;
10073 return 0;
10074 }
10075
10076 /* kernel/module BTF ID */
10077 if (prog->obj->gen_loader) {
10078 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
10079 *btf_obj_fd = 0;
10080 *btf_type_id = 1;
10081 } else {
10082 err = find_kernel_btf_id(prog->obj, attach_name,
10083 attach_type, btf_obj_fd,
10084 btf_type_id);
10085 }
10086 if (err) {
10087 pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %s\n",
10088 prog->name, attach_name, errstr(err));
10089 return err;
10090 }
10091 return 0;
10092 }
10093
libbpf_attach_type_by_name(const char * name,enum bpf_attach_type * attach_type)10094 int libbpf_attach_type_by_name(const char *name,
10095 enum bpf_attach_type *attach_type)
10096 {
10097 char *type_names;
10098 const struct bpf_sec_def *sec_def;
10099
10100 if (!name)
10101 return libbpf_err(-EINVAL);
10102
10103 sec_def = find_sec_def(name);
10104 if (!sec_def) {
10105 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
10106 type_names = libbpf_get_type_names(true);
10107 if (type_names != NULL) {
10108 pr_debug("attachable section(type) names are:%s\n", type_names);
10109 free(type_names);
10110 }
10111
10112 return libbpf_err(-EINVAL);
10113 }
10114
10115 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
10116 return libbpf_err(-EINVAL);
10117 if (!(sec_def->cookie & SEC_ATTACHABLE))
10118 return libbpf_err(-EINVAL);
10119
10120 *attach_type = sec_def->expected_attach_type;
10121 return 0;
10122 }
10123
bpf_map__fd(const struct bpf_map * map)10124 int bpf_map__fd(const struct bpf_map *map)
10125 {
10126 if (!map)
10127 return libbpf_err(-EINVAL);
10128 if (!map_is_created(map))
10129 return -1;
10130 return map->fd;
10131 }
10132
map_uses_real_name(const struct bpf_map * map)10133 static bool map_uses_real_name(const struct bpf_map *map)
10134 {
10135 /* Since libbpf started to support custom .data.* and .rodata.* maps,
10136 * their user-visible name differs from kernel-visible name. Users see
10137 * such map's corresponding ELF section name as a map name.
10138 * This check distinguishes .data/.rodata from .data.* and .rodata.*
10139 * maps to know which name has to be returned to the user.
10140 */
10141 if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
10142 return true;
10143 if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
10144 return true;
10145 return false;
10146 }
10147
bpf_map__name(const struct bpf_map * map)10148 const char *bpf_map__name(const struct bpf_map *map)
10149 {
10150 if (!map)
10151 return NULL;
10152
10153 if (map_uses_real_name(map))
10154 return map->real_name;
10155
10156 return map->name;
10157 }
10158
bpf_map__type(const struct bpf_map * map)10159 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
10160 {
10161 return map->def.type;
10162 }
10163
bpf_map__set_type(struct bpf_map * map,enum bpf_map_type type)10164 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
10165 {
10166 if (map_is_created(map))
10167 return libbpf_err(-EBUSY);
10168 map->def.type = type;
10169 return 0;
10170 }
10171
bpf_map__map_flags(const struct bpf_map * map)10172 __u32 bpf_map__map_flags(const struct bpf_map *map)
10173 {
10174 return map->def.map_flags;
10175 }
10176
bpf_map__set_map_flags(struct bpf_map * map,__u32 flags)10177 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
10178 {
10179 if (map_is_created(map))
10180 return libbpf_err(-EBUSY);
10181 map->def.map_flags = flags;
10182 return 0;
10183 }
10184
bpf_map__map_extra(const struct bpf_map * map)10185 __u64 bpf_map__map_extra(const struct bpf_map *map)
10186 {
10187 return map->map_extra;
10188 }
10189
bpf_map__set_map_extra(struct bpf_map * map,__u64 map_extra)10190 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
10191 {
10192 if (map_is_created(map))
10193 return libbpf_err(-EBUSY);
10194 map->map_extra = map_extra;
10195 return 0;
10196 }
10197
bpf_map__numa_node(const struct bpf_map * map)10198 __u32 bpf_map__numa_node(const struct bpf_map *map)
10199 {
10200 return map->numa_node;
10201 }
10202
bpf_map__set_numa_node(struct bpf_map * map,__u32 numa_node)10203 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
10204 {
10205 if (map_is_created(map))
10206 return libbpf_err(-EBUSY);
10207 map->numa_node = numa_node;
10208 return 0;
10209 }
10210
bpf_map__key_size(const struct bpf_map * map)10211 __u32 bpf_map__key_size(const struct bpf_map *map)
10212 {
10213 return map->def.key_size;
10214 }
10215
bpf_map__set_key_size(struct bpf_map * map,__u32 size)10216 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
10217 {
10218 if (map_is_created(map))
10219 return libbpf_err(-EBUSY);
10220 map->def.key_size = size;
10221 return 0;
10222 }
10223
bpf_map__value_size(const struct bpf_map * map)10224 __u32 bpf_map__value_size(const struct bpf_map *map)
10225 {
10226 return map->def.value_size;
10227 }
10228
map_btf_datasec_resize(struct bpf_map * map,__u32 size)10229 static int map_btf_datasec_resize(struct bpf_map *map, __u32 size)
10230 {
10231 struct btf *btf;
10232 struct btf_type *datasec_type, *var_type;
10233 struct btf_var_secinfo *var;
10234 const struct btf_type *array_type;
10235 const struct btf_array *array;
10236 int vlen, element_sz, new_array_id;
10237 __u32 nr_elements;
10238
10239 /* check btf existence */
10240 btf = bpf_object__btf(map->obj);
10241 if (!btf)
10242 return -ENOENT;
10243
10244 /* verify map is datasec */
10245 datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map));
10246 if (!btf_is_datasec(datasec_type)) {
10247 pr_warn("map '%s': cannot be resized, map value type is not a datasec\n",
10248 bpf_map__name(map));
10249 return -EINVAL;
10250 }
10251
10252 /* verify datasec has at least one var */
10253 vlen = btf_vlen(datasec_type);
10254 if (vlen == 0) {
10255 pr_warn("map '%s': cannot be resized, map value datasec is empty\n",
10256 bpf_map__name(map));
10257 return -EINVAL;
10258 }
10259
10260 /* verify last var in the datasec is an array */
10261 var = &btf_var_secinfos(datasec_type)[vlen - 1];
10262 var_type = btf_type_by_id(btf, var->type);
10263 array_type = skip_mods_and_typedefs(btf, var_type->type, NULL);
10264 if (!btf_is_array(array_type)) {
10265 pr_warn("map '%s': cannot be resized, last var must be an array\n",
10266 bpf_map__name(map));
10267 return -EINVAL;
10268 }
10269
10270 /* verify request size aligns with array */
10271 array = btf_array(array_type);
10272 element_sz = btf__resolve_size(btf, array->type);
10273 if (element_sz <= 0 || (size - var->offset) % element_sz != 0) {
10274 pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n",
10275 bpf_map__name(map), element_sz, size);
10276 return -EINVAL;
10277 }
10278
10279 /* create a new array based on the existing array, but with new length */
10280 nr_elements = (size - var->offset) / element_sz;
10281 new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements);
10282 if (new_array_id < 0)
10283 return new_array_id;
10284
10285 /* adding a new btf type invalidates existing pointers to btf objects,
10286 * so refresh pointers before proceeding
10287 */
10288 datasec_type = btf_type_by_id(btf, map->btf_value_type_id);
10289 var = &btf_var_secinfos(datasec_type)[vlen - 1];
10290 var_type = btf_type_by_id(btf, var->type);
10291
10292 /* finally update btf info */
10293 datasec_type->size = size;
10294 var->size = size - var->offset;
10295 var_type->type = new_array_id;
10296
10297 return 0;
10298 }
10299
bpf_map__set_value_size(struct bpf_map * map,__u32 size)10300 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
10301 {
10302 if (map->obj->loaded || map->reused)
10303 return libbpf_err(-EBUSY);
10304
10305 if (map->mmaped) {
10306 size_t mmap_old_sz, mmap_new_sz;
10307 int err;
10308
10309 if (map->def.type != BPF_MAP_TYPE_ARRAY)
10310 return -EOPNOTSUPP;
10311
10312 mmap_old_sz = bpf_map_mmap_sz(map);
10313 mmap_new_sz = array_map_mmap_sz(size, map->def.max_entries);
10314 err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz);
10315 if (err) {
10316 pr_warn("map '%s': failed to resize memory-mapped region: %s\n",
10317 bpf_map__name(map), errstr(err));
10318 return err;
10319 }
10320 err = map_btf_datasec_resize(map, size);
10321 if (err && err != -ENOENT) {
10322 pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %s\n",
10323 bpf_map__name(map), errstr(err));
10324 map->btf_value_type_id = 0;
10325 map->btf_key_type_id = 0;
10326 }
10327 }
10328
10329 map->def.value_size = size;
10330 return 0;
10331 }
10332
bpf_map__btf_key_type_id(const struct bpf_map * map)10333 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
10334 {
10335 return map ? map->btf_key_type_id : 0;
10336 }
10337
bpf_map__btf_value_type_id(const struct bpf_map * map)10338 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
10339 {
10340 return map ? map->btf_value_type_id : 0;
10341 }
10342
bpf_map__set_initial_value(struct bpf_map * map,const void * data,size_t size)10343 int bpf_map__set_initial_value(struct bpf_map *map,
10344 const void *data, size_t size)
10345 {
10346 size_t actual_sz;
10347
10348 if (map->obj->loaded || map->reused)
10349 return libbpf_err(-EBUSY);
10350
10351 if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG)
10352 return libbpf_err(-EINVAL);
10353
10354 if (map->def.type == BPF_MAP_TYPE_ARENA)
10355 actual_sz = map->obj->arena_data_sz;
10356 else
10357 actual_sz = map->def.value_size;
10358 if (size != actual_sz)
10359 return libbpf_err(-EINVAL);
10360
10361 memcpy(map->mmaped, data, size);
10362 return 0;
10363 }
10364
bpf_map__initial_value(const struct bpf_map * map,size_t * psize)10365 void *bpf_map__initial_value(const struct bpf_map *map, size_t *psize)
10366 {
10367 if (bpf_map__is_struct_ops(map)) {
10368 if (psize)
10369 *psize = map->def.value_size;
10370 return map->st_ops->data;
10371 }
10372
10373 if (!map->mmaped)
10374 return NULL;
10375
10376 if (map->def.type == BPF_MAP_TYPE_ARENA)
10377 *psize = map->obj->arena_data_sz;
10378 else
10379 *psize = map->def.value_size;
10380
10381 return map->mmaped;
10382 }
10383
bpf_map__is_internal(const struct bpf_map * map)10384 bool bpf_map__is_internal(const struct bpf_map *map)
10385 {
10386 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
10387 }
10388
bpf_map__ifindex(const struct bpf_map * map)10389 __u32 bpf_map__ifindex(const struct bpf_map *map)
10390 {
10391 return map->map_ifindex;
10392 }
10393
bpf_map__set_ifindex(struct bpf_map * map,__u32 ifindex)10394 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
10395 {
10396 if (map_is_created(map))
10397 return libbpf_err(-EBUSY);
10398 map->map_ifindex = ifindex;
10399 return 0;
10400 }
10401
bpf_map__set_inner_map_fd(struct bpf_map * map,int fd)10402 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
10403 {
10404 if (!bpf_map_type__is_map_in_map(map->def.type)) {
10405 pr_warn("error: unsupported map type\n");
10406 return libbpf_err(-EINVAL);
10407 }
10408 if (map->inner_map_fd != -1) {
10409 pr_warn("error: inner_map_fd already specified\n");
10410 return libbpf_err(-EINVAL);
10411 }
10412 if (map->inner_map) {
10413 bpf_map__destroy(map->inner_map);
10414 zfree(&map->inner_map);
10415 }
10416 map->inner_map_fd = fd;
10417 return 0;
10418 }
10419
10420 static struct bpf_map *
__bpf_map__iter(const struct bpf_map * m,const struct bpf_object * obj,int i)10421 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
10422 {
10423 ssize_t idx;
10424 struct bpf_map *s, *e;
10425
10426 if (!obj || !obj->maps)
10427 return errno = EINVAL, NULL;
10428
10429 s = obj->maps;
10430 e = obj->maps + obj->nr_maps;
10431
10432 if ((m < s) || (m >= e)) {
10433 pr_warn("error in %s: map handler doesn't belong to object\n",
10434 __func__);
10435 return errno = EINVAL, NULL;
10436 }
10437
10438 idx = (m - obj->maps) + i;
10439 if (idx >= obj->nr_maps || idx < 0)
10440 return NULL;
10441 return &obj->maps[idx];
10442 }
10443
10444 struct bpf_map *
bpf_object__next_map(const struct bpf_object * obj,const struct bpf_map * prev)10445 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
10446 {
10447 if (prev == NULL && obj != NULL)
10448 return obj->maps;
10449
10450 return __bpf_map__iter(prev, obj, 1);
10451 }
10452
10453 struct bpf_map *
bpf_object__prev_map(const struct bpf_object * obj,const struct bpf_map * next)10454 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
10455 {
10456 if (next == NULL && obj != NULL) {
10457 if (!obj->nr_maps)
10458 return NULL;
10459 return obj->maps + obj->nr_maps - 1;
10460 }
10461
10462 return __bpf_map__iter(next, obj, -1);
10463 }
10464
10465 struct bpf_map *
bpf_object__find_map_by_name(const struct bpf_object * obj,const char * name)10466 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
10467 {
10468 struct bpf_map *pos;
10469
10470 bpf_object__for_each_map(pos, obj) {
10471 /* if it's a special internal map name (which always starts
10472 * with dot) then check if that special name matches the
10473 * real map name (ELF section name)
10474 */
10475 if (name[0] == '.') {
10476 if (pos->real_name && strcmp(pos->real_name, name) == 0)
10477 return pos;
10478 continue;
10479 }
10480 /* otherwise map name has to be an exact match */
10481 if (map_uses_real_name(pos)) {
10482 if (strcmp(pos->real_name, name) == 0)
10483 return pos;
10484 continue;
10485 }
10486 if (strcmp(pos->name, name) == 0)
10487 return pos;
10488 }
10489 return errno = ENOENT, NULL;
10490 }
10491
10492 int
bpf_object__find_map_fd_by_name(const struct bpf_object * obj,const char * name)10493 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
10494 {
10495 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
10496 }
10497
validate_map_op(const struct bpf_map * map,size_t key_sz,size_t value_sz,bool check_value_sz)10498 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
10499 size_t value_sz, bool check_value_sz)
10500 {
10501 if (!map_is_created(map)) /* map is not yet created */
10502 return -ENOENT;
10503
10504 if (map->def.key_size != key_sz) {
10505 pr_warn("map '%s': unexpected key size %zu provided, expected %u\n",
10506 map->name, key_sz, map->def.key_size);
10507 return -EINVAL;
10508 }
10509
10510 if (map->fd < 0) {
10511 pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name);
10512 return -EINVAL;
10513 }
10514
10515 if (!check_value_sz)
10516 return 0;
10517
10518 switch (map->def.type) {
10519 case BPF_MAP_TYPE_PERCPU_ARRAY:
10520 case BPF_MAP_TYPE_PERCPU_HASH:
10521 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
10522 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: {
10523 int num_cpu = libbpf_num_possible_cpus();
10524 size_t elem_sz = roundup(map->def.value_size, 8);
10525
10526 if (value_sz != num_cpu * elem_sz) {
10527 pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
10528 map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
10529 return -EINVAL;
10530 }
10531 break;
10532 }
10533 default:
10534 if (map->def.value_size != value_sz) {
10535 pr_warn("map '%s': unexpected value size %zu provided, expected %u\n",
10536 map->name, value_sz, map->def.value_size);
10537 return -EINVAL;
10538 }
10539 break;
10540 }
10541 return 0;
10542 }
10543
bpf_map__lookup_elem(const struct bpf_map * map,const void * key,size_t key_sz,void * value,size_t value_sz,__u64 flags)10544 int bpf_map__lookup_elem(const struct bpf_map *map,
10545 const void *key, size_t key_sz,
10546 void *value, size_t value_sz, __u64 flags)
10547 {
10548 int err;
10549
10550 err = validate_map_op(map, key_sz, value_sz, true);
10551 if (err)
10552 return libbpf_err(err);
10553
10554 return bpf_map_lookup_elem_flags(map->fd, key, value, flags);
10555 }
10556
bpf_map__update_elem(const struct bpf_map * map,const void * key,size_t key_sz,const void * value,size_t value_sz,__u64 flags)10557 int bpf_map__update_elem(const struct bpf_map *map,
10558 const void *key, size_t key_sz,
10559 const void *value, size_t value_sz, __u64 flags)
10560 {
10561 int err;
10562
10563 err = validate_map_op(map, key_sz, value_sz, true);
10564 if (err)
10565 return libbpf_err(err);
10566
10567 return bpf_map_update_elem(map->fd, key, value, flags);
10568 }
10569
bpf_map__delete_elem(const struct bpf_map * map,const void * key,size_t key_sz,__u64 flags)10570 int bpf_map__delete_elem(const struct bpf_map *map,
10571 const void *key, size_t key_sz, __u64 flags)
10572 {
10573 int err;
10574
10575 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
10576 if (err)
10577 return libbpf_err(err);
10578
10579 return bpf_map_delete_elem_flags(map->fd, key, flags);
10580 }
10581
bpf_map__lookup_and_delete_elem(const struct bpf_map * map,const void * key,size_t key_sz,void * value,size_t value_sz,__u64 flags)10582 int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
10583 const void *key, size_t key_sz,
10584 void *value, size_t value_sz, __u64 flags)
10585 {
10586 int err;
10587
10588 err = validate_map_op(map, key_sz, value_sz, true);
10589 if (err)
10590 return libbpf_err(err);
10591
10592 return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags);
10593 }
10594
bpf_map__get_next_key(const struct bpf_map * map,const void * cur_key,void * next_key,size_t key_sz)10595 int bpf_map__get_next_key(const struct bpf_map *map,
10596 const void *cur_key, void *next_key, size_t key_sz)
10597 {
10598 int err;
10599
10600 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
10601 if (err)
10602 return libbpf_err(err);
10603
10604 return bpf_map_get_next_key(map->fd, cur_key, next_key);
10605 }
10606
libbpf_get_error(const void * ptr)10607 long libbpf_get_error(const void *ptr)
10608 {
10609 if (!IS_ERR_OR_NULL(ptr))
10610 return 0;
10611
10612 if (IS_ERR(ptr))
10613 errno = -PTR_ERR(ptr);
10614
10615 /* If ptr == NULL, then errno should be already set by the failing
10616 * API, because libbpf never returns NULL on success and it now always
10617 * sets errno on error. So no extra errno handling for ptr == NULL
10618 * case.
10619 */
10620 return -errno;
10621 }
10622
10623 /* Replace link's underlying BPF program with the new one */
bpf_link__update_program(struct bpf_link * link,struct bpf_program * prog)10624 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
10625 {
10626 int ret;
10627 int prog_fd = bpf_program__fd(prog);
10628
10629 if (prog_fd < 0) {
10630 pr_warn("prog '%s': can't use BPF program without FD (was it loaded?)\n",
10631 prog->name);
10632 return libbpf_err(-EINVAL);
10633 }
10634
10635 ret = bpf_link_update(bpf_link__fd(link), prog_fd, NULL);
10636 return libbpf_err_errno(ret);
10637 }
10638
10639 /* Release "ownership" of underlying BPF resource (typically, BPF program
10640 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
10641 * link, when destructed through bpf_link__destroy() call won't attempt to
10642 * detach/unregisted that BPF resource. This is useful in situations where,
10643 * say, attached BPF program has to outlive userspace program that attached it
10644 * in the system. Depending on type of BPF program, though, there might be
10645 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
10646 * exit of userspace program doesn't trigger automatic detachment and clean up
10647 * inside the kernel.
10648 */
bpf_link__disconnect(struct bpf_link * link)10649 void bpf_link__disconnect(struct bpf_link *link)
10650 {
10651 link->disconnected = true;
10652 }
10653
bpf_link__destroy(struct bpf_link * link)10654 int bpf_link__destroy(struct bpf_link *link)
10655 {
10656 int err = 0;
10657
10658 if (IS_ERR_OR_NULL(link))
10659 return 0;
10660
10661 if (!link->disconnected && link->detach)
10662 err = link->detach(link);
10663 if (link->pin_path)
10664 free(link->pin_path);
10665 if (link->dealloc)
10666 link->dealloc(link);
10667 else
10668 free(link);
10669
10670 return libbpf_err(err);
10671 }
10672
bpf_link__fd(const struct bpf_link * link)10673 int bpf_link__fd(const struct bpf_link *link)
10674 {
10675 return link->fd;
10676 }
10677
bpf_link__pin_path(const struct bpf_link * link)10678 const char *bpf_link__pin_path(const struct bpf_link *link)
10679 {
10680 return link->pin_path;
10681 }
10682
bpf_link__detach_fd(struct bpf_link * link)10683 static int bpf_link__detach_fd(struct bpf_link *link)
10684 {
10685 return libbpf_err_errno(close(link->fd));
10686 }
10687
bpf_link__open(const char * path)10688 struct bpf_link *bpf_link__open(const char *path)
10689 {
10690 struct bpf_link *link;
10691 int fd;
10692
10693 fd = bpf_obj_get(path);
10694 if (fd < 0) {
10695 fd = -errno;
10696 pr_warn("failed to open link at %s: %d\n", path, fd);
10697 return libbpf_err_ptr(fd);
10698 }
10699
10700 link = calloc(1, sizeof(*link));
10701 if (!link) {
10702 close(fd);
10703 return libbpf_err_ptr(-ENOMEM);
10704 }
10705 link->detach = &bpf_link__detach_fd;
10706 link->fd = fd;
10707
10708 link->pin_path = strdup(path);
10709 if (!link->pin_path) {
10710 bpf_link__destroy(link);
10711 return libbpf_err_ptr(-ENOMEM);
10712 }
10713
10714 return link;
10715 }
10716
bpf_link__detach(struct bpf_link * link)10717 int bpf_link__detach(struct bpf_link *link)
10718 {
10719 return bpf_link_detach(link->fd) ? -errno : 0;
10720 }
10721
bpf_link__pin(struct bpf_link * link,const char * path)10722 int bpf_link__pin(struct bpf_link *link, const char *path)
10723 {
10724 int err;
10725
10726 if (link->pin_path)
10727 return libbpf_err(-EBUSY);
10728 err = make_parent_dir(path);
10729 if (err)
10730 return libbpf_err(err);
10731 err = check_path(path);
10732 if (err)
10733 return libbpf_err(err);
10734
10735 link->pin_path = strdup(path);
10736 if (!link->pin_path)
10737 return libbpf_err(-ENOMEM);
10738
10739 if (bpf_obj_pin(link->fd, link->pin_path)) {
10740 err = -errno;
10741 zfree(&link->pin_path);
10742 return libbpf_err(err);
10743 }
10744
10745 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
10746 return 0;
10747 }
10748
bpf_link__unpin(struct bpf_link * link)10749 int bpf_link__unpin(struct bpf_link *link)
10750 {
10751 int err;
10752
10753 if (!link->pin_path)
10754 return libbpf_err(-EINVAL);
10755
10756 err = unlink(link->pin_path);
10757 if (err != 0)
10758 return -errno;
10759
10760 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
10761 zfree(&link->pin_path);
10762 return 0;
10763 }
10764
10765 struct bpf_link_perf {
10766 struct bpf_link link;
10767 int perf_event_fd;
10768 /* legacy kprobe support: keep track of probe identifier and type */
10769 char *legacy_probe_name;
10770 bool legacy_is_kprobe;
10771 bool legacy_is_retprobe;
10772 };
10773
10774 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
10775 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
10776
bpf_link_perf_detach(struct bpf_link * link)10777 static int bpf_link_perf_detach(struct bpf_link *link)
10778 {
10779 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10780 int err = 0;
10781
10782 if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
10783 err = -errno;
10784
10785 if (perf_link->perf_event_fd != link->fd)
10786 close(perf_link->perf_event_fd);
10787 close(link->fd);
10788
10789 /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
10790 if (perf_link->legacy_probe_name) {
10791 if (perf_link->legacy_is_kprobe) {
10792 err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
10793 perf_link->legacy_is_retprobe);
10794 } else {
10795 err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
10796 perf_link->legacy_is_retprobe);
10797 }
10798 }
10799
10800 return err;
10801 }
10802
bpf_link_perf_dealloc(struct bpf_link * link)10803 static void bpf_link_perf_dealloc(struct bpf_link *link)
10804 {
10805 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10806
10807 free(perf_link->legacy_probe_name);
10808 free(perf_link);
10809 }
10810
bpf_program__attach_perf_event_opts(const struct bpf_program * prog,int pfd,const struct bpf_perf_event_opts * opts)10811 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
10812 const struct bpf_perf_event_opts *opts)
10813 {
10814 struct bpf_link_perf *link;
10815 int prog_fd, link_fd = -1, err;
10816 bool force_ioctl_attach;
10817
10818 if (!OPTS_VALID(opts, bpf_perf_event_opts))
10819 return libbpf_err_ptr(-EINVAL);
10820
10821 if (pfd < 0) {
10822 pr_warn("prog '%s': invalid perf event FD %d\n",
10823 prog->name, pfd);
10824 return libbpf_err_ptr(-EINVAL);
10825 }
10826 prog_fd = bpf_program__fd(prog);
10827 if (prog_fd < 0) {
10828 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
10829 prog->name);
10830 return libbpf_err_ptr(-EINVAL);
10831 }
10832
10833 link = calloc(1, sizeof(*link));
10834 if (!link)
10835 return libbpf_err_ptr(-ENOMEM);
10836 link->link.detach = &bpf_link_perf_detach;
10837 link->link.dealloc = &bpf_link_perf_dealloc;
10838 link->perf_event_fd = pfd;
10839
10840 force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false);
10841 if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) {
10842 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
10843 .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
10844
10845 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
10846 if (link_fd < 0) {
10847 err = -errno;
10848 pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %s\n",
10849 prog->name, pfd, errstr(err));
10850 goto err_out;
10851 }
10852 link->link.fd = link_fd;
10853 } else {
10854 if (OPTS_GET(opts, bpf_cookie, 0)) {
10855 pr_warn("prog '%s': user context value is not supported\n", prog->name);
10856 err = -EOPNOTSUPP;
10857 goto err_out;
10858 }
10859
10860 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
10861 err = -errno;
10862 pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
10863 prog->name, pfd, errstr(err));
10864 if (err == -EPROTO)
10865 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
10866 prog->name, pfd);
10867 goto err_out;
10868 }
10869 link->link.fd = pfd;
10870 }
10871 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
10872 err = -errno;
10873 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
10874 prog->name, pfd, errstr(err));
10875 goto err_out;
10876 }
10877
10878 return &link->link;
10879 err_out:
10880 if (link_fd >= 0)
10881 close(link_fd);
10882 free(link);
10883 return libbpf_err_ptr(err);
10884 }
10885
bpf_program__attach_perf_event(const struct bpf_program * prog,int pfd)10886 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
10887 {
10888 return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
10889 }
10890
10891 /*
10892 * this function is expected to parse integer in the range of [0, 2^31-1] from
10893 * given file using scanf format string fmt. If actual parsed value is
10894 * negative, the result might be indistinguishable from error
10895 */
parse_uint_from_file(const char * file,const char * fmt)10896 static int parse_uint_from_file(const char *file, const char *fmt)
10897 {
10898 int err, ret;
10899 FILE *f;
10900
10901 f = fopen(file, "re");
10902 if (!f) {
10903 err = -errno;
10904 pr_debug("failed to open '%s': %s\n", file, errstr(err));
10905 return err;
10906 }
10907 err = fscanf(f, fmt, &ret);
10908 if (err != 1) {
10909 err = err == EOF ? -EIO : -errno;
10910 pr_debug("failed to parse '%s': %s\n", file, errstr(err));
10911 fclose(f);
10912 return err;
10913 }
10914 fclose(f);
10915 return ret;
10916 }
10917
determine_kprobe_perf_type(void)10918 static int determine_kprobe_perf_type(void)
10919 {
10920 const char *file = "/sys/bus/event_source/devices/kprobe/type";
10921
10922 return parse_uint_from_file(file, "%d\n");
10923 }
10924
determine_uprobe_perf_type(void)10925 static int determine_uprobe_perf_type(void)
10926 {
10927 const char *file = "/sys/bus/event_source/devices/uprobe/type";
10928
10929 return parse_uint_from_file(file, "%d\n");
10930 }
10931
determine_kprobe_retprobe_bit(void)10932 static int determine_kprobe_retprobe_bit(void)
10933 {
10934 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
10935
10936 return parse_uint_from_file(file, "config:%d\n");
10937 }
10938
determine_uprobe_retprobe_bit(void)10939 static int determine_uprobe_retprobe_bit(void)
10940 {
10941 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
10942
10943 return parse_uint_from_file(file, "config:%d\n");
10944 }
10945
10946 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
10947 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
10948
perf_event_open_probe(bool uprobe,bool retprobe,const char * name,uint64_t offset,int pid,size_t ref_ctr_off)10949 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
10950 uint64_t offset, int pid, size_t ref_ctr_off)
10951 {
10952 const size_t attr_sz = sizeof(struct perf_event_attr);
10953 struct perf_event_attr attr;
10954 int type, pfd;
10955
10956 if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
10957 return -EINVAL;
10958
10959 memset(&attr, 0, attr_sz);
10960
10961 type = uprobe ? determine_uprobe_perf_type()
10962 : determine_kprobe_perf_type();
10963 if (type < 0) {
10964 pr_warn("failed to determine %s perf type: %s\n",
10965 uprobe ? "uprobe" : "kprobe",
10966 errstr(type));
10967 return type;
10968 }
10969 if (retprobe) {
10970 int bit = uprobe ? determine_uprobe_retprobe_bit()
10971 : determine_kprobe_retprobe_bit();
10972
10973 if (bit < 0) {
10974 pr_warn("failed to determine %s retprobe bit: %s\n",
10975 uprobe ? "uprobe" : "kprobe",
10976 errstr(bit));
10977 return bit;
10978 }
10979 attr.config |= 1 << bit;
10980 }
10981 attr.size = attr_sz;
10982 attr.type = type;
10983 attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
10984 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
10985 attr.config2 = offset; /* kprobe_addr or probe_offset */
10986
10987 /* pid filter is meaningful only for uprobes */
10988 pfd = syscall(__NR_perf_event_open, &attr,
10989 pid < 0 ? -1 : pid /* pid */,
10990 pid == -1 ? 0 : -1 /* cpu */,
10991 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10992 return pfd >= 0 ? pfd : -errno;
10993 }
10994
append_to_file(const char * file,const char * fmt,...)10995 static int append_to_file(const char *file, const char *fmt, ...)
10996 {
10997 int fd, n, err = 0;
10998 va_list ap;
10999 char buf[1024];
11000
11001 va_start(ap, fmt);
11002 n = vsnprintf(buf, sizeof(buf), fmt, ap);
11003 va_end(ap);
11004
11005 if (n < 0 || n >= sizeof(buf))
11006 return -EINVAL;
11007
11008 fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
11009 if (fd < 0)
11010 return -errno;
11011
11012 if (write(fd, buf, n) < 0)
11013 err = -errno;
11014
11015 close(fd);
11016 return err;
11017 }
11018
11019 #define DEBUGFS "/sys/kernel/debug/tracing"
11020 #define TRACEFS "/sys/kernel/tracing"
11021
use_debugfs(void)11022 static bool use_debugfs(void)
11023 {
11024 static int has_debugfs = -1;
11025
11026 if (has_debugfs < 0)
11027 has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0;
11028
11029 return has_debugfs == 1;
11030 }
11031
tracefs_path(void)11032 static const char *tracefs_path(void)
11033 {
11034 return use_debugfs() ? DEBUGFS : TRACEFS;
11035 }
11036
tracefs_kprobe_events(void)11037 static const char *tracefs_kprobe_events(void)
11038 {
11039 return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events";
11040 }
11041
tracefs_uprobe_events(void)11042 static const char *tracefs_uprobe_events(void)
11043 {
11044 return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events";
11045 }
11046
tracefs_available_filter_functions(void)11047 static const char *tracefs_available_filter_functions(void)
11048 {
11049 return use_debugfs() ? DEBUGFS"/available_filter_functions"
11050 : TRACEFS"/available_filter_functions";
11051 }
11052
tracefs_available_filter_functions_addrs(void)11053 static const char *tracefs_available_filter_functions_addrs(void)
11054 {
11055 return use_debugfs() ? DEBUGFS"/available_filter_functions_addrs"
11056 : TRACEFS"/available_filter_functions_addrs";
11057 }
11058
gen_kprobe_legacy_event_name(char * buf,size_t buf_sz,const char * kfunc_name,size_t offset)11059 static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
11060 const char *kfunc_name, size_t offset)
11061 {
11062 static int index = 0;
11063 int i;
11064
11065 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
11066 __sync_fetch_and_add(&index, 1));
11067
11068 /* sanitize binary_path in the probe name */
11069 for (i = 0; buf[i]; i++) {
11070 if (!isalnum(buf[i]))
11071 buf[i] = '_';
11072 }
11073 }
11074
add_kprobe_event_legacy(const char * probe_name,bool retprobe,const char * kfunc_name,size_t offset)11075 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
11076 const char *kfunc_name, size_t offset)
11077 {
11078 return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx",
11079 retprobe ? 'r' : 'p',
11080 retprobe ? "kretprobes" : "kprobes",
11081 probe_name, kfunc_name, offset);
11082 }
11083
remove_kprobe_event_legacy(const char * probe_name,bool retprobe)11084 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
11085 {
11086 return append_to_file(tracefs_kprobe_events(), "-:%s/%s",
11087 retprobe ? "kretprobes" : "kprobes", probe_name);
11088 }
11089
determine_kprobe_perf_type_legacy(const char * probe_name,bool retprobe)11090 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
11091 {
11092 char file[256];
11093
11094 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11095 tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name);
11096
11097 return parse_uint_from_file(file, "%d\n");
11098 }
11099
perf_event_kprobe_open_legacy(const char * probe_name,bool retprobe,const char * kfunc_name,size_t offset,int pid)11100 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
11101 const char *kfunc_name, size_t offset, int pid)
11102 {
11103 const size_t attr_sz = sizeof(struct perf_event_attr);
11104 struct perf_event_attr attr;
11105 int type, pfd, err;
11106
11107 err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
11108 if (err < 0) {
11109 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
11110 kfunc_name, offset,
11111 errstr(err));
11112 return err;
11113 }
11114 type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
11115 if (type < 0) {
11116 err = type;
11117 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
11118 kfunc_name, offset,
11119 errstr(err));
11120 goto err_clean_legacy;
11121 }
11122
11123 memset(&attr, 0, attr_sz);
11124 attr.size = attr_sz;
11125 attr.config = type;
11126 attr.type = PERF_TYPE_TRACEPOINT;
11127
11128 pfd = syscall(__NR_perf_event_open, &attr,
11129 pid < 0 ? -1 : pid, /* pid */
11130 pid == -1 ? 0 : -1, /* cpu */
11131 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11132 if (pfd < 0) {
11133 err = -errno;
11134 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
11135 errstr(err));
11136 goto err_clean_legacy;
11137 }
11138 return pfd;
11139
11140 err_clean_legacy:
11141 /* Clear the newly added legacy kprobe_event */
11142 remove_kprobe_event_legacy(probe_name, retprobe);
11143 return err;
11144 }
11145
arch_specific_syscall_pfx(void)11146 static const char *arch_specific_syscall_pfx(void)
11147 {
11148 #if defined(__x86_64__)
11149 return "x64";
11150 #elif defined(__i386__)
11151 return "ia32";
11152 #elif defined(__s390x__)
11153 return "s390x";
11154 #elif defined(__s390__)
11155 return "s390";
11156 #elif defined(__arm__)
11157 return "arm";
11158 #elif defined(__aarch64__)
11159 return "arm64";
11160 #elif defined(__mips__)
11161 return "mips";
11162 #elif defined(__riscv)
11163 return "riscv";
11164 #elif defined(__powerpc__)
11165 return "powerpc";
11166 #elif defined(__powerpc64__)
11167 return "powerpc64";
11168 #else
11169 return NULL;
11170 #endif
11171 }
11172
probe_kern_syscall_wrapper(int token_fd)11173 int probe_kern_syscall_wrapper(int token_fd)
11174 {
11175 char syscall_name[64];
11176 const char *ksys_pfx;
11177
11178 ksys_pfx = arch_specific_syscall_pfx();
11179 if (!ksys_pfx)
11180 return 0;
11181
11182 snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx);
11183
11184 if (determine_kprobe_perf_type() >= 0) {
11185 int pfd;
11186
11187 pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0);
11188 if (pfd >= 0)
11189 close(pfd);
11190
11191 return pfd >= 0 ? 1 : 0;
11192 } else { /* legacy mode */
11193 char probe_name[128];
11194
11195 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0);
11196 if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0)
11197 return 0;
11198
11199 (void)remove_kprobe_event_legacy(probe_name, false);
11200 return 1;
11201 }
11202 }
11203
11204 struct bpf_link *
bpf_program__attach_kprobe_opts(const struct bpf_program * prog,const char * func_name,const struct bpf_kprobe_opts * opts)11205 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
11206 const char *func_name,
11207 const struct bpf_kprobe_opts *opts)
11208 {
11209 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11210 enum probe_attach_mode attach_mode;
11211 char *legacy_probe = NULL;
11212 struct bpf_link *link;
11213 size_t offset;
11214 bool retprobe, legacy;
11215 int pfd, err;
11216
11217 if (!OPTS_VALID(opts, bpf_kprobe_opts))
11218 return libbpf_err_ptr(-EINVAL);
11219
11220 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
11221 retprobe = OPTS_GET(opts, retprobe, false);
11222 offset = OPTS_GET(opts, offset, 0);
11223 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11224
11225 legacy = determine_kprobe_perf_type() < 0;
11226 switch (attach_mode) {
11227 case PROBE_ATTACH_MODE_LEGACY:
11228 legacy = true;
11229 pe_opts.force_ioctl_attach = true;
11230 break;
11231 case PROBE_ATTACH_MODE_PERF:
11232 if (legacy)
11233 return libbpf_err_ptr(-ENOTSUP);
11234 pe_opts.force_ioctl_attach = true;
11235 break;
11236 case PROBE_ATTACH_MODE_LINK:
11237 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
11238 return libbpf_err_ptr(-ENOTSUP);
11239 break;
11240 case PROBE_ATTACH_MODE_DEFAULT:
11241 break;
11242 default:
11243 return libbpf_err_ptr(-EINVAL);
11244 }
11245
11246 if (!legacy) {
11247 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
11248 func_name, offset,
11249 -1 /* pid */, 0 /* ref_ctr_off */);
11250 } else {
11251 char probe_name[256];
11252
11253 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
11254 func_name, offset);
11255
11256 legacy_probe = strdup(probe_name);
11257 if (!legacy_probe)
11258 return libbpf_err_ptr(-ENOMEM);
11259
11260 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
11261 offset, -1 /* pid */);
11262 }
11263 if (pfd < 0) {
11264 err = -errno;
11265 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
11266 prog->name, retprobe ? "kretprobe" : "kprobe",
11267 func_name, offset,
11268 errstr(err));
11269 goto err_out;
11270 }
11271 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11272 err = libbpf_get_error(link);
11273 if (err) {
11274 close(pfd);
11275 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
11276 prog->name, retprobe ? "kretprobe" : "kprobe",
11277 func_name, offset,
11278 errstr(err));
11279 goto err_clean_legacy;
11280 }
11281 if (legacy) {
11282 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11283
11284 perf_link->legacy_probe_name = legacy_probe;
11285 perf_link->legacy_is_kprobe = true;
11286 perf_link->legacy_is_retprobe = retprobe;
11287 }
11288
11289 return link;
11290
11291 err_clean_legacy:
11292 if (legacy)
11293 remove_kprobe_event_legacy(legacy_probe, retprobe);
11294 err_out:
11295 free(legacy_probe);
11296 return libbpf_err_ptr(err);
11297 }
11298
bpf_program__attach_kprobe(const struct bpf_program * prog,bool retprobe,const char * func_name)11299 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
11300 bool retprobe,
11301 const char *func_name)
11302 {
11303 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
11304 .retprobe = retprobe,
11305 );
11306
11307 return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
11308 }
11309
bpf_program__attach_ksyscall(const struct bpf_program * prog,const char * syscall_name,const struct bpf_ksyscall_opts * opts)11310 struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog,
11311 const char *syscall_name,
11312 const struct bpf_ksyscall_opts *opts)
11313 {
11314 LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts);
11315 char func_name[128];
11316
11317 if (!OPTS_VALID(opts, bpf_ksyscall_opts))
11318 return libbpf_err_ptr(-EINVAL);
11319
11320 if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) {
11321 /* arch_specific_syscall_pfx() should never return NULL here
11322 * because it is guarded by kernel_supports(). However, since
11323 * compiler does not know that we have an explicit conditional
11324 * as well.
11325 */
11326 snprintf(func_name, sizeof(func_name), "__%s_sys_%s",
11327 arch_specific_syscall_pfx() ? : "", syscall_name);
11328 } else {
11329 snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name);
11330 }
11331
11332 kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false);
11333 kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11334
11335 return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts);
11336 }
11337
11338 /* Adapted from perf/util/string.c */
glob_match(const char * str,const char * pat)11339 bool glob_match(const char *str, const char *pat)
11340 {
11341 while (*str && *pat && *pat != '*') {
11342 if (*pat == '?') { /* Matches any single character */
11343 str++;
11344 pat++;
11345 continue;
11346 }
11347 if (*str != *pat)
11348 return false;
11349 str++;
11350 pat++;
11351 }
11352 /* Check wild card */
11353 if (*pat == '*') {
11354 while (*pat == '*')
11355 pat++;
11356 if (!*pat) /* Tail wild card matches all */
11357 return true;
11358 while (*str)
11359 if (glob_match(str++, pat))
11360 return true;
11361 }
11362 return !*str && !*pat;
11363 }
11364
11365 struct kprobe_multi_resolve {
11366 const char *pattern;
11367 unsigned long *addrs;
11368 size_t cap;
11369 size_t cnt;
11370 };
11371
11372 struct avail_kallsyms_data {
11373 char **syms;
11374 size_t cnt;
11375 struct kprobe_multi_resolve *res;
11376 };
11377
avail_func_cmp(const void * a,const void * b)11378 static int avail_func_cmp(const void *a, const void *b)
11379 {
11380 return strcmp(*(const char **)a, *(const char **)b);
11381 }
11382
avail_kallsyms_cb(unsigned long long sym_addr,char sym_type,const char * sym_name,void * ctx)11383 static int avail_kallsyms_cb(unsigned long long sym_addr, char sym_type,
11384 const char *sym_name, void *ctx)
11385 {
11386 struct avail_kallsyms_data *data = ctx;
11387 struct kprobe_multi_resolve *res = data->res;
11388 int err;
11389
11390 if (!glob_match(sym_name, res->pattern))
11391 return 0;
11392
11393 if (!bsearch(&sym_name, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp)) {
11394 /* Some versions of kernel strip out .llvm.<hash> suffix from
11395 * function names reported in available_filter_functions, but
11396 * don't do so for kallsyms. While this is clearly a kernel
11397 * bug (fixed by [0]) we try to accommodate that in libbpf to
11398 * make multi-kprobe usability a bit better: if no match is
11399 * found, we will strip .llvm. suffix and try one more time.
11400 *
11401 * [0] fb6a421fb615 ("kallsyms: Match symbols exactly with CONFIG_LTO_CLANG")
11402 */
11403 char sym_trim[256], *psym_trim = sym_trim, *sym_sfx;
11404
11405 if (!(sym_sfx = strstr(sym_name, ".llvm.")))
11406 return 0;
11407
11408 /* psym_trim vs sym_trim dance is done to avoid pointer vs array
11409 * coercion differences and get proper `const char **` pointer
11410 * which avail_func_cmp() expects
11411 */
11412 snprintf(sym_trim, sizeof(sym_trim), "%.*s", (int)(sym_sfx - sym_name), sym_name);
11413 if (!bsearch(&psym_trim, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp))
11414 return 0;
11415 }
11416
11417 err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, sizeof(*res->addrs), res->cnt + 1);
11418 if (err)
11419 return err;
11420
11421 res->addrs[res->cnt++] = (unsigned long)sym_addr;
11422 return 0;
11423 }
11424
libbpf_available_kallsyms_parse(struct kprobe_multi_resolve * res)11425 static int libbpf_available_kallsyms_parse(struct kprobe_multi_resolve *res)
11426 {
11427 const char *available_functions_file = tracefs_available_filter_functions();
11428 struct avail_kallsyms_data data;
11429 char sym_name[500];
11430 FILE *f;
11431 int err = 0, ret, i;
11432 char **syms = NULL;
11433 size_t cap = 0, cnt = 0;
11434
11435 f = fopen(available_functions_file, "re");
11436 if (!f) {
11437 err = -errno;
11438 pr_warn("failed to open %s: %s\n", available_functions_file, errstr(err));
11439 return err;
11440 }
11441
11442 while (true) {
11443 char *name;
11444
11445 ret = fscanf(f, "%499s%*[^\n]\n", sym_name);
11446 if (ret == EOF && feof(f))
11447 break;
11448
11449 if (ret != 1) {
11450 pr_warn("failed to parse available_filter_functions entry: %d\n", ret);
11451 err = -EINVAL;
11452 goto cleanup;
11453 }
11454
11455 if (!glob_match(sym_name, res->pattern))
11456 continue;
11457
11458 err = libbpf_ensure_mem((void **)&syms, &cap, sizeof(*syms), cnt + 1);
11459 if (err)
11460 goto cleanup;
11461
11462 name = strdup(sym_name);
11463 if (!name) {
11464 err = -errno;
11465 goto cleanup;
11466 }
11467
11468 syms[cnt++] = name;
11469 }
11470
11471 /* no entries found, bail out */
11472 if (cnt == 0) {
11473 err = -ENOENT;
11474 goto cleanup;
11475 }
11476
11477 /* sort available functions */
11478 qsort(syms, cnt, sizeof(*syms), avail_func_cmp);
11479
11480 data.syms = syms;
11481 data.res = res;
11482 data.cnt = cnt;
11483 libbpf_kallsyms_parse(avail_kallsyms_cb, &data);
11484
11485 if (res->cnt == 0)
11486 err = -ENOENT;
11487
11488 cleanup:
11489 for (i = 0; i < cnt; i++)
11490 free((char *)syms[i]);
11491 free(syms);
11492
11493 fclose(f);
11494 return err;
11495 }
11496
has_available_filter_functions_addrs(void)11497 static bool has_available_filter_functions_addrs(void)
11498 {
11499 return access(tracefs_available_filter_functions_addrs(), R_OK) != -1;
11500 }
11501
libbpf_available_kprobes_parse(struct kprobe_multi_resolve * res)11502 static int libbpf_available_kprobes_parse(struct kprobe_multi_resolve *res)
11503 {
11504 const char *available_path = tracefs_available_filter_functions_addrs();
11505 char sym_name[500];
11506 FILE *f;
11507 int ret, err = 0;
11508 unsigned long long sym_addr;
11509
11510 f = fopen(available_path, "re");
11511 if (!f) {
11512 err = -errno;
11513 pr_warn("failed to open %s: %s\n", available_path, errstr(err));
11514 return err;
11515 }
11516
11517 while (true) {
11518 ret = fscanf(f, "%llx %499s%*[^\n]\n", &sym_addr, sym_name);
11519 if (ret == EOF && feof(f))
11520 break;
11521
11522 if (ret != 2) {
11523 pr_warn("failed to parse available_filter_functions_addrs entry: %d\n",
11524 ret);
11525 err = -EINVAL;
11526 goto cleanup;
11527 }
11528
11529 if (!glob_match(sym_name, res->pattern))
11530 continue;
11531
11532 err = libbpf_ensure_mem((void **)&res->addrs, &res->cap,
11533 sizeof(*res->addrs), res->cnt + 1);
11534 if (err)
11535 goto cleanup;
11536
11537 res->addrs[res->cnt++] = (unsigned long)sym_addr;
11538 }
11539
11540 if (res->cnt == 0)
11541 err = -ENOENT;
11542
11543 cleanup:
11544 fclose(f);
11545 return err;
11546 }
11547
11548 struct bpf_link *
bpf_program__attach_kprobe_multi_opts(const struct bpf_program * prog,const char * pattern,const struct bpf_kprobe_multi_opts * opts)11549 bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
11550 const char *pattern,
11551 const struct bpf_kprobe_multi_opts *opts)
11552 {
11553 LIBBPF_OPTS(bpf_link_create_opts, lopts);
11554 struct kprobe_multi_resolve res = {
11555 .pattern = pattern,
11556 };
11557 enum bpf_attach_type attach_type;
11558 struct bpf_link *link = NULL;
11559 const unsigned long *addrs;
11560 int err, link_fd, prog_fd;
11561 bool retprobe, session, unique_match;
11562 const __u64 *cookies;
11563 const char **syms;
11564 size_t cnt;
11565
11566 if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
11567 return libbpf_err_ptr(-EINVAL);
11568
11569 prog_fd = bpf_program__fd(prog);
11570 if (prog_fd < 0) {
11571 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
11572 prog->name);
11573 return libbpf_err_ptr(-EINVAL);
11574 }
11575
11576 syms = OPTS_GET(opts, syms, false);
11577 addrs = OPTS_GET(opts, addrs, false);
11578 cnt = OPTS_GET(opts, cnt, false);
11579 cookies = OPTS_GET(opts, cookies, false);
11580 unique_match = OPTS_GET(opts, unique_match, false);
11581
11582 if (!pattern && !addrs && !syms)
11583 return libbpf_err_ptr(-EINVAL);
11584 if (pattern && (addrs || syms || cookies || cnt))
11585 return libbpf_err_ptr(-EINVAL);
11586 if (!pattern && !cnt)
11587 return libbpf_err_ptr(-EINVAL);
11588 if (!pattern && unique_match)
11589 return libbpf_err_ptr(-EINVAL);
11590 if (addrs && syms)
11591 return libbpf_err_ptr(-EINVAL);
11592
11593 if (pattern) {
11594 if (has_available_filter_functions_addrs())
11595 err = libbpf_available_kprobes_parse(&res);
11596 else
11597 err = libbpf_available_kallsyms_parse(&res);
11598 if (err)
11599 goto error;
11600
11601 if (unique_match && res.cnt != 1) {
11602 pr_warn("prog '%s': failed to find a unique match for '%s' (%zu matches)\n",
11603 prog->name, pattern, res.cnt);
11604 err = -EINVAL;
11605 goto error;
11606 }
11607
11608 addrs = res.addrs;
11609 cnt = res.cnt;
11610 }
11611
11612 retprobe = OPTS_GET(opts, retprobe, false);
11613 session = OPTS_GET(opts, session, false);
11614
11615 if (retprobe && session)
11616 return libbpf_err_ptr(-EINVAL);
11617
11618 attach_type = session ? BPF_TRACE_KPROBE_SESSION : BPF_TRACE_KPROBE_MULTI;
11619
11620 lopts.kprobe_multi.syms = syms;
11621 lopts.kprobe_multi.addrs = addrs;
11622 lopts.kprobe_multi.cookies = cookies;
11623 lopts.kprobe_multi.cnt = cnt;
11624 lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
11625
11626 link = calloc(1, sizeof(*link));
11627 if (!link) {
11628 err = -ENOMEM;
11629 goto error;
11630 }
11631 link->detach = &bpf_link__detach_fd;
11632
11633 link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts);
11634 if (link_fd < 0) {
11635 err = -errno;
11636 pr_warn("prog '%s': failed to attach: %s\n",
11637 prog->name, errstr(err));
11638 goto error;
11639 }
11640 link->fd = link_fd;
11641 free(res.addrs);
11642 return link;
11643
11644 error:
11645 free(link);
11646 free(res.addrs);
11647 return libbpf_err_ptr(err);
11648 }
11649
attach_kprobe(const struct bpf_program * prog,long cookie,struct bpf_link ** link)11650 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11651 {
11652 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
11653 unsigned long offset = 0;
11654 const char *func_name;
11655 char *func;
11656 int n;
11657
11658 *link = NULL;
11659
11660 /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */
11661 if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0)
11662 return 0;
11663
11664 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
11665 if (opts.retprobe)
11666 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
11667 else
11668 func_name = prog->sec_name + sizeof("kprobe/") - 1;
11669
11670 n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
11671 if (n < 1) {
11672 pr_warn("kprobe name is invalid: %s\n", func_name);
11673 return -EINVAL;
11674 }
11675 if (opts.retprobe && offset != 0) {
11676 free(func);
11677 pr_warn("kretprobes do not support offset specification\n");
11678 return -EINVAL;
11679 }
11680
11681 opts.offset = offset;
11682 *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
11683 free(func);
11684 return libbpf_get_error(*link);
11685 }
11686
attach_ksyscall(const struct bpf_program * prog,long cookie,struct bpf_link ** link)11687 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11688 {
11689 LIBBPF_OPTS(bpf_ksyscall_opts, opts);
11690 const char *syscall_name;
11691
11692 *link = NULL;
11693
11694 /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */
11695 if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0)
11696 return 0;
11697
11698 opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/");
11699 if (opts.retprobe)
11700 syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1;
11701 else
11702 syscall_name = prog->sec_name + sizeof("ksyscall/") - 1;
11703
11704 *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts);
11705 return *link ? 0 : -errno;
11706 }
11707
attach_kprobe_multi(const struct bpf_program * prog,long cookie,struct bpf_link ** link)11708 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11709 {
11710 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
11711 const char *spec;
11712 char *pattern;
11713 int n;
11714
11715 *link = NULL;
11716
11717 /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */
11718 if (strcmp(prog->sec_name, "kprobe.multi") == 0 ||
11719 strcmp(prog->sec_name, "kretprobe.multi") == 0)
11720 return 0;
11721
11722 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
11723 if (opts.retprobe)
11724 spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
11725 else
11726 spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
11727
11728 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
11729 if (n < 1) {
11730 pr_warn("kprobe multi pattern is invalid: %s\n", spec);
11731 return -EINVAL;
11732 }
11733
11734 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
11735 free(pattern);
11736 return libbpf_get_error(*link);
11737 }
11738
attach_kprobe_session(const struct bpf_program * prog,long cookie,struct bpf_link ** link)11739 static int attach_kprobe_session(const struct bpf_program *prog, long cookie,
11740 struct bpf_link **link)
11741 {
11742 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts, .session = true);
11743 const char *spec;
11744 char *pattern;
11745 int n;
11746
11747 *link = NULL;
11748
11749 /* no auto-attach for SEC("kprobe.session") */
11750 if (strcmp(prog->sec_name, "kprobe.session") == 0)
11751 return 0;
11752
11753 spec = prog->sec_name + sizeof("kprobe.session/") - 1;
11754 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
11755 if (n < 1) {
11756 pr_warn("kprobe session pattern is invalid: %s\n", spec);
11757 return -EINVAL;
11758 }
11759
11760 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
11761 free(pattern);
11762 return *link ? 0 : -errno;
11763 }
11764
attach_uprobe_multi(const struct bpf_program * prog,long cookie,struct bpf_link ** link)11765 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11766 {
11767 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL;
11768 LIBBPF_OPTS(bpf_uprobe_multi_opts, opts);
11769 int n, ret = -EINVAL;
11770
11771 *link = NULL;
11772
11773 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
11774 &probe_type, &binary_path, &func_name);
11775 switch (n) {
11776 case 1:
11777 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
11778 ret = 0;
11779 break;
11780 case 3:
11781 opts.session = str_has_pfx(probe_type, "uprobe.session");
11782 opts.retprobe = str_has_pfx(probe_type, "uretprobe.multi");
11783
11784 *link = bpf_program__attach_uprobe_multi(prog, -1, binary_path, func_name, &opts);
11785 ret = libbpf_get_error(*link);
11786 break;
11787 default:
11788 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
11789 prog->sec_name);
11790 break;
11791 }
11792 free(probe_type);
11793 free(binary_path);
11794 free(func_name);
11795 return ret;
11796 }
11797
gen_uprobe_legacy_event_name(char * buf,size_t buf_sz,const char * binary_path,uint64_t offset)11798 static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
11799 const char *binary_path, uint64_t offset)
11800 {
11801 int i;
11802
11803 snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
11804
11805 /* sanitize binary_path in the probe name */
11806 for (i = 0; buf[i]; i++) {
11807 if (!isalnum(buf[i]))
11808 buf[i] = '_';
11809 }
11810 }
11811
add_uprobe_event_legacy(const char * probe_name,bool retprobe,const char * binary_path,size_t offset)11812 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
11813 const char *binary_path, size_t offset)
11814 {
11815 return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx",
11816 retprobe ? 'r' : 'p',
11817 retprobe ? "uretprobes" : "uprobes",
11818 probe_name, binary_path, offset);
11819 }
11820
remove_uprobe_event_legacy(const char * probe_name,bool retprobe)11821 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
11822 {
11823 return append_to_file(tracefs_uprobe_events(), "-:%s/%s",
11824 retprobe ? "uretprobes" : "uprobes", probe_name);
11825 }
11826
determine_uprobe_perf_type_legacy(const char * probe_name,bool retprobe)11827 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
11828 {
11829 char file[512];
11830
11831 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11832 tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name);
11833
11834 return parse_uint_from_file(file, "%d\n");
11835 }
11836
perf_event_uprobe_open_legacy(const char * probe_name,bool retprobe,const char * binary_path,size_t offset,int pid)11837 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
11838 const char *binary_path, size_t offset, int pid)
11839 {
11840 const size_t attr_sz = sizeof(struct perf_event_attr);
11841 struct perf_event_attr attr;
11842 int type, pfd, err;
11843
11844 err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
11845 if (err < 0) {
11846 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %s\n",
11847 binary_path, (size_t)offset, errstr(err));
11848 return err;
11849 }
11850 type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
11851 if (type < 0) {
11852 err = type;
11853 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %s\n",
11854 binary_path, offset, errstr(err));
11855 goto err_clean_legacy;
11856 }
11857
11858 memset(&attr, 0, attr_sz);
11859 attr.size = attr_sz;
11860 attr.config = type;
11861 attr.type = PERF_TYPE_TRACEPOINT;
11862
11863 pfd = syscall(__NR_perf_event_open, &attr,
11864 pid < 0 ? -1 : pid, /* pid */
11865 pid == -1 ? 0 : -1, /* cpu */
11866 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11867 if (pfd < 0) {
11868 err = -errno;
11869 pr_warn("legacy uprobe perf_event_open() failed: %s\n", errstr(err));
11870 goto err_clean_legacy;
11871 }
11872 return pfd;
11873
11874 err_clean_legacy:
11875 /* Clear the newly added legacy uprobe_event */
11876 remove_uprobe_event_legacy(probe_name, retprobe);
11877 return err;
11878 }
11879
11880 /* Find offset of function name in archive specified by path. Currently
11881 * supported are .zip files that do not compress their contents, as used on
11882 * Android in the form of APKs, for example. "file_name" is the name of the ELF
11883 * file inside the archive. "func_name" matches symbol name or name@@LIB for
11884 * library functions.
11885 *
11886 * An overview of the APK format specifically provided here:
11887 * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents
11888 */
elf_find_func_offset_from_archive(const char * archive_path,const char * file_name,const char * func_name)11889 static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name,
11890 const char *func_name)
11891 {
11892 struct zip_archive *archive;
11893 struct zip_entry entry;
11894 long ret;
11895 Elf *elf;
11896
11897 archive = zip_archive_open(archive_path);
11898 if (IS_ERR(archive)) {
11899 ret = PTR_ERR(archive);
11900 pr_warn("zip: failed to open %s: %ld\n", archive_path, ret);
11901 return ret;
11902 }
11903
11904 ret = zip_archive_find_entry(archive, file_name, &entry);
11905 if (ret) {
11906 pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name,
11907 archive_path, ret);
11908 goto out;
11909 }
11910 pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path,
11911 (unsigned long)entry.data_offset);
11912
11913 if (entry.compression) {
11914 pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name,
11915 archive_path);
11916 ret = -LIBBPF_ERRNO__FORMAT;
11917 goto out;
11918 }
11919
11920 elf = elf_memory((void *)entry.data, entry.data_length);
11921 if (!elf) {
11922 pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path,
11923 elf_errmsg(-1));
11924 ret = -LIBBPF_ERRNO__LIBELF;
11925 goto out;
11926 }
11927
11928 ret = elf_find_func_offset(elf, file_name, func_name);
11929 if (ret > 0) {
11930 pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n",
11931 func_name, file_name, archive_path, entry.data_offset, ret,
11932 ret + entry.data_offset);
11933 ret += entry.data_offset;
11934 }
11935 elf_end(elf);
11936
11937 out:
11938 zip_archive_close(archive);
11939 return ret;
11940 }
11941
arch_specific_lib_paths(void)11942 static const char *arch_specific_lib_paths(void)
11943 {
11944 /*
11945 * Based on https://packages.debian.org/sid/libc6.
11946 *
11947 * Assume that the traced program is built for the same architecture
11948 * as libbpf, which should cover the vast majority of cases.
11949 */
11950 #if defined(__x86_64__)
11951 return "/lib/x86_64-linux-gnu";
11952 #elif defined(__i386__)
11953 return "/lib/i386-linux-gnu";
11954 #elif defined(__s390x__)
11955 return "/lib/s390x-linux-gnu";
11956 #elif defined(__s390__)
11957 return "/lib/s390-linux-gnu";
11958 #elif defined(__arm__) && defined(__SOFTFP__)
11959 return "/lib/arm-linux-gnueabi";
11960 #elif defined(__arm__) && !defined(__SOFTFP__)
11961 return "/lib/arm-linux-gnueabihf";
11962 #elif defined(__aarch64__)
11963 return "/lib/aarch64-linux-gnu";
11964 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64
11965 return "/lib/mips64el-linux-gnuabi64";
11966 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32
11967 return "/lib/mipsel-linux-gnu";
11968 #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
11969 return "/lib/powerpc64le-linux-gnu";
11970 #elif defined(__sparc__) && defined(__arch64__)
11971 return "/lib/sparc64-linux-gnu";
11972 #elif defined(__riscv) && __riscv_xlen == 64
11973 return "/lib/riscv64-linux-gnu";
11974 #else
11975 return NULL;
11976 #endif
11977 }
11978
11979 /* Get full path to program/shared library. */
resolve_full_path(const char * file,char * result,size_t result_sz)11980 static int resolve_full_path(const char *file, char *result, size_t result_sz)
11981 {
11982 const char *search_paths[3] = {};
11983 int i, perm;
11984
11985 if (str_has_sfx(file, ".so") || strstr(file, ".so.")) {
11986 search_paths[0] = getenv("LD_LIBRARY_PATH");
11987 search_paths[1] = "/usr/lib64:/usr/lib";
11988 search_paths[2] = arch_specific_lib_paths();
11989 perm = R_OK;
11990 } else {
11991 search_paths[0] = getenv("PATH");
11992 search_paths[1] = "/usr/bin:/usr/sbin";
11993 perm = R_OK | X_OK;
11994 }
11995
11996 for (i = 0; i < ARRAY_SIZE(search_paths); i++) {
11997 const char *s;
11998
11999 if (!search_paths[i])
12000 continue;
12001 for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) {
12002 char *next_path;
12003 int seg_len;
12004
12005 if (s[0] == ':')
12006 s++;
12007 next_path = strchr(s, ':');
12008 seg_len = next_path ? next_path - s : strlen(s);
12009 if (!seg_len)
12010 continue;
12011 snprintf(result, result_sz, "%.*s/%s", seg_len, s, file);
12012 /* ensure it has required permissions */
12013 if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0)
12014 continue;
12015 pr_debug("resolved '%s' to '%s'\n", file, result);
12016 return 0;
12017 }
12018 }
12019 return -ENOENT;
12020 }
12021
12022 struct bpf_link *
bpf_program__attach_uprobe_multi(const struct bpf_program * prog,pid_t pid,const char * path,const char * func_pattern,const struct bpf_uprobe_multi_opts * opts)12023 bpf_program__attach_uprobe_multi(const struct bpf_program *prog,
12024 pid_t pid,
12025 const char *path,
12026 const char *func_pattern,
12027 const struct bpf_uprobe_multi_opts *opts)
12028 {
12029 const unsigned long *ref_ctr_offsets = NULL, *offsets = NULL;
12030 LIBBPF_OPTS(bpf_link_create_opts, lopts);
12031 unsigned long *resolved_offsets = NULL;
12032 enum bpf_attach_type attach_type;
12033 int err = 0, link_fd, prog_fd;
12034 struct bpf_link *link = NULL;
12035 char full_path[PATH_MAX];
12036 bool retprobe, session;
12037 const __u64 *cookies;
12038 const char **syms;
12039 size_t cnt;
12040
12041 if (!OPTS_VALID(opts, bpf_uprobe_multi_opts))
12042 return libbpf_err_ptr(-EINVAL);
12043
12044 prog_fd = bpf_program__fd(prog);
12045 if (prog_fd < 0) {
12046 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12047 prog->name);
12048 return libbpf_err_ptr(-EINVAL);
12049 }
12050
12051 syms = OPTS_GET(opts, syms, NULL);
12052 offsets = OPTS_GET(opts, offsets, NULL);
12053 ref_ctr_offsets = OPTS_GET(opts, ref_ctr_offsets, NULL);
12054 cookies = OPTS_GET(opts, cookies, NULL);
12055 cnt = OPTS_GET(opts, cnt, 0);
12056 retprobe = OPTS_GET(opts, retprobe, false);
12057 session = OPTS_GET(opts, session, false);
12058
12059 /*
12060 * User can specify 2 mutually exclusive set of inputs:
12061 *
12062 * 1) use only path/func_pattern/pid arguments
12063 *
12064 * 2) use path/pid with allowed combinations of:
12065 * syms/offsets/ref_ctr_offsets/cookies/cnt
12066 *
12067 * - syms and offsets are mutually exclusive
12068 * - ref_ctr_offsets and cookies are optional
12069 *
12070 * Any other usage results in error.
12071 */
12072
12073 if (!path)
12074 return libbpf_err_ptr(-EINVAL);
12075 if (!func_pattern && cnt == 0)
12076 return libbpf_err_ptr(-EINVAL);
12077
12078 if (func_pattern) {
12079 if (syms || offsets || ref_ctr_offsets || cookies || cnt)
12080 return libbpf_err_ptr(-EINVAL);
12081 } else {
12082 if (!!syms == !!offsets)
12083 return libbpf_err_ptr(-EINVAL);
12084 }
12085
12086 if (retprobe && session)
12087 return libbpf_err_ptr(-EINVAL);
12088
12089 if (func_pattern) {
12090 if (!strchr(path, '/')) {
12091 err = resolve_full_path(path, full_path, sizeof(full_path));
12092 if (err) {
12093 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12094 prog->name, path, errstr(err));
12095 return libbpf_err_ptr(err);
12096 }
12097 path = full_path;
12098 }
12099
12100 err = elf_resolve_pattern_offsets(path, func_pattern,
12101 &resolved_offsets, &cnt);
12102 if (err < 0)
12103 return libbpf_err_ptr(err);
12104 offsets = resolved_offsets;
12105 } else if (syms) {
12106 err = elf_resolve_syms_offsets(path, cnt, syms, &resolved_offsets, STT_FUNC);
12107 if (err < 0)
12108 return libbpf_err_ptr(err);
12109 offsets = resolved_offsets;
12110 }
12111
12112 attach_type = session ? BPF_TRACE_UPROBE_SESSION : BPF_TRACE_UPROBE_MULTI;
12113
12114 lopts.uprobe_multi.path = path;
12115 lopts.uprobe_multi.offsets = offsets;
12116 lopts.uprobe_multi.ref_ctr_offsets = ref_ctr_offsets;
12117 lopts.uprobe_multi.cookies = cookies;
12118 lopts.uprobe_multi.cnt = cnt;
12119 lopts.uprobe_multi.flags = retprobe ? BPF_F_UPROBE_MULTI_RETURN : 0;
12120
12121 if (pid == 0)
12122 pid = getpid();
12123 if (pid > 0)
12124 lopts.uprobe_multi.pid = pid;
12125
12126 link = calloc(1, sizeof(*link));
12127 if (!link) {
12128 err = -ENOMEM;
12129 goto error;
12130 }
12131 link->detach = &bpf_link__detach_fd;
12132
12133 link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts);
12134 if (link_fd < 0) {
12135 err = -errno;
12136 pr_warn("prog '%s': failed to attach multi-uprobe: %s\n",
12137 prog->name, errstr(err));
12138 goto error;
12139 }
12140 link->fd = link_fd;
12141 free(resolved_offsets);
12142 return link;
12143
12144 error:
12145 free(resolved_offsets);
12146 free(link);
12147 return libbpf_err_ptr(err);
12148 }
12149
12150 LIBBPF_API struct bpf_link *
bpf_program__attach_uprobe_opts(const struct bpf_program * prog,pid_t pid,const char * binary_path,size_t func_offset,const struct bpf_uprobe_opts * opts)12151 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
12152 const char *binary_path, size_t func_offset,
12153 const struct bpf_uprobe_opts *opts)
12154 {
12155 const char *archive_path = NULL, *archive_sep = NULL;
12156 char *legacy_probe = NULL;
12157 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
12158 enum probe_attach_mode attach_mode;
12159 char full_path[PATH_MAX];
12160 struct bpf_link *link;
12161 size_t ref_ctr_off;
12162 int pfd, err;
12163 bool retprobe, legacy;
12164 const char *func_name;
12165
12166 if (!OPTS_VALID(opts, bpf_uprobe_opts))
12167 return libbpf_err_ptr(-EINVAL);
12168
12169 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
12170 retprobe = OPTS_GET(opts, retprobe, false);
12171 ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
12172 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
12173
12174 if (!binary_path)
12175 return libbpf_err_ptr(-EINVAL);
12176
12177 /* Check if "binary_path" refers to an archive. */
12178 archive_sep = strstr(binary_path, "!/");
12179 if (archive_sep) {
12180 full_path[0] = '\0';
12181 libbpf_strlcpy(full_path, binary_path,
12182 min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1)));
12183 archive_path = full_path;
12184 binary_path = archive_sep + 2;
12185 } else if (!strchr(binary_path, '/')) {
12186 err = resolve_full_path(binary_path, full_path, sizeof(full_path));
12187 if (err) {
12188 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12189 prog->name, binary_path, errstr(err));
12190 return libbpf_err_ptr(err);
12191 }
12192 binary_path = full_path;
12193 }
12194 func_name = OPTS_GET(opts, func_name, NULL);
12195 if (func_name) {
12196 long sym_off;
12197
12198 if (archive_path) {
12199 sym_off = elf_find_func_offset_from_archive(archive_path, binary_path,
12200 func_name);
12201 binary_path = archive_path;
12202 } else {
12203 sym_off = elf_find_func_offset_from_file(binary_path, func_name);
12204 }
12205 if (sym_off < 0)
12206 return libbpf_err_ptr(sym_off);
12207 func_offset += sym_off;
12208 }
12209
12210 legacy = determine_uprobe_perf_type() < 0;
12211 switch (attach_mode) {
12212 case PROBE_ATTACH_MODE_LEGACY:
12213 legacy = true;
12214 pe_opts.force_ioctl_attach = true;
12215 break;
12216 case PROBE_ATTACH_MODE_PERF:
12217 if (legacy)
12218 return libbpf_err_ptr(-ENOTSUP);
12219 pe_opts.force_ioctl_attach = true;
12220 break;
12221 case PROBE_ATTACH_MODE_LINK:
12222 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
12223 return libbpf_err_ptr(-ENOTSUP);
12224 break;
12225 case PROBE_ATTACH_MODE_DEFAULT:
12226 break;
12227 default:
12228 return libbpf_err_ptr(-EINVAL);
12229 }
12230
12231 if (!legacy) {
12232 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
12233 func_offset, pid, ref_ctr_off);
12234 } else {
12235 char probe_name[PATH_MAX + 64];
12236
12237 if (ref_ctr_off)
12238 return libbpf_err_ptr(-EINVAL);
12239
12240 gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
12241 binary_path, func_offset);
12242
12243 legacy_probe = strdup(probe_name);
12244 if (!legacy_probe)
12245 return libbpf_err_ptr(-ENOMEM);
12246
12247 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
12248 binary_path, func_offset, pid);
12249 }
12250 if (pfd < 0) {
12251 err = -errno;
12252 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
12253 prog->name, retprobe ? "uretprobe" : "uprobe",
12254 binary_path, func_offset,
12255 errstr(err));
12256 goto err_out;
12257 }
12258
12259 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
12260 err = libbpf_get_error(link);
12261 if (err) {
12262 close(pfd);
12263 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
12264 prog->name, retprobe ? "uretprobe" : "uprobe",
12265 binary_path, func_offset,
12266 errstr(err));
12267 goto err_clean_legacy;
12268 }
12269 if (legacy) {
12270 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
12271
12272 perf_link->legacy_probe_name = legacy_probe;
12273 perf_link->legacy_is_kprobe = false;
12274 perf_link->legacy_is_retprobe = retprobe;
12275 }
12276 return link;
12277
12278 err_clean_legacy:
12279 if (legacy)
12280 remove_uprobe_event_legacy(legacy_probe, retprobe);
12281 err_out:
12282 free(legacy_probe);
12283 return libbpf_err_ptr(err);
12284 }
12285
12286 /* Format of u[ret]probe section definition supporting auto-attach:
12287 * u[ret]probe/binary:function[+offset]
12288 *
12289 * binary can be an absolute/relative path or a filename; the latter is resolved to a
12290 * full binary path via bpf_program__attach_uprobe_opts.
12291 *
12292 * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be
12293 * specified (and auto-attach is not possible) or the above format is specified for
12294 * auto-attach.
12295 */
attach_uprobe(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12296 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12297 {
12298 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts);
12299 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL, *func_off;
12300 int n, c, ret = -EINVAL;
12301 long offset = 0;
12302
12303 *link = NULL;
12304
12305 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
12306 &probe_type, &binary_path, &func_name);
12307 switch (n) {
12308 case 1:
12309 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
12310 ret = 0;
12311 break;
12312 case 2:
12313 pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n",
12314 prog->name, prog->sec_name);
12315 break;
12316 case 3:
12317 /* check if user specifies `+offset`, if yes, this should be
12318 * the last part of the string, make sure sscanf read to EOL
12319 */
12320 func_off = strrchr(func_name, '+');
12321 if (func_off) {
12322 n = sscanf(func_off, "+%li%n", &offset, &c);
12323 if (n == 1 && *(func_off + c) == '\0')
12324 func_off[0] = '\0';
12325 else
12326 offset = 0;
12327 }
12328 opts.retprobe = strcmp(probe_type, "uretprobe") == 0 ||
12329 strcmp(probe_type, "uretprobe.s") == 0;
12330 if (opts.retprobe && offset != 0) {
12331 pr_warn("prog '%s': uretprobes do not support offset specification\n",
12332 prog->name);
12333 break;
12334 }
12335 opts.func_name = func_name;
12336 *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts);
12337 ret = libbpf_get_error(*link);
12338 break;
12339 default:
12340 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
12341 prog->sec_name);
12342 break;
12343 }
12344 free(probe_type);
12345 free(binary_path);
12346 free(func_name);
12347
12348 return ret;
12349 }
12350
bpf_program__attach_uprobe(const struct bpf_program * prog,bool retprobe,pid_t pid,const char * binary_path,size_t func_offset)12351 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
12352 bool retprobe, pid_t pid,
12353 const char *binary_path,
12354 size_t func_offset)
12355 {
12356 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
12357
12358 return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
12359 }
12360
bpf_program__attach_usdt(const struct bpf_program * prog,pid_t pid,const char * binary_path,const char * usdt_provider,const char * usdt_name,const struct bpf_usdt_opts * opts)12361 struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog,
12362 pid_t pid, const char *binary_path,
12363 const char *usdt_provider, const char *usdt_name,
12364 const struct bpf_usdt_opts *opts)
12365 {
12366 char resolved_path[512];
12367 struct bpf_object *obj = prog->obj;
12368 struct bpf_link *link;
12369 __u64 usdt_cookie;
12370 int err;
12371
12372 if (!OPTS_VALID(opts, bpf_uprobe_opts))
12373 return libbpf_err_ptr(-EINVAL);
12374
12375 if (bpf_program__fd(prog) < 0) {
12376 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12377 prog->name);
12378 return libbpf_err_ptr(-EINVAL);
12379 }
12380
12381 if (!binary_path)
12382 return libbpf_err_ptr(-EINVAL);
12383
12384 if (!strchr(binary_path, '/')) {
12385 err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path));
12386 if (err) {
12387 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12388 prog->name, binary_path, errstr(err));
12389 return libbpf_err_ptr(err);
12390 }
12391 binary_path = resolved_path;
12392 }
12393
12394 /* USDT manager is instantiated lazily on first USDT attach. It will
12395 * be destroyed together with BPF object in bpf_object__close().
12396 */
12397 if (IS_ERR(obj->usdt_man))
12398 return libbpf_ptr(obj->usdt_man);
12399 if (!obj->usdt_man) {
12400 obj->usdt_man = usdt_manager_new(obj);
12401 if (IS_ERR(obj->usdt_man))
12402 return libbpf_ptr(obj->usdt_man);
12403 }
12404
12405 usdt_cookie = OPTS_GET(opts, usdt_cookie, 0);
12406 link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path,
12407 usdt_provider, usdt_name, usdt_cookie);
12408 err = libbpf_get_error(link);
12409 if (err)
12410 return libbpf_err_ptr(err);
12411 return link;
12412 }
12413
attach_usdt(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12414 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12415 {
12416 char *path = NULL, *provider = NULL, *name = NULL;
12417 const char *sec_name;
12418 int n, err;
12419
12420 sec_name = bpf_program__section_name(prog);
12421 if (strcmp(sec_name, "usdt") == 0) {
12422 /* no auto-attach for just SEC("usdt") */
12423 *link = NULL;
12424 return 0;
12425 }
12426
12427 n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name);
12428 if (n != 3) {
12429 pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n",
12430 sec_name);
12431 err = -EINVAL;
12432 } else {
12433 *link = bpf_program__attach_usdt(prog, -1 /* any process */, path,
12434 provider, name, NULL);
12435 err = libbpf_get_error(*link);
12436 }
12437 free(path);
12438 free(provider);
12439 free(name);
12440 return err;
12441 }
12442
determine_tracepoint_id(const char * tp_category,const char * tp_name)12443 static int determine_tracepoint_id(const char *tp_category,
12444 const char *tp_name)
12445 {
12446 char file[PATH_MAX];
12447 int ret;
12448
12449 ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id",
12450 tracefs_path(), tp_category, tp_name);
12451 if (ret < 0)
12452 return -errno;
12453 if (ret >= sizeof(file)) {
12454 pr_debug("tracepoint %s/%s path is too long\n",
12455 tp_category, tp_name);
12456 return -E2BIG;
12457 }
12458 return parse_uint_from_file(file, "%d\n");
12459 }
12460
perf_event_open_tracepoint(const char * tp_category,const char * tp_name)12461 static int perf_event_open_tracepoint(const char *tp_category,
12462 const char *tp_name)
12463 {
12464 const size_t attr_sz = sizeof(struct perf_event_attr);
12465 struct perf_event_attr attr;
12466 int tp_id, pfd, err;
12467
12468 tp_id = determine_tracepoint_id(tp_category, tp_name);
12469 if (tp_id < 0) {
12470 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
12471 tp_category, tp_name,
12472 errstr(tp_id));
12473 return tp_id;
12474 }
12475
12476 memset(&attr, 0, attr_sz);
12477 attr.type = PERF_TYPE_TRACEPOINT;
12478 attr.size = attr_sz;
12479 attr.config = tp_id;
12480
12481 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
12482 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
12483 if (pfd < 0) {
12484 err = -errno;
12485 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
12486 tp_category, tp_name,
12487 errstr(err));
12488 return err;
12489 }
12490 return pfd;
12491 }
12492
bpf_program__attach_tracepoint_opts(const struct bpf_program * prog,const char * tp_category,const char * tp_name,const struct bpf_tracepoint_opts * opts)12493 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
12494 const char *tp_category,
12495 const char *tp_name,
12496 const struct bpf_tracepoint_opts *opts)
12497 {
12498 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
12499 struct bpf_link *link;
12500 int pfd, err;
12501
12502 if (!OPTS_VALID(opts, bpf_tracepoint_opts))
12503 return libbpf_err_ptr(-EINVAL);
12504
12505 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
12506
12507 pfd = perf_event_open_tracepoint(tp_category, tp_name);
12508 if (pfd < 0) {
12509 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
12510 prog->name, tp_category, tp_name,
12511 errstr(pfd));
12512 return libbpf_err_ptr(pfd);
12513 }
12514 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
12515 err = libbpf_get_error(link);
12516 if (err) {
12517 close(pfd);
12518 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
12519 prog->name, tp_category, tp_name,
12520 errstr(err));
12521 return libbpf_err_ptr(err);
12522 }
12523 return link;
12524 }
12525
bpf_program__attach_tracepoint(const struct bpf_program * prog,const char * tp_category,const char * tp_name)12526 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
12527 const char *tp_category,
12528 const char *tp_name)
12529 {
12530 return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
12531 }
12532
attach_tp(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12533 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12534 {
12535 char *sec_name, *tp_cat, *tp_name;
12536
12537 *link = NULL;
12538
12539 /* no auto-attach for SEC("tp") or SEC("tracepoint") */
12540 if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0)
12541 return 0;
12542
12543 sec_name = strdup(prog->sec_name);
12544 if (!sec_name)
12545 return -ENOMEM;
12546
12547 /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
12548 if (str_has_pfx(prog->sec_name, "tp/"))
12549 tp_cat = sec_name + sizeof("tp/") - 1;
12550 else
12551 tp_cat = sec_name + sizeof("tracepoint/") - 1;
12552 tp_name = strchr(tp_cat, '/');
12553 if (!tp_name) {
12554 free(sec_name);
12555 return -EINVAL;
12556 }
12557 *tp_name = '\0';
12558 tp_name++;
12559
12560 *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
12561 free(sec_name);
12562 return libbpf_get_error(*link);
12563 }
12564
12565 struct bpf_link *
bpf_program__attach_raw_tracepoint_opts(const struct bpf_program * prog,const char * tp_name,struct bpf_raw_tracepoint_opts * opts)12566 bpf_program__attach_raw_tracepoint_opts(const struct bpf_program *prog,
12567 const char *tp_name,
12568 struct bpf_raw_tracepoint_opts *opts)
12569 {
12570 LIBBPF_OPTS(bpf_raw_tp_opts, raw_opts);
12571 struct bpf_link *link;
12572 int prog_fd, pfd;
12573
12574 if (!OPTS_VALID(opts, bpf_raw_tracepoint_opts))
12575 return libbpf_err_ptr(-EINVAL);
12576
12577 prog_fd = bpf_program__fd(prog);
12578 if (prog_fd < 0) {
12579 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12580 return libbpf_err_ptr(-EINVAL);
12581 }
12582
12583 link = calloc(1, sizeof(*link));
12584 if (!link)
12585 return libbpf_err_ptr(-ENOMEM);
12586 link->detach = &bpf_link__detach_fd;
12587
12588 raw_opts.tp_name = tp_name;
12589 raw_opts.cookie = OPTS_GET(opts, cookie, 0);
12590 pfd = bpf_raw_tracepoint_open_opts(prog_fd, &raw_opts);
12591 if (pfd < 0) {
12592 pfd = -errno;
12593 free(link);
12594 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
12595 prog->name, tp_name, errstr(pfd));
12596 return libbpf_err_ptr(pfd);
12597 }
12598 link->fd = pfd;
12599 return link;
12600 }
12601
bpf_program__attach_raw_tracepoint(const struct bpf_program * prog,const char * tp_name)12602 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
12603 const char *tp_name)
12604 {
12605 return bpf_program__attach_raw_tracepoint_opts(prog, tp_name, NULL);
12606 }
12607
attach_raw_tp(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12608 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12609 {
12610 static const char *const prefixes[] = {
12611 "raw_tp",
12612 "raw_tracepoint",
12613 "raw_tp.w",
12614 "raw_tracepoint.w",
12615 };
12616 size_t i;
12617 const char *tp_name = NULL;
12618
12619 *link = NULL;
12620
12621 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
12622 size_t pfx_len;
12623
12624 if (!str_has_pfx(prog->sec_name, prefixes[i]))
12625 continue;
12626
12627 pfx_len = strlen(prefixes[i]);
12628 /* no auto-attach case of, e.g., SEC("raw_tp") */
12629 if (prog->sec_name[pfx_len] == '\0')
12630 return 0;
12631
12632 if (prog->sec_name[pfx_len] != '/')
12633 continue;
12634
12635 tp_name = prog->sec_name + pfx_len + 1;
12636 break;
12637 }
12638
12639 if (!tp_name) {
12640 pr_warn("prog '%s': invalid section name '%s'\n",
12641 prog->name, prog->sec_name);
12642 return -EINVAL;
12643 }
12644
12645 *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
12646 return libbpf_get_error(*link);
12647 }
12648
12649 /* Common logic for all BPF program types that attach to a btf_id */
bpf_program__attach_btf_id(const struct bpf_program * prog,const struct bpf_trace_opts * opts)12650 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog,
12651 const struct bpf_trace_opts *opts)
12652 {
12653 LIBBPF_OPTS(bpf_link_create_opts, link_opts);
12654 struct bpf_link *link;
12655 int prog_fd, pfd;
12656
12657 if (!OPTS_VALID(opts, bpf_trace_opts))
12658 return libbpf_err_ptr(-EINVAL);
12659
12660 prog_fd = bpf_program__fd(prog);
12661 if (prog_fd < 0) {
12662 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12663 return libbpf_err_ptr(-EINVAL);
12664 }
12665
12666 link = calloc(1, sizeof(*link));
12667 if (!link)
12668 return libbpf_err_ptr(-ENOMEM);
12669 link->detach = &bpf_link__detach_fd;
12670
12671 /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */
12672 link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0);
12673 pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts);
12674 if (pfd < 0) {
12675 pfd = -errno;
12676 free(link);
12677 pr_warn("prog '%s': failed to attach: %s\n",
12678 prog->name, errstr(pfd));
12679 return libbpf_err_ptr(pfd);
12680 }
12681 link->fd = pfd;
12682 return link;
12683 }
12684
bpf_program__attach_trace(const struct bpf_program * prog)12685 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
12686 {
12687 return bpf_program__attach_btf_id(prog, NULL);
12688 }
12689
bpf_program__attach_trace_opts(const struct bpf_program * prog,const struct bpf_trace_opts * opts)12690 struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog,
12691 const struct bpf_trace_opts *opts)
12692 {
12693 return bpf_program__attach_btf_id(prog, opts);
12694 }
12695
bpf_program__attach_lsm(const struct bpf_program * prog)12696 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
12697 {
12698 return bpf_program__attach_btf_id(prog, NULL);
12699 }
12700
attach_trace(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12701 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12702 {
12703 *link = bpf_program__attach_trace(prog);
12704 return libbpf_get_error(*link);
12705 }
12706
attach_lsm(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12707 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12708 {
12709 *link = bpf_program__attach_lsm(prog);
12710 return libbpf_get_error(*link);
12711 }
12712
12713 static struct bpf_link *
bpf_program_attach_fd(const struct bpf_program * prog,int target_fd,const char * target_name,const struct bpf_link_create_opts * opts)12714 bpf_program_attach_fd(const struct bpf_program *prog,
12715 int target_fd, const char *target_name,
12716 const struct bpf_link_create_opts *opts)
12717 {
12718 enum bpf_attach_type attach_type;
12719 struct bpf_link *link;
12720 int prog_fd, link_fd;
12721
12722 prog_fd = bpf_program__fd(prog);
12723 if (prog_fd < 0) {
12724 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12725 return libbpf_err_ptr(-EINVAL);
12726 }
12727
12728 link = calloc(1, sizeof(*link));
12729 if (!link)
12730 return libbpf_err_ptr(-ENOMEM);
12731 link->detach = &bpf_link__detach_fd;
12732
12733 attach_type = bpf_program__expected_attach_type(prog);
12734 link_fd = bpf_link_create(prog_fd, target_fd, attach_type, opts);
12735 if (link_fd < 0) {
12736 link_fd = -errno;
12737 free(link);
12738 pr_warn("prog '%s': failed to attach to %s: %s\n",
12739 prog->name, target_name,
12740 errstr(link_fd));
12741 return libbpf_err_ptr(link_fd);
12742 }
12743 link->fd = link_fd;
12744 return link;
12745 }
12746
12747 struct bpf_link *
bpf_program__attach_cgroup(const struct bpf_program * prog,int cgroup_fd)12748 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
12749 {
12750 return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", NULL);
12751 }
12752
12753 struct bpf_link *
bpf_program__attach_netns(const struct bpf_program * prog,int netns_fd)12754 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
12755 {
12756 return bpf_program_attach_fd(prog, netns_fd, "netns", NULL);
12757 }
12758
12759 struct bpf_link *
bpf_program__attach_sockmap(const struct bpf_program * prog,int map_fd)12760 bpf_program__attach_sockmap(const struct bpf_program *prog, int map_fd)
12761 {
12762 return bpf_program_attach_fd(prog, map_fd, "sockmap", NULL);
12763 }
12764
bpf_program__attach_xdp(const struct bpf_program * prog,int ifindex)12765 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
12766 {
12767 /* target_fd/target_ifindex use the same field in LINK_CREATE */
12768 return bpf_program_attach_fd(prog, ifindex, "xdp", NULL);
12769 }
12770
12771 struct bpf_link *
bpf_program__attach_tcx(const struct bpf_program * prog,int ifindex,const struct bpf_tcx_opts * opts)12772 bpf_program__attach_tcx(const struct bpf_program *prog, int ifindex,
12773 const struct bpf_tcx_opts *opts)
12774 {
12775 LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12776 __u32 relative_id;
12777 int relative_fd;
12778
12779 if (!OPTS_VALID(opts, bpf_tcx_opts))
12780 return libbpf_err_ptr(-EINVAL);
12781
12782 relative_id = OPTS_GET(opts, relative_id, 0);
12783 relative_fd = OPTS_GET(opts, relative_fd, 0);
12784
12785 /* validate we don't have unexpected combinations of non-zero fields */
12786 if (!ifindex) {
12787 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
12788 prog->name);
12789 return libbpf_err_ptr(-EINVAL);
12790 }
12791 if (relative_fd && relative_id) {
12792 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
12793 prog->name);
12794 return libbpf_err_ptr(-EINVAL);
12795 }
12796
12797 link_create_opts.tcx.expected_revision = OPTS_GET(opts, expected_revision, 0);
12798 link_create_opts.tcx.relative_fd = relative_fd;
12799 link_create_opts.tcx.relative_id = relative_id;
12800 link_create_opts.flags = OPTS_GET(opts, flags, 0);
12801
12802 /* target_fd/target_ifindex use the same field in LINK_CREATE */
12803 return bpf_program_attach_fd(prog, ifindex, "tcx", &link_create_opts);
12804 }
12805
12806 struct bpf_link *
bpf_program__attach_netkit(const struct bpf_program * prog,int ifindex,const struct bpf_netkit_opts * opts)12807 bpf_program__attach_netkit(const struct bpf_program *prog, int ifindex,
12808 const struct bpf_netkit_opts *opts)
12809 {
12810 LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12811 __u32 relative_id;
12812 int relative_fd;
12813
12814 if (!OPTS_VALID(opts, bpf_netkit_opts))
12815 return libbpf_err_ptr(-EINVAL);
12816
12817 relative_id = OPTS_GET(opts, relative_id, 0);
12818 relative_fd = OPTS_GET(opts, relative_fd, 0);
12819
12820 /* validate we don't have unexpected combinations of non-zero fields */
12821 if (!ifindex) {
12822 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
12823 prog->name);
12824 return libbpf_err_ptr(-EINVAL);
12825 }
12826 if (relative_fd && relative_id) {
12827 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
12828 prog->name);
12829 return libbpf_err_ptr(-EINVAL);
12830 }
12831
12832 link_create_opts.netkit.expected_revision = OPTS_GET(opts, expected_revision, 0);
12833 link_create_opts.netkit.relative_fd = relative_fd;
12834 link_create_opts.netkit.relative_id = relative_id;
12835 link_create_opts.flags = OPTS_GET(opts, flags, 0);
12836
12837 return bpf_program_attach_fd(prog, ifindex, "netkit", &link_create_opts);
12838 }
12839
bpf_program__attach_freplace(const struct bpf_program * prog,int target_fd,const char * attach_func_name)12840 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
12841 int target_fd,
12842 const char *attach_func_name)
12843 {
12844 int btf_id;
12845
12846 if (!!target_fd != !!attach_func_name) {
12847 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
12848 prog->name);
12849 return libbpf_err_ptr(-EINVAL);
12850 }
12851
12852 if (prog->type != BPF_PROG_TYPE_EXT) {
12853 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace\n",
12854 prog->name);
12855 return libbpf_err_ptr(-EINVAL);
12856 }
12857
12858 if (target_fd) {
12859 LIBBPF_OPTS(bpf_link_create_opts, target_opts);
12860
12861 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
12862 if (btf_id < 0)
12863 return libbpf_err_ptr(btf_id);
12864
12865 target_opts.target_btf_id = btf_id;
12866
12867 return bpf_program_attach_fd(prog, target_fd, "freplace",
12868 &target_opts);
12869 } else {
12870 /* no target, so use raw_tracepoint_open for compatibility
12871 * with old kernels
12872 */
12873 return bpf_program__attach_trace(prog);
12874 }
12875 }
12876
12877 struct bpf_link *
bpf_program__attach_iter(const struct bpf_program * prog,const struct bpf_iter_attach_opts * opts)12878 bpf_program__attach_iter(const struct bpf_program *prog,
12879 const struct bpf_iter_attach_opts *opts)
12880 {
12881 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12882 struct bpf_link *link;
12883 int prog_fd, link_fd;
12884 __u32 target_fd = 0;
12885
12886 if (!OPTS_VALID(opts, bpf_iter_attach_opts))
12887 return libbpf_err_ptr(-EINVAL);
12888
12889 link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
12890 link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
12891
12892 prog_fd = bpf_program__fd(prog);
12893 if (prog_fd < 0) {
12894 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12895 return libbpf_err_ptr(-EINVAL);
12896 }
12897
12898 link = calloc(1, sizeof(*link));
12899 if (!link)
12900 return libbpf_err_ptr(-ENOMEM);
12901 link->detach = &bpf_link__detach_fd;
12902
12903 link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
12904 &link_create_opts);
12905 if (link_fd < 0) {
12906 link_fd = -errno;
12907 free(link);
12908 pr_warn("prog '%s': failed to attach to iterator: %s\n",
12909 prog->name, errstr(link_fd));
12910 return libbpf_err_ptr(link_fd);
12911 }
12912 link->fd = link_fd;
12913 return link;
12914 }
12915
attach_iter(const struct bpf_program * prog,long cookie,struct bpf_link ** link)12916 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12917 {
12918 *link = bpf_program__attach_iter(prog, NULL);
12919 return libbpf_get_error(*link);
12920 }
12921
bpf_program__attach_netfilter(const struct bpf_program * prog,const struct bpf_netfilter_opts * opts)12922 struct bpf_link *bpf_program__attach_netfilter(const struct bpf_program *prog,
12923 const struct bpf_netfilter_opts *opts)
12924 {
12925 LIBBPF_OPTS(bpf_link_create_opts, lopts);
12926 struct bpf_link *link;
12927 int prog_fd, link_fd;
12928
12929 if (!OPTS_VALID(opts, bpf_netfilter_opts))
12930 return libbpf_err_ptr(-EINVAL);
12931
12932 prog_fd = bpf_program__fd(prog);
12933 if (prog_fd < 0) {
12934 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12935 return libbpf_err_ptr(-EINVAL);
12936 }
12937
12938 link = calloc(1, sizeof(*link));
12939 if (!link)
12940 return libbpf_err_ptr(-ENOMEM);
12941
12942 link->detach = &bpf_link__detach_fd;
12943
12944 lopts.netfilter.pf = OPTS_GET(opts, pf, 0);
12945 lopts.netfilter.hooknum = OPTS_GET(opts, hooknum, 0);
12946 lopts.netfilter.priority = OPTS_GET(opts, priority, 0);
12947 lopts.netfilter.flags = OPTS_GET(opts, flags, 0);
12948
12949 link_fd = bpf_link_create(prog_fd, 0, BPF_NETFILTER, &lopts);
12950 if (link_fd < 0) {
12951 link_fd = -errno;
12952 free(link);
12953 pr_warn("prog '%s': failed to attach to netfilter: %s\n",
12954 prog->name, errstr(link_fd));
12955 return libbpf_err_ptr(link_fd);
12956 }
12957 link->fd = link_fd;
12958
12959 return link;
12960 }
12961
bpf_program__attach(const struct bpf_program * prog)12962 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
12963 {
12964 struct bpf_link *link = NULL;
12965 int err;
12966
12967 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
12968 return libbpf_err_ptr(-EOPNOTSUPP);
12969
12970 if (bpf_program__fd(prog) < 0) {
12971 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12972 prog->name);
12973 return libbpf_err_ptr(-EINVAL);
12974 }
12975
12976 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
12977 if (err)
12978 return libbpf_err_ptr(err);
12979
12980 /* When calling bpf_program__attach() explicitly, auto-attach support
12981 * is expected to work, so NULL returned link is considered an error.
12982 * This is different for skeleton's attach, see comment in
12983 * bpf_object__attach_skeleton().
12984 */
12985 if (!link)
12986 return libbpf_err_ptr(-EOPNOTSUPP);
12987
12988 return link;
12989 }
12990
12991 struct bpf_link_struct_ops {
12992 struct bpf_link link;
12993 int map_fd;
12994 };
12995
bpf_link__detach_struct_ops(struct bpf_link * link)12996 static int bpf_link__detach_struct_ops(struct bpf_link *link)
12997 {
12998 struct bpf_link_struct_ops *st_link;
12999 __u32 zero = 0;
13000
13001 st_link = container_of(link, struct bpf_link_struct_ops, link);
13002
13003 if (st_link->map_fd < 0)
13004 /* w/o a real link */
13005 return bpf_map_delete_elem(link->fd, &zero);
13006
13007 return close(link->fd);
13008 }
13009
bpf_map__attach_struct_ops(const struct bpf_map * map)13010 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
13011 {
13012 struct bpf_link_struct_ops *link;
13013 __u32 zero = 0;
13014 int err, fd;
13015
13016 if (!bpf_map__is_struct_ops(map)) {
13017 pr_warn("map '%s': can't attach non-struct_ops map\n", map->name);
13018 return libbpf_err_ptr(-EINVAL);
13019 }
13020
13021 if (map->fd < 0) {
13022 pr_warn("map '%s': can't attach BPF map without FD (was it created?)\n", map->name);
13023 return libbpf_err_ptr(-EINVAL);
13024 }
13025
13026 link = calloc(1, sizeof(*link));
13027 if (!link)
13028 return libbpf_err_ptr(-EINVAL);
13029
13030 /* kern_vdata should be prepared during the loading phase. */
13031 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
13032 /* It can be EBUSY if the map has been used to create or
13033 * update a link before. We don't allow updating the value of
13034 * a struct_ops once it is set. That ensures that the value
13035 * never changed. So, it is safe to skip EBUSY.
13036 */
13037 if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) {
13038 free(link);
13039 return libbpf_err_ptr(err);
13040 }
13041
13042 link->link.detach = bpf_link__detach_struct_ops;
13043
13044 if (!(map->def.map_flags & BPF_F_LINK)) {
13045 /* w/o a real link */
13046 link->link.fd = map->fd;
13047 link->map_fd = -1;
13048 return &link->link;
13049 }
13050
13051 fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL);
13052 if (fd < 0) {
13053 free(link);
13054 return libbpf_err_ptr(fd);
13055 }
13056
13057 link->link.fd = fd;
13058 link->map_fd = map->fd;
13059
13060 return &link->link;
13061 }
13062
13063 /*
13064 * Swap the back struct_ops of a link with a new struct_ops map.
13065 */
bpf_link__update_map(struct bpf_link * link,const struct bpf_map * map)13066 int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map)
13067 {
13068 struct bpf_link_struct_ops *st_ops_link;
13069 __u32 zero = 0;
13070 int err;
13071
13072 if (!bpf_map__is_struct_ops(map))
13073 return -EINVAL;
13074
13075 if (map->fd < 0) {
13076 pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name);
13077 return -EINVAL;
13078 }
13079
13080 st_ops_link = container_of(link, struct bpf_link_struct_ops, link);
13081 /* Ensure the type of a link is correct */
13082 if (st_ops_link->map_fd < 0)
13083 return -EINVAL;
13084
13085 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
13086 /* It can be EBUSY if the map has been used to create or
13087 * update a link before. We don't allow updating the value of
13088 * a struct_ops once it is set. That ensures that the value
13089 * never changed. So, it is safe to skip EBUSY.
13090 */
13091 if (err && err != -EBUSY)
13092 return err;
13093
13094 err = bpf_link_update(link->fd, map->fd, NULL);
13095 if (err < 0)
13096 return err;
13097
13098 st_ops_link->map_fd = map->fd;
13099
13100 return 0;
13101 }
13102
13103 typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
13104 void *private_data);
13105
13106 static enum bpf_perf_event_ret
perf_event_read_simple(void * mmap_mem,size_t mmap_size,size_t page_size,void ** copy_mem,size_t * copy_size,bpf_perf_event_print_t fn,void * private_data)13107 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
13108 void **copy_mem, size_t *copy_size,
13109 bpf_perf_event_print_t fn, void *private_data)
13110 {
13111 struct perf_event_mmap_page *header = mmap_mem;
13112 __u64 data_head = ring_buffer_read_head(header);
13113 __u64 data_tail = header->data_tail;
13114 void *base = ((__u8 *)header) + page_size;
13115 int ret = LIBBPF_PERF_EVENT_CONT;
13116 struct perf_event_header *ehdr;
13117 size_t ehdr_size;
13118
13119 while (data_head != data_tail) {
13120 ehdr = base + (data_tail & (mmap_size - 1));
13121 ehdr_size = ehdr->size;
13122
13123 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
13124 void *copy_start = ehdr;
13125 size_t len_first = base + mmap_size - copy_start;
13126 size_t len_secnd = ehdr_size - len_first;
13127
13128 if (*copy_size < ehdr_size) {
13129 free(*copy_mem);
13130 *copy_mem = malloc(ehdr_size);
13131 if (!*copy_mem) {
13132 *copy_size = 0;
13133 ret = LIBBPF_PERF_EVENT_ERROR;
13134 break;
13135 }
13136 *copy_size = ehdr_size;
13137 }
13138
13139 memcpy(*copy_mem, copy_start, len_first);
13140 memcpy(*copy_mem + len_first, base, len_secnd);
13141 ehdr = *copy_mem;
13142 }
13143
13144 ret = fn(ehdr, private_data);
13145 data_tail += ehdr_size;
13146 if (ret != LIBBPF_PERF_EVENT_CONT)
13147 break;
13148 }
13149
13150 ring_buffer_write_tail(header, data_tail);
13151 return libbpf_err(ret);
13152 }
13153
13154 struct perf_buffer;
13155
13156 struct perf_buffer_params {
13157 struct perf_event_attr *attr;
13158 /* if event_cb is specified, it takes precendence */
13159 perf_buffer_event_fn event_cb;
13160 /* sample_cb and lost_cb are higher-level common-case callbacks */
13161 perf_buffer_sample_fn sample_cb;
13162 perf_buffer_lost_fn lost_cb;
13163 void *ctx;
13164 int cpu_cnt;
13165 int *cpus;
13166 int *map_keys;
13167 };
13168
13169 struct perf_cpu_buf {
13170 struct perf_buffer *pb;
13171 void *base; /* mmap()'ed memory */
13172 void *buf; /* for reconstructing segmented data */
13173 size_t buf_size;
13174 int fd;
13175 int cpu;
13176 int map_key;
13177 };
13178
13179 struct perf_buffer {
13180 perf_buffer_event_fn event_cb;
13181 perf_buffer_sample_fn sample_cb;
13182 perf_buffer_lost_fn lost_cb;
13183 void *ctx; /* passed into callbacks */
13184
13185 size_t page_size;
13186 size_t mmap_size;
13187 struct perf_cpu_buf **cpu_bufs;
13188 struct epoll_event *events;
13189 int cpu_cnt; /* number of allocated CPU buffers */
13190 int epoll_fd; /* perf event FD */
13191 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
13192 };
13193
perf_buffer__free_cpu_buf(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)13194 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
13195 struct perf_cpu_buf *cpu_buf)
13196 {
13197 if (!cpu_buf)
13198 return;
13199 if (cpu_buf->base &&
13200 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
13201 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
13202 if (cpu_buf->fd >= 0) {
13203 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
13204 close(cpu_buf->fd);
13205 }
13206 free(cpu_buf->buf);
13207 free(cpu_buf);
13208 }
13209
perf_buffer__free(struct perf_buffer * pb)13210 void perf_buffer__free(struct perf_buffer *pb)
13211 {
13212 int i;
13213
13214 if (IS_ERR_OR_NULL(pb))
13215 return;
13216 if (pb->cpu_bufs) {
13217 for (i = 0; i < pb->cpu_cnt; i++) {
13218 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
13219
13220 if (!cpu_buf)
13221 continue;
13222
13223 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
13224 perf_buffer__free_cpu_buf(pb, cpu_buf);
13225 }
13226 free(pb->cpu_bufs);
13227 }
13228 if (pb->epoll_fd >= 0)
13229 close(pb->epoll_fd);
13230 free(pb->events);
13231 free(pb);
13232 }
13233
13234 static struct perf_cpu_buf *
perf_buffer__open_cpu_buf(struct perf_buffer * pb,struct perf_event_attr * attr,int cpu,int map_key)13235 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
13236 int cpu, int map_key)
13237 {
13238 struct perf_cpu_buf *cpu_buf;
13239 int err;
13240
13241 cpu_buf = calloc(1, sizeof(*cpu_buf));
13242 if (!cpu_buf)
13243 return ERR_PTR(-ENOMEM);
13244
13245 cpu_buf->pb = pb;
13246 cpu_buf->cpu = cpu;
13247 cpu_buf->map_key = map_key;
13248
13249 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
13250 -1, PERF_FLAG_FD_CLOEXEC);
13251 if (cpu_buf->fd < 0) {
13252 err = -errno;
13253 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
13254 cpu, errstr(err));
13255 goto error;
13256 }
13257
13258 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
13259 PROT_READ | PROT_WRITE, MAP_SHARED,
13260 cpu_buf->fd, 0);
13261 if (cpu_buf->base == MAP_FAILED) {
13262 cpu_buf->base = NULL;
13263 err = -errno;
13264 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
13265 cpu, errstr(err));
13266 goto error;
13267 }
13268
13269 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
13270 err = -errno;
13271 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
13272 cpu, errstr(err));
13273 goto error;
13274 }
13275
13276 return cpu_buf;
13277
13278 error:
13279 perf_buffer__free_cpu_buf(pb, cpu_buf);
13280 return (struct perf_cpu_buf *)ERR_PTR(err);
13281 }
13282
13283 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
13284 struct perf_buffer_params *p);
13285
perf_buffer__new(int map_fd,size_t page_cnt,perf_buffer_sample_fn sample_cb,perf_buffer_lost_fn lost_cb,void * ctx,const struct perf_buffer_opts * opts)13286 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
13287 perf_buffer_sample_fn sample_cb,
13288 perf_buffer_lost_fn lost_cb,
13289 void *ctx,
13290 const struct perf_buffer_opts *opts)
13291 {
13292 const size_t attr_sz = sizeof(struct perf_event_attr);
13293 struct perf_buffer_params p = {};
13294 struct perf_event_attr attr;
13295 __u32 sample_period;
13296
13297 if (!OPTS_VALID(opts, perf_buffer_opts))
13298 return libbpf_err_ptr(-EINVAL);
13299
13300 sample_period = OPTS_GET(opts, sample_period, 1);
13301 if (!sample_period)
13302 sample_period = 1;
13303
13304 memset(&attr, 0, attr_sz);
13305 attr.size = attr_sz;
13306 attr.config = PERF_COUNT_SW_BPF_OUTPUT;
13307 attr.type = PERF_TYPE_SOFTWARE;
13308 attr.sample_type = PERF_SAMPLE_RAW;
13309 attr.sample_period = sample_period;
13310 attr.wakeup_events = sample_period;
13311
13312 p.attr = &attr;
13313 p.sample_cb = sample_cb;
13314 p.lost_cb = lost_cb;
13315 p.ctx = ctx;
13316
13317 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
13318 }
13319
perf_buffer__new_raw(int map_fd,size_t page_cnt,struct perf_event_attr * attr,perf_buffer_event_fn event_cb,void * ctx,const struct perf_buffer_raw_opts * opts)13320 struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt,
13321 struct perf_event_attr *attr,
13322 perf_buffer_event_fn event_cb, void *ctx,
13323 const struct perf_buffer_raw_opts *opts)
13324 {
13325 struct perf_buffer_params p = {};
13326
13327 if (!attr)
13328 return libbpf_err_ptr(-EINVAL);
13329
13330 if (!OPTS_VALID(opts, perf_buffer_raw_opts))
13331 return libbpf_err_ptr(-EINVAL);
13332
13333 p.attr = attr;
13334 p.event_cb = event_cb;
13335 p.ctx = ctx;
13336 p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
13337 p.cpus = OPTS_GET(opts, cpus, NULL);
13338 p.map_keys = OPTS_GET(opts, map_keys, NULL);
13339
13340 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
13341 }
13342
__perf_buffer__new(int map_fd,size_t page_cnt,struct perf_buffer_params * p)13343 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
13344 struct perf_buffer_params *p)
13345 {
13346 const char *online_cpus_file = "/sys/devices/system/cpu/online";
13347 struct bpf_map_info map;
13348 struct perf_buffer *pb;
13349 bool *online = NULL;
13350 __u32 map_info_len;
13351 int err, i, j, n;
13352
13353 if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
13354 pr_warn("page count should be power of two, but is %zu\n",
13355 page_cnt);
13356 return ERR_PTR(-EINVAL);
13357 }
13358
13359 /* best-effort sanity checks */
13360 memset(&map, 0, sizeof(map));
13361 map_info_len = sizeof(map);
13362 err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len);
13363 if (err) {
13364 err = -errno;
13365 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
13366 * -EBADFD, -EFAULT, or -E2BIG on real error
13367 */
13368 if (err != -EINVAL) {
13369 pr_warn("failed to get map info for map FD %d: %s\n",
13370 map_fd, errstr(err));
13371 return ERR_PTR(err);
13372 }
13373 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
13374 map_fd);
13375 } else {
13376 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
13377 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
13378 map.name);
13379 return ERR_PTR(-EINVAL);
13380 }
13381 }
13382
13383 pb = calloc(1, sizeof(*pb));
13384 if (!pb)
13385 return ERR_PTR(-ENOMEM);
13386
13387 pb->event_cb = p->event_cb;
13388 pb->sample_cb = p->sample_cb;
13389 pb->lost_cb = p->lost_cb;
13390 pb->ctx = p->ctx;
13391
13392 pb->page_size = getpagesize();
13393 pb->mmap_size = pb->page_size * page_cnt;
13394 pb->map_fd = map_fd;
13395
13396 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
13397 if (pb->epoll_fd < 0) {
13398 err = -errno;
13399 pr_warn("failed to create epoll instance: %s\n",
13400 errstr(err));
13401 goto error;
13402 }
13403
13404 if (p->cpu_cnt > 0) {
13405 pb->cpu_cnt = p->cpu_cnt;
13406 } else {
13407 pb->cpu_cnt = libbpf_num_possible_cpus();
13408 if (pb->cpu_cnt < 0) {
13409 err = pb->cpu_cnt;
13410 goto error;
13411 }
13412 if (map.max_entries && map.max_entries < pb->cpu_cnt)
13413 pb->cpu_cnt = map.max_entries;
13414 }
13415
13416 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
13417 if (!pb->events) {
13418 err = -ENOMEM;
13419 pr_warn("failed to allocate events: out of memory\n");
13420 goto error;
13421 }
13422 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
13423 if (!pb->cpu_bufs) {
13424 err = -ENOMEM;
13425 pr_warn("failed to allocate buffers: out of memory\n");
13426 goto error;
13427 }
13428
13429 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
13430 if (err) {
13431 pr_warn("failed to get online CPU mask: %s\n", errstr(err));
13432 goto error;
13433 }
13434
13435 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
13436 struct perf_cpu_buf *cpu_buf;
13437 int cpu, map_key;
13438
13439 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
13440 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
13441
13442 /* in case user didn't explicitly requested particular CPUs to
13443 * be attached to, skip offline/not present CPUs
13444 */
13445 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
13446 continue;
13447
13448 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
13449 if (IS_ERR(cpu_buf)) {
13450 err = PTR_ERR(cpu_buf);
13451 goto error;
13452 }
13453
13454 pb->cpu_bufs[j] = cpu_buf;
13455
13456 err = bpf_map_update_elem(pb->map_fd, &map_key,
13457 &cpu_buf->fd, 0);
13458 if (err) {
13459 err = -errno;
13460 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
13461 cpu, map_key, cpu_buf->fd,
13462 errstr(err));
13463 goto error;
13464 }
13465
13466 pb->events[j].events = EPOLLIN;
13467 pb->events[j].data.ptr = cpu_buf;
13468 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
13469 &pb->events[j]) < 0) {
13470 err = -errno;
13471 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
13472 cpu, cpu_buf->fd,
13473 errstr(err));
13474 goto error;
13475 }
13476 j++;
13477 }
13478 pb->cpu_cnt = j;
13479 free(online);
13480
13481 return pb;
13482
13483 error:
13484 free(online);
13485 if (pb)
13486 perf_buffer__free(pb);
13487 return ERR_PTR(err);
13488 }
13489
13490 struct perf_sample_raw {
13491 struct perf_event_header header;
13492 uint32_t size;
13493 char data[];
13494 };
13495
13496 struct perf_sample_lost {
13497 struct perf_event_header header;
13498 uint64_t id;
13499 uint64_t lost;
13500 uint64_t sample_id;
13501 };
13502
13503 static enum bpf_perf_event_ret
perf_buffer__process_record(struct perf_event_header * e,void * ctx)13504 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
13505 {
13506 struct perf_cpu_buf *cpu_buf = ctx;
13507 struct perf_buffer *pb = cpu_buf->pb;
13508 void *data = e;
13509
13510 /* user wants full control over parsing perf event */
13511 if (pb->event_cb)
13512 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
13513
13514 switch (e->type) {
13515 case PERF_RECORD_SAMPLE: {
13516 struct perf_sample_raw *s = data;
13517
13518 if (pb->sample_cb)
13519 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
13520 break;
13521 }
13522 case PERF_RECORD_LOST: {
13523 struct perf_sample_lost *s = data;
13524
13525 if (pb->lost_cb)
13526 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
13527 break;
13528 }
13529 default:
13530 pr_warn("unknown perf sample type %d\n", e->type);
13531 return LIBBPF_PERF_EVENT_ERROR;
13532 }
13533 return LIBBPF_PERF_EVENT_CONT;
13534 }
13535
perf_buffer__process_records(struct perf_buffer * pb,struct perf_cpu_buf * cpu_buf)13536 static int perf_buffer__process_records(struct perf_buffer *pb,
13537 struct perf_cpu_buf *cpu_buf)
13538 {
13539 enum bpf_perf_event_ret ret;
13540
13541 ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
13542 pb->page_size, &cpu_buf->buf,
13543 &cpu_buf->buf_size,
13544 perf_buffer__process_record, cpu_buf);
13545 if (ret != LIBBPF_PERF_EVENT_CONT)
13546 return ret;
13547 return 0;
13548 }
13549
perf_buffer__epoll_fd(const struct perf_buffer * pb)13550 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
13551 {
13552 return pb->epoll_fd;
13553 }
13554
perf_buffer__poll(struct perf_buffer * pb,int timeout_ms)13555 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
13556 {
13557 int i, cnt, err;
13558
13559 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
13560 if (cnt < 0)
13561 return -errno;
13562
13563 for (i = 0; i < cnt; i++) {
13564 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
13565
13566 err = perf_buffer__process_records(pb, cpu_buf);
13567 if (err) {
13568 pr_warn("error while processing records: %s\n", errstr(err));
13569 return libbpf_err(err);
13570 }
13571 }
13572 return cnt;
13573 }
13574
13575 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
13576 * manager.
13577 */
perf_buffer__buffer_cnt(const struct perf_buffer * pb)13578 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
13579 {
13580 return pb->cpu_cnt;
13581 }
13582
13583 /*
13584 * Return perf_event FD of a ring buffer in *buf_idx* slot of
13585 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
13586 * select()/poll()/epoll() Linux syscalls.
13587 */
perf_buffer__buffer_fd(const struct perf_buffer * pb,size_t buf_idx)13588 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
13589 {
13590 struct perf_cpu_buf *cpu_buf;
13591
13592 if (buf_idx >= pb->cpu_cnt)
13593 return libbpf_err(-EINVAL);
13594
13595 cpu_buf = pb->cpu_bufs[buf_idx];
13596 if (!cpu_buf)
13597 return libbpf_err(-ENOENT);
13598
13599 return cpu_buf->fd;
13600 }
13601
perf_buffer__buffer(struct perf_buffer * pb,int buf_idx,void ** buf,size_t * buf_size)13602 int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size)
13603 {
13604 struct perf_cpu_buf *cpu_buf;
13605
13606 if (buf_idx >= pb->cpu_cnt)
13607 return libbpf_err(-EINVAL);
13608
13609 cpu_buf = pb->cpu_bufs[buf_idx];
13610 if (!cpu_buf)
13611 return libbpf_err(-ENOENT);
13612
13613 *buf = cpu_buf->base;
13614 *buf_size = pb->mmap_size;
13615 return 0;
13616 }
13617
13618 /*
13619 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
13620 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
13621 * consume, do nothing and return success.
13622 * Returns:
13623 * - 0 on success;
13624 * - <0 on failure.
13625 */
perf_buffer__consume_buffer(struct perf_buffer * pb,size_t buf_idx)13626 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
13627 {
13628 struct perf_cpu_buf *cpu_buf;
13629
13630 if (buf_idx >= pb->cpu_cnt)
13631 return libbpf_err(-EINVAL);
13632
13633 cpu_buf = pb->cpu_bufs[buf_idx];
13634 if (!cpu_buf)
13635 return libbpf_err(-ENOENT);
13636
13637 return perf_buffer__process_records(pb, cpu_buf);
13638 }
13639
perf_buffer__consume(struct perf_buffer * pb)13640 int perf_buffer__consume(struct perf_buffer *pb)
13641 {
13642 int i, err;
13643
13644 for (i = 0; i < pb->cpu_cnt; i++) {
13645 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
13646
13647 if (!cpu_buf)
13648 continue;
13649
13650 err = perf_buffer__process_records(pb, cpu_buf);
13651 if (err) {
13652 pr_warn("perf_buffer: failed to process records in buffer #%d: %s\n",
13653 i, errstr(err));
13654 return libbpf_err(err);
13655 }
13656 }
13657 return 0;
13658 }
13659
bpf_program__set_attach_target(struct bpf_program * prog,int attach_prog_fd,const char * attach_func_name)13660 int bpf_program__set_attach_target(struct bpf_program *prog,
13661 int attach_prog_fd,
13662 const char *attach_func_name)
13663 {
13664 int btf_obj_fd = 0, btf_id = 0, err;
13665
13666 if (!prog || attach_prog_fd < 0)
13667 return libbpf_err(-EINVAL);
13668
13669 if (prog->obj->loaded)
13670 return libbpf_err(-EINVAL);
13671
13672 if (attach_prog_fd && !attach_func_name) {
13673 /* remember attach_prog_fd and let bpf_program__load() find
13674 * BTF ID during the program load
13675 */
13676 prog->attach_prog_fd = attach_prog_fd;
13677 return 0;
13678 }
13679
13680 if (attach_prog_fd) {
13681 btf_id = libbpf_find_prog_btf_id(attach_func_name,
13682 attach_prog_fd);
13683 if (btf_id < 0)
13684 return libbpf_err(btf_id);
13685 } else {
13686 if (!attach_func_name)
13687 return libbpf_err(-EINVAL);
13688
13689 /* load btf_vmlinux, if not yet */
13690 err = bpf_object__load_vmlinux_btf(prog->obj, true);
13691 if (err)
13692 return libbpf_err(err);
13693 err = find_kernel_btf_id(prog->obj, attach_func_name,
13694 prog->expected_attach_type,
13695 &btf_obj_fd, &btf_id);
13696 if (err)
13697 return libbpf_err(err);
13698 }
13699
13700 prog->attach_btf_id = btf_id;
13701 prog->attach_btf_obj_fd = btf_obj_fd;
13702 prog->attach_prog_fd = attach_prog_fd;
13703 return 0;
13704 }
13705
parse_cpu_mask_str(const char * s,bool ** mask,int * mask_sz)13706 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
13707 {
13708 int err = 0, n, len, start, end = -1;
13709 bool *tmp;
13710
13711 *mask = NULL;
13712 *mask_sz = 0;
13713
13714 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
13715 while (*s) {
13716 if (*s == ',' || *s == '\n') {
13717 s++;
13718 continue;
13719 }
13720 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
13721 if (n <= 0 || n > 2) {
13722 pr_warn("Failed to get CPU range %s: %d\n", s, n);
13723 err = -EINVAL;
13724 goto cleanup;
13725 } else if (n == 1) {
13726 end = start;
13727 }
13728 if (start < 0 || start > end) {
13729 pr_warn("Invalid CPU range [%d,%d] in %s\n",
13730 start, end, s);
13731 err = -EINVAL;
13732 goto cleanup;
13733 }
13734 tmp = realloc(*mask, end + 1);
13735 if (!tmp) {
13736 err = -ENOMEM;
13737 goto cleanup;
13738 }
13739 *mask = tmp;
13740 memset(tmp + *mask_sz, 0, start - *mask_sz);
13741 memset(tmp + start, 1, end - start + 1);
13742 *mask_sz = end + 1;
13743 s += len;
13744 }
13745 if (!*mask_sz) {
13746 pr_warn("Empty CPU range\n");
13747 return -EINVAL;
13748 }
13749 return 0;
13750 cleanup:
13751 free(*mask);
13752 *mask = NULL;
13753 return err;
13754 }
13755
parse_cpu_mask_file(const char * fcpu,bool ** mask,int * mask_sz)13756 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
13757 {
13758 int fd, err = 0, len;
13759 char buf[128];
13760
13761 fd = open(fcpu, O_RDONLY | O_CLOEXEC);
13762 if (fd < 0) {
13763 err = -errno;
13764 pr_warn("Failed to open cpu mask file %s: %s\n", fcpu, errstr(err));
13765 return err;
13766 }
13767 len = read(fd, buf, sizeof(buf));
13768 close(fd);
13769 if (len <= 0) {
13770 err = len ? -errno : -EINVAL;
13771 pr_warn("Failed to read cpu mask from %s: %s\n", fcpu, errstr(err));
13772 return err;
13773 }
13774 if (len >= sizeof(buf)) {
13775 pr_warn("CPU mask is too big in file %s\n", fcpu);
13776 return -E2BIG;
13777 }
13778 buf[len] = '\0';
13779
13780 return parse_cpu_mask_str(buf, mask, mask_sz);
13781 }
13782
libbpf_num_possible_cpus(void)13783 int libbpf_num_possible_cpus(void)
13784 {
13785 static const char *fcpu = "/sys/devices/system/cpu/possible";
13786 static int cpus;
13787 int err, n, i, tmp_cpus;
13788 bool *mask;
13789
13790 tmp_cpus = READ_ONCE(cpus);
13791 if (tmp_cpus > 0)
13792 return tmp_cpus;
13793
13794 err = parse_cpu_mask_file(fcpu, &mask, &n);
13795 if (err)
13796 return libbpf_err(err);
13797
13798 tmp_cpus = 0;
13799 for (i = 0; i < n; i++) {
13800 if (mask[i])
13801 tmp_cpus++;
13802 }
13803 free(mask);
13804
13805 WRITE_ONCE(cpus, tmp_cpus);
13806 return tmp_cpus;
13807 }
13808
populate_skeleton_maps(const struct bpf_object * obj,struct bpf_map_skeleton * maps,size_t map_cnt,size_t map_skel_sz)13809 static int populate_skeleton_maps(const struct bpf_object *obj,
13810 struct bpf_map_skeleton *maps,
13811 size_t map_cnt, size_t map_skel_sz)
13812 {
13813 int i;
13814
13815 for (i = 0; i < map_cnt; i++) {
13816 struct bpf_map_skeleton *map_skel = (void *)maps + i * map_skel_sz;
13817 struct bpf_map **map = map_skel->map;
13818 const char *name = map_skel->name;
13819 void **mmaped = map_skel->mmaped;
13820
13821 *map = bpf_object__find_map_by_name(obj, name);
13822 if (!*map) {
13823 pr_warn("failed to find skeleton map '%s'\n", name);
13824 return -ESRCH;
13825 }
13826
13827 /* externs shouldn't be pre-setup from user code */
13828 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
13829 *mmaped = (*map)->mmaped;
13830 }
13831 return 0;
13832 }
13833
populate_skeleton_progs(const struct bpf_object * obj,struct bpf_prog_skeleton * progs,size_t prog_cnt,size_t prog_skel_sz)13834 static int populate_skeleton_progs(const struct bpf_object *obj,
13835 struct bpf_prog_skeleton *progs,
13836 size_t prog_cnt, size_t prog_skel_sz)
13837 {
13838 int i;
13839
13840 for (i = 0; i < prog_cnt; i++) {
13841 struct bpf_prog_skeleton *prog_skel = (void *)progs + i * prog_skel_sz;
13842 struct bpf_program **prog = prog_skel->prog;
13843 const char *name = prog_skel->name;
13844
13845 *prog = bpf_object__find_program_by_name(obj, name);
13846 if (!*prog) {
13847 pr_warn("failed to find skeleton program '%s'\n", name);
13848 return -ESRCH;
13849 }
13850 }
13851 return 0;
13852 }
13853
bpf_object__open_skeleton(struct bpf_object_skeleton * s,const struct bpf_object_open_opts * opts)13854 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
13855 const struct bpf_object_open_opts *opts)
13856 {
13857 struct bpf_object *obj;
13858 int err;
13859
13860 obj = bpf_object_open(NULL, s->data, s->data_sz, s->name, opts);
13861 if (IS_ERR(obj)) {
13862 err = PTR_ERR(obj);
13863 pr_warn("failed to initialize skeleton BPF object '%s': %s\n",
13864 s->name, errstr(err));
13865 return libbpf_err(err);
13866 }
13867
13868 *s->obj = obj;
13869 err = populate_skeleton_maps(obj, s->maps, s->map_cnt, s->map_skel_sz);
13870 if (err) {
13871 pr_warn("failed to populate skeleton maps for '%s': %s\n", s->name, errstr(err));
13872 return libbpf_err(err);
13873 }
13874
13875 err = populate_skeleton_progs(obj, s->progs, s->prog_cnt, s->prog_skel_sz);
13876 if (err) {
13877 pr_warn("failed to populate skeleton progs for '%s': %s\n", s->name, errstr(err));
13878 return libbpf_err(err);
13879 }
13880
13881 return 0;
13882 }
13883
bpf_object__open_subskeleton(struct bpf_object_subskeleton * s)13884 int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
13885 {
13886 int err, len, var_idx, i;
13887 const char *var_name;
13888 const struct bpf_map *map;
13889 struct btf *btf;
13890 __u32 map_type_id;
13891 const struct btf_type *map_type, *var_type;
13892 const struct bpf_var_skeleton *var_skel;
13893 struct btf_var_secinfo *var;
13894
13895 if (!s->obj)
13896 return libbpf_err(-EINVAL);
13897
13898 btf = bpf_object__btf(s->obj);
13899 if (!btf) {
13900 pr_warn("subskeletons require BTF at runtime (object %s)\n",
13901 bpf_object__name(s->obj));
13902 return libbpf_err(-errno);
13903 }
13904
13905 err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt, s->map_skel_sz);
13906 if (err) {
13907 pr_warn("failed to populate subskeleton maps: %s\n", errstr(err));
13908 return libbpf_err(err);
13909 }
13910
13911 err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt, s->prog_skel_sz);
13912 if (err) {
13913 pr_warn("failed to populate subskeleton maps: %s\n", errstr(err));
13914 return libbpf_err(err);
13915 }
13916
13917 for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
13918 var_skel = (void *)s->vars + var_idx * s->var_skel_sz;
13919 map = *var_skel->map;
13920 map_type_id = bpf_map__btf_value_type_id(map);
13921 map_type = btf__type_by_id(btf, map_type_id);
13922
13923 if (!btf_is_datasec(map_type)) {
13924 pr_warn("type for map '%1$s' is not a datasec: %2$s\n",
13925 bpf_map__name(map),
13926 __btf_kind_str(btf_kind(map_type)));
13927 return libbpf_err(-EINVAL);
13928 }
13929
13930 len = btf_vlen(map_type);
13931 var = btf_var_secinfos(map_type);
13932 for (i = 0; i < len; i++, var++) {
13933 var_type = btf__type_by_id(btf, var->type);
13934 var_name = btf__name_by_offset(btf, var_type->name_off);
13935 if (strcmp(var_name, var_skel->name) == 0) {
13936 *var_skel->addr = map->mmaped + var->offset;
13937 break;
13938 }
13939 }
13940 }
13941 return 0;
13942 }
13943
bpf_object__destroy_subskeleton(struct bpf_object_subskeleton * s)13944 void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
13945 {
13946 if (!s)
13947 return;
13948 free(s->maps);
13949 free(s->progs);
13950 free(s->vars);
13951 free(s);
13952 }
13953
bpf_object__load_skeleton(struct bpf_object_skeleton * s)13954 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
13955 {
13956 int i, err;
13957
13958 err = bpf_object__load(*s->obj);
13959 if (err) {
13960 pr_warn("failed to load BPF skeleton '%s': %s\n", s->name, errstr(err));
13961 return libbpf_err(err);
13962 }
13963
13964 for (i = 0; i < s->map_cnt; i++) {
13965 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
13966 struct bpf_map *map = *map_skel->map;
13967
13968 if (!map_skel->mmaped)
13969 continue;
13970
13971 *map_skel->mmaped = map->mmaped;
13972 }
13973
13974 return 0;
13975 }
13976
bpf_object__attach_skeleton(struct bpf_object_skeleton * s)13977 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
13978 {
13979 int i, err;
13980
13981 for (i = 0; i < s->prog_cnt; i++) {
13982 struct bpf_prog_skeleton *prog_skel = (void *)s->progs + i * s->prog_skel_sz;
13983 struct bpf_program *prog = *prog_skel->prog;
13984 struct bpf_link **link = prog_skel->link;
13985
13986 if (!prog->autoload || !prog->autoattach)
13987 continue;
13988
13989 /* auto-attaching not supported for this program */
13990 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
13991 continue;
13992
13993 /* if user already set the link manually, don't attempt auto-attach */
13994 if (*link)
13995 continue;
13996
13997 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
13998 if (err) {
13999 pr_warn("prog '%s': failed to auto-attach: %s\n",
14000 bpf_program__name(prog), errstr(err));
14001 return libbpf_err(err);
14002 }
14003
14004 /* It's possible that for some SEC() definitions auto-attach
14005 * is supported in some cases (e.g., if definition completely
14006 * specifies target information), but is not in other cases.
14007 * SEC("uprobe") is one such case. If user specified target
14008 * binary and function name, such BPF program can be
14009 * auto-attached. But if not, it shouldn't trigger skeleton's
14010 * attach to fail. It should just be skipped.
14011 * attach_fn signals such case with returning 0 (no error) and
14012 * setting link to NULL.
14013 */
14014 }
14015
14016
14017 for (i = 0; i < s->map_cnt; i++) {
14018 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
14019 struct bpf_map *map = *map_skel->map;
14020 struct bpf_link **link;
14021
14022 if (!map->autocreate || !map->autoattach)
14023 continue;
14024
14025 /* only struct_ops maps can be attached */
14026 if (!bpf_map__is_struct_ops(map))
14027 continue;
14028
14029 /* skeleton is created with earlier version of bpftool, notify user */
14030 if (s->map_skel_sz < offsetofend(struct bpf_map_skeleton, link)) {
14031 pr_warn("map '%s': BPF skeleton version is old, skipping map auto-attachment...\n",
14032 bpf_map__name(map));
14033 continue;
14034 }
14035
14036 link = map_skel->link;
14037 if (*link)
14038 continue;
14039
14040 *link = bpf_map__attach_struct_ops(map);
14041 if (!*link) {
14042 err = -errno;
14043 pr_warn("map '%s': failed to auto-attach: %s\n",
14044 bpf_map__name(map), errstr(err));
14045 return libbpf_err(err);
14046 }
14047 }
14048
14049 return 0;
14050 }
14051
bpf_object__detach_skeleton(struct bpf_object_skeleton * s)14052 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
14053 {
14054 int i;
14055
14056 for (i = 0; i < s->prog_cnt; i++) {
14057 struct bpf_prog_skeleton *prog_skel = (void *)s->progs + i * s->prog_skel_sz;
14058 struct bpf_link **link = prog_skel->link;
14059
14060 bpf_link__destroy(*link);
14061 *link = NULL;
14062 }
14063
14064 if (s->map_skel_sz < sizeof(struct bpf_map_skeleton))
14065 return;
14066
14067 for (i = 0; i < s->map_cnt; i++) {
14068 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
14069 struct bpf_link **link = map_skel->link;
14070
14071 if (link) {
14072 bpf_link__destroy(*link);
14073 *link = NULL;
14074 }
14075 }
14076 }
14077
bpf_object__destroy_skeleton(struct bpf_object_skeleton * s)14078 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
14079 {
14080 if (!s)
14081 return;
14082
14083 bpf_object__detach_skeleton(s);
14084 if (s->obj)
14085 bpf_object__close(*s->obj);
14086 free(s->maps);
14087 free(s->progs);
14088 free(s);
14089 }
14090