xref: /linux/tools/lib/bpf/gen_loader.c (revision 7a4ffec9fd54ea27395e24dff726dbf58e2fe06b)
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2021 Facebook */
3 #include <stdio.h>
4 #include <stdlib.h>
5 #include <string.h>
6 #include <errno.h>
7 #include <linux/filter.h>
8 #include <sys/param.h>
9 #include "btf.h"
10 #include "bpf.h"
11 #include "libbpf.h"
12 #include "libbpf_internal.h"
13 #include "hashmap.h"
14 #include "bpf_gen_internal.h"
15 #include "skel_internal.h"
16 #include <asm/byteorder.h>
17 
18 #define MAX_USED_MAPS	64
19 #define MAX_USED_PROGS	32
20 #define MAX_KFUNC_DESCS 256
21 #define MAX_FD_ARRAY_SZ (MAX_USED_MAPS + MAX_KFUNC_DESCS)
22 
23 /* The following structure describes the stack layout of the loader program.
24  * In addition R6 contains the pointer to context.
25  * R7 contains the result of the last sys_bpf command (typically error or FD).
26  * R9 contains the result of the last sys_close command.
27  *
28  * Naming convention:
29  * ctx - bpf program context
30  * stack - bpf program stack
31  * blob - bpf_attr-s, strings, insns, map data.
32  *        All the bytes that loader prog will use for read/write.
33  */
34 struct loader_stack {
35 	__u32 btf_fd;
36 	__u32 inner_map_fd;
37 	__u32 prog_fd[MAX_USED_PROGS];
38 };
39 
40 #define stack_off(field) \
41 	(__s16)(-sizeof(struct loader_stack) + offsetof(struct loader_stack, field))
42 
43 #define attr_field(attr, field) (attr + offsetof(union bpf_attr, field))
44 
45 static int blob_fd_array_off(struct bpf_gen *gen, int index)
46 {
47 	return gen->fd_array + index * sizeof(int);
48 }
49 
50 static int realloc_insn_buf(struct bpf_gen *gen, __u32 size)
51 {
52 	size_t off = gen->insn_cur - gen->insn_start;
53 	void *insn_start;
54 
55 	if (gen->error)
56 		return gen->error;
57 	if (size > INT32_MAX || off + size > INT32_MAX) {
58 		gen->error = -ERANGE;
59 		return -ERANGE;
60 	}
61 	insn_start = realloc(gen->insn_start, off + size);
62 	if (!insn_start) {
63 		gen->error = -ENOMEM;
64 		free(gen->insn_start);
65 		gen->insn_start = NULL;
66 		return -ENOMEM;
67 	}
68 	gen->insn_start = insn_start;
69 	gen->insn_cur = insn_start + off;
70 	return 0;
71 }
72 
73 static int realloc_data_buf(struct bpf_gen *gen, __u32 size)
74 {
75 	size_t off = gen->data_cur - gen->data_start;
76 	void *data_start;
77 
78 	if (gen->error)
79 		return gen->error;
80 	if (size > INT32_MAX || off + size > INT32_MAX) {
81 		gen->error = -ERANGE;
82 		return -ERANGE;
83 	}
84 	data_start = realloc(gen->data_start, off + size);
85 	if (!data_start) {
86 		gen->error = -ENOMEM;
87 		free(gen->data_start);
88 		gen->data_start = NULL;
89 		return -ENOMEM;
90 	}
91 	gen->data_start = data_start;
92 	gen->data_cur = data_start + off;
93 	return 0;
94 }
95 
96 static void emit(struct bpf_gen *gen, struct bpf_insn insn)
97 {
98 	if (realloc_insn_buf(gen, sizeof(insn)))
99 		return;
100 	memcpy(gen->insn_cur, &insn, sizeof(insn));
101 	gen->insn_cur += sizeof(insn);
102 }
103 
104 static void emit2(struct bpf_gen *gen, struct bpf_insn insn1, struct bpf_insn insn2)
105 {
106 	emit(gen, insn1);
107 	emit(gen, insn2);
108 }
109 
110 static int add_data(struct bpf_gen *gen, const void *data, __u32 size);
111 static void emit_sys_close_blob(struct bpf_gen *gen, int blob_off);
112 
113 void bpf_gen__init(struct bpf_gen *gen, int log_level, int nr_progs, int nr_maps)
114 {
115 	size_t stack_sz = sizeof(struct loader_stack), nr_progs_sz;
116 	int i;
117 
118 	gen->fd_array = add_data(gen, NULL, MAX_FD_ARRAY_SZ * sizeof(int));
119 	gen->log_level = log_level;
120 	/* save ctx pointer into R6 */
121 	emit(gen, BPF_MOV64_REG(BPF_REG_6, BPF_REG_1));
122 
123 	/* bzero stack */
124 	emit(gen, BPF_MOV64_REG(BPF_REG_1, BPF_REG_10));
125 	emit(gen, BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -stack_sz));
126 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, stack_sz));
127 	emit(gen, BPF_MOV64_IMM(BPF_REG_3, 0));
128 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_probe_read_kernel));
129 
130 	/* amount of stack actually used, only used to calculate iterations, not stack offset */
131 	nr_progs_sz = offsetof(struct loader_stack, prog_fd[nr_progs]);
132 	/* jump over cleanup code */
133 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0,
134 			      /* size of cleanup code below (including map fd cleanup) */
135 			      (nr_progs_sz / 4) * 3 + 2 +
136 			      /* 6 insns for emit_sys_close_blob,
137 			       * 6 insns for debug_regs in emit_sys_close_blob
138 			       */
139 			      nr_maps * (6 + (gen->log_level ? 6 : 0))));
140 
141 	/* remember the label where all error branches will jump to */
142 	gen->cleanup_label = gen->insn_cur - gen->insn_start;
143 	/* emit cleanup code: close all temp FDs */
144 	for (i = 0; i < nr_progs_sz; i += 4) {
145 		emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_10, -stack_sz + i));
146 		emit(gen, BPF_JMP_IMM(BPF_JSLE, BPF_REG_1, 0, 1));
147 		emit(gen, BPF_EMIT_CALL(BPF_FUNC_sys_close));
148 	}
149 	for (i = 0; i < nr_maps; i++)
150 		emit_sys_close_blob(gen, blob_fd_array_off(gen, i));
151 	/* R7 contains the error code from sys_bpf. Copy it into R0 and exit. */
152 	emit(gen, BPF_MOV64_REG(BPF_REG_0, BPF_REG_7));
153 	emit(gen, BPF_EXIT_INSN());
154 }
155 
156 static int add_data(struct bpf_gen *gen, const void *data, __u32 size)
157 {
158 	__u32 size8 = roundup(size, 8);
159 	__u64 zero = 0;
160 	void *prev;
161 
162 	if (realloc_data_buf(gen, size8))
163 		return 0;
164 	prev = gen->data_cur;
165 	if (data) {
166 		memcpy(gen->data_cur, data, size);
167 		memcpy(gen->data_cur + size, &zero, size8 - size);
168 	} else {
169 		memset(gen->data_cur, 0, size8);
170 	}
171 	gen->data_cur += size8;
172 	return prev - gen->data_start;
173 }
174 
175 /* Get index for map_fd/btf_fd slot in reserved fd_array, or in data relative
176  * to start of fd_array. Caller can decide if it is usable or not.
177  */
178 static int add_map_fd(struct bpf_gen *gen)
179 {
180 	if (gen->nr_maps == MAX_USED_MAPS) {
181 		pr_warn("Total maps exceeds %d\n", MAX_USED_MAPS);
182 		gen->error = -E2BIG;
183 		return 0;
184 	}
185 	return gen->nr_maps++;
186 }
187 
188 static int add_kfunc_btf_fd(struct bpf_gen *gen)
189 {
190 	int cur;
191 
192 	if (gen->nr_fd_array == MAX_KFUNC_DESCS) {
193 		cur = add_data(gen, NULL, sizeof(int));
194 		return (cur - gen->fd_array) / sizeof(int);
195 	}
196 	return MAX_USED_MAPS + gen->nr_fd_array++;
197 }
198 
199 static int insn_bytes_to_bpf_size(__u32 sz)
200 {
201 	switch (sz) {
202 	case 8: return BPF_DW;
203 	case 4: return BPF_W;
204 	case 2: return BPF_H;
205 	case 1: return BPF_B;
206 	default: return -1;
207 	}
208 }
209 
210 /* *(u64 *)(blob + off) = (u64)(void *)(blob + data) */
211 static void emit_rel_store(struct bpf_gen *gen, int off, int data)
212 {
213 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_0, BPF_PSEUDO_MAP_IDX_VALUE,
214 					 0, 0, 0, data));
215 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
216 					 0, 0, 0, off));
217 	emit(gen, BPF_STX_MEM(BPF_DW, BPF_REG_1, BPF_REG_0, 0));
218 }
219 
220 static void move_blob2blob(struct bpf_gen *gen, int off, int size, int blob_off)
221 {
222 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_2, BPF_PSEUDO_MAP_IDX_VALUE,
223 					 0, 0, 0, blob_off));
224 	emit(gen, BPF_LDX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_0, BPF_REG_2, 0));
225 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
226 					 0, 0, 0, off));
227 	emit(gen, BPF_STX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_1, BPF_REG_0, 0));
228 }
229 
230 static void move_blob2ctx(struct bpf_gen *gen, int ctx_off, int size, int blob_off)
231 {
232 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
233 					 0, 0, 0, blob_off));
234 	emit(gen, BPF_LDX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_0, BPF_REG_1, 0));
235 	emit(gen, BPF_STX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_6, BPF_REG_0, ctx_off));
236 }
237 
238 static void move_ctx2blob(struct bpf_gen *gen, int off, int size, int ctx_off,
239 				   bool check_non_zero)
240 {
241 	emit(gen, BPF_LDX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_0, BPF_REG_6, ctx_off));
242 	if (check_non_zero)
243 		/* If value in ctx is zero don't update the blob.
244 		 * For example: when ctx->map.max_entries == 0, keep default max_entries from bpf.c
245 		 */
246 		emit(gen, BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 3));
247 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
248 					 0, 0, 0, off));
249 	emit(gen, BPF_STX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_1, BPF_REG_0, 0));
250 }
251 
252 static void move_stack2blob(struct bpf_gen *gen, int off, int size, int stack_off)
253 {
254 	emit(gen, BPF_LDX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_0, BPF_REG_10, stack_off));
255 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
256 					 0, 0, 0, off));
257 	emit(gen, BPF_STX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_1, BPF_REG_0, 0));
258 }
259 
260 static void move_stack2ctx(struct bpf_gen *gen, int ctx_off, int size, int stack_off)
261 {
262 	emit(gen, BPF_LDX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_0, BPF_REG_10, stack_off));
263 	emit(gen, BPF_STX_MEM(insn_bytes_to_bpf_size(size), BPF_REG_6, BPF_REG_0, ctx_off));
264 }
265 
266 static void emit_sys_bpf(struct bpf_gen *gen, int cmd, int attr, int attr_size)
267 {
268 	emit(gen, BPF_MOV64_IMM(BPF_REG_1, cmd));
269 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_2, BPF_PSEUDO_MAP_IDX_VALUE,
270 					 0, 0, 0, attr));
271 	emit(gen, BPF_MOV64_IMM(BPF_REG_3, attr_size));
272 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_sys_bpf));
273 	/* remember the result in R7 */
274 	emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_0));
275 }
276 
277 static bool is_simm16(__s64 value)
278 {
279 	return value == (__s64)(__s16)value;
280 }
281 
282 static void emit_check_err(struct bpf_gen *gen)
283 {
284 	__s64 off = -(gen->insn_cur - gen->insn_start - gen->cleanup_label) / 8 - 1;
285 
286 	/* R7 contains result of last sys_bpf command.
287 	 * if (R7 < 0) goto cleanup;
288 	 */
289 	if (is_simm16(off)) {
290 		emit(gen, BPF_JMP_IMM(BPF_JSLT, BPF_REG_7, 0, off));
291 	} else {
292 		gen->error = -ERANGE;
293 		emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, -1));
294 	}
295 }
296 
297 /* reg1 and reg2 should not be R1 - R5. They can be R0, R6 - R10 */
298 static void emit_debug(struct bpf_gen *gen, int reg1, int reg2,
299 		       const char *fmt, va_list args)
300 {
301 	char buf[1024];
302 	int addr, len, ret;
303 
304 	if (!gen->log_level)
305 		return;
306 	ret = vsnprintf(buf, sizeof(buf), fmt, args);
307 	if (ret < 1024 - 7 && reg1 >= 0 && reg2 < 0)
308 		/* The special case to accommodate common debug_ret():
309 		 * to avoid specifying BPF_REG_7 and adding " r=%%d" to
310 		 * prints explicitly.
311 		 */
312 		strcat(buf, " r=%d");
313 	len = strlen(buf) + 1;
314 	addr = add_data(gen, buf, len);
315 
316 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
317 					 0, 0, 0, addr));
318 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, len));
319 	if (reg1 >= 0)
320 		emit(gen, BPF_MOV64_REG(BPF_REG_3, reg1));
321 	if (reg2 >= 0)
322 		emit(gen, BPF_MOV64_REG(BPF_REG_4, reg2));
323 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_trace_printk));
324 }
325 
326 static void debug_regs(struct bpf_gen *gen, int reg1, int reg2, const char *fmt, ...)
327 {
328 	va_list args;
329 
330 	va_start(args, fmt);
331 	emit_debug(gen, reg1, reg2, fmt, args);
332 	va_end(args);
333 }
334 
335 static void debug_ret(struct bpf_gen *gen, const char *fmt, ...)
336 {
337 	va_list args;
338 
339 	va_start(args, fmt);
340 	emit_debug(gen, BPF_REG_7, -1, fmt, args);
341 	va_end(args);
342 }
343 
344 static void __emit_sys_close(struct bpf_gen *gen)
345 {
346 	emit(gen, BPF_JMP_IMM(BPF_JSLE, BPF_REG_1, 0,
347 			      /* 2 is the number of the following insns
348 			       * * 6 is additional insns in debug_regs
349 			       */
350 			      2 + (gen->log_level ? 6 : 0)));
351 	emit(gen, BPF_MOV64_REG(BPF_REG_9, BPF_REG_1));
352 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_sys_close));
353 	debug_regs(gen, BPF_REG_9, BPF_REG_0, "close(%%d) = %%d");
354 }
355 
356 static void emit_sys_close_stack(struct bpf_gen *gen, int stack_off)
357 {
358 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_10, stack_off));
359 	__emit_sys_close(gen);
360 }
361 
362 static void emit_sys_close_blob(struct bpf_gen *gen, int blob_off)
363 {
364 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_0, BPF_PSEUDO_MAP_IDX_VALUE,
365 					 0, 0, 0, blob_off));
366 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_0, 0));
367 	__emit_sys_close(gen);
368 }
369 
370 int bpf_gen__finish(struct bpf_gen *gen, int nr_progs, int nr_maps)
371 {
372 	int i;
373 
374 	if (nr_progs < gen->nr_progs || nr_maps != gen->nr_maps) {
375 		pr_warn("nr_progs %d/%d nr_maps %d/%d mismatch\n",
376 			nr_progs, gen->nr_progs, nr_maps, gen->nr_maps);
377 		gen->error = -EFAULT;
378 		return gen->error;
379 	}
380 	emit_sys_close_stack(gen, stack_off(btf_fd));
381 	for (i = 0; i < gen->nr_progs; i++)
382 		move_stack2ctx(gen,
383 			       sizeof(struct bpf_loader_ctx) +
384 			       sizeof(struct bpf_map_desc) * gen->nr_maps +
385 			       sizeof(struct bpf_prog_desc) * i +
386 			       offsetof(struct bpf_prog_desc, prog_fd), 4,
387 			       stack_off(prog_fd[i]));
388 	for (i = 0; i < gen->nr_maps; i++)
389 		move_blob2ctx(gen,
390 			      sizeof(struct bpf_loader_ctx) +
391 			      sizeof(struct bpf_map_desc) * i +
392 			      offsetof(struct bpf_map_desc, map_fd), 4,
393 			      blob_fd_array_off(gen, i));
394 	emit(gen, BPF_MOV64_IMM(BPF_REG_0, 0));
395 	emit(gen, BPF_EXIT_INSN());
396 	pr_debug("gen: finish %d\n", gen->error);
397 	if (!gen->error) {
398 		struct gen_loader_opts *opts = gen->opts;
399 
400 		opts->insns = gen->insn_start;
401 		opts->insns_sz = gen->insn_cur - gen->insn_start;
402 		opts->data = gen->data_start;
403 		opts->data_sz = gen->data_cur - gen->data_start;
404 
405 		/* use target endianness for embedded loader */
406 		if (gen->swapped_endian) {
407 			struct bpf_insn *insn = (struct bpf_insn *)opts->insns;
408 			int insn_cnt = opts->insns_sz / sizeof(struct bpf_insn);
409 
410 			for (i = 0; i < insn_cnt; i++)
411 				bpf_insn_bswap(insn++);
412 		}
413 	}
414 	return gen->error;
415 }
416 
417 void bpf_gen__free(struct bpf_gen *gen)
418 {
419 	if (!gen)
420 		return;
421 	free(gen->data_start);
422 	free(gen->insn_start);
423 	free(gen);
424 }
425 
426 /*
427  * Fields of bpf_attr are set to values in native byte-order before being
428  * written to the target-bound data blob, and may need endian conversion.
429  * This macro allows providing the correct value in situ more simply than
430  * writing a separate converter for *all fields* of *all records* included
431  * in union bpf_attr. Note that sizeof(rval) should match the assignment
432  * target to avoid runtime problems.
433  */
434 #define tgt_endian(rval) ({					\
435 	typeof(rval) _val = (rval);				\
436 	if (gen->swapped_endian) {				\
437 		switch (sizeof(_val)) {				\
438 		case 1: break;					\
439 		case 2: _val = bswap_16(_val); break;		\
440 		case 4: _val = bswap_32(_val); break;		\
441 		case 8: _val = bswap_64(_val); break;		\
442 		default: pr_warn("unsupported bswap size!\n");	\
443 		}						\
444 	}							\
445 	_val;							\
446 })
447 
448 void bpf_gen__load_btf(struct bpf_gen *gen, const void *btf_raw_data,
449 		       __u32 btf_raw_size)
450 {
451 	int attr_size = offsetofend(union bpf_attr, btf_log_level);
452 	int btf_data, btf_load_attr;
453 	union bpf_attr attr;
454 
455 	memset(&attr, 0, attr_size);
456 	btf_data = add_data(gen, btf_raw_data, btf_raw_size);
457 
458 	attr.btf_size = tgt_endian(btf_raw_size);
459 	btf_load_attr = add_data(gen, &attr, attr_size);
460 	pr_debug("gen: load_btf: off %d size %d, attr: off %d size %d\n",
461 		 btf_data, btf_raw_size, btf_load_attr, attr_size);
462 
463 	/* populate union bpf_attr with user provided log details */
464 	move_ctx2blob(gen, attr_field(btf_load_attr, btf_log_level), 4,
465 		      offsetof(struct bpf_loader_ctx, log_level), false);
466 	move_ctx2blob(gen, attr_field(btf_load_attr, btf_log_size), 4,
467 		      offsetof(struct bpf_loader_ctx, log_size), false);
468 	move_ctx2blob(gen, attr_field(btf_load_attr, btf_log_buf), 8,
469 		      offsetof(struct bpf_loader_ctx, log_buf), false);
470 	/* populate union bpf_attr with a pointer to the BTF data */
471 	emit_rel_store(gen, attr_field(btf_load_attr, btf), btf_data);
472 	/* emit BTF_LOAD command */
473 	emit_sys_bpf(gen, BPF_BTF_LOAD, btf_load_attr, attr_size);
474 	debug_ret(gen, "btf_load size %d", btf_raw_size);
475 	emit_check_err(gen);
476 	/* remember btf_fd in the stack, if successful */
477 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_7, stack_off(btf_fd)));
478 }
479 
480 void bpf_gen__map_create(struct bpf_gen *gen,
481 			 enum bpf_map_type map_type,
482 			 const char *map_name,
483 			 __u32 key_size, __u32 value_size, __u32 max_entries,
484 			 struct bpf_map_create_opts *map_attr, int map_idx)
485 {
486 	int attr_size = offsetofend(union bpf_attr, map_extra);
487 	bool close_inner_map_fd = false;
488 	int map_create_attr, idx;
489 	union bpf_attr attr;
490 
491 	memset(&attr, 0, attr_size);
492 	attr.map_type = tgt_endian(map_type);
493 	attr.key_size = tgt_endian(key_size);
494 	attr.value_size = tgt_endian(value_size);
495 	attr.map_flags = tgt_endian(map_attr->map_flags);
496 	attr.map_extra = tgt_endian(map_attr->map_extra);
497 	if (map_name)
498 		libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
499 	attr.numa_node = tgt_endian(map_attr->numa_node);
500 	attr.map_ifindex = tgt_endian(map_attr->map_ifindex);
501 	attr.max_entries = tgt_endian(max_entries);
502 	attr.btf_key_type_id = tgt_endian(map_attr->btf_key_type_id);
503 	attr.btf_value_type_id = tgt_endian(map_attr->btf_value_type_id);
504 
505 	map_create_attr = add_data(gen, &attr, attr_size);
506 	pr_debug("gen: map_create: %s idx %d type %d value_type_id %d, attr: off %d size %d\n",
507 		 map_name, map_idx, map_type, map_attr->btf_value_type_id,
508 		 map_create_attr, attr_size);
509 
510 	if (map_attr->btf_value_type_id)
511 		/* populate union bpf_attr with btf_fd saved in the stack earlier */
512 		move_stack2blob(gen, attr_field(map_create_attr, btf_fd), 4,
513 				stack_off(btf_fd));
514 	switch (map_type) {
515 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
516 	case BPF_MAP_TYPE_HASH_OF_MAPS:
517 		move_stack2blob(gen, attr_field(map_create_attr, inner_map_fd), 4,
518 				stack_off(inner_map_fd));
519 		close_inner_map_fd = true;
520 		break;
521 	default:
522 		break;
523 	}
524 	/* conditionally update max_entries */
525 	if (map_idx >= 0)
526 		move_ctx2blob(gen, attr_field(map_create_attr, max_entries), 4,
527 			      sizeof(struct bpf_loader_ctx) +
528 			      sizeof(struct bpf_map_desc) * map_idx +
529 			      offsetof(struct bpf_map_desc, max_entries),
530 			      true /* check that max_entries != 0 */);
531 	/* emit MAP_CREATE command */
532 	emit_sys_bpf(gen, BPF_MAP_CREATE, map_create_attr, attr_size);
533 	debug_ret(gen, "map_create %s idx %d type %d value_size %d value_btf_id %d",
534 		  map_name, map_idx, map_type, value_size,
535 		  map_attr->btf_value_type_id);
536 	emit_check_err(gen);
537 	/* remember map_fd in the stack, if successful */
538 	if (map_idx < 0) {
539 		/* This bpf_gen__map_create() function is called with map_idx >= 0
540 		 * for all maps that libbpf loading logic tracks.
541 		 * It's called with -1 to create an inner map.
542 		 */
543 		emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_7,
544 				      stack_off(inner_map_fd)));
545 	} else if (map_idx != gen->nr_maps) {
546 		gen->error = -EDOM; /* internal bug */
547 		return;
548 	} else {
549 		/* add_map_fd does gen->nr_maps++ */
550 		idx = add_map_fd(gen);
551 		emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
552 						 0, 0, 0, blob_fd_array_off(gen, idx)));
553 		emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_1, BPF_REG_7, 0));
554 	}
555 	if (close_inner_map_fd)
556 		emit_sys_close_stack(gen, stack_off(inner_map_fd));
557 }
558 
559 void bpf_gen__record_attach_target(struct bpf_gen *gen, const char *attach_name,
560 				   enum bpf_attach_type type)
561 {
562 	const char *prefix;
563 	int kind, ret;
564 
565 	btf_get_kernel_prefix_kind(type, &prefix, &kind);
566 	gen->attach_kind = kind;
567 	ret = snprintf(gen->attach_target, sizeof(gen->attach_target), "%s%s",
568 		       prefix, attach_name);
569 	if (ret >= sizeof(gen->attach_target))
570 		gen->error = -ENOSPC;
571 }
572 
573 static void emit_find_attach_target(struct bpf_gen *gen)
574 {
575 	int name, len = strlen(gen->attach_target) + 1;
576 
577 	pr_debug("gen: find_attach_tgt %s %d\n", gen->attach_target, gen->attach_kind);
578 	name = add_data(gen, gen->attach_target, len);
579 
580 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
581 					 0, 0, 0, name));
582 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, len));
583 	emit(gen, BPF_MOV64_IMM(BPF_REG_3, gen->attach_kind));
584 	emit(gen, BPF_MOV64_IMM(BPF_REG_4, 0));
585 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_btf_find_by_name_kind));
586 	emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_0));
587 	debug_ret(gen, "find_by_name_kind(%s,%d)",
588 		  gen->attach_target, gen->attach_kind);
589 	emit_check_err(gen);
590 	/* if successful, btf_id is in lower 32-bit of R7 and
591 	 * btf_obj_fd is in upper 32-bit
592 	 */
593 }
594 
595 void bpf_gen__record_extern(struct bpf_gen *gen, const char *name, bool is_weak,
596 			    bool is_typeless, bool is_ld64, int kind, int insn_idx)
597 {
598 	struct ksym_relo_desc *relo;
599 
600 	relo = libbpf_reallocarray(gen->relos, gen->relo_cnt + 1, sizeof(*relo));
601 	if (!relo) {
602 		gen->error = -ENOMEM;
603 		return;
604 	}
605 	gen->relos = relo;
606 	relo += gen->relo_cnt;
607 	relo->name = name;
608 	relo->is_weak = is_weak;
609 	relo->is_typeless = is_typeless;
610 	relo->is_ld64 = is_ld64;
611 	relo->kind = kind;
612 	relo->insn_idx = insn_idx;
613 	gen->relo_cnt++;
614 }
615 
616 /* returns existing ksym_desc with ref incremented, or inserts a new one */
617 static struct ksym_desc *get_ksym_desc(struct bpf_gen *gen, struct ksym_relo_desc *relo)
618 {
619 	struct ksym_desc *kdesc;
620 	int i;
621 
622 	for (i = 0; i < gen->nr_ksyms; i++) {
623 		kdesc = &gen->ksyms[i];
624 		if (kdesc->kind == relo->kind && kdesc->is_ld64 == relo->is_ld64 &&
625 		    !strcmp(kdesc->name, relo->name)) {
626 			kdesc->ref++;
627 			return kdesc;
628 		}
629 	}
630 	kdesc = libbpf_reallocarray(gen->ksyms, gen->nr_ksyms + 1, sizeof(*kdesc));
631 	if (!kdesc) {
632 		gen->error = -ENOMEM;
633 		return NULL;
634 	}
635 	gen->ksyms = kdesc;
636 	kdesc = &gen->ksyms[gen->nr_ksyms++];
637 	kdesc->name = relo->name;
638 	kdesc->kind = relo->kind;
639 	kdesc->ref = 1;
640 	kdesc->off = 0;
641 	kdesc->insn = 0;
642 	kdesc->is_ld64 = relo->is_ld64;
643 	return kdesc;
644 }
645 
646 /* Overwrites BPF_REG_{0, 1, 2, 3, 4, 7}
647  * Returns result in BPF_REG_7
648  */
649 static void emit_bpf_find_by_name_kind(struct bpf_gen *gen, struct ksym_relo_desc *relo)
650 {
651 	int name_off, len = strlen(relo->name) + 1;
652 
653 	name_off = add_data(gen, relo->name, len);
654 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
655 					 0, 0, 0, name_off));
656 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, len));
657 	emit(gen, BPF_MOV64_IMM(BPF_REG_3, relo->kind));
658 	emit(gen, BPF_MOV64_IMM(BPF_REG_4, 0));
659 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_btf_find_by_name_kind));
660 	emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_0));
661 	debug_ret(gen, "find_by_name_kind(%s,%d)", relo->name, relo->kind);
662 }
663 
664 /* Overwrites BPF_REG_{0, 1, 2, 3, 4, 7}
665  * Returns result in BPF_REG_7
666  * Returns u64 symbol addr in BPF_REG_9
667  */
668 static void emit_bpf_kallsyms_lookup_name(struct bpf_gen *gen, struct ksym_relo_desc *relo)
669 {
670 	int name_off, len = strlen(relo->name) + 1, res_off;
671 
672 	name_off = add_data(gen, relo->name, len);
673 	res_off = add_data(gen, NULL, 8); /* res is u64 */
674 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
675 					 0, 0, 0, name_off));
676 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, len));
677 	emit(gen, BPF_MOV64_IMM(BPF_REG_3, 0));
678 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_4, BPF_PSEUDO_MAP_IDX_VALUE,
679 					 0, 0, 0, res_off));
680 	emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_4));
681 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_kallsyms_lookup_name));
682 	emit(gen, BPF_LDX_MEM(BPF_DW, BPF_REG_9, BPF_REG_7, 0));
683 	emit(gen, BPF_MOV64_REG(BPF_REG_7, BPF_REG_0));
684 	debug_ret(gen, "kallsyms_lookup_name(%s,%d)", relo->name, relo->kind);
685 }
686 
687 /* Expects:
688  * BPF_REG_8 - pointer to instruction
689  *
690  * We need to reuse BTF fd for same symbol otherwise each relocation takes a new
691  * index, while kernel limits total kfunc BTFs to 256. For duplicate symbols,
692  * this would mean a new BTF fd index for each entry. By pairing symbol name
693  * with index, we get the insn->imm, insn->off pairing that kernel uses for
694  * kfunc_tab, which becomes the effective limit even though all of them may
695  * share same index in fd_array (such that kfunc_btf_tab has 1 element).
696  */
697 static void emit_relo_kfunc_btf(struct bpf_gen *gen, struct ksym_relo_desc *relo, int insn)
698 {
699 	struct ksym_desc *kdesc;
700 	int btf_fd_idx;
701 
702 	kdesc = get_ksym_desc(gen, relo);
703 	if (!kdesc)
704 		return;
705 	/* try to copy from existing bpf_insn */
706 	if (kdesc->ref > 1) {
707 		move_blob2blob(gen, insn + offsetof(struct bpf_insn, imm), 4,
708 			       kdesc->insn + offsetof(struct bpf_insn, imm));
709 		move_blob2blob(gen, insn + offsetof(struct bpf_insn, off), 2,
710 			       kdesc->insn + offsetof(struct bpf_insn, off));
711 		goto log;
712 	}
713 	/* remember insn offset, so we can copy BTF ID and FD later */
714 	kdesc->insn = insn;
715 	emit_bpf_find_by_name_kind(gen, relo);
716 	if (!relo->is_weak)
717 		emit_check_err(gen);
718 	/* get index in fd_array to store BTF FD at */
719 	btf_fd_idx = add_kfunc_btf_fd(gen);
720 	if (btf_fd_idx > INT16_MAX) {
721 		pr_warn("BTF fd off %d for kfunc %s exceeds INT16_MAX, cannot process relocation\n",
722 			btf_fd_idx, relo->name);
723 		gen->error = -E2BIG;
724 		return;
725 	}
726 	kdesc->off = btf_fd_idx;
727 	/* jump to success case */
728 	emit(gen, BPF_JMP_IMM(BPF_JSGE, BPF_REG_7, 0, 3));
729 	/* set value for imm, off as 0 */
730 	emit(gen, BPF_ST_MEM(BPF_W, BPF_REG_8, offsetof(struct bpf_insn, imm), 0));
731 	emit(gen, BPF_ST_MEM(BPF_H, BPF_REG_8, offsetof(struct bpf_insn, off), 0));
732 	/* skip success case for ret < 0 */
733 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, 10));
734 	/* store btf_id into insn[insn_idx].imm */
735 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_7, offsetof(struct bpf_insn, imm)));
736 	/* obtain fd in BPF_REG_9 */
737 	emit(gen, BPF_MOV64_REG(BPF_REG_9, BPF_REG_7));
738 	emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_9, 32));
739 	/* load fd_array slot pointer */
740 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_0, BPF_PSEUDO_MAP_IDX_VALUE,
741 					 0, 0, 0, blob_fd_array_off(gen, btf_fd_idx)));
742 	/* store BTF fd in slot, 0 for vmlinux */
743 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_0, BPF_REG_9, 0));
744 	/* jump to insn[insn_idx].off store if fd denotes module BTF */
745 	emit(gen, BPF_JMP_IMM(BPF_JNE, BPF_REG_9, 0, 2));
746 	/* set the default value for off */
747 	emit(gen, BPF_ST_MEM(BPF_H, BPF_REG_8, offsetof(struct bpf_insn, off), 0));
748 	/* skip BTF fd store for vmlinux BTF */
749 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, 1));
750 	/* store index into insn[insn_idx].off */
751 	emit(gen, BPF_ST_MEM(BPF_H, BPF_REG_8, offsetof(struct bpf_insn, off), btf_fd_idx));
752 log:
753 	if (!gen->log_level)
754 		return;
755 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_8,
756 			      offsetof(struct bpf_insn, imm)));
757 	emit(gen, BPF_LDX_MEM(BPF_H, BPF_REG_9, BPF_REG_8,
758 			      offsetof(struct bpf_insn, off)));
759 	debug_regs(gen, BPF_REG_7, BPF_REG_9, " func (%s:count=%d): imm: %%d, off: %%d",
760 		   relo->name, kdesc->ref);
761 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_0, BPF_PSEUDO_MAP_IDX_VALUE,
762 					 0, 0, 0, blob_fd_array_off(gen, kdesc->off)));
763 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_9, BPF_REG_0, 0));
764 	debug_regs(gen, BPF_REG_9, -1, " func (%s:count=%d): btf_fd",
765 		   relo->name, kdesc->ref);
766 }
767 
768 static void emit_ksym_relo_log(struct bpf_gen *gen, struct ksym_relo_desc *relo,
769 			       int ref)
770 {
771 	if (!gen->log_level)
772 		return;
773 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_8,
774 			      offsetof(struct bpf_insn, imm)));
775 	emit(gen, BPF_LDX_MEM(BPF_H, BPF_REG_9, BPF_REG_8, sizeof(struct bpf_insn) +
776 			      offsetof(struct bpf_insn, imm)));
777 	debug_regs(gen, BPF_REG_7, BPF_REG_9, " var t=%d w=%d (%s:count=%d): imm[0]: %%d, imm[1]: %%d",
778 		   relo->is_typeless, relo->is_weak, relo->name, ref);
779 	emit(gen, BPF_LDX_MEM(BPF_B, BPF_REG_9, BPF_REG_8, offsetofend(struct bpf_insn, code)));
780 	debug_regs(gen, BPF_REG_9, -1, " var t=%d w=%d (%s:count=%d): insn.reg",
781 		   relo->is_typeless, relo->is_weak, relo->name, ref);
782 }
783 
784 /* Expects:
785  * BPF_REG_8 - pointer to instruction
786  */
787 static void emit_relo_ksym_typeless(struct bpf_gen *gen,
788 				    struct ksym_relo_desc *relo, int insn)
789 {
790 	struct ksym_desc *kdesc;
791 
792 	kdesc = get_ksym_desc(gen, relo);
793 	if (!kdesc)
794 		return;
795 	/* try to copy from existing ldimm64 insn */
796 	if (kdesc->ref > 1) {
797 		move_blob2blob(gen, insn + offsetof(struct bpf_insn, imm), 4,
798 			       kdesc->insn + offsetof(struct bpf_insn, imm));
799 		move_blob2blob(gen, insn + sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm), 4,
800 			       kdesc->insn + sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm));
801 		goto log;
802 	}
803 	/* remember insn offset, so we can copy ksym addr later */
804 	kdesc->insn = insn;
805 	/* skip typeless ksym_desc in fd closing loop in cleanup_relos */
806 	kdesc->typeless = true;
807 	emit_bpf_kallsyms_lookup_name(gen, relo);
808 	emit(gen, BPF_JMP_IMM(BPF_JEQ, BPF_REG_7, -ENOENT, 1));
809 	emit_check_err(gen);
810 	/* store lower half of addr into insn[insn_idx].imm */
811 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_9, offsetof(struct bpf_insn, imm)));
812 	/* store upper half of addr into insn[insn_idx + 1].imm */
813 	emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_9, 32));
814 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_9,
815 		      sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm)));
816 log:
817 	emit_ksym_relo_log(gen, relo, kdesc->ref);
818 }
819 
820 static __u32 src_reg_mask(struct bpf_gen *gen)
821 {
822 #if defined(__LITTLE_ENDIAN_BITFIELD) /* src_reg,dst_reg,... */
823 	return gen->swapped_endian ? 0xf0 : 0x0f;
824 #elif defined(__BIG_ENDIAN_BITFIELD) /* dst_reg,src_reg,... */
825 	return gen->swapped_endian ? 0x0f : 0xf0;
826 #else
827 #error "Unsupported bit endianness, cannot proceed"
828 #endif
829 }
830 
831 /* Expects:
832  * BPF_REG_8 - pointer to instruction
833  */
834 static void emit_relo_ksym_btf(struct bpf_gen *gen, struct ksym_relo_desc *relo, int insn)
835 {
836 	struct ksym_desc *kdesc;
837 	__u32 reg_mask;
838 
839 	kdesc = get_ksym_desc(gen, relo);
840 	if (!kdesc)
841 		return;
842 	/* try to copy from existing ldimm64 insn */
843 	if (kdesc->ref > 1) {
844 		move_blob2blob(gen, insn + sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm), 4,
845 			       kdesc->insn + sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm));
846 		move_blob2blob(gen, insn + offsetof(struct bpf_insn, imm), 4,
847 			       kdesc->insn + offsetof(struct bpf_insn, imm));
848 		/* jump over src_reg adjustment if imm (btf_id) is not 0, reuse BPF_REG_0 from move_blob2blob
849 		 * If btf_id is zero, clear BPF_PSEUDO_BTF_ID flag in src_reg of ld_imm64 insn
850 		 */
851 		emit(gen, BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 3));
852 		goto clear_src_reg;
853 	}
854 	/* remember insn offset, so we can copy BTF ID and FD later */
855 	kdesc->insn = insn;
856 	emit_bpf_find_by_name_kind(gen, relo);
857 	if (!relo->is_weak)
858 		emit_check_err(gen);
859 	/* jump to success case */
860 	emit(gen, BPF_JMP_IMM(BPF_JSGE, BPF_REG_7, 0, 3));
861 	/* set values for insn[insn_idx].imm, insn[insn_idx + 1].imm as 0 */
862 	emit(gen, BPF_ST_MEM(BPF_W, BPF_REG_8, offsetof(struct bpf_insn, imm), 0));
863 	emit(gen, BPF_ST_MEM(BPF_W, BPF_REG_8, sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm), 0));
864 	/* skip success case for ret < 0 */
865 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, 4));
866 	/* store btf_id into insn[insn_idx].imm */
867 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_7, offsetof(struct bpf_insn, imm)));
868 	/* store btf_obj_fd into insn[insn_idx + 1].imm */
869 	emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_7, 32));
870 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_8, BPF_REG_7,
871 			      sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm)));
872 	/* skip src_reg adjustment */
873 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, 3));
874 clear_src_reg:
875 	/* clear bpf_object__relocate_data's src_reg assignment, otherwise we get a verifier failure */
876 	reg_mask = src_reg_mask(gen);
877 	emit(gen, BPF_LDX_MEM(BPF_B, BPF_REG_9, BPF_REG_8, offsetofend(struct bpf_insn, code)));
878 	emit(gen, BPF_ALU32_IMM(BPF_AND, BPF_REG_9, reg_mask));
879 	emit(gen, BPF_STX_MEM(BPF_B, BPF_REG_8, BPF_REG_9, offsetofend(struct bpf_insn, code)));
880 
881 	emit_ksym_relo_log(gen, relo, kdesc->ref);
882 }
883 
884 void bpf_gen__record_relo_core(struct bpf_gen *gen,
885 			       const struct bpf_core_relo *core_relo)
886 {
887 	struct bpf_core_relo *relos;
888 
889 	relos = libbpf_reallocarray(gen->core_relos, gen->core_relo_cnt + 1, sizeof(*relos));
890 	if (!relos) {
891 		gen->error = -ENOMEM;
892 		return;
893 	}
894 	gen->core_relos = relos;
895 	relos += gen->core_relo_cnt;
896 	memcpy(relos, core_relo, sizeof(*relos));
897 	gen->core_relo_cnt++;
898 }
899 
900 static void emit_relo(struct bpf_gen *gen, struct ksym_relo_desc *relo, int insns)
901 {
902 	int insn;
903 
904 	pr_debug("gen: emit_relo (%d): %s at %d %s\n",
905 		 relo->kind, relo->name, relo->insn_idx, relo->is_ld64 ? "ld64" : "call");
906 	insn = insns + sizeof(struct bpf_insn) * relo->insn_idx;
907 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_8, BPF_PSEUDO_MAP_IDX_VALUE, 0, 0, 0, insn));
908 	if (relo->is_ld64) {
909 		if (relo->is_typeless)
910 			emit_relo_ksym_typeless(gen, relo, insn);
911 		else
912 			emit_relo_ksym_btf(gen, relo, insn);
913 	} else {
914 		emit_relo_kfunc_btf(gen, relo, insn);
915 	}
916 }
917 
918 static void emit_relos(struct bpf_gen *gen, int insns)
919 {
920 	int i;
921 
922 	for (i = 0; i < gen->relo_cnt; i++)
923 		emit_relo(gen, gen->relos + i, insns);
924 }
925 
926 static void cleanup_core_relo(struct bpf_gen *gen)
927 {
928 	if (!gen->core_relo_cnt)
929 		return;
930 	free(gen->core_relos);
931 	gen->core_relo_cnt = 0;
932 	gen->core_relos = NULL;
933 }
934 
935 static void cleanup_relos(struct bpf_gen *gen, int insns)
936 {
937 	struct ksym_desc *kdesc;
938 	int i, insn;
939 
940 	for (i = 0; i < gen->nr_ksyms; i++) {
941 		kdesc = &gen->ksyms[i];
942 		/* only close fds for typed ksyms and kfuncs */
943 		if (kdesc->is_ld64 && !kdesc->typeless) {
944 			/* close fd recorded in insn[insn_idx + 1].imm */
945 			insn = kdesc->insn;
946 			insn += sizeof(struct bpf_insn) + offsetof(struct bpf_insn, imm);
947 			emit_sys_close_blob(gen, insn);
948 		} else if (!kdesc->is_ld64) {
949 			emit_sys_close_blob(gen, blob_fd_array_off(gen, kdesc->off));
950 			if (kdesc->off < MAX_FD_ARRAY_SZ)
951 				gen->nr_fd_array--;
952 		}
953 	}
954 	if (gen->nr_ksyms) {
955 		free(gen->ksyms);
956 		gen->nr_ksyms = 0;
957 		gen->ksyms = NULL;
958 	}
959 	if (gen->relo_cnt) {
960 		free(gen->relos);
961 		gen->relo_cnt = 0;
962 		gen->relos = NULL;
963 	}
964 	cleanup_core_relo(gen);
965 }
966 
967 /* Convert func, line, and core relo info blobs to target endianness */
968 static void info_blob_bswap(struct bpf_gen *gen, int func_info, int line_info,
969 			    int core_relos, struct bpf_prog_load_opts *load_attr)
970 {
971 	struct bpf_func_info *fi = gen->data_start + func_info;
972 	struct bpf_line_info *li = gen->data_start + line_info;
973 	struct bpf_core_relo *cr = gen->data_start + core_relos;
974 	int i;
975 
976 	for (i = 0; i < load_attr->func_info_cnt; i++)
977 		bpf_func_info_bswap(fi++);
978 
979 	for (i = 0; i < load_attr->line_info_cnt; i++)
980 		bpf_line_info_bswap(li++);
981 
982 	for (i = 0; i < gen->core_relo_cnt; i++)
983 		bpf_core_relo_bswap(cr++);
984 }
985 
986 void bpf_gen__prog_load(struct bpf_gen *gen,
987 			enum bpf_prog_type prog_type, const char *prog_name,
988 			const char *license, struct bpf_insn *insns, size_t insn_cnt,
989 			struct bpf_prog_load_opts *load_attr, int prog_idx)
990 {
991 	int func_info_tot_sz = load_attr->func_info_cnt *
992 			       load_attr->func_info_rec_size;
993 	int line_info_tot_sz = load_attr->line_info_cnt *
994 			       load_attr->line_info_rec_size;
995 	int core_relo_tot_sz = gen->core_relo_cnt *
996 			       sizeof(struct bpf_core_relo);
997 	int prog_load_attr, license_off, insns_off, func_info, line_info, core_relos;
998 	int attr_size = offsetofend(union bpf_attr, core_relo_rec_size);
999 	union bpf_attr attr;
1000 
1001 	memset(&attr, 0, attr_size);
1002 	/* add license string to blob of bytes */
1003 	license_off = add_data(gen, license, strlen(license) + 1);
1004 	/* add insns to blob of bytes */
1005 	insns_off = add_data(gen, insns, insn_cnt * sizeof(struct bpf_insn));
1006 	pr_debug("gen: prog_load: prog_idx %d type %d insn off %d insns_cnt %zd license off %d\n",
1007 		 prog_idx, prog_type, insns_off, insn_cnt, license_off);
1008 
1009 	/* convert blob insns to target endianness */
1010 	if (gen->swapped_endian) {
1011 		struct bpf_insn *insn = gen->data_start + insns_off;
1012 		int i;
1013 
1014 		for (i = 0; i < insn_cnt; i++, insn++)
1015 			bpf_insn_bswap(insn);
1016 	}
1017 
1018 	attr.prog_type = tgt_endian(prog_type);
1019 	attr.expected_attach_type = tgt_endian(load_attr->expected_attach_type);
1020 	attr.attach_btf_id = tgt_endian(load_attr->attach_btf_id);
1021 	attr.prog_ifindex = tgt_endian(load_attr->prog_ifindex);
1022 	attr.kern_version = 0;
1023 	attr.insn_cnt = tgt_endian((__u32)insn_cnt);
1024 	attr.prog_flags = tgt_endian(load_attr->prog_flags);
1025 
1026 	attr.func_info_rec_size = tgt_endian(load_attr->func_info_rec_size);
1027 	attr.func_info_cnt = tgt_endian(load_attr->func_info_cnt);
1028 	func_info = add_data(gen, load_attr->func_info, func_info_tot_sz);
1029 	pr_debug("gen: prog_load: func_info: off %d cnt %d rec size %d\n",
1030 		 func_info, load_attr->func_info_cnt,
1031 		 load_attr->func_info_rec_size);
1032 
1033 	attr.line_info_rec_size = tgt_endian(load_attr->line_info_rec_size);
1034 	attr.line_info_cnt = tgt_endian(load_attr->line_info_cnt);
1035 	line_info = add_data(gen, load_attr->line_info, line_info_tot_sz);
1036 	pr_debug("gen: prog_load: line_info: off %d cnt %d rec size %d\n",
1037 		 line_info, load_attr->line_info_cnt,
1038 		 load_attr->line_info_rec_size);
1039 
1040 	attr.core_relo_rec_size = tgt_endian((__u32)sizeof(struct bpf_core_relo));
1041 	attr.core_relo_cnt = tgt_endian(gen->core_relo_cnt);
1042 	core_relos = add_data(gen, gen->core_relos, core_relo_tot_sz);
1043 	pr_debug("gen: prog_load: core_relos: off %d cnt %d rec size %zd\n",
1044 		 core_relos, gen->core_relo_cnt,
1045 		 sizeof(struct bpf_core_relo));
1046 
1047 	/* convert all info blobs to target endianness */
1048 	if (gen->swapped_endian)
1049 		info_blob_bswap(gen, func_info, line_info, core_relos, load_attr);
1050 
1051 	libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
1052 	prog_load_attr = add_data(gen, &attr, attr_size);
1053 	pr_debug("gen: prog_load: attr: off %d size %d\n",
1054 		 prog_load_attr, attr_size);
1055 
1056 	/* populate union bpf_attr with a pointer to license */
1057 	emit_rel_store(gen, attr_field(prog_load_attr, license), license_off);
1058 
1059 	/* populate union bpf_attr with a pointer to instructions */
1060 	emit_rel_store(gen, attr_field(prog_load_attr, insns), insns_off);
1061 
1062 	/* populate union bpf_attr with a pointer to func_info */
1063 	emit_rel_store(gen, attr_field(prog_load_attr, func_info), func_info);
1064 
1065 	/* populate union bpf_attr with a pointer to line_info */
1066 	emit_rel_store(gen, attr_field(prog_load_attr, line_info), line_info);
1067 
1068 	/* populate union bpf_attr with a pointer to core_relos */
1069 	emit_rel_store(gen, attr_field(prog_load_attr, core_relos), core_relos);
1070 
1071 	/* populate union bpf_attr fd_array with a pointer to data where map_fds are saved */
1072 	emit_rel_store(gen, attr_field(prog_load_attr, fd_array), gen->fd_array);
1073 
1074 	/* populate union bpf_attr with user provided log details */
1075 	move_ctx2blob(gen, attr_field(prog_load_attr, log_level), 4,
1076 		      offsetof(struct bpf_loader_ctx, log_level), false);
1077 	move_ctx2blob(gen, attr_field(prog_load_attr, log_size), 4,
1078 		      offsetof(struct bpf_loader_ctx, log_size), false);
1079 	move_ctx2blob(gen, attr_field(prog_load_attr, log_buf), 8,
1080 		      offsetof(struct bpf_loader_ctx, log_buf), false);
1081 	/* populate union bpf_attr with btf_fd saved in the stack earlier */
1082 	move_stack2blob(gen, attr_field(prog_load_attr, prog_btf_fd), 4,
1083 			stack_off(btf_fd));
1084 	if (gen->attach_kind) {
1085 		emit_find_attach_target(gen);
1086 		/* populate union bpf_attr with btf_id and btf_obj_fd found by helper */
1087 		emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_0, BPF_PSEUDO_MAP_IDX_VALUE,
1088 						 0, 0, 0, prog_load_attr));
1089 		emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_0, BPF_REG_7,
1090 				      offsetof(union bpf_attr, attach_btf_id)));
1091 		emit(gen, BPF_ALU64_IMM(BPF_RSH, BPF_REG_7, 32));
1092 		emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_0, BPF_REG_7,
1093 				      offsetof(union bpf_attr, attach_btf_obj_fd)));
1094 	}
1095 	emit_relos(gen, insns_off);
1096 	/* emit PROG_LOAD command */
1097 	emit_sys_bpf(gen, BPF_PROG_LOAD, prog_load_attr, attr_size);
1098 	debug_ret(gen, "prog_load %s insn_cnt %d", attr.prog_name, attr.insn_cnt);
1099 	/* successful or not, close btf module FDs used in extern ksyms and attach_btf_obj_fd */
1100 	cleanup_relos(gen, insns_off);
1101 	if (gen->attach_kind) {
1102 		emit_sys_close_blob(gen,
1103 				    attr_field(prog_load_attr, attach_btf_obj_fd));
1104 		gen->attach_kind = 0;
1105 	}
1106 	emit_check_err(gen);
1107 	/* remember prog_fd in the stack, if successful */
1108 	emit(gen, BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_7,
1109 			      stack_off(prog_fd[gen->nr_progs])));
1110 	gen->nr_progs++;
1111 }
1112 
1113 void bpf_gen__map_update_elem(struct bpf_gen *gen, int map_idx, void *pvalue,
1114 			      __u32 value_size)
1115 {
1116 	int attr_size = offsetofend(union bpf_attr, flags);
1117 	int map_update_attr, value, key;
1118 	union bpf_attr attr;
1119 	int zero = 0;
1120 
1121 	memset(&attr, 0, attr_size);
1122 
1123 	value = add_data(gen, pvalue, value_size);
1124 	key = add_data(gen, &zero, sizeof(zero));
1125 
1126 	/* if (map_desc[map_idx].initial_value) {
1127 	 *    if (ctx->flags & BPF_SKEL_KERNEL)
1128 	 *        bpf_probe_read_kernel(value, value_size, initial_value);
1129 	 *    else
1130 	 *        bpf_copy_from_user(value, value_size, initial_value);
1131 	 * }
1132 	 */
1133 	emit(gen, BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_6,
1134 			      sizeof(struct bpf_loader_ctx) +
1135 			      sizeof(struct bpf_map_desc) * map_idx +
1136 			      offsetof(struct bpf_map_desc, initial_value)));
1137 	emit(gen, BPF_JMP_IMM(BPF_JEQ, BPF_REG_3, 0, 8));
1138 	emit2(gen, BPF_LD_IMM64_RAW_FULL(BPF_REG_1, BPF_PSEUDO_MAP_IDX_VALUE,
1139 					 0, 0, 0, value));
1140 	emit(gen, BPF_MOV64_IMM(BPF_REG_2, value_size));
1141 	emit(gen, BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_6,
1142 			      offsetof(struct bpf_loader_ctx, flags)));
1143 	emit(gen, BPF_JMP_IMM(BPF_JSET, BPF_REG_0, BPF_SKEL_KERNEL, 2));
1144 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_copy_from_user));
1145 	emit(gen, BPF_JMP_IMM(BPF_JA, 0, 0, 1));
1146 	emit(gen, BPF_EMIT_CALL(BPF_FUNC_probe_read_kernel));
1147 
1148 	map_update_attr = add_data(gen, &attr, attr_size);
1149 	pr_debug("gen: map_update_elem: idx %d, value: off %d size %d, attr: off %d size %d\n",
1150 		 map_idx, value, value_size, map_update_attr, attr_size);
1151 	move_blob2blob(gen, attr_field(map_update_attr, map_fd), 4,
1152 		       blob_fd_array_off(gen, map_idx));
1153 	emit_rel_store(gen, attr_field(map_update_attr, key), key);
1154 	emit_rel_store(gen, attr_field(map_update_attr, value), value);
1155 	/* emit MAP_UPDATE_ELEM command */
1156 	emit_sys_bpf(gen, BPF_MAP_UPDATE_ELEM, map_update_attr, attr_size);
1157 	debug_ret(gen, "update_elem idx %d value_size %d", map_idx, value_size);
1158 	emit_check_err(gen);
1159 }
1160 
1161 void bpf_gen__populate_outer_map(struct bpf_gen *gen, int outer_map_idx, int slot,
1162 				 int inner_map_idx)
1163 {
1164 	int attr_size = offsetofend(union bpf_attr, flags);
1165 	int map_update_attr, key;
1166 	union bpf_attr attr;
1167 	int tgt_slot;
1168 
1169 	memset(&attr, 0, attr_size);
1170 
1171 	tgt_slot = tgt_endian(slot);
1172 	key = add_data(gen, &tgt_slot, sizeof(tgt_slot));
1173 
1174 	map_update_attr = add_data(gen, &attr, attr_size);
1175 	pr_debug("gen: populate_outer_map: outer %d key %d inner %d, attr: off %d size %d\n",
1176 		 outer_map_idx, slot, inner_map_idx, map_update_attr, attr_size);
1177 	move_blob2blob(gen, attr_field(map_update_attr, map_fd), 4,
1178 		       blob_fd_array_off(gen, outer_map_idx));
1179 	emit_rel_store(gen, attr_field(map_update_attr, key), key);
1180 	emit_rel_store(gen, attr_field(map_update_attr, value),
1181 		       blob_fd_array_off(gen, inner_map_idx));
1182 
1183 	/* emit MAP_UPDATE_ELEM command */
1184 	emit_sys_bpf(gen, BPF_MAP_UPDATE_ELEM, map_update_attr, attr_size);
1185 	debug_ret(gen, "populate_outer_map outer %d key %d inner %d",
1186 		  outer_map_idx, slot, inner_map_idx);
1187 	emit_check_err(gen);
1188 }
1189 
1190 void bpf_gen__map_freeze(struct bpf_gen *gen, int map_idx)
1191 {
1192 	int attr_size = offsetofend(union bpf_attr, map_fd);
1193 	int map_freeze_attr;
1194 	union bpf_attr attr;
1195 
1196 	memset(&attr, 0, attr_size);
1197 	map_freeze_attr = add_data(gen, &attr, attr_size);
1198 	pr_debug("gen: map_freeze: idx %d, attr: off %d size %d\n",
1199 		 map_idx, map_freeze_attr, attr_size);
1200 	move_blob2blob(gen, attr_field(map_freeze_attr, map_fd), 4,
1201 		       blob_fd_array_off(gen, map_idx));
1202 	/* emit MAP_FREEZE command */
1203 	emit_sys_bpf(gen, BPF_MAP_FREEZE, map_freeze_attr, attr_size);
1204 	debug_ret(gen, "map_freeze");
1205 	emit_check_err(gen);
1206 }
1207