1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/mm.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
40 #include <net/sock.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <net/xfrm.h>
60 #include <net/udp.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
68 #include <net/flow.h>
69 #include <net/arp.h>
70 #include <net/ipv6.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
73 #include <net/seg6.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
80 #include <net/tls.h>
81 #include <net/xdp.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
84 #include <net/netkit.h>
85 #include <linux/un.h>
86 #include <net/xdp_sock_drv.h>
87 #include <net/inet_dscp.h>
88
89 #include "dev.h"
90
91 /* Keep the struct bpf_fib_lookup small so that it fits into a cacheline */
92 static_assert(sizeof(struct bpf_fib_lookup) == 64, "struct bpf_fib_lookup size check");
93
94 static const struct bpf_func_proto *
95 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
96
copy_bpf_fprog_from_user(struct sock_fprog * dst,sockptr_t src,int len)97 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
98 {
99 if (in_compat_syscall()) {
100 struct compat_sock_fprog f32;
101
102 if (len != sizeof(f32))
103 return -EINVAL;
104 if (copy_from_sockptr(&f32, src, sizeof(f32)))
105 return -EFAULT;
106 memset(dst, 0, sizeof(*dst));
107 dst->len = f32.len;
108 dst->filter = compat_ptr(f32.filter);
109 } else {
110 if (len != sizeof(*dst))
111 return -EINVAL;
112 if (copy_from_sockptr(dst, src, sizeof(*dst)))
113 return -EFAULT;
114 }
115
116 return 0;
117 }
118 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
119
120 /**
121 * sk_filter_trim_cap - run a packet through a socket filter
122 * @sk: sock associated with &sk_buff
123 * @skb: buffer to filter
124 * @cap: limit on how short the eBPF program may trim the packet
125 *
126 * Run the eBPF program and then cut skb->data to correct size returned by
127 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
128 * than pkt_len we keep whole skb->data. This is the socket level
129 * wrapper to bpf_prog_run. It returns 0 if the packet should
130 * be accepted or -EPERM if the packet should be tossed.
131 *
132 */
sk_filter_trim_cap(struct sock * sk,struct sk_buff * skb,unsigned int cap)133 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
134 {
135 int err;
136 struct sk_filter *filter;
137
138 /*
139 * If the skb was allocated from pfmemalloc reserves, only
140 * allow SOCK_MEMALLOC sockets to use it as this socket is
141 * helping free memory
142 */
143 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
144 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
145 return -ENOMEM;
146 }
147 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
148 if (err)
149 return err;
150
151 err = security_sock_rcv_skb(sk, skb);
152 if (err)
153 return err;
154
155 rcu_read_lock();
156 filter = rcu_dereference(sk->sk_filter);
157 if (filter) {
158 struct sock *save_sk = skb->sk;
159 unsigned int pkt_len;
160
161 skb->sk = sk;
162 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
163 skb->sk = save_sk;
164 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
165 }
166 rcu_read_unlock();
167
168 return err;
169 }
170 EXPORT_SYMBOL(sk_filter_trim_cap);
171
BPF_CALL_1(bpf_skb_get_pay_offset,struct sk_buff *,skb)172 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
173 {
174 return skb_get_poff(skb);
175 }
176
BPF_CALL_3(bpf_skb_get_nlattr,struct sk_buff *,skb,u32,a,u32,x)177 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
178 {
179 struct nlattr *nla;
180
181 if (skb_is_nonlinear(skb))
182 return 0;
183
184 if (skb->len < sizeof(struct nlattr))
185 return 0;
186
187 if (a > skb->len - sizeof(struct nlattr))
188 return 0;
189
190 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
191 if (nla)
192 return (void *) nla - (void *) skb->data;
193
194 return 0;
195 }
196
BPF_CALL_3(bpf_skb_get_nlattr_nest,struct sk_buff *,skb,u32,a,u32,x)197 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
198 {
199 struct nlattr *nla;
200
201 if (skb_is_nonlinear(skb))
202 return 0;
203
204 if (skb->len < sizeof(struct nlattr))
205 return 0;
206
207 if (a > skb->len - sizeof(struct nlattr))
208 return 0;
209
210 nla = (struct nlattr *) &skb->data[a];
211 if (!nla_ok(nla, skb->len - a))
212 return 0;
213
214 nla = nla_find_nested(nla, x);
215 if (nla)
216 return (void *) nla - (void *) skb->data;
217
218 return 0;
219 }
220
BPF_CALL_4(bpf_skb_load_helper_8,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)221 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
222 data, int, headlen, int, offset)
223 {
224 u8 tmp, *ptr;
225 const int len = sizeof(tmp);
226
227 if (offset >= 0) {
228 if (headlen - offset >= len)
229 return *(u8 *)(data + offset);
230 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
231 return tmp;
232 } else {
233 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
234 if (likely(ptr))
235 return *(u8 *)ptr;
236 }
237
238 return -EFAULT;
239 }
240
BPF_CALL_2(bpf_skb_load_helper_8_no_cache,const struct sk_buff *,skb,int,offset)241 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
242 int, offset)
243 {
244 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
245 offset);
246 }
247
BPF_CALL_4(bpf_skb_load_helper_16,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)248 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
249 data, int, headlen, int, offset)
250 {
251 __be16 tmp, *ptr;
252 const int len = sizeof(tmp);
253
254 if (offset >= 0) {
255 if (headlen - offset >= len)
256 return get_unaligned_be16(data + offset);
257 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
258 return be16_to_cpu(tmp);
259 } else {
260 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
261 if (likely(ptr))
262 return get_unaligned_be16(ptr);
263 }
264
265 return -EFAULT;
266 }
267
BPF_CALL_2(bpf_skb_load_helper_16_no_cache,const struct sk_buff *,skb,int,offset)268 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
269 int, offset)
270 {
271 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
272 offset);
273 }
274
BPF_CALL_4(bpf_skb_load_helper_32,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)275 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
276 data, int, headlen, int, offset)
277 {
278 __be32 tmp, *ptr;
279 const int len = sizeof(tmp);
280
281 if (likely(offset >= 0)) {
282 if (headlen - offset >= len)
283 return get_unaligned_be32(data + offset);
284 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
285 return be32_to_cpu(tmp);
286 } else {
287 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
288 if (likely(ptr))
289 return get_unaligned_be32(ptr);
290 }
291
292 return -EFAULT;
293 }
294
BPF_CALL_2(bpf_skb_load_helper_32_no_cache,const struct sk_buff *,skb,int,offset)295 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
296 int, offset)
297 {
298 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
299 offset);
300 }
301
convert_skb_access(int skb_field,int dst_reg,int src_reg,struct bpf_insn * insn_buf)302 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
303 struct bpf_insn *insn_buf)
304 {
305 struct bpf_insn *insn = insn_buf;
306
307 switch (skb_field) {
308 case SKF_AD_MARK:
309 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
310
311 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
312 offsetof(struct sk_buff, mark));
313 break;
314
315 case SKF_AD_PKTTYPE:
316 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
317 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
318 #ifdef __BIG_ENDIAN_BITFIELD
319 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
320 #endif
321 break;
322
323 case SKF_AD_QUEUE:
324 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
325
326 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
327 offsetof(struct sk_buff, queue_mapping));
328 break;
329
330 case SKF_AD_VLAN_TAG:
331 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
332
333 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
334 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
335 offsetof(struct sk_buff, vlan_tci));
336 break;
337 case SKF_AD_VLAN_TAG_PRESENT:
338 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
339 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
340 offsetof(struct sk_buff, vlan_all));
341 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
342 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
343 break;
344 }
345
346 return insn - insn_buf;
347 }
348
convert_bpf_extensions(struct sock_filter * fp,struct bpf_insn ** insnp)349 static bool convert_bpf_extensions(struct sock_filter *fp,
350 struct bpf_insn **insnp)
351 {
352 struct bpf_insn *insn = *insnp;
353 u32 cnt;
354
355 switch (fp->k) {
356 case SKF_AD_OFF + SKF_AD_PROTOCOL:
357 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
358
359 /* A = *(u16 *) (CTX + offsetof(protocol)) */
360 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
361 offsetof(struct sk_buff, protocol));
362 /* A = ntohs(A) [emitting a nop or swap16] */
363 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
364 break;
365
366 case SKF_AD_OFF + SKF_AD_PKTTYPE:
367 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
368 insn += cnt - 1;
369 break;
370
371 case SKF_AD_OFF + SKF_AD_IFINDEX:
372 case SKF_AD_OFF + SKF_AD_HATYPE:
373 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
374 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
375
376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
377 BPF_REG_TMP, BPF_REG_CTX,
378 offsetof(struct sk_buff, dev));
379 /* if (tmp != 0) goto pc + 1 */
380 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
381 *insn++ = BPF_EXIT_INSN();
382 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
383 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
384 offsetof(struct net_device, ifindex));
385 else
386 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
387 offsetof(struct net_device, type));
388 break;
389
390 case SKF_AD_OFF + SKF_AD_MARK:
391 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
392 insn += cnt - 1;
393 break;
394
395 case SKF_AD_OFF + SKF_AD_RXHASH:
396 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
397
398 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
399 offsetof(struct sk_buff, hash));
400 break;
401
402 case SKF_AD_OFF + SKF_AD_QUEUE:
403 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
404 insn += cnt - 1;
405 break;
406
407 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
408 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
409 BPF_REG_A, BPF_REG_CTX, insn);
410 insn += cnt - 1;
411 break;
412
413 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
414 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
415 BPF_REG_A, BPF_REG_CTX, insn);
416 insn += cnt - 1;
417 break;
418
419 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
420 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
421
422 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
423 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
424 offsetof(struct sk_buff, vlan_proto));
425 /* A = ntohs(A) [emitting a nop or swap16] */
426 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
427 break;
428
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 case SKF_AD_OFF + SKF_AD_NLATTR:
431 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
432 case SKF_AD_OFF + SKF_AD_CPU:
433 case SKF_AD_OFF + SKF_AD_RANDOM:
434 /* arg1 = CTX */
435 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
436 /* arg2 = A */
437 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
438 /* arg3 = X */
439 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
440 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
441 switch (fp->k) {
442 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
443 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
444 break;
445 case SKF_AD_OFF + SKF_AD_NLATTR:
446 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
447 break;
448 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
449 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
450 break;
451 case SKF_AD_OFF + SKF_AD_CPU:
452 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
453 break;
454 case SKF_AD_OFF + SKF_AD_RANDOM:
455 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
456 bpf_user_rnd_init_once();
457 break;
458 }
459 break;
460
461 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
462 /* A ^= X */
463 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
464 break;
465
466 default:
467 /* This is just a dummy call to avoid letting the compiler
468 * evict __bpf_call_base() as an optimization. Placed here
469 * where no-one bothers.
470 */
471 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
472 return false;
473 }
474
475 *insnp = insn;
476 return true;
477 }
478
convert_bpf_ld_abs(struct sock_filter * fp,struct bpf_insn ** insnp)479 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
480 {
481 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
482 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
483 bool endian = BPF_SIZE(fp->code) == BPF_H ||
484 BPF_SIZE(fp->code) == BPF_W;
485 bool indirect = BPF_MODE(fp->code) == BPF_IND;
486 const int ip_align = NET_IP_ALIGN;
487 struct bpf_insn *insn = *insnp;
488 int offset = fp->k;
489
490 if (!indirect &&
491 ((unaligned_ok && offset >= 0) ||
492 (!unaligned_ok && offset >= 0 &&
493 offset + ip_align >= 0 &&
494 offset + ip_align % size == 0))) {
495 bool ldx_off_ok = offset <= S16_MAX;
496
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
498 if (offset)
499 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
500 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
501 size, 2 + endian + (!ldx_off_ok * 2));
502 if (ldx_off_ok) {
503 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
504 BPF_REG_D, offset);
505 } else {
506 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
507 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
508 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
509 BPF_REG_TMP, 0);
510 }
511 if (endian)
512 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
513 *insn++ = BPF_JMP_A(8);
514 }
515
516 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
517 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
518 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
519 if (!indirect) {
520 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
521 } else {
522 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
523 if (fp->k)
524 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
525 }
526
527 switch (BPF_SIZE(fp->code)) {
528 case BPF_B:
529 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
530 break;
531 case BPF_H:
532 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
533 break;
534 case BPF_W:
535 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
536 break;
537 default:
538 return false;
539 }
540
541 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
542 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
543 *insn = BPF_EXIT_INSN();
544
545 *insnp = insn;
546 return true;
547 }
548
549 /**
550 * bpf_convert_filter - convert filter program
551 * @prog: the user passed filter program
552 * @len: the length of the user passed filter program
553 * @new_prog: allocated 'struct bpf_prog' or NULL
554 * @new_len: pointer to store length of converted program
555 * @seen_ld_abs: bool whether we've seen ld_abs/ind
556 *
557 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
558 * style extended BPF (eBPF).
559 * Conversion workflow:
560 *
561 * 1) First pass for calculating the new program length:
562 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
563 *
564 * 2) 2nd pass to remap in two passes: 1st pass finds new
565 * jump offsets, 2nd pass remapping:
566 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
567 */
bpf_convert_filter(struct sock_filter * prog,int len,struct bpf_prog * new_prog,int * new_len,bool * seen_ld_abs)568 static int bpf_convert_filter(struct sock_filter *prog, int len,
569 struct bpf_prog *new_prog, int *new_len,
570 bool *seen_ld_abs)
571 {
572 int new_flen = 0, pass = 0, target, i, stack_off;
573 struct bpf_insn *new_insn, *first_insn = NULL;
574 struct sock_filter *fp;
575 int *addrs = NULL;
576 u8 bpf_src;
577
578 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
579 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
580
581 if (len <= 0 || len > BPF_MAXINSNS)
582 return -EINVAL;
583
584 if (new_prog) {
585 first_insn = new_prog->insnsi;
586 addrs = kcalloc(len, sizeof(*addrs),
587 GFP_KERNEL | __GFP_NOWARN);
588 if (!addrs)
589 return -ENOMEM;
590 }
591
592 do_pass:
593 new_insn = first_insn;
594 fp = prog;
595
596 /* Classic BPF related prologue emission. */
597 if (new_prog) {
598 /* Classic BPF expects A and X to be reset first. These need
599 * to be guaranteed to be the first two instructions.
600 */
601 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
602 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
603
604 /* All programs must keep CTX in callee saved BPF_REG_CTX.
605 * In eBPF case it's done by the compiler, here we need to
606 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
607 */
608 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
609 if (*seen_ld_abs) {
610 /* For packet access in classic BPF, cache skb->data
611 * in callee-saved BPF R8 and skb->len - skb->data_len
612 * (headlen) in BPF R9. Since classic BPF is read-only
613 * on CTX, we only need to cache it once.
614 */
615 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
616 BPF_REG_D, BPF_REG_CTX,
617 offsetof(struct sk_buff, data));
618 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
619 offsetof(struct sk_buff, len));
620 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
621 offsetof(struct sk_buff, data_len));
622 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
623 }
624 } else {
625 new_insn += 3;
626 }
627
628 for (i = 0; i < len; fp++, i++) {
629 struct bpf_insn tmp_insns[32] = { };
630 struct bpf_insn *insn = tmp_insns;
631
632 if (addrs)
633 addrs[i] = new_insn - first_insn;
634
635 switch (fp->code) {
636 /* All arithmetic insns and skb loads map as-is. */
637 case BPF_ALU | BPF_ADD | BPF_X:
638 case BPF_ALU | BPF_ADD | BPF_K:
639 case BPF_ALU | BPF_SUB | BPF_X:
640 case BPF_ALU | BPF_SUB | BPF_K:
641 case BPF_ALU | BPF_AND | BPF_X:
642 case BPF_ALU | BPF_AND | BPF_K:
643 case BPF_ALU | BPF_OR | BPF_X:
644 case BPF_ALU | BPF_OR | BPF_K:
645 case BPF_ALU | BPF_LSH | BPF_X:
646 case BPF_ALU | BPF_LSH | BPF_K:
647 case BPF_ALU | BPF_RSH | BPF_X:
648 case BPF_ALU | BPF_RSH | BPF_K:
649 case BPF_ALU | BPF_XOR | BPF_X:
650 case BPF_ALU | BPF_XOR | BPF_K:
651 case BPF_ALU | BPF_MUL | BPF_X:
652 case BPF_ALU | BPF_MUL | BPF_K:
653 case BPF_ALU | BPF_DIV | BPF_X:
654 case BPF_ALU | BPF_DIV | BPF_K:
655 case BPF_ALU | BPF_MOD | BPF_X:
656 case BPF_ALU | BPF_MOD | BPF_K:
657 case BPF_ALU | BPF_NEG:
658 case BPF_LD | BPF_ABS | BPF_W:
659 case BPF_LD | BPF_ABS | BPF_H:
660 case BPF_LD | BPF_ABS | BPF_B:
661 case BPF_LD | BPF_IND | BPF_W:
662 case BPF_LD | BPF_IND | BPF_H:
663 case BPF_LD | BPF_IND | BPF_B:
664 /* Check for overloaded BPF extension and
665 * directly convert it if found, otherwise
666 * just move on with mapping.
667 */
668 if (BPF_CLASS(fp->code) == BPF_LD &&
669 BPF_MODE(fp->code) == BPF_ABS &&
670 convert_bpf_extensions(fp, &insn))
671 break;
672 if (BPF_CLASS(fp->code) == BPF_LD &&
673 convert_bpf_ld_abs(fp, &insn)) {
674 *seen_ld_abs = true;
675 break;
676 }
677
678 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
679 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
680 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
681 /* Error with exception code on div/mod by 0.
682 * For cBPF programs, this was always return 0.
683 */
684 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
685 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
686 *insn++ = BPF_EXIT_INSN();
687 }
688
689 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
690 break;
691
692 /* Jump transformation cannot use BPF block macros
693 * everywhere as offset calculation and target updates
694 * require a bit more work than the rest, i.e. jump
695 * opcodes map as-is, but offsets need adjustment.
696 */
697
698 #define BPF_EMIT_JMP \
699 do { \
700 const s32 off_min = S16_MIN, off_max = S16_MAX; \
701 s32 off; \
702 \
703 if (target >= len || target < 0) \
704 goto err; \
705 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
706 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
707 off -= insn - tmp_insns; \
708 /* Reject anything not fitting into insn->off. */ \
709 if (off < off_min || off > off_max) \
710 goto err; \
711 insn->off = off; \
712 } while (0)
713
714 case BPF_JMP | BPF_JA:
715 target = i + fp->k + 1;
716 insn->code = fp->code;
717 BPF_EMIT_JMP;
718 break;
719
720 case BPF_JMP | BPF_JEQ | BPF_K:
721 case BPF_JMP | BPF_JEQ | BPF_X:
722 case BPF_JMP | BPF_JSET | BPF_K:
723 case BPF_JMP | BPF_JSET | BPF_X:
724 case BPF_JMP | BPF_JGT | BPF_K:
725 case BPF_JMP | BPF_JGT | BPF_X:
726 case BPF_JMP | BPF_JGE | BPF_K:
727 case BPF_JMP | BPF_JGE | BPF_X:
728 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
729 /* BPF immediates are signed, zero extend
730 * immediate into tmp register and use it
731 * in compare insn.
732 */
733 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
734
735 insn->dst_reg = BPF_REG_A;
736 insn->src_reg = BPF_REG_TMP;
737 bpf_src = BPF_X;
738 } else {
739 insn->dst_reg = BPF_REG_A;
740 insn->imm = fp->k;
741 bpf_src = BPF_SRC(fp->code);
742 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
743 }
744
745 /* Common case where 'jump_false' is next insn. */
746 if (fp->jf == 0) {
747 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
748 target = i + fp->jt + 1;
749 BPF_EMIT_JMP;
750 break;
751 }
752
753 /* Convert some jumps when 'jump_true' is next insn. */
754 if (fp->jt == 0) {
755 switch (BPF_OP(fp->code)) {
756 case BPF_JEQ:
757 insn->code = BPF_JMP | BPF_JNE | bpf_src;
758 break;
759 case BPF_JGT:
760 insn->code = BPF_JMP | BPF_JLE | bpf_src;
761 break;
762 case BPF_JGE:
763 insn->code = BPF_JMP | BPF_JLT | bpf_src;
764 break;
765 default:
766 goto jmp_rest;
767 }
768
769 target = i + fp->jf + 1;
770 BPF_EMIT_JMP;
771 break;
772 }
773 jmp_rest:
774 /* Other jumps are mapped into two insns: Jxx and JA. */
775 target = i + fp->jt + 1;
776 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
777 BPF_EMIT_JMP;
778 insn++;
779
780 insn->code = BPF_JMP | BPF_JA;
781 target = i + fp->jf + 1;
782 BPF_EMIT_JMP;
783 break;
784
785 /* ldxb 4 * ([14] & 0xf) is remapped into 6 insns. */
786 case BPF_LDX | BPF_MSH | BPF_B: {
787 struct sock_filter tmp = {
788 .code = BPF_LD | BPF_ABS | BPF_B,
789 .k = fp->k,
790 };
791
792 *seen_ld_abs = true;
793
794 /* X = A */
795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
797 convert_bpf_ld_abs(&tmp, &insn);
798 insn++;
799 /* A &= 0xf */
800 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
801 /* A <<= 2 */
802 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
803 /* tmp = X */
804 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
805 /* X = A */
806 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
807 /* A = tmp */
808 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
809 break;
810 }
811 /* RET_K is remapped into 2 insns. RET_A case doesn't need an
812 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
813 */
814 case BPF_RET | BPF_A:
815 case BPF_RET | BPF_K:
816 if (BPF_RVAL(fp->code) == BPF_K)
817 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
818 0, fp->k);
819 *insn = BPF_EXIT_INSN();
820 break;
821
822 /* Store to stack. */
823 case BPF_ST:
824 case BPF_STX:
825 stack_off = fp->k * 4 + 4;
826 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
827 BPF_ST ? BPF_REG_A : BPF_REG_X,
828 -stack_off);
829 /* check_load_and_stores() verifies that classic BPF can
830 * load from stack only after write, so tracking
831 * stack_depth for ST|STX insns is enough
832 */
833 if (new_prog && new_prog->aux->stack_depth < stack_off)
834 new_prog->aux->stack_depth = stack_off;
835 break;
836
837 /* Load from stack. */
838 case BPF_LD | BPF_MEM:
839 case BPF_LDX | BPF_MEM:
840 stack_off = fp->k * 4 + 4;
841 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
842 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
843 -stack_off);
844 break;
845
846 /* A = K or X = K */
847 case BPF_LD | BPF_IMM:
848 case BPF_LDX | BPF_IMM:
849 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
850 BPF_REG_A : BPF_REG_X, fp->k);
851 break;
852
853 /* X = A */
854 case BPF_MISC | BPF_TAX:
855 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
856 break;
857
858 /* A = X */
859 case BPF_MISC | BPF_TXA:
860 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
861 break;
862
863 /* A = skb->len or X = skb->len */
864 case BPF_LD | BPF_W | BPF_LEN:
865 case BPF_LDX | BPF_W | BPF_LEN:
866 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
867 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
868 offsetof(struct sk_buff, len));
869 break;
870
871 /* Access seccomp_data fields. */
872 case BPF_LDX | BPF_ABS | BPF_W:
873 /* A = *(u32 *) (ctx + K) */
874 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
875 break;
876
877 /* Unknown instruction. */
878 default:
879 goto err;
880 }
881
882 insn++;
883 if (new_prog)
884 memcpy(new_insn, tmp_insns,
885 sizeof(*insn) * (insn - tmp_insns));
886 new_insn += insn - tmp_insns;
887 }
888
889 if (!new_prog) {
890 /* Only calculating new length. */
891 *new_len = new_insn - first_insn;
892 if (*seen_ld_abs)
893 *new_len += 4; /* Prologue bits. */
894 return 0;
895 }
896
897 pass++;
898 if (new_flen != new_insn - first_insn) {
899 new_flen = new_insn - first_insn;
900 if (pass > 2)
901 goto err;
902 goto do_pass;
903 }
904
905 kfree(addrs);
906 BUG_ON(*new_len != new_flen);
907 return 0;
908 err:
909 kfree(addrs);
910 return -EINVAL;
911 }
912
913 /* Security:
914 *
915 * As we dont want to clear mem[] array for each packet going through
916 * __bpf_prog_run(), we check that filter loaded by user never try to read
917 * a cell if not previously written, and we check all branches to be sure
918 * a malicious user doesn't try to abuse us.
919 */
check_load_and_stores(const struct sock_filter * filter,int flen)920 static int check_load_and_stores(const struct sock_filter *filter, int flen)
921 {
922 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
923 int pc, ret = 0;
924
925 BUILD_BUG_ON(BPF_MEMWORDS > 16);
926
927 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
928 if (!masks)
929 return -ENOMEM;
930
931 memset(masks, 0xff, flen * sizeof(*masks));
932
933 for (pc = 0; pc < flen; pc++) {
934 memvalid &= masks[pc];
935
936 switch (filter[pc].code) {
937 case BPF_ST:
938 case BPF_STX:
939 memvalid |= (1 << filter[pc].k);
940 break;
941 case BPF_LD | BPF_MEM:
942 case BPF_LDX | BPF_MEM:
943 if (!(memvalid & (1 << filter[pc].k))) {
944 ret = -EINVAL;
945 goto error;
946 }
947 break;
948 case BPF_JMP | BPF_JA:
949 /* A jump must set masks on target */
950 masks[pc + 1 + filter[pc].k] &= memvalid;
951 memvalid = ~0;
952 break;
953 case BPF_JMP | BPF_JEQ | BPF_K:
954 case BPF_JMP | BPF_JEQ | BPF_X:
955 case BPF_JMP | BPF_JGE | BPF_K:
956 case BPF_JMP | BPF_JGE | BPF_X:
957 case BPF_JMP | BPF_JGT | BPF_K:
958 case BPF_JMP | BPF_JGT | BPF_X:
959 case BPF_JMP | BPF_JSET | BPF_K:
960 case BPF_JMP | BPF_JSET | BPF_X:
961 /* A jump must set masks on targets */
962 masks[pc + 1 + filter[pc].jt] &= memvalid;
963 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 memvalid = ~0;
965 break;
966 }
967 }
968 error:
969 kfree(masks);
970 return ret;
971 }
972
chk_code_allowed(u16 code_to_probe)973 static bool chk_code_allowed(u16 code_to_probe)
974 {
975 static const bool codes[] = {
976 /* 32 bit ALU operations */
977 [BPF_ALU | BPF_ADD | BPF_K] = true,
978 [BPF_ALU | BPF_ADD | BPF_X] = true,
979 [BPF_ALU | BPF_SUB | BPF_K] = true,
980 [BPF_ALU | BPF_SUB | BPF_X] = true,
981 [BPF_ALU | BPF_MUL | BPF_K] = true,
982 [BPF_ALU | BPF_MUL | BPF_X] = true,
983 [BPF_ALU | BPF_DIV | BPF_K] = true,
984 [BPF_ALU | BPF_DIV | BPF_X] = true,
985 [BPF_ALU | BPF_MOD | BPF_K] = true,
986 [BPF_ALU | BPF_MOD | BPF_X] = true,
987 [BPF_ALU | BPF_AND | BPF_K] = true,
988 [BPF_ALU | BPF_AND | BPF_X] = true,
989 [BPF_ALU | BPF_OR | BPF_K] = true,
990 [BPF_ALU | BPF_OR | BPF_X] = true,
991 [BPF_ALU | BPF_XOR | BPF_K] = true,
992 [BPF_ALU | BPF_XOR | BPF_X] = true,
993 [BPF_ALU | BPF_LSH | BPF_K] = true,
994 [BPF_ALU | BPF_LSH | BPF_X] = true,
995 [BPF_ALU | BPF_RSH | BPF_K] = true,
996 [BPF_ALU | BPF_RSH | BPF_X] = true,
997 [BPF_ALU | BPF_NEG] = true,
998 /* Load instructions */
999 [BPF_LD | BPF_W | BPF_ABS] = true,
1000 [BPF_LD | BPF_H | BPF_ABS] = true,
1001 [BPF_LD | BPF_B | BPF_ABS] = true,
1002 [BPF_LD | BPF_W | BPF_LEN] = true,
1003 [BPF_LD | BPF_W | BPF_IND] = true,
1004 [BPF_LD | BPF_H | BPF_IND] = true,
1005 [BPF_LD | BPF_B | BPF_IND] = true,
1006 [BPF_LD | BPF_IMM] = true,
1007 [BPF_LD | BPF_MEM] = true,
1008 [BPF_LDX | BPF_W | BPF_LEN] = true,
1009 [BPF_LDX | BPF_B | BPF_MSH] = true,
1010 [BPF_LDX | BPF_IMM] = true,
1011 [BPF_LDX | BPF_MEM] = true,
1012 /* Store instructions */
1013 [BPF_ST] = true,
1014 [BPF_STX] = true,
1015 /* Misc instructions */
1016 [BPF_MISC | BPF_TAX] = true,
1017 [BPF_MISC | BPF_TXA] = true,
1018 /* Return instructions */
1019 [BPF_RET | BPF_K] = true,
1020 [BPF_RET | BPF_A] = true,
1021 /* Jump instructions */
1022 [BPF_JMP | BPF_JA] = true,
1023 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1024 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1025 [BPF_JMP | BPF_JGE | BPF_K] = true,
1026 [BPF_JMP | BPF_JGE | BPF_X] = true,
1027 [BPF_JMP | BPF_JGT | BPF_K] = true,
1028 [BPF_JMP | BPF_JGT | BPF_X] = true,
1029 [BPF_JMP | BPF_JSET | BPF_K] = true,
1030 [BPF_JMP | BPF_JSET | BPF_X] = true,
1031 };
1032
1033 if (code_to_probe >= ARRAY_SIZE(codes))
1034 return false;
1035
1036 return codes[code_to_probe];
1037 }
1038
bpf_check_basics_ok(const struct sock_filter * filter,unsigned int flen)1039 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1040 unsigned int flen)
1041 {
1042 if (filter == NULL)
1043 return false;
1044 if (flen == 0 || flen > BPF_MAXINSNS)
1045 return false;
1046
1047 return true;
1048 }
1049
1050 /**
1051 * bpf_check_classic - verify socket filter code
1052 * @filter: filter to verify
1053 * @flen: length of filter
1054 *
1055 * Check the user's filter code. If we let some ugly
1056 * filter code slip through kaboom! The filter must contain
1057 * no references or jumps that are out of range, no illegal
1058 * instructions, and must end with a RET instruction.
1059 *
1060 * All jumps are forward as they are not signed.
1061 *
1062 * Returns 0 if the rule set is legal or -EINVAL if not.
1063 */
bpf_check_classic(const struct sock_filter * filter,unsigned int flen)1064 static int bpf_check_classic(const struct sock_filter *filter,
1065 unsigned int flen)
1066 {
1067 bool anc_found;
1068 int pc;
1069
1070 /* Check the filter code now */
1071 for (pc = 0; pc < flen; pc++) {
1072 const struct sock_filter *ftest = &filter[pc];
1073
1074 /* May we actually operate on this code? */
1075 if (!chk_code_allowed(ftest->code))
1076 return -EINVAL;
1077
1078 /* Some instructions need special checks */
1079 switch (ftest->code) {
1080 case BPF_ALU | BPF_DIV | BPF_K:
1081 case BPF_ALU | BPF_MOD | BPF_K:
1082 /* Check for division by zero */
1083 if (ftest->k == 0)
1084 return -EINVAL;
1085 break;
1086 case BPF_ALU | BPF_LSH | BPF_K:
1087 case BPF_ALU | BPF_RSH | BPF_K:
1088 if (ftest->k >= 32)
1089 return -EINVAL;
1090 break;
1091 case BPF_LD | BPF_MEM:
1092 case BPF_LDX | BPF_MEM:
1093 case BPF_ST:
1094 case BPF_STX:
1095 /* Check for invalid memory addresses */
1096 if (ftest->k >= BPF_MEMWORDS)
1097 return -EINVAL;
1098 break;
1099 case BPF_JMP | BPF_JA:
1100 /* Note, the large ftest->k might cause loops.
1101 * Compare this with conditional jumps below,
1102 * where offsets are limited. --ANK (981016)
1103 */
1104 if (ftest->k >= (unsigned int)(flen - pc - 1))
1105 return -EINVAL;
1106 break;
1107 case BPF_JMP | BPF_JEQ | BPF_K:
1108 case BPF_JMP | BPF_JEQ | BPF_X:
1109 case BPF_JMP | BPF_JGE | BPF_K:
1110 case BPF_JMP | BPF_JGE | BPF_X:
1111 case BPF_JMP | BPF_JGT | BPF_K:
1112 case BPF_JMP | BPF_JGT | BPF_X:
1113 case BPF_JMP | BPF_JSET | BPF_K:
1114 case BPF_JMP | BPF_JSET | BPF_X:
1115 /* Both conditionals must be safe */
1116 if (pc + ftest->jt + 1 >= flen ||
1117 pc + ftest->jf + 1 >= flen)
1118 return -EINVAL;
1119 break;
1120 case BPF_LD | BPF_W | BPF_ABS:
1121 case BPF_LD | BPF_H | BPF_ABS:
1122 case BPF_LD | BPF_B | BPF_ABS:
1123 anc_found = false;
1124 if (bpf_anc_helper(ftest) & BPF_ANC)
1125 anc_found = true;
1126 /* Ancillary operation unknown or unsupported */
1127 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1128 return -EINVAL;
1129 }
1130 }
1131
1132 /* Last instruction must be a RET code */
1133 switch (filter[flen - 1].code) {
1134 case BPF_RET | BPF_K:
1135 case BPF_RET | BPF_A:
1136 return check_load_and_stores(filter, flen);
1137 }
1138
1139 return -EINVAL;
1140 }
1141
bpf_prog_store_orig_filter(struct bpf_prog * fp,const struct sock_fprog * fprog)1142 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1143 const struct sock_fprog *fprog)
1144 {
1145 unsigned int fsize = bpf_classic_proglen(fprog);
1146 struct sock_fprog_kern *fkprog;
1147
1148 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1149 if (!fp->orig_prog)
1150 return -ENOMEM;
1151
1152 fkprog = fp->orig_prog;
1153 fkprog->len = fprog->len;
1154
1155 fkprog->filter = kmemdup(fp->insns, fsize,
1156 GFP_KERNEL | __GFP_NOWARN);
1157 if (!fkprog->filter) {
1158 kfree(fp->orig_prog);
1159 return -ENOMEM;
1160 }
1161
1162 return 0;
1163 }
1164
bpf_release_orig_filter(struct bpf_prog * fp)1165 static void bpf_release_orig_filter(struct bpf_prog *fp)
1166 {
1167 struct sock_fprog_kern *fprog = fp->orig_prog;
1168
1169 if (fprog) {
1170 kfree(fprog->filter);
1171 kfree(fprog);
1172 }
1173 }
1174
__bpf_prog_release(struct bpf_prog * prog)1175 static void __bpf_prog_release(struct bpf_prog *prog)
1176 {
1177 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1178 bpf_prog_put(prog);
1179 } else {
1180 bpf_release_orig_filter(prog);
1181 bpf_prog_free(prog);
1182 }
1183 }
1184
__sk_filter_release(struct sk_filter * fp)1185 static void __sk_filter_release(struct sk_filter *fp)
1186 {
1187 __bpf_prog_release(fp->prog);
1188 kfree(fp);
1189 }
1190
1191 /**
1192 * sk_filter_release_rcu - Release a socket filter by rcu_head
1193 * @rcu: rcu_head that contains the sk_filter to free
1194 */
sk_filter_release_rcu(struct rcu_head * rcu)1195 static void sk_filter_release_rcu(struct rcu_head *rcu)
1196 {
1197 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1198
1199 __sk_filter_release(fp);
1200 }
1201
1202 /**
1203 * sk_filter_release - release a socket filter
1204 * @fp: filter to remove
1205 *
1206 * Remove a filter from a socket and release its resources.
1207 */
sk_filter_release(struct sk_filter * fp)1208 static void sk_filter_release(struct sk_filter *fp)
1209 {
1210 if (refcount_dec_and_test(&fp->refcnt))
1211 call_rcu(&fp->rcu, sk_filter_release_rcu);
1212 }
1213
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1214 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1215 {
1216 u32 filter_size = bpf_prog_size(fp->prog->len);
1217
1218 atomic_sub(filter_size, &sk->sk_omem_alloc);
1219 sk_filter_release(fp);
1220 }
1221
1222 /* try to charge the socket memory if there is space available
1223 * return true on success
1224 */
__sk_filter_charge(struct sock * sk,struct sk_filter * fp)1225 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1226 {
1227 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1228 u32 filter_size = bpf_prog_size(fp->prog->len);
1229
1230 /* same check as in sock_kmalloc() */
1231 if (filter_size <= optmem_max &&
1232 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1233 atomic_add(filter_size, &sk->sk_omem_alloc);
1234 return true;
1235 }
1236 return false;
1237 }
1238
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1239 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1240 {
1241 if (!refcount_inc_not_zero(&fp->refcnt))
1242 return false;
1243
1244 if (!__sk_filter_charge(sk, fp)) {
1245 sk_filter_release(fp);
1246 return false;
1247 }
1248 return true;
1249 }
1250
bpf_migrate_filter(struct bpf_prog * fp)1251 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1252 {
1253 struct sock_filter *old_prog;
1254 struct bpf_prog *old_fp;
1255 int err, new_len, old_len = fp->len;
1256 bool seen_ld_abs = false;
1257
1258 /* We are free to overwrite insns et al right here as it won't be used at
1259 * this point in time anymore internally after the migration to the eBPF
1260 * instruction representation.
1261 */
1262 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1263 sizeof(struct bpf_insn));
1264
1265 /* Conversion cannot happen on overlapping memory areas,
1266 * so we need to keep the user BPF around until the 2nd
1267 * pass. At this time, the user BPF is stored in fp->insns.
1268 */
1269 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1270 GFP_KERNEL | __GFP_NOWARN);
1271 if (!old_prog) {
1272 err = -ENOMEM;
1273 goto out_err;
1274 }
1275
1276 /* 1st pass: calculate the new program length. */
1277 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1278 &seen_ld_abs);
1279 if (err)
1280 goto out_err_free;
1281
1282 /* Expand fp for appending the new filter representation. */
1283 old_fp = fp;
1284 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1285 if (!fp) {
1286 /* The old_fp is still around in case we couldn't
1287 * allocate new memory, so uncharge on that one.
1288 */
1289 fp = old_fp;
1290 err = -ENOMEM;
1291 goto out_err_free;
1292 }
1293
1294 fp->len = new_len;
1295
1296 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1297 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1298 &seen_ld_abs);
1299 if (err)
1300 /* 2nd bpf_convert_filter() can fail only if it fails
1301 * to allocate memory, remapping must succeed. Note,
1302 * that at this time old_fp has already been released
1303 * by krealloc().
1304 */
1305 goto out_err_free;
1306
1307 fp = bpf_prog_select_runtime(fp, &err);
1308 if (err)
1309 goto out_err_free;
1310
1311 kfree(old_prog);
1312 return fp;
1313
1314 out_err_free:
1315 kfree(old_prog);
1316 out_err:
1317 __bpf_prog_release(fp);
1318 return ERR_PTR(err);
1319 }
1320
bpf_prepare_filter(struct bpf_prog * fp,bpf_aux_classic_check_t trans)1321 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1322 bpf_aux_classic_check_t trans)
1323 {
1324 int err;
1325
1326 fp->bpf_func = NULL;
1327 fp->jited = 0;
1328
1329 err = bpf_check_classic(fp->insns, fp->len);
1330 if (err) {
1331 __bpf_prog_release(fp);
1332 return ERR_PTR(err);
1333 }
1334
1335 /* There might be additional checks and transformations
1336 * needed on classic filters, f.e. in case of seccomp.
1337 */
1338 if (trans) {
1339 err = trans(fp->insns, fp->len);
1340 if (err) {
1341 __bpf_prog_release(fp);
1342 return ERR_PTR(err);
1343 }
1344 }
1345
1346 /* Probe if we can JIT compile the filter and if so, do
1347 * the compilation of the filter.
1348 */
1349 bpf_jit_compile(fp);
1350
1351 /* JIT compiler couldn't process this filter, so do the eBPF translation
1352 * for the optimized interpreter.
1353 */
1354 if (!fp->jited)
1355 fp = bpf_migrate_filter(fp);
1356
1357 return fp;
1358 }
1359
1360 /**
1361 * bpf_prog_create - create an unattached filter
1362 * @pfp: the unattached filter that is created
1363 * @fprog: the filter program
1364 *
1365 * Create a filter independent of any socket. We first run some
1366 * sanity checks on it to make sure it does not explode on us later.
1367 * If an error occurs or there is insufficient memory for the filter
1368 * a negative errno code is returned. On success the return is zero.
1369 */
bpf_prog_create(struct bpf_prog ** pfp,struct sock_fprog_kern * fprog)1370 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1371 {
1372 unsigned int fsize = bpf_classic_proglen(fprog);
1373 struct bpf_prog *fp;
1374
1375 /* Make sure new filter is there and in the right amounts. */
1376 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1377 return -EINVAL;
1378
1379 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1380 if (!fp)
1381 return -ENOMEM;
1382
1383 memcpy(fp->insns, fprog->filter, fsize);
1384
1385 fp->len = fprog->len;
1386 /* Since unattached filters are not copied back to user
1387 * space through sk_get_filter(), we do not need to hold
1388 * a copy here, and can spare us the work.
1389 */
1390 fp->orig_prog = NULL;
1391
1392 /* bpf_prepare_filter() already takes care of freeing
1393 * memory in case something goes wrong.
1394 */
1395 fp = bpf_prepare_filter(fp, NULL);
1396 if (IS_ERR(fp))
1397 return PTR_ERR(fp);
1398
1399 *pfp = fp;
1400 return 0;
1401 }
1402 EXPORT_SYMBOL_GPL(bpf_prog_create);
1403
1404 /**
1405 * bpf_prog_create_from_user - create an unattached filter from user buffer
1406 * @pfp: the unattached filter that is created
1407 * @fprog: the filter program
1408 * @trans: post-classic verifier transformation handler
1409 * @save_orig: save classic BPF program
1410 *
1411 * This function effectively does the same as bpf_prog_create(), only
1412 * that it builds up its insns buffer from user space provided buffer.
1413 * It also allows for passing a bpf_aux_classic_check_t handler.
1414 */
bpf_prog_create_from_user(struct bpf_prog ** pfp,struct sock_fprog * fprog,bpf_aux_classic_check_t trans,bool save_orig)1415 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1416 bpf_aux_classic_check_t trans, bool save_orig)
1417 {
1418 unsigned int fsize = bpf_classic_proglen(fprog);
1419 struct bpf_prog *fp;
1420 int err;
1421
1422 /* Make sure new filter is there and in the right amounts. */
1423 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1424 return -EINVAL;
1425
1426 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1427 if (!fp)
1428 return -ENOMEM;
1429
1430 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1431 __bpf_prog_free(fp);
1432 return -EFAULT;
1433 }
1434
1435 fp->len = fprog->len;
1436 fp->orig_prog = NULL;
1437
1438 if (save_orig) {
1439 err = bpf_prog_store_orig_filter(fp, fprog);
1440 if (err) {
1441 __bpf_prog_free(fp);
1442 return -ENOMEM;
1443 }
1444 }
1445
1446 /* bpf_prepare_filter() already takes care of freeing
1447 * memory in case something goes wrong.
1448 */
1449 fp = bpf_prepare_filter(fp, trans);
1450 if (IS_ERR(fp))
1451 return PTR_ERR(fp);
1452
1453 *pfp = fp;
1454 return 0;
1455 }
1456 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1457
bpf_prog_destroy(struct bpf_prog * fp)1458 void bpf_prog_destroy(struct bpf_prog *fp)
1459 {
1460 __bpf_prog_release(fp);
1461 }
1462 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1463
__sk_attach_prog(struct bpf_prog * prog,struct sock * sk)1464 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1465 {
1466 struct sk_filter *fp, *old_fp;
1467
1468 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1469 if (!fp)
1470 return -ENOMEM;
1471
1472 fp->prog = prog;
1473
1474 if (!__sk_filter_charge(sk, fp)) {
1475 kfree(fp);
1476 return -ENOMEM;
1477 }
1478 refcount_set(&fp->refcnt, 1);
1479
1480 old_fp = rcu_dereference_protected(sk->sk_filter,
1481 lockdep_sock_is_held(sk));
1482 rcu_assign_pointer(sk->sk_filter, fp);
1483
1484 if (old_fp)
1485 sk_filter_uncharge(sk, old_fp);
1486
1487 return 0;
1488 }
1489
1490 static
__get_filter(struct sock_fprog * fprog,struct sock * sk)1491 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1492 {
1493 unsigned int fsize = bpf_classic_proglen(fprog);
1494 struct bpf_prog *prog;
1495 int err;
1496
1497 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1498 return ERR_PTR(-EPERM);
1499
1500 /* Make sure new filter is there and in the right amounts. */
1501 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1502 return ERR_PTR(-EINVAL);
1503
1504 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1505 if (!prog)
1506 return ERR_PTR(-ENOMEM);
1507
1508 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1509 __bpf_prog_free(prog);
1510 return ERR_PTR(-EFAULT);
1511 }
1512
1513 prog->len = fprog->len;
1514
1515 err = bpf_prog_store_orig_filter(prog, fprog);
1516 if (err) {
1517 __bpf_prog_free(prog);
1518 return ERR_PTR(-ENOMEM);
1519 }
1520
1521 /* bpf_prepare_filter() already takes care of freeing
1522 * memory in case something goes wrong.
1523 */
1524 return bpf_prepare_filter(prog, NULL);
1525 }
1526
1527 /**
1528 * sk_attach_filter - attach a socket filter
1529 * @fprog: the filter program
1530 * @sk: the socket to use
1531 *
1532 * Attach the user's filter code. We first run some sanity checks on
1533 * it to make sure it does not explode on us later. If an error
1534 * occurs or there is insufficient memory for the filter a negative
1535 * errno code is returned. On success the return is zero.
1536 */
sk_attach_filter(struct sock_fprog * fprog,struct sock * sk)1537 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1538 {
1539 struct bpf_prog *prog = __get_filter(fprog, sk);
1540 int err;
1541
1542 if (IS_ERR(prog))
1543 return PTR_ERR(prog);
1544
1545 err = __sk_attach_prog(prog, sk);
1546 if (err < 0) {
1547 __bpf_prog_release(prog);
1548 return err;
1549 }
1550
1551 return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(sk_attach_filter);
1554
sk_reuseport_attach_filter(struct sock_fprog * fprog,struct sock * sk)1555 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1556 {
1557 struct bpf_prog *prog = __get_filter(fprog, sk);
1558 int err, optmem_max;
1559
1560 if (IS_ERR(prog))
1561 return PTR_ERR(prog);
1562
1563 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1564 if (bpf_prog_size(prog->len) > optmem_max)
1565 err = -ENOMEM;
1566 else
1567 err = reuseport_attach_prog(sk, prog);
1568
1569 if (err)
1570 __bpf_prog_release(prog);
1571
1572 return err;
1573 }
1574
__get_bpf(u32 ufd,struct sock * sk)1575 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1576 {
1577 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1578 return ERR_PTR(-EPERM);
1579
1580 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1581 }
1582
sk_attach_bpf(u32 ufd,struct sock * sk)1583 int sk_attach_bpf(u32 ufd, struct sock *sk)
1584 {
1585 struct bpf_prog *prog = __get_bpf(ufd, sk);
1586 int err;
1587
1588 if (IS_ERR(prog))
1589 return PTR_ERR(prog);
1590
1591 err = __sk_attach_prog(prog, sk);
1592 if (err < 0) {
1593 bpf_prog_put(prog);
1594 return err;
1595 }
1596
1597 return 0;
1598 }
1599
sk_reuseport_attach_bpf(u32 ufd,struct sock * sk)1600 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1601 {
1602 struct bpf_prog *prog;
1603 int err, optmem_max;
1604
1605 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1606 return -EPERM;
1607
1608 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1609 if (PTR_ERR(prog) == -EINVAL)
1610 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1611 if (IS_ERR(prog))
1612 return PTR_ERR(prog);
1613
1614 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1615 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1616 * bpf prog (e.g. sockmap). It depends on the
1617 * limitation imposed by bpf_prog_load().
1618 * Hence, sysctl_optmem_max is not checked.
1619 */
1620 if ((sk->sk_type != SOCK_STREAM &&
1621 sk->sk_type != SOCK_DGRAM) ||
1622 (sk->sk_protocol != IPPROTO_UDP &&
1623 sk->sk_protocol != IPPROTO_TCP) ||
1624 (sk->sk_family != AF_INET &&
1625 sk->sk_family != AF_INET6)) {
1626 err = -ENOTSUPP;
1627 goto err_prog_put;
1628 }
1629 } else {
1630 /* BPF_PROG_TYPE_SOCKET_FILTER */
1631 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1632 if (bpf_prog_size(prog->len) > optmem_max) {
1633 err = -ENOMEM;
1634 goto err_prog_put;
1635 }
1636 }
1637
1638 err = reuseport_attach_prog(sk, prog);
1639 err_prog_put:
1640 if (err)
1641 bpf_prog_put(prog);
1642
1643 return err;
1644 }
1645
sk_reuseport_prog_free(struct bpf_prog * prog)1646 void sk_reuseport_prog_free(struct bpf_prog *prog)
1647 {
1648 if (!prog)
1649 return;
1650
1651 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1652 bpf_prog_put(prog);
1653 else
1654 bpf_prog_destroy(prog);
1655 }
1656
1657 struct bpf_scratchpad {
1658 union {
1659 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1660 u8 buff[MAX_BPF_STACK];
1661 };
1662 local_lock_t bh_lock;
1663 };
1664
1665 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp) = {
1666 .bh_lock = INIT_LOCAL_LOCK(bh_lock),
1667 };
1668
__bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1669 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1670 unsigned int write_len)
1671 {
1672 #ifdef CONFIG_DEBUG_NET
1673 /* Avoid a splat in pskb_may_pull_reason() */
1674 if (write_len > INT_MAX)
1675 return -EINVAL;
1676 #endif
1677 return skb_ensure_writable(skb, write_len);
1678 }
1679
bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1680 static inline int bpf_try_make_writable(struct sk_buff *skb,
1681 unsigned int write_len)
1682 {
1683 int err = __bpf_try_make_writable(skb, write_len);
1684
1685 bpf_compute_data_pointers(skb);
1686 return err;
1687 }
1688
bpf_try_make_head_writable(struct sk_buff * skb)1689 static int bpf_try_make_head_writable(struct sk_buff *skb)
1690 {
1691 return bpf_try_make_writable(skb, skb_headlen(skb));
1692 }
1693
bpf_push_mac_rcsum(struct sk_buff * skb)1694 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1695 {
1696 if (skb_at_tc_ingress(skb))
1697 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1698 }
1699
bpf_pull_mac_rcsum(struct sk_buff * skb)1700 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1701 {
1702 if (skb_at_tc_ingress(skb))
1703 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1704 }
1705
BPF_CALL_5(bpf_skb_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len,u64,flags)1706 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1707 const void *, from, u32, len, u64, flags)
1708 {
1709 void *ptr;
1710
1711 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1712 return -EINVAL;
1713 if (unlikely(offset > INT_MAX))
1714 return -EFAULT;
1715 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1716 return -EFAULT;
1717
1718 ptr = skb->data + offset;
1719 if (flags & BPF_F_RECOMPUTE_CSUM)
1720 __skb_postpull_rcsum(skb, ptr, len, offset);
1721
1722 memcpy(ptr, from, len);
1723
1724 if (flags & BPF_F_RECOMPUTE_CSUM)
1725 __skb_postpush_rcsum(skb, ptr, len, offset);
1726 if (flags & BPF_F_INVALIDATE_HASH)
1727 skb_clear_hash(skb);
1728
1729 return 0;
1730 }
1731
1732 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1733 .func = bpf_skb_store_bytes,
1734 .gpl_only = false,
1735 .ret_type = RET_INTEGER,
1736 .arg1_type = ARG_PTR_TO_CTX,
1737 .arg2_type = ARG_ANYTHING,
1738 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1739 .arg4_type = ARG_CONST_SIZE,
1740 .arg5_type = ARG_ANYTHING,
1741 };
1742
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1743 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1744 u32 len, u64 flags)
1745 {
1746 return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1747 }
1748
BPF_CALL_4(bpf_skb_load_bytes,const struct sk_buff *,skb,u32,offset,void *,to,u32,len)1749 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1750 void *, to, u32, len)
1751 {
1752 void *ptr;
1753
1754 if (unlikely(offset > INT_MAX))
1755 goto err_clear;
1756
1757 ptr = skb_header_pointer(skb, offset, len, to);
1758 if (unlikely(!ptr))
1759 goto err_clear;
1760 if (ptr != to)
1761 memcpy(to, ptr, len);
1762
1763 return 0;
1764 err_clear:
1765 memset(to, 0, len);
1766 return -EFAULT;
1767 }
1768
1769 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1770 .func = bpf_skb_load_bytes,
1771 .gpl_only = false,
1772 .ret_type = RET_INTEGER,
1773 .arg1_type = ARG_PTR_TO_CTX,
1774 .arg2_type = ARG_ANYTHING,
1775 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1776 .arg4_type = ARG_CONST_SIZE,
1777 };
1778
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1779 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1780 {
1781 return ____bpf_skb_load_bytes(skb, offset, to, len);
1782 }
1783
BPF_CALL_4(bpf_flow_dissector_load_bytes,const struct bpf_flow_dissector *,ctx,u32,offset,void *,to,u32,len)1784 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1785 const struct bpf_flow_dissector *, ctx, u32, offset,
1786 void *, to, u32, len)
1787 {
1788 void *ptr;
1789
1790 if (unlikely(offset > 0xffff))
1791 goto err_clear;
1792
1793 if (unlikely(!ctx->skb))
1794 goto err_clear;
1795
1796 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1797 if (unlikely(!ptr))
1798 goto err_clear;
1799 if (ptr != to)
1800 memcpy(to, ptr, len);
1801
1802 return 0;
1803 err_clear:
1804 memset(to, 0, len);
1805 return -EFAULT;
1806 }
1807
1808 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1809 .func = bpf_flow_dissector_load_bytes,
1810 .gpl_only = false,
1811 .ret_type = RET_INTEGER,
1812 .arg1_type = ARG_PTR_TO_CTX,
1813 .arg2_type = ARG_ANYTHING,
1814 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1815 .arg4_type = ARG_CONST_SIZE,
1816 };
1817
BPF_CALL_5(bpf_skb_load_bytes_relative,const struct sk_buff *,skb,u32,offset,void *,to,u32,len,u32,start_header)1818 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1819 u32, offset, void *, to, u32, len, u32, start_header)
1820 {
1821 u8 *end = skb_tail_pointer(skb);
1822 u8 *start, *ptr;
1823
1824 if (unlikely(offset > 0xffff))
1825 goto err_clear;
1826
1827 switch (start_header) {
1828 case BPF_HDR_START_MAC:
1829 if (unlikely(!skb_mac_header_was_set(skb)))
1830 goto err_clear;
1831 start = skb_mac_header(skb);
1832 break;
1833 case BPF_HDR_START_NET:
1834 start = skb_network_header(skb);
1835 break;
1836 default:
1837 goto err_clear;
1838 }
1839
1840 ptr = start + offset;
1841
1842 if (likely(ptr + len <= end)) {
1843 memcpy(to, ptr, len);
1844 return 0;
1845 }
1846
1847 err_clear:
1848 memset(to, 0, len);
1849 return -EFAULT;
1850 }
1851
1852 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1853 .func = bpf_skb_load_bytes_relative,
1854 .gpl_only = false,
1855 .ret_type = RET_INTEGER,
1856 .arg1_type = ARG_PTR_TO_CTX,
1857 .arg2_type = ARG_ANYTHING,
1858 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1859 .arg4_type = ARG_CONST_SIZE,
1860 .arg5_type = ARG_ANYTHING,
1861 };
1862
BPF_CALL_2(bpf_skb_pull_data,struct sk_buff *,skb,u32,len)1863 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1864 {
1865 /* Idea is the following: should the needed direct read/write
1866 * test fail during runtime, we can pull in more data and redo
1867 * again, since implicitly, we invalidate previous checks here.
1868 *
1869 * Or, since we know how much we need to make read/writeable,
1870 * this can be done once at the program beginning for direct
1871 * access case. By this we overcome limitations of only current
1872 * headroom being accessible.
1873 */
1874 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1875 }
1876
1877 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1878 .func = bpf_skb_pull_data,
1879 .gpl_only = false,
1880 .ret_type = RET_INTEGER,
1881 .arg1_type = ARG_PTR_TO_CTX,
1882 .arg2_type = ARG_ANYTHING,
1883 };
1884
BPF_CALL_1(bpf_sk_fullsock,struct sock *,sk)1885 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1886 {
1887 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1888 }
1889
1890 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1891 .func = bpf_sk_fullsock,
1892 .gpl_only = false,
1893 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1894 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1895 };
1896
sk_skb_try_make_writable(struct sk_buff * skb,unsigned int write_len)1897 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1898 unsigned int write_len)
1899 {
1900 return __bpf_try_make_writable(skb, write_len);
1901 }
1902
BPF_CALL_2(sk_skb_pull_data,struct sk_buff *,skb,u32,len)1903 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1904 {
1905 /* Idea is the following: should the needed direct read/write
1906 * test fail during runtime, we can pull in more data and redo
1907 * again, since implicitly, we invalidate previous checks here.
1908 *
1909 * Or, since we know how much we need to make read/writeable,
1910 * this can be done once at the program beginning for direct
1911 * access case. By this we overcome limitations of only current
1912 * headroom being accessible.
1913 */
1914 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1915 }
1916
1917 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1918 .func = sk_skb_pull_data,
1919 .gpl_only = false,
1920 .ret_type = RET_INTEGER,
1921 .arg1_type = ARG_PTR_TO_CTX,
1922 .arg2_type = ARG_ANYTHING,
1923 };
1924
BPF_CALL_5(bpf_l3_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1925 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1926 u64, from, u64, to, u64, flags)
1927 {
1928 __sum16 *ptr;
1929
1930 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1931 return -EINVAL;
1932 if (unlikely(offset > 0xffff || offset & 1))
1933 return -EFAULT;
1934 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1935 return -EFAULT;
1936
1937 ptr = (__sum16 *)(skb->data + offset);
1938 switch (flags & BPF_F_HDR_FIELD_MASK) {
1939 case 0:
1940 if (unlikely(from != 0))
1941 return -EINVAL;
1942
1943 csum_replace_by_diff(ptr, to);
1944 break;
1945 case 2:
1946 csum_replace2(ptr, from, to);
1947 break;
1948 case 4:
1949 csum_replace4(ptr, from, to);
1950 break;
1951 default:
1952 return -EINVAL;
1953 }
1954
1955 return 0;
1956 }
1957
1958 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1959 .func = bpf_l3_csum_replace,
1960 .gpl_only = false,
1961 .ret_type = RET_INTEGER,
1962 .arg1_type = ARG_PTR_TO_CTX,
1963 .arg2_type = ARG_ANYTHING,
1964 .arg3_type = ARG_ANYTHING,
1965 .arg4_type = ARG_ANYTHING,
1966 .arg5_type = ARG_ANYTHING,
1967 };
1968
BPF_CALL_5(bpf_l4_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1969 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1970 u64, from, u64, to, u64, flags)
1971 {
1972 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1973 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1974 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1975 __sum16 *ptr;
1976
1977 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1978 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1979 return -EINVAL;
1980 if (unlikely(offset > 0xffff || offset & 1))
1981 return -EFAULT;
1982 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1983 return -EFAULT;
1984
1985 ptr = (__sum16 *)(skb->data + offset);
1986 if (is_mmzero && !do_mforce && !*ptr)
1987 return 0;
1988
1989 switch (flags & BPF_F_HDR_FIELD_MASK) {
1990 case 0:
1991 if (unlikely(from != 0))
1992 return -EINVAL;
1993
1994 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1995 break;
1996 case 2:
1997 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1998 break;
1999 case 4:
2000 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
2001 break;
2002 default:
2003 return -EINVAL;
2004 }
2005
2006 if (is_mmzero && !*ptr)
2007 *ptr = CSUM_MANGLED_0;
2008 return 0;
2009 }
2010
2011 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
2012 .func = bpf_l4_csum_replace,
2013 .gpl_only = false,
2014 .ret_type = RET_INTEGER,
2015 .arg1_type = ARG_PTR_TO_CTX,
2016 .arg2_type = ARG_ANYTHING,
2017 .arg3_type = ARG_ANYTHING,
2018 .arg4_type = ARG_ANYTHING,
2019 .arg5_type = ARG_ANYTHING,
2020 };
2021
BPF_CALL_5(bpf_csum_diff,__be32 *,from,u32,from_size,__be32 *,to,u32,to_size,__wsum,seed)2022 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2023 __be32 *, to, u32, to_size, __wsum, seed)
2024 {
2025 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2026 u32 diff_size = from_size + to_size;
2027 int i, j = 0;
2028 __wsum ret;
2029
2030 /* This is quite flexible, some examples:
2031 *
2032 * from_size == 0, to_size > 0, seed := csum --> pushing data
2033 * from_size > 0, to_size == 0, seed := csum --> pulling data
2034 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2035 *
2036 * Even for diffing, from_size and to_size don't need to be equal.
2037 */
2038 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2039 diff_size > sizeof(sp->diff)))
2040 return -EINVAL;
2041
2042 local_lock_nested_bh(&bpf_sp.bh_lock);
2043 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2044 sp->diff[j] = ~from[i];
2045 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2046 sp->diff[j] = to[i];
2047
2048 ret = csum_partial(sp->diff, diff_size, seed);
2049 local_unlock_nested_bh(&bpf_sp.bh_lock);
2050 return ret;
2051 }
2052
2053 static const struct bpf_func_proto bpf_csum_diff_proto = {
2054 .func = bpf_csum_diff,
2055 .gpl_only = false,
2056 .pkt_access = true,
2057 .ret_type = RET_INTEGER,
2058 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2059 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2060 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2061 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2062 .arg5_type = ARG_ANYTHING,
2063 };
2064
BPF_CALL_2(bpf_csum_update,struct sk_buff *,skb,__wsum,csum)2065 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2066 {
2067 /* The interface is to be used in combination with bpf_csum_diff()
2068 * for direct packet writes. csum rotation for alignment as well
2069 * as emulating csum_sub() can be done from the eBPF program.
2070 */
2071 if (skb->ip_summed == CHECKSUM_COMPLETE)
2072 return (skb->csum = csum_add(skb->csum, csum));
2073
2074 return -ENOTSUPP;
2075 }
2076
2077 static const struct bpf_func_proto bpf_csum_update_proto = {
2078 .func = bpf_csum_update,
2079 .gpl_only = false,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2083 };
2084
BPF_CALL_2(bpf_csum_level,struct sk_buff *,skb,u64,level)2085 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2086 {
2087 /* The interface is to be used in combination with bpf_skb_adjust_room()
2088 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2089 * is passed as flags, for example.
2090 */
2091 switch (level) {
2092 case BPF_CSUM_LEVEL_INC:
2093 __skb_incr_checksum_unnecessary(skb);
2094 break;
2095 case BPF_CSUM_LEVEL_DEC:
2096 __skb_decr_checksum_unnecessary(skb);
2097 break;
2098 case BPF_CSUM_LEVEL_RESET:
2099 __skb_reset_checksum_unnecessary(skb);
2100 break;
2101 case BPF_CSUM_LEVEL_QUERY:
2102 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2103 skb->csum_level : -EACCES;
2104 default:
2105 return -EINVAL;
2106 }
2107
2108 return 0;
2109 }
2110
2111 static const struct bpf_func_proto bpf_csum_level_proto = {
2112 .func = bpf_csum_level,
2113 .gpl_only = false,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_PTR_TO_CTX,
2116 .arg2_type = ARG_ANYTHING,
2117 };
2118
__bpf_rx_skb(struct net_device * dev,struct sk_buff * skb)2119 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2120 {
2121 return dev_forward_skb_nomtu(dev, skb);
2122 }
2123
__bpf_rx_skb_no_mac(struct net_device * dev,struct sk_buff * skb)2124 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2125 struct sk_buff *skb)
2126 {
2127 int ret = ____dev_forward_skb(dev, skb, false);
2128
2129 if (likely(!ret)) {
2130 skb->dev = dev;
2131 ret = netif_rx(skb);
2132 }
2133
2134 return ret;
2135 }
2136
__bpf_tx_skb(struct net_device * dev,struct sk_buff * skb)2137 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2138 {
2139 int ret;
2140
2141 if (dev_xmit_recursion()) {
2142 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2143 kfree_skb(skb);
2144 return -ENETDOWN;
2145 }
2146
2147 skb->dev = dev;
2148 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2149 skb_clear_tstamp(skb);
2150
2151 dev_xmit_recursion_inc();
2152 ret = dev_queue_xmit(skb);
2153 dev_xmit_recursion_dec();
2154
2155 return ret;
2156 }
2157
__bpf_redirect_no_mac(struct sk_buff * skb,struct net_device * dev,u32 flags)2158 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2159 u32 flags)
2160 {
2161 unsigned int mlen = skb_network_offset(skb);
2162
2163 if (unlikely(skb->len <= mlen)) {
2164 kfree_skb(skb);
2165 return -ERANGE;
2166 }
2167
2168 if (mlen) {
2169 __skb_pull(skb, mlen);
2170
2171 /* At ingress, the mac header has already been pulled once.
2172 * At egress, skb_pospull_rcsum has to be done in case that
2173 * the skb is originated from ingress (i.e. a forwarded skb)
2174 * to ensure that rcsum starts at net header.
2175 */
2176 if (!skb_at_tc_ingress(skb))
2177 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2178 }
2179 skb_pop_mac_header(skb);
2180 skb_reset_mac_len(skb);
2181 return flags & BPF_F_INGRESS ?
2182 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2183 }
2184
__bpf_redirect_common(struct sk_buff * skb,struct net_device * dev,u32 flags)2185 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2186 u32 flags)
2187 {
2188 /* Verify that a link layer header is carried */
2189 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2190 kfree_skb(skb);
2191 return -ERANGE;
2192 }
2193
2194 bpf_push_mac_rcsum(skb);
2195 return flags & BPF_F_INGRESS ?
2196 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2197 }
2198
__bpf_redirect(struct sk_buff * skb,struct net_device * dev,u32 flags)2199 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2200 u32 flags)
2201 {
2202 if (dev_is_mac_header_xmit(dev))
2203 return __bpf_redirect_common(skb, dev, flags);
2204 else
2205 return __bpf_redirect_no_mac(skb, dev, flags);
2206 }
2207
2208 #if IS_ENABLED(CONFIG_IPV6)
bpf_out_neigh_v6(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2209 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2210 struct net_device *dev, struct bpf_nh_params *nh)
2211 {
2212 u32 hh_len = LL_RESERVED_SPACE(dev);
2213 const struct in6_addr *nexthop;
2214 struct dst_entry *dst = NULL;
2215 struct neighbour *neigh;
2216
2217 if (dev_xmit_recursion()) {
2218 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2219 goto out_drop;
2220 }
2221
2222 skb->dev = dev;
2223 skb_clear_tstamp(skb);
2224
2225 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2226 skb = skb_expand_head(skb, hh_len);
2227 if (!skb)
2228 return -ENOMEM;
2229 }
2230
2231 rcu_read_lock();
2232 if (!nh) {
2233 dst = skb_dst(skb);
2234 nexthop = rt6_nexthop(dst_rt6_info(dst),
2235 &ipv6_hdr(skb)->daddr);
2236 } else {
2237 nexthop = &nh->ipv6_nh;
2238 }
2239 neigh = ip_neigh_gw6(dev, nexthop);
2240 if (likely(!IS_ERR(neigh))) {
2241 int ret;
2242
2243 sock_confirm_neigh(skb, neigh);
2244 local_bh_disable();
2245 dev_xmit_recursion_inc();
2246 ret = neigh_output(neigh, skb, false);
2247 dev_xmit_recursion_dec();
2248 local_bh_enable();
2249 rcu_read_unlock();
2250 return ret;
2251 }
2252 rcu_read_unlock_bh();
2253 if (dst)
2254 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2255 out_drop:
2256 kfree_skb(skb);
2257 return -ENETDOWN;
2258 }
2259
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2260 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2261 struct bpf_nh_params *nh)
2262 {
2263 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2264 struct net *net = dev_net(dev);
2265 int err, ret = NET_XMIT_DROP;
2266
2267 if (!nh) {
2268 struct dst_entry *dst;
2269 struct flowi6 fl6 = {
2270 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2271 .flowi6_mark = skb->mark,
2272 .flowlabel = ip6_flowinfo(ip6h),
2273 .flowi6_oif = dev->ifindex,
2274 .flowi6_proto = ip6h->nexthdr,
2275 .daddr = ip6h->daddr,
2276 .saddr = ip6h->saddr,
2277 };
2278
2279 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2280 if (IS_ERR(dst))
2281 goto out_drop;
2282
2283 skb_dst_set(skb, dst);
2284 } else if (nh->nh_family != AF_INET6) {
2285 goto out_drop;
2286 }
2287
2288 err = bpf_out_neigh_v6(net, skb, dev, nh);
2289 if (unlikely(net_xmit_eval(err)))
2290 DEV_STATS_INC(dev, tx_errors);
2291 else
2292 ret = NET_XMIT_SUCCESS;
2293 goto out_xmit;
2294 out_drop:
2295 DEV_STATS_INC(dev, tx_errors);
2296 kfree_skb(skb);
2297 out_xmit:
2298 return ret;
2299 }
2300 #else
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2301 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2302 struct bpf_nh_params *nh)
2303 {
2304 kfree_skb(skb);
2305 return NET_XMIT_DROP;
2306 }
2307 #endif /* CONFIG_IPV6 */
2308
2309 #if IS_ENABLED(CONFIG_INET)
bpf_out_neigh_v4(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2310 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2311 struct net_device *dev, struct bpf_nh_params *nh)
2312 {
2313 u32 hh_len = LL_RESERVED_SPACE(dev);
2314 struct neighbour *neigh;
2315 bool is_v6gw = false;
2316
2317 if (dev_xmit_recursion()) {
2318 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2319 goto out_drop;
2320 }
2321
2322 skb->dev = dev;
2323 skb_clear_tstamp(skb);
2324
2325 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2326 skb = skb_expand_head(skb, hh_len);
2327 if (!skb)
2328 return -ENOMEM;
2329 }
2330
2331 rcu_read_lock();
2332 if (!nh) {
2333 struct rtable *rt = skb_rtable(skb);
2334
2335 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2336 } else if (nh->nh_family == AF_INET6) {
2337 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2338 is_v6gw = true;
2339 } else if (nh->nh_family == AF_INET) {
2340 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2341 } else {
2342 rcu_read_unlock();
2343 goto out_drop;
2344 }
2345
2346 if (likely(!IS_ERR(neigh))) {
2347 int ret;
2348
2349 sock_confirm_neigh(skb, neigh);
2350 local_bh_disable();
2351 dev_xmit_recursion_inc();
2352 ret = neigh_output(neigh, skb, is_v6gw);
2353 dev_xmit_recursion_dec();
2354 local_bh_enable();
2355 rcu_read_unlock();
2356 return ret;
2357 }
2358 rcu_read_unlock();
2359 out_drop:
2360 kfree_skb(skb);
2361 return -ENETDOWN;
2362 }
2363
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2364 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2365 struct bpf_nh_params *nh)
2366 {
2367 const struct iphdr *ip4h = ip_hdr(skb);
2368 struct net *net = dev_net(dev);
2369 int err, ret = NET_XMIT_DROP;
2370
2371 if (!nh) {
2372 struct flowi4 fl4 = {
2373 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2374 .flowi4_mark = skb->mark,
2375 .flowi4_tos = ip4h->tos & INET_DSCP_MASK,
2376 .flowi4_oif = dev->ifindex,
2377 .flowi4_proto = ip4h->protocol,
2378 .daddr = ip4h->daddr,
2379 .saddr = ip4h->saddr,
2380 };
2381 struct rtable *rt;
2382
2383 rt = ip_route_output_flow(net, &fl4, NULL);
2384 if (IS_ERR(rt))
2385 goto out_drop;
2386 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2387 ip_rt_put(rt);
2388 goto out_drop;
2389 }
2390
2391 skb_dst_set(skb, &rt->dst);
2392 }
2393
2394 err = bpf_out_neigh_v4(net, skb, dev, nh);
2395 if (unlikely(net_xmit_eval(err)))
2396 DEV_STATS_INC(dev, tx_errors);
2397 else
2398 ret = NET_XMIT_SUCCESS;
2399 goto out_xmit;
2400 out_drop:
2401 DEV_STATS_INC(dev, tx_errors);
2402 kfree_skb(skb);
2403 out_xmit:
2404 return ret;
2405 }
2406 #else
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2407 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2408 struct bpf_nh_params *nh)
2409 {
2410 kfree_skb(skb);
2411 return NET_XMIT_DROP;
2412 }
2413 #endif /* CONFIG_INET */
2414
__bpf_redirect_neigh(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2415 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2416 struct bpf_nh_params *nh)
2417 {
2418 struct ethhdr *ethh = eth_hdr(skb);
2419
2420 if (unlikely(skb->mac_header >= skb->network_header))
2421 goto out;
2422 bpf_push_mac_rcsum(skb);
2423 if (is_multicast_ether_addr(ethh->h_dest))
2424 goto out;
2425
2426 skb_pull(skb, sizeof(*ethh));
2427 skb_unset_mac_header(skb);
2428 skb_reset_network_header(skb);
2429
2430 if (skb->protocol == htons(ETH_P_IP))
2431 return __bpf_redirect_neigh_v4(skb, dev, nh);
2432 else if (skb->protocol == htons(ETH_P_IPV6))
2433 return __bpf_redirect_neigh_v6(skb, dev, nh);
2434 out:
2435 kfree_skb(skb);
2436 return -ENOTSUPP;
2437 }
2438
2439 /* Internal, non-exposed redirect flags. */
2440 enum {
2441 BPF_F_NEIGH = (1ULL << 1),
2442 BPF_F_PEER = (1ULL << 2),
2443 BPF_F_NEXTHOP = (1ULL << 3),
2444 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2445 };
2446
BPF_CALL_3(bpf_clone_redirect,struct sk_buff *,skb,u32,ifindex,u64,flags)2447 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2448 {
2449 struct net_device *dev;
2450 struct sk_buff *clone;
2451 int ret;
2452
2453 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2454 return -EINVAL;
2455
2456 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2457 if (unlikely(!dev))
2458 return -EINVAL;
2459
2460 clone = skb_clone(skb, GFP_ATOMIC);
2461 if (unlikely(!clone))
2462 return -ENOMEM;
2463
2464 /* For direct write, we need to keep the invariant that the skbs
2465 * we're dealing with need to be uncloned. Should uncloning fail
2466 * here, we need to free the just generated clone to unclone once
2467 * again.
2468 */
2469 ret = bpf_try_make_head_writable(skb);
2470 if (unlikely(ret)) {
2471 kfree_skb(clone);
2472 return -ENOMEM;
2473 }
2474
2475 return __bpf_redirect(clone, dev, flags);
2476 }
2477
2478 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2479 .func = bpf_clone_redirect,
2480 .gpl_only = false,
2481 .ret_type = RET_INTEGER,
2482 .arg1_type = ARG_PTR_TO_CTX,
2483 .arg2_type = ARG_ANYTHING,
2484 .arg3_type = ARG_ANYTHING,
2485 };
2486
skb_get_peer_dev(struct net_device * dev)2487 static struct net_device *skb_get_peer_dev(struct net_device *dev)
2488 {
2489 const struct net_device_ops *ops = dev->netdev_ops;
2490
2491 if (likely(ops->ndo_get_peer_dev))
2492 return INDIRECT_CALL_1(ops->ndo_get_peer_dev,
2493 netkit_peer_dev, dev);
2494 return NULL;
2495 }
2496
skb_do_redirect(struct sk_buff * skb)2497 int skb_do_redirect(struct sk_buff *skb)
2498 {
2499 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2500 struct net *net = dev_net(skb->dev);
2501 struct net_device *dev;
2502 u32 flags = ri->flags;
2503
2504 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2505 ri->tgt_index = 0;
2506 ri->flags = 0;
2507 if (unlikely(!dev))
2508 goto out_drop;
2509 if (flags & BPF_F_PEER) {
2510 if (unlikely(!skb_at_tc_ingress(skb)))
2511 goto out_drop;
2512 dev = skb_get_peer_dev(dev);
2513 if (unlikely(!dev ||
2514 !(dev->flags & IFF_UP) ||
2515 net_eq(net, dev_net(dev))))
2516 goto out_drop;
2517 skb->dev = dev;
2518 dev_sw_netstats_rx_add(dev, skb->len);
2519 return -EAGAIN;
2520 }
2521 return flags & BPF_F_NEIGH ?
2522 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2523 &ri->nh : NULL) :
2524 __bpf_redirect(skb, dev, flags);
2525 out_drop:
2526 kfree_skb(skb);
2527 return -EINVAL;
2528 }
2529
BPF_CALL_2(bpf_redirect,u32,ifindex,u64,flags)2530 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2531 {
2532 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2533
2534 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2535 return TC_ACT_SHOT;
2536
2537 ri->flags = flags;
2538 ri->tgt_index = ifindex;
2539
2540 return TC_ACT_REDIRECT;
2541 }
2542
2543 static const struct bpf_func_proto bpf_redirect_proto = {
2544 .func = bpf_redirect,
2545 .gpl_only = false,
2546 .ret_type = RET_INTEGER,
2547 .arg1_type = ARG_ANYTHING,
2548 .arg2_type = ARG_ANYTHING,
2549 };
2550
BPF_CALL_2(bpf_redirect_peer,u32,ifindex,u64,flags)2551 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2552 {
2553 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2554
2555 if (unlikely(flags))
2556 return TC_ACT_SHOT;
2557
2558 ri->flags = BPF_F_PEER;
2559 ri->tgt_index = ifindex;
2560
2561 return TC_ACT_REDIRECT;
2562 }
2563
2564 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2565 .func = bpf_redirect_peer,
2566 .gpl_only = false,
2567 .ret_type = RET_INTEGER,
2568 .arg1_type = ARG_ANYTHING,
2569 .arg2_type = ARG_ANYTHING,
2570 };
2571
BPF_CALL_4(bpf_redirect_neigh,u32,ifindex,struct bpf_redir_neigh *,params,int,plen,u64,flags)2572 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2573 int, plen, u64, flags)
2574 {
2575 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2576
2577 if (unlikely((plen && plen < sizeof(*params)) || flags))
2578 return TC_ACT_SHOT;
2579
2580 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2581 ri->tgt_index = ifindex;
2582
2583 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2584 if (plen)
2585 memcpy(&ri->nh, params, sizeof(ri->nh));
2586
2587 return TC_ACT_REDIRECT;
2588 }
2589
2590 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2591 .func = bpf_redirect_neigh,
2592 .gpl_only = false,
2593 .ret_type = RET_INTEGER,
2594 .arg1_type = ARG_ANYTHING,
2595 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2596 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2597 .arg4_type = ARG_ANYTHING,
2598 };
2599
BPF_CALL_2(bpf_msg_apply_bytes,struct sk_msg *,msg,u32,bytes)2600 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2601 {
2602 msg->apply_bytes = bytes;
2603 return 0;
2604 }
2605
2606 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2607 .func = bpf_msg_apply_bytes,
2608 .gpl_only = false,
2609 .ret_type = RET_INTEGER,
2610 .arg1_type = ARG_PTR_TO_CTX,
2611 .arg2_type = ARG_ANYTHING,
2612 };
2613
BPF_CALL_2(bpf_msg_cork_bytes,struct sk_msg *,msg,u32,bytes)2614 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2615 {
2616 msg->cork_bytes = bytes;
2617 return 0;
2618 }
2619
sk_msg_reset_curr(struct sk_msg * msg)2620 static void sk_msg_reset_curr(struct sk_msg *msg)
2621 {
2622 u32 i = msg->sg.start;
2623 u32 len = 0;
2624
2625 do {
2626 len += sk_msg_elem(msg, i)->length;
2627 sk_msg_iter_var_next(i);
2628 if (len >= msg->sg.size)
2629 break;
2630 } while (i != msg->sg.end);
2631
2632 msg->sg.curr = i;
2633 msg->sg.copybreak = 0;
2634 }
2635
2636 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2637 .func = bpf_msg_cork_bytes,
2638 .gpl_only = false,
2639 .ret_type = RET_INTEGER,
2640 .arg1_type = ARG_PTR_TO_CTX,
2641 .arg2_type = ARG_ANYTHING,
2642 };
2643
BPF_CALL_4(bpf_msg_pull_data,struct sk_msg *,msg,u32,start,u32,end,u64,flags)2644 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2645 u32, end, u64, flags)
2646 {
2647 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2648 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2649 struct scatterlist *sge;
2650 u8 *raw, *to, *from;
2651 struct page *page;
2652
2653 if (unlikely(flags || end <= start))
2654 return -EINVAL;
2655
2656 /* First find the starting scatterlist element */
2657 i = msg->sg.start;
2658 do {
2659 offset += len;
2660 len = sk_msg_elem(msg, i)->length;
2661 if (start < offset + len)
2662 break;
2663 sk_msg_iter_var_next(i);
2664 } while (i != msg->sg.end);
2665
2666 if (unlikely(start >= offset + len))
2667 return -EINVAL;
2668
2669 first_sge = i;
2670 /* The start may point into the sg element so we need to also
2671 * account for the headroom.
2672 */
2673 bytes_sg_total = start - offset + bytes;
2674 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2675 goto out;
2676
2677 /* At this point we need to linearize multiple scatterlist
2678 * elements or a single shared page. Either way we need to
2679 * copy into a linear buffer exclusively owned by BPF. Then
2680 * place the buffer in the scatterlist and fixup the original
2681 * entries by removing the entries now in the linear buffer
2682 * and shifting the remaining entries. For now we do not try
2683 * to copy partial entries to avoid complexity of running out
2684 * of sg_entry slots. The downside is reading a single byte
2685 * will copy the entire sg entry.
2686 */
2687 do {
2688 copy += sk_msg_elem(msg, i)->length;
2689 sk_msg_iter_var_next(i);
2690 if (bytes_sg_total <= copy)
2691 break;
2692 } while (i != msg->sg.end);
2693 last_sge = i;
2694
2695 if (unlikely(bytes_sg_total > copy))
2696 return -EINVAL;
2697
2698 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2699 get_order(copy));
2700 if (unlikely(!page))
2701 return -ENOMEM;
2702
2703 raw = page_address(page);
2704 i = first_sge;
2705 do {
2706 sge = sk_msg_elem(msg, i);
2707 from = sg_virt(sge);
2708 len = sge->length;
2709 to = raw + poffset;
2710
2711 memcpy(to, from, len);
2712 poffset += len;
2713 sge->length = 0;
2714 put_page(sg_page(sge));
2715
2716 sk_msg_iter_var_next(i);
2717 } while (i != last_sge);
2718
2719 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2720
2721 /* To repair sg ring we need to shift entries. If we only
2722 * had a single entry though we can just replace it and
2723 * be done. Otherwise walk the ring and shift the entries.
2724 */
2725 WARN_ON_ONCE(last_sge == first_sge);
2726 shift = last_sge > first_sge ?
2727 last_sge - first_sge - 1 :
2728 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2729 if (!shift)
2730 goto out;
2731
2732 i = first_sge;
2733 sk_msg_iter_var_next(i);
2734 do {
2735 u32 move_from;
2736
2737 if (i + shift >= NR_MSG_FRAG_IDS)
2738 move_from = i + shift - NR_MSG_FRAG_IDS;
2739 else
2740 move_from = i + shift;
2741 if (move_from == msg->sg.end)
2742 break;
2743
2744 msg->sg.data[i] = msg->sg.data[move_from];
2745 msg->sg.data[move_from].length = 0;
2746 msg->sg.data[move_from].page_link = 0;
2747 msg->sg.data[move_from].offset = 0;
2748 sk_msg_iter_var_next(i);
2749 } while (1);
2750
2751 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2752 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2753 msg->sg.end - shift;
2754 out:
2755 sk_msg_reset_curr(msg);
2756 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2757 msg->data_end = msg->data + bytes;
2758 return 0;
2759 }
2760
2761 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2762 .func = bpf_msg_pull_data,
2763 .gpl_only = false,
2764 .ret_type = RET_INTEGER,
2765 .arg1_type = ARG_PTR_TO_CTX,
2766 .arg2_type = ARG_ANYTHING,
2767 .arg3_type = ARG_ANYTHING,
2768 .arg4_type = ARG_ANYTHING,
2769 };
2770
BPF_CALL_4(bpf_msg_push_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2771 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2772 u32, len, u64, flags)
2773 {
2774 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2775 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2776 u8 *raw, *to, *from;
2777 struct page *page;
2778
2779 if (unlikely(flags))
2780 return -EINVAL;
2781
2782 if (unlikely(len == 0))
2783 return 0;
2784
2785 /* First find the starting scatterlist element */
2786 i = msg->sg.start;
2787 do {
2788 offset += l;
2789 l = sk_msg_elem(msg, i)->length;
2790
2791 if (start < offset + l)
2792 break;
2793 sk_msg_iter_var_next(i);
2794 } while (i != msg->sg.end);
2795
2796 if (start >= offset + l)
2797 return -EINVAL;
2798
2799 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2800
2801 /* If no space available will fallback to copy, we need at
2802 * least one scatterlist elem available to push data into
2803 * when start aligns to the beginning of an element or two
2804 * when it falls inside an element. We handle the start equals
2805 * offset case because its the common case for inserting a
2806 * header.
2807 */
2808 if (!space || (space == 1 && start != offset))
2809 copy = msg->sg.data[i].length;
2810
2811 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2812 get_order(copy + len));
2813 if (unlikely(!page))
2814 return -ENOMEM;
2815
2816 if (copy) {
2817 int front, back;
2818
2819 raw = page_address(page);
2820
2821 psge = sk_msg_elem(msg, i);
2822 front = start - offset;
2823 back = psge->length - front;
2824 from = sg_virt(psge);
2825
2826 if (front)
2827 memcpy(raw, from, front);
2828
2829 if (back) {
2830 from += front;
2831 to = raw + front + len;
2832
2833 memcpy(to, from, back);
2834 }
2835
2836 put_page(sg_page(psge));
2837 } else if (start - offset) {
2838 psge = sk_msg_elem(msg, i);
2839 rsge = sk_msg_elem_cpy(msg, i);
2840
2841 psge->length = start - offset;
2842 rsge.length -= psge->length;
2843 rsge.offset += start;
2844
2845 sk_msg_iter_var_next(i);
2846 sg_unmark_end(psge);
2847 sg_unmark_end(&rsge);
2848 sk_msg_iter_next(msg, end);
2849 }
2850
2851 /* Slot(s) to place newly allocated data */
2852 new = i;
2853
2854 /* Shift one or two slots as needed */
2855 if (!copy) {
2856 sge = sk_msg_elem_cpy(msg, i);
2857
2858 sk_msg_iter_var_next(i);
2859 sg_unmark_end(&sge);
2860 sk_msg_iter_next(msg, end);
2861
2862 nsge = sk_msg_elem_cpy(msg, i);
2863 if (rsge.length) {
2864 sk_msg_iter_var_next(i);
2865 nnsge = sk_msg_elem_cpy(msg, i);
2866 }
2867
2868 while (i != msg->sg.end) {
2869 msg->sg.data[i] = sge;
2870 sge = nsge;
2871 sk_msg_iter_var_next(i);
2872 if (rsge.length) {
2873 nsge = nnsge;
2874 nnsge = sk_msg_elem_cpy(msg, i);
2875 } else {
2876 nsge = sk_msg_elem_cpy(msg, i);
2877 }
2878 }
2879 }
2880
2881 /* Place newly allocated data buffer */
2882 sk_mem_charge(msg->sk, len);
2883 msg->sg.size += len;
2884 __clear_bit(new, msg->sg.copy);
2885 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2886 if (rsge.length) {
2887 get_page(sg_page(&rsge));
2888 sk_msg_iter_var_next(new);
2889 msg->sg.data[new] = rsge;
2890 }
2891
2892 sk_msg_reset_curr(msg);
2893 sk_msg_compute_data_pointers(msg);
2894 return 0;
2895 }
2896
2897 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2898 .func = bpf_msg_push_data,
2899 .gpl_only = false,
2900 .ret_type = RET_INTEGER,
2901 .arg1_type = ARG_PTR_TO_CTX,
2902 .arg2_type = ARG_ANYTHING,
2903 .arg3_type = ARG_ANYTHING,
2904 .arg4_type = ARG_ANYTHING,
2905 };
2906
sk_msg_shift_left(struct sk_msg * msg,int i)2907 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2908 {
2909 int prev;
2910
2911 do {
2912 prev = i;
2913 sk_msg_iter_var_next(i);
2914 msg->sg.data[prev] = msg->sg.data[i];
2915 } while (i != msg->sg.end);
2916
2917 sk_msg_iter_prev(msg, end);
2918 }
2919
sk_msg_shift_right(struct sk_msg * msg,int i)2920 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2921 {
2922 struct scatterlist tmp, sge;
2923
2924 sk_msg_iter_next(msg, end);
2925 sge = sk_msg_elem_cpy(msg, i);
2926 sk_msg_iter_var_next(i);
2927 tmp = sk_msg_elem_cpy(msg, i);
2928
2929 while (i != msg->sg.end) {
2930 msg->sg.data[i] = sge;
2931 sk_msg_iter_var_next(i);
2932 sge = tmp;
2933 tmp = sk_msg_elem_cpy(msg, i);
2934 }
2935 }
2936
BPF_CALL_4(bpf_msg_pop_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2937 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2938 u32, len, u64, flags)
2939 {
2940 u32 i = 0, l = 0, space, offset = 0;
2941 u64 last = start + len;
2942 int pop;
2943
2944 if (unlikely(flags))
2945 return -EINVAL;
2946
2947 /* First find the starting scatterlist element */
2948 i = msg->sg.start;
2949 do {
2950 offset += l;
2951 l = sk_msg_elem(msg, i)->length;
2952
2953 if (start < offset + l)
2954 break;
2955 sk_msg_iter_var_next(i);
2956 } while (i != msg->sg.end);
2957
2958 /* Bounds checks: start and pop must be inside message */
2959 if (start >= offset + l || last >= msg->sg.size)
2960 return -EINVAL;
2961
2962 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2963
2964 pop = len;
2965 /* --------------| offset
2966 * -| start |-------- len -------|
2967 *
2968 * |----- a ----|-------- pop -------|----- b ----|
2969 * |______________________________________________| length
2970 *
2971 *
2972 * a: region at front of scatter element to save
2973 * b: region at back of scatter element to save when length > A + pop
2974 * pop: region to pop from element, same as input 'pop' here will be
2975 * decremented below per iteration.
2976 *
2977 * Two top-level cases to handle when start != offset, first B is non
2978 * zero and second B is zero corresponding to when a pop includes more
2979 * than one element.
2980 *
2981 * Then if B is non-zero AND there is no space allocate space and
2982 * compact A, B regions into page. If there is space shift ring to
2983 * the right free'ing the next element in ring to place B, leaving
2984 * A untouched except to reduce length.
2985 */
2986 if (start != offset) {
2987 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2988 int a = start;
2989 int b = sge->length - pop - a;
2990
2991 sk_msg_iter_var_next(i);
2992
2993 if (pop < sge->length - a) {
2994 if (space) {
2995 sge->length = a;
2996 sk_msg_shift_right(msg, i);
2997 nsge = sk_msg_elem(msg, i);
2998 get_page(sg_page(sge));
2999 sg_set_page(nsge,
3000 sg_page(sge),
3001 b, sge->offset + pop + a);
3002 } else {
3003 struct page *page, *orig;
3004 u8 *to, *from;
3005
3006 page = alloc_pages(__GFP_NOWARN |
3007 __GFP_COMP | GFP_ATOMIC,
3008 get_order(a + b));
3009 if (unlikely(!page))
3010 return -ENOMEM;
3011
3012 sge->length = a;
3013 orig = sg_page(sge);
3014 from = sg_virt(sge);
3015 to = page_address(page);
3016 memcpy(to, from, a);
3017 memcpy(to + a, from + a + pop, b);
3018 sg_set_page(sge, page, a + b, 0);
3019 put_page(orig);
3020 }
3021 pop = 0;
3022 } else if (pop >= sge->length - a) {
3023 pop -= (sge->length - a);
3024 sge->length = a;
3025 }
3026 }
3027
3028 /* From above the current layout _must_ be as follows,
3029 *
3030 * -| offset
3031 * -| start
3032 *
3033 * |---- pop ---|---------------- b ------------|
3034 * |____________________________________________| length
3035 *
3036 * Offset and start of the current msg elem are equal because in the
3037 * previous case we handled offset != start and either consumed the
3038 * entire element and advanced to the next element OR pop == 0.
3039 *
3040 * Two cases to handle here are first pop is less than the length
3041 * leaving some remainder b above. Simply adjust the element's layout
3042 * in this case. Or pop >= length of the element so that b = 0. In this
3043 * case advance to next element decrementing pop.
3044 */
3045 while (pop) {
3046 struct scatterlist *sge = sk_msg_elem(msg, i);
3047
3048 if (pop < sge->length) {
3049 sge->length -= pop;
3050 sge->offset += pop;
3051 pop = 0;
3052 } else {
3053 pop -= sge->length;
3054 sk_msg_shift_left(msg, i);
3055 }
3056 sk_msg_iter_var_next(i);
3057 }
3058
3059 sk_mem_uncharge(msg->sk, len - pop);
3060 msg->sg.size -= (len - pop);
3061 sk_msg_reset_curr(msg);
3062 sk_msg_compute_data_pointers(msg);
3063 return 0;
3064 }
3065
3066 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3067 .func = bpf_msg_pop_data,
3068 .gpl_only = false,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3071 .arg2_type = ARG_ANYTHING,
3072 .arg3_type = ARG_ANYTHING,
3073 .arg4_type = ARG_ANYTHING,
3074 };
3075
3076 #ifdef CONFIG_CGROUP_NET_CLASSID
BPF_CALL_0(bpf_get_cgroup_classid_curr)3077 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3078 {
3079 return __task_get_classid(current);
3080 }
3081
3082 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3083 .func = bpf_get_cgroup_classid_curr,
3084 .gpl_only = false,
3085 .ret_type = RET_INTEGER,
3086 };
3087
BPF_CALL_1(bpf_skb_cgroup_classid,const struct sk_buff *,skb)3088 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3089 {
3090 struct sock *sk = skb_to_full_sk(skb);
3091
3092 if (!sk || !sk_fullsock(sk))
3093 return 0;
3094
3095 return sock_cgroup_classid(&sk->sk_cgrp_data);
3096 }
3097
3098 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3099 .func = bpf_skb_cgroup_classid,
3100 .gpl_only = false,
3101 .ret_type = RET_INTEGER,
3102 .arg1_type = ARG_PTR_TO_CTX,
3103 };
3104 #endif
3105
BPF_CALL_1(bpf_get_cgroup_classid,const struct sk_buff *,skb)3106 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3107 {
3108 return task_get_classid(skb);
3109 }
3110
3111 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3112 .func = bpf_get_cgroup_classid,
3113 .gpl_only = false,
3114 .ret_type = RET_INTEGER,
3115 .arg1_type = ARG_PTR_TO_CTX,
3116 };
3117
BPF_CALL_1(bpf_get_route_realm,const struct sk_buff *,skb)3118 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3119 {
3120 return dst_tclassid(skb);
3121 }
3122
3123 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3124 .func = bpf_get_route_realm,
3125 .gpl_only = false,
3126 .ret_type = RET_INTEGER,
3127 .arg1_type = ARG_PTR_TO_CTX,
3128 };
3129
BPF_CALL_1(bpf_get_hash_recalc,struct sk_buff *,skb)3130 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3131 {
3132 /* If skb_clear_hash() was called due to mangling, we can
3133 * trigger SW recalculation here. Later access to hash
3134 * can then use the inline skb->hash via context directly
3135 * instead of calling this helper again.
3136 */
3137 return skb_get_hash(skb);
3138 }
3139
3140 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3141 .func = bpf_get_hash_recalc,
3142 .gpl_only = false,
3143 .ret_type = RET_INTEGER,
3144 .arg1_type = ARG_PTR_TO_CTX,
3145 };
3146
BPF_CALL_1(bpf_set_hash_invalid,struct sk_buff *,skb)3147 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3148 {
3149 /* After all direct packet write, this can be used once for
3150 * triggering a lazy recalc on next skb_get_hash() invocation.
3151 */
3152 skb_clear_hash(skb);
3153 return 0;
3154 }
3155
3156 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3157 .func = bpf_set_hash_invalid,
3158 .gpl_only = false,
3159 .ret_type = RET_INTEGER,
3160 .arg1_type = ARG_PTR_TO_CTX,
3161 };
3162
BPF_CALL_2(bpf_set_hash,struct sk_buff *,skb,u32,hash)3163 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3164 {
3165 /* Set user specified hash as L4(+), so that it gets returned
3166 * on skb_get_hash() call unless BPF prog later on triggers a
3167 * skb_clear_hash().
3168 */
3169 __skb_set_sw_hash(skb, hash, true);
3170 return 0;
3171 }
3172
3173 static const struct bpf_func_proto bpf_set_hash_proto = {
3174 .func = bpf_set_hash,
3175 .gpl_only = false,
3176 .ret_type = RET_INTEGER,
3177 .arg1_type = ARG_PTR_TO_CTX,
3178 .arg2_type = ARG_ANYTHING,
3179 };
3180
BPF_CALL_3(bpf_skb_vlan_push,struct sk_buff *,skb,__be16,vlan_proto,u16,vlan_tci)3181 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3182 u16, vlan_tci)
3183 {
3184 int ret;
3185
3186 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3187 vlan_proto != htons(ETH_P_8021AD)))
3188 vlan_proto = htons(ETH_P_8021Q);
3189
3190 bpf_push_mac_rcsum(skb);
3191 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3192 bpf_pull_mac_rcsum(skb);
3193 skb_reset_mac_len(skb);
3194
3195 bpf_compute_data_pointers(skb);
3196 return ret;
3197 }
3198
3199 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3200 .func = bpf_skb_vlan_push,
3201 .gpl_only = false,
3202 .ret_type = RET_INTEGER,
3203 .arg1_type = ARG_PTR_TO_CTX,
3204 .arg2_type = ARG_ANYTHING,
3205 .arg3_type = ARG_ANYTHING,
3206 };
3207
BPF_CALL_1(bpf_skb_vlan_pop,struct sk_buff *,skb)3208 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3209 {
3210 int ret;
3211
3212 bpf_push_mac_rcsum(skb);
3213 ret = skb_vlan_pop(skb);
3214 bpf_pull_mac_rcsum(skb);
3215
3216 bpf_compute_data_pointers(skb);
3217 return ret;
3218 }
3219
3220 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3221 .func = bpf_skb_vlan_pop,
3222 .gpl_only = false,
3223 .ret_type = RET_INTEGER,
3224 .arg1_type = ARG_PTR_TO_CTX,
3225 };
3226
bpf_skb_generic_push(struct sk_buff * skb,u32 off,u32 len)3227 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3228 {
3229 /* Caller already did skb_cow() with len as headroom,
3230 * so no need to do it here.
3231 */
3232 skb_push(skb, len);
3233 memmove(skb->data, skb->data + len, off);
3234 memset(skb->data + off, 0, len);
3235
3236 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3237 * needed here as it does not change the skb->csum
3238 * result for checksum complete when summing over
3239 * zeroed blocks.
3240 */
3241 return 0;
3242 }
3243
bpf_skb_generic_pop(struct sk_buff * skb,u32 off,u32 len)3244 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3245 {
3246 void *old_data;
3247
3248 /* skb_ensure_writable() is not needed here, as we're
3249 * already working on an uncloned skb.
3250 */
3251 if (unlikely(!pskb_may_pull(skb, off + len)))
3252 return -ENOMEM;
3253
3254 old_data = skb->data;
3255 __skb_pull(skb, len);
3256 skb_postpull_rcsum(skb, old_data + off, len);
3257 memmove(skb->data, old_data, off);
3258
3259 return 0;
3260 }
3261
bpf_skb_net_hdr_push(struct sk_buff * skb,u32 off,u32 len)3262 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3263 {
3264 bool trans_same = skb->transport_header == skb->network_header;
3265 int ret;
3266
3267 /* There's no need for __skb_push()/__skb_pull() pair to
3268 * get to the start of the mac header as we're guaranteed
3269 * to always start from here under eBPF.
3270 */
3271 ret = bpf_skb_generic_push(skb, off, len);
3272 if (likely(!ret)) {
3273 skb->mac_header -= len;
3274 skb->network_header -= len;
3275 if (trans_same)
3276 skb->transport_header = skb->network_header;
3277 }
3278
3279 return ret;
3280 }
3281
bpf_skb_net_hdr_pop(struct sk_buff * skb,u32 off,u32 len)3282 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3283 {
3284 bool trans_same = skb->transport_header == skb->network_header;
3285 int ret;
3286
3287 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3288 ret = bpf_skb_generic_pop(skb, off, len);
3289 if (likely(!ret)) {
3290 skb->mac_header += len;
3291 skb->network_header += len;
3292 if (trans_same)
3293 skb->transport_header = skb->network_header;
3294 }
3295
3296 return ret;
3297 }
3298
bpf_skb_proto_4_to_6(struct sk_buff * skb)3299 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3300 {
3301 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3302 u32 off = skb_mac_header_len(skb);
3303 int ret;
3304
3305 ret = skb_cow(skb, len_diff);
3306 if (unlikely(ret < 0))
3307 return ret;
3308
3309 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3310 if (unlikely(ret < 0))
3311 return ret;
3312
3313 if (skb_is_gso(skb)) {
3314 struct skb_shared_info *shinfo = skb_shinfo(skb);
3315
3316 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3317 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3318 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3319 shinfo->gso_type |= SKB_GSO_TCPV6;
3320 }
3321 }
3322
3323 skb->protocol = htons(ETH_P_IPV6);
3324 skb_clear_hash(skb);
3325
3326 return 0;
3327 }
3328
bpf_skb_proto_6_to_4(struct sk_buff * skb)3329 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3330 {
3331 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3332 u32 off = skb_mac_header_len(skb);
3333 int ret;
3334
3335 ret = skb_unclone(skb, GFP_ATOMIC);
3336 if (unlikely(ret < 0))
3337 return ret;
3338
3339 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3340 if (unlikely(ret < 0))
3341 return ret;
3342
3343 if (skb_is_gso(skb)) {
3344 struct skb_shared_info *shinfo = skb_shinfo(skb);
3345
3346 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3347 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3348 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3349 shinfo->gso_type |= SKB_GSO_TCPV4;
3350 }
3351 }
3352
3353 skb->protocol = htons(ETH_P_IP);
3354 skb_clear_hash(skb);
3355
3356 return 0;
3357 }
3358
bpf_skb_proto_xlat(struct sk_buff * skb,__be16 to_proto)3359 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3360 {
3361 __be16 from_proto = skb->protocol;
3362
3363 if (from_proto == htons(ETH_P_IP) &&
3364 to_proto == htons(ETH_P_IPV6))
3365 return bpf_skb_proto_4_to_6(skb);
3366
3367 if (from_proto == htons(ETH_P_IPV6) &&
3368 to_proto == htons(ETH_P_IP))
3369 return bpf_skb_proto_6_to_4(skb);
3370
3371 return -ENOTSUPP;
3372 }
3373
BPF_CALL_3(bpf_skb_change_proto,struct sk_buff *,skb,__be16,proto,u64,flags)3374 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3375 u64, flags)
3376 {
3377 int ret;
3378
3379 if (unlikely(flags))
3380 return -EINVAL;
3381
3382 /* General idea is that this helper does the basic groundwork
3383 * needed for changing the protocol, and eBPF program fills the
3384 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3385 * and other helpers, rather than passing a raw buffer here.
3386 *
3387 * The rationale is to keep this minimal and without a need to
3388 * deal with raw packet data. F.e. even if we would pass buffers
3389 * here, the program still needs to call the bpf_lX_csum_replace()
3390 * helpers anyway. Plus, this way we keep also separation of
3391 * concerns, since f.e. bpf_skb_store_bytes() should only take
3392 * care of stores.
3393 *
3394 * Currently, additional options and extension header space are
3395 * not supported, but flags register is reserved so we can adapt
3396 * that. For offloads, we mark packet as dodgy, so that headers
3397 * need to be verified first.
3398 */
3399 ret = bpf_skb_proto_xlat(skb, proto);
3400 bpf_compute_data_pointers(skb);
3401 return ret;
3402 }
3403
3404 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3405 .func = bpf_skb_change_proto,
3406 .gpl_only = false,
3407 .ret_type = RET_INTEGER,
3408 .arg1_type = ARG_PTR_TO_CTX,
3409 .arg2_type = ARG_ANYTHING,
3410 .arg3_type = ARG_ANYTHING,
3411 };
3412
BPF_CALL_2(bpf_skb_change_type,struct sk_buff *,skb,u32,pkt_type)3413 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3414 {
3415 /* We only allow a restricted subset to be changed for now. */
3416 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3417 !skb_pkt_type_ok(pkt_type)))
3418 return -EINVAL;
3419
3420 skb->pkt_type = pkt_type;
3421 return 0;
3422 }
3423
3424 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3425 .func = bpf_skb_change_type,
3426 .gpl_only = false,
3427 .ret_type = RET_INTEGER,
3428 .arg1_type = ARG_PTR_TO_CTX,
3429 .arg2_type = ARG_ANYTHING,
3430 };
3431
bpf_skb_net_base_len(const struct sk_buff * skb)3432 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3433 {
3434 switch (skb->protocol) {
3435 case htons(ETH_P_IP):
3436 return sizeof(struct iphdr);
3437 case htons(ETH_P_IPV6):
3438 return sizeof(struct ipv6hdr);
3439 default:
3440 return ~0U;
3441 }
3442 }
3443
3444 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3445 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3446
3447 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3448 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3449
3450 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3451 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3452 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3453 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3454 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3455 BPF_F_ADJ_ROOM_ENCAP_L2( \
3456 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3457 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3458
bpf_skb_net_grow(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3459 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3460 u64 flags)
3461 {
3462 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3463 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3464 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3465 unsigned int gso_type = SKB_GSO_DODGY;
3466 int ret;
3467
3468 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3469 /* udp gso_size delineates datagrams, only allow if fixed */
3470 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3471 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3472 return -ENOTSUPP;
3473 }
3474
3475 ret = skb_cow_head(skb, len_diff);
3476 if (unlikely(ret < 0))
3477 return ret;
3478
3479 if (encap) {
3480 if (skb->protocol != htons(ETH_P_IP) &&
3481 skb->protocol != htons(ETH_P_IPV6))
3482 return -ENOTSUPP;
3483
3484 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3485 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3486 return -EINVAL;
3487
3488 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3489 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3490 return -EINVAL;
3491
3492 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3493 inner_mac_len < ETH_HLEN)
3494 return -EINVAL;
3495
3496 if (skb->encapsulation)
3497 return -EALREADY;
3498
3499 mac_len = skb->network_header - skb->mac_header;
3500 inner_net = skb->network_header;
3501 if (inner_mac_len > len_diff)
3502 return -EINVAL;
3503 inner_trans = skb->transport_header;
3504 }
3505
3506 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3507 if (unlikely(ret < 0))
3508 return ret;
3509
3510 if (encap) {
3511 skb->inner_mac_header = inner_net - inner_mac_len;
3512 skb->inner_network_header = inner_net;
3513 skb->inner_transport_header = inner_trans;
3514
3515 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3516 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3517 else
3518 skb_set_inner_protocol(skb, skb->protocol);
3519
3520 skb->encapsulation = 1;
3521 skb_set_network_header(skb, mac_len);
3522
3523 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3524 gso_type |= SKB_GSO_UDP_TUNNEL;
3525 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3526 gso_type |= SKB_GSO_GRE;
3527 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3528 gso_type |= SKB_GSO_IPXIP6;
3529 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3530 gso_type |= SKB_GSO_IPXIP4;
3531
3532 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3533 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3534 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3535 sizeof(struct ipv6hdr) :
3536 sizeof(struct iphdr);
3537
3538 skb_set_transport_header(skb, mac_len + nh_len);
3539 }
3540
3541 /* Match skb->protocol to new outer l3 protocol */
3542 if (skb->protocol == htons(ETH_P_IP) &&
3543 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3544 skb->protocol = htons(ETH_P_IPV6);
3545 else if (skb->protocol == htons(ETH_P_IPV6) &&
3546 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3547 skb->protocol = htons(ETH_P_IP);
3548 }
3549
3550 if (skb_is_gso(skb)) {
3551 struct skb_shared_info *shinfo = skb_shinfo(skb);
3552
3553 /* Header must be checked, and gso_segs recomputed. */
3554 shinfo->gso_type |= gso_type;
3555 shinfo->gso_segs = 0;
3556
3557 /* Due to header growth, MSS needs to be downgraded.
3558 * There is a BUG_ON() when segmenting the frag_list with
3559 * head_frag true, so linearize the skb after downgrading
3560 * the MSS.
3561 */
3562 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) {
3563 skb_decrease_gso_size(shinfo, len_diff);
3564 if (shinfo->frag_list)
3565 return skb_linearize(skb);
3566 }
3567 }
3568
3569 return 0;
3570 }
3571
bpf_skb_net_shrink(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3572 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3573 u64 flags)
3574 {
3575 int ret;
3576
3577 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3578 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3579 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3580 return -EINVAL;
3581
3582 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3583 /* udp gso_size delineates datagrams, only allow if fixed */
3584 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3585 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3586 return -ENOTSUPP;
3587 }
3588
3589 ret = skb_unclone(skb, GFP_ATOMIC);
3590 if (unlikely(ret < 0))
3591 return ret;
3592
3593 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3594 if (unlikely(ret < 0))
3595 return ret;
3596
3597 /* Match skb->protocol to new outer l3 protocol */
3598 if (skb->protocol == htons(ETH_P_IP) &&
3599 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3600 skb->protocol = htons(ETH_P_IPV6);
3601 else if (skb->protocol == htons(ETH_P_IPV6) &&
3602 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3603 skb->protocol = htons(ETH_P_IP);
3604
3605 if (skb_is_gso(skb)) {
3606 struct skb_shared_info *shinfo = skb_shinfo(skb);
3607
3608 /* Due to header shrink, MSS can be upgraded. */
3609 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3610 skb_increase_gso_size(shinfo, len_diff);
3611
3612 /* Header must be checked, and gso_segs recomputed. */
3613 shinfo->gso_type |= SKB_GSO_DODGY;
3614 shinfo->gso_segs = 0;
3615 }
3616
3617 return 0;
3618 }
3619
3620 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3621
BPF_CALL_4(sk_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3622 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3623 u32, mode, u64, flags)
3624 {
3625 u32 len_diff_abs = abs(len_diff);
3626 bool shrink = len_diff < 0;
3627 int ret = 0;
3628
3629 if (unlikely(flags || mode))
3630 return -EINVAL;
3631 if (unlikely(len_diff_abs > 0xfffU))
3632 return -EFAULT;
3633
3634 if (!shrink) {
3635 ret = skb_cow(skb, len_diff);
3636 if (unlikely(ret < 0))
3637 return ret;
3638 __skb_push(skb, len_diff_abs);
3639 memset(skb->data, 0, len_diff_abs);
3640 } else {
3641 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3642 return -ENOMEM;
3643 __skb_pull(skb, len_diff_abs);
3644 }
3645 if (tls_sw_has_ctx_rx(skb->sk)) {
3646 struct strp_msg *rxm = strp_msg(skb);
3647
3648 rxm->full_len += len_diff;
3649 }
3650 return ret;
3651 }
3652
3653 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3654 .func = sk_skb_adjust_room,
3655 .gpl_only = false,
3656 .ret_type = RET_INTEGER,
3657 .arg1_type = ARG_PTR_TO_CTX,
3658 .arg2_type = ARG_ANYTHING,
3659 .arg3_type = ARG_ANYTHING,
3660 .arg4_type = ARG_ANYTHING,
3661 };
3662
BPF_CALL_4(bpf_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3663 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3664 u32, mode, u64, flags)
3665 {
3666 u32 len_cur, len_diff_abs = abs(len_diff);
3667 u32 len_min = bpf_skb_net_base_len(skb);
3668 u32 len_max = BPF_SKB_MAX_LEN;
3669 __be16 proto = skb->protocol;
3670 bool shrink = len_diff < 0;
3671 u32 off;
3672 int ret;
3673
3674 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3675 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3676 return -EINVAL;
3677 if (unlikely(len_diff_abs > 0xfffU))
3678 return -EFAULT;
3679 if (unlikely(proto != htons(ETH_P_IP) &&
3680 proto != htons(ETH_P_IPV6)))
3681 return -ENOTSUPP;
3682
3683 off = skb_mac_header_len(skb);
3684 switch (mode) {
3685 case BPF_ADJ_ROOM_NET:
3686 off += bpf_skb_net_base_len(skb);
3687 break;
3688 case BPF_ADJ_ROOM_MAC:
3689 break;
3690 default:
3691 return -ENOTSUPP;
3692 }
3693
3694 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3695 if (!shrink)
3696 return -EINVAL;
3697
3698 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3699 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3700 len_min = sizeof(struct iphdr);
3701 break;
3702 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3703 len_min = sizeof(struct ipv6hdr);
3704 break;
3705 default:
3706 return -EINVAL;
3707 }
3708 }
3709
3710 len_cur = skb->len - skb_network_offset(skb);
3711 if ((shrink && (len_diff_abs >= len_cur ||
3712 len_cur - len_diff_abs < len_min)) ||
3713 (!shrink && (skb->len + len_diff_abs > len_max &&
3714 !skb_is_gso(skb))))
3715 return -ENOTSUPP;
3716
3717 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3718 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3719 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3720 __skb_reset_checksum_unnecessary(skb);
3721
3722 bpf_compute_data_pointers(skb);
3723 return ret;
3724 }
3725
3726 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3727 .func = bpf_skb_adjust_room,
3728 .gpl_only = false,
3729 .ret_type = RET_INTEGER,
3730 .arg1_type = ARG_PTR_TO_CTX,
3731 .arg2_type = ARG_ANYTHING,
3732 .arg3_type = ARG_ANYTHING,
3733 .arg4_type = ARG_ANYTHING,
3734 };
3735
__bpf_skb_min_len(const struct sk_buff * skb)3736 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3737 {
3738 u32 min_len = skb_network_offset(skb);
3739
3740 if (skb_transport_header_was_set(skb))
3741 min_len = skb_transport_offset(skb);
3742 if (skb->ip_summed == CHECKSUM_PARTIAL)
3743 min_len = skb_checksum_start_offset(skb) +
3744 skb->csum_offset + sizeof(__sum16);
3745 return min_len;
3746 }
3747
bpf_skb_grow_rcsum(struct sk_buff * skb,unsigned int new_len)3748 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3749 {
3750 unsigned int old_len = skb->len;
3751 int ret;
3752
3753 ret = __skb_grow_rcsum(skb, new_len);
3754 if (!ret)
3755 memset(skb->data + old_len, 0, new_len - old_len);
3756 return ret;
3757 }
3758
bpf_skb_trim_rcsum(struct sk_buff * skb,unsigned int new_len)3759 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3760 {
3761 return __skb_trim_rcsum(skb, new_len);
3762 }
3763
__bpf_skb_change_tail(struct sk_buff * skb,u32 new_len,u64 flags)3764 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3765 u64 flags)
3766 {
3767 u32 max_len = BPF_SKB_MAX_LEN;
3768 u32 min_len = __bpf_skb_min_len(skb);
3769 int ret;
3770
3771 if (unlikely(flags || new_len > max_len || new_len < min_len))
3772 return -EINVAL;
3773 if (skb->encapsulation)
3774 return -ENOTSUPP;
3775
3776 /* The basic idea of this helper is that it's performing the
3777 * needed work to either grow or trim an skb, and eBPF program
3778 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3779 * bpf_lX_csum_replace() and others rather than passing a raw
3780 * buffer here. This one is a slow path helper and intended
3781 * for replies with control messages.
3782 *
3783 * Like in bpf_skb_change_proto(), we want to keep this rather
3784 * minimal and without protocol specifics so that we are able
3785 * to separate concerns as in bpf_skb_store_bytes() should only
3786 * be the one responsible for writing buffers.
3787 *
3788 * It's really expected to be a slow path operation here for
3789 * control message replies, so we're implicitly linearizing,
3790 * uncloning and drop offloads from the skb by this.
3791 */
3792 ret = __bpf_try_make_writable(skb, skb->len);
3793 if (!ret) {
3794 if (new_len > skb->len)
3795 ret = bpf_skb_grow_rcsum(skb, new_len);
3796 else if (new_len < skb->len)
3797 ret = bpf_skb_trim_rcsum(skb, new_len);
3798 if (!ret && skb_is_gso(skb))
3799 skb_gso_reset(skb);
3800 }
3801 return ret;
3802 }
3803
BPF_CALL_3(bpf_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3804 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3805 u64, flags)
3806 {
3807 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3808
3809 bpf_compute_data_pointers(skb);
3810 return ret;
3811 }
3812
3813 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3814 .func = bpf_skb_change_tail,
3815 .gpl_only = false,
3816 .ret_type = RET_INTEGER,
3817 .arg1_type = ARG_PTR_TO_CTX,
3818 .arg2_type = ARG_ANYTHING,
3819 .arg3_type = ARG_ANYTHING,
3820 };
3821
BPF_CALL_3(sk_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3822 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3823 u64, flags)
3824 {
3825 return __bpf_skb_change_tail(skb, new_len, flags);
3826 }
3827
3828 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3829 .func = sk_skb_change_tail,
3830 .gpl_only = false,
3831 .ret_type = RET_INTEGER,
3832 .arg1_type = ARG_PTR_TO_CTX,
3833 .arg2_type = ARG_ANYTHING,
3834 .arg3_type = ARG_ANYTHING,
3835 };
3836
__bpf_skb_change_head(struct sk_buff * skb,u32 head_room,u64 flags)3837 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3838 u64 flags)
3839 {
3840 u32 max_len = BPF_SKB_MAX_LEN;
3841 u32 new_len = skb->len + head_room;
3842 int ret;
3843
3844 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3845 new_len < skb->len))
3846 return -EINVAL;
3847
3848 ret = skb_cow(skb, head_room);
3849 if (likely(!ret)) {
3850 /* Idea for this helper is that we currently only
3851 * allow to expand on mac header. This means that
3852 * skb->protocol network header, etc, stay as is.
3853 * Compared to bpf_skb_change_tail(), we're more
3854 * flexible due to not needing to linearize or
3855 * reset GSO. Intention for this helper is to be
3856 * used by an L3 skb that needs to push mac header
3857 * for redirection into L2 device.
3858 */
3859 __skb_push(skb, head_room);
3860 memset(skb->data, 0, head_room);
3861 skb_reset_mac_header(skb);
3862 skb_reset_mac_len(skb);
3863 }
3864
3865 return ret;
3866 }
3867
BPF_CALL_3(bpf_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3868 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3869 u64, flags)
3870 {
3871 int ret = __bpf_skb_change_head(skb, head_room, flags);
3872
3873 bpf_compute_data_pointers(skb);
3874 return ret;
3875 }
3876
3877 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3878 .func = bpf_skb_change_head,
3879 .gpl_only = false,
3880 .ret_type = RET_INTEGER,
3881 .arg1_type = ARG_PTR_TO_CTX,
3882 .arg2_type = ARG_ANYTHING,
3883 .arg3_type = ARG_ANYTHING,
3884 };
3885
BPF_CALL_3(sk_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3886 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3887 u64, flags)
3888 {
3889 return __bpf_skb_change_head(skb, head_room, flags);
3890 }
3891
3892 static const struct bpf_func_proto sk_skb_change_head_proto = {
3893 .func = sk_skb_change_head,
3894 .gpl_only = false,
3895 .ret_type = RET_INTEGER,
3896 .arg1_type = ARG_PTR_TO_CTX,
3897 .arg2_type = ARG_ANYTHING,
3898 .arg3_type = ARG_ANYTHING,
3899 };
3900
BPF_CALL_1(bpf_xdp_get_buff_len,struct xdp_buff *,xdp)3901 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3902 {
3903 return xdp_get_buff_len(xdp);
3904 }
3905
3906 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3907 .func = bpf_xdp_get_buff_len,
3908 .gpl_only = false,
3909 .ret_type = RET_INTEGER,
3910 .arg1_type = ARG_PTR_TO_CTX,
3911 };
3912
3913 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3914
3915 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3916 .func = bpf_xdp_get_buff_len,
3917 .gpl_only = false,
3918 .arg1_type = ARG_PTR_TO_BTF_ID,
3919 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3920 };
3921
xdp_get_metalen(const struct xdp_buff * xdp)3922 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3923 {
3924 return xdp_data_meta_unsupported(xdp) ? 0 :
3925 xdp->data - xdp->data_meta;
3926 }
3927
BPF_CALL_2(bpf_xdp_adjust_head,struct xdp_buff *,xdp,int,offset)3928 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3929 {
3930 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3931 unsigned long metalen = xdp_get_metalen(xdp);
3932 void *data_start = xdp_frame_end + metalen;
3933 void *data = xdp->data + offset;
3934
3935 if (unlikely(data < data_start ||
3936 data > xdp->data_end - ETH_HLEN))
3937 return -EINVAL;
3938
3939 if (metalen)
3940 memmove(xdp->data_meta + offset,
3941 xdp->data_meta, metalen);
3942 xdp->data_meta += offset;
3943 xdp->data = data;
3944
3945 return 0;
3946 }
3947
3948 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3949 .func = bpf_xdp_adjust_head,
3950 .gpl_only = false,
3951 .ret_type = RET_INTEGER,
3952 .arg1_type = ARG_PTR_TO_CTX,
3953 .arg2_type = ARG_ANYTHING,
3954 };
3955
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)3956 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3957 void *buf, unsigned long len, bool flush)
3958 {
3959 unsigned long ptr_len, ptr_off = 0;
3960 skb_frag_t *next_frag, *end_frag;
3961 struct skb_shared_info *sinfo;
3962 void *src, *dst;
3963 u8 *ptr_buf;
3964
3965 if (likely(xdp->data_end - xdp->data >= off + len)) {
3966 src = flush ? buf : xdp->data + off;
3967 dst = flush ? xdp->data + off : buf;
3968 memcpy(dst, src, len);
3969 return;
3970 }
3971
3972 sinfo = xdp_get_shared_info_from_buff(xdp);
3973 end_frag = &sinfo->frags[sinfo->nr_frags];
3974 next_frag = &sinfo->frags[0];
3975
3976 ptr_len = xdp->data_end - xdp->data;
3977 ptr_buf = xdp->data;
3978
3979 while (true) {
3980 if (off < ptr_off + ptr_len) {
3981 unsigned long copy_off = off - ptr_off;
3982 unsigned long copy_len = min(len, ptr_len - copy_off);
3983
3984 src = flush ? buf : ptr_buf + copy_off;
3985 dst = flush ? ptr_buf + copy_off : buf;
3986 memcpy(dst, src, copy_len);
3987
3988 off += copy_len;
3989 len -= copy_len;
3990 buf += copy_len;
3991 }
3992
3993 if (!len || next_frag == end_frag)
3994 break;
3995
3996 ptr_off += ptr_len;
3997 ptr_buf = skb_frag_address(next_frag);
3998 ptr_len = skb_frag_size(next_frag);
3999 next_frag++;
4000 }
4001 }
4002
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)4003 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
4004 {
4005 u32 size = xdp->data_end - xdp->data;
4006 struct skb_shared_info *sinfo;
4007 void *addr = xdp->data;
4008 int i;
4009
4010 if (unlikely(offset > 0xffff || len > 0xffff))
4011 return ERR_PTR(-EFAULT);
4012
4013 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
4014 return ERR_PTR(-EINVAL);
4015
4016 if (likely(offset < size)) /* linear area */
4017 goto out;
4018
4019 sinfo = xdp_get_shared_info_from_buff(xdp);
4020 offset -= size;
4021 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
4022 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
4023
4024 if (offset < frag_size) {
4025 addr = skb_frag_address(&sinfo->frags[i]);
4026 size = frag_size;
4027 break;
4028 }
4029 offset -= frag_size;
4030 }
4031 out:
4032 return offset + len <= size ? addr + offset : NULL;
4033 }
4034
BPF_CALL_4(bpf_xdp_load_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4035 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4036 void *, buf, u32, len)
4037 {
4038 void *ptr;
4039
4040 ptr = bpf_xdp_pointer(xdp, offset, len);
4041 if (IS_ERR(ptr))
4042 return PTR_ERR(ptr);
4043
4044 if (!ptr)
4045 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4046 else
4047 memcpy(buf, ptr, len);
4048
4049 return 0;
4050 }
4051
4052 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4053 .func = bpf_xdp_load_bytes,
4054 .gpl_only = false,
4055 .ret_type = RET_INTEGER,
4056 .arg1_type = ARG_PTR_TO_CTX,
4057 .arg2_type = ARG_ANYTHING,
4058 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4059 .arg4_type = ARG_CONST_SIZE,
4060 };
4061
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4062 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4063 {
4064 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4065 }
4066
BPF_CALL_4(bpf_xdp_store_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4067 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4068 void *, buf, u32, len)
4069 {
4070 void *ptr;
4071
4072 ptr = bpf_xdp_pointer(xdp, offset, len);
4073 if (IS_ERR(ptr))
4074 return PTR_ERR(ptr);
4075
4076 if (!ptr)
4077 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4078 else
4079 memcpy(ptr, buf, len);
4080
4081 return 0;
4082 }
4083
4084 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4085 .func = bpf_xdp_store_bytes,
4086 .gpl_only = false,
4087 .ret_type = RET_INTEGER,
4088 .arg1_type = ARG_PTR_TO_CTX,
4089 .arg2_type = ARG_ANYTHING,
4090 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4091 .arg4_type = ARG_CONST_SIZE,
4092 };
4093
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4094 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4095 {
4096 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4097 }
4098
bpf_xdp_frags_increase_tail(struct xdp_buff * xdp,int offset)4099 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4100 {
4101 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4102 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4103 struct xdp_rxq_info *rxq = xdp->rxq;
4104 unsigned int tailroom;
4105
4106 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4107 return -EOPNOTSUPP;
4108
4109 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4110 if (unlikely(offset > tailroom))
4111 return -EINVAL;
4112
4113 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4114 skb_frag_size_add(frag, offset);
4115 sinfo->xdp_frags_size += offset;
4116 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
4117 xsk_buff_get_tail(xdp)->data_end += offset;
4118
4119 return 0;
4120 }
4121
bpf_xdp_shrink_data_zc(struct xdp_buff * xdp,int shrink,struct xdp_mem_info * mem_info,bool release)4122 static void bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink,
4123 struct xdp_mem_info *mem_info, bool release)
4124 {
4125 struct xdp_buff *zc_frag = xsk_buff_get_tail(xdp);
4126
4127 if (release) {
4128 xsk_buff_del_tail(zc_frag);
4129 __xdp_return(NULL, mem_info, false, zc_frag);
4130 } else {
4131 zc_frag->data_end -= shrink;
4132 }
4133 }
4134
bpf_xdp_shrink_data(struct xdp_buff * xdp,skb_frag_t * frag,int shrink)4135 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag,
4136 int shrink)
4137 {
4138 struct xdp_mem_info *mem_info = &xdp->rxq->mem;
4139 bool release = skb_frag_size(frag) == shrink;
4140
4141 if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
4142 bpf_xdp_shrink_data_zc(xdp, shrink, mem_info, release);
4143 goto out;
4144 }
4145
4146 if (release) {
4147 struct page *page = skb_frag_page(frag);
4148
4149 __xdp_return(page_address(page), mem_info, false, NULL);
4150 }
4151
4152 out:
4153 return release;
4154 }
4155
bpf_xdp_frags_shrink_tail(struct xdp_buff * xdp,int offset)4156 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4157 {
4158 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4159 int i, n_frags_free = 0, len_free = 0;
4160
4161 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4162 return -EINVAL;
4163
4164 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4165 skb_frag_t *frag = &sinfo->frags[i];
4166 int shrink = min_t(int, offset, skb_frag_size(frag));
4167
4168 len_free += shrink;
4169 offset -= shrink;
4170 if (bpf_xdp_shrink_data(xdp, frag, shrink)) {
4171 n_frags_free++;
4172 } else {
4173 skb_frag_size_sub(frag, shrink);
4174 break;
4175 }
4176 }
4177 sinfo->nr_frags -= n_frags_free;
4178 sinfo->xdp_frags_size -= len_free;
4179
4180 if (unlikely(!sinfo->nr_frags)) {
4181 xdp_buff_clear_frags_flag(xdp);
4182 xdp->data_end -= offset;
4183 }
4184
4185 return 0;
4186 }
4187
BPF_CALL_2(bpf_xdp_adjust_tail,struct xdp_buff *,xdp,int,offset)4188 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4189 {
4190 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4191 void *data_end = xdp->data_end + offset;
4192
4193 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4194 if (offset < 0)
4195 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4196
4197 return bpf_xdp_frags_increase_tail(xdp, offset);
4198 }
4199
4200 /* Notice that xdp_data_hard_end have reserved some tailroom */
4201 if (unlikely(data_end > data_hard_end))
4202 return -EINVAL;
4203
4204 if (unlikely(data_end < xdp->data + ETH_HLEN))
4205 return -EINVAL;
4206
4207 /* Clear memory area on grow, can contain uninit kernel memory */
4208 if (offset > 0)
4209 memset(xdp->data_end, 0, offset);
4210
4211 xdp->data_end = data_end;
4212
4213 return 0;
4214 }
4215
4216 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4217 .func = bpf_xdp_adjust_tail,
4218 .gpl_only = false,
4219 .ret_type = RET_INTEGER,
4220 .arg1_type = ARG_PTR_TO_CTX,
4221 .arg2_type = ARG_ANYTHING,
4222 };
4223
BPF_CALL_2(bpf_xdp_adjust_meta,struct xdp_buff *,xdp,int,offset)4224 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4225 {
4226 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4227 void *meta = xdp->data_meta + offset;
4228 unsigned long metalen = xdp->data - meta;
4229
4230 if (xdp_data_meta_unsupported(xdp))
4231 return -ENOTSUPP;
4232 if (unlikely(meta < xdp_frame_end ||
4233 meta > xdp->data))
4234 return -EINVAL;
4235 if (unlikely(xdp_metalen_invalid(metalen)))
4236 return -EACCES;
4237
4238 xdp->data_meta = meta;
4239
4240 return 0;
4241 }
4242
4243 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4244 .func = bpf_xdp_adjust_meta,
4245 .gpl_only = false,
4246 .ret_type = RET_INTEGER,
4247 .arg1_type = ARG_PTR_TO_CTX,
4248 .arg2_type = ARG_ANYTHING,
4249 };
4250
4251 /**
4252 * DOC: xdp redirect
4253 *
4254 * XDP_REDIRECT works by a three-step process, implemented in the functions
4255 * below:
4256 *
4257 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4258 * of the redirect and store it (along with some other metadata) in a per-CPU
4259 * struct bpf_redirect_info.
4260 *
4261 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4262 * call xdp_do_redirect() which will use the information in struct
4263 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4264 * bulk queue structure.
4265 *
4266 * 3. Before exiting its NAPI poll loop, the driver will call
4267 * xdp_do_flush(), which will flush all the different bulk queues,
4268 * thus completing the redirect. Note that xdp_do_flush() must be
4269 * called before napi_complete_done() in the driver, as the
4270 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4271 * through to the xdp_do_flush() call for RCU protection of all
4272 * in-kernel data structures.
4273 */
4274 /*
4275 * Pointers to the map entries will be kept around for this whole sequence of
4276 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4277 * the core code; instead, the RCU protection relies on everything happening
4278 * inside a single NAPI poll sequence, which means it's between a pair of calls
4279 * to local_bh_disable()/local_bh_enable().
4280 *
4281 * The map entries are marked as __rcu and the map code makes sure to
4282 * dereference those pointers with rcu_dereference_check() in a way that works
4283 * for both sections that to hold an rcu_read_lock() and sections that are
4284 * called from NAPI without a separate rcu_read_lock(). The code below does not
4285 * use RCU annotations, but relies on those in the map code.
4286 */
xdp_do_flush(void)4287 void xdp_do_flush(void)
4288 {
4289 struct list_head *lh_map, *lh_dev, *lh_xsk;
4290
4291 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4292 if (lh_dev)
4293 __dev_flush(lh_dev);
4294 if (lh_map)
4295 __cpu_map_flush(lh_map);
4296 if (lh_xsk)
4297 __xsk_map_flush(lh_xsk);
4298 }
4299 EXPORT_SYMBOL_GPL(xdp_do_flush);
4300
4301 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
xdp_do_check_flushed(struct napi_struct * napi)4302 void xdp_do_check_flushed(struct napi_struct *napi)
4303 {
4304 struct list_head *lh_map, *lh_dev, *lh_xsk;
4305 bool missed = false;
4306
4307 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4308 if (lh_dev) {
4309 __dev_flush(lh_dev);
4310 missed = true;
4311 }
4312 if (lh_map) {
4313 __cpu_map_flush(lh_map);
4314 missed = true;
4315 }
4316 if (lh_xsk) {
4317 __xsk_map_flush(lh_xsk);
4318 missed = true;
4319 }
4320
4321 WARN_ONCE(missed, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
4322 napi->poll);
4323 }
4324 #endif
4325
4326 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4327 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4328
xdp_master_redirect(struct xdp_buff * xdp)4329 u32 xdp_master_redirect(struct xdp_buff *xdp)
4330 {
4331 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4332 struct net_device *master, *slave;
4333
4334 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4335 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4336 if (slave && slave != xdp->rxq->dev) {
4337 /* The target device is different from the receiving device, so
4338 * redirect it to the new device.
4339 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4340 * drivers to unmap the packet from their rx ring.
4341 */
4342 ri->tgt_index = slave->ifindex;
4343 ri->map_id = INT_MAX;
4344 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4345 return XDP_REDIRECT;
4346 }
4347 return XDP_TX;
4348 }
4349 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4350
__xdp_do_redirect_xsk(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4351 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4352 struct net_device *dev,
4353 struct xdp_buff *xdp,
4354 struct bpf_prog *xdp_prog)
4355 {
4356 enum bpf_map_type map_type = ri->map_type;
4357 void *fwd = ri->tgt_value;
4358 u32 map_id = ri->map_id;
4359 int err;
4360
4361 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4362 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4363
4364 err = __xsk_map_redirect(fwd, xdp);
4365 if (unlikely(err))
4366 goto err;
4367
4368 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4369 return 0;
4370 err:
4371 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4372 return err;
4373 }
4374
__xdp_do_redirect_frame(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4375 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4376 struct net_device *dev,
4377 struct xdp_frame *xdpf,
4378 struct bpf_prog *xdp_prog)
4379 {
4380 enum bpf_map_type map_type = ri->map_type;
4381 void *fwd = ri->tgt_value;
4382 u32 map_id = ri->map_id;
4383 u32 flags = ri->flags;
4384 struct bpf_map *map;
4385 int err;
4386
4387 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4388 ri->flags = 0;
4389 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4390
4391 if (unlikely(!xdpf)) {
4392 err = -EOVERFLOW;
4393 goto err;
4394 }
4395
4396 switch (map_type) {
4397 case BPF_MAP_TYPE_DEVMAP:
4398 fallthrough;
4399 case BPF_MAP_TYPE_DEVMAP_HASH:
4400 if (unlikely(flags & BPF_F_BROADCAST)) {
4401 map = READ_ONCE(ri->map);
4402
4403 /* The map pointer is cleared when the map is being torn
4404 * down by dev_map_free()
4405 */
4406 if (unlikely(!map)) {
4407 err = -ENOENT;
4408 break;
4409 }
4410
4411 WRITE_ONCE(ri->map, NULL);
4412 err = dev_map_enqueue_multi(xdpf, dev, map,
4413 flags & BPF_F_EXCLUDE_INGRESS);
4414 } else {
4415 err = dev_map_enqueue(fwd, xdpf, dev);
4416 }
4417 break;
4418 case BPF_MAP_TYPE_CPUMAP:
4419 err = cpu_map_enqueue(fwd, xdpf, dev);
4420 break;
4421 case BPF_MAP_TYPE_UNSPEC:
4422 if (map_id == INT_MAX) {
4423 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4424 if (unlikely(!fwd)) {
4425 err = -EINVAL;
4426 break;
4427 }
4428 err = dev_xdp_enqueue(fwd, xdpf, dev);
4429 break;
4430 }
4431 fallthrough;
4432 default:
4433 err = -EBADRQC;
4434 }
4435
4436 if (unlikely(err))
4437 goto err;
4438
4439 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4440 return 0;
4441 err:
4442 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4443 return err;
4444 }
4445
xdp_do_redirect(struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4446 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4447 struct bpf_prog *xdp_prog)
4448 {
4449 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4450 enum bpf_map_type map_type = ri->map_type;
4451
4452 if (map_type == BPF_MAP_TYPE_XSKMAP)
4453 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4454
4455 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4456 xdp_prog);
4457 }
4458 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4459
xdp_do_redirect_frame(struct net_device * dev,struct xdp_buff * xdp,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4460 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4461 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4462 {
4463 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4464 enum bpf_map_type map_type = ri->map_type;
4465
4466 if (map_type == BPF_MAP_TYPE_XSKMAP)
4467 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4468
4469 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4470 }
4471 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4472
xdp_do_generic_redirect_map(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog,void * fwd,enum bpf_map_type map_type,u32 map_id,u32 flags)4473 static int xdp_do_generic_redirect_map(struct net_device *dev,
4474 struct sk_buff *skb,
4475 struct xdp_buff *xdp,
4476 struct bpf_prog *xdp_prog, void *fwd,
4477 enum bpf_map_type map_type, u32 map_id,
4478 u32 flags)
4479 {
4480 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4481 struct bpf_map *map;
4482 int err;
4483
4484 switch (map_type) {
4485 case BPF_MAP_TYPE_DEVMAP:
4486 fallthrough;
4487 case BPF_MAP_TYPE_DEVMAP_HASH:
4488 if (unlikely(flags & BPF_F_BROADCAST)) {
4489 map = READ_ONCE(ri->map);
4490
4491 /* The map pointer is cleared when the map is being torn
4492 * down by dev_map_free()
4493 */
4494 if (unlikely(!map)) {
4495 err = -ENOENT;
4496 break;
4497 }
4498
4499 WRITE_ONCE(ri->map, NULL);
4500 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4501 flags & BPF_F_EXCLUDE_INGRESS);
4502 } else {
4503 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4504 }
4505 if (unlikely(err))
4506 goto err;
4507 break;
4508 case BPF_MAP_TYPE_XSKMAP:
4509 err = xsk_generic_rcv(fwd, xdp);
4510 if (err)
4511 goto err;
4512 consume_skb(skb);
4513 break;
4514 case BPF_MAP_TYPE_CPUMAP:
4515 err = cpu_map_generic_redirect(fwd, skb);
4516 if (unlikely(err))
4517 goto err;
4518 break;
4519 default:
4520 err = -EBADRQC;
4521 goto err;
4522 }
4523
4524 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4525 return 0;
4526 err:
4527 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4528 return err;
4529 }
4530
xdp_do_generic_redirect(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4531 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4532 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4533 {
4534 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4535 enum bpf_map_type map_type = ri->map_type;
4536 void *fwd = ri->tgt_value;
4537 u32 map_id = ri->map_id;
4538 u32 flags = ri->flags;
4539 int err;
4540
4541 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4542 ri->flags = 0;
4543 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4544
4545 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4546 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4547 if (unlikely(!fwd)) {
4548 err = -EINVAL;
4549 goto err;
4550 }
4551
4552 err = xdp_ok_fwd_dev(fwd, skb->len);
4553 if (unlikely(err))
4554 goto err;
4555
4556 skb->dev = fwd;
4557 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4558 generic_xdp_tx(skb, xdp_prog);
4559 return 0;
4560 }
4561
4562 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4563 err:
4564 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4565 return err;
4566 }
4567
BPF_CALL_2(bpf_xdp_redirect,u32,ifindex,u64,flags)4568 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4569 {
4570 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4571
4572 if (unlikely(flags))
4573 return XDP_ABORTED;
4574
4575 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4576 * by map_idr) is used for ifindex based XDP redirect.
4577 */
4578 ri->tgt_index = ifindex;
4579 ri->map_id = INT_MAX;
4580 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4581
4582 return XDP_REDIRECT;
4583 }
4584
4585 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4586 .func = bpf_xdp_redirect,
4587 .gpl_only = false,
4588 .ret_type = RET_INTEGER,
4589 .arg1_type = ARG_ANYTHING,
4590 .arg2_type = ARG_ANYTHING,
4591 };
4592
BPF_CALL_3(bpf_xdp_redirect_map,struct bpf_map *,map,u64,key,u64,flags)4593 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4594 u64, flags)
4595 {
4596 return map->ops->map_redirect(map, key, flags);
4597 }
4598
4599 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4600 .func = bpf_xdp_redirect_map,
4601 .gpl_only = false,
4602 .ret_type = RET_INTEGER,
4603 .arg1_type = ARG_CONST_MAP_PTR,
4604 .arg2_type = ARG_ANYTHING,
4605 .arg3_type = ARG_ANYTHING,
4606 };
4607
bpf_skb_copy(void * dst_buff,const void * skb,unsigned long off,unsigned long len)4608 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4609 unsigned long off, unsigned long len)
4610 {
4611 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4612
4613 if (unlikely(!ptr))
4614 return len;
4615 if (ptr != dst_buff)
4616 memcpy(dst_buff, ptr, len);
4617
4618 return 0;
4619 }
4620
BPF_CALL_5(bpf_skb_event_output,struct sk_buff *,skb,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4621 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4622 u64, flags, void *, meta, u64, meta_size)
4623 {
4624 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4625
4626 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4627 return -EINVAL;
4628 if (unlikely(!skb || skb_size > skb->len))
4629 return -EFAULT;
4630
4631 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4632 bpf_skb_copy);
4633 }
4634
4635 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4636 .func = bpf_skb_event_output,
4637 .gpl_only = true,
4638 .ret_type = RET_INTEGER,
4639 .arg1_type = ARG_PTR_TO_CTX,
4640 .arg2_type = ARG_CONST_MAP_PTR,
4641 .arg3_type = ARG_ANYTHING,
4642 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4643 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4644 };
4645
4646 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4647
4648 const struct bpf_func_proto bpf_skb_output_proto = {
4649 .func = bpf_skb_event_output,
4650 .gpl_only = true,
4651 .ret_type = RET_INTEGER,
4652 .arg1_type = ARG_PTR_TO_BTF_ID,
4653 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4654 .arg2_type = ARG_CONST_MAP_PTR,
4655 .arg3_type = ARG_ANYTHING,
4656 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4657 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4658 };
4659
bpf_tunnel_key_af(u64 flags)4660 static unsigned short bpf_tunnel_key_af(u64 flags)
4661 {
4662 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4663 }
4664
BPF_CALL_4(bpf_skb_get_tunnel_key,struct sk_buff *,skb,struct bpf_tunnel_key *,to,u32,size,u64,flags)4665 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4666 u32, size, u64, flags)
4667 {
4668 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4669 u8 compat[sizeof(struct bpf_tunnel_key)];
4670 void *to_orig = to;
4671 int err;
4672
4673 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4674 BPF_F_TUNINFO_FLAGS)))) {
4675 err = -EINVAL;
4676 goto err_clear;
4677 }
4678 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4679 err = -EPROTO;
4680 goto err_clear;
4681 }
4682 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4683 err = -EINVAL;
4684 switch (size) {
4685 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4686 case offsetof(struct bpf_tunnel_key, tunnel_label):
4687 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4688 goto set_compat;
4689 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4690 /* Fixup deprecated structure layouts here, so we have
4691 * a common path later on.
4692 */
4693 if (ip_tunnel_info_af(info) != AF_INET)
4694 goto err_clear;
4695 set_compat:
4696 to = (struct bpf_tunnel_key *)compat;
4697 break;
4698 default:
4699 goto err_clear;
4700 }
4701 }
4702
4703 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4704 to->tunnel_tos = info->key.tos;
4705 to->tunnel_ttl = info->key.ttl;
4706 if (flags & BPF_F_TUNINFO_FLAGS)
4707 to->tunnel_flags = ip_tunnel_flags_to_be16(info->key.tun_flags);
4708 else
4709 to->tunnel_ext = 0;
4710
4711 if (flags & BPF_F_TUNINFO_IPV6) {
4712 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4713 sizeof(to->remote_ipv6));
4714 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4715 sizeof(to->local_ipv6));
4716 to->tunnel_label = be32_to_cpu(info->key.label);
4717 } else {
4718 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4719 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4720 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4721 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4722 to->tunnel_label = 0;
4723 }
4724
4725 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4726 memcpy(to_orig, to, size);
4727
4728 return 0;
4729 err_clear:
4730 memset(to_orig, 0, size);
4731 return err;
4732 }
4733
4734 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4735 .func = bpf_skb_get_tunnel_key,
4736 .gpl_only = false,
4737 .ret_type = RET_INTEGER,
4738 .arg1_type = ARG_PTR_TO_CTX,
4739 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4740 .arg3_type = ARG_CONST_SIZE,
4741 .arg4_type = ARG_ANYTHING,
4742 };
4743
BPF_CALL_3(bpf_skb_get_tunnel_opt,struct sk_buff *,skb,u8 *,to,u32,size)4744 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4745 {
4746 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4747 int err;
4748
4749 if (unlikely(!info ||
4750 !ip_tunnel_is_options_present(info->key.tun_flags))) {
4751 err = -ENOENT;
4752 goto err_clear;
4753 }
4754 if (unlikely(size < info->options_len)) {
4755 err = -ENOMEM;
4756 goto err_clear;
4757 }
4758
4759 ip_tunnel_info_opts_get(to, info);
4760 if (size > info->options_len)
4761 memset(to + info->options_len, 0, size - info->options_len);
4762
4763 return info->options_len;
4764 err_clear:
4765 memset(to, 0, size);
4766 return err;
4767 }
4768
4769 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4770 .func = bpf_skb_get_tunnel_opt,
4771 .gpl_only = false,
4772 .ret_type = RET_INTEGER,
4773 .arg1_type = ARG_PTR_TO_CTX,
4774 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4775 .arg3_type = ARG_CONST_SIZE,
4776 };
4777
4778 static struct metadata_dst __percpu *md_dst;
4779
BPF_CALL_4(bpf_skb_set_tunnel_key,struct sk_buff *,skb,const struct bpf_tunnel_key *,from,u32,size,u64,flags)4780 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4781 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4782 {
4783 struct metadata_dst *md = this_cpu_ptr(md_dst);
4784 u8 compat[sizeof(struct bpf_tunnel_key)];
4785 struct ip_tunnel_info *info;
4786
4787 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4788 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4789 BPF_F_NO_TUNNEL_KEY)))
4790 return -EINVAL;
4791 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4792 switch (size) {
4793 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4794 case offsetof(struct bpf_tunnel_key, tunnel_label):
4795 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4796 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4797 /* Fixup deprecated structure layouts here, so we have
4798 * a common path later on.
4799 */
4800 memcpy(compat, from, size);
4801 memset(compat + size, 0, sizeof(compat) - size);
4802 from = (const struct bpf_tunnel_key *) compat;
4803 break;
4804 default:
4805 return -EINVAL;
4806 }
4807 }
4808 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4809 from->tunnel_ext))
4810 return -EINVAL;
4811
4812 skb_dst_drop(skb);
4813 dst_hold((struct dst_entry *) md);
4814 skb_dst_set(skb, (struct dst_entry *) md);
4815
4816 info = &md->u.tun_info;
4817 memset(info, 0, sizeof(*info));
4818 info->mode = IP_TUNNEL_INFO_TX;
4819
4820 __set_bit(IP_TUNNEL_NOCACHE_BIT, info->key.tun_flags);
4821 __assign_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, info->key.tun_flags,
4822 flags & BPF_F_DONT_FRAGMENT);
4823 __assign_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags,
4824 !(flags & BPF_F_ZERO_CSUM_TX));
4825 __assign_bit(IP_TUNNEL_SEQ_BIT, info->key.tun_flags,
4826 flags & BPF_F_SEQ_NUMBER);
4827 __assign_bit(IP_TUNNEL_KEY_BIT, info->key.tun_flags,
4828 !(flags & BPF_F_NO_TUNNEL_KEY));
4829
4830 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4831 info->key.tos = from->tunnel_tos;
4832 info->key.ttl = from->tunnel_ttl;
4833
4834 if (flags & BPF_F_TUNINFO_IPV6) {
4835 info->mode |= IP_TUNNEL_INFO_IPV6;
4836 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4837 sizeof(from->remote_ipv6));
4838 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4839 sizeof(from->local_ipv6));
4840 info->key.label = cpu_to_be32(from->tunnel_label) &
4841 IPV6_FLOWLABEL_MASK;
4842 } else {
4843 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4844 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4845 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4846 }
4847
4848 return 0;
4849 }
4850
4851 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4852 .func = bpf_skb_set_tunnel_key,
4853 .gpl_only = false,
4854 .ret_type = RET_INTEGER,
4855 .arg1_type = ARG_PTR_TO_CTX,
4856 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4857 .arg3_type = ARG_CONST_SIZE,
4858 .arg4_type = ARG_ANYTHING,
4859 };
4860
BPF_CALL_3(bpf_skb_set_tunnel_opt,struct sk_buff *,skb,const u8 *,from,u32,size)4861 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4862 const u8 *, from, u32, size)
4863 {
4864 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4865 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4866 IP_TUNNEL_DECLARE_FLAGS(present) = { };
4867
4868 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4869 return -EINVAL;
4870 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4871 return -ENOMEM;
4872
4873 ip_tunnel_set_options_present(present);
4874 ip_tunnel_info_opts_set(info, from, size, present);
4875
4876 return 0;
4877 }
4878
4879 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4880 .func = bpf_skb_set_tunnel_opt,
4881 .gpl_only = false,
4882 .ret_type = RET_INTEGER,
4883 .arg1_type = ARG_PTR_TO_CTX,
4884 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4885 .arg3_type = ARG_CONST_SIZE,
4886 };
4887
4888 static const struct bpf_func_proto *
bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)4889 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4890 {
4891 if (!md_dst) {
4892 struct metadata_dst __percpu *tmp;
4893
4894 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4895 METADATA_IP_TUNNEL,
4896 GFP_KERNEL);
4897 if (!tmp)
4898 return NULL;
4899 if (cmpxchg(&md_dst, NULL, tmp))
4900 metadata_dst_free_percpu(tmp);
4901 }
4902
4903 switch (which) {
4904 case BPF_FUNC_skb_set_tunnel_key:
4905 return &bpf_skb_set_tunnel_key_proto;
4906 case BPF_FUNC_skb_set_tunnel_opt:
4907 return &bpf_skb_set_tunnel_opt_proto;
4908 default:
4909 return NULL;
4910 }
4911 }
4912
BPF_CALL_3(bpf_skb_under_cgroup,struct sk_buff *,skb,struct bpf_map *,map,u32,idx)4913 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4914 u32, idx)
4915 {
4916 struct bpf_array *array = container_of(map, struct bpf_array, map);
4917 struct cgroup *cgrp;
4918 struct sock *sk;
4919
4920 sk = skb_to_full_sk(skb);
4921 if (!sk || !sk_fullsock(sk))
4922 return -ENOENT;
4923 if (unlikely(idx >= array->map.max_entries))
4924 return -E2BIG;
4925
4926 cgrp = READ_ONCE(array->ptrs[idx]);
4927 if (unlikely(!cgrp))
4928 return -EAGAIN;
4929
4930 return sk_under_cgroup_hierarchy(sk, cgrp);
4931 }
4932
4933 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4934 .func = bpf_skb_under_cgroup,
4935 .gpl_only = false,
4936 .ret_type = RET_INTEGER,
4937 .arg1_type = ARG_PTR_TO_CTX,
4938 .arg2_type = ARG_CONST_MAP_PTR,
4939 .arg3_type = ARG_ANYTHING,
4940 };
4941
4942 #ifdef CONFIG_SOCK_CGROUP_DATA
__bpf_sk_cgroup_id(struct sock * sk)4943 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4944 {
4945 struct cgroup *cgrp;
4946
4947 sk = sk_to_full_sk(sk);
4948 if (!sk || !sk_fullsock(sk))
4949 return 0;
4950
4951 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4952 return cgroup_id(cgrp);
4953 }
4954
BPF_CALL_1(bpf_skb_cgroup_id,const struct sk_buff *,skb)4955 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4956 {
4957 return __bpf_sk_cgroup_id(skb->sk);
4958 }
4959
4960 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4961 .func = bpf_skb_cgroup_id,
4962 .gpl_only = false,
4963 .ret_type = RET_INTEGER,
4964 .arg1_type = ARG_PTR_TO_CTX,
4965 };
4966
__bpf_sk_ancestor_cgroup_id(struct sock * sk,int ancestor_level)4967 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4968 int ancestor_level)
4969 {
4970 struct cgroup *ancestor;
4971 struct cgroup *cgrp;
4972
4973 sk = sk_to_full_sk(sk);
4974 if (!sk || !sk_fullsock(sk))
4975 return 0;
4976
4977 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4978 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4979 if (!ancestor)
4980 return 0;
4981
4982 return cgroup_id(ancestor);
4983 }
4984
BPF_CALL_2(bpf_skb_ancestor_cgroup_id,const struct sk_buff *,skb,int,ancestor_level)4985 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4986 ancestor_level)
4987 {
4988 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4989 }
4990
4991 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4992 .func = bpf_skb_ancestor_cgroup_id,
4993 .gpl_only = false,
4994 .ret_type = RET_INTEGER,
4995 .arg1_type = ARG_PTR_TO_CTX,
4996 .arg2_type = ARG_ANYTHING,
4997 };
4998
BPF_CALL_1(bpf_sk_cgroup_id,struct sock *,sk)4999 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
5000 {
5001 return __bpf_sk_cgroup_id(sk);
5002 }
5003
5004 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
5005 .func = bpf_sk_cgroup_id,
5006 .gpl_only = false,
5007 .ret_type = RET_INTEGER,
5008 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5009 };
5010
BPF_CALL_2(bpf_sk_ancestor_cgroup_id,struct sock *,sk,int,ancestor_level)5011 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
5012 {
5013 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
5014 }
5015
5016 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
5017 .func = bpf_sk_ancestor_cgroup_id,
5018 .gpl_only = false,
5019 .ret_type = RET_INTEGER,
5020 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5021 .arg2_type = ARG_ANYTHING,
5022 };
5023 #endif
5024
bpf_xdp_copy(void * dst,const void * ctx,unsigned long off,unsigned long len)5025 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
5026 unsigned long off, unsigned long len)
5027 {
5028 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
5029
5030 bpf_xdp_copy_buf(xdp, off, dst, len, false);
5031 return 0;
5032 }
5033
BPF_CALL_5(bpf_xdp_event_output,struct xdp_buff *,xdp,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)5034 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
5035 u64, flags, void *, meta, u64, meta_size)
5036 {
5037 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
5038
5039 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
5040 return -EINVAL;
5041
5042 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
5043 return -EFAULT;
5044
5045 return bpf_event_output(map, flags, meta, meta_size, xdp,
5046 xdp_size, bpf_xdp_copy);
5047 }
5048
5049 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
5050 .func = bpf_xdp_event_output,
5051 .gpl_only = true,
5052 .ret_type = RET_INTEGER,
5053 .arg1_type = ARG_PTR_TO_CTX,
5054 .arg2_type = ARG_CONST_MAP_PTR,
5055 .arg3_type = ARG_ANYTHING,
5056 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5057 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5058 };
5059
5060 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
5061
5062 const struct bpf_func_proto bpf_xdp_output_proto = {
5063 .func = bpf_xdp_event_output,
5064 .gpl_only = true,
5065 .ret_type = RET_INTEGER,
5066 .arg1_type = ARG_PTR_TO_BTF_ID,
5067 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
5068 .arg2_type = ARG_CONST_MAP_PTR,
5069 .arg3_type = ARG_ANYTHING,
5070 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5071 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5072 };
5073
BPF_CALL_1(bpf_get_socket_cookie,struct sk_buff *,skb)5074 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
5075 {
5076 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
5077 }
5078
5079 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
5080 .func = bpf_get_socket_cookie,
5081 .gpl_only = false,
5082 .ret_type = RET_INTEGER,
5083 .arg1_type = ARG_PTR_TO_CTX,
5084 };
5085
BPF_CALL_1(bpf_get_socket_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5086 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5087 {
5088 return __sock_gen_cookie(ctx->sk);
5089 }
5090
5091 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
5092 .func = bpf_get_socket_cookie_sock_addr,
5093 .gpl_only = false,
5094 .ret_type = RET_INTEGER,
5095 .arg1_type = ARG_PTR_TO_CTX,
5096 };
5097
BPF_CALL_1(bpf_get_socket_cookie_sock,struct sock *,ctx)5098 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
5099 {
5100 return __sock_gen_cookie(ctx);
5101 }
5102
5103 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5104 .func = bpf_get_socket_cookie_sock,
5105 .gpl_only = false,
5106 .ret_type = RET_INTEGER,
5107 .arg1_type = ARG_PTR_TO_CTX,
5108 };
5109
BPF_CALL_1(bpf_get_socket_ptr_cookie,struct sock *,sk)5110 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5111 {
5112 return sk ? sock_gen_cookie(sk) : 0;
5113 }
5114
5115 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5116 .func = bpf_get_socket_ptr_cookie,
5117 .gpl_only = false,
5118 .ret_type = RET_INTEGER,
5119 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5120 };
5121
BPF_CALL_1(bpf_get_socket_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5122 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5123 {
5124 return __sock_gen_cookie(ctx->sk);
5125 }
5126
5127 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5128 .func = bpf_get_socket_cookie_sock_ops,
5129 .gpl_only = false,
5130 .ret_type = RET_INTEGER,
5131 .arg1_type = ARG_PTR_TO_CTX,
5132 };
5133
__bpf_get_netns_cookie(struct sock * sk)5134 static u64 __bpf_get_netns_cookie(struct sock *sk)
5135 {
5136 const struct net *net = sk ? sock_net(sk) : &init_net;
5137
5138 return net->net_cookie;
5139 }
5140
BPF_CALL_1(bpf_get_netns_cookie_sock,struct sock *,ctx)5141 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5142 {
5143 return __bpf_get_netns_cookie(ctx);
5144 }
5145
5146 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5147 .func = bpf_get_netns_cookie_sock,
5148 .gpl_only = false,
5149 .ret_type = RET_INTEGER,
5150 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5151 };
5152
BPF_CALL_1(bpf_get_netns_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5153 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5154 {
5155 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5156 }
5157
5158 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5159 .func = bpf_get_netns_cookie_sock_addr,
5160 .gpl_only = false,
5161 .ret_type = RET_INTEGER,
5162 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5163 };
5164
BPF_CALL_1(bpf_get_netns_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5165 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5166 {
5167 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5168 }
5169
5170 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5171 .func = bpf_get_netns_cookie_sock_ops,
5172 .gpl_only = false,
5173 .ret_type = RET_INTEGER,
5174 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5175 };
5176
BPF_CALL_1(bpf_get_netns_cookie_sk_msg,struct sk_msg *,ctx)5177 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5178 {
5179 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5180 }
5181
5182 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5183 .func = bpf_get_netns_cookie_sk_msg,
5184 .gpl_only = false,
5185 .ret_type = RET_INTEGER,
5186 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5187 };
5188
BPF_CALL_1(bpf_get_socket_uid,struct sk_buff *,skb)5189 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5190 {
5191 struct sock *sk = sk_to_full_sk(skb->sk);
5192 kuid_t kuid;
5193
5194 if (!sk || !sk_fullsock(sk))
5195 return overflowuid;
5196 kuid = sock_net_uid(sock_net(sk), sk);
5197 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5198 }
5199
5200 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5201 .func = bpf_get_socket_uid,
5202 .gpl_only = false,
5203 .ret_type = RET_INTEGER,
5204 .arg1_type = ARG_PTR_TO_CTX,
5205 };
5206
sol_socket_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5207 static int sol_socket_sockopt(struct sock *sk, int optname,
5208 char *optval, int *optlen,
5209 bool getopt)
5210 {
5211 switch (optname) {
5212 case SO_REUSEADDR:
5213 case SO_SNDBUF:
5214 case SO_RCVBUF:
5215 case SO_KEEPALIVE:
5216 case SO_PRIORITY:
5217 case SO_REUSEPORT:
5218 case SO_RCVLOWAT:
5219 case SO_MARK:
5220 case SO_MAX_PACING_RATE:
5221 case SO_BINDTOIFINDEX:
5222 case SO_TXREHASH:
5223 if (*optlen != sizeof(int))
5224 return -EINVAL;
5225 break;
5226 case SO_BINDTODEVICE:
5227 break;
5228 default:
5229 return -EINVAL;
5230 }
5231
5232 if (getopt) {
5233 if (optname == SO_BINDTODEVICE)
5234 return -EINVAL;
5235 return sk_getsockopt(sk, SOL_SOCKET, optname,
5236 KERNEL_SOCKPTR(optval),
5237 KERNEL_SOCKPTR(optlen));
5238 }
5239
5240 return sk_setsockopt(sk, SOL_SOCKET, optname,
5241 KERNEL_SOCKPTR(optval), *optlen);
5242 }
5243
bpf_sol_tcp_setsockopt(struct sock * sk,int optname,char * optval,int optlen)5244 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5245 char *optval, int optlen)
5246 {
5247 struct tcp_sock *tp = tcp_sk(sk);
5248 unsigned long timeout;
5249 int val;
5250
5251 if (optlen != sizeof(int))
5252 return -EINVAL;
5253
5254 val = *(int *)optval;
5255
5256 /* Only some options are supported */
5257 switch (optname) {
5258 case TCP_BPF_IW:
5259 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5260 return -EINVAL;
5261 tcp_snd_cwnd_set(tp, val);
5262 break;
5263 case TCP_BPF_SNDCWND_CLAMP:
5264 if (val <= 0)
5265 return -EINVAL;
5266 tp->snd_cwnd_clamp = val;
5267 tp->snd_ssthresh = val;
5268 break;
5269 case TCP_BPF_DELACK_MAX:
5270 timeout = usecs_to_jiffies(val);
5271 if (timeout > TCP_DELACK_MAX ||
5272 timeout < TCP_TIMEOUT_MIN)
5273 return -EINVAL;
5274 inet_csk(sk)->icsk_delack_max = timeout;
5275 break;
5276 case TCP_BPF_RTO_MIN:
5277 timeout = usecs_to_jiffies(val);
5278 if (timeout > TCP_RTO_MIN ||
5279 timeout < TCP_TIMEOUT_MIN)
5280 return -EINVAL;
5281 inet_csk(sk)->icsk_rto_min = timeout;
5282 break;
5283 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5284 if (val & ~(BPF_SOCK_OPS_ALL_CB_FLAGS))
5285 return -EINVAL;
5286 tp->bpf_sock_ops_cb_flags = val;
5287 break;
5288 default:
5289 return -EINVAL;
5290 }
5291
5292 return 0;
5293 }
5294
sol_tcp_sockopt_congestion(struct sock * sk,char * optval,int * optlen,bool getopt)5295 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5296 int *optlen, bool getopt)
5297 {
5298 struct tcp_sock *tp;
5299 int ret;
5300
5301 if (*optlen < 2)
5302 return -EINVAL;
5303
5304 if (getopt) {
5305 if (!inet_csk(sk)->icsk_ca_ops)
5306 return -EINVAL;
5307 /* BPF expects NULL-terminated tcp-cc string */
5308 optval[--(*optlen)] = '\0';
5309 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5310 KERNEL_SOCKPTR(optval),
5311 KERNEL_SOCKPTR(optlen));
5312 }
5313
5314 /* "cdg" is the only cc that alloc a ptr
5315 * in inet_csk_ca area. The bpf-tcp-cc may
5316 * overwrite this ptr after switching to cdg.
5317 */
5318 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5319 return -ENOTSUPP;
5320
5321 /* It stops this looping
5322 *
5323 * .init => bpf_setsockopt(tcp_cc) => .init =>
5324 * bpf_setsockopt(tcp_cc)" => .init => ....
5325 *
5326 * The second bpf_setsockopt(tcp_cc) is not allowed
5327 * in order to break the loop when both .init
5328 * are the same bpf prog.
5329 *
5330 * This applies even the second bpf_setsockopt(tcp_cc)
5331 * does not cause a loop. This limits only the first
5332 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5333 * pick a fallback cc (eg. peer does not support ECN)
5334 * and the second '.init' cannot fallback to
5335 * another.
5336 */
5337 tp = tcp_sk(sk);
5338 if (tp->bpf_chg_cc_inprogress)
5339 return -EBUSY;
5340
5341 tp->bpf_chg_cc_inprogress = 1;
5342 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5343 KERNEL_SOCKPTR(optval), *optlen);
5344 tp->bpf_chg_cc_inprogress = 0;
5345 return ret;
5346 }
5347
sol_tcp_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5348 static int sol_tcp_sockopt(struct sock *sk, int optname,
5349 char *optval, int *optlen,
5350 bool getopt)
5351 {
5352 if (sk->sk_protocol != IPPROTO_TCP)
5353 return -EINVAL;
5354
5355 switch (optname) {
5356 case TCP_NODELAY:
5357 case TCP_MAXSEG:
5358 case TCP_KEEPIDLE:
5359 case TCP_KEEPINTVL:
5360 case TCP_KEEPCNT:
5361 case TCP_SYNCNT:
5362 case TCP_WINDOW_CLAMP:
5363 case TCP_THIN_LINEAR_TIMEOUTS:
5364 case TCP_USER_TIMEOUT:
5365 case TCP_NOTSENT_LOWAT:
5366 case TCP_SAVE_SYN:
5367 if (*optlen != sizeof(int))
5368 return -EINVAL;
5369 break;
5370 case TCP_CONGESTION:
5371 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5372 case TCP_SAVED_SYN:
5373 if (*optlen < 1)
5374 return -EINVAL;
5375 break;
5376 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5377 if (*optlen != sizeof(int))
5378 return -EINVAL;
5379 if (getopt) {
5380 struct tcp_sock *tp = tcp_sk(sk);
5381 int cb_flags = tp->bpf_sock_ops_cb_flags;
5382
5383 memcpy(optval, &cb_flags, *optlen);
5384 return 0;
5385 }
5386 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5387 default:
5388 if (getopt)
5389 return -EINVAL;
5390 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5391 }
5392
5393 if (getopt) {
5394 if (optname == TCP_SAVED_SYN) {
5395 struct tcp_sock *tp = tcp_sk(sk);
5396
5397 if (!tp->saved_syn ||
5398 *optlen > tcp_saved_syn_len(tp->saved_syn))
5399 return -EINVAL;
5400 memcpy(optval, tp->saved_syn->data, *optlen);
5401 /* It cannot free tp->saved_syn here because it
5402 * does not know if the user space still needs it.
5403 */
5404 return 0;
5405 }
5406
5407 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5408 KERNEL_SOCKPTR(optval),
5409 KERNEL_SOCKPTR(optlen));
5410 }
5411
5412 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5413 KERNEL_SOCKPTR(optval), *optlen);
5414 }
5415
sol_ip_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5416 static int sol_ip_sockopt(struct sock *sk, int optname,
5417 char *optval, int *optlen,
5418 bool getopt)
5419 {
5420 if (sk->sk_family != AF_INET)
5421 return -EINVAL;
5422
5423 switch (optname) {
5424 case IP_TOS:
5425 if (*optlen != sizeof(int))
5426 return -EINVAL;
5427 break;
5428 default:
5429 return -EINVAL;
5430 }
5431
5432 if (getopt)
5433 return do_ip_getsockopt(sk, SOL_IP, optname,
5434 KERNEL_SOCKPTR(optval),
5435 KERNEL_SOCKPTR(optlen));
5436
5437 return do_ip_setsockopt(sk, SOL_IP, optname,
5438 KERNEL_SOCKPTR(optval), *optlen);
5439 }
5440
sol_ipv6_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5441 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5442 char *optval, int *optlen,
5443 bool getopt)
5444 {
5445 if (sk->sk_family != AF_INET6)
5446 return -EINVAL;
5447
5448 switch (optname) {
5449 case IPV6_TCLASS:
5450 case IPV6_AUTOFLOWLABEL:
5451 if (*optlen != sizeof(int))
5452 return -EINVAL;
5453 break;
5454 default:
5455 return -EINVAL;
5456 }
5457
5458 if (getopt)
5459 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5460 KERNEL_SOCKPTR(optval),
5461 KERNEL_SOCKPTR(optlen));
5462
5463 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5464 KERNEL_SOCKPTR(optval), *optlen);
5465 }
5466
__bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5467 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5468 char *optval, int optlen)
5469 {
5470 if (!sk_fullsock(sk))
5471 return -EINVAL;
5472
5473 if (level == SOL_SOCKET)
5474 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5475 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5476 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5477 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5478 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5479 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5480 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5481
5482 return -EINVAL;
5483 }
5484
_bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5485 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5486 char *optval, int optlen)
5487 {
5488 if (sk_fullsock(sk))
5489 sock_owned_by_me(sk);
5490 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5491 }
5492
__bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5493 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5494 char *optval, int optlen)
5495 {
5496 int err, saved_optlen = optlen;
5497
5498 if (!sk_fullsock(sk)) {
5499 err = -EINVAL;
5500 goto done;
5501 }
5502
5503 if (level == SOL_SOCKET)
5504 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5505 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5506 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5507 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5508 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5509 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5510 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5511 else
5512 err = -EINVAL;
5513
5514 done:
5515 if (err)
5516 optlen = 0;
5517 if (optlen < saved_optlen)
5518 memset(optval + optlen, 0, saved_optlen - optlen);
5519 return err;
5520 }
5521
_bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5522 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5523 char *optval, int optlen)
5524 {
5525 if (sk_fullsock(sk))
5526 sock_owned_by_me(sk);
5527 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5528 }
5529
BPF_CALL_5(bpf_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5530 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5531 int, optname, char *, optval, int, optlen)
5532 {
5533 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5534 }
5535
5536 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5537 .func = bpf_sk_setsockopt,
5538 .gpl_only = false,
5539 .ret_type = RET_INTEGER,
5540 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5541 .arg2_type = ARG_ANYTHING,
5542 .arg3_type = ARG_ANYTHING,
5543 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5544 .arg5_type = ARG_CONST_SIZE,
5545 };
5546
BPF_CALL_5(bpf_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5547 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5548 int, optname, char *, optval, int, optlen)
5549 {
5550 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5551 }
5552
5553 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5554 .func = bpf_sk_getsockopt,
5555 .gpl_only = false,
5556 .ret_type = RET_INTEGER,
5557 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5558 .arg2_type = ARG_ANYTHING,
5559 .arg3_type = ARG_ANYTHING,
5560 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5561 .arg5_type = ARG_CONST_SIZE,
5562 };
5563
BPF_CALL_5(bpf_unlocked_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5564 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5565 int, optname, char *, optval, int, optlen)
5566 {
5567 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5568 }
5569
5570 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5571 .func = bpf_unlocked_sk_setsockopt,
5572 .gpl_only = false,
5573 .ret_type = RET_INTEGER,
5574 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5575 .arg2_type = ARG_ANYTHING,
5576 .arg3_type = ARG_ANYTHING,
5577 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5578 .arg5_type = ARG_CONST_SIZE,
5579 };
5580
BPF_CALL_5(bpf_unlocked_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5581 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5582 int, optname, char *, optval, int, optlen)
5583 {
5584 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5585 }
5586
5587 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5588 .func = bpf_unlocked_sk_getsockopt,
5589 .gpl_only = false,
5590 .ret_type = RET_INTEGER,
5591 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5592 .arg2_type = ARG_ANYTHING,
5593 .arg3_type = ARG_ANYTHING,
5594 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5595 .arg5_type = ARG_CONST_SIZE,
5596 };
5597
BPF_CALL_5(bpf_sock_addr_setsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5598 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5599 int, level, int, optname, char *, optval, int, optlen)
5600 {
5601 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5602 }
5603
5604 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5605 .func = bpf_sock_addr_setsockopt,
5606 .gpl_only = false,
5607 .ret_type = RET_INTEGER,
5608 .arg1_type = ARG_PTR_TO_CTX,
5609 .arg2_type = ARG_ANYTHING,
5610 .arg3_type = ARG_ANYTHING,
5611 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5612 .arg5_type = ARG_CONST_SIZE,
5613 };
5614
BPF_CALL_5(bpf_sock_addr_getsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5615 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5616 int, level, int, optname, char *, optval, int, optlen)
5617 {
5618 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5619 }
5620
5621 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5622 .func = bpf_sock_addr_getsockopt,
5623 .gpl_only = false,
5624 .ret_type = RET_INTEGER,
5625 .arg1_type = ARG_PTR_TO_CTX,
5626 .arg2_type = ARG_ANYTHING,
5627 .arg3_type = ARG_ANYTHING,
5628 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5629 .arg5_type = ARG_CONST_SIZE,
5630 };
5631
BPF_CALL_5(bpf_sock_ops_setsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5632 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5633 int, level, int, optname, char *, optval, int, optlen)
5634 {
5635 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5636 }
5637
5638 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5639 .func = bpf_sock_ops_setsockopt,
5640 .gpl_only = false,
5641 .ret_type = RET_INTEGER,
5642 .arg1_type = ARG_PTR_TO_CTX,
5643 .arg2_type = ARG_ANYTHING,
5644 .arg3_type = ARG_ANYTHING,
5645 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5646 .arg5_type = ARG_CONST_SIZE,
5647 };
5648
bpf_sock_ops_get_syn(struct bpf_sock_ops_kern * bpf_sock,int optname,const u8 ** start)5649 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5650 int optname, const u8 **start)
5651 {
5652 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5653 const u8 *hdr_start;
5654 int ret;
5655
5656 if (syn_skb) {
5657 /* sk is a request_sock here */
5658
5659 if (optname == TCP_BPF_SYN) {
5660 hdr_start = syn_skb->data;
5661 ret = tcp_hdrlen(syn_skb);
5662 } else if (optname == TCP_BPF_SYN_IP) {
5663 hdr_start = skb_network_header(syn_skb);
5664 ret = skb_network_header_len(syn_skb) +
5665 tcp_hdrlen(syn_skb);
5666 } else {
5667 /* optname == TCP_BPF_SYN_MAC */
5668 hdr_start = skb_mac_header(syn_skb);
5669 ret = skb_mac_header_len(syn_skb) +
5670 skb_network_header_len(syn_skb) +
5671 tcp_hdrlen(syn_skb);
5672 }
5673 } else {
5674 struct sock *sk = bpf_sock->sk;
5675 struct saved_syn *saved_syn;
5676
5677 if (sk->sk_state == TCP_NEW_SYN_RECV)
5678 /* synack retransmit. bpf_sock->syn_skb will
5679 * not be available. It has to resort to
5680 * saved_syn (if it is saved).
5681 */
5682 saved_syn = inet_reqsk(sk)->saved_syn;
5683 else
5684 saved_syn = tcp_sk(sk)->saved_syn;
5685
5686 if (!saved_syn)
5687 return -ENOENT;
5688
5689 if (optname == TCP_BPF_SYN) {
5690 hdr_start = saved_syn->data +
5691 saved_syn->mac_hdrlen +
5692 saved_syn->network_hdrlen;
5693 ret = saved_syn->tcp_hdrlen;
5694 } else if (optname == TCP_BPF_SYN_IP) {
5695 hdr_start = saved_syn->data +
5696 saved_syn->mac_hdrlen;
5697 ret = saved_syn->network_hdrlen +
5698 saved_syn->tcp_hdrlen;
5699 } else {
5700 /* optname == TCP_BPF_SYN_MAC */
5701
5702 /* TCP_SAVE_SYN may not have saved the mac hdr */
5703 if (!saved_syn->mac_hdrlen)
5704 return -ENOENT;
5705
5706 hdr_start = saved_syn->data;
5707 ret = saved_syn->mac_hdrlen +
5708 saved_syn->network_hdrlen +
5709 saved_syn->tcp_hdrlen;
5710 }
5711 }
5712
5713 *start = hdr_start;
5714 return ret;
5715 }
5716
BPF_CALL_5(bpf_sock_ops_getsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5717 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5718 int, level, int, optname, char *, optval, int, optlen)
5719 {
5720 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5721 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5722 int ret, copy_len = 0;
5723 const u8 *start;
5724
5725 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5726 if (ret > 0) {
5727 copy_len = ret;
5728 if (optlen < copy_len) {
5729 copy_len = optlen;
5730 ret = -ENOSPC;
5731 }
5732
5733 memcpy(optval, start, copy_len);
5734 }
5735
5736 /* Zero out unused buffer at the end */
5737 memset(optval + copy_len, 0, optlen - copy_len);
5738
5739 return ret;
5740 }
5741
5742 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5743 }
5744
5745 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5746 .func = bpf_sock_ops_getsockopt,
5747 .gpl_only = false,
5748 .ret_type = RET_INTEGER,
5749 .arg1_type = ARG_PTR_TO_CTX,
5750 .arg2_type = ARG_ANYTHING,
5751 .arg3_type = ARG_ANYTHING,
5752 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5753 .arg5_type = ARG_CONST_SIZE,
5754 };
5755
BPF_CALL_2(bpf_sock_ops_cb_flags_set,struct bpf_sock_ops_kern *,bpf_sock,int,argval)5756 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5757 int, argval)
5758 {
5759 struct sock *sk = bpf_sock->sk;
5760 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5761
5762 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5763 return -EINVAL;
5764
5765 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5766
5767 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5768 }
5769
5770 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5771 .func = bpf_sock_ops_cb_flags_set,
5772 .gpl_only = false,
5773 .ret_type = RET_INTEGER,
5774 .arg1_type = ARG_PTR_TO_CTX,
5775 .arg2_type = ARG_ANYTHING,
5776 };
5777
5778 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5779 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5780
BPF_CALL_3(bpf_bind,struct bpf_sock_addr_kern *,ctx,struct sockaddr *,addr,int,addr_len)5781 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5782 int, addr_len)
5783 {
5784 #ifdef CONFIG_INET
5785 struct sock *sk = ctx->sk;
5786 u32 flags = BIND_FROM_BPF;
5787 int err;
5788
5789 err = -EINVAL;
5790 if (addr_len < offsetofend(struct sockaddr, sa_family))
5791 return err;
5792 if (addr->sa_family == AF_INET) {
5793 if (addr_len < sizeof(struct sockaddr_in))
5794 return err;
5795 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5796 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5797 return __inet_bind(sk, addr, addr_len, flags);
5798 #if IS_ENABLED(CONFIG_IPV6)
5799 } else if (addr->sa_family == AF_INET6) {
5800 if (addr_len < SIN6_LEN_RFC2133)
5801 return err;
5802 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5803 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5804 /* ipv6_bpf_stub cannot be NULL, since it's called from
5805 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5806 */
5807 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5808 #endif /* CONFIG_IPV6 */
5809 }
5810 #endif /* CONFIG_INET */
5811
5812 return -EAFNOSUPPORT;
5813 }
5814
5815 static const struct bpf_func_proto bpf_bind_proto = {
5816 .func = bpf_bind,
5817 .gpl_only = false,
5818 .ret_type = RET_INTEGER,
5819 .arg1_type = ARG_PTR_TO_CTX,
5820 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5821 .arg3_type = ARG_CONST_SIZE,
5822 };
5823
5824 #ifdef CONFIG_XFRM
5825
5826 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5827 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5828
5829 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5830 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5831
5832 #endif
5833
BPF_CALL_5(bpf_skb_get_xfrm_state,struct sk_buff *,skb,u32,index,struct bpf_xfrm_state *,to,u32,size,u64,flags)5834 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5835 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5836 {
5837 const struct sec_path *sp = skb_sec_path(skb);
5838 const struct xfrm_state *x;
5839
5840 if (!sp || unlikely(index >= sp->len || flags))
5841 goto err_clear;
5842
5843 x = sp->xvec[index];
5844
5845 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5846 goto err_clear;
5847
5848 to->reqid = x->props.reqid;
5849 to->spi = x->id.spi;
5850 to->family = x->props.family;
5851 to->ext = 0;
5852
5853 if (to->family == AF_INET6) {
5854 memcpy(to->remote_ipv6, x->props.saddr.a6,
5855 sizeof(to->remote_ipv6));
5856 } else {
5857 to->remote_ipv4 = x->props.saddr.a4;
5858 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5859 }
5860
5861 return 0;
5862 err_clear:
5863 memset(to, 0, size);
5864 return -EINVAL;
5865 }
5866
5867 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5868 .func = bpf_skb_get_xfrm_state,
5869 .gpl_only = false,
5870 .ret_type = RET_INTEGER,
5871 .arg1_type = ARG_PTR_TO_CTX,
5872 .arg2_type = ARG_ANYTHING,
5873 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5874 .arg4_type = ARG_CONST_SIZE,
5875 .arg5_type = ARG_ANYTHING,
5876 };
5877 #endif
5878
5879 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
bpf_fib_set_fwd_params(struct bpf_fib_lookup * params,u32 mtu)5880 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5881 {
5882 params->h_vlan_TCI = 0;
5883 params->h_vlan_proto = 0;
5884 if (mtu)
5885 params->mtu_result = mtu; /* union with tot_len */
5886
5887 return 0;
5888 }
5889 #endif
5890
5891 #if IS_ENABLED(CONFIG_INET)
bpf_ipv4_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5892 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5893 u32 flags, bool check_mtu)
5894 {
5895 struct fib_nh_common *nhc;
5896 struct in_device *in_dev;
5897 struct neighbour *neigh;
5898 struct net_device *dev;
5899 struct fib_result res;
5900 struct flowi4 fl4;
5901 u32 mtu = 0;
5902 int err;
5903
5904 dev = dev_get_by_index_rcu(net, params->ifindex);
5905 if (unlikely(!dev))
5906 return -ENODEV;
5907
5908 /* verify forwarding is enabled on this interface */
5909 in_dev = __in_dev_get_rcu(dev);
5910 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5911 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5912
5913 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5914 fl4.flowi4_iif = 1;
5915 fl4.flowi4_oif = params->ifindex;
5916 } else {
5917 fl4.flowi4_iif = params->ifindex;
5918 fl4.flowi4_oif = 0;
5919 }
5920 fl4.flowi4_tos = params->tos & INET_DSCP_MASK;
5921 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5922 fl4.flowi4_flags = 0;
5923
5924 fl4.flowi4_proto = params->l4_protocol;
5925 fl4.daddr = params->ipv4_dst;
5926 fl4.saddr = params->ipv4_src;
5927 fl4.fl4_sport = params->sport;
5928 fl4.fl4_dport = params->dport;
5929 fl4.flowi4_multipath_hash = 0;
5930
5931 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5932 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5933 struct fib_table *tb;
5934
5935 if (flags & BPF_FIB_LOOKUP_TBID) {
5936 tbid = params->tbid;
5937 /* zero out for vlan output */
5938 params->tbid = 0;
5939 }
5940
5941 tb = fib_get_table(net, tbid);
5942 if (unlikely(!tb))
5943 return BPF_FIB_LKUP_RET_NOT_FWDED;
5944
5945 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5946 } else {
5947 if (flags & BPF_FIB_LOOKUP_MARK)
5948 fl4.flowi4_mark = params->mark;
5949 else
5950 fl4.flowi4_mark = 0;
5951 fl4.flowi4_secid = 0;
5952 fl4.flowi4_tun_key.tun_id = 0;
5953 fl4.flowi4_uid = sock_net_uid(net, NULL);
5954
5955 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5956 }
5957
5958 if (err) {
5959 /* map fib lookup errors to RTN_ type */
5960 if (err == -EINVAL)
5961 return BPF_FIB_LKUP_RET_BLACKHOLE;
5962 if (err == -EHOSTUNREACH)
5963 return BPF_FIB_LKUP_RET_UNREACHABLE;
5964 if (err == -EACCES)
5965 return BPF_FIB_LKUP_RET_PROHIBIT;
5966
5967 return BPF_FIB_LKUP_RET_NOT_FWDED;
5968 }
5969
5970 if (res.type != RTN_UNICAST)
5971 return BPF_FIB_LKUP_RET_NOT_FWDED;
5972
5973 if (fib_info_num_path(res.fi) > 1)
5974 fib_select_path(net, &res, &fl4, NULL);
5975
5976 if (check_mtu) {
5977 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5978 if (params->tot_len > mtu) {
5979 params->mtu_result = mtu; /* union with tot_len */
5980 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5981 }
5982 }
5983
5984 nhc = res.nhc;
5985
5986 /* do not handle lwt encaps right now */
5987 if (nhc->nhc_lwtstate)
5988 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5989
5990 dev = nhc->nhc_dev;
5991
5992 params->rt_metric = res.fi->fib_priority;
5993 params->ifindex = dev->ifindex;
5994
5995 if (flags & BPF_FIB_LOOKUP_SRC)
5996 params->ipv4_src = fib_result_prefsrc(net, &res);
5997
5998 /* xdp and cls_bpf programs are run in RCU-bh so
5999 * rcu_read_lock_bh is not needed here
6000 */
6001 if (likely(nhc->nhc_gw_family != AF_INET6)) {
6002 if (nhc->nhc_gw_family)
6003 params->ipv4_dst = nhc->nhc_gw.ipv4;
6004 } else {
6005 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
6006
6007 params->family = AF_INET6;
6008 *dst = nhc->nhc_gw.ipv6;
6009 }
6010
6011 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6012 goto set_fwd_params;
6013
6014 if (likely(nhc->nhc_gw_family != AF_INET6))
6015 neigh = __ipv4_neigh_lookup_noref(dev,
6016 (__force u32)params->ipv4_dst);
6017 else
6018 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
6019
6020 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6021 return BPF_FIB_LKUP_RET_NO_NEIGH;
6022 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6023 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6024
6025 set_fwd_params:
6026 return bpf_fib_set_fwd_params(params, mtu);
6027 }
6028 #endif
6029
6030 #if IS_ENABLED(CONFIG_IPV6)
bpf_ipv6_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)6031 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
6032 u32 flags, bool check_mtu)
6033 {
6034 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
6035 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
6036 struct fib6_result res = {};
6037 struct neighbour *neigh;
6038 struct net_device *dev;
6039 struct inet6_dev *idev;
6040 struct flowi6 fl6;
6041 int strict = 0;
6042 int oif, err;
6043 u32 mtu = 0;
6044
6045 /* link local addresses are never forwarded */
6046 if (rt6_need_strict(dst) || rt6_need_strict(src))
6047 return BPF_FIB_LKUP_RET_NOT_FWDED;
6048
6049 dev = dev_get_by_index_rcu(net, params->ifindex);
6050 if (unlikely(!dev))
6051 return -ENODEV;
6052
6053 idev = __in6_dev_get_safely(dev);
6054 if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding)))
6055 return BPF_FIB_LKUP_RET_FWD_DISABLED;
6056
6057 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
6058 fl6.flowi6_iif = 1;
6059 oif = fl6.flowi6_oif = params->ifindex;
6060 } else {
6061 oif = fl6.flowi6_iif = params->ifindex;
6062 fl6.flowi6_oif = 0;
6063 strict = RT6_LOOKUP_F_HAS_SADDR;
6064 }
6065 fl6.flowlabel = params->flowinfo;
6066 fl6.flowi6_scope = 0;
6067 fl6.flowi6_flags = 0;
6068 fl6.mp_hash = 0;
6069
6070 fl6.flowi6_proto = params->l4_protocol;
6071 fl6.daddr = *dst;
6072 fl6.saddr = *src;
6073 fl6.fl6_sport = params->sport;
6074 fl6.fl6_dport = params->dport;
6075
6076 if (flags & BPF_FIB_LOOKUP_DIRECT) {
6077 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
6078 struct fib6_table *tb;
6079
6080 if (flags & BPF_FIB_LOOKUP_TBID) {
6081 tbid = params->tbid;
6082 /* zero out for vlan output */
6083 params->tbid = 0;
6084 }
6085
6086 tb = ipv6_stub->fib6_get_table(net, tbid);
6087 if (unlikely(!tb))
6088 return BPF_FIB_LKUP_RET_NOT_FWDED;
6089
6090 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
6091 strict);
6092 } else {
6093 if (flags & BPF_FIB_LOOKUP_MARK)
6094 fl6.flowi6_mark = params->mark;
6095 else
6096 fl6.flowi6_mark = 0;
6097 fl6.flowi6_secid = 0;
6098 fl6.flowi6_tun_key.tun_id = 0;
6099 fl6.flowi6_uid = sock_net_uid(net, NULL);
6100
6101 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
6102 }
6103
6104 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
6105 res.f6i == net->ipv6.fib6_null_entry))
6106 return BPF_FIB_LKUP_RET_NOT_FWDED;
6107
6108 switch (res.fib6_type) {
6109 /* only unicast is forwarded */
6110 case RTN_UNICAST:
6111 break;
6112 case RTN_BLACKHOLE:
6113 return BPF_FIB_LKUP_RET_BLACKHOLE;
6114 case RTN_UNREACHABLE:
6115 return BPF_FIB_LKUP_RET_UNREACHABLE;
6116 case RTN_PROHIBIT:
6117 return BPF_FIB_LKUP_RET_PROHIBIT;
6118 default:
6119 return BPF_FIB_LKUP_RET_NOT_FWDED;
6120 }
6121
6122 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
6123 fl6.flowi6_oif != 0, NULL, strict);
6124
6125 if (check_mtu) {
6126 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
6127 if (params->tot_len > mtu) {
6128 params->mtu_result = mtu; /* union with tot_len */
6129 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6130 }
6131 }
6132
6133 if (res.nh->fib_nh_lws)
6134 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6135
6136 if (res.nh->fib_nh_gw_family)
6137 *dst = res.nh->fib_nh_gw6;
6138
6139 dev = res.nh->fib_nh_dev;
6140 params->rt_metric = res.f6i->fib6_metric;
6141 params->ifindex = dev->ifindex;
6142
6143 if (flags & BPF_FIB_LOOKUP_SRC) {
6144 if (res.f6i->fib6_prefsrc.plen) {
6145 *src = res.f6i->fib6_prefsrc.addr;
6146 } else {
6147 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
6148 &fl6.daddr, 0,
6149 src);
6150 if (err)
6151 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
6152 }
6153 }
6154
6155 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6156 goto set_fwd_params;
6157
6158 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6159 * not needed here.
6160 */
6161 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6162 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6163 return BPF_FIB_LKUP_RET_NO_NEIGH;
6164 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6165 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6166
6167 set_fwd_params:
6168 return bpf_fib_set_fwd_params(params, mtu);
6169 }
6170 #endif
6171
6172 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6173 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6174 BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_MARK)
6175
BPF_CALL_4(bpf_xdp_fib_lookup,struct xdp_buff *,ctx,struct bpf_fib_lookup *,params,int,plen,u32,flags)6176 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6177 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6178 {
6179 if (plen < sizeof(*params))
6180 return -EINVAL;
6181
6182 if (flags & ~BPF_FIB_LOOKUP_MASK)
6183 return -EINVAL;
6184
6185 switch (params->family) {
6186 #if IS_ENABLED(CONFIG_INET)
6187 case AF_INET:
6188 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6189 flags, true);
6190 #endif
6191 #if IS_ENABLED(CONFIG_IPV6)
6192 case AF_INET6:
6193 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6194 flags, true);
6195 #endif
6196 }
6197 return -EAFNOSUPPORT;
6198 }
6199
6200 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6201 .func = bpf_xdp_fib_lookup,
6202 .gpl_only = true,
6203 .ret_type = RET_INTEGER,
6204 .arg1_type = ARG_PTR_TO_CTX,
6205 .arg2_type = ARG_PTR_TO_MEM,
6206 .arg3_type = ARG_CONST_SIZE,
6207 .arg4_type = ARG_ANYTHING,
6208 };
6209
BPF_CALL_4(bpf_skb_fib_lookup,struct sk_buff *,skb,struct bpf_fib_lookup *,params,int,plen,u32,flags)6210 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6211 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6212 {
6213 struct net *net = dev_net(skb->dev);
6214 int rc = -EAFNOSUPPORT;
6215 bool check_mtu = false;
6216
6217 if (plen < sizeof(*params))
6218 return -EINVAL;
6219
6220 if (flags & ~BPF_FIB_LOOKUP_MASK)
6221 return -EINVAL;
6222
6223 if (params->tot_len)
6224 check_mtu = true;
6225
6226 switch (params->family) {
6227 #if IS_ENABLED(CONFIG_INET)
6228 case AF_INET:
6229 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6230 break;
6231 #endif
6232 #if IS_ENABLED(CONFIG_IPV6)
6233 case AF_INET6:
6234 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6235 break;
6236 #endif
6237 }
6238
6239 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6240 struct net_device *dev;
6241
6242 /* When tot_len isn't provided by user, check skb
6243 * against MTU of FIB lookup resulting net_device
6244 */
6245 dev = dev_get_by_index_rcu(net, params->ifindex);
6246 if (!is_skb_forwardable(dev, skb))
6247 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6248
6249 params->mtu_result = dev->mtu; /* union with tot_len */
6250 }
6251
6252 return rc;
6253 }
6254
6255 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6256 .func = bpf_skb_fib_lookup,
6257 .gpl_only = true,
6258 .ret_type = RET_INTEGER,
6259 .arg1_type = ARG_PTR_TO_CTX,
6260 .arg2_type = ARG_PTR_TO_MEM,
6261 .arg3_type = ARG_CONST_SIZE,
6262 .arg4_type = ARG_ANYTHING,
6263 };
6264
__dev_via_ifindex(struct net_device * dev_curr,u32 ifindex)6265 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6266 u32 ifindex)
6267 {
6268 struct net *netns = dev_net(dev_curr);
6269
6270 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6271 if (ifindex == 0)
6272 return dev_curr;
6273
6274 return dev_get_by_index_rcu(netns, ifindex);
6275 }
6276
BPF_CALL_5(bpf_skb_check_mtu,struct sk_buff *,skb,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6277 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6278 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6279 {
6280 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6281 struct net_device *dev = skb->dev;
6282 int skb_len, dev_len;
6283 int mtu;
6284
6285 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6286 return -EINVAL;
6287
6288 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6289 return -EINVAL;
6290
6291 dev = __dev_via_ifindex(dev, ifindex);
6292 if (unlikely(!dev))
6293 return -ENODEV;
6294
6295 mtu = READ_ONCE(dev->mtu);
6296
6297 dev_len = mtu + dev->hard_header_len;
6298
6299 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6300 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6301
6302 skb_len += len_diff; /* minus result pass check */
6303 if (skb_len <= dev_len) {
6304 ret = BPF_MTU_CHK_RET_SUCCESS;
6305 goto out;
6306 }
6307 /* At this point, skb->len exceed MTU, but as it include length of all
6308 * segments, it can still be below MTU. The SKB can possibly get
6309 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6310 * must choose if segs are to be MTU checked.
6311 */
6312 if (skb_is_gso(skb)) {
6313 ret = BPF_MTU_CHK_RET_SUCCESS;
6314
6315 if (flags & BPF_MTU_CHK_SEGS &&
6316 !skb_gso_validate_network_len(skb, mtu))
6317 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6318 }
6319 out:
6320 /* BPF verifier guarantees valid pointer */
6321 *mtu_len = mtu;
6322
6323 return ret;
6324 }
6325
BPF_CALL_5(bpf_xdp_check_mtu,struct xdp_buff *,xdp,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6326 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6327 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6328 {
6329 struct net_device *dev = xdp->rxq->dev;
6330 int xdp_len = xdp->data_end - xdp->data;
6331 int ret = BPF_MTU_CHK_RET_SUCCESS;
6332 int mtu, dev_len;
6333
6334 /* XDP variant doesn't support multi-buffer segment check (yet) */
6335 if (unlikely(flags))
6336 return -EINVAL;
6337
6338 dev = __dev_via_ifindex(dev, ifindex);
6339 if (unlikely(!dev))
6340 return -ENODEV;
6341
6342 mtu = READ_ONCE(dev->mtu);
6343
6344 /* Add L2-header as dev MTU is L3 size */
6345 dev_len = mtu + dev->hard_header_len;
6346
6347 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6348 if (*mtu_len)
6349 xdp_len = *mtu_len + dev->hard_header_len;
6350
6351 xdp_len += len_diff; /* minus result pass check */
6352 if (xdp_len > dev_len)
6353 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6354
6355 /* BPF verifier guarantees valid pointer */
6356 *mtu_len = mtu;
6357
6358 return ret;
6359 }
6360
6361 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6362 .func = bpf_skb_check_mtu,
6363 .gpl_only = true,
6364 .ret_type = RET_INTEGER,
6365 .arg1_type = ARG_PTR_TO_CTX,
6366 .arg2_type = ARG_ANYTHING,
6367 .arg3_type = ARG_PTR_TO_INT,
6368 .arg4_type = ARG_ANYTHING,
6369 .arg5_type = ARG_ANYTHING,
6370 };
6371
6372 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6373 .func = bpf_xdp_check_mtu,
6374 .gpl_only = true,
6375 .ret_type = RET_INTEGER,
6376 .arg1_type = ARG_PTR_TO_CTX,
6377 .arg2_type = ARG_ANYTHING,
6378 .arg3_type = ARG_PTR_TO_INT,
6379 .arg4_type = ARG_ANYTHING,
6380 .arg5_type = ARG_ANYTHING,
6381 };
6382
6383 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
bpf_push_seg6_encap(struct sk_buff * skb,u32 type,void * hdr,u32 len)6384 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6385 {
6386 int err;
6387 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6388
6389 if (!seg6_validate_srh(srh, len, false))
6390 return -EINVAL;
6391
6392 switch (type) {
6393 case BPF_LWT_ENCAP_SEG6_INLINE:
6394 if (skb->protocol != htons(ETH_P_IPV6))
6395 return -EBADMSG;
6396
6397 err = seg6_do_srh_inline(skb, srh);
6398 break;
6399 case BPF_LWT_ENCAP_SEG6:
6400 skb_reset_inner_headers(skb);
6401 skb->encapsulation = 1;
6402 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6403 break;
6404 default:
6405 return -EINVAL;
6406 }
6407
6408 bpf_compute_data_pointers(skb);
6409 if (err)
6410 return err;
6411
6412 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6413
6414 return seg6_lookup_nexthop(skb, NULL, 0);
6415 }
6416 #endif /* CONFIG_IPV6_SEG6_BPF */
6417
6418 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
bpf_push_ip_encap(struct sk_buff * skb,void * hdr,u32 len,bool ingress)6419 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6420 bool ingress)
6421 {
6422 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6423 }
6424 #endif
6425
BPF_CALL_4(bpf_lwt_in_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6426 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6427 u32, len)
6428 {
6429 switch (type) {
6430 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6431 case BPF_LWT_ENCAP_SEG6:
6432 case BPF_LWT_ENCAP_SEG6_INLINE:
6433 return bpf_push_seg6_encap(skb, type, hdr, len);
6434 #endif
6435 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6436 case BPF_LWT_ENCAP_IP:
6437 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6438 #endif
6439 default:
6440 return -EINVAL;
6441 }
6442 }
6443
BPF_CALL_4(bpf_lwt_xmit_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6444 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6445 void *, hdr, u32, len)
6446 {
6447 switch (type) {
6448 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6449 case BPF_LWT_ENCAP_IP:
6450 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6451 #endif
6452 default:
6453 return -EINVAL;
6454 }
6455 }
6456
6457 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6458 .func = bpf_lwt_in_push_encap,
6459 .gpl_only = false,
6460 .ret_type = RET_INTEGER,
6461 .arg1_type = ARG_PTR_TO_CTX,
6462 .arg2_type = ARG_ANYTHING,
6463 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6464 .arg4_type = ARG_CONST_SIZE
6465 };
6466
6467 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6468 .func = bpf_lwt_xmit_push_encap,
6469 .gpl_only = false,
6470 .ret_type = RET_INTEGER,
6471 .arg1_type = ARG_PTR_TO_CTX,
6472 .arg2_type = ARG_ANYTHING,
6473 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6474 .arg4_type = ARG_CONST_SIZE
6475 };
6476
6477 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
BPF_CALL_4(bpf_lwt_seg6_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len)6478 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6479 const void *, from, u32, len)
6480 {
6481 struct seg6_bpf_srh_state *srh_state =
6482 this_cpu_ptr(&seg6_bpf_srh_states);
6483 struct ipv6_sr_hdr *srh = srh_state->srh;
6484 void *srh_tlvs, *srh_end, *ptr;
6485 int srhoff = 0;
6486
6487 lockdep_assert_held(&srh_state->bh_lock);
6488 if (srh == NULL)
6489 return -EINVAL;
6490
6491 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6492 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6493
6494 ptr = skb->data + offset;
6495 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6496 srh_state->valid = false;
6497 else if (ptr < (void *)&srh->flags ||
6498 ptr + len > (void *)&srh->segments)
6499 return -EFAULT;
6500
6501 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6502 return -EFAULT;
6503 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6504 return -EINVAL;
6505 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6506
6507 memcpy(skb->data + offset, from, len);
6508 return 0;
6509 }
6510
6511 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6512 .func = bpf_lwt_seg6_store_bytes,
6513 .gpl_only = false,
6514 .ret_type = RET_INTEGER,
6515 .arg1_type = ARG_PTR_TO_CTX,
6516 .arg2_type = ARG_ANYTHING,
6517 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6518 .arg4_type = ARG_CONST_SIZE
6519 };
6520
bpf_update_srh_state(struct sk_buff * skb)6521 static void bpf_update_srh_state(struct sk_buff *skb)
6522 {
6523 struct seg6_bpf_srh_state *srh_state =
6524 this_cpu_ptr(&seg6_bpf_srh_states);
6525 int srhoff = 0;
6526
6527 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6528 srh_state->srh = NULL;
6529 } else {
6530 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6531 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6532 srh_state->valid = true;
6533 }
6534 }
6535
BPF_CALL_4(bpf_lwt_seg6_action,struct sk_buff *,skb,u32,action,void *,param,u32,param_len)6536 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6537 u32, action, void *, param, u32, param_len)
6538 {
6539 struct seg6_bpf_srh_state *srh_state =
6540 this_cpu_ptr(&seg6_bpf_srh_states);
6541 int hdroff = 0;
6542 int err;
6543
6544 lockdep_assert_held(&srh_state->bh_lock);
6545 switch (action) {
6546 case SEG6_LOCAL_ACTION_END_X:
6547 if (!seg6_bpf_has_valid_srh(skb))
6548 return -EBADMSG;
6549 if (param_len != sizeof(struct in6_addr))
6550 return -EINVAL;
6551 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6552 case SEG6_LOCAL_ACTION_END_T:
6553 if (!seg6_bpf_has_valid_srh(skb))
6554 return -EBADMSG;
6555 if (param_len != sizeof(int))
6556 return -EINVAL;
6557 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6558 case SEG6_LOCAL_ACTION_END_DT6:
6559 if (!seg6_bpf_has_valid_srh(skb))
6560 return -EBADMSG;
6561 if (param_len != sizeof(int))
6562 return -EINVAL;
6563
6564 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6565 return -EBADMSG;
6566 if (!pskb_pull(skb, hdroff))
6567 return -EBADMSG;
6568
6569 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6570 skb_reset_network_header(skb);
6571 skb_reset_transport_header(skb);
6572 skb->encapsulation = 0;
6573
6574 bpf_compute_data_pointers(skb);
6575 bpf_update_srh_state(skb);
6576 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6577 case SEG6_LOCAL_ACTION_END_B6:
6578 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6579 return -EBADMSG;
6580 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6581 param, param_len);
6582 if (!err)
6583 bpf_update_srh_state(skb);
6584
6585 return err;
6586 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6587 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6588 return -EBADMSG;
6589 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6590 param, param_len);
6591 if (!err)
6592 bpf_update_srh_state(skb);
6593
6594 return err;
6595 default:
6596 return -EINVAL;
6597 }
6598 }
6599
6600 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6601 .func = bpf_lwt_seg6_action,
6602 .gpl_only = false,
6603 .ret_type = RET_INTEGER,
6604 .arg1_type = ARG_PTR_TO_CTX,
6605 .arg2_type = ARG_ANYTHING,
6606 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6607 .arg4_type = ARG_CONST_SIZE
6608 };
6609
BPF_CALL_3(bpf_lwt_seg6_adjust_srh,struct sk_buff *,skb,u32,offset,s32,len)6610 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6611 s32, len)
6612 {
6613 struct seg6_bpf_srh_state *srh_state =
6614 this_cpu_ptr(&seg6_bpf_srh_states);
6615 struct ipv6_sr_hdr *srh = srh_state->srh;
6616 void *srh_end, *srh_tlvs, *ptr;
6617 struct ipv6hdr *hdr;
6618 int srhoff = 0;
6619 int ret;
6620
6621 lockdep_assert_held(&srh_state->bh_lock);
6622 if (unlikely(srh == NULL))
6623 return -EINVAL;
6624
6625 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6626 ((srh->first_segment + 1) << 4));
6627 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6628 srh_state->hdrlen);
6629 ptr = skb->data + offset;
6630
6631 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6632 return -EFAULT;
6633 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6634 return -EFAULT;
6635
6636 if (len > 0) {
6637 ret = skb_cow_head(skb, len);
6638 if (unlikely(ret < 0))
6639 return ret;
6640
6641 ret = bpf_skb_net_hdr_push(skb, offset, len);
6642 } else {
6643 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6644 }
6645
6646 bpf_compute_data_pointers(skb);
6647 if (unlikely(ret < 0))
6648 return ret;
6649
6650 hdr = (struct ipv6hdr *)skb->data;
6651 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6652
6653 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6654 return -EINVAL;
6655 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6656 srh_state->hdrlen += len;
6657 srh_state->valid = false;
6658 return 0;
6659 }
6660
6661 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6662 .func = bpf_lwt_seg6_adjust_srh,
6663 .gpl_only = false,
6664 .ret_type = RET_INTEGER,
6665 .arg1_type = ARG_PTR_TO_CTX,
6666 .arg2_type = ARG_ANYTHING,
6667 .arg3_type = ARG_ANYTHING,
6668 };
6669 #endif /* CONFIG_IPV6_SEG6_BPF */
6670
6671 #ifdef CONFIG_INET
sk_lookup(struct net * net,struct bpf_sock_tuple * tuple,int dif,int sdif,u8 family,u8 proto)6672 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6673 int dif, int sdif, u8 family, u8 proto)
6674 {
6675 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6676 bool refcounted = false;
6677 struct sock *sk = NULL;
6678
6679 if (family == AF_INET) {
6680 __be32 src4 = tuple->ipv4.saddr;
6681 __be32 dst4 = tuple->ipv4.daddr;
6682
6683 if (proto == IPPROTO_TCP)
6684 sk = __inet_lookup(net, hinfo, NULL, 0,
6685 src4, tuple->ipv4.sport,
6686 dst4, tuple->ipv4.dport,
6687 dif, sdif, &refcounted);
6688 else
6689 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6690 dst4, tuple->ipv4.dport,
6691 dif, sdif, net->ipv4.udp_table, NULL);
6692 #if IS_ENABLED(CONFIG_IPV6)
6693 } else {
6694 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6695 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6696
6697 if (proto == IPPROTO_TCP)
6698 sk = __inet6_lookup(net, hinfo, NULL, 0,
6699 src6, tuple->ipv6.sport,
6700 dst6, ntohs(tuple->ipv6.dport),
6701 dif, sdif, &refcounted);
6702 else if (likely(ipv6_bpf_stub))
6703 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6704 src6, tuple->ipv6.sport,
6705 dst6, tuple->ipv6.dport,
6706 dif, sdif,
6707 net->ipv4.udp_table, NULL);
6708 #endif
6709 }
6710
6711 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6712 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6713 sk = NULL;
6714 }
6715 return sk;
6716 }
6717
6718 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6719 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6720 */
6721 static struct sock *
__bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6722 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6723 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6724 u64 flags, int sdif)
6725 {
6726 struct sock *sk = NULL;
6727 struct net *net;
6728 u8 family;
6729
6730 if (len == sizeof(tuple->ipv4))
6731 family = AF_INET;
6732 else if (len == sizeof(tuple->ipv6))
6733 family = AF_INET6;
6734 else
6735 return NULL;
6736
6737 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6738 goto out;
6739
6740 if (sdif < 0) {
6741 if (family == AF_INET)
6742 sdif = inet_sdif(skb);
6743 else
6744 sdif = inet6_sdif(skb);
6745 }
6746
6747 if ((s32)netns_id < 0) {
6748 net = caller_net;
6749 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6750 } else {
6751 net = get_net_ns_by_id(caller_net, netns_id);
6752 if (unlikely(!net))
6753 goto out;
6754 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6755 put_net(net);
6756 }
6757
6758 out:
6759 return sk;
6760 }
6761
6762 static struct sock *
__bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6763 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6764 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6765 u64 flags, int sdif)
6766 {
6767 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6768 ifindex, proto, netns_id, flags,
6769 sdif);
6770
6771 if (sk) {
6772 struct sock *sk2 = sk_to_full_sk(sk);
6773
6774 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6775 * sock refcnt is decremented to prevent a request_sock leak.
6776 */
6777 if (!sk_fullsock(sk2))
6778 sk2 = NULL;
6779 if (sk2 != sk) {
6780 sock_gen_put(sk);
6781 /* Ensure there is no need to bump sk2 refcnt */
6782 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6783 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6784 return NULL;
6785 }
6786 sk = sk2;
6787 }
6788 }
6789
6790 return sk;
6791 }
6792
6793 static struct sock *
bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6794 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6795 u8 proto, u64 netns_id, u64 flags)
6796 {
6797 struct net *caller_net;
6798 int ifindex;
6799
6800 if (skb->dev) {
6801 caller_net = dev_net(skb->dev);
6802 ifindex = skb->dev->ifindex;
6803 } else {
6804 caller_net = sock_net(skb->sk);
6805 ifindex = 0;
6806 }
6807
6808 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6809 netns_id, flags, -1);
6810 }
6811
6812 static struct sock *
bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6813 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6814 u8 proto, u64 netns_id, u64 flags)
6815 {
6816 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6817 flags);
6818
6819 if (sk) {
6820 struct sock *sk2 = sk_to_full_sk(sk);
6821
6822 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6823 * sock refcnt is decremented to prevent a request_sock leak.
6824 */
6825 if (!sk_fullsock(sk2))
6826 sk2 = NULL;
6827 if (sk2 != sk) {
6828 sock_gen_put(sk);
6829 /* Ensure there is no need to bump sk2 refcnt */
6830 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6831 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6832 return NULL;
6833 }
6834 sk = sk2;
6835 }
6836 }
6837
6838 return sk;
6839 }
6840
BPF_CALL_5(bpf_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6841 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6842 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6843 {
6844 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6845 netns_id, flags);
6846 }
6847
6848 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6849 .func = bpf_skc_lookup_tcp,
6850 .gpl_only = false,
6851 .pkt_access = true,
6852 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6853 .arg1_type = ARG_PTR_TO_CTX,
6854 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6855 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6856 .arg4_type = ARG_ANYTHING,
6857 .arg5_type = ARG_ANYTHING,
6858 };
6859
BPF_CALL_5(bpf_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6860 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6861 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6862 {
6863 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6864 netns_id, flags);
6865 }
6866
6867 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6868 .func = bpf_sk_lookup_tcp,
6869 .gpl_only = false,
6870 .pkt_access = true,
6871 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6872 .arg1_type = ARG_PTR_TO_CTX,
6873 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6874 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6875 .arg4_type = ARG_ANYTHING,
6876 .arg5_type = ARG_ANYTHING,
6877 };
6878
BPF_CALL_5(bpf_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6879 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6880 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6881 {
6882 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6883 netns_id, flags);
6884 }
6885
6886 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6887 .func = bpf_sk_lookup_udp,
6888 .gpl_only = false,
6889 .pkt_access = true,
6890 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6891 .arg1_type = ARG_PTR_TO_CTX,
6892 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6893 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6894 .arg4_type = ARG_ANYTHING,
6895 .arg5_type = ARG_ANYTHING,
6896 };
6897
BPF_CALL_5(bpf_tc_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6898 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6899 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6900 {
6901 struct net_device *dev = skb->dev;
6902 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6903 struct net *caller_net = dev_net(dev);
6904
6905 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6906 ifindex, IPPROTO_TCP, netns_id,
6907 flags, sdif);
6908 }
6909
6910 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6911 .func = bpf_tc_skc_lookup_tcp,
6912 .gpl_only = false,
6913 .pkt_access = true,
6914 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6915 .arg1_type = ARG_PTR_TO_CTX,
6916 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6917 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6918 .arg4_type = ARG_ANYTHING,
6919 .arg5_type = ARG_ANYTHING,
6920 };
6921
BPF_CALL_5(bpf_tc_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6922 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6923 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6924 {
6925 struct net_device *dev = skb->dev;
6926 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6927 struct net *caller_net = dev_net(dev);
6928
6929 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6930 ifindex, IPPROTO_TCP, netns_id,
6931 flags, sdif);
6932 }
6933
6934 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6935 .func = bpf_tc_sk_lookup_tcp,
6936 .gpl_only = false,
6937 .pkt_access = true,
6938 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6939 .arg1_type = ARG_PTR_TO_CTX,
6940 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6941 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6942 .arg4_type = ARG_ANYTHING,
6943 .arg5_type = ARG_ANYTHING,
6944 };
6945
BPF_CALL_5(bpf_tc_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6946 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6947 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6948 {
6949 struct net_device *dev = skb->dev;
6950 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6951 struct net *caller_net = dev_net(dev);
6952
6953 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6954 ifindex, IPPROTO_UDP, netns_id,
6955 flags, sdif);
6956 }
6957
6958 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6959 .func = bpf_tc_sk_lookup_udp,
6960 .gpl_only = false,
6961 .pkt_access = true,
6962 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6963 .arg1_type = ARG_PTR_TO_CTX,
6964 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6965 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6966 .arg4_type = ARG_ANYTHING,
6967 .arg5_type = ARG_ANYTHING,
6968 };
6969
BPF_CALL_1(bpf_sk_release,struct sock *,sk)6970 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6971 {
6972 if (sk && sk_is_refcounted(sk))
6973 sock_gen_put(sk);
6974 return 0;
6975 }
6976
6977 static const struct bpf_func_proto bpf_sk_release_proto = {
6978 .func = bpf_sk_release,
6979 .gpl_only = false,
6980 .ret_type = RET_INTEGER,
6981 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6982 };
6983
BPF_CALL_5(bpf_xdp_sk_lookup_udp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6984 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6985 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6986 {
6987 struct net_device *dev = ctx->rxq->dev;
6988 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6989 struct net *caller_net = dev_net(dev);
6990
6991 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6992 ifindex, IPPROTO_UDP, netns_id,
6993 flags, sdif);
6994 }
6995
6996 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6997 .func = bpf_xdp_sk_lookup_udp,
6998 .gpl_only = false,
6999 .pkt_access = true,
7000 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7001 .arg1_type = ARG_PTR_TO_CTX,
7002 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7003 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7004 .arg4_type = ARG_ANYTHING,
7005 .arg5_type = ARG_ANYTHING,
7006 };
7007
BPF_CALL_5(bpf_xdp_skc_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7008 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
7009 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7010 {
7011 struct net_device *dev = ctx->rxq->dev;
7012 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7013 struct net *caller_net = dev_net(dev);
7014
7015 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
7016 ifindex, IPPROTO_TCP, netns_id,
7017 flags, sdif);
7018 }
7019
7020 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
7021 .func = bpf_xdp_skc_lookup_tcp,
7022 .gpl_only = false,
7023 .pkt_access = true,
7024 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7025 .arg1_type = ARG_PTR_TO_CTX,
7026 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7027 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7028 .arg4_type = ARG_ANYTHING,
7029 .arg5_type = ARG_ANYTHING,
7030 };
7031
BPF_CALL_5(bpf_xdp_sk_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7032 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
7033 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7034 {
7035 struct net_device *dev = ctx->rxq->dev;
7036 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7037 struct net *caller_net = dev_net(dev);
7038
7039 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
7040 ifindex, IPPROTO_TCP, netns_id,
7041 flags, sdif);
7042 }
7043
7044 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
7045 .func = bpf_xdp_sk_lookup_tcp,
7046 .gpl_only = false,
7047 .pkt_access = true,
7048 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7049 .arg1_type = ARG_PTR_TO_CTX,
7050 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7051 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7052 .arg4_type = ARG_ANYTHING,
7053 .arg5_type = ARG_ANYTHING,
7054 };
7055
BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7056 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7057 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7058 {
7059 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
7060 sock_net(ctx->sk), 0,
7061 IPPROTO_TCP, netns_id, flags,
7062 -1);
7063 }
7064
7065 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
7066 .func = bpf_sock_addr_skc_lookup_tcp,
7067 .gpl_only = false,
7068 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7069 .arg1_type = ARG_PTR_TO_CTX,
7070 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7071 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7072 .arg4_type = ARG_ANYTHING,
7073 .arg5_type = ARG_ANYTHING,
7074 };
7075
BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7076 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7077 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7078 {
7079 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7080 sock_net(ctx->sk), 0, IPPROTO_TCP,
7081 netns_id, flags, -1);
7082 }
7083
7084 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
7085 .func = bpf_sock_addr_sk_lookup_tcp,
7086 .gpl_only = false,
7087 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7088 .arg1_type = ARG_PTR_TO_CTX,
7089 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7090 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7091 .arg4_type = ARG_ANYTHING,
7092 .arg5_type = ARG_ANYTHING,
7093 };
7094
BPF_CALL_5(bpf_sock_addr_sk_lookup_udp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7095 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
7096 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7097 {
7098 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7099 sock_net(ctx->sk), 0, IPPROTO_UDP,
7100 netns_id, flags, -1);
7101 }
7102
7103 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
7104 .func = bpf_sock_addr_sk_lookup_udp,
7105 .gpl_only = false,
7106 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7107 .arg1_type = ARG_PTR_TO_CTX,
7108 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7109 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7110 .arg4_type = ARG_ANYTHING,
7111 .arg5_type = ARG_ANYTHING,
7112 };
7113
bpf_tcp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7114 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7115 struct bpf_insn_access_aux *info)
7116 {
7117 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
7118 icsk_retransmits))
7119 return false;
7120
7121 if (off % size != 0)
7122 return false;
7123
7124 switch (off) {
7125 case offsetof(struct bpf_tcp_sock, bytes_received):
7126 case offsetof(struct bpf_tcp_sock, bytes_acked):
7127 return size == sizeof(__u64);
7128 default:
7129 return size == sizeof(__u32);
7130 }
7131 }
7132
bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7133 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
7134 const struct bpf_insn *si,
7135 struct bpf_insn *insn_buf,
7136 struct bpf_prog *prog, u32 *target_size)
7137 {
7138 struct bpf_insn *insn = insn_buf;
7139
7140 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
7141 do { \
7142 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
7143 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7144 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7145 si->dst_reg, si->src_reg, \
7146 offsetof(struct tcp_sock, FIELD)); \
7147 } while (0)
7148
7149 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
7150 do { \
7151 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
7152 FIELD) > \
7153 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7154 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7155 struct inet_connection_sock, \
7156 FIELD), \
7157 si->dst_reg, si->src_reg, \
7158 offsetof( \
7159 struct inet_connection_sock, \
7160 FIELD)); \
7161 } while (0)
7162
7163 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7164
7165 switch (si->off) {
7166 case offsetof(struct bpf_tcp_sock, rtt_min):
7167 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7168 sizeof(struct minmax));
7169 BUILD_BUG_ON(sizeof(struct minmax) <
7170 sizeof(struct minmax_sample));
7171
7172 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7173 offsetof(struct tcp_sock, rtt_min) +
7174 offsetof(struct minmax_sample, v));
7175 break;
7176 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7177 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7178 break;
7179 case offsetof(struct bpf_tcp_sock, srtt_us):
7180 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7181 break;
7182 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7183 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7184 break;
7185 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7186 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7187 break;
7188 case offsetof(struct bpf_tcp_sock, snd_nxt):
7189 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7190 break;
7191 case offsetof(struct bpf_tcp_sock, snd_una):
7192 BPF_TCP_SOCK_GET_COMMON(snd_una);
7193 break;
7194 case offsetof(struct bpf_tcp_sock, mss_cache):
7195 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7196 break;
7197 case offsetof(struct bpf_tcp_sock, ecn_flags):
7198 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7199 break;
7200 case offsetof(struct bpf_tcp_sock, rate_delivered):
7201 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7202 break;
7203 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7204 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7205 break;
7206 case offsetof(struct bpf_tcp_sock, packets_out):
7207 BPF_TCP_SOCK_GET_COMMON(packets_out);
7208 break;
7209 case offsetof(struct bpf_tcp_sock, retrans_out):
7210 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7211 break;
7212 case offsetof(struct bpf_tcp_sock, total_retrans):
7213 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7214 break;
7215 case offsetof(struct bpf_tcp_sock, segs_in):
7216 BPF_TCP_SOCK_GET_COMMON(segs_in);
7217 break;
7218 case offsetof(struct bpf_tcp_sock, data_segs_in):
7219 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7220 break;
7221 case offsetof(struct bpf_tcp_sock, segs_out):
7222 BPF_TCP_SOCK_GET_COMMON(segs_out);
7223 break;
7224 case offsetof(struct bpf_tcp_sock, data_segs_out):
7225 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7226 break;
7227 case offsetof(struct bpf_tcp_sock, lost_out):
7228 BPF_TCP_SOCK_GET_COMMON(lost_out);
7229 break;
7230 case offsetof(struct bpf_tcp_sock, sacked_out):
7231 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7232 break;
7233 case offsetof(struct bpf_tcp_sock, bytes_received):
7234 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7235 break;
7236 case offsetof(struct bpf_tcp_sock, bytes_acked):
7237 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7238 break;
7239 case offsetof(struct bpf_tcp_sock, dsack_dups):
7240 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7241 break;
7242 case offsetof(struct bpf_tcp_sock, delivered):
7243 BPF_TCP_SOCK_GET_COMMON(delivered);
7244 break;
7245 case offsetof(struct bpf_tcp_sock, delivered_ce):
7246 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7247 break;
7248 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7249 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7250 break;
7251 }
7252
7253 return insn - insn_buf;
7254 }
7255
BPF_CALL_1(bpf_tcp_sock,struct sock *,sk)7256 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7257 {
7258 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7259 return (unsigned long)sk;
7260
7261 return (unsigned long)NULL;
7262 }
7263
7264 const struct bpf_func_proto bpf_tcp_sock_proto = {
7265 .func = bpf_tcp_sock,
7266 .gpl_only = false,
7267 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7268 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7269 };
7270
BPF_CALL_1(bpf_get_listener_sock,struct sock *,sk)7271 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7272 {
7273 sk = sk_to_full_sk(sk);
7274
7275 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7276 return (unsigned long)sk;
7277
7278 return (unsigned long)NULL;
7279 }
7280
7281 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7282 .func = bpf_get_listener_sock,
7283 .gpl_only = false,
7284 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7285 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7286 };
7287
BPF_CALL_1(bpf_skb_ecn_set_ce,struct sk_buff *,skb)7288 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7289 {
7290 unsigned int iphdr_len;
7291
7292 switch (skb_protocol(skb, true)) {
7293 case cpu_to_be16(ETH_P_IP):
7294 iphdr_len = sizeof(struct iphdr);
7295 break;
7296 case cpu_to_be16(ETH_P_IPV6):
7297 iphdr_len = sizeof(struct ipv6hdr);
7298 break;
7299 default:
7300 return 0;
7301 }
7302
7303 if (skb_headlen(skb) < iphdr_len)
7304 return 0;
7305
7306 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7307 return 0;
7308
7309 return INET_ECN_set_ce(skb);
7310 }
7311
bpf_xdp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7312 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7313 struct bpf_insn_access_aux *info)
7314 {
7315 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7316 return false;
7317
7318 if (off % size != 0)
7319 return false;
7320
7321 switch (off) {
7322 default:
7323 return size == sizeof(__u32);
7324 }
7325 }
7326
bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7327 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7328 const struct bpf_insn *si,
7329 struct bpf_insn *insn_buf,
7330 struct bpf_prog *prog, u32 *target_size)
7331 {
7332 struct bpf_insn *insn = insn_buf;
7333
7334 #define BPF_XDP_SOCK_GET(FIELD) \
7335 do { \
7336 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7337 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7338 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7339 si->dst_reg, si->src_reg, \
7340 offsetof(struct xdp_sock, FIELD)); \
7341 } while (0)
7342
7343 switch (si->off) {
7344 case offsetof(struct bpf_xdp_sock, queue_id):
7345 BPF_XDP_SOCK_GET(queue_id);
7346 break;
7347 }
7348
7349 return insn - insn_buf;
7350 }
7351
7352 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7353 .func = bpf_skb_ecn_set_ce,
7354 .gpl_only = false,
7355 .ret_type = RET_INTEGER,
7356 .arg1_type = ARG_PTR_TO_CTX,
7357 };
7358
BPF_CALL_5(bpf_tcp_check_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7359 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7360 struct tcphdr *, th, u32, th_len)
7361 {
7362 #ifdef CONFIG_SYN_COOKIES
7363 int ret;
7364
7365 if (unlikely(!sk || th_len < sizeof(*th)))
7366 return -EINVAL;
7367
7368 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7369 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7370 return -EINVAL;
7371
7372 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7373 return -EINVAL;
7374
7375 if (!th->ack || th->rst || th->syn)
7376 return -ENOENT;
7377
7378 if (unlikely(iph_len < sizeof(struct iphdr)))
7379 return -EINVAL;
7380
7381 if (tcp_synq_no_recent_overflow(sk))
7382 return -ENOENT;
7383
7384 /* Both struct iphdr and struct ipv6hdr have the version field at the
7385 * same offset so we can cast to the shorter header (struct iphdr).
7386 */
7387 switch (((struct iphdr *)iph)->version) {
7388 case 4:
7389 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7390 return -EINVAL;
7391
7392 ret = __cookie_v4_check((struct iphdr *)iph, th);
7393 break;
7394
7395 #if IS_BUILTIN(CONFIG_IPV6)
7396 case 6:
7397 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7398 return -EINVAL;
7399
7400 if (sk->sk_family != AF_INET6)
7401 return -EINVAL;
7402
7403 ret = __cookie_v6_check((struct ipv6hdr *)iph, th);
7404 break;
7405 #endif /* CONFIG_IPV6 */
7406
7407 default:
7408 return -EPROTONOSUPPORT;
7409 }
7410
7411 if (ret > 0)
7412 return 0;
7413
7414 return -ENOENT;
7415 #else
7416 return -ENOTSUPP;
7417 #endif
7418 }
7419
7420 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7421 .func = bpf_tcp_check_syncookie,
7422 .gpl_only = true,
7423 .pkt_access = true,
7424 .ret_type = RET_INTEGER,
7425 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7426 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7427 .arg3_type = ARG_CONST_SIZE,
7428 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7429 .arg5_type = ARG_CONST_SIZE,
7430 };
7431
BPF_CALL_5(bpf_tcp_gen_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7432 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7433 struct tcphdr *, th, u32, th_len)
7434 {
7435 #ifdef CONFIG_SYN_COOKIES
7436 u32 cookie;
7437 u16 mss;
7438
7439 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7440 return -EINVAL;
7441
7442 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7443 return -EINVAL;
7444
7445 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7446 return -ENOENT;
7447
7448 if (!th->syn || th->ack || th->fin || th->rst)
7449 return -EINVAL;
7450
7451 if (unlikely(iph_len < sizeof(struct iphdr)))
7452 return -EINVAL;
7453
7454 /* Both struct iphdr and struct ipv6hdr have the version field at the
7455 * same offset so we can cast to the shorter header (struct iphdr).
7456 */
7457 switch (((struct iphdr *)iph)->version) {
7458 case 4:
7459 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7460 return -EINVAL;
7461
7462 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7463 break;
7464
7465 #if IS_BUILTIN(CONFIG_IPV6)
7466 case 6:
7467 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7468 return -EINVAL;
7469
7470 if (sk->sk_family != AF_INET6)
7471 return -EINVAL;
7472
7473 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7474 break;
7475 #endif /* CONFIG_IPV6 */
7476
7477 default:
7478 return -EPROTONOSUPPORT;
7479 }
7480 if (mss == 0)
7481 return -ENOENT;
7482
7483 return cookie | ((u64)mss << 32);
7484 #else
7485 return -EOPNOTSUPP;
7486 #endif /* CONFIG_SYN_COOKIES */
7487 }
7488
7489 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7490 .func = bpf_tcp_gen_syncookie,
7491 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7492 .pkt_access = true,
7493 .ret_type = RET_INTEGER,
7494 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7495 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7496 .arg3_type = ARG_CONST_SIZE,
7497 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7498 .arg5_type = ARG_CONST_SIZE,
7499 };
7500
BPF_CALL_3(bpf_sk_assign,struct sk_buff *,skb,struct sock *,sk,u64,flags)7501 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7502 {
7503 if (!sk || flags != 0)
7504 return -EINVAL;
7505 if (!skb_at_tc_ingress(skb))
7506 return -EOPNOTSUPP;
7507 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7508 return -ENETUNREACH;
7509 if (sk_unhashed(sk))
7510 return -EOPNOTSUPP;
7511 if (sk_is_refcounted(sk) &&
7512 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7513 return -ENOENT;
7514
7515 skb_orphan(skb);
7516 skb->sk = sk;
7517 skb->destructor = sock_pfree;
7518
7519 return 0;
7520 }
7521
7522 static const struct bpf_func_proto bpf_sk_assign_proto = {
7523 .func = bpf_sk_assign,
7524 .gpl_only = false,
7525 .ret_type = RET_INTEGER,
7526 .arg1_type = ARG_PTR_TO_CTX,
7527 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7528 .arg3_type = ARG_ANYTHING,
7529 };
7530
bpf_search_tcp_opt(const u8 * op,const u8 * opend,u8 search_kind,const u8 * magic,u8 magic_len,bool * eol)7531 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7532 u8 search_kind, const u8 *magic,
7533 u8 magic_len, bool *eol)
7534 {
7535 u8 kind, kind_len;
7536
7537 *eol = false;
7538
7539 while (op < opend) {
7540 kind = op[0];
7541
7542 if (kind == TCPOPT_EOL) {
7543 *eol = true;
7544 return ERR_PTR(-ENOMSG);
7545 } else if (kind == TCPOPT_NOP) {
7546 op++;
7547 continue;
7548 }
7549
7550 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7551 /* Something is wrong in the received header.
7552 * Follow the TCP stack's tcp_parse_options()
7553 * and just bail here.
7554 */
7555 return ERR_PTR(-EFAULT);
7556
7557 kind_len = op[1];
7558 if (search_kind == kind) {
7559 if (!magic_len)
7560 return op;
7561
7562 if (magic_len > kind_len - 2)
7563 return ERR_PTR(-ENOMSG);
7564
7565 if (!memcmp(&op[2], magic, magic_len))
7566 return op;
7567 }
7568
7569 op += kind_len;
7570 }
7571
7572 return ERR_PTR(-ENOMSG);
7573 }
7574
BPF_CALL_4(bpf_sock_ops_load_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,void *,search_res,u32,len,u64,flags)7575 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7576 void *, search_res, u32, len, u64, flags)
7577 {
7578 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7579 const u8 *op, *opend, *magic, *search = search_res;
7580 u8 search_kind, search_len, copy_len, magic_len;
7581 int ret;
7582
7583 /* 2 byte is the minimal option len except TCPOPT_NOP and
7584 * TCPOPT_EOL which are useless for the bpf prog to learn
7585 * and this helper disallow loading them also.
7586 */
7587 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7588 return -EINVAL;
7589
7590 search_kind = search[0];
7591 search_len = search[1];
7592
7593 if (search_len > len || search_kind == TCPOPT_NOP ||
7594 search_kind == TCPOPT_EOL)
7595 return -EINVAL;
7596
7597 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7598 /* 16 or 32 bit magic. +2 for kind and kind length */
7599 if (search_len != 4 && search_len != 6)
7600 return -EINVAL;
7601 magic = &search[2];
7602 magic_len = search_len - 2;
7603 } else {
7604 if (search_len)
7605 return -EINVAL;
7606 magic = NULL;
7607 magic_len = 0;
7608 }
7609
7610 if (load_syn) {
7611 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7612 if (ret < 0)
7613 return ret;
7614
7615 opend = op + ret;
7616 op += sizeof(struct tcphdr);
7617 } else {
7618 if (!bpf_sock->skb ||
7619 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7620 /* This bpf_sock->op cannot call this helper */
7621 return -EPERM;
7622
7623 opend = bpf_sock->skb_data_end;
7624 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7625 }
7626
7627 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7628 &eol);
7629 if (IS_ERR(op))
7630 return PTR_ERR(op);
7631
7632 copy_len = op[1];
7633 ret = copy_len;
7634 if (copy_len > len) {
7635 ret = -ENOSPC;
7636 copy_len = len;
7637 }
7638
7639 memcpy(search_res, op, copy_len);
7640 return ret;
7641 }
7642
7643 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7644 .func = bpf_sock_ops_load_hdr_opt,
7645 .gpl_only = false,
7646 .ret_type = RET_INTEGER,
7647 .arg1_type = ARG_PTR_TO_CTX,
7648 .arg2_type = ARG_PTR_TO_MEM,
7649 .arg3_type = ARG_CONST_SIZE,
7650 .arg4_type = ARG_ANYTHING,
7651 };
7652
BPF_CALL_4(bpf_sock_ops_store_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,const void *,from,u32,len,u64,flags)7653 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7654 const void *, from, u32, len, u64, flags)
7655 {
7656 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7657 const u8 *op, *new_op, *magic = NULL;
7658 struct sk_buff *skb;
7659 bool eol;
7660
7661 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7662 return -EPERM;
7663
7664 if (len < 2 || flags)
7665 return -EINVAL;
7666
7667 new_op = from;
7668 new_kind = new_op[0];
7669 new_kind_len = new_op[1];
7670
7671 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7672 new_kind == TCPOPT_EOL)
7673 return -EINVAL;
7674
7675 if (new_kind_len > bpf_sock->remaining_opt_len)
7676 return -ENOSPC;
7677
7678 /* 253 is another experimental kind */
7679 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7680 if (new_kind_len < 4)
7681 return -EINVAL;
7682 /* Match for the 2 byte magic also.
7683 * RFC 6994: the magic could be 2 or 4 bytes.
7684 * Hence, matching by 2 byte only is on the
7685 * conservative side but it is the right
7686 * thing to do for the 'search-for-duplication'
7687 * purpose.
7688 */
7689 magic = &new_op[2];
7690 magic_len = 2;
7691 }
7692
7693 /* Check for duplication */
7694 skb = bpf_sock->skb;
7695 op = skb->data + sizeof(struct tcphdr);
7696 opend = bpf_sock->skb_data_end;
7697
7698 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7699 &eol);
7700 if (!IS_ERR(op))
7701 return -EEXIST;
7702
7703 if (PTR_ERR(op) != -ENOMSG)
7704 return PTR_ERR(op);
7705
7706 if (eol)
7707 /* The option has been ended. Treat it as no more
7708 * header option can be written.
7709 */
7710 return -ENOSPC;
7711
7712 /* No duplication found. Store the header option. */
7713 memcpy(opend, from, new_kind_len);
7714
7715 bpf_sock->remaining_opt_len -= new_kind_len;
7716 bpf_sock->skb_data_end += new_kind_len;
7717
7718 return 0;
7719 }
7720
7721 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7722 .func = bpf_sock_ops_store_hdr_opt,
7723 .gpl_only = false,
7724 .ret_type = RET_INTEGER,
7725 .arg1_type = ARG_PTR_TO_CTX,
7726 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7727 .arg3_type = ARG_CONST_SIZE,
7728 .arg4_type = ARG_ANYTHING,
7729 };
7730
BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,u32,len,u64,flags)7731 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7732 u32, len, u64, flags)
7733 {
7734 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7735 return -EPERM;
7736
7737 if (flags || len < 2)
7738 return -EINVAL;
7739
7740 if (len > bpf_sock->remaining_opt_len)
7741 return -ENOSPC;
7742
7743 bpf_sock->remaining_opt_len -= len;
7744
7745 return 0;
7746 }
7747
7748 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7749 .func = bpf_sock_ops_reserve_hdr_opt,
7750 .gpl_only = false,
7751 .ret_type = RET_INTEGER,
7752 .arg1_type = ARG_PTR_TO_CTX,
7753 .arg2_type = ARG_ANYTHING,
7754 .arg3_type = ARG_ANYTHING,
7755 };
7756
BPF_CALL_3(bpf_skb_set_tstamp,struct sk_buff *,skb,u64,tstamp,u32,tstamp_type)7757 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7758 u64, tstamp, u32, tstamp_type)
7759 {
7760 /* skb_clear_delivery_time() is done for inet protocol */
7761 if (skb->protocol != htons(ETH_P_IP) &&
7762 skb->protocol != htons(ETH_P_IPV6))
7763 return -EOPNOTSUPP;
7764
7765 switch (tstamp_type) {
7766 case BPF_SKB_CLOCK_REALTIME:
7767 skb->tstamp = tstamp;
7768 skb->tstamp_type = SKB_CLOCK_REALTIME;
7769 break;
7770 case BPF_SKB_CLOCK_MONOTONIC:
7771 if (!tstamp)
7772 return -EINVAL;
7773 skb->tstamp = tstamp;
7774 skb->tstamp_type = SKB_CLOCK_MONOTONIC;
7775 break;
7776 case BPF_SKB_CLOCK_TAI:
7777 if (!tstamp)
7778 return -EINVAL;
7779 skb->tstamp = tstamp;
7780 skb->tstamp_type = SKB_CLOCK_TAI;
7781 break;
7782 default:
7783 return -EINVAL;
7784 }
7785
7786 return 0;
7787 }
7788
7789 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7790 .func = bpf_skb_set_tstamp,
7791 .gpl_only = false,
7792 .ret_type = RET_INTEGER,
7793 .arg1_type = ARG_PTR_TO_CTX,
7794 .arg2_type = ARG_ANYTHING,
7795 .arg3_type = ARG_ANYTHING,
7796 };
7797
7798 #ifdef CONFIG_SYN_COOKIES
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th,u32,th_len)7799 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7800 struct tcphdr *, th, u32, th_len)
7801 {
7802 u32 cookie;
7803 u16 mss;
7804
7805 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7806 return -EINVAL;
7807
7808 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7809 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7810
7811 return cookie | ((u64)mss << 32);
7812 }
7813
7814 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7815 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7816 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7817 .pkt_access = true,
7818 .ret_type = RET_INTEGER,
7819 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7820 .arg1_size = sizeof(struct iphdr),
7821 .arg2_type = ARG_PTR_TO_MEM,
7822 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7823 };
7824
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th,u32,th_len)7825 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7826 struct tcphdr *, th, u32, th_len)
7827 {
7828 #if IS_BUILTIN(CONFIG_IPV6)
7829 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7830 sizeof(struct ipv6hdr);
7831 u32 cookie;
7832 u16 mss;
7833
7834 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7835 return -EINVAL;
7836
7837 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7838 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7839
7840 return cookie | ((u64)mss << 32);
7841 #else
7842 return -EPROTONOSUPPORT;
7843 #endif
7844 }
7845
7846 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7847 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7848 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7849 .pkt_access = true,
7850 .ret_type = RET_INTEGER,
7851 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7852 .arg1_size = sizeof(struct ipv6hdr),
7853 .arg2_type = ARG_PTR_TO_MEM,
7854 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7855 };
7856
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th)7857 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7858 struct tcphdr *, th)
7859 {
7860 if (__cookie_v4_check(iph, th) > 0)
7861 return 0;
7862
7863 return -EACCES;
7864 }
7865
7866 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7867 .func = bpf_tcp_raw_check_syncookie_ipv4,
7868 .gpl_only = true, /* __cookie_v4_check is GPL */
7869 .pkt_access = true,
7870 .ret_type = RET_INTEGER,
7871 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7872 .arg1_size = sizeof(struct iphdr),
7873 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7874 .arg2_size = sizeof(struct tcphdr),
7875 };
7876
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th)7877 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7878 struct tcphdr *, th)
7879 {
7880 #if IS_BUILTIN(CONFIG_IPV6)
7881 if (__cookie_v6_check(iph, th) > 0)
7882 return 0;
7883
7884 return -EACCES;
7885 #else
7886 return -EPROTONOSUPPORT;
7887 #endif
7888 }
7889
7890 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7891 .func = bpf_tcp_raw_check_syncookie_ipv6,
7892 .gpl_only = true, /* __cookie_v6_check is GPL */
7893 .pkt_access = true,
7894 .ret_type = RET_INTEGER,
7895 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7896 .arg1_size = sizeof(struct ipv6hdr),
7897 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7898 .arg2_size = sizeof(struct tcphdr),
7899 };
7900 #endif /* CONFIG_SYN_COOKIES */
7901
7902 #endif /* CONFIG_INET */
7903
bpf_helper_changes_pkt_data(void * func)7904 bool bpf_helper_changes_pkt_data(void *func)
7905 {
7906 if (func == bpf_skb_vlan_push ||
7907 func == bpf_skb_vlan_pop ||
7908 func == bpf_skb_store_bytes ||
7909 func == bpf_skb_change_proto ||
7910 func == bpf_skb_change_head ||
7911 func == sk_skb_change_head ||
7912 func == bpf_skb_change_tail ||
7913 func == sk_skb_change_tail ||
7914 func == bpf_skb_adjust_room ||
7915 func == sk_skb_adjust_room ||
7916 func == bpf_skb_pull_data ||
7917 func == sk_skb_pull_data ||
7918 func == bpf_clone_redirect ||
7919 func == bpf_l3_csum_replace ||
7920 func == bpf_l4_csum_replace ||
7921 func == bpf_xdp_adjust_head ||
7922 func == bpf_xdp_adjust_meta ||
7923 func == bpf_msg_pull_data ||
7924 func == bpf_msg_push_data ||
7925 func == bpf_msg_pop_data ||
7926 func == bpf_xdp_adjust_tail ||
7927 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7928 func == bpf_lwt_seg6_store_bytes ||
7929 func == bpf_lwt_seg6_adjust_srh ||
7930 func == bpf_lwt_seg6_action ||
7931 #endif
7932 #ifdef CONFIG_INET
7933 func == bpf_sock_ops_store_hdr_opt ||
7934 #endif
7935 func == bpf_lwt_in_push_encap ||
7936 func == bpf_lwt_xmit_push_encap)
7937 return true;
7938
7939 return false;
7940 }
7941
7942 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7943 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7944
7945 static const struct bpf_func_proto *
sock_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7946 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7947 {
7948 const struct bpf_func_proto *func_proto;
7949
7950 func_proto = cgroup_common_func_proto(func_id, prog);
7951 if (func_proto)
7952 return func_proto;
7953
7954 func_proto = cgroup_current_func_proto(func_id, prog);
7955 if (func_proto)
7956 return func_proto;
7957
7958 switch (func_id) {
7959 case BPF_FUNC_get_socket_cookie:
7960 return &bpf_get_socket_cookie_sock_proto;
7961 case BPF_FUNC_get_netns_cookie:
7962 return &bpf_get_netns_cookie_sock_proto;
7963 case BPF_FUNC_perf_event_output:
7964 return &bpf_event_output_data_proto;
7965 case BPF_FUNC_sk_storage_get:
7966 return &bpf_sk_storage_get_cg_sock_proto;
7967 case BPF_FUNC_ktime_get_coarse_ns:
7968 return &bpf_ktime_get_coarse_ns_proto;
7969 default:
7970 return bpf_base_func_proto(func_id, prog);
7971 }
7972 }
7973
7974 static const struct bpf_func_proto *
sock_addr_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7975 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7976 {
7977 const struct bpf_func_proto *func_proto;
7978
7979 func_proto = cgroup_common_func_proto(func_id, prog);
7980 if (func_proto)
7981 return func_proto;
7982
7983 func_proto = cgroup_current_func_proto(func_id, prog);
7984 if (func_proto)
7985 return func_proto;
7986
7987 switch (func_id) {
7988 case BPF_FUNC_bind:
7989 switch (prog->expected_attach_type) {
7990 case BPF_CGROUP_INET4_CONNECT:
7991 case BPF_CGROUP_INET6_CONNECT:
7992 return &bpf_bind_proto;
7993 default:
7994 return NULL;
7995 }
7996 case BPF_FUNC_get_socket_cookie:
7997 return &bpf_get_socket_cookie_sock_addr_proto;
7998 case BPF_FUNC_get_netns_cookie:
7999 return &bpf_get_netns_cookie_sock_addr_proto;
8000 case BPF_FUNC_perf_event_output:
8001 return &bpf_event_output_data_proto;
8002 #ifdef CONFIG_INET
8003 case BPF_FUNC_sk_lookup_tcp:
8004 return &bpf_sock_addr_sk_lookup_tcp_proto;
8005 case BPF_FUNC_sk_lookup_udp:
8006 return &bpf_sock_addr_sk_lookup_udp_proto;
8007 case BPF_FUNC_sk_release:
8008 return &bpf_sk_release_proto;
8009 case BPF_FUNC_skc_lookup_tcp:
8010 return &bpf_sock_addr_skc_lookup_tcp_proto;
8011 #endif /* CONFIG_INET */
8012 case BPF_FUNC_sk_storage_get:
8013 return &bpf_sk_storage_get_proto;
8014 case BPF_FUNC_sk_storage_delete:
8015 return &bpf_sk_storage_delete_proto;
8016 case BPF_FUNC_setsockopt:
8017 switch (prog->expected_attach_type) {
8018 case BPF_CGROUP_INET4_BIND:
8019 case BPF_CGROUP_INET6_BIND:
8020 case BPF_CGROUP_INET4_CONNECT:
8021 case BPF_CGROUP_INET6_CONNECT:
8022 case BPF_CGROUP_UNIX_CONNECT:
8023 case BPF_CGROUP_UDP4_RECVMSG:
8024 case BPF_CGROUP_UDP6_RECVMSG:
8025 case BPF_CGROUP_UNIX_RECVMSG:
8026 case BPF_CGROUP_UDP4_SENDMSG:
8027 case BPF_CGROUP_UDP6_SENDMSG:
8028 case BPF_CGROUP_UNIX_SENDMSG:
8029 case BPF_CGROUP_INET4_GETPEERNAME:
8030 case BPF_CGROUP_INET6_GETPEERNAME:
8031 case BPF_CGROUP_UNIX_GETPEERNAME:
8032 case BPF_CGROUP_INET4_GETSOCKNAME:
8033 case BPF_CGROUP_INET6_GETSOCKNAME:
8034 case BPF_CGROUP_UNIX_GETSOCKNAME:
8035 return &bpf_sock_addr_setsockopt_proto;
8036 default:
8037 return NULL;
8038 }
8039 case BPF_FUNC_getsockopt:
8040 switch (prog->expected_attach_type) {
8041 case BPF_CGROUP_INET4_BIND:
8042 case BPF_CGROUP_INET6_BIND:
8043 case BPF_CGROUP_INET4_CONNECT:
8044 case BPF_CGROUP_INET6_CONNECT:
8045 case BPF_CGROUP_UNIX_CONNECT:
8046 case BPF_CGROUP_UDP4_RECVMSG:
8047 case BPF_CGROUP_UDP6_RECVMSG:
8048 case BPF_CGROUP_UNIX_RECVMSG:
8049 case BPF_CGROUP_UDP4_SENDMSG:
8050 case BPF_CGROUP_UDP6_SENDMSG:
8051 case BPF_CGROUP_UNIX_SENDMSG:
8052 case BPF_CGROUP_INET4_GETPEERNAME:
8053 case BPF_CGROUP_INET6_GETPEERNAME:
8054 case BPF_CGROUP_UNIX_GETPEERNAME:
8055 case BPF_CGROUP_INET4_GETSOCKNAME:
8056 case BPF_CGROUP_INET6_GETSOCKNAME:
8057 case BPF_CGROUP_UNIX_GETSOCKNAME:
8058 return &bpf_sock_addr_getsockopt_proto;
8059 default:
8060 return NULL;
8061 }
8062 default:
8063 return bpf_sk_base_func_proto(func_id, prog);
8064 }
8065 }
8066
8067 static const struct bpf_func_proto *
sk_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8068 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8069 {
8070 switch (func_id) {
8071 case BPF_FUNC_skb_load_bytes:
8072 return &bpf_skb_load_bytes_proto;
8073 case BPF_FUNC_skb_load_bytes_relative:
8074 return &bpf_skb_load_bytes_relative_proto;
8075 case BPF_FUNC_get_socket_cookie:
8076 return &bpf_get_socket_cookie_proto;
8077 case BPF_FUNC_get_socket_uid:
8078 return &bpf_get_socket_uid_proto;
8079 case BPF_FUNC_perf_event_output:
8080 return &bpf_skb_event_output_proto;
8081 default:
8082 return bpf_sk_base_func_proto(func_id, prog);
8083 }
8084 }
8085
8086 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
8087 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
8088
8089 static const struct bpf_func_proto *
cg_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8090 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8091 {
8092 const struct bpf_func_proto *func_proto;
8093
8094 func_proto = cgroup_common_func_proto(func_id, prog);
8095 if (func_proto)
8096 return func_proto;
8097
8098 switch (func_id) {
8099 case BPF_FUNC_sk_fullsock:
8100 return &bpf_sk_fullsock_proto;
8101 case BPF_FUNC_sk_storage_get:
8102 return &bpf_sk_storage_get_proto;
8103 case BPF_FUNC_sk_storage_delete:
8104 return &bpf_sk_storage_delete_proto;
8105 case BPF_FUNC_perf_event_output:
8106 return &bpf_skb_event_output_proto;
8107 #ifdef CONFIG_SOCK_CGROUP_DATA
8108 case BPF_FUNC_skb_cgroup_id:
8109 return &bpf_skb_cgroup_id_proto;
8110 case BPF_FUNC_skb_ancestor_cgroup_id:
8111 return &bpf_skb_ancestor_cgroup_id_proto;
8112 case BPF_FUNC_sk_cgroup_id:
8113 return &bpf_sk_cgroup_id_proto;
8114 case BPF_FUNC_sk_ancestor_cgroup_id:
8115 return &bpf_sk_ancestor_cgroup_id_proto;
8116 #endif
8117 #ifdef CONFIG_INET
8118 case BPF_FUNC_sk_lookup_tcp:
8119 return &bpf_sk_lookup_tcp_proto;
8120 case BPF_FUNC_sk_lookup_udp:
8121 return &bpf_sk_lookup_udp_proto;
8122 case BPF_FUNC_sk_release:
8123 return &bpf_sk_release_proto;
8124 case BPF_FUNC_skc_lookup_tcp:
8125 return &bpf_skc_lookup_tcp_proto;
8126 case BPF_FUNC_tcp_sock:
8127 return &bpf_tcp_sock_proto;
8128 case BPF_FUNC_get_listener_sock:
8129 return &bpf_get_listener_sock_proto;
8130 case BPF_FUNC_skb_ecn_set_ce:
8131 return &bpf_skb_ecn_set_ce_proto;
8132 #endif
8133 default:
8134 return sk_filter_func_proto(func_id, prog);
8135 }
8136 }
8137
8138 static const struct bpf_func_proto *
tc_cls_act_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8139 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8140 {
8141 switch (func_id) {
8142 case BPF_FUNC_skb_store_bytes:
8143 return &bpf_skb_store_bytes_proto;
8144 case BPF_FUNC_skb_load_bytes:
8145 return &bpf_skb_load_bytes_proto;
8146 case BPF_FUNC_skb_load_bytes_relative:
8147 return &bpf_skb_load_bytes_relative_proto;
8148 case BPF_FUNC_skb_pull_data:
8149 return &bpf_skb_pull_data_proto;
8150 case BPF_FUNC_csum_diff:
8151 return &bpf_csum_diff_proto;
8152 case BPF_FUNC_csum_update:
8153 return &bpf_csum_update_proto;
8154 case BPF_FUNC_csum_level:
8155 return &bpf_csum_level_proto;
8156 case BPF_FUNC_l3_csum_replace:
8157 return &bpf_l3_csum_replace_proto;
8158 case BPF_FUNC_l4_csum_replace:
8159 return &bpf_l4_csum_replace_proto;
8160 case BPF_FUNC_clone_redirect:
8161 return &bpf_clone_redirect_proto;
8162 case BPF_FUNC_get_cgroup_classid:
8163 return &bpf_get_cgroup_classid_proto;
8164 case BPF_FUNC_skb_vlan_push:
8165 return &bpf_skb_vlan_push_proto;
8166 case BPF_FUNC_skb_vlan_pop:
8167 return &bpf_skb_vlan_pop_proto;
8168 case BPF_FUNC_skb_change_proto:
8169 return &bpf_skb_change_proto_proto;
8170 case BPF_FUNC_skb_change_type:
8171 return &bpf_skb_change_type_proto;
8172 case BPF_FUNC_skb_adjust_room:
8173 return &bpf_skb_adjust_room_proto;
8174 case BPF_FUNC_skb_change_tail:
8175 return &bpf_skb_change_tail_proto;
8176 case BPF_FUNC_skb_change_head:
8177 return &bpf_skb_change_head_proto;
8178 case BPF_FUNC_skb_get_tunnel_key:
8179 return &bpf_skb_get_tunnel_key_proto;
8180 case BPF_FUNC_skb_set_tunnel_key:
8181 return bpf_get_skb_set_tunnel_proto(func_id);
8182 case BPF_FUNC_skb_get_tunnel_opt:
8183 return &bpf_skb_get_tunnel_opt_proto;
8184 case BPF_FUNC_skb_set_tunnel_opt:
8185 return bpf_get_skb_set_tunnel_proto(func_id);
8186 case BPF_FUNC_redirect:
8187 return &bpf_redirect_proto;
8188 case BPF_FUNC_redirect_neigh:
8189 return &bpf_redirect_neigh_proto;
8190 case BPF_FUNC_redirect_peer:
8191 return &bpf_redirect_peer_proto;
8192 case BPF_FUNC_get_route_realm:
8193 return &bpf_get_route_realm_proto;
8194 case BPF_FUNC_get_hash_recalc:
8195 return &bpf_get_hash_recalc_proto;
8196 case BPF_FUNC_set_hash_invalid:
8197 return &bpf_set_hash_invalid_proto;
8198 case BPF_FUNC_set_hash:
8199 return &bpf_set_hash_proto;
8200 case BPF_FUNC_perf_event_output:
8201 return &bpf_skb_event_output_proto;
8202 case BPF_FUNC_get_smp_processor_id:
8203 return &bpf_get_smp_processor_id_proto;
8204 case BPF_FUNC_skb_under_cgroup:
8205 return &bpf_skb_under_cgroup_proto;
8206 case BPF_FUNC_get_socket_cookie:
8207 return &bpf_get_socket_cookie_proto;
8208 case BPF_FUNC_get_socket_uid:
8209 return &bpf_get_socket_uid_proto;
8210 case BPF_FUNC_fib_lookup:
8211 return &bpf_skb_fib_lookup_proto;
8212 case BPF_FUNC_check_mtu:
8213 return &bpf_skb_check_mtu_proto;
8214 case BPF_FUNC_sk_fullsock:
8215 return &bpf_sk_fullsock_proto;
8216 case BPF_FUNC_sk_storage_get:
8217 return &bpf_sk_storage_get_proto;
8218 case BPF_FUNC_sk_storage_delete:
8219 return &bpf_sk_storage_delete_proto;
8220 #ifdef CONFIG_XFRM
8221 case BPF_FUNC_skb_get_xfrm_state:
8222 return &bpf_skb_get_xfrm_state_proto;
8223 #endif
8224 #ifdef CONFIG_CGROUP_NET_CLASSID
8225 case BPF_FUNC_skb_cgroup_classid:
8226 return &bpf_skb_cgroup_classid_proto;
8227 #endif
8228 #ifdef CONFIG_SOCK_CGROUP_DATA
8229 case BPF_FUNC_skb_cgroup_id:
8230 return &bpf_skb_cgroup_id_proto;
8231 case BPF_FUNC_skb_ancestor_cgroup_id:
8232 return &bpf_skb_ancestor_cgroup_id_proto;
8233 #endif
8234 #ifdef CONFIG_INET
8235 case BPF_FUNC_sk_lookup_tcp:
8236 return &bpf_tc_sk_lookup_tcp_proto;
8237 case BPF_FUNC_sk_lookup_udp:
8238 return &bpf_tc_sk_lookup_udp_proto;
8239 case BPF_FUNC_sk_release:
8240 return &bpf_sk_release_proto;
8241 case BPF_FUNC_tcp_sock:
8242 return &bpf_tcp_sock_proto;
8243 case BPF_FUNC_get_listener_sock:
8244 return &bpf_get_listener_sock_proto;
8245 case BPF_FUNC_skc_lookup_tcp:
8246 return &bpf_tc_skc_lookup_tcp_proto;
8247 case BPF_FUNC_tcp_check_syncookie:
8248 return &bpf_tcp_check_syncookie_proto;
8249 case BPF_FUNC_skb_ecn_set_ce:
8250 return &bpf_skb_ecn_set_ce_proto;
8251 case BPF_FUNC_tcp_gen_syncookie:
8252 return &bpf_tcp_gen_syncookie_proto;
8253 case BPF_FUNC_sk_assign:
8254 return &bpf_sk_assign_proto;
8255 case BPF_FUNC_skb_set_tstamp:
8256 return &bpf_skb_set_tstamp_proto;
8257 #ifdef CONFIG_SYN_COOKIES
8258 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8259 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8260 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8261 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8262 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8263 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8264 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8265 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8266 #endif
8267 #endif
8268 default:
8269 return bpf_sk_base_func_proto(func_id, prog);
8270 }
8271 }
8272
8273 static const struct bpf_func_proto *
xdp_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8274 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8275 {
8276 switch (func_id) {
8277 case BPF_FUNC_perf_event_output:
8278 return &bpf_xdp_event_output_proto;
8279 case BPF_FUNC_get_smp_processor_id:
8280 return &bpf_get_smp_processor_id_proto;
8281 case BPF_FUNC_csum_diff:
8282 return &bpf_csum_diff_proto;
8283 case BPF_FUNC_xdp_adjust_head:
8284 return &bpf_xdp_adjust_head_proto;
8285 case BPF_FUNC_xdp_adjust_meta:
8286 return &bpf_xdp_adjust_meta_proto;
8287 case BPF_FUNC_redirect:
8288 return &bpf_xdp_redirect_proto;
8289 case BPF_FUNC_redirect_map:
8290 return &bpf_xdp_redirect_map_proto;
8291 case BPF_FUNC_xdp_adjust_tail:
8292 return &bpf_xdp_adjust_tail_proto;
8293 case BPF_FUNC_xdp_get_buff_len:
8294 return &bpf_xdp_get_buff_len_proto;
8295 case BPF_FUNC_xdp_load_bytes:
8296 return &bpf_xdp_load_bytes_proto;
8297 case BPF_FUNC_xdp_store_bytes:
8298 return &bpf_xdp_store_bytes_proto;
8299 case BPF_FUNC_fib_lookup:
8300 return &bpf_xdp_fib_lookup_proto;
8301 case BPF_FUNC_check_mtu:
8302 return &bpf_xdp_check_mtu_proto;
8303 #ifdef CONFIG_INET
8304 case BPF_FUNC_sk_lookup_udp:
8305 return &bpf_xdp_sk_lookup_udp_proto;
8306 case BPF_FUNC_sk_lookup_tcp:
8307 return &bpf_xdp_sk_lookup_tcp_proto;
8308 case BPF_FUNC_sk_release:
8309 return &bpf_sk_release_proto;
8310 case BPF_FUNC_skc_lookup_tcp:
8311 return &bpf_xdp_skc_lookup_tcp_proto;
8312 case BPF_FUNC_tcp_check_syncookie:
8313 return &bpf_tcp_check_syncookie_proto;
8314 case BPF_FUNC_tcp_gen_syncookie:
8315 return &bpf_tcp_gen_syncookie_proto;
8316 #ifdef CONFIG_SYN_COOKIES
8317 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8318 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8319 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8320 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8321 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8322 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8323 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8324 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8325 #endif
8326 #endif
8327 default:
8328 return bpf_sk_base_func_proto(func_id, prog);
8329 }
8330
8331 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8332 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8333 * kfuncs are defined in two different modules, and we want to be able
8334 * to use them interchangeably with the same BTF type ID. Because modules
8335 * can't de-duplicate BTF IDs between each other, we need the type to be
8336 * referenced in the vmlinux BTF or the verifier will get confused about
8337 * the different types. So we add this dummy type reference which will
8338 * be included in vmlinux BTF, allowing both modules to refer to the
8339 * same type ID.
8340 */
8341 BTF_TYPE_EMIT(struct nf_conn___init);
8342 #endif
8343 }
8344
8345 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8346 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8347
8348 static const struct bpf_func_proto *
sock_ops_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8349 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8350 {
8351 const struct bpf_func_proto *func_proto;
8352
8353 func_proto = cgroup_common_func_proto(func_id, prog);
8354 if (func_proto)
8355 return func_proto;
8356
8357 switch (func_id) {
8358 case BPF_FUNC_setsockopt:
8359 return &bpf_sock_ops_setsockopt_proto;
8360 case BPF_FUNC_getsockopt:
8361 return &bpf_sock_ops_getsockopt_proto;
8362 case BPF_FUNC_sock_ops_cb_flags_set:
8363 return &bpf_sock_ops_cb_flags_set_proto;
8364 case BPF_FUNC_sock_map_update:
8365 return &bpf_sock_map_update_proto;
8366 case BPF_FUNC_sock_hash_update:
8367 return &bpf_sock_hash_update_proto;
8368 case BPF_FUNC_get_socket_cookie:
8369 return &bpf_get_socket_cookie_sock_ops_proto;
8370 case BPF_FUNC_perf_event_output:
8371 return &bpf_event_output_data_proto;
8372 case BPF_FUNC_sk_storage_get:
8373 return &bpf_sk_storage_get_proto;
8374 case BPF_FUNC_sk_storage_delete:
8375 return &bpf_sk_storage_delete_proto;
8376 case BPF_FUNC_get_netns_cookie:
8377 return &bpf_get_netns_cookie_sock_ops_proto;
8378 #ifdef CONFIG_INET
8379 case BPF_FUNC_load_hdr_opt:
8380 return &bpf_sock_ops_load_hdr_opt_proto;
8381 case BPF_FUNC_store_hdr_opt:
8382 return &bpf_sock_ops_store_hdr_opt_proto;
8383 case BPF_FUNC_reserve_hdr_opt:
8384 return &bpf_sock_ops_reserve_hdr_opt_proto;
8385 case BPF_FUNC_tcp_sock:
8386 return &bpf_tcp_sock_proto;
8387 #endif /* CONFIG_INET */
8388 default:
8389 return bpf_sk_base_func_proto(func_id, prog);
8390 }
8391 }
8392
8393 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8394 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8395
8396 static const struct bpf_func_proto *
sk_msg_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8397 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8398 {
8399 switch (func_id) {
8400 case BPF_FUNC_msg_redirect_map:
8401 return &bpf_msg_redirect_map_proto;
8402 case BPF_FUNC_msg_redirect_hash:
8403 return &bpf_msg_redirect_hash_proto;
8404 case BPF_FUNC_msg_apply_bytes:
8405 return &bpf_msg_apply_bytes_proto;
8406 case BPF_FUNC_msg_cork_bytes:
8407 return &bpf_msg_cork_bytes_proto;
8408 case BPF_FUNC_msg_pull_data:
8409 return &bpf_msg_pull_data_proto;
8410 case BPF_FUNC_msg_push_data:
8411 return &bpf_msg_push_data_proto;
8412 case BPF_FUNC_msg_pop_data:
8413 return &bpf_msg_pop_data_proto;
8414 case BPF_FUNC_perf_event_output:
8415 return &bpf_event_output_data_proto;
8416 case BPF_FUNC_get_current_uid_gid:
8417 return &bpf_get_current_uid_gid_proto;
8418 case BPF_FUNC_sk_storage_get:
8419 return &bpf_sk_storage_get_proto;
8420 case BPF_FUNC_sk_storage_delete:
8421 return &bpf_sk_storage_delete_proto;
8422 case BPF_FUNC_get_netns_cookie:
8423 return &bpf_get_netns_cookie_sk_msg_proto;
8424 #ifdef CONFIG_CGROUP_NET_CLASSID
8425 case BPF_FUNC_get_cgroup_classid:
8426 return &bpf_get_cgroup_classid_curr_proto;
8427 #endif
8428 default:
8429 return bpf_sk_base_func_proto(func_id, prog);
8430 }
8431 }
8432
8433 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8434 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8435
8436 static const struct bpf_func_proto *
sk_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8437 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8438 {
8439 switch (func_id) {
8440 case BPF_FUNC_skb_store_bytes:
8441 return &bpf_skb_store_bytes_proto;
8442 case BPF_FUNC_skb_load_bytes:
8443 return &bpf_skb_load_bytes_proto;
8444 case BPF_FUNC_skb_pull_data:
8445 return &sk_skb_pull_data_proto;
8446 case BPF_FUNC_skb_change_tail:
8447 return &sk_skb_change_tail_proto;
8448 case BPF_FUNC_skb_change_head:
8449 return &sk_skb_change_head_proto;
8450 case BPF_FUNC_skb_adjust_room:
8451 return &sk_skb_adjust_room_proto;
8452 case BPF_FUNC_get_socket_cookie:
8453 return &bpf_get_socket_cookie_proto;
8454 case BPF_FUNC_get_socket_uid:
8455 return &bpf_get_socket_uid_proto;
8456 case BPF_FUNC_sk_redirect_map:
8457 return &bpf_sk_redirect_map_proto;
8458 case BPF_FUNC_sk_redirect_hash:
8459 return &bpf_sk_redirect_hash_proto;
8460 case BPF_FUNC_perf_event_output:
8461 return &bpf_skb_event_output_proto;
8462 #ifdef CONFIG_INET
8463 case BPF_FUNC_sk_lookup_tcp:
8464 return &bpf_sk_lookup_tcp_proto;
8465 case BPF_FUNC_sk_lookup_udp:
8466 return &bpf_sk_lookup_udp_proto;
8467 case BPF_FUNC_sk_release:
8468 return &bpf_sk_release_proto;
8469 case BPF_FUNC_skc_lookup_tcp:
8470 return &bpf_skc_lookup_tcp_proto;
8471 #endif
8472 default:
8473 return bpf_sk_base_func_proto(func_id, prog);
8474 }
8475 }
8476
8477 static const struct bpf_func_proto *
flow_dissector_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8478 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8479 {
8480 switch (func_id) {
8481 case BPF_FUNC_skb_load_bytes:
8482 return &bpf_flow_dissector_load_bytes_proto;
8483 default:
8484 return bpf_sk_base_func_proto(func_id, prog);
8485 }
8486 }
8487
8488 static const struct bpf_func_proto *
lwt_out_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8489 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8490 {
8491 switch (func_id) {
8492 case BPF_FUNC_skb_load_bytes:
8493 return &bpf_skb_load_bytes_proto;
8494 case BPF_FUNC_skb_pull_data:
8495 return &bpf_skb_pull_data_proto;
8496 case BPF_FUNC_csum_diff:
8497 return &bpf_csum_diff_proto;
8498 case BPF_FUNC_get_cgroup_classid:
8499 return &bpf_get_cgroup_classid_proto;
8500 case BPF_FUNC_get_route_realm:
8501 return &bpf_get_route_realm_proto;
8502 case BPF_FUNC_get_hash_recalc:
8503 return &bpf_get_hash_recalc_proto;
8504 case BPF_FUNC_perf_event_output:
8505 return &bpf_skb_event_output_proto;
8506 case BPF_FUNC_get_smp_processor_id:
8507 return &bpf_get_smp_processor_id_proto;
8508 case BPF_FUNC_skb_under_cgroup:
8509 return &bpf_skb_under_cgroup_proto;
8510 default:
8511 return bpf_sk_base_func_proto(func_id, prog);
8512 }
8513 }
8514
8515 static const struct bpf_func_proto *
lwt_in_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8516 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8517 {
8518 switch (func_id) {
8519 case BPF_FUNC_lwt_push_encap:
8520 return &bpf_lwt_in_push_encap_proto;
8521 default:
8522 return lwt_out_func_proto(func_id, prog);
8523 }
8524 }
8525
8526 static const struct bpf_func_proto *
lwt_xmit_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8527 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8528 {
8529 switch (func_id) {
8530 case BPF_FUNC_skb_get_tunnel_key:
8531 return &bpf_skb_get_tunnel_key_proto;
8532 case BPF_FUNC_skb_set_tunnel_key:
8533 return bpf_get_skb_set_tunnel_proto(func_id);
8534 case BPF_FUNC_skb_get_tunnel_opt:
8535 return &bpf_skb_get_tunnel_opt_proto;
8536 case BPF_FUNC_skb_set_tunnel_opt:
8537 return bpf_get_skb_set_tunnel_proto(func_id);
8538 case BPF_FUNC_redirect:
8539 return &bpf_redirect_proto;
8540 case BPF_FUNC_clone_redirect:
8541 return &bpf_clone_redirect_proto;
8542 case BPF_FUNC_skb_change_tail:
8543 return &bpf_skb_change_tail_proto;
8544 case BPF_FUNC_skb_change_head:
8545 return &bpf_skb_change_head_proto;
8546 case BPF_FUNC_skb_store_bytes:
8547 return &bpf_skb_store_bytes_proto;
8548 case BPF_FUNC_csum_update:
8549 return &bpf_csum_update_proto;
8550 case BPF_FUNC_csum_level:
8551 return &bpf_csum_level_proto;
8552 case BPF_FUNC_l3_csum_replace:
8553 return &bpf_l3_csum_replace_proto;
8554 case BPF_FUNC_l4_csum_replace:
8555 return &bpf_l4_csum_replace_proto;
8556 case BPF_FUNC_set_hash_invalid:
8557 return &bpf_set_hash_invalid_proto;
8558 case BPF_FUNC_lwt_push_encap:
8559 return &bpf_lwt_xmit_push_encap_proto;
8560 default:
8561 return lwt_out_func_proto(func_id, prog);
8562 }
8563 }
8564
8565 static const struct bpf_func_proto *
lwt_seg6local_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8566 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8567 {
8568 switch (func_id) {
8569 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8570 case BPF_FUNC_lwt_seg6_store_bytes:
8571 return &bpf_lwt_seg6_store_bytes_proto;
8572 case BPF_FUNC_lwt_seg6_action:
8573 return &bpf_lwt_seg6_action_proto;
8574 case BPF_FUNC_lwt_seg6_adjust_srh:
8575 return &bpf_lwt_seg6_adjust_srh_proto;
8576 #endif
8577 default:
8578 return lwt_out_func_proto(func_id, prog);
8579 }
8580 }
8581
bpf_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8582 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8583 const struct bpf_prog *prog,
8584 struct bpf_insn_access_aux *info)
8585 {
8586 const int size_default = sizeof(__u32);
8587
8588 if (off < 0 || off >= sizeof(struct __sk_buff))
8589 return false;
8590
8591 /* The verifier guarantees that size > 0. */
8592 if (off % size != 0)
8593 return false;
8594
8595 switch (off) {
8596 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8597 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8598 return false;
8599 break;
8600 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8601 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8602 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8603 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8604 case bpf_ctx_range(struct __sk_buff, data):
8605 case bpf_ctx_range(struct __sk_buff, data_meta):
8606 case bpf_ctx_range(struct __sk_buff, data_end):
8607 if (size != size_default)
8608 return false;
8609 break;
8610 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8611 return false;
8612 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8613 if (type == BPF_WRITE || size != sizeof(__u64))
8614 return false;
8615 break;
8616 case bpf_ctx_range(struct __sk_buff, tstamp):
8617 if (size != sizeof(__u64))
8618 return false;
8619 break;
8620 case offsetof(struct __sk_buff, sk):
8621 if (type == BPF_WRITE || size != sizeof(__u64))
8622 return false;
8623 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8624 break;
8625 case offsetof(struct __sk_buff, tstamp_type):
8626 return false;
8627 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8628 /* Explicitly prohibit access to padding in __sk_buff. */
8629 return false;
8630 default:
8631 /* Only narrow read access allowed for now. */
8632 if (type == BPF_WRITE) {
8633 if (size != size_default)
8634 return false;
8635 } else {
8636 bpf_ctx_record_field_size(info, size_default);
8637 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8638 return false;
8639 }
8640 }
8641
8642 return true;
8643 }
8644
sk_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8645 static bool sk_filter_is_valid_access(int off, int size,
8646 enum bpf_access_type type,
8647 const struct bpf_prog *prog,
8648 struct bpf_insn_access_aux *info)
8649 {
8650 switch (off) {
8651 case bpf_ctx_range(struct __sk_buff, tc_classid):
8652 case bpf_ctx_range(struct __sk_buff, data):
8653 case bpf_ctx_range(struct __sk_buff, data_meta):
8654 case bpf_ctx_range(struct __sk_buff, data_end):
8655 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8656 case bpf_ctx_range(struct __sk_buff, tstamp):
8657 case bpf_ctx_range(struct __sk_buff, wire_len):
8658 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8659 return false;
8660 }
8661
8662 if (type == BPF_WRITE) {
8663 switch (off) {
8664 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8665 break;
8666 default:
8667 return false;
8668 }
8669 }
8670
8671 return bpf_skb_is_valid_access(off, size, type, prog, info);
8672 }
8673
cg_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8674 static bool cg_skb_is_valid_access(int off, int size,
8675 enum bpf_access_type type,
8676 const struct bpf_prog *prog,
8677 struct bpf_insn_access_aux *info)
8678 {
8679 switch (off) {
8680 case bpf_ctx_range(struct __sk_buff, tc_classid):
8681 case bpf_ctx_range(struct __sk_buff, data_meta):
8682 case bpf_ctx_range(struct __sk_buff, wire_len):
8683 return false;
8684 case bpf_ctx_range(struct __sk_buff, data):
8685 case bpf_ctx_range(struct __sk_buff, data_end):
8686 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8687 return false;
8688 break;
8689 }
8690
8691 if (type == BPF_WRITE) {
8692 switch (off) {
8693 case bpf_ctx_range(struct __sk_buff, mark):
8694 case bpf_ctx_range(struct __sk_buff, priority):
8695 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8696 break;
8697 case bpf_ctx_range(struct __sk_buff, tstamp):
8698 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8699 return false;
8700 break;
8701 default:
8702 return false;
8703 }
8704 }
8705
8706 switch (off) {
8707 case bpf_ctx_range(struct __sk_buff, data):
8708 info->reg_type = PTR_TO_PACKET;
8709 break;
8710 case bpf_ctx_range(struct __sk_buff, data_end):
8711 info->reg_type = PTR_TO_PACKET_END;
8712 break;
8713 }
8714
8715 return bpf_skb_is_valid_access(off, size, type, prog, info);
8716 }
8717
lwt_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8718 static bool lwt_is_valid_access(int off, int size,
8719 enum bpf_access_type type,
8720 const struct bpf_prog *prog,
8721 struct bpf_insn_access_aux *info)
8722 {
8723 switch (off) {
8724 case bpf_ctx_range(struct __sk_buff, tc_classid):
8725 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8726 case bpf_ctx_range(struct __sk_buff, data_meta):
8727 case bpf_ctx_range(struct __sk_buff, tstamp):
8728 case bpf_ctx_range(struct __sk_buff, wire_len):
8729 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8730 return false;
8731 }
8732
8733 if (type == BPF_WRITE) {
8734 switch (off) {
8735 case bpf_ctx_range(struct __sk_buff, mark):
8736 case bpf_ctx_range(struct __sk_buff, priority):
8737 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8738 break;
8739 default:
8740 return false;
8741 }
8742 }
8743
8744 switch (off) {
8745 case bpf_ctx_range(struct __sk_buff, data):
8746 info->reg_type = PTR_TO_PACKET;
8747 break;
8748 case bpf_ctx_range(struct __sk_buff, data_end):
8749 info->reg_type = PTR_TO_PACKET_END;
8750 break;
8751 }
8752
8753 return bpf_skb_is_valid_access(off, size, type, prog, info);
8754 }
8755
8756 /* Attach type specific accesses */
__sock_filter_check_attach_type(int off,enum bpf_access_type access_type,enum bpf_attach_type attach_type)8757 static bool __sock_filter_check_attach_type(int off,
8758 enum bpf_access_type access_type,
8759 enum bpf_attach_type attach_type)
8760 {
8761 switch (off) {
8762 case offsetof(struct bpf_sock, bound_dev_if):
8763 case offsetof(struct bpf_sock, mark):
8764 case offsetof(struct bpf_sock, priority):
8765 switch (attach_type) {
8766 case BPF_CGROUP_INET_SOCK_CREATE:
8767 case BPF_CGROUP_INET_SOCK_RELEASE:
8768 goto full_access;
8769 default:
8770 return false;
8771 }
8772 case bpf_ctx_range(struct bpf_sock, src_ip4):
8773 switch (attach_type) {
8774 case BPF_CGROUP_INET4_POST_BIND:
8775 goto read_only;
8776 default:
8777 return false;
8778 }
8779 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8780 switch (attach_type) {
8781 case BPF_CGROUP_INET6_POST_BIND:
8782 goto read_only;
8783 default:
8784 return false;
8785 }
8786 case bpf_ctx_range(struct bpf_sock, src_port):
8787 switch (attach_type) {
8788 case BPF_CGROUP_INET4_POST_BIND:
8789 case BPF_CGROUP_INET6_POST_BIND:
8790 goto read_only;
8791 default:
8792 return false;
8793 }
8794 }
8795 read_only:
8796 return access_type == BPF_READ;
8797 full_access:
8798 return true;
8799 }
8800
bpf_sock_common_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8801 bool bpf_sock_common_is_valid_access(int off, int size,
8802 enum bpf_access_type type,
8803 struct bpf_insn_access_aux *info)
8804 {
8805 switch (off) {
8806 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8807 return false;
8808 default:
8809 return bpf_sock_is_valid_access(off, size, type, info);
8810 }
8811 }
8812
bpf_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8813 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8814 struct bpf_insn_access_aux *info)
8815 {
8816 const int size_default = sizeof(__u32);
8817 int field_size;
8818
8819 if (off < 0 || off >= sizeof(struct bpf_sock))
8820 return false;
8821 if (off % size != 0)
8822 return false;
8823
8824 switch (off) {
8825 case offsetof(struct bpf_sock, state):
8826 case offsetof(struct bpf_sock, family):
8827 case offsetof(struct bpf_sock, type):
8828 case offsetof(struct bpf_sock, protocol):
8829 case offsetof(struct bpf_sock, src_port):
8830 case offsetof(struct bpf_sock, rx_queue_mapping):
8831 case bpf_ctx_range(struct bpf_sock, src_ip4):
8832 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8833 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8834 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8835 bpf_ctx_record_field_size(info, size_default);
8836 return bpf_ctx_narrow_access_ok(off, size, size_default);
8837 case bpf_ctx_range(struct bpf_sock, dst_port):
8838 field_size = size == size_default ?
8839 size_default : sizeof_field(struct bpf_sock, dst_port);
8840 bpf_ctx_record_field_size(info, field_size);
8841 return bpf_ctx_narrow_access_ok(off, size, field_size);
8842 case offsetofend(struct bpf_sock, dst_port) ...
8843 offsetof(struct bpf_sock, dst_ip4) - 1:
8844 return false;
8845 }
8846
8847 return size == size_default;
8848 }
8849
sock_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8850 static bool sock_filter_is_valid_access(int off, int size,
8851 enum bpf_access_type type,
8852 const struct bpf_prog *prog,
8853 struct bpf_insn_access_aux *info)
8854 {
8855 if (!bpf_sock_is_valid_access(off, size, type, info))
8856 return false;
8857 return __sock_filter_check_attach_type(off, type,
8858 prog->expected_attach_type);
8859 }
8860
bpf_noop_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8861 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8862 const struct bpf_prog *prog)
8863 {
8864 /* Neither direct read nor direct write requires any preliminary
8865 * action.
8866 */
8867 return 0;
8868 }
8869
bpf_unclone_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog,int drop_verdict)8870 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8871 const struct bpf_prog *prog, int drop_verdict)
8872 {
8873 struct bpf_insn *insn = insn_buf;
8874
8875 if (!direct_write)
8876 return 0;
8877
8878 /* if (!skb->cloned)
8879 * goto start;
8880 *
8881 * (Fast-path, otherwise approximation that we might be
8882 * a clone, do the rest in helper.)
8883 */
8884 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8885 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8886 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8887
8888 /* ret = bpf_skb_pull_data(skb, 0); */
8889 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8890 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8891 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8892 BPF_FUNC_skb_pull_data);
8893 /* if (!ret)
8894 * goto restore;
8895 * return TC_ACT_SHOT;
8896 */
8897 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8898 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8899 *insn++ = BPF_EXIT_INSN();
8900
8901 /* restore: */
8902 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8903 /* start: */
8904 *insn++ = prog->insnsi[0];
8905
8906 return insn - insn_buf;
8907 }
8908
bpf_gen_ld_abs(const struct bpf_insn * orig,struct bpf_insn * insn_buf)8909 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8910 struct bpf_insn *insn_buf)
8911 {
8912 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8913 struct bpf_insn *insn = insn_buf;
8914
8915 if (!indirect) {
8916 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8917 } else {
8918 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8919 if (orig->imm)
8920 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8921 }
8922 /* We're guaranteed here that CTX is in R6. */
8923 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8924
8925 switch (BPF_SIZE(orig->code)) {
8926 case BPF_B:
8927 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8928 break;
8929 case BPF_H:
8930 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8931 break;
8932 case BPF_W:
8933 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8934 break;
8935 }
8936
8937 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8938 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8939 *insn++ = BPF_EXIT_INSN();
8940
8941 return insn - insn_buf;
8942 }
8943
tc_cls_act_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8944 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8945 const struct bpf_prog *prog)
8946 {
8947 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8948 }
8949
tc_cls_act_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8950 static bool tc_cls_act_is_valid_access(int off, int size,
8951 enum bpf_access_type type,
8952 const struct bpf_prog *prog,
8953 struct bpf_insn_access_aux *info)
8954 {
8955 if (type == BPF_WRITE) {
8956 switch (off) {
8957 case bpf_ctx_range(struct __sk_buff, mark):
8958 case bpf_ctx_range(struct __sk_buff, tc_index):
8959 case bpf_ctx_range(struct __sk_buff, priority):
8960 case bpf_ctx_range(struct __sk_buff, tc_classid):
8961 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8962 case bpf_ctx_range(struct __sk_buff, tstamp):
8963 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8964 break;
8965 default:
8966 return false;
8967 }
8968 }
8969
8970 switch (off) {
8971 case bpf_ctx_range(struct __sk_buff, data):
8972 info->reg_type = PTR_TO_PACKET;
8973 break;
8974 case bpf_ctx_range(struct __sk_buff, data_meta):
8975 info->reg_type = PTR_TO_PACKET_META;
8976 break;
8977 case bpf_ctx_range(struct __sk_buff, data_end):
8978 info->reg_type = PTR_TO_PACKET_END;
8979 break;
8980 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8981 return false;
8982 case offsetof(struct __sk_buff, tstamp_type):
8983 /* The convert_ctx_access() on reading and writing
8984 * __sk_buff->tstamp depends on whether the bpf prog
8985 * has used __sk_buff->tstamp_type or not.
8986 * Thus, we need to set prog->tstamp_type_access
8987 * earlier during is_valid_access() here.
8988 */
8989 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8990 return size == sizeof(__u8);
8991 }
8992
8993 return bpf_skb_is_valid_access(off, size, type, prog, info);
8994 }
8995
8996 DEFINE_MUTEX(nf_conn_btf_access_lock);
8997 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8998
8999 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
9000 const struct bpf_reg_state *reg,
9001 int off, int size);
9002 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
9003
tc_cls_act_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9004 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
9005 const struct bpf_reg_state *reg,
9006 int off, int size)
9007 {
9008 int ret = -EACCES;
9009
9010 mutex_lock(&nf_conn_btf_access_lock);
9011 if (nfct_btf_struct_access)
9012 ret = nfct_btf_struct_access(log, reg, off, size);
9013 mutex_unlock(&nf_conn_btf_access_lock);
9014
9015 return ret;
9016 }
9017
__is_valid_xdp_access(int off,int size)9018 static bool __is_valid_xdp_access(int off, int size)
9019 {
9020 if (off < 0 || off >= sizeof(struct xdp_md))
9021 return false;
9022 if (off % size != 0)
9023 return false;
9024 if (size != sizeof(__u32))
9025 return false;
9026
9027 return true;
9028 }
9029
xdp_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9030 static bool xdp_is_valid_access(int off, int size,
9031 enum bpf_access_type type,
9032 const struct bpf_prog *prog,
9033 struct bpf_insn_access_aux *info)
9034 {
9035 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
9036 switch (off) {
9037 case offsetof(struct xdp_md, egress_ifindex):
9038 return false;
9039 }
9040 }
9041
9042 if (type == BPF_WRITE) {
9043 if (bpf_prog_is_offloaded(prog->aux)) {
9044 switch (off) {
9045 case offsetof(struct xdp_md, rx_queue_index):
9046 return __is_valid_xdp_access(off, size);
9047 }
9048 }
9049 return false;
9050 }
9051
9052 switch (off) {
9053 case offsetof(struct xdp_md, data):
9054 info->reg_type = PTR_TO_PACKET;
9055 break;
9056 case offsetof(struct xdp_md, data_meta):
9057 info->reg_type = PTR_TO_PACKET_META;
9058 break;
9059 case offsetof(struct xdp_md, data_end):
9060 info->reg_type = PTR_TO_PACKET_END;
9061 break;
9062 }
9063
9064 return __is_valid_xdp_access(off, size);
9065 }
9066
bpf_warn_invalid_xdp_action(struct net_device * dev,struct bpf_prog * prog,u32 act)9067 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
9068 {
9069 const u32 act_max = XDP_REDIRECT;
9070
9071 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
9072 act > act_max ? "Illegal" : "Driver unsupported",
9073 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
9074 }
9075 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
9076
xdp_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9077 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
9078 const struct bpf_reg_state *reg,
9079 int off, int size)
9080 {
9081 int ret = -EACCES;
9082
9083 mutex_lock(&nf_conn_btf_access_lock);
9084 if (nfct_btf_struct_access)
9085 ret = nfct_btf_struct_access(log, reg, off, size);
9086 mutex_unlock(&nf_conn_btf_access_lock);
9087
9088 return ret;
9089 }
9090
sock_addr_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9091 static bool sock_addr_is_valid_access(int off, int size,
9092 enum bpf_access_type type,
9093 const struct bpf_prog *prog,
9094 struct bpf_insn_access_aux *info)
9095 {
9096 const int size_default = sizeof(__u32);
9097
9098 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
9099 return false;
9100 if (off % size != 0)
9101 return false;
9102
9103 /* Disallow access to fields not belonging to the attach type's address
9104 * family.
9105 */
9106 switch (off) {
9107 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9108 switch (prog->expected_attach_type) {
9109 case BPF_CGROUP_INET4_BIND:
9110 case BPF_CGROUP_INET4_CONNECT:
9111 case BPF_CGROUP_INET4_GETPEERNAME:
9112 case BPF_CGROUP_INET4_GETSOCKNAME:
9113 case BPF_CGROUP_UDP4_SENDMSG:
9114 case BPF_CGROUP_UDP4_RECVMSG:
9115 break;
9116 default:
9117 return false;
9118 }
9119 break;
9120 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9121 switch (prog->expected_attach_type) {
9122 case BPF_CGROUP_INET6_BIND:
9123 case BPF_CGROUP_INET6_CONNECT:
9124 case BPF_CGROUP_INET6_GETPEERNAME:
9125 case BPF_CGROUP_INET6_GETSOCKNAME:
9126 case BPF_CGROUP_UDP6_SENDMSG:
9127 case BPF_CGROUP_UDP6_RECVMSG:
9128 break;
9129 default:
9130 return false;
9131 }
9132 break;
9133 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9134 switch (prog->expected_attach_type) {
9135 case BPF_CGROUP_UDP4_SENDMSG:
9136 break;
9137 default:
9138 return false;
9139 }
9140 break;
9141 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9142 msg_src_ip6[3]):
9143 switch (prog->expected_attach_type) {
9144 case BPF_CGROUP_UDP6_SENDMSG:
9145 break;
9146 default:
9147 return false;
9148 }
9149 break;
9150 }
9151
9152 switch (off) {
9153 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9154 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9155 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9156 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9157 msg_src_ip6[3]):
9158 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9159 if (type == BPF_READ) {
9160 bpf_ctx_record_field_size(info, size_default);
9161
9162 if (bpf_ctx_wide_access_ok(off, size,
9163 struct bpf_sock_addr,
9164 user_ip6))
9165 return true;
9166
9167 if (bpf_ctx_wide_access_ok(off, size,
9168 struct bpf_sock_addr,
9169 msg_src_ip6))
9170 return true;
9171
9172 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9173 return false;
9174 } else {
9175 if (bpf_ctx_wide_access_ok(off, size,
9176 struct bpf_sock_addr,
9177 user_ip6))
9178 return true;
9179
9180 if (bpf_ctx_wide_access_ok(off, size,
9181 struct bpf_sock_addr,
9182 msg_src_ip6))
9183 return true;
9184
9185 if (size != size_default)
9186 return false;
9187 }
9188 break;
9189 case offsetof(struct bpf_sock_addr, sk):
9190 if (type != BPF_READ)
9191 return false;
9192 if (size != sizeof(__u64))
9193 return false;
9194 info->reg_type = PTR_TO_SOCKET;
9195 break;
9196 default:
9197 if (type == BPF_READ) {
9198 if (size != size_default)
9199 return false;
9200 } else {
9201 return false;
9202 }
9203 }
9204
9205 return true;
9206 }
9207
sock_ops_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9208 static bool sock_ops_is_valid_access(int off, int size,
9209 enum bpf_access_type type,
9210 const struct bpf_prog *prog,
9211 struct bpf_insn_access_aux *info)
9212 {
9213 const int size_default = sizeof(__u32);
9214
9215 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9216 return false;
9217
9218 /* The verifier guarantees that size > 0. */
9219 if (off % size != 0)
9220 return false;
9221
9222 if (type == BPF_WRITE) {
9223 switch (off) {
9224 case offsetof(struct bpf_sock_ops, reply):
9225 case offsetof(struct bpf_sock_ops, sk_txhash):
9226 if (size != size_default)
9227 return false;
9228 break;
9229 default:
9230 return false;
9231 }
9232 } else {
9233 switch (off) {
9234 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9235 bytes_acked):
9236 if (size != sizeof(__u64))
9237 return false;
9238 break;
9239 case offsetof(struct bpf_sock_ops, sk):
9240 if (size != sizeof(__u64))
9241 return false;
9242 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9243 break;
9244 case offsetof(struct bpf_sock_ops, skb_data):
9245 if (size != sizeof(__u64))
9246 return false;
9247 info->reg_type = PTR_TO_PACKET;
9248 break;
9249 case offsetof(struct bpf_sock_ops, skb_data_end):
9250 if (size != sizeof(__u64))
9251 return false;
9252 info->reg_type = PTR_TO_PACKET_END;
9253 break;
9254 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9255 bpf_ctx_record_field_size(info, size_default);
9256 return bpf_ctx_narrow_access_ok(off, size,
9257 size_default);
9258 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9259 if (size != sizeof(__u64))
9260 return false;
9261 break;
9262 default:
9263 if (size != size_default)
9264 return false;
9265 break;
9266 }
9267 }
9268
9269 return true;
9270 }
9271
sk_skb_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)9272 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9273 const struct bpf_prog *prog)
9274 {
9275 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9276 }
9277
sk_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9278 static bool sk_skb_is_valid_access(int off, int size,
9279 enum bpf_access_type type,
9280 const struct bpf_prog *prog,
9281 struct bpf_insn_access_aux *info)
9282 {
9283 switch (off) {
9284 case bpf_ctx_range(struct __sk_buff, tc_classid):
9285 case bpf_ctx_range(struct __sk_buff, data_meta):
9286 case bpf_ctx_range(struct __sk_buff, tstamp):
9287 case bpf_ctx_range(struct __sk_buff, wire_len):
9288 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9289 return false;
9290 }
9291
9292 if (type == BPF_WRITE) {
9293 switch (off) {
9294 case bpf_ctx_range(struct __sk_buff, tc_index):
9295 case bpf_ctx_range(struct __sk_buff, priority):
9296 break;
9297 default:
9298 return false;
9299 }
9300 }
9301
9302 switch (off) {
9303 case bpf_ctx_range(struct __sk_buff, mark):
9304 return false;
9305 case bpf_ctx_range(struct __sk_buff, data):
9306 info->reg_type = PTR_TO_PACKET;
9307 break;
9308 case bpf_ctx_range(struct __sk_buff, data_end):
9309 info->reg_type = PTR_TO_PACKET_END;
9310 break;
9311 }
9312
9313 return bpf_skb_is_valid_access(off, size, type, prog, info);
9314 }
9315
sk_msg_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9316 static bool sk_msg_is_valid_access(int off, int size,
9317 enum bpf_access_type type,
9318 const struct bpf_prog *prog,
9319 struct bpf_insn_access_aux *info)
9320 {
9321 if (type == BPF_WRITE)
9322 return false;
9323
9324 if (off % size != 0)
9325 return false;
9326
9327 switch (off) {
9328 case offsetof(struct sk_msg_md, data):
9329 info->reg_type = PTR_TO_PACKET;
9330 if (size != sizeof(__u64))
9331 return false;
9332 break;
9333 case offsetof(struct sk_msg_md, data_end):
9334 info->reg_type = PTR_TO_PACKET_END;
9335 if (size != sizeof(__u64))
9336 return false;
9337 break;
9338 case offsetof(struct sk_msg_md, sk):
9339 if (size != sizeof(__u64))
9340 return false;
9341 info->reg_type = PTR_TO_SOCKET;
9342 break;
9343 case bpf_ctx_range(struct sk_msg_md, family):
9344 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9345 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9346 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9347 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9348 case bpf_ctx_range(struct sk_msg_md, remote_port):
9349 case bpf_ctx_range(struct sk_msg_md, local_port):
9350 case bpf_ctx_range(struct sk_msg_md, size):
9351 if (size != sizeof(__u32))
9352 return false;
9353 break;
9354 default:
9355 return false;
9356 }
9357 return true;
9358 }
9359
flow_dissector_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9360 static bool flow_dissector_is_valid_access(int off, int size,
9361 enum bpf_access_type type,
9362 const struct bpf_prog *prog,
9363 struct bpf_insn_access_aux *info)
9364 {
9365 const int size_default = sizeof(__u32);
9366
9367 if (off < 0 || off >= sizeof(struct __sk_buff))
9368 return false;
9369
9370 if (type == BPF_WRITE)
9371 return false;
9372
9373 switch (off) {
9374 case bpf_ctx_range(struct __sk_buff, data):
9375 if (size != size_default)
9376 return false;
9377 info->reg_type = PTR_TO_PACKET;
9378 return true;
9379 case bpf_ctx_range(struct __sk_buff, data_end):
9380 if (size != size_default)
9381 return false;
9382 info->reg_type = PTR_TO_PACKET_END;
9383 return true;
9384 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9385 if (size != sizeof(__u64))
9386 return false;
9387 info->reg_type = PTR_TO_FLOW_KEYS;
9388 return true;
9389 default:
9390 return false;
9391 }
9392 }
9393
flow_dissector_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9394 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9395 const struct bpf_insn *si,
9396 struct bpf_insn *insn_buf,
9397 struct bpf_prog *prog,
9398 u32 *target_size)
9399
9400 {
9401 struct bpf_insn *insn = insn_buf;
9402
9403 switch (si->off) {
9404 case offsetof(struct __sk_buff, data):
9405 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9406 si->dst_reg, si->src_reg,
9407 offsetof(struct bpf_flow_dissector, data));
9408 break;
9409
9410 case offsetof(struct __sk_buff, data_end):
9411 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9412 si->dst_reg, si->src_reg,
9413 offsetof(struct bpf_flow_dissector, data_end));
9414 break;
9415
9416 case offsetof(struct __sk_buff, flow_keys):
9417 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9418 si->dst_reg, si->src_reg,
9419 offsetof(struct bpf_flow_dissector, flow_keys));
9420 break;
9421 }
9422
9423 return insn - insn_buf;
9424 }
9425
bpf_convert_tstamp_type_read(const struct bpf_insn * si,struct bpf_insn * insn)9426 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9427 struct bpf_insn *insn)
9428 {
9429 __u8 value_reg = si->dst_reg;
9430 __u8 skb_reg = si->src_reg;
9431 BUILD_BUG_ON(__SKB_CLOCK_MAX != (int)BPF_SKB_CLOCK_TAI);
9432 BUILD_BUG_ON(SKB_CLOCK_REALTIME != (int)BPF_SKB_CLOCK_REALTIME);
9433 BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != (int)BPF_SKB_CLOCK_MONOTONIC);
9434 BUILD_BUG_ON(SKB_CLOCK_TAI != (int)BPF_SKB_CLOCK_TAI);
9435 *insn++ = BPF_LDX_MEM(BPF_B, value_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9436 *insn++ = BPF_ALU32_IMM(BPF_AND, value_reg, SKB_TSTAMP_TYPE_MASK);
9437 #ifdef __BIG_ENDIAN_BITFIELD
9438 *insn++ = BPF_ALU32_IMM(BPF_RSH, value_reg, SKB_TSTAMP_TYPE_RSHIFT);
9439 #else
9440 BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1));
9441 #endif
9442
9443 return insn;
9444 }
9445
bpf_convert_shinfo_access(__u8 dst_reg,__u8 skb_reg,struct bpf_insn * insn)9446 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9447 struct bpf_insn *insn)
9448 {
9449 /* si->dst_reg = skb_shinfo(SKB); */
9450 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9452 BPF_REG_AX, skb_reg,
9453 offsetof(struct sk_buff, end));
9454 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9455 dst_reg, skb_reg,
9456 offsetof(struct sk_buff, head));
9457 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9458 #else
9459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9460 dst_reg, skb_reg,
9461 offsetof(struct sk_buff, end));
9462 #endif
9463
9464 return insn;
9465 }
9466
bpf_convert_tstamp_read(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9467 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9468 const struct bpf_insn *si,
9469 struct bpf_insn *insn)
9470 {
9471 __u8 value_reg = si->dst_reg;
9472 __u8 skb_reg = si->src_reg;
9473
9474 #ifdef CONFIG_NET_XGRESS
9475 /* If the tstamp_type is read,
9476 * the bpf prog is aware the tstamp could have delivery time.
9477 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9478 */
9479 if (!prog->tstamp_type_access) {
9480 /* AX is needed because src_reg and dst_reg could be the same */
9481 __u8 tmp_reg = BPF_REG_AX;
9482
9483 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9484 /* check if ingress mask bits is set */
9485 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9486 *insn++ = BPF_JMP_A(4);
9487 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, SKB_TSTAMP_TYPE_MASK, 1);
9488 *insn++ = BPF_JMP_A(2);
9489 /* skb->tc_at_ingress && skb->tstamp_type,
9490 * read 0 as the (rcv) timestamp.
9491 */
9492 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9493 *insn++ = BPF_JMP_A(1);
9494 }
9495 #endif
9496
9497 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9498 offsetof(struct sk_buff, tstamp));
9499 return insn;
9500 }
9501
bpf_convert_tstamp_write(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9502 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9503 const struct bpf_insn *si,
9504 struct bpf_insn *insn)
9505 {
9506 __u8 value_reg = si->src_reg;
9507 __u8 skb_reg = si->dst_reg;
9508
9509 #ifdef CONFIG_NET_XGRESS
9510 /* If the tstamp_type is read,
9511 * the bpf prog is aware the tstamp could have delivery time.
9512 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9513 * Otherwise, writing at ingress will have to clear the
9514 * skb->tstamp_type bit also.
9515 */
9516 if (!prog->tstamp_type_access) {
9517 __u8 tmp_reg = BPF_REG_AX;
9518
9519 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9520 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9521 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9522 /* goto <store> */
9523 *insn++ = BPF_JMP_A(2);
9524 /* <clear>: skb->tstamp_type */
9525 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_TSTAMP_TYPE_MASK);
9526 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9527 }
9528 #endif
9529
9530 /* <store>: skb->tstamp = tstamp */
9531 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9532 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9533 return insn;
9534 }
9535
9536 #define BPF_EMIT_STORE(size, si, off) \
9537 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9538 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9539
bpf_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9540 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9541 const struct bpf_insn *si,
9542 struct bpf_insn *insn_buf,
9543 struct bpf_prog *prog, u32 *target_size)
9544 {
9545 struct bpf_insn *insn = insn_buf;
9546 int off;
9547
9548 switch (si->off) {
9549 case offsetof(struct __sk_buff, len):
9550 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9551 bpf_target_off(struct sk_buff, len, 4,
9552 target_size));
9553 break;
9554
9555 case offsetof(struct __sk_buff, protocol):
9556 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9557 bpf_target_off(struct sk_buff, protocol, 2,
9558 target_size));
9559 break;
9560
9561 case offsetof(struct __sk_buff, vlan_proto):
9562 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9563 bpf_target_off(struct sk_buff, vlan_proto, 2,
9564 target_size));
9565 break;
9566
9567 case offsetof(struct __sk_buff, priority):
9568 if (type == BPF_WRITE)
9569 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9570 bpf_target_off(struct sk_buff, priority, 4,
9571 target_size));
9572 else
9573 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9574 bpf_target_off(struct sk_buff, priority, 4,
9575 target_size));
9576 break;
9577
9578 case offsetof(struct __sk_buff, ingress_ifindex):
9579 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9580 bpf_target_off(struct sk_buff, skb_iif, 4,
9581 target_size));
9582 break;
9583
9584 case offsetof(struct __sk_buff, ifindex):
9585 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9586 si->dst_reg, si->src_reg,
9587 offsetof(struct sk_buff, dev));
9588 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9589 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9590 bpf_target_off(struct net_device, ifindex, 4,
9591 target_size));
9592 break;
9593
9594 case offsetof(struct __sk_buff, hash):
9595 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9596 bpf_target_off(struct sk_buff, hash, 4,
9597 target_size));
9598 break;
9599
9600 case offsetof(struct __sk_buff, mark):
9601 if (type == BPF_WRITE)
9602 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9603 bpf_target_off(struct sk_buff, mark, 4,
9604 target_size));
9605 else
9606 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9607 bpf_target_off(struct sk_buff, mark, 4,
9608 target_size));
9609 break;
9610
9611 case offsetof(struct __sk_buff, pkt_type):
9612 *target_size = 1;
9613 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9614 PKT_TYPE_OFFSET);
9615 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9616 #ifdef __BIG_ENDIAN_BITFIELD
9617 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9618 #endif
9619 break;
9620
9621 case offsetof(struct __sk_buff, queue_mapping):
9622 if (type == BPF_WRITE) {
9623 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9624
9625 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9626 *insn++ = BPF_JMP_A(0); /* noop */
9627 break;
9628 }
9629
9630 if (BPF_CLASS(si->code) == BPF_STX)
9631 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9632 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9633 } else {
9634 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9635 bpf_target_off(struct sk_buff,
9636 queue_mapping,
9637 2, target_size));
9638 }
9639 break;
9640
9641 case offsetof(struct __sk_buff, vlan_present):
9642 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9643 bpf_target_off(struct sk_buff,
9644 vlan_all, 4, target_size));
9645 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9646 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9647 break;
9648
9649 case offsetof(struct __sk_buff, vlan_tci):
9650 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9651 bpf_target_off(struct sk_buff, vlan_tci, 2,
9652 target_size));
9653 break;
9654
9655 case offsetof(struct __sk_buff, cb[0]) ...
9656 offsetofend(struct __sk_buff, cb[4]) - 1:
9657 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9658 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9659 offsetof(struct qdisc_skb_cb, data)) %
9660 sizeof(__u64));
9661
9662 prog->cb_access = 1;
9663 off = si->off;
9664 off -= offsetof(struct __sk_buff, cb[0]);
9665 off += offsetof(struct sk_buff, cb);
9666 off += offsetof(struct qdisc_skb_cb, data);
9667 if (type == BPF_WRITE)
9668 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9669 else
9670 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9671 si->src_reg, off);
9672 break;
9673
9674 case offsetof(struct __sk_buff, tc_classid):
9675 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9676
9677 off = si->off;
9678 off -= offsetof(struct __sk_buff, tc_classid);
9679 off += offsetof(struct sk_buff, cb);
9680 off += offsetof(struct qdisc_skb_cb, tc_classid);
9681 *target_size = 2;
9682 if (type == BPF_WRITE)
9683 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9684 else
9685 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9686 si->src_reg, off);
9687 break;
9688
9689 case offsetof(struct __sk_buff, data):
9690 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9691 si->dst_reg, si->src_reg,
9692 offsetof(struct sk_buff, data));
9693 break;
9694
9695 case offsetof(struct __sk_buff, data_meta):
9696 off = si->off;
9697 off -= offsetof(struct __sk_buff, data_meta);
9698 off += offsetof(struct sk_buff, cb);
9699 off += offsetof(struct bpf_skb_data_end, data_meta);
9700 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9701 si->src_reg, off);
9702 break;
9703
9704 case offsetof(struct __sk_buff, data_end):
9705 off = si->off;
9706 off -= offsetof(struct __sk_buff, data_end);
9707 off += offsetof(struct sk_buff, cb);
9708 off += offsetof(struct bpf_skb_data_end, data_end);
9709 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9710 si->src_reg, off);
9711 break;
9712
9713 case offsetof(struct __sk_buff, tc_index):
9714 #ifdef CONFIG_NET_SCHED
9715 if (type == BPF_WRITE)
9716 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9717 bpf_target_off(struct sk_buff, tc_index, 2,
9718 target_size));
9719 else
9720 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9721 bpf_target_off(struct sk_buff, tc_index, 2,
9722 target_size));
9723 #else
9724 *target_size = 2;
9725 if (type == BPF_WRITE)
9726 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9727 else
9728 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9729 #endif
9730 break;
9731
9732 case offsetof(struct __sk_buff, napi_id):
9733 #if defined(CONFIG_NET_RX_BUSY_POLL)
9734 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9735 bpf_target_off(struct sk_buff, napi_id, 4,
9736 target_size));
9737 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9738 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9739 #else
9740 *target_size = 4;
9741 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9742 #endif
9743 break;
9744 case offsetof(struct __sk_buff, family):
9745 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9746
9747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9748 si->dst_reg, si->src_reg,
9749 offsetof(struct sk_buff, sk));
9750 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9751 bpf_target_off(struct sock_common,
9752 skc_family,
9753 2, target_size));
9754 break;
9755 case offsetof(struct __sk_buff, remote_ip4):
9756 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9757
9758 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9759 si->dst_reg, si->src_reg,
9760 offsetof(struct sk_buff, sk));
9761 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9762 bpf_target_off(struct sock_common,
9763 skc_daddr,
9764 4, target_size));
9765 break;
9766 case offsetof(struct __sk_buff, local_ip4):
9767 BUILD_BUG_ON(sizeof_field(struct sock_common,
9768 skc_rcv_saddr) != 4);
9769
9770 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9771 si->dst_reg, si->src_reg,
9772 offsetof(struct sk_buff, sk));
9773 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9774 bpf_target_off(struct sock_common,
9775 skc_rcv_saddr,
9776 4, target_size));
9777 break;
9778 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9779 offsetof(struct __sk_buff, remote_ip6[3]):
9780 #if IS_ENABLED(CONFIG_IPV6)
9781 BUILD_BUG_ON(sizeof_field(struct sock_common,
9782 skc_v6_daddr.s6_addr32[0]) != 4);
9783
9784 off = si->off;
9785 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9786
9787 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9788 si->dst_reg, si->src_reg,
9789 offsetof(struct sk_buff, sk));
9790 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9791 offsetof(struct sock_common,
9792 skc_v6_daddr.s6_addr32[0]) +
9793 off);
9794 #else
9795 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9796 #endif
9797 break;
9798 case offsetof(struct __sk_buff, local_ip6[0]) ...
9799 offsetof(struct __sk_buff, local_ip6[3]):
9800 #if IS_ENABLED(CONFIG_IPV6)
9801 BUILD_BUG_ON(sizeof_field(struct sock_common,
9802 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9803
9804 off = si->off;
9805 off -= offsetof(struct __sk_buff, local_ip6[0]);
9806
9807 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9808 si->dst_reg, si->src_reg,
9809 offsetof(struct sk_buff, sk));
9810 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9811 offsetof(struct sock_common,
9812 skc_v6_rcv_saddr.s6_addr32[0]) +
9813 off);
9814 #else
9815 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9816 #endif
9817 break;
9818
9819 case offsetof(struct __sk_buff, remote_port):
9820 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9821
9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9823 si->dst_reg, si->src_reg,
9824 offsetof(struct sk_buff, sk));
9825 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9826 bpf_target_off(struct sock_common,
9827 skc_dport,
9828 2, target_size));
9829 #ifndef __BIG_ENDIAN_BITFIELD
9830 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9831 #endif
9832 break;
9833
9834 case offsetof(struct __sk_buff, local_port):
9835 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9836
9837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9838 si->dst_reg, si->src_reg,
9839 offsetof(struct sk_buff, sk));
9840 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9841 bpf_target_off(struct sock_common,
9842 skc_num, 2, target_size));
9843 break;
9844
9845 case offsetof(struct __sk_buff, tstamp):
9846 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9847
9848 if (type == BPF_WRITE)
9849 insn = bpf_convert_tstamp_write(prog, si, insn);
9850 else
9851 insn = bpf_convert_tstamp_read(prog, si, insn);
9852 break;
9853
9854 case offsetof(struct __sk_buff, tstamp_type):
9855 insn = bpf_convert_tstamp_type_read(si, insn);
9856 break;
9857
9858 case offsetof(struct __sk_buff, gso_segs):
9859 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9860 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9861 si->dst_reg, si->dst_reg,
9862 bpf_target_off(struct skb_shared_info,
9863 gso_segs, 2,
9864 target_size));
9865 break;
9866 case offsetof(struct __sk_buff, gso_size):
9867 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9868 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9869 si->dst_reg, si->dst_reg,
9870 bpf_target_off(struct skb_shared_info,
9871 gso_size, 2,
9872 target_size));
9873 break;
9874 case offsetof(struct __sk_buff, wire_len):
9875 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9876
9877 off = si->off;
9878 off -= offsetof(struct __sk_buff, wire_len);
9879 off += offsetof(struct sk_buff, cb);
9880 off += offsetof(struct qdisc_skb_cb, pkt_len);
9881 *target_size = 4;
9882 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9883 break;
9884
9885 case offsetof(struct __sk_buff, sk):
9886 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9887 si->dst_reg, si->src_reg,
9888 offsetof(struct sk_buff, sk));
9889 break;
9890 case offsetof(struct __sk_buff, hwtstamp):
9891 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9892 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9893
9894 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9895 *insn++ = BPF_LDX_MEM(BPF_DW,
9896 si->dst_reg, si->dst_reg,
9897 bpf_target_off(struct skb_shared_info,
9898 hwtstamps, 8,
9899 target_size));
9900 break;
9901 }
9902
9903 return insn - insn_buf;
9904 }
9905
bpf_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9906 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9907 const struct bpf_insn *si,
9908 struct bpf_insn *insn_buf,
9909 struct bpf_prog *prog, u32 *target_size)
9910 {
9911 struct bpf_insn *insn = insn_buf;
9912 int off;
9913
9914 switch (si->off) {
9915 case offsetof(struct bpf_sock, bound_dev_if):
9916 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9917
9918 if (type == BPF_WRITE)
9919 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9920 offsetof(struct sock, sk_bound_dev_if));
9921 else
9922 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9923 offsetof(struct sock, sk_bound_dev_if));
9924 break;
9925
9926 case offsetof(struct bpf_sock, mark):
9927 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9928
9929 if (type == BPF_WRITE)
9930 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9931 offsetof(struct sock, sk_mark));
9932 else
9933 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9934 offsetof(struct sock, sk_mark));
9935 break;
9936
9937 case offsetof(struct bpf_sock, priority):
9938 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9939
9940 if (type == BPF_WRITE)
9941 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9942 offsetof(struct sock, sk_priority));
9943 else
9944 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9945 offsetof(struct sock, sk_priority));
9946 break;
9947
9948 case offsetof(struct bpf_sock, family):
9949 *insn++ = BPF_LDX_MEM(
9950 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9951 si->dst_reg, si->src_reg,
9952 bpf_target_off(struct sock_common,
9953 skc_family,
9954 sizeof_field(struct sock_common,
9955 skc_family),
9956 target_size));
9957 break;
9958
9959 case offsetof(struct bpf_sock, type):
9960 *insn++ = BPF_LDX_MEM(
9961 BPF_FIELD_SIZEOF(struct sock, sk_type),
9962 si->dst_reg, si->src_reg,
9963 bpf_target_off(struct sock, sk_type,
9964 sizeof_field(struct sock, sk_type),
9965 target_size));
9966 break;
9967
9968 case offsetof(struct bpf_sock, protocol):
9969 *insn++ = BPF_LDX_MEM(
9970 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9971 si->dst_reg, si->src_reg,
9972 bpf_target_off(struct sock, sk_protocol,
9973 sizeof_field(struct sock, sk_protocol),
9974 target_size));
9975 break;
9976
9977 case offsetof(struct bpf_sock, src_ip4):
9978 *insn++ = BPF_LDX_MEM(
9979 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9980 bpf_target_off(struct sock_common, skc_rcv_saddr,
9981 sizeof_field(struct sock_common,
9982 skc_rcv_saddr),
9983 target_size));
9984 break;
9985
9986 case offsetof(struct bpf_sock, dst_ip4):
9987 *insn++ = BPF_LDX_MEM(
9988 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9989 bpf_target_off(struct sock_common, skc_daddr,
9990 sizeof_field(struct sock_common,
9991 skc_daddr),
9992 target_size));
9993 break;
9994
9995 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9996 #if IS_ENABLED(CONFIG_IPV6)
9997 off = si->off;
9998 off -= offsetof(struct bpf_sock, src_ip6[0]);
9999 *insn++ = BPF_LDX_MEM(
10000 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10001 bpf_target_off(
10002 struct sock_common,
10003 skc_v6_rcv_saddr.s6_addr32[0],
10004 sizeof_field(struct sock_common,
10005 skc_v6_rcv_saddr.s6_addr32[0]),
10006 target_size) + off);
10007 #else
10008 (void)off;
10009 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10010 #endif
10011 break;
10012
10013 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
10014 #if IS_ENABLED(CONFIG_IPV6)
10015 off = si->off;
10016 off -= offsetof(struct bpf_sock, dst_ip6[0]);
10017 *insn++ = BPF_LDX_MEM(
10018 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10019 bpf_target_off(struct sock_common,
10020 skc_v6_daddr.s6_addr32[0],
10021 sizeof_field(struct sock_common,
10022 skc_v6_daddr.s6_addr32[0]),
10023 target_size) + off);
10024 #else
10025 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10026 *target_size = 4;
10027 #endif
10028 break;
10029
10030 case offsetof(struct bpf_sock, src_port):
10031 *insn++ = BPF_LDX_MEM(
10032 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
10033 si->dst_reg, si->src_reg,
10034 bpf_target_off(struct sock_common, skc_num,
10035 sizeof_field(struct sock_common,
10036 skc_num),
10037 target_size));
10038 break;
10039
10040 case offsetof(struct bpf_sock, dst_port):
10041 *insn++ = BPF_LDX_MEM(
10042 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
10043 si->dst_reg, si->src_reg,
10044 bpf_target_off(struct sock_common, skc_dport,
10045 sizeof_field(struct sock_common,
10046 skc_dport),
10047 target_size));
10048 break;
10049
10050 case offsetof(struct bpf_sock, state):
10051 *insn++ = BPF_LDX_MEM(
10052 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
10053 si->dst_reg, si->src_reg,
10054 bpf_target_off(struct sock_common, skc_state,
10055 sizeof_field(struct sock_common,
10056 skc_state),
10057 target_size));
10058 break;
10059 case offsetof(struct bpf_sock, rx_queue_mapping):
10060 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
10061 *insn++ = BPF_LDX_MEM(
10062 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
10063 si->dst_reg, si->src_reg,
10064 bpf_target_off(struct sock, sk_rx_queue_mapping,
10065 sizeof_field(struct sock,
10066 sk_rx_queue_mapping),
10067 target_size));
10068 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
10069 1);
10070 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10071 #else
10072 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10073 *target_size = 2;
10074 #endif
10075 break;
10076 }
10077
10078 return insn - insn_buf;
10079 }
10080
tc_cls_act_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10081 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
10082 const struct bpf_insn *si,
10083 struct bpf_insn *insn_buf,
10084 struct bpf_prog *prog, u32 *target_size)
10085 {
10086 struct bpf_insn *insn = insn_buf;
10087
10088 switch (si->off) {
10089 case offsetof(struct __sk_buff, ifindex):
10090 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
10091 si->dst_reg, si->src_reg,
10092 offsetof(struct sk_buff, dev));
10093 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10094 bpf_target_off(struct net_device, ifindex, 4,
10095 target_size));
10096 break;
10097 default:
10098 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10099 target_size);
10100 }
10101
10102 return insn - insn_buf;
10103 }
10104
xdp_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10105 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
10106 const struct bpf_insn *si,
10107 struct bpf_insn *insn_buf,
10108 struct bpf_prog *prog, u32 *target_size)
10109 {
10110 struct bpf_insn *insn = insn_buf;
10111
10112 switch (si->off) {
10113 case offsetof(struct xdp_md, data):
10114 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
10115 si->dst_reg, si->src_reg,
10116 offsetof(struct xdp_buff, data));
10117 break;
10118 case offsetof(struct xdp_md, data_meta):
10119 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
10120 si->dst_reg, si->src_reg,
10121 offsetof(struct xdp_buff, data_meta));
10122 break;
10123 case offsetof(struct xdp_md, data_end):
10124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
10125 si->dst_reg, si->src_reg,
10126 offsetof(struct xdp_buff, data_end));
10127 break;
10128 case offsetof(struct xdp_md, ingress_ifindex):
10129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10130 si->dst_reg, si->src_reg,
10131 offsetof(struct xdp_buff, rxq));
10132 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
10133 si->dst_reg, si->dst_reg,
10134 offsetof(struct xdp_rxq_info, dev));
10135 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10136 offsetof(struct net_device, ifindex));
10137 break;
10138 case offsetof(struct xdp_md, rx_queue_index):
10139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10140 si->dst_reg, si->src_reg,
10141 offsetof(struct xdp_buff, rxq));
10142 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10143 offsetof(struct xdp_rxq_info,
10144 queue_index));
10145 break;
10146 case offsetof(struct xdp_md, egress_ifindex):
10147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
10148 si->dst_reg, si->src_reg,
10149 offsetof(struct xdp_buff, txq));
10150 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10151 si->dst_reg, si->dst_reg,
10152 offsetof(struct xdp_txq_info, dev));
10153 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10154 offsetof(struct net_device, ifindex));
10155 break;
10156 }
10157
10158 return insn - insn_buf;
10159 }
10160
10161 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10162 * context Structure, F is Field in context structure that contains a pointer
10163 * to Nested Structure of type NS that has the field NF.
10164 *
10165 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10166 * sure that SIZE is not greater than actual size of S.F.NF.
10167 *
10168 * If offset OFF is provided, the load happens from that offset relative to
10169 * offset of NF.
10170 */
10171 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10172 do { \
10173 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10174 si->src_reg, offsetof(S, F)); \
10175 *insn++ = BPF_LDX_MEM( \
10176 SIZE, si->dst_reg, si->dst_reg, \
10177 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10178 target_size) \
10179 + OFF); \
10180 } while (0)
10181
10182 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10183 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10184 BPF_FIELD_SIZEOF(NS, NF), 0)
10185
10186 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10187 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10188 *
10189 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10190 * "register" since two registers available in convert_ctx_access are not
10191 * enough: we can't override neither SRC, since it contains value to store, nor
10192 * DST since it contains pointer to context that may be used by later
10193 * instructions. But we need a temporary place to save pointer to nested
10194 * structure whose field we want to store to.
10195 */
10196 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10197 do { \
10198 int tmp_reg = BPF_REG_9; \
10199 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10200 --tmp_reg; \
10201 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10202 --tmp_reg; \
10203 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10204 offsetof(S, TF)); \
10205 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10206 si->dst_reg, offsetof(S, F)); \
10207 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10208 tmp_reg, si->src_reg, \
10209 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10210 target_size) \
10211 + OFF, \
10212 si->imm); \
10213 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10214 offsetof(S, TF)); \
10215 } while (0)
10216
10217 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10218 TF) \
10219 do { \
10220 if (type == BPF_WRITE) { \
10221 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10222 OFF, TF); \
10223 } else { \
10224 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10225 S, NS, F, NF, SIZE, OFF); \
10226 } \
10227 } while (0)
10228
10229 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10230 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10231 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10232
sock_addr_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10233 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10234 const struct bpf_insn *si,
10235 struct bpf_insn *insn_buf,
10236 struct bpf_prog *prog, u32 *target_size)
10237 {
10238 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10239 struct bpf_insn *insn = insn_buf;
10240
10241 switch (si->off) {
10242 case offsetof(struct bpf_sock_addr, user_family):
10243 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10244 struct sockaddr, uaddr, sa_family);
10245 break;
10246
10247 case offsetof(struct bpf_sock_addr, user_ip4):
10248 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10249 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10250 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10251 break;
10252
10253 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10254 off = si->off;
10255 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10256 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10257 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10258 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10259 tmp_reg);
10260 break;
10261
10262 case offsetof(struct bpf_sock_addr, user_port):
10263 /* To get port we need to know sa_family first and then treat
10264 * sockaddr as either sockaddr_in or sockaddr_in6.
10265 * Though we can simplify since port field has same offset and
10266 * size in both structures.
10267 * Here we check this invariant and use just one of the
10268 * structures if it's true.
10269 */
10270 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10271 offsetof(struct sockaddr_in6, sin6_port));
10272 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10273 sizeof_field(struct sockaddr_in6, sin6_port));
10274 /* Account for sin6_port being smaller than user_port. */
10275 port_size = min(port_size, BPF_LDST_BYTES(si));
10276 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10277 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10278 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10279 break;
10280
10281 case offsetof(struct bpf_sock_addr, family):
10282 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10283 struct sock, sk, sk_family);
10284 break;
10285
10286 case offsetof(struct bpf_sock_addr, type):
10287 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10288 struct sock, sk, sk_type);
10289 break;
10290
10291 case offsetof(struct bpf_sock_addr, protocol):
10292 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10293 struct sock, sk, sk_protocol);
10294 break;
10295
10296 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10297 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10298 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10299 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10300 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10301 break;
10302
10303 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10304 msg_src_ip6[3]):
10305 off = si->off;
10306 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10307 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10308 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10309 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10310 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10311 break;
10312 case offsetof(struct bpf_sock_addr, sk):
10313 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10314 si->dst_reg, si->src_reg,
10315 offsetof(struct bpf_sock_addr_kern, sk));
10316 break;
10317 }
10318
10319 return insn - insn_buf;
10320 }
10321
sock_ops_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10322 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10323 const struct bpf_insn *si,
10324 struct bpf_insn *insn_buf,
10325 struct bpf_prog *prog,
10326 u32 *target_size)
10327 {
10328 struct bpf_insn *insn = insn_buf;
10329 int off;
10330
10331 /* Helper macro for adding read access to tcp_sock or sock fields. */
10332 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10333 do { \
10334 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10335 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10336 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10337 if (si->dst_reg == reg || si->src_reg == reg) \
10338 reg--; \
10339 if (si->dst_reg == reg || si->src_reg == reg) \
10340 reg--; \
10341 if (si->dst_reg == si->src_reg) { \
10342 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10343 offsetof(struct bpf_sock_ops_kern, \
10344 temp)); \
10345 fullsock_reg = reg; \
10346 jmp += 2; \
10347 } \
10348 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10349 struct bpf_sock_ops_kern, \
10350 is_fullsock), \
10351 fullsock_reg, si->src_reg, \
10352 offsetof(struct bpf_sock_ops_kern, \
10353 is_fullsock)); \
10354 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10355 if (si->dst_reg == si->src_reg) \
10356 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10357 offsetof(struct bpf_sock_ops_kern, \
10358 temp)); \
10359 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10360 struct bpf_sock_ops_kern, sk),\
10361 si->dst_reg, si->src_reg, \
10362 offsetof(struct bpf_sock_ops_kern, sk));\
10363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10364 OBJ_FIELD), \
10365 si->dst_reg, si->dst_reg, \
10366 offsetof(OBJ, OBJ_FIELD)); \
10367 if (si->dst_reg == si->src_reg) { \
10368 *insn++ = BPF_JMP_A(1); \
10369 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10370 offsetof(struct bpf_sock_ops_kern, \
10371 temp)); \
10372 } \
10373 } while (0)
10374
10375 #define SOCK_OPS_GET_SK() \
10376 do { \
10377 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10378 if (si->dst_reg == reg || si->src_reg == reg) \
10379 reg--; \
10380 if (si->dst_reg == reg || si->src_reg == reg) \
10381 reg--; \
10382 if (si->dst_reg == si->src_reg) { \
10383 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10384 offsetof(struct bpf_sock_ops_kern, \
10385 temp)); \
10386 fullsock_reg = reg; \
10387 jmp += 2; \
10388 } \
10389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10390 struct bpf_sock_ops_kern, \
10391 is_fullsock), \
10392 fullsock_reg, si->src_reg, \
10393 offsetof(struct bpf_sock_ops_kern, \
10394 is_fullsock)); \
10395 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10396 if (si->dst_reg == si->src_reg) \
10397 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10398 offsetof(struct bpf_sock_ops_kern, \
10399 temp)); \
10400 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10401 struct bpf_sock_ops_kern, sk),\
10402 si->dst_reg, si->src_reg, \
10403 offsetof(struct bpf_sock_ops_kern, sk));\
10404 if (si->dst_reg == si->src_reg) { \
10405 *insn++ = BPF_JMP_A(1); \
10406 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10407 offsetof(struct bpf_sock_ops_kern, \
10408 temp)); \
10409 } \
10410 } while (0)
10411
10412 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10413 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10414
10415 /* Helper macro for adding write access to tcp_sock or sock fields.
10416 * The macro is called with two registers, dst_reg which contains a pointer
10417 * to ctx (context) and src_reg which contains the value that should be
10418 * stored. However, we need an additional register since we cannot overwrite
10419 * dst_reg because it may be used later in the program.
10420 * Instead we "borrow" one of the other register. We first save its value
10421 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10422 * it at the end of the macro.
10423 */
10424 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10425 do { \
10426 int reg = BPF_REG_9; \
10427 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10428 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10429 if (si->dst_reg == reg || si->src_reg == reg) \
10430 reg--; \
10431 if (si->dst_reg == reg || si->src_reg == reg) \
10432 reg--; \
10433 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10434 offsetof(struct bpf_sock_ops_kern, \
10435 temp)); \
10436 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10437 struct bpf_sock_ops_kern, \
10438 is_fullsock), \
10439 reg, si->dst_reg, \
10440 offsetof(struct bpf_sock_ops_kern, \
10441 is_fullsock)); \
10442 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10444 struct bpf_sock_ops_kern, sk),\
10445 reg, si->dst_reg, \
10446 offsetof(struct bpf_sock_ops_kern, sk));\
10447 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10448 BPF_MEM | BPF_CLASS(si->code), \
10449 reg, si->src_reg, \
10450 offsetof(OBJ, OBJ_FIELD), \
10451 si->imm); \
10452 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10453 offsetof(struct bpf_sock_ops_kern, \
10454 temp)); \
10455 } while (0)
10456
10457 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10458 do { \
10459 if (TYPE == BPF_WRITE) \
10460 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10461 else \
10462 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10463 } while (0)
10464
10465 switch (si->off) {
10466 case offsetof(struct bpf_sock_ops, op):
10467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10468 op),
10469 si->dst_reg, si->src_reg,
10470 offsetof(struct bpf_sock_ops_kern, op));
10471 break;
10472
10473 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10474 offsetof(struct bpf_sock_ops, replylong[3]):
10475 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10476 sizeof_field(struct bpf_sock_ops_kern, reply));
10477 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10478 sizeof_field(struct bpf_sock_ops_kern, replylong));
10479 off = si->off;
10480 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10481 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10482 if (type == BPF_WRITE)
10483 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10484 else
10485 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10486 off);
10487 break;
10488
10489 case offsetof(struct bpf_sock_ops, family):
10490 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10491
10492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10493 struct bpf_sock_ops_kern, sk),
10494 si->dst_reg, si->src_reg,
10495 offsetof(struct bpf_sock_ops_kern, sk));
10496 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10497 offsetof(struct sock_common, skc_family));
10498 break;
10499
10500 case offsetof(struct bpf_sock_ops, remote_ip4):
10501 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10502
10503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10504 struct bpf_sock_ops_kern, sk),
10505 si->dst_reg, si->src_reg,
10506 offsetof(struct bpf_sock_ops_kern, sk));
10507 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10508 offsetof(struct sock_common, skc_daddr));
10509 break;
10510
10511 case offsetof(struct bpf_sock_ops, local_ip4):
10512 BUILD_BUG_ON(sizeof_field(struct sock_common,
10513 skc_rcv_saddr) != 4);
10514
10515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10516 struct bpf_sock_ops_kern, sk),
10517 si->dst_reg, si->src_reg,
10518 offsetof(struct bpf_sock_ops_kern, sk));
10519 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10520 offsetof(struct sock_common,
10521 skc_rcv_saddr));
10522 break;
10523
10524 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10525 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10526 #if IS_ENABLED(CONFIG_IPV6)
10527 BUILD_BUG_ON(sizeof_field(struct sock_common,
10528 skc_v6_daddr.s6_addr32[0]) != 4);
10529
10530 off = si->off;
10531 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10532 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10533 struct bpf_sock_ops_kern, sk),
10534 si->dst_reg, si->src_reg,
10535 offsetof(struct bpf_sock_ops_kern, sk));
10536 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10537 offsetof(struct sock_common,
10538 skc_v6_daddr.s6_addr32[0]) +
10539 off);
10540 #else
10541 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10542 #endif
10543 break;
10544
10545 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10546 offsetof(struct bpf_sock_ops, local_ip6[3]):
10547 #if IS_ENABLED(CONFIG_IPV6)
10548 BUILD_BUG_ON(sizeof_field(struct sock_common,
10549 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10550
10551 off = si->off;
10552 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10553 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10554 struct bpf_sock_ops_kern, sk),
10555 si->dst_reg, si->src_reg,
10556 offsetof(struct bpf_sock_ops_kern, sk));
10557 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10558 offsetof(struct sock_common,
10559 skc_v6_rcv_saddr.s6_addr32[0]) +
10560 off);
10561 #else
10562 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10563 #endif
10564 break;
10565
10566 case offsetof(struct bpf_sock_ops, remote_port):
10567 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10568
10569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10570 struct bpf_sock_ops_kern, sk),
10571 si->dst_reg, si->src_reg,
10572 offsetof(struct bpf_sock_ops_kern, sk));
10573 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10574 offsetof(struct sock_common, skc_dport));
10575 #ifndef __BIG_ENDIAN_BITFIELD
10576 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10577 #endif
10578 break;
10579
10580 case offsetof(struct bpf_sock_ops, local_port):
10581 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10582
10583 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10584 struct bpf_sock_ops_kern, sk),
10585 si->dst_reg, si->src_reg,
10586 offsetof(struct bpf_sock_ops_kern, sk));
10587 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10588 offsetof(struct sock_common, skc_num));
10589 break;
10590
10591 case offsetof(struct bpf_sock_ops, is_fullsock):
10592 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10593 struct bpf_sock_ops_kern,
10594 is_fullsock),
10595 si->dst_reg, si->src_reg,
10596 offsetof(struct bpf_sock_ops_kern,
10597 is_fullsock));
10598 break;
10599
10600 case offsetof(struct bpf_sock_ops, state):
10601 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10602
10603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10604 struct bpf_sock_ops_kern, sk),
10605 si->dst_reg, si->src_reg,
10606 offsetof(struct bpf_sock_ops_kern, sk));
10607 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10608 offsetof(struct sock_common, skc_state));
10609 break;
10610
10611 case offsetof(struct bpf_sock_ops, rtt_min):
10612 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10613 sizeof(struct minmax));
10614 BUILD_BUG_ON(sizeof(struct minmax) <
10615 sizeof(struct minmax_sample));
10616
10617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10618 struct bpf_sock_ops_kern, sk),
10619 si->dst_reg, si->src_reg,
10620 offsetof(struct bpf_sock_ops_kern, sk));
10621 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10622 offsetof(struct tcp_sock, rtt_min) +
10623 sizeof_field(struct minmax_sample, t));
10624 break;
10625
10626 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10627 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10628 struct tcp_sock);
10629 break;
10630
10631 case offsetof(struct bpf_sock_ops, sk_txhash):
10632 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10633 struct sock, type);
10634 break;
10635 case offsetof(struct bpf_sock_ops, snd_cwnd):
10636 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10637 break;
10638 case offsetof(struct bpf_sock_ops, srtt_us):
10639 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10640 break;
10641 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10642 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10643 break;
10644 case offsetof(struct bpf_sock_ops, rcv_nxt):
10645 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10646 break;
10647 case offsetof(struct bpf_sock_ops, snd_nxt):
10648 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10649 break;
10650 case offsetof(struct bpf_sock_ops, snd_una):
10651 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10652 break;
10653 case offsetof(struct bpf_sock_ops, mss_cache):
10654 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10655 break;
10656 case offsetof(struct bpf_sock_ops, ecn_flags):
10657 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10658 break;
10659 case offsetof(struct bpf_sock_ops, rate_delivered):
10660 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10661 break;
10662 case offsetof(struct bpf_sock_ops, rate_interval_us):
10663 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10664 break;
10665 case offsetof(struct bpf_sock_ops, packets_out):
10666 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10667 break;
10668 case offsetof(struct bpf_sock_ops, retrans_out):
10669 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10670 break;
10671 case offsetof(struct bpf_sock_ops, total_retrans):
10672 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10673 break;
10674 case offsetof(struct bpf_sock_ops, segs_in):
10675 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10676 break;
10677 case offsetof(struct bpf_sock_ops, data_segs_in):
10678 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10679 break;
10680 case offsetof(struct bpf_sock_ops, segs_out):
10681 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10682 break;
10683 case offsetof(struct bpf_sock_ops, data_segs_out):
10684 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10685 break;
10686 case offsetof(struct bpf_sock_ops, lost_out):
10687 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10688 break;
10689 case offsetof(struct bpf_sock_ops, sacked_out):
10690 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10691 break;
10692 case offsetof(struct bpf_sock_ops, bytes_received):
10693 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10694 break;
10695 case offsetof(struct bpf_sock_ops, bytes_acked):
10696 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10697 break;
10698 case offsetof(struct bpf_sock_ops, sk):
10699 SOCK_OPS_GET_SK();
10700 break;
10701 case offsetof(struct bpf_sock_ops, skb_data_end):
10702 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10703 skb_data_end),
10704 si->dst_reg, si->src_reg,
10705 offsetof(struct bpf_sock_ops_kern,
10706 skb_data_end));
10707 break;
10708 case offsetof(struct bpf_sock_ops, skb_data):
10709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10710 skb),
10711 si->dst_reg, si->src_reg,
10712 offsetof(struct bpf_sock_ops_kern,
10713 skb));
10714 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10716 si->dst_reg, si->dst_reg,
10717 offsetof(struct sk_buff, data));
10718 break;
10719 case offsetof(struct bpf_sock_ops, skb_len):
10720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10721 skb),
10722 si->dst_reg, si->src_reg,
10723 offsetof(struct bpf_sock_ops_kern,
10724 skb));
10725 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10726 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10727 si->dst_reg, si->dst_reg,
10728 offsetof(struct sk_buff, len));
10729 break;
10730 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10731 off = offsetof(struct sk_buff, cb);
10732 off += offsetof(struct tcp_skb_cb, tcp_flags);
10733 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10735 skb),
10736 si->dst_reg, si->src_reg,
10737 offsetof(struct bpf_sock_ops_kern,
10738 skb));
10739 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10740 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10741 tcp_flags),
10742 si->dst_reg, si->dst_reg, off);
10743 break;
10744 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10745 struct bpf_insn *jmp_on_null_skb;
10746
10747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10748 skb),
10749 si->dst_reg, si->src_reg,
10750 offsetof(struct bpf_sock_ops_kern,
10751 skb));
10752 /* Reserve one insn to test skb == NULL */
10753 jmp_on_null_skb = insn++;
10754 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10755 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10756 bpf_target_off(struct skb_shared_info,
10757 hwtstamps, 8,
10758 target_size));
10759 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10760 insn - jmp_on_null_skb - 1);
10761 break;
10762 }
10763 }
10764 return insn - insn_buf;
10765 }
10766
10767 /* data_end = skb->data + skb_headlen() */
bpf_convert_data_end_access(const struct bpf_insn * si,struct bpf_insn * insn)10768 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10769 struct bpf_insn *insn)
10770 {
10771 int reg;
10772 int temp_reg_off = offsetof(struct sk_buff, cb) +
10773 offsetof(struct sk_skb_cb, temp_reg);
10774
10775 if (si->src_reg == si->dst_reg) {
10776 /* We need an extra register, choose and save a register. */
10777 reg = BPF_REG_9;
10778 if (si->src_reg == reg || si->dst_reg == reg)
10779 reg--;
10780 if (si->src_reg == reg || si->dst_reg == reg)
10781 reg--;
10782 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10783 } else {
10784 reg = si->dst_reg;
10785 }
10786
10787 /* reg = skb->data */
10788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10789 reg, si->src_reg,
10790 offsetof(struct sk_buff, data));
10791 /* AX = skb->len */
10792 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10793 BPF_REG_AX, si->src_reg,
10794 offsetof(struct sk_buff, len));
10795 /* reg = skb->data + skb->len */
10796 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10797 /* AX = skb->data_len */
10798 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10799 BPF_REG_AX, si->src_reg,
10800 offsetof(struct sk_buff, data_len));
10801
10802 /* reg = skb->data + skb->len - skb->data_len */
10803 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10804
10805 if (si->src_reg == si->dst_reg) {
10806 /* Restore the saved register */
10807 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10808 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10809 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10810 }
10811
10812 return insn;
10813 }
10814
sk_skb_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10815 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10816 const struct bpf_insn *si,
10817 struct bpf_insn *insn_buf,
10818 struct bpf_prog *prog, u32 *target_size)
10819 {
10820 struct bpf_insn *insn = insn_buf;
10821 int off;
10822
10823 switch (si->off) {
10824 case offsetof(struct __sk_buff, data_end):
10825 insn = bpf_convert_data_end_access(si, insn);
10826 break;
10827 case offsetof(struct __sk_buff, cb[0]) ...
10828 offsetofend(struct __sk_buff, cb[4]) - 1:
10829 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10830 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10831 offsetof(struct sk_skb_cb, data)) %
10832 sizeof(__u64));
10833
10834 prog->cb_access = 1;
10835 off = si->off;
10836 off -= offsetof(struct __sk_buff, cb[0]);
10837 off += offsetof(struct sk_buff, cb);
10838 off += offsetof(struct sk_skb_cb, data);
10839 if (type == BPF_WRITE)
10840 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10841 else
10842 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10843 si->src_reg, off);
10844 break;
10845
10846
10847 default:
10848 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10849 target_size);
10850 }
10851
10852 return insn - insn_buf;
10853 }
10854
sk_msg_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10855 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10856 const struct bpf_insn *si,
10857 struct bpf_insn *insn_buf,
10858 struct bpf_prog *prog, u32 *target_size)
10859 {
10860 struct bpf_insn *insn = insn_buf;
10861 #if IS_ENABLED(CONFIG_IPV6)
10862 int off;
10863 #endif
10864
10865 /* convert ctx uses the fact sg element is first in struct */
10866 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10867
10868 switch (si->off) {
10869 case offsetof(struct sk_msg_md, data):
10870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10871 si->dst_reg, si->src_reg,
10872 offsetof(struct sk_msg, data));
10873 break;
10874 case offsetof(struct sk_msg_md, data_end):
10875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10876 si->dst_reg, si->src_reg,
10877 offsetof(struct sk_msg, data_end));
10878 break;
10879 case offsetof(struct sk_msg_md, family):
10880 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10881
10882 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10883 struct sk_msg, sk),
10884 si->dst_reg, si->src_reg,
10885 offsetof(struct sk_msg, sk));
10886 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10887 offsetof(struct sock_common, skc_family));
10888 break;
10889
10890 case offsetof(struct sk_msg_md, remote_ip4):
10891 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10892
10893 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10894 struct sk_msg, sk),
10895 si->dst_reg, si->src_reg,
10896 offsetof(struct sk_msg, sk));
10897 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10898 offsetof(struct sock_common, skc_daddr));
10899 break;
10900
10901 case offsetof(struct sk_msg_md, local_ip4):
10902 BUILD_BUG_ON(sizeof_field(struct sock_common,
10903 skc_rcv_saddr) != 4);
10904
10905 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10906 struct sk_msg, sk),
10907 si->dst_reg, si->src_reg,
10908 offsetof(struct sk_msg, sk));
10909 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10910 offsetof(struct sock_common,
10911 skc_rcv_saddr));
10912 break;
10913
10914 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10915 offsetof(struct sk_msg_md, remote_ip6[3]):
10916 #if IS_ENABLED(CONFIG_IPV6)
10917 BUILD_BUG_ON(sizeof_field(struct sock_common,
10918 skc_v6_daddr.s6_addr32[0]) != 4);
10919
10920 off = si->off;
10921 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10922 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10923 struct sk_msg, sk),
10924 si->dst_reg, si->src_reg,
10925 offsetof(struct sk_msg, sk));
10926 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10927 offsetof(struct sock_common,
10928 skc_v6_daddr.s6_addr32[0]) +
10929 off);
10930 #else
10931 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10932 #endif
10933 break;
10934
10935 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10936 offsetof(struct sk_msg_md, local_ip6[3]):
10937 #if IS_ENABLED(CONFIG_IPV6)
10938 BUILD_BUG_ON(sizeof_field(struct sock_common,
10939 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10940
10941 off = si->off;
10942 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10943 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10944 struct sk_msg, sk),
10945 si->dst_reg, si->src_reg,
10946 offsetof(struct sk_msg, sk));
10947 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10948 offsetof(struct sock_common,
10949 skc_v6_rcv_saddr.s6_addr32[0]) +
10950 off);
10951 #else
10952 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10953 #endif
10954 break;
10955
10956 case offsetof(struct sk_msg_md, remote_port):
10957 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10958
10959 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10960 struct sk_msg, sk),
10961 si->dst_reg, si->src_reg,
10962 offsetof(struct sk_msg, sk));
10963 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10964 offsetof(struct sock_common, skc_dport));
10965 #ifndef __BIG_ENDIAN_BITFIELD
10966 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10967 #endif
10968 break;
10969
10970 case offsetof(struct sk_msg_md, local_port):
10971 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10972
10973 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10974 struct sk_msg, sk),
10975 si->dst_reg, si->src_reg,
10976 offsetof(struct sk_msg, sk));
10977 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10978 offsetof(struct sock_common, skc_num));
10979 break;
10980
10981 case offsetof(struct sk_msg_md, size):
10982 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10983 si->dst_reg, si->src_reg,
10984 offsetof(struct sk_msg_sg, size));
10985 break;
10986
10987 case offsetof(struct sk_msg_md, sk):
10988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10989 si->dst_reg, si->src_reg,
10990 offsetof(struct sk_msg, sk));
10991 break;
10992 }
10993
10994 return insn - insn_buf;
10995 }
10996
10997 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10998 .get_func_proto = sk_filter_func_proto,
10999 .is_valid_access = sk_filter_is_valid_access,
11000 .convert_ctx_access = bpf_convert_ctx_access,
11001 .gen_ld_abs = bpf_gen_ld_abs,
11002 };
11003
11004 const struct bpf_prog_ops sk_filter_prog_ops = {
11005 .test_run = bpf_prog_test_run_skb,
11006 };
11007
11008 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
11009 .get_func_proto = tc_cls_act_func_proto,
11010 .is_valid_access = tc_cls_act_is_valid_access,
11011 .convert_ctx_access = tc_cls_act_convert_ctx_access,
11012 .gen_prologue = tc_cls_act_prologue,
11013 .gen_ld_abs = bpf_gen_ld_abs,
11014 .btf_struct_access = tc_cls_act_btf_struct_access,
11015 };
11016
11017 const struct bpf_prog_ops tc_cls_act_prog_ops = {
11018 .test_run = bpf_prog_test_run_skb,
11019 };
11020
11021 const struct bpf_verifier_ops xdp_verifier_ops = {
11022 .get_func_proto = xdp_func_proto,
11023 .is_valid_access = xdp_is_valid_access,
11024 .convert_ctx_access = xdp_convert_ctx_access,
11025 .gen_prologue = bpf_noop_prologue,
11026 .btf_struct_access = xdp_btf_struct_access,
11027 };
11028
11029 const struct bpf_prog_ops xdp_prog_ops = {
11030 .test_run = bpf_prog_test_run_xdp,
11031 };
11032
11033 const struct bpf_verifier_ops cg_skb_verifier_ops = {
11034 .get_func_proto = cg_skb_func_proto,
11035 .is_valid_access = cg_skb_is_valid_access,
11036 .convert_ctx_access = bpf_convert_ctx_access,
11037 };
11038
11039 const struct bpf_prog_ops cg_skb_prog_ops = {
11040 .test_run = bpf_prog_test_run_skb,
11041 };
11042
11043 const struct bpf_verifier_ops lwt_in_verifier_ops = {
11044 .get_func_proto = lwt_in_func_proto,
11045 .is_valid_access = lwt_is_valid_access,
11046 .convert_ctx_access = bpf_convert_ctx_access,
11047 };
11048
11049 const struct bpf_prog_ops lwt_in_prog_ops = {
11050 .test_run = bpf_prog_test_run_skb,
11051 };
11052
11053 const struct bpf_verifier_ops lwt_out_verifier_ops = {
11054 .get_func_proto = lwt_out_func_proto,
11055 .is_valid_access = lwt_is_valid_access,
11056 .convert_ctx_access = bpf_convert_ctx_access,
11057 };
11058
11059 const struct bpf_prog_ops lwt_out_prog_ops = {
11060 .test_run = bpf_prog_test_run_skb,
11061 };
11062
11063 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
11064 .get_func_proto = lwt_xmit_func_proto,
11065 .is_valid_access = lwt_is_valid_access,
11066 .convert_ctx_access = bpf_convert_ctx_access,
11067 .gen_prologue = tc_cls_act_prologue,
11068 };
11069
11070 const struct bpf_prog_ops lwt_xmit_prog_ops = {
11071 .test_run = bpf_prog_test_run_skb,
11072 };
11073
11074 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
11075 .get_func_proto = lwt_seg6local_func_proto,
11076 .is_valid_access = lwt_is_valid_access,
11077 .convert_ctx_access = bpf_convert_ctx_access,
11078 };
11079
11080 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
11081 };
11082
11083 const struct bpf_verifier_ops cg_sock_verifier_ops = {
11084 .get_func_proto = sock_filter_func_proto,
11085 .is_valid_access = sock_filter_is_valid_access,
11086 .convert_ctx_access = bpf_sock_convert_ctx_access,
11087 };
11088
11089 const struct bpf_prog_ops cg_sock_prog_ops = {
11090 };
11091
11092 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
11093 .get_func_proto = sock_addr_func_proto,
11094 .is_valid_access = sock_addr_is_valid_access,
11095 .convert_ctx_access = sock_addr_convert_ctx_access,
11096 };
11097
11098 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
11099 };
11100
11101 const struct bpf_verifier_ops sock_ops_verifier_ops = {
11102 .get_func_proto = sock_ops_func_proto,
11103 .is_valid_access = sock_ops_is_valid_access,
11104 .convert_ctx_access = sock_ops_convert_ctx_access,
11105 };
11106
11107 const struct bpf_prog_ops sock_ops_prog_ops = {
11108 };
11109
11110 const struct bpf_verifier_ops sk_skb_verifier_ops = {
11111 .get_func_proto = sk_skb_func_proto,
11112 .is_valid_access = sk_skb_is_valid_access,
11113 .convert_ctx_access = sk_skb_convert_ctx_access,
11114 .gen_prologue = sk_skb_prologue,
11115 };
11116
11117 const struct bpf_prog_ops sk_skb_prog_ops = {
11118 };
11119
11120 const struct bpf_verifier_ops sk_msg_verifier_ops = {
11121 .get_func_proto = sk_msg_func_proto,
11122 .is_valid_access = sk_msg_is_valid_access,
11123 .convert_ctx_access = sk_msg_convert_ctx_access,
11124 .gen_prologue = bpf_noop_prologue,
11125 };
11126
11127 const struct bpf_prog_ops sk_msg_prog_ops = {
11128 };
11129
11130 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
11131 .get_func_proto = flow_dissector_func_proto,
11132 .is_valid_access = flow_dissector_is_valid_access,
11133 .convert_ctx_access = flow_dissector_convert_ctx_access,
11134 };
11135
11136 const struct bpf_prog_ops flow_dissector_prog_ops = {
11137 .test_run = bpf_prog_test_run_flow_dissector,
11138 };
11139
sk_detach_filter(struct sock * sk)11140 int sk_detach_filter(struct sock *sk)
11141 {
11142 int ret = -ENOENT;
11143 struct sk_filter *filter;
11144
11145 if (sock_flag(sk, SOCK_FILTER_LOCKED))
11146 return -EPERM;
11147
11148 filter = rcu_dereference_protected(sk->sk_filter,
11149 lockdep_sock_is_held(sk));
11150 if (filter) {
11151 RCU_INIT_POINTER(sk->sk_filter, NULL);
11152 sk_filter_uncharge(sk, filter);
11153 ret = 0;
11154 }
11155
11156 return ret;
11157 }
11158 EXPORT_SYMBOL_GPL(sk_detach_filter);
11159
sk_get_filter(struct sock * sk,sockptr_t optval,unsigned int len)11160 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11161 {
11162 struct sock_fprog_kern *fprog;
11163 struct sk_filter *filter;
11164 int ret = 0;
11165
11166 sockopt_lock_sock(sk);
11167 filter = rcu_dereference_protected(sk->sk_filter,
11168 lockdep_sock_is_held(sk));
11169 if (!filter)
11170 goto out;
11171
11172 /* We're copying the filter that has been originally attached,
11173 * so no conversion/decode needed anymore. eBPF programs that
11174 * have no original program cannot be dumped through this.
11175 */
11176 ret = -EACCES;
11177 fprog = filter->prog->orig_prog;
11178 if (!fprog)
11179 goto out;
11180
11181 ret = fprog->len;
11182 if (!len)
11183 /* User space only enquires number of filter blocks. */
11184 goto out;
11185
11186 ret = -EINVAL;
11187 if (len < fprog->len)
11188 goto out;
11189
11190 ret = -EFAULT;
11191 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11192 goto out;
11193
11194 /* Instead of bytes, the API requests to return the number
11195 * of filter blocks.
11196 */
11197 ret = fprog->len;
11198 out:
11199 sockopt_release_sock(sk);
11200 return ret;
11201 }
11202
11203 #ifdef CONFIG_INET
bpf_init_reuseport_kern(struct sk_reuseport_kern * reuse_kern,struct sock_reuseport * reuse,struct sock * sk,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11204 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11205 struct sock_reuseport *reuse,
11206 struct sock *sk, struct sk_buff *skb,
11207 struct sock *migrating_sk,
11208 u32 hash)
11209 {
11210 reuse_kern->skb = skb;
11211 reuse_kern->sk = sk;
11212 reuse_kern->selected_sk = NULL;
11213 reuse_kern->migrating_sk = migrating_sk;
11214 reuse_kern->data_end = skb->data + skb_headlen(skb);
11215 reuse_kern->hash = hash;
11216 reuse_kern->reuseport_id = reuse->reuseport_id;
11217 reuse_kern->bind_inany = reuse->bind_inany;
11218 }
11219
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11220 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11221 struct bpf_prog *prog, struct sk_buff *skb,
11222 struct sock *migrating_sk,
11223 u32 hash)
11224 {
11225 struct sk_reuseport_kern reuse_kern;
11226 enum sk_action action;
11227
11228 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11229 action = bpf_prog_run(prog, &reuse_kern);
11230
11231 if (action == SK_PASS)
11232 return reuse_kern.selected_sk;
11233 else
11234 return ERR_PTR(-ECONNREFUSED);
11235 }
11236
BPF_CALL_4(sk_select_reuseport,struct sk_reuseport_kern *,reuse_kern,struct bpf_map *,map,void *,key,u32,flags)11237 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11238 struct bpf_map *, map, void *, key, u32, flags)
11239 {
11240 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11241 struct sock_reuseport *reuse;
11242 struct sock *selected_sk;
11243
11244 selected_sk = map->ops->map_lookup_elem(map, key);
11245 if (!selected_sk)
11246 return -ENOENT;
11247
11248 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11249 if (!reuse) {
11250 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11251 if (sk_is_refcounted(selected_sk))
11252 sock_put(selected_sk);
11253
11254 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11255 * The only (!reuse) case here is - the sk has already been
11256 * unhashed (e.g. by close()), so treat it as -ENOENT.
11257 *
11258 * Other maps (e.g. sock_map) do not provide this guarantee and
11259 * the sk may never be in the reuseport group to begin with.
11260 */
11261 return is_sockarray ? -ENOENT : -EINVAL;
11262 }
11263
11264 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11265 struct sock *sk = reuse_kern->sk;
11266
11267 if (sk->sk_protocol != selected_sk->sk_protocol)
11268 return -EPROTOTYPE;
11269 else if (sk->sk_family != selected_sk->sk_family)
11270 return -EAFNOSUPPORT;
11271
11272 /* Catch all. Likely bound to a different sockaddr. */
11273 return -EBADFD;
11274 }
11275
11276 reuse_kern->selected_sk = selected_sk;
11277
11278 return 0;
11279 }
11280
11281 static const struct bpf_func_proto sk_select_reuseport_proto = {
11282 .func = sk_select_reuseport,
11283 .gpl_only = false,
11284 .ret_type = RET_INTEGER,
11285 .arg1_type = ARG_PTR_TO_CTX,
11286 .arg2_type = ARG_CONST_MAP_PTR,
11287 .arg3_type = ARG_PTR_TO_MAP_KEY,
11288 .arg4_type = ARG_ANYTHING,
11289 };
11290
BPF_CALL_4(sk_reuseport_load_bytes,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len)11291 BPF_CALL_4(sk_reuseport_load_bytes,
11292 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11293 void *, to, u32, len)
11294 {
11295 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11296 }
11297
11298 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11299 .func = sk_reuseport_load_bytes,
11300 .gpl_only = false,
11301 .ret_type = RET_INTEGER,
11302 .arg1_type = ARG_PTR_TO_CTX,
11303 .arg2_type = ARG_ANYTHING,
11304 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11305 .arg4_type = ARG_CONST_SIZE,
11306 };
11307
BPF_CALL_5(sk_reuseport_load_bytes_relative,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len,u32,start_header)11308 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11309 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11310 void *, to, u32, len, u32, start_header)
11311 {
11312 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11313 len, start_header);
11314 }
11315
11316 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11317 .func = sk_reuseport_load_bytes_relative,
11318 .gpl_only = false,
11319 .ret_type = RET_INTEGER,
11320 .arg1_type = ARG_PTR_TO_CTX,
11321 .arg2_type = ARG_ANYTHING,
11322 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11323 .arg4_type = ARG_CONST_SIZE,
11324 .arg5_type = ARG_ANYTHING,
11325 };
11326
11327 static const struct bpf_func_proto *
sk_reuseport_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11328 sk_reuseport_func_proto(enum bpf_func_id func_id,
11329 const struct bpf_prog *prog)
11330 {
11331 switch (func_id) {
11332 case BPF_FUNC_sk_select_reuseport:
11333 return &sk_select_reuseport_proto;
11334 case BPF_FUNC_skb_load_bytes:
11335 return &sk_reuseport_load_bytes_proto;
11336 case BPF_FUNC_skb_load_bytes_relative:
11337 return &sk_reuseport_load_bytes_relative_proto;
11338 case BPF_FUNC_get_socket_cookie:
11339 return &bpf_get_socket_ptr_cookie_proto;
11340 case BPF_FUNC_ktime_get_coarse_ns:
11341 return &bpf_ktime_get_coarse_ns_proto;
11342 default:
11343 return bpf_base_func_proto(func_id, prog);
11344 }
11345 }
11346
11347 static bool
sk_reuseport_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11348 sk_reuseport_is_valid_access(int off, int size,
11349 enum bpf_access_type type,
11350 const struct bpf_prog *prog,
11351 struct bpf_insn_access_aux *info)
11352 {
11353 const u32 size_default = sizeof(__u32);
11354
11355 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11356 off % size || type != BPF_READ)
11357 return false;
11358
11359 switch (off) {
11360 case offsetof(struct sk_reuseport_md, data):
11361 info->reg_type = PTR_TO_PACKET;
11362 return size == sizeof(__u64);
11363
11364 case offsetof(struct sk_reuseport_md, data_end):
11365 info->reg_type = PTR_TO_PACKET_END;
11366 return size == sizeof(__u64);
11367
11368 case offsetof(struct sk_reuseport_md, hash):
11369 return size == size_default;
11370
11371 case offsetof(struct sk_reuseport_md, sk):
11372 info->reg_type = PTR_TO_SOCKET;
11373 return size == sizeof(__u64);
11374
11375 case offsetof(struct sk_reuseport_md, migrating_sk):
11376 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11377 return size == sizeof(__u64);
11378
11379 /* Fields that allow narrowing */
11380 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11381 if (size < sizeof_field(struct sk_buff, protocol))
11382 return false;
11383 fallthrough;
11384 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11385 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11386 case bpf_ctx_range(struct sk_reuseport_md, len):
11387 bpf_ctx_record_field_size(info, size_default);
11388 return bpf_ctx_narrow_access_ok(off, size, size_default);
11389
11390 default:
11391 return false;
11392 }
11393 }
11394
11395 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11396 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11397 si->dst_reg, si->src_reg, \
11398 bpf_target_off(struct sk_reuseport_kern, F, \
11399 sizeof_field(struct sk_reuseport_kern, F), \
11400 target_size)); \
11401 })
11402
11403 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11404 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11405 struct sk_buff, \
11406 skb, \
11407 SKB_FIELD)
11408
11409 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11410 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11411 struct sock, \
11412 sk, \
11413 SK_FIELD)
11414
sk_reuseport_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11415 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11416 const struct bpf_insn *si,
11417 struct bpf_insn *insn_buf,
11418 struct bpf_prog *prog,
11419 u32 *target_size)
11420 {
11421 struct bpf_insn *insn = insn_buf;
11422
11423 switch (si->off) {
11424 case offsetof(struct sk_reuseport_md, data):
11425 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11426 break;
11427
11428 case offsetof(struct sk_reuseport_md, len):
11429 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11430 break;
11431
11432 case offsetof(struct sk_reuseport_md, eth_protocol):
11433 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11434 break;
11435
11436 case offsetof(struct sk_reuseport_md, ip_protocol):
11437 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11438 break;
11439
11440 case offsetof(struct sk_reuseport_md, data_end):
11441 SK_REUSEPORT_LOAD_FIELD(data_end);
11442 break;
11443
11444 case offsetof(struct sk_reuseport_md, hash):
11445 SK_REUSEPORT_LOAD_FIELD(hash);
11446 break;
11447
11448 case offsetof(struct sk_reuseport_md, bind_inany):
11449 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11450 break;
11451
11452 case offsetof(struct sk_reuseport_md, sk):
11453 SK_REUSEPORT_LOAD_FIELD(sk);
11454 break;
11455
11456 case offsetof(struct sk_reuseport_md, migrating_sk):
11457 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11458 break;
11459 }
11460
11461 return insn - insn_buf;
11462 }
11463
11464 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11465 .get_func_proto = sk_reuseport_func_proto,
11466 .is_valid_access = sk_reuseport_is_valid_access,
11467 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11468 };
11469
11470 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11471 };
11472
11473 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11474 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11475
BPF_CALL_3(bpf_sk_lookup_assign,struct bpf_sk_lookup_kern *,ctx,struct sock *,sk,u64,flags)11476 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11477 struct sock *, sk, u64, flags)
11478 {
11479 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11480 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11481 return -EINVAL;
11482 if (unlikely(sk && sk_is_refcounted(sk)))
11483 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11484 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11485 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11486 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11487 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11488
11489 /* Check if socket is suitable for packet L3/L4 protocol */
11490 if (sk && sk->sk_protocol != ctx->protocol)
11491 return -EPROTOTYPE;
11492 if (sk && sk->sk_family != ctx->family &&
11493 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11494 return -EAFNOSUPPORT;
11495
11496 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11497 return -EEXIST;
11498
11499 /* Select socket as lookup result */
11500 ctx->selected_sk = sk;
11501 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11502 return 0;
11503 }
11504
11505 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11506 .func = bpf_sk_lookup_assign,
11507 .gpl_only = false,
11508 .ret_type = RET_INTEGER,
11509 .arg1_type = ARG_PTR_TO_CTX,
11510 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11511 .arg3_type = ARG_ANYTHING,
11512 };
11513
11514 static const struct bpf_func_proto *
sk_lookup_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11515 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11516 {
11517 switch (func_id) {
11518 case BPF_FUNC_perf_event_output:
11519 return &bpf_event_output_data_proto;
11520 case BPF_FUNC_sk_assign:
11521 return &bpf_sk_lookup_assign_proto;
11522 case BPF_FUNC_sk_release:
11523 return &bpf_sk_release_proto;
11524 default:
11525 return bpf_sk_base_func_proto(func_id, prog);
11526 }
11527 }
11528
sk_lookup_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11529 static bool sk_lookup_is_valid_access(int off, int size,
11530 enum bpf_access_type type,
11531 const struct bpf_prog *prog,
11532 struct bpf_insn_access_aux *info)
11533 {
11534 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11535 return false;
11536 if (off % size != 0)
11537 return false;
11538 if (type != BPF_READ)
11539 return false;
11540
11541 switch (off) {
11542 case offsetof(struct bpf_sk_lookup, sk):
11543 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11544 return size == sizeof(__u64);
11545
11546 case bpf_ctx_range(struct bpf_sk_lookup, family):
11547 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11548 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11549 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11550 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11551 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11552 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11553 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11554 bpf_ctx_record_field_size(info, sizeof(__u32));
11555 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11556
11557 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11558 /* Allow 4-byte access to 2-byte field for backward compatibility */
11559 if (size == sizeof(__u32))
11560 return true;
11561 bpf_ctx_record_field_size(info, sizeof(__be16));
11562 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11563
11564 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11565 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11566 /* Allow access to zero padding for backward compatibility */
11567 bpf_ctx_record_field_size(info, sizeof(__u16));
11568 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11569
11570 default:
11571 return false;
11572 }
11573 }
11574
sk_lookup_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11575 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11576 const struct bpf_insn *si,
11577 struct bpf_insn *insn_buf,
11578 struct bpf_prog *prog,
11579 u32 *target_size)
11580 {
11581 struct bpf_insn *insn = insn_buf;
11582
11583 switch (si->off) {
11584 case offsetof(struct bpf_sk_lookup, sk):
11585 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11586 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11587 break;
11588
11589 case offsetof(struct bpf_sk_lookup, family):
11590 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11591 bpf_target_off(struct bpf_sk_lookup_kern,
11592 family, 2, target_size));
11593 break;
11594
11595 case offsetof(struct bpf_sk_lookup, protocol):
11596 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11597 bpf_target_off(struct bpf_sk_lookup_kern,
11598 protocol, 2, target_size));
11599 break;
11600
11601 case offsetof(struct bpf_sk_lookup, remote_ip4):
11602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11603 bpf_target_off(struct bpf_sk_lookup_kern,
11604 v4.saddr, 4, target_size));
11605 break;
11606
11607 case offsetof(struct bpf_sk_lookup, local_ip4):
11608 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11609 bpf_target_off(struct bpf_sk_lookup_kern,
11610 v4.daddr, 4, target_size));
11611 break;
11612
11613 case bpf_ctx_range_till(struct bpf_sk_lookup,
11614 remote_ip6[0], remote_ip6[3]): {
11615 #if IS_ENABLED(CONFIG_IPV6)
11616 int off = si->off;
11617
11618 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11619 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11620 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11621 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11622 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11624 #else
11625 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11626 #endif
11627 break;
11628 }
11629 case bpf_ctx_range_till(struct bpf_sk_lookup,
11630 local_ip6[0], local_ip6[3]): {
11631 #if IS_ENABLED(CONFIG_IPV6)
11632 int off = si->off;
11633
11634 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11635 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11636 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11637 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11638 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11639 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11640 #else
11641 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11642 #endif
11643 break;
11644 }
11645 case offsetof(struct bpf_sk_lookup, remote_port):
11646 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11647 bpf_target_off(struct bpf_sk_lookup_kern,
11648 sport, 2, target_size));
11649 break;
11650
11651 case offsetofend(struct bpf_sk_lookup, remote_port):
11652 *target_size = 2;
11653 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11654 break;
11655
11656 case offsetof(struct bpf_sk_lookup, local_port):
11657 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11658 bpf_target_off(struct bpf_sk_lookup_kern,
11659 dport, 2, target_size));
11660 break;
11661
11662 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11663 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11664 bpf_target_off(struct bpf_sk_lookup_kern,
11665 ingress_ifindex, 4, target_size));
11666 break;
11667 }
11668
11669 return insn - insn_buf;
11670 }
11671
11672 const struct bpf_prog_ops sk_lookup_prog_ops = {
11673 .test_run = bpf_prog_test_run_sk_lookup,
11674 };
11675
11676 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11677 .get_func_proto = sk_lookup_func_proto,
11678 .is_valid_access = sk_lookup_is_valid_access,
11679 .convert_ctx_access = sk_lookup_convert_ctx_access,
11680 };
11681
11682 #endif /* CONFIG_INET */
11683
DEFINE_BPF_DISPATCHER(xdp)11684 DEFINE_BPF_DISPATCHER(xdp)
11685
11686 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11687 {
11688 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11689 }
11690
BTF_ID_LIST_GLOBAL(btf_sock_ids,MAX_BTF_SOCK_TYPE)11691 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11692 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11693 BTF_SOCK_TYPE_xxx
11694 #undef BTF_SOCK_TYPE
11695
11696 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11697 {
11698 /* tcp6_sock type is not generated in dwarf and hence btf,
11699 * trigger an explicit type generation here.
11700 */
11701 BTF_TYPE_EMIT(struct tcp6_sock);
11702 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11703 sk->sk_family == AF_INET6)
11704 return (unsigned long)sk;
11705
11706 return (unsigned long)NULL;
11707 }
11708
11709 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11710 .func = bpf_skc_to_tcp6_sock,
11711 .gpl_only = false,
11712 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11713 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11714 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11715 };
11716
BPF_CALL_1(bpf_skc_to_tcp_sock,struct sock *,sk)11717 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11718 {
11719 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11720 return (unsigned long)sk;
11721
11722 return (unsigned long)NULL;
11723 }
11724
11725 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11726 .func = bpf_skc_to_tcp_sock,
11727 .gpl_only = false,
11728 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11729 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11730 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11731 };
11732
BPF_CALL_1(bpf_skc_to_tcp_timewait_sock,struct sock *,sk)11733 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11734 {
11735 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11736 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11737 */
11738 BTF_TYPE_EMIT(struct inet_timewait_sock);
11739 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11740
11741 #ifdef CONFIG_INET
11742 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11743 return (unsigned long)sk;
11744 #endif
11745
11746 #if IS_BUILTIN(CONFIG_IPV6)
11747 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11748 return (unsigned long)sk;
11749 #endif
11750
11751 return (unsigned long)NULL;
11752 }
11753
11754 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11755 .func = bpf_skc_to_tcp_timewait_sock,
11756 .gpl_only = false,
11757 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11758 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11759 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11760 };
11761
BPF_CALL_1(bpf_skc_to_tcp_request_sock,struct sock *,sk)11762 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11763 {
11764 #ifdef CONFIG_INET
11765 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11766 return (unsigned long)sk;
11767 #endif
11768
11769 #if IS_BUILTIN(CONFIG_IPV6)
11770 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11771 return (unsigned long)sk;
11772 #endif
11773
11774 return (unsigned long)NULL;
11775 }
11776
11777 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11778 .func = bpf_skc_to_tcp_request_sock,
11779 .gpl_only = false,
11780 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11781 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11782 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11783 };
11784
BPF_CALL_1(bpf_skc_to_udp6_sock,struct sock *,sk)11785 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11786 {
11787 /* udp6_sock type is not generated in dwarf and hence btf,
11788 * trigger an explicit type generation here.
11789 */
11790 BTF_TYPE_EMIT(struct udp6_sock);
11791 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11792 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11793 return (unsigned long)sk;
11794
11795 return (unsigned long)NULL;
11796 }
11797
11798 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11799 .func = bpf_skc_to_udp6_sock,
11800 .gpl_only = false,
11801 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11802 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11803 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11804 };
11805
BPF_CALL_1(bpf_skc_to_unix_sock,struct sock *,sk)11806 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11807 {
11808 /* unix_sock type is not generated in dwarf and hence btf,
11809 * trigger an explicit type generation here.
11810 */
11811 BTF_TYPE_EMIT(struct unix_sock);
11812 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11813 return (unsigned long)sk;
11814
11815 return (unsigned long)NULL;
11816 }
11817
11818 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11819 .func = bpf_skc_to_unix_sock,
11820 .gpl_only = false,
11821 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11822 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11823 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11824 };
11825
BPF_CALL_1(bpf_skc_to_mptcp_sock,struct sock *,sk)11826 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11827 {
11828 BTF_TYPE_EMIT(struct mptcp_sock);
11829 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11830 }
11831
11832 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11833 .func = bpf_skc_to_mptcp_sock,
11834 .gpl_only = false,
11835 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11836 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11837 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11838 };
11839
BPF_CALL_1(bpf_sock_from_file,struct file *,file)11840 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11841 {
11842 return (unsigned long)sock_from_file(file);
11843 }
11844
11845 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11846 BTF_ID(struct, socket)
11847 BTF_ID(struct, file)
11848
11849 const struct bpf_func_proto bpf_sock_from_file_proto = {
11850 .func = bpf_sock_from_file,
11851 .gpl_only = false,
11852 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11853 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11854 .arg1_type = ARG_PTR_TO_BTF_ID,
11855 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11856 };
11857
11858 static const struct bpf_func_proto *
bpf_sk_base_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11859 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11860 {
11861 const struct bpf_func_proto *func;
11862
11863 switch (func_id) {
11864 case BPF_FUNC_skc_to_tcp6_sock:
11865 func = &bpf_skc_to_tcp6_sock_proto;
11866 break;
11867 case BPF_FUNC_skc_to_tcp_sock:
11868 func = &bpf_skc_to_tcp_sock_proto;
11869 break;
11870 case BPF_FUNC_skc_to_tcp_timewait_sock:
11871 func = &bpf_skc_to_tcp_timewait_sock_proto;
11872 break;
11873 case BPF_FUNC_skc_to_tcp_request_sock:
11874 func = &bpf_skc_to_tcp_request_sock_proto;
11875 break;
11876 case BPF_FUNC_skc_to_udp6_sock:
11877 func = &bpf_skc_to_udp6_sock_proto;
11878 break;
11879 case BPF_FUNC_skc_to_unix_sock:
11880 func = &bpf_skc_to_unix_sock_proto;
11881 break;
11882 case BPF_FUNC_skc_to_mptcp_sock:
11883 func = &bpf_skc_to_mptcp_sock_proto;
11884 break;
11885 case BPF_FUNC_ktime_get_coarse_ns:
11886 return &bpf_ktime_get_coarse_ns_proto;
11887 default:
11888 return bpf_base_func_proto(func_id, prog);
11889 }
11890
11891 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
11892 return NULL;
11893
11894 return func;
11895 }
11896
11897 __bpf_kfunc_start_defs();
bpf_dynptr_from_skb(struct __sk_buff * s,u64 flags,struct bpf_dynptr * ptr__uninit)11898 __bpf_kfunc int bpf_dynptr_from_skb(struct __sk_buff *s, u64 flags,
11899 struct bpf_dynptr *ptr__uninit)
11900 {
11901 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11902 struct sk_buff *skb = (struct sk_buff *)s;
11903
11904 if (flags) {
11905 bpf_dynptr_set_null(ptr);
11906 return -EINVAL;
11907 }
11908
11909 bpf_dynptr_init(ptr, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11910
11911 return 0;
11912 }
11913
bpf_dynptr_from_xdp(struct xdp_md * x,u64 flags,struct bpf_dynptr * ptr__uninit)11914 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_md *x, u64 flags,
11915 struct bpf_dynptr *ptr__uninit)
11916 {
11917 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11918 struct xdp_buff *xdp = (struct xdp_buff *)x;
11919
11920 if (flags) {
11921 bpf_dynptr_set_null(ptr);
11922 return -EINVAL;
11923 }
11924
11925 bpf_dynptr_init(ptr, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11926
11927 return 0;
11928 }
11929
bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern * sa_kern,const u8 * sun_path,u32 sun_path__sz)11930 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
11931 const u8 *sun_path, u32 sun_path__sz)
11932 {
11933 struct sockaddr_un *un;
11934
11935 if (sa_kern->sk->sk_family != AF_UNIX)
11936 return -EINVAL;
11937
11938 /* We do not allow changing the address to unnamed or larger than the
11939 * maximum allowed address size for a unix sockaddr.
11940 */
11941 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX)
11942 return -EINVAL;
11943
11944 un = (struct sockaddr_un *)sa_kern->uaddr;
11945 memcpy(un->sun_path, sun_path, sun_path__sz);
11946 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz;
11947
11948 return 0;
11949 }
11950
bpf_sk_assign_tcp_reqsk(struct __sk_buff * s,struct sock * sk,struct bpf_tcp_req_attrs * attrs,int attrs__sz)11951 __bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct __sk_buff *s, struct sock *sk,
11952 struct bpf_tcp_req_attrs *attrs, int attrs__sz)
11953 {
11954 #if IS_ENABLED(CONFIG_SYN_COOKIES)
11955 struct sk_buff *skb = (struct sk_buff *)s;
11956 const struct request_sock_ops *ops;
11957 struct inet_request_sock *ireq;
11958 struct tcp_request_sock *treq;
11959 struct request_sock *req;
11960 struct net *net;
11961 __u16 min_mss;
11962 u32 tsoff = 0;
11963
11964 if (attrs__sz != sizeof(*attrs) ||
11965 attrs->reserved[0] || attrs->reserved[1] || attrs->reserved[2])
11966 return -EINVAL;
11967
11968 if (!skb_at_tc_ingress(skb))
11969 return -EINVAL;
11970
11971 net = dev_net(skb->dev);
11972 if (net != sock_net(sk))
11973 return -ENETUNREACH;
11974
11975 switch (skb->protocol) {
11976 case htons(ETH_P_IP):
11977 ops = &tcp_request_sock_ops;
11978 min_mss = 536;
11979 break;
11980 #if IS_BUILTIN(CONFIG_IPV6)
11981 case htons(ETH_P_IPV6):
11982 ops = &tcp6_request_sock_ops;
11983 min_mss = IPV6_MIN_MTU - 60;
11984 break;
11985 #endif
11986 default:
11987 return -EINVAL;
11988 }
11989
11990 if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN ||
11991 sk_is_mptcp(sk))
11992 return -EINVAL;
11993
11994 if (attrs->mss < min_mss)
11995 return -EINVAL;
11996
11997 if (attrs->wscale_ok) {
11998 if (!READ_ONCE(net->ipv4.sysctl_tcp_window_scaling))
11999 return -EINVAL;
12000
12001 if (attrs->snd_wscale > TCP_MAX_WSCALE ||
12002 attrs->rcv_wscale > TCP_MAX_WSCALE)
12003 return -EINVAL;
12004 }
12005
12006 if (attrs->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
12007 return -EINVAL;
12008
12009 if (attrs->tstamp_ok) {
12010 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
12011 return -EINVAL;
12012
12013 tsoff = attrs->rcv_tsecr - tcp_ns_to_ts(attrs->usec_ts_ok, tcp_clock_ns());
12014 }
12015
12016 req = inet_reqsk_alloc(ops, sk, false);
12017 if (!req)
12018 return -ENOMEM;
12019
12020 ireq = inet_rsk(req);
12021 treq = tcp_rsk(req);
12022
12023 req->rsk_listener = sk;
12024 req->syncookie = 1;
12025 req->mss = attrs->mss;
12026 req->ts_recent = attrs->rcv_tsval;
12027
12028 ireq->snd_wscale = attrs->snd_wscale;
12029 ireq->rcv_wscale = attrs->rcv_wscale;
12030 ireq->tstamp_ok = !!attrs->tstamp_ok;
12031 ireq->sack_ok = !!attrs->sack_ok;
12032 ireq->wscale_ok = !!attrs->wscale_ok;
12033 ireq->ecn_ok = !!attrs->ecn_ok;
12034
12035 treq->req_usec_ts = !!attrs->usec_ts_ok;
12036 treq->ts_off = tsoff;
12037
12038 skb_orphan(skb);
12039 skb->sk = req_to_sk(req);
12040 skb->destructor = sock_pfree;
12041
12042 return 0;
12043 #else
12044 return -EOPNOTSUPP;
12045 #endif
12046 }
12047
12048 __bpf_kfunc_end_defs();
12049
bpf_dynptr_from_skb_rdonly(struct __sk_buff * skb,u64 flags,struct bpf_dynptr * ptr__uninit)12050 int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
12051 struct bpf_dynptr *ptr__uninit)
12052 {
12053 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
12054 int err;
12055
12056 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
12057 if (err)
12058 return err;
12059
12060 bpf_dynptr_set_rdonly(ptr);
12061
12062 return 0;
12063 }
12064
12065 BTF_KFUNCS_START(bpf_kfunc_check_set_skb)
12066 BTF_ID_FLAGS(func, bpf_dynptr_from_skb, KF_TRUSTED_ARGS)
12067 BTF_KFUNCS_END(bpf_kfunc_check_set_skb)
12068
12069 BTF_KFUNCS_START(bpf_kfunc_check_set_xdp)
12070 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
12071 BTF_KFUNCS_END(bpf_kfunc_check_set_xdp)
12072
12073 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_addr)
12074 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
12075 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_addr)
12076
12077 BTF_KFUNCS_START(bpf_kfunc_check_set_tcp_reqsk)
12078 BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk, KF_TRUSTED_ARGS)
12079 BTF_KFUNCS_END(bpf_kfunc_check_set_tcp_reqsk)
12080
12081 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
12082 .owner = THIS_MODULE,
12083 .set = &bpf_kfunc_check_set_skb,
12084 };
12085
12086 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
12087 .owner = THIS_MODULE,
12088 .set = &bpf_kfunc_check_set_xdp,
12089 };
12090
12091 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
12092 .owner = THIS_MODULE,
12093 .set = &bpf_kfunc_check_set_sock_addr,
12094 };
12095
12096 static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = {
12097 .owner = THIS_MODULE,
12098 .set = &bpf_kfunc_check_set_tcp_reqsk,
12099 };
12100
bpf_kfunc_init(void)12101 static int __init bpf_kfunc_init(void)
12102 {
12103 int ret;
12104
12105 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
12106 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
12107 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
12108 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
12109 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
12110 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
12111 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
12112 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
12113 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
12114 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
12115 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_kfunc_set_skb);
12116 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
12117 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
12118 &bpf_kfunc_set_sock_addr);
12119 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk);
12120 }
12121 late_initcall(bpf_kfunc_init);
12122
12123 __bpf_kfunc_start_defs();
12124
12125 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
12126 *
12127 * The function expects a non-NULL pointer to a socket, and invokes the
12128 * protocol specific socket destroy handlers.
12129 *
12130 * The helper can only be called from BPF contexts that have acquired the socket
12131 * locks.
12132 *
12133 * Parameters:
12134 * @sock: Pointer to socket to be destroyed
12135 *
12136 * Return:
12137 * On error, may return EPROTONOSUPPORT, EINVAL.
12138 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
12139 * 0 otherwise
12140 */
bpf_sock_destroy(struct sock_common * sock)12141 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
12142 {
12143 struct sock *sk = (struct sock *)sock;
12144
12145 /* The locking semantics that allow for synchronous execution of the
12146 * destroy handlers are only supported for TCP and UDP.
12147 * Supporting protocols will need to acquire sock lock in the BPF context
12148 * prior to invoking this kfunc.
12149 */
12150 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
12151 sk->sk_protocol != IPPROTO_UDP))
12152 return -EOPNOTSUPP;
12153
12154 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
12155 }
12156
12157 __bpf_kfunc_end_defs();
12158
12159 BTF_KFUNCS_START(bpf_sk_iter_kfunc_ids)
BTF_ID_FLAGS(func,bpf_sock_destroy,KF_TRUSTED_ARGS)12160 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
12161 BTF_KFUNCS_END(bpf_sk_iter_kfunc_ids)
12162
12163 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
12164 {
12165 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
12166 prog->expected_attach_type != BPF_TRACE_ITER)
12167 return -EACCES;
12168 return 0;
12169 }
12170
12171 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
12172 .owner = THIS_MODULE,
12173 .set = &bpf_sk_iter_kfunc_ids,
12174 .filter = tracing_iter_filter,
12175 };
12176
init_subsystem(void)12177 static int init_subsystem(void)
12178 {
12179 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
12180 }
12181 late_initcall(init_subsystem);
12182