1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20#include <linux/module.h>
21#include <linux/types.h>
22#include <linux/mm.h>
23#include <linux/fcntl.h>
24#include <linux/socket.h>
25#include <linux/sock_diag.h>
26#include <linux/in.h>
27#include <linux/inet.h>
28#include <linux/netdevice.h>
29#include <linux/if_packet.h>
30#include <linux/if_arp.h>
31#include <linux/gfp.h>
32#include <net/inet_common.h>
33#include <net/ip.h>
34#include <net/protocol.h>
35#include <net/netlink.h>
36#include <linux/skbuff.h>
37#include <linux/skmsg.h>
38#include <net/sock.h>
39#include <net/flow_dissector.h>
40#include <linux/errno.h>
41#include <linux/timer.h>
42#include <linux/uaccess.h>
43#include <asm/unaligned.h>
44#include <asm/cmpxchg.h>
45#include <linux/filter.h>
46#include <linux/ratelimit.h>
47#include <linux/seccomp.h>
48#include <linux/if_vlan.h>
49#include <linux/bpf.h>
50#include <net/sch_generic.h>
51#include <net/cls_cgroup.h>
52#include <net/dst_metadata.h>
53#include <net/dst.h>
54#include <net/sock_reuseport.h>
55#include <net/busy_poll.h>
56#include <net/tcp.h>
57#include <net/xfrm.h>
58#include <net/udp.h>
59#include <linux/bpf_trace.h>
60#include <net/xdp_sock.h>
61#include <linux/inetdevice.h>
62#include <net/inet_hashtables.h>
63#include <net/inet6_hashtables.h>
64#include <net/ip_fib.h>
65#include <net/flow.h>
66#include <net/arp.h>
67#include <net/ipv6.h>
68#include <net/net_namespace.h>
69#include <linux/seg6_local.h>
70#include <net/seg6.h>
71#include <net/seg6_local.h>
72#include <net/lwtunnel.h>
73#include <net/ipv6_stubs.h>
74#include <net/bpf_sk_storage.h>
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
90{
91 int err;
92 struct sk_filter *filter;
93
94
95
96
97
98
99 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
100 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
101 return -ENOMEM;
102 }
103 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
104 if (err)
105 return err;
106
107 err = security_sock_rcv_skb(sk, skb);
108 if (err)
109 return err;
110
111 rcu_read_lock();
112 filter = rcu_dereference(sk->sk_filter);
113 if (filter) {
114 struct sock *save_sk = skb->sk;
115 unsigned int pkt_len;
116
117 skb->sk = sk;
118 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
119 skb->sk = save_sk;
120 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
121 }
122 rcu_read_unlock();
123
124 return err;
125}
126EXPORT_SYMBOL(sk_filter_trim_cap);
127
128BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
129{
130 return skb_get_poff(skb);
131}
132
133BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
134{
135 struct nlattr *nla;
136
137 if (skb_is_nonlinear(skb))
138 return 0;
139
140 if (skb->len < sizeof(struct nlattr))
141 return 0;
142
143 if (a > skb->len - sizeof(struct nlattr))
144 return 0;
145
146 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
147 if (nla)
148 return (void *) nla - (void *) skb->data;
149
150 return 0;
151}
152
153BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
154{
155 struct nlattr *nla;
156
157 if (skb_is_nonlinear(skb))
158 return 0;
159
160 if (skb->len < sizeof(struct nlattr))
161 return 0;
162
163 if (a > skb->len - sizeof(struct nlattr))
164 return 0;
165
166 nla = (struct nlattr *) &skb->data[a];
167 if (nla->nla_len > skb->len - a)
168 return 0;
169
170 nla = nla_find_nested(nla, x);
171 if (nla)
172 return (void *) nla - (void *) skb->data;
173
174 return 0;
175}
176
177BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
178 data, int, headlen, int, offset)
179{
180 u8 tmp, *ptr;
181 const int len = sizeof(tmp);
182
183 if (offset >= 0) {
184 if (headlen - offset >= len)
185 return *(u8 *)(data + offset);
186 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
187 return tmp;
188 } else {
189 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
190 if (likely(ptr))
191 return *(u8 *)ptr;
192 }
193
194 return -EFAULT;
195}
196
197BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
198 int, offset)
199{
200 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
201 offset);
202}
203
204BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
205 data, int, headlen, int, offset)
206{
207 u16 tmp, *ptr;
208 const int len = sizeof(tmp);
209
210 if (offset >= 0) {
211 if (headlen - offset >= len)
212 return get_unaligned_be16(data + offset);
213 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
214 return be16_to_cpu(tmp);
215 } else {
216 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
217 if (likely(ptr))
218 return get_unaligned_be16(ptr);
219 }
220
221 return -EFAULT;
222}
223
224BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
225 int, offset)
226{
227 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
228 offset);
229}
230
231BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
232 data, int, headlen, int, offset)
233{
234 u32 tmp, *ptr;
235 const int len = sizeof(tmp);
236
237 if (likely(offset >= 0)) {
238 if (headlen - offset >= len)
239 return get_unaligned_be32(data + offset);
240 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
241 return be32_to_cpu(tmp);
242 } else {
243 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
244 if (likely(ptr))
245 return get_unaligned_be32(ptr);
246 }
247
248 return -EFAULT;
249}
250
251BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
252 int, offset)
253{
254 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
255 offset);
256}
257
258BPF_CALL_0(bpf_get_raw_cpu_id)
259{
260 return raw_smp_processor_id();
261}
262
263static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
264 .func = bpf_get_raw_cpu_id,
265 .gpl_only = false,
266 .ret_type = RET_INTEGER,
267};
268
269static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
270 struct bpf_insn *insn_buf)
271{
272 struct bpf_insn *insn = insn_buf;
273
274 switch (skb_field) {
275 case SKF_AD_MARK:
276 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
277
278 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
279 offsetof(struct sk_buff, mark));
280 break;
281
282 case SKF_AD_PKTTYPE:
283 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
284 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
285#ifdef __BIG_ENDIAN_BITFIELD
286 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
287#endif
288 break;
289
290 case SKF_AD_QUEUE:
291 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
292
293 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
294 offsetof(struct sk_buff, queue_mapping));
295 break;
296
297 case SKF_AD_VLAN_TAG:
298 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
299
300
301 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
302 offsetof(struct sk_buff, vlan_tci));
303 break;
304 case SKF_AD_VLAN_TAG_PRESENT:
305 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
306 if (PKT_VLAN_PRESENT_BIT)
307 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
308 if (PKT_VLAN_PRESENT_BIT < 7)
309 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
310 break;
311 }
312
313 return insn - insn_buf;
314}
315
316static bool convert_bpf_extensions(struct sock_filter *fp,
317 struct bpf_insn **insnp)
318{
319 struct bpf_insn *insn = *insnp;
320 u32 cnt;
321
322 switch (fp->k) {
323 case SKF_AD_OFF + SKF_AD_PROTOCOL:
324 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
325
326
327 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
328 offsetof(struct sk_buff, protocol));
329
330 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
331 break;
332
333 case SKF_AD_OFF + SKF_AD_PKTTYPE:
334 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
335 insn += cnt - 1;
336 break;
337
338 case SKF_AD_OFF + SKF_AD_IFINDEX:
339 case SKF_AD_OFF + SKF_AD_HATYPE:
340 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
341 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
342
343 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
344 BPF_REG_TMP, BPF_REG_CTX,
345 offsetof(struct sk_buff, dev));
346
347 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
348 *insn++ = BPF_EXIT_INSN();
349 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
350 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
351 offsetof(struct net_device, ifindex));
352 else
353 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
354 offsetof(struct net_device, type));
355 break;
356
357 case SKF_AD_OFF + SKF_AD_MARK:
358 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
359 insn += cnt - 1;
360 break;
361
362 case SKF_AD_OFF + SKF_AD_RXHASH:
363 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
364
365 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
366 offsetof(struct sk_buff, hash));
367 break;
368
369 case SKF_AD_OFF + SKF_AD_QUEUE:
370 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
371 insn += cnt - 1;
372 break;
373
374 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
375 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
376 BPF_REG_A, BPF_REG_CTX, insn);
377 insn += cnt - 1;
378 break;
379
380 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
381 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
382 BPF_REG_A, BPF_REG_CTX, insn);
383 insn += cnt - 1;
384 break;
385
386 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
387 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
388
389
390 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
391 offsetof(struct sk_buff, vlan_proto));
392
393 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
394 break;
395
396 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
397 case SKF_AD_OFF + SKF_AD_NLATTR:
398 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
399 case SKF_AD_OFF + SKF_AD_CPU:
400 case SKF_AD_OFF + SKF_AD_RANDOM:
401
402 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
403
404 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
405
406 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
407
408 switch (fp->k) {
409 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
410 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
411 break;
412 case SKF_AD_OFF + SKF_AD_NLATTR:
413 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
414 break;
415 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
416 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
417 break;
418 case SKF_AD_OFF + SKF_AD_CPU:
419 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
420 break;
421 case SKF_AD_OFF + SKF_AD_RANDOM:
422 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
423 bpf_user_rnd_init_once();
424 break;
425 }
426 break;
427
428 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
429
430 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
431 break;
432
433 default:
434
435
436
437
438 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
439 return false;
440 }
441
442 *insnp = insn;
443 return true;
444}
445
446static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
447{
448 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
449 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
450 bool endian = BPF_SIZE(fp->code) == BPF_H ||
451 BPF_SIZE(fp->code) == BPF_W;
452 bool indirect = BPF_MODE(fp->code) == BPF_IND;
453 const int ip_align = NET_IP_ALIGN;
454 struct bpf_insn *insn = *insnp;
455 int offset = fp->k;
456
457 if (!indirect &&
458 ((unaligned_ok && offset >= 0) ||
459 (!unaligned_ok && offset >= 0 &&
460 offset + ip_align >= 0 &&
461 offset + ip_align % size == 0))) {
462 bool ldx_off_ok = offset <= S16_MAX;
463
464 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
465 if (offset)
466 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
467 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
468 size, 2 + endian + (!ldx_off_ok * 2));
469 if (ldx_off_ok) {
470 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
471 BPF_REG_D, offset);
472 } else {
473 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
474 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
475 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
476 BPF_REG_TMP, 0);
477 }
478 if (endian)
479 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
480 *insn++ = BPF_JMP_A(8);
481 }
482
483 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
484 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
485 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
486 if (!indirect) {
487 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
488 } else {
489 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
490 if (fp->k)
491 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
492 }
493
494 switch (BPF_SIZE(fp->code)) {
495 case BPF_B:
496 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
497 break;
498 case BPF_H:
499 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
500 break;
501 case BPF_W:
502 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
503 break;
504 default:
505 return false;
506 }
507
508 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
509 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
510 *insn = BPF_EXIT_INSN();
511
512 *insnp = insn;
513 return true;
514}
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535static int bpf_convert_filter(struct sock_filter *prog, int len,
536 struct bpf_prog *new_prog, int *new_len,
537 bool *seen_ld_abs)
538{
539 int new_flen = 0, pass = 0, target, i, stack_off;
540 struct bpf_insn *new_insn, *first_insn = NULL;
541 struct sock_filter *fp;
542 int *addrs = NULL;
543 u8 bpf_src;
544
545 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
546 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
547
548 if (len <= 0 || len > BPF_MAXINSNS)
549 return -EINVAL;
550
551 if (new_prog) {
552 first_insn = new_prog->insnsi;
553 addrs = kcalloc(len, sizeof(*addrs),
554 GFP_KERNEL | __GFP_NOWARN);
555 if (!addrs)
556 return -ENOMEM;
557 }
558
559do_pass:
560 new_insn = first_insn;
561 fp = prog;
562
563
564 if (new_prog) {
565
566
567
568 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
569 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
570
571
572
573
574
575 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
576 if (*seen_ld_abs) {
577
578
579
580
581
582 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
583 BPF_REG_D, BPF_REG_CTX,
584 offsetof(struct sk_buff, data));
585 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
586 offsetof(struct sk_buff, len));
587 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
588 offsetof(struct sk_buff, data_len));
589 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
590 }
591 } else {
592 new_insn += 3;
593 }
594
595 for (i = 0; i < len; fp++, i++) {
596 struct bpf_insn tmp_insns[32] = { };
597 struct bpf_insn *insn = tmp_insns;
598
599 if (addrs)
600 addrs[i] = new_insn - first_insn;
601
602 switch (fp->code) {
603
604 case BPF_ALU | BPF_ADD | BPF_X:
605 case BPF_ALU | BPF_ADD | BPF_K:
606 case BPF_ALU | BPF_SUB | BPF_X:
607 case BPF_ALU | BPF_SUB | BPF_K:
608 case BPF_ALU | BPF_AND | BPF_X:
609 case BPF_ALU | BPF_AND | BPF_K:
610 case BPF_ALU | BPF_OR | BPF_X:
611 case BPF_ALU | BPF_OR | BPF_K:
612 case BPF_ALU | BPF_LSH | BPF_X:
613 case BPF_ALU | BPF_LSH | BPF_K:
614 case BPF_ALU | BPF_RSH | BPF_X:
615 case BPF_ALU | BPF_RSH | BPF_K:
616 case BPF_ALU | BPF_XOR | BPF_X:
617 case BPF_ALU | BPF_XOR | BPF_K:
618 case BPF_ALU | BPF_MUL | BPF_X:
619 case BPF_ALU | BPF_MUL | BPF_K:
620 case BPF_ALU | BPF_DIV | BPF_X:
621 case BPF_ALU | BPF_DIV | BPF_K:
622 case BPF_ALU | BPF_MOD | BPF_X:
623 case BPF_ALU | BPF_MOD | BPF_K:
624 case BPF_ALU | BPF_NEG:
625 case BPF_LD | BPF_ABS | BPF_W:
626 case BPF_LD | BPF_ABS | BPF_H:
627 case BPF_LD | BPF_ABS | BPF_B:
628 case BPF_LD | BPF_IND | BPF_W:
629 case BPF_LD | BPF_IND | BPF_H:
630 case BPF_LD | BPF_IND | BPF_B:
631
632
633
634
635 if (BPF_CLASS(fp->code) == BPF_LD &&
636 BPF_MODE(fp->code) == BPF_ABS &&
637 convert_bpf_extensions(fp, &insn))
638 break;
639 if (BPF_CLASS(fp->code) == BPF_LD &&
640 convert_bpf_ld_abs(fp, &insn)) {
641 *seen_ld_abs = true;
642 break;
643 }
644
645 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
646 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
647 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
648
649
650
651 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
652 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
653 *insn++ = BPF_EXIT_INSN();
654 }
655
656 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
657 break;
658
659
660
661
662
663
664
665#define BPF_EMIT_JMP \
666 do { \
667 const s32 off_min = S16_MIN, off_max = S16_MAX; \
668 s32 off; \
669 \
670 if (target >= len || target < 0) \
671 goto err; \
672 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
673 \
674 off -= insn - tmp_insns; \
675 \
676 if (off < off_min || off > off_max) \
677 goto err; \
678 insn->off = off; \
679 } while (0)
680
681 case BPF_JMP | BPF_JA:
682 target = i + fp->k + 1;
683 insn->code = fp->code;
684 BPF_EMIT_JMP;
685 break;
686
687 case BPF_JMP | BPF_JEQ | BPF_K:
688 case BPF_JMP | BPF_JEQ | BPF_X:
689 case BPF_JMP | BPF_JSET | BPF_K:
690 case BPF_JMP | BPF_JSET | BPF_X:
691 case BPF_JMP | BPF_JGT | BPF_K:
692 case BPF_JMP | BPF_JGT | BPF_X:
693 case BPF_JMP | BPF_JGE | BPF_K:
694 case BPF_JMP | BPF_JGE | BPF_X:
695 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
696
697
698
699
700 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
701
702 insn->dst_reg = BPF_REG_A;
703 insn->src_reg = BPF_REG_TMP;
704 bpf_src = BPF_X;
705 } else {
706 insn->dst_reg = BPF_REG_A;
707 insn->imm = fp->k;
708 bpf_src = BPF_SRC(fp->code);
709 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
710 }
711
712
713 if (fp->jf == 0) {
714 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
715 target = i + fp->jt + 1;
716 BPF_EMIT_JMP;
717 break;
718 }
719
720
721 if (fp->jt == 0) {
722 switch (BPF_OP(fp->code)) {
723 case BPF_JEQ:
724 insn->code = BPF_JMP | BPF_JNE | bpf_src;
725 break;
726 case BPF_JGT:
727 insn->code = BPF_JMP | BPF_JLE | bpf_src;
728 break;
729 case BPF_JGE:
730 insn->code = BPF_JMP | BPF_JLT | bpf_src;
731 break;
732 default:
733 goto jmp_rest;
734 }
735
736 target = i + fp->jf + 1;
737 BPF_EMIT_JMP;
738 break;
739 }
740jmp_rest:
741
742 target = i + fp->jt + 1;
743 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
744 BPF_EMIT_JMP;
745 insn++;
746
747 insn->code = BPF_JMP | BPF_JA;
748 target = i + fp->jf + 1;
749 BPF_EMIT_JMP;
750 break;
751
752
753 case BPF_LDX | BPF_MSH | BPF_B: {
754 struct sock_filter tmp = {
755 .code = BPF_LD | BPF_ABS | BPF_B,
756 .k = fp->k,
757 };
758
759 *seen_ld_abs = true;
760
761
762 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
763
764 convert_bpf_ld_abs(&tmp, &insn);
765 insn++;
766
767 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
768
769 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
770
771 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
772
773 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
774
775 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
776 break;
777 }
778
779
780
781 case BPF_RET | BPF_A:
782 case BPF_RET | BPF_K:
783 if (BPF_RVAL(fp->code) == BPF_K)
784 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
785 0, fp->k);
786 *insn = BPF_EXIT_INSN();
787 break;
788
789
790 case BPF_ST:
791 case BPF_STX:
792 stack_off = fp->k * 4 + 4;
793 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
794 BPF_ST ? BPF_REG_A : BPF_REG_X,
795 -stack_off);
796
797
798
799
800 if (new_prog && new_prog->aux->stack_depth < stack_off)
801 new_prog->aux->stack_depth = stack_off;
802 break;
803
804
805 case BPF_LD | BPF_MEM:
806 case BPF_LDX | BPF_MEM:
807 stack_off = fp->k * 4 + 4;
808 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
809 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
810 -stack_off);
811 break;
812
813
814 case BPF_LD | BPF_IMM:
815 case BPF_LDX | BPF_IMM:
816 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
817 BPF_REG_A : BPF_REG_X, fp->k);
818 break;
819
820
821 case BPF_MISC | BPF_TAX:
822 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
823 break;
824
825
826 case BPF_MISC | BPF_TXA:
827 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
828 break;
829
830
831 case BPF_LD | BPF_W | BPF_LEN:
832 case BPF_LDX | BPF_W | BPF_LEN:
833 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
834 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
835 offsetof(struct sk_buff, len));
836 break;
837
838
839 case BPF_LDX | BPF_ABS | BPF_W:
840
841 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
842 break;
843
844
845 default:
846 goto err;
847 }
848
849 insn++;
850 if (new_prog)
851 memcpy(new_insn, tmp_insns,
852 sizeof(*insn) * (insn - tmp_insns));
853 new_insn += insn - tmp_insns;
854 }
855
856 if (!new_prog) {
857
858 *new_len = new_insn - first_insn;
859 if (*seen_ld_abs)
860 *new_len += 4;
861 return 0;
862 }
863
864 pass++;
865 if (new_flen != new_insn - first_insn) {
866 new_flen = new_insn - first_insn;
867 if (pass > 2)
868 goto err;
869 goto do_pass;
870 }
871
872 kfree(addrs);
873 BUG_ON(*new_len != new_flen);
874 return 0;
875err:
876 kfree(addrs);
877 return -EINVAL;
878}
879
880
881
882
883
884
885
886
887static int check_load_and_stores(const struct sock_filter *filter, int flen)
888{
889 u16 *masks, memvalid = 0;
890 int pc, ret = 0;
891
892 BUILD_BUG_ON(BPF_MEMWORDS > 16);
893
894 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
895 if (!masks)
896 return -ENOMEM;
897
898 memset(masks, 0xff, flen * sizeof(*masks));
899
900 for (pc = 0; pc < flen; pc++) {
901 memvalid &= masks[pc];
902
903 switch (filter[pc].code) {
904 case BPF_ST:
905 case BPF_STX:
906 memvalid |= (1 << filter[pc].k);
907 break;
908 case BPF_LD | BPF_MEM:
909 case BPF_LDX | BPF_MEM:
910 if (!(memvalid & (1 << filter[pc].k))) {
911 ret = -EINVAL;
912 goto error;
913 }
914 break;
915 case BPF_JMP | BPF_JA:
916
917 masks[pc + 1 + filter[pc].k] &= memvalid;
918 memvalid = ~0;
919 break;
920 case BPF_JMP | BPF_JEQ | BPF_K:
921 case BPF_JMP | BPF_JEQ | BPF_X:
922 case BPF_JMP | BPF_JGE | BPF_K:
923 case BPF_JMP | BPF_JGE | BPF_X:
924 case BPF_JMP | BPF_JGT | BPF_K:
925 case BPF_JMP | BPF_JGT | BPF_X:
926 case BPF_JMP | BPF_JSET | BPF_K:
927 case BPF_JMP | BPF_JSET | BPF_X:
928
929 masks[pc + 1 + filter[pc].jt] &= memvalid;
930 masks[pc + 1 + filter[pc].jf] &= memvalid;
931 memvalid = ~0;
932 break;
933 }
934 }
935error:
936 kfree(masks);
937 return ret;
938}
939
940static bool chk_code_allowed(u16 code_to_probe)
941{
942 static const bool codes[] = {
943
944 [BPF_ALU | BPF_ADD | BPF_K] = true,
945 [BPF_ALU | BPF_ADD | BPF_X] = true,
946 [BPF_ALU | BPF_SUB | BPF_K] = true,
947 [BPF_ALU | BPF_SUB | BPF_X] = true,
948 [BPF_ALU | BPF_MUL | BPF_K] = true,
949 [BPF_ALU | BPF_MUL | BPF_X] = true,
950 [BPF_ALU | BPF_DIV | BPF_K] = true,
951 [BPF_ALU | BPF_DIV | BPF_X] = true,
952 [BPF_ALU | BPF_MOD | BPF_K] = true,
953 [BPF_ALU | BPF_MOD | BPF_X] = true,
954 [BPF_ALU | BPF_AND | BPF_K] = true,
955 [BPF_ALU | BPF_AND | BPF_X] = true,
956 [BPF_ALU | BPF_OR | BPF_K] = true,
957 [BPF_ALU | BPF_OR | BPF_X] = true,
958 [BPF_ALU | BPF_XOR | BPF_K] = true,
959 [BPF_ALU | BPF_XOR | BPF_X] = true,
960 [BPF_ALU | BPF_LSH | BPF_K] = true,
961 [BPF_ALU | BPF_LSH | BPF_X] = true,
962 [BPF_ALU | BPF_RSH | BPF_K] = true,
963 [BPF_ALU | BPF_RSH | BPF_X] = true,
964 [BPF_ALU | BPF_NEG] = true,
965
966 [BPF_LD | BPF_W | BPF_ABS] = true,
967 [BPF_LD | BPF_H | BPF_ABS] = true,
968 [BPF_LD | BPF_B | BPF_ABS] = true,
969 [BPF_LD | BPF_W | BPF_LEN] = true,
970 [BPF_LD | BPF_W | BPF_IND] = true,
971 [BPF_LD | BPF_H | BPF_IND] = true,
972 [BPF_LD | BPF_B | BPF_IND] = true,
973 [BPF_LD | BPF_IMM] = true,
974 [BPF_LD | BPF_MEM] = true,
975 [BPF_LDX | BPF_W | BPF_LEN] = true,
976 [BPF_LDX | BPF_B | BPF_MSH] = true,
977 [BPF_LDX | BPF_IMM] = true,
978 [BPF_LDX | BPF_MEM] = true,
979
980 [BPF_ST] = true,
981 [BPF_STX] = true,
982
983 [BPF_MISC | BPF_TAX] = true,
984 [BPF_MISC | BPF_TXA] = true,
985
986 [BPF_RET | BPF_K] = true,
987 [BPF_RET | BPF_A] = true,
988
989 [BPF_JMP | BPF_JA] = true,
990 [BPF_JMP | BPF_JEQ | BPF_K] = true,
991 [BPF_JMP | BPF_JEQ | BPF_X] = true,
992 [BPF_JMP | BPF_JGE | BPF_K] = true,
993 [BPF_JMP | BPF_JGE | BPF_X] = true,
994 [BPF_JMP | BPF_JGT | BPF_K] = true,
995 [BPF_JMP | BPF_JGT | BPF_X] = true,
996 [BPF_JMP | BPF_JSET | BPF_K] = true,
997 [BPF_JMP | BPF_JSET | BPF_X] = true,
998 };
999
1000 if (code_to_probe >= ARRAY_SIZE(codes))
1001 return false;
1002
1003 return codes[code_to_probe];
1004}
1005
1006static bool bpf_check_basics_ok(const struct sock_filter *filter,
1007 unsigned int flen)
1008{
1009 if (filter == NULL)
1010 return false;
1011 if (flen == 0 || flen > BPF_MAXINSNS)
1012 return false;
1013
1014 return true;
1015}
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031static int bpf_check_classic(const struct sock_filter *filter,
1032 unsigned int flen)
1033{
1034 bool anc_found;
1035 int pc;
1036
1037
1038 for (pc = 0; pc < flen; pc++) {
1039 const struct sock_filter *ftest = &filter[pc];
1040
1041
1042 if (!chk_code_allowed(ftest->code))
1043 return -EINVAL;
1044
1045
1046 switch (ftest->code) {
1047 case BPF_ALU | BPF_DIV | BPF_K:
1048 case BPF_ALU | BPF_MOD | BPF_K:
1049
1050 if (ftest->k == 0)
1051 return -EINVAL;
1052 break;
1053 case BPF_ALU | BPF_LSH | BPF_K:
1054 case BPF_ALU | BPF_RSH | BPF_K:
1055 if (ftest->k >= 32)
1056 return -EINVAL;
1057 break;
1058 case BPF_LD | BPF_MEM:
1059 case BPF_LDX | BPF_MEM:
1060 case BPF_ST:
1061 case BPF_STX:
1062
1063 if (ftest->k >= BPF_MEMWORDS)
1064 return -EINVAL;
1065 break;
1066 case BPF_JMP | BPF_JA:
1067
1068
1069
1070
1071 if (ftest->k >= (unsigned int)(flen - pc - 1))
1072 return -EINVAL;
1073 break;
1074 case BPF_JMP | BPF_JEQ | BPF_K:
1075 case BPF_JMP | BPF_JEQ | BPF_X:
1076 case BPF_JMP | BPF_JGE | BPF_K:
1077 case BPF_JMP | BPF_JGE | BPF_X:
1078 case BPF_JMP | BPF_JGT | BPF_K:
1079 case BPF_JMP | BPF_JGT | BPF_X:
1080 case BPF_JMP | BPF_JSET | BPF_K:
1081 case BPF_JMP | BPF_JSET | BPF_X:
1082
1083 if (pc + ftest->jt + 1 >= flen ||
1084 pc + ftest->jf + 1 >= flen)
1085 return -EINVAL;
1086 break;
1087 case BPF_LD | BPF_W | BPF_ABS:
1088 case BPF_LD | BPF_H | BPF_ABS:
1089 case BPF_LD | BPF_B | BPF_ABS:
1090 anc_found = false;
1091 if (bpf_anc_helper(ftest) & BPF_ANC)
1092 anc_found = true;
1093
1094 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1095 return -EINVAL;
1096 }
1097 }
1098
1099
1100 switch (filter[flen - 1].code) {
1101 case BPF_RET | BPF_K:
1102 case BPF_RET | BPF_A:
1103 return check_load_and_stores(filter, flen);
1104 }
1105
1106 return -EINVAL;
1107}
1108
1109static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1110 const struct sock_fprog *fprog)
1111{
1112 unsigned int fsize = bpf_classic_proglen(fprog);
1113 struct sock_fprog_kern *fkprog;
1114
1115 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1116 if (!fp->orig_prog)
1117 return -ENOMEM;
1118
1119 fkprog = fp->orig_prog;
1120 fkprog->len = fprog->len;
1121
1122 fkprog->filter = kmemdup(fp->insns, fsize,
1123 GFP_KERNEL | __GFP_NOWARN);
1124 if (!fkprog->filter) {
1125 kfree(fp->orig_prog);
1126 return -ENOMEM;
1127 }
1128
1129 return 0;
1130}
1131
1132static void bpf_release_orig_filter(struct bpf_prog *fp)
1133{
1134 struct sock_fprog_kern *fprog = fp->orig_prog;
1135
1136 if (fprog) {
1137 kfree(fprog->filter);
1138 kfree(fprog);
1139 }
1140}
1141
1142static void __bpf_prog_release(struct bpf_prog *prog)
1143{
1144 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1145 bpf_prog_put(prog);
1146 } else {
1147 bpf_release_orig_filter(prog);
1148 bpf_prog_free(prog);
1149 }
1150}
1151
1152static void __sk_filter_release(struct sk_filter *fp)
1153{
1154 __bpf_prog_release(fp->prog);
1155 kfree(fp);
1156}
1157
1158
1159
1160
1161
1162static void sk_filter_release_rcu(struct rcu_head *rcu)
1163{
1164 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1165
1166 __sk_filter_release(fp);
1167}
1168
1169
1170
1171
1172
1173
1174
1175static void sk_filter_release(struct sk_filter *fp)
1176{
1177 if (refcount_dec_and_test(&fp->refcnt))
1178 call_rcu(&fp->rcu, sk_filter_release_rcu);
1179}
1180
1181void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1182{
1183 u32 filter_size = bpf_prog_size(fp->prog->len);
1184
1185 atomic_sub(filter_size, &sk->sk_omem_alloc);
1186 sk_filter_release(fp);
1187}
1188
1189
1190
1191
1192static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1193{
1194 u32 filter_size = bpf_prog_size(fp->prog->len);
1195
1196
1197 if (filter_size <= sysctl_optmem_max &&
1198 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1199 atomic_add(filter_size, &sk->sk_omem_alloc);
1200 return true;
1201 }
1202 return false;
1203}
1204
1205bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1206{
1207 if (!refcount_inc_not_zero(&fp->refcnt))
1208 return false;
1209
1210 if (!__sk_filter_charge(sk, fp)) {
1211 sk_filter_release(fp);
1212 return false;
1213 }
1214 return true;
1215}
1216
1217static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1218{
1219 struct sock_filter *old_prog;
1220 struct bpf_prog *old_fp;
1221 int err, new_len, old_len = fp->len;
1222 bool seen_ld_abs = false;
1223
1224
1225
1226
1227
1228
1229 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1230 sizeof(struct bpf_insn));
1231
1232
1233
1234
1235
1236 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1237 GFP_KERNEL | __GFP_NOWARN);
1238 if (!old_prog) {
1239 err = -ENOMEM;
1240 goto out_err;
1241 }
1242
1243
1244 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1245 &seen_ld_abs);
1246 if (err)
1247 goto out_err_free;
1248
1249
1250 old_fp = fp;
1251 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1252 if (!fp) {
1253
1254
1255
1256 fp = old_fp;
1257 err = -ENOMEM;
1258 goto out_err_free;
1259 }
1260
1261 fp->len = new_len;
1262
1263
1264 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1265 &seen_ld_abs);
1266 if (err)
1267
1268
1269
1270
1271
1272 goto out_err_free;
1273
1274 fp = bpf_prog_select_runtime(fp, &err);
1275 if (err)
1276 goto out_err_free;
1277
1278 kfree(old_prog);
1279 return fp;
1280
1281out_err_free:
1282 kfree(old_prog);
1283out_err:
1284 __bpf_prog_release(fp);
1285 return ERR_PTR(err);
1286}
1287
1288static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1289 bpf_aux_classic_check_t trans)
1290{
1291 int err;
1292
1293 fp->bpf_func = NULL;
1294 fp->jited = 0;
1295
1296 err = bpf_check_classic(fp->insns, fp->len);
1297 if (err) {
1298 __bpf_prog_release(fp);
1299 return ERR_PTR(err);
1300 }
1301
1302
1303
1304
1305 if (trans) {
1306 err = trans(fp->insns, fp->len);
1307 if (err) {
1308 __bpf_prog_release(fp);
1309 return ERR_PTR(err);
1310 }
1311 }
1312
1313
1314
1315
1316 bpf_jit_compile(fp);
1317
1318
1319
1320
1321 if (!fp->jited)
1322 fp = bpf_migrate_filter(fp);
1323
1324 return fp;
1325}
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1338{
1339 unsigned int fsize = bpf_classic_proglen(fprog);
1340 struct bpf_prog *fp;
1341
1342
1343 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1344 return -EINVAL;
1345
1346 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1347 if (!fp)
1348 return -ENOMEM;
1349
1350 memcpy(fp->insns, fprog->filter, fsize);
1351
1352 fp->len = fprog->len;
1353
1354
1355
1356
1357 fp->orig_prog = NULL;
1358
1359
1360
1361
1362 fp = bpf_prepare_filter(fp, NULL);
1363 if (IS_ERR(fp))
1364 return PTR_ERR(fp);
1365
1366 *pfp = fp;
1367 return 0;
1368}
1369EXPORT_SYMBOL_GPL(bpf_prog_create);
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1383 bpf_aux_classic_check_t trans, bool save_orig)
1384{
1385 unsigned int fsize = bpf_classic_proglen(fprog);
1386 struct bpf_prog *fp;
1387 int err;
1388
1389
1390 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1391 return -EINVAL;
1392
1393 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1394 if (!fp)
1395 return -ENOMEM;
1396
1397 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1398 __bpf_prog_free(fp);
1399 return -EFAULT;
1400 }
1401
1402 fp->len = fprog->len;
1403 fp->orig_prog = NULL;
1404
1405 if (save_orig) {
1406 err = bpf_prog_store_orig_filter(fp, fprog);
1407 if (err) {
1408 __bpf_prog_free(fp);
1409 return -ENOMEM;
1410 }
1411 }
1412
1413
1414
1415
1416 fp = bpf_prepare_filter(fp, trans);
1417 if (IS_ERR(fp))
1418 return PTR_ERR(fp);
1419
1420 *pfp = fp;
1421 return 0;
1422}
1423EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1424
1425void bpf_prog_destroy(struct bpf_prog *fp)
1426{
1427 __bpf_prog_release(fp);
1428}
1429EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1430
1431static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1432{
1433 struct sk_filter *fp, *old_fp;
1434
1435 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1436 if (!fp)
1437 return -ENOMEM;
1438
1439 fp->prog = prog;
1440
1441 if (!__sk_filter_charge(sk, fp)) {
1442 kfree(fp);
1443 return -ENOMEM;
1444 }
1445 refcount_set(&fp->refcnt, 1);
1446
1447 old_fp = rcu_dereference_protected(sk->sk_filter,
1448 lockdep_sock_is_held(sk));
1449 rcu_assign_pointer(sk->sk_filter, fp);
1450
1451 if (old_fp)
1452 sk_filter_uncharge(sk, old_fp);
1453
1454 return 0;
1455}
1456
1457static
1458struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1459{
1460 unsigned int fsize = bpf_classic_proglen(fprog);
1461 struct bpf_prog *prog;
1462 int err;
1463
1464 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1465 return ERR_PTR(-EPERM);
1466
1467
1468 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1469 return ERR_PTR(-EINVAL);
1470
1471 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1472 if (!prog)
1473 return ERR_PTR(-ENOMEM);
1474
1475 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1476 __bpf_prog_free(prog);
1477 return ERR_PTR(-EFAULT);
1478 }
1479
1480 prog->len = fprog->len;
1481
1482 err = bpf_prog_store_orig_filter(prog, fprog);
1483 if (err) {
1484 __bpf_prog_free(prog);
1485 return ERR_PTR(-ENOMEM);
1486 }
1487
1488
1489
1490
1491 return bpf_prepare_filter(prog, NULL);
1492}
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1505{
1506 struct bpf_prog *prog = __get_filter(fprog, sk);
1507 int err;
1508
1509 if (IS_ERR(prog))
1510 return PTR_ERR(prog);
1511
1512 err = __sk_attach_prog(prog, sk);
1513 if (err < 0) {
1514 __bpf_prog_release(prog);
1515 return err;
1516 }
1517
1518 return 0;
1519}
1520EXPORT_SYMBOL_GPL(sk_attach_filter);
1521
1522int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1523{
1524 struct bpf_prog *prog = __get_filter(fprog, sk);
1525 int err;
1526
1527 if (IS_ERR(prog))
1528 return PTR_ERR(prog);
1529
1530 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1531 err = -ENOMEM;
1532 else
1533 err = reuseport_attach_prog(sk, prog);
1534
1535 if (err)
1536 __bpf_prog_release(prog);
1537
1538 return err;
1539}
1540
1541static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1542{
1543 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1544 return ERR_PTR(-EPERM);
1545
1546 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1547}
1548
1549int sk_attach_bpf(u32 ufd, struct sock *sk)
1550{
1551 struct bpf_prog *prog = __get_bpf(ufd, sk);
1552 int err;
1553
1554 if (IS_ERR(prog))
1555 return PTR_ERR(prog);
1556
1557 err = __sk_attach_prog(prog, sk);
1558 if (err < 0) {
1559 bpf_prog_put(prog);
1560 return err;
1561 }
1562
1563 return 0;
1564}
1565
1566int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1567{
1568 struct bpf_prog *prog;
1569 int err;
1570
1571 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1572 return -EPERM;
1573
1574 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1575 if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1576 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1577 if (IS_ERR(prog))
1578 return PTR_ERR(prog);
1579
1580 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1581
1582
1583
1584
1585
1586 if ((sk->sk_type != SOCK_STREAM &&
1587 sk->sk_type != SOCK_DGRAM) ||
1588 (sk->sk_protocol != IPPROTO_UDP &&
1589 sk->sk_protocol != IPPROTO_TCP) ||
1590 (sk->sk_family != AF_INET &&
1591 sk->sk_family != AF_INET6)) {
1592 err = -ENOTSUPP;
1593 goto err_prog_put;
1594 }
1595 } else {
1596
1597 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1598 err = -ENOMEM;
1599 goto err_prog_put;
1600 }
1601 }
1602
1603 err = reuseport_attach_prog(sk, prog);
1604err_prog_put:
1605 if (err)
1606 bpf_prog_put(prog);
1607
1608 return err;
1609}
1610
1611void sk_reuseport_prog_free(struct bpf_prog *prog)
1612{
1613 if (!prog)
1614 return;
1615
1616 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1617 bpf_prog_put(prog);
1618 else
1619 bpf_prog_destroy(prog);
1620}
1621
1622struct bpf_scratchpad {
1623 union {
1624 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1625 u8 buff[MAX_BPF_STACK];
1626 };
1627};
1628
1629static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1630
1631static inline int __bpf_try_make_writable(struct sk_buff *skb,
1632 unsigned int write_len)
1633{
1634 return skb_ensure_writable(skb, write_len);
1635}
1636
1637static inline int bpf_try_make_writable(struct sk_buff *skb,
1638 unsigned int write_len)
1639{
1640 int err = __bpf_try_make_writable(skb, write_len);
1641
1642 bpf_compute_data_pointers(skb);
1643 return err;
1644}
1645
1646static int bpf_try_make_head_writable(struct sk_buff *skb)
1647{
1648 return bpf_try_make_writable(skb, skb_headlen(skb));
1649}
1650
1651static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1652{
1653 if (skb_at_tc_ingress(skb))
1654 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1655}
1656
1657static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1658{
1659 if (skb_at_tc_ingress(skb))
1660 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1661}
1662
1663BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1664 const void *, from, u32, len, u64, flags)
1665{
1666 void *ptr;
1667
1668 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1669 return -EINVAL;
1670 if (unlikely(offset > 0xffff))
1671 return -EFAULT;
1672 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1673 return -EFAULT;
1674
1675 ptr = skb->data + offset;
1676 if (flags & BPF_F_RECOMPUTE_CSUM)
1677 __skb_postpull_rcsum(skb, ptr, len, offset);
1678
1679 memcpy(ptr, from, len);
1680
1681 if (flags & BPF_F_RECOMPUTE_CSUM)
1682 __skb_postpush_rcsum(skb, ptr, len, offset);
1683 if (flags & BPF_F_INVALIDATE_HASH)
1684 skb_clear_hash(skb);
1685
1686 return 0;
1687}
1688
1689static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1690 .func = bpf_skb_store_bytes,
1691 .gpl_only = false,
1692 .ret_type = RET_INTEGER,
1693 .arg1_type = ARG_PTR_TO_CTX,
1694 .arg2_type = ARG_ANYTHING,
1695 .arg3_type = ARG_PTR_TO_MEM,
1696 .arg4_type = ARG_CONST_SIZE,
1697 .arg5_type = ARG_ANYTHING,
1698};
1699
1700BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1701 void *, to, u32, len)
1702{
1703 void *ptr;
1704
1705 if (unlikely(offset > 0xffff))
1706 goto err_clear;
1707
1708 ptr = skb_header_pointer(skb, offset, len, to);
1709 if (unlikely(!ptr))
1710 goto err_clear;
1711 if (ptr != to)
1712 memcpy(to, ptr, len);
1713
1714 return 0;
1715err_clear:
1716 memset(to, 0, len);
1717 return -EFAULT;
1718}
1719
1720static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1721 .func = bpf_skb_load_bytes,
1722 .gpl_only = false,
1723 .ret_type = RET_INTEGER,
1724 .arg1_type = ARG_PTR_TO_CTX,
1725 .arg2_type = ARG_ANYTHING,
1726 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1727 .arg4_type = ARG_CONST_SIZE,
1728};
1729
1730BPF_CALL_4(bpf_flow_dissector_load_bytes,
1731 const struct bpf_flow_dissector *, ctx, u32, offset,
1732 void *, to, u32, len)
1733{
1734 void *ptr;
1735
1736 if (unlikely(offset > 0xffff))
1737 goto err_clear;
1738
1739 if (unlikely(!ctx->skb))
1740 goto err_clear;
1741
1742 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1743 if (unlikely(!ptr))
1744 goto err_clear;
1745 if (ptr != to)
1746 memcpy(to, ptr, len);
1747
1748 return 0;
1749err_clear:
1750 memset(to, 0, len);
1751 return -EFAULT;
1752}
1753
1754static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1755 .func = bpf_flow_dissector_load_bytes,
1756 .gpl_only = false,
1757 .ret_type = RET_INTEGER,
1758 .arg1_type = ARG_PTR_TO_CTX,
1759 .arg2_type = ARG_ANYTHING,
1760 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1761 .arg4_type = ARG_CONST_SIZE,
1762};
1763
1764BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1765 u32, offset, void *, to, u32, len, u32, start_header)
1766{
1767 u8 *end = skb_tail_pointer(skb);
1768 u8 *net = skb_network_header(skb);
1769 u8 *mac = skb_mac_header(skb);
1770 u8 *ptr;
1771
1772 if (unlikely(offset > 0xffff || len > (end - mac)))
1773 goto err_clear;
1774
1775 switch (start_header) {
1776 case BPF_HDR_START_MAC:
1777 ptr = mac + offset;
1778 break;
1779 case BPF_HDR_START_NET:
1780 ptr = net + offset;
1781 break;
1782 default:
1783 goto err_clear;
1784 }
1785
1786 if (likely(ptr >= mac && ptr + len <= end)) {
1787 memcpy(to, ptr, len);
1788 return 0;
1789 }
1790
1791err_clear:
1792 memset(to, 0, len);
1793 return -EFAULT;
1794}
1795
1796static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1797 .func = bpf_skb_load_bytes_relative,
1798 .gpl_only = false,
1799 .ret_type = RET_INTEGER,
1800 .arg1_type = ARG_PTR_TO_CTX,
1801 .arg2_type = ARG_ANYTHING,
1802 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1803 .arg4_type = ARG_CONST_SIZE,
1804 .arg5_type = ARG_ANYTHING,
1805};
1806
1807BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1808{
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1819}
1820
1821static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1822 .func = bpf_skb_pull_data,
1823 .gpl_only = false,
1824 .ret_type = RET_INTEGER,
1825 .arg1_type = ARG_PTR_TO_CTX,
1826 .arg2_type = ARG_ANYTHING,
1827};
1828
1829BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1830{
1831 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1832}
1833
1834static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1835 .func = bpf_sk_fullsock,
1836 .gpl_only = false,
1837 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1838 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1839};
1840
1841static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1842 unsigned int write_len)
1843{
1844 int err = __bpf_try_make_writable(skb, write_len);
1845
1846 bpf_compute_data_end_sk_skb(skb);
1847 return err;
1848}
1849
1850BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1851{
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1862}
1863
1864static const struct bpf_func_proto sk_skb_pull_data_proto = {
1865 .func = sk_skb_pull_data,
1866 .gpl_only = false,
1867 .ret_type = RET_INTEGER,
1868 .arg1_type = ARG_PTR_TO_CTX,
1869 .arg2_type = ARG_ANYTHING,
1870};
1871
1872BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1873 u64, from, u64, to, u64, flags)
1874{
1875 __sum16 *ptr;
1876
1877 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1878 return -EINVAL;
1879 if (unlikely(offset > 0xffff || offset & 1))
1880 return -EFAULT;
1881 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1882 return -EFAULT;
1883
1884 ptr = (__sum16 *)(skb->data + offset);
1885 switch (flags & BPF_F_HDR_FIELD_MASK) {
1886 case 0:
1887 if (unlikely(from != 0))
1888 return -EINVAL;
1889
1890 csum_replace_by_diff(ptr, to);
1891 break;
1892 case 2:
1893 csum_replace2(ptr, from, to);
1894 break;
1895 case 4:
1896 csum_replace4(ptr, from, to);
1897 break;
1898 default:
1899 return -EINVAL;
1900 }
1901
1902 return 0;
1903}
1904
1905static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1906 .func = bpf_l3_csum_replace,
1907 .gpl_only = false,
1908 .ret_type = RET_INTEGER,
1909 .arg1_type = ARG_PTR_TO_CTX,
1910 .arg2_type = ARG_ANYTHING,
1911 .arg3_type = ARG_ANYTHING,
1912 .arg4_type = ARG_ANYTHING,
1913 .arg5_type = ARG_ANYTHING,
1914};
1915
1916BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1917 u64, from, u64, to, u64, flags)
1918{
1919 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1920 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1921 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1922 __sum16 *ptr;
1923
1924 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1925 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1926 return -EINVAL;
1927 if (unlikely(offset > 0xffff || offset & 1))
1928 return -EFAULT;
1929 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1930 return -EFAULT;
1931
1932 ptr = (__sum16 *)(skb->data + offset);
1933 if (is_mmzero && !do_mforce && !*ptr)
1934 return 0;
1935
1936 switch (flags & BPF_F_HDR_FIELD_MASK) {
1937 case 0:
1938 if (unlikely(from != 0))
1939 return -EINVAL;
1940
1941 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1942 break;
1943 case 2:
1944 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1945 break;
1946 case 4:
1947 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1948 break;
1949 default:
1950 return -EINVAL;
1951 }
1952
1953 if (is_mmzero && !*ptr)
1954 *ptr = CSUM_MANGLED_0;
1955 return 0;
1956}
1957
1958static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1959 .func = bpf_l4_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
1969BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1970 __be32 *, to, u32, to_size, __wsum, seed)
1971{
1972 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1973 u32 diff_size = from_size + to_size;
1974 int i, j = 0;
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1985 diff_size > sizeof(sp->diff)))
1986 return -EINVAL;
1987
1988 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1989 sp->diff[j] = ~from[i];
1990 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1991 sp->diff[j] = to[i];
1992
1993 return csum_partial(sp->diff, diff_size, seed);
1994}
1995
1996static const struct bpf_func_proto bpf_csum_diff_proto = {
1997 .func = bpf_csum_diff,
1998 .gpl_only = false,
1999 .pkt_access = true,
2000 .ret_type = RET_INTEGER,
2001 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2002 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2003 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2004 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2005 .arg5_type = ARG_ANYTHING,
2006};
2007
2008BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2009{
2010
2011
2012
2013
2014 if (skb->ip_summed == CHECKSUM_COMPLETE)
2015 return (skb->csum = csum_add(skb->csum, csum));
2016
2017 return -ENOTSUPP;
2018}
2019
2020static const struct bpf_func_proto bpf_csum_update_proto = {
2021 .func = bpf_csum_update,
2022 .gpl_only = false,
2023 .ret_type = RET_INTEGER,
2024 .arg1_type = ARG_PTR_TO_CTX,
2025 .arg2_type = ARG_ANYTHING,
2026};
2027
2028static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2029{
2030 return dev_forward_skb(dev, skb);
2031}
2032
2033static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2034 struct sk_buff *skb)
2035{
2036 int ret = ____dev_forward_skb(dev, skb);
2037
2038 if (likely(!ret)) {
2039 skb->dev = dev;
2040 ret = netif_rx(skb);
2041 }
2042
2043 return ret;
2044}
2045
2046static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2047{
2048 int ret;
2049
2050 if (dev_xmit_recursion()) {
2051 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2052 kfree_skb(skb);
2053 return -ENETDOWN;
2054 }
2055
2056 skb->dev = dev;
2057
2058 dev_xmit_recursion_inc();
2059 ret = dev_queue_xmit(skb);
2060 dev_xmit_recursion_dec();
2061
2062 return ret;
2063}
2064
2065static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2066 u32 flags)
2067{
2068 unsigned int mlen = skb_network_offset(skb);
2069
2070 if (mlen) {
2071 __skb_pull(skb, mlen);
2072
2073
2074
2075
2076
2077
2078 if (!skb_at_tc_ingress(skb))
2079 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2080 }
2081 skb_pop_mac_header(skb);
2082 skb_reset_mac_len(skb);
2083 return flags & BPF_F_INGRESS ?
2084 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2085}
2086
2087static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2088 u32 flags)
2089{
2090
2091 if (unlikely(skb->mac_header >= skb->network_header)) {
2092 kfree_skb(skb);
2093 return -ERANGE;
2094 }
2095
2096 bpf_push_mac_rcsum(skb);
2097 return flags & BPF_F_INGRESS ?
2098 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2099}
2100
2101static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2102 u32 flags)
2103{
2104 if (dev_is_mac_header_xmit(dev))
2105 return __bpf_redirect_common(skb, dev, flags);
2106 else
2107 return __bpf_redirect_no_mac(skb, dev, flags);
2108}
2109
2110BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2111{
2112 struct net_device *dev;
2113 struct sk_buff *clone;
2114 int ret;
2115
2116 if (unlikely(flags & ~(BPF_F_INGRESS)))
2117 return -EINVAL;
2118
2119 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2120 if (unlikely(!dev))
2121 return -EINVAL;
2122
2123 clone = skb_clone(skb, GFP_ATOMIC);
2124 if (unlikely(!clone))
2125 return -ENOMEM;
2126
2127
2128
2129
2130
2131
2132 ret = bpf_try_make_head_writable(skb);
2133 if (unlikely(ret)) {
2134 kfree_skb(clone);
2135 return -ENOMEM;
2136 }
2137
2138 return __bpf_redirect(clone, dev, flags);
2139}
2140
2141static const struct bpf_func_proto bpf_clone_redirect_proto = {
2142 .func = bpf_clone_redirect,
2143 .gpl_only = false,
2144 .ret_type = RET_INTEGER,
2145 .arg1_type = ARG_PTR_TO_CTX,
2146 .arg2_type = ARG_ANYTHING,
2147 .arg3_type = ARG_ANYTHING,
2148};
2149
2150DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2151EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2152
2153BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2154{
2155 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2156
2157 if (unlikely(flags & ~(BPF_F_INGRESS)))
2158 return TC_ACT_SHOT;
2159
2160 ri->ifindex = ifindex;
2161 ri->flags = flags;
2162
2163 return TC_ACT_REDIRECT;
2164}
2165
2166int skb_do_redirect(struct sk_buff *skb)
2167{
2168 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2169 struct net_device *dev;
2170
2171 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2172 ri->ifindex = 0;
2173 if (unlikely(!dev)) {
2174 kfree_skb(skb);
2175 return -EINVAL;
2176 }
2177
2178 return __bpf_redirect(skb, dev, ri->flags);
2179}
2180
2181static const struct bpf_func_proto bpf_redirect_proto = {
2182 .func = bpf_redirect,
2183 .gpl_only = false,
2184 .ret_type = RET_INTEGER,
2185 .arg1_type = ARG_ANYTHING,
2186 .arg2_type = ARG_ANYTHING,
2187};
2188
2189BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2190{
2191 msg->apply_bytes = bytes;
2192 return 0;
2193}
2194
2195static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2196 .func = bpf_msg_apply_bytes,
2197 .gpl_only = false,
2198 .ret_type = RET_INTEGER,
2199 .arg1_type = ARG_PTR_TO_CTX,
2200 .arg2_type = ARG_ANYTHING,
2201};
2202
2203BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2204{
2205 msg->cork_bytes = bytes;
2206 return 0;
2207}
2208
2209static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2210 .func = bpf_msg_cork_bytes,
2211 .gpl_only = false,
2212 .ret_type = RET_INTEGER,
2213 .arg1_type = ARG_PTR_TO_CTX,
2214 .arg2_type = ARG_ANYTHING,
2215};
2216
2217BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2218 u32, end, u64, flags)
2219{
2220 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2221 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2222 struct scatterlist *sge;
2223 u8 *raw, *to, *from;
2224 struct page *page;
2225
2226 if (unlikely(flags || end <= start))
2227 return -EINVAL;
2228
2229
2230 i = msg->sg.start;
2231 do {
2232 len = sk_msg_elem(msg, i)->length;
2233 if (start < offset + len)
2234 break;
2235 offset += len;
2236 sk_msg_iter_var_next(i);
2237 } while (i != msg->sg.end);
2238
2239 if (unlikely(start >= offset + len))
2240 return -EINVAL;
2241
2242 first_sge = i;
2243
2244
2245
2246 bytes_sg_total = start - offset + bytes;
2247 if (!msg->sg.copy[i] && bytes_sg_total <= len)
2248 goto out;
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260 do {
2261 copy += sk_msg_elem(msg, i)->length;
2262 sk_msg_iter_var_next(i);
2263 if (bytes_sg_total <= copy)
2264 break;
2265 } while (i != msg->sg.end);
2266 last_sge = i;
2267
2268 if (unlikely(bytes_sg_total > copy))
2269 return -EINVAL;
2270
2271 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2272 get_order(copy));
2273 if (unlikely(!page))
2274 return -ENOMEM;
2275
2276 raw = page_address(page);
2277 i = first_sge;
2278 do {
2279 sge = sk_msg_elem(msg, i);
2280 from = sg_virt(sge);
2281 len = sge->length;
2282 to = raw + poffset;
2283
2284 memcpy(to, from, len);
2285 poffset += len;
2286 sge->length = 0;
2287 put_page(sg_page(sge));
2288
2289 sk_msg_iter_var_next(i);
2290 } while (i != last_sge);
2291
2292 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2293
2294
2295
2296
2297
2298 WARN_ON_ONCE(last_sge == first_sge);
2299 shift = last_sge > first_sge ?
2300 last_sge - first_sge - 1 :
2301 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2302 if (!shift)
2303 goto out;
2304
2305 i = first_sge;
2306 sk_msg_iter_var_next(i);
2307 do {
2308 u32 move_from;
2309
2310 if (i + shift >= MAX_MSG_FRAGS)
2311 move_from = i + shift - MAX_MSG_FRAGS;
2312 else
2313 move_from = i + shift;
2314 if (move_from == msg->sg.end)
2315 break;
2316
2317 msg->sg.data[i] = msg->sg.data[move_from];
2318 msg->sg.data[move_from].length = 0;
2319 msg->sg.data[move_from].page_link = 0;
2320 msg->sg.data[move_from].offset = 0;
2321 sk_msg_iter_var_next(i);
2322 } while (1);
2323
2324 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2325 msg->sg.end - shift + MAX_MSG_FRAGS :
2326 msg->sg.end - shift;
2327out:
2328 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2329 msg->data_end = msg->data + bytes;
2330 return 0;
2331}
2332
2333static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2334 .func = bpf_msg_pull_data,
2335 .gpl_only = false,
2336 .ret_type = RET_INTEGER,
2337 .arg1_type = ARG_PTR_TO_CTX,
2338 .arg2_type = ARG_ANYTHING,
2339 .arg3_type = ARG_ANYTHING,
2340 .arg4_type = ARG_ANYTHING,
2341};
2342
2343BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2344 u32, len, u64, flags)
2345{
2346 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2347 u32 new, i = 0, l, space, copy = 0, offset = 0;
2348 u8 *raw, *to, *from;
2349 struct page *page;
2350
2351 if (unlikely(flags))
2352 return -EINVAL;
2353
2354
2355 i = msg->sg.start;
2356 do {
2357 l = sk_msg_elem(msg, i)->length;
2358
2359 if (start < offset + l)
2360 break;
2361 offset += l;
2362 sk_msg_iter_var_next(i);
2363 } while (i != msg->sg.end);
2364
2365 if (start >= offset + l)
2366 return -EINVAL;
2367
2368 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2369
2370
2371
2372
2373
2374
2375
2376
2377 if (!space || (space == 1 && start != offset))
2378 copy = msg->sg.data[i].length;
2379
2380 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2381 get_order(copy + len));
2382 if (unlikely(!page))
2383 return -ENOMEM;
2384
2385 if (copy) {
2386 int front, back;
2387
2388 raw = page_address(page);
2389
2390 psge = sk_msg_elem(msg, i);
2391 front = start - offset;
2392 back = psge->length - front;
2393 from = sg_virt(psge);
2394
2395 if (front)
2396 memcpy(raw, from, front);
2397
2398 if (back) {
2399 from += front;
2400 to = raw + front + len;
2401
2402 memcpy(to, from, back);
2403 }
2404
2405 put_page(sg_page(psge));
2406 } else if (start - offset) {
2407 psge = sk_msg_elem(msg, i);
2408 rsge = sk_msg_elem_cpy(msg, i);
2409
2410 psge->length = start - offset;
2411 rsge.length -= psge->length;
2412 rsge.offset += start;
2413
2414 sk_msg_iter_var_next(i);
2415 sg_unmark_end(psge);
2416 sk_msg_iter_next(msg, end);
2417 }
2418
2419
2420 new = i;
2421
2422
2423 if (!copy) {
2424 sge = sk_msg_elem_cpy(msg, i);
2425
2426 sk_msg_iter_var_next(i);
2427 sg_unmark_end(&sge);
2428 sk_msg_iter_next(msg, end);
2429
2430 nsge = sk_msg_elem_cpy(msg, i);
2431 if (rsge.length) {
2432 sk_msg_iter_var_next(i);
2433 nnsge = sk_msg_elem_cpy(msg, i);
2434 }
2435
2436 while (i != msg->sg.end) {
2437 msg->sg.data[i] = sge;
2438 sge = nsge;
2439 sk_msg_iter_var_next(i);
2440 if (rsge.length) {
2441 nsge = nnsge;
2442 nnsge = sk_msg_elem_cpy(msg, i);
2443 } else {
2444 nsge = sk_msg_elem_cpy(msg, i);
2445 }
2446 }
2447 }
2448
2449
2450 sk_mem_charge(msg->sk, len);
2451 msg->sg.size += len;
2452 msg->sg.copy[new] = false;
2453 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2454 if (rsge.length) {
2455 get_page(sg_page(&rsge));
2456 sk_msg_iter_var_next(new);
2457 msg->sg.data[new] = rsge;
2458 }
2459
2460 sk_msg_compute_data_pointers(msg);
2461 return 0;
2462}
2463
2464static const struct bpf_func_proto bpf_msg_push_data_proto = {
2465 .func = bpf_msg_push_data,
2466 .gpl_only = false,
2467 .ret_type = RET_INTEGER,
2468 .arg1_type = ARG_PTR_TO_CTX,
2469 .arg2_type = ARG_ANYTHING,
2470 .arg3_type = ARG_ANYTHING,
2471 .arg4_type = ARG_ANYTHING,
2472};
2473
2474static void sk_msg_shift_left(struct sk_msg *msg, int i)
2475{
2476 int prev;
2477
2478 do {
2479 prev = i;
2480 sk_msg_iter_var_next(i);
2481 msg->sg.data[prev] = msg->sg.data[i];
2482 } while (i != msg->sg.end);
2483
2484 sk_msg_iter_prev(msg, end);
2485}
2486
2487static void sk_msg_shift_right(struct sk_msg *msg, int i)
2488{
2489 struct scatterlist tmp, sge;
2490
2491 sk_msg_iter_next(msg, end);
2492 sge = sk_msg_elem_cpy(msg, i);
2493 sk_msg_iter_var_next(i);
2494 tmp = sk_msg_elem_cpy(msg, i);
2495
2496 while (i != msg->sg.end) {
2497 msg->sg.data[i] = sge;
2498 sk_msg_iter_var_next(i);
2499 sge = tmp;
2500 tmp = sk_msg_elem_cpy(msg, i);
2501 }
2502}
2503
2504BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2505 u32, len, u64, flags)
2506{
2507 u32 i = 0, l, space, offset = 0;
2508 u64 last = start + len;
2509 int pop;
2510
2511 if (unlikely(flags))
2512 return -EINVAL;
2513
2514
2515 i = msg->sg.start;
2516 do {
2517 l = sk_msg_elem(msg, i)->length;
2518
2519 if (start < offset + l)
2520 break;
2521 offset += l;
2522 sk_msg_iter_var_next(i);
2523 } while (i != msg->sg.end);
2524
2525
2526 if (start >= offset + l || last >= msg->sg.size)
2527 return -EINVAL;
2528
2529 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2530
2531 pop = len;
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553 if (start != offset) {
2554 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2555 int a = start;
2556 int b = sge->length - pop - a;
2557
2558 sk_msg_iter_var_next(i);
2559
2560 if (pop < sge->length - a) {
2561 if (space) {
2562 sge->length = a;
2563 sk_msg_shift_right(msg, i);
2564 nsge = sk_msg_elem(msg, i);
2565 get_page(sg_page(sge));
2566 sg_set_page(nsge,
2567 sg_page(sge),
2568 b, sge->offset + pop + a);
2569 } else {
2570 struct page *page, *orig;
2571 u8 *to, *from;
2572
2573 page = alloc_pages(__GFP_NOWARN |
2574 __GFP_COMP | GFP_ATOMIC,
2575 get_order(a + b));
2576 if (unlikely(!page))
2577 return -ENOMEM;
2578
2579 sge->length = a;
2580 orig = sg_page(sge);
2581 from = sg_virt(sge);
2582 to = page_address(page);
2583 memcpy(to, from, a);
2584 memcpy(to + a, from + a + pop, b);
2585 sg_set_page(sge, page, a + b, 0);
2586 put_page(orig);
2587 }
2588 pop = 0;
2589 } else if (pop >= sge->length - a) {
2590 sge->length = a;
2591 pop -= (sge->length - a);
2592 }
2593 }
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612 while (pop) {
2613 struct scatterlist *sge = sk_msg_elem(msg, i);
2614
2615 if (pop < sge->length) {
2616 sge->length -= pop;
2617 sge->offset += pop;
2618 pop = 0;
2619 } else {
2620 pop -= sge->length;
2621 sk_msg_shift_left(msg, i);
2622 }
2623 sk_msg_iter_var_next(i);
2624 }
2625
2626 sk_mem_uncharge(msg->sk, len - pop);
2627 msg->sg.size -= (len - pop);
2628 sk_msg_compute_data_pointers(msg);
2629 return 0;
2630}
2631
2632static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2633 .func = bpf_msg_pop_data,
2634 .gpl_only = false,
2635 .ret_type = RET_INTEGER,
2636 .arg1_type = ARG_PTR_TO_CTX,
2637 .arg2_type = ARG_ANYTHING,
2638 .arg3_type = ARG_ANYTHING,
2639 .arg4_type = ARG_ANYTHING,
2640};
2641
2642BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2643{
2644 return task_get_classid(skb);
2645}
2646
2647static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2648 .func = bpf_get_cgroup_classid,
2649 .gpl_only = false,
2650 .ret_type = RET_INTEGER,
2651 .arg1_type = ARG_PTR_TO_CTX,
2652};
2653
2654BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2655{
2656 return dst_tclassid(skb);
2657}
2658
2659static const struct bpf_func_proto bpf_get_route_realm_proto = {
2660 .func = bpf_get_route_realm,
2661 .gpl_only = false,
2662 .ret_type = RET_INTEGER,
2663 .arg1_type = ARG_PTR_TO_CTX,
2664};
2665
2666BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2667{
2668
2669
2670
2671
2672
2673 return skb_get_hash(skb);
2674}
2675
2676static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2677 .func = bpf_get_hash_recalc,
2678 .gpl_only = false,
2679 .ret_type = RET_INTEGER,
2680 .arg1_type = ARG_PTR_TO_CTX,
2681};
2682
2683BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2684{
2685
2686
2687
2688 skb_clear_hash(skb);
2689 return 0;
2690}
2691
2692static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2693 .func = bpf_set_hash_invalid,
2694 .gpl_only = false,
2695 .ret_type = RET_INTEGER,
2696 .arg1_type = ARG_PTR_TO_CTX,
2697};
2698
2699BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2700{
2701
2702
2703
2704
2705 __skb_set_sw_hash(skb, hash, true);
2706 return 0;
2707}
2708
2709static const struct bpf_func_proto bpf_set_hash_proto = {
2710 .func = bpf_set_hash,
2711 .gpl_only = false,
2712 .ret_type = RET_INTEGER,
2713 .arg1_type = ARG_PTR_TO_CTX,
2714 .arg2_type = ARG_ANYTHING,
2715};
2716
2717BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2718 u16, vlan_tci)
2719{
2720 int ret;
2721
2722 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2723 vlan_proto != htons(ETH_P_8021AD)))
2724 vlan_proto = htons(ETH_P_8021Q);
2725
2726 bpf_push_mac_rcsum(skb);
2727 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2728 bpf_pull_mac_rcsum(skb);
2729
2730 bpf_compute_data_pointers(skb);
2731 return ret;
2732}
2733
2734static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2735 .func = bpf_skb_vlan_push,
2736 .gpl_only = false,
2737 .ret_type = RET_INTEGER,
2738 .arg1_type = ARG_PTR_TO_CTX,
2739 .arg2_type = ARG_ANYTHING,
2740 .arg3_type = ARG_ANYTHING,
2741};
2742
2743BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2744{
2745 int ret;
2746
2747 bpf_push_mac_rcsum(skb);
2748 ret = skb_vlan_pop(skb);
2749 bpf_pull_mac_rcsum(skb);
2750
2751 bpf_compute_data_pointers(skb);
2752 return ret;
2753}
2754
2755static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2756 .func = bpf_skb_vlan_pop,
2757 .gpl_only = false,
2758 .ret_type = RET_INTEGER,
2759 .arg1_type = ARG_PTR_TO_CTX,
2760};
2761
2762static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2763{
2764
2765
2766
2767 skb_push(skb, len);
2768 memmove(skb->data, skb->data + len, off);
2769 memset(skb->data + off, 0, len);
2770
2771
2772
2773
2774
2775
2776 return 0;
2777}
2778
2779static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2780{
2781
2782
2783
2784 if (unlikely(!pskb_may_pull(skb, off + len)))
2785 return -ENOMEM;
2786
2787 skb_postpull_rcsum(skb, skb->data + off, len);
2788 memmove(skb->data + len, skb->data, off);
2789 __skb_pull(skb, len);
2790
2791 return 0;
2792}
2793
2794static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2795{
2796 bool trans_same = skb->transport_header == skb->network_header;
2797 int ret;
2798
2799
2800
2801
2802
2803 ret = bpf_skb_generic_push(skb, off, len);
2804 if (likely(!ret)) {
2805 skb->mac_header -= len;
2806 skb->network_header -= len;
2807 if (trans_same)
2808 skb->transport_header = skb->network_header;
2809 }
2810
2811 return ret;
2812}
2813
2814static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2815{
2816 bool trans_same = skb->transport_header == skb->network_header;
2817 int ret;
2818
2819
2820 ret = bpf_skb_generic_pop(skb, off, len);
2821 if (likely(!ret)) {
2822 skb->mac_header += len;
2823 skb->network_header += len;
2824 if (trans_same)
2825 skb->transport_header = skb->network_header;
2826 }
2827
2828 return ret;
2829}
2830
2831static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2832{
2833 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2834 u32 off = skb_mac_header_len(skb);
2835 int ret;
2836
2837 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2838 return -ENOTSUPP;
2839
2840 ret = skb_cow(skb, len_diff);
2841 if (unlikely(ret < 0))
2842 return ret;
2843
2844 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2845 if (unlikely(ret < 0))
2846 return ret;
2847
2848 if (skb_is_gso(skb)) {
2849 struct skb_shared_info *shinfo = skb_shinfo(skb);
2850
2851
2852
2853
2854 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2855 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2856 shinfo->gso_type |= SKB_GSO_TCPV6;
2857 }
2858
2859
2860 skb_decrease_gso_size(shinfo, len_diff);
2861
2862 shinfo->gso_type |= SKB_GSO_DODGY;
2863 shinfo->gso_segs = 0;
2864 }
2865
2866 skb->protocol = htons(ETH_P_IPV6);
2867 skb_clear_hash(skb);
2868
2869 return 0;
2870}
2871
2872static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2873{
2874 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2875 u32 off = skb_mac_header_len(skb);
2876 int ret;
2877
2878 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2879 return -ENOTSUPP;
2880
2881 ret = skb_unclone(skb, GFP_ATOMIC);
2882 if (unlikely(ret < 0))
2883 return ret;
2884
2885 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2886 if (unlikely(ret < 0))
2887 return ret;
2888
2889 if (skb_is_gso(skb)) {
2890 struct skb_shared_info *shinfo = skb_shinfo(skb);
2891
2892
2893
2894
2895 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2896 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2897 shinfo->gso_type |= SKB_GSO_TCPV4;
2898 }
2899
2900
2901 skb_increase_gso_size(shinfo, len_diff);
2902
2903 shinfo->gso_type |= SKB_GSO_DODGY;
2904 shinfo->gso_segs = 0;
2905 }
2906
2907 skb->protocol = htons(ETH_P_IP);
2908 skb_clear_hash(skb);
2909
2910 return 0;
2911}
2912
2913static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2914{
2915 __be16 from_proto = skb->protocol;
2916
2917 if (from_proto == htons(ETH_P_IP) &&
2918 to_proto == htons(ETH_P_IPV6))
2919 return bpf_skb_proto_4_to_6(skb);
2920
2921 if (from_proto == htons(ETH_P_IPV6) &&
2922 to_proto == htons(ETH_P_IP))
2923 return bpf_skb_proto_6_to_4(skb);
2924
2925 return -ENOTSUPP;
2926}
2927
2928BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2929 u64, flags)
2930{
2931 int ret;
2932
2933 if (unlikely(flags))
2934 return -EINVAL;
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953 ret = bpf_skb_proto_xlat(skb, proto);
2954 bpf_compute_data_pointers(skb);
2955 return ret;
2956}
2957
2958static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2959 .func = bpf_skb_change_proto,
2960 .gpl_only = false,
2961 .ret_type = RET_INTEGER,
2962 .arg1_type = ARG_PTR_TO_CTX,
2963 .arg2_type = ARG_ANYTHING,
2964 .arg3_type = ARG_ANYTHING,
2965};
2966
2967BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2968{
2969
2970 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2971 !skb_pkt_type_ok(pkt_type)))
2972 return -EINVAL;
2973
2974 skb->pkt_type = pkt_type;
2975 return 0;
2976}
2977
2978static const struct bpf_func_proto bpf_skb_change_type_proto = {
2979 .func = bpf_skb_change_type,
2980 .gpl_only = false,
2981 .ret_type = RET_INTEGER,
2982 .arg1_type = ARG_PTR_TO_CTX,
2983 .arg2_type = ARG_ANYTHING,
2984};
2985
2986static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2987{
2988 switch (skb->protocol) {
2989 case htons(ETH_P_IP):
2990 return sizeof(struct iphdr);
2991 case htons(ETH_P_IPV6):
2992 return sizeof(struct ipv6hdr);
2993 default:
2994 return ~0U;
2995 }
2996}
2997
2998#define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
2999 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3000
3001#define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3002 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3003 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3004 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3005 BPF_F_ADJ_ROOM_ENCAP_L2( \
3006 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3007
3008static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3009 u64 flags)
3010{
3011 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3012 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3013 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3014 unsigned int gso_type = SKB_GSO_DODGY;
3015 int ret;
3016
3017 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3018
3019 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3020 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3021 return -ENOTSUPP;
3022 }
3023
3024 ret = skb_cow_head(skb, len_diff);
3025 if (unlikely(ret < 0))
3026 return ret;
3027
3028 if (encap) {
3029 if (skb->protocol != htons(ETH_P_IP) &&
3030 skb->protocol != htons(ETH_P_IPV6))
3031 return -ENOTSUPP;
3032
3033 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3034 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3035 return -EINVAL;
3036
3037 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3038 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3039 return -EINVAL;
3040
3041 if (skb->encapsulation)
3042 return -EALREADY;
3043
3044 mac_len = skb->network_header - skb->mac_header;
3045 inner_net = skb->network_header;
3046 if (inner_mac_len > len_diff)
3047 return -EINVAL;
3048 inner_trans = skb->transport_header;
3049 }
3050
3051 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3052 if (unlikely(ret < 0))
3053 return ret;
3054
3055 if (encap) {
3056 skb->inner_mac_header = inner_net - inner_mac_len;
3057 skb->inner_network_header = inner_net;
3058 skb->inner_transport_header = inner_trans;
3059 skb_set_inner_protocol(skb, skb->protocol);
3060
3061 skb->encapsulation = 1;
3062 skb_set_network_header(skb, mac_len);
3063
3064 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3065 gso_type |= SKB_GSO_UDP_TUNNEL;
3066 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3067 gso_type |= SKB_GSO_GRE;
3068 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3069 gso_type |= SKB_GSO_IPXIP6;
3070 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3071 gso_type |= SKB_GSO_IPXIP4;
3072
3073 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3074 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3075 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3076 sizeof(struct ipv6hdr) :
3077 sizeof(struct iphdr);
3078
3079 skb_set_transport_header(skb, mac_len + nh_len);
3080 }
3081
3082
3083 if (skb->protocol == htons(ETH_P_IP) &&
3084 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3085 skb->protocol = htons(ETH_P_IPV6);
3086 else if (skb->protocol == htons(ETH_P_IPV6) &&
3087 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3088 skb->protocol = htons(ETH_P_IP);
3089 }
3090
3091 if (skb_is_gso(skb)) {
3092 struct skb_shared_info *shinfo = skb_shinfo(skb);
3093
3094
3095 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3096 skb_decrease_gso_size(shinfo, len_diff);
3097
3098
3099 shinfo->gso_type |= gso_type;
3100 shinfo->gso_segs = 0;
3101 }
3102
3103 return 0;
3104}
3105
3106static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3107 u64 flags)
3108{
3109 int ret;
3110
3111 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3112 return -EINVAL;
3113
3114 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3115
3116 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3117 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3118 return -ENOTSUPP;
3119 }
3120
3121 ret = skb_unclone(skb, GFP_ATOMIC);
3122 if (unlikely(ret < 0))
3123 return ret;
3124
3125 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3126 if (unlikely(ret < 0))
3127 return ret;
3128
3129 if (skb_is_gso(skb)) {
3130 struct skb_shared_info *shinfo = skb_shinfo(skb);
3131
3132
3133 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3134 skb_increase_gso_size(shinfo, len_diff);
3135
3136
3137 shinfo->gso_type |= SKB_GSO_DODGY;
3138 shinfo->gso_segs = 0;
3139 }
3140
3141 return 0;
3142}
3143
3144static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3145{
3146 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3147 SKB_MAX_ALLOC;
3148}
3149
3150BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3151 u32, mode, u64, flags)
3152{
3153 u32 len_cur, len_diff_abs = abs(len_diff);
3154 u32 len_min = bpf_skb_net_base_len(skb);
3155 u32 len_max = __bpf_skb_max_len(skb);
3156 __be16 proto = skb->protocol;
3157 bool shrink = len_diff < 0;
3158 u32 off;
3159 int ret;
3160
3161 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3162 return -EINVAL;
3163 if (unlikely(len_diff_abs > 0xfffU))
3164 return -EFAULT;
3165 if (unlikely(proto != htons(ETH_P_IP) &&
3166 proto != htons(ETH_P_IPV6)))
3167 return -ENOTSUPP;
3168
3169 off = skb_mac_header_len(skb);
3170 switch (mode) {
3171 case BPF_ADJ_ROOM_NET:
3172 off += bpf_skb_net_base_len(skb);
3173 break;
3174 case BPF_ADJ_ROOM_MAC:
3175 break;
3176 default:
3177 return -ENOTSUPP;
3178 }
3179
3180 len_cur = skb->len - skb_network_offset(skb);
3181 if ((shrink && (len_diff_abs >= len_cur ||
3182 len_cur - len_diff_abs < len_min)) ||
3183 (!shrink && (skb->len + len_diff_abs > len_max &&
3184 !skb_is_gso(skb))))
3185 return -ENOTSUPP;
3186
3187 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3188 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3189
3190 bpf_compute_data_pointers(skb);
3191 return ret;
3192}
3193
3194static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3195 .func = bpf_skb_adjust_room,
3196 .gpl_only = false,
3197 .ret_type = RET_INTEGER,
3198 .arg1_type = ARG_PTR_TO_CTX,
3199 .arg2_type = ARG_ANYTHING,
3200 .arg3_type = ARG_ANYTHING,
3201 .arg4_type = ARG_ANYTHING,
3202};
3203
3204static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3205{
3206 u32 min_len = skb_network_offset(skb);
3207
3208 if (skb_transport_header_was_set(skb))
3209 min_len = skb_transport_offset(skb);
3210 if (skb->ip_summed == CHECKSUM_PARTIAL)
3211 min_len = skb_checksum_start_offset(skb) +
3212 skb->csum_offset + sizeof(__sum16);
3213 return min_len;
3214}
3215
3216static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3217{
3218 unsigned int old_len = skb->len;
3219 int ret;
3220
3221 ret = __skb_grow_rcsum(skb, new_len);
3222 if (!ret)
3223 memset(skb->data + old_len, 0, new_len - old_len);
3224 return ret;
3225}
3226
3227static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3228{
3229 return __skb_trim_rcsum(skb, new_len);
3230}
3231
3232static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3233 u64 flags)
3234{
3235 u32 max_len = __bpf_skb_max_len(skb);
3236 u32 min_len = __bpf_skb_min_len(skb);
3237 int ret;
3238
3239 if (unlikely(flags || new_len > max_len || new_len < min_len))
3240 return -EINVAL;
3241 if (skb->encapsulation)
3242 return -ENOTSUPP;
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260 ret = __bpf_try_make_writable(skb, skb->len);
3261 if (!ret) {
3262 if (new_len > skb->len)
3263 ret = bpf_skb_grow_rcsum(skb, new_len);
3264 else if (new_len < skb->len)
3265 ret = bpf_skb_trim_rcsum(skb, new_len);
3266 if (!ret && skb_is_gso(skb))
3267 skb_gso_reset(skb);
3268 }
3269 return ret;
3270}
3271
3272BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3273 u64, flags)
3274{
3275 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3276
3277 bpf_compute_data_pointers(skb);
3278 return ret;
3279}
3280
3281static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3282 .func = bpf_skb_change_tail,
3283 .gpl_only = false,
3284 .ret_type = RET_INTEGER,
3285 .arg1_type = ARG_PTR_TO_CTX,
3286 .arg2_type = ARG_ANYTHING,
3287 .arg3_type = ARG_ANYTHING,
3288};
3289
3290BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3291 u64, flags)
3292{
3293 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3294
3295 bpf_compute_data_end_sk_skb(skb);
3296 return ret;
3297}
3298
3299static const struct bpf_func_proto sk_skb_change_tail_proto = {
3300 .func = sk_skb_change_tail,
3301 .gpl_only = false,
3302 .ret_type = RET_INTEGER,
3303 .arg1_type = ARG_PTR_TO_CTX,
3304 .arg2_type = ARG_ANYTHING,
3305 .arg3_type = ARG_ANYTHING,
3306};
3307
3308static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3309 u64 flags)
3310{
3311 u32 max_len = __bpf_skb_max_len(skb);
3312 u32 new_len = skb->len + head_room;
3313 int ret;
3314
3315 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3316 new_len < skb->len))
3317 return -EINVAL;
3318
3319 ret = skb_cow(skb, head_room);
3320 if (likely(!ret)) {
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330 __skb_push(skb, head_room);
3331 memset(skb->data, 0, head_room);
3332 skb_reset_mac_header(skb);
3333 }
3334
3335 return ret;
3336}
3337
3338BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3339 u64, flags)
3340{
3341 int ret = __bpf_skb_change_head(skb, head_room, flags);
3342
3343 bpf_compute_data_pointers(skb);
3344 return ret;
3345}
3346
3347static const struct bpf_func_proto bpf_skb_change_head_proto = {
3348 .func = bpf_skb_change_head,
3349 .gpl_only = false,
3350 .ret_type = RET_INTEGER,
3351 .arg1_type = ARG_PTR_TO_CTX,
3352 .arg2_type = ARG_ANYTHING,
3353 .arg3_type = ARG_ANYTHING,
3354};
3355
3356BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3357 u64, flags)
3358{
3359 int ret = __bpf_skb_change_head(skb, head_room, flags);
3360
3361 bpf_compute_data_end_sk_skb(skb);
3362 return ret;
3363}
3364
3365static const struct bpf_func_proto sk_skb_change_head_proto = {
3366 .func = sk_skb_change_head,
3367 .gpl_only = false,
3368 .ret_type = RET_INTEGER,
3369 .arg1_type = ARG_PTR_TO_CTX,
3370 .arg2_type = ARG_ANYTHING,
3371 .arg3_type = ARG_ANYTHING,
3372};
3373static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3374{
3375 return xdp_data_meta_unsupported(xdp) ? 0 :
3376 xdp->data - xdp->data_meta;
3377}
3378
3379BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3380{
3381 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3382 unsigned long metalen = xdp_get_metalen(xdp);
3383 void *data_start = xdp_frame_end + metalen;
3384 void *data = xdp->data + offset;
3385
3386 if (unlikely(data < data_start ||
3387 data > xdp->data_end - ETH_HLEN))
3388 return -EINVAL;
3389
3390 if (metalen)
3391 memmove(xdp->data_meta + offset,
3392 xdp->data_meta, metalen);
3393 xdp->data_meta += offset;
3394 xdp->data = data;
3395
3396 return 0;
3397}
3398
3399static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3400 .func = bpf_xdp_adjust_head,
3401 .gpl_only = false,
3402 .ret_type = RET_INTEGER,
3403 .arg1_type = ARG_PTR_TO_CTX,
3404 .arg2_type = ARG_ANYTHING,
3405};
3406
3407BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3408{
3409 void *data_end = xdp->data_end + offset;
3410
3411
3412 if (unlikely(offset >= 0))
3413 return -EINVAL;
3414
3415 if (unlikely(data_end < xdp->data + ETH_HLEN))
3416 return -EINVAL;
3417
3418 xdp->data_end = data_end;
3419
3420 return 0;
3421}
3422
3423static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3424 .func = bpf_xdp_adjust_tail,
3425 .gpl_only = false,
3426 .ret_type = RET_INTEGER,
3427 .arg1_type = ARG_PTR_TO_CTX,
3428 .arg2_type = ARG_ANYTHING,
3429};
3430
3431BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3432{
3433 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3434 void *meta = xdp->data_meta + offset;
3435 unsigned long metalen = xdp->data - meta;
3436
3437 if (xdp_data_meta_unsupported(xdp))
3438 return -ENOTSUPP;
3439 if (unlikely(meta < xdp_frame_end ||
3440 meta > xdp->data))
3441 return -EINVAL;
3442 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3443 (metalen > 32)))
3444 return -EACCES;
3445
3446 xdp->data_meta = meta;
3447
3448 return 0;
3449}
3450
3451static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3452 .func = bpf_xdp_adjust_meta,
3453 .gpl_only = false,
3454 .ret_type = RET_INTEGER,
3455 .arg1_type = ARG_PTR_TO_CTX,
3456 .arg2_type = ARG_ANYTHING,
3457};
3458
3459static int __bpf_tx_xdp(struct net_device *dev,
3460 struct bpf_map *map,
3461 struct xdp_buff *xdp,
3462 u32 index)
3463{
3464 struct xdp_frame *xdpf;
3465 int err, sent;
3466
3467 if (!dev->netdev_ops->ndo_xdp_xmit) {
3468 return -EOPNOTSUPP;
3469 }
3470
3471 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3472 if (unlikely(err))
3473 return err;
3474
3475 xdpf = convert_to_xdp_frame(xdp);
3476 if (unlikely(!xdpf))
3477 return -EOVERFLOW;
3478
3479 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3480 if (sent <= 0)
3481 return sent;
3482 return 0;
3483}
3484
3485static noinline int
3486xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3487 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3488{
3489 struct net_device *fwd;
3490 u32 index = ri->ifindex;
3491 int err;
3492
3493 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3494 ri->ifindex = 0;
3495 if (unlikely(!fwd)) {
3496 err = -EINVAL;
3497 goto err;
3498 }
3499
3500 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3501 if (unlikely(err))
3502 goto err;
3503
3504 _trace_xdp_redirect(dev, xdp_prog, index);
3505 return 0;
3506err:
3507 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3508 return err;
3509}
3510
3511static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3512 struct bpf_map *map,
3513 struct xdp_buff *xdp,
3514 u32 index)
3515{
3516 int err;
3517
3518 switch (map->map_type) {
3519 case BPF_MAP_TYPE_DEVMAP: {
3520 struct bpf_dtab_netdev *dst = fwd;
3521
3522 err = dev_map_enqueue(dst, xdp, dev_rx);
3523 if (unlikely(err))
3524 return err;
3525 __dev_map_insert_ctx(map, index);
3526 break;
3527 }
3528 case BPF_MAP_TYPE_CPUMAP: {
3529 struct bpf_cpu_map_entry *rcpu = fwd;
3530
3531 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3532 if (unlikely(err))
3533 return err;
3534 __cpu_map_insert_ctx(map, index);
3535 break;
3536 }
3537 case BPF_MAP_TYPE_XSKMAP: {
3538 struct xdp_sock *xs = fwd;
3539
3540 err = __xsk_map_redirect(map, xdp, xs);
3541 return err;
3542 }
3543 default:
3544 break;
3545 }
3546 return 0;
3547}
3548
3549void xdp_do_flush_map(void)
3550{
3551 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3552 struct bpf_map *map = ri->map_to_flush;
3553
3554 ri->map_to_flush = NULL;
3555 if (map) {
3556 switch (map->map_type) {
3557 case BPF_MAP_TYPE_DEVMAP:
3558 __dev_map_flush(map);
3559 break;
3560 case BPF_MAP_TYPE_CPUMAP:
3561 __cpu_map_flush(map);
3562 break;
3563 case BPF_MAP_TYPE_XSKMAP:
3564 __xsk_map_flush(map);
3565 break;
3566 default:
3567 break;
3568 }
3569 }
3570}
3571EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3572
3573static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3574{
3575 switch (map->map_type) {
3576 case BPF_MAP_TYPE_DEVMAP:
3577 return __dev_map_lookup_elem(map, index);
3578 case BPF_MAP_TYPE_CPUMAP:
3579 return __cpu_map_lookup_elem(map, index);
3580 case BPF_MAP_TYPE_XSKMAP:
3581 return __xsk_map_lookup_elem(map, index);
3582 default:
3583 return NULL;
3584 }
3585}
3586
3587void bpf_clear_redirect_map(struct bpf_map *map)
3588{
3589 struct bpf_redirect_info *ri;
3590 int cpu;
3591
3592 for_each_possible_cpu(cpu) {
3593 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3594
3595
3596
3597
3598
3599 if (unlikely(READ_ONCE(ri->map) == map))
3600 cmpxchg(&ri->map, map, NULL);
3601 }
3602}
3603
3604static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3605 struct bpf_prog *xdp_prog, struct bpf_map *map,
3606 struct bpf_redirect_info *ri)
3607{
3608 u32 index = ri->ifindex;
3609 void *fwd = NULL;
3610 int err;
3611
3612 ri->ifindex = 0;
3613 WRITE_ONCE(ri->map, NULL);
3614
3615 fwd = __xdp_map_lookup_elem(map, index);
3616 if (unlikely(!fwd)) {
3617 err = -EINVAL;
3618 goto err;
3619 }
3620 if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3621 xdp_do_flush_map();
3622
3623 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3624 if (unlikely(err))
3625 goto err;
3626
3627 ri->map_to_flush = map;
3628 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3629 return 0;
3630err:
3631 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3632 return err;
3633}
3634
3635int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3636 struct bpf_prog *xdp_prog)
3637{
3638 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3639 struct bpf_map *map = READ_ONCE(ri->map);
3640
3641 if (likely(map))
3642 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3643
3644 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3645}
3646EXPORT_SYMBOL_GPL(xdp_do_redirect);
3647
3648static int xdp_do_generic_redirect_map(struct net_device *dev,
3649 struct sk_buff *skb,
3650 struct xdp_buff *xdp,
3651 struct bpf_prog *xdp_prog,
3652 struct bpf_map *map)
3653{
3654 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3655 u32 index = ri->ifindex;
3656 void *fwd = NULL;
3657 int err = 0;
3658
3659 ri->ifindex = 0;
3660 WRITE_ONCE(ri->map, NULL);
3661
3662 fwd = __xdp_map_lookup_elem(map, index);
3663 if (unlikely(!fwd)) {
3664 err = -EINVAL;
3665 goto err;
3666 }
3667
3668 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3669 struct bpf_dtab_netdev *dst = fwd;
3670
3671 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3672 if (unlikely(err))
3673 goto err;
3674 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3675 struct xdp_sock *xs = fwd;
3676
3677 err = xsk_generic_rcv(xs, xdp);
3678 if (err)
3679 goto err;
3680 consume_skb(skb);
3681 } else {
3682
3683 err = -EBADRQC;
3684 goto err;
3685 }
3686
3687 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3688 return 0;
3689err:
3690 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3691 return err;
3692}
3693
3694int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3695 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3696{
3697 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3698 struct bpf_map *map = READ_ONCE(ri->map);
3699 u32 index = ri->ifindex;
3700 struct net_device *fwd;
3701 int err = 0;
3702
3703 if (map)
3704 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3705 map);
3706 ri->ifindex = 0;
3707 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3708 if (unlikely(!fwd)) {
3709 err = -EINVAL;
3710 goto err;
3711 }
3712
3713 err = xdp_ok_fwd_dev(fwd, skb->len);
3714 if (unlikely(err))
3715 goto err;
3716
3717 skb->dev = fwd;
3718 _trace_xdp_redirect(dev, xdp_prog, index);
3719 generic_xdp_tx(skb, xdp_prog);
3720 return 0;
3721err:
3722 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3723 return err;
3724}
3725EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3726
3727BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3728{
3729 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3730
3731 if (unlikely(flags))
3732 return XDP_ABORTED;
3733
3734 ri->ifindex = ifindex;
3735 ri->flags = flags;
3736 WRITE_ONCE(ri->map, NULL);
3737
3738 return XDP_REDIRECT;
3739}
3740
3741static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3742 .func = bpf_xdp_redirect,
3743 .gpl_only = false,
3744 .ret_type = RET_INTEGER,
3745 .arg1_type = ARG_ANYTHING,
3746 .arg2_type = ARG_ANYTHING,
3747};
3748
3749BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3750 u64, flags)
3751{
3752 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3753
3754 if (unlikely(flags))
3755 return XDP_ABORTED;
3756
3757 ri->ifindex = ifindex;
3758 ri->flags = flags;
3759 WRITE_ONCE(ri->map, map);
3760
3761 return XDP_REDIRECT;
3762}
3763
3764static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3765 .func = bpf_xdp_redirect_map,
3766 .gpl_only = false,
3767 .ret_type = RET_INTEGER,
3768 .arg1_type = ARG_CONST_MAP_PTR,
3769 .arg2_type = ARG_ANYTHING,
3770 .arg3_type = ARG_ANYTHING,
3771};
3772
3773static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3774 unsigned long off, unsigned long len)
3775{
3776 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3777
3778 if (unlikely(!ptr))
3779 return len;
3780 if (ptr != dst_buff)
3781 memcpy(dst_buff, ptr, len);
3782
3783 return 0;
3784}
3785
3786BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3787 u64, flags, void *, meta, u64, meta_size)
3788{
3789 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3790
3791 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3792 return -EINVAL;
3793 if (unlikely(skb_size > skb->len))
3794 return -EFAULT;
3795
3796 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3797 bpf_skb_copy);
3798}
3799
3800static const struct bpf_func_proto bpf_skb_event_output_proto = {
3801 .func = bpf_skb_event_output,
3802 .gpl_only = true,
3803 .ret_type = RET_INTEGER,
3804 .arg1_type = ARG_PTR_TO_CTX,
3805 .arg2_type = ARG_CONST_MAP_PTR,
3806 .arg3_type = ARG_ANYTHING,
3807 .arg4_type = ARG_PTR_TO_MEM,
3808 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3809};
3810
3811static unsigned short bpf_tunnel_key_af(u64 flags)
3812{
3813 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3814}
3815
3816BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3817 u32, size, u64, flags)
3818{
3819 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3820 u8 compat[sizeof(struct bpf_tunnel_key)];
3821 void *to_orig = to;
3822 int err;
3823
3824 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3825 err = -EINVAL;
3826 goto err_clear;
3827 }
3828 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3829 err = -EPROTO;
3830 goto err_clear;
3831 }
3832 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3833 err = -EINVAL;
3834 switch (size) {
3835 case offsetof(struct bpf_tunnel_key, tunnel_label):
3836 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3837 goto set_compat;
3838 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3839
3840
3841
3842 if (ip_tunnel_info_af(info) != AF_INET)
3843 goto err_clear;
3844set_compat:
3845 to = (struct bpf_tunnel_key *)compat;
3846 break;
3847 default:
3848 goto err_clear;
3849 }
3850 }
3851
3852 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3853 to->tunnel_tos = info->key.tos;
3854 to->tunnel_ttl = info->key.ttl;
3855 to->tunnel_ext = 0;
3856
3857 if (flags & BPF_F_TUNINFO_IPV6) {
3858 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3859 sizeof(to->remote_ipv6));
3860 to->tunnel_label = be32_to_cpu(info->key.label);
3861 } else {
3862 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3863 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3864 to->tunnel_label = 0;
3865 }
3866
3867 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3868 memcpy(to_orig, to, size);
3869
3870 return 0;
3871err_clear:
3872 memset(to_orig, 0, size);
3873 return err;
3874}
3875
3876static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3877 .func = bpf_skb_get_tunnel_key,
3878 .gpl_only = false,
3879 .ret_type = RET_INTEGER,
3880 .arg1_type = ARG_PTR_TO_CTX,
3881 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3882 .arg3_type = ARG_CONST_SIZE,
3883 .arg4_type = ARG_ANYTHING,
3884};
3885
3886BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3887{
3888 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3889 int err;
3890
3891 if (unlikely(!info ||
3892 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3893 err = -ENOENT;
3894 goto err_clear;
3895 }
3896 if (unlikely(size < info->options_len)) {
3897 err = -ENOMEM;
3898 goto err_clear;
3899 }
3900
3901 ip_tunnel_info_opts_get(to, info);
3902 if (size > info->options_len)
3903 memset(to + info->options_len, 0, size - info->options_len);
3904
3905 return info->options_len;
3906err_clear:
3907 memset(to, 0, size);
3908 return err;
3909}
3910
3911static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3912 .func = bpf_skb_get_tunnel_opt,
3913 .gpl_only = false,
3914 .ret_type = RET_INTEGER,
3915 .arg1_type = ARG_PTR_TO_CTX,
3916 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3917 .arg3_type = ARG_CONST_SIZE,
3918};
3919
3920static struct metadata_dst __percpu *md_dst;
3921
3922BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3923 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3924{
3925 struct metadata_dst *md = this_cpu_ptr(md_dst);
3926 u8 compat[sizeof(struct bpf_tunnel_key)];
3927 struct ip_tunnel_info *info;
3928
3929 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3930 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3931 return -EINVAL;
3932 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3933 switch (size) {
3934 case offsetof(struct bpf_tunnel_key, tunnel_label):
3935 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3936 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3937
3938
3939
3940 memcpy(compat, from, size);
3941 memset(compat + size, 0, sizeof(compat) - size);
3942 from = (const struct bpf_tunnel_key *) compat;
3943 break;
3944 default:
3945 return -EINVAL;
3946 }
3947 }
3948 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3949 from->tunnel_ext))
3950 return -EINVAL;
3951
3952 skb_dst_drop(skb);
3953 dst_hold((struct dst_entry *) md);
3954 skb_dst_set(skb, (struct dst_entry *) md);
3955
3956 info = &md->u.tun_info;
3957 memset(info, 0, sizeof(*info));
3958 info->mode = IP_TUNNEL_INFO_TX;
3959
3960 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3961 if (flags & BPF_F_DONT_FRAGMENT)
3962 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3963 if (flags & BPF_F_ZERO_CSUM_TX)
3964 info->key.tun_flags &= ~TUNNEL_CSUM;
3965 if (flags & BPF_F_SEQ_NUMBER)
3966 info->key.tun_flags |= TUNNEL_SEQ;
3967
3968 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3969 info->key.tos = from->tunnel_tos;
3970 info->key.ttl = from->tunnel_ttl;
3971
3972 if (flags & BPF_F_TUNINFO_IPV6) {
3973 info->mode |= IP_TUNNEL_INFO_IPV6;
3974 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3975 sizeof(from->remote_ipv6));
3976 info->key.label = cpu_to_be32(from->tunnel_label) &
3977 IPV6_FLOWLABEL_MASK;
3978 } else {
3979 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3980 }
3981
3982 return 0;
3983}
3984
3985static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3986 .func = bpf_skb_set_tunnel_key,
3987 .gpl_only = false,
3988 .ret_type = RET_INTEGER,
3989 .arg1_type = ARG_PTR_TO_CTX,
3990 .arg2_type = ARG_PTR_TO_MEM,
3991 .arg3_type = ARG_CONST_SIZE,
3992 .arg4_type = ARG_ANYTHING,
3993};
3994
3995BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3996 const u8 *, from, u32, size)
3997{
3998 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3999 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4000
4001 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4002 return -EINVAL;
4003 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4004 return -ENOMEM;
4005
4006 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4007
4008 return 0;
4009}
4010
4011static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4012 .func = bpf_skb_set_tunnel_opt,
4013 .gpl_only = false,
4014 .ret_type = RET_INTEGER,
4015 .arg1_type = ARG_PTR_TO_CTX,
4016 .arg2_type = ARG_PTR_TO_MEM,
4017 .arg3_type = ARG_CONST_SIZE,
4018};
4019
4020static const struct bpf_func_proto *
4021bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4022{
4023 if (!md_dst) {
4024 struct metadata_dst __percpu *tmp;
4025
4026 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4027 METADATA_IP_TUNNEL,
4028 GFP_KERNEL);
4029 if (!tmp)
4030 return NULL;
4031 if (cmpxchg(&md_dst, NULL, tmp))
4032 metadata_dst_free_percpu(tmp);
4033 }
4034
4035 switch (which) {
4036 case BPF_FUNC_skb_set_tunnel_key:
4037 return &bpf_skb_set_tunnel_key_proto;
4038 case BPF_FUNC_skb_set_tunnel_opt:
4039 return &bpf_skb_set_tunnel_opt_proto;
4040 default:
4041 return NULL;
4042 }
4043}
4044
4045BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4046 u32, idx)
4047{
4048 struct bpf_array *array = container_of(map, struct bpf_array, map);
4049 struct cgroup *cgrp;
4050 struct sock *sk;
4051
4052 sk = skb_to_full_sk(skb);
4053 if (!sk || !sk_fullsock(sk))
4054 return -ENOENT;
4055 if (unlikely(idx >= array->map.max_entries))
4056 return -E2BIG;
4057
4058 cgrp = READ_ONCE(array->ptrs[idx]);
4059 if (unlikely(!cgrp))
4060 return -EAGAIN;
4061
4062 return sk_under_cgroup_hierarchy(sk, cgrp);
4063}
4064
4065static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4066 .func = bpf_skb_under_cgroup,
4067 .gpl_only = false,
4068 .ret_type = RET_INTEGER,
4069 .arg1_type = ARG_PTR_TO_CTX,
4070 .arg2_type = ARG_CONST_MAP_PTR,
4071 .arg3_type = ARG_ANYTHING,
4072};
4073
4074#ifdef CONFIG_SOCK_CGROUP_DATA
4075BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4076{
4077 struct sock *sk = skb_to_full_sk(skb);
4078 struct cgroup *cgrp;
4079
4080 if (!sk || !sk_fullsock(sk))
4081 return 0;
4082
4083 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4084 return cgrp->kn->id.id;
4085}
4086
4087static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4088 .func = bpf_skb_cgroup_id,
4089 .gpl_only = false,
4090 .ret_type = RET_INTEGER,
4091 .arg1_type = ARG_PTR_TO_CTX,
4092};
4093
4094BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4095 ancestor_level)
4096{
4097 struct sock *sk = skb_to_full_sk(skb);
4098 struct cgroup *ancestor;
4099 struct cgroup *cgrp;
4100
4101 if (!sk || !sk_fullsock(sk))
4102 return 0;
4103
4104 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4105 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4106 if (!ancestor)
4107 return 0;
4108
4109 return ancestor->kn->id.id;
4110}
4111
4112static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4113 .func = bpf_skb_ancestor_cgroup_id,
4114 .gpl_only = false,
4115 .ret_type = RET_INTEGER,
4116 .arg1_type = ARG_PTR_TO_CTX,
4117 .arg2_type = ARG_ANYTHING,
4118};
4119#endif
4120
4121static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4122 unsigned long off, unsigned long len)
4123{
4124 memcpy(dst_buff, src_buff + off, len);
4125 return 0;
4126}
4127
4128BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4129 u64, flags, void *, meta, u64, meta_size)
4130{
4131 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4132
4133 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4134 return -EINVAL;
4135 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4136 return -EFAULT;
4137
4138 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4139 xdp_size, bpf_xdp_copy);
4140}
4141
4142static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4143 .func = bpf_xdp_event_output,
4144 .gpl_only = true,
4145 .ret_type = RET_INTEGER,
4146 .arg1_type = ARG_PTR_TO_CTX,
4147 .arg2_type = ARG_CONST_MAP_PTR,
4148 .arg3_type = ARG_ANYTHING,
4149 .arg4_type = ARG_PTR_TO_MEM,
4150 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4151};
4152
4153BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4154{
4155 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4156}
4157
4158static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4159 .func = bpf_get_socket_cookie,
4160 .gpl_only = false,
4161 .ret_type = RET_INTEGER,
4162 .arg1_type = ARG_PTR_TO_CTX,
4163};
4164
4165BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4166{
4167 return sock_gen_cookie(ctx->sk);
4168}
4169
4170static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4171 .func = bpf_get_socket_cookie_sock_addr,
4172 .gpl_only = false,
4173 .ret_type = RET_INTEGER,
4174 .arg1_type = ARG_PTR_TO_CTX,
4175};
4176
4177BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4178{
4179 return sock_gen_cookie(ctx->sk);
4180}
4181
4182static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4183 .func = bpf_get_socket_cookie_sock_ops,
4184 .gpl_only = false,
4185 .ret_type = RET_INTEGER,
4186 .arg1_type = ARG_PTR_TO_CTX,
4187};
4188
4189BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4190{
4191 struct sock *sk = sk_to_full_sk(skb->sk);
4192 kuid_t kuid;
4193
4194 if (!sk || !sk_fullsock(sk))
4195 return overflowuid;
4196 kuid = sock_net_uid(sock_net(sk), sk);
4197 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4198}
4199
4200static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4201 .func = bpf_get_socket_uid,
4202 .gpl_only = false,
4203 .ret_type = RET_INTEGER,
4204 .arg1_type = ARG_PTR_TO_CTX,
4205};
4206
4207BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4208 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4209{
4210 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4211 return -EINVAL;
4212
4213 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4214}
4215
4216static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4217 .func = bpf_sockopt_event_output,
4218 .gpl_only = true,
4219 .ret_type = RET_INTEGER,
4220 .arg1_type = ARG_PTR_TO_CTX,
4221 .arg2_type = ARG_CONST_MAP_PTR,
4222 .arg3_type = ARG_ANYTHING,
4223 .arg4_type = ARG_PTR_TO_MEM,
4224 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4225};
4226
4227BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4228 int, level, int, optname, char *, optval, int, optlen)
4229{
4230 struct sock *sk = bpf_sock->sk;
4231 int ret = 0;
4232 int val;
4233
4234 if (!sk_fullsock(sk))
4235 return -EINVAL;
4236
4237 if (level == SOL_SOCKET) {
4238 if (optlen != sizeof(int))
4239 return -EINVAL;
4240 val = *((int *)optval);
4241
4242
4243 switch (optname) {
4244 case SO_RCVBUF:
4245 val = min_t(u32, val, sysctl_rmem_max);
4246 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4247 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4248 break;
4249 case SO_SNDBUF:
4250 val = min_t(u32, val, sysctl_wmem_max);
4251 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4252 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4253 break;
4254 case SO_MAX_PACING_RATE:
4255 if (val != ~0U)
4256 cmpxchg(&sk->sk_pacing_status,
4257 SK_PACING_NONE,
4258 SK_PACING_NEEDED);
4259 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4260 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4261 sk->sk_max_pacing_rate);
4262 break;
4263 case SO_PRIORITY:
4264 sk->sk_priority = val;
4265 break;
4266 case SO_RCVLOWAT:
4267 if (val < 0)
4268 val = INT_MAX;
4269 sk->sk_rcvlowat = val ? : 1;
4270 break;
4271 case SO_MARK:
4272 if (sk->sk_mark != val) {
4273 sk->sk_mark = val;
4274 sk_dst_reset(sk);
4275 }
4276 break;
4277 default:
4278 ret = -EINVAL;
4279 }
4280#ifdef CONFIG_INET
4281 } else if (level == SOL_IP) {
4282 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4283 return -EINVAL;
4284
4285 val = *((int *)optval);
4286
4287 switch (optname) {
4288 case IP_TOS:
4289 if (val < -1 || val > 0xff) {
4290 ret = -EINVAL;
4291 } else {
4292 struct inet_sock *inet = inet_sk(sk);
4293
4294 if (val == -1)
4295 val = 0;
4296 inet->tos = val;
4297 }
4298 break;
4299 default:
4300 ret = -EINVAL;
4301 }
4302#if IS_ENABLED(CONFIG_IPV6)
4303 } else if (level == SOL_IPV6) {
4304 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4305 return -EINVAL;
4306
4307 val = *((int *)optval);
4308
4309 switch (optname) {
4310 case IPV6_TCLASS:
4311 if (val < -1 || val > 0xff) {
4312 ret = -EINVAL;
4313 } else {
4314 struct ipv6_pinfo *np = inet6_sk(sk);
4315
4316 if (val == -1)
4317 val = 0;
4318 np->tclass = val;
4319 }
4320 break;
4321 default:
4322 ret = -EINVAL;
4323 }
4324#endif
4325 } else if (level == SOL_TCP &&
4326 sk->sk_prot->setsockopt == tcp_setsockopt) {
4327 if (optname == TCP_CONGESTION) {
4328 char name[TCP_CA_NAME_MAX];
4329 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4330
4331 strncpy(name, optval, min_t(long, optlen,
4332 TCP_CA_NAME_MAX-1));
4333 name[TCP_CA_NAME_MAX-1] = 0;
4334 ret = tcp_set_congestion_control(sk, name, false,
4335 reinit);
4336 } else {
4337 struct tcp_sock *tp = tcp_sk(sk);
4338
4339 if (optlen != sizeof(int))
4340 return -EINVAL;
4341
4342 val = *((int *)optval);
4343
4344 switch (optname) {
4345 case TCP_BPF_IW:
4346 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4347 ret = -EINVAL;
4348 else
4349 tp->snd_cwnd = val;
4350 break;
4351 case TCP_BPF_SNDCWND_CLAMP:
4352 if (val <= 0) {
4353 ret = -EINVAL;
4354 } else {
4355 tp->snd_cwnd_clamp = val;
4356 tp->snd_ssthresh = val;
4357 }
4358 break;
4359 case TCP_SAVE_SYN:
4360 if (val < 0 || val > 1)
4361 ret = -EINVAL;
4362 else
4363 tp->save_syn = val;
4364 break;
4365 default:
4366 ret = -EINVAL;
4367 }
4368 }
4369#endif
4370 } else {
4371 ret = -EINVAL;
4372 }
4373 return ret;
4374}
4375
4376static const struct bpf_func_proto bpf_setsockopt_proto = {
4377 .func = bpf_setsockopt,
4378 .gpl_only = false,
4379 .ret_type = RET_INTEGER,
4380 .arg1_type = ARG_PTR_TO_CTX,
4381 .arg2_type = ARG_ANYTHING,
4382 .arg3_type = ARG_ANYTHING,
4383 .arg4_type = ARG_PTR_TO_MEM,
4384 .arg5_type = ARG_CONST_SIZE,
4385};
4386
4387BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4388 int, level, int, optname, char *, optval, int, optlen)
4389{
4390 struct sock *sk = bpf_sock->sk;
4391
4392 if (!sk_fullsock(sk))
4393 goto err_clear;
4394#ifdef CONFIG_INET
4395 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4396 struct inet_connection_sock *icsk;
4397 struct tcp_sock *tp;
4398
4399 switch (optname) {
4400 case TCP_CONGESTION:
4401 icsk = inet_csk(sk);
4402
4403 if (!icsk->icsk_ca_ops || optlen <= 1)
4404 goto err_clear;
4405 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4406 optval[optlen - 1] = 0;
4407 break;
4408 case TCP_SAVED_SYN:
4409 tp = tcp_sk(sk);
4410
4411 if (optlen <= 0 || !tp->saved_syn ||
4412 optlen > tp->saved_syn[0])
4413 goto err_clear;
4414 memcpy(optval, tp->saved_syn + 1, optlen);
4415 break;
4416 default:
4417 goto err_clear;
4418 }
4419 } else if (level == SOL_IP) {
4420 struct inet_sock *inet = inet_sk(sk);
4421
4422 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4423 goto err_clear;
4424
4425
4426 switch (optname) {
4427 case IP_TOS:
4428 *((int *)optval) = (int)inet->tos;
4429 break;
4430 default:
4431 goto err_clear;
4432 }
4433#if IS_ENABLED(CONFIG_IPV6)
4434 } else if (level == SOL_IPV6) {
4435 struct ipv6_pinfo *np = inet6_sk(sk);
4436
4437 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4438 goto err_clear;
4439
4440
4441 switch (optname) {
4442 case IPV6_TCLASS:
4443 *((int *)optval) = (int)np->tclass;
4444 break;
4445 default:
4446 goto err_clear;
4447 }
4448#endif
4449 } else {
4450 goto err_clear;
4451 }
4452 return 0;
4453#endif
4454err_clear:
4455 memset(optval, 0, optlen);
4456 return -EINVAL;
4457}
4458
4459static const struct bpf_func_proto bpf_getsockopt_proto = {
4460 .func = bpf_getsockopt,
4461 .gpl_only = false,
4462 .ret_type = RET_INTEGER,
4463 .arg1_type = ARG_PTR_TO_CTX,
4464 .arg2_type = ARG_ANYTHING,
4465 .arg3_type = ARG_ANYTHING,
4466 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4467 .arg5_type = ARG_CONST_SIZE,
4468};
4469
4470BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4471 int, argval)
4472{
4473 struct sock *sk = bpf_sock->sk;
4474 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4475
4476 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4477 return -EINVAL;
4478
4479 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4480
4481 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4482}
4483
4484static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4485 .func = bpf_sock_ops_cb_flags_set,
4486 .gpl_only = false,
4487 .ret_type = RET_INTEGER,
4488 .arg1_type = ARG_PTR_TO_CTX,
4489 .arg2_type = ARG_ANYTHING,
4490};
4491
4492const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4493EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4494
4495BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4496 int, addr_len)
4497{
4498#ifdef CONFIG_INET
4499 struct sock *sk = ctx->sk;
4500 int err;
4501
4502
4503
4504
4505 err = -EINVAL;
4506 if (addr_len < offsetofend(struct sockaddr, sa_family))
4507 return err;
4508 if (addr->sa_family == AF_INET) {
4509 if (addr_len < sizeof(struct sockaddr_in))
4510 return err;
4511 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4512 return err;
4513 return __inet_bind(sk, addr, addr_len, true, false);
4514#if IS_ENABLED(CONFIG_IPV6)
4515 } else if (addr->sa_family == AF_INET6) {
4516 if (addr_len < SIN6_LEN_RFC2133)
4517 return err;
4518 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4519 return err;
4520
4521
4522
4523 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4524#endif
4525 }
4526#endif
4527
4528 return -EAFNOSUPPORT;
4529}
4530
4531static const struct bpf_func_proto bpf_bind_proto = {
4532 .func = bpf_bind,
4533 .gpl_only = false,
4534 .ret_type = RET_INTEGER,
4535 .arg1_type = ARG_PTR_TO_CTX,
4536 .arg2_type = ARG_PTR_TO_MEM,
4537 .arg3_type = ARG_CONST_SIZE,
4538};
4539
4540#ifdef CONFIG_XFRM
4541BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4542 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4543{
4544 const struct sec_path *sp = skb_sec_path(skb);
4545 const struct xfrm_state *x;
4546
4547 if (!sp || unlikely(index >= sp->len || flags))
4548 goto err_clear;
4549
4550 x = sp->xvec[index];
4551
4552 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4553 goto err_clear;
4554
4555 to->reqid = x->props.reqid;
4556 to->spi = x->id.spi;
4557 to->family = x->props.family;
4558 to->ext = 0;
4559
4560 if (to->family == AF_INET6) {
4561 memcpy(to->remote_ipv6, x->props.saddr.a6,
4562 sizeof(to->remote_ipv6));
4563 } else {
4564 to->remote_ipv4 = x->props.saddr.a4;
4565 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4566 }
4567
4568 return 0;
4569err_clear:
4570 memset(to, 0, size);
4571 return -EINVAL;
4572}
4573
4574static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4575 .func = bpf_skb_get_xfrm_state,
4576 .gpl_only = false,
4577 .ret_type = RET_INTEGER,
4578 .arg1_type = ARG_PTR_TO_CTX,
4579 .arg2_type = ARG_ANYTHING,
4580 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4581 .arg4_type = ARG_CONST_SIZE,
4582 .arg5_type = ARG_ANYTHING,
4583};
4584#endif
4585
4586#if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4587static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4588 const struct neighbour *neigh,
4589 const struct net_device *dev)
4590{
4591 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4592 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4593 params->h_vlan_TCI = 0;
4594 params->h_vlan_proto = 0;
4595 params->ifindex = dev->ifindex;
4596
4597 return 0;
4598}
4599#endif
4600
4601#if IS_ENABLED(CONFIG_INET)
4602static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4603 u32 flags, bool check_mtu)
4604{
4605 struct fib_nh_common *nhc;
4606 struct in_device *in_dev;
4607 struct neighbour *neigh;
4608 struct net_device *dev;
4609 struct fib_result res;
4610 struct flowi4 fl4;
4611 int err;
4612 u32 mtu;
4613
4614 dev = dev_get_by_index_rcu(net, params->ifindex);
4615 if (unlikely(!dev))
4616 return -ENODEV;
4617
4618
4619 in_dev = __in_dev_get_rcu(dev);
4620 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4621 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4622
4623 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4624 fl4.flowi4_iif = 1;
4625 fl4.flowi4_oif = params->ifindex;
4626 } else {
4627 fl4.flowi4_iif = params->ifindex;
4628 fl4.flowi4_oif = 0;
4629 }
4630 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4631 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4632 fl4.flowi4_flags = 0;
4633
4634 fl4.flowi4_proto = params->l4_protocol;
4635 fl4.daddr = params->ipv4_dst;
4636 fl4.saddr = params->ipv4_src;
4637 fl4.fl4_sport = params->sport;
4638 fl4.fl4_dport = params->dport;
4639
4640 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4641 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4642 struct fib_table *tb;
4643
4644 tb = fib_get_table(net, tbid);
4645 if (unlikely(!tb))
4646 return BPF_FIB_LKUP_RET_NOT_FWDED;
4647
4648 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4649 } else {
4650 fl4.flowi4_mark = 0;
4651 fl4.flowi4_secid = 0;
4652 fl4.flowi4_tun_key.tun_id = 0;
4653 fl4.flowi4_uid = sock_net_uid(net, NULL);
4654
4655 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4656 }
4657
4658 if (err) {
4659
4660 if (err == -EINVAL)
4661 return BPF_FIB_LKUP_RET_BLACKHOLE;
4662 if (err == -EHOSTUNREACH)
4663 return BPF_FIB_LKUP_RET_UNREACHABLE;
4664 if (err == -EACCES)
4665 return BPF_FIB_LKUP_RET_PROHIBIT;
4666
4667 return BPF_FIB_LKUP_RET_NOT_FWDED;
4668 }
4669
4670 if (res.type != RTN_UNICAST)
4671 return BPF_FIB_LKUP_RET_NOT_FWDED;
4672
4673 if (res.fi->fib_nhs > 1)
4674 fib_select_path(net, &res, &fl4, NULL);
4675
4676 if (check_mtu) {
4677 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4678 if (params->tot_len > mtu)
4679 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4680 }
4681
4682 nhc = res.nhc;
4683
4684
4685 if (nhc->nhc_lwtstate)
4686 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4687
4688 dev = nhc->nhc_dev;
4689
4690 params->rt_metric = res.fi->fib_priority;
4691
4692
4693
4694
4695 if (likely(nhc->nhc_gw_family != AF_INET6)) {
4696 if (nhc->nhc_gw_family)
4697 params->ipv4_dst = nhc->nhc_gw.ipv4;
4698
4699 neigh = __ipv4_neigh_lookup_noref(dev,
4700 (__force u32)params->ipv4_dst);
4701 } else {
4702 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4703
4704 params->family = AF_INET6;
4705 *dst = nhc->nhc_gw.ipv6;
4706 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4707 }
4708
4709 if (!neigh)
4710 return BPF_FIB_LKUP_RET_NO_NEIGH;
4711
4712 return bpf_fib_set_fwd_params(params, neigh, dev);
4713}
4714#endif
4715
4716#if IS_ENABLED(CONFIG_IPV6)
4717static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4718 u32 flags, bool check_mtu)
4719{
4720 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4721 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4722 struct fib6_result res = {};
4723 struct neighbour *neigh;
4724 struct net_device *dev;
4725 struct inet6_dev *idev;
4726 struct flowi6 fl6;
4727 int strict = 0;
4728 int oif, err;
4729 u32 mtu;
4730
4731
4732 if (rt6_need_strict(dst) || rt6_need_strict(src))
4733 return BPF_FIB_LKUP_RET_NOT_FWDED;
4734
4735 dev = dev_get_by_index_rcu(net, params->ifindex);
4736 if (unlikely(!dev))
4737 return -ENODEV;
4738
4739 idev = __in6_dev_get_safely(dev);
4740 if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4741 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4742
4743 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4744 fl6.flowi6_iif = 1;
4745 oif = fl6.flowi6_oif = params->ifindex;
4746 } else {
4747 oif = fl6.flowi6_iif = params->ifindex;
4748 fl6.flowi6_oif = 0;
4749 strict = RT6_LOOKUP_F_HAS_SADDR;
4750 }
4751 fl6.flowlabel = params->flowinfo;
4752 fl6.flowi6_scope = 0;
4753 fl6.flowi6_flags = 0;
4754 fl6.mp_hash = 0;
4755
4756 fl6.flowi6_proto = params->l4_protocol;
4757 fl6.daddr = *dst;
4758 fl6.saddr = *src;
4759 fl6.fl6_sport = params->sport;
4760 fl6.fl6_dport = params->dport;
4761
4762 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4763 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4764 struct fib6_table *tb;
4765
4766 tb = ipv6_stub->fib6_get_table(net, tbid);
4767 if (unlikely(!tb))
4768 return BPF_FIB_LKUP_RET_NOT_FWDED;
4769
4770 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4771 strict);
4772 } else {
4773 fl6.flowi6_mark = 0;
4774 fl6.flowi6_secid = 0;
4775 fl6.flowi6_tun_key.tun_id = 0;
4776 fl6.flowi6_uid = sock_net_uid(net, NULL);
4777
4778 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4779 }
4780
4781 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4782 res.f6i == net->ipv6.fib6_null_entry))
4783 return BPF_FIB_LKUP_RET_NOT_FWDED;
4784
4785 switch (res.fib6_type) {
4786
4787 case RTN_UNICAST:
4788 break;
4789 case RTN_BLACKHOLE:
4790 return BPF_FIB_LKUP_RET_BLACKHOLE;
4791 case RTN_UNREACHABLE:
4792 return BPF_FIB_LKUP_RET_UNREACHABLE;
4793 case RTN_PROHIBIT:
4794 return BPF_FIB_LKUP_RET_PROHIBIT;
4795 default:
4796 return BPF_FIB_LKUP_RET_NOT_FWDED;
4797 }
4798
4799 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4800 fl6.flowi6_oif != 0, NULL, strict);
4801
4802 if (check_mtu) {
4803 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4804 if (params->tot_len > mtu)
4805 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4806 }
4807
4808 if (res.nh->fib_nh_lws)
4809 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4810
4811 if (res.nh->fib_nh_gw_family)
4812 *dst = res.nh->fib_nh_gw6;
4813
4814 dev = res.nh->fib_nh_dev;
4815 params->rt_metric = res.f6i->fib6_metric;
4816
4817
4818
4819
4820 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4821 if (!neigh)
4822 return BPF_FIB_LKUP_RET_NO_NEIGH;
4823
4824 return bpf_fib_set_fwd_params(params, neigh, dev);
4825}
4826#endif
4827
4828BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4829 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4830{
4831 if (plen < sizeof(*params))
4832 return -EINVAL;
4833
4834 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4835 return -EINVAL;
4836
4837 switch (params->family) {
4838#if IS_ENABLED(CONFIG_INET)
4839 case AF_INET:
4840 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4841 flags, true);
4842#endif
4843#if IS_ENABLED(CONFIG_IPV6)
4844 case AF_INET6:
4845 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4846 flags, true);
4847#endif
4848 }
4849 return -EAFNOSUPPORT;
4850}
4851
4852static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4853 .func = bpf_xdp_fib_lookup,
4854 .gpl_only = true,
4855 .ret_type = RET_INTEGER,
4856 .arg1_type = ARG_PTR_TO_CTX,
4857 .arg2_type = ARG_PTR_TO_MEM,
4858 .arg3_type = ARG_CONST_SIZE,
4859 .arg4_type = ARG_ANYTHING,
4860};
4861
4862BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4863 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4864{
4865 struct net *net = dev_net(skb->dev);
4866 int rc = -EAFNOSUPPORT;
4867
4868 if (plen < sizeof(*params))
4869 return -EINVAL;
4870
4871 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4872 return -EINVAL;
4873
4874 switch (params->family) {
4875#if IS_ENABLED(CONFIG_INET)
4876 case AF_INET:
4877 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4878 break;
4879#endif
4880#if IS_ENABLED(CONFIG_IPV6)
4881 case AF_INET6:
4882 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4883 break;
4884#endif
4885 }
4886
4887 if (!rc) {
4888 struct net_device *dev;
4889
4890 dev = dev_get_by_index_rcu(net, params->ifindex);
4891 if (!is_skb_forwardable(dev, skb))
4892 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4893 }
4894
4895 return rc;
4896}
4897
4898static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4899 .func = bpf_skb_fib_lookup,
4900 .gpl_only = true,
4901 .ret_type = RET_INTEGER,
4902 .arg1_type = ARG_PTR_TO_CTX,
4903 .arg2_type = ARG_PTR_TO_MEM,
4904 .arg3_type = ARG_CONST_SIZE,
4905 .arg4_type = ARG_ANYTHING,
4906};
4907
4908#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4909static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4910{
4911 int err;
4912 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4913
4914 if (!seg6_validate_srh(srh, len))
4915 return -EINVAL;
4916
4917 switch (type) {
4918 case BPF_LWT_ENCAP_SEG6_INLINE:
4919 if (skb->protocol != htons(ETH_P_IPV6))
4920 return -EBADMSG;
4921
4922 err = seg6_do_srh_inline(skb, srh);
4923 break;
4924 case BPF_LWT_ENCAP_SEG6:
4925 skb_reset_inner_headers(skb);
4926 skb->encapsulation = 1;
4927 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4928 break;
4929 default:
4930 return -EINVAL;
4931 }
4932
4933 bpf_compute_data_pointers(skb);
4934 if (err)
4935 return err;
4936
4937 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4938 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4939
4940 return seg6_lookup_nexthop(skb, NULL, 0);
4941}
4942#endif
4943
4944#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4945static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4946 bool ingress)
4947{
4948 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4949}
4950#endif
4951
4952BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4953 u32, len)
4954{
4955 switch (type) {
4956#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4957 case BPF_LWT_ENCAP_SEG6:
4958 case BPF_LWT_ENCAP_SEG6_INLINE:
4959 return bpf_push_seg6_encap(skb, type, hdr, len);
4960#endif
4961#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4962 case BPF_LWT_ENCAP_IP:
4963 return bpf_push_ip_encap(skb, hdr, len, true );
4964#endif
4965 default:
4966 return -EINVAL;
4967 }
4968}
4969
4970BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4971 void *, hdr, u32, len)
4972{
4973 switch (type) {
4974#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4975 case BPF_LWT_ENCAP_IP:
4976 return bpf_push_ip_encap(skb, hdr, len, false );
4977#endif
4978 default:
4979 return -EINVAL;
4980 }
4981}
4982
4983static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4984 .func = bpf_lwt_in_push_encap,
4985 .gpl_only = false,
4986 .ret_type = RET_INTEGER,
4987 .arg1_type = ARG_PTR_TO_CTX,
4988 .arg2_type = ARG_ANYTHING,
4989 .arg3_type = ARG_PTR_TO_MEM,
4990 .arg4_type = ARG_CONST_SIZE
4991};
4992
4993static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4994 .func = bpf_lwt_xmit_push_encap,
4995 .gpl_only = false,
4996 .ret_type = RET_INTEGER,
4997 .arg1_type = ARG_PTR_TO_CTX,
4998 .arg2_type = ARG_ANYTHING,
4999 .arg3_type = ARG_PTR_TO_MEM,
5000 .arg4_type = ARG_CONST_SIZE
5001};
5002
5003#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5004BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5005 const void *, from, u32, len)
5006{
5007 struct seg6_bpf_srh_state *srh_state =
5008 this_cpu_ptr(&seg6_bpf_srh_states);
5009 struct ipv6_sr_hdr *srh = srh_state->srh;
5010 void *srh_tlvs, *srh_end, *ptr;
5011 int srhoff = 0;
5012
5013 if (srh == NULL)
5014 return -EINVAL;
5015
5016 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5017 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5018
5019 ptr = skb->data + offset;
5020 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5021 srh_state->valid = false;
5022 else if (ptr < (void *)&srh->flags ||
5023 ptr + len > (void *)&srh->segments)
5024 return -EFAULT;
5025
5026 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5027 return -EFAULT;
5028 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5029 return -EINVAL;
5030 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5031
5032 memcpy(skb->data + offset, from, len);
5033 return 0;
5034}
5035
5036static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5037 .func = bpf_lwt_seg6_store_bytes,
5038 .gpl_only = false,
5039 .ret_type = RET_INTEGER,
5040 .arg1_type = ARG_PTR_TO_CTX,
5041 .arg2_type = ARG_ANYTHING,
5042 .arg3_type = ARG_PTR_TO_MEM,
5043 .arg4_type = ARG_CONST_SIZE
5044};
5045
5046static void bpf_update_srh_state(struct sk_buff *skb)
5047{
5048 struct seg6_bpf_srh_state *srh_state =
5049 this_cpu_ptr(&seg6_bpf_srh_states);
5050 int srhoff = 0;
5051
5052 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5053 srh_state->srh = NULL;
5054 } else {
5055 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5056 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5057 srh_state->valid = true;
5058 }
5059}
5060
5061BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5062 u32, action, void *, param, u32, param_len)
5063{
5064 struct seg6_bpf_srh_state *srh_state =
5065 this_cpu_ptr(&seg6_bpf_srh_states);
5066 int hdroff = 0;
5067 int err;
5068
5069 switch (action) {
5070 case SEG6_LOCAL_ACTION_END_X:
5071 if (!seg6_bpf_has_valid_srh(skb))
5072 return -EBADMSG;
5073 if (param_len != sizeof(struct in6_addr))
5074 return -EINVAL;
5075 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5076 case SEG6_LOCAL_ACTION_END_T:
5077 if (!seg6_bpf_has_valid_srh(skb))
5078 return -EBADMSG;
5079 if (param_len != sizeof(int))
5080 return -EINVAL;
5081 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5082 case SEG6_LOCAL_ACTION_END_DT6:
5083 if (!seg6_bpf_has_valid_srh(skb))
5084 return -EBADMSG;
5085 if (param_len != sizeof(int))
5086 return -EINVAL;
5087
5088 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5089 return -EBADMSG;
5090 if (!pskb_pull(skb, hdroff))
5091 return -EBADMSG;
5092
5093 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5094 skb_reset_network_header(skb);
5095 skb_reset_transport_header(skb);
5096 skb->encapsulation = 0;
5097
5098 bpf_compute_data_pointers(skb);
5099 bpf_update_srh_state(skb);
5100 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5101 case SEG6_LOCAL_ACTION_END_B6:
5102 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5103 return -EBADMSG;
5104 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5105 param, param_len);
5106 if (!err)
5107 bpf_update_srh_state(skb);
5108
5109 return err;
5110 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5111 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5112 return -EBADMSG;
5113 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5114 param, param_len);
5115 if (!err)
5116 bpf_update_srh_state(skb);
5117
5118 return err;
5119 default:
5120 return -EINVAL;
5121 }
5122}
5123
5124static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5125 .func = bpf_lwt_seg6_action,
5126 .gpl_only = false,
5127 .ret_type = RET_INTEGER,
5128 .arg1_type = ARG_PTR_TO_CTX,
5129 .arg2_type = ARG_ANYTHING,
5130 .arg3_type = ARG_PTR_TO_MEM,
5131 .arg4_type = ARG_CONST_SIZE
5132};
5133
5134BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5135 s32, len)
5136{
5137 struct seg6_bpf_srh_state *srh_state =
5138 this_cpu_ptr(&seg6_bpf_srh_states);
5139 struct ipv6_sr_hdr *srh = srh_state->srh;
5140 void *srh_end, *srh_tlvs, *ptr;
5141 struct ipv6hdr *hdr;
5142 int srhoff = 0;
5143 int ret;
5144
5145 if (unlikely(srh == NULL))
5146 return -EINVAL;
5147
5148 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5149 ((srh->first_segment + 1) << 4));
5150 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5151 srh_state->hdrlen);
5152 ptr = skb->data + offset;
5153
5154 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5155 return -EFAULT;
5156 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5157 return -EFAULT;
5158
5159 if (len > 0) {
5160 ret = skb_cow_head(skb, len);
5161 if (unlikely(ret < 0))
5162 return ret;
5163
5164 ret = bpf_skb_net_hdr_push(skb, offset, len);
5165 } else {
5166 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5167 }
5168
5169 bpf_compute_data_pointers(skb);
5170 if (unlikely(ret < 0))
5171 return ret;
5172
5173 hdr = (struct ipv6hdr *)skb->data;
5174 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5175
5176 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5177 return -EINVAL;
5178 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5179 srh_state->hdrlen += len;
5180 srh_state->valid = false;
5181 return 0;
5182}
5183
5184static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5185 .func = bpf_lwt_seg6_adjust_srh,
5186 .gpl_only = false,
5187 .ret_type = RET_INTEGER,
5188 .arg1_type = ARG_PTR_TO_CTX,
5189 .arg2_type = ARG_ANYTHING,
5190 .arg3_type = ARG_ANYTHING,
5191};
5192#endif
5193
5194#define CONVERT_COMMON_TCP_SOCK_FIELDS(md_type, CONVERT) \
5195do { \
5196 switch (si->off) { \
5197 case offsetof(md_type, snd_cwnd): \
5198 CONVERT(snd_cwnd); break; \
5199 case offsetof(md_type, srtt_us): \
5200 CONVERT(srtt_us); break; \
5201 case offsetof(md_type, snd_ssthresh): \
5202 CONVERT(snd_ssthresh); break; \
5203 case offsetof(md_type, rcv_nxt): \
5204 CONVERT(rcv_nxt); break; \
5205 case offsetof(md_type, snd_nxt): \
5206 CONVERT(snd_nxt); break; \
5207 case offsetof(md_type, snd_una): \
5208 CONVERT(snd_una); break; \
5209 case offsetof(md_type, mss_cache): \
5210 CONVERT(mss_cache); break; \
5211 case offsetof(md_type, ecn_flags): \
5212 CONVERT(ecn_flags); break; \
5213 case offsetof(md_type, rate_delivered): \
5214 CONVERT(rate_delivered); break; \
5215 case offsetof(md_type, rate_interval_us): \
5216 CONVERT(rate_interval_us); break; \
5217 case offsetof(md_type, packets_out): \
5218 CONVERT(packets_out); break; \
5219 case offsetof(md_type, retrans_out): \
5220 CONVERT(retrans_out); break; \
5221 case offsetof(md_type, total_retrans): \
5222 CONVERT(total_retrans); break; \
5223 case offsetof(md_type, segs_in): \
5224 CONVERT(segs_in); break; \
5225 case offsetof(md_type, data_segs_in): \
5226 CONVERT(data_segs_in); break; \
5227 case offsetof(md_type, segs_out): \
5228 CONVERT(segs_out); break; \
5229 case offsetof(md_type, data_segs_out): \
5230 CONVERT(data_segs_out); break; \
5231 case offsetof(md_type, lost_out): \
5232 CONVERT(lost_out); break; \
5233 case offsetof(md_type, sacked_out): \
5234 CONVERT(sacked_out); break; \
5235 case offsetof(md_type, bytes_received): \
5236 CONVERT(bytes_received); break; \
5237 case offsetof(md_type, bytes_acked): \
5238 CONVERT(bytes_acked); break; \
5239 } \
5240} while (0)
5241
5242#ifdef CONFIG_INET
5243static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5244 int dif, int sdif, u8 family, u8 proto)
5245{
5246 bool refcounted = false;
5247 struct sock *sk = NULL;
5248
5249 if (family == AF_INET) {
5250 __be32 src4 = tuple->ipv4.saddr;
5251 __be32 dst4 = tuple->ipv4.daddr;
5252
5253 if (proto == IPPROTO_TCP)
5254 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5255 src4, tuple->ipv4.sport,
5256 dst4, tuple->ipv4.dport,
5257 dif, sdif, &refcounted);
5258 else
5259 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5260 dst4, tuple->ipv4.dport,
5261 dif, sdif, &udp_table, NULL);
5262#if IS_ENABLED(CONFIG_IPV6)
5263 } else {
5264 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5265 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5266
5267 if (proto == IPPROTO_TCP)
5268 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5269 src6, tuple->ipv6.sport,
5270 dst6, ntohs(tuple->ipv6.dport),
5271 dif, sdif, &refcounted);
5272 else if (likely(ipv6_bpf_stub))
5273 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5274 src6, tuple->ipv6.sport,
5275 dst6, tuple->ipv6.dport,
5276 dif, sdif,
5277 &udp_table, NULL);
5278#endif
5279 }
5280
5281 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5282 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5283 sk = NULL;
5284 }
5285 return sk;
5286}
5287
5288
5289
5290
5291
5292
5293static struct sock *
5294__bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5295 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5296 u64 flags)
5297{
5298 struct sock *sk = NULL;
5299 u8 family = AF_UNSPEC;
5300 struct net *net;
5301 int sdif;
5302
5303 if (len == sizeof(tuple->ipv4))
5304 family = AF_INET;
5305 else if (len == sizeof(tuple->ipv6))
5306 family = AF_INET6;
5307 else
5308 return NULL;
5309
5310 if (unlikely(family == AF_UNSPEC || flags ||
5311 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5312 goto out;
5313
5314 if (family == AF_INET)
5315 sdif = inet_sdif(skb);
5316 else
5317 sdif = inet6_sdif(skb);
5318
5319 if ((s32)netns_id < 0) {
5320 net = caller_net;
5321 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5322 } else {
5323 net = get_net_ns_by_id(caller_net, netns_id);
5324 if (unlikely(!net))
5325 goto out;
5326 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5327 put_net(net);
5328 }
5329
5330out:
5331 return sk;
5332}
5333
5334static struct sock *
5335__bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5336 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5337 u64 flags)
5338{
5339 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5340 ifindex, proto, netns_id, flags);
5341
5342 if (sk) {
5343 sk = sk_to_full_sk(sk);
5344 if (!sk_fullsock(sk)) {
5345 if (!sock_flag(sk, SOCK_RCU_FREE))
5346 sock_gen_put(sk);
5347 return NULL;
5348 }
5349 }
5350
5351 return sk;
5352}
5353
5354static struct sock *
5355bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5356 u8 proto, u64 netns_id, u64 flags)
5357{
5358 struct net *caller_net;
5359 int ifindex;
5360
5361 if (skb->dev) {
5362 caller_net = dev_net(skb->dev);
5363 ifindex = skb->dev->ifindex;
5364 } else {
5365 caller_net = sock_net(skb->sk);
5366 ifindex = 0;
5367 }
5368
5369 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5370 netns_id, flags);
5371}
5372
5373static struct sock *
5374bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5375 u8 proto, u64 netns_id, u64 flags)
5376{
5377 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5378 flags);
5379
5380 if (sk) {
5381 sk = sk_to_full_sk(sk);
5382 if (!sk_fullsock(sk)) {
5383 if (!sock_flag(sk, SOCK_RCU_FREE))
5384 sock_gen_put(sk);
5385 return NULL;
5386 }
5387 }
5388
5389 return sk;
5390}
5391
5392BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5393 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5394{
5395 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5396 netns_id, flags);
5397}
5398
5399static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5400 .func = bpf_skc_lookup_tcp,
5401 .gpl_only = false,
5402 .pkt_access = true,
5403 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5404 .arg1_type = ARG_PTR_TO_CTX,
5405 .arg2_type = ARG_PTR_TO_MEM,
5406 .arg3_type = ARG_CONST_SIZE,
5407 .arg4_type = ARG_ANYTHING,
5408 .arg5_type = ARG_ANYTHING,
5409};
5410
5411BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5412 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5413{
5414 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5415 netns_id, flags);
5416}
5417
5418static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5419 .func = bpf_sk_lookup_tcp,
5420 .gpl_only = false,
5421 .pkt_access = true,
5422 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5423 .arg1_type = ARG_PTR_TO_CTX,
5424 .arg2_type = ARG_PTR_TO_MEM,
5425 .arg3_type = ARG_CONST_SIZE,
5426 .arg4_type = ARG_ANYTHING,
5427 .arg5_type = ARG_ANYTHING,
5428};
5429
5430BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5431 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5432{
5433 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5434 netns_id, flags);
5435}
5436
5437static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5438 .func = bpf_sk_lookup_udp,
5439 .gpl_only = false,
5440 .pkt_access = true,
5441 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5442 .arg1_type = ARG_PTR_TO_CTX,
5443 .arg2_type = ARG_PTR_TO_MEM,
5444 .arg3_type = ARG_CONST_SIZE,
5445 .arg4_type = ARG_ANYTHING,
5446 .arg5_type = ARG_ANYTHING,
5447};
5448
5449BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5450{
5451 if (!sock_flag(sk, SOCK_RCU_FREE))
5452 sock_gen_put(sk);
5453 return 0;
5454}
5455
5456static const struct bpf_func_proto bpf_sk_release_proto = {
5457 .func = bpf_sk_release,
5458 .gpl_only = false,
5459 .ret_type = RET_INTEGER,
5460 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5461};
5462
5463BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5464 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5465{
5466 struct net *caller_net = dev_net(ctx->rxq->dev);
5467 int ifindex = ctx->rxq->dev->ifindex;
5468
5469 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5470 ifindex, IPPROTO_UDP, netns_id,
5471 flags);
5472}
5473
5474static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5475 .func = bpf_xdp_sk_lookup_udp,
5476 .gpl_only = false,
5477 .pkt_access = true,
5478 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5479 .arg1_type = ARG_PTR_TO_CTX,
5480 .arg2_type = ARG_PTR_TO_MEM,
5481 .arg3_type = ARG_CONST_SIZE,
5482 .arg4_type = ARG_ANYTHING,
5483 .arg5_type = ARG_ANYTHING,
5484};
5485
5486BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5487 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5488{
5489 struct net *caller_net = dev_net(ctx->rxq->dev);
5490 int ifindex = ctx->rxq->dev->ifindex;
5491
5492 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5493 ifindex, IPPROTO_TCP, netns_id,
5494 flags);
5495}
5496
5497static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5498 .func = bpf_xdp_skc_lookup_tcp,
5499 .gpl_only = false,
5500 .pkt_access = true,
5501 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5502 .arg1_type = ARG_PTR_TO_CTX,
5503 .arg2_type = ARG_PTR_TO_MEM,
5504 .arg3_type = ARG_CONST_SIZE,
5505 .arg4_type = ARG_ANYTHING,
5506 .arg5_type = ARG_ANYTHING,
5507};
5508
5509BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5510 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5511{
5512 struct net *caller_net = dev_net(ctx->rxq->dev);
5513 int ifindex = ctx->rxq->dev->ifindex;
5514
5515 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5516 ifindex, IPPROTO_TCP, netns_id,
5517 flags);
5518}
5519
5520static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5521 .func = bpf_xdp_sk_lookup_tcp,
5522 .gpl_only = false,
5523 .pkt_access = true,
5524 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5525 .arg1_type = ARG_PTR_TO_CTX,
5526 .arg2_type = ARG_PTR_TO_MEM,
5527 .arg3_type = ARG_CONST_SIZE,
5528 .arg4_type = ARG_ANYTHING,
5529 .arg5_type = ARG_ANYTHING,
5530};
5531
5532BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5533 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5534{
5535 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5536 sock_net(ctx->sk), 0,
5537 IPPROTO_TCP, netns_id, flags);
5538}
5539
5540static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5541 .func = bpf_sock_addr_skc_lookup_tcp,
5542 .gpl_only = false,
5543 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5544 .arg1_type = ARG_PTR_TO_CTX,
5545 .arg2_type = ARG_PTR_TO_MEM,
5546 .arg3_type = ARG_CONST_SIZE,
5547 .arg4_type = ARG_ANYTHING,
5548 .arg5_type = ARG_ANYTHING,
5549};
5550
5551BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5552 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5553{
5554 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5555 sock_net(ctx->sk), 0, IPPROTO_TCP,
5556 netns_id, flags);
5557}
5558
5559static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5560 .func = bpf_sock_addr_sk_lookup_tcp,
5561 .gpl_only = false,
5562 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5563 .arg1_type = ARG_PTR_TO_CTX,
5564 .arg2_type = ARG_PTR_TO_MEM,
5565 .arg3_type = ARG_CONST_SIZE,
5566 .arg4_type = ARG_ANYTHING,
5567 .arg5_type = ARG_ANYTHING,
5568};
5569
5570BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5571 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5572{
5573 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5574 sock_net(ctx->sk), 0, IPPROTO_UDP,
5575 netns_id, flags);
5576}
5577
5578static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5579 .func = bpf_sock_addr_sk_lookup_udp,
5580 .gpl_only = false,
5581 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5582 .arg1_type = ARG_PTR_TO_CTX,
5583 .arg2_type = ARG_PTR_TO_MEM,
5584 .arg3_type = ARG_CONST_SIZE,
5585 .arg4_type = ARG_ANYTHING,
5586 .arg5_type = ARG_ANYTHING,
5587};
5588
5589bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5590 struct bpf_insn_access_aux *info)
5591{
5592 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, bytes_acked))
5593 return false;
5594
5595 if (off % size != 0)
5596 return false;
5597
5598 switch (off) {
5599 case offsetof(struct bpf_tcp_sock, bytes_received):
5600 case offsetof(struct bpf_tcp_sock, bytes_acked):
5601 return size == sizeof(__u64);
5602 default:
5603 return size == sizeof(__u32);
5604 }
5605}
5606
5607u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5608 const struct bpf_insn *si,
5609 struct bpf_insn *insn_buf,
5610 struct bpf_prog *prog, u32 *target_size)
5611{
5612 struct bpf_insn *insn = insn_buf;
5613
5614#define BPF_TCP_SOCK_GET_COMMON(FIELD) \
5615 do { \
5616 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) > \
5617 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5618 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5619 si->dst_reg, si->src_reg, \
5620 offsetof(struct tcp_sock, FIELD)); \
5621 } while (0)
5622
5623 CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_tcp_sock,
5624 BPF_TCP_SOCK_GET_COMMON);
5625
5626 if (insn > insn_buf)
5627 return insn - insn_buf;
5628
5629 switch (si->off) {
5630 case offsetof(struct bpf_tcp_sock, rtt_min):
5631 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5632 sizeof(struct minmax));
5633 BUILD_BUG_ON(sizeof(struct minmax) <
5634 sizeof(struct minmax_sample));
5635
5636 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5637 offsetof(struct tcp_sock, rtt_min) +
5638 offsetof(struct minmax_sample, v));
5639 break;
5640 }
5641
5642 return insn - insn_buf;
5643}
5644
5645BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5646{
5647 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5648 return (unsigned long)sk;
5649
5650 return (unsigned long)NULL;
5651}
5652
5653static const struct bpf_func_proto bpf_tcp_sock_proto = {
5654 .func = bpf_tcp_sock,
5655 .gpl_only = false,
5656 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5657 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5658};
5659
5660BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5661{
5662 sk = sk_to_full_sk(sk);
5663
5664 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5665 return (unsigned long)sk;
5666
5667 return (unsigned long)NULL;
5668}
5669
5670static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5671 .func = bpf_get_listener_sock,
5672 .gpl_only = false,
5673 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5674 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5675};
5676
5677BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5678{
5679 unsigned int iphdr_len;
5680
5681 if (skb->protocol == cpu_to_be16(ETH_P_IP))
5682 iphdr_len = sizeof(struct iphdr);
5683 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5684 iphdr_len = sizeof(struct ipv6hdr);
5685 else
5686 return 0;
5687
5688 if (skb_headlen(skb) < iphdr_len)
5689 return 0;
5690
5691 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5692 return 0;
5693
5694 return INET_ECN_set_ce(skb);
5695}
5696
5697static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5698 .func = bpf_skb_ecn_set_ce,
5699 .gpl_only = false,
5700 .ret_type = RET_INTEGER,
5701 .arg1_type = ARG_PTR_TO_CTX,
5702};
5703
5704BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5705 struct tcphdr *, th, u32, th_len)
5706{
5707#ifdef CONFIG_SYN_COOKIES
5708 u32 cookie;
5709 int ret;
5710
5711 if (unlikely(th_len < sizeof(*th)))
5712 return -EINVAL;
5713
5714
5715 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5716 return -EINVAL;
5717
5718 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5719 return -EINVAL;
5720
5721 if (!th->ack || th->rst || th->syn)
5722 return -ENOENT;
5723
5724 if (tcp_synq_no_recent_overflow(sk))
5725 return -ENOENT;
5726
5727 cookie = ntohl(th->ack_seq) - 1;
5728
5729 switch (sk->sk_family) {
5730 case AF_INET:
5731 if (unlikely(iph_len < sizeof(struct iphdr)))
5732 return -EINVAL;
5733
5734 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5735 break;
5736
5737#if IS_BUILTIN(CONFIG_IPV6)
5738 case AF_INET6:
5739 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5740 return -EINVAL;
5741
5742 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5743 break;
5744#endif
5745
5746 default:
5747 return -EPROTONOSUPPORT;
5748 }
5749
5750 if (ret > 0)
5751 return 0;
5752
5753 return -ENOENT;
5754#else
5755 return -ENOTSUPP;
5756#endif
5757}
5758
5759static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5760 .func = bpf_tcp_check_syncookie,
5761 .gpl_only = true,
5762 .pkt_access = true,
5763 .ret_type = RET_INTEGER,
5764 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5765 .arg2_type = ARG_PTR_TO_MEM,
5766 .arg3_type = ARG_CONST_SIZE,
5767 .arg4_type = ARG_PTR_TO_MEM,
5768 .arg5_type = ARG_CONST_SIZE,
5769};
5770
5771#endif
5772
5773bool bpf_helper_changes_pkt_data(void *func)
5774{
5775 if (func == bpf_skb_vlan_push ||
5776 func == bpf_skb_vlan_pop ||
5777 func == bpf_skb_store_bytes ||
5778 func == bpf_skb_change_proto ||
5779 func == bpf_skb_change_head ||
5780 func == sk_skb_change_head ||
5781 func == bpf_skb_change_tail ||
5782 func == sk_skb_change_tail ||
5783 func == bpf_skb_adjust_room ||
5784 func == bpf_skb_pull_data ||
5785 func == sk_skb_pull_data ||
5786 func == bpf_clone_redirect ||
5787 func == bpf_l3_csum_replace ||
5788 func == bpf_l4_csum_replace ||
5789 func == bpf_xdp_adjust_head ||
5790 func == bpf_xdp_adjust_meta ||
5791 func == bpf_msg_pull_data ||
5792 func == bpf_msg_push_data ||
5793 func == bpf_msg_pop_data ||
5794 func == bpf_xdp_adjust_tail ||
5795#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5796 func == bpf_lwt_seg6_store_bytes ||
5797 func == bpf_lwt_seg6_adjust_srh ||
5798 func == bpf_lwt_seg6_action ||
5799#endif
5800 func == bpf_lwt_in_push_encap ||
5801 func == bpf_lwt_xmit_push_encap)
5802 return true;
5803
5804 return false;
5805}
5806
5807static const struct bpf_func_proto *
5808bpf_base_func_proto(enum bpf_func_id func_id)
5809{
5810 switch (func_id) {
5811 case BPF_FUNC_map_lookup_elem:
5812 return &bpf_map_lookup_elem_proto;
5813 case BPF_FUNC_map_update_elem:
5814 return &bpf_map_update_elem_proto;
5815 case BPF_FUNC_map_delete_elem:
5816 return &bpf_map_delete_elem_proto;
5817 case BPF_FUNC_map_push_elem:
5818 return &bpf_map_push_elem_proto;
5819 case BPF_FUNC_map_pop_elem:
5820 return &bpf_map_pop_elem_proto;
5821 case BPF_FUNC_map_peek_elem:
5822 return &bpf_map_peek_elem_proto;
5823 case BPF_FUNC_get_prandom_u32:
5824 return &bpf_get_prandom_u32_proto;
5825 case BPF_FUNC_get_smp_processor_id:
5826 return &bpf_get_raw_smp_processor_id_proto;
5827 case BPF_FUNC_get_numa_node_id:
5828 return &bpf_get_numa_node_id_proto;
5829 case BPF_FUNC_tail_call:
5830 return &bpf_tail_call_proto;
5831 case BPF_FUNC_ktime_get_ns:
5832 return &bpf_ktime_get_ns_proto;
5833 default:
5834 break;
5835 }
5836
5837 if (!capable(CAP_SYS_ADMIN))
5838 return NULL;
5839
5840 switch (func_id) {
5841 case BPF_FUNC_spin_lock:
5842 return &bpf_spin_lock_proto;
5843 case BPF_FUNC_spin_unlock:
5844 return &bpf_spin_unlock_proto;
5845 case BPF_FUNC_trace_printk:
5846 return bpf_get_trace_printk_proto();
5847 default:
5848 return NULL;
5849 }
5850}
5851
5852static const struct bpf_func_proto *
5853sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5854{
5855 switch (func_id) {
5856
5857
5858
5859 case BPF_FUNC_get_current_uid_gid:
5860 return &bpf_get_current_uid_gid_proto;
5861 case BPF_FUNC_get_local_storage:
5862 return &bpf_get_local_storage_proto;
5863 default:
5864 return bpf_base_func_proto(func_id);
5865 }
5866}
5867
5868static const struct bpf_func_proto *
5869sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5870{
5871 switch (func_id) {
5872
5873
5874
5875 case BPF_FUNC_get_current_uid_gid:
5876 return &bpf_get_current_uid_gid_proto;
5877 case BPF_FUNC_bind:
5878 switch (prog->expected_attach_type) {
5879 case BPF_CGROUP_INET4_CONNECT:
5880 case BPF_CGROUP_INET6_CONNECT:
5881 return &bpf_bind_proto;
5882 default:
5883 return NULL;
5884 }
5885 case BPF_FUNC_get_socket_cookie:
5886 return &bpf_get_socket_cookie_sock_addr_proto;
5887 case BPF_FUNC_get_local_storage:
5888 return &bpf_get_local_storage_proto;
5889#ifdef CONFIG_INET
5890 case BPF_FUNC_sk_lookup_tcp:
5891 return &bpf_sock_addr_sk_lookup_tcp_proto;
5892 case BPF_FUNC_sk_lookup_udp:
5893 return &bpf_sock_addr_sk_lookup_udp_proto;
5894 case BPF_FUNC_sk_release:
5895 return &bpf_sk_release_proto;
5896 case BPF_FUNC_skc_lookup_tcp:
5897 return &bpf_sock_addr_skc_lookup_tcp_proto;
5898#endif
5899 default:
5900 return bpf_base_func_proto(func_id);
5901 }
5902}
5903
5904static const struct bpf_func_proto *
5905sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5906{
5907 switch (func_id) {
5908 case BPF_FUNC_skb_load_bytes:
5909 return &bpf_skb_load_bytes_proto;
5910 case BPF_FUNC_skb_load_bytes_relative:
5911 return &bpf_skb_load_bytes_relative_proto;
5912 case BPF_FUNC_get_socket_cookie:
5913 return &bpf_get_socket_cookie_proto;
5914 case BPF_FUNC_get_socket_uid:
5915 return &bpf_get_socket_uid_proto;
5916 default:
5917 return bpf_base_func_proto(func_id);
5918 }
5919}
5920
5921const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
5922const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
5923
5924static const struct bpf_func_proto *
5925cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5926{
5927 switch (func_id) {
5928 case BPF_FUNC_get_local_storage:
5929 return &bpf_get_local_storage_proto;
5930 case BPF_FUNC_sk_fullsock:
5931 return &bpf_sk_fullsock_proto;
5932 case BPF_FUNC_sk_storage_get:
5933 return &bpf_sk_storage_get_proto;
5934 case BPF_FUNC_sk_storage_delete:
5935 return &bpf_sk_storage_delete_proto;
5936#ifdef CONFIG_INET
5937 case BPF_FUNC_tcp_sock:
5938 return &bpf_tcp_sock_proto;
5939 case BPF_FUNC_get_listener_sock:
5940 return &bpf_get_listener_sock_proto;
5941 case BPF_FUNC_skb_ecn_set_ce:
5942 return &bpf_skb_ecn_set_ce_proto;
5943#endif
5944 default:
5945 return sk_filter_func_proto(func_id, prog);
5946 }
5947}
5948
5949static const struct bpf_func_proto *
5950tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5951{
5952 switch (func_id) {
5953 case BPF_FUNC_skb_store_bytes:
5954 return &bpf_skb_store_bytes_proto;
5955 case BPF_FUNC_skb_load_bytes:
5956 return &bpf_skb_load_bytes_proto;
5957 case BPF_FUNC_skb_load_bytes_relative:
5958 return &bpf_skb_load_bytes_relative_proto;
5959 case BPF_FUNC_skb_pull_data:
5960 return &bpf_skb_pull_data_proto;
5961 case BPF_FUNC_csum_diff:
5962 return &bpf_csum_diff_proto;
5963 case BPF_FUNC_csum_update:
5964 return &bpf_csum_update_proto;
5965 case BPF_FUNC_l3_csum_replace:
5966 return &bpf_l3_csum_replace_proto;
5967 case BPF_FUNC_l4_csum_replace:
5968 return &bpf_l4_csum_replace_proto;
5969 case BPF_FUNC_clone_redirect:
5970 return &bpf_clone_redirect_proto;
5971 case BPF_FUNC_get_cgroup_classid:
5972 return &bpf_get_cgroup_classid_proto;
5973 case BPF_FUNC_skb_vlan_push:
5974 return &bpf_skb_vlan_push_proto;
5975 case BPF_FUNC_skb_vlan_pop:
5976 return &bpf_skb_vlan_pop_proto;
5977 case BPF_FUNC_skb_change_proto:
5978 return &bpf_skb_change_proto_proto;
5979 case BPF_FUNC_skb_change_type:
5980 return &bpf_skb_change_type_proto;
5981 case BPF_FUNC_skb_adjust_room:
5982 return &bpf_skb_adjust_room_proto;
5983 case BPF_FUNC_skb_change_tail:
5984 return &bpf_skb_change_tail_proto;
5985 case BPF_FUNC_skb_get_tunnel_key:
5986 return &bpf_skb_get_tunnel_key_proto;
5987 case BPF_FUNC_skb_set_tunnel_key:
5988 return bpf_get_skb_set_tunnel_proto(func_id);
5989 case BPF_FUNC_skb_get_tunnel_opt:
5990 return &bpf_skb_get_tunnel_opt_proto;
5991 case BPF_FUNC_skb_set_tunnel_opt:
5992 return bpf_get_skb_set_tunnel_proto(func_id);
5993 case BPF_FUNC_redirect:
5994 return &bpf_redirect_proto;
5995 case BPF_FUNC_get_route_realm:
5996 return &bpf_get_route_realm_proto;
5997 case BPF_FUNC_get_hash_recalc:
5998 return &bpf_get_hash_recalc_proto;
5999 case BPF_FUNC_set_hash_invalid:
6000 return &bpf_set_hash_invalid_proto;
6001 case BPF_FUNC_set_hash:
6002 return &bpf_set_hash_proto;
6003 case BPF_FUNC_perf_event_output:
6004 return &bpf_skb_event_output_proto;
6005 case BPF_FUNC_get_smp_processor_id:
6006 return &bpf_get_smp_processor_id_proto;
6007 case BPF_FUNC_skb_under_cgroup:
6008 return &bpf_skb_under_cgroup_proto;
6009 case BPF_FUNC_get_socket_cookie:
6010 return &bpf_get_socket_cookie_proto;
6011 case BPF_FUNC_get_socket_uid:
6012 return &bpf_get_socket_uid_proto;
6013 case BPF_FUNC_fib_lookup:
6014 return &bpf_skb_fib_lookup_proto;
6015 case BPF_FUNC_sk_fullsock:
6016 return &bpf_sk_fullsock_proto;
6017 case BPF_FUNC_sk_storage_get:
6018 return &bpf_sk_storage_get_proto;
6019 case BPF_FUNC_sk_storage_delete:
6020 return &bpf_sk_storage_delete_proto;
6021#ifdef CONFIG_XFRM
6022 case BPF_FUNC_skb_get_xfrm_state:
6023 return &bpf_skb_get_xfrm_state_proto;
6024#endif
6025#ifdef CONFIG_SOCK_CGROUP_DATA
6026 case BPF_FUNC_skb_cgroup_id:
6027 return &bpf_skb_cgroup_id_proto;
6028 case BPF_FUNC_skb_ancestor_cgroup_id:
6029 return &bpf_skb_ancestor_cgroup_id_proto;
6030#endif
6031#ifdef CONFIG_INET
6032 case BPF_FUNC_sk_lookup_tcp:
6033 return &bpf_sk_lookup_tcp_proto;
6034 case BPF_FUNC_sk_lookup_udp:
6035 return &bpf_sk_lookup_udp_proto;
6036 case BPF_FUNC_sk_release:
6037 return &bpf_sk_release_proto;
6038 case BPF_FUNC_tcp_sock:
6039 return &bpf_tcp_sock_proto;
6040 case BPF_FUNC_get_listener_sock:
6041 return &bpf_get_listener_sock_proto;
6042 case BPF_FUNC_skc_lookup_tcp:
6043 return &bpf_skc_lookup_tcp_proto;
6044 case BPF_FUNC_tcp_check_syncookie:
6045 return &bpf_tcp_check_syncookie_proto;
6046 case BPF_FUNC_skb_ecn_set_ce:
6047 return &bpf_skb_ecn_set_ce_proto;
6048#endif
6049 default:
6050 return bpf_base_func_proto(func_id);
6051 }
6052}
6053
6054static const struct bpf_func_proto *
6055xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6056{
6057 switch (func_id) {
6058 case BPF_FUNC_perf_event_output:
6059 return &bpf_xdp_event_output_proto;
6060 case BPF_FUNC_get_smp_processor_id:
6061 return &bpf_get_smp_processor_id_proto;
6062 case BPF_FUNC_csum_diff:
6063 return &bpf_csum_diff_proto;
6064 case BPF_FUNC_xdp_adjust_head:
6065 return &bpf_xdp_adjust_head_proto;
6066 case BPF_FUNC_xdp_adjust_meta:
6067 return &bpf_xdp_adjust_meta_proto;
6068 case BPF_FUNC_redirect:
6069 return &bpf_xdp_redirect_proto;
6070 case BPF_FUNC_redirect_map:
6071 return &bpf_xdp_redirect_map_proto;
6072 case BPF_FUNC_xdp_adjust_tail:
6073 return &bpf_xdp_adjust_tail_proto;
6074 case BPF_FUNC_fib_lookup:
6075 return &bpf_xdp_fib_lookup_proto;
6076#ifdef CONFIG_INET
6077 case BPF_FUNC_sk_lookup_udp:
6078 return &bpf_xdp_sk_lookup_udp_proto;
6079 case BPF_FUNC_sk_lookup_tcp:
6080 return &bpf_xdp_sk_lookup_tcp_proto;
6081 case BPF_FUNC_sk_release:
6082 return &bpf_sk_release_proto;
6083 case BPF_FUNC_skc_lookup_tcp:
6084 return &bpf_xdp_skc_lookup_tcp_proto;
6085 case BPF_FUNC_tcp_check_syncookie:
6086 return &bpf_tcp_check_syncookie_proto;
6087#endif
6088 default:
6089 return bpf_base_func_proto(func_id);
6090 }
6091}
6092
6093const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6094const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6095
6096static const struct bpf_func_proto *
6097sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6098{
6099 switch (func_id) {
6100 case BPF_FUNC_setsockopt:
6101 return &bpf_setsockopt_proto;
6102 case BPF_FUNC_getsockopt:
6103 return &bpf_getsockopt_proto;
6104 case BPF_FUNC_sock_ops_cb_flags_set:
6105 return &bpf_sock_ops_cb_flags_set_proto;
6106 case BPF_FUNC_sock_map_update:
6107 return &bpf_sock_map_update_proto;
6108 case BPF_FUNC_sock_hash_update:
6109 return &bpf_sock_hash_update_proto;
6110 case BPF_FUNC_get_socket_cookie:
6111 return &bpf_get_socket_cookie_sock_ops_proto;
6112 case BPF_FUNC_get_local_storage:
6113 return &bpf_get_local_storage_proto;
6114 case BPF_FUNC_perf_event_output:
6115 return &bpf_sockopt_event_output_proto;
6116 default:
6117 return bpf_base_func_proto(func_id);
6118 }
6119}
6120
6121const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6122const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6123
6124static const struct bpf_func_proto *
6125sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6126{
6127 switch (func_id) {
6128 case BPF_FUNC_msg_redirect_map:
6129 return &bpf_msg_redirect_map_proto;
6130 case BPF_FUNC_msg_redirect_hash:
6131 return &bpf_msg_redirect_hash_proto;
6132 case BPF_FUNC_msg_apply_bytes:
6133 return &bpf_msg_apply_bytes_proto;
6134 case BPF_FUNC_msg_cork_bytes:
6135 return &bpf_msg_cork_bytes_proto;
6136 case BPF_FUNC_msg_pull_data:
6137 return &bpf_msg_pull_data_proto;
6138 case BPF_FUNC_msg_push_data:
6139 return &bpf_msg_push_data_proto;
6140 case BPF_FUNC_msg_pop_data:
6141 return &bpf_msg_pop_data_proto;
6142 default:
6143 return bpf_base_func_proto(func_id);
6144 }
6145}
6146
6147const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6148const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6149
6150static const struct bpf_func_proto *
6151sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6152{
6153 switch (func_id) {
6154 case BPF_FUNC_skb_store_bytes:
6155 return &bpf_skb_store_bytes_proto;
6156 case BPF_FUNC_skb_load_bytes:
6157 return &bpf_skb_load_bytes_proto;
6158 case BPF_FUNC_skb_pull_data:
6159 return &sk_skb_pull_data_proto;
6160 case BPF_FUNC_skb_change_tail:
6161 return &sk_skb_change_tail_proto;
6162 case BPF_FUNC_skb_change_head:
6163 return &sk_skb_change_head_proto;
6164 case BPF_FUNC_get_socket_cookie:
6165 return &bpf_get_socket_cookie_proto;
6166 case BPF_FUNC_get_socket_uid:
6167 return &bpf_get_socket_uid_proto;
6168 case BPF_FUNC_sk_redirect_map:
6169 return &bpf_sk_redirect_map_proto;
6170 case BPF_FUNC_sk_redirect_hash:
6171 return &bpf_sk_redirect_hash_proto;
6172#ifdef CONFIG_INET
6173 case BPF_FUNC_sk_lookup_tcp:
6174 return &bpf_sk_lookup_tcp_proto;
6175 case BPF_FUNC_sk_lookup_udp:
6176 return &bpf_sk_lookup_udp_proto;
6177 case BPF_FUNC_sk_release:
6178 return &bpf_sk_release_proto;
6179 case BPF_FUNC_skc_lookup_tcp:
6180 return &bpf_skc_lookup_tcp_proto;
6181#endif
6182 default:
6183 return bpf_base_func_proto(func_id);
6184 }
6185}
6186
6187static const struct bpf_func_proto *
6188flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6189{
6190 switch (func_id) {
6191 case BPF_FUNC_skb_load_bytes:
6192 return &bpf_flow_dissector_load_bytes_proto;
6193 default:
6194 return bpf_base_func_proto(func_id);
6195 }
6196}
6197
6198static const struct bpf_func_proto *
6199lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6200{
6201 switch (func_id) {
6202 case BPF_FUNC_skb_load_bytes:
6203 return &bpf_skb_load_bytes_proto;
6204 case BPF_FUNC_skb_pull_data:
6205 return &bpf_skb_pull_data_proto;
6206 case BPF_FUNC_csum_diff:
6207 return &bpf_csum_diff_proto;
6208 case BPF_FUNC_get_cgroup_classid:
6209 return &bpf_get_cgroup_classid_proto;
6210 case BPF_FUNC_get_route_realm:
6211 return &bpf_get_route_realm_proto;
6212 case BPF_FUNC_get_hash_recalc:
6213 return &bpf_get_hash_recalc_proto;
6214 case BPF_FUNC_perf_event_output:
6215 return &bpf_skb_event_output_proto;
6216 case BPF_FUNC_get_smp_processor_id:
6217 return &bpf_get_smp_processor_id_proto;
6218 case BPF_FUNC_skb_under_cgroup:
6219 return &bpf_skb_under_cgroup_proto;
6220 default:
6221 return bpf_base_func_proto(func_id);
6222 }
6223}
6224
6225static const struct bpf_func_proto *
6226lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6227{
6228 switch (func_id) {
6229 case BPF_FUNC_lwt_push_encap:
6230 return &bpf_lwt_in_push_encap_proto;
6231 default:
6232 return lwt_out_func_proto(func_id, prog);
6233 }
6234}
6235
6236static const struct bpf_func_proto *
6237lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6238{
6239 switch (func_id) {
6240 case BPF_FUNC_skb_get_tunnel_key:
6241 return &bpf_skb_get_tunnel_key_proto;
6242 case BPF_FUNC_skb_set_tunnel_key:
6243 return bpf_get_skb_set_tunnel_proto(func_id);
6244 case BPF_FUNC_skb_get_tunnel_opt:
6245 return &bpf_skb_get_tunnel_opt_proto;
6246 case BPF_FUNC_skb_set_tunnel_opt:
6247 return bpf_get_skb_set_tunnel_proto(func_id);
6248 case BPF_FUNC_redirect:
6249 return &bpf_redirect_proto;
6250 case BPF_FUNC_clone_redirect:
6251 return &bpf_clone_redirect_proto;
6252 case BPF_FUNC_skb_change_tail:
6253 return &bpf_skb_change_tail_proto;
6254 case BPF_FUNC_skb_change_head:
6255 return &bpf_skb_change_head_proto;
6256 case BPF_FUNC_skb_store_bytes:
6257 return &bpf_skb_store_bytes_proto;
6258 case BPF_FUNC_csum_update:
6259 return &bpf_csum_update_proto;
6260 case BPF_FUNC_l3_csum_replace:
6261 return &bpf_l3_csum_replace_proto;
6262 case BPF_FUNC_l4_csum_replace:
6263 return &bpf_l4_csum_replace_proto;
6264 case BPF_FUNC_set_hash_invalid:
6265 return &bpf_set_hash_invalid_proto;
6266 case BPF_FUNC_lwt_push_encap:
6267 return &bpf_lwt_xmit_push_encap_proto;
6268 default:
6269 return lwt_out_func_proto(func_id, prog);
6270 }
6271}
6272
6273static const struct bpf_func_proto *
6274lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6275{
6276 switch (func_id) {
6277#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6278 case BPF_FUNC_lwt_seg6_store_bytes:
6279 return &bpf_lwt_seg6_store_bytes_proto;
6280 case BPF_FUNC_lwt_seg6_action:
6281 return &bpf_lwt_seg6_action_proto;
6282 case BPF_FUNC_lwt_seg6_adjust_srh:
6283 return &bpf_lwt_seg6_adjust_srh_proto;
6284#endif
6285 default:
6286 return lwt_out_func_proto(func_id, prog);
6287 }
6288}
6289
6290static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6291 const struct bpf_prog *prog,
6292 struct bpf_insn_access_aux *info)
6293{
6294 const int size_default = sizeof(__u32);
6295
6296 if (off < 0 || off >= sizeof(struct __sk_buff))
6297 return false;
6298
6299
6300 if (off % size != 0)
6301 return false;
6302
6303 switch (off) {
6304 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6305 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6306 return false;
6307 break;
6308 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6309 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6310 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6311 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6312 case bpf_ctx_range(struct __sk_buff, data):
6313 case bpf_ctx_range(struct __sk_buff, data_meta):
6314 case bpf_ctx_range(struct __sk_buff, data_end):
6315 if (size != size_default)
6316 return false;
6317 break;
6318 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6319 return false;
6320 case bpf_ctx_range(struct __sk_buff, tstamp):
6321 if (size != sizeof(__u64))
6322 return false;
6323 break;
6324 case offsetof(struct __sk_buff, sk):
6325 if (type == BPF_WRITE || size != sizeof(__u64))
6326 return false;
6327 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6328 break;
6329 default:
6330
6331 if (type == BPF_WRITE) {
6332 if (size != size_default)
6333 return false;
6334 } else {
6335 bpf_ctx_record_field_size(info, size_default);
6336 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6337 return false;
6338 }
6339 }
6340
6341 return true;
6342}
6343
6344static bool sk_filter_is_valid_access(int off, int size,
6345 enum bpf_access_type type,
6346 const struct bpf_prog *prog,
6347 struct bpf_insn_access_aux *info)
6348{
6349 switch (off) {
6350 case bpf_ctx_range(struct __sk_buff, tc_classid):
6351 case bpf_ctx_range(struct __sk_buff, data):
6352 case bpf_ctx_range(struct __sk_buff, data_meta):
6353 case bpf_ctx_range(struct __sk_buff, data_end):
6354 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6355 case bpf_ctx_range(struct __sk_buff, tstamp):
6356 case bpf_ctx_range(struct __sk_buff, wire_len):
6357 return false;
6358 }
6359
6360 if (type == BPF_WRITE) {
6361 switch (off) {
6362 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6363 break;
6364 default:
6365 return false;
6366 }
6367 }
6368
6369 return bpf_skb_is_valid_access(off, size, type, prog, info);
6370}
6371
6372static bool cg_skb_is_valid_access(int off, int size,
6373 enum bpf_access_type type,
6374 const struct bpf_prog *prog,
6375 struct bpf_insn_access_aux *info)
6376{
6377 switch (off) {
6378 case bpf_ctx_range(struct __sk_buff, tc_classid):
6379 case bpf_ctx_range(struct __sk_buff, data_meta):
6380 case bpf_ctx_range(struct __sk_buff, wire_len):
6381 return false;
6382 case bpf_ctx_range(struct __sk_buff, data):
6383 case bpf_ctx_range(struct __sk_buff, data_end):
6384 if (!capable(CAP_SYS_ADMIN))
6385 return false;
6386 break;
6387 }
6388
6389 if (type == BPF_WRITE) {
6390 switch (off) {
6391 case bpf_ctx_range(struct __sk_buff, mark):
6392 case bpf_ctx_range(struct __sk_buff, priority):
6393 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6394 break;
6395 case bpf_ctx_range(struct __sk_buff, tstamp):
6396 if (!capable(CAP_SYS_ADMIN))
6397 return false;
6398 break;
6399 default:
6400 return false;
6401 }
6402 }
6403
6404 switch (off) {
6405 case bpf_ctx_range(struct __sk_buff, data):
6406 info->reg_type = PTR_TO_PACKET;
6407 break;
6408 case bpf_ctx_range(struct __sk_buff, data_end):
6409 info->reg_type = PTR_TO_PACKET_END;
6410 break;
6411 }
6412
6413 return bpf_skb_is_valid_access(off, size, type, prog, info);
6414}
6415
6416static bool lwt_is_valid_access(int off, int size,
6417 enum bpf_access_type type,
6418 const struct bpf_prog *prog,
6419 struct bpf_insn_access_aux *info)
6420{
6421 switch (off) {
6422 case bpf_ctx_range(struct __sk_buff, tc_classid):
6423 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6424 case bpf_ctx_range(struct __sk_buff, data_meta):
6425 case bpf_ctx_range(struct __sk_buff, tstamp):
6426 case bpf_ctx_range(struct __sk_buff, wire_len):
6427 return false;
6428 }
6429
6430 if (type == BPF_WRITE) {
6431 switch (off) {
6432 case bpf_ctx_range(struct __sk_buff, mark):
6433 case bpf_ctx_range(struct __sk_buff, priority):
6434 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6435 break;
6436 default:
6437 return false;
6438 }
6439 }
6440
6441 switch (off) {
6442 case bpf_ctx_range(struct __sk_buff, data):
6443 info->reg_type = PTR_TO_PACKET;
6444 break;
6445 case bpf_ctx_range(struct __sk_buff, data_end):
6446 info->reg_type = PTR_TO_PACKET_END;
6447 break;
6448 }
6449
6450 return bpf_skb_is_valid_access(off, size, type, prog, info);
6451}
6452
6453
6454static bool __sock_filter_check_attach_type(int off,
6455 enum bpf_access_type access_type,
6456 enum bpf_attach_type attach_type)
6457{
6458 switch (off) {
6459 case offsetof(struct bpf_sock, bound_dev_if):
6460 case offsetof(struct bpf_sock, mark):
6461 case offsetof(struct bpf_sock, priority):
6462 switch (attach_type) {
6463 case BPF_CGROUP_INET_SOCK_CREATE:
6464 goto full_access;
6465 default:
6466 return false;
6467 }
6468 case bpf_ctx_range(struct bpf_sock, src_ip4):
6469 switch (attach_type) {
6470 case BPF_CGROUP_INET4_POST_BIND:
6471 goto read_only;
6472 default:
6473 return false;
6474 }
6475 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6476 switch (attach_type) {
6477 case BPF_CGROUP_INET6_POST_BIND:
6478 goto read_only;
6479 default:
6480 return false;
6481 }
6482 case bpf_ctx_range(struct bpf_sock, src_port):
6483 switch (attach_type) {
6484 case BPF_CGROUP_INET4_POST_BIND:
6485 case BPF_CGROUP_INET6_POST_BIND:
6486 goto read_only;
6487 default:
6488 return false;
6489 }
6490 }
6491read_only:
6492 return access_type == BPF_READ;
6493full_access:
6494 return true;
6495}
6496
6497bool bpf_sock_common_is_valid_access(int off, int size,
6498 enum bpf_access_type type,
6499 struct bpf_insn_access_aux *info)
6500{
6501 switch (off) {
6502 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6503 return false;
6504 default:
6505 return bpf_sock_is_valid_access(off, size, type, info);
6506 }
6507}
6508
6509bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6510 struct bpf_insn_access_aux *info)
6511{
6512 const int size_default = sizeof(__u32);
6513
6514 if (off < 0 || off >= sizeof(struct bpf_sock))
6515 return false;
6516 if (off % size != 0)
6517 return false;
6518
6519 switch (off) {
6520 case offsetof(struct bpf_sock, state):
6521 case offsetof(struct bpf_sock, family):
6522 case offsetof(struct bpf_sock, type):
6523 case offsetof(struct bpf_sock, protocol):
6524 case offsetof(struct bpf_sock, dst_port):
6525 case offsetof(struct bpf_sock, src_port):
6526 case bpf_ctx_range(struct bpf_sock, src_ip4):
6527 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6528 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6529 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6530 bpf_ctx_record_field_size(info, size_default);
6531 return bpf_ctx_narrow_access_ok(off, size, size_default);
6532 }
6533
6534 return size == size_default;
6535}
6536
6537static bool sock_filter_is_valid_access(int off, int size,
6538 enum bpf_access_type type,
6539 const struct bpf_prog *prog,
6540 struct bpf_insn_access_aux *info)
6541{
6542 if (!bpf_sock_is_valid_access(off, size, type, info))
6543 return false;
6544 return __sock_filter_check_attach_type(off, type,
6545 prog->expected_attach_type);
6546}
6547
6548static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6549 const struct bpf_prog *prog)
6550{
6551
6552
6553
6554 return 0;
6555}
6556
6557static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6558 const struct bpf_prog *prog, int drop_verdict)
6559{
6560 struct bpf_insn *insn = insn_buf;
6561
6562 if (!direct_write)
6563 return 0;
6564
6565
6566
6567
6568
6569
6570
6571 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6572 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6573 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6574
6575
6576 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6577 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6578 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6579 BPF_FUNC_skb_pull_data);
6580
6581
6582
6583
6584 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6585 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6586 *insn++ = BPF_EXIT_INSN();
6587
6588
6589 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6590
6591 *insn++ = prog->insnsi[0];
6592
6593 return insn - insn_buf;
6594}
6595
6596static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6597 struct bpf_insn *insn_buf)
6598{
6599 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6600 struct bpf_insn *insn = insn_buf;
6601
6602
6603 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6604 if (!indirect) {
6605 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6606 } else {
6607 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6608 if (orig->imm)
6609 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6610 }
6611
6612 switch (BPF_SIZE(orig->code)) {
6613 case BPF_B:
6614 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6615 break;
6616 case BPF_H:
6617 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6618 break;
6619 case BPF_W:
6620 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6621 break;
6622 }
6623
6624 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6625 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6626 *insn++ = BPF_EXIT_INSN();
6627
6628 return insn - insn_buf;
6629}
6630
6631static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6632 const struct bpf_prog *prog)
6633{
6634 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6635}
6636
6637static bool tc_cls_act_is_valid_access(int off, int size,
6638 enum bpf_access_type type,
6639 const struct bpf_prog *prog,
6640 struct bpf_insn_access_aux *info)
6641{
6642 if (type == BPF_WRITE) {
6643 switch (off) {
6644 case bpf_ctx_range(struct __sk_buff, mark):
6645 case bpf_ctx_range(struct __sk_buff, tc_index):
6646 case bpf_ctx_range(struct __sk_buff, priority):
6647 case bpf_ctx_range(struct __sk_buff, tc_classid):
6648 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6649 case bpf_ctx_range(struct __sk_buff, tstamp):
6650 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6651 break;
6652 default:
6653 return false;
6654 }
6655 }
6656
6657 switch (off) {
6658 case bpf_ctx_range(struct __sk_buff, data):
6659 info->reg_type = PTR_TO_PACKET;
6660 break;
6661 case bpf_ctx_range(struct __sk_buff, data_meta):
6662 info->reg_type = PTR_TO_PACKET_META;
6663 break;
6664 case bpf_ctx_range(struct __sk_buff, data_end):
6665 info->reg_type = PTR_TO_PACKET_END;
6666 break;
6667 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6668 return false;
6669 }
6670
6671 return bpf_skb_is_valid_access(off, size, type, prog, info);
6672}
6673
6674static bool __is_valid_xdp_access(int off, int size)
6675{
6676 if (off < 0 || off >= sizeof(struct xdp_md))
6677 return false;
6678 if (off % size != 0)
6679 return false;
6680 if (size != sizeof(__u32))
6681 return false;
6682
6683 return true;
6684}
6685
6686static bool xdp_is_valid_access(int off, int size,
6687 enum bpf_access_type type,
6688 const struct bpf_prog *prog,
6689 struct bpf_insn_access_aux *info)
6690{
6691 if (type == BPF_WRITE) {
6692 if (bpf_prog_is_dev_bound(prog->aux)) {
6693 switch (off) {
6694 case offsetof(struct xdp_md, rx_queue_index):
6695 return __is_valid_xdp_access(off, size);
6696 }
6697 }
6698 return false;
6699 }
6700
6701 switch (off) {
6702 case offsetof(struct xdp_md, data):
6703 info->reg_type = PTR_TO_PACKET;
6704 break;
6705 case offsetof(struct xdp_md, data_meta):
6706 info->reg_type = PTR_TO_PACKET_META;
6707 break;
6708 case offsetof(struct xdp_md, data_end):
6709 info->reg_type = PTR_TO_PACKET_END;
6710 break;
6711 }
6712
6713 return __is_valid_xdp_access(off, size);
6714}
6715
6716void bpf_warn_invalid_xdp_action(u32 act)
6717{
6718 const u32 act_max = XDP_REDIRECT;
6719
6720 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6721 act > act_max ? "Illegal" : "Driver unsupported",
6722 act);
6723}
6724EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6725
6726static bool sock_addr_is_valid_access(int off, int size,
6727 enum bpf_access_type type,
6728 const struct bpf_prog *prog,
6729 struct bpf_insn_access_aux *info)
6730{
6731 const int size_default = sizeof(__u32);
6732
6733 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6734 return false;
6735 if (off % size != 0)
6736 return false;
6737
6738
6739
6740
6741 switch (off) {
6742 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6743 switch (prog->expected_attach_type) {
6744 case BPF_CGROUP_INET4_BIND:
6745 case BPF_CGROUP_INET4_CONNECT:
6746 case BPF_CGROUP_UDP4_SENDMSG:
6747 case BPF_CGROUP_UDP4_RECVMSG:
6748 break;
6749 default:
6750 return false;
6751 }
6752 break;
6753 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6754 switch (prog->expected_attach_type) {
6755 case BPF_CGROUP_INET6_BIND:
6756 case BPF_CGROUP_INET6_CONNECT:
6757 case BPF_CGROUP_UDP6_SENDMSG:
6758 case BPF_CGROUP_UDP6_RECVMSG:
6759 break;
6760 default:
6761 return false;
6762 }
6763 break;
6764 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6765 switch (prog->expected_attach_type) {
6766 case BPF_CGROUP_UDP4_SENDMSG:
6767 break;
6768 default:
6769 return false;
6770 }
6771 break;
6772 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6773 msg_src_ip6[3]):
6774 switch (prog->expected_attach_type) {
6775 case BPF_CGROUP_UDP6_SENDMSG:
6776 break;
6777 default:
6778 return false;
6779 }
6780 break;
6781 }
6782
6783 switch (off) {
6784 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6785 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6786 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6787 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6788 msg_src_ip6[3]):
6789
6790 if (type == BPF_READ) {
6791 bpf_ctx_record_field_size(info, size_default);
6792 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6793 return false;
6794 } else {
6795 if (size != size_default)
6796 return false;
6797 }
6798 break;
6799 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6800 if (size != size_default)
6801 return false;
6802 break;
6803 default:
6804 if (type == BPF_READ) {
6805 if (size != size_default)
6806 return false;
6807 } else {
6808 return false;
6809 }
6810 }
6811
6812 return true;
6813}
6814
6815static bool sock_ops_is_valid_access(int off, int size,
6816 enum bpf_access_type type,
6817 const struct bpf_prog *prog,
6818 struct bpf_insn_access_aux *info)
6819{
6820 const int size_default = sizeof(__u32);
6821
6822 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6823 return false;
6824
6825
6826 if (off % size != 0)
6827 return false;
6828
6829 if (type == BPF_WRITE) {
6830 switch (off) {
6831 case offsetof(struct bpf_sock_ops, reply):
6832 case offsetof(struct bpf_sock_ops, sk_txhash):
6833 if (size != size_default)
6834 return false;
6835 break;
6836 default:
6837 return false;
6838 }
6839 } else {
6840 switch (off) {
6841 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6842 bytes_acked):
6843 if (size != sizeof(__u64))
6844 return false;
6845 break;
6846 default:
6847 if (size != size_default)
6848 return false;
6849 break;
6850 }
6851 }
6852
6853 return true;
6854}
6855
6856static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6857 const struct bpf_prog *prog)
6858{
6859 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6860}
6861
6862static bool sk_skb_is_valid_access(int off, int size,
6863 enum bpf_access_type type,
6864 const struct bpf_prog *prog,
6865 struct bpf_insn_access_aux *info)
6866{
6867 switch (off) {
6868 case bpf_ctx_range(struct __sk_buff, tc_classid):
6869 case bpf_ctx_range(struct __sk_buff, data_meta):
6870 case bpf_ctx_range(struct __sk_buff, tstamp):
6871 case bpf_ctx_range(struct __sk_buff, wire_len):
6872 return false;
6873 }
6874
6875 if (type == BPF_WRITE) {
6876 switch (off) {
6877 case bpf_ctx_range(struct __sk_buff, tc_index):
6878 case bpf_ctx_range(struct __sk_buff, priority):
6879 break;
6880 default:
6881 return false;
6882 }
6883 }
6884
6885 switch (off) {
6886 case bpf_ctx_range(struct __sk_buff, mark):
6887 return false;
6888 case bpf_ctx_range(struct __sk_buff, data):
6889 info->reg_type = PTR_TO_PACKET;
6890 break;
6891 case bpf_ctx_range(struct __sk_buff, data_end):
6892 info->reg_type = PTR_TO_PACKET_END;
6893 break;
6894 }
6895
6896 return bpf_skb_is_valid_access(off, size, type, prog, info);
6897}
6898
6899static bool sk_msg_is_valid_access(int off, int size,
6900 enum bpf_access_type type,
6901 const struct bpf_prog *prog,
6902 struct bpf_insn_access_aux *info)
6903{
6904 if (type == BPF_WRITE)
6905 return false;
6906
6907 if (off % size != 0)
6908 return false;
6909
6910 switch (off) {
6911 case offsetof(struct sk_msg_md, data):
6912 info->reg_type = PTR_TO_PACKET;
6913 if (size != sizeof(__u64))
6914 return false;
6915 break;
6916 case offsetof(struct sk_msg_md, data_end):
6917 info->reg_type = PTR_TO_PACKET_END;
6918 if (size != sizeof(__u64))
6919 return false;
6920 break;
6921 case bpf_ctx_range(struct sk_msg_md, family):
6922 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
6923 case bpf_ctx_range(struct sk_msg_md, local_ip4):
6924 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
6925 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
6926 case bpf_ctx_range(struct sk_msg_md, remote_port):
6927 case bpf_ctx_range(struct sk_msg_md, local_port):
6928 case bpf_ctx_range(struct sk_msg_md, size):
6929 if (size != sizeof(__u32))
6930 return false;
6931 break;
6932 default:
6933 return false;
6934 }
6935 return true;
6936}
6937
6938static bool flow_dissector_is_valid_access(int off, int size,
6939 enum bpf_access_type type,
6940 const struct bpf_prog *prog,
6941 struct bpf_insn_access_aux *info)
6942{
6943 const int size_default = sizeof(__u32);
6944
6945 if (off < 0 || off >= sizeof(struct __sk_buff))
6946 return false;
6947
6948 if (type == BPF_WRITE)
6949 return false;
6950
6951 switch (off) {
6952 case bpf_ctx_range(struct __sk_buff, data):
6953 if (size != size_default)
6954 return false;
6955 info->reg_type = PTR_TO_PACKET;
6956 return true;
6957 case bpf_ctx_range(struct __sk_buff, data_end):
6958 if (size != size_default)
6959 return false;
6960 info->reg_type = PTR_TO_PACKET_END;
6961 return true;
6962 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6963 if (size != sizeof(__u64))
6964 return false;
6965 info->reg_type = PTR_TO_FLOW_KEYS;
6966 return true;
6967 default:
6968 return false;
6969 }
6970}
6971
6972static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
6973 const struct bpf_insn *si,
6974 struct bpf_insn *insn_buf,
6975 struct bpf_prog *prog,
6976 u32 *target_size)
6977
6978{
6979 struct bpf_insn *insn = insn_buf;
6980
6981 switch (si->off) {
6982 case offsetof(struct __sk_buff, data):
6983 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
6984 si->dst_reg, si->src_reg,
6985 offsetof(struct bpf_flow_dissector, data));
6986 break;
6987
6988 case offsetof(struct __sk_buff, data_end):
6989 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
6990 si->dst_reg, si->src_reg,
6991 offsetof(struct bpf_flow_dissector, data_end));
6992 break;
6993
6994 case offsetof(struct __sk_buff, flow_keys):
6995 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
6996 si->dst_reg, si->src_reg,
6997 offsetof(struct bpf_flow_dissector, flow_keys));
6998 break;
6999 }
7000
7001 return insn - insn_buf;
7002}
7003
7004static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7005 const struct bpf_insn *si,
7006 struct bpf_insn *insn_buf,
7007 struct bpf_prog *prog, u32 *target_size)
7008{
7009 struct bpf_insn *insn = insn_buf;
7010 int off;
7011
7012 switch (si->off) {
7013 case offsetof(struct __sk_buff, len):
7014 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7015 bpf_target_off(struct sk_buff, len, 4,
7016 target_size));
7017 break;
7018
7019 case offsetof(struct __sk_buff, protocol):
7020 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7021 bpf_target_off(struct sk_buff, protocol, 2,
7022 target_size));
7023 break;
7024
7025 case offsetof(struct __sk_buff, vlan_proto):
7026 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7027 bpf_target_off(struct sk_buff, vlan_proto, 2,
7028 target_size));
7029 break;
7030
7031 case offsetof(struct __sk_buff, priority):
7032 if (type == BPF_WRITE)
7033 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7034 bpf_target_off(struct sk_buff, priority, 4,
7035 target_size));
7036 else
7037 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7038 bpf_target_off(struct sk_buff, priority, 4,
7039 target_size));
7040 break;
7041
7042 case offsetof(struct __sk_buff, ingress_ifindex):
7043 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7044 bpf_target_off(struct sk_buff, skb_iif, 4,
7045 target_size));
7046 break;
7047
7048 case offsetof(struct __sk_buff, ifindex):
7049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7050 si->dst_reg, si->src_reg,
7051 offsetof(struct sk_buff, dev));
7052 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7053 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7054 bpf_target_off(struct net_device, ifindex, 4,
7055 target_size));
7056 break;
7057
7058 case offsetof(struct __sk_buff, hash):
7059 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7060 bpf_target_off(struct sk_buff, hash, 4,
7061 target_size));
7062 break;
7063
7064 case offsetof(struct __sk_buff, mark):
7065 if (type == BPF_WRITE)
7066 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7067 bpf_target_off(struct sk_buff, mark, 4,
7068 target_size));
7069 else
7070 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7071 bpf_target_off(struct sk_buff, mark, 4,
7072 target_size));
7073 break;
7074
7075 case offsetof(struct __sk_buff, pkt_type):
7076 *target_size = 1;
7077 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7078 PKT_TYPE_OFFSET());
7079 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7080#ifdef __BIG_ENDIAN_BITFIELD
7081 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7082#endif
7083 break;
7084
7085 case offsetof(struct __sk_buff, queue_mapping):
7086 if (type == BPF_WRITE) {
7087 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7088 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7089 bpf_target_off(struct sk_buff,
7090 queue_mapping,
7091 2, target_size));
7092 } else {
7093 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7094 bpf_target_off(struct sk_buff,
7095 queue_mapping,
7096 2, target_size));
7097 }
7098 break;
7099
7100 case offsetof(struct __sk_buff, vlan_present):
7101 *target_size = 1;
7102 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7103 PKT_VLAN_PRESENT_OFFSET());
7104 if (PKT_VLAN_PRESENT_BIT)
7105 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7106 if (PKT_VLAN_PRESENT_BIT < 7)
7107 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7108 break;
7109
7110 case offsetof(struct __sk_buff, vlan_tci):
7111 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7112 bpf_target_off(struct sk_buff, vlan_tci, 2,
7113 target_size));
7114 break;
7115
7116 case offsetof(struct __sk_buff, cb[0]) ...
7117 offsetofend(struct __sk_buff, cb[4]) - 1:
7118 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
7119 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7120 offsetof(struct qdisc_skb_cb, data)) %
7121 sizeof(__u64));
7122
7123 prog->cb_access = 1;
7124 off = si->off;
7125 off -= offsetof(struct __sk_buff, cb[0]);
7126 off += offsetof(struct sk_buff, cb);
7127 off += offsetof(struct qdisc_skb_cb, data);
7128 if (type == BPF_WRITE)
7129 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7130 si->src_reg, off);
7131 else
7132 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7133 si->src_reg, off);
7134 break;
7135
7136 case offsetof(struct __sk_buff, tc_classid):
7137 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
7138
7139 off = si->off;
7140 off -= offsetof(struct __sk_buff, tc_classid);
7141 off += offsetof(struct sk_buff, cb);
7142 off += offsetof(struct qdisc_skb_cb, tc_classid);
7143 *target_size = 2;
7144 if (type == BPF_WRITE)
7145 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7146 si->src_reg, off);
7147 else
7148 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7149 si->src_reg, off);
7150 break;
7151
7152 case offsetof(struct __sk_buff, data):
7153 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7154 si->dst_reg, si->src_reg,
7155 offsetof(struct sk_buff, data));
7156 break;
7157
7158 case offsetof(struct __sk_buff, data_meta):
7159 off = si->off;
7160 off -= offsetof(struct __sk_buff, data_meta);
7161 off += offsetof(struct sk_buff, cb);
7162 off += offsetof(struct bpf_skb_data_end, data_meta);
7163 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7164 si->src_reg, off);
7165 break;
7166
7167 case offsetof(struct __sk_buff, data_end):
7168 off = si->off;
7169 off -= offsetof(struct __sk_buff, data_end);
7170 off += offsetof(struct sk_buff, cb);
7171 off += offsetof(struct bpf_skb_data_end, data_end);
7172 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7173 si->src_reg, off);
7174 break;
7175
7176 case offsetof(struct __sk_buff, tc_index):
7177#ifdef CONFIG_NET_SCHED
7178 if (type == BPF_WRITE)
7179 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7180 bpf_target_off(struct sk_buff, tc_index, 2,
7181 target_size));
7182 else
7183 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7184 bpf_target_off(struct sk_buff, tc_index, 2,
7185 target_size));
7186#else
7187 *target_size = 2;
7188 if (type == BPF_WRITE)
7189 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7190 else
7191 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7192#endif
7193 break;
7194
7195 case offsetof(struct __sk_buff, napi_id):
7196#if defined(CONFIG_NET_RX_BUSY_POLL)
7197 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7198 bpf_target_off(struct sk_buff, napi_id, 4,
7199 target_size));
7200 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7201 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7202#else
7203 *target_size = 4;
7204 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7205#endif
7206 break;
7207 case offsetof(struct __sk_buff, family):
7208 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7209
7210 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7211 si->dst_reg, si->src_reg,
7212 offsetof(struct sk_buff, sk));
7213 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7214 bpf_target_off(struct sock_common,
7215 skc_family,
7216 2, target_size));
7217 break;
7218 case offsetof(struct __sk_buff, remote_ip4):
7219 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7220
7221 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7222 si->dst_reg, si->src_reg,
7223 offsetof(struct sk_buff, sk));
7224 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7225 bpf_target_off(struct sock_common,
7226 skc_daddr,
7227 4, target_size));
7228 break;
7229 case offsetof(struct __sk_buff, local_ip4):
7230 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7231 skc_rcv_saddr) != 4);
7232
7233 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7234 si->dst_reg, si->src_reg,
7235 offsetof(struct sk_buff, sk));
7236 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7237 bpf_target_off(struct sock_common,
7238 skc_rcv_saddr,
7239 4, target_size));
7240 break;
7241 case offsetof(struct __sk_buff, remote_ip6[0]) ...
7242 offsetof(struct __sk_buff, remote_ip6[3]):
7243#if IS_ENABLED(CONFIG_IPV6)
7244 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7245 skc_v6_daddr.s6_addr32[0]) != 4);
7246
7247 off = si->off;
7248 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7249
7250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7251 si->dst_reg, si->src_reg,
7252 offsetof(struct sk_buff, sk));
7253 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7254 offsetof(struct sock_common,
7255 skc_v6_daddr.s6_addr32[0]) +
7256 off);
7257#else
7258 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7259#endif
7260 break;
7261 case offsetof(struct __sk_buff, local_ip6[0]) ...
7262 offsetof(struct __sk_buff, local_ip6[3]):
7263#if IS_ENABLED(CONFIG_IPV6)
7264 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7265 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7266
7267 off = si->off;
7268 off -= offsetof(struct __sk_buff, local_ip6[0]);
7269
7270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7271 si->dst_reg, si->src_reg,
7272 offsetof(struct sk_buff, sk));
7273 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7274 offsetof(struct sock_common,
7275 skc_v6_rcv_saddr.s6_addr32[0]) +
7276 off);
7277#else
7278 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7279#endif
7280 break;
7281
7282 case offsetof(struct __sk_buff, remote_port):
7283 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7284
7285 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7286 si->dst_reg, si->src_reg,
7287 offsetof(struct sk_buff, sk));
7288 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7289 bpf_target_off(struct sock_common,
7290 skc_dport,
7291 2, target_size));
7292#ifndef __BIG_ENDIAN_BITFIELD
7293 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7294#endif
7295 break;
7296
7297 case offsetof(struct __sk_buff, local_port):
7298 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7299
7300 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7301 si->dst_reg, si->src_reg,
7302 offsetof(struct sk_buff, sk));
7303 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7304 bpf_target_off(struct sock_common,
7305 skc_num, 2, target_size));
7306 break;
7307
7308 case offsetof(struct __sk_buff, tstamp):
7309 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
7310
7311 if (type == BPF_WRITE)
7312 *insn++ = BPF_STX_MEM(BPF_DW,
7313 si->dst_reg, si->src_reg,
7314 bpf_target_off(struct sk_buff,
7315 tstamp, 8,
7316 target_size));
7317 else
7318 *insn++ = BPF_LDX_MEM(BPF_DW,
7319 si->dst_reg, si->src_reg,
7320 bpf_target_off(struct sk_buff,
7321 tstamp, 8,
7322 target_size));
7323 break;
7324
7325 case offsetof(struct __sk_buff, gso_segs):
7326
7327#ifdef NET_SKBUFF_DATA_USES_OFFSET
7328 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7329 si->dst_reg, si->src_reg,
7330 offsetof(struct sk_buff, head));
7331 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7332 BPF_REG_AX, si->src_reg,
7333 offsetof(struct sk_buff, end));
7334 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7335#else
7336 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7337 si->dst_reg, si->src_reg,
7338 offsetof(struct sk_buff, end));
7339#endif
7340 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7341 si->dst_reg, si->dst_reg,
7342 bpf_target_off(struct skb_shared_info,
7343 gso_segs, 2,
7344 target_size));
7345 break;
7346 case offsetof(struct __sk_buff, wire_len):
7347 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
7348
7349 off = si->off;
7350 off -= offsetof(struct __sk_buff, wire_len);
7351 off += offsetof(struct sk_buff, cb);
7352 off += offsetof(struct qdisc_skb_cb, pkt_len);
7353 *target_size = 4;
7354 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7355 break;
7356
7357 case offsetof(struct __sk_buff, sk):
7358 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7359 si->dst_reg, si->src_reg,
7360 offsetof(struct sk_buff, sk));
7361 break;
7362 }
7363
7364 return insn - insn_buf;
7365}
7366
7367u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7368 const struct bpf_insn *si,
7369 struct bpf_insn *insn_buf,
7370 struct bpf_prog *prog, u32 *target_size)
7371{
7372 struct bpf_insn *insn = insn_buf;
7373 int off;
7374
7375 switch (si->off) {
7376 case offsetof(struct bpf_sock, bound_dev_if):
7377 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7378
7379 if (type == BPF_WRITE)
7380 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7381 offsetof(struct sock, sk_bound_dev_if));
7382 else
7383 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7384 offsetof(struct sock, sk_bound_dev_if));
7385 break;
7386
7387 case offsetof(struct bpf_sock, mark):
7388 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7389
7390 if (type == BPF_WRITE)
7391 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7392 offsetof(struct sock, sk_mark));
7393 else
7394 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7395 offsetof(struct sock, sk_mark));
7396 break;
7397
7398 case offsetof(struct bpf_sock, priority):
7399 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7400
7401 if (type == BPF_WRITE)
7402 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7403 offsetof(struct sock, sk_priority));
7404 else
7405 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7406 offsetof(struct sock, sk_priority));
7407 break;
7408
7409 case offsetof(struct bpf_sock, family):
7410 *insn++ = BPF_LDX_MEM(
7411 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7412 si->dst_reg, si->src_reg,
7413 bpf_target_off(struct sock_common,
7414 skc_family,
7415 FIELD_SIZEOF(struct sock_common,
7416 skc_family),
7417 target_size));
7418 break;
7419
7420 case offsetof(struct bpf_sock, type):
7421 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7422 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7423 offsetof(struct sock, __sk_flags_offset));
7424 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7425 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7426 *target_size = 2;
7427 break;
7428
7429 case offsetof(struct bpf_sock, protocol):
7430 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7431 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7432 offsetof(struct sock, __sk_flags_offset));
7433 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7434 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7435 *target_size = 1;
7436 break;
7437
7438 case offsetof(struct bpf_sock, src_ip4):
7439 *insn++ = BPF_LDX_MEM(
7440 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7441 bpf_target_off(struct sock_common, skc_rcv_saddr,
7442 FIELD_SIZEOF(struct sock_common,
7443 skc_rcv_saddr),
7444 target_size));
7445 break;
7446
7447 case offsetof(struct bpf_sock, dst_ip4):
7448 *insn++ = BPF_LDX_MEM(
7449 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7450 bpf_target_off(struct sock_common, skc_daddr,
7451 FIELD_SIZEOF(struct sock_common,
7452 skc_daddr),
7453 target_size));
7454 break;
7455
7456 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7457#if IS_ENABLED(CONFIG_IPV6)
7458 off = si->off;
7459 off -= offsetof(struct bpf_sock, src_ip6[0]);
7460 *insn++ = BPF_LDX_MEM(
7461 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7462 bpf_target_off(
7463 struct sock_common,
7464 skc_v6_rcv_saddr.s6_addr32[0],
7465 FIELD_SIZEOF(struct sock_common,
7466 skc_v6_rcv_saddr.s6_addr32[0]),
7467 target_size) + off);
7468#else
7469 (void)off;
7470 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7471#endif
7472 break;
7473
7474 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7475#if IS_ENABLED(CONFIG_IPV6)
7476 off = si->off;
7477 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7478 *insn++ = BPF_LDX_MEM(
7479 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7480 bpf_target_off(struct sock_common,
7481 skc_v6_daddr.s6_addr32[0],
7482 FIELD_SIZEOF(struct sock_common,
7483 skc_v6_daddr.s6_addr32[0]),
7484 target_size) + off);
7485#else
7486 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7487 *target_size = 4;
7488#endif
7489 break;
7490
7491 case offsetof(struct bpf_sock, src_port):
7492 *insn++ = BPF_LDX_MEM(
7493 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7494 si->dst_reg, si->src_reg,
7495 bpf_target_off(struct sock_common, skc_num,
7496 FIELD_SIZEOF(struct sock_common,
7497 skc_num),
7498 target_size));
7499 break;
7500
7501 case offsetof(struct bpf_sock, dst_port):
7502 *insn++ = BPF_LDX_MEM(
7503 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7504 si->dst_reg, si->src_reg,
7505 bpf_target_off(struct sock_common, skc_dport,
7506 FIELD_SIZEOF(struct sock_common,
7507 skc_dport),
7508 target_size));
7509 break;
7510
7511 case offsetof(struct bpf_sock, state):
7512 *insn++ = BPF_LDX_MEM(
7513 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7514 si->dst_reg, si->src_reg,
7515 bpf_target_off(struct sock_common, skc_state,
7516 FIELD_SIZEOF(struct sock_common,
7517 skc_state),
7518 target_size));
7519 break;
7520 }
7521
7522 return insn - insn_buf;
7523}
7524
7525static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7526 const struct bpf_insn *si,
7527 struct bpf_insn *insn_buf,
7528 struct bpf_prog *prog, u32 *target_size)
7529{
7530 struct bpf_insn *insn = insn_buf;
7531
7532 switch (si->off) {
7533 case offsetof(struct __sk_buff, ifindex):
7534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7535 si->dst_reg, si->src_reg,
7536 offsetof(struct sk_buff, dev));
7537 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7538 bpf_target_off(struct net_device, ifindex, 4,
7539 target_size));
7540 break;
7541 default:
7542 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7543 target_size);
7544 }
7545
7546 return insn - insn_buf;
7547}
7548
7549static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7550 const struct bpf_insn *si,
7551 struct bpf_insn *insn_buf,
7552 struct bpf_prog *prog, u32 *target_size)
7553{
7554 struct bpf_insn *insn = insn_buf;
7555
7556 switch (si->off) {
7557 case offsetof(struct xdp_md, data):
7558 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7559 si->dst_reg, si->src_reg,
7560 offsetof(struct xdp_buff, data));
7561 break;
7562 case offsetof(struct xdp_md, data_meta):
7563 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7564 si->dst_reg, si->src_reg,
7565 offsetof(struct xdp_buff, data_meta));
7566 break;
7567 case offsetof(struct xdp_md, data_end):
7568 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7569 si->dst_reg, si->src_reg,
7570 offsetof(struct xdp_buff, data_end));
7571 break;
7572 case offsetof(struct xdp_md, ingress_ifindex):
7573 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7574 si->dst_reg, si->src_reg,
7575 offsetof(struct xdp_buff, rxq));
7576 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7577 si->dst_reg, si->dst_reg,
7578 offsetof(struct xdp_rxq_info, dev));
7579 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7580 offsetof(struct net_device, ifindex));
7581 break;
7582 case offsetof(struct xdp_md, rx_queue_index):
7583 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7584 si->dst_reg, si->src_reg,
7585 offsetof(struct xdp_buff, rxq));
7586 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7587 offsetof(struct xdp_rxq_info,
7588 queue_index));
7589 break;
7590 }
7591
7592 return insn - insn_buf;
7593}
7594
7595
7596
7597
7598
7599
7600
7601
7602
7603
7604
7605#define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7606 do { \
7607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7608 si->src_reg, offsetof(S, F)); \
7609 *insn++ = BPF_LDX_MEM( \
7610 SIZE, si->dst_reg, si->dst_reg, \
7611 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7612 target_size) \
7613 + OFF); \
7614 } while (0)
7615
7616#define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7617 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7618 BPF_FIELD_SIZEOF(NS, NF), 0)
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633#define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
7634 do { \
7635 int tmp_reg = BPF_REG_9; \
7636 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7637 --tmp_reg; \
7638 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7639 --tmp_reg; \
7640 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7641 offsetof(S, TF)); \
7642 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7643 si->dst_reg, offsetof(S, F)); \
7644 *insn++ = BPF_STX_MEM( \
7645 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
7646 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7647 target_size) \
7648 + OFF); \
7649 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7650 offsetof(S, TF)); \
7651 } while (0)
7652
7653#define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7654 TF) \
7655 do { \
7656 if (type == BPF_WRITE) { \
7657 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
7658 TF); \
7659 } else { \
7660 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7661 S, NS, F, NF, SIZE, OFF); \
7662 } \
7663 } while (0)
7664
7665#define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7666 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7667 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7668
7669static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7670 const struct bpf_insn *si,
7671 struct bpf_insn *insn_buf,
7672 struct bpf_prog *prog, u32 *target_size)
7673{
7674 struct bpf_insn *insn = insn_buf;
7675 int off;
7676
7677 switch (si->off) {
7678 case offsetof(struct bpf_sock_addr, user_family):
7679 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7680 struct sockaddr, uaddr, sa_family);
7681 break;
7682
7683 case offsetof(struct bpf_sock_addr, user_ip4):
7684 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7685 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7686 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7687 break;
7688
7689 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7690 off = si->off;
7691 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7692 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7693 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7694 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7695 tmp_reg);
7696 break;
7697
7698 case offsetof(struct bpf_sock_addr, user_port):
7699
7700
7701
7702
7703
7704
7705
7706 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7707 offsetof(struct sockaddr_in6, sin6_port));
7708 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7709 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7710 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7711 struct sockaddr_in6, uaddr,
7712 sin6_port, tmp_reg);
7713 break;
7714
7715 case offsetof(struct bpf_sock_addr, family):
7716 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7717 struct sock, sk, sk_family);
7718 break;
7719
7720 case offsetof(struct bpf_sock_addr, type):
7721 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7722 struct bpf_sock_addr_kern, struct sock, sk,
7723 __sk_flags_offset, BPF_W, 0);
7724 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7725 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7726 break;
7727
7728 case offsetof(struct bpf_sock_addr, protocol):
7729 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7730 struct bpf_sock_addr_kern, struct sock, sk,
7731 __sk_flags_offset, BPF_W, 0);
7732 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7733 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7734 SK_FL_PROTO_SHIFT);
7735 break;
7736
7737 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7738
7739 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7740 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7741 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7742 break;
7743
7744 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7745 msg_src_ip6[3]):
7746 off = si->off;
7747 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7748
7749 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7750 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7751 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7752 break;
7753 }
7754
7755 return insn - insn_buf;
7756}
7757
7758static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7759 const struct bpf_insn *si,
7760 struct bpf_insn *insn_buf,
7761 struct bpf_prog *prog,
7762 u32 *target_size)
7763{
7764 struct bpf_insn *insn = insn_buf;
7765 int off;
7766
7767
7768#define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7769 do { \
7770 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7771 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7773 struct bpf_sock_ops_kern, \
7774 is_fullsock), \
7775 si->dst_reg, si->src_reg, \
7776 offsetof(struct bpf_sock_ops_kern, \
7777 is_fullsock)); \
7778 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7779 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7780 struct bpf_sock_ops_kern, sk),\
7781 si->dst_reg, si->src_reg, \
7782 offsetof(struct bpf_sock_ops_kern, sk));\
7783 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7784 OBJ_FIELD), \
7785 si->dst_reg, si->dst_reg, \
7786 offsetof(OBJ, OBJ_FIELD)); \
7787 } while (0)
7788
7789#define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7790 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7791
7792
7793
7794
7795
7796
7797
7798
7799
7800
7801#define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7802 do { \
7803 int reg = BPF_REG_9; \
7804 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7805 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7806 if (si->dst_reg == reg || si->src_reg == reg) \
7807 reg--; \
7808 if (si->dst_reg == reg || si->src_reg == reg) \
7809 reg--; \
7810 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7811 offsetof(struct bpf_sock_ops_kern, \
7812 temp)); \
7813 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7814 struct bpf_sock_ops_kern, \
7815 is_fullsock), \
7816 reg, si->dst_reg, \
7817 offsetof(struct bpf_sock_ops_kern, \
7818 is_fullsock)); \
7819 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7820 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7821 struct bpf_sock_ops_kern, sk),\
7822 reg, si->dst_reg, \
7823 offsetof(struct bpf_sock_ops_kern, sk));\
7824 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7825 reg, si->src_reg, \
7826 offsetof(OBJ, OBJ_FIELD)); \
7827 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7828 offsetof(struct bpf_sock_ops_kern, \
7829 temp)); \
7830 } while (0)
7831
7832#define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7833 do { \
7834 if (TYPE == BPF_WRITE) \
7835 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7836 else \
7837 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7838 } while (0)
7839
7840 CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_sock_ops,
7841 SOCK_OPS_GET_TCP_SOCK_FIELD);
7842
7843 if (insn > insn_buf)
7844 return insn - insn_buf;
7845
7846 switch (si->off) {
7847 case offsetof(struct bpf_sock_ops, op) ...
7848 offsetof(struct bpf_sock_ops, replylong[3]):
7849 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7850 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7851 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7852 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7853 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7854 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7855 off = si->off;
7856 off -= offsetof(struct bpf_sock_ops, op);
7857 off += offsetof(struct bpf_sock_ops_kern, op);
7858 if (type == BPF_WRITE)
7859 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7860 off);
7861 else
7862 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7863 off);
7864 break;
7865
7866 case offsetof(struct bpf_sock_ops, family):
7867 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7868
7869 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7870 struct bpf_sock_ops_kern, sk),
7871 si->dst_reg, si->src_reg,
7872 offsetof(struct bpf_sock_ops_kern, sk));
7873 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7874 offsetof(struct sock_common, skc_family));
7875 break;
7876
7877 case offsetof(struct bpf_sock_ops, remote_ip4):
7878 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7879
7880 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7881 struct bpf_sock_ops_kern, sk),
7882 si->dst_reg, si->src_reg,
7883 offsetof(struct bpf_sock_ops_kern, sk));
7884 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7885 offsetof(struct sock_common, skc_daddr));
7886 break;
7887
7888 case offsetof(struct bpf_sock_ops, local_ip4):
7889 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7890 skc_rcv_saddr) != 4);
7891
7892 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7893 struct bpf_sock_ops_kern, sk),
7894 si->dst_reg, si->src_reg,
7895 offsetof(struct bpf_sock_ops_kern, sk));
7896 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7897 offsetof(struct sock_common,
7898 skc_rcv_saddr));
7899 break;
7900
7901 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
7902 offsetof(struct bpf_sock_ops, remote_ip6[3]):
7903#if IS_ENABLED(CONFIG_IPV6)
7904 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7905 skc_v6_daddr.s6_addr32[0]) != 4);
7906
7907 off = si->off;
7908 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
7909 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7910 struct bpf_sock_ops_kern, sk),
7911 si->dst_reg, si->src_reg,
7912 offsetof(struct bpf_sock_ops_kern, sk));
7913 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7914 offsetof(struct sock_common,
7915 skc_v6_daddr.s6_addr32[0]) +
7916 off);
7917#else
7918 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7919#endif
7920 break;
7921
7922 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
7923 offsetof(struct bpf_sock_ops, local_ip6[3]):
7924#if IS_ENABLED(CONFIG_IPV6)
7925 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7926 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7927
7928 off = si->off;
7929 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
7930 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7931 struct bpf_sock_ops_kern, sk),
7932 si->dst_reg, si->src_reg,
7933 offsetof(struct bpf_sock_ops_kern, sk));
7934 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7935 offsetof(struct sock_common,
7936 skc_v6_rcv_saddr.s6_addr32[0]) +
7937 off);
7938#else
7939 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7940#endif
7941 break;
7942
7943 case offsetof(struct bpf_sock_ops, remote_port):
7944 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7945
7946 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7947 struct bpf_sock_ops_kern, sk),
7948 si->dst_reg, si->src_reg,
7949 offsetof(struct bpf_sock_ops_kern, sk));
7950 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7951 offsetof(struct sock_common, skc_dport));
7952#ifndef __BIG_ENDIAN_BITFIELD
7953 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7954#endif
7955 break;
7956
7957 case offsetof(struct bpf_sock_ops, local_port):
7958 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7959
7960 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7961 struct bpf_sock_ops_kern, sk),
7962 si->dst_reg, si->src_reg,
7963 offsetof(struct bpf_sock_ops_kern, sk));
7964 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7965 offsetof(struct sock_common, skc_num));
7966 break;
7967
7968 case offsetof(struct bpf_sock_ops, is_fullsock):
7969 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7970 struct bpf_sock_ops_kern,
7971 is_fullsock),
7972 si->dst_reg, si->src_reg,
7973 offsetof(struct bpf_sock_ops_kern,
7974 is_fullsock));
7975 break;
7976
7977 case offsetof(struct bpf_sock_ops, state):
7978 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
7979
7980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7981 struct bpf_sock_ops_kern, sk),
7982 si->dst_reg, si->src_reg,
7983 offsetof(struct bpf_sock_ops_kern, sk));
7984 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
7985 offsetof(struct sock_common, skc_state));
7986 break;
7987
7988 case offsetof(struct bpf_sock_ops, rtt_min):
7989 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
7990 sizeof(struct minmax));
7991 BUILD_BUG_ON(sizeof(struct minmax) <
7992 sizeof(struct minmax_sample));
7993
7994 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7995 struct bpf_sock_ops_kern, sk),
7996 si->dst_reg, si->src_reg,
7997 offsetof(struct bpf_sock_ops_kern, sk));
7998 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7999 offsetof(struct tcp_sock, rtt_min) +
8000 FIELD_SIZEOF(struct minmax_sample, t));
8001 break;
8002
8003 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8004 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8005 struct tcp_sock);
8006 break;
8007
8008 case offsetof(struct bpf_sock_ops, sk_txhash):
8009 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8010 struct sock, type);
8011 break;
8012 }
8013 return insn - insn_buf;
8014}
8015
8016static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8017 const struct bpf_insn *si,
8018 struct bpf_insn *insn_buf,
8019 struct bpf_prog *prog, u32 *target_size)
8020{
8021 struct bpf_insn *insn = insn_buf;
8022 int off;
8023
8024 switch (si->off) {
8025 case offsetof(struct __sk_buff, data_end):
8026 off = si->off;
8027 off -= offsetof(struct __sk_buff, data_end);
8028 off += offsetof(struct sk_buff, cb);
8029 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8030 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8031 si->src_reg, off);
8032 break;
8033 default:
8034 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8035 target_size);
8036 }
8037
8038 return insn - insn_buf;
8039}
8040
8041static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8042 const struct bpf_insn *si,
8043 struct bpf_insn *insn_buf,
8044 struct bpf_prog *prog, u32 *target_size)
8045{
8046 struct bpf_insn *insn = insn_buf;
8047#if IS_ENABLED(CONFIG_IPV6)
8048 int off;
8049#endif
8050
8051
8052 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8053
8054 switch (si->off) {
8055 case offsetof(struct sk_msg_md, data):
8056 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8057 si->dst_reg, si->src_reg,
8058 offsetof(struct sk_msg, data));
8059 break;
8060 case offsetof(struct sk_msg_md, data_end):
8061 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8062 si->dst_reg, si->src_reg,
8063 offsetof(struct sk_msg, data_end));
8064 break;
8065 case offsetof(struct sk_msg_md, family):
8066 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
8067
8068 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8069 struct sk_msg, sk),
8070 si->dst_reg, si->src_reg,
8071 offsetof(struct sk_msg, sk));
8072 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8073 offsetof(struct sock_common, skc_family));
8074 break;
8075
8076 case offsetof(struct sk_msg_md, remote_ip4):
8077 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8078
8079 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8080 struct sk_msg, sk),
8081 si->dst_reg, si->src_reg,
8082 offsetof(struct sk_msg, sk));
8083 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8084 offsetof(struct sock_common, skc_daddr));
8085 break;
8086
8087 case offsetof(struct sk_msg_md, local_ip4):
8088 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8089 skc_rcv_saddr) != 4);
8090
8091 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8092 struct sk_msg, sk),
8093 si->dst_reg, si->src_reg,
8094 offsetof(struct sk_msg, sk));
8095 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8096 offsetof(struct sock_common,
8097 skc_rcv_saddr));
8098 break;
8099
8100 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8101 offsetof(struct sk_msg_md, remote_ip6[3]):
8102#if IS_ENABLED(CONFIG_IPV6)
8103 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8104 skc_v6_daddr.s6_addr32[0]) != 4);
8105
8106 off = si->off;
8107 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8108 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8109 struct sk_msg, sk),
8110 si->dst_reg, si->src_reg,
8111 offsetof(struct sk_msg, sk));
8112 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8113 offsetof(struct sock_common,
8114 skc_v6_daddr.s6_addr32[0]) +
8115 off);
8116#else
8117 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8118#endif
8119 break;
8120
8121 case offsetof(struct sk_msg_md, local_ip6[0]) ...
8122 offsetof(struct sk_msg_md, local_ip6[3]):
8123#if IS_ENABLED(CONFIG_IPV6)
8124 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8125 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8126
8127 off = si->off;
8128 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8130 struct sk_msg, sk),
8131 si->dst_reg, si->src_reg,
8132 offsetof(struct sk_msg, sk));
8133 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8134 offsetof(struct sock_common,
8135 skc_v6_rcv_saddr.s6_addr32[0]) +
8136 off);
8137#else
8138 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8139#endif
8140 break;
8141
8142 case offsetof(struct sk_msg_md, remote_port):
8143 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8144
8145 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8146 struct sk_msg, sk),
8147 si->dst_reg, si->src_reg,
8148 offsetof(struct sk_msg, sk));
8149 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8150 offsetof(struct sock_common, skc_dport));
8151#ifndef __BIG_ENDIAN_BITFIELD
8152 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8153#endif
8154 break;
8155
8156 case offsetof(struct sk_msg_md, local_port):
8157 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8158
8159 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8160 struct sk_msg, sk),
8161 si->dst_reg, si->src_reg,
8162 offsetof(struct sk_msg, sk));
8163 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8164 offsetof(struct sock_common, skc_num));
8165 break;
8166
8167 case offsetof(struct sk_msg_md, size):
8168 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8169 si->dst_reg, si->src_reg,
8170 offsetof(struct sk_msg_sg, size));
8171 break;
8172 }
8173
8174 return insn - insn_buf;
8175}
8176
8177const struct bpf_verifier_ops sk_filter_verifier_ops = {
8178 .get_func_proto = sk_filter_func_proto,
8179 .is_valid_access = sk_filter_is_valid_access,
8180 .convert_ctx_access = bpf_convert_ctx_access,
8181 .gen_ld_abs = bpf_gen_ld_abs,
8182};
8183
8184const struct bpf_prog_ops sk_filter_prog_ops = {
8185 .test_run = bpf_prog_test_run_skb,
8186};
8187
8188const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8189 .get_func_proto = tc_cls_act_func_proto,
8190 .is_valid_access = tc_cls_act_is_valid_access,
8191 .convert_ctx_access = tc_cls_act_convert_ctx_access,
8192 .gen_prologue = tc_cls_act_prologue,
8193 .gen_ld_abs = bpf_gen_ld_abs,
8194};
8195
8196const struct bpf_prog_ops tc_cls_act_prog_ops = {
8197 .test_run = bpf_prog_test_run_skb,
8198};
8199
8200const struct bpf_verifier_ops xdp_verifier_ops = {
8201 .get_func_proto = xdp_func_proto,
8202 .is_valid_access = xdp_is_valid_access,
8203 .convert_ctx_access = xdp_convert_ctx_access,
8204 .gen_prologue = bpf_noop_prologue,
8205};
8206
8207const struct bpf_prog_ops xdp_prog_ops = {
8208 .test_run = bpf_prog_test_run_xdp,
8209};
8210
8211const struct bpf_verifier_ops cg_skb_verifier_ops = {
8212 .get_func_proto = cg_skb_func_proto,
8213 .is_valid_access = cg_skb_is_valid_access,
8214 .convert_ctx_access = bpf_convert_ctx_access,
8215};
8216
8217const struct bpf_prog_ops cg_skb_prog_ops = {
8218 .test_run = bpf_prog_test_run_skb,
8219};
8220
8221const struct bpf_verifier_ops lwt_in_verifier_ops = {
8222 .get_func_proto = lwt_in_func_proto,
8223 .is_valid_access = lwt_is_valid_access,
8224 .convert_ctx_access = bpf_convert_ctx_access,
8225};
8226
8227const struct bpf_prog_ops lwt_in_prog_ops = {
8228 .test_run = bpf_prog_test_run_skb,
8229};
8230
8231const struct bpf_verifier_ops lwt_out_verifier_ops = {
8232 .get_func_proto = lwt_out_func_proto,
8233 .is_valid_access = lwt_is_valid_access,
8234 .convert_ctx_access = bpf_convert_ctx_access,
8235};
8236
8237const struct bpf_prog_ops lwt_out_prog_ops = {
8238 .test_run = bpf_prog_test_run_skb,
8239};
8240
8241const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8242 .get_func_proto = lwt_xmit_func_proto,
8243 .is_valid_access = lwt_is_valid_access,
8244 .convert_ctx_access = bpf_convert_ctx_access,
8245 .gen_prologue = tc_cls_act_prologue,
8246};
8247
8248const struct bpf_prog_ops lwt_xmit_prog_ops = {
8249 .test_run = bpf_prog_test_run_skb,
8250};
8251
8252const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8253 .get_func_proto = lwt_seg6local_func_proto,
8254 .is_valid_access = lwt_is_valid_access,
8255 .convert_ctx_access = bpf_convert_ctx_access,
8256};
8257
8258const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8259 .test_run = bpf_prog_test_run_skb,
8260};
8261
8262const struct bpf_verifier_ops cg_sock_verifier_ops = {
8263 .get_func_proto = sock_filter_func_proto,
8264 .is_valid_access = sock_filter_is_valid_access,
8265 .convert_ctx_access = bpf_sock_convert_ctx_access,
8266};
8267
8268const struct bpf_prog_ops cg_sock_prog_ops = {
8269};
8270
8271const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8272 .get_func_proto = sock_addr_func_proto,
8273 .is_valid_access = sock_addr_is_valid_access,
8274 .convert_ctx_access = sock_addr_convert_ctx_access,
8275};
8276
8277const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8278};
8279
8280const struct bpf_verifier_ops sock_ops_verifier_ops = {
8281 .get_func_proto = sock_ops_func_proto,
8282 .is_valid_access = sock_ops_is_valid_access,
8283 .convert_ctx_access = sock_ops_convert_ctx_access,
8284};
8285
8286const struct bpf_prog_ops sock_ops_prog_ops = {
8287};
8288
8289const struct bpf_verifier_ops sk_skb_verifier_ops = {
8290 .get_func_proto = sk_skb_func_proto,
8291 .is_valid_access = sk_skb_is_valid_access,
8292 .convert_ctx_access = sk_skb_convert_ctx_access,
8293 .gen_prologue = sk_skb_prologue,
8294};
8295
8296const struct bpf_prog_ops sk_skb_prog_ops = {
8297};
8298
8299const struct bpf_verifier_ops sk_msg_verifier_ops = {
8300 .get_func_proto = sk_msg_func_proto,
8301 .is_valid_access = sk_msg_is_valid_access,
8302 .convert_ctx_access = sk_msg_convert_ctx_access,
8303 .gen_prologue = bpf_noop_prologue,
8304};
8305
8306const struct bpf_prog_ops sk_msg_prog_ops = {
8307};
8308
8309const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8310 .get_func_proto = flow_dissector_func_proto,
8311 .is_valid_access = flow_dissector_is_valid_access,
8312 .convert_ctx_access = flow_dissector_convert_ctx_access,
8313};
8314
8315const struct bpf_prog_ops flow_dissector_prog_ops = {
8316 .test_run = bpf_prog_test_run_flow_dissector,
8317};
8318
8319int sk_detach_filter(struct sock *sk)
8320{
8321 int ret = -ENOENT;
8322 struct sk_filter *filter;
8323
8324 if (sock_flag(sk, SOCK_FILTER_LOCKED))
8325 return -EPERM;
8326
8327 filter = rcu_dereference_protected(sk->sk_filter,
8328 lockdep_sock_is_held(sk));
8329 if (filter) {
8330 RCU_INIT_POINTER(sk->sk_filter, NULL);
8331 sk_filter_uncharge(sk, filter);
8332 ret = 0;
8333 }
8334
8335 return ret;
8336}
8337EXPORT_SYMBOL_GPL(sk_detach_filter);
8338
8339int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8340 unsigned int len)
8341{
8342 struct sock_fprog_kern *fprog;
8343 struct sk_filter *filter;
8344 int ret = 0;
8345
8346 lock_sock(sk);
8347 filter = rcu_dereference_protected(sk->sk_filter,
8348 lockdep_sock_is_held(sk));
8349 if (!filter)
8350 goto out;
8351
8352
8353
8354
8355
8356 ret = -EACCES;
8357 fprog = filter->prog->orig_prog;
8358 if (!fprog)
8359 goto out;
8360
8361 ret = fprog->len;
8362 if (!len)
8363
8364 goto out;
8365
8366 ret = -EINVAL;
8367 if (len < fprog->len)
8368 goto out;
8369
8370 ret = -EFAULT;
8371 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8372 goto out;
8373
8374
8375
8376
8377 ret = fprog->len;
8378out:
8379 release_sock(sk);
8380 return ret;
8381}
8382
8383#ifdef CONFIG_INET
8384struct sk_reuseport_kern {
8385 struct sk_buff *skb;
8386 struct sock *sk;
8387 struct sock *selected_sk;
8388 void *data_end;
8389 u32 hash;
8390 u32 reuseport_id;
8391 bool bind_inany;
8392};
8393
8394static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8395 struct sock_reuseport *reuse,
8396 struct sock *sk, struct sk_buff *skb,
8397 u32 hash)
8398{
8399 reuse_kern->skb = skb;
8400 reuse_kern->sk = sk;
8401 reuse_kern->selected_sk = NULL;
8402 reuse_kern->data_end = skb->data + skb_headlen(skb);
8403 reuse_kern->hash = hash;
8404 reuse_kern->reuseport_id = reuse->reuseport_id;
8405 reuse_kern->bind_inany = reuse->bind_inany;
8406}
8407
8408struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8409 struct bpf_prog *prog, struct sk_buff *skb,
8410 u32 hash)
8411{
8412 struct sk_reuseport_kern reuse_kern;
8413 enum sk_action action;
8414
8415 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8416 action = BPF_PROG_RUN(prog, &reuse_kern);
8417
8418 if (action == SK_PASS)
8419 return reuse_kern.selected_sk;
8420 else
8421 return ERR_PTR(-ECONNREFUSED);
8422}
8423
8424BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8425 struct bpf_map *, map, void *, key, u32, flags)
8426{
8427 struct sock_reuseport *reuse;
8428 struct sock *selected_sk;
8429
8430 selected_sk = map->ops->map_lookup_elem(map, key);
8431 if (!selected_sk)
8432 return -ENOENT;
8433
8434 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8435 if (!reuse)
8436
8437
8438
8439
8440 return -ENOENT;
8441
8442 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8443 struct sock *sk;
8444
8445 if (unlikely(!reuse_kern->reuseport_id))
8446
8447
8448
8449
8450
8451
8452 return -ENOENT;
8453
8454 sk = reuse_kern->sk;
8455 if (sk->sk_protocol != selected_sk->sk_protocol)
8456 return -EPROTOTYPE;
8457 else if (sk->sk_family != selected_sk->sk_family)
8458 return -EAFNOSUPPORT;
8459
8460
8461 return -EBADFD;
8462 }
8463
8464 reuse_kern->selected_sk = selected_sk;
8465
8466 return 0;
8467}
8468
8469static const struct bpf_func_proto sk_select_reuseport_proto = {
8470 .func = sk_select_reuseport,
8471 .gpl_only = false,
8472 .ret_type = RET_INTEGER,
8473 .arg1_type = ARG_PTR_TO_CTX,
8474 .arg2_type = ARG_CONST_MAP_PTR,
8475 .arg3_type = ARG_PTR_TO_MAP_KEY,
8476 .arg4_type = ARG_ANYTHING,
8477};
8478
8479BPF_CALL_4(sk_reuseport_load_bytes,
8480 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8481 void *, to, u32, len)
8482{
8483 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8484}
8485
8486static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8487 .func = sk_reuseport_load_bytes,
8488 .gpl_only = false,
8489 .ret_type = RET_INTEGER,
8490 .arg1_type = ARG_PTR_TO_CTX,
8491 .arg2_type = ARG_ANYTHING,
8492 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8493 .arg4_type = ARG_CONST_SIZE,
8494};
8495
8496BPF_CALL_5(sk_reuseport_load_bytes_relative,
8497 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8498 void *, to, u32, len, u32, start_header)
8499{
8500 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8501 len, start_header);
8502}
8503
8504static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8505 .func = sk_reuseport_load_bytes_relative,
8506 .gpl_only = false,
8507 .ret_type = RET_INTEGER,
8508 .arg1_type = ARG_PTR_TO_CTX,
8509 .arg2_type = ARG_ANYTHING,
8510 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8511 .arg4_type = ARG_CONST_SIZE,
8512 .arg5_type = ARG_ANYTHING,
8513};
8514
8515static const struct bpf_func_proto *
8516sk_reuseport_func_proto(enum bpf_func_id func_id,
8517 const struct bpf_prog *prog)
8518{
8519 switch (func_id) {
8520 case BPF_FUNC_sk_select_reuseport:
8521 return &sk_select_reuseport_proto;
8522 case BPF_FUNC_skb_load_bytes:
8523 return &sk_reuseport_load_bytes_proto;
8524 case BPF_FUNC_skb_load_bytes_relative:
8525 return &sk_reuseport_load_bytes_relative_proto;
8526 default:
8527 return bpf_base_func_proto(func_id);
8528 }
8529}
8530
8531static bool
8532sk_reuseport_is_valid_access(int off, int size,
8533 enum bpf_access_type type,
8534 const struct bpf_prog *prog,
8535 struct bpf_insn_access_aux *info)
8536{
8537 const u32 size_default = sizeof(__u32);
8538
8539 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8540 off % size || type != BPF_READ)
8541 return false;
8542
8543 switch (off) {
8544 case offsetof(struct sk_reuseport_md, data):
8545 info->reg_type = PTR_TO_PACKET;
8546 return size == sizeof(__u64);
8547
8548 case offsetof(struct sk_reuseport_md, data_end):
8549 info->reg_type = PTR_TO_PACKET_END;
8550 return size == sizeof(__u64);
8551
8552 case offsetof(struct sk_reuseport_md, hash):
8553 return size == size_default;
8554
8555
8556 case offsetof(struct sk_reuseport_md, eth_protocol):
8557 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8558 return false;
8559
8560 case offsetof(struct sk_reuseport_md, ip_protocol):
8561 case offsetof(struct sk_reuseport_md, bind_inany):
8562 case offsetof(struct sk_reuseport_md, len):
8563 bpf_ctx_record_field_size(info, size_default);
8564 return bpf_ctx_narrow_access_ok(off, size, size_default);
8565
8566 default:
8567 return false;
8568 }
8569}
8570
8571#define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8572 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8573 si->dst_reg, si->src_reg, \
8574 bpf_target_off(struct sk_reuseport_kern, F, \
8575 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8576 target_size)); \
8577 })
8578
8579#define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8580 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8581 struct sk_buff, \
8582 skb, \
8583 SKB_FIELD)
8584
8585#define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8586 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
8587 struct sock, \
8588 sk, \
8589 SK_FIELD, BPF_SIZE, EXTRA_OFF)
8590
8591static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8592 const struct bpf_insn *si,
8593 struct bpf_insn *insn_buf,
8594 struct bpf_prog *prog,
8595 u32 *target_size)
8596{
8597 struct bpf_insn *insn = insn_buf;
8598
8599 switch (si->off) {
8600 case offsetof(struct sk_reuseport_md, data):
8601 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8602 break;
8603
8604 case offsetof(struct sk_reuseport_md, len):
8605 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8606 break;
8607
8608 case offsetof(struct sk_reuseport_md, eth_protocol):
8609 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8610 break;
8611
8612 case offsetof(struct sk_reuseport_md, ip_protocol):
8613 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8614 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8615 BPF_W, 0);
8616 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8617 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8618 SK_FL_PROTO_SHIFT);
8619
8620
8621
8622 *target_size = 1;
8623 break;
8624
8625 case offsetof(struct sk_reuseport_md, data_end):
8626 SK_REUSEPORT_LOAD_FIELD(data_end);
8627 break;
8628
8629 case offsetof(struct sk_reuseport_md, hash):
8630 SK_REUSEPORT_LOAD_FIELD(hash);
8631 break;
8632
8633 case offsetof(struct sk_reuseport_md, bind_inany):
8634 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8635 break;
8636 }
8637
8638 return insn - insn_buf;
8639}
8640
8641const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8642 .get_func_proto = sk_reuseport_func_proto,
8643 .is_valid_access = sk_reuseport_is_valid_access,
8644 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8645};
8646
8647const struct bpf_prog_ops sk_reuseport_prog_ops = {
8648};
8649#endif
8650