]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - net/core/filter.c
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:


46d713d8c2ad6e6cc2c9095cb62f44ef7c1f3fec
[mirror_ubuntu-eoan-kernel.git] / net / core / filter.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/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/nexthop.h>
66 #include <net/flow.h>
67 #include <net/arp.h>
68 #include <net/ipv6.h>
69 #include <net/net_namespace.h>
70 #include <linux/seg6_local.h>
71 #include <net/seg6.h>
72 #include <net/seg6_local.h>
73 #include <net/lwtunnel.h>
74 #include <net/ipv6_stubs.h>
75 #include <net/bpf_sk_storage.h>
76
77 /**
78 * sk_filter_trim_cap - run a packet through a socket filter
79 * @sk: sock associated with &sk_buff
80 * @skb: buffer to filter
81 * @cap: limit on how short the eBPF program may trim the packet
82 *
83 * Run the eBPF program and then cut skb->data to correct size returned by
84 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
85 * than pkt_len we keep whole skb->data. This is the socket level
86 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
87 * be accepted or -EPERM if the packet should be tossed.
88 *
89 */
90 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
91 {
92 int err;
93 struct sk_filter *filter;
94
95 /*
96 * If the skb was allocated from pfmemalloc reserves, only
97 * allow SOCK_MEMALLOC sockets to use it as this socket is
98 * helping free memory
99 */
100 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
101 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
102 return -ENOMEM;
103 }
104 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
105 if (err)
106 return err;
107
108 err = security_sock_rcv_skb(sk, skb);
109 if (err)
110 return err;
111
112 rcu_read_lock();
113 filter = rcu_dereference(sk->sk_filter);
114 if (filter) {
115 struct sock *save_sk = skb->sk;
116 unsigned int pkt_len;
117
118 skb->sk = sk;
119 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
120 skb->sk = save_sk;
121 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
122 }
123 rcu_read_unlock();
124
125 return err;
126 }
127 EXPORT_SYMBOL(sk_filter_trim_cap);
128
129 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
130 {
131 return skb_get_poff(skb);
132 }
133
134 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
135 {
136 struct nlattr *nla;
137
138 if (skb_is_nonlinear(skb))
139 return 0;
140
141 if (skb->len < sizeof(struct nlattr))
142 return 0;
143
144 if (a > skb->len - sizeof(struct nlattr))
145 return 0;
146
147 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
148 if (nla)
149 return (void *) nla - (void *) skb->data;
150
151 return 0;
152 }
153
154 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
155 {
156 struct nlattr *nla;
157
158 if (skb_is_nonlinear(skb))
159 return 0;
160
161 if (skb->len < sizeof(struct nlattr))
162 return 0;
163
164 if (a > skb->len - sizeof(struct nlattr))
165 return 0;
166
167 nla = (struct nlattr *) &skb->data[a];
168 if (nla->nla_len > skb->len - a)
169 return 0;
170
171 nla = nla_find_nested(nla, x);
172 if (nla)
173 return (void *) nla - (void *) skb->data;
174
175 return 0;
176 }
177
178 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
179 data, int, headlen, int, offset)
180 {
181 u8 tmp, *ptr;
182 const int len = sizeof(tmp);
183
184 if (offset >= 0) {
185 if (headlen - offset >= len)
186 return *(u8 *)(data + offset);
187 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
188 return tmp;
189 } else {
190 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
191 if (likely(ptr))
192 return *(u8 *)ptr;
193 }
194
195 return -EFAULT;
196 }
197
198 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
199 int, offset)
200 {
201 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
202 offset);
203 }
204
205 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
206 data, int, headlen, int, offset)
207 {
208 u16 tmp, *ptr;
209 const int len = sizeof(tmp);
210
211 if (offset >= 0) {
212 if (headlen - offset >= len)
213 return get_unaligned_be16(data + offset);
214 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
215 return be16_to_cpu(tmp);
216 } else {
217 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
218 if (likely(ptr))
219 return get_unaligned_be16(ptr);
220 }
221
222 return -EFAULT;
223 }
224
225 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
226 int, offset)
227 {
228 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
229 offset);
230 }
231
232 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
233 data, int, headlen, int, offset)
234 {
235 u32 tmp, *ptr;
236 const int len = sizeof(tmp);
237
238 if (likely(offset >= 0)) {
239 if (headlen - offset >= len)
240 return get_unaligned_be32(data + offset);
241 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
242 return be32_to_cpu(tmp);
243 } else {
244 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
245 if (likely(ptr))
246 return get_unaligned_be32(ptr);
247 }
248
249 return -EFAULT;
250 }
251
252 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
253 int, offset)
254 {
255 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
256 offset);
257 }
258
259 BPF_CALL_0(bpf_get_raw_cpu_id)
260 {
261 return raw_smp_processor_id();
262 }
263
264 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
265 .func = bpf_get_raw_cpu_id,
266 .gpl_only = false,
267 .ret_type = RET_INTEGER,
268 };
269
270 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
271 struct bpf_insn *insn_buf)
272 {
273 struct bpf_insn *insn = insn_buf;
274
275 switch (skb_field) {
276 case SKF_AD_MARK:
277 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
278
279 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
280 offsetof(struct sk_buff, mark));
281 break;
282
283 case SKF_AD_PKTTYPE:
284 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
285 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
286 #ifdef __BIG_ENDIAN_BITFIELD
287 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
288 #endif
289 break;
290
291 case SKF_AD_QUEUE:
292 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
293
294 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
295 offsetof(struct sk_buff, queue_mapping));
296 break;
297
298 case SKF_AD_VLAN_TAG:
299 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
300
301 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
302 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
303 offsetof(struct sk_buff, vlan_tci));
304 break;
305 case SKF_AD_VLAN_TAG_PRESENT:
306 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
307 if (PKT_VLAN_PRESENT_BIT)
308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
309 if (PKT_VLAN_PRESENT_BIT < 7)
310 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
311 break;
312 }
313
314 return insn - insn_buf;
315 }
316
317 static bool convert_bpf_extensions(struct sock_filter *fp,
318 struct bpf_insn **insnp)
319 {
320 struct bpf_insn *insn = *insnp;
321 u32 cnt;
322
323 switch (fp->k) {
324 case SKF_AD_OFF + SKF_AD_PROTOCOL:
325 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
326
327 /* A = *(u16 *) (CTX + offsetof(protocol)) */
328 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
329 offsetof(struct sk_buff, protocol));
330 /* A = ntohs(A) [emitting a nop or swap16] */
331 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
332 break;
333
334 case SKF_AD_OFF + SKF_AD_PKTTYPE:
335 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
336 insn += cnt - 1;
337 break;
338
339 case SKF_AD_OFF + SKF_AD_IFINDEX:
340 case SKF_AD_OFF + SKF_AD_HATYPE:
341 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
342 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
343
344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
345 BPF_REG_TMP, BPF_REG_CTX,
346 offsetof(struct sk_buff, dev));
347 /* if (tmp != 0) goto pc + 1 */
348 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
349 *insn++ = BPF_EXIT_INSN();
350 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
351 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
352 offsetof(struct net_device, ifindex));
353 else
354 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
355 offsetof(struct net_device, type));
356 break;
357
358 case SKF_AD_OFF + SKF_AD_MARK:
359 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
360 insn += cnt - 1;
361 break;
362
363 case SKF_AD_OFF + SKF_AD_RXHASH:
364 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
365
366 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
367 offsetof(struct sk_buff, hash));
368 break;
369
370 case SKF_AD_OFF + SKF_AD_QUEUE:
371 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
372 insn += cnt - 1;
373 break;
374
375 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
376 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
377 BPF_REG_A, BPF_REG_CTX, insn);
378 insn += cnt - 1;
379 break;
380
381 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
382 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
383 BPF_REG_A, BPF_REG_CTX, insn);
384 insn += cnt - 1;
385 break;
386
387 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
388 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
389
390 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
391 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
392 offsetof(struct sk_buff, vlan_proto));
393 /* A = ntohs(A) [emitting a nop or swap16] */
394 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
395 break;
396
397 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
398 case SKF_AD_OFF + SKF_AD_NLATTR:
399 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
400 case SKF_AD_OFF + SKF_AD_CPU:
401 case SKF_AD_OFF + SKF_AD_RANDOM:
402 /* arg1 = CTX */
403 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
404 /* arg2 = A */
405 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
406 /* arg3 = X */
407 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
408 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
409 switch (fp->k) {
410 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
411 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
412 break;
413 case SKF_AD_OFF + SKF_AD_NLATTR:
414 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
415 break;
416 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
417 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
418 break;
419 case SKF_AD_OFF + SKF_AD_CPU:
420 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
421 break;
422 case SKF_AD_OFF + SKF_AD_RANDOM:
423 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
424 bpf_user_rnd_init_once();
425 break;
426 }
427 break;
428
429 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
430 /* A ^= X */
431 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
432 break;
433
434 default:
435 /* This is just a dummy call to avoid letting the compiler
436 * evict __bpf_call_base() as an optimization. Placed here
437 * where no-one bothers.
438 */
439 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
440 return false;
441 }
442
443 *insnp = insn;
444 return true;
445 }
446
447 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
448 {
449 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
450 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
451 bool endian = BPF_SIZE(fp->code) == BPF_H ||
452 BPF_SIZE(fp->code) == BPF_W;
453 bool indirect = BPF_MODE(fp->code) == BPF_IND;
454 const int ip_align = NET_IP_ALIGN;
455 struct bpf_insn *insn = *insnp;
456 int offset = fp->k;
457
458 if (!indirect &&
459 ((unaligned_ok && offset >= 0) ||
460 (!unaligned_ok && offset >= 0 &&
461 offset + ip_align >= 0 &&
462 offset + ip_align % size == 0))) {
463 bool ldx_off_ok = offset <= S16_MAX;
464
465 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
466 if (offset)
467 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
468 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
469 size, 2 + endian + (!ldx_off_ok * 2));
470 if (ldx_off_ok) {
471 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
472 BPF_REG_D, offset);
473 } else {
474 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
475 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
476 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
477 BPF_REG_TMP, 0);
478 }
479 if (endian)
480 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
481 *insn++ = BPF_JMP_A(8);
482 }
483
484 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
485 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
486 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
487 if (!indirect) {
488 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
489 } else {
490 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
491 if (fp->k)
492 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
493 }
494
495 switch (BPF_SIZE(fp->code)) {
496 case BPF_B:
497 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
498 break;
499 case BPF_H:
500 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
501 break;
502 case BPF_W:
503 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
504 break;
505 default:
506 return false;
507 }
508
509 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
510 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
511 *insn = BPF_EXIT_INSN();
512
513 *insnp = insn;
514 return true;
515 }
516
517 /**
518 * bpf_convert_filter - convert filter program
519 * @prog: the user passed filter program
520 * @len: the length of the user passed filter program
521 * @new_prog: allocated 'struct bpf_prog' or NULL
522 * @new_len: pointer to store length of converted program
523 * @seen_ld_abs: bool whether we've seen ld_abs/ind
524 *
525 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
526 * style extended BPF (eBPF).
527 * Conversion workflow:
528 *
529 * 1) First pass for calculating the new program length:
530 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
531 *
532 * 2) 2nd pass to remap in two passes: 1st pass finds new
533 * jump offsets, 2nd pass remapping:
534 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
535 */
536 static int bpf_convert_filter(struct sock_filter *prog, int len,
537 struct bpf_prog *new_prog, int *new_len,
538 bool *seen_ld_abs)
539 {
540 int new_flen = 0, pass = 0, target, i, stack_off;
541 struct bpf_insn *new_insn, *first_insn = NULL;
542 struct sock_filter *fp;
543 int *addrs = NULL;
544 u8 bpf_src;
545
546 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
547 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
548
549 if (len <= 0 || len > BPF_MAXINSNS)
550 return -EINVAL;
551
552 if (new_prog) {
553 first_insn = new_prog->insnsi;
554 addrs = kcalloc(len, sizeof(*addrs),
555 GFP_KERNEL | __GFP_NOWARN);
556 if (!addrs)
557 return -ENOMEM;
558 }
559
560 do_pass:
561 new_insn = first_insn;
562 fp = prog;
563
564 /* Classic BPF related prologue emission. */
565 if (new_prog) {
566 /* Classic BPF expects A and X to be reset first. These need
567 * to be guaranteed to be the first two instructions.
568 */
569 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
570 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
571
572 /* All programs must keep CTX in callee saved BPF_REG_CTX.
573 * In eBPF case it's done by the compiler, here we need to
574 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
575 */
576 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
577 if (*seen_ld_abs) {
578 /* For packet access in classic BPF, cache skb->data
579 * in callee-saved BPF R8 and skb->len - skb->data_len
580 * (headlen) in BPF R9. Since classic BPF is read-only
581 * on CTX, we only need to cache it once.
582 */
583 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
584 BPF_REG_D, BPF_REG_CTX,
585 offsetof(struct sk_buff, data));
586 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
587 offsetof(struct sk_buff, len));
588 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
589 offsetof(struct sk_buff, data_len));
590 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
591 }
592 } else {
593 new_insn += 3;
594 }
595
596 for (i = 0; i < len; fp++, i++) {
597 struct bpf_insn tmp_insns[32] = { };
598 struct bpf_insn *insn = tmp_insns;
599
600 if (addrs)
601 addrs[i] = new_insn - first_insn;
602
603 switch (fp->code) {
604 /* All arithmetic insns and skb loads map as-is. */
605 case BPF_ALU | BPF_ADD | BPF_X:
606 case BPF_ALU | BPF_ADD | BPF_K:
607 case BPF_ALU | BPF_SUB | BPF_X:
608 case BPF_ALU | BPF_SUB | BPF_K:
609 case BPF_ALU | BPF_AND | BPF_X:
610 case BPF_ALU | BPF_AND | BPF_K:
611 case BPF_ALU | BPF_OR | BPF_X:
612 case BPF_ALU | BPF_OR | BPF_K:
613 case BPF_ALU | BPF_LSH | BPF_X:
614 case BPF_ALU | BPF_LSH | BPF_K:
615 case BPF_ALU | BPF_RSH | BPF_X:
616 case BPF_ALU | BPF_RSH | BPF_K:
617 case BPF_ALU | BPF_XOR | BPF_X:
618 case BPF_ALU | BPF_XOR | BPF_K:
619 case BPF_ALU | BPF_MUL | BPF_X:
620 case BPF_ALU | BPF_MUL | BPF_K:
621 case BPF_ALU | BPF_DIV | BPF_X:
622 case BPF_ALU | BPF_DIV | BPF_K:
623 case BPF_ALU | BPF_MOD | BPF_X:
624 case BPF_ALU | BPF_MOD | BPF_K:
625 case BPF_ALU | BPF_NEG:
626 case BPF_LD | BPF_ABS | BPF_W:
627 case BPF_LD | BPF_ABS | BPF_H:
628 case BPF_LD | BPF_ABS | BPF_B:
629 case BPF_LD | BPF_IND | BPF_W:
630 case BPF_LD | BPF_IND | BPF_H:
631 case BPF_LD | BPF_IND | BPF_B:
632 /* Check for overloaded BPF extension and
633 * directly convert it if found, otherwise
634 * just move on with mapping.
635 */
636 if (BPF_CLASS(fp->code) == BPF_LD &&
637 BPF_MODE(fp->code) == BPF_ABS &&
638 convert_bpf_extensions(fp, &insn))
639 break;
640 if (BPF_CLASS(fp->code) == BPF_LD &&
641 convert_bpf_ld_abs(fp, &insn)) {
642 *seen_ld_abs = true;
643 break;
644 }
645
646 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
647 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
648 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
649 /* Error with exception code on div/mod by 0.
650 * For cBPF programs, this was always return 0.
651 */
652 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
653 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
654 *insn++ = BPF_EXIT_INSN();
655 }
656
657 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
658 break;
659
660 /* Jump transformation cannot use BPF block macros
661 * everywhere as offset calculation and target updates
662 * require a bit more work than the rest, i.e. jump
663 * opcodes map as-is, but offsets need adjustment.
664 */
665
666 #define BPF_EMIT_JMP \
667 do { \
668 const s32 off_min = S16_MIN, off_max = S16_MAX; \
669 s32 off; \
670 \
671 if (target >= len || target < 0) \
672 goto err; \
673 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
674 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
675 off -= insn - tmp_insns; \
676 /* Reject anything not fitting into insn->off. */ \
677 if (off < off_min || off > off_max) \
678 goto err; \
679 insn->off = off; \
680 } while (0)
681
682 case BPF_JMP | BPF_JA:
683 target = i + fp->k + 1;
684 insn->code = fp->code;
685 BPF_EMIT_JMP;
686 break;
687
688 case BPF_JMP | BPF_JEQ | BPF_K:
689 case BPF_JMP | BPF_JEQ | BPF_X:
690 case BPF_JMP | BPF_JSET | BPF_K:
691 case BPF_JMP | BPF_JSET | BPF_X:
692 case BPF_JMP | BPF_JGT | BPF_K:
693 case BPF_JMP | BPF_JGT | BPF_X:
694 case BPF_JMP | BPF_JGE | BPF_K:
695 case BPF_JMP | BPF_JGE | BPF_X:
696 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
697 /* BPF immediates are signed, zero extend
698 * immediate into tmp register and use it
699 * in compare insn.
700 */
701 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
702
703 insn->dst_reg = BPF_REG_A;
704 insn->src_reg = BPF_REG_TMP;
705 bpf_src = BPF_X;
706 } else {
707 insn->dst_reg = BPF_REG_A;
708 insn->imm = fp->k;
709 bpf_src = BPF_SRC(fp->code);
710 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
711 }
712
713 /* Common case where 'jump_false' is next insn. */
714 if (fp->jf == 0) {
715 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
716 target = i + fp->jt + 1;
717 BPF_EMIT_JMP;
718 break;
719 }
720
721 /* Convert some jumps when 'jump_true' is next insn. */
722 if (fp->jt == 0) {
723 switch (BPF_OP(fp->code)) {
724 case BPF_JEQ:
725 insn->code = BPF_JMP | BPF_JNE | bpf_src;
726 break;
727 case BPF_JGT:
728 insn->code = BPF_JMP | BPF_JLE | bpf_src;
729 break;
730 case BPF_JGE:
731 insn->code = BPF_JMP | BPF_JLT | bpf_src;
732 break;
733 default:
734 goto jmp_rest;
735 }
736
737 target = i + fp->jf + 1;
738 BPF_EMIT_JMP;
739 break;
740 }
741 jmp_rest:
742 /* Other jumps are mapped into two insns: Jxx and JA. */
743 target = i + fp->jt + 1;
744 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
745 BPF_EMIT_JMP;
746 insn++;
747
748 insn->code = BPF_JMP | BPF_JA;
749 target = i + fp->jf + 1;
750 BPF_EMIT_JMP;
751 break;
752
753 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
754 case BPF_LDX | BPF_MSH | BPF_B: {
755 struct sock_filter tmp = {
756 .code = BPF_LD | BPF_ABS | BPF_B,
757 .k = fp->k,
758 };
759
760 *seen_ld_abs = true;
761
762 /* X = A */
763 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
764 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
765 convert_bpf_ld_abs(&tmp, &insn);
766 insn++;
767 /* A &= 0xf */
768 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
769 /* A <<= 2 */
770 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
771 /* tmp = X */
772 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
773 /* X = A */
774 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
775 /* A = tmp */
776 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
777 break;
778 }
779 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
780 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
781 */
782 case BPF_RET | BPF_A:
783 case BPF_RET | BPF_K:
784 if (BPF_RVAL(fp->code) == BPF_K)
785 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
786 0, fp->k);
787 *insn = BPF_EXIT_INSN();
788 break;
789
790 /* Store to stack. */
791 case BPF_ST:
792 case BPF_STX:
793 stack_off = fp->k * 4 + 4;
794 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
795 BPF_ST ? BPF_REG_A : BPF_REG_X,
796 -stack_off);
797 /* check_load_and_stores() verifies that classic BPF can
798 * load from stack only after write, so tracking
799 * stack_depth for ST|STX insns is enough
800 */
801 if (new_prog && new_prog->aux->stack_depth < stack_off)
802 new_prog->aux->stack_depth = stack_off;
803 break;
804
805 /* Load from stack. */
806 case BPF_LD | BPF_MEM:
807 case BPF_LDX | BPF_MEM:
808 stack_off = fp->k * 4 + 4;
809 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
810 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
811 -stack_off);
812 break;
813
814 /* A = K or X = K */
815 case BPF_LD | BPF_IMM:
816 case BPF_LDX | BPF_IMM:
817 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
818 BPF_REG_A : BPF_REG_X, fp->k);
819 break;
820
821 /* X = A */
822 case BPF_MISC | BPF_TAX:
823 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
824 break;
825
826 /* A = X */
827 case BPF_MISC | BPF_TXA:
828 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
829 break;
830
831 /* A = skb->len or X = skb->len */
832 case BPF_LD | BPF_W | BPF_LEN:
833 case BPF_LDX | BPF_W | BPF_LEN:
834 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
835 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
836 offsetof(struct sk_buff, len));
837 break;
838
839 /* Access seccomp_data fields. */
840 case BPF_LDX | BPF_ABS | BPF_W:
841 /* A = *(u32 *) (ctx + K) */
842 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
843 break;
844
845 /* Unknown instruction. */
846 default:
847 goto err;
848 }
849
850 insn++;
851 if (new_prog)
852 memcpy(new_insn, tmp_insns,
853 sizeof(*insn) * (insn - tmp_insns));
854 new_insn += insn - tmp_insns;
855 }
856
857 if (!new_prog) {
858 /* Only calculating new length. */
859 *new_len = new_insn - first_insn;
860 if (*seen_ld_abs)
861 *new_len += 4; /* Prologue bits. */
862 return 0;
863 }
864
865 pass++;
866 if (new_flen != new_insn - first_insn) {
867 new_flen = new_insn - first_insn;
868 if (pass > 2)
869 goto err;
870 goto do_pass;
871 }
872
873 kfree(addrs);
874 BUG_ON(*new_len != new_flen);
875 return 0;
876 err:
877 kfree(addrs);
878 return -EINVAL;
879 }
880
881 /* Security:
882 *
883 * As we dont want to clear mem[] array for each packet going through
884 * __bpf_prog_run(), we check that filter loaded by user never try to read
885 * a cell if not previously written, and we check all branches to be sure
886 * a malicious user doesn't try to abuse us.
887 */
888 static int check_load_and_stores(const struct sock_filter *filter, int flen)
889 {
890 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
891 int pc, ret = 0;
892
893 BUILD_BUG_ON(BPF_MEMWORDS > 16);
894
895 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
896 if (!masks)
897 return -ENOMEM;
898
899 memset(masks, 0xff, flen * sizeof(*masks));
900
901 for (pc = 0; pc < flen; pc++) {
902 memvalid &= masks[pc];
903
904 switch (filter[pc].code) {
905 case BPF_ST:
906 case BPF_STX:
907 memvalid |= (1 << filter[pc].k);
908 break;
909 case BPF_LD | BPF_MEM:
910 case BPF_LDX | BPF_MEM:
911 if (!(memvalid & (1 << filter[pc].k))) {
912 ret = -EINVAL;
913 goto error;
914 }
915 break;
916 case BPF_JMP | BPF_JA:
917 /* A jump must set masks on target */
918 masks[pc + 1 + filter[pc].k] &= memvalid;
919 memvalid = ~0;
920 break;
921 case BPF_JMP | BPF_JEQ | BPF_K:
922 case BPF_JMP | BPF_JEQ | BPF_X:
923 case BPF_JMP | BPF_JGE | BPF_K:
924 case BPF_JMP | BPF_JGE | BPF_X:
925 case BPF_JMP | BPF_JGT | BPF_K:
926 case BPF_JMP | BPF_JGT | BPF_X:
927 case BPF_JMP | BPF_JSET | BPF_K:
928 case BPF_JMP | BPF_JSET | BPF_X:
929 /* A jump must set masks on targets */
930 masks[pc + 1 + filter[pc].jt] &= memvalid;
931 masks[pc + 1 + filter[pc].jf] &= memvalid;
932 memvalid = ~0;
933 break;
934 }
935 }
936 error:
937 kfree(masks);
938 return ret;
939 }
940
941 static bool chk_code_allowed(u16 code_to_probe)
942 {
943 static const bool codes[] = {
944 /* 32 bit ALU operations */
945 [BPF_ALU | BPF_ADD | BPF_K] = true,
946 [BPF_ALU | BPF_ADD | BPF_X] = true,
947 [BPF_ALU | BPF_SUB | BPF_K] = true,
948 [BPF_ALU | BPF_SUB | BPF_X] = true,
949 [BPF_ALU | BPF_MUL | BPF_K] = true,
950 [BPF_ALU | BPF_MUL | BPF_X] = true,
951 [BPF_ALU | BPF_DIV | BPF_K] = true,
952 [BPF_ALU | BPF_DIV | BPF_X] = true,
953 [BPF_ALU | BPF_MOD | BPF_K] = true,
954 [BPF_ALU | BPF_MOD | BPF_X] = true,
955 [BPF_ALU | BPF_AND | BPF_K] = true,
956 [BPF_ALU | BPF_AND | BPF_X] = true,
957 [BPF_ALU | BPF_OR | BPF_K] = true,
958 [BPF_ALU | BPF_OR | BPF_X] = true,
959 [BPF_ALU | BPF_XOR | BPF_K] = true,
960 [BPF_ALU | BPF_XOR | BPF_X] = true,
961 [BPF_ALU | BPF_LSH | BPF_K] = true,
962 [BPF_ALU | BPF_LSH | BPF_X] = true,
963 [BPF_ALU | BPF_RSH | BPF_K] = true,
964 [BPF_ALU | BPF_RSH | BPF_X] = true,
965 [BPF_ALU | BPF_NEG] = true,
966 /* Load instructions */
967 [BPF_LD | BPF_W | BPF_ABS] = true,
968 [BPF_LD | BPF_H | BPF_ABS] = true,
969 [BPF_LD | BPF_B | BPF_ABS] = true,
970 [BPF_LD | BPF_W | BPF_LEN] = true,
971 [BPF_LD | BPF_W | BPF_IND] = true,
972 [BPF_LD | BPF_H | BPF_IND] = true,
973 [BPF_LD | BPF_B | BPF_IND] = true,
974 [BPF_LD | BPF_IMM] = true,
975 [BPF_LD | BPF_MEM] = true,
976 [BPF_LDX | BPF_W | BPF_LEN] = true,
977 [BPF_LDX | BPF_B | BPF_MSH] = true,
978 [BPF_LDX | BPF_IMM] = true,
979 [BPF_LDX | BPF_MEM] = true,
980 /* Store instructions */
981 [BPF_ST] = true,
982 [BPF_STX] = true,
983 /* Misc instructions */
984 [BPF_MISC | BPF_TAX] = true,
985 [BPF_MISC | BPF_TXA] = true,
986 /* Return instructions */
987 [BPF_RET | BPF_K] = true,
988 [BPF_RET | BPF_A] = true,
989 /* Jump instructions */
990 [BPF_JMP | BPF_JA] = true,
991 [BPF_JMP | BPF_JEQ | BPF_K] = true,
992 [BPF_JMP | BPF_JEQ | BPF_X] = true,
993 [BPF_JMP | BPF_JGE | BPF_K] = true,
994 [BPF_JMP | BPF_JGE | BPF_X] = true,
995 [BPF_JMP | BPF_JGT | BPF_K] = true,
996 [BPF_JMP | BPF_JGT | BPF_X] = true,
997 [BPF_JMP | BPF_JSET | BPF_K] = true,
998 [BPF_JMP | BPF_JSET | BPF_X] = true,
999 };
1000
1001 if (code_to_probe >= ARRAY_SIZE(codes))
1002 return false;
1003
1004 return codes[code_to_probe];
1005 }
1006
1007 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1008 unsigned int flen)
1009 {
1010 if (filter == NULL)
1011 return false;
1012 if (flen == 0 || flen > BPF_MAXINSNS)
1013 return false;
1014
1015 return true;
1016 }
1017
1018 /**
1019 * bpf_check_classic - verify socket filter code
1020 * @filter: filter to verify
1021 * @flen: length of filter
1022 *
1023 * Check the user's filter code. If we let some ugly
1024 * filter code slip through kaboom! The filter must contain
1025 * no references or jumps that are out of range, no illegal
1026 * instructions, and must end with a RET instruction.
1027 *
1028 * All jumps are forward as they are not signed.
1029 *
1030 * Returns 0 if the rule set is legal or -EINVAL if not.
1031 */
1032 static int bpf_check_classic(const struct sock_filter *filter,
1033 unsigned int flen)
1034 {
1035 bool anc_found;
1036 int pc;
1037
1038 /* Check the filter code now */
1039 for (pc = 0; pc < flen; pc++) {
1040 const struct sock_filter *ftest = &filter[pc];
1041
1042 /* May we actually operate on this code? */
1043 if (!chk_code_allowed(ftest->code))
1044 return -EINVAL;
1045
1046 /* Some instructions need special checks */
1047 switch (ftest->code) {
1048 case BPF_ALU | BPF_DIV | BPF_K:
1049 case BPF_ALU | BPF_MOD | BPF_K:
1050 /* Check for division by zero */
1051 if (ftest->k == 0)
1052 return -EINVAL;
1053 break;
1054 case BPF_ALU | BPF_LSH | BPF_K:
1055 case BPF_ALU | BPF_RSH | BPF_K:
1056 if (ftest->k >= 32)
1057 return -EINVAL;
1058 break;
1059 case BPF_LD | BPF_MEM:
1060 case BPF_LDX | BPF_MEM:
1061 case BPF_ST:
1062 case BPF_STX:
1063 /* Check for invalid memory addresses */
1064 if (ftest->k >= BPF_MEMWORDS)
1065 return -EINVAL;
1066 break;
1067 case BPF_JMP | BPF_JA:
1068 /* Note, the large ftest->k might cause loops.
1069 * Compare this with conditional jumps below,
1070 * where offsets are limited. --ANK (981016)
1071 */
1072 if (ftest->k >= (unsigned int)(flen - pc - 1))
1073 return -EINVAL;
1074 break;
1075 case BPF_JMP | BPF_JEQ | BPF_K:
1076 case BPF_JMP | BPF_JEQ | BPF_X:
1077 case BPF_JMP | BPF_JGE | BPF_K:
1078 case BPF_JMP | BPF_JGE | BPF_X:
1079 case BPF_JMP | BPF_JGT | BPF_K:
1080 case BPF_JMP | BPF_JGT | BPF_X:
1081 case BPF_JMP | BPF_JSET | BPF_K:
1082 case BPF_JMP | BPF_JSET | BPF_X:
1083 /* Both conditionals must be safe */
1084 if (pc + ftest->jt + 1 >= flen ||
1085 pc + ftest->jf + 1 >= flen)
1086 return -EINVAL;
1087 break;
1088 case BPF_LD | BPF_W | BPF_ABS:
1089 case BPF_LD | BPF_H | BPF_ABS:
1090 case BPF_LD | BPF_B | BPF_ABS:
1091 anc_found = false;
1092 if (bpf_anc_helper(ftest) & BPF_ANC)
1093 anc_found = true;
1094 /* Ancillary operation unknown or unsupported */
1095 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1096 return -EINVAL;
1097 }
1098 }
1099
1100 /* Last instruction must be a RET code */
1101 switch (filter[flen - 1].code) {
1102 case BPF_RET | BPF_K:
1103 case BPF_RET | BPF_A:
1104 return check_load_and_stores(filter, flen);
1105 }
1106
1107 return -EINVAL;
1108 }
1109
1110 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1111 const struct sock_fprog *fprog)
1112 {
1113 unsigned int fsize = bpf_classic_proglen(fprog);
1114 struct sock_fprog_kern *fkprog;
1115
1116 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1117 if (!fp->orig_prog)
1118 return -ENOMEM;
1119
1120 fkprog = fp->orig_prog;
1121 fkprog->len = fprog->len;
1122
1123 fkprog->filter = kmemdup(fp->insns, fsize,
1124 GFP_KERNEL | __GFP_NOWARN);
1125 if (!fkprog->filter) {
1126 kfree(fp->orig_prog);
1127 return -ENOMEM;
1128 }
1129
1130 return 0;
1131 }
1132
1133 static void bpf_release_orig_filter(struct bpf_prog *fp)
1134 {
1135 struct sock_fprog_kern *fprog = fp->orig_prog;
1136
1137 if (fprog) {
1138 kfree(fprog->filter);
1139 kfree(fprog);
1140 }
1141 }
1142
1143 static void __bpf_prog_release(struct bpf_prog *prog)
1144 {
1145 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1146 bpf_prog_put(prog);
1147 } else {
1148 bpf_release_orig_filter(prog);
1149 bpf_prog_free(prog);
1150 }
1151 }
1152
1153 static void __sk_filter_release(struct sk_filter *fp)
1154 {
1155 __bpf_prog_release(fp->prog);
1156 kfree(fp);
1157 }
1158
1159 /**
1160 * sk_filter_release_rcu - Release a socket filter by rcu_head
1161 * @rcu: rcu_head that contains the sk_filter to free
1162 */
1163 static void sk_filter_release_rcu(struct rcu_head *rcu)
1164 {
1165 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1166
1167 __sk_filter_release(fp);
1168 }
1169
1170 /**
1171 * sk_filter_release - release a socket filter
1172 * @fp: filter to remove
1173 *
1174 * Remove a filter from a socket and release its resources.
1175 */
1176 static void sk_filter_release(struct sk_filter *fp)
1177 {
1178 if (refcount_dec_and_test(&fp->refcnt))
1179 call_rcu(&fp->rcu, sk_filter_release_rcu);
1180 }
1181
1182 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1183 {
1184 u32 filter_size = bpf_prog_size(fp->prog->len);
1185
1186 atomic_sub(filter_size, &sk->sk_omem_alloc);
1187 sk_filter_release(fp);
1188 }
1189
1190 /* try to charge the socket memory if there is space available
1191 * return true on success
1192 */
1193 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1194 {
1195 u32 filter_size = bpf_prog_size(fp->prog->len);
1196
1197 /* same check as in sock_kmalloc() */
1198 if (filter_size <= sysctl_optmem_max &&
1199 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1200 atomic_add(filter_size, &sk->sk_omem_alloc);
1201 return true;
1202 }
1203 return false;
1204 }
1205
1206 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1207 {
1208 if (!refcount_inc_not_zero(&fp->refcnt))
1209 return false;
1210
1211 if (!__sk_filter_charge(sk, fp)) {
1212 sk_filter_release(fp);
1213 return false;
1214 }
1215 return true;
1216 }
1217
1218 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1219 {
1220 struct sock_filter *old_prog;
1221 struct bpf_prog *old_fp;
1222 int err, new_len, old_len = fp->len;
1223 bool seen_ld_abs = false;
1224
1225 /* We are free to overwrite insns et al right here as it
1226 * won't be used at this point in time anymore internally
1227 * after the migration to the internal BPF instruction
1228 * representation.
1229 */
1230 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1231 sizeof(struct bpf_insn));
1232
1233 /* Conversion cannot happen on overlapping memory areas,
1234 * so we need to keep the user BPF around until the 2nd
1235 * pass. At this time, the user BPF is stored in fp->insns.
1236 */
1237 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1238 GFP_KERNEL | __GFP_NOWARN);
1239 if (!old_prog) {
1240 err = -ENOMEM;
1241 goto out_err;
1242 }
1243
1244 /* 1st pass: calculate the new program length. */
1245 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1246 &seen_ld_abs);
1247 if (err)
1248 goto out_err_free;
1249
1250 /* Expand fp for appending the new filter representation. */
1251 old_fp = fp;
1252 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1253 if (!fp) {
1254 /* The old_fp is still around in case we couldn't
1255 * allocate new memory, so uncharge on that one.
1256 */
1257 fp = old_fp;
1258 err = -ENOMEM;
1259 goto out_err_free;
1260 }
1261
1262 fp->len = new_len;
1263
1264 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1265 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1266 &seen_ld_abs);
1267 if (err)
1268 /* 2nd bpf_convert_filter() can fail only if it fails
1269 * to allocate memory, remapping must succeed. Note,
1270 * that at this time old_fp has already been released
1271 * by krealloc().
1272 */
1273 goto out_err_free;
1274
1275 fp = bpf_prog_select_runtime(fp, &err);
1276 if (err)
1277 goto out_err_free;
1278
1279 kfree(old_prog);
1280 return fp;
1281
1282 out_err_free:
1283 kfree(old_prog);
1284 out_err:
1285 __bpf_prog_release(fp);
1286 return ERR_PTR(err);
1287 }
1288
1289 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1290 bpf_aux_classic_check_t trans)
1291 {
1292 int err;
1293
1294 fp->bpf_func = NULL;
1295 fp->jited = 0;
1296
1297 err = bpf_check_classic(fp->insns, fp->len);
1298 if (err) {
1299 __bpf_prog_release(fp);
1300 return ERR_PTR(err);
1301 }
1302
1303 /* There might be additional checks and transformations
1304 * needed on classic filters, f.e. in case of seccomp.
1305 */
1306 if (trans) {
1307 err = trans(fp->insns, fp->len);
1308 if (err) {
1309 __bpf_prog_release(fp);
1310 return ERR_PTR(err);
1311 }
1312 }
1313
1314 /* Probe if we can JIT compile the filter and if so, do
1315 * the compilation of the filter.
1316 */
1317 bpf_jit_compile(fp);
1318
1319 /* JIT compiler couldn't process this filter, so do the
1320 * internal BPF translation for the optimized interpreter.
1321 */
1322 if (!fp->jited)
1323 fp = bpf_migrate_filter(fp);
1324
1325 return fp;
1326 }
1327
1328 /**
1329 * bpf_prog_create - create an unattached filter
1330 * @pfp: the unattached filter that is created
1331 * @fprog: the filter program
1332 *
1333 * Create a filter independent of any socket. We first run some
1334 * sanity checks on it to make sure it does not explode on us later.
1335 * If an error occurs or there is insufficient memory for the filter
1336 * a negative errno code is returned. On success the return is zero.
1337 */
1338 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1339 {
1340 unsigned int fsize = bpf_classic_proglen(fprog);
1341 struct bpf_prog *fp;
1342
1343 /* Make sure new filter is there and in the right amounts. */
1344 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1345 return -EINVAL;
1346
1347 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1348 if (!fp)
1349 return -ENOMEM;
1350
1351 memcpy(fp->insns, fprog->filter, fsize);
1352
1353 fp->len = fprog->len;
1354 /* Since unattached filters are not copied back to user
1355 * space through sk_get_filter(), we do not need to hold
1356 * a copy here, and can spare us the work.
1357 */
1358 fp->orig_prog = NULL;
1359
1360 /* bpf_prepare_filter() already takes care of freeing
1361 * memory in case something goes wrong.
1362 */
1363 fp = bpf_prepare_filter(fp, NULL);
1364 if (IS_ERR(fp))
1365 return PTR_ERR(fp);
1366
1367 *pfp = fp;
1368 return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(bpf_prog_create);
1371
1372 /**
1373 * bpf_prog_create_from_user - create an unattached filter from user buffer
1374 * @pfp: the unattached filter that is created
1375 * @fprog: the filter program
1376 * @trans: post-classic verifier transformation handler
1377 * @save_orig: save classic BPF program
1378 *
1379 * This function effectively does the same as bpf_prog_create(), only
1380 * that it builds up its insns buffer from user space provided buffer.
1381 * It also allows for passing a bpf_aux_classic_check_t handler.
1382 */
1383 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1384 bpf_aux_classic_check_t trans, bool save_orig)
1385 {
1386 unsigned int fsize = bpf_classic_proglen(fprog);
1387 struct bpf_prog *fp;
1388 int err;
1389
1390 /* Make sure new filter is there and in the right amounts. */
1391 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1392 return -EINVAL;
1393
1394 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1395 if (!fp)
1396 return -ENOMEM;
1397
1398 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1399 __bpf_prog_free(fp);
1400 return -EFAULT;
1401 }
1402
1403 fp->len = fprog->len;
1404 fp->orig_prog = NULL;
1405
1406 if (save_orig) {
1407 err = bpf_prog_store_orig_filter(fp, fprog);
1408 if (err) {
1409 __bpf_prog_free(fp);
1410 return -ENOMEM;
1411 }
1412 }
1413
1414 /* bpf_prepare_filter() already takes care of freeing
1415 * memory in case something goes wrong.
1416 */
1417 fp = bpf_prepare_filter(fp, trans);
1418 if (IS_ERR(fp))
1419 return PTR_ERR(fp);
1420
1421 *pfp = fp;
1422 return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1425
1426 void bpf_prog_destroy(struct bpf_prog *fp)
1427 {
1428 __bpf_prog_release(fp);
1429 }
1430 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1431
1432 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1433 {
1434 struct sk_filter *fp, *old_fp;
1435
1436 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1437 if (!fp)
1438 return -ENOMEM;
1439
1440 fp->prog = prog;
1441
1442 if (!__sk_filter_charge(sk, fp)) {
1443 kfree(fp);
1444 return -ENOMEM;
1445 }
1446 refcount_set(&fp->refcnt, 1);
1447
1448 old_fp = rcu_dereference_protected(sk->sk_filter,
1449 lockdep_sock_is_held(sk));
1450 rcu_assign_pointer(sk->sk_filter, fp);
1451
1452 if (old_fp)
1453 sk_filter_uncharge(sk, old_fp);
1454
1455 return 0;
1456 }
1457
1458 static
1459 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1460 {
1461 unsigned int fsize = bpf_classic_proglen(fprog);
1462 struct bpf_prog *prog;
1463 int err;
1464
1465 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1466 return ERR_PTR(-EPERM);
1467
1468 /* Make sure new filter is there and in the right amounts. */
1469 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1470 return ERR_PTR(-EINVAL);
1471
1472 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1473 if (!prog)
1474 return ERR_PTR(-ENOMEM);
1475
1476 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1477 __bpf_prog_free(prog);
1478 return ERR_PTR(-EFAULT);
1479 }
1480
1481 prog->len = fprog->len;
1482
1483 err = bpf_prog_store_orig_filter(prog, fprog);
1484 if (err) {
1485 __bpf_prog_free(prog);
1486 return ERR_PTR(-ENOMEM);
1487 }
1488
1489 /* bpf_prepare_filter() already takes care of freeing
1490 * memory in case something goes wrong.
1491 */
1492 return bpf_prepare_filter(prog, NULL);
1493 }
1494
1495 /**
1496 * sk_attach_filter - attach a socket filter
1497 * @fprog: the filter program
1498 * @sk: the socket to use
1499 *
1500 * Attach the user's filter code. We first run some sanity checks on
1501 * it to make sure it does not explode on us later. If an error
1502 * occurs or there is insufficient memory for the filter a negative
1503 * errno code is returned. On success the return is zero.
1504 */
1505 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1506 {
1507 struct bpf_prog *prog = __get_filter(fprog, sk);
1508 int err;
1509
1510 if (IS_ERR(prog))
1511 return PTR_ERR(prog);
1512
1513 err = __sk_attach_prog(prog, sk);
1514 if (err < 0) {
1515 __bpf_prog_release(prog);
1516 return err;
1517 }
1518
1519 return 0;
1520 }
1521 EXPORT_SYMBOL_GPL(sk_attach_filter);
1522
1523 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1524 {
1525 struct bpf_prog *prog = __get_filter(fprog, sk);
1526 int err;
1527
1528 if (IS_ERR(prog))
1529 return PTR_ERR(prog);
1530
1531 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1532 err = -ENOMEM;
1533 else
1534 err = reuseport_attach_prog(sk, prog);
1535
1536 if (err)
1537 __bpf_prog_release(prog);
1538
1539 return err;
1540 }
1541
1542 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1543 {
1544 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1545 return ERR_PTR(-EPERM);
1546
1547 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1548 }
1549
1550 int sk_attach_bpf(u32 ufd, struct sock *sk)
1551 {
1552 struct bpf_prog *prog = __get_bpf(ufd, sk);
1553 int err;
1554
1555 if (IS_ERR(prog))
1556 return PTR_ERR(prog);
1557
1558 err = __sk_attach_prog(prog, sk);
1559 if (err < 0) {
1560 bpf_prog_put(prog);
1561 return err;
1562 }
1563
1564 return 0;
1565 }
1566
1567 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1568 {
1569 struct bpf_prog *prog;
1570 int err;
1571
1572 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1573 return -EPERM;
1574
1575 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1576 if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1577 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1578 if (IS_ERR(prog))
1579 return PTR_ERR(prog);
1580
1581 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1582 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1583 * bpf prog (e.g. sockmap). It depends on the
1584 * limitation imposed by bpf_prog_load().
1585 * Hence, sysctl_optmem_max is not checked.
1586 */
1587 if ((sk->sk_type != SOCK_STREAM &&
1588 sk->sk_type != SOCK_DGRAM) ||
1589 (sk->sk_protocol != IPPROTO_UDP &&
1590 sk->sk_protocol != IPPROTO_TCP) ||
1591 (sk->sk_family != AF_INET &&
1592 sk->sk_family != AF_INET6)) {
1593 err = -ENOTSUPP;
1594 goto err_prog_put;
1595 }
1596 } else {
1597 /* BPF_PROG_TYPE_SOCKET_FILTER */
1598 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1599 err = -ENOMEM;
1600 goto err_prog_put;
1601 }
1602 }
1603
1604 err = reuseport_attach_prog(sk, prog);
1605 err_prog_put:
1606 if (err)
1607 bpf_prog_put(prog);
1608
1609 return err;
1610 }
1611
1612 void sk_reuseport_prog_free(struct bpf_prog *prog)
1613 {
1614 if (!prog)
1615 return;
1616
1617 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1618 bpf_prog_put(prog);
1619 else
1620 bpf_prog_destroy(prog);
1621 }
1622
1623 struct bpf_scratchpad {
1624 union {
1625 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1626 u8 buff[MAX_BPF_STACK];
1627 };
1628 };
1629
1630 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1631
1632 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1633 unsigned int write_len)
1634 {
1635 return skb_ensure_writable(skb, write_len);
1636 }
1637
1638 static inline int bpf_try_make_writable(struct sk_buff *skb,
1639 unsigned int write_len)
1640 {
1641 int err = __bpf_try_make_writable(skb, write_len);
1642
1643 bpf_compute_data_pointers(skb);
1644 return err;
1645 }
1646
1647 static int bpf_try_make_head_writable(struct sk_buff *skb)
1648 {
1649 return bpf_try_make_writable(skb, skb_headlen(skb));
1650 }
1651
1652 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1653 {
1654 if (skb_at_tc_ingress(skb))
1655 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1656 }
1657
1658 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1659 {
1660 if (skb_at_tc_ingress(skb))
1661 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1662 }
1663
1664 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1665 const void *, from, u32, len, u64, flags)
1666 {
1667 void *ptr;
1668
1669 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1670 return -EINVAL;
1671 if (unlikely(offset > 0xffff))
1672 return -EFAULT;
1673 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1674 return -EFAULT;
1675
1676 ptr = skb->data + offset;
1677 if (flags & BPF_F_RECOMPUTE_CSUM)
1678 __skb_postpull_rcsum(skb, ptr, len, offset);
1679
1680 memcpy(ptr, from, len);
1681
1682 if (flags & BPF_F_RECOMPUTE_CSUM)
1683 __skb_postpush_rcsum(skb, ptr, len, offset);
1684 if (flags & BPF_F_INVALIDATE_HASH)
1685 skb_clear_hash(skb);
1686
1687 return 0;
1688 }
1689
1690 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1691 .func = bpf_skb_store_bytes,
1692 .gpl_only = false,
1693 .ret_type = RET_INTEGER,
1694 .arg1_type = ARG_PTR_TO_CTX,
1695 .arg2_type = ARG_ANYTHING,
1696 .arg3_type = ARG_PTR_TO_MEM,
1697 .arg4_type = ARG_CONST_SIZE,
1698 .arg5_type = ARG_ANYTHING,
1699 };
1700
1701 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1702 void *, to, u32, len)
1703 {
1704 void *ptr;
1705
1706 if (unlikely(offset > 0xffff))
1707 goto err_clear;
1708
1709 ptr = skb_header_pointer(skb, offset, len, to);
1710 if (unlikely(!ptr))
1711 goto err_clear;
1712 if (ptr != to)
1713 memcpy(to, ptr, len);
1714
1715 return 0;
1716 err_clear:
1717 memset(to, 0, len);
1718 return -EFAULT;
1719 }
1720
1721 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1722 .func = bpf_skb_load_bytes,
1723 .gpl_only = false,
1724 .ret_type = RET_INTEGER,
1725 .arg1_type = ARG_PTR_TO_CTX,
1726 .arg2_type = ARG_ANYTHING,
1727 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1728 .arg4_type = ARG_CONST_SIZE,
1729 };
1730
1731 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1732 const struct bpf_flow_dissector *, ctx, u32, offset,
1733 void *, to, u32, len)
1734 {
1735 void *ptr;
1736
1737 if (unlikely(offset > 0xffff))
1738 goto err_clear;
1739
1740 if (unlikely(!ctx->skb))
1741 goto err_clear;
1742
1743 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1744 if (unlikely(!ptr))
1745 goto err_clear;
1746 if (ptr != to)
1747 memcpy(to, ptr, len);
1748
1749 return 0;
1750 err_clear:
1751 memset(to, 0, len);
1752 return -EFAULT;
1753 }
1754
1755 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1756 .func = bpf_flow_dissector_load_bytes,
1757 .gpl_only = false,
1758 .ret_type = RET_INTEGER,
1759 .arg1_type = ARG_PTR_TO_CTX,
1760 .arg2_type = ARG_ANYTHING,
1761 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1762 .arg4_type = ARG_CONST_SIZE,
1763 };
1764
1765 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1766 u32, offset, void *, to, u32, len, u32, start_header)
1767 {
1768 u8 *end = skb_tail_pointer(skb);
1769 u8 *net = skb_network_header(skb);
1770 u8 *mac = skb_mac_header(skb);
1771 u8 *ptr;
1772
1773 if (unlikely(offset > 0xffff || len > (end - mac)))
1774 goto err_clear;
1775
1776 switch (start_header) {
1777 case BPF_HDR_START_MAC:
1778 ptr = mac + offset;
1779 break;
1780 case BPF_HDR_START_NET:
1781 ptr = net + offset;
1782 break;
1783 default:
1784 goto err_clear;
1785 }
1786
1787 if (likely(ptr >= mac && ptr + len <= end)) {
1788 memcpy(to, ptr, len);
1789 return 0;
1790 }
1791
1792 err_clear:
1793 memset(to, 0, len);
1794 return -EFAULT;
1795 }
1796
1797 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1798 .func = bpf_skb_load_bytes_relative,
1799 .gpl_only = false,
1800 .ret_type = RET_INTEGER,
1801 .arg1_type = ARG_PTR_TO_CTX,
1802 .arg2_type = ARG_ANYTHING,
1803 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1804 .arg4_type = ARG_CONST_SIZE,
1805 .arg5_type = ARG_ANYTHING,
1806 };
1807
1808 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1809 {
1810 /* Idea is the following: should the needed direct read/write
1811 * test fail during runtime, we can pull in more data and redo
1812 * again, since implicitly, we invalidate previous checks here.
1813 *
1814 * Or, since we know how much we need to make read/writeable,
1815 * this can be done once at the program beginning for direct
1816 * access case. By this we overcome limitations of only current
1817 * headroom being accessible.
1818 */
1819 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1820 }
1821
1822 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1823 .func = bpf_skb_pull_data,
1824 .gpl_only = false,
1825 .ret_type = RET_INTEGER,
1826 .arg1_type = ARG_PTR_TO_CTX,
1827 .arg2_type = ARG_ANYTHING,
1828 };
1829
1830 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1831 {
1832 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1833 }
1834
1835 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1836 .func = bpf_sk_fullsock,
1837 .gpl_only = false,
1838 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1839 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1840 };
1841
1842 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1843 unsigned int write_len)
1844 {
1845 int err = __bpf_try_make_writable(skb, write_len);
1846
1847 bpf_compute_data_end_sk_skb(skb);
1848 return err;
1849 }
1850
1851 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1852 {
1853 /* Idea is the following: should the needed direct read/write
1854 * test fail during runtime, we can pull in more data and redo
1855 * again, since implicitly, we invalidate previous checks here.
1856 *
1857 * Or, since we know how much we need to make read/writeable,
1858 * this can be done once at the program beginning for direct
1859 * access case. By this we overcome limitations of only current
1860 * headroom being accessible.
1861 */
1862 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1863 }
1864
1865 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1866 .func = sk_skb_pull_data,
1867 .gpl_only = false,
1868 .ret_type = RET_INTEGER,
1869 .arg1_type = ARG_PTR_TO_CTX,
1870 .arg2_type = ARG_ANYTHING,
1871 };
1872
1873 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1874 u64, from, u64, to, u64, flags)
1875 {
1876 __sum16 *ptr;
1877
1878 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1879 return -EINVAL;
1880 if (unlikely(offset > 0xffff || offset & 1))
1881 return -EFAULT;
1882 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1883 return -EFAULT;
1884
1885 ptr = (__sum16 *)(skb->data + offset);
1886 switch (flags & BPF_F_HDR_FIELD_MASK) {
1887 case 0:
1888 if (unlikely(from != 0))
1889 return -EINVAL;
1890
1891 csum_replace_by_diff(ptr, to);
1892 break;
1893 case 2:
1894 csum_replace2(ptr, from, to);
1895 break;
1896 case 4:
1897 csum_replace4(ptr, from, to);
1898 break;
1899 default:
1900 return -EINVAL;
1901 }
1902
1903 return 0;
1904 }
1905
1906 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1907 .func = bpf_l3_csum_replace,
1908 .gpl_only = false,
1909 .ret_type = RET_INTEGER,
1910 .arg1_type = ARG_PTR_TO_CTX,
1911 .arg2_type = ARG_ANYTHING,
1912 .arg3_type = ARG_ANYTHING,
1913 .arg4_type = ARG_ANYTHING,
1914 .arg5_type = ARG_ANYTHING,
1915 };
1916
1917 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1918 u64, from, u64, to, u64, flags)
1919 {
1920 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1921 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1922 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1923 __sum16 *ptr;
1924
1925 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1926 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1927 return -EINVAL;
1928 if (unlikely(offset > 0xffff || offset & 1))
1929 return -EFAULT;
1930 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1931 return -EFAULT;
1932
1933 ptr = (__sum16 *)(skb->data + offset);
1934 if (is_mmzero && !do_mforce && !*ptr)
1935 return 0;
1936
1937 switch (flags & BPF_F_HDR_FIELD_MASK) {
1938 case 0:
1939 if (unlikely(from != 0))
1940 return -EINVAL;
1941
1942 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1943 break;
1944 case 2:
1945 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1946 break;
1947 case 4:
1948 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1949 break;
1950 default:
1951 return -EINVAL;
1952 }
1953
1954 if (is_mmzero && !*ptr)
1955 *ptr = CSUM_MANGLED_0;
1956 return 0;
1957 }
1958
1959 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1960 .func = bpf_l4_csum_replace,
1961 .gpl_only = false,
1962 .ret_type = RET_INTEGER,
1963 .arg1_type = ARG_PTR_TO_CTX,
1964 .arg2_type = ARG_ANYTHING,
1965 .arg3_type = ARG_ANYTHING,
1966 .arg4_type = ARG_ANYTHING,
1967 .arg5_type = ARG_ANYTHING,
1968 };
1969
1970 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1971 __be32 *, to, u32, to_size, __wsum, seed)
1972 {
1973 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1974 u32 diff_size = from_size + to_size;
1975 int i, j = 0;
1976
1977 /* This is quite flexible, some examples:
1978 *
1979 * from_size == 0, to_size > 0, seed := csum --> pushing data
1980 * from_size > 0, to_size == 0, seed := csum --> pulling data
1981 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1982 *
1983 * Even for diffing, from_size and to_size don't need to be equal.
1984 */
1985 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1986 diff_size > sizeof(sp->diff)))
1987 return -EINVAL;
1988
1989 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1990 sp->diff[j] = ~from[i];
1991 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1992 sp->diff[j] = to[i];
1993
1994 return csum_partial(sp->diff, diff_size, seed);
1995 }
1996
1997 static const struct bpf_func_proto bpf_csum_diff_proto = {
1998 .func = bpf_csum_diff,
1999 .gpl_only = false,
2000 .pkt_access = true,
2001 .ret_type = RET_INTEGER,
2002 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2003 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2004 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2005 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2006 .arg5_type = ARG_ANYTHING,
2007 };
2008
2009 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2010 {
2011 /* The interface is to be used in combination with bpf_csum_diff()
2012 * for direct packet writes. csum rotation for alignment as well
2013 * as emulating csum_sub() can be done from the eBPF program.
2014 */
2015 if (skb->ip_summed == CHECKSUM_COMPLETE)
2016 return (skb->csum = csum_add(skb->csum, csum));
2017
2018 return -ENOTSUPP;
2019 }
2020
2021 static const struct bpf_func_proto bpf_csum_update_proto = {
2022 .func = bpf_csum_update,
2023 .gpl_only = false,
2024 .ret_type = RET_INTEGER,
2025 .arg1_type = ARG_PTR_TO_CTX,
2026 .arg2_type = ARG_ANYTHING,
2027 };
2028
2029 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2030 {
2031 return dev_forward_skb(dev, skb);
2032 }
2033
2034 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2035 struct sk_buff *skb)
2036 {
2037 int ret = ____dev_forward_skb(dev, skb);
2038
2039 if (likely(!ret)) {
2040 skb->dev = dev;
2041 ret = netif_rx(skb);
2042 }
2043
2044 return ret;
2045 }
2046
2047 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2048 {
2049 int ret;
2050
2051 if (dev_xmit_recursion()) {
2052 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2053 kfree_skb(skb);
2054 return -ENETDOWN;
2055 }
2056
2057 skb->dev = dev;
2058 skb->tstamp = 0;
2059
2060 dev_xmit_recursion_inc();
2061 ret = dev_queue_xmit(skb);
2062 dev_xmit_recursion_dec();
2063
2064 return ret;
2065 }
2066
2067 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2068 u32 flags)
2069 {
2070 unsigned int mlen = skb_network_offset(skb);
2071
2072 if (mlen) {
2073 __skb_pull(skb, mlen);
2074
2075 /* At ingress, the mac header has already been pulled once.
2076 * At egress, skb_pospull_rcsum has to be done in case that
2077 * the skb is originated from ingress (i.e. a forwarded skb)
2078 * to ensure that rcsum starts at net header.
2079 */
2080 if (!skb_at_tc_ingress(skb))
2081 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2082 }
2083 skb_pop_mac_header(skb);
2084 skb_reset_mac_len(skb);
2085 return flags & BPF_F_INGRESS ?
2086 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2087 }
2088
2089 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2090 u32 flags)
2091 {
2092 /* Verify that a link layer header is carried */
2093 if (unlikely(skb->mac_header >= skb->network_header)) {
2094 kfree_skb(skb);
2095 return -ERANGE;
2096 }
2097
2098 bpf_push_mac_rcsum(skb);
2099 return flags & BPF_F_INGRESS ?
2100 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2101 }
2102
2103 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2104 u32 flags)
2105 {
2106 if (dev_is_mac_header_xmit(dev))
2107 return __bpf_redirect_common(skb, dev, flags);
2108 else
2109 return __bpf_redirect_no_mac(skb, dev, flags);
2110 }
2111
2112 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2113 {
2114 struct net_device *dev;
2115 struct sk_buff *clone;
2116 int ret;
2117
2118 if (unlikely(flags & ~(BPF_F_INGRESS)))
2119 return -EINVAL;
2120
2121 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2122 if (unlikely(!dev))
2123 return -EINVAL;
2124
2125 clone = skb_clone(skb, GFP_ATOMIC);
2126 if (unlikely(!clone))
2127 return -ENOMEM;
2128
2129 /* For direct write, we need to keep the invariant that the skbs
2130 * we're dealing with need to be uncloned. Should uncloning fail
2131 * here, we need to free the just generated clone to unclone once
2132 * again.
2133 */
2134 ret = bpf_try_make_head_writable(skb);
2135 if (unlikely(ret)) {
2136 kfree_skb(clone);
2137 return -ENOMEM;
2138 }
2139
2140 return __bpf_redirect(clone, dev, flags);
2141 }
2142
2143 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2144 .func = bpf_clone_redirect,
2145 .gpl_only = false,
2146 .ret_type = RET_INTEGER,
2147 .arg1_type = ARG_PTR_TO_CTX,
2148 .arg2_type = ARG_ANYTHING,
2149 .arg3_type = ARG_ANYTHING,
2150 };
2151
2152 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2153 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2154
2155 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2156 {
2157 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2158
2159 if (unlikely(flags & ~(BPF_F_INGRESS)))
2160 return TC_ACT_SHOT;
2161
2162 ri->flags = flags;
2163 ri->tgt_index = ifindex;
2164
2165 return TC_ACT_REDIRECT;
2166 }
2167
2168 int skb_do_redirect(struct sk_buff *skb)
2169 {
2170 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2171 struct net_device *dev;
2172
2173 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2174 ri->tgt_index = 0;
2175 if (unlikely(!dev)) {
2176 kfree_skb(skb);
2177 return -EINVAL;
2178 }
2179
2180 return __bpf_redirect(skb, dev, ri->flags);
2181 }
2182
2183 static const struct bpf_func_proto bpf_redirect_proto = {
2184 .func = bpf_redirect,
2185 .gpl_only = false,
2186 .ret_type = RET_INTEGER,
2187 .arg1_type = ARG_ANYTHING,
2188 .arg2_type = ARG_ANYTHING,
2189 };
2190
2191 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2192 {
2193 msg->apply_bytes = bytes;
2194 return 0;
2195 }
2196
2197 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2198 .func = bpf_msg_apply_bytes,
2199 .gpl_only = false,
2200 .ret_type = RET_INTEGER,
2201 .arg1_type = ARG_PTR_TO_CTX,
2202 .arg2_type = ARG_ANYTHING,
2203 };
2204
2205 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2206 {
2207 msg->cork_bytes = bytes;
2208 return 0;
2209 }
2210
2211 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2212 .func = bpf_msg_cork_bytes,
2213 .gpl_only = false,
2214 .ret_type = RET_INTEGER,
2215 .arg1_type = ARG_PTR_TO_CTX,
2216 .arg2_type = ARG_ANYTHING,
2217 };
2218
2219 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2220 u32, end, u64, flags)
2221 {
2222 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2223 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2224 struct scatterlist *sge;
2225 u8 *raw, *to, *from;
2226 struct page *page;
2227
2228 if (unlikely(flags || end <= start))
2229 return -EINVAL;
2230
2231 /* First find the starting scatterlist element */
2232 i = msg->sg.start;
2233 do {
2234 offset += len;
2235 len = sk_msg_elem(msg, i)->length;
2236 if (start < offset + len)
2237 break;
2238 sk_msg_iter_var_next(i);
2239 } while (i != msg->sg.end);
2240
2241 if (unlikely(start >= offset + len))
2242 return -EINVAL;
2243
2244 first_sge = i;
2245 /* The start may point into the sg element so we need to also
2246 * account for the headroom.
2247 */
2248 bytes_sg_total = start - offset + bytes;
2249 if (!msg->sg.copy[i] && bytes_sg_total <= len)
2250 goto out;
2251
2252 /* At this point we need to linearize multiple scatterlist
2253 * elements or a single shared page. Either way we need to
2254 * copy into a linear buffer exclusively owned by BPF. Then
2255 * place the buffer in the scatterlist and fixup the original
2256 * entries by removing the entries now in the linear buffer
2257 * and shifting the remaining entries. For now we do not try
2258 * to copy partial entries to avoid complexity of running out
2259 * of sg_entry slots. The downside is reading a single byte
2260 * will copy the entire sg entry.
2261 */
2262 do {
2263 copy += sk_msg_elem(msg, i)->length;
2264 sk_msg_iter_var_next(i);
2265 if (bytes_sg_total <= copy)
2266 break;
2267 } while (i != msg->sg.end);
2268 last_sge = i;
2269
2270 if (unlikely(bytes_sg_total > copy))
2271 return -EINVAL;
2272
2273 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2274 get_order(copy));
2275 if (unlikely(!page))
2276 return -ENOMEM;
2277
2278 raw = page_address(page);
2279 i = first_sge;
2280 do {
2281 sge = sk_msg_elem(msg, i);
2282 from = sg_virt(sge);
2283 len = sge->length;
2284 to = raw + poffset;
2285
2286 memcpy(to, from, len);
2287 poffset += len;
2288 sge->length = 0;
2289 put_page(sg_page(sge));
2290
2291 sk_msg_iter_var_next(i);
2292 } while (i != last_sge);
2293
2294 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2295
2296 /* To repair sg ring we need to shift entries. If we only
2297 * had a single entry though we can just replace it and
2298 * be done. Otherwise walk the ring and shift the entries.
2299 */
2300 WARN_ON_ONCE(last_sge == first_sge);
2301 shift = last_sge > first_sge ?
2302 last_sge - first_sge - 1 :
2303 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2304 if (!shift)
2305 goto out;
2306
2307 i = first_sge;
2308 sk_msg_iter_var_next(i);
2309 do {
2310 u32 move_from;
2311
2312 if (i + shift >= NR_MSG_FRAG_IDS)
2313 move_from = i + shift - NR_MSG_FRAG_IDS;
2314 else
2315 move_from = i + shift;
2316 if (move_from == msg->sg.end)
2317 break;
2318
2319 msg->sg.data[i] = msg->sg.data[move_from];
2320 msg->sg.data[move_from].length = 0;
2321 msg->sg.data[move_from].page_link = 0;
2322 msg->sg.data[move_from].offset = 0;
2323 sk_msg_iter_var_next(i);
2324 } while (1);
2325
2326 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2327 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2328 msg->sg.end - shift;
2329 out:
2330 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2331 msg->data_end = msg->data + bytes;
2332 return 0;
2333 }
2334
2335 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2336 .func = bpf_msg_pull_data,
2337 .gpl_only = false,
2338 .ret_type = RET_INTEGER,
2339 .arg1_type = ARG_PTR_TO_CTX,
2340 .arg2_type = ARG_ANYTHING,
2341 .arg3_type = ARG_ANYTHING,
2342 .arg4_type = ARG_ANYTHING,
2343 };
2344
2345 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2346 u32, len, u64, flags)
2347 {
2348 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2349 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2350 u8 *raw, *to, *from;
2351 struct page *page;
2352
2353 if (unlikely(flags))
2354 return -EINVAL;
2355
2356 /* First find the starting scatterlist element */
2357 i = msg->sg.start;
2358 do {
2359 offset += l;
2360 l = sk_msg_elem(msg, i)->length;
2361
2362 if (start < offset + l)
2363 break;
2364 sk_msg_iter_var_next(i);
2365 } while (i != msg->sg.end);
2366
2367 if (start >= offset + l)
2368 return -EINVAL;
2369
2370 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2371
2372 /* If no space available will fallback to copy, we need at
2373 * least one scatterlist elem available to push data into
2374 * when start aligns to the beginning of an element or two
2375 * when it falls inside an element. We handle the start equals
2376 * offset case because its the common case for inserting a
2377 * header.
2378 */
2379 if (!space || (space == 1 && start != offset))
2380 copy = msg->sg.data[i].length;
2381
2382 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2383 get_order(copy + len));
2384 if (unlikely(!page))
2385 return -ENOMEM;
2386
2387 if (copy) {
2388 int front, back;
2389
2390 raw = page_address(page);
2391
2392 psge = sk_msg_elem(msg, i);
2393 front = start - offset;
2394 back = psge->length - front;
2395 from = sg_virt(psge);
2396
2397 if (front)
2398 memcpy(raw, from, front);
2399
2400 if (back) {
2401 from += front;
2402 to = raw + front + len;
2403
2404 memcpy(to, from, back);
2405 }
2406
2407 put_page(sg_page(psge));
2408 } else if (start - offset) {
2409 psge = sk_msg_elem(msg, i);
2410 rsge = sk_msg_elem_cpy(msg, i);
2411
2412 psge->length = start - offset;
2413 rsge.length -= psge->length;
2414 rsge.offset += start;
2415
2416 sk_msg_iter_var_next(i);
2417 sg_unmark_end(psge);
2418 sk_msg_iter_next(msg, end);
2419 }
2420
2421 /* Slot(s) to place newly allocated data */
2422 new = i;
2423
2424 /* Shift one or two slots as needed */
2425 if (!copy) {
2426 sge = sk_msg_elem_cpy(msg, i);
2427
2428 sk_msg_iter_var_next(i);
2429 sg_unmark_end(&sge);
2430 sk_msg_iter_next(msg, end);
2431
2432 nsge = sk_msg_elem_cpy(msg, i);
2433 if (rsge.length) {
2434 sk_msg_iter_var_next(i);
2435 nnsge = sk_msg_elem_cpy(msg, i);
2436 }
2437
2438 while (i != msg->sg.end) {
2439 msg->sg.data[i] = sge;
2440 sge = nsge;
2441 sk_msg_iter_var_next(i);
2442 if (rsge.length) {
2443 nsge = nnsge;
2444 nnsge = sk_msg_elem_cpy(msg, i);
2445 } else {
2446 nsge = sk_msg_elem_cpy(msg, i);
2447 }
2448 }
2449 }
2450
2451 /* Place newly allocated data buffer */
2452 sk_mem_charge(msg->sk, len);
2453 msg->sg.size += len;
2454 msg->sg.copy[new] = false;
2455 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2456 if (rsge.length) {
2457 get_page(sg_page(&rsge));
2458 sk_msg_iter_var_next(new);
2459 msg->sg.data[new] = rsge;
2460 }
2461
2462 sk_msg_compute_data_pointers(msg);
2463 return 0;
2464 }
2465
2466 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2467 .func = bpf_msg_push_data,
2468 .gpl_only = false,
2469 .ret_type = RET_INTEGER,
2470 .arg1_type = ARG_PTR_TO_CTX,
2471 .arg2_type = ARG_ANYTHING,
2472 .arg3_type = ARG_ANYTHING,
2473 .arg4_type = ARG_ANYTHING,
2474 };
2475
2476 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2477 {
2478 int prev;
2479
2480 do {
2481 prev = i;
2482 sk_msg_iter_var_next(i);
2483 msg->sg.data[prev] = msg->sg.data[i];
2484 } while (i != msg->sg.end);
2485
2486 sk_msg_iter_prev(msg, end);
2487 }
2488
2489 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2490 {
2491 struct scatterlist tmp, sge;
2492
2493 sk_msg_iter_next(msg, end);
2494 sge = sk_msg_elem_cpy(msg, i);
2495 sk_msg_iter_var_next(i);
2496 tmp = sk_msg_elem_cpy(msg, i);
2497
2498 while (i != msg->sg.end) {
2499 msg->sg.data[i] = sge;
2500 sk_msg_iter_var_next(i);
2501 sge = tmp;
2502 tmp = sk_msg_elem_cpy(msg, i);
2503 }
2504 }
2505
2506 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2507 u32, len, u64, flags)
2508 {
2509 u32 i = 0, l = 0, space, offset = 0;
2510 u64 last = start + len;
2511 int pop;
2512
2513 if (unlikely(flags))
2514 return -EINVAL;
2515
2516 /* First find the starting scatterlist element */
2517 i = msg->sg.start;
2518 do {
2519 offset += l;
2520 l = sk_msg_elem(msg, i)->length;
2521
2522 if (start < offset + l)
2523 break;
2524 sk_msg_iter_var_next(i);
2525 } while (i != msg->sg.end);
2526
2527 /* Bounds checks: start and pop must be inside message */
2528 if (start >= offset + l || last >= msg->sg.size)
2529 return -EINVAL;
2530
2531 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2532
2533 pop = len;
2534 /* --------------| offset
2535 * -| start |-------- len -------|
2536 *
2537 * |----- a ----|-------- pop -------|----- b ----|
2538 * |______________________________________________| length
2539 *
2540 *
2541 * a: region at front of scatter element to save
2542 * b: region at back of scatter element to save when length > A + pop
2543 * pop: region to pop from element, same as input 'pop' here will be
2544 * decremented below per iteration.
2545 *
2546 * Two top-level cases to handle when start != offset, first B is non
2547 * zero and second B is zero corresponding to when a pop includes more
2548 * than one element.
2549 *
2550 * Then if B is non-zero AND there is no space allocate space and
2551 * compact A, B regions into page. If there is space shift ring to
2552 * the rigth free'ing the next element in ring to place B, leaving
2553 * A untouched except to reduce length.
2554 */
2555 if (start != offset) {
2556 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2557 int a = start;
2558 int b = sge->length - pop - a;
2559
2560 sk_msg_iter_var_next(i);
2561
2562 if (pop < sge->length - a) {
2563 if (space) {
2564 sge->length = a;
2565 sk_msg_shift_right(msg, i);
2566 nsge = sk_msg_elem(msg, i);
2567 get_page(sg_page(sge));
2568 sg_set_page(nsge,
2569 sg_page(sge),
2570 b, sge->offset + pop + a);
2571 } else {
2572 struct page *page, *orig;
2573 u8 *to, *from;
2574
2575 page = alloc_pages(__GFP_NOWARN |
2576 __GFP_COMP | GFP_ATOMIC,
2577 get_order(a + b));
2578 if (unlikely(!page))
2579 return -ENOMEM;
2580
2581 sge->length = a;
2582 orig = sg_page(sge);
2583 from = sg_virt(sge);
2584 to = page_address(page);
2585 memcpy(to, from, a);
2586 memcpy(to + a, from + a + pop, b);
2587 sg_set_page(sge, page, a + b, 0);
2588 put_page(orig);
2589 }
2590 pop = 0;
2591 } else if (pop >= sge->length - a) {
2592 sge->length = a;
2593 pop -= (sge->length - a);
2594 }
2595 }
2596
2597 /* From above the current layout _must_ be as follows,
2598 *
2599 * -| offset
2600 * -| start
2601 *
2602 * |---- pop ---|---------------- b ------------|
2603 * |____________________________________________| length
2604 *
2605 * Offset and start of the current msg elem are equal because in the
2606 * previous case we handled offset != start and either consumed the
2607 * entire element and advanced to the next element OR pop == 0.
2608 *
2609 * Two cases to handle here are first pop is less than the length
2610 * leaving some remainder b above. Simply adjust the element's layout
2611 * in this case. Or pop >= length of the element so that b = 0. In this
2612 * case advance to next element decrementing pop.
2613 */
2614 while (pop) {
2615 struct scatterlist *sge = sk_msg_elem(msg, i);
2616
2617 if (pop < sge->length) {
2618 sge->length -= pop;
2619 sge->offset += pop;
2620 pop = 0;
2621 } else {
2622 pop -= sge->length;
2623 sk_msg_shift_left(msg, i);
2624 }
2625 sk_msg_iter_var_next(i);
2626 }
2627
2628 sk_mem_uncharge(msg->sk, len - pop);
2629 msg->sg.size -= (len - pop);
2630 sk_msg_compute_data_pointers(msg);
2631 return 0;
2632 }
2633
2634 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2635 .func = bpf_msg_pop_data,
2636 .gpl_only = false,
2637 .ret_type = RET_INTEGER,
2638 .arg1_type = ARG_PTR_TO_CTX,
2639 .arg2_type = ARG_ANYTHING,
2640 .arg3_type = ARG_ANYTHING,
2641 .arg4_type = ARG_ANYTHING,
2642 };
2643
2644 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2645 {
2646 return task_get_classid(skb);
2647 }
2648
2649 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2650 .func = bpf_get_cgroup_classid,
2651 .gpl_only = false,
2652 .ret_type = RET_INTEGER,
2653 .arg1_type = ARG_PTR_TO_CTX,
2654 };
2655
2656 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2657 {
2658 return dst_tclassid(skb);
2659 }
2660
2661 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2662 .func = bpf_get_route_realm,
2663 .gpl_only = false,
2664 .ret_type = RET_INTEGER,
2665 .arg1_type = ARG_PTR_TO_CTX,
2666 };
2667
2668 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2669 {
2670 /* If skb_clear_hash() was called due to mangling, we can
2671 * trigger SW recalculation here. Later access to hash
2672 * can then use the inline skb->hash via context directly
2673 * instead of calling this helper again.
2674 */
2675 return skb_get_hash(skb);
2676 }
2677
2678 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2679 .func = bpf_get_hash_recalc,
2680 .gpl_only = false,
2681 .ret_type = RET_INTEGER,
2682 .arg1_type = ARG_PTR_TO_CTX,
2683 };
2684
2685 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2686 {
2687 /* After all direct packet write, this can be used once for
2688 * triggering a lazy recalc on next skb_get_hash() invocation.
2689 */
2690 skb_clear_hash(skb);
2691 return 0;
2692 }
2693
2694 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2695 .func = bpf_set_hash_invalid,
2696 .gpl_only = false,
2697 .ret_type = RET_INTEGER,
2698 .arg1_type = ARG_PTR_TO_CTX,
2699 };
2700
2701 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2702 {
2703 /* Set user specified hash as L4(+), so that it gets returned
2704 * on skb_get_hash() call unless BPF prog later on triggers a
2705 * skb_clear_hash().
2706 */
2707 __skb_set_sw_hash(skb, hash, true);
2708 return 0;
2709 }
2710
2711 static const struct bpf_func_proto bpf_set_hash_proto = {
2712 .func = bpf_set_hash,
2713 .gpl_only = false,
2714 .ret_type = RET_INTEGER,
2715 .arg1_type = ARG_PTR_TO_CTX,
2716 .arg2_type = ARG_ANYTHING,
2717 };
2718
2719 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2720 u16, vlan_tci)
2721 {
2722 int ret;
2723
2724 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2725 vlan_proto != htons(ETH_P_8021AD)))
2726 vlan_proto = htons(ETH_P_8021Q);
2727
2728 bpf_push_mac_rcsum(skb);
2729 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2730 bpf_pull_mac_rcsum(skb);
2731
2732 bpf_compute_data_pointers(skb);
2733 return ret;
2734 }
2735
2736 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2737 .func = bpf_skb_vlan_push,
2738 .gpl_only = false,
2739 .ret_type = RET_INTEGER,
2740 .arg1_type = ARG_PTR_TO_CTX,
2741 .arg2_type = ARG_ANYTHING,
2742 .arg3_type = ARG_ANYTHING,
2743 };
2744
2745 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2746 {
2747 int ret;
2748
2749 bpf_push_mac_rcsum(skb);
2750 ret = skb_vlan_pop(skb);
2751 bpf_pull_mac_rcsum(skb);
2752
2753 bpf_compute_data_pointers(skb);
2754 return ret;
2755 }
2756
2757 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2758 .func = bpf_skb_vlan_pop,
2759 .gpl_only = false,
2760 .ret_type = RET_INTEGER,
2761 .arg1_type = ARG_PTR_TO_CTX,
2762 };
2763
2764 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2765 {
2766 /* Caller already did skb_cow() with len as headroom,
2767 * so no need to do it here.
2768 */
2769 skb_push(skb, len);
2770 memmove(skb->data, skb->data + len, off);
2771 memset(skb->data + off, 0, len);
2772
2773 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2774 * needed here as it does not change the skb->csum
2775 * result for checksum complete when summing over
2776 * zeroed blocks.
2777 */
2778 return 0;
2779 }
2780
2781 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2782 {
2783 /* skb_ensure_writable() is not needed here, as we're
2784 * already working on an uncloned skb.
2785 */
2786 if (unlikely(!pskb_may_pull(skb, off + len)))
2787 return -ENOMEM;
2788
2789 skb_postpull_rcsum(skb, skb->data + off, len);
2790 memmove(skb->data + len, skb->data, off);
2791 __skb_pull(skb, len);
2792
2793 return 0;
2794 }
2795
2796 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2797 {
2798 bool trans_same = skb->transport_header == skb->network_header;
2799 int ret;
2800
2801 /* There's no need for __skb_push()/__skb_pull() pair to
2802 * get to the start of the mac header as we're guaranteed
2803 * to always start from here under eBPF.
2804 */
2805 ret = bpf_skb_generic_push(skb, off, len);
2806 if (likely(!ret)) {
2807 skb->mac_header -= len;
2808 skb->network_header -= len;
2809 if (trans_same)
2810 skb->transport_header = skb->network_header;
2811 }
2812
2813 return ret;
2814 }
2815
2816 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2817 {
2818 bool trans_same = skb->transport_header == skb->network_header;
2819 int ret;
2820
2821 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2822 ret = bpf_skb_generic_pop(skb, off, len);
2823 if (likely(!ret)) {
2824 skb->mac_header += len;
2825 skb->network_header += len;
2826 if (trans_same)
2827 skb->transport_header = skb->network_header;
2828 }
2829
2830 return ret;
2831 }
2832
2833 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2834 {
2835 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2836 u32 off = skb_mac_header_len(skb);
2837 int ret;
2838
2839 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2840 return -ENOTSUPP;
2841
2842 ret = skb_cow(skb, len_diff);
2843 if (unlikely(ret < 0))
2844 return ret;
2845
2846 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2847 if (unlikely(ret < 0))
2848 return ret;
2849
2850 if (skb_is_gso(skb)) {
2851 struct skb_shared_info *shinfo = skb_shinfo(skb);
2852
2853 /* SKB_GSO_TCPV4 needs to be changed into
2854 * SKB_GSO_TCPV6.
2855 */
2856 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2857 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2858 shinfo->gso_type |= SKB_GSO_TCPV6;
2859 }
2860
2861 /* Due to IPv6 header, MSS needs to be downgraded. */
2862 skb_decrease_gso_size(shinfo, len_diff);
2863 /* Header must be checked, and gso_segs recomputed. */
2864 shinfo->gso_type |= SKB_GSO_DODGY;
2865 shinfo->gso_segs = 0;
2866 }
2867
2868 skb->protocol = htons(ETH_P_IPV6);
2869 skb_clear_hash(skb);
2870
2871 return 0;
2872 }
2873
2874 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2875 {
2876 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2877 u32 off = skb_mac_header_len(skb);
2878 int ret;
2879
2880 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2881 return -ENOTSUPP;
2882
2883 ret = skb_unclone(skb, GFP_ATOMIC);
2884 if (unlikely(ret < 0))
2885 return ret;
2886
2887 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2888 if (unlikely(ret < 0))
2889 return ret;
2890
2891 if (skb_is_gso(skb)) {
2892 struct skb_shared_info *shinfo = skb_shinfo(skb);
2893
2894 /* SKB_GSO_TCPV6 needs to be changed into
2895 * SKB_GSO_TCPV4.
2896 */
2897 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2898 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2899 shinfo->gso_type |= SKB_GSO_TCPV4;
2900 }
2901
2902 /* Due to IPv4 header, MSS can be upgraded. */
2903 skb_increase_gso_size(shinfo, len_diff);
2904 /* Header must be checked, and gso_segs recomputed. */
2905 shinfo->gso_type |= SKB_GSO_DODGY;
2906 shinfo->gso_segs = 0;
2907 }
2908
2909 skb->protocol = htons(ETH_P_IP);
2910 skb_clear_hash(skb);
2911
2912 return 0;
2913 }
2914
2915 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2916 {
2917 __be16 from_proto = skb->protocol;
2918
2919 if (from_proto == htons(ETH_P_IP) &&
2920 to_proto == htons(ETH_P_IPV6))
2921 return bpf_skb_proto_4_to_6(skb);
2922
2923 if (from_proto == htons(ETH_P_IPV6) &&
2924 to_proto == htons(ETH_P_IP))
2925 return bpf_skb_proto_6_to_4(skb);
2926
2927 return -ENOTSUPP;
2928 }
2929
2930 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2931 u64, flags)
2932 {
2933 int ret;
2934
2935 if (unlikely(flags))
2936 return -EINVAL;
2937
2938 /* General idea is that this helper does the basic groundwork
2939 * needed for changing the protocol, and eBPF program fills the
2940 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2941 * and other helpers, rather than passing a raw buffer here.
2942 *
2943 * The rationale is to keep this minimal and without a need to
2944 * deal with raw packet data. F.e. even if we would pass buffers
2945 * here, the program still needs to call the bpf_lX_csum_replace()
2946 * helpers anyway. Plus, this way we keep also separation of
2947 * concerns, since f.e. bpf_skb_store_bytes() should only take
2948 * care of stores.
2949 *
2950 * Currently, additional options and extension header space are
2951 * not supported, but flags register is reserved so we can adapt
2952 * that. For offloads, we mark packet as dodgy, so that headers
2953 * need to be verified first.
2954 */
2955 ret = bpf_skb_proto_xlat(skb, proto);
2956 bpf_compute_data_pointers(skb);
2957 return ret;
2958 }
2959
2960 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2961 .func = bpf_skb_change_proto,
2962 .gpl_only = false,
2963 .ret_type = RET_INTEGER,
2964 .arg1_type = ARG_PTR_TO_CTX,
2965 .arg2_type = ARG_ANYTHING,
2966 .arg3_type = ARG_ANYTHING,
2967 };
2968
2969 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2970 {
2971 /* We only allow a restricted subset to be changed for now. */
2972 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2973 !skb_pkt_type_ok(pkt_type)))
2974 return -EINVAL;
2975
2976 skb->pkt_type = pkt_type;
2977 return 0;
2978 }
2979
2980 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2981 .func = bpf_skb_change_type,
2982 .gpl_only = false,
2983 .ret_type = RET_INTEGER,
2984 .arg1_type = ARG_PTR_TO_CTX,
2985 .arg2_type = ARG_ANYTHING,
2986 };
2987
2988 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2989 {
2990 switch (skb->protocol) {
2991 case htons(ETH_P_IP):
2992 return sizeof(struct iphdr);
2993 case htons(ETH_P_IPV6):
2994 return sizeof(struct ipv6hdr);
2995 default:
2996 return ~0U;
2997 }
2998 }
2999
3000 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3001 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3002
3003 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3004 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3005 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3006 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3007 BPF_F_ADJ_ROOM_ENCAP_L2( \
3008 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3009
3010 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3011 u64 flags)
3012 {
3013 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3014 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3015 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3016 unsigned int gso_type = SKB_GSO_DODGY;
3017 int ret;
3018
3019 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3020 /* udp gso_size delineates datagrams, only allow if fixed */
3021 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3022 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3023 return -ENOTSUPP;
3024 }
3025
3026 ret = skb_cow_head(skb, len_diff);
3027 if (unlikely(ret < 0))
3028 return ret;
3029
3030 if (encap) {
3031 if (skb->protocol != htons(ETH_P_IP) &&
3032 skb->protocol != htons(ETH_P_IPV6))
3033 return -ENOTSUPP;
3034
3035 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3036 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3037 return -EINVAL;
3038
3039 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3040 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3041 return -EINVAL;
3042
3043 if (skb->encapsulation)
3044 return -EALREADY;
3045
3046 mac_len = skb->network_header - skb->mac_header;
3047 inner_net = skb->network_header;
3048 if (inner_mac_len > len_diff)
3049 return -EINVAL;
3050 inner_trans = skb->transport_header;
3051 }
3052
3053 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3054 if (unlikely(ret < 0))
3055 return ret;
3056
3057 if (encap) {
3058 skb->inner_mac_header = inner_net - inner_mac_len;
3059 skb->inner_network_header = inner_net;
3060 skb->inner_transport_header = inner_trans;
3061 skb_set_inner_protocol(skb, skb->protocol);
3062
3063 skb->encapsulation = 1;
3064 skb_set_network_header(skb, mac_len);
3065
3066 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3067 gso_type |= SKB_GSO_UDP_TUNNEL;
3068 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3069 gso_type |= SKB_GSO_GRE;
3070 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3071 gso_type |= SKB_GSO_IPXIP6;
3072 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3073 gso_type |= SKB_GSO_IPXIP4;
3074
3075 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3076 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3077 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3078 sizeof(struct ipv6hdr) :
3079 sizeof(struct iphdr);
3080
3081 skb_set_transport_header(skb, mac_len + nh_len);
3082 }
3083
3084 /* Match skb->protocol to new outer l3 protocol */
3085 if (skb->protocol == htons(ETH_P_IP) &&
3086 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3087 skb->protocol = htons(ETH_P_IPV6);
3088 else if (skb->protocol == htons(ETH_P_IPV6) &&
3089 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3090 skb->protocol = htons(ETH_P_IP);
3091 }
3092
3093 if (skb_is_gso(skb)) {
3094 struct skb_shared_info *shinfo = skb_shinfo(skb);
3095
3096 /* Due to header grow, MSS needs to be downgraded. */
3097 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3098 skb_decrease_gso_size(shinfo, len_diff);
3099
3100 /* Header must be checked, and gso_segs recomputed. */
3101 shinfo->gso_type |= gso_type;
3102 shinfo->gso_segs = 0;
3103 }
3104
3105 return 0;
3106 }
3107
3108 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3109 u64 flags)
3110 {
3111 int ret;
3112
3113 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3114 return -EINVAL;
3115
3116 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3117 /* udp gso_size delineates datagrams, only allow if fixed */
3118 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3119 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3120 return -ENOTSUPP;
3121 }
3122
3123 ret = skb_unclone(skb, GFP_ATOMIC);
3124 if (unlikely(ret < 0))
3125 return ret;
3126
3127 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3128 if (unlikely(ret < 0))
3129 return ret;
3130
3131 if (skb_is_gso(skb)) {
3132 struct skb_shared_info *shinfo = skb_shinfo(skb);
3133
3134 /* Due to header shrink, MSS can be upgraded. */
3135 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3136 skb_increase_gso_size(shinfo, len_diff);
3137
3138 /* Header must be checked, and gso_segs recomputed. */
3139 shinfo->gso_type |= SKB_GSO_DODGY;
3140 shinfo->gso_segs = 0;
3141 }
3142
3143 return 0;
3144 }
3145
3146 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3147 {
3148 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3149 SKB_MAX_ALLOC;
3150 }
3151
3152 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3153 u32, mode, u64, flags)
3154 {
3155 u32 len_cur, len_diff_abs = abs(len_diff);
3156 u32 len_min = bpf_skb_net_base_len(skb);
3157 u32 len_max = __bpf_skb_max_len(skb);
3158 __be16 proto = skb->protocol;
3159 bool shrink = len_diff < 0;
3160 u32 off;
3161 int ret;
3162
3163 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3164 return -EINVAL;
3165 if (unlikely(len_diff_abs > 0xfffU))
3166 return -EFAULT;
3167 if (unlikely(proto != htons(ETH_P_IP) &&
3168 proto != htons(ETH_P_IPV6)))
3169 return -ENOTSUPP;
3170
3171 off = skb_mac_header_len(skb);
3172 switch (mode) {
3173 case BPF_ADJ_ROOM_NET:
3174 off += bpf_skb_net_base_len(skb);
3175 break;
3176 case BPF_ADJ_ROOM_MAC:
3177 break;
3178 default:
3179 return -ENOTSUPP;
3180 }
3181
3182 len_cur = skb->len - skb_network_offset(skb);
3183 if ((shrink && (len_diff_abs >= len_cur ||
3184 len_cur - len_diff_abs < len_min)) ||
3185 (!shrink && (skb->len + len_diff_abs > len_max &&
3186 !skb_is_gso(skb))))
3187 return -ENOTSUPP;
3188
3189 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3190 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3191
3192 bpf_compute_data_pointers(skb);
3193 return ret;
3194 }
3195
3196 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3197 .func = bpf_skb_adjust_room,
3198 .gpl_only = false,
3199 .ret_type = RET_INTEGER,
3200 .arg1_type = ARG_PTR_TO_CTX,
3201 .arg2_type = ARG_ANYTHING,
3202 .arg3_type = ARG_ANYTHING,
3203 .arg4_type = ARG_ANYTHING,
3204 };
3205
3206 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3207 {
3208 u32 min_len = skb_network_offset(skb);
3209
3210 if (skb_transport_header_was_set(skb))
3211 min_len = skb_transport_offset(skb);
3212 if (skb->ip_summed == CHECKSUM_PARTIAL)
3213 min_len = skb_checksum_start_offset(skb) +
3214 skb->csum_offset + sizeof(__sum16);
3215 return min_len;
3216 }
3217
3218 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3219 {
3220 unsigned int old_len = skb->len;
3221 int ret;
3222
3223 ret = __skb_grow_rcsum(skb, new_len);
3224 if (!ret)
3225 memset(skb->data + old_len, 0, new_len - old_len);
3226 return ret;
3227 }
3228
3229 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3230 {
3231 return __skb_trim_rcsum(skb, new_len);
3232 }
3233
3234 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3235 u64 flags)
3236 {
3237 u32 max_len = __bpf_skb_max_len(skb);
3238 u32 min_len = __bpf_skb_min_len(skb);
3239 int ret;
3240
3241 if (unlikely(flags || new_len > max_len || new_len < min_len))
3242 return -EINVAL;
3243 if (skb->encapsulation)
3244 return -ENOTSUPP;
3245
3246 /* The basic idea of this helper is that it's performing the
3247 * needed work to either grow or trim an skb, and eBPF program
3248 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3249 * bpf_lX_csum_replace() and others rather than passing a raw
3250 * buffer here. This one is a slow path helper and intended
3251 * for replies with control messages.
3252 *
3253 * Like in bpf_skb_change_proto(), we want to keep this rather
3254 * minimal and without protocol specifics so that we are able
3255 * to separate concerns as in bpf_skb_store_bytes() should only
3256 * be the one responsible for writing buffers.
3257 *
3258 * It's really expected to be a slow path operation here for
3259 * control message replies, so we're implicitly linearizing,
3260 * uncloning and drop offloads from the skb by this.
3261 */
3262 ret = __bpf_try_make_writable(skb, skb->len);
3263 if (!ret) {
3264 if (new_len > skb->len)
3265 ret = bpf_skb_grow_rcsum(skb, new_len);
3266 else if (new_len < skb->len)
3267 ret = bpf_skb_trim_rcsum(skb, new_len);
3268 if (!ret && skb_is_gso(skb))
3269 skb_gso_reset(skb);
3270 }
3271 return ret;
3272 }
3273
3274 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3275 u64, flags)
3276 {
3277 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3278
3279 bpf_compute_data_pointers(skb);
3280 return ret;
3281 }
3282
3283 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3284 .func = bpf_skb_change_tail,
3285 .gpl_only = false,
3286 .ret_type = RET_INTEGER,
3287 .arg1_type = ARG_PTR_TO_CTX,
3288 .arg2_type = ARG_ANYTHING,
3289 .arg3_type = ARG_ANYTHING,
3290 };
3291
3292 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3293 u64, flags)
3294 {
3295 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3296
3297 bpf_compute_data_end_sk_skb(skb);
3298 return ret;
3299 }
3300
3301 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3302 .func = sk_skb_change_tail,
3303 .gpl_only = false,
3304 .ret_type = RET_INTEGER,
3305 .arg1_type = ARG_PTR_TO_CTX,
3306 .arg2_type = ARG_ANYTHING,
3307 .arg3_type = ARG_ANYTHING,
3308 };
3309
3310 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3311 u64 flags)
3312 {
3313 u32 max_len = __bpf_skb_max_len(skb);
3314 u32 new_len = skb->len + head_room;
3315 int ret;
3316
3317 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3318 new_len < skb->len))
3319 return -EINVAL;
3320
3321 ret = skb_cow(skb, head_room);
3322 if (likely(!ret)) {
3323 /* Idea for this helper is that we currently only
3324 * allow to expand on mac header. This means that
3325 * skb->protocol network header, etc, stay as is.
3326 * Compared to bpf_skb_change_tail(), we're more
3327 * flexible due to not needing to linearize or
3328 * reset GSO. Intention for this helper is to be
3329 * used by an L3 skb that needs to push mac header
3330 * for redirection into L2 device.
3331 */
3332 __skb_push(skb, head_room);
3333 memset(skb->data, 0, head_room);
3334 skb_reset_mac_header(skb);
3335 }
3336
3337 return ret;
3338 }
3339
3340 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3341 u64, flags)
3342 {
3343 int ret = __bpf_skb_change_head(skb, head_room, flags);
3344
3345 bpf_compute_data_pointers(skb);
3346 return ret;
3347 }
3348
3349 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3350 .func = bpf_skb_change_head,
3351 .gpl_only = false,
3352 .ret_type = RET_INTEGER,
3353 .arg1_type = ARG_PTR_TO_CTX,
3354 .arg2_type = ARG_ANYTHING,
3355 .arg3_type = ARG_ANYTHING,
3356 };
3357
3358 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3359 u64, flags)
3360 {
3361 int ret = __bpf_skb_change_head(skb, head_room, flags);
3362
3363 bpf_compute_data_end_sk_skb(skb);
3364 return ret;
3365 }
3366
3367 static const struct bpf_func_proto sk_skb_change_head_proto = {
3368 .func = sk_skb_change_head,
3369 .gpl_only = false,
3370 .ret_type = RET_INTEGER,
3371 .arg1_type = ARG_PTR_TO_CTX,
3372 .arg2_type = ARG_ANYTHING,
3373 .arg3_type = ARG_ANYTHING,
3374 };
3375 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3376 {
3377 return xdp_data_meta_unsupported(xdp) ? 0 :
3378 xdp->data - xdp->data_meta;
3379 }
3380
3381 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3382 {
3383 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3384 unsigned long metalen = xdp_get_metalen(xdp);
3385 void *data_start = xdp_frame_end + metalen;
3386 void *data = xdp->data + offset;
3387
3388 if (unlikely(data < data_start ||
3389 data > xdp->data_end - ETH_HLEN))
3390 return -EINVAL;
3391
3392 if (metalen)
3393 memmove(xdp->data_meta + offset,
3394 xdp->data_meta, metalen);
3395 xdp->data_meta += offset;
3396 xdp->data = data;
3397
3398 return 0;
3399 }
3400
3401 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3402 .func = bpf_xdp_adjust_head,
3403 .gpl_only = false,
3404 .ret_type = RET_INTEGER,
3405 .arg1_type = ARG_PTR_TO_CTX,
3406 .arg2_type = ARG_ANYTHING,
3407 };
3408
3409 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3410 {
3411 void *data_end = xdp->data_end + offset;
3412
3413 /* only shrinking is allowed for now. */
3414 if (unlikely(offset >= 0))
3415 return -EINVAL;
3416
3417 if (unlikely(data_end < xdp->data + ETH_HLEN))
3418 return -EINVAL;
3419
3420 xdp->data_end = data_end;
3421
3422 return 0;
3423 }
3424
3425 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3426 .func = bpf_xdp_adjust_tail,
3427 .gpl_only = false,
3428 .ret_type = RET_INTEGER,
3429 .arg1_type = ARG_PTR_TO_CTX,
3430 .arg2_type = ARG_ANYTHING,
3431 };
3432
3433 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3434 {
3435 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3436 void *meta = xdp->data_meta + offset;
3437 unsigned long metalen = xdp->data - meta;
3438
3439 if (xdp_data_meta_unsupported(xdp))
3440 return -ENOTSUPP;
3441 if (unlikely(meta < xdp_frame_end ||
3442 meta > xdp->data))
3443 return -EINVAL;
3444 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3445 (metalen > 32)))
3446 return -EACCES;
3447
3448 xdp->data_meta = meta;
3449
3450 return 0;
3451 }
3452
3453 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3454 .func = bpf_xdp_adjust_meta,
3455 .gpl_only = false,
3456 .ret_type = RET_INTEGER,
3457 .arg1_type = ARG_PTR_TO_CTX,
3458 .arg2_type = ARG_ANYTHING,
3459 };
3460
3461 static int __bpf_tx_xdp(struct net_device *dev,
3462 struct bpf_map *map,
3463 struct xdp_buff *xdp,
3464 u32 index)
3465 {
3466 struct xdp_frame *xdpf;
3467 int err, sent;
3468
3469 if (!dev->netdev_ops->ndo_xdp_xmit) {
3470 return -EOPNOTSUPP;
3471 }
3472
3473 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3474 if (unlikely(err))
3475 return err;
3476
3477 xdpf = convert_to_xdp_frame(xdp);
3478 if (unlikely(!xdpf))
3479 return -EOVERFLOW;
3480
3481 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3482 if (sent <= 0)
3483 return sent;
3484 return 0;
3485 }
3486
3487 static noinline int
3488 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3489 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3490 {
3491 struct net_device *fwd;
3492 u32 index = ri->tgt_index;
3493 int err;
3494
3495 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3496 ri->tgt_index = 0;
3497 if (unlikely(!fwd)) {
3498 err = -EINVAL;
3499 goto err;
3500 }
3501
3502 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3503 if (unlikely(err))
3504 goto err;
3505
3506 _trace_xdp_redirect(dev, xdp_prog, index);
3507 return 0;
3508 err:
3509 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3510 return err;
3511 }
3512
3513 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3514 struct bpf_map *map,
3515 struct xdp_buff *xdp,
3516 u32 index)
3517 {
3518 int err;
3519
3520 switch (map->map_type) {
3521 case BPF_MAP_TYPE_DEVMAP: {
3522 struct bpf_dtab_netdev *dst = fwd;
3523
3524 err = dev_map_enqueue(dst, xdp, dev_rx);
3525 if (unlikely(err))
3526 return err;
3527 break;
3528 }
3529 case BPF_MAP_TYPE_CPUMAP: {
3530 struct bpf_cpu_map_entry *rcpu = fwd;
3531
3532 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3533 if (unlikely(err))
3534 return err;
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
3549 void 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 }
3571 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3572
3573 static 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
3587 void 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 /* Avoid polluting remote cacheline due to writes if
3595 * not needed. Once we pass this test, we need the
3596 * cmpxchg() to make sure it hasn't been changed in
3597 * the meantime by remote CPU.
3598 */
3599 if (unlikely(READ_ONCE(ri->map) == map))
3600 cmpxchg(&ri->map, map, NULL);
3601 }
3602 }
3603
3604 static 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->tgt_index;
3609 void *fwd = ri->tgt_value;
3610 int err;
3611
3612 ri->tgt_index = 0;
3613 ri->tgt_value = NULL;
3614 WRITE_ONCE(ri->map, NULL);
3615
3616 if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3617 xdp_do_flush_map();
3618
3619 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3620 if (unlikely(err))
3621 goto err;
3622
3623 ri->map_to_flush = map;
3624 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3625 return 0;
3626 err:
3627 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3628 return err;
3629 }
3630
3631 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3632 struct bpf_prog *xdp_prog)
3633 {
3634 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3635 struct bpf_map *map = READ_ONCE(ri->map);
3636
3637 if (likely(map))
3638 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3639
3640 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3641 }
3642 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3643
3644 static int xdp_do_generic_redirect_map(struct net_device *dev,
3645 struct sk_buff *skb,
3646 struct xdp_buff *xdp,
3647 struct bpf_prog *xdp_prog,
3648 struct bpf_map *map)
3649 {
3650 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3651 u32 index = ri->tgt_index;
3652 void *fwd = ri->tgt_value;
3653 int err = 0;
3654
3655 ri->tgt_index = 0;
3656 ri->tgt_value = NULL;
3657 WRITE_ONCE(ri->map, NULL);
3658
3659 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3660 struct bpf_dtab_netdev *dst = fwd;
3661
3662 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3663 if (unlikely(err))
3664 goto err;
3665 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3666 struct xdp_sock *xs = fwd;
3667
3668 err = xsk_generic_rcv(xs, xdp);
3669 if (err)
3670 goto err;
3671 consume_skb(skb);
3672 } else {
3673 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3674 err = -EBADRQC;
3675 goto err;
3676 }
3677
3678 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3679 return 0;
3680 err:
3681 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3682 return err;
3683 }
3684
3685 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3686 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3687 {
3688 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3689 struct bpf_map *map = READ_ONCE(ri->map);
3690 u32 index = ri->tgt_index;
3691 struct net_device *fwd;
3692 int err = 0;
3693
3694 if (map)
3695 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3696 map);
3697 ri->tgt_index = 0;
3698 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3699 if (unlikely(!fwd)) {
3700 err = -EINVAL;
3701 goto err;
3702 }
3703
3704 err = xdp_ok_fwd_dev(fwd, skb->len);
3705 if (unlikely(err))
3706 goto err;
3707
3708 skb->dev = fwd;
3709 _trace_xdp_redirect(dev, xdp_prog, index);
3710 generic_xdp_tx(skb, xdp_prog);
3711 return 0;
3712 err:
3713 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3714 return err;
3715 }
3716 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3717
3718 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3719 {
3720 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3721
3722 if (unlikely(flags))
3723 return XDP_ABORTED;
3724
3725 ri->flags = flags;
3726 ri->tgt_index = ifindex;
3727 ri->tgt_value = NULL;
3728 WRITE_ONCE(ri->map, NULL);
3729
3730 return XDP_REDIRECT;
3731 }
3732
3733 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3734 .func = bpf_xdp_redirect,
3735 .gpl_only = false,
3736 .ret_type = RET_INTEGER,
3737 .arg1_type = ARG_ANYTHING,
3738 .arg2_type = ARG_ANYTHING,
3739 };
3740
3741 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3742 u64, flags)
3743 {
3744 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3745
3746 /* Lower bits of the flags are used as return code on lookup failure */
3747 if (unlikely(flags > XDP_TX))
3748 return XDP_ABORTED;
3749
3750 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
3751 if (unlikely(!ri->tgt_value)) {
3752 /* If the lookup fails we want to clear out the state in the
3753 * redirect_info struct completely, so that if an eBPF program
3754 * performs multiple lookups, the last one always takes
3755 * precedence.
3756 */
3757 WRITE_ONCE(ri->map, NULL);
3758 return flags;
3759 }
3760
3761 ri->flags = flags;
3762 ri->tgt_index = ifindex;
3763 WRITE_ONCE(ri->map, map);
3764
3765 return XDP_REDIRECT;
3766 }
3767
3768 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3769 .func = bpf_xdp_redirect_map,
3770 .gpl_only = false,
3771 .ret_type = RET_INTEGER,
3772 .arg1_type = ARG_CONST_MAP_PTR,
3773 .arg2_type = ARG_ANYTHING,
3774 .arg3_type = ARG_ANYTHING,
3775 };
3776
3777 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3778 unsigned long off, unsigned long len)
3779 {
3780 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3781
3782 if (unlikely(!ptr))
3783 return len;
3784 if (ptr != dst_buff)
3785 memcpy(dst_buff, ptr, len);
3786
3787 return 0;
3788 }
3789
3790 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3791 u64, flags, void *, meta, u64, meta_size)
3792 {
3793 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3794
3795 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3796 return -EINVAL;
3797 if (unlikely(skb_size > skb->len))
3798 return -EFAULT;
3799
3800 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3801 bpf_skb_copy);
3802 }
3803
3804 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3805 .func = bpf_skb_event_output,
3806 .gpl_only = true,
3807 .ret_type = RET_INTEGER,
3808 .arg1_type = ARG_PTR_TO_CTX,
3809 .arg2_type = ARG_CONST_MAP_PTR,
3810 .arg3_type = ARG_ANYTHING,
3811 .arg4_type = ARG_PTR_TO_MEM,
3812 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3813 };
3814
3815 static unsigned short bpf_tunnel_key_af(u64 flags)
3816 {
3817 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3818 }
3819
3820 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3821 u32, size, u64, flags)
3822 {
3823 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3824 u8 compat[sizeof(struct bpf_tunnel_key)];
3825 void *to_orig = to;
3826 int err;
3827
3828 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3829 err = -EINVAL;
3830 goto err_clear;
3831 }
3832 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3833 err = -EPROTO;
3834 goto err_clear;
3835 }
3836 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3837 err = -EINVAL;
3838 switch (size) {
3839 case offsetof(struct bpf_tunnel_key, tunnel_label):
3840 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3841 goto set_compat;
3842 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3843 /* Fixup deprecated structure layouts here, so we have
3844 * a common path later on.
3845 */
3846 if (ip_tunnel_info_af(info) != AF_INET)
3847 goto err_clear;
3848 set_compat:
3849 to = (struct bpf_tunnel_key *)compat;
3850 break;
3851 default:
3852 goto err_clear;
3853 }
3854 }
3855
3856 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3857 to->tunnel_tos = info->key.tos;
3858 to->tunnel_ttl = info->key.ttl;
3859 to->tunnel_ext = 0;
3860
3861 if (flags & BPF_F_TUNINFO_IPV6) {
3862 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3863 sizeof(to->remote_ipv6));
3864 to->tunnel_label = be32_to_cpu(info->key.label);
3865 } else {
3866 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3867 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3868 to->tunnel_label = 0;
3869 }
3870
3871 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3872 memcpy(to_orig, to, size);
3873
3874 return 0;
3875 err_clear:
3876 memset(to_orig, 0, size);
3877 return err;
3878 }
3879
3880 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3881 .func = bpf_skb_get_tunnel_key,
3882 .gpl_only = false,
3883 .ret_type = RET_INTEGER,
3884 .arg1_type = ARG_PTR_TO_CTX,
3885 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3886 .arg3_type = ARG_CONST_SIZE,
3887 .arg4_type = ARG_ANYTHING,
3888 };
3889
3890 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3891 {
3892 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3893 int err;
3894
3895 if (unlikely(!info ||
3896 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3897 err = -ENOENT;
3898 goto err_clear;
3899 }
3900 if (unlikely(size < info->options_len)) {
3901 err = -ENOMEM;
3902 goto err_clear;
3903 }
3904
3905 ip_tunnel_info_opts_get(to, info);
3906 if (size > info->options_len)
3907 memset(to + info->options_len, 0, size - info->options_len);
3908
3909 return info->options_len;
3910 err_clear:
3911 memset(to, 0, size);
3912 return err;
3913 }
3914
3915 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3916 .func = bpf_skb_get_tunnel_opt,
3917 .gpl_only = false,
3918 .ret_type = RET_INTEGER,
3919 .arg1_type = ARG_PTR_TO_CTX,
3920 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3921 .arg3_type = ARG_CONST_SIZE,
3922 };
3923
3924 static struct metadata_dst __percpu *md_dst;
3925
3926 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3927 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3928 {
3929 struct metadata_dst *md = this_cpu_ptr(md_dst);
3930 u8 compat[sizeof(struct bpf_tunnel_key)];
3931 struct ip_tunnel_info *info;
3932
3933 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3934 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3935 return -EINVAL;
3936 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3937 switch (size) {
3938 case offsetof(struct bpf_tunnel_key, tunnel_label):
3939 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3940 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3941 /* Fixup deprecated structure layouts here, so we have
3942 * a common path later on.
3943 */
3944 memcpy(compat, from, size);
3945 memset(compat + size, 0, sizeof(compat) - size);
3946 from = (const struct bpf_tunnel_key *) compat;
3947 break;
3948 default:
3949 return -EINVAL;
3950 }
3951 }
3952 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3953 from->tunnel_ext))
3954 return -EINVAL;
3955
3956 skb_dst_drop(skb);
3957 dst_hold((struct dst_entry *) md);
3958 skb_dst_set(skb, (struct dst_entry *) md);
3959
3960 info = &md->u.tun_info;
3961 memset(info, 0, sizeof(*info));
3962 info->mode = IP_TUNNEL_INFO_TX;
3963
3964 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3965 if (flags & BPF_F_DONT_FRAGMENT)
3966 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3967 if (flags & BPF_F_ZERO_CSUM_TX)
3968 info->key.tun_flags &= ~TUNNEL_CSUM;
3969 if (flags & BPF_F_SEQ_NUMBER)
3970 info->key.tun_flags |= TUNNEL_SEQ;
3971
3972 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3973 info->key.tos = from->tunnel_tos;
3974 info->key.ttl = from->tunnel_ttl;
3975
3976 if (flags & BPF_F_TUNINFO_IPV6) {
3977 info->mode |= IP_TUNNEL_INFO_IPV6;
3978 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3979 sizeof(from->remote_ipv6));
3980 info->key.label = cpu_to_be32(from->tunnel_label) &
3981 IPV6_FLOWLABEL_MASK;
3982 } else {
3983 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3984 }
3985
3986 return 0;
3987 }
3988
3989 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3990 .func = bpf_skb_set_tunnel_key,
3991 .gpl_only = false,
3992 .ret_type = RET_INTEGER,
3993 .arg1_type = ARG_PTR_TO_CTX,
3994 .arg2_type = ARG_PTR_TO_MEM,
3995 .arg3_type = ARG_CONST_SIZE,
3996 .arg4_type = ARG_ANYTHING,
3997 };
3998
3999 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4000 const u8 *, from, u32, size)
4001 {
4002 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4003 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4004
4005 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4006 return -EINVAL;
4007 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4008 return -ENOMEM;
4009
4010 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4011
4012 return 0;
4013 }
4014
4015 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4016 .func = bpf_skb_set_tunnel_opt,
4017 .gpl_only = false,
4018 .ret_type = RET_INTEGER,
4019 .arg1_type = ARG_PTR_TO_CTX,
4020 .arg2_type = ARG_PTR_TO_MEM,
4021 .arg3_type = ARG_CONST_SIZE,
4022 };
4023
4024 static const struct bpf_func_proto *
4025 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4026 {
4027 if (!md_dst) {
4028 struct metadata_dst __percpu *tmp;
4029
4030 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4031 METADATA_IP_TUNNEL,
4032 GFP_KERNEL);
4033 if (!tmp)
4034 return NULL;
4035 if (cmpxchg(&md_dst, NULL, tmp))
4036 metadata_dst_free_percpu(tmp);
4037 }
4038
4039 switch (which) {
4040 case BPF_FUNC_skb_set_tunnel_key:
4041 return &bpf_skb_set_tunnel_key_proto;
4042 case BPF_FUNC_skb_set_tunnel_opt:
4043 return &bpf_skb_set_tunnel_opt_proto;
4044 default:
4045 return NULL;
4046 }
4047 }
4048
4049 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4050 u32, idx)
4051 {
4052 struct bpf_array *array = container_of(map, struct bpf_array, map);
4053 struct cgroup *cgrp;
4054 struct sock *sk;
4055
4056 sk = skb_to_full_sk(skb);
4057 if (!sk || !sk_fullsock(sk))
4058 return -ENOENT;
4059 if (unlikely(idx >= array->map.max_entries))
4060 return -E2BIG;
4061
4062 cgrp = READ_ONCE(array->ptrs[idx]);
4063 if (unlikely(!cgrp))
4064 return -EAGAIN;
4065
4066 return sk_under_cgroup_hierarchy(sk, cgrp);
4067 }
4068
4069 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4070 .func = bpf_skb_under_cgroup,
4071 .gpl_only = false,
4072 .ret_type = RET_INTEGER,
4073 .arg1_type = ARG_PTR_TO_CTX,
4074 .arg2_type = ARG_CONST_MAP_PTR,
4075 .arg3_type = ARG_ANYTHING,
4076 };
4077
4078 #ifdef CONFIG_SOCK_CGROUP_DATA
4079 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4080 {
4081 struct sock *sk = skb_to_full_sk(skb);
4082 struct cgroup *cgrp;
4083
4084 if (!sk || !sk_fullsock(sk))
4085 return 0;
4086
4087 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4088 return cgrp->kn->id.id;
4089 }
4090
4091 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4092 .func = bpf_skb_cgroup_id,
4093 .gpl_only = false,
4094 .ret_type = RET_INTEGER,
4095 .arg1_type = ARG_PTR_TO_CTX,
4096 };
4097
4098 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4099 ancestor_level)
4100 {
4101 struct sock *sk = skb_to_full_sk(skb);
4102 struct cgroup *ancestor;
4103 struct cgroup *cgrp;
4104
4105 if (!sk || !sk_fullsock(sk))
4106 return 0;
4107
4108 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4109 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4110 if (!ancestor)
4111 return 0;
4112
4113 return ancestor->kn->id.id;
4114 }
4115
4116 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4117 .func = bpf_skb_ancestor_cgroup_id,
4118 .gpl_only = false,
4119 .ret_type = RET_INTEGER,
4120 .arg1_type = ARG_PTR_TO_CTX,
4121 .arg2_type = ARG_ANYTHING,
4122 };
4123 #endif
4124
4125 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4126 unsigned long off, unsigned long len)
4127 {
4128 memcpy(dst_buff, src_buff + off, len);
4129 return 0;
4130 }
4131
4132 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4133 u64, flags, void *, meta, u64, meta_size)
4134 {
4135 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4136
4137 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4138 return -EINVAL;
4139 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4140 return -EFAULT;
4141
4142 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4143 xdp_size, bpf_xdp_copy);
4144 }
4145
4146 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4147 .func = bpf_xdp_event_output,
4148 .gpl_only = true,
4149 .ret_type = RET_INTEGER,
4150 .arg1_type = ARG_PTR_TO_CTX,
4151 .arg2_type = ARG_CONST_MAP_PTR,
4152 .arg3_type = ARG_ANYTHING,
4153 .arg4_type = ARG_PTR_TO_MEM,
4154 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4155 };
4156
4157 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4158 {
4159 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4160 }
4161
4162 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4163 .func = bpf_get_socket_cookie,
4164 .gpl_only = false,
4165 .ret_type = RET_INTEGER,
4166 .arg1_type = ARG_PTR_TO_CTX,
4167 };
4168
4169 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4170 {
4171 return sock_gen_cookie(ctx->sk);
4172 }
4173
4174 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4175 .func = bpf_get_socket_cookie_sock_addr,
4176 .gpl_only = false,
4177 .ret_type = RET_INTEGER,
4178 .arg1_type = ARG_PTR_TO_CTX,
4179 };
4180
4181 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4182 {
4183 return sock_gen_cookie(ctx->sk);
4184 }
4185
4186 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4187 .func = bpf_get_socket_cookie_sock_ops,
4188 .gpl_only = false,
4189 .ret_type = RET_INTEGER,
4190 .arg1_type = ARG_PTR_TO_CTX,
4191 };
4192
4193 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4194 {
4195 struct sock *sk = sk_to_full_sk(skb->sk);
4196 kuid_t kuid;
4197
4198 if (!sk || !sk_fullsock(sk))
4199 return overflowuid;
4200 kuid = sock_net_uid(sock_net(sk), sk);
4201 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4202 }
4203
4204 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4205 .func = bpf_get_socket_uid,
4206 .gpl_only = false,
4207 .ret_type = RET_INTEGER,
4208 .arg1_type = ARG_PTR_TO_CTX,
4209 };
4210
4211 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4212 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4213 {
4214 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4215 return -EINVAL;
4216
4217 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4218 }
4219
4220 static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4221 .func = bpf_sockopt_event_output,
4222 .gpl_only = true,
4223 .ret_type = RET_INTEGER,
4224 .arg1_type = ARG_PTR_TO_CTX,
4225 .arg2_type = ARG_CONST_MAP_PTR,
4226 .arg3_type = ARG_ANYTHING,
4227 .arg4_type = ARG_PTR_TO_MEM,
4228 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4229 };
4230
4231 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4232 int, level, int, optname, char *, optval, int, optlen)
4233 {
4234 struct sock *sk = bpf_sock->sk;
4235 int ret = 0;
4236 int val;
4237
4238 if (!sk_fullsock(sk))
4239 return -EINVAL;
4240
4241 if (level == SOL_SOCKET) {
4242 if (optlen != sizeof(int))
4243 return -EINVAL;
4244 val = *((int *)optval);
4245
4246 /* Only some socketops are supported */
4247 switch (optname) {
4248 case SO_RCVBUF:
4249 val = min_t(u32, val, sysctl_rmem_max);
4250 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4251 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4252 break;
4253 case SO_SNDBUF:
4254 val = min_t(u32, val, sysctl_wmem_max);
4255 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4256 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4257 break;
4258 case SO_MAX_PACING_RATE: /* 32bit version */
4259 if (val != ~0U)
4260 cmpxchg(&sk->sk_pacing_status,
4261 SK_PACING_NONE,
4262 SK_PACING_NEEDED);
4263 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4264 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4265 sk->sk_max_pacing_rate);
4266 break;
4267 case SO_PRIORITY:
4268 sk->sk_priority = val;
4269 break;
4270 case SO_RCVLOWAT:
4271 if (val < 0)
4272 val = INT_MAX;
4273 sk->sk_rcvlowat = val ? : 1;
4274 break;
4275 case SO_MARK:
4276 if (sk->sk_mark != val) {
4277 sk->sk_mark = val;
4278 sk_dst_reset(sk);
4279 }
4280 break;
4281 default:
4282 ret = -EINVAL;
4283 }
4284 #ifdef CONFIG_INET
4285 } else if (level == SOL_IP) {
4286 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4287 return -EINVAL;
4288
4289 val = *((int *)optval);
4290 /* Only some options are supported */
4291 switch (optname) {
4292 case IP_TOS:
4293 if (val < -1 || val > 0xff) {
4294 ret = -EINVAL;
4295 } else {
4296 struct inet_sock *inet = inet_sk(sk);
4297
4298 if (val == -1)
4299 val = 0;
4300 inet->tos = val;
4301 }
4302 break;
4303 default:
4304 ret = -EINVAL;
4305 }
4306 #if IS_ENABLED(CONFIG_IPV6)
4307 } else if (level == SOL_IPV6) {
4308 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4309 return -EINVAL;
4310
4311 val = *((int *)optval);
4312 /* Only some options are supported */
4313 switch (optname) {
4314 case IPV6_TCLASS:
4315 if (val < -1 || val > 0xff) {
4316 ret = -EINVAL;
4317 } else {
4318 struct ipv6_pinfo *np = inet6_sk(sk);
4319
4320 if (val == -1)
4321 val = 0;
4322 np->tclass = val;
4323 }
4324 break;
4325 default:
4326 ret = -EINVAL;
4327 }
4328 #endif
4329 } else if (level == SOL_TCP &&
4330 sk->sk_prot->setsockopt == tcp_setsockopt) {
4331 if (optname == TCP_CONGESTION) {
4332 char name[TCP_CA_NAME_MAX];
4333 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4334
4335 strncpy(name, optval, min_t(long, optlen,
4336 TCP_CA_NAME_MAX-1));
4337 name[TCP_CA_NAME_MAX-1] = 0;
4338 ret = tcp_set_congestion_control(sk, name, false,
4339 reinit, true);
4340 } else {
4341 struct tcp_sock *tp = tcp_sk(sk);
4342
4343 if (optlen != sizeof(int))
4344 return -EINVAL;
4345
4346 val = *((int *)optval);
4347 /* Only some options are supported */
4348 switch (optname) {
4349 case TCP_BPF_IW:
4350 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4351 ret = -EINVAL;
4352 else
4353 tp->snd_cwnd = val;
4354 break;
4355 case TCP_BPF_SNDCWND_CLAMP:
4356 if (val <= 0) {
4357 ret = -EINVAL;
4358 } else {
4359 tp->snd_cwnd_clamp = val;
4360 tp->snd_ssthresh = val;
4361 }
4362 break;
4363 case TCP_SAVE_SYN:
4364 if (val < 0 || val > 1)
4365 ret = -EINVAL;
4366 else
4367 tp->save_syn = val;
4368 break;
4369 default:
4370 ret = -EINVAL;
4371 }
4372 }
4373 #endif
4374 } else {
4375 ret = -EINVAL;
4376 }
4377 return ret;
4378 }
4379
4380 static const struct bpf_func_proto bpf_setsockopt_proto = {
4381 .func = bpf_setsockopt,
4382 .gpl_only = false,
4383 .ret_type = RET_INTEGER,
4384 .arg1_type = ARG_PTR_TO_CTX,
4385 .arg2_type = ARG_ANYTHING,
4386 .arg3_type = ARG_ANYTHING,
4387 .arg4_type = ARG_PTR_TO_MEM,
4388 .arg5_type = ARG_CONST_SIZE,
4389 };
4390
4391 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4392 int, level, int, optname, char *, optval, int, optlen)
4393 {
4394 struct sock *sk = bpf_sock->sk;
4395
4396 if (!sk_fullsock(sk))
4397 goto err_clear;
4398 #ifdef CONFIG_INET
4399 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4400 struct inet_connection_sock *icsk;
4401 struct tcp_sock *tp;
4402
4403 switch (optname) {
4404 case TCP_CONGESTION:
4405 icsk = inet_csk(sk);
4406
4407 if (!icsk->icsk_ca_ops || optlen <= 1)
4408 goto err_clear;
4409 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4410 optval[optlen - 1] = 0;
4411 break;
4412 case TCP_SAVED_SYN:
4413 tp = tcp_sk(sk);
4414
4415 if (optlen <= 0 || !tp->saved_syn ||
4416 optlen > tp->saved_syn[0])
4417 goto err_clear;
4418 memcpy(optval, tp->saved_syn + 1, optlen);
4419 break;
4420 default:
4421 goto err_clear;
4422 }
4423 } else if (level == SOL_IP) {
4424 struct inet_sock *inet = inet_sk(sk);
4425
4426 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4427 goto err_clear;
4428
4429 /* Only some options are supported */
4430 switch (optname) {
4431 case IP_TOS:
4432 *((int *)optval) = (int)inet->tos;
4433 break;
4434 default:
4435 goto err_clear;
4436 }
4437 #if IS_ENABLED(CONFIG_IPV6)
4438 } else if (level == SOL_IPV6) {
4439 struct ipv6_pinfo *np = inet6_sk(sk);
4440
4441 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4442 goto err_clear;
4443
4444 /* Only some options are supported */
4445 switch (optname) {
4446 case IPV6_TCLASS:
4447 *((int *)optval) = (int)np->tclass;
4448 break;
4449 default:
4450 goto err_clear;
4451 }
4452 #endif
4453 } else {
4454 goto err_clear;
4455 }
4456 return 0;
4457 #endif
4458 err_clear:
4459 memset(optval, 0, optlen);
4460 return -EINVAL;
4461 }
4462
4463 static const struct bpf_func_proto bpf_getsockopt_proto = {
4464 .func = bpf_getsockopt,
4465 .gpl_only = false,
4466 .ret_type = RET_INTEGER,
4467 .arg1_type = ARG_PTR_TO_CTX,
4468 .arg2_type = ARG_ANYTHING,
4469 .arg3_type = ARG_ANYTHING,
4470 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4471 .arg5_type = ARG_CONST_SIZE,
4472 };
4473
4474 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4475 int, argval)
4476 {
4477 struct sock *sk = bpf_sock->sk;
4478 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4479
4480 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4481 return -EINVAL;
4482
4483 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4484
4485 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4486 }
4487
4488 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4489 .func = bpf_sock_ops_cb_flags_set,
4490 .gpl_only = false,
4491 .ret_type = RET_INTEGER,
4492 .arg1_type = ARG_PTR_TO_CTX,
4493 .arg2_type = ARG_ANYTHING,
4494 };
4495
4496 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4497 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4498
4499 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4500 int, addr_len)
4501 {
4502 #ifdef CONFIG_INET
4503 struct sock *sk = ctx->sk;
4504 int err;
4505
4506 /* Binding to port can be expensive so it's prohibited in the helper.
4507 * Only binding to IP is supported.
4508 */
4509 err = -EINVAL;
4510 if (addr_len < offsetofend(struct sockaddr, sa_family))
4511 return err;
4512 if (addr->sa_family == AF_INET) {
4513 if (addr_len < sizeof(struct sockaddr_in))
4514 return err;
4515 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4516 return err;
4517 return __inet_bind(sk, addr, addr_len, true, false);
4518 #if IS_ENABLED(CONFIG_IPV6)
4519 } else if (addr->sa_family == AF_INET6) {
4520 if (addr_len < SIN6_LEN_RFC2133)
4521 return err;
4522 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4523 return err;
4524 /* ipv6_bpf_stub cannot be NULL, since it's called from
4525 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4526 */
4527 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4528 #endif /* CONFIG_IPV6 */
4529 }
4530 #endif /* CONFIG_INET */
4531
4532 return -EAFNOSUPPORT;
4533 }
4534
4535 static const struct bpf_func_proto bpf_bind_proto = {
4536 .func = bpf_bind,
4537 .gpl_only = false,
4538 .ret_type = RET_INTEGER,
4539 .arg1_type = ARG_PTR_TO_CTX,
4540 .arg2_type = ARG_PTR_TO_MEM,
4541 .arg3_type = ARG_CONST_SIZE,
4542 };
4543
4544 #ifdef CONFIG_XFRM
4545 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4546 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4547 {
4548 const struct sec_path *sp = skb_sec_path(skb);
4549 const struct xfrm_state *x;
4550
4551 if (!sp || unlikely(index >= sp->len || flags))
4552 goto err_clear;
4553
4554 x = sp->xvec[index];
4555
4556 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4557 goto err_clear;
4558
4559 to->reqid = x->props.reqid;
4560 to->spi = x->id.spi;
4561 to->family = x->props.family;
4562 to->ext = 0;
4563
4564 if (to->family == AF_INET6) {
4565 memcpy(to->remote_ipv6, x->props.saddr.a6,
4566 sizeof(to->remote_ipv6));
4567 } else {
4568 to->remote_ipv4 = x->props.saddr.a4;
4569 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4570 }
4571
4572 return 0;
4573 err_clear:
4574 memset(to, 0, size);
4575 return -EINVAL;
4576 }
4577
4578 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4579 .func = bpf_skb_get_xfrm_state,
4580 .gpl_only = false,
4581 .ret_type = RET_INTEGER,
4582 .arg1_type = ARG_PTR_TO_CTX,
4583 .arg2_type = ARG_ANYTHING,
4584 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4585 .arg4_type = ARG_CONST_SIZE,
4586 .arg5_type = ARG_ANYTHING,
4587 };
4588 #endif
4589
4590 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4591 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4592 const struct neighbour *neigh,
4593 const struct net_device *dev)
4594 {
4595 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4596 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4597 params->h_vlan_TCI = 0;
4598 params->h_vlan_proto = 0;
4599 params->ifindex = dev->ifindex;
4600
4601 return 0;
4602 }
4603 #endif
4604
4605 #if IS_ENABLED(CONFIG_INET)
4606 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4607 u32 flags, bool check_mtu)
4608 {
4609 struct fib_nh_common *nhc;
4610 struct in_device *in_dev;
4611 struct neighbour *neigh;
4612 struct net_device *dev;
4613 struct fib_result res;
4614 struct flowi4 fl4;
4615 int err;
4616 u32 mtu;
4617
4618 dev = dev_get_by_index_rcu(net, params->ifindex);
4619 if (unlikely(!dev))
4620 return -ENODEV;
4621
4622 /* verify forwarding is enabled on this interface */
4623 in_dev = __in_dev_get_rcu(dev);
4624 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4625 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4626
4627 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4628 fl4.flowi4_iif = 1;
4629 fl4.flowi4_oif = params->ifindex;
4630 } else {
4631 fl4.flowi4_iif = params->ifindex;
4632 fl4.flowi4_oif = 0;
4633 }
4634 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4635 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4636 fl4.flowi4_flags = 0;
4637
4638 fl4.flowi4_proto = params->l4_protocol;
4639 fl4.daddr = params->ipv4_dst;
4640 fl4.saddr = params->ipv4_src;
4641 fl4.fl4_sport = params->sport;
4642 fl4.fl4_dport = params->dport;
4643
4644 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4645 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4646 struct fib_table *tb;
4647
4648 tb = fib_get_table(net, tbid);
4649 if (unlikely(!tb))
4650 return BPF_FIB_LKUP_RET_NOT_FWDED;
4651
4652 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4653 } else {
4654 fl4.flowi4_mark = 0;
4655 fl4.flowi4_secid = 0;
4656 fl4.flowi4_tun_key.tun_id = 0;
4657 fl4.flowi4_uid = sock_net_uid(net, NULL);
4658
4659 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4660 }
4661
4662 if (err) {
4663 /* map fib lookup errors to RTN_ type */
4664 if (err == -EINVAL)
4665 return BPF_FIB_LKUP_RET_BLACKHOLE;
4666 if (err == -EHOSTUNREACH)
4667 return BPF_FIB_LKUP_RET_UNREACHABLE;
4668 if (err == -EACCES)
4669 return BPF_FIB_LKUP_RET_PROHIBIT;
4670
4671 return BPF_FIB_LKUP_RET_NOT_FWDED;
4672 }
4673
4674 if (res.type != RTN_UNICAST)
4675 return BPF_FIB_LKUP_RET_NOT_FWDED;
4676
4677 if (fib_info_num_path(res.fi) > 1)
4678 fib_select_path(net, &res, &fl4, NULL);
4679
4680 if (check_mtu) {
4681 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4682 if (params->tot_len > mtu)
4683 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4684 }
4685
4686 nhc = res.nhc;
4687
4688 /* do not handle lwt encaps right now */
4689 if (nhc->nhc_lwtstate)
4690 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4691
4692 dev = nhc->nhc_dev;
4693
4694 params->rt_metric = res.fi->fib_priority;
4695
4696 /* xdp and cls_bpf programs are run in RCU-bh so
4697 * rcu_read_lock_bh is not needed here
4698 */
4699 if (likely(nhc->nhc_gw_family != AF_INET6)) {
4700 if (nhc->nhc_gw_family)
4701 params->ipv4_dst = nhc->nhc_gw.ipv4;
4702
4703 neigh = __ipv4_neigh_lookup_noref(dev,
4704 (__force u32)params->ipv4_dst);
4705 } else {
4706 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4707
4708 params->family = AF_INET6;
4709 *dst = nhc->nhc_gw.ipv6;
4710 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4711 }
4712
4713 if (!neigh)
4714 return BPF_FIB_LKUP_RET_NO_NEIGH;
4715
4716 return bpf_fib_set_fwd_params(params, neigh, dev);
4717 }
4718 #endif
4719
4720 #if IS_ENABLED(CONFIG_IPV6)
4721 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4722 u32 flags, bool check_mtu)
4723 {
4724 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4725 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4726 struct fib6_result res = {};
4727 struct neighbour *neigh;
4728 struct net_device *dev;
4729 struct inet6_dev *idev;
4730 struct flowi6 fl6;
4731 int strict = 0;
4732 int oif, err;
4733 u32 mtu;
4734
4735 /* link local addresses are never forwarded */
4736 if (rt6_need_strict(dst) || rt6_need_strict(src))
4737 return BPF_FIB_LKUP_RET_NOT_FWDED;
4738
4739 dev = dev_get_by_index_rcu(net, params->ifindex);
4740 if (unlikely(!dev))
4741 return -ENODEV;
4742
4743 idev = __in6_dev_get_safely(dev);
4744 if (unlikely(!idev || !idev->cnf.forwarding))
4745 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4746
4747 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4748 fl6.flowi6_iif = 1;
4749 oif = fl6.flowi6_oif = params->ifindex;
4750 } else {
4751 oif = fl6.flowi6_iif = params->ifindex;
4752 fl6.flowi6_oif = 0;
4753 strict = RT6_LOOKUP_F_HAS_SADDR;
4754 }
4755 fl6.flowlabel = params->flowinfo;
4756 fl6.flowi6_scope = 0;
4757 fl6.flowi6_flags = 0;
4758 fl6.mp_hash = 0;
4759
4760 fl6.flowi6_proto = params->l4_protocol;
4761 fl6.daddr = *dst;
4762 fl6.saddr = *src;
4763 fl6.fl6_sport = params->sport;
4764 fl6.fl6_dport = params->dport;
4765
4766 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4767 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4768 struct fib6_table *tb;
4769
4770 tb = ipv6_stub->fib6_get_table(net, tbid);
4771 if (unlikely(!tb))
4772 return BPF_FIB_LKUP_RET_NOT_FWDED;
4773
4774 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4775 strict);
4776 } else {
4777 fl6.flowi6_mark = 0;
4778 fl6.flowi6_secid = 0;
4779 fl6.flowi6_tun_key.tun_id = 0;
4780 fl6.flowi6_uid = sock_net_uid(net, NULL);
4781
4782 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4783 }
4784
4785 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4786 res.f6i == net->ipv6.fib6_null_entry))
4787 return BPF_FIB_LKUP_RET_NOT_FWDED;
4788
4789 switch (res.fib6_type) {
4790 /* only unicast is forwarded */
4791 case RTN_UNICAST:
4792 break;
4793 case RTN_BLACKHOLE:
4794 return BPF_FIB_LKUP_RET_BLACKHOLE;
4795 case RTN_UNREACHABLE:
4796 return BPF_FIB_LKUP_RET_UNREACHABLE;
4797 case RTN_PROHIBIT:
4798 return BPF_FIB_LKUP_RET_PROHIBIT;
4799 default:
4800 return BPF_FIB_LKUP_RET_NOT_FWDED;
4801 }
4802
4803 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4804 fl6.flowi6_oif != 0, NULL, strict);
4805
4806 if (check_mtu) {
4807 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4808 if (params->tot_len > mtu)
4809 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4810 }
4811
4812 if (res.nh->fib_nh_lws)
4813 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4814
4815 if (res.nh->fib_nh_gw_family)
4816 *dst = res.nh->fib_nh_gw6;
4817
4818 dev = res.nh->fib_nh_dev;
4819 params->rt_metric = res.f6i->fib6_metric;
4820
4821 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4822 * not needed here.
4823 */
4824 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4825 if (!neigh)
4826 return BPF_FIB_LKUP_RET_NO_NEIGH;
4827
4828 return bpf_fib_set_fwd_params(params, neigh, dev);
4829 }
4830 #endif
4831
4832 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4833 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4834 {
4835 if (plen < sizeof(*params))
4836 return -EINVAL;
4837
4838 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4839 return -EINVAL;
4840
4841 switch (params->family) {
4842 #if IS_ENABLED(CONFIG_INET)
4843 case AF_INET:
4844 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4845 flags, true);
4846 #endif
4847 #if IS_ENABLED(CONFIG_IPV6)
4848 case AF_INET6:
4849 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4850 flags, true);
4851 #endif
4852 }
4853 return -EAFNOSUPPORT;
4854 }
4855
4856 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4857 .func = bpf_xdp_fib_lookup,
4858 .gpl_only = true,
4859 .ret_type = RET_INTEGER,
4860 .arg1_type = ARG_PTR_TO_CTX,
4861 .arg2_type = ARG_PTR_TO_MEM,
4862 .arg3_type = ARG_CONST_SIZE,
4863 .arg4_type = ARG_ANYTHING,
4864 };
4865
4866 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4867 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4868 {
4869 struct net *net = dev_net(skb->dev);
4870 int rc = -EAFNOSUPPORT;
4871
4872 if (plen < sizeof(*params))
4873 return -EINVAL;
4874
4875 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4876 return -EINVAL;
4877
4878 switch (params->family) {
4879 #if IS_ENABLED(CONFIG_INET)
4880 case AF_INET:
4881 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4882 break;
4883 #endif
4884 #if IS_ENABLED(CONFIG_IPV6)
4885 case AF_INET6:
4886 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4887 break;
4888 #endif
4889 }
4890
4891 if (!rc) {
4892 struct net_device *dev;
4893
4894 dev = dev_get_by_index_rcu(net, params->ifindex);
4895 if (!is_skb_forwardable(dev, skb))
4896 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4897 }
4898
4899 return rc;
4900 }
4901
4902 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4903 .func = bpf_skb_fib_lookup,
4904 .gpl_only = true,
4905 .ret_type = RET_INTEGER,
4906 .arg1_type = ARG_PTR_TO_CTX,
4907 .arg2_type = ARG_PTR_TO_MEM,
4908 .arg3_type = ARG_CONST_SIZE,
4909 .arg4_type = ARG_ANYTHING,
4910 };
4911
4912 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4913 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4914 {
4915 int err;
4916 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4917
4918 if (!seg6_validate_srh(srh, len))
4919 return -EINVAL;
4920
4921 switch (type) {
4922 case BPF_LWT_ENCAP_SEG6_INLINE:
4923 if (skb->protocol != htons(ETH_P_IPV6))
4924 return -EBADMSG;
4925
4926 err = seg6_do_srh_inline(skb, srh);
4927 break;
4928 case BPF_LWT_ENCAP_SEG6:
4929 skb_reset_inner_headers(skb);
4930 skb->encapsulation = 1;
4931 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4932 break;
4933 default:
4934 return -EINVAL;
4935 }
4936
4937 bpf_compute_data_pointers(skb);
4938 if (err)
4939 return err;
4940
4941 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4942 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4943
4944 return seg6_lookup_nexthop(skb, NULL, 0);
4945 }
4946 #endif /* CONFIG_IPV6_SEG6_BPF */
4947
4948 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4949 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4950 bool ingress)
4951 {
4952 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4953 }
4954 #endif
4955
4956 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4957 u32, len)
4958 {
4959 switch (type) {
4960 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4961 case BPF_LWT_ENCAP_SEG6:
4962 case BPF_LWT_ENCAP_SEG6_INLINE:
4963 return bpf_push_seg6_encap(skb, type, hdr, len);
4964 #endif
4965 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4966 case BPF_LWT_ENCAP_IP:
4967 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4968 #endif
4969 default:
4970 return -EINVAL;
4971 }
4972 }
4973
4974 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4975 void *, hdr, u32, len)
4976 {
4977 switch (type) {
4978 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4979 case BPF_LWT_ENCAP_IP:
4980 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4981 #endif
4982 default:
4983 return -EINVAL;
4984 }
4985 }
4986
4987 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4988 .func = bpf_lwt_in_push_encap,
4989 .gpl_only = false,
4990 .ret_type = RET_INTEGER,
4991 .arg1_type = ARG_PTR_TO_CTX,
4992 .arg2_type = ARG_ANYTHING,
4993 .arg3_type = ARG_PTR_TO_MEM,
4994 .arg4_type = ARG_CONST_SIZE
4995 };
4996
4997 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4998 .func = bpf_lwt_xmit_push_encap,
4999 .gpl_only = false,
5000 .ret_type = RET_INTEGER,
5001 .arg1_type = ARG_PTR_TO_CTX,
5002 .arg2_type = ARG_ANYTHING,
5003 .arg3_type = ARG_PTR_TO_MEM,
5004 .arg4_type = ARG_CONST_SIZE
5005 };
5006
5007 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5008 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5009 const void *, from, u32, len)
5010 {
5011 struct seg6_bpf_srh_state *srh_state =
5012 this_cpu_ptr(&seg6_bpf_srh_states);
5013 struct ipv6_sr_hdr *srh = srh_state->srh;
5014 void *srh_tlvs, *srh_end, *ptr;
5015 int srhoff = 0;
5016
5017 if (srh == NULL)
5018 return -EINVAL;
5019
5020 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5021 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5022
5023 ptr = skb->data + offset;
5024 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5025 srh_state->valid = false;
5026 else if (ptr < (void *)&srh->flags ||
5027 ptr + len > (void *)&srh->segments)
5028 return -EFAULT;
5029
5030 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5031 return -EFAULT;
5032 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5033 return -EINVAL;
5034 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5035
5036 memcpy(skb->data + offset, from, len);
5037 return 0;
5038 }
5039
5040 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5041 .func = bpf_lwt_seg6_store_bytes,
5042 .gpl_only = false,
5043 .ret_type = RET_INTEGER,
5044 .arg1_type = ARG_PTR_TO_CTX,
5045 .arg2_type = ARG_ANYTHING,
5046 .arg3_type = ARG_PTR_TO_MEM,
5047 .arg4_type = ARG_CONST_SIZE
5048 };
5049
5050 static void bpf_update_srh_state(struct sk_buff *skb)
5051 {
5052 struct seg6_bpf_srh_state *srh_state =
5053 this_cpu_ptr(&seg6_bpf_srh_states);
5054 int srhoff = 0;
5055
5056 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5057 srh_state->srh = NULL;
5058 } else {
5059 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5060 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5061 srh_state->valid = true;
5062 }
5063 }
5064
5065 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5066 u32, action, void *, param, u32, param_len)
5067 {
5068 struct seg6_bpf_srh_state *srh_state =
5069 this_cpu_ptr(&seg6_bpf_srh_states);
5070 int hdroff = 0;
5071 int err;
5072
5073 switch (action) {
5074 case SEG6_LOCAL_ACTION_END_X:
5075 if (!seg6_bpf_has_valid_srh(skb))
5076 return -EBADMSG;
5077 if (param_len != sizeof(struct in6_addr))
5078 return -EINVAL;
5079 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5080 case SEG6_LOCAL_ACTION_END_T:
5081 if (!seg6_bpf_has_valid_srh(skb))
5082 return -EBADMSG;
5083 if (param_len != sizeof(int))
5084 return -EINVAL;
5085 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5086 case SEG6_LOCAL_ACTION_END_DT6:
5087 if (!seg6_bpf_has_valid_srh(skb))
5088 return -EBADMSG;
5089 if (param_len != sizeof(int))
5090 return -EINVAL;
5091
5092 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5093 return -EBADMSG;
5094 if (!pskb_pull(skb, hdroff))
5095 return -EBADMSG;
5096
5097 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5098 skb_reset_network_header(skb);
5099 skb_reset_transport_header(skb);
5100 skb->encapsulation = 0;
5101
5102 bpf_compute_data_pointers(skb);
5103 bpf_update_srh_state(skb);
5104 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5105 case SEG6_LOCAL_ACTION_END_B6:
5106 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5107 return -EBADMSG;
5108 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5109 param, param_len);
5110 if (!err)
5111 bpf_update_srh_state(skb);
5112
5113 return err;
5114 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5115 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5116 return -EBADMSG;
5117 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5118 param, param_len);
5119 if (!err)
5120 bpf_update_srh_state(skb);
5121
5122 return err;
5123 default:
5124 return -EINVAL;
5125 }
5126 }
5127
5128 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5129 .func = bpf_lwt_seg6_action,
5130 .gpl_only = false,
5131 .ret_type = RET_INTEGER,
5132 .arg1_type = ARG_PTR_TO_CTX,
5133 .arg2_type = ARG_ANYTHING,
5134 .arg3_type = ARG_PTR_TO_MEM,
5135 .arg4_type = ARG_CONST_SIZE
5136 };
5137
5138 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5139 s32, len)
5140 {
5141 struct seg6_bpf_srh_state *srh_state =
5142 this_cpu_ptr(&seg6_bpf_srh_states);
5143 struct ipv6_sr_hdr *srh = srh_state->srh;
5144 void *srh_end, *srh_tlvs, *ptr;
5145 struct ipv6hdr *hdr;
5146 int srhoff = 0;
5147 int ret;
5148
5149 if (unlikely(srh == NULL))
5150 return -EINVAL;
5151
5152 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5153 ((srh->first_segment + 1) << 4));
5154 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5155 srh_state->hdrlen);
5156 ptr = skb->data + offset;
5157
5158 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5159 return -EFAULT;
5160 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5161 return -EFAULT;
5162
5163 if (len > 0) {
5164 ret = skb_cow_head(skb, len);
5165 if (unlikely(ret < 0))
5166 return ret;
5167
5168 ret = bpf_skb_net_hdr_push(skb, offset, len);
5169 } else {
5170 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5171 }
5172
5173 bpf_compute_data_pointers(skb);
5174 if (unlikely(ret < 0))
5175 return ret;
5176
5177 hdr = (struct ipv6hdr *)skb->data;
5178 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5179
5180 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5181 return -EINVAL;
5182 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5183 srh_state->hdrlen += len;
5184 srh_state->valid = false;
5185 return 0;
5186 }
5187
5188 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5189 .func = bpf_lwt_seg6_adjust_srh,
5190 .gpl_only = false,
5191 .ret_type = RET_INTEGER,
5192 .arg1_type = ARG_PTR_TO_CTX,
5193 .arg2_type = ARG_ANYTHING,
5194 .arg3_type = ARG_ANYTHING,
5195 };
5196 #endif /* CONFIG_IPV6_SEG6_BPF */
5197
5198 #ifdef CONFIG_INET
5199 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5200 int dif, int sdif, u8 family, u8 proto)
5201 {
5202 bool refcounted = false;
5203 struct sock *sk = NULL;
5204
5205 if (family == AF_INET) {
5206 __be32 src4 = tuple->ipv4.saddr;
5207 __be32 dst4 = tuple->ipv4.daddr;
5208
5209 if (proto == IPPROTO_TCP)
5210 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5211 src4, tuple->ipv4.sport,
5212 dst4, tuple->ipv4.dport,
5213 dif, sdif, &refcounted);
5214 else
5215 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5216 dst4, tuple->ipv4.dport,
5217 dif, sdif, &udp_table, NULL);
5218 #if IS_ENABLED(CONFIG_IPV6)
5219 } else {
5220 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5221 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5222
5223 if (proto == IPPROTO_TCP)
5224 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5225 src6, tuple->ipv6.sport,
5226 dst6, ntohs(tuple->ipv6.dport),
5227 dif, sdif, &refcounted);
5228 else if (likely(ipv6_bpf_stub))
5229 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5230 src6, tuple->ipv6.sport,
5231 dst6, tuple->ipv6.dport,
5232 dif, sdif,
5233 &udp_table, NULL);
5234 #endif
5235 }
5236
5237 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5238 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5239 sk = NULL;
5240 }
5241 return sk;
5242 }
5243
5244 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5245 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5246 * Returns the socket as an 'unsigned long' to simplify the casting in the
5247 * callers to satisfy BPF_CALL declarations.
5248 */
5249 static struct sock *
5250 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5251 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5252 u64 flags)
5253 {
5254 struct sock *sk = NULL;
5255 u8 family = AF_UNSPEC;
5256 struct net *net;
5257 int sdif;
5258
5259 if (len == sizeof(tuple->ipv4))
5260 family = AF_INET;
5261 else if (len == sizeof(tuple->ipv6))
5262 family = AF_INET6;
5263 else
5264 return NULL;
5265
5266 if (unlikely(family == AF_UNSPEC || flags ||
5267 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5268 goto out;
5269
5270 if (family == AF_INET)
5271 sdif = inet_sdif(skb);
5272 else
5273 sdif = inet6_sdif(skb);
5274
5275 if ((s32)netns_id < 0) {
5276 net = caller_net;
5277 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5278 } else {
5279 net = get_net_ns_by_id(caller_net, netns_id);
5280 if (unlikely(!net))
5281 goto out;
5282 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5283 put_net(net);
5284 }
5285
5286 out:
5287 return sk;
5288 }
5289
5290 static struct sock *
5291 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5292 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5293 u64 flags)
5294 {
5295 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5296 ifindex, proto, netns_id, flags);
5297
5298 if (sk) {
5299 sk = sk_to_full_sk(sk);
5300 if (!sk_fullsock(sk)) {
5301 if (!sock_flag(sk, SOCK_RCU_FREE))
5302 sock_gen_put(sk);
5303 return NULL;
5304 }
5305 }
5306
5307 return sk;
5308 }
5309
5310 static struct sock *
5311 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5312 u8 proto, u64 netns_id, u64 flags)
5313 {
5314 struct net *caller_net;
5315 int ifindex;
5316
5317 if (skb->dev) {
5318 caller_net = dev_net(skb->dev);
5319 ifindex = skb->dev->ifindex;
5320 } else {
5321 caller_net = sock_net(skb->sk);
5322 ifindex = 0;
5323 }
5324
5325 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5326 netns_id, flags);
5327 }
5328
5329 static struct sock *
5330 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5331 u8 proto, u64 netns_id, u64 flags)
5332 {
5333 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5334 flags);
5335
5336 if (sk) {
5337 sk = sk_to_full_sk(sk);
5338 if (!sk_fullsock(sk)) {
5339 if (!sock_flag(sk, SOCK_RCU_FREE))
5340 sock_gen_put(sk);
5341 return NULL;
5342 }
5343 }
5344
5345 return sk;
5346 }
5347
5348 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5349 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5350 {
5351 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5352 netns_id, flags);
5353 }
5354
5355 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5356 .func = bpf_skc_lookup_tcp,
5357 .gpl_only = false,
5358 .pkt_access = true,
5359 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5360 .arg1_type = ARG_PTR_TO_CTX,
5361 .arg2_type = ARG_PTR_TO_MEM,
5362 .arg3_type = ARG_CONST_SIZE,
5363 .arg4_type = ARG_ANYTHING,
5364 .arg5_type = ARG_ANYTHING,
5365 };
5366
5367 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5368 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5369 {
5370 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5371 netns_id, flags);
5372 }
5373
5374 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5375 .func = bpf_sk_lookup_tcp,
5376 .gpl_only = false,
5377 .pkt_access = true,
5378 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5379 .arg1_type = ARG_PTR_TO_CTX,
5380 .arg2_type = ARG_PTR_TO_MEM,
5381 .arg3_type = ARG_CONST_SIZE,
5382 .arg4_type = ARG_ANYTHING,
5383 .arg5_type = ARG_ANYTHING,
5384 };
5385
5386 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5387 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5388 {
5389 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5390 netns_id, flags);
5391 }
5392
5393 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5394 .func = bpf_sk_lookup_udp,
5395 .gpl_only = false,
5396 .pkt_access = true,
5397 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5398 .arg1_type = ARG_PTR_TO_CTX,
5399 .arg2_type = ARG_PTR_TO_MEM,
5400 .arg3_type = ARG_CONST_SIZE,
5401 .arg4_type = ARG_ANYTHING,
5402 .arg5_type = ARG_ANYTHING,
5403 };
5404
5405 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5406 {
5407 if (!sock_flag(sk, SOCK_RCU_FREE))
5408 sock_gen_put(sk);
5409 return 0;
5410 }
5411
5412 static const struct bpf_func_proto bpf_sk_release_proto = {
5413 .func = bpf_sk_release,
5414 .gpl_only = false,
5415 .ret_type = RET_INTEGER,
5416 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5417 };
5418
5419 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5420 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5421 {
5422 struct net *caller_net = dev_net(ctx->rxq->dev);
5423 int ifindex = ctx->rxq->dev->ifindex;
5424
5425 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5426 ifindex, IPPROTO_UDP, netns_id,
5427 flags);
5428 }
5429
5430 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5431 .func = bpf_xdp_sk_lookup_udp,
5432 .gpl_only = false,
5433 .pkt_access = true,
5434 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5435 .arg1_type = ARG_PTR_TO_CTX,
5436 .arg2_type = ARG_PTR_TO_MEM,
5437 .arg3_type = ARG_CONST_SIZE,
5438 .arg4_type = ARG_ANYTHING,
5439 .arg5_type = ARG_ANYTHING,
5440 };
5441
5442 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5443 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5444 {
5445 struct net *caller_net = dev_net(ctx->rxq->dev);
5446 int ifindex = ctx->rxq->dev->ifindex;
5447
5448 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5449 ifindex, IPPROTO_TCP, netns_id,
5450 flags);
5451 }
5452
5453 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5454 .func = bpf_xdp_skc_lookup_tcp,
5455 .gpl_only = false,
5456 .pkt_access = true,
5457 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5458 .arg1_type = ARG_PTR_TO_CTX,
5459 .arg2_type = ARG_PTR_TO_MEM,
5460 .arg3_type = ARG_CONST_SIZE,
5461 .arg4_type = ARG_ANYTHING,
5462 .arg5_type = ARG_ANYTHING,
5463 };
5464
5465 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5466 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5467 {
5468 struct net *caller_net = dev_net(ctx->rxq->dev);
5469 int ifindex = ctx->rxq->dev->ifindex;
5470
5471 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5472 ifindex, IPPROTO_TCP, netns_id,
5473 flags);
5474 }
5475
5476 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5477 .func = bpf_xdp_sk_lookup_tcp,
5478 .gpl_only = false,
5479 .pkt_access = true,
5480 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5481 .arg1_type = ARG_PTR_TO_CTX,
5482 .arg2_type = ARG_PTR_TO_MEM,
5483 .arg3_type = ARG_CONST_SIZE,
5484 .arg4_type = ARG_ANYTHING,
5485 .arg5_type = ARG_ANYTHING,
5486 };
5487
5488 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5489 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5490 {
5491 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5492 sock_net(ctx->sk), 0,
5493 IPPROTO_TCP, netns_id, flags);
5494 }
5495
5496 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5497 .func = bpf_sock_addr_skc_lookup_tcp,
5498 .gpl_only = false,
5499 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5500 .arg1_type = ARG_PTR_TO_CTX,
5501 .arg2_type = ARG_PTR_TO_MEM,
5502 .arg3_type = ARG_CONST_SIZE,
5503 .arg4_type = ARG_ANYTHING,
5504 .arg5_type = ARG_ANYTHING,
5505 };
5506
5507 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5508 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5509 {
5510 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5511 sock_net(ctx->sk), 0, IPPROTO_TCP,
5512 netns_id, flags);
5513 }
5514
5515 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5516 .func = bpf_sock_addr_sk_lookup_tcp,
5517 .gpl_only = false,
5518 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5519 .arg1_type = ARG_PTR_TO_CTX,
5520 .arg2_type = ARG_PTR_TO_MEM,
5521 .arg3_type = ARG_CONST_SIZE,
5522 .arg4_type = ARG_ANYTHING,
5523 .arg5_type = ARG_ANYTHING,
5524 };
5525
5526 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5527 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5528 {
5529 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5530 sock_net(ctx->sk), 0, IPPROTO_UDP,
5531 netns_id, flags);
5532 }
5533
5534 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5535 .func = bpf_sock_addr_sk_lookup_udp,
5536 .gpl_only = false,
5537 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5538 .arg1_type = ARG_PTR_TO_CTX,
5539 .arg2_type = ARG_PTR_TO_MEM,
5540 .arg3_type = ARG_CONST_SIZE,
5541 .arg4_type = ARG_ANYTHING,
5542 .arg5_type = ARG_ANYTHING,
5543 };
5544
5545 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5546 struct bpf_insn_access_aux *info)
5547 {
5548 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
5549 icsk_retransmits))
5550 return false;
5551
5552 if (off % size != 0)
5553 return false;
5554
5555 switch (off) {
5556 case offsetof(struct bpf_tcp_sock, bytes_received):
5557 case offsetof(struct bpf_tcp_sock, bytes_acked):
5558 return size == sizeof(__u64);
5559 default:
5560 return size == sizeof(__u32);
5561 }
5562 }
5563
5564 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5565 const struct bpf_insn *si,
5566 struct bpf_insn *insn_buf,
5567 struct bpf_prog *prog, u32 *target_size)
5568 {
5569 struct bpf_insn *insn = insn_buf;
5570
5571 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
5572 do { \
5573 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) > \
5574 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5575 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5576 si->dst_reg, si->src_reg, \
5577 offsetof(struct tcp_sock, FIELD)); \
5578 } while (0)
5579
5580 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
5581 do { \
5582 BUILD_BUG_ON(FIELD_SIZEOF(struct inet_connection_sock, \
5583 FIELD) > \
5584 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5585 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
5586 struct inet_connection_sock, \
5587 FIELD), \
5588 si->dst_reg, si->src_reg, \
5589 offsetof( \
5590 struct inet_connection_sock, \
5591 FIELD)); \
5592 } while (0)
5593
5594 if (insn > insn_buf)
5595 return insn - insn_buf;
5596
5597 switch (si->off) {
5598 case offsetof(struct bpf_tcp_sock, rtt_min):
5599 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5600 sizeof(struct minmax));
5601 BUILD_BUG_ON(sizeof(struct minmax) <
5602 sizeof(struct minmax_sample));
5603
5604 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5605 offsetof(struct tcp_sock, rtt_min) +
5606 offsetof(struct minmax_sample, v));
5607 break;
5608 case offsetof(struct bpf_tcp_sock, snd_cwnd):
5609 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
5610 break;
5611 case offsetof(struct bpf_tcp_sock, srtt_us):
5612 BPF_TCP_SOCK_GET_COMMON(srtt_us);
5613 break;
5614 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
5615 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
5616 break;
5617 case offsetof(struct bpf_tcp_sock, rcv_nxt):
5618 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
5619 break;
5620 case offsetof(struct bpf_tcp_sock, snd_nxt):
5621 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
5622 break;
5623 case offsetof(struct bpf_tcp_sock, snd_una):
5624 BPF_TCP_SOCK_GET_COMMON(snd_una);
5625 break;
5626 case offsetof(struct bpf_tcp_sock, mss_cache):
5627 BPF_TCP_SOCK_GET_COMMON(mss_cache);
5628 break;
5629 case offsetof(struct bpf_tcp_sock, ecn_flags):
5630 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
5631 break;
5632 case offsetof(struct bpf_tcp_sock, rate_delivered):
5633 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
5634 break;
5635 case offsetof(struct bpf_tcp_sock, rate_interval_us):
5636 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
5637 break;
5638 case offsetof(struct bpf_tcp_sock, packets_out):
5639 BPF_TCP_SOCK_GET_COMMON(packets_out);
5640 break;
5641 case offsetof(struct bpf_tcp_sock, retrans_out):
5642 BPF_TCP_SOCK_GET_COMMON(retrans_out);
5643 break;
5644 case offsetof(struct bpf_tcp_sock, total_retrans):
5645 BPF_TCP_SOCK_GET_COMMON(total_retrans);
5646 break;
5647 case offsetof(struct bpf_tcp_sock, segs_in):
5648 BPF_TCP_SOCK_GET_COMMON(segs_in);
5649 break;
5650 case offsetof(struct bpf_tcp_sock, data_segs_in):
5651 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
5652 break;
5653 case offsetof(struct bpf_tcp_sock, segs_out):
5654 BPF_TCP_SOCK_GET_COMMON(segs_out);
5655 break;
5656 case offsetof(struct bpf_tcp_sock, data_segs_out):
5657 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
5658 break;
5659 case offsetof(struct bpf_tcp_sock, lost_out):
5660 BPF_TCP_SOCK_GET_COMMON(lost_out);
5661 break;
5662 case offsetof(struct bpf_tcp_sock, sacked_out):
5663 BPF_TCP_SOCK_GET_COMMON(sacked_out);
5664 break;
5665 case offsetof(struct bpf_tcp_sock, bytes_received):
5666 BPF_TCP_SOCK_GET_COMMON(bytes_received);
5667 break;
5668 case offsetof(struct bpf_tcp_sock, bytes_acked):
5669 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
5670 break;
5671 case offsetof(struct bpf_tcp_sock, dsack_dups):
5672 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
5673 break;
5674 case offsetof(struct bpf_tcp_sock, delivered):
5675 BPF_TCP_SOCK_GET_COMMON(delivered);
5676 break;
5677 case offsetof(struct bpf_tcp_sock, delivered_ce):
5678 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
5679 break;
5680 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
5681 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
5682 break;
5683 }
5684
5685 return insn - insn_buf;
5686 }
5687
5688 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5689 {
5690 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5691 return (unsigned long)sk;
5692
5693 return (unsigned long)NULL;
5694 }
5695
5696 const struct bpf_func_proto bpf_tcp_sock_proto = {
5697 .func = bpf_tcp_sock,
5698 .gpl_only = false,
5699 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5700 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5701 };
5702
5703 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5704 {
5705 sk = sk_to_full_sk(sk);
5706
5707 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5708 return (unsigned long)sk;
5709
5710 return (unsigned long)NULL;
5711 }
5712
5713 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5714 .func = bpf_get_listener_sock,
5715 .gpl_only = false,
5716 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5717 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5718 };
5719
5720 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5721 {
5722 unsigned int iphdr_len;
5723
5724 if (skb->protocol == cpu_to_be16(ETH_P_IP))
5725 iphdr_len = sizeof(struct iphdr);
5726 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5727 iphdr_len = sizeof(struct ipv6hdr);
5728 else
5729 return 0;
5730
5731 if (skb_headlen(skb) < iphdr_len)
5732 return 0;
5733
5734 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5735 return 0;
5736
5737 return INET_ECN_set_ce(skb);
5738 }
5739
5740 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5741 struct bpf_insn_access_aux *info)
5742 {
5743 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
5744 return false;
5745
5746 if (off % size != 0)
5747 return false;
5748
5749 switch (off) {
5750 default:
5751 return size == sizeof(__u32);
5752 }
5753 }
5754
5755 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
5756 const struct bpf_insn *si,
5757 struct bpf_insn *insn_buf,
5758 struct bpf_prog *prog, u32 *target_size)
5759 {
5760 struct bpf_insn *insn = insn_buf;
5761
5762 #define BPF_XDP_SOCK_GET(FIELD) \
5763 do { \
5764 BUILD_BUG_ON(FIELD_SIZEOF(struct xdp_sock, FIELD) > \
5765 FIELD_SIZEOF(struct bpf_xdp_sock, FIELD)); \
5766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
5767 si->dst_reg, si->src_reg, \
5768 offsetof(struct xdp_sock, FIELD)); \
5769 } while (0)
5770
5771 switch (si->off) {
5772 case offsetof(struct bpf_xdp_sock, queue_id):
5773 BPF_XDP_SOCK_GET(queue_id);
5774 break;
5775 }
5776
5777 return insn - insn_buf;
5778 }
5779
5780 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5781 .func = bpf_skb_ecn_set_ce,
5782 .gpl_only = false,
5783 .ret_type = RET_INTEGER,
5784 .arg1_type = ARG_PTR_TO_CTX,
5785 };
5786
5787 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5788 struct tcphdr *, th, u32, th_len)
5789 {
5790 #ifdef CONFIG_SYN_COOKIES
5791 u32 cookie;
5792 int ret;
5793
5794 if (unlikely(th_len < sizeof(*th)))
5795 return -EINVAL;
5796
5797 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
5798 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5799 return -EINVAL;
5800
5801 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5802 return -EINVAL;
5803
5804 if (!th->ack || th->rst || th->syn)
5805 return -ENOENT;
5806
5807 if (tcp_synq_no_recent_overflow(sk))
5808 return -ENOENT;
5809
5810 cookie = ntohl(th->ack_seq) - 1;
5811
5812 switch (sk->sk_family) {
5813 case AF_INET:
5814 if (unlikely(iph_len < sizeof(struct iphdr)))
5815 return -EINVAL;
5816
5817 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5818 break;
5819
5820 #if IS_BUILTIN(CONFIG_IPV6)
5821 case AF_INET6:
5822 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5823 return -EINVAL;
5824
5825 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5826 break;
5827 #endif /* CONFIG_IPV6 */
5828
5829 default:
5830 return -EPROTONOSUPPORT;
5831 }
5832
5833 if (ret > 0)
5834 return 0;
5835
5836 return -ENOENT;
5837 #else
5838 return -ENOTSUPP;
5839 #endif
5840 }
5841
5842 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5843 .func = bpf_tcp_check_syncookie,
5844 .gpl_only = true,
5845 .pkt_access = true,
5846 .ret_type = RET_INTEGER,
5847 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5848 .arg2_type = ARG_PTR_TO_MEM,
5849 .arg3_type = ARG_CONST_SIZE,
5850 .arg4_type = ARG_PTR_TO_MEM,
5851 .arg5_type = ARG_CONST_SIZE,
5852 };
5853
5854 #endif /* CONFIG_INET */
5855
5856 bool bpf_helper_changes_pkt_data(void *func)
5857 {
5858 if (func == bpf_skb_vlan_push ||
5859 func == bpf_skb_vlan_pop ||
5860 func == bpf_skb_store_bytes ||
5861 func == bpf_skb_change_proto ||
5862 func == bpf_skb_change_head ||
5863 func == sk_skb_change_head ||
5864 func == bpf_skb_change_tail ||
5865 func == sk_skb_change_tail ||
5866 func == bpf_skb_adjust_room ||
5867 func == bpf_skb_pull_data ||
5868 func == sk_skb_pull_data ||
5869 func == bpf_clone_redirect ||
5870 func == bpf_l3_csum_replace ||
5871 func == bpf_l4_csum_replace ||
5872 func == bpf_xdp_adjust_head ||
5873 func == bpf_xdp_adjust_meta ||
5874 func == bpf_msg_pull_data ||
5875 func == bpf_msg_push_data ||
5876 func == bpf_msg_pop_data ||
5877 func == bpf_xdp_adjust_tail ||
5878 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5879 func == bpf_lwt_seg6_store_bytes ||
5880 func == bpf_lwt_seg6_adjust_srh ||
5881 func == bpf_lwt_seg6_action ||
5882 #endif
5883 func == bpf_lwt_in_push_encap ||
5884 func == bpf_lwt_xmit_push_encap)
5885 return true;
5886
5887 return false;
5888 }
5889
5890 static const struct bpf_func_proto *
5891 bpf_base_func_proto(enum bpf_func_id func_id)
5892 {
5893 switch (func_id) {
5894 case BPF_FUNC_map_lookup_elem:
5895 return &bpf_map_lookup_elem_proto;
5896 case BPF_FUNC_map_update_elem:
5897 return &bpf_map_update_elem_proto;
5898 case BPF_FUNC_map_delete_elem:
5899 return &bpf_map_delete_elem_proto;
5900 case BPF_FUNC_map_push_elem:
5901 return &bpf_map_push_elem_proto;
5902 case BPF_FUNC_map_pop_elem:
5903 return &bpf_map_pop_elem_proto;
5904 case BPF_FUNC_map_peek_elem:
5905 return &bpf_map_peek_elem_proto;
5906 case BPF_FUNC_get_prandom_u32:
5907 return &bpf_get_prandom_u32_proto;
5908 case BPF_FUNC_get_smp_processor_id:
5909 return &bpf_get_raw_smp_processor_id_proto;
5910 case BPF_FUNC_get_numa_node_id:
5911 return &bpf_get_numa_node_id_proto;
5912 case BPF_FUNC_tail_call:
5913 return &bpf_tail_call_proto;
5914 case BPF_FUNC_ktime_get_ns:
5915 return &bpf_ktime_get_ns_proto;
5916 default:
5917 break;
5918 }
5919
5920 if (!capable(CAP_SYS_ADMIN))
5921 return NULL;
5922
5923 switch (func_id) {
5924 case BPF_FUNC_spin_lock:
5925 return &bpf_spin_lock_proto;
5926 case BPF_FUNC_spin_unlock:
5927 return &bpf_spin_unlock_proto;
5928 case BPF_FUNC_trace_printk:
5929 return bpf_get_trace_printk_proto();
5930 default:
5931 return NULL;
5932 }
5933 }
5934
5935 static const struct bpf_func_proto *
5936 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5937 {
5938 switch (func_id) {
5939 /* inet and inet6 sockets are created in a process
5940 * context so there is always a valid uid/gid
5941 */
5942 case BPF_FUNC_get_current_uid_gid:
5943 return &bpf_get_current_uid_gid_proto;
5944 case BPF_FUNC_get_local_storage:
5945 return &bpf_get_local_storage_proto;
5946 default:
5947 return bpf_base_func_proto(func_id);
5948 }
5949 }
5950
5951 static const struct bpf_func_proto *
5952 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5953 {
5954 switch (func_id) {
5955 /* inet and inet6 sockets are created in a process
5956 * context so there is always a valid uid/gid
5957 */
5958 case BPF_FUNC_get_current_uid_gid:
5959 return &bpf_get_current_uid_gid_proto;
5960 case BPF_FUNC_bind:
5961 switch (prog->expected_attach_type) {
5962 case BPF_CGROUP_INET4_CONNECT:
5963 case BPF_CGROUP_INET6_CONNECT:
5964 return &bpf_bind_proto;
5965 default:
5966 return NULL;
5967 }
5968 case BPF_FUNC_get_socket_cookie:
5969 return &bpf_get_socket_cookie_sock_addr_proto;
5970 case BPF_FUNC_get_local_storage:
5971 return &bpf_get_local_storage_proto;
5972 #ifdef CONFIG_INET
5973 case BPF_FUNC_sk_lookup_tcp:
5974 return &bpf_sock_addr_sk_lookup_tcp_proto;
5975 case BPF_FUNC_sk_lookup_udp:
5976 return &bpf_sock_addr_sk_lookup_udp_proto;
5977 case BPF_FUNC_sk_release:
5978 return &bpf_sk_release_proto;
5979 case BPF_FUNC_skc_lookup_tcp:
5980 return &bpf_sock_addr_skc_lookup_tcp_proto;
5981 #endif /* CONFIG_INET */
5982 case BPF_FUNC_sk_storage_get:
5983 return &bpf_sk_storage_get_proto;
5984 case BPF_FUNC_sk_storage_delete:
5985 return &bpf_sk_storage_delete_proto;
5986 default:
5987 return bpf_base_func_proto(func_id);
5988 }
5989 }
5990
5991 static const struct bpf_func_proto *
5992 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5993 {
5994 switch (func_id) {
5995 case BPF_FUNC_skb_load_bytes:
5996 return &bpf_skb_load_bytes_proto;
5997 case BPF_FUNC_skb_load_bytes_relative:
5998 return &bpf_skb_load_bytes_relative_proto;
5999 case BPF_FUNC_get_socket_cookie:
6000 return &bpf_get_socket_cookie_proto;
6001 case BPF_FUNC_get_socket_uid:
6002 return &bpf_get_socket_uid_proto;
6003 default:
6004 return bpf_base_func_proto(func_id);
6005 }
6006 }
6007
6008 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6009 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6010
6011 static const struct bpf_func_proto *
6012 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6013 {
6014 switch (func_id) {
6015 case BPF_FUNC_get_local_storage:
6016 return &bpf_get_local_storage_proto;
6017 case BPF_FUNC_sk_fullsock:
6018 return &bpf_sk_fullsock_proto;
6019 case BPF_FUNC_sk_storage_get:
6020 return &bpf_sk_storage_get_proto;
6021 case BPF_FUNC_sk_storage_delete:
6022 return &bpf_sk_storage_delete_proto;
6023 #ifdef CONFIG_SOCK_CGROUP_DATA
6024 case BPF_FUNC_skb_cgroup_id:
6025 return &bpf_skb_cgroup_id_proto;
6026 #endif
6027 #ifdef CONFIG_INET
6028 case BPF_FUNC_tcp_sock:
6029 return &bpf_tcp_sock_proto;
6030 case BPF_FUNC_get_listener_sock:
6031 return &bpf_get_listener_sock_proto;
6032 case BPF_FUNC_skb_ecn_set_ce:
6033 return &bpf_skb_ecn_set_ce_proto;
6034 #endif
6035 default:
6036 return sk_filter_func_proto(func_id, prog);
6037 }
6038 }
6039
6040 static const struct bpf_func_proto *
6041 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6042 {
6043 switch (func_id) {
6044 case BPF_FUNC_skb_store_bytes:
6045 return &bpf_skb_store_bytes_proto;
6046 case BPF_FUNC_skb_load_bytes:
6047 return &bpf_skb_load_bytes_proto;
6048 case BPF_FUNC_skb_load_bytes_relative:
6049 return &bpf_skb_load_bytes_relative_proto;
6050 case BPF_FUNC_skb_pull_data:
6051 return &bpf_skb_pull_data_proto;
6052 case BPF_FUNC_csum_diff:
6053 return &bpf_csum_diff_proto;
6054 case BPF_FUNC_csum_update:
6055 return &bpf_csum_update_proto;
6056 case BPF_FUNC_l3_csum_replace:
6057 return &bpf_l3_csum_replace_proto;
6058 case BPF_FUNC_l4_csum_replace:
6059 return &bpf_l4_csum_replace_proto;
6060 case BPF_FUNC_clone_redirect:
6061 return &bpf_clone_redirect_proto;
6062 case BPF_FUNC_get_cgroup_classid:
6063 return &bpf_get_cgroup_classid_proto;
6064 case BPF_FUNC_skb_vlan_push:
6065 return &bpf_skb_vlan_push_proto;
6066 case BPF_FUNC_skb_vlan_pop:
6067 return &bpf_skb_vlan_pop_proto;
6068 case BPF_FUNC_skb_change_proto:
6069 return &bpf_skb_change_proto_proto;
6070 case BPF_FUNC_skb_change_type:
6071 return &bpf_skb_change_type_proto;
6072 case BPF_FUNC_skb_adjust_room:
6073 return &bpf_skb_adjust_room_proto;
6074 case BPF_FUNC_skb_change_tail:
6075 return &bpf_skb_change_tail_proto;
6076 case BPF_FUNC_skb_get_tunnel_key:
6077 return &bpf_skb_get_tunnel_key_proto;
6078 case BPF_FUNC_skb_set_tunnel_key:
6079 return bpf_get_skb_set_tunnel_proto(func_id);
6080 case BPF_FUNC_skb_get_tunnel_opt:
6081 return &bpf_skb_get_tunnel_opt_proto;
6082 case BPF_FUNC_skb_set_tunnel_opt:
6083 return bpf_get_skb_set_tunnel_proto(func_id);
6084 case BPF_FUNC_redirect:
6085 return &bpf_redirect_proto;
6086 case BPF_FUNC_get_route_realm:
6087 return &bpf_get_route_realm_proto;
6088 case BPF_FUNC_get_hash_recalc:
6089 return &bpf_get_hash_recalc_proto;
6090 case BPF_FUNC_set_hash_invalid:
6091 return &bpf_set_hash_invalid_proto;
6092 case BPF_FUNC_set_hash:
6093 return &bpf_set_hash_proto;
6094 case BPF_FUNC_perf_event_output:
6095 return &bpf_skb_event_output_proto;
6096 case BPF_FUNC_get_smp_processor_id:
6097 return &bpf_get_smp_processor_id_proto;
6098 case BPF_FUNC_skb_under_cgroup:
6099 return &bpf_skb_under_cgroup_proto;
6100 case BPF_FUNC_get_socket_cookie:
6101 return &bpf_get_socket_cookie_proto;
6102 case BPF_FUNC_get_socket_uid:
6103 return &bpf_get_socket_uid_proto;
6104 case BPF_FUNC_fib_lookup:
6105 return &bpf_skb_fib_lookup_proto;
6106 case BPF_FUNC_sk_fullsock:
6107 return &bpf_sk_fullsock_proto;
6108 case BPF_FUNC_sk_storage_get:
6109 return &bpf_sk_storage_get_proto;
6110 case BPF_FUNC_sk_storage_delete:
6111 return &bpf_sk_storage_delete_proto;
6112 #ifdef CONFIG_XFRM
6113 case BPF_FUNC_skb_get_xfrm_state:
6114 return &bpf_skb_get_xfrm_state_proto;
6115 #endif
6116 #ifdef CONFIG_SOCK_CGROUP_DATA
6117 case BPF_FUNC_skb_cgroup_id:
6118 return &bpf_skb_cgroup_id_proto;
6119 case BPF_FUNC_skb_ancestor_cgroup_id:
6120 return &bpf_skb_ancestor_cgroup_id_proto;
6121 #endif
6122 #ifdef CONFIG_INET
6123 case BPF_FUNC_sk_lookup_tcp:
6124 return &bpf_sk_lookup_tcp_proto;
6125 case BPF_FUNC_sk_lookup_udp:
6126 return &bpf_sk_lookup_udp_proto;
6127 case BPF_FUNC_sk_release:
6128 return &bpf_sk_release_proto;
6129 case BPF_FUNC_tcp_sock:
6130 return &bpf_tcp_sock_proto;
6131 case BPF_FUNC_get_listener_sock:
6132 return &bpf_get_listener_sock_proto;
6133 case BPF_FUNC_skc_lookup_tcp:
6134 return &bpf_skc_lookup_tcp_proto;
6135 case BPF_FUNC_tcp_check_syncookie:
6136 return &bpf_tcp_check_syncookie_proto;
6137 case BPF_FUNC_skb_ecn_set_ce:
6138 return &bpf_skb_ecn_set_ce_proto;
6139 #endif
6140 default:
6141 return bpf_base_func_proto(func_id);
6142 }
6143 }
6144
6145 static const struct bpf_func_proto *
6146 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6147 {
6148 switch (func_id) {
6149 case BPF_FUNC_perf_event_output:
6150 return &bpf_xdp_event_output_proto;
6151 case BPF_FUNC_get_smp_processor_id:
6152 return &bpf_get_smp_processor_id_proto;
6153 case BPF_FUNC_csum_diff:
6154 return &bpf_csum_diff_proto;
6155 case BPF_FUNC_xdp_adjust_head:
6156 return &bpf_xdp_adjust_head_proto;
6157 case BPF_FUNC_xdp_adjust_meta:
6158 return &bpf_xdp_adjust_meta_proto;
6159 case BPF_FUNC_redirect:
6160 return &bpf_xdp_redirect_proto;
6161 case BPF_FUNC_redirect_map:
6162 return &bpf_xdp_redirect_map_proto;
6163 case BPF_FUNC_xdp_adjust_tail:
6164 return &bpf_xdp_adjust_tail_proto;
6165 case BPF_FUNC_fib_lookup:
6166 return &bpf_xdp_fib_lookup_proto;
6167 #ifdef CONFIG_INET
6168 case BPF_FUNC_sk_lookup_udp:
6169 return &bpf_xdp_sk_lookup_udp_proto;
6170 case BPF_FUNC_sk_lookup_tcp:
6171 return &bpf_xdp_sk_lookup_tcp_proto;
6172 case BPF_FUNC_sk_release:
6173 return &bpf_sk_release_proto;
6174 case BPF_FUNC_skc_lookup_tcp:
6175 return &bpf_xdp_skc_lookup_tcp_proto;
6176 case BPF_FUNC_tcp_check_syncookie:
6177 return &bpf_tcp_check_syncookie_proto;
6178 #endif
6179 default:
6180 return bpf_base_func_proto(func_id);
6181 }
6182 }
6183
6184 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6185 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6186
6187 static const struct bpf_func_proto *
6188 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6189 {
6190 switch (func_id) {
6191 case BPF_FUNC_setsockopt:
6192 return &bpf_setsockopt_proto;
6193 case BPF_FUNC_getsockopt:
6194 return &bpf_getsockopt_proto;
6195 case BPF_FUNC_sock_ops_cb_flags_set:
6196 return &bpf_sock_ops_cb_flags_set_proto;
6197 case BPF_FUNC_sock_map_update:
6198 return &bpf_sock_map_update_proto;
6199 case BPF_FUNC_sock_hash_update:
6200 return &bpf_sock_hash_update_proto;
6201 case BPF_FUNC_get_socket_cookie:
6202 return &bpf_get_socket_cookie_sock_ops_proto;
6203 case BPF_FUNC_get_local_storage:
6204 return &bpf_get_local_storage_proto;
6205 case BPF_FUNC_perf_event_output:
6206 return &bpf_sockopt_event_output_proto;
6207 case BPF_FUNC_sk_storage_get:
6208 return &bpf_sk_storage_get_proto;
6209 case BPF_FUNC_sk_storage_delete:
6210 return &bpf_sk_storage_delete_proto;
6211 #ifdef CONFIG_INET
6212 case BPF_FUNC_tcp_sock:
6213 return &bpf_tcp_sock_proto;
6214 #endif /* CONFIG_INET */
6215 default:
6216 return bpf_base_func_proto(func_id);
6217 }
6218 }
6219
6220 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6221 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6222
6223 static const struct bpf_func_proto *
6224 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6225 {
6226 switch (func_id) {
6227 case BPF_FUNC_msg_redirect_map:
6228 return &bpf_msg_redirect_map_proto;
6229 case BPF_FUNC_msg_redirect_hash:
6230 return &bpf_msg_redirect_hash_proto;
6231 case BPF_FUNC_msg_apply_bytes:
6232 return &bpf_msg_apply_bytes_proto;
6233 case BPF_FUNC_msg_cork_bytes:
6234 return &bpf_msg_cork_bytes_proto;
6235 case BPF_FUNC_msg_pull_data:
6236 return &bpf_msg_pull_data_proto;
6237 case BPF_FUNC_msg_push_data:
6238 return &bpf_msg_push_data_proto;
6239 case BPF_FUNC_msg_pop_data:
6240 return &bpf_msg_pop_data_proto;
6241 default:
6242 return bpf_base_func_proto(func_id);
6243 }
6244 }
6245
6246 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6247 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6248
6249 static const struct bpf_func_proto *
6250 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6251 {
6252 switch (func_id) {
6253 case BPF_FUNC_skb_store_bytes:
6254 return &bpf_skb_store_bytes_proto;
6255 case BPF_FUNC_skb_load_bytes:
6256 return &bpf_skb_load_bytes_proto;
6257 case BPF_FUNC_skb_pull_data:
6258 return &sk_skb_pull_data_proto;
6259 case BPF_FUNC_skb_change_tail:
6260 return &sk_skb_change_tail_proto;
6261 case BPF_FUNC_skb_change_head:
6262 return &sk_skb_change_head_proto;
6263 case BPF_FUNC_get_socket_cookie:
6264 return &bpf_get_socket_cookie_proto;
6265 case BPF_FUNC_get_socket_uid:
6266 return &bpf_get_socket_uid_proto;
6267 case BPF_FUNC_sk_redirect_map:
6268 return &bpf_sk_redirect_map_proto;
6269 case BPF_FUNC_sk_redirect_hash:
6270 return &bpf_sk_redirect_hash_proto;
6271 #ifdef CONFIG_INET
6272 case BPF_FUNC_sk_lookup_tcp:
6273 return &bpf_sk_lookup_tcp_proto;
6274 case BPF_FUNC_sk_lookup_udp:
6275 return &bpf_sk_lookup_udp_proto;
6276 case BPF_FUNC_sk_release:
6277 return &bpf_sk_release_proto;
6278 case BPF_FUNC_skc_lookup_tcp:
6279 return &bpf_skc_lookup_tcp_proto;
6280 #endif
6281 default:
6282 return bpf_base_func_proto(func_id);
6283 }
6284 }
6285
6286 static const struct bpf_func_proto *
6287 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6288 {
6289 switch (func_id) {
6290 case BPF_FUNC_skb_load_bytes:
6291 return &bpf_flow_dissector_load_bytes_proto;
6292 default:
6293 return bpf_base_func_proto(func_id);
6294 }
6295 }
6296
6297 static const struct bpf_func_proto *
6298 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6299 {
6300 switch (func_id) {
6301 case BPF_FUNC_skb_load_bytes:
6302 return &bpf_skb_load_bytes_proto;
6303 case BPF_FUNC_skb_pull_data:
6304 return &bpf_skb_pull_data_proto;
6305 case BPF_FUNC_csum_diff:
6306 return &bpf_csum_diff_proto;
6307 case BPF_FUNC_get_cgroup_classid:
6308 return &bpf_get_cgroup_classid_proto;
6309 case BPF_FUNC_get_route_realm:
6310 return &bpf_get_route_realm_proto;
6311 case BPF_FUNC_get_hash_recalc:
6312 return &bpf_get_hash_recalc_proto;
6313 case BPF_FUNC_perf_event_output:
6314 return &bpf_skb_event_output_proto;
6315 case BPF_FUNC_get_smp_processor_id:
6316 return &bpf_get_smp_processor_id_proto;
6317 case BPF_FUNC_skb_under_cgroup:
6318 return &bpf_skb_under_cgroup_proto;
6319 default:
6320 return bpf_base_func_proto(func_id);
6321 }
6322 }
6323
6324 static const struct bpf_func_proto *
6325 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6326 {
6327 switch (func_id) {
6328 case BPF_FUNC_lwt_push_encap:
6329 return &bpf_lwt_in_push_encap_proto;
6330 default:
6331 return lwt_out_func_proto(func_id, prog);
6332 }
6333 }
6334
6335 static const struct bpf_func_proto *
6336 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6337 {
6338 switch (func_id) {
6339 case BPF_FUNC_skb_get_tunnel_key:
6340 return &bpf_skb_get_tunnel_key_proto;
6341 case BPF_FUNC_skb_set_tunnel_key:
6342 return bpf_get_skb_set_tunnel_proto(func_id);
6343 case BPF_FUNC_skb_get_tunnel_opt:
6344 return &bpf_skb_get_tunnel_opt_proto;
6345 case BPF_FUNC_skb_set_tunnel_opt:
6346 return bpf_get_skb_set_tunnel_proto(func_id);
6347 case BPF_FUNC_redirect:
6348 return &bpf_redirect_proto;
6349 case BPF_FUNC_clone_redirect:
6350 return &bpf_clone_redirect_proto;
6351 case BPF_FUNC_skb_change_tail:
6352 return &bpf_skb_change_tail_proto;
6353 case BPF_FUNC_skb_change_head:
6354 return &bpf_skb_change_head_proto;
6355 case BPF_FUNC_skb_store_bytes:
6356 return &bpf_skb_store_bytes_proto;
6357 case BPF_FUNC_csum_update:
6358 return &bpf_csum_update_proto;
6359 case BPF_FUNC_l3_csum_replace:
6360 return &bpf_l3_csum_replace_proto;
6361 case BPF_FUNC_l4_csum_replace:
6362 return &bpf_l4_csum_replace_proto;
6363 case BPF_FUNC_set_hash_invalid:
6364 return &bpf_set_hash_invalid_proto;
6365 case BPF_FUNC_lwt_push_encap:
6366 return &bpf_lwt_xmit_push_encap_proto;
6367 default:
6368 return lwt_out_func_proto(func_id, prog);
6369 }
6370 }
6371
6372 static const struct bpf_func_proto *
6373 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6374 {
6375 switch (func_id) {
6376 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6377 case BPF_FUNC_lwt_seg6_store_bytes:
6378 return &bpf_lwt_seg6_store_bytes_proto;
6379 case BPF_FUNC_lwt_seg6_action:
6380 return &bpf_lwt_seg6_action_proto;
6381 case BPF_FUNC_lwt_seg6_adjust_srh:
6382 return &bpf_lwt_seg6_adjust_srh_proto;
6383 #endif
6384 default:
6385 return lwt_out_func_proto(func_id, prog);
6386 }
6387 }
6388
6389 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6390 const struct bpf_prog *prog,
6391 struct bpf_insn_access_aux *info)
6392 {
6393 const int size_default = sizeof(__u32);
6394
6395 if (off < 0 || off >= sizeof(struct __sk_buff))
6396 return false;
6397
6398 /* The verifier guarantees that size > 0. */
6399 if (off % size != 0)
6400 return false;
6401
6402 switch (off) {
6403 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6404 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6405 return false;
6406 break;
6407 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6408 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6409 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6410 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6411 case bpf_ctx_range(struct __sk_buff, data):
6412 case bpf_ctx_range(struct __sk_buff, data_meta):
6413 case bpf_ctx_range(struct __sk_buff, data_end):
6414 if (size != size_default)
6415 return false;
6416 break;
6417 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6418 return false;
6419 case bpf_ctx_range(struct __sk_buff, tstamp):
6420 if (size != sizeof(__u64))
6421 return false;
6422 break;
6423 case offsetof(struct __sk_buff, sk):
6424 if (type == BPF_WRITE || size != sizeof(__u64))
6425 return false;
6426 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6427 break;
6428 default:
6429 /* Only narrow read access allowed for now. */
6430 if (type == BPF_WRITE) {
6431 if (size != size_default)
6432 return false;
6433 } else {
6434 bpf_ctx_record_field_size(info, size_default);
6435 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6436 return false;
6437 }
6438 }
6439
6440 return true;
6441 }
6442
6443 static bool sk_filter_is_valid_access(int off, int size,
6444 enum bpf_access_type type,
6445 const struct bpf_prog *prog,
6446 struct bpf_insn_access_aux *info)
6447 {
6448 switch (off) {
6449 case bpf_ctx_range(struct __sk_buff, tc_classid):
6450 case bpf_ctx_range(struct __sk_buff, data):
6451 case bpf_ctx_range(struct __sk_buff, data_meta):
6452 case bpf_ctx_range(struct __sk_buff, data_end):
6453 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6454 case bpf_ctx_range(struct __sk_buff, tstamp):
6455 case bpf_ctx_range(struct __sk_buff, wire_len):
6456 return false;
6457 }
6458
6459 if (type == BPF_WRITE) {
6460 switch (off) {
6461 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6462 break;
6463 default:
6464 return false;
6465 }
6466 }
6467
6468 return bpf_skb_is_valid_access(off, size, type, prog, info);
6469 }
6470
6471 static bool cg_skb_is_valid_access(int off, int size,
6472 enum bpf_access_type type,
6473 const struct bpf_prog *prog,
6474 struct bpf_insn_access_aux *info)
6475 {
6476 switch (off) {
6477 case bpf_ctx_range(struct __sk_buff, tc_classid):
6478 case bpf_ctx_range(struct __sk_buff, data_meta):
6479 case bpf_ctx_range(struct __sk_buff, wire_len):
6480 return false;
6481 case bpf_ctx_range(struct __sk_buff, data):
6482 case bpf_ctx_range(struct __sk_buff, data_end):
6483 if (!capable(CAP_SYS_ADMIN))
6484 return false;
6485 break;
6486 }
6487
6488 if (type == BPF_WRITE) {
6489 switch (off) {
6490 case bpf_ctx_range(struct __sk_buff, mark):
6491 case bpf_ctx_range(struct __sk_buff, priority):
6492 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6493 break;
6494 case bpf_ctx_range(struct __sk_buff, tstamp):
6495 if (!capable(CAP_SYS_ADMIN))
6496 return false;
6497 break;
6498 default:
6499 return false;
6500 }
6501 }
6502
6503 switch (off) {
6504 case bpf_ctx_range(struct __sk_buff, data):
6505 info->reg_type = PTR_TO_PACKET;
6506 break;
6507 case bpf_ctx_range(struct __sk_buff, data_end):
6508 info->reg_type = PTR_TO_PACKET_END;
6509 break;
6510 }
6511
6512 return bpf_skb_is_valid_access(off, size, type, prog, info);
6513 }
6514
6515 static bool lwt_is_valid_access(int off, int size,
6516 enum bpf_access_type type,
6517 const struct bpf_prog *prog,
6518 struct bpf_insn_access_aux *info)
6519 {
6520 switch (off) {
6521 case bpf_ctx_range(struct __sk_buff, tc_classid):
6522 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6523 case bpf_ctx_range(struct __sk_buff, data_meta):
6524 case bpf_ctx_range(struct __sk_buff, tstamp):
6525 case bpf_ctx_range(struct __sk_buff, wire_len):
6526 return false;
6527 }
6528
6529 if (type == BPF_WRITE) {
6530 switch (off) {
6531 case bpf_ctx_range(struct __sk_buff, mark):
6532 case bpf_ctx_range(struct __sk_buff, priority):
6533 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6534 break;
6535 default:
6536 return false;
6537 }
6538 }
6539
6540 switch (off) {
6541 case bpf_ctx_range(struct __sk_buff, data):
6542 info->reg_type = PTR_TO_PACKET;
6543 break;
6544 case bpf_ctx_range(struct __sk_buff, data_end):
6545 info->reg_type = PTR_TO_PACKET_END;
6546 break;
6547 }
6548
6549 return bpf_skb_is_valid_access(off, size, type, prog, info);
6550 }
6551
6552 /* Attach type specific accesses */
6553 static bool __sock_filter_check_attach_type(int off,
6554 enum bpf_access_type access_type,
6555 enum bpf_attach_type attach_type)
6556 {
6557 switch (off) {
6558 case offsetof(struct bpf_sock, bound_dev_if):
6559 case offsetof(struct bpf_sock, mark):
6560 case offsetof(struct bpf_sock, priority):
6561 switch (attach_type) {
6562 case BPF_CGROUP_INET_SOCK_CREATE:
6563 goto full_access;
6564 default:
6565 return false;
6566 }
6567 case bpf_ctx_range(struct bpf_sock, src_ip4):
6568 switch (attach_type) {
6569 case BPF_CGROUP_INET4_POST_BIND:
6570 goto read_only;
6571 default:
6572 return false;
6573 }
6574 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6575 switch (attach_type) {
6576 case BPF_CGROUP_INET6_POST_BIND:
6577 goto read_only;
6578 default:
6579 return false;
6580 }
6581 case bpf_ctx_range(struct bpf_sock, src_port):
6582 switch (attach_type) {
6583 case BPF_CGROUP_INET4_POST_BIND:
6584 case BPF_CGROUP_INET6_POST_BIND:
6585 goto read_only;
6586 default:
6587 return false;
6588 }
6589 }
6590 read_only:
6591 return access_type == BPF_READ;
6592 full_access:
6593 return true;
6594 }
6595
6596 bool bpf_sock_common_is_valid_access(int off, int size,
6597 enum bpf_access_type type,
6598 struct bpf_insn_access_aux *info)
6599 {
6600 switch (off) {
6601 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6602 return false;
6603 default:
6604 return bpf_sock_is_valid_access(off, size, type, info);
6605 }
6606 }
6607
6608 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6609 struct bpf_insn_access_aux *info)
6610 {
6611 const int size_default = sizeof(__u32);
6612
6613 if (off < 0 || off >= sizeof(struct bpf_sock))
6614 return false;
6615 if (off % size != 0)
6616 return false;
6617
6618 switch (off) {
6619 case offsetof(struct bpf_sock, state):
6620 case offsetof(struct bpf_sock, family):
6621 case offsetof(struct bpf_sock, type):
6622 case offsetof(struct bpf_sock, protocol):
6623 case offsetof(struct bpf_sock, dst_port):
6624 case offsetof(struct bpf_sock, src_port):
6625 case bpf_ctx_range(struct bpf_sock, src_ip4):
6626 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6627 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6628 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6629 bpf_ctx_record_field_size(info, size_default);
6630 return bpf_ctx_narrow_access_ok(off, size, size_default);
6631 }
6632
6633 return size == size_default;
6634 }
6635
6636 static bool sock_filter_is_valid_access(int off, int size,
6637 enum bpf_access_type type,
6638 const struct bpf_prog *prog,
6639 struct bpf_insn_access_aux *info)
6640 {
6641 if (!bpf_sock_is_valid_access(off, size, type, info))
6642 return false;
6643 return __sock_filter_check_attach_type(off, type,
6644 prog->expected_attach_type);
6645 }
6646
6647 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6648 const struct bpf_prog *prog)
6649 {
6650 /* Neither direct read nor direct write requires any preliminary
6651 * action.
6652 */
6653 return 0;
6654 }
6655
6656 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6657 const struct bpf_prog *prog, int drop_verdict)
6658 {
6659 struct bpf_insn *insn = insn_buf;
6660
6661 if (!direct_write)
6662 return 0;
6663
6664 /* if (!skb->cloned)
6665 * goto start;
6666 *
6667 * (Fast-path, otherwise approximation that we might be
6668 * a clone, do the rest in helper.)
6669 */
6670 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6671 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6672 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6673
6674 /* ret = bpf_skb_pull_data(skb, 0); */
6675 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6676 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6677 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6678 BPF_FUNC_skb_pull_data);
6679 /* if (!ret)
6680 * goto restore;
6681 * return TC_ACT_SHOT;
6682 */
6683 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6684 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6685 *insn++ = BPF_EXIT_INSN();
6686
6687 /* restore: */
6688 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6689 /* start: */
6690 *insn++ = prog->insnsi[0];
6691
6692 return insn - insn_buf;
6693 }
6694
6695 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6696 struct bpf_insn *insn_buf)
6697 {
6698 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6699 struct bpf_insn *insn = insn_buf;
6700
6701 /* We're guaranteed here that CTX is in R6. */
6702 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6703 if (!indirect) {
6704 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6705 } else {
6706 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6707 if (orig->imm)
6708 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6709 }
6710
6711 switch (BPF_SIZE(orig->code)) {
6712 case BPF_B:
6713 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6714 break;
6715 case BPF_H:
6716 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6717 break;
6718 case BPF_W:
6719 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6720 break;
6721 }
6722
6723 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6724 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6725 *insn++ = BPF_EXIT_INSN();
6726
6727 return insn - insn_buf;
6728 }
6729
6730 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6731 const struct bpf_prog *prog)
6732 {
6733 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6734 }
6735
6736 static bool tc_cls_act_is_valid_access(int off, int size,
6737 enum bpf_access_type type,
6738 const struct bpf_prog *prog,
6739 struct bpf_insn_access_aux *info)
6740 {
6741 if (type == BPF_WRITE) {
6742 switch (off) {
6743 case bpf_ctx_range(struct __sk_buff, mark):
6744 case bpf_ctx_range(struct __sk_buff, tc_index):
6745 case bpf_ctx_range(struct __sk_buff, priority):
6746 case bpf_ctx_range(struct __sk_buff, tc_classid):
6747 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6748 case bpf_ctx_range(struct __sk_buff, tstamp):
6749 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6750 break;
6751 default:
6752 return false;
6753 }
6754 }
6755
6756 switch (off) {
6757 case bpf_ctx_range(struct __sk_buff, data):
6758 info->reg_type = PTR_TO_PACKET;
6759 break;
6760 case bpf_ctx_range(struct __sk_buff, data_meta):
6761 info->reg_type = PTR_TO_PACKET_META;
6762 break;
6763 case bpf_ctx_range(struct __sk_buff, data_end):
6764 info->reg_type = PTR_TO_PACKET_END;
6765 break;
6766 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6767 return false;
6768 }
6769
6770 return bpf_skb_is_valid_access(off, size, type, prog, info);
6771 }
6772
6773 static bool __is_valid_xdp_access(int off, int size)
6774 {
6775 if (off < 0 || off >= sizeof(struct xdp_md))
6776 return false;
6777 if (off % size != 0)
6778 return false;
6779 if (size != sizeof(__u32))
6780 return false;
6781
6782 return true;
6783 }
6784
6785 static bool xdp_is_valid_access(int off, int size,
6786 enum bpf_access_type type,
6787 const struct bpf_prog *prog,
6788 struct bpf_insn_access_aux *info)
6789 {
6790 if (type == BPF_WRITE) {
6791 if (bpf_prog_is_dev_bound(prog->aux)) {
6792 switch (off) {
6793 case offsetof(struct xdp_md, rx_queue_index):
6794 return __is_valid_xdp_access(off, size);
6795 }
6796 }
6797 return false;
6798 }
6799
6800 switch (off) {
6801 case offsetof(struct xdp_md, data):
6802 info->reg_type = PTR_TO_PACKET;
6803 break;
6804 case offsetof(struct xdp_md, data_meta):
6805 info->reg_type = PTR_TO_PACKET_META;
6806 break;
6807 case offsetof(struct xdp_md, data_end):
6808 info->reg_type = PTR_TO_PACKET_END;
6809 break;
6810 }
6811
6812 return __is_valid_xdp_access(off, size);
6813 }
6814
6815 void bpf_warn_invalid_xdp_action(u32 act)
6816 {
6817 const u32 act_max = XDP_REDIRECT;
6818
6819 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6820 act > act_max ? "Illegal" : "Driver unsupported",
6821 act);
6822 }
6823 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6824
6825 static bool sock_addr_is_valid_access(int off, int size,
6826 enum bpf_access_type type,
6827 const struct bpf_prog *prog,
6828 struct bpf_insn_access_aux *info)
6829 {
6830 const int size_default = sizeof(__u32);
6831
6832 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6833 return false;
6834 if (off % size != 0)
6835 return false;
6836
6837 /* Disallow access to IPv6 fields from IPv4 contex and vise
6838 * versa.
6839 */
6840 switch (off) {
6841 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6842 switch (prog->expected_attach_type) {
6843 case BPF_CGROUP_INET4_BIND:
6844 case BPF_CGROUP_INET4_CONNECT:
6845 case BPF_CGROUP_UDP4_SENDMSG:
6846 case BPF_CGROUP_UDP4_RECVMSG:
6847 break;
6848 default:
6849 return false;
6850 }
6851 break;
6852 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6853 switch (prog->expected_attach_type) {
6854 case BPF_CGROUP_INET6_BIND:
6855 case BPF_CGROUP_INET6_CONNECT:
6856 case BPF_CGROUP_UDP6_SENDMSG:
6857 case BPF_CGROUP_UDP6_RECVMSG:
6858 break;
6859 default:
6860 return false;
6861 }
6862 break;
6863 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6864 switch (prog->expected_attach_type) {
6865 case BPF_CGROUP_UDP4_SENDMSG:
6866 break;
6867 default:
6868 return false;
6869 }
6870 break;
6871 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6872 msg_src_ip6[3]):
6873 switch (prog->expected_attach_type) {
6874 case BPF_CGROUP_UDP6_SENDMSG:
6875 break;
6876 default:
6877 return false;
6878 }
6879 break;
6880 }
6881
6882 switch (off) {
6883 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6884 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6885 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6886 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6887 msg_src_ip6[3]):
6888 if (type == BPF_READ) {
6889 bpf_ctx_record_field_size(info, size_default);
6890
6891 if (bpf_ctx_wide_access_ok(off, size,
6892 struct bpf_sock_addr,
6893 user_ip6))
6894 return true;
6895
6896 if (bpf_ctx_wide_access_ok(off, size,
6897 struct bpf_sock_addr,
6898 msg_src_ip6))
6899 return true;
6900
6901 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6902 return false;
6903 } else {
6904 if (bpf_ctx_wide_access_ok(off, size,
6905 struct bpf_sock_addr,
6906 user_ip6))
6907 return true;
6908
6909 if (bpf_ctx_wide_access_ok(off, size,
6910 struct bpf_sock_addr,
6911 msg_src_ip6))
6912 return true;
6913
6914 if (size != size_default)
6915 return false;
6916 }
6917 break;
6918 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6919 if (size != size_default)
6920 return false;
6921 break;
6922 case offsetof(struct bpf_sock_addr, sk):
6923 if (type != BPF_READ)
6924 return false;
6925 if (size != sizeof(__u64))
6926 return false;
6927 info->reg_type = PTR_TO_SOCKET;
6928 break;
6929 default:
6930 if (type == BPF_READ) {
6931 if (size != size_default)
6932 return false;
6933 } else {
6934 return false;
6935 }
6936 }
6937
6938 return true;
6939 }
6940
6941 static bool sock_ops_is_valid_access(int off, int size,
6942 enum bpf_access_type type,
6943 const struct bpf_prog *prog,
6944 struct bpf_insn_access_aux *info)
6945 {
6946 const int size_default = sizeof(__u32);
6947
6948 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6949 return false;
6950
6951 /* The verifier guarantees that size > 0. */
6952 if (off % size != 0)
6953 return false;
6954
6955 if (type == BPF_WRITE) {
6956 switch (off) {
6957 case offsetof(struct bpf_sock_ops, reply):
6958 case offsetof(struct bpf_sock_ops, sk_txhash):
6959 if (size != size_default)
6960 return false;
6961 break;
6962 default:
6963 return false;
6964 }
6965 } else {
6966 switch (off) {
6967 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6968 bytes_acked):
6969 if (size != sizeof(__u64))
6970 return false;
6971 break;
6972 case offsetof(struct bpf_sock_ops, sk):
6973 if (size != sizeof(__u64))
6974 return false;
6975 info->reg_type = PTR_TO_SOCKET_OR_NULL;
6976 break;
6977 default:
6978 if (size != size_default)
6979 return false;
6980 break;
6981 }
6982 }
6983
6984 return true;
6985 }
6986
6987 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6988 const struct bpf_prog *prog)
6989 {
6990 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6991 }
6992
6993 static bool sk_skb_is_valid_access(int off, int size,
6994 enum bpf_access_type type,
6995 const struct bpf_prog *prog,
6996 struct bpf_insn_access_aux *info)
6997 {
6998 switch (off) {
6999 case bpf_ctx_range(struct __sk_buff, tc_classid):
7000 case bpf_ctx_range(struct __sk_buff, data_meta):
7001 case bpf_ctx_range(struct __sk_buff, tstamp):
7002 case bpf_ctx_range(struct __sk_buff, wire_len):
7003 return false;
7004 }
7005
7006 if (type == BPF_WRITE) {
7007 switch (off) {
7008 case bpf_ctx_range(struct __sk_buff, tc_index):
7009 case bpf_ctx_range(struct __sk_buff, priority):
7010 break;
7011 default:
7012 return false;
7013 }
7014 }
7015
7016 switch (off) {
7017 case bpf_ctx_range(struct __sk_buff, mark):
7018 return false;
7019 case bpf_ctx_range(struct __sk_buff, data):
7020 info->reg_type = PTR_TO_PACKET;
7021 break;
7022 case bpf_ctx_range(struct __sk_buff, data_end):
7023 info->reg_type = PTR_TO_PACKET_END;
7024 break;
7025 }
7026
7027 return bpf_skb_is_valid_access(off, size, type, prog, info);
7028 }
7029
7030 static bool sk_msg_is_valid_access(int off, int size,
7031 enum bpf_access_type type,
7032 const struct bpf_prog *prog,
7033 struct bpf_insn_access_aux *info)
7034 {
7035 if (type == BPF_WRITE)
7036 return false;
7037
7038 if (off % size != 0)
7039 return false;
7040
7041 switch (off) {
7042 case offsetof(struct sk_msg_md, data):
7043 info->reg_type = PTR_TO_PACKET;
7044 if (size != sizeof(__u64))
7045 return false;
7046 break;
7047 case offsetof(struct sk_msg_md, data_end):
7048 info->reg_type = PTR_TO_PACKET_END;
7049 if (size != sizeof(__u64))
7050 return false;
7051 break;
7052 case bpf_ctx_range(struct sk_msg_md, family):
7053 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
7054 case bpf_ctx_range(struct sk_msg_md, local_ip4):
7055 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
7056 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
7057 case bpf_ctx_range(struct sk_msg_md, remote_port):
7058 case bpf_ctx_range(struct sk_msg_md, local_port):
7059 case bpf_ctx_range(struct sk_msg_md, size):
7060 if (size != sizeof(__u32))
7061 return false;
7062 break;
7063 default:
7064 return false;
7065 }
7066 return true;
7067 }
7068
7069 static bool flow_dissector_is_valid_access(int off, int size,
7070 enum bpf_access_type type,
7071 const struct bpf_prog *prog,
7072 struct bpf_insn_access_aux *info)
7073 {
7074 const int size_default = sizeof(__u32);
7075
7076 if (off < 0 || off >= sizeof(struct __sk_buff))
7077 return false;
7078
7079 if (type == BPF_WRITE)
7080 return false;
7081
7082 switch (off) {
7083 case bpf_ctx_range(struct __sk_buff, data):
7084 if (size != size_default)
7085 return false;
7086 info->reg_type = PTR_TO_PACKET;
7087 return true;
7088 case bpf_ctx_range(struct __sk_buff, data_end):
7089 if (size != size_default)
7090 return false;
7091 info->reg_type = PTR_TO_PACKET_END;
7092 return true;
7093 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7094 if (size != sizeof(__u64))
7095 return false;
7096 info->reg_type = PTR_TO_FLOW_KEYS;
7097 return true;
7098 default:
7099 return false;
7100 }
7101 }
7102
7103 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
7104 const struct bpf_insn *si,
7105 struct bpf_insn *insn_buf,
7106 struct bpf_prog *prog,
7107 u32 *target_size)
7108
7109 {
7110 struct bpf_insn *insn = insn_buf;
7111
7112 switch (si->off) {
7113 case offsetof(struct __sk_buff, data):
7114 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
7115 si->dst_reg, si->src_reg,
7116 offsetof(struct bpf_flow_dissector, data));
7117 break;
7118
7119 case offsetof(struct __sk_buff, data_end):
7120 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
7121 si->dst_reg, si->src_reg,
7122 offsetof(struct bpf_flow_dissector, data_end));
7123 break;
7124
7125 case offsetof(struct __sk_buff, flow_keys):
7126 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
7127 si->dst_reg, si->src_reg,
7128 offsetof(struct bpf_flow_dissector, flow_keys));
7129 break;
7130 }
7131
7132 return insn - insn_buf;
7133 }
7134
7135 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7136 const struct bpf_insn *si,
7137 struct bpf_insn *insn_buf,
7138 struct bpf_prog *prog, u32 *target_size)
7139 {
7140 struct bpf_insn *insn = insn_buf;
7141 int off;
7142
7143 switch (si->off) {
7144 case offsetof(struct __sk_buff, len):
7145 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7146 bpf_target_off(struct sk_buff, len, 4,
7147 target_size));
7148 break;
7149
7150 case offsetof(struct __sk_buff, protocol):
7151 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7152 bpf_target_off(struct sk_buff, protocol, 2,
7153 target_size));
7154 break;
7155
7156 case offsetof(struct __sk_buff, vlan_proto):
7157 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7158 bpf_target_off(struct sk_buff, vlan_proto, 2,
7159 target_size));
7160 break;
7161
7162 case offsetof(struct __sk_buff, priority):
7163 if (type == BPF_WRITE)
7164 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7165 bpf_target_off(struct sk_buff, priority, 4,
7166 target_size));
7167 else
7168 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7169 bpf_target_off(struct sk_buff, priority, 4,
7170 target_size));
7171 break;
7172
7173 case offsetof(struct __sk_buff, ingress_ifindex):
7174 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7175 bpf_target_off(struct sk_buff, skb_iif, 4,
7176 target_size));
7177 break;
7178
7179 case offsetof(struct __sk_buff, ifindex):
7180 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7181 si->dst_reg, si->src_reg,
7182 offsetof(struct sk_buff, dev));
7183 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7184 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7185 bpf_target_off(struct net_device, ifindex, 4,
7186 target_size));
7187 break;
7188
7189 case offsetof(struct __sk_buff, hash):
7190 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7191 bpf_target_off(struct sk_buff, hash, 4,
7192 target_size));
7193 break;
7194
7195 case offsetof(struct __sk_buff, mark):
7196 if (type == BPF_WRITE)
7197 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7198 bpf_target_off(struct sk_buff, mark, 4,
7199 target_size));
7200 else
7201 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7202 bpf_target_off(struct sk_buff, mark, 4,
7203 target_size));
7204 break;
7205
7206 case offsetof(struct __sk_buff, pkt_type):
7207 *target_size = 1;
7208 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7209 PKT_TYPE_OFFSET());
7210 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7211 #ifdef __BIG_ENDIAN_BITFIELD
7212 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7213 #endif
7214 break;
7215
7216 case offsetof(struct __sk_buff, queue_mapping):
7217 if (type == BPF_WRITE) {
7218 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7219 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7220 bpf_target_off(struct sk_buff,
7221 queue_mapping,
7222 2, target_size));
7223 } else {
7224 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7225 bpf_target_off(struct sk_buff,
7226 queue_mapping,
7227 2, target_size));
7228 }
7229 break;
7230
7231 case offsetof(struct __sk_buff, vlan_present):
7232 *target_size = 1;
7233 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7234 PKT_VLAN_PRESENT_OFFSET());
7235 if (PKT_VLAN_PRESENT_BIT)
7236 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7237 if (PKT_VLAN_PRESENT_BIT < 7)
7238 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7239 break;
7240
7241 case offsetof(struct __sk_buff, vlan_tci):
7242 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7243 bpf_target_off(struct sk_buff, vlan_tci, 2,
7244 target_size));
7245 break;
7246
7247 case offsetof(struct __sk_buff, cb[0]) ...
7248 offsetofend(struct __sk_buff, cb[4]) - 1:
7249 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
7250 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7251 offsetof(struct qdisc_skb_cb, data)) %
7252 sizeof(__u64));
7253
7254 prog->cb_access = 1;
7255 off = si->off;
7256 off -= offsetof(struct __sk_buff, cb[0]);
7257 off += offsetof(struct sk_buff, cb);
7258 off += offsetof(struct qdisc_skb_cb, data);
7259 if (type == BPF_WRITE)
7260 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7261 si->src_reg, off);
7262 else
7263 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7264 si->src_reg, off);
7265 break;
7266
7267 case offsetof(struct __sk_buff, tc_classid):
7268 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
7269
7270 off = si->off;
7271 off -= offsetof(struct __sk_buff, tc_classid);
7272 off += offsetof(struct sk_buff, cb);
7273 off += offsetof(struct qdisc_skb_cb, tc_classid);
7274 *target_size = 2;
7275 if (type == BPF_WRITE)
7276 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7277 si->src_reg, off);
7278 else
7279 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7280 si->src_reg, off);
7281 break;
7282
7283 case offsetof(struct __sk_buff, data):
7284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7285 si->dst_reg, si->src_reg,
7286 offsetof(struct sk_buff, data));
7287 break;
7288
7289 case offsetof(struct __sk_buff, data_meta):
7290 off = si->off;
7291 off -= offsetof(struct __sk_buff, data_meta);
7292 off += offsetof(struct sk_buff, cb);
7293 off += offsetof(struct bpf_skb_data_end, data_meta);
7294 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7295 si->src_reg, off);
7296 break;
7297
7298 case offsetof(struct __sk_buff, data_end):
7299 off = si->off;
7300 off -= offsetof(struct __sk_buff, data_end);
7301 off += offsetof(struct sk_buff, cb);
7302 off += offsetof(struct bpf_skb_data_end, data_end);
7303 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7304 si->src_reg, off);
7305 break;
7306
7307 case offsetof(struct __sk_buff, tc_index):
7308 #ifdef CONFIG_NET_SCHED
7309 if (type == BPF_WRITE)
7310 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7311 bpf_target_off(struct sk_buff, tc_index, 2,
7312 target_size));
7313 else
7314 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7315 bpf_target_off(struct sk_buff, tc_index, 2,
7316 target_size));
7317 #else
7318 *target_size = 2;
7319 if (type == BPF_WRITE)
7320 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7321 else
7322 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7323 #endif
7324 break;
7325
7326 case offsetof(struct __sk_buff, napi_id):
7327 #if defined(CONFIG_NET_RX_BUSY_POLL)
7328 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7329 bpf_target_off(struct sk_buff, napi_id, 4,
7330 target_size));
7331 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7332 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7333 #else
7334 *target_size = 4;
7335 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7336 #endif
7337 break;
7338 case offsetof(struct __sk_buff, family):
7339 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7340
7341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7342 si->dst_reg, si->src_reg,
7343 offsetof(struct sk_buff, sk));
7344 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7345 bpf_target_off(struct sock_common,
7346 skc_family,
7347 2, target_size));
7348 break;
7349 case offsetof(struct __sk_buff, remote_ip4):
7350 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7351
7352 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7353 si->dst_reg, si->src_reg,
7354 offsetof(struct sk_buff, sk));
7355 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7356 bpf_target_off(struct sock_common,
7357 skc_daddr,
7358 4, target_size));
7359 break;
7360 case offsetof(struct __sk_buff, local_ip4):
7361 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7362 skc_rcv_saddr) != 4);
7363
7364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7365 si->dst_reg, si->src_reg,
7366 offsetof(struct sk_buff, sk));
7367 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7368 bpf_target_off(struct sock_common,
7369 skc_rcv_saddr,
7370 4, target_size));
7371 break;
7372 case offsetof(struct __sk_buff, remote_ip6[0]) ...
7373 offsetof(struct __sk_buff, remote_ip6[3]):
7374 #if IS_ENABLED(CONFIG_IPV6)
7375 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7376 skc_v6_daddr.s6_addr32[0]) != 4);
7377
7378 off = si->off;
7379 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7380
7381 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7382 si->dst_reg, si->src_reg,
7383 offsetof(struct sk_buff, sk));
7384 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7385 offsetof(struct sock_common,
7386 skc_v6_daddr.s6_addr32[0]) +
7387 off);
7388 #else
7389 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7390 #endif
7391 break;
7392 case offsetof(struct __sk_buff, local_ip6[0]) ...
7393 offsetof(struct __sk_buff, local_ip6[3]):
7394 #if IS_ENABLED(CONFIG_IPV6)
7395 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7396 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7397
7398 off = si->off;
7399 off -= offsetof(struct __sk_buff, local_ip6[0]);
7400
7401 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7402 si->dst_reg, si->src_reg,
7403 offsetof(struct sk_buff, sk));
7404 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7405 offsetof(struct sock_common,
7406 skc_v6_rcv_saddr.s6_addr32[0]) +
7407 off);
7408 #else
7409 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7410 #endif
7411 break;
7412
7413 case offsetof(struct __sk_buff, remote_port):
7414 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7415
7416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7417 si->dst_reg, si->src_reg,
7418 offsetof(struct sk_buff, sk));
7419 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7420 bpf_target_off(struct sock_common,
7421 skc_dport,
7422 2, target_size));
7423 #ifndef __BIG_ENDIAN_BITFIELD
7424 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7425 #endif
7426 break;
7427
7428 case offsetof(struct __sk_buff, local_port):
7429 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7430
7431 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7432 si->dst_reg, si->src_reg,
7433 offsetof(struct sk_buff, sk));
7434 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7435 bpf_target_off(struct sock_common,
7436 skc_num, 2, target_size));
7437 break;
7438
7439 case offsetof(struct __sk_buff, tstamp):
7440 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
7441
7442 if (type == BPF_WRITE)
7443 *insn++ = BPF_STX_MEM(BPF_DW,
7444 si->dst_reg, si->src_reg,
7445 bpf_target_off(struct sk_buff,
7446 tstamp, 8,
7447 target_size));
7448 else
7449 *insn++ = BPF_LDX_MEM(BPF_DW,
7450 si->dst_reg, si->src_reg,
7451 bpf_target_off(struct sk_buff,
7452 tstamp, 8,
7453 target_size));
7454 break;
7455
7456 case offsetof(struct __sk_buff, gso_segs):
7457 /* si->dst_reg = skb_shinfo(SKB); */
7458 #ifdef NET_SKBUFF_DATA_USES_OFFSET
7459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7460 BPF_REG_AX, si->src_reg,
7461 offsetof(struct sk_buff, end));
7462 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7463 si->dst_reg, si->src_reg,
7464 offsetof(struct sk_buff, head));
7465 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7466 #else
7467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7468 si->dst_reg, si->src_reg,
7469 offsetof(struct sk_buff, end));
7470 #endif
7471 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7472 si->dst_reg, si->dst_reg,
7473 bpf_target_off(struct skb_shared_info,
7474 gso_segs, 2,
7475 target_size));
7476 break;
7477 case offsetof(struct __sk_buff, wire_len):
7478 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
7479
7480 off = si->off;
7481 off -= offsetof(struct __sk_buff, wire_len);
7482 off += offsetof(struct sk_buff, cb);
7483 off += offsetof(struct qdisc_skb_cb, pkt_len);
7484 *target_size = 4;
7485 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7486 break;
7487
7488 case offsetof(struct __sk_buff, sk):
7489 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7490 si->dst_reg, si->src_reg,
7491 offsetof(struct sk_buff, sk));
7492 break;
7493 }
7494
7495 return insn - insn_buf;
7496 }
7497
7498 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7499 const struct bpf_insn *si,
7500 struct bpf_insn *insn_buf,
7501 struct bpf_prog *prog, u32 *target_size)
7502 {
7503 struct bpf_insn *insn = insn_buf;
7504 int off;
7505
7506 switch (si->off) {
7507 case offsetof(struct bpf_sock, bound_dev_if):
7508 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7509
7510 if (type == BPF_WRITE)
7511 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7512 offsetof(struct sock, sk_bound_dev_if));
7513 else
7514 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7515 offsetof(struct sock, sk_bound_dev_if));
7516 break;
7517
7518 case offsetof(struct bpf_sock, mark):
7519 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7520
7521 if (type == BPF_WRITE)
7522 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7523 offsetof(struct sock, sk_mark));
7524 else
7525 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7526 offsetof(struct sock, sk_mark));
7527 break;
7528
7529 case offsetof(struct bpf_sock, priority):
7530 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7531
7532 if (type == BPF_WRITE)
7533 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7534 offsetof(struct sock, sk_priority));
7535 else
7536 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7537 offsetof(struct sock, sk_priority));
7538 break;
7539
7540 case offsetof(struct bpf_sock, family):
7541 *insn++ = BPF_LDX_MEM(
7542 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7543 si->dst_reg, si->src_reg,
7544 bpf_target_off(struct sock_common,
7545 skc_family,
7546 FIELD_SIZEOF(struct sock_common,
7547 skc_family),
7548 target_size));
7549 break;
7550
7551 case offsetof(struct bpf_sock, type):
7552 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7554 offsetof(struct sock, __sk_flags_offset));
7555 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7556 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7557 *target_size = 2;
7558 break;
7559
7560 case offsetof(struct bpf_sock, protocol):
7561 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7562 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7563 offsetof(struct sock, __sk_flags_offset));
7564 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7565 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7566 *target_size = 1;
7567 break;
7568
7569 case offsetof(struct bpf_sock, src_ip4):
7570 *insn++ = BPF_LDX_MEM(
7571 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7572 bpf_target_off(struct sock_common, skc_rcv_saddr,
7573 FIELD_SIZEOF(struct sock_common,
7574 skc_rcv_saddr),
7575 target_size));
7576 break;
7577
7578 case offsetof(struct bpf_sock, dst_ip4):
7579 *insn++ = BPF_LDX_MEM(
7580 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7581 bpf_target_off(struct sock_common, skc_daddr,
7582 FIELD_SIZEOF(struct sock_common,
7583 skc_daddr),
7584 target_size));
7585 break;
7586
7587 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7588 #if IS_ENABLED(CONFIG_IPV6)
7589 off = si->off;
7590 off -= offsetof(struct bpf_sock, src_ip6[0]);
7591 *insn++ = BPF_LDX_MEM(
7592 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7593 bpf_target_off(
7594 struct sock_common,
7595 skc_v6_rcv_saddr.s6_addr32[0],
7596 FIELD_SIZEOF(struct sock_common,
7597 skc_v6_rcv_saddr.s6_addr32[0]),
7598 target_size) + off);
7599 #else
7600 (void)off;
7601 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7602 #endif
7603 break;
7604
7605 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7606 #if IS_ENABLED(CONFIG_IPV6)
7607 off = si->off;
7608 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7609 *insn++ = BPF_LDX_MEM(
7610 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7611 bpf_target_off(struct sock_common,
7612 skc_v6_daddr.s6_addr32[0],
7613 FIELD_SIZEOF(struct sock_common,
7614 skc_v6_daddr.s6_addr32[0]),
7615 target_size) + off);
7616 #else
7617 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7618 *target_size = 4;
7619 #endif
7620 break;
7621
7622 case offsetof(struct bpf_sock, src_port):
7623 *insn++ = BPF_LDX_MEM(
7624 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7625 si->dst_reg, si->src_reg,
7626 bpf_target_off(struct sock_common, skc_num,
7627 FIELD_SIZEOF(struct sock_common,
7628 skc_num),
7629 target_size));
7630 break;
7631
7632 case offsetof(struct bpf_sock, dst_port):
7633 *insn++ = BPF_LDX_MEM(
7634 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7635 si->dst_reg, si->src_reg,
7636 bpf_target_off(struct sock_common, skc_dport,
7637 FIELD_SIZEOF(struct sock_common,
7638 skc_dport),
7639 target_size));
7640 break;
7641
7642 case offsetof(struct bpf_sock, state):
7643 *insn++ = BPF_LDX_MEM(
7644 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7645 si->dst_reg, si->src_reg,
7646 bpf_target_off(struct sock_common, skc_state,
7647 FIELD_SIZEOF(struct sock_common,
7648 skc_state),
7649 target_size));
7650 break;
7651 }
7652
7653 return insn - insn_buf;
7654 }
7655
7656 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7657 const struct bpf_insn *si,
7658 struct bpf_insn *insn_buf,
7659 struct bpf_prog *prog, u32 *target_size)
7660 {
7661 struct bpf_insn *insn = insn_buf;
7662
7663 switch (si->off) {
7664 case offsetof(struct __sk_buff, ifindex):
7665 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7666 si->dst_reg, si->src_reg,
7667 offsetof(struct sk_buff, dev));
7668 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7669 bpf_target_off(struct net_device, ifindex, 4,
7670 target_size));
7671 break;
7672 default:
7673 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7674 target_size);
7675 }
7676
7677 return insn - insn_buf;
7678 }
7679
7680 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7681 const struct bpf_insn *si,
7682 struct bpf_insn *insn_buf,
7683 struct bpf_prog *prog, u32 *target_size)
7684 {
7685 struct bpf_insn *insn = insn_buf;
7686
7687 switch (si->off) {
7688 case offsetof(struct xdp_md, data):
7689 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7690 si->dst_reg, si->src_reg,
7691 offsetof(struct xdp_buff, data));
7692 break;
7693 case offsetof(struct xdp_md, data_meta):
7694 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7695 si->dst_reg, si->src_reg,
7696 offsetof(struct xdp_buff, data_meta));
7697 break;
7698 case offsetof(struct xdp_md, data_end):
7699 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7700 si->dst_reg, si->src_reg,
7701 offsetof(struct xdp_buff, data_end));
7702 break;
7703 case offsetof(struct xdp_md, ingress_ifindex):
7704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7705 si->dst_reg, si->src_reg,
7706 offsetof(struct xdp_buff, rxq));
7707 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7708 si->dst_reg, si->dst_reg,
7709 offsetof(struct xdp_rxq_info, dev));
7710 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7711 offsetof(struct net_device, ifindex));
7712 break;
7713 case offsetof(struct xdp_md, rx_queue_index):
7714 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7715 si->dst_reg, si->src_reg,
7716 offsetof(struct xdp_buff, rxq));
7717 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7718 offsetof(struct xdp_rxq_info,
7719 queue_index));
7720 break;
7721 }
7722
7723 return insn - insn_buf;
7724 }
7725
7726 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7727 * context Structure, F is Field in context structure that contains a pointer
7728 * to Nested Structure of type NS that has the field NF.
7729 *
7730 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7731 * sure that SIZE is not greater than actual size of S.F.NF.
7732 *
7733 * If offset OFF is provided, the load happens from that offset relative to
7734 * offset of NF.
7735 */
7736 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7737 do { \
7738 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7739 si->src_reg, offsetof(S, F)); \
7740 *insn++ = BPF_LDX_MEM( \
7741 SIZE, si->dst_reg, si->dst_reg, \
7742 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7743 target_size) \
7744 + OFF); \
7745 } while (0)
7746
7747 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7748 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7749 BPF_FIELD_SIZEOF(NS, NF), 0)
7750
7751 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7752 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7753 *
7754 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7755 * "register" since two registers available in convert_ctx_access are not
7756 * enough: we can't override neither SRC, since it contains value to store, nor
7757 * DST since it contains pointer to context that may be used by later
7758 * instructions. But we need a temporary place to save pointer to nested
7759 * structure whose field we want to store to.
7760 */
7761 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
7762 do { \
7763 int tmp_reg = BPF_REG_9; \
7764 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7765 --tmp_reg; \
7766 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7767 --tmp_reg; \
7768 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7769 offsetof(S, TF)); \
7770 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7771 si->dst_reg, offsetof(S, F)); \
7772 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
7773 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7774 target_size) \
7775 + OFF); \
7776 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7777 offsetof(S, TF)); \
7778 } while (0)
7779
7780 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7781 TF) \
7782 do { \
7783 if (type == BPF_WRITE) { \
7784 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
7785 OFF, TF); \
7786 } else { \
7787 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7788 S, NS, F, NF, SIZE, OFF); \
7789 } \
7790 } while (0)
7791
7792 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7793 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7794 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7795
7796 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7797 const struct bpf_insn *si,
7798 struct bpf_insn *insn_buf,
7799 struct bpf_prog *prog, u32 *target_size)
7800 {
7801 struct bpf_insn *insn = insn_buf;
7802 int off;
7803
7804 switch (si->off) {
7805 case offsetof(struct bpf_sock_addr, user_family):
7806 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7807 struct sockaddr, uaddr, sa_family);
7808 break;
7809
7810 case offsetof(struct bpf_sock_addr, user_ip4):
7811 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7812 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7813 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7814 break;
7815
7816 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7817 off = si->off;
7818 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7819 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7820 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7821 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7822 tmp_reg);
7823 break;
7824
7825 case offsetof(struct bpf_sock_addr, user_port):
7826 /* To get port we need to know sa_family first and then treat
7827 * sockaddr as either sockaddr_in or sockaddr_in6.
7828 * Though we can simplify since port field has same offset and
7829 * size in both structures.
7830 * Here we check this invariant and use just one of the
7831 * structures if it's true.
7832 */
7833 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7834 offsetof(struct sockaddr_in6, sin6_port));
7835 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7836 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7837 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7838 struct sockaddr_in6, uaddr,
7839 sin6_port, tmp_reg);
7840 break;
7841
7842 case offsetof(struct bpf_sock_addr, family):
7843 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7844 struct sock, sk, sk_family);
7845 break;
7846
7847 case offsetof(struct bpf_sock_addr, type):
7848 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7849 struct bpf_sock_addr_kern, struct sock, sk,
7850 __sk_flags_offset, BPF_W, 0);
7851 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7852 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7853 break;
7854
7855 case offsetof(struct bpf_sock_addr, protocol):
7856 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7857 struct bpf_sock_addr_kern, struct sock, sk,
7858 __sk_flags_offset, BPF_W, 0);
7859 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7860 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7861 SK_FL_PROTO_SHIFT);
7862 break;
7863
7864 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7865 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7866 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7867 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7868 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7869 break;
7870
7871 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7872 msg_src_ip6[3]):
7873 off = si->off;
7874 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7875 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7876 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7877 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7878 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7879 break;
7880 case offsetof(struct bpf_sock_addr, sk):
7881 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
7882 si->dst_reg, si->src_reg,
7883 offsetof(struct bpf_sock_addr_kern, sk));
7884 break;
7885 }
7886
7887 return insn - insn_buf;
7888 }
7889
7890 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7891 const struct bpf_insn *si,
7892 struct bpf_insn *insn_buf,
7893 struct bpf_prog *prog,
7894 u32 *target_size)
7895 {
7896 struct bpf_insn *insn = insn_buf;
7897 int off;
7898
7899 /* Helper macro for adding read access to tcp_sock or sock fields. */
7900 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7901 do { \
7902 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7903 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7905 struct bpf_sock_ops_kern, \
7906 is_fullsock), \
7907 si->dst_reg, si->src_reg, \
7908 offsetof(struct bpf_sock_ops_kern, \
7909 is_fullsock)); \
7910 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7912 struct bpf_sock_ops_kern, sk),\
7913 si->dst_reg, si->src_reg, \
7914 offsetof(struct bpf_sock_ops_kern, sk));\
7915 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7916 OBJ_FIELD), \
7917 si->dst_reg, si->dst_reg, \
7918 offsetof(OBJ, OBJ_FIELD)); \
7919 } while (0)
7920
7921 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7922 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7923
7924 /* Helper macro for adding write access to tcp_sock or sock fields.
7925 * The macro is called with two registers, dst_reg which contains a pointer
7926 * to ctx (context) and src_reg which contains the value that should be
7927 * stored. However, we need an additional register since we cannot overwrite
7928 * dst_reg because it may be used later in the program.
7929 * Instead we "borrow" one of the other register. We first save its value
7930 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7931 * it at the end of the macro.
7932 */
7933 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7934 do { \
7935 int reg = BPF_REG_9; \
7936 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7937 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7938 if (si->dst_reg == reg || si->src_reg == reg) \
7939 reg--; \
7940 if (si->dst_reg == reg || si->src_reg == reg) \
7941 reg--; \
7942 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7943 offsetof(struct bpf_sock_ops_kern, \
7944 temp)); \
7945 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7946 struct bpf_sock_ops_kern, \
7947 is_fullsock), \
7948 reg, si->dst_reg, \
7949 offsetof(struct bpf_sock_ops_kern, \
7950 is_fullsock)); \
7951 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7952 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7953 struct bpf_sock_ops_kern, sk),\
7954 reg, si->dst_reg, \
7955 offsetof(struct bpf_sock_ops_kern, sk));\
7956 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7957 reg, si->src_reg, \
7958 offsetof(OBJ, OBJ_FIELD)); \
7959 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7960 offsetof(struct bpf_sock_ops_kern, \
7961 temp)); \
7962 } while (0)
7963
7964 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7965 do { \
7966 if (TYPE == BPF_WRITE) \
7967 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7968 else \
7969 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7970 } while (0)
7971
7972 if (insn > insn_buf)
7973 return insn - insn_buf;
7974
7975 switch (si->off) {
7976 case offsetof(struct bpf_sock_ops, op) ...
7977 offsetof(struct bpf_sock_ops, replylong[3]):
7978 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7979 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7980 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7981 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7982 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7983 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7984 off = si->off;
7985 off -= offsetof(struct bpf_sock_ops, op);
7986 off += offsetof(struct bpf_sock_ops_kern, op);
7987 if (type == BPF_WRITE)
7988 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7989 off);
7990 else
7991 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7992 off);
7993 break;
7994
7995 case offsetof(struct bpf_sock_ops, family):
7996 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7997
7998 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7999 struct bpf_sock_ops_kern, sk),
8000 si->dst_reg, si->src_reg,
8001 offsetof(struct bpf_sock_ops_kern, sk));
8002 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8003 offsetof(struct sock_common, skc_family));
8004 break;
8005
8006 case offsetof(struct bpf_sock_ops, remote_ip4):
8007 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8008
8009 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8010 struct bpf_sock_ops_kern, sk),
8011 si->dst_reg, si->src_reg,
8012 offsetof(struct bpf_sock_ops_kern, sk));
8013 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8014 offsetof(struct sock_common, skc_daddr));
8015 break;
8016
8017 case offsetof(struct bpf_sock_ops, local_ip4):
8018 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8019 skc_rcv_saddr) != 4);
8020
8021 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8022 struct bpf_sock_ops_kern, sk),
8023 si->dst_reg, si->src_reg,
8024 offsetof(struct bpf_sock_ops_kern, sk));
8025 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8026 offsetof(struct sock_common,
8027 skc_rcv_saddr));
8028 break;
8029
8030 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
8031 offsetof(struct bpf_sock_ops, remote_ip6[3]):
8032 #if IS_ENABLED(CONFIG_IPV6)
8033 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8034 skc_v6_daddr.s6_addr32[0]) != 4);
8035
8036 off = si->off;
8037 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
8038 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8039 struct bpf_sock_ops_kern, sk),
8040 si->dst_reg, si->src_reg,
8041 offsetof(struct bpf_sock_ops_kern, sk));
8042 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8043 offsetof(struct sock_common,
8044 skc_v6_daddr.s6_addr32[0]) +
8045 off);
8046 #else
8047 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8048 #endif
8049 break;
8050
8051 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
8052 offsetof(struct bpf_sock_ops, local_ip6[3]):
8053 #if IS_ENABLED(CONFIG_IPV6)
8054 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8055 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8056
8057 off = si->off;
8058 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
8059 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8060 struct bpf_sock_ops_kern, sk),
8061 si->dst_reg, si->src_reg,
8062 offsetof(struct bpf_sock_ops_kern, sk));
8063 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8064 offsetof(struct sock_common,
8065 skc_v6_rcv_saddr.s6_addr32[0]) +
8066 off);
8067 #else
8068 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8069 #endif
8070 break;
8071
8072 case offsetof(struct bpf_sock_ops, remote_port):
8073 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8074
8075 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8076 struct bpf_sock_ops_kern, sk),
8077 si->dst_reg, si->src_reg,
8078 offsetof(struct bpf_sock_ops_kern, sk));
8079 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8080 offsetof(struct sock_common, skc_dport));
8081 #ifndef __BIG_ENDIAN_BITFIELD
8082 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8083 #endif
8084 break;
8085
8086 case offsetof(struct bpf_sock_ops, local_port):
8087 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8088
8089 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8090 struct bpf_sock_ops_kern, sk),
8091 si->dst_reg, si->src_reg,
8092 offsetof(struct bpf_sock_ops_kern, sk));
8093 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8094 offsetof(struct sock_common, skc_num));
8095 break;
8096
8097 case offsetof(struct bpf_sock_ops, is_fullsock):
8098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8099 struct bpf_sock_ops_kern,
8100 is_fullsock),
8101 si->dst_reg, si->src_reg,
8102 offsetof(struct bpf_sock_ops_kern,
8103 is_fullsock));
8104 break;
8105
8106 case offsetof(struct bpf_sock_ops, state):
8107 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
8108
8109 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8110 struct bpf_sock_ops_kern, sk),
8111 si->dst_reg, si->src_reg,
8112 offsetof(struct bpf_sock_ops_kern, sk));
8113 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
8114 offsetof(struct sock_common, skc_state));
8115 break;
8116
8117 case offsetof(struct bpf_sock_ops, rtt_min):
8118 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
8119 sizeof(struct minmax));
8120 BUILD_BUG_ON(sizeof(struct minmax) <
8121 sizeof(struct minmax_sample));
8122
8123 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8124 struct bpf_sock_ops_kern, sk),
8125 si->dst_reg, si->src_reg,
8126 offsetof(struct bpf_sock_ops_kern, sk));
8127 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8128 offsetof(struct tcp_sock, rtt_min) +
8129 FIELD_SIZEOF(struct minmax_sample, t));
8130 break;
8131
8132 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8133 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8134 struct tcp_sock);
8135 break;
8136
8137 case offsetof(struct bpf_sock_ops, sk_txhash):
8138 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8139 struct sock, type);
8140 break;
8141 case offsetof(struct bpf_sock_ops, snd_cwnd):
8142 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
8143 break;
8144 case offsetof(struct bpf_sock_ops, srtt_us):
8145 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
8146 break;
8147 case offsetof(struct bpf_sock_ops, snd_ssthresh):
8148 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
8149 break;
8150 case offsetof(struct bpf_sock_ops, rcv_nxt):
8151 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
8152 break;
8153 case offsetof(struct bpf_sock_ops, snd_nxt):
8154 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
8155 break;
8156 case offsetof(struct bpf_sock_ops, snd_una):
8157 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
8158 break;
8159 case offsetof(struct bpf_sock_ops, mss_cache):
8160 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
8161 break;
8162 case offsetof(struct bpf_sock_ops, ecn_flags):
8163 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
8164 break;
8165 case offsetof(struct bpf_sock_ops, rate_delivered):
8166 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
8167 break;
8168 case offsetof(struct bpf_sock_ops, rate_interval_us):
8169 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
8170 break;
8171 case offsetof(struct bpf_sock_ops, packets_out):
8172 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
8173 break;
8174 case offsetof(struct bpf_sock_ops, retrans_out):
8175 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
8176 break;
8177 case offsetof(struct bpf_sock_ops, total_retrans):
8178 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
8179 break;
8180 case offsetof(struct bpf_sock_ops, segs_in):
8181 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
8182 break;
8183 case offsetof(struct bpf_sock_ops, data_segs_in):
8184 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
8185 break;
8186 case offsetof(struct bpf_sock_ops, segs_out):
8187 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
8188 break;
8189 case offsetof(struct bpf_sock_ops, data_segs_out):
8190 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
8191 break;
8192 case offsetof(struct bpf_sock_ops, lost_out):
8193 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
8194 break;
8195 case offsetof(struct bpf_sock_ops, sacked_out):
8196 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
8197 break;
8198 case offsetof(struct bpf_sock_ops, bytes_received):
8199 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
8200 break;
8201 case offsetof(struct bpf_sock_ops, bytes_acked):
8202 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
8203 break;
8204 case offsetof(struct bpf_sock_ops, sk):
8205 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8206 struct bpf_sock_ops_kern,
8207 is_fullsock),
8208 si->dst_reg, si->src_reg,
8209 offsetof(struct bpf_sock_ops_kern,
8210 is_fullsock));
8211 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8212 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8213 struct bpf_sock_ops_kern, sk),
8214 si->dst_reg, si->src_reg,
8215 offsetof(struct bpf_sock_ops_kern, sk));
8216 break;
8217 }
8218 return insn - insn_buf;
8219 }
8220
8221 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8222 const struct bpf_insn *si,
8223 struct bpf_insn *insn_buf,
8224 struct bpf_prog *prog, u32 *target_size)
8225 {
8226 struct bpf_insn *insn = insn_buf;
8227 int off;
8228
8229 switch (si->off) {
8230 case offsetof(struct __sk_buff, data_end):
8231 off = si->off;
8232 off -= offsetof(struct __sk_buff, data_end);
8233 off += offsetof(struct sk_buff, cb);
8234 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8235 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8236 si->src_reg, off);
8237 break;
8238 default:
8239 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8240 target_size);
8241 }
8242
8243 return insn - insn_buf;
8244 }
8245
8246 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8247 const struct bpf_insn *si,
8248 struct bpf_insn *insn_buf,
8249 struct bpf_prog *prog, u32 *target_size)
8250 {
8251 struct bpf_insn *insn = insn_buf;
8252 #if IS_ENABLED(CONFIG_IPV6)
8253 int off;
8254 #endif
8255
8256 /* convert ctx uses the fact sg element is first in struct */
8257 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8258
8259 switch (si->off) {
8260 case offsetof(struct sk_msg_md, data):
8261 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8262 si->dst_reg, si->src_reg,
8263 offsetof(struct sk_msg, data));
8264 break;
8265 case offsetof(struct sk_msg_md, data_end):
8266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8267 si->dst_reg, si->src_reg,
8268 offsetof(struct sk_msg, data_end));
8269 break;
8270 case offsetof(struct sk_msg_md, family):
8271 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
8272
8273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8274 struct sk_msg, sk),
8275 si->dst_reg, si->src_reg,
8276 offsetof(struct sk_msg, sk));
8277 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8278 offsetof(struct sock_common, skc_family));
8279 break;
8280
8281 case offsetof(struct sk_msg_md, remote_ip4):
8282 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8283
8284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8285 struct sk_msg, sk),
8286 si->dst_reg, si->src_reg,
8287 offsetof(struct sk_msg, sk));
8288 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8289 offsetof(struct sock_common, skc_daddr));
8290 break;
8291
8292 case offsetof(struct sk_msg_md, local_ip4):
8293 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8294 skc_rcv_saddr) != 4);
8295
8296 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8297 struct sk_msg, sk),
8298 si->dst_reg, si->src_reg,
8299 offsetof(struct sk_msg, sk));
8300 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8301 offsetof(struct sock_common,
8302 skc_rcv_saddr));
8303 break;
8304
8305 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8306 offsetof(struct sk_msg_md, remote_ip6[3]):
8307 #if IS_ENABLED(CONFIG_IPV6)
8308 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8309 skc_v6_daddr.s6_addr32[0]) != 4);
8310
8311 off = si->off;
8312 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8313 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8314 struct sk_msg, sk),
8315 si->dst_reg, si->src_reg,
8316 offsetof(struct sk_msg, sk));
8317 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8318 offsetof(struct sock_common,
8319 skc_v6_daddr.s6_addr32[0]) +
8320 off);
8321 #else
8322 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8323 #endif
8324 break;
8325
8326 case offsetof(struct sk_msg_md, local_ip6[0]) ...
8327 offsetof(struct sk_msg_md, local_ip6[3]):
8328 #if IS_ENABLED(CONFIG_IPV6)
8329 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8330 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8331
8332 off = si->off;
8333 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8334 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8335 struct sk_msg, sk),
8336 si->dst_reg, si->src_reg,
8337 offsetof(struct sk_msg, sk));
8338 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8339 offsetof(struct sock_common,
8340 skc_v6_rcv_saddr.s6_addr32[0]) +
8341 off);
8342 #else
8343 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8344 #endif
8345 break;
8346
8347 case offsetof(struct sk_msg_md, remote_port):
8348 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8349
8350 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8351 struct sk_msg, sk),
8352 si->dst_reg, si->src_reg,
8353 offsetof(struct sk_msg, sk));
8354 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8355 offsetof(struct sock_common, skc_dport));
8356 #ifndef __BIG_ENDIAN_BITFIELD
8357 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8358 #endif
8359 break;
8360
8361 case offsetof(struct sk_msg_md, local_port):
8362 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8363
8364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8365 struct sk_msg, sk),
8366 si->dst_reg, si->src_reg,
8367 offsetof(struct sk_msg, sk));
8368 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8369 offsetof(struct sock_common, skc_num));
8370 break;
8371
8372 case offsetof(struct sk_msg_md, size):
8373 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8374 si->dst_reg, si->src_reg,
8375 offsetof(struct sk_msg_sg, size));
8376 break;
8377 }
8378
8379 return insn - insn_buf;
8380 }
8381
8382 const struct bpf_verifier_ops sk_filter_verifier_ops = {
8383 .get_func_proto = sk_filter_func_proto,
8384 .is_valid_access = sk_filter_is_valid_access,
8385 .convert_ctx_access = bpf_convert_ctx_access,
8386 .gen_ld_abs = bpf_gen_ld_abs,
8387 };
8388
8389 const struct bpf_prog_ops sk_filter_prog_ops = {
8390 .test_run = bpf_prog_test_run_skb,
8391 };
8392
8393 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8394 .get_func_proto = tc_cls_act_func_proto,
8395 .is_valid_access = tc_cls_act_is_valid_access,
8396 .convert_ctx_access = tc_cls_act_convert_ctx_access,
8397 .gen_prologue = tc_cls_act_prologue,
8398 .gen_ld_abs = bpf_gen_ld_abs,
8399 };
8400
8401 const struct bpf_prog_ops tc_cls_act_prog_ops = {
8402 .test_run = bpf_prog_test_run_skb,
8403 };
8404
8405 const struct bpf_verifier_ops xdp_verifier_ops = {
8406 .get_func_proto = xdp_func_proto,
8407 .is_valid_access = xdp_is_valid_access,
8408 .convert_ctx_access = xdp_convert_ctx_access,
8409 .gen_prologue = bpf_noop_prologue,
8410 };
8411
8412 const struct bpf_prog_ops xdp_prog_ops = {
8413 .test_run = bpf_prog_test_run_xdp,
8414 };
8415
8416 const struct bpf_verifier_ops cg_skb_verifier_ops = {
8417 .get_func_proto = cg_skb_func_proto,
8418 .is_valid_access = cg_skb_is_valid_access,
8419 .convert_ctx_access = bpf_convert_ctx_access,
8420 };
8421
8422 const struct bpf_prog_ops cg_skb_prog_ops = {
8423 .test_run = bpf_prog_test_run_skb,
8424 };
8425
8426 const struct bpf_verifier_ops lwt_in_verifier_ops = {
8427 .get_func_proto = lwt_in_func_proto,
8428 .is_valid_access = lwt_is_valid_access,
8429 .convert_ctx_access = bpf_convert_ctx_access,
8430 };
8431
8432 const struct bpf_prog_ops lwt_in_prog_ops = {
8433 .test_run = bpf_prog_test_run_skb,
8434 };
8435
8436 const struct bpf_verifier_ops lwt_out_verifier_ops = {
8437 .get_func_proto = lwt_out_func_proto,
8438 .is_valid_access = lwt_is_valid_access,
8439 .convert_ctx_access = bpf_convert_ctx_access,
8440 };
8441
8442 const struct bpf_prog_ops lwt_out_prog_ops = {
8443 .test_run = bpf_prog_test_run_skb,
8444 };
8445
8446 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8447 .get_func_proto = lwt_xmit_func_proto,
8448 .is_valid_access = lwt_is_valid_access,
8449 .convert_ctx_access = bpf_convert_ctx_access,
8450 .gen_prologue = tc_cls_act_prologue,
8451 };
8452
8453 const struct bpf_prog_ops lwt_xmit_prog_ops = {
8454 .test_run = bpf_prog_test_run_skb,
8455 };
8456
8457 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8458 .get_func_proto = lwt_seg6local_func_proto,
8459 .is_valid_access = lwt_is_valid_access,
8460 .convert_ctx_access = bpf_convert_ctx_access,
8461 };
8462
8463 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8464 .test_run = bpf_prog_test_run_skb,
8465 };
8466
8467 const struct bpf_verifier_ops cg_sock_verifier_ops = {
8468 .get_func_proto = sock_filter_func_proto,
8469 .is_valid_access = sock_filter_is_valid_access,
8470 .convert_ctx_access = bpf_sock_convert_ctx_access,
8471 };
8472
8473 const struct bpf_prog_ops cg_sock_prog_ops = {
8474 };
8475
8476 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8477 .get_func_proto = sock_addr_func_proto,
8478 .is_valid_access = sock_addr_is_valid_access,
8479 .convert_ctx_access = sock_addr_convert_ctx_access,
8480 };
8481
8482 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8483 };
8484
8485 const struct bpf_verifier_ops sock_ops_verifier_ops = {
8486 .get_func_proto = sock_ops_func_proto,
8487 .is_valid_access = sock_ops_is_valid_access,
8488 .convert_ctx_access = sock_ops_convert_ctx_access,
8489 };
8490
8491 const struct bpf_prog_ops sock_ops_prog_ops = {
8492 };
8493
8494 const struct bpf_verifier_ops sk_skb_verifier_ops = {
8495 .get_func_proto = sk_skb_func_proto,
8496 .is_valid_access = sk_skb_is_valid_access,
8497 .convert_ctx_access = sk_skb_convert_ctx_access,
8498 .gen_prologue = sk_skb_prologue,
8499 };
8500
8501 const struct bpf_prog_ops sk_skb_prog_ops = {
8502 };
8503
8504 const struct bpf_verifier_ops sk_msg_verifier_ops = {
8505 .get_func_proto = sk_msg_func_proto,
8506 .is_valid_access = sk_msg_is_valid_access,
8507 .convert_ctx_access = sk_msg_convert_ctx_access,
8508 .gen_prologue = bpf_noop_prologue,
8509 };
8510
8511 const struct bpf_prog_ops sk_msg_prog_ops = {
8512 };
8513
8514 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8515 .get_func_proto = flow_dissector_func_proto,
8516 .is_valid_access = flow_dissector_is_valid_access,
8517 .convert_ctx_access = flow_dissector_convert_ctx_access,
8518 };
8519
8520 const struct bpf_prog_ops flow_dissector_prog_ops = {
8521 .test_run = bpf_prog_test_run_flow_dissector,
8522 };
8523
8524 int sk_detach_filter(struct sock *sk)
8525 {
8526 int ret = -ENOENT;
8527 struct sk_filter *filter;
8528
8529 if (sock_flag(sk, SOCK_FILTER_LOCKED))
8530 return -EPERM;
8531
8532 filter = rcu_dereference_protected(sk->sk_filter,
8533 lockdep_sock_is_held(sk));
8534 if (filter) {
8535 RCU_INIT_POINTER(sk->sk_filter, NULL);
8536 sk_filter_uncharge(sk, filter);
8537 ret = 0;
8538 }
8539
8540 return ret;
8541 }
8542 EXPORT_SYMBOL_GPL(sk_detach_filter);
8543
8544 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8545 unsigned int len)
8546 {
8547 struct sock_fprog_kern *fprog;
8548 struct sk_filter *filter;
8549 int ret = 0;
8550
8551 lock_sock(sk);
8552 filter = rcu_dereference_protected(sk->sk_filter,
8553 lockdep_sock_is_held(sk));
8554 if (!filter)
8555 goto out;
8556
8557 /* We're copying the filter that has been originally attached,
8558 * so no conversion/decode needed anymore. eBPF programs that
8559 * have no original program cannot be dumped through this.
8560 */
8561 ret = -EACCES;
8562 fprog = filter->prog->orig_prog;
8563 if (!fprog)
8564 goto out;
8565
8566 ret = fprog->len;
8567 if (!len)
8568 /* User space only enquires number of filter blocks. */
8569 goto out;
8570
8571 ret = -EINVAL;
8572 if (len < fprog->len)
8573 goto out;
8574
8575 ret = -EFAULT;
8576 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8577 goto out;
8578
8579 /* Instead of bytes, the API requests to return the number
8580 * of filter blocks.
8581 */
8582 ret = fprog->len;
8583 out:
8584 release_sock(sk);
8585 return ret;
8586 }
8587
8588 #ifdef CONFIG_INET
8589 struct sk_reuseport_kern {
8590 struct sk_buff *skb;
8591 struct sock *sk;
8592 struct sock *selected_sk;
8593 void *data_end;
8594 u32 hash;
8595 u32 reuseport_id;
8596 bool bind_inany;
8597 };
8598
8599 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8600 struct sock_reuseport *reuse,
8601 struct sock *sk, struct sk_buff *skb,
8602 u32 hash)
8603 {
8604 reuse_kern->skb = skb;
8605 reuse_kern->sk = sk;
8606 reuse_kern->selected_sk = NULL;
8607 reuse_kern->data_end = skb->data + skb_headlen(skb);
8608 reuse_kern->hash = hash;
8609 reuse_kern->reuseport_id = reuse->reuseport_id;
8610 reuse_kern->bind_inany = reuse->bind_inany;
8611 }
8612
8613 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8614 struct bpf_prog *prog, struct sk_buff *skb,
8615 u32 hash)
8616 {
8617 struct sk_reuseport_kern reuse_kern;
8618 enum sk_action action;
8619
8620 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8621 action = BPF_PROG_RUN(prog, &reuse_kern);
8622
8623 if (action == SK_PASS)
8624 return reuse_kern.selected_sk;
8625 else
8626 return ERR_PTR(-ECONNREFUSED);
8627 }
8628
8629 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8630 struct bpf_map *, map, void *, key, u32, flags)
8631 {
8632 struct sock_reuseport *reuse;
8633 struct sock *selected_sk;
8634
8635 selected_sk = map->ops->map_lookup_elem(map, key);
8636 if (!selected_sk)
8637 return -ENOENT;
8638
8639 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8640 if (!reuse)
8641 /* selected_sk is unhashed (e.g. by close()) after the
8642 * above map_lookup_elem(). Treat selected_sk has already
8643 * been removed from the map.
8644 */
8645 return -ENOENT;
8646
8647 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8648 struct sock *sk;
8649
8650 if (unlikely(!reuse_kern->reuseport_id))
8651 /* There is a small race between adding the
8652 * sk to the map and setting the
8653 * reuse_kern->reuseport_id.
8654 * Treat it as the sk has not been added to
8655 * the bpf map yet.
8656 */
8657 return -ENOENT;
8658
8659 sk = reuse_kern->sk;
8660 if (sk->sk_protocol != selected_sk->sk_protocol)
8661 return -EPROTOTYPE;
8662 else if (sk->sk_family != selected_sk->sk_family)
8663 return -EAFNOSUPPORT;
8664
8665 /* Catch all. Likely bound to a different sockaddr. */
8666 return -EBADFD;
8667 }
8668
8669 reuse_kern->selected_sk = selected_sk;
8670
8671 return 0;
8672 }
8673
8674 static const struct bpf_func_proto sk_select_reuseport_proto = {
8675 .func = sk_select_reuseport,
8676 .gpl_only = false,
8677 .ret_type = RET_INTEGER,
8678 .arg1_type = ARG_PTR_TO_CTX,
8679 .arg2_type = ARG_CONST_MAP_PTR,
8680 .arg3_type = ARG_PTR_TO_MAP_KEY,
8681 .arg4_type = ARG_ANYTHING,
8682 };
8683
8684 BPF_CALL_4(sk_reuseport_load_bytes,
8685 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8686 void *, to, u32, len)
8687 {
8688 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8689 }
8690
8691 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8692 .func = sk_reuseport_load_bytes,
8693 .gpl_only = false,
8694 .ret_type = RET_INTEGER,
8695 .arg1_type = ARG_PTR_TO_CTX,
8696 .arg2_type = ARG_ANYTHING,
8697 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8698 .arg4_type = ARG_CONST_SIZE,
8699 };
8700
8701 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8702 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8703 void *, to, u32, len, u32, start_header)
8704 {
8705 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8706 len, start_header);
8707 }
8708
8709 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8710 .func = sk_reuseport_load_bytes_relative,
8711 .gpl_only = false,
8712 .ret_type = RET_INTEGER,
8713 .arg1_type = ARG_PTR_TO_CTX,
8714 .arg2_type = ARG_ANYTHING,
8715 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8716 .arg4_type = ARG_CONST_SIZE,
8717 .arg5_type = ARG_ANYTHING,
8718 };
8719
8720 static const struct bpf_func_proto *
8721 sk_reuseport_func_proto(enum bpf_func_id func_id,
8722 const struct bpf_prog *prog)
8723 {
8724 switch (func_id) {
8725 case BPF_FUNC_sk_select_reuseport:
8726 return &sk_select_reuseport_proto;
8727 case BPF_FUNC_skb_load_bytes:
8728 return &sk_reuseport_load_bytes_proto;
8729 case BPF_FUNC_skb_load_bytes_relative:
8730 return &sk_reuseport_load_bytes_relative_proto;
8731 default:
8732 return bpf_base_func_proto(func_id);
8733 }
8734 }
8735
8736 static bool
8737 sk_reuseport_is_valid_access(int off, int size,
8738 enum bpf_access_type type,
8739 const struct bpf_prog *prog,
8740 struct bpf_insn_access_aux *info)
8741 {
8742 const u32 size_default = sizeof(__u32);
8743
8744 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8745 off % size || type != BPF_READ)
8746 return false;
8747
8748 switch (off) {
8749 case offsetof(struct sk_reuseport_md, data):
8750 info->reg_type = PTR_TO_PACKET;
8751 return size == sizeof(__u64);
8752
8753 case offsetof(struct sk_reuseport_md, data_end):
8754 info->reg_type = PTR_TO_PACKET_END;
8755 return size == sizeof(__u64);
8756
8757 case offsetof(struct sk_reuseport_md, hash):
8758 return size == size_default;
8759
8760 /* Fields that allow narrowing */
8761 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
8762 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8763 return false;
8764 /* fall through */
8765 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
8766 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
8767 case bpf_ctx_range(struct sk_reuseport_md, len):
8768 bpf_ctx_record_field_size(info, size_default);
8769 return bpf_ctx_narrow_access_ok(off, size, size_default);
8770
8771 default:
8772 return false;
8773 }
8774 }
8775
8776 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8778 si->dst_reg, si->src_reg, \
8779 bpf_target_off(struct sk_reuseport_kern, F, \
8780 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8781 target_size)); \
8782 })
8783
8784 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8785 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8786 struct sk_buff, \
8787 skb, \
8788 SKB_FIELD)
8789
8790 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8791 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
8792 struct sock, \
8793 sk, \
8794 SK_FIELD, BPF_SIZE, EXTRA_OFF)
8795
8796 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8797 const struct bpf_insn *si,
8798 struct bpf_insn *insn_buf,
8799 struct bpf_prog *prog,
8800 u32 *target_size)
8801 {
8802 struct bpf_insn *insn = insn_buf;
8803
8804 switch (si->off) {
8805 case offsetof(struct sk_reuseport_md, data):
8806 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8807 break;
8808
8809 case offsetof(struct sk_reuseport_md, len):
8810 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8811 break;
8812
8813 case offsetof(struct sk_reuseport_md, eth_protocol):
8814 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8815 break;
8816
8817 case offsetof(struct sk_reuseport_md, ip_protocol):
8818 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8819 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8820 BPF_W, 0);
8821 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8822 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8823 SK_FL_PROTO_SHIFT);
8824 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8825 * aware. No further narrowing or masking is needed.
8826 */
8827 *target_size = 1;
8828 break;
8829
8830 case offsetof(struct sk_reuseport_md, data_end):
8831 SK_REUSEPORT_LOAD_FIELD(data_end);
8832 break;
8833
8834 case offsetof(struct sk_reuseport_md, hash):
8835 SK_REUSEPORT_LOAD_FIELD(hash);
8836 break;
8837
8838 case offsetof(struct sk_reuseport_md, bind_inany):
8839 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8840 break;
8841 }
8842
8843 return insn - insn_buf;
8844 }
8845
8846 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8847 .get_func_proto = sk_reuseport_func_proto,
8848 .is_valid_access = sk_reuseport_is_valid_access,
8849 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8850 };
8851
8852 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8853 };
8854 #endif /* CONFIG_INET */
Ubuntu Eoan Linux kernel mirror