1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/atomic.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
23 #include <linux/mm.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
39 #include <net/sock.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.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 <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
54 #include <net/dst.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
57 #include <net/tcp.h>
58 #include <net/xfrm.h>
59 #include <net/udp.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
67 #include <net/flow.h>
68 #include <net/arp.h>
69 #include <net/ipv6.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
72 #include <net/seg6.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
79 #include <net/tls.h>
80 #include <net/xdp.h>
81 #include <net/mptcp.h>
82
83 static const struct bpf_func_proto *
84 bpf_sk_base_func_proto(enum bpf_func_id func_id);
85
86 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
87 {
88 if (in_compat_syscall()) {
89 struct compat_sock_fprog f32;
90
91 if (len != sizeof(f32))
92 return -EINVAL;
93 if (copy_from_sockptr(&f32, src, sizeof(f32)))
94 return -EFAULT;
95 memset(dst, 0, sizeof(*dst));
96 dst->len = f32.len;
97 dst->filter = compat_ptr(f32.filter);
98 } else {
99 if (len != sizeof(*dst))
100 return -EINVAL;
101 if (copy_from_sockptr(dst, src, sizeof(*dst)))
102 return -EFAULT;
103 }
104
105 return 0;
106 }
107 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
108
109 /**
110 * sk_filter_trim_cap - run a packet through a socket filter
111 * @sk: sock associated with &sk_buff
112 * @skb: buffer to filter
113 * @cap: limit on how short the eBPF program may trim the packet
114 *
115 * Run the eBPF program and then cut skb->data to correct size returned by
116 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
117 * than pkt_len we keep whole skb->data. This is the socket level
118 * wrapper to bpf_prog_run. It returns 0 if the packet should
119 * be accepted or -EPERM if the packet should be tossed.
120 *
121 */
122 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
123 {
124 int err;
125 struct sk_filter *filter;
126
127 /*
128 * If the skb was allocated from pfmemalloc reserves, only
129 * allow SOCK_MEMALLOC sockets to use it as this socket is
130 * helping free memory
131 */
132 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
133 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
134 return -ENOMEM;
135 }
136 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
137 if (err)
138 return err;
139
140 err = security_sock_rcv_skb(sk, skb);
141 if (err)
142 return err;
143
144 rcu_read_lock();
145 filter = rcu_dereference(sk->sk_filter);
146 if (filter) {
147 struct sock *save_sk = skb->sk;
148 unsigned int pkt_len;
149
150 skb->sk = sk;
151 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
152 skb->sk = save_sk;
153 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
154 }
155 rcu_read_unlock();
156
157 return err;
158 }
159 EXPORT_SYMBOL(sk_filter_trim_cap);
160
161 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
162 {
163 return skb_get_poff(skb);
164 }
165
166 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
167 {
168 struct nlattr *nla;
169
170 if (skb_is_nonlinear(skb))
171 return 0;
172
173 if (skb->len < sizeof(struct nlattr))
174 return 0;
175
176 if (a > skb->len - sizeof(struct nlattr))
177 return 0;
178
179 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
180 if (nla)
181 return (void *) nla - (void *) skb->data;
182
183 return 0;
184 }
185
186 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
187 {
188 struct nlattr *nla;
189
190 if (skb_is_nonlinear(skb))
191 return 0;
192
193 if (skb->len < sizeof(struct nlattr))
194 return 0;
195
196 if (a > skb->len - sizeof(struct nlattr))
197 return 0;
198
199 nla = (struct nlattr *) &skb->data[a];
200 if (nla->nla_len > skb->len - a)
201 return 0;
202
203 nla = nla_find_nested(nla, x);
204 if (nla)
205 return (void *) nla - (void *) skb->data;
206
207 return 0;
208 }
209
210 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
211 data, int, headlen, int, offset)
212 {
213 u8 tmp, *ptr;
214 const int len = sizeof(tmp);
215
216 if (offset >= 0) {
217 if (headlen - offset >= len)
218 return *(u8 *)(data + offset);
219 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
220 return tmp;
221 } else {
222 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
223 if (likely(ptr))
224 return *(u8 *)ptr;
225 }
226
227 return -EFAULT;
228 }
229
230 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
231 int, offset)
232 {
233 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
234 offset);
235 }
236
237 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
238 data, int, headlen, int, offset)
239 {
240 __be16 tmp, *ptr;
241 const int len = sizeof(tmp);
242
243 if (offset >= 0) {
244 if (headlen - offset >= len)
245 return get_unaligned_be16(data + offset);
246 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
247 return be16_to_cpu(tmp);
248 } else {
249 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
250 if (likely(ptr))
251 return get_unaligned_be16(ptr);
252 }
253
254 return -EFAULT;
255 }
256
257 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
258 int, offset)
259 {
260 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
261 offset);
262 }
263
264 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
265 data, int, headlen, int, offset)
266 {
267 __be32 tmp, *ptr;
268 const int len = sizeof(tmp);
269
270 if (likely(offset >= 0)) {
271 if (headlen - offset >= len)
272 return get_unaligned_be32(data + offset);
273 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
274 return be32_to_cpu(tmp);
275 } else {
276 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
277 if (likely(ptr))
278 return get_unaligned_be32(ptr);
279 }
280
281 return -EFAULT;
282 }
283
284 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
285 int, offset)
286 {
287 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
288 offset);
289 }
290
291 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
292 struct bpf_insn *insn_buf)
293 {
294 struct bpf_insn *insn = insn_buf;
295
296 switch (skb_field) {
297 case SKF_AD_MARK:
298 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
299
300 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
301 offsetof(struct sk_buff, mark));
302 break;
303
304 case SKF_AD_PKTTYPE:
305 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
306 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
307 #ifdef __BIG_ENDIAN_BITFIELD
308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
309 #endif
310 break;
311
312 case SKF_AD_QUEUE:
313 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
314
315 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
316 offsetof(struct sk_buff, queue_mapping));
317 break;
318
319 case SKF_AD_VLAN_TAG:
320 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
321
322 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
323 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
324 offsetof(struct sk_buff, vlan_tci));
325 break;
326 case SKF_AD_VLAN_TAG_PRESENT:
327 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
328 if (PKT_VLAN_PRESENT_BIT)
329 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
330 if (PKT_VLAN_PRESENT_BIT < 7)
331 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
332 break;
333 }
334
335 return insn - insn_buf;
336 }
337
338 static bool convert_bpf_extensions(struct sock_filter *fp,
339 struct bpf_insn **insnp)
340 {
341 struct bpf_insn *insn = *insnp;
342 u32 cnt;
343
344 switch (fp->k) {
345 case SKF_AD_OFF + SKF_AD_PROTOCOL:
346 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
347
348 /* A = *(u16 *) (CTX + offsetof(protocol)) */
349 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
350 offsetof(struct sk_buff, protocol));
351 /* A = ntohs(A) [emitting a nop or swap16] */
352 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
353 break;
354
355 case SKF_AD_OFF + SKF_AD_PKTTYPE:
356 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
357 insn += cnt - 1;
358 break;
359
360 case SKF_AD_OFF + SKF_AD_IFINDEX:
361 case SKF_AD_OFF + SKF_AD_HATYPE:
362 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
363 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
364
365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
366 BPF_REG_TMP, BPF_REG_CTX,
367 offsetof(struct sk_buff, dev));
368 /* if (tmp != 0) goto pc + 1 */
369 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
370 *insn++ = BPF_EXIT_INSN();
371 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
372 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
373 offsetof(struct net_device, ifindex));
374 else
375 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
376 offsetof(struct net_device, type));
377 break;
378
379 case SKF_AD_OFF + SKF_AD_MARK:
380 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
381 insn += cnt - 1;
382 break;
383
384 case SKF_AD_OFF + SKF_AD_RXHASH:
385 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
386
387 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
388 offsetof(struct sk_buff, hash));
389 break;
390
391 case SKF_AD_OFF + SKF_AD_QUEUE:
392 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
393 insn += cnt - 1;
394 break;
395
396 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
397 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
398 BPF_REG_A, BPF_REG_CTX, insn);
399 insn += cnt - 1;
400 break;
401
402 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
403 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
404 BPF_REG_A, BPF_REG_CTX, insn);
405 insn += cnt - 1;
406 break;
407
408 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
409 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
410
411 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
412 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
413 offsetof(struct sk_buff, vlan_proto));
414 /* A = ntohs(A) [emitting a nop or swap16] */
415 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
416 break;
417
418 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
419 case SKF_AD_OFF + SKF_AD_NLATTR:
420 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
421 case SKF_AD_OFF + SKF_AD_CPU:
422 case SKF_AD_OFF + SKF_AD_RANDOM:
423 /* arg1 = CTX */
424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
425 /* arg2 = A */
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
427 /* arg3 = X */
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
429 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
430 switch (fp->k) {
431 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
432 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
433 break;
434 case SKF_AD_OFF + SKF_AD_NLATTR:
435 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
436 break;
437 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
438 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
439 break;
440 case SKF_AD_OFF + SKF_AD_CPU:
441 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
442 break;
443 case SKF_AD_OFF + SKF_AD_RANDOM:
444 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
445 bpf_user_rnd_init_once();
446 break;
447 }
448 break;
449
450 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
451 /* A ^= X */
452 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
453 break;
454
455 default:
456 /* This is just a dummy call to avoid letting the compiler
457 * evict __bpf_call_base() as an optimization. Placed here
458 * where no-one bothers.
459 */
460 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
461 return false;
462 }
463
464 *insnp = insn;
465 return true;
466 }
467
468 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
469 {
470 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
471 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
472 bool endian = BPF_SIZE(fp->code) == BPF_H ||
473 BPF_SIZE(fp->code) == BPF_W;
474 bool indirect = BPF_MODE(fp->code) == BPF_IND;
475 const int ip_align = NET_IP_ALIGN;
476 struct bpf_insn *insn = *insnp;
477 int offset = fp->k;
478
479 if (!indirect &&
480 ((unaligned_ok && offset >= 0) ||
481 (!unaligned_ok && offset >= 0 &&
482 offset + ip_align >= 0 &&
483 offset + ip_align % size == 0))) {
484 bool ldx_off_ok = offset <= S16_MAX;
485
486 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
487 if (offset)
488 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
489 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
490 size, 2 + endian + (!ldx_off_ok * 2));
491 if (ldx_off_ok) {
492 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
493 BPF_REG_D, offset);
494 } else {
495 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
496 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
497 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
498 BPF_REG_TMP, 0);
499 }
500 if (endian)
501 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
502 *insn++ = BPF_JMP_A(8);
503 }
504
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
508 if (!indirect) {
509 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
510 } else {
511 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
512 if (fp->k)
513 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
514 }
515
516 switch (BPF_SIZE(fp->code)) {
517 case BPF_B:
518 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
519 break;
520 case BPF_H:
521 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
522 break;
523 case BPF_W:
524 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
525 break;
526 default:
527 return false;
528 }
529
530 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
531 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
532 *insn = BPF_EXIT_INSN();
533
534 *insnp = insn;
535 return true;
536 }
537
538 /**
539 * bpf_convert_filter - convert filter program
540 * @prog: the user passed filter program
541 * @len: the length of the user passed filter program
542 * @new_prog: allocated 'struct bpf_prog' or NULL
543 * @new_len: pointer to store length of converted program
544 * @seen_ld_abs: bool whether we've seen ld_abs/ind
545 *
546 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
547 * style extended BPF (eBPF).
548 * Conversion workflow:
549 *
550 * 1) First pass for calculating the new program length:
551 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
552 *
553 * 2) 2nd pass to remap in two passes: 1st pass finds new
554 * jump offsets, 2nd pass remapping:
555 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
556 */
557 static int bpf_convert_filter(struct sock_filter *prog, int len,
558 struct bpf_prog *new_prog, int *new_len,
559 bool *seen_ld_abs)
560 {
561 int new_flen = 0, pass = 0, target, i, stack_off;
562 struct bpf_insn *new_insn, *first_insn = NULL;
563 struct sock_filter *fp;
564 int *addrs = NULL;
565 u8 bpf_src;
566
567 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
568 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
569
570 if (len <= 0 || len > BPF_MAXINSNS)
571 return -EINVAL;
572
573 if (new_prog) {
574 first_insn = new_prog->insnsi;
575 addrs = kcalloc(len, sizeof(*addrs),
576 GFP_KERNEL | __GFP_NOWARN);
577 if (!addrs)
578 return -ENOMEM;
579 }
580
581 do_pass:
582 new_insn = first_insn;
583 fp = prog;
584
585 /* Classic BPF related prologue emission. */
586 if (new_prog) {
587 /* Classic BPF expects A and X to be reset first. These need
588 * to be guaranteed to be the first two instructions.
589 */
590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
591 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
592
593 /* All programs must keep CTX in callee saved BPF_REG_CTX.
594 * In eBPF case it's done by the compiler, here we need to
595 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
596 */
597 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
598 if (*seen_ld_abs) {
599 /* For packet access in classic BPF, cache skb->data
600 * in callee-saved BPF R8 and skb->len - skb->data_len
601 * (headlen) in BPF R9. Since classic BPF is read-only
602 * on CTX, we only need to cache it once.
603 */
604 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
605 BPF_REG_D, BPF_REG_CTX,
606 offsetof(struct sk_buff, data));
607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
608 offsetof(struct sk_buff, len));
609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
610 offsetof(struct sk_buff, data_len));
611 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
612 }
613 } else {
614 new_insn += 3;
615 }
616
617 for (i = 0; i < len; fp++, i++) {
618 struct bpf_insn tmp_insns[32] = { };
619 struct bpf_insn *insn = tmp_insns;
620
621 if (addrs)
622 addrs[i] = new_insn - first_insn;
623
624 switch (fp->code) {
625 /* All arithmetic insns and skb loads map as-is. */
626 case BPF_ALU | BPF_ADD | BPF_X:
627 case BPF_ALU | BPF_ADD | BPF_K:
628 case BPF_ALU | BPF_SUB | BPF_X:
629 case BPF_ALU | BPF_SUB | BPF_K:
630 case BPF_ALU | BPF_AND | BPF_X:
631 case BPF_ALU | BPF_AND | BPF_K:
632 case BPF_ALU | BPF_OR | BPF_X:
633 case BPF_ALU | BPF_OR | BPF_K:
634 case BPF_ALU | BPF_LSH | BPF_X:
635 case BPF_ALU | BPF_LSH | BPF_K:
636 case BPF_ALU | BPF_RSH | BPF_X:
637 case BPF_ALU | BPF_RSH | BPF_K:
638 case BPF_ALU | BPF_XOR | BPF_X:
639 case BPF_ALU | BPF_XOR | BPF_K:
640 case BPF_ALU | BPF_MUL | BPF_X:
641 case BPF_ALU | BPF_MUL | BPF_K:
642 case BPF_ALU | BPF_DIV | BPF_X:
643 case BPF_ALU | BPF_DIV | BPF_K:
644 case BPF_ALU | BPF_MOD | BPF_X:
645 case BPF_ALU | BPF_MOD | BPF_K:
646 case BPF_ALU | BPF_NEG:
647 case BPF_LD | BPF_ABS | BPF_W:
648 case BPF_LD | BPF_ABS | BPF_H:
649 case BPF_LD | BPF_ABS | BPF_B:
650 case BPF_LD | BPF_IND | BPF_W:
651 case BPF_LD | BPF_IND | BPF_H:
652 case BPF_LD | BPF_IND | BPF_B:
653 /* Check for overloaded BPF extension and
654 * directly convert it if found, otherwise
655 * just move on with mapping.
656 */
657 if (BPF_CLASS(fp->code) == BPF_LD &&
658 BPF_MODE(fp->code) == BPF_ABS &&
659 convert_bpf_extensions(fp, &insn))
660 break;
661 if (BPF_CLASS(fp->code) == BPF_LD &&
662 convert_bpf_ld_abs(fp, &insn)) {
663 *seen_ld_abs = true;
664 break;
665 }
666
667 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
668 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
669 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
670 /* Error with exception code on div/mod by 0.
671 * For cBPF programs, this was always return 0.
672 */
673 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
674 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
675 *insn++ = BPF_EXIT_INSN();
676 }
677
678 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
679 break;
680
681 /* Jump transformation cannot use BPF block macros
682 * everywhere as offset calculation and target updates
683 * require a bit more work than the rest, i.e. jump
684 * opcodes map as-is, but offsets need adjustment.
685 */
686
687 #define BPF_EMIT_JMP \
688 do { \
689 const s32 off_min = S16_MIN, off_max = S16_MAX; \
690 s32 off; \
691 \
692 if (target >= len || target < 0) \
693 goto err; \
694 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
695 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
696 off -= insn - tmp_insns; \
697 /* Reject anything not fitting into insn->off. */ \
698 if (off < off_min || off > off_max) \
699 goto err; \
700 insn->off = off; \
701 } while (0)
702
703 case BPF_JMP | BPF_JA:
704 target = i + fp->k + 1;
705 insn->code = fp->code;
706 BPF_EMIT_JMP;
707 break;
708
709 case BPF_JMP | BPF_JEQ | BPF_K:
710 case BPF_JMP | BPF_JEQ | BPF_X:
711 case BPF_JMP | BPF_JSET | BPF_K:
712 case BPF_JMP | BPF_JSET | BPF_X:
713 case BPF_JMP | BPF_JGT | BPF_K:
714 case BPF_JMP | BPF_JGT | BPF_X:
715 case BPF_JMP | BPF_JGE | BPF_K:
716 case BPF_JMP | BPF_JGE | BPF_X:
717 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
718 /* BPF immediates are signed, zero extend
719 * immediate into tmp register and use it
720 * in compare insn.
721 */
722 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
723
724 insn->dst_reg = BPF_REG_A;
725 insn->src_reg = BPF_REG_TMP;
726 bpf_src = BPF_X;
727 } else {
728 insn->dst_reg = BPF_REG_A;
729 insn->imm = fp->k;
730 bpf_src = BPF_SRC(fp->code);
731 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
732 }
733
734 /* Common case where 'jump_false' is next insn. */
735 if (fp->jf == 0) {
736 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
737 target = i + fp->jt + 1;
738 BPF_EMIT_JMP;
739 break;
740 }
741
742 /* Convert some jumps when 'jump_true' is next insn. */
743 if (fp->jt == 0) {
744 switch (BPF_OP(fp->code)) {
745 case BPF_JEQ:
746 insn->code = BPF_JMP | BPF_JNE | bpf_src;
747 break;
748 case BPF_JGT:
749 insn->code = BPF_JMP | BPF_JLE | bpf_src;
750 break;
751 case BPF_JGE:
752 insn->code = BPF_JMP | BPF_JLT | bpf_src;
753 break;
754 default:
755 goto jmp_rest;
756 }
757
758 target = i + fp->jf + 1;
759 BPF_EMIT_JMP;
760 break;
761 }
762 jmp_rest:
763 /* Other jumps are mapped into two insns: Jxx and JA. */
764 target = i + fp->jt + 1;
765 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
766 BPF_EMIT_JMP;
767 insn++;
768
769 insn->code = BPF_JMP | BPF_JA;
770 target = i + fp->jf + 1;
771 BPF_EMIT_JMP;
772 break;
773
774 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
775 case BPF_LDX | BPF_MSH | BPF_B: {
776 struct sock_filter tmp = {
777 .code = BPF_LD | BPF_ABS | BPF_B,
778 .k = fp->k,
779 };
780
781 *seen_ld_abs = true;
782
783 /* X = A */
784 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
785 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
786 convert_bpf_ld_abs(&tmp, &insn);
787 insn++;
788 /* A &= 0xf */
789 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
790 /* A <<= 2 */
791 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
792 /* tmp = X */
793 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
794 /* X = A */
795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 /* A = tmp */
797 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
798 break;
799 }
800 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
801 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
802 */
803 case BPF_RET | BPF_A:
804 case BPF_RET | BPF_K:
805 if (BPF_RVAL(fp->code) == BPF_K)
806 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
807 0, fp->k);
808 *insn = BPF_EXIT_INSN();
809 break;
810
811 /* Store to stack. */
812 case BPF_ST:
813 case BPF_STX:
814 stack_off = fp->k * 4 + 4;
815 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
816 BPF_ST ? BPF_REG_A : BPF_REG_X,
817 -stack_off);
818 /* check_load_and_stores() verifies that classic BPF can
819 * load from stack only after write, so tracking
820 * stack_depth for ST|STX insns is enough
821 */
822 if (new_prog && new_prog->aux->stack_depth < stack_off)
823 new_prog->aux->stack_depth = stack_off;
824 break;
825
826 /* Load from stack. */
827 case BPF_LD | BPF_MEM:
828 case BPF_LDX | BPF_MEM:
829 stack_off = fp->k * 4 + 4;
830 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
831 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
832 -stack_off);
833 break;
834
835 /* A = K or X = K */
836 case BPF_LD | BPF_IMM:
837 case BPF_LDX | BPF_IMM:
838 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
839 BPF_REG_A : BPF_REG_X, fp->k);
840 break;
841
842 /* X = A */
843 case BPF_MISC | BPF_TAX:
844 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
845 break;
846
847 /* A = X */
848 case BPF_MISC | BPF_TXA:
849 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
850 break;
851
852 /* A = skb->len or X = skb->len */
853 case BPF_LD | BPF_W | BPF_LEN:
854 case BPF_LDX | BPF_W | BPF_LEN:
855 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
856 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
857 offsetof(struct sk_buff, len));
858 break;
859
860 /* Access seccomp_data fields. */
861 case BPF_LDX | BPF_ABS | BPF_W:
862 /* A = *(u32 *) (ctx + K) */
863 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
864 break;
865
866 /* Unknown instruction. */
867 default:
868 goto err;
869 }
870
871 insn++;
872 if (new_prog)
873 memcpy(new_insn, tmp_insns,
874 sizeof(*insn) * (insn - tmp_insns));
875 new_insn += insn - tmp_insns;
876 }
877
878 if (!new_prog) {
879 /* Only calculating new length. */
880 *new_len = new_insn - first_insn;
881 if (*seen_ld_abs)
882 *new_len += 4; /* Prologue bits. */
883 return 0;
884 }
885
886 pass++;
887 if (new_flen != new_insn - first_insn) {
888 new_flen = new_insn - first_insn;
889 if (pass > 2)
890 goto err;
891 goto do_pass;
892 }
893
894 kfree(addrs);
895 BUG_ON(*new_len != new_flen);
896 return 0;
897 err:
898 kfree(addrs);
899 return -EINVAL;
900 }
901
902 /* Security:
903 *
904 * As we dont want to clear mem[] array for each packet going through
905 * __bpf_prog_run(), we check that filter loaded by user never try to read
906 * a cell if not previously written, and we check all branches to be sure
907 * a malicious user doesn't try to abuse us.
908 */
909 static int check_load_and_stores(const struct sock_filter *filter, int flen)
910 {
911 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
912 int pc, ret = 0;
913
914 BUILD_BUG_ON(BPF_MEMWORDS > 16);
915
916 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
917 if (!masks)
918 return -ENOMEM;
919
920 memset(masks, 0xff, flen * sizeof(*masks));
921
922 for (pc = 0; pc < flen; pc++) {
923 memvalid &= masks[pc];
924
925 switch (filter[pc].code) {
926 case BPF_ST:
927 case BPF_STX:
928 memvalid |= (1 << filter[pc].k);
929 break;
930 case BPF_LD | BPF_MEM:
931 case BPF_LDX | BPF_MEM:
932 if (!(memvalid & (1 << filter[pc].k))) {
933 ret = -EINVAL;
934 goto error;
935 }
936 break;
937 case BPF_JMP | BPF_JA:
938 /* A jump must set masks on target */
939 masks[pc + 1 + filter[pc].k] &= memvalid;
940 memvalid = ~0;
941 break;
942 case BPF_JMP | BPF_JEQ | BPF_K:
943 case BPF_JMP | BPF_JEQ | BPF_X:
944 case BPF_JMP | BPF_JGE | BPF_K:
945 case BPF_JMP | BPF_JGE | BPF_X:
946 case BPF_JMP | BPF_JGT | BPF_K:
947 case BPF_JMP | BPF_JGT | BPF_X:
948 case BPF_JMP | BPF_JSET | BPF_K:
949 case BPF_JMP | BPF_JSET | BPF_X:
950 /* A jump must set masks on targets */
951 masks[pc + 1 + filter[pc].jt] &= memvalid;
952 masks[pc + 1 + filter[pc].jf] &= memvalid;
953 memvalid = ~0;
954 break;
955 }
956 }
957 error:
958 kfree(masks);
959 return ret;
960 }
961
962 static bool chk_code_allowed(u16 code_to_probe)
963 {
964 static const bool codes[] = {
965 /* 32 bit ALU operations */
966 [BPF_ALU | BPF_ADD | BPF_K] = true,
967 [BPF_ALU | BPF_ADD | BPF_X] = true,
968 [BPF_ALU | BPF_SUB | BPF_K] = true,
969 [BPF_ALU | BPF_SUB | BPF_X] = true,
970 [BPF_ALU | BPF_MUL | BPF_K] = true,
971 [BPF_ALU | BPF_MUL | BPF_X] = true,
972 [BPF_ALU | BPF_DIV | BPF_K] = true,
973 [BPF_ALU | BPF_DIV | BPF_X] = true,
974 [BPF_ALU | BPF_MOD | BPF_K] = true,
975 [BPF_ALU | BPF_MOD | BPF_X] = true,
976 [BPF_ALU | BPF_AND | BPF_K] = true,
977 [BPF_ALU | BPF_AND | BPF_X] = true,
978 [BPF_ALU | BPF_OR | BPF_K] = true,
979 [BPF_ALU | BPF_OR | BPF_X] = true,
980 [BPF_ALU | BPF_XOR | BPF_K] = true,
981 [BPF_ALU | BPF_XOR | BPF_X] = true,
982 [BPF_ALU | BPF_LSH | BPF_K] = true,
983 [BPF_ALU | BPF_LSH | BPF_X] = true,
984 [BPF_ALU | BPF_RSH | BPF_K] = true,
985 [BPF_ALU | BPF_RSH | BPF_X] = true,
986 [BPF_ALU | BPF_NEG] = true,
987 /* Load instructions */
988 [BPF_LD | BPF_W | BPF_ABS] = true,
989 [BPF_LD | BPF_H | BPF_ABS] = true,
990 [BPF_LD | BPF_B | BPF_ABS] = true,
991 [BPF_LD | BPF_W | BPF_LEN] = true,
992 [BPF_LD | BPF_W | BPF_IND] = true,
993 [BPF_LD | BPF_H | BPF_IND] = true,
994 [BPF_LD | BPF_B | BPF_IND] = true,
995 [BPF_LD | BPF_IMM] = true,
996 [BPF_LD | BPF_MEM] = true,
997 [BPF_LDX | BPF_W | BPF_LEN] = true,
998 [BPF_LDX | BPF_B | BPF_MSH] = true,
999 [BPF_LDX | BPF_IMM] = true,
1000 [BPF_LDX | BPF_MEM] = true,
1001 /* Store instructions */
1002 [BPF_ST] = true,
1003 [BPF_STX] = true,
1004 /* Misc instructions */
1005 [BPF_MISC | BPF_TAX] = true,
1006 [BPF_MISC | BPF_TXA] = true,
1007 /* Return instructions */
1008 [BPF_RET | BPF_K] = true,
1009 [BPF_RET | BPF_A] = true,
1010 /* Jump instructions */
1011 [BPF_JMP | BPF_JA] = true,
1012 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1013 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1014 [BPF_JMP | BPF_JGE | BPF_K] = true,
1015 [BPF_JMP | BPF_JGE | BPF_X] = true,
1016 [BPF_JMP | BPF_JGT | BPF_K] = true,
1017 [BPF_JMP | BPF_JGT | BPF_X] = true,
1018 [BPF_JMP | BPF_JSET | BPF_K] = true,
1019 [BPF_JMP | BPF_JSET | BPF_X] = true,
1020 };
1021
1022 if (code_to_probe >= ARRAY_SIZE(codes))
1023 return false;
1024
1025 return codes[code_to_probe];
1026 }
1027
1028 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1029 unsigned int flen)
1030 {
1031 if (filter == NULL)
1032 return false;
1033 if (flen == 0 || flen > BPF_MAXINSNS)
1034 return false;
1035
1036 return true;
1037 }
1038
1039 /**
1040 * bpf_check_classic - verify socket filter code
1041 * @filter: filter to verify
1042 * @flen: length of filter
1043 *
1044 * Check the user's filter code. If we let some ugly
1045 * filter code slip through kaboom! The filter must contain
1046 * no references or jumps that are out of range, no illegal
1047 * instructions, and must end with a RET instruction.
1048 *
1049 * All jumps are forward as they are not signed.
1050 *
1051 * Returns 0 if the rule set is legal or -EINVAL if not.
1052 */
1053 static int bpf_check_classic(const struct sock_filter *filter,
1054 unsigned int flen)
1055 {
1056 bool anc_found;
1057 int pc;
1058
1059 /* Check the filter code now */
1060 for (pc = 0; pc < flen; pc++) {
1061 const struct sock_filter *ftest = &filter[pc];
1062
1063 /* May we actually operate on this code? */
1064 if (!chk_code_allowed(ftest->code))
1065 return -EINVAL;
1066
1067 /* Some instructions need special checks */
1068 switch (ftest->code) {
1069 case BPF_ALU | BPF_DIV | BPF_K:
1070 case BPF_ALU | BPF_MOD | BPF_K:
1071 /* Check for division by zero */
1072 if (ftest->k == 0)
1073 return -EINVAL;
1074 break;
1075 case BPF_ALU | BPF_LSH | BPF_K:
1076 case BPF_ALU | BPF_RSH | BPF_K:
1077 if (ftest->k >= 32)
1078 return -EINVAL;
1079 break;
1080 case BPF_LD | BPF_MEM:
1081 case BPF_LDX | BPF_MEM:
1082 case BPF_ST:
1083 case BPF_STX:
1084 /* Check for invalid memory addresses */
1085 if (ftest->k >= BPF_MEMWORDS)
1086 return -EINVAL;
1087 break;
1088 case BPF_JMP | BPF_JA:
1089 /* Note, the large ftest->k might cause loops.
1090 * Compare this with conditional jumps below,
1091 * where offsets are limited. --ANK (981016)
1092 */
1093 if (ftest->k >= (unsigned int)(flen - pc - 1))
1094 return -EINVAL;
1095 break;
1096 case BPF_JMP | BPF_JEQ | BPF_K:
1097 case BPF_JMP | BPF_JEQ | BPF_X:
1098 case BPF_JMP | BPF_JGE | BPF_K:
1099 case BPF_JMP | BPF_JGE | BPF_X:
1100 case BPF_JMP | BPF_JGT | BPF_K:
1101 case BPF_JMP | BPF_JGT | BPF_X:
1102 case BPF_JMP | BPF_JSET | BPF_K:
1103 case BPF_JMP | BPF_JSET | BPF_X:
1104 /* Both conditionals must be safe */
1105 if (pc + ftest->jt + 1 >= flen ||
1106 pc + ftest->jf + 1 >= flen)
1107 return -EINVAL;
1108 break;
1109 case BPF_LD | BPF_W | BPF_ABS:
1110 case BPF_LD | BPF_H | BPF_ABS:
1111 case BPF_LD | BPF_B | BPF_ABS:
1112 anc_found = false;
1113 if (bpf_anc_helper(ftest) & BPF_ANC)
1114 anc_found = true;
1115 /* Ancillary operation unknown or unsupported */
1116 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1117 return -EINVAL;
1118 }
1119 }
1120
1121 /* Last instruction must be a RET code */
1122 switch (filter[flen - 1].code) {
1123 case BPF_RET | BPF_K:
1124 case BPF_RET | BPF_A:
1125 return check_load_and_stores(filter, flen);
1126 }
1127
1128 return -EINVAL;
1129 }
1130
1131 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1132 const struct sock_fprog *fprog)
1133 {
1134 unsigned int fsize = bpf_classic_proglen(fprog);
1135 struct sock_fprog_kern *fkprog;
1136
1137 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1138 if (!fp->orig_prog)
1139 return -ENOMEM;
1140
1141 fkprog = fp->orig_prog;
1142 fkprog->len = fprog->len;
1143
1144 fkprog->filter = kmemdup(fp->insns, fsize,
1145 GFP_KERNEL | __GFP_NOWARN);
1146 if (!fkprog->filter) {
1147 kfree(fp->orig_prog);
1148 return -ENOMEM;
1149 }
1150
1151 return 0;
1152 }
1153
1154 static void bpf_release_orig_filter(struct bpf_prog *fp)
1155 {
1156 struct sock_fprog_kern *fprog = fp->orig_prog;
1157
1158 if (fprog) {
1159 kfree(fprog->filter);
1160 kfree(fprog);
1161 }
1162 }
1163
1164 static void __bpf_prog_release(struct bpf_prog *prog)
1165 {
1166 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1167 bpf_prog_put(prog);
1168 } else {
1169 bpf_release_orig_filter(prog);
1170 bpf_prog_free(prog);
1171 }
1172 }
1173
1174 static void __sk_filter_release(struct sk_filter *fp)
1175 {
1176 __bpf_prog_release(fp->prog);
1177 kfree(fp);
1178 }
1179
1180 /**
1181 * sk_filter_release_rcu - Release a socket filter by rcu_head
1182 * @rcu: rcu_head that contains the sk_filter to free
1183 */
1184 static void sk_filter_release_rcu(struct rcu_head *rcu)
1185 {
1186 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1187
1188 __sk_filter_release(fp);
1189 }
1190
1191 /**
1192 * sk_filter_release - release a socket filter
1193 * @fp: filter to remove
1194 *
1195 * Remove a filter from a socket and release its resources.
1196 */
1197 static void sk_filter_release(struct sk_filter *fp)
1198 {
1199 if (refcount_dec_and_test(&fp->refcnt))
1200 call_rcu(&fp->rcu, sk_filter_release_rcu);
1201 }
1202
1203 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1204 {
1205 u32 filter_size = bpf_prog_size(fp->prog->len);
1206
1207 atomic_sub(filter_size, &sk->sk_omem_alloc);
1208 sk_filter_release(fp);
1209 }
1210
1211 /* try to charge the socket memory if there is space available
1212 * return true on success
1213 */
1214 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1215 {
1216 u32 filter_size = bpf_prog_size(fp->prog->len);
1217 int optmem_max = READ_ONCE(sysctl_optmem_max);
1218
1219 /* same check as in sock_kmalloc() */
1220 if (filter_size <= optmem_max &&
1221 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1222 atomic_add(filter_size, &sk->sk_omem_alloc);
1223 return true;
1224 }
1225 return false;
1226 }
1227
1228 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1229 {
1230 if (!refcount_inc_not_zero(&fp->refcnt))
1231 return false;
1232
1233 if (!__sk_filter_charge(sk, fp)) {
1234 sk_filter_release(fp);
1235 return false;
1236 }
1237 return true;
1238 }
1239
1240 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1241 {
1242 struct sock_filter *old_prog;
1243 struct bpf_prog *old_fp;
1244 int err, new_len, old_len = fp->len;
1245 bool seen_ld_abs = false;
1246
1247 /* We are free to overwrite insns et al right here as it won't be used at
1248 * this point in time anymore internally after the migration to the eBPF
1249 * instruction representation.
1250 */
1251 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1252 sizeof(struct bpf_insn));
1253
1254 /* Conversion cannot happen on overlapping memory areas,
1255 * so we need to keep the user BPF around until the 2nd
1256 * pass. At this time, the user BPF is stored in fp->insns.
1257 */
1258 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1259 GFP_KERNEL | __GFP_NOWARN);
1260 if (!old_prog) {
1261 err = -ENOMEM;
1262 goto out_err;
1263 }
1264
1265 /* 1st pass: calculate the new program length. */
1266 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1267 &seen_ld_abs);
1268 if (err)
1269 goto out_err_free;
1270
1271 /* Expand fp for appending the new filter representation. */
1272 old_fp = fp;
1273 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274 if (!fp) {
1275 /* The old_fp is still around in case we couldn't
1276 * allocate new memory, so uncharge on that one.
1277 */
1278 fp = old_fp;
1279 err = -ENOMEM;
1280 goto out_err_free;
1281 }
1282
1283 fp->len = new_len;
1284
1285 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1286 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287 &seen_ld_abs);
1288 if (err)
1289 /* 2nd bpf_convert_filter() can fail only if it fails
1290 * to allocate memory, remapping must succeed. Note,
1291 * that at this time old_fp has already been released
1292 * by krealloc().
1293 */
1294 goto out_err_free;
1295
1296 fp = bpf_prog_select_runtime(fp, &err);
1297 if (err)
1298 goto out_err_free;
1299
1300 kfree(old_prog);
1301 return fp;
1302
1303 out_err_free:
1304 kfree(old_prog);
1305 out_err:
1306 __bpf_prog_release(fp);
1307 return ERR_PTR(err);
1308 }
1309
1310 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1311 bpf_aux_classic_check_t trans)
1312 {
1313 int err;
1314
1315 fp->bpf_func = NULL;
1316 fp->jited = 0;
1317
1318 err = bpf_check_classic(fp->insns, fp->len);
1319 if (err) {
1320 __bpf_prog_release(fp);
1321 return ERR_PTR(err);
1322 }
1323
1324 /* There might be additional checks and transformations
1325 * needed on classic filters, f.e. in case of seccomp.
1326 */
1327 if (trans) {
1328 err = trans(fp->insns, fp->len);
1329 if (err) {
1330 __bpf_prog_release(fp);
1331 return ERR_PTR(err);
1332 }
1333 }
1334
1335 /* Probe if we can JIT compile the filter and if so, do
1336 * the compilation of the filter.
1337 */
1338 bpf_jit_compile(fp);
1339
1340 /* JIT compiler couldn't process this filter, so do the eBPF translation
1341 * for the optimized interpreter.
1342 */
1343 if (!fp->jited)
1344 fp = bpf_migrate_filter(fp);
1345
1346 return fp;
1347 }
1348
1349 /**
1350 * bpf_prog_create - create an unattached filter
1351 * @pfp: the unattached filter that is created
1352 * @fprog: the filter program
1353 *
1354 * Create a filter independent of any socket. We first run some
1355 * sanity checks on it to make sure it does not explode on us later.
1356 * If an error occurs or there is insufficient memory for the filter
1357 * a negative errno code is returned. On success the return is zero.
1358 */
1359 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 {
1361 unsigned int fsize = bpf_classic_proglen(fprog);
1362 struct bpf_prog *fp;
1363
1364 /* Make sure new filter is there and in the right amounts. */
1365 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366 return -EINVAL;
1367
1368 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1369 if (!fp)
1370 return -ENOMEM;
1371
1372 memcpy(fp->insns, fprog->filter, fsize);
1373
1374 fp->len = fprog->len;
1375 /* Since unattached filters are not copied back to user
1376 * space through sk_get_filter(), we do not need to hold
1377 * a copy here, and can spare us the work.
1378 */
1379 fp->orig_prog = NULL;
1380
1381 /* bpf_prepare_filter() already takes care of freeing
1382 * memory in case something goes wrong.
1383 */
1384 fp = bpf_prepare_filter(fp, NULL);
1385 if (IS_ERR(fp))
1386 return PTR_ERR(fp);
1387
1388 *pfp = fp;
1389 return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392
1393 /**
1394 * bpf_prog_create_from_user - create an unattached filter from user buffer
1395 * @pfp: the unattached filter that is created
1396 * @fprog: the filter program
1397 * @trans: post-classic verifier transformation handler
1398 * @save_orig: save classic BPF program
1399 *
1400 * This function effectively does the same as bpf_prog_create(), only
1401 * that it builds up its insns buffer from user space provided buffer.
1402 * It also allows for passing a bpf_aux_classic_check_t handler.
1403 */
1404 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1405 bpf_aux_classic_check_t trans, bool save_orig)
1406 {
1407 unsigned int fsize = bpf_classic_proglen(fprog);
1408 struct bpf_prog *fp;
1409 int err;
1410
1411 /* Make sure new filter is there and in the right amounts. */
1412 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413 return -EINVAL;
1414
1415 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1416 if (!fp)
1417 return -ENOMEM;
1418
1419 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1420 __bpf_prog_free(fp);
1421 return -EFAULT;
1422 }
1423
1424 fp->len = fprog->len;
1425 fp->orig_prog = NULL;
1426
1427 if (save_orig) {
1428 err = bpf_prog_store_orig_filter(fp, fprog);
1429 if (err) {
1430 __bpf_prog_free(fp);
1431 return -ENOMEM;
1432 }
1433 }
1434
1435 /* bpf_prepare_filter() already takes care of freeing
1436 * memory in case something goes wrong.
1437 */
1438 fp = bpf_prepare_filter(fp, trans);
1439 if (IS_ERR(fp))
1440 return PTR_ERR(fp);
1441
1442 *pfp = fp;
1443 return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446
1447 void bpf_prog_destroy(struct bpf_prog *fp)
1448 {
1449 __bpf_prog_release(fp);
1450 }
1451 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452
1453 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 {
1455 struct sk_filter *fp, *old_fp;
1456
1457 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1458 if (!fp)
1459 return -ENOMEM;
1460
1461 fp->prog = prog;
1462
1463 if (!__sk_filter_charge(sk, fp)) {
1464 kfree(fp);
1465 return -ENOMEM;
1466 }
1467 refcount_set(&fp->refcnt, 1);
1468
1469 old_fp = rcu_dereference_protected(sk->sk_filter,
1470 lockdep_sock_is_held(sk));
1471 rcu_assign_pointer(sk->sk_filter, fp);
1472
1473 if (old_fp)
1474 sk_filter_uncharge(sk, old_fp);
1475
1476 return 0;
1477 }
1478
1479 static
1480 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 {
1482 unsigned int fsize = bpf_classic_proglen(fprog);
1483 struct bpf_prog *prog;
1484 int err;
1485
1486 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1487 return ERR_PTR(-EPERM);
1488
1489 /* Make sure new filter is there and in the right amounts. */
1490 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1491 return ERR_PTR(-EINVAL);
1492
1493 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494 if (!prog)
1495 return ERR_PTR(-ENOMEM);
1496
1497 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1498 __bpf_prog_free(prog);
1499 return ERR_PTR(-EFAULT);
1500 }
1501
1502 prog->len = fprog->len;
1503
1504 err = bpf_prog_store_orig_filter(prog, fprog);
1505 if (err) {
1506 __bpf_prog_free(prog);
1507 return ERR_PTR(-ENOMEM);
1508 }
1509
1510 /* bpf_prepare_filter() already takes care of freeing
1511 * memory in case something goes wrong.
1512 */
1513 return bpf_prepare_filter(prog, NULL);
1514 }
1515
1516 /**
1517 * sk_attach_filter - attach a socket filter
1518 * @fprog: the filter program
1519 * @sk: the socket to use
1520 *
1521 * Attach the user's filter code. We first run some sanity checks on
1522 * it to make sure it does not explode on us later. If an error
1523 * occurs or there is insufficient memory for the filter a negative
1524 * errno code is returned. On success the return is zero.
1525 */
1526 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 {
1528 struct bpf_prog *prog = __get_filter(fprog, sk);
1529 int err;
1530
1531 if (IS_ERR(prog))
1532 return PTR_ERR(prog);
1533
1534 err = __sk_attach_prog(prog, sk);
1535 if (err < 0) {
1536 __bpf_prog_release(prog);
1537 return err;
1538 }
1539
1540 return 0;
1541 }
1542 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543
1544 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 {
1546 struct bpf_prog *prog = __get_filter(fprog, sk);
1547 int err;
1548
1549 if (IS_ERR(prog))
1550 return PTR_ERR(prog);
1551
1552 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1553 err = -ENOMEM;
1554 else
1555 err = reuseport_attach_prog(sk, prog);
1556
1557 if (err)
1558 __bpf_prog_release(prog);
1559
1560 return err;
1561 }
1562
1563 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 {
1565 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1566 return ERR_PTR(-EPERM);
1567
1568 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 }
1570
1571 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 {
1573 struct bpf_prog *prog = __get_bpf(ufd, sk);
1574 int err;
1575
1576 if (IS_ERR(prog))
1577 return PTR_ERR(prog);
1578
1579 err = __sk_attach_prog(prog, sk);
1580 if (err < 0) {
1581 bpf_prog_put(prog);
1582 return err;
1583 }
1584
1585 return 0;
1586 }
1587
1588 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 {
1590 struct bpf_prog *prog;
1591 int err;
1592
1593 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594 return -EPERM;
1595
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1597 if (PTR_ERR(prog) == -EINVAL)
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599 if (IS_ERR(prog))
1600 return PTR_ERR(prog);
1601
1602 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1603 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1604 * bpf prog (e.g. sockmap). It depends on the
1605 * limitation imposed by bpf_prog_load().
1606 * Hence, sysctl_optmem_max is not checked.
1607 */
1608 if ((sk->sk_type != SOCK_STREAM &&
1609 sk->sk_type != SOCK_DGRAM) ||
1610 (sk->sk_protocol != IPPROTO_UDP &&
1611 sk->sk_protocol != IPPROTO_TCP) ||
1612 (sk->sk_family != AF_INET &&
1613 sk->sk_family != AF_INET6)) {
1614 err = -ENOTSUPP;
1615 goto err_prog_put;
1616 }
1617 } else {
1618 /* BPF_PROG_TYPE_SOCKET_FILTER */
1619 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1620 err = -ENOMEM;
1621 goto err_prog_put;
1622 }
1623 }
1624
1625 err = reuseport_attach_prog(sk, prog);
1626 err_prog_put:
1627 if (err)
1628 bpf_prog_put(prog);
1629
1630 return err;
1631 }
1632
1633 void sk_reuseport_prog_free(struct bpf_prog *prog)
1634 {
1635 if (!prog)
1636 return;
1637
1638 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639 bpf_prog_put(prog);
1640 else
1641 bpf_prog_destroy(prog);
1642 }
1643
1644 struct bpf_scratchpad {
1645 union {
1646 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1647 u8 buff[MAX_BPF_STACK];
1648 };
1649 };
1650
1651 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652
1653 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1654 unsigned int write_len)
1655 {
1656 return skb_ensure_writable(skb, write_len);
1657 }
1658
1659 static inline int bpf_try_make_writable(struct sk_buff *skb,
1660 unsigned int write_len)
1661 {
1662 int err = __bpf_try_make_writable(skb, write_len);
1663
1664 bpf_compute_data_pointers(skb);
1665 return err;
1666 }
1667
1668 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 {
1670 return bpf_try_make_writable(skb, skb_headlen(skb));
1671 }
1672
1673 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 {
1675 if (skb_at_tc_ingress(skb))
1676 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 }
1678
1679 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 {
1681 if (skb_at_tc_ingress(skb))
1682 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 }
1684
1685 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1686 const void *, from, u32, len, u64, flags)
1687 {
1688 void *ptr;
1689
1690 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691 return -EINVAL;
1692 if (unlikely(offset > INT_MAX))
1693 return -EFAULT;
1694 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695 return -EFAULT;
1696
1697 ptr = skb->data + offset;
1698 if (flags & BPF_F_RECOMPUTE_CSUM)
1699 __skb_postpull_rcsum(skb, ptr, len, offset);
1700
1701 memcpy(ptr, from, len);
1702
1703 if (flags & BPF_F_RECOMPUTE_CSUM)
1704 __skb_postpush_rcsum(skb, ptr, len, offset);
1705 if (flags & BPF_F_INVALIDATE_HASH)
1706 skb_clear_hash(skb);
1707
1708 return 0;
1709 }
1710
1711 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1712 .func = bpf_skb_store_bytes,
1713 .gpl_only = false,
1714 .ret_type = RET_INTEGER,
1715 .arg1_type = ARG_PTR_TO_CTX,
1716 .arg2_type = ARG_ANYTHING,
1717 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1718 .arg4_type = ARG_CONST_SIZE,
1719 .arg5_type = ARG_ANYTHING,
1720 };
1721
1722 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1723 void *, to, u32, len)
1724 {
1725 void *ptr;
1726
1727 if (unlikely(offset > INT_MAX))
1728 goto err_clear;
1729
1730 ptr = skb_header_pointer(skb, offset, len, to);
1731 if (unlikely(!ptr))
1732 goto err_clear;
1733 if (ptr != to)
1734 memcpy(to, ptr, len);
1735
1736 return 0;
1737 err_clear:
1738 memset(to, 0, len);
1739 return -EFAULT;
1740 }
1741
1742 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1743 .func = bpf_skb_load_bytes,
1744 .gpl_only = false,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747 .arg2_type = ARG_ANYTHING,
1748 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1749 .arg4_type = ARG_CONST_SIZE,
1750 };
1751
1752 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1753 const struct bpf_flow_dissector *, ctx, u32, offset,
1754 void *, to, u32, len)
1755 {
1756 void *ptr;
1757
1758 if (unlikely(offset > 0xffff))
1759 goto err_clear;
1760
1761 if (unlikely(!ctx->skb))
1762 goto err_clear;
1763
1764 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1765 if (unlikely(!ptr))
1766 goto err_clear;
1767 if (ptr != to)
1768 memcpy(to, ptr, len);
1769
1770 return 0;
1771 err_clear:
1772 memset(to, 0, len);
1773 return -EFAULT;
1774 }
1775
1776 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1777 .func = bpf_flow_dissector_load_bytes,
1778 .gpl_only = false,
1779 .ret_type = RET_INTEGER,
1780 .arg1_type = ARG_PTR_TO_CTX,
1781 .arg2_type = ARG_ANYTHING,
1782 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1783 .arg4_type = ARG_CONST_SIZE,
1784 };
1785
1786 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1787 u32, offset, void *, to, u32, len, u32, start_header)
1788 {
1789 u8 *end = skb_tail_pointer(skb);
1790 u8 *start, *ptr;
1791
1792 if (unlikely(offset > 0xffff))
1793 goto err_clear;
1794
1795 switch (start_header) {
1796 case BPF_HDR_START_MAC:
1797 if (unlikely(!skb_mac_header_was_set(skb)))
1798 goto err_clear;
1799 start = skb_mac_header(skb);
1800 break;
1801 case BPF_HDR_START_NET:
1802 start = skb_network_header(skb);
1803 break;
1804 default:
1805 goto err_clear;
1806 }
1807
1808 ptr = start + offset;
1809
1810 if (likely(ptr + len <= end)) {
1811 memcpy(to, ptr, len);
1812 return 0;
1813 }
1814
1815 err_clear:
1816 memset(to, 0, len);
1817 return -EFAULT;
1818 }
1819
1820 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1821 .func = bpf_skb_load_bytes_relative,
1822 .gpl_only = false,
1823 .ret_type = RET_INTEGER,
1824 .arg1_type = ARG_PTR_TO_CTX,
1825 .arg2_type = ARG_ANYTHING,
1826 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1827 .arg4_type = ARG_CONST_SIZE,
1828 .arg5_type = ARG_ANYTHING,
1829 };
1830
1831 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 {
1833 /* Idea is the following: should the needed direct read/write
1834 * test fail during runtime, we can pull in more data and redo
1835 * again, since implicitly, we invalidate previous checks here.
1836 *
1837 * Or, since we know how much we need to make read/writeable,
1838 * this can be done once at the program beginning for direct
1839 * access case. By this we overcome limitations of only current
1840 * headroom being accessible.
1841 */
1842 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 }
1844
1845 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1846 .func = bpf_skb_pull_data,
1847 .gpl_only = false,
1848 .ret_type = RET_INTEGER,
1849 .arg1_type = ARG_PTR_TO_CTX,
1850 .arg2_type = ARG_ANYTHING,
1851 };
1852
1853 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 {
1855 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 }
1857
1858 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1859 .func = bpf_sk_fullsock,
1860 .gpl_only = false,
1861 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1862 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1863 };
1864
1865 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1866 unsigned int write_len)
1867 {
1868 return __bpf_try_make_writable(skb, write_len);
1869 }
1870
1871 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 {
1873 /* Idea is the following: should the needed direct read/write
1874 * test fail during runtime, we can pull in more data and redo
1875 * again, since implicitly, we invalidate previous checks here.
1876 *
1877 * Or, since we know how much we need to make read/writeable,
1878 * this can be done once at the program beginning for direct
1879 * access case. By this we overcome limitations of only current
1880 * headroom being accessible.
1881 */
1882 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883 }
1884
1885 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886 .func = sk_skb_pull_data,
1887 .gpl_only = false,
1888 .ret_type = RET_INTEGER,
1889 .arg1_type = ARG_PTR_TO_CTX,
1890 .arg2_type = ARG_ANYTHING,
1891 };
1892
1893 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894 u64, from, u64, to, u64, flags)
1895 {
1896 __sum16 *ptr;
1897
1898 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899 return -EINVAL;
1900 if (unlikely(offset > 0xffff || offset & 1))
1901 return -EFAULT;
1902 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1903 return -EFAULT;
1904
1905 ptr = (__sum16 *)(skb->data + offset);
1906 switch (flags & BPF_F_HDR_FIELD_MASK) {
1907 case 0:
1908 if (unlikely(from != 0))
1909 return -EINVAL;
1910
1911 csum_replace_by_diff(ptr, to);
1912 break;
1913 case 2:
1914 csum_replace2(ptr, from, to);
1915 break;
1916 case 4:
1917 csum_replace4(ptr, from, to);
1918 break;
1919 default:
1920 return -EINVAL;
1921 }
1922
1923 return 0;
1924 }
1925
1926 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1927 .func = bpf_l3_csum_replace,
1928 .gpl_only = false,
1929 .ret_type = RET_INTEGER,
1930 .arg1_type = ARG_PTR_TO_CTX,
1931 .arg2_type = ARG_ANYTHING,
1932 .arg3_type = ARG_ANYTHING,
1933 .arg4_type = ARG_ANYTHING,
1934 .arg5_type = ARG_ANYTHING,
1935 };
1936
1937 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938 u64, from, u64, to, u64, flags)
1939 {
1940 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1941 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1942 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1943 __sum16 *ptr;
1944
1945 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947 return -EINVAL;
1948 if (unlikely(offset > 0xffff || offset & 1))
1949 return -EFAULT;
1950 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1951 return -EFAULT;
1952
1953 ptr = (__sum16 *)(skb->data + offset);
1954 if (is_mmzero && !do_mforce && !*ptr)
1955 return 0;
1956
1957 switch (flags & BPF_F_HDR_FIELD_MASK) {
1958 case 0:
1959 if (unlikely(from != 0))
1960 return -EINVAL;
1961
1962 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963 break;
1964 case 2:
1965 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966 break;
1967 case 4:
1968 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1969 break;
1970 default:
1971 return -EINVAL;
1972 }
1973
1974 if (is_mmzero && !*ptr)
1975 *ptr = CSUM_MANGLED_0;
1976 return 0;
1977 }
1978
1979 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1980 .func = bpf_l4_csum_replace,
1981 .gpl_only = false,
1982 .ret_type = RET_INTEGER,
1983 .arg1_type = ARG_PTR_TO_CTX,
1984 .arg2_type = ARG_ANYTHING,
1985 .arg3_type = ARG_ANYTHING,
1986 .arg4_type = ARG_ANYTHING,
1987 .arg5_type = ARG_ANYTHING,
1988 };
1989
1990 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991 __be32 *, to, u32, to_size, __wsum, seed)
1992 {
1993 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1994 u32 diff_size = from_size + to_size;
1995 int i, j = 0;
1996
1997 /* This is quite flexible, some examples:
1998 *
1999 * from_size == 0, to_size > 0, seed := csum --> pushing data
2000 * from_size > 0, to_size == 0, seed := csum --> pulling data
2001 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2002 *
2003 * Even for diffing, from_size and to_size don't need to be equal.
2004 */
2005 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006 diff_size > sizeof(sp->diff)))
2007 return -EINVAL;
2008
2009 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = ~from[i];
2011 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = to[i];
2013
2014 return csum_partial(sp->diff, diff_size, seed);
2015 }
2016
2017 static const struct bpf_func_proto bpf_csum_diff_proto = {
2018 .func = bpf_csum_diff,
2019 .gpl_only = false,
2020 .pkt_access = true,
2021 .ret_type = RET_INTEGER,
2022 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2023 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2024 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg5_type = ARG_ANYTHING,
2027 };
2028
2029 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 {
2031 /* The interface is to be used in combination with bpf_csum_diff()
2032 * for direct packet writes. csum rotation for alignment as well
2033 * as emulating csum_sub() can be done from the eBPF program.
2034 */
2035 if (skb->ip_summed == CHECKSUM_COMPLETE)
2036 return (skb->csum = csum_add(skb->csum, csum));
2037
2038 return -ENOTSUPP;
2039 }
2040
2041 static const struct bpf_func_proto bpf_csum_update_proto = {
2042 .func = bpf_csum_update,
2043 .gpl_only = false,
2044 .ret_type = RET_INTEGER,
2045 .arg1_type = ARG_PTR_TO_CTX,
2046 .arg2_type = ARG_ANYTHING,
2047 };
2048
2049 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 {
2051 /* The interface is to be used in combination with bpf_skb_adjust_room()
2052 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053 * is passed as flags, for example.
2054 */
2055 switch (level) {
2056 case BPF_CSUM_LEVEL_INC:
2057 __skb_incr_checksum_unnecessary(skb);
2058 break;
2059 case BPF_CSUM_LEVEL_DEC:
2060 __skb_decr_checksum_unnecessary(skb);
2061 break;
2062 case BPF_CSUM_LEVEL_RESET:
2063 __skb_reset_checksum_unnecessary(skb);
2064 break;
2065 case BPF_CSUM_LEVEL_QUERY:
2066 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067 skb->csum_level : -EACCES;
2068 default:
2069 return -EINVAL;
2070 }
2071
2072 return 0;
2073 }
2074
2075 static const struct bpf_func_proto bpf_csum_level_proto = {
2076 .func = bpf_csum_level,
2077 .gpl_only = false,
2078 .ret_type = RET_INTEGER,
2079 .arg1_type = ARG_PTR_TO_CTX,
2080 .arg2_type = ARG_ANYTHING,
2081 };
2082
2083 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 {
2085 return dev_forward_skb_nomtu(dev, skb);
2086 }
2087
2088 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089 struct sk_buff *skb)
2090 {
2091 int ret = ____dev_forward_skb(dev, skb, false);
2092
2093 if (likely(!ret)) {
2094 skb->dev = dev;
2095 ret = netif_rx(skb);
2096 }
2097
2098 return ret;
2099 }
2100
2101 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2102 {
2103 int ret;
2104
2105 if (dev_xmit_recursion()) {
2106 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2107 kfree_skb(skb);
2108 return -ENETDOWN;
2109 }
2110
2111 skb->dev = dev;
2112 skb_clear_tstamp(skb);
2113
2114 dev_xmit_recursion_inc();
2115 ret = dev_queue_xmit(skb);
2116 dev_xmit_recursion_dec();
2117
2118 return ret;
2119 }
2120
2121 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122 u32 flags)
2123 {
2124 unsigned int mlen = skb_network_offset(skb);
2125
2126 if (unlikely(skb->len <= mlen)) {
2127 kfree_skb(skb);
2128 return -ERANGE;
2129 }
2130
2131 if (mlen) {
2132 __skb_pull(skb, mlen);
2133 if (unlikely(!skb->len)) {
2134 kfree_skb(skb);
2135 return -ERANGE;
2136 }
2137
2138 /* At ingress, the mac header has already been pulled once.
2139 * At egress, skb_pospull_rcsum has to be done in case that
2140 * the skb is originated from ingress (i.e. a forwarded skb)
2141 * to ensure that rcsum starts at net header.
2142 */
2143 if (!skb_at_tc_ingress(skb))
2144 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2145 }
2146 skb_pop_mac_header(skb);
2147 skb_reset_mac_len(skb);
2148 return flags & BPF_F_INGRESS ?
2149 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2150 }
2151
2152 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2153 u32 flags)
2154 {
2155 /* Verify that a link layer header is carried */
2156 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2157 kfree_skb(skb);
2158 return -ERANGE;
2159 }
2160
2161 bpf_push_mac_rcsum(skb);
2162 return flags & BPF_F_INGRESS ?
2163 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2164 }
2165
2166 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2167 u32 flags)
2168 {
2169 if (dev_is_mac_header_xmit(dev))
2170 return __bpf_redirect_common(skb, dev, flags);
2171 else
2172 return __bpf_redirect_no_mac(skb, dev, flags);
2173 }
2174
2175 #if IS_ENABLED(CONFIG_IPV6)
2176 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2177 struct net_device *dev, struct bpf_nh_params *nh)
2178 {
2179 u32 hh_len = LL_RESERVED_SPACE(dev);
2180 const struct in6_addr *nexthop;
2181 struct dst_entry *dst = NULL;
2182 struct neighbour *neigh;
2183
2184 if (dev_xmit_recursion()) {
2185 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2186 goto out_drop;
2187 }
2188
2189 skb->dev = dev;
2190 skb_clear_tstamp(skb);
2191
2192 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2193 skb = skb_expand_head(skb, hh_len);
2194 if (!skb)
2195 return -ENOMEM;
2196 }
2197
2198 rcu_read_lock_bh();
2199 if (!nh) {
2200 dst = skb_dst(skb);
2201 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202 &ipv6_hdr(skb)->daddr);
2203 } else {
2204 nexthop = &nh->ipv6_nh;
2205 }
2206 neigh = ip_neigh_gw6(dev, nexthop);
2207 if (likely(!IS_ERR(neigh))) {
2208 int ret;
2209
2210 sock_confirm_neigh(skb, neigh);
2211 dev_xmit_recursion_inc();
2212 ret = neigh_output(neigh, skb, false);
2213 dev_xmit_recursion_dec();
2214 rcu_read_unlock_bh();
2215 return ret;
2216 }
2217 rcu_read_unlock_bh();
2218 if (dst)
2219 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2220 out_drop:
2221 kfree_skb(skb);
2222 return -ENETDOWN;
2223 }
2224
2225 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2226 struct bpf_nh_params *nh)
2227 {
2228 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2229 struct net *net = dev_net(dev);
2230 int err, ret = NET_XMIT_DROP;
2231
2232 if (!nh) {
2233 struct dst_entry *dst;
2234 struct flowi6 fl6 = {
2235 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2236 .flowi6_mark = skb->mark,
2237 .flowlabel = ip6_flowinfo(ip6h),
2238 .flowi6_oif = dev->ifindex,
2239 .flowi6_proto = ip6h->nexthdr,
2240 .daddr = ip6h->daddr,
2241 .saddr = ip6h->saddr,
2242 };
2243
2244 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2245 if (IS_ERR(dst))
2246 goto out_drop;
2247
2248 skb_dst_set(skb, dst);
2249 } else if (nh->nh_family != AF_INET6) {
2250 goto out_drop;
2251 }
2252
2253 err = bpf_out_neigh_v6(net, skb, dev, nh);
2254 if (unlikely(net_xmit_eval(err)))
2255 dev->stats.tx_errors++;
2256 else
2257 ret = NET_XMIT_SUCCESS;
2258 goto out_xmit;
2259 out_drop:
2260 dev->stats.tx_errors++;
2261 kfree_skb(skb);
2262 out_xmit:
2263 return ret;
2264 }
2265 #else
2266 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2267 struct bpf_nh_params *nh)
2268 {
2269 kfree_skb(skb);
2270 return NET_XMIT_DROP;
2271 }
2272 #endif /* CONFIG_IPV6 */
2273
2274 #if IS_ENABLED(CONFIG_INET)
2275 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2276 struct net_device *dev, struct bpf_nh_params *nh)
2277 {
2278 u32 hh_len = LL_RESERVED_SPACE(dev);
2279 struct neighbour *neigh;
2280 bool is_v6gw = false;
2281
2282 if (dev_xmit_recursion()) {
2283 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2284 goto out_drop;
2285 }
2286
2287 skb->dev = dev;
2288 skb_clear_tstamp(skb);
2289
2290 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2291 skb = skb_expand_head(skb, hh_len);
2292 if (!skb)
2293 return -ENOMEM;
2294 }
2295
2296 rcu_read_lock_bh();
2297 if (!nh) {
2298 struct dst_entry *dst = skb_dst(skb);
2299 struct rtable *rt = container_of(dst, struct rtable, dst);
2300
2301 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2302 } else if (nh->nh_family == AF_INET6) {
2303 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2304 is_v6gw = true;
2305 } else if (nh->nh_family == AF_INET) {
2306 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2307 } else {
2308 rcu_read_unlock_bh();
2309 goto out_drop;
2310 }
2311
2312 if (likely(!IS_ERR(neigh))) {
2313 int ret;
2314
2315 sock_confirm_neigh(skb, neigh);
2316 dev_xmit_recursion_inc();
2317 ret = neigh_output(neigh, skb, is_v6gw);
2318 dev_xmit_recursion_dec();
2319 rcu_read_unlock_bh();
2320 return ret;
2321 }
2322 rcu_read_unlock_bh();
2323 out_drop:
2324 kfree_skb(skb);
2325 return -ENETDOWN;
2326 }
2327
2328 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2329 struct bpf_nh_params *nh)
2330 {
2331 const struct iphdr *ip4h = ip_hdr(skb);
2332 struct net *net = dev_net(dev);
2333 int err, ret = NET_XMIT_DROP;
2334
2335 if (!nh) {
2336 struct flowi4 fl4 = {
2337 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2338 .flowi4_mark = skb->mark,
2339 .flowi4_tos = RT_TOS(ip4h->tos),
2340 .flowi4_oif = dev->ifindex,
2341 .flowi4_proto = ip4h->protocol,
2342 .daddr = ip4h->daddr,
2343 .saddr = ip4h->saddr,
2344 };
2345 struct rtable *rt;
2346
2347 rt = ip_route_output_flow(net, &fl4, NULL);
2348 if (IS_ERR(rt))
2349 goto out_drop;
2350 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2351 ip_rt_put(rt);
2352 goto out_drop;
2353 }
2354
2355 skb_dst_set(skb, &rt->dst);
2356 }
2357
2358 err = bpf_out_neigh_v4(net, skb, dev, nh);
2359 if (unlikely(net_xmit_eval(err)))
2360 dev->stats.tx_errors++;
2361 else
2362 ret = NET_XMIT_SUCCESS;
2363 goto out_xmit;
2364 out_drop:
2365 dev->stats.tx_errors++;
2366 kfree_skb(skb);
2367 out_xmit:
2368 return ret;
2369 }
2370 #else
2371 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2372 struct bpf_nh_params *nh)
2373 {
2374 kfree_skb(skb);
2375 return NET_XMIT_DROP;
2376 }
2377 #endif /* CONFIG_INET */
2378
2379 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2380 struct bpf_nh_params *nh)
2381 {
2382 struct ethhdr *ethh = eth_hdr(skb);
2383
2384 if (unlikely(skb->mac_header >= skb->network_header))
2385 goto out;
2386 bpf_push_mac_rcsum(skb);
2387 if (is_multicast_ether_addr(ethh->h_dest))
2388 goto out;
2389
2390 skb_pull(skb, sizeof(*ethh));
2391 skb_unset_mac_header(skb);
2392 skb_reset_network_header(skb);
2393
2394 if (skb->protocol == htons(ETH_P_IP))
2395 return __bpf_redirect_neigh_v4(skb, dev, nh);
2396 else if (skb->protocol == htons(ETH_P_IPV6))
2397 return __bpf_redirect_neigh_v6(skb, dev, nh);
2398 out:
2399 kfree_skb(skb);
2400 return -ENOTSUPP;
2401 }
2402
2403 /* Internal, non-exposed redirect flags. */
2404 enum {
2405 BPF_F_NEIGH = (1ULL << 1),
2406 BPF_F_PEER = (1ULL << 2),
2407 BPF_F_NEXTHOP = (1ULL << 3),
2408 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2409 };
2410
2411 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2412 {
2413 struct net_device *dev;
2414 struct sk_buff *clone;
2415 int ret;
2416
2417 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2418 return -EINVAL;
2419
2420 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2421 if (unlikely(!dev))
2422 return -EINVAL;
2423
2424 clone = skb_clone(skb, GFP_ATOMIC);
2425 if (unlikely(!clone))
2426 return -ENOMEM;
2427
2428 /* For direct write, we need to keep the invariant that the skbs
2429 * we're dealing with need to be uncloned. Should uncloning fail
2430 * here, we need to free the just generated clone to unclone once
2431 * again.
2432 */
2433 ret = bpf_try_make_head_writable(skb);
2434 if (unlikely(ret)) {
2435 kfree_skb(clone);
2436 return -ENOMEM;
2437 }
2438
2439 return __bpf_redirect(clone, dev, flags);
2440 }
2441
2442 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2443 .func = bpf_clone_redirect,
2444 .gpl_only = false,
2445 .ret_type = RET_INTEGER,
2446 .arg1_type = ARG_PTR_TO_CTX,
2447 .arg2_type = ARG_ANYTHING,
2448 .arg3_type = ARG_ANYTHING,
2449 };
2450
2451 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2452 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2453
2454 int skb_do_redirect(struct sk_buff *skb)
2455 {
2456 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2457 struct net *net = dev_net(skb->dev);
2458 struct net_device *dev;
2459 u32 flags = ri->flags;
2460
2461 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2462 ri->tgt_index = 0;
2463 ri->flags = 0;
2464 if (unlikely(!dev))
2465 goto out_drop;
2466 if (flags & BPF_F_PEER) {
2467 const struct net_device_ops *ops = dev->netdev_ops;
2468
2469 if (unlikely(!ops->ndo_get_peer_dev ||
2470 !skb_at_tc_ingress(skb)))
2471 goto out_drop;
2472 dev = ops->ndo_get_peer_dev(dev);
2473 if (unlikely(!dev ||
2474 !(dev->flags & IFF_UP) ||
2475 net_eq(net, dev_net(dev))))
2476 goto out_drop;
2477 skb->dev = dev;
2478 return -EAGAIN;
2479 }
2480 return flags & BPF_F_NEIGH ?
2481 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2482 &ri->nh : NULL) :
2483 __bpf_redirect(skb, dev, flags);
2484 out_drop:
2485 kfree_skb(skb);
2486 return -EINVAL;
2487 }
2488
2489 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2490 {
2491 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2492
2493 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2494 return TC_ACT_SHOT;
2495
2496 ri->flags = flags;
2497 ri->tgt_index = ifindex;
2498
2499 return TC_ACT_REDIRECT;
2500 }
2501
2502 static const struct bpf_func_proto bpf_redirect_proto = {
2503 .func = bpf_redirect,
2504 .gpl_only = false,
2505 .ret_type = RET_INTEGER,
2506 .arg1_type = ARG_ANYTHING,
2507 .arg2_type = ARG_ANYTHING,
2508 };
2509
2510 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2511 {
2512 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2513
2514 if (unlikely(flags))
2515 return TC_ACT_SHOT;
2516
2517 ri->flags = BPF_F_PEER;
2518 ri->tgt_index = ifindex;
2519
2520 return TC_ACT_REDIRECT;
2521 }
2522
2523 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2524 .func = bpf_redirect_peer,
2525 .gpl_only = false,
2526 .ret_type = RET_INTEGER,
2527 .arg1_type = ARG_ANYTHING,
2528 .arg2_type = ARG_ANYTHING,
2529 };
2530
2531 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2532 int, plen, u64, flags)
2533 {
2534 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2535
2536 if (unlikely((plen && plen < sizeof(*params)) || flags))
2537 return TC_ACT_SHOT;
2538
2539 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2540 ri->tgt_index = ifindex;
2541
2542 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2543 if (plen)
2544 memcpy(&ri->nh, params, sizeof(ri->nh));
2545
2546 return TC_ACT_REDIRECT;
2547 }
2548
2549 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2550 .func = bpf_redirect_neigh,
2551 .gpl_only = false,
2552 .ret_type = RET_INTEGER,
2553 .arg1_type = ARG_ANYTHING,
2554 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2555 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2556 .arg4_type = ARG_ANYTHING,
2557 };
2558
2559 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2560 {
2561 msg->apply_bytes = bytes;
2562 return 0;
2563 }
2564
2565 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2566 .func = bpf_msg_apply_bytes,
2567 .gpl_only = false,
2568 .ret_type = RET_INTEGER,
2569 .arg1_type = ARG_PTR_TO_CTX,
2570 .arg2_type = ARG_ANYTHING,
2571 };
2572
2573 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2574 {
2575 msg->cork_bytes = bytes;
2576 return 0;
2577 }
2578
2579 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2580 .func = bpf_msg_cork_bytes,
2581 .gpl_only = false,
2582 .ret_type = RET_INTEGER,
2583 .arg1_type = ARG_PTR_TO_CTX,
2584 .arg2_type = ARG_ANYTHING,
2585 };
2586
2587 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2588 u32, end, u64, flags)
2589 {
2590 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2591 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2592 struct scatterlist *sge;
2593 u8 *raw, *to, *from;
2594 struct page *page;
2595
2596 if (unlikely(flags || end <= start))
2597 return -EINVAL;
2598
2599 /* First find the starting scatterlist element */
2600 i = msg->sg.start;
2601 do {
2602 offset += len;
2603 len = sk_msg_elem(msg, i)->length;
2604 if (start < offset + len)
2605 break;
2606 sk_msg_iter_var_next(i);
2607 } while (i != msg->sg.end);
2608
2609 if (unlikely(start >= offset + len))
2610 return -EINVAL;
2611
2612 first_sge = i;
2613 /* The start may point into the sg element so we need to also
2614 * account for the headroom.
2615 */
2616 bytes_sg_total = start - offset + bytes;
2617 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2618 goto out;
2619
2620 /* At this point we need to linearize multiple scatterlist
2621 * elements or a single shared page. Either way we need to
2622 * copy into a linear buffer exclusively owned by BPF. Then
2623 * place the buffer in the scatterlist and fixup the original
2624 * entries by removing the entries now in the linear buffer
2625 * and shifting the remaining entries. For now we do not try
2626 * to copy partial entries to avoid complexity of running out
2627 * of sg_entry slots. The downside is reading a single byte
2628 * will copy the entire sg entry.
2629 */
2630 do {
2631 copy += sk_msg_elem(msg, i)->length;
2632 sk_msg_iter_var_next(i);
2633 if (bytes_sg_total <= copy)
2634 break;
2635 } while (i != msg->sg.end);
2636 last_sge = i;
2637
2638 if (unlikely(bytes_sg_total > copy))
2639 return -EINVAL;
2640
2641 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2642 get_order(copy));
2643 if (unlikely(!page))
2644 return -ENOMEM;
2645
2646 raw = page_address(page);
2647 i = first_sge;
2648 do {
2649 sge = sk_msg_elem(msg, i);
2650 from = sg_virt(sge);
2651 len = sge->length;
2652 to = raw + poffset;
2653
2654 memcpy(to, from, len);
2655 poffset += len;
2656 sge->length = 0;
2657 put_page(sg_page(sge));
2658
2659 sk_msg_iter_var_next(i);
2660 } while (i != last_sge);
2661
2662 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2663
2664 /* To repair sg ring we need to shift entries. If we only
2665 * had a single entry though we can just replace it and
2666 * be done. Otherwise walk the ring and shift the entries.
2667 */
2668 WARN_ON_ONCE(last_sge == first_sge);
2669 shift = last_sge > first_sge ?
2670 last_sge - first_sge - 1 :
2671 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2672 if (!shift)
2673 goto out;
2674
2675 i = first_sge;
2676 sk_msg_iter_var_next(i);
2677 do {
2678 u32 move_from;
2679
2680 if (i + shift >= NR_MSG_FRAG_IDS)
2681 move_from = i + shift - NR_MSG_FRAG_IDS;
2682 else
2683 move_from = i + shift;
2684 if (move_from == msg->sg.end)
2685 break;
2686
2687 msg->sg.data[i] = msg->sg.data[move_from];
2688 msg->sg.data[move_from].length = 0;
2689 msg->sg.data[move_from].page_link = 0;
2690 msg->sg.data[move_from].offset = 0;
2691 sk_msg_iter_var_next(i);
2692 } while (1);
2693
2694 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2695 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2696 msg->sg.end - shift;
2697 out:
2698 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2699 msg->data_end = msg->data + bytes;
2700 return 0;
2701 }
2702
2703 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2704 .func = bpf_msg_pull_data,
2705 .gpl_only = false,
2706 .ret_type = RET_INTEGER,
2707 .arg1_type = ARG_PTR_TO_CTX,
2708 .arg2_type = ARG_ANYTHING,
2709 .arg3_type = ARG_ANYTHING,
2710 .arg4_type = ARG_ANYTHING,
2711 };
2712
2713 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2714 u32, len, u64, flags)
2715 {
2716 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2717 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2718 u8 *raw, *to, *from;
2719 struct page *page;
2720
2721 if (unlikely(flags))
2722 return -EINVAL;
2723
2724 if (unlikely(len == 0))
2725 return 0;
2726
2727 /* First find the starting scatterlist element */
2728 i = msg->sg.start;
2729 do {
2730 offset += l;
2731 l = sk_msg_elem(msg, i)->length;
2732
2733 if (start < offset + l)
2734 break;
2735 sk_msg_iter_var_next(i);
2736 } while (i != msg->sg.end);
2737
2738 if (start >= offset + l)
2739 return -EINVAL;
2740
2741 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2742
2743 /* If no space available will fallback to copy, we need at
2744 * least one scatterlist elem available to push data into
2745 * when start aligns to the beginning of an element or two
2746 * when it falls inside an element. We handle the start equals
2747 * offset case because its the common case for inserting a
2748 * header.
2749 */
2750 if (!space || (space == 1 && start != offset))
2751 copy = msg->sg.data[i].length;
2752
2753 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2754 get_order(copy + len));
2755 if (unlikely(!page))
2756 return -ENOMEM;
2757
2758 if (copy) {
2759 int front, back;
2760
2761 raw = page_address(page);
2762
2763 psge = sk_msg_elem(msg, i);
2764 front = start - offset;
2765 back = psge->length - front;
2766 from = sg_virt(psge);
2767
2768 if (front)
2769 memcpy(raw, from, front);
2770
2771 if (back) {
2772 from += front;
2773 to = raw + front + len;
2774
2775 memcpy(to, from, back);
2776 }
2777
2778 put_page(sg_page(psge));
2779 } else if (start - offset) {
2780 psge = sk_msg_elem(msg, i);
2781 rsge = sk_msg_elem_cpy(msg, i);
2782
2783 psge->length = start - offset;
2784 rsge.length -= psge->length;
2785 rsge.offset += start;
2786
2787 sk_msg_iter_var_next(i);
2788 sg_unmark_end(psge);
2789 sg_unmark_end(&rsge);
2790 sk_msg_iter_next(msg, end);
2791 }
2792
2793 /* Slot(s) to place newly allocated data */
2794 new = i;
2795
2796 /* Shift one or two slots as needed */
2797 if (!copy) {
2798 sge = sk_msg_elem_cpy(msg, i);
2799
2800 sk_msg_iter_var_next(i);
2801 sg_unmark_end(&sge);
2802 sk_msg_iter_next(msg, end);
2803
2804 nsge = sk_msg_elem_cpy(msg, i);
2805 if (rsge.length) {
2806 sk_msg_iter_var_next(i);
2807 nnsge = sk_msg_elem_cpy(msg, i);
2808 }
2809
2810 while (i != msg->sg.end) {
2811 msg->sg.data[i] = sge;
2812 sge = nsge;
2813 sk_msg_iter_var_next(i);
2814 if (rsge.length) {
2815 nsge = nnsge;
2816 nnsge = sk_msg_elem_cpy(msg, i);
2817 } else {
2818 nsge = sk_msg_elem_cpy(msg, i);
2819 }
2820 }
2821 }
2822
2823 /* Place newly allocated data buffer */
2824 sk_mem_charge(msg->sk, len);
2825 msg->sg.size += len;
2826 __clear_bit(new, msg->sg.copy);
2827 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2828 if (rsge.length) {
2829 get_page(sg_page(&rsge));
2830 sk_msg_iter_var_next(new);
2831 msg->sg.data[new] = rsge;
2832 }
2833
2834 sk_msg_compute_data_pointers(msg);
2835 return 0;
2836 }
2837
2838 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2839 .func = bpf_msg_push_data,
2840 .gpl_only = false,
2841 .ret_type = RET_INTEGER,
2842 .arg1_type = ARG_PTR_TO_CTX,
2843 .arg2_type = ARG_ANYTHING,
2844 .arg3_type = ARG_ANYTHING,
2845 .arg4_type = ARG_ANYTHING,
2846 };
2847
2848 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2849 {
2850 int prev;
2851
2852 do {
2853 prev = i;
2854 sk_msg_iter_var_next(i);
2855 msg->sg.data[prev] = msg->sg.data[i];
2856 } while (i != msg->sg.end);
2857
2858 sk_msg_iter_prev(msg, end);
2859 }
2860
2861 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2862 {
2863 struct scatterlist tmp, sge;
2864
2865 sk_msg_iter_next(msg, end);
2866 sge = sk_msg_elem_cpy(msg, i);
2867 sk_msg_iter_var_next(i);
2868 tmp = sk_msg_elem_cpy(msg, i);
2869
2870 while (i != msg->sg.end) {
2871 msg->sg.data[i] = sge;
2872 sk_msg_iter_var_next(i);
2873 sge = tmp;
2874 tmp = sk_msg_elem_cpy(msg, i);
2875 }
2876 }
2877
2878 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2879 u32, len, u64, flags)
2880 {
2881 u32 i = 0, l = 0, space, offset = 0;
2882 u64 last = start + len;
2883 int pop;
2884
2885 if (unlikely(flags))
2886 return -EINVAL;
2887
2888 /* First find the starting scatterlist element */
2889 i = msg->sg.start;
2890 do {
2891 offset += l;
2892 l = sk_msg_elem(msg, i)->length;
2893
2894 if (start < offset + l)
2895 break;
2896 sk_msg_iter_var_next(i);
2897 } while (i != msg->sg.end);
2898
2899 /* Bounds checks: start and pop must be inside message */
2900 if (start >= offset + l || last >= msg->sg.size)
2901 return -EINVAL;
2902
2903 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2904
2905 pop = len;
2906 /* --------------| offset
2907 * -| start |-------- len -------|
2908 *
2909 * |----- a ----|-------- pop -------|----- b ----|
2910 * |______________________________________________| length
2911 *
2912 *
2913 * a: region at front of scatter element to save
2914 * b: region at back of scatter element to save when length > A + pop
2915 * pop: region to pop from element, same as input 'pop' here will be
2916 * decremented below per iteration.
2917 *
2918 * Two top-level cases to handle when start != offset, first B is non
2919 * zero and second B is zero corresponding to when a pop includes more
2920 * than one element.
2921 *
2922 * Then if B is non-zero AND there is no space allocate space and
2923 * compact A, B regions into page. If there is space shift ring to
2924 * the rigth free'ing the next element in ring to place B, leaving
2925 * A untouched except to reduce length.
2926 */
2927 if (start != offset) {
2928 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2929 int a = start;
2930 int b = sge->length - pop - a;
2931
2932 sk_msg_iter_var_next(i);
2933
2934 if (pop < sge->length - a) {
2935 if (space) {
2936 sge->length = a;
2937 sk_msg_shift_right(msg, i);
2938 nsge = sk_msg_elem(msg, i);
2939 get_page(sg_page(sge));
2940 sg_set_page(nsge,
2941 sg_page(sge),
2942 b, sge->offset + pop + a);
2943 } else {
2944 struct page *page, *orig;
2945 u8 *to, *from;
2946
2947 page = alloc_pages(__GFP_NOWARN |
2948 __GFP_COMP | GFP_ATOMIC,
2949 get_order(a + b));
2950 if (unlikely(!page))
2951 return -ENOMEM;
2952
2953 sge->length = a;
2954 orig = sg_page(sge);
2955 from = sg_virt(sge);
2956 to = page_address(page);
2957 memcpy(to, from, a);
2958 memcpy(to + a, from + a + pop, b);
2959 sg_set_page(sge, page, a + b, 0);
2960 put_page(orig);
2961 }
2962 pop = 0;
2963 } else if (pop >= sge->length - a) {
2964 pop -= (sge->length - a);
2965 sge->length = a;
2966 }
2967 }
2968
2969 /* From above the current layout _must_ be as follows,
2970 *
2971 * -| offset
2972 * -| start
2973 *
2974 * |---- pop ---|---------------- b ------------|
2975 * |____________________________________________| length
2976 *
2977 * Offset and start of the current msg elem are equal because in the
2978 * previous case we handled offset != start and either consumed the
2979 * entire element and advanced to the next element OR pop == 0.
2980 *
2981 * Two cases to handle here are first pop is less than the length
2982 * leaving some remainder b above. Simply adjust the element's layout
2983 * in this case. Or pop >= length of the element so that b = 0. In this
2984 * case advance to next element decrementing pop.
2985 */
2986 while (pop) {
2987 struct scatterlist *sge = sk_msg_elem(msg, i);
2988
2989 if (pop < sge->length) {
2990 sge->length -= pop;
2991 sge->offset += pop;
2992 pop = 0;
2993 } else {
2994 pop -= sge->length;
2995 sk_msg_shift_left(msg, i);
2996 }
2997 sk_msg_iter_var_next(i);
2998 }
2999
3000 sk_mem_uncharge(msg->sk, len - pop);
3001 msg->sg.size -= (len - pop);
3002 sk_msg_compute_data_pointers(msg);
3003 return 0;
3004 }
3005
3006 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3007 .func = bpf_msg_pop_data,
3008 .gpl_only = false,
3009 .ret_type = RET_INTEGER,
3010 .arg1_type = ARG_PTR_TO_CTX,
3011 .arg2_type = ARG_ANYTHING,
3012 .arg3_type = ARG_ANYTHING,
3013 .arg4_type = ARG_ANYTHING,
3014 };
3015
3016 #ifdef CONFIG_CGROUP_NET_CLASSID
3017 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3018 {
3019 return __task_get_classid(current);
3020 }
3021
3022 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3023 .func = bpf_get_cgroup_classid_curr,
3024 .gpl_only = false,
3025 .ret_type = RET_INTEGER,
3026 };
3027
3028 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3029 {
3030 struct sock *sk = skb_to_full_sk(skb);
3031
3032 if (!sk || !sk_fullsock(sk))
3033 return 0;
3034
3035 return sock_cgroup_classid(&sk->sk_cgrp_data);
3036 }
3037
3038 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3039 .func = bpf_skb_cgroup_classid,
3040 .gpl_only = false,
3041 .ret_type = RET_INTEGER,
3042 .arg1_type = ARG_PTR_TO_CTX,
3043 };
3044 #endif
3045
3046 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3047 {
3048 return task_get_classid(skb);
3049 }
3050
3051 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3052 .func = bpf_get_cgroup_classid,
3053 .gpl_only = false,
3054 .ret_type = RET_INTEGER,
3055 .arg1_type = ARG_PTR_TO_CTX,
3056 };
3057
3058 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3059 {
3060 return dst_tclassid(skb);
3061 }
3062
3063 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3064 .func = bpf_get_route_realm,
3065 .gpl_only = false,
3066 .ret_type = RET_INTEGER,
3067 .arg1_type = ARG_PTR_TO_CTX,
3068 };
3069
3070 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3071 {
3072 /* If skb_clear_hash() was called due to mangling, we can
3073 * trigger SW recalculation here. Later access to hash
3074 * can then use the inline skb->hash via context directly
3075 * instead of calling this helper again.
3076 */
3077 return skb_get_hash(skb);
3078 }
3079
3080 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3081 .func = bpf_get_hash_recalc,
3082 .gpl_only = false,
3083 .ret_type = RET_INTEGER,
3084 .arg1_type = ARG_PTR_TO_CTX,
3085 };
3086
3087 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3088 {
3089 /* After all direct packet write, this can be used once for
3090 * triggering a lazy recalc on next skb_get_hash() invocation.
3091 */
3092 skb_clear_hash(skb);
3093 return 0;
3094 }
3095
3096 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3097 .func = bpf_set_hash_invalid,
3098 .gpl_only = false,
3099 .ret_type = RET_INTEGER,
3100 .arg1_type = ARG_PTR_TO_CTX,
3101 };
3102
3103 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3104 {
3105 /* Set user specified hash as L4(+), so that it gets returned
3106 * on skb_get_hash() call unless BPF prog later on triggers a
3107 * skb_clear_hash().
3108 */
3109 __skb_set_sw_hash(skb, hash, true);
3110 return 0;
3111 }
3112
3113 static const struct bpf_func_proto bpf_set_hash_proto = {
3114 .func = bpf_set_hash,
3115 .gpl_only = false,
3116 .ret_type = RET_INTEGER,
3117 .arg1_type = ARG_PTR_TO_CTX,
3118 .arg2_type = ARG_ANYTHING,
3119 };
3120
3121 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3122 u16, vlan_tci)
3123 {
3124 int ret;
3125
3126 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3127 vlan_proto != htons(ETH_P_8021AD)))
3128 vlan_proto = htons(ETH_P_8021Q);
3129
3130 bpf_push_mac_rcsum(skb);
3131 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3132 bpf_pull_mac_rcsum(skb);
3133
3134 bpf_compute_data_pointers(skb);
3135 return ret;
3136 }
3137
3138 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3139 .func = bpf_skb_vlan_push,
3140 .gpl_only = false,
3141 .ret_type = RET_INTEGER,
3142 .arg1_type = ARG_PTR_TO_CTX,
3143 .arg2_type = ARG_ANYTHING,
3144 .arg3_type = ARG_ANYTHING,
3145 };
3146
3147 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3148 {
3149 int ret;
3150
3151 bpf_push_mac_rcsum(skb);
3152 ret = skb_vlan_pop(skb);
3153 bpf_pull_mac_rcsum(skb);
3154
3155 bpf_compute_data_pointers(skb);
3156 return ret;
3157 }
3158
3159 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3160 .func = bpf_skb_vlan_pop,
3161 .gpl_only = false,
3162 .ret_type = RET_INTEGER,
3163 .arg1_type = ARG_PTR_TO_CTX,
3164 };
3165
3166 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3167 {
3168 /* Caller already did skb_cow() with len as headroom,
3169 * so no need to do it here.
3170 */
3171 skb_push(skb, len);
3172 memmove(skb->data, skb->data + len, off);
3173 memset(skb->data + off, 0, len);
3174
3175 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3176 * needed here as it does not change the skb->csum
3177 * result for checksum complete when summing over
3178 * zeroed blocks.
3179 */
3180 return 0;
3181 }
3182
3183 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3184 {
3185 void *old_data;
3186
3187 /* skb_ensure_writable() is not needed here, as we're
3188 * already working on an uncloned skb.
3189 */
3190 if (unlikely(!pskb_may_pull(skb, off + len)))
3191 return -ENOMEM;
3192
3193 old_data = skb->data;
3194 __skb_pull(skb, len);
3195 skb_postpull_rcsum(skb, old_data + off, len);
3196 memmove(skb->data, old_data, off);
3197
3198 return 0;
3199 }
3200
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3202 {
3203 bool trans_same = skb->transport_header == skb->network_header;
3204 int ret;
3205
3206 /* There's no need for __skb_push()/__skb_pull() pair to
3207 * get to the start of the mac header as we're guaranteed
3208 * to always start from here under eBPF.
3209 */
3210 ret = bpf_skb_generic_push(skb, off, len);
3211 if (likely(!ret)) {
3212 skb->mac_header -= len;
3213 skb->network_header -= len;
3214 if (trans_same)
3215 skb->transport_header = skb->network_header;
3216 }
3217
3218 return ret;
3219 }
3220
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3222 {
3223 bool trans_same = skb->transport_header == skb->network_header;
3224 int ret;
3225
3226 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227 ret = bpf_skb_generic_pop(skb, off, len);
3228 if (likely(!ret)) {
3229 skb->mac_header += len;
3230 skb->network_header += len;
3231 if (trans_same)
3232 skb->transport_header = skb->network_header;
3233 }
3234
3235 return ret;
3236 }
3237
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3239 {
3240 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241 u32 off = skb_mac_header_len(skb);
3242 int ret;
3243
3244 ret = skb_cow(skb, len_diff);
3245 if (unlikely(ret < 0))
3246 return ret;
3247
3248 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3249 if (unlikely(ret < 0))
3250 return ret;
3251
3252 if (skb_is_gso(skb)) {
3253 struct skb_shared_info *shinfo = skb_shinfo(skb);
3254
3255 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3256 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3257 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3258 shinfo->gso_type |= SKB_GSO_TCPV6;
3259 }
3260 }
3261
3262 skb->protocol = htons(ETH_P_IPV6);
3263 skb_clear_hash(skb);
3264
3265 return 0;
3266 }
3267
3268 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3269 {
3270 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3271 u32 off = skb_mac_header_len(skb);
3272 int ret;
3273
3274 ret = skb_unclone(skb, GFP_ATOMIC);
3275 if (unlikely(ret < 0))
3276 return ret;
3277
3278 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3279 if (unlikely(ret < 0))
3280 return ret;
3281
3282 if (skb_is_gso(skb)) {
3283 struct skb_shared_info *shinfo = skb_shinfo(skb);
3284
3285 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3286 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3287 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3288 shinfo->gso_type |= SKB_GSO_TCPV4;
3289 }
3290 }
3291
3292 skb->protocol = htons(ETH_P_IP);
3293 skb_clear_hash(skb);
3294
3295 return 0;
3296 }
3297
3298 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3299 {
3300 __be16 from_proto = skb->protocol;
3301
3302 if (from_proto == htons(ETH_P_IP) &&
3303 to_proto == htons(ETH_P_IPV6))
3304 return bpf_skb_proto_4_to_6(skb);
3305
3306 if (from_proto == htons(ETH_P_IPV6) &&
3307 to_proto == htons(ETH_P_IP))
3308 return bpf_skb_proto_6_to_4(skb);
3309
3310 return -ENOTSUPP;
3311 }
3312
3313 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3314 u64, flags)
3315 {
3316 int ret;
3317
3318 if (unlikely(flags))
3319 return -EINVAL;
3320
3321 /* General idea is that this helper does the basic groundwork
3322 * needed for changing the protocol, and eBPF program fills the
3323 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3324 * and other helpers, rather than passing a raw buffer here.
3325 *
3326 * The rationale is to keep this minimal and without a need to
3327 * deal with raw packet data. F.e. even if we would pass buffers
3328 * here, the program still needs to call the bpf_lX_csum_replace()
3329 * helpers anyway. Plus, this way we keep also separation of
3330 * concerns, since f.e. bpf_skb_store_bytes() should only take
3331 * care of stores.
3332 *
3333 * Currently, additional options and extension header space are
3334 * not supported, but flags register is reserved so we can adapt
3335 * that. For offloads, we mark packet as dodgy, so that headers
3336 * need to be verified first.
3337 */
3338 ret = bpf_skb_proto_xlat(skb, proto);
3339 bpf_compute_data_pointers(skb);
3340 return ret;
3341 }
3342
3343 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3344 .func = bpf_skb_change_proto,
3345 .gpl_only = false,
3346 .ret_type = RET_INTEGER,
3347 .arg1_type = ARG_PTR_TO_CTX,
3348 .arg2_type = ARG_ANYTHING,
3349 .arg3_type = ARG_ANYTHING,
3350 };
3351
3352 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3353 {
3354 /* We only allow a restricted subset to be changed for now. */
3355 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3356 !skb_pkt_type_ok(pkt_type)))
3357 return -EINVAL;
3358
3359 skb->pkt_type = pkt_type;
3360 return 0;
3361 }
3362
3363 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3364 .func = bpf_skb_change_type,
3365 .gpl_only = false,
3366 .ret_type = RET_INTEGER,
3367 .arg1_type = ARG_PTR_TO_CTX,
3368 .arg2_type = ARG_ANYTHING,
3369 };
3370
3371 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3372 {
3373 switch (skb->protocol) {
3374 case htons(ETH_P_IP):
3375 return sizeof(struct iphdr);
3376 case htons(ETH_P_IPV6):
3377 return sizeof(struct ipv6hdr);
3378 default:
3379 return ~0U;
3380 }
3381 }
3382
3383 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3384 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3385
3386 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3387 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3388 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3389 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3390 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3391 BPF_F_ADJ_ROOM_ENCAP_L2( \
3392 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3393
3394 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3395 u64 flags)
3396 {
3397 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3398 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3399 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3400 unsigned int gso_type = SKB_GSO_DODGY;
3401 int ret;
3402
3403 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3404 /* udp gso_size delineates datagrams, only allow if fixed */
3405 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3406 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3407 return -ENOTSUPP;
3408 }
3409
3410 ret = skb_cow_head(skb, len_diff);
3411 if (unlikely(ret < 0))
3412 return ret;
3413
3414 if (encap) {
3415 if (skb->protocol != htons(ETH_P_IP) &&
3416 skb->protocol != htons(ETH_P_IPV6))
3417 return -ENOTSUPP;
3418
3419 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3420 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3421 return -EINVAL;
3422
3423 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3424 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3425 return -EINVAL;
3426
3427 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3428 inner_mac_len < ETH_HLEN)
3429 return -EINVAL;
3430
3431 if (skb->encapsulation)
3432 return -EALREADY;
3433
3434 mac_len = skb->network_header - skb->mac_header;
3435 inner_net = skb->network_header;
3436 if (inner_mac_len > len_diff)
3437 return -EINVAL;
3438 inner_trans = skb->transport_header;
3439 }
3440
3441 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3442 if (unlikely(ret < 0))
3443 return ret;
3444
3445 if (encap) {
3446 skb->inner_mac_header = inner_net - inner_mac_len;
3447 skb->inner_network_header = inner_net;
3448 skb->inner_transport_header = inner_trans;
3449
3450 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3451 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3452 else
3453 skb_set_inner_protocol(skb, skb->protocol);
3454
3455 skb->encapsulation = 1;
3456 skb_set_network_header(skb, mac_len);
3457
3458 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3459 gso_type |= SKB_GSO_UDP_TUNNEL;
3460 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3461 gso_type |= SKB_GSO_GRE;
3462 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463 gso_type |= SKB_GSO_IPXIP6;
3464 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465 gso_type |= SKB_GSO_IPXIP4;
3466
3467 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3468 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3469 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3470 sizeof(struct ipv6hdr) :
3471 sizeof(struct iphdr);
3472
3473 skb_set_transport_header(skb, mac_len + nh_len);
3474 }
3475
3476 /* Match skb->protocol to new outer l3 protocol */
3477 if (skb->protocol == htons(ETH_P_IP) &&
3478 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 skb->protocol = htons(ETH_P_IPV6);
3480 else if (skb->protocol == htons(ETH_P_IPV6) &&
3481 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3482 skb->protocol = htons(ETH_P_IP);
3483 }
3484
3485 if (skb_is_gso(skb)) {
3486 struct skb_shared_info *shinfo = skb_shinfo(skb);
3487
3488 /* Due to header grow, MSS needs to be downgraded. */
3489 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3490 skb_decrease_gso_size(shinfo, len_diff);
3491
3492 /* Header must be checked, and gso_segs recomputed. */
3493 shinfo->gso_type |= gso_type;
3494 shinfo->gso_segs = 0;
3495 }
3496
3497 return 0;
3498 }
3499
3500 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3501 u64 flags)
3502 {
3503 int ret;
3504
3505 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3506 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3507 return -EINVAL;
3508
3509 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3510 /* udp gso_size delineates datagrams, only allow if fixed */
3511 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3512 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3513 return -ENOTSUPP;
3514 }
3515
3516 ret = skb_unclone(skb, GFP_ATOMIC);
3517 if (unlikely(ret < 0))
3518 return ret;
3519
3520 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3521 if (unlikely(ret < 0))
3522 return ret;
3523
3524 if (skb_is_gso(skb)) {
3525 struct skb_shared_info *shinfo = skb_shinfo(skb);
3526
3527 /* Due to header shrink, MSS can be upgraded. */
3528 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529 skb_increase_gso_size(shinfo, len_diff);
3530
3531 /* Header must be checked, and gso_segs recomputed. */
3532 shinfo->gso_type |= SKB_GSO_DODGY;
3533 shinfo->gso_segs = 0;
3534 }
3535
3536 return 0;
3537 }
3538
3539 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3540
3541 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3542 u32, mode, u64, flags)
3543 {
3544 u32 len_diff_abs = abs(len_diff);
3545 bool shrink = len_diff < 0;
3546 int ret = 0;
3547
3548 if (unlikely(flags || mode))
3549 return -EINVAL;
3550 if (unlikely(len_diff_abs > 0xfffU))
3551 return -EFAULT;
3552
3553 if (!shrink) {
3554 ret = skb_cow(skb, len_diff);
3555 if (unlikely(ret < 0))
3556 return ret;
3557 __skb_push(skb, len_diff_abs);
3558 memset(skb->data, 0, len_diff_abs);
3559 } else {
3560 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3561 return -ENOMEM;
3562 __skb_pull(skb, len_diff_abs);
3563 }
3564 if (tls_sw_has_ctx_rx(skb->sk)) {
3565 struct strp_msg *rxm = strp_msg(skb);
3566
3567 rxm->full_len += len_diff;
3568 }
3569 return ret;
3570 }
3571
3572 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3573 .func = sk_skb_adjust_room,
3574 .gpl_only = false,
3575 .ret_type = RET_INTEGER,
3576 .arg1_type = ARG_PTR_TO_CTX,
3577 .arg2_type = ARG_ANYTHING,
3578 .arg3_type = ARG_ANYTHING,
3579 .arg4_type = ARG_ANYTHING,
3580 };
3581
3582 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3583 u32, mode, u64, flags)
3584 {
3585 u32 len_cur, len_diff_abs = abs(len_diff);
3586 u32 len_min = bpf_skb_net_base_len(skb);
3587 u32 len_max = BPF_SKB_MAX_LEN;
3588 __be16 proto = skb->protocol;
3589 bool shrink = len_diff < 0;
3590 u32 off;
3591 int ret;
3592
3593 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3594 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3595 return -EINVAL;
3596 if (unlikely(len_diff_abs > 0xfffU))
3597 return -EFAULT;
3598 if (unlikely(proto != htons(ETH_P_IP) &&
3599 proto != htons(ETH_P_IPV6)))
3600 return -ENOTSUPP;
3601
3602 off = skb_mac_header_len(skb);
3603 switch (mode) {
3604 case BPF_ADJ_ROOM_NET:
3605 off += bpf_skb_net_base_len(skb);
3606 break;
3607 case BPF_ADJ_ROOM_MAC:
3608 break;
3609 default:
3610 return -ENOTSUPP;
3611 }
3612
3613 len_cur = skb->len - skb_network_offset(skb);
3614 if ((shrink && (len_diff_abs >= len_cur ||
3615 len_cur - len_diff_abs < len_min)) ||
3616 (!shrink && (skb->len + len_diff_abs > len_max &&
3617 !skb_is_gso(skb))))
3618 return -ENOTSUPP;
3619
3620 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3621 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3622 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3623 __skb_reset_checksum_unnecessary(skb);
3624
3625 bpf_compute_data_pointers(skb);
3626 return ret;
3627 }
3628
3629 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3630 .func = bpf_skb_adjust_room,
3631 .gpl_only = false,
3632 .ret_type = RET_INTEGER,
3633 .arg1_type = ARG_PTR_TO_CTX,
3634 .arg2_type = ARG_ANYTHING,
3635 .arg3_type = ARG_ANYTHING,
3636 .arg4_type = ARG_ANYTHING,
3637 };
3638
3639 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3640 {
3641 u32 min_len = skb_network_offset(skb);
3642
3643 if (skb_transport_header_was_set(skb))
3644 min_len = skb_transport_offset(skb);
3645 if (skb->ip_summed == CHECKSUM_PARTIAL)
3646 min_len = skb_checksum_start_offset(skb) +
3647 skb->csum_offset + sizeof(__sum16);
3648 return min_len;
3649 }
3650
3651 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3652 {
3653 unsigned int old_len = skb->len;
3654 int ret;
3655
3656 ret = __skb_grow_rcsum(skb, new_len);
3657 if (!ret)
3658 memset(skb->data + old_len, 0, new_len - old_len);
3659 return ret;
3660 }
3661
3662 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3663 {
3664 return __skb_trim_rcsum(skb, new_len);
3665 }
3666
3667 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3668 u64 flags)
3669 {
3670 u32 max_len = BPF_SKB_MAX_LEN;
3671 u32 min_len = __bpf_skb_min_len(skb);
3672 int ret;
3673
3674 if (unlikely(flags || new_len > max_len || new_len < min_len))
3675 return -EINVAL;
3676 if (skb->encapsulation)
3677 return -ENOTSUPP;
3678
3679 /* The basic idea of this helper is that it's performing the
3680 * needed work to either grow or trim an skb, and eBPF program
3681 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3682 * bpf_lX_csum_replace() and others rather than passing a raw
3683 * buffer here. This one is a slow path helper and intended
3684 * for replies with control messages.
3685 *
3686 * Like in bpf_skb_change_proto(), we want to keep this rather
3687 * minimal and without protocol specifics so that we are able
3688 * to separate concerns as in bpf_skb_store_bytes() should only
3689 * be the one responsible for writing buffers.
3690 *
3691 * It's really expected to be a slow path operation here for
3692 * control message replies, so we're implicitly linearizing,
3693 * uncloning and drop offloads from the skb by this.
3694 */
3695 ret = __bpf_try_make_writable(skb, skb->len);
3696 if (!ret) {
3697 if (new_len > skb->len)
3698 ret = bpf_skb_grow_rcsum(skb, new_len);
3699 else if (new_len < skb->len)
3700 ret = bpf_skb_trim_rcsum(skb, new_len);
3701 if (!ret && skb_is_gso(skb))
3702 skb_gso_reset(skb);
3703 }
3704 return ret;
3705 }
3706
3707 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3708 u64, flags)
3709 {
3710 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3711
3712 bpf_compute_data_pointers(skb);
3713 return ret;
3714 }
3715
3716 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3717 .func = bpf_skb_change_tail,
3718 .gpl_only = false,
3719 .ret_type = RET_INTEGER,
3720 .arg1_type = ARG_PTR_TO_CTX,
3721 .arg2_type = ARG_ANYTHING,
3722 .arg3_type = ARG_ANYTHING,
3723 };
3724
3725 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3726 u64, flags)
3727 {
3728 return __bpf_skb_change_tail(skb, new_len, flags);
3729 }
3730
3731 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3732 .func = sk_skb_change_tail,
3733 .gpl_only = false,
3734 .ret_type = RET_INTEGER,
3735 .arg1_type = ARG_PTR_TO_CTX,
3736 .arg2_type = ARG_ANYTHING,
3737 .arg3_type = ARG_ANYTHING,
3738 };
3739
3740 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3741 u64 flags)
3742 {
3743 u32 max_len = BPF_SKB_MAX_LEN;
3744 u32 new_len = skb->len + head_room;
3745 int ret;
3746
3747 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3748 new_len < skb->len))
3749 return -EINVAL;
3750
3751 ret = skb_cow(skb, head_room);
3752 if (likely(!ret)) {
3753 /* Idea for this helper is that we currently only
3754 * allow to expand on mac header. This means that
3755 * skb->protocol network header, etc, stay as is.
3756 * Compared to bpf_skb_change_tail(), we're more
3757 * flexible due to not needing to linearize or
3758 * reset GSO. Intention for this helper is to be
3759 * used by an L3 skb that needs to push mac header
3760 * for redirection into L2 device.
3761 */
3762 __skb_push(skb, head_room);
3763 memset(skb->data, 0, head_room);
3764 skb_reset_mac_header(skb);
3765 skb_reset_mac_len(skb);
3766 }
3767
3768 return ret;
3769 }
3770
3771 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3772 u64, flags)
3773 {
3774 int ret = __bpf_skb_change_head(skb, head_room, flags);
3775
3776 bpf_compute_data_pointers(skb);
3777 return ret;
3778 }
3779
3780 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3781 .func = bpf_skb_change_head,
3782 .gpl_only = false,
3783 .ret_type = RET_INTEGER,
3784 .arg1_type = ARG_PTR_TO_CTX,
3785 .arg2_type = ARG_ANYTHING,
3786 .arg3_type = ARG_ANYTHING,
3787 };
3788
3789 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3790 u64, flags)
3791 {
3792 return __bpf_skb_change_head(skb, head_room, flags);
3793 }
3794
3795 static const struct bpf_func_proto sk_skb_change_head_proto = {
3796 .func = sk_skb_change_head,
3797 .gpl_only = false,
3798 .ret_type = RET_INTEGER,
3799 .arg1_type = ARG_PTR_TO_CTX,
3800 .arg2_type = ARG_ANYTHING,
3801 .arg3_type = ARG_ANYTHING,
3802 };
3803
3804 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3805 {
3806 return xdp_get_buff_len(xdp);
3807 }
3808
3809 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3810 .func = bpf_xdp_get_buff_len,
3811 .gpl_only = false,
3812 .ret_type = RET_INTEGER,
3813 .arg1_type = ARG_PTR_TO_CTX,
3814 };
3815
3816 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3817
3818 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3819 .func = bpf_xdp_get_buff_len,
3820 .gpl_only = false,
3821 .arg1_type = ARG_PTR_TO_BTF_ID,
3822 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3823 };
3824
3825 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3826 {
3827 return xdp_data_meta_unsupported(xdp) ? 0 :
3828 xdp->data - xdp->data_meta;
3829 }
3830
3831 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3832 {
3833 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3834 unsigned long metalen = xdp_get_metalen(xdp);
3835 void *data_start = xdp_frame_end + metalen;
3836 void *data = xdp->data + offset;
3837
3838 if (unlikely(data < data_start ||
3839 data > xdp->data_end - ETH_HLEN))
3840 return -EINVAL;
3841
3842 if (metalen)
3843 memmove(xdp->data_meta + offset,
3844 xdp->data_meta, metalen);
3845 xdp->data_meta += offset;
3846 xdp->data = data;
3847
3848 return 0;
3849 }
3850
3851 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3852 .func = bpf_xdp_adjust_head,
3853 .gpl_only = false,
3854 .ret_type = RET_INTEGER,
3855 .arg1_type = ARG_PTR_TO_CTX,
3856 .arg2_type = ARG_ANYTHING,
3857 };
3858
3859 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3860 void *buf, unsigned long len, bool flush)
3861 {
3862 unsigned long ptr_len, ptr_off = 0;
3863 skb_frag_t *next_frag, *end_frag;
3864 struct skb_shared_info *sinfo;
3865 void *src, *dst;
3866 u8 *ptr_buf;
3867
3868 if (likely(xdp->data_end - xdp->data >= off + len)) {
3869 src = flush ? buf : xdp->data + off;
3870 dst = flush ? xdp->data + off : buf;
3871 memcpy(dst, src, len);
3872 return;
3873 }
3874
3875 sinfo = xdp_get_shared_info_from_buff(xdp);
3876 end_frag = &sinfo->frags[sinfo->nr_frags];
3877 next_frag = &sinfo->frags[0];
3878
3879 ptr_len = xdp->data_end - xdp->data;
3880 ptr_buf = xdp->data;
3881
3882 while (true) {
3883 if (off < ptr_off + ptr_len) {
3884 unsigned long copy_off = off - ptr_off;
3885 unsigned long copy_len = min(len, ptr_len - copy_off);
3886
3887 src = flush ? buf : ptr_buf + copy_off;
3888 dst = flush ? ptr_buf + copy_off : buf;
3889 memcpy(dst, src, copy_len);
3890
3891 off += copy_len;
3892 len -= copy_len;
3893 buf += copy_len;
3894 }
3895
3896 if (!len || next_frag == end_frag)
3897 break;
3898
3899 ptr_off += ptr_len;
3900 ptr_buf = skb_frag_address(next_frag);
3901 ptr_len = skb_frag_size(next_frag);
3902 next_frag++;
3903 }
3904 }
3905
3906 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3907 {
3908 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3909 u32 size = xdp->data_end - xdp->data;
3910 void *addr = xdp->data;
3911 int i;
3912
3913 if (unlikely(offset > 0xffff || len > 0xffff))
3914 return ERR_PTR(-EFAULT);
3915
3916 if (offset + len > xdp_get_buff_len(xdp))
3917 return ERR_PTR(-EINVAL);
3918
3919 if (offset < size) /* linear area */
3920 goto out;
3921
3922 offset -= size;
3923 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3924 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3925
3926 if (offset < frag_size) {
3927 addr = skb_frag_address(&sinfo->frags[i]);
3928 size = frag_size;
3929 break;
3930 }
3931 offset -= frag_size;
3932 }
3933 out:
3934 return offset + len <= size ? addr + offset : NULL;
3935 }
3936
3937 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3938 void *, buf, u32, len)
3939 {
3940 void *ptr;
3941
3942 ptr = bpf_xdp_pointer(xdp, offset, len);
3943 if (IS_ERR(ptr))
3944 return PTR_ERR(ptr);
3945
3946 if (!ptr)
3947 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3948 else
3949 memcpy(buf, ptr, len);
3950
3951 return 0;
3952 }
3953
3954 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3955 .func = bpf_xdp_load_bytes,
3956 .gpl_only = false,
3957 .ret_type = RET_INTEGER,
3958 .arg1_type = ARG_PTR_TO_CTX,
3959 .arg2_type = ARG_ANYTHING,
3960 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3961 .arg4_type = ARG_CONST_SIZE,
3962 };
3963
3964 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3965 void *, buf, u32, len)
3966 {
3967 void *ptr;
3968
3969 ptr = bpf_xdp_pointer(xdp, offset, len);
3970 if (IS_ERR(ptr))
3971 return PTR_ERR(ptr);
3972
3973 if (!ptr)
3974 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
3975 else
3976 memcpy(ptr, buf, len);
3977
3978 return 0;
3979 }
3980
3981 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
3982 .func = bpf_xdp_store_bytes,
3983 .gpl_only = false,
3984 .ret_type = RET_INTEGER,
3985 .arg1_type = ARG_PTR_TO_CTX,
3986 .arg2_type = ARG_ANYTHING,
3987 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3988 .arg4_type = ARG_CONST_SIZE,
3989 };
3990
3991 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
3992 {
3993 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3994 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
3995 struct xdp_rxq_info *rxq = xdp->rxq;
3996 unsigned int tailroom;
3997
3998 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
3999 return -EOPNOTSUPP;
4000
4001 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4002 if (unlikely(offset > tailroom))
4003 return -EINVAL;
4004
4005 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4006 skb_frag_size_add(frag, offset);
4007 sinfo->xdp_frags_size += offset;
4008
4009 return 0;
4010 }
4011
4012 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4013 {
4014 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4015 int i, n_frags_free = 0, len_free = 0;
4016
4017 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4018 return -EINVAL;
4019
4020 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4021 skb_frag_t *frag = &sinfo->frags[i];
4022 int shrink = min_t(int, offset, skb_frag_size(frag));
4023
4024 len_free += shrink;
4025 offset -= shrink;
4026
4027 if (skb_frag_size(frag) == shrink) {
4028 struct page *page = skb_frag_page(frag);
4029
4030 __xdp_return(page_address(page), &xdp->rxq->mem,
4031 false, NULL);
4032 n_frags_free++;
4033 } else {
4034 skb_frag_size_sub(frag, shrink);
4035 break;
4036 }
4037 }
4038 sinfo->nr_frags -= n_frags_free;
4039 sinfo->xdp_frags_size -= len_free;
4040
4041 if (unlikely(!sinfo->nr_frags)) {
4042 xdp_buff_clear_frags_flag(xdp);
4043 xdp->data_end -= offset;
4044 }
4045
4046 return 0;
4047 }
4048
4049 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4050 {
4051 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4052 void *data_end = xdp->data_end + offset;
4053
4054 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4055 if (offset < 0)
4056 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4057
4058 return bpf_xdp_frags_increase_tail(xdp, offset);
4059 }
4060
4061 /* Notice that xdp_data_hard_end have reserved some tailroom */
4062 if (unlikely(data_end > data_hard_end))
4063 return -EINVAL;
4064
4065 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4066 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4067 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4068 return -EINVAL;
4069 }
4070
4071 if (unlikely(data_end < xdp->data + ETH_HLEN))
4072 return -EINVAL;
4073
4074 /* Clear memory area on grow, can contain uninit kernel memory */
4075 if (offset > 0)
4076 memset(xdp->data_end, 0, offset);
4077
4078 xdp->data_end = data_end;
4079
4080 return 0;
4081 }
4082
4083 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4084 .func = bpf_xdp_adjust_tail,
4085 .gpl_only = false,
4086 .ret_type = RET_INTEGER,
4087 .arg1_type = ARG_PTR_TO_CTX,
4088 .arg2_type = ARG_ANYTHING,
4089 };
4090
4091 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4092 {
4093 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4094 void *meta = xdp->data_meta + offset;
4095 unsigned long metalen = xdp->data - meta;
4096
4097 if (xdp_data_meta_unsupported(xdp))
4098 return -ENOTSUPP;
4099 if (unlikely(meta < xdp_frame_end ||
4100 meta > xdp->data))
4101 return -EINVAL;
4102 if (unlikely(xdp_metalen_invalid(metalen)))
4103 return -EACCES;
4104
4105 xdp->data_meta = meta;
4106
4107 return 0;
4108 }
4109
4110 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4111 .func = bpf_xdp_adjust_meta,
4112 .gpl_only = false,
4113 .ret_type = RET_INTEGER,
4114 .arg1_type = ARG_PTR_TO_CTX,
4115 .arg2_type = ARG_ANYTHING,
4116 };
4117
4118 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4119 * below:
4120 *
4121 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4122 * of the redirect and store it (along with some other metadata) in a per-CPU
4123 * struct bpf_redirect_info.
4124 *
4125 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4126 * call xdp_do_redirect() which will use the information in struct
4127 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4128 * bulk queue structure.
4129 *
4130 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4131 * which will flush all the different bulk queues, thus completing the
4132 * redirect.
4133 *
4134 * Pointers to the map entries will be kept around for this whole sequence of
4135 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4136 * the core code; instead, the RCU protection relies on everything happening
4137 * inside a single NAPI poll sequence, which means it's between a pair of calls
4138 * to local_bh_disable()/local_bh_enable().
4139 *
4140 * The map entries are marked as __rcu and the map code makes sure to
4141 * dereference those pointers with rcu_dereference_check() in a way that works
4142 * for both sections that to hold an rcu_read_lock() and sections that are
4143 * called from NAPI without a separate rcu_read_lock(). The code below does not
4144 * use RCU annotations, but relies on those in the map code.
4145 */
4146 void xdp_do_flush(void)
4147 {
4148 __dev_flush();
4149 __cpu_map_flush();
4150 __xsk_map_flush();
4151 }
4152 EXPORT_SYMBOL_GPL(xdp_do_flush);
4153
4154 void bpf_clear_redirect_map(struct bpf_map *map)
4155 {
4156 struct bpf_redirect_info *ri;
4157 int cpu;
4158
4159 for_each_possible_cpu(cpu) {
4160 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4161 /* Avoid polluting remote cacheline due to writes if
4162 * not needed. Once we pass this test, we need the
4163 * cmpxchg() to make sure it hasn't been changed in
4164 * the meantime by remote CPU.
4165 */
4166 if (unlikely(READ_ONCE(ri->map) == map))
4167 cmpxchg(&ri->map, map, NULL);
4168 }
4169 }
4170
4171 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4172 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4173
4174 u32 xdp_master_redirect(struct xdp_buff *xdp)
4175 {
4176 struct net_device *master, *slave;
4177 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4178
4179 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4180 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4181 if (slave && slave != xdp->rxq->dev) {
4182 /* The target device is different from the receiving device, so
4183 * redirect it to the new device.
4184 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4185 * drivers to unmap the packet from their rx ring.
4186 */
4187 ri->tgt_index = slave->ifindex;
4188 ri->map_id = INT_MAX;
4189 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4190 return XDP_REDIRECT;
4191 }
4192 return XDP_TX;
4193 }
4194 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4195
4196 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4197 struct net_device *dev,
4198 struct xdp_buff *xdp,
4199 struct bpf_prog *xdp_prog)
4200 {
4201 enum bpf_map_type map_type = ri->map_type;
4202 void *fwd = ri->tgt_value;
4203 u32 map_id = ri->map_id;
4204 int err;
4205
4206 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4207 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4208
4209 err = __xsk_map_redirect(fwd, xdp);
4210 if (unlikely(err))
4211 goto err;
4212
4213 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4214 return 0;
4215 err:
4216 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4217 return err;
4218 }
4219
4220 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4221 struct net_device *dev,
4222 struct xdp_frame *xdpf,
4223 struct bpf_prog *xdp_prog)
4224 {
4225 enum bpf_map_type map_type = ri->map_type;
4226 void *fwd = ri->tgt_value;
4227 u32 map_id = ri->map_id;
4228 struct bpf_map *map;
4229 int err;
4230
4231 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4232 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4233
4234 if (unlikely(!xdpf)) {
4235 err = -EOVERFLOW;
4236 goto err;
4237 }
4238
4239 switch (map_type) {
4240 case BPF_MAP_TYPE_DEVMAP:
4241 fallthrough;
4242 case BPF_MAP_TYPE_DEVMAP_HASH:
4243 map = READ_ONCE(ri->map);
4244 if (unlikely(map)) {
4245 WRITE_ONCE(ri->map, NULL);
4246 err = dev_map_enqueue_multi(xdpf, dev, map,
4247 ri->flags & BPF_F_EXCLUDE_INGRESS);
4248 } else {
4249 err = dev_map_enqueue(fwd, xdpf, dev);
4250 }
4251 break;
4252 case BPF_MAP_TYPE_CPUMAP:
4253 err = cpu_map_enqueue(fwd, xdpf, dev);
4254 break;
4255 case BPF_MAP_TYPE_UNSPEC:
4256 if (map_id == INT_MAX) {
4257 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4258 if (unlikely(!fwd)) {
4259 err = -EINVAL;
4260 break;
4261 }
4262 err = dev_xdp_enqueue(fwd, xdpf, dev);
4263 break;
4264 }
4265 fallthrough;
4266 default:
4267 err = -EBADRQC;
4268 }
4269
4270 if (unlikely(err))
4271 goto err;
4272
4273 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4274 return 0;
4275 err:
4276 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4277 return err;
4278 }
4279
4280 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4281 struct bpf_prog *xdp_prog)
4282 {
4283 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4284 enum bpf_map_type map_type = ri->map_type;
4285
4286 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4287 * not all XDP capable drivers can map non-linear xdp_frame in
4288 * ndo_xdp_xmit.
4289 */
4290 if (unlikely(xdp_buff_has_frags(xdp) &&
4291 map_type != BPF_MAP_TYPE_CPUMAP))
4292 return -EOPNOTSUPP;
4293
4294 if (map_type == BPF_MAP_TYPE_XSKMAP)
4295 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4296
4297 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4298 xdp_prog);
4299 }
4300 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4301
4302 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4303 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4304 {
4305 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4306 enum bpf_map_type map_type = ri->map_type;
4307
4308 if (map_type == BPF_MAP_TYPE_XSKMAP)
4309 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4310
4311 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4312 }
4313 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4314
4315 static int xdp_do_generic_redirect_map(struct net_device *dev,
4316 struct sk_buff *skb,
4317 struct xdp_buff *xdp,
4318 struct bpf_prog *xdp_prog,
4319 void *fwd,
4320 enum bpf_map_type map_type, u32 map_id)
4321 {
4322 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4323 struct bpf_map *map;
4324 int err;
4325
4326 switch (map_type) {
4327 case BPF_MAP_TYPE_DEVMAP:
4328 fallthrough;
4329 case BPF_MAP_TYPE_DEVMAP_HASH:
4330 map = READ_ONCE(ri->map);
4331 if (unlikely(map)) {
4332 WRITE_ONCE(ri->map, NULL);
4333 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4334 ri->flags & BPF_F_EXCLUDE_INGRESS);
4335 } else {
4336 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4337 }
4338 if (unlikely(err))
4339 goto err;
4340 break;
4341 case BPF_MAP_TYPE_XSKMAP:
4342 err = xsk_generic_rcv(fwd, xdp);
4343 if (err)
4344 goto err;
4345 consume_skb(skb);
4346 break;
4347 case BPF_MAP_TYPE_CPUMAP:
4348 err = cpu_map_generic_redirect(fwd, skb);
4349 if (unlikely(err))
4350 goto err;
4351 break;
4352 default:
4353 err = -EBADRQC;
4354 goto err;
4355 }
4356
4357 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4358 return 0;
4359 err:
4360 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4361 return err;
4362 }
4363
4364 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4365 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4366 {
4367 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4368 enum bpf_map_type map_type = ri->map_type;
4369 void *fwd = ri->tgt_value;
4370 u32 map_id = ri->map_id;
4371 int err;
4372
4373 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4374 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4375
4376 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4377 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4378 if (unlikely(!fwd)) {
4379 err = -EINVAL;
4380 goto err;
4381 }
4382
4383 err = xdp_ok_fwd_dev(fwd, skb->len);
4384 if (unlikely(err))
4385 goto err;
4386
4387 skb->dev = fwd;
4388 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4389 generic_xdp_tx(skb, xdp_prog);
4390 return 0;
4391 }
4392
4393 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4394 err:
4395 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4396 return err;
4397 }
4398
4399 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4400 {
4401 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4402
4403 if (unlikely(flags))
4404 return XDP_ABORTED;
4405
4406 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4407 * by map_idr) is used for ifindex based XDP redirect.
4408 */
4409 ri->tgt_index = ifindex;
4410 ri->map_id = INT_MAX;
4411 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4412
4413 return XDP_REDIRECT;
4414 }
4415
4416 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4417 .func = bpf_xdp_redirect,
4418 .gpl_only = false,
4419 .ret_type = RET_INTEGER,
4420 .arg1_type = ARG_ANYTHING,
4421 .arg2_type = ARG_ANYTHING,
4422 };
4423
4424 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4425 u64, flags)
4426 {
4427 return map->ops->map_redirect(map, ifindex, flags);
4428 }
4429
4430 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4431 .func = bpf_xdp_redirect_map,
4432 .gpl_only = false,
4433 .ret_type = RET_INTEGER,
4434 .arg1_type = ARG_CONST_MAP_PTR,
4435 .arg2_type = ARG_ANYTHING,
4436 .arg3_type = ARG_ANYTHING,
4437 };
4438
4439 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4440 unsigned long off, unsigned long len)
4441 {
4442 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4443
4444 if (unlikely(!ptr))
4445 return len;
4446 if (ptr != dst_buff)
4447 memcpy(dst_buff, ptr, len);
4448
4449 return 0;
4450 }
4451
4452 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4453 u64, flags, void *, meta, u64, meta_size)
4454 {
4455 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4456
4457 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4458 return -EINVAL;
4459 if (unlikely(!skb || skb_size > skb->len))
4460 return -EFAULT;
4461
4462 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4463 bpf_skb_copy);
4464 }
4465
4466 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4467 .func = bpf_skb_event_output,
4468 .gpl_only = true,
4469 .ret_type = RET_INTEGER,
4470 .arg1_type = ARG_PTR_TO_CTX,
4471 .arg2_type = ARG_CONST_MAP_PTR,
4472 .arg3_type = ARG_ANYTHING,
4473 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4474 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4475 };
4476
4477 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4478
4479 const struct bpf_func_proto bpf_skb_output_proto = {
4480 .func = bpf_skb_event_output,
4481 .gpl_only = true,
4482 .ret_type = RET_INTEGER,
4483 .arg1_type = ARG_PTR_TO_BTF_ID,
4484 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4485 .arg2_type = ARG_CONST_MAP_PTR,
4486 .arg3_type = ARG_ANYTHING,
4487 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4488 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4489 };
4490
4491 static unsigned short bpf_tunnel_key_af(u64 flags)
4492 {
4493 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4494 }
4495
4496 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4497 u32, size, u64, flags)
4498 {
4499 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4500 u8 compat[sizeof(struct bpf_tunnel_key)];
4501 void *to_orig = to;
4502 int err;
4503
4504 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4505 err = -EINVAL;
4506 goto err_clear;
4507 }
4508 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4509 err = -EPROTO;
4510 goto err_clear;
4511 }
4512 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4513 err = -EINVAL;
4514 switch (size) {
4515 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4516 case offsetof(struct bpf_tunnel_key, tunnel_label):
4517 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4518 goto set_compat;
4519 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4520 /* Fixup deprecated structure layouts here, so we have
4521 * a common path later on.
4522 */
4523 if (ip_tunnel_info_af(info) != AF_INET)
4524 goto err_clear;
4525 set_compat:
4526 to = (struct bpf_tunnel_key *)compat;
4527 break;
4528 default:
4529 goto err_clear;
4530 }
4531 }
4532
4533 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4534 to->tunnel_tos = info->key.tos;
4535 to->tunnel_ttl = info->key.ttl;
4536 to->tunnel_ext = 0;
4537
4538 if (flags & BPF_F_TUNINFO_IPV6) {
4539 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4540 sizeof(to->remote_ipv6));
4541 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4542 sizeof(to->local_ipv6));
4543 to->tunnel_label = be32_to_cpu(info->key.label);
4544 } else {
4545 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4546 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4547 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4548 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4549 to->tunnel_label = 0;
4550 }
4551
4552 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4553 memcpy(to_orig, to, size);
4554
4555 return 0;
4556 err_clear:
4557 memset(to_orig, 0, size);
4558 return err;
4559 }
4560
4561 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4562 .func = bpf_skb_get_tunnel_key,
4563 .gpl_only = false,
4564 .ret_type = RET_INTEGER,
4565 .arg1_type = ARG_PTR_TO_CTX,
4566 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4567 .arg3_type = ARG_CONST_SIZE,
4568 .arg4_type = ARG_ANYTHING,
4569 };
4570
4571 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4572 {
4573 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4574 int err;
4575
4576 if (unlikely(!info ||
4577 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4578 err = -ENOENT;
4579 goto err_clear;
4580 }
4581 if (unlikely(size < info->options_len)) {
4582 err = -ENOMEM;
4583 goto err_clear;
4584 }
4585
4586 ip_tunnel_info_opts_get(to, info);
4587 if (size > info->options_len)
4588 memset(to + info->options_len, 0, size - info->options_len);
4589
4590 return info->options_len;
4591 err_clear:
4592 memset(to, 0, size);
4593 return err;
4594 }
4595
4596 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4597 .func = bpf_skb_get_tunnel_opt,
4598 .gpl_only = false,
4599 .ret_type = RET_INTEGER,
4600 .arg1_type = ARG_PTR_TO_CTX,
4601 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4602 .arg3_type = ARG_CONST_SIZE,
4603 };
4604
4605 static struct metadata_dst __percpu *md_dst;
4606
4607 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4608 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4609 {
4610 struct metadata_dst *md = this_cpu_ptr(md_dst);
4611 u8 compat[sizeof(struct bpf_tunnel_key)];
4612 struct ip_tunnel_info *info;
4613
4614 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4615 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4616 return -EINVAL;
4617 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4618 switch (size) {
4619 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4620 case offsetof(struct bpf_tunnel_key, tunnel_label):
4621 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4622 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4623 /* Fixup deprecated structure layouts here, so we have
4624 * a common path later on.
4625 */
4626 memcpy(compat, from, size);
4627 memset(compat + size, 0, sizeof(compat) - size);
4628 from = (const struct bpf_tunnel_key *) compat;
4629 break;
4630 default:
4631 return -EINVAL;
4632 }
4633 }
4634 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4635 from->tunnel_ext))
4636 return -EINVAL;
4637
4638 skb_dst_drop(skb);
4639 dst_hold((struct dst_entry *) md);
4640 skb_dst_set(skb, (struct dst_entry *) md);
4641
4642 info = &md->u.tun_info;
4643 memset(info, 0, sizeof(*info));
4644 info->mode = IP_TUNNEL_INFO_TX;
4645
4646 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4647 if (flags & BPF_F_DONT_FRAGMENT)
4648 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4649 if (flags & BPF_F_ZERO_CSUM_TX)
4650 info->key.tun_flags &= ~TUNNEL_CSUM;
4651 if (flags & BPF_F_SEQ_NUMBER)
4652 info->key.tun_flags |= TUNNEL_SEQ;
4653
4654 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4655 info->key.tos = from->tunnel_tos;
4656 info->key.ttl = from->tunnel_ttl;
4657
4658 if (flags & BPF_F_TUNINFO_IPV6) {
4659 info->mode |= IP_TUNNEL_INFO_IPV6;
4660 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4661 sizeof(from->remote_ipv6));
4662 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4663 sizeof(from->local_ipv6));
4664 info->key.label = cpu_to_be32(from->tunnel_label) &
4665 IPV6_FLOWLABEL_MASK;
4666 } else {
4667 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4668 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4669 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4670 }
4671
4672 return 0;
4673 }
4674
4675 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4676 .func = bpf_skb_set_tunnel_key,
4677 .gpl_only = false,
4678 .ret_type = RET_INTEGER,
4679 .arg1_type = ARG_PTR_TO_CTX,
4680 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4681 .arg3_type = ARG_CONST_SIZE,
4682 .arg4_type = ARG_ANYTHING,
4683 };
4684
4685 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4686 const u8 *, from, u32, size)
4687 {
4688 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4689 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4690
4691 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4692 return -EINVAL;
4693 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4694 return -ENOMEM;
4695
4696 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4697
4698 return 0;
4699 }
4700
4701 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4702 .func = bpf_skb_set_tunnel_opt,
4703 .gpl_only = false,
4704 .ret_type = RET_INTEGER,
4705 .arg1_type = ARG_PTR_TO_CTX,
4706 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4707 .arg3_type = ARG_CONST_SIZE,
4708 };
4709
4710 static const struct bpf_func_proto *
4711 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4712 {
4713 if (!md_dst) {
4714 struct metadata_dst __percpu *tmp;
4715
4716 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4717 METADATA_IP_TUNNEL,
4718 GFP_KERNEL);
4719 if (!tmp)
4720 return NULL;
4721 if (cmpxchg(&md_dst, NULL, tmp))
4722 metadata_dst_free_percpu(tmp);
4723 }
4724
4725 switch (which) {
4726 case BPF_FUNC_skb_set_tunnel_key:
4727 return &bpf_skb_set_tunnel_key_proto;
4728 case BPF_FUNC_skb_set_tunnel_opt:
4729 return &bpf_skb_set_tunnel_opt_proto;
4730 default:
4731 return NULL;
4732 }
4733 }
4734
4735 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4736 u32, idx)
4737 {
4738 struct bpf_array *array = container_of(map, struct bpf_array, map);
4739 struct cgroup *cgrp;
4740 struct sock *sk;
4741
4742 sk = skb_to_full_sk(skb);
4743 if (!sk || !sk_fullsock(sk))
4744 return -ENOENT;
4745 if (unlikely(idx >= array->map.max_entries))
4746 return -E2BIG;
4747
4748 cgrp = READ_ONCE(array->ptrs[idx]);
4749 if (unlikely(!cgrp))
4750 return -EAGAIN;
4751
4752 return sk_under_cgroup_hierarchy(sk, cgrp);
4753 }
4754
4755 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4756 .func = bpf_skb_under_cgroup,
4757 .gpl_only = false,
4758 .ret_type = RET_INTEGER,
4759 .arg1_type = ARG_PTR_TO_CTX,
4760 .arg2_type = ARG_CONST_MAP_PTR,
4761 .arg3_type = ARG_ANYTHING,
4762 };
4763
4764 #ifdef CONFIG_SOCK_CGROUP_DATA
4765 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4766 {
4767 struct cgroup *cgrp;
4768
4769 sk = sk_to_full_sk(sk);
4770 if (!sk || !sk_fullsock(sk))
4771 return 0;
4772
4773 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4774 return cgroup_id(cgrp);
4775 }
4776
4777 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4778 {
4779 return __bpf_sk_cgroup_id(skb->sk);
4780 }
4781
4782 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4783 .func = bpf_skb_cgroup_id,
4784 .gpl_only = false,
4785 .ret_type = RET_INTEGER,
4786 .arg1_type = ARG_PTR_TO_CTX,
4787 };
4788
4789 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4790 int ancestor_level)
4791 {
4792 struct cgroup *ancestor;
4793 struct cgroup *cgrp;
4794
4795 sk = sk_to_full_sk(sk);
4796 if (!sk || !sk_fullsock(sk))
4797 return 0;
4798
4799 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4800 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4801 if (!ancestor)
4802 return 0;
4803
4804 return cgroup_id(ancestor);
4805 }
4806
4807 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4808 ancestor_level)
4809 {
4810 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4811 }
4812
4813 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4814 .func = bpf_skb_ancestor_cgroup_id,
4815 .gpl_only = false,
4816 .ret_type = RET_INTEGER,
4817 .arg1_type = ARG_PTR_TO_CTX,
4818 .arg2_type = ARG_ANYTHING,
4819 };
4820
4821 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4822 {
4823 return __bpf_sk_cgroup_id(sk);
4824 }
4825
4826 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4827 .func = bpf_sk_cgroup_id,
4828 .gpl_only = false,
4829 .ret_type = RET_INTEGER,
4830 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4831 };
4832
4833 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4834 {
4835 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4836 }
4837
4838 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4839 .func = bpf_sk_ancestor_cgroup_id,
4840 .gpl_only = false,
4841 .ret_type = RET_INTEGER,
4842 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4843 .arg2_type = ARG_ANYTHING,
4844 };
4845 #endif
4846
4847 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4848 unsigned long off, unsigned long len)
4849 {
4850 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4851
4852 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4853 return 0;
4854 }
4855
4856 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4857 u64, flags, void *, meta, u64, meta_size)
4858 {
4859 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4860
4861 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4862 return -EINVAL;
4863
4864 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4865 return -EFAULT;
4866
4867 return bpf_event_output(map, flags, meta, meta_size, xdp,
4868 xdp_size, bpf_xdp_copy);
4869 }
4870
4871 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4872 .func = bpf_xdp_event_output,
4873 .gpl_only = true,
4874 .ret_type = RET_INTEGER,
4875 .arg1_type = ARG_PTR_TO_CTX,
4876 .arg2_type = ARG_CONST_MAP_PTR,
4877 .arg3_type = ARG_ANYTHING,
4878 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4879 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4880 };
4881
4882 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4883
4884 const struct bpf_func_proto bpf_xdp_output_proto = {
4885 .func = bpf_xdp_event_output,
4886 .gpl_only = true,
4887 .ret_type = RET_INTEGER,
4888 .arg1_type = ARG_PTR_TO_BTF_ID,
4889 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4890 .arg2_type = ARG_CONST_MAP_PTR,
4891 .arg3_type = ARG_ANYTHING,
4892 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4893 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4894 };
4895
4896 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4897 {
4898 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4899 }
4900
4901 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4902 .func = bpf_get_socket_cookie,
4903 .gpl_only = false,
4904 .ret_type = RET_INTEGER,
4905 .arg1_type = ARG_PTR_TO_CTX,
4906 };
4907
4908 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4909 {
4910 return __sock_gen_cookie(ctx->sk);
4911 }
4912
4913 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4914 .func = bpf_get_socket_cookie_sock_addr,
4915 .gpl_only = false,
4916 .ret_type = RET_INTEGER,
4917 .arg1_type = ARG_PTR_TO_CTX,
4918 };
4919
4920 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4921 {
4922 return __sock_gen_cookie(ctx);
4923 }
4924
4925 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4926 .func = bpf_get_socket_cookie_sock,
4927 .gpl_only = false,
4928 .ret_type = RET_INTEGER,
4929 .arg1_type = ARG_PTR_TO_CTX,
4930 };
4931
4932 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4933 {
4934 return sk ? sock_gen_cookie(sk) : 0;
4935 }
4936
4937 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4938 .func = bpf_get_socket_ptr_cookie,
4939 .gpl_only = false,
4940 .ret_type = RET_INTEGER,
4941 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4942 };
4943
4944 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4945 {
4946 return __sock_gen_cookie(ctx->sk);
4947 }
4948
4949 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4950 .func = bpf_get_socket_cookie_sock_ops,
4951 .gpl_only = false,
4952 .ret_type = RET_INTEGER,
4953 .arg1_type = ARG_PTR_TO_CTX,
4954 };
4955
4956 static u64 __bpf_get_netns_cookie(struct sock *sk)
4957 {
4958 const struct net *net = sk ? sock_net(sk) : &init_net;
4959
4960 return net->net_cookie;
4961 }
4962
4963 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4964 {
4965 return __bpf_get_netns_cookie(ctx);
4966 }
4967
4968 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4969 .func = bpf_get_netns_cookie_sock,
4970 .gpl_only = false,
4971 .ret_type = RET_INTEGER,
4972 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4973 };
4974
4975 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4976 {
4977 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4978 }
4979
4980 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4981 .func = bpf_get_netns_cookie_sock_addr,
4982 .gpl_only = false,
4983 .ret_type = RET_INTEGER,
4984 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4985 };
4986
4987 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4988 {
4989 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4990 }
4991
4992 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4993 .func = bpf_get_netns_cookie_sock_ops,
4994 .gpl_only = false,
4995 .ret_type = RET_INTEGER,
4996 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4997 };
4998
4999 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5000 {
5001 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5002 }
5003
5004 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5005 .func = bpf_get_netns_cookie_sk_msg,
5006 .gpl_only = false,
5007 .ret_type = RET_INTEGER,
5008 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5009 };
5010
5011 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5012 {
5013 struct sock *sk = sk_to_full_sk(skb->sk);
5014 kuid_t kuid;
5015
5016 if (!sk || !sk_fullsock(sk))
5017 return overflowuid;
5018 kuid = sock_net_uid(sock_net(sk), sk);
5019 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5020 }
5021
5022 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5023 .func = bpf_get_socket_uid,
5024 .gpl_only = false,
5025 .ret_type = RET_INTEGER,
5026 .arg1_type = ARG_PTR_TO_CTX,
5027 };
5028
5029 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5030 char *optval, int optlen)
5031 {
5032 char devname[IFNAMSIZ];
5033 int val, valbool;
5034 struct net *net;
5035 int ifindex;
5036 int ret = 0;
5037
5038 if (!sk_fullsock(sk))
5039 return -EINVAL;
5040
5041 if (level == SOL_SOCKET) {
5042 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
5043 return -EINVAL;
5044 val = *((int *)optval);
5045 valbool = val ? 1 : 0;
5046
5047 /* Only some socketops are supported */
5048 switch (optname) {
5049 case SO_RCVBUF:
5050 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
5051 val = min_t(int, val, INT_MAX / 2);
5052 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
5053 WRITE_ONCE(sk->sk_rcvbuf,
5054 max_t(int, val * 2, SOCK_MIN_RCVBUF));
5055 break;
5056 case SO_SNDBUF:
5057 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
5058 val = min_t(int, val, INT_MAX / 2);
5059 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
5060 WRITE_ONCE(sk->sk_sndbuf,
5061 max_t(int, val * 2, SOCK_MIN_SNDBUF));
5062 break;
5063 case SO_MAX_PACING_RATE: /* 32bit version */
5064 if (val != ~0U)
5065 cmpxchg(&sk->sk_pacing_status,
5066 SK_PACING_NONE,
5067 SK_PACING_NEEDED);
5068 sk->sk_max_pacing_rate = (val == ~0U) ?
5069 ~0UL : (unsigned int)val;
5070 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
5071 sk->sk_max_pacing_rate);
5072 break;
5073 case SO_PRIORITY:
5074 sk->sk_priority = val;
5075 break;
5076 case SO_RCVLOWAT:
5077 if (val < 0)
5078 val = INT_MAX;
5079 if (sk->sk_socket && sk->sk_socket->ops->set_rcvlowat)
5080 ret = sk->sk_socket->ops->set_rcvlowat(sk, val);
5081 else
5082 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
5083 break;
5084 case SO_MARK:
5085 if (sk->sk_mark != val) {
5086 sk->sk_mark = val;
5087 sk_dst_reset(sk);
5088 }
5089 break;
5090 case SO_BINDTODEVICE:
5091 optlen = min_t(long, optlen, IFNAMSIZ - 1);
5092 strncpy(devname, optval, optlen);
5093 devname[optlen] = 0;
5094
5095 ifindex = 0;
5096 if (devname[0] != '\0') {
5097 struct net_device *dev;
5098
5099 ret = -ENODEV;
5100
5101 net = sock_net(sk);
5102 dev = dev_get_by_name(net, devname);
5103 if (!dev)
5104 break;
5105 ifindex = dev->ifindex;
5106 dev_put(dev);
5107 }
5108 fallthrough;
5109 case SO_BINDTOIFINDEX:
5110 if (optname == SO_BINDTOIFINDEX)
5111 ifindex = val;
5112 ret = sock_bindtoindex(sk, ifindex, false);
5113 break;
5114 case SO_KEEPALIVE:
5115 if (sk->sk_prot->keepalive)
5116 sk->sk_prot->keepalive(sk, valbool);
5117 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
5118 break;
5119 case SO_REUSEPORT:
5120 sk->sk_reuseport = valbool;
5121 break;
5122 case SO_TXREHASH:
5123 if (val < -1 || val > 1) {
5124 ret = -EINVAL;
5125 break;
5126 }
5127 sk->sk_txrehash = (u8)val;
5128 break;
5129 default:
5130 ret = -EINVAL;
5131 }
5132 #ifdef CONFIG_INET
5133 } else if (level == SOL_IP) {
5134 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5135 return -EINVAL;
5136
5137 val = *((int *)optval);
5138 /* Only some options are supported */
5139 switch (optname) {
5140 case IP_TOS:
5141 if (val < -1 || val > 0xff) {
5142 ret = -EINVAL;
5143 } else {
5144 struct inet_sock *inet = inet_sk(sk);
5145
5146 if (val == -1)
5147 val = 0;
5148 inet->tos = val;
5149 }
5150 break;
5151 default:
5152 ret = -EINVAL;
5153 }
5154 #if IS_ENABLED(CONFIG_IPV6)
5155 } else if (level == SOL_IPV6) {
5156 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5157 return -EINVAL;
5158
5159 val = *((int *)optval);
5160 /* Only some options are supported */
5161 switch (optname) {
5162 case IPV6_TCLASS:
5163 if (val < -1 || val > 0xff) {
5164 ret = -EINVAL;
5165 } else {
5166 struct ipv6_pinfo *np = inet6_sk(sk);
5167
5168 if (val == -1)
5169 val = 0;
5170 np->tclass = val;
5171 }
5172 break;
5173 default:
5174 ret = -EINVAL;
5175 }
5176 #endif
5177 } else if (level == SOL_TCP &&
5178 sk->sk_prot->setsockopt == tcp_setsockopt) {
5179 if (optname == TCP_CONGESTION) {
5180 char name[TCP_CA_NAME_MAX];
5181
5182 strncpy(name, optval, min_t(long, optlen,
5183 TCP_CA_NAME_MAX-1));
5184 name[TCP_CA_NAME_MAX-1] = 0;
5185 ret = tcp_set_congestion_control(sk, name, false, true);
5186 } else {
5187 struct inet_connection_sock *icsk = inet_csk(sk);
5188 struct tcp_sock *tp = tcp_sk(sk);
5189 unsigned long timeout;
5190
5191 if (optlen != sizeof(int))
5192 return -EINVAL;
5193
5194 val = *((int *)optval);
5195 /* Only some options are supported */
5196 switch (optname) {
5197 case TCP_BPF_IW:
5198 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5199 ret = -EINVAL;
5200 else
5201 tcp_snd_cwnd_set(tp, val);
5202 break;
5203 case TCP_BPF_SNDCWND_CLAMP:
5204 if (val <= 0) {
5205 ret = -EINVAL;
5206 } else {
5207 tp->snd_cwnd_clamp = val;
5208 tp->snd_ssthresh = val;
5209 }
5210 break;
5211 case TCP_BPF_DELACK_MAX:
5212 timeout = usecs_to_jiffies(val);
5213 if (timeout > TCP_DELACK_MAX ||
5214 timeout < TCP_TIMEOUT_MIN)
5215 return -EINVAL;
5216 inet_csk(sk)->icsk_delack_max = timeout;
5217 break;
5218 case TCP_BPF_RTO_MIN:
5219 timeout = usecs_to_jiffies(val);
5220 if (timeout > TCP_RTO_MIN ||
5221 timeout < TCP_TIMEOUT_MIN)
5222 return -EINVAL;
5223 inet_csk(sk)->icsk_rto_min = timeout;
5224 break;
5225 case TCP_SAVE_SYN:
5226 if (val < 0 || val > 1)
5227 ret = -EINVAL;
5228 else
5229 tp->save_syn = val;
5230 break;
5231 case TCP_KEEPIDLE:
5232 ret = tcp_sock_set_keepidle_locked(sk, val);
5233 break;
5234 case TCP_KEEPINTVL:
5235 if (val < 1 || val > MAX_TCP_KEEPINTVL)
5236 ret = -EINVAL;
5237 else
5238 tp->keepalive_intvl = val * HZ;
5239 break;
5240 case TCP_KEEPCNT:
5241 if (val < 1 || val > MAX_TCP_KEEPCNT)
5242 ret = -EINVAL;
5243 else
5244 tp->keepalive_probes = val;
5245 break;
5246 case TCP_SYNCNT:
5247 if (val < 1 || val > MAX_TCP_SYNCNT)
5248 ret = -EINVAL;
5249 else
5250 icsk->icsk_syn_retries = val;
5251 break;
5252 case TCP_USER_TIMEOUT:
5253 if (val < 0)
5254 ret = -EINVAL;
5255 else
5256 icsk->icsk_user_timeout = val;
5257 break;
5258 case TCP_NOTSENT_LOWAT:
5259 tp->notsent_lowat = val;
5260 sk->sk_write_space(sk);
5261 break;
5262 case TCP_WINDOW_CLAMP:
5263 ret = tcp_set_window_clamp(sk, val);
5264 break;
5265 default:
5266 ret = -EINVAL;
5267 }
5268 }
5269 #endif
5270 } else {
5271 ret = -EINVAL;
5272 }
5273 return ret;
5274 }
5275
5276 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5277 char *optval, int optlen)
5278 {
5279 if (sk_fullsock(sk))
5280 sock_owned_by_me(sk);
5281 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5282 }
5283
5284 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5285 char *optval, int optlen)
5286 {
5287 if (!sk_fullsock(sk))
5288 goto err_clear;
5289
5290 if (level == SOL_SOCKET) {
5291 if (optlen != sizeof(int))
5292 goto err_clear;
5293
5294 switch (optname) {
5295 case SO_RCVBUF:
5296 *((int *)optval) = sk->sk_rcvbuf;
5297 break;
5298 case SO_SNDBUF:
5299 *((int *)optval) = sk->sk_sndbuf;
5300 break;
5301 case SO_MARK:
5302 *((int *)optval) = sk->sk_mark;
5303 break;
5304 case SO_PRIORITY:
5305 *((int *)optval) = sk->sk_priority;
5306 break;
5307 case SO_BINDTOIFINDEX:
5308 *((int *)optval) = sk->sk_bound_dev_if;
5309 break;
5310 case SO_REUSEPORT:
5311 *((int *)optval) = sk->sk_reuseport;
5312 break;
5313 case SO_TXREHASH:
5314 *((int *)optval) = sk->sk_txrehash;
5315 break;
5316 default:
5317 goto err_clear;
5318 }
5319 #ifdef CONFIG_INET
5320 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
5321 struct inet_connection_sock *icsk;
5322 struct tcp_sock *tp;
5323
5324 switch (optname) {
5325 case TCP_CONGESTION:
5326 icsk = inet_csk(sk);
5327
5328 if (!icsk->icsk_ca_ops || optlen <= 1)
5329 goto err_clear;
5330 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5331 optval[optlen - 1] = 0;
5332 break;
5333 case TCP_SAVED_SYN:
5334 tp = tcp_sk(sk);
5335
5336 if (optlen <= 0 || !tp->saved_syn ||
5337 optlen > tcp_saved_syn_len(tp->saved_syn))
5338 goto err_clear;
5339 memcpy(optval, tp->saved_syn->data, optlen);
5340 break;
5341 default:
5342 goto err_clear;
5343 }
5344 } else if (level == SOL_IP) {
5345 struct inet_sock *inet = inet_sk(sk);
5346
5347 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5348 goto err_clear;
5349
5350 /* Only some options are supported */
5351 switch (optname) {
5352 case IP_TOS:
5353 *((int *)optval) = (int)inet->tos;
5354 break;
5355 default:
5356 goto err_clear;
5357 }
5358 #if IS_ENABLED(CONFIG_IPV6)
5359 } else if (level == SOL_IPV6) {
5360 struct ipv6_pinfo *np = inet6_sk(sk);
5361
5362 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5363 goto err_clear;
5364
5365 /* Only some options are supported */
5366 switch (optname) {
5367 case IPV6_TCLASS:
5368 *((int *)optval) = (int)np->tclass;
5369 break;
5370 default:
5371 goto err_clear;
5372 }
5373 #endif
5374 #endif
5375 } else {
5376 goto err_clear;
5377 }
5378 return 0;
5379 err_clear:
5380 memset(optval, 0, optlen);
5381 return -EINVAL;
5382 }
5383
5384 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5385 char *optval, int optlen)
5386 {
5387 if (sk_fullsock(sk))
5388 sock_owned_by_me(sk);
5389 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5390 }
5391
5392 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5393 int, optname, char *, optval, int, optlen)
5394 {
5395 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5396 if (optlen >= sizeof("cdg") - 1 &&
5397 !strncmp("cdg", optval, optlen))
5398 return -ENOTSUPP;
5399 }
5400
5401 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5402 }
5403
5404 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5405 .func = bpf_sk_setsockopt,
5406 .gpl_only = false,
5407 .ret_type = RET_INTEGER,
5408 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5409 .arg2_type = ARG_ANYTHING,
5410 .arg3_type = ARG_ANYTHING,
5411 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5412 .arg5_type = ARG_CONST_SIZE,
5413 };
5414
5415 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5416 int, optname, char *, optval, int, optlen)
5417 {
5418 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5419 }
5420
5421 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5422 .func = bpf_sk_getsockopt,
5423 .gpl_only = false,
5424 .ret_type = RET_INTEGER,
5425 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5426 .arg2_type = ARG_ANYTHING,
5427 .arg3_type = ARG_ANYTHING,
5428 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5429 .arg5_type = ARG_CONST_SIZE,
5430 };
5431
5432 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5433 int, optname, char *, optval, int, optlen)
5434 {
5435 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5436 }
5437
5438 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5439 .func = bpf_unlocked_sk_setsockopt,
5440 .gpl_only = false,
5441 .ret_type = RET_INTEGER,
5442 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5443 .arg2_type = ARG_ANYTHING,
5444 .arg3_type = ARG_ANYTHING,
5445 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5446 .arg5_type = ARG_CONST_SIZE,
5447 };
5448
5449 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5450 int, optname, char *, optval, int, optlen)
5451 {
5452 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5453 }
5454
5455 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5456 .func = bpf_unlocked_sk_getsockopt,
5457 .gpl_only = false,
5458 .ret_type = RET_INTEGER,
5459 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5460 .arg2_type = ARG_ANYTHING,
5461 .arg3_type = ARG_ANYTHING,
5462 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5463 .arg5_type = ARG_CONST_SIZE,
5464 };
5465
5466 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5467 int, level, int, optname, char *, optval, int, optlen)
5468 {
5469 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5470 }
5471
5472 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5473 .func = bpf_sock_addr_setsockopt,
5474 .gpl_only = false,
5475 .ret_type = RET_INTEGER,
5476 .arg1_type = ARG_PTR_TO_CTX,
5477 .arg2_type = ARG_ANYTHING,
5478 .arg3_type = ARG_ANYTHING,
5479 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5480 .arg5_type = ARG_CONST_SIZE,
5481 };
5482
5483 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5484 int, level, int, optname, char *, optval, int, optlen)
5485 {
5486 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5487 }
5488
5489 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5490 .func = bpf_sock_addr_getsockopt,
5491 .gpl_only = false,
5492 .ret_type = RET_INTEGER,
5493 .arg1_type = ARG_PTR_TO_CTX,
5494 .arg2_type = ARG_ANYTHING,
5495 .arg3_type = ARG_ANYTHING,
5496 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5497 .arg5_type = ARG_CONST_SIZE,
5498 };
5499
5500 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5501 int, level, int, optname, char *, optval, int, optlen)
5502 {
5503 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5504 }
5505
5506 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5507 .func = bpf_sock_ops_setsockopt,
5508 .gpl_only = false,
5509 .ret_type = RET_INTEGER,
5510 .arg1_type = ARG_PTR_TO_CTX,
5511 .arg2_type = ARG_ANYTHING,
5512 .arg3_type = ARG_ANYTHING,
5513 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5514 .arg5_type = ARG_CONST_SIZE,
5515 };
5516
5517 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5518 int optname, const u8 **start)
5519 {
5520 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5521 const u8 *hdr_start;
5522 int ret;
5523
5524 if (syn_skb) {
5525 /* sk is a request_sock here */
5526
5527 if (optname == TCP_BPF_SYN) {
5528 hdr_start = syn_skb->data;
5529 ret = tcp_hdrlen(syn_skb);
5530 } else if (optname == TCP_BPF_SYN_IP) {
5531 hdr_start = skb_network_header(syn_skb);
5532 ret = skb_network_header_len(syn_skb) +
5533 tcp_hdrlen(syn_skb);
5534 } else {
5535 /* optname == TCP_BPF_SYN_MAC */
5536 hdr_start = skb_mac_header(syn_skb);
5537 ret = skb_mac_header_len(syn_skb) +
5538 skb_network_header_len(syn_skb) +
5539 tcp_hdrlen(syn_skb);
5540 }
5541 } else {
5542 struct sock *sk = bpf_sock->sk;
5543 struct saved_syn *saved_syn;
5544
5545 if (sk->sk_state == TCP_NEW_SYN_RECV)
5546 /* synack retransmit. bpf_sock->syn_skb will
5547 * not be available. It has to resort to
5548 * saved_syn (if it is saved).
5549 */
5550 saved_syn = inet_reqsk(sk)->saved_syn;
5551 else
5552 saved_syn = tcp_sk(sk)->saved_syn;
5553
5554 if (!saved_syn)
5555 return -ENOENT;
5556
5557 if (optname == TCP_BPF_SYN) {
5558 hdr_start = saved_syn->data +
5559 saved_syn->mac_hdrlen +
5560 saved_syn->network_hdrlen;
5561 ret = saved_syn->tcp_hdrlen;
5562 } else if (optname == TCP_BPF_SYN_IP) {
5563 hdr_start = saved_syn->data +
5564 saved_syn->mac_hdrlen;
5565 ret = saved_syn->network_hdrlen +
5566 saved_syn->tcp_hdrlen;
5567 } else {
5568 /* optname == TCP_BPF_SYN_MAC */
5569
5570 /* TCP_SAVE_SYN may not have saved the mac hdr */
5571 if (!saved_syn->mac_hdrlen)
5572 return -ENOENT;
5573
5574 hdr_start = saved_syn->data;
5575 ret = saved_syn->mac_hdrlen +
5576 saved_syn->network_hdrlen +
5577 saved_syn->tcp_hdrlen;
5578 }
5579 }
5580
5581 *start = hdr_start;
5582 return ret;
5583 }
5584
5585 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5586 int, level, int, optname, char *, optval, int, optlen)
5587 {
5588 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5589 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5590 int ret, copy_len = 0;
5591 const u8 *start;
5592
5593 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5594 if (ret > 0) {
5595 copy_len = ret;
5596 if (optlen < copy_len) {
5597 copy_len = optlen;
5598 ret = -ENOSPC;
5599 }
5600
5601 memcpy(optval, start, copy_len);
5602 }
5603
5604 /* Zero out unused buffer at the end */
5605 memset(optval + copy_len, 0, optlen - copy_len);
5606
5607 return ret;
5608 }
5609
5610 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5611 }
5612
5613 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5614 .func = bpf_sock_ops_getsockopt,
5615 .gpl_only = false,
5616 .ret_type = RET_INTEGER,
5617 .arg1_type = ARG_PTR_TO_CTX,
5618 .arg2_type = ARG_ANYTHING,
5619 .arg3_type = ARG_ANYTHING,
5620 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5621 .arg5_type = ARG_CONST_SIZE,
5622 };
5623
5624 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5625 int, argval)
5626 {
5627 struct sock *sk = bpf_sock->sk;
5628 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5629
5630 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5631 return -EINVAL;
5632
5633 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5634
5635 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5636 }
5637
5638 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5639 .func = bpf_sock_ops_cb_flags_set,
5640 .gpl_only = false,
5641 .ret_type = RET_INTEGER,
5642 .arg1_type = ARG_PTR_TO_CTX,
5643 .arg2_type = ARG_ANYTHING,
5644 };
5645
5646 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5647 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5648
5649 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5650 int, addr_len)
5651 {
5652 #ifdef CONFIG_INET
5653 struct sock *sk = ctx->sk;
5654 u32 flags = BIND_FROM_BPF;
5655 int err;
5656
5657 err = -EINVAL;
5658 if (addr_len < offsetofend(struct sockaddr, sa_family))
5659 return err;
5660 if (addr->sa_family == AF_INET) {
5661 if (addr_len < sizeof(struct sockaddr_in))
5662 return err;
5663 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5664 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5665 return __inet_bind(sk, addr, addr_len, flags);
5666 #if IS_ENABLED(CONFIG_IPV6)
5667 } else if (addr->sa_family == AF_INET6) {
5668 if (addr_len < SIN6_LEN_RFC2133)
5669 return err;
5670 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5671 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5672 /* ipv6_bpf_stub cannot be NULL, since it's called from
5673 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5674 */
5675 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5676 #endif /* CONFIG_IPV6 */
5677 }
5678 #endif /* CONFIG_INET */
5679
5680 return -EAFNOSUPPORT;
5681 }
5682
5683 static const struct bpf_func_proto bpf_bind_proto = {
5684 .func = bpf_bind,
5685 .gpl_only = false,
5686 .ret_type = RET_INTEGER,
5687 .arg1_type = ARG_PTR_TO_CTX,
5688 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5689 .arg3_type = ARG_CONST_SIZE,
5690 };
5691
5692 #ifdef CONFIG_XFRM
5693 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5694 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5695 {
5696 const struct sec_path *sp = skb_sec_path(skb);
5697 const struct xfrm_state *x;
5698
5699 if (!sp || unlikely(index >= sp->len || flags))
5700 goto err_clear;
5701
5702 x = sp->xvec[index];
5703
5704 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5705 goto err_clear;
5706
5707 to->reqid = x->props.reqid;
5708 to->spi = x->id.spi;
5709 to->family = x->props.family;
5710 to->ext = 0;
5711
5712 if (to->family == AF_INET6) {
5713 memcpy(to->remote_ipv6, x->props.saddr.a6,
5714 sizeof(to->remote_ipv6));
5715 } else {
5716 to->remote_ipv4 = x->props.saddr.a4;
5717 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5718 }
5719
5720 return 0;
5721 err_clear:
5722 memset(to, 0, size);
5723 return -EINVAL;
5724 }
5725
5726 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5727 .func = bpf_skb_get_xfrm_state,
5728 .gpl_only = false,
5729 .ret_type = RET_INTEGER,
5730 .arg1_type = ARG_PTR_TO_CTX,
5731 .arg2_type = ARG_ANYTHING,
5732 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5733 .arg4_type = ARG_CONST_SIZE,
5734 .arg5_type = ARG_ANYTHING,
5735 };
5736 #endif
5737
5738 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5739 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5740 const struct neighbour *neigh,
5741 const struct net_device *dev, u32 mtu)
5742 {
5743 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5744 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5745 params->h_vlan_TCI = 0;
5746 params->h_vlan_proto = 0;
5747 if (mtu)
5748 params->mtu_result = mtu; /* union with tot_len */
5749
5750 return 0;
5751 }
5752 #endif
5753
5754 #if IS_ENABLED(CONFIG_INET)
5755 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5756 u32 flags, bool check_mtu)
5757 {
5758 struct fib_nh_common *nhc;
5759 struct in_device *in_dev;
5760 struct neighbour *neigh;
5761 struct net_device *dev;
5762 struct fib_result res;
5763 struct flowi4 fl4;
5764 u32 mtu = 0;
5765 int err;
5766
5767 dev = dev_get_by_index_rcu(net, params->ifindex);
5768 if (unlikely(!dev))
5769 return -ENODEV;
5770
5771 /* verify forwarding is enabled on this interface */
5772 in_dev = __in_dev_get_rcu(dev);
5773 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5774 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5775
5776 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5777 fl4.flowi4_iif = 1;
5778 fl4.flowi4_oif = params->ifindex;
5779 } else {
5780 fl4.flowi4_iif = params->ifindex;
5781 fl4.flowi4_oif = 0;
5782 }
5783 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5784 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5785 fl4.flowi4_flags = 0;
5786
5787 fl4.flowi4_proto = params->l4_protocol;
5788 fl4.daddr = params->ipv4_dst;
5789 fl4.saddr = params->ipv4_src;
5790 fl4.fl4_sport = params->sport;
5791 fl4.fl4_dport = params->dport;
5792 fl4.flowi4_multipath_hash = 0;
5793
5794 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5795 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5796 struct fib_table *tb;
5797
5798 tb = fib_get_table(net, tbid);
5799 if (unlikely(!tb))
5800 return BPF_FIB_LKUP_RET_NOT_FWDED;
5801
5802 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5803 } else {
5804 fl4.flowi4_mark = 0;
5805 fl4.flowi4_secid = 0;
5806 fl4.flowi4_tun_key.tun_id = 0;
5807 fl4.flowi4_uid = sock_net_uid(net, NULL);
5808
5809 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5810 }
5811
5812 if (err) {
5813 /* map fib lookup errors to RTN_ type */
5814 if (err == -EINVAL)
5815 return BPF_FIB_LKUP_RET_BLACKHOLE;
5816 if (err == -EHOSTUNREACH)
5817 return BPF_FIB_LKUP_RET_UNREACHABLE;
5818 if (err == -EACCES)
5819 return BPF_FIB_LKUP_RET_PROHIBIT;
5820
5821 return BPF_FIB_LKUP_RET_NOT_FWDED;
5822 }
5823
5824 if (res.type != RTN_UNICAST)
5825 return BPF_FIB_LKUP_RET_NOT_FWDED;
5826
5827 if (fib_info_num_path(res.fi) > 1)
5828 fib_select_path(net, &res, &fl4, NULL);
5829
5830 if (check_mtu) {
5831 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5832 if (params->tot_len > mtu) {
5833 params->mtu_result = mtu; /* union with tot_len */
5834 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5835 }
5836 }
5837
5838 nhc = res.nhc;
5839
5840 /* do not handle lwt encaps right now */
5841 if (nhc->nhc_lwtstate)
5842 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5843
5844 dev = nhc->nhc_dev;
5845
5846 params->rt_metric = res.fi->fib_priority;
5847 params->ifindex = dev->ifindex;
5848
5849 /* xdp and cls_bpf programs are run in RCU-bh so
5850 * rcu_read_lock_bh is not needed here
5851 */
5852 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5853 if (nhc->nhc_gw_family)
5854 params->ipv4_dst = nhc->nhc_gw.ipv4;
5855
5856 neigh = __ipv4_neigh_lookup_noref(dev,
5857 (__force u32)params->ipv4_dst);
5858 } else {
5859 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5860
5861 params->family = AF_INET6;
5862 *dst = nhc->nhc_gw.ipv6;
5863 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5864 }
5865
5866 if (!neigh)
5867 return BPF_FIB_LKUP_RET_NO_NEIGH;
5868
5869 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5870 }
5871 #endif
5872
5873 #if IS_ENABLED(CONFIG_IPV6)
5874 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5875 u32 flags, bool check_mtu)
5876 {
5877 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5878 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5879 struct fib6_result res = {};
5880 struct neighbour *neigh;
5881 struct net_device *dev;
5882 struct inet6_dev *idev;
5883 struct flowi6 fl6;
5884 int strict = 0;
5885 int oif, err;
5886 u32 mtu = 0;
5887
5888 /* link local addresses are never forwarded */
5889 if (rt6_need_strict(dst) || rt6_need_strict(src))
5890 return BPF_FIB_LKUP_RET_NOT_FWDED;
5891
5892 dev = dev_get_by_index_rcu(net,