1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/mm.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
40 #include <net/sock.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <net/xfrm.h>
60 #include <net/udp.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
68 #include <net/flow.h>
69 #include <net/arp.h>
70 #include <net/ipv6.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
73 #include <net/seg6.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
80 #include <net/tls.h>
81 #include <net/xdp.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
84 #include <linux/un.h>
85 #include <net/xdp_sock_drv.h>
86
87 static const struct bpf_func_proto *
88 bpf_sk_base_func_proto(enum bpf_func_id func_id);
89
copy_bpf_fprog_from_user(struct sock_fprog * dst,sockptr_t src,int len)90 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
91 {
92 if (in_compat_syscall()) {
93 struct compat_sock_fprog f32;
94
95 if (len != sizeof(f32))
96 return -EINVAL;
97 if (copy_from_sockptr(&f32, src, sizeof(f32)))
98 return -EFAULT;
99 memset(dst, 0, sizeof(*dst));
100 dst->len = f32.len;
101 dst->filter = compat_ptr(f32.filter);
102 } else {
103 if (len != sizeof(*dst))
104 return -EINVAL;
105 if (copy_from_sockptr(dst, src, sizeof(*dst)))
106 return -EFAULT;
107 }
108
109 return 0;
110 }
111 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
112
113 /**
114 * sk_filter_trim_cap - run a packet through a socket filter
115 * @sk: sock associated with &sk_buff
116 * @skb: buffer to filter
117 * @cap: limit on how short the eBPF program may trim the packet
118 *
119 * Run the eBPF program and then cut skb->data to correct size returned by
120 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
121 * than pkt_len we keep whole skb->data. This is the socket level
122 * wrapper to bpf_prog_run. It returns 0 if the packet should
123 * be accepted or -EPERM if the packet should be tossed.
124 *
125 */
sk_filter_trim_cap(struct sock * sk,struct sk_buff * skb,unsigned int cap)126 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
127 {
128 int err;
129 struct sk_filter *filter;
130
131 /*
132 * If the skb was allocated from pfmemalloc reserves, only
133 * allow SOCK_MEMALLOC sockets to use it as this socket is
134 * helping free memory
135 */
136 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
137 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
138 return -ENOMEM;
139 }
140 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
141 if (err)
142 return err;
143
144 err = security_sock_rcv_skb(sk, skb);
145 if (err)
146 return err;
147
148 rcu_read_lock();
149 filter = rcu_dereference(sk->sk_filter);
150 if (filter) {
151 struct sock *save_sk = skb->sk;
152 unsigned int pkt_len;
153
154 skb->sk = sk;
155 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
156 skb->sk = save_sk;
157 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
158 }
159 rcu_read_unlock();
160
161 return err;
162 }
163 EXPORT_SYMBOL(sk_filter_trim_cap);
164
BPF_CALL_1(bpf_skb_get_pay_offset,struct sk_buff *,skb)165 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
166 {
167 return skb_get_poff(skb);
168 }
169
BPF_CALL_3(bpf_skb_get_nlattr,struct sk_buff *,skb,u32,a,u32,x)170 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
171 {
172 struct nlattr *nla;
173
174 if (skb_is_nonlinear(skb))
175 return 0;
176
177 if (skb->len < sizeof(struct nlattr))
178 return 0;
179
180 if (a > skb->len - sizeof(struct nlattr))
181 return 0;
182
183 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
184 if (nla)
185 return (void *) nla - (void *) skb->data;
186
187 return 0;
188 }
189
BPF_CALL_3(bpf_skb_get_nlattr_nest,struct sk_buff *,skb,u32,a,u32,x)190 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
191 {
192 struct nlattr *nla;
193
194 if (skb_is_nonlinear(skb))
195 return 0;
196
197 if (skb->len < sizeof(struct nlattr))
198 return 0;
199
200 if (a > skb->len - sizeof(struct nlattr))
201 return 0;
202
203 nla = (struct nlattr *) &skb->data[a];
204 if (nla->nla_len > skb->len - a)
205 return 0;
206
207 nla = nla_find_nested(nla, x);
208 if (nla)
209 return (void *) nla - (void *) skb->data;
210
211 return 0;
212 }
213
BPF_CALL_4(bpf_skb_load_helper_8,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)214 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
215 data, int, headlen, int, offset)
216 {
217 u8 tmp, *ptr;
218 const int len = sizeof(tmp);
219
220 if (offset >= 0) {
221 if (headlen - offset >= len)
222 return *(u8 *)(data + offset);
223 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
224 return tmp;
225 } else {
226 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
227 if (likely(ptr))
228 return *(u8 *)ptr;
229 }
230
231 return -EFAULT;
232 }
233
BPF_CALL_2(bpf_skb_load_helper_8_no_cache,const struct sk_buff *,skb,int,offset)234 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
235 int, offset)
236 {
237 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
238 offset);
239 }
240
BPF_CALL_4(bpf_skb_load_helper_16,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)241 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
242 data, int, headlen, int, offset)
243 {
244 __be16 tmp, *ptr;
245 const int len = sizeof(tmp);
246
247 if (offset >= 0) {
248 if (headlen - offset >= len)
249 return get_unaligned_be16(data + offset);
250 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
251 return be16_to_cpu(tmp);
252 } else {
253 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
254 if (likely(ptr))
255 return get_unaligned_be16(ptr);
256 }
257
258 return -EFAULT;
259 }
260
BPF_CALL_2(bpf_skb_load_helper_16_no_cache,const struct sk_buff *,skb,int,offset)261 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
262 int, offset)
263 {
264 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
265 offset);
266 }
267
BPF_CALL_4(bpf_skb_load_helper_32,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)268 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
269 data, int, headlen, int, offset)
270 {
271 __be32 tmp, *ptr;
272 const int len = sizeof(tmp);
273
274 if (likely(offset >= 0)) {
275 if (headlen - offset >= len)
276 return get_unaligned_be32(data + offset);
277 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
278 return be32_to_cpu(tmp);
279 } else {
280 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
281 if (likely(ptr))
282 return get_unaligned_be32(ptr);
283 }
284
285 return -EFAULT;
286 }
287
BPF_CALL_2(bpf_skb_load_helper_32_no_cache,const struct sk_buff *,skb,int,offset)288 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
289 int, offset)
290 {
291 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
292 offset);
293 }
294
convert_skb_access(int skb_field,int dst_reg,int src_reg,struct bpf_insn * insn_buf)295 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
296 struct bpf_insn *insn_buf)
297 {
298 struct bpf_insn *insn = insn_buf;
299
300 switch (skb_field) {
301 case SKF_AD_MARK:
302 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
303
304 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
305 offsetof(struct sk_buff, mark));
306 break;
307
308 case SKF_AD_PKTTYPE:
309 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
310 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
311 #ifdef __BIG_ENDIAN_BITFIELD
312 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
313 #endif
314 break;
315
316 case SKF_AD_QUEUE:
317 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
318
319 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
320 offsetof(struct sk_buff, queue_mapping));
321 break;
322
323 case SKF_AD_VLAN_TAG:
324 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
325
326 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
327 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
328 offsetof(struct sk_buff, vlan_tci));
329 break;
330 case SKF_AD_VLAN_TAG_PRESENT:
331 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
332 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
333 offsetof(struct sk_buff, vlan_all));
334 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
335 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
336 break;
337 }
338
339 return insn - insn_buf;
340 }
341
convert_bpf_extensions(struct sock_filter * fp,struct bpf_insn ** insnp)342 static bool convert_bpf_extensions(struct sock_filter *fp,
343 struct bpf_insn **insnp)
344 {
345 struct bpf_insn *insn = *insnp;
346 u32 cnt;
347
348 switch (fp->k) {
349 case SKF_AD_OFF + SKF_AD_PROTOCOL:
350 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
351
352 /* A = *(u16 *) (CTX + offsetof(protocol)) */
353 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
354 offsetof(struct sk_buff, protocol));
355 /* A = ntohs(A) [emitting a nop or swap16] */
356 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
357 break;
358
359 case SKF_AD_OFF + SKF_AD_PKTTYPE:
360 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
361 insn += cnt - 1;
362 break;
363
364 case SKF_AD_OFF + SKF_AD_IFINDEX:
365 case SKF_AD_OFF + SKF_AD_HATYPE:
366 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
367 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
368
369 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
370 BPF_REG_TMP, BPF_REG_CTX,
371 offsetof(struct sk_buff, dev));
372 /* if (tmp != 0) goto pc + 1 */
373 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
374 *insn++ = BPF_EXIT_INSN();
375 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
376 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
377 offsetof(struct net_device, ifindex));
378 else
379 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
380 offsetof(struct net_device, type));
381 break;
382
383 case SKF_AD_OFF + SKF_AD_MARK:
384 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
385 insn += cnt - 1;
386 break;
387
388 case SKF_AD_OFF + SKF_AD_RXHASH:
389 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
390
391 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
392 offsetof(struct sk_buff, hash));
393 break;
394
395 case SKF_AD_OFF + SKF_AD_QUEUE:
396 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
397 insn += cnt - 1;
398 break;
399
400 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
401 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
402 BPF_REG_A, BPF_REG_CTX, insn);
403 insn += cnt - 1;
404 break;
405
406 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
407 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
408 BPF_REG_A, BPF_REG_CTX, insn);
409 insn += cnt - 1;
410 break;
411
412 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
413 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
414
415 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
416 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
417 offsetof(struct sk_buff, vlan_proto));
418 /* A = ntohs(A) [emitting a nop or swap16] */
419 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
420 break;
421
422 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
423 case SKF_AD_OFF + SKF_AD_NLATTR:
424 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
425 case SKF_AD_OFF + SKF_AD_CPU:
426 case SKF_AD_OFF + SKF_AD_RANDOM:
427 /* arg1 = CTX */
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
429 /* arg2 = A */
430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
431 /* arg3 = X */
432 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
433 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
434 switch (fp->k) {
435 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
436 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
437 break;
438 case SKF_AD_OFF + SKF_AD_NLATTR:
439 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
440 break;
441 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
442 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
443 break;
444 case SKF_AD_OFF + SKF_AD_CPU:
445 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
446 break;
447 case SKF_AD_OFF + SKF_AD_RANDOM:
448 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
449 bpf_user_rnd_init_once();
450 break;
451 }
452 break;
453
454 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
455 /* A ^= X */
456 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
457 break;
458
459 default:
460 /* This is just a dummy call to avoid letting the compiler
461 * evict __bpf_call_base() as an optimization. Placed here
462 * where no-one bothers.
463 */
464 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
465 return false;
466 }
467
468 *insnp = insn;
469 return true;
470 }
471
convert_bpf_ld_abs(struct sock_filter * fp,struct bpf_insn ** insnp)472 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
473 {
474 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
475 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
476 bool endian = BPF_SIZE(fp->code) == BPF_H ||
477 BPF_SIZE(fp->code) == BPF_W;
478 bool indirect = BPF_MODE(fp->code) == BPF_IND;
479 const int ip_align = NET_IP_ALIGN;
480 struct bpf_insn *insn = *insnp;
481 int offset = fp->k;
482
483 if (!indirect &&
484 ((unaligned_ok && offset >= 0) ||
485 (!unaligned_ok && offset >= 0 &&
486 offset + ip_align >= 0 &&
487 offset + ip_align % size == 0))) {
488 bool ldx_off_ok = offset <= S16_MAX;
489
490 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
491 if (offset)
492 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
493 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
494 size, 2 + endian + (!ldx_off_ok * 2));
495 if (ldx_off_ok) {
496 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
497 BPF_REG_D, offset);
498 } else {
499 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
500 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
501 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
502 BPF_REG_TMP, 0);
503 }
504 if (endian)
505 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
506 *insn++ = BPF_JMP_A(8);
507 }
508
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
510 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
511 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
512 if (!indirect) {
513 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
514 } else {
515 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
516 if (fp->k)
517 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
518 }
519
520 switch (BPF_SIZE(fp->code)) {
521 case BPF_B:
522 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
523 break;
524 case BPF_H:
525 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
526 break;
527 case BPF_W:
528 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
529 break;
530 default:
531 return false;
532 }
533
534 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
535 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
536 *insn = BPF_EXIT_INSN();
537
538 *insnp = insn;
539 return true;
540 }
541
542 /**
543 * bpf_convert_filter - convert filter program
544 * @prog: the user passed filter program
545 * @len: the length of the user passed filter program
546 * @new_prog: allocated 'struct bpf_prog' or NULL
547 * @new_len: pointer to store length of converted program
548 * @seen_ld_abs: bool whether we've seen ld_abs/ind
549 *
550 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
551 * style extended BPF (eBPF).
552 * Conversion workflow:
553 *
554 * 1) First pass for calculating the new program length:
555 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
556 *
557 * 2) 2nd pass to remap in two passes: 1st pass finds new
558 * jump offsets, 2nd pass remapping:
559 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
560 */
bpf_convert_filter(struct sock_filter * prog,int len,struct bpf_prog * new_prog,int * new_len,bool * seen_ld_abs)561 static int bpf_convert_filter(struct sock_filter *prog, int len,
562 struct bpf_prog *new_prog, int *new_len,
563 bool *seen_ld_abs)
564 {
565 int new_flen = 0, pass = 0, target, i, stack_off;
566 struct bpf_insn *new_insn, *first_insn = NULL;
567 struct sock_filter *fp;
568 int *addrs = NULL;
569 u8 bpf_src;
570
571 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
572 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
573
574 if (len <= 0 || len > BPF_MAXINSNS)
575 return -EINVAL;
576
577 if (new_prog) {
578 first_insn = new_prog->insnsi;
579 addrs = kcalloc(len, sizeof(*addrs),
580 GFP_KERNEL | __GFP_NOWARN);
581 if (!addrs)
582 return -ENOMEM;
583 }
584
585 do_pass:
586 new_insn = first_insn;
587 fp = prog;
588
589 /* Classic BPF related prologue emission. */
590 if (new_prog) {
591 /* Classic BPF expects A and X to be reset first. These need
592 * to be guaranteed to be the first two instructions.
593 */
594 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
595 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
596
597 /* All programs must keep CTX in callee saved BPF_REG_CTX.
598 * In eBPF case it's done by the compiler, here we need to
599 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
600 */
601 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
602 if (*seen_ld_abs) {
603 /* For packet access in classic BPF, cache skb->data
604 * in callee-saved BPF R8 and skb->len - skb->data_len
605 * (headlen) in BPF R9. Since classic BPF is read-only
606 * on CTX, we only need to cache it once.
607 */
608 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
609 BPF_REG_D, BPF_REG_CTX,
610 offsetof(struct sk_buff, data));
611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
612 offsetof(struct sk_buff, len));
613 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
614 offsetof(struct sk_buff, data_len));
615 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
616 }
617 } else {
618 new_insn += 3;
619 }
620
621 for (i = 0; i < len; fp++, i++) {
622 struct bpf_insn tmp_insns[32] = { };
623 struct bpf_insn *insn = tmp_insns;
624
625 if (addrs)
626 addrs[i] = new_insn - first_insn;
627
628 switch (fp->code) {
629 /* All arithmetic insns and skb loads map as-is. */
630 case BPF_ALU | BPF_ADD | BPF_X:
631 case BPF_ALU | BPF_ADD | BPF_K:
632 case BPF_ALU | BPF_SUB | BPF_X:
633 case BPF_ALU | BPF_SUB | BPF_K:
634 case BPF_ALU | BPF_AND | BPF_X:
635 case BPF_ALU | BPF_AND | BPF_K:
636 case BPF_ALU | BPF_OR | BPF_X:
637 case BPF_ALU | BPF_OR | BPF_K:
638 case BPF_ALU | BPF_LSH | BPF_X:
639 case BPF_ALU | BPF_LSH | BPF_K:
640 case BPF_ALU | BPF_RSH | BPF_X:
641 case BPF_ALU | BPF_RSH | BPF_K:
642 case BPF_ALU | BPF_XOR | BPF_X:
643 case BPF_ALU | BPF_XOR | BPF_K:
644 case BPF_ALU | BPF_MUL | BPF_X:
645 case BPF_ALU | BPF_MUL | BPF_K:
646 case BPF_ALU | BPF_DIV | BPF_X:
647 case BPF_ALU | BPF_DIV | BPF_K:
648 case BPF_ALU | BPF_MOD | BPF_X:
649 case BPF_ALU | BPF_MOD | BPF_K:
650 case BPF_ALU | BPF_NEG:
651 case BPF_LD | BPF_ABS | BPF_W:
652 case BPF_LD | BPF_ABS | BPF_H:
653 case BPF_LD | BPF_ABS | BPF_B:
654 case BPF_LD | BPF_IND | BPF_W:
655 case BPF_LD | BPF_IND | BPF_H:
656 case BPF_LD | BPF_IND | BPF_B:
657 /* Check for overloaded BPF extension and
658 * directly convert it if found, otherwise
659 * just move on with mapping.
660 */
661 if (BPF_CLASS(fp->code) == BPF_LD &&
662 BPF_MODE(fp->code) == BPF_ABS &&
663 convert_bpf_extensions(fp, &insn))
664 break;
665 if (BPF_CLASS(fp->code) == BPF_LD &&
666 convert_bpf_ld_abs(fp, &insn)) {
667 *seen_ld_abs = true;
668 break;
669 }
670
671 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
672 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
673 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
674 /* Error with exception code on div/mod by 0.
675 * For cBPF programs, this was always return 0.
676 */
677 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
678 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
679 *insn++ = BPF_EXIT_INSN();
680 }
681
682 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
683 break;
684
685 /* Jump transformation cannot use BPF block macros
686 * everywhere as offset calculation and target updates
687 * require a bit more work than the rest, i.e. jump
688 * opcodes map as-is, but offsets need adjustment.
689 */
690
691 #define BPF_EMIT_JMP \
692 do { \
693 const s32 off_min = S16_MIN, off_max = S16_MAX; \
694 s32 off; \
695 \
696 if (target >= len || target < 0) \
697 goto err; \
698 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
699 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
700 off -= insn - tmp_insns; \
701 /* Reject anything not fitting into insn->off. */ \
702 if (off < off_min || off > off_max) \
703 goto err; \
704 insn->off = off; \
705 } while (0)
706
707 case BPF_JMP | BPF_JA:
708 target = i + fp->k + 1;
709 insn->code = fp->code;
710 BPF_EMIT_JMP;
711 break;
712
713 case BPF_JMP | BPF_JEQ | BPF_K:
714 case BPF_JMP | BPF_JEQ | BPF_X:
715 case BPF_JMP | BPF_JSET | BPF_K:
716 case BPF_JMP | BPF_JSET | BPF_X:
717 case BPF_JMP | BPF_JGT | BPF_K:
718 case BPF_JMP | BPF_JGT | BPF_X:
719 case BPF_JMP | BPF_JGE | BPF_K:
720 case BPF_JMP | BPF_JGE | BPF_X:
721 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
722 /* BPF immediates are signed, zero extend
723 * immediate into tmp register and use it
724 * in compare insn.
725 */
726 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
727
728 insn->dst_reg = BPF_REG_A;
729 insn->src_reg = BPF_REG_TMP;
730 bpf_src = BPF_X;
731 } else {
732 insn->dst_reg = BPF_REG_A;
733 insn->imm = fp->k;
734 bpf_src = BPF_SRC(fp->code);
735 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
736 }
737
738 /* Common case where 'jump_false' is next insn. */
739 if (fp->jf == 0) {
740 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
741 target = i + fp->jt + 1;
742 BPF_EMIT_JMP;
743 break;
744 }
745
746 /* Convert some jumps when 'jump_true' is next insn. */
747 if (fp->jt == 0) {
748 switch (BPF_OP(fp->code)) {
749 case BPF_JEQ:
750 insn->code = BPF_JMP | BPF_JNE | bpf_src;
751 break;
752 case BPF_JGT:
753 insn->code = BPF_JMP | BPF_JLE | bpf_src;
754 break;
755 case BPF_JGE:
756 insn->code = BPF_JMP | BPF_JLT | bpf_src;
757 break;
758 default:
759 goto jmp_rest;
760 }
761
762 target = i + fp->jf + 1;
763 BPF_EMIT_JMP;
764 break;
765 }
766 jmp_rest:
767 /* Other jumps are mapped into two insns: Jxx and JA. */
768 target = i + fp->jt + 1;
769 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
770 BPF_EMIT_JMP;
771 insn++;
772
773 insn->code = BPF_JMP | BPF_JA;
774 target = i + fp->jf + 1;
775 BPF_EMIT_JMP;
776 break;
777
778 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
779 case BPF_LDX | BPF_MSH | BPF_B: {
780 struct sock_filter tmp = {
781 .code = BPF_LD | BPF_ABS | BPF_B,
782 .k = fp->k,
783 };
784
785 *seen_ld_abs = true;
786
787 /* X = A */
788 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
789 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
790 convert_bpf_ld_abs(&tmp, &insn);
791 insn++;
792 /* A &= 0xf */
793 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
794 /* A <<= 2 */
795 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
796 /* tmp = X */
797 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
798 /* X = A */
799 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
800 /* A = tmp */
801 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
802 break;
803 }
804 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
805 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
806 */
807 case BPF_RET | BPF_A:
808 case BPF_RET | BPF_K:
809 if (BPF_RVAL(fp->code) == BPF_K)
810 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
811 0, fp->k);
812 *insn = BPF_EXIT_INSN();
813 break;
814
815 /* Store to stack. */
816 case BPF_ST:
817 case BPF_STX:
818 stack_off = fp->k * 4 + 4;
819 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
820 BPF_ST ? BPF_REG_A : BPF_REG_X,
821 -stack_off);
822 /* check_load_and_stores() verifies that classic BPF can
823 * load from stack only after write, so tracking
824 * stack_depth for ST|STX insns is enough
825 */
826 if (new_prog && new_prog->aux->stack_depth < stack_off)
827 new_prog->aux->stack_depth = stack_off;
828 break;
829
830 /* Load from stack. */
831 case BPF_LD | BPF_MEM:
832 case BPF_LDX | BPF_MEM:
833 stack_off = fp->k * 4 + 4;
834 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
835 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
836 -stack_off);
837 break;
838
839 /* A = K or X = K */
840 case BPF_LD | BPF_IMM:
841 case BPF_LDX | BPF_IMM:
842 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
843 BPF_REG_A : BPF_REG_X, fp->k);
844 break;
845
846 /* X = A */
847 case BPF_MISC | BPF_TAX:
848 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
849 break;
850
851 /* A = X */
852 case BPF_MISC | BPF_TXA:
853 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
854 break;
855
856 /* A = skb->len or X = skb->len */
857 case BPF_LD | BPF_W | BPF_LEN:
858 case BPF_LDX | BPF_W | BPF_LEN:
859 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
860 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
861 offsetof(struct sk_buff, len));
862 break;
863
864 /* Access seccomp_data fields. */
865 case BPF_LDX | BPF_ABS | BPF_W:
866 /* A = *(u32 *) (ctx + K) */
867 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
868 break;
869
870 /* Unknown instruction. */
871 default:
872 goto err;
873 }
874
875 insn++;
876 if (new_prog)
877 memcpy(new_insn, tmp_insns,
878 sizeof(*insn) * (insn - tmp_insns));
879 new_insn += insn - tmp_insns;
880 }
881
882 if (!new_prog) {
883 /* Only calculating new length. */
884 *new_len = new_insn - first_insn;
885 if (*seen_ld_abs)
886 *new_len += 4; /* Prologue bits. */
887 return 0;
888 }
889
890 pass++;
891 if (new_flen != new_insn - first_insn) {
892 new_flen = new_insn - first_insn;
893 if (pass > 2)
894 goto err;
895 goto do_pass;
896 }
897
898 kfree(addrs);
899 BUG_ON(*new_len != new_flen);
900 return 0;
901 err:
902 kfree(addrs);
903 return -EINVAL;
904 }
905
906 /* Security:
907 *
908 * As we dont want to clear mem[] array for each packet going through
909 * __bpf_prog_run(), we check that filter loaded by user never try to read
910 * a cell if not previously written, and we check all branches to be sure
911 * a malicious user doesn't try to abuse us.
912 */
check_load_and_stores(const struct sock_filter * filter,int flen)913 static int check_load_and_stores(const struct sock_filter *filter, int flen)
914 {
915 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
916 int pc, ret = 0;
917
918 BUILD_BUG_ON(BPF_MEMWORDS > 16);
919
920 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
921 if (!masks)
922 return -ENOMEM;
923
924 memset(masks, 0xff, flen * sizeof(*masks));
925
926 for (pc = 0; pc < flen; pc++) {
927 memvalid &= masks[pc];
928
929 switch (filter[pc].code) {
930 case BPF_ST:
931 case BPF_STX:
932 memvalid |= (1 << filter[pc].k);
933 break;
934 case BPF_LD | BPF_MEM:
935 case BPF_LDX | BPF_MEM:
936 if (!(memvalid & (1 << filter[pc].k))) {
937 ret = -EINVAL;
938 goto error;
939 }
940 break;
941 case BPF_JMP | BPF_JA:
942 /* A jump must set masks on target */
943 masks[pc + 1 + filter[pc].k] &= memvalid;
944 memvalid = ~0;
945 break;
946 case BPF_JMP | BPF_JEQ | BPF_K:
947 case BPF_JMP | BPF_JEQ | BPF_X:
948 case BPF_JMP | BPF_JGE | BPF_K:
949 case BPF_JMP | BPF_JGE | BPF_X:
950 case BPF_JMP | BPF_JGT | BPF_K:
951 case BPF_JMP | BPF_JGT | BPF_X:
952 case BPF_JMP | BPF_JSET | BPF_K:
953 case BPF_JMP | BPF_JSET | BPF_X:
954 /* A jump must set masks on targets */
955 masks[pc + 1 + filter[pc].jt] &= memvalid;
956 masks[pc + 1 + filter[pc].jf] &= memvalid;
957 memvalid = ~0;
958 break;
959 }
960 }
961 error:
962 kfree(masks);
963 return ret;
964 }
965
chk_code_allowed(u16 code_to_probe)966 static bool chk_code_allowed(u16 code_to_probe)
967 {
968 static const bool codes[] = {
969 /* 32 bit ALU operations */
970 [BPF_ALU | BPF_ADD | BPF_K] = true,
971 [BPF_ALU | BPF_ADD | BPF_X] = true,
972 [BPF_ALU | BPF_SUB | BPF_K] = true,
973 [BPF_ALU | BPF_SUB | BPF_X] = true,
974 [BPF_ALU | BPF_MUL | BPF_K] = true,
975 [BPF_ALU | BPF_MUL | BPF_X] = true,
976 [BPF_ALU | BPF_DIV | BPF_K] = true,
977 [BPF_ALU | BPF_DIV | BPF_X] = true,
978 [BPF_ALU | BPF_MOD | BPF_K] = true,
979 [BPF_ALU | BPF_MOD | BPF_X] = true,
980 [BPF_ALU | BPF_AND | BPF_K] = true,
981 [BPF_ALU | BPF_AND | BPF_X] = true,
982 [BPF_ALU | BPF_OR | BPF_K] = true,
983 [BPF_ALU | BPF_OR | BPF_X] = true,
984 [BPF_ALU | BPF_XOR | BPF_K] = true,
985 [BPF_ALU | BPF_XOR | BPF_X] = true,
986 [BPF_ALU | BPF_LSH | BPF_K] = true,
987 [BPF_ALU | BPF_LSH | BPF_X] = true,
988 [BPF_ALU | BPF_RSH | BPF_K] = true,
989 [BPF_ALU | BPF_RSH | BPF_X] = true,
990 [BPF_ALU | BPF_NEG] = true,
991 /* Load instructions */
992 [BPF_LD | BPF_W | BPF_ABS] = true,
993 [BPF_LD | BPF_H | BPF_ABS] = true,
994 [BPF_LD | BPF_B | BPF_ABS] = true,
995 [BPF_LD | BPF_W | BPF_LEN] = true,
996 [BPF_LD | BPF_W | BPF_IND] = true,
997 [BPF_LD | BPF_H | BPF_IND] = true,
998 [BPF_LD | BPF_B | BPF_IND] = true,
999 [BPF_LD | BPF_IMM] = true,
1000 [BPF_LD | BPF_MEM] = true,
1001 [BPF_LDX | BPF_W | BPF_LEN] = true,
1002 [BPF_LDX | BPF_B | BPF_MSH] = true,
1003 [BPF_LDX | BPF_IMM] = true,
1004 [BPF_LDX | BPF_MEM] = true,
1005 /* Store instructions */
1006 [BPF_ST] = true,
1007 [BPF_STX] = true,
1008 /* Misc instructions */
1009 [BPF_MISC | BPF_TAX] = true,
1010 [BPF_MISC | BPF_TXA] = true,
1011 /* Return instructions */
1012 [BPF_RET | BPF_K] = true,
1013 [BPF_RET | BPF_A] = true,
1014 /* Jump instructions */
1015 [BPF_JMP | BPF_JA] = true,
1016 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1017 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1018 [BPF_JMP | BPF_JGE | BPF_K] = true,
1019 [BPF_JMP | BPF_JGE | BPF_X] = true,
1020 [BPF_JMP | BPF_JGT | BPF_K] = true,
1021 [BPF_JMP | BPF_JGT | BPF_X] = true,
1022 [BPF_JMP | BPF_JSET | BPF_K] = true,
1023 [BPF_JMP | BPF_JSET | BPF_X] = true,
1024 };
1025
1026 if (code_to_probe >= ARRAY_SIZE(codes))
1027 return false;
1028
1029 return codes[code_to_probe];
1030 }
1031
bpf_check_basics_ok(const struct sock_filter * filter,unsigned int flen)1032 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1033 unsigned int flen)
1034 {
1035 if (filter == NULL)
1036 return false;
1037 if (flen == 0 || flen > BPF_MAXINSNS)
1038 return false;
1039
1040 return true;
1041 }
1042
1043 /**
1044 * bpf_check_classic - verify socket filter code
1045 * @filter: filter to verify
1046 * @flen: length of filter
1047 *
1048 * Check the user's filter code. If we let some ugly
1049 * filter code slip through kaboom! The filter must contain
1050 * no references or jumps that are out of range, no illegal
1051 * instructions, and must end with a RET instruction.
1052 *
1053 * All jumps are forward as they are not signed.
1054 *
1055 * Returns 0 if the rule set is legal or -EINVAL if not.
1056 */
bpf_check_classic(const struct sock_filter * filter,unsigned int flen)1057 static int bpf_check_classic(const struct sock_filter *filter,
1058 unsigned int flen)
1059 {
1060 bool anc_found;
1061 int pc;
1062
1063 /* Check the filter code now */
1064 for (pc = 0; pc < flen; pc++) {
1065 const struct sock_filter *ftest = &filter[pc];
1066
1067 /* May we actually operate on this code? */
1068 if (!chk_code_allowed(ftest->code))
1069 return -EINVAL;
1070
1071 /* Some instructions need special checks */
1072 switch (ftest->code) {
1073 case BPF_ALU | BPF_DIV | BPF_K:
1074 case BPF_ALU | BPF_MOD | BPF_K:
1075 /* Check for division by zero */
1076 if (ftest->k == 0)
1077 return -EINVAL;
1078 break;
1079 case BPF_ALU | BPF_LSH | BPF_K:
1080 case BPF_ALU | BPF_RSH | BPF_K:
1081 if (ftest->k >= 32)
1082 return -EINVAL;
1083 break;
1084 case BPF_LD | BPF_MEM:
1085 case BPF_LDX | BPF_MEM:
1086 case BPF_ST:
1087 case BPF_STX:
1088 /* Check for invalid memory addresses */
1089 if (ftest->k >= BPF_MEMWORDS)
1090 return -EINVAL;
1091 break;
1092 case BPF_JMP | BPF_JA:
1093 /* Note, the large ftest->k might cause loops.
1094 * Compare this with conditional jumps below,
1095 * where offsets are limited. --ANK (981016)
1096 */
1097 if (ftest->k >= (unsigned int)(flen - pc - 1))
1098 return -EINVAL;
1099 break;
1100 case BPF_JMP | BPF_JEQ | BPF_K:
1101 case BPF_JMP | BPF_JEQ | BPF_X:
1102 case BPF_JMP | BPF_JGE | BPF_K:
1103 case BPF_JMP | BPF_JGE | BPF_X:
1104 case BPF_JMP | BPF_JGT | BPF_K:
1105 case BPF_JMP | BPF_JGT | BPF_X:
1106 case BPF_JMP | BPF_JSET | BPF_K:
1107 case BPF_JMP | BPF_JSET | BPF_X:
1108 /* Both conditionals must be safe */
1109 if (pc + ftest->jt + 1 >= flen ||
1110 pc + ftest->jf + 1 >= flen)
1111 return -EINVAL;
1112 break;
1113 case BPF_LD | BPF_W | BPF_ABS:
1114 case BPF_LD | BPF_H | BPF_ABS:
1115 case BPF_LD | BPF_B | BPF_ABS:
1116 anc_found = false;
1117 if (bpf_anc_helper(ftest) & BPF_ANC)
1118 anc_found = true;
1119 /* Ancillary operation unknown or unsupported */
1120 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1121 return -EINVAL;
1122 }
1123 }
1124
1125 /* Last instruction must be a RET code */
1126 switch (filter[flen - 1].code) {
1127 case BPF_RET | BPF_K:
1128 case BPF_RET | BPF_A:
1129 return check_load_and_stores(filter, flen);
1130 }
1131
1132 return -EINVAL;
1133 }
1134
bpf_prog_store_orig_filter(struct bpf_prog * fp,const struct sock_fprog * fprog)1135 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1136 const struct sock_fprog *fprog)
1137 {
1138 unsigned int fsize = bpf_classic_proglen(fprog);
1139 struct sock_fprog_kern *fkprog;
1140
1141 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1142 if (!fp->orig_prog)
1143 return -ENOMEM;
1144
1145 fkprog = fp->orig_prog;
1146 fkprog->len = fprog->len;
1147
1148 fkprog->filter = kmemdup(fp->insns, fsize,
1149 GFP_KERNEL | __GFP_NOWARN);
1150 if (!fkprog->filter) {
1151 kfree(fp->orig_prog);
1152 return -ENOMEM;
1153 }
1154
1155 return 0;
1156 }
1157
bpf_release_orig_filter(struct bpf_prog * fp)1158 static void bpf_release_orig_filter(struct bpf_prog *fp)
1159 {
1160 struct sock_fprog_kern *fprog = fp->orig_prog;
1161
1162 if (fprog) {
1163 kfree(fprog->filter);
1164 kfree(fprog);
1165 }
1166 }
1167
__bpf_prog_release(struct bpf_prog * prog)1168 static void __bpf_prog_release(struct bpf_prog *prog)
1169 {
1170 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1171 bpf_prog_put(prog);
1172 } else {
1173 bpf_release_orig_filter(prog);
1174 bpf_prog_free(prog);
1175 }
1176 }
1177
__sk_filter_release(struct sk_filter * fp)1178 static void __sk_filter_release(struct sk_filter *fp)
1179 {
1180 __bpf_prog_release(fp->prog);
1181 kfree(fp);
1182 }
1183
1184 /**
1185 * sk_filter_release_rcu - Release a socket filter by rcu_head
1186 * @rcu: rcu_head that contains the sk_filter to free
1187 */
sk_filter_release_rcu(struct rcu_head * rcu)1188 static void sk_filter_release_rcu(struct rcu_head *rcu)
1189 {
1190 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1191
1192 __sk_filter_release(fp);
1193 }
1194
1195 /**
1196 * sk_filter_release - release a socket filter
1197 * @fp: filter to remove
1198 *
1199 * Remove a filter from a socket and release its resources.
1200 */
sk_filter_release(struct sk_filter * fp)1201 static void sk_filter_release(struct sk_filter *fp)
1202 {
1203 if (refcount_dec_and_test(&fp->refcnt))
1204 call_rcu(&fp->rcu, sk_filter_release_rcu);
1205 }
1206
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1207 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1208 {
1209 u32 filter_size = bpf_prog_size(fp->prog->len);
1210
1211 atomic_sub(filter_size, &sk->sk_omem_alloc);
1212 sk_filter_release(fp);
1213 }
1214
1215 /* try to charge the socket memory if there is space available
1216 * return true on success
1217 */
__sk_filter_charge(struct sock * sk,struct sk_filter * fp)1218 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1219 {
1220 u32 filter_size = bpf_prog_size(fp->prog->len);
1221 int optmem_max = READ_ONCE(sysctl_optmem_max);
1222
1223 /* same check as in sock_kmalloc() */
1224 if (filter_size <= optmem_max &&
1225 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1226 atomic_add(filter_size, &sk->sk_omem_alloc);
1227 return true;
1228 }
1229 return false;
1230 }
1231
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1232 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1233 {
1234 if (!refcount_inc_not_zero(&fp->refcnt))
1235 return false;
1236
1237 if (!__sk_filter_charge(sk, fp)) {
1238 sk_filter_release(fp);
1239 return false;
1240 }
1241 return true;
1242 }
1243
bpf_migrate_filter(struct bpf_prog * fp)1244 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1245 {
1246 struct sock_filter *old_prog;
1247 struct bpf_prog *old_fp;
1248 int err, new_len, old_len = fp->len;
1249 bool seen_ld_abs = false;
1250
1251 /* We are free to overwrite insns et al right here as it won't be used at
1252 * this point in time anymore internally after the migration to the eBPF
1253 * instruction representation.
1254 */
1255 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1256 sizeof(struct bpf_insn));
1257
1258 /* Conversion cannot happen on overlapping memory areas,
1259 * so we need to keep the user BPF around until the 2nd
1260 * pass. At this time, the user BPF is stored in fp->insns.
1261 */
1262 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1263 GFP_KERNEL | __GFP_NOWARN);
1264 if (!old_prog) {
1265 err = -ENOMEM;
1266 goto out_err;
1267 }
1268
1269 /* 1st pass: calculate the new program length. */
1270 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1271 &seen_ld_abs);
1272 if (err)
1273 goto out_err_free;
1274
1275 /* Expand fp for appending the new filter representation. */
1276 old_fp = fp;
1277 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1278 if (!fp) {
1279 /* The old_fp is still around in case we couldn't
1280 * allocate new memory, so uncharge on that one.
1281 */
1282 fp = old_fp;
1283 err = -ENOMEM;
1284 goto out_err_free;
1285 }
1286
1287 fp->len = new_len;
1288
1289 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1290 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1291 &seen_ld_abs);
1292 if (err)
1293 /* 2nd bpf_convert_filter() can fail only if it fails
1294 * to allocate memory, remapping must succeed. Note,
1295 * that at this time old_fp has already been released
1296 * by krealloc().
1297 */
1298 goto out_err_free;
1299
1300 fp = bpf_prog_select_runtime(fp, &err);
1301 if (err)
1302 goto out_err_free;
1303
1304 kfree(old_prog);
1305 return fp;
1306
1307 out_err_free:
1308 kfree(old_prog);
1309 out_err:
1310 __bpf_prog_release(fp);
1311 return ERR_PTR(err);
1312 }
1313
bpf_prepare_filter(struct bpf_prog * fp,bpf_aux_classic_check_t trans)1314 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1315 bpf_aux_classic_check_t trans)
1316 {
1317 int err;
1318
1319 fp->bpf_func = NULL;
1320 fp->jited = 0;
1321
1322 err = bpf_check_classic(fp->insns, fp->len);
1323 if (err) {
1324 __bpf_prog_release(fp);
1325 return ERR_PTR(err);
1326 }
1327
1328 /* There might be additional checks and transformations
1329 * needed on classic filters, f.e. in case of seccomp.
1330 */
1331 if (trans) {
1332 err = trans(fp->insns, fp->len);
1333 if (err) {
1334 __bpf_prog_release(fp);
1335 return ERR_PTR(err);
1336 }
1337 }
1338
1339 /* Probe if we can JIT compile the filter and if so, do
1340 * the compilation of the filter.
1341 */
1342 bpf_jit_compile(fp);
1343
1344 /* JIT compiler couldn't process this filter, so do the eBPF translation
1345 * for the optimized interpreter.
1346 */
1347 if (!fp->jited)
1348 fp = bpf_migrate_filter(fp);
1349
1350 return fp;
1351 }
1352
1353 /**
1354 * bpf_prog_create - create an unattached filter
1355 * @pfp: the unattached filter that is created
1356 * @fprog: the filter program
1357 *
1358 * Create a filter independent of any socket. We first run some
1359 * sanity checks on it to make sure it does not explode on us later.
1360 * If an error occurs or there is insufficient memory for the filter
1361 * a negative errno code is returned. On success the return is zero.
1362 */
bpf_prog_create(struct bpf_prog ** pfp,struct sock_fprog_kern * fprog)1363 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1364 {
1365 unsigned int fsize = bpf_classic_proglen(fprog);
1366 struct bpf_prog *fp;
1367
1368 /* Make sure new filter is there and in the right amounts. */
1369 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1370 return -EINVAL;
1371
1372 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1373 if (!fp)
1374 return -ENOMEM;
1375
1376 memcpy(fp->insns, fprog->filter, fsize);
1377
1378 fp->len = fprog->len;
1379 /* Since unattached filters are not copied back to user
1380 * space through sk_get_filter(), we do not need to hold
1381 * a copy here, and can spare us the work.
1382 */
1383 fp->orig_prog = NULL;
1384
1385 /* bpf_prepare_filter() already takes care of freeing
1386 * memory in case something goes wrong.
1387 */
1388 fp = bpf_prepare_filter(fp, NULL);
1389 if (IS_ERR(fp))
1390 return PTR_ERR(fp);
1391
1392 *pfp = fp;
1393 return 0;
1394 }
1395 EXPORT_SYMBOL_GPL(bpf_prog_create);
1396
1397 /**
1398 * bpf_prog_create_from_user - create an unattached filter from user buffer
1399 * @pfp: the unattached filter that is created
1400 * @fprog: the filter program
1401 * @trans: post-classic verifier transformation handler
1402 * @save_orig: save classic BPF program
1403 *
1404 * This function effectively does the same as bpf_prog_create(), only
1405 * that it builds up its insns buffer from user space provided buffer.
1406 * It also allows for passing a bpf_aux_classic_check_t handler.
1407 */
bpf_prog_create_from_user(struct bpf_prog ** pfp,struct sock_fprog * fprog,bpf_aux_classic_check_t trans,bool save_orig)1408 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1409 bpf_aux_classic_check_t trans, bool save_orig)
1410 {
1411 unsigned int fsize = bpf_classic_proglen(fprog);
1412 struct bpf_prog *fp;
1413 int err;
1414
1415 /* Make sure new filter is there and in the right amounts. */
1416 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1417 return -EINVAL;
1418
1419 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1420 if (!fp)
1421 return -ENOMEM;
1422
1423 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1424 __bpf_prog_free(fp);
1425 return -EFAULT;
1426 }
1427
1428 fp->len = fprog->len;
1429 fp->orig_prog = NULL;
1430
1431 if (save_orig) {
1432 err = bpf_prog_store_orig_filter(fp, fprog);
1433 if (err) {
1434 __bpf_prog_free(fp);
1435 return -ENOMEM;
1436 }
1437 }
1438
1439 /* bpf_prepare_filter() already takes care of freeing
1440 * memory in case something goes wrong.
1441 */
1442 fp = bpf_prepare_filter(fp, trans);
1443 if (IS_ERR(fp))
1444 return PTR_ERR(fp);
1445
1446 *pfp = fp;
1447 return 0;
1448 }
1449 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1450
bpf_prog_destroy(struct bpf_prog * fp)1451 void bpf_prog_destroy(struct bpf_prog *fp)
1452 {
1453 __bpf_prog_release(fp);
1454 }
1455 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1456
__sk_attach_prog(struct bpf_prog * prog,struct sock * sk)1457 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1458 {
1459 struct sk_filter *fp, *old_fp;
1460
1461 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1462 if (!fp)
1463 return -ENOMEM;
1464
1465 fp->prog = prog;
1466
1467 if (!__sk_filter_charge(sk, fp)) {
1468 kfree(fp);
1469 return -ENOMEM;
1470 }
1471 refcount_set(&fp->refcnt, 1);
1472
1473 old_fp = rcu_dereference_protected(sk->sk_filter,
1474 lockdep_sock_is_held(sk));
1475 rcu_assign_pointer(sk->sk_filter, fp);
1476
1477 if (old_fp)
1478 sk_filter_uncharge(sk, old_fp);
1479
1480 return 0;
1481 }
1482
1483 static
__get_filter(struct sock_fprog * fprog,struct sock * sk)1484 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1485 {
1486 unsigned int fsize = bpf_classic_proglen(fprog);
1487 struct bpf_prog *prog;
1488 int err;
1489
1490 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1491 return ERR_PTR(-EPERM);
1492
1493 /* Make sure new filter is there and in the right amounts. */
1494 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1495 return ERR_PTR(-EINVAL);
1496
1497 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1498 if (!prog)
1499 return ERR_PTR(-ENOMEM);
1500
1501 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1502 __bpf_prog_free(prog);
1503 return ERR_PTR(-EFAULT);
1504 }
1505
1506 prog->len = fprog->len;
1507
1508 err = bpf_prog_store_orig_filter(prog, fprog);
1509 if (err) {
1510 __bpf_prog_free(prog);
1511 return ERR_PTR(-ENOMEM);
1512 }
1513
1514 /* bpf_prepare_filter() already takes care of freeing
1515 * memory in case something goes wrong.
1516 */
1517 return bpf_prepare_filter(prog, NULL);
1518 }
1519
1520 /**
1521 * sk_attach_filter - attach a socket filter
1522 * @fprog: the filter program
1523 * @sk: the socket to use
1524 *
1525 * Attach the user's filter code. We first run some sanity checks on
1526 * it to make sure it does not explode on us later. If an error
1527 * occurs or there is insufficient memory for the filter a negative
1528 * errno code is returned. On success the return is zero.
1529 */
sk_attach_filter(struct sock_fprog * fprog,struct sock * sk)1530 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1531 {
1532 struct bpf_prog *prog = __get_filter(fprog, sk);
1533 int err;
1534
1535 if (IS_ERR(prog))
1536 return PTR_ERR(prog);
1537
1538 err = __sk_attach_prog(prog, sk);
1539 if (err < 0) {
1540 __bpf_prog_release(prog);
1541 return err;
1542 }
1543
1544 return 0;
1545 }
1546 EXPORT_SYMBOL_GPL(sk_attach_filter);
1547
sk_reuseport_attach_filter(struct sock_fprog * fprog,struct sock * sk)1548 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1549 {
1550 struct bpf_prog *prog = __get_filter(fprog, sk);
1551 int err;
1552
1553 if (IS_ERR(prog))
1554 return PTR_ERR(prog);
1555
1556 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1557 err = -ENOMEM;
1558 else
1559 err = reuseport_attach_prog(sk, prog);
1560
1561 if (err)
1562 __bpf_prog_release(prog);
1563
1564 return err;
1565 }
1566
__get_bpf(u32 ufd,struct sock * sk)1567 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1568 {
1569 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1570 return ERR_PTR(-EPERM);
1571
1572 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1573 }
1574
sk_attach_bpf(u32 ufd,struct sock * sk)1575 int sk_attach_bpf(u32 ufd, struct sock *sk)
1576 {
1577 struct bpf_prog *prog = __get_bpf(ufd, sk);
1578 int err;
1579
1580 if (IS_ERR(prog))
1581 return PTR_ERR(prog);
1582
1583 err = __sk_attach_prog(prog, sk);
1584 if (err < 0) {
1585 bpf_prog_put(prog);
1586 return err;
1587 }
1588
1589 return 0;
1590 }
1591
sk_reuseport_attach_bpf(u32 ufd,struct sock * sk)1592 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1593 {
1594 struct bpf_prog *prog;
1595 int err;
1596
1597 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1598 return -EPERM;
1599
1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1601 if (PTR_ERR(prog) == -EINVAL)
1602 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1603 if (IS_ERR(prog))
1604 return PTR_ERR(prog);
1605
1606 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1607 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1608 * bpf prog (e.g. sockmap). It depends on the
1609 * limitation imposed by bpf_prog_load().
1610 * Hence, sysctl_optmem_max is not checked.
1611 */
1612 if ((sk->sk_type != SOCK_STREAM &&
1613 sk->sk_type != SOCK_DGRAM) ||
1614 (sk->sk_protocol != IPPROTO_UDP &&
1615 sk->sk_protocol != IPPROTO_TCP) ||
1616 (sk->sk_family != AF_INET &&
1617 sk->sk_family != AF_INET6)) {
1618 err = -ENOTSUPP;
1619 goto err_prog_put;
1620 }
1621 } else {
1622 /* BPF_PROG_TYPE_SOCKET_FILTER */
1623 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1624 err = -ENOMEM;
1625 goto err_prog_put;
1626 }
1627 }
1628
1629 err = reuseport_attach_prog(sk, prog);
1630 err_prog_put:
1631 if (err)
1632 bpf_prog_put(prog);
1633
1634 return err;
1635 }
1636
sk_reuseport_prog_free(struct bpf_prog * prog)1637 void sk_reuseport_prog_free(struct bpf_prog *prog)
1638 {
1639 if (!prog)
1640 return;
1641
1642 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1643 bpf_prog_put(prog);
1644 else
1645 bpf_prog_destroy(prog);
1646 }
1647
1648 struct bpf_scratchpad {
1649 union {
1650 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1651 u8 buff[MAX_BPF_STACK];
1652 };
1653 };
1654
1655 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1656
__bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1657 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1658 unsigned int write_len)
1659 {
1660 #ifdef CONFIG_DEBUG_NET
1661 /* Avoid a splat in pskb_may_pull_reason() */
1662 if (write_len > INT_MAX)
1663 return -EINVAL;
1664 #endif
1665 return skb_ensure_writable(skb, write_len);
1666 }
1667
bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1668 static inline int bpf_try_make_writable(struct sk_buff *skb,
1669 unsigned int write_len)
1670 {
1671 int err = __bpf_try_make_writable(skb, write_len);
1672
1673 bpf_compute_data_pointers(skb);
1674 return err;
1675 }
1676
bpf_try_make_head_writable(struct sk_buff * skb)1677 static int bpf_try_make_head_writable(struct sk_buff *skb)
1678 {
1679 return bpf_try_make_writable(skb, skb_headlen(skb));
1680 }
1681
bpf_push_mac_rcsum(struct sk_buff * skb)1682 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1683 {
1684 if (skb_at_tc_ingress(skb))
1685 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1686 }
1687
bpf_pull_mac_rcsum(struct sk_buff * skb)1688 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1689 {
1690 if (skb_at_tc_ingress(skb))
1691 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1692 }
1693
BPF_CALL_5(bpf_skb_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len,u64,flags)1694 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1695 const void *, from, u32, len, u64, flags)
1696 {
1697 void *ptr;
1698
1699 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1700 return -EINVAL;
1701 if (unlikely(offset > INT_MAX))
1702 return -EFAULT;
1703 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1704 return -EFAULT;
1705
1706 ptr = skb->data + offset;
1707 if (flags & BPF_F_RECOMPUTE_CSUM)
1708 __skb_postpull_rcsum(skb, ptr, len, offset);
1709
1710 memcpy(ptr, from, len);
1711
1712 if (flags & BPF_F_RECOMPUTE_CSUM)
1713 __skb_postpush_rcsum(skb, ptr, len, offset);
1714 if (flags & BPF_F_INVALIDATE_HASH)
1715 skb_clear_hash(skb);
1716
1717 return 0;
1718 }
1719
1720 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1721 .func = bpf_skb_store_bytes,
1722 .gpl_only = false,
1723 .ret_type = RET_INTEGER,
1724 .arg1_type = ARG_PTR_TO_CTX,
1725 .arg2_type = ARG_ANYTHING,
1726 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1727 .arg4_type = ARG_CONST_SIZE,
1728 .arg5_type = ARG_ANYTHING,
1729 };
1730
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1731 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1732 u32 len, u64 flags)
1733 {
1734 return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1735 }
1736
BPF_CALL_4(bpf_skb_load_bytes,const struct sk_buff *,skb,u32,offset,void *,to,u32,len)1737 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1738 void *, to, u32, len)
1739 {
1740 void *ptr;
1741
1742 if (unlikely(offset > INT_MAX))
1743 goto err_clear;
1744
1745 ptr = skb_header_pointer(skb, offset, len, to);
1746 if (unlikely(!ptr))
1747 goto err_clear;
1748 if (ptr != to)
1749 memcpy(to, ptr, len);
1750
1751 return 0;
1752 err_clear:
1753 memset(to, 0, len);
1754 return -EFAULT;
1755 }
1756
1757 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1758 .func = bpf_skb_load_bytes,
1759 .gpl_only = false,
1760 .ret_type = RET_INTEGER,
1761 .arg1_type = ARG_PTR_TO_CTX,
1762 .arg2_type = ARG_ANYTHING,
1763 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1764 .arg4_type = ARG_CONST_SIZE,
1765 };
1766
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1767 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1768 {
1769 return ____bpf_skb_load_bytes(skb, offset, to, len);
1770 }
1771
BPF_CALL_4(bpf_flow_dissector_load_bytes,const struct bpf_flow_dissector *,ctx,u32,offset,void *,to,u32,len)1772 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1773 const struct bpf_flow_dissector *, ctx, u32, offset,
1774 void *, to, u32, len)
1775 {
1776 void *ptr;
1777
1778 if (unlikely(offset > 0xffff))
1779 goto err_clear;
1780
1781 if (unlikely(!ctx->skb))
1782 goto err_clear;
1783
1784 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1785 if (unlikely(!ptr))
1786 goto err_clear;
1787 if (ptr != to)
1788 memcpy(to, ptr, len);
1789
1790 return 0;
1791 err_clear:
1792 memset(to, 0, len);
1793 return -EFAULT;
1794 }
1795
1796 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1797 .func = bpf_flow_dissector_load_bytes,
1798 .gpl_only = false,
1799 .ret_type = RET_INTEGER,
1800 .arg1_type = ARG_PTR_TO_CTX,
1801 .arg2_type = ARG_ANYTHING,
1802 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1803 .arg4_type = ARG_CONST_SIZE,
1804 };
1805
BPF_CALL_5(bpf_skb_load_bytes_relative,const struct sk_buff *,skb,u32,offset,void *,to,u32,len,u32,start_header)1806 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1807 u32, offset, void *, to, u32, len, u32, start_header)
1808 {
1809 u8 *end = skb_tail_pointer(skb);
1810 u8 *start, *ptr;
1811
1812 if (unlikely(offset > 0xffff))
1813 goto err_clear;
1814
1815 switch (start_header) {
1816 case BPF_HDR_START_MAC:
1817 if (unlikely(!skb_mac_header_was_set(skb)))
1818 goto err_clear;
1819 start = skb_mac_header(skb);
1820 break;
1821 case BPF_HDR_START_NET:
1822 start = skb_network_header(skb);
1823 break;
1824 default:
1825 goto err_clear;
1826 }
1827
1828 ptr = start + offset;
1829
1830 if (likely(ptr + len <= end)) {
1831 memcpy(to, ptr, len);
1832 return 0;
1833 }
1834
1835 err_clear:
1836 memset(to, 0, len);
1837 return -EFAULT;
1838 }
1839
1840 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1841 .func = bpf_skb_load_bytes_relative,
1842 .gpl_only = false,
1843 .ret_type = RET_INTEGER,
1844 .arg1_type = ARG_PTR_TO_CTX,
1845 .arg2_type = ARG_ANYTHING,
1846 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1847 .arg4_type = ARG_CONST_SIZE,
1848 .arg5_type = ARG_ANYTHING,
1849 };
1850
BPF_CALL_2(bpf_skb_pull_data,struct sk_buff *,skb,u32,len)1851 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1852 {
1853 /* Idea is the following: should the needed direct read/write
1854 * test fail during runtime, we can pull in more data and redo
1855 * again, since implicitly, we invalidate previous checks here.
1856 *
1857 * Or, since we know how much we need to make read/writeable,
1858 * this can be done once at the program beginning for direct
1859 * access case. By this we overcome limitations of only current
1860 * headroom being accessible.
1861 */
1862 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1863 }
1864
1865 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1866 .func = bpf_skb_pull_data,
1867 .gpl_only = false,
1868 .ret_type = RET_INTEGER,
1869 .arg1_type = ARG_PTR_TO_CTX,
1870 .arg2_type = ARG_ANYTHING,
1871 };
1872
BPF_CALL_1(bpf_sk_fullsock,struct sock *,sk)1873 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1874 {
1875 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1876 }
1877
1878 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1879 .func = bpf_sk_fullsock,
1880 .gpl_only = false,
1881 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1882 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1883 };
1884
sk_skb_try_make_writable(struct sk_buff * skb,unsigned int write_len)1885 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1886 unsigned int write_len)
1887 {
1888 return __bpf_try_make_writable(skb, write_len);
1889 }
1890
BPF_CALL_2(sk_skb_pull_data,struct sk_buff *,skb,u32,len)1891 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1892 {
1893 /* Idea is the following: should the needed direct read/write
1894 * test fail during runtime, we can pull in more data and redo
1895 * again, since implicitly, we invalidate previous checks here.
1896 *
1897 * Or, since we know how much we need to make read/writeable,
1898 * this can be done once at the program beginning for direct
1899 * access case. By this we overcome limitations of only current
1900 * headroom being accessible.
1901 */
1902 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1903 }
1904
1905 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1906 .func = sk_skb_pull_data,
1907 .gpl_only = false,
1908 .ret_type = RET_INTEGER,
1909 .arg1_type = ARG_PTR_TO_CTX,
1910 .arg2_type = ARG_ANYTHING,
1911 };
1912
BPF_CALL_5(bpf_l3_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1913 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1914 u64, from, u64, to, u64, flags)
1915 {
1916 __sum16 *ptr;
1917
1918 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1919 return -EINVAL;
1920 if (unlikely(offset > 0xffff || offset & 1))
1921 return -EFAULT;
1922 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1923 return -EFAULT;
1924
1925 ptr = (__sum16 *)(skb->data + offset);
1926 switch (flags & BPF_F_HDR_FIELD_MASK) {
1927 case 0:
1928 if (unlikely(from != 0))
1929 return -EINVAL;
1930
1931 csum_replace_by_diff(ptr, to);
1932 break;
1933 case 2:
1934 csum_replace2(ptr, from, to);
1935 break;
1936 case 4:
1937 csum_replace4(ptr, from, to);
1938 break;
1939 default:
1940 return -EINVAL;
1941 }
1942
1943 return 0;
1944 }
1945
1946 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1947 .func = bpf_l3_csum_replace,
1948 .gpl_only = false,
1949 .ret_type = RET_INTEGER,
1950 .arg1_type = ARG_PTR_TO_CTX,
1951 .arg2_type = ARG_ANYTHING,
1952 .arg3_type = ARG_ANYTHING,
1953 .arg4_type = ARG_ANYTHING,
1954 .arg5_type = ARG_ANYTHING,
1955 };
1956
BPF_CALL_5(bpf_l4_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1957 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1958 u64, from, u64, to, u64, flags)
1959 {
1960 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1961 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1962 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1963 __sum16 *ptr;
1964
1965 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1966 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1967 return -EINVAL;
1968 if (unlikely(offset > 0xffff || offset & 1))
1969 return -EFAULT;
1970 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1971 return -EFAULT;
1972
1973 ptr = (__sum16 *)(skb->data + offset);
1974 if (is_mmzero && !do_mforce && !*ptr)
1975 return 0;
1976
1977 switch (flags & BPF_F_HDR_FIELD_MASK) {
1978 case 0:
1979 if (unlikely(from != 0))
1980 return -EINVAL;
1981
1982 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1983 break;
1984 case 2:
1985 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1986 break;
1987 case 4:
1988 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1989 break;
1990 default:
1991 return -EINVAL;
1992 }
1993
1994 if (is_mmzero && !*ptr)
1995 *ptr = CSUM_MANGLED_0;
1996 return 0;
1997 }
1998
1999 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
2000 .func = bpf_l4_csum_replace,
2001 .gpl_only = false,
2002 .ret_type = RET_INTEGER,
2003 .arg1_type = ARG_PTR_TO_CTX,
2004 .arg2_type = ARG_ANYTHING,
2005 .arg3_type = ARG_ANYTHING,
2006 .arg4_type = ARG_ANYTHING,
2007 .arg5_type = ARG_ANYTHING,
2008 };
2009
BPF_CALL_5(bpf_csum_diff,__be32 *,from,u32,from_size,__be32 *,to,u32,to_size,__wsum,seed)2010 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2011 __be32 *, to, u32, to_size, __wsum, seed)
2012 {
2013 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2014 u32 diff_size = from_size + to_size;
2015 int i, j = 0;
2016
2017 /* This is quite flexible, some examples:
2018 *
2019 * from_size == 0, to_size > 0, seed := csum --> pushing data
2020 * from_size > 0, to_size == 0, seed := csum --> pulling data
2021 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2022 *
2023 * Even for diffing, from_size and to_size don't need to be equal.
2024 */
2025 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2026 diff_size > sizeof(sp->diff)))
2027 return -EINVAL;
2028
2029 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2030 sp->diff[j] = ~from[i];
2031 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2032 sp->diff[j] = to[i];
2033
2034 return csum_partial(sp->diff, diff_size, seed);
2035 }
2036
2037 static const struct bpf_func_proto bpf_csum_diff_proto = {
2038 .func = bpf_csum_diff,
2039 .gpl_only = false,
2040 .pkt_access = true,
2041 .ret_type = RET_INTEGER,
2042 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2043 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2044 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2045 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2046 .arg5_type = ARG_ANYTHING,
2047 };
2048
BPF_CALL_2(bpf_csum_update,struct sk_buff *,skb,__wsum,csum)2049 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2050 {
2051 /* The interface is to be used in combination with bpf_csum_diff()
2052 * for direct packet writes. csum rotation for alignment as well
2053 * as emulating csum_sub() can be done from the eBPF program.
2054 */
2055 if (skb->ip_summed == CHECKSUM_COMPLETE)
2056 return (skb->csum = csum_add(skb->csum, csum));
2057
2058 return -ENOTSUPP;
2059 }
2060
2061 static const struct bpf_func_proto bpf_csum_update_proto = {
2062 .func = bpf_csum_update,
2063 .gpl_only = false,
2064 .ret_type = RET_INTEGER,
2065 .arg1_type = ARG_PTR_TO_CTX,
2066 .arg2_type = ARG_ANYTHING,
2067 };
2068
BPF_CALL_2(bpf_csum_level,struct sk_buff *,skb,u64,level)2069 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2070 {
2071 /* The interface is to be used in combination with bpf_skb_adjust_room()
2072 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2073 * is passed as flags, for example.
2074 */
2075 switch (level) {
2076 case BPF_CSUM_LEVEL_INC:
2077 __skb_incr_checksum_unnecessary(skb);
2078 break;
2079 case BPF_CSUM_LEVEL_DEC:
2080 __skb_decr_checksum_unnecessary(skb);
2081 break;
2082 case BPF_CSUM_LEVEL_RESET:
2083 __skb_reset_checksum_unnecessary(skb);
2084 break;
2085 case BPF_CSUM_LEVEL_QUERY:
2086 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2087 skb->csum_level : -EACCES;
2088 default:
2089 return -EINVAL;
2090 }
2091
2092 return 0;
2093 }
2094
2095 static const struct bpf_func_proto bpf_csum_level_proto = {
2096 .func = bpf_csum_level,
2097 .gpl_only = false,
2098 .ret_type = RET_INTEGER,
2099 .arg1_type = ARG_PTR_TO_CTX,
2100 .arg2_type = ARG_ANYTHING,
2101 };
2102
__bpf_rx_skb(struct net_device * dev,struct sk_buff * skb)2103 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2104 {
2105 return dev_forward_skb_nomtu(dev, skb);
2106 }
2107
__bpf_rx_skb_no_mac(struct net_device * dev,struct sk_buff * skb)2108 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2109 struct sk_buff *skb)
2110 {
2111 int ret = ____dev_forward_skb(dev, skb, false);
2112
2113 if (likely(!ret)) {
2114 skb->dev = dev;
2115 ret = netif_rx(skb);
2116 }
2117
2118 return ret;
2119 }
2120
__bpf_tx_skb(struct net_device * dev,struct sk_buff * skb)2121 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2122 {
2123 int ret;
2124
2125 if (dev_xmit_recursion()) {
2126 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2127 kfree_skb(skb);
2128 return -ENETDOWN;
2129 }
2130
2131 skb->dev = dev;
2132 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2133 skb_clear_tstamp(skb);
2134
2135 dev_xmit_recursion_inc();
2136 ret = dev_queue_xmit(skb);
2137 dev_xmit_recursion_dec();
2138
2139 return ret;
2140 }
2141
__bpf_redirect_no_mac(struct sk_buff * skb,struct net_device * dev,u32 flags)2142 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2143 u32 flags)
2144 {
2145 unsigned int mlen = skb_network_offset(skb);
2146
2147 if (unlikely(skb->len <= mlen)) {
2148 kfree_skb(skb);
2149 return -ERANGE;
2150 }
2151
2152 if (mlen) {
2153 __skb_pull(skb, mlen);
2154
2155 /* At ingress, the mac header has already been pulled once.
2156 * At egress, skb_pospull_rcsum has to be done in case that
2157 * the skb is originated from ingress (i.e. a forwarded skb)
2158 * to ensure that rcsum starts at net header.
2159 */
2160 if (!skb_at_tc_ingress(skb))
2161 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2162 }
2163 skb_pop_mac_header(skb);
2164 skb_reset_mac_len(skb);
2165 return flags & BPF_F_INGRESS ?
2166 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2167 }
2168
__bpf_redirect_common(struct sk_buff * skb,struct net_device * dev,u32 flags)2169 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2170 u32 flags)
2171 {
2172 /* Verify that a link layer header is carried */
2173 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2174 kfree_skb(skb);
2175 return -ERANGE;
2176 }
2177
2178 bpf_push_mac_rcsum(skb);
2179 return flags & BPF_F_INGRESS ?
2180 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2181 }
2182
__bpf_redirect(struct sk_buff * skb,struct net_device * dev,u32 flags)2183 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2184 u32 flags)
2185 {
2186 if (dev_is_mac_header_xmit(dev))
2187 return __bpf_redirect_common(skb, dev, flags);
2188 else
2189 return __bpf_redirect_no_mac(skb, dev, flags);
2190 }
2191
2192 #if IS_ENABLED(CONFIG_IPV6)
bpf_out_neigh_v6(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2193 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2194 struct net_device *dev, struct bpf_nh_params *nh)
2195 {
2196 u32 hh_len = LL_RESERVED_SPACE(dev);
2197 const struct in6_addr *nexthop;
2198 struct dst_entry *dst = NULL;
2199 struct neighbour *neigh;
2200
2201 if (dev_xmit_recursion()) {
2202 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2203 goto out_drop;
2204 }
2205
2206 skb->dev = dev;
2207 skb_clear_tstamp(skb);
2208
2209 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2210 skb = skb_expand_head(skb, hh_len);
2211 if (!skb)
2212 return -ENOMEM;
2213 }
2214
2215 rcu_read_lock();
2216 if (!nh) {
2217 dst = skb_dst(skb);
2218 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2219 &ipv6_hdr(skb)->daddr);
2220 } else {
2221 nexthop = &nh->ipv6_nh;
2222 }
2223 neigh = ip_neigh_gw6(dev, nexthop);
2224 if (likely(!IS_ERR(neigh))) {
2225 int ret;
2226
2227 sock_confirm_neigh(skb, neigh);
2228 local_bh_disable();
2229 dev_xmit_recursion_inc();
2230 ret = neigh_output(neigh, skb, false);
2231 dev_xmit_recursion_dec();
2232 local_bh_enable();
2233 rcu_read_unlock();
2234 return ret;
2235 }
2236 rcu_read_unlock_bh();
2237 if (dst)
2238 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2239 out_drop:
2240 kfree_skb(skb);
2241 return -ENETDOWN;
2242 }
2243
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2244 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2245 struct bpf_nh_params *nh)
2246 {
2247 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2248 struct net *net = dev_net(dev);
2249 int err, ret = NET_XMIT_DROP;
2250
2251 if (!nh) {
2252 struct dst_entry *dst;
2253 struct flowi6 fl6 = {
2254 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2255 .flowi6_mark = skb->mark,
2256 .flowlabel = ip6_flowinfo(ip6h),
2257 .flowi6_oif = dev->ifindex,
2258 .flowi6_proto = ip6h->nexthdr,
2259 .daddr = ip6h->daddr,
2260 .saddr = ip6h->saddr,
2261 };
2262
2263 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2264 if (IS_ERR(dst))
2265 goto out_drop;
2266
2267 skb_dst_set(skb, dst);
2268 } else if (nh->nh_family != AF_INET6) {
2269 goto out_drop;
2270 }
2271
2272 err = bpf_out_neigh_v6(net, skb, dev, nh);
2273 if (unlikely(net_xmit_eval(err)))
2274 DEV_STATS_INC(dev, tx_errors);
2275 else
2276 ret = NET_XMIT_SUCCESS;
2277 goto out_xmit;
2278 out_drop:
2279 DEV_STATS_INC(dev, tx_errors);
2280 kfree_skb(skb);
2281 out_xmit:
2282 return ret;
2283 }
2284 #else
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2285 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2286 struct bpf_nh_params *nh)
2287 {
2288 kfree_skb(skb);
2289 return NET_XMIT_DROP;
2290 }
2291 #endif /* CONFIG_IPV6 */
2292
2293 #if IS_ENABLED(CONFIG_INET)
bpf_out_neigh_v4(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2294 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2295 struct net_device *dev, struct bpf_nh_params *nh)
2296 {
2297 u32 hh_len = LL_RESERVED_SPACE(dev);
2298 struct neighbour *neigh;
2299 bool is_v6gw = false;
2300
2301 if (dev_xmit_recursion()) {
2302 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2303 goto out_drop;
2304 }
2305
2306 skb->dev = dev;
2307 skb_clear_tstamp(skb);
2308
2309 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2310 skb = skb_expand_head(skb, hh_len);
2311 if (!skb)
2312 return -ENOMEM;
2313 }
2314
2315 rcu_read_lock();
2316 if (!nh) {
2317 struct dst_entry *dst = skb_dst(skb);
2318 struct rtable *rt = container_of(dst, struct rtable, dst);
2319
2320 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2321 } else if (nh->nh_family == AF_INET6) {
2322 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2323 is_v6gw = true;
2324 } else if (nh->nh_family == AF_INET) {
2325 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2326 } else {
2327 rcu_read_unlock();
2328 goto out_drop;
2329 }
2330
2331 if (likely(!IS_ERR(neigh))) {
2332 int ret;
2333
2334 sock_confirm_neigh(skb, neigh);
2335 local_bh_disable();
2336 dev_xmit_recursion_inc();
2337 ret = neigh_output(neigh, skb, is_v6gw);
2338 dev_xmit_recursion_dec();
2339 local_bh_enable();
2340 rcu_read_unlock();
2341 return ret;
2342 }
2343 rcu_read_unlock();
2344 out_drop:
2345 kfree_skb(skb);
2346 return -ENETDOWN;
2347 }
2348
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2349 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2350 struct bpf_nh_params *nh)
2351 {
2352 const struct iphdr *ip4h = ip_hdr(skb);
2353 struct net *net = dev_net(dev);
2354 int err, ret = NET_XMIT_DROP;
2355
2356 if (!nh) {
2357 struct flowi4 fl4 = {
2358 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2359 .flowi4_mark = skb->mark,
2360 .flowi4_tos = RT_TOS(ip4h->tos),
2361 .flowi4_oif = dev->ifindex,
2362 .flowi4_proto = ip4h->protocol,
2363 .daddr = ip4h->daddr,
2364 .saddr = ip4h->saddr,
2365 };
2366 struct rtable *rt;
2367
2368 rt = ip_route_output_flow(net, &fl4, NULL);
2369 if (IS_ERR(rt))
2370 goto out_drop;
2371 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2372 ip_rt_put(rt);
2373 goto out_drop;
2374 }
2375
2376 skb_dst_set(skb, &rt->dst);
2377 }
2378
2379 err = bpf_out_neigh_v4(net, skb, dev, nh);
2380 if (unlikely(net_xmit_eval(err)))
2381 DEV_STATS_INC(dev, tx_errors);
2382 else
2383 ret = NET_XMIT_SUCCESS;
2384 goto out_xmit;
2385 out_drop:
2386 DEV_STATS_INC(dev, tx_errors);
2387 kfree_skb(skb);
2388 out_xmit:
2389 return ret;
2390 }
2391 #else
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2392 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2393 struct bpf_nh_params *nh)
2394 {
2395 kfree_skb(skb);
2396 return NET_XMIT_DROP;
2397 }
2398 #endif /* CONFIG_INET */
2399
__bpf_redirect_neigh(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2400 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2401 struct bpf_nh_params *nh)
2402 {
2403 struct ethhdr *ethh = eth_hdr(skb);
2404
2405 if (unlikely(skb->mac_header >= skb->network_header))
2406 goto out;
2407 bpf_push_mac_rcsum(skb);
2408 if (is_multicast_ether_addr(ethh->h_dest))
2409 goto out;
2410
2411 skb_pull(skb, sizeof(*ethh));
2412 skb_unset_mac_header(skb);
2413 skb_reset_network_header(skb);
2414
2415 if (skb->protocol == htons(ETH_P_IP))
2416 return __bpf_redirect_neigh_v4(skb, dev, nh);
2417 else if (skb->protocol == htons(ETH_P_IPV6))
2418 return __bpf_redirect_neigh_v6(skb, dev, nh);
2419 out:
2420 kfree_skb(skb);
2421 return -ENOTSUPP;
2422 }
2423
2424 /* Internal, non-exposed redirect flags. */
2425 enum {
2426 BPF_F_NEIGH = (1ULL << 1),
2427 BPF_F_PEER = (1ULL << 2),
2428 BPF_F_NEXTHOP = (1ULL << 3),
2429 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2430 };
2431
BPF_CALL_3(bpf_clone_redirect,struct sk_buff *,skb,u32,ifindex,u64,flags)2432 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2433 {
2434 struct net_device *dev;
2435 struct sk_buff *clone;
2436 int ret;
2437
2438 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2439 return -EINVAL;
2440
2441 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2442 if (unlikely(!dev))
2443 return -EINVAL;
2444
2445 clone = skb_clone(skb, GFP_ATOMIC);
2446 if (unlikely(!clone))
2447 return -ENOMEM;
2448
2449 /* For direct write, we need to keep the invariant that the skbs
2450 * we're dealing with need to be uncloned. Should uncloning fail
2451 * here, we need to free the just generated clone to unclone once
2452 * again.
2453 */
2454 ret = bpf_try_make_head_writable(skb);
2455 if (unlikely(ret)) {
2456 kfree_skb(clone);
2457 return -ENOMEM;
2458 }
2459
2460 return __bpf_redirect(clone, dev, flags);
2461 }
2462
2463 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2464 .func = bpf_clone_redirect,
2465 .gpl_only = false,
2466 .ret_type = RET_INTEGER,
2467 .arg1_type = ARG_PTR_TO_CTX,
2468 .arg2_type = ARG_ANYTHING,
2469 .arg3_type = ARG_ANYTHING,
2470 };
2471
2472 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2473 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2474
skb_do_redirect(struct sk_buff * skb)2475 int skb_do_redirect(struct sk_buff *skb)
2476 {
2477 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2478 struct net *net = dev_net(skb->dev);
2479 struct net_device *dev;
2480 u32 flags = ri->flags;
2481
2482 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2483 ri->tgt_index = 0;
2484 ri->flags = 0;
2485 if (unlikely(!dev))
2486 goto out_drop;
2487 if (flags & BPF_F_PEER) {
2488 const struct net_device_ops *ops = dev->netdev_ops;
2489
2490 if (unlikely(!ops->ndo_get_peer_dev ||
2491 !skb_at_tc_ingress(skb)))
2492 goto out_drop;
2493 dev = ops->ndo_get_peer_dev(dev);
2494 if (unlikely(!dev ||
2495 !(dev->flags & IFF_UP) ||
2496 net_eq(net, dev_net(dev))))
2497 goto out_drop;
2498 skb->dev = dev;
2499 dev_sw_netstats_rx_add(dev, skb->len);
2500 return -EAGAIN;
2501 }
2502 return flags & BPF_F_NEIGH ?
2503 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2504 &ri->nh : NULL) :
2505 __bpf_redirect(skb, dev, flags);
2506 out_drop:
2507 kfree_skb(skb);
2508 return -EINVAL;
2509 }
2510
BPF_CALL_2(bpf_redirect,u32,ifindex,u64,flags)2511 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2512 {
2513 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2514
2515 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2516 return TC_ACT_SHOT;
2517
2518 ri->flags = flags;
2519 ri->tgt_index = ifindex;
2520
2521 return TC_ACT_REDIRECT;
2522 }
2523
2524 static const struct bpf_func_proto bpf_redirect_proto = {
2525 .func = bpf_redirect,
2526 .gpl_only = false,
2527 .ret_type = RET_INTEGER,
2528 .arg1_type = ARG_ANYTHING,
2529 .arg2_type = ARG_ANYTHING,
2530 };
2531
BPF_CALL_2(bpf_redirect_peer,u32,ifindex,u64,flags)2532 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2533 {
2534 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2535
2536 if (unlikely(flags))
2537 return TC_ACT_SHOT;
2538
2539 ri->flags = BPF_F_PEER;
2540 ri->tgt_index = ifindex;
2541
2542 return TC_ACT_REDIRECT;
2543 }
2544
2545 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2546 .func = bpf_redirect_peer,
2547 .gpl_only = false,
2548 .ret_type = RET_INTEGER,
2549 .arg1_type = ARG_ANYTHING,
2550 .arg2_type = ARG_ANYTHING,
2551 };
2552
BPF_CALL_4(bpf_redirect_neigh,u32,ifindex,struct bpf_redir_neigh *,params,int,plen,u64,flags)2553 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2554 int, plen, u64, flags)
2555 {
2556 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2557
2558 if (unlikely((plen && plen < sizeof(*params)) || flags))
2559 return TC_ACT_SHOT;
2560
2561 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2562 ri->tgt_index = ifindex;
2563
2564 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2565 if (plen)
2566 memcpy(&ri->nh, params, sizeof(ri->nh));
2567
2568 return TC_ACT_REDIRECT;
2569 }
2570
2571 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2572 .func = bpf_redirect_neigh,
2573 .gpl_only = false,
2574 .ret_type = RET_INTEGER,
2575 .arg1_type = ARG_ANYTHING,
2576 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2577 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2578 .arg4_type = ARG_ANYTHING,
2579 };
2580
BPF_CALL_2(bpf_msg_apply_bytes,struct sk_msg *,msg,u32,bytes)2581 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2582 {
2583 msg->apply_bytes = bytes;
2584 return 0;
2585 }
2586
2587 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2588 .func = bpf_msg_apply_bytes,
2589 .gpl_only = false,
2590 .ret_type = RET_INTEGER,
2591 .arg1_type = ARG_PTR_TO_CTX,
2592 .arg2_type = ARG_ANYTHING,
2593 };
2594
BPF_CALL_2(bpf_msg_cork_bytes,struct sk_msg *,msg,u32,bytes)2595 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2596 {
2597 msg->cork_bytes = bytes;
2598 return 0;
2599 }
2600
sk_msg_reset_curr(struct sk_msg * msg)2601 static void sk_msg_reset_curr(struct sk_msg *msg)
2602 {
2603 u32 i = msg->sg.start;
2604 u32 len = 0;
2605
2606 do {
2607 len += sk_msg_elem(msg, i)->length;
2608 sk_msg_iter_var_next(i);
2609 if (len >= msg->sg.size)
2610 break;
2611 } while (i != msg->sg.end);
2612
2613 msg->sg.curr = i;
2614 msg->sg.copybreak = 0;
2615 }
2616
2617 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2618 .func = bpf_msg_cork_bytes,
2619 .gpl_only = false,
2620 .ret_type = RET_INTEGER,
2621 .arg1_type = ARG_PTR_TO_CTX,
2622 .arg2_type = ARG_ANYTHING,
2623 };
2624
BPF_CALL_4(bpf_msg_pull_data,struct sk_msg *,msg,u32,start,u32,end,u64,flags)2625 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2626 u32, end, u64, flags)
2627 {
2628 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2629 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2630 struct scatterlist *sge;
2631 u8 *raw, *to, *from;
2632 struct page *page;
2633
2634 if (unlikely(flags || end <= start))
2635 return -EINVAL;
2636
2637 /* First find the starting scatterlist element */
2638 i = msg->sg.start;
2639 do {
2640 offset += len;
2641 len = sk_msg_elem(msg, i)->length;
2642 if (start < offset + len)
2643 break;
2644 sk_msg_iter_var_next(i);
2645 } while (i != msg->sg.end);
2646
2647 if (unlikely(start >= offset + len))
2648 return -EINVAL;
2649
2650 first_sge = i;
2651 /* The start may point into the sg element so we need to also
2652 * account for the headroom.
2653 */
2654 bytes_sg_total = start - offset + bytes;
2655 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2656 goto out;
2657
2658 /* At this point we need to linearize multiple scatterlist
2659 * elements or a single shared page. Either way we need to
2660 * copy into a linear buffer exclusively owned by BPF. Then
2661 * place the buffer in the scatterlist and fixup the original
2662 * entries by removing the entries now in the linear buffer
2663 * and shifting the remaining entries. For now we do not try
2664 * to copy partial entries to avoid complexity of running out
2665 * of sg_entry slots. The downside is reading a single byte
2666 * will copy the entire sg entry.
2667 */
2668 do {
2669 copy += sk_msg_elem(msg, i)->length;
2670 sk_msg_iter_var_next(i);
2671 if (bytes_sg_total <= copy)
2672 break;
2673 } while (i != msg->sg.end);
2674 last_sge = i;
2675
2676 if (unlikely(bytes_sg_total > copy))
2677 return -EINVAL;
2678
2679 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2680 get_order(copy));
2681 if (unlikely(!page))
2682 return -ENOMEM;
2683
2684 raw = page_address(page);
2685 i = first_sge;
2686 do {
2687 sge = sk_msg_elem(msg, i);
2688 from = sg_virt(sge);
2689 len = sge->length;
2690 to = raw + poffset;
2691
2692 memcpy(to, from, len);
2693 poffset += len;
2694 sge->length = 0;
2695 put_page(sg_page(sge));
2696
2697 sk_msg_iter_var_next(i);
2698 } while (i != last_sge);
2699
2700 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2701
2702 /* To repair sg ring we need to shift entries. If we only
2703 * had a single entry though we can just replace it and
2704 * be done. Otherwise walk the ring and shift the entries.
2705 */
2706 WARN_ON_ONCE(last_sge == first_sge);
2707 shift = last_sge > first_sge ?
2708 last_sge - first_sge - 1 :
2709 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2710 if (!shift)
2711 goto out;
2712
2713 i = first_sge;
2714 sk_msg_iter_var_next(i);
2715 do {
2716 u32 move_from;
2717
2718 if (i + shift >= NR_MSG_FRAG_IDS)
2719 move_from = i + shift - NR_MSG_FRAG_IDS;
2720 else
2721 move_from = i + shift;
2722 if (move_from == msg->sg.end)
2723 break;
2724
2725 msg->sg.data[i] = msg->sg.data[move_from];
2726 msg->sg.data[move_from].length = 0;
2727 msg->sg.data[move_from].page_link = 0;
2728 msg->sg.data[move_from].offset = 0;
2729 sk_msg_iter_var_next(i);
2730 } while (1);
2731
2732 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2733 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2734 msg->sg.end - shift;
2735 out:
2736 sk_msg_reset_curr(msg);
2737 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2738 msg->data_end = msg->data + bytes;
2739 return 0;
2740 }
2741
2742 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2743 .func = bpf_msg_pull_data,
2744 .gpl_only = false,
2745 .ret_type = RET_INTEGER,
2746 .arg1_type = ARG_PTR_TO_CTX,
2747 .arg2_type = ARG_ANYTHING,
2748 .arg3_type = ARG_ANYTHING,
2749 .arg4_type = ARG_ANYTHING,
2750 };
2751
BPF_CALL_4(bpf_msg_push_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2752 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2753 u32, len, u64, flags)
2754 {
2755 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2756 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2757 u8 *raw, *to, *from;
2758 struct page *page;
2759
2760 if (unlikely(flags))
2761 return -EINVAL;
2762
2763 if (unlikely(len == 0))
2764 return 0;
2765
2766 /* First find the starting scatterlist element */
2767 i = msg->sg.start;
2768 do {
2769 offset += l;
2770 l = sk_msg_elem(msg, i)->length;
2771
2772 if (start < offset + l)
2773 break;
2774 sk_msg_iter_var_next(i);
2775 } while (i != msg->sg.end);
2776
2777 if (start >= offset + l)
2778 return -EINVAL;
2779
2780 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2781
2782 /* If no space available will fallback to copy, we need at
2783 * least one scatterlist elem available to push data into
2784 * when start aligns to the beginning of an element or two
2785 * when it falls inside an element. We handle the start equals
2786 * offset case because its the common case for inserting a
2787 * header.
2788 */
2789 if (!space || (space == 1 && start != offset))
2790 copy = msg->sg.data[i].length;
2791
2792 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2793 get_order(copy + len));
2794 if (unlikely(!page))
2795 return -ENOMEM;
2796
2797 if (copy) {
2798 int front, back;
2799
2800 raw = page_address(page);
2801
2802 psge = sk_msg_elem(msg, i);
2803 front = start - offset;
2804 back = psge->length - front;
2805 from = sg_virt(psge);
2806
2807 if (front)
2808 memcpy(raw, from, front);
2809
2810 if (back) {
2811 from += front;
2812 to = raw + front + len;
2813
2814 memcpy(to, from, back);
2815 }
2816
2817 put_page(sg_page(psge));
2818 } else if (start - offset) {
2819 psge = sk_msg_elem(msg, i);
2820 rsge = sk_msg_elem_cpy(msg, i);
2821
2822 psge->length = start - offset;
2823 rsge.length -= psge->length;
2824 rsge.offset += start;
2825
2826 sk_msg_iter_var_next(i);
2827 sg_unmark_end(psge);
2828 sg_unmark_end(&rsge);
2829 sk_msg_iter_next(msg, end);
2830 }
2831
2832 /* Slot(s) to place newly allocated data */
2833 new = i;
2834
2835 /* Shift one or two slots as needed */
2836 if (!copy) {
2837 sge = sk_msg_elem_cpy(msg, i);
2838
2839 sk_msg_iter_var_next(i);
2840 sg_unmark_end(&sge);
2841 sk_msg_iter_next(msg, end);
2842
2843 nsge = sk_msg_elem_cpy(msg, i);
2844 if (rsge.length) {
2845 sk_msg_iter_var_next(i);
2846 nnsge = sk_msg_elem_cpy(msg, i);
2847 }
2848
2849 while (i != msg->sg.end) {
2850 msg->sg.data[i] = sge;
2851 sge = nsge;
2852 sk_msg_iter_var_next(i);
2853 if (rsge.length) {
2854 nsge = nnsge;
2855 nnsge = sk_msg_elem_cpy(msg, i);
2856 } else {
2857 nsge = sk_msg_elem_cpy(msg, i);
2858 }
2859 }
2860 }
2861
2862 /* Place newly allocated data buffer */
2863 sk_mem_charge(msg->sk, len);
2864 msg->sg.size += len;
2865 __clear_bit(new, msg->sg.copy);
2866 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2867 if (rsge.length) {
2868 get_page(sg_page(&rsge));
2869 sk_msg_iter_var_next(new);
2870 msg->sg.data[new] = rsge;
2871 }
2872
2873 sk_msg_reset_curr(msg);
2874 sk_msg_compute_data_pointers(msg);
2875 return 0;
2876 }
2877
2878 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2879 .func = bpf_msg_push_data,
2880 .gpl_only = false,
2881 .ret_type = RET_INTEGER,
2882 .arg1_type = ARG_PTR_TO_CTX,
2883 .arg2_type = ARG_ANYTHING,
2884 .arg3_type = ARG_ANYTHING,
2885 .arg4_type = ARG_ANYTHING,
2886 };
2887
sk_msg_shift_left(struct sk_msg * msg,int i)2888 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2889 {
2890 int prev;
2891
2892 do {
2893 prev = i;
2894 sk_msg_iter_var_next(i);
2895 msg->sg.data[prev] = msg->sg.data[i];
2896 } while (i != msg->sg.end);
2897
2898 sk_msg_iter_prev(msg, end);
2899 }
2900
sk_msg_shift_right(struct sk_msg * msg,int i)2901 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2902 {
2903 struct scatterlist tmp, sge;
2904
2905 sk_msg_iter_next(msg, end);
2906 sge = sk_msg_elem_cpy(msg, i);
2907 sk_msg_iter_var_next(i);
2908 tmp = sk_msg_elem_cpy(msg, i);
2909
2910 while (i != msg->sg.end) {
2911 msg->sg.data[i] = sge;
2912 sk_msg_iter_var_next(i);
2913 sge = tmp;
2914 tmp = sk_msg_elem_cpy(msg, i);
2915 }
2916 }
2917
BPF_CALL_4(bpf_msg_pop_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2918 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2919 u32, len, u64, flags)
2920 {
2921 u32 i = 0, l = 0, space, offset = 0;
2922 u64 last = start + len;
2923 int pop;
2924
2925 if (unlikely(flags))
2926 return -EINVAL;
2927
2928 /* First find the starting scatterlist element */
2929 i = msg->sg.start;
2930 do {
2931 offset += l;
2932 l = sk_msg_elem(msg, i)->length;
2933
2934 if (start < offset + l)
2935 break;
2936 sk_msg_iter_var_next(i);
2937 } while (i != msg->sg.end);
2938
2939 /* Bounds checks: start and pop must be inside message */
2940 if (start >= offset + l || last >= msg->sg.size)
2941 return -EINVAL;
2942
2943 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2944
2945 pop = len;
2946 /* --------------| offset
2947 * -| start |-------- len -------|
2948 *
2949 * |----- a ----|-------- pop -------|----- b ----|
2950 * |______________________________________________| length
2951 *
2952 *
2953 * a: region at front of scatter element to save
2954 * b: region at back of scatter element to save when length > A + pop
2955 * pop: region to pop from element, same as input 'pop' here will be
2956 * decremented below per iteration.
2957 *
2958 * Two top-level cases to handle when start != offset, first B is non
2959 * zero and second B is zero corresponding to when a pop includes more
2960 * than one element.
2961 *
2962 * Then if B is non-zero AND there is no space allocate space and
2963 * compact A, B regions into page. If there is space shift ring to
2964 * the rigth free'ing the next element in ring to place B, leaving
2965 * A untouched except to reduce length.
2966 */
2967 if (start != offset) {
2968 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2969 int a = start;
2970 int b = sge->length - pop - a;
2971
2972 sk_msg_iter_var_next(i);
2973
2974 if (pop < sge->length - a) {
2975 if (space) {
2976 sge->length = a;
2977 sk_msg_shift_right(msg, i);
2978 nsge = sk_msg_elem(msg, i);
2979 get_page(sg_page(sge));
2980 sg_set_page(nsge,
2981 sg_page(sge),
2982 b, sge->offset + pop + a);
2983 } else {
2984 struct page *page, *orig;
2985 u8 *to, *from;
2986
2987 page = alloc_pages(__GFP_NOWARN |
2988 __GFP_COMP | GFP_ATOMIC,
2989 get_order(a + b));
2990 if (unlikely(!page))
2991 return -ENOMEM;
2992
2993 sge->length = a;
2994 orig = sg_page(sge);
2995 from = sg_virt(sge);
2996 to = page_address(page);
2997 memcpy(to, from, a);
2998 memcpy(to + a, from + a + pop, b);
2999 sg_set_page(sge, page, a + b, 0);
3000 put_page(orig);
3001 }
3002 pop = 0;
3003 } else if (pop >= sge->length - a) {
3004 pop -= (sge->length - a);
3005 sge->length = a;
3006 }
3007 }
3008
3009 /* From above the current layout _must_ be as follows,
3010 *
3011 * -| offset
3012 * -| start
3013 *
3014 * |---- pop ---|---------------- b ------------|
3015 * |____________________________________________| length
3016 *
3017 * Offset and start of the current msg elem are equal because in the
3018 * previous case we handled offset != start and either consumed the
3019 * entire element and advanced to the next element OR pop == 0.
3020 *
3021 * Two cases to handle here are first pop is less than the length
3022 * leaving some remainder b above. Simply adjust the element's layout
3023 * in this case. Or pop >= length of the element so that b = 0. In this
3024 * case advance to next element decrementing pop.
3025 */
3026 while (pop) {
3027 struct scatterlist *sge = sk_msg_elem(msg, i);
3028
3029 if (pop < sge->length) {
3030 sge->length -= pop;
3031 sge->offset += pop;
3032 pop = 0;
3033 } else {
3034 pop -= sge->length;
3035 sk_msg_shift_left(msg, i);
3036 }
3037 sk_msg_iter_var_next(i);
3038 }
3039
3040 sk_mem_uncharge(msg->sk, len - pop);
3041 msg->sg.size -= (len - pop);
3042 sk_msg_reset_curr(msg);
3043 sk_msg_compute_data_pointers(msg);
3044 return 0;
3045 }
3046
3047 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3048 .func = bpf_msg_pop_data,
3049 .gpl_only = false,
3050 .ret_type = RET_INTEGER,
3051 .arg1_type = ARG_PTR_TO_CTX,
3052 .arg2_type = ARG_ANYTHING,
3053 .arg3_type = ARG_ANYTHING,
3054 .arg4_type = ARG_ANYTHING,
3055 };
3056
3057 #ifdef CONFIG_CGROUP_NET_CLASSID
BPF_CALL_0(bpf_get_cgroup_classid_curr)3058 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3059 {
3060 return __task_get_classid(current);
3061 }
3062
3063 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3064 .func = bpf_get_cgroup_classid_curr,
3065 .gpl_only = false,
3066 .ret_type = RET_INTEGER,
3067 };
3068
BPF_CALL_1(bpf_skb_cgroup_classid,const struct sk_buff *,skb)3069 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3070 {
3071 struct sock *sk = skb_to_full_sk(skb);
3072
3073 if (!sk || !sk_fullsock(sk))
3074 return 0;
3075
3076 return sock_cgroup_classid(&sk->sk_cgrp_data);
3077 }
3078
3079 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3080 .func = bpf_skb_cgroup_classid,
3081 .gpl_only = false,
3082 .ret_type = RET_INTEGER,
3083 .arg1_type = ARG_PTR_TO_CTX,
3084 };
3085 #endif
3086
BPF_CALL_1(bpf_get_cgroup_classid,const struct sk_buff *,skb)3087 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3088 {
3089 return task_get_classid(skb);
3090 }
3091
3092 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3093 .func = bpf_get_cgroup_classid,
3094 .gpl_only = false,
3095 .ret_type = RET_INTEGER,
3096 .arg1_type = ARG_PTR_TO_CTX,
3097 };
3098
BPF_CALL_1(bpf_get_route_realm,const struct sk_buff *,skb)3099 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3100 {
3101 return dst_tclassid(skb);
3102 }
3103
3104 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3105 .func = bpf_get_route_realm,
3106 .gpl_only = false,
3107 .ret_type = RET_INTEGER,
3108 .arg1_type = ARG_PTR_TO_CTX,
3109 };
3110
BPF_CALL_1(bpf_get_hash_recalc,struct sk_buff *,skb)3111 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3112 {
3113 /* If skb_clear_hash() was called due to mangling, we can
3114 * trigger SW recalculation here. Later access to hash
3115 * can then use the inline skb->hash via context directly
3116 * instead of calling this helper again.
3117 */
3118 return skb_get_hash(skb);
3119 }
3120
3121 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3122 .func = bpf_get_hash_recalc,
3123 .gpl_only = false,
3124 .ret_type = RET_INTEGER,
3125 .arg1_type = ARG_PTR_TO_CTX,
3126 };
3127
BPF_CALL_1(bpf_set_hash_invalid,struct sk_buff *,skb)3128 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3129 {
3130 /* After all direct packet write, this can be used once for
3131 * triggering a lazy recalc on next skb_get_hash() invocation.
3132 */
3133 skb_clear_hash(skb);
3134 return 0;
3135 }
3136
3137 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3138 .func = bpf_set_hash_invalid,
3139 .gpl_only = false,
3140 .ret_type = RET_INTEGER,
3141 .arg1_type = ARG_PTR_TO_CTX,
3142 };
3143
BPF_CALL_2(bpf_set_hash,struct sk_buff *,skb,u32,hash)3144 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3145 {
3146 /* Set user specified hash as L4(+), so that it gets returned
3147 * on skb_get_hash() call unless BPF prog later on triggers a
3148 * skb_clear_hash().
3149 */
3150 __skb_set_sw_hash(skb, hash, true);
3151 return 0;
3152 }
3153
3154 static const struct bpf_func_proto bpf_set_hash_proto = {
3155 .func = bpf_set_hash,
3156 .gpl_only = false,
3157 .ret_type = RET_INTEGER,
3158 .arg1_type = ARG_PTR_TO_CTX,
3159 .arg2_type = ARG_ANYTHING,
3160 };
3161
BPF_CALL_3(bpf_skb_vlan_push,struct sk_buff *,skb,__be16,vlan_proto,u16,vlan_tci)3162 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3163 u16, vlan_tci)
3164 {
3165 int ret;
3166
3167 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3168 vlan_proto != htons(ETH_P_8021AD)))
3169 vlan_proto = htons(ETH_P_8021Q);
3170
3171 bpf_push_mac_rcsum(skb);
3172 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3173 bpf_pull_mac_rcsum(skb);
3174
3175 bpf_compute_data_pointers(skb);
3176 return ret;
3177 }
3178
3179 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3180 .func = bpf_skb_vlan_push,
3181 .gpl_only = false,
3182 .ret_type = RET_INTEGER,
3183 .arg1_type = ARG_PTR_TO_CTX,
3184 .arg2_type = ARG_ANYTHING,
3185 .arg3_type = ARG_ANYTHING,
3186 };
3187
BPF_CALL_1(bpf_skb_vlan_pop,struct sk_buff *,skb)3188 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3189 {
3190 int ret;
3191
3192 bpf_push_mac_rcsum(skb);
3193 ret = skb_vlan_pop(skb);
3194 bpf_pull_mac_rcsum(skb);
3195
3196 bpf_compute_data_pointers(skb);
3197 return ret;
3198 }
3199
3200 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3201 .func = bpf_skb_vlan_pop,
3202 .gpl_only = false,
3203 .ret_type = RET_INTEGER,
3204 .arg1_type = ARG_PTR_TO_CTX,
3205 };
3206
bpf_skb_generic_push(struct sk_buff * skb,u32 off,u32 len)3207 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3208 {
3209 /* Caller already did skb_cow() with len as headroom,
3210 * so no need to do it here.
3211 */
3212 skb_push(skb, len);
3213 memmove(skb->data, skb->data + len, off);
3214 memset(skb->data + off, 0, len);
3215
3216 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3217 * needed here as it does not change the skb->csum
3218 * result for checksum complete when summing over
3219 * zeroed blocks.
3220 */
3221 return 0;
3222 }
3223
bpf_skb_generic_pop(struct sk_buff * skb,u32 off,u32 len)3224 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3225 {
3226 void *old_data;
3227
3228 /* skb_ensure_writable() is not needed here, as we're
3229 * already working on an uncloned skb.
3230 */
3231 if (unlikely(!pskb_may_pull(skb, off + len)))
3232 return -ENOMEM;
3233
3234 old_data = skb->data;
3235 __skb_pull(skb, len);
3236 skb_postpull_rcsum(skb, old_data + off, len);
3237 memmove(skb->data, old_data, off);
3238
3239 return 0;
3240 }
3241
bpf_skb_net_hdr_push(struct sk_buff * skb,u32 off,u32 len)3242 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3243 {
3244 bool trans_same = skb->transport_header == skb->network_header;
3245 int ret;
3246
3247 /* There's no need for __skb_push()/__skb_pull() pair to
3248 * get to the start of the mac header as we're guaranteed
3249 * to always start from here under eBPF.
3250 */
3251 ret = bpf_skb_generic_push(skb, off, len);
3252 if (likely(!ret)) {
3253 skb->mac_header -= len;
3254 skb->network_header -= len;
3255 if (trans_same)
3256 skb->transport_header = skb->network_header;
3257 }
3258
3259 return ret;
3260 }
3261
bpf_skb_net_hdr_pop(struct sk_buff * skb,u32 off,u32 len)3262 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3263 {
3264 bool trans_same = skb->transport_header == skb->network_header;
3265 int ret;
3266
3267 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3268 ret = bpf_skb_generic_pop(skb, off, len);
3269 if (likely(!ret)) {
3270 skb->mac_header += len;
3271 skb->network_header += len;
3272 if (trans_same)
3273 skb->transport_header = skb->network_header;
3274 }
3275
3276 return ret;
3277 }
3278
bpf_skb_proto_4_to_6(struct sk_buff * skb)3279 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3280 {
3281 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3282 u32 off = skb_mac_header_len(skb);
3283 int ret;
3284
3285 ret = skb_cow(skb, len_diff);
3286 if (unlikely(ret < 0))
3287 return ret;
3288
3289 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3290 if (unlikely(ret < 0))
3291 return ret;
3292
3293 if (skb_is_gso(skb)) {
3294 struct skb_shared_info *shinfo = skb_shinfo(skb);
3295
3296 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3297 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3298 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3299 shinfo->gso_type |= SKB_GSO_TCPV6;
3300 }
3301 }
3302
3303 skb->protocol = htons(ETH_P_IPV6);
3304 skb_clear_hash(skb);
3305
3306 return 0;
3307 }
3308
bpf_skb_proto_6_to_4(struct sk_buff * skb)3309 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3310 {
3311 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3312 u32 off = skb_mac_header_len(skb);
3313 int ret;
3314
3315 ret = skb_unclone(skb, GFP_ATOMIC);
3316 if (unlikely(ret < 0))
3317 return ret;
3318
3319 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3320 if (unlikely(ret < 0))
3321 return ret;
3322
3323 if (skb_is_gso(skb)) {
3324 struct skb_shared_info *shinfo = skb_shinfo(skb);
3325
3326 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3327 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3328 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3329 shinfo->gso_type |= SKB_GSO_TCPV4;
3330 }
3331 }
3332
3333 skb->protocol = htons(ETH_P_IP);
3334 skb_clear_hash(skb);
3335
3336 return 0;
3337 }
3338
bpf_skb_proto_xlat(struct sk_buff * skb,__be16 to_proto)3339 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3340 {
3341 __be16 from_proto = skb->protocol;
3342
3343 if (from_proto == htons(ETH_P_IP) &&
3344 to_proto == htons(ETH_P_IPV6))
3345 return bpf_skb_proto_4_to_6(skb);
3346
3347 if (from_proto == htons(ETH_P_IPV6) &&
3348 to_proto == htons(ETH_P_IP))
3349 return bpf_skb_proto_6_to_4(skb);
3350
3351 return -ENOTSUPP;
3352 }
3353
BPF_CALL_3(bpf_skb_change_proto,struct sk_buff *,skb,__be16,proto,u64,flags)3354 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3355 u64, flags)
3356 {
3357 int ret;
3358
3359 if (unlikely(flags))
3360 return -EINVAL;
3361
3362 /* General idea is that this helper does the basic groundwork
3363 * needed for changing the protocol, and eBPF program fills the
3364 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3365 * and other helpers, rather than passing a raw buffer here.
3366 *
3367 * The rationale is to keep this minimal and without a need to
3368 * deal with raw packet data. F.e. even if we would pass buffers
3369 * here, the program still needs to call the bpf_lX_csum_replace()
3370 * helpers anyway. Plus, this way we keep also separation of
3371 * concerns, since f.e. bpf_skb_store_bytes() should only take
3372 * care of stores.
3373 *
3374 * Currently, additional options and extension header space are
3375 * not supported, but flags register is reserved so we can adapt
3376 * that. For offloads, we mark packet as dodgy, so that headers
3377 * need to be verified first.
3378 */
3379 ret = bpf_skb_proto_xlat(skb, proto);
3380 bpf_compute_data_pointers(skb);
3381 return ret;
3382 }
3383
3384 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3385 .func = bpf_skb_change_proto,
3386 .gpl_only = false,
3387 .ret_type = RET_INTEGER,
3388 .arg1_type = ARG_PTR_TO_CTX,
3389 .arg2_type = ARG_ANYTHING,
3390 .arg3_type = ARG_ANYTHING,
3391 };
3392
BPF_CALL_2(bpf_skb_change_type,struct sk_buff *,skb,u32,pkt_type)3393 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3394 {
3395 /* We only allow a restricted subset to be changed for now. */
3396 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3397 !skb_pkt_type_ok(pkt_type)))
3398 return -EINVAL;
3399
3400 skb->pkt_type = pkt_type;
3401 return 0;
3402 }
3403
3404 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3405 .func = bpf_skb_change_type,
3406 .gpl_only = false,
3407 .ret_type = RET_INTEGER,
3408 .arg1_type = ARG_PTR_TO_CTX,
3409 .arg2_type = ARG_ANYTHING,
3410 };
3411
bpf_skb_net_base_len(const struct sk_buff * skb)3412 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3413 {
3414 switch (skb->protocol) {
3415 case htons(ETH_P_IP):
3416 return sizeof(struct iphdr);
3417 case htons(ETH_P_IPV6):
3418 return sizeof(struct ipv6hdr);
3419 default:
3420 return ~0U;
3421 }
3422 }
3423
3424 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3425 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3426
3427 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3428 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3429
3430 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3431 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3432 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3433 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3434 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3435 BPF_F_ADJ_ROOM_ENCAP_L2( \
3436 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3437 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3438
bpf_skb_net_grow(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3439 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3440 u64 flags)
3441 {
3442 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3443 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3444 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3445 unsigned int gso_type = SKB_GSO_DODGY;
3446 int ret;
3447
3448 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3449 /* udp gso_size delineates datagrams, only allow if fixed */
3450 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3451 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3452 return -ENOTSUPP;
3453 }
3454
3455 ret = skb_cow_head(skb, len_diff);
3456 if (unlikely(ret < 0))
3457 return ret;
3458
3459 if (encap) {
3460 if (skb->protocol != htons(ETH_P_IP) &&
3461 skb->protocol != htons(ETH_P_IPV6))
3462 return -ENOTSUPP;
3463
3464 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3466 return -EINVAL;
3467
3468 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3469 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3470 return -EINVAL;
3471
3472 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3473 inner_mac_len < ETH_HLEN)
3474 return -EINVAL;
3475
3476 if (skb->encapsulation)
3477 return -EALREADY;
3478
3479 mac_len = skb->network_header - skb->mac_header;
3480 inner_net = skb->network_header;
3481 if (inner_mac_len > len_diff)
3482 return -EINVAL;
3483 inner_trans = skb->transport_header;
3484 }
3485
3486 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3487 if (unlikely(ret < 0))
3488 return ret;
3489
3490 if (encap) {
3491 skb->inner_mac_header = inner_net - inner_mac_len;
3492 skb->inner_network_header = inner_net;
3493 skb->inner_transport_header = inner_trans;
3494
3495 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3496 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3497 else
3498 skb_set_inner_protocol(skb, skb->protocol);
3499
3500 skb->encapsulation = 1;
3501 skb_set_network_header(skb, mac_len);
3502
3503 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3504 gso_type |= SKB_GSO_UDP_TUNNEL;
3505 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3506 gso_type |= SKB_GSO_GRE;
3507 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3508 gso_type |= SKB_GSO_IPXIP6;
3509 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3510 gso_type |= SKB_GSO_IPXIP4;
3511
3512 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3513 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3514 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3515 sizeof(struct ipv6hdr) :
3516 sizeof(struct iphdr);
3517
3518 skb_set_transport_header(skb, mac_len + nh_len);
3519 }
3520
3521 /* Match skb->protocol to new outer l3 protocol */
3522 if (skb->protocol == htons(ETH_P_IP) &&
3523 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3524 skb->protocol = htons(ETH_P_IPV6);
3525 else if (skb->protocol == htons(ETH_P_IPV6) &&
3526 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3527 skb->protocol = htons(ETH_P_IP);
3528 }
3529
3530 if (skb_is_gso(skb)) {
3531 struct skb_shared_info *shinfo = skb_shinfo(skb);
3532
3533 /* Header must be checked, and gso_segs recomputed. */
3534 shinfo->gso_type |= gso_type;
3535 shinfo->gso_segs = 0;
3536
3537 /* Due to header growth, MSS needs to be downgraded.
3538 * There is a BUG_ON() when segmenting the frag_list with
3539 * head_frag true, so linearize the skb after downgrading
3540 * the MSS.
3541 */
3542 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) {
3543 skb_decrease_gso_size(shinfo, len_diff);
3544 if (shinfo->frag_list)
3545 return skb_linearize(skb);
3546 }
3547 }
3548
3549 return 0;
3550 }
3551
bpf_skb_net_shrink(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3552 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3553 u64 flags)
3554 {
3555 int ret;
3556
3557 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3558 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3559 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3560 return -EINVAL;
3561
3562 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3563 /* udp gso_size delineates datagrams, only allow if fixed */
3564 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3565 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3566 return -ENOTSUPP;
3567 }
3568
3569 ret = skb_unclone(skb, GFP_ATOMIC);
3570 if (unlikely(ret < 0))
3571 return ret;
3572
3573 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3574 if (unlikely(ret < 0))
3575 return ret;
3576
3577 /* Match skb->protocol to new outer l3 protocol */
3578 if (skb->protocol == htons(ETH_P_IP) &&
3579 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3580 skb->protocol = htons(ETH_P_IPV6);
3581 else if (skb->protocol == htons(ETH_P_IPV6) &&
3582 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3583 skb->protocol = htons(ETH_P_IP);
3584
3585 if (skb_is_gso(skb)) {
3586 struct skb_shared_info *shinfo = skb_shinfo(skb);
3587
3588 /* Due to header shrink, MSS can be upgraded. */
3589 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3590 skb_increase_gso_size(shinfo, len_diff);
3591
3592 /* Header must be checked, and gso_segs recomputed. */
3593 shinfo->gso_type |= SKB_GSO_DODGY;
3594 shinfo->gso_segs = 0;
3595 }
3596
3597 return 0;
3598 }
3599
3600 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3601
BPF_CALL_4(sk_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3602 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3603 u32, mode, u64, flags)
3604 {
3605 u32 len_diff_abs = abs(len_diff);
3606 bool shrink = len_diff < 0;
3607 int ret = 0;
3608
3609 if (unlikely(flags || mode))
3610 return -EINVAL;
3611 if (unlikely(len_diff_abs > 0xfffU))
3612 return -EFAULT;
3613
3614 if (!shrink) {
3615 ret = skb_cow(skb, len_diff);
3616 if (unlikely(ret < 0))
3617 return ret;
3618 __skb_push(skb, len_diff_abs);
3619 memset(skb->data, 0, len_diff_abs);
3620 } else {
3621 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3622 return -ENOMEM;
3623 __skb_pull(skb, len_diff_abs);
3624 }
3625 if (tls_sw_has_ctx_rx(skb->sk)) {
3626 struct strp_msg *rxm = strp_msg(skb);
3627
3628 rxm->full_len += len_diff;
3629 }
3630 return ret;
3631 }
3632
3633 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3634 .func = sk_skb_adjust_room,
3635 .gpl_only = false,
3636 .ret_type = RET_INTEGER,
3637 .arg1_type = ARG_PTR_TO_CTX,
3638 .arg2_type = ARG_ANYTHING,
3639 .arg3_type = ARG_ANYTHING,
3640 .arg4_type = ARG_ANYTHING,
3641 };
3642
BPF_CALL_4(bpf_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3643 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3644 u32, mode, u64, flags)
3645 {
3646 u32 len_cur, len_diff_abs = abs(len_diff);
3647 u32 len_min = bpf_skb_net_base_len(skb);
3648 u32 len_max = BPF_SKB_MAX_LEN;
3649 __be16 proto = skb->protocol;
3650 bool shrink = len_diff < 0;
3651 u32 off;
3652 int ret;
3653
3654 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3655 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3656 return -EINVAL;
3657 if (unlikely(len_diff_abs > 0xfffU))
3658 return -EFAULT;
3659 if (unlikely(proto != htons(ETH_P_IP) &&
3660 proto != htons(ETH_P_IPV6)))
3661 return -ENOTSUPP;
3662
3663 off = skb_mac_header_len(skb);
3664 switch (mode) {
3665 case BPF_ADJ_ROOM_NET:
3666 off += bpf_skb_net_base_len(skb);
3667 break;
3668 case BPF_ADJ_ROOM_MAC:
3669 break;
3670 default:
3671 return -ENOTSUPP;
3672 }
3673
3674 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3675 if (!shrink)
3676 return -EINVAL;
3677
3678 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3679 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3680 len_min = sizeof(struct iphdr);
3681 break;
3682 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3683 len_min = sizeof(struct ipv6hdr);
3684 break;
3685 default:
3686 return -EINVAL;
3687 }
3688 }
3689
3690 len_cur = skb->len - skb_network_offset(skb);
3691 if ((shrink && (len_diff_abs >= len_cur ||
3692 len_cur - len_diff_abs < len_min)) ||
3693 (!shrink && (skb->len + len_diff_abs > len_max &&
3694 !skb_is_gso(skb))))
3695 return -ENOTSUPP;
3696
3697 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3698 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3699 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3700 __skb_reset_checksum_unnecessary(skb);
3701
3702 bpf_compute_data_pointers(skb);
3703 return ret;
3704 }
3705
3706 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3707 .func = bpf_skb_adjust_room,
3708 .gpl_only = false,
3709 .ret_type = RET_INTEGER,
3710 .arg1_type = ARG_PTR_TO_CTX,
3711 .arg2_type = ARG_ANYTHING,
3712 .arg3_type = ARG_ANYTHING,
3713 .arg4_type = ARG_ANYTHING,
3714 };
3715
__bpf_skb_min_len(const struct sk_buff * skb)3716 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3717 {
3718 u32 min_len = skb_network_offset(skb);
3719
3720 if (skb_transport_header_was_set(skb))
3721 min_len = skb_transport_offset(skb);
3722 if (skb->ip_summed == CHECKSUM_PARTIAL)
3723 min_len = skb_checksum_start_offset(skb) +
3724 skb->csum_offset + sizeof(__sum16);
3725 return min_len;
3726 }
3727
bpf_skb_grow_rcsum(struct sk_buff * skb,unsigned int new_len)3728 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3729 {
3730 unsigned int old_len = skb->len;
3731 int ret;
3732
3733 ret = __skb_grow_rcsum(skb, new_len);
3734 if (!ret)
3735 memset(skb->data + old_len, 0, new_len - old_len);
3736 return ret;
3737 }
3738
bpf_skb_trim_rcsum(struct sk_buff * skb,unsigned int new_len)3739 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3740 {
3741 return __skb_trim_rcsum(skb, new_len);
3742 }
3743
__bpf_skb_change_tail(struct sk_buff * skb,u32 new_len,u64 flags)3744 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3745 u64 flags)
3746 {
3747 u32 max_len = BPF_SKB_MAX_LEN;
3748 u32 min_len = __bpf_skb_min_len(skb);
3749 int ret;
3750
3751 if (unlikely(flags || new_len > max_len || new_len < min_len))
3752 return -EINVAL;
3753 if (skb->encapsulation)
3754 return -ENOTSUPP;
3755
3756 /* The basic idea of this helper is that it's performing the
3757 * needed work to either grow or trim an skb, and eBPF program
3758 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3759 * bpf_lX_csum_replace() and others rather than passing a raw
3760 * buffer here. This one is a slow path helper and intended
3761 * for replies with control messages.
3762 *
3763 * Like in bpf_skb_change_proto(), we want to keep this rather
3764 * minimal and without protocol specifics so that we are able
3765 * to separate concerns as in bpf_skb_store_bytes() should only
3766 * be the one responsible for writing buffers.
3767 *
3768 * It's really expected to be a slow path operation here for
3769 * control message replies, so we're implicitly linearizing,
3770 * uncloning and drop offloads from the skb by this.
3771 */
3772 ret = __bpf_try_make_writable(skb, skb->len);
3773 if (!ret) {
3774 if (new_len > skb->len)
3775 ret = bpf_skb_grow_rcsum(skb, new_len);
3776 else if (new_len < skb->len)
3777 ret = bpf_skb_trim_rcsum(skb, new_len);
3778 if (!ret && skb_is_gso(skb))
3779 skb_gso_reset(skb);
3780 }
3781 return ret;
3782 }
3783
BPF_CALL_3(bpf_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3784 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3785 u64, flags)
3786 {
3787 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3788
3789 bpf_compute_data_pointers(skb);
3790 return ret;
3791 }
3792
3793 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3794 .func = bpf_skb_change_tail,
3795 .gpl_only = false,
3796 .ret_type = RET_INTEGER,
3797 .arg1_type = ARG_PTR_TO_CTX,
3798 .arg2_type = ARG_ANYTHING,
3799 .arg3_type = ARG_ANYTHING,
3800 };
3801
BPF_CALL_3(sk_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3802 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3803 u64, flags)
3804 {
3805 return __bpf_skb_change_tail(skb, new_len, flags);
3806 }
3807
3808 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3809 .func = sk_skb_change_tail,
3810 .gpl_only = false,
3811 .ret_type = RET_INTEGER,
3812 .arg1_type = ARG_PTR_TO_CTX,
3813 .arg2_type = ARG_ANYTHING,
3814 .arg3_type = ARG_ANYTHING,
3815 };
3816
__bpf_skb_change_head(struct sk_buff * skb,u32 head_room,u64 flags)3817 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3818 u64 flags)
3819 {
3820 u32 max_len = BPF_SKB_MAX_LEN;
3821 u32 new_len = skb->len + head_room;
3822 int ret;
3823
3824 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3825 new_len < skb->len))
3826 return -EINVAL;
3827
3828 ret = skb_cow(skb, head_room);
3829 if (likely(!ret)) {
3830 /* Idea for this helper is that we currently only
3831 * allow to expand on mac header. This means that
3832 * skb->protocol network header, etc, stay as is.
3833 * Compared to bpf_skb_change_tail(), we're more
3834 * flexible due to not needing to linearize or
3835 * reset GSO. Intention for this helper is to be
3836 * used by an L3 skb that needs to push mac header
3837 * for redirection into L2 device.
3838 */
3839 __skb_push(skb, head_room);
3840 memset(skb->data, 0, head_room);
3841 skb_reset_mac_header(skb);
3842 skb_reset_mac_len(skb);
3843 }
3844
3845 return ret;
3846 }
3847
BPF_CALL_3(bpf_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3848 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3849 u64, flags)
3850 {
3851 int ret = __bpf_skb_change_head(skb, head_room, flags);
3852
3853 bpf_compute_data_pointers(skb);
3854 return ret;
3855 }
3856
3857 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3858 .func = bpf_skb_change_head,
3859 .gpl_only = false,
3860 .ret_type = RET_INTEGER,
3861 .arg1_type = ARG_PTR_TO_CTX,
3862 .arg2_type = ARG_ANYTHING,
3863 .arg3_type = ARG_ANYTHING,
3864 };
3865
BPF_CALL_3(sk_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3866 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3867 u64, flags)
3868 {
3869 return __bpf_skb_change_head(skb, head_room, flags);
3870 }
3871
3872 static const struct bpf_func_proto sk_skb_change_head_proto = {
3873 .func = sk_skb_change_head,
3874 .gpl_only = false,
3875 .ret_type = RET_INTEGER,
3876 .arg1_type = ARG_PTR_TO_CTX,
3877 .arg2_type = ARG_ANYTHING,
3878 .arg3_type = ARG_ANYTHING,
3879 };
3880
BPF_CALL_1(bpf_xdp_get_buff_len,struct xdp_buff *,xdp)3881 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3882 {
3883 return xdp_get_buff_len(xdp);
3884 }
3885
3886 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3887 .func = bpf_xdp_get_buff_len,
3888 .gpl_only = false,
3889 .ret_type = RET_INTEGER,
3890 .arg1_type = ARG_PTR_TO_CTX,
3891 };
3892
3893 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3894
3895 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3896 .func = bpf_xdp_get_buff_len,
3897 .gpl_only = false,
3898 .arg1_type = ARG_PTR_TO_BTF_ID,
3899 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3900 };
3901
xdp_get_metalen(const struct xdp_buff * xdp)3902 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3903 {
3904 return xdp_data_meta_unsupported(xdp) ? 0 :
3905 xdp->data - xdp->data_meta;
3906 }
3907
BPF_CALL_2(bpf_xdp_adjust_head,struct xdp_buff *,xdp,int,offset)3908 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3909 {
3910 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3911 unsigned long metalen = xdp_get_metalen(xdp);
3912 void *data_start = xdp_frame_end + metalen;
3913 void *data = xdp->data + offset;
3914
3915 if (unlikely(data < data_start ||
3916 data > xdp->data_end - ETH_HLEN))
3917 return -EINVAL;
3918
3919 if (metalen)
3920 memmove(xdp->data_meta + offset,
3921 xdp->data_meta, metalen);
3922 xdp->data_meta += offset;
3923 xdp->data = data;
3924
3925 return 0;
3926 }
3927
3928 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3929 .func = bpf_xdp_adjust_head,
3930 .gpl_only = false,
3931 .ret_type = RET_INTEGER,
3932 .arg1_type = ARG_PTR_TO_CTX,
3933 .arg2_type = ARG_ANYTHING,
3934 };
3935
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)3936 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3937 void *buf, unsigned long len, bool flush)
3938 {
3939 unsigned long ptr_len, ptr_off = 0;
3940 skb_frag_t *next_frag, *end_frag;
3941 struct skb_shared_info *sinfo;
3942 void *src, *dst;
3943 u8 *ptr_buf;
3944
3945 if (likely(xdp->data_end - xdp->data >= off + len)) {
3946 src = flush ? buf : xdp->data + off;
3947 dst = flush ? xdp->data + off : buf;
3948 memcpy(dst, src, len);
3949 return;
3950 }
3951
3952 sinfo = xdp_get_shared_info_from_buff(xdp);
3953 end_frag = &sinfo->frags[sinfo->nr_frags];
3954 next_frag = &sinfo->frags[0];
3955
3956 ptr_len = xdp->data_end - xdp->data;
3957 ptr_buf = xdp->data;
3958
3959 while (true) {
3960 if (off < ptr_off + ptr_len) {
3961 unsigned long copy_off = off - ptr_off;
3962 unsigned long copy_len = min(len, ptr_len - copy_off);
3963
3964 src = flush ? buf : ptr_buf + copy_off;
3965 dst = flush ? ptr_buf + copy_off : buf;
3966 memcpy(dst, src, copy_len);
3967
3968 off += copy_len;
3969 len -= copy_len;
3970 buf += copy_len;
3971 }
3972
3973 if (!len || next_frag == end_frag)
3974 break;
3975
3976 ptr_off += ptr_len;
3977 ptr_buf = skb_frag_address(next_frag);
3978 ptr_len = skb_frag_size(next_frag);
3979 next_frag++;
3980 }
3981 }
3982
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)3983 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3984 {
3985 u32 size = xdp->data_end - xdp->data;
3986 struct skb_shared_info *sinfo;
3987 void *addr = xdp->data;
3988 int i;
3989
3990 if (unlikely(offset > 0xffff || len > 0xffff))
3991 return ERR_PTR(-EFAULT);
3992
3993 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3994 return ERR_PTR(-EINVAL);
3995
3996 if (likely(offset < size)) /* linear area */
3997 goto out;
3998
3999 sinfo = xdp_get_shared_info_from_buff(xdp);
4000 offset -= size;
4001 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
4002 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
4003
4004 if (offset < frag_size) {
4005 addr = skb_frag_address(&sinfo->frags[i]);
4006 size = frag_size;
4007 break;
4008 }
4009 offset -= frag_size;
4010 }
4011 out:
4012 return offset + len <= size ? addr + offset : NULL;
4013 }
4014
BPF_CALL_4(bpf_xdp_load_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4015 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4016 void *, buf, u32, len)
4017 {
4018 void *ptr;
4019
4020 ptr = bpf_xdp_pointer(xdp, offset, len);
4021 if (IS_ERR(ptr))
4022 return PTR_ERR(ptr);
4023
4024 if (!ptr)
4025 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4026 else
4027 memcpy(buf, ptr, len);
4028
4029 return 0;
4030 }
4031
4032 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4033 .func = bpf_xdp_load_bytes,
4034 .gpl_only = false,
4035 .ret_type = RET_INTEGER,
4036 .arg1_type = ARG_PTR_TO_CTX,
4037 .arg2_type = ARG_ANYTHING,
4038 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4039 .arg4_type = ARG_CONST_SIZE,
4040 };
4041
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4042 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4043 {
4044 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4045 }
4046
BPF_CALL_4(bpf_xdp_store_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4047 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4048 void *, buf, u32, len)
4049 {
4050 void *ptr;
4051
4052 ptr = bpf_xdp_pointer(xdp, offset, len);
4053 if (IS_ERR(ptr))
4054 return PTR_ERR(ptr);
4055
4056 if (!ptr)
4057 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4058 else
4059 memcpy(ptr, buf, len);
4060
4061 return 0;
4062 }
4063
4064 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4065 .func = bpf_xdp_store_bytes,
4066 .gpl_only = false,
4067 .ret_type = RET_INTEGER,
4068 .arg1_type = ARG_PTR_TO_CTX,
4069 .arg2_type = ARG_ANYTHING,
4070 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4071 .arg4_type = ARG_CONST_SIZE,
4072 };
4073
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4074 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4075 {
4076 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4077 }
4078
bpf_xdp_frags_increase_tail(struct xdp_buff * xdp,int offset)4079 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4080 {
4081 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4082 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4083 struct xdp_rxq_info *rxq = xdp->rxq;
4084 unsigned int tailroom;
4085
4086 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4087 return -EOPNOTSUPP;
4088
4089 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4090 if (unlikely(offset > tailroom))
4091 return -EINVAL;
4092
4093 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4094 skb_frag_size_add(frag, offset);
4095 sinfo->xdp_frags_size += offset;
4096 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
4097 xsk_buff_get_tail(xdp)->data_end += offset;
4098
4099 return 0;
4100 }
4101
bpf_xdp_shrink_data_zc(struct xdp_buff * xdp,int shrink,struct xdp_mem_info * mem_info,bool release)4102 static void bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink,
4103 struct xdp_mem_info *mem_info, bool release)
4104 {
4105 struct xdp_buff *zc_frag = xsk_buff_get_tail(xdp);
4106
4107 if (release) {
4108 xsk_buff_del_tail(zc_frag);
4109 __xdp_return(NULL, mem_info, false, zc_frag);
4110 } else {
4111 zc_frag->data_end -= shrink;
4112 }
4113 }
4114
bpf_xdp_shrink_data(struct xdp_buff * xdp,skb_frag_t * frag,int shrink)4115 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag,
4116 int shrink)
4117 {
4118 struct xdp_mem_info *mem_info = &xdp->rxq->mem;
4119 bool release = skb_frag_size(frag) == shrink;
4120
4121 if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
4122 bpf_xdp_shrink_data_zc(xdp, shrink, mem_info, release);
4123 goto out;
4124 }
4125
4126 if (release) {
4127 struct page *page = skb_frag_page(frag);
4128
4129 __xdp_return(page_address(page), mem_info, false, NULL);
4130 }
4131
4132 out:
4133 return release;
4134 }
4135
bpf_xdp_frags_shrink_tail(struct xdp_buff * xdp,int offset)4136 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4137 {
4138 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4139 int i, n_frags_free = 0, len_free = 0;
4140
4141 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4142 return -EINVAL;
4143
4144 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4145 skb_frag_t *frag = &sinfo->frags[i];
4146 int shrink = min_t(int, offset, skb_frag_size(frag));
4147
4148 len_free += shrink;
4149 offset -= shrink;
4150 if (bpf_xdp_shrink_data(xdp, frag, shrink)) {
4151 n_frags_free++;
4152 } else {
4153 skb_frag_size_sub(frag, shrink);
4154 break;
4155 }
4156 }
4157 sinfo->nr_frags -= n_frags_free;
4158 sinfo->xdp_frags_size -= len_free;
4159
4160 if (unlikely(!sinfo->nr_frags)) {
4161 xdp_buff_clear_frags_flag(xdp);
4162 xdp->data_end -= offset;
4163 }
4164
4165 return 0;
4166 }
4167
BPF_CALL_2(bpf_xdp_adjust_tail,struct xdp_buff *,xdp,int,offset)4168 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4169 {
4170 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4171 void *data_end = xdp->data_end + offset;
4172
4173 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4174 if (offset < 0)
4175 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4176
4177 return bpf_xdp_frags_increase_tail(xdp, offset);
4178 }
4179
4180 /* Notice that xdp_data_hard_end have reserved some tailroom */
4181 if (unlikely(data_end > data_hard_end))
4182 return -EINVAL;
4183
4184 if (unlikely(data_end < xdp->data + ETH_HLEN))
4185 return -EINVAL;
4186
4187 /* Clear memory area on grow, can contain uninit kernel memory */
4188 if (offset > 0)
4189 memset(xdp->data_end, 0, offset);
4190
4191 xdp->data_end = data_end;
4192
4193 return 0;
4194 }
4195
4196 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4197 .func = bpf_xdp_adjust_tail,
4198 .gpl_only = false,
4199 .ret_type = RET_INTEGER,
4200 .arg1_type = ARG_PTR_TO_CTX,
4201 .arg2_type = ARG_ANYTHING,
4202 };
4203
BPF_CALL_2(bpf_xdp_adjust_meta,struct xdp_buff *,xdp,int,offset)4204 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4205 {
4206 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4207 void *meta = xdp->data_meta + offset;
4208 unsigned long metalen = xdp->data - meta;
4209
4210 if (xdp_data_meta_unsupported(xdp))
4211 return -ENOTSUPP;
4212 if (unlikely(meta < xdp_frame_end ||
4213 meta > xdp->data))
4214 return -EINVAL;
4215 if (unlikely(xdp_metalen_invalid(metalen)))
4216 return -EACCES;
4217
4218 xdp->data_meta = meta;
4219
4220 return 0;
4221 }
4222
4223 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4224 .func = bpf_xdp_adjust_meta,
4225 .gpl_only = false,
4226 .ret_type = RET_INTEGER,
4227 .arg1_type = ARG_PTR_TO_CTX,
4228 .arg2_type = ARG_ANYTHING,
4229 };
4230
4231 /**
4232 * DOC: xdp redirect
4233 *
4234 * XDP_REDIRECT works by a three-step process, implemented in the functions
4235 * below:
4236 *
4237 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4238 * of the redirect and store it (along with some other metadata) in a per-CPU
4239 * struct bpf_redirect_info.
4240 *
4241 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4242 * call xdp_do_redirect() which will use the information in struct
4243 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4244 * bulk queue structure.
4245 *
4246 * 3. Before exiting its NAPI poll loop, the driver will call
4247 * xdp_do_flush(), which will flush all the different bulk queues,
4248 * thus completing the redirect. Note that xdp_do_flush() must be
4249 * called before napi_complete_done() in the driver, as the
4250 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4251 * through to the xdp_do_flush() call for RCU protection of all
4252 * in-kernel data structures.
4253 */
4254 /*
4255 * Pointers to the map entries will be kept around for this whole sequence of
4256 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4257 * the core code; instead, the RCU protection relies on everything happening
4258 * inside a single NAPI poll sequence, which means it's between a pair of calls
4259 * to local_bh_disable()/local_bh_enable().
4260 *
4261 * The map entries are marked as __rcu and the map code makes sure to
4262 * dereference those pointers with rcu_dereference_check() in a way that works
4263 * for both sections that to hold an rcu_read_lock() and sections that are
4264 * called from NAPI without a separate rcu_read_lock(). The code below does not
4265 * use RCU annotations, but relies on those in the map code.
4266 */
xdp_do_flush(void)4267 void xdp_do_flush(void)
4268 {
4269 __dev_flush();
4270 __cpu_map_flush();
4271 __xsk_map_flush();
4272 }
4273 EXPORT_SYMBOL_GPL(xdp_do_flush);
4274
bpf_clear_redirect_map(struct bpf_map * map)4275 void bpf_clear_redirect_map(struct bpf_map *map)
4276 {
4277 struct bpf_redirect_info *ri;
4278 int cpu;
4279
4280 for_each_possible_cpu(cpu) {
4281 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4282 /* Avoid polluting remote cacheline due to writes if
4283 * not needed. Once we pass this test, we need the
4284 * cmpxchg() to make sure it hasn't been changed in
4285 * the meantime by remote CPU.
4286 */
4287 if (unlikely(READ_ONCE(ri->map) == map))
4288 cmpxchg(&ri->map, map, NULL);
4289 }
4290 }
4291
4292 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4293 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4294
xdp_master_redirect(struct xdp_buff * xdp)4295 u32 xdp_master_redirect(struct xdp_buff *xdp)
4296 {
4297 struct net_device *master, *slave;
4298 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4299
4300 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4301 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4302 if (slave && slave != xdp->rxq->dev) {
4303 /* The target device is different from the receiving device, so
4304 * redirect it to the new device.
4305 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4306 * drivers to unmap the packet from their rx ring.
4307 */
4308 ri->tgt_index = slave->ifindex;
4309 ri->map_id = INT_MAX;
4310 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4311 return XDP_REDIRECT;
4312 }
4313 return XDP_TX;
4314 }
4315 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4316
__xdp_do_redirect_xsk(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4317 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4318 struct net_device *dev,
4319 struct xdp_buff *xdp,
4320 struct bpf_prog *xdp_prog)
4321 {
4322 enum bpf_map_type map_type = ri->map_type;
4323 void *fwd = ri->tgt_value;
4324 u32 map_id = ri->map_id;
4325 int err;
4326
4327 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4328 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4329
4330 err = __xsk_map_redirect(fwd, xdp);
4331 if (unlikely(err))
4332 goto err;
4333
4334 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4335 return 0;
4336 err:
4337 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4338 return err;
4339 }
4340
__xdp_do_redirect_frame(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4341 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4342 struct net_device *dev,
4343 struct xdp_frame *xdpf,
4344 struct bpf_prog *xdp_prog)
4345 {
4346 enum bpf_map_type map_type = ri->map_type;
4347 void *fwd = ri->tgt_value;
4348 u32 map_id = ri->map_id;
4349 u32 flags = ri->flags;
4350 struct bpf_map *map;
4351 int err;
4352
4353 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4354 ri->flags = 0;
4355 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4356
4357 if (unlikely(!xdpf)) {
4358 err = -EOVERFLOW;
4359 goto err;
4360 }
4361
4362 switch (map_type) {
4363 case BPF_MAP_TYPE_DEVMAP:
4364 fallthrough;
4365 case BPF_MAP_TYPE_DEVMAP_HASH:
4366 if (unlikely(flags & BPF_F_BROADCAST)) {
4367 map = READ_ONCE(ri->map);
4368
4369 /* The map pointer is cleared when the map is being torn
4370 * down by bpf_clear_redirect_map()
4371 */
4372 if (unlikely(!map)) {
4373 err = -ENOENT;
4374 break;
4375 }
4376
4377 WRITE_ONCE(ri->map, NULL);
4378 err = dev_map_enqueue_multi(xdpf, dev, map,
4379 flags & BPF_F_EXCLUDE_INGRESS);
4380 } else {
4381 err = dev_map_enqueue(fwd, xdpf, dev);
4382 }
4383 break;
4384 case BPF_MAP_TYPE_CPUMAP:
4385 err = cpu_map_enqueue(fwd, xdpf, dev);
4386 break;
4387 case BPF_MAP_TYPE_UNSPEC:
4388 if (map_id == INT_MAX) {
4389 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4390 if (unlikely(!fwd)) {
4391 err = -EINVAL;
4392 break;
4393 }
4394 err = dev_xdp_enqueue(fwd, xdpf, dev);
4395 break;
4396 }
4397 fallthrough;
4398 default:
4399 err = -EBADRQC;
4400 }
4401
4402 if (unlikely(err))
4403 goto err;
4404
4405 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4406 return 0;
4407 err:
4408 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4409 return err;
4410 }
4411
xdp_do_redirect(struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4412 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4413 struct bpf_prog *xdp_prog)
4414 {
4415 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4416 enum bpf_map_type map_type = ri->map_type;
4417
4418 if (map_type == BPF_MAP_TYPE_XSKMAP)
4419 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4420
4421 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4422 xdp_prog);
4423 }
4424 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4425
xdp_do_redirect_frame(struct net_device * dev,struct xdp_buff * xdp,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4426 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4427 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4428 {
4429 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4430 enum bpf_map_type map_type = ri->map_type;
4431
4432 if (map_type == BPF_MAP_TYPE_XSKMAP)
4433 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4434
4435 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4436 }
4437 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4438
xdp_do_generic_redirect_map(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog,void * fwd,enum bpf_map_type map_type,u32 map_id,u32 flags)4439 static int xdp_do_generic_redirect_map(struct net_device *dev,
4440 struct sk_buff *skb,
4441 struct xdp_buff *xdp,
4442 struct bpf_prog *xdp_prog, void *fwd,
4443 enum bpf_map_type map_type, u32 map_id,
4444 u32 flags)
4445 {
4446 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4447 struct bpf_map *map;
4448 int err;
4449
4450 switch (map_type) {
4451 case BPF_MAP_TYPE_DEVMAP:
4452 fallthrough;
4453 case BPF_MAP_TYPE_DEVMAP_HASH:
4454 if (unlikely(flags & BPF_F_BROADCAST)) {
4455 map = READ_ONCE(ri->map);
4456
4457 /* The map pointer is cleared when the map is being torn
4458 * down by bpf_clear_redirect_map()
4459 */
4460 if (unlikely(!map)) {
4461 err = -ENOENT;
4462 break;
4463 }
4464
4465 WRITE_ONCE(ri->map, NULL);
4466 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4467 flags & BPF_F_EXCLUDE_INGRESS);
4468 } else {
4469 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4470 }
4471 if (unlikely(err))
4472 goto err;
4473 break;
4474 case BPF_MAP_TYPE_XSKMAP:
4475 err = xsk_generic_rcv(fwd, xdp);
4476 if (err)
4477 goto err;
4478 consume_skb(skb);
4479 break;
4480 case BPF_MAP_TYPE_CPUMAP:
4481 err = cpu_map_generic_redirect(fwd, skb);
4482 if (unlikely(err))
4483 goto err;
4484 break;
4485 default:
4486 err = -EBADRQC;
4487 goto err;
4488 }
4489
4490 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4491 return 0;
4492 err:
4493 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4494 return err;
4495 }
4496
xdp_do_generic_redirect(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4497 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4498 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4499 {
4500 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4501 enum bpf_map_type map_type = ri->map_type;
4502 void *fwd = ri->tgt_value;
4503 u32 map_id = ri->map_id;
4504 u32 flags = ri->flags;
4505 int err;
4506
4507 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4508 ri->flags = 0;
4509 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4510
4511 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4512 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4513 if (unlikely(!fwd)) {
4514 err = -EINVAL;
4515 goto err;
4516 }
4517
4518 err = xdp_ok_fwd_dev(fwd, skb->len);
4519 if (unlikely(err))
4520 goto err;
4521
4522 skb->dev = fwd;
4523 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4524 generic_xdp_tx(skb, xdp_prog);
4525 return 0;
4526 }
4527
4528 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4529 err:
4530 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4531 return err;
4532 }
4533
BPF_CALL_2(bpf_xdp_redirect,u32,ifindex,u64,flags)4534 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4535 {
4536 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4537
4538 if (unlikely(flags))
4539 return XDP_ABORTED;
4540
4541 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4542 * by map_idr) is used for ifindex based XDP redirect.
4543 */
4544 ri->tgt_index = ifindex;
4545 ri->map_id = INT_MAX;
4546 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4547
4548 return XDP_REDIRECT;
4549 }
4550
4551 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4552 .func = bpf_xdp_redirect,
4553 .gpl_only = false,
4554 .ret_type = RET_INTEGER,
4555 .arg1_type = ARG_ANYTHING,
4556 .arg2_type = ARG_ANYTHING,
4557 };
4558
BPF_CALL_3(bpf_xdp_redirect_map,struct bpf_map *,map,u64,key,u64,flags)4559 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4560 u64, flags)
4561 {
4562 return map->ops->map_redirect(map, key, flags);
4563 }
4564
4565 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4566 .func = bpf_xdp_redirect_map,
4567 .gpl_only = false,
4568 .ret_type = RET_INTEGER,
4569 .arg1_type = ARG_CONST_MAP_PTR,
4570 .arg2_type = ARG_ANYTHING,
4571 .arg3_type = ARG_ANYTHING,
4572 };
4573
bpf_skb_copy(void * dst_buff,const void * skb,unsigned long off,unsigned long len)4574 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4575 unsigned long off, unsigned long len)
4576 {
4577 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4578
4579 if (unlikely(!ptr))
4580 return len;
4581 if (ptr != dst_buff)
4582 memcpy(dst_buff, ptr, len);
4583
4584 return 0;
4585 }
4586
BPF_CALL_5(bpf_skb_event_output,struct sk_buff *,skb,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4587 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4588 u64, flags, void *, meta, u64, meta_size)
4589 {
4590 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4591
4592 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4593 return -EINVAL;
4594 if (unlikely(!skb || skb_size > skb->len))
4595 return -EFAULT;
4596
4597 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4598 bpf_skb_copy);
4599 }
4600
4601 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4602 .func = bpf_skb_event_output,
4603 .gpl_only = true,
4604 .ret_type = RET_INTEGER,
4605 .arg1_type = ARG_PTR_TO_CTX,
4606 .arg2_type = ARG_CONST_MAP_PTR,
4607 .arg3_type = ARG_ANYTHING,
4608 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4609 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4610 };
4611
4612 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4613
4614 const struct bpf_func_proto bpf_skb_output_proto = {
4615 .func = bpf_skb_event_output,
4616 .gpl_only = true,
4617 .ret_type = RET_INTEGER,
4618 .arg1_type = ARG_PTR_TO_BTF_ID,
4619 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4620 .arg2_type = ARG_CONST_MAP_PTR,
4621 .arg3_type = ARG_ANYTHING,
4622 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4623 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4624 };
4625
bpf_tunnel_key_af(u64 flags)4626 static unsigned short bpf_tunnel_key_af(u64 flags)
4627 {
4628 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4629 }
4630
BPF_CALL_4(bpf_skb_get_tunnel_key,struct sk_buff *,skb,struct bpf_tunnel_key *,to,u32,size,u64,flags)4631 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4632 u32, size, u64, flags)
4633 {
4634 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4635 u8 compat[sizeof(struct bpf_tunnel_key)];
4636 void *to_orig = to;
4637 int err;
4638
4639 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4640 BPF_F_TUNINFO_FLAGS)))) {
4641 err = -EINVAL;
4642 goto err_clear;
4643 }
4644 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4645 err = -EPROTO;
4646 goto err_clear;
4647 }
4648 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4649 err = -EINVAL;
4650 switch (size) {
4651 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4652 case offsetof(struct bpf_tunnel_key, tunnel_label):
4653 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4654 goto set_compat;
4655 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4656 /* Fixup deprecated structure layouts here, so we have
4657 * a common path later on.
4658 */
4659 if (ip_tunnel_info_af(info) != AF_INET)
4660 goto err_clear;
4661 set_compat:
4662 to = (struct bpf_tunnel_key *)compat;
4663 break;
4664 default:
4665 goto err_clear;
4666 }
4667 }
4668
4669 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4670 to->tunnel_tos = info->key.tos;
4671 to->tunnel_ttl = info->key.ttl;
4672 if (flags & BPF_F_TUNINFO_FLAGS)
4673 to->tunnel_flags = info->key.tun_flags;
4674 else
4675 to->tunnel_ext = 0;
4676
4677 if (flags & BPF_F_TUNINFO_IPV6) {
4678 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4679 sizeof(to->remote_ipv6));
4680 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4681 sizeof(to->local_ipv6));
4682 to->tunnel_label = be32_to_cpu(info->key.label);
4683 } else {
4684 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4685 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4686 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4687 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4688 to->tunnel_label = 0;
4689 }
4690
4691 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4692 memcpy(to_orig, to, size);
4693
4694 return 0;
4695 err_clear:
4696 memset(to_orig, 0, size);
4697 return err;
4698 }
4699
4700 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4701 .func = bpf_skb_get_tunnel_key,
4702 .gpl_only = false,
4703 .ret_type = RET_INTEGER,
4704 .arg1_type = ARG_PTR_TO_CTX,
4705 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4706 .arg3_type = ARG_CONST_SIZE,
4707 .arg4_type = ARG_ANYTHING,
4708 };
4709
BPF_CALL_3(bpf_skb_get_tunnel_opt,struct sk_buff *,skb,u8 *,to,u32,size)4710 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4711 {
4712 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4713 int err;
4714
4715 if (unlikely(!info ||
4716 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4717 err = -ENOENT;
4718 goto err_clear;
4719 }
4720 if (unlikely(size < info->options_len)) {
4721 err = -ENOMEM;
4722 goto err_clear;
4723 }
4724
4725 ip_tunnel_info_opts_get(to, info);
4726 if (size > info->options_len)
4727 memset(to + info->options_len, 0, size - info->options_len);
4728
4729 return info->options_len;
4730 err_clear:
4731 memset(to, 0, size);
4732 return err;
4733 }
4734
4735 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4736 .func = bpf_skb_get_tunnel_opt,
4737 .gpl_only = false,
4738 .ret_type = RET_INTEGER,
4739 .arg1_type = ARG_PTR_TO_CTX,
4740 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4741 .arg3_type = ARG_CONST_SIZE,
4742 };
4743
4744 static struct metadata_dst __percpu *md_dst;
4745
BPF_CALL_4(bpf_skb_set_tunnel_key,struct sk_buff *,skb,const struct bpf_tunnel_key *,from,u32,size,u64,flags)4746 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4747 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4748 {
4749 struct metadata_dst *md = this_cpu_ptr(md_dst);
4750 u8 compat[sizeof(struct bpf_tunnel_key)];
4751 struct ip_tunnel_info *info;
4752
4753 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4754 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4755 BPF_F_NO_TUNNEL_KEY)))
4756 return -EINVAL;
4757 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4758 switch (size) {
4759 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4760 case offsetof(struct bpf_tunnel_key, tunnel_label):
4761 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4762 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4763 /* Fixup deprecated structure layouts here, so we have
4764 * a common path later on.
4765 */
4766 memcpy(compat, from, size);
4767 memset(compat + size, 0, sizeof(compat) - size);
4768 from = (const struct bpf_tunnel_key *) compat;
4769 break;
4770 default:
4771 return -EINVAL;
4772 }
4773 }
4774 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4775 from->tunnel_ext))
4776 return -EINVAL;
4777
4778 skb_dst_drop(skb);
4779 dst_hold((struct dst_entry *) md);
4780 skb_dst_set(skb, (struct dst_entry *) md);
4781
4782 info = &md->u.tun_info;
4783 memset(info, 0, sizeof(*info));
4784 info->mode = IP_TUNNEL_INFO_TX;
4785
4786 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4787 if (flags & BPF_F_DONT_FRAGMENT)
4788 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4789 if (flags & BPF_F_ZERO_CSUM_TX)
4790 info->key.tun_flags &= ~TUNNEL_CSUM;
4791 if (flags & BPF_F_SEQ_NUMBER)
4792 info->key.tun_flags |= TUNNEL_SEQ;
4793 if (flags & BPF_F_NO_TUNNEL_KEY)
4794 info->key.tun_flags &= ~TUNNEL_KEY;
4795
4796 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4797 info->key.tos = from->tunnel_tos;
4798 info->key.ttl = from->tunnel_ttl;
4799
4800 if (flags & BPF_F_TUNINFO_IPV6) {
4801 info->mode |= IP_TUNNEL_INFO_IPV6;
4802 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4803 sizeof(from->remote_ipv6));
4804 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4805 sizeof(from->local_ipv6));
4806 info->key.label = cpu_to_be32(from->tunnel_label) &
4807 IPV6_FLOWLABEL_MASK;
4808 } else {
4809 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4810 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4811 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4812 }
4813
4814 return 0;
4815 }
4816
4817 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4818 .func = bpf_skb_set_tunnel_key,
4819 .gpl_only = false,
4820 .ret_type = RET_INTEGER,
4821 .arg1_type = ARG_PTR_TO_CTX,
4822 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4823 .arg3_type = ARG_CONST_SIZE,
4824 .arg4_type = ARG_ANYTHING,
4825 };
4826
BPF_CALL_3(bpf_skb_set_tunnel_opt,struct sk_buff *,skb,const u8 *,from,u32,size)4827 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4828 const u8 *, from, u32, size)
4829 {
4830 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4831 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4832
4833 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4834 return -EINVAL;
4835 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4836 return -ENOMEM;
4837
4838 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4839
4840 return 0;
4841 }
4842
4843 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4844 .func = bpf_skb_set_tunnel_opt,
4845 .gpl_only = false,
4846 .ret_type = RET_INTEGER,
4847 .arg1_type = ARG_PTR_TO_CTX,
4848 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4849 .arg3_type = ARG_CONST_SIZE,
4850 };
4851
4852 static const struct bpf_func_proto *
bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)4853 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4854 {
4855 if (!md_dst) {
4856 struct metadata_dst __percpu *tmp;
4857
4858 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4859 METADATA_IP_TUNNEL,
4860 GFP_KERNEL);
4861 if (!tmp)
4862 return NULL;
4863 if (cmpxchg(&md_dst, NULL, tmp))
4864 metadata_dst_free_percpu(tmp);
4865 }
4866
4867 switch (which) {
4868 case BPF_FUNC_skb_set_tunnel_key:
4869 return &bpf_skb_set_tunnel_key_proto;
4870 case BPF_FUNC_skb_set_tunnel_opt:
4871 return &bpf_skb_set_tunnel_opt_proto;
4872 default:
4873 return NULL;
4874 }
4875 }
4876
BPF_CALL_3(bpf_skb_under_cgroup,struct sk_buff *,skb,struct bpf_map *,map,u32,idx)4877 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4878 u32, idx)
4879 {
4880 struct bpf_array *array = container_of(map, struct bpf_array, map);
4881 struct cgroup *cgrp;
4882 struct sock *sk;
4883
4884 sk = skb_to_full_sk(skb);
4885 if (!sk || !sk_fullsock(sk))
4886 return -ENOENT;
4887 if (unlikely(idx >= array->map.max_entries))
4888 return -E2BIG;
4889
4890 cgrp = READ_ONCE(array->ptrs[idx]);
4891 if (unlikely(!cgrp))
4892 return -EAGAIN;
4893
4894 return sk_under_cgroup_hierarchy(sk, cgrp);
4895 }
4896
4897 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4898 .func = bpf_skb_under_cgroup,
4899 .gpl_only = false,
4900 .ret_type = RET_INTEGER,
4901 .arg1_type = ARG_PTR_TO_CTX,
4902 .arg2_type = ARG_CONST_MAP_PTR,
4903 .arg3_type = ARG_ANYTHING,
4904 };
4905
4906 #ifdef CONFIG_SOCK_CGROUP_DATA
__bpf_sk_cgroup_id(struct sock * sk)4907 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4908 {
4909 struct cgroup *cgrp;
4910
4911 sk = sk_to_full_sk(sk);
4912 if (!sk || !sk_fullsock(sk))
4913 return 0;
4914
4915 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4916 return cgroup_id(cgrp);
4917 }
4918
BPF_CALL_1(bpf_skb_cgroup_id,const struct sk_buff *,skb)4919 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4920 {
4921 return __bpf_sk_cgroup_id(skb->sk);
4922 }
4923
4924 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4925 .func = bpf_skb_cgroup_id,
4926 .gpl_only = false,
4927 .ret_type = RET_INTEGER,
4928 .arg1_type = ARG_PTR_TO_CTX,
4929 };
4930
__bpf_sk_ancestor_cgroup_id(struct sock * sk,int ancestor_level)4931 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4932 int ancestor_level)
4933 {
4934 struct cgroup *ancestor;
4935 struct cgroup *cgrp;
4936
4937 sk = sk_to_full_sk(sk);
4938 if (!sk || !sk_fullsock(sk))
4939 return 0;
4940
4941 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4942 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4943 if (!ancestor)
4944 return 0;
4945
4946 return cgroup_id(ancestor);
4947 }
4948
BPF_CALL_2(bpf_skb_ancestor_cgroup_id,const struct sk_buff *,skb,int,ancestor_level)4949 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4950 ancestor_level)
4951 {
4952 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4953 }
4954
4955 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4956 .func = bpf_skb_ancestor_cgroup_id,
4957 .gpl_only = false,
4958 .ret_type = RET_INTEGER,
4959 .arg1_type = ARG_PTR_TO_CTX,
4960 .arg2_type = ARG_ANYTHING,
4961 };
4962
BPF_CALL_1(bpf_sk_cgroup_id,struct sock *,sk)4963 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4964 {
4965 return __bpf_sk_cgroup_id(sk);
4966 }
4967
4968 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4969 .func = bpf_sk_cgroup_id,
4970 .gpl_only = false,
4971 .ret_type = RET_INTEGER,
4972 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4973 };
4974
BPF_CALL_2(bpf_sk_ancestor_cgroup_id,struct sock *,sk,int,ancestor_level)4975 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4976 {
4977 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4978 }
4979
4980 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4981 .func = bpf_sk_ancestor_cgroup_id,
4982 .gpl_only = false,
4983 .ret_type = RET_INTEGER,
4984 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4985 .arg2_type = ARG_ANYTHING,
4986 };
4987 #endif
4988
bpf_xdp_copy(void * dst,const void * ctx,unsigned long off,unsigned long len)4989 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4990 unsigned long off, unsigned long len)
4991 {
4992 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4993
4994 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4995 return 0;
4996 }
4997
BPF_CALL_5(bpf_xdp_event_output,struct xdp_buff *,xdp,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4998 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4999 u64, flags, void *, meta, u64, meta_size)
5000 {
5001 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
5002
5003 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
5004 return -EINVAL;
5005
5006 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
5007 return -EFAULT;
5008
5009 return bpf_event_output(map, flags, meta, meta_size, xdp,
5010 xdp_size, bpf_xdp_copy);
5011 }
5012
5013 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
5014 .func = bpf_xdp_event_output,
5015 .gpl_only = true,
5016 .ret_type = RET_INTEGER,
5017 .arg1_type = ARG_PTR_TO_CTX,
5018 .arg2_type = ARG_CONST_MAP_PTR,
5019 .arg3_type = ARG_ANYTHING,
5020 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5021 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5022 };
5023
5024 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
5025
5026 const struct bpf_func_proto bpf_xdp_output_proto = {
5027 .func = bpf_xdp_event_output,
5028 .gpl_only = true,
5029 .ret_type = RET_INTEGER,
5030 .arg1_type = ARG_PTR_TO_BTF_ID,
5031 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
5032 .arg2_type = ARG_CONST_MAP_PTR,
5033 .arg3_type = ARG_ANYTHING,
5034 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5035 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5036 };
5037
BPF_CALL_1(bpf_get_socket_cookie,struct sk_buff *,skb)5038 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
5039 {
5040 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
5041 }
5042
5043 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
5044 .func = bpf_get_socket_cookie,
5045 .gpl_only = false,
5046 .ret_type = RET_INTEGER,
5047 .arg1_type = ARG_PTR_TO_CTX,
5048 };
5049
BPF_CALL_1(bpf_get_socket_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5050 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5051 {
5052 return __sock_gen_cookie(ctx->sk);
5053 }
5054
5055 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
5056 .func = bpf_get_socket_cookie_sock_addr,
5057 .gpl_only = false,
5058 .ret_type = RET_INTEGER,
5059 .arg1_type = ARG_PTR_TO_CTX,
5060 };
5061
BPF_CALL_1(bpf_get_socket_cookie_sock,struct sock *,ctx)5062 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
5063 {
5064 return __sock_gen_cookie(ctx);
5065 }
5066
5067 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5068 .func = bpf_get_socket_cookie_sock,
5069 .gpl_only = false,
5070 .ret_type = RET_INTEGER,
5071 .arg1_type = ARG_PTR_TO_CTX,
5072 };
5073
BPF_CALL_1(bpf_get_socket_ptr_cookie,struct sock *,sk)5074 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5075 {
5076 return sk ? sock_gen_cookie(sk) : 0;
5077 }
5078
5079 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5080 .func = bpf_get_socket_ptr_cookie,
5081 .gpl_only = false,
5082 .ret_type = RET_INTEGER,
5083 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5084 };
5085
BPF_CALL_1(bpf_get_socket_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5086 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5087 {
5088 return __sock_gen_cookie(ctx->sk);
5089 }
5090
5091 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5092 .func = bpf_get_socket_cookie_sock_ops,
5093 .gpl_only = false,
5094 .ret_type = RET_INTEGER,
5095 .arg1_type = ARG_PTR_TO_CTX,
5096 };
5097
__bpf_get_netns_cookie(struct sock * sk)5098 static u64 __bpf_get_netns_cookie(struct sock *sk)
5099 {
5100 const struct net *net = sk ? sock_net(sk) : &init_net;
5101
5102 return net->net_cookie;
5103 }
5104
BPF_CALL_1(bpf_get_netns_cookie_sock,struct sock *,ctx)5105 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5106 {
5107 return __bpf_get_netns_cookie(ctx);
5108 }
5109
5110 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5111 .func = bpf_get_netns_cookie_sock,
5112 .gpl_only = false,
5113 .ret_type = RET_INTEGER,
5114 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5115 };
5116
BPF_CALL_1(bpf_get_netns_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5117 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5118 {
5119 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5120 }
5121
5122 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5123 .func = bpf_get_netns_cookie_sock_addr,
5124 .gpl_only = false,
5125 .ret_type = RET_INTEGER,
5126 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5127 };
5128
BPF_CALL_1(bpf_get_netns_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5129 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5130 {
5131 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5132 }
5133
5134 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5135 .func = bpf_get_netns_cookie_sock_ops,
5136 .gpl_only = false,
5137 .ret_type = RET_INTEGER,
5138 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5139 };
5140
BPF_CALL_1(bpf_get_netns_cookie_sk_msg,struct sk_msg *,ctx)5141 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5142 {
5143 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5144 }
5145
5146 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5147 .func = bpf_get_netns_cookie_sk_msg,
5148 .gpl_only = false,
5149 .ret_type = RET_INTEGER,
5150 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5151 };
5152
BPF_CALL_1(bpf_get_socket_uid,struct sk_buff *,skb)5153 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5154 {
5155 struct sock *sk = sk_to_full_sk(skb->sk);
5156 kuid_t kuid;
5157
5158 if (!sk || !sk_fullsock(sk))
5159 return overflowuid;
5160 kuid = sock_net_uid(sock_net(sk), sk);
5161 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5162 }
5163
5164 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5165 .func = bpf_get_socket_uid,
5166 .gpl_only = false,
5167 .ret_type = RET_INTEGER,
5168 .arg1_type = ARG_PTR_TO_CTX,
5169 };
5170
sol_socket_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5171 static int sol_socket_sockopt(struct sock *sk, int optname,
5172 char *optval, int *optlen,
5173 bool getopt)
5174 {
5175 switch (optname) {
5176 case SO_REUSEADDR:
5177 case SO_SNDBUF:
5178 case SO_RCVBUF:
5179 case SO_KEEPALIVE:
5180 case SO_PRIORITY:
5181 case SO_REUSEPORT:
5182 case SO_RCVLOWAT:
5183 case SO_MARK:
5184 case SO_MAX_PACING_RATE:
5185 case SO_BINDTOIFINDEX:
5186 case SO_TXREHASH:
5187 if (*optlen != sizeof(int))
5188 return -EINVAL;
5189 break;
5190 case SO_BINDTODEVICE:
5191 break;
5192 default:
5193 return -EINVAL;
5194 }
5195
5196 if (getopt) {
5197 if (optname == SO_BINDTODEVICE)
5198 return -EINVAL;
5199 return sk_getsockopt(sk, SOL_SOCKET, optname,
5200 KERNEL_SOCKPTR(optval),
5201 KERNEL_SOCKPTR(optlen));
5202 }
5203
5204 return sk_setsockopt(sk, SOL_SOCKET, optname,
5205 KERNEL_SOCKPTR(optval), *optlen);
5206 }
5207
bpf_sol_tcp_setsockopt(struct sock * sk,int optname,char * optval,int optlen)5208 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5209 char *optval, int optlen)
5210 {
5211 struct tcp_sock *tp = tcp_sk(sk);
5212 unsigned long timeout;
5213 int val;
5214
5215 if (optlen != sizeof(int))
5216 return -EINVAL;
5217
5218 val = *(int *)optval;
5219
5220 /* Only some options are supported */
5221 switch (optname) {
5222 case TCP_BPF_IW:
5223 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5224 return -EINVAL;
5225 tcp_snd_cwnd_set(tp, val);
5226 break;
5227 case TCP_BPF_SNDCWND_CLAMP:
5228 if (val <= 0)
5229 return -EINVAL;
5230 tp->snd_cwnd_clamp = val;
5231 tp->snd_ssthresh = val;
5232 break;
5233 case TCP_BPF_DELACK_MAX:
5234 timeout = usecs_to_jiffies(val);
5235 if (timeout > TCP_DELACK_MAX ||
5236 timeout < TCP_TIMEOUT_MIN)
5237 return -EINVAL;
5238 inet_csk(sk)->icsk_delack_max = timeout;
5239 break;
5240 case TCP_BPF_RTO_MIN:
5241 timeout = usecs_to_jiffies(val);
5242 if (timeout > TCP_RTO_MIN ||
5243 timeout < TCP_TIMEOUT_MIN)
5244 return -EINVAL;
5245 inet_csk(sk)->icsk_rto_min = timeout;
5246 break;
5247 default:
5248 return -EINVAL;
5249 }
5250
5251 return 0;
5252 }
5253
sol_tcp_sockopt_congestion(struct sock * sk,char * optval,int * optlen,bool getopt)5254 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5255 int *optlen, bool getopt)
5256 {
5257 struct tcp_sock *tp;
5258 int ret;
5259
5260 if (*optlen < 2)
5261 return -EINVAL;
5262
5263 if (getopt) {
5264 if (!inet_csk(sk)->icsk_ca_ops)
5265 return -EINVAL;
5266 /* BPF expects NULL-terminated tcp-cc string */
5267 optval[--(*optlen)] = '\0';
5268 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5269 KERNEL_SOCKPTR(optval),
5270 KERNEL_SOCKPTR(optlen));
5271 }
5272
5273 /* "cdg" is the only cc that alloc a ptr
5274 * in inet_csk_ca area. The bpf-tcp-cc may
5275 * overwrite this ptr after switching to cdg.
5276 */
5277 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5278 return -ENOTSUPP;
5279
5280 /* It stops this looping
5281 *
5282 * .init => bpf_setsockopt(tcp_cc) => .init =>
5283 * bpf_setsockopt(tcp_cc)" => .init => ....
5284 *
5285 * The second bpf_setsockopt(tcp_cc) is not allowed
5286 * in order to break the loop when both .init
5287 * are the same bpf prog.
5288 *
5289 * This applies even the second bpf_setsockopt(tcp_cc)
5290 * does not cause a loop. This limits only the first
5291 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5292 * pick a fallback cc (eg. peer does not support ECN)
5293 * and the second '.init' cannot fallback to
5294 * another.
5295 */
5296 tp = tcp_sk(sk);
5297 if (tp->bpf_chg_cc_inprogress)
5298 return -EBUSY;
5299
5300 tp->bpf_chg_cc_inprogress = 1;
5301 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5302 KERNEL_SOCKPTR(optval), *optlen);
5303 tp->bpf_chg_cc_inprogress = 0;
5304 return ret;
5305 }
5306
sol_tcp_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5307 static int sol_tcp_sockopt(struct sock *sk, int optname,
5308 char *optval, int *optlen,
5309 bool getopt)
5310 {
5311 if (sk->sk_protocol != IPPROTO_TCP)
5312 return -EINVAL;
5313
5314 switch (optname) {
5315 case TCP_NODELAY:
5316 case TCP_MAXSEG:
5317 case TCP_KEEPIDLE:
5318 case TCP_KEEPINTVL:
5319 case TCP_KEEPCNT:
5320 case TCP_SYNCNT:
5321 case TCP_WINDOW_CLAMP:
5322 case TCP_THIN_LINEAR_TIMEOUTS:
5323 case TCP_USER_TIMEOUT:
5324 case TCP_NOTSENT_LOWAT:
5325 case TCP_SAVE_SYN:
5326 if (*optlen != sizeof(int))
5327 return -EINVAL;
5328 break;
5329 case TCP_CONGESTION:
5330 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5331 case TCP_SAVED_SYN:
5332 if (*optlen < 1)
5333 return -EINVAL;
5334 break;
5335 default:
5336 if (getopt)
5337 return -EINVAL;
5338 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5339 }
5340
5341 if (getopt) {
5342 if (optname == TCP_SAVED_SYN) {
5343 struct tcp_sock *tp = tcp_sk(sk);
5344
5345 if (!tp->saved_syn ||
5346 *optlen > tcp_saved_syn_len(tp->saved_syn))
5347 return -EINVAL;
5348 memcpy(optval, tp->saved_syn->data, *optlen);
5349 /* It cannot free tp->saved_syn here because it
5350 * does not know if the user space still needs it.
5351 */
5352 return 0;
5353 }
5354
5355 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5356 KERNEL_SOCKPTR(optval),
5357 KERNEL_SOCKPTR(optlen));
5358 }
5359
5360 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5361 KERNEL_SOCKPTR(optval), *optlen);
5362 }
5363
sol_ip_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5364 static int sol_ip_sockopt(struct sock *sk, int optname,
5365 char *optval, int *optlen,
5366 bool getopt)
5367 {
5368 if (sk->sk_family != AF_INET)
5369 return -EINVAL;
5370
5371 switch (optname) {
5372 case IP_TOS:
5373 if (*optlen != sizeof(int))
5374 return -EINVAL;
5375 break;
5376 default:
5377 return -EINVAL;
5378 }
5379
5380 if (getopt)
5381 return do_ip_getsockopt(sk, SOL_IP, optname,
5382 KERNEL_SOCKPTR(optval),
5383 KERNEL_SOCKPTR(optlen));
5384
5385 return do_ip_setsockopt(sk, SOL_IP, optname,
5386 KERNEL_SOCKPTR(optval), *optlen);
5387 }
5388
sol_ipv6_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5389 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5390 char *optval, int *optlen,
5391 bool getopt)
5392 {
5393 if (sk->sk_family != AF_INET6)
5394 return -EINVAL;
5395
5396 switch (optname) {
5397 case IPV6_TCLASS:
5398 case IPV6_AUTOFLOWLABEL:
5399 if (*optlen != sizeof(int))
5400 return -EINVAL;
5401 break;
5402 default:
5403 return -EINVAL;
5404 }
5405
5406 if (getopt)
5407 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5408 KERNEL_SOCKPTR(optval),
5409 KERNEL_SOCKPTR(optlen));
5410
5411 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5412 KERNEL_SOCKPTR(optval), *optlen);
5413 }
5414
__bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5415 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5416 char *optval, int optlen)
5417 {
5418 if (!sk_fullsock(sk))
5419 return -EINVAL;
5420
5421 if (level == SOL_SOCKET)
5422 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5423 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5424 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5425 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5426 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5427 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5428 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5429
5430 return -EINVAL;
5431 }
5432
_bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5433 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5434 char *optval, int optlen)
5435 {
5436 if (sk_fullsock(sk))
5437 sock_owned_by_me(sk);
5438 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5439 }
5440
__bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5441 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5442 char *optval, int optlen)
5443 {
5444 int err, saved_optlen = optlen;
5445
5446 if (!sk_fullsock(sk)) {
5447 err = -EINVAL;
5448 goto done;
5449 }
5450
5451 if (level == SOL_SOCKET)
5452 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5453 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5454 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5455 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5456 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5457 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5458 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5459 else
5460 err = -EINVAL;
5461
5462 done:
5463 if (err)
5464 optlen = 0;
5465 if (optlen < saved_optlen)
5466 memset(optval + optlen, 0, saved_optlen - optlen);
5467 return err;
5468 }
5469
_bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5470 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5471 char *optval, int optlen)
5472 {
5473 if (sk_fullsock(sk))
5474 sock_owned_by_me(sk);
5475 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5476 }
5477
BPF_CALL_5(bpf_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5478 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5479 int, optname, char *, optval, int, optlen)
5480 {
5481 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5482 }
5483
5484 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5485 .func = bpf_sk_setsockopt,
5486 .gpl_only = false,
5487 .ret_type = RET_INTEGER,
5488 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5489 .arg2_type = ARG_ANYTHING,
5490 .arg3_type = ARG_ANYTHING,
5491 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5492 .arg5_type = ARG_CONST_SIZE,
5493 };
5494
BPF_CALL_5(bpf_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5495 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5496 int, optname, char *, optval, int, optlen)
5497 {
5498 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5499 }
5500
5501 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5502 .func = bpf_sk_getsockopt,
5503 .gpl_only = false,
5504 .ret_type = RET_INTEGER,
5505 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5506 .arg2_type = ARG_ANYTHING,
5507 .arg3_type = ARG_ANYTHING,
5508 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5509 .arg5_type = ARG_CONST_SIZE,
5510 };
5511
BPF_CALL_5(bpf_unlocked_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5512 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5513 int, optname, char *, optval, int, optlen)
5514 {
5515 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5516 }
5517
5518 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5519 .func = bpf_unlocked_sk_setsockopt,
5520 .gpl_only = false,
5521 .ret_type = RET_INTEGER,
5522 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5523 .arg2_type = ARG_ANYTHING,
5524 .arg3_type = ARG_ANYTHING,
5525 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5526 .arg5_type = ARG_CONST_SIZE,
5527 };
5528
BPF_CALL_5(bpf_unlocked_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5529 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5530 int, optname, char *, optval, int, optlen)
5531 {
5532 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5533 }
5534
5535 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5536 .func = bpf_unlocked_sk_getsockopt,
5537 .gpl_only = false,
5538 .ret_type = RET_INTEGER,
5539 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5540 .arg2_type = ARG_ANYTHING,
5541 .arg3_type = ARG_ANYTHING,
5542 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5543 .arg5_type = ARG_CONST_SIZE,
5544 };
5545
BPF_CALL_5(bpf_sock_addr_setsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5546 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5547 int, level, int, optname, char *, optval, int, optlen)
5548 {
5549 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5550 }
5551
5552 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5553 .func = bpf_sock_addr_setsockopt,
5554 .gpl_only = false,
5555 .ret_type = RET_INTEGER,
5556 .arg1_type = ARG_PTR_TO_CTX,
5557 .arg2_type = ARG_ANYTHING,
5558 .arg3_type = ARG_ANYTHING,
5559 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5560 .arg5_type = ARG_CONST_SIZE,
5561 };
5562
BPF_CALL_5(bpf_sock_addr_getsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5563 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5564 int, level, int, optname, char *, optval, int, optlen)
5565 {
5566 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5567 }
5568
5569 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5570 .func = bpf_sock_addr_getsockopt,
5571 .gpl_only = false,
5572 .ret_type = RET_INTEGER,
5573 .arg1_type = ARG_PTR_TO_CTX,
5574 .arg2_type = ARG_ANYTHING,
5575 .arg3_type = ARG_ANYTHING,
5576 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5577 .arg5_type = ARG_CONST_SIZE,
5578 };
5579
BPF_CALL_5(bpf_sock_ops_setsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5580 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5581 int, level, int, optname, char *, optval, int, optlen)
5582 {
5583 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5584 }
5585
5586 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5587 .func = bpf_sock_ops_setsockopt,
5588 .gpl_only = false,
5589 .ret_type = RET_INTEGER,
5590 .arg1_type = ARG_PTR_TO_CTX,
5591 .arg2_type = ARG_ANYTHING,
5592 .arg3_type = ARG_ANYTHING,
5593 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5594 .arg5_type = ARG_CONST_SIZE,
5595 };
5596
bpf_sock_ops_get_syn(struct bpf_sock_ops_kern * bpf_sock,int optname,const u8 ** start)5597 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5598 int optname, const u8 **start)
5599 {
5600 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5601 const u8 *hdr_start;
5602 int ret;
5603
5604 if (syn_skb) {
5605 /* sk is a request_sock here */
5606
5607 if (optname == TCP_BPF_SYN) {
5608 hdr_start = syn_skb->data;
5609 ret = tcp_hdrlen(syn_skb);
5610 } else if (optname == TCP_BPF_SYN_IP) {
5611 hdr_start = skb_network_header(syn_skb);
5612 ret = skb_network_header_len(syn_skb) +
5613 tcp_hdrlen(syn_skb);
5614 } else {
5615 /* optname == TCP_BPF_SYN_MAC */
5616 hdr_start = skb_mac_header(syn_skb);
5617 ret = skb_mac_header_len(syn_skb) +
5618 skb_network_header_len(syn_skb) +
5619 tcp_hdrlen(syn_skb);
5620 }
5621 } else {
5622 struct sock *sk = bpf_sock->sk;
5623 struct saved_syn *saved_syn;
5624
5625 if (sk->sk_state == TCP_NEW_SYN_RECV)
5626 /* synack retransmit. bpf_sock->syn_skb will
5627 * not be available. It has to resort to
5628 * saved_syn (if it is saved).
5629 */
5630 saved_syn = inet_reqsk(sk)->saved_syn;
5631 else
5632 saved_syn = tcp_sk(sk)->saved_syn;
5633
5634 if (!saved_syn)
5635 return -ENOENT;
5636
5637 if (optname == TCP_BPF_SYN) {
5638 hdr_start = saved_syn->data +
5639 saved_syn->mac_hdrlen +
5640 saved_syn->network_hdrlen;
5641 ret = saved_syn->tcp_hdrlen;
5642 } else if (optname == TCP_BPF_SYN_IP) {
5643 hdr_start = saved_syn->data +
5644 saved_syn->mac_hdrlen;
5645 ret = saved_syn->network_hdrlen +
5646 saved_syn->tcp_hdrlen;
5647 } else {
5648 /* optname == TCP_BPF_SYN_MAC */
5649
5650 /* TCP_SAVE_SYN may not have saved the mac hdr */
5651 if (!saved_syn->mac_hdrlen)
5652 return -ENOENT;
5653
5654 hdr_start = saved_syn->data;
5655 ret = saved_syn->mac_hdrlen +
5656 saved_syn->network_hdrlen +
5657 saved_syn->tcp_hdrlen;
5658 }
5659 }
5660
5661 *start = hdr_start;
5662 return ret;
5663 }
5664
BPF_CALL_5(bpf_sock_ops_getsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5665 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5666 int, level, int, optname, char *, optval, int, optlen)
5667 {
5668 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5669 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5670 int ret, copy_len = 0;
5671 const u8 *start;
5672
5673 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5674 if (ret > 0) {
5675 copy_len = ret;
5676 if (optlen < copy_len) {
5677 copy_len = optlen;
5678 ret = -ENOSPC;
5679 }
5680
5681 memcpy(optval, start, copy_len);
5682 }
5683
5684 /* Zero out unused buffer at the end */
5685 memset(optval + copy_len, 0, optlen - copy_len);
5686
5687 return ret;
5688 }
5689
5690 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5691 }
5692
5693 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5694 .func = bpf_sock_ops_getsockopt,
5695 .gpl_only = false,
5696 .ret_type = RET_INTEGER,
5697 .arg1_type = ARG_PTR_TO_CTX,
5698 .arg2_type = ARG_ANYTHING,
5699 .arg3_type = ARG_ANYTHING,
5700 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5701 .arg5_type = ARG_CONST_SIZE,
5702 };
5703
BPF_CALL_2(bpf_sock_ops_cb_flags_set,struct bpf_sock_ops_kern *,bpf_sock,int,argval)5704 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5705 int, argval)
5706 {
5707 struct sock *sk = bpf_sock->sk;
5708 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5709
5710 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5711 return -EINVAL;
5712
5713 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5714
5715 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5716 }
5717
5718 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5719 .func = bpf_sock_ops_cb_flags_set,
5720 .gpl_only = false,
5721 .ret_type = RET_INTEGER,
5722 .arg1_type = ARG_PTR_TO_CTX,
5723 .arg2_type = ARG_ANYTHING,
5724 };
5725
5726 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5727 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5728
BPF_CALL_3(bpf_bind,struct bpf_sock_addr_kern *,ctx,struct sockaddr *,addr,int,addr_len)5729 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5730 int, addr_len)
5731 {
5732 #ifdef CONFIG_INET
5733 struct sock *sk = ctx->sk;
5734 u32 flags = BIND_FROM_BPF;
5735 int err;
5736
5737 err = -EINVAL;
5738 if (addr_len < offsetofend(struct sockaddr, sa_family))
5739 return err;
5740 if (addr->sa_family == AF_INET) {
5741 if (addr_len < sizeof(struct sockaddr_in))
5742 return err;
5743 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5744 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5745 return __inet_bind(sk, addr, addr_len, flags);
5746 #if IS_ENABLED(CONFIG_IPV6)
5747 } else if (addr->sa_family == AF_INET6) {
5748 if (addr_len < SIN6_LEN_RFC2133)
5749 return err;
5750 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5751 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5752 /* ipv6_bpf_stub cannot be NULL, since it's called from
5753 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5754 */
5755 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5756 #endif /* CONFIG_IPV6 */
5757 }
5758 #endif /* CONFIG_INET */
5759
5760 return -EAFNOSUPPORT;
5761 }
5762
5763 static const struct bpf_func_proto bpf_bind_proto = {
5764 .func = bpf_bind,
5765 .gpl_only = false,
5766 .ret_type = RET_INTEGER,
5767 .arg1_type = ARG_PTR_TO_CTX,
5768 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5769 .arg3_type = ARG_CONST_SIZE,
5770 };
5771
5772 #ifdef CONFIG_XFRM
5773
5774 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5775 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5776
5777 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5778 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5779
5780 #endif
5781
BPF_CALL_5(bpf_skb_get_xfrm_state,struct sk_buff *,skb,u32,index,struct bpf_xfrm_state *,to,u32,size,u64,flags)5782 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5783 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5784 {
5785 const struct sec_path *sp = skb_sec_path(skb);
5786 const struct xfrm_state *x;
5787
5788 if (!sp || unlikely(index >= sp->len || flags))
5789 goto err_clear;
5790
5791 x = sp->xvec[index];
5792
5793 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5794 goto err_clear;
5795
5796 to->reqid = x->props.reqid;
5797 to->spi = x->id.spi;
5798 to->family = x->props.family;
5799 to->ext = 0;
5800
5801 if (to->family == AF_INET6) {
5802 memcpy(to->remote_ipv6, x->props.saddr.a6,
5803 sizeof(to->remote_ipv6));
5804 } else {
5805 to->remote_ipv4 = x->props.saddr.a4;
5806 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5807 }
5808
5809 return 0;
5810 err_clear:
5811 memset(to, 0, size);
5812 return -EINVAL;
5813 }
5814
5815 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5816 .func = bpf_skb_get_xfrm_state,
5817 .gpl_only = false,
5818 .ret_type = RET_INTEGER,
5819 .arg1_type = ARG_PTR_TO_CTX,
5820 .arg2_type = ARG_ANYTHING,
5821 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5822 .arg4_type = ARG_CONST_SIZE,
5823 .arg5_type = ARG_ANYTHING,
5824 };
5825 #endif
5826
5827 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
bpf_fib_set_fwd_params(struct bpf_fib_lookup * params,u32 mtu)5828 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5829 {
5830 params->h_vlan_TCI = 0;
5831 params->h_vlan_proto = 0;
5832 if (mtu)
5833 params->mtu_result = mtu; /* union with tot_len */
5834
5835 return 0;
5836 }
5837 #endif
5838
5839 #if IS_ENABLED(CONFIG_INET)
bpf_ipv4_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5840 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5841 u32 flags, bool check_mtu)
5842 {
5843 struct fib_nh_common *nhc;
5844 struct in_device *in_dev;
5845 struct neighbour *neigh;
5846 struct net_device *dev;
5847 struct fib_result res;
5848 struct flowi4 fl4;
5849 u32 mtu = 0;
5850 int err;
5851
5852 dev = dev_get_by_index_rcu(net, params->ifindex);
5853 if (unlikely(!dev))
5854 return -ENODEV;
5855
5856 /* verify forwarding is enabled on this interface */
5857 in_dev = __in_dev_get_rcu(dev);
5858 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5859 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5860
5861 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5862 fl4.flowi4_iif = 1;
5863 fl4.flowi4_oif = params->ifindex;
5864 } else {
5865 fl4.flowi4_iif = params->ifindex;
5866 fl4.flowi4_oif = 0;
5867 }
5868 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5869 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5870 fl4.flowi4_flags = 0;
5871
5872 fl4.flowi4_proto = params->l4_protocol;
5873 fl4.daddr = params->ipv4_dst;
5874 fl4.saddr = params->ipv4_src;
5875 fl4.fl4_sport = params->sport;
5876 fl4.fl4_dport = params->dport;
5877 fl4.flowi4_multipath_hash = 0;
5878
5879 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5880 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5881 struct fib_table *tb;
5882
5883 if (flags & BPF_FIB_LOOKUP_TBID) {
5884 tbid = params->tbid;
5885 /* zero out for vlan output */
5886 params->tbid = 0;
5887 }
5888
5889 tb = fib_get_table(net, tbid);
5890 if (unlikely(!tb))
5891 return BPF_FIB_LKUP_RET_NOT_FWDED;
5892
5893 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5894 } else {
5895 fl4.flowi4_mark = 0;
5896 fl4.flowi4_secid = 0;
5897 fl4.flowi4_tun_key.tun_id = 0;
5898 fl4.flowi4_uid = sock_net_uid(net, NULL);
5899
5900 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5901 }
5902
5903 if (err) {
5904 /* map fib lookup errors to RTN_ type */
5905 if (err == -EINVAL)
5906 return BPF_FIB_LKUP_RET_BLACKHOLE;
5907 if (err == -EHOSTUNREACH)
5908 return BPF_FIB_LKUP_RET_UNREACHABLE;
5909 if (err == -EACCES)
5910 return BPF_FIB_LKUP_RET_PROHIBIT;
5911
5912 return BPF_FIB_LKUP_RET_NOT_FWDED;
5913 }
5914
5915 if (res.type != RTN_UNICAST)
5916 return BPF_FIB_LKUP_RET_NOT_FWDED;
5917
5918 if (fib_info_num_path(res.fi) > 1)
5919 fib_select_path(net, &res, &fl4, NULL);
5920
5921 if (check_mtu) {
5922 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5923 if (params->tot_len > mtu) {
5924 params->mtu_result = mtu; /* union with tot_len */
5925 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5926 }
5927 }
5928
5929 nhc = res.nhc;
5930
5931 /* do not handle lwt encaps right now */
5932 if (nhc->nhc_lwtstate)
5933 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5934
5935 dev = nhc->nhc_dev;
5936
5937 params->rt_metric = res.fi->fib_priority;
5938 params->ifindex = dev->ifindex;
5939
5940 if (flags & BPF_FIB_LOOKUP_SRC)
5941 params->ipv4_src = fib_result_prefsrc(net, &res);
5942
5943 /* xdp and cls_bpf programs are run in RCU-bh so
5944 * rcu_read_lock_bh is not needed here
5945 */
5946 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5947 if (nhc->nhc_gw_family)
5948 params->ipv4_dst = nhc->nhc_gw.ipv4;
5949 } else {
5950 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5951
5952 params->family = AF_INET6;
5953 *dst = nhc->nhc_gw.ipv6;
5954 }
5955
5956 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5957 goto set_fwd_params;
5958
5959 if (likely(nhc->nhc_gw_family != AF_INET6))
5960 neigh = __ipv4_neigh_lookup_noref(dev,
5961 (__force u32)params->ipv4_dst);
5962 else
5963 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5964
5965 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5966 return BPF_FIB_LKUP_RET_NO_NEIGH;
5967 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5968 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5969
5970 set_fwd_params:
5971 return bpf_fib_set_fwd_params(params, mtu);
5972 }
5973 #endif
5974
5975 #if IS_ENABLED(CONFIG_IPV6)
bpf_ipv6_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5976 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5977 u32 flags, bool check_mtu)
5978 {
5979 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5980 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5981 struct fib6_result res = {};
5982 struct neighbour *neigh;
5983 struct net_device *dev;
5984 struct inet6_dev *idev;
5985 struct flowi6 fl6;
5986 int strict = 0;
5987 int oif, err;
5988 u32 mtu = 0;
5989
5990 /* link local addresses are never forwarded */
5991 if (rt6_need_strict(dst) || rt6_need_strict(src))
5992 return BPF_FIB_LKUP_RET_NOT_FWDED;
5993
5994 dev = dev_get_by_index_rcu(net, params->ifindex);
5995 if (unlikely(!dev))
5996 return -ENODEV;
5997
5998 idev = __in6_dev_get_safely(dev);
5999 if (unlikely(!idev || !idev->cnf.forwarding))
6000 return BPF_FIB_LKUP_RET_FWD_DISABLED;
6001
6002 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
6003 fl6.flowi6_iif = 1;
6004 oif = fl6.flowi6_oif = params->ifindex;
6005 } else {
6006 oif = fl6.flowi6_iif = params->ifindex;
6007 fl6.flowi6_oif = 0;
6008 strict = RT6_LOOKUP_F_HAS_SADDR;
6009 }
6010 fl6.flowlabel = params->flowinfo;
6011 fl6.flowi6_scope = 0;
6012 fl6.flowi6_flags = 0;
6013 fl6.mp_hash = 0;
6014
6015 fl6.flowi6_proto = params->l4_protocol;
6016 fl6.daddr = *dst;
6017 fl6.saddr = *src;
6018 fl6.fl6_sport = params->sport;
6019 fl6.fl6_dport = params->dport;
6020
6021 if (flags & BPF_FIB_LOOKUP_DIRECT) {
6022 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
6023 struct fib6_table *tb;
6024
6025 if (flags & BPF_FIB_LOOKUP_TBID) {
6026 tbid = params->tbid;
6027 /* zero out for vlan output */
6028 params->tbid = 0;
6029 }
6030
6031 tb = ipv6_stub->fib6_get_table(net, tbid);
6032 if (unlikely(!tb))
6033 return BPF_FIB_LKUP_RET_NOT_FWDED;
6034
6035 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
6036 strict);
6037 } else {
6038 fl6.flowi6_mark = 0;
6039 fl6.flowi6_secid = 0;
6040 fl6.flowi6_tun_key.tun_id = 0;
6041 fl6.flowi6_uid = sock_net_uid(net, NULL);
6042
6043 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
6044 }
6045
6046 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
6047 res.f6i == net->ipv6.fib6_null_entry))
6048 return BPF_FIB_LKUP_RET_NOT_FWDED;
6049
6050 switch (res.fib6_type) {
6051 /* only unicast is forwarded */
6052 case RTN_UNICAST:
6053 break;
6054 case RTN_BLACKHOLE:
6055 return BPF_FIB_LKUP_RET_BLACKHOLE;
6056 case RTN_UNREACHABLE:
6057 return BPF_FIB_LKUP_RET_UNREACHABLE;
6058 case RTN_PROHIBIT:
6059 return BPF_FIB_LKUP_RET_PROHIBIT;
6060 default:
6061 return BPF_FIB_LKUP_RET_NOT_FWDED;
6062 }
6063
6064 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
6065 fl6.flowi6_oif != 0, NULL, strict);
6066
6067 if (check_mtu) {
6068 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
6069 if (params->tot_len > mtu) {
6070 params->mtu_result = mtu; /* union with tot_len */
6071 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6072 }
6073 }
6074
6075 if (res.nh->fib_nh_lws)
6076 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6077
6078 if (res.nh->fib_nh_gw_family)
6079 *dst = res.nh->fib_nh_gw6;
6080
6081 dev = res.nh->fib_nh_dev;
6082 params->rt_metric = res.f6i->fib6_metric;
6083 params->ifindex = dev->ifindex;
6084
6085 if (flags & BPF_FIB_LOOKUP_SRC) {
6086 if (res.f6i->fib6_prefsrc.plen) {
6087 *src = res.f6i->fib6_prefsrc.addr;
6088 } else {
6089 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
6090 &fl6.daddr, 0,
6091 src);
6092 if (err)
6093 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
6094 }
6095 }
6096
6097 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6098 goto set_fwd_params;
6099
6100 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6101 * not needed here.
6102 */
6103 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6104 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6105 return BPF_FIB_LKUP_RET_NO_NEIGH;
6106 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6107 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6108
6109 set_fwd_params:
6110 return bpf_fib_set_fwd_params(params, mtu);
6111 }
6112 #endif
6113
6114 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6115 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6116 BPF_FIB_LOOKUP_SRC)
6117
BPF_CALL_4(bpf_xdp_fib_lookup,struct xdp_buff *,ctx,struct bpf_fib_lookup *,params,int,plen,u32,flags)6118 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6119 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6120 {
6121 if (plen < sizeof(*params))
6122 return -EINVAL;
6123
6124 if (flags & ~BPF_FIB_LOOKUP_MASK)
6125 return -EINVAL;
6126
6127 switch (params->family) {
6128 #if IS_ENABLED(CONFIG_INET)
6129 case AF_INET:
6130 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6131 flags, true);
6132 #endif
6133 #if IS_ENABLED(CONFIG_IPV6)
6134 case AF_INET6:
6135 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6136 flags, true);
6137 #endif
6138 }
6139 return -EAFNOSUPPORT;
6140 }
6141
6142 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6143 .func = bpf_xdp_fib_lookup,
6144 .gpl_only = true,
6145 .ret_type = RET_INTEGER,
6146 .arg1_type = ARG_PTR_TO_CTX,
6147 .arg2_type = ARG_PTR_TO_MEM,
6148 .arg3_type = ARG_CONST_SIZE,
6149 .arg4_type = ARG_ANYTHING,
6150 };
6151
BPF_CALL_4(bpf_skb_fib_lookup,struct sk_buff *,skb,struct bpf_fib_lookup *,params,int,plen,u32,flags)6152 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6153 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6154 {
6155 struct net *net = dev_net(skb->dev);
6156 int rc = -EAFNOSUPPORT;
6157 bool check_mtu = false;
6158
6159 if (plen < sizeof(*params))
6160 return -EINVAL;
6161
6162 if (flags & ~BPF_FIB_LOOKUP_MASK)
6163 return -EINVAL;
6164
6165 if (params->tot_len)
6166 check_mtu = true;
6167
6168 switch (params->family) {
6169 #if IS_ENABLED(CONFIG_INET)
6170 case AF_INET:
6171 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6172 break;
6173 #endif
6174 #if IS_ENABLED(CONFIG_IPV6)
6175 case AF_INET6:
6176 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6177 break;
6178 #endif
6179 }
6180
6181 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6182 struct net_device *dev;
6183
6184 /* When tot_len isn't provided by user, check skb
6185 * against MTU of FIB lookup resulting net_device
6186 */
6187 dev = dev_get_by_index_rcu(net, params->ifindex);
6188 if (!is_skb_forwardable(dev, skb))
6189 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6190
6191 params->mtu_result = dev->mtu; /* union with tot_len */
6192 }
6193
6194 return rc;
6195 }
6196
6197 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6198 .func = bpf_skb_fib_lookup,
6199 .gpl_only = true,
6200 .ret_type = RET_INTEGER,
6201 .arg1_type = ARG_PTR_TO_CTX,
6202 .arg2_type = ARG_PTR_TO_MEM,
6203 .arg3_type = ARG_CONST_SIZE,
6204 .arg4_type = ARG_ANYTHING,
6205 };
6206
__dev_via_ifindex(struct net_device * dev_curr,u32 ifindex)6207 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6208 u32 ifindex)
6209 {
6210 struct net *netns = dev_net(dev_curr);
6211
6212 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6213 if (ifindex == 0)
6214 return dev_curr;
6215
6216 return dev_get_by_index_rcu(netns, ifindex);
6217 }
6218
BPF_CALL_5(bpf_skb_check_mtu,struct sk_buff *,skb,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6219 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6220 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6221 {
6222 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6223 struct net_device *dev = skb->dev;
6224 int skb_len, dev_len;
6225 int mtu = 0;
6226
6227 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) {
6228 ret = -EINVAL;
6229 goto out;
6230 }
6231
6232 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) {
6233 ret = -EINVAL;
6234 goto out;
6235 }
6236
6237 dev = __dev_via_ifindex(dev, ifindex);
6238 if (unlikely(!dev)) {
6239 ret = -ENODEV;
6240 goto out;
6241 }
6242
6243 mtu = READ_ONCE(dev->mtu);
6244 dev_len = mtu + dev->hard_header_len;
6245
6246 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6247 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6248
6249 skb_len += len_diff; /* minus result pass check */
6250 if (skb_len <= dev_len) {
6251 ret = BPF_MTU_CHK_RET_SUCCESS;
6252 goto out;
6253 }
6254 /* At this point, skb->len exceed MTU, but as it include length of all
6255 * segments, it can still be below MTU. The SKB can possibly get
6256 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6257 * must choose if segs are to be MTU checked.
6258 */
6259 if (skb_is_gso(skb)) {
6260 ret = BPF_MTU_CHK_RET_SUCCESS;
6261 if (flags & BPF_MTU_CHK_SEGS &&
6262 !skb_gso_validate_network_len(skb, mtu))
6263 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6264 }
6265 out:
6266 *mtu_len = mtu;
6267 return ret;
6268 }
6269
BPF_CALL_5(bpf_xdp_check_mtu,struct xdp_buff *,xdp,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6270 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6271 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6272 {
6273 struct net_device *dev = xdp->rxq->dev;
6274 int xdp_len = xdp->data_end - xdp->data;
6275 int ret = BPF_MTU_CHK_RET_SUCCESS;
6276 int mtu = 0, dev_len;
6277
6278 /* XDP variant doesn't support multi-buffer segment check (yet) */
6279 if (unlikely(flags)) {
6280 ret = -EINVAL;
6281 goto out;
6282 }
6283
6284 dev = __dev_via_ifindex(dev, ifindex);
6285 if (unlikely(!dev)) {
6286 ret = -ENODEV;
6287 goto out;
6288 }
6289
6290 mtu = READ_ONCE(dev->mtu);
6291 dev_len = mtu + dev->hard_header_len;
6292
6293 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6294 if (*mtu_len)
6295 xdp_len = *mtu_len + dev->hard_header_len;
6296
6297 xdp_len += len_diff; /* minus result pass check */
6298 if (xdp_len > dev_len)
6299 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6300 out:
6301 *mtu_len = mtu;
6302 return ret;
6303 }
6304
6305 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6306 .func = bpf_skb_check_mtu,
6307 .gpl_only = true,
6308 .ret_type = RET_INTEGER,
6309 .arg1_type = ARG_PTR_TO_CTX,
6310 .arg2_type = ARG_ANYTHING,
6311 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_ALIGNED,
6312 .arg3_size = sizeof(u32),
6313 .arg4_type = ARG_ANYTHING,
6314 .arg5_type = ARG_ANYTHING,
6315 };
6316
6317 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6318 .func = bpf_xdp_check_mtu,
6319 .gpl_only = true,
6320 .ret_type = RET_INTEGER,
6321 .arg1_type = ARG_PTR_TO_CTX,
6322 .arg2_type = ARG_ANYTHING,
6323 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_ALIGNED,
6324 .arg3_size = sizeof(u32),
6325 .arg4_type = ARG_ANYTHING,
6326 .arg5_type = ARG_ANYTHING,
6327 };
6328
6329 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
bpf_push_seg6_encap(struct sk_buff * skb,u32 type,void * hdr,u32 len)6330 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6331 {
6332 int err;
6333 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6334
6335 if (!seg6_validate_srh(srh, len, false))
6336 return -EINVAL;
6337
6338 switch (type) {
6339 case BPF_LWT_ENCAP_SEG6_INLINE:
6340 if (skb->protocol != htons(ETH_P_IPV6))
6341 return -EBADMSG;
6342
6343 err = seg6_do_srh_inline(skb, srh);
6344 break;
6345 case BPF_LWT_ENCAP_SEG6:
6346 skb_reset_inner_headers(skb);
6347 skb->encapsulation = 1;
6348 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6349 break;
6350 default:
6351 return -EINVAL;
6352 }
6353
6354 bpf_compute_data_pointers(skb);
6355 if (err)
6356 return err;
6357
6358 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6359
6360 return seg6_lookup_nexthop(skb, NULL, 0);
6361 }
6362 #endif /* CONFIG_IPV6_SEG6_BPF */
6363
6364 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
bpf_push_ip_encap(struct sk_buff * skb,void * hdr,u32 len,bool ingress)6365 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6366 bool ingress)
6367 {
6368 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6369 }
6370 #endif
6371
BPF_CALL_4(bpf_lwt_in_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6372 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6373 u32, len)
6374 {
6375 switch (type) {
6376 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6377 case BPF_LWT_ENCAP_SEG6:
6378 case BPF_LWT_ENCAP_SEG6_INLINE:
6379 return bpf_push_seg6_encap(skb, type, hdr, len);
6380 #endif
6381 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6382 case BPF_LWT_ENCAP_IP:
6383 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6384 #endif
6385 default:
6386 return -EINVAL;
6387 }
6388 }
6389
BPF_CALL_4(bpf_lwt_xmit_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6390 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6391 void *, hdr, u32, len)
6392 {
6393 switch (type) {
6394 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6395 case BPF_LWT_ENCAP_IP:
6396 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6397 #endif
6398 default:
6399 return -EINVAL;
6400 }
6401 }
6402
6403 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6404 .func = bpf_lwt_in_push_encap,
6405 .gpl_only = false,
6406 .ret_type = RET_INTEGER,
6407 .arg1_type = ARG_PTR_TO_CTX,
6408 .arg2_type = ARG_ANYTHING,
6409 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6410 .arg4_type = ARG_CONST_SIZE
6411 };
6412
6413 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6414 .func = bpf_lwt_xmit_push_encap,
6415 .gpl_only = false,
6416 .ret_type = RET_INTEGER,
6417 .arg1_type = ARG_PTR_TO_CTX,
6418 .arg2_type = ARG_ANYTHING,
6419 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6420 .arg4_type = ARG_CONST_SIZE
6421 };
6422
6423 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
BPF_CALL_4(bpf_lwt_seg6_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len)6424 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6425 const void *, from, u32, len)
6426 {
6427 struct seg6_bpf_srh_state *srh_state =
6428 this_cpu_ptr(&seg6_bpf_srh_states);
6429 struct ipv6_sr_hdr *srh = srh_state->srh;
6430 void *srh_tlvs, *srh_end, *ptr;
6431 int srhoff = 0;
6432
6433 if (srh == NULL)
6434 return -EINVAL;
6435
6436 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6437 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6438
6439 ptr = skb->data + offset;
6440 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6441 srh_state->valid = false;
6442 else if (ptr < (void *)&srh->flags ||
6443 ptr + len > (void *)&srh->segments)
6444 return -EFAULT;
6445
6446 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6447 return -EFAULT;
6448 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6449 return -EINVAL;
6450 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6451
6452 memcpy(skb->data + offset, from, len);
6453 return 0;
6454 }
6455
6456 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6457 .func = bpf_lwt_seg6_store_bytes,
6458 .gpl_only = false,
6459 .ret_type = RET_INTEGER,
6460 .arg1_type = ARG_PTR_TO_CTX,
6461 .arg2_type = ARG_ANYTHING,
6462 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6463 .arg4_type = ARG_CONST_SIZE
6464 };
6465
bpf_update_srh_state(struct sk_buff * skb)6466 static void bpf_update_srh_state(struct sk_buff *skb)
6467 {
6468 struct seg6_bpf_srh_state *srh_state =
6469 this_cpu_ptr(&seg6_bpf_srh_states);
6470 int srhoff = 0;
6471
6472 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6473 srh_state->srh = NULL;
6474 } else {
6475 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6476 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6477 srh_state->valid = true;
6478 }
6479 }
6480
BPF_CALL_4(bpf_lwt_seg6_action,struct sk_buff *,skb,u32,action,void *,param,u32,param_len)6481 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6482 u32, action, void *, param, u32, param_len)
6483 {
6484 struct seg6_bpf_srh_state *srh_state =
6485 this_cpu_ptr(&seg6_bpf_srh_states);
6486 int hdroff = 0;
6487 int err;
6488
6489 switch (action) {
6490 case SEG6_LOCAL_ACTION_END_X:
6491 if (!seg6_bpf_has_valid_srh(skb))
6492 return -EBADMSG;
6493 if (param_len != sizeof(struct in6_addr))
6494 return -EINVAL;
6495 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6496 case SEG6_LOCAL_ACTION_END_T:
6497 if (!seg6_bpf_has_valid_srh(skb))
6498 return -EBADMSG;
6499 if (param_len != sizeof(int))
6500 return -EINVAL;
6501 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6502 case SEG6_LOCAL_ACTION_END_DT6:
6503 if (!seg6_bpf_has_valid_srh(skb))
6504 return -EBADMSG;
6505 if (param_len != sizeof(int))
6506 return -EINVAL;
6507
6508 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6509 return -EBADMSG;
6510 if (!pskb_pull(skb, hdroff))
6511 return -EBADMSG;
6512
6513 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6514 skb_reset_network_header(skb);
6515 skb_reset_transport_header(skb);
6516 skb->encapsulation = 0;
6517
6518 bpf_compute_data_pointers(skb);
6519 bpf_update_srh_state(skb);
6520 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6521 case SEG6_LOCAL_ACTION_END_B6:
6522 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6523 return -EBADMSG;
6524 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6525 param, param_len);
6526 if (!err)
6527 bpf_update_srh_state(skb);
6528
6529 return err;
6530 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6531 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6532 return -EBADMSG;
6533 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6534 param, param_len);
6535 if (!err)
6536 bpf_update_srh_state(skb);
6537
6538 return err;
6539 default:
6540 return -EINVAL;
6541 }
6542 }
6543
6544 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6545 .func = bpf_lwt_seg6_action,
6546 .gpl_only = false,
6547 .ret_type = RET_INTEGER,
6548 .arg1_type = ARG_PTR_TO_CTX,
6549 .arg2_type = ARG_ANYTHING,
6550 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6551 .arg4_type = ARG_CONST_SIZE
6552 };
6553
BPF_CALL_3(bpf_lwt_seg6_adjust_srh,struct sk_buff *,skb,u32,offset,s32,len)6554 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6555 s32, len)
6556 {
6557 struct seg6_bpf_srh_state *srh_state =
6558 this_cpu_ptr(&seg6_bpf_srh_states);
6559 struct ipv6_sr_hdr *srh = srh_state->srh;
6560 void *srh_end, *srh_tlvs, *ptr;
6561 struct ipv6hdr *hdr;
6562 int srhoff = 0;
6563 int ret;
6564
6565 if (unlikely(srh == NULL))
6566 return -EINVAL;
6567
6568 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6569 ((srh->first_segment + 1) << 4));
6570 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6571 srh_state->hdrlen);
6572 ptr = skb->data + offset;
6573
6574 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6575 return -EFAULT;
6576 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6577 return -EFAULT;
6578
6579 if (len > 0) {
6580 ret = skb_cow_head(skb, len);
6581 if (unlikely(ret < 0))
6582 return ret;
6583
6584 ret = bpf_skb_net_hdr_push(skb, offset, len);
6585 } else {
6586 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6587 }
6588
6589 bpf_compute_data_pointers(skb);
6590 if (unlikely(ret < 0))
6591 return ret;
6592
6593 hdr = (struct ipv6hdr *)skb->data;
6594 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6595
6596 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6597 return -EINVAL;
6598 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6599 srh_state->hdrlen += len;
6600 srh_state->valid = false;
6601 return 0;
6602 }
6603
6604 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6605 .func = bpf_lwt_seg6_adjust_srh,
6606 .gpl_only = false,
6607 .ret_type = RET_INTEGER,
6608 .arg1_type = ARG_PTR_TO_CTX,
6609 .arg2_type = ARG_ANYTHING,
6610 .arg3_type = ARG_ANYTHING,
6611 };
6612 #endif /* CONFIG_IPV6_SEG6_BPF */
6613
6614 #ifdef CONFIG_INET
sk_lookup(struct net * net,struct bpf_sock_tuple * tuple,int dif,int sdif,u8 family,u8 proto)6615 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6616 int dif, int sdif, u8 family, u8 proto)
6617 {
6618 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6619 bool refcounted = false;
6620 struct sock *sk = NULL;
6621
6622 if (family == AF_INET) {
6623 __be32 src4 = tuple->ipv4.saddr;
6624 __be32 dst4 = tuple->ipv4.daddr;
6625
6626 if (proto == IPPROTO_TCP)
6627 sk = __inet_lookup(net, hinfo, NULL, 0,
6628 src4, tuple->ipv4.sport,
6629 dst4, tuple->ipv4.dport,
6630 dif, sdif, &refcounted);
6631 else
6632 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6633 dst4, tuple->ipv4.dport,
6634 dif, sdif, net->ipv4.udp_table, NULL);
6635 #if IS_ENABLED(CONFIG_IPV6)
6636 } else {
6637 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6638 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6639
6640 if (proto == IPPROTO_TCP)
6641 sk = __inet6_lookup(net, hinfo, NULL, 0,
6642 src6, tuple->ipv6.sport,
6643 dst6, ntohs(tuple->ipv6.dport),
6644 dif, sdif, &refcounted);
6645 else if (likely(ipv6_bpf_stub))
6646 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6647 src6, tuple->ipv6.sport,
6648 dst6, tuple->ipv6.dport,
6649 dif, sdif,
6650 net->ipv4.udp_table, NULL);
6651 #endif
6652 }
6653
6654 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6655 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6656 sk = NULL;
6657 }
6658 return sk;
6659 }
6660
6661 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6662 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6663 */
6664 static struct sock *
__bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6665 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6666 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6667 u64 flags, int sdif)
6668 {
6669 struct sock *sk = NULL;
6670 struct net *net;
6671 u8 family;
6672
6673 if (len == sizeof(tuple->ipv4))
6674 family = AF_INET;
6675 else if (len == sizeof(tuple->ipv6))
6676 family = AF_INET6;
6677 else
6678 return NULL;
6679
6680 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6681 goto out;
6682
6683 if (sdif < 0) {
6684 if (family == AF_INET)
6685 sdif = inet_sdif(skb);
6686 else
6687 sdif = inet6_sdif(skb);
6688 }
6689
6690 if ((s32)netns_id < 0) {
6691 net = caller_net;
6692 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6693 } else {
6694 net = get_net_ns_by_id(caller_net, netns_id);
6695 if (unlikely(!net))
6696 goto out;
6697 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6698 put_net(net);
6699 }
6700
6701 out:
6702 return sk;
6703 }
6704
6705 static struct sock *
__bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6706 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6707 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6708 u64 flags, int sdif)
6709 {
6710 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6711 ifindex, proto, netns_id, flags,
6712 sdif);
6713
6714 if (sk) {
6715 struct sock *sk2 = sk_to_full_sk(sk);
6716
6717 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6718 * sock refcnt is decremented to prevent a request_sock leak.
6719 */
6720 if (!sk_fullsock(sk2))
6721 sk2 = NULL;
6722 if (sk2 != sk) {
6723 sock_gen_put(sk);
6724 /* Ensure there is no need to bump sk2 refcnt */
6725 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6726 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6727 return NULL;
6728 }
6729 sk = sk2;
6730 }
6731 }
6732
6733 return sk;
6734 }
6735
6736 static struct sock *
bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6737 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6738 u8 proto, u64 netns_id, u64 flags)
6739 {
6740 struct net *caller_net;
6741 int ifindex;
6742
6743 if (skb->dev) {
6744 caller_net = dev_net(skb->dev);
6745 ifindex = skb->dev->ifindex;
6746 } else {
6747 caller_net = sock_net(skb->sk);
6748 ifindex = 0;
6749 }
6750
6751 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6752 netns_id, flags, -1);
6753 }
6754
6755 static struct sock *
bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6756 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6757 u8 proto, u64 netns_id, u64 flags)
6758 {
6759 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6760 flags);
6761
6762 if (sk) {
6763 struct sock *sk2 = sk_to_full_sk(sk);
6764
6765 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6766 * sock refcnt is decremented to prevent a request_sock leak.
6767 */
6768 if (!sk_fullsock(sk2))
6769 sk2 = NULL;
6770 if (sk2 != sk) {
6771 sock_gen_put(sk);
6772 /* Ensure there is no need to bump sk2 refcnt */
6773 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6774 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6775 return NULL;
6776 }
6777 sk = sk2;
6778 }
6779 }
6780
6781 return sk;
6782 }
6783
BPF_CALL_5(bpf_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6784 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6785 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6786 {
6787 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6788 netns_id, flags);
6789 }
6790
6791 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6792 .func = bpf_skc_lookup_tcp,
6793 .gpl_only = false,
6794 .pkt_access = true,
6795 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6796 .arg1_type = ARG_PTR_TO_CTX,
6797 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6798 .arg3_type = ARG_CONST_SIZE,
6799 .arg4_type = ARG_ANYTHING,
6800 .arg5_type = ARG_ANYTHING,
6801 };
6802
BPF_CALL_5(bpf_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6803 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6804 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6805 {
6806 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6807 netns_id, flags);
6808 }
6809
6810 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6811 .func = bpf_sk_lookup_tcp,
6812 .gpl_only = false,
6813 .pkt_access = true,
6814 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6815 .arg1_type = ARG_PTR_TO_CTX,
6816 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6817 .arg3_type = ARG_CONST_SIZE,
6818 .arg4_type = ARG_ANYTHING,
6819 .arg5_type = ARG_ANYTHING,
6820 };
6821
BPF_CALL_5(bpf_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6822 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6823 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6824 {
6825 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6826 netns_id, flags);
6827 }
6828
6829 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6830 .func = bpf_sk_lookup_udp,
6831 .gpl_only = false,
6832 .pkt_access = true,
6833 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6834 .arg1_type = ARG_PTR_TO_CTX,
6835 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6836 .arg3_type = ARG_CONST_SIZE,
6837 .arg4_type = ARG_ANYTHING,
6838 .arg5_type = ARG_ANYTHING,
6839 };
6840
BPF_CALL_5(bpf_tc_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6841 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6842 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6843 {
6844 struct net_device *dev = skb->dev;
6845 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6846 struct net *caller_net = dev_net(dev);
6847
6848 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6849 ifindex, IPPROTO_TCP, netns_id,
6850 flags, sdif);
6851 }
6852
6853 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6854 .func = bpf_tc_skc_lookup_tcp,
6855 .gpl_only = false,
6856 .pkt_access = true,
6857 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6858 .arg1_type = ARG_PTR_TO_CTX,
6859 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6860 .arg3_type = ARG_CONST_SIZE,
6861 .arg4_type = ARG_ANYTHING,
6862 .arg5_type = ARG_ANYTHING,
6863 };
6864
BPF_CALL_5(bpf_tc_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6865 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6866 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6867 {
6868 struct net_device *dev = skb->dev;
6869 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6870 struct net *caller_net = dev_net(dev);
6871
6872 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6873 ifindex, IPPROTO_TCP, netns_id,
6874 flags, sdif);
6875 }
6876
6877 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6878 .func = bpf_tc_sk_lookup_tcp,
6879 .gpl_only = false,
6880 .pkt_access = true,
6881 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6882 .arg1_type = ARG_PTR_TO_CTX,
6883 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6884 .arg3_type = ARG_CONST_SIZE,
6885 .arg4_type = ARG_ANYTHING,
6886 .arg5_type = ARG_ANYTHING,
6887 };
6888
BPF_CALL_5(bpf_tc_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6889 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6890 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6891 {
6892 struct net_device *dev = skb->dev;
6893 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6894 struct net *caller_net = dev_net(dev);
6895
6896 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6897 ifindex, IPPROTO_UDP, netns_id,
6898 flags, sdif);
6899 }
6900
6901 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6902 .func = bpf_tc_sk_lookup_udp,
6903 .gpl_only = false,
6904 .pkt_access = true,
6905 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6906 .arg1_type = ARG_PTR_TO_CTX,
6907 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6908 .arg3_type = ARG_CONST_SIZE,
6909 .arg4_type = ARG_ANYTHING,
6910 .arg5_type = ARG_ANYTHING,
6911 };
6912
BPF_CALL_1(bpf_sk_release,struct sock *,sk)6913 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6914 {
6915 if (sk && sk_is_refcounted(sk))
6916 sock_gen_put(sk);
6917 return 0;
6918 }
6919
6920 static const struct bpf_func_proto bpf_sk_release_proto = {
6921 .func = bpf_sk_release,
6922 .gpl_only = false,
6923 .ret_type = RET_INTEGER,
6924 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6925 };
6926
BPF_CALL_5(bpf_xdp_sk_lookup_udp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6927 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6928 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6929 {
6930 struct net_device *dev = ctx->rxq->dev;
6931 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6932 struct net *caller_net = dev_net(dev);
6933
6934 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6935 ifindex, IPPROTO_UDP, netns_id,
6936 flags, sdif);
6937 }
6938
6939 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6940 .func = bpf_xdp_sk_lookup_udp,
6941 .gpl_only = false,
6942 .pkt_access = true,
6943 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6944 .arg1_type = ARG_PTR_TO_CTX,
6945 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6946 .arg3_type = ARG_CONST_SIZE,
6947 .arg4_type = ARG_ANYTHING,
6948 .arg5_type = ARG_ANYTHING,
6949 };
6950
BPF_CALL_5(bpf_xdp_skc_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6951 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6952 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6953 {
6954 struct net_device *dev = ctx->rxq->dev;
6955 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6956 struct net *caller_net = dev_net(dev);
6957
6958 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6959 ifindex, IPPROTO_TCP, netns_id,
6960 flags, sdif);
6961 }
6962
6963 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6964 .func = bpf_xdp_skc_lookup_tcp,
6965 .gpl_only = false,
6966 .pkt_access = true,
6967 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6968 .arg1_type = ARG_PTR_TO_CTX,
6969 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6970 .arg3_type = ARG_CONST_SIZE,
6971 .arg4_type = ARG_ANYTHING,
6972 .arg5_type = ARG_ANYTHING,
6973 };
6974
BPF_CALL_5(bpf_xdp_sk_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6975 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6976 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6977 {
6978 struct net_device *dev = ctx->rxq->dev;
6979 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6980 struct net *caller_net = dev_net(dev);
6981
6982 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6983 ifindex, IPPROTO_TCP, netns_id,
6984 flags, sdif);
6985 }
6986
6987 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6988 .func = bpf_xdp_sk_lookup_tcp,
6989 .gpl_only = false,
6990 .pkt_access = true,
6991 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6992 .arg1_type = ARG_PTR_TO_CTX,
6993 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6994 .arg3_type = ARG_CONST_SIZE,
6995 .arg4_type = ARG_ANYTHING,
6996 .arg5_type = ARG_ANYTHING,
6997 };
6998
BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6999 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7000 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7001 {
7002 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
7003 sock_net(ctx->sk), 0,
7004 IPPROTO_TCP, netns_id, flags,
7005 -1);
7006 }
7007
7008 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
7009 .func = bpf_sock_addr_skc_lookup_tcp,
7010 .gpl_only = false,
7011 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7012 .arg1_type = ARG_PTR_TO_CTX,
7013 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7014 .arg3_type = ARG_CONST_SIZE,
7015 .arg4_type = ARG_ANYTHING,
7016 .arg5_type = ARG_ANYTHING,
7017 };
7018
BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7019 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7020 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7021 {
7022 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7023 sock_net(ctx->sk), 0, IPPROTO_TCP,
7024 netns_id, flags, -1);
7025 }
7026
7027 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
7028 .func = bpf_sock_addr_sk_lookup_tcp,
7029 .gpl_only = false,
7030 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7031 .arg1_type = ARG_PTR_TO_CTX,
7032 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7033 .arg3_type = ARG_CONST_SIZE,
7034 .arg4_type = ARG_ANYTHING,
7035 .arg5_type = ARG_ANYTHING,
7036 };
7037
BPF_CALL_5(bpf_sock_addr_sk_lookup_udp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7038 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
7039 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7040 {
7041 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7042 sock_net(ctx->sk), 0, IPPROTO_UDP,
7043 netns_id, flags, -1);
7044 }
7045
7046 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
7047 .func = bpf_sock_addr_sk_lookup_udp,
7048 .gpl_only = false,
7049 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7050 .arg1_type = ARG_PTR_TO_CTX,
7051 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7052 .arg3_type = ARG_CONST_SIZE,
7053 .arg4_type = ARG_ANYTHING,
7054 .arg5_type = ARG_ANYTHING,
7055 };
7056
bpf_tcp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7057 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7058 struct bpf_insn_access_aux *info)
7059 {
7060 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
7061 icsk_retransmits))
7062 return false;
7063
7064 if (off % size != 0)
7065 return false;
7066
7067 switch (off) {
7068 case offsetof(struct bpf_tcp_sock, bytes_received):
7069 case offsetof(struct bpf_tcp_sock, bytes_acked):
7070 return size == sizeof(__u64);
7071 default:
7072 return size == sizeof(__u32);
7073 }
7074 }
7075
bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7076 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
7077 const struct bpf_insn *si,
7078 struct bpf_insn *insn_buf,
7079 struct bpf_prog *prog, u32 *target_size)
7080 {
7081 struct bpf_insn *insn = insn_buf;
7082
7083 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
7084 do { \
7085 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
7086 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7087 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7088 si->dst_reg, si->src_reg, \
7089 offsetof(struct tcp_sock, FIELD)); \
7090 } while (0)
7091
7092 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
7093 do { \
7094 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
7095 FIELD) > \
7096 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7097 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7098 struct inet_connection_sock, \
7099 FIELD), \
7100 si->dst_reg, si->src_reg, \
7101 offsetof( \
7102 struct inet_connection_sock, \
7103 FIELD)); \
7104 } while (0)
7105
7106 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7107
7108 switch (si->off) {
7109 case offsetof(struct bpf_tcp_sock, rtt_min):
7110 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7111 sizeof(struct minmax));
7112 BUILD_BUG_ON(sizeof(struct minmax) <
7113 sizeof(struct minmax_sample));
7114
7115 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7116 offsetof(struct tcp_sock, rtt_min) +
7117 offsetof(struct minmax_sample, v));
7118 break;
7119 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7120 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7121 break;
7122 case offsetof(struct bpf_tcp_sock, srtt_us):
7123 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7124 break;
7125 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7126 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7127 break;
7128 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7129 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7130 break;
7131 case offsetof(struct bpf_tcp_sock, snd_nxt):
7132 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7133 break;
7134 case offsetof(struct bpf_tcp_sock, snd_una):
7135 BPF_TCP_SOCK_GET_COMMON(snd_una);
7136 break;
7137 case offsetof(struct bpf_tcp_sock, mss_cache):
7138 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7139 break;
7140 case offsetof(struct bpf_tcp_sock, ecn_flags):
7141 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7142 break;
7143 case offsetof(struct bpf_tcp_sock, rate_delivered):
7144 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7145 break;
7146 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7147 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7148 break;
7149 case offsetof(struct bpf_tcp_sock, packets_out):
7150 BPF_TCP_SOCK_GET_COMMON(packets_out);
7151 break;
7152 case offsetof(struct bpf_tcp_sock, retrans_out):
7153 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7154 break;
7155 case offsetof(struct bpf_tcp_sock, total_retrans):
7156 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7157 break;
7158 case offsetof(struct bpf_tcp_sock, segs_in):
7159 BPF_TCP_SOCK_GET_COMMON(segs_in);
7160 break;
7161 case offsetof(struct bpf_tcp_sock, data_segs_in):
7162 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7163 break;
7164 case offsetof(struct bpf_tcp_sock, segs_out):
7165 BPF_TCP_SOCK_GET_COMMON(segs_out);
7166 break;
7167 case offsetof(struct bpf_tcp_sock, data_segs_out):
7168 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7169 break;
7170 case offsetof(struct bpf_tcp_sock, lost_out):
7171 BPF_TCP_SOCK_GET_COMMON(lost_out);
7172 break;
7173 case offsetof(struct bpf_tcp_sock, sacked_out):
7174 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7175 break;
7176 case offsetof(struct bpf_tcp_sock, bytes_received):
7177 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7178 break;
7179 case offsetof(struct bpf_tcp_sock, bytes_acked):
7180 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7181 break;
7182 case offsetof(struct bpf_tcp_sock, dsack_dups):
7183 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7184 break;
7185 case offsetof(struct bpf_tcp_sock, delivered):
7186 BPF_TCP_SOCK_GET_COMMON(delivered);
7187 break;
7188 case offsetof(struct bpf_tcp_sock, delivered_ce):
7189 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7190 break;
7191 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7192 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7193 break;
7194 }
7195
7196 return insn - insn_buf;
7197 }
7198
BPF_CALL_1(bpf_tcp_sock,struct sock *,sk)7199 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7200 {
7201 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7202 return (unsigned long)sk;
7203
7204 return (unsigned long)NULL;
7205 }
7206
7207 const struct bpf_func_proto bpf_tcp_sock_proto = {
7208 .func = bpf_tcp_sock,
7209 .gpl_only = false,
7210 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7211 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7212 };
7213
BPF_CALL_1(bpf_get_listener_sock,struct sock *,sk)7214 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7215 {
7216 sk = sk_to_full_sk(sk);
7217
7218 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7219 return (unsigned long)sk;
7220
7221 return (unsigned long)NULL;
7222 }
7223
7224 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7225 .func = bpf_get_listener_sock,
7226 .gpl_only = false,
7227 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7228 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7229 };
7230
BPF_CALL_1(bpf_skb_ecn_set_ce,struct sk_buff *,skb)7231 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7232 {
7233 unsigned int iphdr_len;
7234
7235 switch (skb_protocol(skb, true)) {
7236 case cpu_to_be16(ETH_P_IP):
7237 iphdr_len = sizeof(struct iphdr);
7238 break;
7239 case cpu_to_be16(ETH_P_IPV6):
7240 iphdr_len = sizeof(struct ipv6hdr);
7241 break;
7242 default:
7243 return 0;
7244 }
7245
7246 if (skb_headlen(skb) < iphdr_len)
7247 return 0;
7248
7249 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7250 return 0;
7251
7252 return INET_ECN_set_ce(skb);
7253 }
7254
bpf_xdp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7255 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7256 struct bpf_insn_access_aux *info)
7257 {
7258 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7259 return false;
7260
7261 if (off % size != 0)
7262 return false;
7263
7264 switch (off) {
7265 default:
7266 return size == sizeof(__u32);
7267 }
7268 }
7269
bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7270 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7271 const struct bpf_insn *si,
7272 struct bpf_insn *insn_buf,
7273 struct bpf_prog *prog, u32 *target_size)
7274 {
7275 struct bpf_insn *insn = insn_buf;
7276
7277 #define BPF_XDP_SOCK_GET(FIELD) \
7278 do { \
7279 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7280 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7281 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7282 si->dst_reg, si->src_reg, \
7283 offsetof(struct xdp_sock, FIELD)); \
7284 } while (0)
7285
7286 switch (si->off) {
7287 case offsetof(struct bpf_xdp_sock, queue_id):
7288 BPF_XDP_SOCK_GET(queue_id);
7289 break;
7290 }
7291
7292 return insn - insn_buf;
7293 }
7294
7295 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7296 .func = bpf_skb_ecn_set_ce,
7297 .gpl_only = false,
7298 .ret_type = RET_INTEGER,
7299 .arg1_type = ARG_PTR_TO_CTX,
7300 };
7301
BPF_CALL_5(bpf_tcp_check_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7302 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7303 struct tcphdr *, th, u32, th_len)
7304 {
7305 #ifdef CONFIG_SYN_COOKIES
7306 u32 cookie;
7307 int ret;
7308
7309 if (unlikely(!sk || th_len < sizeof(*th)))
7310 return -EINVAL;
7311
7312 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7313 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7314 return -EINVAL;
7315
7316 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7317 return -EINVAL;
7318
7319 if (!th->ack || th->rst || th->syn)
7320 return -ENOENT;
7321
7322 if (unlikely(iph_len < sizeof(struct iphdr)))
7323 return -EINVAL;
7324
7325 if (tcp_synq_no_recent_overflow(sk))
7326 return -ENOENT;
7327
7328 cookie = ntohl(th->ack_seq) - 1;
7329
7330 /* Both struct iphdr and struct ipv6hdr have the version field at the
7331 * same offset so we can cast to the shorter header (struct iphdr).
7332 */
7333 switch (((struct iphdr *)iph)->version) {
7334 case 4:
7335 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7336 return -EINVAL;
7337
7338 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7339 break;
7340
7341 #if IS_BUILTIN(CONFIG_IPV6)
7342 case 6:
7343 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7344 return -EINVAL;
7345
7346 if (sk->sk_family != AF_INET6)
7347 return -EINVAL;
7348
7349 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7350 break;
7351 #endif /* CONFIG_IPV6 */
7352
7353 default:
7354 return -EPROTONOSUPPORT;
7355 }
7356
7357 if (ret > 0)
7358 return 0;
7359
7360 return -ENOENT;
7361 #else
7362 return -ENOTSUPP;
7363 #endif
7364 }
7365
7366 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7367 .func = bpf_tcp_check_syncookie,
7368 .gpl_only = true,
7369 .pkt_access = true,
7370 .ret_type = RET_INTEGER,
7371 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7372 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7373 .arg3_type = ARG_CONST_SIZE,
7374 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7375 .arg5_type = ARG_CONST_SIZE,
7376 };
7377
BPF_CALL_5(bpf_tcp_gen_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7378 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7379 struct tcphdr *, th, u32, th_len)
7380 {
7381 #ifdef CONFIG_SYN_COOKIES
7382 u32 cookie;
7383 u16 mss;
7384
7385 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7386 return -EINVAL;
7387
7388 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7389 return -EINVAL;
7390
7391 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7392 return -ENOENT;
7393
7394 if (!th->syn || th->ack || th->fin || th->rst)
7395 return -EINVAL;
7396
7397 if (unlikely(iph_len < sizeof(struct iphdr)))
7398 return -EINVAL;
7399
7400 /* Both struct iphdr and struct ipv6hdr have the version field at the
7401 * same offset so we can cast to the shorter header (struct iphdr).
7402 */
7403 switch (((struct iphdr *)iph)->version) {
7404 case 4:
7405 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7406 return -EINVAL;
7407
7408 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7409 break;
7410
7411 #if IS_BUILTIN(CONFIG_IPV6)
7412 case 6:
7413 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7414 return -EINVAL;
7415
7416 if (sk->sk_family != AF_INET6)
7417 return -EINVAL;
7418
7419 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7420 break;
7421 #endif /* CONFIG_IPV6 */
7422
7423 default:
7424 return -EPROTONOSUPPORT;
7425 }
7426 if (mss == 0)
7427 return -ENOENT;
7428
7429 return cookie | ((u64)mss << 32);
7430 #else
7431 return -EOPNOTSUPP;
7432 #endif /* CONFIG_SYN_COOKIES */
7433 }
7434
7435 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7436 .func = bpf_tcp_gen_syncookie,
7437 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7438 .pkt_access = true,
7439 .ret_type = RET_INTEGER,
7440 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7441 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7442 .arg3_type = ARG_CONST_SIZE,
7443 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7444 .arg5_type = ARG_CONST_SIZE,
7445 };
7446
BPF_CALL_3(bpf_sk_assign,struct sk_buff *,skb,struct sock *,sk,u64,flags)7447 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7448 {
7449 if (!sk || flags != 0)
7450 return -EINVAL;
7451 if (!skb_at_tc_ingress(skb))
7452 return -EOPNOTSUPP;
7453 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7454 return -ENETUNREACH;
7455 if (sk_unhashed(sk))
7456 return -EOPNOTSUPP;
7457 if (sk_is_refcounted(sk) &&
7458 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7459 return -ENOENT;
7460
7461 skb_orphan(skb);
7462 skb->sk = sk;
7463 skb->destructor = sock_pfree;
7464
7465 return 0;
7466 }
7467
7468 static const struct bpf_func_proto bpf_sk_assign_proto = {
7469 .func = bpf_sk_assign,
7470 .gpl_only = false,
7471 .ret_type = RET_INTEGER,
7472 .arg1_type = ARG_PTR_TO_CTX,
7473 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7474 .arg3_type = ARG_ANYTHING,
7475 };
7476
bpf_search_tcp_opt(const u8 * op,const u8 * opend,u8 search_kind,const u8 * magic,u8 magic_len,bool * eol)7477 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7478 u8 search_kind, const u8 *magic,
7479 u8 magic_len, bool *eol)
7480 {
7481 u8 kind, kind_len;
7482
7483 *eol = false;
7484
7485 while (op < opend) {
7486 kind = op[0];
7487
7488 if (kind == TCPOPT_EOL) {
7489 *eol = true;
7490 return ERR_PTR(-ENOMSG);
7491 } else if (kind == TCPOPT_NOP) {
7492 op++;
7493 continue;
7494 }
7495
7496 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7497 /* Something is wrong in the received header.
7498 * Follow the TCP stack's tcp_parse_options()
7499 * and just bail here.
7500 */
7501 return ERR_PTR(-EFAULT);
7502
7503 kind_len = op[1];
7504 if (search_kind == kind) {
7505 if (!magic_len)
7506 return op;
7507
7508 if (magic_len > kind_len - 2)
7509 return ERR_PTR(-ENOMSG);
7510
7511 if (!memcmp(&op[2], magic, magic_len))
7512 return op;
7513 }
7514
7515 op += kind_len;
7516 }
7517
7518 return ERR_PTR(-ENOMSG);
7519 }
7520
BPF_CALL_4(bpf_sock_ops_load_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,void *,search_res,u32,len,u64,flags)7521 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7522 void *, search_res, u32, len, u64, flags)
7523 {
7524 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7525 const u8 *op, *opend, *magic, *search = search_res;
7526 u8 search_kind, search_len, copy_len, magic_len;
7527 int ret;
7528
7529 /* 2 byte is the minimal option len except TCPOPT_NOP and
7530 * TCPOPT_EOL which are useless for the bpf prog to learn
7531 * and this helper disallow loading them also.
7532 */
7533 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7534 return -EINVAL;
7535
7536 search_kind = search[0];
7537 search_len = search[1];
7538
7539 if (search_len > len || search_kind == TCPOPT_NOP ||
7540 search_kind == TCPOPT_EOL)
7541 return -EINVAL;
7542
7543 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7544 /* 16 or 32 bit magic. +2 for kind and kind length */
7545 if (search_len != 4 && search_len != 6)
7546 return -EINVAL;
7547 magic = &search[2];
7548 magic_len = search_len - 2;
7549 } else {
7550 if (search_len)
7551 return -EINVAL;
7552 magic = NULL;
7553 magic_len = 0;
7554 }
7555
7556 if (load_syn) {
7557 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7558 if (ret < 0)
7559 return ret;
7560
7561 opend = op + ret;
7562 op += sizeof(struct tcphdr);
7563 } else {
7564 if (!bpf_sock->skb ||
7565 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7566 /* This bpf_sock->op cannot call this helper */
7567 return -EPERM;
7568
7569 opend = bpf_sock->skb_data_end;
7570 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7571 }
7572
7573 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7574 &eol);
7575 if (IS_ERR(op))
7576 return PTR_ERR(op);
7577
7578 copy_len = op[1];
7579 ret = copy_len;
7580 if (copy_len > len) {
7581 ret = -ENOSPC;
7582 copy_len = len;
7583 }
7584
7585 memcpy(search_res, op, copy_len);
7586 return ret;
7587 }
7588
7589 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7590 .func = bpf_sock_ops_load_hdr_opt,
7591 .gpl_only = false,
7592 .ret_type = RET_INTEGER,
7593 .arg1_type = ARG_PTR_TO_CTX,
7594 .arg2_type = ARG_PTR_TO_MEM,
7595 .arg3_type = ARG_CONST_SIZE,
7596 .arg4_type = ARG_ANYTHING,
7597 };
7598
BPF_CALL_4(bpf_sock_ops_store_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,const void *,from,u32,len,u64,flags)7599 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7600 const void *, from, u32, len, u64, flags)
7601 {
7602 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7603 const u8 *op, *new_op, *magic = NULL;
7604 struct sk_buff *skb;
7605 bool eol;
7606
7607 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7608 return -EPERM;
7609
7610 if (len < 2 || flags)
7611 return -EINVAL;
7612
7613 new_op = from;
7614 new_kind = new_op[0];
7615 new_kind_len = new_op[1];
7616
7617 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7618 new_kind == TCPOPT_EOL)
7619 return -EINVAL;
7620
7621 if (new_kind_len > bpf_sock->remaining_opt_len)
7622 return -ENOSPC;
7623
7624 /* 253 is another experimental kind */
7625 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7626 if (new_kind_len < 4)
7627 return -EINVAL;
7628 /* Match for the 2 byte magic also.
7629 * RFC 6994: the magic could be 2 or 4 bytes.
7630 * Hence, matching by 2 byte only is on the
7631 * conservative side but it is the right
7632 * thing to do for the 'search-for-duplication'
7633 * purpose.
7634 */
7635 magic = &new_op[2];
7636 magic_len = 2;
7637 }
7638
7639 /* Check for duplication */
7640 skb = bpf_sock->skb;
7641 op = skb->data + sizeof(struct tcphdr);
7642 opend = bpf_sock->skb_data_end;
7643
7644 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7645 &eol);
7646 if (!IS_ERR(op))
7647 return -EEXIST;
7648
7649 if (PTR_ERR(op) != -ENOMSG)
7650 return PTR_ERR(op);
7651
7652 if (eol)
7653 /* The option has been ended. Treat it as no more
7654 * header option can be written.
7655 */
7656 return -ENOSPC;
7657
7658 /* No duplication found. Store the header option. */
7659 memcpy(opend, from, new_kind_len);
7660
7661 bpf_sock->remaining_opt_len -= new_kind_len;
7662 bpf_sock->skb_data_end += new_kind_len;
7663
7664 return 0;
7665 }
7666
7667 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7668 .func = bpf_sock_ops_store_hdr_opt,
7669 .gpl_only = false,
7670 .ret_type = RET_INTEGER,
7671 .arg1_type = ARG_PTR_TO_CTX,
7672 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7673 .arg3_type = ARG_CONST_SIZE,
7674 .arg4_type = ARG_ANYTHING,
7675 };
7676
BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,u32,len,u64,flags)7677 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7678 u32, len, u64, flags)
7679 {
7680 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7681 return -EPERM;
7682
7683 if (flags || len < 2)
7684 return -EINVAL;
7685
7686 if (len > bpf_sock->remaining_opt_len)
7687 return -ENOSPC;
7688
7689 bpf_sock->remaining_opt_len -= len;
7690
7691 return 0;
7692 }
7693
7694 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7695 .func = bpf_sock_ops_reserve_hdr_opt,
7696 .gpl_only = false,
7697 .ret_type = RET_INTEGER,
7698 .arg1_type = ARG_PTR_TO_CTX,
7699 .arg2_type = ARG_ANYTHING,
7700 .arg3_type = ARG_ANYTHING,
7701 };
7702
BPF_CALL_3(bpf_skb_set_tstamp,struct sk_buff *,skb,u64,tstamp,u32,tstamp_type)7703 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7704 u64, tstamp, u32, tstamp_type)
7705 {
7706 /* skb_clear_delivery_time() is done for inet protocol */
7707 if (skb->protocol != htons(ETH_P_IP) &&
7708 skb->protocol != htons(ETH_P_IPV6))
7709 return -EOPNOTSUPP;
7710
7711 switch (tstamp_type) {
7712 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7713 if (!tstamp)
7714 return -EINVAL;
7715 skb->tstamp = tstamp;
7716 skb->mono_delivery_time = 1;
7717 break;
7718 case BPF_SKB_TSTAMP_UNSPEC:
7719 if (tstamp)
7720 return -EINVAL;
7721 skb->tstamp = 0;
7722 skb->mono_delivery_time = 0;
7723 break;
7724 default:
7725 return -EINVAL;
7726 }
7727
7728 return 0;
7729 }
7730
7731 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7732 .func = bpf_skb_set_tstamp,
7733 .gpl_only = false,
7734 .ret_type = RET_INTEGER,
7735 .arg1_type = ARG_PTR_TO_CTX,
7736 .arg2_type = ARG_ANYTHING,
7737 .arg3_type = ARG_ANYTHING,
7738 };
7739
7740 #ifdef CONFIG_SYN_COOKIES
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th,u32,th_len)7741 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7742 struct tcphdr *, th, u32, th_len)
7743 {
7744 u32 cookie;
7745 u16 mss;
7746
7747 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7748 return -EINVAL;
7749
7750 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7751 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7752
7753 return cookie | ((u64)mss << 32);
7754 }
7755
7756 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7757 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7758 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7759 .pkt_access = true,
7760 .ret_type = RET_INTEGER,
7761 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7762 .arg1_size = sizeof(struct iphdr),
7763 .arg2_type = ARG_PTR_TO_MEM,
7764 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7765 };
7766
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th,u32,th_len)7767 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7768 struct tcphdr *, th, u32, th_len)
7769 {
7770 #if IS_BUILTIN(CONFIG_IPV6)
7771 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7772 sizeof(struct ipv6hdr);
7773 u32 cookie;
7774 u16 mss;
7775
7776 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7777 return -EINVAL;
7778
7779 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7780 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7781
7782 return cookie | ((u64)mss << 32);
7783 #else
7784 return -EPROTONOSUPPORT;
7785 #endif
7786 }
7787
7788 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7789 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7790 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7791 .pkt_access = true,
7792 .ret_type = RET_INTEGER,
7793 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7794 .arg1_size = sizeof(struct ipv6hdr),
7795 .arg2_type = ARG_PTR_TO_MEM,
7796 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7797 };
7798
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th)7799 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7800 struct tcphdr *, th)
7801 {
7802 u32 cookie = ntohl(th->ack_seq) - 1;
7803
7804 if (__cookie_v4_check(iph, th, cookie) > 0)
7805 return 0;
7806
7807 return -EACCES;
7808 }
7809
7810 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7811 .func = bpf_tcp_raw_check_syncookie_ipv4,
7812 .gpl_only = true, /* __cookie_v4_check is GPL */
7813 .pkt_access = true,
7814 .ret_type = RET_INTEGER,
7815 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7816 .arg1_size = sizeof(struct iphdr),
7817 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7818 .arg2_size = sizeof(struct tcphdr),
7819 };
7820
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th)7821 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7822 struct tcphdr *, th)
7823 {
7824 #if IS_BUILTIN(CONFIG_IPV6)
7825 u32 cookie = ntohl(th->ack_seq) - 1;
7826
7827 if (__cookie_v6_check(iph, th, cookie) > 0)
7828 return 0;
7829
7830 return -EACCES;
7831 #else
7832 return -EPROTONOSUPPORT;
7833 #endif
7834 }
7835
7836 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7837 .func = bpf_tcp_raw_check_syncookie_ipv6,
7838 .gpl_only = true, /* __cookie_v6_check is GPL */
7839 .pkt_access = true,
7840 .ret_type = RET_INTEGER,
7841 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7842 .arg1_size = sizeof(struct ipv6hdr),
7843 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7844 .arg2_size = sizeof(struct tcphdr),
7845 };
7846 #endif /* CONFIG_SYN_COOKIES */
7847
7848 #endif /* CONFIG_INET */
7849
bpf_helper_changes_pkt_data(void * func)7850 bool bpf_helper_changes_pkt_data(void *func)
7851 {
7852 if (func == bpf_skb_vlan_push ||
7853 func == bpf_skb_vlan_pop ||
7854 func == bpf_skb_store_bytes ||
7855 func == bpf_skb_change_proto ||
7856 func == bpf_skb_change_head ||
7857 func == sk_skb_change_head ||
7858 func == bpf_skb_change_tail ||
7859 func == sk_skb_change_tail ||
7860 func == bpf_skb_adjust_room ||
7861 func == sk_skb_adjust_room ||
7862 func == bpf_skb_pull_data ||
7863 func == sk_skb_pull_data ||
7864 func == bpf_clone_redirect ||
7865 func == bpf_l3_csum_replace ||
7866 func == bpf_l4_csum_replace ||
7867 func == bpf_xdp_adjust_head ||
7868 func == bpf_xdp_adjust_meta ||
7869 func == bpf_msg_pull_data ||
7870 func == bpf_msg_push_data ||
7871 func == bpf_msg_pop_data ||
7872 func == bpf_xdp_adjust_tail ||
7873 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7874 func == bpf_lwt_seg6_store_bytes ||
7875 func == bpf_lwt_seg6_adjust_srh ||
7876 func == bpf_lwt_seg6_action ||
7877 #endif
7878 #ifdef CONFIG_INET
7879 func == bpf_sock_ops_store_hdr_opt ||
7880 #endif
7881 func == bpf_lwt_in_push_encap ||
7882 func == bpf_lwt_xmit_push_encap)
7883 return true;
7884
7885 return false;
7886 }
7887
7888 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7889 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7890
7891 static const struct bpf_func_proto *
sock_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7892 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7893 {
7894 const struct bpf_func_proto *func_proto;
7895
7896 func_proto = cgroup_common_func_proto(func_id, prog);
7897 if (func_proto)
7898 return func_proto;
7899
7900 func_proto = cgroup_current_func_proto(func_id, prog);
7901 if (func_proto)
7902 return func_proto;
7903
7904 switch (func_id) {
7905 case BPF_FUNC_get_socket_cookie:
7906 return &bpf_get_socket_cookie_sock_proto;
7907 case BPF_FUNC_get_netns_cookie:
7908 return &bpf_get_netns_cookie_sock_proto;
7909 case BPF_FUNC_perf_event_output:
7910 return &bpf_event_output_data_proto;
7911 case BPF_FUNC_sk_storage_get:
7912 return &bpf_sk_storage_get_cg_sock_proto;
7913 case BPF_FUNC_ktime_get_coarse_ns:
7914 return &bpf_ktime_get_coarse_ns_proto;
7915 default:
7916 return bpf_base_func_proto(func_id);
7917 }
7918 }
7919
7920 static const struct bpf_func_proto *
sock_addr_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7921 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7922 {
7923 const struct bpf_func_proto *func_proto;
7924
7925 func_proto = cgroup_common_func_proto(func_id, prog);
7926 if (func_proto)
7927 return func_proto;
7928
7929 func_proto = cgroup_current_func_proto(func_id, prog);
7930 if (func_proto)
7931 return func_proto;
7932
7933 switch (func_id) {
7934 case BPF_FUNC_bind:
7935 switch (prog->expected_attach_type) {
7936 case BPF_CGROUP_INET4_CONNECT:
7937 case BPF_CGROUP_INET6_CONNECT:
7938 return &bpf_bind_proto;
7939 default:
7940 return NULL;
7941 }
7942 case BPF_FUNC_get_socket_cookie:
7943 return &bpf_get_socket_cookie_sock_addr_proto;
7944 case BPF_FUNC_get_netns_cookie:
7945 return &bpf_get_netns_cookie_sock_addr_proto;
7946 case BPF_FUNC_perf_event_output:
7947 return &bpf_event_output_data_proto;
7948 #ifdef CONFIG_INET
7949 case BPF_FUNC_sk_lookup_tcp:
7950 return &bpf_sock_addr_sk_lookup_tcp_proto;
7951 case BPF_FUNC_sk_lookup_udp:
7952 return &bpf_sock_addr_sk_lookup_udp_proto;
7953 case BPF_FUNC_sk_release:
7954 return &bpf_sk_release_proto;
7955 case BPF_FUNC_skc_lookup_tcp:
7956 return &bpf_sock_addr_skc_lookup_tcp_proto;
7957 #endif /* CONFIG_INET */
7958 case BPF_FUNC_sk_storage_get:
7959 return &bpf_sk_storage_get_proto;
7960 case BPF_FUNC_sk_storage_delete:
7961 return &bpf_sk_storage_delete_proto;
7962 case BPF_FUNC_setsockopt:
7963 switch (prog->expected_attach_type) {
7964 case BPF_CGROUP_INET4_BIND:
7965 case BPF_CGROUP_INET6_BIND:
7966 case BPF_CGROUP_INET4_CONNECT:
7967 case BPF_CGROUP_INET6_CONNECT:
7968 case BPF_CGROUP_UDP4_RECVMSG:
7969 case BPF_CGROUP_UDP6_RECVMSG:
7970 case BPF_CGROUP_UDP4_SENDMSG:
7971 case BPF_CGROUP_UDP6_SENDMSG:
7972 case BPF_CGROUP_INET4_GETPEERNAME:
7973 case BPF_CGROUP_INET6_GETPEERNAME:
7974 case BPF_CGROUP_INET4_GETSOCKNAME:
7975 case BPF_CGROUP_INET6_GETSOCKNAME:
7976 return &bpf_sock_addr_setsockopt_proto;
7977 default:
7978 return NULL;
7979 }
7980 case BPF_FUNC_getsockopt:
7981 switch (prog->expected_attach_type) {
7982 case BPF_CGROUP_INET4_BIND:
7983 case BPF_CGROUP_INET6_BIND:
7984 case BPF_CGROUP_INET4_CONNECT:
7985 case BPF_CGROUP_INET6_CONNECT:
7986 case BPF_CGROUP_UDP4_RECVMSG:
7987 case BPF_CGROUP_UDP6_RECVMSG:
7988 case BPF_CGROUP_UDP4_SENDMSG:
7989 case BPF_CGROUP_UDP6_SENDMSG:
7990 case BPF_CGROUP_INET4_GETPEERNAME:
7991 case BPF_CGROUP_INET6_GETPEERNAME:
7992 case BPF_CGROUP_INET4_GETSOCKNAME:
7993 case BPF_CGROUP_INET6_GETSOCKNAME:
7994 return &bpf_sock_addr_getsockopt_proto;
7995 default:
7996 return NULL;
7997 }
7998 default:
7999 return bpf_sk_base_func_proto(func_id);
8000 }
8001 }
8002
8003 static const struct bpf_func_proto *
sk_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8004 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8005 {
8006 switch (func_id) {
8007 case BPF_FUNC_skb_load_bytes:
8008 return &bpf_skb_load_bytes_proto;
8009 case BPF_FUNC_skb_load_bytes_relative:
8010 return &bpf_skb_load_bytes_relative_proto;
8011 case BPF_FUNC_get_socket_cookie:
8012 return &bpf_get_socket_cookie_proto;
8013 case BPF_FUNC_get_socket_uid:
8014 return &bpf_get_socket_uid_proto;
8015 case BPF_FUNC_perf_event_output:
8016 return &bpf_skb_event_output_proto;
8017 default:
8018 return bpf_sk_base_func_proto(func_id);
8019 }
8020 }
8021
8022 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
8023 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
8024
8025 static const struct bpf_func_proto *
cg_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8026 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8027 {
8028 const struct bpf_func_proto *func_proto;
8029
8030 func_proto = cgroup_common_func_proto(func_id, prog);
8031 if (func_proto)
8032 return func_proto;
8033
8034 switch (func_id) {
8035 case BPF_FUNC_sk_fullsock:
8036 return &bpf_sk_fullsock_proto;
8037 case BPF_FUNC_sk_storage_get:
8038 return &bpf_sk_storage_get_proto;
8039 case BPF_FUNC_sk_storage_delete:
8040 return &bpf_sk_storage_delete_proto;
8041 case BPF_FUNC_perf_event_output:
8042 return &bpf_skb_event_output_proto;
8043 #ifdef CONFIG_SOCK_CGROUP_DATA
8044 case BPF_FUNC_skb_cgroup_id:
8045 return &bpf_skb_cgroup_id_proto;
8046 case BPF_FUNC_skb_ancestor_cgroup_id:
8047 return &bpf_skb_ancestor_cgroup_id_proto;
8048 case BPF_FUNC_sk_cgroup_id:
8049 return &bpf_sk_cgroup_id_proto;
8050 case BPF_FUNC_sk_ancestor_cgroup_id:
8051 return &bpf_sk_ancestor_cgroup_id_proto;
8052 #endif
8053 #ifdef CONFIG_INET
8054 case BPF_FUNC_sk_lookup_tcp:
8055 return &bpf_sk_lookup_tcp_proto;
8056 case BPF_FUNC_sk_lookup_udp:
8057 return &bpf_sk_lookup_udp_proto;
8058 case BPF_FUNC_sk_release:
8059 return &bpf_sk_release_proto;
8060 case BPF_FUNC_skc_lookup_tcp:
8061 return &bpf_skc_lookup_tcp_proto;
8062 case BPF_FUNC_tcp_sock:
8063 return &bpf_tcp_sock_proto;
8064 case BPF_FUNC_get_listener_sock:
8065 return &bpf_get_listener_sock_proto;
8066 case BPF_FUNC_skb_ecn_set_ce:
8067 return &bpf_skb_ecn_set_ce_proto;
8068 #endif
8069 default:
8070 return sk_filter_func_proto(func_id, prog);
8071 }
8072 }
8073
8074 static const struct bpf_func_proto *
tc_cls_act_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8075 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8076 {
8077 switch (func_id) {
8078 case BPF_FUNC_skb_store_bytes:
8079 return &bpf_skb_store_bytes_proto;
8080 case BPF_FUNC_skb_load_bytes:
8081 return &bpf_skb_load_bytes_proto;
8082 case BPF_FUNC_skb_load_bytes_relative:
8083 return &bpf_skb_load_bytes_relative_proto;
8084 case BPF_FUNC_skb_pull_data:
8085 return &bpf_skb_pull_data_proto;
8086 case BPF_FUNC_csum_diff:
8087 return &bpf_csum_diff_proto;
8088 case BPF_FUNC_csum_update:
8089 return &bpf_csum_update_proto;
8090 case BPF_FUNC_csum_level:
8091 return &bpf_csum_level_proto;
8092 case BPF_FUNC_l3_csum_replace:
8093 return &bpf_l3_csum_replace_proto;
8094 case BPF_FUNC_l4_csum_replace:
8095 return &bpf_l4_csum_replace_proto;
8096 case BPF_FUNC_clone_redirect:
8097 return &bpf_clone_redirect_proto;
8098 case BPF_FUNC_get_cgroup_classid:
8099 return &bpf_get_cgroup_classid_proto;
8100 case BPF_FUNC_skb_vlan_push:
8101 return &bpf_skb_vlan_push_proto;
8102 case BPF_FUNC_skb_vlan_pop:
8103 return &bpf_skb_vlan_pop_proto;
8104 case BPF_FUNC_skb_change_proto:
8105 return &bpf_skb_change_proto_proto;
8106 case BPF_FUNC_skb_change_type:
8107 return &bpf_skb_change_type_proto;
8108 case BPF_FUNC_skb_adjust_room:
8109 return &bpf_skb_adjust_room_proto;
8110 case BPF_FUNC_skb_change_tail:
8111 return &bpf_skb_change_tail_proto;
8112 case BPF_FUNC_skb_change_head:
8113 return &bpf_skb_change_head_proto;
8114 case BPF_FUNC_skb_get_tunnel_key:
8115 return &bpf_skb_get_tunnel_key_proto;
8116 case BPF_FUNC_skb_set_tunnel_key:
8117 return bpf_get_skb_set_tunnel_proto(func_id);
8118 case BPF_FUNC_skb_get_tunnel_opt:
8119 return &bpf_skb_get_tunnel_opt_proto;
8120 case BPF_FUNC_skb_set_tunnel_opt:
8121 return bpf_get_skb_set_tunnel_proto(func_id);
8122 case BPF_FUNC_redirect:
8123 return &bpf_redirect_proto;
8124 case BPF_FUNC_redirect_neigh:
8125 return &bpf_redirect_neigh_proto;
8126 case BPF_FUNC_redirect_peer:
8127 return &bpf_redirect_peer_proto;
8128 case BPF_FUNC_get_route_realm:
8129 return &bpf_get_route_realm_proto;
8130 case BPF_FUNC_get_hash_recalc:
8131 return &bpf_get_hash_recalc_proto;
8132 case BPF_FUNC_set_hash_invalid:
8133 return &bpf_set_hash_invalid_proto;
8134 case BPF_FUNC_set_hash:
8135 return &bpf_set_hash_proto;
8136 case BPF_FUNC_perf_event_output:
8137 return &bpf_skb_event_output_proto;
8138 case BPF_FUNC_get_smp_processor_id:
8139 return &bpf_get_smp_processor_id_proto;
8140 case BPF_FUNC_skb_under_cgroup:
8141 return &bpf_skb_under_cgroup_proto;
8142 case BPF_FUNC_get_socket_cookie:
8143 return &bpf_get_socket_cookie_proto;
8144 case BPF_FUNC_get_socket_uid:
8145 return &bpf_get_socket_uid_proto;
8146 case BPF_FUNC_fib_lookup:
8147 return &bpf_skb_fib_lookup_proto;
8148 case BPF_FUNC_check_mtu:
8149 return &bpf_skb_check_mtu_proto;
8150 case BPF_FUNC_sk_fullsock:
8151 return &bpf_sk_fullsock_proto;
8152 case BPF_FUNC_sk_storage_get:
8153 return &bpf_sk_storage_get_proto;
8154 case BPF_FUNC_sk_storage_delete:
8155 return &bpf_sk_storage_delete_proto;
8156 #ifdef CONFIG_XFRM
8157 case BPF_FUNC_skb_get_xfrm_state:
8158 return &bpf_skb_get_xfrm_state_proto;
8159 #endif
8160 #ifdef CONFIG_CGROUP_NET_CLASSID
8161 case BPF_FUNC_skb_cgroup_classid:
8162 return &bpf_skb_cgroup_classid_proto;
8163 #endif
8164 #ifdef CONFIG_SOCK_CGROUP_DATA
8165 case BPF_FUNC_skb_cgroup_id:
8166 return &bpf_skb_cgroup_id_proto;
8167 case BPF_FUNC_skb_ancestor_cgroup_id:
8168 return &bpf_skb_ancestor_cgroup_id_proto;
8169 #endif
8170 #ifdef CONFIG_INET
8171 case BPF_FUNC_sk_lookup_tcp:
8172 return &bpf_tc_sk_lookup_tcp_proto;
8173 case BPF_FUNC_sk_lookup_udp:
8174 return &bpf_tc_sk_lookup_udp_proto;
8175 case BPF_FUNC_sk_release:
8176 return &bpf_sk_release_proto;
8177 case BPF_FUNC_tcp_sock:
8178 return &bpf_tcp_sock_proto;
8179 case BPF_FUNC_get_listener_sock:
8180 return &bpf_get_listener_sock_proto;
8181 case BPF_FUNC_skc_lookup_tcp:
8182 return &bpf_tc_skc_lookup_tcp_proto;
8183 case BPF_FUNC_tcp_check_syncookie:
8184 return &bpf_tcp_check_syncookie_proto;
8185 case BPF_FUNC_skb_ecn_set_ce:
8186 return &bpf_skb_ecn_set_ce_proto;
8187 case BPF_FUNC_tcp_gen_syncookie:
8188 return &bpf_tcp_gen_syncookie_proto;
8189 case BPF_FUNC_sk_assign:
8190 return &bpf_sk_assign_proto;
8191 case BPF_FUNC_skb_set_tstamp:
8192 return &bpf_skb_set_tstamp_proto;
8193 #ifdef CONFIG_SYN_COOKIES
8194 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8195 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8196 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8197 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8198 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8199 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8200 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8201 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8202 #endif
8203 #endif
8204 default:
8205 return bpf_sk_base_func_proto(func_id);
8206 }
8207 }
8208
8209 static const struct bpf_func_proto *
xdp_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8210 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8211 {
8212 switch (func_id) {
8213 case BPF_FUNC_perf_event_output:
8214 return &bpf_xdp_event_output_proto;
8215 case BPF_FUNC_get_smp_processor_id:
8216 return &bpf_get_smp_processor_id_proto;
8217 case BPF_FUNC_csum_diff:
8218 return &bpf_csum_diff_proto;
8219 case BPF_FUNC_xdp_adjust_head:
8220 return &bpf_xdp_adjust_head_proto;
8221 case BPF_FUNC_xdp_adjust_meta:
8222 return &bpf_xdp_adjust_meta_proto;
8223 case BPF_FUNC_redirect:
8224 return &bpf_xdp_redirect_proto;
8225 case BPF_FUNC_redirect_map:
8226 return &bpf_xdp_redirect_map_proto;
8227 case BPF_FUNC_xdp_adjust_tail:
8228 return &bpf_xdp_adjust_tail_proto;
8229 case BPF_FUNC_xdp_get_buff_len:
8230 return &bpf_xdp_get_buff_len_proto;
8231 case BPF_FUNC_xdp_load_bytes:
8232 return &bpf_xdp_load_bytes_proto;
8233 case BPF_FUNC_xdp_store_bytes:
8234 return &bpf_xdp_store_bytes_proto;
8235 case BPF_FUNC_fib_lookup:
8236 return &bpf_xdp_fib_lookup_proto;
8237 case BPF_FUNC_check_mtu:
8238 return &bpf_xdp_check_mtu_proto;
8239 #ifdef CONFIG_INET
8240 case BPF_FUNC_sk_lookup_udp:
8241 return &bpf_xdp_sk_lookup_udp_proto;
8242 case BPF_FUNC_sk_lookup_tcp:
8243 return &bpf_xdp_sk_lookup_tcp_proto;
8244 case BPF_FUNC_sk_release:
8245 return &bpf_sk_release_proto;
8246 case BPF_FUNC_skc_lookup_tcp:
8247 return &bpf_xdp_skc_lookup_tcp_proto;
8248 case BPF_FUNC_tcp_check_syncookie:
8249 return &bpf_tcp_check_syncookie_proto;
8250 case BPF_FUNC_tcp_gen_syncookie:
8251 return &bpf_tcp_gen_syncookie_proto;
8252 #ifdef CONFIG_SYN_COOKIES
8253 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8254 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8255 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8256 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8257 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8258 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8259 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8260 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8261 #endif
8262 #endif
8263 default:
8264 return bpf_sk_base_func_proto(func_id);
8265 }
8266
8267 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8268 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8269 * kfuncs are defined in two different modules, and we want to be able
8270 * to use them interchangably with the same BTF type ID. Because modules
8271 * can't de-duplicate BTF IDs between each other, we need the type to be
8272 * referenced in the vmlinux BTF or the verifier will get confused about
8273 * the different types. So we add this dummy type reference which will
8274 * be included in vmlinux BTF, allowing both modules to refer to the
8275 * same type ID.
8276 */
8277 BTF_TYPE_EMIT(struct nf_conn___init);
8278 #endif
8279 }
8280
8281 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8282 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8283
8284 static const struct bpf_func_proto *
sock_ops_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8285 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8286 {
8287 const struct bpf_func_proto *func_proto;
8288
8289 func_proto = cgroup_common_func_proto(func_id, prog);
8290 if (func_proto)
8291 return func_proto;
8292
8293 switch (func_id) {
8294 case BPF_FUNC_setsockopt:
8295 return &bpf_sock_ops_setsockopt_proto;
8296 case BPF_FUNC_getsockopt:
8297 return &bpf_sock_ops_getsockopt_proto;
8298 case BPF_FUNC_sock_ops_cb_flags_set:
8299 return &bpf_sock_ops_cb_flags_set_proto;
8300 case BPF_FUNC_sock_map_update:
8301 return &bpf_sock_map_update_proto;
8302 case BPF_FUNC_sock_hash_update:
8303 return &bpf_sock_hash_update_proto;
8304 case BPF_FUNC_get_socket_cookie:
8305 return &bpf_get_socket_cookie_sock_ops_proto;
8306 case BPF_FUNC_perf_event_output:
8307 return &bpf_event_output_data_proto;
8308 case BPF_FUNC_sk_storage_get:
8309 return &bpf_sk_storage_get_proto;
8310 case BPF_FUNC_sk_storage_delete:
8311 return &bpf_sk_storage_delete_proto;
8312 case BPF_FUNC_get_netns_cookie:
8313 return &bpf_get_netns_cookie_sock_ops_proto;
8314 #ifdef CONFIG_INET
8315 case BPF_FUNC_load_hdr_opt:
8316 return &bpf_sock_ops_load_hdr_opt_proto;
8317 case BPF_FUNC_store_hdr_opt:
8318 return &bpf_sock_ops_store_hdr_opt_proto;
8319 case BPF_FUNC_reserve_hdr_opt:
8320 return &bpf_sock_ops_reserve_hdr_opt_proto;
8321 case BPF_FUNC_tcp_sock:
8322 return &bpf_tcp_sock_proto;
8323 #endif /* CONFIG_INET */
8324 default:
8325 return bpf_sk_base_func_proto(func_id);
8326 }
8327 }
8328
8329 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8330 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8331
8332 static const struct bpf_func_proto *
sk_msg_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8333 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8334 {
8335 switch (func_id) {
8336 case BPF_FUNC_msg_redirect_map:
8337 return &bpf_msg_redirect_map_proto;
8338 case BPF_FUNC_msg_redirect_hash:
8339 return &bpf_msg_redirect_hash_proto;
8340 case BPF_FUNC_msg_apply_bytes:
8341 return &bpf_msg_apply_bytes_proto;
8342 case BPF_FUNC_msg_cork_bytes:
8343 return &bpf_msg_cork_bytes_proto;
8344 case BPF_FUNC_msg_pull_data:
8345 return &bpf_msg_pull_data_proto;
8346 case BPF_FUNC_msg_push_data:
8347 return &bpf_msg_push_data_proto;
8348 case BPF_FUNC_msg_pop_data:
8349 return &bpf_msg_pop_data_proto;
8350 case BPF_FUNC_perf_event_output:
8351 return &bpf_event_output_data_proto;
8352 case BPF_FUNC_get_current_uid_gid:
8353 return &bpf_get_current_uid_gid_proto;
8354 case BPF_FUNC_get_current_pid_tgid:
8355 return &bpf_get_current_pid_tgid_proto;
8356 case BPF_FUNC_sk_storage_get:
8357 return &bpf_sk_storage_get_proto;
8358 case BPF_FUNC_sk_storage_delete:
8359 return &bpf_sk_storage_delete_proto;
8360 case BPF_FUNC_get_netns_cookie:
8361 return &bpf_get_netns_cookie_sk_msg_proto;
8362 #ifdef CONFIG_CGROUP_NET_CLASSID
8363 case BPF_FUNC_get_cgroup_classid:
8364 return &bpf_get_cgroup_classid_curr_proto;
8365 #endif
8366 default:
8367 return bpf_sk_base_func_proto(func_id);
8368 }
8369 }
8370
8371 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8372 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8373
8374 static const struct bpf_func_proto *
sk_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8375 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8376 {
8377 switch (func_id) {
8378 case BPF_FUNC_skb_store_bytes:
8379 return &bpf_skb_store_bytes_proto;
8380 case BPF_FUNC_skb_load_bytes:
8381 return &bpf_skb_load_bytes_proto;
8382 case BPF_FUNC_skb_pull_data:
8383 return &sk_skb_pull_data_proto;
8384 case BPF_FUNC_skb_change_tail:
8385 return &sk_skb_change_tail_proto;
8386 case BPF_FUNC_skb_change_head:
8387 return &sk_skb_change_head_proto;
8388 case BPF_FUNC_skb_adjust_room:
8389 return &sk_skb_adjust_room_proto;
8390 case BPF_FUNC_get_socket_cookie:
8391 return &bpf_get_socket_cookie_proto;
8392 case BPF_FUNC_get_socket_uid:
8393 return &bpf_get_socket_uid_proto;
8394 case BPF_FUNC_sk_redirect_map:
8395 return &bpf_sk_redirect_map_proto;
8396 case BPF_FUNC_sk_redirect_hash:
8397 return &bpf_sk_redirect_hash_proto;
8398 case BPF_FUNC_perf_event_output:
8399 return &bpf_skb_event_output_proto;
8400 #ifdef CONFIG_INET
8401 case BPF_FUNC_sk_lookup_tcp:
8402 return &bpf_sk_lookup_tcp_proto;
8403 case BPF_FUNC_sk_lookup_udp:
8404 return &bpf_sk_lookup_udp_proto;
8405 case BPF_FUNC_sk_release:
8406 return &bpf_sk_release_proto;
8407 case BPF_FUNC_skc_lookup_tcp:
8408 return &bpf_skc_lookup_tcp_proto;
8409 #endif
8410 default:
8411 return bpf_sk_base_func_proto(func_id);
8412 }
8413 }
8414
8415 static const struct bpf_func_proto *
flow_dissector_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8416 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8417 {
8418 switch (func_id) {
8419 case BPF_FUNC_skb_load_bytes:
8420 return &bpf_flow_dissector_load_bytes_proto;
8421 default:
8422 return bpf_sk_base_func_proto(func_id);
8423 }
8424 }
8425
8426 static const struct bpf_func_proto *
lwt_out_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8427 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8428 {
8429 switch (func_id) {
8430 case BPF_FUNC_skb_load_bytes:
8431 return &bpf_skb_load_bytes_proto;
8432 case BPF_FUNC_skb_pull_data:
8433 return &bpf_skb_pull_data_proto;
8434 case BPF_FUNC_csum_diff:
8435 return &bpf_csum_diff_proto;
8436 case BPF_FUNC_get_cgroup_classid:
8437 return &bpf_get_cgroup_classid_proto;
8438 case BPF_FUNC_get_route_realm:
8439 return &bpf_get_route_realm_proto;
8440 case BPF_FUNC_get_hash_recalc:
8441 return &bpf_get_hash_recalc_proto;
8442 case BPF_FUNC_perf_event_output:
8443 return &bpf_skb_event_output_proto;
8444 case BPF_FUNC_get_smp_processor_id:
8445 return &bpf_get_smp_processor_id_proto;
8446 case BPF_FUNC_skb_under_cgroup:
8447 return &bpf_skb_under_cgroup_proto;
8448 default:
8449 return bpf_sk_base_func_proto(func_id);
8450 }
8451 }
8452
8453 static const struct bpf_func_proto *
lwt_in_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8454 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8455 {
8456 switch (func_id) {
8457 case BPF_FUNC_lwt_push_encap:
8458 return &bpf_lwt_in_push_encap_proto;
8459 default:
8460 return lwt_out_func_proto(func_id, prog);
8461 }
8462 }
8463
8464 static const struct bpf_func_proto *
lwt_xmit_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8465 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8466 {
8467 switch (func_id) {
8468 case BPF_FUNC_skb_get_tunnel_key:
8469 return &bpf_skb_get_tunnel_key_proto;
8470 case BPF_FUNC_skb_set_tunnel_key:
8471 return bpf_get_skb_set_tunnel_proto(func_id);
8472 case BPF_FUNC_skb_get_tunnel_opt:
8473 return &bpf_skb_get_tunnel_opt_proto;
8474 case BPF_FUNC_skb_set_tunnel_opt:
8475 return bpf_get_skb_set_tunnel_proto(func_id);
8476 case BPF_FUNC_redirect:
8477 return &bpf_redirect_proto;
8478 case BPF_FUNC_clone_redirect:
8479 return &bpf_clone_redirect_proto;
8480 case BPF_FUNC_skb_change_tail:
8481 return &bpf_skb_change_tail_proto;
8482 case BPF_FUNC_skb_change_head:
8483 return &bpf_skb_change_head_proto;
8484 case BPF_FUNC_skb_store_bytes:
8485 return &bpf_skb_store_bytes_proto;
8486 case BPF_FUNC_csum_update:
8487 return &bpf_csum_update_proto;
8488 case BPF_FUNC_csum_level:
8489 return &bpf_csum_level_proto;
8490 case BPF_FUNC_l3_csum_replace:
8491 return &bpf_l3_csum_replace_proto;
8492 case BPF_FUNC_l4_csum_replace:
8493 return &bpf_l4_csum_replace_proto;
8494 case BPF_FUNC_set_hash_invalid:
8495 return &bpf_set_hash_invalid_proto;
8496 case BPF_FUNC_lwt_push_encap:
8497 return &bpf_lwt_xmit_push_encap_proto;
8498 default:
8499 return lwt_out_func_proto(func_id, prog);
8500 }
8501 }
8502
8503 static const struct bpf_func_proto *
lwt_seg6local_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8504 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8505 {
8506 switch (func_id) {
8507 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8508 case BPF_FUNC_lwt_seg6_store_bytes:
8509 return &bpf_lwt_seg6_store_bytes_proto;
8510 case BPF_FUNC_lwt_seg6_action:
8511 return &bpf_lwt_seg6_action_proto;
8512 case BPF_FUNC_lwt_seg6_adjust_srh:
8513 return &bpf_lwt_seg6_adjust_srh_proto;
8514 #endif
8515 default:
8516 return lwt_out_func_proto(func_id, prog);
8517 }
8518 }
8519
bpf_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8520 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8521 const struct bpf_prog *prog,
8522 struct bpf_insn_access_aux *info)
8523 {
8524 const int size_default = sizeof(__u32);
8525
8526 if (off < 0 || off >= sizeof(struct __sk_buff))
8527 return false;
8528
8529 /* The verifier guarantees that size > 0. */
8530 if (off % size != 0)
8531 return false;
8532
8533 switch (off) {
8534 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8535 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8536 return false;
8537 break;
8538 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8539 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8540 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8541 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8542 case bpf_ctx_range(struct __sk_buff, data):
8543 case bpf_ctx_range(struct __sk_buff, data_meta):
8544 case bpf_ctx_range(struct __sk_buff, data_end):
8545 if (size != size_default)
8546 return false;
8547 break;
8548 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8549 return false;
8550 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8551 if (type == BPF_WRITE || size != sizeof(__u64))
8552 return false;
8553 break;
8554 case bpf_ctx_range(struct __sk_buff, tstamp):
8555 if (size != sizeof(__u64))
8556 return false;
8557 break;
8558 case offsetof(struct __sk_buff, sk):
8559 if (type == BPF_WRITE || size != sizeof(__u64))
8560 return false;
8561 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8562 break;
8563 case offsetof(struct __sk_buff, tstamp_type):
8564 return false;
8565 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8566 /* Explicitly prohibit access to padding in __sk_buff. */
8567 return false;
8568 default:
8569 /* Only narrow read access allowed for now. */
8570 if (type == BPF_WRITE) {
8571 if (size != size_default)
8572 return false;
8573 } else {
8574 bpf_ctx_record_field_size(info, size_default);
8575 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8576 return false;
8577 }
8578 }
8579
8580 return true;
8581 }
8582
sk_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8583 static bool sk_filter_is_valid_access(int off, int size,
8584 enum bpf_access_type type,
8585 const struct bpf_prog *prog,
8586 struct bpf_insn_access_aux *info)
8587 {
8588 switch (off) {
8589 case bpf_ctx_range(struct __sk_buff, tc_classid):
8590 case bpf_ctx_range(struct __sk_buff, data):
8591 case bpf_ctx_range(struct __sk_buff, data_meta):
8592 case bpf_ctx_range(struct __sk_buff, data_end):
8593 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8594 case bpf_ctx_range(struct __sk_buff, tstamp):
8595 case bpf_ctx_range(struct __sk_buff, wire_len):
8596 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8597 return false;
8598 }
8599
8600 if (type == BPF_WRITE) {
8601 switch (off) {
8602 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8603 break;
8604 default:
8605 return false;
8606 }
8607 }
8608
8609 return bpf_skb_is_valid_access(off, size, type, prog, info);
8610 }
8611
cg_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8612 static bool cg_skb_is_valid_access(int off, int size,
8613 enum bpf_access_type type,
8614 const struct bpf_prog *prog,
8615 struct bpf_insn_access_aux *info)
8616 {
8617 switch (off) {
8618 case bpf_ctx_range(struct __sk_buff, tc_classid):
8619 case bpf_ctx_range(struct __sk_buff, data_meta):
8620 case bpf_ctx_range(struct __sk_buff, wire_len):
8621 return false;
8622 case bpf_ctx_range(struct __sk_buff, data):
8623 case bpf_ctx_range(struct __sk_buff, data_end):
8624 if (!bpf_capable())
8625 return false;
8626 break;
8627 }
8628
8629 if (type == BPF_WRITE) {
8630 switch (off) {
8631 case bpf_ctx_range(struct __sk_buff, mark):
8632 case bpf_ctx_range(struct __sk_buff, priority):
8633 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8634 break;
8635 case bpf_ctx_range(struct __sk_buff, tstamp):
8636 if (!bpf_capable())
8637 return false;
8638 break;
8639 default:
8640 return false;
8641 }
8642 }
8643
8644 switch (off) {
8645 case bpf_ctx_range(struct __sk_buff, data):
8646 info->reg_type = PTR_TO_PACKET;
8647 break;
8648 case bpf_ctx_range(struct __sk_buff, data_end):
8649 info->reg_type = PTR_TO_PACKET_END;
8650 break;
8651 }
8652
8653 return bpf_skb_is_valid_access(off, size, type, prog, info);
8654 }
8655
lwt_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8656 static bool lwt_is_valid_access(int off, int size,
8657 enum bpf_access_type type,
8658 const struct bpf_prog *prog,
8659 struct bpf_insn_access_aux *info)
8660 {
8661 switch (off) {
8662 case bpf_ctx_range(struct __sk_buff, tc_classid):
8663 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8664 case bpf_ctx_range(struct __sk_buff, data_meta):
8665 case bpf_ctx_range(struct __sk_buff, tstamp):
8666 case bpf_ctx_range(struct __sk_buff, wire_len):
8667 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8668 return false;
8669 }
8670
8671 if (type == BPF_WRITE) {
8672 switch (off) {
8673 case bpf_ctx_range(struct __sk_buff, mark):
8674 case bpf_ctx_range(struct __sk_buff, priority):
8675 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8676 break;
8677 default:
8678 return false;
8679 }
8680 }
8681
8682 switch (off) {
8683 case bpf_ctx_range(struct __sk_buff, data):
8684 info->reg_type = PTR_TO_PACKET;
8685 break;
8686 case bpf_ctx_range(struct __sk_buff, data_end):
8687 info->reg_type = PTR_TO_PACKET_END;
8688 break;
8689 }
8690
8691 return bpf_skb_is_valid_access(off, size, type, prog, info);
8692 }
8693
8694 /* Attach type specific accesses */
__sock_filter_check_attach_type(int off,enum bpf_access_type access_type,enum bpf_attach_type attach_type)8695 static bool __sock_filter_check_attach_type(int off,
8696 enum bpf_access_type access_type,
8697 enum bpf_attach_type attach_type)
8698 {
8699 switch (off) {
8700 case offsetof(struct bpf_sock, bound_dev_if):
8701 case offsetof(struct bpf_sock, mark):
8702 case offsetof(struct bpf_sock, priority):
8703 switch (attach_type) {
8704 case BPF_CGROUP_INET_SOCK_CREATE:
8705 case BPF_CGROUP_INET_SOCK_RELEASE:
8706 goto full_access;
8707 default:
8708 return false;
8709 }
8710 case bpf_ctx_range(struct bpf_sock, src_ip4):
8711 switch (attach_type) {
8712 case BPF_CGROUP_INET4_POST_BIND:
8713 goto read_only;
8714 default:
8715 return false;
8716 }
8717 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8718 switch (attach_type) {
8719 case BPF_CGROUP_INET6_POST_BIND:
8720 goto read_only;
8721 default:
8722 return false;
8723 }
8724 case bpf_ctx_range(struct bpf_sock, src_port):
8725 switch (attach_type) {
8726 case BPF_CGROUP_INET4_POST_BIND:
8727 case BPF_CGROUP_INET6_POST_BIND:
8728 goto read_only;
8729 default:
8730 return false;
8731 }
8732 }
8733 read_only:
8734 return access_type == BPF_READ;
8735 full_access:
8736 return true;
8737 }
8738
bpf_sock_common_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8739 bool bpf_sock_common_is_valid_access(int off, int size,
8740 enum bpf_access_type type,
8741 struct bpf_insn_access_aux *info)
8742 {
8743 switch (off) {
8744 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8745 return false;
8746 default:
8747 return bpf_sock_is_valid_access(off, size, type, info);
8748 }
8749 }
8750
bpf_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8751 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8752 struct bpf_insn_access_aux *info)
8753 {
8754 const int size_default = sizeof(__u32);
8755 int field_size;
8756
8757 if (off < 0 || off >= sizeof(struct bpf_sock))
8758 return false;
8759 if (off % size != 0)
8760 return false;
8761
8762 switch (off) {
8763 case offsetof(struct bpf_sock, state):
8764 case offsetof(struct bpf_sock, family):
8765 case offsetof(struct bpf_sock, type):
8766 case offsetof(struct bpf_sock, protocol):
8767 case offsetof(struct bpf_sock, src_port):
8768 case offsetof(struct bpf_sock, rx_queue_mapping):
8769 case bpf_ctx_range(struct bpf_sock, src_ip4):
8770 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8771 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8772 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8773 bpf_ctx_record_field_size(info, size_default);
8774 return bpf_ctx_narrow_access_ok(off, size, size_default);
8775 case bpf_ctx_range(struct bpf_sock, dst_port):
8776 field_size = size == size_default ?
8777 size_default : sizeof_field(struct bpf_sock, dst_port);
8778 bpf_ctx_record_field_size(info, field_size);
8779 return bpf_ctx_narrow_access_ok(off, size, field_size);
8780 case offsetofend(struct bpf_sock, dst_port) ...
8781 offsetof(struct bpf_sock, dst_ip4) - 1:
8782 return false;
8783 }
8784
8785 return size == size_default;
8786 }
8787
sock_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8788 static bool sock_filter_is_valid_access(int off, int size,
8789 enum bpf_access_type type,
8790 const struct bpf_prog *prog,
8791 struct bpf_insn_access_aux *info)
8792 {
8793 if (!bpf_sock_is_valid_access(off, size, type, info))
8794 return false;
8795 return __sock_filter_check_attach_type(off, type,
8796 prog->expected_attach_type);
8797 }
8798
bpf_noop_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8799 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8800 const struct bpf_prog *prog)
8801 {
8802 /* Neither direct read nor direct write requires any preliminary
8803 * action.
8804 */
8805 return 0;
8806 }
8807
bpf_unclone_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog,int drop_verdict)8808 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8809 const struct bpf_prog *prog, int drop_verdict)
8810 {
8811 struct bpf_insn *insn = insn_buf;
8812
8813 if (!direct_write)
8814 return 0;
8815
8816 /* if (!skb->cloned)
8817 * goto start;
8818 *
8819 * (Fast-path, otherwise approximation that we might be
8820 * a clone, do the rest in helper.)
8821 */
8822 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8823 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8824 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8825
8826 /* ret = bpf_skb_pull_data(skb, 0); */
8827 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8828 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8829 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8830 BPF_FUNC_skb_pull_data);
8831 /* if (!ret)
8832 * goto restore;
8833 * return TC_ACT_SHOT;
8834 */
8835 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8836 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8837 *insn++ = BPF_EXIT_INSN();
8838
8839 /* restore: */
8840 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8841 /* start: */
8842 *insn++ = prog->insnsi[0];
8843
8844 return insn - insn_buf;
8845 }
8846
bpf_gen_ld_abs(const struct bpf_insn * orig,struct bpf_insn * insn_buf)8847 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8848 struct bpf_insn *insn_buf)
8849 {
8850 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8851 struct bpf_insn *insn = insn_buf;
8852
8853 if (!indirect) {
8854 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8855 } else {
8856 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8857 if (orig->imm)
8858 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8859 }
8860 /* We're guaranteed here that CTX is in R6. */
8861 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8862
8863 switch (BPF_SIZE(orig->code)) {
8864 case BPF_B:
8865 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8866 break;
8867 case BPF_H:
8868 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8869 break;
8870 case BPF_W:
8871 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8872 break;
8873 }
8874
8875 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8876 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8877 *insn++ = BPF_EXIT_INSN();
8878
8879 return insn - insn_buf;
8880 }
8881
tc_cls_act_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8882 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8883 const struct bpf_prog *prog)
8884 {
8885 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8886 }
8887
tc_cls_act_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8888 static bool tc_cls_act_is_valid_access(int off, int size,
8889 enum bpf_access_type type,
8890 const struct bpf_prog *prog,
8891 struct bpf_insn_access_aux *info)
8892 {
8893 if (type == BPF_WRITE) {
8894 switch (off) {
8895 case bpf_ctx_range(struct __sk_buff, mark):
8896 case bpf_ctx_range(struct __sk_buff, tc_index):
8897 case bpf_ctx_range(struct __sk_buff, priority):
8898 case bpf_ctx_range(struct __sk_buff, tc_classid):
8899 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8900 case bpf_ctx_range(struct __sk_buff, tstamp):
8901 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8902 break;
8903 default:
8904 return false;
8905 }
8906 }
8907
8908 switch (off) {
8909 case bpf_ctx_range(struct __sk_buff, data):
8910 info->reg_type = PTR_TO_PACKET;
8911 break;
8912 case bpf_ctx_range(struct __sk_buff, data_meta):
8913 info->reg_type = PTR_TO_PACKET_META;
8914 break;
8915 case bpf_ctx_range(struct __sk_buff, data_end):
8916 info->reg_type = PTR_TO_PACKET_END;
8917 break;
8918 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8919 return false;
8920 case offsetof(struct __sk_buff, tstamp_type):
8921 /* The convert_ctx_access() on reading and writing
8922 * __sk_buff->tstamp depends on whether the bpf prog
8923 * has used __sk_buff->tstamp_type or not.
8924 * Thus, we need to set prog->tstamp_type_access
8925 * earlier during is_valid_access() here.
8926 */
8927 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8928 return size == sizeof(__u8);
8929 }
8930
8931 return bpf_skb_is_valid_access(off, size, type, prog, info);
8932 }
8933
8934 DEFINE_MUTEX(nf_conn_btf_access_lock);
8935 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8936
8937 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8938 const struct bpf_reg_state *reg,
8939 int off, int size);
8940 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8941
tc_cls_act_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)8942 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8943 const struct bpf_reg_state *reg,
8944 int off, int size)
8945 {
8946 int ret = -EACCES;
8947
8948 mutex_lock(&nf_conn_btf_access_lock);
8949 if (nfct_btf_struct_access)
8950 ret = nfct_btf_struct_access(log, reg, off, size);
8951 mutex_unlock(&nf_conn_btf_access_lock);
8952
8953 return ret;
8954 }
8955
__is_valid_xdp_access(int off,int size)8956 static bool __is_valid_xdp_access(int off, int size)
8957 {
8958 if (off < 0 || off >= sizeof(struct xdp_md))
8959 return false;
8960 if (off % size != 0)
8961 return false;
8962 if (size != sizeof(__u32))
8963 return false;
8964
8965 return true;
8966 }
8967
xdp_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8968 static bool xdp_is_valid_access(int off, int size,
8969 enum bpf_access_type type,
8970 const struct bpf_prog *prog,
8971 struct bpf_insn_access_aux *info)
8972 {
8973 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8974 switch (off) {
8975 case offsetof(struct xdp_md, egress_ifindex):
8976 return false;
8977 }
8978 }
8979
8980 if (type == BPF_WRITE) {
8981 if (bpf_prog_is_offloaded(prog->aux)) {
8982 switch (off) {
8983 case offsetof(struct xdp_md, rx_queue_index):
8984 return __is_valid_xdp_access(off, size);
8985 }
8986 }
8987 return false;
8988 }
8989
8990 switch (off) {
8991 case offsetof(struct xdp_md, data):
8992 info->reg_type = PTR_TO_PACKET;
8993 break;
8994 case offsetof(struct xdp_md, data_meta):
8995 info->reg_type = PTR_TO_PACKET_META;
8996 break;
8997 case offsetof(struct xdp_md, data_end):
8998 info->reg_type = PTR_TO_PACKET_END;
8999 break;
9000 }
9001
9002 return __is_valid_xdp_access(off, size);
9003 }
9004
bpf_warn_invalid_xdp_action(struct net_device * dev,struct bpf_prog * prog,u32 act)9005 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
9006 {
9007 const u32 act_max = XDP_REDIRECT;
9008
9009 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
9010 act > act_max ? "Illegal" : "Driver unsupported",
9011 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
9012 }
9013 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
9014
xdp_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9015 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
9016 const struct bpf_reg_state *reg,
9017 int off, int size)
9018 {
9019 int ret = -EACCES;
9020
9021 mutex_lock(&nf_conn_btf_access_lock);
9022 if (nfct_btf_struct_access)
9023 ret = nfct_btf_struct_access(log, reg, off, size);
9024 mutex_unlock(&nf_conn_btf_access_lock);
9025
9026 return ret;
9027 }
9028
sock_addr_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9029 static bool sock_addr_is_valid_access(int off, int size,
9030 enum bpf_access_type type,
9031 const struct bpf_prog *prog,
9032 struct bpf_insn_access_aux *info)
9033 {
9034 const int size_default = sizeof(__u32);
9035
9036 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
9037 return false;
9038 if (off % size != 0)
9039 return false;
9040
9041 /* Disallow access to IPv6 fields from IPv4 contex and vise
9042 * versa.
9043 */
9044 switch (off) {
9045 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9046 switch (prog->expected_attach_type) {
9047 case BPF_CGROUP_INET4_BIND:
9048 case BPF_CGROUP_INET4_CONNECT:
9049 case BPF_CGROUP_INET4_GETPEERNAME:
9050 case BPF_CGROUP_INET4_GETSOCKNAME:
9051 case BPF_CGROUP_UDP4_SENDMSG:
9052 case BPF_CGROUP_UDP4_RECVMSG:
9053 break;
9054 default:
9055 return false;
9056 }
9057 break;
9058 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9059 switch (prog->expected_attach_type) {
9060 case BPF_CGROUP_INET6_BIND:
9061 case BPF_CGROUP_INET6_CONNECT:
9062 case BPF_CGROUP_INET6_GETPEERNAME:
9063 case BPF_CGROUP_INET6_GETSOCKNAME:
9064 case BPF_CGROUP_UDP6_SENDMSG:
9065 case BPF_CGROUP_UDP6_RECVMSG:
9066 break;
9067 default:
9068 return false;
9069 }
9070 break;
9071 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9072 switch (prog->expected_attach_type) {
9073 case BPF_CGROUP_UDP4_SENDMSG:
9074 break;
9075 default:
9076 return false;
9077 }
9078 break;
9079 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9080 msg_src_ip6[3]):
9081 switch (prog->expected_attach_type) {
9082 case BPF_CGROUP_UDP6_SENDMSG:
9083 break;
9084 default:
9085 return false;
9086 }
9087 break;
9088 }
9089
9090 switch (off) {
9091 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9092 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9093 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9094 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9095 msg_src_ip6[3]):
9096 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9097 if (type == BPF_READ) {
9098 bpf_ctx_record_field_size(info, size_default);
9099
9100 if (bpf_ctx_wide_access_ok(off, size,
9101 struct bpf_sock_addr,
9102 user_ip6))
9103 return true;
9104
9105 if (bpf_ctx_wide_access_ok(off, size,
9106 struct bpf_sock_addr,
9107 msg_src_ip6))
9108 return true;
9109
9110 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9111 return false;
9112 } else {
9113 if (bpf_ctx_wide_access_ok(off, size,
9114 struct bpf_sock_addr,
9115 user_ip6))
9116 return true;
9117
9118 if (bpf_ctx_wide_access_ok(off, size,
9119 struct bpf_sock_addr,
9120 msg_src_ip6))
9121 return true;
9122
9123 if (size != size_default)
9124 return false;
9125 }
9126 break;
9127 case offsetof(struct bpf_sock_addr, sk):
9128 if (type != BPF_READ)
9129 return false;
9130 if (size != sizeof(__u64))
9131 return false;
9132 info->reg_type = PTR_TO_SOCKET;
9133 break;
9134 default:
9135 if (type == BPF_READ) {
9136 if (size != size_default)
9137 return false;
9138 } else {
9139 return false;
9140 }
9141 }
9142
9143 return true;
9144 }
9145
sock_ops_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9146 static bool sock_ops_is_valid_access(int off, int size,
9147 enum bpf_access_type type,
9148 const struct bpf_prog *prog,
9149 struct bpf_insn_access_aux *info)
9150 {
9151 const int size_default = sizeof(__u32);
9152
9153 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9154 return false;
9155
9156 /* The verifier guarantees that size > 0. */
9157 if (off % size != 0)
9158 return false;
9159
9160 if (type == BPF_WRITE) {
9161 switch (off) {
9162 case offsetof(struct bpf_sock_ops, reply):
9163 case offsetof(struct bpf_sock_ops, sk_txhash):
9164 if (size != size_default)
9165 return false;
9166 break;
9167 default:
9168 return false;
9169 }
9170 } else {
9171 switch (off) {
9172 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9173 bytes_acked):
9174 if (size != sizeof(__u64))
9175 return false;
9176 break;
9177 case offsetof(struct bpf_sock_ops, sk):
9178 if (size != sizeof(__u64))
9179 return false;
9180 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9181 break;
9182 case offsetof(struct bpf_sock_ops, skb_data):
9183 if (size != sizeof(__u64))
9184 return false;
9185 info->reg_type = PTR_TO_PACKET;
9186 break;
9187 case offsetof(struct bpf_sock_ops, skb_data_end):
9188 if (size != sizeof(__u64))
9189 return false;
9190 info->reg_type = PTR_TO_PACKET_END;
9191 break;
9192 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9193 bpf_ctx_record_field_size(info, size_default);
9194 return bpf_ctx_narrow_access_ok(off, size,
9195 size_default);
9196 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9197 if (size != sizeof(__u64))
9198 return false;
9199 break;
9200 default:
9201 if (size != size_default)
9202 return false;
9203 break;
9204 }
9205 }
9206
9207 return true;
9208 }
9209
sk_skb_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)9210 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9211 const struct bpf_prog *prog)
9212 {
9213 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9214 }
9215
sk_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9216 static bool sk_skb_is_valid_access(int off, int size,
9217 enum bpf_access_type type,
9218 const struct bpf_prog *prog,
9219 struct bpf_insn_access_aux *info)
9220 {
9221 switch (off) {
9222 case bpf_ctx_range(struct __sk_buff, tc_classid):
9223 case bpf_ctx_range(struct __sk_buff, data_meta):
9224 case bpf_ctx_range(struct __sk_buff, tstamp):
9225 case bpf_ctx_range(struct __sk_buff, wire_len):
9226 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9227 return false;
9228 }
9229
9230 if (type == BPF_WRITE) {
9231 switch (off) {
9232 case bpf_ctx_range(struct __sk_buff, tc_index):
9233 case bpf_ctx_range(struct __sk_buff, priority):
9234 break;
9235 default:
9236 return false;
9237 }
9238 }
9239
9240 switch (off) {
9241 case bpf_ctx_range(struct __sk_buff, mark):
9242 return false;
9243 case bpf_ctx_range(struct __sk_buff, data):
9244 info->reg_type = PTR_TO_PACKET;
9245 break;
9246 case bpf_ctx_range(struct __sk_buff, data_end):
9247 info->reg_type = PTR_TO_PACKET_END;
9248 break;
9249 }
9250
9251 return bpf_skb_is_valid_access(off, size, type, prog, info);
9252 }
9253
sk_msg_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9254 static bool sk_msg_is_valid_access(int off, int size,
9255 enum bpf_access_type type,
9256 const struct bpf_prog *prog,
9257 struct bpf_insn_access_aux *info)
9258 {
9259 if (type == BPF_WRITE)
9260 return false;
9261
9262 if (off % size != 0)
9263 return false;
9264
9265 switch (off) {
9266 case offsetof(struct sk_msg_md, data):
9267 info->reg_type = PTR_TO_PACKET;
9268 if (size != sizeof(__u64))
9269 return false;
9270 break;
9271 case offsetof(struct sk_msg_md, data_end):
9272 info->reg_type = PTR_TO_PACKET_END;
9273 if (size != sizeof(__u64))
9274 return false;
9275 break;
9276 case offsetof(struct sk_msg_md, sk):
9277 if (size != sizeof(__u64))
9278 return false;
9279 info->reg_type = PTR_TO_SOCKET;
9280 break;
9281 case bpf_ctx_range(struct sk_msg_md, family):
9282 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9283 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9284 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9285 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9286 case bpf_ctx_range(struct sk_msg_md, remote_port):
9287 case bpf_ctx_range(struct sk_msg_md, local_port):
9288 case bpf_ctx_range(struct sk_msg_md, size):
9289 if (size != sizeof(__u32))
9290 return false;
9291 break;
9292 default:
9293 return false;
9294 }
9295 return true;
9296 }
9297
flow_dissector_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9298 static bool flow_dissector_is_valid_access(int off, int size,
9299 enum bpf_access_type type,
9300 const struct bpf_prog *prog,
9301 struct bpf_insn_access_aux *info)
9302 {
9303 const int size_default = sizeof(__u32);
9304
9305 if (off < 0 || off >= sizeof(struct __sk_buff))
9306 return false;
9307
9308 if (type == BPF_WRITE)
9309 return false;
9310
9311 switch (off) {
9312 case bpf_ctx_range(struct __sk_buff, data):
9313 if (size != size_default)
9314 return false;
9315 info->reg_type = PTR_TO_PACKET;
9316 return true;
9317 case bpf_ctx_range(struct __sk_buff, data_end):
9318 if (size != size_default)
9319 return false;
9320 info->reg_type = PTR_TO_PACKET_END;
9321 return true;
9322 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9323 if (size != sizeof(__u64))
9324 return false;
9325 info->reg_type = PTR_TO_FLOW_KEYS;
9326 return true;
9327 default:
9328 return false;
9329 }
9330 }
9331
flow_dissector_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9332 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9333 const struct bpf_insn *si,
9334 struct bpf_insn *insn_buf,
9335 struct bpf_prog *prog,
9336 u32 *target_size)
9337
9338 {
9339 struct bpf_insn *insn = insn_buf;
9340
9341 switch (si->off) {
9342 case offsetof(struct __sk_buff, data):
9343 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9344 si->dst_reg, si->src_reg,
9345 offsetof(struct bpf_flow_dissector, data));
9346 break;
9347
9348 case offsetof(struct __sk_buff, data_end):
9349 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9350 si->dst_reg, si->src_reg,
9351 offsetof(struct bpf_flow_dissector, data_end));
9352 break;
9353
9354 case offsetof(struct __sk_buff, flow_keys):
9355 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9356 si->dst_reg, si->src_reg,
9357 offsetof(struct bpf_flow_dissector, flow_keys));
9358 break;
9359 }
9360
9361 return insn - insn_buf;
9362 }
9363
bpf_convert_tstamp_type_read(const struct bpf_insn * si,struct bpf_insn * insn)9364 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9365 struct bpf_insn *insn)
9366 {
9367 __u8 value_reg = si->dst_reg;
9368 __u8 skb_reg = si->src_reg;
9369 /* AX is needed because src_reg and dst_reg could be the same */
9370 __u8 tmp_reg = BPF_REG_AX;
9371
9372 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9373 SKB_BF_MONO_TC_OFFSET);
9374 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9375 SKB_MONO_DELIVERY_TIME_MASK, 2);
9376 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9377 *insn++ = BPF_JMP_A(1);
9378 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9379
9380 return insn;
9381 }
9382
bpf_convert_shinfo_access(__u8 dst_reg,__u8 skb_reg,struct bpf_insn * insn)9383 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9384 struct bpf_insn *insn)
9385 {
9386 /* si->dst_reg = skb_shinfo(SKB); */
9387 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9388 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9389 BPF_REG_AX, skb_reg,
9390 offsetof(struct sk_buff, end));
9391 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9392 dst_reg, skb_reg,
9393 offsetof(struct sk_buff, head));
9394 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9395 #else
9396 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9397 dst_reg, skb_reg,
9398 offsetof(struct sk_buff, end));
9399 #endif
9400
9401 return insn;
9402 }
9403
bpf_convert_tstamp_read(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9404 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9405 const struct bpf_insn *si,
9406 struct bpf_insn *insn)
9407 {
9408 __u8 value_reg = si->dst_reg;
9409 __u8 skb_reg = si->src_reg;
9410
9411 #ifdef CONFIG_NET_XGRESS
9412 /* If the tstamp_type is read,
9413 * the bpf prog is aware the tstamp could have delivery time.
9414 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9415 */
9416 if (!prog->tstamp_type_access) {
9417 /* AX is needed because src_reg and dst_reg could be the same */
9418 __u8 tmp_reg = BPF_REG_AX;
9419
9420 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9421 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9422 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9423 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9424 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9425 /* skb->tc_at_ingress && skb->mono_delivery_time,
9426 * read 0 as the (rcv) timestamp.
9427 */
9428 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9429 *insn++ = BPF_JMP_A(1);
9430 }
9431 #endif
9432
9433 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9434 offsetof(struct sk_buff, tstamp));
9435 return insn;
9436 }
9437
bpf_convert_tstamp_write(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9438 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9439 const struct bpf_insn *si,
9440 struct bpf_insn *insn)
9441 {
9442 __u8 value_reg = si->src_reg;
9443 __u8 skb_reg = si->dst_reg;
9444
9445 #ifdef CONFIG_NET_XGRESS
9446 /* If the tstamp_type is read,
9447 * the bpf prog is aware the tstamp could have delivery time.
9448 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9449 * Otherwise, writing at ingress will have to clear the
9450 * mono_delivery_time bit also.
9451 */
9452 if (!prog->tstamp_type_access) {
9453 __u8 tmp_reg = BPF_REG_AX;
9454
9455 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9456 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9457 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9458 /* goto <store> */
9459 *insn++ = BPF_JMP_A(2);
9460 /* <clear>: mono_delivery_time */
9461 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9462 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9463 }
9464 #endif
9465
9466 /* <store>: skb->tstamp = tstamp */
9467 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9468 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9469 return insn;
9470 }
9471
9472 #define BPF_EMIT_STORE(size, si, off) \
9473 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9474 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9475
bpf_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9476 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9477 const struct bpf_insn *si,
9478 struct bpf_insn *insn_buf,
9479 struct bpf_prog *prog, u32 *target_size)
9480 {
9481 struct bpf_insn *insn = insn_buf;
9482 int off;
9483
9484 switch (si->off) {
9485 case offsetof(struct __sk_buff, len):
9486 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9487 bpf_target_off(struct sk_buff, len, 4,
9488 target_size));
9489 break;
9490
9491 case offsetof(struct __sk_buff, protocol):
9492 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9493 bpf_target_off(struct sk_buff, protocol, 2,
9494 target_size));
9495 break;
9496
9497 case offsetof(struct __sk_buff, vlan_proto):
9498 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9499 bpf_target_off(struct sk_buff, vlan_proto, 2,
9500 target_size));
9501 break;
9502
9503 case offsetof(struct __sk_buff, priority):
9504 if (type == BPF_WRITE)
9505 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9506 bpf_target_off(struct sk_buff, priority, 4,
9507 target_size));
9508 else
9509 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9510 bpf_target_off(struct sk_buff, priority, 4,
9511 target_size));
9512 break;
9513
9514 case offsetof(struct __sk_buff, ingress_ifindex):
9515 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9516 bpf_target_off(struct sk_buff, skb_iif, 4,
9517 target_size));
9518 break;
9519
9520 case offsetof(struct __sk_buff, ifindex):
9521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9522 si->dst_reg, si->src_reg,
9523 offsetof(struct sk_buff, dev));
9524 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9525 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9526 bpf_target_off(struct net_device, ifindex, 4,
9527 target_size));
9528 break;
9529
9530 case offsetof(struct __sk_buff, hash):
9531 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9532 bpf_target_off(struct sk_buff, hash, 4,
9533 target_size));
9534 break;
9535
9536 case offsetof(struct __sk_buff, mark):
9537 if (type == BPF_WRITE)
9538 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9539 bpf_target_off(struct sk_buff, mark, 4,
9540 target_size));
9541 else
9542 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9543 bpf_target_off(struct sk_buff, mark, 4,
9544 target_size));
9545 break;
9546
9547 case offsetof(struct __sk_buff, pkt_type):
9548 *target_size = 1;
9549 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9550 PKT_TYPE_OFFSET);
9551 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9552 #ifdef __BIG_ENDIAN_BITFIELD
9553 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9554 #endif
9555 break;
9556
9557 case offsetof(struct __sk_buff, queue_mapping):
9558 if (type == BPF_WRITE) {
9559 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9560
9561 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9562 *insn++ = BPF_JMP_A(0); /* noop */
9563 break;
9564 }
9565
9566 if (BPF_CLASS(si->code) == BPF_STX)
9567 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9568 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9569 } else {
9570 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9571 bpf_target_off(struct sk_buff,
9572 queue_mapping,
9573 2, target_size));
9574 }
9575 break;
9576
9577 case offsetof(struct __sk_buff, vlan_present):
9578 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9579 bpf_target_off(struct sk_buff,
9580 vlan_all, 4, target_size));
9581 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9582 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9583 break;
9584
9585 case offsetof(struct __sk_buff, vlan_tci):
9586 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9587 bpf_target_off(struct sk_buff, vlan_tci, 2,
9588 target_size));
9589 break;
9590
9591 case offsetof(struct __sk_buff, cb[0]) ...
9592 offsetofend(struct __sk_buff, cb[4]) - 1:
9593 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9594 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9595 offsetof(struct qdisc_skb_cb, data)) %
9596 sizeof(__u64));
9597
9598 prog->cb_access = 1;
9599 off = si->off;
9600 off -= offsetof(struct __sk_buff, cb[0]);
9601 off += offsetof(struct sk_buff, cb);
9602 off += offsetof(struct qdisc_skb_cb, data);
9603 if (type == BPF_WRITE)
9604 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9605 else
9606 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9607 si->src_reg, off);
9608 break;
9609
9610 case offsetof(struct __sk_buff, tc_classid):
9611 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9612
9613 off = si->off;
9614 off -= offsetof(struct __sk_buff, tc_classid);
9615 off += offsetof(struct sk_buff, cb);
9616 off += offsetof(struct qdisc_skb_cb, tc_classid);
9617 *target_size = 2;
9618 if (type == BPF_WRITE)
9619 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9620 else
9621 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9622 si->src_reg, off);
9623 break;
9624
9625 case offsetof(struct __sk_buff, data):
9626 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9627 si->dst_reg, si->src_reg,
9628 offsetof(struct sk_buff, data));
9629 break;
9630
9631 case offsetof(struct __sk_buff, data_meta):
9632 off = si->off;
9633 off -= offsetof(struct __sk_buff, data_meta);
9634 off += offsetof(struct sk_buff, cb);
9635 off += offsetof(struct bpf_skb_data_end, data_meta);
9636 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9637 si->src_reg, off);
9638 break;
9639
9640 case offsetof(struct __sk_buff, data_end):
9641 off = si->off;
9642 off -= offsetof(struct __sk_buff, data_end);
9643 off += offsetof(struct sk_buff, cb);
9644 off += offsetof(struct bpf_skb_data_end, data_end);
9645 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9646 si->src_reg, off);
9647 break;
9648
9649 case offsetof(struct __sk_buff, tc_index):
9650 #ifdef CONFIG_NET_SCHED
9651 if (type == BPF_WRITE)
9652 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9653 bpf_target_off(struct sk_buff, tc_index, 2,
9654 target_size));
9655 else
9656 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9657 bpf_target_off(struct sk_buff, tc_index, 2,
9658 target_size));
9659 #else
9660 *target_size = 2;
9661 if (type == BPF_WRITE)
9662 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9663 else
9664 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9665 #endif
9666 break;
9667
9668 case offsetof(struct __sk_buff, napi_id):
9669 #if defined(CONFIG_NET_RX_BUSY_POLL)
9670 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9671 bpf_target_off(struct sk_buff, napi_id, 4,
9672 target_size));
9673 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9674 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9675 #else
9676 *target_size = 4;
9677 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9678 #endif
9679 break;
9680 case offsetof(struct __sk_buff, family):
9681 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9682
9683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9684 si->dst_reg, si->src_reg,
9685 offsetof(struct sk_buff, sk));
9686 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9687 bpf_target_off(struct sock_common,
9688 skc_family,
9689 2, target_size));
9690 break;
9691 case offsetof(struct __sk_buff, remote_ip4):
9692 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9693
9694 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9695 si->dst_reg, si->src_reg,
9696 offsetof(struct sk_buff, sk));
9697 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9698 bpf_target_off(struct sock_common,
9699 skc_daddr,
9700 4, target_size));
9701 break;
9702 case offsetof(struct __sk_buff, local_ip4):
9703 BUILD_BUG_ON(sizeof_field(struct sock_common,
9704 skc_rcv_saddr) != 4);
9705
9706 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9707 si->dst_reg, si->src_reg,
9708 offsetof(struct sk_buff, sk));
9709 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9710 bpf_target_off(struct sock_common,
9711 skc_rcv_saddr,
9712 4, target_size));
9713 break;
9714 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9715 offsetof(struct __sk_buff, remote_ip6[3]):
9716 #if IS_ENABLED(CONFIG_IPV6)
9717 BUILD_BUG_ON(sizeof_field(struct sock_common,
9718 skc_v6_daddr.s6_addr32[0]) != 4);
9719
9720 off = si->off;
9721 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9722
9723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9724 si->dst_reg, si->src_reg,
9725 offsetof(struct sk_buff, sk));
9726 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9727 offsetof(struct sock_common,
9728 skc_v6_daddr.s6_addr32[0]) +
9729 off);
9730 #else
9731 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9732 #endif
9733 break;
9734 case offsetof(struct __sk_buff, local_ip6[0]) ...
9735 offsetof(struct __sk_buff, local_ip6[3]):
9736 #if IS_ENABLED(CONFIG_IPV6)
9737 BUILD_BUG_ON(sizeof_field(struct sock_common,
9738 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9739
9740 off = si->off;
9741 off -= offsetof(struct __sk_buff, local_ip6[0]);
9742
9743 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9744 si->dst_reg, si->src_reg,
9745 offsetof(struct sk_buff, sk));
9746 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9747 offsetof(struct sock_common,
9748 skc_v6_rcv_saddr.s6_addr32[0]) +
9749 off);
9750 #else
9751 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9752 #endif
9753 break;
9754
9755 case offsetof(struct __sk_buff, remote_port):
9756 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9757
9758 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9759 si->dst_reg, si->src_reg,
9760 offsetof(struct sk_buff, sk));
9761 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9762 bpf_target_off(struct sock_common,
9763 skc_dport,
9764 2, target_size));
9765 #ifndef __BIG_ENDIAN_BITFIELD
9766 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9767 #endif
9768 break;
9769
9770 case offsetof(struct __sk_buff, local_port):
9771 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9772
9773 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9774 si->dst_reg, si->src_reg,
9775 offsetof(struct sk_buff, sk));
9776 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9777 bpf_target_off(struct sock_common,
9778 skc_num, 2, target_size));
9779 break;
9780
9781 case offsetof(struct __sk_buff, tstamp):
9782 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9783
9784 if (type == BPF_WRITE)
9785 insn = bpf_convert_tstamp_write(prog, si, insn);
9786 else
9787 insn = bpf_convert_tstamp_read(prog, si, insn);
9788 break;
9789
9790 case offsetof(struct __sk_buff, tstamp_type):
9791 insn = bpf_convert_tstamp_type_read(si, insn);
9792 break;
9793
9794 case offsetof(struct __sk_buff, gso_segs):
9795 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9796 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9797 si->dst_reg, si->dst_reg,
9798 bpf_target_off(struct skb_shared_info,
9799 gso_segs, 2,
9800 target_size));
9801 break;
9802 case offsetof(struct __sk_buff, gso_size):
9803 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9804 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9805 si->dst_reg, si->dst_reg,
9806 bpf_target_off(struct skb_shared_info,
9807 gso_size, 2,
9808 target_size));
9809 break;
9810 case offsetof(struct __sk_buff, wire_len):
9811 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9812
9813 off = si->off;
9814 off -= offsetof(struct __sk_buff, wire_len);
9815 off += offsetof(struct sk_buff, cb);
9816 off += offsetof(struct qdisc_skb_cb, pkt_len);
9817 *target_size = 4;
9818 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9819 break;
9820
9821 case offsetof(struct __sk_buff, sk):
9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9823 si->dst_reg, si->src_reg,
9824 offsetof(struct sk_buff, sk));
9825 break;
9826 case offsetof(struct __sk_buff, hwtstamp):
9827 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9828 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9829
9830 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9831 *insn++ = BPF_LDX_MEM(BPF_DW,
9832 si->dst_reg, si->dst_reg,
9833 bpf_target_off(struct skb_shared_info,
9834 hwtstamps, 8,
9835 target_size));
9836 break;
9837 }
9838
9839 return insn - insn_buf;
9840 }
9841
bpf_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9842 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9843 const struct bpf_insn *si,
9844 struct bpf_insn *insn_buf,
9845 struct bpf_prog *prog, u32 *target_size)
9846 {
9847 struct bpf_insn *insn = insn_buf;
9848 int off;
9849
9850 switch (si->off) {
9851 case offsetof(struct bpf_sock, bound_dev_if):
9852 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9853
9854 if (type == BPF_WRITE)
9855 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9856 offsetof(struct sock, sk_bound_dev_if));
9857 else
9858 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9859 offsetof(struct sock, sk_bound_dev_if));
9860 break;
9861
9862 case offsetof(struct bpf_sock, mark):
9863 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9864
9865 if (type == BPF_WRITE)
9866 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9867 offsetof(struct sock, sk_mark));
9868 else
9869 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9870 offsetof(struct sock, sk_mark));
9871 break;
9872
9873 case offsetof(struct bpf_sock, priority):
9874 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9875
9876 if (type == BPF_WRITE)
9877 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9878 offsetof(struct sock, sk_priority));
9879 else
9880 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9881 offsetof(struct sock, sk_priority));
9882 break;
9883
9884 case offsetof(struct bpf_sock, family):
9885 *insn++ = BPF_LDX_MEM(
9886 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9887 si->dst_reg, si->src_reg,
9888 bpf_target_off(struct sock_common,
9889 skc_family,
9890 sizeof_field(struct sock_common,
9891 skc_family),
9892 target_size));
9893 break;
9894
9895 case offsetof(struct bpf_sock, type):
9896 *insn++ = BPF_LDX_MEM(
9897 BPF_FIELD_SIZEOF(struct sock, sk_type),
9898 si->dst_reg, si->src_reg,
9899 bpf_target_off(struct sock, sk_type,
9900 sizeof_field(struct sock, sk_type),
9901 target_size));
9902 break;
9903
9904 case offsetof(struct bpf_sock, protocol):
9905 *insn++ = BPF_LDX_MEM(
9906 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9907 si->dst_reg, si->src_reg,
9908 bpf_target_off(struct sock, sk_protocol,
9909 sizeof_field(struct sock, sk_protocol),
9910 target_size));
9911 break;
9912
9913 case offsetof(struct bpf_sock, src_ip4):
9914 *insn++ = BPF_LDX_MEM(
9915 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9916 bpf_target_off(struct sock_common, skc_rcv_saddr,
9917 sizeof_field(struct sock_common,
9918 skc_rcv_saddr),
9919 target_size));
9920 break;
9921
9922 case offsetof(struct bpf_sock, dst_ip4):
9923 *insn++ = BPF_LDX_MEM(
9924 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9925 bpf_target_off(struct sock_common, skc_daddr,
9926 sizeof_field(struct sock_common,
9927 skc_daddr),
9928 target_size));
9929 break;
9930
9931 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9932 #if IS_ENABLED(CONFIG_IPV6)
9933 off = si->off;
9934 off -= offsetof(struct bpf_sock, src_ip6[0]);
9935 *insn++ = BPF_LDX_MEM(
9936 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9937 bpf_target_off(
9938 struct sock_common,
9939 skc_v6_rcv_saddr.s6_addr32[0],
9940 sizeof_field(struct sock_common,
9941 skc_v6_rcv_saddr.s6_addr32[0]),
9942 target_size) + off);
9943 #else
9944 (void)off;
9945 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9946 #endif
9947 break;
9948
9949 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9950 #if IS_ENABLED(CONFIG_IPV6)
9951 off = si->off;
9952 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9953 *insn++ = BPF_LDX_MEM(
9954 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9955 bpf_target_off(struct sock_common,
9956 skc_v6_daddr.s6_addr32[0],
9957 sizeof_field(struct sock_common,
9958 skc_v6_daddr.s6_addr32[0]),
9959 target_size) + off);
9960 #else
9961 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9962 *target_size = 4;
9963 #endif
9964 break;
9965
9966 case offsetof(struct bpf_sock, src_port):
9967 *insn++ = BPF_LDX_MEM(
9968 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9969 si->dst_reg, si->src_reg,
9970 bpf_target_off(struct sock_common, skc_num,
9971 sizeof_field(struct sock_common,
9972 skc_num),
9973 target_size));
9974 break;
9975
9976 case offsetof(struct bpf_sock, dst_port):
9977 *insn++ = BPF_LDX_MEM(
9978 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9979 si->dst_reg, si->src_reg,
9980 bpf_target_off(struct sock_common, skc_dport,
9981 sizeof_field(struct sock_common,
9982 skc_dport),
9983 target_size));
9984 break;
9985
9986 case offsetof(struct bpf_sock, state):
9987 *insn++ = BPF_LDX_MEM(
9988 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9989 si->dst_reg, si->src_reg,
9990 bpf_target_off(struct sock_common, skc_state,
9991 sizeof_field(struct sock_common,
9992 skc_state),
9993 target_size));
9994 break;
9995 case offsetof(struct bpf_sock, rx_queue_mapping):
9996 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9997 *insn++ = BPF_LDX_MEM(
9998 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9999 si->dst_reg, si->src_reg,
10000 bpf_target_off(struct sock, sk_rx_queue_mapping,
10001 sizeof_field(struct sock,
10002 sk_rx_queue_mapping),
10003 target_size));
10004 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
10005 1);
10006 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10007 #else
10008 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10009 *target_size = 2;
10010 #endif
10011 break;
10012 }
10013
10014 return insn - insn_buf;
10015 }
10016
tc_cls_act_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10017 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
10018 const struct bpf_insn *si,
10019 struct bpf_insn *insn_buf,
10020 struct bpf_prog *prog, u32 *target_size)
10021 {
10022 struct bpf_insn *insn = insn_buf;
10023
10024 switch (si->off) {
10025 case offsetof(struct __sk_buff, ifindex):
10026 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
10027 si->dst_reg, si->src_reg,
10028 offsetof(struct sk_buff, dev));
10029 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10030 bpf_target_off(struct net_device, ifindex, 4,
10031 target_size));
10032 break;
10033 default:
10034 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10035 target_size);
10036 }
10037
10038 return insn - insn_buf;
10039 }
10040
xdp_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10041 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
10042 const struct bpf_insn *si,
10043 struct bpf_insn *insn_buf,
10044 struct bpf_prog *prog, u32 *target_size)
10045 {
10046 struct bpf_insn *insn = insn_buf;
10047
10048 switch (si->off) {
10049 case offsetof(struct xdp_md, data):
10050 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
10051 si->dst_reg, si->src_reg,
10052 offsetof(struct xdp_buff, data));
10053 break;
10054 case offsetof(struct xdp_md, data_meta):
10055 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
10056 si->dst_reg, si->src_reg,
10057 offsetof(struct xdp_buff, data_meta));
10058 break;
10059 case offsetof(struct xdp_md, data_end):
10060 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
10061 si->dst_reg, si->src_reg,
10062 offsetof(struct xdp_buff, data_end));
10063 break;
10064 case offsetof(struct xdp_md, ingress_ifindex):
10065 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10066 si->dst_reg, si->src_reg,
10067 offsetof(struct xdp_buff, rxq));
10068 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
10069 si->dst_reg, si->dst_reg,
10070 offsetof(struct xdp_rxq_info, dev));
10071 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10072 offsetof(struct net_device, ifindex));
10073 break;
10074 case offsetof(struct xdp_md, rx_queue_index):
10075 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10076 si->dst_reg, si->src_reg,
10077 offsetof(struct xdp_buff, rxq));
10078 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10079 offsetof(struct xdp_rxq_info,
10080 queue_index));
10081 break;
10082 case offsetof(struct xdp_md, egress_ifindex):
10083 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
10084 si->dst_reg, si->src_reg,
10085 offsetof(struct xdp_buff, txq));
10086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10087 si->dst_reg, si->dst_reg,
10088 offsetof(struct xdp_txq_info, dev));
10089 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10090 offsetof(struct net_device, ifindex));
10091 break;
10092 }
10093
10094 return insn - insn_buf;
10095 }
10096
10097 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10098 * context Structure, F is Field in context structure that contains a pointer
10099 * to Nested Structure of type NS that has the field NF.
10100 *
10101 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10102 * sure that SIZE is not greater than actual size of S.F.NF.
10103 *
10104 * If offset OFF is provided, the load happens from that offset relative to
10105 * offset of NF.
10106 */
10107 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10108 do { \
10109 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10110 si->src_reg, offsetof(S, F)); \
10111 *insn++ = BPF_LDX_MEM( \
10112 SIZE, si->dst_reg, si->dst_reg, \
10113 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10114 target_size) \
10115 + OFF); \
10116 } while (0)
10117
10118 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10119 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10120 BPF_FIELD_SIZEOF(NS, NF), 0)
10121
10122 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10123 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10124 *
10125 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10126 * "register" since two registers available in convert_ctx_access are not
10127 * enough: we can't override neither SRC, since it contains value to store, nor
10128 * DST since it contains pointer to context that may be used by later
10129 * instructions. But we need a temporary place to save pointer to nested
10130 * structure whose field we want to store to.
10131 */
10132 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10133 do { \
10134 int tmp_reg = BPF_REG_9; \
10135 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10136 --tmp_reg; \
10137 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10138 --tmp_reg; \
10139 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10140 offsetof(S, TF)); \
10141 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10142 si->dst_reg, offsetof(S, F)); \
10143 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10144 tmp_reg, si->src_reg, \
10145 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10146 target_size) \
10147 + OFF, \
10148 si->imm); \
10149 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10150 offsetof(S, TF)); \
10151 } while (0)
10152
10153 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10154 TF) \
10155 do { \
10156 if (type == BPF_WRITE) { \
10157 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10158 OFF, TF); \
10159 } else { \
10160 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10161 S, NS, F, NF, SIZE, OFF); \
10162 } \
10163 } while (0)
10164
10165 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10166 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10167 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10168
sock_addr_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10169 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10170 const struct bpf_insn *si,
10171 struct bpf_insn *insn_buf,
10172 struct bpf_prog *prog, u32 *target_size)
10173 {
10174 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10175 struct bpf_insn *insn = insn_buf;
10176
10177 switch (si->off) {
10178 case offsetof(struct bpf_sock_addr, user_family):
10179 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10180 struct sockaddr, uaddr, sa_family);
10181 break;
10182
10183 case offsetof(struct bpf_sock_addr, user_ip4):
10184 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10185 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10186 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10187 break;
10188
10189 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10190 off = si->off;
10191 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10192 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10193 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10194 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10195 tmp_reg);
10196 break;
10197
10198 case offsetof(struct bpf_sock_addr, user_port):
10199 /* To get port we need to know sa_family first and then treat
10200 * sockaddr as either sockaddr_in or sockaddr_in6.
10201 * Though we can simplify since port field has same offset and
10202 * size in both structures.
10203 * Here we check this invariant and use just one of the
10204 * structures if it's true.
10205 */
10206 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10207 offsetof(struct sockaddr_in6, sin6_port));
10208 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10209 sizeof_field(struct sockaddr_in6, sin6_port));
10210 /* Account for sin6_port being smaller than user_port. */
10211 port_size = min(port_size, BPF_LDST_BYTES(si));
10212 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10213 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10214 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10215 break;
10216
10217 case offsetof(struct bpf_sock_addr, family):
10218 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10219 struct sock, sk, sk_family);
10220 break;
10221
10222 case offsetof(struct bpf_sock_addr, type):
10223 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10224 struct sock, sk, sk_type);
10225 break;
10226
10227 case offsetof(struct bpf_sock_addr, protocol):
10228 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10229 struct sock, sk, sk_protocol);
10230 break;
10231
10232 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10233 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10234 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10235 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10236 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10237 break;
10238
10239 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10240 msg_src_ip6[3]):
10241 off = si->off;
10242 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10243 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10244 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10245 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10246 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10247 break;
10248 case offsetof(struct bpf_sock_addr, sk):
10249 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10250 si->dst_reg, si->src_reg,
10251 offsetof(struct bpf_sock_addr_kern, sk));
10252 break;
10253 }
10254
10255 return insn - insn_buf;
10256 }
10257
sock_ops_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10258 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10259 const struct bpf_insn *si,
10260 struct bpf_insn *insn_buf,
10261 struct bpf_prog *prog,
10262 u32 *target_size)
10263 {
10264 struct bpf_insn *insn = insn_buf;
10265 int off;
10266
10267 /* Helper macro for adding read access to tcp_sock or sock fields. */
10268 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10269 do { \
10270 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10271 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10272 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10273 if (si->dst_reg == reg || si->src_reg == reg) \
10274 reg--; \
10275 if (si->dst_reg == reg || si->src_reg == reg) \
10276 reg--; \
10277 if (si->dst_reg == si->src_reg) { \
10278 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10279 offsetof(struct bpf_sock_ops_kern, \
10280 temp)); \
10281 fullsock_reg = reg; \
10282 jmp += 2; \
10283 } \
10284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10285 struct bpf_sock_ops_kern, \
10286 is_fullsock), \
10287 fullsock_reg, si->src_reg, \
10288 offsetof(struct bpf_sock_ops_kern, \
10289 is_fullsock)); \
10290 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10291 if (si->dst_reg == si->src_reg) \
10292 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10293 offsetof(struct bpf_sock_ops_kern, \
10294 temp)); \
10295 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10296 struct bpf_sock_ops_kern, sk),\
10297 si->dst_reg, si->src_reg, \
10298 offsetof(struct bpf_sock_ops_kern, sk));\
10299 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10300 OBJ_FIELD), \
10301 si->dst_reg, si->dst_reg, \
10302 offsetof(OBJ, OBJ_FIELD)); \
10303 if (si->dst_reg == si->src_reg) { \
10304 *insn++ = BPF_JMP_A(1); \
10305 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10306 offsetof(struct bpf_sock_ops_kern, \
10307 temp)); \
10308 } \
10309 } while (0)
10310
10311 #define SOCK_OPS_GET_SK() \
10312 do { \
10313 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10314 if (si->dst_reg == reg || si->src_reg == reg) \
10315 reg--; \
10316 if (si->dst_reg == reg || si->src_reg == reg) \
10317 reg--; \
10318 if (si->dst_reg == si->src_reg) { \
10319 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10320 offsetof(struct bpf_sock_ops_kern, \
10321 temp)); \
10322 fullsock_reg = reg; \
10323 jmp += 2; \
10324 } \
10325 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10326 struct bpf_sock_ops_kern, \
10327 is_fullsock), \
10328 fullsock_reg, si->src_reg, \
10329 offsetof(struct bpf_sock_ops_kern, \
10330 is_fullsock)); \
10331 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10332 if (si->dst_reg == si->src_reg) \
10333 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10334 offsetof(struct bpf_sock_ops_kern, \
10335 temp)); \
10336 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10337 struct bpf_sock_ops_kern, sk),\
10338 si->dst_reg, si->src_reg, \
10339 offsetof(struct bpf_sock_ops_kern, sk));\
10340 if (si->dst_reg == si->src_reg) { \
10341 *insn++ = BPF_JMP_A(1); \
10342 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10343 offsetof(struct bpf_sock_ops_kern, \
10344 temp)); \
10345 } \
10346 } while (0)
10347
10348 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10349 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10350
10351 /* Helper macro for adding write access to tcp_sock or sock fields.
10352 * The macro is called with two registers, dst_reg which contains a pointer
10353 * to ctx (context) and src_reg which contains the value that should be
10354 * stored. However, we need an additional register since we cannot overwrite
10355 * dst_reg because it may be used later in the program.
10356 * Instead we "borrow" one of the other register. We first save its value
10357 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10358 * it at the end of the macro.
10359 */
10360 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10361 do { \
10362 int reg = BPF_REG_9; \
10363 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10364 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10365 if (si->dst_reg == reg || si->src_reg == reg) \
10366 reg--; \
10367 if (si->dst_reg == reg || si->src_reg == reg) \
10368 reg--; \
10369 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10370 offsetof(struct bpf_sock_ops_kern, \
10371 temp)); \
10372 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10373 struct bpf_sock_ops_kern, \
10374 is_fullsock), \
10375 reg, si->dst_reg, \
10376 offsetof(struct bpf_sock_ops_kern, \
10377 is_fullsock)); \
10378 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10379 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10380 struct bpf_sock_ops_kern, sk),\
10381 reg, si->dst_reg, \
10382 offsetof(struct bpf_sock_ops_kern, sk));\
10383 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10384 BPF_MEM | BPF_CLASS(si->code), \
10385 reg, si->src_reg, \
10386 offsetof(OBJ, OBJ_FIELD), \
10387 si->imm); \
10388 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10389 offsetof(struct bpf_sock_ops_kern, \
10390 temp)); \
10391 } while (0)
10392
10393 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10394 do { \
10395 if (TYPE == BPF_WRITE) \
10396 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10397 else \
10398 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10399 } while (0)
10400
10401 switch (si->off) {
10402 case offsetof(struct bpf_sock_ops, op):
10403 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10404 op),
10405 si->dst_reg, si->src_reg,
10406 offsetof(struct bpf_sock_ops_kern, op));
10407 break;
10408
10409 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10410 offsetof(struct bpf_sock_ops, replylong[3]):
10411 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10412 sizeof_field(struct bpf_sock_ops_kern, reply));
10413 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10414 sizeof_field(struct bpf_sock_ops_kern, replylong));
10415 off = si->off;
10416 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10417 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10418 if (type == BPF_WRITE)
10419 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10420 else
10421 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10422 off);
10423 break;
10424
10425 case offsetof(struct bpf_sock_ops, family):
10426 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10427
10428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10429 struct bpf_sock_ops_kern, sk),
10430 si->dst_reg, si->src_reg,
10431 offsetof(struct bpf_sock_ops_kern, sk));
10432 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10433 offsetof(struct sock_common, skc_family));
10434 break;
10435
10436 case offsetof(struct bpf_sock_ops, remote_ip4):
10437 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10438
10439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10440 struct bpf_sock_ops_kern, sk),
10441 si->dst_reg, si->src_reg,
10442 offsetof(struct bpf_sock_ops_kern, sk));
10443 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10444 offsetof(struct sock_common, skc_daddr));
10445 break;
10446
10447 case offsetof(struct bpf_sock_ops, local_ip4):
10448 BUILD_BUG_ON(sizeof_field(struct sock_common,
10449 skc_rcv_saddr) != 4);
10450
10451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10452 struct bpf_sock_ops_kern, sk),
10453 si->dst_reg, si->src_reg,
10454 offsetof(struct bpf_sock_ops_kern, sk));
10455 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10456 offsetof(struct sock_common,
10457 skc_rcv_saddr));
10458 break;
10459
10460 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10461 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10462 #if IS_ENABLED(CONFIG_IPV6)
10463 BUILD_BUG_ON(sizeof_field(struct sock_common,
10464 skc_v6_daddr.s6_addr32[0]) != 4);
10465
10466 off = si->off;
10467 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10469 struct bpf_sock_ops_kern, sk),
10470 si->dst_reg, si->src_reg,
10471 offsetof(struct bpf_sock_ops_kern, sk));
10472 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10473 offsetof(struct sock_common,
10474 skc_v6_daddr.s6_addr32[0]) +
10475 off);
10476 #else
10477 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10478 #endif
10479 break;
10480
10481 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10482 offsetof(struct bpf_sock_ops, local_ip6[3]):
10483 #if IS_ENABLED(CONFIG_IPV6)
10484 BUILD_BUG_ON(sizeof_field(struct sock_common,
10485 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10486
10487 off = si->off;
10488 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10489 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10490 struct bpf_sock_ops_kern, sk),
10491 si->dst_reg, si->src_reg,
10492 offsetof(struct bpf_sock_ops_kern, sk));
10493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10494 offsetof(struct sock_common,
10495 skc_v6_rcv_saddr.s6_addr32[0]) +
10496 off);
10497 #else
10498 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10499 #endif
10500 break;
10501
10502 case offsetof(struct bpf_sock_ops, remote_port):
10503 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10504
10505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10506 struct bpf_sock_ops_kern, sk),
10507 si->dst_reg, si->src_reg,
10508 offsetof(struct bpf_sock_ops_kern, sk));
10509 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10510 offsetof(struct sock_common, skc_dport));
10511 #ifndef __BIG_ENDIAN_BITFIELD
10512 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10513 #endif
10514 break;
10515
10516 case offsetof(struct bpf_sock_ops, local_port):
10517 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10518
10519 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10520 struct bpf_sock_ops_kern, sk),
10521 si->dst_reg, si->src_reg,
10522 offsetof(struct bpf_sock_ops_kern, sk));
10523 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10524 offsetof(struct sock_common, skc_num));
10525 break;
10526
10527 case offsetof(struct bpf_sock_ops, is_fullsock):
10528 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10529 struct bpf_sock_ops_kern,
10530 is_fullsock),
10531 si->dst_reg, si->src_reg,
10532 offsetof(struct bpf_sock_ops_kern,
10533 is_fullsock));
10534 break;
10535
10536 case offsetof(struct bpf_sock_ops, state):
10537 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10538
10539 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10540 struct bpf_sock_ops_kern, sk),
10541 si->dst_reg, si->src_reg,
10542 offsetof(struct bpf_sock_ops_kern, sk));
10543 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10544 offsetof(struct sock_common, skc_state));
10545 break;
10546
10547 case offsetof(struct bpf_sock_ops, rtt_min):
10548 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10549 sizeof(struct minmax));
10550 BUILD_BUG_ON(sizeof(struct minmax) <
10551 sizeof(struct minmax_sample));
10552
10553 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10554 struct bpf_sock_ops_kern, sk),
10555 si->dst_reg, si->src_reg,
10556 offsetof(struct bpf_sock_ops_kern, sk));
10557 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10558 offsetof(struct tcp_sock, rtt_min) +
10559 sizeof_field(struct minmax_sample, t));
10560 break;
10561
10562 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10563 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10564 struct tcp_sock);
10565 break;
10566
10567 case offsetof(struct bpf_sock_ops, sk_txhash):
10568 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10569 struct sock, type);
10570 break;
10571 case offsetof(struct bpf_sock_ops, snd_cwnd):
10572 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10573 break;
10574 case offsetof(struct bpf_sock_ops, srtt_us):
10575 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10576 break;
10577 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10578 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10579 break;
10580 case offsetof(struct bpf_sock_ops, rcv_nxt):
10581 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10582 break;
10583 case offsetof(struct bpf_sock_ops, snd_nxt):
10584 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10585 break;
10586 case offsetof(struct bpf_sock_ops, snd_una):
10587 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10588 break;
10589 case offsetof(struct bpf_sock_ops, mss_cache):
10590 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10591 break;
10592 case offsetof(struct bpf_sock_ops, ecn_flags):
10593 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10594 break;
10595 case offsetof(struct bpf_sock_ops, rate_delivered):
10596 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10597 break;
10598 case offsetof(struct bpf_sock_ops, rate_interval_us):
10599 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10600 break;
10601 case offsetof(struct bpf_sock_ops, packets_out):
10602 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10603 break;
10604 case offsetof(struct bpf_sock_ops, retrans_out):
10605 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10606 break;
10607 case offsetof(struct bpf_sock_ops, total_retrans):
10608 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10609 break;
10610 case offsetof(struct bpf_sock_ops, segs_in):
10611 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10612 break;
10613 case offsetof(struct bpf_sock_ops, data_segs_in):
10614 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10615 break;
10616 case offsetof(struct bpf_sock_ops, segs_out):
10617 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10618 break;
10619 case offsetof(struct bpf_sock_ops, data_segs_out):
10620 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10621 break;
10622 case offsetof(struct bpf_sock_ops, lost_out):
10623 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10624 break;
10625 case offsetof(struct bpf_sock_ops, sacked_out):
10626 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10627 break;
10628 case offsetof(struct bpf_sock_ops, bytes_received):
10629 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10630 break;
10631 case offsetof(struct bpf_sock_ops, bytes_acked):
10632 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10633 break;
10634 case offsetof(struct bpf_sock_ops, sk):
10635 SOCK_OPS_GET_SK();
10636 break;
10637 case offsetof(struct bpf_sock_ops, skb_data_end):
10638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10639 skb_data_end),
10640 si->dst_reg, si->src_reg,
10641 offsetof(struct bpf_sock_ops_kern,
10642 skb_data_end));
10643 break;
10644 case offsetof(struct bpf_sock_ops, skb_data):
10645 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10646 skb),
10647 si->dst_reg, si->src_reg,
10648 offsetof(struct bpf_sock_ops_kern,
10649 skb));
10650 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10651 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10652 si->dst_reg, si->dst_reg,
10653 offsetof(struct sk_buff, data));
10654 break;
10655 case offsetof(struct bpf_sock_ops, skb_len):
10656 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10657 skb),
10658 si->dst_reg, si->src_reg,
10659 offsetof(struct bpf_sock_ops_kern,
10660 skb));
10661 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10662 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10663 si->dst_reg, si->dst_reg,
10664 offsetof(struct sk_buff, len));
10665 break;
10666 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10667 off = offsetof(struct sk_buff, cb);
10668 off += offsetof(struct tcp_skb_cb, tcp_flags);
10669 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10670 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10671 skb),
10672 si->dst_reg, si->src_reg,
10673 offsetof(struct bpf_sock_ops_kern,
10674 skb));
10675 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10676 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10677 tcp_flags),
10678 si->dst_reg, si->dst_reg, off);
10679 break;
10680 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10681 struct bpf_insn *jmp_on_null_skb;
10682
10683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10684 skb),
10685 si->dst_reg, si->src_reg,
10686 offsetof(struct bpf_sock_ops_kern,
10687 skb));
10688 /* Reserve one insn to test skb == NULL */
10689 jmp_on_null_skb = insn++;
10690 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10691 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10692 bpf_target_off(struct skb_shared_info,
10693 hwtstamps, 8,
10694 target_size));
10695 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10696 insn - jmp_on_null_skb - 1);
10697 break;
10698 }
10699 }
10700 return insn - insn_buf;
10701 }
10702
10703 /* data_end = skb->data + skb_headlen() */
bpf_convert_data_end_access(const struct bpf_insn * si,struct bpf_insn * insn)10704 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10705 struct bpf_insn *insn)
10706 {
10707 int reg;
10708 int temp_reg_off = offsetof(struct sk_buff, cb) +
10709 offsetof(struct sk_skb_cb, temp_reg);
10710
10711 if (si->src_reg == si->dst_reg) {
10712 /* We need an extra register, choose and save a register. */
10713 reg = BPF_REG_9;
10714 if (si->src_reg == reg || si->dst_reg == reg)
10715 reg--;
10716 if (si->src_reg == reg || si->dst_reg == reg)
10717 reg--;
10718 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10719 } else {
10720 reg = si->dst_reg;
10721 }
10722
10723 /* reg = skb->data */
10724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10725 reg, si->src_reg,
10726 offsetof(struct sk_buff, data));
10727 /* AX = skb->len */
10728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10729 BPF_REG_AX, si->src_reg,
10730 offsetof(struct sk_buff, len));
10731 /* reg = skb->data + skb->len */
10732 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10733 /* AX = skb->data_len */
10734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10735 BPF_REG_AX, si->src_reg,
10736 offsetof(struct sk_buff, data_len));
10737
10738 /* reg = skb->data + skb->len - skb->data_len */
10739 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10740
10741 if (si->src_reg == si->dst_reg) {
10742 /* Restore the saved register */
10743 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10744 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10745 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10746 }
10747
10748 return insn;
10749 }
10750
sk_skb_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10751 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10752 const struct bpf_insn *si,
10753 struct bpf_insn *insn_buf,
10754 struct bpf_prog *prog, u32 *target_size)
10755 {
10756 struct bpf_insn *insn = insn_buf;
10757 int off;
10758
10759 switch (si->off) {
10760 case offsetof(struct __sk_buff, data_end):
10761 insn = bpf_convert_data_end_access(si, insn);
10762 break;
10763 case offsetof(struct __sk_buff, cb[0]) ...
10764 offsetofend(struct __sk_buff, cb[4]) - 1:
10765 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10766 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10767 offsetof(struct sk_skb_cb, data)) %
10768 sizeof(__u64));
10769
10770 prog->cb_access = 1;
10771 off = si->off;
10772 off -= offsetof(struct __sk_buff, cb[0]);
10773 off += offsetof(struct sk_buff, cb);
10774 off += offsetof(struct sk_skb_cb, data);
10775 if (type == BPF_WRITE)
10776 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10777 else
10778 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10779 si->src_reg, off);
10780 break;
10781
10782
10783 default:
10784 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10785 target_size);
10786 }
10787
10788 return insn - insn_buf;
10789 }
10790
sk_msg_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10791 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10792 const struct bpf_insn *si,
10793 struct bpf_insn *insn_buf,
10794 struct bpf_prog *prog, u32 *target_size)
10795 {
10796 struct bpf_insn *insn = insn_buf;
10797 #if IS_ENABLED(CONFIG_IPV6)
10798 int off;
10799 #endif
10800
10801 /* convert ctx uses the fact sg element is first in struct */
10802 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10803
10804 switch (si->off) {
10805 case offsetof(struct sk_msg_md, data):
10806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10807 si->dst_reg, si->src_reg,
10808 offsetof(struct sk_msg, data));
10809 break;
10810 case offsetof(struct sk_msg_md, data_end):
10811 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10812 si->dst_reg, si->src_reg,
10813 offsetof(struct sk_msg, data_end));
10814 break;
10815 case offsetof(struct sk_msg_md, family):
10816 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10817
10818 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10819 struct sk_msg, sk),
10820 si->dst_reg, si->src_reg,
10821 offsetof(struct sk_msg, sk));
10822 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10823 offsetof(struct sock_common, skc_family));
10824 break;
10825
10826 case offsetof(struct sk_msg_md, remote_ip4):
10827 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10828
10829 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10830 struct sk_msg, sk),
10831 si->dst_reg, si->src_reg,
10832 offsetof(struct sk_msg, sk));
10833 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10834 offsetof(struct sock_common, skc_daddr));
10835 break;
10836
10837 case offsetof(struct sk_msg_md, local_ip4):
10838 BUILD_BUG_ON(sizeof_field(struct sock_common,
10839 skc_rcv_saddr) != 4);
10840
10841 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10842 struct sk_msg, sk),
10843 si->dst_reg, si->src_reg,
10844 offsetof(struct sk_msg, sk));
10845 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10846 offsetof(struct sock_common,
10847 skc_rcv_saddr));
10848 break;
10849
10850 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10851 offsetof(struct sk_msg_md, remote_ip6[3]):
10852 #if IS_ENABLED(CONFIG_IPV6)
10853 BUILD_BUG_ON(sizeof_field(struct sock_common,
10854 skc_v6_daddr.s6_addr32[0]) != 4);
10855
10856 off = si->off;
10857 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10858 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10859 struct sk_msg, sk),
10860 si->dst_reg, si->src_reg,
10861 offsetof(struct sk_msg, sk));
10862 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10863 offsetof(struct sock_common,
10864 skc_v6_daddr.s6_addr32[0]) +
10865 off);
10866 #else
10867 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10868 #endif
10869 break;
10870
10871 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10872 offsetof(struct sk_msg_md, local_ip6[3]):
10873 #if IS_ENABLED(CONFIG_IPV6)
10874 BUILD_BUG_ON(sizeof_field(struct sock_common,
10875 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10876
10877 off = si->off;
10878 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10880 struct sk_msg, sk),
10881 si->dst_reg, si->src_reg,
10882 offsetof(struct sk_msg, sk));
10883 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10884 offsetof(struct sock_common,
10885 skc_v6_rcv_saddr.s6_addr32[0]) +
10886 off);
10887 #else
10888 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10889 #endif
10890 break;
10891
10892 case offsetof(struct sk_msg_md, remote_port):
10893 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10894
10895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10896 struct sk_msg, sk),
10897 si->dst_reg, si->src_reg,
10898 offsetof(struct sk_msg, sk));
10899 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10900 offsetof(struct sock_common, skc_dport));
10901 #ifndef __BIG_ENDIAN_BITFIELD
10902 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10903 #endif
10904 break;
10905
10906 case offsetof(struct sk_msg_md, local_port):
10907 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10908
10909 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10910 struct sk_msg, sk),
10911 si->dst_reg, si->src_reg,
10912 offsetof(struct sk_msg, sk));
10913 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10914 offsetof(struct sock_common, skc_num));
10915 break;
10916
10917 case offsetof(struct sk_msg_md, size):
10918 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10919 si->dst_reg, si->src_reg,
10920 offsetof(struct sk_msg_sg, size));
10921 break;
10922
10923 case offsetof(struct sk_msg_md, sk):
10924 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10925 si->dst_reg, si->src_reg,
10926 offsetof(struct sk_msg, sk));
10927 break;
10928 }
10929
10930 return insn - insn_buf;
10931 }
10932
10933 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10934 .get_func_proto = sk_filter_func_proto,
10935 .is_valid_access = sk_filter_is_valid_access,
10936 .convert_ctx_access = bpf_convert_ctx_access,
10937 .gen_ld_abs = bpf_gen_ld_abs,
10938 };
10939
10940 const struct bpf_prog_ops sk_filter_prog_ops = {
10941 .test_run = bpf_prog_test_run_skb,
10942 };
10943
10944 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10945 .get_func_proto = tc_cls_act_func_proto,
10946 .is_valid_access = tc_cls_act_is_valid_access,
10947 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10948 .gen_prologue = tc_cls_act_prologue,
10949 .gen_ld_abs = bpf_gen_ld_abs,
10950 .btf_struct_access = tc_cls_act_btf_struct_access,
10951 };
10952
10953 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10954 .test_run = bpf_prog_test_run_skb,
10955 };
10956
10957 const struct bpf_verifier_ops xdp_verifier_ops = {
10958 .get_func_proto = xdp_func_proto,
10959 .is_valid_access = xdp_is_valid_access,
10960 .convert_ctx_access = xdp_convert_ctx_access,
10961 .gen_prologue = bpf_noop_prologue,
10962 .btf_struct_access = xdp_btf_struct_access,
10963 };
10964
10965 const struct bpf_prog_ops xdp_prog_ops = {
10966 .test_run = bpf_prog_test_run_xdp,
10967 };
10968
10969 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10970 .get_func_proto = cg_skb_func_proto,
10971 .is_valid_access = cg_skb_is_valid_access,
10972 .convert_ctx_access = bpf_convert_ctx_access,
10973 };
10974
10975 const struct bpf_prog_ops cg_skb_prog_ops = {
10976 .test_run = bpf_prog_test_run_skb,
10977 };
10978
10979 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10980 .get_func_proto = lwt_in_func_proto,
10981 .is_valid_access = lwt_is_valid_access,
10982 .convert_ctx_access = bpf_convert_ctx_access,
10983 };
10984
10985 const struct bpf_prog_ops lwt_in_prog_ops = {
10986 .test_run = bpf_prog_test_run_skb,
10987 };
10988
10989 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10990 .get_func_proto = lwt_out_func_proto,
10991 .is_valid_access = lwt_is_valid_access,
10992 .convert_ctx_access = bpf_convert_ctx_access,
10993 };
10994
10995 const struct bpf_prog_ops lwt_out_prog_ops = {
10996 .test_run = bpf_prog_test_run_skb,
10997 };
10998
10999 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
11000 .get_func_proto = lwt_xmit_func_proto,
11001 .is_valid_access = lwt_is_valid_access,
11002 .convert_ctx_access = bpf_convert_ctx_access,
11003 .gen_prologue = tc_cls_act_prologue,
11004 };
11005
11006 const struct bpf_prog_ops lwt_xmit_prog_ops = {
11007 .test_run = bpf_prog_test_run_skb,
11008 };
11009
11010 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
11011 .get_func_proto = lwt_seg6local_func_proto,
11012 .is_valid_access = lwt_is_valid_access,
11013 .convert_ctx_access = bpf_convert_ctx_access,
11014 };
11015
11016 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
11017 .test_run = bpf_prog_test_run_skb,
11018 };
11019
11020 const struct bpf_verifier_ops cg_sock_verifier_ops = {
11021 .get_func_proto = sock_filter_func_proto,
11022 .is_valid_access = sock_filter_is_valid_access,
11023 .convert_ctx_access = bpf_sock_convert_ctx_access,
11024 };
11025
11026 const struct bpf_prog_ops cg_sock_prog_ops = {
11027 };
11028
11029 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
11030 .get_func_proto = sock_addr_func_proto,
11031 .is_valid_access = sock_addr_is_valid_access,
11032 .convert_ctx_access = sock_addr_convert_ctx_access,
11033 };
11034
11035 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
11036 };
11037
11038 const struct bpf_verifier_ops sock_ops_verifier_ops = {
11039 .get_func_proto = sock_ops_func_proto,
11040 .is_valid_access = sock_ops_is_valid_access,
11041 .convert_ctx_access = sock_ops_convert_ctx_access,
11042 };
11043
11044 const struct bpf_prog_ops sock_ops_prog_ops = {
11045 };
11046
11047 const struct bpf_verifier_ops sk_skb_verifier_ops = {
11048 .get_func_proto = sk_skb_func_proto,
11049 .is_valid_access = sk_skb_is_valid_access,
11050 .convert_ctx_access = sk_skb_convert_ctx_access,
11051 .gen_prologue = sk_skb_prologue,
11052 };
11053
11054 const struct bpf_prog_ops sk_skb_prog_ops = {
11055 };
11056
11057 const struct bpf_verifier_ops sk_msg_verifier_ops = {
11058 .get_func_proto = sk_msg_func_proto,
11059 .is_valid_access = sk_msg_is_valid_access,
11060 .convert_ctx_access = sk_msg_convert_ctx_access,
11061 .gen_prologue = bpf_noop_prologue,
11062 };
11063
11064 const struct bpf_prog_ops sk_msg_prog_ops = {
11065 };
11066
11067 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
11068 .get_func_proto = flow_dissector_func_proto,
11069 .is_valid_access = flow_dissector_is_valid_access,
11070 .convert_ctx_access = flow_dissector_convert_ctx_access,
11071 };
11072
11073 const struct bpf_prog_ops flow_dissector_prog_ops = {
11074 .test_run = bpf_prog_test_run_flow_dissector,
11075 };
11076
sk_detach_filter(struct sock * sk)11077 int sk_detach_filter(struct sock *sk)
11078 {
11079 int ret = -ENOENT;
11080 struct sk_filter *filter;
11081
11082 if (sock_flag(sk, SOCK_FILTER_LOCKED))
11083 return -EPERM;
11084
11085 filter = rcu_dereference_protected(sk->sk_filter,
11086 lockdep_sock_is_held(sk));
11087 if (filter) {
11088 RCU_INIT_POINTER(sk->sk_filter, NULL);
11089 sk_filter_uncharge(sk, filter);
11090 ret = 0;
11091 }
11092
11093 return ret;
11094 }
11095 EXPORT_SYMBOL_GPL(sk_detach_filter);
11096
sk_get_filter(struct sock * sk,sockptr_t optval,unsigned int len)11097 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11098 {
11099 struct sock_fprog_kern *fprog;
11100 struct sk_filter *filter;
11101 int ret = 0;
11102
11103 sockopt_lock_sock(sk);
11104 filter = rcu_dereference_protected(sk->sk_filter,
11105 lockdep_sock_is_held(sk));
11106 if (!filter)
11107 goto out;
11108
11109 /* We're copying the filter that has been originally attached,
11110 * so no conversion/decode needed anymore. eBPF programs that
11111 * have no original program cannot be dumped through this.
11112 */
11113 ret = -EACCES;
11114 fprog = filter->prog->orig_prog;
11115 if (!fprog)
11116 goto out;
11117
11118 ret = fprog->len;
11119 if (!len)
11120 /* User space only enquires number of filter blocks. */
11121 goto out;
11122
11123 ret = -EINVAL;
11124 if (len < fprog->len)
11125 goto out;
11126
11127 ret = -EFAULT;
11128 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11129 goto out;
11130
11131 /* Instead of bytes, the API requests to return the number
11132 * of filter blocks.
11133 */
11134 ret = fprog->len;
11135 out:
11136 sockopt_release_sock(sk);
11137 return ret;
11138 }
11139
11140 #ifdef CONFIG_INET
bpf_init_reuseport_kern(struct sk_reuseport_kern * reuse_kern,struct sock_reuseport * reuse,struct sock * sk,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11141 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11142 struct sock_reuseport *reuse,
11143 struct sock *sk, struct sk_buff *skb,
11144 struct sock *migrating_sk,
11145 u32 hash)
11146 {
11147 reuse_kern->skb = skb;
11148 reuse_kern->sk = sk;
11149 reuse_kern->selected_sk = NULL;
11150 reuse_kern->migrating_sk = migrating_sk;
11151 reuse_kern->data_end = skb->data + skb_headlen(skb);
11152 reuse_kern->hash = hash;
11153 reuse_kern->reuseport_id = reuse->reuseport_id;
11154 reuse_kern->bind_inany = reuse->bind_inany;
11155 }
11156
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11157 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11158 struct bpf_prog *prog, struct sk_buff *skb,
11159 struct sock *migrating_sk,
11160 u32 hash)
11161 {
11162 struct sk_reuseport_kern reuse_kern;
11163 enum sk_action action;
11164
11165 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11166 action = bpf_prog_run(prog, &reuse_kern);
11167
11168 if (action == SK_PASS)
11169 return reuse_kern.selected_sk;
11170 else
11171 return ERR_PTR(-ECONNREFUSED);
11172 }
11173
BPF_CALL_4(sk_select_reuseport,struct sk_reuseport_kern *,reuse_kern,struct bpf_map *,map,void *,key,u32,flags)11174 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11175 struct bpf_map *, map, void *, key, u32, flags)
11176 {
11177 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11178 struct sock_reuseport *reuse;
11179 struct sock *selected_sk;
11180
11181 selected_sk = map->ops->map_lookup_elem(map, key);
11182 if (!selected_sk)
11183 return -ENOENT;
11184
11185 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11186 if (!reuse) {
11187 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11188 if (sk_is_refcounted(selected_sk))
11189 sock_put(selected_sk);
11190
11191 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11192 * The only (!reuse) case here is - the sk has already been
11193 * unhashed (e.g. by close()), so treat it as -ENOENT.
11194 *
11195 * Other maps (e.g. sock_map) do not provide this guarantee and
11196 * the sk may never be in the reuseport group to begin with.
11197 */
11198 return is_sockarray ? -ENOENT : -EINVAL;
11199 }
11200
11201 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11202 struct sock *sk = reuse_kern->sk;
11203
11204 if (sk->sk_protocol != selected_sk->sk_protocol)
11205 return -EPROTOTYPE;
11206 else if (sk->sk_family != selected_sk->sk_family)
11207 return -EAFNOSUPPORT;
11208
11209 /* Catch all. Likely bound to a different sockaddr. */
11210 return -EBADFD;
11211 }
11212
11213 reuse_kern->selected_sk = selected_sk;
11214
11215 return 0;
11216 }
11217
11218 static const struct bpf_func_proto sk_select_reuseport_proto = {
11219 .func = sk_select_reuseport,
11220 .gpl_only = false,
11221 .ret_type = RET_INTEGER,
11222 .arg1_type = ARG_PTR_TO_CTX,
11223 .arg2_type = ARG_CONST_MAP_PTR,
11224 .arg3_type = ARG_PTR_TO_MAP_KEY,
11225 .arg4_type = ARG_ANYTHING,
11226 };
11227
BPF_CALL_4(sk_reuseport_load_bytes,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len)11228 BPF_CALL_4(sk_reuseport_load_bytes,
11229 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11230 void *, to, u32, len)
11231 {
11232 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11233 }
11234
11235 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11236 .func = sk_reuseport_load_bytes,
11237 .gpl_only = false,
11238 .ret_type = RET_INTEGER,
11239 .arg1_type = ARG_PTR_TO_CTX,
11240 .arg2_type = ARG_ANYTHING,
11241 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11242 .arg4_type = ARG_CONST_SIZE,
11243 };
11244
BPF_CALL_5(sk_reuseport_load_bytes_relative,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len,u32,start_header)11245 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11246 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11247 void *, to, u32, len, u32, start_header)
11248 {
11249 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11250 len, start_header);
11251 }
11252
11253 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11254 .func = sk_reuseport_load_bytes_relative,
11255 .gpl_only = false,
11256 .ret_type = RET_INTEGER,
11257 .arg1_type = ARG_PTR_TO_CTX,
11258 .arg2_type = ARG_ANYTHING,
11259 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11260 .arg4_type = ARG_CONST_SIZE,
11261 .arg5_type = ARG_ANYTHING,
11262 };
11263
11264 static const struct bpf_func_proto *
sk_reuseport_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11265 sk_reuseport_func_proto(enum bpf_func_id func_id,
11266 const struct bpf_prog *prog)
11267 {
11268 switch (func_id) {
11269 case BPF_FUNC_sk_select_reuseport:
11270 return &sk_select_reuseport_proto;
11271 case BPF_FUNC_skb_load_bytes:
11272 return &sk_reuseport_load_bytes_proto;
11273 case BPF_FUNC_skb_load_bytes_relative:
11274 return &sk_reuseport_load_bytes_relative_proto;
11275 case BPF_FUNC_get_socket_cookie:
11276 return &bpf_get_socket_ptr_cookie_proto;
11277 case BPF_FUNC_ktime_get_coarse_ns:
11278 return &bpf_ktime_get_coarse_ns_proto;
11279 default:
11280 return bpf_base_func_proto(func_id);
11281 }
11282 }
11283
11284 static bool
sk_reuseport_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11285 sk_reuseport_is_valid_access(int off, int size,
11286 enum bpf_access_type type,
11287 const struct bpf_prog *prog,
11288 struct bpf_insn_access_aux *info)
11289 {
11290 const u32 size_default = sizeof(__u32);
11291
11292 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11293 off % size || type != BPF_READ)
11294 return false;
11295
11296 switch (off) {
11297 case offsetof(struct sk_reuseport_md, data):
11298 info->reg_type = PTR_TO_PACKET;
11299 return size == sizeof(__u64);
11300
11301 case offsetof(struct sk_reuseport_md, data_end):
11302 info->reg_type = PTR_TO_PACKET_END;
11303 return size == sizeof(__u64);
11304
11305 case offsetof(struct sk_reuseport_md, hash):
11306 return size == size_default;
11307
11308 case offsetof(struct sk_reuseport_md, sk):
11309 info->reg_type = PTR_TO_SOCKET;
11310 return size == sizeof(__u64);
11311
11312 case offsetof(struct sk_reuseport_md, migrating_sk):
11313 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11314 return size == sizeof(__u64);
11315
11316 /* Fields that allow narrowing */
11317 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11318 if (size < sizeof_field(struct sk_buff, protocol))
11319 return false;
11320 fallthrough;
11321 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11322 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11323 case bpf_ctx_range(struct sk_reuseport_md, len):
11324 bpf_ctx_record_field_size(info, size_default);
11325 return bpf_ctx_narrow_access_ok(off, size, size_default);
11326
11327 default:
11328 return false;
11329 }
11330 }
11331
11332 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11333 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11334 si->dst_reg, si->src_reg, \
11335 bpf_target_off(struct sk_reuseport_kern, F, \
11336 sizeof_field(struct sk_reuseport_kern, F), \
11337 target_size)); \
11338 })
11339
11340 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11341 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11342 struct sk_buff, \
11343 skb, \
11344 SKB_FIELD)
11345
11346 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11347 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11348 struct sock, \
11349 sk, \
11350 SK_FIELD)
11351
sk_reuseport_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11352 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11353 const struct bpf_insn *si,
11354 struct bpf_insn *insn_buf,
11355 struct bpf_prog *prog,
11356 u32 *target_size)
11357 {
11358 struct bpf_insn *insn = insn_buf;
11359
11360 switch (si->off) {
11361 case offsetof(struct sk_reuseport_md, data):
11362 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11363 break;
11364
11365 case offsetof(struct sk_reuseport_md, len):
11366 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11367 break;
11368
11369 case offsetof(struct sk_reuseport_md, eth_protocol):
11370 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11371 break;
11372
11373 case offsetof(struct sk_reuseport_md, ip_protocol):
11374 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11375 break;
11376
11377 case offsetof(struct sk_reuseport_md, data_end):
11378 SK_REUSEPORT_LOAD_FIELD(data_end);
11379 break;
11380
11381 case offsetof(struct sk_reuseport_md, hash):
11382 SK_REUSEPORT_LOAD_FIELD(hash);
11383 break;
11384
11385 case offsetof(struct sk_reuseport_md, bind_inany):
11386 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11387 break;
11388
11389 case offsetof(struct sk_reuseport_md, sk):
11390 SK_REUSEPORT_LOAD_FIELD(sk);
11391 break;
11392
11393 case offsetof(struct sk_reuseport_md, migrating_sk):
11394 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11395 break;
11396 }
11397
11398 return insn - insn_buf;
11399 }
11400
11401 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11402 .get_func_proto = sk_reuseport_func_proto,
11403 .is_valid_access = sk_reuseport_is_valid_access,
11404 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11405 };
11406
11407 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11408 };
11409
11410 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11411 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11412
BPF_CALL_3(bpf_sk_lookup_assign,struct bpf_sk_lookup_kern *,ctx,struct sock *,sk,u64,flags)11413 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11414 struct sock *, sk, u64, flags)
11415 {
11416 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11417 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11418 return -EINVAL;
11419 if (unlikely(sk && sk_is_refcounted(sk)))
11420 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11421 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11422 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11423 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11424 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11425
11426 /* Check if socket is suitable for packet L3/L4 protocol */
11427 if (sk && sk->sk_protocol != ctx->protocol)
11428 return -EPROTOTYPE;
11429 if (sk && sk->sk_family != ctx->family &&
11430 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11431 return -EAFNOSUPPORT;
11432
11433 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11434 return -EEXIST;
11435
11436 /* Select socket as lookup result */
11437 ctx->selected_sk = sk;
11438 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11439 return 0;
11440 }
11441
11442 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11443 .func = bpf_sk_lookup_assign,
11444 .gpl_only = false,
11445 .ret_type = RET_INTEGER,
11446 .arg1_type = ARG_PTR_TO_CTX,
11447 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11448 .arg3_type = ARG_ANYTHING,
11449 };
11450
11451 static const struct bpf_func_proto *
sk_lookup_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11452 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11453 {
11454 switch (func_id) {
11455 case BPF_FUNC_perf_event_output:
11456 return &bpf_event_output_data_proto;
11457 case BPF_FUNC_sk_assign:
11458 return &bpf_sk_lookup_assign_proto;
11459 case BPF_FUNC_sk_release:
11460 return &bpf_sk_release_proto;
11461 default:
11462 return bpf_sk_base_func_proto(func_id);
11463 }
11464 }
11465
sk_lookup_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11466 static bool sk_lookup_is_valid_access(int off, int size,
11467 enum bpf_access_type type,
11468 const struct bpf_prog *prog,
11469 struct bpf_insn_access_aux *info)
11470 {
11471 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11472 return false;
11473 if (off % size != 0)
11474 return false;
11475 if (type != BPF_READ)
11476 return false;
11477
11478 switch (off) {
11479 case offsetof(struct bpf_sk_lookup, sk):
11480 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11481 return size == sizeof(__u64);
11482
11483 case bpf_ctx_range(struct bpf_sk_lookup, family):
11484 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11485 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11486 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11487 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11488 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11489 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11490 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11491 bpf_ctx_record_field_size(info, sizeof(__u32));
11492 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11493
11494 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11495 /* Allow 4-byte access to 2-byte field for backward compatibility */
11496 if (size == sizeof(__u32))
11497 return true;
11498 bpf_ctx_record_field_size(info, sizeof(__be16));
11499 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11500
11501 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11502 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11503 /* Allow access to zero padding for backward compatibility */
11504 bpf_ctx_record_field_size(info, sizeof(__u16));
11505 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11506
11507 default:
11508 return false;
11509 }
11510 }
11511
sk_lookup_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11512 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11513 const struct bpf_insn *si,
11514 struct bpf_insn *insn_buf,
11515 struct bpf_prog *prog,
11516 u32 *target_size)
11517 {
11518 struct bpf_insn *insn = insn_buf;
11519
11520 switch (si->off) {
11521 case offsetof(struct bpf_sk_lookup, sk):
11522 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11523 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11524 break;
11525
11526 case offsetof(struct bpf_sk_lookup, family):
11527 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11528 bpf_target_off(struct bpf_sk_lookup_kern,
11529 family, 2, target_size));
11530 break;
11531
11532 case offsetof(struct bpf_sk_lookup, protocol):
11533 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11534 bpf_target_off(struct bpf_sk_lookup_kern,
11535 protocol, 2, target_size));
11536 break;
11537
11538 case offsetof(struct bpf_sk_lookup, remote_ip4):
11539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11540 bpf_target_off(struct bpf_sk_lookup_kern,
11541 v4.saddr, 4, target_size));
11542 break;
11543
11544 case offsetof(struct bpf_sk_lookup, local_ip4):
11545 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11546 bpf_target_off(struct bpf_sk_lookup_kern,
11547 v4.daddr, 4, target_size));
11548 break;
11549
11550 case bpf_ctx_range_till(struct bpf_sk_lookup,
11551 remote_ip6[0], remote_ip6[3]): {
11552 #if IS_ENABLED(CONFIG_IPV6)
11553 int off = si->off;
11554
11555 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11556 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11557 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11558 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11559 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11560 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11561 #else
11562 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11563 #endif
11564 break;
11565 }
11566 case bpf_ctx_range_till(struct bpf_sk_lookup,
11567 local_ip6[0], local_ip6[3]): {
11568 #if IS_ENABLED(CONFIG_IPV6)
11569 int off = si->off;
11570
11571 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11572 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11573 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11574 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11575 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11576 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11577 #else
11578 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11579 #endif
11580 break;
11581 }
11582 case offsetof(struct bpf_sk_lookup, remote_port):
11583 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11584 bpf_target_off(struct bpf_sk_lookup_kern,
11585 sport, 2, target_size));
11586 break;
11587
11588 case offsetofend(struct bpf_sk_lookup, remote_port):
11589 *target_size = 2;
11590 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11591 break;
11592
11593 case offsetof(struct bpf_sk_lookup, local_port):
11594 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11595 bpf_target_off(struct bpf_sk_lookup_kern,
11596 dport, 2, target_size));
11597 break;
11598
11599 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11600 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11601 bpf_target_off(struct bpf_sk_lookup_kern,
11602 ingress_ifindex, 4, target_size));
11603 break;
11604 }
11605
11606 return insn - insn_buf;
11607 }
11608
11609 const struct bpf_prog_ops sk_lookup_prog_ops = {
11610 .test_run = bpf_prog_test_run_sk_lookup,
11611 };
11612
11613 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11614 .get_func_proto = sk_lookup_func_proto,
11615 .is_valid_access = sk_lookup_is_valid_access,
11616 .convert_ctx_access = sk_lookup_convert_ctx_access,
11617 };
11618
11619 #endif /* CONFIG_INET */
11620
DEFINE_BPF_DISPATCHER(xdp)11621 DEFINE_BPF_DISPATCHER(xdp)
11622
11623 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11624 {
11625 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11626 }
11627
BTF_ID_LIST_GLOBAL(btf_sock_ids,MAX_BTF_SOCK_TYPE)11628 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11629 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11630 BTF_SOCK_TYPE_xxx
11631 #undef BTF_SOCK_TYPE
11632
11633 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11634 {
11635 /* tcp6_sock type is not generated in dwarf and hence btf,
11636 * trigger an explicit type generation here.
11637 */
11638 BTF_TYPE_EMIT(struct tcp6_sock);
11639 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11640 sk->sk_family == AF_INET6)
11641 return (unsigned long)sk;
11642
11643 return (unsigned long)NULL;
11644 }
11645
11646 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11647 .func = bpf_skc_to_tcp6_sock,
11648 .gpl_only = false,
11649 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11650 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11651 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11652 };
11653
BPF_CALL_1(bpf_skc_to_tcp_sock,struct sock *,sk)11654 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11655 {
11656 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11657 return (unsigned long)sk;
11658
11659 return (unsigned long)NULL;
11660 }
11661
11662 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11663 .func = bpf_skc_to_tcp_sock,
11664 .gpl_only = false,
11665 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11666 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11667 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11668 };
11669
BPF_CALL_1(bpf_skc_to_tcp_timewait_sock,struct sock *,sk)11670 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11671 {
11672 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11673 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11674 */
11675 BTF_TYPE_EMIT(struct inet_timewait_sock);
11676 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11677
11678 #ifdef CONFIG_INET
11679 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11680 return (unsigned long)sk;
11681 #endif
11682
11683 #if IS_BUILTIN(CONFIG_IPV6)
11684 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11685 return (unsigned long)sk;
11686 #endif
11687
11688 return (unsigned long)NULL;
11689 }
11690
11691 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11692 .func = bpf_skc_to_tcp_timewait_sock,
11693 .gpl_only = false,
11694 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11695 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11696 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11697 };
11698
BPF_CALL_1(bpf_skc_to_tcp_request_sock,struct sock *,sk)11699 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11700 {
11701 #ifdef CONFIG_INET
11702 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11703 return (unsigned long)sk;
11704 #endif
11705
11706 #if IS_BUILTIN(CONFIG_IPV6)
11707 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11708 return (unsigned long)sk;
11709 #endif
11710
11711 return (unsigned long)NULL;
11712 }
11713
11714 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11715 .func = bpf_skc_to_tcp_request_sock,
11716 .gpl_only = false,
11717 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11718 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11719 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11720 };
11721
BPF_CALL_1(bpf_skc_to_udp6_sock,struct sock *,sk)11722 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11723 {
11724 /* udp6_sock type is not generated in dwarf and hence btf,
11725 * trigger an explicit type generation here.
11726 */
11727 BTF_TYPE_EMIT(struct udp6_sock);
11728 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11729 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11730 return (unsigned long)sk;
11731
11732 return (unsigned long)NULL;
11733 }
11734
11735 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11736 .func = bpf_skc_to_udp6_sock,
11737 .gpl_only = false,
11738 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11739 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11740 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11741 };
11742
BPF_CALL_1(bpf_skc_to_unix_sock,struct sock *,sk)11743 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11744 {
11745 /* unix_sock type is not generated in dwarf and hence btf,
11746 * trigger an explicit type generation here.
11747 */
11748 BTF_TYPE_EMIT(struct unix_sock);
11749 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11750 return (unsigned long)sk;
11751
11752 return (unsigned long)NULL;
11753 }
11754
11755 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11756 .func = bpf_skc_to_unix_sock,
11757 .gpl_only = false,
11758 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11759 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11760 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11761 };
11762
BPF_CALL_1(bpf_skc_to_mptcp_sock,struct sock *,sk)11763 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11764 {
11765 BTF_TYPE_EMIT(struct mptcp_sock);
11766 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11767 }
11768
11769 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11770 .func = bpf_skc_to_mptcp_sock,
11771 .gpl_only = false,
11772 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11773 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11774 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11775 };
11776
BPF_CALL_1(bpf_sock_from_file,struct file *,file)11777 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11778 {
11779 return (unsigned long)sock_from_file(file);
11780 }
11781
11782 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11783 BTF_ID(struct, socket)
11784 BTF_ID(struct, file)
11785
11786 const struct bpf_func_proto bpf_sock_from_file_proto = {
11787 .func = bpf_sock_from_file,
11788 .gpl_only = false,
11789 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11790 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11791 .arg1_type = ARG_PTR_TO_BTF_ID,
11792 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11793 };
11794
11795 static const struct bpf_func_proto *
bpf_sk_base_func_proto(enum bpf_func_id func_id)11796 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11797 {
11798 const struct bpf_func_proto *func;
11799
11800 switch (func_id) {
11801 case BPF_FUNC_skc_to_tcp6_sock:
11802 func = &bpf_skc_to_tcp6_sock_proto;
11803 break;
11804 case BPF_FUNC_skc_to_tcp_sock:
11805 func = &bpf_skc_to_tcp_sock_proto;
11806 break;
11807 case BPF_FUNC_skc_to_tcp_timewait_sock:
11808 func = &bpf_skc_to_tcp_timewait_sock_proto;
11809 break;
11810 case BPF_FUNC_skc_to_tcp_request_sock:
11811 func = &bpf_skc_to_tcp_request_sock_proto;
11812 break;
11813 case BPF_FUNC_skc_to_udp6_sock:
11814 func = &bpf_skc_to_udp6_sock_proto;
11815 break;
11816 case BPF_FUNC_skc_to_unix_sock:
11817 func = &bpf_skc_to_unix_sock_proto;
11818 break;
11819 case BPF_FUNC_skc_to_mptcp_sock:
11820 func = &bpf_skc_to_mptcp_sock_proto;
11821 break;
11822 case BPF_FUNC_ktime_get_coarse_ns:
11823 return &bpf_ktime_get_coarse_ns_proto;
11824 default:
11825 return bpf_base_func_proto(func_id);
11826 }
11827
11828 if (!perfmon_capable())
11829 return NULL;
11830
11831 return func;
11832 }
11833
11834 __diag_push();
11835 __diag_ignore_all("-Wmissing-prototypes",
11836 "Global functions as their definitions will be in vmlinux BTF");
bpf_dynptr_from_skb(struct sk_buff * skb,u64 flags,struct bpf_dynptr_kern * ptr__uninit)11837 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11838 struct bpf_dynptr_kern *ptr__uninit)
11839 {
11840 if (flags) {
11841 bpf_dynptr_set_null(ptr__uninit);
11842 return -EINVAL;
11843 }
11844
11845 bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11846
11847 return 0;
11848 }
11849
bpf_dynptr_from_xdp(struct xdp_buff * xdp,u64 flags,struct bpf_dynptr_kern * ptr__uninit)11850 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11851 struct bpf_dynptr_kern *ptr__uninit)
11852 {
11853 if (flags) {
11854 bpf_dynptr_set_null(ptr__uninit);
11855 return -EINVAL;
11856 }
11857
11858 bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11859
11860 return 0;
11861 }
11862
bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern * sa_kern,const u8 * sun_path,u32 sun_path__sz)11863 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
11864 const u8 *sun_path, u32 sun_path__sz)
11865 {
11866 struct sockaddr_un *un;
11867
11868 if (sa_kern->sk->sk_family != AF_UNIX)
11869 return -EINVAL;
11870
11871 /* We do not allow changing the address to unnamed or larger than the
11872 * maximum allowed address size for a unix sockaddr.
11873 */
11874 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX)
11875 return -EINVAL;
11876
11877 un = (struct sockaddr_un *)sa_kern->uaddr;
11878 memcpy(un->sun_path, sun_path, sun_path__sz);
11879 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz;
11880
11881 return 0;
11882 }
11883 __diag_pop();
11884
bpf_dynptr_from_skb_rdonly(struct sk_buff * skb,u64 flags,struct bpf_dynptr_kern * ptr__uninit)11885 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11886 struct bpf_dynptr_kern *ptr__uninit)
11887 {
11888 int err;
11889
11890 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11891 if (err)
11892 return err;
11893
11894 bpf_dynptr_set_rdonly(ptr__uninit);
11895
11896 return 0;
11897 }
11898
11899 BTF_SET8_START(bpf_kfunc_check_set_skb)
11900 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11901 BTF_SET8_END(bpf_kfunc_check_set_skb)
11902
11903 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11904 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11905 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11906
11907 BTF_SET8_START(bpf_kfunc_check_set_sock_addr)
11908 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
11909 BTF_SET8_END(bpf_kfunc_check_set_sock_addr)
11910
11911 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11912 .owner = THIS_MODULE,
11913 .set = &bpf_kfunc_check_set_skb,
11914 };
11915
11916 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11917 .owner = THIS_MODULE,
11918 .set = &bpf_kfunc_check_set_xdp,
11919 };
11920
11921 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
11922 .owner = THIS_MODULE,
11923 .set = &bpf_kfunc_check_set_sock_addr,
11924 };
11925
bpf_kfunc_init(void)11926 static int __init bpf_kfunc_init(void)
11927 {
11928 int ret;
11929
11930 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11931 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11932 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11933 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11934 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11935 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11936 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11937 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11938 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11939 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11940 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11941 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
11942 &bpf_kfunc_set_sock_addr);
11943 }
11944 late_initcall(bpf_kfunc_init);
11945
11946 /* Disables missing prototype warnings */
11947 __diag_push();
11948 __diag_ignore_all("-Wmissing-prototypes",
11949 "Global functions as their definitions will be in vmlinux BTF");
11950
11951 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11952 *
11953 * The function expects a non-NULL pointer to a socket, and invokes the
11954 * protocol specific socket destroy handlers.
11955 *
11956 * The helper can only be called from BPF contexts that have acquired the socket
11957 * locks.
11958 *
11959 * Parameters:
11960 * @sock: Pointer to socket to be destroyed
11961 *
11962 * Return:
11963 * On error, may return EPROTONOSUPPORT, EINVAL.
11964 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11965 * 0 otherwise
11966 */
bpf_sock_destroy(struct sock_common * sock)11967 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11968 {
11969 struct sock *sk = (struct sock *)sock;
11970
11971 /* The locking semantics that allow for synchronous execution of the
11972 * destroy handlers are only supported for TCP and UDP.
11973 * Supporting protocols will need to acquire sock lock in the BPF context
11974 * prior to invoking this kfunc.
11975 */
11976 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11977 sk->sk_protocol != IPPROTO_UDP))
11978 return -EOPNOTSUPP;
11979
11980 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11981 }
11982
11983 __diag_pop()
11984
BTF_SET8_START(bpf_sk_iter_kfunc_ids)11985 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11986 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11987 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11988
11989 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11990 {
11991 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11992 prog->expected_attach_type != BPF_TRACE_ITER)
11993 return -EACCES;
11994 return 0;
11995 }
11996
11997 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11998 .owner = THIS_MODULE,
11999 .set = &bpf_sk_iter_kfunc_ids,
12000 .filter = tracing_iter_filter,
12001 };
12002
init_subsystem(void)12003 static int init_subsystem(void)
12004 {
12005 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
12006 }
12007 late_initcall(init_subsystem);
12008