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 <linux/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <net/xfrm.h>
60 #include <net/udp.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
68 #include <net/flow.h>
69 #include <net/arp.h>
70 #include <net/ipv6.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
73 #include <net/seg6.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
80 #include <net/tls.h>
81 #include <net/xdp.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
84 #include <net/netkit.h>
85 #include <linux/un.h>
86 #include <net/xdp_sock_drv.h>
87 #include <net/inet_dscp.h>
88
89 #include "dev.h"
90
91 /* Keep the struct bpf_fib_lookup small so that it fits into a cacheline */
92 static_assert(sizeof(struct bpf_fib_lookup) == 64, "struct bpf_fib_lookup size check");
93
94 static const struct bpf_func_proto *
95 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
96
copy_bpf_fprog_from_user(struct sock_fprog * dst,sockptr_t src,int len)97 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
98 {
99 if (in_compat_syscall()) {
100 struct compat_sock_fprog f32;
101
102 if (len != sizeof(f32))
103 return -EINVAL;
104 if (copy_from_sockptr(&f32, src, sizeof(f32)))
105 return -EFAULT;
106 memset(dst, 0, sizeof(*dst));
107 dst->len = f32.len;
108 dst->filter = compat_ptr(f32.filter);
109 } else {
110 if (len != sizeof(*dst))
111 return -EINVAL;
112 if (copy_from_sockptr(dst, src, sizeof(*dst)))
113 return -EFAULT;
114 }
115
116 return 0;
117 }
118 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
119
120 /**
121 * sk_filter_trim_cap - run a packet through a socket filter
122 * @sk: sock associated with &sk_buff
123 * @skb: buffer to filter
124 * @cap: limit on how short the eBPF program may trim the packet
125 *
126 * Run the eBPF program and then cut skb->data to correct size returned by
127 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
128 * than pkt_len we keep whole skb->data. This is the socket level
129 * wrapper to bpf_prog_run. It returns 0 if the packet should
130 * be accepted or -EPERM if the packet should be tossed.
131 *
132 */
sk_filter_trim_cap(struct sock * sk,struct sk_buff * skb,unsigned int cap)133 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
134 {
135 int err;
136 struct sk_filter *filter;
137
138 /*
139 * If the skb was allocated from pfmemalloc reserves, only
140 * allow SOCK_MEMALLOC sockets to use it as this socket is
141 * helping free memory
142 */
143 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
144 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
145 return -ENOMEM;
146 }
147 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
148 if (err)
149 return err;
150
151 err = security_sock_rcv_skb(sk, skb);
152 if (err)
153 return err;
154
155 rcu_read_lock();
156 filter = rcu_dereference(sk->sk_filter);
157 if (filter) {
158 struct sock *save_sk = skb->sk;
159 unsigned int pkt_len;
160
161 skb->sk = sk;
162 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
163 skb->sk = save_sk;
164 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
165 }
166 rcu_read_unlock();
167
168 return err;
169 }
170 EXPORT_SYMBOL(sk_filter_trim_cap);
171
BPF_CALL_1(bpf_skb_get_pay_offset,struct sk_buff *,skb)172 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
173 {
174 return skb_get_poff(skb);
175 }
176
BPF_CALL_3(bpf_skb_get_nlattr,struct sk_buff *,skb,u32,a,u32,x)177 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
178 {
179 struct nlattr *nla;
180
181 if (skb_is_nonlinear(skb))
182 return 0;
183
184 if (skb->len < sizeof(struct nlattr))
185 return 0;
186
187 if (a > skb->len - sizeof(struct nlattr))
188 return 0;
189
190 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
191 if (nla)
192 return (void *) nla - (void *) skb->data;
193
194 return 0;
195 }
196
BPF_CALL_3(bpf_skb_get_nlattr_nest,struct sk_buff *,skb,u32,a,u32,x)197 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
198 {
199 struct nlattr *nla;
200
201 if (skb_is_nonlinear(skb))
202 return 0;
203
204 if (skb->len < sizeof(struct nlattr))
205 return 0;
206
207 if (a > skb->len - sizeof(struct nlattr))
208 return 0;
209
210 nla = (struct nlattr *) &skb->data[a];
211 if (!nla_ok(nla, skb->len - a))
212 return 0;
213
214 nla = nla_find_nested(nla, x);
215 if (nla)
216 return (void *) nla - (void *) skb->data;
217
218 return 0;
219 }
220
BPF_CALL_4(bpf_skb_load_helper_8,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)221 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
222 data, int, headlen, int, offset)
223 {
224 u8 tmp, *ptr;
225 const int len = sizeof(tmp);
226
227 if (offset >= 0) {
228 if (headlen - offset >= len)
229 return *(u8 *)(data + offset);
230 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
231 return tmp;
232 } else {
233 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
234 if (likely(ptr))
235 return *(u8 *)ptr;
236 }
237
238 return -EFAULT;
239 }
240
BPF_CALL_2(bpf_skb_load_helper_8_no_cache,const struct sk_buff *,skb,int,offset)241 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
242 int, offset)
243 {
244 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
245 offset);
246 }
247
BPF_CALL_4(bpf_skb_load_helper_16,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)248 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
249 data, int, headlen, int, offset)
250 {
251 __be16 tmp, *ptr;
252 const int len = sizeof(tmp);
253
254 if (offset >= 0) {
255 if (headlen - offset >= len)
256 return get_unaligned_be16(data + offset);
257 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
258 return be16_to_cpu(tmp);
259 } else {
260 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
261 if (likely(ptr))
262 return get_unaligned_be16(ptr);
263 }
264
265 return -EFAULT;
266 }
267
BPF_CALL_2(bpf_skb_load_helper_16_no_cache,const struct sk_buff *,skb,int,offset)268 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
269 int, offset)
270 {
271 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
272 offset);
273 }
274
BPF_CALL_4(bpf_skb_load_helper_32,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)275 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
276 data, int, headlen, int, offset)
277 {
278 __be32 tmp, *ptr;
279 const int len = sizeof(tmp);
280
281 if (likely(offset >= 0)) {
282 if (headlen - offset >= len)
283 return get_unaligned_be32(data + offset);
284 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
285 return be32_to_cpu(tmp);
286 } else {
287 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
288 if (likely(ptr))
289 return get_unaligned_be32(ptr);
290 }
291
292 return -EFAULT;
293 }
294
BPF_CALL_2(bpf_skb_load_helper_32_no_cache,const struct sk_buff *,skb,int,offset)295 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
296 int, offset)
297 {
298 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
299 offset);
300 }
301
convert_skb_access(int skb_field,int dst_reg,int src_reg,struct bpf_insn * insn_buf)302 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
303 struct bpf_insn *insn_buf)
304 {
305 struct bpf_insn *insn = insn_buf;
306
307 switch (skb_field) {
308 case SKF_AD_MARK:
309 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
310
311 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
312 offsetof(struct sk_buff, mark));
313 break;
314
315 case SKF_AD_PKTTYPE:
316 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
317 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
318 #ifdef __BIG_ENDIAN_BITFIELD
319 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
320 #endif
321 break;
322
323 case SKF_AD_QUEUE:
324 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
325
326 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
327 offsetof(struct sk_buff, queue_mapping));
328 break;
329
330 case SKF_AD_VLAN_TAG:
331 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
332
333 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
334 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
335 offsetof(struct sk_buff, vlan_tci));
336 break;
337 case SKF_AD_VLAN_TAG_PRESENT:
338 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
339 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
340 offsetof(struct sk_buff, vlan_all));
341 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
342 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
343 break;
344 }
345
346 return insn - insn_buf;
347 }
348
convert_bpf_extensions(struct sock_filter * fp,struct bpf_insn ** insnp)349 static bool convert_bpf_extensions(struct sock_filter *fp,
350 struct bpf_insn **insnp)
351 {
352 struct bpf_insn *insn = *insnp;
353 u32 cnt;
354
355 switch (fp->k) {
356 case SKF_AD_OFF + SKF_AD_PROTOCOL:
357 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
358
359 /* A = *(u16 *) (CTX + offsetof(protocol)) */
360 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
361 offsetof(struct sk_buff, protocol));
362 /* A = ntohs(A) [emitting a nop or swap16] */
363 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
364 break;
365
366 case SKF_AD_OFF + SKF_AD_PKTTYPE:
367 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
368 insn += cnt - 1;
369 break;
370
371 case SKF_AD_OFF + SKF_AD_IFINDEX:
372 case SKF_AD_OFF + SKF_AD_HATYPE:
373 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
374 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
375
376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
377 BPF_REG_TMP, BPF_REG_CTX,
378 offsetof(struct sk_buff, dev));
379 /* if (tmp != 0) goto pc + 1 */
380 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
381 *insn++ = BPF_EXIT_INSN();
382 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
383 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
384 offsetof(struct net_device, ifindex));
385 else
386 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
387 offsetof(struct net_device, type));
388 break;
389
390 case SKF_AD_OFF + SKF_AD_MARK:
391 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
392 insn += cnt - 1;
393 break;
394
395 case SKF_AD_OFF + SKF_AD_RXHASH:
396 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
397
398 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
399 offsetof(struct sk_buff, hash));
400 break;
401
402 case SKF_AD_OFF + SKF_AD_QUEUE:
403 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
404 insn += cnt - 1;
405 break;
406
407 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
408 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
409 BPF_REG_A, BPF_REG_CTX, insn);
410 insn += cnt - 1;
411 break;
412
413 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
414 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
415 BPF_REG_A, BPF_REG_CTX, insn);
416 insn += cnt - 1;
417 break;
418
419 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
420 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
421
422 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
423 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
424 offsetof(struct sk_buff, vlan_proto));
425 /* A = ntohs(A) [emitting a nop or swap16] */
426 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
427 break;
428
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 case SKF_AD_OFF + SKF_AD_NLATTR:
431 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
432 case SKF_AD_OFF + SKF_AD_CPU:
433 case SKF_AD_OFF + SKF_AD_RANDOM:
434 /* arg1 = CTX */
435 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
436 /* arg2 = A */
437 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
438 /* arg3 = X */
439 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
440 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
441 switch (fp->k) {
442 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
443 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
444 break;
445 case SKF_AD_OFF + SKF_AD_NLATTR:
446 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
447 break;
448 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
449 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
450 break;
451 case SKF_AD_OFF + SKF_AD_CPU:
452 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
453 break;
454 case SKF_AD_OFF + SKF_AD_RANDOM:
455 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
456 bpf_user_rnd_init_once();
457 break;
458 }
459 break;
460
461 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
462 /* A ^= X */
463 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
464 break;
465
466 default:
467 /* This is just a dummy call to avoid letting the compiler
468 * evict __bpf_call_base() as an optimization. Placed here
469 * where no-one bothers.
470 */
471 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
472 return false;
473 }
474
475 *insnp = insn;
476 return true;
477 }
478
convert_bpf_ld_abs(struct sock_filter * fp,struct bpf_insn ** insnp)479 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
480 {
481 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
482 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
483 bool endian = BPF_SIZE(fp->code) == BPF_H ||
484 BPF_SIZE(fp->code) == BPF_W;
485 bool indirect = BPF_MODE(fp->code) == BPF_IND;
486 const int ip_align = NET_IP_ALIGN;
487 struct bpf_insn *insn = *insnp;
488 int offset = fp->k;
489
490 if (!indirect &&
491 ((unaligned_ok && offset >= 0) ||
492 (!unaligned_ok && offset >= 0 &&
493 offset + ip_align >= 0 &&
494 offset + ip_align % size == 0))) {
495 bool ldx_off_ok = offset <= S16_MAX;
496
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
498 if (offset)
499 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
500 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
501 size, 2 + endian + (!ldx_off_ok * 2));
502 if (ldx_off_ok) {
503 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
504 BPF_REG_D, offset);
505 } else {
506 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
507 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
508 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
509 BPF_REG_TMP, 0);
510 }
511 if (endian)
512 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
513 *insn++ = BPF_JMP_A(8);
514 }
515
516 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
517 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
518 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
519 if (!indirect) {
520 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
521 } else {
522 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
523 if (fp->k)
524 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
525 }
526
527 switch (BPF_SIZE(fp->code)) {
528 case BPF_B:
529 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
530 break;
531 case BPF_H:
532 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
533 break;
534 case BPF_W:
535 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
536 break;
537 default:
538 return false;
539 }
540
541 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
542 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
543 *insn = BPF_EXIT_INSN();
544
545 *insnp = insn;
546 return true;
547 }
548
549 /**
550 * bpf_convert_filter - convert filter program
551 * @prog: the user passed filter program
552 * @len: the length of the user passed filter program
553 * @new_prog: allocated 'struct bpf_prog' or NULL
554 * @new_len: pointer to store length of converted program
555 * @seen_ld_abs: bool whether we've seen ld_abs/ind
556 *
557 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
558 * style extended BPF (eBPF).
559 * Conversion workflow:
560 *
561 * 1) First pass for calculating the new program length:
562 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
563 *
564 * 2) 2nd pass to remap in two passes: 1st pass finds new
565 * jump offsets, 2nd pass remapping:
566 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
567 */
bpf_convert_filter(struct sock_filter * prog,int len,struct bpf_prog * new_prog,int * new_len,bool * seen_ld_abs)568 static int bpf_convert_filter(struct sock_filter *prog, int len,
569 struct bpf_prog *new_prog, int *new_len,
570 bool *seen_ld_abs)
571 {
572 int new_flen = 0, pass = 0, target, i, stack_off;
573 struct bpf_insn *new_insn, *first_insn = NULL;
574 struct sock_filter *fp;
575 int *addrs = NULL;
576 u8 bpf_src;
577
578 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
579 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
580
581 if (len <= 0 || len > BPF_MAXINSNS)
582 return -EINVAL;
583
584 if (new_prog) {
585 first_insn = new_prog->insnsi;
586 addrs = kcalloc(len, sizeof(*addrs),
587 GFP_KERNEL | __GFP_NOWARN);
588 if (!addrs)
589 return -ENOMEM;
590 }
591
592 do_pass:
593 new_insn = first_insn;
594 fp = prog;
595
596 /* Classic BPF related prologue emission. */
597 if (new_prog) {
598 /* Classic BPF expects A and X to be reset first. These need
599 * to be guaranteed to be the first two instructions.
600 */
601 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
602 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
603
604 /* All programs must keep CTX in callee saved BPF_REG_CTX.
605 * In eBPF case it's done by the compiler, here we need to
606 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
607 */
608 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
609 if (*seen_ld_abs) {
610 /* For packet access in classic BPF, cache skb->data
611 * in callee-saved BPF R8 and skb->len - skb->data_len
612 * (headlen) in BPF R9. Since classic BPF is read-only
613 * on CTX, we only need to cache it once.
614 */
615 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
616 BPF_REG_D, BPF_REG_CTX,
617 offsetof(struct sk_buff, data));
618 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
619 offsetof(struct sk_buff, len));
620 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
621 offsetof(struct sk_buff, data_len));
622 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
623 }
624 } else {
625 new_insn += 3;
626 }
627
628 for (i = 0; i < len; fp++, i++) {
629 struct bpf_insn tmp_insns[32] = { };
630 struct bpf_insn *insn = tmp_insns;
631
632 if (addrs)
633 addrs[i] = new_insn - first_insn;
634
635 switch (fp->code) {
636 /* All arithmetic insns and skb loads map as-is. */
637 case BPF_ALU | BPF_ADD | BPF_X:
638 case BPF_ALU | BPF_ADD | BPF_K:
639 case BPF_ALU | BPF_SUB | BPF_X:
640 case BPF_ALU | BPF_SUB | BPF_K:
641 case BPF_ALU | BPF_AND | BPF_X:
642 case BPF_ALU | BPF_AND | BPF_K:
643 case BPF_ALU | BPF_OR | BPF_X:
644 case BPF_ALU | BPF_OR | BPF_K:
645 case BPF_ALU | BPF_LSH | BPF_X:
646 case BPF_ALU | BPF_LSH | BPF_K:
647 case BPF_ALU | BPF_RSH | BPF_X:
648 case BPF_ALU | BPF_RSH | BPF_K:
649 case BPF_ALU | BPF_XOR | BPF_X:
650 case BPF_ALU | BPF_XOR | BPF_K:
651 case BPF_ALU | BPF_MUL | BPF_X:
652 case BPF_ALU | BPF_MUL | BPF_K:
653 case BPF_ALU | BPF_DIV | BPF_X:
654 case BPF_ALU | BPF_DIV | BPF_K:
655 case BPF_ALU | BPF_MOD | BPF_X:
656 case BPF_ALU | BPF_MOD | BPF_K:
657 case BPF_ALU | BPF_NEG:
658 case BPF_LD | BPF_ABS | BPF_W:
659 case BPF_LD | BPF_ABS | BPF_H:
660 case BPF_LD | BPF_ABS | BPF_B:
661 case BPF_LD | BPF_IND | BPF_W:
662 case BPF_LD | BPF_IND | BPF_H:
663 case BPF_LD | BPF_IND | BPF_B:
664 /* Check for overloaded BPF extension and
665 * directly convert it if found, otherwise
666 * just move on with mapping.
667 */
668 if (BPF_CLASS(fp->code) == BPF_LD &&
669 BPF_MODE(fp->code) == BPF_ABS &&
670 convert_bpf_extensions(fp, &insn))
671 break;
672 if (BPF_CLASS(fp->code) == BPF_LD &&
673 convert_bpf_ld_abs(fp, &insn)) {
674 *seen_ld_abs = true;
675 break;
676 }
677
678 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
679 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
680 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
681 /* Error with exception code on div/mod by 0.
682 * For cBPF programs, this was always return 0.
683 */
684 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
685 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
686 *insn++ = BPF_EXIT_INSN();
687 }
688
689 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
690 break;
691
692 /* Jump transformation cannot use BPF block macros
693 * everywhere as offset calculation and target updates
694 * require a bit more work than the rest, i.e. jump
695 * opcodes map as-is, but offsets need adjustment.
696 */
697
698 #define BPF_EMIT_JMP \
699 do { \
700 const s32 off_min = S16_MIN, off_max = S16_MAX; \
701 s32 off; \
702 \
703 if (target >= len || target < 0) \
704 goto err; \
705 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
706 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
707 off -= insn - tmp_insns; \
708 /* Reject anything not fitting into insn->off. */ \
709 if (off < off_min || off > off_max) \
710 goto err; \
711 insn->off = off; \
712 } while (0)
713
714 case BPF_JMP | BPF_JA:
715 target = i + fp->k + 1;
716 insn->code = fp->code;
717 BPF_EMIT_JMP;
718 break;
719
720 case BPF_JMP | BPF_JEQ | BPF_K:
721 case BPF_JMP | BPF_JEQ | BPF_X:
722 case BPF_JMP | BPF_JSET | BPF_K:
723 case BPF_JMP | BPF_JSET | BPF_X:
724 case BPF_JMP | BPF_JGT | BPF_K:
725 case BPF_JMP | BPF_JGT | BPF_X:
726 case BPF_JMP | BPF_JGE | BPF_K:
727 case BPF_JMP | BPF_JGE | BPF_X:
728 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
729 /* BPF immediates are signed, zero extend
730 * immediate into tmp register and use it
731 * in compare insn.
732 */
733 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
734
735 insn->dst_reg = BPF_REG_A;
736 insn->src_reg = BPF_REG_TMP;
737 bpf_src = BPF_X;
738 } else {
739 insn->dst_reg = BPF_REG_A;
740 insn->imm = fp->k;
741 bpf_src = BPF_SRC(fp->code);
742 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
743 }
744
745 /* Common case where 'jump_false' is next insn. */
746 if (fp->jf == 0) {
747 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
748 target = i + fp->jt + 1;
749 BPF_EMIT_JMP;
750 break;
751 }
752
753 /* Convert some jumps when 'jump_true' is next insn. */
754 if (fp->jt == 0) {
755 switch (BPF_OP(fp->code)) {
756 case BPF_JEQ:
757 insn->code = BPF_JMP | BPF_JNE | bpf_src;
758 break;
759 case BPF_JGT:
760 insn->code = BPF_JMP | BPF_JLE | bpf_src;
761 break;
762 case BPF_JGE:
763 insn->code = BPF_JMP | BPF_JLT | bpf_src;
764 break;
765 default:
766 goto jmp_rest;
767 }
768
769 target = i + fp->jf + 1;
770 BPF_EMIT_JMP;
771 break;
772 }
773 jmp_rest:
774 /* Other jumps are mapped into two insns: Jxx and JA. */
775 target = i + fp->jt + 1;
776 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
777 BPF_EMIT_JMP;
778 insn++;
779
780 insn->code = BPF_JMP | BPF_JA;
781 target = i + fp->jf + 1;
782 BPF_EMIT_JMP;
783 break;
784
785 /* ldxb 4 * ([14] & 0xf) is remapped into 6 insns. */
786 case BPF_LDX | BPF_MSH | BPF_B: {
787 struct sock_filter tmp = {
788 .code = BPF_LD | BPF_ABS | BPF_B,
789 .k = fp->k,
790 };
791
792 *seen_ld_abs = true;
793
794 /* X = A */
795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
797 convert_bpf_ld_abs(&tmp, &insn);
798 insn++;
799 /* A &= 0xf */
800 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
801 /* A <<= 2 */
802 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
803 /* tmp = X */
804 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
805 /* X = A */
806 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
807 /* A = tmp */
808 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
809 break;
810 }
811 /* RET_K is remapped into 2 insns. RET_A case doesn't need an
812 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
813 */
814 case BPF_RET | BPF_A:
815 case BPF_RET | BPF_K:
816 if (BPF_RVAL(fp->code) == BPF_K)
817 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
818 0, fp->k);
819 *insn = BPF_EXIT_INSN();
820 break;
821
822 /* Store to stack. */
823 case BPF_ST:
824 case BPF_STX:
825 stack_off = fp->k * 4 + 4;
826 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
827 BPF_ST ? BPF_REG_A : BPF_REG_X,
828 -stack_off);
829 /* check_load_and_stores() verifies that classic BPF can
830 * load from stack only after write, so tracking
831 * stack_depth for ST|STX insns is enough
832 */
833 if (new_prog && new_prog->aux->stack_depth < stack_off)
834 new_prog->aux->stack_depth = stack_off;
835 break;
836
837 /* Load from stack. */
838 case BPF_LD | BPF_MEM:
839 case BPF_LDX | BPF_MEM:
840 stack_off = fp->k * 4 + 4;
841 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
842 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
843 -stack_off);
844 break;
845
846 /* A = K or X = K */
847 case BPF_LD | BPF_IMM:
848 case BPF_LDX | BPF_IMM:
849 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
850 BPF_REG_A : BPF_REG_X, fp->k);
851 break;
852
853 /* X = A */
854 case BPF_MISC | BPF_TAX:
855 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
856 break;
857
858 /* A = X */
859 case BPF_MISC | BPF_TXA:
860 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
861 break;
862
863 /* A = skb->len or X = skb->len */
864 case BPF_LD | BPF_W | BPF_LEN:
865 case BPF_LDX | BPF_W | BPF_LEN:
866 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
867 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
868 offsetof(struct sk_buff, len));
869 break;
870
871 /* Access seccomp_data fields. */
872 case BPF_LDX | BPF_ABS | BPF_W:
873 /* A = *(u32 *) (ctx + K) */
874 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
875 break;
876
877 /* Unknown instruction. */
878 default:
879 goto err;
880 }
881
882 insn++;
883 if (new_prog)
884 memcpy(new_insn, tmp_insns,
885 sizeof(*insn) * (insn - tmp_insns));
886 new_insn += insn - tmp_insns;
887 }
888
889 if (!new_prog) {
890 /* Only calculating new length. */
891 *new_len = new_insn - first_insn;
892 if (*seen_ld_abs)
893 *new_len += 4; /* Prologue bits. */
894 return 0;
895 }
896
897 pass++;
898 if (new_flen != new_insn - first_insn) {
899 new_flen = new_insn - first_insn;
900 if (pass > 2)
901 goto err;
902 goto do_pass;
903 }
904
905 kfree(addrs);
906 BUG_ON(*new_len != new_flen);
907 return 0;
908 err:
909 kfree(addrs);
910 return -EINVAL;
911 }
912
913 /* Security:
914 *
915 * As we dont want to clear mem[] array for each packet going through
916 * __bpf_prog_run(), we check that filter loaded by user never try to read
917 * a cell if not previously written, and we check all branches to be sure
918 * a malicious user doesn't try to abuse us.
919 */
check_load_and_stores(const struct sock_filter * filter,int flen)920 static int check_load_and_stores(const struct sock_filter *filter, int flen)
921 {
922 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
923 int pc, ret = 0;
924
925 BUILD_BUG_ON(BPF_MEMWORDS > 16);
926
927 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
928 if (!masks)
929 return -ENOMEM;
930
931 memset(masks, 0xff, flen * sizeof(*masks));
932
933 for (pc = 0; pc < flen; pc++) {
934 memvalid &= masks[pc];
935
936 switch (filter[pc].code) {
937 case BPF_ST:
938 case BPF_STX:
939 memvalid |= (1 << filter[pc].k);
940 break;
941 case BPF_LD | BPF_MEM:
942 case BPF_LDX | BPF_MEM:
943 if (!(memvalid & (1 << filter[pc].k))) {
944 ret = -EINVAL;
945 goto error;
946 }
947 break;
948 case BPF_JMP | BPF_JA:
949 /* A jump must set masks on target */
950 masks[pc + 1 + filter[pc].k] &= memvalid;
951 memvalid = ~0;
952 break;
953 case BPF_JMP | BPF_JEQ | BPF_K:
954 case BPF_JMP | BPF_JEQ | BPF_X:
955 case BPF_JMP | BPF_JGE | BPF_K:
956 case BPF_JMP | BPF_JGE | BPF_X:
957 case BPF_JMP | BPF_JGT | BPF_K:
958 case BPF_JMP | BPF_JGT | BPF_X:
959 case BPF_JMP | BPF_JSET | BPF_K:
960 case BPF_JMP | BPF_JSET | BPF_X:
961 /* A jump must set masks on targets */
962 masks[pc + 1 + filter[pc].jt] &= memvalid;
963 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 memvalid = ~0;
965 break;
966 }
967 }
968 error:
969 kfree(masks);
970 return ret;
971 }
972
chk_code_allowed(u16 code_to_probe)973 static bool chk_code_allowed(u16 code_to_probe)
974 {
975 static const bool codes[] = {
976 /* 32 bit ALU operations */
977 [BPF_ALU | BPF_ADD | BPF_K] = true,
978 [BPF_ALU | BPF_ADD | BPF_X] = true,
979 [BPF_ALU | BPF_SUB | BPF_K] = true,
980 [BPF_ALU | BPF_SUB | BPF_X] = true,
981 [BPF_ALU | BPF_MUL | BPF_K] = true,
982 [BPF_ALU | BPF_MUL | BPF_X] = true,
983 [BPF_ALU | BPF_DIV | BPF_K] = true,
984 [BPF_ALU | BPF_DIV | BPF_X] = true,
985 [BPF_ALU | BPF_MOD | BPF_K] = true,
986 [BPF_ALU | BPF_MOD | BPF_X] = true,
987 [BPF_ALU | BPF_AND | BPF_K] = true,
988 [BPF_ALU | BPF_AND | BPF_X] = true,
989 [BPF_ALU | BPF_OR | BPF_K] = true,
990 [BPF_ALU | BPF_OR | BPF_X] = true,
991 [BPF_ALU | BPF_XOR | BPF_K] = true,
992 [BPF_ALU | BPF_XOR | BPF_X] = true,
993 [BPF_ALU | BPF_LSH | BPF_K] = true,
994 [BPF_ALU | BPF_LSH | BPF_X] = true,
995 [BPF_ALU | BPF_RSH | BPF_K] = true,
996 [BPF_ALU | BPF_RSH | BPF_X] = true,
997 [BPF_ALU | BPF_NEG] = true,
998 /* Load instructions */
999 [BPF_LD | BPF_W | BPF_ABS] = true,
1000 [BPF_LD | BPF_H | BPF_ABS] = true,
1001 [BPF_LD | BPF_B | BPF_ABS] = true,
1002 [BPF_LD | BPF_W | BPF_LEN] = true,
1003 [BPF_LD | BPF_W | BPF_IND] = true,
1004 [BPF_LD | BPF_H | BPF_IND] = true,
1005 [BPF_LD | BPF_B | BPF_IND] = true,
1006 [BPF_LD | BPF_IMM] = true,
1007 [BPF_LD | BPF_MEM] = true,
1008 [BPF_LDX | BPF_W | BPF_LEN] = true,
1009 [BPF_LDX | BPF_B | BPF_MSH] = true,
1010 [BPF_LDX | BPF_IMM] = true,
1011 [BPF_LDX | BPF_MEM] = true,
1012 /* Store instructions */
1013 [BPF_ST] = true,
1014 [BPF_STX] = true,
1015 /* Misc instructions */
1016 [BPF_MISC | BPF_TAX] = true,
1017 [BPF_MISC | BPF_TXA] = true,
1018 /* Return instructions */
1019 [BPF_RET | BPF_K] = true,
1020 [BPF_RET | BPF_A] = true,
1021 /* Jump instructions */
1022 [BPF_JMP | BPF_JA] = true,
1023 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1024 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1025 [BPF_JMP | BPF_JGE | BPF_K] = true,
1026 [BPF_JMP | BPF_JGE | BPF_X] = true,
1027 [BPF_JMP | BPF_JGT | BPF_K] = true,
1028 [BPF_JMP | BPF_JGT | BPF_X] = true,
1029 [BPF_JMP | BPF_JSET | BPF_K] = true,
1030 [BPF_JMP | BPF_JSET | BPF_X] = true,
1031 };
1032
1033 if (code_to_probe >= ARRAY_SIZE(codes))
1034 return false;
1035
1036 return codes[code_to_probe];
1037 }
1038
bpf_check_basics_ok(const struct sock_filter * filter,unsigned int flen)1039 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1040 unsigned int flen)
1041 {
1042 if (filter == NULL)
1043 return false;
1044 if (flen == 0 || flen > BPF_MAXINSNS)
1045 return false;
1046
1047 return true;
1048 }
1049
1050 /**
1051 * bpf_check_classic - verify socket filter code
1052 * @filter: filter to verify
1053 * @flen: length of filter
1054 *
1055 * Check the user's filter code. If we let some ugly
1056 * filter code slip through kaboom! The filter must contain
1057 * no references or jumps that are out of range, no illegal
1058 * instructions, and must end with a RET instruction.
1059 *
1060 * All jumps are forward as they are not signed.
1061 *
1062 * Returns 0 if the rule set is legal or -EINVAL if not.
1063 */
bpf_check_classic(const struct sock_filter * filter,unsigned int flen)1064 static int bpf_check_classic(const struct sock_filter *filter,
1065 unsigned int flen)
1066 {
1067 bool anc_found;
1068 int pc;
1069
1070 /* Check the filter code now */
1071 for (pc = 0; pc < flen; pc++) {
1072 const struct sock_filter *ftest = &filter[pc];
1073
1074 /* May we actually operate on this code? */
1075 if (!chk_code_allowed(ftest->code))
1076 return -EINVAL;
1077
1078 /* Some instructions need special checks */
1079 switch (ftest->code) {
1080 case BPF_ALU | BPF_DIV | BPF_K:
1081 case BPF_ALU | BPF_MOD | BPF_K:
1082 /* Check for division by zero */
1083 if (ftest->k == 0)
1084 return -EINVAL;
1085 break;
1086 case BPF_ALU | BPF_LSH | BPF_K:
1087 case BPF_ALU | BPF_RSH | BPF_K:
1088 if (ftest->k >= 32)
1089 return -EINVAL;
1090 break;
1091 case BPF_LD | BPF_MEM:
1092 case BPF_LDX | BPF_MEM:
1093 case BPF_ST:
1094 case BPF_STX:
1095 /* Check for invalid memory addresses */
1096 if (ftest->k >= BPF_MEMWORDS)
1097 return -EINVAL;
1098 break;
1099 case BPF_JMP | BPF_JA:
1100 /* Note, the large ftest->k might cause loops.
1101 * Compare this with conditional jumps below,
1102 * where offsets are limited. --ANK (981016)
1103 */
1104 if (ftest->k >= (unsigned int)(flen - pc - 1))
1105 return -EINVAL;
1106 break;
1107 case BPF_JMP | BPF_JEQ | BPF_K:
1108 case BPF_JMP | BPF_JEQ | BPF_X:
1109 case BPF_JMP | BPF_JGE | BPF_K:
1110 case BPF_JMP | BPF_JGE | BPF_X:
1111 case BPF_JMP | BPF_JGT | BPF_K:
1112 case BPF_JMP | BPF_JGT | BPF_X:
1113 case BPF_JMP | BPF_JSET | BPF_K:
1114 case BPF_JMP | BPF_JSET | BPF_X:
1115 /* Both conditionals must be safe */
1116 if (pc + ftest->jt + 1 >= flen ||
1117 pc + ftest->jf + 1 >= flen)
1118 return -EINVAL;
1119 break;
1120 case BPF_LD | BPF_W | BPF_ABS:
1121 case BPF_LD | BPF_H | BPF_ABS:
1122 case BPF_LD | BPF_B | BPF_ABS:
1123 anc_found = false;
1124 if (bpf_anc_helper(ftest) & BPF_ANC)
1125 anc_found = true;
1126 /* Ancillary operation unknown or unsupported */
1127 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1128 return -EINVAL;
1129 }
1130 }
1131
1132 /* Last instruction must be a RET code */
1133 switch (filter[flen - 1].code) {
1134 case BPF_RET | BPF_K:
1135 case BPF_RET | BPF_A:
1136 return check_load_and_stores(filter, flen);
1137 }
1138
1139 return -EINVAL;
1140 }
1141
bpf_prog_store_orig_filter(struct bpf_prog * fp,const struct sock_fprog * fprog)1142 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1143 const struct sock_fprog *fprog)
1144 {
1145 unsigned int fsize = bpf_classic_proglen(fprog);
1146 struct sock_fprog_kern *fkprog;
1147
1148 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1149 if (!fp->orig_prog)
1150 return -ENOMEM;
1151
1152 fkprog = fp->orig_prog;
1153 fkprog->len = fprog->len;
1154
1155 fkprog->filter = kmemdup(fp->insns, fsize,
1156 GFP_KERNEL | __GFP_NOWARN);
1157 if (!fkprog->filter) {
1158 kfree(fp->orig_prog);
1159 return -ENOMEM;
1160 }
1161
1162 return 0;
1163 }
1164
bpf_release_orig_filter(struct bpf_prog * fp)1165 static void bpf_release_orig_filter(struct bpf_prog *fp)
1166 {
1167 struct sock_fprog_kern *fprog = fp->orig_prog;
1168
1169 if (fprog) {
1170 kfree(fprog->filter);
1171 kfree(fprog);
1172 }
1173 }
1174
__bpf_prog_release(struct bpf_prog * prog)1175 static void __bpf_prog_release(struct bpf_prog *prog)
1176 {
1177 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1178 bpf_prog_put(prog);
1179 } else {
1180 bpf_release_orig_filter(prog);
1181 bpf_prog_free(prog);
1182 }
1183 }
1184
__sk_filter_release(struct sk_filter * fp)1185 static void __sk_filter_release(struct sk_filter *fp)
1186 {
1187 __bpf_prog_release(fp->prog);
1188 kfree(fp);
1189 }
1190
1191 /**
1192 * sk_filter_release_rcu - Release a socket filter by rcu_head
1193 * @rcu: rcu_head that contains the sk_filter to free
1194 */
sk_filter_release_rcu(struct rcu_head * rcu)1195 static void sk_filter_release_rcu(struct rcu_head *rcu)
1196 {
1197 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1198
1199 __sk_filter_release(fp);
1200 }
1201
1202 /**
1203 * sk_filter_release - release a socket filter
1204 * @fp: filter to remove
1205 *
1206 * Remove a filter from a socket and release its resources.
1207 */
sk_filter_release(struct sk_filter * fp)1208 static void sk_filter_release(struct sk_filter *fp)
1209 {
1210 if (refcount_dec_and_test(&fp->refcnt))
1211 call_rcu(&fp->rcu, sk_filter_release_rcu);
1212 }
1213
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1214 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1215 {
1216 u32 filter_size = bpf_prog_size(fp->prog->len);
1217
1218 atomic_sub(filter_size, &sk->sk_omem_alloc);
1219 sk_filter_release(fp);
1220 }
1221
1222 /* try to charge the socket memory if there is space available
1223 * return true on success
1224 */
__sk_filter_charge(struct sock * sk,struct sk_filter * fp)1225 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1226 {
1227 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1228 u32 filter_size = bpf_prog_size(fp->prog->len);
1229
1230 /* same check as in sock_kmalloc() */
1231 if (filter_size <= optmem_max &&
1232 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1233 atomic_add(filter_size, &sk->sk_omem_alloc);
1234 return true;
1235 }
1236 return false;
1237 }
1238
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1239 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1240 {
1241 if (!refcount_inc_not_zero(&fp->refcnt))
1242 return false;
1243
1244 if (!__sk_filter_charge(sk, fp)) {
1245 sk_filter_release(fp);
1246 return false;
1247 }
1248 return true;
1249 }
1250
bpf_migrate_filter(struct bpf_prog * fp)1251 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1252 {
1253 struct sock_filter *old_prog;
1254 struct bpf_prog *old_fp;
1255 int err, new_len, old_len = fp->len;
1256 bool seen_ld_abs = false;
1257
1258 /* We are free to overwrite insns et al right here as it won't be used at
1259 * this point in time anymore internally after the migration to the eBPF
1260 * instruction representation.
1261 */
1262 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1263 sizeof(struct bpf_insn));
1264
1265 /* Conversion cannot happen on overlapping memory areas,
1266 * so we need to keep the user BPF around until the 2nd
1267 * pass. At this time, the user BPF is stored in fp->insns.
1268 */
1269 old_prog = kmemdup_array(fp->insns, old_len, sizeof(struct sock_filter),
1270 GFP_KERNEL | __GFP_NOWARN);
1271 if (!old_prog) {
1272 err = -ENOMEM;
1273 goto out_err;
1274 }
1275
1276 /* 1st pass: calculate the new program length. */
1277 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1278 &seen_ld_abs);
1279 if (err)
1280 goto out_err_free;
1281
1282 /* Expand fp for appending the new filter representation. */
1283 old_fp = fp;
1284 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1285 if (!fp) {
1286 /* The old_fp is still around in case we couldn't
1287 * allocate new memory, so uncharge on that one.
1288 */
1289 fp = old_fp;
1290 err = -ENOMEM;
1291 goto out_err_free;
1292 }
1293
1294 fp->len = new_len;
1295
1296 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1297 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1298 &seen_ld_abs);
1299 if (err)
1300 /* 2nd bpf_convert_filter() can fail only if it fails
1301 * to allocate memory, remapping must succeed. Note,
1302 * that at this time old_fp has already been released
1303 * by krealloc().
1304 */
1305 goto out_err_free;
1306
1307 fp = bpf_prog_select_runtime(fp, &err);
1308 if (err)
1309 goto out_err_free;
1310
1311 kfree(old_prog);
1312 return fp;
1313
1314 out_err_free:
1315 kfree(old_prog);
1316 out_err:
1317 __bpf_prog_release(fp);
1318 return ERR_PTR(err);
1319 }
1320
bpf_prepare_filter(struct bpf_prog * fp,bpf_aux_classic_check_t trans)1321 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1322 bpf_aux_classic_check_t trans)
1323 {
1324 int err;
1325
1326 fp->bpf_func = NULL;
1327 fp->jited = 0;
1328
1329 err = bpf_check_classic(fp->insns, fp->len);
1330 if (err) {
1331 __bpf_prog_release(fp);
1332 return ERR_PTR(err);
1333 }
1334
1335 /* There might be additional checks and transformations
1336 * needed on classic filters, f.e. in case of seccomp.
1337 */
1338 if (trans) {
1339 err = trans(fp->insns, fp->len);
1340 if (err) {
1341 __bpf_prog_release(fp);
1342 return ERR_PTR(err);
1343 }
1344 }
1345
1346 /* Probe if we can JIT compile the filter and if so, do
1347 * the compilation of the filter.
1348 */
1349 bpf_jit_compile(fp);
1350
1351 /* JIT compiler couldn't process this filter, so do the eBPF translation
1352 * for the optimized interpreter.
1353 */
1354 if (!fp->jited)
1355 fp = bpf_migrate_filter(fp);
1356
1357 return fp;
1358 }
1359
1360 /**
1361 * bpf_prog_create - create an unattached filter
1362 * @pfp: the unattached filter that is created
1363 * @fprog: the filter program
1364 *
1365 * Create a filter independent of any socket. We first run some
1366 * sanity checks on it to make sure it does not explode on us later.
1367 * If an error occurs or there is insufficient memory for the filter
1368 * a negative errno code is returned. On success the return is zero.
1369 */
bpf_prog_create(struct bpf_prog ** pfp,struct sock_fprog_kern * fprog)1370 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1371 {
1372 unsigned int fsize = bpf_classic_proglen(fprog);
1373 struct bpf_prog *fp;
1374
1375 /* Make sure new filter is there and in the right amounts. */
1376 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1377 return -EINVAL;
1378
1379 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1380 if (!fp)
1381 return -ENOMEM;
1382
1383 memcpy(fp->insns, fprog->filter, fsize);
1384
1385 fp->len = fprog->len;
1386 /* Since unattached filters are not copied back to user
1387 * space through sk_get_filter(), we do not need to hold
1388 * a copy here, and can spare us the work.
1389 */
1390 fp->orig_prog = NULL;
1391
1392 /* bpf_prepare_filter() already takes care of freeing
1393 * memory in case something goes wrong.
1394 */
1395 fp = bpf_prepare_filter(fp, NULL);
1396 if (IS_ERR(fp))
1397 return PTR_ERR(fp);
1398
1399 *pfp = fp;
1400 return 0;
1401 }
1402 EXPORT_SYMBOL_GPL(bpf_prog_create);
1403
1404 /**
1405 * bpf_prog_create_from_user - create an unattached filter from user buffer
1406 * @pfp: the unattached filter that is created
1407 * @fprog: the filter program
1408 * @trans: post-classic verifier transformation handler
1409 * @save_orig: save classic BPF program
1410 *
1411 * This function effectively does the same as bpf_prog_create(), only
1412 * that it builds up its insns buffer from user space provided buffer.
1413 * It also allows for passing a bpf_aux_classic_check_t handler.
1414 */
bpf_prog_create_from_user(struct bpf_prog ** pfp,struct sock_fprog * fprog,bpf_aux_classic_check_t trans,bool save_orig)1415 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1416 bpf_aux_classic_check_t trans, bool save_orig)
1417 {
1418 unsigned int fsize = bpf_classic_proglen(fprog);
1419 struct bpf_prog *fp;
1420 int err;
1421
1422 /* Make sure new filter is there and in the right amounts. */
1423 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1424 return -EINVAL;
1425
1426 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1427 if (!fp)
1428 return -ENOMEM;
1429
1430 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1431 __bpf_prog_free(fp);
1432 return -EFAULT;
1433 }
1434
1435 fp->len = fprog->len;
1436 fp->orig_prog = NULL;
1437
1438 if (save_orig) {
1439 err = bpf_prog_store_orig_filter(fp, fprog);
1440 if (err) {
1441 __bpf_prog_free(fp);
1442 return -ENOMEM;
1443 }
1444 }
1445
1446 /* bpf_prepare_filter() already takes care of freeing
1447 * memory in case something goes wrong.
1448 */
1449 fp = bpf_prepare_filter(fp, trans);
1450 if (IS_ERR(fp))
1451 return PTR_ERR(fp);
1452
1453 *pfp = fp;
1454 return 0;
1455 }
1456 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1457
bpf_prog_destroy(struct bpf_prog * fp)1458 void bpf_prog_destroy(struct bpf_prog *fp)
1459 {
1460 __bpf_prog_release(fp);
1461 }
1462 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1463
__sk_attach_prog(struct bpf_prog * prog,struct sock * sk)1464 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1465 {
1466 struct sk_filter *fp, *old_fp;
1467
1468 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1469 if (!fp)
1470 return -ENOMEM;
1471
1472 fp->prog = prog;
1473
1474 if (!__sk_filter_charge(sk, fp)) {
1475 kfree(fp);
1476 return -ENOMEM;
1477 }
1478 refcount_set(&fp->refcnt, 1);
1479
1480 old_fp = rcu_dereference_protected(sk->sk_filter,
1481 lockdep_sock_is_held(sk));
1482 rcu_assign_pointer(sk->sk_filter, fp);
1483
1484 if (old_fp)
1485 sk_filter_uncharge(sk, old_fp);
1486
1487 return 0;
1488 }
1489
1490 static
__get_filter(struct sock_fprog * fprog,struct sock * sk)1491 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1492 {
1493 unsigned int fsize = bpf_classic_proglen(fprog);
1494 struct bpf_prog *prog;
1495 int err;
1496
1497 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1498 return ERR_PTR(-EPERM);
1499
1500 /* Make sure new filter is there and in the right amounts. */
1501 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1502 return ERR_PTR(-EINVAL);
1503
1504 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1505 if (!prog)
1506 return ERR_PTR(-ENOMEM);
1507
1508 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1509 __bpf_prog_free(prog);
1510 return ERR_PTR(-EFAULT);
1511 }
1512
1513 prog->len = fprog->len;
1514
1515 err = bpf_prog_store_orig_filter(prog, fprog);
1516 if (err) {
1517 __bpf_prog_free(prog);
1518 return ERR_PTR(-ENOMEM);
1519 }
1520
1521 /* bpf_prepare_filter() already takes care of freeing
1522 * memory in case something goes wrong.
1523 */
1524 return bpf_prepare_filter(prog, NULL);
1525 }
1526
1527 /**
1528 * sk_attach_filter - attach a socket filter
1529 * @fprog: the filter program
1530 * @sk: the socket to use
1531 *
1532 * Attach the user's filter code. We first run some sanity checks on
1533 * it to make sure it does not explode on us later. If an error
1534 * occurs or there is insufficient memory for the filter a negative
1535 * errno code is returned. On success the return is zero.
1536 */
sk_attach_filter(struct sock_fprog * fprog,struct sock * sk)1537 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1538 {
1539 struct bpf_prog *prog = __get_filter(fprog, sk);
1540 int err;
1541
1542 if (IS_ERR(prog))
1543 return PTR_ERR(prog);
1544
1545 err = __sk_attach_prog(prog, sk);
1546 if (err < 0) {
1547 __bpf_prog_release(prog);
1548 return err;
1549 }
1550
1551 return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(sk_attach_filter);
1554
sk_reuseport_attach_filter(struct sock_fprog * fprog,struct sock * sk)1555 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1556 {
1557 struct bpf_prog *prog = __get_filter(fprog, sk);
1558 int err, optmem_max;
1559
1560 if (IS_ERR(prog))
1561 return PTR_ERR(prog);
1562
1563 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1564 if (bpf_prog_size(prog->len) > optmem_max)
1565 err = -ENOMEM;
1566 else
1567 err = reuseport_attach_prog(sk, prog);
1568
1569 if (err)
1570 __bpf_prog_release(prog);
1571
1572 return err;
1573 }
1574
__get_bpf(u32 ufd,struct sock * sk)1575 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1576 {
1577 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1578 return ERR_PTR(-EPERM);
1579
1580 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1581 }
1582
sk_attach_bpf(u32 ufd,struct sock * sk)1583 int sk_attach_bpf(u32 ufd, struct sock *sk)
1584 {
1585 struct bpf_prog *prog = __get_bpf(ufd, sk);
1586 int err;
1587
1588 if (IS_ERR(prog))
1589 return PTR_ERR(prog);
1590
1591 err = __sk_attach_prog(prog, sk);
1592 if (err < 0) {
1593 bpf_prog_put(prog);
1594 return err;
1595 }
1596
1597 return 0;
1598 }
1599
sk_reuseport_attach_bpf(u32 ufd,struct sock * sk)1600 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1601 {
1602 struct bpf_prog *prog;
1603 int err, optmem_max;
1604
1605 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1606 return -EPERM;
1607
1608 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1609 if (PTR_ERR(prog) == -EINVAL)
1610 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1611 if (IS_ERR(prog))
1612 return PTR_ERR(prog);
1613
1614 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1615 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1616 * bpf prog (e.g. sockmap). It depends on the
1617 * limitation imposed by bpf_prog_load().
1618 * Hence, sysctl_optmem_max is not checked.
1619 */
1620 if ((sk->sk_type != SOCK_STREAM &&
1621 sk->sk_type != SOCK_DGRAM) ||
1622 (sk->sk_protocol != IPPROTO_UDP &&
1623 sk->sk_protocol != IPPROTO_TCP) ||
1624 (sk->sk_family != AF_INET &&
1625 sk->sk_family != AF_INET6)) {
1626 err = -ENOTSUPP;
1627 goto err_prog_put;
1628 }
1629 } else {
1630 /* BPF_PROG_TYPE_SOCKET_FILTER */
1631 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1632 if (bpf_prog_size(prog->len) > optmem_max) {
1633 err = -ENOMEM;
1634 goto err_prog_put;
1635 }
1636 }
1637
1638 err = reuseport_attach_prog(sk, prog);
1639 err_prog_put:
1640 if (err)
1641 bpf_prog_put(prog);
1642
1643 return err;
1644 }
1645
sk_reuseport_prog_free(struct bpf_prog * prog)1646 void sk_reuseport_prog_free(struct bpf_prog *prog)
1647 {
1648 if (!prog)
1649 return;
1650
1651 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1652 bpf_prog_put(prog);
1653 else
1654 bpf_prog_destroy(prog);
1655 }
1656
__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 /* This is quite flexible, some examples:
2014 *
2015 * from_size == 0, to_size > 0, seed := csum --> pushing data
2016 * from_size > 0, to_size == 0, seed := csum --> pulling data
2017 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2018 *
2019 * Even for diffing, from_size and to_size don't need to be equal.
2020 */
2021
2022 __wsum ret = seed;
2023
2024 if (from_size && to_size)
2025 ret = csum_sub(csum_partial(to, to_size, ret),
2026 csum_partial(from, from_size, 0));
2027 else if (to_size)
2028 ret = csum_partial(to, to_size, ret);
2029
2030 else if (from_size)
2031 ret = ~csum_partial(from, from_size, ~ret);
2032
2033 return csum_from32to16((__force unsigned int)ret);
2034 }
2035
2036 static const struct bpf_func_proto bpf_csum_diff_proto = {
2037 .func = bpf_csum_diff,
2038 .gpl_only = false,
2039 .pkt_access = true,
2040 .ret_type = RET_INTEGER,
2041 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2042 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2043 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2044 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2045 .arg5_type = ARG_ANYTHING,
2046 };
2047
BPF_CALL_2(bpf_csum_update,struct sk_buff *,skb,__wsum,csum)2048 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2049 {
2050 /* The interface is to be used in combination with bpf_csum_diff()
2051 * for direct packet writes. csum rotation for alignment as well
2052 * as emulating csum_sub() can be done from the eBPF program.
2053 */
2054 if (skb->ip_summed == CHECKSUM_COMPLETE)
2055 return (skb->csum = csum_add(skb->csum, csum));
2056
2057 return -ENOTSUPP;
2058 }
2059
2060 static const struct bpf_func_proto bpf_csum_update_proto = {
2061 .func = bpf_csum_update,
2062 .gpl_only = false,
2063 .ret_type = RET_INTEGER,
2064 .arg1_type = ARG_PTR_TO_CTX,
2065 .arg2_type = ARG_ANYTHING,
2066 };
2067
BPF_CALL_2(bpf_csum_level,struct sk_buff *,skb,u64,level)2068 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2069 {
2070 /* The interface is to be used in combination with bpf_skb_adjust_room()
2071 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2072 * is passed as flags, for example.
2073 */
2074 switch (level) {
2075 case BPF_CSUM_LEVEL_INC:
2076 __skb_incr_checksum_unnecessary(skb);
2077 break;
2078 case BPF_CSUM_LEVEL_DEC:
2079 __skb_decr_checksum_unnecessary(skb);
2080 break;
2081 case BPF_CSUM_LEVEL_RESET:
2082 __skb_reset_checksum_unnecessary(skb);
2083 break;
2084 case BPF_CSUM_LEVEL_QUERY:
2085 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2086 skb->csum_level : -EACCES;
2087 default:
2088 return -EINVAL;
2089 }
2090
2091 return 0;
2092 }
2093
2094 static const struct bpf_func_proto bpf_csum_level_proto = {
2095 .func = bpf_csum_level,
2096 .gpl_only = false,
2097 .ret_type = RET_INTEGER,
2098 .arg1_type = ARG_PTR_TO_CTX,
2099 .arg2_type = ARG_ANYTHING,
2100 };
2101
__bpf_rx_skb(struct net_device * dev,struct sk_buff * skb)2102 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2103 {
2104 return dev_forward_skb_nomtu(dev, skb);
2105 }
2106
__bpf_rx_skb_no_mac(struct net_device * dev,struct sk_buff * skb)2107 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2108 struct sk_buff *skb)
2109 {
2110 int ret = ____dev_forward_skb(dev, skb, false);
2111
2112 if (likely(!ret)) {
2113 skb->dev = dev;
2114 ret = netif_rx(skb);
2115 }
2116
2117 return ret;
2118 }
2119
__bpf_tx_skb(struct net_device * dev,struct sk_buff * skb)2120 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2121 {
2122 int ret;
2123
2124 if (dev_xmit_recursion()) {
2125 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2126 kfree_skb(skb);
2127 return -ENETDOWN;
2128 }
2129
2130 skb->dev = dev;
2131 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2132 skb_clear_tstamp(skb);
2133
2134 dev_xmit_recursion_inc();
2135 ret = dev_queue_xmit(skb);
2136 dev_xmit_recursion_dec();
2137
2138 return ret;
2139 }
2140
__bpf_redirect_no_mac(struct sk_buff * skb,struct net_device * dev,u32 flags)2141 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2142 u32 flags)
2143 {
2144 unsigned int mlen = skb_network_offset(skb);
2145
2146 if (unlikely(skb->len <= mlen)) {
2147 kfree_skb(skb);
2148 return -ERANGE;
2149 }
2150
2151 if (mlen) {
2152 __skb_pull(skb, mlen);
2153
2154 /* At ingress, the mac header has already been pulled once.
2155 * At egress, skb_pospull_rcsum has to be done in case that
2156 * the skb is originated from ingress (i.e. a forwarded skb)
2157 * to ensure that rcsum starts at net header.
2158 */
2159 if (!skb_at_tc_ingress(skb))
2160 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2161 }
2162 skb_pop_mac_header(skb);
2163 skb_reset_mac_len(skb);
2164 return flags & BPF_F_INGRESS ?
2165 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2166 }
2167
__bpf_redirect_common(struct sk_buff * skb,struct net_device * dev,u32 flags)2168 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2169 u32 flags)
2170 {
2171 /* Verify that a link layer header is carried */
2172 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2173 kfree_skb(skb);
2174 return -ERANGE;
2175 }
2176
2177 bpf_push_mac_rcsum(skb);
2178 return flags & BPF_F_INGRESS ?
2179 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2180 }
2181
__bpf_redirect(struct sk_buff * skb,struct net_device * dev,u32 flags)2182 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2183 u32 flags)
2184 {
2185 if (dev_is_mac_header_xmit(dev))
2186 return __bpf_redirect_common(skb, dev, flags);
2187 else
2188 return __bpf_redirect_no_mac(skb, dev, flags);
2189 }
2190
2191 #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)2192 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2193 struct net_device *dev, struct bpf_nh_params *nh)
2194 {
2195 u32 hh_len = LL_RESERVED_SPACE(dev);
2196 const struct in6_addr *nexthop;
2197 struct dst_entry *dst = NULL;
2198 struct neighbour *neigh;
2199
2200 if (dev_xmit_recursion()) {
2201 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2202 goto out_drop;
2203 }
2204
2205 skb->dev = dev;
2206 skb_clear_tstamp(skb);
2207
2208 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2209 skb = skb_expand_head(skb, hh_len);
2210 if (!skb)
2211 return -ENOMEM;
2212 }
2213
2214 rcu_read_lock();
2215 if (!nh) {
2216 dst = skb_dst(skb);
2217 nexthop = rt6_nexthop(dst_rt6_info(dst),
2218 &ipv6_hdr(skb)->daddr);
2219 } else {
2220 nexthop = &nh->ipv6_nh;
2221 }
2222 neigh = ip_neigh_gw6(dev, nexthop);
2223 if (likely(!IS_ERR(neigh))) {
2224 int ret;
2225
2226 sock_confirm_neigh(skb, neigh);
2227 local_bh_disable();
2228 dev_xmit_recursion_inc();
2229 ret = neigh_output(neigh, skb, false);
2230 dev_xmit_recursion_dec();
2231 local_bh_enable();
2232 rcu_read_unlock();
2233 return ret;
2234 }
2235 rcu_read_unlock();
2236 if (dst)
2237 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2238 out_drop:
2239 kfree_skb(skb);
2240 return -ENETDOWN;
2241 }
2242
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2243 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2244 struct bpf_nh_params *nh)
2245 {
2246 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2247 struct net *net = dev_net(dev);
2248 int err, ret = NET_XMIT_DROP;
2249
2250 if (!nh) {
2251 struct dst_entry *dst;
2252 struct flowi6 fl6 = {
2253 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2254 .flowi6_mark = skb->mark,
2255 .flowlabel = ip6_flowinfo(ip6h),
2256 .flowi6_oif = dev->ifindex,
2257 .flowi6_proto = ip6h->nexthdr,
2258 .daddr = ip6h->daddr,
2259 .saddr = ip6h->saddr,
2260 };
2261
2262 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2263 if (IS_ERR(dst))
2264 goto out_drop;
2265
2266 skb_dst_set(skb, dst);
2267 } else if (nh->nh_family != AF_INET6) {
2268 goto out_drop;
2269 }
2270
2271 err = bpf_out_neigh_v6(net, skb, dev, nh);
2272 if (unlikely(net_xmit_eval(err)))
2273 DEV_STATS_INC(dev, tx_errors);
2274 else
2275 ret = NET_XMIT_SUCCESS;
2276 goto out_xmit;
2277 out_drop:
2278 DEV_STATS_INC(dev, tx_errors);
2279 kfree_skb(skb);
2280 out_xmit:
2281 return ret;
2282 }
2283 #else
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2284 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2285 struct bpf_nh_params *nh)
2286 {
2287 kfree_skb(skb);
2288 return NET_XMIT_DROP;
2289 }
2290 #endif /* CONFIG_IPV6 */
2291
2292 #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)2293 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2294 struct net_device *dev, struct bpf_nh_params *nh)
2295 {
2296 u32 hh_len = LL_RESERVED_SPACE(dev);
2297 struct neighbour *neigh;
2298 bool is_v6gw = false;
2299
2300 if (dev_xmit_recursion()) {
2301 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2302 goto out_drop;
2303 }
2304
2305 skb->dev = dev;
2306 skb_clear_tstamp(skb);
2307
2308 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2309 skb = skb_expand_head(skb, hh_len);
2310 if (!skb)
2311 return -ENOMEM;
2312 }
2313
2314 rcu_read_lock();
2315 if (!nh) {
2316 struct rtable *rt = skb_rtable(skb);
2317
2318 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2319 } else if (nh->nh_family == AF_INET6) {
2320 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2321 is_v6gw = true;
2322 } else if (nh->nh_family == AF_INET) {
2323 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2324 } else {
2325 rcu_read_unlock();
2326 goto out_drop;
2327 }
2328
2329 if (likely(!IS_ERR(neigh))) {
2330 int ret;
2331
2332 sock_confirm_neigh(skb, neigh);
2333 local_bh_disable();
2334 dev_xmit_recursion_inc();
2335 ret = neigh_output(neigh, skb, is_v6gw);
2336 dev_xmit_recursion_dec();
2337 local_bh_enable();
2338 rcu_read_unlock();
2339 return ret;
2340 }
2341 rcu_read_unlock();
2342 out_drop:
2343 kfree_skb(skb);
2344 return -ENETDOWN;
2345 }
2346
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2347 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2348 struct bpf_nh_params *nh)
2349 {
2350 const struct iphdr *ip4h = ip_hdr(skb);
2351 struct net *net = dev_net(dev);
2352 int err, ret = NET_XMIT_DROP;
2353
2354 if (!nh) {
2355 struct flowi4 fl4 = {
2356 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2357 .flowi4_mark = skb->mark,
2358 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip4h)),
2359 .flowi4_oif = dev->ifindex,
2360 .flowi4_proto = ip4h->protocol,
2361 .daddr = ip4h->daddr,
2362 .saddr = ip4h->saddr,
2363 };
2364 struct rtable *rt;
2365
2366 rt = ip_route_output_flow(net, &fl4, NULL);
2367 if (IS_ERR(rt))
2368 goto out_drop;
2369 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2370 ip_rt_put(rt);
2371 goto out_drop;
2372 }
2373
2374 skb_dst_set(skb, &rt->dst);
2375 }
2376
2377 err = bpf_out_neigh_v4(net, skb, dev, nh);
2378 if (unlikely(net_xmit_eval(err)))
2379 DEV_STATS_INC(dev, tx_errors);
2380 else
2381 ret = NET_XMIT_SUCCESS;
2382 goto out_xmit;
2383 out_drop:
2384 DEV_STATS_INC(dev, tx_errors);
2385 kfree_skb(skb);
2386 out_xmit:
2387 return ret;
2388 }
2389 #else
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2390 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2391 struct bpf_nh_params *nh)
2392 {
2393 kfree_skb(skb);
2394 return NET_XMIT_DROP;
2395 }
2396 #endif /* CONFIG_INET */
2397
__bpf_redirect_neigh(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2398 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2399 struct bpf_nh_params *nh)
2400 {
2401 struct ethhdr *ethh = eth_hdr(skb);
2402
2403 if (unlikely(skb->mac_header >= skb->network_header))
2404 goto out;
2405 bpf_push_mac_rcsum(skb);
2406 if (is_multicast_ether_addr(ethh->h_dest))
2407 goto out;
2408
2409 skb_pull(skb, sizeof(*ethh));
2410 skb_unset_mac_header(skb);
2411 skb_reset_network_header(skb);
2412
2413 if (skb->protocol == htons(ETH_P_IP))
2414 return __bpf_redirect_neigh_v4(skb, dev, nh);
2415 else if (skb->protocol == htons(ETH_P_IPV6))
2416 return __bpf_redirect_neigh_v6(skb, dev, nh);
2417 out:
2418 kfree_skb(skb);
2419 return -ENOTSUPP;
2420 }
2421
2422 /* Internal, non-exposed redirect flags. */
2423 enum {
2424 BPF_F_NEIGH = (1ULL << 16),
2425 BPF_F_PEER = (1ULL << 17),
2426 BPF_F_NEXTHOP = (1ULL << 18),
2427 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2428 };
2429
BPF_CALL_3(bpf_clone_redirect,struct sk_buff *,skb,u32,ifindex,u64,flags)2430 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2431 {
2432 struct net_device *dev;
2433 struct sk_buff *clone;
2434 int ret;
2435
2436 BUILD_BUG_ON(BPF_F_REDIRECT_INTERNAL & BPF_F_REDIRECT_FLAGS);
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
skb_get_peer_dev(struct net_device * dev)2472 static struct net_device *skb_get_peer_dev(struct net_device *dev)
2473 {
2474 const struct net_device_ops *ops = dev->netdev_ops;
2475
2476 if (likely(ops->ndo_get_peer_dev))
2477 return INDIRECT_CALL_1(ops->ndo_get_peer_dev,
2478 netkit_peer_dev, dev);
2479 return NULL;
2480 }
2481
skb_do_redirect(struct sk_buff * skb)2482 int skb_do_redirect(struct sk_buff *skb)
2483 {
2484 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2485 struct net *net = dev_net(skb->dev);
2486 struct net_device *dev;
2487 u32 flags = ri->flags;
2488
2489 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2490 ri->tgt_index = 0;
2491 ri->flags = 0;
2492 if (unlikely(!dev))
2493 goto out_drop;
2494 if (flags & BPF_F_PEER) {
2495 if (unlikely(!skb_at_tc_ingress(skb)))
2496 goto out_drop;
2497 dev = skb_get_peer_dev(dev);
2498 if (unlikely(!dev ||
2499 !(dev->flags & IFF_UP) ||
2500 net_eq(net, dev_net(dev))))
2501 goto out_drop;
2502 skb->dev = dev;
2503 dev_sw_netstats_rx_add(dev, skb->len);
2504 return -EAGAIN;
2505 }
2506 return flags & BPF_F_NEIGH ?
2507 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2508 &ri->nh : NULL) :
2509 __bpf_redirect(skb, dev, flags);
2510 out_drop:
2511 kfree_skb(skb);
2512 return -EINVAL;
2513 }
2514
BPF_CALL_2(bpf_redirect,u32,ifindex,u64,flags)2515 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2516 {
2517 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2518
2519 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2520 return TC_ACT_SHOT;
2521
2522 ri->flags = flags;
2523 ri->tgt_index = ifindex;
2524
2525 return TC_ACT_REDIRECT;
2526 }
2527
2528 static const struct bpf_func_proto bpf_redirect_proto = {
2529 .func = bpf_redirect,
2530 .gpl_only = false,
2531 .ret_type = RET_INTEGER,
2532 .arg1_type = ARG_ANYTHING,
2533 .arg2_type = ARG_ANYTHING,
2534 };
2535
BPF_CALL_2(bpf_redirect_peer,u32,ifindex,u64,flags)2536 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2537 {
2538 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2539
2540 if (unlikely(flags))
2541 return TC_ACT_SHOT;
2542
2543 ri->flags = BPF_F_PEER;
2544 ri->tgt_index = ifindex;
2545
2546 return TC_ACT_REDIRECT;
2547 }
2548
2549 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2550 .func = bpf_redirect_peer,
2551 .gpl_only = false,
2552 .ret_type = RET_INTEGER,
2553 .arg1_type = ARG_ANYTHING,
2554 .arg2_type = ARG_ANYTHING,
2555 };
2556
BPF_CALL_4(bpf_redirect_neigh,u32,ifindex,struct bpf_redir_neigh *,params,int,plen,u64,flags)2557 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2558 int, plen, u64, flags)
2559 {
2560 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2561
2562 if (unlikely((plen && plen < sizeof(*params)) || flags))
2563 return TC_ACT_SHOT;
2564
2565 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2566 ri->tgt_index = ifindex;
2567
2568 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2569 if (plen)
2570 memcpy(&ri->nh, params, sizeof(ri->nh));
2571
2572 return TC_ACT_REDIRECT;
2573 }
2574
2575 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2576 .func = bpf_redirect_neigh,
2577 .gpl_only = false,
2578 .ret_type = RET_INTEGER,
2579 .arg1_type = ARG_ANYTHING,
2580 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2581 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2582 .arg4_type = ARG_ANYTHING,
2583 };
2584
BPF_CALL_2(bpf_msg_apply_bytes,struct sk_msg *,msg,u32,bytes)2585 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2586 {
2587 msg->apply_bytes = bytes;
2588 return 0;
2589 }
2590
2591 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2592 .func = bpf_msg_apply_bytes,
2593 .gpl_only = false,
2594 .ret_type = RET_INTEGER,
2595 .arg1_type = ARG_PTR_TO_CTX,
2596 .arg2_type = ARG_ANYTHING,
2597 };
2598
BPF_CALL_2(bpf_msg_cork_bytes,struct sk_msg *,msg,u32,bytes)2599 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2600 {
2601 msg->cork_bytes = bytes;
2602 return 0;
2603 }
2604
sk_msg_reset_curr(struct sk_msg * msg)2605 static void sk_msg_reset_curr(struct sk_msg *msg)
2606 {
2607 if (!msg->sg.size) {
2608 msg->sg.curr = msg->sg.start;
2609 msg->sg.copybreak = 0;
2610 } else {
2611 u32 i = msg->sg.end;
2612
2613 sk_msg_iter_var_prev(i);
2614 msg->sg.curr = i;
2615 msg->sg.copybreak = msg->sg.data[i].length;
2616 }
2617 }
2618
2619 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2620 .func = bpf_msg_cork_bytes,
2621 .gpl_only = false,
2622 .ret_type = RET_INTEGER,
2623 .arg1_type = ARG_PTR_TO_CTX,
2624 .arg2_type = ARG_ANYTHING,
2625 };
2626
BPF_CALL_4(bpf_msg_pull_data,struct sk_msg *,msg,u32,start,u32,end,u64,flags)2627 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2628 u32, end, u64, flags)
2629 {
2630 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2631 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2632 struct scatterlist *sge;
2633 u8 *raw, *to, *from;
2634 struct page *page;
2635
2636 if (unlikely(flags || end <= start))
2637 return -EINVAL;
2638
2639 /* First find the starting scatterlist element */
2640 i = msg->sg.start;
2641 do {
2642 offset += len;
2643 len = sk_msg_elem(msg, i)->length;
2644 if (start < offset + len)
2645 break;
2646 sk_msg_iter_var_next(i);
2647 } while (i != msg->sg.end);
2648
2649 if (unlikely(start >= offset + len))
2650 return -EINVAL;
2651
2652 first_sge = i;
2653 /* The start may point into the sg element so we need to also
2654 * account for the headroom.
2655 */
2656 bytes_sg_total = start - offset + bytes;
2657 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2658 goto out;
2659
2660 /* At this point we need to linearize multiple scatterlist
2661 * elements or a single shared page. Either way we need to
2662 * copy into a linear buffer exclusively owned by BPF. Then
2663 * place the buffer in the scatterlist and fixup the original
2664 * entries by removing the entries now in the linear buffer
2665 * and shifting the remaining entries. For now we do not try
2666 * to copy partial entries to avoid complexity of running out
2667 * of sg_entry slots. The downside is reading a single byte
2668 * will copy the entire sg entry.
2669 */
2670 do {
2671 copy += sk_msg_elem(msg, i)->length;
2672 sk_msg_iter_var_next(i);
2673 if (bytes_sg_total <= copy)
2674 break;
2675 } while (i != msg->sg.end);
2676 last_sge = i;
2677
2678 if (unlikely(bytes_sg_total > copy))
2679 return -EINVAL;
2680
2681 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2682 get_order(copy));
2683 if (unlikely(!page))
2684 return -ENOMEM;
2685
2686 raw = page_address(page);
2687 i = first_sge;
2688 do {
2689 sge = sk_msg_elem(msg, i);
2690 from = sg_virt(sge);
2691 len = sge->length;
2692 to = raw + poffset;
2693
2694 memcpy(to, from, len);
2695 poffset += len;
2696 sge->length = 0;
2697 put_page(sg_page(sge));
2698
2699 sk_msg_iter_var_next(i);
2700 } while (i != last_sge);
2701
2702 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2703
2704 /* To repair sg ring we need to shift entries. If we only
2705 * had a single entry though we can just replace it and
2706 * be done. Otherwise walk the ring and shift the entries.
2707 */
2708 WARN_ON_ONCE(last_sge == first_sge);
2709 shift = last_sge > first_sge ?
2710 last_sge - first_sge - 1 :
2711 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2712 if (!shift)
2713 goto out;
2714
2715 i = first_sge;
2716 sk_msg_iter_var_next(i);
2717 do {
2718 u32 move_from;
2719
2720 if (i + shift >= NR_MSG_FRAG_IDS)
2721 move_from = i + shift - NR_MSG_FRAG_IDS;
2722 else
2723 move_from = i + shift;
2724 if (move_from == msg->sg.end)
2725 break;
2726
2727 msg->sg.data[i] = msg->sg.data[move_from];
2728 msg->sg.data[move_from].length = 0;
2729 msg->sg.data[move_from].page_link = 0;
2730 msg->sg.data[move_from].offset = 0;
2731 sk_msg_iter_var_next(i);
2732 } while (1);
2733
2734 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2735 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2736 msg->sg.end - shift;
2737 out:
2738 sk_msg_reset_curr(msg);
2739 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2740 msg->data_end = msg->data + bytes;
2741 return 0;
2742 }
2743
2744 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2745 .func = bpf_msg_pull_data,
2746 .gpl_only = false,
2747 .ret_type = RET_INTEGER,
2748 .arg1_type = ARG_PTR_TO_CTX,
2749 .arg2_type = ARG_ANYTHING,
2750 .arg3_type = ARG_ANYTHING,
2751 .arg4_type = ARG_ANYTHING,
2752 };
2753
BPF_CALL_4(bpf_msg_push_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2754 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2755 u32, len, u64, flags)
2756 {
2757 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2758 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2759 u8 *raw, *to, *from;
2760 struct page *page;
2761
2762 if (unlikely(flags))
2763 return -EINVAL;
2764
2765 if (unlikely(len == 0))
2766 return 0;
2767
2768 /* First find the starting scatterlist element */
2769 i = msg->sg.start;
2770 do {
2771 offset += l;
2772 l = sk_msg_elem(msg, i)->length;
2773
2774 if (start < offset + l)
2775 break;
2776 sk_msg_iter_var_next(i);
2777 } while (i != msg->sg.end);
2778
2779 if (start > offset + l)
2780 return -EINVAL;
2781
2782 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2783
2784 /* If no space available will fallback to copy, we need at
2785 * least one scatterlist elem available to push data into
2786 * when start aligns to the beginning of an element or two
2787 * when it falls inside an element. We handle the start equals
2788 * offset case because its the common case for inserting a
2789 * header.
2790 */
2791 if (!space || (space == 1 && start != offset))
2792 copy = msg->sg.data[i].length;
2793
2794 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2795 get_order(copy + len));
2796 if (unlikely(!page))
2797 return -ENOMEM;
2798
2799 if (copy) {
2800 int front, back;
2801
2802 raw = page_address(page);
2803
2804 if (i == msg->sg.end)
2805 sk_msg_iter_var_prev(i);
2806 psge = sk_msg_elem(msg, i);
2807 front = start - offset;
2808 back = psge->length - front;
2809 from = sg_virt(psge);
2810
2811 if (front)
2812 memcpy(raw, from, front);
2813
2814 if (back) {
2815 from += front;
2816 to = raw + front + len;
2817
2818 memcpy(to, from, back);
2819 }
2820
2821 put_page(sg_page(psge));
2822 new = i;
2823 goto place_new;
2824 }
2825
2826 if (start - offset) {
2827 if (i == msg->sg.end)
2828 sk_msg_iter_var_prev(i);
2829 psge = sk_msg_elem(msg, i);
2830 rsge = sk_msg_elem_cpy(msg, i);
2831
2832 psge->length = start - offset;
2833 rsge.length -= psge->length;
2834 rsge.offset += start;
2835
2836 sk_msg_iter_var_next(i);
2837 sg_unmark_end(psge);
2838 sg_unmark_end(&rsge);
2839 }
2840
2841 /* Slot(s) to place newly allocated data */
2842 sk_msg_iter_next(msg, end);
2843 new = i;
2844 sk_msg_iter_var_next(i);
2845
2846 if (i == msg->sg.end) {
2847 if (!rsge.length)
2848 goto place_new;
2849 sk_msg_iter_next(msg, end);
2850 goto place_new;
2851 }
2852
2853 /* Shift one or two slots as needed */
2854 sge = sk_msg_elem_cpy(msg, new);
2855 sg_unmark_end(&sge);
2856
2857 nsge = sk_msg_elem_cpy(msg, i);
2858 if (rsge.length) {
2859 sk_msg_iter_var_next(i);
2860 nnsge = sk_msg_elem_cpy(msg, i);
2861 sk_msg_iter_next(msg, end);
2862 }
2863
2864 while (i != msg->sg.end) {
2865 msg->sg.data[i] = sge;
2866 sge = nsge;
2867 sk_msg_iter_var_next(i);
2868 if (rsge.length) {
2869 nsge = nnsge;
2870 nnsge = sk_msg_elem_cpy(msg, i);
2871 } else {
2872 nsge = sk_msg_elem_cpy(msg, i);
2873 }
2874 }
2875
2876 place_new:
2877 /* Place newly allocated data buffer */
2878 sk_mem_charge(msg->sk, len);
2879 msg->sg.size += len;
2880 __clear_bit(new, msg->sg.copy);
2881 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2882 if (rsge.length) {
2883 get_page(sg_page(&rsge));
2884 sk_msg_iter_var_next(new);
2885 msg->sg.data[new] = rsge;
2886 }
2887
2888 sk_msg_reset_curr(msg);
2889 sk_msg_compute_data_pointers(msg);
2890 return 0;
2891 }
2892
2893 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2894 .func = bpf_msg_push_data,
2895 .gpl_only = false,
2896 .ret_type = RET_INTEGER,
2897 .arg1_type = ARG_PTR_TO_CTX,
2898 .arg2_type = ARG_ANYTHING,
2899 .arg3_type = ARG_ANYTHING,
2900 .arg4_type = ARG_ANYTHING,
2901 };
2902
sk_msg_shift_left(struct sk_msg * msg,int i)2903 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2904 {
2905 struct scatterlist *sge = sk_msg_elem(msg, i);
2906 int prev;
2907
2908 put_page(sg_page(sge));
2909 do {
2910 prev = i;
2911 sk_msg_iter_var_next(i);
2912 msg->sg.data[prev] = msg->sg.data[i];
2913 } while (i != msg->sg.end);
2914
2915 sk_msg_iter_prev(msg, end);
2916 }
2917
sk_msg_shift_right(struct sk_msg * msg,int i)2918 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2919 {
2920 struct scatterlist tmp, sge;
2921
2922 sk_msg_iter_next(msg, end);
2923 sge = sk_msg_elem_cpy(msg, i);
2924 sk_msg_iter_var_next(i);
2925 tmp = sk_msg_elem_cpy(msg, i);
2926
2927 while (i != msg->sg.end) {
2928 msg->sg.data[i] = sge;
2929 sk_msg_iter_var_next(i);
2930 sge = tmp;
2931 tmp = sk_msg_elem_cpy(msg, i);
2932 }
2933 }
2934
BPF_CALL_4(bpf_msg_pop_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2935 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2936 u32, len, u64, flags)
2937 {
2938 u32 i = 0, l = 0, space, offset = 0;
2939 u64 last = start + len;
2940 int pop;
2941
2942 if (unlikely(flags))
2943 return -EINVAL;
2944
2945 if (unlikely(len == 0))
2946 return 0;
2947
2948 /* First find the starting scatterlist element */
2949 i = msg->sg.start;
2950 do {
2951 offset += l;
2952 l = sk_msg_elem(msg, i)->length;
2953
2954 if (start < offset + l)
2955 break;
2956 sk_msg_iter_var_next(i);
2957 } while (i != msg->sg.end);
2958
2959 /* Bounds checks: start and pop must be inside message */
2960 if (start >= offset + l || last > msg->sg.size)
2961 return -EINVAL;
2962
2963 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2964
2965 pop = len;
2966 /* --------------| offset
2967 * -| start |-------- len -------|
2968 *
2969 * |----- a ----|-------- pop -------|----- b ----|
2970 * |______________________________________________| length
2971 *
2972 *
2973 * a: region at front of scatter element to save
2974 * b: region at back of scatter element to save when length > A + pop
2975 * pop: region to pop from element, same as input 'pop' here will be
2976 * decremented below per iteration.
2977 *
2978 * Two top-level cases to handle when start != offset, first B is non
2979 * zero and second B is zero corresponding to when a pop includes more
2980 * than one element.
2981 *
2982 * Then if B is non-zero AND there is no space allocate space and
2983 * compact A, B regions into page. If there is space shift ring to
2984 * the right free'ing the next element in ring to place B, leaving
2985 * A untouched except to reduce length.
2986 */
2987 if (start != offset) {
2988 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2989 int a = start - offset;
2990 int b = sge->length - pop - a;
2991
2992 sk_msg_iter_var_next(i);
2993
2994 if (b > 0) {
2995 if (space) {
2996 sge->length = a;
2997 sk_msg_shift_right(msg, i);
2998 nsge = sk_msg_elem(msg, i);
2999 get_page(sg_page(sge));
3000 sg_set_page(nsge,
3001 sg_page(sge),
3002 b, sge->offset + pop + a);
3003 } else {
3004 struct page *page, *orig;
3005 u8 *to, *from;
3006
3007 page = alloc_pages(__GFP_NOWARN |
3008 __GFP_COMP | GFP_ATOMIC,
3009 get_order(a + b));
3010 if (unlikely(!page))
3011 return -ENOMEM;
3012
3013 orig = sg_page(sge);
3014 from = sg_virt(sge);
3015 to = page_address(page);
3016 memcpy(to, from, a);
3017 memcpy(to + a, from + a + pop, b);
3018 sg_set_page(sge, page, a + b, 0);
3019 put_page(orig);
3020 }
3021 pop = 0;
3022 } else {
3023 pop -= (sge->length - a);
3024 sge->length = a;
3025 }
3026 }
3027
3028 /* From above the current layout _must_ be as follows,
3029 *
3030 * -| offset
3031 * -| start
3032 *
3033 * |---- pop ---|---------------- b ------------|
3034 * |____________________________________________| length
3035 *
3036 * Offset and start of the current msg elem are equal because in the
3037 * previous case we handled offset != start and either consumed the
3038 * entire element and advanced to the next element OR pop == 0.
3039 *
3040 * Two cases to handle here are first pop is less than the length
3041 * leaving some remainder b above. Simply adjust the element's layout
3042 * in this case. Or pop >= length of the element so that b = 0. In this
3043 * case advance to next element decrementing pop.
3044 */
3045 while (pop) {
3046 struct scatterlist *sge = sk_msg_elem(msg, i);
3047
3048 if (pop < sge->length) {
3049 sge->length -= pop;
3050 sge->offset += pop;
3051 pop = 0;
3052 } else {
3053 pop -= sge->length;
3054 sk_msg_shift_left(msg, i);
3055 }
3056 }
3057
3058 sk_mem_uncharge(msg->sk, len - pop);
3059 msg->sg.size -= (len - pop);
3060 sk_msg_reset_curr(msg);
3061 sk_msg_compute_data_pointers(msg);
3062 return 0;
3063 }
3064
3065 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3066 .func = bpf_msg_pop_data,
3067 .gpl_only = false,
3068 .ret_type = RET_INTEGER,
3069 .arg1_type = ARG_PTR_TO_CTX,
3070 .arg2_type = ARG_ANYTHING,
3071 .arg3_type = ARG_ANYTHING,
3072 .arg4_type = ARG_ANYTHING,
3073 };
3074
3075 #ifdef CONFIG_CGROUP_NET_CLASSID
BPF_CALL_0(bpf_get_cgroup_classid_curr)3076 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3077 {
3078 return __task_get_classid(current);
3079 }
3080
3081 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3082 .func = bpf_get_cgroup_classid_curr,
3083 .gpl_only = false,
3084 .ret_type = RET_INTEGER,
3085 };
3086
BPF_CALL_1(bpf_skb_cgroup_classid,const struct sk_buff *,skb)3087 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3088 {
3089 struct sock *sk = skb_to_full_sk(skb);
3090
3091 if (!sk || !sk_fullsock(sk))
3092 return 0;
3093
3094 return sock_cgroup_classid(&sk->sk_cgrp_data);
3095 }
3096
3097 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3098 .func = bpf_skb_cgroup_classid,
3099 .gpl_only = false,
3100 .ret_type = RET_INTEGER,
3101 .arg1_type = ARG_PTR_TO_CTX,
3102 };
3103 #endif
3104
BPF_CALL_1(bpf_get_cgroup_classid,const struct sk_buff *,skb)3105 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3106 {
3107 return task_get_classid(skb);
3108 }
3109
3110 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3111 .func = bpf_get_cgroup_classid,
3112 .gpl_only = false,
3113 .ret_type = RET_INTEGER,
3114 .arg1_type = ARG_PTR_TO_CTX,
3115 };
3116
BPF_CALL_1(bpf_get_route_realm,const struct sk_buff *,skb)3117 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3118 {
3119 return dst_tclassid(skb);
3120 }
3121
3122 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3123 .func = bpf_get_route_realm,
3124 .gpl_only = false,
3125 .ret_type = RET_INTEGER,
3126 .arg1_type = ARG_PTR_TO_CTX,
3127 };
3128
BPF_CALL_1(bpf_get_hash_recalc,struct sk_buff *,skb)3129 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3130 {
3131 /* If skb_clear_hash() was called due to mangling, we can
3132 * trigger SW recalculation here. Later access to hash
3133 * can then use the inline skb->hash via context directly
3134 * instead of calling this helper again.
3135 */
3136 return skb_get_hash(skb);
3137 }
3138
3139 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3140 .func = bpf_get_hash_recalc,
3141 .gpl_only = false,
3142 .ret_type = RET_INTEGER,
3143 .arg1_type = ARG_PTR_TO_CTX,
3144 };
3145
BPF_CALL_1(bpf_set_hash_invalid,struct sk_buff *,skb)3146 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3147 {
3148 /* After all direct packet write, this can be used once for
3149 * triggering a lazy recalc on next skb_get_hash() invocation.
3150 */
3151 skb_clear_hash(skb);
3152 return 0;
3153 }
3154
3155 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3156 .func = bpf_set_hash_invalid,
3157 .gpl_only = false,
3158 .ret_type = RET_INTEGER,
3159 .arg1_type = ARG_PTR_TO_CTX,
3160 };
3161
BPF_CALL_2(bpf_set_hash,struct sk_buff *,skb,u32,hash)3162 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3163 {
3164 /* Set user specified hash as L4(+), so that it gets returned
3165 * on skb_get_hash() call unless BPF prog later on triggers a
3166 * skb_clear_hash().
3167 */
3168 __skb_set_sw_hash(skb, hash, true);
3169 return 0;
3170 }
3171
3172 static const struct bpf_func_proto bpf_set_hash_proto = {
3173 .func = bpf_set_hash,
3174 .gpl_only = false,
3175 .ret_type = RET_INTEGER,
3176 .arg1_type = ARG_PTR_TO_CTX,
3177 .arg2_type = ARG_ANYTHING,
3178 };
3179
BPF_CALL_3(bpf_skb_vlan_push,struct sk_buff *,skb,__be16,vlan_proto,u16,vlan_tci)3180 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3181 u16, vlan_tci)
3182 {
3183 int ret;
3184
3185 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3186 vlan_proto != htons(ETH_P_8021AD)))
3187 vlan_proto = htons(ETH_P_8021Q);
3188
3189 bpf_push_mac_rcsum(skb);
3190 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3191 bpf_pull_mac_rcsum(skb);
3192 skb_reset_mac_len(skb);
3193
3194 bpf_compute_data_pointers(skb);
3195 return ret;
3196 }
3197
3198 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3199 .func = bpf_skb_vlan_push,
3200 .gpl_only = false,
3201 .ret_type = RET_INTEGER,
3202 .arg1_type = ARG_PTR_TO_CTX,
3203 .arg2_type = ARG_ANYTHING,
3204 .arg3_type = ARG_ANYTHING,
3205 };
3206
BPF_CALL_1(bpf_skb_vlan_pop,struct sk_buff *,skb)3207 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3208 {
3209 int ret;
3210
3211 bpf_push_mac_rcsum(skb);
3212 ret = skb_vlan_pop(skb);
3213 bpf_pull_mac_rcsum(skb);
3214
3215 bpf_compute_data_pointers(skb);
3216 return ret;
3217 }
3218
3219 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3220 .func = bpf_skb_vlan_pop,
3221 .gpl_only = false,
3222 .ret_type = RET_INTEGER,
3223 .arg1_type = ARG_PTR_TO_CTX,
3224 };
3225
bpf_skb_generic_push(struct sk_buff * skb,u32 off,u32 len)3226 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3227 {
3228 /* Caller already did skb_cow() with len as headroom,
3229 * so no need to do it here.
3230 */
3231 skb_push(skb, len);
3232 memmove(skb->data, skb->data + len, off);
3233 memset(skb->data + off, 0, len);
3234
3235 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3236 * needed here as it does not change the skb->csum
3237 * result for checksum complete when summing over
3238 * zeroed blocks.
3239 */
3240 return 0;
3241 }
3242
bpf_skb_generic_pop(struct sk_buff * skb,u32 off,u32 len)3243 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3244 {
3245 void *old_data;
3246
3247 /* skb_ensure_writable() is not needed here, as we're
3248 * already working on an uncloned skb.
3249 */
3250 if (unlikely(!pskb_may_pull(skb, off + len)))
3251 return -ENOMEM;
3252
3253 old_data = skb->data;
3254 __skb_pull(skb, len);
3255 skb_postpull_rcsum(skb, old_data + off, len);
3256 memmove(skb->data, old_data, off);
3257
3258 return 0;
3259 }
3260
bpf_skb_net_hdr_push(struct sk_buff * skb,u32 off,u32 len)3261 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3262 {
3263 bool trans_same = skb->transport_header == skb->network_header;
3264 int ret;
3265
3266 /* There's no need for __skb_push()/__skb_pull() pair to
3267 * get to the start of the mac header as we're guaranteed
3268 * to always start from here under eBPF.
3269 */
3270 ret = bpf_skb_generic_push(skb, off, len);
3271 if (likely(!ret)) {
3272 skb->mac_header -= len;
3273 skb->network_header -= len;
3274 if (trans_same)
3275 skb->transport_header = skb->network_header;
3276 }
3277
3278 return ret;
3279 }
3280
bpf_skb_net_hdr_pop(struct sk_buff * skb,u32 off,u32 len)3281 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3282 {
3283 bool trans_same = skb->transport_header == skb->network_header;
3284 int ret;
3285
3286 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3287 ret = bpf_skb_generic_pop(skb, off, len);
3288 if (likely(!ret)) {
3289 skb->mac_header += len;
3290 skb->network_header += len;
3291 if (trans_same)
3292 skb->transport_header = skb->network_header;
3293 }
3294
3295 return ret;
3296 }
3297
bpf_skb_proto_4_to_6(struct sk_buff * skb)3298 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3299 {
3300 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3301 u32 off = skb_mac_header_len(skb);
3302 int ret;
3303
3304 ret = skb_cow(skb, len_diff);
3305 if (unlikely(ret < 0))
3306 return ret;
3307
3308 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3309 if (unlikely(ret < 0))
3310 return ret;
3311
3312 if (skb_is_gso(skb)) {
3313 struct skb_shared_info *shinfo = skb_shinfo(skb);
3314
3315 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3316 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3317 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3318 shinfo->gso_type |= SKB_GSO_TCPV6;
3319 }
3320 }
3321
3322 skb->protocol = htons(ETH_P_IPV6);
3323 skb_clear_hash(skb);
3324
3325 return 0;
3326 }
3327
bpf_skb_proto_6_to_4(struct sk_buff * skb)3328 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3329 {
3330 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3331 u32 off = skb_mac_header_len(skb);
3332 int ret;
3333
3334 ret = skb_unclone(skb, GFP_ATOMIC);
3335 if (unlikely(ret < 0))
3336 return ret;
3337
3338 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3339 if (unlikely(ret < 0))
3340 return ret;
3341
3342 if (skb_is_gso(skb)) {
3343 struct skb_shared_info *shinfo = skb_shinfo(skb);
3344
3345 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3346 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3347 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3348 shinfo->gso_type |= SKB_GSO_TCPV4;
3349 }
3350 }
3351
3352 skb->protocol = htons(ETH_P_IP);
3353 skb_clear_hash(skb);
3354
3355 return 0;
3356 }
3357
bpf_skb_proto_xlat(struct sk_buff * skb,__be16 to_proto)3358 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3359 {
3360 __be16 from_proto = skb->protocol;
3361
3362 if (from_proto == htons(ETH_P_IP) &&
3363 to_proto == htons(ETH_P_IPV6))
3364 return bpf_skb_proto_4_to_6(skb);
3365
3366 if (from_proto == htons(ETH_P_IPV6) &&
3367 to_proto == htons(ETH_P_IP))
3368 return bpf_skb_proto_6_to_4(skb);
3369
3370 return -ENOTSUPP;
3371 }
3372
BPF_CALL_3(bpf_skb_change_proto,struct sk_buff *,skb,__be16,proto,u64,flags)3373 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3374 u64, flags)
3375 {
3376 int ret;
3377
3378 if (unlikely(flags))
3379 return -EINVAL;
3380
3381 /* General idea is that this helper does the basic groundwork
3382 * needed for changing the protocol, and eBPF program fills the
3383 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3384 * and other helpers, rather than passing a raw buffer here.
3385 *
3386 * The rationale is to keep this minimal and without a need to
3387 * deal with raw packet data. F.e. even if we would pass buffers
3388 * here, the program still needs to call the bpf_lX_csum_replace()
3389 * helpers anyway. Plus, this way we keep also separation of
3390 * concerns, since f.e. bpf_skb_store_bytes() should only take
3391 * care of stores.
3392 *
3393 * Currently, additional options and extension header space are
3394 * not supported, but flags register is reserved so we can adapt
3395 * that. For offloads, we mark packet as dodgy, so that headers
3396 * need to be verified first.
3397 */
3398 ret = bpf_skb_proto_xlat(skb, proto);
3399 bpf_compute_data_pointers(skb);
3400 return ret;
3401 }
3402
3403 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3404 .func = bpf_skb_change_proto,
3405 .gpl_only = false,
3406 .ret_type = RET_INTEGER,
3407 .arg1_type = ARG_PTR_TO_CTX,
3408 .arg2_type = ARG_ANYTHING,
3409 .arg3_type = ARG_ANYTHING,
3410 };
3411
BPF_CALL_2(bpf_skb_change_type,struct sk_buff *,skb,u32,pkt_type)3412 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3413 {
3414 /* We only allow a restricted subset to be changed for now. */
3415 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3416 !skb_pkt_type_ok(pkt_type)))
3417 return -EINVAL;
3418
3419 skb->pkt_type = pkt_type;
3420 return 0;
3421 }
3422
3423 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3424 .func = bpf_skb_change_type,
3425 .gpl_only = false,
3426 .ret_type = RET_INTEGER,
3427 .arg1_type = ARG_PTR_TO_CTX,
3428 .arg2_type = ARG_ANYTHING,
3429 };
3430
bpf_skb_net_base_len(const struct sk_buff * skb)3431 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3432 {
3433 switch (skb->protocol) {
3434 case htons(ETH_P_IP):
3435 return sizeof(struct iphdr);
3436 case htons(ETH_P_IPV6):
3437 return sizeof(struct ipv6hdr);
3438 default:
3439 return ~0U;
3440 }
3441 }
3442
3443 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3444 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3445
3446 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3447 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3448
3449 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3450 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3451 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3452 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3453 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3454 BPF_F_ADJ_ROOM_ENCAP_L2( \
3455 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3456 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3457
bpf_skb_net_grow(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3458 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3459 u64 flags)
3460 {
3461 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3462 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3463 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3464 unsigned int gso_type = SKB_GSO_DODGY;
3465 int ret;
3466
3467 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3468 /* udp gso_size delineates datagrams, only allow if fixed */
3469 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3470 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3471 return -ENOTSUPP;
3472 }
3473
3474 ret = skb_cow_head(skb, len_diff);
3475 if (unlikely(ret < 0))
3476 return ret;
3477
3478 if (encap) {
3479 if (skb->protocol != htons(ETH_P_IP) &&
3480 skb->protocol != htons(ETH_P_IPV6))
3481 return -ENOTSUPP;
3482
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3484 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3485 return -EINVAL;
3486
3487 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3488 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3489 return -EINVAL;
3490
3491 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3492 inner_mac_len < ETH_HLEN)
3493 return -EINVAL;
3494
3495 if (skb->encapsulation)
3496 return -EALREADY;
3497
3498 mac_len = skb->network_header - skb->mac_header;
3499 inner_net = skb->network_header;
3500 if (inner_mac_len > len_diff)
3501 return -EINVAL;
3502 inner_trans = skb->transport_header;
3503 }
3504
3505 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3506 if (unlikely(ret < 0))
3507 return ret;
3508
3509 if (encap) {
3510 skb->inner_mac_header = inner_net - inner_mac_len;
3511 skb->inner_network_header = inner_net;
3512 skb->inner_transport_header = inner_trans;
3513
3514 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3515 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3516 else
3517 skb_set_inner_protocol(skb, skb->protocol);
3518
3519 skb->encapsulation = 1;
3520 skb_set_network_header(skb, mac_len);
3521
3522 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3523 gso_type |= SKB_GSO_UDP_TUNNEL;
3524 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3525 gso_type |= SKB_GSO_GRE;
3526 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3527 gso_type |= SKB_GSO_IPXIP6;
3528 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3529 gso_type |= SKB_GSO_IPXIP4;
3530
3531 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3532 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3533 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3534 sizeof(struct ipv6hdr) :
3535 sizeof(struct iphdr);
3536
3537 skb_set_transport_header(skb, mac_len + nh_len);
3538 }
3539
3540 /* Match skb->protocol to new outer l3 protocol */
3541 if (skb->protocol == htons(ETH_P_IP) &&
3542 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3543 skb->protocol = htons(ETH_P_IPV6);
3544 else if (skb->protocol == htons(ETH_P_IPV6) &&
3545 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3546 skb->protocol = htons(ETH_P_IP);
3547 }
3548
3549 if (skb_is_gso(skb)) {
3550 struct skb_shared_info *shinfo = skb_shinfo(skb);
3551
3552 /* Header must be checked, and gso_segs recomputed. */
3553 shinfo->gso_type |= gso_type;
3554 shinfo->gso_segs = 0;
3555
3556 /* Due to header growth, MSS needs to be downgraded.
3557 * There is a BUG_ON() when segmenting the frag_list with
3558 * head_frag true, so linearize the skb after downgrading
3559 * the MSS.
3560 */
3561 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) {
3562 skb_decrease_gso_size(shinfo, len_diff);
3563 if (shinfo->frag_list)
3564 return skb_linearize(skb);
3565 }
3566 }
3567
3568 return 0;
3569 }
3570
bpf_skb_net_shrink(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3571 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3572 u64 flags)
3573 {
3574 int ret;
3575
3576 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3577 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3578 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3579 return -EINVAL;
3580
3581 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3582 /* udp gso_size delineates datagrams, only allow if fixed */
3583 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3584 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3585 return -ENOTSUPP;
3586 }
3587
3588 ret = skb_unclone(skb, GFP_ATOMIC);
3589 if (unlikely(ret < 0))
3590 return ret;
3591
3592 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3593 if (unlikely(ret < 0))
3594 return ret;
3595
3596 /* Match skb->protocol to new outer l3 protocol */
3597 if (skb->protocol == htons(ETH_P_IP) &&
3598 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3599 skb->protocol = htons(ETH_P_IPV6);
3600 else if (skb->protocol == htons(ETH_P_IPV6) &&
3601 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3602 skb->protocol = htons(ETH_P_IP);
3603
3604 if (skb_is_gso(skb)) {
3605 struct skb_shared_info *shinfo = skb_shinfo(skb);
3606
3607 /* Due to header shrink, MSS can be upgraded. */
3608 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3609 skb_increase_gso_size(shinfo, len_diff);
3610
3611 /* Header must be checked, and gso_segs recomputed. */
3612 shinfo->gso_type |= SKB_GSO_DODGY;
3613 shinfo->gso_segs = 0;
3614 }
3615
3616 return 0;
3617 }
3618
3619 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3620
BPF_CALL_4(sk_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3621 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3622 u32, mode, u64, flags)
3623 {
3624 u32 len_diff_abs = abs(len_diff);
3625 bool shrink = len_diff < 0;
3626 int ret = 0;
3627
3628 if (unlikely(flags || mode))
3629 return -EINVAL;
3630 if (unlikely(len_diff_abs > 0xfffU))
3631 return -EFAULT;
3632
3633 if (!shrink) {
3634 ret = skb_cow(skb, len_diff);
3635 if (unlikely(ret < 0))
3636 return ret;
3637 __skb_push(skb, len_diff_abs);
3638 memset(skb->data, 0, len_diff_abs);
3639 } else {
3640 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3641 return -ENOMEM;
3642 __skb_pull(skb, len_diff_abs);
3643 }
3644 if (tls_sw_has_ctx_rx(skb->sk)) {
3645 struct strp_msg *rxm = strp_msg(skb);
3646
3647 rxm->full_len += len_diff;
3648 }
3649 return ret;
3650 }
3651
3652 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3653 .func = sk_skb_adjust_room,
3654 .gpl_only = false,
3655 .ret_type = RET_INTEGER,
3656 .arg1_type = ARG_PTR_TO_CTX,
3657 .arg2_type = ARG_ANYTHING,
3658 .arg3_type = ARG_ANYTHING,
3659 .arg4_type = ARG_ANYTHING,
3660 };
3661
BPF_CALL_4(bpf_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3662 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3663 u32, mode, u64, flags)
3664 {
3665 u32 len_cur, len_diff_abs = abs(len_diff);
3666 u32 len_min = bpf_skb_net_base_len(skb);
3667 u32 len_max = BPF_SKB_MAX_LEN;
3668 __be16 proto = skb->protocol;
3669 bool shrink = len_diff < 0;
3670 u32 off;
3671 int ret;
3672
3673 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3674 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3675 return -EINVAL;
3676 if (unlikely(len_diff_abs > 0xfffU))
3677 return -EFAULT;
3678 if (unlikely(proto != htons(ETH_P_IP) &&
3679 proto != htons(ETH_P_IPV6)))
3680 return -ENOTSUPP;
3681
3682 off = skb_mac_header_len(skb);
3683 switch (mode) {
3684 case BPF_ADJ_ROOM_NET:
3685 off += bpf_skb_net_base_len(skb);
3686 break;
3687 case BPF_ADJ_ROOM_MAC:
3688 break;
3689 default:
3690 return -ENOTSUPP;
3691 }
3692
3693 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3694 if (!shrink)
3695 return -EINVAL;
3696
3697 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3698 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3699 len_min = sizeof(struct iphdr);
3700 break;
3701 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3702 len_min = sizeof(struct ipv6hdr);
3703 break;
3704 default:
3705 return -EINVAL;
3706 }
3707 }
3708
3709 len_cur = skb->len - skb_network_offset(skb);
3710 if ((shrink && (len_diff_abs >= len_cur ||
3711 len_cur - len_diff_abs < len_min)) ||
3712 (!shrink && (skb->len + len_diff_abs > len_max &&
3713 !skb_is_gso(skb))))
3714 return -ENOTSUPP;
3715
3716 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3717 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3718 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3719 __skb_reset_checksum_unnecessary(skb);
3720
3721 bpf_compute_data_pointers(skb);
3722 return ret;
3723 }
3724
3725 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3726 .func = bpf_skb_adjust_room,
3727 .gpl_only = false,
3728 .ret_type = RET_INTEGER,
3729 .arg1_type = ARG_PTR_TO_CTX,
3730 .arg2_type = ARG_ANYTHING,
3731 .arg3_type = ARG_ANYTHING,
3732 .arg4_type = ARG_ANYTHING,
3733 };
3734
__bpf_skb_min_len(const struct sk_buff * skb)3735 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3736 {
3737 int offset = skb_network_offset(skb);
3738 u32 min_len = 0;
3739
3740 if (offset > 0)
3741 min_len = offset;
3742 if (skb_transport_header_was_set(skb)) {
3743 offset = skb_transport_offset(skb);
3744 if (offset > 0)
3745 min_len = offset;
3746 }
3747 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3748 offset = skb_checksum_start_offset(skb) +
3749 skb->csum_offset + sizeof(__sum16);
3750 if (offset > 0)
3751 min_len = offset;
3752 }
3753 return min_len;
3754 }
3755
bpf_skb_grow_rcsum(struct sk_buff * skb,unsigned int new_len)3756 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3757 {
3758 unsigned int old_len = skb->len;
3759 int ret;
3760
3761 ret = __skb_grow_rcsum(skb, new_len);
3762 if (!ret)
3763 memset(skb->data + old_len, 0, new_len - old_len);
3764 return ret;
3765 }
3766
bpf_skb_trim_rcsum(struct sk_buff * skb,unsigned int new_len)3767 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3768 {
3769 return __skb_trim_rcsum(skb, new_len);
3770 }
3771
__bpf_skb_change_tail(struct sk_buff * skb,u32 new_len,u64 flags)3772 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3773 u64 flags)
3774 {
3775 u32 max_len = BPF_SKB_MAX_LEN;
3776 u32 min_len = __bpf_skb_min_len(skb);
3777 int ret;
3778
3779 if (unlikely(flags || new_len > max_len || new_len < min_len))
3780 return -EINVAL;
3781 if (skb->encapsulation)
3782 return -ENOTSUPP;
3783
3784 /* The basic idea of this helper is that it's performing the
3785 * needed work to either grow or trim an skb, and eBPF program
3786 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3787 * bpf_lX_csum_replace() and others rather than passing a raw
3788 * buffer here. This one is a slow path helper and intended
3789 * for replies with control messages.
3790 *
3791 * Like in bpf_skb_change_proto(), we want to keep this rather
3792 * minimal and without protocol specifics so that we are able
3793 * to separate concerns as in bpf_skb_store_bytes() should only
3794 * be the one responsible for writing buffers.
3795 *
3796 * It's really expected to be a slow path operation here for
3797 * control message replies, so we're implicitly linearizing,
3798 * uncloning and drop offloads from the skb by this.
3799 */
3800 ret = __bpf_try_make_writable(skb, skb->len);
3801 if (!ret) {
3802 if (new_len > skb->len)
3803 ret = bpf_skb_grow_rcsum(skb, new_len);
3804 else if (new_len < skb->len)
3805 ret = bpf_skb_trim_rcsum(skb, new_len);
3806 if (!ret && skb_is_gso(skb))
3807 skb_gso_reset(skb);
3808 }
3809 return ret;
3810 }
3811
BPF_CALL_3(bpf_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3812 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3813 u64, flags)
3814 {
3815 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3816
3817 bpf_compute_data_pointers(skb);
3818 return ret;
3819 }
3820
3821 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3822 .func = bpf_skb_change_tail,
3823 .gpl_only = false,
3824 .ret_type = RET_INTEGER,
3825 .arg1_type = ARG_PTR_TO_CTX,
3826 .arg2_type = ARG_ANYTHING,
3827 .arg3_type = ARG_ANYTHING,
3828 };
3829
BPF_CALL_3(sk_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3830 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3831 u64, flags)
3832 {
3833 return __bpf_skb_change_tail(skb, new_len, flags);
3834 }
3835
3836 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3837 .func = sk_skb_change_tail,
3838 .gpl_only = false,
3839 .ret_type = RET_INTEGER,
3840 .arg1_type = ARG_PTR_TO_CTX,
3841 .arg2_type = ARG_ANYTHING,
3842 .arg3_type = ARG_ANYTHING,
3843 };
3844
__bpf_skb_change_head(struct sk_buff * skb,u32 head_room,u64 flags)3845 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3846 u64 flags)
3847 {
3848 u32 max_len = BPF_SKB_MAX_LEN;
3849 u32 new_len = skb->len + head_room;
3850 int ret;
3851
3852 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3853 new_len < skb->len))
3854 return -EINVAL;
3855
3856 ret = skb_cow(skb, head_room);
3857 if (likely(!ret)) {
3858 /* Idea for this helper is that we currently only
3859 * allow to expand on mac header. This means that
3860 * skb->protocol network header, etc, stay as is.
3861 * Compared to bpf_skb_change_tail(), we're more
3862 * flexible due to not needing to linearize or
3863 * reset GSO. Intention for this helper is to be
3864 * used by an L3 skb that needs to push mac header
3865 * for redirection into L2 device.
3866 */
3867 __skb_push(skb, head_room);
3868 memset(skb->data, 0, head_room);
3869 skb_reset_mac_header(skb);
3870 skb_reset_mac_len(skb);
3871 }
3872
3873 return ret;
3874 }
3875
BPF_CALL_3(bpf_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3876 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3877 u64, flags)
3878 {
3879 int ret = __bpf_skb_change_head(skb, head_room, flags);
3880
3881 bpf_compute_data_pointers(skb);
3882 return ret;
3883 }
3884
3885 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3886 .func = bpf_skb_change_head,
3887 .gpl_only = false,
3888 .ret_type = RET_INTEGER,
3889 .arg1_type = ARG_PTR_TO_CTX,
3890 .arg2_type = ARG_ANYTHING,
3891 .arg3_type = ARG_ANYTHING,
3892 };
3893
BPF_CALL_3(sk_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3894 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3895 u64, flags)
3896 {
3897 return __bpf_skb_change_head(skb, head_room, flags);
3898 }
3899
3900 static const struct bpf_func_proto sk_skb_change_head_proto = {
3901 .func = sk_skb_change_head,
3902 .gpl_only = false,
3903 .ret_type = RET_INTEGER,
3904 .arg1_type = ARG_PTR_TO_CTX,
3905 .arg2_type = ARG_ANYTHING,
3906 .arg3_type = ARG_ANYTHING,
3907 };
3908
BPF_CALL_1(bpf_xdp_get_buff_len,struct xdp_buff *,xdp)3909 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3910 {
3911 return xdp_get_buff_len(xdp);
3912 }
3913
3914 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3915 .func = bpf_xdp_get_buff_len,
3916 .gpl_only = false,
3917 .ret_type = RET_INTEGER,
3918 .arg1_type = ARG_PTR_TO_CTX,
3919 };
3920
3921 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3922
3923 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3924 .func = bpf_xdp_get_buff_len,
3925 .gpl_only = false,
3926 .arg1_type = ARG_PTR_TO_BTF_ID,
3927 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3928 };
3929
xdp_get_metalen(const struct xdp_buff * xdp)3930 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3931 {
3932 return xdp_data_meta_unsupported(xdp) ? 0 :
3933 xdp->data - xdp->data_meta;
3934 }
3935
BPF_CALL_2(bpf_xdp_adjust_head,struct xdp_buff *,xdp,int,offset)3936 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3937 {
3938 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3939 unsigned long metalen = xdp_get_metalen(xdp);
3940 void *data_start = xdp_frame_end + metalen;
3941 void *data = xdp->data + offset;
3942
3943 if (unlikely(data < data_start ||
3944 data > xdp->data_end - ETH_HLEN))
3945 return -EINVAL;
3946
3947 if (metalen)
3948 memmove(xdp->data_meta + offset,
3949 xdp->data_meta, metalen);
3950 xdp->data_meta += offset;
3951 xdp->data = data;
3952
3953 return 0;
3954 }
3955
3956 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3957 .func = bpf_xdp_adjust_head,
3958 .gpl_only = false,
3959 .ret_type = RET_INTEGER,
3960 .arg1_type = ARG_PTR_TO_CTX,
3961 .arg2_type = ARG_ANYTHING,
3962 };
3963
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)3964 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3965 void *buf, unsigned long len, bool flush)
3966 {
3967 unsigned long ptr_len, ptr_off = 0;
3968 skb_frag_t *next_frag, *end_frag;
3969 struct skb_shared_info *sinfo;
3970 void *src, *dst;
3971 u8 *ptr_buf;
3972
3973 if (likely(xdp->data_end - xdp->data >= off + len)) {
3974 src = flush ? buf : xdp->data + off;
3975 dst = flush ? xdp->data + off : buf;
3976 memcpy(dst, src, len);
3977 return;
3978 }
3979
3980 sinfo = xdp_get_shared_info_from_buff(xdp);
3981 end_frag = &sinfo->frags[sinfo->nr_frags];
3982 next_frag = &sinfo->frags[0];
3983
3984 ptr_len = xdp->data_end - xdp->data;
3985 ptr_buf = xdp->data;
3986
3987 while (true) {
3988 if (off < ptr_off + ptr_len) {
3989 unsigned long copy_off = off - ptr_off;
3990 unsigned long copy_len = min(len, ptr_len - copy_off);
3991
3992 src = flush ? buf : ptr_buf + copy_off;
3993 dst = flush ? ptr_buf + copy_off : buf;
3994 memcpy(dst, src, copy_len);
3995
3996 off += copy_len;
3997 len -= copy_len;
3998 buf += copy_len;
3999 }
4000
4001 if (!len || next_frag == end_frag)
4002 break;
4003
4004 ptr_off += ptr_len;
4005 ptr_buf = skb_frag_address(next_frag);
4006 ptr_len = skb_frag_size(next_frag);
4007 next_frag++;
4008 }
4009 }
4010
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)4011 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
4012 {
4013 u32 size = xdp->data_end - xdp->data;
4014 struct skb_shared_info *sinfo;
4015 void *addr = xdp->data;
4016 int i;
4017
4018 if (unlikely(offset > 0xffff || len > 0xffff))
4019 return ERR_PTR(-EFAULT);
4020
4021 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
4022 return ERR_PTR(-EINVAL);
4023
4024 if (likely(offset < size)) /* linear area */
4025 goto out;
4026
4027 sinfo = xdp_get_shared_info_from_buff(xdp);
4028 offset -= size;
4029 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
4030 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
4031
4032 if (offset < frag_size) {
4033 addr = skb_frag_address(&sinfo->frags[i]);
4034 size = frag_size;
4035 break;
4036 }
4037 offset -= frag_size;
4038 }
4039 out:
4040 return offset + len <= size ? addr + offset : NULL;
4041 }
4042
BPF_CALL_4(bpf_xdp_load_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4043 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4044 void *, buf, u32, len)
4045 {
4046 void *ptr;
4047
4048 ptr = bpf_xdp_pointer(xdp, offset, len);
4049 if (IS_ERR(ptr))
4050 return PTR_ERR(ptr);
4051
4052 if (!ptr)
4053 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4054 else
4055 memcpy(buf, ptr, len);
4056
4057 return 0;
4058 }
4059
4060 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4061 .func = bpf_xdp_load_bytes,
4062 .gpl_only = false,
4063 .ret_type = RET_INTEGER,
4064 .arg1_type = ARG_PTR_TO_CTX,
4065 .arg2_type = ARG_ANYTHING,
4066 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4067 .arg4_type = ARG_CONST_SIZE,
4068 };
4069
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4070 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4071 {
4072 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4073 }
4074
BPF_CALL_4(bpf_xdp_store_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4075 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4076 void *, buf, u32, len)
4077 {
4078 void *ptr;
4079
4080 ptr = bpf_xdp_pointer(xdp, offset, len);
4081 if (IS_ERR(ptr))
4082 return PTR_ERR(ptr);
4083
4084 if (!ptr)
4085 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4086 else
4087 memcpy(ptr, buf, len);
4088
4089 return 0;
4090 }
4091
4092 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4093 .func = bpf_xdp_store_bytes,
4094 .gpl_only = false,
4095 .ret_type = RET_INTEGER,
4096 .arg1_type = ARG_PTR_TO_CTX,
4097 .arg2_type = ARG_ANYTHING,
4098 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4099 .arg4_type = ARG_CONST_SIZE,
4100 };
4101
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4102 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4103 {
4104 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4105 }
4106
bpf_xdp_frags_increase_tail(struct xdp_buff * xdp,int offset)4107 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4108 {
4109 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4110 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4111 struct xdp_rxq_info *rxq = xdp->rxq;
4112 unsigned int tailroom;
4113
4114 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4115 return -EOPNOTSUPP;
4116
4117 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4118 if (unlikely(offset > tailroom))
4119 return -EINVAL;
4120
4121 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4122 skb_frag_size_add(frag, offset);
4123 sinfo->xdp_frags_size += offset;
4124 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
4125 xsk_buff_get_tail(xdp)->data_end += offset;
4126
4127 return 0;
4128 }
4129
bpf_xdp_shrink_data_zc(struct xdp_buff * xdp,int shrink,enum xdp_mem_type mem_type,bool release)4130 static void bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink,
4131 enum xdp_mem_type mem_type, bool release)
4132 {
4133 struct xdp_buff *zc_frag = xsk_buff_get_tail(xdp);
4134
4135 if (release) {
4136 xsk_buff_del_tail(zc_frag);
4137 __xdp_return(0, mem_type, false, zc_frag);
4138 } else {
4139 zc_frag->data_end -= shrink;
4140 }
4141 }
4142
bpf_xdp_shrink_data(struct xdp_buff * xdp,skb_frag_t * frag,int shrink)4143 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag,
4144 int shrink)
4145 {
4146 enum xdp_mem_type mem_type = xdp->rxq->mem.type;
4147 bool release = skb_frag_size(frag) == shrink;
4148
4149 if (mem_type == MEM_TYPE_XSK_BUFF_POOL) {
4150 bpf_xdp_shrink_data_zc(xdp, shrink, mem_type, release);
4151 goto out;
4152 }
4153
4154 if (release)
4155 __xdp_return(skb_frag_netmem(frag), mem_type, false, NULL);
4156
4157 out:
4158 return release;
4159 }
4160
bpf_xdp_frags_shrink_tail(struct xdp_buff * xdp,int offset)4161 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4162 {
4163 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4164 int i, n_frags_free = 0, len_free = 0;
4165
4166 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4167 return -EINVAL;
4168
4169 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4170 skb_frag_t *frag = &sinfo->frags[i];
4171 int shrink = min_t(int, offset, skb_frag_size(frag));
4172
4173 len_free += shrink;
4174 offset -= shrink;
4175 if (bpf_xdp_shrink_data(xdp, frag, shrink)) {
4176 n_frags_free++;
4177 } else {
4178 skb_frag_size_sub(frag, shrink);
4179 break;
4180 }
4181 }
4182 sinfo->nr_frags -= n_frags_free;
4183 sinfo->xdp_frags_size -= len_free;
4184
4185 if (unlikely(!sinfo->nr_frags)) {
4186 xdp_buff_clear_frags_flag(xdp);
4187 xdp->data_end -= offset;
4188 }
4189
4190 return 0;
4191 }
4192
BPF_CALL_2(bpf_xdp_adjust_tail,struct xdp_buff *,xdp,int,offset)4193 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4194 {
4195 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4196 void *data_end = xdp->data_end + offset;
4197
4198 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4199 if (offset < 0)
4200 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4201
4202 return bpf_xdp_frags_increase_tail(xdp, offset);
4203 }
4204
4205 /* Notice that xdp_data_hard_end have reserved some tailroom */
4206 if (unlikely(data_end > data_hard_end))
4207 return -EINVAL;
4208
4209 if (unlikely(data_end < xdp->data + ETH_HLEN))
4210 return -EINVAL;
4211
4212 /* Clear memory area on grow, can contain uninit kernel memory */
4213 if (offset > 0)
4214 memset(xdp->data_end, 0, offset);
4215
4216 xdp->data_end = data_end;
4217
4218 return 0;
4219 }
4220
4221 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4222 .func = bpf_xdp_adjust_tail,
4223 .gpl_only = false,
4224 .ret_type = RET_INTEGER,
4225 .arg1_type = ARG_PTR_TO_CTX,
4226 .arg2_type = ARG_ANYTHING,
4227 };
4228
BPF_CALL_2(bpf_xdp_adjust_meta,struct xdp_buff *,xdp,int,offset)4229 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4230 {
4231 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4232 void *meta = xdp->data_meta + offset;
4233 unsigned long metalen = xdp->data - meta;
4234
4235 if (xdp_data_meta_unsupported(xdp))
4236 return -ENOTSUPP;
4237 if (unlikely(meta < xdp_frame_end ||
4238 meta > xdp->data))
4239 return -EINVAL;
4240 if (unlikely(xdp_metalen_invalid(metalen)))
4241 return -EACCES;
4242
4243 xdp->data_meta = meta;
4244
4245 return 0;
4246 }
4247
4248 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4249 .func = bpf_xdp_adjust_meta,
4250 .gpl_only = false,
4251 .ret_type = RET_INTEGER,
4252 .arg1_type = ARG_PTR_TO_CTX,
4253 .arg2_type = ARG_ANYTHING,
4254 };
4255
4256 /**
4257 * DOC: xdp redirect
4258 *
4259 * XDP_REDIRECT works by a three-step process, implemented in the functions
4260 * below:
4261 *
4262 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4263 * of the redirect and store it (along with some other metadata) in a per-CPU
4264 * struct bpf_redirect_info.
4265 *
4266 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4267 * call xdp_do_redirect() which will use the information in struct
4268 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4269 * bulk queue structure.
4270 *
4271 * 3. Before exiting its NAPI poll loop, the driver will call
4272 * xdp_do_flush(), which will flush all the different bulk queues,
4273 * thus completing the redirect. Note that xdp_do_flush() must be
4274 * called before napi_complete_done() in the driver, as the
4275 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4276 * through to the xdp_do_flush() call for RCU protection of all
4277 * in-kernel data structures.
4278 */
4279 /*
4280 * Pointers to the map entries will be kept around for this whole sequence of
4281 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4282 * the core code; instead, the RCU protection relies on everything happening
4283 * inside a single NAPI poll sequence, which means it's between a pair of calls
4284 * to local_bh_disable()/local_bh_enable().
4285 *
4286 * The map entries are marked as __rcu and the map code makes sure to
4287 * dereference those pointers with rcu_dereference_check() in a way that works
4288 * for both sections that to hold an rcu_read_lock() and sections that are
4289 * called from NAPI without a separate rcu_read_lock(). The code below does not
4290 * use RCU annotations, but relies on those in the map code.
4291 */
xdp_do_flush(void)4292 void xdp_do_flush(void)
4293 {
4294 struct list_head *lh_map, *lh_dev, *lh_xsk;
4295
4296 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4297 if (lh_dev)
4298 __dev_flush(lh_dev);
4299 if (lh_map)
4300 __cpu_map_flush(lh_map);
4301 if (lh_xsk)
4302 __xsk_map_flush(lh_xsk);
4303 }
4304 EXPORT_SYMBOL_GPL(xdp_do_flush);
4305
4306 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
xdp_do_check_flushed(struct napi_struct * napi)4307 void xdp_do_check_flushed(struct napi_struct *napi)
4308 {
4309 struct list_head *lh_map, *lh_dev, *lh_xsk;
4310 bool missed = false;
4311
4312 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4313 if (lh_dev) {
4314 __dev_flush(lh_dev);
4315 missed = true;
4316 }
4317 if (lh_map) {
4318 __cpu_map_flush(lh_map);
4319 missed = true;
4320 }
4321 if (lh_xsk) {
4322 __xsk_map_flush(lh_xsk);
4323 missed = true;
4324 }
4325
4326 WARN_ONCE(missed, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
4327 napi->poll);
4328 }
4329 #endif
4330
4331 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4332 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4333
xdp_master_redirect(struct xdp_buff * xdp)4334 u32 xdp_master_redirect(struct xdp_buff *xdp)
4335 {
4336 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4337 struct net_device *master, *slave;
4338
4339 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4340 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4341 if (slave && slave != xdp->rxq->dev) {
4342 /* The target device is different from the receiving device, so
4343 * redirect it to the new device.
4344 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4345 * drivers to unmap the packet from their rx ring.
4346 */
4347 ri->tgt_index = slave->ifindex;
4348 ri->map_id = INT_MAX;
4349 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4350 return XDP_REDIRECT;
4351 }
4352 return XDP_TX;
4353 }
4354 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4355
__xdp_do_redirect_xsk(struct bpf_redirect_info * ri,const struct net_device * dev,struct xdp_buff * xdp,const struct bpf_prog * xdp_prog)4356 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4357 const struct net_device *dev,
4358 struct xdp_buff *xdp,
4359 const struct bpf_prog *xdp_prog)
4360 {
4361 enum bpf_map_type map_type = ri->map_type;
4362 void *fwd = ri->tgt_value;
4363 u32 map_id = ri->map_id;
4364 int err;
4365
4366 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4367 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4368
4369 err = __xsk_map_redirect(fwd, xdp);
4370 if (unlikely(err))
4371 goto err;
4372
4373 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4374 return 0;
4375 err:
4376 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4377 return err;
4378 }
4379
4380 static __always_inline int
__xdp_do_redirect_frame(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_frame * xdpf,const struct bpf_prog * xdp_prog)4381 __xdp_do_redirect_frame(struct bpf_redirect_info *ri, struct net_device *dev,
4382 struct xdp_frame *xdpf,
4383 const struct bpf_prog *xdp_prog)
4384 {
4385 enum bpf_map_type map_type = ri->map_type;
4386 void *fwd = ri->tgt_value;
4387 u32 map_id = ri->map_id;
4388 u32 flags = ri->flags;
4389 struct bpf_map *map;
4390 int err;
4391
4392 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4393 ri->flags = 0;
4394 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4395
4396 if (unlikely(!xdpf)) {
4397 err = -EOVERFLOW;
4398 goto err;
4399 }
4400
4401 switch (map_type) {
4402 case BPF_MAP_TYPE_DEVMAP:
4403 fallthrough;
4404 case BPF_MAP_TYPE_DEVMAP_HASH:
4405 if (unlikely(flags & BPF_F_BROADCAST)) {
4406 map = READ_ONCE(ri->map);
4407
4408 /* The map pointer is cleared when the map is being torn
4409 * down by dev_map_free()
4410 */
4411 if (unlikely(!map)) {
4412 err = -ENOENT;
4413 break;
4414 }
4415
4416 WRITE_ONCE(ri->map, NULL);
4417 err = dev_map_enqueue_multi(xdpf, dev, map,
4418 flags & BPF_F_EXCLUDE_INGRESS);
4419 } else {
4420 err = dev_map_enqueue(fwd, xdpf, dev);
4421 }
4422 break;
4423 case BPF_MAP_TYPE_CPUMAP:
4424 err = cpu_map_enqueue(fwd, xdpf, dev);
4425 break;
4426 case BPF_MAP_TYPE_UNSPEC:
4427 if (map_id == INT_MAX) {
4428 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4429 if (unlikely(!fwd)) {
4430 err = -EINVAL;
4431 break;
4432 }
4433 err = dev_xdp_enqueue(fwd, xdpf, dev);
4434 break;
4435 }
4436 fallthrough;
4437 default:
4438 err = -EBADRQC;
4439 }
4440
4441 if (unlikely(err))
4442 goto err;
4443
4444 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4445 return 0;
4446 err:
4447 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4448 return err;
4449 }
4450
xdp_do_redirect(struct net_device * dev,struct xdp_buff * xdp,const struct bpf_prog * xdp_prog)4451 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4452 const struct bpf_prog *xdp_prog)
4453 {
4454 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4455 enum bpf_map_type map_type = ri->map_type;
4456
4457 if (map_type == BPF_MAP_TYPE_XSKMAP)
4458 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4459
4460 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4461 xdp_prog);
4462 }
4463 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4464
xdp_do_redirect_frame(struct net_device * dev,struct xdp_buff * xdp,struct xdp_frame * xdpf,const struct bpf_prog * xdp_prog)4465 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4466 struct xdp_frame *xdpf,
4467 const struct bpf_prog *xdp_prog)
4468 {
4469 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4470 enum bpf_map_type map_type = ri->map_type;
4471
4472 if (map_type == BPF_MAP_TYPE_XSKMAP)
4473 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4474
4475 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4476 }
4477 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4478
xdp_do_generic_redirect_map(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,const struct bpf_prog * xdp_prog,void * fwd,enum bpf_map_type map_type,u32 map_id,u32 flags)4479 static int xdp_do_generic_redirect_map(struct net_device *dev,
4480 struct sk_buff *skb,
4481 struct xdp_buff *xdp,
4482 const struct bpf_prog *xdp_prog,
4483 void *fwd, enum bpf_map_type map_type,
4484 u32 map_id, u32 flags)
4485 {
4486 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4487 struct bpf_map *map;
4488 int err;
4489
4490 switch (map_type) {
4491 case BPF_MAP_TYPE_DEVMAP:
4492 fallthrough;
4493 case BPF_MAP_TYPE_DEVMAP_HASH:
4494 if (unlikely(flags & BPF_F_BROADCAST)) {
4495 map = READ_ONCE(ri->map);
4496
4497 /* The map pointer is cleared when the map is being torn
4498 * down by dev_map_free()
4499 */
4500 if (unlikely(!map)) {
4501 err = -ENOENT;
4502 break;
4503 }
4504
4505 WRITE_ONCE(ri->map, NULL);
4506 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4507 flags & BPF_F_EXCLUDE_INGRESS);
4508 } else {
4509 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4510 }
4511 if (unlikely(err))
4512 goto err;
4513 break;
4514 case BPF_MAP_TYPE_XSKMAP:
4515 err = xsk_generic_rcv(fwd, xdp);
4516 if (err)
4517 goto err;
4518 consume_skb(skb);
4519 break;
4520 case BPF_MAP_TYPE_CPUMAP:
4521 err = cpu_map_generic_redirect(fwd, skb);
4522 if (unlikely(err))
4523 goto err;
4524 break;
4525 default:
4526 err = -EBADRQC;
4527 goto err;
4528 }
4529
4530 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4531 return 0;
4532 err:
4533 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4534 return err;
4535 }
4536
xdp_do_generic_redirect(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,const struct bpf_prog * xdp_prog)4537 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4538 struct xdp_buff *xdp,
4539 const struct bpf_prog *xdp_prog)
4540 {
4541 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4542 enum bpf_map_type map_type = ri->map_type;
4543 void *fwd = ri->tgt_value;
4544 u32 map_id = ri->map_id;
4545 u32 flags = ri->flags;
4546 int err;
4547
4548 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4549 ri->flags = 0;
4550 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4551
4552 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4553 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4554 if (unlikely(!fwd)) {
4555 err = -EINVAL;
4556 goto err;
4557 }
4558
4559 err = xdp_ok_fwd_dev(fwd, skb->len);
4560 if (unlikely(err))
4561 goto err;
4562
4563 skb->dev = fwd;
4564 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4565 generic_xdp_tx(skb, xdp_prog);
4566 return 0;
4567 }
4568
4569 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4570 err:
4571 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4572 return err;
4573 }
4574
BPF_CALL_2(bpf_xdp_redirect,u32,ifindex,u64,flags)4575 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4576 {
4577 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4578
4579 if (unlikely(flags))
4580 return XDP_ABORTED;
4581
4582 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4583 * by map_idr) is used for ifindex based XDP redirect.
4584 */
4585 ri->tgt_index = ifindex;
4586 ri->map_id = INT_MAX;
4587 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4588
4589 return XDP_REDIRECT;
4590 }
4591
4592 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4593 .func = bpf_xdp_redirect,
4594 .gpl_only = false,
4595 .ret_type = RET_INTEGER,
4596 .arg1_type = ARG_ANYTHING,
4597 .arg2_type = ARG_ANYTHING,
4598 };
4599
BPF_CALL_3(bpf_xdp_redirect_map,struct bpf_map *,map,u64,key,u64,flags)4600 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4601 u64, flags)
4602 {
4603 return map->ops->map_redirect(map, key, flags);
4604 }
4605
4606 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4607 .func = bpf_xdp_redirect_map,
4608 .gpl_only = false,
4609 .ret_type = RET_INTEGER,
4610 .arg1_type = ARG_CONST_MAP_PTR,
4611 .arg2_type = ARG_ANYTHING,
4612 .arg3_type = ARG_ANYTHING,
4613 };
4614
bpf_skb_copy(void * dst_buff,const void * skb,unsigned long off,unsigned long len)4615 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4616 unsigned long off, unsigned long len)
4617 {
4618 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4619
4620 if (unlikely(!ptr))
4621 return len;
4622 if (ptr != dst_buff)
4623 memcpy(dst_buff, ptr, len);
4624
4625 return 0;
4626 }
4627
BPF_CALL_5(bpf_skb_event_output,struct sk_buff *,skb,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4628 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4629 u64, flags, void *, meta, u64, meta_size)
4630 {
4631 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4632
4633 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4634 return -EINVAL;
4635 if (unlikely(!skb || skb_size > skb->len))
4636 return -EFAULT;
4637
4638 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4639 bpf_skb_copy);
4640 }
4641
4642 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4643 .func = bpf_skb_event_output,
4644 .gpl_only = true,
4645 .ret_type = RET_INTEGER,
4646 .arg1_type = ARG_PTR_TO_CTX,
4647 .arg2_type = ARG_CONST_MAP_PTR,
4648 .arg3_type = ARG_ANYTHING,
4649 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4650 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4651 };
4652
4653 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4654
4655 const struct bpf_func_proto bpf_skb_output_proto = {
4656 .func = bpf_skb_event_output,
4657 .gpl_only = true,
4658 .ret_type = RET_INTEGER,
4659 .arg1_type = ARG_PTR_TO_BTF_ID,
4660 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4661 .arg2_type = ARG_CONST_MAP_PTR,
4662 .arg3_type = ARG_ANYTHING,
4663 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4664 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4665 };
4666
bpf_tunnel_key_af(u64 flags)4667 static unsigned short bpf_tunnel_key_af(u64 flags)
4668 {
4669 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4670 }
4671
BPF_CALL_4(bpf_skb_get_tunnel_key,struct sk_buff *,skb,struct bpf_tunnel_key *,to,u32,size,u64,flags)4672 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4673 u32, size, u64, flags)
4674 {
4675 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4676 u8 compat[sizeof(struct bpf_tunnel_key)];
4677 void *to_orig = to;
4678 int err;
4679
4680 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4681 BPF_F_TUNINFO_FLAGS)))) {
4682 err = -EINVAL;
4683 goto err_clear;
4684 }
4685 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4686 err = -EPROTO;
4687 goto err_clear;
4688 }
4689 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4690 err = -EINVAL;
4691 switch (size) {
4692 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4693 case offsetof(struct bpf_tunnel_key, tunnel_label):
4694 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4695 goto set_compat;
4696 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4697 /* Fixup deprecated structure layouts here, so we have
4698 * a common path later on.
4699 */
4700 if (ip_tunnel_info_af(info) != AF_INET)
4701 goto err_clear;
4702 set_compat:
4703 to = (struct bpf_tunnel_key *)compat;
4704 break;
4705 default:
4706 goto err_clear;
4707 }
4708 }
4709
4710 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4711 to->tunnel_tos = info->key.tos;
4712 to->tunnel_ttl = info->key.ttl;
4713 if (flags & BPF_F_TUNINFO_FLAGS)
4714 to->tunnel_flags = ip_tunnel_flags_to_be16(info->key.tun_flags);
4715 else
4716 to->tunnel_ext = 0;
4717
4718 if (flags & BPF_F_TUNINFO_IPV6) {
4719 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4720 sizeof(to->remote_ipv6));
4721 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4722 sizeof(to->local_ipv6));
4723 to->tunnel_label = be32_to_cpu(info->key.label);
4724 } else {
4725 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4726 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4727 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4728 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4729 to->tunnel_label = 0;
4730 }
4731
4732 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4733 memcpy(to_orig, to, size);
4734
4735 return 0;
4736 err_clear:
4737 memset(to_orig, 0, size);
4738 return err;
4739 }
4740
4741 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4742 .func = bpf_skb_get_tunnel_key,
4743 .gpl_only = false,
4744 .ret_type = RET_INTEGER,
4745 .arg1_type = ARG_PTR_TO_CTX,
4746 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4747 .arg3_type = ARG_CONST_SIZE,
4748 .arg4_type = ARG_ANYTHING,
4749 };
4750
BPF_CALL_3(bpf_skb_get_tunnel_opt,struct sk_buff *,skb,u8 *,to,u32,size)4751 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4752 {
4753 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4754 int err;
4755
4756 if (unlikely(!info ||
4757 !ip_tunnel_is_options_present(info->key.tun_flags))) {
4758 err = -ENOENT;
4759 goto err_clear;
4760 }
4761 if (unlikely(size < info->options_len)) {
4762 err = -ENOMEM;
4763 goto err_clear;
4764 }
4765
4766 ip_tunnel_info_opts_get(to, info);
4767 if (size > info->options_len)
4768 memset(to + info->options_len, 0, size - info->options_len);
4769
4770 return info->options_len;
4771 err_clear:
4772 memset(to, 0, size);
4773 return err;
4774 }
4775
4776 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4777 .func = bpf_skb_get_tunnel_opt,
4778 .gpl_only = false,
4779 .ret_type = RET_INTEGER,
4780 .arg1_type = ARG_PTR_TO_CTX,
4781 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4782 .arg3_type = ARG_CONST_SIZE,
4783 };
4784
4785 static struct metadata_dst __percpu *md_dst;
4786
BPF_CALL_4(bpf_skb_set_tunnel_key,struct sk_buff *,skb,const struct bpf_tunnel_key *,from,u32,size,u64,flags)4787 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4788 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4789 {
4790 struct metadata_dst *md = this_cpu_ptr(md_dst);
4791 u8 compat[sizeof(struct bpf_tunnel_key)];
4792 struct ip_tunnel_info *info;
4793
4794 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4795 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4796 BPF_F_NO_TUNNEL_KEY)))
4797 return -EINVAL;
4798 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4799 switch (size) {
4800 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4801 case offsetof(struct bpf_tunnel_key, tunnel_label):
4802 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4803 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4804 /* Fixup deprecated structure layouts here, so we have
4805 * a common path later on.
4806 */
4807 memcpy(compat, from, size);
4808 memset(compat + size, 0, sizeof(compat) - size);
4809 from = (const struct bpf_tunnel_key *) compat;
4810 break;
4811 default:
4812 return -EINVAL;
4813 }
4814 }
4815 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4816 from->tunnel_ext))
4817 return -EINVAL;
4818
4819 skb_dst_drop(skb);
4820 dst_hold((struct dst_entry *) md);
4821 skb_dst_set(skb, (struct dst_entry *) md);
4822
4823 info = &md->u.tun_info;
4824 memset(info, 0, sizeof(*info));
4825 info->mode = IP_TUNNEL_INFO_TX;
4826
4827 __set_bit(IP_TUNNEL_NOCACHE_BIT, info->key.tun_flags);
4828 __assign_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, info->key.tun_flags,
4829 flags & BPF_F_DONT_FRAGMENT);
4830 __assign_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags,
4831 !(flags & BPF_F_ZERO_CSUM_TX));
4832 __assign_bit(IP_TUNNEL_SEQ_BIT, info->key.tun_flags,
4833 flags & BPF_F_SEQ_NUMBER);
4834 __assign_bit(IP_TUNNEL_KEY_BIT, info->key.tun_flags,
4835 !(flags & BPF_F_NO_TUNNEL_KEY));
4836
4837 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4838 info->key.tos = from->tunnel_tos;
4839 info->key.ttl = from->tunnel_ttl;
4840
4841 if (flags & BPF_F_TUNINFO_IPV6) {
4842 info->mode |= IP_TUNNEL_INFO_IPV6;
4843 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4844 sizeof(from->remote_ipv6));
4845 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4846 sizeof(from->local_ipv6));
4847 info->key.label = cpu_to_be32(from->tunnel_label) &
4848 IPV6_FLOWLABEL_MASK;
4849 } else {
4850 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4851 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4852 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4853 }
4854
4855 return 0;
4856 }
4857
4858 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4859 .func = bpf_skb_set_tunnel_key,
4860 .gpl_only = false,
4861 .ret_type = RET_INTEGER,
4862 .arg1_type = ARG_PTR_TO_CTX,
4863 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4864 .arg3_type = ARG_CONST_SIZE,
4865 .arg4_type = ARG_ANYTHING,
4866 };
4867
BPF_CALL_3(bpf_skb_set_tunnel_opt,struct sk_buff *,skb,const u8 *,from,u32,size)4868 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4869 const u8 *, from, u32, size)
4870 {
4871 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4872 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4873 IP_TUNNEL_DECLARE_FLAGS(present) = { };
4874
4875 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4876 return -EINVAL;
4877 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4878 return -ENOMEM;
4879
4880 ip_tunnel_set_options_present(present);
4881 ip_tunnel_info_opts_set(info, from, size, present);
4882
4883 return 0;
4884 }
4885
4886 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4887 .func = bpf_skb_set_tunnel_opt,
4888 .gpl_only = false,
4889 .ret_type = RET_INTEGER,
4890 .arg1_type = ARG_PTR_TO_CTX,
4891 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4892 .arg3_type = ARG_CONST_SIZE,
4893 };
4894
4895 static const struct bpf_func_proto *
bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)4896 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4897 {
4898 if (!md_dst) {
4899 struct metadata_dst __percpu *tmp;
4900
4901 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4902 METADATA_IP_TUNNEL,
4903 GFP_KERNEL);
4904 if (!tmp)
4905 return NULL;
4906 if (cmpxchg(&md_dst, NULL, tmp))
4907 metadata_dst_free_percpu(tmp);
4908 }
4909
4910 switch (which) {
4911 case BPF_FUNC_skb_set_tunnel_key:
4912 return &bpf_skb_set_tunnel_key_proto;
4913 case BPF_FUNC_skb_set_tunnel_opt:
4914 return &bpf_skb_set_tunnel_opt_proto;
4915 default:
4916 return NULL;
4917 }
4918 }
4919
BPF_CALL_3(bpf_skb_under_cgroup,struct sk_buff *,skb,struct bpf_map *,map,u32,idx)4920 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4921 u32, idx)
4922 {
4923 struct bpf_array *array = container_of(map, struct bpf_array, map);
4924 struct cgroup *cgrp;
4925 struct sock *sk;
4926
4927 sk = skb_to_full_sk(skb);
4928 if (!sk || !sk_fullsock(sk))
4929 return -ENOENT;
4930 if (unlikely(idx >= array->map.max_entries))
4931 return -E2BIG;
4932
4933 cgrp = READ_ONCE(array->ptrs[idx]);
4934 if (unlikely(!cgrp))
4935 return -EAGAIN;
4936
4937 return sk_under_cgroup_hierarchy(sk, cgrp);
4938 }
4939
4940 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4941 .func = bpf_skb_under_cgroup,
4942 .gpl_only = false,
4943 .ret_type = RET_INTEGER,
4944 .arg1_type = ARG_PTR_TO_CTX,
4945 .arg2_type = ARG_CONST_MAP_PTR,
4946 .arg3_type = ARG_ANYTHING,
4947 };
4948
4949 #ifdef CONFIG_SOCK_CGROUP_DATA
__bpf_sk_cgroup_id(struct sock * sk)4950 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4951 {
4952 struct cgroup *cgrp;
4953
4954 sk = sk_to_full_sk(sk);
4955 if (!sk || !sk_fullsock(sk))
4956 return 0;
4957
4958 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4959 return cgroup_id(cgrp);
4960 }
4961
BPF_CALL_1(bpf_skb_cgroup_id,const struct sk_buff *,skb)4962 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4963 {
4964 return __bpf_sk_cgroup_id(skb->sk);
4965 }
4966
4967 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4968 .func = bpf_skb_cgroup_id,
4969 .gpl_only = false,
4970 .ret_type = RET_INTEGER,
4971 .arg1_type = ARG_PTR_TO_CTX,
4972 };
4973
__bpf_sk_ancestor_cgroup_id(struct sock * sk,int ancestor_level)4974 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4975 int ancestor_level)
4976 {
4977 struct cgroup *ancestor;
4978 struct cgroup *cgrp;
4979
4980 sk = sk_to_full_sk(sk);
4981 if (!sk || !sk_fullsock(sk))
4982 return 0;
4983
4984 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4985 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4986 if (!ancestor)
4987 return 0;
4988
4989 return cgroup_id(ancestor);
4990 }
4991
BPF_CALL_2(bpf_skb_ancestor_cgroup_id,const struct sk_buff *,skb,int,ancestor_level)4992 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4993 ancestor_level)
4994 {
4995 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4996 }
4997
4998 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4999 .func = bpf_skb_ancestor_cgroup_id,
5000 .gpl_only = false,
5001 .ret_type = RET_INTEGER,
5002 .arg1_type = ARG_PTR_TO_CTX,
5003 .arg2_type = ARG_ANYTHING,
5004 };
5005
BPF_CALL_1(bpf_sk_cgroup_id,struct sock *,sk)5006 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
5007 {
5008 return __bpf_sk_cgroup_id(sk);
5009 }
5010
5011 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
5012 .func = bpf_sk_cgroup_id,
5013 .gpl_only = false,
5014 .ret_type = RET_INTEGER,
5015 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5016 };
5017
BPF_CALL_2(bpf_sk_ancestor_cgroup_id,struct sock *,sk,int,ancestor_level)5018 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
5019 {
5020 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
5021 }
5022
5023 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
5024 .func = bpf_sk_ancestor_cgroup_id,
5025 .gpl_only = false,
5026 .ret_type = RET_INTEGER,
5027 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5028 .arg2_type = ARG_ANYTHING,
5029 };
5030 #endif
5031
bpf_xdp_copy(void * dst,const void * ctx,unsigned long off,unsigned long len)5032 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
5033 unsigned long off, unsigned long len)
5034 {
5035 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
5036
5037 bpf_xdp_copy_buf(xdp, off, dst, len, false);
5038 return 0;
5039 }
5040
BPF_CALL_5(bpf_xdp_event_output,struct xdp_buff *,xdp,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)5041 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
5042 u64, flags, void *, meta, u64, meta_size)
5043 {
5044 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
5045
5046 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
5047 return -EINVAL;
5048
5049 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
5050 return -EFAULT;
5051
5052 return bpf_event_output(map, flags, meta, meta_size, xdp,
5053 xdp_size, bpf_xdp_copy);
5054 }
5055
5056 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
5057 .func = bpf_xdp_event_output,
5058 .gpl_only = true,
5059 .ret_type = RET_INTEGER,
5060 .arg1_type = ARG_PTR_TO_CTX,
5061 .arg2_type = ARG_CONST_MAP_PTR,
5062 .arg3_type = ARG_ANYTHING,
5063 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5064 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5065 };
5066
5067 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
5068
5069 const struct bpf_func_proto bpf_xdp_output_proto = {
5070 .func = bpf_xdp_event_output,
5071 .gpl_only = true,
5072 .ret_type = RET_INTEGER,
5073 .arg1_type = ARG_PTR_TO_BTF_ID,
5074 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
5075 .arg2_type = ARG_CONST_MAP_PTR,
5076 .arg3_type = ARG_ANYTHING,
5077 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5078 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5079 };
5080
BPF_CALL_1(bpf_get_socket_cookie,struct sk_buff *,skb)5081 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
5082 {
5083 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
5084 }
5085
5086 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
5087 .func = bpf_get_socket_cookie,
5088 .gpl_only = false,
5089 .ret_type = RET_INTEGER,
5090 .arg1_type = ARG_PTR_TO_CTX,
5091 };
5092
BPF_CALL_1(bpf_get_socket_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5093 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5094 {
5095 return __sock_gen_cookie(ctx->sk);
5096 }
5097
5098 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
5099 .func = bpf_get_socket_cookie_sock_addr,
5100 .gpl_only = false,
5101 .ret_type = RET_INTEGER,
5102 .arg1_type = ARG_PTR_TO_CTX,
5103 };
5104
BPF_CALL_1(bpf_get_socket_cookie_sock,struct sock *,ctx)5105 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
5106 {
5107 return __sock_gen_cookie(ctx);
5108 }
5109
5110 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5111 .func = bpf_get_socket_cookie_sock,
5112 .gpl_only = false,
5113 .ret_type = RET_INTEGER,
5114 .arg1_type = ARG_PTR_TO_CTX,
5115 };
5116
BPF_CALL_1(bpf_get_socket_ptr_cookie,struct sock *,sk)5117 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5118 {
5119 return sk ? sock_gen_cookie(sk) : 0;
5120 }
5121
5122 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5123 .func = bpf_get_socket_ptr_cookie,
5124 .gpl_only = false,
5125 .ret_type = RET_INTEGER,
5126 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5127 };
5128
BPF_CALL_1(bpf_get_socket_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5129 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5130 {
5131 return __sock_gen_cookie(ctx->sk);
5132 }
5133
5134 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5135 .func = bpf_get_socket_cookie_sock_ops,
5136 .gpl_only = false,
5137 .ret_type = RET_INTEGER,
5138 .arg1_type = ARG_PTR_TO_CTX,
5139 };
5140
__bpf_get_netns_cookie(struct sock * sk)5141 static u64 __bpf_get_netns_cookie(struct sock *sk)
5142 {
5143 const struct net *net = sk ? sock_net(sk) : &init_net;
5144
5145 return net->net_cookie;
5146 }
5147
BPF_CALL_1(bpf_get_netns_cookie,struct sk_buff *,skb)5148 BPF_CALL_1(bpf_get_netns_cookie, struct sk_buff *, skb)
5149 {
5150 return __bpf_get_netns_cookie(skb && skb->sk ? skb->sk : NULL);
5151 }
5152
5153 static const struct bpf_func_proto bpf_get_netns_cookie_proto = {
5154 .func = bpf_get_netns_cookie,
5155 .ret_type = RET_INTEGER,
5156 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5157 };
5158
BPF_CALL_1(bpf_get_netns_cookie_sock,struct sock *,ctx)5159 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5160 {
5161 return __bpf_get_netns_cookie(ctx);
5162 }
5163
5164 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5165 .func = bpf_get_netns_cookie_sock,
5166 .gpl_only = false,
5167 .ret_type = RET_INTEGER,
5168 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5169 };
5170
BPF_CALL_1(bpf_get_netns_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5171 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5172 {
5173 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5174 }
5175
5176 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5177 .func = bpf_get_netns_cookie_sock_addr,
5178 .gpl_only = false,
5179 .ret_type = RET_INTEGER,
5180 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5181 };
5182
BPF_CALL_1(bpf_get_netns_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5183 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5184 {
5185 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5186 }
5187
5188 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5189 .func = bpf_get_netns_cookie_sock_ops,
5190 .gpl_only = false,
5191 .ret_type = RET_INTEGER,
5192 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5193 };
5194
BPF_CALL_1(bpf_get_netns_cookie_sk_msg,struct sk_msg *,ctx)5195 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5196 {
5197 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5198 }
5199
5200 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5201 .func = bpf_get_netns_cookie_sk_msg,
5202 .gpl_only = false,
5203 .ret_type = RET_INTEGER,
5204 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5205 };
5206
BPF_CALL_1(bpf_get_socket_uid,struct sk_buff *,skb)5207 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5208 {
5209 struct sock *sk = sk_to_full_sk(skb->sk);
5210 kuid_t kuid;
5211
5212 if (!sk || !sk_fullsock(sk))
5213 return overflowuid;
5214 kuid = sock_net_uid(sock_net(sk), sk);
5215 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5216 }
5217
5218 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5219 .func = bpf_get_socket_uid,
5220 .gpl_only = false,
5221 .ret_type = RET_INTEGER,
5222 .arg1_type = ARG_PTR_TO_CTX,
5223 };
5224
sol_socket_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5225 static int sol_socket_sockopt(struct sock *sk, int optname,
5226 char *optval, int *optlen,
5227 bool getopt)
5228 {
5229 switch (optname) {
5230 case SO_REUSEADDR:
5231 case SO_SNDBUF:
5232 case SO_RCVBUF:
5233 case SO_KEEPALIVE:
5234 case SO_PRIORITY:
5235 case SO_REUSEPORT:
5236 case SO_RCVLOWAT:
5237 case SO_MARK:
5238 case SO_MAX_PACING_RATE:
5239 case SO_BINDTOIFINDEX:
5240 case SO_TXREHASH:
5241 if (*optlen != sizeof(int))
5242 return -EINVAL;
5243 break;
5244 case SO_BINDTODEVICE:
5245 break;
5246 default:
5247 return -EINVAL;
5248 }
5249
5250 if (getopt) {
5251 if (optname == SO_BINDTODEVICE)
5252 return -EINVAL;
5253 return sk_getsockopt(sk, SOL_SOCKET, optname,
5254 KERNEL_SOCKPTR(optval),
5255 KERNEL_SOCKPTR(optlen));
5256 }
5257
5258 return sk_setsockopt(sk, SOL_SOCKET, optname,
5259 KERNEL_SOCKPTR(optval), *optlen);
5260 }
5261
bpf_sol_tcp_setsockopt(struct sock * sk,int optname,char * optval,int optlen)5262 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5263 char *optval, int optlen)
5264 {
5265 struct tcp_sock *tp = tcp_sk(sk);
5266 unsigned long timeout;
5267 int val;
5268
5269 if (optlen != sizeof(int))
5270 return -EINVAL;
5271
5272 val = *(int *)optval;
5273
5274 /* Only some options are supported */
5275 switch (optname) {
5276 case TCP_BPF_IW:
5277 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5278 return -EINVAL;
5279 tcp_snd_cwnd_set(tp, val);
5280 break;
5281 case TCP_BPF_SNDCWND_CLAMP:
5282 if (val <= 0)
5283 return -EINVAL;
5284 tp->snd_cwnd_clamp = val;
5285 tp->snd_ssthresh = val;
5286 break;
5287 case TCP_BPF_DELACK_MAX:
5288 timeout = usecs_to_jiffies(val);
5289 if (timeout > TCP_DELACK_MAX ||
5290 timeout < TCP_TIMEOUT_MIN)
5291 return -EINVAL;
5292 inet_csk(sk)->icsk_delack_max = timeout;
5293 break;
5294 case TCP_BPF_RTO_MIN:
5295 timeout = usecs_to_jiffies(val);
5296 if (timeout > TCP_RTO_MIN ||
5297 timeout < TCP_TIMEOUT_MIN)
5298 return -EINVAL;
5299 inet_csk(sk)->icsk_rto_min = timeout;
5300 break;
5301 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5302 if (val & ~(BPF_SOCK_OPS_ALL_CB_FLAGS))
5303 return -EINVAL;
5304 tp->bpf_sock_ops_cb_flags = val;
5305 break;
5306 default:
5307 return -EINVAL;
5308 }
5309
5310 return 0;
5311 }
5312
sol_tcp_sockopt_congestion(struct sock * sk,char * optval,int * optlen,bool getopt)5313 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5314 int *optlen, bool getopt)
5315 {
5316 struct tcp_sock *tp;
5317 int ret;
5318
5319 if (*optlen < 2)
5320 return -EINVAL;
5321
5322 if (getopt) {
5323 if (!inet_csk(sk)->icsk_ca_ops)
5324 return -EINVAL;
5325 /* BPF expects NULL-terminated tcp-cc string */
5326 optval[--(*optlen)] = '\0';
5327 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5328 KERNEL_SOCKPTR(optval),
5329 KERNEL_SOCKPTR(optlen));
5330 }
5331
5332 /* "cdg" is the only cc that alloc a ptr
5333 * in inet_csk_ca area. The bpf-tcp-cc may
5334 * overwrite this ptr after switching to cdg.
5335 */
5336 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5337 return -ENOTSUPP;
5338
5339 /* It stops this looping
5340 *
5341 * .init => bpf_setsockopt(tcp_cc) => .init =>
5342 * bpf_setsockopt(tcp_cc)" => .init => ....
5343 *
5344 * The second bpf_setsockopt(tcp_cc) is not allowed
5345 * in order to break the loop when both .init
5346 * are the same bpf prog.
5347 *
5348 * This applies even the second bpf_setsockopt(tcp_cc)
5349 * does not cause a loop. This limits only the first
5350 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5351 * pick a fallback cc (eg. peer does not support ECN)
5352 * and the second '.init' cannot fallback to
5353 * another.
5354 */
5355 tp = tcp_sk(sk);
5356 if (tp->bpf_chg_cc_inprogress)
5357 return -EBUSY;
5358
5359 tp->bpf_chg_cc_inprogress = 1;
5360 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5361 KERNEL_SOCKPTR(optval), *optlen);
5362 tp->bpf_chg_cc_inprogress = 0;
5363 return ret;
5364 }
5365
sol_tcp_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5366 static int sol_tcp_sockopt(struct sock *sk, int optname,
5367 char *optval, int *optlen,
5368 bool getopt)
5369 {
5370 if (sk->sk_protocol != IPPROTO_TCP)
5371 return -EINVAL;
5372
5373 switch (optname) {
5374 case TCP_NODELAY:
5375 case TCP_MAXSEG:
5376 case TCP_KEEPIDLE:
5377 case TCP_KEEPINTVL:
5378 case TCP_KEEPCNT:
5379 case TCP_SYNCNT:
5380 case TCP_WINDOW_CLAMP:
5381 case TCP_THIN_LINEAR_TIMEOUTS:
5382 case TCP_USER_TIMEOUT:
5383 case TCP_NOTSENT_LOWAT:
5384 case TCP_SAVE_SYN:
5385 if (*optlen != sizeof(int))
5386 return -EINVAL;
5387 break;
5388 case TCP_CONGESTION:
5389 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5390 case TCP_SAVED_SYN:
5391 if (*optlen < 1)
5392 return -EINVAL;
5393 break;
5394 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5395 if (*optlen != sizeof(int))
5396 return -EINVAL;
5397 if (getopt) {
5398 struct tcp_sock *tp = tcp_sk(sk);
5399 int cb_flags = tp->bpf_sock_ops_cb_flags;
5400
5401 memcpy(optval, &cb_flags, *optlen);
5402 return 0;
5403 }
5404 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5405 default:
5406 if (getopt)
5407 return -EINVAL;
5408 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5409 }
5410
5411 if (getopt) {
5412 if (optname == TCP_SAVED_SYN) {
5413 struct tcp_sock *tp = tcp_sk(sk);
5414
5415 if (!tp->saved_syn ||
5416 *optlen > tcp_saved_syn_len(tp->saved_syn))
5417 return -EINVAL;
5418 memcpy(optval, tp->saved_syn->data, *optlen);
5419 /* It cannot free tp->saved_syn here because it
5420 * does not know if the user space still needs it.
5421 */
5422 return 0;
5423 }
5424
5425 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5426 KERNEL_SOCKPTR(optval),
5427 KERNEL_SOCKPTR(optlen));
5428 }
5429
5430 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5431 KERNEL_SOCKPTR(optval), *optlen);
5432 }
5433
sol_ip_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5434 static int sol_ip_sockopt(struct sock *sk, int optname,
5435 char *optval, int *optlen,
5436 bool getopt)
5437 {
5438 if (sk->sk_family != AF_INET)
5439 return -EINVAL;
5440
5441 switch (optname) {
5442 case IP_TOS:
5443 if (*optlen != sizeof(int))
5444 return -EINVAL;
5445 break;
5446 default:
5447 return -EINVAL;
5448 }
5449
5450 if (getopt)
5451 return do_ip_getsockopt(sk, SOL_IP, optname,
5452 KERNEL_SOCKPTR(optval),
5453 KERNEL_SOCKPTR(optlen));
5454
5455 return do_ip_setsockopt(sk, SOL_IP, optname,
5456 KERNEL_SOCKPTR(optval), *optlen);
5457 }
5458
sol_ipv6_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5459 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5460 char *optval, int *optlen,
5461 bool getopt)
5462 {
5463 if (sk->sk_family != AF_INET6)
5464 return -EINVAL;
5465
5466 switch (optname) {
5467 case IPV6_TCLASS:
5468 case IPV6_AUTOFLOWLABEL:
5469 if (*optlen != sizeof(int))
5470 return -EINVAL;
5471 break;
5472 default:
5473 return -EINVAL;
5474 }
5475
5476 if (getopt)
5477 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5478 KERNEL_SOCKPTR(optval),
5479 KERNEL_SOCKPTR(optlen));
5480
5481 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5482 KERNEL_SOCKPTR(optval), *optlen);
5483 }
5484
__bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5485 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5486 char *optval, int optlen)
5487 {
5488 if (!sk_fullsock(sk))
5489 return -EINVAL;
5490
5491 if (level == SOL_SOCKET)
5492 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5493 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5494 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5495 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5496 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5497 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5498 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5499
5500 return -EINVAL;
5501 }
5502
_bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5503 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5504 char *optval, int optlen)
5505 {
5506 if (sk_fullsock(sk))
5507 sock_owned_by_me(sk);
5508 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5509 }
5510
__bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5511 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5512 char *optval, int optlen)
5513 {
5514 int err, saved_optlen = optlen;
5515
5516 if (!sk_fullsock(sk)) {
5517 err = -EINVAL;
5518 goto done;
5519 }
5520
5521 if (level == SOL_SOCKET)
5522 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5523 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5524 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5525 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5526 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5527 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5528 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5529 else
5530 err = -EINVAL;
5531
5532 done:
5533 if (err)
5534 optlen = 0;
5535 if (optlen < saved_optlen)
5536 memset(optval + optlen, 0, saved_optlen - optlen);
5537 return err;
5538 }
5539
_bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5540 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5541 char *optval, int optlen)
5542 {
5543 if (sk_fullsock(sk))
5544 sock_owned_by_me(sk);
5545 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5546 }
5547
BPF_CALL_5(bpf_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5548 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5549 int, optname, char *, optval, int, optlen)
5550 {
5551 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5552 }
5553
5554 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5555 .func = bpf_sk_setsockopt,
5556 .gpl_only = false,
5557 .ret_type = RET_INTEGER,
5558 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5559 .arg2_type = ARG_ANYTHING,
5560 .arg3_type = ARG_ANYTHING,
5561 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5562 .arg5_type = ARG_CONST_SIZE,
5563 };
5564
BPF_CALL_5(bpf_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5565 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5566 int, optname, char *, optval, int, optlen)
5567 {
5568 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5569 }
5570
5571 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5572 .func = bpf_sk_getsockopt,
5573 .gpl_only = false,
5574 .ret_type = RET_INTEGER,
5575 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5576 .arg2_type = ARG_ANYTHING,
5577 .arg3_type = ARG_ANYTHING,
5578 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5579 .arg5_type = ARG_CONST_SIZE,
5580 };
5581
BPF_CALL_5(bpf_unlocked_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5582 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5583 int, optname, char *, optval, int, optlen)
5584 {
5585 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5586 }
5587
5588 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5589 .func = bpf_unlocked_sk_setsockopt,
5590 .gpl_only = false,
5591 .ret_type = RET_INTEGER,
5592 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5593 .arg2_type = ARG_ANYTHING,
5594 .arg3_type = ARG_ANYTHING,
5595 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5596 .arg5_type = ARG_CONST_SIZE,
5597 };
5598
BPF_CALL_5(bpf_unlocked_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5599 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5600 int, optname, char *, optval, int, optlen)
5601 {
5602 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5603 }
5604
5605 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5606 .func = bpf_unlocked_sk_getsockopt,
5607 .gpl_only = false,
5608 .ret_type = RET_INTEGER,
5609 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5610 .arg2_type = ARG_ANYTHING,
5611 .arg3_type = ARG_ANYTHING,
5612 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5613 .arg5_type = ARG_CONST_SIZE,
5614 };
5615
BPF_CALL_5(bpf_sock_addr_setsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5616 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5617 int, level, int, optname, char *, optval, int, optlen)
5618 {
5619 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5620 }
5621
5622 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5623 .func = bpf_sock_addr_setsockopt,
5624 .gpl_only = false,
5625 .ret_type = RET_INTEGER,
5626 .arg1_type = ARG_PTR_TO_CTX,
5627 .arg2_type = ARG_ANYTHING,
5628 .arg3_type = ARG_ANYTHING,
5629 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5630 .arg5_type = ARG_CONST_SIZE,
5631 };
5632
BPF_CALL_5(bpf_sock_addr_getsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5633 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5634 int, level, int, optname, char *, optval, int, optlen)
5635 {
5636 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5637 }
5638
5639 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5640 .func = bpf_sock_addr_getsockopt,
5641 .gpl_only = false,
5642 .ret_type = RET_INTEGER,
5643 .arg1_type = ARG_PTR_TO_CTX,
5644 .arg2_type = ARG_ANYTHING,
5645 .arg3_type = ARG_ANYTHING,
5646 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5647 .arg5_type = ARG_CONST_SIZE,
5648 };
5649
BPF_CALL_5(bpf_sock_ops_setsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5650 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5651 int, level, int, optname, char *, optval, int, optlen)
5652 {
5653 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5654 }
5655
5656 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5657 .func = bpf_sock_ops_setsockopt,
5658 .gpl_only = false,
5659 .ret_type = RET_INTEGER,
5660 .arg1_type = ARG_PTR_TO_CTX,
5661 .arg2_type = ARG_ANYTHING,
5662 .arg3_type = ARG_ANYTHING,
5663 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5664 .arg5_type = ARG_CONST_SIZE,
5665 };
5666
bpf_sock_ops_get_syn(struct bpf_sock_ops_kern * bpf_sock,int optname,const u8 ** start)5667 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5668 int optname, const u8 **start)
5669 {
5670 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5671 const u8 *hdr_start;
5672 int ret;
5673
5674 if (syn_skb) {
5675 /* sk is a request_sock here */
5676
5677 if (optname == TCP_BPF_SYN) {
5678 hdr_start = syn_skb->data;
5679 ret = tcp_hdrlen(syn_skb);
5680 } else if (optname == TCP_BPF_SYN_IP) {
5681 hdr_start = skb_network_header(syn_skb);
5682 ret = skb_network_header_len(syn_skb) +
5683 tcp_hdrlen(syn_skb);
5684 } else {
5685 /* optname == TCP_BPF_SYN_MAC */
5686 hdr_start = skb_mac_header(syn_skb);
5687 ret = skb_mac_header_len(syn_skb) +
5688 skb_network_header_len(syn_skb) +
5689 tcp_hdrlen(syn_skb);
5690 }
5691 } else {
5692 struct sock *sk = bpf_sock->sk;
5693 struct saved_syn *saved_syn;
5694
5695 if (sk->sk_state == TCP_NEW_SYN_RECV)
5696 /* synack retransmit. bpf_sock->syn_skb will
5697 * not be available. It has to resort to
5698 * saved_syn (if it is saved).
5699 */
5700 saved_syn = inet_reqsk(sk)->saved_syn;
5701 else
5702 saved_syn = tcp_sk(sk)->saved_syn;
5703
5704 if (!saved_syn)
5705 return -ENOENT;
5706
5707 if (optname == TCP_BPF_SYN) {
5708 hdr_start = saved_syn->data +
5709 saved_syn->mac_hdrlen +
5710 saved_syn->network_hdrlen;
5711 ret = saved_syn->tcp_hdrlen;
5712 } else if (optname == TCP_BPF_SYN_IP) {
5713 hdr_start = saved_syn->data +
5714 saved_syn->mac_hdrlen;
5715 ret = saved_syn->network_hdrlen +
5716 saved_syn->tcp_hdrlen;
5717 } else {
5718 /* optname == TCP_BPF_SYN_MAC */
5719
5720 /* TCP_SAVE_SYN may not have saved the mac hdr */
5721 if (!saved_syn->mac_hdrlen)
5722 return -ENOENT;
5723
5724 hdr_start = saved_syn->data;
5725 ret = saved_syn->mac_hdrlen +
5726 saved_syn->network_hdrlen +
5727 saved_syn->tcp_hdrlen;
5728 }
5729 }
5730
5731 *start = hdr_start;
5732 return ret;
5733 }
5734
BPF_CALL_5(bpf_sock_ops_getsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5735 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5736 int, level, int, optname, char *, optval, int, optlen)
5737 {
5738 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5739 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5740 int ret, copy_len = 0;
5741 const u8 *start;
5742
5743 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5744 if (ret > 0) {
5745 copy_len = ret;
5746 if (optlen < copy_len) {
5747 copy_len = optlen;
5748 ret = -ENOSPC;
5749 }
5750
5751 memcpy(optval, start, copy_len);
5752 }
5753
5754 /* Zero out unused buffer at the end */
5755 memset(optval + copy_len, 0, optlen - copy_len);
5756
5757 return ret;
5758 }
5759
5760 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5761 }
5762
5763 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5764 .func = bpf_sock_ops_getsockopt,
5765 .gpl_only = false,
5766 .ret_type = RET_INTEGER,
5767 .arg1_type = ARG_PTR_TO_CTX,
5768 .arg2_type = ARG_ANYTHING,
5769 .arg3_type = ARG_ANYTHING,
5770 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5771 .arg5_type = ARG_CONST_SIZE,
5772 };
5773
BPF_CALL_2(bpf_sock_ops_cb_flags_set,struct bpf_sock_ops_kern *,bpf_sock,int,argval)5774 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5775 int, argval)
5776 {
5777 struct sock *sk = bpf_sock->sk;
5778 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5779
5780 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5781 return -EINVAL;
5782
5783 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5784
5785 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5786 }
5787
5788 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5789 .func = bpf_sock_ops_cb_flags_set,
5790 .gpl_only = false,
5791 .ret_type = RET_INTEGER,
5792 .arg1_type = ARG_PTR_TO_CTX,
5793 .arg2_type = ARG_ANYTHING,
5794 };
5795
5796 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5797 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5798
BPF_CALL_3(bpf_bind,struct bpf_sock_addr_kern *,ctx,struct sockaddr *,addr,int,addr_len)5799 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5800 int, addr_len)
5801 {
5802 #ifdef CONFIG_INET
5803 struct sock *sk = ctx->sk;
5804 u32 flags = BIND_FROM_BPF;
5805 int err;
5806
5807 err = -EINVAL;
5808 if (addr_len < offsetofend(struct sockaddr, sa_family))
5809 return err;
5810 if (addr->sa_family == AF_INET) {
5811 if (addr_len < sizeof(struct sockaddr_in))
5812 return err;
5813 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5814 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5815 return __inet_bind(sk, addr, addr_len, flags);
5816 #if IS_ENABLED(CONFIG_IPV6)
5817 } else if (addr->sa_family == AF_INET6) {
5818 if (addr_len < SIN6_LEN_RFC2133)
5819 return err;
5820 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5821 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5822 /* ipv6_bpf_stub cannot be NULL, since it's called from
5823 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5824 */
5825 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5826 #endif /* CONFIG_IPV6 */
5827 }
5828 #endif /* CONFIG_INET */
5829
5830 return -EAFNOSUPPORT;
5831 }
5832
5833 static const struct bpf_func_proto bpf_bind_proto = {
5834 .func = bpf_bind,
5835 .gpl_only = false,
5836 .ret_type = RET_INTEGER,
5837 .arg1_type = ARG_PTR_TO_CTX,
5838 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5839 .arg3_type = ARG_CONST_SIZE,
5840 };
5841
5842 #ifdef CONFIG_XFRM
5843
5844 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5845 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5846
5847 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5848 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5849
5850 #endif
5851
BPF_CALL_5(bpf_skb_get_xfrm_state,struct sk_buff *,skb,u32,index,struct bpf_xfrm_state *,to,u32,size,u64,flags)5852 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5853 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5854 {
5855 const struct sec_path *sp = skb_sec_path(skb);
5856 const struct xfrm_state *x;
5857
5858 if (!sp || unlikely(index >= sp->len || flags))
5859 goto err_clear;
5860
5861 x = sp->xvec[index];
5862
5863 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5864 goto err_clear;
5865
5866 to->reqid = x->props.reqid;
5867 to->spi = x->id.spi;
5868 to->family = x->props.family;
5869 to->ext = 0;
5870
5871 if (to->family == AF_INET6) {
5872 memcpy(to->remote_ipv6, x->props.saddr.a6,
5873 sizeof(to->remote_ipv6));
5874 } else {
5875 to->remote_ipv4 = x->props.saddr.a4;
5876 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5877 }
5878
5879 return 0;
5880 err_clear:
5881 memset(to, 0, size);
5882 return -EINVAL;
5883 }
5884
5885 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5886 .func = bpf_skb_get_xfrm_state,
5887 .gpl_only = false,
5888 .ret_type = RET_INTEGER,
5889 .arg1_type = ARG_PTR_TO_CTX,
5890 .arg2_type = ARG_ANYTHING,
5891 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5892 .arg4_type = ARG_CONST_SIZE,
5893 .arg5_type = ARG_ANYTHING,
5894 };
5895 #endif
5896
5897 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
bpf_fib_set_fwd_params(struct bpf_fib_lookup * params,u32 mtu)5898 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5899 {
5900 params->h_vlan_TCI = 0;
5901 params->h_vlan_proto = 0;
5902 if (mtu)
5903 params->mtu_result = mtu; /* union with tot_len */
5904
5905 return 0;
5906 }
5907 #endif
5908
5909 #if IS_ENABLED(CONFIG_INET)
bpf_ipv4_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5910 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5911 u32 flags, bool check_mtu)
5912 {
5913 struct fib_nh_common *nhc;
5914 struct in_device *in_dev;
5915 struct neighbour *neigh;
5916 struct net_device *dev;
5917 struct fib_result res;
5918 struct flowi4 fl4;
5919 u32 mtu = 0;
5920 int err;
5921
5922 dev = dev_get_by_index_rcu(net, params->ifindex);
5923 if (unlikely(!dev))
5924 return -ENODEV;
5925
5926 /* verify forwarding is enabled on this interface */
5927 in_dev = __in_dev_get_rcu(dev);
5928 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5929 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5930
5931 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5932 fl4.flowi4_iif = 1;
5933 fl4.flowi4_oif = params->ifindex;
5934 } else {
5935 fl4.flowi4_iif = params->ifindex;
5936 fl4.flowi4_oif = 0;
5937 }
5938 fl4.flowi4_tos = params->tos & INET_DSCP_MASK;
5939 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5940 fl4.flowi4_flags = 0;
5941
5942 fl4.flowi4_proto = params->l4_protocol;
5943 fl4.daddr = params->ipv4_dst;
5944 fl4.saddr = params->ipv4_src;
5945 fl4.fl4_sport = params->sport;
5946 fl4.fl4_dport = params->dport;
5947 fl4.flowi4_multipath_hash = 0;
5948
5949 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5950 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5951 struct fib_table *tb;
5952
5953 if (flags & BPF_FIB_LOOKUP_TBID) {
5954 tbid = params->tbid;
5955 /* zero out for vlan output */
5956 params->tbid = 0;
5957 }
5958
5959 tb = fib_get_table(net, tbid);
5960 if (unlikely(!tb))
5961 return BPF_FIB_LKUP_RET_NOT_FWDED;
5962
5963 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5964 } else {
5965 if (flags & BPF_FIB_LOOKUP_MARK)
5966 fl4.flowi4_mark = params->mark;
5967 else
5968 fl4.flowi4_mark = 0;
5969 fl4.flowi4_secid = 0;
5970 fl4.flowi4_tun_key.tun_id = 0;
5971 fl4.flowi4_uid = sock_net_uid(net, NULL);
5972
5973 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5974 }
5975
5976 if (err) {
5977 /* map fib lookup errors to RTN_ type */
5978 if (err == -EINVAL)
5979 return BPF_FIB_LKUP_RET_BLACKHOLE;
5980 if (err == -EHOSTUNREACH)
5981 return BPF_FIB_LKUP_RET_UNREACHABLE;
5982 if (err == -EACCES)
5983 return BPF_FIB_LKUP_RET_PROHIBIT;
5984
5985 return BPF_FIB_LKUP_RET_NOT_FWDED;
5986 }
5987
5988 if (res.type != RTN_UNICAST)
5989 return BPF_FIB_LKUP_RET_NOT_FWDED;
5990
5991 if (fib_info_num_path(res.fi) > 1)
5992 fib_select_path(net, &res, &fl4, NULL);
5993
5994 if (check_mtu) {
5995 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5996 if (params->tot_len > mtu) {
5997 params->mtu_result = mtu; /* union with tot_len */
5998 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5999 }
6000 }
6001
6002 nhc = res.nhc;
6003
6004 /* do not handle lwt encaps right now */
6005 if (nhc->nhc_lwtstate)
6006 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6007
6008 dev = nhc->nhc_dev;
6009
6010 params->rt_metric = res.fi->fib_priority;
6011 params->ifindex = dev->ifindex;
6012
6013 if (flags & BPF_FIB_LOOKUP_SRC)
6014 params->ipv4_src = fib_result_prefsrc(net, &res);
6015
6016 /* xdp and cls_bpf programs are run in RCU-bh so
6017 * rcu_read_lock_bh is not needed here
6018 */
6019 if (likely(nhc->nhc_gw_family != AF_INET6)) {
6020 if (nhc->nhc_gw_family)
6021 params->ipv4_dst = nhc->nhc_gw.ipv4;
6022 } else {
6023 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
6024
6025 params->family = AF_INET6;
6026 *dst = nhc->nhc_gw.ipv6;
6027 }
6028
6029 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6030 goto set_fwd_params;
6031
6032 if (likely(nhc->nhc_gw_family != AF_INET6))
6033 neigh = __ipv4_neigh_lookup_noref(dev,
6034 (__force u32)params->ipv4_dst);
6035 else
6036 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
6037
6038 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6039 return BPF_FIB_LKUP_RET_NO_NEIGH;
6040 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6041 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6042
6043 set_fwd_params:
6044 return bpf_fib_set_fwd_params(params, mtu);
6045 }
6046 #endif
6047
6048 #if IS_ENABLED(CONFIG_IPV6)
bpf_ipv6_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)6049 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
6050 u32 flags, bool check_mtu)
6051 {
6052 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
6053 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
6054 struct fib6_result res = {};
6055 struct neighbour *neigh;
6056 struct net_device *dev;
6057 struct inet6_dev *idev;
6058 struct flowi6 fl6;
6059 int strict = 0;
6060 int oif, err;
6061 u32 mtu = 0;
6062
6063 /* link local addresses are never forwarded */
6064 if (rt6_need_strict(dst) || rt6_need_strict(src))
6065 return BPF_FIB_LKUP_RET_NOT_FWDED;
6066
6067 dev = dev_get_by_index_rcu(net, params->ifindex);
6068 if (unlikely(!dev))
6069 return -ENODEV;
6070
6071 idev = __in6_dev_get_safely(dev);
6072 if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding)))
6073 return BPF_FIB_LKUP_RET_FWD_DISABLED;
6074
6075 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
6076 fl6.flowi6_iif = 1;
6077 oif = fl6.flowi6_oif = params->ifindex;
6078 } else {
6079 oif = fl6.flowi6_iif = params->ifindex;
6080 fl6.flowi6_oif = 0;
6081 strict = RT6_LOOKUP_F_HAS_SADDR;
6082 }
6083 fl6.flowlabel = params->flowinfo;
6084 fl6.flowi6_scope = 0;
6085 fl6.flowi6_flags = 0;
6086 fl6.mp_hash = 0;
6087
6088 fl6.flowi6_proto = params->l4_protocol;
6089 fl6.daddr = *dst;
6090 fl6.saddr = *src;
6091 fl6.fl6_sport = params->sport;
6092 fl6.fl6_dport = params->dport;
6093
6094 if (flags & BPF_FIB_LOOKUP_DIRECT) {
6095 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
6096 struct fib6_table *tb;
6097
6098 if (flags & BPF_FIB_LOOKUP_TBID) {
6099 tbid = params->tbid;
6100 /* zero out for vlan output */
6101 params->tbid = 0;
6102 }
6103
6104 tb = ipv6_stub->fib6_get_table(net, tbid);
6105 if (unlikely(!tb))
6106 return BPF_FIB_LKUP_RET_NOT_FWDED;
6107
6108 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
6109 strict);
6110 } else {
6111 if (flags & BPF_FIB_LOOKUP_MARK)
6112 fl6.flowi6_mark = params->mark;
6113 else
6114 fl6.flowi6_mark = 0;
6115 fl6.flowi6_secid = 0;
6116 fl6.flowi6_tun_key.tun_id = 0;
6117 fl6.flowi6_uid = sock_net_uid(net, NULL);
6118
6119 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
6120 }
6121
6122 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
6123 res.f6i == net->ipv6.fib6_null_entry))
6124 return BPF_FIB_LKUP_RET_NOT_FWDED;
6125
6126 switch (res.fib6_type) {
6127 /* only unicast is forwarded */
6128 case RTN_UNICAST:
6129 break;
6130 case RTN_BLACKHOLE:
6131 return BPF_FIB_LKUP_RET_BLACKHOLE;
6132 case RTN_UNREACHABLE:
6133 return BPF_FIB_LKUP_RET_UNREACHABLE;
6134 case RTN_PROHIBIT:
6135 return BPF_FIB_LKUP_RET_PROHIBIT;
6136 default:
6137 return BPF_FIB_LKUP_RET_NOT_FWDED;
6138 }
6139
6140 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
6141 fl6.flowi6_oif != 0, NULL, strict);
6142
6143 if (check_mtu) {
6144 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
6145 if (params->tot_len > mtu) {
6146 params->mtu_result = mtu; /* union with tot_len */
6147 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6148 }
6149 }
6150
6151 if (res.nh->fib_nh_lws)
6152 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6153
6154 if (res.nh->fib_nh_gw_family)
6155 *dst = res.nh->fib_nh_gw6;
6156
6157 dev = res.nh->fib_nh_dev;
6158 params->rt_metric = res.f6i->fib6_metric;
6159 params->ifindex = dev->ifindex;
6160
6161 if (flags & BPF_FIB_LOOKUP_SRC) {
6162 if (res.f6i->fib6_prefsrc.plen) {
6163 *src = res.f6i->fib6_prefsrc.addr;
6164 } else {
6165 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
6166 &fl6.daddr, 0,
6167 src);
6168 if (err)
6169 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
6170 }
6171 }
6172
6173 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6174 goto set_fwd_params;
6175
6176 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6177 * not needed here.
6178 */
6179 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6180 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6181 return BPF_FIB_LKUP_RET_NO_NEIGH;
6182 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6183 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6184
6185 set_fwd_params:
6186 return bpf_fib_set_fwd_params(params, mtu);
6187 }
6188 #endif
6189
6190 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6191 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6192 BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_MARK)
6193
BPF_CALL_4(bpf_xdp_fib_lookup,struct xdp_buff *,ctx,struct bpf_fib_lookup *,params,int,plen,u32,flags)6194 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6195 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6196 {
6197 if (plen < sizeof(*params))
6198 return -EINVAL;
6199
6200 if (flags & ~BPF_FIB_LOOKUP_MASK)
6201 return -EINVAL;
6202
6203 switch (params->family) {
6204 #if IS_ENABLED(CONFIG_INET)
6205 case AF_INET:
6206 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6207 flags, true);
6208 #endif
6209 #if IS_ENABLED(CONFIG_IPV6)
6210 case AF_INET6:
6211 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6212 flags, true);
6213 #endif
6214 }
6215 return -EAFNOSUPPORT;
6216 }
6217
6218 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6219 .func = bpf_xdp_fib_lookup,
6220 .gpl_only = true,
6221 .ret_type = RET_INTEGER,
6222 .arg1_type = ARG_PTR_TO_CTX,
6223 .arg2_type = ARG_PTR_TO_MEM,
6224 .arg3_type = ARG_CONST_SIZE,
6225 .arg4_type = ARG_ANYTHING,
6226 };
6227
BPF_CALL_4(bpf_skb_fib_lookup,struct sk_buff *,skb,struct bpf_fib_lookup *,params,int,plen,u32,flags)6228 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6229 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6230 {
6231 struct net *net = dev_net(skb->dev);
6232 int rc = -EAFNOSUPPORT;
6233 bool check_mtu = false;
6234
6235 if (plen < sizeof(*params))
6236 return -EINVAL;
6237
6238 if (flags & ~BPF_FIB_LOOKUP_MASK)
6239 return -EINVAL;
6240
6241 if (params->tot_len)
6242 check_mtu = true;
6243
6244 switch (params->family) {
6245 #if IS_ENABLED(CONFIG_INET)
6246 case AF_INET:
6247 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6248 break;
6249 #endif
6250 #if IS_ENABLED(CONFIG_IPV6)
6251 case AF_INET6:
6252 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6253 break;
6254 #endif
6255 }
6256
6257 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6258 struct net_device *dev;
6259
6260 /* When tot_len isn't provided by user, check skb
6261 * against MTU of FIB lookup resulting net_device
6262 */
6263 dev = dev_get_by_index_rcu(net, params->ifindex);
6264 if (!is_skb_forwardable(dev, skb))
6265 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6266
6267 params->mtu_result = dev->mtu; /* union with tot_len */
6268 }
6269
6270 return rc;
6271 }
6272
6273 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6274 .func = bpf_skb_fib_lookup,
6275 .gpl_only = true,
6276 .ret_type = RET_INTEGER,
6277 .arg1_type = ARG_PTR_TO_CTX,
6278 .arg2_type = ARG_PTR_TO_MEM,
6279 .arg3_type = ARG_CONST_SIZE,
6280 .arg4_type = ARG_ANYTHING,
6281 };
6282
__dev_via_ifindex(struct net_device * dev_curr,u32 ifindex)6283 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6284 u32 ifindex)
6285 {
6286 struct net *netns = dev_net(dev_curr);
6287
6288 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6289 if (ifindex == 0)
6290 return dev_curr;
6291
6292 return dev_get_by_index_rcu(netns, ifindex);
6293 }
6294
BPF_CALL_5(bpf_skb_check_mtu,struct sk_buff *,skb,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6295 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6296 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6297 {
6298 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6299 struct net_device *dev = skb->dev;
6300 int mtu, dev_len, skb_len;
6301
6302 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6303 return -EINVAL;
6304 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6305 return -EINVAL;
6306
6307 dev = __dev_via_ifindex(dev, ifindex);
6308 if (unlikely(!dev))
6309 return -ENODEV;
6310
6311 mtu = READ_ONCE(dev->mtu);
6312 dev_len = mtu + dev->hard_header_len;
6313
6314 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6315 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6316
6317 skb_len += len_diff; /* minus result pass check */
6318 if (skb_len <= dev_len) {
6319 ret = BPF_MTU_CHK_RET_SUCCESS;
6320 goto out;
6321 }
6322 /* At this point, skb->len exceed MTU, but as it include length of all
6323 * segments, it can still be below MTU. The SKB can possibly get
6324 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6325 * must choose if segs are to be MTU checked.
6326 */
6327 if (skb_is_gso(skb)) {
6328 ret = BPF_MTU_CHK_RET_SUCCESS;
6329 if (flags & BPF_MTU_CHK_SEGS &&
6330 !skb_gso_validate_network_len(skb, mtu))
6331 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6332 }
6333 out:
6334 *mtu_len = mtu;
6335 return ret;
6336 }
6337
BPF_CALL_5(bpf_xdp_check_mtu,struct xdp_buff *,xdp,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6338 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6339 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6340 {
6341 struct net_device *dev = xdp->rxq->dev;
6342 int xdp_len = xdp->data_end - xdp->data;
6343 int ret = BPF_MTU_CHK_RET_SUCCESS;
6344 int mtu, dev_len;
6345
6346 /* XDP variant doesn't support multi-buffer segment check (yet) */
6347 if (unlikely(flags))
6348 return -EINVAL;
6349
6350 dev = __dev_via_ifindex(dev, ifindex);
6351 if (unlikely(!dev))
6352 return -ENODEV;
6353
6354 mtu = READ_ONCE(dev->mtu);
6355 dev_len = mtu + dev->hard_header_len;
6356
6357 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6358 if (*mtu_len)
6359 xdp_len = *mtu_len + dev->hard_header_len;
6360
6361 xdp_len += len_diff; /* minus result pass check */
6362 if (xdp_len > dev_len)
6363 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6364
6365 *mtu_len = mtu;
6366 return ret;
6367 }
6368
6369 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6370 .func = bpf_skb_check_mtu,
6371 .gpl_only = true,
6372 .ret_type = RET_INTEGER,
6373 .arg1_type = ARG_PTR_TO_CTX,
6374 .arg2_type = ARG_ANYTHING,
6375 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_WRITE | MEM_ALIGNED,
6376 .arg3_size = sizeof(u32),
6377 .arg4_type = ARG_ANYTHING,
6378 .arg5_type = ARG_ANYTHING,
6379 };
6380
6381 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6382 .func = bpf_xdp_check_mtu,
6383 .gpl_only = true,
6384 .ret_type = RET_INTEGER,
6385 .arg1_type = ARG_PTR_TO_CTX,
6386 .arg2_type = ARG_ANYTHING,
6387 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_WRITE | MEM_ALIGNED,
6388 .arg3_size = sizeof(u32),
6389 .arg4_type = ARG_ANYTHING,
6390 .arg5_type = ARG_ANYTHING,
6391 };
6392
6393 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
bpf_push_seg6_encap(struct sk_buff * skb,u32 type,void * hdr,u32 len)6394 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6395 {
6396 int err;
6397 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6398
6399 if (!seg6_validate_srh(srh, len, false))
6400 return -EINVAL;
6401
6402 switch (type) {
6403 case BPF_LWT_ENCAP_SEG6_INLINE:
6404 if (skb->protocol != htons(ETH_P_IPV6))
6405 return -EBADMSG;
6406
6407 err = seg6_do_srh_inline(skb, srh);
6408 break;
6409 case BPF_LWT_ENCAP_SEG6:
6410 skb_reset_inner_headers(skb);
6411 skb->encapsulation = 1;
6412 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6413 break;
6414 default:
6415 return -EINVAL;
6416 }
6417
6418 bpf_compute_data_pointers(skb);
6419 if (err)
6420 return err;
6421
6422 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6423
6424 return seg6_lookup_nexthop(skb, NULL, 0);
6425 }
6426 #endif /* CONFIG_IPV6_SEG6_BPF */
6427
6428 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
bpf_push_ip_encap(struct sk_buff * skb,void * hdr,u32 len,bool ingress)6429 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6430 bool ingress)
6431 {
6432 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6433 }
6434 #endif
6435
BPF_CALL_4(bpf_lwt_in_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6436 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6437 u32, len)
6438 {
6439 switch (type) {
6440 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6441 case BPF_LWT_ENCAP_SEG6:
6442 case BPF_LWT_ENCAP_SEG6_INLINE:
6443 return bpf_push_seg6_encap(skb, type, hdr, len);
6444 #endif
6445 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6446 case BPF_LWT_ENCAP_IP:
6447 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6448 #endif
6449 default:
6450 return -EINVAL;
6451 }
6452 }
6453
BPF_CALL_4(bpf_lwt_xmit_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6454 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6455 void *, hdr, u32, len)
6456 {
6457 switch (type) {
6458 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6459 case BPF_LWT_ENCAP_IP:
6460 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6461 #endif
6462 default:
6463 return -EINVAL;
6464 }
6465 }
6466
6467 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6468 .func = bpf_lwt_in_push_encap,
6469 .gpl_only = false,
6470 .ret_type = RET_INTEGER,
6471 .arg1_type = ARG_PTR_TO_CTX,
6472 .arg2_type = ARG_ANYTHING,
6473 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6474 .arg4_type = ARG_CONST_SIZE
6475 };
6476
6477 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6478 .func = bpf_lwt_xmit_push_encap,
6479 .gpl_only = false,
6480 .ret_type = RET_INTEGER,
6481 .arg1_type = ARG_PTR_TO_CTX,
6482 .arg2_type = ARG_ANYTHING,
6483 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6484 .arg4_type = ARG_CONST_SIZE
6485 };
6486
6487 #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)6488 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6489 const void *, from, u32, len)
6490 {
6491 struct seg6_bpf_srh_state *srh_state =
6492 this_cpu_ptr(&seg6_bpf_srh_states);
6493 struct ipv6_sr_hdr *srh = srh_state->srh;
6494 void *srh_tlvs, *srh_end, *ptr;
6495 int srhoff = 0;
6496
6497 lockdep_assert_held(&srh_state->bh_lock);
6498 if (srh == NULL)
6499 return -EINVAL;
6500
6501 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6502 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6503
6504 ptr = skb->data + offset;
6505 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6506 srh_state->valid = false;
6507 else if (ptr < (void *)&srh->flags ||
6508 ptr + len > (void *)&srh->segments)
6509 return -EFAULT;
6510
6511 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6512 return -EFAULT;
6513 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6514 return -EINVAL;
6515 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6516
6517 memcpy(skb->data + offset, from, len);
6518 return 0;
6519 }
6520
6521 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6522 .func = bpf_lwt_seg6_store_bytes,
6523 .gpl_only = false,
6524 .ret_type = RET_INTEGER,
6525 .arg1_type = ARG_PTR_TO_CTX,
6526 .arg2_type = ARG_ANYTHING,
6527 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6528 .arg4_type = ARG_CONST_SIZE
6529 };
6530
bpf_update_srh_state(struct sk_buff * skb)6531 static void bpf_update_srh_state(struct sk_buff *skb)
6532 {
6533 struct seg6_bpf_srh_state *srh_state =
6534 this_cpu_ptr(&seg6_bpf_srh_states);
6535 int srhoff = 0;
6536
6537 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6538 srh_state->srh = NULL;
6539 } else {
6540 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6541 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6542 srh_state->valid = true;
6543 }
6544 }
6545
BPF_CALL_4(bpf_lwt_seg6_action,struct sk_buff *,skb,u32,action,void *,param,u32,param_len)6546 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6547 u32, action, void *, param, u32, param_len)
6548 {
6549 struct seg6_bpf_srh_state *srh_state =
6550 this_cpu_ptr(&seg6_bpf_srh_states);
6551 int hdroff = 0;
6552 int err;
6553
6554 lockdep_assert_held(&srh_state->bh_lock);
6555 switch (action) {
6556 case SEG6_LOCAL_ACTION_END_X:
6557 if (!seg6_bpf_has_valid_srh(skb))
6558 return -EBADMSG;
6559 if (param_len != sizeof(struct in6_addr))
6560 return -EINVAL;
6561 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6562 case SEG6_LOCAL_ACTION_END_T:
6563 if (!seg6_bpf_has_valid_srh(skb))
6564 return -EBADMSG;
6565 if (param_len != sizeof(int))
6566 return -EINVAL;
6567 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6568 case SEG6_LOCAL_ACTION_END_DT6:
6569 if (!seg6_bpf_has_valid_srh(skb))
6570 return -EBADMSG;
6571 if (param_len != sizeof(int))
6572 return -EINVAL;
6573
6574 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6575 return -EBADMSG;
6576 if (!pskb_pull(skb, hdroff))
6577 return -EBADMSG;
6578
6579 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6580 skb_reset_network_header(skb);
6581 skb_reset_transport_header(skb);
6582 skb->encapsulation = 0;
6583
6584 bpf_compute_data_pointers(skb);
6585 bpf_update_srh_state(skb);
6586 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6587 case SEG6_LOCAL_ACTION_END_B6:
6588 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6589 return -EBADMSG;
6590 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6591 param, param_len);
6592 if (!err)
6593 bpf_update_srh_state(skb);
6594
6595 return err;
6596 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6597 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6598 return -EBADMSG;
6599 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6600 param, param_len);
6601 if (!err)
6602 bpf_update_srh_state(skb);
6603
6604 return err;
6605 default:
6606 return -EINVAL;
6607 }
6608 }
6609
6610 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6611 .func = bpf_lwt_seg6_action,
6612 .gpl_only = false,
6613 .ret_type = RET_INTEGER,
6614 .arg1_type = ARG_PTR_TO_CTX,
6615 .arg2_type = ARG_ANYTHING,
6616 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6617 .arg4_type = ARG_CONST_SIZE
6618 };
6619
BPF_CALL_3(bpf_lwt_seg6_adjust_srh,struct sk_buff *,skb,u32,offset,s32,len)6620 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6621 s32, len)
6622 {
6623 struct seg6_bpf_srh_state *srh_state =
6624 this_cpu_ptr(&seg6_bpf_srh_states);
6625 struct ipv6_sr_hdr *srh = srh_state->srh;
6626 void *srh_end, *srh_tlvs, *ptr;
6627 struct ipv6hdr *hdr;
6628 int srhoff = 0;
6629 int ret;
6630
6631 lockdep_assert_held(&srh_state->bh_lock);
6632 if (unlikely(srh == NULL))
6633 return -EINVAL;
6634
6635 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6636 ((srh->first_segment + 1) << 4));
6637 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6638 srh_state->hdrlen);
6639 ptr = skb->data + offset;
6640
6641 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6642 return -EFAULT;
6643 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6644 return -EFAULT;
6645
6646 if (len > 0) {
6647 ret = skb_cow_head(skb, len);
6648 if (unlikely(ret < 0))
6649 return ret;
6650
6651 ret = bpf_skb_net_hdr_push(skb, offset, len);
6652 } else {
6653 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6654 }
6655
6656 bpf_compute_data_pointers(skb);
6657 if (unlikely(ret < 0))
6658 return ret;
6659
6660 hdr = (struct ipv6hdr *)skb->data;
6661 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6662
6663 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6664 return -EINVAL;
6665 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6666 srh_state->hdrlen += len;
6667 srh_state->valid = false;
6668 return 0;
6669 }
6670
6671 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6672 .func = bpf_lwt_seg6_adjust_srh,
6673 .gpl_only = false,
6674 .ret_type = RET_INTEGER,
6675 .arg1_type = ARG_PTR_TO_CTX,
6676 .arg2_type = ARG_ANYTHING,
6677 .arg3_type = ARG_ANYTHING,
6678 };
6679 #endif /* CONFIG_IPV6_SEG6_BPF */
6680
6681 #ifdef CONFIG_INET
sk_lookup(struct net * net,struct bpf_sock_tuple * tuple,int dif,int sdif,u8 family,u8 proto)6682 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6683 int dif, int sdif, u8 family, u8 proto)
6684 {
6685 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6686 bool refcounted = false;
6687 struct sock *sk = NULL;
6688
6689 if (family == AF_INET) {
6690 __be32 src4 = tuple->ipv4.saddr;
6691 __be32 dst4 = tuple->ipv4.daddr;
6692
6693 if (proto == IPPROTO_TCP)
6694 sk = __inet_lookup(net, hinfo, NULL, 0,
6695 src4, tuple->ipv4.sport,
6696 dst4, tuple->ipv4.dport,
6697 dif, sdif, &refcounted);
6698 else
6699 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6700 dst4, tuple->ipv4.dport,
6701 dif, sdif, net->ipv4.udp_table, NULL);
6702 #if IS_ENABLED(CONFIG_IPV6)
6703 } else {
6704 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6705 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6706
6707 if (proto == IPPROTO_TCP)
6708 sk = __inet6_lookup(net, hinfo, NULL, 0,
6709 src6, tuple->ipv6.sport,
6710 dst6, ntohs(tuple->ipv6.dport),
6711 dif, sdif, &refcounted);
6712 else if (likely(ipv6_bpf_stub))
6713 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6714 src6, tuple->ipv6.sport,
6715 dst6, tuple->ipv6.dport,
6716 dif, sdif,
6717 net->ipv4.udp_table, NULL);
6718 #endif
6719 }
6720
6721 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6722 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6723 sk = NULL;
6724 }
6725 return sk;
6726 }
6727
6728 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6729 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6730 */
6731 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)6732 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6733 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6734 u64 flags, int sdif)
6735 {
6736 struct sock *sk = NULL;
6737 struct net *net;
6738 u8 family;
6739
6740 if (len == sizeof(tuple->ipv4))
6741 family = AF_INET;
6742 else if (len == sizeof(tuple->ipv6))
6743 family = AF_INET6;
6744 else
6745 return NULL;
6746
6747 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6748 goto out;
6749
6750 if (sdif < 0) {
6751 if (family == AF_INET)
6752 sdif = inet_sdif(skb);
6753 else
6754 sdif = inet6_sdif(skb);
6755 }
6756
6757 if ((s32)netns_id < 0) {
6758 net = caller_net;
6759 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6760 } else {
6761 net = get_net_ns_by_id(caller_net, netns_id);
6762 if (unlikely(!net))
6763 goto out;
6764 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6765 put_net(net);
6766 }
6767
6768 out:
6769 return sk;
6770 }
6771
6772 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)6773 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6774 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6775 u64 flags, int sdif)
6776 {
6777 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6778 ifindex, proto, netns_id, flags,
6779 sdif);
6780
6781 if (sk) {
6782 struct sock *sk2 = sk_to_full_sk(sk);
6783
6784 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6785 * sock refcnt is decremented to prevent a request_sock leak.
6786 */
6787 if (sk2 != sk) {
6788 sock_gen_put(sk);
6789 /* Ensure there is no need to bump sk2 refcnt */
6790 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6791 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6792 return NULL;
6793 }
6794 sk = sk2;
6795 }
6796 }
6797
6798 return sk;
6799 }
6800
6801 static struct sock *
bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6802 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6803 u8 proto, u64 netns_id, u64 flags)
6804 {
6805 struct net *caller_net;
6806 int ifindex;
6807
6808 if (skb->dev) {
6809 caller_net = dev_net(skb->dev);
6810 ifindex = skb->dev->ifindex;
6811 } else {
6812 caller_net = sock_net(skb->sk);
6813 ifindex = 0;
6814 }
6815
6816 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6817 netns_id, flags, -1);
6818 }
6819
6820 static struct sock *
bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6821 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6822 u8 proto, u64 netns_id, u64 flags)
6823 {
6824 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6825 flags);
6826
6827 if (sk) {
6828 struct sock *sk2 = sk_to_full_sk(sk);
6829
6830 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6831 * sock refcnt is decremented to prevent a request_sock leak.
6832 */
6833 if (sk2 != sk) {
6834 sock_gen_put(sk);
6835 /* Ensure there is no need to bump sk2 refcnt */
6836 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6837 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6838 return NULL;
6839 }
6840 sk = sk2;
6841 }
6842 }
6843
6844 return sk;
6845 }
6846
BPF_CALL_5(bpf_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6847 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6848 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6849 {
6850 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6851 netns_id, flags);
6852 }
6853
6854 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6855 .func = bpf_skc_lookup_tcp,
6856 .gpl_only = false,
6857 .pkt_access = true,
6858 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6859 .arg1_type = ARG_PTR_TO_CTX,
6860 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6861 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6862 .arg4_type = ARG_ANYTHING,
6863 .arg5_type = ARG_ANYTHING,
6864 };
6865
BPF_CALL_5(bpf_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6866 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6867 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6868 {
6869 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6870 netns_id, flags);
6871 }
6872
6873 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6874 .func = bpf_sk_lookup_tcp,
6875 .gpl_only = false,
6876 .pkt_access = true,
6877 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6878 .arg1_type = ARG_PTR_TO_CTX,
6879 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6880 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6881 .arg4_type = ARG_ANYTHING,
6882 .arg5_type = ARG_ANYTHING,
6883 };
6884
BPF_CALL_5(bpf_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6885 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6886 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6887 {
6888 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6889 netns_id, flags);
6890 }
6891
6892 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6893 .func = bpf_sk_lookup_udp,
6894 .gpl_only = false,
6895 .pkt_access = true,
6896 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6897 .arg1_type = ARG_PTR_TO_CTX,
6898 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6899 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6900 .arg4_type = ARG_ANYTHING,
6901 .arg5_type = ARG_ANYTHING,
6902 };
6903
BPF_CALL_5(bpf_tc_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6904 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6905 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6906 {
6907 struct net_device *dev = skb->dev;
6908 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6909 struct net *caller_net = dev_net(dev);
6910
6911 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6912 ifindex, IPPROTO_TCP, netns_id,
6913 flags, sdif);
6914 }
6915
6916 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6917 .func = bpf_tc_skc_lookup_tcp,
6918 .gpl_only = false,
6919 .pkt_access = true,
6920 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6921 .arg1_type = ARG_PTR_TO_CTX,
6922 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6923 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6924 .arg4_type = ARG_ANYTHING,
6925 .arg5_type = ARG_ANYTHING,
6926 };
6927
BPF_CALL_5(bpf_tc_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6928 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6929 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6930 {
6931 struct net_device *dev = skb->dev;
6932 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6933 struct net *caller_net = dev_net(dev);
6934
6935 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6936 ifindex, IPPROTO_TCP, netns_id,
6937 flags, sdif);
6938 }
6939
6940 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6941 .func = bpf_tc_sk_lookup_tcp,
6942 .gpl_only = false,
6943 .pkt_access = true,
6944 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6945 .arg1_type = ARG_PTR_TO_CTX,
6946 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6947 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6948 .arg4_type = ARG_ANYTHING,
6949 .arg5_type = ARG_ANYTHING,
6950 };
6951
BPF_CALL_5(bpf_tc_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6952 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6953 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6954 {
6955 struct net_device *dev = skb->dev;
6956 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6957 struct net *caller_net = dev_net(dev);
6958
6959 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6960 ifindex, IPPROTO_UDP, netns_id,
6961 flags, sdif);
6962 }
6963
6964 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6965 .func = bpf_tc_sk_lookup_udp,
6966 .gpl_only = false,
6967 .pkt_access = true,
6968 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6969 .arg1_type = ARG_PTR_TO_CTX,
6970 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6971 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6972 .arg4_type = ARG_ANYTHING,
6973 .arg5_type = ARG_ANYTHING,
6974 };
6975
BPF_CALL_1(bpf_sk_release,struct sock *,sk)6976 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6977 {
6978 if (sk && sk_is_refcounted(sk))
6979 sock_gen_put(sk);
6980 return 0;
6981 }
6982
6983 static const struct bpf_func_proto bpf_sk_release_proto = {
6984 .func = bpf_sk_release,
6985 .gpl_only = false,
6986 .ret_type = RET_INTEGER,
6987 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6988 };
6989
BPF_CALL_5(bpf_xdp_sk_lookup_udp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6990 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6991 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6992 {
6993 struct net_device *dev = ctx->rxq->dev;
6994 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6995 struct net *caller_net = dev_net(dev);
6996
6997 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6998 ifindex, IPPROTO_UDP, netns_id,
6999 flags, sdif);
7000 }
7001
7002 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
7003 .func = bpf_xdp_sk_lookup_udp,
7004 .gpl_only = false,
7005 .pkt_access = true,
7006 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7007 .arg1_type = ARG_PTR_TO_CTX,
7008 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7009 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7010 .arg4_type = ARG_ANYTHING,
7011 .arg5_type = ARG_ANYTHING,
7012 };
7013
BPF_CALL_5(bpf_xdp_skc_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7014 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
7015 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7016 {
7017 struct net_device *dev = ctx->rxq->dev;
7018 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7019 struct net *caller_net = dev_net(dev);
7020
7021 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
7022 ifindex, IPPROTO_TCP, netns_id,
7023 flags, sdif);
7024 }
7025
7026 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
7027 .func = bpf_xdp_skc_lookup_tcp,
7028 .gpl_only = false,
7029 .pkt_access = true,
7030 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7031 .arg1_type = ARG_PTR_TO_CTX,
7032 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7033 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7034 .arg4_type = ARG_ANYTHING,
7035 .arg5_type = ARG_ANYTHING,
7036 };
7037
BPF_CALL_5(bpf_xdp_sk_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7038 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
7039 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7040 {
7041 struct net_device *dev = ctx->rxq->dev;
7042 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7043 struct net *caller_net = dev_net(dev);
7044
7045 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
7046 ifindex, IPPROTO_TCP, netns_id,
7047 flags, sdif);
7048 }
7049
7050 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
7051 .func = bpf_xdp_sk_lookup_tcp,
7052 .gpl_only = false,
7053 .pkt_access = true,
7054 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7055 .arg1_type = ARG_PTR_TO_CTX,
7056 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7057 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7058 .arg4_type = ARG_ANYTHING,
7059 .arg5_type = ARG_ANYTHING,
7060 };
7061
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)7062 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7063 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7064 {
7065 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
7066 sock_net(ctx->sk), 0,
7067 IPPROTO_TCP, netns_id, flags,
7068 -1);
7069 }
7070
7071 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
7072 .func = bpf_sock_addr_skc_lookup_tcp,
7073 .gpl_only = false,
7074 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7075 .arg1_type = ARG_PTR_TO_CTX,
7076 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7077 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7078 .arg4_type = ARG_ANYTHING,
7079 .arg5_type = ARG_ANYTHING,
7080 };
7081
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)7082 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7083 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7084 {
7085 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7086 sock_net(ctx->sk), 0, IPPROTO_TCP,
7087 netns_id, flags, -1);
7088 }
7089
7090 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
7091 .func = bpf_sock_addr_sk_lookup_tcp,
7092 .gpl_only = false,
7093 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7094 .arg1_type = ARG_PTR_TO_CTX,
7095 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7096 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7097 .arg4_type = ARG_ANYTHING,
7098 .arg5_type = ARG_ANYTHING,
7099 };
7100
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)7101 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
7102 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7103 {
7104 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7105 sock_net(ctx->sk), 0, IPPROTO_UDP,
7106 netns_id, flags, -1);
7107 }
7108
7109 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
7110 .func = bpf_sock_addr_sk_lookup_udp,
7111 .gpl_only = false,
7112 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7113 .arg1_type = ARG_PTR_TO_CTX,
7114 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7115 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7116 .arg4_type = ARG_ANYTHING,
7117 .arg5_type = ARG_ANYTHING,
7118 };
7119
bpf_tcp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7120 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7121 struct bpf_insn_access_aux *info)
7122 {
7123 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
7124 icsk_retransmits))
7125 return false;
7126
7127 if (off % size != 0)
7128 return false;
7129
7130 switch (off) {
7131 case offsetof(struct bpf_tcp_sock, bytes_received):
7132 case offsetof(struct bpf_tcp_sock, bytes_acked):
7133 return size == sizeof(__u64);
7134 default:
7135 return size == sizeof(__u32);
7136 }
7137 }
7138
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)7139 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
7140 const struct bpf_insn *si,
7141 struct bpf_insn *insn_buf,
7142 struct bpf_prog *prog, u32 *target_size)
7143 {
7144 struct bpf_insn *insn = insn_buf;
7145
7146 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
7147 do { \
7148 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
7149 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7150 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7151 si->dst_reg, si->src_reg, \
7152 offsetof(struct tcp_sock, FIELD)); \
7153 } while (0)
7154
7155 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
7156 do { \
7157 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
7158 FIELD) > \
7159 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7160 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7161 struct inet_connection_sock, \
7162 FIELD), \
7163 si->dst_reg, si->src_reg, \
7164 offsetof( \
7165 struct inet_connection_sock, \
7166 FIELD)); \
7167 } while (0)
7168
7169 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7170
7171 switch (si->off) {
7172 case offsetof(struct bpf_tcp_sock, rtt_min):
7173 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7174 sizeof(struct minmax));
7175 BUILD_BUG_ON(sizeof(struct minmax) <
7176 sizeof(struct minmax_sample));
7177
7178 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7179 offsetof(struct tcp_sock, rtt_min) +
7180 offsetof(struct minmax_sample, v));
7181 break;
7182 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7183 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7184 break;
7185 case offsetof(struct bpf_tcp_sock, srtt_us):
7186 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7187 break;
7188 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7189 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7190 break;
7191 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7192 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7193 break;
7194 case offsetof(struct bpf_tcp_sock, snd_nxt):
7195 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7196 break;
7197 case offsetof(struct bpf_tcp_sock, snd_una):
7198 BPF_TCP_SOCK_GET_COMMON(snd_una);
7199 break;
7200 case offsetof(struct bpf_tcp_sock, mss_cache):
7201 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7202 break;
7203 case offsetof(struct bpf_tcp_sock, ecn_flags):
7204 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7205 break;
7206 case offsetof(struct bpf_tcp_sock, rate_delivered):
7207 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7208 break;
7209 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7210 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7211 break;
7212 case offsetof(struct bpf_tcp_sock, packets_out):
7213 BPF_TCP_SOCK_GET_COMMON(packets_out);
7214 break;
7215 case offsetof(struct bpf_tcp_sock, retrans_out):
7216 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7217 break;
7218 case offsetof(struct bpf_tcp_sock, total_retrans):
7219 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7220 break;
7221 case offsetof(struct bpf_tcp_sock, segs_in):
7222 BPF_TCP_SOCK_GET_COMMON(segs_in);
7223 break;
7224 case offsetof(struct bpf_tcp_sock, data_segs_in):
7225 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7226 break;
7227 case offsetof(struct bpf_tcp_sock, segs_out):
7228 BPF_TCP_SOCK_GET_COMMON(segs_out);
7229 break;
7230 case offsetof(struct bpf_tcp_sock, data_segs_out):
7231 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7232 break;
7233 case offsetof(struct bpf_tcp_sock, lost_out):
7234 BPF_TCP_SOCK_GET_COMMON(lost_out);
7235 break;
7236 case offsetof(struct bpf_tcp_sock, sacked_out):
7237 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7238 break;
7239 case offsetof(struct bpf_tcp_sock, bytes_received):
7240 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7241 break;
7242 case offsetof(struct bpf_tcp_sock, bytes_acked):
7243 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7244 break;
7245 case offsetof(struct bpf_tcp_sock, dsack_dups):
7246 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7247 break;
7248 case offsetof(struct bpf_tcp_sock, delivered):
7249 BPF_TCP_SOCK_GET_COMMON(delivered);
7250 break;
7251 case offsetof(struct bpf_tcp_sock, delivered_ce):
7252 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7253 break;
7254 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7255 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7256 break;
7257 }
7258
7259 return insn - insn_buf;
7260 }
7261
BPF_CALL_1(bpf_tcp_sock,struct sock *,sk)7262 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7263 {
7264 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7265 return (unsigned long)sk;
7266
7267 return (unsigned long)NULL;
7268 }
7269
7270 const struct bpf_func_proto bpf_tcp_sock_proto = {
7271 .func = bpf_tcp_sock,
7272 .gpl_only = false,
7273 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7274 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7275 };
7276
BPF_CALL_1(bpf_get_listener_sock,struct sock *,sk)7277 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7278 {
7279 sk = sk_to_full_sk(sk);
7280
7281 if (sk && sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7282 return (unsigned long)sk;
7283
7284 return (unsigned long)NULL;
7285 }
7286
7287 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7288 .func = bpf_get_listener_sock,
7289 .gpl_only = false,
7290 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7291 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7292 };
7293
BPF_CALL_1(bpf_skb_ecn_set_ce,struct sk_buff *,skb)7294 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7295 {
7296 unsigned int iphdr_len;
7297
7298 switch (skb_protocol(skb, true)) {
7299 case cpu_to_be16(ETH_P_IP):
7300 iphdr_len = sizeof(struct iphdr);
7301 break;
7302 case cpu_to_be16(ETH_P_IPV6):
7303 iphdr_len = sizeof(struct ipv6hdr);
7304 break;
7305 default:
7306 return 0;
7307 }
7308
7309 if (skb_headlen(skb) < iphdr_len)
7310 return 0;
7311
7312 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7313 return 0;
7314
7315 return INET_ECN_set_ce(skb);
7316 }
7317
bpf_xdp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7318 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7319 struct bpf_insn_access_aux *info)
7320 {
7321 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7322 return false;
7323
7324 if (off % size != 0)
7325 return false;
7326
7327 switch (off) {
7328 default:
7329 return size == sizeof(__u32);
7330 }
7331 }
7332
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)7333 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7334 const struct bpf_insn *si,
7335 struct bpf_insn *insn_buf,
7336 struct bpf_prog *prog, u32 *target_size)
7337 {
7338 struct bpf_insn *insn = insn_buf;
7339
7340 #define BPF_XDP_SOCK_GET(FIELD) \
7341 do { \
7342 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7343 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7345 si->dst_reg, si->src_reg, \
7346 offsetof(struct xdp_sock, FIELD)); \
7347 } while (0)
7348
7349 switch (si->off) {
7350 case offsetof(struct bpf_xdp_sock, queue_id):
7351 BPF_XDP_SOCK_GET(queue_id);
7352 break;
7353 }
7354
7355 return insn - insn_buf;
7356 }
7357
7358 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7359 .func = bpf_skb_ecn_set_ce,
7360 .gpl_only = false,
7361 .ret_type = RET_INTEGER,
7362 .arg1_type = ARG_PTR_TO_CTX,
7363 };
7364
BPF_CALL_5(bpf_tcp_check_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7365 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7366 struct tcphdr *, th, u32, th_len)
7367 {
7368 #ifdef CONFIG_SYN_COOKIES
7369 int ret;
7370
7371 if (unlikely(!sk || th_len < sizeof(*th)))
7372 return -EINVAL;
7373
7374 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7375 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7376 return -EINVAL;
7377
7378 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7379 return -EINVAL;
7380
7381 if (!th->ack || th->rst || th->syn)
7382 return -ENOENT;
7383
7384 if (unlikely(iph_len < sizeof(struct iphdr)))
7385 return -EINVAL;
7386
7387 if (tcp_synq_no_recent_overflow(sk))
7388 return -ENOENT;
7389
7390 /* Both struct iphdr and struct ipv6hdr have the version field at the
7391 * same offset so we can cast to the shorter header (struct iphdr).
7392 */
7393 switch (((struct iphdr *)iph)->version) {
7394 case 4:
7395 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7396 return -EINVAL;
7397
7398 ret = __cookie_v4_check((struct iphdr *)iph, th);
7399 break;
7400
7401 #if IS_BUILTIN(CONFIG_IPV6)
7402 case 6:
7403 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7404 return -EINVAL;
7405
7406 if (sk->sk_family != AF_INET6)
7407 return -EINVAL;
7408
7409 ret = __cookie_v6_check((struct ipv6hdr *)iph, th);
7410 break;
7411 #endif /* CONFIG_IPV6 */
7412
7413 default:
7414 return -EPROTONOSUPPORT;
7415 }
7416
7417 if (ret > 0)
7418 return 0;
7419
7420 return -ENOENT;
7421 #else
7422 return -ENOTSUPP;
7423 #endif
7424 }
7425
7426 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7427 .func = bpf_tcp_check_syncookie,
7428 .gpl_only = true,
7429 .pkt_access = true,
7430 .ret_type = RET_INTEGER,
7431 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7432 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7433 .arg3_type = ARG_CONST_SIZE,
7434 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7435 .arg5_type = ARG_CONST_SIZE,
7436 };
7437
BPF_CALL_5(bpf_tcp_gen_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7438 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7439 struct tcphdr *, th, u32, th_len)
7440 {
7441 #ifdef CONFIG_SYN_COOKIES
7442 u32 cookie;
7443 u16 mss;
7444
7445 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7446 return -EINVAL;
7447
7448 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7449 return -EINVAL;
7450
7451 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7452 return -ENOENT;
7453
7454 if (!th->syn || th->ack || th->fin || th->rst)
7455 return -EINVAL;
7456
7457 if (unlikely(iph_len < sizeof(struct iphdr)))
7458 return -EINVAL;
7459
7460 /* Both struct iphdr and struct ipv6hdr have the version field at the
7461 * same offset so we can cast to the shorter header (struct iphdr).
7462 */
7463 switch (((struct iphdr *)iph)->version) {
7464 case 4:
7465 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7466 return -EINVAL;
7467
7468 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7469 break;
7470
7471 #if IS_BUILTIN(CONFIG_IPV6)
7472 case 6:
7473 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7474 return -EINVAL;
7475
7476 if (sk->sk_family != AF_INET6)
7477 return -EINVAL;
7478
7479 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7480 break;
7481 #endif /* CONFIG_IPV6 */
7482
7483 default:
7484 return -EPROTONOSUPPORT;
7485 }
7486 if (mss == 0)
7487 return -ENOENT;
7488
7489 return cookie | ((u64)mss << 32);
7490 #else
7491 return -EOPNOTSUPP;
7492 #endif /* CONFIG_SYN_COOKIES */
7493 }
7494
7495 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7496 .func = bpf_tcp_gen_syncookie,
7497 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7498 .pkt_access = true,
7499 .ret_type = RET_INTEGER,
7500 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7501 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7502 .arg3_type = ARG_CONST_SIZE,
7503 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7504 .arg5_type = ARG_CONST_SIZE,
7505 };
7506
BPF_CALL_3(bpf_sk_assign,struct sk_buff *,skb,struct sock *,sk,u64,flags)7507 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7508 {
7509 if (!sk || flags != 0)
7510 return -EINVAL;
7511 if (!skb_at_tc_ingress(skb))
7512 return -EOPNOTSUPP;
7513 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7514 return -ENETUNREACH;
7515 if (sk_unhashed(sk))
7516 return -EOPNOTSUPP;
7517 if (sk_is_refcounted(sk) &&
7518 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7519 return -ENOENT;
7520
7521 skb_orphan(skb);
7522 skb->sk = sk;
7523 skb->destructor = sock_pfree;
7524
7525 return 0;
7526 }
7527
7528 static const struct bpf_func_proto bpf_sk_assign_proto = {
7529 .func = bpf_sk_assign,
7530 .gpl_only = false,
7531 .ret_type = RET_INTEGER,
7532 .arg1_type = ARG_PTR_TO_CTX,
7533 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7534 .arg3_type = ARG_ANYTHING,
7535 };
7536
bpf_search_tcp_opt(const u8 * op,const u8 * opend,u8 search_kind,const u8 * magic,u8 magic_len,bool * eol)7537 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7538 u8 search_kind, const u8 *magic,
7539 u8 magic_len, bool *eol)
7540 {
7541 u8 kind, kind_len;
7542
7543 *eol = false;
7544
7545 while (op < opend) {
7546 kind = op[0];
7547
7548 if (kind == TCPOPT_EOL) {
7549 *eol = true;
7550 return ERR_PTR(-ENOMSG);
7551 } else if (kind == TCPOPT_NOP) {
7552 op++;
7553 continue;
7554 }
7555
7556 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7557 /* Something is wrong in the received header.
7558 * Follow the TCP stack's tcp_parse_options()
7559 * and just bail here.
7560 */
7561 return ERR_PTR(-EFAULT);
7562
7563 kind_len = op[1];
7564 if (search_kind == kind) {
7565 if (!magic_len)
7566 return op;
7567
7568 if (magic_len > kind_len - 2)
7569 return ERR_PTR(-ENOMSG);
7570
7571 if (!memcmp(&op[2], magic, magic_len))
7572 return op;
7573 }
7574
7575 op += kind_len;
7576 }
7577
7578 return ERR_PTR(-ENOMSG);
7579 }
7580
BPF_CALL_4(bpf_sock_ops_load_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,void *,search_res,u32,len,u64,flags)7581 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7582 void *, search_res, u32, len, u64, flags)
7583 {
7584 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7585 const u8 *op, *opend, *magic, *search = search_res;
7586 u8 search_kind, search_len, copy_len, magic_len;
7587 int ret;
7588
7589 /* 2 byte is the minimal option len except TCPOPT_NOP and
7590 * TCPOPT_EOL which are useless for the bpf prog to learn
7591 * and this helper disallow loading them also.
7592 */
7593 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7594 return -EINVAL;
7595
7596 search_kind = search[0];
7597 search_len = search[1];
7598
7599 if (search_len > len || search_kind == TCPOPT_NOP ||
7600 search_kind == TCPOPT_EOL)
7601 return -EINVAL;
7602
7603 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7604 /* 16 or 32 bit magic. +2 for kind and kind length */
7605 if (search_len != 4 && search_len != 6)
7606 return -EINVAL;
7607 magic = &search[2];
7608 magic_len = search_len - 2;
7609 } else {
7610 if (search_len)
7611 return -EINVAL;
7612 magic = NULL;
7613 magic_len = 0;
7614 }
7615
7616 if (load_syn) {
7617 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7618 if (ret < 0)
7619 return ret;
7620
7621 opend = op + ret;
7622 op += sizeof(struct tcphdr);
7623 } else {
7624 if (!bpf_sock->skb ||
7625 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7626 /* This bpf_sock->op cannot call this helper */
7627 return -EPERM;
7628
7629 opend = bpf_sock->skb_data_end;
7630 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7631 }
7632
7633 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7634 &eol);
7635 if (IS_ERR(op))
7636 return PTR_ERR(op);
7637
7638 copy_len = op[1];
7639 ret = copy_len;
7640 if (copy_len > len) {
7641 ret = -ENOSPC;
7642 copy_len = len;
7643 }
7644
7645 memcpy(search_res, op, copy_len);
7646 return ret;
7647 }
7648
7649 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7650 .func = bpf_sock_ops_load_hdr_opt,
7651 .gpl_only = false,
7652 .ret_type = RET_INTEGER,
7653 .arg1_type = ARG_PTR_TO_CTX,
7654 .arg2_type = ARG_PTR_TO_MEM | MEM_WRITE,
7655 .arg3_type = ARG_CONST_SIZE,
7656 .arg4_type = ARG_ANYTHING,
7657 };
7658
BPF_CALL_4(bpf_sock_ops_store_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,const void *,from,u32,len,u64,flags)7659 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7660 const void *, from, u32, len, u64, flags)
7661 {
7662 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7663 const u8 *op, *new_op, *magic = NULL;
7664 struct sk_buff *skb;
7665 bool eol;
7666
7667 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7668 return -EPERM;
7669
7670 if (len < 2 || flags)
7671 return -EINVAL;
7672
7673 new_op = from;
7674 new_kind = new_op[0];
7675 new_kind_len = new_op[1];
7676
7677 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7678 new_kind == TCPOPT_EOL)
7679 return -EINVAL;
7680
7681 if (new_kind_len > bpf_sock->remaining_opt_len)
7682 return -ENOSPC;
7683
7684 /* 253 is another experimental kind */
7685 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7686 if (new_kind_len < 4)
7687 return -EINVAL;
7688 /* Match for the 2 byte magic also.
7689 * RFC 6994: the magic could be 2 or 4 bytes.
7690 * Hence, matching by 2 byte only is on the
7691 * conservative side but it is the right
7692 * thing to do for the 'search-for-duplication'
7693 * purpose.
7694 */
7695 magic = &new_op[2];
7696 magic_len = 2;
7697 }
7698
7699 /* Check for duplication */
7700 skb = bpf_sock->skb;
7701 op = skb->data + sizeof(struct tcphdr);
7702 opend = bpf_sock->skb_data_end;
7703
7704 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7705 &eol);
7706 if (!IS_ERR(op))
7707 return -EEXIST;
7708
7709 if (PTR_ERR(op) != -ENOMSG)
7710 return PTR_ERR(op);
7711
7712 if (eol)
7713 /* The option has been ended. Treat it as no more
7714 * header option can be written.
7715 */
7716 return -ENOSPC;
7717
7718 /* No duplication found. Store the header option. */
7719 memcpy(opend, from, new_kind_len);
7720
7721 bpf_sock->remaining_opt_len -= new_kind_len;
7722 bpf_sock->skb_data_end += new_kind_len;
7723
7724 return 0;
7725 }
7726
7727 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7728 .func = bpf_sock_ops_store_hdr_opt,
7729 .gpl_only = false,
7730 .ret_type = RET_INTEGER,
7731 .arg1_type = ARG_PTR_TO_CTX,
7732 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7733 .arg3_type = ARG_CONST_SIZE,
7734 .arg4_type = ARG_ANYTHING,
7735 };
7736
BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,u32,len,u64,flags)7737 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7738 u32, len, u64, flags)
7739 {
7740 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7741 return -EPERM;
7742
7743 if (flags || len < 2)
7744 return -EINVAL;
7745
7746 if (len > bpf_sock->remaining_opt_len)
7747 return -ENOSPC;
7748
7749 bpf_sock->remaining_opt_len -= len;
7750
7751 return 0;
7752 }
7753
7754 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7755 .func = bpf_sock_ops_reserve_hdr_opt,
7756 .gpl_only = false,
7757 .ret_type = RET_INTEGER,
7758 .arg1_type = ARG_PTR_TO_CTX,
7759 .arg2_type = ARG_ANYTHING,
7760 .arg3_type = ARG_ANYTHING,
7761 };
7762
BPF_CALL_3(bpf_skb_set_tstamp,struct sk_buff *,skb,u64,tstamp,u32,tstamp_type)7763 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7764 u64, tstamp, u32, tstamp_type)
7765 {
7766 /* skb_clear_delivery_time() is done for inet protocol */
7767 if (skb->protocol != htons(ETH_P_IP) &&
7768 skb->protocol != htons(ETH_P_IPV6))
7769 return -EOPNOTSUPP;
7770
7771 switch (tstamp_type) {
7772 case BPF_SKB_CLOCK_REALTIME:
7773 skb->tstamp = tstamp;
7774 skb->tstamp_type = SKB_CLOCK_REALTIME;
7775 break;
7776 case BPF_SKB_CLOCK_MONOTONIC:
7777 if (!tstamp)
7778 return -EINVAL;
7779 skb->tstamp = tstamp;
7780 skb->tstamp_type = SKB_CLOCK_MONOTONIC;
7781 break;
7782 case BPF_SKB_CLOCK_TAI:
7783 if (!tstamp)
7784 return -EINVAL;
7785 skb->tstamp = tstamp;
7786 skb->tstamp_type = SKB_CLOCK_TAI;
7787 break;
7788 default:
7789 return -EINVAL;
7790 }
7791
7792 return 0;
7793 }
7794
7795 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7796 .func = bpf_skb_set_tstamp,
7797 .gpl_only = false,
7798 .ret_type = RET_INTEGER,
7799 .arg1_type = ARG_PTR_TO_CTX,
7800 .arg2_type = ARG_ANYTHING,
7801 .arg3_type = ARG_ANYTHING,
7802 };
7803
7804 #ifdef CONFIG_SYN_COOKIES
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th,u32,th_len)7805 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7806 struct tcphdr *, th, u32, th_len)
7807 {
7808 u32 cookie;
7809 u16 mss;
7810
7811 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7812 return -EINVAL;
7813
7814 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7815 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7816
7817 return cookie | ((u64)mss << 32);
7818 }
7819
7820 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7821 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7822 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7823 .pkt_access = true,
7824 .ret_type = RET_INTEGER,
7825 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7826 .arg1_size = sizeof(struct iphdr),
7827 .arg2_type = ARG_PTR_TO_MEM,
7828 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7829 };
7830
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th,u32,th_len)7831 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7832 struct tcphdr *, th, u32, th_len)
7833 {
7834 #if IS_BUILTIN(CONFIG_IPV6)
7835 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7836 sizeof(struct ipv6hdr);
7837 u32 cookie;
7838 u16 mss;
7839
7840 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7841 return -EINVAL;
7842
7843 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7844 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7845
7846 return cookie | ((u64)mss << 32);
7847 #else
7848 return -EPROTONOSUPPORT;
7849 #endif
7850 }
7851
7852 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7853 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7854 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7855 .pkt_access = true,
7856 .ret_type = RET_INTEGER,
7857 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7858 .arg1_size = sizeof(struct ipv6hdr),
7859 .arg2_type = ARG_PTR_TO_MEM,
7860 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7861 };
7862
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th)7863 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7864 struct tcphdr *, th)
7865 {
7866 if (__cookie_v4_check(iph, th) > 0)
7867 return 0;
7868
7869 return -EACCES;
7870 }
7871
7872 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7873 .func = bpf_tcp_raw_check_syncookie_ipv4,
7874 .gpl_only = true, /* __cookie_v4_check is GPL */
7875 .pkt_access = true,
7876 .ret_type = RET_INTEGER,
7877 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7878 .arg1_size = sizeof(struct iphdr),
7879 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7880 .arg2_size = sizeof(struct tcphdr),
7881 };
7882
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th)7883 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7884 struct tcphdr *, th)
7885 {
7886 #if IS_BUILTIN(CONFIG_IPV6)
7887 if (__cookie_v6_check(iph, th) > 0)
7888 return 0;
7889
7890 return -EACCES;
7891 #else
7892 return -EPROTONOSUPPORT;
7893 #endif
7894 }
7895
7896 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7897 .func = bpf_tcp_raw_check_syncookie_ipv6,
7898 .gpl_only = true, /* __cookie_v6_check is GPL */
7899 .pkt_access = true,
7900 .ret_type = RET_INTEGER,
7901 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7902 .arg1_size = sizeof(struct ipv6hdr),
7903 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7904 .arg2_size = sizeof(struct tcphdr),
7905 };
7906 #endif /* CONFIG_SYN_COOKIES */
7907
7908 #endif /* CONFIG_INET */
7909
bpf_helper_changes_pkt_data(enum bpf_func_id func_id)7910 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id)
7911 {
7912 switch (func_id) {
7913 case BPF_FUNC_clone_redirect:
7914 case BPF_FUNC_l3_csum_replace:
7915 case BPF_FUNC_l4_csum_replace:
7916 case BPF_FUNC_lwt_push_encap:
7917 case BPF_FUNC_lwt_seg6_action:
7918 case BPF_FUNC_lwt_seg6_adjust_srh:
7919 case BPF_FUNC_lwt_seg6_store_bytes:
7920 case BPF_FUNC_msg_pop_data:
7921 case BPF_FUNC_msg_pull_data:
7922 case BPF_FUNC_msg_push_data:
7923 case BPF_FUNC_skb_adjust_room:
7924 case BPF_FUNC_skb_change_head:
7925 case BPF_FUNC_skb_change_proto:
7926 case BPF_FUNC_skb_change_tail:
7927 case BPF_FUNC_skb_pull_data:
7928 case BPF_FUNC_skb_store_bytes:
7929 case BPF_FUNC_skb_vlan_pop:
7930 case BPF_FUNC_skb_vlan_push:
7931 case BPF_FUNC_store_hdr_opt:
7932 case BPF_FUNC_xdp_adjust_head:
7933 case BPF_FUNC_xdp_adjust_meta:
7934 case BPF_FUNC_xdp_adjust_tail:
7935 /* tail-called program could call any of the above */
7936 case BPF_FUNC_tail_call:
7937 return true;
7938 default:
7939 return false;
7940 }
7941 }
7942
7943 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7944 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7945
7946 static const struct bpf_func_proto *
sock_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7947 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7948 {
7949 const struct bpf_func_proto *func_proto;
7950
7951 func_proto = cgroup_common_func_proto(func_id, prog);
7952 if (func_proto)
7953 return func_proto;
7954
7955 func_proto = cgroup_current_func_proto(func_id, prog);
7956 if (func_proto)
7957 return func_proto;
7958
7959 switch (func_id) {
7960 case BPF_FUNC_get_socket_cookie:
7961 return &bpf_get_socket_cookie_sock_proto;
7962 case BPF_FUNC_get_netns_cookie:
7963 return &bpf_get_netns_cookie_sock_proto;
7964 case BPF_FUNC_perf_event_output:
7965 return &bpf_event_output_data_proto;
7966 case BPF_FUNC_sk_storage_get:
7967 return &bpf_sk_storage_get_cg_sock_proto;
7968 case BPF_FUNC_ktime_get_coarse_ns:
7969 return &bpf_ktime_get_coarse_ns_proto;
7970 default:
7971 return bpf_base_func_proto(func_id, prog);
7972 }
7973 }
7974
7975 static const struct bpf_func_proto *
sock_addr_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7976 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7977 {
7978 const struct bpf_func_proto *func_proto;
7979
7980 func_proto = cgroup_common_func_proto(func_id, prog);
7981 if (func_proto)
7982 return func_proto;
7983
7984 func_proto = cgroup_current_func_proto(func_id, prog);
7985 if (func_proto)
7986 return func_proto;
7987
7988 switch (func_id) {
7989 case BPF_FUNC_bind:
7990 switch (prog->expected_attach_type) {
7991 case BPF_CGROUP_INET4_CONNECT:
7992 case BPF_CGROUP_INET6_CONNECT:
7993 return &bpf_bind_proto;
7994 default:
7995 return NULL;
7996 }
7997 case BPF_FUNC_get_socket_cookie:
7998 return &bpf_get_socket_cookie_sock_addr_proto;
7999 case BPF_FUNC_get_netns_cookie:
8000 return &bpf_get_netns_cookie_sock_addr_proto;
8001 case BPF_FUNC_perf_event_output:
8002 return &bpf_event_output_data_proto;
8003 #ifdef CONFIG_INET
8004 case BPF_FUNC_sk_lookup_tcp:
8005 return &bpf_sock_addr_sk_lookup_tcp_proto;
8006 case BPF_FUNC_sk_lookup_udp:
8007 return &bpf_sock_addr_sk_lookup_udp_proto;
8008 case BPF_FUNC_sk_release:
8009 return &bpf_sk_release_proto;
8010 case BPF_FUNC_skc_lookup_tcp:
8011 return &bpf_sock_addr_skc_lookup_tcp_proto;
8012 #endif /* CONFIG_INET */
8013 case BPF_FUNC_sk_storage_get:
8014 return &bpf_sk_storage_get_proto;
8015 case BPF_FUNC_sk_storage_delete:
8016 return &bpf_sk_storage_delete_proto;
8017 case BPF_FUNC_setsockopt:
8018 switch (prog->expected_attach_type) {
8019 case BPF_CGROUP_INET4_BIND:
8020 case BPF_CGROUP_INET6_BIND:
8021 case BPF_CGROUP_INET4_CONNECT:
8022 case BPF_CGROUP_INET6_CONNECT:
8023 case BPF_CGROUP_UNIX_CONNECT:
8024 case BPF_CGROUP_UDP4_RECVMSG:
8025 case BPF_CGROUP_UDP6_RECVMSG:
8026 case BPF_CGROUP_UNIX_RECVMSG:
8027 case BPF_CGROUP_UDP4_SENDMSG:
8028 case BPF_CGROUP_UDP6_SENDMSG:
8029 case BPF_CGROUP_UNIX_SENDMSG:
8030 case BPF_CGROUP_INET4_GETPEERNAME:
8031 case BPF_CGROUP_INET6_GETPEERNAME:
8032 case BPF_CGROUP_UNIX_GETPEERNAME:
8033 case BPF_CGROUP_INET4_GETSOCKNAME:
8034 case BPF_CGROUP_INET6_GETSOCKNAME:
8035 case BPF_CGROUP_UNIX_GETSOCKNAME:
8036 return &bpf_sock_addr_setsockopt_proto;
8037 default:
8038 return NULL;
8039 }
8040 case BPF_FUNC_getsockopt:
8041 switch (prog->expected_attach_type) {
8042 case BPF_CGROUP_INET4_BIND:
8043 case BPF_CGROUP_INET6_BIND:
8044 case BPF_CGROUP_INET4_CONNECT:
8045 case BPF_CGROUP_INET6_CONNECT:
8046 case BPF_CGROUP_UNIX_CONNECT:
8047 case BPF_CGROUP_UDP4_RECVMSG:
8048 case BPF_CGROUP_UDP6_RECVMSG:
8049 case BPF_CGROUP_UNIX_RECVMSG:
8050 case BPF_CGROUP_UDP4_SENDMSG:
8051 case BPF_CGROUP_UDP6_SENDMSG:
8052 case BPF_CGROUP_UNIX_SENDMSG:
8053 case BPF_CGROUP_INET4_GETPEERNAME:
8054 case BPF_CGROUP_INET6_GETPEERNAME:
8055 case BPF_CGROUP_UNIX_GETPEERNAME:
8056 case BPF_CGROUP_INET4_GETSOCKNAME:
8057 case BPF_CGROUP_INET6_GETSOCKNAME:
8058 case BPF_CGROUP_UNIX_GETSOCKNAME:
8059 return &bpf_sock_addr_getsockopt_proto;
8060 default:
8061 return NULL;
8062 }
8063 default:
8064 return bpf_sk_base_func_proto(func_id, prog);
8065 }
8066 }
8067
8068 static const struct bpf_func_proto *
sk_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8069 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8070 {
8071 switch (func_id) {
8072 case BPF_FUNC_skb_load_bytes:
8073 return &bpf_skb_load_bytes_proto;
8074 case BPF_FUNC_skb_load_bytes_relative:
8075 return &bpf_skb_load_bytes_relative_proto;
8076 case BPF_FUNC_get_socket_cookie:
8077 return &bpf_get_socket_cookie_proto;
8078 case BPF_FUNC_get_socket_uid:
8079 return &bpf_get_socket_uid_proto;
8080 case BPF_FUNC_perf_event_output:
8081 return &bpf_skb_event_output_proto;
8082 default:
8083 return bpf_sk_base_func_proto(func_id, prog);
8084 }
8085 }
8086
8087 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
8088 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
8089
8090 static const struct bpf_func_proto *
cg_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8091 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8092 {
8093 const struct bpf_func_proto *func_proto;
8094
8095 func_proto = cgroup_common_func_proto(func_id, prog);
8096 if (func_proto)
8097 return func_proto;
8098
8099 switch (func_id) {
8100 case BPF_FUNC_sk_fullsock:
8101 return &bpf_sk_fullsock_proto;
8102 case BPF_FUNC_sk_storage_get:
8103 return &bpf_sk_storage_get_proto;
8104 case BPF_FUNC_sk_storage_delete:
8105 return &bpf_sk_storage_delete_proto;
8106 case BPF_FUNC_perf_event_output:
8107 return &bpf_skb_event_output_proto;
8108 #ifdef CONFIG_SOCK_CGROUP_DATA
8109 case BPF_FUNC_skb_cgroup_id:
8110 return &bpf_skb_cgroup_id_proto;
8111 case BPF_FUNC_skb_ancestor_cgroup_id:
8112 return &bpf_skb_ancestor_cgroup_id_proto;
8113 case BPF_FUNC_sk_cgroup_id:
8114 return &bpf_sk_cgroup_id_proto;
8115 case BPF_FUNC_sk_ancestor_cgroup_id:
8116 return &bpf_sk_ancestor_cgroup_id_proto;
8117 #endif
8118 #ifdef CONFIG_INET
8119 case BPF_FUNC_sk_lookup_tcp:
8120 return &bpf_sk_lookup_tcp_proto;
8121 case BPF_FUNC_sk_lookup_udp:
8122 return &bpf_sk_lookup_udp_proto;
8123 case BPF_FUNC_sk_release:
8124 return &bpf_sk_release_proto;
8125 case BPF_FUNC_skc_lookup_tcp:
8126 return &bpf_skc_lookup_tcp_proto;
8127 case BPF_FUNC_tcp_sock:
8128 return &bpf_tcp_sock_proto;
8129 case BPF_FUNC_get_listener_sock:
8130 return &bpf_get_listener_sock_proto;
8131 case BPF_FUNC_skb_ecn_set_ce:
8132 return &bpf_skb_ecn_set_ce_proto;
8133 #endif
8134 default:
8135 return sk_filter_func_proto(func_id, prog);
8136 }
8137 }
8138
8139 static const struct bpf_func_proto *
tc_cls_act_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8140 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8141 {
8142 switch (func_id) {
8143 case BPF_FUNC_skb_store_bytes:
8144 return &bpf_skb_store_bytes_proto;
8145 case BPF_FUNC_skb_load_bytes:
8146 return &bpf_skb_load_bytes_proto;
8147 case BPF_FUNC_skb_load_bytes_relative:
8148 return &bpf_skb_load_bytes_relative_proto;
8149 case BPF_FUNC_skb_pull_data:
8150 return &bpf_skb_pull_data_proto;
8151 case BPF_FUNC_csum_diff:
8152 return &bpf_csum_diff_proto;
8153 case BPF_FUNC_csum_update:
8154 return &bpf_csum_update_proto;
8155 case BPF_FUNC_csum_level:
8156 return &bpf_csum_level_proto;
8157 case BPF_FUNC_l3_csum_replace:
8158 return &bpf_l3_csum_replace_proto;
8159 case BPF_FUNC_l4_csum_replace:
8160 return &bpf_l4_csum_replace_proto;
8161 case BPF_FUNC_clone_redirect:
8162 return &bpf_clone_redirect_proto;
8163 case BPF_FUNC_get_cgroup_classid:
8164 return &bpf_get_cgroup_classid_proto;
8165 case BPF_FUNC_skb_vlan_push:
8166 return &bpf_skb_vlan_push_proto;
8167 case BPF_FUNC_skb_vlan_pop:
8168 return &bpf_skb_vlan_pop_proto;
8169 case BPF_FUNC_skb_change_proto:
8170 return &bpf_skb_change_proto_proto;
8171 case BPF_FUNC_skb_change_type:
8172 return &bpf_skb_change_type_proto;
8173 case BPF_FUNC_skb_adjust_room:
8174 return &bpf_skb_adjust_room_proto;
8175 case BPF_FUNC_skb_change_tail:
8176 return &bpf_skb_change_tail_proto;
8177 case BPF_FUNC_skb_change_head:
8178 return &bpf_skb_change_head_proto;
8179 case BPF_FUNC_skb_get_tunnel_key:
8180 return &bpf_skb_get_tunnel_key_proto;
8181 case BPF_FUNC_skb_set_tunnel_key:
8182 return bpf_get_skb_set_tunnel_proto(func_id);
8183 case BPF_FUNC_skb_get_tunnel_opt:
8184 return &bpf_skb_get_tunnel_opt_proto;
8185 case BPF_FUNC_skb_set_tunnel_opt:
8186 return bpf_get_skb_set_tunnel_proto(func_id);
8187 case BPF_FUNC_redirect:
8188 return &bpf_redirect_proto;
8189 case BPF_FUNC_redirect_neigh:
8190 return &bpf_redirect_neigh_proto;
8191 case BPF_FUNC_redirect_peer:
8192 return &bpf_redirect_peer_proto;
8193 case BPF_FUNC_get_route_realm:
8194 return &bpf_get_route_realm_proto;
8195 case BPF_FUNC_get_hash_recalc:
8196 return &bpf_get_hash_recalc_proto;
8197 case BPF_FUNC_set_hash_invalid:
8198 return &bpf_set_hash_invalid_proto;
8199 case BPF_FUNC_set_hash:
8200 return &bpf_set_hash_proto;
8201 case BPF_FUNC_perf_event_output:
8202 return &bpf_skb_event_output_proto;
8203 case BPF_FUNC_get_smp_processor_id:
8204 return &bpf_get_smp_processor_id_proto;
8205 case BPF_FUNC_skb_under_cgroup:
8206 return &bpf_skb_under_cgroup_proto;
8207 case BPF_FUNC_get_socket_cookie:
8208 return &bpf_get_socket_cookie_proto;
8209 case BPF_FUNC_get_netns_cookie:
8210 return &bpf_get_netns_cookie_proto;
8211 case BPF_FUNC_get_socket_uid:
8212 return &bpf_get_socket_uid_proto;
8213 case BPF_FUNC_fib_lookup:
8214 return &bpf_skb_fib_lookup_proto;
8215 case BPF_FUNC_check_mtu:
8216 return &bpf_skb_check_mtu_proto;
8217 case BPF_FUNC_sk_fullsock:
8218 return &bpf_sk_fullsock_proto;
8219 case BPF_FUNC_sk_storage_get:
8220 return &bpf_sk_storage_get_proto;
8221 case BPF_FUNC_sk_storage_delete:
8222 return &bpf_sk_storage_delete_proto;
8223 #ifdef CONFIG_XFRM
8224 case BPF_FUNC_skb_get_xfrm_state:
8225 return &bpf_skb_get_xfrm_state_proto;
8226 #endif
8227 #ifdef CONFIG_CGROUP_NET_CLASSID
8228 case BPF_FUNC_skb_cgroup_classid:
8229 return &bpf_skb_cgroup_classid_proto;
8230 #endif
8231 #ifdef CONFIG_SOCK_CGROUP_DATA
8232 case BPF_FUNC_skb_cgroup_id:
8233 return &bpf_skb_cgroup_id_proto;
8234 case BPF_FUNC_skb_ancestor_cgroup_id:
8235 return &bpf_skb_ancestor_cgroup_id_proto;
8236 #endif
8237 #ifdef CONFIG_INET
8238 case BPF_FUNC_sk_lookup_tcp:
8239 return &bpf_tc_sk_lookup_tcp_proto;
8240 case BPF_FUNC_sk_lookup_udp:
8241 return &bpf_tc_sk_lookup_udp_proto;
8242 case BPF_FUNC_sk_release:
8243 return &bpf_sk_release_proto;
8244 case BPF_FUNC_tcp_sock:
8245 return &bpf_tcp_sock_proto;
8246 case BPF_FUNC_get_listener_sock:
8247 return &bpf_get_listener_sock_proto;
8248 case BPF_FUNC_skc_lookup_tcp:
8249 return &bpf_tc_skc_lookup_tcp_proto;
8250 case BPF_FUNC_tcp_check_syncookie:
8251 return &bpf_tcp_check_syncookie_proto;
8252 case BPF_FUNC_skb_ecn_set_ce:
8253 return &bpf_skb_ecn_set_ce_proto;
8254 case BPF_FUNC_tcp_gen_syncookie:
8255 return &bpf_tcp_gen_syncookie_proto;
8256 case BPF_FUNC_sk_assign:
8257 return &bpf_sk_assign_proto;
8258 case BPF_FUNC_skb_set_tstamp:
8259 return &bpf_skb_set_tstamp_proto;
8260 #ifdef CONFIG_SYN_COOKIES
8261 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8262 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8263 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8264 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8265 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8266 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8267 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8268 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8269 #endif
8270 #endif
8271 default:
8272 return bpf_sk_base_func_proto(func_id, prog);
8273 }
8274 }
8275
8276 static const struct bpf_func_proto *
xdp_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8277 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8278 {
8279 switch (func_id) {
8280 case BPF_FUNC_perf_event_output:
8281 return &bpf_xdp_event_output_proto;
8282 case BPF_FUNC_get_smp_processor_id:
8283 return &bpf_get_smp_processor_id_proto;
8284 case BPF_FUNC_csum_diff:
8285 return &bpf_csum_diff_proto;
8286 case BPF_FUNC_xdp_adjust_head:
8287 return &bpf_xdp_adjust_head_proto;
8288 case BPF_FUNC_xdp_adjust_meta:
8289 return &bpf_xdp_adjust_meta_proto;
8290 case BPF_FUNC_redirect:
8291 return &bpf_xdp_redirect_proto;
8292 case BPF_FUNC_redirect_map:
8293 return &bpf_xdp_redirect_map_proto;
8294 case BPF_FUNC_xdp_adjust_tail:
8295 return &bpf_xdp_adjust_tail_proto;
8296 case BPF_FUNC_xdp_get_buff_len:
8297 return &bpf_xdp_get_buff_len_proto;
8298 case BPF_FUNC_xdp_load_bytes:
8299 return &bpf_xdp_load_bytes_proto;
8300 case BPF_FUNC_xdp_store_bytes:
8301 return &bpf_xdp_store_bytes_proto;
8302 case BPF_FUNC_fib_lookup:
8303 return &bpf_xdp_fib_lookup_proto;
8304 case BPF_FUNC_check_mtu:
8305 return &bpf_xdp_check_mtu_proto;
8306 #ifdef CONFIG_INET
8307 case BPF_FUNC_sk_lookup_udp:
8308 return &bpf_xdp_sk_lookup_udp_proto;
8309 case BPF_FUNC_sk_lookup_tcp:
8310 return &bpf_xdp_sk_lookup_tcp_proto;
8311 case BPF_FUNC_sk_release:
8312 return &bpf_sk_release_proto;
8313 case BPF_FUNC_skc_lookup_tcp:
8314 return &bpf_xdp_skc_lookup_tcp_proto;
8315 case BPF_FUNC_tcp_check_syncookie:
8316 return &bpf_tcp_check_syncookie_proto;
8317 case BPF_FUNC_tcp_gen_syncookie:
8318 return &bpf_tcp_gen_syncookie_proto;
8319 #ifdef CONFIG_SYN_COOKIES
8320 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8321 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8322 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8323 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8324 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8325 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8326 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8327 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8328 #endif
8329 #endif
8330 default:
8331 return bpf_sk_base_func_proto(func_id, prog);
8332 }
8333
8334 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8335 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8336 * kfuncs are defined in two different modules, and we want to be able
8337 * to use them interchangeably with the same BTF type ID. Because modules
8338 * can't de-duplicate BTF IDs between each other, we need the type to be
8339 * referenced in the vmlinux BTF or the verifier will get confused about
8340 * the different types. So we add this dummy type reference which will
8341 * be included in vmlinux BTF, allowing both modules to refer to the
8342 * same type ID.
8343 */
8344 BTF_TYPE_EMIT(struct nf_conn___init);
8345 #endif
8346 }
8347
8348 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8349 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8350
8351 static const struct bpf_func_proto *
sock_ops_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8352 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8353 {
8354 const struct bpf_func_proto *func_proto;
8355
8356 func_proto = cgroup_common_func_proto(func_id, prog);
8357 if (func_proto)
8358 return func_proto;
8359
8360 switch (func_id) {
8361 case BPF_FUNC_setsockopt:
8362 return &bpf_sock_ops_setsockopt_proto;
8363 case BPF_FUNC_getsockopt:
8364 return &bpf_sock_ops_getsockopt_proto;
8365 case BPF_FUNC_sock_ops_cb_flags_set:
8366 return &bpf_sock_ops_cb_flags_set_proto;
8367 case BPF_FUNC_sock_map_update:
8368 return &bpf_sock_map_update_proto;
8369 case BPF_FUNC_sock_hash_update:
8370 return &bpf_sock_hash_update_proto;
8371 case BPF_FUNC_get_socket_cookie:
8372 return &bpf_get_socket_cookie_sock_ops_proto;
8373 case BPF_FUNC_perf_event_output:
8374 return &bpf_event_output_data_proto;
8375 case BPF_FUNC_sk_storage_get:
8376 return &bpf_sk_storage_get_proto;
8377 case BPF_FUNC_sk_storage_delete:
8378 return &bpf_sk_storage_delete_proto;
8379 case BPF_FUNC_get_netns_cookie:
8380 return &bpf_get_netns_cookie_sock_ops_proto;
8381 #ifdef CONFIG_INET
8382 case BPF_FUNC_load_hdr_opt:
8383 return &bpf_sock_ops_load_hdr_opt_proto;
8384 case BPF_FUNC_store_hdr_opt:
8385 return &bpf_sock_ops_store_hdr_opt_proto;
8386 case BPF_FUNC_reserve_hdr_opt:
8387 return &bpf_sock_ops_reserve_hdr_opt_proto;
8388 case BPF_FUNC_tcp_sock:
8389 return &bpf_tcp_sock_proto;
8390 #endif /* CONFIG_INET */
8391 default:
8392 return bpf_sk_base_func_proto(func_id, prog);
8393 }
8394 }
8395
8396 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8397 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8398
8399 static const struct bpf_func_proto *
sk_msg_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8400 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8401 {
8402 switch (func_id) {
8403 case BPF_FUNC_msg_redirect_map:
8404 return &bpf_msg_redirect_map_proto;
8405 case BPF_FUNC_msg_redirect_hash:
8406 return &bpf_msg_redirect_hash_proto;
8407 case BPF_FUNC_msg_apply_bytes:
8408 return &bpf_msg_apply_bytes_proto;
8409 case BPF_FUNC_msg_cork_bytes:
8410 return &bpf_msg_cork_bytes_proto;
8411 case BPF_FUNC_msg_pull_data:
8412 return &bpf_msg_pull_data_proto;
8413 case BPF_FUNC_msg_push_data:
8414 return &bpf_msg_push_data_proto;
8415 case BPF_FUNC_msg_pop_data:
8416 return &bpf_msg_pop_data_proto;
8417 case BPF_FUNC_perf_event_output:
8418 return &bpf_event_output_data_proto;
8419 case BPF_FUNC_get_current_uid_gid:
8420 return &bpf_get_current_uid_gid_proto;
8421 case BPF_FUNC_sk_storage_get:
8422 return &bpf_sk_storage_get_proto;
8423 case BPF_FUNC_sk_storage_delete:
8424 return &bpf_sk_storage_delete_proto;
8425 case BPF_FUNC_get_netns_cookie:
8426 return &bpf_get_netns_cookie_sk_msg_proto;
8427 #ifdef CONFIG_CGROUP_NET_CLASSID
8428 case BPF_FUNC_get_cgroup_classid:
8429 return &bpf_get_cgroup_classid_curr_proto;
8430 #endif
8431 default:
8432 return bpf_sk_base_func_proto(func_id, prog);
8433 }
8434 }
8435
8436 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8437 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8438
8439 static const struct bpf_func_proto *
sk_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8440 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8441 {
8442 switch (func_id) {
8443 case BPF_FUNC_skb_store_bytes:
8444 return &bpf_skb_store_bytes_proto;
8445 case BPF_FUNC_skb_load_bytes:
8446 return &bpf_skb_load_bytes_proto;
8447 case BPF_FUNC_skb_pull_data:
8448 return &sk_skb_pull_data_proto;
8449 case BPF_FUNC_skb_change_tail:
8450 return &sk_skb_change_tail_proto;
8451 case BPF_FUNC_skb_change_head:
8452 return &sk_skb_change_head_proto;
8453 case BPF_FUNC_skb_adjust_room:
8454 return &sk_skb_adjust_room_proto;
8455 case BPF_FUNC_get_socket_cookie:
8456 return &bpf_get_socket_cookie_proto;
8457 case BPF_FUNC_get_socket_uid:
8458 return &bpf_get_socket_uid_proto;
8459 case BPF_FUNC_sk_redirect_map:
8460 return &bpf_sk_redirect_map_proto;
8461 case BPF_FUNC_sk_redirect_hash:
8462 return &bpf_sk_redirect_hash_proto;
8463 case BPF_FUNC_perf_event_output:
8464 return &bpf_skb_event_output_proto;
8465 #ifdef CONFIG_INET
8466 case BPF_FUNC_sk_lookup_tcp:
8467 return &bpf_sk_lookup_tcp_proto;
8468 case BPF_FUNC_sk_lookup_udp:
8469 return &bpf_sk_lookup_udp_proto;
8470 case BPF_FUNC_sk_release:
8471 return &bpf_sk_release_proto;
8472 case BPF_FUNC_skc_lookup_tcp:
8473 return &bpf_skc_lookup_tcp_proto;
8474 #endif
8475 default:
8476 return bpf_sk_base_func_proto(func_id, prog);
8477 }
8478 }
8479
8480 static const struct bpf_func_proto *
flow_dissector_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8481 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8482 {
8483 switch (func_id) {
8484 case BPF_FUNC_skb_load_bytes:
8485 return &bpf_flow_dissector_load_bytes_proto;
8486 default:
8487 return bpf_sk_base_func_proto(func_id, prog);
8488 }
8489 }
8490
8491 static const struct bpf_func_proto *
lwt_out_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8492 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8493 {
8494 switch (func_id) {
8495 case BPF_FUNC_skb_load_bytes:
8496 return &bpf_skb_load_bytes_proto;
8497 case BPF_FUNC_skb_pull_data:
8498 return &bpf_skb_pull_data_proto;
8499 case BPF_FUNC_csum_diff:
8500 return &bpf_csum_diff_proto;
8501 case BPF_FUNC_get_cgroup_classid:
8502 return &bpf_get_cgroup_classid_proto;
8503 case BPF_FUNC_get_route_realm:
8504 return &bpf_get_route_realm_proto;
8505 case BPF_FUNC_get_hash_recalc:
8506 return &bpf_get_hash_recalc_proto;
8507 case BPF_FUNC_perf_event_output:
8508 return &bpf_skb_event_output_proto;
8509 case BPF_FUNC_get_smp_processor_id:
8510 return &bpf_get_smp_processor_id_proto;
8511 case BPF_FUNC_skb_under_cgroup:
8512 return &bpf_skb_under_cgroup_proto;
8513 default:
8514 return bpf_sk_base_func_proto(func_id, prog);
8515 }
8516 }
8517
8518 static const struct bpf_func_proto *
lwt_in_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8519 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8520 {
8521 switch (func_id) {
8522 case BPF_FUNC_lwt_push_encap:
8523 return &bpf_lwt_in_push_encap_proto;
8524 default:
8525 return lwt_out_func_proto(func_id, prog);
8526 }
8527 }
8528
8529 static const struct bpf_func_proto *
lwt_xmit_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8530 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8531 {
8532 switch (func_id) {
8533 case BPF_FUNC_skb_get_tunnel_key:
8534 return &bpf_skb_get_tunnel_key_proto;
8535 case BPF_FUNC_skb_set_tunnel_key:
8536 return bpf_get_skb_set_tunnel_proto(func_id);
8537 case BPF_FUNC_skb_get_tunnel_opt:
8538 return &bpf_skb_get_tunnel_opt_proto;
8539 case BPF_FUNC_skb_set_tunnel_opt:
8540 return bpf_get_skb_set_tunnel_proto(func_id);
8541 case BPF_FUNC_redirect:
8542 return &bpf_redirect_proto;
8543 case BPF_FUNC_clone_redirect:
8544 return &bpf_clone_redirect_proto;
8545 case BPF_FUNC_skb_change_tail:
8546 return &bpf_skb_change_tail_proto;
8547 case BPF_FUNC_skb_change_head:
8548 return &bpf_skb_change_head_proto;
8549 case BPF_FUNC_skb_store_bytes:
8550 return &bpf_skb_store_bytes_proto;
8551 case BPF_FUNC_csum_update:
8552 return &bpf_csum_update_proto;
8553 case BPF_FUNC_csum_level:
8554 return &bpf_csum_level_proto;
8555 case BPF_FUNC_l3_csum_replace:
8556 return &bpf_l3_csum_replace_proto;
8557 case BPF_FUNC_l4_csum_replace:
8558 return &bpf_l4_csum_replace_proto;
8559 case BPF_FUNC_set_hash_invalid:
8560 return &bpf_set_hash_invalid_proto;
8561 case BPF_FUNC_lwt_push_encap:
8562 return &bpf_lwt_xmit_push_encap_proto;
8563 default:
8564 return lwt_out_func_proto(func_id, prog);
8565 }
8566 }
8567
8568 static const struct bpf_func_proto *
lwt_seg6local_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8569 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8570 {
8571 switch (func_id) {
8572 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8573 case BPF_FUNC_lwt_seg6_store_bytes:
8574 return &bpf_lwt_seg6_store_bytes_proto;
8575 case BPF_FUNC_lwt_seg6_action:
8576 return &bpf_lwt_seg6_action_proto;
8577 case BPF_FUNC_lwt_seg6_adjust_srh:
8578 return &bpf_lwt_seg6_adjust_srh_proto;
8579 #endif
8580 default:
8581 return lwt_out_func_proto(func_id, prog);
8582 }
8583 }
8584
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)8585 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8586 const struct bpf_prog *prog,
8587 struct bpf_insn_access_aux *info)
8588 {
8589 const int size_default = sizeof(__u32);
8590
8591 if (off < 0 || off >= sizeof(struct __sk_buff))
8592 return false;
8593
8594 /* The verifier guarantees that size > 0. */
8595 if (off % size != 0)
8596 return false;
8597
8598 switch (off) {
8599 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8600 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8601 return false;
8602 break;
8603 case bpf_ctx_range(struct __sk_buff, data):
8604 case bpf_ctx_range(struct __sk_buff, data_meta):
8605 case bpf_ctx_range(struct __sk_buff, data_end):
8606 if (info->is_ldsx || size != size_default)
8607 return false;
8608 break;
8609 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8610 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8611 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8612 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8613 if (size != size_default)
8614 return false;
8615 break;
8616 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8617 return false;
8618 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8619 if (type == BPF_WRITE || size != sizeof(__u64))
8620 return false;
8621 break;
8622 case bpf_ctx_range(struct __sk_buff, tstamp):
8623 if (size != sizeof(__u64))
8624 return false;
8625 break;
8626 case offsetof(struct __sk_buff, sk):
8627 if (type == BPF_WRITE || size != sizeof(__u64))
8628 return false;
8629 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8630 break;
8631 case offsetof(struct __sk_buff, tstamp_type):
8632 return false;
8633 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8634 /* Explicitly prohibit access to padding in __sk_buff. */
8635 return false;
8636 default:
8637 /* Only narrow read access allowed for now. */
8638 if (type == BPF_WRITE) {
8639 if (size != size_default)
8640 return false;
8641 } else {
8642 bpf_ctx_record_field_size(info, size_default);
8643 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8644 return false;
8645 }
8646 }
8647
8648 return true;
8649 }
8650
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)8651 static bool sk_filter_is_valid_access(int off, int size,
8652 enum bpf_access_type type,
8653 const struct bpf_prog *prog,
8654 struct bpf_insn_access_aux *info)
8655 {
8656 switch (off) {
8657 case bpf_ctx_range(struct __sk_buff, tc_classid):
8658 case bpf_ctx_range(struct __sk_buff, data):
8659 case bpf_ctx_range(struct __sk_buff, data_meta):
8660 case bpf_ctx_range(struct __sk_buff, data_end):
8661 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8662 case bpf_ctx_range(struct __sk_buff, tstamp):
8663 case bpf_ctx_range(struct __sk_buff, wire_len):
8664 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8665 return false;
8666 }
8667
8668 if (type == BPF_WRITE) {
8669 switch (off) {
8670 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8671 break;
8672 default:
8673 return false;
8674 }
8675 }
8676
8677 return bpf_skb_is_valid_access(off, size, type, prog, info);
8678 }
8679
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)8680 static bool cg_skb_is_valid_access(int off, int size,
8681 enum bpf_access_type type,
8682 const struct bpf_prog *prog,
8683 struct bpf_insn_access_aux *info)
8684 {
8685 switch (off) {
8686 case bpf_ctx_range(struct __sk_buff, tc_classid):
8687 case bpf_ctx_range(struct __sk_buff, data_meta):
8688 case bpf_ctx_range(struct __sk_buff, wire_len):
8689 return false;
8690 case bpf_ctx_range(struct __sk_buff, data):
8691 case bpf_ctx_range(struct __sk_buff, data_end):
8692 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8693 return false;
8694 break;
8695 }
8696
8697 if (type == BPF_WRITE) {
8698 switch (off) {
8699 case bpf_ctx_range(struct __sk_buff, mark):
8700 case bpf_ctx_range(struct __sk_buff, priority):
8701 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8702 break;
8703 case bpf_ctx_range(struct __sk_buff, tstamp):
8704 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8705 return false;
8706 break;
8707 default:
8708 return false;
8709 }
8710 }
8711
8712 switch (off) {
8713 case bpf_ctx_range(struct __sk_buff, data):
8714 info->reg_type = PTR_TO_PACKET;
8715 break;
8716 case bpf_ctx_range(struct __sk_buff, data_end):
8717 info->reg_type = PTR_TO_PACKET_END;
8718 break;
8719 }
8720
8721 return bpf_skb_is_valid_access(off, size, type, prog, info);
8722 }
8723
lwt_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8724 static bool lwt_is_valid_access(int off, int size,
8725 enum bpf_access_type type,
8726 const struct bpf_prog *prog,
8727 struct bpf_insn_access_aux *info)
8728 {
8729 switch (off) {
8730 case bpf_ctx_range(struct __sk_buff, tc_classid):
8731 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8732 case bpf_ctx_range(struct __sk_buff, data_meta):
8733 case bpf_ctx_range(struct __sk_buff, tstamp):
8734 case bpf_ctx_range(struct __sk_buff, wire_len):
8735 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8736 return false;
8737 }
8738
8739 if (type == BPF_WRITE) {
8740 switch (off) {
8741 case bpf_ctx_range(struct __sk_buff, mark):
8742 case bpf_ctx_range(struct __sk_buff, priority):
8743 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8744 break;
8745 default:
8746 return false;
8747 }
8748 }
8749
8750 switch (off) {
8751 case bpf_ctx_range(struct __sk_buff, data):
8752 info->reg_type = PTR_TO_PACKET;
8753 break;
8754 case bpf_ctx_range(struct __sk_buff, data_end):
8755 info->reg_type = PTR_TO_PACKET_END;
8756 break;
8757 }
8758
8759 return bpf_skb_is_valid_access(off, size, type, prog, info);
8760 }
8761
8762 /* Attach type specific accesses */
__sock_filter_check_attach_type(int off,enum bpf_access_type access_type,enum bpf_attach_type attach_type)8763 static bool __sock_filter_check_attach_type(int off,
8764 enum bpf_access_type access_type,
8765 enum bpf_attach_type attach_type)
8766 {
8767 switch (off) {
8768 case offsetof(struct bpf_sock, bound_dev_if):
8769 case offsetof(struct bpf_sock, mark):
8770 case offsetof(struct bpf_sock, priority):
8771 switch (attach_type) {
8772 case BPF_CGROUP_INET_SOCK_CREATE:
8773 case BPF_CGROUP_INET_SOCK_RELEASE:
8774 goto full_access;
8775 default:
8776 return false;
8777 }
8778 case bpf_ctx_range(struct bpf_sock, src_ip4):
8779 switch (attach_type) {
8780 case BPF_CGROUP_INET4_POST_BIND:
8781 goto read_only;
8782 default:
8783 return false;
8784 }
8785 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8786 switch (attach_type) {
8787 case BPF_CGROUP_INET6_POST_BIND:
8788 goto read_only;
8789 default:
8790 return false;
8791 }
8792 case bpf_ctx_range(struct bpf_sock, src_port):
8793 switch (attach_type) {
8794 case BPF_CGROUP_INET4_POST_BIND:
8795 case BPF_CGROUP_INET6_POST_BIND:
8796 goto read_only;
8797 default:
8798 return false;
8799 }
8800 }
8801 read_only:
8802 return access_type == BPF_READ;
8803 full_access:
8804 return true;
8805 }
8806
bpf_sock_common_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8807 bool bpf_sock_common_is_valid_access(int off, int size,
8808 enum bpf_access_type type,
8809 struct bpf_insn_access_aux *info)
8810 {
8811 switch (off) {
8812 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8813 return false;
8814 default:
8815 return bpf_sock_is_valid_access(off, size, type, info);
8816 }
8817 }
8818
bpf_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8819 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8820 struct bpf_insn_access_aux *info)
8821 {
8822 const int size_default = sizeof(__u32);
8823 int field_size;
8824
8825 if (off < 0 || off >= sizeof(struct bpf_sock))
8826 return false;
8827 if (off % size != 0)
8828 return false;
8829
8830 switch (off) {
8831 case offsetof(struct bpf_sock, state):
8832 case offsetof(struct bpf_sock, family):
8833 case offsetof(struct bpf_sock, type):
8834 case offsetof(struct bpf_sock, protocol):
8835 case offsetof(struct bpf_sock, src_port):
8836 case offsetof(struct bpf_sock, rx_queue_mapping):
8837 case bpf_ctx_range(struct bpf_sock, src_ip4):
8838 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8839 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8840 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8841 bpf_ctx_record_field_size(info, size_default);
8842 return bpf_ctx_narrow_access_ok(off, size, size_default);
8843 case bpf_ctx_range(struct bpf_sock, dst_port):
8844 field_size = size == size_default ?
8845 size_default : sizeof_field(struct bpf_sock, dst_port);
8846 bpf_ctx_record_field_size(info, field_size);
8847 return bpf_ctx_narrow_access_ok(off, size, field_size);
8848 case offsetofend(struct bpf_sock, dst_port) ...
8849 offsetof(struct bpf_sock, dst_ip4) - 1:
8850 return false;
8851 }
8852
8853 return size == size_default;
8854 }
8855
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)8856 static bool sock_filter_is_valid_access(int off, int size,
8857 enum bpf_access_type type,
8858 const struct bpf_prog *prog,
8859 struct bpf_insn_access_aux *info)
8860 {
8861 if (!bpf_sock_is_valid_access(off, size, type, info))
8862 return false;
8863 return __sock_filter_check_attach_type(off, type,
8864 prog->expected_attach_type);
8865 }
8866
bpf_noop_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8867 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8868 const struct bpf_prog *prog)
8869 {
8870 /* Neither direct read nor direct write requires any preliminary
8871 * action.
8872 */
8873 return 0;
8874 }
8875
bpf_unclone_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog,int drop_verdict)8876 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8877 const struct bpf_prog *prog, int drop_verdict)
8878 {
8879 struct bpf_insn *insn = insn_buf;
8880
8881 if (!direct_write)
8882 return 0;
8883
8884 /* if (!skb->cloned)
8885 * goto start;
8886 *
8887 * (Fast-path, otherwise approximation that we might be
8888 * a clone, do the rest in helper.)
8889 */
8890 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8891 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8892 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8893
8894 /* ret = bpf_skb_pull_data(skb, 0); */
8895 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8896 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8897 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8898 BPF_FUNC_skb_pull_data);
8899 /* if (!ret)
8900 * goto restore;
8901 * return TC_ACT_SHOT;
8902 */
8903 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8904 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8905 *insn++ = BPF_EXIT_INSN();
8906
8907 /* restore: */
8908 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8909 /* start: */
8910 *insn++ = prog->insnsi[0];
8911
8912 return insn - insn_buf;
8913 }
8914
bpf_gen_ld_abs(const struct bpf_insn * orig,struct bpf_insn * insn_buf)8915 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8916 struct bpf_insn *insn_buf)
8917 {
8918 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8919 struct bpf_insn *insn = insn_buf;
8920
8921 if (!indirect) {
8922 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8923 } else {
8924 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8925 if (orig->imm)
8926 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8927 }
8928 /* We're guaranteed here that CTX is in R6. */
8929 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8930
8931 switch (BPF_SIZE(orig->code)) {
8932 case BPF_B:
8933 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8934 break;
8935 case BPF_H:
8936 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8937 break;
8938 case BPF_W:
8939 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8940 break;
8941 }
8942
8943 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8944 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8945 *insn++ = BPF_EXIT_INSN();
8946
8947 return insn - insn_buf;
8948 }
8949
tc_cls_act_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8950 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8951 const struct bpf_prog *prog)
8952 {
8953 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8954 }
8955
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)8956 static bool tc_cls_act_is_valid_access(int off, int size,
8957 enum bpf_access_type type,
8958 const struct bpf_prog *prog,
8959 struct bpf_insn_access_aux *info)
8960 {
8961 if (type == BPF_WRITE) {
8962 switch (off) {
8963 case bpf_ctx_range(struct __sk_buff, mark):
8964 case bpf_ctx_range(struct __sk_buff, tc_index):
8965 case bpf_ctx_range(struct __sk_buff, priority):
8966 case bpf_ctx_range(struct __sk_buff, tc_classid):
8967 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8968 case bpf_ctx_range(struct __sk_buff, tstamp):
8969 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8970 break;
8971 default:
8972 return false;
8973 }
8974 }
8975
8976 switch (off) {
8977 case bpf_ctx_range(struct __sk_buff, data):
8978 info->reg_type = PTR_TO_PACKET;
8979 break;
8980 case bpf_ctx_range(struct __sk_buff, data_meta):
8981 info->reg_type = PTR_TO_PACKET_META;
8982 break;
8983 case bpf_ctx_range(struct __sk_buff, data_end):
8984 info->reg_type = PTR_TO_PACKET_END;
8985 break;
8986 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8987 return false;
8988 case offsetof(struct __sk_buff, tstamp_type):
8989 /* The convert_ctx_access() on reading and writing
8990 * __sk_buff->tstamp depends on whether the bpf prog
8991 * has used __sk_buff->tstamp_type or not.
8992 * Thus, we need to set prog->tstamp_type_access
8993 * earlier during is_valid_access() here.
8994 */
8995 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8996 return size == sizeof(__u8);
8997 }
8998
8999 return bpf_skb_is_valid_access(off, size, type, prog, info);
9000 }
9001
9002 DEFINE_MUTEX(nf_conn_btf_access_lock);
9003 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
9004
9005 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
9006 const struct bpf_reg_state *reg,
9007 int off, int size);
9008 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
9009
tc_cls_act_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9010 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
9011 const struct bpf_reg_state *reg,
9012 int off, int size)
9013 {
9014 int ret = -EACCES;
9015
9016 mutex_lock(&nf_conn_btf_access_lock);
9017 if (nfct_btf_struct_access)
9018 ret = nfct_btf_struct_access(log, reg, off, size);
9019 mutex_unlock(&nf_conn_btf_access_lock);
9020
9021 return ret;
9022 }
9023
__is_valid_xdp_access(int off,int size)9024 static bool __is_valid_xdp_access(int off, int size)
9025 {
9026 if (off < 0 || off >= sizeof(struct xdp_md))
9027 return false;
9028 if (off % size != 0)
9029 return false;
9030 if (size != sizeof(__u32))
9031 return false;
9032
9033 return true;
9034 }
9035
xdp_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9036 static bool xdp_is_valid_access(int off, int size,
9037 enum bpf_access_type type,
9038 const struct bpf_prog *prog,
9039 struct bpf_insn_access_aux *info)
9040 {
9041 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
9042 switch (off) {
9043 case offsetof(struct xdp_md, egress_ifindex):
9044 return false;
9045 }
9046 }
9047
9048 if (type == BPF_WRITE) {
9049 if (bpf_prog_is_offloaded(prog->aux)) {
9050 switch (off) {
9051 case offsetof(struct xdp_md, rx_queue_index):
9052 return __is_valid_xdp_access(off, size);
9053 }
9054 }
9055 return false;
9056 } else {
9057 switch (off) {
9058 case offsetof(struct xdp_md, data_meta):
9059 case offsetof(struct xdp_md, data):
9060 case offsetof(struct xdp_md, data_end):
9061 if (info->is_ldsx)
9062 return false;
9063 }
9064 }
9065
9066 switch (off) {
9067 case offsetof(struct xdp_md, data):
9068 info->reg_type = PTR_TO_PACKET;
9069 break;
9070 case offsetof(struct xdp_md, data_meta):
9071 info->reg_type = PTR_TO_PACKET_META;
9072 break;
9073 case offsetof(struct xdp_md, data_end):
9074 info->reg_type = PTR_TO_PACKET_END;
9075 break;
9076 }
9077
9078 return __is_valid_xdp_access(off, size);
9079 }
9080
bpf_warn_invalid_xdp_action(const struct net_device * dev,const struct bpf_prog * prog,u32 act)9081 void bpf_warn_invalid_xdp_action(const struct net_device *dev,
9082 const struct bpf_prog *prog, u32 act)
9083 {
9084 const u32 act_max = XDP_REDIRECT;
9085
9086 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
9087 act > act_max ? "Illegal" : "Driver unsupported",
9088 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
9089 }
9090 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
9091
xdp_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9092 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
9093 const struct bpf_reg_state *reg,
9094 int off, int size)
9095 {
9096 int ret = -EACCES;
9097
9098 mutex_lock(&nf_conn_btf_access_lock);
9099 if (nfct_btf_struct_access)
9100 ret = nfct_btf_struct_access(log, reg, off, size);
9101 mutex_unlock(&nf_conn_btf_access_lock);
9102
9103 return ret;
9104 }
9105
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)9106 static bool sock_addr_is_valid_access(int off, int size,
9107 enum bpf_access_type type,
9108 const struct bpf_prog *prog,
9109 struct bpf_insn_access_aux *info)
9110 {
9111 const int size_default = sizeof(__u32);
9112
9113 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
9114 return false;
9115 if (off % size != 0)
9116 return false;
9117
9118 /* Disallow access to fields not belonging to the attach type's address
9119 * family.
9120 */
9121 switch (off) {
9122 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9123 switch (prog->expected_attach_type) {
9124 case BPF_CGROUP_INET4_BIND:
9125 case BPF_CGROUP_INET4_CONNECT:
9126 case BPF_CGROUP_INET4_GETPEERNAME:
9127 case BPF_CGROUP_INET4_GETSOCKNAME:
9128 case BPF_CGROUP_UDP4_SENDMSG:
9129 case BPF_CGROUP_UDP4_RECVMSG:
9130 break;
9131 default:
9132 return false;
9133 }
9134 break;
9135 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9136 switch (prog->expected_attach_type) {
9137 case BPF_CGROUP_INET6_BIND:
9138 case BPF_CGROUP_INET6_CONNECT:
9139 case BPF_CGROUP_INET6_GETPEERNAME:
9140 case BPF_CGROUP_INET6_GETSOCKNAME:
9141 case BPF_CGROUP_UDP6_SENDMSG:
9142 case BPF_CGROUP_UDP6_RECVMSG:
9143 break;
9144 default:
9145 return false;
9146 }
9147 break;
9148 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9149 switch (prog->expected_attach_type) {
9150 case BPF_CGROUP_UDP4_SENDMSG:
9151 break;
9152 default:
9153 return false;
9154 }
9155 break;
9156 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9157 msg_src_ip6[3]):
9158 switch (prog->expected_attach_type) {
9159 case BPF_CGROUP_UDP6_SENDMSG:
9160 break;
9161 default:
9162 return false;
9163 }
9164 break;
9165 }
9166
9167 switch (off) {
9168 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9169 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9170 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9171 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9172 msg_src_ip6[3]):
9173 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9174 if (type == BPF_READ) {
9175 bpf_ctx_record_field_size(info, size_default);
9176
9177 if (bpf_ctx_wide_access_ok(off, size,
9178 struct bpf_sock_addr,
9179 user_ip6))
9180 return true;
9181
9182 if (bpf_ctx_wide_access_ok(off, size,
9183 struct bpf_sock_addr,
9184 msg_src_ip6))
9185 return true;
9186
9187 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9188 return false;
9189 } else {
9190 if (bpf_ctx_wide_access_ok(off, size,
9191 struct bpf_sock_addr,
9192 user_ip6))
9193 return true;
9194
9195 if (bpf_ctx_wide_access_ok(off, size,
9196 struct bpf_sock_addr,
9197 msg_src_ip6))
9198 return true;
9199
9200 if (size != size_default)
9201 return false;
9202 }
9203 break;
9204 case offsetof(struct bpf_sock_addr, sk):
9205 if (type != BPF_READ)
9206 return false;
9207 if (size != sizeof(__u64))
9208 return false;
9209 info->reg_type = PTR_TO_SOCKET;
9210 break;
9211 default:
9212 if (type == BPF_READ) {
9213 if (size != size_default)
9214 return false;
9215 } else {
9216 return false;
9217 }
9218 }
9219
9220 return true;
9221 }
9222
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)9223 static bool sock_ops_is_valid_access(int off, int size,
9224 enum bpf_access_type type,
9225 const struct bpf_prog *prog,
9226 struct bpf_insn_access_aux *info)
9227 {
9228 const int size_default = sizeof(__u32);
9229
9230 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9231 return false;
9232
9233 /* The verifier guarantees that size > 0. */
9234 if (off % size != 0)
9235 return false;
9236
9237 if (type == BPF_WRITE) {
9238 switch (off) {
9239 case offsetof(struct bpf_sock_ops, reply):
9240 case offsetof(struct bpf_sock_ops, sk_txhash):
9241 if (size != size_default)
9242 return false;
9243 break;
9244 default:
9245 return false;
9246 }
9247 } else {
9248 switch (off) {
9249 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9250 bytes_acked):
9251 if (size != sizeof(__u64))
9252 return false;
9253 break;
9254 case offsetof(struct bpf_sock_ops, sk):
9255 if (size != sizeof(__u64))
9256 return false;
9257 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9258 break;
9259 case offsetof(struct bpf_sock_ops, skb_data):
9260 if (size != sizeof(__u64))
9261 return false;
9262 info->reg_type = PTR_TO_PACKET;
9263 break;
9264 case offsetof(struct bpf_sock_ops, skb_data_end):
9265 if (size != sizeof(__u64))
9266 return false;
9267 info->reg_type = PTR_TO_PACKET_END;
9268 break;
9269 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9270 bpf_ctx_record_field_size(info, size_default);
9271 return bpf_ctx_narrow_access_ok(off, size,
9272 size_default);
9273 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9274 if (size != sizeof(__u64))
9275 return false;
9276 break;
9277 default:
9278 if (size != size_default)
9279 return false;
9280 break;
9281 }
9282 }
9283
9284 return true;
9285 }
9286
sk_skb_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)9287 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9288 const struct bpf_prog *prog)
9289 {
9290 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9291 }
9292
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)9293 static bool sk_skb_is_valid_access(int off, int size,
9294 enum bpf_access_type type,
9295 const struct bpf_prog *prog,
9296 struct bpf_insn_access_aux *info)
9297 {
9298 switch (off) {
9299 case bpf_ctx_range(struct __sk_buff, tc_classid):
9300 case bpf_ctx_range(struct __sk_buff, data_meta):
9301 case bpf_ctx_range(struct __sk_buff, tstamp):
9302 case bpf_ctx_range(struct __sk_buff, wire_len):
9303 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9304 return false;
9305 }
9306
9307 if (type == BPF_WRITE) {
9308 switch (off) {
9309 case bpf_ctx_range(struct __sk_buff, tc_index):
9310 case bpf_ctx_range(struct __sk_buff, priority):
9311 break;
9312 default:
9313 return false;
9314 }
9315 }
9316
9317 switch (off) {
9318 case bpf_ctx_range(struct __sk_buff, mark):
9319 return false;
9320 case bpf_ctx_range(struct __sk_buff, data):
9321 info->reg_type = PTR_TO_PACKET;
9322 break;
9323 case bpf_ctx_range(struct __sk_buff, data_end):
9324 info->reg_type = PTR_TO_PACKET_END;
9325 break;
9326 }
9327
9328 return bpf_skb_is_valid_access(off, size, type, prog, info);
9329 }
9330
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)9331 static bool sk_msg_is_valid_access(int off, int size,
9332 enum bpf_access_type type,
9333 const struct bpf_prog *prog,
9334 struct bpf_insn_access_aux *info)
9335 {
9336 if (type == BPF_WRITE)
9337 return false;
9338
9339 if (off % size != 0)
9340 return false;
9341
9342 switch (off) {
9343 case offsetof(struct sk_msg_md, data):
9344 info->reg_type = PTR_TO_PACKET;
9345 if (size != sizeof(__u64))
9346 return false;
9347 break;
9348 case offsetof(struct sk_msg_md, data_end):
9349 info->reg_type = PTR_TO_PACKET_END;
9350 if (size != sizeof(__u64))
9351 return false;
9352 break;
9353 case offsetof(struct sk_msg_md, sk):
9354 if (size != sizeof(__u64))
9355 return false;
9356 info->reg_type = PTR_TO_SOCKET;
9357 break;
9358 case bpf_ctx_range(struct sk_msg_md, family):
9359 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9360 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9361 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9362 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9363 case bpf_ctx_range(struct sk_msg_md, remote_port):
9364 case bpf_ctx_range(struct sk_msg_md, local_port):
9365 case bpf_ctx_range(struct sk_msg_md, size):
9366 if (size != sizeof(__u32))
9367 return false;
9368 break;
9369 default:
9370 return false;
9371 }
9372 return true;
9373 }
9374
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)9375 static bool flow_dissector_is_valid_access(int off, int size,
9376 enum bpf_access_type type,
9377 const struct bpf_prog *prog,
9378 struct bpf_insn_access_aux *info)
9379 {
9380 const int size_default = sizeof(__u32);
9381
9382 if (off < 0 || off >= sizeof(struct __sk_buff))
9383 return false;
9384
9385 if (type == BPF_WRITE)
9386 return false;
9387
9388 switch (off) {
9389 case bpf_ctx_range(struct __sk_buff, data):
9390 if (info->is_ldsx || size != size_default)
9391 return false;
9392 info->reg_type = PTR_TO_PACKET;
9393 return true;
9394 case bpf_ctx_range(struct __sk_buff, data_end):
9395 if (info->is_ldsx || size != size_default)
9396 return false;
9397 info->reg_type = PTR_TO_PACKET_END;
9398 return true;
9399 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9400 if (size != sizeof(__u64))
9401 return false;
9402 info->reg_type = PTR_TO_FLOW_KEYS;
9403 return true;
9404 default:
9405 return false;
9406 }
9407 }
9408
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)9409 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9410 const struct bpf_insn *si,
9411 struct bpf_insn *insn_buf,
9412 struct bpf_prog *prog,
9413 u32 *target_size)
9414
9415 {
9416 struct bpf_insn *insn = insn_buf;
9417
9418 switch (si->off) {
9419 case offsetof(struct __sk_buff, data):
9420 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9421 si->dst_reg, si->src_reg,
9422 offsetof(struct bpf_flow_dissector, data));
9423 break;
9424
9425 case offsetof(struct __sk_buff, data_end):
9426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9427 si->dst_reg, si->src_reg,
9428 offsetof(struct bpf_flow_dissector, data_end));
9429 break;
9430
9431 case offsetof(struct __sk_buff, flow_keys):
9432 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9433 si->dst_reg, si->src_reg,
9434 offsetof(struct bpf_flow_dissector, flow_keys));
9435 break;
9436 }
9437
9438 return insn - insn_buf;
9439 }
9440
bpf_convert_tstamp_type_read(const struct bpf_insn * si,struct bpf_insn * insn)9441 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9442 struct bpf_insn *insn)
9443 {
9444 __u8 value_reg = si->dst_reg;
9445 __u8 skb_reg = si->src_reg;
9446 BUILD_BUG_ON(__SKB_CLOCK_MAX != (int)BPF_SKB_CLOCK_TAI);
9447 BUILD_BUG_ON(SKB_CLOCK_REALTIME != (int)BPF_SKB_CLOCK_REALTIME);
9448 BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != (int)BPF_SKB_CLOCK_MONOTONIC);
9449 BUILD_BUG_ON(SKB_CLOCK_TAI != (int)BPF_SKB_CLOCK_TAI);
9450 *insn++ = BPF_LDX_MEM(BPF_B, value_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9451 *insn++ = BPF_ALU32_IMM(BPF_AND, value_reg, SKB_TSTAMP_TYPE_MASK);
9452 #ifdef __BIG_ENDIAN_BITFIELD
9453 *insn++ = BPF_ALU32_IMM(BPF_RSH, value_reg, SKB_TSTAMP_TYPE_RSHIFT);
9454 #else
9455 BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1));
9456 #endif
9457
9458 return insn;
9459 }
9460
bpf_convert_shinfo_access(__u8 dst_reg,__u8 skb_reg,struct bpf_insn * insn)9461 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9462 struct bpf_insn *insn)
9463 {
9464 /* si->dst_reg = skb_shinfo(SKB); */
9465 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9466 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9467 BPF_REG_AX, skb_reg,
9468 offsetof(struct sk_buff, end));
9469 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9470 dst_reg, skb_reg,
9471 offsetof(struct sk_buff, head));
9472 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9473 #else
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9475 dst_reg, skb_reg,
9476 offsetof(struct sk_buff, end));
9477 #endif
9478
9479 return insn;
9480 }
9481
bpf_convert_tstamp_read(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9482 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9483 const struct bpf_insn *si,
9484 struct bpf_insn *insn)
9485 {
9486 __u8 value_reg = si->dst_reg;
9487 __u8 skb_reg = si->src_reg;
9488
9489 #ifdef CONFIG_NET_XGRESS
9490 /* If the tstamp_type is read,
9491 * the bpf prog is aware the tstamp could have delivery time.
9492 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9493 */
9494 if (!prog->tstamp_type_access) {
9495 /* AX is needed because src_reg and dst_reg could be the same */
9496 __u8 tmp_reg = BPF_REG_AX;
9497
9498 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9499 /* check if ingress mask bits is set */
9500 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9501 *insn++ = BPF_JMP_A(4);
9502 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, SKB_TSTAMP_TYPE_MASK, 1);
9503 *insn++ = BPF_JMP_A(2);
9504 /* skb->tc_at_ingress && skb->tstamp_type,
9505 * read 0 as the (rcv) timestamp.
9506 */
9507 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9508 *insn++ = BPF_JMP_A(1);
9509 }
9510 #endif
9511
9512 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9513 offsetof(struct sk_buff, tstamp));
9514 return insn;
9515 }
9516
bpf_convert_tstamp_write(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9517 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9518 const struct bpf_insn *si,
9519 struct bpf_insn *insn)
9520 {
9521 __u8 value_reg = si->src_reg;
9522 __u8 skb_reg = si->dst_reg;
9523
9524 #ifdef CONFIG_NET_XGRESS
9525 /* If the tstamp_type is read,
9526 * the bpf prog is aware the tstamp could have delivery time.
9527 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9528 * Otherwise, writing at ingress will have to clear the
9529 * skb->tstamp_type bit also.
9530 */
9531 if (!prog->tstamp_type_access) {
9532 __u8 tmp_reg = BPF_REG_AX;
9533
9534 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9535 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9536 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9537 /* goto <store> */
9538 *insn++ = BPF_JMP_A(2);
9539 /* <clear>: skb->tstamp_type */
9540 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_TSTAMP_TYPE_MASK);
9541 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9542 }
9543 #endif
9544
9545 /* <store>: skb->tstamp = tstamp */
9546 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9547 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9548 return insn;
9549 }
9550
9551 #define BPF_EMIT_STORE(size, si, off) \
9552 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9553 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9554
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)9555 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9556 const struct bpf_insn *si,
9557 struct bpf_insn *insn_buf,
9558 struct bpf_prog *prog, u32 *target_size)
9559 {
9560 struct bpf_insn *insn = insn_buf;
9561 int off;
9562
9563 switch (si->off) {
9564 case offsetof(struct __sk_buff, len):
9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9566 bpf_target_off(struct sk_buff, len, 4,
9567 target_size));
9568 break;
9569
9570 case offsetof(struct __sk_buff, protocol):
9571 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9572 bpf_target_off(struct sk_buff, protocol, 2,
9573 target_size));
9574 break;
9575
9576 case offsetof(struct __sk_buff, vlan_proto):
9577 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9578 bpf_target_off(struct sk_buff, vlan_proto, 2,
9579 target_size));
9580 break;
9581
9582 case offsetof(struct __sk_buff, priority):
9583 if (type == BPF_WRITE)
9584 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9585 bpf_target_off(struct sk_buff, priority, 4,
9586 target_size));
9587 else
9588 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9589 bpf_target_off(struct sk_buff, priority, 4,
9590 target_size));
9591 break;
9592
9593 case offsetof(struct __sk_buff, ingress_ifindex):
9594 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9595 bpf_target_off(struct sk_buff, skb_iif, 4,
9596 target_size));
9597 break;
9598
9599 case offsetof(struct __sk_buff, ifindex):
9600 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9601 si->dst_reg, si->src_reg,
9602 offsetof(struct sk_buff, dev));
9603 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9604 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9605 bpf_target_off(struct net_device, ifindex, 4,
9606 target_size));
9607 break;
9608
9609 case offsetof(struct __sk_buff, hash):
9610 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9611 bpf_target_off(struct sk_buff, hash, 4,
9612 target_size));
9613 break;
9614
9615 case offsetof(struct __sk_buff, mark):
9616 if (type == BPF_WRITE)
9617 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9618 bpf_target_off(struct sk_buff, mark, 4,
9619 target_size));
9620 else
9621 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9622 bpf_target_off(struct sk_buff, mark, 4,
9623 target_size));
9624 break;
9625
9626 case offsetof(struct __sk_buff, pkt_type):
9627 *target_size = 1;
9628 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9629 PKT_TYPE_OFFSET);
9630 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9631 #ifdef __BIG_ENDIAN_BITFIELD
9632 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9633 #endif
9634 break;
9635
9636 case offsetof(struct __sk_buff, queue_mapping):
9637 if (type == BPF_WRITE) {
9638 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9639
9640 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9641 *insn++ = BPF_JMP_A(0); /* noop */
9642 break;
9643 }
9644
9645 if (BPF_CLASS(si->code) == BPF_STX)
9646 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9647 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9648 } else {
9649 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9650 bpf_target_off(struct sk_buff,
9651 queue_mapping,
9652 2, target_size));
9653 }
9654 break;
9655
9656 case offsetof(struct __sk_buff, vlan_present):
9657 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9658 bpf_target_off(struct sk_buff,
9659 vlan_all, 4, target_size));
9660 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9661 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9662 break;
9663
9664 case offsetof(struct __sk_buff, vlan_tci):
9665 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9666 bpf_target_off(struct sk_buff, vlan_tci, 2,
9667 target_size));
9668 break;
9669
9670 case offsetof(struct __sk_buff, cb[0]) ...
9671 offsetofend(struct __sk_buff, cb[4]) - 1:
9672 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9673 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9674 offsetof(struct qdisc_skb_cb, data)) %
9675 sizeof(__u64));
9676
9677 prog->cb_access = 1;
9678 off = si->off;
9679 off -= offsetof(struct __sk_buff, cb[0]);
9680 off += offsetof(struct sk_buff, cb);
9681 off += offsetof(struct qdisc_skb_cb, data);
9682 if (type == BPF_WRITE)
9683 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9684 else
9685 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9686 si->src_reg, off);
9687 break;
9688
9689 case offsetof(struct __sk_buff, tc_classid):
9690 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9691
9692 off = si->off;
9693 off -= offsetof(struct __sk_buff, tc_classid);
9694 off += offsetof(struct sk_buff, cb);
9695 off += offsetof(struct qdisc_skb_cb, tc_classid);
9696 *target_size = 2;
9697 if (type == BPF_WRITE)
9698 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9699 else
9700 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9701 si->src_reg, off);
9702 break;
9703
9704 case offsetof(struct __sk_buff, data):
9705 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9706 si->dst_reg, si->src_reg,
9707 offsetof(struct sk_buff, data));
9708 break;
9709
9710 case offsetof(struct __sk_buff, data_meta):
9711 off = si->off;
9712 off -= offsetof(struct __sk_buff, data_meta);
9713 off += offsetof(struct sk_buff, cb);
9714 off += offsetof(struct bpf_skb_data_end, data_meta);
9715 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9716 si->src_reg, off);
9717 break;
9718
9719 case offsetof(struct __sk_buff, data_end):
9720 off = si->off;
9721 off -= offsetof(struct __sk_buff, data_end);
9722 off += offsetof(struct sk_buff, cb);
9723 off += offsetof(struct bpf_skb_data_end, data_end);
9724 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9725 si->src_reg, off);
9726 break;
9727
9728 case offsetof(struct __sk_buff, tc_index):
9729 #ifdef CONFIG_NET_SCHED
9730 if (type == BPF_WRITE)
9731 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9732 bpf_target_off(struct sk_buff, tc_index, 2,
9733 target_size));
9734 else
9735 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9736 bpf_target_off(struct sk_buff, tc_index, 2,
9737 target_size));
9738 #else
9739 *target_size = 2;
9740 if (type == BPF_WRITE)
9741 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9742 else
9743 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9744 #endif
9745 break;
9746
9747 case offsetof(struct __sk_buff, napi_id):
9748 #if defined(CONFIG_NET_RX_BUSY_POLL)
9749 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9750 bpf_target_off(struct sk_buff, napi_id, 4,
9751 target_size));
9752 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9753 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9754 #else
9755 *target_size = 4;
9756 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9757 #endif
9758 break;
9759 case offsetof(struct __sk_buff, family):
9760 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9761
9762 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9763 si->dst_reg, si->src_reg,
9764 offsetof(struct sk_buff, sk));
9765 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9766 bpf_target_off(struct sock_common,
9767 skc_family,
9768 2, target_size));
9769 break;
9770 case offsetof(struct __sk_buff, remote_ip4):
9771 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
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_W, si->dst_reg, si->dst_reg,
9777 bpf_target_off(struct sock_common,
9778 skc_daddr,
9779 4, target_size));
9780 break;
9781 case offsetof(struct __sk_buff, local_ip4):
9782 BUILD_BUG_ON(sizeof_field(struct sock_common,
9783 skc_rcv_saddr) != 4);
9784
9785 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9786 si->dst_reg, si->src_reg,
9787 offsetof(struct sk_buff, sk));
9788 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9789 bpf_target_off(struct sock_common,
9790 skc_rcv_saddr,
9791 4, target_size));
9792 break;
9793 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9794 offsetof(struct __sk_buff, remote_ip6[3]):
9795 #if IS_ENABLED(CONFIG_IPV6)
9796 BUILD_BUG_ON(sizeof_field(struct sock_common,
9797 skc_v6_daddr.s6_addr32[0]) != 4);
9798
9799 off = si->off;
9800 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9801
9802 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9803 si->dst_reg, si->src_reg,
9804 offsetof(struct sk_buff, sk));
9805 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9806 offsetof(struct sock_common,
9807 skc_v6_daddr.s6_addr32[0]) +
9808 off);
9809 #else
9810 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9811 #endif
9812 break;
9813 case offsetof(struct __sk_buff, local_ip6[0]) ...
9814 offsetof(struct __sk_buff, local_ip6[3]):
9815 #if IS_ENABLED(CONFIG_IPV6)
9816 BUILD_BUG_ON(sizeof_field(struct sock_common,
9817 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9818
9819 off = si->off;
9820 off -= offsetof(struct __sk_buff, local_ip6[0]);
9821
9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9823 si->dst_reg, si->src_reg,
9824 offsetof(struct sk_buff, sk));
9825 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9826 offsetof(struct sock_common,
9827 skc_v6_rcv_saddr.s6_addr32[0]) +
9828 off);
9829 #else
9830 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9831 #endif
9832 break;
9833
9834 case offsetof(struct __sk_buff, remote_port):
9835 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9836
9837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9838 si->dst_reg, si->src_reg,
9839 offsetof(struct sk_buff, sk));
9840 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9841 bpf_target_off(struct sock_common,
9842 skc_dport,
9843 2, target_size));
9844 #ifndef __BIG_ENDIAN_BITFIELD
9845 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9846 #endif
9847 break;
9848
9849 case offsetof(struct __sk_buff, local_port):
9850 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9851
9852 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9853 si->dst_reg, si->src_reg,
9854 offsetof(struct sk_buff, sk));
9855 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9856 bpf_target_off(struct sock_common,
9857 skc_num, 2, target_size));
9858 break;
9859
9860 case offsetof(struct __sk_buff, tstamp):
9861 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9862
9863 if (type == BPF_WRITE)
9864 insn = bpf_convert_tstamp_write(prog, si, insn);
9865 else
9866 insn = bpf_convert_tstamp_read(prog, si, insn);
9867 break;
9868
9869 case offsetof(struct __sk_buff, tstamp_type):
9870 insn = bpf_convert_tstamp_type_read(si, insn);
9871 break;
9872
9873 case offsetof(struct __sk_buff, gso_segs):
9874 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9876 si->dst_reg, si->dst_reg,
9877 bpf_target_off(struct skb_shared_info,
9878 gso_segs, 2,
9879 target_size));
9880 break;
9881 case offsetof(struct __sk_buff, gso_size):
9882 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9884 si->dst_reg, si->dst_reg,
9885 bpf_target_off(struct skb_shared_info,
9886 gso_size, 2,
9887 target_size));
9888 break;
9889 case offsetof(struct __sk_buff, wire_len):
9890 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9891
9892 off = si->off;
9893 off -= offsetof(struct __sk_buff, wire_len);
9894 off += offsetof(struct sk_buff, cb);
9895 off += offsetof(struct qdisc_skb_cb, pkt_len);
9896 *target_size = 4;
9897 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9898 break;
9899
9900 case offsetof(struct __sk_buff, sk):
9901 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9902 si->dst_reg, si->src_reg,
9903 offsetof(struct sk_buff, sk));
9904 break;
9905 case offsetof(struct __sk_buff, hwtstamp):
9906 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9907 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9908
9909 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9910 *insn++ = BPF_LDX_MEM(BPF_DW,
9911 si->dst_reg, si->dst_reg,
9912 bpf_target_off(struct skb_shared_info,
9913 hwtstamps, 8,
9914 target_size));
9915 break;
9916 }
9917
9918 return insn - insn_buf;
9919 }
9920
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)9921 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9922 const struct bpf_insn *si,
9923 struct bpf_insn *insn_buf,
9924 struct bpf_prog *prog, u32 *target_size)
9925 {
9926 struct bpf_insn *insn = insn_buf;
9927 int off;
9928
9929 switch (si->off) {
9930 case offsetof(struct bpf_sock, bound_dev_if):
9931 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9932
9933 if (type == BPF_WRITE)
9934 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9935 offsetof(struct sock, sk_bound_dev_if));
9936 else
9937 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9938 offsetof(struct sock, sk_bound_dev_if));
9939 break;
9940
9941 case offsetof(struct bpf_sock, mark):
9942 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9943
9944 if (type == BPF_WRITE)
9945 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9946 offsetof(struct sock, sk_mark));
9947 else
9948 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9949 offsetof(struct sock, sk_mark));
9950 break;
9951
9952 case offsetof(struct bpf_sock, priority):
9953 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9954
9955 if (type == BPF_WRITE)
9956 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9957 offsetof(struct sock, sk_priority));
9958 else
9959 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9960 offsetof(struct sock, sk_priority));
9961 break;
9962
9963 case offsetof(struct bpf_sock, family):
9964 *insn++ = BPF_LDX_MEM(
9965 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9966 si->dst_reg, si->src_reg,
9967 bpf_target_off(struct sock_common,
9968 skc_family,
9969 sizeof_field(struct sock_common,
9970 skc_family),
9971 target_size));
9972 break;
9973
9974 case offsetof(struct bpf_sock, type):
9975 *insn++ = BPF_LDX_MEM(
9976 BPF_FIELD_SIZEOF(struct sock, sk_type),
9977 si->dst_reg, si->src_reg,
9978 bpf_target_off(struct sock, sk_type,
9979 sizeof_field(struct sock, sk_type),
9980 target_size));
9981 break;
9982
9983 case offsetof(struct bpf_sock, protocol):
9984 *insn++ = BPF_LDX_MEM(
9985 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9986 si->dst_reg, si->src_reg,
9987 bpf_target_off(struct sock, sk_protocol,
9988 sizeof_field(struct sock, sk_protocol),
9989 target_size));
9990 break;
9991
9992 case offsetof(struct bpf_sock, src_ip4):
9993 *insn++ = BPF_LDX_MEM(
9994 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9995 bpf_target_off(struct sock_common, skc_rcv_saddr,
9996 sizeof_field(struct sock_common,
9997 skc_rcv_saddr),
9998 target_size));
9999 break;
10000
10001 case offsetof(struct bpf_sock, dst_ip4):
10002 *insn++ = BPF_LDX_MEM(
10003 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10004 bpf_target_off(struct sock_common, skc_daddr,
10005 sizeof_field(struct sock_common,
10006 skc_daddr),
10007 target_size));
10008 break;
10009
10010 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
10011 #if IS_ENABLED(CONFIG_IPV6)
10012 off = si->off;
10013 off -= offsetof(struct bpf_sock, src_ip6[0]);
10014 *insn++ = BPF_LDX_MEM(
10015 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10016 bpf_target_off(
10017 struct sock_common,
10018 skc_v6_rcv_saddr.s6_addr32[0],
10019 sizeof_field(struct sock_common,
10020 skc_v6_rcv_saddr.s6_addr32[0]),
10021 target_size) + off);
10022 #else
10023 (void)off;
10024 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10025 #endif
10026 break;
10027
10028 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
10029 #if IS_ENABLED(CONFIG_IPV6)
10030 off = si->off;
10031 off -= offsetof(struct bpf_sock, dst_ip6[0]);
10032 *insn++ = BPF_LDX_MEM(
10033 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10034 bpf_target_off(struct sock_common,
10035 skc_v6_daddr.s6_addr32[0],
10036 sizeof_field(struct sock_common,
10037 skc_v6_daddr.s6_addr32[0]),
10038 target_size) + off);
10039 #else
10040 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10041 *target_size = 4;
10042 #endif
10043 break;
10044
10045 case offsetof(struct bpf_sock, src_port):
10046 *insn++ = BPF_LDX_MEM(
10047 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
10048 si->dst_reg, si->src_reg,
10049 bpf_target_off(struct sock_common, skc_num,
10050 sizeof_field(struct sock_common,
10051 skc_num),
10052 target_size));
10053 break;
10054
10055 case offsetof(struct bpf_sock, dst_port):
10056 *insn++ = BPF_LDX_MEM(
10057 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
10058 si->dst_reg, si->src_reg,
10059 bpf_target_off(struct sock_common, skc_dport,
10060 sizeof_field(struct sock_common,
10061 skc_dport),
10062 target_size));
10063 break;
10064
10065 case offsetof(struct bpf_sock, state):
10066 *insn++ = BPF_LDX_MEM(
10067 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
10068 si->dst_reg, si->src_reg,
10069 bpf_target_off(struct sock_common, skc_state,
10070 sizeof_field(struct sock_common,
10071 skc_state),
10072 target_size));
10073 break;
10074 case offsetof(struct bpf_sock, rx_queue_mapping):
10075 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
10076 *insn++ = BPF_LDX_MEM(
10077 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
10078 si->dst_reg, si->src_reg,
10079 bpf_target_off(struct sock, sk_rx_queue_mapping,
10080 sizeof_field(struct sock,
10081 sk_rx_queue_mapping),
10082 target_size));
10083 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
10084 1);
10085 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10086 #else
10087 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10088 *target_size = 2;
10089 #endif
10090 break;
10091 }
10092
10093 return insn - insn_buf;
10094 }
10095
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)10096 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
10097 const struct bpf_insn *si,
10098 struct bpf_insn *insn_buf,
10099 struct bpf_prog *prog, u32 *target_size)
10100 {
10101 struct bpf_insn *insn = insn_buf;
10102
10103 switch (si->off) {
10104 case offsetof(struct __sk_buff, ifindex):
10105 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
10106 si->dst_reg, si->src_reg,
10107 offsetof(struct sk_buff, dev));
10108 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10109 bpf_target_off(struct net_device, ifindex, 4,
10110 target_size));
10111 break;
10112 default:
10113 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10114 target_size);
10115 }
10116
10117 return insn - insn_buf;
10118 }
10119
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)10120 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
10121 const struct bpf_insn *si,
10122 struct bpf_insn *insn_buf,
10123 struct bpf_prog *prog, u32 *target_size)
10124 {
10125 struct bpf_insn *insn = insn_buf;
10126
10127 switch (si->off) {
10128 case offsetof(struct xdp_md, data):
10129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
10130 si->dst_reg, si->src_reg,
10131 offsetof(struct xdp_buff, data));
10132 break;
10133 case offsetof(struct xdp_md, data_meta):
10134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
10135 si->dst_reg, si->src_reg,
10136 offsetof(struct xdp_buff, data_meta));
10137 break;
10138 case offsetof(struct xdp_md, data_end):
10139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
10140 si->dst_reg, si->src_reg,
10141 offsetof(struct xdp_buff, data_end));
10142 break;
10143 case offsetof(struct xdp_md, ingress_ifindex):
10144 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10145 si->dst_reg, si->src_reg,
10146 offsetof(struct xdp_buff, rxq));
10147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
10148 si->dst_reg, si->dst_reg,
10149 offsetof(struct xdp_rxq_info, dev));
10150 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10151 offsetof(struct net_device, ifindex));
10152 break;
10153 case offsetof(struct xdp_md, rx_queue_index):
10154 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10155 si->dst_reg, si->src_reg,
10156 offsetof(struct xdp_buff, rxq));
10157 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10158 offsetof(struct xdp_rxq_info,
10159 queue_index));
10160 break;
10161 case offsetof(struct xdp_md, egress_ifindex):
10162 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
10163 si->dst_reg, si->src_reg,
10164 offsetof(struct xdp_buff, txq));
10165 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10166 si->dst_reg, si->dst_reg,
10167 offsetof(struct xdp_txq_info, dev));
10168 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10169 offsetof(struct net_device, ifindex));
10170 break;
10171 }
10172
10173 return insn - insn_buf;
10174 }
10175
10176 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10177 * context Structure, F is Field in context structure that contains a pointer
10178 * to Nested Structure of type NS that has the field NF.
10179 *
10180 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10181 * sure that SIZE is not greater than actual size of S.F.NF.
10182 *
10183 * If offset OFF is provided, the load happens from that offset relative to
10184 * offset of NF.
10185 */
10186 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10187 do { \
10188 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10189 si->src_reg, offsetof(S, F)); \
10190 *insn++ = BPF_LDX_MEM( \
10191 SIZE, si->dst_reg, si->dst_reg, \
10192 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10193 target_size) \
10194 + OFF); \
10195 } while (0)
10196
10197 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10198 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10199 BPF_FIELD_SIZEOF(NS, NF), 0)
10200
10201 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10202 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10203 *
10204 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10205 * "register" since two registers available in convert_ctx_access are not
10206 * enough: we can't override neither SRC, since it contains value to store, nor
10207 * DST since it contains pointer to context that may be used by later
10208 * instructions. But we need a temporary place to save pointer to nested
10209 * structure whose field we want to store to.
10210 */
10211 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10212 do { \
10213 int tmp_reg = BPF_REG_9; \
10214 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10215 --tmp_reg; \
10216 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10217 --tmp_reg; \
10218 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10219 offsetof(S, TF)); \
10220 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10221 si->dst_reg, offsetof(S, F)); \
10222 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10223 tmp_reg, si->src_reg, \
10224 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10225 target_size) \
10226 + OFF, \
10227 si->imm); \
10228 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10229 offsetof(S, TF)); \
10230 } while (0)
10231
10232 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10233 TF) \
10234 do { \
10235 if (type == BPF_WRITE) { \
10236 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10237 OFF, TF); \
10238 } else { \
10239 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10240 S, NS, F, NF, SIZE, OFF); \
10241 } \
10242 } while (0)
10243
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)10244 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10245 const struct bpf_insn *si,
10246 struct bpf_insn *insn_buf,
10247 struct bpf_prog *prog, u32 *target_size)
10248 {
10249 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10250 struct bpf_insn *insn = insn_buf;
10251
10252 switch (si->off) {
10253 case offsetof(struct bpf_sock_addr, user_family):
10254 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10255 struct sockaddr, uaddr, sa_family);
10256 break;
10257
10258 case offsetof(struct bpf_sock_addr, user_ip4):
10259 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10260 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10261 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10262 break;
10263
10264 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10265 off = si->off;
10266 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10267 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10268 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10269 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10270 tmp_reg);
10271 break;
10272
10273 case offsetof(struct bpf_sock_addr, user_port):
10274 /* To get port we need to know sa_family first and then treat
10275 * sockaddr as either sockaddr_in or sockaddr_in6.
10276 * Though we can simplify since port field has same offset and
10277 * size in both structures.
10278 * Here we check this invariant and use just one of the
10279 * structures if it's true.
10280 */
10281 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10282 offsetof(struct sockaddr_in6, sin6_port));
10283 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10284 sizeof_field(struct sockaddr_in6, sin6_port));
10285 /* Account for sin6_port being smaller than user_port. */
10286 port_size = min(port_size, BPF_LDST_BYTES(si));
10287 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10288 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10289 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10290 break;
10291
10292 case offsetof(struct bpf_sock_addr, family):
10293 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10294 struct sock, sk, sk_family);
10295 break;
10296
10297 case offsetof(struct bpf_sock_addr, type):
10298 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10299 struct sock, sk, sk_type);
10300 break;
10301
10302 case offsetof(struct bpf_sock_addr, protocol):
10303 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10304 struct sock, sk, sk_protocol);
10305 break;
10306
10307 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10308 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10309 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10310 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10311 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10312 break;
10313
10314 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10315 msg_src_ip6[3]):
10316 off = si->off;
10317 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10318 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10319 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10320 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10321 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10322 break;
10323 case offsetof(struct bpf_sock_addr, sk):
10324 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10325 si->dst_reg, si->src_reg,
10326 offsetof(struct bpf_sock_addr_kern, sk));
10327 break;
10328 }
10329
10330 return insn - insn_buf;
10331 }
10332
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)10333 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10334 const struct bpf_insn *si,
10335 struct bpf_insn *insn_buf,
10336 struct bpf_prog *prog,
10337 u32 *target_size)
10338 {
10339 struct bpf_insn *insn = insn_buf;
10340 int off;
10341
10342 /* Helper macro for adding read access to tcp_sock or sock fields. */
10343 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10344 do { \
10345 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10346 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10347 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10348 if (si->dst_reg == reg || si->src_reg == reg) \
10349 reg--; \
10350 if (si->dst_reg == reg || si->src_reg == reg) \
10351 reg--; \
10352 if (si->dst_reg == si->src_reg) { \
10353 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10354 offsetof(struct bpf_sock_ops_kern, \
10355 temp)); \
10356 fullsock_reg = reg; \
10357 jmp += 2; \
10358 } \
10359 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10360 struct bpf_sock_ops_kern, \
10361 is_fullsock), \
10362 fullsock_reg, si->src_reg, \
10363 offsetof(struct bpf_sock_ops_kern, \
10364 is_fullsock)); \
10365 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10366 if (si->dst_reg == si->src_reg) \
10367 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10368 offsetof(struct bpf_sock_ops_kern, \
10369 temp)); \
10370 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10371 struct bpf_sock_ops_kern, sk),\
10372 si->dst_reg, si->src_reg, \
10373 offsetof(struct bpf_sock_ops_kern, sk));\
10374 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10375 OBJ_FIELD), \
10376 si->dst_reg, si->dst_reg, \
10377 offsetof(OBJ, OBJ_FIELD)); \
10378 if (si->dst_reg == si->src_reg) { \
10379 *insn++ = BPF_JMP_A(1); \
10380 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10381 offsetof(struct bpf_sock_ops_kern, \
10382 temp)); \
10383 } \
10384 } while (0)
10385
10386 #define SOCK_OPS_GET_SK() \
10387 do { \
10388 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10389 if (si->dst_reg == reg || si->src_reg == reg) \
10390 reg--; \
10391 if (si->dst_reg == reg || si->src_reg == reg) \
10392 reg--; \
10393 if (si->dst_reg == si->src_reg) { \
10394 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10395 offsetof(struct bpf_sock_ops_kern, \
10396 temp)); \
10397 fullsock_reg = reg; \
10398 jmp += 2; \
10399 } \
10400 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10401 struct bpf_sock_ops_kern, \
10402 is_fullsock), \
10403 fullsock_reg, si->src_reg, \
10404 offsetof(struct bpf_sock_ops_kern, \
10405 is_fullsock)); \
10406 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10407 if (si->dst_reg == si->src_reg) \
10408 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10409 offsetof(struct bpf_sock_ops_kern, \
10410 temp)); \
10411 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10412 struct bpf_sock_ops_kern, sk),\
10413 si->dst_reg, si->src_reg, \
10414 offsetof(struct bpf_sock_ops_kern, sk));\
10415 if (si->dst_reg == si->src_reg) { \
10416 *insn++ = BPF_JMP_A(1); \
10417 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10418 offsetof(struct bpf_sock_ops_kern, \
10419 temp)); \
10420 } \
10421 } while (0)
10422
10423 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10424 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10425
10426 /* Helper macro for adding write access to tcp_sock or sock fields.
10427 * The macro is called with two registers, dst_reg which contains a pointer
10428 * to ctx (context) and src_reg which contains the value that should be
10429 * stored. However, we need an additional register since we cannot overwrite
10430 * dst_reg because it may be used later in the program.
10431 * Instead we "borrow" one of the other register. We first save its value
10432 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10433 * it at the end of the macro.
10434 */
10435 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10436 do { \
10437 int reg = BPF_REG_9; \
10438 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10439 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10440 if (si->dst_reg == reg || si->src_reg == reg) \
10441 reg--; \
10442 if (si->dst_reg == reg || si->src_reg == reg) \
10443 reg--; \
10444 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10445 offsetof(struct bpf_sock_ops_kern, \
10446 temp)); \
10447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10448 struct bpf_sock_ops_kern, \
10449 is_fullsock), \
10450 reg, si->dst_reg, \
10451 offsetof(struct bpf_sock_ops_kern, \
10452 is_fullsock)); \
10453 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10454 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10455 struct bpf_sock_ops_kern, sk),\
10456 reg, si->dst_reg, \
10457 offsetof(struct bpf_sock_ops_kern, sk));\
10458 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10459 BPF_MEM | BPF_CLASS(si->code), \
10460 reg, si->src_reg, \
10461 offsetof(OBJ, OBJ_FIELD), \
10462 si->imm); \
10463 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10464 offsetof(struct bpf_sock_ops_kern, \
10465 temp)); \
10466 } while (0)
10467
10468 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10469 do { \
10470 if (TYPE == BPF_WRITE) \
10471 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10472 else \
10473 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10474 } while (0)
10475
10476 switch (si->off) {
10477 case offsetof(struct bpf_sock_ops, op):
10478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10479 op),
10480 si->dst_reg, si->src_reg,
10481 offsetof(struct bpf_sock_ops_kern, op));
10482 break;
10483
10484 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10485 offsetof(struct bpf_sock_ops, replylong[3]):
10486 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10487 sizeof_field(struct bpf_sock_ops_kern, reply));
10488 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10489 sizeof_field(struct bpf_sock_ops_kern, replylong));
10490 off = si->off;
10491 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10492 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10493 if (type == BPF_WRITE)
10494 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10495 else
10496 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10497 off);
10498 break;
10499
10500 case offsetof(struct bpf_sock_ops, family):
10501 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10502
10503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10504 struct bpf_sock_ops_kern, sk),
10505 si->dst_reg, si->src_reg,
10506 offsetof(struct bpf_sock_ops_kern, sk));
10507 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10508 offsetof(struct sock_common, skc_family));
10509 break;
10510
10511 case offsetof(struct bpf_sock_ops, remote_ip4):
10512 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10513
10514 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10515 struct bpf_sock_ops_kern, sk),
10516 si->dst_reg, si->src_reg,
10517 offsetof(struct bpf_sock_ops_kern, sk));
10518 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10519 offsetof(struct sock_common, skc_daddr));
10520 break;
10521
10522 case offsetof(struct bpf_sock_ops, local_ip4):
10523 BUILD_BUG_ON(sizeof_field(struct sock_common,
10524 skc_rcv_saddr) != 4);
10525
10526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10527 struct bpf_sock_ops_kern, sk),
10528 si->dst_reg, si->src_reg,
10529 offsetof(struct bpf_sock_ops_kern, sk));
10530 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10531 offsetof(struct sock_common,
10532 skc_rcv_saddr));
10533 break;
10534
10535 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10536 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10537 #if IS_ENABLED(CONFIG_IPV6)
10538 BUILD_BUG_ON(sizeof_field(struct sock_common,
10539 skc_v6_daddr.s6_addr32[0]) != 4);
10540
10541 off = si->off;
10542 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10543 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10544 struct bpf_sock_ops_kern, sk),
10545 si->dst_reg, si->src_reg,
10546 offsetof(struct bpf_sock_ops_kern, sk));
10547 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10548 offsetof(struct sock_common,
10549 skc_v6_daddr.s6_addr32[0]) +
10550 off);
10551 #else
10552 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10553 #endif
10554 break;
10555
10556 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10557 offsetof(struct bpf_sock_ops, local_ip6[3]):
10558 #if IS_ENABLED(CONFIG_IPV6)
10559 BUILD_BUG_ON(sizeof_field(struct sock_common,
10560 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10561
10562 off = si->off;
10563 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10564 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10565 struct bpf_sock_ops_kern, sk),
10566 si->dst_reg, si->src_reg,
10567 offsetof(struct bpf_sock_ops_kern, sk));
10568 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10569 offsetof(struct sock_common,
10570 skc_v6_rcv_saddr.s6_addr32[0]) +
10571 off);
10572 #else
10573 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10574 #endif
10575 break;
10576
10577 case offsetof(struct bpf_sock_ops, remote_port):
10578 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10579
10580 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10581 struct bpf_sock_ops_kern, sk),
10582 si->dst_reg, si->src_reg,
10583 offsetof(struct bpf_sock_ops_kern, sk));
10584 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10585 offsetof(struct sock_common, skc_dport));
10586 #ifndef __BIG_ENDIAN_BITFIELD
10587 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10588 #endif
10589 break;
10590
10591 case offsetof(struct bpf_sock_ops, local_port):
10592 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10593
10594 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10595 struct bpf_sock_ops_kern, sk),
10596 si->dst_reg, si->src_reg,
10597 offsetof(struct bpf_sock_ops_kern, sk));
10598 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10599 offsetof(struct sock_common, skc_num));
10600 break;
10601
10602 case offsetof(struct bpf_sock_ops, is_fullsock):
10603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10604 struct bpf_sock_ops_kern,
10605 is_fullsock),
10606 si->dst_reg, si->src_reg,
10607 offsetof(struct bpf_sock_ops_kern,
10608 is_fullsock));
10609 break;
10610
10611 case offsetof(struct bpf_sock_ops, state):
10612 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10613
10614 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10615 struct bpf_sock_ops_kern, sk),
10616 si->dst_reg, si->src_reg,
10617 offsetof(struct bpf_sock_ops_kern, sk));
10618 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10619 offsetof(struct sock_common, skc_state));
10620 break;
10621
10622 case offsetof(struct bpf_sock_ops, rtt_min):
10623 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10624 sizeof(struct minmax));
10625 BUILD_BUG_ON(sizeof(struct minmax) <
10626 sizeof(struct minmax_sample));
10627
10628 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10629 struct bpf_sock_ops_kern, sk),
10630 si->dst_reg, si->src_reg,
10631 offsetof(struct bpf_sock_ops_kern, sk));
10632 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10633 offsetof(struct tcp_sock, rtt_min) +
10634 sizeof_field(struct minmax_sample, t));
10635 break;
10636
10637 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10638 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10639 struct tcp_sock);
10640 break;
10641
10642 case offsetof(struct bpf_sock_ops, sk_txhash):
10643 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10644 struct sock, type);
10645 break;
10646 case offsetof(struct bpf_sock_ops, snd_cwnd):
10647 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10648 break;
10649 case offsetof(struct bpf_sock_ops, srtt_us):
10650 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10651 break;
10652 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10653 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10654 break;
10655 case offsetof(struct bpf_sock_ops, rcv_nxt):
10656 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10657 break;
10658 case offsetof(struct bpf_sock_ops, snd_nxt):
10659 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10660 break;
10661 case offsetof(struct bpf_sock_ops, snd_una):
10662 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10663 break;
10664 case offsetof(struct bpf_sock_ops, mss_cache):
10665 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10666 break;
10667 case offsetof(struct bpf_sock_ops, ecn_flags):
10668 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10669 break;
10670 case offsetof(struct bpf_sock_ops, rate_delivered):
10671 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10672 break;
10673 case offsetof(struct bpf_sock_ops, rate_interval_us):
10674 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10675 break;
10676 case offsetof(struct bpf_sock_ops, packets_out):
10677 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10678 break;
10679 case offsetof(struct bpf_sock_ops, retrans_out):
10680 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10681 break;
10682 case offsetof(struct bpf_sock_ops, total_retrans):
10683 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10684 break;
10685 case offsetof(struct bpf_sock_ops, segs_in):
10686 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10687 break;
10688 case offsetof(struct bpf_sock_ops, data_segs_in):
10689 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10690 break;
10691 case offsetof(struct bpf_sock_ops, segs_out):
10692 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10693 break;
10694 case offsetof(struct bpf_sock_ops, data_segs_out):
10695 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10696 break;
10697 case offsetof(struct bpf_sock_ops, lost_out):
10698 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10699 break;
10700 case offsetof(struct bpf_sock_ops, sacked_out):
10701 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10702 break;
10703 case offsetof(struct bpf_sock_ops, bytes_received):
10704 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10705 break;
10706 case offsetof(struct bpf_sock_ops, bytes_acked):
10707 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10708 break;
10709 case offsetof(struct bpf_sock_ops, sk):
10710 SOCK_OPS_GET_SK();
10711 break;
10712 case offsetof(struct bpf_sock_ops, skb_data_end):
10713 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10714 skb_data_end),
10715 si->dst_reg, si->src_reg,
10716 offsetof(struct bpf_sock_ops_kern,
10717 skb_data_end));
10718 break;
10719 case offsetof(struct bpf_sock_ops, skb_data):
10720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10721 skb),
10722 si->dst_reg, si->src_reg,
10723 offsetof(struct bpf_sock_ops_kern,
10724 skb));
10725 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10726 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10727 si->dst_reg, si->dst_reg,
10728 offsetof(struct sk_buff, data));
10729 break;
10730 case offsetof(struct bpf_sock_ops, skb_len):
10731 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10732 skb),
10733 si->dst_reg, si->src_reg,
10734 offsetof(struct bpf_sock_ops_kern,
10735 skb));
10736 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10737 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10738 si->dst_reg, si->dst_reg,
10739 offsetof(struct sk_buff, len));
10740 break;
10741 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10742 off = offsetof(struct sk_buff, cb);
10743 off += offsetof(struct tcp_skb_cb, tcp_flags);
10744 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10745 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10746 skb),
10747 si->dst_reg, si->src_reg,
10748 offsetof(struct bpf_sock_ops_kern,
10749 skb));
10750 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10751 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10752 tcp_flags),
10753 si->dst_reg, si->dst_reg, off);
10754 break;
10755 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10756 struct bpf_insn *jmp_on_null_skb;
10757
10758 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10759 skb),
10760 si->dst_reg, si->src_reg,
10761 offsetof(struct bpf_sock_ops_kern,
10762 skb));
10763 /* Reserve one insn to test skb == NULL */
10764 jmp_on_null_skb = insn++;
10765 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10766 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10767 bpf_target_off(struct skb_shared_info,
10768 hwtstamps, 8,
10769 target_size));
10770 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10771 insn - jmp_on_null_skb - 1);
10772 break;
10773 }
10774 }
10775 return insn - insn_buf;
10776 }
10777
10778 /* data_end = skb->data + skb_headlen() */
bpf_convert_data_end_access(const struct bpf_insn * si,struct bpf_insn * insn)10779 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10780 struct bpf_insn *insn)
10781 {
10782 int reg;
10783 int temp_reg_off = offsetof(struct sk_buff, cb) +
10784 offsetof(struct sk_skb_cb, temp_reg);
10785
10786 if (si->src_reg == si->dst_reg) {
10787 /* We need an extra register, choose and save a register. */
10788 reg = BPF_REG_9;
10789 if (si->src_reg == reg || si->dst_reg == reg)
10790 reg--;
10791 if (si->src_reg == reg || si->dst_reg == reg)
10792 reg--;
10793 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10794 } else {
10795 reg = si->dst_reg;
10796 }
10797
10798 /* reg = skb->data */
10799 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10800 reg, si->src_reg,
10801 offsetof(struct sk_buff, data));
10802 /* AX = skb->len */
10803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10804 BPF_REG_AX, si->src_reg,
10805 offsetof(struct sk_buff, len));
10806 /* reg = skb->data + skb->len */
10807 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10808 /* AX = skb->data_len */
10809 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10810 BPF_REG_AX, si->src_reg,
10811 offsetof(struct sk_buff, data_len));
10812
10813 /* reg = skb->data + skb->len - skb->data_len */
10814 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10815
10816 if (si->src_reg == si->dst_reg) {
10817 /* Restore the saved register */
10818 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10819 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10820 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10821 }
10822
10823 return insn;
10824 }
10825
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)10826 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10827 const struct bpf_insn *si,
10828 struct bpf_insn *insn_buf,
10829 struct bpf_prog *prog, u32 *target_size)
10830 {
10831 struct bpf_insn *insn = insn_buf;
10832 int off;
10833
10834 switch (si->off) {
10835 case offsetof(struct __sk_buff, data_end):
10836 insn = bpf_convert_data_end_access(si, insn);
10837 break;
10838 case offsetof(struct __sk_buff, cb[0]) ...
10839 offsetofend(struct __sk_buff, cb[4]) - 1:
10840 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10841 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10842 offsetof(struct sk_skb_cb, data)) %
10843 sizeof(__u64));
10844
10845 prog->cb_access = 1;
10846 off = si->off;
10847 off -= offsetof(struct __sk_buff, cb[0]);
10848 off += offsetof(struct sk_buff, cb);
10849 off += offsetof(struct sk_skb_cb, data);
10850 if (type == BPF_WRITE)
10851 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10852 else
10853 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10854 si->src_reg, off);
10855 break;
10856
10857
10858 default:
10859 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10860 target_size);
10861 }
10862
10863 return insn - insn_buf;
10864 }
10865
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)10866 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10867 const struct bpf_insn *si,
10868 struct bpf_insn *insn_buf,
10869 struct bpf_prog *prog, u32 *target_size)
10870 {
10871 struct bpf_insn *insn = insn_buf;
10872 #if IS_ENABLED(CONFIG_IPV6)
10873 int off;
10874 #endif
10875
10876 /* convert ctx uses the fact sg element is first in struct */
10877 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10878
10879 switch (si->off) {
10880 case offsetof(struct sk_msg_md, data):
10881 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10882 si->dst_reg, si->src_reg,
10883 offsetof(struct sk_msg, data));
10884 break;
10885 case offsetof(struct sk_msg_md, data_end):
10886 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10887 si->dst_reg, si->src_reg,
10888 offsetof(struct sk_msg, data_end));
10889 break;
10890 case offsetof(struct sk_msg_md, family):
10891 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10892
10893 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10894 struct sk_msg, sk),
10895 si->dst_reg, si->src_reg,
10896 offsetof(struct sk_msg, sk));
10897 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10898 offsetof(struct sock_common, skc_family));
10899 break;
10900
10901 case offsetof(struct sk_msg_md, remote_ip4):
10902 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10903
10904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10905 struct sk_msg, sk),
10906 si->dst_reg, si->src_reg,
10907 offsetof(struct sk_msg, sk));
10908 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10909 offsetof(struct sock_common, skc_daddr));
10910 break;
10911
10912 case offsetof(struct sk_msg_md, local_ip4):
10913 BUILD_BUG_ON(sizeof_field(struct sock_common,
10914 skc_rcv_saddr) != 4);
10915
10916 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10917 struct sk_msg, sk),
10918 si->dst_reg, si->src_reg,
10919 offsetof(struct sk_msg, sk));
10920 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10921 offsetof(struct sock_common,
10922 skc_rcv_saddr));
10923 break;
10924
10925 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10926 offsetof(struct sk_msg_md, remote_ip6[3]):
10927 #if IS_ENABLED(CONFIG_IPV6)
10928 BUILD_BUG_ON(sizeof_field(struct sock_common,
10929 skc_v6_daddr.s6_addr32[0]) != 4);
10930
10931 off = si->off;
10932 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10933 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10934 struct sk_msg, sk),
10935 si->dst_reg, si->src_reg,
10936 offsetof(struct sk_msg, sk));
10937 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10938 offsetof(struct sock_common,
10939 skc_v6_daddr.s6_addr32[0]) +
10940 off);
10941 #else
10942 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10943 #endif
10944 break;
10945
10946 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10947 offsetof(struct sk_msg_md, local_ip6[3]):
10948 #if IS_ENABLED(CONFIG_IPV6)
10949 BUILD_BUG_ON(sizeof_field(struct sock_common,
10950 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10951
10952 off = si->off;
10953 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10954 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10955 struct sk_msg, sk),
10956 si->dst_reg, si->src_reg,
10957 offsetof(struct sk_msg, sk));
10958 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10959 offsetof(struct sock_common,
10960 skc_v6_rcv_saddr.s6_addr32[0]) +
10961 off);
10962 #else
10963 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10964 #endif
10965 break;
10966
10967 case offsetof(struct sk_msg_md, remote_port):
10968 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10969
10970 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10971 struct sk_msg, sk),
10972 si->dst_reg, si->src_reg,
10973 offsetof(struct sk_msg, sk));
10974 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10975 offsetof(struct sock_common, skc_dport));
10976 #ifndef __BIG_ENDIAN_BITFIELD
10977 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10978 #endif
10979 break;
10980
10981 case offsetof(struct sk_msg_md, local_port):
10982 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10983
10984 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10985 struct sk_msg, sk),
10986 si->dst_reg, si->src_reg,
10987 offsetof(struct sk_msg, sk));
10988 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10989 offsetof(struct sock_common, skc_num));
10990 break;
10991
10992 case offsetof(struct sk_msg_md, size):
10993 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10994 si->dst_reg, si->src_reg,
10995 offsetof(struct sk_msg_sg, size));
10996 break;
10997
10998 case offsetof(struct sk_msg_md, sk):
10999 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
11000 si->dst_reg, si->src_reg,
11001 offsetof(struct sk_msg, sk));
11002 break;
11003 }
11004
11005 return insn - insn_buf;
11006 }
11007
11008 const struct bpf_verifier_ops sk_filter_verifier_ops = {
11009 .get_func_proto = sk_filter_func_proto,
11010 .is_valid_access = sk_filter_is_valid_access,
11011 .convert_ctx_access = bpf_convert_ctx_access,
11012 .gen_ld_abs = bpf_gen_ld_abs,
11013 };
11014
11015 const struct bpf_prog_ops sk_filter_prog_ops = {
11016 .test_run = bpf_prog_test_run_skb,
11017 };
11018
11019 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
11020 .get_func_proto = tc_cls_act_func_proto,
11021 .is_valid_access = tc_cls_act_is_valid_access,
11022 .convert_ctx_access = tc_cls_act_convert_ctx_access,
11023 .gen_prologue = tc_cls_act_prologue,
11024 .gen_ld_abs = bpf_gen_ld_abs,
11025 .btf_struct_access = tc_cls_act_btf_struct_access,
11026 };
11027
11028 const struct bpf_prog_ops tc_cls_act_prog_ops = {
11029 .test_run = bpf_prog_test_run_skb,
11030 };
11031
11032 const struct bpf_verifier_ops xdp_verifier_ops = {
11033 .get_func_proto = xdp_func_proto,
11034 .is_valid_access = xdp_is_valid_access,
11035 .convert_ctx_access = xdp_convert_ctx_access,
11036 .gen_prologue = bpf_noop_prologue,
11037 .btf_struct_access = xdp_btf_struct_access,
11038 };
11039
11040 const struct bpf_prog_ops xdp_prog_ops = {
11041 .test_run = bpf_prog_test_run_xdp,
11042 };
11043
11044 const struct bpf_verifier_ops cg_skb_verifier_ops = {
11045 .get_func_proto = cg_skb_func_proto,
11046 .is_valid_access = cg_skb_is_valid_access,
11047 .convert_ctx_access = bpf_convert_ctx_access,
11048 };
11049
11050 const struct bpf_prog_ops cg_skb_prog_ops = {
11051 .test_run = bpf_prog_test_run_skb,
11052 };
11053
11054 const struct bpf_verifier_ops lwt_in_verifier_ops = {
11055 .get_func_proto = lwt_in_func_proto,
11056 .is_valid_access = lwt_is_valid_access,
11057 .convert_ctx_access = bpf_convert_ctx_access,
11058 };
11059
11060 const struct bpf_prog_ops lwt_in_prog_ops = {
11061 .test_run = bpf_prog_test_run_skb,
11062 };
11063
11064 const struct bpf_verifier_ops lwt_out_verifier_ops = {
11065 .get_func_proto = lwt_out_func_proto,
11066 .is_valid_access = lwt_is_valid_access,
11067 .convert_ctx_access = bpf_convert_ctx_access,
11068 };
11069
11070 const struct bpf_prog_ops lwt_out_prog_ops = {
11071 .test_run = bpf_prog_test_run_skb,
11072 };
11073
11074 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
11075 .get_func_proto = lwt_xmit_func_proto,
11076 .is_valid_access = lwt_is_valid_access,
11077 .convert_ctx_access = bpf_convert_ctx_access,
11078 .gen_prologue = tc_cls_act_prologue,
11079 };
11080
11081 const struct bpf_prog_ops lwt_xmit_prog_ops = {
11082 .test_run = bpf_prog_test_run_skb,
11083 };
11084
11085 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
11086 .get_func_proto = lwt_seg6local_func_proto,
11087 .is_valid_access = lwt_is_valid_access,
11088 .convert_ctx_access = bpf_convert_ctx_access,
11089 };
11090
11091 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
11092 };
11093
11094 const struct bpf_verifier_ops cg_sock_verifier_ops = {
11095 .get_func_proto = sock_filter_func_proto,
11096 .is_valid_access = sock_filter_is_valid_access,
11097 .convert_ctx_access = bpf_sock_convert_ctx_access,
11098 };
11099
11100 const struct bpf_prog_ops cg_sock_prog_ops = {
11101 };
11102
11103 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
11104 .get_func_proto = sock_addr_func_proto,
11105 .is_valid_access = sock_addr_is_valid_access,
11106 .convert_ctx_access = sock_addr_convert_ctx_access,
11107 };
11108
11109 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
11110 };
11111
11112 const struct bpf_verifier_ops sock_ops_verifier_ops = {
11113 .get_func_proto = sock_ops_func_proto,
11114 .is_valid_access = sock_ops_is_valid_access,
11115 .convert_ctx_access = sock_ops_convert_ctx_access,
11116 };
11117
11118 const struct bpf_prog_ops sock_ops_prog_ops = {
11119 };
11120
11121 const struct bpf_verifier_ops sk_skb_verifier_ops = {
11122 .get_func_proto = sk_skb_func_proto,
11123 .is_valid_access = sk_skb_is_valid_access,
11124 .convert_ctx_access = sk_skb_convert_ctx_access,
11125 .gen_prologue = sk_skb_prologue,
11126 };
11127
11128 const struct bpf_prog_ops sk_skb_prog_ops = {
11129 };
11130
11131 const struct bpf_verifier_ops sk_msg_verifier_ops = {
11132 .get_func_proto = sk_msg_func_proto,
11133 .is_valid_access = sk_msg_is_valid_access,
11134 .convert_ctx_access = sk_msg_convert_ctx_access,
11135 .gen_prologue = bpf_noop_prologue,
11136 };
11137
11138 const struct bpf_prog_ops sk_msg_prog_ops = {
11139 };
11140
11141 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
11142 .get_func_proto = flow_dissector_func_proto,
11143 .is_valid_access = flow_dissector_is_valid_access,
11144 .convert_ctx_access = flow_dissector_convert_ctx_access,
11145 };
11146
11147 const struct bpf_prog_ops flow_dissector_prog_ops = {
11148 .test_run = bpf_prog_test_run_flow_dissector,
11149 };
11150
sk_detach_filter(struct sock * sk)11151 int sk_detach_filter(struct sock *sk)
11152 {
11153 int ret = -ENOENT;
11154 struct sk_filter *filter;
11155
11156 if (sock_flag(sk, SOCK_FILTER_LOCKED))
11157 return -EPERM;
11158
11159 filter = rcu_dereference_protected(sk->sk_filter,
11160 lockdep_sock_is_held(sk));
11161 if (filter) {
11162 RCU_INIT_POINTER(sk->sk_filter, NULL);
11163 sk_filter_uncharge(sk, filter);
11164 ret = 0;
11165 }
11166
11167 return ret;
11168 }
11169 EXPORT_SYMBOL_GPL(sk_detach_filter);
11170
sk_get_filter(struct sock * sk,sockptr_t optval,unsigned int len)11171 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11172 {
11173 struct sock_fprog_kern *fprog;
11174 struct sk_filter *filter;
11175 int ret = 0;
11176
11177 sockopt_lock_sock(sk);
11178 filter = rcu_dereference_protected(sk->sk_filter,
11179 lockdep_sock_is_held(sk));
11180 if (!filter)
11181 goto out;
11182
11183 /* We're copying the filter that has been originally attached,
11184 * so no conversion/decode needed anymore. eBPF programs that
11185 * have no original program cannot be dumped through this.
11186 */
11187 ret = -EACCES;
11188 fprog = filter->prog->orig_prog;
11189 if (!fprog)
11190 goto out;
11191
11192 ret = fprog->len;
11193 if (!len)
11194 /* User space only enquires number of filter blocks. */
11195 goto out;
11196
11197 ret = -EINVAL;
11198 if (len < fprog->len)
11199 goto out;
11200
11201 ret = -EFAULT;
11202 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11203 goto out;
11204
11205 /* Instead of bytes, the API requests to return the number
11206 * of filter blocks.
11207 */
11208 ret = fprog->len;
11209 out:
11210 sockopt_release_sock(sk);
11211 return ret;
11212 }
11213
11214 #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)11215 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11216 struct sock_reuseport *reuse,
11217 struct sock *sk, struct sk_buff *skb,
11218 struct sock *migrating_sk,
11219 u32 hash)
11220 {
11221 reuse_kern->skb = skb;
11222 reuse_kern->sk = sk;
11223 reuse_kern->selected_sk = NULL;
11224 reuse_kern->migrating_sk = migrating_sk;
11225 reuse_kern->data_end = skb->data + skb_headlen(skb);
11226 reuse_kern->hash = hash;
11227 reuse_kern->reuseport_id = reuse->reuseport_id;
11228 reuse_kern->bind_inany = reuse->bind_inany;
11229 }
11230
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)11231 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11232 struct bpf_prog *prog, struct sk_buff *skb,
11233 struct sock *migrating_sk,
11234 u32 hash)
11235 {
11236 struct sk_reuseport_kern reuse_kern;
11237 enum sk_action action;
11238
11239 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11240 action = bpf_prog_run(prog, &reuse_kern);
11241
11242 if (action == SK_PASS)
11243 return reuse_kern.selected_sk;
11244 else
11245 return ERR_PTR(-ECONNREFUSED);
11246 }
11247
BPF_CALL_4(sk_select_reuseport,struct sk_reuseport_kern *,reuse_kern,struct bpf_map *,map,void *,key,u32,flags)11248 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11249 struct bpf_map *, map, void *, key, u32, flags)
11250 {
11251 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11252 struct sock_reuseport *reuse;
11253 struct sock *selected_sk;
11254 int err;
11255
11256 selected_sk = map->ops->map_lookup_elem(map, key);
11257 if (!selected_sk)
11258 return -ENOENT;
11259
11260 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11261 if (!reuse) {
11262 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11263 * The only (!reuse) case here is - the sk has already been
11264 * unhashed (e.g. by close()), so treat it as -ENOENT.
11265 *
11266 * Other maps (e.g. sock_map) do not provide this guarantee and
11267 * the sk may never be in the reuseport group to begin with.
11268 */
11269 err = is_sockarray ? -ENOENT : -EINVAL;
11270 goto error;
11271 }
11272
11273 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11274 struct sock *sk = reuse_kern->sk;
11275
11276 if (sk->sk_protocol != selected_sk->sk_protocol) {
11277 err = -EPROTOTYPE;
11278 } else if (sk->sk_family != selected_sk->sk_family) {
11279 err = -EAFNOSUPPORT;
11280 } else {
11281 /* Catch all. Likely bound to a different sockaddr. */
11282 err = -EBADFD;
11283 }
11284 goto error;
11285 }
11286
11287 reuse_kern->selected_sk = selected_sk;
11288
11289 return 0;
11290 error:
11291 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11292 if (sk_is_refcounted(selected_sk))
11293 sock_put(selected_sk);
11294
11295 return err;
11296 }
11297
11298 static const struct bpf_func_proto sk_select_reuseport_proto = {
11299 .func = sk_select_reuseport,
11300 .gpl_only = false,
11301 .ret_type = RET_INTEGER,
11302 .arg1_type = ARG_PTR_TO_CTX,
11303 .arg2_type = ARG_CONST_MAP_PTR,
11304 .arg3_type = ARG_PTR_TO_MAP_KEY,
11305 .arg4_type = ARG_ANYTHING,
11306 };
11307
BPF_CALL_4(sk_reuseport_load_bytes,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len)11308 BPF_CALL_4(sk_reuseport_load_bytes,
11309 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11310 void *, to, u32, len)
11311 {
11312 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11313 }
11314
11315 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11316 .func = sk_reuseport_load_bytes,
11317 .gpl_only = false,
11318 .ret_type = RET_INTEGER,
11319 .arg1_type = ARG_PTR_TO_CTX,
11320 .arg2_type = ARG_ANYTHING,
11321 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11322 .arg4_type = ARG_CONST_SIZE,
11323 };
11324
BPF_CALL_5(sk_reuseport_load_bytes_relative,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len,u32,start_header)11325 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11326 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11327 void *, to, u32, len, u32, start_header)
11328 {
11329 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11330 len, start_header);
11331 }
11332
11333 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11334 .func = sk_reuseport_load_bytes_relative,
11335 .gpl_only = false,
11336 .ret_type = RET_INTEGER,
11337 .arg1_type = ARG_PTR_TO_CTX,
11338 .arg2_type = ARG_ANYTHING,
11339 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11340 .arg4_type = ARG_CONST_SIZE,
11341 .arg5_type = ARG_ANYTHING,
11342 };
11343
11344 static const struct bpf_func_proto *
sk_reuseport_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11345 sk_reuseport_func_proto(enum bpf_func_id func_id,
11346 const struct bpf_prog *prog)
11347 {
11348 switch (func_id) {
11349 case BPF_FUNC_sk_select_reuseport:
11350 return &sk_select_reuseport_proto;
11351 case BPF_FUNC_skb_load_bytes:
11352 return &sk_reuseport_load_bytes_proto;
11353 case BPF_FUNC_skb_load_bytes_relative:
11354 return &sk_reuseport_load_bytes_relative_proto;
11355 case BPF_FUNC_get_socket_cookie:
11356 return &bpf_get_socket_ptr_cookie_proto;
11357 case BPF_FUNC_ktime_get_coarse_ns:
11358 return &bpf_ktime_get_coarse_ns_proto;
11359 default:
11360 return bpf_base_func_proto(func_id, prog);
11361 }
11362 }
11363
11364 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)11365 sk_reuseport_is_valid_access(int off, int size,
11366 enum bpf_access_type type,
11367 const struct bpf_prog *prog,
11368 struct bpf_insn_access_aux *info)
11369 {
11370 const u32 size_default = sizeof(__u32);
11371
11372 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11373 off % size || type != BPF_READ)
11374 return false;
11375
11376 switch (off) {
11377 case offsetof(struct sk_reuseport_md, data):
11378 info->reg_type = PTR_TO_PACKET;
11379 return size == sizeof(__u64);
11380
11381 case offsetof(struct sk_reuseport_md, data_end):
11382 info->reg_type = PTR_TO_PACKET_END;
11383 return size == sizeof(__u64);
11384
11385 case offsetof(struct sk_reuseport_md, hash):
11386 return size == size_default;
11387
11388 case offsetof(struct sk_reuseport_md, sk):
11389 info->reg_type = PTR_TO_SOCKET;
11390 return size == sizeof(__u64);
11391
11392 case offsetof(struct sk_reuseport_md, migrating_sk):
11393 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11394 return size == sizeof(__u64);
11395
11396 /* Fields that allow narrowing */
11397 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11398 if (size < sizeof_field(struct sk_buff, protocol))
11399 return false;
11400 fallthrough;
11401 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11402 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11403 case bpf_ctx_range(struct sk_reuseport_md, len):
11404 bpf_ctx_record_field_size(info, size_default);
11405 return bpf_ctx_narrow_access_ok(off, size, size_default);
11406
11407 default:
11408 return false;
11409 }
11410 }
11411
11412 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11413 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11414 si->dst_reg, si->src_reg, \
11415 bpf_target_off(struct sk_reuseport_kern, F, \
11416 sizeof_field(struct sk_reuseport_kern, F), \
11417 target_size)); \
11418 })
11419
11420 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11421 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11422 struct sk_buff, \
11423 skb, \
11424 SKB_FIELD)
11425
11426 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11427 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11428 struct sock, \
11429 sk, \
11430 SK_FIELD)
11431
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)11432 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11433 const struct bpf_insn *si,
11434 struct bpf_insn *insn_buf,
11435 struct bpf_prog *prog,
11436 u32 *target_size)
11437 {
11438 struct bpf_insn *insn = insn_buf;
11439
11440 switch (si->off) {
11441 case offsetof(struct sk_reuseport_md, data):
11442 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11443 break;
11444
11445 case offsetof(struct sk_reuseport_md, len):
11446 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11447 break;
11448
11449 case offsetof(struct sk_reuseport_md, eth_protocol):
11450 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11451 break;
11452
11453 case offsetof(struct sk_reuseport_md, ip_protocol):
11454 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11455 break;
11456
11457 case offsetof(struct sk_reuseport_md, data_end):
11458 SK_REUSEPORT_LOAD_FIELD(data_end);
11459 break;
11460
11461 case offsetof(struct sk_reuseport_md, hash):
11462 SK_REUSEPORT_LOAD_FIELD(hash);
11463 break;
11464
11465 case offsetof(struct sk_reuseport_md, bind_inany):
11466 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11467 break;
11468
11469 case offsetof(struct sk_reuseport_md, sk):
11470 SK_REUSEPORT_LOAD_FIELD(sk);
11471 break;
11472
11473 case offsetof(struct sk_reuseport_md, migrating_sk):
11474 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11475 break;
11476 }
11477
11478 return insn - insn_buf;
11479 }
11480
11481 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11482 .get_func_proto = sk_reuseport_func_proto,
11483 .is_valid_access = sk_reuseport_is_valid_access,
11484 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11485 };
11486
11487 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11488 };
11489
11490 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11491 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11492
BPF_CALL_3(bpf_sk_lookup_assign,struct bpf_sk_lookup_kern *,ctx,struct sock *,sk,u64,flags)11493 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11494 struct sock *, sk, u64, flags)
11495 {
11496 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11497 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11498 return -EINVAL;
11499 if (unlikely(sk && sk_is_refcounted(sk)))
11500 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11501 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11502 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11503 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11504 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11505
11506 /* Check if socket is suitable for packet L3/L4 protocol */
11507 if (sk && sk->sk_protocol != ctx->protocol)
11508 return -EPROTOTYPE;
11509 if (sk && sk->sk_family != ctx->family &&
11510 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11511 return -EAFNOSUPPORT;
11512
11513 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11514 return -EEXIST;
11515
11516 /* Select socket as lookup result */
11517 ctx->selected_sk = sk;
11518 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11519 return 0;
11520 }
11521
11522 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11523 .func = bpf_sk_lookup_assign,
11524 .gpl_only = false,
11525 .ret_type = RET_INTEGER,
11526 .arg1_type = ARG_PTR_TO_CTX,
11527 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11528 .arg3_type = ARG_ANYTHING,
11529 };
11530
11531 static const struct bpf_func_proto *
sk_lookup_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11532 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11533 {
11534 switch (func_id) {
11535 case BPF_FUNC_perf_event_output:
11536 return &bpf_event_output_data_proto;
11537 case BPF_FUNC_sk_assign:
11538 return &bpf_sk_lookup_assign_proto;
11539 case BPF_FUNC_sk_release:
11540 return &bpf_sk_release_proto;
11541 default:
11542 return bpf_sk_base_func_proto(func_id, prog);
11543 }
11544 }
11545
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)11546 static bool sk_lookup_is_valid_access(int off, int size,
11547 enum bpf_access_type type,
11548 const struct bpf_prog *prog,
11549 struct bpf_insn_access_aux *info)
11550 {
11551 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11552 return false;
11553 if (off % size != 0)
11554 return false;
11555 if (type != BPF_READ)
11556 return false;
11557
11558 switch (off) {
11559 case offsetof(struct bpf_sk_lookup, sk):
11560 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11561 return size == sizeof(__u64);
11562
11563 case bpf_ctx_range(struct bpf_sk_lookup, family):
11564 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11565 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11566 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11567 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11568 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11569 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11570 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11571 bpf_ctx_record_field_size(info, sizeof(__u32));
11572 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11573
11574 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11575 /* Allow 4-byte access to 2-byte field for backward compatibility */
11576 if (size == sizeof(__u32))
11577 return true;
11578 bpf_ctx_record_field_size(info, sizeof(__be16));
11579 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11580
11581 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11582 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11583 /* Allow access to zero padding for backward compatibility */
11584 bpf_ctx_record_field_size(info, sizeof(__u16));
11585 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11586
11587 default:
11588 return false;
11589 }
11590 }
11591
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)11592 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11593 const struct bpf_insn *si,
11594 struct bpf_insn *insn_buf,
11595 struct bpf_prog *prog,
11596 u32 *target_size)
11597 {
11598 struct bpf_insn *insn = insn_buf;
11599
11600 switch (si->off) {
11601 case offsetof(struct bpf_sk_lookup, sk):
11602 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11603 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11604 break;
11605
11606 case offsetof(struct bpf_sk_lookup, family):
11607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11608 bpf_target_off(struct bpf_sk_lookup_kern,
11609 family, 2, target_size));
11610 break;
11611
11612 case offsetof(struct bpf_sk_lookup, protocol):
11613 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11614 bpf_target_off(struct bpf_sk_lookup_kern,
11615 protocol, 2, target_size));
11616 break;
11617
11618 case offsetof(struct bpf_sk_lookup, remote_ip4):
11619 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11620 bpf_target_off(struct bpf_sk_lookup_kern,
11621 v4.saddr, 4, target_size));
11622 break;
11623
11624 case offsetof(struct bpf_sk_lookup, local_ip4):
11625 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11626 bpf_target_off(struct bpf_sk_lookup_kern,
11627 v4.daddr, 4, target_size));
11628 break;
11629
11630 case bpf_ctx_range_till(struct bpf_sk_lookup,
11631 remote_ip6[0], remote_ip6[3]): {
11632 #if IS_ENABLED(CONFIG_IPV6)
11633 int off = si->off;
11634
11635 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11636 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11637 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11638 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11639 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11640 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11641 #else
11642 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11643 #endif
11644 break;
11645 }
11646 case bpf_ctx_range_till(struct bpf_sk_lookup,
11647 local_ip6[0], local_ip6[3]): {
11648 #if IS_ENABLED(CONFIG_IPV6)
11649 int off = si->off;
11650
11651 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11652 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11653 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11654 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11655 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11656 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11657 #else
11658 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11659 #endif
11660 break;
11661 }
11662 case offsetof(struct bpf_sk_lookup, remote_port):
11663 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11664 bpf_target_off(struct bpf_sk_lookup_kern,
11665 sport, 2, target_size));
11666 break;
11667
11668 case offsetofend(struct bpf_sk_lookup, remote_port):
11669 *target_size = 2;
11670 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11671 break;
11672
11673 case offsetof(struct bpf_sk_lookup, local_port):
11674 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11675 bpf_target_off(struct bpf_sk_lookup_kern,
11676 dport, 2, target_size));
11677 break;
11678
11679 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11680 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11681 bpf_target_off(struct bpf_sk_lookup_kern,
11682 ingress_ifindex, 4, target_size));
11683 break;
11684 }
11685
11686 return insn - insn_buf;
11687 }
11688
11689 const struct bpf_prog_ops sk_lookup_prog_ops = {
11690 .test_run = bpf_prog_test_run_sk_lookup,
11691 };
11692
11693 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11694 .get_func_proto = sk_lookup_func_proto,
11695 .is_valid_access = sk_lookup_is_valid_access,
11696 .convert_ctx_access = sk_lookup_convert_ctx_access,
11697 };
11698
11699 #endif /* CONFIG_INET */
11700
DEFINE_BPF_DISPATCHER(xdp)11701 DEFINE_BPF_DISPATCHER(xdp)
11702
11703 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11704 {
11705 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11706 }
11707
BTF_ID_LIST_GLOBAL(btf_sock_ids,MAX_BTF_SOCK_TYPE)11708 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11709 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11710 BTF_SOCK_TYPE_xxx
11711 #undef BTF_SOCK_TYPE
11712
11713 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11714 {
11715 /* tcp6_sock type is not generated in dwarf and hence btf,
11716 * trigger an explicit type generation here.
11717 */
11718 BTF_TYPE_EMIT(struct tcp6_sock);
11719 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11720 sk->sk_family == AF_INET6)
11721 return (unsigned long)sk;
11722
11723 return (unsigned long)NULL;
11724 }
11725
11726 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11727 .func = bpf_skc_to_tcp6_sock,
11728 .gpl_only = false,
11729 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11730 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11731 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11732 };
11733
BPF_CALL_1(bpf_skc_to_tcp_sock,struct sock *,sk)11734 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11735 {
11736 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11737 return (unsigned long)sk;
11738
11739 return (unsigned long)NULL;
11740 }
11741
11742 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11743 .func = bpf_skc_to_tcp_sock,
11744 .gpl_only = false,
11745 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11746 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11747 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11748 };
11749
BPF_CALL_1(bpf_skc_to_tcp_timewait_sock,struct sock *,sk)11750 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11751 {
11752 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11753 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11754 */
11755 BTF_TYPE_EMIT(struct inet_timewait_sock);
11756 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11757
11758 #ifdef CONFIG_INET
11759 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11760 return (unsigned long)sk;
11761 #endif
11762
11763 #if IS_BUILTIN(CONFIG_IPV6)
11764 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11765 return (unsigned long)sk;
11766 #endif
11767
11768 return (unsigned long)NULL;
11769 }
11770
11771 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11772 .func = bpf_skc_to_tcp_timewait_sock,
11773 .gpl_only = false,
11774 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11775 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11776 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11777 };
11778
BPF_CALL_1(bpf_skc_to_tcp_request_sock,struct sock *,sk)11779 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11780 {
11781 #ifdef CONFIG_INET
11782 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11783 return (unsigned long)sk;
11784 #endif
11785
11786 #if IS_BUILTIN(CONFIG_IPV6)
11787 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11788 return (unsigned long)sk;
11789 #endif
11790
11791 return (unsigned long)NULL;
11792 }
11793
11794 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11795 .func = bpf_skc_to_tcp_request_sock,
11796 .gpl_only = false,
11797 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11798 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11799 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11800 };
11801
BPF_CALL_1(bpf_skc_to_udp6_sock,struct sock *,sk)11802 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11803 {
11804 /* udp6_sock type is not generated in dwarf and hence btf,
11805 * trigger an explicit type generation here.
11806 */
11807 BTF_TYPE_EMIT(struct udp6_sock);
11808 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11809 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11810 return (unsigned long)sk;
11811
11812 return (unsigned long)NULL;
11813 }
11814
11815 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11816 .func = bpf_skc_to_udp6_sock,
11817 .gpl_only = false,
11818 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11819 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11820 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11821 };
11822
BPF_CALL_1(bpf_skc_to_unix_sock,struct sock *,sk)11823 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11824 {
11825 /* unix_sock type is not generated in dwarf and hence btf,
11826 * trigger an explicit type generation here.
11827 */
11828 BTF_TYPE_EMIT(struct unix_sock);
11829 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11830 return (unsigned long)sk;
11831
11832 return (unsigned long)NULL;
11833 }
11834
11835 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11836 .func = bpf_skc_to_unix_sock,
11837 .gpl_only = false,
11838 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11839 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11840 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11841 };
11842
BPF_CALL_1(bpf_skc_to_mptcp_sock,struct sock *,sk)11843 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11844 {
11845 BTF_TYPE_EMIT(struct mptcp_sock);
11846 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11847 }
11848
11849 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11850 .func = bpf_skc_to_mptcp_sock,
11851 .gpl_only = false,
11852 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11853 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11854 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11855 };
11856
BPF_CALL_1(bpf_sock_from_file,struct file *,file)11857 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11858 {
11859 return (unsigned long)sock_from_file(file);
11860 }
11861
11862 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11863 BTF_ID(struct, socket)
11864 BTF_ID(struct, file)
11865
11866 const struct bpf_func_proto bpf_sock_from_file_proto = {
11867 .func = bpf_sock_from_file,
11868 .gpl_only = false,
11869 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11870 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11871 .arg1_type = ARG_PTR_TO_BTF_ID,
11872 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11873 };
11874
11875 static const struct bpf_func_proto *
bpf_sk_base_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11876 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11877 {
11878 const struct bpf_func_proto *func;
11879
11880 switch (func_id) {
11881 case BPF_FUNC_skc_to_tcp6_sock:
11882 func = &bpf_skc_to_tcp6_sock_proto;
11883 break;
11884 case BPF_FUNC_skc_to_tcp_sock:
11885 func = &bpf_skc_to_tcp_sock_proto;
11886 break;
11887 case BPF_FUNC_skc_to_tcp_timewait_sock:
11888 func = &bpf_skc_to_tcp_timewait_sock_proto;
11889 break;
11890 case BPF_FUNC_skc_to_tcp_request_sock:
11891 func = &bpf_skc_to_tcp_request_sock_proto;
11892 break;
11893 case BPF_FUNC_skc_to_udp6_sock:
11894 func = &bpf_skc_to_udp6_sock_proto;
11895 break;
11896 case BPF_FUNC_skc_to_unix_sock:
11897 func = &bpf_skc_to_unix_sock_proto;
11898 break;
11899 case BPF_FUNC_skc_to_mptcp_sock:
11900 func = &bpf_skc_to_mptcp_sock_proto;
11901 break;
11902 case BPF_FUNC_ktime_get_coarse_ns:
11903 return &bpf_ktime_get_coarse_ns_proto;
11904 default:
11905 return bpf_base_func_proto(func_id, prog);
11906 }
11907
11908 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
11909 return NULL;
11910
11911 return func;
11912 }
11913
11914 __bpf_kfunc_start_defs();
bpf_dynptr_from_skb(struct __sk_buff * s,u64 flags,struct bpf_dynptr * ptr__uninit)11915 __bpf_kfunc int bpf_dynptr_from_skb(struct __sk_buff *s, u64 flags,
11916 struct bpf_dynptr *ptr__uninit)
11917 {
11918 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11919 struct sk_buff *skb = (struct sk_buff *)s;
11920
11921 if (flags) {
11922 bpf_dynptr_set_null(ptr);
11923 return -EINVAL;
11924 }
11925
11926 bpf_dynptr_init(ptr, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11927
11928 return 0;
11929 }
11930
bpf_dynptr_from_xdp(struct xdp_md * x,u64 flags,struct bpf_dynptr * ptr__uninit)11931 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_md *x, u64 flags,
11932 struct bpf_dynptr *ptr__uninit)
11933 {
11934 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11935 struct xdp_buff *xdp = (struct xdp_buff *)x;
11936
11937 if (flags) {
11938 bpf_dynptr_set_null(ptr);
11939 return -EINVAL;
11940 }
11941
11942 bpf_dynptr_init(ptr, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11943
11944 return 0;
11945 }
11946
bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern * sa_kern,const u8 * sun_path,u32 sun_path__sz)11947 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
11948 const u8 *sun_path, u32 sun_path__sz)
11949 {
11950 struct sockaddr_un *un;
11951
11952 if (sa_kern->sk->sk_family != AF_UNIX)
11953 return -EINVAL;
11954
11955 /* We do not allow changing the address to unnamed or larger than the
11956 * maximum allowed address size for a unix sockaddr.
11957 */
11958 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX)
11959 return -EINVAL;
11960
11961 un = (struct sockaddr_un *)sa_kern->uaddr;
11962 memcpy(un->sun_path, sun_path, sun_path__sz);
11963 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz;
11964
11965 return 0;
11966 }
11967
bpf_sk_assign_tcp_reqsk(struct __sk_buff * s,struct sock * sk,struct bpf_tcp_req_attrs * attrs,int attrs__sz)11968 __bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct __sk_buff *s, struct sock *sk,
11969 struct bpf_tcp_req_attrs *attrs, int attrs__sz)
11970 {
11971 #if IS_ENABLED(CONFIG_SYN_COOKIES)
11972 struct sk_buff *skb = (struct sk_buff *)s;
11973 const struct request_sock_ops *ops;
11974 struct inet_request_sock *ireq;
11975 struct tcp_request_sock *treq;
11976 struct request_sock *req;
11977 struct net *net;
11978 __u16 min_mss;
11979 u32 tsoff = 0;
11980
11981 if (attrs__sz != sizeof(*attrs) ||
11982 attrs->reserved[0] || attrs->reserved[1] || attrs->reserved[2])
11983 return -EINVAL;
11984
11985 if (!skb_at_tc_ingress(skb))
11986 return -EINVAL;
11987
11988 net = dev_net(skb->dev);
11989 if (net != sock_net(sk))
11990 return -ENETUNREACH;
11991
11992 switch (skb->protocol) {
11993 case htons(ETH_P_IP):
11994 ops = &tcp_request_sock_ops;
11995 min_mss = 536;
11996 break;
11997 #if IS_BUILTIN(CONFIG_IPV6)
11998 case htons(ETH_P_IPV6):
11999 ops = &tcp6_request_sock_ops;
12000 min_mss = IPV6_MIN_MTU - 60;
12001 break;
12002 #endif
12003 default:
12004 return -EINVAL;
12005 }
12006
12007 if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN ||
12008 sk_is_mptcp(sk))
12009 return -EINVAL;
12010
12011 if (attrs->mss < min_mss)
12012 return -EINVAL;
12013
12014 if (attrs->wscale_ok) {
12015 if (!READ_ONCE(net->ipv4.sysctl_tcp_window_scaling))
12016 return -EINVAL;
12017
12018 if (attrs->snd_wscale > TCP_MAX_WSCALE ||
12019 attrs->rcv_wscale > TCP_MAX_WSCALE)
12020 return -EINVAL;
12021 }
12022
12023 if (attrs->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
12024 return -EINVAL;
12025
12026 if (attrs->tstamp_ok) {
12027 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
12028 return -EINVAL;
12029
12030 tsoff = attrs->rcv_tsecr - tcp_ns_to_ts(attrs->usec_ts_ok, tcp_clock_ns());
12031 }
12032
12033 req = inet_reqsk_alloc(ops, sk, false);
12034 if (!req)
12035 return -ENOMEM;
12036
12037 ireq = inet_rsk(req);
12038 treq = tcp_rsk(req);
12039
12040 req->rsk_listener = sk;
12041 req->syncookie = 1;
12042 req->mss = attrs->mss;
12043 req->ts_recent = attrs->rcv_tsval;
12044
12045 ireq->snd_wscale = attrs->snd_wscale;
12046 ireq->rcv_wscale = attrs->rcv_wscale;
12047 ireq->tstamp_ok = !!attrs->tstamp_ok;
12048 ireq->sack_ok = !!attrs->sack_ok;
12049 ireq->wscale_ok = !!attrs->wscale_ok;
12050 ireq->ecn_ok = !!attrs->ecn_ok;
12051
12052 treq->req_usec_ts = !!attrs->usec_ts_ok;
12053 treq->ts_off = tsoff;
12054
12055 skb_orphan(skb);
12056 skb->sk = req_to_sk(req);
12057 skb->destructor = sock_pfree;
12058
12059 return 0;
12060 #else
12061 return -EOPNOTSUPP;
12062 #endif
12063 }
12064
12065 __bpf_kfunc_end_defs();
12066
bpf_dynptr_from_skb_rdonly(struct __sk_buff * skb,u64 flags,struct bpf_dynptr * ptr__uninit)12067 int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
12068 struct bpf_dynptr *ptr__uninit)
12069 {
12070 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
12071 int err;
12072
12073 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
12074 if (err)
12075 return err;
12076
12077 bpf_dynptr_set_rdonly(ptr);
12078
12079 return 0;
12080 }
12081
12082 BTF_KFUNCS_START(bpf_kfunc_check_set_skb)
12083 BTF_ID_FLAGS(func, bpf_dynptr_from_skb, KF_TRUSTED_ARGS)
12084 BTF_KFUNCS_END(bpf_kfunc_check_set_skb)
12085
12086 BTF_KFUNCS_START(bpf_kfunc_check_set_xdp)
12087 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
12088 BTF_KFUNCS_END(bpf_kfunc_check_set_xdp)
12089
12090 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_addr)
12091 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
12092 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_addr)
12093
12094 BTF_KFUNCS_START(bpf_kfunc_check_set_tcp_reqsk)
12095 BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk, KF_TRUSTED_ARGS)
12096 BTF_KFUNCS_END(bpf_kfunc_check_set_tcp_reqsk)
12097
12098 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
12099 .owner = THIS_MODULE,
12100 .set = &bpf_kfunc_check_set_skb,
12101 };
12102
12103 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
12104 .owner = THIS_MODULE,
12105 .set = &bpf_kfunc_check_set_xdp,
12106 };
12107
12108 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
12109 .owner = THIS_MODULE,
12110 .set = &bpf_kfunc_check_set_sock_addr,
12111 };
12112
12113 static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = {
12114 .owner = THIS_MODULE,
12115 .set = &bpf_kfunc_check_set_tcp_reqsk,
12116 };
12117
bpf_kfunc_init(void)12118 static int __init bpf_kfunc_init(void)
12119 {
12120 int ret;
12121
12122 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
12123 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
12124 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
12125 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
12126 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
12127 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
12128 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
12129 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
12130 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
12131 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
12132 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_kfunc_set_skb);
12133 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
12134 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
12135 &bpf_kfunc_set_sock_addr);
12136 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk);
12137 }
12138 late_initcall(bpf_kfunc_init);
12139
12140 __bpf_kfunc_start_defs();
12141
12142 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
12143 *
12144 * The function expects a non-NULL pointer to a socket, and invokes the
12145 * protocol specific socket destroy handlers.
12146 *
12147 * The helper can only be called from BPF contexts that have acquired the socket
12148 * locks.
12149 *
12150 * Parameters:
12151 * @sock: Pointer to socket to be destroyed
12152 *
12153 * Return:
12154 * On error, may return EPROTONOSUPPORT, EINVAL.
12155 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
12156 * 0 otherwise
12157 */
bpf_sock_destroy(struct sock_common * sock)12158 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
12159 {
12160 struct sock *sk = (struct sock *)sock;
12161
12162 /* The locking semantics that allow for synchronous execution of the
12163 * destroy handlers are only supported for TCP and UDP.
12164 * Supporting protocols will need to acquire sock lock in the BPF context
12165 * prior to invoking this kfunc.
12166 */
12167 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
12168 sk->sk_protocol != IPPROTO_UDP))
12169 return -EOPNOTSUPP;
12170
12171 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
12172 }
12173
12174 __bpf_kfunc_end_defs();
12175
12176 BTF_KFUNCS_START(bpf_sk_iter_kfunc_ids)
BTF_ID_FLAGS(func,bpf_sock_destroy,KF_TRUSTED_ARGS)12177 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
12178 BTF_KFUNCS_END(bpf_sk_iter_kfunc_ids)
12179
12180 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
12181 {
12182 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
12183 prog->expected_attach_type != BPF_TRACE_ITER)
12184 return -EACCES;
12185 return 0;
12186 }
12187
12188 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
12189 .owner = THIS_MODULE,
12190 .set = &bpf_sk_iter_kfunc_ids,
12191 .filter = tracing_iter_filter,
12192 };
12193
init_subsystem(void)12194 static int init_subsystem(void)
12195 {
12196 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
12197 }
12198 late_initcall(init_subsystem);
12199