projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'bpf: BTF support for ksyms'
[linux-2.6-block.git] / net / core / filter.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
bd4cf0ed
AS		 5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
1da177e4 7 *
bd4cf0ed
AS		 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>
1da177e4 15 *
1da177e4 16 * Andi Kleen - Fix a few bad bugs and races.
4df95ff4 17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
1da177e4
LT		 18 */
19
20#include <linux/module.h>
21#include <linux/types.h>
1da177e4
LT		 22#include <linux/mm.h>
23#include <linux/fcntl.h>
24#include <linux/socket.h>
91b8270f 25#include <linux/sock_diag.h>
1da177e4
LT		 26#include <linux/in.h>
27#include <linux/inet.h>
28#include <linux/netdevice.h>
29#include <linux/if_packet.h>
c491680f 30#include <linux/if_arp.h>
5a0e3ad6 31#include <linux/gfp.h>
d74bad4e 32#include <net/inet_common.h>
1da177e4
LT		 33#include <net/ip.h>
34#include <net/protocol.h>
4738c1db 35#include <net/netlink.h>
1da177e4 36#include <linux/skbuff.h>
604326b4 37#include <linux/skmsg.h>
1da177e4 38#include <net/sock.h>
10b89ee4 39#include <net/flow_dissector.h>
1da177e4
LT		 40#include <linux/errno.h>
41#include <linux/timer.h>
7c0f6ba6 42#include <linux/uaccess.h>
40daafc8 43#include <asm/unaligned.h>
d66f2b91 44#include <asm/cmpxchg.h>
1da177e4 45#include <linux/filter.h>
86e4ca66 46#include <linux/ratelimit.h>
46b325c7 47#include <linux/seccomp.h>
f3335031 48#include <linux/if_vlan.h>
89aa0758 49#include <linux/bpf.h>
af7ec138 50#include <linux/btf.h>
d691f9e8 51#include <net/sch_generic.h>
8d20aabe 52#include <net/cls_cgroup.h>
d3aa45ce 53#include <net/dst_metadata.h>
c46646d0 54#include <net/dst.h>
538950a1 55#include <net/sock_reuseport.h>
b1d9fc41 56#include <net/busy_poll.h>
8c4b4c7e 57#include <net/tcp.h>
12bed760 58#include <net/xfrm.h>
6acc9b43 59#include <net/udp.h>
5acaee0a 60#include <linux/bpf_trace.h>
02671e23 61#include <net/xdp_sock.h>
87f5fc7e 62#include <linux/inetdevice.h>
6acc9b43
JS		 63#include <net/inet_hashtables.h>
64#include <net/inet6_hashtables.h>
87f5fc7e 65#include <net/ip_fib.h>
5481d73f 66#include <net/nexthop.h>
87f5fc7e
DA		 67#include <net/flow.h>
68#include <net/arp.h>
fe94cc29 69#include <net/ipv6.h>
6acc9b43 70#include <net/net_namespace.h>
fe94cc29
MX		 71#include <linux/seg6_local.h>
72#include <net/seg6.h>
73#include <net/seg6_local.h>
52f27877 74#include <net/lwtunnel.h>
3616d08b 75#include <net/ipv6_stubs.h>
6ac99e8f 76#include <net/bpf_sk_storage.h>
478cfbdf 77#include <net/transp_v6.h>
c9a0f3b8 78#include <linux/btf_ids.h>
1da177e4 79
1df8f55a
MKL		 80static const struct bpf_func_proto *
81bpf_sk_base_func_proto(enum bpf_func_id func_id);
82
b1ea9ff6 83int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
4d295e54
CH		 84{
85 if (in_compat_syscall()) {
86 struct compat_sock_fprog f32;
87
88 if (len != sizeof(f32))
89 return -EINVAL;
b1ea9ff6 90 if (copy_from_sockptr(&f32, src, sizeof(f32)))
4d295e54
CH		 91 return -EFAULT;
92 memset(dst, 0, sizeof(*dst));
93 dst->len = f32.len;
94 dst->filter = compat_ptr(f32.filter);
95 } else {
96 if (len != sizeof(*dst))
97 return -EINVAL;
b1ea9ff6 98 if (copy_from_sockptr(dst, src, sizeof(*dst)))
4d295e54
CH		 99 return -EFAULT;
100 }
101
102 return 0;
103}
104EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
105
43db6d65 106/**
f4979fce 107 * sk_filter_trim_cap - run a packet through a socket filter
43db6d65
SH		 108 * @sk: sock associated with &sk_buff
109 * @skb: buffer to filter
f4979fce 110 * @cap: limit on how short the eBPF program may trim the packet
43db6d65 111 *
ff936a04
AS		 112 * Run the eBPF program and then cut skb->data to correct size returned by
113 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
43db6d65 114 * than pkt_len we keep whole skb->data. This is the socket level
ff936a04 115 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
43db6d65
SH		 116 * be accepted or -EPERM if the packet should be tossed.
117 *
118 */
f4979fce 119int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
43db6d65
SH		 120{
121 int err;
122 struct sk_filter *filter;
123
c93bdd0e
MG		 124 /*
125 * If the skb was allocated from pfmemalloc reserves, only
126 * allow SOCK_MEMALLOC sockets to use it as this socket is
127 * helping free memory
128 */
8fe809a9
ED		 129 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
130 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
c93bdd0e 131 return -ENOMEM;
8fe809a9 132 }
c11cd3a6
DM		 133 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
134 if (err)
135 return err;
136
43db6d65
SH		 137 err = security_sock_rcv_skb(sk, skb);
138 if (err)
139 return err;
140
80f8f102
ED		 141 rcu_read_lock();
142 filter = rcu_dereference(sk->sk_filter);
43db6d65 143 if (filter) {
8f917bba
WB		 144 struct sock *save_sk = skb->sk;
145 unsigned int pkt_len;
146
147 skb->sk = sk;
148 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
8f917bba 149 skb->sk = save_sk;
d1f496fd 150 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
43db6d65 151 }
80f8f102 152 rcu_read_unlock();
43db6d65
SH		 153
154 return err;
155}
f4979fce 156EXPORT_SYMBOL(sk_filter_trim_cap);
43db6d65 157
b390134c 158BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
bd4cf0ed 159{
f3694e00 160 return skb_get_poff(skb);
bd4cf0ed
AS		 161}
162
b390134c 163BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
bd4cf0ed 164{
bd4cf0ed
AS		 165 struct nlattr *nla;
166
167 if (skb_is_nonlinear(skb))
168 return 0;
169
05ab8f26
MK		 170 if (skb->len < sizeof(struct nlattr))
171 return 0;
172
30743837 173 if (a > skb->len - sizeof(struct nlattr))
bd4cf0ed
AS		 174 return 0;
175
30743837 176 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
bd4cf0ed
AS		 177 if (nla)
178 return (void *) nla - (void *) skb->data;
179
180 return 0;
181}
182
b390134c 183BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
bd4cf0ed 184{
bd4cf0ed
AS		 185 struct nlattr *nla;
186
187 if (skb_is_nonlinear(skb))
188 return 0;
189
05ab8f26
MK		 190 if (skb->len < sizeof(struct nlattr))
191 return 0;
192
30743837 193 if (a > skb->len - sizeof(struct nlattr))
bd4cf0ed
AS		 194 return 0;
195
30743837
DB		 196 nla = (struct nlattr *) &skb->data[a];
197 if (nla->nla_len > skb->len - a)
bd4cf0ed
AS		 198 return 0;
199
30743837 200 nla = nla_find_nested(nla, x);
bd4cf0ed
AS		 201 if (nla)
202 return (void *) nla - (void *) skb->data;
203
204 return 0;
205}
206
e0cea7ce
DB		 207BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
208 data, int, headlen, int, offset)
209{
210 u8 tmp, *ptr;
211 const int len = sizeof(tmp);
212
213 if (offset >= 0) {
214 if (headlen - offset >= len)
215 return *(u8 *)(data + offset);
216 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
217 return tmp;
218 } else {
219 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
220 if (likely(ptr))
221 return *(u8 *)ptr;
222 }
223
224 return -EFAULT;
225}
226
227BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
228 int, offset)
229{
230 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
231 offset);
232}
233
234BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
235 data, int, headlen, int, offset)
236{
237 u16 tmp, *ptr;
238 const int len = sizeof(tmp);
239
240 if (offset >= 0) {
241 if (headlen - offset >= len)
242 return get_unaligned_be16(data + offset);
243 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
244 return be16_to_cpu(tmp);
245 } else {
246 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
247 if (likely(ptr))
248 return get_unaligned_be16(ptr);
249 }
250
251 return -EFAULT;
252}
253
254BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
255 int, offset)
256{
257 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
258 offset);
259}
260
261BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
262 data, int, headlen, int, offset)
263{
264 u32 tmp, *ptr;
265 const int len = sizeof(tmp);
266
267 if (likely(offset >= 0)) {
268 if (headlen - offset >= len)
269 return get_unaligned_be32(data + offset);
270 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
271 return be32_to_cpu(tmp);
272 } else {
273 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
274 if (likely(ptr))
275 return get_unaligned_be32(ptr);
276 }
277
278 return -EFAULT;
279}
280
281BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
282 int, offset)
283{
284 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
285 offset);
286}
287
9bac3d6d
AS		 288static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
289 struct bpf_insn *insn_buf)
290{
291 struct bpf_insn *insn = insn_buf;
292
293 switch (skb_field) {
294 case SKF_AD_MARK:
c593642c 295 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
9bac3d6d
AS		 296
297 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
298 offsetof(struct sk_buff, mark));
299 break;
300
301 case SKF_AD_PKTTYPE:
302 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
303 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
304#ifdef __BIG_ENDIAN_BITFIELD
305 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
306#endif
307 break;
308
309 case SKF_AD_QUEUE:
c593642c 310 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
9bac3d6d
AS		 311
312 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
313 offsetof(struct sk_buff, queue_mapping));
314 break;
c2497395 315
c2497395 316 case SKF_AD_VLAN_TAG:
c593642c 317 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
c2497395
AS		 318
319 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
320 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
321 offsetof(struct sk_buff, vlan_tci));
9c212255
MM		 322 break;
323 case SKF_AD_VLAN_TAG_PRESENT:
324 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
325 if (PKT_VLAN_PRESENT_BIT)
326 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
327 if (PKT_VLAN_PRESENT_BIT < 7)
c2497395 328 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
c2497395 329 break;
9bac3d6d
AS		 330 }
331
332 return insn - insn_buf;
333}
334
bd4cf0ed 335static bool convert_bpf_extensions(struct sock_filter *fp,
2695fb55 336 struct bpf_insn **insnp)
bd4cf0ed 337{
2695fb55 338 struct bpf_insn *insn = *insnp;
9bac3d6d 339 u32 cnt;
bd4cf0ed
AS		 340
341 switch (fp->k) {
342 case SKF_AD_OFF + SKF_AD_PROTOCOL:
c593642c 343 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
0b8c707d
DB		 344
345 /* A = *(u16 *) (CTX + offsetof(protocol)) */
346 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
347 offsetof(struct sk_buff, protocol));
348 /* A = ntohs(A) [emitting a nop or swap16] */
349 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
bd4cf0ed
AS		 350 break;
351
352 case SKF_AD_OFF + SKF_AD_PKTTYPE:
9bac3d6d
AS		 353 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
354 insn += cnt - 1;
bd4cf0ed
AS		 355 break;
356
357 case SKF_AD_OFF + SKF_AD_IFINDEX:
358 case SKF_AD_OFF + SKF_AD_HATYPE:
c593642c
PB		 359 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
360 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
f8f6d679 361
f035a515 362 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
f8f6d679
DB		 363 BPF_REG_TMP, BPF_REG_CTX,
364 offsetof(struct sk_buff, dev));
365 /* if (tmp != 0) goto pc + 1 */
366 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
367 *insn++ = BPF_EXIT_INSN();
368 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
369 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
370 offsetof(struct net_device, ifindex));
371 else
372 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
373 offsetof(struct net_device, type));
bd4cf0ed
AS		 374 break;
375
376 case SKF_AD_OFF + SKF_AD_MARK:
9bac3d6d
AS		 377 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
378 insn += cnt - 1;
bd4cf0ed
AS		 379 break;
380
381 case SKF_AD_OFF + SKF_AD_RXHASH:
c593642c 382 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
bd4cf0ed 383
9739eef1
AS		 384 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
385 offsetof(struct sk_buff, hash));
bd4cf0ed
AS		 386 break;
387
388 case SKF_AD_OFF + SKF_AD_QUEUE:
9bac3d6d
AS		 389 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
390 insn += cnt - 1;
bd4cf0ed
AS		 391 break;
392
393 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
c2497395
AS		 394 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
395 BPF_REG_A, BPF_REG_CTX, insn);
396 insn += cnt - 1;
397 break;
bd4cf0ed 398
c2497395
AS		 399 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
400 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
401 BPF_REG_A, BPF_REG_CTX, insn);
402 insn += cnt - 1;
bd4cf0ed
AS		 403 break;
404
27cd5452 405 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
c593642c 406 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
27cd5452
MS		 407
408 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
409 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
410 offsetof(struct sk_buff, vlan_proto));
411 /* A = ntohs(A) [emitting a nop or swap16] */
412 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
413 break;
414
bd4cf0ed
AS		 415 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
416 case SKF_AD_OFF + SKF_AD_NLATTR:
417 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
418 case SKF_AD_OFF + SKF_AD_CPU:
4cd3675e 419 case SKF_AD_OFF + SKF_AD_RANDOM:
e430f34e 420 /* arg1 = CTX */
f8f6d679 421 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
bd4cf0ed 422 /* arg2 = A */
f8f6d679 423 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
bd4cf0ed 424 /* arg3 = X */
f8f6d679 425 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
e430f34e 426 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
bd4cf0ed
AS		 427 switch (fp->k) {
428 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
b390134c 429 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
bd4cf0ed
AS		 430 break;
431 case SKF_AD_OFF + SKF_AD_NLATTR:
b390134c 432 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
bd4cf0ed
AS		 433 break;
434 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
b390134c 435 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
bd4cf0ed
AS		 436 break;
437 case SKF_AD_OFF + SKF_AD_CPU:
b390134c 438 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
bd4cf0ed 439 break;
4cd3675e 440 case SKF_AD_OFF + SKF_AD_RANDOM:
3ad00405
DB		 441 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
442 bpf_user_rnd_init_once();
4cd3675e 443 break;
bd4cf0ed
AS		 444 }
445 break;
446
447 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
9739eef1
AS		 448 /* A ^= X */
449 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
bd4cf0ed
AS		 450 break;
451
452 default:
453 /* This is just a dummy call to avoid letting the compiler
454 * evict __bpf_call_base() as an optimization. Placed here
455 * where no-one bothers.
456 */
457 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
458 return false;
459 }
460
461 *insnp = insn;
462 return true;
463}
464
e0cea7ce
DB		 465static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
466{
467 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
468 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
469 bool endian = BPF_SIZE(fp->code) == BPF_H ||
470 BPF_SIZE(fp->code) == BPF_W;
471 bool indirect = BPF_MODE(fp->code) == BPF_IND;
472 const int ip_align = NET_IP_ALIGN;
473 struct bpf_insn *insn = *insnp;
474 int offset = fp->k;
475
476 if (!indirect &&
477 ((unaligned_ok && offset >= 0) ||
478 (!unaligned_ok && offset >= 0 &&
479 offset + ip_align >= 0 &&
480 offset + ip_align % size == 0))) {
59ee4129
DB		 481 bool ldx_off_ok = offset <= S16_MAX;
482
e0cea7ce 483 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
d8f3e978
DM		 484 if (offset)
485 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
59ee4129
DB		 486 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
487 size, 2 + endian + (!ldx_off_ok * 2));
488 if (ldx_off_ok) {
489 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
490 BPF_REG_D, offset);
491 } else {
492 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
493 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
495 BPF_REG_TMP, 0);
496 }
e0cea7ce
DB		 497 if (endian)
498 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
499 *insn++ = BPF_JMP_A(8);
500 }
501
502 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
503 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
505 if (!indirect) {
506 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
507 } else {
508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
509 if (fp->k)
510 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
511 }
512
513 switch (BPF_SIZE(fp->code)) {
514 case BPF_B:
515 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
516 break;
517 case BPF_H:
518 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
519 break;
520 case BPF_W:
521 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
522 break;
523 default:
524 return false;
525 }
526
527 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
528 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
529 *insn = BPF_EXIT_INSN();
530
531 *insnp = insn;
532 return true;
533}
534
bd4cf0ed 535/**
8fb575ca 536 * bpf_convert_filter - convert filter program
bd4cf0ed
AS		 537 * @prog: the user passed filter program
538 * @len: the length of the user passed filter program
50bbfed9 539 * @new_prog: allocated 'struct bpf_prog' or NULL
bd4cf0ed 540 * @new_len: pointer to store length of converted program
e0cea7ce 541 * @seen_ld_abs: bool whether we've seen ld_abs/ind
bd4cf0ed 542 *
1f504ec9
TK		 543 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
544 * style extended BPF (eBPF).
bd4cf0ed
AS		 545 * Conversion workflow:
546 *
547 * 1) First pass for calculating the new program length:
e0cea7ce 548 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
bd4cf0ed
AS		 549 *
550 * 2) 2nd pass to remap in two passes: 1st pass finds new
551 * jump offsets, 2nd pass remapping:
e0cea7ce 552 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
bd4cf0ed 553 */
d9e12f42 554static int bpf_convert_filter(struct sock_filter *prog, int len,
e0cea7ce
DB		 555 struct bpf_prog *new_prog, int *new_len,
556 bool *seen_ld_abs)
bd4cf0ed 557{
50bbfed9
AS		 558 int new_flen = 0, pass = 0, target, i, stack_off;
559 struct bpf_insn *new_insn, *first_insn = NULL;
bd4cf0ed
AS		 560 struct sock_filter *fp;
561 int *addrs = NULL;
562 u8 bpf_src;
563
564 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
30743837 565 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
bd4cf0ed 566
6f9a093b 567 if (len <= 0 || len > BPF_MAXINSNS)
bd4cf0ed
AS		 568 return -EINVAL;
569
570 if (new_prog) {
50bbfed9 571 first_insn = new_prog->insnsi;
658da937
DB		 572 addrs = kcalloc(len, sizeof(*addrs),
573 GFP_KERNEL | __GFP_NOWARN);
bd4cf0ed
AS		 574 if (!addrs)
575 return -ENOMEM;
576 }
577
578do_pass:
50bbfed9 579 new_insn = first_insn;
bd4cf0ed
AS		 580 fp = prog;
581
8b614aeb 582 /* Classic BPF related prologue emission. */
50bbfed9 583 if (new_prog) {
8b614aeb
DB		 584 /* Classic BPF expects A and X to be reset first. These need
585 * to be guaranteed to be the first two instructions.
586 */
1d621674
DB		 587 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
588 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
8b614aeb
DB		 589
590 /* All programs must keep CTX in callee saved BPF_REG_CTX.
591 * In eBPF case it's done by the compiler, here we need to
592 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
593 */
594 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
e0cea7ce
DB		 595 if (*seen_ld_abs) {
596 /* For packet access in classic BPF, cache skb->data
597 * in callee-saved BPF R8 and skb->len - skb->data_len
598 * (headlen) in BPF R9. Since classic BPF is read-only
599 * on CTX, we only need to cache it once.
600 */
601 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
602 BPF_REG_D, BPF_REG_CTX,
603 offsetof(struct sk_buff, data));
604 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
605 offsetof(struct sk_buff, len));
606 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
607 offsetof(struct sk_buff, data_len));
608 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
609 }
8b614aeb
DB		 610 } else {
611 new_insn += 3;
612 }
bd4cf0ed
AS		 613
614 for (i = 0; i < len; fp++, i++) {
e0cea7ce 615 struct bpf_insn tmp_insns[32] = { };
2695fb55 616 struct bpf_insn *insn = tmp_insns;
bd4cf0ed
AS		 617
618 if (addrs)
50bbfed9 619 addrs[i] = new_insn - first_insn;
bd4cf0ed
AS		 620
621 switch (fp->code) {
622 /* All arithmetic insns and skb loads map as-is. */
623 case BPF_ALU | BPF_ADD | BPF_X:
624 case BPF_ALU | BPF_ADD | BPF_K:
625 case BPF_ALU | BPF_SUB | BPF_X:
626 case BPF_ALU | BPF_SUB | BPF_K:
627 case BPF_ALU | BPF_AND | BPF_X:
628 case BPF_ALU | BPF_AND | BPF_K:
629 case BPF_ALU | BPF_OR | BPF_X:
630 case BPF_ALU | BPF_OR | BPF_K:
631 case BPF_ALU | BPF_LSH | BPF_X:
632 case BPF_ALU | BPF_LSH | BPF_K:
633 case BPF_ALU | BPF_RSH | BPF_X:
634 case BPF_ALU | BPF_RSH | BPF_K:
635 case BPF_ALU | BPF_XOR | BPF_X:
636 case BPF_ALU | BPF_XOR | BPF_K:
637 case BPF_ALU | BPF_MUL | BPF_X:
638 case BPF_ALU | BPF_MUL | BPF_K:
639 case BPF_ALU | BPF_DIV | BPF_X:
640 case BPF_ALU | BPF_DIV | BPF_K:
641 case BPF_ALU | BPF_MOD | BPF_X:
642 case BPF_ALU | BPF_MOD | BPF_K:
643 case BPF_ALU | BPF_NEG:
644 case BPF_LD | BPF_ABS | BPF_W:
645 case BPF_LD | BPF_ABS | BPF_H:
646 case BPF_LD | BPF_ABS | BPF_B:
647 case BPF_LD | BPF_IND | BPF_W:
648 case BPF_LD | BPF_IND | BPF_H:
649 case BPF_LD | BPF_IND | BPF_B:
650 /* Check for overloaded BPF extension and
651 * directly convert it if found, otherwise
652 * just move on with mapping.
653 */
654 if (BPF_CLASS(fp->code) == BPF_LD &&
655 BPF_MODE(fp->code) == BPF_ABS &&
656 convert_bpf_extensions(fp, &insn))
657 break;
e0cea7ce
DB		 658 if (BPF_CLASS(fp->code) == BPF_LD &&
659 convert_bpf_ld_abs(fp, &insn)) {
660 *seen_ld_abs = true;
661 break;
662 }
bd4cf0ed 663
68fda450 664 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
f6b1b3bf 665 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
68fda450 666 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
f6b1b3bf
DB		 667 /* Error with exception code on div/mod by 0.
668 * For cBPF programs, this was always return 0.
669 */
670 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
671 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
672 *insn++ = BPF_EXIT_INSN();
673 }
68fda450 674
f8f6d679 675 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
bd4cf0ed
AS		 676 break;
677
f8f6d679
DB		 678 /* Jump transformation cannot use BPF block macros
679 * everywhere as offset calculation and target updates
680 * require a bit more work than the rest, i.e. jump
681 * opcodes map as-is, but offsets need adjustment.
682 */
683
684#define BPF_EMIT_JMP \
bd4cf0ed 685 do { \
050fad7c
DB		 686 const s32 off_min = S16_MIN, off_max = S16_MAX; \
687 s32 off; \
688 \
bd4cf0ed
AS		 689 if (target >= len || target < 0) \
690 goto err; \
050fad7c 691 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
bd4cf0ed 692 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
050fad7c
DB		 693 off -= insn - tmp_insns; \
694 /* Reject anything not fitting into insn->off. */ \
695 if (off < off_min || off > off_max) \
696 goto err; \
697 insn->off = off; \
bd4cf0ed
AS		 698 } while (0)
699
f8f6d679
DB		 700 case BPF_JMP | BPF_JA:
701 target = i + fp->k + 1;
702 insn->code = fp->code;
703 BPF_EMIT_JMP;
bd4cf0ed
AS		 704 break;
705
706 case BPF_JMP | BPF_JEQ | BPF_K:
707 case BPF_JMP | BPF_JEQ | BPF_X:
708 case BPF_JMP | BPF_JSET | BPF_K:
709 case BPF_JMP | BPF_JSET | BPF_X:
710 case BPF_JMP | BPF_JGT | BPF_K:
711 case BPF_JMP | BPF_JGT | BPF_X:
712 case BPF_JMP | BPF_JGE | BPF_K:
713 case BPF_JMP | BPF_JGE | BPF_X:
714 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
715 /* BPF immediates are signed, zero extend
716 * immediate into tmp register and use it
717 * in compare insn.
718 */
f8f6d679 719 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
bd4cf0ed 720
e430f34e
AS		 721 insn->dst_reg = BPF_REG_A;
722 insn->src_reg = BPF_REG_TMP;
bd4cf0ed
AS		 723 bpf_src = BPF_X;
724 } else {
e430f34e 725 insn->dst_reg = BPF_REG_A;
bd4cf0ed
AS		 726 insn->imm = fp->k;
727 bpf_src = BPF_SRC(fp->code);
19539ce7 728 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
1da177e4 729 }
bd4cf0ed
AS		 730
731 /* Common case where 'jump_false' is next insn. */
732 if (fp->jf == 0) {
733 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
734 target = i + fp->jt + 1;
f8f6d679 735 BPF_EMIT_JMP;
bd4cf0ed 736 break;
1da177e4 737 }
bd4cf0ed 738
92b31a9a
DB		 739 /* Convert some jumps when 'jump_true' is next insn. */
740 if (fp->jt == 0) {
741 switch (BPF_OP(fp->code)) {
742 case BPF_JEQ:
743 insn->code = BPF_JMP | BPF_JNE | bpf_src;
744 break;
745 case BPF_JGT:
746 insn->code = BPF_JMP | BPF_JLE | bpf_src;
747 break;
748 case BPF_JGE:
749 insn->code = BPF_JMP | BPF_JLT | bpf_src;
750 break;
751 default:
752 goto jmp_rest;
753 }
754
bd4cf0ed 755 target = i + fp->jf + 1;
f8f6d679 756 BPF_EMIT_JMP;
bd4cf0ed 757 break;
0b05b2a4 758 }
92b31a9a 759jmp_rest:
bd4cf0ed
AS		 760 /* Other jumps are mapped into two insns: Jxx and JA. */
761 target = i + fp->jt + 1;
762 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
f8f6d679 763 BPF_EMIT_JMP;
bd4cf0ed
AS		 764 insn++;
765
766 insn->code = BPF_JMP | BPF_JA;
767 target = i + fp->jf + 1;
f8f6d679 768 BPF_EMIT_JMP;
bd4cf0ed
AS		 769 break;
770
771 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
e0cea7ce
DB		 772 case BPF_LDX | BPF_MSH | BPF_B: {
773 struct sock_filter tmp = {
774 .code = BPF_LD | BPF_ABS | BPF_B,
775 .k = fp->k,
776 };
777
778 *seen_ld_abs = true;
779
780 /* X = A */
781 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
1268e253 782 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
e0cea7ce
DB		 783 convert_bpf_ld_abs(&tmp, &insn);
784 insn++;
9739eef1 785 /* A &= 0xf */
f8f6d679 786 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
9739eef1 787 /* A <<= 2 */
f8f6d679 788 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
e0cea7ce
DB		 789 /* tmp = X */
790 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
9739eef1 791 /* X = A */
f8f6d679 792 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
9739eef1 793 /* A = tmp */
f8f6d679 794 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
bd4cf0ed 795 break;
e0cea7ce 796 }
6205b9cf
DB		 797 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
798 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
799 */
bd4cf0ed
AS		 800 case BPF_RET | BPF_A:
801 case BPF_RET | BPF_K:
6205b9cf
DB		 802 if (BPF_RVAL(fp->code) == BPF_K)
803 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
804 0, fp->k);
9739eef1 805 *insn = BPF_EXIT_INSN();
bd4cf0ed
AS		 806 break;
807
808 /* Store to stack. */
809 case BPF_ST:
810 case BPF_STX:
50bbfed9 811 stack_off = fp->k * 4 + 4;
f8f6d679
DB		 812 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
813 BPF_ST ? BPF_REG_A : BPF_REG_X,
50bbfed9
AS		 814 -stack_off);
815 /* check_load_and_stores() verifies that classic BPF can
816 * load from stack only after write, so tracking
817 * stack_depth for ST|STX insns is enough
818 */
819 if (new_prog && new_prog->aux->stack_depth < stack_off)
820 new_prog->aux->stack_depth = stack_off;
bd4cf0ed
AS		 821 break;
822
823 /* Load from stack. */
824 case BPF_LD | BPF_MEM:
825 case BPF_LDX | BPF_MEM:
50bbfed9 826 stack_off = fp->k * 4 + 4;
f8f6d679
DB		 827 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
828 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
50bbfed9 829 -stack_off);
bd4cf0ed
AS		 830 break;
831
832 /* A = K or X = K */
833 case BPF_LD | BPF_IMM:
834 case BPF_LDX | BPF_IMM:
f8f6d679
DB		 835 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
836 BPF_REG_A : BPF_REG_X, fp->k);
bd4cf0ed
AS		 837 break;
838
839 /* X = A */
840 case BPF_MISC | BPF_TAX:
f8f6d679 841 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
bd4cf0ed
AS		 842 break;
843
844 /* A = X */
845 case BPF_MISC | BPF_TXA:
f8f6d679 846 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
bd4cf0ed
AS		 847 break;
848
849 /* A = skb->len or X = skb->len */
850 case BPF_LD | BPF_W | BPF_LEN:
851 case BPF_LDX | BPF_W | BPF_LEN:
f8f6d679
DB		 852 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
853 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
854 offsetof(struct sk_buff, len));
bd4cf0ed
AS		 855 break;
856
f8f6d679 857 /* Access seccomp_data fields. */
bd4cf0ed 858 case BPF_LDX | BPF_ABS | BPF_W:
9739eef1
AS		 859 /* A = *(u32 *) (ctx + K) */
860 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
bd4cf0ed
AS		 861 break;
862
ca9f1fd2 863 /* Unknown instruction. */
1da177e4 864 default:
bd4cf0ed 865 goto err;
1da177e4 866 }
bd4cf0ed
AS		 867
868 insn++;
869 if (new_prog)
870 memcpy(new_insn, tmp_insns,
871 sizeof(*insn) * (insn - tmp_insns));
bd4cf0ed 872 new_insn += insn - tmp_insns;
1da177e4
LT		 873 }
874
bd4cf0ed
AS		 875 if (!new_prog) {
876 /* Only calculating new length. */
50bbfed9 877 *new_len = new_insn - first_insn;
e0cea7ce
DB		 878 if (*seen_ld_abs)
879 *new_len += 4; /* Prologue bits. */
bd4cf0ed
AS		 880 return 0;
881 }
882
883 pass++;
50bbfed9
AS		 884 if (new_flen != new_insn - first_insn) {
885 new_flen = new_insn - first_insn;
bd4cf0ed
AS		 886 if (pass > 2)
887 goto err;
bd4cf0ed
AS		 888 goto do_pass;
889 }
890
891 kfree(addrs);
892 BUG_ON(*new_len != new_flen);
1da177e4 893 return 0;
bd4cf0ed
AS		 894err:
895 kfree(addrs);
896 return -EINVAL;
1da177e4
LT		 897}
898
bd4cf0ed 899/* Security:
bd4cf0ed 900 *
2d5311e4 901 * As we dont want to clear mem[] array for each packet going through
8ea6e345 902 * __bpf_prog_run(), we check that filter loaded by user never try to read
2d5311e4 903 * a cell if not previously written, and we check all branches to be sure
25985edc 904 * a malicious user doesn't try to abuse us.
2d5311e4 905 */
ec31a05c 906static int check_load_and_stores(const struct sock_filter *filter, int flen)
2d5311e4 907{
34805931 908 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
2d5311e4
ED		 909 int pc, ret = 0;
910
911 BUILD_BUG_ON(BPF_MEMWORDS > 16);
34805931 912
99e72a0f 913 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
2d5311e4
ED		 914 if (!masks)
915 return -ENOMEM;
34805931 916
2d5311e4
ED		 917 memset(masks, 0xff, flen * sizeof(*masks));
918
919 for (pc = 0; pc < flen; pc++) {
920 memvalid &= masks[pc];
921
922 switch (filter[pc].code) {
34805931
DB		 923 case BPF_ST:
924 case BPF_STX:
2d5311e4
ED		 925 memvalid |= (1 << filter[pc].k);
926 break;
34805931
DB		 927 case BPF_LD | BPF_MEM:
928 case BPF_LDX | BPF_MEM:
2d5311e4
ED		 929 if (!(memvalid & (1 << filter[pc].k))) {
930 ret = -EINVAL;
931 goto error;
932 }
933 break;
34805931
DB		 934 case BPF_JMP | BPF_JA:
935 /* A jump must set masks on target */
2d5311e4
ED		 936 masks[pc + 1 + filter[pc].k] &= memvalid;
937 memvalid = ~0;
938 break;
34805931
DB		 939 case BPF_JMP | BPF_JEQ | BPF_K:
940 case BPF_JMP | BPF_JEQ | BPF_X:
941 case BPF_JMP | BPF_JGE | BPF_K:
942 case BPF_JMP | BPF_JGE | BPF_X:
943 case BPF_JMP | BPF_JGT | BPF_K:
944 case BPF_JMP | BPF_JGT | BPF_X:
945 case BPF_JMP | BPF_JSET | BPF_K:
946 case BPF_JMP | BPF_JSET | BPF_X:
947 /* A jump must set masks on targets */
2d5311e4
ED		 948 masks[pc + 1 + filter[pc].jt] &= memvalid;
949 masks[pc + 1 + filter[pc].jf] &= memvalid;
950 memvalid = ~0;
951 break;
952 }
953 }
954error:
955 kfree(masks);
956 return ret;
957}
958
34805931
DB		 959static bool chk_code_allowed(u16 code_to_probe)
960{
961 static const bool codes[] = {
962 /* 32 bit ALU operations */
963 [BPF_ALU | BPF_ADD | BPF_K] = true,
964 [BPF_ALU | BPF_ADD | BPF_X] = true,
965 [BPF_ALU | BPF_SUB | BPF_K] = true,
966 [BPF_ALU | BPF_SUB | BPF_X] = true,
967 [BPF_ALU | BPF_MUL | BPF_K] = true,
968 [BPF_ALU | BPF_MUL | BPF_X] = true,
969 [BPF_ALU | BPF_DIV | BPF_K] = true,
970 [BPF_ALU | BPF_DIV | BPF_X] = true,
971 [BPF_ALU | BPF_MOD | BPF_K] = true,
972 [BPF_ALU | BPF_MOD | BPF_X] = true,
973 [BPF_ALU | BPF_AND | BPF_K] = true,
974 [BPF_ALU | BPF_AND | BPF_X] = true,
975 [BPF_ALU | BPF_OR | BPF_K] = true,
976 [BPF_ALU | BPF_OR | BPF_X] = true,
977 [BPF_ALU | BPF_XOR | BPF_K] = true,
978 [BPF_ALU | BPF_XOR | BPF_X] = true,
979 [BPF_ALU | BPF_LSH | BPF_K] = true,
980 [BPF_ALU | BPF_LSH | BPF_X] = true,
981 [BPF_ALU | BPF_RSH | BPF_K] = true,
982 [BPF_ALU | BPF_RSH | BPF_X] = true,
983 [BPF_ALU | BPF_NEG] = true,
984 /* Load instructions */
985 [BPF_LD | BPF_W | BPF_ABS] = true,
986 [BPF_LD | BPF_H | BPF_ABS] = true,
987 [BPF_LD | BPF_B | BPF_ABS] = true,
988 [BPF_LD | BPF_W | BPF_LEN] = true,
989 [BPF_LD | BPF_W | BPF_IND] = true,
990 [BPF_LD | BPF_H | BPF_IND] = true,
991 [BPF_LD | BPF_B | BPF_IND] = true,
992 [BPF_LD | BPF_IMM] = true,
993 [BPF_LD | BPF_MEM] = true,
994 [BPF_LDX | BPF_W | BPF_LEN] = true,
995 [BPF_LDX | BPF_B | BPF_MSH] = true,
996 [BPF_LDX | BPF_IMM] = true,
997 [BPF_LDX | BPF_MEM] = true,
998 /* Store instructions */
999 [BPF_ST] = true,
1000 [BPF_STX] = true,
1001 /* Misc instructions */
1002 [BPF_MISC | BPF_TAX] = true,
1003 [BPF_MISC | BPF_TXA] = true,
1004 /* Return instructions */
1005 [BPF_RET | BPF_K] = true,
1006 [BPF_RET | BPF_A] = true,
1007 /* Jump instructions */
1008 [BPF_JMP | BPF_JA] = true,
1009 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1010 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1011 [BPF_JMP | BPF_JGE | BPF_K] = true,
1012 [BPF_JMP | BPF_JGE | BPF_X] = true,
1013 [BPF_JMP | BPF_JGT | BPF_K] = true,
1014 [BPF_JMP | BPF_JGT | BPF_X] = true,
1015 [BPF_JMP | BPF_JSET | BPF_K] = true,
1016 [BPF_JMP | BPF_JSET | BPF_X] = true,
1017 };
1018
1019 if (code_to_probe >= ARRAY_SIZE(codes))
1020 return false;
1021
1022 return codes[code_to_probe];
1023}
1024
f7bd9e36
DB		 1025static bool bpf_check_basics_ok(const struct sock_filter *filter,
1026 unsigned int flen)
1027{
1028 if (filter == NULL)
1029 return false;
1030 if (flen == 0 || flen > BPF_MAXINSNS)
1031 return false;
1032
1033 return true;
1034}
1035
1da177e4 1036/**
4df95ff4 1037 * bpf_check_classic - verify socket filter code
1da177e4
LT		 1038 * @filter: filter to verify
1039 * @flen: length of filter
1040 *
1041 * Check the user's filter code. If we let some ugly
1042 * filter code slip through kaboom! The filter must contain
93699863
KK		 1043 * no references or jumps that are out of range, no illegal
1044 * instructions, and must end with a RET instruction.
1da177e4 1045 *
7b11f69f
KK		 1046 * All jumps are forward as they are not signed.
1047 *
1048 * Returns 0 if the rule set is legal or -EINVAL if not.
1da177e4 1049 */
d9e12f42
NS		 1050static int bpf_check_classic(const struct sock_filter *filter,
1051 unsigned int flen)
1da177e4 1052{
aa1113d9 1053 bool anc_found;
34805931 1054 int pc;
1da177e4 1055
34805931 1056 /* Check the filter code now */
1da177e4 1057 for (pc = 0; pc < flen; pc++) {
ec31a05c 1058 const struct sock_filter *ftest = &filter[pc];
93699863 1059
34805931
DB		 1060 /* May we actually operate on this code? */
1061 if (!chk_code_allowed(ftest->code))
cba328fc 1062 return -EINVAL;
34805931 1063
93699863 1064 /* Some instructions need special checks */
34805931
DB		 1065 switch (ftest->code) {
1066 case BPF_ALU | BPF_DIV | BPF_K:
1067 case BPF_ALU | BPF_MOD | BPF_K:
1068 /* Check for division by zero */
b6069a95
ED		 1069 if (ftest->k == 0)
1070 return -EINVAL;
1071 break;
229394e8
RV		 1072 case BPF_ALU | BPF_LSH | BPF_K:
1073 case BPF_ALU | BPF_RSH | BPF_K:
1074 if (ftest->k >= 32)
1075 return -EINVAL;
1076 break;
34805931
DB		 1077 case BPF_LD | BPF_MEM:
1078 case BPF_LDX | BPF_MEM:
1079 case BPF_ST:
1080 case BPF_STX:
1081 /* Check for invalid memory addresses */
93699863
KK		 1082 if (ftest->k >= BPF_MEMWORDS)
1083 return -EINVAL;
1084 break;
34805931
DB		 1085 case BPF_JMP | BPF_JA:
1086 /* Note, the large ftest->k might cause loops.
93699863
KK		 1087 * Compare this with conditional jumps below,
1088 * where offsets are limited. --ANK (981016)
1089 */
34805931 1090 if (ftest->k >= (unsigned int)(flen - pc - 1))
93699863 1091 return -EINVAL;
01f2f3f6 1092 break;
34805931
DB		 1093 case BPF_JMP | BPF_JEQ | BPF_K:
1094 case BPF_JMP | BPF_JEQ | BPF_X:
1095 case BPF_JMP | BPF_JGE | BPF_K:
1096 case BPF_JMP | BPF_JGE | BPF_X:
1097 case BPF_JMP | BPF_JGT | BPF_K:
1098 case BPF_JMP | BPF_JGT | BPF_X:
1099 case BPF_JMP | BPF_JSET | BPF_K:
1100 case BPF_JMP | BPF_JSET | BPF_X:
1101 /* Both conditionals must be safe */
e35bedf3 1102 if (pc + ftest->jt + 1 >= flen ||
93699863
KK		 1103 pc + ftest->jf + 1 >= flen)
1104 return -EINVAL;
cba328fc 1105 break;
34805931
DB		 1106 case BPF_LD | BPF_W | BPF_ABS:
1107 case BPF_LD | BPF_H | BPF_ABS:
1108 case BPF_LD | BPF_B | BPF_ABS:
aa1113d9 1109 anc_found = false;
34805931
DB		 1110 if (bpf_anc_helper(ftest) & BPF_ANC)
1111 anc_found = true;
1112 /* Ancillary operation unknown or unsupported */
aa1113d9
DB		 1113 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1114 return -EINVAL;
01f2f3f6
HPP		 1115 }
1116 }
93699863 1117
34805931 1118 /* Last instruction must be a RET code */
01f2f3f6 1119 switch (filter[flen - 1].code) {
34805931
DB		 1120 case BPF_RET | BPF_K:
1121 case BPF_RET | BPF_A:
2d5311e4 1122 return check_load_and_stores(filter, flen);
cba328fc 1123 }
34805931 1124
cba328fc 1125 return -EINVAL;
1da177e4
LT		 1126}
1127
7ae457c1
AS		 1128static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1129 const struct sock_fprog *fprog)
a3ea269b 1130{
009937e7 1131 unsigned int fsize = bpf_classic_proglen(fprog);
a3ea269b
DB		 1132 struct sock_fprog_kern *fkprog;
1133
1134 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1135 if (!fp->orig_prog)
1136 return -ENOMEM;
1137
1138 fkprog = fp->orig_prog;
1139 fkprog->len = fprog->len;
658da937
DB		 1140
1141 fkprog->filter = kmemdup(fp->insns, fsize,
1142 GFP_KERNEL | __GFP_NOWARN);
a3ea269b
DB		 1143 if (!fkprog->filter) {
1144 kfree(fp->orig_prog);
1145 return -ENOMEM;
1146 }
1147
1148 return 0;
1149}
1150
7ae457c1 1151static void bpf_release_orig_filter(struct bpf_prog *fp)
a3ea269b
DB		 1152{
1153 struct sock_fprog_kern *fprog = fp->orig_prog;
1154
1155 if (fprog) {
1156 kfree(fprog->filter);
1157 kfree(fprog);
1158 }
1159}
1160
7ae457c1
AS		 1161static void __bpf_prog_release(struct bpf_prog *prog)
1162{
24701ece 1163 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
89aa0758
AS		 1164 bpf_prog_put(prog);
1165 } else {
1166 bpf_release_orig_filter(prog);
1167 bpf_prog_free(prog);
1168 }
7ae457c1
AS		 1169}
1170
34c5bd66
PN		 1171static void __sk_filter_release(struct sk_filter *fp)
1172{
7ae457c1
AS		 1173 __bpf_prog_release(fp->prog);
1174 kfree(fp);
34c5bd66
PN		 1175}
1176
47e958ea 1177/**
46bcf14f 1178 * sk_filter_release_rcu - Release a socket filter by rcu_head
47e958ea
PE		 1179 * @rcu: rcu_head that contains the sk_filter to free
1180 */
fbc907f0 1181static void sk_filter_release_rcu(struct rcu_head *rcu)
47e958ea
PE		 1182{
1183 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1184
34c5bd66 1185 __sk_filter_release(fp);
47e958ea 1186}
fbc907f0
DB		 1187
1188/**
1189 * sk_filter_release - release a socket filter
1190 * @fp: filter to remove
1191 *
1192 * Remove a filter from a socket and release its resources.
1193 */
1194static void sk_filter_release(struct sk_filter *fp)
1195{
4c355cdf 1196 if (refcount_dec_and_test(&fp->refcnt))
fbc907f0
DB		 1197 call_rcu(&fp->rcu, sk_filter_release_rcu);
1198}
1199
1200void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1201{
7ae457c1 1202 u32 filter_size = bpf_prog_size(fp->prog->len);
fbc907f0 1203
278571ba
AS		 1204 atomic_sub(filter_size, &sk->sk_omem_alloc);
1205 sk_filter_release(fp);
fbc907f0 1206}
47e958ea 1207
278571ba
AS		 1208/* try to charge the socket memory if there is space available
1209 * return true on success
1210 */
4c355cdf 1211static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
bd4cf0ed 1212{
7ae457c1 1213 u32 filter_size = bpf_prog_size(fp->prog->len);
278571ba
AS		 1214
1215 /* same check as in sock_kmalloc() */
1216 if (filter_size <= sysctl_optmem_max &&
1217 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
278571ba
AS		 1218 atomic_add(filter_size, &sk->sk_omem_alloc);
1219 return true;
bd4cf0ed 1220 }
278571ba 1221 return false;
bd4cf0ed
AS		 1222}
1223
4c355cdf
RE		 1224bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1225{
eefca20e
ED		 1226 if (!refcount_inc_not_zero(&fp->refcnt))
1227 return false;
1228
1229 if (!__sk_filter_charge(sk, fp)) {
1230 sk_filter_release(fp);
1231 return false;
1232 }
1233 return true;
4c355cdf
RE		 1234}
1235
7ae457c1 1236static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
bd4cf0ed
AS		 1237{
1238 struct sock_filter *old_prog;
7ae457c1 1239 struct bpf_prog *old_fp;
34805931 1240 int err, new_len, old_len = fp->len;
e0cea7ce 1241 bool seen_ld_abs = false;
bd4cf0ed
AS		 1242
1243 /* We are free to overwrite insns et al right here as it
1244 * won't be used at this point in time anymore internally
1245 * after the migration to the internal BPF instruction
1246 * representation.
1247 */
1248 BUILD_BUG_ON(sizeof(struct sock_filter) !=
2695fb55 1249 sizeof(struct bpf_insn));
bd4cf0ed 1250
bd4cf0ed
AS		 1251 /* Conversion cannot happen on overlapping memory areas,
1252 * so we need to keep the user BPF around until the 2nd
1253 * pass. At this time, the user BPF is stored in fp->insns.
1254 */
1255 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
658da937 1256 GFP_KERNEL | __GFP_NOWARN);
bd4cf0ed
AS		 1257 if (!old_prog) {
1258 err = -ENOMEM;
1259 goto out_err;
1260 }
1261
1262 /* 1st pass: calculate the new program length. */
e0cea7ce
DB		 1263 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1264 &seen_ld_abs);
bd4cf0ed
AS		 1265 if (err)
1266 goto out_err_free;
1267
1268 /* Expand fp for appending the new filter representation. */
1269 old_fp = fp;
60a3b225 1270 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
bd4cf0ed
AS		 1271 if (!fp) {
1272 /* The old_fp is still around in case we couldn't
1273 * allocate new memory, so uncharge on that one.
1274 */
1275 fp = old_fp;
1276 err = -ENOMEM;
1277 goto out_err_free;
1278 }
1279
bd4cf0ed
AS		 1280 fp->len = new_len;
1281
2695fb55 1282 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
e0cea7ce
DB		 1283 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1284 &seen_ld_abs);
bd4cf0ed 1285 if (err)
8fb575ca 1286 /* 2nd bpf_convert_filter() can fail only if it fails
bd4cf0ed
AS		 1287 * to allocate memory, remapping must succeed. Note,
1288 * that at this time old_fp has already been released
278571ba 1289 * by krealloc().
bd4cf0ed
AS		 1290 */
1291 goto out_err_free;
1292
d1c55ab5 1293 fp = bpf_prog_select_runtime(fp, &err);
290af866
AS		 1294 if (err)
1295 goto out_err_free;
5fe821a9 1296
bd4cf0ed
AS		 1297 kfree(old_prog);
1298 return fp;
1299
1300out_err_free:
1301 kfree(old_prog);
1302out_err:
7ae457c1 1303 __bpf_prog_release(fp);
bd4cf0ed
AS		 1304 return ERR_PTR(err);
1305}
1306
ac67eb2c
DB		 1307static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1308 bpf_aux_classic_check_t trans)
302d6637
JP		 1309{
1310 int err;
1311
bd4cf0ed 1312 fp->bpf_func = NULL;
a91263d5 1313 fp->jited = 0;
302d6637 1314
4df95ff4 1315 err = bpf_check_classic(fp->insns, fp->len);
418c96ac 1316 if (err) {
7ae457c1 1317 __bpf_prog_release(fp);
bd4cf0ed 1318 return ERR_PTR(err);
418c96ac 1319 }
302d6637 1320
4ae92bc7
NS		 1321 /* There might be additional checks and transformations
1322 * needed on classic filters, f.e. in case of seccomp.
1323 */
1324 if (trans) {
1325 err = trans(fp->insns, fp->len);
1326 if (err) {
1327 __bpf_prog_release(fp);
1328 return ERR_PTR(err);
1329 }
1330 }
1331
bd4cf0ed
AS		 1332 /* Probe if we can JIT compile the filter and if so, do
1333 * the compilation of the filter.
1334 */
302d6637 1335 bpf_jit_compile(fp);
bd4cf0ed
AS		 1336
1337 /* JIT compiler couldn't process this filter, so do the
1338 * internal BPF translation for the optimized interpreter.
1339 */
5fe821a9 1340 if (!fp->jited)
7ae457c1 1341 fp = bpf_migrate_filter(fp);
bd4cf0ed
AS		 1342
1343 return fp;
302d6637
JP		 1344}
1345
1346/**
7ae457c1 1347 * bpf_prog_create - create an unattached filter
c6c4b97c 1348 * @pfp: the unattached filter that is created
677a9fd3 1349 * @fprog: the filter program
302d6637 1350 *
c6c4b97c 1351 * Create a filter independent of any socket. We first run some
302d6637
JP		 1352 * sanity checks on it to make sure it does not explode on us later.
1353 * If an error occurs or there is insufficient memory for the filter
1354 * a negative errno code is returned. On success the return is zero.
1355 */
7ae457c1 1356int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
302d6637 1357{
009937e7 1358 unsigned int fsize = bpf_classic_proglen(fprog);
7ae457c1 1359 struct bpf_prog *fp;
302d6637
JP		 1360
1361 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1362 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
302d6637
JP		 1363 return -EINVAL;
1364
60a3b225 1365 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
302d6637
JP		 1366 if (!fp)
1367 return -ENOMEM;
a3ea269b 1368
302d6637
JP		 1369 memcpy(fp->insns, fprog->filter, fsize);
1370
302d6637 1371 fp->len = fprog->len;
a3ea269b
DB		 1372 /* Since unattached filters are not copied back to user
1373 * space through sk_get_filter(), we do not need to hold
1374 * a copy here, and can spare us the work.
1375 */
1376 fp->orig_prog = NULL;
302d6637 1377
7ae457c1 1378 /* bpf_prepare_filter() already takes care of freeing
bd4cf0ed
AS		 1379 * memory in case something goes wrong.
1380 */
4ae92bc7 1381 fp = bpf_prepare_filter(fp, NULL);
bd4cf0ed
AS		 1382 if (IS_ERR(fp))
1383 return PTR_ERR(fp);
302d6637
JP		 1384
1385 *pfp = fp;
1386 return 0;
302d6637 1387}
7ae457c1 1388EXPORT_SYMBOL_GPL(bpf_prog_create);
302d6637 1389
ac67eb2c
DB		 1390/**
1391 * bpf_prog_create_from_user - create an unattached filter from user buffer
1392 * @pfp: the unattached filter that is created
1393 * @fprog: the filter program
1394 * @trans: post-classic verifier transformation handler
bab18991 1395 * @save_orig: save classic BPF program
ac67eb2c
DB		 1396 *
1397 * This function effectively does the same as bpf_prog_create(), only
1398 * that it builds up its insns buffer from user space provided buffer.
1399 * It also allows for passing a bpf_aux_classic_check_t handler.
1400 */
1401int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
bab18991 1402 bpf_aux_classic_check_t trans, bool save_orig)
ac67eb2c
DB		 1403{
1404 unsigned int fsize = bpf_classic_proglen(fprog);
1405 struct bpf_prog *fp;
bab18991 1406 int err;
ac67eb2c
DB		 1407
1408 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1409 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
ac67eb2c
DB		 1410 return -EINVAL;
1411
1412 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1413 if (!fp)
1414 return -ENOMEM;
1415
1416 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1417 __bpf_prog_free(fp);
1418 return -EFAULT;
1419 }
1420
1421 fp->len = fprog->len;
ac67eb2c
DB		 1422 fp->orig_prog = NULL;
1423
bab18991
DB		 1424 if (save_orig) {
1425 err = bpf_prog_store_orig_filter(fp, fprog);
1426 if (err) {
1427 __bpf_prog_free(fp);
1428 return -ENOMEM;
1429 }
1430 }
1431
ac67eb2c
DB		 1432 /* bpf_prepare_filter() already takes care of freeing
1433 * memory in case something goes wrong.
1434 */
1435 fp = bpf_prepare_filter(fp, trans);
1436 if (IS_ERR(fp))
1437 return PTR_ERR(fp);
1438
1439 *pfp = fp;
1440 return 0;
1441}
2ea273d7 1442EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
ac67eb2c 1443
7ae457c1 1444void bpf_prog_destroy(struct bpf_prog *fp)
302d6637 1445{
7ae457c1 1446 __bpf_prog_release(fp);
302d6637 1447}
7ae457c1 1448EXPORT_SYMBOL_GPL(bpf_prog_destroy);
302d6637 1449
8ced425e 1450static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
49b31e57
DB		 1451{
1452 struct sk_filter *fp, *old_fp;
1453
1454 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1455 if (!fp)
1456 return -ENOMEM;
1457
1458 fp->prog = prog;
49b31e57 1459
4c355cdf 1460 if (!__sk_filter_charge(sk, fp)) {
49b31e57
DB		 1461 kfree(fp);
1462 return -ENOMEM;
1463 }
4c355cdf 1464 refcount_set(&fp->refcnt, 1);
49b31e57 1465
8ced425e
HFS		 1466 old_fp = rcu_dereference_protected(sk->sk_filter,
1467 lockdep_sock_is_held(sk));
49b31e57 1468 rcu_assign_pointer(sk->sk_filter, fp);
8ced425e 1469
49b31e57
DB		 1470 if (old_fp)
1471 sk_filter_uncharge(sk, old_fp);
1472
1473 return 0;
1474}
1475
538950a1
CG		 1476static
1477struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1da177e4 1478{
009937e7 1479 unsigned int fsize = bpf_classic_proglen(fprog);
7ae457c1 1480 struct bpf_prog *prog;
1da177e4
LT		 1481 int err;
1482
d59577b6 1483 if (sock_flag(sk, SOCK_FILTER_LOCKED))
538950a1 1484 return ERR_PTR(-EPERM);
d59577b6 1485
1da177e4 1486 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1487 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
538950a1 1488 return ERR_PTR(-EINVAL);
1da177e4 1489
f7bd9e36 1490 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
7ae457c1 1491 if (!prog)
538950a1 1492 return ERR_PTR(-ENOMEM);
a3ea269b 1493
7ae457c1 1494 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
c0d1379a 1495 __bpf_prog_free(prog);
538950a1 1496 return ERR_PTR(-EFAULT);
1da177e4
LT		 1497 }
1498
7ae457c1 1499 prog->len = fprog->len;
1da177e4 1500
7ae457c1 1501 err = bpf_prog_store_orig_filter(prog, fprog);
a3ea269b 1502 if (err) {
c0d1379a 1503 __bpf_prog_free(prog);
538950a1 1504 return ERR_PTR(-ENOMEM);
a3ea269b
DB		 1505 }
1506
7ae457c1 1507 /* bpf_prepare_filter() already takes care of freeing
bd4cf0ed
AS		 1508 * memory in case something goes wrong.
1509 */
538950a1
CG		 1510 return bpf_prepare_filter(prog, NULL);
1511}
1512
1513/**
1514 * sk_attach_filter - attach a socket filter
1515 * @fprog: the filter program
1516 * @sk: the socket to use
1517 *
1518 * Attach the user's filter code. We first run some sanity checks on
1519 * it to make sure it does not explode on us later. If an error
1520 * occurs or there is insufficient memory for the filter a negative
1521 * errno code is returned. On success the return is zero.
1522 */
8ced425e 1523int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
538950a1
CG		 1524{
1525 struct bpf_prog *prog = __get_filter(fprog, sk);
1526 int err;
1527
7ae457c1
AS		 1528 if (IS_ERR(prog))
1529 return PTR_ERR(prog);
1530
8ced425e 1531 err = __sk_attach_prog(prog, sk);
49b31e57 1532 if (err < 0) {
7ae457c1 1533 __bpf_prog_release(prog);
49b31e57 1534 return err;
278571ba
AS		 1535 }
1536
d3904b73 1537 return 0;
1da177e4 1538}
8ced425e 1539EXPORT_SYMBOL_GPL(sk_attach_filter);
1da177e4 1540
538950a1 1541int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
89aa0758 1542{
538950a1 1543 struct bpf_prog *prog = __get_filter(fprog, sk);
49b31e57 1544 int err;
89aa0758 1545
538950a1
CG		 1546 if (IS_ERR(prog))
1547 return PTR_ERR(prog);
1548
8217ca65
MKL		 1549 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1550 err = -ENOMEM;
1551 else
1552 err = reuseport_attach_prog(sk, prog);
1553
1554 if (err)
538950a1 1555 __bpf_prog_release(prog);
538950a1 1556
8217ca65 1557 return err;
538950a1
CG		 1558}
1559
1560static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1561{
89aa0758 1562 if (sock_flag(sk, SOCK_FILTER_LOCKED))
538950a1 1563 return ERR_PTR(-EPERM);
89aa0758 1564
113214be 1565 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
538950a1
CG		 1566}
1567
1568int sk_attach_bpf(u32 ufd, struct sock *sk)
1569{
1570 struct bpf_prog *prog = __get_bpf(ufd, sk);
1571 int err;
1572
1573 if (IS_ERR(prog))
1574 return PTR_ERR(prog);
1575
8ced425e 1576 err = __sk_attach_prog(prog, sk);
49b31e57 1577 if (err < 0) {
89aa0758 1578 bpf_prog_put(prog);
49b31e57 1579 return err;
89aa0758
AS		 1580 }
1581
89aa0758
AS		 1582 return 0;
1583}
1584
538950a1
CG		 1585int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1586{
8217ca65 1587 struct bpf_prog *prog;
538950a1
CG		 1588 int err;
1589
8217ca65
MKL		 1590 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1591 return -EPERM;
1592
1593 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
45586c70 1594 if (PTR_ERR(prog) == -EINVAL)
8217ca65 1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
538950a1
CG		 1596 if (IS_ERR(prog))
1597 return PTR_ERR(prog);
1598
8217ca65
MKL		 1599 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1600 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1601 * bpf prog (e.g. sockmap). It depends on the
1602 * limitation imposed by bpf_prog_load().
1603 * Hence, sysctl_optmem_max is not checked.
1604 */
1605 if ((sk->sk_type != SOCK_STREAM &&
1606 sk->sk_type != SOCK_DGRAM) ||
1607 (sk->sk_protocol != IPPROTO_UDP &&
1608 sk->sk_protocol != IPPROTO_TCP) ||
1609 (sk->sk_family != AF_INET &&
1610 sk->sk_family != AF_INET6)) {
1611 err = -ENOTSUPP;
1612 goto err_prog_put;
1613 }
1614 } else {
1615 /* BPF_PROG_TYPE_SOCKET_FILTER */
1616 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1617 err = -ENOMEM;
1618 goto err_prog_put;
1619 }
538950a1
CG		 1620 }
1621
8217ca65
MKL		 1622 err = reuseport_attach_prog(sk, prog);
1623err_prog_put:
1624 if (err)
1625 bpf_prog_put(prog);
1626
1627 return err;
1628}
1629
1630void sk_reuseport_prog_free(struct bpf_prog *prog)
1631{
1632 if (!prog)
1633 return;
1634
1635 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1636 bpf_prog_put(prog);
1637 else
1638 bpf_prog_destroy(prog);
538950a1
CG		 1639}
1640
21cafc1d
DB		 1641struct bpf_scratchpad {
1642 union {
1643 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1644 u8 buff[MAX_BPF_STACK];
1645 };
1646};
1647
1648static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
91bc4822 1649
5293efe6
DB		 1650static inline int __bpf_try_make_writable(struct sk_buff *skb,
1651 unsigned int write_len)
1652{
1653 return skb_ensure_writable(skb, write_len);
1654}
1655
db58ba45
AS		 1656static inline int bpf_try_make_writable(struct sk_buff *skb,
1657 unsigned int write_len)
1658{
5293efe6 1659 int err = __bpf_try_make_writable(skb, write_len);
db58ba45 1660
6aaae2b6 1661 bpf_compute_data_pointers(skb);
db58ba45
AS		 1662 return err;
1663}
1664
36bbef52
DB		 1665static int bpf_try_make_head_writable(struct sk_buff *skb)
1666{
1667 return bpf_try_make_writable(skb, skb_headlen(skb));
1668}
1669
a2bfe6bf
DB		 1670static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1671{
1672 if (skb_at_tc_ingress(skb))
1673 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1674}
1675
8065694e
DB		 1676static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1677{
1678 if (skb_at_tc_ingress(skb))
1679 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1680}
1681
f3694e00
DB		 1682BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1683 const void *, from, u32, len, u64, flags)
608cd71a 1684{
608cd71a
AS		 1685 void *ptr;
1686
8afd54c8 1687 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
781c53bc 1688 return -EINVAL;
0ed661d5 1689 if (unlikely(offset > 0xffff))
608cd71a 1690 return -EFAULT;
db58ba45 1691 if (unlikely(bpf_try_make_writable(skb, offset + len)))
608cd71a
AS		 1692 return -EFAULT;
1693
0ed661d5 1694 ptr = skb->data + offset;
781c53bc 1695 if (flags & BPF_F_RECOMPUTE_CSUM)
479ffccc 1696 __skb_postpull_rcsum(skb, ptr, len, offset);
608cd71a
AS		 1697
1698 memcpy(ptr, from, len);
1699
781c53bc 1700 if (flags & BPF_F_RECOMPUTE_CSUM)
479ffccc 1701 __skb_postpush_rcsum(skb, ptr, len, offset);
8afd54c8
DB		 1702 if (flags & BPF_F_INVALIDATE_HASH)
1703 skb_clear_hash(skb);
f8ffad69 1704
608cd71a
AS		 1705 return 0;
1706}
1707
577c50aa 1708static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
608cd71a
AS		 1709 .func = bpf_skb_store_bytes,
1710 .gpl_only = false,
1711 .ret_type = RET_INTEGER,
1712 .arg1_type = ARG_PTR_TO_CTX,
1713 .arg2_type = ARG_ANYTHING,
39f19ebb
AS		 1714 .arg3_type = ARG_PTR_TO_MEM,
1715 .arg4_type = ARG_CONST_SIZE,
91bc4822
AS		 1716 .arg5_type = ARG_ANYTHING,
1717};
1718
f3694e00
DB		 1719BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1720 void *, to, u32, len)
05c74e5e 1721{
05c74e5e
DB		 1722 void *ptr;
1723
0ed661d5 1724 if (unlikely(offset > 0xffff))
074f528e 1725 goto err_clear;
05c74e5e
DB		 1726
1727 ptr = skb_header_pointer(skb, offset, len, to);
1728 if (unlikely(!ptr))
074f528e 1729 goto err_clear;
05c74e5e
DB		 1730 if (ptr != to)
1731 memcpy(to, ptr, len);
1732
1733 return 0;
074f528e
DB		 1734err_clear:
1735 memset(to, 0, len);
1736 return -EFAULT;
05c74e5e
DB		 1737}
1738
577c50aa 1739static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
05c74e5e
DB		 1740 .func = bpf_skb_load_bytes,
1741 .gpl_only = false,
1742 .ret_type = RET_INTEGER,
1743 .arg1_type = ARG_PTR_TO_CTX,
1744 .arg2_type = ARG_ANYTHING,
39f19ebb
AS		 1745 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1746 .arg4_type = ARG_CONST_SIZE,
05c74e5e
DB		 1747};
1748
089b19a9
SF		 1749BPF_CALL_4(bpf_flow_dissector_load_bytes,
1750 const struct bpf_flow_dissector *, ctx, u32, offset,
1751 void *, to, u32, len)
1752{
1753 void *ptr;
1754
1755 if (unlikely(offset > 0xffff))
1756 goto err_clear;
1757
1758 if (unlikely(!ctx->skb))
1759 goto err_clear;
1760
1761 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1762 if (unlikely(!ptr))
1763 goto err_clear;
1764 if (ptr != to)
1765 memcpy(to, ptr, len);
1766
1767 return 0;
1768err_clear:
1769 memset(to, 0, len);
1770 return -EFAULT;
1771}
1772
1773static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1774 .func = bpf_flow_dissector_load_bytes,
1775 .gpl_only = false,
1776 .ret_type = RET_INTEGER,
1777 .arg1_type = ARG_PTR_TO_CTX,
1778 .arg2_type = ARG_ANYTHING,
1779 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1780 .arg4_type = ARG_CONST_SIZE,
1781};
1782
4e1ec56c
DB		 1783BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1784 u32, offset, void *, to, u32, len, u32, start_header)
1785{
3eee1f75 1786 u8 *end = skb_tail_pointer(skb);
0f5d82f1 1787 u8 *start, *ptr;
4e1ec56c 1788
0f5d82f1 1789 if (unlikely(offset > 0xffff))
4e1ec56c
DB		 1790 goto err_clear;
1791
1792 switch (start_header) {
1793 case BPF_HDR_START_MAC:
0f5d82f1
YZ		 1794 if (unlikely(!skb_mac_header_was_set(skb)))
1795 goto err_clear;
1796 start = skb_mac_header(skb);
4e1ec56c
DB		 1797 break;
1798 case BPF_HDR_START_NET:
0f5d82f1 1799 start = skb_network_header(skb);
4e1ec56c
DB		 1800 break;
1801 default:
1802 goto err_clear;
1803 }
1804
0f5d82f1
YZ		 1805 ptr = start + offset;
1806
1807 if (likely(ptr + len <= end)) {
4e1ec56c
DB		 1808 memcpy(to, ptr, len);
1809 return 0;
1810 }
1811
1812err_clear:
1813 memset(to, 0, len);
1814 return -EFAULT;
1815}
1816
1817static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1818 .func = bpf_skb_load_bytes_relative,
1819 .gpl_only = false,
1820 .ret_type = RET_INTEGER,
1821 .arg1_type = ARG_PTR_TO_CTX,
1822 .arg2_type = ARG_ANYTHING,
1823 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1824 .arg4_type = ARG_CONST_SIZE,
1825 .arg5_type = ARG_ANYTHING,
1826};
1827
36bbef52
DB		 1828BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1829{
1830 /* Idea is the following: should the needed direct read/write
1831 * test fail during runtime, we can pull in more data and redo
1832 * again, since implicitly, we invalidate previous checks here.
1833 *
1834 * Or, since we know how much we need to make read/writeable,
1835 * this can be done once at the program beginning for direct
1836 * access case. By this we overcome limitations of only current
1837 * headroom being accessible.
1838 */
1839 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1840}
1841
1842static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1843 .func = bpf_skb_pull_data,
1844 .gpl_only = false,
1845 .ret_type = RET_INTEGER,
1846 .arg1_type = ARG_PTR_TO_CTX,
1847 .arg2_type = ARG_ANYTHING,
1848};
1849
46f8bc92
MKL		 1850BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1851{
46f8bc92
MKL		 1852 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1853}
1854
1855static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1856 .func = bpf_sk_fullsock,
1857 .gpl_only = false,
1858 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1859 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1860};
1861
0ea488ff
JF		 1862static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1863 unsigned int write_len)
1864{
1865 int err = __bpf_try_make_writable(skb, write_len);
1866
1867 bpf_compute_data_end_sk_skb(skb);
1868 return err;
1869}
1870
1871BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872{
1873 /* Idea is the following: should the needed direct read/write
1874 * test fail during runtime, we can pull in more data and redo
1875 * again, since implicitly, we invalidate previous checks here.
1876 *
1877 * Or, since we know how much we need to make read/writeable,
1878 * this can be done once at the program beginning for direct
1879 * access case. By this we overcome limitations of only current
1880 * headroom being accessible.
1881 */
1882 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883}
1884
1885static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886 .func = sk_skb_pull_data,
1887 .gpl_only = false,
1888 .ret_type = RET_INTEGER,
1889 .arg1_type = ARG_PTR_TO_CTX,
1890 .arg2_type = ARG_ANYTHING,
1891};
1892
f3694e00
DB		 1893BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894 u64, from, u64, to, u64, flags)
91bc4822 1895{
0ed661d5 1896 __sum16 *ptr;
91bc4822 1897
781c53bc
DB		 1898 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899 return -EINVAL;
0ed661d5 1900 if (unlikely(offset > 0xffff || offset & 1))
91bc4822 1901 return -EFAULT;
0ed661d5 1902 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
91bc4822
AS		 1903 return -EFAULT;
1904
0ed661d5 1905 ptr = (__sum16 *)(skb->data + offset);
781c53bc 1906 switch (flags & BPF_F_HDR_FIELD_MASK) {
8050c0f0
DB		 1907 case 0:
1908 if (unlikely(from != 0))
1909 return -EINVAL;
1910
1911 csum_replace_by_diff(ptr, to);
1912 break;
91bc4822
AS		 1913 case 2:
1914 csum_replace2(ptr, from, to);
1915 break;
1916 case 4:
1917 csum_replace4(ptr, from, to);
1918 break;
1919 default:
1920 return -EINVAL;
1921 }
1922
91bc4822
AS		 1923 return 0;
1924}
1925
577c50aa 1926static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
91bc4822
AS		 1927 .func = bpf_l3_csum_replace,
1928 .gpl_only = false,
1929 .ret_type = RET_INTEGER,
1930 .arg1_type = ARG_PTR_TO_CTX,
1931 .arg2_type = ARG_ANYTHING,
1932 .arg3_type = ARG_ANYTHING,
1933 .arg4_type = ARG_ANYTHING,
1934 .arg5_type = ARG_ANYTHING,
1935};
1936
f3694e00
DB		 1937BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938 u64, from, u64, to, u64, flags)
91bc4822 1939{
781c53bc 1940 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
2f72959a 1941 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
d1b662ad 1942 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
0ed661d5 1943 __sum16 *ptr;
91bc4822 1944
d1b662ad
DB		 1945 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
781c53bc 1947 return -EINVAL;
0ed661d5 1948 if (unlikely(offset > 0xffff || offset & 1))
91bc4822 1949 return -EFAULT;
0ed661d5 1950 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
91bc4822
AS		 1951 return -EFAULT;
1952
0ed661d5 1953 ptr = (__sum16 *)(skb->data + offset);
d1b662ad 1954 if (is_mmzero && !do_mforce && !*ptr)
2f72959a 1955 return 0;
91bc4822 1956
781c53bc 1957 switch (flags & BPF_F_HDR_FIELD_MASK) {
7d672345
DB		 1958 case 0:
1959 if (unlikely(from != 0))
1960 return -EINVAL;
1961
1962 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963 break;
91bc4822
AS		 1964 case 2:
1965 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966 break;
1967 case 4:
1968 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1969 break;
1970 default:
1971 return -EINVAL;
1972 }
1973
2f72959a
DB		 1974 if (is_mmzero && !*ptr)
1975 *ptr = CSUM_MANGLED_0;
91bc4822
AS		 1976 return 0;
1977}
1978
577c50aa 1979static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
91bc4822
AS		 1980 .func = bpf_l4_csum_replace,
1981 .gpl_only = false,
1982 .ret_type = RET_INTEGER,
1983 .arg1_type = ARG_PTR_TO_CTX,
1984 .arg2_type = ARG_ANYTHING,
1985 .arg3_type = ARG_ANYTHING,
1986 .arg4_type = ARG_ANYTHING,
1987 .arg5_type = ARG_ANYTHING,
608cd71a
AS		 1988};
1989
f3694e00
DB		 1990BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991 __be32 *, to, u32, to_size, __wsum, seed)
7d672345 1992{
21cafc1d 1993 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
f3694e00 1994 u32 diff_size = from_size + to_size;
7d672345
DB		 1995 int i, j = 0;
1996
1997 /* This is quite flexible, some examples:
1998 *
1999 * from_size == 0, to_size > 0, seed := csum --> pushing data
2000 * from_size > 0, to_size == 0, seed := csum --> pulling data
2001 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2002 *
2003 * Even for diffing, from_size and to_size don't need to be equal.
2004 */
2005 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006 diff_size > sizeof(sp->diff)))
2007 return -EINVAL;
2008
2009 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = ~from[i];
2011 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = to[i];
2013
2014 return csum_partial(sp->diff, diff_size, seed);
2015}
2016
577c50aa 2017static const struct bpf_func_proto bpf_csum_diff_proto = {
7d672345
DB		 2018 .func = bpf_csum_diff,
2019 .gpl_only = false,
36bbef52 2020 .pkt_access = true,
7d672345 2021 .ret_type = RET_INTEGER,
db1ac496 2022 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
39f19ebb 2023 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
db1ac496 2024 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
39f19ebb 2025 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
7d672345
DB		 2026 .arg5_type = ARG_ANYTHING,
2027};
2028
36bbef52
DB		 2029BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030{
2031 /* The interface is to be used in combination with bpf_csum_diff()
2032 * for direct packet writes. csum rotation for alignment as well
2033 * as emulating csum_sub() can be done from the eBPF program.
2034 */
2035 if (skb->ip_summed == CHECKSUM_COMPLETE)
2036 return (skb->csum = csum_add(skb->csum, csum));
2037
2038 return -ENOTSUPP;
2039}
2040
2041static const struct bpf_func_proto bpf_csum_update_proto = {
2042 .func = bpf_csum_update,
2043 .gpl_only = false,
2044 .ret_type = RET_INTEGER,
2045 .arg1_type = ARG_PTR_TO_CTX,
2046 .arg2_type = ARG_ANYTHING,
2047};
2048
7cdec54f
DB		 2049BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050{
2051 /* The interface is to be used in combination with bpf_skb_adjust_room()
2052 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053 * is passed as flags, for example.
2054 */
2055 switch (level) {
2056 case BPF_CSUM_LEVEL_INC:
2057 __skb_incr_checksum_unnecessary(skb);
2058 break;
2059 case BPF_CSUM_LEVEL_DEC:
2060 __skb_decr_checksum_unnecessary(skb);
2061 break;
2062 case BPF_CSUM_LEVEL_RESET:
2063 __skb_reset_checksum_unnecessary(skb);
2064 break;
2065 case BPF_CSUM_LEVEL_QUERY:
2066 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067 skb->csum_level : -EACCES;
2068 default:
2069 return -EINVAL;
2070 }
2071
2072 return 0;
2073}
2074
2075static const struct bpf_func_proto bpf_csum_level_proto = {
2076 .func = bpf_csum_level,
2077 .gpl_only = false,
2078 .ret_type = RET_INTEGER,
2079 .arg1_type = ARG_PTR_TO_CTX,
2080 .arg2_type = ARG_ANYTHING,
2081};
2082
a70b506e
DB		 2083static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084{
a70b506e
DB		 2085 return dev_forward_skb(dev, skb);
2086}
2087
4e3264d2
MKL		 2088static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089 struct sk_buff *skb)
2090{
2091 int ret = ____dev_forward_skb(dev, skb);
2092
2093 if (likely(!ret)) {
2094 skb->dev = dev;
2095 ret = netif_rx(skb);
2096 }
2097
2098 return ret;
2099}
2100
a70b506e
DB		 2101static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2102{
2103 int ret;
2104
97cdcf37 2105 if (dev_xmit_recursion()) {
a70b506e
DB		 2106 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2107 kfree_skb(skb);
2108 return -ENETDOWN;
2109 }
2110
2111 skb->dev = dev;
5133498f 2112 skb->tstamp = 0;
a70b506e 2113
97cdcf37 2114 dev_xmit_recursion_inc();
a70b506e 2115 ret = dev_queue_xmit(skb);
97cdcf37 2116 dev_xmit_recursion_dec();
a70b506e
DB		 2117
2118 return ret;
2119}
2120
4e3264d2
MKL		 2121static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122 u32 flags)
2123{
e7c87bd6 2124 unsigned int mlen = skb_network_offset(skb);
4e3264d2 2125
e7c87bd6
WB		 2126 if (mlen) {
2127 __skb_pull(skb, mlen);
4e3264d2 2128
e7c87bd6
WB		 2129 /* At ingress, the mac header has already been pulled once.
2130 * At egress, skb_pospull_rcsum has to be done in case that
2131 * the skb is originated from ingress (i.e. a forwarded skb)
2132 * to ensure that rcsum starts at net header.
2133 */
2134 if (!skb_at_tc_ingress(skb))
2135 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2136 }
4e3264d2
MKL		 2137 skb_pop_mac_header(skb);
2138 skb_reset_mac_len(skb);
2139 return flags & BPF_F_INGRESS ?
2140 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2141}
2142
2143static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2144 u32 flags)
2145{
3a0af8fd
TG		 2146 /* Verify that a link layer header is carried */
2147 if (unlikely(skb->mac_header >= skb->network_header)) {
2148 kfree_skb(skb);
2149 return -ERANGE;
2150 }
2151
4e3264d2
MKL		 2152 bpf_push_mac_rcsum(skb);
2153 return flags & BPF_F_INGRESS ?
2154 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2155}
2156
2157static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2158 u32 flags)
2159{
c491680f 2160 if (dev_is_mac_header_xmit(dev))
4e3264d2 2161 return __bpf_redirect_common(skb, dev, flags);
c491680f
DB		 2162 else
2163 return __bpf_redirect_no_mac(skb, dev, flags);
4e3264d2
MKL		 2164}
2165
b4ab3141
DB		 2166#if IS_ENABLED(CONFIG_IPV6)
2167static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb)
2168{
2169 struct dst_entry *dst = skb_dst(skb);
2170 struct net_device *dev = dst->dev;
2171 u32 hh_len = LL_RESERVED_SPACE(dev);
2172 const struct in6_addr *nexthop;
2173 struct neighbour *neigh;
2174
2175 if (dev_xmit_recursion()) {
2176 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2177 goto out_drop;
2178 }
2179
2180 skb->dev = dev;
2181 skb->tstamp = 0;
2182
2183 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2184 struct sk_buff *skb2;
2185
2186 skb2 = skb_realloc_headroom(skb, hh_len);
2187 if (unlikely(!skb2)) {
2188 kfree_skb(skb);
2189 return -ENOMEM;
2190 }
2191 if (skb->sk)
2192 skb_set_owner_w(skb2, skb->sk);
2193 consume_skb(skb);
2194 skb = skb2;
2195 }
2196
2197 rcu_read_lock_bh();
2198 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2199 &ipv6_hdr(skb)->daddr);
2200 neigh = ip_neigh_gw6(dev, nexthop);
2201 if (likely(!IS_ERR(neigh))) {
2202 int ret;
2203
2204 sock_confirm_neigh(skb, neigh);
2205 dev_xmit_recursion_inc();
2206 ret = neigh_output(neigh, skb, false);
2207 dev_xmit_recursion_dec();
2208 rcu_read_unlock_bh();
2209 return ret;
2210 }
2211 rcu_read_unlock_bh();
2212 IP6_INC_STATS(dev_net(dst->dev),
2213 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2214out_drop:
2215 kfree_skb(skb);
2216 return -ENETDOWN;
2217}
2218
2219static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2220{
2221 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2222 struct net *net = dev_net(dev);
2223 int err, ret = NET_XMIT_DROP;
2224 struct dst_entry *dst;
2225 struct flowi6 fl6 = {
2226 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2227 .flowi6_mark = skb->mark,
2228 .flowlabel = ip6_flowinfo(ip6h),
2229 .flowi6_oif = dev->ifindex,
2230 .flowi6_proto = ip6h->nexthdr,
2231 .daddr = ip6h->daddr,
2232 .saddr = ip6h->saddr,
2233 };
2234
2235 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2236 if (IS_ERR(dst))
2237 goto out_drop;
2238
2239 skb_dst_set(skb, dst);
2240
2241 err = bpf_out_neigh_v6(net, skb);
2242 if (unlikely(net_xmit_eval(err)))
2243 dev->stats.tx_errors++;
2244 else
2245 ret = NET_XMIT_SUCCESS;
2246 goto out_xmit;
2247out_drop:
2248 dev->stats.tx_errors++;
2249 kfree_skb(skb);
2250out_xmit:
2251 return ret;
2252}
2253#else
2254static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2255{
2256 kfree_skb(skb);
2257 return NET_XMIT_DROP;
2258}
2259#endif /* CONFIG_IPV6 */
2260
2261#if IS_ENABLED(CONFIG_INET)
2262static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb)
2263{
2264 struct dst_entry *dst = skb_dst(skb);
2265 struct rtable *rt = container_of(dst, struct rtable, dst);
2266 struct net_device *dev = dst->dev;
2267 u32 hh_len = LL_RESERVED_SPACE(dev);
2268 struct neighbour *neigh;
2269 bool is_v6gw = false;
2270
2271 if (dev_xmit_recursion()) {
2272 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2273 goto out_drop;
2274 }
2275
2276 skb->dev = dev;
2277 skb->tstamp = 0;
2278
2279 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2280 struct sk_buff *skb2;
2281
2282 skb2 = skb_realloc_headroom(skb, hh_len);
2283 if (unlikely(!skb2)) {
2284 kfree_skb(skb);
2285 return -ENOMEM;
2286 }
2287 if (skb->sk)
2288 skb_set_owner_w(skb2, skb->sk);
2289 consume_skb(skb);
2290 skb = skb2;
2291 }
2292
2293 rcu_read_lock_bh();
2294 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2295 if (likely(!IS_ERR(neigh))) {
2296 int ret;
2297
2298 sock_confirm_neigh(skb, neigh);
2299 dev_xmit_recursion_inc();
2300 ret = neigh_output(neigh, skb, is_v6gw);
2301 dev_xmit_recursion_dec();
2302 rcu_read_unlock_bh();
2303 return ret;
2304 }
2305 rcu_read_unlock_bh();
2306out_drop:
2307 kfree_skb(skb);
2308 return -ENETDOWN;
2309}
2310
2311static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2312{
2313 const struct iphdr *ip4h = ip_hdr(skb);
2314 struct net *net = dev_net(dev);
2315 int err, ret = NET_XMIT_DROP;
2316 struct rtable *rt;
2317 struct flowi4 fl4 = {
2318 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2319 .flowi4_mark = skb->mark,
2320 .flowi4_tos = RT_TOS(ip4h->tos),
2321 .flowi4_oif = dev->ifindex,
2322 .flowi4_proto = ip4h->protocol,
2323 .daddr = ip4h->daddr,
2324 .saddr = ip4h->saddr,
2325 };
2326
2327 rt = ip_route_output_flow(net, &fl4, NULL);
2328 if (IS_ERR(rt))
2329 goto out_drop;
2330 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2331 ip_rt_put(rt);
2332 goto out_drop;
2333 }
2334
2335 skb_dst_set(skb, &rt->dst);
2336
2337 err = bpf_out_neigh_v4(net, skb);
2338 if (unlikely(net_xmit_eval(err)))
2339 dev->stats.tx_errors++;
2340 else
2341 ret = NET_XMIT_SUCCESS;
2342 goto out_xmit;
2343out_drop:
2344 dev->stats.tx_errors++;
2345 kfree_skb(skb);
2346out_xmit:
2347 return ret;
2348}
2349#else
2350static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2351{
2352 kfree_skb(skb);
2353 return NET_XMIT_DROP;
2354}
2355#endif /* CONFIG_INET */
2356
2357static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev)
2358{
2359 struct ethhdr *ethh = eth_hdr(skb);
2360
2361 if (unlikely(skb->mac_header >= skb->network_header))
2362 goto out;
2363 bpf_push_mac_rcsum(skb);
2364 if (is_multicast_ether_addr(ethh->h_dest))
2365 goto out;
2366
2367 skb_pull(skb, sizeof(*ethh));
2368 skb_unset_mac_header(skb);
2369 skb_reset_network_header(skb);
2370
2371 if (skb->protocol == htons(ETH_P_IP))
2372 return __bpf_redirect_neigh_v4(skb, dev);
2373 else if (skb->protocol == htons(ETH_P_IPV6))
2374 return __bpf_redirect_neigh_v6(skb, dev);
2375out:
2376 kfree_skb(skb);
2377 return -ENOTSUPP;
2378}
2379
2380/* Internal, non-exposed redirect flags. */
2381enum {
2382 BPF_F_NEIGH = (1ULL << 1),
2383#define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH)
2384};
2385
f3694e00 2386BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
3896d655 2387{
3896d655 2388 struct net_device *dev;
36bbef52
DB		 2389 struct sk_buff *clone;
2390 int ret;
3896d655 2391
b4ab3141 2392 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
781c53bc
DB		 2393 return -EINVAL;
2394
3896d655
AS		 2395 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2396 if (unlikely(!dev))
2397 return -EINVAL;
2398
36bbef52
DB		 2399 clone = skb_clone(skb, GFP_ATOMIC);
2400 if (unlikely(!clone))
3896d655
AS		 2401 return -ENOMEM;
2402
36bbef52
DB		 2403 /* For direct write, we need to keep the invariant that the skbs
2404 * we're dealing with need to be uncloned. Should uncloning fail
2405 * here, we need to free the just generated clone to unclone once
2406 * again.
2407 */
2408 ret = bpf_try_make_head_writable(skb);
2409 if (unlikely(ret)) {
2410 kfree_skb(clone);
2411 return -ENOMEM;
2412 }
2413
4e3264d2 2414 return __bpf_redirect(clone, dev, flags);
3896d655
AS		 2415}
2416
577c50aa 2417static const struct bpf_func_proto bpf_clone_redirect_proto = {
3896d655
AS		 2418 .func = bpf_clone_redirect,
2419 .gpl_only = false,
2420 .ret_type = RET_INTEGER,
2421 .arg1_type = ARG_PTR_TO_CTX,
2422 .arg2_type = ARG_ANYTHING,
2423 .arg3_type = ARG_ANYTHING,
2424};
2425
0b19cc0a
TM		 2426DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2427EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
781c53bc 2428
27b29f63
AS		 2429int skb_do_redirect(struct sk_buff *skb)
2430{
0b19cc0a 2431 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
27b29f63 2432 struct net_device *dev;
b4ab3141 2433 u32 flags = ri->flags;
27b29f63 2434
4b55cf29
THJ		 2435 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2436 ri->tgt_index = 0;
27b29f63
AS		 2437 if (unlikely(!dev)) {
2438 kfree_skb(skb);
2439 return -EINVAL;
2440 }
2441
b4ab3141
DB		 2442 return flags & BPF_F_NEIGH ?
2443 __bpf_redirect_neigh(skb, dev) :
2444 __bpf_redirect(skb, dev, flags);
2445}
2446
2447BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2448{
2449 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2450
2451 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2452 return TC_ACT_SHOT;
2453
2454 ri->flags = flags;
2455 ri->tgt_index = ifindex;
2456
2457 return TC_ACT_REDIRECT;
27b29f63
AS		 2458}
2459
577c50aa 2460static const struct bpf_func_proto bpf_redirect_proto = {
27b29f63
AS		 2461 .func = bpf_redirect,
2462 .gpl_only = false,
2463 .ret_type = RET_INTEGER,
2464 .arg1_type = ARG_ANYTHING,
2465 .arg2_type = ARG_ANYTHING,
2466};
2467
b4ab3141
DB		 2468BPF_CALL_2(bpf_redirect_neigh, u32, ifindex, u64, flags)
2469{
2470 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2471
2472 if (unlikely(flags))
2473 return TC_ACT_SHOT;
2474
2475 ri->flags = BPF_F_NEIGH;
2476 ri->tgt_index = ifindex;
2477
2478 return TC_ACT_REDIRECT;
2479}
2480
2481static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2482 .func = bpf_redirect_neigh,
2483 .gpl_only = false,
2484 .ret_type = RET_INTEGER,
2485 .arg1_type = ARG_ANYTHING,
2486 .arg2_type = ARG_ANYTHING,
2487};
2488
604326b4 2489BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2a100317
JF		 2490{
2491 msg->apply_bytes = bytes;
2492 return 0;
2493}
2494
2495static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2496 .func = bpf_msg_apply_bytes,
2497 .gpl_only = false,
2498 .ret_type = RET_INTEGER,
2499 .arg1_type = ARG_PTR_TO_CTX,
2500 .arg2_type = ARG_ANYTHING,
2501};
2502
604326b4 2503BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
91843d54
JF		 2504{
2505 msg->cork_bytes = bytes;
2506 return 0;
2507}
2508
2509static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2510 .func = bpf_msg_cork_bytes,
2511 .gpl_only = false,
2512 .ret_type = RET_INTEGER,
2513 .arg1_type = ARG_PTR_TO_CTX,
2514 .arg2_type = ARG_ANYTHING,
2515};
2516
604326b4
DB		 2517BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2518 u32, end, u64, flags)
015632bb 2519{
604326b4
DB		 2520 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2521 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2522 struct scatterlist *sge;
2523 u8 *raw, *to, *from;
015632bb
JF		 2524 struct page *page;
2525
2526 if (unlikely(flags || end <= start))
2527 return -EINVAL;
2528
2529 /* First find the starting scatterlist element */
604326b4 2530 i = msg->sg.start;
015632bb 2531 do {
6562e29c 2532 offset += len;
604326b4 2533 len = sk_msg_elem(msg, i)->length;
015632bb
JF		 2534 if (start < offset + len)
2535 break;
604326b4
DB		 2536 sk_msg_iter_var_next(i);
2537 } while (i != msg->sg.end);
015632bb
JF		 2538
2539 if (unlikely(start >= offset + len))
2540 return -EINVAL;
2541
604326b4 2542 first_sge = i;
5b24109b
DB		 2543 /* The start may point into the sg element so we need to also
2544 * account for the headroom.
2545 */
2546 bytes_sg_total = start - offset + bytes;
163ab96b 2547 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
015632bb 2548 goto out;
015632bb
JF		 2549
2550 /* At this point we need to linearize multiple scatterlist
2551 * elements or a single shared page. Either way we need to
2552 * copy into a linear buffer exclusively owned by BPF. Then
2553 * place the buffer in the scatterlist and fixup the original
2554 * entries by removing the entries now in the linear buffer
2555 * and shifting the remaining entries. For now we do not try
2556 * to copy partial entries to avoid complexity of running out
2557 * of sg_entry slots. The downside is reading a single byte
2558 * will copy the entire sg entry.
2559 */
2560 do {
604326b4
DB		 2561 copy += sk_msg_elem(msg, i)->length;
2562 sk_msg_iter_var_next(i);
5b24109b 2563 if (bytes_sg_total <= copy)
015632bb 2564 break;
604326b4
DB		 2565 } while (i != msg->sg.end);
2566 last_sge = i;
015632bb 2567
5b24109b 2568 if (unlikely(bytes_sg_total > copy))
015632bb
JF		 2569 return -EINVAL;
2570
4c3d795c
TD		 2571 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2572 get_order(copy));
015632bb
JF		 2573 if (unlikely(!page))
2574 return -ENOMEM;
015632bb 2575
604326b4
DB		 2576 raw = page_address(page);
2577 i = first_sge;
015632bb 2578 do {
604326b4
DB		 2579 sge = sk_msg_elem(msg, i);
2580 from = sg_virt(sge);
2581 len = sge->length;
2582 to = raw + poffset;
015632bb
JF		 2583
2584 memcpy(to, from, len);
9db39f4d 2585 poffset += len;
604326b4
DB		 2586 sge->length = 0;
2587 put_page(sg_page(sge));
015632bb 2588
604326b4
DB		 2589 sk_msg_iter_var_next(i);
2590 } while (i != last_sge);
015632bb 2591
604326b4 2592 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
015632bb
JF		 2593
2594 /* To repair sg ring we need to shift entries. If we only
2595 * had a single entry though we can just replace it and
2596 * be done. Otherwise walk the ring and shift the entries.
2597 */
604326b4
DB		 2598 WARN_ON_ONCE(last_sge == first_sge);
2599 shift = last_sge > first_sge ?
2600 last_sge - first_sge - 1 :
031097d9 2601 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
015632bb
JF		 2602 if (!shift)
2603 goto out;
2604
604326b4
DB		 2605 i = first_sge;
2606 sk_msg_iter_var_next(i);
015632bb 2607 do {
604326b4 2608 u32 move_from;
015632bb 2609
031097d9
JK		 2610 if (i + shift >= NR_MSG_FRAG_IDS)
2611 move_from = i + shift - NR_MSG_FRAG_IDS;
015632bb
JF		 2612 else
2613 move_from = i + shift;
604326b4 2614 if (move_from == msg->sg.end)
015632bb
JF		 2615 break;
2616
604326b4
DB		 2617 msg->sg.data[i] = msg->sg.data[move_from];
2618 msg->sg.data[move_from].length = 0;
2619 msg->sg.data[move_from].page_link = 0;
2620 msg->sg.data[move_from].offset = 0;
2621 sk_msg_iter_var_next(i);
015632bb 2622 } while (1);
604326b4
DB		 2623
2624 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
031097d9 2625 msg->sg.end - shift + NR_MSG_FRAG_IDS :
604326b4 2626 msg->sg.end - shift;
015632bb 2627out:
604326b4 2628 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
015632bb 2629 msg->data_end = msg->data + bytes;
015632bb
JF		 2630 return 0;
2631}
2632
2633static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2634 .func = bpf_msg_pull_data,
2635 .gpl_only = false,
2636 .ret_type = RET_INTEGER,
2637 .arg1_type = ARG_PTR_TO_CTX,
2638 .arg2_type = ARG_ANYTHING,
2639 .arg3_type = ARG_ANYTHING,
2640 .arg4_type = ARG_ANYTHING,
2641};
2642
6fff607e
JF		 2643BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2644 u32, len, u64, flags)
2645{
2646 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
6562e29c 2647 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
6fff607e
JF		 2648 u8 *raw, *to, *from;
2649 struct page *page;
2650
2651 if (unlikely(flags))
2652 return -EINVAL;
2653
2654 /* First find the starting scatterlist element */
2655 i = msg->sg.start;
2656 do {
6562e29c 2657 offset += l;
6fff607e
JF		 2658 l = sk_msg_elem(msg, i)->length;
2659
2660 if (start < offset + l)
2661 break;
6fff607e
JF		 2662 sk_msg_iter_var_next(i);
2663 } while (i != msg->sg.end);
2664
2665 if (start >= offset + l)
2666 return -EINVAL;
2667
2668 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2669
2670 /* If no space available will fallback to copy, we need at
2671 * least one scatterlist elem available to push data into
2672 * when start aligns to the beginning of an element or two
2673 * when it falls inside an element. We handle the start equals
2674 * offset case because its the common case for inserting a
2675 * header.
2676 */
2677 if (!space || (space == 1 && start != offset))
2678 copy = msg->sg.data[i].length;
2679
2680 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2681 get_order(copy + len));
2682 if (unlikely(!page))
2683 return -ENOMEM;
2684
2685 if (copy) {
2686 int front, back;
2687
2688 raw = page_address(page);
2689
2690 psge = sk_msg_elem(msg, i);
2691 front = start - offset;
2692 back = psge->length - front;
2693 from = sg_virt(psge);
2694
2695 if (front)
2696 memcpy(raw, from, front);
2697
2698 if (back) {
2699 from += front;
2700 to = raw + front + len;
2701
2702 memcpy(to, from, back);
2703 }
2704
2705 put_page(sg_page(psge));
2706 } else if (start - offset) {
2707 psge = sk_msg_elem(msg, i);
2708 rsge = sk_msg_elem_cpy(msg, i);
2709
2710 psge->length = start - offset;
2711 rsge.length -= psge->length;
2712 rsge.offset += start;
2713
2714 sk_msg_iter_var_next(i);
2715 sg_unmark_end(psge);
cf21e9ba 2716 sg_unmark_end(&rsge);
6fff607e
JF		 2717 sk_msg_iter_next(msg, end);
2718 }
2719
2720 /* Slot(s) to place newly allocated data */
2721 new = i;
2722
2723 /* Shift one or two slots as needed */
2724 if (!copy) {
2725 sge = sk_msg_elem_cpy(msg, i);
2726
2727 sk_msg_iter_var_next(i);
2728 sg_unmark_end(&sge);
2729 sk_msg_iter_next(msg, end);
2730
2731 nsge = sk_msg_elem_cpy(msg, i);
2732 if (rsge.length) {
2733 sk_msg_iter_var_next(i);
2734 nnsge = sk_msg_elem_cpy(msg, i);
2735 }
2736
2737 while (i != msg->sg.end) {
2738 msg->sg.data[i] = sge;
2739 sge = nsge;
2740 sk_msg_iter_var_next(i);
2741 if (rsge.length) {
2742 nsge = nnsge;
2743 nnsge = sk_msg_elem_cpy(msg, i);
2744 } else {
2745 nsge = sk_msg_elem_cpy(msg, i);
2746 }
2747 }
2748 }
2749
2750 /* Place newly allocated data buffer */
2751 sk_mem_charge(msg->sk, len);
2752 msg->sg.size += len;
163ab96b 2753 __clear_bit(new, &msg->sg.copy);
6fff607e
JF		 2754 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2755 if (rsge.length) {
2756 get_page(sg_page(&rsge));
2757 sk_msg_iter_var_next(new);
2758 msg->sg.data[new] = rsge;
2759 }
2760
2761 sk_msg_compute_data_pointers(msg);
2762 return 0;
2763}
2764
2765static const struct bpf_func_proto bpf_msg_push_data_proto = {
2766 .func = bpf_msg_push_data,
2767 .gpl_only = false,
2768 .ret_type = RET_INTEGER,
2769 .arg1_type = ARG_PTR_TO_CTX,
2770 .arg2_type = ARG_ANYTHING,
2771 .arg3_type = ARG_ANYTHING,
2772 .arg4_type = ARG_ANYTHING,
2773};
2774
7246d8ed
JF		 2775static void sk_msg_shift_left(struct sk_msg *msg, int i)
2776{
2777 int prev;
2778
2779 do {
2780 prev = i;
2781 sk_msg_iter_var_next(i);
2782 msg->sg.data[prev] = msg->sg.data[i];
2783 } while (i != msg->sg.end);
2784
2785 sk_msg_iter_prev(msg, end);
2786}
2787
2788static void sk_msg_shift_right(struct sk_msg *msg, int i)
2789{
2790 struct scatterlist tmp, sge;
2791
2792 sk_msg_iter_next(msg, end);
2793 sge = sk_msg_elem_cpy(msg, i);
2794 sk_msg_iter_var_next(i);
2795 tmp = sk_msg_elem_cpy(msg, i);
2796
2797 while (i != msg->sg.end) {
2798 msg->sg.data[i] = sge;
2799 sk_msg_iter_var_next(i);
2800 sge = tmp;
2801 tmp = sk_msg_elem_cpy(msg, i);
2802 }
2803}
2804
2805BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2806 u32, len, u64, flags)
2807{
6562e29c 2808 u32 i = 0, l = 0, space, offset = 0;
7246d8ed
JF		 2809 u64 last = start + len;
2810 int pop;
2811
2812 if (unlikely(flags))
2813 return -EINVAL;
2814
2815 /* First find the starting scatterlist element */
2816 i = msg->sg.start;
2817 do {
6562e29c 2818 offset += l;
7246d8ed
JF		 2819 l = sk_msg_elem(msg, i)->length;
2820
2821 if (start < offset + l)
2822 break;
7246d8ed
JF		 2823 sk_msg_iter_var_next(i);
2824 } while (i != msg->sg.end);
2825
2826 /* Bounds checks: start and pop must be inside message */
2827 if (start >= offset + l || last >= msg->sg.size)
2828 return -EINVAL;
2829
2830 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2831
2832 pop = len;
2833 /* --------------| offset
2834 * -| start |-------- len -------|
2835 *
2836 * |----- a ----|-------- pop -------|----- b ----|
2837 * |______________________________________________| length
2838 *
2839 *
2840 * a: region at front of scatter element to save
2841 * b: region at back of scatter element to save when length > A + pop
2842 * pop: region to pop from element, same as input 'pop' here will be
2843 * decremented below per iteration.
2844 *
2845 * Two top-level cases to handle when start != offset, first B is non
2846 * zero and second B is zero corresponding to when a pop includes more
2847 * than one element.
2848 *
2849 * Then if B is non-zero AND there is no space allocate space and
2850 * compact A, B regions into page. If there is space shift ring to
2851 * the rigth free'ing the next element in ring to place B, leaving
2852 * A untouched except to reduce length.
2853 */
2854 if (start != offset) {
2855 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2856 int a = start;
2857 int b = sge->length - pop - a;
2858
2859 sk_msg_iter_var_next(i);
2860
2861 if (pop < sge->length - a) {
2862 if (space) {
2863 sge->length = a;
2864 sk_msg_shift_right(msg, i);
2865 nsge = sk_msg_elem(msg, i);
2866 get_page(sg_page(sge));
2867 sg_set_page(nsge,
2868 sg_page(sge),
2869 b, sge->offset + pop + a);
2870 } else {
2871 struct page *page, *orig;
2872 u8 *to, *from;
2873
2874 page = alloc_pages(__GFP_NOWARN |
2875 __GFP_COMP | GFP_ATOMIC,
2876 get_order(a + b));
2877 if (unlikely(!page))
2878 return -ENOMEM;
2879
2880 sge->length = a;
2881 orig = sg_page(sge);
2882 from = sg_virt(sge);
2883 to = page_address(page);
2884 memcpy(to, from, a);
2885 memcpy(to + a, from + a + pop, b);
2886 sg_set_page(sge, page, a + b, 0);
2887 put_page(orig);
2888 }
2889 pop = 0;
2890 } else if (pop >= sge->length - a) {
7246d8ed 2891 pop -= (sge->length - a);
3e104c23 2892 sge->length = a;
7246d8ed
JF		 2893 }
2894 }
2895
2896 /* From above the current layout _must_ be as follows,
2897 *
2898 * -| offset
2899 * -| start
2900 *
2901 * |---- pop ---|---------------- b ------------|
2902 * |____________________________________________| length
2903 *
2904 * Offset and start of the current msg elem are equal because in the
2905 * previous case we handled offset != start and either consumed the
2906 * entire element and advanced to the next element OR pop == 0.
2907 *
2908 * Two cases to handle here are first pop is less than the length
2909 * leaving some remainder b above. Simply adjust the element's layout
2910 * in this case. Or pop >= length of the element so that b = 0. In this
2911 * case advance to next element decrementing pop.
2912 */
2913 while (pop) {
2914 struct scatterlist *sge = sk_msg_elem(msg, i);
2915
2916 if (pop < sge->length) {
2917 sge->length -= pop;
2918 sge->offset += pop;
2919 pop = 0;
2920 } else {
2921 pop -= sge->length;
2922 sk_msg_shift_left(msg, i);
2923 }
2924 sk_msg_iter_var_next(i);
2925 }
2926
2927 sk_mem_uncharge(msg->sk, len - pop);
2928 msg->sg.size -= (len - pop);
2929 sk_msg_compute_data_pointers(msg);
2930 return 0;
2931}
2932
2933static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2934 .func = bpf_msg_pop_data,
2935 .gpl_only = false,
2936 .ret_type = RET_INTEGER,
2937 .arg1_type = ARG_PTR_TO_CTX,
2938 .arg2_type = ARG_ANYTHING,
2939 .arg3_type = ARG_ANYTHING,
2940 .arg4_type = ARG_ANYTHING,
2941};
2942
5a52ae4e
DB		 2943#ifdef CONFIG_CGROUP_NET_CLASSID
2944BPF_CALL_0(bpf_get_cgroup_classid_curr)
2945{
2946 return __task_get_classid(current);
2947}
2948
2949static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
2950 .func = bpf_get_cgroup_classid_curr,
2951 .gpl_only = false,
2952 .ret_type = RET_INTEGER,
2953};
b426ce83
DB		 2954
2955BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
2956{
2957 struct sock *sk = skb_to_full_sk(skb);
2958
2959 if (!sk || !sk_fullsock(sk))
2960 return 0;
2961
2962 return sock_cgroup_classid(&sk->sk_cgrp_data);
2963}
2964
2965static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
2966 .func = bpf_skb_cgroup_classid,
2967 .gpl_only = false,
2968 .ret_type = RET_INTEGER,
2969 .arg1_type = ARG_PTR_TO_CTX,
2970};
5a52ae4e
DB		 2971#endif
2972
f3694e00 2973BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
8d20aabe 2974{
f3694e00 2975 return task_get_classid(skb);
8d20aabe
DB		 2976}
2977
2978static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2979 .func = bpf_get_cgroup_classid,
2980 .gpl_only = false,
2981 .ret_type = RET_INTEGER,
2982 .arg1_type = ARG_PTR_TO_CTX,
2983};
2984
f3694e00 2985BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
c46646d0 2986{
f3694e00 2987 return dst_tclassid(skb);
c46646d0
DB		 2988}
2989
2990static const struct bpf_func_proto bpf_get_route_realm_proto = {
2991 .func = bpf_get_route_realm,
2992 .gpl_only = false,
2993 .ret_type = RET_INTEGER,
2994 .arg1_type = ARG_PTR_TO_CTX,
2995};
2996
f3694e00 2997BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
13c5c240
DB		 2998{
2999 /* If skb_clear_hash() was called due to mangling, we can
3000 * trigger SW recalculation here. Later access to hash
3001 * can then use the inline skb->hash via context directly
3002 * instead of calling this helper again.
3003 */
f3694e00 3004 return skb_get_hash(skb);
13c5c240
DB		 3005}
3006
3007static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3008 .func = bpf_get_hash_recalc,
3009 .gpl_only = false,
3010 .ret_type = RET_INTEGER,
3011 .arg1_type = ARG_PTR_TO_CTX,
3012};
3013
7a4b28c6
DB		 3014BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3015{
3016 /* After all direct packet write, this can be used once for
3017 * triggering a lazy recalc on next skb_get_hash() invocation.
3018 */
3019 skb_clear_hash(skb);
3020 return 0;
3021}
3022
3023static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3024 .func = bpf_set_hash_invalid,
3025 .gpl_only = false,
3026 .ret_type = RET_INTEGER,
3027 .arg1_type = ARG_PTR_TO_CTX,
3028};
3029
ded092cd
DB		 3030BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3031{
3032 /* Set user specified hash as L4(+), so that it gets returned
3033 * on skb_get_hash() call unless BPF prog later on triggers a
3034 * skb_clear_hash().
3035 */
3036 __skb_set_sw_hash(skb, hash, true);
3037 return 0;
3038}
3039
3040static const struct bpf_func_proto bpf_set_hash_proto = {
3041 .func = bpf_set_hash,
3042 .gpl_only = false,
3043 .ret_type = RET_INTEGER,
3044 .arg1_type = ARG_PTR_TO_CTX,
3045 .arg2_type = ARG_ANYTHING,
3046};
3047
f3694e00
DB		 3048BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3049 u16, vlan_tci)
4e10df9a 3050{
db58ba45 3051 int ret;
4e10df9a
AS		 3052
3053 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3054 vlan_proto != htons(ETH_P_8021AD)))
3055 vlan_proto = htons(ETH_P_8021Q);
3056
8065694e 3057 bpf_push_mac_rcsum(skb);
db58ba45 3058 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
8065694e
DB		 3059 bpf_pull_mac_rcsum(skb);
3060
6aaae2b6 3061 bpf_compute_data_pointers(skb);
db58ba45 3062 return ret;
4e10df9a
AS		 3063}
3064
93731ef0 3065static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
4e10df9a
AS		 3066 .func = bpf_skb_vlan_push,
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};
3073
f3694e00 3074BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
4e10df9a 3075{
db58ba45 3076 int ret;
4e10df9a 3077
8065694e 3078 bpf_push_mac_rcsum(skb);
db58ba45 3079 ret = skb_vlan_pop(skb);
8065694e
DB		 3080 bpf_pull_mac_rcsum(skb);
3081
6aaae2b6 3082 bpf_compute_data_pointers(skb);
db58ba45 3083 return ret;
4e10df9a
AS		 3084}
3085
93731ef0 3086static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
4e10df9a
AS		 3087 .func = bpf_skb_vlan_pop,
3088 .gpl_only = false,
3089 .ret_type = RET_INTEGER,
3090 .arg1_type = ARG_PTR_TO_CTX,
3091};
3092
6578171a
DB		 3093static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3094{
3095 /* Caller already did skb_cow() with len as headroom,
3096 * so no need to do it here.
3097 */
3098 skb_push(skb, len);
3099 memmove(skb->data, skb->data + len, off);
3100 memset(skb->data + off, 0, len);
3101
3102 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3103 * needed here as it does not change the skb->csum
3104 * result for checksum complete when summing over
3105 * zeroed blocks.
3106 */
3107 return 0;
3108}
3109
3110static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3111{
3112 /* skb_ensure_writable() is not needed here, as we're
3113 * already working on an uncloned skb.
3114 */
3115 if (unlikely(!pskb_may_pull(skb, off + len)))
3116 return -ENOMEM;
3117
3118 skb_postpull_rcsum(skb, skb->data + off, len);
3119 memmove(skb->data + len, skb->data, off);
3120 __skb_pull(skb, len);
3121
3122 return 0;
3123}
3124
3125static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3126{
3127 bool trans_same = skb->transport_header == skb->network_header;
3128 int ret;
3129
3130 /* There's no need for __skb_push()/__skb_pull() pair to
3131 * get to the start of the mac header as we're guaranteed
3132 * to always start from here under eBPF.
3133 */
3134 ret = bpf_skb_generic_push(skb, off, len);
3135 if (likely(!ret)) {
3136 skb->mac_header -= len;
3137 skb->network_header -= len;
3138 if (trans_same)
3139 skb->transport_header = skb->network_header;
3140 }
3141
3142 return ret;
3143}
3144
3145static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3146{
3147 bool trans_same = skb->transport_header == skb->network_header;
3148 int ret;
3149
3150 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3151 ret = bpf_skb_generic_pop(skb, off, len);
3152 if (likely(!ret)) {
3153 skb->mac_header += len;
3154 skb->network_header += len;
3155 if (trans_same)
3156 skb->transport_header = skb->network_header;
3157 }
3158
3159 return ret;
3160}
3161
3162static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3163{
3164 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
0daf4349 3165 u32 off = skb_mac_header_len(skb);
6578171a
DB		 3166 int ret;
3167
4c3024de 3168 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
d02f51cb
DA		 3169 return -ENOTSUPP;
3170
6578171a
DB		 3171 ret = skb_cow(skb, len_diff);
3172 if (unlikely(ret < 0))
3173 return ret;
3174
3175 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3176 if (unlikely(ret < 0))
3177 return ret;
3178
3179 if (skb_is_gso(skb)) {
d02f51cb
DA		 3180 struct skb_shared_info *shinfo = skb_shinfo(skb);
3181
880388aa
DM		 3182 /* SKB_GSO_TCPV4 needs to be changed into
3183 * SKB_GSO_TCPV6.
6578171a 3184 */
d02f51cb
DA		 3185 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3186 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3187 shinfo->gso_type |= SKB_GSO_TCPV6;
6578171a
DB		 3188 }
3189
3190 /* Due to IPv6 header, MSS needs to be downgraded. */
d02f51cb 3191 skb_decrease_gso_size(shinfo, len_diff);
6578171a 3192 /* Header must be checked, and gso_segs recomputed. */
d02f51cb
DA		 3193 shinfo->gso_type |= SKB_GSO_DODGY;
3194 shinfo->gso_segs = 0;
6578171a
DB		 3195 }
3196
3197 skb->protocol = htons(ETH_P_IPV6);
3198 skb_clear_hash(skb);
3199
3200 return 0;
3201}
3202
3203static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3204{
3205 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
0daf4349 3206 u32 off = skb_mac_header_len(skb);
6578171a
DB		 3207 int ret;
3208
4c3024de 3209 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
d02f51cb
DA		 3210 return -ENOTSUPP;
3211
6578171a
DB		 3212 ret = skb_unclone(skb, GFP_ATOMIC);
3213 if (unlikely(ret < 0))
3214 return ret;
3215
3216 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3217 if (unlikely(ret < 0))
3218 return ret;
3219
3220 if (skb_is_gso(skb)) {
d02f51cb
DA		 3221 struct skb_shared_info *shinfo = skb_shinfo(skb);
3222
880388aa
DM		 3223 /* SKB_GSO_TCPV6 needs to be changed into
3224 * SKB_GSO_TCPV4.
6578171a 3225 */
d02f51cb
DA		 3226 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3227 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3228 shinfo->gso_type |= SKB_GSO_TCPV4;
6578171a
DB		 3229 }
3230
3231 /* Due to IPv4 header, MSS can be upgraded. */
d02f51cb 3232 skb_increase_gso_size(shinfo, len_diff);
6578171a 3233 /* Header must be checked, and gso_segs recomputed. */
d02f51cb
DA		 3234 shinfo->gso_type |= SKB_GSO_DODGY;
3235 shinfo->gso_segs = 0;
6578171a
DB		 3236 }
3237
3238 skb->protocol = htons(ETH_P_IP);
3239 skb_clear_hash(skb);
3240
3241 return 0;
3242}
3243
3244static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3245{
3246 __be16 from_proto = skb->protocol;
3247
3248 if (from_proto == htons(ETH_P_IP) &&
3249 to_proto == htons(ETH_P_IPV6))
3250 return bpf_skb_proto_4_to_6(skb);
3251
3252 if (from_proto == htons(ETH_P_IPV6) &&
3253 to_proto == htons(ETH_P_IP))
3254 return bpf_skb_proto_6_to_4(skb);
3255
3256 return -ENOTSUPP;
3257}
3258
f3694e00
DB		 3259BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3260 u64, flags)
6578171a 3261{
6578171a
DB		 3262 int ret;
3263
3264 if (unlikely(flags))
3265 return -EINVAL;
3266
3267 /* General idea is that this helper does the basic groundwork
3268 * needed for changing the protocol, and eBPF program fills the
3269 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3270 * and other helpers, rather than passing a raw buffer here.
3271 *
3272 * The rationale is to keep this minimal and without a need to
3273 * deal with raw packet data. F.e. even if we would pass buffers
3274 * here, the program still needs to call the bpf_lX_csum_replace()
3275 * helpers anyway. Plus, this way we keep also separation of
3276 * concerns, since f.e. bpf_skb_store_bytes() should only take
3277 * care of stores.
3278 *
3279 * Currently, additional options and extension header space are
3280 * not supported, but flags register is reserved so we can adapt
3281 * that. For offloads, we mark packet as dodgy, so that headers
3282 * need to be verified first.
3283 */
3284 ret = bpf_skb_proto_xlat(skb, proto);
6aaae2b6 3285 bpf_compute_data_pointers(skb);
6578171a
DB		 3286 return ret;
3287}
3288
3289static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3290 .func = bpf_skb_change_proto,
3291 .gpl_only = false,
3292 .ret_type = RET_INTEGER,
3293 .arg1_type = ARG_PTR_TO_CTX,
3294 .arg2_type = ARG_ANYTHING,
3295 .arg3_type = ARG_ANYTHING,
3296};
3297
f3694e00 3298BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
d2485c42 3299{
d2485c42 3300 /* We only allow a restricted subset to be changed for now. */
45c7fffa
DB		 3301 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3302 !skb_pkt_type_ok(pkt_type)))
d2485c42
DB		 3303 return -EINVAL;
3304
3305 skb->pkt_type = pkt_type;
3306 return 0;
3307}
3308
3309static const struct bpf_func_proto bpf_skb_change_type_proto = {
3310 .func = bpf_skb_change_type,
3311 .gpl_only = false,
3312 .ret_type = RET_INTEGER,
3313 .arg1_type = ARG_PTR_TO_CTX,
3314 .arg2_type = ARG_ANYTHING,
3315};
3316
2be7e212
DB		 3317static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3318{
3319 switch (skb->protocol) {
3320 case htons(ETH_P_IP):
3321 return sizeof(struct iphdr);
3322 case htons(ETH_P_IPV6):
3323 return sizeof(struct ipv6hdr);
3324 default:
3325 return ~0U;
3326 }
3327}
3328
868d5235
WB		 3329#define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3330 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3331
3332#define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3333 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3334 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
58dfc900
AM		 3335 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3336 BPF_F_ADJ_ROOM_ENCAP_L2( \
3337 BPF_ADJ_ROOM_ENCAP_L2_MASK))
2278f6cc
WB		 3338
3339static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3340 u64 flags)
2be7e212 3341{
58dfc900 3342 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
868d5235 3343 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
62b31b42 3344 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
868d5235 3345 unsigned int gso_type = SKB_GSO_DODGY;
2be7e212
DB		 3346 int ret;
3347
2278f6cc
WB		 3348 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3349 /* udp gso_size delineates datagrams, only allow if fixed */
3350 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3351 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3352 return -ENOTSUPP;
3353 }
d02f51cb 3354
908adce6 3355 ret = skb_cow_head(skb, len_diff);
2be7e212
DB		 3356 if (unlikely(ret < 0))
3357 return ret;
3358
868d5235
WB		 3359 if (encap) {
3360 if (skb->protocol != htons(ETH_P_IP) &&
3361 skb->protocol != htons(ETH_P_IPV6))
3362 return -ENOTSUPP;
3363
3364 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3365 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3366 return -EINVAL;
3367
3368 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3369 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3370 return -EINVAL;
3371
3372 if (skb->encapsulation)
3373 return -EALREADY;
3374
3375 mac_len = skb->network_header - skb->mac_header;
3376 inner_net = skb->network_header;
58dfc900
AM		 3377 if (inner_mac_len > len_diff)
3378 return -EINVAL;
868d5235
WB		 3379 inner_trans = skb->transport_header;
3380 }
3381
2be7e212
DB		 3382 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3383 if (unlikely(ret < 0))
3384 return ret;
3385
868d5235 3386 if (encap) {
58dfc900 3387 skb->inner_mac_header = inner_net - inner_mac_len;
868d5235
WB		 3388 skb->inner_network_header = inner_net;
3389 skb->inner_transport_header = inner_trans;
3390 skb_set_inner_protocol(skb, skb->protocol);
3391
3392 skb->encapsulation = 1;
3393 skb_set_network_header(skb, mac_len);
3394
3395 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3396 gso_type |= SKB_GSO_UDP_TUNNEL;
3397 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3398 gso_type |= SKB_GSO_GRE;
3399 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3400 gso_type |= SKB_GSO_IPXIP6;
58dfc900 3401 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
868d5235
WB		 3402 gso_type |= SKB_GSO_IPXIP4;
3403
3404 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3405 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3406 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3407 sizeof(struct ipv6hdr) :
3408 sizeof(struct iphdr);
3409
3410 skb_set_transport_header(skb, mac_len + nh_len);
3411 }
1b00e0df
WB		 3412
3413 /* Match skb->protocol to new outer l3 protocol */
3414 if (skb->protocol == htons(ETH_P_IP) &&
3415 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3416 skb->protocol = htons(ETH_P_IPV6);
3417 else if (skb->protocol == htons(ETH_P_IPV6) &&
3418 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3419 skb->protocol = htons(ETH_P_IP);
868d5235
WB		 3420 }
3421
2be7e212 3422 if (skb_is_gso(skb)) {
d02f51cb
DA		 3423 struct skb_shared_info *shinfo = skb_shinfo(skb);
3424
2be7e212 3425 /* Due to header grow, MSS needs to be downgraded. */
2278f6cc
WB		 3426 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3427 skb_decrease_gso_size(shinfo, len_diff);
3428
2be7e212 3429 /* Header must be checked, and gso_segs recomputed. */
868d5235 3430 shinfo->gso_type |= gso_type;
d02f51cb 3431 shinfo->gso_segs = 0;
2be7e212
DB		 3432 }
3433
3434 return 0;
3435}
3436
2278f6cc
WB		 3437static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3438 u64 flags)
2be7e212 3439{
2be7e212
DB		 3440 int ret;
3441
836e66c2
DB		 3442 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3443 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
43537b8e
WB		 3444 return -EINVAL;
3445
2278f6cc
WB		 3446 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3447 /* udp gso_size delineates datagrams, only allow if fixed */
3448 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3449 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3450 return -ENOTSUPP;
3451 }
d02f51cb 3452
2be7e212
DB		 3453 ret = skb_unclone(skb, GFP_ATOMIC);
3454 if (unlikely(ret < 0))
3455 return ret;
3456
3457 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3458 if (unlikely(ret < 0))
3459 return ret;
3460
3461 if (skb_is_gso(skb)) {
d02f51cb
DA		 3462 struct skb_shared_info *shinfo = skb_shinfo(skb);
3463
2be7e212 3464 /* Due to header shrink, MSS can be upgraded. */
2278f6cc
WB		 3465 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3466 skb_increase_gso_size(shinfo, len_diff);
3467
2be7e212 3468 /* Header must be checked, and gso_segs recomputed. */
d02f51cb
DA		 3469 shinfo->gso_type |= SKB_GSO_DODGY;
3470 shinfo->gso_segs = 0;
2be7e212
DB		 3471 }
3472
3473 return 0;
3474}
3475
3476static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3477{
0c6bc6e5
JF		 3478 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3479 SKB_MAX_ALLOC;
2be7e212
DB		 3480}
3481
14aa3192
WB		 3482BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3483 u32, mode, u64, flags)
2be7e212 3484{
2be7e212
DB		 3485 u32 len_cur, len_diff_abs = abs(len_diff);
3486 u32 len_min = bpf_skb_net_base_len(skb);
3487 u32 len_max = __bpf_skb_max_len(skb);
3488 __be16 proto = skb->protocol;
3489 bool shrink = len_diff < 0;
14aa3192 3490 u32 off;
2be7e212
DB		 3491 int ret;
3492
836e66c2
DB		 3493 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3494 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
14aa3192 3495 return -EINVAL;
2be7e212
DB		 3496 if (unlikely(len_diff_abs > 0xfffU))
3497 return -EFAULT;
3498 if (unlikely(proto != htons(ETH_P_IP) &&
3499 proto != htons(ETH_P_IPV6)))
3500 return -ENOTSUPP;
3501
14aa3192
WB		 3502 off = skb_mac_header_len(skb);
3503 switch (mode) {
3504 case BPF_ADJ_ROOM_NET:
3505 off += bpf_skb_net_base_len(skb);
3506 break;
3507 case BPF_ADJ_ROOM_MAC:
3508 break;
3509 default:
3510 return -ENOTSUPP;
3511 }
3512
2be7e212 3513 len_cur = skb->len - skb_network_offset(skb);
2be7e212
DB		 3514 if ((shrink && (len_diff_abs >= len_cur ||
3515 len_cur - len_diff_abs < len_min)) ||
3516 (!shrink && (skb->len + len_diff_abs > len_max &&
3517 !skb_is_gso(skb))))
3518 return -ENOTSUPP;
3519
2278f6cc
WB		 3520 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3521 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
836e66c2
DB		 3522 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3523 __skb_reset_checksum_unnecessary(skb);
2be7e212 3524
6aaae2b6 3525 bpf_compute_data_pointers(skb);
e4a6a342 3526 return ret;
2be7e212
DB		 3527}
3528
2be7e212
DB		 3529static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3530 .func = bpf_skb_adjust_room,
3531 .gpl_only = false,
3532 .ret_type = RET_INTEGER,
3533 .arg1_type = ARG_PTR_TO_CTX,
3534 .arg2_type = ARG_ANYTHING,
3535 .arg3_type = ARG_ANYTHING,
3536 .arg4_type = ARG_ANYTHING,
3537};
3538
5293efe6
DB		 3539static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3540{
3541 u32 min_len = skb_network_offset(skb);
3542
3543 if (skb_transport_header_was_set(skb))
3544 min_len = skb_transport_offset(skb);
3545 if (skb->ip_summed == CHECKSUM_PARTIAL)
3546 min_len = skb_checksum_start_offset(skb) +
3547 skb->csum_offset + sizeof(__sum16);
3548 return min_len;
3549}
3550
5293efe6
DB		 3551static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3552{
3553 unsigned int old_len = skb->len;
3554 int ret;
3555
3556 ret = __skb_grow_rcsum(skb, new_len);
3557 if (!ret)
3558 memset(skb->data + old_len, 0, new_len - old_len);
3559 return ret;
3560}
3561
3562static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3563{
3564 return __skb_trim_rcsum(skb, new_len);
3565}
3566
0ea488ff
JF		 3567static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3568 u64 flags)
5293efe6 3569{
5293efe6
DB		 3570 u32 max_len = __bpf_skb_max_len(skb);
3571 u32 min_len = __bpf_skb_min_len(skb);
5293efe6
DB		 3572 int ret;
3573
3574 if (unlikely(flags || new_len > max_len || new_len < min_len))
3575 return -EINVAL;
3576 if (skb->encapsulation)
3577 return -ENOTSUPP;
3578
3579 /* The basic idea of this helper is that it's performing the
3580 * needed work to either grow or trim an skb, and eBPF program
3581 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3582 * bpf_lX_csum_replace() and others rather than passing a raw
3583 * buffer here. This one is a slow path helper and intended
3584 * for replies with control messages.
3585 *
3586 * Like in bpf_skb_change_proto(), we want to keep this rather
3587 * minimal and without protocol specifics so that we are able
3588 * to separate concerns as in bpf_skb_store_bytes() should only
3589 * be the one responsible for writing buffers.
3590 *
3591 * It's really expected to be a slow path operation here for
3592 * control message replies, so we're implicitly linearizing,
3593 * uncloning and drop offloads from the skb by this.
3594 */
3595 ret = __bpf_try_make_writable(skb, skb->len);
3596 if (!ret) {
3597 if (new_len > skb->len)
3598 ret = bpf_skb_grow_rcsum(skb, new_len);
3599 else if (new_len < skb->len)
3600 ret = bpf_skb_trim_rcsum(skb, new_len);
3601 if (!ret && skb_is_gso(skb))
3602 skb_gso_reset(skb);
3603 }
0ea488ff
JF		 3604 return ret;
3605}
3606
3607BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3608 u64, flags)
3609{
3610 int ret = __bpf_skb_change_tail(skb, new_len, flags);
5293efe6 3611
6aaae2b6 3612 bpf_compute_data_pointers(skb);
5293efe6
DB		 3613 return ret;
3614}
3615
3616static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3617 .func = bpf_skb_change_tail,
3618 .gpl_only = false,
3619 .ret_type = RET_INTEGER,
3620 .arg1_type = ARG_PTR_TO_CTX,
3621 .arg2_type = ARG_ANYTHING,
3622 .arg3_type = ARG_ANYTHING,
3623};
3624
0ea488ff 3625BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3a0af8fd 3626 u64, flags)
0ea488ff
JF		 3627{
3628 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3629
3630 bpf_compute_data_end_sk_skb(skb);
3631 return ret;
3632}
3633
3634static const struct bpf_func_proto sk_skb_change_tail_proto = {
3635 .func = sk_skb_change_tail,
3636 .gpl_only = false,
3637 .ret_type = RET_INTEGER,
3638 .arg1_type = ARG_PTR_TO_CTX,
3639 .arg2_type = ARG_ANYTHING,
3640 .arg3_type = ARG_ANYTHING,
3641};
3642
3643static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3644 u64 flags)
3a0af8fd
TG		 3645{
3646 u32 max_len = __bpf_skb_max_len(skb);
3647 u32 new_len = skb->len + head_room;
3648 int ret;
3649
3650 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3651 new_len < skb->len))
3652 return -EINVAL;
3653
3654 ret = skb_cow(skb, head_room);
3655 if (likely(!ret)) {
3656 /* Idea for this helper is that we currently only
3657 * allow to expand on mac header. This means that
3658 * skb->protocol network header, etc, stay as is.
3659 * Compared to bpf_skb_change_tail(), we're more
3660 * flexible due to not needing to linearize or
3661 * reset GSO. Intention for this helper is to be
3662 * used by an L3 skb that needs to push mac header
3663 * for redirection into L2 device.
3664 */
3665 __skb_push(skb, head_room);
3666 memset(skb->data, 0, head_room);
3667 skb_reset_mac_header(skb);
3668 }
3669
0ea488ff
JF		 3670 return ret;
3671}
3672
3673BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3674 u64, flags)
3675{
3676 int ret = __bpf_skb_change_head(skb, head_room, flags);
3677
6aaae2b6 3678 bpf_compute_data_pointers(skb);
0ea488ff 3679 return ret;
3a0af8fd
TG		 3680}
3681
3682static const struct bpf_func_proto bpf_skb_change_head_proto = {
3683 .func = bpf_skb_change_head,
3684 .gpl_only = false,
3685 .ret_type = RET_INTEGER,
3686 .arg1_type = ARG_PTR_TO_CTX,
3687 .arg2_type = ARG_ANYTHING,
3688 .arg3_type = ARG_ANYTHING,
3689};
3690
0ea488ff
JF		 3691BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3692 u64, flags)
3693{
3694 int ret = __bpf_skb_change_head(skb, head_room, flags);
3695
3696 bpf_compute_data_end_sk_skb(skb);
3697 return ret;
3698}
3699
3700static const struct bpf_func_proto sk_skb_change_head_proto = {
3701 .func = sk_skb_change_head,
3702 .gpl_only = false,
3703 .ret_type = RET_INTEGER,
3704 .arg1_type = ARG_PTR_TO_CTX,
3705 .arg2_type = ARG_ANYTHING,
3706 .arg3_type = ARG_ANYTHING,
3707};
de8f3a83
DB		 3708static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3709{
3710 return xdp_data_meta_unsupported(xdp) ? 0 :
3711 xdp->data - xdp->data_meta;
3712}
3713
17bedab2
MKL		 3714BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3715{
6dfb970d 3716 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
de8f3a83 3717 unsigned long metalen = xdp_get_metalen(xdp);
97e19cce 3718 void *data_start = xdp_frame_end + metalen;
17bedab2
MKL		 3719 void *data = xdp->data + offset;
3720
de8f3a83 3721 if (unlikely(data < data_start ||
17bedab2
MKL		 3722 data > xdp->data_end - ETH_HLEN))
3723 return -EINVAL;
3724
de8f3a83
DB		 3725 if (metalen)
3726 memmove(xdp->data_meta + offset,
3727 xdp->data_meta, metalen);
3728 xdp->data_meta += offset;
17bedab2
MKL		 3729 xdp->data = data;
3730
3731 return 0;
3732}
3733
3734static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3735 .func = bpf_xdp_adjust_head,
3736 .gpl_only = false,
3737 .ret_type = RET_INTEGER,
3738 .arg1_type = ARG_PTR_TO_CTX,
3739 .arg2_type = ARG_ANYTHING,
3740};
3741
b32cc5b9
NS		 3742BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3743{
c8741e2b 3744 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
b32cc5b9
NS		 3745 void *data_end = xdp->data_end + offset;
3746
c8741e2b
JDB		 3747 /* Notice that xdp_data_hard_end have reserved some tailroom */
3748 if (unlikely(data_end > data_hard_end))
b32cc5b9
NS		 3749 return -EINVAL;
3750
c8741e2b
JDB		 3751 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3752 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3753 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3754 return -EINVAL;
3755 }
3756
b32cc5b9
NS		 3757 if (unlikely(data_end < xdp->data + ETH_HLEN))
3758 return -EINVAL;
3759
ddb47d51
JDB		 3760 /* Clear memory area on grow, can contain uninit kernel memory */
3761 if (offset > 0)
3762 memset(xdp->data_end, 0, offset);
3763
b32cc5b9
NS		 3764 xdp->data_end = data_end;
3765
3766 return 0;
3767}
3768
3769static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3770 .func = bpf_xdp_adjust_tail,
3771 .gpl_only = false,
3772 .ret_type = RET_INTEGER,
3773 .arg1_type = ARG_PTR_TO_CTX,
3774 .arg2_type = ARG_ANYTHING,
3775};
3776
de8f3a83
DB		 3777BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3778{
97e19cce 3779 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
de8f3a83
DB		 3780 void *meta = xdp->data_meta + offset;
3781 unsigned long metalen = xdp->data - meta;
3782
3783 if (xdp_data_meta_unsupported(xdp))
3784 return -ENOTSUPP;
97e19cce 3785 if (unlikely(meta < xdp_frame_end ||
de8f3a83
DB		 3786 meta > xdp->data))
3787 return -EINVAL;
3788 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3789 (metalen > 32)))
3790 return -EACCES;
3791
3792 xdp->data_meta = meta;
3793
3794 return 0;
3795}
3796
3797static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3798 .func = bpf_xdp_adjust_meta,
3799 .gpl_only = false,
3800 .ret_type = RET_INTEGER,
3801 .arg1_type = ARG_PTR_TO_CTX,
3802 .arg2_type = ARG_ANYTHING,
3803};
3804
9c270af3 3805static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
1170beaa 3806 struct bpf_map *map, struct xdp_buff *xdp)
9c270af3 3807{
1b1a251c 3808 switch (map->map_type) {
6f9d451a 3809 case BPF_MAP_TYPE_DEVMAP:
1170beaa
BT		 3810 case BPF_MAP_TYPE_DEVMAP_HASH:
3811 return dev_map_enqueue(fwd, xdp, dev_rx);
3812 case BPF_MAP_TYPE_CPUMAP:
3813 return cpu_map_enqueue(fwd, xdp, dev_rx);
3814 case BPF_MAP_TYPE_XSKMAP:
3815 return __xsk_map_redirect(fwd, xdp);
1b1a251c 3816 default:
0a29275b 3817 return -EBADRQC;
9c270af3 3818 }
e4a8e817 3819 return 0;
814abfab
JF		 3820}
3821
1d233886 3822void xdp_do_flush(void)
11393cc9 3823{
1d233886 3824 __dev_flush();
332f22a6
BT		 3825 __cpu_map_flush();
3826 __xsk_map_flush();
11393cc9 3827}
1d233886 3828EXPORT_SYMBOL_GPL(xdp_do_flush);
11393cc9 3829
2a68d85f 3830static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
9c270af3
JDB		 3831{
3832 switch (map->map_type) {
3833 case BPF_MAP_TYPE_DEVMAP:
3834 return __dev_map_lookup_elem(map, index);
6f9d451a
THJ		 3835 case BPF_MAP_TYPE_DEVMAP_HASH:
3836 return __dev_map_hash_lookup_elem(map, index);
9c270af3
JDB		 3837 case BPF_MAP_TYPE_CPUMAP:
3838 return __cpu_map_lookup_elem(map, index);
1b1a251c
BT		 3839 case BPF_MAP_TYPE_XSKMAP:
3840 return __xsk_map_lookup_elem(map, index);
9c270af3
JDB		 3841 default:
3842 return NULL;
3843 }
3844}
3845
f6069b9a 3846void bpf_clear_redirect_map(struct bpf_map *map)
7c300131 3847{
f6069b9a
DB		 3848 struct bpf_redirect_info *ri;
3849 int cpu;
3850
3851 for_each_possible_cpu(cpu) {
3852 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3853 /* Avoid polluting remote cacheline due to writes if
3854 * not needed. Once we pass this test, we need the
3855 * cmpxchg() to make sure it hasn't been changed in
3856 * the meantime by remote CPU.
3857 */
3858 if (unlikely(READ_ONCE(ri->map) == map))
3859 cmpxchg(&ri->map, map, NULL);
3860 }
7c300131
DB		 3861}
3862
1d233886
THJ		 3863int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3864 struct bpf_prog *xdp_prog)
97f91a7c 3865{
1d233886
THJ		 3866 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3867 struct bpf_map *map = READ_ONCE(ri->map);
4b55cf29 3868 u32 index = ri->tgt_index;
43e74c02 3869 void *fwd = ri->tgt_value;
4c03bdd7 3870 int err;
97f91a7c 3871
4b55cf29 3872 ri->tgt_index = 0;
43e74c02 3873 ri->tgt_value = NULL;
f6069b9a 3874 WRITE_ONCE(ri->map, NULL);
97f91a7c 3875
1d233886
THJ		 3876 if (unlikely(!map)) {
3877 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3878 if (unlikely(!fwd)) {
3879 err = -EINVAL;
3880 goto err;
3881 }
3882
3883 err = dev_xdp_enqueue(fwd, xdp, dev);
3884 } else {
3885 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
3886 }
3887
f5836ca5
JDB		 3888 if (unlikely(err))
3889 goto err;
3890
59a30896 3891 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
f5836ca5
JDB		 3892 return 0;
3893err:
59a30896 3894 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
97f91a7c
JF		 3895 return err;
3896}
814abfab
JF		 3897EXPORT_SYMBOL_GPL(xdp_do_redirect);
3898
c060bc61
XS		 3899static int xdp_do_generic_redirect_map(struct net_device *dev,
3900 struct sk_buff *skb,
02671e23 3901 struct xdp_buff *xdp,
f6069b9a
DB		 3902 struct bpf_prog *xdp_prog,
3903 struct bpf_map *map)
6103aa96 3904{
0b19cc0a 3905 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4b55cf29 3906 u32 index = ri->tgt_index;
43e74c02 3907 void *fwd = ri->tgt_value;
2facaad6 3908 int err = 0;
6103aa96 3909
4b55cf29 3910 ri->tgt_index = 0;
43e74c02 3911 ri->tgt_value = NULL;
f6069b9a 3912 WRITE_ONCE(ri->map, NULL);
96c5508e 3913
6f9d451a
THJ		 3914 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3915 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
6d5fc195
TM		 3916 struct bpf_dtab_netdev *dst = fwd;
3917
3918 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3919 if (unlikely(err))
9c270af3 3920 goto err;
02671e23
BT		 3921 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3922 struct xdp_sock *xs = fwd;
3923
3924 err = xsk_generic_rcv(xs, xdp);
3925 if (err)
3926 goto err;
3927 consume_skb(skb);
9c270af3
JDB		 3928 } else {
3929 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3930 err = -EBADRQC;
f5836ca5 3931 goto err;
2facaad6 3932 }
6103aa96 3933
9c270af3
JDB		 3934 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3935 return 0;
3936err:
3937 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3938 return err;
3939}
3940
3941int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
02671e23 3942 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
9c270af3 3943{
0b19cc0a 3944 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
f6069b9a 3945 struct bpf_map *map = READ_ONCE(ri->map);
4b55cf29 3946 u32 index = ri->tgt_index;
9c270af3
JDB		 3947 struct net_device *fwd;
3948 int err = 0;
3949
f6069b9a
DB		 3950 if (map)
3951 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3952 map);
4b55cf29 3953 ri->tgt_index = 0;
9c270af3
JDB		 3954 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3955 if (unlikely(!fwd)) {
3956 err = -EINVAL;
f5836ca5 3957 goto err;
2facaad6
JDB		 3958 }
3959
d8d7218a
TM		 3960 err = xdp_ok_fwd_dev(fwd, skb->len);
3961 if (unlikely(err))
9c270af3
JDB		 3962 goto err;
3963
2facaad6 3964 skb->dev = fwd;
9c270af3 3965 _trace_xdp_redirect(dev, xdp_prog, index);
02671e23 3966 generic_xdp_tx(skb, xdp_prog);
f5836ca5
JDB		 3967 return 0;
3968err:
9c270af3 3969 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2facaad6 3970 return err;
6103aa96 3971}
6103aa96 3972
814abfab
JF		 3973BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3974{
0b19cc0a 3975 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
814abfab
JF		 3976
3977 if (unlikely(flags))
3978 return XDP_ABORTED;
3979
814abfab 3980 ri->flags = flags;
4b55cf29 3981 ri->tgt_index = ifindex;
43e74c02 3982 ri->tgt_value = NULL;
f6069b9a 3983 WRITE_ONCE(ri->map, NULL);
e4a8e817 3984
814abfab
JF		 3985 return XDP_REDIRECT;
3986}
3987
3988static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3989 .func = bpf_xdp_redirect,
3990 .gpl_only = false,
3991 .ret_type = RET_INTEGER,
3992 .arg1_type = ARG_ANYTHING,
3993 .arg2_type = ARG_ANYTHING,
3994};
3995
f6069b9a
DB		 3996BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3997 u64, flags)
e4a8e817 3998{
0b19cc0a 3999 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
e4a8e817 4000
43e74c02
THJ		 4001 /* Lower bits of the flags are used as return code on lookup failure */
4002 if (unlikely(flags > XDP_TX))
e4a8e817
DB		 4003 return XDP_ABORTED;
4004
43e74c02
THJ		 4005 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4006 if (unlikely(!ri->tgt_value)) {
4007 /* If the lookup fails we want to clear out the state in the
4008 * redirect_info struct completely, so that if an eBPF program
4009 * performs multiple lookups, the last one always takes
4010 * precedence.
4011 */
4012 WRITE_ONCE(ri->map, NULL);
4013 return flags;
4014 }
4015
e4a8e817 4016 ri->flags = flags;
4b55cf29 4017 ri->tgt_index = ifindex;
f6069b9a 4018 WRITE_ONCE(ri->map, map);
e4a8e817
DB		 4019
4020 return XDP_REDIRECT;
4021}
4022
4023static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4024 .func = bpf_xdp_redirect_map,
4025 .gpl_only = false,
4026 .ret_type = RET_INTEGER,
4027 .arg1_type = ARG_CONST_MAP_PTR,
4028 .arg2_type = ARG_ANYTHING,
4029 .arg3_type = ARG_ANYTHING,
4030};
4031
555c8a86 4032static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
aa7145c1 4033 unsigned long off, unsigned long len)
555c8a86 4034{
aa7145c1 4035 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
555c8a86
DB		 4036
4037 if (unlikely(!ptr))
4038 return len;
4039 if (ptr != dst_buff)
4040 memcpy(dst_buff, ptr, len);
4041
4042 return 0;
4043}
4044
f3694e00
DB		 4045BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4046 u64, flags, void *, meta, u64, meta_size)
555c8a86 4047{
555c8a86 4048 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
555c8a86
DB		 4049
4050 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4051 return -EINVAL;
a7658e1a 4052 if (unlikely(!skb || skb_size > skb->len))
555c8a86
DB		 4053 return -EFAULT;
4054
4055 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4056 bpf_skb_copy);
4057}
4058
4059static const struct bpf_func_proto bpf_skb_event_output_proto = {
4060 .func = bpf_skb_event_output,
4061 .gpl_only = true,
4062 .ret_type = RET_INTEGER,
4063 .arg1_type = ARG_PTR_TO_CTX,
4064 .arg2_type = ARG_CONST_MAP_PTR,
4065 .arg3_type = ARG_ANYTHING,
39f19ebb 4066 .arg4_type = ARG_PTR_TO_MEM,
1728a4f2 4067 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
555c8a86
DB		 4068};
4069
9436ef6e 4070BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
c9a0f3b8 4071
a7658e1a
AS		 4072const struct bpf_func_proto bpf_skb_output_proto = {
4073 .func = bpf_skb_event_output,
4074 .gpl_only = true,
4075 .ret_type = RET_INTEGER,
4076 .arg1_type = ARG_PTR_TO_BTF_ID,
9436ef6e 4077 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
a7658e1a
AS		 4078 .arg2_type = ARG_CONST_MAP_PTR,
4079 .arg3_type = ARG_ANYTHING,
4080 .arg4_type = ARG_PTR_TO_MEM,
4081 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
a7658e1a
AS		 4082};
4083
c6c33454
DB		 4084static unsigned short bpf_tunnel_key_af(u64 flags)
4085{
4086 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4087}
4088
f3694e00
DB		 4089BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4090 u32, size, u64, flags)
d3aa45ce 4091{
c6c33454
DB		 4092 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4093 u8 compat[sizeof(struct bpf_tunnel_key)];
074f528e
DB		 4094 void *to_orig = to;
4095 int err;
d3aa45ce 4096
074f528e
DB		 4097 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4098 err = -EINVAL;
4099 goto err_clear;
4100 }
4101 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4102 err = -EPROTO;
4103 goto err_clear;
4104 }
c6c33454 4105 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
074f528e 4106 err = -EINVAL;
c6c33454 4107 switch (size) {
4018ab18 4108 case offsetof(struct bpf_tunnel_key, tunnel_label):
c0e760c9 4109 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4018ab18 4110 goto set_compat;
c6c33454
DB		 4111 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4112 /* Fixup deprecated structure layouts here, so we have
4113 * a common path later on.
4114 */
4115 if (ip_tunnel_info_af(info) != AF_INET)
074f528e 4116 goto err_clear;
4018ab18 4117set_compat:
c6c33454
DB		 4118 to = (struct bpf_tunnel_key *)compat;
4119 break;
4120 default:
074f528e 4121 goto err_clear;
c6c33454
DB		 4122 }
4123 }
d3aa45ce
AS		 4124
4125 to->tunnel_id = be64_to_cpu(info->key.tun_id);
c6c33454
DB		 4126 to->tunnel_tos = info->key.tos;
4127 to->tunnel_ttl = info->key.ttl;
1fbc2e0c 4128 to->tunnel_ext = 0;
c6c33454 4129
4018ab18 4130 if (flags & BPF_F_TUNINFO_IPV6) {
c6c33454
DB		 4131 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4132 sizeof(to->remote_ipv6));
4018ab18
DB		 4133 to->tunnel_label = be32_to_cpu(info->key.label);
4134 } else {
c6c33454 4135 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
1fbc2e0c
DB		 4136 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4137 to->tunnel_label = 0;
4018ab18 4138 }
c6c33454
DB		 4139
4140 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
074f528e 4141 memcpy(to_orig, to, size);
d3aa45ce
AS		 4142
4143 return 0;
074f528e
DB		 4144err_clear:
4145 memset(to_orig, 0, size);
4146 return err;
d3aa45ce
AS		 4147}
4148
577c50aa 4149static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
d3aa45ce
AS		 4150 .func = bpf_skb_get_tunnel_key,
4151 .gpl_only = false,
4152 .ret_type = RET_INTEGER,
4153 .arg1_type = ARG_PTR_TO_CTX,
39f19ebb
AS		 4154 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4155 .arg3_type = ARG_CONST_SIZE,
d3aa45ce
AS		 4156 .arg4_type = ARG_ANYTHING,
4157};
4158
f3694e00 4159BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
14ca0751 4160{
14ca0751 4161 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
074f528e 4162 int err;
14ca0751
DB		 4163
4164 if (unlikely(!info ||
074f528e
DB		 4165 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4166 err = -ENOENT;
4167 goto err_clear;
4168 }
4169 if (unlikely(size < info->options_len)) {
4170 err = -ENOMEM;
4171 goto err_clear;
4172 }
14ca0751
DB		 4173
4174 ip_tunnel_info_opts_get(to, info);
074f528e
DB		 4175 if (size > info->options_len)
4176 memset(to + info->options_len, 0, size - info->options_len);
14ca0751
DB		 4177
4178 return info->options_len;
074f528e
DB		 4179err_clear:
4180 memset(to, 0, size);
4181 return err;
14ca0751
DB		 4182}
4183
4184static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4185 .func = bpf_skb_get_tunnel_opt,
4186 .gpl_only = false,
4187 .ret_type = RET_INTEGER,
4188 .arg1_type = ARG_PTR_TO_CTX,
39f19ebb
AS		 4189 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4190 .arg3_type = ARG_CONST_SIZE,
14ca0751
DB		 4191};
4192
d3aa45ce
AS		 4193static struct metadata_dst __percpu *md_dst;
4194
f3694e00
DB		 4195BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4196 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
d3aa45ce 4197{
d3aa45ce 4198 struct metadata_dst *md = this_cpu_ptr(md_dst);
c6c33454 4199 u8 compat[sizeof(struct bpf_tunnel_key)];
d3aa45ce
AS		 4200 struct ip_tunnel_info *info;
4201
22080870 4202 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
77a5196a 4203 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
d3aa45ce 4204 return -EINVAL;
c6c33454
DB		 4205 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4206 switch (size) {
4018ab18 4207 case offsetof(struct bpf_tunnel_key, tunnel_label):
c0e760c9 4208 case offsetof(struct bpf_tunnel_key, tunnel_ext):
c6c33454
DB		 4209 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4210 /* Fixup deprecated structure layouts here, so we have
4211 * a common path later on.
4212 */
4213 memcpy(compat, from, size);
4214 memset(compat + size, 0, sizeof(compat) - size);
f3694e00 4215 from = (const struct bpf_tunnel_key *) compat;
c6c33454
DB		 4216 break;
4217 default:
4218 return -EINVAL;
4219 }
4220 }
c0e760c9
DB		 4221 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4222 from->tunnel_ext))
4018ab18 4223 return -EINVAL;
d3aa45ce
AS		 4224
4225 skb_dst_drop(skb);
4226 dst_hold((struct dst_entry *) md);
4227 skb_dst_set(skb, (struct dst_entry *) md);
4228
4229 info = &md->u.tun_info;
5540fbf4 4230 memset(info, 0, sizeof(*info));
d3aa45ce 4231 info->mode = IP_TUNNEL_INFO_TX;
c6c33454 4232
db3c6139 4233 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
22080870
DB		 4234 if (flags & BPF_F_DONT_FRAGMENT)
4235 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
792f3dd6
WT		 4236 if (flags & BPF_F_ZERO_CSUM_TX)
4237 info->key.tun_flags &= ~TUNNEL_CSUM;
77a5196a
WT		 4238 if (flags & BPF_F_SEQ_NUMBER)
4239 info->key.tun_flags |= TUNNEL_SEQ;
22080870 4240
d3aa45ce 4241 info->key.tun_id = cpu_to_be64(from->tunnel_id);
c6c33454
DB		 4242 info->key.tos = from->tunnel_tos;
4243 info->key.ttl = from->tunnel_ttl;
4244
4245 if (flags & BPF_F_TUNINFO_IPV6) {
4246 info->mode |= IP_TUNNEL_INFO_IPV6;
4247 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4248 sizeof(from->remote_ipv6));
4018ab18
DB		 4249 info->key.label = cpu_to_be32(from->tunnel_label) &
4250 IPV6_FLOWLABEL_MASK;
c6c33454
DB		 4251 } else {
4252 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4253 }
d3aa45ce
AS		 4254
4255 return 0;
4256}
4257
577c50aa 4258static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
d3aa45ce
AS		 4259 .func = bpf_skb_set_tunnel_key,
4260 .gpl_only = false,
4261 .ret_type = RET_INTEGER,
4262 .arg1_type = ARG_PTR_TO_CTX,
39f19ebb
AS		 4263 .arg2_type = ARG_PTR_TO_MEM,
4264 .arg3_type = ARG_CONST_SIZE,
d3aa45ce
AS		 4265 .arg4_type = ARG_ANYTHING,
4266};
4267
f3694e00
DB		 4268BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4269 const u8 *, from, u32, size)
14ca0751 4270{
14ca0751
DB		 4271 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4272 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4273
4274 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4275 return -EINVAL;
fca5fdf6 4276 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
14ca0751
DB		 4277 return -ENOMEM;
4278
256c87c1 4279 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
14ca0751
DB		 4280
4281 return 0;
4282}
4283
4284static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4285 .func = bpf_skb_set_tunnel_opt,
4286 .gpl_only = false,
4287 .ret_type = RET_INTEGER,
4288 .arg1_type = ARG_PTR_TO_CTX,
39f19ebb
AS		 4289 .arg2_type = ARG_PTR_TO_MEM,
4290 .arg3_type = ARG_CONST_SIZE,
14ca0751
DB		 4291};
4292
4293static const struct bpf_func_proto *
4294bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
d3aa45ce
AS		 4295{
4296 if (!md_dst) {
d66f2b91
JK		 4297 struct metadata_dst __percpu *tmp;
4298
4299 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4300 METADATA_IP_TUNNEL,
4301 GFP_KERNEL);
4302 if (!tmp)
d3aa45ce 4303 return NULL;
d66f2b91
JK		 4304 if (cmpxchg(&md_dst, NULL, tmp))
4305 metadata_dst_free_percpu(tmp);
d3aa45ce 4306 }
14ca0751
DB		 4307
4308 switch (which) {
4309 case BPF_FUNC_skb_set_tunnel_key:
4310 return &bpf_skb_set_tunnel_key_proto;
4311 case BPF_FUNC_skb_set_tunnel_opt:
4312 return &bpf_skb_set_tunnel_opt_proto;
4313 default:
4314 return NULL;
4315 }
d3aa45ce
AS		 4316}
4317
f3694e00
DB		 4318BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4319 u32, idx)
4a482f34 4320{
4a482f34
MKL		 4321 struct bpf_array *array = container_of(map, struct bpf_array, map);
4322 struct cgroup *cgrp;
4323 struct sock *sk;
4a482f34 4324
2d48c5f9 4325 sk = skb_to_full_sk(skb);
4a482f34
MKL		 4326 if (!sk || !sk_fullsock(sk))
4327 return -ENOENT;
f3694e00 4328 if (unlikely(idx >= array->map.max_entries))
4a482f34
MKL		 4329 return -E2BIG;
4330
f3694e00 4331 cgrp = READ_ONCE(array->ptrs[idx]);
4a482f34
MKL		 4332 if (unlikely(!cgrp))
4333 return -EAGAIN;
4334
54fd9c2d 4335 return sk_under_cgroup_hierarchy(sk, cgrp);
4a482f34
MKL		 4336}
4337
747ea55e
DB		 4338static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4339 .func = bpf_skb_under_cgroup,
4a482f34
MKL		 4340 .gpl_only = false,
4341 .ret_type = RET_INTEGER,
4342 .arg1_type = ARG_PTR_TO_CTX,
4343 .arg2_type = ARG_CONST_MAP_PTR,
4344 .arg3_type = ARG_ANYTHING,
4345};
4a482f34 4346
cb20b08e 4347#ifdef CONFIG_SOCK_CGROUP_DATA
f307fa2c
AI		 4348static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4349{
4350 struct cgroup *cgrp;
4351
a5fa25ad
MKL		 4352 sk = sk_to_full_sk(sk);
4353 if (!sk || !sk_fullsock(sk))
4354 return 0;
4355
f307fa2c
AI		 4356 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4357 return cgroup_id(cgrp);
4358}
4359
cb20b08e
DB		 4360BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4361{
a5fa25ad 4362 return __bpf_sk_cgroup_id(skb->sk);
cb20b08e
DB		 4363}
4364
4365static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4366 .func = bpf_skb_cgroup_id,
4367 .gpl_only = false,
4368 .ret_type = RET_INTEGER,
4369 .arg1_type = ARG_PTR_TO_CTX,
4370};
77236281 4371
f307fa2c
AI		 4372static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4373 int ancestor_level)
77236281 4374{
77236281
AI		 4375 struct cgroup *ancestor;
4376 struct cgroup *cgrp;
4377
a5fa25ad
MKL		 4378 sk = sk_to_full_sk(sk);
4379 if (!sk || !sk_fullsock(sk))
4380 return 0;
4381
77236281
AI		 4382 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4383 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4384 if (!ancestor)
4385 return 0;
4386
74321038 4387 return cgroup_id(ancestor);
77236281
AI		 4388}
4389
f307fa2c
AI		 4390BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4391 ancestor_level)
4392{
a5fa25ad 4393 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
f307fa2c
AI		 4394}
4395
77236281
AI		 4396static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4397 .func = bpf_skb_ancestor_cgroup_id,
4398 .gpl_only = false,
4399 .ret_type = RET_INTEGER,
4400 .arg1_type = ARG_PTR_TO_CTX,
4401 .arg2_type = ARG_ANYTHING,
4402};
f307fa2c
AI		 4403
4404BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4405{
4406 return __bpf_sk_cgroup_id(sk);
4407}
4408
4409static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4410 .func = bpf_sk_cgroup_id,
4411 .gpl_only = false,
4412 .ret_type = RET_INTEGER,
a5fa25ad 4413 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
f307fa2c
AI		 4414};
4415
4416BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4417{
4418 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4419}
4420
4421static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4422 .func = bpf_sk_ancestor_cgroup_id,
4423 .gpl_only = false,
4424 .ret_type = RET_INTEGER,
a5fa25ad 4425 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
f307fa2c
AI		 4426 .arg2_type = ARG_ANYTHING,
4427};
cb20b08e
DB		 4428#endif
4429
4de16969
DB		 4430static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4431 unsigned long off, unsigned long len)
4432{
4433 memcpy(dst_buff, src_buff + off, len);
4434 return 0;
4435}
4436
f3694e00
DB		 4437BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4438 u64, flags, void *, meta, u64, meta_size)
4de16969 4439{
4de16969 4440 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4de16969
DB		 4441
4442 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4443 return -EINVAL;
d831ee84
EC		 4444 if (unlikely(!xdp ||
4445 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4de16969
DB		 4446 return -EFAULT;
4447
9c471370
MKL		 4448 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4449 xdp_size, bpf_xdp_copy);
4de16969
DB		 4450}
4451
4452static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4453 .func = bpf_xdp_event_output,
4454 .gpl_only = true,
4455 .ret_type = RET_INTEGER,
4456 .arg1_type = ARG_PTR_TO_CTX,
4457 .arg2_type = ARG_CONST_MAP_PTR,
4458 .arg3_type = ARG_ANYTHING,
39f19ebb 4459 .arg4_type = ARG_PTR_TO_MEM,
1728a4f2 4460 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4de16969
DB		 4461};
4462
9436ef6e 4463BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
c9a0f3b8 4464
d831ee84
EC		 4465const struct bpf_func_proto bpf_xdp_output_proto = {
4466 .func = bpf_xdp_event_output,
4467 .gpl_only = true,
4468 .ret_type = RET_INTEGER,
4469 .arg1_type = ARG_PTR_TO_BTF_ID,
9436ef6e 4470 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
d831ee84
EC		 4471 .arg2_type = ARG_CONST_MAP_PTR,
4472 .arg3_type = ARG_ANYTHING,
4473 .arg4_type = ARG_PTR_TO_MEM,
4474 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
d831ee84
EC		 4475};
4476
91b8270f
CF		 4477BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4478{
92acdc58 4479 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
91b8270f
CF		 4480}
4481
4482static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4483 .func = bpf_get_socket_cookie,
4484 .gpl_only = false,
4485 .ret_type = RET_INTEGER,
4486 .arg1_type = ARG_PTR_TO_CTX,
4487};
4488
d692f113
AI		 4489BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4490{
92acdc58 4491 return __sock_gen_cookie(ctx->sk);
d692f113
AI		 4492}
4493
4494static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4495 .func = bpf_get_socket_cookie_sock_addr,
4496 .gpl_only = false,
4497 .ret_type = RET_INTEGER,
4498 .arg1_type = ARG_PTR_TO_CTX,
4499};
4500
0e53d9e5
DB		 4501BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4502{
92acdc58 4503 return __sock_gen_cookie(ctx);
0e53d9e5
DB		 4504}
4505
4506static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4507 .func = bpf_get_socket_cookie_sock,
4508 .gpl_only = false,
4509 .ret_type = RET_INTEGER,
4510 .arg1_type = ARG_PTR_TO_CTX,
4511};
4512
d692f113
AI		 4513BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4514{
92acdc58 4515 return __sock_gen_cookie(ctx->sk);
d692f113
AI		 4516}
4517
4518static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4519 .func = bpf_get_socket_cookie_sock_ops,
4520 .gpl_only = false,
4521 .ret_type = RET_INTEGER,
4522 .arg1_type = ARG_PTR_TO_CTX,
4523};
4524
f318903c
DB		 4525static u64 __bpf_get_netns_cookie(struct sock *sk)
4526{
4527#ifdef CONFIG_NET_NS
92acdc58 4528 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
f318903c
DB		 4529#else
4530 return 0;
4531#endif
4532}
4533
4534BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4535{
4536 return __bpf_get_netns_cookie(ctx);
4537}
4538
4539static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4540 .func = bpf_get_netns_cookie_sock,
4541 .gpl_only = false,
4542 .ret_type = RET_INTEGER,
4543 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4544};
4545
4546BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4547{
4548 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4549}
4550
4551static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4552 .func = bpf_get_netns_cookie_sock_addr,
4553 .gpl_only = false,
4554 .ret_type = RET_INTEGER,
4555 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4556};
4557
6acc5c29
CF		 4558BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4559{
4560 struct sock *sk = sk_to_full_sk(skb->sk);
4561 kuid_t kuid;
4562
4563 if (!sk || !sk_fullsock(sk))
4564 return overflowuid;
4565 kuid = sock_net_uid(sock_net(sk), sk);
4566 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4567}
4568
4569static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4570 .func = bpf_get_socket_uid,
4571 .gpl_only = false,
4572 .ret_type = RET_INTEGER,
4573 .arg1_type = ARG_PTR_TO_CTX,
4574};
4575
beecf11b 4576static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5cdc744c 4577 char *optval, int optlen)
8c4b4c7e 4578{
70c58997 4579 char devname[IFNAMSIZ];
f9bcf968 4580 int val, valbool;
70c58997
FF		 4581 struct net *net;
4582 int ifindex;
8c4b4c7e 4583 int ret = 0;
8c4b4c7e
LB		 4584
4585 if (!sk_fullsock(sk))
4586 return -EINVAL;
4587
beecf11b
SF		 4588 sock_owned_by_me(sk);
4589
8c4b4c7e 4590 if (level == SOL_SOCKET) {
70c58997 4591 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
8c4b4c7e
LB		 4592 return -EINVAL;
4593 val = *((int *)optval);
f9bcf968 4594 valbool = val ? 1 : 0;
8c4b4c7e
LB		 4595
4596 /* Only some socketops are supported */
4597 switch (optname) {
4598 case SO_RCVBUF:
c9e45767 4599 val = min_t(u32, val, sysctl_rmem_max);
8c4b4c7e 4600 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
ebb3b78d
ED		 4601 WRITE_ONCE(sk->sk_rcvbuf,
4602 max_t(int, val * 2, SOCK_MIN_RCVBUF));
8c4b4c7e
LB		 4603 break;
4604 case SO_SNDBUF:
c9e45767 4605 val = min_t(u32, val, sysctl_wmem_max);
8c4b4c7e 4606 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
e292f05e
ED		 4607 WRITE_ONCE(sk->sk_sndbuf,
4608 max_t(int, val * 2, SOCK_MIN_SNDBUF));
8c4b4c7e 4609 break;
76a9ebe8 4610 case SO_MAX_PACING_RATE: /* 32bit version */
e224c390
YC		 4611 if (val != ~0U)
4612 cmpxchg(&sk->sk_pacing_status,
4613 SK_PACING_NONE,
4614 SK_PACING_NEEDED);
76a9ebe8 4615 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
8c4b4c7e
LB		 4616 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4617 sk->sk_max_pacing_rate);
4618 break;
4619 case SO_PRIORITY:
4620 sk->sk_priority = val;
4621 break;
4622 case SO_RCVLOWAT:
4623 if (val < 0)
4624 val = INT_MAX;
eac66402 4625 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
8c4b4c7e
LB		 4626 break;
4627 case SO_MARK:
f4924f24
PO		 4628 if (sk->sk_mark != val) {
4629 sk->sk_mark = val;
4630 sk_dst_reset(sk);
4631 }
8c4b4c7e 4632 break;
70c58997 4633 case SO_BINDTODEVICE:
70c58997
FF		 4634 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4635 strncpy(devname, optval, optlen);
4636 devname[optlen] = 0;
4637
4638 ifindex = 0;
4639 if (devname[0] != '\0') {
4640 struct net_device *dev;
4641
4642 ret = -ENODEV;
4643
4644 net = sock_net(sk);
4645 dev = dev_get_by_name(net, devname);
4646 if (!dev)
4647 break;
4648 ifindex = dev->ifindex;
4649 dev_put(dev);
4650 }
4651 ret = sock_bindtoindex(sk, ifindex, false);
70c58997 4652 break;
f9bcf968
DY		 4653 case SO_KEEPALIVE:
4654 if (sk->sk_prot->keepalive)
4655 sk->sk_prot->keepalive(sk, valbool);
4656 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4657 break;
8c4b4c7e
LB		 4658 default:
4659 ret = -EINVAL;
4660 }
a5192c52 4661#ifdef CONFIG_INET
6f5c39fa
NS		 4662 } else if (level == SOL_IP) {
4663 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4664 return -EINVAL;
4665
4666 val = *((int *)optval);
4667 /* Only some options are supported */
4668 switch (optname) {
4669 case IP_TOS:
4670 if (val < -1 || val > 0xff) {
4671 ret = -EINVAL;
4672 } else {
4673 struct inet_sock *inet = inet_sk(sk);
4674
4675 if (val == -1)
4676 val = 0;
4677 inet->tos = val;
4678 }
4679 break;
4680 default:
4681 ret = -EINVAL;
4682 }
6f9bd3d7
LB		 4683#if IS_ENABLED(CONFIG_IPV6)
4684 } else if (level == SOL_IPV6) {
4685 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4686 return -EINVAL;
4687
4688 val = *((int *)optval);
4689 /* Only some options are supported */
4690 switch (optname) {
4691 case IPV6_TCLASS:
4692 if (val < -1 || val > 0xff) {
4693 ret = -EINVAL;
4694 } else {
4695 struct ipv6_pinfo *np = inet6_sk(sk);
4696
4697 if (val == -1)
4698 val = 0;
4699 np->tclass = val;
4700 }
4701 break;
4702 default:
4703 ret = -EINVAL;
4704 }
4705#endif
8c4b4c7e
LB		 4706 } else if (level == SOL_TCP &&
4707 sk->sk_prot->setsockopt == tcp_setsockopt) {
91b5b21c
LB		 4708 if (optname == TCP_CONGESTION) {
4709 char name[TCP_CA_NAME_MAX];
4710
4711 strncpy(name, optval, min_t(long, optlen,
4712 TCP_CA_NAME_MAX-1));
4713 name[TCP_CA_NAME_MAX-1] = 0;
29a94932 4714 ret = tcp_set_congestion_control(sk, name, false, true);
91b5b21c 4715 } else {
f9bcf968 4716 struct inet_connection_sock *icsk = inet_csk(sk);
fc747810 4717 struct tcp_sock *tp = tcp_sk(sk);
2b8ee4f0 4718 unsigned long timeout;
fc747810
LB		 4719
4720 if (optlen != sizeof(int))
4721 return -EINVAL;
4722
4723 val = *((int *)optval);
4724 /* Only some options are supported */
4725 switch (optname) {
4726 case TCP_BPF_IW:
31aa6503 4727 if (val <= 0 || tp->data_segs_out > tp->syn_data)
fc747810
LB		 4728 ret = -EINVAL;
4729 else
4730 tp->snd_cwnd = val;
4731 break;
13bf9641
LB		 4732 case TCP_BPF_SNDCWND_CLAMP:
4733 if (val <= 0) {
4734 ret = -EINVAL;
4735 } else {
4736 tp->snd_cwnd_clamp = val;
4737 tp->snd_ssthresh = val;
4738 }
6d3f06a0 4739 break;
2b8ee4f0
MKL		 4740 case TCP_BPF_DELACK_MAX:
4741 timeout = usecs_to_jiffies(val);
4742 if (timeout > TCP_DELACK_MAX ||
4743 timeout < TCP_TIMEOUT_MIN)
4744 return -EINVAL;
4745 inet_csk(sk)->icsk_delack_max = timeout;
4746 break;
ca584ba0
MKL		 4747 case TCP_BPF_RTO_MIN:
4748 timeout = usecs_to_jiffies(val);
4749 if (timeout > TCP_RTO_MIN ||
4750 timeout < TCP_TIMEOUT_MIN)
4751 return -EINVAL;
4752 inet_csk(sk)->icsk_rto_min = timeout;
4753 break;
1e215300
NS		 4754 case TCP_SAVE_SYN:
4755 if (val < 0 || val > 1)
4756 ret = -EINVAL;
4757 else
4758 tp->save_syn = val;
4759 break;
f9bcf968
DY		 4760 case TCP_KEEPIDLE:
4761 ret = tcp_sock_set_keepidle_locked(sk, val);
4762 break;
4763 case TCP_KEEPINTVL:
4764 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4765 ret = -EINVAL;
4766 else
4767 tp->keepalive_intvl = val * HZ;
4768 break;
4769 case TCP_KEEPCNT:
4770 if (val < 1 || val > MAX_TCP_KEEPCNT)
4771 ret = -EINVAL;
4772 else
4773 tp->keepalive_probes = val;
4774 break;
4775 case TCP_SYNCNT:
4776 if (val < 1 || val > MAX_TCP_SYNCNT)
4777 ret = -EINVAL;
4778 else
4779 icsk->icsk_syn_retries = val;
4780 break;
4781 case TCP_USER_TIMEOUT:
4782 if (val < 0)
4783 ret = -EINVAL;
4784 else
4785 icsk->icsk_user_timeout = val;
4786 break;
fc747810
LB		 4787 default:
4788 ret = -EINVAL;
4789 }
91b5b21c 4790 }
91b5b21c 4791#endif
8c4b4c7e
LB		 4792 } else {
4793 ret = -EINVAL;
4794 }
4795 return ret;
4796}
4797
beecf11b
SF		 4798static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4799 char *optval, int optlen)
cd86d1fd 4800{
cd86d1fd
LB		 4801 if (!sk_fullsock(sk))
4802 goto err_clear;
beecf11b
SF		 4803
4804 sock_owned_by_me(sk);
4805
cd86d1fd
LB		 4806#ifdef CONFIG_INET
4807 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
1edb6e03
AR		 4808 struct inet_connection_sock *icsk;
4809 struct tcp_sock *tp;
4810
1e215300
NS		 4811 switch (optname) {
4812 case TCP_CONGESTION:
4813 icsk = inet_csk(sk);
cd86d1fd
LB		 4814
4815 if (!icsk->icsk_ca_ops || optlen <= 1)
4816 goto err_clear;
4817 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4818 optval[optlen - 1] = 0;
1e215300
NS		 4819 break;
4820 case TCP_SAVED_SYN:
4821 tp = tcp_sk(sk);
4822
4823 if (optlen <= 0 || !tp->saved_syn ||
70a217f1 4824 optlen > tcp_saved_syn_len(tp->saved_syn))
1e215300 4825 goto err_clear;
70a217f1 4826 memcpy(optval, tp->saved_syn->data, optlen);
1e215300
NS		 4827 break;
4828 default:
cd86d1fd
LB		 4829 goto err_clear;
4830 }
6f5c39fa
NS		 4831 } else if (level == SOL_IP) {
4832 struct inet_sock *inet = inet_sk(sk);
4833
4834 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4835 goto err_clear;
4836
4837 /* Only some options are supported */
4838 switch (optname) {
4839 case IP_TOS:
4840 *((int *)optval) = (int)inet->tos;
4841 break;
4842 default:
4843 goto err_clear;
4844 }
6f9bd3d7
LB		 4845#if IS_ENABLED(CONFIG_IPV6)
4846 } else if (level == SOL_IPV6) {
4847 struct ipv6_pinfo *np = inet6_sk(sk);
4848
4849 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4850 goto err_clear;
4851
4852 /* Only some options are supported */
4853 switch (optname) {
4854 case IPV6_TCLASS:
4855 *((int *)optval) = (int)np->tclass;
4856 break;
4857 default:
4858 goto err_clear;
4859 }
4860#endif
cd86d1fd
LB		 4861 } else {
4862 goto err_clear;
4863 }
aa2bc739 4864 return 0;
cd86d1fd
LB		 4865#endif
4866err_clear:
4867 memset(optval, 0, optlen);
4868 return -EINVAL;
4869}
4870
beecf11b
SF		 4871BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4872 int, level, int, optname, char *, optval, int, optlen)
4873{
5cdc744c 4874 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
beecf11b
SF		 4875}
4876
4877static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
4878 .func = bpf_sock_addr_setsockopt,
4879 .gpl_only = false,
4880 .ret_type = RET_INTEGER,
4881 .arg1_type = ARG_PTR_TO_CTX,
4882 .arg2_type = ARG_ANYTHING,
4883 .arg3_type = ARG_ANYTHING,
4884 .arg4_type = ARG_PTR_TO_MEM,
4885 .arg5_type = ARG_CONST_SIZE,
4886};
4887
4888BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
4889 int, level, int, optname, char *, optval, int, optlen)
4890{
4891 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
4892}
4893
4894static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
4895 .func = bpf_sock_addr_getsockopt,
4896 .gpl_only = false,
4897 .ret_type = RET_INTEGER,
4898 .arg1_type = ARG_PTR_TO_CTX,
4899 .arg2_type = ARG_ANYTHING,
4900 .arg3_type = ARG_ANYTHING,
4901 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4902 .arg5_type = ARG_CONST_SIZE,
4903};
4904
4905BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4906 int, level, int, optname, char *, optval, int, optlen)
4907{
5cdc744c 4908 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
beecf11b
SF		 4909}
4910
4911static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
4912 .func = bpf_sock_ops_setsockopt,
4913 .gpl_only = false,
4914 .ret_type = RET_INTEGER,
4915 .arg1_type = ARG_PTR_TO_CTX,
4916 .arg2_type = ARG_ANYTHING,
4917 .arg3_type = ARG_ANYTHING,
4918 .arg4_type = ARG_PTR_TO_MEM,
4919 .arg5_type = ARG_CONST_SIZE,
4920};
4921
0813a841
MKL		 4922static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
4923 int optname, const u8 **start)
4924{
4925 struct sk_buff *syn_skb = bpf_sock->syn_skb;
4926 const u8 *hdr_start;
4927 int ret;
4928
4929 if (syn_skb) {
4930 /* sk is a request_sock here */
4931
4932 if (optname == TCP_BPF_SYN) {
4933 hdr_start = syn_skb->data;
4934 ret = tcp_hdrlen(syn_skb);
267cf9fa 4935 } else if (optname == TCP_BPF_SYN_IP) {
0813a841
MKL		 4936 hdr_start = skb_network_header(syn_skb);
4937 ret = skb_network_header_len(syn_skb) +
4938 tcp_hdrlen(syn_skb);
267cf9fa
MKL		 4939 } else {
4940 /* optname == TCP_BPF_SYN_MAC */
4941 hdr_start = skb_mac_header(syn_skb);
4942 ret = skb_mac_header_len(syn_skb) +
4943 skb_network_header_len(syn_skb) +
4944 tcp_hdrlen(syn_skb);
0813a841
MKL		 4945 }
4946 } else {
4947 struct sock *sk = bpf_sock->sk;
4948 struct saved_syn *saved_syn;
4949
4950 if (sk->sk_state == TCP_NEW_SYN_RECV)
4951 /* synack retransmit. bpf_sock->syn_skb will
4952 * not be available. It has to resort to
4953 * saved_syn (if it is saved).
4954 */
4955 saved_syn = inet_reqsk(sk)->saved_syn;
4956 else
4957 saved_syn = tcp_sk(sk)->saved_syn;
4958
4959 if (!saved_syn)
4960 return -ENOENT;
4961
4962 if (optname == TCP_BPF_SYN) {
4963 hdr_start = saved_syn->data +
267cf9fa 4964 saved_syn->mac_hdrlen +
0813a841
MKL		 4965 saved_syn->network_hdrlen;
4966 ret = saved_syn->tcp_hdrlen;
267cf9fa
MKL		 4967 } else if (optname == TCP_BPF_SYN_IP) {
4968 hdr_start = saved_syn->data +
4969 saved_syn->mac_hdrlen;
4970 ret = saved_syn->network_hdrlen +
4971 saved_syn->tcp_hdrlen;
0813a841 4972 } else {
267cf9fa
MKL		 4973 /* optname == TCP_BPF_SYN_MAC */
4974
4975 /* TCP_SAVE_SYN may not have saved the mac hdr */
4976 if (!saved_syn->mac_hdrlen)
4977 return -ENOENT;
4978
0813a841 4979 hdr_start = saved_syn->data;
267cf9fa
MKL		 4980 ret = saved_syn->mac_hdrlen +
4981 saved_syn->network_hdrlen +
0813a841
MKL		 4982 saved_syn->tcp_hdrlen;
4983 }
4984 }
4985
4986 *start = hdr_start;
4987 return ret;
4988}
4989
beecf11b
SF		 4990BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4991 int, level, int, optname, char *, optval, int, optlen)
4992{
0813a841 4993 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
267cf9fa 4994 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
0813a841
MKL		 4995 int ret, copy_len = 0;
4996 const u8 *start;
4997
4998 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
4999 if (ret > 0) {
5000 copy_len = ret;
5001 if (optlen < copy_len) {
5002 copy_len = optlen;
5003 ret = -ENOSPC;
5004 }
5005
5006 memcpy(optval, start, copy_len);
5007 }
5008
5009 /* Zero out unused buffer at the end */
5010 memset(optval + copy_len, 0, optlen - copy_len);
5011
5012 return ret;
5013 }
5014
beecf11b
SF		 5015 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5016}
5017
5018static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5019 .func = bpf_sock_ops_getsockopt,
cd86d1fd
LB		 5020 .gpl_only = false,
5021 .ret_type = RET_INTEGER,
5022 .arg1_type = ARG_PTR_TO_CTX,
5023 .arg2_type = ARG_ANYTHING,
5024 .arg3_type = ARG_ANYTHING,
5025 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5026 .arg5_type = ARG_CONST_SIZE,
5027};
5028
b13d8807
LB		 5029BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5030 int, argval)
5031{
5032 struct sock *sk = bpf_sock->sk;
5033 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5034
a7dcdf6e 5035 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
b13d8807
LB		 5036 return -EINVAL;
5037
725721a6 5038 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
b13d8807
LB		 5039
5040 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
b13d8807
LB		 5041}
5042
5043static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5044 .func = bpf_sock_ops_cb_flags_set,
5045 .gpl_only = false,
5046 .ret_type = RET_INTEGER,
5047 .arg1_type = ARG_PTR_TO_CTX,
5048 .arg2_type = ARG_ANYTHING,
5049};
5050
d74bad4e
AI		 5051const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5052EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5053
5054BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5055 int, addr_len)
5056{
5057#ifdef CONFIG_INET
5058 struct sock *sk = ctx->sk;
8086fbaf 5059 u32 flags = BIND_FROM_BPF;
d74bad4e
AI		 5060 int err;
5061
d74bad4e 5062 err = -EINVAL;
ba024f25
TH		 5063 if (addr_len < offsetofend(struct sockaddr, sa_family))
5064 return err;
d74bad4e
AI		 5065 if (addr->sa_family == AF_INET) {
5066 if (addr_len < sizeof(struct sockaddr_in))
5067 return err;
8086fbaf
SF		 5068 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5069 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5070 return __inet_bind(sk, addr, addr_len, flags);
d74bad4e
AI		 5071#if IS_ENABLED(CONFIG_IPV6)
5072 } else if (addr->sa_family == AF_INET6) {
5073 if (addr_len < SIN6_LEN_RFC2133)
5074 return err;
8086fbaf
SF		 5075 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5076 flags |= BIND_FORCE_ADDRESS_NO_PORT;
d74bad4e
AI		 5077 /* ipv6_bpf_stub cannot be NULL, since it's called from
5078 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5079 */
8086fbaf 5080 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
d74bad4e
AI		 5081#endif /* CONFIG_IPV6 */
5082 }
5083#endif /* CONFIG_INET */
5084
5085 return -EAFNOSUPPORT;
5086}
5087
5088static const struct bpf_func_proto bpf_bind_proto = {
5089 .func = bpf_bind,
5090 .gpl_only = false,
5091 .ret_type = RET_INTEGER,
5092 .arg1_type = ARG_PTR_TO_CTX,
5093 .arg2_type = ARG_PTR_TO_MEM,
5094 .arg3_type = ARG_CONST_SIZE,
5095};
5096
12bed760
EB		 5097#ifdef CONFIG_XFRM
5098BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5099 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5100{
5101 const struct sec_path *sp = skb_sec_path(skb);
5102 const struct xfrm_state *x;
5103
5104 if (!sp || unlikely(index >= sp->len || flags))
5105 goto err_clear;
5106
5107 x = sp->xvec[index];
5108
5109 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5110 goto err_clear;
5111
5112 to->reqid = x->props.reqid;
5113 to->spi = x->id.spi;
5114 to->family = x->props.family;
1fbc2e0c
DB		 5115 to->ext = 0;
5116
12bed760
EB		 5117 if (to->family == AF_INET6) {
5118 memcpy(to->remote_ipv6, x->props.saddr.a6,
5119 sizeof(to->remote_ipv6));
5120 } else {
5121 to->remote_ipv4 = x->props.saddr.a4;
1fbc2e0c 5122 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
12bed760
EB		 5123 }
5124
5125 return 0;
5126err_clear:
5127 memset(to, 0, size);
5128 return -EINVAL;
5129}
5130
5131static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5132 .func = bpf_skb_get_xfrm_state,
5133 .gpl_only = false,
5134 .ret_type = RET_INTEGER,
5135 .arg1_type = ARG_PTR_TO_CTX,
5136 .arg2_type = ARG_ANYTHING,
5137 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5138 .arg4_type = ARG_CONST_SIZE,
5139 .arg5_type = ARG_ANYTHING,
5140};
5141#endif
5142
87f5fc7e
DA		 5143#if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5144static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5145 const struct neighbour *neigh,
5146 const struct net_device *dev)
5147{
5148 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5149 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5150 params->h_vlan_TCI = 0;
5151 params->h_vlan_proto = 0;
4c79579b 5152 params->ifindex = dev->ifindex;
87f5fc7e 5153
4c79579b 5154 return 0;
87f5fc7e
DA		 5155}
5156#endif
5157
5158#if IS_ENABLED(CONFIG_INET)
5159static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4f74fede 5160 u32 flags, bool check_mtu)
87f5fc7e 5161{
eba618ab 5162 struct fib_nh_common *nhc;
87f5fc7e
DA		 5163 struct in_device *in_dev;
5164 struct neighbour *neigh;
5165 struct net_device *dev;
5166 struct fib_result res;
87f5fc7e
DA		 5167 struct flowi4 fl4;
5168 int err;
4f74fede 5169 u32 mtu;
87f5fc7e
DA		 5170
5171 dev = dev_get_by_index_rcu(net, params->ifindex);
5172 if (unlikely(!dev))
5173 return -ENODEV;
5174
5175 /* verify forwarding is enabled on this interface */
5176 in_dev = __in_dev_get_rcu(dev);
5177 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4c79579b 5178 return BPF_FIB_LKUP_RET_FWD_DISABLED;
87f5fc7e
DA		 5179
5180 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5181 fl4.flowi4_iif = 1;
5182 fl4.flowi4_oif = params->ifindex;
5183 } else {
5184 fl4.flowi4_iif = params->ifindex;
5185 fl4.flowi4_oif = 0;
5186 }
5187 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5188 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5189 fl4.flowi4_flags = 0;
5190
5191 fl4.flowi4_proto = params->l4_protocol;
5192 fl4.daddr = params->ipv4_dst;
5193 fl4.saddr = params->ipv4_src;
5194 fl4.fl4_sport = params->sport;
5195 fl4.fl4_dport = params->dport;
1869e226 5196 fl4.flowi4_multipath_hash = 0;
87f5fc7e
DA		 5197
5198 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5199 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5200 struct fib_table *tb;
5201
5202 tb = fib_get_table(net, tbid);
5203 if (unlikely(!tb))
4c79579b 5204 return BPF_FIB_LKUP_RET_NOT_FWDED;
87f5fc7e
DA		 5205
5206 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5207 } else {
5208 fl4.flowi4_mark = 0;
5209 fl4.flowi4_secid = 0;
5210 fl4.flowi4_tun_key.tun_id = 0;
5211 fl4.flowi4_uid = sock_net_uid(net, NULL);
5212
5213 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5214 }
5215
4c79579b
DA		 5216 if (err) {
5217 /* map fib lookup errors to RTN_ type */
5218 if (err == -EINVAL)
5219 return BPF_FIB_LKUP_RET_BLACKHOLE;
5220 if (err == -EHOSTUNREACH)
5221 return BPF_FIB_LKUP_RET_UNREACHABLE;
5222 if (err == -EACCES)
5223 return BPF_FIB_LKUP_RET_PROHIBIT;
5224
5225 return BPF_FIB_LKUP_RET_NOT_FWDED;
5226 }
5227
5228 if (res.type != RTN_UNICAST)
5229 return BPF_FIB_LKUP_RET_NOT_FWDED;
87f5fc7e 5230
5481d73f 5231 if (fib_info_num_path(res.fi) > 1)
87f5fc7e
DA		 5232 fib_select_path(net, &res, &fl4, NULL);
5233
4f74fede
DA		 5234 if (check_mtu) {
5235 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5236 if (params->tot_len > mtu)
4c79579b 5237 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4f74fede
DA		 5238 }
5239
eba618ab 5240 nhc = res.nhc;
87f5fc7e
DA		 5241
5242 /* do not handle lwt encaps right now */
eba618ab 5243 if (nhc->nhc_lwtstate)
4c79579b 5244 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
87f5fc7e 5245
eba618ab 5246 dev = nhc->nhc_dev;
87f5fc7e
DA		 5247
5248 params->rt_metric = res.fi->fib_priority;
5249
5250 /* xdp and cls_bpf programs are run in RCU-bh so
5251 * rcu_read_lock_bh is not needed here
5252 */
6f5f68d0
DA		 5253 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5254 if (nhc->nhc_gw_family)
5255 params->ipv4_dst = nhc->nhc_gw.ipv4;
5256
5257 neigh = __ipv4_neigh_lookup_noref(dev,
5258 (__force u32)params->ipv4_dst);
5259 } else {
5260 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5261
5262 params->family = AF_INET6;
5263 *dst = nhc->nhc_gw.ipv6;
5264 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5265 }
5266
4c79579b
DA		 5267 if (!neigh)
5268 return BPF_FIB_LKUP_RET_NO_NEIGH;
87f5fc7e 5269
4c79579b 5270 return bpf_fib_set_fwd_params(params, neigh, dev);
87f5fc7e
DA		 5271}
5272#endif
5273
5274#if IS_ENABLED(CONFIG_IPV6)
5275static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4f74fede 5276 u32 flags, bool check_mtu)
87f5fc7e
DA		 5277{
5278 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5279 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
e55449e7 5280 struct fib6_result res = {};
87f5fc7e
DA		 5281 struct neighbour *neigh;
5282 struct net_device *dev;
5283 struct inet6_dev *idev;
87f5fc7e
DA		 5284 struct flowi6 fl6;
5285 int strict = 0;
effda4dd 5286 int oif, err;
4f74fede 5287 u32 mtu;
87f5fc7e
DA		 5288
5289 /* link local addresses are never forwarded */
5290 if (rt6_need_strict(dst) || rt6_need_strict(src))
4c79579b 5291 return BPF_FIB_LKUP_RET_NOT_FWDED;
87f5fc7e
DA		 5292
5293 dev = dev_get_by_index_rcu(net, params->ifindex);
5294 if (unlikely(!dev))
5295 return -ENODEV;
5296
5297 idev = __in6_dev_get_safely(dev);
56f0f84e 5298 if (unlikely(!idev || !idev->cnf.forwarding))
4c79579b 5299 return BPF_FIB_LKUP_RET_FWD_DISABLED;
87f5fc7e
DA		 5300
5301 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5302 fl6.flowi6_iif = 1;
5303 oif = fl6.flowi6_oif = params->ifindex;
5304 } else {
5305 oif = fl6.flowi6_iif = params->ifindex;
5306 fl6.flowi6_oif = 0;
5307 strict = RT6_LOOKUP_F_HAS_SADDR;
5308 }
bd3a08aa 5309 fl6.flowlabel = params->flowinfo;
87f5fc7e
DA		 5310 fl6.flowi6_scope = 0;
5311 fl6.flowi6_flags = 0;
5312 fl6.mp_hash = 0;
5313
5314 fl6.flowi6_proto = params->l4_protocol;
5315 fl6.daddr = *dst;
5316 fl6.saddr = *src;
5317 fl6.fl6_sport = params->sport;
5318 fl6.fl6_dport = params->dport;
5319
5320 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5321 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5322 struct fib6_table *tb;
5323
5324 tb = ipv6_stub->fib6_get_table(net, tbid);
5325 if (unlikely(!tb))
4c79579b 5326 return BPF_FIB_LKUP_RET_NOT_FWDED;
87f5fc7e 5327
effda4dd
DA		 5328 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5329 strict);
87f5fc7e
DA		 5330 } else {
5331 fl6.flowi6_mark = 0;
5332 fl6.flowi6_secid = 0;
5333 fl6.flowi6_tun_key.tun_id = 0;
5334 fl6.flowi6_uid = sock_net_uid(net, NULL);
5335
effda4dd 5336 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
87f5fc7e
DA		 5337 }
5338
effda4dd 5339 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
b1d40991 5340 res.f6i == net->ipv6.fib6_null_entry))
4c79579b
DA		 5341 return BPF_FIB_LKUP_RET_NOT_FWDED;
5342
7d21fec9
DA		 5343 switch (res.fib6_type) {
5344 /* only unicast is forwarded */
5345 case RTN_UNICAST:
5346 break;
5347 case RTN_BLACKHOLE:
5348 return BPF_FIB_LKUP_RET_BLACKHOLE;
5349 case RTN_UNREACHABLE:
5350 return BPF_FIB_LKUP_RET_UNREACHABLE;
5351 case RTN_PROHIBIT:
5352 return BPF_FIB_LKUP_RET_PROHIBIT;
5353 default:
4c79579b 5354 return BPF_FIB_LKUP_RET_NOT_FWDED;
7d21fec9 5355 }
87f5fc7e 5356
b1d40991
DA		 5357 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5358 fl6.flowi6_oif != 0, NULL, strict);
87f5fc7e 5359
4f74fede 5360 if (check_mtu) {
b748f260 5361 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4f74fede 5362 if (params->tot_len > mtu)
4c79579b 5363 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4f74fede
DA		 5364 }
5365
b1d40991 5366 if (res.nh->fib_nh_lws)
4c79579b 5367 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
87f5fc7e 5368
b1d40991
DA		 5369 if (res.nh->fib_nh_gw_family)
5370 *dst = res.nh->fib_nh_gw6;
87f5fc7e 5371
b1d40991
DA		 5372 dev = res.nh->fib_nh_dev;
5373 params->rt_metric = res.f6i->fib6_metric;
87f5fc7e
DA		 5374
5375 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
71df5777 5376 * not needed here.
87f5fc7e 5377 */
71df5777 5378 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4c79579b
DA		 5379 if (!neigh)
5380 return BPF_FIB_LKUP_RET_NO_NEIGH;
87f5fc7e 5381
4c79579b 5382 return bpf_fib_set_fwd_params(params, neigh, dev);
87f5fc7e
DA		 5383}
5384#endif
5385
5386BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5387 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5388{
5389 if (plen < sizeof(*params))
5390 return -EINVAL;
5391
9ce64f19
DA		 5392 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5393 return -EINVAL;
5394
87f5fc7e
DA		 5395 switch (params->family) {
5396#if IS_ENABLED(CONFIG_INET)
5397 case AF_INET:
5398 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4f74fede 5399 flags, true);
87f5fc7e
DA		 5400#endif
5401#if IS_ENABLED(CONFIG_IPV6)
5402 case AF_INET6:
5403 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4f74fede 5404 flags, true);
87f5fc7e
DA		 5405#endif
5406 }
bcece5dc 5407 return -EAFNOSUPPORT;
87f5fc7e
DA		 5408}
5409
5410static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5411 .func = bpf_xdp_fib_lookup,
5412 .gpl_only = true,
5413 .ret_type = RET_INTEGER,
5414 .arg1_type = ARG_PTR_TO_CTX,
5415 .arg2_type = ARG_PTR_TO_MEM,
5416 .arg3_type = ARG_CONST_SIZE,
5417 .arg4_type = ARG_ANYTHING,
5418};
5419
5420BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5421 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5422{
4f74fede 5423 struct net *net = dev_net(skb->dev);
4c79579b 5424 int rc = -EAFNOSUPPORT;
4f74fede 5425
87f5fc7e
DA		 5426 if (plen < sizeof(*params))
5427 return -EINVAL;
5428
9ce64f19
DA		 5429 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5430 return -EINVAL;
5431
87f5fc7e
DA		 5432 switch (params->family) {
5433#if IS_ENABLED(CONFIG_INET)
5434 case AF_INET:
4c79579b 5435 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4f74fede 5436 break;
87f5fc7e
DA		 5437#endif
5438#if IS_ENABLED(CONFIG_IPV6)
5439 case AF_INET6:
4c79579b 5440 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4f74fede 5441 break;
87f5fc7e
DA		 5442#endif
5443 }
4f74fede 5444
4c79579b 5445 if (!rc) {
4f74fede
DA		 5446 struct net_device *dev;
5447
4c79579b 5448 dev = dev_get_by_index_rcu(net, params->ifindex);
4f74fede 5449 if (!is_skb_forwardable(dev, skb))
4c79579b 5450 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4f74fede
DA		 5451 }
5452
4c79579b 5453 return rc;
87f5fc7e
DA		 5454}
5455
5456static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5457 .func = bpf_skb_fib_lookup,
5458 .gpl_only = true,
5459 .ret_type = RET_INTEGER,
5460 .arg1_type = ARG_PTR_TO_CTX,
5461 .arg2_type = ARG_PTR_TO_MEM,
5462 .arg3_type = ARG_CONST_SIZE,
5463 .arg4_type = ARG_ANYTHING,
5464};
5465
fe94cc29
MX		 5466#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5467static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5468{
5469 int err;
5470 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5471
bb986a50 5472 if (!seg6_validate_srh(srh, len, false))
fe94cc29
MX		 5473 return -EINVAL;
5474
5475 switch (type) {
5476 case BPF_LWT_ENCAP_SEG6_INLINE:
5477 if (skb->protocol != htons(ETH_P_IPV6))
5478 return -EBADMSG;
5479
5480 err = seg6_do_srh_inline(skb, srh);
5481 break;
5482 case BPF_LWT_ENCAP_SEG6:
5483 skb_reset_inner_headers(skb);
5484 skb->encapsulation = 1;
5485 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5486 break;
5487 default:
5488 return -EINVAL;
5489 }
5490
5491 bpf_compute_data_pointers(skb);
5492 if (err)
5493 return err;
5494
5495 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5496 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5497
5498 return seg6_lookup_nexthop(skb, NULL, 0);
5499}
5500#endif /* CONFIG_IPV6_SEG6_BPF */
5501
3e0bd37c
PO		 5502#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5503static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5504 bool ingress)
5505{
52f27877 5506 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
3e0bd37c
PO		 5507}
5508#endif
5509
5510BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
fe94cc29
MX		 5511 u32, len)
5512{
5513 switch (type) {
5514#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5515 case BPF_LWT_ENCAP_SEG6:
5516 case BPF_LWT_ENCAP_SEG6_INLINE:
5517 return bpf_push_seg6_encap(skb, type, hdr, len);
3e0bd37c
PO		 5518#endif
5519#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5520 case BPF_LWT_ENCAP_IP:
5521 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
fe94cc29
MX		 5522#endif
5523 default:
5524 return -EINVAL;
5525 }
5526}
5527
3e0bd37c
PO		 5528BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5529 void *, hdr, u32, len)
5530{
5531 switch (type) {
5532#if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5533 case BPF_LWT_ENCAP_IP:
5534 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
fe94cc29
MX		 5535#endif
5536 default:
5537 return -EINVAL;
5538 }
5539}
5540
3e0bd37c
PO		 5541static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5542 .func = bpf_lwt_in_push_encap,
5543 .gpl_only = false,
5544 .ret_type = RET_INTEGER,
5545 .arg1_type = ARG_PTR_TO_CTX,
5546 .arg2_type = ARG_ANYTHING,
5547 .arg3_type = ARG_PTR_TO_MEM,
5548 .arg4_type = ARG_CONST_SIZE
5549};
5550
5551static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5552 .func = bpf_lwt_xmit_push_encap,
fe94cc29
MX		 5553 .gpl_only = false,
5554 .ret_type = RET_INTEGER,
5555 .arg1_type = ARG_PTR_TO_CTX,
5556 .arg2_type = ARG_ANYTHING,
5557 .arg3_type = ARG_PTR_TO_MEM,
5558 .arg4_type = ARG_CONST_SIZE
5559};
5560
61d76980 5561#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
fe94cc29
MX		 5562BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5563 const void *, from, u32, len)
5564{
fe94cc29
MX		 5565 struct seg6_bpf_srh_state *srh_state =
5566 this_cpu_ptr(&seg6_bpf_srh_states);
486cdf21 5567 struct ipv6_sr_hdr *srh = srh_state->srh;
fe94cc29 5568 void *srh_tlvs, *srh_end, *ptr;
fe94cc29
MX		 5569 int srhoff = 0;
5570
486cdf21 5571 if (srh == NULL)
fe94cc29
MX		 5572 return -EINVAL;
5573
fe94cc29
MX		 5574 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5575 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5576
5577 ptr = skb->data + offset;
5578 if (ptr >= srh_tlvs && ptr + len <= srh_end)
486cdf21 5579 srh_state->valid = false;
fe94cc29
MX		 5580 else if (ptr < (void *)&srh->flags ||
5581 ptr + len > (void *)&srh->segments)
5582 return -EFAULT;
5583
5584 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5585 return -EFAULT;
486cdf21
MX		 5586 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5587 return -EINVAL;
5588 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
fe94cc29
MX		 5589
5590 memcpy(skb->data + offset, from, len);
5591 return 0;
fe94cc29
MX		 5592}
5593
5594static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5595 .func = bpf_lwt_seg6_store_bytes,
5596 .gpl_only = false,
5597 .ret_type = RET_INTEGER,
5598 .arg1_type = ARG_PTR_TO_CTX,
5599 .arg2_type = ARG_ANYTHING,
5600 .arg3_type = ARG_PTR_TO_MEM,
5601 .arg4_type = ARG_CONST_SIZE
5602};
5603
486cdf21 5604static void bpf_update_srh_state(struct sk_buff *skb)
fe94cc29 5605{
fe94cc29
MX		 5606 struct seg6_bpf_srh_state *srh_state =
5607 this_cpu_ptr(&seg6_bpf_srh_states);
fe94cc29 5608 int srhoff = 0;
fe94cc29 5609
486cdf21
MX		 5610 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5611 srh_state->srh = NULL;
5612 } else {
5613 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5614 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5615 srh_state->valid = true;
fe94cc29 5616 }
486cdf21
MX		 5617}
5618
5619BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5620 u32, action, void *, param, u32, param_len)
5621{
5622 struct seg6_bpf_srh_state *srh_state =
5623 this_cpu_ptr(&seg6_bpf_srh_states);
5624 int hdroff = 0;
5625 int err;
fe94cc29
MX		 5626
5627 switch (action) {
5628 case SEG6_LOCAL_ACTION_END_X:
486cdf21
MX		 5629 if (!seg6_bpf_has_valid_srh(skb))
5630 return -EBADMSG;
fe94cc29
MX		 5631 if (param_len != sizeof(struct in6_addr))
5632 return -EINVAL;
5633 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5634 case SEG6_LOCAL_ACTION_END_T:
486cdf21
MX		 5635 if (!seg6_bpf_has_valid_srh(skb))
5636 return -EBADMSG;
fe94cc29
MX		 5637 if (param_len != sizeof(int))
5638 return -EINVAL;
5639 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
486cdf21
MX		 5640 case SEG6_LOCAL_ACTION_END_DT6:
5641 if (!seg6_bpf_has_valid_srh(skb))
5642 return -EBADMSG;
fe94cc29
MX		 5643 if (param_len != sizeof(int))
5644 return -EINVAL;
486cdf21
MX		 5645
5646 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5647 return -EBADMSG;
5648 if (!pskb_pull(skb, hdroff))
5649 return -EBADMSG;
5650
5651 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5652 skb_reset_network_header(skb);
5653 skb_reset_transport_header(skb);
5654 skb->encapsulation = 0;
5655
5656 bpf_compute_data_pointers(skb);
5657 bpf_update_srh_state(skb);
fe94cc29
MX		 5658 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5659 case SEG6_LOCAL_ACTION_END_B6:
486cdf21
MX		 5660 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5661 return -EBADMSG;
fe94cc29
MX		 5662 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5663 param, param_len);
5664 if (!err)
486cdf21
MX		 5665 bpf_update_srh_state(skb);
5666
fe94cc29
MX		 5667 return err;
5668 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
486cdf21
MX		 5669 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5670 return -EBADMSG;
fe94cc29
MX		 5671 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5672 param, param_len);
5673 if (!err)
486cdf21
MX		 5674 bpf_update_srh_state(skb);
5675
fe94cc29
MX		 5676 return err;
5677 default:
5678 return -EINVAL;
5679 }
fe94cc29
MX		 5680}
5681
5682static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5683 .func = bpf_lwt_seg6_action,
5684 .gpl_only = false,
5685 .ret_type = RET_INTEGER,
5686 .arg1_type = ARG_PTR_TO_CTX,
5687 .arg2_type = ARG_ANYTHING,
5688 .arg3_type = ARG_PTR_TO_MEM,
5689 .arg4_type = ARG_CONST_SIZE
5690};
5691
5692BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5693 s32, len)
5694{
fe94cc29
MX		 5695 struct seg6_bpf_srh_state *srh_state =
5696 this_cpu_ptr(&seg6_bpf_srh_states);
486cdf21 5697 struct ipv6_sr_hdr *srh = srh_state->srh;
fe94cc29 5698 void *srh_end, *srh_tlvs, *ptr;
fe94cc29
MX		 5699 struct ipv6hdr *hdr;
5700 int srhoff = 0;
5701 int ret;
5702
486cdf21 5703 if (unlikely(srh == NULL))
fe94cc29 5704 return -EINVAL;
fe94cc29
MX		 5705
5706 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5707 ((srh->first_segment + 1) << 4));
5708 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5709 srh_state->hdrlen);
5710 ptr = skb->data + offset;
5711
5712 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5713 return -EFAULT;
5714 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5715 return -EFAULT;
5716
5717 if (len > 0) {
5718 ret = skb_cow_head(skb, len);
5719 if (unlikely(ret < 0))
5720 return ret;
5721
5722 ret = bpf_skb_net_hdr_push(skb, offset, len);
5723 } else {
5724 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5725 }
5726
5727 bpf_compute_data_pointers(skb);
5728 if (unlikely(ret < 0))
5729 return ret;
5730
5731 hdr = (struct ipv6hdr *)skb->data;
5732 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5733
486cdf21
MX		 5734 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5735 return -EINVAL;
5736 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
fe94cc29 5737 srh_state->hdrlen += len;
486cdf21 5738 srh_state->valid = false;
fe94cc29 5739 return 0;
fe94cc29
MX		 5740}
5741
5742static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5743 .func = bpf_lwt_seg6_adjust_srh,
5744 .gpl_only = false,
5745 .ret_type = RET_INTEGER,
5746 .arg1_type = ARG_PTR_TO_CTX,
5747 .arg2_type = ARG_ANYTHING,
5748 .arg3_type = ARG_ANYTHING,
5749};
61d76980 5750#endif /* CONFIG_IPV6_SEG6_BPF */
fe94cc29 5751
df3f94a0
AB		 5752#ifdef CONFIG_INET
5753static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
c8123ead 5754 int dif, int sdif, u8 family, u8 proto)
6acc9b43 5755{
6acc9b43
JS		 5756 bool refcounted = false;
5757 struct sock *sk = NULL;
5758
5759 if (family == AF_INET) {
5760 __be32 src4 = tuple->ipv4.saddr;
5761 __be32 dst4 = tuple->ipv4.daddr;
6acc9b43
JS		 5762
5763 if (proto == IPPROTO_TCP)
c8123ead 5764 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6acc9b43
JS		 5765 src4, tuple->ipv4.sport,
5766 dst4, tuple->ipv4.dport,
5767 dif, sdif, &refcounted);
5768 else
5769 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5770 dst4, tuple->ipv4.dport,
c8123ead 5771 dif, sdif, &udp_table, NULL);
8a615c6b 5772#if IS_ENABLED(CONFIG_IPV6)
6acc9b43
JS		 5773 } else {
5774 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5775 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6acc9b43
JS		 5776
5777 if (proto == IPPROTO_TCP)
c8123ead 5778 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6acc9b43 5779 src6, tuple->ipv6.sport,
cac6cc2f 5780 dst6, ntohs(tuple->ipv6.dport),
6acc9b43 5781 dif, sdif, &refcounted);
8a615c6b
JS		 5782 else if (likely(ipv6_bpf_stub))
5783 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5784 src6, tuple->ipv6.sport,
cac6cc2f 5785 dst6, tuple->ipv6.dport,
8a615c6b 5786 dif, sdif,
c8123ead 5787 &udp_table, NULL);
6acc9b43
JS		 5788#endif
5789 }
5790
5791 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5792 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5793 sk = NULL;
5794 }
5795 return sk;
5796}
5797
edbf8c01 5798/* bpf_skc_lookup performs the core lookup for different types of sockets,
6acc9b43
JS		 5799 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5800 * Returns the socket as an 'unsigned long' to simplify the casting in the
5801 * callers to satisfy BPF_CALL declarations.
5802 */
edbf8c01
LB		 5803static struct sock *
5804__bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5805 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5806 u64 flags)
6acc9b43 5807{
6acc9b43
JS		 5808 struct sock *sk = NULL;
5809 u8 family = AF_UNSPEC;
5810 struct net *net;
c8123ead 5811 int sdif;
6acc9b43 5812
9b28ae24
LB		 5813 if (len == sizeof(tuple->ipv4))
5814 family = AF_INET;
5815 else if (len == sizeof(tuple->ipv6))
5816 family = AF_INET6;
5817 else
5818 return NULL;
5819
f71c6143
JS		 5820 if (unlikely(family == AF_UNSPEC || flags ||
5821 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6acc9b43
JS		 5822 goto out;
5823
c8123ead
NH		 5824 if (family == AF_INET)
5825 sdif = inet_sdif(skb);
6acc9b43 5826 else
c8123ead
NH		 5827 sdif = inet6_sdif(skb);
5828
f71c6143
JS		 5829 if ((s32)netns_id < 0) {
5830 net = caller_net;
4cc1feeb 5831 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
f71c6143 5832 } else {
6acc9b43
JS		 5833 net = get_net_ns_by_id(caller_net, netns_id);
5834 if (unlikely(!net))
5835 goto out;
c8123ead 5836 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6acc9b43 5837 put_net(net);
6acc9b43
JS		 5838 }
5839
edbf8c01
LB		 5840out:
5841 return sk;
5842}
5843
5844static struct sock *
5845__bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5846 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5847 u64 flags)
5848{
5849 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5850 ifindex, proto, netns_id, flags);
5851
f7355a6c 5852 if (sk) {
6acc9b43 5853 sk = sk_to_full_sk(sk);
f7355a6c 5854 if (!sk_fullsock(sk)) {
2e012c74 5855 sock_gen_put(sk);
f7355a6c
MKL		 5856 return NULL;
5857 }
5858 }
edbf8c01
LB		 5859
5860 return sk;
6acc9b43
JS		 5861}
5862
edbf8c01
LB		 5863static struct sock *
5864bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5865 u8 proto, u64 netns_id, u64 flags)
c8123ead
NH		 5866{
5867 struct net *caller_net;
5868 int ifindex;
5869
5870 if (skb->dev) {
5871 caller_net = dev_net(skb->dev);
5872 ifindex = skb->dev->ifindex;
5873 } else {
5874 caller_net = sock_net(skb->sk);
5875 ifindex = 0;
5876 }
5877
edbf8c01
LB		 5878 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5879 netns_id, flags);
c8123ead
NH		 5880}
5881
edbf8c01
LB		 5882static struct sock *
5883bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5884 u8 proto, u64 netns_id, u64 flags)
5885{
5886 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5887 flags);
5888
f7355a6c 5889 if (sk) {
edbf8c01 5890 sk = sk_to_full_sk(sk);
f7355a6c 5891 if (!sk_fullsock(sk)) {
2e012c74 5892 sock_gen_put(sk);
f7355a6c
MKL		 5893 return NULL;
5894 }
5895 }
edbf8c01
LB		 5896
5897 return sk;
5898}
5899
5900BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5901 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5902{
5903 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5904 netns_id, flags);
5905}
5906
5907static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5908 .func = bpf_skc_lookup_tcp,
5909 .gpl_only = false,
5910 .pkt_access = true,
5911 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5912 .arg1_type = ARG_PTR_TO_CTX,
5913 .arg2_type = ARG_PTR_TO_MEM,
5914 .arg3_type = ARG_CONST_SIZE,
5915 .arg4_type = ARG_ANYTHING,
5916 .arg5_type = ARG_ANYTHING,
5917};
5918
6acc9b43
JS		 5919BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5920 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5921{
edbf8c01
LB		 5922 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5923 netns_id, flags);
6acc9b43
JS		 5924}
5925
5926static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5927 .func = bpf_sk_lookup_tcp,
5928 .gpl_only = false,
5929 .pkt_access = true,
5930 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5931 .arg1_type = ARG_PTR_TO_CTX,
5932 .arg2_type = ARG_PTR_TO_MEM,
5933 .arg3_type = ARG_CONST_SIZE,
5934 .arg4_type = ARG_ANYTHING,
5935 .arg5_type = ARG_ANYTHING,
5936};
5937
5938BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5939 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5940{
edbf8c01
LB		 5941 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5942 netns_id, flags);
6acc9b43
JS		 5943}
5944
5945static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5946 .func = bpf_sk_lookup_udp,
5947 .gpl_only = false,
5948 .pkt_access = true,
5949 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5950 .arg1_type = ARG_PTR_TO_CTX,
5951 .arg2_type = ARG_PTR_TO_MEM,
5952 .arg3_type = ARG_CONST_SIZE,
5953 .arg4_type = ARG_ANYTHING,
5954 .arg5_type = ARG_ANYTHING,
5955};
5956
5957BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5958{
a5fa25ad 5959 if (sk && sk_is_refcounted(sk))
6acc9b43
JS		 5960 sock_gen_put(sk);
5961 return 0;
5962}
5963
5964static const struct bpf_func_proto bpf_sk_release_proto = {
5965 .func = bpf_sk_release,
5966 .gpl_only = false,
5967 .ret_type = RET_INTEGER,
a5fa25ad 5968 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6acc9b43 5969};
c8123ead
NH		 5970
5971BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5972 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5973{
5974 struct net *caller_net = dev_net(ctx->rxq->dev);
5975 int ifindex = ctx->rxq->dev->ifindex;
5976
edbf8c01
LB		 5977 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5978 ifindex, IPPROTO_UDP, netns_id,
5979 flags);
c8123ead
NH		 5980}
5981
5982static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5983 .func = bpf_xdp_sk_lookup_udp,
5984 .gpl_only = false,
5985 .pkt_access = true,
5986 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5987 .arg1_type = ARG_PTR_TO_CTX,
5988 .arg2_type = ARG_PTR_TO_MEM,
5989 .arg3_type = ARG_CONST_SIZE,
5990 .arg4_type = ARG_ANYTHING,
5991 .arg5_type = ARG_ANYTHING,
5992};
5993
edbf8c01
LB		 5994BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5995 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5996{
5997 struct net *caller_net = dev_net(ctx->rxq->dev);
5998 int ifindex = ctx->rxq->dev->ifindex;
5999
6000 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6001 ifindex, IPPROTO_TCP, netns_id,
6002 flags);
6003}
6004
6005static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6006 .func = bpf_xdp_skc_lookup_tcp,
6007 .gpl_only = false,
6008 .pkt_access = true,
6009 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6010 .arg1_type = ARG_PTR_TO_CTX,
6011 .arg2_type = ARG_PTR_TO_MEM,
6012 .arg3_type = ARG_CONST_SIZE,
6013 .arg4_type = ARG_ANYTHING,
6014 .arg5_type = ARG_ANYTHING,
6015};
6016
c8123ead
NH		 6017BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6018 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6019{
6020 struct net *caller_net = dev_net(ctx->rxq->dev);
6021 int ifindex = ctx->rxq->dev->ifindex;
6022
edbf8c01
LB		 6023 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6024 ifindex, IPPROTO_TCP, netns_id,
6025 flags);
c8123ead
NH		 6026}
6027
6028static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6029 .func = bpf_xdp_sk_lookup_tcp,
6030 .gpl_only = false,
6031 .pkt_access = true,
6032 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6033 .arg1_type = ARG_PTR_TO_CTX,
6034 .arg2_type = ARG_PTR_TO_MEM,
6035 .arg3_type = ARG_CONST_SIZE,
6036 .arg4_type = ARG_ANYTHING,
6037 .arg5_type = ARG_ANYTHING,
6038};
6c49e65e 6039
edbf8c01
LB		 6040BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6041 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6042{
6043 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6044 sock_net(ctx->sk), 0,
6045 IPPROTO_TCP, netns_id, flags);
6046}
6047
6048static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6049 .func = bpf_sock_addr_skc_lookup_tcp,
6050 .gpl_only = false,
6051 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6052 .arg1_type = ARG_PTR_TO_CTX,
6053 .arg2_type = ARG_PTR_TO_MEM,
6054 .arg3_type = ARG_CONST_SIZE,
6055 .arg4_type = ARG_ANYTHING,
6056 .arg5_type = ARG_ANYTHING,
6057};
6058
6c49e65e
AI		 6059BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6060 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6061{
edbf8c01
LB		 6062 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6063 sock_net(ctx->sk), 0, IPPROTO_TCP,
6064 netns_id, flags);
6c49e65e
AI		 6065}
6066
6067static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6068 .func = bpf_sock_addr_sk_lookup_tcp,
6069 .gpl_only = false,
6070 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6071 .arg1_type = ARG_PTR_TO_CTX,
6072 .arg2_type = ARG_PTR_TO_MEM,
6073 .arg3_type = ARG_CONST_SIZE,
6074 .arg4_type = ARG_ANYTHING,
6075 .arg5_type = ARG_ANYTHING,
6076};
6077
6078BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6079 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6080{
edbf8c01
LB		 6081 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6082 sock_net(ctx->sk), 0, IPPROTO_UDP,
6083 netns_id, flags);
6c49e65e
AI		 6084}
6085
6086static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6087 .func = bpf_sock_addr_sk_lookup_udp,
6088 .gpl_only = false,
6089 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6090 .arg1_type = ARG_PTR_TO_CTX,
6091 .arg2_type = ARG_PTR_TO_MEM,
6092 .arg3_type = ARG_CONST_SIZE,
6093 .arg4_type = ARG_ANYTHING,
6094 .arg5_type = ARG_ANYTHING,
6095};
6096
655a51e5
MKL		 6097bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6098 struct bpf_insn_access_aux *info)
6099{
c2cb5e82
SF		 6100 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6101 icsk_retransmits))
655a51e5
MKL		 6102 return false;
6103
6104 if (off % size != 0)
6105 return false;
6106
6107 switch (off) {
6108 case offsetof(struct bpf_tcp_sock, bytes_received):
6109 case offsetof(struct bpf_tcp_sock, bytes_acked):
6110 return size == sizeof(__u64);
6111 default:
6112 return size == sizeof(__u32);
6113 }
6114}
6115
6116u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6117 const struct bpf_insn *si,
6118 struct bpf_insn *insn_buf,
6119 struct bpf_prog *prog, u32 *target_size)
6120{
6121 struct bpf_insn *insn = insn_buf;
6122
6123#define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6124 do { \
c593642c
PB		 6125 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6126 sizeof_field(struct bpf_tcp_sock, FIELD)); \
655a51e5
MKL		 6127 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6128 si->dst_reg, si->src_reg, \
6129 offsetof(struct tcp_sock, FIELD)); \
6130 } while (0)
6131
c2cb5e82
SF		 6132#define BPF_INET_SOCK_GET_COMMON(FIELD) \
6133 do { \
c593642c 6134 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
c2cb5e82 6135 FIELD) > \
c593642c 6136 sizeof_field(struct bpf_tcp_sock, FIELD)); \
c2cb5e82
SF		 6137 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6138 struct inet_connection_sock, \
6139 FIELD), \
6140 si->dst_reg, si->src_reg, \
6141 offsetof( \
6142 struct inet_connection_sock, \
6143 FIELD)); \
6144 } while (0)
6145
655a51e5
MKL		 6146 if (insn > insn_buf)
6147 return insn - insn_buf;
6148
6149 switch (si->off) {
6150 case offsetof(struct bpf_tcp_sock, rtt_min):
c593642c 6151 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
655a51e5
MKL		 6152 sizeof(struct minmax));
6153 BUILD_BUG_ON(sizeof(struct minmax) <
6154 sizeof(struct minmax_sample));
6155
6156 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6157 offsetof(struct tcp_sock, rtt_min) +
6158 offsetof(struct minmax_sample, v));
6159 break;
2377b81d
SF		 6160 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6161 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6162 break;
6163 case offsetof(struct bpf_tcp_sock, srtt_us):
6164 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6165 break;
6166 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6167 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6168 break;
6169 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6170 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6171 break;
6172 case offsetof(struct bpf_tcp_sock, snd_nxt):
6173 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6174 break;
6175 case offsetof(struct bpf_tcp_sock, snd_una):
6176 BPF_TCP_SOCK_GET_COMMON(snd_una);
6177 break;
6178 case offsetof(struct bpf_tcp_sock, mss_cache):
6179 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6180 break;
6181 case offsetof(struct bpf_tcp_sock, ecn_flags):
6182 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6183 break;
6184 case offsetof(struct bpf_tcp_sock, rate_delivered):
6185 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6186 break;
6187 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6188 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6189 break;
6190 case offsetof(struct bpf_tcp_sock, packets_out):
6191 BPF_TCP_SOCK_GET_COMMON(packets_out);
6192 break;
6193 case offsetof(struct bpf_tcp_sock, retrans_out):
6194 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6195 break;
6196 case offsetof(struct bpf_tcp_sock, total_retrans):
6197 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6198 break;
6199 case offsetof(struct bpf_tcp_sock, segs_in):
6200 BPF_TCP_SOCK_GET_COMMON(segs_in);
6201 break;
6202 case offsetof(struct bpf_tcp_sock, data_segs_in):
6203 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6204 break;
6205 case offsetof(struct bpf_tcp_sock, segs_out):
6206 BPF_TCP_SOCK_GET_COMMON(segs_out);
6207 break;
6208 case offsetof(struct bpf_tcp_sock, data_segs_out):
6209 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6210 break;
6211 case offsetof(struct bpf_tcp_sock, lost_out):
6212 BPF_TCP_SOCK_GET_COMMON(lost_out);
6213 break;
6214 case offsetof(struct bpf_tcp_sock, sacked_out):
6215 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6216 break;
6217 case offsetof(struct bpf_tcp_sock, bytes_received):
6218 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6219 break;
6220 case offsetof(struct bpf_tcp_sock, bytes_acked):
6221 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6222 break;
0357746d
SF		 6223 case offsetof(struct bpf_tcp_sock, dsack_dups):
6224 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6225 break;
6226 case offsetof(struct bpf_tcp_sock, delivered):
6227 BPF_TCP_SOCK_GET_COMMON(delivered);
6228 break;
6229 case offsetof(struct bpf_tcp_sock, delivered_ce):
6230 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6231 break;
c2cb5e82
SF		 6232 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6233 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6234 break;
655a51e5
MKL		 6235 }
6236
6237 return insn - insn_buf;
6238}
6239
6240BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6241{
655a51e5
MKL		 6242 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6243 return (unsigned long)sk;
6244
6245 return (unsigned long)NULL;
6246}
6247
0d01da6a 6248const struct bpf_func_proto bpf_tcp_sock_proto = {
655a51e5
MKL		 6249 .func = bpf_tcp_sock,
6250 .gpl_only = false,
6251 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6252 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6253};
6254
dbafd7dd
MKL		 6255BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6256{
6257 sk = sk_to_full_sk(sk);
6258
6259 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6260 return (unsigned long)sk;
6261
6262 return (unsigned long)NULL;
6263}
6264
6265static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6266 .func = bpf_get_listener_sock,
6267 .gpl_only = false,
6268 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6269 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6270};
6271
f7c917ba 6272BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6273{
6274 unsigned int iphdr_len;
6275
d7bf2ebe
THJ		 6276 switch (skb_protocol(skb, true)) {
6277 case cpu_to_be16(ETH_P_IP):
f7c917ba 6278 iphdr_len = sizeof(struct iphdr);
d7bf2ebe
THJ		 6279 break;
6280 case cpu_to_be16(ETH_P_IPV6):
f7c917ba 6281 iphdr_len = sizeof(struct ipv6hdr);
d7bf2ebe
THJ		 6282 break;
6283 default:
f7c917ba 6284 return 0;
d7bf2ebe 6285 }
f7c917ba 6286
6287 if (skb_headlen(skb) < iphdr_len)
6288 return 0;
6289
6290 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6291 return 0;
6292
6293 return INET_ECN_set_ce(skb);
6294}
6295
fada7fdc
JL		 6296bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6297 struct bpf_insn_access_aux *info)
6298{
6299 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6300 return false;
6301
6302 if (off % size != 0)
6303 return false;
6304
6305 switch (off) {
6306 default:
6307 return size == sizeof(__u32);
6308 }
6309}
6310
6311u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6312 const struct bpf_insn *si,
6313 struct bpf_insn *insn_buf,
6314 struct bpf_prog *prog, u32 *target_size)
6315{
6316 struct bpf_insn *insn = insn_buf;
6317
6318#define BPF_XDP_SOCK_GET(FIELD) \
6319 do { \
c593642c
PB		 6320 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6321 sizeof_field(struct bpf_xdp_sock, FIELD)); \
fada7fdc
JL		 6322 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6323 si->dst_reg, si->src_reg, \
6324 offsetof(struct xdp_sock, FIELD)); \
6325 } while (0)
6326
6327 switch (si->off) {
6328 case offsetof(struct bpf_xdp_sock, queue_id):
6329 BPF_XDP_SOCK_GET(queue_id);
6330 break;
6331 }
6332
6333 return insn - insn_buf;
6334}
6335
f7c917ba 6336static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6337 .func = bpf_skb_ecn_set_ce,
6338 .gpl_only = false,
6339 .ret_type = RET_INTEGER,
6340 .arg1_type = ARG_PTR_TO_CTX,
6341};
39904084
LB		 6342
6343BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6344 struct tcphdr *, th, u32, th_len)
6345{
6346#ifdef CONFIG_SYN_COOKIES
6347 u32 cookie;
6348 int ret;
6349
c0df236e 6350 if (unlikely(!sk || th_len < sizeof(*th)))
39904084
LB		 6351 return -EINVAL;
6352
6353 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6354 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6355 return -EINVAL;
6356
6357 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6358 return -EINVAL;
6359
6360 if (!th->ack || th->rst || th->syn)
6361 return -ENOENT;
6362
6363 if (tcp_synq_no_recent_overflow(sk))
6364 return -ENOENT;
6365
6366 cookie = ntohl(th->ack_seq) - 1;
6367
6368 switch (sk->sk_family) {
6369 case AF_INET:
6370 if (unlikely(iph_len < sizeof(struct iphdr)))
6371 return -EINVAL;
6372
6373 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6374 break;
6375
6376#if IS_BUILTIN(CONFIG_IPV6)
6377 case AF_INET6:
6378 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6379 return -EINVAL;
6380
6381 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6382 break;
6383#endif /* CONFIG_IPV6 */
6384
6385 default:
6386 return -EPROTONOSUPPORT;
6387 }
6388
6389 if (ret > 0)
6390 return 0;
6391
6392 return -ENOENT;
6393#else
6394 return -ENOTSUPP;
6395#endif
6396}
6397
6398static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6399 .func = bpf_tcp_check_syncookie,
6400 .gpl_only = true,
6401 .pkt_access = true,
6402 .ret_type = RET_INTEGER,
c0df236e 6403 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
39904084
LB		 6404 .arg2_type = ARG_PTR_TO_MEM,
6405 .arg3_type = ARG_CONST_SIZE,
6406 .arg4_type = ARG_PTR_TO_MEM,
6407 .arg5_type = ARG_CONST_SIZE,
6408};
6409
70d66244
PP		 6410BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6411 struct tcphdr *, th, u32, th_len)
6412{
6413#ifdef CONFIG_SYN_COOKIES
6414 u32 cookie;
6415 u16 mss;
6416
c0df236e 6417 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
70d66244
PP		 6418 return -EINVAL;
6419
6420 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6421 return -EINVAL;
6422
6423 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6424 return -ENOENT;
6425
6426 if (!th->syn || th->ack || th->fin || th->rst)
6427 return -EINVAL;
6428
6429 if (unlikely(iph_len < sizeof(struct iphdr)))
6430 return -EINVAL;
6431
6432 /* Both struct iphdr and struct ipv6hdr have the version field at the
6433 * same offset so we can cast to the shorter header (struct iphdr).
6434 */
6435 switch (((struct iphdr *)iph)->version) {
6436 case 4:
6437 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6438 return -EINVAL;
6439
6440 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6441 break;
6442
6443#if IS_BUILTIN(CONFIG_IPV6)
6444 case 6:
6445 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6446 return -EINVAL;
6447
6448 if (sk->sk_family != AF_INET6)
6449 return -EINVAL;
6450
6451 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6452 break;
6453#endif /* CONFIG_IPV6 */
6454
6455 default:
6456 return -EPROTONOSUPPORT;
6457 }
0741be35 6458 if (mss == 0)
70d66244
PP		 6459 return -ENOENT;
6460
6461 return cookie | ((u64)mss << 32);
6462#else
6463 return -EOPNOTSUPP;
6464#endif /* CONFIG_SYN_COOKIES */
6465}
6466
6467static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6468 .func = bpf_tcp_gen_syncookie,
6469 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6470 .pkt_access = true,
6471 .ret_type = RET_INTEGER,
c0df236e 6472 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
70d66244
PP		 6473 .arg2_type = ARG_PTR_TO_MEM,
6474 .arg3_type = ARG_CONST_SIZE,
6475 .arg4_type = ARG_PTR_TO_MEM,
6476 .arg5_type = ARG_CONST_SIZE,
6477};
6478
cf7fbe66
JS		 6479BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6480{
27e5203b 6481 if (!sk || flags != 0)
cf7fbe66
JS		 6482 return -EINVAL;
6483 if (!skb_at_tc_ingress(skb))
6484 return -EOPNOTSUPP;
6485 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6486 return -ENETUNREACH;
8e368dc7 6487 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
cf7fbe66 6488 return -ESOCKTNOSUPPORT;
7ae215d2
JS		 6489 if (sk_is_refcounted(sk) &&
6490 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
cf7fbe66
JS		 6491 return -ENOENT;
6492
6493 skb_orphan(skb);
6494 skb->sk = sk;
6495 skb->destructor = sock_pfree;
6496
6497 return 0;
6498}
6499
6500static const struct bpf_func_proto bpf_sk_assign_proto = {
6501 .func = bpf_sk_assign,
6502 .gpl_only = false,
6503 .ret_type = RET_INTEGER,
6504 .arg1_type = ARG_PTR_TO_CTX,
27e5203b 6505 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
cf7fbe66
JS		 6506 .arg3_type = ARG_ANYTHING,
6507};
6508
0813a841
MKL		 6509static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6510 u8 search_kind, const u8 *magic,
6511 u8 magic_len, bool *eol)
6512{
6513 u8 kind, kind_len;
6514
6515 *eol = false;
6516
6517 while (op < opend) {
6518 kind = op[0];
6519
6520 if (kind == TCPOPT_EOL) {
6521 *eol = true;
6522 return ERR_PTR(-ENOMSG);
6523 } else if (kind == TCPOPT_NOP) {
6524 op++;
6525 continue;
6526 }
6527
6528 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6529 /* Something is wrong in the received header.
6530 * Follow the TCP stack's tcp_parse_options()
6531 * and just bail here.
6532 */
6533 return ERR_PTR(-EFAULT);
6534
6535 kind_len = op[1];
6536 if (search_kind == kind) {
6537 if (!magic_len)
6538 return op;
6539
6540 if (magic_len > kind_len - 2)
6541 return ERR_PTR(-ENOMSG);
6542
6543 if (!memcmp(&op[2], magic, magic_len))
6544 return op;
6545 }
6546
6547 op += kind_len;
6548 }
6549
6550 return ERR_PTR(-ENOMSG);
6551}
6552
6553BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6554 void *, search_res, u32, len, u64, flags)
6555{
6556 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6557 const u8 *op, *opend, *magic, *search = search_res;
6558 u8 search_kind, search_len, copy_len, magic_len;
6559 int ret;
6560
6561 /* 2 byte is the minimal option len except TCPOPT_NOP and
6562 * TCPOPT_EOL which are useless for the bpf prog to learn
6563 * and this helper disallow loading them also.
6564 */
6565 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6566 return -EINVAL;
6567
6568 search_kind = search[0];
6569 search_len = search[1];
6570
6571 if (search_len > len || search_kind == TCPOPT_NOP ||
6572 search_kind == TCPOPT_EOL)
6573 return -EINVAL;
6574
6575 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6576 /* 16 or 32 bit magic. +2 for kind and kind length */
6577 if (search_len != 4 && search_len != 6)
6578 return -EINVAL;
6579 magic = &search[2];
6580 magic_len = search_len - 2;
6581 } else {
6582 if (search_len)
6583 return -EINVAL;
6584 magic = NULL;
6585 magic_len = 0;
6586 }
6587
6588 if (load_syn) {
6589 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6590 if (ret < 0)
6591 return ret;
6592
6593 opend = op + ret;
6594 op += sizeof(struct tcphdr);
6595 } else {
6596 if (!bpf_sock->skb ||
6597 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6598 /* This bpf_sock->op cannot call this helper */
6599 return -EPERM;
6600
6601 opend = bpf_sock->skb_data_end;
6602 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6603 }
6604
6605 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6606 &eol);
6607 if (IS_ERR(op))
6608 return PTR_ERR(op);
6609
6610 copy_len = op[1];
6611 ret = copy_len;
6612 if (copy_len > len) {
6613 ret = -ENOSPC;
6614 copy_len = len;
6615 }
6616
6617 memcpy(search_res, op, copy_len);
6618 return ret;
6619}
6620
6621static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6622 .func = bpf_sock_ops_load_hdr_opt,
6623 .gpl_only = false,
6624 .ret_type = RET_INTEGER,
6625 .arg1_type = ARG_PTR_TO_CTX,
6626 .arg2_type = ARG_PTR_TO_MEM,
6627 .arg3_type = ARG_CONST_SIZE,
6628 .arg4_type = ARG_ANYTHING,
6629};
6630
6631BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6632 const void *, from, u32, len, u64, flags)
6633{
6634 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6635 const u8 *op, *new_op, *magic = NULL;
6636 struct sk_buff *skb;
6637 bool eol;
6638
6639 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6640 return -EPERM;
6641
6642 if (len < 2 || flags)
6643 return -EINVAL;
6644
6645 new_op = from;
6646 new_kind = new_op[0];
6647 new_kind_len = new_op[1];
6648
6649 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6650 new_kind == TCPOPT_EOL)
6651 return -EINVAL;
6652
6653 if (new_kind_len > bpf_sock->remaining_opt_len)
6654 return -ENOSPC;
6655
6656 /* 253 is another experimental kind */
6657 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6658 if (new_kind_len < 4)
6659 return -EINVAL;
6660 /* Match for the 2 byte magic also.
6661 * RFC 6994: the magic could be 2 or 4 bytes.
6662 * Hence, matching by 2 byte only is on the
6663 * conservative side but it is the right
6664 * thing to do for the 'search-for-duplication'
6665 * purpose.
6666 */
6667 magic = &new_op[2];
6668 magic_len = 2;
6669 }
6670
6671 /* Check for duplication */
6672 skb = bpf_sock->skb;
6673 op = skb->data + sizeof(struct tcphdr);
6674 opend = bpf_sock->skb_data_end;
6675
6676 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6677 &eol);
6678 if (!IS_ERR(op))
6679 return -EEXIST;
6680
6681 if (PTR_ERR(op) != -ENOMSG)
6682 return PTR_ERR(op);
6683
6684 if (eol)
6685 /* The option has been ended. Treat it as no more
6686 * header option can be written.
6687 */
6688 return -ENOSPC;
6689
6690 /* No duplication found. Store the header option. */
6691 memcpy(opend, from, new_kind_len);
6692
6693 bpf_sock->remaining_opt_len -= new_kind_len;
6694 bpf_sock->skb_data_end += new_kind_len;
6695
6696 return 0;
6697}
6698
6699static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6700 .func = bpf_sock_ops_store_hdr_opt,
6701 .gpl_only = false,
6702 .ret_type = RET_INTEGER,
6703 .arg1_type = ARG_PTR_TO_CTX,
6704 .arg2_type = ARG_PTR_TO_MEM,
6705 .arg3_type = ARG_CONST_SIZE,
6706 .arg4_type = ARG_ANYTHING,
6707};
6708
6709BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6710 u32, len, u64, flags)
6711{
6712 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6713 return -EPERM;
6714
6715 if (flags || len < 2)
6716 return -EINVAL;
6717
6718 if (len > bpf_sock->remaining_opt_len)
6719 return -ENOSPC;
6720
6721 bpf_sock->remaining_opt_len -= len;
6722
6723 return 0;
6724}
6725
6726static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6727 .func = bpf_sock_ops_reserve_hdr_opt,
6728 .gpl_only = false,
6729 .ret_type = RET_INTEGER,
6730 .arg1_type = ARG_PTR_TO_CTX,
6731 .arg2_type = ARG_ANYTHING,
6732 .arg3_type = ARG_ANYTHING,
6733};
6734
df3f94a0 6735#endif /* CONFIG_INET */
6acc9b43 6736
fe94cc29
MX		 6737bool bpf_helper_changes_pkt_data(void *func)
6738{
6739 if (func == bpf_skb_vlan_push ||
6740 func == bpf_skb_vlan_pop ||
6741 func == bpf_skb_store_bytes ||
6742 func == bpf_skb_change_proto ||
6743 func == bpf_skb_change_head ||
0ea488ff 6744 func == sk_skb_change_head ||
fe94cc29 6745 func == bpf_skb_change_tail ||
0ea488ff 6746 func == sk_skb_change_tail ||
fe94cc29
MX		 6747 func == bpf_skb_adjust_room ||
6748 func == bpf_skb_pull_data ||
0ea488ff 6749 func == sk_skb_pull_data ||
fe94cc29
MX		 6750 func == bpf_clone_redirect ||
6751 func == bpf_l3_csum_replace ||
6752 func == bpf_l4_csum_replace ||
6753 func == bpf_xdp_adjust_head ||
6754 func == bpf_xdp_adjust_meta ||
6755 func == bpf_msg_pull_data ||
6fff607e 6756 func == bpf_msg_push_data ||
7246d8ed 6757 func == bpf_msg_pop_data ||
fe94cc29 6758 func == bpf_xdp_adjust_tail ||
61d76980 6759#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
fe94cc29
MX		 6760 func == bpf_lwt_seg6_store_bytes ||
6761 func == bpf_lwt_seg6_adjust_srh ||
61d76980 6762 func == bpf_lwt_seg6_action ||
0813a841
MKL		 6763#endif
6764#ifdef CONFIG_INET
6765 func == bpf_sock_ops_store_hdr_opt ||
61d76980 6766#endif
3e0bd37c
PO		 6767 func == bpf_lwt_in_push_encap ||
6768 func == bpf_lwt_xmit_push_encap)
fe94cc29
MX		 6769 return true;
6770
6771 return false;
6772}
6773
6890896b 6774const struct bpf_func_proto bpf_event_output_data_proto __weak;
f7c6cb1d 6775const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
89aa0758 6776
ae2cf1c4 6777static const struct bpf_func_proto *
5e43f899 6778sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
ae2cf1c4
DA		 6779{
6780 switch (func_id) {
6781 /* inet and inet6 sockets are created in a process
6782 * context so there is always a valid uid/gid
6783 */
6784 case BPF_FUNC_get_current_uid_gid:
6785 return &bpf_get_current_uid_gid_proto;
cd339431
RG		 6786 case BPF_FUNC_get_local_storage:
6787 return &bpf_get_local_storage_proto;
0e53d9e5
DB		 6788 case BPF_FUNC_get_socket_cookie:
6789 return &bpf_get_socket_cookie_sock_proto;
f318903c
DB		 6790 case BPF_FUNC_get_netns_cookie:
6791 return &bpf_get_netns_cookie_sock_proto;
fcf752ea
DB		 6792 case BPF_FUNC_perf_event_output:
6793 return &bpf_event_output_data_proto;
834ebca8
DB		 6794 case BPF_FUNC_get_current_pid_tgid:
6795 return &bpf_get_current_pid_tgid_proto;
6796 case BPF_FUNC_get_current_comm:
6797 return &bpf_get_current_comm_proto;
0f09abd1
DB		 6798#ifdef CONFIG_CGROUPS
6799 case BPF_FUNC_get_current_cgroup_id:
6800 return &bpf_get_current_cgroup_id_proto;
6801 case BPF_FUNC_get_current_ancestor_cgroup_id:
6802 return &bpf_get_current_ancestor_cgroup_id_proto;
6803#endif
5a52ae4e
DB		 6804#ifdef CONFIG_CGROUP_NET_CLASSID
6805 case BPF_FUNC_get_cgroup_classid:
6806 return &bpf_get_cgroup_classid_curr_proto;
6807#endif
f7c6cb1d
SF		 6808 case BPF_FUNC_sk_storage_get:
6809 return &bpf_sk_storage_get_cg_sock_proto;
ae2cf1c4
DA		 6810 default:
6811 return bpf_base_func_proto(func_id);
6812 }
6813}
6814
4fbac77d
AI		 6815static const struct bpf_func_proto *
6816sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6817{
6818 switch (func_id) {
6819 /* inet and inet6 sockets are created in a process
6820 * context so there is always a valid uid/gid
6821 */
6822 case BPF_FUNC_get_current_uid_gid:
6823 return &bpf_get_current_uid_gid_proto;
d74bad4e
AI		 6824 case BPF_FUNC_bind:
6825 switch (prog->expected_attach_type) {
6826 case BPF_CGROUP_INET4_CONNECT:
6827 case BPF_CGROUP_INET6_CONNECT:
6828 return &bpf_bind_proto;
6829 default:
6830 return NULL;
6831 }
d692f113
AI		 6832 case BPF_FUNC_get_socket_cookie:
6833 return &bpf_get_socket_cookie_sock_addr_proto;
f318903c
DB		 6834 case BPF_FUNC_get_netns_cookie:
6835 return &bpf_get_netns_cookie_sock_addr_proto;
cd339431
RG		 6836 case BPF_FUNC_get_local_storage:
6837 return &bpf_get_local_storage_proto;
fcf752ea
DB		 6838 case BPF_FUNC_perf_event_output:
6839 return &bpf_event_output_data_proto;
834ebca8
DB		 6840 case BPF_FUNC_get_current_pid_tgid:
6841 return &bpf_get_current_pid_tgid_proto;
6842 case BPF_FUNC_get_current_comm:
6843 return &bpf_get_current_comm_proto;
0f09abd1
DB		 6844#ifdef CONFIG_CGROUPS
6845 case BPF_FUNC_get_current_cgroup_id:
6846 return &bpf_get_current_cgroup_id_proto;
6847 case BPF_FUNC_get_current_ancestor_cgroup_id:
6848 return &bpf_get_current_ancestor_cgroup_id_proto;
6849#endif
5a52ae4e
DB		 6850#ifdef CONFIG_CGROUP_NET_CLASSID
6851 case BPF_FUNC_get_cgroup_classid:
6852 return &bpf_get_cgroup_classid_curr_proto;
6853#endif
6c49e65e
AI		 6854#ifdef CONFIG_INET
6855 case BPF_FUNC_sk_lookup_tcp:
6856 return &bpf_sock_addr_sk_lookup_tcp_proto;
6857 case BPF_FUNC_sk_lookup_udp:
6858 return &bpf_sock_addr_sk_lookup_udp_proto;
6859 case BPF_FUNC_sk_release:
6860 return &bpf_sk_release_proto;
edbf8c01
LB		 6861 case BPF_FUNC_skc_lookup_tcp:
6862 return &bpf_sock_addr_skc_lookup_tcp_proto;
6c49e65e 6863#endif /* CONFIG_INET */
fb85c4a7
SF		 6864 case BPF_FUNC_sk_storage_get:
6865 return &bpf_sk_storage_get_proto;
6866 case BPF_FUNC_sk_storage_delete:
6867 return &bpf_sk_storage_delete_proto;
beecf11b
SF		 6868 case BPF_FUNC_setsockopt:
6869 switch (prog->expected_attach_type) {
6870 case BPF_CGROUP_INET4_CONNECT:
6871 case BPF_CGROUP_INET6_CONNECT:
6872 return &bpf_sock_addr_setsockopt_proto;
6873 default:
6874 return NULL;
6875 }
6876 case BPF_FUNC_getsockopt:
6877 switch (prog->expected_attach_type) {
6878 case BPF_CGROUP_INET4_CONNECT:
6879 case BPF_CGROUP_INET6_CONNECT:
6880 return &bpf_sock_addr_getsockopt_proto;
6881 default:
6882 return NULL;
6883 }
4fbac77d 6884 default:
1df8f55a 6885 return bpf_sk_base_func_proto(func_id);
4fbac77d
AI		 6886 }
6887}
6888
2492d3b8 6889static const struct bpf_func_proto *
5e43f899 6890sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2492d3b8
DB		 6891{
6892 switch (func_id) {
6893 case BPF_FUNC_skb_load_bytes:
6894 return &bpf_skb_load_bytes_proto;
4e1ec56c
DB		 6895 case BPF_FUNC_skb_load_bytes_relative:
6896 return &bpf_skb_load_bytes_relative_proto;
91b8270f
CF		 6897 case BPF_FUNC_get_socket_cookie:
6898 return &bpf_get_socket_cookie_proto;
6acc5c29
CF		 6899 case BPF_FUNC_get_socket_uid:
6900 return &bpf_get_socket_uid_proto;
7c4b90d7
AZ		 6901 case BPF_FUNC_perf_event_output:
6902 return &bpf_skb_event_output_proto;
2492d3b8 6903 default:
1df8f55a 6904 return bpf_sk_base_func_proto(func_id);
2492d3b8
DB		 6905 }
6906}
6907
6ac99e8f
MKL		 6908const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6909const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6910
cd339431
RG		 6911static const struct bpf_func_proto *
6912cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6913{
6914 switch (func_id) {
6915 case BPF_FUNC_get_local_storage:
6916 return &bpf_get_local_storage_proto;
46f8bc92
MKL		 6917 case BPF_FUNC_sk_fullsock:
6918 return &bpf_sk_fullsock_proto;
6ac99e8f
MKL		 6919 case BPF_FUNC_sk_storage_get:
6920 return &bpf_sk_storage_get_proto;
6921 case BPF_FUNC_sk_storage_delete:
6922 return &bpf_sk_storage_delete_proto;
7c4b90d7
AZ		 6923 case BPF_FUNC_perf_event_output:
6924 return &bpf_skb_event_output_proto;
4ecabd55
RG		 6925#ifdef CONFIG_SOCK_CGROUP_DATA
6926 case BPF_FUNC_skb_cgroup_id:
6927 return &bpf_skb_cgroup_id_proto;
06d3e4c9
AI		 6928 case BPF_FUNC_skb_ancestor_cgroup_id:
6929 return &bpf_skb_ancestor_cgroup_id_proto;
f307fa2c
AI		 6930 case BPF_FUNC_sk_cgroup_id:
6931 return &bpf_sk_cgroup_id_proto;
6932 case BPF_FUNC_sk_ancestor_cgroup_id:
6933 return &bpf_sk_ancestor_cgroup_id_proto;
4ecabd55 6934#endif
655a51e5 6935#ifdef CONFIG_INET
d56c2f95
AI		 6936 case BPF_FUNC_sk_lookup_tcp:
6937 return &bpf_sk_lookup_tcp_proto;
6938 case BPF_FUNC_sk_lookup_udp:
6939 return &bpf_sk_lookup_udp_proto;
6940 case BPF_FUNC_sk_release:
6941 return &bpf_sk_release_proto;
6942 case BPF_FUNC_skc_lookup_tcp:
6943 return &bpf_skc_lookup_tcp_proto;
655a51e5
MKL		 6944 case BPF_FUNC_tcp_sock:
6945 return &bpf_tcp_sock_proto;
dbafd7dd
MKL		 6946 case BPF_FUNC_get_listener_sock:
6947 return &bpf_get_listener_sock_proto;
f7c917ba 6948 case BPF_FUNC_skb_ecn_set_ce:
6949 return &bpf_skb_ecn_set_ce_proto;
655a51e5 6950#endif
cd339431
RG		 6951 default:
6952 return sk_filter_func_proto(func_id, prog);
6953 }
6954}
6955
608cd71a 6956static const struct bpf_func_proto *
5e43f899 6957tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
608cd71a
AS		 6958{
6959 switch (func_id) {
6960 case BPF_FUNC_skb_store_bytes:
6961 return &bpf_skb_store_bytes_proto;
05c74e5e
DB		 6962 case BPF_FUNC_skb_load_bytes:
6963 return &bpf_skb_load_bytes_proto;
4e1ec56c
DB		 6964 case BPF_FUNC_skb_load_bytes_relative:
6965 return &bpf_skb_load_bytes_relative_proto;
36bbef52
DB		 6966 case BPF_FUNC_skb_pull_data:
6967 return &bpf_skb_pull_data_proto;
7d672345
DB		 6968 case BPF_FUNC_csum_diff:
6969 return &bpf_csum_diff_proto;
36bbef52
DB		 6970 case BPF_FUNC_csum_update:
6971 return &bpf_csum_update_proto;
7cdec54f
DB		 6972 case BPF_FUNC_csum_level:
6973 return &bpf_csum_level_proto;
91bc4822
AS		 6974 case BPF_FUNC_l3_csum_replace:
6975 return &bpf_l3_csum_replace_proto;
6976 case BPF_FUNC_l4_csum_replace:
6977 return &bpf_l4_csum_replace_proto;
3896d655
AS		 6978 case BPF_FUNC_clone_redirect:
6979 return &bpf_clone_redirect_proto;
8d20aabe
DB		 6980 case BPF_FUNC_get_cgroup_classid:
6981 return &bpf_get_cgroup_classid_proto;
4e10df9a
AS		 6982 case BPF_FUNC_skb_vlan_push:
6983 return &bpf_skb_vlan_push_proto;
6984 case BPF_FUNC_skb_vlan_pop:
6985 return &bpf_skb_vlan_pop_proto;
6578171a
DB		 6986 case BPF_FUNC_skb_change_proto:
6987 return &bpf_skb_change_proto_proto;
d2485c42
DB		 6988 case BPF_FUNC_skb_change_type:
6989 return &bpf_skb_change_type_proto;
2be7e212
DB		 6990 case BPF_FUNC_skb_adjust_room:
6991 return &bpf_skb_adjust_room_proto;
5293efe6
DB		 6992 case BPF_FUNC_skb_change_tail:
6993 return &bpf_skb_change_tail_proto;
6f3f65d8
LC		 6994 case BPF_FUNC_skb_change_head:
6995 return &bpf_skb_change_head_proto;
d3aa45ce
AS		 6996 case BPF_FUNC_skb_get_tunnel_key:
6997 return &bpf_skb_get_tunnel_key_proto;
6998 case BPF_FUNC_skb_set_tunnel_key:
14ca0751
DB		 6999 return bpf_get_skb_set_tunnel_proto(func_id);
7000 case BPF_FUNC_skb_get_tunnel_opt:
7001 return &bpf_skb_get_tunnel_opt_proto;
7002 case BPF_FUNC_skb_set_tunnel_opt:
7003 return bpf_get_skb_set_tunnel_proto(func_id);
27b29f63
AS		 7004 case BPF_FUNC_redirect:
7005 return &bpf_redirect_proto;
b4ab3141
DB		 7006 case BPF_FUNC_redirect_neigh:
7007 return &bpf_redirect_neigh_proto;
c46646d0
DB		 7008 case BPF_FUNC_get_route_realm:
7009 return &bpf_get_route_realm_proto;
13c5c240
DB		 7010 case BPF_FUNC_get_hash_recalc:
7011 return &bpf_get_hash_recalc_proto;
7a4b28c6
DB		 7012 case BPF_FUNC_set_hash_invalid:
7013 return &bpf_set_hash_invalid_proto;
ded092cd
DB		 7014 case BPF_FUNC_set_hash:
7015 return &bpf_set_hash_proto;
bd570ff9 7016 case BPF_FUNC_perf_event_output:
555c8a86 7017 return &bpf_skb_event_output_proto;
80b48c44
DB		 7018 case BPF_FUNC_get_smp_processor_id:
7019 return &bpf_get_smp_processor_id_proto;
747ea55e
DB		 7020 case BPF_FUNC_skb_under_cgroup:
7021 return &bpf_skb_under_cgroup_proto;
91b8270f
CF		 7022 case BPF_FUNC_get_socket_cookie:
7023 return &bpf_get_socket_cookie_proto;
6acc5c29
CF		 7024 case BPF_FUNC_get_socket_uid:
7025 return &bpf_get_socket_uid_proto;
cb20b08e
DB		 7026 case BPF_FUNC_fib_lookup:
7027 return &bpf_skb_fib_lookup_proto;
46f8bc92
MKL		 7028 case BPF_FUNC_sk_fullsock:
7029 return &bpf_sk_fullsock_proto;
6ac99e8f
MKL		 7030 case BPF_FUNC_sk_storage_get:
7031 return &bpf_sk_storage_get_proto;
7032 case BPF_FUNC_sk_storage_delete:
7033 return &bpf_sk_storage_delete_proto;
12bed760
EB		 7034#ifdef CONFIG_XFRM
7035 case BPF_FUNC_skb_get_xfrm_state:
7036 return &bpf_skb_get_xfrm_state_proto;
7037#endif
b426ce83
DB		 7038#ifdef CONFIG_CGROUP_NET_CLASSID
7039 case BPF_FUNC_skb_cgroup_classid:
7040 return &bpf_skb_cgroup_classid_proto;
7041#endif
cb20b08e
DB		 7042#ifdef CONFIG_SOCK_CGROUP_DATA
7043 case BPF_FUNC_skb_cgroup_id:
7044 return &bpf_skb_cgroup_id_proto;
77236281
AI		 7045 case BPF_FUNC_skb_ancestor_cgroup_id:
7046 return &bpf_skb_ancestor_cgroup_id_proto;
cb20b08e 7047#endif
df3f94a0 7048#ifdef CONFIG_INET
6acc9b43
JS		 7049 case BPF_FUNC_sk_lookup_tcp:
7050 return &bpf_sk_lookup_tcp_proto;
7051 case BPF_FUNC_sk_lookup_udp:
7052 return &bpf_sk_lookup_udp_proto;
7053 case BPF_FUNC_sk_release:
7054 return &bpf_sk_release_proto;
655a51e5
MKL		 7055 case BPF_FUNC_tcp_sock:
7056 return &bpf_tcp_sock_proto;
dbafd7dd
MKL		 7057 case BPF_FUNC_get_listener_sock:
7058 return &bpf_get_listener_sock_proto;
edbf8c01
LB		 7059 case BPF_FUNC_skc_lookup_tcp:
7060 return &bpf_skc_lookup_tcp_proto;
39904084
LB		 7061 case BPF_FUNC_tcp_check_syncookie:
7062 return &bpf_tcp_check_syncookie_proto;
315a2029
PO		 7063 case BPF_FUNC_skb_ecn_set_ce:
7064 return &bpf_skb_ecn_set_ce_proto;
70d66244
PP		 7065 case BPF_FUNC_tcp_gen_syncookie:
7066 return &bpf_tcp_gen_syncookie_proto;
cf7fbe66
JS		 7067 case BPF_FUNC_sk_assign:
7068 return &bpf_sk_assign_proto;
df3f94a0 7069#endif
608cd71a 7070 default:
1df8f55a 7071 return bpf_sk_base_func_proto(func_id);
608cd71a
AS		 7072 }
7073}
7074
6a773a15 7075static const struct bpf_func_proto *
5e43f899 7076xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6a773a15 7077{
4de16969
DB		 7078 switch (func_id) {
7079 case BPF_FUNC_perf_event_output:
7080 return &bpf_xdp_event_output_proto;
669dc4d7
DB		 7081 case BPF_FUNC_get_smp_processor_id:
7082 return &bpf_get_smp_processor_id_proto;
205c3807
DB		 7083 case BPF_FUNC_csum_diff:
7084 return &bpf_csum_diff_proto;
17bedab2
MKL		 7085 case BPF_FUNC_xdp_adjust_head:
7086 return &bpf_xdp_adjust_head_proto;
de8f3a83
DB		 7087 case BPF_FUNC_xdp_adjust_meta:
7088 return &bpf_xdp_adjust_meta_proto;
814abfab
JF		 7089 case BPF_FUNC_redirect:
7090 return &bpf_xdp_redirect_proto;
97f91a7c 7091 case BPF_FUNC_redirect_map:
e4a8e817 7092 return &bpf_xdp_redirect_map_proto;
b32cc5b9
NS		 7093 case BPF_FUNC_xdp_adjust_tail:
7094 return &bpf_xdp_adjust_tail_proto;
87f5fc7e
DA		 7095 case BPF_FUNC_fib_lookup:
7096 return &bpf_xdp_fib_lookup_proto;
c8123ead
NH		 7097#ifdef CONFIG_INET
7098 case BPF_FUNC_sk_lookup_udp:
7099 return &bpf_xdp_sk_lookup_udp_proto;
7100 case BPF_FUNC_sk_lookup_tcp:
7101 return &bpf_xdp_sk_lookup_tcp_proto;
7102 case BPF_FUNC_sk_release:
7103 return &bpf_sk_release_proto;
edbf8c01
LB		 7104 case BPF_FUNC_skc_lookup_tcp:
7105 return &bpf_xdp_skc_lookup_tcp_proto;
39904084
LB		 7106 case BPF_FUNC_tcp_check_syncookie:
7107 return &bpf_tcp_check_syncookie_proto;
70d66244
PP		 7108 case BPF_FUNC_tcp_gen_syncookie:
7109 return &bpf_tcp_gen_syncookie_proto;
c8123ead 7110#endif
4de16969 7111 default:
1df8f55a 7112 return bpf_sk_base_func_proto(func_id);
4de16969 7113 }
6a773a15
BB		 7114}
7115
604326b4
DB		 7116const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7117const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7118
8c4b4c7e 7119static const struct bpf_func_proto *
5e43f899 7120sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8c4b4c7e
LB		 7121{
7122 switch (func_id) {
7123 case BPF_FUNC_setsockopt:
beecf11b 7124 return &bpf_sock_ops_setsockopt_proto;
cd86d1fd 7125 case BPF_FUNC_getsockopt:
beecf11b 7126 return &bpf_sock_ops_getsockopt_proto;
b13d8807
LB		 7127 case BPF_FUNC_sock_ops_cb_flags_set:
7128 return &bpf_sock_ops_cb_flags_set_proto;
174a79ff
JF		 7129 case BPF_FUNC_sock_map_update:
7130 return &bpf_sock_map_update_proto;
81110384
JF		 7131 case BPF_FUNC_sock_hash_update:
7132 return &bpf_sock_hash_update_proto;
d692f113
AI		 7133 case BPF_FUNC_get_socket_cookie:
7134 return &bpf_get_socket_cookie_sock_ops_proto;
cd339431
RG		 7135 case BPF_FUNC_get_local_storage:
7136 return &bpf_get_local_storage_proto;
a5a3a828 7137 case BPF_FUNC_perf_event_output:
fcf752ea 7138 return &bpf_event_output_data_proto;
1314ef56
SF		 7139 case BPF_FUNC_sk_storage_get:
7140 return &bpf_sk_storage_get_proto;
7141 case BPF_FUNC_sk_storage_delete:
7142 return &bpf_sk_storage_delete_proto;
7143#ifdef CONFIG_INET
0813a841
MKL		 7144 case BPF_FUNC_load_hdr_opt:
7145 return &bpf_sock_ops_load_hdr_opt_proto;
7146 case BPF_FUNC_store_hdr_opt:
7147 return &bpf_sock_ops_store_hdr_opt_proto;
7148 case BPF_FUNC_reserve_hdr_opt:
7149 return &bpf_sock_ops_reserve_hdr_opt_proto;
1314ef56
SF		 7150 case BPF_FUNC_tcp_sock:
7151 return &bpf_tcp_sock_proto;
7152#endif /* CONFIG_INET */
8c4b4c7e 7153 default:
1df8f55a 7154 return bpf_sk_base_func_proto(func_id);
8c4b4c7e
LB		 7155 }
7156}
7157
604326b4
DB		 7158const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7159const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7160
5e43f899
AI		 7161static const struct bpf_func_proto *
7162sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4f738adb
JF		 7163{
7164 switch (func_id) {
7165 case BPF_FUNC_msg_redirect_map:
7166 return &bpf_msg_redirect_map_proto;
81110384
JF		 7167 case BPF_FUNC_msg_redirect_hash:
7168 return &bpf_msg_redirect_hash_proto;
2a100317
JF		 7169 case BPF_FUNC_msg_apply_bytes:
7170 return &bpf_msg_apply_bytes_proto;
91843d54
JF		 7171 case BPF_FUNC_msg_cork_bytes:
7172 return &bpf_msg_cork_bytes_proto;
015632bb
JF		 7173 case BPF_FUNC_msg_pull_data:
7174 return &bpf_msg_pull_data_proto;
6fff607e
JF		 7175 case BPF_FUNC_msg_push_data:
7176 return &bpf_msg_push_data_proto;
7246d8ed
JF		 7177 case BPF_FUNC_msg_pop_data:
7178 return &bpf_msg_pop_data_proto;
abe3cac8
JF		 7179 case BPF_FUNC_perf_event_output:
7180 return &bpf_event_output_data_proto;
7181 case BPF_FUNC_get_current_uid_gid:
7182 return &bpf_get_current_uid_gid_proto;
7183 case BPF_FUNC_get_current_pid_tgid:
7184 return &bpf_get_current_pid_tgid_proto;
13d70f5a
JF		 7185 case BPF_FUNC_sk_storage_get:
7186 return &bpf_sk_storage_get_proto;
7187 case BPF_FUNC_sk_storage_delete:
7188 return &bpf_sk_storage_delete_proto;
abe3cac8
JF		 7189#ifdef CONFIG_CGROUPS
7190 case BPF_FUNC_get_current_cgroup_id:
7191 return &bpf_get_current_cgroup_id_proto;
7192 case BPF_FUNC_get_current_ancestor_cgroup_id:
7193 return &bpf_get_current_ancestor_cgroup_id_proto;
7194#endif
7195#ifdef CONFIG_CGROUP_NET_CLASSID
7196 case BPF_FUNC_get_cgroup_classid:
7197 return &bpf_get_cgroup_classid_curr_proto;
7198#endif
4f738adb 7199 default:
1df8f55a 7200 return bpf_sk_base_func_proto(func_id);
4f738adb
JF		 7201 }
7202}
7203
604326b4
DB		 7204const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7205const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7206
5e43f899
AI		 7207static const struct bpf_func_proto *
7208sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
b005fd18
JF		 7209{
7210 switch (func_id) {
8a31db56
JF		 7211 case BPF_FUNC_skb_store_bytes:
7212 return &bpf_skb_store_bytes_proto;
b005fd18
JF		 7213 case BPF_FUNC_skb_load_bytes:
7214 return &bpf_skb_load_bytes_proto;
8a31db56 7215 case BPF_FUNC_skb_pull_data:
0ea488ff 7216 return &sk_skb_pull_data_proto;
8a31db56 7217 case BPF_FUNC_skb_change_tail:
0ea488ff 7218 return &sk_skb_change_tail_proto;
8a31db56 7219 case BPF_FUNC_skb_change_head:
0ea488ff 7220 return &sk_skb_change_head_proto;
b005fd18
JF		 7221 case BPF_FUNC_get_socket_cookie:
7222 return &bpf_get_socket_cookie_proto;
7223 case BPF_FUNC_get_socket_uid:
7224 return &bpf_get_socket_uid_proto;
174a79ff
JF		 7225 case BPF_FUNC_sk_redirect_map:
7226 return &bpf_sk_redirect_map_proto;
81110384
JF		 7227 case BPF_FUNC_sk_redirect_hash:
7228 return &bpf_sk_redirect_hash_proto;
7c4b90d7
AZ		 7229 case BPF_FUNC_perf_event_output:
7230 return &bpf_skb_event_output_proto;
df3f94a0 7231#ifdef CONFIG_INET
6acc9b43
JS		 7232 case BPF_FUNC_sk_lookup_tcp:
7233 return &bpf_sk_lookup_tcp_proto;
7234 case BPF_FUNC_sk_lookup_udp:
7235 return &bpf_sk_lookup_udp_proto;
7236 case BPF_FUNC_sk_release:
7237 return &bpf_sk_release_proto;
edbf8c01
LB		 7238 case BPF_FUNC_skc_lookup_tcp:
7239 return &bpf_skc_lookup_tcp_proto;
df3f94a0 7240#endif
b005fd18 7241 default:
1df8f55a 7242 return bpf_sk_base_func_proto(func_id);
b005fd18
JF		 7243 }
7244}
7245
d58e468b
PP		 7246static const struct bpf_func_proto *
7247flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7248{
7249 switch (func_id) {
7250 case BPF_FUNC_skb_load_bytes:
089b19a9 7251 return &bpf_flow_dissector_load_bytes_proto;
d58e468b 7252 default:
1df8f55a 7253 return bpf_sk_base_func_proto(func_id);
d58e468b
PP		 7254 }
7255}
7256
cd3092c7
MX		 7257static const struct bpf_func_proto *
7258lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7259{
7260 switch (func_id) {
7261 case BPF_FUNC_skb_load_bytes:
7262 return &bpf_skb_load_bytes_proto;
7263 case BPF_FUNC_skb_pull_data:
7264 return &bpf_skb_pull_data_proto;
7265 case BPF_FUNC_csum_diff:
7266 return &bpf_csum_diff_proto;
7267 case BPF_FUNC_get_cgroup_classid:
7268 return &bpf_get_cgroup_classid_proto;
7269 case BPF_FUNC_get_route_realm:
7270 return &bpf_get_route_realm_proto;
7271 case BPF_FUNC_get_hash_recalc:
7272 return &bpf_get_hash_recalc_proto;
7273 case BPF_FUNC_perf_event_output:
7274 return &bpf_skb_event_output_proto;
7275 case BPF_FUNC_get_smp_processor_id:
7276 return &bpf_get_smp_processor_id_proto;
7277 case BPF_FUNC_skb_under_cgroup:
7278 return &bpf_skb_under_cgroup_proto;
7279 default:
1df8f55a 7280 return bpf_sk_base_func_proto(func_id);
cd3092c7
MX		 7281 }
7282}
7283
7284static const struct bpf_func_proto *
7285lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7286{
7287 switch (func_id) {
7288 case BPF_FUNC_lwt_push_encap:
3e0bd37c 7289 return &bpf_lwt_in_push_encap_proto;
cd3092c7
MX		 7290 default:
7291 return lwt_out_func_proto(func_id, prog);
7292 }
7293}
7294
3a0af8fd 7295static const struct bpf_func_proto *
5e43f899 7296lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
3a0af8fd
TG		 7297{
7298 switch (func_id) {
7299 case BPF_FUNC_skb_get_tunnel_key:
7300 return &bpf_skb_get_tunnel_key_proto;
7301 case BPF_FUNC_skb_set_tunnel_key:
7302 return bpf_get_skb_set_tunnel_proto(func_id);
7303 case BPF_FUNC_skb_get_tunnel_opt:
7304 return &bpf_skb_get_tunnel_opt_proto;
7305 case BPF_FUNC_skb_set_tunnel_opt:
7306 return bpf_get_skb_set_tunnel_proto(func_id);
7307 case BPF_FUNC_redirect:
7308 return &bpf_redirect_proto;
7309 case BPF_FUNC_clone_redirect:
7310 return &bpf_clone_redirect_proto;
7311 case BPF_FUNC_skb_change_tail:
7312 return &bpf_skb_change_tail_proto;
7313 case BPF_FUNC_skb_change_head:
7314 return &bpf_skb_change_head_proto;
7315 case BPF_FUNC_skb_store_bytes:
7316 return &bpf_skb_store_bytes_proto;
7317 case BPF_FUNC_csum_update:
7318 return &bpf_csum_update_proto;
7cdec54f
DB		 7319 case BPF_FUNC_csum_level:
7320 return &bpf_csum_level_proto;
3a0af8fd
TG		 7321 case BPF_FUNC_l3_csum_replace:
7322 return &bpf_l3_csum_replace_proto;
7323 case BPF_FUNC_l4_csum_replace:
7324 return &bpf_l4_csum_replace_proto;
7325 case BPF_FUNC_set_hash_invalid:
7326 return &bpf_set_hash_invalid_proto;
3e0bd37c
PO		 7327 case BPF_FUNC_lwt_push_encap:
7328 return &bpf_lwt_xmit_push_encap_proto;
3a0af8fd 7329 default:
cd3092c7 7330 return lwt_out_func_proto(func_id, prog);
3a0af8fd
TG		 7331 }
7332}
7333
004d4b27
MX		 7334static const struct bpf_func_proto *
7335lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7336{
7337 switch (func_id) {
61d76980 7338#if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
004d4b27
MX		 7339 case BPF_FUNC_lwt_seg6_store_bytes:
7340 return &bpf_lwt_seg6_store_bytes_proto;
7341 case BPF_FUNC_lwt_seg6_action:
7342 return &bpf_lwt_seg6_action_proto;
7343 case BPF_FUNC_lwt_seg6_adjust_srh:
7344 return &bpf_lwt_seg6_adjust_srh_proto;
61d76980 7345#endif
004d4b27
MX		 7346 default:
7347 return lwt_out_func_proto(func_id, prog);
3a0af8fd
TG		 7348 }
7349}
7350
f96da094 7351static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5e43f899 7352 const struct bpf_prog *prog,
f96da094 7353 struct bpf_insn_access_aux *info)
23994631 7354{
f96da094 7355 const int size_default = sizeof(__u32);
23994631 7356
9bac3d6d
AS		 7357 if (off < 0 || off >= sizeof(struct __sk_buff))
7358 return false;
62c7989b 7359
4936e352 7360 /* The verifier guarantees that size > 0. */
9bac3d6d
AS		 7361 if (off % size != 0)
7362 return false;
62c7989b
DB		 7363
7364 switch (off) {
f96da094
DB		 7365 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7366 if (off + size > offsetofend(struct __sk_buff, cb[4]))
62c7989b
DB		 7367 return false;
7368 break;
8a31db56
JF		 7369 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7370 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7371 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7372 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
f96da094 7373 case bpf_ctx_range(struct __sk_buff, data):
de8f3a83 7374 case bpf_ctx_range(struct __sk_buff, data_meta):
f96da094
DB		 7375 case bpf_ctx_range(struct __sk_buff, data_end):
7376 if (size != size_default)
23994631 7377 return false;
31fd8581 7378 break;
b7df9ada 7379 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
089b19a9 7380 return false;
f11216b2
VD		 7381 case bpf_ctx_range(struct __sk_buff, tstamp):
7382 if (size != sizeof(__u64))
7383 return false;
7384 break;
46f8bc92
MKL		 7385 case offsetof(struct __sk_buff, sk):
7386 if (type == BPF_WRITE || size != sizeof(__u64))
7387 return false;
7388 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7389 break;
31fd8581 7390 default:
f96da094 7391 /* Only narrow read access allowed for now. */
31fd8581 7392 if (type == BPF_WRITE) {
f96da094 7393 if (size != size_default)
31fd8581
YS		 7394 return false;
7395 } else {
f96da094
DB		 7396 bpf_ctx_record_field_size(info, size_default);
7397 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
23994631 7398 return false;
31fd8581 7399 }
62c7989b 7400 }
9bac3d6d
AS		 7401
7402 return true;
7403}
7404
d691f9e8 7405static bool sk_filter_is_valid_access(int off, int size,
19de99f7 7406 enum bpf_access_type type,
5e43f899 7407 const struct bpf_prog *prog,
23994631 7408 struct bpf_insn_access_aux *info)
d691f9e8 7409{
db58ba45 7410 switch (off) {
f96da094
DB		 7411 case bpf_ctx_range(struct __sk_buff, tc_classid):
7412 case bpf_ctx_range(struct __sk_buff, data):
de8f3a83 7413 case bpf_ctx_range(struct __sk_buff, data_meta):
f96da094 7414 case bpf_ctx_range(struct __sk_buff, data_end):
8a31db56 7415 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
f11216b2 7416 case bpf_ctx_range(struct __sk_buff, tstamp):
e3da08d0 7417 case bpf_ctx_range(struct __sk_buff, wire_len):
045efa82 7418 return false;
db58ba45 7419 }
045efa82 7420
d691f9e8
AS		 7421 if (type == BPF_WRITE) {
7422 switch (off) {
f96da094 7423 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
d691f9e8
AS		 7424 break;
7425 default:
7426 return false;
7427 }
7428 }
7429
5e43f899 7430 return bpf_skb_is_valid_access(off, size, type, prog, info);
d691f9e8
AS		 7431}
7432
b39b5f41
SL		 7433static bool cg_skb_is_valid_access(int off, int size,
7434 enum bpf_access_type type,
7435 const struct bpf_prog *prog,
7436 struct bpf_insn_access_aux *info)
7437{
7438 switch (off) {
7439 case bpf_ctx_range(struct __sk_buff, tc_classid):
7440 case bpf_ctx_range(struct __sk_buff, data_meta):
e3da08d0 7441 case bpf_ctx_range(struct __sk_buff, wire_len):
b39b5f41 7442 return false;
ab21c1b5
DB		 7443 case bpf_ctx_range(struct __sk_buff, data):
7444 case bpf_ctx_range(struct __sk_buff, data_end):
2c78ee89 7445 if (!bpf_capable())
ab21c1b5
DB		 7446 return false;
7447 break;
b39b5f41 7448 }
ab21c1b5 7449
b39b5f41
SL		 7450 if (type == BPF_WRITE) {
7451 switch (off) {
7452 case bpf_ctx_range(struct __sk_buff, mark):
7453 case bpf_ctx_range(struct __sk_buff, priority):
7454 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7455 break;
f11216b2 7456 case bpf_ctx_range(struct __sk_buff, tstamp):
2c78ee89 7457 if (!bpf_capable())
f11216b2
VD		 7458 return false;
7459 break;
b39b5f41
SL		 7460 default:
7461 return false;
7462 }
7463 }
7464
7465 switch (off) {
7466 case bpf_ctx_range(struct __sk_buff, data):
7467 info->reg_type = PTR_TO_PACKET;
7468 break;
7469 case bpf_ctx_range(struct __sk_buff, data_end):
7470 info->reg_type = PTR_TO_PACKET_END;
7471 break;
7472 }
7473
7474 return bpf_skb_is_valid_access(off, size, type, prog, info);
7475}
7476
3a0af8fd
TG		 7477static bool lwt_is_valid_access(int off, int size,
7478 enum bpf_access_type type,
5e43f899 7479 const struct bpf_prog *prog,
23994631 7480 struct bpf_insn_access_aux *info)
3a0af8fd
TG		 7481{
7482 switch (off) {
f96da094 7483 case bpf_ctx_range(struct __sk_buff, tc_classid):
8a31db56 7484 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
de8f3a83 7485 case bpf_ctx_range(struct __sk_buff, data_meta):
f11216b2 7486 case bpf_ctx_range(struct __sk_buff, tstamp):
e3da08d0 7487 case bpf_ctx_range(struct __sk_buff, wire_len):
3a0af8fd
TG		 7488 return false;
7489 }
7490
7491 if (type == BPF_WRITE) {
7492 switch (off) {
f96da094
DB		 7493 case bpf_ctx_range(struct __sk_buff, mark):
7494 case bpf_ctx_range(struct __sk_buff, priority):
7495 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3a0af8fd
TG		 7496 break;
7497 default:
7498 return false;
7499 }
7500 }
7501
f96da094
DB		 7502 switch (off) {
7503 case bpf_ctx_range(struct __sk_buff, data):
7504 info->reg_type = PTR_TO_PACKET;
7505 break;
7506 case bpf_ctx_range(struct __sk_buff, data_end):
7507 info->reg_type = PTR_TO_PACKET_END;
7508 break;
7509 }
7510
5e43f899 7511 return bpf_skb_is_valid_access(off, size, type, prog, info);
3a0af8fd
TG		 7512}
7513
aac3fc32
AI		 7514/* Attach type specific accesses */
7515static bool __sock_filter_check_attach_type(int off,
7516 enum bpf_access_type access_type,
7517 enum bpf_attach_type attach_type)
61023658 7518{
aac3fc32
AI		 7519 switch (off) {
7520 case offsetof(struct bpf_sock, bound_dev_if):
7521 case offsetof(struct bpf_sock, mark):
7522 case offsetof(struct bpf_sock, priority):
7523 switch (attach_type) {
7524 case BPF_CGROUP_INET_SOCK_CREATE:
f5836749 7525 case BPF_CGROUP_INET_SOCK_RELEASE:
aac3fc32
AI		 7526 goto full_access;
7527 default:
7528 return false;
7529 }
7530 case bpf_ctx_range(struct bpf_sock, src_ip4):
7531 switch (attach_type) {
7532 case BPF_CGROUP_INET4_POST_BIND:
7533 goto read_only;
7534 default:
7535 return false;
7536 }
7537 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7538 switch (attach_type) {
7539 case BPF_CGROUP_INET6_POST_BIND:
7540 goto read_only;
7541 default:
7542 return false;
7543 }
7544 case bpf_ctx_range(struct bpf_sock, src_port):
7545 switch (attach_type) {
7546 case BPF_CGROUP_INET4_POST_BIND:
7547 case BPF_CGROUP_INET6_POST_BIND:
7548 goto read_only;
61023658
DA		 7549 default:
7550 return false;
7551 }
7552 }
aac3fc32
AI		 7553read_only:
7554 return access_type == BPF_READ;
7555full_access:
7556 return true;
7557}
7558
46f8bc92
MKL		 7559bool bpf_sock_common_is_valid_access(int off, int size,
7560 enum bpf_access_type type,
aac3fc32
AI		 7561 struct bpf_insn_access_aux *info)
7562{
aac3fc32 7563 switch (off) {
46f8bc92
MKL		 7564 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7565 return false;
7566 default:
7567 return bpf_sock_is_valid_access(off, size, type, info);
aac3fc32 7568 }
aac3fc32
AI		 7569}
7570
c64b7983
JS		 7571bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7572 struct bpf_insn_access_aux *info)
aac3fc32 7573{
aa65d696
MKL		 7574 const int size_default = sizeof(__u32);
7575
aac3fc32 7576 if (off < 0 || off >= sizeof(struct bpf_sock))
61023658 7577 return false;
61023658
DA		 7578 if (off % size != 0)
7579 return false;
aa65d696
MKL		 7580
7581 switch (off) {
7582 case offsetof(struct bpf_sock, state):
7583 case offsetof(struct bpf_sock, family):
7584 case offsetof(struct bpf_sock, type):
7585 case offsetof(struct bpf_sock, protocol):
7586 case offsetof(struct bpf_sock, dst_port):
7587 case offsetof(struct bpf_sock, src_port):
c3c16f2e 7588 case offsetof(struct bpf_sock, rx_queue_mapping):
aa65d696
MKL		 7589 case bpf_ctx_range(struct bpf_sock, src_ip4):
7590 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7591 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7592 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7593 bpf_ctx_record_field_size(info, size_default);
7594 return bpf_ctx_narrow_access_ok(off, size, size_default);
7595 }
7596
7597 return size == size_default;
61023658
DA		 7598}
7599
c64b7983
JS		 7600static bool sock_filter_is_valid_access(int off, int size,
7601 enum bpf_access_type type,
7602 const struct bpf_prog *prog,
7603 struct bpf_insn_access_aux *info)
7604{
7605 if (!bpf_sock_is_valid_access(off, size, type, info))
7606 return false;
7607 return __sock_filter_check_attach_type(off, type,
7608 prog->expected_attach_type);
7609}
7610
b09928b9
DB		 7611static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7612 const struct bpf_prog *prog)
7613{
7614 /* Neither direct read nor direct write requires any preliminary
7615 * action.
7616 */
7617 return 0;
7618}
7619
047b0ecd
DB		 7620static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7621 const struct bpf_prog *prog, int drop_verdict)
36bbef52
DB		 7622{
7623 struct bpf_insn *insn = insn_buf;
7624
7625 if (!direct_write)
7626 return 0;
7627
7628 /* if (!skb->cloned)
7629 * goto start;
7630 *
7631 * (Fast-path, otherwise approximation that we might be
7632 * a clone, do the rest in helper.)
7633 */
7634 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7635 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7636 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7637
7638 /* ret = bpf_skb_pull_data(skb, 0); */
7639 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7640 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7641 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7642 BPF_FUNC_skb_pull_data);
7643 /* if (!ret)
7644 * goto restore;
7645 * return TC_ACT_SHOT;
7646 */
7647 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
047b0ecd 7648 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
36bbef52
DB		 7649 *insn++ = BPF_EXIT_INSN();
7650
7651 /* restore: */
7652 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7653 /* start: */
7654 *insn++ = prog->insnsi[0];
7655
7656 return insn - insn_buf;
7657}
7658
e0cea7ce
DB		 7659static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7660 struct bpf_insn *insn_buf)
7661{
7662 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7663 struct bpf_insn *insn = insn_buf;
7664
e0cea7ce
DB		 7665 if (!indirect) {
7666 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7667 } else {
7668 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7669 if (orig->imm)
7670 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7671 }
e6a18d36
DB		 7672 /* We're guaranteed here that CTX is in R6. */
7673 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
e0cea7ce
DB		 7674
7675 switch (BPF_SIZE(orig->code)) {
7676 case BPF_B:
7677 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7678 break;
7679 case BPF_H:
7680 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7681 break;
7682 case BPF_W:
7683 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7684 break;
7685 }
7686
7687 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7688 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7689 *insn++ = BPF_EXIT_INSN();
7690
7691 return insn - insn_buf;
7692}
7693
047b0ecd
DB		 7694static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7695 const struct bpf_prog *prog)
7696{
7697 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7698}
7699
d691f9e8 7700static bool tc_cls_act_is_valid_access(int off, int size,
19de99f7 7701 enum bpf_access_type type,
5e43f899 7702 const struct bpf_prog *prog,
23994631 7703 struct bpf_insn_access_aux *info)
d691f9e8
AS		 7704{
7705 if (type == BPF_WRITE) {
7706 switch (off) {
f96da094
DB		 7707 case bpf_ctx_range(struct __sk_buff, mark):
7708 case bpf_ctx_range(struct __sk_buff, tc_index):
7709 case bpf_ctx_range(struct __sk_buff, priority):
7710 case bpf_ctx_range(struct __sk_buff, tc_classid):
7711 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
f11216b2 7712 case bpf_ctx_range(struct __sk_buff, tstamp):
74e31ca8 7713 case bpf_ctx_range(struct __sk_buff, queue_mapping):
d691f9e8
AS		 7714 break;
7715 default:
7716 return false;
7717 }
7718 }
19de99f7 7719
f96da094
DB		 7720 switch (off) {
7721 case bpf_ctx_range(struct __sk_buff, data):
7722 info->reg_type = PTR_TO_PACKET;
7723 break;
de8f3a83
DB		 7724 case bpf_ctx_range(struct __sk_buff, data_meta):
7725 info->reg_type = PTR_TO_PACKET_META;
7726 break;
f96da094
DB		 7727 case bpf_ctx_range(struct __sk_buff, data_end):
7728 info->reg_type = PTR_TO_PACKET_END;
7729 break;
8a31db56
JF		 7730 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7731 return false;
f96da094
DB		 7732 }
7733
5e43f899 7734 return bpf_skb_is_valid_access(off, size, type, prog, info);
d691f9e8
AS		 7735}
7736
1afaf661 7737static bool __is_valid_xdp_access(int off, int size)
6a773a15
BB		 7738{
7739 if (off < 0 || off >= sizeof(struct xdp_md))
7740 return false;
7741 if (off % size != 0)
7742 return false;
6088b582 7743 if (size != sizeof(__u32))
6a773a15
BB		 7744 return false;
7745
7746 return true;
7747}
7748
7749static bool xdp_is_valid_access(int off, int size,
7750 enum bpf_access_type type,
5e43f899 7751 const struct bpf_prog *prog,
23994631 7752 struct bpf_insn_access_aux *info)
6a773a15 7753{
64b59025
DA		 7754 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7755 switch (off) {
7756 case offsetof(struct xdp_md, egress_ifindex):
7757 return false;
7758 }
7759 }
7760
0d830032
JK		 7761 if (type == BPF_WRITE) {
7762 if (bpf_prog_is_dev_bound(prog->aux)) {
7763 switch (off) {
7764 case offsetof(struct xdp_md, rx_queue_index):
7765 return __is_valid_xdp_access(off, size);
7766 }
7767 }
6a773a15 7768 return false;
0d830032 7769 }
6a773a15
BB		 7770
7771 switch (off) {
7772 case offsetof(struct xdp_md, data):
23994631 7773 info->reg_type = PTR_TO_PACKET;
6a773a15 7774 break;
de8f3a83
DB		 7775 case offsetof(struct xdp_md, data_meta):
7776 info->reg_type = PTR_TO_PACKET_META;
7777 break;
6a773a15 7778 case offsetof(struct xdp_md, data_end):
23994631 7779 info->reg_type = PTR_TO_PACKET_END;
6a773a15
BB		 7780 break;
7781 }
7782
1afaf661 7783 return __is_valid_xdp_access(off, size);
6a773a15
BB		 7784}
7785
7786void bpf_warn_invalid_xdp_action(u32 act)
7787{
9beb8bed
DB		 7788 const u32 act_max = XDP_REDIRECT;
7789
7790 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
7791 act > act_max ? "Illegal" : "Driver unsupported",
7792 act);
6a773a15
BB		 7793}
7794EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7795
4fbac77d
AI		 7796static bool sock_addr_is_valid_access(int off, int size,
7797 enum bpf_access_type type,
7798 const struct bpf_prog *prog,
7799 struct bpf_insn_access_aux *info)
7800{
7801 const int size_default = sizeof(__u32);
7802
7803 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7804 return false;
7805 if (off % size != 0)
7806 return false;
7807
7808 /* Disallow access to IPv6 fields from IPv4 contex and vise
7809 * versa.
7810 */
7811 switch (off) {
7812 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7813 switch (prog->expected_attach_type) {
7814 case BPF_CGROUP_INET4_BIND:
d74bad4e 7815 case BPF_CGROUP_INET4_CONNECT:
1b66d253
DB		 7816 case BPF_CGROUP_INET4_GETPEERNAME:
7817 case BPF_CGROUP_INET4_GETSOCKNAME:
1cedee13 7818 case BPF_CGROUP_UDP4_SENDMSG:
983695fa 7819 case BPF_CGROUP_UDP4_RECVMSG:
4fbac77d
AI		 7820 break;
7821 default:
7822 return false;
7823 }
7824 break;
7825 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7826 switch (prog->expected_attach_type) {
7827 case BPF_CGROUP_INET6_BIND:
d74bad4e 7828 case BPF_CGROUP_INET6_CONNECT:
1b66d253
DB		 7829 case BPF_CGROUP_INET6_GETPEERNAME:
7830 case BPF_CGROUP_INET6_GETSOCKNAME:
1cedee13 7831 case BPF_CGROUP_UDP6_SENDMSG:
983695fa 7832 case BPF_CGROUP_UDP6_RECVMSG:
1cedee13
AI		 7833 break;
7834 default:
7835 return false;
7836 }
7837 break;
7838 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7839 switch (prog->expected_attach_type) {
7840 case BPF_CGROUP_UDP4_SENDMSG:
7841 break;
7842 default:
7843 return false;
7844 }
7845 break;
7846 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7847 msg_src_ip6[3]):
7848 switch (prog->expected_attach_type) {
7849 case BPF_CGROUP_UDP6_SENDMSG:
4fbac77d
AI		 7850 break;
7851 default:
7852 return false;
7853 }
7854 break;
7855 }
7856
7857 switch (off) {
7858 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7859 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
1cedee13
AI		 7860 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7861 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7862 msg_src_ip6[3]):
7aebfa1b 7863 case bpf_ctx_range(struct bpf_sock_addr, user_port):
4fbac77d
AI		 7864 if (type == BPF_READ) {
7865 bpf_ctx_record_field_size(info, size_default);
d4ecfeb1
SF		 7866
7867 if (bpf_ctx_wide_access_ok(off, size,
7868 struct bpf_sock_addr,
7869 user_ip6))
7870 return true;
7871
7872 if (bpf_ctx_wide_access_ok(off, size,
7873 struct bpf_sock_addr,
7874 msg_src_ip6))
7875 return true;
7876
4fbac77d
AI		 7877 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7878 return false;
7879 } else {
b4399546
SF		 7880 if (bpf_ctx_wide_access_ok(off, size,
7881 struct bpf_sock_addr,
7882 user_ip6))
600c70ba
SF		 7883 return true;
7884
b4399546
SF		 7885 if (bpf_ctx_wide_access_ok(off, size,
7886 struct bpf_sock_addr,
7887 msg_src_ip6))
600c70ba
SF		 7888 return true;
7889
4fbac77d
AI		 7890 if (size != size_default)
7891 return false;
7892 }
7893 break;
fb85c4a7
SF		 7894 case offsetof(struct bpf_sock_addr, sk):
7895 if (type != BPF_READ)
7896 return false;
7897 if (size != sizeof(__u64))
7898 return false;
7899 info->reg_type = PTR_TO_SOCKET;
7900 break;
4fbac77d
AI		 7901 default:
7902 if (type == BPF_READ) {
7903 if (size != size_default)
7904 return false;
7905 } else {
7906 return false;
7907 }
7908 }
7909
7910 return true;
7911}
7912
44f0e430
LB		 7913static bool sock_ops_is_valid_access(int off, int size,
7914 enum bpf_access_type type,
5e43f899 7915 const struct bpf_prog *prog,
44f0e430 7916 struct bpf_insn_access_aux *info)
40304b2a 7917{
44f0e430
LB		 7918 const int size_default = sizeof(__u32);
7919
40304b2a
LB		 7920 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
7921 return false;
44f0e430 7922
40304b2a
LB		 7923 /* The verifier guarantees that size > 0. */
7924 if (off % size != 0)
7925 return false;
40304b2a 7926
40304b2a
LB		 7927 if (type == BPF_WRITE) {
7928 switch (off) {
2585cd62 7929 case offsetof(struct bpf_sock_ops, reply):
6f9bd3d7 7930 case offsetof(struct bpf_sock_ops, sk_txhash):
44f0e430
LB		 7931 if (size != size_default)
7932 return false;
40304b2a
LB		 7933 break;
7934 default:
7935 return false;
7936 }
44f0e430
LB		 7937 } else {
7938 switch (off) {
7939 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
7940 bytes_acked):
7941 if (size != sizeof(__u64))
7942 return false;
7943 break;
1314ef56
SF		 7944 case offsetof(struct bpf_sock_ops, sk):
7945 if (size != sizeof(__u64))
7946 return false;
7947 info->reg_type = PTR_TO_SOCKET_OR_NULL;
7948 break;
0813a841
MKL		 7949 case offsetof(struct bpf_sock_ops, skb_data):
7950 if (size != sizeof(__u64))
7951 return false;
7952 info->reg_type = PTR_TO_PACKET;
7953 break;
7954 case offsetof(struct bpf_sock_ops, skb_data_end):
7955 if (size != sizeof(__u64))
7956 return false;
7957 info->reg_type = PTR_TO_PACKET_END;
7958 break;
7959 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
7960 bpf_ctx_record_field_size(info, size_default);
7961 return bpf_ctx_narrow_access_ok(off, size,
7962 size_default);
44f0e430
LB		 7963 default:
7964 if (size != size_default)
7965 return false;
7966 break;
7967 }
40304b2a
LB		 7968 }
7969
44f0e430 7970 return true;
40304b2a
LB		 7971}
7972
8a31db56
JF		 7973static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
7974 const struct bpf_prog *prog)
7975{
047b0ecd 7976 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8a31db56
JF		 7977}
7978
b005fd18
JF		 7979static bool sk_skb_is_valid_access(int off, int size,
7980 enum bpf_access_type type,
5e43f899 7981 const struct bpf_prog *prog,
b005fd18
JF		 7982 struct bpf_insn_access_aux *info)
7983{
de8f3a83
DB		 7984 switch (off) {
7985 case bpf_ctx_range(struct __sk_buff, tc_classid):
7986 case bpf_ctx_range(struct __sk_buff, data_meta):
f11216b2 7987 case bpf_ctx_range(struct __sk_buff, tstamp):
e3da08d0 7988 case bpf_ctx_range(struct __sk_buff, wire_len):
de8f3a83
DB		 7989 return false;
7990 }
7991
8a31db56
JF		 7992 if (type == BPF_WRITE) {
7993 switch (off) {
8a31db56
JF		 7994 case bpf_ctx_range(struct __sk_buff, tc_index):
7995 case bpf_ctx_range(struct __sk_buff, priority):
7996 break;
7997 default:
7998 return false;
7999 }
8000 }
8001
b005fd18 8002 switch (off) {
f7e9cb1e 8003 case bpf_ctx_range(struct __sk_buff, mark):
8a31db56 8004 return false;
b005fd18
JF		 8005 case bpf_ctx_range(struct __sk_buff, data):
8006 info->reg_type = PTR_TO_PACKET;
8007 break;
8008 case bpf_ctx_range(struct __sk_buff, data_end):
8009 info->reg_type = PTR_TO_PACKET_END;
8010 break;
8011 }
8012
5e43f899 8013 return bpf_skb_is_valid_access(off, size, type, prog, info);
b005fd18
JF		 8014}
8015
4f738adb
JF		 8016static bool sk_msg_is_valid_access(int off, int size,
8017 enum bpf_access_type type,
5e43f899 8018 const struct bpf_prog *prog,
4f738adb
JF		 8019 struct bpf_insn_access_aux *info)
8020{
8021 if (type == BPF_WRITE)
8022 return false;
8023
bc1b4f01
JF		 8024 if (off % size != 0)
8025 return false;
8026
4f738adb
JF		 8027 switch (off) {
8028 case offsetof(struct sk_msg_md, data):
8029 info->reg_type = PTR_TO_PACKET;
303def35
JF		 8030 if (size != sizeof(__u64))
8031 return false;
4f738adb
JF		 8032 break;
8033 case offsetof(struct sk_msg_md, data_end):
8034 info->reg_type = PTR_TO_PACKET_END;
303def35
JF		 8035 if (size != sizeof(__u64))
8036 return false;
4f738adb 8037 break;
13d70f5a
JF		 8038 case offsetof(struct sk_msg_md, sk):
8039 if (size != sizeof(__u64))
8040 return false;
8041 info->reg_type = PTR_TO_SOCKET;
8042 break;
bc1b4f01
JF		 8043 case bpf_ctx_range(struct sk_msg_md, family):
8044 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8045 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8046 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8047 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8048 case bpf_ctx_range(struct sk_msg_md, remote_port):
8049 case bpf_ctx_range(struct sk_msg_md, local_port):
8050 case bpf_ctx_range(struct sk_msg_md, size):
303def35
JF		 8051 if (size != sizeof(__u32))
8052 return false;
bc1b4f01
JF		 8053 break;
8054 default:
4f738adb 8055 return false;
bc1b4f01 8056 }
4f738adb
JF		 8057 return true;
8058}
8059
d58e468b
PP		 8060static bool flow_dissector_is_valid_access(int off, int size,
8061 enum bpf_access_type type,
8062 const struct bpf_prog *prog,
8063 struct bpf_insn_access_aux *info)
8064{
089b19a9
SF		 8065 const int size_default = sizeof(__u32);
8066
8067 if (off < 0 || off >= sizeof(struct __sk_buff))
8068 return false;
8069
2ee7fba0
SF		 8070 if (type == BPF_WRITE)
8071 return false;
d58e468b
PP		 8072
8073 switch (off) {
8074 case bpf_ctx_range(struct __sk_buff, data):
089b19a9
SF		 8075 if (size != size_default)
8076 return false;
d58e468b 8077 info->reg_type = PTR_TO_PACKET;
089b19a9 8078 return true;
d58e468b 8079 case bpf_ctx_range(struct __sk_buff, data_end):
089b19a9
SF		 8080 if (size != size_default)
8081 return false;
d58e468b 8082 info->reg_type = PTR_TO_PACKET_END;
089b19a9 8083 return true;
b7df9ada 8084 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
089b19a9
SF		 8085 if (size != sizeof(__u64))
8086 return false;
d58e468b 8087 info->reg_type = PTR_TO_FLOW_KEYS;
089b19a9 8088 return true;
2ee7fba0 8089 default:
d58e468b
PP		 8090 return false;
8091 }
089b19a9 8092}
d58e468b 8093
089b19a9
SF		 8094static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8095 const struct bpf_insn *si,
8096 struct bpf_insn *insn_buf,
8097 struct bpf_prog *prog,
8098 u32 *target_size)
8099
8100{
8101 struct bpf_insn *insn = insn_buf;
8102
8103 switch (si->off) {
8104 case offsetof(struct __sk_buff, data):
8105 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8106 si->dst_reg, si->src_reg,
8107 offsetof(struct bpf_flow_dissector, data));
8108 break;
8109
8110 case offsetof(struct __sk_buff, data_end):
8111 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8112 si->dst_reg, si->src_reg,
8113 offsetof(struct bpf_flow_dissector, data_end));
8114 break;
8115
8116 case offsetof(struct __sk_buff, flow_keys):
8117 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8118 si->dst_reg, si->src_reg,
8119 offsetof(struct bpf_flow_dissector, flow_keys));
8120 break;
8121 }
8122
8123 return insn - insn_buf;
d58e468b
PP		 8124}
8125
cf62089b
WB		 8126static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8127 struct bpf_insn *insn)
8128{
8129 /* si->dst_reg = skb_shinfo(SKB); */
8130#ifdef NET_SKBUFF_DATA_USES_OFFSET
8131 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8132 BPF_REG_AX, si->src_reg,
8133 offsetof(struct sk_buff, end));
8134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8135 si->dst_reg, si->src_reg,
8136 offsetof(struct sk_buff, head));
8137 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8138#else
8139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8140 si->dst_reg, si->src_reg,
8141 offsetof(struct sk_buff, end));
8142#endif
8143
8144 return insn;
8145}
8146
2492d3b8
DB		 8147static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8148 const struct bpf_insn *si,
8149 struct bpf_insn *insn_buf,
f96da094 8150 struct bpf_prog *prog, u32 *target_size)
9bac3d6d
AS		 8151{
8152 struct bpf_insn *insn = insn_buf;
6b8cc1d1 8153 int off;
9bac3d6d 8154
6b8cc1d1 8155 switch (si->off) {
9bac3d6d 8156 case offsetof(struct __sk_buff, len):
6b8cc1d1 8157 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8158 bpf_target_off(struct sk_buff, len, 4,
8159 target_size));
9bac3d6d
AS		 8160 break;
8161
0b8c707d 8162 case offsetof(struct __sk_buff, protocol):
6b8cc1d1 8163 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
f96da094
DB		 8164 bpf_target_off(struct sk_buff, protocol, 2,
8165 target_size));
0b8c707d
DB		 8166 break;
8167
27cd5452 8168 case offsetof(struct __sk_buff, vlan_proto):
6b8cc1d1 8169 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
f96da094
DB		 8170 bpf_target_off(struct sk_buff, vlan_proto, 2,
8171 target_size));
27cd5452
MS		 8172 break;
8173
bcad5718 8174 case offsetof(struct __sk_buff, priority):
754f1e6a 8175 if (type == BPF_WRITE)
6b8cc1d1 8176 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8177 bpf_target_off(struct sk_buff, priority, 4,
8178 target_size));
754f1e6a 8179 else
6b8cc1d1 8180 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8181 bpf_target_off(struct sk_buff, priority, 4,
8182 target_size));
bcad5718
DB		 8183 break;
8184
37e82c2f 8185 case offsetof(struct __sk_buff, ingress_ifindex):
6b8cc1d1 8186 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8187 bpf_target_off(struct sk_buff, skb_iif, 4,
8188 target_size));
37e82c2f
AS		 8189 break;
8190
8191 case offsetof(struct __sk_buff, ifindex):
f035a515 8192 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6b8cc1d1 8193 si->dst_reg, si->src_reg,
37e82c2f 8194 offsetof(struct sk_buff, dev));
6b8cc1d1
DB		 8195 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8196 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
f96da094
DB		 8197 bpf_target_off(struct net_device, ifindex, 4,
8198 target_size));
37e82c2f
AS		 8199 break;
8200
ba7591d8 8201 case offsetof(struct __sk_buff, hash):
6b8cc1d1 8202 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8203 bpf_target_off(struct sk_buff, hash, 4,
8204 target_size));
ba7591d8
DB		 8205 break;
8206
9bac3d6d 8207 case offsetof(struct __sk_buff, mark):
d691f9e8 8208 if (type == BPF_WRITE)
6b8cc1d1 8209 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8210 bpf_target_off(struct sk_buff, mark, 4,
8211 target_size));
d691f9e8 8212 else
6b8cc1d1 8213 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8214 bpf_target_off(struct sk_buff, mark, 4,
8215 target_size));
d691f9e8 8216 break;
9bac3d6d
AS		 8217
8218 case offsetof(struct __sk_buff, pkt_type):
f96da094
DB		 8219 *target_size = 1;
8220 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8221 PKT_TYPE_OFFSET());
8222 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8223#ifdef __BIG_ENDIAN_BITFIELD
8224 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8225#endif
8226 break;
9bac3d6d
AS		 8227
8228 case offsetof(struct __sk_buff, queue_mapping):
74e31ca8
JDB		 8229 if (type == BPF_WRITE) {
8230 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8231 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8232 bpf_target_off(struct sk_buff,
8233 queue_mapping,
8234 2, target_size));
8235 } else {
8236 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8237 bpf_target_off(struct sk_buff,
8238 queue_mapping,
8239 2, target_size));
8240 }
f96da094 8241 break;
c2497395 8242
c2497395 8243 case offsetof(struct __sk_buff, vlan_present):
9c212255
MM		 8244 *target_size = 1;
8245 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8246 PKT_VLAN_PRESENT_OFFSET());
8247 if (PKT_VLAN_PRESENT_BIT)
8248 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8249 if (PKT_VLAN_PRESENT_BIT < 7)
8250 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8251 break;
f96da094 8252
9c212255 8253 case offsetof(struct __sk_buff, vlan_tci):
f96da094
DB		 8254 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8255 bpf_target_off(struct sk_buff, vlan_tci, 2,
8256 target_size));
f96da094 8257 break;
d691f9e8
AS		 8258
8259 case offsetof(struct __sk_buff, cb[0]) ...
f96da094 8260 offsetofend(struct __sk_buff, cb[4]) - 1:
c593642c 8261 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
62c7989b
DB		 8262 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8263 offsetof(struct qdisc_skb_cb, data)) %
8264 sizeof(__u64));
d691f9e8 8265
ff936a04 8266 prog->cb_access = 1;
6b8cc1d1
DB		 8267 off = si->off;
8268 off -= offsetof(struct __sk_buff, cb[0]);
8269 off += offsetof(struct sk_buff, cb);
8270 off += offsetof(struct qdisc_skb_cb, data);
d691f9e8 8271 if (type == BPF_WRITE)
62c7989b 8272 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6b8cc1d1 8273 si->src_reg, off);
d691f9e8 8274 else
62c7989b 8275 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6b8cc1d1 8276 si->src_reg, off);
d691f9e8
AS		 8277 break;
8278
045efa82 8279 case offsetof(struct __sk_buff, tc_classid):
c593642c 8280 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
6b8cc1d1
DB		 8281
8282 off = si->off;
8283 off -= offsetof(struct __sk_buff, tc_classid);
8284 off += offsetof(struct sk_buff, cb);
8285 off += offsetof(struct qdisc_skb_cb, tc_classid);
f96da094 8286 *target_size = 2;
09c37a2c 8287 if (type == BPF_WRITE)
6b8cc1d1
DB		 8288 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8289 si->src_reg, off);
09c37a2c 8290 else
6b8cc1d1
DB		 8291 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8292 si->src_reg, off);
045efa82
DB		 8293 break;
8294
db58ba45 8295 case offsetof(struct __sk_buff, data):
f035a515 8296 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6b8cc1d1 8297 si->dst_reg, si->src_reg,
db58ba45
AS		 8298 offsetof(struct sk_buff, data));
8299 break;
8300
de8f3a83
DB		 8301 case offsetof(struct __sk_buff, data_meta):
8302 off = si->off;
8303 off -= offsetof(struct __sk_buff, data_meta);
8304 off += offsetof(struct sk_buff, cb);
8305 off += offsetof(struct bpf_skb_data_end, data_meta);
8306 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8307 si->src_reg, off);
8308 break;
8309
db58ba45 8310 case offsetof(struct __sk_buff, data_end):
6b8cc1d1
DB		 8311 off = si->off;
8312 off -= offsetof(struct __sk_buff, data_end);
8313 off += offsetof(struct sk_buff, cb);
8314 off += offsetof(struct bpf_skb_data_end, data_end);
8315 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8316 si->src_reg, off);
db58ba45
AS		 8317 break;
8318
d691f9e8
AS		 8319 case offsetof(struct __sk_buff, tc_index):
8320#ifdef CONFIG_NET_SCHED
d691f9e8 8321 if (type == BPF_WRITE)
6b8cc1d1 8322 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
f96da094
DB		 8323 bpf_target_off(struct sk_buff, tc_index, 2,
8324 target_size));
d691f9e8 8325 else
6b8cc1d1 8326 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
f96da094
DB		 8327 bpf_target_off(struct sk_buff, tc_index, 2,
8328 target_size));
d691f9e8 8329#else
2ed46ce4 8330 *target_size = 2;
d691f9e8 8331 if (type == BPF_WRITE)
6b8cc1d1 8332 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
d691f9e8 8333 else
6b8cc1d1 8334 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
b1d9fc41
DB		 8335#endif
8336 break;
8337
8338 case offsetof(struct __sk_buff, napi_id):
8339#if defined(CONFIG_NET_RX_BUSY_POLL)
b1d9fc41 8340 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
f96da094
DB		 8341 bpf_target_off(struct sk_buff, napi_id, 4,
8342 target_size));
b1d9fc41
DB		 8343 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8344 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8345#else
2ed46ce4 8346 *target_size = 4;
b1d9fc41 8347 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
d691f9e8 8348#endif
6b8cc1d1 8349 break;
8a31db56 8350 case offsetof(struct __sk_buff, family):
c593642c 8351 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8a31db56
JF		 8352
8353 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8354 si->dst_reg, si->src_reg,
8355 offsetof(struct sk_buff, sk));
8356 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8357 bpf_target_off(struct sock_common,
8358 skc_family,
8359 2, target_size));
8360 break;
8361 case offsetof(struct __sk_buff, remote_ip4):
c593642c 8362 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8a31db56
JF		 8363
8364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8365 si->dst_reg, si->src_reg,
8366 offsetof(struct sk_buff, sk));
8367 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8368 bpf_target_off(struct sock_common,
8369 skc_daddr,
8370 4, target_size));
8371 break;
8372 case offsetof(struct __sk_buff, local_ip4):
c593642c 8373 BUILD_BUG_ON(sizeof_field(struct sock_common,
8a31db56
JF		 8374 skc_rcv_saddr) != 4);
8375
8376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8377 si->dst_reg, si->src_reg,
8378 offsetof(struct sk_buff, sk));
8379 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8380 bpf_target_off(struct sock_common,
8381 skc_rcv_saddr,
8382 4, target_size));
8383 break;
8384 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8385 offsetof(struct __sk_buff, remote_ip6[3]):
8386#if IS_ENABLED(CONFIG_IPV6)
c593642c 8387 BUILD_BUG_ON(sizeof_field(struct sock_common,
8a31db56
JF		 8388 skc_v6_daddr.s6_addr32[0]) != 4);
8389
8390 off = si->off;
8391 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8392
8393 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8394 si->dst_reg, si->src_reg,
8395 offsetof(struct sk_buff, sk));
8396 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8397 offsetof(struct sock_common,
8398 skc_v6_daddr.s6_addr32[0]) +
8399 off);
8400#else
8401 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8402#endif
8403 break;
8404 case offsetof(struct __sk_buff, local_ip6[0]) ...
8405 offsetof(struct __sk_buff, local_ip6[3]):
8406#if IS_ENABLED(CONFIG_IPV6)
c593642c 8407 BUILD_BUG_ON(sizeof_field(struct sock_common,
8a31db56
JF		 8408 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8409
8410 off = si->off;
8411 off -= offsetof(struct __sk_buff, local_ip6[0]);
8412
8413 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8414 si->dst_reg, si->src_reg,
8415 offsetof(struct sk_buff, sk));
8416 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8417 offsetof(struct sock_common,
8418 skc_v6_rcv_saddr.s6_addr32[0]) +
8419 off);
8420#else
8421 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8422#endif
8423 break;
8424
8425 case offsetof(struct __sk_buff, remote_port):
c593642c 8426 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8a31db56
JF		 8427
8428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8429 si->dst_reg, si->src_reg,
8430 offsetof(struct sk_buff, sk));
8431 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8432 bpf_target_off(struct sock_common,
8433 skc_dport,
8434 2, target_size));
8435#ifndef __BIG_ENDIAN_BITFIELD
8436 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8437#endif
8438 break;
8439
8440 case offsetof(struct __sk_buff, local_port):
c593642c 8441 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8a31db56
JF		 8442
8443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8444 si->dst_reg, si->src_reg,
8445 offsetof(struct sk_buff, sk));
8446 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8447 bpf_target_off(struct sock_common,
8448 skc_num, 2, target_size));
8449 break;
d58e468b 8450
f11216b2 8451 case offsetof(struct __sk_buff, tstamp):
c593642c 8452 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
f11216b2
VD		 8453
8454 if (type == BPF_WRITE)
8455 *insn++ = BPF_STX_MEM(BPF_DW,
8456 si->dst_reg, si->src_reg,
8457 bpf_target_off(struct sk_buff,
8458 tstamp, 8,
8459 target_size));
8460 else
8461 *insn++ = BPF_LDX_MEM(BPF_DW,
8462 si->dst_reg, si->src_reg,
8463 bpf_target_off(struct sk_buff,
8464 tstamp, 8,
8465 target_size));
e3da08d0
PP		 8466 break;
8467
d9ff286a 8468 case offsetof(struct __sk_buff, gso_segs):
cf62089b 8469 insn = bpf_convert_shinfo_access(si, insn);
d9ff286a
ED		 8470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8471 si->dst_reg, si->dst_reg,
8472 bpf_target_off(struct skb_shared_info,
8473 gso_segs, 2,
8474 target_size));
8475 break;
cf62089b
WB		 8476 case offsetof(struct __sk_buff, gso_size):
8477 insn = bpf_convert_shinfo_access(si, insn);
8478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8479 si->dst_reg, si->dst_reg,
8480 bpf_target_off(struct skb_shared_info,
8481 gso_size, 2,
8482 target_size));
8483 break;
e3da08d0 8484 case offsetof(struct __sk_buff, wire_len):
c593642c 8485 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
e3da08d0
PP		 8486
8487 off = si->off;
8488 off -= offsetof(struct __sk_buff, wire_len);
8489 off += offsetof(struct sk_buff, cb);
8490 off += offsetof(struct qdisc_skb_cb, pkt_len);
8491 *target_size = 4;
8492 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
46f8bc92
MKL		 8493 break;
8494
8495 case offsetof(struct __sk_buff, sk):
8496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8497 si->dst_reg, si->src_reg,
8498 offsetof(struct sk_buff, sk));
8499 break;
9bac3d6d
AS		 8500 }
8501
8502 return insn - insn_buf;
89aa0758
AS		 8503}
8504
c64b7983
JS		 8505u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8506 const struct bpf_insn *si,
8507 struct bpf_insn *insn_buf,
8508 struct bpf_prog *prog, u32 *target_size)
61023658
DA		 8509{
8510 struct bpf_insn *insn = insn_buf;
aac3fc32 8511 int off;
61023658 8512
6b8cc1d1 8513 switch (si->off) {
61023658 8514 case offsetof(struct bpf_sock, bound_dev_if):
c593642c 8515 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
61023658
DA		 8516
8517 if (type == BPF_WRITE)
6b8cc1d1 8518 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
61023658
DA		 8519 offsetof(struct sock, sk_bound_dev_if));
8520 else
6b8cc1d1 8521 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
61023658
DA		 8522 offsetof(struct sock, sk_bound_dev_if));
8523 break;
aa4c1037 8524
482dca93 8525 case offsetof(struct bpf_sock, mark):
c593642c 8526 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
482dca93
DA		 8527
8528 if (type == BPF_WRITE)
8529 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8530 offsetof(struct sock, sk_mark));
8531 else
8532 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8533 offsetof(struct sock, sk_mark));
8534 break;
8535
8536 case offsetof(struct bpf_sock, priority):
c593642c 8537 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
482dca93
DA		 8538
8539 if (type == BPF_WRITE)
8540 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8541 offsetof(struct sock, sk_priority));
8542 else
8543 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8544 offsetof(struct sock, sk_priority));
8545 break;
8546
aa4c1037 8547 case offsetof(struct bpf_sock, family):
aa65d696
MKL		 8548 *insn++ = BPF_LDX_MEM(
8549 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8550 si->dst_reg, si->src_reg,
8551 bpf_target_off(struct sock_common,
8552 skc_family,
c593642c 8553 sizeof_field(struct sock_common,
aa65d696
MKL		 8554 skc_family),
8555 target_size));
aa4c1037
DA		 8556 break;
8557
8558 case offsetof(struct bpf_sock, type):
bf976514
MM		 8559 *insn++ = BPF_LDX_MEM(
8560 BPF_FIELD_SIZEOF(struct sock, sk_type),
8561 si->dst_reg, si->src_reg,
8562 bpf_target_off(struct sock, sk_type,
8563 sizeof_field(struct sock, sk_type),
8564 target_size));
aa4c1037
DA		 8565 break;
8566
8567 case offsetof(struct bpf_sock, protocol):
bf976514
MM		 8568 *insn++ = BPF_LDX_MEM(
8569 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8570 si->dst_reg, si->src_reg,
8571 bpf_target_off(struct sock, sk_protocol,
8572 sizeof_field(struct sock, sk_protocol),
8573 target_size));
aa4c1037 8574 break;
aac3fc32
AI		 8575
8576 case offsetof(struct bpf_sock, src_ip4):
8577 *insn++ = BPF_LDX_MEM(
8578 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8579 bpf_target_off(struct sock_common, skc_rcv_saddr,
c593642c 8580 sizeof_field(struct sock_common,
aac3fc32
AI		 8581 skc_rcv_saddr),
8582 target_size));
8583 break;
8584
aa65d696
MKL		 8585 case offsetof(struct bpf_sock, dst_ip4):
8586 *insn++ = BPF_LDX_MEM(
8587 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8588 bpf_target_off(struct sock_common, skc_daddr,
c593642c 8589 sizeof_field(struct sock_common,
aa65d696
MKL		 8590 skc_daddr),
8591 target_size));
8592 break;
8593
aac3fc32
AI		 8594 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8595#if IS_ENABLED(CONFIG_IPV6)
8596 off = si->off;
8597 off -= offsetof(struct bpf_sock, src_ip6[0]);
8598 *insn++ = BPF_LDX_MEM(
8599 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8600 bpf_target_off(
8601 struct sock_common,
8602 skc_v6_rcv_saddr.s6_addr32[0],
c593642c 8603 sizeof_field(struct sock_common,
aac3fc32
AI		 8604 skc_v6_rcv_saddr.s6_addr32[0]),
8605 target_size) + off);
8606#else
8607 (void)off;
8608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8609#endif
8610 break;
8611
aa65d696
MKL		 8612 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8613#if IS_ENABLED(CONFIG_IPV6)
8614 off = si->off;
8615 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8616 *insn++ = BPF_LDX_MEM(
8617 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8618 bpf_target_off(struct sock_common,
8619 skc_v6_daddr.s6_addr32[0],
c593642c 8620 sizeof_field(struct sock_common,
aa65d696
MKL		 8621 skc_v6_daddr.s6_addr32[0]),
8622 target_size) + off);
8623#else
8624 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8625 *target_size = 4;
8626#endif
8627 break;
8628
aac3fc32
AI		 8629 case offsetof(struct bpf_sock, src_port):
8630 *insn++ = BPF_LDX_MEM(
8631 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8632 si->dst_reg, si->src_reg,
8633 bpf_target_off(struct sock_common, skc_num,
c593642c 8634 sizeof_field(struct sock_common,
aac3fc32
AI		 8635 skc_num),
8636 target_size));
8637 break;
aa65d696
MKL		 8638
8639 case offsetof(struct bpf_sock, dst_port):
8640 *insn++ = BPF_LDX_MEM(
8641 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8642 si->dst_reg, si->src_reg,
8643 bpf_target_off(struct sock_common, skc_dport,
c593642c 8644 sizeof_field(struct sock_common,
aa65d696
MKL		 8645 skc_dport),
8646 target_size));
8647 break;
8648
8649 case offsetof(struct bpf_sock, state):
8650 *insn++ = BPF_LDX_MEM(
8651 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8652 si->dst_reg, si->src_reg,
8653 bpf_target_off(struct sock_common, skc_state,
c593642c 8654 sizeof_field(struct sock_common,
aa65d696
MKL		 8655 skc_state),
8656 target_size));
8657 break;
c3c16f2e
AN		 8658 case offsetof(struct bpf_sock, rx_queue_mapping):
8659#ifdef CONFIG_XPS
8660 *insn++ = BPF_LDX_MEM(
8661 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8662 si->dst_reg, si->src_reg,
8663 bpf_target_off(struct sock, sk_rx_queue_mapping,
8664 sizeof_field(struct sock,
8665 sk_rx_queue_mapping),
8666 target_size));
8667 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8668 1);
8669 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8670#else
8671 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8672 *target_size = 2;
8673#endif
8674 break;
61023658
DA		 8675 }
8676
8677 return insn - insn_buf;
8678}
8679
6b8cc1d1
DB		 8680static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8681 const struct bpf_insn *si,
374fb54e 8682 struct bpf_insn *insn_buf,
f96da094 8683 struct bpf_prog *prog, u32 *target_size)
374fb54e
DB		 8684{
8685 struct bpf_insn *insn = insn_buf;
8686
6b8cc1d1 8687 switch (si->off) {
374fb54e 8688 case offsetof(struct __sk_buff, ifindex):
374fb54e 8689 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6b8cc1d1 8690 si->dst_reg, si->src_reg,
374fb54e 8691 offsetof(struct sk_buff, dev));
6b8cc1d1 8692 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
f96da094
DB		 8693 bpf_target_off(struct net_device, ifindex, 4,
8694 target_size));
374fb54e
DB		 8695 break;
8696 default:
f96da094
DB		 8697 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8698 target_size);
374fb54e
DB		 8699 }
8700
8701 return insn - insn_buf;
8702}
8703
6b8cc1d1
DB		 8704static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8705 const struct bpf_insn *si,
6a773a15 8706 struct bpf_insn *insn_buf,
f96da094 8707 struct bpf_prog *prog, u32 *target_size)
6a773a15
BB		 8708{
8709 struct bpf_insn *insn = insn_buf;
8710
6b8cc1d1 8711 switch (si->off) {
6a773a15 8712 case offsetof(struct xdp_md, data):
f035a515 8713 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6b8cc1d1 8714 si->dst_reg, si->src_reg,
6a773a15
BB		 8715 offsetof(struct xdp_buff, data));
8716 break;
de8f3a83
DB		 8717 case offsetof(struct xdp_md, data_meta):
8718 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8719 si->dst_reg, si->src_reg,
8720 offsetof(struct xdp_buff, data_meta));
8721 break;
6a773a15 8722 case offsetof(struct xdp_md, data_end):
f035a515 8723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6b8cc1d1 8724 si->dst_reg, si->src_reg,
6a773a15
BB		 8725 offsetof(struct xdp_buff, data_end));
8726 break;
02dd3291
JDB		 8727 case offsetof(struct xdp_md, ingress_ifindex):
8728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8729 si->dst_reg, si->src_reg,
8730 offsetof(struct xdp_buff, rxq));
8731 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8732 si->dst_reg, si->dst_reg,
8733 offsetof(struct xdp_rxq_info, dev));
8734 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
daaf24c6 8735 offsetof(struct net_device, ifindex));
02dd3291
JDB		 8736 break;
8737 case offsetof(struct xdp_md, rx_queue_index):
8738 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8739 si->dst_reg, si->src_reg,
8740 offsetof(struct xdp_buff, rxq));
8741 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
daaf24c6
JDB		 8742 offsetof(struct xdp_rxq_info,
8743 queue_index));
02dd3291 8744 break;
64b59025
DA		 8745 case offsetof(struct xdp_md, egress_ifindex):
8746 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8747 si->dst_reg, si->src_reg,
8748 offsetof(struct xdp_buff, txq));
8749 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8750 si->dst_reg, si->dst_reg,
8751 offsetof(struct xdp_txq_info, dev));
8752 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8753 offsetof(struct net_device, ifindex));
8754 break;
6a773a15
BB		 8755 }
8756
8757 return insn - insn_buf;
8758}
8759
4fbac77d
AI		 8760/* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8761 * context Structure, F is Field in context structure that contains a pointer
8762 * to Nested Structure of type NS that has the field NF.
8763 *
8764 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8765 * sure that SIZE is not greater than actual size of S.F.NF.
8766 *
8767 * If offset OFF is provided, the load happens from that offset relative to
8768 * offset of NF.
8769 */
8770#define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8771 do { \
8772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8773 si->src_reg, offsetof(S, F)); \
8774 *insn++ = BPF_LDX_MEM( \
8775 SIZE, si->dst_reg, si->dst_reg, \
c593642c 8776 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
4fbac77d
AI		 8777 target_size) \
8778 + OFF); \
8779 } while (0)
8780
8781#define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8782 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8783 BPF_FIELD_SIZEOF(NS, NF), 0)
8784
8785/* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8786 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8787 *
4fbac77d
AI		 8788 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8789 * "register" since two registers available in convert_ctx_access are not
8790 * enough: we can't override neither SRC, since it contains value to store, nor
8791 * DST since it contains pointer to context that may be used by later
8792 * instructions. But we need a temporary place to save pointer to nested
8793 * structure whose field we want to store to.
8794 */
600c70ba 8795#define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
4fbac77d
AI		 8796 do { \
8797 int tmp_reg = BPF_REG_9; \
8798 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8799 --tmp_reg; \
8800 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8801 --tmp_reg; \
8802 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8803 offsetof(S, TF)); \
8804 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8805 si->dst_reg, offsetof(S, F)); \
600c70ba 8806 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
c593642c 8807 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
4fbac77d
AI		 8808 target_size) \
8809 + OFF); \
8810 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8811 offsetof(S, TF)); \
8812 } while (0)
8813
8814#define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8815 TF) \
8816 do { \
8817 if (type == BPF_WRITE) { \
600c70ba
SF		 8818 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8819 OFF, TF); \
4fbac77d
AI		 8820 } else { \
8821 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8822 S, NS, F, NF, SIZE, OFF); \
8823 } \
8824 } while (0)
8825
8826#define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
8827 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
8828 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
8829
8830static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
8831 const struct bpf_insn *si,
8832 struct bpf_insn *insn_buf,
8833 struct bpf_prog *prog, u32 *target_size)
8834{
7aebfa1b 8835 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
4fbac77d 8836 struct bpf_insn *insn = insn_buf;
4fbac77d
AI		 8837
8838 switch (si->off) {
8839 case offsetof(struct bpf_sock_addr, user_family):
8840 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8841 struct sockaddr, uaddr, sa_family);
8842 break;
8843
8844 case offsetof(struct bpf_sock_addr, user_ip4):
8845 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8846 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
8847 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
8848 break;
8849
8850 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8851 off = si->off;
8852 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
8853 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8854 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8855 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
8856 tmp_reg);
8857 break;
8858
8859 case offsetof(struct bpf_sock_addr, user_port):
8860 /* To get port we need to know sa_family first and then treat
8861 * sockaddr as either sockaddr_in or sockaddr_in6.
8862 * Though we can simplify since port field has same offset and
8863 * size in both structures.
8864 * Here we check this invariant and use just one of the
8865 * structures if it's true.
8866 */
8867 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
8868 offsetof(struct sockaddr_in6, sin6_port));
c593642c
PB		 8869 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
8870 sizeof_field(struct sockaddr_in6, sin6_port));
7aebfa1b
AI		 8871 /* Account for sin6_port being smaller than user_port. */
8872 port_size = min(port_size, BPF_LDST_BYTES(si));
8873 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8874 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8875 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
4fbac77d
AI		 8876 break;
8877
8878 case offsetof(struct bpf_sock_addr, family):
8879 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8880 struct sock, sk, sk_family);
8881 break;
8882
8883 case offsetof(struct bpf_sock_addr, type):
bf976514
MM		 8884 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8885 struct sock, sk, sk_type);
4fbac77d
AI		 8886 break;
8887
8888 case offsetof(struct bpf_sock_addr, protocol):
bf976514
MM		 8889 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8890 struct sock, sk, sk_protocol);
4fbac77d 8891 break;
1cedee13
AI		 8892
8893 case offsetof(struct bpf_sock_addr, msg_src_ip4):
8894 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
8895 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8896 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
8897 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
8898 break;
8899
8900 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8901 msg_src_ip6[3]):
8902 off = si->off;
8903 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
8904 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
8905 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8906 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
8907 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
8908 break;
fb85c4a7
SF		 8909 case offsetof(struct bpf_sock_addr, sk):
8910 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
8911 si->dst_reg, si->src_reg,
8912 offsetof(struct bpf_sock_addr_kern, sk));
8913 break;
4fbac77d
AI		 8914 }
8915
8916 return insn - insn_buf;
8917}
8918
40304b2a
LB		 8919static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
8920 const struct bpf_insn *si,
8921 struct bpf_insn *insn_buf,
f96da094
DB		 8922 struct bpf_prog *prog,
8923 u32 *target_size)
40304b2a
LB		 8924{
8925 struct bpf_insn *insn = insn_buf;
8926 int off;
8927
9b1f3d6e
MKL		 8928/* Helper macro for adding read access to tcp_sock or sock fields. */
8929#define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
8930 do { \
fd09af01 8931 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
c593642c
PB		 8932 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
8933 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
fd09af01
JF		 8934 if (si->dst_reg == reg || si->src_reg == reg) \
8935 reg--; \
8936 if (si->dst_reg == reg || si->src_reg == reg) \
8937 reg--; \
8938 if (si->dst_reg == si->src_reg) { \
8939 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
8940 offsetof(struct bpf_sock_ops_kern, \
8941 temp)); \
8942 fullsock_reg = reg; \
8943 jmp += 2; \
8944 } \
9b1f3d6e
MKL		 8945 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
8946 struct bpf_sock_ops_kern, \
8947 is_fullsock), \
fd09af01 8948 fullsock_reg, si->src_reg, \
9b1f3d6e
MKL		 8949 offsetof(struct bpf_sock_ops_kern, \
8950 is_fullsock)); \
fd09af01
JF		 8951 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
8952 if (si->dst_reg == si->src_reg) \
8953 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
8954 offsetof(struct bpf_sock_ops_kern, \
8955 temp)); \
9b1f3d6e
MKL		 8956 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
8957 struct bpf_sock_ops_kern, sk),\
8958 si->dst_reg, si->src_reg, \
8959 offsetof(struct bpf_sock_ops_kern, sk));\
8960 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
8961 OBJ_FIELD), \
8962 si->dst_reg, si->dst_reg, \
8963 offsetof(OBJ, OBJ_FIELD)); \
fd09af01
JF		 8964 if (si->dst_reg == si->src_reg) { \
8965 *insn++ = BPF_JMP_A(1); \
8966 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
8967 offsetof(struct bpf_sock_ops_kern, \
8968 temp)); \
8969 } \
9b1f3d6e
MKL		 8970 } while (0)
8971
84f44df6
JF		 8972#define SOCK_OPS_GET_SK() \
8973 do { \
8974 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
8975 if (si->dst_reg == reg || si->src_reg == reg) \
8976 reg--; \
8977 if (si->dst_reg == reg || si->src_reg == reg) \
8978 reg--; \
8979 if (si->dst_reg == si->src_reg) { \
8980 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
8981 offsetof(struct bpf_sock_ops_kern, \
8982 temp)); \
8983 fullsock_reg = reg; \
8984 jmp += 2; \
8985 } \
8986 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
8987 struct bpf_sock_ops_kern, \
8988 is_fullsock), \
8989 fullsock_reg, si->src_reg, \
8990 offsetof(struct bpf_sock_ops_kern, \
8991 is_fullsock)); \
8992 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
8993 if (si->dst_reg == si->src_reg) \
8994 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
8995 offsetof(struct bpf_sock_ops_kern, \
8996 temp)); \
8997 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
8998 struct bpf_sock_ops_kern, sk),\
8999 si->dst_reg, si->src_reg, \
9000 offsetof(struct bpf_sock_ops_kern, sk));\
9001 if (si->dst_reg == si->src_reg) { \
9002 *insn++ = BPF_JMP_A(1); \
9003 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9004 offsetof(struct bpf_sock_ops_kern, \
9005 temp)); \
9006 } \
9007 } while (0)
9008
9b1f3d6e
MKL		 9009#define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9010 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9011
9012/* Helper macro for adding write access to tcp_sock or sock fields.
9013 * The macro is called with two registers, dst_reg which contains a pointer
9014 * to ctx (context) and src_reg which contains the value that should be
9015 * stored. However, we need an additional register since we cannot overwrite
9016 * dst_reg because it may be used later in the program.
9017 * Instead we "borrow" one of the other register. We first save its value
9018 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9019 * it at the end of the macro.
9020 */
9021#define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9022 do { \
9023 int reg = BPF_REG_9; \
c593642c
PB		 9024 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9025 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9b1f3d6e
MKL		 9026 if (si->dst_reg == reg || si->src_reg == reg) \
9027 reg--; \
9028 if (si->dst_reg == reg || si->src_reg == reg) \
9029 reg--; \
9030 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9031 offsetof(struct bpf_sock_ops_kern, \
9032 temp)); \
9033 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9034 struct bpf_sock_ops_kern, \
9035 is_fullsock), \
9036 reg, si->dst_reg, \
9037 offsetof(struct bpf_sock_ops_kern, \
9038 is_fullsock)); \
9039 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9040 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9041 struct bpf_sock_ops_kern, sk),\
9042 reg, si->dst_reg, \
9043 offsetof(struct bpf_sock_ops_kern, sk));\
9044 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9045 reg, si->src_reg, \
9046 offsetof(OBJ, OBJ_FIELD)); \
9047 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9048 offsetof(struct bpf_sock_ops_kern, \
9049 temp)); \
9050 } while (0)
9051
9052#define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9053 do { \
9054 if (TYPE == BPF_WRITE) \
9055 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9056 else \
9057 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9058 } while (0)
9059
9b1f3d6e
MKL		 9060 if (insn > insn_buf)
9061 return insn - insn_buf;
9062
40304b2a 9063 switch (si->off) {
c9985d09
MKL		 9064 case offsetof(struct bpf_sock_ops, op):
9065 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9066 op),
9067 si->dst_reg, si->src_reg,
9068 offsetof(struct bpf_sock_ops_kern, op));
9069 break;
9070
9071 case offsetof(struct bpf_sock_ops, replylong[0]) ...
40304b2a 9072 offsetof(struct bpf_sock_ops, replylong[3]):
c593642c
PB		 9073 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9074 sizeof_field(struct bpf_sock_ops_kern, reply));
9075 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9076 sizeof_field(struct bpf_sock_ops_kern, replylong));
40304b2a 9077 off = si->off;
c9985d09
MKL		 9078 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9079 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
40304b2a
LB		 9080 if (type == BPF_WRITE)
9081 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9082 off);
9083 else
9084 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9085 off);
9086 break;
9087
9088 case offsetof(struct bpf_sock_ops, family):
c593642c 9089 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
40304b2a
LB		 9090
9091 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9092 struct bpf_sock_ops_kern, sk),
9093 si->dst_reg, si->src_reg,
9094 offsetof(struct bpf_sock_ops_kern, sk));
9095 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9096 offsetof(struct sock_common, skc_family));
9097 break;
9098
9099 case offsetof(struct bpf_sock_ops, remote_ip4):
c593642c 9100 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
40304b2a
LB		 9101
9102 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9103 struct bpf_sock_ops_kern, sk),
9104 si->dst_reg, si->src_reg,
9105 offsetof(struct bpf_sock_ops_kern, sk));
9106 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9107 offsetof(struct sock_common, skc_daddr));
9108 break;
9109
9110 case offsetof(struct bpf_sock_ops, local_ip4):
c593642c 9111 BUILD_BUG_ON(sizeof_field(struct sock_common,
303def35 9112 skc_rcv_saddr) != 4);
40304b2a
LB		 9113
9114 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9115 struct bpf_sock_ops_kern, sk),
9116 si->dst_reg, si->src_reg,
9117 offsetof(struct bpf_sock_ops_kern, sk));
9118 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9119 offsetof(struct sock_common,
9120 skc_rcv_saddr));
9121 break;
9122
9123 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9124 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9125#if IS_ENABLED(CONFIG_IPV6)
c593642c 9126 BUILD_BUG_ON(sizeof_field(struct sock_common,
40304b2a
LB		 9127 skc_v6_daddr.s6_addr32[0]) != 4);
9128
9129 off = si->off;
9130 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9131 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9132 struct bpf_sock_ops_kern, sk),
9133 si->dst_reg, si->src_reg,
9134 offsetof(struct bpf_sock_ops_kern, sk));
9135 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9136 offsetof(struct sock_common,
9137 skc_v6_daddr.s6_addr32[0]) +
9138 off);
9139#else
9140 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9141#endif
9142 break;
9143
9144 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9145 offsetof(struct bpf_sock_ops, local_ip6[3]):
9146#if IS_ENABLED(CONFIG_IPV6)
c593642c 9147 BUILD_BUG_ON(sizeof_field(struct sock_common,
40304b2a
LB		 9148 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9149
9150 off = si->off;
9151 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9152 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9153 struct bpf_sock_ops_kern, sk),
9154 si->dst_reg, si->src_reg,
9155 offsetof(struct bpf_sock_ops_kern, sk));
9156 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9157 offsetof(struct sock_common,
9158 skc_v6_rcv_saddr.s6_addr32[0]) +
9159 off);
9160#else
9161 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9162#endif
9163 break;
9164
9165 case offsetof(struct bpf_sock_ops, remote_port):
c593642c 9166 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
40304b2a
LB		 9167
9168 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9169 struct bpf_sock_ops_kern, sk),
9170 si->dst_reg, si->src_reg,
9171 offsetof(struct bpf_sock_ops_kern, sk));
9172 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9173 offsetof(struct sock_common, skc_dport));
9174#ifndef __BIG_ENDIAN_BITFIELD
9175 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9176#endif
9177 break;
9178
9179 case offsetof(struct bpf_sock_ops, local_port):
c593642c 9180 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
40304b2a
LB		 9181
9182 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9183 struct bpf_sock_ops_kern, sk),
9184 si->dst_reg, si->src_reg,
9185 offsetof(struct bpf_sock_ops_kern, sk));
9186 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9187 offsetof(struct sock_common, skc_num));
9188 break;
f19397a5
LB		 9189
9190 case offsetof(struct bpf_sock_ops, is_fullsock):
9191 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9192 struct bpf_sock_ops_kern,
9193 is_fullsock),
9194 si->dst_reg, si->src_reg,
9195 offsetof(struct bpf_sock_ops_kern,
9196 is_fullsock));
9197 break;
9198
44f0e430 9199 case offsetof(struct bpf_sock_ops, state):
c593642c 9200 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
44f0e430
LB		 9201
9202 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9203 struct bpf_sock_ops_kern, sk),
9204 si->dst_reg, si->src_reg,
9205 offsetof(struct bpf_sock_ops_kern, sk));
9206 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9207 offsetof(struct sock_common, skc_state));
9208 break;
9209
9210 case offsetof(struct bpf_sock_ops, rtt_min):
c593642c 9211 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
44f0e430
LB		 9212 sizeof(struct minmax));
9213 BUILD_BUG_ON(sizeof(struct minmax) <
9214 sizeof(struct minmax_sample));
9215
9216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9217 struct bpf_sock_ops_kern, sk),
9218 si->dst_reg, si->src_reg,
9219 offsetof(struct bpf_sock_ops_kern, sk));
9220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9221 offsetof(struct tcp_sock, rtt_min) +
c593642c 9222 sizeof_field(struct minmax_sample, t));
44f0e430
LB		 9223 break;
9224
b13d8807
LB		 9225 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9226 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9227 struct tcp_sock);
9228 break;
44f0e430 9229
44f0e430 9230 case offsetof(struct bpf_sock_ops, sk_txhash):
6f9bd3d7
LB		 9231 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9232 struct sock, type);
44f0e430 9233 break;
2377b81d
SF		 9234 case offsetof(struct bpf_sock_ops, snd_cwnd):
9235 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9236 break;
9237 case offsetof(struct bpf_sock_ops, srtt_us):
9238 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9239 break;
9240 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9241 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9242 break;
9243 case offsetof(struct bpf_sock_ops, rcv_nxt):
9244 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9245 break;
9246 case offsetof(struct bpf_sock_ops, snd_nxt):
9247 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9248 break;
9249 case offsetof(struct bpf_sock_ops, snd_una):
9250 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9251 break;
9252 case offsetof(struct bpf_sock_ops, mss_cache):
9253 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9254 break;
9255 case offsetof(struct bpf_sock_ops, ecn_flags):
9256 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9257 break;
9258 case offsetof(struct bpf_sock_ops, rate_delivered):
9259 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9260 break;
9261 case offsetof(struct bpf_sock_ops, rate_interval_us):
9262 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9263 break;
9264 case offsetof(struct bpf_sock_ops, packets_out):
9265 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9266 break;
9267 case offsetof(struct bpf_sock_ops, retrans_out):
9268 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9269 break;
9270 case offsetof(struct bpf_sock_ops, total_retrans):
9271 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9272 break;
9273 case offsetof(struct bpf_sock_ops, segs_in):
9274 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9275 break;
9276 case offsetof(struct bpf_sock_ops, data_segs_in):
9277 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9278 break;
9279 case offsetof(struct bpf_sock_ops, segs_out):
9280 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9281 break;
9282 case offsetof(struct bpf_sock_ops, data_segs_out):
9283 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9284 break;
9285 case offsetof(struct bpf_sock_ops, lost_out):
9286 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9287 break;
9288 case offsetof(struct bpf_sock_ops, sacked_out):
9289 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9290 break;
9291 case offsetof(struct bpf_sock_ops, bytes_received):
9292 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9293 break;
9294 case offsetof(struct bpf_sock_ops, bytes_acked):
9295 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9296 break;
1314ef56 9297 case offsetof(struct bpf_sock_ops, sk):
84f44df6 9298 SOCK_OPS_GET_SK();
1314ef56 9299 break;
0813a841
MKL		 9300 case offsetof(struct bpf_sock_ops, skb_data_end):
9301 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9302 skb_data_end),
9303 si->dst_reg, si->src_reg,
9304 offsetof(struct bpf_sock_ops_kern,
9305 skb_data_end));
9306 break;
9307 case offsetof(struct bpf_sock_ops, skb_data):
9308 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9309 skb),
9310 si->dst_reg, si->src_reg,
9311 offsetof(struct bpf_sock_ops_kern,
9312 skb));
9313 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9315 si->dst_reg, si->dst_reg,
9316 offsetof(struct sk_buff, data));
9317 break;
9318 case offsetof(struct bpf_sock_ops, skb_len):
9319 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9320 skb),
9321 si->dst_reg, si->src_reg,
9322 offsetof(struct bpf_sock_ops_kern,
9323 skb));
9324 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9325 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9326 si->dst_reg, si->dst_reg,
9327 offsetof(struct sk_buff, len));
9328 break;
9329 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9330 off = offsetof(struct sk_buff, cb);
9331 off += offsetof(struct tcp_skb_cb, tcp_flags);
9332 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9333 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9334 skb),
9335 si->dst_reg, si->src_reg,
9336 offsetof(struct bpf_sock_ops_kern,
9337 skb));
9338 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9339 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9340 tcp_flags),
9341 si->dst_reg, si->dst_reg, off);
9342 break;
40304b2a
LB		 9343 }
9344 return insn - insn_buf;
9345}
9346
8108a775
JF		 9347static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9348 const struct bpf_insn *si,
9349 struct bpf_insn *insn_buf,
9350 struct bpf_prog *prog, u32 *target_size)
9351{
9352 struct bpf_insn *insn = insn_buf;
9353 int off;
9354
9355 switch (si->off) {
9356 case offsetof(struct __sk_buff, data_end):
9357 off = si->off;
9358 off -= offsetof(struct __sk_buff, data_end);
9359 off += offsetof(struct sk_buff, cb);
9360 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9361 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9362 si->src_reg, off);
9363 break;
9364 default:
9365 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9366 target_size);
9367 }
9368
9369 return insn - insn_buf;
9370}
9371
4f738adb
JF		 9372static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9373 const struct bpf_insn *si,
9374 struct bpf_insn *insn_buf,
9375 struct bpf_prog *prog, u32 *target_size)
9376{
9377 struct bpf_insn *insn = insn_buf;
720e7f38 9378#if IS_ENABLED(CONFIG_IPV6)
303def35 9379 int off;
720e7f38 9380#endif
4f738adb 9381
7a69c0f2
JF		 9382 /* convert ctx uses the fact sg element is first in struct */
9383 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9384
4f738adb
JF		 9385 switch (si->off) {
9386 case offsetof(struct sk_msg_md, data):
604326b4 9387 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
4f738adb 9388 si->dst_reg, si->src_reg,
604326b4 9389 offsetof(struct sk_msg, data));
4f738adb
JF		 9390 break;
9391 case offsetof(struct sk_msg_md, data_end):
604326b4 9392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
4f738adb 9393 si->dst_reg, si->src_reg,
604326b4 9394 offsetof(struct sk_msg, data_end));
4f738adb 9395 break;
303def35 9396 case offsetof(struct sk_msg_md, family):
c593642c 9397 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
303def35
JF		 9398
9399 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9400 struct sk_msg, sk),
303def35 9401 si->dst_reg, si->src_reg,
604326b4 9402 offsetof(struct sk_msg, sk));
303def35
JF		 9403 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9404 offsetof(struct sock_common, skc_family));
9405 break;
9406
9407 case offsetof(struct sk_msg_md, remote_ip4):
c593642c 9408 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
303def35
JF		 9409
9410 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9411 struct sk_msg, sk),
303def35 9412 si->dst_reg, si->src_reg,
604326b4 9413 offsetof(struct sk_msg, sk));
303def35
JF		 9414 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9415 offsetof(struct sock_common, skc_daddr));
9416 break;
9417
9418 case offsetof(struct sk_msg_md, local_ip4):
c593642c 9419 BUILD_BUG_ON(sizeof_field(struct sock_common,
303def35
JF		 9420 skc_rcv_saddr) != 4);
9421
9422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9423 struct sk_msg, sk),
303def35 9424 si->dst_reg, si->src_reg,
604326b4 9425 offsetof(struct sk_msg, sk));
303def35
JF		 9426 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9427 offsetof(struct sock_common,
9428 skc_rcv_saddr));
9429 break;
9430
9431 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9432 offsetof(struct sk_msg_md, remote_ip6[3]):
9433#if IS_ENABLED(CONFIG_IPV6)
c593642c 9434 BUILD_BUG_ON(sizeof_field(struct sock_common,
303def35
JF		 9435 skc_v6_daddr.s6_addr32[0]) != 4);
9436
9437 off = si->off;
9438 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9440 struct sk_msg, sk),
303def35 9441 si->dst_reg, si->src_reg,
604326b4 9442 offsetof(struct sk_msg, sk));
303def35
JF		 9443 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9444 offsetof(struct sock_common,
9445 skc_v6_daddr.s6_addr32[0]) +
9446 off);
9447#else
9448 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9449#endif
9450 break;
9451
9452 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9453 offsetof(struct sk_msg_md, local_ip6[3]):
9454#if IS_ENABLED(CONFIG_IPV6)
c593642c 9455 BUILD_BUG_ON(sizeof_field(struct sock_common,
303def35
JF		 9456 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9457
9458 off = si->off;
9459 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9460 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9461 struct sk_msg, sk),
303def35 9462 si->dst_reg, si->src_reg,
604326b4 9463 offsetof(struct sk_msg, sk));
303def35
JF		 9464 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9465 offsetof(struct sock_common,
9466 skc_v6_rcv_saddr.s6_addr32[0]) +
9467 off);
9468#else
9469 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9470#endif
9471 break;
9472
9473 case offsetof(struct sk_msg_md, remote_port):
c593642c 9474 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
303def35
JF		 9475
9476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9477 struct sk_msg, sk),
303def35 9478 si->dst_reg, si->src_reg,
604326b4 9479 offsetof(struct sk_msg, sk));
303def35
JF		 9480 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9481 offsetof(struct sock_common, skc_dport));
9482#ifndef __BIG_ENDIAN_BITFIELD
9483 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9484#endif
9485 break;
9486
9487 case offsetof(struct sk_msg_md, local_port):
c593642c 9488 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
303def35
JF		 9489
9490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
604326b4 9491 struct sk_msg, sk),
303def35 9492 si->dst_reg, si->src_reg,
604326b4 9493 offsetof(struct sk_msg, sk));
303def35
JF		 9494 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9495 offsetof(struct sock_common, skc_num));
9496 break;
3bdbd022
JF		 9497
9498 case offsetof(struct sk_msg_md, size):
9499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9500 si->dst_reg, si->src_reg,
9501 offsetof(struct sk_msg_sg, size));
9502 break;
13d70f5a
JF		 9503
9504 case offsetof(struct sk_msg_md, sk):
9505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9506 si->dst_reg, si->src_reg,
9507 offsetof(struct sk_msg, sk));
9508 break;
4f738adb
JF		 9509 }
9510
9511 return insn - insn_buf;
9512}
9513
7de16e3a 9514const struct bpf_verifier_ops sk_filter_verifier_ops = {
4936e352
DB		 9515 .get_func_proto = sk_filter_func_proto,
9516 .is_valid_access = sk_filter_is_valid_access,
2492d3b8 9517 .convert_ctx_access = bpf_convert_ctx_access,
e0cea7ce 9518 .gen_ld_abs = bpf_gen_ld_abs,
89aa0758
AS		 9519};
9520
7de16e3a 9521const struct bpf_prog_ops sk_filter_prog_ops = {
61f3c964 9522 .test_run = bpf_prog_test_run_skb,
7de16e3a
JK		 9523};
9524
9525const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
4936e352
DB		 9526 .get_func_proto = tc_cls_act_func_proto,
9527 .is_valid_access = tc_cls_act_is_valid_access,
374fb54e 9528 .convert_ctx_access = tc_cls_act_convert_ctx_access,
36bbef52 9529 .gen_prologue = tc_cls_act_prologue,
e0cea7ce 9530 .gen_ld_abs = bpf_gen_ld_abs,
7de16e3a
JK		 9531};
9532
9533const struct bpf_prog_ops tc_cls_act_prog_ops = {
1cf1cae9 9534 .test_run = bpf_prog_test_run_skb,
608cd71a
AS		 9535};
9536
7de16e3a 9537const struct bpf_verifier_ops xdp_verifier_ops = {
6a773a15
BB		 9538 .get_func_proto = xdp_func_proto,
9539 .is_valid_access = xdp_is_valid_access,
9540 .convert_ctx_access = xdp_convert_ctx_access,
b09928b9 9541 .gen_prologue = bpf_noop_prologue,
7de16e3a
JK		 9542};
9543
9544const struct bpf_prog_ops xdp_prog_ops = {
1cf1cae9 9545 .test_run = bpf_prog_test_run_xdp,
6a773a15
BB		 9546};
9547
7de16e3a 9548const struct bpf_verifier_ops cg_skb_verifier_ops = {
cd339431 9549 .get_func_proto = cg_skb_func_proto,
b39b5f41 9550 .is_valid_access = cg_skb_is_valid_access,
2492d3b8 9551 .convert_ctx_access = bpf_convert_ctx_access,
7de16e3a
JK		 9552};
9553
9554const struct bpf_prog_ops cg_skb_prog_ops = {
1cf1cae9 9555 .test_run = bpf_prog_test_run_skb,
0e33661d
DM		 9556};
9557
cd3092c7
MX		 9558const struct bpf_verifier_ops lwt_in_verifier_ops = {
9559 .get_func_proto = lwt_in_func_proto,
3a0af8fd 9560 .is_valid_access = lwt_is_valid_access,
2492d3b8 9561 .convert_ctx_access = bpf_convert_ctx_access,
7de16e3a
JK		 9562};
9563
cd3092c7
MX		 9564const struct bpf_prog_ops lwt_in_prog_ops = {
9565 .test_run = bpf_prog_test_run_skb,
9566};
9567
9568const struct bpf_verifier_ops lwt_out_verifier_ops = {
9569 .get_func_proto = lwt_out_func_proto,
3a0af8fd 9570 .is_valid_access = lwt_is_valid_access,
2492d3b8 9571 .convert_ctx_access = bpf_convert_ctx_access,
7de16e3a
JK		 9572};
9573
cd3092c7 9574const struct bpf_prog_ops lwt_out_prog_ops = {
1cf1cae9 9575 .test_run = bpf_prog_test_run_skb,
3a0af8fd
TG		 9576};
9577
7de16e3a 9578const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
3a0af8fd
TG		 9579 .get_func_proto = lwt_xmit_func_proto,
9580 .is_valid_access = lwt_is_valid_access,
2492d3b8 9581 .convert_ctx_access = bpf_convert_ctx_access,
3a0af8fd 9582 .gen_prologue = tc_cls_act_prologue,
7de16e3a
JK		 9583};
9584
9585const struct bpf_prog_ops lwt_xmit_prog_ops = {
1cf1cae9 9586 .test_run = bpf_prog_test_run_skb,
3a0af8fd
TG		 9587};
9588
004d4b27
MX		 9589const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9590 .get_func_proto = lwt_seg6local_func_proto,
9591 .is_valid_access = lwt_is_valid_access,
9592 .convert_ctx_access = bpf_convert_ctx_access,
9593};
9594
9595const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9596 .test_run = bpf_prog_test_run_skb,
9597};
9598
7de16e3a 9599const struct bpf_verifier_ops cg_sock_verifier_ops = {
ae2cf1c4 9600 .get_func_proto = sock_filter_func_proto,
61023658 9601 .is_valid_access = sock_filter_is_valid_access,
c64b7983 9602 .convert_ctx_access = bpf_sock_convert_ctx_access,
61023658
DA		 9603};
9604
7de16e3a
JK		 9605const struct bpf_prog_ops cg_sock_prog_ops = {
9606};
9607
4fbac77d
AI		 9608const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9609 .get_func_proto = sock_addr_func_proto,
9610 .is_valid_access = sock_addr_is_valid_access,
9611 .convert_ctx_access = sock_addr_convert_ctx_access,
9612};
9613
9614const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9615};
9616
7de16e3a 9617const struct bpf_verifier_ops sock_ops_verifier_ops = {
8c4b4c7e 9618 .get_func_proto = sock_ops_func_proto,
40304b2a
LB		 9619 .is_valid_access = sock_ops_is_valid_access,
9620 .convert_ctx_access = sock_ops_convert_ctx_access,
9621};
9622
7de16e3a
JK		 9623const struct bpf_prog_ops sock_ops_prog_ops = {
9624};
9625
9626const struct bpf_verifier_ops sk_skb_verifier_ops = {
b005fd18
JF		 9627 .get_func_proto = sk_skb_func_proto,
9628 .is_valid_access = sk_skb_is_valid_access,
8108a775 9629 .convert_ctx_access = sk_skb_convert_ctx_access,
8a31db56 9630 .gen_prologue = sk_skb_prologue,
b005fd18
JF		 9631};
9632
7de16e3a
JK		 9633const struct bpf_prog_ops sk_skb_prog_ops = {
9634};
9635
4f738adb
JF		 9636const struct bpf_verifier_ops sk_msg_verifier_ops = {
9637 .get_func_proto = sk_msg_func_proto,
9638 .is_valid_access = sk_msg_is_valid_access,
9639 .convert_ctx_access = sk_msg_convert_ctx_access,
b09928b9 9640 .gen_prologue = bpf_noop_prologue,
4f738adb
JF		 9641};
9642
9643const struct bpf_prog_ops sk_msg_prog_ops = {
9644};
9645
d58e468b
PP		 9646const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9647 .get_func_proto = flow_dissector_func_proto,
9648 .is_valid_access = flow_dissector_is_valid_access,
089b19a9 9649 .convert_ctx_access = flow_dissector_convert_ctx_access,
d58e468b
PP		 9650};
9651
9652const struct bpf_prog_ops flow_dissector_prog_ops = {
b7a1848e 9653 .test_run = bpf_prog_test_run_flow_dissector,
d58e468b
PP		 9654};
9655
8ced425e 9656int sk_detach_filter(struct sock *sk)
55b33325
PE		 9657{
9658 int ret = -ENOENT;
9659 struct sk_filter *filter;
9660
d59577b6
VB		 9661 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9662 return -EPERM;
9663
8ced425e
HFS		 9664 filter = rcu_dereference_protected(sk->sk_filter,
9665 lockdep_sock_is_held(sk));
55b33325 9666 if (filter) {
a9b3cd7f 9667 RCU_INIT_POINTER(sk->sk_filter, NULL);
46bcf14f 9668 sk_filter_uncharge(sk, filter);
55b33325
PE		 9669 ret = 0;
9670 }
a3ea269b 9671
55b33325
PE		 9672 return ret;
9673}
8ced425e 9674EXPORT_SYMBOL_GPL(sk_detach_filter);
a8fc9277 9675
a3ea269b
DB		 9676int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9677 unsigned int len)
a8fc9277 9678{
a3ea269b 9679 struct sock_fprog_kern *fprog;
a8fc9277 9680 struct sk_filter *filter;
a3ea269b 9681 int ret = 0;
a8fc9277
PE		 9682
9683 lock_sock(sk);
9684 filter = rcu_dereference_protected(sk->sk_filter,
8ced425e 9685 lockdep_sock_is_held(sk));
a8fc9277
PE		 9686 if (!filter)
9687 goto out;
a3ea269b
DB		 9688
9689 /* We're copying the filter that has been originally attached,
93d08b69
DB		 9690 * so no conversion/decode needed anymore. eBPF programs that
9691 * have no original program cannot be dumped through this.
a3ea269b 9692 */
93d08b69 9693 ret = -EACCES;
7ae457c1 9694 fprog = filter->prog->orig_prog;
93d08b69
DB		 9695 if (!fprog)
9696 goto out;
a3ea269b
DB		 9697
9698 ret = fprog->len;
a8fc9277 9699 if (!len)
a3ea269b 9700 /* User space only enquires number of filter blocks. */
a8fc9277 9701 goto out;
a3ea269b 9702
a8fc9277 9703 ret = -EINVAL;
a3ea269b 9704 if (len < fprog->len)
a8fc9277
PE		 9705 goto out;
9706
9707 ret = -EFAULT;
009937e7 9708 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
a3ea269b 9709 goto out;
a8fc9277 9710
a3ea269b
DB		 9711 /* Instead of bytes, the API requests to return the number
9712 * of filter blocks.
9713 */
9714 ret = fprog->len;
a8fc9277
PE		 9715out:
9716 release_sock(sk);
9717 return ret;
9718}
2dbb9b9e
MKL		 9719
9720#ifdef CONFIG_INET
2dbb9b9e
MKL		 9721static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9722 struct sock_reuseport *reuse,
9723 struct sock *sk, struct sk_buff *skb,
9724 u32 hash)
9725{
9726 reuse_kern->skb = skb;
9727 reuse_kern->sk = sk;
9728 reuse_kern->selected_sk = NULL;
9729 reuse_kern->data_end = skb->data + skb_headlen(skb);
9730 reuse_kern->hash = hash;
9731 reuse_kern->reuseport_id = reuse->reuseport_id;
9732 reuse_kern->bind_inany = reuse->bind_inany;
9733}
9734
9735struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9736 struct bpf_prog *prog, struct sk_buff *skb,
9737 u32 hash)
9738{
9739 struct sk_reuseport_kern reuse_kern;
9740 enum sk_action action;
9741
9742 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9743 action = BPF_PROG_RUN(prog, &reuse_kern);
9744
9745 if (action == SK_PASS)
9746 return reuse_kern.selected_sk;
9747 else
9748 return ERR_PTR(-ECONNREFUSED);
9749}
9750
9751BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9752 struct bpf_map *, map, void *, key, u32, flags)
9753{
9fed9000 9754 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
2dbb9b9e
MKL		 9755 struct sock_reuseport *reuse;
9756 struct sock *selected_sk;
9757
9758 selected_sk = map->ops->map_lookup_elem(map, key);
9759 if (!selected_sk)
9760 return -ENOENT;
9761
9762 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9fed9000 9763 if (!reuse) {
64d85290
JS		 9764 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9765 if (sk_is_refcounted(selected_sk))
9766 sock_put(selected_sk);
9767
9fed9000
JS		 9768 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9769 * The only (!reuse) case here is - the sk has already been
9770 * unhashed (e.g. by close()), so treat it as -ENOENT.
9771 *
9772 * Other maps (e.g. sock_map) do not provide this guarantee and
9773 * the sk may never be in the reuseport group to begin with.
2dbb9b9e 9774 */
9fed9000
JS		 9775 return is_sockarray ? -ENOENT : -EINVAL;
9776 }
2dbb9b9e
MKL		 9777
9778 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
035ff358 9779 struct sock *sk = reuse_kern->sk;
2dbb9b9e 9780
2dbb9b9e
MKL		 9781 if (sk->sk_protocol != selected_sk->sk_protocol)
9782 return -EPROTOTYPE;
9783 else if (sk->sk_family != selected_sk->sk_family)
9784 return -EAFNOSUPPORT;
9785
9786 /* Catch all. Likely bound to a different sockaddr. */
9787 return -EBADFD;
9788 }
9789
9790 reuse_kern->selected_sk = selected_sk;
9791
9792 return 0;
9793}
9794
9795static const struct bpf_func_proto sk_select_reuseport_proto = {
9796 .func = sk_select_reuseport,
9797 .gpl_only = false,
9798 .ret_type = RET_INTEGER,
9799 .arg1_type = ARG_PTR_TO_CTX,
9800 .arg2_type = ARG_CONST_MAP_PTR,
9801 .arg3_type = ARG_PTR_TO_MAP_KEY,
9802 .arg4_type = ARG_ANYTHING,
9803};
9804
9805BPF_CALL_4(sk_reuseport_load_bytes,
9806 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9807 void *, to, u32, len)
9808{
9809 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
9810}
9811
9812static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
9813 .func = sk_reuseport_load_bytes,
9814 .gpl_only = false,
9815 .ret_type = RET_INTEGER,
9816 .arg1_type = ARG_PTR_TO_CTX,
9817 .arg2_type = ARG_ANYTHING,
9818 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9819 .arg4_type = ARG_CONST_SIZE,
9820};
9821
9822BPF_CALL_5(sk_reuseport_load_bytes_relative,
9823 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9824 void *, to, u32, len, u32, start_header)
9825{
9826 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
9827 len, start_header);
9828}
9829
9830static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
9831 .func = sk_reuseport_load_bytes_relative,
9832 .gpl_only = false,
9833 .ret_type = RET_INTEGER,
9834 .arg1_type = ARG_PTR_TO_CTX,
9835 .arg2_type = ARG_ANYTHING,
9836 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9837 .arg4_type = ARG_CONST_SIZE,
9838 .arg5_type = ARG_ANYTHING,
9839};
9840
9841static const struct bpf_func_proto *
9842sk_reuseport_func_proto(enum bpf_func_id func_id,
9843 const struct bpf_prog *prog)
9844{
9845 switch (func_id) {
9846 case BPF_FUNC_sk_select_reuseport:
9847 return &sk_select_reuseport_proto;
9848 case BPF_FUNC_skb_load_bytes:
9849 return &sk_reuseport_load_bytes_proto;
9850 case BPF_FUNC_skb_load_bytes_relative:
9851 return &sk_reuseport_load_bytes_relative_proto;
9852 default:
9853 return bpf_base_func_proto(func_id);
9854 }
9855}
9856
9857static bool
9858sk_reuseport_is_valid_access(int off, int size,
9859 enum bpf_access_type type,
9860 const struct bpf_prog *prog,
9861 struct bpf_insn_access_aux *info)
9862{
9863 const u32 size_default = sizeof(__u32);
9864
9865 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
9866 off % size || type != BPF_READ)
9867 return false;
9868
9869 switch (off) {
9870 case offsetof(struct sk_reuseport_md, data):
9871 info->reg_type = PTR_TO_PACKET;
9872 return size == sizeof(__u64);
9873
9874 case offsetof(struct sk_reuseport_md, data_end):
9875 info->reg_type = PTR_TO_PACKET_END;
9876 return size == sizeof(__u64);
9877
9878 case offsetof(struct sk_reuseport_md, hash):
9879 return size == size_default;
9880
9881 /* Fields that allow narrowing */
2c238177 9882 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
c593642c 9883 if (size < sizeof_field(struct sk_buff, protocol))
2dbb9b9e 9884 return false;
df561f66 9885 fallthrough;
2c238177
IL		 9886 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
9887 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
9888 case bpf_ctx_range(struct sk_reuseport_md, len):
2dbb9b9e
MKL		 9889 bpf_ctx_record_field_size(info, size_default);
9890 return bpf_ctx_narrow_access_ok(off, size, size_default);
9891
9892 default:
9893 return false;
9894 }
9895}
9896
9897#define SK_REUSEPORT_LOAD_FIELD(F) ({ \
9898 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
9899 si->dst_reg, si->src_reg, \
9900 bpf_target_off(struct sk_reuseport_kern, F, \
c593642c 9901 sizeof_field(struct sk_reuseport_kern, F), \
2dbb9b9e
MKL		 9902 target_size)); \
9903 })
9904
9905#define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
9906 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
9907 struct sk_buff, \
9908 skb, \
9909 SKB_FIELD)
9910
bf976514
MM		 9911#define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
9912 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
9913 struct sock, \
9914 sk, \
9915 SK_FIELD)
2dbb9b9e
MKL		 9916
9917static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
9918 const struct bpf_insn *si,
9919 struct bpf_insn *insn_buf,
9920 struct bpf_prog *prog,
9921 u32 *target_size)
9922{
9923 struct bpf_insn *insn = insn_buf;
9924
9925 switch (si->off) {
9926 case offsetof(struct sk_reuseport_md, data):
9927 SK_REUSEPORT_LOAD_SKB_FIELD(data);
9928 break;
9929
9930 case offsetof(struct sk_reuseport_md, len):
9931 SK_REUSEPORT_LOAD_SKB_FIELD(len);
9932 break;
9933
9934 case offsetof(struct sk_reuseport_md, eth_protocol):
9935 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
9936 break;
9937
9938 case offsetof(struct sk_reuseport_md, ip_protocol):
bf976514 9939 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
2dbb9b9e
MKL		 9940 break;
9941
9942 case offsetof(struct sk_reuseport_md, data_end):
9943 SK_REUSEPORT_LOAD_FIELD(data_end);
9944 break;
9945
9946 case offsetof(struct sk_reuseport_md, hash):
9947 SK_REUSEPORT_LOAD_FIELD(hash);
9948 break;
9949
9950 case offsetof(struct sk_reuseport_md, bind_inany):
9951 SK_REUSEPORT_LOAD_FIELD(bind_inany);
9952 break;
9953 }
9954
9955 return insn - insn_buf;
9956}
9957
9958const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
9959 .get_func_proto = sk_reuseport_func_proto,
9960 .is_valid_access = sk_reuseport_is_valid_access,
9961 .convert_ctx_access = sk_reuseport_convert_ctx_access,
9962};
9963
9964const struct bpf_prog_ops sk_reuseport_prog_ops = {
9965};
7e6897f9 9966
1559b4aa
JS		 9967DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
9968EXPORT_SYMBOL(bpf_sk_lookup_enabled);
7e6897f9 9969
e9ddbb77
JS		 9970BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
9971 struct sock *, sk, u64, flags)
7e6897f9 9972{
e9ddbb77
JS		 9973 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
9974 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
9975 return -EINVAL;
9976 if (unlikely(sk && sk_is_refcounted(sk)))
9977 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
9978 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
9979 return -ESOCKTNOSUPPORT; /* reject connected sockets */
9980
9981 /* Check if socket is suitable for packet L3/L4 protocol */
9982 if (sk && sk->sk_protocol != ctx->protocol)
9983 return -EPROTOTYPE;
9984 if (sk && sk->sk_family != ctx->family &&
9985 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
9986 return -EAFNOSUPPORT;
9987
9988 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
9989 return -EEXIST;
9990
9991 /* Select socket as lookup result */
9992 ctx->selected_sk = sk;
9993 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
9994 return 0;
7e6897f9 9995}
af7ec138 9996
e9ddbb77
JS		 9997static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
9998 .func = bpf_sk_lookup_assign,
9999 .gpl_only = false,
10000 .ret_type = RET_INTEGER,
10001 .arg1_type = ARG_PTR_TO_CTX,
10002 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10003 .arg3_type = ARG_ANYTHING,
af7ec138
YS		 10004};
10005
e9ddbb77
JS		 10006static const struct bpf_func_proto *
10007sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10008{
10009 switch (func_id) {
10010 case BPF_FUNC_perf_event_output:
10011 return &bpf_event_output_data_proto;
10012 case BPF_FUNC_sk_assign:
10013 return &bpf_sk_lookup_assign_proto;
10014 case BPF_FUNC_sk_release:
10015 return &bpf_sk_release_proto;
10016 default:
1df8f55a 10017 return bpf_sk_base_func_proto(func_id);
e9ddbb77
JS		 10018 }
10019}
af7ec138 10020
e9ddbb77
JS		 10021static bool sk_lookup_is_valid_access(int off, int size,
10022 enum bpf_access_type type,
10023 const struct bpf_prog *prog,
10024 struct bpf_insn_access_aux *info)
10025{
10026 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10027 return false;
10028 if (off % size != 0)
10029 return false;
10030 if (type != BPF_READ)
10031 return false;
10032
10033 switch (off) {
10034 case offsetof(struct bpf_sk_lookup, sk):
10035 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10036 return size == sizeof(__u64);
af7ec138 10037
e9ddbb77
JS		 10038 case bpf_ctx_range(struct bpf_sk_lookup, family):
10039 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10040 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10041 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10042 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10043 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10044 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10045 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10046 bpf_ctx_record_field_size(info, sizeof(__u32));
10047 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10048
10049 default:
10050 return false;
10051 }
10052}
10053
10054static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10055 const struct bpf_insn *si,
10056 struct bpf_insn *insn_buf,
10057 struct bpf_prog *prog,
10058 u32 *target_size)
af7ec138 10059{
e9ddbb77
JS		 10060 struct bpf_insn *insn = insn_buf;
10061
10062 switch (si->off) {
10063 case offsetof(struct bpf_sk_lookup, sk):
10064 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10065 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10066 break;
af7ec138 10067
e9ddbb77
JS		 10068 case offsetof(struct bpf_sk_lookup, family):
10069 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10070 bpf_target_off(struct bpf_sk_lookup_kern,
10071 family, 2, target_size));
10072 break;
10073
10074 case offsetof(struct bpf_sk_lookup, protocol):
10075 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10076 bpf_target_off(struct bpf_sk_lookup_kern,
10077 protocol, 2, target_size));
10078 break;
10079
10080 case offsetof(struct bpf_sk_lookup, remote_ip4):
10081 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10082 bpf_target_off(struct bpf_sk_lookup_kern,
10083 v4.saddr, 4, target_size));
10084 break;
10085
10086 case offsetof(struct bpf_sk_lookup, local_ip4):
10087 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10088 bpf_target_off(struct bpf_sk_lookup_kern,
10089 v4.daddr, 4, target_size));
10090 break;
10091
10092 case bpf_ctx_range_till(struct bpf_sk_lookup,
10093 remote_ip6[0], remote_ip6[3]): {
10094#if IS_ENABLED(CONFIG_IPV6)
10095 int off = si->off;
10096
10097 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10098 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10099 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10100 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10101 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10102 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10103#else
10104 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10105#endif
10106 break;
10107 }
10108 case bpf_ctx_range_till(struct bpf_sk_lookup,
10109 local_ip6[0], local_ip6[3]): {
10110#if IS_ENABLED(CONFIG_IPV6)
10111 int off = si->off;
10112
10113 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10114 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10115 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10116 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10117 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10118 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10119#else
10120 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10121#endif
10122 break;
af7ec138 10123 }
e9ddbb77
JS		 10124 case offsetof(struct bpf_sk_lookup, remote_port):
10125 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10126 bpf_target_off(struct bpf_sk_lookup_kern,
10127 sport, 2, target_size));
10128 break;
10129
10130 case offsetof(struct bpf_sk_lookup, local_port):
10131 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10132 bpf_target_off(struct bpf_sk_lookup_kern,
10133 dport, 2, target_size));
10134 break;
10135 }
10136
10137 return insn - insn_buf;
af7ec138 10138}
e9ddbb77
JS		 10139
10140const struct bpf_prog_ops sk_lookup_prog_ops = {
10141};
10142
10143const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10144 .get_func_proto = sk_lookup_func_proto,
10145 .is_valid_access = sk_lookup_is_valid_access,
10146 .convert_ctx_access = sk_lookup_convert_ctx_access,
10147};
10148
2dbb9b9e 10149#endif /* CONFIG_INET */
7e6897f9 10150
6a64037d 10151DEFINE_BPF_DISPATCHER(xdp)
7e6897f9
BT		 10152
10153void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10154{
6a64037d 10155 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
7e6897f9 10156}
af7ec138 10157
bc4f0548 10158#ifdef CONFIG_DEBUG_INFO_BTF
fce557bc 10159BTF_ID_LIST_GLOBAL(btf_sock_ids)
bc4f0548 10160#define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
af7ec138
YS		 10161BTF_SOCK_TYPE_xxx
10162#undef BTF_SOCK_TYPE
bc4f0548 10163#else
fce557bc 10164u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
af7ec138
YS		 10165#endif
10166
af7ec138
YS		 10167BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10168{
10169 /* tcp6_sock type is not generated in dwarf and hence btf,
10170 * trigger an explicit type generation here.
10171 */
10172 BTF_TYPE_EMIT(struct tcp6_sock);
8c33dadc 10173 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
af7ec138
YS		 10174 sk->sk_family == AF_INET6)
10175 return (unsigned long)sk;
10176
10177 return (unsigned long)NULL;
10178}
10179
10180const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10181 .func = bpf_skc_to_tcp6_sock,
10182 .gpl_only = false,
10183 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1df8f55a 10184 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
af7ec138
YS		 10185 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10186};
478cfbdf
YS		 10187
10188BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10189{
8c33dadc 10190 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
478cfbdf
YS		 10191 return (unsigned long)sk;
10192
10193 return (unsigned long)NULL;
10194}
10195
10196const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10197 .func = bpf_skc_to_tcp_sock,
10198 .gpl_only = false,
10199 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1df8f55a 10200 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
478cfbdf
YS		 10201 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10202};
10203
10204BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10205{
6b207d66 10206#ifdef CONFIG_INET
8c33dadc 10207 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
478cfbdf 10208 return (unsigned long)sk;
6b207d66 10209#endif
478cfbdf
YS		 10210
10211#if IS_BUILTIN(CONFIG_IPV6)
8c33dadc 10212 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
478cfbdf
YS		 10213 return (unsigned long)sk;
10214#endif
10215
10216 return (unsigned long)NULL;
10217}
10218
10219const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10220 .func = bpf_skc_to_tcp_timewait_sock,
10221 .gpl_only = false,
10222 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1df8f55a 10223 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
478cfbdf
YS		 10224 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10225};
10226
10227BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10228{
6b207d66 10229#ifdef CONFIG_INET
8c33dadc 10230 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
478cfbdf 10231 return (unsigned long)sk;
6b207d66 10232#endif
478cfbdf
YS		 10233
10234#if IS_BUILTIN(CONFIG_IPV6)
8c33dadc 10235 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
478cfbdf
YS		 10236 return (unsigned long)sk;
10237#endif
10238
10239 return (unsigned long)NULL;
10240}
10241
10242const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10243 .func = bpf_skc_to_tcp_request_sock,
10244 .gpl_only = false,
10245 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1df8f55a 10246 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
478cfbdf
YS		 10247 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10248};
0d4fad3e
YS		 10249
10250BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10251{
10252 /* udp6_sock type is not generated in dwarf and hence btf,
10253 * trigger an explicit type generation here.
10254 */
10255 BTF_TYPE_EMIT(struct udp6_sock);
8c33dadc 10256 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
0d4fad3e
YS		 10257 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10258 return (unsigned long)sk;
10259
10260 return (unsigned long)NULL;
10261}
10262
10263const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10264 .func = bpf_skc_to_udp6_sock,
10265 .gpl_only = false,
10266 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1df8f55a 10267 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
0d4fad3e
YS		 10268 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10269};
1df8f55a
MKL		 10270
10271static const struct bpf_func_proto *
10272bpf_sk_base_func_proto(enum bpf_func_id func_id)
10273{
10274 const struct bpf_func_proto *func;
10275
10276 switch (func_id) {
10277 case BPF_FUNC_skc_to_tcp6_sock:
10278 func = &bpf_skc_to_tcp6_sock_proto;
10279 break;
10280 case BPF_FUNC_skc_to_tcp_sock:
10281 func = &bpf_skc_to_tcp_sock_proto;
10282 break;
10283 case BPF_FUNC_skc_to_tcp_timewait_sock:
10284 func = &bpf_skc_to_tcp_timewait_sock_proto;
10285 break;
10286 case BPF_FUNC_skc_to_tcp_request_sock:
10287 func = &bpf_skc_to_tcp_request_sock_proto;
10288 break;
10289 case BPF_FUNC_skc_to_udp6_sock:
10290 func = &bpf_skc_to_udp6_sock_proto;
10291 break;
10292 default:
10293 return bpf_base_func_proto(func_id);
10294 }
10295
10296 if (!perfmon_capable())
10297 return NULL;
10298
10299 return func;
10300}
Linux 6.x block layer and io_uring tree(s)