]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/core/filter.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: fix ifindex collision during namespace removal
[mirror_ubuntu-bionic-kernel.git] / net / core / filter.c
1 /*
2 * Linux Socket Filter - Kernel level socket filtering
3 *
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
6 *
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
8 *
9 * Authors:
10 *
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 *
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
22 */
23
24 #include <linux/module.h>
25 #include <linux/types.h>
26 #include <linux/mm.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/sock_diag.h>
30 #include <linux/in.h>
31 #include <linux/inet.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_packet.h>
34 #include <linux/if_arp.h>
35 #include <linux/gfp.h>
36 #include <net/ip.h>
37 #include <net/protocol.h>
38 #include <net/netlink.h>
39 #include <linux/skbuff.h>
40 #include <net/sock.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <asm/cmpxchg.h>
47 #include <linux/filter.h>
48 #include <linux/ratelimit.h>
49 #include <linux/seccomp.h>
50 #include <linux/if_vlan.h>
51 #include <linux/bpf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <linux/bpf_trace.h>
60
61 /**
62 * sk_filter_trim_cap - run a packet through a socket filter
63 * @sk: sock associated with &sk_buff
64 * @skb: buffer to filter
65 * @cap: limit on how short the eBPF program may trim the packet
66 *
67 * Run the eBPF program and then cut skb->data to correct size returned by
68 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
69 * than pkt_len we keep whole skb->data. This is the socket level
70 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
71 * be accepted or -EPERM if the packet should be tossed.
72 *
73 */
74 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
75 {
76 int err;
77 struct sk_filter *filter;
78
79 /*
80 * If the skb was allocated from pfmemalloc reserves, only
81 * allow SOCK_MEMALLOC sockets to use it as this socket is
82 * helping free memory
83 */
84 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
85 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
86 return -ENOMEM;
87 }
88 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
89 if (err)
90 return err;
91
92 err = security_sock_rcv_skb(sk, skb);
93 if (err)
94 return err;
95
96 rcu_read_lock();
97 filter = rcu_dereference(sk->sk_filter);
98 if (filter) {
99 struct sock *save_sk = skb->sk;
100 unsigned int pkt_len;
101
102 skb->sk = sk;
103 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
104 skb->sk = save_sk;
105 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
106 }
107 rcu_read_unlock();
108
109 return err;
110 }
111 EXPORT_SYMBOL(sk_filter_trim_cap);
112
113 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb)
114 {
115 return skb_get_poff(skb);
116 }
117
118 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
119 {
120 struct nlattr *nla;
121
122 if (skb_is_nonlinear(skb))
123 return 0;
124
125 if (skb->len < sizeof(struct nlattr))
126 return 0;
127
128 if (a > skb->len - sizeof(struct nlattr))
129 return 0;
130
131 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
132 if (nla)
133 return (void *) nla - (void *) skb->data;
134
135 return 0;
136 }
137
138 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
139 {
140 struct nlattr *nla;
141
142 if (skb_is_nonlinear(skb))
143 return 0;
144
145 if (skb->len < sizeof(struct nlattr))
146 return 0;
147
148 if (a > skb->len - sizeof(struct nlattr))
149 return 0;
150
151 nla = (struct nlattr *) &skb->data[a];
152 if (nla->nla_len > skb->len - a)
153 return 0;
154
155 nla = nla_find_nested(nla, x);
156 if (nla)
157 return (void *) nla - (void *) skb->data;
158
159 return 0;
160 }
161
162 BPF_CALL_0(__get_raw_cpu_id)
163 {
164 return raw_smp_processor_id();
165 }
166
167 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
168 .func = __get_raw_cpu_id,
169 .gpl_only = false,
170 .ret_type = RET_INTEGER,
171 };
172
173 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
174 struct bpf_insn *insn_buf)
175 {
176 struct bpf_insn *insn = insn_buf;
177
178 switch (skb_field) {
179 case SKF_AD_MARK:
180 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
181
182 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
183 offsetof(struct sk_buff, mark));
184 break;
185
186 case SKF_AD_PKTTYPE:
187 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
188 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
189 #ifdef __BIG_ENDIAN_BITFIELD
190 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
191 #endif
192 break;
193
194 case SKF_AD_QUEUE:
195 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
196
197 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
198 offsetof(struct sk_buff, queue_mapping));
199 break;
200
201 case SKF_AD_VLAN_TAG:
202 case SKF_AD_VLAN_TAG_PRESENT:
203 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
204 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
205
206 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
207 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
208 offsetof(struct sk_buff, vlan_tci));
209 if (skb_field == SKF_AD_VLAN_TAG) {
210 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
211 ~VLAN_TAG_PRESENT);
212 } else {
213 /* dst_reg >>= 12 */
214 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
215 /* dst_reg &= 1 */
216 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
217 }
218 break;
219 }
220
221 return insn - insn_buf;
222 }
223
224 static bool convert_bpf_extensions(struct sock_filter *fp,
225 struct bpf_insn **insnp)
226 {
227 struct bpf_insn *insn = *insnp;
228 u32 cnt;
229
230 switch (fp->k) {
231 case SKF_AD_OFF + SKF_AD_PROTOCOL:
232 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
233
234 /* A = *(u16 *) (CTX + offsetof(protocol)) */
235 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
236 offsetof(struct sk_buff, protocol));
237 /* A = ntohs(A) [emitting a nop or swap16] */
238 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
239 break;
240
241 case SKF_AD_OFF + SKF_AD_PKTTYPE:
242 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
243 insn += cnt - 1;
244 break;
245
246 case SKF_AD_OFF + SKF_AD_IFINDEX:
247 case SKF_AD_OFF + SKF_AD_HATYPE:
248 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
249 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
250
251 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
252 BPF_REG_TMP, BPF_REG_CTX,
253 offsetof(struct sk_buff, dev));
254 /* if (tmp != 0) goto pc + 1 */
255 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
256 *insn++ = BPF_EXIT_INSN();
257 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
258 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
259 offsetof(struct net_device, ifindex));
260 else
261 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
262 offsetof(struct net_device, type));
263 break;
264
265 case SKF_AD_OFF + SKF_AD_MARK:
266 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
267 insn += cnt - 1;
268 break;
269
270 case SKF_AD_OFF + SKF_AD_RXHASH:
271 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
272
273 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
274 offsetof(struct sk_buff, hash));
275 break;
276
277 case SKF_AD_OFF + SKF_AD_QUEUE:
278 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
279 insn += cnt - 1;
280 break;
281
282 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
283 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
284 BPF_REG_A, BPF_REG_CTX, insn);
285 insn += cnt - 1;
286 break;
287
288 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
289 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
290 BPF_REG_A, BPF_REG_CTX, insn);
291 insn += cnt - 1;
292 break;
293
294 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
295 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
296
297 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
298 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
299 offsetof(struct sk_buff, vlan_proto));
300 /* A = ntohs(A) [emitting a nop or swap16] */
301 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
302 break;
303
304 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
305 case SKF_AD_OFF + SKF_AD_NLATTR:
306 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
307 case SKF_AD_OFF + SKF_AD_CPU:
308 case SKF_AD_OFF + SKF_AD_RANDOM:
309 /* arg1 = CTX */
310 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
311 /* arg2 = A */
312 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
313 /* arg3 = X */
314 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
315 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
316 switch (fp->k) {
317 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
318 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
319 break;
320 case SKF_AD_OFF + SKF_AD_NLATTR:
321 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
322 break;
323 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
324 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
325 break;
326 case SKF_AD_OFF + SKF_AD_CPU:
327 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
328 break;
329 case SKF_AD_OFF + SKF_AD_RANDOM:
330 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
331 bpf_user_rnd_init_once();
332 break;
333 }
334 break;
335
336 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
337 /* A ^= X */
338 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
339 break;
340
341 default:
342 /* This is just a dummy call to avoid letting the compiler
343 * evict __bpf_call_base() as an optimization. Placed here
344 * where no-one bothers.
345 */
346 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
347 return false;
348 }
349
350 *insnp = insn;
351 return true;
352 }
353
354 /**
355 * bpf_convert_filter - convert filter program
356 * @prog: the user passed filter program
357 * @len: the length of the user passed filter program
358 * @new_prog: allocated 'struct bpf_prog' or NULL
359 * @new_len: pointer to store length of converted program
360 *
361 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
362 * style extended BPF (eBPF).
363 * Conversion workflow:
364 *
365 * 1) First pass for calculating the new program length:
366 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
367 *
368 * 2) 2nd pass to remap in two passes: 1st pass finds new
369 * jump offsets, 2nd pass remapping:
370 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
371 */
372 static int bpf_convert_filter(struct sock_filter *prog, int len,
373 struct bpf_prog *new_prog, int *new_len)
374 {
375 int new_flen = 0, pass = 0, target, i, stack_off;
376 struct bpf_insn *new_insn, *first_insn = NULL;
377 struct sock_filter *fp;
378 int *addrs = NULL;
379 u8 bpf_src;
380
381 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
382 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
383
384 if (len <= 0 || len > BPF_MAXINSNS)
385 return -EINVAL;
386
387 if (new_prog) {
388 first_insn = new_prog->insnsi;
389 addrs = kcalloc(len, sizeof(*addrs),
390 GFP_KERNEL | __GFP_NOWARN);
391 if (!addrs)
392 return -ENOMEM;
393 }
394
395 do_pass:
396 new_insn = first_insn;
397 fp = prog;
398
399 /* Classic BPF related prologue emission. */
400 if (new_prog) {
401 /* Classic BPF expects A and X to be reset first. These need
402 * to be guaranteed to be the first two instructions.
403 */
404 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
405 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
406
407 /* All programs must keep CTX in callee saved BPF_REG_CTX.
408 * In eBPF case it's done by the compiler, here we need to
409 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
410 */
411 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
412 } else {
413 new_insn += 3;
414 }
415
416 for (i = 0; i < len; fp++, i++) {
417 struct bpf_insn tmp_insns[6] = { };
418 struct bpf_insn *insn = tmp_insns;
419
420 if (addrs)
421 addrs[i] = new_insn - first_insn;
422
423 switch (fp->code) {
424 /* All arithmetic insns and skb loads map as-is. */
425 case BPF_ALU | BPF_ADD | BPF_X:
426 case BPF_ALU | BPF_ADD | BPF_K:
427 case BPF_ALU | BPF_SUB | BPF_X:
428 case BPF_ALU | BPF_SUB | BPF_K:
429 case BPF_ALU | BPF_AND | BPF_X:
430 case BPF_ALU | BPF_AND | BPF_K:
431 case BPF_ALU | BPF_OR | BPF_X:
432 case BPF_ALU | BPF_OR | BPF_K:
433 case BPF_ALU | BPF_LSH | BPF_X:
434 case BPF_ALU | BPF_LSH | BPF_K:
435 case BPF_ALU | BPF_RSH | BPF_X:
436 case BPF_ALU | BPF_RSH | BPF_K:
437 case BPF_ALU | BPF_XOR | BPF_X:
438 case BPF_ALU | BPF_XOR | BPF_K:
439 case BPF_ALU | BPF_MUL | BPF_X:
440 case BPF_ALU | BPF_MUL | BPF_K:
441 case BPF_ALU | BPF_DIV | BPF_X:
442 case BPF_ALU | BPF_DIV | BPF_K:
443 case BPF_ALU | BPF_MOD | BPF_X:
444 case BPF_ALU | BPF_MOD | BPF_K:
445 case BPF_ALU | BPF_NEG:
446 case BPF_LD | BPF_ABS | BPF_W:
447 case BPF_LD | BPF_ABS | BPF_H:
448 case BPF_LD | BPF_ABS | BPF_B:
449 case BPF_LD | BPF_IND | BPF_W:
450 case BPF_LD | BPF_IND | BPF_H:
451 case BPF_LD | BPF_IND | BPF_B:
452 /* Check for overloaded BPF extension and
453 * directly convert it if found, otherwise
454 * just move on with mapping.
455 */
456 if (BPF_CLASS(fp->code) == BPF_LD &&
457 BPF_MODE(fp->code) == BPF_ABS &&
458 convert_bpf_extensions(fp, &insn))
459 break;
460
461 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
462 fp->code == (BPF_ALU | BPF_MOD | BPF_X))
463 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
464
465 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
466 break;
467
468 /* Jump transformation cannot use BPF block macros
469 * everywhere as offset calculation and target updates
470 * require a bit more work than the rest, i.e. jump
471 * opcodes map as-is, but offsets need adjustment.
472 */
473
474 #define BPF_EMIT_JMP \
475 do { \
476 if (target >= len || target < 0) \
477 goto err; \
478 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
479 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
480 insn->off -= insn - tmp_insns; \
481 } while (0)
482
483 case BPF_JMP | BPF_JA:
484 target = i + fp->k + 1;
485 insn->code = fp->code;
486 BPF_EMIT_JMP;
487 break;
488
489 case BPF_JMP | BPF_JEQ | BPF_K:
490 case BPF_JMP | BPF_JEQ | BPF_X:
491 case BPF_JMP | BPF_JSET | BPF_K:
492 case BPF_JMP | BPF_JSET | BPF_X:
493 case BPF_JMP | BPF_JGT | BPF_K:
494 case BPF_JMP | BPF_JGT | BPF_X:
495 case BPF_JMP | BPF_JGE | BPF_K:
496 case BPF_JMP | BPF_JGE | BPF_X:
497 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
498 /* BPF immediates are signed, zero extend
499 * immediate into tmp register and use it
500 * in compare insn.
501 */
502 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
503
504 insn->dst_reg = BPF_REG_A;
505 insn->src_reg = BPF_REG_TMP;
506 bpf_src = BPF_X;
507 } else {
508 insn->dst_reg = BPF_REG_A;
509 insn->imm = fp->k;
510 bpf_src = BPF_SRC(fp->code);
511 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
512 }
513
514 /* Common case where 'jump_false' is next insn. */
515 if (fp->jf == 0) {
516 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
517 target = i + fp->jt + 1;
518 BPF_EMIT_JMP;
519 break;
520 }
521
522 /* Convert some jumps when 'jump_true' is next insn. */
523 if (fp->jt == 0) {
524 switch (BPF_OP(fp->code)) {
525 case BPF_JEQ:
526 insn->code = BPF_JMP | BPF_JNE | bpf_src;
527 break;
528 case BPF_JGT:
529 insn->code = BPF_JMP | BPF_JLE | bpf_src;
530 break;
531 case BPF_JGE:
532 insn->code = BPF_JMP | BPF_JLT | bpf_src;
533 break;
534 default:
535 goto jmp_rest;
536 }
537
538 target = i + fp->jf + 1;
539 BPF_EMIT_JMP;
540 break;
541 }
542 jmp_rest:
543 /* Other jumps are mapped into two insns: Jxx and JA. */
544 target = i + fp->jt + 1;
545 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
546 BPF_EMIT_JMP;
547 insn++;
548
549 insn->code = BPF_JMP | BPF_JA;
550 target = i + fp->jf + 1;
551 BPF_EMIT_JMP;
552 break;
553
554 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
555 case BPF_LDX | BPF_MSH | BPF_B:
556 /* tmp = A */
557 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
558 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
559 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
560 /* A &= 0xf */
561 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
562 /* A <<= 2 */
563 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
564 /* X = A */
565 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
566 /* A = tmp */
567 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
568 break;
569
570 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
571 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
572 */
573 case BPF_RET | BPF_A:
574 case BPF_RET | BPF_K:
575 if (BPF_RVAL(fp->code) == BPF_K)
576 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
577 0, fp->k);
578 *insn = BPF_EXIT_INSN();
579 break;
580
581 /* Store to stack. */
582 case BPF_ST:
583 case BPF_STX:
584 stack_off = fp->k * 4 + 4;
585 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
586 BPF_ST ? BPF_REG_A : BPF_REG_X,
587 -stack_off);
588 /* check_load_and_stores() verifies that classic BPF can
589 * load from stack only after write, so tracking
590 * stack_depth for ST|STX insns is enough
591 */
592 if (new_prog && new_prog->aux->stack_depth < stack_off)
593 new_prog->aux->stack_depth = stack_off;
594 break;
595
596 /* Load from stack. */
597 case BPF_LD | BPF_MEM:
598 case BPF_LDX | BPF_MEM:
599 stack_off = fp->k * 4 + 4;
600 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
601 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
602 -stack_off);
603 break;
604
605 /* A = K or X = K */
606 case BPF_LD | BPF_IMM:
607 case BPF_LDX | BPF_IMM:
608 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
609 BPF_REG_A : BPF_REG_X, fp->k);
610 break;
611
612 /* X = A */
613 case BPF_MISC | BPF_TAX:
614 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
615 break;
616
617 /* A = X */
618 case BPF_MISC | BPF_TXA:
619 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
620 break;
621
622 /* A = skb->len or X = skb->len */
623 case BPF_LD | BPF_W | BPF_LEN:
624 case BPF_LDX | BPF_W | BPF_LEN:
625 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
626 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
627 offsetof(struct sk_buff, len));
628 break;
629
630 /* Access seccomp_data fields. */
631 case BPF_LDX | BPF_ABS | BPF_W:
632 /* A = *(u32 *) (ctx + K) */
633 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
634 break;
635
636 /* Unknown instruction. */
637 default:
638 goto err;
639 }
640
641 insn++;
642 if (new_prog)
643 memcpy(new_insn, tmp_insns,
644 sizeof(*insn) * (insn - tmp_insns));
645 new_insn += insn - tmp_insns;
646 }
647
648 if (!new_prog) {
649 /* Only calculating new length. */
650 *new_len = new_insn - first_insn;
651 return 0;
652 }
653
654 pass++;
655 if (new_flen != new_insn - first_insn) {
656 new_flen = new_insn - first_insn;
657 if (pass > 2)
658 goto err;
659 goto do_pass;
660 }
661
662 kfree(addrs);
663 BUG_ON(*new_len != new_flen);
664 return 0;
665 err:
666 kfree(addrs);
667 return -EINVAL;
668 }
669
670 /* Security:
671 *
672 * As we dont want to clear mem[] array for each packet going through
673 * __bpf_prog_run(), we check that filter loaded by user never try to read
674 * a cell if not previously written, and we check all branches to be sure
675 * a malicious user doesn't try to abuse us.
676 */
677 static int check_load_and_stores(const struct sock_filter *filter, int flen)
678 {
679 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
680 int pc, ret = 0;
681
682 BUILD_BUG_ON(BPF_MEMWORDS > 16);
683
684 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
685 if (!masks)
686 return -ENOMEM;
687
688 memset(masks, 0xff, flen * sizeof(*masks));
689
690 for (pc = 0; pc < flen; pc++) {
691 memvalid &= masks[pc];
692
693 switch (filter[pc].code) {
694 case BPF_ST:
695 case BPF_STX:
696 memvalid |= (1 << filter[pc].k);
697 break;
698 case BPF_LD | BPF_MEM:
699 case BPF_LDX | BPF_MEM:
700 if (!(memvalid & (1 << filter[pc].k))) {
701 ret = -EINVAL;
702 goto error;
703 }
704 break;
705 case BPF_JMP | BPF_JA:
706 /* A jump must set masks on target */
707 masks[pc + 1 + filter[pc].k] &= memvalid;
708 memvalid = ~0;
709 break;
710 case BPF_JMP | BPF_JEQ | BPF_K:
711 case BPF_JMP | BPF_JEQ | BPF_X:
712 case BPF_JMP | BPF_JGE | BPF_K:
713 case BPF_JMP | BPF_JGE | BPF_X:
714 case BPF_JMP | BPF_JGT | BPF_K:
715 case BPF_JMP | BPF_JGT | BPF_X:
716 case BPF_JMP | BPF_JSET | BPF_K:
717 case BPF_JMP | BPF_JSET | BPF_X:
718 /* A jump must set masks on targets */
719 masks[pc + 1 + filter[pc].jt] &= memvalid;
720 masks[pc + 1 + filter[pc].jf] &= memvalid;
721 memvalid = ~0;
722 break;
723 }
724 }
725 error:
726 kfree(masks);
727 return ret;
728 }
729
730 static bool chk_code_allowed(u16 code_to_probe)
731 {
732 static const bool codes[] = {
733 /* 32 bit ALU operations */
734 [BPF_ALU | BPF_ADD | BPF_K] = true,
735 [BPF_ALU | BPF_ADD | BPF_X] = true,
736 [BPF_ALU | BPF_SUB | BPF_K] = true,
737 [BPF_ALU | BPF_SUB | BPF_X] = true,
738 [BPF_ALU | BPF_MUL | BPF_K] = true,
739 [BPF_ALU | BPF_MUL | BPF_X] = true,
740 [BPF_ALU | BPF_DIV | BPF_K] = true,
741 [BPF_ALU | BPF_DIV | BPF_X] = true,
742 [BPF_ALU | BPF_MOD | BPF_K] = true,
743 [BPF_ALU | BPF_MOD | BPF_X] = true,
744 [BPF_ALU | BPF_AND | BPF_K] = true,
745 [BPF_ALU | BPF_AND | BPF_X] = true,
746 [BPF_ALU | BPF_OR | BPF_K] = true,
747 [BPF_ALU | BPF_OR | BPF_X] = true,
748 [BPF_ALU | BPF_XOR | BPF_K] = true,
749 [BPF_ALU | BPF_XOR | BPF_X] = true,
750 [BPF_ALU | BPF_LSH | BPF_K] = true,
751 [BPF_ALU | BPF_LSH | BPF_X] = true,
752 [BPF_ALU | BPF_RSH | BPF_K] = true,
753 [BPF_ALU | BPF_RSH | BPF_X] = true,
754 [BPF_ALU | BPF_NEG] = true,
755 /* Load instructions */
756 [BPF_LD | BPF_W | BPF_ABS] = true,
757 [BPF_LD | BPF_H | BPF_ABS] = true,
758 [BPF_LD | BPF_B | BPF_ABS] = true,
759 [BPF_LD | BPF_W | BPF_LEN] = true,
760 [BPF_LD | BPF_W | BPF_IND] = true,
761 [BPF_LD | BPF_H | BPF_IND] = true,
762 [BPF_LD | BPF_B | BPF_IND] = true,
763 [BPF_LD | BPF_IMM] = true,
764 [BPF_LD | BPF_MEM] = true,
765 [BPF_LDX | BPF_W | BPF_LEN] = true,
766 [BPF_LDX | BPF_B | BPF_MSH] = true,
767 [BPF_LDX | BPF_IMM] = true,
768 [BPF_LDX | BPF_MEM] = true,
769 /* Store instructions */
770 [BPF_ST] = true,
771 [BPF_STX] = true,
772 /* Misc instructions */
773 [BPF_MISC | BPF_TAX] = true,
774 [BPF_MISC | BPF_TXA] = true,
775 /* Return instructions */
776 [BPF_RET | BPF_K] = true,
777 [BPF_RET | BPF_A] = true,
778 /* Jump instructions */
779 [BPF_JMP | BPF_JA] = true,
780 [BPF_JMP | BPF_JEQ | BPF_K] = true,
781 [BPF_JMP | BPF_JEQ | BPF_X] = true,
782 [BPF_JMP | BPF_JGE | BPF_K] = true,
783 [BPF_JMP | BPF_JGE | BPF_X] = true,
784 [BPF_JMP | BPF_JGT | BPF_K] = true,
785 [BPF_JMP | BPF_JGT | BPF_X] = true,
786 [BPF_JMP | BPF_JSET | BPF_K] = true,
787 [BPF_JMP | BPF_JSET | BPF_X] = true,
788 };
789
790 if (code_to_probe >= ARRAY_SIZE(codes))
791 return false;
792
793 return codes[code_to_probe];
794 }
795
796 static bool bpf_check_basics_ok(const struct sock_filter *filter,
797 unsigned int flen)
798 {
799 if (filter == NULL)
800 return false;
801 if (flen == 0 || flen > BPF_MAXINSNS)
802 return false;
803
804 return true;
805 }
806
807 /**
808 * bpf_check_classic - verify socket filter code
809 * @filter: filter to verify
810 * @flen: length of filter
811 *
812 * Check the user's filter code. If we let some ugly
813 * filter code slip through kaboom! The filter must contain
814 * no references or jumps that are out of range, no illegal
815 * instructions, and must end with a RET instruction.
816 *
817 * All jumps are forward as they are not signed.
818 *
819 * Returns 0 if the rule set is legal or -EINVAL if not.
820 */
821 static int bpf_check_classic(const struct sock_filter *filter,
822 unsigned int flen)
823 {
824 bool anc_found;
825 int pc;
826
827 /* Check the filter code now */
828 for (pc = 0; pc < flen; pc++) {
829 const struct sock_filter *ftest = &filter[pc];
830
831 /* May we actually operate on this code? */
832 if (!chk_code_allowed(ftest->code))
833 return -EINVAL;
834
835 /* Some instructions need special checks */
836 switch (ftest->code) {
837 case BPF_ALU | BPF_DIV | BPF_K:
838 case BPF_ALU | BPF_MOD | BPF_K:
839 /* Check for division by zero */
840 if (ftest->k == 0)
841 return -EINVAL;
842 break;
843 case BPF_ALU | BPF_LSH | BPF_K:
844 case BPF_ALU | BPF_RSH | BPF_K:
845 if (ftest->k >= 32)
846 return -EINVAL;
847 break;
848 case BPF_LD | BPF_MEM:
849 case BPF_LDX | BPF_MEM:
850 case BPF_ST:
851 case BPF_STX:
852 /* Check for invalid memory addresses */
853 if (ftest->k >= BPF_MEMWORDS)
854 return -EINVAL;
855 break;
856 case BPF_JMP | BPF_JA:
857 /* Note, the large ftest->k might cause loops.
858 * Compare this with conditional jumps below,
859 * where offsets are limited. --ANK (981016)
860 */
861 if (ftest->k >= (unsigned int)(flen - pc - 1))
862 return -EINVAL;
863 break;
864 case BPF_JMP | BPF_JEQ | BPF_K:
865 case BPF_JMP | BPF_JEQ | BPF_X:
866 case BPF_JMP | BPF_JGE | BPF_K:
867 case BPF_JMP | BPF_JGE | BPF_X:
868 case BPF_JMP | BPF_JGT | BPF_K:
869 case BPF_JMP | BPF_JGT | BPF_X:
870 case BPF_JMP | BPF_JSET | BPF_K:
871 case BPF_JMP | BPF_JSET | BPF_X:
872 /* Both conditionals must be safe */
873 if (pc + ftest->jt + 1 >= flen ||
874 pc + ftest->jf + 1 >= flen)
875 return -EINVAL;
876 break;
877 case BPF_LD | BPF_W | BPF_ABS:
878 case BPF_LD | BPF_H | BPF_ABS:
879 case BPF_LD | BPF_B | BPF_ABS:
880 anc_found = false;
881 if (bpf_anc_helper(ftest) & BPF_ANC)
882 anc_found = true;
883 /* Ancillary operation unknown or unsupported */
884 if (anc_found == false && ftest->k >= SKF_AD_OFF)
885 return -EINVAL;
886 }
887 }
888
889 /* Last instruction must be a RET code */
890 switch (filter[flen - 1].code) {
891 case BPF_RET | BPF_K:
892 case BPF_RET | BPF_A:
893 return check_load_and_stores(filter, flen);
894 }
895
896 return -EINVAL;
897 }
898
899 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
900 const struct sock_fprog *fprog)
901 {
902 unsigned int fsize = bpf_classic_proglen(fprog);
903 struct sock_fprog_kern *fkprog;
904
905 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
906 if (!fp->orig_prog)
907 return -ENOMEM;
908
909 fkprog = fp->orig_prog;
910 fkprog->len = fprog->len;
911
912 fkprog->filter = kmemdup(fp->insns, fsize,
913 GFP_KERNEL | __GFP_NOWARN);
914 if (!fkprog->filter) {
915 kfree(fp->orig_prog);
916 return -ENOMEM;
917 }
918
919 return 0;
920 }
921
922 static void bpf_release_orig_filter(struct bpf_prog *fp)
923 {
924 struct sock_fprog_kern *fprog = fp->orig_prog;
925
926 if (fprog) {
927 kfree(fprog->filter);
928 kfree(fprog);
929 }
930 }
931
932 static void __bpf_prog_release(struct bpf_prog *prog)
933 {
934 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
935 bpf_prog_put(prog);
936 } else {
937 bpf_release_orig_filter(prog);
938 bpf_prog_free(prog);
939 }
940 }
941
942 static void __sk_filter_release(struct sk_filter *fp)
943 {
944 __bpf_prog_release(fp->prog);
945 kfree(fp);
946 }
947
948 /**
949 * sk_filter_release_rcu - Release a socket filter by rcu_head
950 * @rcu: rcu_head that contains the sk_filter to free
951 */
952 static void sk_filter_release_rcu(struct rcu_head *rcu)
953 {
954 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
955
956 __sk_filter_release(fp);
957 }
958
959 /**
960 * sk_filter_release - release a socket filter
961 * @fp: filter to remove
962 *
963 * Remove a filter from a socket and release its resources.
964 */
965 static void sk_filter_release(struct sk_filter *fp)
966 {
967 if (refcount_dec_and_test(&fp->refcnt))
968 call_rcu(&fp->rcu, sk_filter_release_rcu);
969 }
970
971 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
972 {
973 u32 filter_size = bpf_prog_size(fp->prog->len);
974
975 atomic_sub(filter_size, &sk->sk_omem_alloc);
976 sk_filter_release(fp);
977 }
978
979 /* try to charge the socket memory if there is space available
980 * return true on success
981 */
982 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
983 {
984 u32 filter_size = bpf_prog_size(fp->prog->len);
985
986 /* same check as in sock_kmalloc() */
987 if (filter_size <= sysctl_optmem_max &&
988 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
989 atomic_add(filter_size, &sk->sk_omem_alloc);
990 return true;
991 }
992 return false;
993 }
994
995 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
996 {
997 if (!refcount_inc_not_zero(&fp->refcnt))
998 return false;
999
1000 if (!__sk_filter_charge(sk, fp)) {
1001 sk_filter_release(fp);
1002 return false;
1003 }
1004 return true;
1005 }
1006
1007 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1008 {
1009 struct sock_filter *old_prog;
1010 struct bpf_prog *old_fp;
1011 int err, new_len, old_len = fp->len;
1012
1013 /* We are free to overwrite insns et al right here as it
1014 * won't be used at this point in time anymore internally
1015 * after the migration to the internal BPF instruction
1016 * representation.
1017 */
1018 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1019 sizeof(struct bpf_insn));
1020
1021 /* Conversion cannot happen on overlapping memory areas,
1022 * so we need to keep the user BPF around until the 2nd
1023 * pass. At this time, the user BPF is stored in fp->insns.
1024 */
1025 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1026 GFP_KERNEL | __GFP_NOWARN);
1027 if (!old_prog) {
1028 err = -ENOMEM;
1029 goto out_err;
1030 }
1031
1032 /* 1st pass: calculate the new program length. */
1033 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
1034 if (err)
1035 goto out_err_free;
1036
1037 /* Expand fp for appending the new filter representation. */
1038 old_fp = fp;
1039 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1040 if (!fp) {
1041 /* The old_fp is still around in case we couldn't
1042 * allocate new memory, so uncharge on that one.
1043 */
1044 fp = old_fp;
1045 err = -ENOMEM;
1046 goto out_err_free;
1047 }
1048
1049 fp->len = new_len;
1050
1051 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1052 err = bpf_convert_filter(old_prog, old_len, fp, &new_len);
1053 if (err)
1054 /* 2nd bpf_convert_filter() can fail only if it fails
1055 * to allocate memory, remapping must succeed. Note,
1056 * that at this time old_fp has already been released
1057 * by krealloc().
1058 */
1059 goto out_err_free;
1060
1061 fp = bpf_prog_select_runtime(fp, &err);
1062 if (err)
1063 goto out_err_free;
1064
1065 kfree(old_prog);
1066 return fp;
1067
1068 out_err_free:
1069 kfree(old_prog);
1070 out_err:
1071 __bpf_prog_release(fp);
1072 return ERR_PTR(err);
1073 }
1074
1075 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1076 bpf_aux_classic_check_t trans)
1077 {
1078 int err;
1079
1080 fp->bpf_func = NULL;
1081 fp->jited = 0;
1082
1083 err = bpf_check_classic(fp->insns, fp->len);
1084 if (err) {
1085 __bpf_prog_release(fp);
1086 return ERR_PTR(err);
1087 }
1088
1089 /* There might be additional checks and transformations
1090 * needed on classic filters, f.e. in case of seccomp.
1091 */
1092 if (trans) {
1093 err = trans(fp->insns, fp->len);
1094 if (err) {
1095 __bpf_prog_release(fp);
1096 return ERR_PTR(err);
1097 }
1098 }
1099
1100 /* Probe if we can JIT compile the filter and if so, do
1101 * the compilation of the filter.
1102 */
1103 bpf_jit_compile(fp);
1104
1105 /* JIT compiler couldn't process this filter, so do the
1106 * internal BPF translation for the optimized interpreter.
1107 */
1108 if (!fp->jited)
1109 fp = bpf_migrate_filter(fp);
1110
1111 return fp;
1112 }
1113
1114 /**
1115 * bpf_prog_create - create an unattached filter
1116 * @pfp: the unattached filter that is created
1117 * @fprog: the filter program
1118 *
1119 * Create a filter independent of any socket. We first run some
1120 * sanity checks on it to make sure it does not explode on us later.
1121 * If an error occurs or there is insufficient memory for the filter
1122 * a negative errno code is returned. On success the return is zero.
1123 */
1124 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1125 {
1126 unsigned int fsize = bpf_classic_proglen(fprog);
1127 struct bpf_prog *fp;
1128
1129 /* Make sure new filter is there and in the right amounts. */
1130 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1131 return -EINVAL;
1132
1133 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1134 if (!fp)
1135 return -ENOMEM;
1136
1137 memcpy(fp->insns, fprog->filter, fsize);
1138
1139 fp->len = fprog->len;
1140 /* Since unattached filters are not copied back to user
1141 * space through sk_get_filter(), we do not need to hold
1142 * a copy here, and can spare us the work.
1143 */
1144 fp->orig_prog = NULL;
1145
1146 /* bpf_prepare_filter() already takes care of freeing
1147 * memory in case something goes wrong.
1148 */
1149 fp = bpf_prepare_filter(fp, NULL);
1150 if (IS_ERR(fp))
1151 return PTR_ERR(fp);
1152
1153 *pfp = fp;
1154 return 0;
1155 }
1156 EXPORT_SYMBOL_GPL(bpf_prog_create);
1157
1158 /**
1159 * bpf_prog_create_from_user - create an unattached filter from user buffer
1160 * @pfp: the unattached filter that is created
1161 * @fprog: the filter program
1162 * @trans: post-classic verifier transformation handler
1163 * @save_orig: save classic BPF program
1164 *
1165 * This function effectively does the same as bpf_prog_create(), only
1166 * that it builds up its insns buffer from user space provided buffer.
1167 * It also allows for passing a bpf_aux_classic_check_t handler.
1168 */
1169 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1170 bpf_aux_classic_check_t trans, bool save_orig)
1171 {
1172 unsigned int fsize = bpf_classic_proglen(fprog);
1173 struct bpf_prog *fp;
1174 int err;
1175
1176 /* Make sure new filter is there and in the right amounts. */
1177 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1178 return -EINVAL;
1179
1180 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1181 if (!fp)
1182 return -ENOMEM;
1183
1184 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1185 __bpf_prog_free(fp);
1186 return -EFAULT;
1187 }
1188
1189 fp->len = fprog->len;
1190 fp->orig_prog = NULL;
1191
1192 if (save_orig) {
1193 err = bpf_prog_store_orig_filter(fp, fprog);
1194 if (err) {
1195 __bpf_prog_free(fp);
1196 return -ENOMEM;
1197 }
1198 }
1199
1200 /* bpf_prepare_filter() already takes care of freeing
1201 * memory in case something goes wrong.
1202 */
1203 fp = bpf_prepare_filter(fp, trans);
1204 if (IS_ERR(fp))
1205 return PTR_ERR(fp);
1206
1207 *pfp = fp;
1208 return 0;
1209 }
1210 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1211
1212 void bpf_prog_destroy(struct bpf_prog *fp)
1213 {
1214 __bpf_prog_release(fp);
1215 }
1216 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1217
1218 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1219 {
1220 struct sk_filter *fp, *old_fp;
1221
1222 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1223 if (!fp)
1224 return -ENOMEM;
1225
1226 fp->prog = prog;
1227
1228 if (!__sk_filter_charge(sk, fp)) {
1229 kfree(fp);
1230 return -ENOMEM;
1231 }
1232 refcount_set(&fp->refcnt, 1);
1233
1234 old_fp = rcu_dereference_protected(sk->sk_filter,
1235 lockdep_sock_is_held(sk));
1236 rcu_assign_pointer(sk->sk_filter, fp);
1237
1238 if (old_fp)
1239 sk_filter_uncharge(sk, old_fp);
1240
1241 return 0;
1242 }
1243
1244 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1245 {
1246 struct bpf_prog *old_prog;
1247 int err;
1248
1249 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1250 return -ENOMEM;
1251
1252 if (sk_unhashed(sk) && sk->sk_reuseport) {
1253 err = reuseport_alloc(sk);
1254 if (err)
1255 return err;
1256 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1257 /* The socket wasn't bound with SO_REUSEPORT */
1258 return -EINVAL;
1259 }
1260
1261 old_prog = reuseport_attach_prog(sk, prog);
1262 if (old_prog)
1263 bpf_prog_destroy(old_prog);
1264
1265 return 0;
1266 }
1267
1268 static
1269 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1270 {
1271 unsigned int fsize = bpf_classic_proglen(fprog);
1272 struct bpf_prog *prog;
1273 int err;
1274
1275 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1276 return ERR_PTR(-EPERM);
1277
1278 /* Make sure new filter is there and in the right amounts. */
1279 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1280 return ERR_PTR(-EINVAL);
1281
1282 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1283 if (!prog)
1284 return ERR_PTR(-ENOMEM);
1285
1286 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1287 __bpf_prog_free(prog);
1288 return ERR_PTR(-EFAULT);
1289 }
1290
1291 prog->len = fprog->len;
1292
1293 err = bpf_prog_store_orig_filter(prog, fprog);
1294 if (err) {
1295 __bpf_prog_free(prog);
1296 return ERR_PTR(-ENOMEM);
1297 }
1298
1299 /* bpf_prepare_filter() already takes care of freeing
1300 * memory in case something goes wrong.
1301 */
1302 return bpf_prepare_filter(prog, NULL);
1303 }
1304
1305 /**
1306 * sk_attach_filter - attach a socket filter
1307 * @fprog: the filter program
1308 * @sk: the socket to use
1309 *
1310 * Attach the user's filter code. We first run some sanity checks on
1311 * it to make sure it does not explode on us later. If an error
1312 * occurs or there is insufficient memory for the filter a negative
1313 * errno code is returned. On success the return is zero.
1314 */
1315 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1316 {
1317 struct bpf_prog *prog = __get_filter(fprog, sk);
1318 int err;
1319
1320 if (IS_ERR(prog))
1321 return PTR_ERR(prog);
1322
1323 err = __sk_attach_prog(prog, sk);
1324 if (err < 0) {
1325 __bpf_prog_release(prog);
1326 return err;
1327 }
1328
1329 return 0;
1330 }
1331 EXPORT_SYMBOL_GPL(sk_attach_filter);
1332
1333 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1334 {
1335 struct bpf_prog *prog = __get_filter(fprog, sk);
1336 int err;
1337
1338 if (IS_ERR(prog))
1339 return PTR_ERR(prog);
1340
1341 err = __reuseport_attach_prog(prog, sk);
1342 if (err < 0) {
1343 __bpf_prog_release(prog);
1344 return err;
1345 }
1346
1347 return 0;
1348 }
1349
1350 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1351 {
1352 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1353 return ERR_PTR(-EPERM);
1354
1355 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1356 }
1357
1358 int sk_attach_bpf(u32 ufd, struct sock *sk)
1359 {
1360 struct bpf_prog *prog = __get_bpf(ufd, sk);
1361 int err;
1362
1363 if (IS_ERR(prog))
1364 return PTR_ERR(prog);
1365
1366 err = __sk_attach_prog(prog, sk);
1367 if (err < 0) {
1368 bpf_prog_put(prog);
1369 return err;
1370 }
1371
1372 return 0;
1373 }
1374
1375 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1376 {
1377 struct bpf_prog *prog = __get_bpf(ufd, sk);
1378 int err;
1379
1380 if (IS_ERR(prog))
1381 return PTR_ERR(prog);
1382
1383 err = __reuseport_attach_prog(prog, sk);
1384 if (err < 0) {
1385 bpf_prog_put(prog);
1386 return err;
1387 }
1388
1389 return 0;
1390 }
1391
1392 struct bpf_scratchpad {
1393 union {
1394 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1395 u8 buff[MAX_BPF_STACK];
1396 };
1397 };
1398
1399 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1400
1401 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1402 unsigned int write_len)
1403 {
1404 return skb_ensure_writable(skb, write_len);
1405 }
1406
1407 static inline int bpf_try_make_writable(struct sk_buff *skb,
1408 unsigned int write_len)
1409 {
1410 int err = __bpf_try_make_writable(skb, write_len);
1411
1412 bpf_compute_data_pointers(skb);
1413 return err;
1414 }
1415
1416 static int bpf_try_make_head_writable(struct sk_buff *skb)
1417 {
1418 return bpf_try_make_writable(skb, skb_headlen(skb));
1419 }
1420
1421 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1422 {
1423 if (skb_at_tc_ingress(skb))
1424 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1425 }
1426
1427 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1428 {
1429 if (skb_at_tc_ingress(skb))
1430 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1431 }
1432
1433 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1434 const void *, from, u32, len, u64, flags)
1435 {
1436 void *ptr;
1437
1438 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1439 return -EINVAL;
1440 if (unlikely(offset > 0xffff))
1441 return -EFAULT;
1442 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1443 return -EFAULT;
1444
1445 ptr = skb->data + offset;
1446 if (flags & BPF_F_RECOMPUTE_CSUM)
1447 __skb_postpull_rcsum(skb, ptr, len, offset);
1448
1449 memcpy(ptr, from, len);
1450
1451 if (flags & BPF_F_RECOMPUTE_CSUM)
1452 __skb_postpush_rcsum(skb, ptr, len, offset);
1453 if (flags & BPF_F_INVALIDATE_HASH)
1454 skb_clear_hash(skb);
1455
1456 return 0;
1457 }
1458
1459 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1460 .func = bpf_skb_store_bytes,
1461 .gpl_only = false,
1462 .ret_type = RET_INTEGER,
1463 .arg1_type = ARG_PTR_TO_CTX,
1464 .arg2_type = ARG_ANYTHING,
1465 .arg3_type = ARG_PTR_TO_MEM,
1466 .arg4_type = ARG_CONST_SIZE,
1467 .arg5_type = ARG_ANYTHING,
1468 };
1469
1470 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1471 void *, to, u32, len)
1472 {
1473 void *ptr;
1474
1475 if (unlikely(offset > 0xffff))
1476 goto err_clear;
1477
1478 ptr = skb_header_pointer(skb, offset, len, to);
1479 if (unlikely(!ptr))
1480 goto err_clear;
1481 if (ptr != to)
1482 memcpy(to, ptr, len);
1483
1484 return 0;
1485 err_clear:
1486 memset(to, 0, len);
1487 return -EFAULT;
1488 }
1489
1490 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1491 .func = bpf_skb_load_bytes,
1492 .gpl_only = false,
1493 .ret_type = RET_INTEGER,
1494 .arg1_type = ARG_PTR_TO_CTX,
1495 .arg2_type = ARG_ANYTHING,
1496 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1497 .arg4_type = ARG_CONST_SIZE,
1498 };
1499
1500 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1501 {
1502 /* Idea is the following: should the needed direct read/write
1503 * test fail during runtime, we can pull in more data and redo
1504 * again, since implicitly, we invalidate previous checks here.
1505 *
1506 * Or, since we know how much we need to make read/writeable,
1507 * this can be done once at the program beginning for direct
1508 * access case. By this we overcome limitations of only current
1509 * headroom being accessible.
1510 */
1511 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1512 }
1513
1514 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1515 .func = bpf_skb_pull_data,
1516 .gpl_only = false,
1517 .ret_type = RET_INTEGER,
1518 .arg1_type = ARG_PTR_TO_CTX,
1519 .arg2_type = ARG_ANYTHING,
1520 };
1521
1522 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1523 u64, from, u64, to, u64, flags)
1524 {
1525 __sum16 *ptr;
1526
1527 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1528 return -EINVAL;
1529 if (unlikely(offset > 0xffff || offset & 1))
1530 return -EFAULT;
1531 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1532 return -EFAULT;
1533
1534 ptr = (__sum16 *)(skb->data + offset);
1535 switch (flags & BPF_F_HDR_FIELD_MASK) {
1536 case 0:
1537 if (unlikely(from != 0))
1538 return -EINVAL;
1539
1540 csum_replace_by_diff(ptr, to);
1541 break;
1542 case 2:
1543 csum_replace2(ptr, from, to);
1544 break;
1545 case 4:
1546 csum_replace4(ptr, from, to);
1547 break;
1548 default:
1549 return -EINVAL;
1550 }
1551
1552 return 0;
1553 }
1554
1555 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1556 .func = bpf_l3_csum_replace,
1557 .gpl_only = false,
1558 .ret_type = RET_INTEGER,
1559 .arg1_type = ARG_PTR_TO_CTX,
1560 .arg2_type = ARG_ANYTHING,
1561 .arg3_type = ARG_ANYTHING,
1562 .arg4_type = ARG_ANYTHING,
1563 .arg5_type = ARG_ANYTHING,
1564 };
1565
1566 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1567 u64, from, u64, to, u64, flags)
1568 {
1569 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1570 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1571 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1572 __sum16 *ptr;
1573
1574 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1575 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1576 return -EINVAL;
1577 if (unlikely(offset > 0xffff || offset & 1))
1578 return -EFAULT;
1579 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1580 return -EFAULT;
1581
1582 ptr = (__sum16 *)(skb->data + offset);
1583 if (is_mmzero && !do_mforce && !*ptr)
1584 return 0;
1585
1586 switch (flags & BPF_F_HDR_FIELD_MASK) {
1587 case 0:
1588 if (unlikely(from != 0))
1589 return -EINVAL;
1590
1591 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1592 break;
1593 case 2:
1594 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1595 break;
1596 case 4:
1597 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1598 break;
1599 default:
1600 return -EINVAL;
1601 }
1602
1603 if (is_mmzero && !*ptr)
1604 *ptr = CSUM_MANGLED_0;
1605 return 0;
1606 }
1607
1608 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1609 .func = bpf_l4_csum_replace,
1610 .gpl_only = false,
1611 .ret_type = RET_INTEGER,
1612 .arg1_type = ARG_PTR_TO_CTX,
1613 .arg2_type = ARG_ANYTHING,
1614 .arg3_type = ARG_ANYTHING,
1615 .arg4_type = ARG_ANYTHING,
1616 .arg5_type = ARG_ANYTHING,
1617 };
1618
1619 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1620 __be32 *, to, u32, to_size, __wsum, seed)
1621 {
1622 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1623 u32 diff_size = from_size + to_size;
1624 int i, j = 0;
1625
1626 /* This is quite flexible, some examples:
1627 *
1628 * from_size == 0, to_size > 0, seed := csum --> pushing data
1629 * from_size > 0, to_size == 0, seed := csum --> pulling data
1630 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1631 *
1632 * Even for diffing, from_size and to_size don't need to be equal.
1633 */
1634 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1635 diff_size > sizeof(sp->diff)))
1636 return -EINVAL;
1637
1638 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1639 sp->diff[j] = ~from[i];
1640 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1641 sp->diff[j] = to[i];
1642
1643 return csum_partial(sp->diff, diff_size, seed);
1644 }
1645
1646 static const struct bpf_func_proto bpf_csum_diff_proto = {
1647 .func = bpf_csum_diff,
1648 .gpl_only = false,
1649 .pkt_access = true,
1650 .ret_type = RET_INTEGER,
1651 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1652 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1653 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1654 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1655 .arg5_type = ARG_ANYTHING,
1656 };
1657
1658 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1659 {
1660 /* The interface is to be used in combination with bpf_csum_diff()
1661 * for direct packet writes. csum rotation for alignment as well
1662 * as emulating csum_sub() can be done from the eBPF program.
1663 */
1664 if (skb->ip_summed == CHECKSUM_COMPLETE)
1665 return (skb->csum = csum_add(skb->csum, csum));
1666
1667 return -ENOTSUPP;
1668 }
1669
1670 static const struct bpf_func_proto bpf_csum_update_proto = {
1671 .func = bpf_csum_update,
1672 .gpl_only = false,
1673 .ret_type = RET_INTEGER,
1674 .arg1_type = ARG_PTR_TO_CTX,
1675 .arg2_type = ARG_ANYTHING,
1676 };
1677
1678 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1679 {
1680 return dev_forward_skb(dev, skb);
1681 }
1682
1683 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1684 struct sk_buff *skb)
1685 {
1686 int ret = ____dev_forward_skb(dev, skb);
1687
1688 if (likely(!ret)) {
1689 skb->dev = dev;
1690 ret = netif_rx(skb);
1691 }
1692
1693 return ret;
1694 }
1695
1696 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1697 {
1698 int ret;
1699
1700 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1701 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1702 kfree_skb(skb);
1703 return -ENETDOWN;
1704 }
1705
1706 skb->dev = dev;
1707
1708 __this_cpu_inc(xmit_recursion);
1709 ret = dev_queue_xmit(skb);
1710 __this_cpu_dec(xmit_recursion);
1711
1712 return ret;
1713 }
1714
1715 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
1716 u32 flags)
1717 {
1718 unsigned int mlen = skb_network_offset(skb);
1719
1720 if (mlen) {
1721 __skb_pull(skb, mlen);
1722
1723 /* At ingress, the mac header has already been pulled once.
1724 * At egress, skb_pospull_rcsum has to be done in case that
1725 * the skb is originated from ingress (i.e. a forwarded skb)
1726 * to ensure that rcsum starts at net header.
1727 */
1728 if (!skb_at_tc_ingress(skb))
1729 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
1730 }
1731 skb_pop_mac_header(skb);
1732 skb_reset_mac_len(skb);
1733 return flags & BPF_F_INGRESS ?
1734 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
1735 }
1736
1737 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
1738 u32 flags)
1739 {
1740 /* Verify that a link layer header is carried */
1741 if (unlikely(skb->mac_header >= skb->network_header)) {
1742 kfree_skb(skb);
1743 return -ERANGE;
1744 }
1745
1746 bpf_push_mac_rcsum(skb);
1747 return flags & BPF_F_INGRESS ?
1748 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1749 }
1750
1751 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
1752 u32 flags)
1753 {
1754 if (dev_is_mac_header_xmit(dev))
1755 return __bpf_redirect_common(skb, dev, flags);
1756 else
1757 return __bpf_redirect_no_mac(skb, dev, flags);
1758 }
1759
1760 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
1761 {
1762 struct net_device *dev;
1763 struct sk_buff *clone;
1764 int ret;
1765
1766 if (unlikely(flags & ~(BPF_F_INGRESS)))
1767 return -EINVAL;
1768
1769 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1770 if (unlikely(!dev))
1771 return -EINVAL;
1772
1773 clone = skb_clone(skb, GFP_ATOMIC);
1774 if (unlikely(!clone))
1775 return -ENOMEM;
1776
1777 /* For direct write, we need to keep the invariant that the skbs
1778 * we're dealing with need to be uncloned. Should uncloning fail
1779 * here, we need to free the just generated clone to unclone once
1780 * again.
1781 */
1782 ret = bpf_try_make_head_writable(skb);
1783 if (unlikely(ret)) {
1784 kfree_skb(clone);
1785 return -ENOMEM;
1786 }
1787
1788 return __bpf_redirect(clone, dev, flags);
1789 }
1790
1791 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1792 .func = bpf_clone_redirect,
1793 .gpl_only = false,
1794 .ret_type = RET_INTEGER,
1795 .arg1_type = ARG_PTR_TO_CTX,
1796 .arg2_type = ARG_ANYTHING,
1797 .arg3_type = ARG_ANYTHING,
1798 };
1799
1800 struct redirect_info {
1801 u32 ifindex;
1802 u32 flags;
1803 struct bpf_map *map;
1804 struct bpf_map *map_to_flush;
1805 unsigned long map_owner;
1806 };
1807
1808 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1809
1810 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
1811 {
1812 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1813
1814 if (unlikely(flags & ~(BPF_F_INGRESS)))
1815 return TC_ACT_SHOT;
1816
1817 ri->ifindex = ifindex;
1818 ri->flags = flags;
1819
1820 return TC_ACT_REDIRECT;
1821 }
1822
1823 int skb_do_redirect(struct sk_buff *skb)
1824 {
1825 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1826 struct net_device *dev;
1827
1828 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1829 ri->ifindex = 0;
1830 if (unlikely(!dev)) {
1831 kfree_skb(skb);
1832 return -EINVAL;
1833 }
1834
1835 return __bpf_redirect(skb, dev, ri->flags);
1836 }
1837
1838 static const struct bpf_func_proto bpf_redirect_proto = {
1839 .func = bpf_redirect,
1840 .gpl_only = false,
1841 .ret_type = RET_INTEGER,
1842 .arg1_type = ARG_ANYTHING,
1843 .arg2_type = ARG_ANYTHING,
1844 };
1845
1846 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
1847 struct bpf_map *, map, u32, key, u64, flags)
1848 {
1849 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1850
1851 /* If user passes invalid input drop the packet. */
1852 if (unlikely(flags))
1853 return SK_DROP;
1854
1855 tcb->bpf.key = key;
1856 tcb->bpf.flags = flags;
1857 tcb->bpf.map = map;
1858
1859 return SK_PASS;
1860 }
1861
1862 struct sock *do_sk_redirect_map(struct sk_buff *skb)
1863 {
1864 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1865 struct sock *sk = NULL;
1866
1867 if (tcb->bpf.map) {
1868 sk = __sock_map_lookup_elem(tcb->bpf.map, tcb->bpf.key);
1869
1870 tcb->bpf.key = 0;
1871 tcb->bpf.map = NULL;
1872 }
1873
1874 return sk;
1875 }
1876
1877 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
1878 .func = bpf_sk_redirect_map,
1879 .gpl_only = false,
1880 .ret_type = RET_INTEGER,
1881 .arg1_type = ARG_PTR_TO_CTX,
1882 .arg2_type = ARG_CONST_MAP_PTR,
1883 .arg3_type = ARG_ANYTHING,
1884 .arg4_type = ARG_ANYTHING,
1885 };
1886
1887 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
1888 {
1889 return task_get_classid(skb);
1890 }
1891
1892 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1893 .func = bpf_get_cgroup_classid,
1894 .gpl_only = false,
1895 .ret_type = RET_INTEGER,
1896 .arg1_type = ARG_PTR_TO_CTX,
1897 };
1898
1899 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
1900 {
1901 return dst_tclassid(skb);
1902 }
1903
1904 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1905 .func = bpf_get_route_realm,
1906 .gpl_only = false,
1907 .ret_type = RET_INTEGER,
1908 .arg1_type = ARG_PTR_TO_CTX,
1909 };
1910
1911 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
1912 {
1913 /* If skb_clear_hash() was called due to mangling, we can
1914 * trigger SW recalculation here. Later access to hash
1915 * can then use the inline skb->hash via context directly
1916 * instead of calling this helper again.
1917 */
1918 return skb_get_hash(skb);
1919 }
1920
1921 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1922 .func = bpf_get_hash_recalc,
1923 .gpl_only = false,
1924 .ret_type = RET_INTEGER,
1925 .arg1_type = ARG_PTR_TO_CTX,
1926 };
1927
1928 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
1929 {
1930 /* After all direct packet write, this can be used once for
1931 * triggering a lazy recalc on next skb_get_hash() invocation.
1932 */
1933 skb_clear_hash(skb);
1934 return 0;
1935 }
1936
1937 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
1938 .func = bpf_set_hash_invalid,
1939 .gpl_only = false,
1940 .ret_type = RET_INTEGER,
1941 .arg1_type = ARG_PTR_TO_CTX,
1942 };
1943
1944 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
1945 {
1946 /* Set user specified hash as L4(+), so that it gets returned
1947 * on skb_get_hash() call unless BPF prog later on triggers a
1948 * skb_clear_hash().
1949 */
1950 __skb_set_sw_hash(skb, hash, true);
1951 return 0;
1952 }
1953
1954 static const struct bpf_func_proto bpf_set_hash_proto = {
1955 .func = bpf_set_hash,
1956 .gpl_only = false,
1957 .ret_type = RET_INTEGER,
1958 .arg1_type = ARG_PTR_TO_CTX,
1959 .arg2_type = ARG_ANYTHING,
1960 };
1961
1962 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
1963 u16, vlan_tci)
1964 {
1965 int ret;
1966
1967 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1968 vlan_proto != htons(ETH_P_8021AD)))
1969 vlan_proto = htons(ETH_P_8021Q);
1970
1971 bpf_push_mac_rcsum(skb);
1972 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1973 bpf_pull_mac_rcsum(skb);
1974
1975 bpf_compute_data_pointers(skb);
1976 return ret;
1977 }
1978
1979 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1980 .func = bpf_skb_vlan_push,
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 };
1987 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1988
1989 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
1990 {
1991 int ret;
1992
1993 bpf_push_mac_rcsum(skb);
1994 ret = skb_vlan_pop(skb);
1995 bpf_pull_mac_rcsum(skb);
1996
1997 bpf_compute_data_pointers(skb);
1998 return ret;
1999 }
2000
2001 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2002 .func = bpf_skb_vlan_pop,
2003 .gpl_only = false,
2004 .ret_type = RET_INTEGER,
2005 .arg1_type = ARG_PTR_TO_CTX,
2006 };
2007 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
2008
2009 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2010 {
2011 /* Caller already did skb_cow() with len as headroom,
2012 * so no need to do it here.
2013 */
2014 skb_push(skb, len);
2015 memmove(skb->data, skb->data + len, off);
2016 memset(skb->data + off, 0, len);
2017
2018 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2019 * needed here as it does not change the skb->csum
2020 * result for checksum complete when summing over
2021 * zeroed blocks.
2022 */
2023 return 0;
2024 }
2025
2026 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2027 {
2028 /* skb_ensure_writable() is not needed here, as we're
2029 * already working on an uncloned skb.
2030 */
2031 if (unlikely(!pskb_may_pull(skb, off + len)))
2032 return -ENOMEM;
2033
2034 skb_postpull_rcsum(skb, skb->data + off, len);
2035 memmove(skb->data + len, skb->data, off);
2036 __skb_pull(skb, len);
2037
2038 return 0;
2039 }
2040
2041 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2042 {
2043 bool trans_same = skb->transport_header == skb->network_header;
2044 int ret;
2045
2046 /* There's no need for __skb_push()/__skb_pull() pair to
2047 * get to the start of the mac header as we're guaranteed
2048 * to always start from here under eBPF.
2049 */
2050 ret = bpf_skb_generic_push(skb, off, len);
2051 if (likely(!ret)) {
2052 skb->mac_header -= len;
2053 skb->network_header -= len;
2054 if (trans_same)
2055 skb->transport_header = skb->network_header;
2056 }
2057
2058 return ret;
2059 }
2060
2061 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2062 {
2063 bool trans_same = skb->transport_header == skb->network_header;
2064 int ret;
2065
2066 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2067 ret = bpf_skb_generic_pop(skb, off, len);
2068 if (likely(!ret)) {
2069 skb->mac_header += len;
2070 skb->network_header += len;
2071 if (trans_same)
2072 skb->transport_header = skb->network_header;
2073 }
2074
2075 return ret;
2076 }
2077
2078 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2079 {
2080 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2081 u32 off = skb_mac_header_len(skb);
2082 int ret;
2083
2084 ret = skb_cow(skb, len_diff);
2085 if (unlikely(ret < 0))
2086 return ret;
2087
2088 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2089 if (unlikely(ret < 0))
2090 return ret;
2091
2092 if (skb_is_gso(skb)) {
2093 /* SKB_GSO_TCPV4 needs to be changed into
2094 * SKB_GSO_TCPV6.
2095 */
2096 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2097 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
2098 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
2099 }
2100
2101 /* Due to IPv6 header, MSS needs to be downgraded. */
2102 skb_shinfo(skb)->gso_size -= len_diff;
2103 /* Header must be checked, and gso_segs recomputed. */
2104 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2105 skb_shinfo(skb)->gso_segs = 0;
2106 }
2107
2108 skb->protocol = htons(ETH_P_IPV6);
2109 skb_clear_hash(skb);
2110
2111 return 0;
2112 }
2113
2114 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2115 {
2116 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2117 u32 off = skb_mac_header_len(skb);
2118 int ret;
2119
2120 ret = skb_unclone(skb, GFP_ATOMIC);
2121 if (unlikely(ret < 0))
2122 return ret;
2123
2124 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2125 if (unlikely(ret < 0))
2126 return ret;
2127
2128 if (skb_is_gso(skb)) {
2129 /* SKB_GSO_TCPV6 needs to be changed into
2130 * SKB_GSO_TCPV4.
2131 */
2132 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
2133 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
2134 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
2135 }
2136
2137 /* Due to IPv4 header, MSS can be upgraded. */
2138 skb_shinfo(skb)->gso_size += len_diff;
2139 /* Header must be checked, and gso_segs recomputed. */
2140 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2141 skb_shinfo(skb)->gso_segs = 0;
2142 }
2143
2144 skb->protocol = htons(ETH_P_IP);
2145 skb_clear_hash(skb);
2146
2147 return 0;
2148 }
2149
2150 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2151 {
2152 __be16 from_proto = skb->protocol;
2153
2154 if (from_proto == htons(ETH_P_IP) &&
2155 to_proto == htons(ETH_P_IPV6))
2156 return bpf_skb_proto_4_to_6(skb);
2157
2158 if (from_proto == htons(ETH_P_IPV6) &&
2159 to_proto == htons(ETH_P_IP))
2160 return bpf_skb_proto_6_to_4(skb);
2161
2162 return -ENOTSUPP;
2163 }
2164
2165 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2166 u64, flags)
2167 {
2168 int ret;
2169
2170 if (unlikely(flags))
2171 return -EINVAL;
2172
2173 /* General idea is that this helper does the basic groundwork
2174 * needed for changing the protocol, and eBPF program fills the
2175 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2176 * and other helpers, rather than passing a raw buffer here.
2177 *
2178 * The rationale is to keep this minimal and without a need to
2179 * deal with raw packet data. F.e. even if we would pass buffers
2180 * here, the program still needs to call the bpf_lX_csum_replace()
2181 * helpers anyway. Plus, this way we keep also separation of
2182 * concerns, since f.e. bpf_skb_store_bytes() should only take
2183 * care of stores.
2184 *
2185 * Currently, additional options and extension header space are
2186 * not supported, but flags register is reserved so we can adapt
2187 * that. For offloads, we mark packet as dodgy, so that headers
2188 * need to be verified first.
2189 */
2190 ret = bpf_skb_proto_xlat(skb, proto);
2191 bpf_compute_data_pointers(skb);
2192 return ret;
2193 }
2194
2195 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2196 .func = bpf_skb_change_proto,
2197 .gpl_only = false,
2198 .ret_type = RET_INTEGER,
2199 .arg1_type = ARG_PTR_TO_CTX,
2200 .arg2_type = ARG_ANYTHING,
2201 .arg3_type = ARG_ANYTHING,
2202 };
2203
2204 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2205 {
2206 /* We only allow a restricted subset to be changed for now. */
2207 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2208 !skb_pkt_type_ok(pkt_type)))
2209 return -EINVAL;
2210
2211 skb->pkt_type = pkt_type;
2212 return 0;
2213 }
2214
2215 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2216 .func = bpf_skb_change_type,
2217 .gpl_only = false,
2218 .ret_type = RET_INTEGER,
2219 .arg1_type = ARG_PTR_TO_CTX,
2220 .arg2_type = ARG_ANYTHING,
2221 };
2222
2223 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2224 {
2225 switch (skb->protocol) {
2226 case htons(ETH_P_IP):
2227 return sizeof(struct iphdr);
2228 case htons(ETH_P_IPV6):
2229 return sizeof(struct ipv6hdr);
2230 default:
2231 return ~0U;
2232 }
2233 }
2234
2235 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2236 {
2237 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2238 int ret;
2239
2240 ret = skb_cow(skb, len_diff);
2241 if (unlikely(ret < 0))
2242 return ret;
2243
2244 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2245 if (unlikely(ret < 0))
2246 return ret;
2247
2248 if (skb_is_gso(skb)) {
2249 /* Due to header grow, MSS needs to be downgraded. */
2250 skb_shinfo(skb)->gso_size -= len_diff;
2251 /* Header must be checked, and gso_segs recomputed. */
2252 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2253 skb_shinfo(skb)->gso_segs = 0;
2254 }
2255
2256 return 0;
2257 }
2258
2259 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2260 {
2261 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2262 int ret;
2263
2264 ret = skb_unclone(skb, GFP_ATOMIC);
2265 if (unlikely(ret < 0))
2266 return ret;
2267
2268 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2269 if (unlikely(ret < 0))
2270 return ret;
2271
2272 if (skb_is_gso(skb)) {
2273 /* Due to header shrink, MSS can be upgraded. */
2274 skb_shinfo(skb)->gso_size += len_diff;
2275 /* Header must be checked, and gso_segs recomputed. */
2276 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2277 skb_shinfo(skb)->gso_segs = 0;
2278 }
2279
2280 return 0;
2281 }
2282
2283 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2284 {
2285 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2286 SKB_MAX_ALLOC;
2287 }
2288
2289 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2290 {
2291 bool trans_same = skb->transport_header == skb->network_header;
2292 u32 len_cur, len_diff_abs = abs(len_diff);
2293 u32 len_min = bpf_skb_net_base_len(skb);
2294 u32 len_max = __bpf_skb_max_len(skb);
2295 __be16 proto = skb->protocol;
2296 bool shrink = len_diff < 0;
2297 int ret;
2298
2299 if (unlikely(len_diff_abs > 0xfffU))
2300 return -EFAULT;
2301 if (unlikely(proto != htons(ETH_P_IP) &&
2302 proto != htons(ETH_P_IPV6)))
2303 return -ENOTSUPP;
2304
2305 len_cur = skb->len - skb_network_offset(skb);
2306 if (skb_transport_header_was_set(skb) && !trans_same)
2307 len_cur = skb_network_header_len(skb);
2308 if ((shrink && (len_diff_abs >= len_cur ||
2309 len_cur - len_diff_abs < len_min)) ||
2310 (!shrink && (skb->len + len_diff_abs > len_max &&
2311 !skb_is_gso(skb))))
2312 return -ENOTSUPP;
2313
2314 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2315 bpf_skb_net_grow(skb, len_diff_abs);
2316
2317 bpf_compute_data_pointers(skb);
2318 return ret;
2319 }
2320
2321 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2322 u32, mode, u64, flags)
2323 {
2324 if (unlikely(flags))
2325 return -EINVAL;
2326 if (likely(mode == BPF_ADJ_ROOM_NET))
2327 return bpf_skb_adjust_net(skb, len_diff);
2328
2329 return -ENOTSUPP;
2330 }
2331
2332 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2333 .func = bpf_skb_adjust_room,
2334 .gpl_only = false,
2335 .ret_type = RET_INTEGER,
2336 .arg1_type = ARG_PTR_TO_CTX,
2337 .arg2_type = ARG_ANYTHING,
2338 .arg3_type = ARG_ANYTHING,
2339 .arg4_type = ARG_ANYTHING,
2340 };
2341
2342 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2343 {
2344 u32 min_len = skb_network_offset(skb);
2345
2346 if (skb_transport_header_was_set(skb))
2347 min_len = skb_transport_offset(skb);
2348 if (skb->ip_summed == CHECKSUM_PARTIAL)
2349 min_len = skb_checksum_start_offset(skb) +
2350 skb->csum_offset + sizeof(__sum16);
2351 return min_len;
2352 }
2353
2354 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2355 {
2356 unsigned int old_len = skb->len;
2357 int ret;
2358
2359 ret = __skb_grow_rcsum(skb, new_len);
2360 if (!ret)
2361 memset(skb->data + old_len, 0, new_len - old_len);
2362 return ret;
2363 }
2364
2365 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2366 {
2367 return __skb_trim_rcsum(skb, new_len);
2368 }
2369
2370 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2371 u64, flags)
2372 {
2373 u32 max_len = __bpf_skb_max_len(skb);
2374 u32 min_len = __bpf_skb_min_len(skb);
2375 int ret;
2376
2377 if (unlikely(flags || new_len > max_len || new_len < min_len))
2378 return -EINVAL;
2379 if (skb->encapsulation)
2380 return -ENOTSUPP;
2381
2382 /* The basic idea of this helper is that it's performing the
2383 * needed work to either grow or trim an skb, and eBPF program
2384 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2385 * bpf_lX_csum_replace() and others rather than passing a raw
2386 * buffer here. This one is a slow path helper and intended
2387 * for replies with control messages.
2388 *
2389 * Like in bpf_skb_change_proto(), we want to keep this rather
2390 * minimal and without protocol specifics so that we are able
2391 * to separate concerns as in bpf_skb_store_bytes() should only
2392 * be the one responsible for writing buffers.
2393 *
2394 * It's really expected to be a slow path operation here for
2395 * control message replies, so we're implicitly linearizing,
2396 * uncloning and drop offloads from the skb by this.
2397 */
2398 ret = __bpf_try_make_writable(skb, skb->len);
2399 if (!ret) {
2400 if (new_len > skb->len)
2401 ret = bpf_skb_grow_rcsum(skb, new_len);
2402 else if (new_len < skb->len)
2403 ret = bpf_skb_trim_rcsum(skb, new_len);
2404 if (!ret && skb_is_gso(skb))
2405 skb_gso_reset(skb);
2406 }
2407
2408 bpf_compute_data_pointers(skb);
2409 return ret;
2410 }
2411
2412 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2413 .func = bpf_skb_change_tail,
2414 .gpl_only = false,
2415 .ret_type = RET_INTEGER,
2416 .arg1_type = ARG_PTR_TO_CTX,
2417 .arg2_type = ARG_ANYTHING,
2418 .arg3_type = ARG_ANYTHING,
2419 };
2420
2421 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2422 u64, flags)
2423 {
2424 u32 max_len = __bpf_skb_max_len(skb);
2425 u32 new_len = skb->len + head_room;
2426 int ret;
2427
2428 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2429 new_len < skb->len))
2430 return -EINVAL;
2431
2432 ret = skb_cow(skb, head_room);
2433 if (likely(!ret)) {
2434 /* Idea for this helper is that we currently only
2435 * allow to expand on mac header. This means that
2436 * skb->protocol network header, etc, stay as is.
2437 * Compared to bpf_skb_change_tail(), we're more
2438 * flexible due to not needing to linearize or
2439 * reset GSO. Intention for this helper is to be
2440 * used by an L3 skb that needs to push mac header
2441 * for redirection into L2 device.
2442 */
2443 __skb_push(skb, head_room);
2444 memset(skb->data, 0, head_room);
2445 skb_reset_mac_header(skb);
2446 }
2447
2448 bpf_compute_data_pointers(skb);
2449 return 0;
2450 }
2451
2452 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2453 .func = bpf_skb_change_head,
2454 .gpl_only = false,
2455 .ret_type = RET_INTEGER,
2456 .arg1_type = ARG_PTR_TO_CTX,
2457 .arg2_type = ARG_ANYTHING,
2458 .arg3_type = ARG_ANYTHING,
2459 };
2460
2461 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
2462 {
2463 return xdp_data_meta_unsupported(xdp) ? 0 :
2464 xdp->data - xdp->data_meta;
2465 }
2466
2467 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2468 {
2469 unsigned long metalen = xdp_get_metalen(xdp);
2470 void *data_start = xdp->data_hard_start + metalen;
2471 void *data = xdp->data + offset;
2472
2473 if (unlikely(data < data_start ||
2474 data > xdp->data_end - ETH_HLEN))
2475 return -EINVAL;
2476
2477 if (metalen)
2478 memmove(xdp->data_meta + offset,
2479 xdp->data_meta, metalen);
2480 xdp->data_meta += offset;
2481 xdp->data = data;
2482
2483 return 0;
2484 }
2485
2486 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2487 .func = bpf_xdp_adjust_head,
2488 .gpl_only = false,
2489 .ret_type = RET_INTEGER,
2490 .arg1_type = ARG_PTR_TO_CTX,
2491 .arg2_type = ARG_ANYTHING,
2492 };
2493
2494 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
2495 {
2496 void *meta = xdp->data_meta + offset;
2497 unsigned long metalen = xdp->data - meta;
2498
2499 if (xdp_data_meta_unsupported(xdp))
2500 return -ENOTSUPP;
2501 if (unlikely(meta < xdp->data_hard_start ||
2502 meta > xdp->data))
2503 return -EINVAL;
2504 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
2505 (metalen > 32)))
2506 return -EACCES;
2507
2508 xdp->data_meta = meta;
2509
2510 return 0;
2511 }
2512
2513 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
2514 .func = bpf_xdp_adjust_meta,
2515 .gpl_only = false,
2516 .ret_type = RET_INTEGER,
2517 .arg1_type = ARG_PTR_TO_CTX,
2518 .arg2_type = ARG_ANYTHING,
2519 };
2520
2521 static int __bpf_tx_xdp(struct net_device *dev,
2522 struct bpf_map *map,
2523 struct xdp_buff *xdp,
2524 u32 index)
2525 {
2526 int err;
2527
2528 if (!dev->netdev_ops->ndo_xdp_xmit) {
2529 return -EOPNOTSUPP;
2530 }
2531
2532 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2533 if (err)
2534 return err;
2535 dev->netdev_ops->ndo_xdp_flush(dev);
2536 return 0;
2537 }
2538
2539 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
2540 struct bpf_map *map,
2541 struct xdp_buff *xdp,
2542 u32 index)
2543 {
2544 int err;
2545
2546 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2547 struct net_device *dev = fwd;
2548
2549 if (!dev->netdev_ops->ndo_xdp_xmit)
2550 return -EOPNOTSUPP;
2551
2552 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2553 if (err)
2554 return err;
2555 __dev_map_insert_ctx(map, index);
2556
2557 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
2558 struct bpf_cpu_map_entry *rcpu = fwd;
2559
2560 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
2561 if (err)
2562 return err;
2563 __cpu_map_insert_ctx(map, index);
2564 }
2565 return 0;
2566 }
2567
2568 void xdp_do_flush_map(void)
2569 {
2570 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2571 struct bpf_map *map = ri->map_to_flush;
2572
2573 ri->map_to_flush = NULL;
2574 if (map) {
2575 switch (map->map_type) {
2576 case BPF_MAP_TYPE_DEVMAP:
2577 __dev_map_flush(map);
2578 break;
2579 case BPF_MAP_TYPE_CPUMAP:
2580 __cpu_map_flush(map);
2581 break;
2582 default:
2583 break;
2584 }
2585 }
2586 }
2587 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
2588
2589 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
2590 {
2591 switch (map->map_type) {
2592 case BPF_MAP_TYPE_DEVMAP:
2593 return __dev_map_lookup_elem(map, index);
2594 case BPF_MAP_TYPE_CPUMAP:
2595 return __cpu_map_lookup_elem(map, index);
2596 default:
2597 return NULL;
2598 }
2599 }
2600
2601 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
2602 unsigned long aux)
2603 {
2604 return (unsigned long)xdp_prog->aux != aux;
2605 }
2606
2607 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
2608 struct bpf_prog *xdp_prog)
2609 {
2610 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2611 unsigned long map_owner = ri->map_owner;
2612 struct bpf_map *map = ri->map;
2613 u32 index = ri->ifindex;
2614 void *fwd = NULL;
2615 int err;
2616
2617 ri->ifindex = 0;
2618 ri->map = NULL;
2619 ri->map_owner = 0;
2620
2621 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2622 err = -EFAULT;
2623 map = NULL;
2624 goto err;
2625 }
2626
2627 fwd = __xdp_map_lookup_elem(map, index);
2628 if (!fwd) {
2629 err = -EINVAL;
2630 goto err;
2631 }
2632 if (ri->map_to_flush && ri->map_to_flush != map)
2633 xdp_do_flush_map();
2634
2635 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
2636 if (unlikely(err))
2637 goto err;
2638
2639 ri->map_to_flush = map;
2640 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2641 return 0;
2642 err:
2643 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2644 return err;
2645 }
2646
2647 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
2648 struct bpf_prog *xdp_prog)
2649 {
2650 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2651 struct net_device *fwd;
2652 u32 index = ri->ifindex;
2653 int err;
2654
2655 if (ri->map)
2656 return xdp_do_redirect_map(dev, xdp, xdp_prog);
2657
2658 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2659 ri->ifindex = 0;
2660 if (unlikely(!fwd)) {
2661 err = -EINVAL;
2662 goto err;
2663 }
2664
2665 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
2666 if (unlikely(err))
2667 goto err;
2668
2669 _trace_xdp_redirect(dev, xdp_prog, index);
2670 return 0;
2671 err:
2672 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2673 return err;
2674 }
2675 EXPORT_SYMBOL_GPL(xdp_do_redirect);
2676
2677 static int __xdp_generic_ok_fwd_dev(struct sk_buff *skb, struct net_device *fwd)
2678 {
2679 unsigned int len;
2680
2681 if (unlikely(!(fwd->flags & IFF_UP)))
2682 return -ENETDOWN;
2683
2684 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
2685 if (skb->len > len)
2686 return -EMSGSIZE;
2687
2688 return 0;
2689 }
2690
2691 int xdp_do_generic_redirect_map(struct net_device *dev, struct sk_buff *skb,
2692 struct bpf_prog *xdp_prog)
2693 {
2694 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2695 unsigned long map_owner = ri->map_owner;
2696 struct bpf_map *map = ri->map;
2697 struct net_device *fwd = NULL;
2698 u32 index = ri->ifindex;
2699 int err = 0;
2700
2701 ri->ifindex = 0;
2702 ri->map = NULL;
2703 ri->map_owner = 0;
2704
2705 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2706 err = -EFAULT;
2707 map = NULL;
2708 goto err;
2709 }
2710 fwd = __xdp_map_lookup_elem(map, index);
2711 if (unlikely(!fwd)) {
2712 err = -EINVAL;
2713 goto err;
2714 }
2715
2716 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
2717 if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2718 goto err;
2719 skb->dev = fwd;
2720 } else {
2721 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
2722 err = -EBADRQC;
2723 goto err;
2724 }
2725
2726 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2727 return 0;
2728 err:
2729 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2730 return err;
2731 }
2732
2733 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
2734 struct bpf_prog *xdp_prog)
2735 {
2736 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2737 u32 index = ri->ifindex;
2738 struct net_device *fwd;
2739 int err = 0;
2740
2741 if (ri->map)
2742 return xdp_do_generic_redirect_map(dev, skb, xdp_prog);
2743
2744 ri->ifindex = 0;
2745 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2746 if (unlikely(!fwd)) {
2747 err = -EINVAL;
2748 goto err;
2749 }
2750
2751 if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
2752 goto err;
2753
2754 skb->dev = fwd;
2755 _trace_xdp_redirect(dev, xdp_prog, index);
2756 return 0;
2757 err:
2758 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2759 return err;
2760 }
2761 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
2762
2763 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
2764 {
2765 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2766
2767 if (unlikely(flags))
2768 return XDP_ABORTED;
2769
2770 ri->ifindex = ifindex;
2771 ri->flags = flags;
2772 ri->map = NULL;
2773 ri->map_owner = 0;
2774
2775 return XDP_REDIRECT;
2776 }
2777
2778 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
2779 .func = bpf_xdp_redirect,
2780 .gpl_only = false,
2781 .ret_type = RET_INTEGER,
2782 .arg1_type = ARG_ANYTHING,
2783 .arg2_type = ARG_ANYTHING,
2784 };
2785
2786 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
2787 unsigned long, map_owner)
2788 {
2789 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2790
2791 if (unlikely(flags))
2792 return XDP_ABORTED;
2793
2794 ri->ifindex = ifindex;
2795 ri->flags = flags;
2796 ri->map = map;
2797 ri->map_owner = map_owner;
2798
2799 return XDP_REDIRECT;
2800 }
2801
2802 /* Note, arg4 is hidden from users and populated by the verifier
2803 * with the right pointer.
2804 */
2805 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
2806 .func = bpf_xdp_redirect_map,
2807 .gpl_only = false,
2808 .ret_type = RET_INTEGER,
2809 .arg1_type = ARG_CONST_MAP_PTR,
2810 .arg2_type = ARG_ANYTHING,
2811 .arg3_type = ARG_ANYTHING,
2812 };
2813
2814 bool bpf_helper_changes_pkt_data(void *func)
2815 {
2816 if (func == bpf_skb_vlan_push ||
2817 func == bpf_skb_vlan_pop ||
2818 func == bpf_skb_store_bytes ||
2819 func == bpf_skb_change_proto ||
2820 func == bpf_skb_change_head ||
2821 func == bpf_skb_change_tail ||
2822 func == bpf_skb_adjust_room ||
2823 func == bpf_skb_pull_data ||
2824 func == bpf_clone_redirect ||
2825 func == bpf_l3_csum_replace ||
2826 func == bpf_l4_csum_replace ||
2827 func == bpf_xdp_adjust_head ||
2828 func == bpf_xdp_adjust_meta)
2829 return true;
2830
2831 return false;
2832 }
2833
2834 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2835 unsigned long off, unsigned long len)
2836 {
2837 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2838
2839 if (unlikely(!ptr))
2840 return len;
2841 if (ptr != dst_buff)
2842 memcpy(dst_buff, ptr, len);
2843
2844 return 0;
2845 }
2846
2847 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2848 u64, flags, void *, meta, u64, meta_size)
2849 {
2850 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2851
2852 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2853 return -EINVAL;
2854 if (unlikely(skb_size > skb->len))
2855 return -EFAULT;
2856
2857 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2858 bpf_skb_copy);
2859 }
2860
2861 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2862 .func = bpf_skb_event_output,
2863 .gpl_only = true,
2864 .ret_type = RET_INTEGER,
2865 .arg1_type = ARG_PTR_TO_CTX,
2866 .arg2_type = ARG_CONST_MAP_PTR,
2867 .arg3_type = ARG_ANYTHING,
2868 .arg4_type = ARG_PTR_TO_MEM,
2869 .arg5_type = ARG_CONST_SIZE,
2870 };
2871
2872 static unsigned short bpf_tunnel_key_af(u64 flags)
2873 {
2874 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2875 }
2876
2877 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2878 u32, size, u64, flags)
2879 {
2880 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2881 u8 compat[sizeof(struct bpf_tunnel_key)];
2882 void *to_orig = to;
2883 int err;
2884
2885 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2886 err = -EINVAL;
2887 goto err_clear;
2888 }
2889 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2890 err = -EPROTO;
2891 goto err_clear;
2892 }
2893 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2894 err = -EINVAL;
2895 switch (size) {
2896 case offsetof(struct bpf_tunnel_key, tunnel_label):
2897 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2898 goto set_compat;
2899 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2900 /* Fixup deprecated structure layouts here, so we have
2901 * a common path later on.
2902 */
2903 if (ip_tunnel_info_af(info) != AF_INET)
2904 goto err_clear;
2905 set_compat:
2906 to = (struct bpf_tunnel_key *)compat;
2907 break;
2908 default:
2909 goto err_clear;
2910 }
2911 }
2912
2913 to->tunnel_id = be64_to_cpu(info->key.tun_id);
2914 to->tunnel_tos = info->key.tos;
2915 to->tunnel_ttl = info->key.ttl;
2916
2917 if (flags & BPF_F_TUNINFO_IPV6) {
2918 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2919 sizeof(to->remote_ipv6));
2920 to->tunnel_label = be32_to_cpu(info->key.label);
2921 } else {
2922 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2923 }
2924
2925 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2926 memcpy(to_orig, to, size);
2927
2928 return 0;
2929 err_clear:
2930 memset(to_orig, 0, size);
2931 return err;
2932 }
2933
2934 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2935 .func = bpf_skb_get_tunnel_key,
2936 .gpl_only = false,
2937 .ret_type = RET_INTEGER,
2938 .arg1_type = ARG_PTR_TO_CTX,
2939 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2940 .arg3_type = ARG_CONST_SIZE,
2941 .arg4_type = ARG_ANYTHING,
2942 };
2943
2944 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2945 {
2946 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2947 int err;
2948
2949 if (unlikely(!info ||
2950 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2951 err = -ENOENT;
2952 goto err_clear;
2953 }
2954 if (unlikely(size < info->options_len)) {
2955 err = -ENOMEM;
2956 goto err_clear;
2957 }
2958
2959 ip_tunnel_info_opts_get(to, info);
2960 if (size > info->options_len)
2961 memset(to + info->options_len, 0, size - info->options_len);
2962
2963 return info->options_len;
2964 err_clear:
2965 memset(to, 0, size);
2966 return err;
2967 }
2968
2969 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2970 .func = bpf_skb_get_tunnel_opt,
2971 .gpl_only = false,
2972 .ret_type = RET_INTEGER,
2973 .arg1_type = ARG_PTR_TO_CTX,
2974 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2975 .arg3_type = ARG_CONST_SIZE,
2976 };
2977
2978 static struct metadata_dst __percpu *md_dst;
2979
2980 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
2981 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
2982 {
2983 struct metadata_dst *md = this_cpu_ptr(md_dst);
2984 u8 compat[sizeof(struct bpf_tunnel_key)];
2985 struct ip_tunnel_info *info;
2986
2987 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2988 BPF_F_DONT_FRAGMENT)))
2989 return -EINVAL;
2990 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2991 switch (size) {
2992 case offsetof(struct bpf_tunnel_key, tunnel_label):
2993 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2994 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2995 /* Fixup deprecated structure layouts here, so we have
2996 * a common path later on.
2997 */
2998 memcpy(compat, from, size);
2999 memset(compat + size, 0, sizeof(compat) - size);
3000 from = (const struct bpf_tunnel_key *) compat;
3001 break;
3002 default:
3003 return -EINVAL;
3004 }
3005 }
3006 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3007 from->tunnel_ext))
3008 return -EINVAL;
3009
3010 skb_dst_drop(skb);
3011 dst_hold((struct dst_entry *) md);
3012 skb_dst_set(skb, (struct dst_entry *) md);
3013
3014 info = &md->u.tun_info;
3015 info->mode = IP_TUNNEL_INFO_TX;
3016
3017 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3018 if (flags & BPF_F_DONT_FRAGMENT)
3019 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3020
3021 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3022 info->key.tos = from->tunnel_tos;
3023 info->key.ttl = from->tunnel_ttl;
3024
3025 if (flags & BPF_F_TUNINFO_IPV6) {
3026 info->mode |= IP_TUNNEL_INFO_IPV6;
3027 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3028 sizeof(from->remote_ipv6));
3029 info->key.label = cpu_to_be32(from->tunnel_label) &
3030 IPV6_FLOWLABEL_MASK;
3031 } else {
3032 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3033 if (flags & BPF_F_ZERO_CSUM_TX)
3034 info->key.tun_flags &= ~TUNNEL_CSUM;
3035 }
3036
3037 return 0;
3038 }
3039
3040 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3041 .func = bpf_skb_set_tunnel_key,
3042 .gpl_only = false,
3043 .ret_type = RET_INTEGER,
3044 .arg1_type = ARG_PTR_TO_CTX,
3045 .arg2_type = ARG_PTR_TO_MEM,
3046 .arg3_type = ARG_CONST_SIZE,
3047 .arg4_type = ARG_ANYTHING,
3048 };
3049
3050 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3051 const u8 *, from, u32, size)
3052 {
3053 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3054 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3055
3056 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3057 return -EINVAL;
3058 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3059 return -ENOMEM;
3060
3061 ip_tunnel_info_opts_set(info, from, size);
3062
3063 return 0;
3064 }
3065
3066 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3067 .func = bpf_skb_set_tunnel_opt,
3068 .gpl_only = false,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3071 .arg2_type = ARG_PTR_TO_MEM,
3072 .arg3_type = ARG_CONST_SIZE,
3073 };
3074
3075 static const struct bpf_func_proto *
3076 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3077 {
3078 if (!md_dst) {
3079 struct metadata_dst __percpu *tmp;
3080
3081 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3082 METADATA_IP_TUNNEL,
3083 GFP_KERNEL);
3084 if (!tmp)
3085 return NULL;
3086 if (cmpxchg(&md_dst, NULL, tmp))
3087 metadata_dst_free_percpu(tmp);
3088 }
3089
3090 switch (which) {
3091 case BPF_FUNC_skb_set_tunnel_key:
3092 return &bpf_skb_set_tunnel_key_proto;
3093 case BPF_FUNC_skb_set_tunnel_opt:
3094 return &bpf_skb_set_tunnel_opt_proto;
3095 default:
3096 return NULL;
3097 }
3098 }
3099
3100 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3101 u32, idx)
3102 {
3103 struct bpf_array *array = container_of(map, struct bpf_array, map);
3104 struct cgroup *cgrp;
3105 struct sock *sk;
3106
3107 sk = skb_to_full_sk(skb);
3108 if (!sk || !sk_fullsock(sk))
3109 return -ENOENT;
3110 if (unlikely(idx >= array->map.max_entries))
3111 return -E2BIG;
3112
3113 cgrp = READ_ONCE(array->ptrs[idx]);
3114 if (unlikely(!cgrp))
3115 return -EAGAIN;
3116
3117 return sk_under_cgroup_hierarchy(sk, cgrp);
3118 }
3119
3120 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3121 .func = bpf_skb_under_cgroup,
3122 .gpl_only = false,
3123 .ret_type = RET_INTEGER,
3124 .arg1_type = ARG_PTR_TO_CTX,
3125 .arg2_type = ARG_CONST_MAP_PTR,
3126 .arg3_type = ARG_ANYTHING,
3127 };
3128
3129 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3130 unsigned long off, unsigned long len)
3131 {
3132 memcpy(dst_buff, src_buff + off, len);
3133 return 0;
3134 }
3135
3136 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3137 u64, flags, void *, meta, u64, meta_size)
3138 {
3139 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3140
3141 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3142 return -EINVAL;
3143 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3144 return -EFAULT;
3145
3146 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3147 xdp_size, bpf_xdp_copy);
3148 }
3149
3150 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3151 .func = bpf_xdp_event_output,
3152 .gpl_only = true,
3153 .ret_type = RET_INTEGER,
3154 .arg1_type = ARG_PTR_TO_CTX,
3155 .arg2_type = ARG_CONST_MAP_PTR,
3156 .arg3_type = ARG_ANYTHING,
3157 .arg4_type = ARG_PTR_TO_MEM,
3158 .arg5_type = ARG_CONST_SIZE,
3159 };
3160
3161 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3162 {
3163 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3164 }
3165
3166 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3167 .func = bpf_get_socket_cookie,
3168 .gpl_only = false,
3169 .ret_type = RET_INTEGER,
3170 .arg1_type = ARG_PTR_TO_CTX,
3171 };
3172
3173 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3174 {
3175 struct sock *sk = sk_to_full_sk(skb->sk);
3176 kuid_t kuid;
3177
3178 if (!sk || !sk_fullsock(sk))
3179 return overflowuid;
3180 kuid = sock_net_uid(sock_net(sk), sk);
3181 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3182 }
3183
3184 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3185 .func = bpf_get_socket_uid,
3186 .gpl_only = false,
3187 .ret_type = RET_INTEGER,
3188 .arg1_type = ARG_PTR_TO_CTX,
3189 };
3190
3191 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3192 int, level, int, optname, char *, optval, int, optlen)
3193 {
3194 struct sock *sk = bpf_sock->sk;
3195 int ret = 0;
3196 int val;
3197
3198 if (!sk_fullsock(sk))
3199 return -EINVAL;
3200
3201 if (level == SOL_SOCKET) {
3202 if (optlen != sizeof(int))
3203 return -EINVAL;
3204 val = *((int *)optval);
3205
3206 /* Only some socketops are supported */
3207 switch (optname) {
3208 case SO_RCVBUF:
3209 val = min_t(u32, val, sysctl_rmem_max);
3210 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3211 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3212 break;
3213 case SO_SNDBUF:
3214 val = min_t(u32, val, sysctl_wmem_max);
3215 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3216 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3217 break;
3218 case SO_MAX_PACING_RATE:
3219 sk->sk_max_pacing_rate = val;
3220 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3221 sk->sk_max_pacing_rate);
3222 break;
3223 case SO_PRIORITY:
3224 sk->sk_priority = val;
3225 break;
3226 case SO_RCVLOWAT:
3227 if (val < 0)
3228 val = INT_MAX;
3229 sk->sk_rcvlowat = val ? : 1;
3230 break;
3231 case SO_MARK:
3232 if (sk->sk_mark != val) {
3233 sk->sk_mark = val;
3234 sk_dst_reset(sk);
3235 }
3236 break;
3237 default:
3238 ret = -EINVAL;
3239 }
3240 #ifdef CONFIG_INET
3241 } else if (level == SOL_TCP &&
3242 sk->sk_prot->setsockopt == tcp_setsockopt) {
3243 if (optname == TCP_CONGESTION) {
3244 char name[TCP_CA_NAME_MAX];
3245 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3246
3247 strncpy(name, optval, min_t(long, optlen,
3248 TCP_CA_NAME_MAX-1));
3249 name[TCP_CA_NAME_MAX-1] = 0;
3250 ret = tcp_set_congestion_control(sk, name, false,
3251 reinit, true);
3252 } else {
3253 struct tcp_sock *tp = tcp_sk(sk);
3254
3255 if (optlen != sizeof(int))
3256 return -EINVAL;
3257
3258 val = *((int *)optval);
3259 /* Only some options are supported */
3260 switch (optname) {
3261 case TCP_BPF_IW:
3262 if (val <= 0 || tp->data_segs_out > tp->syn_data)
3263 ret = -EINVAL;
3264 else
3265 tp->snd_cwnd = val;
3266 break;
3267 case TCP_BPF_SNDCWND_CLAMP:
3268 if (val <= 0) {
3269 ret = -EINVAL;
3270 } else {
3271 tp->snd_cwnd_clamp = val;
3272 tp->snd_ssthresh = val;
3273 }
3274 break;
3275 default:
3276 ret = -EINVAL;
3277 }
3278 }
3279 #endif
3280 } else {
3281 ret = -EINVAL;
3282 }
3283 return ret;
3284 }
3285
3286 static const struct bpf_func_proto bpf_setsockopt_proto = {
3287 .func = bpf_setsockopt,
3288 .gpl_only = false,
3289 .ret_type = RET_INTEGER,
3290 .arg1_type = ARG_PTR_TO_CTX,
3291 .arg2_type = ARG_ANYTHING,
3292 .arg3_type = ARG_ANYTHING,
3293 .arg4_type = ARG_PTR_TO_MEM,
3294 .arg5_type = ARG_CONST_SIZE,
3295 };
3296
3297 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3298 int, level, int, optname, char *, optval, int, optlen)
3299 {
3300 struct sock *sk = bpf_sock->sk;
3301
3302 if (!sk_fullsock(sk))
3303 goto err_clear;
3304
3305 #ifdef CONFIG_INET
3306 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3307 if (optname == TCP_CONGESTION) {
3308 struct inet_connection_sock *icsk = inet_csk(sk);
3309
3310 if (!icsk->icsk_ca_ops || optlen <= 1)
3311 goto err_clear;
3312 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3313 optval[optlen - 1] = 0;
3314 } else {
3315 goto err_clear;
3316 }
3317 } else {
3318 goto err_clear;
3319 }
3320 return 0;
3321 #endif
3322 err_clear:
3323 memset(optval, 0, optlen);
3324 return -EINVAL;
3325 }
3326
3327 static const struct bpf_func_proto bpf_getsockopt_proto = {
3328 .func = bpf_getsockopt,
3329 .gpl_only = false,
3330 .ret_type = RET_INTEGER,
3331 .arg1_type = ARG_PTR_TO_CTX,
3332 .arg2_type = ARG_ANYTHING,
3333 .arg3_type = ARG_ANYTHING,
3334 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
3335 .arg5_type = ARG_CONST_SIZE,
3336 };
3337
3338 static const struct bpf_func_proto *
3339 bpf_base_func_proto(enum bpf_func_id func_id)
3340 {
3341 switch (func_id) {
3342 case BPF_FUNC_map_lookup_elem:
3343 return &bpf_map_lookup_elem_proto;
3344 case BPF_FUNC_map_update_elem:
3345 return &bpf_map_update_elem_proto;
3346 case BPF_FUNC_map_delete_elem:
3347 return &bpf_map_delete_elem_proto;
3348 case BPF_FUNC_get_prandom_u32:
3349 return &bpf_get_prandom_u32_proto;
3350 case BPF_FUNC_get_smp_processor_id:
3351 return &bpf_get_raw_smp_processor_id_proto;
3352 case BPF_FUNC_get_numa_node_id:
3353 return &bpf_get_numa_node_id_proto;
3354 case BPF_FUNC_tail_call:
3355 return &bpf_tail_call_proto;
3356 case BPF_FUNC_ktime_get_ns:
3357 return &bpf_ktime_get_ns_proto;
3358 case BPF_FUNC_trace_printk:
3359 if (capable(CAP_SYS_ADMIN))
3360 return bpf_get_trace_printk_proto();
3361 default:
3362 return NULL;
3363 }
3364 }
3365
3366 static const struct bpf_func_proto *
3367 sock_filter_func_proto(enum bpf_func_id func_id)
3368 {
3369 switch (func_id) {
3370 /* inet and inet6 sockets are created in a process
3371 * context so there is always a valid uid/gid
3372 */
3373 case BPF_FUNC_get_current_uid_gid:
3374 return &bpf_get_current_uid_gid_proto;
3375 default:
3376 return bpf_base_func_proto(func_id);
3377 }
3378 }
3379
3380 static const struct bpf_func_proto *
3381 sk_filter_func_proto(enum bpf_func_id func_id)
3382 {
3383 switch (func_id) {
3384 case BPF_FUNC_skb_load_bytes:
3385 return &bpf_skb_load_bytes_proto;
3386 case BPF_FUNC_get_socket_cookie:
3387 return &bpf_get_socket_cookie_proto;
3388 case BPF_FUNC_get_socket_uid:
3389 return &bpf_get_socket_uid_proto;
3390 default:
3391 return bpf_base_func_proto(func_id);
3392 }
3393 }
3394
3395 static const struct bpf_func_proto *
3396 tc_cls_act_func_proto(enum bpf_func_id func_id)
3397 {
3398 switch (func_id) {
3399 case BPF_FUNC_skb_store_bytes:
3400 return &bpf_skb_store_bytes_proto;
3401 case BPF_FUNC_skb_load_bytes:
3402 return &bpf_skb_load_bytes_proto;
3403 case BPF_FUNC_skb_pull_data:
3404 return &bpf_skb_pull_data_proto;
3405 case BPF_FUNC_csum_diff:
3406 return &bpf_csum_diff_proto;
3407 case BPF_FUNC_csum_update:
3408 return &bpf_csum_update_proto;
3409 case BPF_FUNC_l3_csum_replace:
3410 return &bpf_l3_csum_replace_proto;
3411 case BPF_FUNC_l4_csum_replace:
3412 return &bpf_l4_csum_replace_proto;
3413 case BPF_FUNC_clone_redirect:
3414 return &bpf_clone_redirect_proto;
3415 case BPF_FUNC_get_cgroup_classid:
3416 return &bpf_get_cgroup_classid_proto;
3417 case BPF_FUNC_skb_vlan_push:
3418 return &bpf_skb_vlan_push_proto;
3419 case BPF_FUNC_skb_vlan_pop:
3420 return &bpf_skb_vlan_pop_proto;
3421 case BPF_FUNC_skb_change_proto:
3422 return &bpf_skb_change_proto_proto;
3423 case BPF_FUNC_skb_change_type:
3424 return &bpf_skb_change_type_proto;
3425 case BPF_FUNC_skb_adjust_room:
3426 return &bpf_skb_adjust_room_proto;
3427 case BPF_FUNC_skb_change_tail:
3428 return &bpf_skb_change_tail_proto;
3429 case BPF_FUNC_skb_get_tunnel_key:
3430 return &bpf_skb_get_tunnel_key_proto;
3431 case BPF_FUNC_skb_set_tunnel_key:
3432 return bpf_get_skb_set_tunnel_proto(func_id);
3433 case BPF_FUNC_skb_get_tunnel_opt:
3434 return &bpf_skb_get_tunnel_opt_proto;
3435 case BPF_FUNC_skb_set_tunnel_opt:
3436 return bpf_get_skb_set_tunnel_proto(func_id);
3437 case BPF_FUNC_redirect:
3438 return &bpf_redirect_proto;
3439 case BPF_FUNC_get_route_realm:
3440 return &bpf_get_route_realm_proto;
3441 case BPF_FUNC_get_hash_recalc:
3442 return &bpf_get_hash_recalc_proto;
3443 case BPF_FUNC_set_hash_invalid:
3444 return &bpf_set_hash_invalid_proto;
3445 case BPF_FUNC_set_hash:
3446 return &bpf_set_hash_proto;
3447 case BPF_FUNC_perf_event_output:
3448 return &bpf_skb_event_output_proto;
3449 case BPF_FUNC_get_smp_processor_id:
3450 return &bpf_get_smp_processor_id_proto;
3451 case BPF_FUNC_skb_under_cgroup:
3452 return &bpf_skb_under_cgroup_proto;
3453 case BPF_FUNC_get_socket_cookie:
3454 return &bpf_get_socket_cookie_proto;
3455 case BPF_FUNC_get_socket_uid:
3456 return &bpf_get_socket_uid_proto;
3457 default:
3458 return bpf_base_func_proto(func_id);
3459 }
3460 }
3461
3462 static const struct bpf_func_proto *
3463 xdp_func_proto(enum bpf_func_id func_id)
3464 {
3465 switch (func_id) {
3466 case BPF_FUNC_perf_event_output:
3467 return &bpf_xdp_event_output_proto;
3468 case BPF_FUNC_get_smp_processor_id:
3469 return &bpf_get_smp_processor_id_proto;
3470 case BPF_FUNC_xdp_adjust_head:
3471 return &bpf_xdp_adjust_head_proto;
3472 case BPF_FUNC_xdp_adjust_meta:
3473 return &bpf_xdp_adjust_meta_proto;
3474 case BPF_FUNC_redirect:
3475 return &bpf_xdp_redirect_proto;
3476 case BPF_FUNC_redirect_map:
3477 return &bpf_xdp_redirect_map_proto;
3478 default:
3479 return bpf_base_func_proto(func_id);
3480 }
3481 }
3482
3483 static const struct bpf_func_proto *
3484 lwt_inout_func_proto(enum bpf_func_id func_id)
3485 {
3486 switch (func_id) {
3487 case BPF_FUNC_skb_load_bytes:
3488 return &bpf_skb_load_bytes_proto;
3489 case BPF_FUNC_skb_pull_data:
3490 return &bpf_skb_pull_data_proto;
3491 case BPF_FUNC_csum_diff:
3492 return &bpf_csum_diff_proto;
3493 case BPF_FUNC_get_cgroup_classid:
3494 return &bpf_get_cgroup_classid_proto;
3495 case BPF_FUNC_get_route_realm:
3496 return &bpf_get_route_realm_proto;
3497 case BPF_FUNC_get_hash_recalc:
3498 return &bpf_get_hash_recalc_proto;
3499 case BPF_FUNC_perf_event_output:
3500 return &bpf_skb_event_output_proto;
3501 case BPF_FUNC_get_smp_processor_id:
3502 return &bpf_get_smp_processor_id_proto;
3503 case BPF_FUNC_skb_under_cgroup:
3504 return &bpf_skb_under_cgroup_proto;
3505 default:
3506 return bpf_base_func_proto(func_id);
3507 }
3508 }
3509
3510 static const struct bpf_func_proto *
3511 sock_ops_func_proto(enum bpf_func_id func_id)
3512 {
3513 switch (func_id) {
3514 case BPF_FUNC_setsockopt:
3515 return &bpf_setsockopt_proto;
3516 case BPF_FUNC_getsockopt:
3517 return &bpf_getsockopt_proto;
3518 case BPF_FUNC_sock_map_update:
3519 return &bpf_sock_map_update_proto;
3520 default:
3521 return bpf_base_func_proto(func_id);
3522 }
3523 }
3524
3525 static const struct bpf_func_proto *sk_skb_func_proto(enum bpf_func_id func_id)
3526 {
3527 switch (func_id) {
3528 case BPF_FUNC_skb_store_bytes:
3529 return &bpf_skb_store_bytes_proto;
3530 case BPF_FUNC_skb_load_bytes:
3531 return &bpf_skb_load_bytes_proto;
3532 case BPF_FUNC_skb_pull_data:
3533 return &bpf_skb_pull_data_proto;
3534 case BPF_FUNC_skb_change_tail:
3535 return &bpf_skb_change_tail_proto;
3536 case BPF_FUNC_skb_change_head:
3537 return &bpf_skb_change_head_proto;
3538 case BPF_FUNC_get_socket_cookie:
3539 return &bpf_get_socket_cookie_proto;
3540 case BPF_FUNC_get_socket_uid:
3541 return &bpf_get_socket_uid_proto;
3542 case BPF_FUNC_sk_redirect_map:
3543 return &bpf_sk_redirect_map_proto;
3544 default:
3545 return bpf_base_func_proto(func_id);
3546 }
3547 }
3548
3549 static const struct bpf_func_proto *
3550 lwt_xmit_func_proto(enum bpf_func_id func_id)
3551 {
3552 switch (func_id) {
3553 case BPF_FUNC_skb_get_tunnel_key:
3554 return &bpf_skb_get_tunnel_key_proto;
3555 case BPF_FUNC_skb_set_tunnel_key:
3556 return bpf_get_skb_set_tunnel_proto(func_id);
3557 case BPF_FUNC_skb_get_tunnel_opt:
3558 return &bpf_skb_get_tunnel_opt_proto;
3559 case BPF_FUNC_skb_set_tunnel_opt:
3560 return bpf_get_skb_set_tunnel_proto(func_id);
3561 case BPF_FUNC_redirect:
3562 return &bpf_redirect_proto;
3563 case BPF_FUNC_clone_redirect:
3564 return &bpf_clone_redirect_proto;
3565 case BPF_FUNC_skb_change_tail:
3566 return &bpf_skb_change_tail_proto;
3567 case BPF_FUNC_skb_change_head:
3568 return &bpf_skb_change_head_proto;
3569 case BPF_FUNC_skb_store_bytes:
3570 return &bpf_skb_store_bytes_proto;
3571 case BPF_FUNC_csum_update:
3572 return &bpf_csum_update_proto;
3573 case BPF_FUNC_l3_csum_replace:
3574 return &bpf_l3_csum_replace_proto;
3575 case BPF_FUNC_l4_csum_replace:
3576 return &bpf_l4_csum_replace_proto;
3577 case BPF_FUNC_set_hash_invalid:
3578 return &bpf_set_hash_invalid_proto;
3579 default:
3580 return lwt_inout_func_proto(func_id);
3581 }
3582 }
3583
3584 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
3585 struct bpf_insn_access_aux *info)
3586 {
3587 const int size_default = sizeof(__u32);
3588
3589 if (off < 0 || off >= sizeof(struct __sk_buff))
3590 return false;
3591
3592 /* The verifier guarantees that size > 0. */
3593 if (off % size != 0)
3594 return false;
3595
3596 switch (off) {
3597 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3598 if (off + size > offsetofend(struct __sk_buff, cb[4]))
3599 return false;
3600 break;
3601 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
3602 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
3603 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
3604 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
3605 case bpf_ctx_range(struct __sk_buff, data):
3606 case bpf_ctx_range(struct __sk_buff, data_meta):
3607 case bpf_ctx_range(struct __sk_buff, data_end):
3608 if (size != size_default)
3609 return false;
3610 break;
3611 default:
3612 /* Only narrow read access allowed for now. */
3613 if (type == BPF_WRITE) {
3614 if (size != size_default)
3615 return false;
3616 } else {
3617 bpf_ctx_record_field_size(info, size_default);
3618 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
3619 return false;
3620 }
3621 }
3622
3623 return true;
3624 }
3625
3626 static bool sk_filter_is_valid_access(int off, int size,
3627 enum bpf_access_type type,
3628 struct bpf_insn_access_aux *info)
3629 {
3630 switch (off) {
3631 case bpf_ctx_range(struct __sk_buff, tc_classid):
3632 case bpf_ctx_range(struct __sk_buff, data):
3633 case bpf_ctx_range(struct __sk_buff, data_meta):
3634 case bpf_ctx_range(struct __sk_buff, data_end):
3635 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3636 return false;
3637 }
3638
3639 if (type == BPF_WRITE) {
3640 switch (off) {
3641 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3642 break;
3643 default:
3644 return false;
3645 }
3646 }
3647
3648 return bpf_skb_is_valid_access(off, size, type, info);
3649 }
3650
3651 static bool lwt_is_valid_access(int off, int size,
3652 enum bpf_access_type type,
3653 struct bpf_insn_access_aux *info)
3654 {
3655 switch (off) {
3656 case bpf_ctx_range(struct __sk_buff, tc_classid):
3657 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3658 case bpf_ctx_range(struct __sk_buff, data_meta):
3659 return false;
3660 }
3661
3662 if (type == BPF_WRITE) {
3663 switch (off) {
3664 case bpf_ctx_range(struct __sk_buff, mark):
3665 case bpf_ctx_range(struct __sk_buff, priority):
3666 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3667 break;
3668 default:
3669 return false;
3670 }
3671 }
3672
3673 switch (off) {
3674 case bpf_ctx_range(struct __sk_buff, data):
3675 info->reg_type = PTR_TO_PACKET;
3676 break;
3677 case bpf_ctx_range(struct __sk_buff, data_end):
3678 info->reg_type = PTR_TO_PACKET_END;
3679 break;
3680 }
3681
3682 return bpf_skb_is_valid_access(off, size, type, info);
3683 }
3684
3685 static bool sock_filter_is_valid_access(int off, int size,
3686 enum bpf_access_type type,
3687 struct bpf_insn_access_aux *info)
3688 {
3689 if (type == BPF_WRITE) {
3690 switch (off) {
3691 case offsetof(struct bpf_sock, bound_dev_if):
3692 case offsetof(struct bpf_sock, mark):
3693 case offsetof(struct bpf_sock, priority):
3694 break;
3695 default:
3696 return false;
3697 }
3698 }
3699
3700 if (off < 0 || off + size > sizeof(struct bpf_sock))
3701 return false;
3702 /* The verifier guarantees that size > 0. */
3703 if (off % size != 0)
3704 return false;
3705 if (size != sizeof(__u32))
3706 return false;
3707
3708 return true;
3709 }
3710
3711 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
3712 const struct bpf_prog *prog, int drop_verdict)
3713 {
3714 struct bpf_insn *insn = insn_buf;
3715
3716 if (!direct_write)
3717 return 0;
3718
3719 /* if (!skb->cloned)
3720 * goto start;
3721 *
3722 * (Fast-path, otherwise approximation that we might be
3723 * a clone, do the rest in helper.)
3724 */
3725 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
3726 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
3727 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
3728
3729 /* ret = bpf_skb_pull_data(skb, 0); */
3730 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
3731 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
3732 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
3733 BPF_FUNC_skb_pull_data);
3734 /* if (!ret)
3735 * goto restore;
3736 * return TC_ACT_SHOT;
3737 */
3738 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
3739 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
3740 *insn++ = BPF_EXIT_INSN();
3741
3742 /* restore: */
3743 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
3744 /* start: */
3745 *insn++ = prog->insnsi[0];
3746
3747 return insn - insn_buf;
3748 }
3749
3750 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
3751 const struct bpf_prog *prog)
3752 {
3753 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
3754 }
3755
3756 static bool tc_cls_act_is_valid_access(int off, int size,
3757 enum bpf_access_type type,
3758 struct bpf_insn_access_aux *info)
3759 {
3760 if (type == BPF_WRITE) {
3761 switch (off) {
3762 case bpf_ctx_range(struct __sk_buff, mark):
3763 case bpf_ctx_range(struct __sk_buff, tc_index):
3764 case bpf_ctx_range(struct __sk_buff, priority):
3765 case bpf_ctx_range(struct __sk_buff, tc_classid):
3766 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3767 break;
3768 default:
3769 return false;
3770 }
3771 }
3772
3773 switch (off) {
3774 case bpf_ctx_range(struct __sk_buff, data):
3775 info->reg_type = PTR_TO_PACKET;
3776 break;
3777 case bpf_ctx_range(struct __sk_buff, data_meta):
3778 info->reg_type = PTR_TO_PACKET_META;
3779 break;
3780 case bpf_ctx_range(struct __sk_buff, data_end):
3781 info->reg_type = PTR_TO_PACKET_END;
3782 break;
3783 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3784 return false;
3785 }
3786
3787 return bpf_skb_is_valid_access(off, size, type, info);
3788 }
3789
3790 static bool __is_valid_xdp_access(int off, int size)
3791 {
3792 if (off < 0 || off >= sizeof(struct xdp_md))
3793 return false;
3794 if (off % size != 0)
3795 return false;
3796 if (size != sizeof(__u32))
3797 return false;
3798
3799 return true;
3800 }
3801
3802 static bool xdp_is_valid_access(int off, int size,
3803 enum bpf_access_type type,
3804 struct bpf_insn_access_aux *info)
3805 {
3806 if (type == BPF_WRITE)
3807 return false;
3808
3809 switch (off) {
3810 case offsetof(struct xdp_md, data):
3811 info->reg_type = PTR_TO_PACKET;
3812 break;
3813 case offsetof(struct xdp_md, data_meta):
3814 info->reg_type = PTR_TO_PACKET_META;
3815 break;
3816 case offsetof(struct xdp_md, data_end):
3817 info->reg_type = PTR_TO_PACKET_END;
3818 break;
3819 }
3820
3821 return __is_valid_xdp_access(off, size);
3822 }
3823
3824 void bpf_warn_invalid_xdp_action(u32 act)
3825 {
3826 const u32 act_max = XDP_REDIRECT;
3827
3828 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
3829 act > act_max ? "Illegal" : "Driver unsupported",
3830 act);
3831 }
3832 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
3833
3834 static bool __is_valid_sock_ops_access(int off, int size)
3835 {
3836 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
3837 return false;
3838 /* The verifier guarantees that size > 0. */
3839 if (off % size != 0)
3840 return false;
3841 if (size != sizeof(__u32))
3842 return false;
3843
3844 return true;
3845 }
3846
3847 static bool sock_ops_is_valid_access(int off, int size,
3848 enum bpf_access_type type,
3849 struct bpf_insn_access_aux *info)
3850 {
3851 if (type == BPF_WRITE) {
3852 switch (off) {
3853 case offsetof(struct bpf_sock_ops, op) ...
3854 offsetof(struct bpf_sock_ops, replylong[3]):
3855 break;
3856 default:
3857 return false;
3858 }
3859 }
3860
3861 return __is_valid_sock_ops_access(off, size);
3862 }
3863
3864 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
3865 const struct bpf_prog *prog)
3866 {
3867 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
3868 }
3869
3870 static bool sk_skb_is_valid_access(int off, int size,
3871 enum bpf_access_type type,
3872 struct bpf_insn_access_aux *info)
3873 {
3874 switch (off) {
3875 case bpf_ctx_range(struct __sk_buff, tc_classid):
3876 case bpf_ctx_range(struct __sk_buff, data_meta):
3877 return false;
3878 }
3879
3880 if (type == BPF_WRITE) {
3881 switch (off) {
3882 case bpf_ctx_range(struct __sk_buff, tc_index):
3883 case bpf_ctx_range(struct __sk_buff, priority):
3884 break;
3885 default:
3886 return false;
3887 }
3888 }
3889
3890 switch (off) {
3891 case bpf_ctx_range(struct __sk_buff, mark):
3892 return false;
3893 case bpf_ctx_range(struct __sk_buff, data):
3894 info->reg_type = PTR_TO_PACKET;
3895 break;
3896 case bpf_ctx_range(struct __sk_buff, data_end):
3897 info->reg_type = PTR_TO_PACKET_END;
3898 break;
3899 }
3900
3901 return bpf_skb_is_valid_access(off, size, type, info);
3902 }
3903
3904 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
3905 const struct bpf_insn *si,
3906 struct bpf_insn *insn_buf,
3907 struct bpf_prog *prog, u32 *target_size)
3908 {
3909 struct bpf_insn *insn = insn_buf;
3910 int off;
3911
3912 switch (si->off) {
3913 case offsetof(struct __sk_buff, len):
3914 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3915 bpf_target_off(struct sk_buff, len, 4,
3916 target_size));
3917 break;
3918
3919 case offsetof(struct __sk_buff, protocol):
3920 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3921 bpf_target_off(struct sk_buff, protocol, 2,
3922 target_size));
3923 break;
3924
3925 case offsetof(struct __sk_buff, vlan_proto):
3926 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3927 bpf_target_off(struct sk_buff, vlan_proto, 2,
3928 target_size));
3929 break;
3930
3931 case offsetof(struct __sk_buff, priority):
3932 if (type == BPF_WRITE)
3933 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3934 bpf_target_off(struct sk_buff, priority, 4,
3935 target_size));
3936 else
3937 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3938 bpf_target_off(struct sk_buff, priority, 4,
3939 target_size));
3940 break;
3941
3942 case offsetof(struct __sk_buff, ingress_ifindex):
3943 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3944 bpf_target_off(struct sk_buff, skb_iif, 4,
3945 target_size));
3946 break;
3947
3948 case offsetof(struct __sk_buff, ifindex):
3949 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
3950 si->dst_reg, si->src_reg,
3951 offsetof(struct sk_buff, dev));
3952 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
3953 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3954 bpf_target_off(struct net_device, ifindex, 4,
3955 target_size));
3956 break;
3957
3958 case offsetof(struct __sk_buff, hash):
3959 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3960 bpf_target_off(struct sk_buff, hash, 4,
3961 target_size));
3962 break;
3963
3964 case offsetof(struct __sk_buff, mark):
3965 if (type == BPF_WRITE)
3966 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3967 bpf_target_off(struct sk_buff, mark, 4,
3968 target_size));
3969 else
3970 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3971 bpf_target_off(struct sk_buff, mark, 4,
3972 target_size));
3973 break;
3974
3975 case offsetof(struct __sk_buff, pkt_type):
3976 *target_size = 1;
3977 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
3978 PKT_TYPE_OFFSET());
3979 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
3980 #ifdef __BIG_ENDIAN_BITFIELD
3981 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
3982 #endif
3983 break;
3984
3985 case offsetof(struct __sk_buff, queue_mapping):
3986 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3987 bpf_target_off(struct sk_buff, queue_mapping, 2,
3988 target_size));
3989 break;
3990
3991 case offsetof(struct __sk_buff, vlan_present):
3992 case offsetof(struct __sk_buff, vlan_tci):
3993 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
3994
3995 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3996 bpf_target_off(struct sk_buff, vlan_tci, 2,
3997 target_size));
3998 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
3999 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
4000 ~VLAN_TAG_PRESENT);
4001 } else {
4002 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
4003 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
4004 }
4005 break;
4006
4007 case offsetof(struct __sk_buff, cb[0]) ...
4008 offsetofend(struct __sk_buff, cb[4]) - 1:
4009 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
4010 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
4011 offsetof(struct qdisc_skb_cb, data)) %
4012 sizeof(__u64));
4013
4014 prog->cb_access = 1;
4015 off = si->off;
4016 off -= offsetof(struct __sk_buff, cb[0]);
4017 off += offsetof(struct sk_buff, cb);
4018 off += offsetof(struct qdisc_skb_cb, data);
4019 if (type == BPF_WRITE)
4020 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
4021 si->src_reg, off);
4022 else
4023 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
4024 si->src_reg, off);
4025 break;
4026
4027 case offsetof(struct __sk_buff, tc_classid):
4028 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
4029
4030 off = si->off;
4031 off -= offsetof(struct __sk_buff, tc_classid);
4032 off += offsetof(struct sk_buff, cb);
4033 off += offsetof(struct qdisc_skb_cb, tc_classid);
4034 *target_size = 2;
4035 if (type == BPF_WRITE)
4036 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
4037 si->src_reg, off);
4038 else
4039 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
4040 si->src_reg, off);
4041 break;
4042
4043 case offsetof(struct __sk_buff, data):
4044 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
4045 si->dst_reg, si->src_reg,
4046 offsetof(struct sk_buff, data));
4047 break;
4048
4049 case offsetof(struct __sk_buff, data_meta):
4050 off = si->off;
4051 off -= offsetof(struct __sk_buff, data_meta);
4052 off += offsetof(struct sk_buff, cb);
4053 off += offsetof(struct bpf_skb_data_end, data_meta);
4054 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4055 si->src_reg, off);
4056 break;
4057
4058 case offsetof(struct __sk_buff, data_end):
4059 off = si->off;
4060 off -= offsetof(struct __sk_buff, data_end);
4061 off += offsetof(struct sk_buff, cb);
4062 off += offsetof(struct bpf_skb_data_end, data_end);
4063 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4064 si->src_reg, off);
4065 break;
4066
4067 case offsetof(struct __sk_buff, tc_index):
4068 #ifdef CONFIG_NET_SCHED
4069 if (type == BPF_WRITE)
4070 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
4071 bpf_target_off(struct sk_buff, tc_index, 2,
4072 target_size));
4073 else
4074 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4075 bpf_target_off(struct sk_buff, tc_index, 2,
4076 target_size));
4077 #else
4078 *target_size = 2;
4079 if (type == BPF_WRITE)
4080 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
4081 else
4082 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4083 #endif
4084 break;
4085
4086 case offsetof(struct __sk_buff, napi_id):
4087 #if defined(CONFIG_NET_RX_BUSY_POLL)
4088 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4089 bpf_target_off(struct sk_buff, napi_id, 4,
4090 target_size));
4091 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
4092 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4093 #else
4094 *target_size = 4;
4095 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
4096 #endif
4097 break;
4098 case offsetof(struct __sk_buff, family):
4099 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4100
4101 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4102 si->dst_reg, si->src_reg,
4103 offsetof(struct sk_buff, sk));
4104 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4105 bpf_target_off(struct sock_common,
4106 skc_family,
4107 2, target_size));
4108 break;
4109 case offsetof(struct __sk_buff, remote_ip4):
4110 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4111
4112 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4113 si->dst_reg, si->src_reg,
4114 offsetof(struct sk_buff, sk));
4115 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4116 bpf_target_off(struct sock_common,
4117 skc_daddr,
4118 4, target_size));
4119 break;
4120 case offsetof(struct __sk_buff, local_ip4):
4121 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4122 skc_rcv_saddr) != 4);
4123
4124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4125 si->dst_reg, si->src_reg,
4126 offsetof(struct sk_buff, sk));
4127 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4128 bpf_target_off(struct sock_common,
4129 skc_rcv_saddr,
4130 4, target_size));
4131 break;
4132 case offsetof(struct __sk_buff, remote_ip6[0]) ...
4133 offsetof(struct __sk_buff, remote_ip6[3]):
4134 #if IS_ENABLED(CONFIG_IPV6)
4135 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4136 skc_v6_daddr.s6_addr32[0]) != 4);
4137
4138 off = si->off;
4139 off -= offsetof(struct __sk_buff, remote_ip6[0]);
4140
4141 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4142 si->dst_reg, si->src_reg,
4143 offsetof(struct sk_buff, sk));
4144 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4145 offsetof(struct sock_common,
4146 skc_v6_daddr.s6_addr32[0]) +
4147 off);
4148 #else
4149 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4150 #endif
4151 break;
4152 case offsetof(struct __sk_buff, local_ip6[0]) ...
4153 offsetof(struct __sk_buff, local_ip6[3]):
4154 #if IS_ENABLED(CONFIG_IPV6)
4155 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4156 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4157
4158 off = si->off;
4159 off -= offsetof(struct __sk_buff, local_ip6[0]);
4160
4161 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4162 si->dst_reg, si->src_reg,
4163 offsetof(struct sk_buff, sk));
4164 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4165 offsetof(struct sock_common,
4166 skc_v6_rcv_saddr.s6_addr32[0]) +
4167 off);
4168 #else
4169 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4170 #endif
4171 break;
4172
4173 case offsetof(struct __sk_buff, remote_port):
4174 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4175
4176 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4177 si->dst_reg, si->src_reg,
4178 offsetof(struct sk_buff, sk));
4179 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4180 bpf_target_off(struct sock_common,
4181 skc_dport,
4182 2, target_size));
4183 #ifndef __BIG_ENDIAN_BITFIELD
4184 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4185 #endif
4186 break;
4187
4188 case offsetof(struct __sk_buff, local_port):
4189 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4190
4191 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4192 si->dst_reg, si->src_reg,
4193 offsetof(struct sk_buff, sk));
4194 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4195 bpf_target_off(struct sock_common,
4196 skc_num, 2, target_size));
4197 break;
4198 }
4199
4200 return insn - insn_buf;
4201 }
4202
4203 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
4204 const struct bpf_insn *si,
4205 struct bpf_insn *insn_buf,
4206 struct bpf_prog *prog, u32 *target_size)
4207 {
4208 struct bpf_insn *insn = insn_buf;
4209
4210 switch (si->off) {
4211 case offsetof(struct bpf_sock, bound_dev_if):
4212 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
4213
4214 if (type == BPF_WRITE)
4215 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4216 offsetof(struct sock, sk_bound_dev_if));
4217 else
4218 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4219 offsetof(struct sock, sk_bound_dev_if));
4220 break;
4221
4222 case offsetof(struct bpf_sock, mark):
4223 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
4224
4225 if (type == BPF_WRITE)
4226 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4227 offsetof(struct sock, sk_mark));
4228 else
4229 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4230 offsetof(struct sock, sk_mark));
4231 break;
4232
4233 case offsetof(struct bpf_sock, priority):
4234 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
4235
4236 if (type == BPF_WRITE)
4237 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4238 offsetof(struct sock, sk_priority));
4239 else
4240 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4241 offsetof(struct sock, sk_priority));
4242 break;
4243
4244 case offsetof(struct bpf_sock, family):
4245 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
4246
4247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4248 offsetof(struct sock, sk_family));
4249 break;
4250
4251 case offsetof(struct bpf_sock, type):
4252 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4253 offsetof(struct sock, __sk_flags_offset));
4254 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
4255 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
4256 break;
4257
4258 case offsetof(struct bpf_sock, protocol):
4259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4260 offsetof(struct sock, __sk_flags_offset));
4261 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
4262 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
4263 break;
4264 }
4265
4266 return insn - insn_buf;
4267 }
4268
4269 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
4270 const struct bpf_insn *si,
4271 struct bpf_insn *insn_buf,
4272 struct bpf_prog *prog, u32 *target_size)
4273 {
4274 struct bpf_insn *insn = insn_buf;
4275
4276 switch (si->off) {
4277 case offsetof(struct __sk_buff, ifindex):
4278 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4279 si->dst_reg, si->src_reg,
4280 offsetof(struct sk_buff, dev));
4281 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4282 bpf_target_off(struct net_device, ifindex, 4,
4283 target_size));
4284 break;
4285 default:
4286 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4287 target_size);
4288 }
4289
4290 return insn - insn_buf;
4291 }
4292
4293 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
4294 const struct bpf_insn *si,
4295 struct bpf_insn *insn_buf,
4296 struct bpf_prog *prog, u32 *target_size)
4297 {
4298 struct bpf_insn *insn = insn_buf;
4299
4300 switch (si->off) {
4301 case offsetof(struct xdp_md, data):
4302 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
4303 si->dst_reg, si->src_reg,
4304 offsetof(struct xdp_buff, data));
4305 break;
4306 case offsetof(struct xdp_md, data_meta):
4307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
4308 si->dst_reg, si->src_reg,
4309 offsetof(struct xdp_buff, data_meta));
4310 break;
4311 case offsetof(struct xdp_md, data_end):
4312 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
4313 si->dst_reg, si->src_reg,
4314 offsetof(struct xdp_buff, data_end));
4315 break;
4316 }
4317
4318 return insn - insn_buf;
4319 }
4320
4321 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
4322 const struct bpf_insn *si,
4323 struct bpf_insn *insn_buf,
4324 struct bpf_prog *prog,
4325 u32 *target_size)
4326 {
4327 struct bpf_insn *insn = insn_buf;
4328 int off;
4329
4330 switch (si->off) {
4331 case offsetof(struct bpf_sock_ops, op) ...
4332 offsetof(struct bpf_sock_ops, replylong[3]):
4333 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
4334 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
4335 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
4336 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
4337 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
4338 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
4339 off = si->off;
4340 off -= offsetof(struct bpf_sock_ops, op);
4341 off += offsetof(struct bpf_sock_ops_kern, op);
4342 if (type == BPF_WRITE)
4343 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4344 off);
4345 else
4346 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4347 off);
4348 break;
4349
4350 case offsetof(struct bpf_sock_ops, family):
4351 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4352
4353 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4354 struct bpf_sock_ops_kern, sk),
4355 si->dst_reg, si->src_reg,
4356 offsetof(struct bpf_sock_ops_kern, sk));
4357 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4358 offsetof(struct sock_common, skc_family));
4359 break;
4360
4361 case offsetof(struct bpf_sock_ops, remote_ip4):
4362 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4363
4364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4365 struct bpf_sock_ops_kern, sk),
4366 si->dst_reg, si->src_reg,
4367 offsetof(struct bpf_sock_ops_kern, sk));
4368 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4369 offsetof(struct sock_common, skc_daddr));
4370 break;
4371
4372 case offsetof(struct bpf_sock_ops, local_ip4):
4373 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4);
4374
4375 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4376 struct bpf_sock_ops_kern, sk),
4377 si->dst_reg, si->src_reg,
4378 offsetof(struct bpf_sock_ops_kern, sk));
4379 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4380 offsetof(struct sock_common,
4381 skc_rcv_saddr));
4382 break;
4383
4384 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
4385 offsetof(struct bpf_sock_ops, remote_ip6[3]):
4386 #if IS_ENABLED(CONFIG_IPV6)
4387 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4388 skc_v6_daddr.s6_addr32[0]) != 4);
4389
4390 off = si->off;
4391 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
4392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4393 struct bpf_sock_ops_kern, sk),
4394 si->dst_reg, si->src_reg,
4395 offsetof(struct bpf_sock_ops_kern, sk));
4396 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4397 offsetof(struct sock_common,
4398 skc_v6_daddr.s6_addr32[0]) +
4399 off);
4400 #else
4401 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4402 #endif
4403 break;
4404
4405 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
4406 offsetof(struct bpf_sock_ops, local_ip6[3]):
4407 #if IS_ENABLED(CONFIG_IPV6)
4408 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4409 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4410
4411 off = si->off;
4412 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
4413 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4414 struct bpf_sock_ops_kern, sk),
4415 si->dst_reg, si->src_reg,
4416 offsetof(struct bpf_sock_ops_kern, sk));
4417 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4418 offsetof(struct sock_common,
4419 skc_v6_rcv_saddr.s6_addr32[0]) +
4420 off);
4421 #else
4422 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4423 #endif
4424 break;
4425
4426 case offsetof(struct bpf_sock_ops, remote_port):
4427 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4428
4429 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4430 struct bpf_sock_ops_kern, sk),
4431 si->dst_reg, si->src_reg,
4432 offsetof(struct bpf_sock_ops_kern, sk));
4433 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4434 offsetof(struct sock_common, skc_dport));
4435 #ifndef __BIG_ENDIAN_BITFIELD
4436 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4437 #endif
4438 break;
4439
4440 case offsetof(struct bpf_sock_ops, local_port):
4441 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4442
4443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4444 struct bpf_sock_ops_kern, sk),
4445 si->dst_reg, si->src_reg,
4446 offsetof(struct bpf_sock_ops_kern, sk));
4447 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4448 offsetof(struct sock_common, skc_num));
4449 break;
4450 }
4451 return insn - insn_buf;
4452 }
4453
4454 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
4455 const struct bpf_insn *si,
4456 struct bpf_insn *insn_buf,
4457 struct bpf_prog *prog, u32 *target_size)
4458 {
4459 struct bpf_insn *insn = insn_buf;
4460 int off;
4461
4462 switch (si->off) {
4463 case offsetof(struct __sk_buff, data_end):
4464 off = si->off;
4465 off -= offsetof(struct __sk_buff, data_end);
4466 off += offsetof(struct sk_buff, cb);
4467 off += offsetof(struct tcp_skb_cb, bpf.data_end);
4468 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4469 si->src_reg, off);
4470 break;
4471 default:
4472 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4473 target_size);
4474 }
4475
4476 return insn - insn_buf;
4477 }
4478
4479 const struct bpf_verifier_ops sk_filter_verifier_ops = {
4480 .get_func_proto = sk_filter_func_proto,
4481 .is_valid_access = sk_filter_is_valid_access,
4482 .convert_ctx_access = bpf_convert_ctx_access,
4483 };
4484
4485 const struct bpf_prog_ops sk_filter_prog_ops = {
4486 };
4487
4488 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
4489 .get_func_proto = tc_cls_act_func_proto,
4490 .is_valid_access = tc_cls_act_is_valid_access,
4491 .convert_ctx_access = tc_cls_act_convert_ctx_access,
4492 .gen_prologue = tc_cls_act_prologue,
4493 };
4494
4495 const struct bpf_prog_ops tc_cls_act_prog_ops = {
4496 .test_run = bpf_prog_test_run_skb,
4497 };
4498
4499 const struct bpf_verifier_ops xdp_verifier_ops = {
4500 .get_func_proto = xdp_func_proto,
4501 .is_valid_access = xdp_is_valid_access,
4502 .convert_ctx_access = xdp_convert_ctx_access,
4503 };
4504
4505 const struct bpf_prog_ops xdp_prog_ops = {
4506 .test_run = bpf_prog_test_run_xdp,
4507 };
4508
4509 const struct bpf_verifier_ops cg_skb_verifier_ops = {
4510 .get_func_proto = sk_filter_func_proto,
4511 .is_valid_access = sk_filter_is_valid_access,
4512 .convert_ctx_access = bpf_convert_ctx_access,
4513 };
4514
4515 const struct bpf_prog_ops cg_skb_prog_ops = {
4516 .test_run = bpf_prog_test_run_skb,
4517 };
4518
4519 const struct bpf_verifier_ops lwt_inout_verifier_ops = {
4520 .get_func_proto = lwt_inout_func_proto,
4521 .is_valid_access = lwt_is_valid_access,
4522 .convert_ctx_access = bpf_convert_ctx_access,
4523 };
4524
4525 const struct bpf_prog_ops lwt_inout_prog_ops = {
4526 .test_run = bpf_prog_test_run_skb,
4527 };
4528
4529 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
4530 .get_func_proto = lwt_xmit_func_proto,
4531 .is_valid_access = lwt_is_valid_access,
4532 .convert_ctx_access = bpf_convert_ctx_access,
4533 .gen_prologue = tc_cls_act_prologue,
4534 };
4535
4536 const struct bpf_prog_ops lwt_xmit_prog_ops = {
4537 .test_run = bpf_prog_test_run_skb,
4538 };
4539
4540 const struct bpf_verifier_ops cg_sock_verifier_ops = {
4541 .get_func_proto = sock_filter_func_proto,
4542 .is_valid_access = sock_filter_is_valid_access,
4543 .convert_ctx_access = sock_filter_convert_ctx_access,
4544 };
4545
4546 const struct bpf_prog_ops cg_sock_prog_ops = {
4547 };
4548
4549 const struct bpf_verifier_ops sock_ops_verifier_ops = {
4550 .get_func_proto = sock_ops_func_proto,
4551 .is_valid_access = sock_ops_is_valid_access,
4552 .convert_ctx_access = sock_ops_convert_ctx_access,
4553 };
4554
4555 const struct bpf_prog_ops sock_ops_prog_ops = {
4556 };
4557
4558 const struct bpf_verifier_ops sk_skb_verifier_ops = {
4559 .get_func_proto = sk_skb_func_proto,
4560 .is_valid_access = sk_skb_is_valid_access,
4561 .convert_ctx_access = sk_skb_convert_ctx_access,
4562 .gen_prologue = sk_skb_prologue,
4563 };
4564
4565 const struct bpf_prog_ops sk_skb_prog_ops = {
4566 };
4567
4568 int sk_detach_filter(struct sock *sk)
4569 {
4570 int ret = -ENOENT;
4571 struct sk_filter *filter;
4572
4573 if (sock_flag(sk, SOCK_FILTER_LOCKED))
4574 return -EPERM;
4575
4576 filter = rcu_dereference_protected(sk->sk_filter,
4577 lockdep_sock_is_held(sk));
4578 if (filter) {
4579 RCU_INIT_POINTER(sk->sk_filter, NULL);
4580 sk_filter_uncharge(sk, filter);
4581 ret = 0;
4582 }
4583
4584 return ret;
4585 }
4586 EXPORT_SYMBOL_GPL(sk_detach_filter);
4587
4588 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
4589 unsigned int len)
4590 {
4591 struct sock_fprog_kern *fprog;
4592 struct sk_filter *filter;
4593 int ret = 0;
4594
4595 lock_sock(sk);
4596 filter = rcu_dereference_protected(sk->sk_filter,
4597 lockdep_sock_is_held(sk));
4598 if (!filter)
4599 goto out;
4600
4601 /* We're copying the filter that has been originally attached,
4602 * so no conversion/decode needed anymore. eBPF programs that
4603 * have no original program cannot be dumped through this.
4604 */
4605 ret = -EACCES;
4606 fprog = filter->prog->orig_prog;
4607 if (!fprog)
4608 goto out;
4609
4610 ret = fprog->len;
4611 if (!len)
4612 /* User space only enquires number of filter blocks. */
4613 goto out;
4614
4615 ret = -EINVAL;
4616 if (len < fprog->len)
4617 goto out;
4618
4619 ret = -EFAULT;
4620 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
4621 goto out;
4622
4623 /* Instead of bytes, the API requests to return the number
4624 * of filter blocks.
4625 */
4626 ret = fprog->len;
4627 out:
4628 release_sock(sk);
4629 return ret;
4630 }
ubuntu-bionic-kernel mirror