2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
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.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/sock_diag.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>
37 #include <net/protocol.h>
38 #include <net/netlink.h>
39 #include <linux/skbuff.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
58 #include <linux/bpf_trace.h>
61 * sk_filter_trim_cap - run a packet through a socket filter
62 * @sk: sock associated with &sk_buff
63 * @skb: buffer to filter
64 * @cap: limit on how short the eBPF program may trim the packet
66 * Run the eBPF program and then cut skb->data to correct size returned by
67 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
68 * than pkt_len we keep whole skb->data. This is the socket level
69 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
70 * be accepted or -EPERM if the packet should be tossed.
73 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
76 struct sk_filter *filter;
79 * If the skb was allocated from pfmemalloc reserves, only
80 * allow SOCK_MEMALLOC sockets to use it as this socket is
83 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
84 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
87 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
91 err = security_sock_rcv_skb(sk, skb);
96 filter = rcu_dereference(sk->sk_filter);
98 struct sock *save_sk = skb->sk;
102 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
104 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
110 EXPORT_SYMBOL(sk_filter_trim_cap);
112 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb)
114 return skb_get_poff(skb);
117 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
121 if (skb_is_nonlinear(skb))
124 if (skb->len < sizeof(struct nlattr))
127 if (a > skb->len - sizeof(struct nlattr))
130 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
132 return (void *) nla - (void *) skb->data;
137 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
141 if (skb_is_nonlinear(skb))
144 if (skb->len < sizeof(struct nlattr))
147 if (a > skb->len - sizeof(struct nlattr))
150 nla = (struct nlattr *) &skb->data[a];
151 if (nla->nla_len > skb->len - a)
154 nla = nla_find_nested(nla, x);
156 return (void *) nla - (void *) skb->data;
161 BPF_CALL_0(__get_raw_cpu_id)
163 return raw_smp_processor_id();
166 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
167 .func = __get_raw_cpu_id,
169 .ret_type = RET_INTEGER,
172 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
173 struct bpf_insn *insn_buf)
175 struct bpf_insn *insn = insn_buf;
179 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
181 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
182 offsetof(struct sk_buff, mark));
186 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
187 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
188 #ifdef __BIG_ENDIAN_BITFIELD
189 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
194 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
196 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
197 offsetof(struct sk_buff, queue_mapping));
200 case SKF_AD_VLAN_TAG:
201 case SKF_AD_VLAN_TAG_PRESENT:
202 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
203 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
205 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
206 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
207 offsetof(struct sk_buff, vlan_tci));
208 if (skb_field == SKF_AD_VLAN_TAG) {
209 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
213 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
215 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
220 return insn - insn_buf;
223 static bool convert_bpf_extensions(struct sock_filter *fp,
224 struct bpf_insn **insnp)
226 struct bpf_insn *insn = *insnp;
230 case SKF_AD_OFF + SKF_AD_PROTOCOL:
231 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
233 /* A = *(u16 *) (CTX + offsetof(protocol)) */
234 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
235 offsetof(struct sk_buff, protocol));
236 /* A = ntohs(A) [emitting a nop or swap16] */
237 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
240 case SKF_AD_OFF + SKF_AD_PKTTYPE:
241 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
245 case SKF_AD_OFF + SKF_AD_IFINDEX:
246 case SKF_AD_OFF + SKF_AD_HATYPE:
247 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
248 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
251 BPF_REG_TMP, BPF_REG_CTX,
252 offsetof(struct sk_buff, dev));
253 /* if (tmp != 0) goto pc + 1 */
254 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
255 *insn++ = BPF_EXIT_INSN();
256 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
257 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
258 offsetof(struct net_device, ifindex));
260 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
261 offsetof(struct net_device, type));
264 case SKF_AD_OFF + SKF_AD_MARK:
265 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
269 case SKF_AD_OFF + SKF_AD_RXHASH:
270 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
272 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
273 offsetof(struct sk_buff, hash));
276 case SKF_AD_OFF + SKF_AD_QUEUE:
277 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
281 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
282 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
283 BPF_REG_A, BPF_REG_CTX, insn);
287 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
288 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
289 BPF_REG_A, BPF_REG_CTX, insn);
293 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
294 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
296 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
297 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
298 offsetof(struct sk_buff, vlan_proto));
299 /* A = ntohs(A) [emitting a nop or swap16] */
300 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
303 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
304 case SKF_AD_OFF + SKF_AD_NLATTR:
305 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
306 case SKF_AD_OFF + SKF_AD_CPU:
307 case SKF_AD_OFF + SKF_AD_RANDOM:
309 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
311 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
313 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
314 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
316 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
317 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
319 case SKF_AD_OFF + SKF_AD_NLATTR:
320 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
322 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
323 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
325 case SKF_AD_OFF + SKF_AD_CPU:
326 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
328 case SKF_AD_OFF + SKF_AD_RANDOM:
329 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
330 bpf_user_rnd_init_once();
335 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
337 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
341 /* This is just a dummy call to avoid letting the compiler
342 * evict __bpf_call_base() as an optimization. Placed here
343 * where no-one bothers.
345 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
354 * bpf_convert_filter - convert filter program
355 * @prog: the user passed filter program
356 * @len: the length of the user passed filter program
357 * @new_prog: allocated 'struct bpf_prog' or NULL
358 * @new_len: pointer to store length of converted program
360 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
361 * style extended BPF (eBPF).
362 * Conversion workflow:
364 * 1) First pass for calculating the new program length:
365 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
367 * 2) 2nd pass to remap in two passes: 1st pass finds new
368 * jump offsets, 2nd pass remapping:
369 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
371 static int bpf_convert_filter(struct sock_filter *prog, int len,
372 struct bpf_prog *new_prog, int *new_len)
374 int new_flen = 0, pass = 0, target, i, stack_off;
375 struct bpf_insn *new_insn, *first_insn = NULL;
376 struct sock_filter *fp;
380 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
381 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
383 if (len <= 0 || len > BPF_MAXINSNS)
387 first_insn = new_prog->insnsi;
388 addrs = kcalloc(len, sizeof(*addrs),
389 GFP_KERNEL | __GFP_NOWARN);
395 new_insn = first_insn;
398 /* Classic BPF related prologue emission. */
400 /* Classic BPF expects A and X to be reset first. These need
401 * to be guaranteed to be the first two instructions.
403 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
404 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
406 /* All programs must keep CTX in callee saved BPF_REG_CTX.
407 * In eBPF case it's done by the compiler, here we need to
408 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
410 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
415 for (i = 0; i < len; fp++, i++) {
416 struct bpf_insn tmp_insns[6] = { };
417 struct bpf_insn *insn = tmp_insns;
420 addrs[i] = new_insn - first_insn;
423 /* All arithmetic insns and skb loads map as-is. */
424 case BPF_ALU | BPF_ADD | BPF_X:
425 case BPF_ALU | BPF_ADD | BPF_K:
426 case BPF_ALU | BPF_SUB | BPF_X:
427 case BPF_ALU | BPF_SUB | BPF_K:
428 case BPF_ALU | BPF_AND | BPF_X:
429 case BPF_ALU | BPF_AND | BPF_K:
430 case BPF_ALU | BPF_OR | BPF_X:
431 case BPF_ALU | BPF_OR | BPF_K:
432 case BPF_ALU | BPF_LSH | BPF_X:
433 case BPF_ALU | BPF_LSH | BPF_K:
434 case BPF_ALU | BPF_RSH | BPF_X:
435 case BPF_ALU | BPF_RSH | BPF_K:
436 case BPF_ALU | BPF_XOR | BPF_X:
437 case BPF_ALU | BPF_XOR | BPF_K:
438 case BPF_ALU | BPF_MUL | BPF_X:
439 case BPF_ALU | BPF_MUL | BPF_K:
440 case BPF_ALU | BPF_DIV | BPF_X:
441 case BPF_ALU | BPF_DIV | BPF_K:
442 case BPF_ALU | BPF_MOD | BPF_X:
443 case BPF_ALU | BPF_MOD | BPF_K:
444 case BPF_ALU | BPF_NEG:
445 case BPF_LD | BPF_ABS | BPF_W:
446 case BPF_LD | BPF_ABS | BPF_H:
447 case BPF_LD | BPF_ABS | BPF_B:
448 case BPF_LD | BPF_IND | BPF_W:
449 case BPF_LD | BPF_IND | BPF_H:
450 case BPF_LD | BPF_IND | BPF_B:
451 /* Check for overloaded BPF extension and
452 * directly convert it if found, otherwise
453 * just move on with mapping.
455 if (BPF_CLASS(fp->code) == BPF_LD &&
456 BPF_MODE(fp->code) == BPF_ABS &&
457 convert_bpf_extensions(fp, &insn))
460 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
461 fp->code == (BPF_ALU | BPF_MOD | BPF_X))
462 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
464 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
467 /* Jump transformation cannot use BPF block macros
468 * everywhere as offset calculation and target updates
469 * require a bit more work than the rest, i.e. jump
470 * opcodes map as-is, but offsets need adjustment.
473 #define BPF_EMIT_JMP \
475 if (target >= len || target < 0) \
477 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
478 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
479 insn->off -= insn - tmp_insns; \
482 case BPF_JMP | BPF_JA:
483 target = i + fp->k + 1;
484 insn->code = fp->code;
488 case BPF_JMP | BPF_JEQ | BPF_K:
489 case BPF_JMP | BPF_JEQ | BPF_X:
490 case BPF_JMP | BPF_JSET | BPF_K:
491 case BPF_JMP | BPF_JSET | BPF_X:
492 case BPF_JMP | BPF_JGT | BPF_K:
493 case BPF_JMP | BPF_JGT | BPF_X:
494 case BPF_JMP | BPF_JGE | BPF_K:
495 case BPF_JMP | BPF_JGE | BPF_X:
496 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
497 /* BPF immediates are signed, zero extend
498 * immediate into tmp register and use it
501 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
503 insn->dst_reg = BPF_REG_A;
504 insn->src_reg = BPF_REG_TMP;
507 insn->dst_reg = BPF_REG_A;
509 bpf_src = BPF_SRC(fp->code);
510 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
513 /* Common case where 'jump_false' is next insn. */
515 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
516 target = i + fp->jt + 1;
521 /* Convert some jumps when 'jump_true' is next insn. */
523 switch (BPF_OP(fp->code)) {
525 insn->code = BPF_JMP | BPF_JNE | bpf_src;
528 insn->code = BPF_JMP | BPF_JLE | bpf_src;
531 insn->code = BPF_JMP | BPF_JLT | bpf_src;
537 target = i + fp->jf + 1;
542 /* Other jumps are mapped into two insns: Jxx and JA. */
543 target = i + fp->jt + 1;
544 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
548 insn->code = BPF_JMP | BPF_JA;
549 target = i + fp->jf + 1;
553 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
554 case BPF_LDX | BPF_MSH | BPF_B:
556 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
557 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
558 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
560 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
562 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
564 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
566 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
569 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
570 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
572 case BPF_RET | BPF_A:
573 case BPF_RET | BPF_K:
574 if (BPF_RVAL(fp->code) == BPF_K)
575 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
577 *insn = BPF_EXIT_INSN();
580 /* Store to stack. */
583 stack_off = fp->k * 4 + 4;
584 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
585 BPF_ST ? BPF_REG_A : BPF_REG_X,
587 /* check_load_and_stores() verifies that classic BPF can
588 * load from stack only after write, so tracking
589 * stack_depth for ST|STX insns is enough
591 if (new_prog && new_prog->aux->stack_depth < stack_off)
592 new_prog->aux->stack_depth = stack_off;
595 /* Load from stack. */
596 case BPF_LD | BPF_MEM:
597 case BPF_LDX | BPF_MEM:
598 stack_off = fp->k * 4 + 4;
599 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
600 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
605 case BPF_LD | BPF_IMM:
606 case BPF_LDX | BPF_IMM:
607 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
608 BPF_REG_A : BPF_REG_X, fp->k);
612 case BPF_MISC | BPF_TAX:
613 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
617 case BPF_MISC | BPF_TXA:
618 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
621 /* A = skb->len or X = skb->len */
622 case BPF_LD | BPF_W | BPF_LEN:
623 case BPF_LDX | BPF_W | BPF_LEN:
624 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
625 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
626 offsetof(struct sk_buff, len));
629 /* Access seccomp_data fields. */
630 case BPF_LDX | BPF_ABS | BPF_W:
631 /* A = *(u32 *) (ctx + K) */
632 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
635 /* Unknown instruction. */
642 memcpy(new_insn, tmp_insns,
643 sizeof(*insn) * (insn - tmp_insns));
644 new_insn += insn - tmp_insns;
648 /* Only calculating new length. */
649 *new_len = new_insn - first_insn;
654 if (new_flen != new_insn - first_insn) {
655 new_flen = new_insn - first_insn;
662 BUG_ON(*new_len != new_flen);
671 * As we dont want to clear mem[] array for each packet going through
672 * __bpf_prog_run(), we check that filter loaded by user never try to read
673 * a cell if not previously written, and we check all branches to be sure
674 * a malicious user doesn't try to abuse us.
676 static int check_load_and_stores(const struct sock_filter *filter, int flen)
678 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
681 BUILD_BUG_ON(BPF_MEMWORDS > 16);
683 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
687 memset(masks, 0xff, flen * sizeof(*masks));
689 for (pc = 0; pc < flen; pc++) {
690 memvalid &= masks[pc];
692 switch (filter[pc].code) {
695 memvalid |= (1 << filter[pc].k);
697 case BPF_LD | BPF_MEM:
698 case BPF_LDX | BPF_MEM:
699 if (!(memvalid & (1 << filter[pc].k))) {
704 case BPF_JMP | BPF_JA:
705 /* A jump must set masks on target */
706 masks[pc + 1 + filter[pc].k] &= memvalid;
709 case BPF_JMP | BPF_JEQ | BPF_K:
710 case BPF_JMP | BPF_JEQ | BPF_X:
711 case BPF_JMP | BPF_JGE | BPF_K:
712 case BPF_JMP | BPF_JGE | BPF_X:
713 case BPF_JMP | BPF_JGT | BPF_K:
714 case BPF_JMP | BPF_JGT | BPF_X:
715 case BPF_JMP | BPF_JSET | BPF_K:
716 case BPF_JMP | BPF_JSET | BPF_X:
717 /* A jump must set masks on targets */
718 masks[pc + 1 + filter[pc].jt] &= memvalid;
719 masks[pc + 1 + filter[pc].jf] &= memvalid;
729 static bool chk_code_allowed(u16 code_to_probe)
731 static const bool codes[] = {
732 /* 32 bit ALU operations */
733 [BPF_ALU | BPF_ADD | BPF_K] = true,
734 [BPF_ALU | BPF_ADD | BPF_X] = true,
735 [BPF_ALU | BPF_SUB | BPF_K] = true,
736 [BPF_ALU | BPF_SUB | BPF_X] = true,
737 [BPF_ALU | BPF_MUL | BPF_K] = true,
738 [BPF_ALU | BPF_MUL | BPF_X] = true,
739 [BPF_ALU | BPF_DIV | BPF_K] = true,
740 [BPF_ALU | BPF_DIV | BPF_X] = true,
741 [BPF_ALU | BPF_MOD | BPF_K] = true,
742 [BPF_ALU | BPF_MOD | BPF_X] = true,
743 [BPF_ALU | BPF_AND | BPF_K] = true,
744 [BPF_ALU | BPF_AND | BPF_X] = true,
745 [BPF_ALU | BPF_OR | BPF_K] = true,
746 [BPF_ALU | BPF_OR | BPF_X] = true,
747 [BPF_ALU | BPF_XOR | BPF_K] = true,
748 [BPF_ALU | BPF_XOR | BPF_X] = true,
749 [BPF_ALU | BPF_LSH | BPF_K] = true,
750 [BPF_ALU | BPF_LSH | BPF_X] = true,
751 [BPF_ALU | BPF_RSH | BPF_K] = true,
752 [BPF_ALU | BPF_RSH | BPF_X] = true,
753 [BPF_ALU | BPF_NEG] = true,
754 /* Load instructions */
755 [BPF_LD | BPF_W | BPF_ABS] = true,
756 [BPF_LD | BPF_H | BPF_ABS] = true,
757 [BPF_LD | BPF_B | BPF_ABS] = true,
758 [BPF_LD | BPF_W | BPF_LEN] = true,
759 [BPF_LD | BPF_W | BPF_IND] = true,
760 [BPF_LD | BPF_H | BPF_IND] = true,
761 [BPF_LD | BPF_B | BPF_IND] = true,
762 [BPF_LD | BPF_IMM] = true,
763 [BPF_LD | BPF_MEM] = true,
764 [BPF_LDX | BPF_W | BPF_LEN] = true,
765 [BPF_LDX | BPF_B | BPF_MSH] = true,
766 [BPF_LDX | BPF_IMM] = true,
767 [BPF_LDX | BPF_MEM] = true,
768 /* Store instructions */
771 /* Misc instructions */
772 [BPF_MISC | BPF_TAX] = true,
773 [BPF_MISC | BPF_TXA] = true,
774 /* Return instructions */
775 [BPF_RET | BPF_K] = true,
776 [BPF_RET | BPF_A] = true,
777 /* Jump instructions */
778 [BPF_JMP | BPF_JA] = true,
779 [BPF_JMP | BPF_JEQ | BPF_K] = true,
780 [BPF_JMP | BPF_JEQ | BPF_X] = true,
781 [BPF_JMP | BPF_JGE | BPF_K] = true,
782 [BPF_JMP | BPF_JGE | BPF_X] = true,
783 [BPF_JMP | BPF_JGT | BPF_K] = true,
784 [BPF_JMP | BPF_JGT | BPF_X] = true,
785 [BPF_JMP | BPF_JSET | BPF_K] = true,
786 [BPF_JMP | BPF_JSET | BPF_X] = true,
789 if (code_to_probe >= ARRAY_SIZE(codes))
792 return codes[code_to_probe];
795 static bool bpf_check_basics_ok(const struct sock_filter *filter,
800 if (flen == 0 || flen > BPF_MAXINSNS)
807 * bpf_check_classic - verify socket filter code
808 * @filter: filter to verify
809 * @flen: length of filter
811 * Check the user's filter code. If we let some ugly
812 * filter code slip through kaboom! The filter must contain
813 * no references or jumps that are out of range, no illegal
814 * instructions, and must end with a RET instruction.
816 * All jumps are forward as they are not signed.
818 * Returns 0 if the rule set is legal or -EINVAL if not.
820 static int bpf_check_classic(const struct sock_filter *filter,
826 /* Check the filter code now */
827 for (pc = 0; pc < flen; pc++) {
828 const struct sock_filter *ftest = &filter[pc];
830 /* May we actually operate on this code? */
831 if (!chk_code_allowed(ftest->code))
834 /* Some instructions need special checks */
835 switch (ftest->code) {
836 case BPF_ALU | BPF_DIV | BPF_K:
837 case BPF_ALU | BPF_MOD | BPF_K:
838 /* Check for division by zero */
842 case BPF_ALU | BPF_LSH | BPF_K:
843 case BPF_ALU | BPF_RSH | BPF_K:
847 case BPF_LD | BPF_MEM:
848 case BPF_LDX | BPF_MEM:
851 /* Check for invalid memory addresses */
852 if (ftest->k >= BPF_MEMWORDS)
855 case BPF_JMP | BPF_JA:
856 /* Note, the large ftest->k might cause loops.
857 * Compare this with conditional jumps below,
858 * where offsets are limited. --ANK (981016)
860 if (ftest->k >= (unsigned int)(flen - pc - 1))
863 case BPF_JMP | BPF_JEQ | BPF_K:
864 case BPF_JMP | BPF_JEQ | BPF_X:
865 case BPF_JMP | BPF_JGE | BPF_K:
866 case BPF_JMP | BPF_JGE | BPF_X:
867 case BPF_JMP | BPF_JGT | BPF_K:
868 case BPF_JMP | BPF_JGT | BPF_X:
869 case BPF_JMP | BPF_JSET | BPF_K:
870 case BPF_JMP | BPF_JSET | BPF_X:
871 /* Both conditionals must be safe */
872 if (pc + ftest->jt + 1 >= flen ||
873 pc + ftest->jf + 1 >= flen)
876 case BPF_LD | BPF_W | BPF_ABS:
877 case BPF_LD | BPF_H | BPF_ABS:
878 case BPF_LD | BPF_B | BPF_ABS:
880 if (bpf_anc_helper(ftest) & BPF_ANC)
882 /* Ancillary operation unknown or unsupported */
883 if (anc_found == false && ftest->k >= SKF_AD_OFF)
888 /* Last instruction must be a RET code */
889 switch (filter[flen - 1].code) {
890 case BPF_RET | BPF_K:
891 case BPF_RET | BPF_A:
892 return check_load_and_stores(filter, flen);
898 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
899 const struct sock_fprog *fprog)
901 unsigned int fsize = bpf_classic_proglen(fprog);
902 struct sock_fprog_kern *fkprog;
904 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
908 fkprog = fp->orig_prog;
909 fkprog->len = fprog->len;
911 fkprog->filter = kmemdup(fp->insns, fsize,
912 GFP_KERNEL | __GFP_NOWARN);
913 if (!fkprog->filter) {
914 kfree(fp->orig_prog);
921 static void bpf_release_orig_filter(struct bpf_prog *fp)
923 struct sock_fprog_kern *fprog = fp->orig_prog;
926 kfree(fprog->filter);
931 static void __bpf_prog_release(struct bpf_prog *prog)
933 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
936 bpf_release_orig_filter(prog);
941 static void __sk_filter_release(struct sk_filter *fp)
943 __bpf_prog_release(fp->prog);
948 * sk_filter_release_rcu - Release a socket filter by rcu_head
949 * @rcu: rcu_head that contains the sk_filter to free
951 static void sk_filter_release_rcu(struct rcu_head *rcu)
953 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
955 __sk_filter_release(fp);
959 * sk_filter_release - release a socket filter
960 * @fp: filter to remove
962 * Remove a filter from a socket and release its resources.
964 static void sk_filter_release(struct sk_filter *fp)
966 if (refcount_dec_and_test(&fp->refcnt))
967 call_rcu(&fp->rcu, sk_filter_release_rcu);
970 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
972 u32 filter_size = bpf_prog_size(fp->prog->len);
974 atomic_sub(filter_size, &sk->sk_omem_alloc);
975 sk_filter_release(fp);
978 /* try to charge the socket memory if there is space available
979 * return true on success
981 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
983 u32 filter_size = bpf_prog_size(fp->prog->len);
985 /* same check as in sock_kmalloc() */
986 if (filter_size <= sysctl_optmem_max &&
987 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
988 atomic_add(filter_size, &sk->sk_omem_alloc);
994 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
996 if (!refcount_inc_not_zero(&fp->refcnt))
999 if (!__sk_filter_charge(sk, fp)) {
1000 sk_filter_release(fp);
1006 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1008 struct sock_filter *old_prog;
1009 struct bpf_prog *old_fp;
1010 int err, new_len, old_len = fp->len;
1012 /* We are free to overwrite insns et al right here as it
1013 * won't be used at this point in time anymore internally
1014 * after the migration to the internal BPF instruction
1017 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1018 sizeof(struct bpf_insn));
1020 /* Conversion cannot happen on overlapping memory areas,
1021 * so we need to keep the user BPF around until the 2nd
1022 * pass. At this time, the user BPF is stored in fp->insns.
1024 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1025 GFP_KERNEL | __GFP_NOWARN);
1031 /* 1st pass: calculate the new program length. */
1032 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
1036 /* Expand fp for appending the new filter representation. */
1038 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1040 /* The old_fp is still around in case we couldn't
1041 * allocate new memory, so uncharge on that one.
1050 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1051 err = bpf_convert_filter(old_prog, old_len, fp, &new_len);
1053 /* 2nd bpf_convert_filter() can fail only if it fails
1054 * to allocate memory, remapping must succeed. Note,
1055 * that at this time old_fp has already been released
1060 fp = bpf_prog_select_runtime(fp, &err);
1070 __bpf_prog_release(fp);
1071 return ERR_PTR(err);
1074 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1075 bpf_aux_classic_check_t trans)
1079 fp->bpf_func = NULL;
1082 err = bpf_check_classic(fp->insns, fp->len);
1084 __bpf_prog_release(fp);
1085 return ERR_PTR(err);
1088 /* There might be additional checks and transformations
1089 * needed on classic filters, f.e. in case of seccomp.
1092 err = trans(fp->insns, fp->len);
1094 __bpf_prog_release(fp);
1095 return ERR_PTR(err);
1099 /* Probe if we can JIT compile the filter and if so, do
1100 * the compilation of the filter.
1102 bpf_jit_compile(fp);
1104 /* JIT compiler couldn't process this filter, so do the
1105 * internal BPF translation for the optimized interpreter.
1108 fp = bpf_migrate_filter(fp);
1114 * bpf_prog_create - create an unattached filter
1115 * @pfp: the unattached filter that is created
1116 * @fprog: the filter program
1118 * Create a filter independent of any socket. We first run some
1119 * sanity checks on it to make sure it does not explode on us later.
1120 * If an error occurs or there is insufficient memory for the filter
1121 * a negative errno code is returned. On success the return is zero.
1123 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1125 unsigned int fsize = bpf_classic_proglen(fprog);
1126 struct bpf_prog *fp;
1128 /* Make sure new filter is there and in the right amounts. */
1129 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1132 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1136 memcpy(fp->insns, fprog->filter, fsize);
1138 fp->len = fprog->len;
1139 /* Since unattached filters are not copied back to user
1140 * space through sk_get_filter(), we do not need to hold
1141 * a copy here, and can spare us the work.
1143 fp->orig_prog = NULL;
1145 /* bpf_prepare_filter() already takes care of freeing
1146 * memory in case something goes wrong.
1148 fp = bpf_prepare_filter(fp, NULL);
1155 EXPORT_SYMBOL_GPL(bpf_prog_create);
1158 * bpf_prog_create_from_user - create an unattached filter from user buffer
1159 * @pfp: the unattached filter that is created
1160 * @fprog: the filter program
1161 * @trans: post-classic verifier transformation handler
1162 * @save_orig: save classic BPF program
1164 * This function effectively does the same as bpf_prog_create(), only
1165 * that it builds up its insns buffer from user space provided buffer.
1166 * It also allows for passing a bpf_aux_classic_check_t handler.
1168 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1169 bpf_aux_classic_check_t trans, bool save_orig)
1171 unsigned int fsize = bpf_classic_proglen(fprog);
1172 struct bpf_prog *fp;
1175 /* Make sure new filter is there and in the right amounts. */
1176 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1179 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1183 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1184 __bpf_prog_free(fp);
1188 fp->len = fprog->len;
1189 fp->orig_prog = NULL;
1192 err = bpf_prog_store_orig_filter(fp, fprog);
1194 __bpf_prog_free(fp);
1199 /* bpf_prepare_filter() already takes care of freeing
1200 * memory in case something goes wrong.
1202 fp = bpf_prepare_filter(fp, trans);
1209 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1211 void bpf_prog_destroy(struct bpf_prog *fp)
1213 __bpf_prog_release(fp);
1215 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1217 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1219 struct sk_filter *fp, *old_fp;
1221 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1227 if (!__sk_filter_charge(sk, fp)) {
1231 refcount_set(&fp->refcnt, 1);
1233 old_fp = rcu_dereference_protected(sk->sk_filter,
1234 lockdep_sock_is_held(sk));
1235 rcu_assign_pointer(sk->sk_filter, fp);
1238 sk_filter_uncharge(sk, old_fp);
1243 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1245 struct bpf_prog *old_prog;
1248 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1251 if (sk_unhashed(sk) && sk->sk_reuseport) {
1252 err = reuseport_alloc(sk);
1255 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1256 /* The socket wasn't bound with SO_REUSEPORT */
1260 old_prog = reuseport_attach_prog(sk, prog);
1262 bpf_prog_destroy(old_prog);
1268 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1270 unsigned int fsize = bpf_classic_proglen(fprog);
1271 struct bpf_prog *prog;
1274 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1275 return ERR_PTR(-EPERM);
1277 /* Make sure new filter is there and in the right amounts. */
1278 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1279 return ERR_PTR(-EINVAL);
1281 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1283 return ERR_PTR(-ENOMEM);
1285 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1286 __bpf_prog_free(prog);
1287 return ERR_PTR(-EFAULT);
1290 prog->len = fprog->len;
1292 err = bpf_prog_store_orig_filter(prog, fprog);
1294 __bpf_prog_free(prog);
1295 return ERR_PTR(-ENOMEM);
1298 /* bpf_prepare_filter() already takes care of freeing
1299 * memory in case something goes wrong.
1301 return bpf_prepare_filter(prog, NULL);
1305 * sk_attach_filter - attach a socket filter
1306 * @fprog: the filter program
1307 * @sk: the socket to use
1309 * Attach the user's filter code. We first run some sanity checks on
1310 * it to make sure it does not explode on us later. If an error
1311 * occurs or there is insufficient memory for the filter a negative
1312 * errno code is returned. On success the return is zero.
1314 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1316 struct bpf_prog *prog = __get_filter(fprog, sk);
1320 return PTR_ERR(prog);
1322 err = __sk_attach_prog(prog, sk);
1324 __bpf_prog_release(prog);
1330 EXPORT_SYMBOL_GPL(sk_attach_filter);
1332 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1334 struct bpf_prog *prog = __get_filter(fprog, sk);
1338 return PTR_ERR(prog);
1340 err = __reuseport_attach_prog(prog, sk);
1342 __bpf_prog_release(prog);
1349 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1351 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1352 return ERR_PTR(-EPERM);
1354 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1357 int sk_attach_bpf(u32 ufd, struct sock *sk)
1359 struct bpf_prog *prog = __get_bpf(ufd, sk);
1363 return PTR_ERR(prog);
1365 err = __sk_attach_prog(prog, sk);
1374 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1376 struct bpf_prog *prog = __get_bpf(ufd, sk);
1380 return PTR_ERR(prog);
1382 err = __reuseport_attach_prog(prog, sk);
1391 struct bpf_scratchpad {
1393 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1394 u8 buff[MAX_BPF_STACK];
1398 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1400 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1401 unsigned int write_len)
1403 return skb_ensure_writable(skb, write_len);
1406 static inline int bpf_try_make_writable(struct sk_buff *skb,
1407 unsigned int write_len)
1409 int err = __bpf_try_make_writable(skb, write_len);
1411 bpf_compute_data_end(skb);
1415 static int bpf_try_make_head_writable(struct sk_buff *skb)
1417 return bpf_try_make_writable(skb, skb_headlen(skb));
1420 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1422 if (skb_at_tc_ingress(skb))
1423 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1426 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1428 if (skb_at_tc_ingress(skb))
1429 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1432 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1433 const void *, from, u32, len, u64, flags)
1437 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1439 if (unlikely(offset > 0xffff))
1441 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1444 ptr = skb->data + offset;
1445 if (flags & BPF_F_RECOMPUTE_CSUM)
1446 __skb_postpull_rcsum(skb, ptr, len, offset);
1448 memcpy(ptr, from, len);
1450 if (flags & BPF_F_RECOMPUTE_CSUM)
1451 __skb_postpush_rcsum(skb, ptr, len, offset);
1452 if (flags & BPF_F_INVALIDATE_HASH)
1453 skb_clear_hash(skb);
1458 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1459 .func = bpf_skb_store_bytes,
1461 .ret_type = RET_INTEGER,
1462 .arg1_type = ARG_PTR_TO_CTX,
1463 .arg2_type = ARG_ANYTHING,
1464 .arg3_type = ARG_PTR_TO_MEM,
1465 .arg4_type = ARG_CONST_SIZE,
1466 .arg5_type = ARG_ANYTHING,
1469 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1470 void *, to, u32, len)
1474 if (unlikely(offset > 0xffff))
1477 ptr = skb_header_pointer(skb, offset, len, to);
1481 memcpy(to, ptr, len);
1489 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1490 .func = bpf_skb_load_bytes,
1492 .ret_type = RET_INTEGER,
1493 .arg1_type = ARG_PTR_TO_CTX,
1494 .arg2_type = ARG_ANYTHING,
1495 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1496 .arg4_type = ARG_CONST_SIZE,
1499 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1501 /* Idea is the following: should the needed direct read/write
1502 * test fail during runtime, we can pull in more data and redo
1503 * again, since implicitly, we invalidate previous checks here.
1505 * Or, since we know how much we need to make read/writeable,
1506 * this can be done once at the program beginning for direct
1507 * access case. By this we overcome limitations of only current
1508 * headroom being accessible.
1510 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1513 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1514 .func = bpf_skb_pull_data,
1516 .ret_type = RET_INTEGER,
1517 .arg1_type = ARG_PTR_TO_CTX,
1518 .arg2_type = ARG_ANYTHING,
1521 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1522 u64, from, u64, to, u64, flags)
1526 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1528 if (unlikely(offset > 0xffff || offset & 1))
1530 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1533 ptr = (__sum16 *)(skb->data + offset);
1534 switch (flags & BPF_F_HDR_FIELD_MASK) {
1536 if (unlikely(from != 0))
1539 csum_replace_by_diff(ptr, to);
1542 csum_replace2(ptr, from, to);
1545 csum_replace4(ptr, from, to);
1554 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1555 .func = bpf_l3_csum_replace,
1557 .ret_type = RET_INTEGER,
1558 .arg1_type = ARG_PTR_TO_CTX,
1559 .arg2_type = ARG_ANYTHING,
1560 .arg3_type = ARG_ANYTHING,
1561 .arg4_type = ARG_ANYTHING,
1562 .arg5_type = ARG_ANYTHING,
1565 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1566 u64, from, u64, to, u64, flags)
1568 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1569 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1570 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1573 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1574 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1576 if (unlikely(offset > 0xffff || offset & 1))
1578 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1581 ptr = (__sum16 *)(skb->data + offset);
1582 if (is_mmzero && !do_mforce && !*ptr)
1585 switch (flags & BPF_F_HDR_FIELD_MASK) {
1587 if (unlikely(from != 0))
1590 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1593 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1596 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1602 if (is_mmzero && !*ptr)
1603 *ptr = CSUM_MANGLED_0;
1607 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1608 .func = bpf_l4_csum_replace,
1610 .ret_type = RET_INTEGER,
1611 .arg1_type = ARG_PTR_TO_CTX,
1612 .arg2_type = ARG_ANYTHING,
1613 .arg3_type = ARG_ANYTHING,
1614 .arg4_type = ARG_ANYTHING,
1615 .arg5_type = ARG_ANYTHING,
1618 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1619 __be32 *, to, u32, to_size, __wsum, seed)
1621 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1622 u32 diff_size = from_size + to_size;
1625 /* This is quite flexible, some examples:
1627 * from_size == 0, to_size > 0, seed := csum --> pushing data
1628 * from_size > 0, to_size == 0, seed := csum --> pulling data
1629 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1631 * Even for diffing, from_size and to_size don't need to be equal.
1633 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1634 diff_size > sizeof(sp->diff)))
1637 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1638 sp->diff[j] = ~from[i];
1639 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1640 sp->diff[j] = to[i];
1642 return csum_partial(sp->diff, diff_size, seed);
1645 static const struct bpf_func_proto bpf_csum_diff_proto = {
1646 .func = bpf_csum_diff,
1649 .ret_type = RET_INTEGER,
1650 .arg1_type = ARG_PTR_TO_MEM,
1651 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1652 .arg3_type = ARG_PTR_TO_MEM,
1653 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1654 .arg5_type = ARG_ANYTHING,
1657 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1659 /* The interface is to be used in combination with bpf_csum_diff()
1660 * for direct packet writes. csum rotation for alignment as well
1661 * as emulating csum_sub() can be done from the eBPF program.
1663 if (skb->ip_summed == CHECKSUM_COMPLETE)
1664 return (skb->csum = csum_add(skb->csum, csum));
1669 static const struct bpf_func_proto bpf_csum_update_proto = {
1670 .func = bpf_csum_update,
1672 .ret_type = RET_INTEGER,
1673 .arg1_type = ARG_PTR_TO_CTX,
1674 .arg2_type = ARG_ANYTHING,
1677 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1679 return dev_forward_skb(dev, skb);
1682 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1683 struct sk_buff *skb)
1685 int ret = ____dev_forward_skb(dev, skb);
1689 ret = netif_rx(skb);
1695 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1699 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1700 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1707 __this_cpu_inc(xmit_recursion);
1708 ret = dev_queue_xmit(skb);
1709 __this_cpu_dec(xmit_recursion);
1714 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
1717 unsigned int mlen = skb_network_offset(skb);
1720 __skb_pull(skb, mlen);
1722 /* At ingress, the mac header has already been pulled once.
1723 * At egress, skb_pospull_rcsum has to be done in case that
1724 * the skb is originated from ingress (i.e. a forwarded skb)
1725 * to ensure that rcsum starts at net header.
1727 if (!skb_at_tc_ingress(skb))
1728 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
1730 skb_pop_mac_header(skb);
1731 skb_reset_mac_len(skb);
1732 return flags & BPF_F_INGRESS ?
1733 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
1736 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
1739 /* Verify that a link layer header is carried */
1740 if (unlikely(skb->mac_header >= skb->network_header)) {
1745 bpf_push_mac_rcsum(skb);
1746 return flags & BPF_F_INGRESS ?
1747 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1750 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
1753 if (dev_is_mac_header_xmit(dev))
1754 return __bpf_redirect_common(skb, dev, flags);
1756 return __bpf_redirect_no_mac(skb, dev, flags);
1759 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
1761 struct net_device *dev;
1762 struct sk_buff *clone;
1765 if (unlikely(flags & ~(BPF_F_INGRESS)))
1768 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1772 clone = skb_clone(skb, GFP_ATOMIC);
1773 if (unlikely(!clone))
1776 /* For direct write, we need to keep the invariant that the skbs
1777 * we're dealing with need to be uncloned. Should uncloning fail
1778 * here, we need to free the just generated clone to unclone once
1781 ret = bpf_try_make_head_writable(skb);
1782 if (unlikely(ret)) {
1787 return __bpf_redirect(clone, dev, flags);
1790 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1791 .func = bpf_clone_redirect,
1793 .ret_type = RET_INTEGER,
1794 .arg1_type = ARG_PTR_TO_CTX,
1795 .arg2_type = ARG_ANYTHING,
1796 .arg3_type = ARG_ANYTHING,
1799 struct redirect_info {
1802 struct bpf_map *map;
1803 struct bpf_map *map_to_flush;
1804 unsigned long map_owner;
1807 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1809 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
1811 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1813 if (unlikely(flags & ~(BPF_F_INGRESS)))
1816 ri->ifindex = ifindex;
1819 return TC_ACT_REDIRECT;
1822 int skb_do_redirect(struct sk_buff *skb)
1824 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1825 struct net_device *dev;
1827 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1829 if (unlikely(!dev)) {
1834 return __bpf_redirect(skb, dev, ri->flags);
1837 static const struct bpf_func_proto bpf_redirect_proto = {
1838 .func = bpf_redirect,
1840 .ret_type = RET_INTEGER,
1841 .arg1_type = ARG_ANYTHING,
1842 .arg2_type = ARG_ANYTHING,
1845 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
1846 struct bpf_map *, map, u32, key, u64, flags)
1848 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1850 /* If user passes invalid input drop the packet. */
1851 if (unlikely(flags))
1855 tcb->bpf.flags = flags;
1861 struct sock *do_sk_redirect_map(struct sk_buff *skb)
1863 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1864 struct sock *sk = NULL;
1867 sk = __sock_map_lookup_elem(tcb->bpf.map, tcb->bpf.key);
1870 tcb->bpf.map = NULL;
1876 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
1877 .func = bpf_sk_redirect_map,
1879 .ret_type = RET_INTEGER,
1880 .arg1_type = ARG_PTR_TO_CTX,
1881 .arg2_type = ARG_CONST_MAP_PTR,
1882 .arg3_type = ARG_ANYTHING,
1883 .arg4_type = ARG_ANYTHING,
1886 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
1888 return task_get_classid(skb);
1891 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1892 .func = bpf_get_cgroup_classid,
1894 .ret_type = RET_INTEGER,
1895 .arg1_type = ARG_PTR_TO_CTX,
1898 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
1900 return dst_tclassid(skb);
1903 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1904 .func = bpf_get_route_realm,
1906 .ret_type = RET_INTEGER,
1907 .arg1_type = ARG_PTR_TO_CTX,
1910 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
1912 /* If skb_clear_hash() was called due to mangling, we can
1913 * trigger SW recalculation here. Later access to hash
1914 * can then use the inline skb->hash via context directly
1915 * instead of calling this helper again.
1917 return skb_get_hash(skb);
1920 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1921 .func = bpf_get_hash_recalc,
1923 .ret_type = RET_INTEGER,
1924 .arg1_type = ARG_PTR_TO_CTX,
1927 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
1929 /* After all direct packet write, this can be used once for
1930 * triggering a lazy recalc on next skb_get_hash() invocation.
1932 skb_clear_hash(skb);
1936 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
1937 .func = bpf_set_hash_invalid,
1939 .ret_type = RET_INTEGER,
1940 .arg1_type = ARG_PTR_TO_CTX,
1943 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
1945 /* Set user specified hash as L4(+), so that it gets returned
1946 * on skb_get_hash() call unless BPF prog later on triggers a
1949 __skb_set_sw_hash(skb, hash, true);
1953 static const struct bpf_func_proto bpf_set_hash_proto = {
1954 .func = bpf_set_hash,
1956 .ret_type = RET_INTEGER,
1957 .arg1_type = ARG_PTR_TO_CTX,
1958 .arg2_type = ARG_ANYTHING,
1961 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
1966 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1967 vlan_proto != htons(ETH_P_8021AD)))
1968 vlan_proto = htons(ETH_P_8021Q);
1970 bpf_push_mac_rcsum(skb);
1971 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1972 bpf_pull_mac_rcsum(skb);
1974 bpf_compute_data_end(skb);
1978 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1979 .func = bpf_skb_vlan_push,
1981 .ret_type = RET_INTEGER,
1982 .arg1_type = ARG_PTR_TO_CTX,
1983 .arg2_type = ARG_ANYTHING,
1984 .arg3_type = ARG_ANYTHING,
1986 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1988 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
1992 bpf_push_mac_rcsum(skb);
1993 ret = skb_vlan_pop(skb);
1994 bpf_pull_mac_rcsum(skb);
1996 bpf_compute_data_end(skb);
2000 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2001 .func = bpf_skb_vlan_pop,
2003 .ret_type = RET_INTEGER,
2004 .arg1_type = ARG_PTR_TO_CTX,
2006 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
2008 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2010 /* Caller already did skb_cow() with len as headroom,
2011 * so no need to do it here.
2014 memmove(skb->data, skb->data + len, off);
2015 memset(skb->data + off, 0, len);
2017 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2018 * needed here as it does not change the skb->csum
2019 * result for checksum complete when summing over
2025 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2027 /* skb_ensure_writable() is not needed here, as we're
2028 * already working on an uncloned skb.
2030 if (unlikely(!pskb_may_pull(skb, off + len)))
2033 skb_postpull_rcsum(skb, skb->data + off, len);
2034 memmove(skb->data + len, skb->data, off);
2035 __skb_pull(skb, len);
2040 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2042 bool trans_same = skb->transport_header == skb->network_header;
2045 /* There's no need for __skb_push()/__skb_pull() pair to
2046 * get to the start of the mac header as we're guaranteed
2047 * to always start from here under eBPF.
2049 ret = bpf_skb_generic_push(skb, off, len);
2051 skb->mac_header -= len;
2052 skb->network_header -= len;
2054 skb->transport_header = skb->network_header;
2060 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2062 bool trans_same = skb->transport_header == skb->network_header;
2065 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2066 ret = bpf_skb_generic_pop(skb, off, len);
2068 skb->mac_header += len;
2069 skb->network_header += len;
2071 skb->transport_header = skb->network_header;
2077 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2079 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2080 u32 off = skb_mac_header_len(skb);
2083 ret = skb_cow(skb, len_diff);
2084 if (unlikely(ret < 0))
2087 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2088 if (unlikely(ret < 0))
2091 if (skb_is_gso(skb)) {
2092 /* SKB_GSO_TCPV4 needs to be changed into
2095 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2096 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
2097 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
2100 /* Due to IPv6 header, MSS needs to be downgraded. */
2101 skb_shinfo(skb)->gso_size -= len_diff;
2102 /* Header must be checked, and gso_segs recomputed. */
2103 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2104 skb_shinfo(skb)->gso_segs = 0;
2107 skb->protocol = htons(ETH_P_IPV6);
2108 skb_clear_hash(skb);
2113 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2115 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2116 u32 off = skb_mac_header_len(skb);
2119 ret = skb_unclone(skb, GFP_ATOMIC);
2120 if (unlikely(ret < 0))
2123 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2124 if (unlikely(ret < 0))
2127 if (skb_is_gso(skb)) {
2128 /* SKB_GSO_TCPV6 needs to be changed into
2131 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
2132 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
2133 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
2136 /* Due to IPv4 header, MSS can be upgraded. */
2137 skb_shinfo(skb)->gso_size += len_diff;
2138 /* Header must be checked, and gso_segs recomputed. */
2139 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2140 skb_shinfo(skb)->gso_segs = 0;
2143 skb->protocol = htons(ETH_P_IP);
2144 skb_clear_hash(skb);
2149 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2151 __be16 from_proto = skb->protocol;
2153 if (from_proto == htons(ETH_P_IP) &&
2154 to_proto == htons(ETH_P_IPV6))
2155 return bpf_skb_proto_4_to_6(skb);
2157 if (from_proto == htons(ETH_P_IPV6) &&
2158 to_proto == htons(ETH_P_IP))
2159 return bpf_skb_proto_6_to_4(skb);
2164 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2169 if (unlikely(flags))
2172 /* General idea is that this helper does the basic groundwork
2173 * needed for changing the protocol, and eBPF program fills the
2174 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2175 * and other helpers, rather than passing a raw buffer here.
2177 * The rationale is to keep this minimal and without a need to
2178 * deal with raw packet data. F.e. even if we would pass buffers
2179 * here, the program still needs to call the bpf_lX_csum_replace()
2180 * helpers anyway. Plus, this way we keep also separation of
2181 * concerns, since f.e. bpf_skb_store_bytes() should only take
2184 * Currently, additional options and extension header space are
2185 * not supported, but flags register is reserved so we can adapt
2186 * that. For offloads, we mark packet as dodgy, so that headers
2187 * need to be verified first.
2189 ret = bpf_skb_proto_xlat(skb, proto);
2190 bpf_compute_data_end(skb);
2194 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2195 .func = bpf_skb_change_proto,
2197 .ret_type = RET_INTEGER,
2198 .arg1_type = ARG_PTR_TO_CTX,
2199 .arg2_type = ARG_ANYTHING,
2200 .arg3_type = ARG_ANYTHING,
2203 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2205 /* We only allow a restricted subset to be changed for now. */
2206 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2207 !skb_pkt_type_ok(pkt_type)))
2210 skb->pkt_type = pkt_type;
2214 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2215 .func = bpf_skb_change_type,
2217 .ret_type = RET_INTEGER,
2218 .arg1_type = ARG_PTR_TO_CTX,
2219 .arg2_type = ARG_ANYTHING,
2222 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2224 switch (skb->protocol) {
2225 case htons(ETH_P_IP):
2226 return sizeof(struct iphdr);
2227 case htons(ETH_P_IPV6):
2228 return sizeof(struct ipv6hdr);
2234 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2236 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2239 ret = skb_cow(skb, len_diff);
2240 if (unlikely(ret < 0))
2243 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2244 if (unlikely(ret < 0))
2247 if (skb_is_gso(skb)) {
2248 /* Due to header grow, MSS needs to be downgraded. */
2249 skb_shinfo(skb)->gso_size -= len_diff;
2250 /* Header must be checked, and gso_segs recomputed. */
2251 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2252 skb_shinfo(skb)->gso_segs = 0;
2258 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2260 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2263 ret = skb_unclone(skb, GFP_ATOMIC);
2264 if (unlikely(ret < 0))
2267 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2268 if (unlikely(ret < 0))
2271 if (skb_is_gso(skb)) {
2272 /* Due to header shrink, MSS can be upgraded. */
2273 skb_shinfo(skb)->gso_size += len_diff;
2274 /* Header must be checked, and gso_segs recomputed. */
2275 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2276 skb_shinfo(skb)->gso_segs = 0;
2282 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
2284 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2286 bool trans_same = skb->transport_header == skb->network_header;
2287 u32 len_cur, len_diff_abs = abs(len_diff);
2288 u32 len_min = bpf_skb_net_base_len(skb);
2289 u32 len_max = BPF_SKB_MAX_LEN;
2290 __be16 proto = skb->protocol;
2291 bool shrink = len_diff < 0;
2294 if (unlikely(len_diff_abs > 0xfffU))
2296 if (unlikely(proto != htons(ETH_P_IP) &&
2297 proto != htons(ETH_P_IPV6)))
2300 len_cur = skb->len - skb_network_offset(skb);
2301 if (skb_transport_header_was_set(skb) && !trans_same)
2302 len_cur = skb_network_header_len(skb);
2303 if ((shrink && (len_diff_abs >= len_cur ||
2304 len_cur - len_diff_abs < len_min)) ||
2305 (!shrink && (skb->len + len_diff_abs > len_max &&
2309 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2310 bpf_skb_net_grow(skb, len_diff_abs);
2312 bpf_compute_data_end(skb);
2316 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2317 u32, mode, u64, flags)
2319 if (unlikely(flags))
2321 if (likely(mode == BPF_ADJ_ROOM_NET))
2322 return bpf_skb_adjust_net(skb, len_diff);
2327 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2328 .func = bpf_skb_adjust_room,
2330 .ret_type = RET_INTEGER,
2331 .arg1_type = ARG_PTR_TO_CTX,
2332 .arg2_type = ARG_ANYTHING,
2333 .arg3_type = ARG_ANYTHING,
2334 .arg4_type = ARG_ANYTHING,
2337 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2339 u32 min_len = skb_network_offset(skb);
2341 if (skb_transport_header_was_set(skb))
2342 min_len = skb_transport_offset(skb);
2343 if (skb->ip_summed == CHECKSUM_PARTIAL)
2344 min_len = skb_checksum_start_offset(skb) +
2345 skb->csum_offset + sizeof(__sum16);
2349 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2351 unsigned int old_len = skb->len;
2354 ret = __skb_grow_rcsum(skb, new_len);
2356 memset(skb->data + old_len, 0, new_len - old_len);
2360 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2362 return __skb_trim_rcsum(skb, new_len);
2365 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2368 u32 max_len = BPF_SKB_MAX_LEN;
2369 u32 min_len = __bpf_skb_min_len(skb);
2372 if (unlikely(flags || new_len > max_len || new_len < min_len))
2374 if (skb->encapsulation)
2377 /* The basic idea of this helper is that it's performing the
2378 * needed work to either grow or trim an skb, and eBPF program
2379 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2380 * bpf_lX_csum_replace() and others rather than passing a raw
2381 * buffer here. This one is a slow path helper and intended
2382 * for replies with control messages.
2384 * Like in bpf_skb_change_proto(), we want to keep this rather
2385 * minimal and without protocol specifics so that we are able
2386 * to separate concerns as in bpf_skb_store_bytes() should only
2387 * be the one responsible for writing buffers.
2389 * It's really expected to be a slow path operation here for
2390 * control message replies, so we're implicitly linearizing,
2391 * uncloning and drop offloads from the skb by this.
2393 ret = __bpf_try_make_writable(skb, skb->len);
2395 if (new_len > skb->len)
2396 ret = bpf_skb_grow_rcsum(skb, new_len);
2397 else if (new_len < skb->len)
2398 ret = bpf_skb_trim_rcsum(skb, new_len);
2399 if (!ret && skb_is_gso(skb))
2403 bpf_compute_data_end(skb);
2407 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2408 .func = bpf_skb_change_tail,
2410 .ret_type = RET_INTEGER,
2411 .arg1_type = ARG_PTR_TO_CTX,
2412 .arg2_type = ARG_ANYTHING,
2413 .arg3_type = ARG_ANYTHING,
2416 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2419 u32 max_len = BPF_SKB_MAX_LEN;
2420 u32 new_len = skb->len + head_room;
2423 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2424 new_len < skb->len))
2427 ret = skb_cow(skb, head_room);
2429 /* Idea for this helper is that we currently only
2430 * allow to expand on mac header. This means that
2431 * skb->protocol network header, etc, stay as is.
2432 * Compared to bpf_skb_change_tail(), we're more
2433 * flexible due to not needing to linearize or
2434 * reset GSO. Intention for this helper is to be
2435 * used by an L3 skb that needs to push mac header
2436 * for redirection into L2 device.
2438 __skb_push(skb, head_room);
2439 memset(skb->data, 0, head_room);
2440 skb_reset_mac_header(skb);
2441 skb_reset_mac_len(skb);
2444 bpf_compute_data_end(skb);
2448 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2449 .func = bpf_skb_change_head,
2451 .ret_type = RET_INTEGER,
2452 .arg1_type = ARG_PTR_TO_CTX,
2453 .arg2_type = ARG_ANYTHING,
2454 .arg3_type = ARG_ANYTHING,
2457 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2459 void *data = xdp->data + offset;
2461 if (unlikely(data < xdp->data_hard_start ||
2462 data > xdp->data_end - ETH_HLEN))
2470 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2471 .func = bpf_xdp_adjust_head,
2473 .ret_type = RET_INTEGER,
2474 .arg1_type = ARG_PTR_TO_CTX,
2475 .arg2_type = ARG_ANYTHING,
2478 static int __bpf_tx_xdp(struct net_device *dev,
2479 struct bpf_map *map,
2480 struct xdp_buff *xdp,
2485 if (!dev->netdev_ops->ndo_xdp_xmit) {
2489 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2493 __dev_map_insert_ctx(map, index);
2495 dev->netdev_ops->ndo_xdp_flush(dev);
2499 void xdp_do_flush_map(void)
2501 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2502 struct bpf_map *map = ri->map_to_flush;
2504 ri->map_to_flush = NULL;
2506 __dev_map_flush(map);
2508 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
2510 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
2513 return (unsigned long)xdp_prog->aux != aux;
2516 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
2517 struct bpf_prog *xdp_prog)
2519 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2520 unsigned long map_owner = ri->map_owner;
2521 struct bpf_map *map = ri->map;
2522 struct net_device *fwd = NULL;
2523 u32 index = ri->ifindex;
2530 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2536 fwd = __dev_map_lookup_elem(map, index);
2541 if (ri->map_to_flush && ri->map_to_flush != map)
2544 err = __bpf_tx_xdp(fwd, map, xdp, index);
2548 ri->map_to_flush = map;
2549 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2552 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2556 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
2557 struct bpf_prog *xdp_prog)
2559 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2560 struct net_device *fwd;
2561 u32 index = ri->ifindex;
2565 return xdp_do_redirect_map(dev, xdp, xdp_prog);
2567 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2569 if (unlikely(!fwd)) {
2574 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
2578 _trace_xdp_redirect(dev, xdp_prog, index);
2581 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2584 EXPORT_SYMBOL_GPL(xdp_do_redirect);
2586 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
2587 struct bpf_prog *xdp_prog)
2589 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2590 unsigned long map_owner = ri->map_owner;
2591 struct bpf_map *map = ri->map;
2592 struct net_device *fwd = NULL;
2593 u32 index = ri->ifindex;
2602 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2607 fwd = __dev_map_lookup_elem(map, index);
2609 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2611 if (unlikely(!fwd)) {
2616 if (unlikely(!(fwd->flags & IFF_UP))) {
2621 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
2622 if (skb->len > len) {
2628 map ? _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index)
2629 : _trace_xdp_redirect(dev, xdp_prog, index);
2632 map ? _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err)
2633 : _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2636 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
2638 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
2640 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2642 if (unlikely(flags))
2645 ri->ifindex = ifindex;
2650 return XDP_REDIRECT;
2653 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
2654 .func = bpf_xdp_redirect,
2656 .ret_type = RET_INTEGER,
2657 .arg1_type = ARG_ANYTHING,
2658 .arg2_type = ARG_ANYTHING,
2661 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
2662 unsigned long, map_owner)
2664 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2666 if (unlikely(flags))
2669 ri->ifindex = ifindex;
2672 ri->map_owner = map_owner;
2674 return XDP_REDIRECT;
2677 /* Note, arg4 is hidden from users and populated by the verifier
2678 * with the right pointer.
2680 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
2681 .func = bpf_xdp_redirect_map,
2683 .ret_type = RET_INTEGER,
2684 .arg1_type = ARG_CONST_MAP_PTR,
2685 .arg2_type = ARG_ANYTHING,
2686 .arg3_type = ARG_ANYTHING,
2689 bool bpf_helper_changes_pkt_data(void *func)
2691 if (func == bpf_skb_vlan_push ||
2692 func == bpf_skb_vlan_pop ||
2693 func == bpf_skb_store_bytes ||
2694 func == bpf_skb_change_proto ||
2695 func == bpf_skb_change_head ||
2696 func == bpf_skb_change_tail ||
2697 func == bpf_skb_adjust_room ||
2698 func == bpf_skb_pull_data ||
2699 func == bpf_clone_redirect ||
2700 func == bpf_l3_csum_replace ||
2701 func == bpf_l4_csum_replace ||
2702 func == bpf_xdp_adjust_head)
2708 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2709 unsigned long off, unsigned long len)
2711 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2715 if (ptr != dst_buff)
2716 memcpy(dst_buff, ptr, len);
2721 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2722 u64, flags, void *, meta, u64, meta_size)
2724 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2726 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2728 if (unlikely(skb_size > skb->len))
2731 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2735 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2736 .func = bpf_skb_event_output,
2738 .ret_type = RET_INTEGER,
2739 .arg1_type = ARG_PTR_TO_CTX,
2740 .arg2_type = ARG_CONST_MAP_PTR,
2741 .arg3_type = ARG_ANYTHING,
2742 .arg4_type = ARG_PTR_TO_MEM,
2743 .arg5_type = ARG_CONST_SIZE,
2746 static unsigned short bpf_tunnel_key_af(u64 flags)
2748 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2751 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2752 u32, size, u64, flags)
2754 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2755 u8 compat[sizeof(struct bpf_tunnel_key)];
2759 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2763 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2767 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2770 case offsetof(struct bpf_tunnel_key, tunnel_label):
2771 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2773 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2774 /* Fixup deprecated structure layouts here, so we have
2775 * a common path later on.
2777 if (ip_tunnel_info_af(info) != AF_INET)
2780 to = (struct bpf_tunnel_key *)compat;
2787 to->tunnel_id = be64_to_cpu(info->key.tun_id);
2788 to->tunnel_tos = info->key.tos;
2789 to->tunnel_ttl = info->key.ttl;
2791 if (flags & BPF_F_TUNINFO_IPV6) {
2792 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2793 sizeof(to->remote_ipv6));
2794 to->tunnel_label = be32_to_cpu(info->key.label);
2796 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2799 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2800 memcpy(to_orig, to, size);
2804 memset(to_orig, 0, size);
2808 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2809 .func = bpf_skb_get_tunnel_key,
2811 .ret_type = RET_INTEGER,
2812 .arg1_type = ARG_PTR_TO_CTX,
2813 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2814 .arg3_type = ARG_CONST_SIZE,
2815 .arg4_type = ARG_ANYTHING,
2818 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2820 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2823 if (unlikely(!info ||
2824 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2828 if (unlikely(size < info->options_len)) {
2833 ip_tunnel_info_opts_get(to, info);
2834 if (size > info->options_len)
2835 memset(to + info->options_len, 0, size - info->options_len);
2837 return info->options_len;
2839 memset(to, 0, size);
2843 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2844 .func = bpf_skb_get_tunnel_opt,
2846 .ret_type = RET_INTEGER,
2847 .arg1_type = ARG_PTR_TO_CTX,
2848 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2849 .arg3_type = ARG_CONST_SIZE,
2852 static struct metadata_dst __percpu *md_dst;
2854 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
2855 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
2857 struct metadata_dst *md = this_cpu_ptr(md_dst);
2858 u8 compat[sizeof(struct bpf_tunnel_key)];
2859 struct ip_tunnel_info *info;
2861 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2862 BPF_F_DONT_FRAGMENT)))
2864 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2866 case offsetof(struct bpf_tunnel_key, tunnel_label):
2867 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2868 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2869 /* Fixup deprecated structure layouts here, so we have
2870 * a common path later on.
2872 memcpy(compat, from, size);
2873 memset(compat + size, 0, sizeof(compat) - size);
2874 from = (const struct bpf_tunnel_key *) compat;
2880 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
2885 dst_hold((struct dst_entry *) md);
2886 skb_dst_set(skb, (struct dst_entry *) md);
2888 info = &md->u.tun_info;
2889 info->mode = IP_TUNNEL_INFO_TX;
2891 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
2892 if (flags & BPF_F_DONT_FRAGMENT)
2893 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
2895 info->key.tun_id = cpu_to_be64(from->tunnel_id);
2896 info->key.tos = from->tunnel_tos;
2897 info->key.ttl = from->tunnel_ttl;
2899 if (flags & BPF_F_TUNINFO_IPV6) {
2900 info->mode |= IP_TUNNEL_INFO_IPV6;
2901 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
2902 sizeof(from->remote_ipv6));
2903 info->key.label = cpu_to_be32(from->tunnel_label) &
2904 IPV6_FLOWLABEL_MASK;
2906 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
2907 if (flags & BPF_F_ZERO_CSUM_TX)
2908 info->key.tun_flags &= ~TUNNEL_CSUM;
2914 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
2915 .func = bpf_skb_set_tunnel_key,
2917 .ret_type = RET_INTEGER,
2918 .arg1_type = ARG_PTR_TO_CTX,
2919 .arg2_type = ARG_PTR_TO_MEM,
2920 .arg3_type = ARG_CONST_SIZE,
2921 .arg4_type = ARG_ANYTHING,
2924 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
2925 const u8 *, from, u32, size)
2927 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2928 const struct metadata_dst *md = this_cpu_ptr(md_dst);
2930 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
2932 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
2935 ip_tunnel_info_opts_set(info, from, size);
2940 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
2941 .func = bpf_skb_set_tunnel_opt,
2943 .ret_type = RET_INTEGER,
2944 .arg1_type = ARG_PTR_TO_CTX,
2945 .arg2_type = ARG_PTR_TO_MEM,
2946 .arg3_type = ARG_CONST_SIZE,
2949 static const struct bpf_func_proto *
2950 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
2953 /* Race is not possible, since it's called from verifier
2954 * that is holding verifier mutex.
2956 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
2964 case BPF_FUNC_skb_set_tunnel_key:
2965 return &bpf_skb_set_tunnel_key_proto;
2966 case BPF_FUNC_skb_set_tunnel_opt:
2967 return &bpf_skb_set_tunnel_opt_proto;
2973 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
2976 struct bpf_array *array = container_of(map, struct bpf_array, map);
2977 struct cgroup *cgrp;
2980 sk = skb_to_full_sk(skb);
2981 if (!sk || !sk_fullsock(sk))
2983 if (unlikely(idx >= array->map.max_entries))
2986 cgrp = READ_ONCE(array->ptrs[idx]);
2987 if (unlikely(!cgrp))
2990 return sk_under_cgroup_hierarchy(sk, cgrp);
2993 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
2994 .func = bpf_skb_under_cgroup,
2996 .ret_type = RET_INTEGER,
2997 .arg1_type = ARG_PTR_TO_CTX,
2998 .arg2_type = ARG_CONST_MAP_PTR,
2999 .arg3_type = ARG_ANYTHING,
3002 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3003 unsigned long off, unsigned long len)
3005 memcpy(dst_buff, src_buff + off, len);
3009 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3010 u64, flags, void *, meta, u64, meta_size)
3012 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3014 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3016 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3019 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3020 xdp_size, bpf_xdp_copy);
3023 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3024 .func = bpf_xdp_event_output,
3026 .ret_type = RET_INTEGER,
3027 .arg1_type = ARG_PTR_TO_CTX,
3028 .arg2_type = ARG_CONST_MAP_PTR,
3029 .arg3_type = ARG_ANYTHING,
3030 .arg4_type = ARG_PTR_TO_MEM,
3031 .arg5_type = ARG_CONST_SIZE,
3034 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3036 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3039 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3040 .func = bpf_get_socket_cookie,
3042 .ret_type = RET_INTEGER,
3043 .arg1_type = ARG_PTR_TO_CTX,
3046 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3048 struct sock *sk = sk_to_full_sk(skb->sk);
3051 if (!sk || !sk_fullsock(sk))
3053 kuid = sock_net_uid(sock_net(sk), sk);
3054 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3057 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3058 .func = bpf_get_socket_uid,
3060 .ret_type = RET_INTEGER,
3061 .arg1_type = ARG_PTR_TO_CTX,
3064 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3065 int, level, int, optname, char *, optval, int, optlen)
3067 struct sock *sk = bpf_sock->sk;
3071 if (!sk_fullsock(sk))
3074 if (level == SOL_SOCKET) {
3075 if (optlen != sizeof(int))
3077 val = *((int *)optval);
3079 /* Only some socketops are supported */
3082 val = min_t(u32, val, sysctl_rmem_max);
3083 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3084 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3087 val = min_t(u32, val, sysctl_wmem_max);
3088 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3089 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3091 case SO_MAX_PACING_RATE:
3092 sk->sk_max_pacing_rate = val;
3093 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3094 sk->sk_max_pacing_rate);
3097 sk->sk_priority = val;
3102 sk->sk_rcvlowat = val ? : 1;
3105 if (sk->sk_mark != val) {
3114 } else if (level == SOL_TCP &&
3115 sk->sk_prot->setsockopt == tcp_setsockopt) {
3116 if (optname == TCP_CONGESTION) {
3117 char name[TCP_CA_NAME_MAX];
3118 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3120 strncpy(name, optval, min_t(long, optlen,
3121 TCP_CA_NAME_MAX-1));
3122 name[TCP_CA_NAME_MAX-1] = 0;
3123 ret = tcp_set_congestion_control(sk, name, false,
3126 struct tcp_sock *tp = tcp_sk(sk);
3128 if (optlen != sizeof(int))
3131 val = *((int *)optval);
3132 /* Only some options are supported */
3135 if (val <= 0 || tp->data_segs_out > tp->syn_data)
3140 case TCP_BPF_SNDCWND_CLAMP:
3144 tp->snd_cwnd_clamp = val;
3145 tp->snd_ssthresh = val;
3159 static const struct bpf_func_proto bpf_setsockopt_proto = {
3160 .func = bpf_setsockopt,
3162 .ret_type = RET_INTEGER,
3163 .arg1_type = ARG_PTR_TO_CTX,
3164 .arg2_type = ARG_ANYTHING,
3165 .arg3_type = ARG_ANYTHING,
3166 .arg4_type = ARG_PTR_TO_MEM,
3167 .arg5_type = ARG_CONST_SIZE,
3170 static const struct bpf_func_proto *
3171 bpf_base_func_proto(enum bpf_func_id func_id)
3174 case BPF_FUNC_map_lookup_elem:
3175 return &bpf_map_lookup_elem_proto;
3176 case BPF_FUNC_map_update_elem:
3177 return &bpf_map_update_elem_proto;
3178 case BPF_FUNC_map_delete_elem:
3179 return &bpf_map_delete_elem_proto;
3180 case BPF_FUNC_get_prandom_u32:
3181 return &bpf_get_prandom_u32_proto;
3182 case BPF_FUNC_get_smp_processor_id:
3183 return &bpf_get_raw_smp_processor_id_proto;
3184 case BPF_FUNC_get_numa_node_id:
3185 return &bpf_get_numa_node_id_proto;
3186 case BPF_FUNC_tail_call:
3187 return &bpf_tail_call_proto;
3188 case BPF_FUNC_ktime_get_ns:
3189 return &bpf_ktime_get_ns_proto;
3190 case BPF_FUNC_trace_printk:
3191 if (capable(CAP_SYS_ADMIN))
3192 return bpf_get_trace_printk_proto();
3198 static const struct bpf_func_proto *
3199 sock_filter_func_proto(enum bpf_func_id func_id)
3202 /* inet and inet6 sockets are created in a process
3203 * context so there is always a valid uid/gid
3205 case BPF_FUNC_get_current_uid_gid:
3206 return &bpf_get_current_uid_gid_proto;
3208 return bpf_base_func_proto(func_id);
3212 static const struct bpf_func_proto *
3213 sk_filter_func_proto(enum bpf_func_id func_id)
3216 case BPF_FUNC_skb_load_bytes:
3217 return &bpf_skb_load_bytes_proto;
3218 case BPF_FUNC_get_socket_cookie:
3219 return &bpf_get_socket_cookie_proto;
3220 case BPF_FUNC_get_socket_uid:
3221 return &bpf_get_socket_uid_proto;
3223 return bpf_base_func_proto(func_id);
3227 static const struct bpf_func_proto *
3228 tc_cls_act_func_proto(enum bpf_func_id func_id)
3231 case BPF_FUNC_skb_store_bytes:
3232 return &bpf_skb_store_bytes_proto;
3233 case BPF_FUNC_skb_load_bytes:
3234 return &bpf_skb_load_bytes_proto;
3235 case BPF_FUNC_skb_pull_data:
3236 return &bpf_skb_pull_data_proto;
3237 case BPF_FUNC_csum_diff:
3238 return &bpf_csum_diff_proto;
3239 case BPF_FUNC_csum_update:
3240 return &bpf_csum_update_proto;
3241 case BPF_FUNC_l3_csum_replace:
3242 return &bpf_l3_csum_replace_proto;
3243 case BPF_FUNC_l4_csum_replace:
3244 return &bpf_l4_csum_replace_proto;
3245 case BPF_FUNC_clone_redirect:
3246 return &bpf_clone_redirect_proto;
3247 case BPF_FUNC_get_cgroup_classid:
3248 return &bpf_get_cgroup_classid_proto;
3249 case BPF_FUNC_skb_vlan_push:
3250 return &bpf_skb_vlan_push_proto;
3251 case BPF_FUNC_skb_vlan_pop:
3252 return &bpf_skb_vlan_pop_proto;
3253 case BPF_FUNC_skb_change_proto:
3254 return &bpf_skb_change_proto_proto;
3255 case BPF_FUNC_skb_change_type:
3256 return &bpf_skb_change_type_proto;
3257 case BPF_FUNC_skb_adjust_room:
3258 return &bpf_skb_adjust_room_proto;
3259 case BPF_FUNC_skb_change_tail:
3260 return &bpf_skb_change_tail_proto;
3261 case BPF_FUNC_skb_get_tunnel_key:
3262 return &bpf_skb_get_tunnel_key_proto;
3263 case BPF_FUNC_skb_set_tunnel_key:
3264 return bpf_get_skb_set_tunnel_proto(func_id);
3265 case BPF_FUNC_skb_get_tunnel_opt:
3266 return &bpf_skb_get_tunnel_opt_proto;
3267 case BPF_FUNC_skb_set_tunnel_opt:
3268 return bpf_get_skb_set_tunnel_proto(func_id);
3269 case BPF_FUNC_redirect:
3270 return &bpf_redirect_proto;
3271 case BPF_FUNC_get_route_realm:
3272 return &bpf_get_route_realm_proto;
3273 case BPF_FUNC_get_hash_recalc:
3274 return &bpf_get_hash_recalc_proto;
3275 case BPF_FUNC_set_hash_invalid:
3276 return &bpf_set_hash_invalid_proto;
3277 case BPF_FUNC_set_hash:
3278 return &bpf_set_hash_proto;
3279 case BPF_FUNC_perf_event_output:
3280 return &bpf_skb_event_output_proto;
3281 case BPF_FUNC_get_smp_processor_id:
3282 return &bpf_get_smp_processor_id_proto;
3283 case BPF_FUNC_skb_under_cgroup:
3284 return &bpf_skb_under_cgroup_proto;
3285 case BPF_FUNC_get_socket_cookie:
3286 return &bpf_get_socket_cookie_proto;
3287 case BPF_FUNC_get_socket_uid:
3288 return &bpf_get_socket_uid_proto;
3290 return bpf_base_func_proto(func_id);
3294 static const struct bpf_func_proto *
3295 xdp_func_proto(enum bpf_func_id func_id)
3298 case BPF_FUNC_perf_event_output:
3299 return &bpf_xdp_event_output_proto;
3300 case BPF_FUNC_get_smp_processor_id:
3301 return &bpf_get_smp_processor_id_proto;
3302 case BPF_FUNC_xdp_adjust_head:
3303 return &bpf_xdp_adjust_head_proto;
3304 case BPF_FUNC_redirect:
3305 return &bpf_xdp_redirect_proto;
3306 case BPF_FUNC_redirect_map:
3307 return &bpf_xdp_redirect_map_proto;
3309 return bpf_base_func_proto(func_id);
3313 static const struct bpf_func_proto *
3314 lwt_inout_func_proto(enum bpf_func_id func_id)
3317 case BPF_FUNC_skb_load_bytes:
3318 return &bpf_skb_load_bytes_proto;
3319 case BPF_FUNC_skb_pull_data:
3320 return &bpf_skb_pull_data_proto;
3321 case BPF_FUNC_csum_diff:
3322 return &bpf_csum_diff_proto;
3323 case BPF_FUNC_get_cgroup_classid:
3324 return &bpf_get_cgroup_classid_proto;
3325 case BPF_FUNC_get_route_realm:
3326 return &bpf_get_route_realm_proto;
3327 case BPF_FUNC_get_hash_recalc:
3328 return &bpf_get_hash_recalc_proto;
3329 case BPF_FUNC_perf_event_output:
3330 return &bpf_skb_event_output_proto;
3331 case BPF_FUNC_get_smp_processor_id:
3332 return &bpf_get_smp_processor_id_proto;
3333 case BPF_FUNC_skb_under_cgroup:
3334 return &bpf_skb_under_cgroup_proto;
3336 return bpf_base_func_proto(func_id);
3340 static const struct bpf_func_proto *
3341 sock_ops_func_proto(enum bpf_func_id func_id)
3344 case BPF_FUNC_setsockopt:
3345 return &bpf_setsockopt_proto;
3346 case BPF_FUNC_sock_map_update:
3347 return &bpf_sock_map_update_proto;
3349 return bpf_base_func_proto(func_id);
3353 static const struct bpf_func_proto *sk_skb_func_proto(enum bpf_func_id func_id)
3356 case BPF_FUNC_skb_store_bytes:
3357 return &bpf_skb_store_bytes_proto;
3358 case BPF_FUNC_skb_load_bytes:
3359 return &bpf_skb_load_bytes_proto;
3360 case BPF_FUNC_skb_pull_data:
3361 return &bpf_skb_pull_data_proto;
3362 case BPF_FUNC_skb_change_tail:
3363 return &bpf_skb_change_tail_proto;
3364 case BPF_FUNC_skb_change_head:
3365 return &bpf_skb_change_head_proto;
3366 case BPF_FUNC_get_socket_cookie:
3367 return &bpf_get_socket_cookie_proto;
3368 case BPF_FUNC_get_socket_uid:
3369 return &bpf_get_socket_uid_proto;
3370 case BPF_FUNC_sk_redirect_map:
3371 return &bpf_sk_redirect_map_proto;
3373 return bpf_base_func_proto(func_id);
3377 static const struct bpf_func_proto *
3378 lwt_xmit_func_proto(enum bpf_func_id func_id)
3381 case BPF_FUNC_skb_get_tunnel_key:
3382 return &bpf_skb_get_tunnel_key_proto;
3383 case BPF_FUNC_skb_set_tunnel_key:
3384 return bpf_get_skb_set_tunnel_proto(func_id);
3385 case BPF_FUNC_skb_get_tunnel_opt:
3386 return &bpf_skb_get_tunnel_opt_proto;
3387 case BPF_FUNC_skb_set_tunnel_opt:
3388 return bpf_get_skb_set_tunnel_proto(func_id);
3389 case BPF_FUNC_redirect:
3390 return &bpf_redirect_proto;
3391 case BPF_FUNC_clone_redirect:
3392 return &bpf_clone_redirect_proto;
3393 case BPF_FUNC_skb_change_tail:
3394 return &bpf_skb_change_tail_proto;
3395 case BPF_FUNC_skb_change_head:
3396 return &bpf_skb_change_head_proto;
3397 case BPF_FUNC_skb_store_bytes:
3398 return &bpf_skb_store_bytes_proto;
3399 case BPF_FUNC_csum_update:
3400 return &bpf_csum_update_proto;
3401 case BPF_FUNC_l3_csum_replace:
3402 return &bpf_l3_csum_replace_proto;
3403 case BPF_FUNC_l4_csum_replace:
3404 return &bpf_l4_csum_replace_proto;
3405 case BPF_FUNC_set_hash_invalid:
3406 return &bpf_set_hash_invalid_proto;
3408 return lwt_inout_func_proto(func_id);
3412 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
3413 struct bpf_insn_access_aux *info)
3415 const int size_default = sizeof(__u32);
3417 if (off < 0 || off >= sizeof(struct __sk_buff))
3420 /* The verifier guarantees that size > 0. */
3421 if (off % size != 0)
3425 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3426 if (off + size > offsetofend(struct __sk_buff, cb[4]))
3429 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
3430 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
3431 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
3432 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
3433 case bpf_ctx_range(struct __sk_buff, data):
3434 case bpf_ctx_range(struct __sk_buff, data_end):
3435 if (size != size_default)
3439 /* Only narrow read access allowed for now. */
3440 if (type == BPF_WRITE) {
3441 if (size != size_default)
3444 bpf_ctx_record_field_size(info, size_default);
3445 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
3453 static bool sk_filter_is_valid_access(int off, int size,
3454 enum bpf_access_type type,
3455 struct bpf_insn_access_aux *info)
3458 case bpf_ctx_range(struct __sk_buff, tc_classid):
3459 case bpf_ctx_range(struct __sk_buff, data):
3460 case bpf_ctx_range(struct __sk_buff, data_end):
3461 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3465 if (type == BPF_WRITE) {
3467 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3474 return bpf_skb_is_valid_access(off, size, type, info);
3477 static bool lwt_is_valid_access(int off, int size,
3478 enum bpf_access_type type,
3479 struct bpf_insn_access_aux *info)
3482 case bpf_ctx_range(struct __sk_buff, tc_classid):
3483 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3487 if (type == BPF_WRITE) {
3489 case bpf_ctx_range(struct __sk_buff, mark):
3490 case bpf_ctx_range(struct __sk_buff, priority):
3491 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3499 case bpf_ctx_range(struct __sk_buff, data):
3500 info->reg_type = PTR_TO_PACKET;
3502 case bpf_ctx_range(struct __sk_buff, data_end):
3503 info->reg_type = PTR_TO_PACKET_END;
3507 return bpf_skb_is_valid_access(off, size, type, info);
3510 static bool sock_filter_is_valid_access(int off, int size,
3511 enum bpf_access_type type,
3512 struct bpf_insn_access_aux *info)
3514 if (type == BPF_WRITE) {
3516 case offsetof(struct bpf_sock, bound_dev_if):
3517 case offsetof(struct bpf_sock, mark):
3518 case offsetof(struct bpf_sock, priority):
3525 if (off < 0 || off + size > sizeof(struct bpf_sock))
3527 /* The verifier guarantees that size > 0. */
3528 if (off % size != 0)
3530 if (size != sizeof(__u32))
3536 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
3537 const struct bpf_prog *prog, int drop_verdict)
3539 struct bpf_insn *insn = insn_buf;
3544 /* if (!skb->cloned)
3547 * (Fast-path, otherwise approximation that we might be
3548 * a clone, do the rest in helper.)
3550 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
3551 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
3552 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
3554 /* ret = bpf_skb_pull_data(skb, 0); */
3555 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
3556 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
3557 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
3558 BPF_FUNC_skb_pull_data);
3561 * return TC_ACT_SHOT;
3563 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
3564 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
3565 *insn++ = BPF_EXIT_INSN();
3568 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
3570 *insn++ = prog->insnsi[0];
3572 return insn - insn_buf;
3575 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
3576 const struct bpf_prog *prog)
3578 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
3581 static bool tc_cls_act_is_valid_access(int off, int size,
3582 enum bpf_access_type type,
3583 struct bpf_insn_access_aux *info)
3585 if (type == BPF_WRITE) {
3587 case bpf_ctx_range(struct __sk_buff, mark):
3588 case bpf_ctx_range(struct __sk_buff, tc_index):
3589 case bpf_ctx_range(struct __sk_buff, priority):
3590 case bpf_ctx_range(struct __sk_buff, tc_classid):
3591 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3599 case bpf_ctx_range(struct __sk_buff, data):
3600 info->reg_type = PTR_TO_PACKET;
3602 case bpf_ctx_range(struct __sk_buff, data_end):
3603 info->reg_type = PTR_TO_PACKET_END;
3605 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3609 return bpf_skb_is_valid_access(off, size, type, info);
3612 static bool __is_valid_xdp_access(int off, int size)
3614 if (off < 0 || off >= sizeof(struct xdp_md))
3616 if (off % size != 0)
3618 if (size != sizeof(__u32))
3624 static bool xdp_is_valid_access(int off, int size,
3625 enum bpf_access_type type,
3626 struct bpf_insn_access_aux *info)
3628 if (type == BPF_WRITE)
3632 case offsetof(struct xdp_md, data):
3633 info->reg_type = PTR_TO_PACKET;
3635 case offsetof(struct xdp_md, data_end):
3636 info->reg_type = PTR_TO_PACKET_END;
3640 return __is_valid_xdp_access(off, size);
3643 void bpf_warn_invalid_xdp_action(u32 act)
3645 const u32 act_max = XDP_REDIRECT;
3647 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
3648 act > act_max ? "Illegal" : "Driver unsupported",
3651 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
3653 static bool __is_valid_sock_ops_access(int off, int size)
3655 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
3657 /* The verifier guarantees that size > 0. */
3658 if (off % size != 0)
3660 if (size != sizeof(__u32))
3666 static bool sock_ops_is_valid_access(int off, int size,
3667 enum bpf_access_type type,
3668 struct bpf_insn_access_aux *info)
3670 if (type == BPF_WRITE) {
3672 case offsetof(struct bpf_sock_ops, op) ...
3673 offsetof(struct bpf_sock_ops, replylong[3]):
3680 return __is_valid_sock_ops_access(off, size);
3683 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
3684 const struct bpf_prog *prog)
3686 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
3689 static bool sk_skb_is_valid_access(int off, int size,
3690 enum bpf_access_type type,
3691 struct bpf_insn_access_aux *info)
3693 if (type == BPF_WRITE) {
3695 case bpf_ctx_range(struct __sk_buff, tc_index):
3696 case bpf_ctx_range(struct __sk_buff, priority):
3704 case bpf_ctx_range(struct __sk_buff, mark):
3705 case bpf_ctx_range(struct __sk_buff, tc_classid):
3707 case bpf_ctx_range(struct __sk_buff, data):
3708 info->reg_type = PTR_TO_PACKET;
3710 case bpf_ctx_range(struct __sk_buff, data_end):
3711 info->reg_type = PTR_TO_PACKET_END;
3715 return bpf_skb_is_valid_access(off, size, type, info);
3718 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
3719 const struct bpf_insn *si,
3720 struct bpf_insn *insn_buf,
3721 struct bpf_prog *prog, u32 *target_size)
3723 struct bpf_insn *insn = insn_buf;
3727 case offsetof(struct __sk_buff, len):
3728 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3729 bpf_target_off(struct sk_buff, len, 4,
3733 case offsetof(struct __sk_buff, protocol):
3734 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3735 bpf_target_off(struct sk_buff, protocol, 2,
3739 case offsetof(struct __sk_buff, vlan_proto):
3740 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3741 bpf_target_off(struct sk_buff, vlan_proto, 2,
3745 case offsetof(struct __sk_buff, priority):
3746 if (type == BPF_WRITE)
3747 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3748 bpf_target_off(struct sk_buff, priority, 4,
3751 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3752 bpf_target_off(struct sk_buff, priority, 4,
3756 case offsetof(struct __sk_buff, ingress_ifindex):
3757 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3758 bpf_target_off(struct sk_buff, skb_iif, 4,
3762 case offsetof(struct __sk_buff, ifindex):
3763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
3764 si->dst_reg, si->src_reg,
3765 offsetof(struct sk_buff, dev));
3766 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
3767 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3768 bpf_target_off(struct net_device, ifindex, 4,
3772 case offsetof(struct __sk_buff, hash):
3773 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3774 bpf_target_off(struct sk_buff, hash, 4,
3778 case offsetof(struct __sk_buff, mark):
3779 if (type == BPF_WRITE)
3780 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3781 bpf_target_off(struct sk_buff, mark, 4,
3784 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3785 bpf_target_off(struct sk_buff, mark, 4,
3789 case offsetof(struct __sk_buff, pkt_type):
3791 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
3793 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
3794 #ifdef __BIG_ENDIAN_BITFIELD
3795 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
3799 case offsetof(struct __sk_buff, queue_mapping):
3800 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3801 bpf_target_off(struct sk_buff, queue_mapping, 2,
3805 case offsetof(struct __sk_buff, vlan_present):
3806 case offsetof(struct __sk_buff, vlan_tci):
3807 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
3809 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3810 bpf_target_off(struct sk_buff, vlan_tci, 2,
3812 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
3813 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
3816 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
3817 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
3821 case offsetof(struct __sk_buff, cb[0]) ...
3822 offsetofend(struct __sk_buff, cb[4]) - 1:
3823 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
3824 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
3825 offsetof(struct qdisc_skb_cb, data)) %
3828 prog->cb_access = 1;
3830 off -= offsetof(struct __sk_buff, cb[0]);
3831 off += offsetof(struct sk_buff, cb);
3832 off += offsetof(struct qdisc_skb_cb, data);
3833 if (type == BPF_WRITE)
3834 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
3837 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
3841 case offsetof(struct __sk_buff, tc_classid):
3842 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
3845 off -= offsetof(struct __sk_buff, tc_classid);
3846 off += offsetof(struct sk_buff, cb);
3847 off += offsetof(struct qdisc_skb_cb, tc_classid);
3849 if (type == BPF_WRITE)
3850 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
3853 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
3857 case offsetof(struct __sk_buff, data):
3858 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
3859 si->dst_reg, si->src_reg,
3860 offsetof(struct sk_buff, data));
3863 case offsetof(struct __sk_buff, data_end):
3865 off -= offsetof(struct __sk_buff, data_end);
3866 off += offsetof(struct sk_buff, cb);
3867 off += offsetof(struct bpf_skb_data_end, data_end);
3868 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
3872 case offsetof(struct __sk_buff, tc_index):
3873 #ifdef CONFIG_NET_SCHED
3874 if (type == BPF_WRITE)
3875 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
3876 bpf_target_off(struct sk_buff, tc_index, 2,
3879 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3880 bpf_target_off(struct sk_buff, tc_index, 2,
3884 if (type == BPF_WRITE)
3885 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
3887 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3891 case offsetof(struct __sk_buff, napi_id):
3892 #if defined(CONFIG_NET_RX_BUSY_POLL)
3893 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3894 bpf_target_off(struct sk_buff, napi_id, 4,
3896 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
3897 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3900 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3903 case offsetof(struct __sk_buff, family):
3904 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
3906 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3907 si->dst_reg, si->src_reg,
3908 offsetof(struct sk_buff, sk));
3909 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
3910 bpf_target_off(struct sock_common,
3914 case offsetof(struct __sk_buff, remote_ip4):
3915 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
3917 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3918 si->dst_reg, si->src_reg,
3919 offsetof(struct sk_buff, sk));
3920 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3921 bpf_target_off(struct sock_common,
3925 case offsetof(struct __sk_buff, local_ip4):
3926 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
3927 skc_rcv_saddr) != 4);
3929 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3930 si->dst_reg, si->src_reg,
3931 offsetof(struct sk_buff, sk));
3932 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3933 bpf_target_off(struct sock_common,
3937 case offsetof(struct __sk_buff, remote_ip6[0]) ...
3938 offsetof(struct __sk_buff, remote_ip6[3]):
3939 #if IS_ENABLED(CONFIG_IPV6)
3940 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
3941 skc_v6_daddr.s6_addr32[0]) != 4);
3944 off -= offsetof(struct __sk_buff, remote_ip6[0]);
3946 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3947 si->dst_reg, si->src_reg,
3948 offsetof(struct sk_buff, sk));
3949 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3950 offsetof(struct sock_common,
3951 skc_v6_daddr.s6_addr32[0]) +
3954 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
3957 case offsetof(struct __sk_buff, local_ip6[0]) ...
3958 offsetof(struct __sk_buff, local_ip6[3]):
3959 #if IS_ENABLED(CONFIG_IPV6)
3960 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
3961 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
3964 off -= offsetof(struct __sk_buff, local_ip6[0]);
3966 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3967 si->dst_reg, si->src_reg,
3968 offsetof(struct sk_buff, sk));
3969 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3970 offsetof(struct sock_common,
3971 skc_v6_rcv_saddr.s6_addr32[0]) +
3974 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
3978 case offsetof(struct __sk_buff, remote_port):
3979 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
3981 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3982 si->dst_reg, si->src_reg,
3983 offsetof(struct sk_buff, sk));
3984 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
3985 bpf_target_off(struct sock_common,
3988 #ifndef __BIG_ENDIAN_BITFIELD
3989 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
3993 case offsetof(struct __sk_buff, local_port):
3994 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
3996 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3997 si->dst_reg, si->src_reg,
3998 offsetof(struct sk_buff, sk));
3999 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4000 bpf_target_off(struct sock_common,
4001 skc_num, 2, target_size));
4005 return insn - insn_buf;
4008 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
4009 const struct bpf_insn *si,
4010 struct bpf_insn *insn_buf,
4011 struct bpf_prog *prog, u32 *target_size)
4013 struct bpf_insn *insn = insn_buf;
4016 case offsetof(struct bpf_sock, bound_dev_if):
4017 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
4019 if (type == BPF_WRITE)
4020 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4021 offsetof(struct sock, sk_bound_dev_if));
4023 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4024 offsetof(struct sock, sk_bound_dev_if));
4027 case offsetof(struct bpf_sock, mark):
4028 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
4030 if (type == BPF_WRITE)
4031 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4032 offsetof(struct sock, sk_mark));
4034 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4035 offsetof(struct sock, sk_mark));
4038 case offsetof(struct bpf_sock, priority):
4039 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
4041 if (type == BPF_WRITE)
4042 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4043 offsetof(struct sock, sk_priority));
4045 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4046 offsetof(struct sock, sk_priority));
4049 case offsetof(struct bpf_sock, family):
4050 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
4052 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4053 offsetof(struct sock, sk_family));
4056 case offsetof(struct bpf_sock, type):
4057 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4058 offsetof(struct sock, __sk_flags_offset));
4059 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
4060 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
4063 case offsetof(struct bpf_sock, protocol):
4064 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4065 offsetof(struct sock, __sk_flags_offset));
4066 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
4067 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
4071 return insn - insn_buf;
4074 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
4075 const struct bpf_insn *si,
4076 struct bpf_insn *insn_buf,
4077 struct bpf_prog *prog, u32 *target_size)
4079 struct bpf_insn *insn = insn_buf;
4082 case offsetof(struct __sk_buff, ifindex):
4083 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4084 si->dst_reg, si->src_reg,
4085 offsetof(struct sk_buff, dev));
4086 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4087 bpf_target_off(struct net_device, ifindex, 4,
4091 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4095 return insn - insn_buf;
4098 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
4099 const struct bpf_insn *si,
4100 struct bpf_insn *insn_buf,
4101 struct bpf_prog *prog, u32 *target_size)
4103 struct bpf_insn *insn = insn_buf;
4106 case offsetof(struct xdp_md, data):
4107 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
4108 si->dst_reg, si->src_reg,
4109 offsetof(struct xdp_buff, data));
4111 case offsetof(struct xdp_md, data_end):
4112 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
4113 si->dst_reg, si->src_reg,
4114 offsetof(struct xdp_buff, data_end));
4118 return insn - insn_buf;
4121 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
4122 const struct bpf_insn *si,
4123 struct bpf_insn *insn_buf,
4124 struct bpf_prog *prog,
4127 struct bpf_insn *insn = insn_buf;
4131 case offsetof(struct bpf_sock_ops, op) ...
4132 offsetof(struct bpf_sock_ops, replylong[3]):
4133 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
4134 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
4135 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
4136 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
4137 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
4138 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
4140 off -= offsetof(struct bpf_sock_ops, op);
4141 off += offsetof(struct bpf_sock_ops_kern, op);
4142 if (type == BPF_WRITE)
4143 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4146 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4150 case offsetof(struct bpf_sock_ops, family):
4151 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4153 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4154 struct bpf_sock_ops_kern, sk),
4155 si->dst_reg, si->src_reg,
4156 offsetof(struct bpf_sock_ops_kern, sk));
4157 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4158 offsetof(struct sock_common, skc_family));
4161 case offsetof(struct bpf_sock_ops, remote_ip4):
4162 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4164 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4165 struct bpf_sock_ops_kern, sk),
4166 si->dst_reg, si->src_reg,
4167 offsetof(struct bpf_sock_ops_kern, sk));
4168 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4169 offsetof(struct sock_common, skc_daddr));
4172 case offsetof(struct bpf_sock_ops, local_ip4):
4173 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4);
4175 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4176 struct bpf_sock_ops_kern, sk),
4177 si->dst_reg, si->src_reg,
4178 offsetof(struct bpf_sock_ops_kern, sk));
4179 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4180 offsetof(struct sock_common,
4184 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
4185 offsetof(struct bpf_sock_ops, remote_ip6[3]):
4186 #if IS_ENABLED(CONFIG_IPV6)
4187 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4188 skc_v6_daddr.s6_addr32[0]) != 4);
4191 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
4192 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4193 struct bpf_sock_ops_kern, sk),
4194 si->dst_reg, si->src_reg,
4195 offsetof(struct bpf_sock_ops_kern, sk));
4196 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4197 offsetof(struct sock_common,
4198 skc_v6_daddr.s6_addr32[0]) +
4201 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4205 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
4206 offsetof(struct bpf_sock_ops, local_ip6[3]):
4207 #if IS_ENABLED(CONFIG_IPV6)
4208 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4209 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4212 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
4213 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4214 struct bpf_sock_ops_kern, sk),
4215 si->dst_reg, si->src_reg,
4216 offsetof(struct bpf_sock_ops_kern, sk));
4217 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4218 offsetof(struct sock_common,
4219 skc_v6_rcv_saddr.s6_addr32[0]) +
4222 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4226 case offsetof(struct bpf_sock_ops, remote_port):
4227 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4229 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4230 struct bpf_sock_ops_kern, sk),
4231 si->dst_reg, si->src_reg,
4232 offsetof(struct bpf_sock_ops_kern, sk));
4233 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4234 offsetof(struct sock_common, skc_dport));
4235 #ifndef __BIG_ENDIAN_BITFIELD
4236 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4240 case offsetof(struct bpf_sock_ops, local_port):
4241 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4244 struct bpf_sock_ops_kern, sk),
4245 si->dst_reg, si->src_reg,
4246 offsetof(struct bpf_sock_ops_kern, sk));
4247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4248 offsetof(struct sock_common, skc_num));
4251 return insn - insn_buf;
4254 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
4255 const struct bpf_insn *si,
4256 struct bpf_insn *insn_buf,
4257 struct bpf_prog *prog, u32 *target_size)
4259 struct bpf_insn *insn = insn_buf;
4263 case offsetof(struct __sk_buff, data_end):
4265 off -= offsetof(struct __sk_buff, data_end);
4266 off += offsetof(struct sk_buff, cb);
4267 off += offsetof(struct tcp_skb_cb, bpf.data_end);
4268 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
4272 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4276 return insn - insn_buf;
4279 const struct bpf_verifier_ops sk_filter_prog_ops = {
4280 .get_func_proto = sk_filter_func_proto,
4281 .is_valid_access = sk_filter_is_valid_access,
4282 .convert_ctx_access = bpf_convert_ctx_access,
4285 const struct bpf_verifier_ops tc_cls_act_prog_ops = {
4286 .get_func_proto = tc_cls_act_func_proto,
4287 .is_valid_access = tc_cls_act_is_valid_access,
4288 .convert_ctx_access = tc_cls_act_convert_ctx_access,
4289 .gen_prologue = tc_cls_act_prologue,
4290 .test_run = bpf_prog_test_run_skb,
4293 const struct bpf_verifier_ops xdp_prog_ops = {
4294 .get_func_proto = xdp_func_proto,
4295 .is_valid_access = xdp_is_valid_access,
4296 .convert_ctx_access = xdp_convert_ctx_access,
4297 .test_run = bpf_prog_test_run_xdp,
4300 const struct bpf_verifier_ops cg_skb_prog_ops = {
4301 .get_func_proto = sk_filter_func_proto,
4302 .is_valid_access = sk_filter_is_valid_access,
4303 .convert_ctx_access = bpf_convert_ctx_access,
4304 .test_run = bpf_prog_test_run_skb,
4307 const struct bpf_verifier_ops lwt_inout_prog_ops = {
4308 .get_func_proto = lwt_inout_func_proto,
4309 .is_valid_access = lwt_is_valid_access,
4310 .convert_ctx_access = bpf_convert_ctx_access,
4311 .test_run = bpf_prog_test_run_skb,
4314 const struct bpf_verifier_ops lwt_xmit_prog_ops = {
4315 .get_func_proto = lwt_xmit_func_proto,
4316 .is_valid_access = lwt_is_valid_access,
4317 .convert_ctx_access = bpf_convert_ctx_access,
4318 .gen_prologue = tc_cls_act_prologue,
4319 .test_run = bpf_prog_test_run_skb,
4322 const struct bpf_verifier_ops cg_sock_prog_ops = {
4323 .get_func_proto = sock_filter_func_proto,
4324 .is_valid_access = sock_filter_is_valid_access,
4325 .convert_ctx_access = sock_filter_convert_ctx_access,
4328 const struct bpf_verifier_ops sock_ops_prog_ops = {
4329 .get_func_proto = sock_ops_func_proto,
4330 .is_valid_access = sock_ops_is_valid_access,
4331 .convert_ctx_access = sock_ops_convert_ctx_access,
4334 const struct bpf_verifier_ops sk_skb_prog_ops = {
4335 .get_func_proto = sk_skb_func_proto,
4336 .is_valid_access = sk_skb_is_valid_access,
4337 .convert_ctx_access = sk_skb_convert_ctx_access,
4338 .gen_prologue = sk_skb_prologue,
4341 int sk_detach_filter(struct sock *sk)
4344 struct sk_filter *filter;
4346 if (sock_flag(sk, SOCK_FILTER_LOCKED))
4349 filter = rcu_dereference_protected(sk->sk_filter,
4350 lockdep_sock_is_held(sk));
4352 RCU_INIT_POINTER(sk->sk_filter, NULL);
4353 sk_filter_uncharge(sk, filter);
4359 EXPORT_SYMBOL_GPL(sk_detach_filter);
4361 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
4364 struct sock_fprog_kern *fprog;
4365 struct sk_filter *filter;
4369 filter = rcu_dereference_protected(sk->sk_filter,
4370 lockdep_sock_is_held(sk));
4374 /* We're copying the filter that has been originally attached,
4375 * so no conversion/decode needed anymore. eBPF programs that
4376 * have no original program cannot be dumped through this.
4379 fprog = filter->prog->orig_prog;
4385 /* User space only enquires number of filter blocks. */
4389 if (len < fprog->len)
4393 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
4396 /* Instead of bytes, the API requests to return the number