]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame_incremental - net/core/filter.c
projects / mirror_ubuntu-artful-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
net: samples: pktgen mode samples/tests for qdisc layer
[mirror_ubuntu-artful-kernel.git] / net / core / filter.c
... / ...
CommitLineData
1/*
2 * Linux Socket Filter - Kernel level socket filtering
3 *
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
6 *
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
8 *
9 * Authors:
10 *
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 *
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
22 */
23
24#include <linux/module.h>
25#include <linux/types.h>
26#include <linux/mm.h>
27#include <linux/fcntl.h>
28#include <linux/socket.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/netdevice.h>
32#include <linux/if_packet.h>
33#include <linux/gfp.h>
34#include <net/ip.h>
35#include <net/protocol.h>
36#include <net/netlink.h>
37#include <linux/skbuff.h>
38#include <net/sock.h>
39#include <net/flow_dissector.h>
40#include <linux/errno.h>
41#include <linux/timer.h>
42#include <asm/uaccess.h>
43#include <asm/unaligned.h>
44#include <linux/filter.h>
45#include <linux/ratelimit.h>
46#include <linux/seccomp.h>
47#include <linux/if_vlan.h>
48#include <linux/bpf.h>
49#include <net/sch_generic.h>
50#include <net/cls_cgroup.h>
51#include <net/dst_metadata.h>
52#include <net/dst.h>
53#include <net/sock_reuseport.h>
54
55/**
56 * sk_filter - run a packet through a socket filter
57 * @sk: sock associated with &sk_buff
58 * @skb: buffer to filter
59 *
60 * Run the eBPF program and then cut skb->data to correct size returned by
61 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
62 * than pkt_len we keep whole skb->data. This is the socket level
63 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
64 * be accepted or -EPERM if the packet should be tossed.
65 *
66 */
67int sk_filter(struct sock *sk, struct sk_buff *skb)
68{
69 int err;
70 struct sk_filter *filter;
71
72 /*
73 * If the skb was allocated from pfmemalloc reserves, only
74 * allow SOCK_MEMALLOC sockets to use it as this socket is
75 * helping free memory
76 */
77 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
78 return -ENOMEM;
79
80 err = security_sock_rcv_skb(sk, skb);
81 if (err)
82 return err;
83
84 rcu_read_lock();
85 filter = rcu_dereference(sk->sk_filter);
86 if (filter) {
87 unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
88
89 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
90 }
91 rcu_read_unlock();
92
93 return err;
94}
95EXPORT_SYMBOL(sk_filter);
96
97static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
98{
99 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
100}
101
102static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
103{
104 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
105 struct nlattr *nla;
106
107 if (skb_is_nonlinear(skb))
108 return 0;
109
110 if (skb->len < sizeof(struct nlattr))
111 return 0;
112
113 if (a > skb->len - sizeof(struct nlattr))
114 return 0;
115
116 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
117 if (nla)
118 return (void *) nla - (void *) skb->data;
119
120 return 0;
121}
122
123static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
124{
125 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
126 struct nlattr *nla;
127
128 if (skb_is_nonlinear(skb))
129 return 0;
130
131 if (skb->len < sizeof(struct nlattr))
132 return 0;
133
134 if (a > skb->len - sizeof(struct nlattr))
135 return 0;
136
137 nla = (struct nlattr *) &skb->data[a];
138 if (nla->nla_len > skb->len - a)
139 return 0;
140
141 nla = nla_find_nested(nla, x);
142 if (nla)
143 return (void *) nla - (void *) skb->data;
144
145 return 0;
146}
147
148static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
149{
150 return raw_smp_processor_id();
151}
152
153static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
154 .func = __get_raw_cpu_id,
155 .gpl_only = false,
156 .ret_type = RET_INTEGER,
157};
158
159static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
160 struct bpf_insn *insn_buf)
161{
162 struct bpf_insn *insn = insn_buf;
163
164 switch (skb_field) {
165 case SKF_AD_MARK:
166 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
167
168 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
169 offsetof(struct sk_buff, mark));
170 break;
171
172 case SKF_AD_PKTTYPE:
173 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
174 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
175#ifdef __BIG_ENDIAN_BITFIELD
176 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
177#endif
178 break;
179
180 case SKF_AD_QUEUE:
181 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
182
183 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
184 offsetof(struct sk_buff, queue_mapping));
185 break;
186
187 case SKF_AD_VLAN_TAG:
188 case SKF_AD_VLAN_TAG_PRESENT:
189 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
190 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
191
192 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
193 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
194 offsetof(struct sk_buff, vlan_tci));
195 if (skb_field == SKF_AD_VLAN_TAG) {
196 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
197 ~VLAN_TAG_PRESENT);
198 } else {
199 /* dst_reg >>= 12 */
200 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
201 /* dst_reg &= 1 */
202 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
203 }
204 break;
205 }
206
207 return insn - insn_buf;
208}
209
210static bool convert_bpf_extensions(struct sock_filter *fp,
211 struct bpf_insn **insnp)
212{
213 struct bpf_insn *insn = *insnp;
214 u32 cnt;
215
216 switch (fp->k) {
217 case SKF_AD_OFF + SKF_AD_PROTOCOL:
218 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
219
220 /* A = *(u16 *) (CTX + offsetof(protocol)) */
221 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
222 offsetof(struct sk_buff, protocol));
223 /* A = ntohs(A) [emitting a nop or swap16] */
224 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
225 break;
226
227 case SKF_AD_OFF + SKF_AD_PKTTYPE:
228 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
229 insn += cnt - 1;
230 break;
231
232 case SKF_AD_OFF + SKF_AD_IFINDEX:
233 case SKF_AD_OFF + SKF_AD_HATYPE:
234 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
235 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
236 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
237
238 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
239 BPF_REG_TMP, BPF_REG_CTX,
240 offsetof(struct sk_buff, dev));
241 /* if (tmp != 0) goto pc + 1 */
242 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
243 *insn++ = BPF_EXIT_INSN();
244 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
245 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
246 offsetof(struct net_device, ifindex));
247 else
248 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
249 offsetof(struct net_device, type));
250 break;
251
252 case SKF_AD_OFF + SKF_AD_MARK:
253 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
254 insn += cnt - 1;
255 break;
256
257 case SKF_AD_OFF + SKF_AD_RXHASH:
258 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
259
260 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
261 offsetof(struct sk_buff, hash));
262 break;
263
264 case SKF_AD_OFF + SKF_AD_QUEUE:
265 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
266 insn += cnt - 1;
267 break;
268
269 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
270 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
271 BPF_REG_A, BPF_REG_CTX, insn);
272 insn += cnt - 1;
273 break;
274
275 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
276 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
277 BPF_REG_A, BPF_REG_CTX, insn);
278 insn += cnt - 1;
279 break;
280
281 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
282 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
283
284 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
285 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
286 offsetof(struct sk_buff, vlan_proto));
287 /* A = ntohs(A) [emitting a nop or swap16] */
288 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
289 break;
290
291 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
292 case SKF_AD_OFF + SKF_AD_NLATTR:
293 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
294 case SKF_AD_OFF + SKF_AD_CPU:
295 case SKF_AD_OFF + SKF_AD_RANDOM:
296 /* arg1 = CTX */
297 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
298 /* arg2 = A */
299 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
300 /* arg3 = X */
301 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
302 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
303 switch (fp->k) {
304 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
305 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
306 break;
307 case SKF_AD_OFF + SKF_AD_NLATTR:
308 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
309 break;
310 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
311 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
312 break;
313 case SKF_AD_OFF + SKF_AD_CPU:
314 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
315 break;
316 case SKF_AD_OFF + SKF_AD_RANDOM:
317 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
318 bpf_user_rnd_init_once();
319 break;
320 }
321 break;
322
323 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
324 /* A ^= X */
325 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
326 break;
327
328 default:
329 /* This is just a dummy call to avoid letting the compiler
330 * evict __bpf_call_base() as an optimization. Placed here
331 * where no-one bothers.
332 */
333 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
334 return false;
335 }
336
337 *insnp = insn;
338 return true;
339}
340
341/**
342 * bpf_convert_filter - convert filter program
343 * @prog: the user passed filter program
344 * @len: the length of the user passed filter program
345 * @new_prog: buffer where converted program will be stored
346 * @new_len: pointer to store length of converted program
347 *
348 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
349 * Conversion workflow:
350 *
351 * 1) First pass for calculating the new program length:
352 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
353 *
354 * 2) 2nd pass to remap in two passes: 1st pass finds new
355 * jump offsets, 2nd pass remapping:
356 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
357 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
358 */
359static int bpf_convert_filter(struct sock_filter *prog, int len,
360 struct bpf_insn *new_prog, int *new_len)
361{
362 int new_flen = 0, pass = 0, target, i;
363 struct bpf_insn *new_insn;
364 struct sock_filter *fp;
365 int *addrs = NULL;
366 u8 bpf_src;
367
368 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
369 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
370
371 if (len <= 0 || len > BPF_MAXINSNS)
372 return -EINVAL;
373
374 if (new_prog) {
375 addrs = kcalloc(len, sizeof(*addrs),
376 GFP_KERNEL | __GFP_NOWARN);
377 if (!addrs)
378 return -ENOMEM;
379 }
380
381do_pass:
382 new_insn = new_prog;
383 fp = prog;
384
385 /* Classic BPF related prologue emission. */
386 if (new_insn) {
387 /* Classic BPF expects A and X to be reset first. These need
388 * to be guaranteed to be the first two instructions.
389 */
390 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
391 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
392
393 /* All programs must keep CTX in callee saved BPF_REG_CTX.
394 * In eBPF case it's done by the compiler, here we need to
395 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
396 */
397 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
398 } else {
399 new_insn += 3;
400 }
401
402 for (i = 0; i < len; fp++, i++) {
403 struct bpf_insn tmp_insns[6] = { };
404 struct bpf_insn *insn = tmp_insns;
405
406 if (addrs)
407 addrs[i] = new_insn - new_prog;
408
409 switch (fp->code) {
410 /* All arithmetic insns and skb loads map as-is. */
411 case BPF_ALU | BPF_ADD | BPF_X:
412 case BPF_ALU | BPF_ADD | BPF_K:
413 case BPF_ALU | BPF_SUB | BPF_X:
414 case BPF_ALU | BPF_SUB | BPF_K:
415 case BPF_ALU | BPF_AND | BPF_X:
416 case BPF_ALU | BPF_AND | BPF_K:
417 case BPF_ALU | BPF_OR | BPF_X:
418 case BPF_ALU | BPF_OR | BPF_K:
419 case BPF_ALU | BPF_LSH | BPF_X:
420 case BPF_ALU | BPF_LSH | BPF_K:
421 case BPF_ALU | BPF_RSH | BPF_X:
422 case BPF_ALU | BPF_RSH | BPF_K:
423 case BPF_ALU | BPF_XOR | BPF_X:
424 case BPF_ALU | BPF_XOR | BPF_K:
425 case BPF_ALU | BPF_MUL | BPF_X:
426 case BPF_ALU | BPF_MUL | BPF_K:
427 case BPF_ALU | BPF_DIV | BPF_X:
428 case BPF_ALU | BPF_DIV | BPF_K:
429 case BPF_ALU | BPF_MOD | BPF_X:
430 case BPF_ALU | BPF_MOD | BPF_K:
431 case BPF_ALU | BPF_NEG:
432 case BPF_LD | BPF_ABS | BPF_W:
433 case BPF_LD | BPF_ABS | BPF_H:
434 case BPF_LD | BPF_ABS | BPF_B:
435 case BPF_LD | BPF_IND | BPF_W:
436 case BPF_LD | BPF_IND | BPF_H:
437 case BPF_LD | BPF_IND | BPF_B:
438 /* Check for overloaded BPF extension and
439 * directly convert it if found, otherwise
440 * just move on with mapping.
441 */
442 if (BPF_CLASS(fp->code) == BPF_LD &&
443 BPF_MODE(fp->code) == BPF_ABS &&
444 convert_bpf_extensions(fp, &insn))
445 break;
446
447 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
448 break;
449
450 /* Jump transformation cannot use BPF block macros
451 * everywhere as offset calculation and target updates
452 * require a bit more work than the rest, i.e. jump
453 * opcodes map as-is, but offsets need adjustment.
454 */
455
456#define BPF_EMIT_JMP \
457 do { \
458 if (target >= len || target < 0) \
459 goto err; \
460 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
461 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
462 insn->off -= insn - tmp_insns; \
463 } while (0)
464
465 case BPF_JMP | BPF_JA:
466 target = i + fp->k + 1;
467 insn->code = fp->code;
468 BPF_EMIT_JMP;
469 break;
470
471 case BPF_JMP | BPF_JEQ | BPF_K:
472 case BPF_JMP | BPF_JEQ | BPF_X:
473 case BPF_JMP | BPF_JSET | BPF_K:
474 case BPF_JMP | BPF_JSET | BPF_X:
475 case BPF_JMP | BPF_JGT | BPF_K:
476 case BPF_JMP | BPF_JGT | BPF_X:
477 case BPF_JMP | BPF_JGE | BPF_K:
478 case BPF_JMP | BPF_JGE | BPF_X:
479 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
480 /* BPF immediates are signed, zero extend
481 * immediate into tmp register and use it
482 * in compare insn.
483 */
484 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
485
486 insn->dst_reg = BPF_REG_A;
487 insn->src_reg = BPF_REG_TMP;
488 bpf_src = BPF_X;
489 } else {
490 insn->dst_reg = BPF_REG_A;
491 insn->imm = fp->k;
492 bpf_src = BPF_SRC(fp->code);
493 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
494 }
495
496 /* Common case where 'jump_false' is next insn. */
497 if (fp->jf == 0) {
498 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
499 target = i + fp->jt + 1;
500 BPF_EMIT_JMP;
501 break;
502 }
503
504 /* Convert JEQ into JNE when 'jump_true' is next insn. */
505 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
506 insn->code = BPF_JMP | BPF_JNE | bpf_src;
507 target = i + fp->jf + 1;
508 BPF_EMIT_JMP;
509 break;
510 }
511
512 /* Other jumps are mapped into two insns: Jxx and JA. */
513 target = i + fp->jt + 1;
514 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
515 BPF_EMIT_JMP;
516 insn++;
517
518 insn->code = BPF_JMP | BPF_JA;
519 target = i + fp->jf + 1;
520 BPF_EMIT_JMP;
521 break;
522
523 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
524 case BPF_LDX | BPF_MSH | BPF_B:
525 /* tmp = A */
526 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
527 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
528 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
529 /* A &= 0xf */
530 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
531 /* A <<= 2 */
532 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
533 /* X = A */
534 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
535 /* A = tmp */
536 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
537 break;
538
539 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
540 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
541 */
542 case BPF_RET | BPF_A:
543 case BPF_RET | BPF_K:
544 if (BPF_RVAL(fp->code) == BPF_K)
545 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
546 0, fp->k);
547 *insn = BPF_EXIT_INSN();
548 break;
549
550 /* Store to stack. */
551 case BPF_ST:
552 case BPF_STX:
553 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
554 BPF_ST ? BPF_REG_A : BPF_REG_X,
555 -(BPF_MEMWORDS - fp->k) * 4);
556 break;
557
558 /* Load from stack. */
559 case BPF_LD | BPF_MEM:
560 case BPF_LDX | BPF_MEM:
561 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
562 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
563 -(BPF_MEMWORDS - fp->k) * 4);
564 break;
565
566 /* A = K or X = K */
567 case BPF_LD | BPF_IMM:
568 case BPF_LDX | BPF_IMM:
569 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
570 BPF_REG_A : BPF_REG_X, fp->k);
571 break;
572
573 /* X = A */
574 case BPF_MISC | BPF_TAX:
575 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
576 break;
577
578 /* A = X */
579 case BPF_MISC | BPF_TXA:
580 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
581 break;
582
583 /* A = skb->len or X = skb->len */
584 case BPF_LD | BPF_W | BPF_LEN:
585 case BPF_LDX | BPF_W | BPF_LEN:
586 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
587 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
588 offsetof(struct sk_buff, len));
589 break;
590
591 /* Access seccomp_data fields. */
592 case BPF_LDX | BPF_ABS | BPF_W:
593 /* A = *(u32 *) (ctx + K) */
594 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
595 break;
596
597 /* Unknown instruction. */
598 default:
599 goto err;
600 }
601
602 insn++;
603 if (new_prog)
604 memcpy(new_insn, tmp_insns,
605 sizeof(*insn) * (insn - tmp_insns));
606 new_insn += insn - tmp_insns;
607 }
608
609 if (!new_prog) {
610 /* Only calculating new length. */
611 *new_len = new_insn - new_prog;
612 return 0;
613 }
614
615 pass++;
616 if (new_flen != new_insn - new_prog) {
617 new_flen = new_insn - new_prog;
618 if (pass > 2)
619 goto err;
620 goto do_pass;
621 }
622
623 kfree(addrs);
624 BUG_ON(*new_len != new_flen);
625 return 0;
626err:
627 kfree(addrs);
628 return -EINVAL;
629}
630
631/* Security:
632 *
633 * As we dont want to clear mem[] array for each packet going through
634 * __bpf_prog_run(), we check that filter loaded by user never try to read
635 * a cell if not previously written, and we check all branches to be sure
636 * a malicious user doesn't try to abuse us.
637 */
638static int check_load_and_stores(const struct sock_filter *filter, int flen)
639{
640 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
641 int pc, ret = 0;
642
643 BUILD_BUG_ON(BPF_MEMWORDS > 16);
644
645 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
646 if (!masks)
647 return -ENOMEM;
648
649 memset(masks, 0xff, flen * sizeof(*masks));
650
651 for (pc = 0; pc < flen; pc++) {
652 memvalid &= masks[pc];
653
654 switch (filter[pc].code) {
655 case BPF_ST:
656 case BPF_STX:
657 memvalid |= (1 << filter[pc].k);
658 break;
659 case BPF_LD | BPF_MEM:
660 case BPF_LDX | BPF_MEM:
661 if (!(memvalid & (1 << filter[pc].k))) {
662 ret = -EINVAL;
663 goto error;
664 }
665 break;
666 case BPF_JMP | BPF_JA:
667 /* A jump must set masks on target */
668 masks[pc + 1 + filter[pc].k] &= memvalid;
669 memvalid = ~0;
670 break;
671 case BPF_JMP | BPF_JEQ | BPF_K:
672 case BPF_JMP | BPF_JEQ | BPF_X:
673 case BPF_JMP | BPF_JGE | BPF_K:
674 case BPF_JMP | BPF_JGE | BPF_X:
675 case BPF_JMP | BPF_JGT | BPF_K:
676 case BPF_JMP | BPF_JGT | BPF_X:
677 case BPF_JMP | BPF_JSET | BPF_K:
678 case BPF_JMP | BPF_JSET | BPF_X:
679 /* A jump must set masks on targets */
680 masks[pc + 1 + filter[pc].jt] &= memvalid;
681 masks[pc + 1 + filter[pc].jf] &= memvalid;
682 memvalid = ~0;
683 break;
684 }
685 }
686error:
687 kfree(masks);
688 return ret;
689}
690
691static bool chk_code_allowed(u16 code_to_probe)
692{
693 static const bool codes[] = {
694 /* 32 bit ALU operations */
695 [BPF_ALU | BPF_ADD | BPF_K] = true,
696 [BPF_ALU | BPF_ADD | BPF_X] = true,
697 [BPF_ALU | BPF_SUB | BPF_K] = true,
698 [BPF_ALU | BPF_SUB | BPF_X] = true,
699 [BPF_ALU | BPF_MUL | BPF_K] = true,
700 [BPF_ALU | BPF_MUL | BPF_X] = true,
701 [BPF_ALU | BPF_DIV | BPF_K] = true,
702 [BPF_ALU | BPF_DIV | BPF_X] = true,
703 [BPF_ALU | BPF_MOD | BPF_K] = true,
704 [BPF_ALU | BPF_MOD | BPF_X] = true,
705 [BPF_ALU | BPF_AND | BPF_K] = true,
706 [BPF_ALU | BPF_AND | BPF_X] = true,
707 [BPF_ALU | BPF_OR | BPF_K] = true,
708 [BPF_ALU | BPF_OR | BPF_X] = true,
709 [BPF_ALU | BPF_XOR | BPF_K] = true,
710 [BPF_ALU | BPF_XOR | BPF_X] = true,
711 [BPF_ALU | BPF_LSH | BPF_K] = true,
712 [BPF_ALU | BPF_LSH | BPF_X] = true,
713 [BPF_ALU | BPF_RSH | BPF_K] = true,
714 [BPF_ALU | BPF_RSH | BPF_X] = true,
715 [BPF_ALU | BPF_NEG] = true,
716 /* Load instructions */
717 [BPF_LD | BPF_W | BPF_ABS] = true,
718 [BPF_LD | BPF_H | BPF_ABS] = true,
719 [BPF_LD | BPF_B | BPF_ABS] = true,
720 [BPF_LD | BPF_W | BPF_LEN] = true,
721 [BPF_LD | BPF_W | BPF_IND] = true,
722 [BPF_LD | BPF_H | BPF_IND] = true,
723 [BPF_LD | BPF_B | BPF_IND] = true,
724 [BPF_LD | BPF_IMM] = true,
725 [BPF_LD | BPF_MEM] = true,
726 [BPF_LDX | BPF_W | BPF_LEN] = true,
727 [BPF_LDX | BPF_B | BPF_MSH] = true,
728 [BPF_LDX | BPF_IMM] = true,
729 [BPF_LDX | BPF_MEM] = true,
730 /* Store instructions */
731 [BPF_ST] = true,
732 [BPF_STX] = true,
733 /* Misc instructions */
734 [BPF_MISC | BPF_TAX] = true,
735 [BPF_MISC | BPF_TXA] = true,
736 /* Return instructions */
737 [BPF_RET | BPF_K] = true,
738 [BPF_RET | BPF_A] = true,
739 /* Jump instructions */
740 [BPF_JMP | BPF_JA] = true,
741 [BPF_JMP | BPF_JEQ | BPF_K] = true,
742 [BPF_JMP | BPF_JEQ | BPF_X] = true,
743 [BPF_JMP | BPF_JGE | BPF_K] = true,
744 [BPF_JMP | BPF_JGE | BPF_X] = true,
745 [BPF_JMP | BPF_JGT | BPF_K] = true,
746 [BPF_JMP | BPF_JGT | BPF_X] = true,
747 [BPF_JMP | BPF_JSET | BPF_K] = true,
748 [BPF_JMP | BPF_JSET | BPF_X] = true,
749 };
750
751 if (code_to_probe >= ARRAY_SIZE(codes))
752 return false;
753
754 return codes[code_to_probe];
755}
756
757static bool bpf_check_basics_ok(const struct sock_filter *filter,
758 unsigned int flen)
759{
760 if (filter == NULL)
761 return false;
762 if (flen == 0 || flen > BPF_MAXINSNS)
763 return false;
764
765 return true;
766}
767
768/**
769 * bpf_check_classic - verify socket filter code
770 * @filter: filter to verify
771 * @flen: length of filter
772 *
773 * Check the user's filter code. If we let some ugly
774 * filter code slip through kaboom! The filter must contain
775 * no references or jumps that are out of range, no illegal
776 * instructions, and must end with a RET instruction.
777 *
778 * All jumps are forward as they are not signed.
779 *
780 * Returns 0 if the rule set is legal or -EINVAL if not.
781 */
782static int bpf_check_classic(const struct sock_filter *filter,
783 unsigned int flen)
784{
785 bool anc_found;
786 int pc;
787
788 /* Check the filter code now */
789 for (pc = 0; pc < flen; pc++) {
790 const struct sock_filter *ftest = &filter[pc];
791
792 /* May we actually operate on this code? */
793 if (!chk_code_allowed(ftest->code))
794 return -EINVAL;
795
796 /* Some instructions need special checks */
797 switch (ftest->code) {
798 case BPF_ALU | BPF_DIV | BPF_K:
799 case BPF_ALU | BPF_MOD | BPF_K:
800 /* Check for division by zero */
801 if (ftest->k == 0)
802 return -EINVAL;
803 break;
804 case BPF_ALU | BPF_LSH | BPF_K:
805 case BPF_ALU | BPF_RSH | BPF_K:
806 if (ftest->k >= 32)
807 return -EINVAL;
808 break;
809 case BPF_LD | BPF_MEM:
810 case BPF_LDX | BPF_MEM:
811 case BPF_ST:
812 case BPF_STX:
813 /* Check for invalid memory addresses */
814 if (ftest->k >= BPF_MEMWORDS)
815 return -EINVAL;
816 break;
817 case BPF_JMP | BPF_JA:
818 /* Note, the large ftest->k might cause loops.
819 * Compare this with conditional jumps below,
820 * where offsets are limited. --ANK (981016)
821 */
822 if (ftest->k >= (unsigned int)(flen - pc - 1))
823 return -EINVAL;
824 break;
825 case BPF_JMP | BPF_JEQ | BPF_K:
826 case BPF_JMP | BPF_JEQ | BPF_X:
827 case BPF_JMP | BPF_JGE | BPF_K:
828 case BPF_JMP | BPF_JGE | BPF_X:
829 case BPF_JMP | BPF_JGT | BPF_K:
830 case BPF_JMP | BPF_JGT | BPF_X:
831 case BPF_JMP | BPF_JSET | BPF_K:
832 case BPF_JMP | BPF_JSET | BPF_X:
833 /* Both conditionals must be safe */
834 if (pc + ftest->jt + 1 >= flen ||
835 pc + ftest->jf + 1 >= flen)
836 return -EINVAL;
837 break;
838 case BPF_LD | BPF_W | BPF_ABS:
839 case BPF_LD | BPF_H | BPF_ABS:
840 case BPF_LD | BPF_B | BPF_ABS:
841 anc_found = false;
842 if (bpf_anc_helper(ftest) & BPF_ANC)
843 anc_found = true;
844 /* Ancillary operation unknown or unsupported */
845 if (anc_found == false && ftest->k >= SKF_AD_OFF)
846 return -EINVAL;
847 }
848 }
849
850 /* Last instruction must be a RET code */
851 switch (filter[flen - 1].code) {
852 case BPF_RET | BPF_K:
853 case BPF_RET | BPF_A:
854 return check_load_and_stores(filter, flen);
855 }
856
857 return -EINVAL;
858}
859
860static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
861 const struct sock_fprog *fprog)
862{
863 unsigned int fsize = bpf_classic_proglen(fprog);
864 struct sock_fprog_kern *fkprog;
865
866 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
867 if (!fp->orig_prog)
868 return -ENOMEM;
869
870 fkprog = fp->orig_prog;
871 fkprog->len = fprog->len;
872
873 fkprog->filter = kmemdup(fp->insns, fsize,
874 GFP_KERNEL | __GFP_NOWARN);
875 if (!fkprog->filter) {
876 kfree(fp->orig_prog);
877 return -ENOMEM;
878 }
879
880 return 0;
881}
882
883static void bpf_release_orig_filter(struct bpf_prog *fp)
884{
885 struct sock_fprog_kern *fprog = fp->orig_prog;
886
887 if (fprog) {
888 kfree(fprog->filter);
889 kfree(fprog);
890 }
891}
892
893static void __bpf_prog_release(struct bpf_prog *prog)
894{
895 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
896 bpf_prog_put(prog);
897 } else {
898 bpf_release_orig_filter(prog);
899 bpf_prog_free(prog);
900 }
901}
902
903static void __sk_filter_release(struct sk_filter *fp)
904{
905 __bpf_prog_release(fp->prog);
906 kfree(fp);
907}
908
909/**
910 * sk_filter_release_rcu - Release a socket filter by rcu_head
911 * @rcu: rcu_head that contains the sk_filter to free
912 */
913static void sk_filter_release_rcu(struct rcu_head *rcu)
914{
915 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
916
917 __sk_filter_release(fp);
918}
919
920/**
921 * sk_filter_release - release a socket filter
922 * @fp: filter to remove
923 *
924 * Remove a filter from a socket and release its resources.
925 */
926static void sk_filter_release(struct sk_filter *fp)
927{
928 if (atomic_dec_and_test(&fp->refcnt))
929 call_rcu(&fp->rcu, sk_filter_release_rcu);
930}
931
932void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
933{
934 u32 filter_size = bpf_prog_size(fp->prog->len);
935
936 atomic_sub(filter_size, &sk->sk_omem_alloc);
937 sk_filter_release(fp);
938}
939
940/* try to charge the socket memory if there is space available
941 * return true on success
942 */
943bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
944{
945 u32 filter_size = bpf_prog_size(fp->prog->len);
946
947 /* same check as in sock_kmalloc() */
948 if (filter_size <= sysctl_optmem_max &&
949 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
950 atomic_inc(&fp->refcnt);
951 atomic_add(filter_size, &sk->sk_omem_alloc);
952 return true;
953 }
954 return false;
955}
956
957static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
958{
959 struct sock_filter *old_prog;
960 struct bpf_prog *old_fp;
961 int err, new_len, old_len = fp->len;
962
963 /* We are free to overwrite insns et al right here as it
964 * won't be used at this point in time anymore internally
965 * after the migration to the internal BPF instruction
966 * representation.
967 */
968 BUILD_BUG_ON(sizeof(struct sock_filter) !=
969 sizeof(struct bpf_insn));
970
971 /* Conversion cannot happen on overlapping memory areas,
972 * so we need to keep the user BPF around until the 2nd
973 * pass. At this time, the user BPF is stored in fp->insns.
974 */
975 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
976 GFP_KERNEL | __GFP_NOWARN);
977 if (!old_prog) {
978 err = -ENOMEM;
979 goto out_err;
980 }
981
982 /* 1st pass: calculate the new program length. */
983 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
984 if (err)
985 goto out_err_free;
986
987 /* Expand fp for appending the new filter representation. */
988 old_fp = fp;
989 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
990 if (!fp) {
991 /* The old_fp is still around in case we couldn't
992 * allocate new memory, so uncharge on that one.
993 */
994 fp = old_fp;
995 err = -ENOMEM;
996 goto out_err_free;
997 }
998
999 fp->len = new_len;
1000
1001 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1002 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
1003 if (err)
1004 /* 2nd bpf_convert_filter() can fail only if it fails
1005 * to allocate memory, remapping must succeed. Note,
1006 * that at this time old_fp has already been released
1007 * by krealloc().
1008 */
1009 goto out_err_free;
1010
1011 /* We are guaranteed to never error here with cBPF to eBPF
1012 * transitions, since there's no issue with type compatibility
1013 * checks on program arrays.
1014 */
1015 fp = bpf_prog_select_runtime(fp, &err);
1016
1017 kfree(old_prog);
1018 return fp;
1019
1020out_err_free:
1021 kfree(old_prog);
1022out_err:
1023 __bpf_prog_release(fp);
1024 return ERR_PTR(err);
1025}
1026
1027static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1028 bpf_aux_classic_check_t trans)
1029{
1030 int err;
1031
1032 fp->bpf_func = NULL;
1033 fp->jited = 0;
1034
1035 err = bpf_check_classic(fp->insns, fp->len);
1036 if (err) {
1037 __bpf_prog_release(fp);
1038 return ERR_PTR(err);
1039 }
1040
1041 /* There might be additional checks and transformations
1042 * needed on classic filters, f.e. in case of seccomp.
1043 */
1044 if (trans) {
1045 err = trans(fp->insns, fp->len);
1046 if (err) {
1047 __bpf_prog_release(fp);
1048 return ERR_PTR(err);
1049 }
1050 }
1051
1052 /* Probe if we can JIT compile the filter and if so, do
1053 * the compilation of the filter.
1054 */
1055 bpf_jit_compile(fp);
1056
1057 /* JIT compiler couldn't process this filter, so do the
1058 * internal BPF translation for the optimized interpreter.
1059 */
1060 if (!fp->jited)
1061 fp = bpf_migrate_filter(fp);
1062
1063 return fp;
1064}
1065
1066/**
1067 * bpf_prog_create - create an unattached filter
1068 * @pfp: the unattached filter that is created
1069 * @fprog: the filter program
1070 *
1071 * Create a filter independent of any socket. We first run some
1072 * sanity checks on it to make sure it does not explode on us later.
1073 * If an error occurs or there is insufficient memory for the filter
1074 * a negative errno code is returned. On success the return is zero.
1075 */
1076int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1077{
1078 unsigned int fsize = bpf_classic_proglen(fprog);
1079 struct bpf_prog *fp;
1080
1081 /* Make sure new filter is there and in the right amounts. */
1082 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1083 return -EINVAL;
1084
1085 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1086 if (!fp)
1087 return -ENOMEM;
1088
1089 memcpy(fp->insns, fprog->filter, fsize);
1090
1091 fp->len = fprog->len;
1092 /* Since unattached filters are not copied back to user
1093 * space through sk_get_filter(), we do not need to hold
1094 * a copy here, and can spare us the work.
1095 */
1096 fp->orig_prog = NULL;
1097
1098 /* bpf_prepare_filter() already takes care of freeing
1099 * memory in case something goes wrong.
1100 */
1101 fp = bpf_prepare_filter(fp, NULL);
1102 if (IS_ERR(fp))
1103 return PTR_ERR(fp);
1104
1105 *pfp = fp;
1106 return 0;
1107}
1108EXPORT_SYMBOL_GPL(bpf_prog_create);
1109
1110/**
1111 * bpf_prog_create_from_user - create an unattached filter from user buffer
1112 * @pfp: the unattached filter that is created
1113 * @fprog: the filter program
1114 * @trans: post-classic verifier transformation handler
1115 * @save_orig: save classic BPF program
1116 *
1117 * This function effectively does the same as bpf_prog_create(), only
1118 * that it builds up its insns buffer from user space provided buffer.
1119 * It also allows for passing a bpf_aux_classic_check_t handler.
1120 */
1121int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1122 bpf_aux_classic_check_t trans, bool save_orig)
1123{
1124 unsigned int fsize = bpf_classic_proglen(fprog);
1125 struct bpf_prog *fp;
1126 int err;
1127
1128 /* Make sure new filter is there and in the right amounts. */
1129 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1130 return -EINVAL;
1131
1132 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1133 if (!fp)
1134 return -ENOMEM;
1135
1136 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1137 __bpf_prog_free(fp);
1138 return -EFAULT;
1139 }
1140
1141 fp->len = fprog->len;
1142 fp->orig_prog = NULL;
1143
1144 if (save_orig) {
1145 err = bpf_prog_store_orig_filter(fp, fprog);
1146 if (err) {
1147 __bpf_prog_free(fp);
1148 return -ENOMEM;
1149 }
1150 }
1151
1152 /* bpf_prepare_filter() already takes care of freeing
1153 * memory in case something goes wrong.
1154 */
1155 fp = bpf_prepare_filter(fp, trans);
1156 if (IS_ERR(fp))
1157 return PTR_ERR(fp);
1158
1159 *pfp = fp;
1160 return 0;
1161}
1162EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1163
1164void bpf_prog_destroy(struct bpf_prog *fp)
1165{
1166 __bpf_prog_release(fp);
1167}
1168EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1169
1170static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1171{
1172 struct sk_filter *fp, *old_fp;
1173
1174 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1175 if (!fp)
1176 return -ENOMEM;
1177
1178 fp->prog = prog;
1179 atomic_set(&fp->refcnt, 0);
1180
1181 if (!sk_filter_charge(sk, fp)) {
1182 kfree(fp);
1183 return -ENOMEM;
1184 }
1185
1186 old_fp = rcu_dereference_protected(sk->sk_filter,
1187 lockdep_sock_is_held(sk));
1188 rcu_assign_pointer(sk->sk_filter, fp);
1189
1190 if (old_fp)
1191 sk_filter_uncharge(sk, old_fp);
1192
1193 return 0;
1194}
1195
1196static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1197{
1198 struct bpf_prog *old_prog;
1199 int err;
1200
1201 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1202 return -ENOMEM;
1203
1204 if (sk_unhashed(sk) && sk->sk_reuseport) {
1205 err = reuseport_alloc(sk);
1206 if (err)
1207 return err;
1208 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1209 /* The socket wasn't bound with SO_REUSEPORT */
1210 return -EINVAL;
1211 }
1212
1213 old_prog = reuseport_attach_prog(sk, prog);
1214 if (old_prog)
1215 bpf_prog_destroy(old_prog);
1216
1217 return 0;
1218}
1219
1220static
1221struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1222{
1223 unsigned int fsize = bpf_classic_proglen(fprog);
1224 struct bpf_prog *prog;
1225 int err;
1226
1227 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1228 return ERR_PTR(-EPERM);
1229
1230 /* Make sure new filter is there and in the right amounts. */
1231 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1232 return ERR_PTR(-EINVAL);
1233
1234 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1235 if (!prog)
1236 return ERR_PTR(-ENOMEM);
1237
1238 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1239 __bpf_prog_free(prog);
1240 return ERR_PTR(-EFAULT);
1241 }
1242
1243 prog->len = fprog->len;
1244
1245 err = bpf_prog_store_orig_filter(prog, fprog);
1246 if (err) {
1247 __bpf_prog_free(prog);
1248 return ERR_PTR(-ENOMEM);
1249 }
1250
1251 /* bpf_prepare_filter() already takes care of freeing
1252 * memory in case something goes wrong.
1253 */
1254 return bpf_prepare_filter(prog, NULL);
1255}
1256
1257/**
1258 * sk_attach_filter - attach a socket filter
1259 * @fprog: the filter program
1260 * @sk: the socket to use
1261 *
1262 * Attach the user's filter code. We first run some sanity checks on
1263 * it to make sure it does not explode on us later. If an error
1264 * occurs or there is insufficient memory for the filter a negative
1265 * errno code is returned. On success the return is zero.
1266 */
1267int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1268{
1269 struct bpf_prog *prog = __get_filter(fprog, sk);
1270 int err;
1271
1272 if (IS_ERR(prog))
1273 return PTR_ERR(prog);
1274
1275 err = __sk_attach_prog(prog, sk);
1276 if (err < 0) {
1277 __bpf_prog_release(prog);
1278 return err;
1279 }
1280
1281 return 0;
1282}
1283EXPORT_SYMBOL_GPL(sk_attach_filter);
1284
1285int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1286{
1287 struct bpf_prog *prog = __get_filter(fprog, sk);
1288 int err;
1289
1290 if (IS_ERR(prog))
1291 return PTR_ERR(prog);
1292
1293 err = __reuseport_attach_prog(prog, sk);
1294 if (err < 0) {
1295 __bpf_prog_release(prog);
1296 return err;
1297 }
1298
1299 return 0;
1300}
1301
1302static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1303{
1304 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1305 return ERR_PTR(-EPERM);
1306
1307 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1308}
1309
1310int sk_attach_bpf(u32 ufd, struct sock *sk)
1311{
1312 struct bpf_prog *prog = __get_bpf(ufd, sk);
1313 int err;
1314
1315 if (IS_ERR(prog))
1316 return PTR_ERR(prog);
1317
1318 err = __sk_attach_prog(prog, sk);
1319 if (err < 0) {
1320 bpf_prog_put(prog);
1321 return err;
1322 }
1323
1324 return 0;
1325}
1326
1327int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1328{
1329 struct bpf_prog *prog = __get_bpf(ufd, sk);
1330 int err;
1331
1332 if (IS_ERR(prog))
1333 return PTR_ERR(prog);
1334
1335 err = __reuseport_attach_prog(prog, sk);
1336 if (err < 0) {
1337 bpf_prog_put(prog);
1338 return err;
1339 }
1340
1341 return 0;
1342}
1343
1344struct bpf_scratchpad {
1345 union {
1346 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1347 u8 buff[MAX_BPF_STACK];
1348 };
1349};
1350
1351static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1352
1353static inline int bpf_try_make_writable(struct sk_buff *skb,
1354 unsigned int write_len)
1355{
1356 int err;
1357
1358 if (!skb_cloned(skb))
1359 return 0;
1360 if (skb_clone_writable(skb, write_len))
1361 return 0;
1362 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1363 if (!err)
1364 bpf_compute_data_end(skb);
1365 return err;
1366}
1367
1368static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1369{
1370 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1371 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1372 int offset = (int) r2;
1373 void *from = (void *) (long) r3;
1374 unsigned int len = (unsigned int) r4;
1375 void *ptr;
1376
1377 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1378 return -EINVAL;
1379
1380 /* bpf verifier guarantees that:
1381 * 'from' pointer points to bpf program stack
1382 * 'len' bytes of it were initialized
1383 * 'len' > 0
1384 * 'skb' is a valid pointer to 'struct sk_buff'
1385 *
1386 * so check for invalid 'offset' and too large 'len'
1387 */
1388 if (unlikely((u32) offset > 0xffff || len > sizeof(sp->buff)))
1389 return -EFAULT;
1390 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1391 return -EFAULT;
1392
1393 ptr = skb_header_pointer(skb, offset, len, sp->buff);
1394 if (unlikely(!ptr))
1395 return -EFAULT;
1396
1397 if (flags & BPF_F_RECOMPUTE_CSUM)
1398 skb_postpull_rcsum(skb, ptr, len);
1399
1400 memcpy(ptr, from, len);
1401
1402 if (ptr == sp->buff)
1403 /* skb_store_bits cannot return -EFAULT here */
1404 skb_store_bits(skb, offset, ptr, len);
1405
1406 if (flags & BPF_F_RECOMPUTE_CSUM)
1407 skb_postpush_rcsum(skb, ptr, len);
1408 if (flags & BPF_F_INVALIDATE_HASH)
1409 skb_clear_hash(skb);
1410
1411 return 0;
1412}
1413
1414static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1415 .func = bpf_skb_store_bytes,
1416 .gpl_only = false,
1417 .ret_type = RET_INTEGER,
1418 .arg1_type = ARG_PTR_TO_CTX,
1419 .arg2_type = ARG_ANYTHING,
1420 .arg3_type = ARG_PTR_TO_STACK,
1421 .arg4_type = ARG_CONST_STACK_SIZE,
1422 .arg5_type = ARG_ANYTHING,
1423};
1424
1425static u64 bpf_skb_load_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1426{
1427 const struct sk_buff *skb = (const struct sk_buff *)(unsigned long) r1;
1428 int offset = (int) r2;
1429 void *to = (void *)(unsigned long) r3;
1430 unsigned int len = (unsigned int) r4;
1431 void *ptr;
1432
1433 if (unlikely((u32) offset > 0xffff))
1434 goto err_clear;
1435
1436 ptr = skb_header_pointer(skb, offset, len, to);
1437 if (unlikely(!ptr))
1438 goto err_clear;
1439 if (ptr != to)
1440 memcpy(to, ptr, len);
1441
1442 return 0;
1443err_clear:
1444 memset(to, 0, len);
1445 return -EFAULT;
1446}
1447
1448static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1449 .func = bpf_skb_load_bytes,
1450 .gpl_only = false,
1451 .ret_type = RET_INTEGER,
1452 .arg1_type = ARG_PTR_TO_CTX,
1453 .arg2_type = ARG_ANYTHING,
1454 .arg3_type = ARG_PTR_TO_RAW_STACK,
1455 .arg4_type = ARG_CONST_STACK_SIZE,
1456};
1457
1458static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1459{
1460 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1461 int offset = (int) r2;
1462 __sum16 sum, *ptr;
1463
1464 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1465 return -EINVAL;
1466 if (unlikely((u32) offset > 0xffff))
1467 return -EFAULT;
1468 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum))))
1469 return -EFAULT;
1470
1471 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1472 if (unlikely(!ptr))
1473 return -EFAULT;
1474
1475 switch (flags & BPF_F_HDR_FIELD_MASK) {
1476 case 0:
1477 if (unlikely(from != 0))
1478 return -EINVAL;
1479
1480 csum_replace_by_diff(ptr, to);
1481 break;
1482 case 2:
1483 csum_replace2(ptr, from, to);
1484 break;
1485 case 4:
1486 csum_replace4(ptr, from, to);
1487 break;
1488 default:
1489 return -EINVAL;
1490 }
1491
1492 if (ptr == &sum)
1493 /* skb_store_bits guaranteed to not return -EFAULT here */
1494 skb_store_bits(skb, offset, ptr, sizeof(sum));
1495
1496 return 0;
1497}
1498
1499static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1500 .func = bpf_l3_csum_replace,
1501 .gpl_only = false,
1502 .ret_type = RET_INTEGER,
1503 .arg1_type = ARG_PTR_TO_CTX,
1504 .arg2_type = ARG_ANYTHING,
1505 .arg3_type = ARG_ANYTHING,
1506 .arg4_type = ARG_ANYTHING,
1507 .arg5_type = ARG_ANYTHING,
1508};
1509
1510static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1511{
1512 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1513 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1514 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1515 int offset = (int) r2;
1516 __sum16 sum, *ptr;
1517
1518 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR |
1519 BPF_F_HDR_FIELD_MASK)))
1520 return -EINVAL;
1521 if (unlikely((u32) offset > 0xffff))
1522 return -EFAULT;
1523 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum))))
1524 return -EFAULT;
1525
1526 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1527 if (unlikely(!ptr))
1528 return -EFAULT;
1529 if (is_mmzero && !*ptr)
1530 return 0;
1531
1532 switch (flags & BPF_F_HDR_FIELD_MASK) {
1533 case 0:
1534 if (unlikely(from != 0))
1535 return -EINVAL;
1536
1537 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1538 break;
1539 case 2:
1540 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1541 break;
1542 case 4:
1543 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1544 break;
1545 default:
1546 return -EINVAL;
1547 }
1548
1549 if (is_mmzero && !*ptr)
1550 *ptr = CSUM_MANGLED_0;
1551 if (ptr == &sum)
1552 /* skb_store_bits guaranteed to not return -EFAULT here */
1553 skb_store_bits(skb, offset, ptr, sizeof(sum));
1554
1555 return 0;
1556}
1557
1558static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1559 .func = bpf_l4_csum_replace,
1560 .gpl_only = false,
1561 .ret_type = RET_INTEGER,
1562 .arg1_type = ARG_PTR_TO_CTX,
1563 .arg2_type = ARG_ANYTHING,
1564 .arg3_type = ARG_ANYTHING,
1565 .arg4_type = ARG_ANYTHING,
1566 .arg5_type = ARG_ANYTHING,
1567};
1568
1569static u64 bpf_csum_diff(u64 r1, u64 from_size, u64 r3, u64 to_size, u64 seed)
1570{
1571 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1572 u64 diff_size = from_size + to_size;
1573 __be32 *from = (__be32 *) (long) r1;
1574 __be32 *to = (__be32 *) (long) r3;
1575 int i, j = 0;
1576
1577 /* This is quite flexible, some examples:
1578 *
1579 * from_size == 0, to_size > 0, seed := csum --> pushing data
1580 * from_size > 0, to_size == 0, seed := csum --> pulling data
1581 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1582 *
1583 * Even for diffing, from_size and to_size don't need to be equal.
1584 */
1585 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1586 diff_size > sizeof(sp->diff)))
1587 return -EINVAL;
1588
1589 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1590 sp->diff[j] = ~from[i];
1591 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1592 sp->diff[j] = to[i];
1593
1594 return csum_partial(sp->diff, diff_size, seed);
1595}
1596
1597static const struct bpf_func_proto bpf_csum_diff_proto = {
1598 .func = bpf_csum_diff,
1599 .gpl_only = false,
1600 .ret_type = RET_INTEGER,
1601 .arg1_type = ARG_PTR_TO_STACK,
1602 .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1603 .arg3_type = ARG_PTR_TO_STACK,
1604 .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1605 .arg5_type = ARG_ANYTHING,
1606};
1607
1608static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1609{
1610 if (skb_at_tc_ingress(skb))
1611 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1612
1613 return dev_forward_skb(dev, skb);
1614}
1615
1616static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1617{
1618 int ret;
1619
1620 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1621 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1622 kfree_skb(skb);
1623 return -ENETDOWN;
1624 }
1625
1626 skb->dev = dev;
1627
1628 __this_cpu_inc(xmit_recursion);
1629 ret = dev_queue_xmit(skb);
1630 __this_cpu_dec(xmit_recursion);
1631
1632 return ret;
1633}
1634
1635static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1636{
1637 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1638 struct net_device *dev;
1639
1640 if (unlikely(flags & ~(BPF_F_INGRESS)))
1641 return -EINVAL;
1642
1643 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1644 if (unlikely(!dev))
1645 return -EINVAL;
1646
1647 skb = skb_clone(skb, GFP_ATOMIC);
1648 if (unlikely(!skb))
1649 return -ENOMEM;
1650
1651 return flags & BPF_F_INGRESS ?
1652 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1653}
1654
1655static const struct bpf_func_proto bpf_clone_redirect_proto = {
1656 .func = bpf_clone_redirect,
1657 .gpl_only = false,
1658 .ret_type = RET_INTEGER,
1659 .arg1_type = ARG_PTR_TO_CTX,
1660 .arg2_type = ARG_ANYTHING,
1661 .arg3_type = ARG_ANYTHING,
1662};
1663
1664struct redirect_info {
1665 u32 ifindex;
1666 u32 flags;
1667};
1668
1669static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1670
1671static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5)
1672{
1673 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1674
1675 if (unlikely(flags & ~(BPF_F_INGRESS)))
1676 return TC_ACT_SHOT;
1677
1678 ri->ifindex = ifindex;
1679 ri->flags = flags;
1680
1681 return TC_ACT_REDIRECT;
1682}
1683
1684int skb_do_redirect(struct sk_buff *skb)
1685{
1686 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1687 struct net_device *dev;
1688
1689 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1690 ri->ifindex = 0;
1691 if (unlikely(!dev)) {
1692 kfree_skb(skb);
1693 return -EINVAL;
1694 }
1695
1696 return ri->flags & BPF_F_INGRESS ?
1697 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1698}
1699
1700static const struct bpf_func_proto bpf_redirect_proto = {
1701 .func = bpf_redirect,
1702 .gpl_only = false,
1703 .ret_type = RET_INTEGER,
1704 .arg1_type = ARG_ANYTHING,
1705 .arg2_type = ARG_ANYTHING,
1706};
1707
1708static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1709{
1710 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1711}
1712
1713static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1714 .func = bpf_get_cgroup_classid,
1715 .gpl_only = false,
1716 .ret_type = RET_INTEGER,
1717 .arg1_type = ARG_PTR_TO_CTX,
1718};
1719
1720static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1721{
1722 return dst_tclassid((struct sk_buff *) (unsigned long) r1);
1723}
1724
1725static const struct bpf_func_proto bpf_get_route_realm_proto = {
1726 .func = bpf_get_route_realm,
1727 .gpl_only = false,
1728 .ret_type = RET_INTEGER,
1729 .arg1_type = ARG_PTR_TO_CTX,
1730};
1731
1732static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1733{
1734 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1735 __be16 vlan_proto = (__force __be16) r2;
1736 int ret;
1737
1738 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1739 vlan_proto != htons(ETH_P_8021AD)))
1740 vlan_proto = htons(ETH_P_8021Q);
1741
1742 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1743 bpf_compute_data_end(skb);
1744 return ret;
1745}
1746
1747const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1748 .func = bpf_skb_vlan_push,
1749 .gpl_only = false,
1750 .ret_type = RET_INTEGER,
1751 .arg1_type = ARG_PTR_TO_CTX,
1752 .arg2_type = ARG_ANYTHING,
1753 .arg3_type = ARG_ANYTHING,
1754};
1755EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1756
1757static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1758{
1759 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1760 int ret;
1761
1762 ret = skb_vlan_pop(skb);
1763 bpf_compute_data_end(skb);
1764 return ret;
1765}
1766
1767const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1768 .func = bpf_skb_vlan_pop,
1769 .gpl_only = false,
1770 .ret_type = RET_INTEGER,
1771 .arg1_type = ARG_PTR_TO_CTX,
1772};
1773EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1774
1775static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
1776{
1777 /* Caller already did skb_cow() with len as headroom,
1778 * so no need to do it here.
1779 */
1780 skb_push(skb, len);
1781 memmove(skb->data, skb->data + len, off);
1782 memset(skb->data + off, 0, len);
1783
1784 /* No skb_postpush_rcsum(skb, skb->data + off, len)
1785 * needed here as it does not change the skb->csum
1786 * result for checksum complete when summing over
1787 * zeroed blocks.
1788 */
1789 return 0;
1790}
1791
1792static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
1793{
1794 /* skb_ensure_writable() is not needed here, as we're
1795 * already working on an uncloned skb.
1796 */
1797 if (unlikely(!pskb_may_pull(skb, off + len)))
1798 return -ENOMEM;
1799
1800 skb_postpull_rcsum(skb, skb->data + off, len);
1801 memmove(skb->data + len, skb->data, off);
1802 __skb_pull(skb, len);
1803
1804 return 0;
1805}
1806
1807static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
1808{
1809 bool trans_same = skb->transport_header == skb->network_header;
1810 int ret;
1811
1812 /* There's no need for __skb_push()/__skb_pull() pair to
1813 * get to the start of the mac header as we're guaranteed
1814 * to always start from here under eBPF.
1815 */
1816 ret = bpf_skb_generic_push(skb, off, len);
1817 if (likely(!ret)) {
1818 skb->mac_header -= len;
1819 skb->network_header -= len;
1820 if (trans_same)
1821 skb->transport_header = skb->network_header;
1822 }
1823
1824 return ret;
1825}
1826
1827static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
1828{
1829 bool trans_same = skb->transport_header == skb->network_header;
1830 int ret;
1831
1832 /* Same here, __skb_push()/__skb_pull() pair not needed. */
1833 ret = bpf_skb_generic_pop(skb, off, len);
1834 if (likely(!ret)) {
1835 skb->mac_header += len;
1836 skb->network_header += len;
1837 if (trans_same)
1838 skb->transport_header = skb->network_header;
1839 }
1840
1841 return ret;
1842}
1843
1844static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
1845{
1846 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1847 u32 off = skb->network_header - skb->mac_header;
1848 int ret;
1849
1850 ret = skb_cow(skb, len_diff);
1851 if (unlikely(ret < 0))
1852 return ret;
1853
1854 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
1855 if (unlikely(ret < 0))
1856 return ret;
1857
1858 if (skb_is_gso(skb)) {
1859 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV4 needs to
1860 * be changed into SKB_GSO_TCPV6.
1861 */
1862 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
1863 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
1864 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
1865 }
1866
1867 /* Due to IPv6 header, MSS needs to be downgraded. */
1868 skb_shinfo(skb)->gso_size -= len_diff;
1869 /* Header must be checked, and gso_segs recomputed. */
1870 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1871 skb_shinfo(skb)->gso_segs = 0;
1872 }
1873
1874 skb->protocol = htons(ETH_P_IPV6);
1875 skb_clear_hash(skb);
1876
1877 return 0;
1878}
1879
1880static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
1881{
1882 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1883 u32 off = skb->network_header - skb->mac_header;
1884 int ret;
1885
1886 ret = skb_unclone(skb, GFP_ATOMIC);
1887 if (unlikely(ret < 0))
1888 return ret;
1889
1890 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
1891 if (unlikely(ret < 0))
1892 return ret;
1893
1894 if (skb_is_gso(skb)) {
1895 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV6 needs to
1896 * be changed into SKB_GSO_TCPV4.
1897 */
1898 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
1899 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
1900 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
1901 }
1902
1903 /* Due to IPv4 header, MSS can be upgraded. */
1904 skb_shinfo(skb)->gso_size += len_diff;
1905 /* Header must be checked, and gso_segs recomputed. */
1906 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1907 skb_shinfo(skb)->gso_segs = 0;
1908 }
1909
1910 skb->protocol = htons(ETH_P_IP);
1911 skb_clear_hash(skb);
1912
1913 return 0;
1914}
1915
1916static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
1917{
1918 __be16 from_proto = skb->protocol;
1919
1920 if (from_proto == htons(ETH_P_IP) &&
1921 to_proto == htons(ETH_P_IPV6))
1922 return bpf_skb_proto_4_to_6(skb);
1923
1924 if (from_proto == htons(ETH_P_IPV6) &&
1925 to_proto == htons(ETH_P_IP))
1926 return bpf_skb_proto_6_to_4(skb);
1927
1928 return -ENOTSUPP;
1929}
1930
1931static u64 bpf_skb_change_proto(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5)
1932{
1933 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1934 __be16 proto = (__force __be16) r2;
1935 int ret;
1936
1937 if (unlikely(flags))
1938 return -EINVAL;
1939
1940 /* General idea is that this helper does the basic groundwork
1941 * needed for changing the protocol, and eBPF program fills the
1942 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
1943 * and other helpers, rather than passing a raw buffer here.
1944 *
1945 * The rationale is to keep this minimal and without a need to
1946 * deal with raw packet data. F.e. even if we would pass buffers
1947 * here, the program still needs to call the bpf_lX_csum_replace()
1948 * helpers anyway. Plus, this way we keep also separation of
1949 * concerns, since f.e. bpf_skb_store_bytes() should only take
1950 * care of stores.
1951 *
1952 * Currently, additional options and extension header space are
1953 * not supported, but flags register is reserved so we can adapt
1954 * that. For offloads, we mark packet as dodgy, so that headers
1955 * need to be verified first.
1956 */
1957 ret = bpf_skb_proto_xlat(skb, proto);
1958 bpf_compute_data_end(skb);
1959 return ret;
1960}
1961
1962static const struct bpf_func_proto bpf_skb_change_proto_proto = {
1963 .func = bpf_skb_change_proto,
1964 .gpl_only = false,
1965 .ret_type = RET_INTEGER,
1966 .arg1_type = ARG_PTR_TO_CTX,
1967 .arg2_type = ARG_ANYTHING,
1968 .arg3_type = ARG_ANYTHING,
1969};
1970
1971static u64 bpf_skb_change_type(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1972{
1973 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1974 u32 pkt_type = r2;
1975
1976 /* We only allow a restricted subset to be changed for now. */
1977 if (unlikely(skb->pkt_type > PACKET_OTHERHOST ||
1978 pkt_type > PACKET_OTHERHOST))
1979 return -EINVAL;
1980
1981 skb->pkt_type = pkt_type;
1982 return 0;
1983}
1984
1985static const struct bpf_func_proto bpf_skb_change_type_proto = {
1986 .func = bpf_skb_change_type,
1987 .gpl_only = false,
1988 .ret_type = RET_INTEGER,
1989 .arg1_type = ARG_PTR_TO_CTX,
1990 .arg2_type = ARG_ANYTHING,
1991};
1992
1993bool bpf_helper_changes_skb_data(void *func)
1994{
1995 if (func == bpf_skb_vlan_push)
1996 return true;
1997 if (func == bpf_skb_vlan_pop)
1998 return true;
1999 if (func == bpf_skb_store_bytes)
2000 return true;
2001 if (func == bpf_skb_change_proto)
2002 return true;
2003 if (func == bpf_l3_csum_replace)
2004 return true;
2005 if (func == bpf_l4_csum_replace)
2006 return true;
2007
2008 return false;
2009}
2010
2011static unsigned short bpf_tunnel_key_af(u64 flags)
2012{
2013 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2014}
2015
2016static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
2017{
2018 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2019 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
2020 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2021 u8 compat[sizeof(struct bpf_tunnel_key)];
2022 void *to_orig = to;
2023 int err;
2024
2025 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2026 err = -EINVAL;
2027 goto err_clear;
2028 }
2029 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2030 err = -EPROTO;
2031 goto err_clear;
2032 }
2033 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2034 err = -EINVAL;
2035 switch (size) {
2036 case offsetof(struct bpf_tunnel_key, tunnel_label):
2037 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2038 goto set_compat;
2039 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2040 /* Fixup deprecated structure layouts here, so we have
2041 * a common path later on.
2042 */
2043 if (ip_tunnel_info_af(info) != AF_INET)
2044 goto err_clear;
2045set_compat:
2046 to = (struct bpf_tunnel_key *)compat;
2047 break;
2048 default:
2049 goto err_clear;
2050 }
2051 }
2052
2053 to->tunnel_id = be64_to_cpu(info->key.tun_id);
2054 to->tunnel_tos = info->key.tos;
2055 to->tunnel_ttl = info->key.ttl;
2056
2057 if (flags & BPF_F_TUNINFO_IPV6) {
2058 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2059 sizeof(to->remote_ipv6));
2060 to->tunnel_label = be32_to_cpu(info->key.label);
2061 } else {
2062 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2063 }
2064
2065 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2066 memcpy(to_orig, to, size);
2067
2068 return 0;
2069err_clear:
2070 memset(to_orig, 0, size);
2071 return err;
2072}
2073
2074static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2075 .func = bpf_skb_get_tunnel_key,
2076 .gpl_only = false,
2077 .ret_type = RET_INTEGER,
2078 .arg1_type = ARG_PTR_TO_CTX,
2079 .arg2_type = ARG_PTR_TO_RAW_STACK,
2080 .arg3_type = ARG_CONST_STACK_SIZE,
2081 .arg4_type = ARG_ANYTHING,
2082};
2083
2084static u64 bpf_skb_get_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
2085{
2086 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2087 u8 *to = (u8 *) (long) r2;
2088 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2089 int err;
2090
2091 if (unlikely(!info ||
2092 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2093 err = -ENOENT;
2094 goto err_clear;
2095 }
2096 if (unlikely(size < info->options_len)) {
2097 err = -ENOMEM;
2098 goto err_clear;
2099 }
2100
2101 ip_tunnel_info_opts_get(to, info);
2102 if (size > info->options_len)
2103 memset(to + info->options_len, 0, size - info->options_len);
2104
2105 return info->options_len;
2106err_clear:
2107 memset(to, 0, size);
2108 return err;
2109}
2110
2111static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2112 .func = bpf_skb_get_tunnel_opt,
2113 .gpl_only = false,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_PTR_TO_CTX,
2116 .arg2_type = ARG_PTR_TO_RAW_STACK,
2117 .arg3_type = ARG_CONST_STACK_SIZE,
2118};
2119
2120static struct metadata_dst __percpu *md_dst;
2121
2122static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
2123{
2124 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2125 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
2126 struct metadata_dst *md = this_cpu_ptr(md_dst);
2127 u8 compat[sizeof(struct bpf_tunnel_key)];
2128 struct ip_tunnel_info *info;
2129
2130 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2131 BPF_F_DONT_FRAGMENT)))
2132 return -EINVAL;
2133 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2134 switch (size) {
2135 case offsetof(struct bpf_tunnel_key, tunnel_label):
2136 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2137 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2138 /* Fixup deprecated structure layouts here, so we have
2139 * a common path later on.
2140 */
2141 memcpy(compat, from, size);
2142 memset(compat + size, 0, sizeof(compat) - size);
2143 from = (struct bpf_tunnel_key *)compat;
2144 break;
2145 default:
2146 return -EINVAL;
2147 }
2148 }
2149 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
2150 from->tunnel_ext))
2151 return -EINVAL;
2152
2153 skb_dst_drop(skb);
2154 dst_hold((struct dst_entry *) md);
2155 skb_dst_set(skb, (struct dst_entry *) md);
2156
2157 info = &md->u.tun_info;
2158 info->mode = IP_TUNNEL_INFO_TX;
2159
2160 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
2161 if (flags & BPF_F_DONT_FRAGMENT)
2162 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
2163
2164 info->key.tun_id = cpu_to_be64(from->tunnel_id);
2165 info->key.tos = from->tunnel_tos;
2166 info->key.ttl = from->tunnel_ttl;
2167
2168 if (flags & BPF_F_TUNINFO_IPV6) {
2169 info->mode |= IP_TUNNEL_INFO_IPV6;
2170 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
2171 sizeof(from->remote_ipv6));
2172 info->key.label = cpu_to_be32(from->tunnel_label) &
2173 IPV6_FLOWLABEL_MASK;
2174 } else {
2175 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
2176 if (flags & BPF_F_ZERO_CSUM_TX)
2177 info->key.tun_flags &= ~TUNNEL_CSUM;
2178 }
2179
2180 return 0;
2181}
2182
2183static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
2184 .func = bpf_skb_set_tunnel_key,
2185 .gpl_only = false,
2186 .ret_type = RET_INTEGER,
2187 .arg1_type = ARG_PTR_TO_CTX,
2188 .arg2_type = ARG_PTR_TO_STACK,
2189 .arg3_type = ARG_CONST_STACK_SIZE,
2190 .arg4_type = ARG_ANYTHING,
2191};
2192
2193static u64 bpf_skb_set_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
2194{
2195 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2196 u8 *from = (u8 *) (long) r2;
2197 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2198 const struct metadata_dst *md = this_cpu_ptr(md_dst);
2199
2200 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
2201 return -EINVAL;
2202 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
2203 return -ENOMEM;
2204
2205 ip_tunnel_info_opts_set(info, from, size);
2206
2207 return 0;
2208}
2209
2210static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
2211 .func = bpf_skb_set_tunnel_opt,
2212 .gpl_only = false,
2213 .ret_type = RET_INTEGER,
2214 .arg1_type = ARG_PTR_TO_CTX,
2215 .arg2_type = ARG_PTR_TO_STACK,
2216 .arg3_type = ARG_CONST_STACK_SIZE,
2217};
2218
2219static const struct bpf_func_proto *
2220bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
2221{
2222 if (!md_dst) {
2223 /* Race is not possible, since it's called from verifier
2224 * that is holding verifier mutex.
2225 */
2226 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
2227 GFP_KERNEL);
2228 if (!md_dst)
2229 return NULL;
2230 }
2231
2232 switch (which) {
2233 case BPF_FUNC_skb_set_tunnel_key:
2234 return &bpf_skb_set_tunnel_key_proto;
2235 case BPF_FUNC_skb_set_tunnel_opt:
2236 return &bpf_skb_set_tunnel_opt_proto;
2237 default:
2238 return NULL;
2239 }
2240}
2241
2242#ifdef CONFIG_SOCK_CGROUP_DATA
2243static u64 bpf_skb_in_cgroup(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
2244{
2245 struct sk_buff *skb = (struct sk_buff *)(long)r1;
2246 struct bpf_map *map = (struct bpf_map *)(long)r2;
2247 struct bpf_array *array = container_of(map, struct bpf_array, map);
2248 struct cgroup *cgrp;
2249 struct sock *sk;
2250 u32 i = (u32)r3;
2251
2252 sk = skb->sk;
2253 if (!sk || !sk_fullsock(sk))
2254 return -ENOENT;
2255
2256 if (unlikely(i >= array->map.max_entries))
2257 return -E2BIG;
2258
2259 cgrp = READ_ONCE(array->ptrs[i]);
2260 if (unlikely(!cgrp))
2261 return -EAGAIN;
2262
2263 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), cgrp);
2264}
2265
2266static const struct bpf_func_proto bpf_skb_in_cgroup_proto = {
2267 .func = bpf_skb_in_cgroup,
2268 .gpl_only = false,
2269 .ret_type = RET_INTEGER,
2270 .arg1_type = ARG_PTR_TO_CTX,
2271 .arg2_type = ARG_CONST_MAP_PTR,
2272 .arg3_type = ARG_ANYTHING,
2273};
2274#endif
2275
2276static const struct bpf_func_proto *
2277sk_filter_func_proto(enum bpf_func_id func_id)
2278{
2279 switch (func_id) {
2280 case BPF_FUNC_map_lookup_elem:
2281 return &bpf_map_lookup_elem_proto;
2282 case BPF_FUNC_map_update_elem:
2283 return &bpf_map_update_elem_proto;
2284 case BPF_FUNC_map_delete_elem:
2285 return &bpf_map_delete_elem_proto;
2286 case BPF_FUNC_get_prandom_u32:
2287 return &bpf_get_prandom_u32_proto;
2288 case BPF_FUNC_get_smp_processor_id:
2289 return &bpf_get_raw_smp_processor_id_proto;
2290 case BPF_FUNC_tail_call:
2291 return &bpf_tail_call_proto;
2292 case BPF_FUNC_ktime_get_ns:
2293 return &bpf_ktime_get_ns_proto;
2294 case BPF_FUNC_trace_printk:
2295 if (capable(CAP_SYS_ADMIN))
2296 return bpf_get_trace_printk_proto();
2297 default:
2298 return NULL;
2299 }
2300}
2301
2302static const struct bpf_func_proto *
2303tc_cls_act_func_proto(enum bpf_func_id func_id)
2304{
2305 switch (func_id) {
2306 case BPF_FUNC_skb_store_bytes:
2307 return &bpf_skb_store_bytes_proto;
2308 case BPF_FUNC_skb_load_bytes:
2309 return &bpf_skb_load_bytes_proto;
2310 case BPF_FUNC_csum_diff:
2311 return &bpf_csum_diff_proto;
2312 case BPF_FUNC_l3_csum_replace:
2313 return &bpf_l3_csum_replace_proto;
2314 case BPF_FUNC_l4_csum_replace:
2315 return &bpf_l4_csum_replace_proto;
2316 case BPF_FUNC_clone_redirect:
2317 return &bpf_clone_redirect_proto;
2318 case BPF_FUNC_get_cgroup_classid:
2319 return &bpf_get_cgroup_classid_proto;
2320 case BPF_FUNC_skb_vlan_push:
2321 return &bpf_skb_vlan_push_proto;
2322 case BPF_FUNC_skb_vlan_pop:
2323 return &bpf_skb_vlan_pop_proto;
2324 case BPF_FUNC_skb_change_proto:
2325 return &bpf_skb_change_proto_proto;
2326 case BPF_FUNC_skb_change_type:
2327 return &bpf_skb_change_type_proto;
2328 case BPF_FUNC_skb_get_tunnel_key:
2329 return &bpf_skb_get_tunnel_key_proto;
2330 case BPF_FUNC_skb_set_tunnel_key:
2331 return bpf_get_skb_set_tunnel_proto(func_id);
2332 case BPF_FUNC_skb_get_tunnel_opt:
2333 return &bpf_skb_get_tunnel_opt_proto;
2334 case BPF_FUNC_skb_set_tunnel_opt:
2335 return bpf_get_skb_set_tunnel_proto(func_id);
2336 case BPF_FUNC_redirect:
2337 return &bpf_redirect_proto;
2338 case BPF_FUNC_get_route_realm:
2339 return &bpf_get_route_realm_proto;
2340 case BPF_FUNC_perf_event_output:
2341 return bpf_get_event_output_proto();
2342 case BPF_FUNC_get_smp_processor_id:
2343 return &bpf_get_smp_processor_id_proto;
2344#ifdef CONFIG_SOCK_CGROUP_DATA
2345 case BPF_FUNC_skb_in_cgroup:
2346 return &bpf_skb_in_cgroup_proto;
2347#endif
2348 default:
2349 return sk_filter_func_proto(func_id);
2350 }
2351}
2352
2353static bool __is_valid_access(int off, int size, enum bpf_access_type type)
2354{
2355 if (off < 0 || off >= sizeof(struct __sk_buff))
2356 return false;
2357 /* The verifier guarantees that size > 0. */
2358 if (off % size != 0)
2359 return false;
2360 if (size != sizeof(__u32))
2361 return false;
2362
2363 return true;
2364}
2365
2366static bool sk_filter_is_valid_access(int off, int size,
2367 enum bpf_access_type type,
2368 enum bpf_reg_type *reg_type)
2369{
2370 switch (off) {
2371 case offsetof(struct __sk_buff, tc_classid):
2372 case offsetof(struct __sk_buff, data):
2373 case offsetof(struct __sk_buff, data_end):
2374 return false;
2375 }
2376
2377 if (type == BPF_WRITE) {
2378 switch (off) {
2379 case offsetof(struct __sk_buff, cb[0]) ...
2380 offsetof(struct __sk_buff, cb[4]):
2381 break;
2382 default:
2383 return false;
2384 }
2385 }
2386
2387 return __is_valid_access(off, size, type);
2388}
2389
2390static bool tc_cls_act_is_valid_access(int off, int size,
2391 enum bpf_access_type type,
2392 enum bpf_reg_type *reg_type)
2393{
2394 if (type == BPF_WRITE) {
2395 switch (off) {
2396 case offsetof(struct __sk_buff, mark):
2397 case offsetof(struct __sk_buff, tc_index):
2398 case offsetof(struct __sk_buff, priority):
2399 case offsetof(struct __sk_buff, cb[0]) ...
2400 offsetof(struct __sk_buff, cb[4]):
2401 case offsetof(struct __sk_buff, tc_classid):
2402 break;
2403 default:
2404 return false;
2405 }
2406 }
2407
2408 switch (off) {
2409 case offsetof(struct __sk_buff, data):
2410 *reg_type = PTR_TO_PACKET;
2411 break;
2412 case offsetof(struct __sk_buff, data_end):
2413 *reg_type = PTR_TO_PACKET_END;
2414 break;
2415 }
2416
2417 return __is_valid_access(off, size, type);
2418}
2419
2420static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2421 int src_reg, int ctx_off,
2422 struct bpf_insn *insn_buf,
2423 struct bpf_prog *prog)
2424{
2425 struct bpf_insn *insn = insn_buf;
2426
2427 switch (ctx_off) {
2428 case offsetof(struct __sk_buff, len):
2429 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
2430
2431 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2432 offsetof(struct sk_buff, len));
2433 break;
2434
2435 case offsetof(struct __sk_buff, protocol):
2436 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
2437
2438 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2439 offsetof(struct sk_buff, protocol));
2440 break;
2441
2442 case offsetof(struct __sk_buff, vlan_proto):
2443 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
2444
2445 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2446 offsetof(struct sk_buff, vlan_proto));
2447 break;
2448
2449 case offsetof(struct __sk_buff, priority):
2450 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
2451
2452 if (type == BPF_WRITE)
2453 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2454 offsetof(struct sk_buff, priority));
2455 else
2456 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2457 offsetof(struct sk_buff, priority));
2458 break;
2459
2460 case offsetof(struct __sk_buff, ingress_ifindex):
2461 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
2462
2463 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2464 offsetof(struct sk_buff, skb_iif));
2465 break;
2466
2467 case offsetof(struct __sk_buff, ifindex):
2468 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
2469
2470 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
2471 dst_reg, src_reg,
2472 offsetof(struct sk_buff, dev));
2473 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
2474 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
2475 offsetof(struct net_device, ifindex));
2476 break;
2477
2478 case offsetof(struct __sk_buff, hash):
2479 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
2480
2481 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2482 offsetof(struct sk_buff, hash));
2483 break;
2484
2485 case offsetof(struct __sk_buff, mark):
2486 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
2487
2488 if (type == BPF_WRITE)
2489 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2490 offsetof(struct sk_buff, mark));
2491 else
2492 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2493 offsetof(struct sk_buff, mark));
2494 break;
2495
2496 case offsetof(struct __sk_buff, pkt_type):
2497 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
2498
2499 case offsetof(struct __sk_buff, queue_mapping):
2500 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
2501
2502 case offsetof(struct __sk_buff, vlan_present):
2503 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
2504 dst_reg, src_reg, insn);
2505
2506 case offsetof(struct __sk_buff, vlan_tci):
2507 return convert_skb_access(SKF_AD_VLAN_TAG,
2508 dst_reg, src_reg, insn);
2509
2510 case offsetof(struct __sk_buff, cb[0]) ...
2511 offsetof(struct __sk_buff, cb[4]):
2512 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
2513
2514 prog->cb_access = 1;
2515 ctx_off -= offsetof(struct __sk_buff, cb[0]);
2516 ctx_off += offsetof(struct sk_buff, cb);
2517 ctx_off += offsetof(struct qdisc_skb_cb, data);
2518 if (type == BPF_WRITE)
2519 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2520 else
2521 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2522 break;
2523
2524 case offsetof(struct __sk_buff, tc_classid):
2525 ctx_off -= offsetof(struct __sk_buff, tc_classid);
2526 ctx_off += offsetof(struct sk_buff, cb);
2527 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
2528 if (type == BPF_WRITE)
2529 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2530 else
2531 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2532 break;
2533
2534 case offsetof(struct __sk_buff, data):
2535 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, data)),
2536 dst_reg, src_reg,
2537 offsetof(struct sk_buff, data));
2538 break;
2539
2540 case offsetof(struct __sk_buff, data_end):
2541 ctx_off -= offsetof(struct __sk_buff, data_end);
2542 ctx_off += offsetof(struct sk_buff, cb);
2543 ctx_off += offsetof(struct bpf_skb_data_end, data_end);
2544 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(sizeof(void *)),
2545 dst_reg, src_reg, ctx_off);
2546 break;
2547
2548 case offsetof(struct __sk_buff, tc_index):
2549#ifdef CONFIG_NET_SCHED
2550 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
2551
2552 if (type == BPF_WRITE)
2553 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
2554 offsetof(struct sk_buff, tc_index));
2555 else
2556 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2557 offsetof(struct sk_buff, tc_index));
2558 break;
2559#else
2560 if (type == BPF_WRITE)
2561 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
2562 else
2563 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
2564 break;
2565#endif
2566 }
2567
2568 return insn - insn_buf;
2569}
2570
2571static const struct bpf_verifier_ops sk_filter_ops = {
2572 .get_func_proto = sk_filter_func_proto,
2573 .is_valid_access = sk_filter_is_valid_access,
2574 .convert_ctx_access = bpf_net_convert_ctx_access,
2575};
2576
2577static const struct bpf_verifier_ops tc_cls_act_ops = {
2578 .get_func_proto = tc_cls_act_func_proto,
2579 .is_valid_access = tc_cls_act_is_valid_access,
2580 .convert_ctx_access = bpf_net_convert_ctx_access,
2581};
2582
2583static struct bpf_prog_type_list sk_filter_type __read_mostly = {
2584 .ops = &sk_filter_ops,
2585 .type = BPF_PROG_TYPE_SOCKET_FILTER,
2586};
2587
2588static struct bpf_prog_type_list sched_cls_type __read_mostly = {
2589 .ops = &tc_cls_act_ops,
2590 .type = BPF_PROG_TYPE_SCHED_CLS,
2591};
2592
2593static struct bpf_prog_type_list sched_act_type __read_mostly = {
2594 .ops = &tc_cls_act_ops,
2595 .type = BPF_PROG_TYPE_SCHED_ACT,
2596};
2597
2598static int __init register_sk_filter_ops(void)
2599{
2600 bpf_register_prog_type(&sk_filter_type);
2601 bpf_register_prog_type(&sched_cls_type);
2602 bpf_register_prog_type(&sched_act_type);
2603
2604 return 0;
2605}
2606late_initcall(register_sk_filter_ops);
2607
2608int sk_detach_filter(struct sock *sk)
2609{
2610 int ret = -ENOENT;
2611 struct sk_filter *filter;
2612
2613 if (sock_flag(sk, SOCK_FILTER_LOCKED))
2614 return -EPERM;
2615
2616 filter = rcu_dereference_protected(sk->sk_filter,
2617 lockdep_sock_is_held(sk));
2618 if (filter) {
2619 RCU_INIT_POINTER(sk->sk_filter, NULL);
2620 sk_filter_uncharge(sk, filter);
2621 ret = 0;
2622 }
2623
2624 return ret;
2625}
2626EXPORT_SYMBOL_GPL(sk_detach_filter);
2627
2628int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
2629 unsigned int len)
2630{
2631 struct sock_fprog_kern *fprog;
2632 struct sk_filter *filter;
2633 int ret = 0;
2634
2635 lock_sock(sk);
2636 filter = rcu_dereference_protected(sk->sk_filter,
2637 lockdep_sock_is_held(sk));
2638 if (!filter)
2639 goto out;
2640
2641 /* We're copying the filter that has been originally attached,
2642 * so no conversion/decode needed anymore. eBPF programs that
2643 * have no original program cannot be dumped through this.
2644 */
2645 ret = -EACCES;
2646 fprog = filter->prog->orig_prog;
2647 if (!fprog)
2648 goto out;
2649
2650 ret = fprog->len;
2651 if (!len)
2652 /* User space only enquires number of filter blocks. */
2653 goto out;
2654
2655 ret = -EINVAL;
2656 if (len < fprog->len)
2657 goto out;
2658
2659 ret = -EFAULT;
2660 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
2661 goto out;
2662
2663 /* Instead of bytes, the API requests to return the number
2664 * of filter blocks.
2665 */
2666 ret = fprog->len;
2667out:
2668 release_sock(sk);
2669 return ret;
2670}
Ubuntu Artful kernel mirror