projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'lightnvm/for-next'
[deliverable/linux.git] / net / core / filter.c
1 /*
2 * Linux Socket Filter - Kernel level socket filtering
3 *
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
6 *
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
8 *
9 * Authors:
10 *
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 *
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
22 */
23
24 #include <linux/module.h>
25 #include <linux/types.h>
26 #include <linux/mm.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/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_trim_cap - run a packet through a socket filter
57 * @sk: sock associated with &sk_buff
58 * @skb: buffer to filter
59 * @cap: limit on how short the eBPF program may trim the packet
60 *
61 * Run the eBPF program and then cut skb->data to correct size returned by
62 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
63 * than pkt_len we keep whole skb->data. This is the socket level
64 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
65 * be accepted or -EPERM if the packet should be tossed.
66 *
67 */
68 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
69 {
70 int err;
71 struct sk_filter *filter;
72
73 /*
74 * If the skb was allocated from pfmemalloc reserves, only
75 * allow SOCK_MEMALLOC sockets to use it as this socket is
76 * helping free memory
77 */
78 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
79 return -ENOMEM;
80
81 err = security_sock_rcv_skb(sk, skb);
82 if (err)
83 return err;
84
85 rcu_read_lock();
86 filter = rcu_dereference(sk->sk_filter);
87 if (filter) {
88 unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
89 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
90 }
91 rcu_read_unlock();
92
93 return err;
94 }
95 EXPORT_SYMBOL(sk_filter_trim_cap);
96
97 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb)
98 {
99 return skb_get_poff(skb);
100 }
101
102 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
103 {
104 struct nlattr *nla;
105
106 if (skb_is_nonlinear(skb))
107 return 0;
108
109 if (skb->len < sizeof(struct nlattr))
110 return 0;
111
112 if (a > skb->len - sizeof(struct nlattr))
113 return 0;
114
115 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
116 if (nla)
117 return (void *) nla - (void *) skb->data;
118
119 return 0;
120 }
121
122 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
123 {
124 struct nlattr *nla;
125
126 if (skb_is_nonlinear(skb))
127 return 0;
128
129 if (skb->len < sizeof(struct nlattr))
130 return 0;
131
132 if (a > skb->len - sizeof(struct nlattr))
133 return 0;
134
135 nla = (struct nlattr *) &skb->data[a];
136 if (nla->nla_len > skb->len - a)
137 return 0;
138
139 nla = nla_find_nested(nla, x);
140 if (nla)
141 return (void *) nla - (void *) skb->data;
142
143 return 0;
144 }
145
146 BPF_CALL_0(__get_raw_cpu_id)
147 {
148 return raw_smp_processor_id();
149 }
150
151 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
152 .func = __get_raw_cpu_id,
153 .gpl_only = false,
154 .ret_type = RET_INTEGER,
155 };
156
157 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
158 struct bpf_insn *insn_buf)
159 {
160 struct bpf_insn *insn = insn_buf;
161
162 switch (skb_field) {
163 case SKF_AD_MARK:
164 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
165
166 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
167 offsetof(struct sk_buff, mark));
168 break;
169
170 case SKF_AD_PKTTYPE:
171 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
172 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
173 #ifdef __BIG_ENDIAN_BITFIELD
174 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
175 #endif
176 break;
177
178 case SKF_AD_QUEUE:
179 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
180
181 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
182 offsetof(struct sk_buff, queue_mapping));
183 break;
184
185 case SKF_AD_VLAN_TAG:
186 case SKF_AD_VLAN_TAG_PRESENT:
187 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
188 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
189
190 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
191 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
192 offsetof(struct sk_buff, vlan_tci));
193 if (skb_field == SKF_AD_VLAN_TAG) {
194 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
195 ~VLAN_TAG_PRESENT);
196 } else {
197 /* dst_reg >>= 12 */
198 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
199 /* dst_reg &= 1 */
200 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
201 }
202 break;
203 }
204
205 return insn - insn_buf;
206 }
207
208 static bool convert_bpf_extensions(struct sock_filter *fp,
209 struct bpf_insn **insnp)
210 {
211 struct bpf_insn *insn = *insnp;
212 u32 cnt;
213
214 switch (fp->k) {
215 case SKF_AD_OFF + SKF_AD_PROTOCOL:
216 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
217
218 /* A = *(u16 *) (CTX + offsetof(protocol)) */
219 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
220 offsetof(struct sk_buff, protocol));
221 /* A = ntohs(A) [emitting a nop or swap16] */
222 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
223 break;
224
225 case SKF_AD_OFF + SKF_AD_PKTTYPE:
226 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
227 insn += cnt - 1;
228 break;
229
230 case SKF_AD_OFF + SKF_AD_IFINDEX:
231 case SKF_AD_OFF + SKF_AD_HATYPE:
232 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
233 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
234
235 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
236 BPF_REG_TMP, BPF_REG_CTX,
237 offsetof(struct sk_buff, dev));
238 /* if (tmp != 0) goto pc + 1 */
239 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
240 *insn++ = BPF_EXIT_INSN();
241 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
242 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
243 offsetof(struct net_device, ifindex));
244 else
245 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
246 offsetof(struct net_device, type));
247 break;
248
249 case SKF_AD_OFF + SKF_AD_MARK:
250 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
251 insn += cnt - 1;
252 break;
253
254 case SKF_AD_OFF + SKF_AD_RXHASH:
255 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
256
257 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
258 offsetof(struct sk_buff, hash));
259 break;
260
261 case SKF_AD_OFF + SKF_AD_QUEUE:
262 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
263 insn += cnt - 1;
264 break;
265
266 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
267 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
268 BPF_REG_A, BPF_REG_CTX, insn);
269 insn += cnt - 1;
270 break;
271
272 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
273 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
274 BPF_REG_A, BPF_REG_CTX, insn);
275 insn += cnt - 1;
276 break;
277
278 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
279 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
280
281 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
282 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
283 offsetof(struct sk_buff, vlan_proto));
284 /* A = ntohs(A) [emitting a nop or swap16] */
285 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
286 break;
287
288 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
289 case SKF_AD_OFF + SKF_AD_NLATTR:
290 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
291 case SKF_AD_OFF + SKF_AD_CPU:
292 case SKF_AD_OFF + SKF_AD_RANDOM:
293 /* arg1 = CTX */
294 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
295 /* arg2 = A */
296 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
297 /* arg3 = X */
298 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
299 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
300 switch (fp->k) {
301 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
302 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
303 break;
304 case SKF_AD_OFF + SKF_AD_NLATTR:
305 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
306 break;
307 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
308 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
309 break;
310 case SKF_AD_OFF + SKF_AD_CPU:
311 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
312 break;
313 case SKF_AD_OFF + SKF_AD_RANDOM:
314 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
315 bpf_user_rnd_init_once();
316 break;
317 }
318 break;
319
320 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
321 /* A ^= X */
322 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
323 break;
324
325 default:
326 /* This is just a dummy call to avoid letting the compiler
327 * evict __bpf_call_base() as an optimization. Placed here
328 * where no-one bothers.
329 */
330 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
331 return false;
332 }
333
334 *insnp = insn;
335 return true;
336 }
337
338 /**
339 * bpf_convert_filter - convert filter program
340 * @prog: the user passed filter program
341 * @len: the length of the user passed filter program
342 * @new_prog: buffer where converted program will be stored
343 * @new_len: pointer to store length of converted program
344 *
345 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
346 * Conversion workflow:
347 *
348 * 1) First pass for calculating the new program length:
349 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
350 *
351 * 2) 2nd pass to remap in two passes: 1st pass finds new
352 * jump offsets, 2nd pass remapping:
353 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
354 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
355 */
356 static int bpf_convert_filter(struct sock_filter *prog, int len,
357 struct bpf_insn *new_prog, int *new_len)
358 {
359 int new_flen = 0, pass = 0, target, i;
360 struct bpf_insn *new_insn;
361 struct sock_filter *fp;
362 int *addrs = NULL;
363 u8 bpf_src;
364
365 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
366 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
367
368 if (len <= 0 || len > BPF_MAXINSNS)
369 return -EINVAL;
370
371 if (new_prog) {
372 addrs = kcalloc(len, sizeof(*addrs),
373 GFP_KERNEL | __GFP_NOWARN);
374 if (!addrs)
375 return -ENOMEM;
376 }
377
378 do_pass:
379 new_insn = new_prog;
380 fp = prog;
381
382 /* Classic BPF related prologue emission. */
383 if (new_insn) {
384 /* Classic BPF expects A and X to be reset first. These need
385 * to be guaranteed to be the first two instructions.
386 */
387 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
388 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
389
390 /* All programs must keep CTX in callee saved BPF_REG_CTX.
391 * In eBPF case it's done by the compiler, here we need to
392 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
393 */
394 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
395 } else {
396 new_insn += 3;
397 }
398
399 for (i = 0; i < len; fp++, i++) {
400 struct bpf_insn tmp_insns[6] = { };
401 struct bpf_insn *insn = tmp_insns;
402
403 if (addrs)
404 addrs[i] = new_insn - new_prog;
405
406 switch (fp->code) {
407 /* All arithmetic insns and skb loads map as-is. */
408 case BPF_ALU | BPF_ADD | BPF_X:
409 case BPF_ALU | BPF_ADD | BPF_K:
410 case BPF_ALU | BPF_SUB | BPF_X:
411 case BPF_ALU | BPF_SUB | BPF_K:
412 case BPF_ALU | BPF_AND | BPF_X:
413 case BPF_ALU | BPF_AND | BPF_K:
414 case BPF_ALU | BPF_OR | BPF_X:
415 case BPF_ALU | BPF_OR | BPF_K:
416 case BPF_ALU | BPF_LSH | BPF_X:
417 case BPF_ALU | BPF_LSH | BPF_K:
418 case BPF_ALU | BPF_RSH | BPF_X:
419 case BPF_ALU | BPF_RSH | BPF_K:
420 case BPF_ALU | BPF_XOR | BPF_X:
421 case BPF_ALU | BPF_XOR | BPF_K:
422 case BPF_ALU | BPF_MUL | BPF_X:
423 case BPF_ALU | BPF_MUL | BPF_K:
424 case BPF_ALU | BPF_DIV | BPF_X:
425 case BPF_ALU | BPF_DIV | BPF_K:
426 case BPF_ALU | BPF_MOD | BPF_X:
427 case BPF_ALU | BPF_MOD | BPF_K:
428 case BPF_ALU | BPF_NEG:
429 case BPF_LD | BPF_ABS | BPF_W:
430 case BPF_LD | BPF_ABS | BPF_H:
431 case BPF_LD | BPF_ABS | BPF_B:
432 case BPF_LD | BPF_IND | BPF_W:
433 case BPF_LD | BPF_IND | BPF_H:
434 case BPF_LD | BPF_IND | BPF_B:
435 /* Check for overloaded BPF extension and
436 * directly convert it if found, otherwise
437 * just move on with mapping.
438 */
439 if (BPF_CLASS(fp->code) == BPF_LD &&
440 BPF_MODE(fp->code) == BPF_ABS &&
441 convert_bpf_extensions(fp, &insn))
442 break;
443
444 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
445 break;
446
447 /* Jump transformation cannot use BPF block macros
448 * everywhere as offset calculation and target updates
449 * require a bit more work than the rest, i.e. jump
450 * opcodes map as-is, but offsets need adjustment.
451 */
452
453 #define BPF_EMIT_JMP \
454 do { \
455 if (target >= len || target < 0) \
456 goto err; \
457 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
458 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
459 insn->off -= insn - tmp_insns; \
460 } while (0)
461
462 case BPF_JMP | BPF_JA:
463 target = i + fp->k + 1;
464 insn->code = fp->code;
465 BPF_EMIT_JMP;
466 break;
467
468 case BPF_JMP | BPF_JEQ | BPF_K:
469 case BPF_JMP | BPF_JEQ | BPF_X:
470 case BPF_JMP | BPF_JSET | BPF_K:
471 case BPF_JMP | BPF_JSET | BPF_X:
472 case BPF_JMP | BPF_JGT | BPF_K:
473 case BPF_JMP | BPF_JGT | BPF_X:
474 case BPF_JMP | BPF_JGE | BPF_K:
475 case BPF_JMP | BPF_JGE | BPF_X:
476 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
477 /* BPF immediates are signed, zero extend
478 * immediate into tmp register and use it
479 * in compare insn.
480 */
481 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
482
483 insn->dst_reg = BPF_REG_A;
484 insn->src_reg = BPF_REG_TMP;
485 bpf_src = BPF_X;
486 } else {
487 insn->dst_reg = BPF_REG_A;
488 insn->imm = fp->k;
489 bpf_src = BPF_SRC(fp->code);
490 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
491 }
492
493 /* Common case where 'jump_false' is next insn. */
494 if (fp->jf == 0) {
495 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
496 target = i + fp->jt + 1;
497 BPF_EMIT_JMP;
498 break;
499 }
500
501 /* Convert JEQ into JNE when 'jump_true' is next insn. */
502 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
503 insn->code = BPF_JMP | BPF_JNE | bpf_src;
504 target = i + fp->jf + 1;
505 BPF_EMIT_JMP;
506 break;
507 }
508
509 /* Other jumps are mapped into two insns: Jxx and JA. */
510 target = i + fp->jt + 1;
511 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
512 BPF_EMIT_JMP;
513 insn++;
514
515 insn->code = BPF_JMP | BPF_JA;
516 target = i + fp->jf + 1;
517 BPF_EMIT_JMP;
518 break;
519
520 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
521 case BPF_LDX | BPF_MSH | BPF_B:
522 /* tmp = A */
523 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
524 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
525 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
526 /* A &= 0xf */
527 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
528 /* A <<= 2 */
529 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
530 /* X = A */
531 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
532 /* A = tmp */
533 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
534 break;
535
536 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
537 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
538 */
539 case BPF_RET | BPF_A:
540 case BPF_RET | BPF_K:
541 if (BPF_RVAL(fp->code) == BPF_K)
542 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
543 0, fp->k);
544 *insn = BPF_EXIT_INSN();
545 break;
546
547 /* Store to stack. */
548 case BPF_ST:
549 case BPF_STX:
550 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
551 BPF_ST ? BPF_REG_A : BPF_REG_X,
552 -(BPF_MEMWORDS - fp->k) * 4);
553 break;
554
555 /* Load from stack. */
556 case BPF_LD | BPF_MEM:
557 case BPF_LDX | BPF_MEM:
558 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
559 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
560 -(BPF_MEMWORDS - fp->k) * 4);
561 break;
562
563 /* A = K or X = K */
564 case BPF_LD | BPF_IMM:
565 case BPF_LDX | BPF_IMM:
566 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
567 BPF_REG_A : BPF_REG_X, fp->k);
568 break;
569
570 /* X = A */
571 case BPF_MISC | BPF_TAX:
572 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
573 break;
574
575 /* A = X */
576 case BPF_MISC | BPF_TXA:
577 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
578 break;
579
580 /* A = skb->len or X = skb->len */
581 case BPF_LD | BPF_W | BPF_LEN:
582 case BPF_LDX | BPF_W | BPF_LEN:
583 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
584 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
585 offsetof(struct sk_buff, len));
586 break;
587
588 /* Access seccomp_data fields. */
589 case BPF_LDX | BPF_ABS | BPF_W:
590 /* A = *(u32 *) (ctx + K) */
591 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
592 break;
593
594 /* Unknown instruction. */
595 default:
596 goto err;
597 }
598
599 insn++;
600 if (new_prog)
601 memcpy(new_insn, tmp_insns,
602 sizeof(*insn) * (insn - tmp_insns));
603 new_insn += insn - tmp_insns;
604 }
605
606 if (!new_prog) {
607 /* Only calculating new length. */
608 *new_len = new_insn - new_prog;
609 return 0;
610 }
611
612 pass++;
613 if (new_flen != new_insn - new_prog) {
614 new_flen = new_insn - new_prog;
615 if (pass > 2)
616 goto err;
617 goto do_pass;
618 }
619
620 kfree(addrs);
621 BUG_ON(*new_len != new_flen);
622 return 0;
623 err:
624 kfree(addrs);
625 return -EINVAL;
626 }
627
628 /* Security:
629 *
630 * As we dont want to clear mem[] array for each packet going through
631 * __bpf_prog_run(), we check that filter loaded by user never try to read
632 * a cell if not previously written, and we check all branches to be sure
633 * a malicious user doesn't try to abuse us.
634 */
635 static int check_load_and_stores(const struct sock_filter *filter, int flen)
636 {
637 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
638 int pc, ret = 0;
639
640 BUILD_BUG_ON(BPF_MEMWORDS > 16);
641
642 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
643 if (!masks)
644 return -ENOMEM;
645
646 memset(masks, 0xff, flen * sizeof(*masks));
647
648 for (pc = 0; pc < flen; pc++) {
649 memvalid &= masks[pc];
650
651 switch (filter[pc].code) {
652 case BPF_ST:
653 case BPF_STX:
654 memvalid |= (1 << filter[pc].k);
655 break;
656 case BPF_LD | BPF_MEM:
657 case BPF_LDX | BPF_MEM:
658 if (!(memvalid & (1 << filter[pc].k))) {
659 ret = -EINVAL;
660 goto error;
661 }
662 break;
663 case BPF_JMP | BPF_JA:
664 /* A jump must set masks on target */
665 masks[pc + 1 + filter[pc].k] &= memvalid;
666 memvalid = ~0;
667 break;
668 case BPF_JMP | BPF_JEQ | BPF_K:
669 case BPF_JMP | BPF_JEQ | BPF_X:
670 case BPF_JMP | BPF_JGE | BPF_K:
671 case BPF_JMP | BPF_JGE | BPF_X:
672 case BPF_JMP | BPF_JGT | BPF_K:
673 case BPF_JMP | BPF_JGT | BPF_X:
674 case BPF_JMP | BPF_JSET | BPF_K:
675 case BPF_JMP | BPF_JSET | BPF_X:
676 /* A jump must set masks on targets */
677 masks[pc + 1 + filter[pc].jt] &= memvalid;
678 masks[pc + 1 + filter[pc].jf] &= memvalid;
679 memvalid = ~0;
680 break;
681 }
682 }
683 error:
684 kfree(masks);
685 return ret;
686 }
687
688 static bool chk_code_allowed(u16 code_to_probe)
689 {
690 static const bool codes[] = {
691 /* 32 bit ALU operations */
692 [BPF_ALU | BPF_ADD | BPF_K] = true,
693 [BPF_ALU | BPF_ADD | BPF_X] = true,
694 [BPF_ALU | BPF_SUB | BPF_K] = true,
695 [BPF_ALU | BPF_SUB | BPF_X] = true,
696 [BPF_ALU | BPF_MUL | BPF_K] = true,
697 [BPF_ALU | BPF_MUL | BPF_X] = true,
698 [BPF_ALU | BPF_DIV | BPF_K] = true,
699 [BPF_ALU | BPF_DIV | BPF_X] = true,
700 [BPF_ALU | BPF_MOD | BPF_K] = true,
701 [BPF_ALU | BPF_MOD | BPF_X] = true,
702 [BPF_ALU | BPF_AND | BPF_K] = true,
703 [BPF_ALU | BPF_AND | BPF_X] = true,
704 [BPF_ALU | BPF_OR | BPF_K] = true,
705 [BPF_ALU | BPF_OR | BPF_X] = true,
706 [BPF_ALU | BPF_XOR | BPF_K] = true,
707 [BPF_ALU | BPF_XOR | BPF_X] = true,
708 [BPF_ALU | BPF_LSH | BPF_K] = true,
709 [BPF_ALU | BPF_LSH | BPF_X] = true,
710 [BPF_ALU | BPF_RSH | BPF_K] = true,
711 [BPF_ALU | BPF_RSH | BPF_X] = true,
712 [BPF_ALU | BPF_NEG] = true,
713 /* Load instructions */
714 [BPF_LD | BPF_W | BPF_ABS] = true,
715 [BPF_LD | BPF_H | BPF_ABS] = true,
716 [BPF_LD | BPF_B | BPF_ABS] = true,
717 [BPF_LD | BPF_W | BPF_LEN] = true,
718 [BPF_LD | BPF_W | BPF_IND] = true,
719 [BPF_LD | BPF_H | BPF_IND] = true,
720 [BPF_LD | BPF_B | BPF_IND] = true,
721 [BPF_LD | BPF_IMM] = true,
722 [BPF_LD | BPF_MEM] = true,
723 [BPF_LDX | BPF_W | BPF_LEN] = true,
724 [BPF_LDX | BPF_B | BPF_MSH] = true,
725 [BPF_LDX | BPF_IMM] = true,
726 [BPF_LDX | BPF_MEM] = true,
727 /* Store instructions */
728 [BPF_ST] = true,
729 [BPF_STX] = true,
730 /* Misc instructions */
731 [BPF_MISC | BPF_TAX] = true,
732 [BPF_MISC | BPF_TXA] = true,
733 /* Return instructions */
734 [BPF_RET | BPF_K] = true,
735 [BPF_RET | BPF_A] = true,
736 /* Jump instructions */
737 [BPF_JMP | BPF_JA] = true,
738 [BPF_JMP | BPF_JEQ | BPF_K] = true,
739 [BPF_JMP | BPF_JEQ | BPF_X] = true,
740 [BPF_JMP | BPF_JGE | BPF_K] = true,
741 [BPF_JMP | BPF_JGE | BPF_X] = true,
742 [BPF_JMP | BPF_JGT | BPF_K] = true,
743 [BPF_JMP | BPF_JGT | BPF_X] = true,
744 [BPF_JMP | BPF_JSET | BPF_K] = true,
745 [BPF_JMP | BPF_JSET | BPF_X] = true,
746 };
747
748 if (code_to_probe >= ARRAY_SIZE(codes))
749 return false;
750
751 return codes[code_to_probe];
752 }
753
754 static bool bpf_check_basics_ok(const struct sock_filter *filter,
755 unsigned int flen)
756 {
757 if (filter == NULL)
758 return false;
759 if (flen == 0 || flen > BPF_MAXINSNS)
760 return false;
761
762 return true;
763 }
764
765 /**
766 * bpf_check_classic - verify socket filter code
767 * @filter: filter to verify
768 * @flen: length of filter
769 *
770 * Check the user's filter code. If we let some ugly
771 * filter code slip through kaboom! The filter must contain
772 * no references or jumps that are out of range, no illegal
773 * instructions, and must end with a RET instruction.
774 *
775 * All jumps are forward as they are not signed.
776 *
777 * Returns 0 if the rule set is legal or -EINVAL if not.
778 */
779 static int bpf_check_classic(const struct sock_filter *filter,
780 unsigned int flen)
781 {
782 bool anc_found;
783 int pc;
784
785 /* Check the filter code now */
786 for (pc = 0; pc < flen; pc++) {
787 const struct sock_filter *ftest = &filter[pc];
788
789 /* May we actually operate on this code? */
790 if (!chk_code_allowed(ftest->code))
791 return -EINVAL;
792
793 /* Some instructions need special checks */
794 switch (ftest->code) {
795 case BPF_ALU | BPF_DIV | BPF_K:
796 case BPF_ALU | BPF_MOD | BPF_K:
797 /* Check for division by zero */
798 if (ftest->k == 0)
799 return -EINVAL;
800 break;
801 case BPF_ALU | BPF_LSH | BPF_K:
802 case BPF_ALU | BPF_RSH | BPF_K:
803 if (ftest->k >= 32)
804 return -EINVAL;
805 break;
806 case BPF_LD | BPF_MEM:
807 case BPF_LDX | BPF_MEM:
808 case BPF_ST:
809 case BPF_STX:
810 /* Check for invalid memory addresses */
811 if (ftest->k >= BPF_MEMWORDS)
812 return -EINVAL;
813 break;
814 case BPF_JMP | BPF_JA:
815 /* Note, the large ftest->k might cause loops.
816 * Compare this with conditional jumps below,
817 * where offsets are limited. --ANK (981016)
818 */
819 if (ftest->k >= (unsigned int)(flen - pc - 1))
820 return -EINVAL;
821 break;
822 case BPF_JMP | BPF_JEQ | BPF_K:
823 case BPF_JMP | BPF_JEQ | BPF_X:
824 case BPF_JMP | BPF_JGE | BPF_K:
825 case BPF_JMP | BPF_JGE | BPF_X:
826 case BPF_JMP | BPF_JGT | BPF_K:
827 case BPF_JMP | BPF_JGT | BPF_X:
828 case BPF_JMP | BPF_JSET | BPF_K:
829 case BPF_JMP | BPF_JSET | BPF_X:
830 /* Both conditionals must be safe */
831 if (pc + ftest->jt + 1 >= flen ||
832 pc + ftest->jf + 1 >= flen)
833 return -EINVAL;
834 break;
835 case BPF_LD | BPF_W | BPF_ABS:
836 case BPF_LD | BPF_H | BPF_ABS:
837 case BPF_LD | BPF_B | BPF_ABS:
838 anc_found = false;
839 if (bpf_anc_helper(ftest) & BPF_ANC)
840 anc_found = true;
841 /* Ancillary operation unknown or unsupported */
842 if (anc_found == false && ftest->k >= SKF_AD_OFF)
843 return -EINVAL;
844 }
845 }
846
847 /* Last instruction must be a RET code */
848 switch (filter[flen - 1].code) {
849 case BPF_RET | BPF_K:
850 case BPF_RET | BPF_A:
851 return check_load_and_stores(filter, flen);
852 }
853
854 return -EINVAL;
855 }
856
857 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
858 const struct sock_fprog *fprog)
859 {
860 unsigned int fsize = bpf_classic_proglen(fprog);
861 struct sock_fprog_kern *fkprog;
862
863 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
864 if (!fp->orig_prog)
865 return -ENOMEM;
866
867 fkprog = fp->orig_prog;
868 fkprog->len = fprog->len;
869
870 fkprog->filter = kmemdup(fp->insns, fsize,
871 GFP_KERNEL | __GFP_NOWARN);
872 if (!fkprog->filter) {
873 kfree(fp->orig_prog);
874 return -ENOMEM;
875 }
876
877 return 0;
878 }
879
880 static void bpf_release_orig_filter(struct bpf_prog *fp)
881 {
882 struct sock_fprog_kern *fprog = fp->orig_prog;
883
884 if (fprog) {
885 kfree(fprog->filter);
886 kfree(fprog);
887 }
888 }
889
890 static void __bpf_prog_release(struct bpf_prog *prog)
891 {
892 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
893 bpf_prog_put(prog);
894 } else {
895 bpf_release_orig_filter(prog);
896 bpf_prog_free(prog);
897 }
898 }
899
900 static void __sk_filter_release(struct sk_filter *fp)
901 {
902 __bpf_prog_release(fp->prog);
903 kfree(fp);
904 }
905
906 /**
907 * sk_filter_release_rcu - Release a socket filter by rcu_head
908 * @rcu: rcu_head that contains the sk_filter to free
909 */
910 static void sk_filter_release_rcu(struct rcu_head *rcu)
911 {
912 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
913
914 __sk_filter_release(fp);
915 }
916
917 /**
918 * sk_filter_release - release a socket filter
919 * @fp: filter to remove
920 *
921 * Remove a filter from a socket and release its resources.
922 */
923 static void sk_filter_release(struct sk_filter *fp)
924 {
925 if (atomic_dec_and_test(&fp->refcnt))
926 call_rcu(&fp->rcu, sk_filter_release_rcu);
927 }
928
929 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
930 {
931 u32 filter_size = bpf_prog_size(fp->prog->len);
932
933 atomic_sub(filter_size, &sk->sk_omem_alloc);
934 sk_filter_release(fp);
935 }
936
937 /* try to charge the socket memory if there is space available
938 * return true on success
939 */
940 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
941 {
942 u32 filter_size = bpf_prog_size(fp->prog->len);
943
944 /* same check as in sock_kmalloc() */
945 if (filter_size <= sysctl_optmem_max &&
946 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
947 atomic_inc(&fp->refcnt);
948 atomic_add(filter_size, &sk->sk_omem_alloc);
949 return true;
950 }
951 return false;
952 }
953
954 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
955 {
956 struct sock_filter *old_prog;
957 struct bpf_prog *old_fp;
958 int err, new_len, old_len = fp->len;
959
960 /* We are free to overwrite insns et al right here as it
961 * won't be used at this point in time anymore internally
962 * after the migration to the internal BPF instruction
963 * representation.
964 */
965 BUILD_BUG_ON(sizeof(struct sock_filter) !=
966 sizeof(struct bpf_insn));
967
968 /* Conversion cannot happen on overlapping memory areas,
969 * so we need to keep the user BPF around until the 2nd
970 * pass. At this time, the user BPF is stored in fp->insns.
971 */
972 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
973 GFP_KERNEL | __GFP_NOWARN);
974 if (!old_prog) {
975 err = -ENOMEM;
976 goto out_err;
977 }
978
979 /* 1st pass: calculate the new program length. */
980 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
981 if (err)
982 goto out_err_free;
983
984 /* Expand fp for appending the new filter representation. */
985 old_fp = fp;
986 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
987 if (!fp) {
988 /* The old_fp is still around in case we couldn't
989 * allocate new memory, so uncharge on that one.
990 */
991 fp = old_fp;
992 err = -ENOMEM;
993 goto out_err_free;
994 }
995
996 fp->len = new_len;
997
998 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
999 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
1000 if (err)
1001 /* 2nd bpf_convert_filter() can fail only if it fails
1002 * to allocate memory, remapping must succeed. Note,
1003 * that at this time old_fp has already been released
1004 * by krealloc().
1005 */
1006 goto out_err_free;
1007
1008 /* We are guaranteed to never error here with cBPF to eBPF
1009 * transitions, since there's no issue with type compatibility
1010 * checks on program arrays.
1011 */
1012 fp = bpf_prog_select_runtime(fp, &err);
1013
1014 kfree(old_prog);
1015 return fp;
1016
1017 out_err_free:
1018 kfree(old_prog);
1019 out_err:
1020 __bpf_prog_release(fp);
1021 return ERR_PTR(err);
1022 }
1023
1024 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1025 bpf_aux_classic_check_t trans)
1026 {
1027 int err;
1028
1029 fp->bpf_func = NULL;
1030 fp->jited = 0;
1031
1032 err = bpf_check_classic(fp->insns, fp->len);
1033 if (err) {
1034 __bpf_prog_release(fp);
1035 return ERR_PTR(err);
1036 }
1037
1038 /* There might be additional checks and transformations
1039 * needed on classic filters, f.e. in case of seccomp.
1040 */
1041 if (trans) {
1042 err = trans(fp->insns, fp->len);
1043 if (err) {
1044 __bpf_prog_release(fp);
1045 return ERR_PTR(err);
1046 }
1047 }
1048
1049 /* Probe if we can JIT compile the filter and if so, do
1050 * the compilation of the filter.
1051 */
1052 bpf_jit_compile(fp);
1053
1054 /* JIT compiler couldn't process this filter, so do the
1055 * internal BPF translation for the optimized interpreter.
1056 */
1057 if (!fp->jited)
1058 fp = bpf_migrate_filter(fp);
1059
1060 return fp;
1061 }
1062
1063 /**
1064 * bpf_prog_create - create an unattached filter
1065 * @pfp: the unattached filter that is created
1066 * @fprog: the filter program
1067 *
1068 * Create a filter independent of any socket. We first run some
1069 * sanity checks on it to make sure it does not explode on us later.
1070 * If an error occurs or there is insufficient memory for the filter
1071 * a negative errno code is returned. On success the return is zero.
1072 */
1073 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1074 {
1075 unsigned int fsize = bpf_classic_proglen(fprog);
1076 struct bpf_prog *fp;
1077
1078 /* Make sure new filter is there and in the right amounts. */
1079 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1080 return -EINVAL;
1081
1082 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1083 if (!fp)
1084 return -ENOMEM;
1085
1086 memcpy(fp->insns, fprog->filter, fsize);
1087
1088 fp->len = fprog->len;
1089 /* Since unattached filters are not copied back to user
1090 * space through sk_get_filter(), we do not need to hold
1091 * a copy here, and can spare us the work.
1092 */
1093 fp->orig_prog = NULL;
1094
1095 /* bpf_prepare_filter() already takes care of freeing
1096 * memory in case something goes wrong.
1097 */
1098 fp = bpf_prepare_filter(fp, NULL);
1099 if (IS_ERR(fp))
1100 return PTR_ERR(fp);
1101
1102 *pfp = fp;
1103 return 0;
1104 }
1105 EXPORT_SYMBOL_GPL(bpf_prog_create);
1106
1107 /**
1108 * bpf_prog_create_from_user - create an unattached filter from user buffer
1109 * @pfp: the unattached filter that is created
1110 * @fprog: the filter program
1111 * @trans: post-classic verifier transformation handler
1112 * @save_orig: save classic BPF program
1113 *
1114 * This function effectively does the same as bpf_prog_create(), only
1115 * that it builds up its insns buffer from user space provided buffer.
1116 * It also allows for passing a bpf_aux_classic_check_t handler.
1117 */
1118 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1119 bpf_aux_classic_check_t trans, bool save_orig)
1120 {
1121 unsigned int fsize = bpf_classic_proglen(fprog);
1122 struct bpf_prog *fp;
1123 int err;
1124
1125 /* Make sure new filter is there and in the right amounts. */
1126 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1127 return -EINVAL;
1128
1129 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1130 if (!fp)
1131 return -ENOMEM;
1132
1133 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1134 __bpf_prog_free(fp);
1135 return -EFAULT;
1136 }
1137
1138 fp->len = fprog->len;
1139 fp->orig_prog = NULL;
1140
1141 if (save_orig) {
1142 err = bpf_prog_store_orig_filter(fp, fprog);
1143 if (err) {
1144 __bpf_prog_free(fp);
1145 return -ENOMEM;
1146 }
1147 }
1148
1149 /* bpf_prepare_filter() already takes care of freeing
1150 * memory in case something goes wrong.
1151 */
1152 fp = bpf_prepare_filter(fp, trans);
1153 if (IS_ERR(fp))
1154 return PTR_ERR(fp);
1155
1156 *pfp = fp;
1157 return 0;
1158 }
1159 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1160
1161 void bpf_prog_destroy(struct bpf_prog *fp)
1162 {
1163 __bpf_prog_release(fp);
1164 }
1165 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1166
1167 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1168 {
1169 struct sk_filter *fp, *old_fp;
1170
1171 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1172 if (!fp)
1173 return -ENOMEM;
1174
1175 fp->prog = prog;
1176 atomic_set(&fp->refcnt, 0);
1177
1178 if (!sk_filter_charge(sk, fp)) {
1179 kfree(fp);
1180 return -ENOMEM;
1181 }
1182
1183 old_fp = rcu_dereference_protected(sk->sk_filter,
1184 lockdep_sock_is_held(sk));
1185 rcu_assign_pointer(sk->sk_filter, fp);
1186
1187 if (old_fp)
1188 sk_filter_uncharge(sk, old_fp);
1189
1190 return 0;
1191 }
1192
1193 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1194 {
1195 struct bpf_prog *old_prog;
1196 int err;
1197
1198 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1199 return -ENOMEM;
1200
1201 if (sk_unhashed(sk) && sk->sk_reuseport) {
1202 err = reuseport_alloc(sk);
1203 if (err)
1204 return err;
1205 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1206 /* The socket wasn't bound with SO_REUSEPORT */
1207 return -EINVAL;
1208 }
1209
1210 old_prog = reuseport_attach_prog(sk, prog);
1211 if (old_prog)
1212 bpf_prog_destroy(old_prog);
1213
1214 return 0;
1215 }
1216
1217 static
1218 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1219 {
1220 unsigned int fsize = bpf_classic_proglen(fprog);
1221 struct bpf_prog *prog;
1222 int err;
1223
1224 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1225 return ERR_PTR(-EPERM);
1226
1227 /* Make sure new filter is there and in the right amounts. */
1228 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1229 return ERR_PTR(-EINVAL);
1230
1231 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1232 if (!prog)
1233 return ERR_PTR(-ENOMEM);
1234
1235 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1236 __bpf_prog_free(prog);
1237 return ERR_PTR(-EFAULT);
1238 }
1239
1240 prog->len = fprog->len;
1241
1242 err = bpf_prog_store_orig_filter(prog, fprog);
1243 if (err) {
1244 __bpf_prog_free(prog);
1245 return ERR_PTR(-ENOMEM);
1246 }
1247
1248 /* bpf_prepare_filter() already takes care of freeing
1249 * memory in case something goes wrong.
1250 */
1251 return bpf_prepare_filter(prog, NULL);
1252 }
1253
1254 /**
1255 * sk_attach_filter - attach a socket filter
1256 * @fprog: the filter program
1257 * @sk: the socket to use
1258 *
1259 * Attach the user's filter code. We first run some sanity checks on
1260 * it to make sure it does not explode on us later. If an error
1261 * occurs or there is insufficient memory for the filter a negative
1262 * errno code is returned. On success the return is zero.
1263 */
1264 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1265 {
1266 struct bpf_prog *prog = __get_filter(fprog, sk);
1267 int err;
1268
1269 if (IS_ERR(prog))
1270 return PTR_ERR(prog);
1271
1272 err = __sk_attach_prog(prog, sk);
1273 if (err < 0) {
1274 __bpf_prog_release(prog);
1275 return err;
1276 }
1277
1278 return 0;
1279 }
1280 EXPORT_SYMBOL_GPL(sk_attach_filter);
1281
1282 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1283 {
1284 struct bpf_prog *prog = __get_filter(fprog, sk);
1285 int err;
1286
1287 if (IS_ERR(prog))
1288 return PTR_ERR(prog);
1289
1290 err = __reuseport_attach_prog(prog, sk);
1291 if (err < 0) {
1292 __bpf_prog_release(prog);
1293 return err;
1294 }
1295
1296 return 0;
1297 }
1298
1299 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1300 {
1301 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1302 return ERR_PTR(-EPERM);
1303
1304 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1305 }
1306
1307 int sk_attach_bpf(u32 ufd, struct sock *sk)
1308 {
1309 struct bpf_prog *prog = __get_bpf(ufd, sk);
1310 int err;
1311
1312 if (IS_ERR(prog))
1313 return PTR_ERR(prog);
1314
1315 err = __sk_attach_prog(prog, sk);
1316 if (err < 0) {
1317 bpf_prog_put(prog);
1318 return err;
1319 }
1320
1321 return 0;
1322 }
1323
1324 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1325 {
1326 struct bpf_prog *prog = __get_bpf(ufd, sk);
1327 int err;
1328
1329 if (IS_ERR(prog))
1330 return PTR_ERR(prog);
1331
1332 err = __reuseport_attach_prog(prog, sk);
1333 if (err < 0) {
1334 bpf_prog_put(prog);
1335 return err;
1336 }
1337
1338 return 0;
1339 }
1340
1341 struct bpf_scratchpad {
1342 union {
1343 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1344 u8 buff[MAX_BPF_STACK];
1345 };
1346 };
1347
1348 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1349
1350 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1351 unsigned int write_len)
1352 {
1353 return skb_ensure_writable(skb, write_len);
1354 }
1355
1356 static inline int bpf_try_make_writable(struct sk_buff *skb,
1357 unsigned int write_len)
1358 {
1359 int err = __bpf_try_make_writable(skb, write_len);
1360
1361 bpf_compute_data_end(skb);
1362 return err;
1363 }
1364
1365 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1366 {
1367 if (skb_at_tc_ingress(skb))
1368 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1369 }
1370
1371 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1372 {
1373 if (skb_at_tc_ingress(skb))
1374 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1375 }
1376
1377 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1378 const void *, from, u32, len, u64, flags)
1379 {
1380 void *ptr;
1381
1382 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1383 return -EINVAL;
1384 if (unlikely(offset > 0xffff))
1385 return -EFAULT;
1386 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1387 return -EFAULT;
1388
1389 ptr = skb->data + offset;
1390 if (flags & BPF_F_RECOMPUTE_CSUM)
1391 __skb_postpull_rcsum(skb, ptr, len, offset);
1392
1393 memcpy(ptr, from, len);
1394
1395 if (flags & BPF_F_RECOMPUTE_CSUM)
1396 __skb_postpush_rcsum(skb, ptr, len, offset);
1397 if (flags & BPF_F_INVALIDATE_HASH)
1398 skb_clear_hash(skb);
1399
1400 return 0;
1401 }
1402
1403 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1404 .func = bpf_skb_store_bytes,
1405 .gpl_only = false,
1406 .ret_type = RET_INTEGER,
1407 .arg1_type = ARG_PTR_TO_CTX,
1408 .arg2_type = ARG_ANYTHING,
1409 .arg3_type = ARG_PTR_TO_STACK,
1410 .arg4_type = ARG_CONST_STACK_SIZE,
1411 .arg5_type = ARG_ANYTHING,
1412 };
1413
1414 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1415 void *, to, u32, len)
1416 {
1417 void *ptr;
1418
1419 if (unlikely(offset > 0xffff))
1420 goto err_clear;
1421
1422 ptr = skb_header_pointer(skb, offset, len, to);
1423 if (unlikely(!ptr))
1424 goto err_clear;
1425 if (ptr != to)
1426 memcpy(to, ptr, len);
1427
1428 return 0;
1429 err_clear:
1430 memset(to, 0, len);
1431 return -EFAULT;
1432 }
1433
1434 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1435 .func = bpf_skb_load_bytes,
1436 .gpl_only = false,
1437 .ret_type = RET_INTEGER,
1438 .arg1_type = ARG_PTR_TO_CTX,
1439 .arg2_type = ARG_ANYTHING,
1440 .arg3_type = ARG_PTR_TO_RAW_STACK,
1441 .arg4_type = ARG_CONST_STACK_SIZE,
1442 };
1443
1444 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1445 u64, from, u64, to, u64, flags)
1446 {
1447 __sum16 *ptr;
1448
1449 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1450 return -EINVAL;
1451 if (unlikely(offset > 0xffff || offset & 1))
1452 return -EFAULT;
1453 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1454 return -EFAULT;
1455
1456 ptr = (__sum16 *)(skb->data + offset);
1457 switch (flags & BPF_F_HDR_FIELD_MASK) {
1458 case 0:
1459 if (unlikely(from != 0))
1460 return -EINVAL;
1461
1462 csum_replace_by_diff(ptr, to);
1463 break;
1464 case 2:
1465 csum_replace2(ptr, from, to);
1466 break;
1467 case 4:
1468 csum_replace4(ptr, from, to);
1469 break;
1470 default:
1471 return -EINVAL;
1472 }
1473
1474 return 0;
1475 }
1476
1477 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1478 .func = bpf_l3_csum_replace,
1479 .gpl_only = false,
1480 .ret_type = RET_INTEGER,
1481 .arg1_type = ARG_PTR_TO_CTX,
1482 .arg2_type = ARG_ANYTHING,
1483 .arg3_type = ARG_ANYTHING,
1484 .arg4_type = ARG_ANYTHING,
1485 .arg5_type = ARG_ANYTHING,
1486 };
1487
1488 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1489 u64, from, u64, to, u64, flags)
1490 {
1491 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1492 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1493 __sum16 *ptr;
1494
1495 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR |
1496 BPF_F_HDR_FIELD_MASK)))
1497 return -EINVAL;
1498 if (unlikely(offset > 0xffff || offset & 1))
1499 return -EFAULT;
1500 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1501 return -EFAULT;
1502
1503 ptr = (__sum16 *)(skb->data + offset);
1504 if (is_mmzero && !*ptr)
1505 return 0;
1506
1507 switch (flags & BPF_F_HDR_FIELD_MASK) {
1508 case 0:
1509 if (unlikely(from != 0))
1510 return -EINVAL;
1511
1512 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1513 break;
1514 case 2:
1515 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1516 break;
1517 case 4:
1518 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1519 break;
1520 default:
1521 return -EINVAL;
1522 }
1523
1524 if (is_mmzero && !*ptr)
1525 *ptr = CSUM_MANGLED_0;
1526 return 0;
1527 }
1528
1529 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1530 .func = bpf_l4_csum_replace,
1531 .gpl_only = false,
1532 .ret_type = RET_INTEGER,
1533 .arg1_type = ARG_PTR_TO_CTX,
1534 .arg2_type = ARG_ANYTHING,
1535 .arg3_type = ARG_ANYTHING,
1536 .arg4_type = ARG_ANYTHING,
1537 .arg5_type = ARG_ANYTHING,
1538 };
1539
1540 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1541 __be32 *, to, u32, to_size, __wsum, seed)
1542 {
1543 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1544 u32 diff_size = from_size + to_size;
1545 int i, j = 0;
1546
1547 /* This is quite flexible, some examples:
1548 *
1549 * from_size == 0, to_size > 0, seed := csum --> pushing data
1550 * from_size > 0, to_size == 0, seed := csum --> pulling data
1551 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1552 *
1553 * Even for diffing, from_size and to_size don't need to be equal.
1554 */
1555 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1556 diff_size > sizeof(sp->diff)))
1557 return -EINVAL;
1558
1559 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1560 sp->diff[j] = ~from[i];
1561 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1562 sp->diff[j] = to[i];
1563
1564 return csum_partial(sp->diff, diff_size, seed);
1565 }
1566
1567 static const struct bpf_func_proto bpf_csum_diff_proto = {
1568 .func = bpf_csum_diff,
1569 .gpl_only = false,
1570 .ret_type = RET_INTEGER,
1571 .arg1_type = ARG_PTR_TO_STACK,
1572 .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1573 .arg3_type = ARG_PTR_TO_STACK,
1574 .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1575 .arg5_type = ARG_ANYTHING,
1576 };
1577
1578 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1579 {
1580 return dev_forward_skb(dev, skb);
1581 }
1582
1583 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1584 {
1585 int ret;
1586
1587 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1588 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1589 kfree_skb(skb);
1590 return -ENETDOWN;
1591 }
1592
1593 skb->dev = dev;
1594
1595 __this_cpu_inc(xmit_recursion);
1596 ret = dev_queue_xmit(skb);
1597 __this_cpu_dec(xmit_recursion);
1598
1599 return ret;
1600 }
1601
1602 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
1603 {
1604 struct net_device *dev;
1605
1606 if (unlikely(flags & ~(BPF_F_INGRESS)))
1607 return -EINVAL;
1608
1609 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1610 if (unlikely(!dev))
1611 return -EINVAL;
1612
1613 skb = skb_clone(skb, GFP_ATOMIC);
1614 if (unlikely(!skb))
1615 return -ENOMEM;
1616
1617 bpf_push_mac_rcsum(skb);
1618
1619 return flags & BPF_F_INGRESS ?
1620 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1621 }
1622
1623 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1624 .func = bpf_clone_redirect,
1625 .gpl_only = false,
1626 .ret_type = RET_INTEGER,
1627 .arg1_type = ARG_PTR_TO_CTX,
1628 .arg2_type = ARG_ANYTHING,
1629 .arg3_type = ARG_ANYTHING,
1630 };
1631
1632 struct redirect_info {
1633 u32 ifindex;
1634 u32 flags;
1635 };
1636
1637 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1638
1639 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
1640 {
1641 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1642
1643 if (unlikely(flags & ~(BPF_F_INGRESS)))
1644 return TC_ACT_SHOT;
1645
1646 ri->ifindex = ifindex;
1647 ri->flags = flags;
1648
1649 return TC_ACT_REDIRECT;
1650 }
1651
1652 int skb_do_redirect(struct sk_buff *skb)
1653 {
1654 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1655 struct net_device *dev;
1656
1657 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1658 ri->ifindex = 0;
1659 if (unlikely(!dev)) {
1660 kfree_skb(skb);
1661 return -EINVAL;
1662 }
1663
1664 bpf_push_mac_rcsum(skb);
1665
1666 return ri->flags & BPF_F_INGRESS ?
1667 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1668 }
1669
1670 static const struct bpf_func_proto bpf_redirect_proto = {
1671 .func = bpf_redirect,
1672 .gpl_only = false,
1673 .ret_type = RET_INTEGER,
1674 .arg1_type = ARG_ANYTHING,
1675 .arg2_type = ARG_ANYTHING,
1676 };
1677
1678 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
1679 {
1680 return task_get_classid(skb);
1681 }
1682
1683 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1684 .func = bpf_get_cgroup_classid,
1685 .gpl_only = false,
1686 .ret_type = RET_INTEGER,
1687 .arg1_type = ARG_PTR_TO_CTX,
1688 };
1689
1690 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
1691 {
1692 return dst_tclassid(skb);
1693 }
1694
1695 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1696 .func = bpf_get_route_realm,
1697 .gpl_only = false,
1698 .ret_type = RET_INTEGER,
1699 .arg1_type = ARG_PTR_TO_CTX,
1700 };
1701
1702 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
1703 {
1704 /* If skb_clear_hash() was called due to mangling, we can
1705 * trigger SW recalculation here. Later access to hash
1706 * can then use the inline skb->hash via context directly
1707 * instead of calling this helper again.
1708 */
1709 return skb_get_hash(skb);
1710 }
1711
1712 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1713 .func = bpf_get_hash_recalc,
1714 .gpl_only = false,
1715 .ret_type = RET_INTEGER,
1716 .arg1_type = ARG_PTR_TO_CTX,
1717 };
1718
1719 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
1720 u16, vlan_tci)
1721 {
1722 int ret;
1723
1724 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1725 vlan_proto != htons(ETH_P_8021AD)))
1726 vlan_proto = htons(ETH_P_8021Q);
1727
1728 bpf_push_mac_rcsum(skb);
1729 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1730 bpf_pull_mac_rcsum(skb);
1731
1732 bpf_compute_data_end(skb);
1733 return ret;
1734 }
1735
1736 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1737 .func = bpf_skb_vlan_push,
1738 .gpl_only = false,
1739 .ret_type = RET_INTEGER,
1740 .arg1_type = ARG_PTR_TO_CTX,
1741 .arg2_type = ARG_ANYTHING,
1742 .arg3_type = ARG_ANYTHING,
1743 };
1744 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1745
1746 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
1747 {
1748 int ret;
1749
1750 bpf_push_mac_rcsum(skb);
1751 ret = skb_vlan_pop(skb);
1752 bpf_pull_mac_rcsum(skb);
1753
1754 bpf_compute_data_end(skb);
1755 return ret;
1756 }
1757
1758 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1759 .func = bpf_skb_vlan_pop,
1760 .gpl_only = false,
1761 .ret_type = RET_INTEGER,
1762 .arg1_type = ARG_PTR_TO_CTX,
1763 };
1764 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1765
1766 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
1767 {
1768 /* Caller already did skb_cow() with len as headroom,
1769 * so no need to do it here.
1770 */
1771 skb_push(skb, len);
1772 memmove(skb->data, skb->data + len, off);
1773 memset(skb->data + off, 0, len);
1774
1775 /* No skb_postpush_rcsum(skb, skb->data + off, len)
1776 * needed here as it does not change the skb->csum
1777 * result for checksum complete when summing over
1778 * zeroed blocks.
1779 */
1780 return 0;
1781 }
1782
1783 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
1784 {
1785 /* skb_ensure_writable() is not needed here, as we're
1786 * already working on an uncloned skb.
1787 */
1788 if (unlikely(!pskb_may_pull(skb, off + len)))
1789 return -ENOMEM;
1790
1791 skb_postpull_rcsum(skb, skb->data + off, len);
1792 memmove(skb->data + len, skb->data, off);
1793 __skb_pull(skb, len);
1794
1795 return 0;
1796 }
1797
1798 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
1799 {
1800 bool trans_same = skb->transport_header == skb->network_header;
1801 int ret;
1802
1803 /* There's no need for __skb_push()/__skb_pull() pair to
1804 * get to the start of the mac header as we're guaranteed
1805 * to always start from here under eBPF.
1806 */
1807 ret = bpf_skb_generic_push(skb, off, len);
1808 if (likely(!ret)) {
1809 skb->mac_header -= len;
1810 skb->network_header -= len;
1811 if (trans_same)
1812 skb->transport_header = skb->network_header;
1813 }
1814
1815 return ret;
1816 }
1817
1818 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
1819 {
1820 bool trans_same = skb->transport_header == skb->network_header;
1821 int ret;
1822
1823 /* Same here, __skb_push()/__skb_pull() pair not needed. */
1824 ret = bpf_skb_generic_pop(skb, off, len);
1825 if (likely(!ret)) {
1826 skb->mac_header += len;
1827 skb->network_header += len;
1828 if (trans_same)
1829 skb->transport_header = skb->network_header;
1830 }
1831
1832 return ret;
1833 }
1834
1835 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
1836 {
1837 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1838 u32 off = skb->network_header - skb->mac_header;
1839 int ret;
1840
1841 ret = skb_cow(skb, len_diff);
1842 if (unlikely(ret < 0))
1843 return ret;
1844
1845 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
1846 if (unlikely(ret < 0))
1847 return ret;
1848
1849 if (skb_is_gso(skb)) {
1850 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV4 needs to
1851 * be changed into SKB_GSO_TCPV6.
1852 */
1853 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
1854 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
1855 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
1856 }
1857
1858 /* Due to IPv6 header, MSS needs to be downgraded. */
1859 skb_shinfo(skb)->gso_size -= len_diff;
1860 /* Header must be checked, and gso_segs recomputed. */
1861 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1862 skb_shinfo(skb)->gso_segs = 0;
1863 }
1864
1865 skb->protocol = htons(ETH_P_IPV6);
1866 skb_clear_hash(skb);
1867
1868 return 0;
1869 }
1870
1871 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
1872 {
1873 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1874 u32 off = skb->network_header - skb->mac_header;
1875 int ret;
1876
1877 ret = skb_unclone(skb, GFP_ATOMIC);
1878 if (unlikely(ret < 0))
1879 return ret;
1880
1881 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
1882 if (unlikely(ret < 0))
1883 return ret;
1884
1885 if (skb_is_gso(skb)) {
1886 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV6 needs to
1887 * be changed into SKB_GSO_TCPV4.
1888 */
1889 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
1890 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
1891 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
1892 }
1893
1894 /* Due to IPv4 header, MSS can be upgraded. */
1895 skb_shinfo(skb)->gso_size += len_diff;
1896 /* Header must be checked, and gso_segs recomputed. */
1897 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1898 skb_shinfo(skb)->gso_segs = 0;
1899 }
1900
1901 skb->protocol = htons(ETH_P_IP);
1902 skb_clear_hash(skb);
1903
1904 return 0;
1905 }
1906
1907 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
1908 {
1909 __be16 from_proto = skb->protocol;
1910
1911 if (from_proto == htons(ETH_P_IP) &&
1912 to_proto == htons(ETH_P_IPV6))
1913 return bpf_skb_proto_4_to_6(skb);
1914
1915 if (from_proto == htons(ETH_P_IPV6) &&
1916 to_proto == htons(ETH_P_IP))
1917 return bpf_skb_proto_6_to_4(skb);
1918
1919 return -ENOTSUPP;
1920 }
1921
1922 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
1923 u64, flags)
1924 {
1925 int ret;
1926
1927 if (unlikely(flags))
1928 return -EINVAL;
1929
1930 /* General idea is that this helper does the basic groundwork
1931 * needed for changing the protocol, and eBPF program fills the
1932 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
1933 * and other helpers, rather than passing a raw buffer here.
1934 *
1935 * The rationale is to keep this minimal and without a need to
1936 * deal with raw packet data. F.e. even if we would pass buffers
1937 * here, the program still needs to call the bpf_lX_csum_replace()
1938 * helpers anyway. Plus, this way we keep also separation of
1939 * concerns, since f.e. bpf_skb_store_bytes() should only take
1940 * care of stores.
1941 *
1942 * Currently, additional options and extension header space are
1943 * not supported, but flags register is reserved so we can adapt
1944 * that. For offloads, we mark packet as dodgy, so that headers
1945 * need to be verified first.
1946 */
1947 ret = bpf_skb_proto_xlat(skb, proto);
1948 bpf_compute_data_end(skb);
1949 return ret;
1950 }
1951
1952 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
1953 .func = bpf_skb_change_proto,
1954 .gpl_only = false,
1955 .ret_type = RET_INTEGER,
1956 .arg1_type = ARG_PTR_TO_CTX,
1957 .arg2_type = ARG_ANYTHING,
1958 .arg3_type = ARG_ANYTHING,
1959 };
1960
1961 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
1962 {
1963 /* We only allow a restricted subset to be changed for now. */
1964 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
1965 !skb_pkt_type_ok(pkt_type)))
1966 return -EINVAL;
1967
1968 skb->pkt_type = pkt_type;
1969 return 0;
1970 }
1971
1972 static const struct bpf_func_proto bpf_skb_change_type_proto = {
1973 .func = bpf_skb_change_type,
1974 .gpl_only = false,
1975 .ret_type = RET_INTEGER,
1976 .arg1_type = ARG_PTR_TO_CTX,
1977 .arg2_type = ARG_ANYTHING,
1978 };
1979
1980 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
1981 {
1982 u32 min_len = skb_network_offset(skb);
1983
1984 if (skb_transport_header_was_set(skb))
1985 min_len = skb_transport_offset(skb);
1986 if (skb->ip_summed == CHECKSUM_PARTIAL)
1987 min_len = skb_checksum_start_offset(skb) +
1988 skb->csum_offset + sizeof(__sum16);
1989 return min_len;
1990 }
1991
1992 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
1993 {
1994 return skb->dev->mtu + skb->dev->hard_header_len;
1995 }
1996
1997 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
1998 {
1999 unsigned int old_len = skb->len;
2000 int ret;
2001
2002 ret = __skb_grow_rcsum(skb, new_len);
2003 if (!ret)
2004 memset(skb->data + old_len, 0, new_len - old_len);
2005 return ret;
2006 }
2007
2008 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2009 {
2010 return __skb_trim_rcsum(skb, new_len);
2011 }
2012
2013 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2014 u64, flags)
2015 {
2016 u32 max_len = __bpf_skb_max_len(skb);
2017 u32 min_len = __bpf_skb_min_len(skb);
2018 int ret;
2019
2020 if (unlikely(flags || new_len > max_len || new_len < min_len))
2021 return -EINVAL;
2022 if (skb->encapsulation)
2023 return -ENOTSUPP;
2024
2025 /* The basic idea of this helper is that it's performing the
2026 * needed work to either grow or trim an skb, and eBPF program
2027 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2028 * bpf_lX_csum_replace() and others rather than passing a raw
2029 * buffer here. This one is a slow path helper and intended
2030 * for replies with control messages.
2031 *
2032 * Like in bpf_skb_change_proto(), we want to keep this rather
2033 * minimal and without protocol specifics so that we are able
2034 * to separate concerns as in bpf_skb_store_bytes() should only
2035 * be the one responsible for writing buffers.
2036 *
2037 * It's really expected to be a slow path operation here for
2038 * control message replies, so we're implicitly linearizing,
2039 * uncloning and drop offloads from the skb by this.
2040 */
2041 ret = __bpf_try_make_writable(skb, skb->len);
2042 if (!ret) {
2043 if (new_len > skb->len)
2044 ret = bpf_skb_grow_rcsum(skb, new_len);
2045 else if (new_len < skb->len)
2046 ret = bpf_skb_trim_rcsum(skb, new_len);
2047 if (!ret && skb_is_gso(skb))
2048 skb_gso_reset(skb);
2049 }
2050
2051 bpf_compute_data_end(skb);
2052 return ret;
2053 }
2054
2055 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2056 .func = bpf_skb_change_tail,
2057 .gpl_only = false,
2058 .ret_type = RET_INTEGER,
2059 .arg1_type = ARG_PTR_TO_CTX,
2060 .arg2_type = ARG_ANYTHING,
2061 .arg3_type = ARG_ANYTHING,
2062 };
2063
2064 bool bpf_helper_changes_skb_data(void *func)
2065 {
2066 if (func == bpf_skb_vlan_push)
2067 return true;
2068 if (func == bpf_skb_vlan_pop)
2069 return true;
2070 if (func == bpf_skb_store_bytes)
2071 return true;
2072 if (func == bpf_skb_change_proto)
2073 return true;
2074 if (func == bpf_skb_change_tail)
2075 return true;
2076 if (func == bpf_l3_csum_replace)
2077 return true;
2078 if (func == bpf_l4_csum_replace)
2079 return true;
2080
2081 return false;
2082 }
2083
2084 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2085 unsigned long off, unsigned long len)
2086 {
2087 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2088
2089 if (unlikely(!ptr))
2090 return len;
2091 if (ptr != dst_buff)
2092 memcpy(dst_buff, ptr, len);
2093
2094 return 0;
2095 }
2096
2097 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2098 u64, flags, void *, meta, u64, meta_size)
2099 {
2100 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2101
2102 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2103 return -EINVAL;
2104 if (unlikely(skb_size > skb->len))
2105 return -EFAULT;
2106
2107 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2108 bpf_skb_copy);
2109 }
2110
2111 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2112 .func = bpf_skb_event_output,
2113 .gpl_only = true,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_PTR_TO_CTX,
2116 .arg2_type = ARG_CONST_MAP_PTR,
2117 .arg3_type = ARG_ANYTHING,
2118 .arg4_type = ARG_PTR_TO_STACK,
2119 .arg5_type = ARG_CONST_STACK_SIZE,
2120 };
2121
2122 static unsigned short bpf_tunnel_key_af(u64 flags)
2123 {
2124 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2125 }
2126
2127 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2128 u32, size, u64, flags)
2129 {
2130 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2131 u8 compat[sizeof(struct bpf_tunnel_key)];
2132 void *to_orig = to;
2133 int err;
2134
2135 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2136 err = -EINVAL;
2137 goto err_clear;
2138 }
2139 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2140 err = -EPROTO;
2141 goto err_clear;
2142 }
2143 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2144 err = -EINVAL;
2145 switch (size) {
2146 case offsetof(struct bpf_tunnel_key, tunnel_label):
2147 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2148 goto set_compat;
2149 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2150 /* Fixup deprecated structure layouts here, so we have
2151 * a common path later on.
2152 */
2153 if (ip_tunnel_info_af(info) != AF_INET)
2154 goto err_clear;
2155 set_compat:
2156 to = (struct bpf_tunnel_key *)compat;
2157 break;
2158 default:
2159 goto err_clear;
2160 }
2161 }
2162
2163 to->tunnel_id = be64_to_cpu(info->key.tun_id);
2164 to->tunnel_tos = info->key.tos;
2165 to->tunnel_ttl = info->key.ttl;
2166
2167 if (flags & BPF_F_TUNINFO_IPV6) {
2168 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2169 sizeof(to->remote_ipv6));
2170 to->tunnel_label = be32_to_cpu(info->key.label);
2171 } else {
2172 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2173 }
2174
2175 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2176 memcpy(to_orig, to, size);
2177
2178 return 0;
2179 err_clear:
2180 memset(to_orig, 0, size);
2181 return err;
2182 }
2183
2184 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2185 .func = bpf_skb_get_tunnel_key,
2186 .gpl_only = false,
2187 .ret_type = RET_INTEGER,
2188 .arg1_type = ARG_PTR_TO_CTX,
2189 .arg2_type = ARG_PTR_TO_RAW_STACK,
2190 .arg3_type = ARG_CONST_STACK_SIZE,
2191 .arg4_type = ARG_ANYTHING,
2192 };
2193
2194 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2195 {
2196 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2197 int err;
2198
2199 if (unlikely(!info ||
2200 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2201 err = -ENOENT;
2202 goto err_clear;
2203 }
2204 if (unlikely(size < info->options_len)) {
2205 err = -ENOMEM;
2206 goto err_clear;
2207 }
2208
2209 ip_tunnel_info_opts_get(to, info);
2210 if (size > info->options_len)
2211 memset(to + info->options_len, 0, size - info->options_len);
2212
2213 return info->options_len;
2214 err_clear:
2215 memset(to, 0, size);
2216 return err;
2217 }
2218
2219 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2220 .func = bpf_skb_get_tunnel_opt,
2221 .gpl_only = false,
2222 .ret_type = RET_INTEGER,
2223 .arg1_type = ARG_PTR_TO_CTX,
2224 .arg2_type = ARG_PTR_TO_RAW_STACK,
2225 .arg3_type = ARG_CONST_STACK_SIZE,
2226 };
2227
2228 static struct metadata_dst __percpu *md_dst;
2229
2230 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
2231 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
2232 {
2233 struct metadata_dst *md = this_cpu_ptr(md_dst);
2234 u8 compat[sizeof(struct bpf_tunnel_key)];
2235 struct ip_tunnel_info *info;
2236
2237 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2238 BPF_F_DONT_FRAGMENT)))
2239 return -EINVAL;
2240 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2241 switch (size) {
2242 case offsetof(struct bpf_tunnel_key, tunnel_label):
2243 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2244 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2245 /* Fixup deprecated structure layouts here, so we have
2246 * a common path later on.
2247 */
2248 memcpy(compat, from, size);
2249 memset(compat + size, 0, sizeof(compat) - size);
2250 from = (const struct bpf_tunnel_key *) compat;
2251 break;
2252 default:
2253 return -EINVAL;
2254 }
2255 }
2256 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
2257 from->tunnel_ext))
2258 return -EINVAL;
2259
2260 skb_dst_drop(skb);
2261 dst_hold((struct dst_entry *) md);
2262 skb_dst_set(skb, (struct dst_entry *) md);
2263
2264 info = &md->u.tun_info;
2265 info->mode = IP_TUNNEL_INFO_TX;
2266
2267 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
2268 if (flags & BPF_F_DONT_FRAGMENT)
2269 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
2270
2271 info->key.tun_id = cpu_to_be64(from->tunnel_id);
2272 info->key.tos = from->tunnel_tos;
2273 info->key.ttl = from->tunnel_ttl;
2274
2275 if (flags & BPF_F_TUNINFO_IPV6) {
2276 info->mode |= IP_TUNNEL_INFO_IPV6;
2277 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
2278 sizeof(from->remote_ipv6));
2279 info->key.label = cpu_to_be32(from->tunnel_label) &
2280 IPV6_FLOWLABEL_MASK;
2281 } else {
2282 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
2283 if (flags & BPF_F_ZERO_CSUM_TX)
2284 info->key.tun_flags &= ~TUNNEL_CSUM;
2285 }
2286
2287 return 0;
2288 }
2289
2290 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
2291 .func = bpf_skb_set_tunnel_key,
2292 .gpl_only = false,
2293 .ret_type = RET_INTEGER,
2294 .arg1_type = ARG_PTR_TO_CTX,
2295 .arg2_type = ARG_PTR_TO_STACK,
2296 .arg3_type = ARG_CONST_STACK_SIZE,
2297 .arg4_type = ARG_ANYTHING,
2298 };
2299
2300 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
2301 const u8 *, from, u32, size)
2302 {
2303 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2304 const struct metadata_dst *md = this_cpu_ptr(md_dst);
2305
2306 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
2307 return -EINVAL;
2308 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
2309 return -ENOMEM;
2310
2311 ip_tunnel_info_opts_set(info, from, size);
2312
2313 return 0;
2314 }
2315
2316 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
2317 .func = bpf_skb_set_tunnel_opt,
2318 .gpl_only = false,
2319 .ret_type = RET_INTEGER,
2320 .arg1_type = ARG_PTR_TO_CTX,
2321 .arg2_type = ARG_PTR_TO_STACK,
2322 .arg3_type = ARG_CONST_STACK_SIZE,
2323 };
2324
2325 static const struct bpf_func_proto *
2326 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
2327 {
2328 if (!md_dst) {
2329 /* Race is not possible, since it's called from verifier
2330 * that is holding verifier mutex.
2331 */
2332 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
2333 GFP_KERNEL);
2334 if (!md_dst)
2335 return NULL;
2336 }
2337
2338 switch (which) {
2339 case BPF_FUNC_skb_set_tunnel_key:
2340 return &bpf_skb_set_tunnel_key_proto;
2341 case BPF_FUNC_skb_set_tunnel_opt:
2342 return &bpf_skb_set_tunnel_opt_proto;
2343 default:
2344 return NULL;
2345 }
2346 }
2347
2348 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
2349 u32, idx)
2350 {
2351 struct bpf_array *array = container_of(map, struct bpf_array, map);
2352 struct cgroup *cgrp;
2353 struct sock *sk;
2354
2355 sk = skb->sk;
2356 if (!sk || !sk_fullsock(sk))
2357 return -ENOENT;
2358 if (unlikely(idx >= array->map.max_entries))
2359 return -E2BIG;
2360
2361 cgrp = READ_ONCE(array->ptrs[idx]);
2362 if (unlikely(!cgrp))
2363 return -EAGAIN;
2364
2365 return sk_under_cgroup_hierarchy(sk, cgrp);
2366 }
2367
2368 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
2369 .func = bpf_skb_under_cgroup,
2370 .gpl_only = false,
2371 .ret_type = RET_INTEGER,
2372 .arg1_type = ARG_PTR_TO_CTX,
2373 .arg2_type = ARG_CONST_MAP_PTR,
2374 .arg3_type = ARG_ANYTHING,
2375 };
2376
2377 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
2378 unsigned long off, unsigned long len)
2379 {
2380 memcpy(dst_buff, src_buff + off, len);
2381 return 0;
2382 }
2383
2384 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
2385 u64, flags, void *, meta, u64, meta_size)
2386 {
2387 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2388
2389 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2390 return -EINVAL;
2391 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
2392 return -EFAULT;
2393
2394 return bpf_event_output(map, flags, meta, meta_size, xdp, xdp_size,
2395 bpf_xdp_copy);
2396 }
2397
2398 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
2399 .func = bpf_xdp_event_output,
2400 .gpl_only = true,
2401 .ret_type = RET_INTEGER,
2402 .arg1_type = ARG_PTR_TO_CTX,
2403 .arg2_type = ARG_CONST_MAP_PTR,
2404 .arg3_type = ARG_ANYTHING,
2405 .arg4_type = ARG_PTR_TO_STACK,
2406 .arg5_type = ARG_CONST_STACK_SIZE,
2407 };
2408
2409 static const struct bpf_func_proto *
2410 sk_filter_func_proto(enum bpf_func_id func_id)
2411 {
2412 switch (func_id) {
2413 case BPF_FUNC_map_lookup_elem:
2414 return &bpf_map_lookup_elem_proto;
2415 case BPF_FUNC_map_update_elem:
2416 return &bpf_map_update_elem_proto;
2417 case BPF_FUNC_map_delete_elem:
2418 return &bpf_map_delete_elem_proto;
2419 case BPF_FUNC_get_prandom_u32:
2420 return &bpf_get_prandom_u32_proto;
2421 case BPF_FUNC_get_smp_processor_id:
2422 return &bpf_get_raw_smp_processor_id_proto;
2423 case BPF_FUNC_tail_call:
2424 return &bpf_tail_call_proto;
2425 case BPF_FUNC_ktime_get_ns:
2426 return &bpf_ktime_get_ns_proto;
2427 case BPF_FUNC_trace_printk:
2428 if (capable(CAP_SYS_ADMIN))
2429 return bpf_get_trace_printk_proto();
2430 default:
2431 return NULL;
2432 }
2433 }
2434
2435 static const struct bpf_func_proto *
2436 tc_cls_act_func_proto(enum bpf_func_id func_id)
2437 {
2438 switch (func_id) {
2439 case BPF_FUNC_skb_store_bytes:
2440 return &bpf_skb_store_bytes_proto;
2441 case BPF_FUNC_skb_load_bytes:
2442 return &bpf_skb_load_bytes_proto;
2443 case BPF_FUNC_csum_diff:
2444 return &bpf_csum_diff_proto;
2445 case BPF_FUNC_l3_csum_replace:
2446 return &bpf_l3_csum_replace_proto;
2447 case BPF_FUNC_l4_csum_replace:
2448 return &bpf_l4_csum_replace_proto;
2449 case BPF_FUNC_clone_redirect:
2450 return &bpf_clone_redirect_proto;
2451 case BPF_FUNC_get_cgroup_classid:
2452 return &bpf_get_cgroup_classid_proto;
2453 case BPF_FUNC_skb_vlan_push:
2454 return &bpf_skb_vlan_push_proto;
2455 case BPF_FUNC_skb_vlan_pop:
2456 return &bpf_skb_vlan_pop_proto;
2457 case BPF_FUNC_skb_change_proto:
2458 return &bpf_skb_change_proto_proto;
2459 case BPF_FUNC_skb_change_type:
2460 return &bpf_skb_change_type_proto;
2461 case BPF_FUNC_skb_change_tail:
2462 return &bpf_skb_change_tail_proto;
2463 case BPF_FUNC_skb_get_tunnel_key:
2464 return &bpf_skb_get_tunnel_key_proto;
2465 case BPF_FUNC_skb_set_tunnel_key:
2466 return bpf_get_skb_set_tunnel_proto(func_id);
2467 case BPF_FUNC_skb_get_tunnel_opt:
2468 return &bpf_skb_get_tunnel_opt_proto;
2469 case BPF_FUNC_skb_set_tunnel_opt:
2470 return bpf_get_skb_set_tunnel_proto(func_id);
2471 case BPF_FUNC_redirect:
2472 return &bpf_redirect_proto;
2473 case BPF_FUNC_get_route_realm:
2474 return &bpf_get_route_realm_proto;
2475 case BPF_FUNC_get_hash_recalc:
2476 return &bpf_get_hash_recalc_proto;
2477 case BPF_FUNC_perf_event_output:
2478 return &bpf_skb_event_output_proto;
2479 case BPF_FUNC_get_smp_processor_id:
2480 return &bpf_get_smp_processor_id_proto;
2481 case BPF_FUNC_skb_under_cgroup:
2482 return &bpf_skb_under_cgroup_proto;
2483 default:
2484 return sk_filter_func_proto(func_id);
2485 }
2486 }
2487
2488 static const struct bpf_func_proto *
2489 xdp_func_proto(enum bpf_func_id func_id)
2490 {
2491 switch (func_id) {
2492 case BPF_FUNC_perf_event_output:
2493 return &bpf_xdp_event_output_proto;
2494 default:
2495 return sk_filter_func_proto(func_id);
2496 }
2497 }
2498
2499 static bool __is_valid_access(int off, int size, enum bpf_access_type type)
2500 {
2501 if (off < 0 || off >= sizeof(struct __sk_buff))
2502 return false;
2503 /* The verifier guarantees that size > 0. */
2504 if (off % size != 0)
2505 return false;
2506 if (size != sizeof(__u32))
2507 return false;
2508
2509 return true;
2510 }
2511
2512 static bool sk_filter_is_valid_access(int off, int size,
2513 enum bpf_access_type type,
2514 enum bpf_reg_type *reg_type)
2515 {
2516 switch (off) {
2517 case offsetof(struct __sk_buff, tc_classid):
2518 case offsetof(struct __sk_buff, data):
2519 case offsetof(struct __sk_buff, data_end):
2520 return false;
2521 }
2522
2523 if (type == BPF_WRITE) {
2524 switch (off) {
2525 case offsetof(struct __sk_buff, cb[0]) ...
2526 offsetof(struct __sk_buff, cb[4]):
2527 break;
2528 default:
2529 return false;
2530 }
2531 }
2532
2533 return __is_valid_access(off, size, type);
2534 }
2535
2536 static bool tc_cls_act_is_valid_access(int off, int size,
2537 enum bpf_access_type type,
2538 enum bpf_reg_type *reg_type)
2539 {
2540 if (type == BPF_WRITE) {
2541 switch (off) {
2542 case offsetof(struct __sk_buff, mark):
2543 case offsetof(struct __sk_buff, tc_index):
2544 case offsetof(struct __sk_buff, priority):
2545 case offsetof(struct __sk_buff, cb[0]) ...
2546 offsetof(struct __sk_buff, cb[4]):
2547 case offsetof(struct __sk_buff, tc_classid):
2548 break;
2549 default:
2550 return false;
2551 }
2552 }
2553
2554 switch (off) {
2555 case offsetof(struct __sk_buff, data):
2556 *reg_type = PTR_TO_PACKET;
2557 break;
2558 case offsetof(struct __sk_buff, data_end):
2559 *reg_type = PTR_TO_PACKET_END;
2560 break;
2561 }
2562
2563 return __is_valid_access(off, size, type);
2564 }
2565
2566 static bool __is_valid_xdp_access(int off, int size,
2567 enum bpf_access_type type)
2568 {
2569 if (off < 0 || off >= sizeof(struct xdp_md))
2570 return false;
2571 if (off % size != 0)
2572 return false;
2573 if (size != sizeof(__u32))
2574 return false;
2575
2576 return true;
2577 }
2578
2579 static bool xdp_is_valid_access(int off, int size,
2580 enum bpf_access_type type,
2581 enum bpf_reg_type *reg_type)
2582 {
2583 if (type == BPF_WRITE)
2584 return false;
2585
2586 switch (off) {
2587 case offsetof(struct xdp_md, data):
2588 *reg_type = PTR_TO_PACKET;
2589 break;
2590 case offsetof(struct xdp_md, data_end):
2591 *reg_type = PTR_TO_PACKET_END;
2592 break;
2593 }
2594
2595 return __is_valid_xdp_access(off, size, type);
2596 }
2597
2598 void bpf_warn_invalid_xdp_action(u32 act)
2599 {
2600 WARN_ONCE(1, "Illegal XDP return value %u, expect packet loss\n", act);
2601 }
2602 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
2603
2604 static u32 sk_filter_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2605 int src_reg, int ctx_off,
2606 struct bpf_insn *insn_buf,
2607 struct bpf_prog *prog)
2608 {
2609 struct bpf_insn *insn = insn_buf;
2610
2611 switch (ctx_off) {
2612 case offsetof(struct __sk_buff, len):
2613 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
2614
2615 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2616 offsetof(struct sk_buff, len));
2617 break;
2618
2619 case offsetof(struct __sk_buff, protocol):
2620 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
2621
2622 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2623 offsetof(struct sk_buff, protocol));
2624 break;
2625
2626 case offsetof(struct __sk_buff, vlan_proto):
2627 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
2628
2629 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2630 offsetof(struct sk_buff, vlan_proto));
2631 break;
2632
2633 case offsetof(struct __sk_buff, priority):
2634 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
2635
2636 if (type == BPF_WRITE)
2637 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2638 offsetof(struct sk_buff, priority));
2639 else
2640 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2641 offsetof(struct sk_buff, priority));
2642 break;
2643
2644 case offsetof(struct __sk_buff, ingress_ifindex):
2645 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
2646
2647 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2648 offsetof(struct sk_buff, skb_iif));
2649 break;
2650
2651 case offsetof(struct __sk_buff, ifindex):
2652 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
2653
2654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
2655 dst_reg, src_reg,
2656 offsetof(struct sk_buff, dev));
2657 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
2658 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
2659 offsetof(struct net_device, ifindex));
2660 break;
2661
2662 case offsetof(struct __sk_buff, hash):
2663 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
2664
2665 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2666 offsetof(struct sk_buff, hash));
2667 break;
2668
2669 case offsetof(struct __sk_buff, mark):
2670 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
2671
2672 if (type == BPF_WRITE)
2673 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2674 offsetof(struct sk_buff, mark));
2675 else
2676 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2677 offsetof(struct sk_buff, mark));
2678 break;
2679
2680 case offsetof(struct __sk_buff, pkt_type):
2681 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
2682
2683 case offsetof(struct __sk_buff, queue_mapping):
2684 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
2685
2686 case offsetof(struct __sk_buff, vlan_present):
2687 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
2688 dst_reg, src_reg, insn);
2689
2690 case offsetof(struct __sk_buff, vlan_tci):
2691 return convert_skb_access(SKF_AD_VLAN_TAG,
2692 dst_reg, src_reg, insn);
2693
2694 case offsetof(struct __sk_buff, cb[0]) ...
2695 offsetof(struct __sk_buff, cb[4]):
2696 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
2697
2698 prog->cb_access = 1;
2699 ctx_off -= offsetof(struct __sk_buff, cb[0]);
2700 ctx_off += offsetof(struct sk_buff, cb);
2701 ctx_off += offsetof(struct qdisc_skb_cb, data);
2702 if (type == BPF_WRITE)
2703 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2704 else
2705 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2706 break;
2707
2708 case offsetof(struct __sk_buff, tc_classid):
2709 ctx_off -= offsetof(struct __sk_buff, tc_classid);
2710 ctx_off += offsetof(struct sk_buff, cb);
2711 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
2712 if (type == BPF_WRITE)
2713 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2714 else
2715 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2716 break;
2717
2718 case offsetof(struct __sk_buff, data):
2719 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
2720 dst_reg, src_reg,
2721 offsetof(struct sk_buff, data));
2722 break;
2723
2724 case offsetof(struct __sk_buff, data_end):
2725 ctx_off -= offsetof(struct __sk_buff, data_end);
2726 ctx_off += offsetof(struct sk_buff, cb);
2727 ctx_off += offsetof(struct bpf_skb_data_end, data_end);
2728 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), dst_reg, src_reg,
2729 ctx_off);
2730 break;
2731
2732 case offsetof(struct __sk_buff, tc_index):
2733 #ifdef CONFIG_NET_SCHED
2734 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
2735
2736 if (type == BPF_WRITE)
2737 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
2738 offsetof(struct sk_buff, tc_index));
2739 else
2740 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2741 offsetof(struct sk_buff, tc_index));
2742 break;
2743 #else
2744 if (type == BPF_WRITE)
2745 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
2746 else
2747 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
2748 break;
2749 #endif
2750 }
2751
2752 return insn - insn_buf;
2753 }
2754
2755 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2756 int src_reg, int ctx_off,
2757 struct bpf_insn *insn_buf,
2758 struct bpf_prog *prog)
2759 {
2760 struct bpf_insn *insn = insn_buf;
2761
2762 switch (ctx_off) {
2763 case offsetof(struct __sk_buff, ifindex):
2764 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
2765
2766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
2767 dst_reg, src_reg,
2768 offsetof(struct sk_buff, dev));
2769 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
2770 offsetof(struct net_device, ifindex));
2771 break;
2772 default:
2773 return sk_filter_convert_ctx_access(type, dst_reg, src_reg,
2774 ctx_off, insn_buf, prog);
2775 }
2776
2777 return insn - insn_buf;
2778 }
2779
2780 static u32 xdp_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2781 int src_reg, int ctx_off,
2782 struct bpf_insn *insn_buf,
2783 struct bpf_prog *prog)
2784 {
2785 struct bpf_insn *insn = insn_buf;
2786
2787 switch (ctx_off) {
2788 case offsetof(struct xdp_md, data):
2789 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
2790 dst_reg, src_reg,
2791 offsetof(struct xdp_buff, data));
2792 break;
2793 case offsetof(struct xdp_md, data_end):
2794 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
2795 dst_reg, src_reg,
2796 offsetof(struct xdp_buff, data_end));
2797 break;
2798 }
2799
2800 return insn - insn_buf;
2801 }
2802
2803 static const struct bpf_verifier_ops sk_filter_ops = {
2804 .get_func_proto = sk_filter_func_proto,
2805 .is_valid_access = sk_filter_is_valid_access,
2806 .convert_ctx_access = sk_filter_convert_ctx_access,
2807 };
2808
2809 static const struct bpf_verifier_ops tc_cls_act_ops = {
2810 .get_func_proto = tc_cls_act_func_proto,
2811 .is_valid_access = tc_cls_act_is_valid_access,
2812 .convert_ctx_access = tc_cls_act_convert_ctx_access,
2813 };
2814
2815 static const struct bpf_verifier_ops xdp_ops = {
2816 .get_func_proto = xdp_func_proto,
2817 .is_valid_access = xdp_is_valid_access,
2818 .convert_ctx_access = xdp_convert_ctx_access,
2819 };
2820
2821 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
2822 .ops = &sk_filter_ops,
2823 .type = BPF_PROG_TYPE_SOCKET_FILTER,
2824 };
2825
2826 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
2827 .ops = &tc_cls_act_ops,
2828 .type = BPF_PROG_TYPE_SCHED_CLS,
2829 };
2830
2831 static struct bpf_prog_type_list sched_act_type __read_mostly = {
2832 .ops = &tc_cls_act_ops,
2833 .type = BPF_PROG_TYPE_SCHED_ACT,
2834 };
2835
2836 static struct bpf_prog_type_list xdp_type __read_mostly = {
2837 .ops = &xdp_ops,
2838 .type = BPF_PROG_TYPE_XDP,
2839 };
2840
2841 static int __init register_sk_filter_ops(void)
2842 {
2843 bpf_register_prog_type(&sk_filter_type);
2844 bpf_register_prog_type(&sched_cls_type);
2845 bpf_register_prog_type(&sched_act_type);
2846 bpf_register_prog_type(&xdp_type);
2847
2848 return 0;
2849 }
2850 late_initcall(register_sk_filter_ops);
2851
2852 int sk_detach_filter(struct sock *sk)
2853 {
2854 int ret = -ENOENT;
2855 struct sk_filter *filter;
2856
2857 if (sock_flag(sk, SOCK_FILTER_LOCKED))
2858 return -EPERM;
2859
2860 filter = rcu_dereference_protected(sk->sk_filter,
2861 lockdep_sock_is_held(sk));
2862 if (filter) {
2863 RCU_INIT_POINTER(sk->sk_filter, NULL);
2864 sk_filter_uncharge(sk, filter);
2865 ret = 0;
2866 }
2867
2868 return ret;
2869 }
2870 EXPORT_SYMBOL_GPL(sk_detach_filter);
2871
2872 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
2873 unsigned int len)
2874 {
2875 struct sock_fprog_kern *fprog;
2876 struct sk_filter *filter;
2877 int ret = 0;
2878
2879 lock_sock(sk);
2880 filter = rcu_dereference_protected(sk->sk_filter,
2881 lockdep_sock_is_held(sk));
2882 if (!filter)
2883 goto out;
2884
2885 /* We're copying the filter that has been originally attached,
2886 * so no conversion/decode needed anymore. eBPF programs that
2887 * have no original program cannot be dumped through this.
2888 */
2889 ret = -EACCES;
2890 fprog = filter->prog->orig_prog;
2891 if (!fprog)
2892 goto out;
2893
2894 ret = fprog->len;
2895 if (!len)
2896 /* User space only enquires number of filter blocks. */
2897 goto out;
2898
2899 ret = -EINVAL;
2900 if (len < fprog->len)
2901 goto out;
2902
2903 ret = -EFAULT;
2904 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
2905 goto out;
2906
2907 /* Instead of bytes, the API requests to return the number
2908 * of filter blocks.
2909 */
2910 ret = fprog->len;
2911 out:
2912 release_sock(sk);
2913 return ret;
2914 }
Linux kernel - Deliverables
This page took 0.126183 seconds and 5 git commands to generate.