?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Input: synaptics-rmi4 - register F03 port as pass-through serio
[linux-2.6/btrfs-unstable.git] / kernel / bpf / core.c
blobdedf367f59bba529ded5603d61782bccb80782cb
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/filter.h>
25 #include <linux/skbuff.h>
26 #include <linux/vmalloc.h>
27 #include <linux/random.h>
28 #include <linux/moduleloader.h>
29 #include <linux/bpf.h>
30 #include <linux/frame.h>
31 #include <linux/rbtree_latch.h>
32 #include <linux/kallsyms.h>
33 #include <linux/rcupdate.h>
34
35 #include <asm/unaligned.h>
36
37 /* Registers */
38 #define BPF_R0 regs[BPF_REG_0]
39 #define BPF_R1 regs[BPF_REG_1]
40 #define BPF_R2 regs[BPF_REG_2]
41 #define BPF_R3 regs[BPF_REG_3]
42 #define BPF_R4 regs[BPF_REG_4]
43 #define BPF_R5 regs[BPF_REG_5]
44 #define BPF_R6 regs[BPF_REG_6]
45 #define BPF_R7 regs[BPF_REG_7]
46 #define BPF_R8 regs[BPF_REG_8]
47 #define BPF_R9 regs[BPF_REG_9]
48 #define BPF_R10 regs[BPF_REG_10]
49
50 /* Named registers */
51 #define DST regs[insn->dst_reg]
52 #define SRC regs[insn->src_reg]
53 #define FP regs[BPF_REG_FP]
54 #define ARG1 regs[BPF_REG_ARG1]
55 #define CTX regs[BPF_REG_CTX]
56 #define IMM insn->imm
57
58 /* No hurry in this branch
59 *
60 * Exported for the bpf jit load helper.
61 */
62 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
63 {
64 u8 *ptr = NULL;
65
66 if (k >= SKF_NET_OFF)
67 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
68 else if (k >= SKF_LL_OFF)
69 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
70
71 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
72 return ptr;
73
74 return NULL;
75 }
76
77 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
78 {
79 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
80 struct bpf_prog_aux *aux;
81 struct bpf_prog *fp;
82
83 size = round_up(size, PAGE_SIZE);
84 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
85 if (fp == NULL)
86 return NULL;
87
88 kmemcheck_annotate_bitfield(fp, meta);
89
90 aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
91 if (aux == NULL) {
92 vfree(fp);
93 return NULL;
94 }
95
96 fp->pages = size / PAGE_SIZE;
97 fp->aux = aux;
98 fp->aux->prog = fp;
99
100 INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode);
101
102 return fp;
103 }
104 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
105
106 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
107 gfp_t gfp_extra_flags)
108 {
109 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
110 struct bpf_prog *fp;
111 u32 pages, delta;
112 int ret;
113
114 BUG_ON(fp_old == NULL);
115
116 size = round_up(size, PAGE_SIZE);
117 pages = size / PAGE_SIZE;
118 if (pages <= fp_old->pages)
119 return fp_old;
120
121 delta = pages - fp_old->pages;
122 ret = __bpf_prog_charge(fp_old->aux->user, delta);
123 if (ret)
124 return NULL;
125
126 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
127 if (fp == NULL) {
128 __bpf_prog_uncharge(fp_old->aux->user, delta);
129 } else {
130 kmemcheck_annotate_bitfield(fp, meta);
131
132 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
133 fp->pages = pages;
134 fp->aux->prog = fp;
135
136 /* We keep fp->aux from fp_old around in the new
137 * reallocated structure.
138 */
139 fp_old->aux = NULL;
140 __bpf_prog_free(fp_old);
141 }
142
143 return fp;
144 }
145
146 void __bpf_prog_free(struct bpf_prog *fp)
147 {
148 kfree(fp->aux);
149 vfree(fp);
150 }
151
152 int bpf_prog_calc_tag(struct bpf_prog *fp)
153 {
154 const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
155 u32 raw_size = bpf_prog_tag_scratch_size(fp);
156 u32 digest[SHA_DIGEST_WORDS];
157 u32 ws[SHA_WORKSPACE_WORDS];
158 u32 i, bsize, psize, blocks;
159 struct bpf_insn *dst;
160 bool was_ld_map;
161 u8 *raw, *todo;
162 __be32 *result;
163 __be64 *bits;
164
165 raw = vmalloc(raw_size);
166 if (!raw)
167 return -ENOMEM;
168
169 sha_init(digest);
170 memset(ws, 0, sizeof(ws));
171
172 /* We need to take out the map fd for the digest calculation
173 * since they are unstable from user space side.
174 */
175 dst = (void *)raw;
176 for (i = 0, was_ld_map = false; i < fp->len; i++) {
177 dst[i] = fp->insnsi[i];
178 if (!was_ld_map &&
179 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
180 dst[i].src_reg == BPF_PSEUDO_MAP_FD) {
181 was_ld_map = true;
182 dst[i].imm = 0;
183 } else if (was_ld_map &&
184 dst[i].code == 0 &&
185 dst[i].dst_reg == 0 &&
186 dst[i].src_reg == 0 &&
187 dst[i].off == 0) {
188 was_ld_map = false;
189 dst[i].imm = 0;
190 } else {
191 was_ld_map = false;
192 }
193 }
194
195 psize = bpf_prog_insn_size(fp);
196 memset(&raw[psize], 0, raw_size - psize);
197 raw[psize++] = 0x80;
198
199 bsize = round_up(psize, SHA_MESSAGE_BYTES);
200 blocks = bsize / SHA_MESSAGE_BYTES;
201 todo = raw;
202 if (bsize - psize >= sizeof(__be64)) {
203 bits = (__be64 *)(todo + bsize - sizeof(__be64));
204 } else {
205 bits = (__be64 *)(todo + bsize + bits_offset);
206 blocks++;
207 }
208 *bits = cpu_to_be64((psize - 1) << 3);
209
210 while (blocks--) {
211 sha_transform(digest, todo, ws);
212 todo += SHA_MESSAGE_BYTES;
213 }
214
215 result = (__force __be32 *)digest;
216 for (i = 0; i < SHA_DIGEST_WORDS; i++)
217 result[i] = cpu_to_be32(digest[i]);
218 memcpy(fp->tag, result, sizeof(fp->tag));
219
220 vfree(raw);
221 return 0;
222 }
223
224 static bool bpf_is_jmp_and_has_target(const struct bpf_insn *insn)
225 {
226 return BPF_CLASS(insn->code) == BPF_JMP &&
227 /* Call and Exit are both special jumps with no
228 * target inside the BPF instruction image.
229 */
230 BPF_OP(insn->code) != BPF_CALL &&
231 BPF_OP(insn->code) != BPF_EXIT;
232 }
233
234 static void bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta)
235 {
236 struct bpf_insn *insn = prog->insnsi;
237 u32 i, insn_cnt = prog->len;
238
239 for (i = 0; i < insn_cnt; i++, insn++) {
240 if (!bpf_is_jmp_and_has_target(insn))
241 continue;
242
243 /* Adjust offset of jmps if we cross boundaries. */
244 if (i < pos && i + insn->off + 1 > pos)
245 insn->off += delta;
246 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
247 insn->off -= delta;
248 }
249 }
250
251 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
252 const struct bpf_insn *patch, u32 len)
253 {
254 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
255 struct bpf_prog *prog_adj;
256
257 /* Since our patchlet doesn't expand the image, we're done. */
258 if (insn_delta == 0) {
259 memcpy(prog->insnsi + off, patch, sizeof(*patch));
260 return prog;
261 }
262
263 insn_adj_cnt = prog->len + insn_delta;
264
265 /* Several new instructions need to be inserted. Make room
266 * for them. Likely, there's no need for a new allocation as
267 * last page could have large enough tailroom.
268 */
269 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
270 GFP_USER);
271 if (!prog_adj)
272 return NULL;
273
274 prog_adj->len = insn_adj_cnt;
275
276 /* Patching happens in 3 steps:
277 *
278 * 1) Move over tail of insnsi from next instruction onwards,
279 * so we can patch the single target insn with one or more
280 * new ones (patching is always from 1 to n insns, n > 0).
281 * 2) Inject new instructions at the target location.
282 * 3) Adjust branch offsets if necessary.
283 */
284 insn_rest = insn_adj_cnt - off - len;
285
286 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
287 sizeof(*patch) * insn_rest);
288 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
289
290 bpf_adj_branches(prog_adj, off, insn_delta);
291
292 return prog_adj;
293 }
294
295 #ifdef CONFIG_BPF_JIT
296 static __always_inline void
297 bpf_get_prog_addr_region(const struct bpf_prog *prog,
298 unsigned long *symbol_start,
299 unsigned long *symbol_end)
300 {
301 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
302 unsigned long addr = (unsigned long)hdr;
303
304 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
305
306 *symbol_start = addr;
307 *symbol_end = addr + hdr->pages * PAGE_SIZE;
308 }
309
310 static void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
311 {
312 BUILD_BUG_ON(sizeof("bpf_prog_") +
313 sizeof(prog->tag) * 2 + 1 > KSYM_NAME_LEN);
314
315 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
316 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
317 *sym = 0;
318 }
319
320 static __always_inline unsigned long
321 bpf_get_prog_addr_start(struct latch_tree_node *n)
322 {
323 unsigned long symbol_start, symbol_end;
324 const struct bpf_prog_aux *aux;
325
326 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
327 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
328
329 return symbol_start;
330 }
331
332 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
333 struct latch_tree_node *b)
334 {
335 return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
336 }
337
338 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
339 {
340 unsigned long val = (unsigned long)key;
341 unsigned long symbol_start, symbol_end;
342 const struct bpf_prog_aux *aux;
343
344 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
345 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
346
347 if (val < symbol_start)
348 return -1;
349 if (val >= symbol_end)
350 return 1;
351
352 return 0;
353 }
354
355 static const struct latch_tree_ops bpf_tree_ops = {
356 .less = bpf_tree_less,
357 .comp = bpf_tree_comp,
358 };
359
360 static DEFINE_SPINLOCK(bpf_lock);
361 static LIST_HEAD(bpf_kallsyms);
362 static struct latch_tree_root bpf_tree __cacheline_aligned;
363
364 int bpf_jit_kallsyms __read_mostly;
365
366 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
367 {
368 WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
369 list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
370 latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
371 }
372
373 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
374 {
375 if (list_empty(&aux->ksym_lnode))
376 return;
377
378 latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
379 list_del_rcu(&aux->ksym_lnode);
380 }
381
382 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
383 {
384 return fp->jited && !bpf_prog_was_classic(fp);
385 }
386
387 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
388 {
389 return list_empty(&fp->aux->ksym_lnode) ||
390 fp->aux->ksym_lnode.prev == LIST_POISON2;
391 }
392
393 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
394 {
395 if (!bpf_prog_kallsyms_candidate(fp) ||
396 !capable(CAP_SYS_ADMIN))
397 return;
398
399 spin_lock_bh(&bpf_lock);
400 bpf_prog_ksym_node_add(fp->aux);
401 spin_unlock_bh(&bpf_lock);
402 }
403
404 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
405 {
406 if (!bpf_prog_kallsyms_candidate(fp))
407 return;
408
409 spin_lock_bh(&bpf_lock);
410 bpf_prog_ksym_node_del(fp->aux);
411 spin_unlock_bh(&bpf_lock);
412 }
413
414 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
415 {
416 struct latch_tree_node *n;
417
418 if (!bpf_jit_kallsyms_enabled())
419 return NULL;
420
421 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
422 return n ?
423 container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
424 NULL;
425 }
426
427 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
428 unsigned long *off, char *sym)
429 {
430 unsigned long symbol_start, symbol_end;
431 struct bpf_prog *prog;
432 char *ret = NULL;
433
434 rcu_read_lock();
435 prog = bpf_prog_kallsyms_find(addr);
436 if (prog) {
437 bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
438 bpf_get_prog_name(prog, sym);
439
440 ret = sym;
441 if (size)
442 *size = symbol_end - symbol_start;
443 if (off)
444 *off = addr - symbol_start;
445 }
446 rcu_read_unlock();
447
448 return ret;
449 }
450
451 bool is_bpf_text_address(unsigned long addr)
452 {
453 bool ret;
454
455 rcu_read_lock();
456 ret = bpf_prog_kallsyms_find(addr) != NULL;
457 rcu_read_unlock();
458
459 return ret;
460 }
461
462 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
463 char *sym)
464 {
465 unsigned long symbol_start, symbol_end;
466 struct bpf_prog_aux *aux;
467 unsigned int it = 0;
468 int ret = -ERANGE;
469
470 if (!bpf_jit_kallsyms_enabled())
471 return ret;
472
473 rcu_read_lock();
474 list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
475 if (it++ != symnum)
476 continue;
477
478 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
479 bpf_get_prog_name(aux->prog, sym);
480
481 *value = symbol_start;
482 *type = BPF_SYM_ELF_TYPE;
483
484 ret = 0;
485 break;
486 }
487 rcu_read_unlock();
488
489 return ret;
490 }
491
492 struct bpf_binary_header *
493 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
494 unsigned int alignment,
495 bpf_jit_fill_hole_t bpf_fill_ill_insns)
496 {
497 struct bpf_binary_header *hdr;
498 unsigned int size, hole, start;
499
500 /* Most of BPF filters are really small, but if some of them
501 * fill a page, allow at least 128 extra bytes to insert a
502 * random section of illegal instructions.
503 */
504 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
505 hdr = module_alloc(size);
506 if (hdr == NULL)
507 return NULL;
508
509 /* Fill space with illegal/arch-dep instructions. */
510 bpf_fill_ill_insns(hdr, size);
511
512 hdr->pages = size / PAGE_SIZE;
513 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
514 PAGE_SIZE - sizeof(*hdr));
515 start = (get_random_int() % hole) & ~(alignment - 1);
516
517 /* Leave a random number of instructions before BPF code. */
518 *image_ptr = &hdr->image[start];
519
520 return hdr;
521 }
522
523 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
524 {
525 module_memfree(hdr);
526 }
527
528 /* This symbol is only overridden by archs that have different
529 * requirements than the usual eBPF JITs, f.e. when they only
530 * implement cBPF JIT, do not set images read-only, etc.
531 */
532 void __weak bpf_jit_free(struct bpf_prog *fp)
533 {
534 if (fp->jited) {
535 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
536
537 bpf_jit_binary_unlock_ro(hdr);
538 bpf_jit_binary_free(hdr);
539
540 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
541 }
542
543 bpf_prog_unlock_free(fp);
544 }
545
546 int bpf_jit_harden __read_mostly;
547
548 static int bpf_jit_blind_insn(const struct bpf_insn *from,
549 const struct bpf_insn *aux,
550 struct bpf_insn *to_buff)
551 {
552 struct bpf_insn *to = to_buff;
553 u32 imm_rnd = get_random_int();
554 s16 off;
555
556 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
557 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
558
559 if (from->imm == 0 &&
560 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
561 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
562 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
563 goto out;
564 }
565
566 switch (from->code) {
567 case BPF_ALU | BPF_ADD | BPF_K:
568 case BPF_ALU | BPF_SUB | BPF_K:
569 case BPF_ALU | BPF_AND | BPF_K:
570 case BPF_ALU | BPF_OR | BPF_K:
571 case BPF_ALU | BPF_XOR | BPF_K:
572 case BPF_ALU | BPF_MUL | BPF_K:
573 case BPF_ALU | BPF_MOV | BPF_K:
574 case BPF_ALU | BPF_DIV | BPF_K:
575 case BPF_ALU | BPF_MOD | BPF_K:
576 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
577 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
578 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
579 break;
580
581 case BPF_ALU64 | BPF_ADD | BPF_K:
582 case BPF_ALU64 | BPF_SUB | BPF_K:
583 case BPF_ALU64 | BPF_AND | BPF_K:
584 case BPF_ALU64 | BPF_OR | BPF_K:
585 case BPF_ALU64 | BPF_XOR | BPF_K:
586 case BPF_ALU64 | BPF_MUL | BPF_K:
587 case BPF_ALU64 | BPF_MOV | BPF_K:
588 case BPF_ALU64 | BPF_DIV | BPF_K:
589 case BPF_ALU64 | BPF_MOD | BPF_K:
590 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
591 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
592 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
593 break;
594
595 case BPF_JMP | BPF_JEQ | BPF_K:
596 case BPF_JMP | BPF_JNE | BPF_K:
597 case BPF_JMP | BPF_JGT | BPF_K:
598 case BPF_JMP | BPF_JGE | BPF_K:
599 case BPF_JMP | BPF_JSGT | BPF_K:
600 case BPF_JMP | BPF_JSGE | BPF_K:
601 case BPF_JMP | BPF_JSET | BPF_K:
602 /* Accommodate for extra offset in case of a backjump. */
603 off = from->off;
604 if (off < 0)
605 off -= 2;
606 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
607 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
608 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
609 break;
610
611 case BPF_LD | BPF_ABS | BPF_W:
612 case BPF_LD | BPF_ABS | BPF_H:
613 case BPF_LD | BPF_ABS | BPF_B:
614 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
615 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
616 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
617 break;
618
619 case BPF_LD | BPF_IND | BPF_W:
620 case BPF_LD | BPF_IND | BPF_H:
621 case BPF_LD | BPF_IND | BPF_B:
622 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
623 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
624 *to++ = BPF_ALU32_REG(BPF_ADD, BPF_REG_AX, from->src_reg);
625 *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0);
626 break;
627
628 case BPF_LD | BPF_IMM | BPF_DW:
629 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
630 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
631 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
632 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
633 break;
634 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
635 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
636 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
637 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
638 break;
639
640 case BPF_ST | BPF_MEM | BPF_DW:
641 case BPF_ST | BPF_MEM | BPF_W:
642 case BPF_ST | BPF_MEM | BPF_H:
643 case BPF_ST | BPF_MEM | BPF_B:
644 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
645 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
646 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
647 break;
648 }
649 out:
650 return to - to_buff;
651 }
652
653 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
654 gfp_t gfp_extra_flags)
655 {
656 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
657 struct bpf_prog *fp;
658
659 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
660 if (fp != NULL) {
661 kmemcheck_annotate_bitfield(fp, meta);
662
663 /* aux->prog still points to the fp_other one, so
664 * when promoting the clone to the real program,
665 * this still needs to be adapted.
666 */
667 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
668 }
669
670 return fp;
671 }
672
673 static void bpf_prog_clone_free(struct bpf_prog *fp)
674 {
675 /* aux was stolen by the other clone, so we cannot free
676 * it from this path! It will be freed eventually by the
677 * other program on release.
678 *
679 * At this point, we don't need a deferred release since
680 * clone is guaranteed to not be locked.
681 */
682 fp->aux = NULL;
683 __bpf_prog_free(fp);
684 }
685
686 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
687 {
688 /* We have to repoint aux->prog to self, as we don't
689 * know whether fp here is the clone or the original.
690 */
691 fp->aux->prog = fp;
692 bpf_prog_clone_free(fp_other);
693 }
694
695 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
696 {
697 struct bpf_insn insn_buff[16], aux[2];
698 struct bpf_prog *clone, *tmp;
699 int insn_delta, insn_cnt;
700 struct bpf_insn *insn;
701 int i, rewritten;
702
703 if (!bpf_jit_blinding_enabled())
704 return prog;
705
706 clone = bpf_prog_clone_create(prog, GFP_USER);
707 if (!clone)
708 return ERR_PTR(-ENOMEM);
709
710 insn_cnt = clone->len;
711 insn = clone->insnsi;
712
713 for (i = 0; i < insn_cnt; i++, insn++) {
714 /* We temporarily need to hold the original ld64 insn
715 * so that we can still access the first part in the
716 * second blinding run.
717 */
718 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
719 insn[1].code == 0)
720 memcpy(aux, insn, sizeof(aux));
721
722 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff);
723 if (!rewritten)
724 continue;
725
726 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
727 if (!tmp) {
728 /* Patching may have repointed aux->prog during
729 * realloc from the original one, so we need to
730 * fix it up here on error.
731 */
732 bpf_jit_prog_release_other(prog, clone);
733 return ERR_PTR(-ENOMEM);
734 }
735
736 clone = tmp;
737 insn_delta = rewritten - 1;
738
739 /* Walk new program and skip insns we just inserted. */
740 insn = clone->insnsi + i + insn_delta;
741 insn_cnt += insn_delta;
742 i += insn_delta;
743 }
744
745 return clone;
746 }
747 #endif /* CONFIG_BPF_JIT */
748
749 /* Base function for offset calculation. Needs to go into .text section,
750 * therefore keeping it non-static as well; will also be used by JITs
751 * anyway later on, so do not let the compiler omit it.
752 */
753 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
754 {
755 return 0;
756 }
757 EXPORT_SYMBOL_GPL(__bpf_call_base);
758
759 /**
760 * __bpf_prog_run - run eBPF program on a given context
761 * @ctx: is the data we are operating on
762 * @insn: is the array of eBPF instructions
763 *
764 * Decode and execute eBPF instructions.
765 */
766 static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
767 {
768 u64 stack[MAX_BPF_STACK / sizeof(u64)];
769 u64 regs[MAX_BPF_REG], tmp;
770 static const void *jumptable[256] = {
771 [0 ... 255] = &&default_label,
772 /* Now overwrite non-defaults ... */
773 /* 32 bit ALU operations */
774 [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X,
775 [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K,
776 [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X,
777 [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K,
778 [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X,
779 [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K,
780 [BPF_ALU | BPF_OR | BPF_X] = &&ALU_OR_X,
781 [BPF_ALU | BPF_OR | BPF_K] = &&ALU_OR_K,
782 [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X,
783 [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K,
784 [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X,
785 [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K,
786 [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X,
787 [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K,
788 [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X,
789 [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K,
790 [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X,
791 [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K,
792 [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X,
793 [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K,
794 [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X,
795 [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K,
796 [BPF_ALU | BPF_NEG] = &&ALU_NEG,
797 [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE,
798 [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE,
799 /* 64 bit ALU operations */
800 [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X,
801 [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K,
802 [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X,
803 [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K,
804 [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X,
805 [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K,
806 [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X,
807 [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K,
808 [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X,
809 [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K,
810 [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X,
811 [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K,
812 [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X,
813 [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K,
814 [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X,
815 [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K,
816 [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X,
817 [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K,
818 [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X,
819 [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K,
820 [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X,
821 [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K,
822 [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X,
823 [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K,
824 [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG,
825 /* Call instruction */
826 [BPF_JMP | BPF_CALL] = &&JMP_CALL,
827 [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL,
828 /* Jumps */
829 [BPF_JMP | BPF_JA] = &&JMP_JA,
830 [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X,
831 [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K,
832 [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X,
833 [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K,
834 [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X,
835 [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K,
836 [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X,
837 [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K,
838 [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X,
839 [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K,
840 [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X,
841 [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K,
842 [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X,
843 [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K,
844 /* Program return */
845 [BPF_JMP | BPF_EXIT] = &&JMP_EXIT,
846 /* Store instructions */
847 [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B,
848 [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H,
849 [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W,
850 [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW,
851 [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W,
852 [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW,
853 [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B,
854 [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H,
855 [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W,
856 [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW,
857 /* Load instructions */
858 [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B,
859 [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H,
860 [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W,
861 [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW,
862 [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W,
863 [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H,
864 [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B,
865 [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W,
866 [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H,
867 [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B,
868 [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW,
869 };
870 u32 tail_call_cnt = 0;
871 void *ptr;
872 int off;
873
874 #define CONT ({ insn++; goto select_insn; })
875 #define CONT_JMP ({ insn++; goto select_insn; })
876
877 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
878 ARG1 = (u64) (unsigned long) ctx;
879
880 select_insn:
881 goto *jumptable[insn->code];
882
883 /* ALU */
884 #define ALU(OPCODE, OP) \
885 ALU64_##OPCODE##_X: \
886 DST = DST OP SRC; \
887 CONT; \
888 ALU_##OPCODE##_X: \
889 DST = (u32) DST OP (u32) SRC; \
890 CONT; \
891 ALU64_##OPCODE##_K: \
892 DST = DST OP IMM; \
893 CONT; \
894 ALU_##OPCODE##_K: \
895 DST = (u32) DST OP (u32) IMM; \
896 CONT;
897
898 ALU(ADD, +)
899 ALU(SUB, -)
900 ALU(AND, &)
901 ALU(OR, |)
902 ALU(LSH, <<)
903 ALU(RSH, >>)
904 ALU(XOR, ^)
905 ALU(MUL, *)
906 #undef ALU
907 ALU_NEG:
908 DST = (u32) -DST;
909 CONT;
910 ALU64_NEG:
911 DST = -DST;
912 CONT;
913 ALU_MOV_X:
914 DST = (u32) SRC;
915 CONT;
916 ALU_MOV_K:
917 DST = (u32) IMM;
918 CONT;
919 ALU64_MOV_X:
920 DST = SRC;
921 CONT;
922 ALU64_MOV_K:
923 DST = IMM;
924 CONT;
925 LD_IMM_DW:
926 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
927 insn++;
928 CONT;
929 ALU64_ARSH_X:
930 (*(s64 *) &DST) >>= SRC;
931 CONT;
932 ALU64_ARSH_K:
933 (*(s64 *) &DST) >>= IMM;
934 CONT;
935 ALU64_MOD_X:
936 if (unlikely(SRC == 0))
937 return 0;
938 div64_u64_rem(DST, SRC, &tmp);
939 DST = tmp;
940 CONT;
941 ALU_MOD_X:
942 if (unlikely(SRC == 0))
943 return 0;
944 tmp = (u32) DST;
945 DST = do_div(tmp, (u32) SRC);
946 CONT;
947 ALU64_MOD_K:
948 div64_u64_rem(DST, IMM, &tmp);
949 DST = tmp;
950 CONT;
951 ALU_MOD_K:
952 tmp = (u32) DST;
953 DST = do_div(tmp, (u32) IMM);
954 CONT;
955 ALU64_DIV_X:
956 if (unlikely(SRC == 0))
957 return 0;
958 DST = div64_u64(DST, SRC);
959 CONT;
960 ALU_DIV_X:
961 if (unlikely(SRC == 0))
962 return 0;
963 tmp = (u32) DST;
964 do_div(tmp, (u32) SRC);
965 DST = (u32) tmp;
966 CONT;
967 ALU64_DIV_K:
968 DST = div64_u64(DST, IMM);
969 CONT;
970 ALU_DIV_K:
971 tmp = (u32) DST;
972 do_div(tmp, (u32) IMM);
973 DST = (u32) tmp;
974 CONT;
975 ALU_END_TO_BE:
976 switch (IMM) {
977 case 16:
978 DST = (__force u16) cpu_to_be16(DST);
979 break;
980 case 32:
981 DST = (__force u32) cpu_to_be32(DST);
982 break;
983 case 64:
984 DST = (__force u64) cpu_to_be64(DST);
985 break;
986 }
987 CONT;
988 ALU_END_TO_LE:
989 switch (IMM) {
990 case 16:
991 DST = (__force u16) cpu_to_le16(DST);
992 break;
993 case 32:
994 DST = (__force u32) cpu_to_le32(DST);
995 break;
996 case 64:
997 DST = (__force u64) cpu_to_le64(DST);
998 break;
999 }
1000 CONT;
1001
1002 /* CALL */
1003 JMP_CALL:
1004 /* Function call scratches BPF_R1-BPF_R5 registers,
1005 * preserves BPF_R6-BPF_R9, and stores return value
1006 * into BPF_R0.
1007 */
1008 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1009 BPF_R4, BPF_R5);
1010 CONT;
1011
1012 JMP_TAIL_CALL: {
1013 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1014 struct bpf_array *array = container_of(map, struct bpf_array, map);
1015 struct bpf_prog *prog;
1016 u64 index = BPF_R3;
1017
1018 if (unlikely(index >= array->map.max_entries))
1019 goto out;
1020 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1021 goto out;
1022
1023 tail_call_cnt++;
1024
1025 prog = READ_ONCE(array->ptrs[index]);
1026 if (!prog)
1027 goto out;
1028
1029 /* ARG1 at this point is guaranteed to point to CTX from
1030 * the verifier side due to the fact that the tail call is
1031 * handeled like a helper, that is, bpf_tail_call_proto,
1032 * where arg1_type is ARG_PTR_TO_CTX.
1033 */
1034 insn = prog->insnsi;
1035 goto select_insn;
1036 out:
1037 CONT;
1038 }
1039 /* JMP */
1040 JMP_JA:
1041 insn += insn->off;
1042 CONT;
1043 JMP_JEQ_X:
1044 if (DST == SRC) {
1045 insn += insn->off;
1046 CONT_JMP;
1047 }
1048 CONT;
1049 JMP_JEQ_K:
1050 if (DST == IMM) {
1051 insn += insn->off;
1052 CONT_JMP;
1053 }
1054 CONT;
1055 JMP_JNE_X:
1056 if (DST != SRC) {
1057 insn += insn->off;
1058 CONT_JMP;
1059 }
1060 CONT;
1061 JMP_JNE_K:
1062 if (DST != IMM) {
1063 insn += insn->off;
1064 CONT_JMP;
1065 }
1066 CONT;
1067 JMP_JGT_X:
1068 if (DST > SRC) {
1069 insn += insn->off;
1070 CONT_JMP;
1071 }
1072 CONT;
1073 JMP_JGT_K:
1074 if (DST > IMM) {
1075 insn += insn->off;
1076 CONT_JMP;
1077 }
1078 CONT;
1079 JMP_JGE_X:
1080 if (DST >= SRC) {
1081 insn += insn->off;
1082 CONT_JMP;
1083 }
1084 CONT;
1085 JMP_JGE_K:
1086 if (DST >= IMM) {
1087 insn += insn->off;
1088 CONT_JMP;
1089 }
1090 CONT;
1091 JMP_JSGT_X:
1092 if (((s64) DST) > ((s64) SRC)) {
1093 insn += insn->off;
1094 CONT_JMP;
1095 }
1096 CONT;
1097 JMP_JSGT_K:
1098 if (((s64) DST) > ((s64) IMM)) {
1099 insn += insn->off;
1100 CONT_JMP;
1101 }
1102 CONT;
1103 JMP_JSGE_X:
1104 if (((s64) DST) >= ((s64) SRC)) {
1105 insn += insn->off;
1106 CONT_JMP;
1107 }
1108 CONT;
1109 JMP_JSGE_K:
1110 if (((s64) DST) >= ((s64) IMM)) {
1111 insn += insn->off;
1112 CONT_JMP;
1113 }
1114 CONT;
1115 JMP_JSET_X:
1116 if (DST & SRC) {
1117 insn += insn->off;
1118 CONT_JMP;
1119 }
1120 CONT;
1121 JMP_JSET_K:
1122 if (DST & IMM) {
1123 insn += insn->off;
1124 CONT_JMP;
1125 }
1126 CONT;
1127 JMP_EXIT:
1128 return BPF_R0;
1129
1130 /* STX and ST and LDX*/
1131 #define LDST(SIZEOP, SIZE) \
1132 STX_MEM_##SIZEOP: \
1133 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1134 CONT; \
1135 ST_MEM_##SIZEOP: \
1136 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1137 CONT; \
1138 LDX_MEM_##SIZEOP: \
1139 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1140 CONT;
1141
1142 LDST(B, u8)
1143 LDST(H, u16)
1144 LDST(W, u32)
1145 LDST(DW, u64)
1146 #undef LDST
1147 STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1148 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1149 (DST + insn->off));
1150 CONT;
1151 STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1152 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1153 (DST + insn->off));
1154 CONT;
1155 LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
1156 off = IMM;
1157 load_word:
1158 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are only
1159 * appearing in the programs where ctx == skb
1160 * (see may_access_skb() in the verifier). All programs
1161 * keep 'ctx' in regs[BPF_REG_CTX] == BPF_R6,
1162 * bpf_convert_filter() saves it in BPF_R6, internal BPF
1163 * verifier will check that BPF_R6 == ctx.
1164 *
1165 * BPF_ABS and BPF_IND are wrappers of function calls,
1166 * so they scratch BPF_R1-BPF_R5 registers, preserve
1167 * BPF_R6-BPF_R9, and store return value into BPF_R0.
1168 *
1169 * Implicit input:
1170 * ctx == skb == BPF_R6 == CTX
1171 *
1172 * Explicit input:
1173 * SRC == any register
1174 * IMM == 32-bit immediate
1175 *
1176 * Output:
1177 * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
1178 */
1179
1180 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp);
1181 if (likely(ptr != NULL)) {
1182 BPF_R0 = get_unaligned_be32(ptr);
1183 CONT;
1184 }
1185
1186 return 0;
1187 LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
1188 off = IMM;
1189 load_half:
1190 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp);
1191 if (likely(ptr != NULL)) {
1192 BPF_R0 = get_unaligned_be16(ptr);
1193 CONT;
1194 }
1195
1196 return 0;
1197 LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
1198 off = IMM;
1199 load_byte:
1200 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp);
1201 if (likely(ptr != NULL)) {
1202 BPF_R0 = *(u8 *)ptr;
1203 CONT;
1204 }
1205
1206 return 0;
1207 LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
1208 off = IMM + SRC;
1209 goto load_word;
1210 LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
1211 off = IMM + SRC;
1212 goto load_half;
1213 LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
1214 off = IMM + SRC;
1215 goto load_byte;
1216
1217 default_label:
1218 /* If we ever reach this, we have a bug somewhere. */
1219 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
1220 return 0;
1221 }
1222 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
1223
1224 bool bpf_prog_array_compatible(struct bpf_array *array,
1225 const struct bpf_prog *fp)
1226 {
1227 if (!array->owner_prog_type) {
1228 /* There's no owner yet where we could check for
1229 * compatibility.
1230 */
1231 array->owner_prog_type = fp->type;
1232 array->owner_jited = fp->jited;
1233
1234 return true;
1235 }
1236
1237 return array->owner_prog_type == fp->type &&
1238 array->owner_jited == fp->jited;
1239 }
1240
1241 static int bpf_check_tail_call(const struct bpf_prog *fp)
1242 {
1243 struct bpf_prog_aux *aux = fp->aux;
1244 int i;
1245
1246 for (i = 0; i < aux->used_map_cnt; i++) {
1247 struct bpf_map *map = aux->used_maps[i];
1248 struct bpf_array *array;
1249
1250 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1251 continue;
1252
1253 array = container_of(map, struct bpf_array, map);
1254 if (!bpf_prog_array_compatible(array, fp))
1255 return -EINVAL;
1256 }
1257
1258 return 0;
1259 }
1260
1261 /**
1262 * bpf_prog_select_runtime - select exec runtime for BPF program
1263 * @fp: bpf_prog populated with internal BPF program
1264 * @err: pointer to error variable
1265 *
1266 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1267 * The BPF program will be executed via BPF_PROG_RUN() macro.
1268 */
1269 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1270 {
1271 fp->bpf_func = (void *) __bpf_prog_run;
1272
1273 /* eBPF JITs can rewrite the program in case constant
1274 * blinding is active. However, in case of error during
1275 * blinding, bpf_int_jit_compile() must always return a
1276 * valid program, which in this case would simply not
1277 * be JITed, but falls back to the interpreter.
1278 */
1279 fp = bpf_int_jit_compile(fp);
1280 bpf_prog_lock_ro(fp);
1281
1282 /* The tail call compatibility check can only be done at
1283 * this late stage as we need to determine, if we deal
1284 * with JITed or non JITed program concatenations and not
1285 * all eBPF JITs might immediately support all features.
1286 */
1287 *err = bpf_check_tail_call(fp);
1288
1289 return fp;
1290 }
1291 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1292
1293 static void bpf_prog_free_deferred(struct work_struct *work)
1294 {
1295 struct bpf_prog_aux *aux;
1296
1297 aux = container_of(work, struct bpf_prog_aux, work);
1298 bpf_jit_free(aux->prog);
1299 }
1300
1301 /* Free internal BPF program */
1302 void bpf_prog_free(struct bpf_prog *fp)
1303 {
1304 struct bpf_prog_aux *aux = fp->aux;
1305
1306 INIT_WORK(&aux->work, bpf_prog_free_deferred);
1307 schedule_work(&aux->work);
1308 }
1309 EXPORT_SYMBOL_GPL(bpf_prog_free);
1310
1311 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
1312 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
1313
1314 void bpf_user_rnd_init_once(void)
1315 {
1316 prandom_init_once(&bpf_user_rnd_state);
1317 }
1318
1319 BPF_CALL_0(bpf_user_rnd_u32)
1320 {
1321 /* Should someone ever have the rather unwise idea to use some
1322 * of the registers passed into this function, then note that
1323 * this function is called from native eBPF and classic-to-eBPF
1324 * transformations. Register assignments from both sides are
1325 * different, f.e. classic always sets fn(ctx, A, X) here.
1326 */
1327 struct rnd_state *state;
1328 u32 res;
1329
1330 state = &get_cpu_var(bpf_user_rnd_state);
1331 res = prandom_u32_state(state);
1332 put_cpu_var(bpf_user_rnd_state);
1333
1334 return res;
1335 }
1336
1337 /* Weak definitions of helper functions in case we don't have bpf syscall. */
1338 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
1339 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
1340 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
1341
1342 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
1343 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
1344 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
1345 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
1346
1347 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
1348 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
1349 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
1350
1351 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
1352 {
1353 return NULL;
1354 }
1355
1356 u64 __weak
1357 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
1358 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
1359 {
1360 return -ENOTSUPP;
1361 }
1362
1363 /* Always built-in helper functions. */
1364 const struct bpf_func_proto bpf_tail_call_proto = {
1365 .func = NULL,
1366 .gpl_only = false,
1367 .ret_type = RET_VOID,
1368 .arg1_type = ARG_PTR_TO_CTX,
1369 .arg2_type = ARG_CONST_MAP_PTR,
1370 .arg3_type = ARG_ANYTHING,
1371 };
1372
1373 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
1374 * It is encouraged to implement bpf_int_jit_compile() instead, so that
1375 * eBPF and implicitly also cBPF can get JITed!
1376 */
1377 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
1378 {
1379 return prog;
1380 }
1381
1382 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
1383 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
1384 */
1385 void __weak bpf_jit_compile(struct bpf_prog *prog)
1386 {
1387 }
1388
1389 bool __weak bpf_helper_changes_pkt_data(void *func)
1390 {
1391 return false;
1392 }
1393
1394 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
1395 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
1396 */
1397 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
1398 int len)
1399 {
1400 return -EFAULT;
1401 }
1402
1403 /* All definitions of tracepoints related to BPF. */
1404 #define CREATE_TRACE_POINTS
1405 #include <linux/bpf_trace.h>
1406
1407 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
1408
1409 EXPORT_TRACEPOINT_SYMBOL_GPL(bpf_prog_get_type);
1410 EXPORT_TRACEPOINT_SYMBOL_GPL(bpf_prog_put_rcu);
(deprecated) Btrfs devel tree