]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - kernel/bpf/core.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bpf: Add bpf_snprintf_btf helper
[mirror_ubuntu-hirsute-kernel.git] / kernel / bpf / core.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <uapi/linux/btf.h>
21 #include <linux/filter.h>
22 #include <linux/skbuff.h>
23 #include <linux/vmalloc.h>
24 #include <linux/random.h>
25 #include <linux/moduleloader.h>
26 #include <linux/bpf.h>
27 #include <linux/btf.h>
28 #include <linux/frame.h>
29 #include <linux/rbtree_latch.h>
30 #include <linux/kallsyms.h>
31 #include <linux/rcupdate.h>
32 #include <linux/perf_event.h>
33 #include <linux/extable.h>
34 #include <linux/log2.h>
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 AX regs[BPF_REG_AX]
55 #define ARG1 regs[BPF_REG_ARG1]
56 #define CTX regs[BPF_REG_CTX]
57 #define IMM insn->imm
58
59 /* No hurry in this branch
60 *
61 * Exported for the bpf jit load helper.
62 */
63 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
64 {
65 u8 *ptr = NULL;
66
67 if (k >= SKF_NET_OFF)
68 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
69 else if (k >= SKF_LL_OFF)
70 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
71
72 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
73 return ptr;
74
75 return NULL;
76 }
77
78 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
79 {
80 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
81 struct bpf_prog_aux *aux;
82 struct bpf_prog *fp;
83
84 size = round_up(size, PAGE_SIZE);
85 fp = __vmalloc(size, gfp_flags);
86 if (fp == NULL)
87 return NULL;
88
89 aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
90 if (aux == NULL) {
91 vfree(fp);
92 return NULL;
93 }
94
95 fp->pages = size / PAGE_SIZE;
96 fp->aux = aux;
97 fp->aux->prog = fp;
98 fp->jit_requested = ebpf_jit_enabled();
99
100 INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
101 mutex_init(&fp->aux->used_maps_mutex);
102
103 return fp;
104 }
105
106 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
107 {
108 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
109 struct bpf_prog *prog;
110 int cpu;
111
112 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
113 if (!prog)
114 return NULL;
115
116 prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
117 if (!prog->aux->stats) {
118 kfree(prog->aux);
119 vfree(prog);
120 return NULL;
121 }
122
123 for_each_possible_cpu(cpu) {
124 struct bpf_prog_stats *pstats;
125
126 pstats = per_cpu_ptr(prog->aux->stats, cpu);
127 u64_stats_init(&pstats->syncp);
128 }
129 return prog;
130 }
131 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
132
133 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
134 {
135 if (!prog->aux->nr_linfo || !prog->jit_requested)
136 return 0;
137
138 prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo,
139 sizeof(*prog->aux->jited_linfo),
140 GFP_KERNEL | __GFP_NOWARN);
141 if (!prog->aux->jited_linfo)
142 return -ENOMEM;
143
144 return 0;
145 }
146
147 void bpf_prog_free_jited_linfo(struct bpf_prog *prog)
148 {
149 kfree(prog->aux->jited_linfo);
150 prog->aux->jited_linfo = NULL;
151 }
152
153 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog)
154 {
155 if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0])
156 bpf_prog_free_jited_linfo(prog);
157 }
158
159 /* The jit engine is responsible to provide an array
160 * for insn_off to the jited_off mapping (insn_to_jit_off).
161 *
162 * The idx to this array is the insn_off. Hence, the insn_off
163 * here is relative to the prog itself instead of the main prog.
164 * This array has one entry for each xlated bpf insn.
165 *
166 * jited_off is the byte off to the last byte of the jited insn.
167 *
168 * Hence, with
169 * insn_start:
170 * The first bpf insn off of the prog. The insn off
171 * here is relative to the main prog.
172 * e.g. if prog is a subprog, insn_start > 0
173 * linfo_idx:
174 * The prog's idx to prog->aux->linfo and jited_linfo
175 *
176 * jited_linfo[linfo_idx] = prog->bpf_func
177 *
178 * For i > linfo_idx,
179 *
180 * jited_linfo[i] = prog->bpf_func +
181 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
182 */
183 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
184 const u32 *insn_to_jit_off)
185 {
186 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
187 const struct bpf_line_info *linfo;
188 void **jited_linfo;
189
190 if (!prog->aux->jited_linfo)
191 /* Userspace did not provide linfo */
192 return;
193
194 linfo_idx = prog->aux->linfo_idx;
195 linfo = &prog->aux->linfo[linfo_idx];
196 insn_start = linfo[0].insn_off;
197 insn_end = insn_start + prog->len;
198
199 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
200 jited_linfo[0] = prog->bpf_func;
201
202 nr_linfo = prog->aux->nr_linfo - linfo_idx;
203
204 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
205 /* The verifier ensures that linfo[i].insn_off is
206 * strictly increasing
207 */
208 jited_linfo[i] = prog->bpf_func +
209 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
210 }
211
212 void bpf_prog_free_linfo(struct bpf_prog *prog)
213 {
214 bpf_prog_free_jited_linfo(prog);
215 kvfree(prog->aux->linfo);
216 }
217
218 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
219 gfp_t gfp_extra_flags)
220 {
221 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
222 struct bpf_prog *fp;
223 u32 pages, delta;
224 int ret;
225
226 size = round_up(size, PAGE_SIZE);
227 pages = size / PAGE_SIZE;
228 if (pages <= fp_old->pages)
229 return fp_old;
230
231 delta = pages - fp_old->pages;
232 ret = __bpf_prog_charge(fp_old->aux->user, delta);
233 if (ret)
234 return NULL;
235
236 fp = __vmalloc(size, gfp_flags);
237 if (fp == NULL) {
238 __bpf_prog_uncharge(fp_old->aux->user, delta);
239 } else {
240 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
241 fp->pages = pages;
242 fp->aux->prog = fp;
243
244 /* We keep fp->aux from fp_old around in the new
245 * reallocated structure.
246 */
247 fp_old->aux = NULL;
248 __bpf_prog_free(fp_old);
249 }
250
251 return fp;
252 }
253
254 void __bpf_prog_free(struct bpf_prog *fp)
255 {
256 if (fp->aux) {
257 mutex_destroy(&fp->aux->used_maps_mutex);
258 free_percpu(fp->aux->stats);
259 kfree(fp->aux->poke_tab);
260 kfree(fp->aux);
261 }
262 vfree(fp);
263 }
264
265 int bpf_prog_calc_tag(struct bpf_prog *fp)
266 {
267 const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
268 u32 raw_size = bpf_prog_tag_scratch_size(fp);
269 u32 digest[SHA1_DIGEST_WORDS];
270 u32 ws[SHA1_WORKSPACE_WORDS];
271 u32 i, bsize, psize, blocks;
272 struct bpf_insn *dst;
273 bool was_ld_map;
274 u8 *raw, *todo;
275 __be32 *result;
276 __be64 *bits;
277
278 raw = vmalloc(raw_size);
279 if (!raw)
280 return -ENOMEM;
281
282 sha1_init(digest);
283 memset(ws, 0, sizeof(ws));
284
285 /* We need to take out the map fd for the digest calculation
286 * since they are unstable from user space side.
287 */
288 dst = (void *)raw;
289 for (i = 0, was_ld_map = false; i < fp->len; i++) {
290 dst[i] = fp->insnsi[i];
291 if (!was_ld_map &&
292 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
293 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
294 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
295 was_ld_map = true;
296 dst[i].imm = 0;
297 } else if (was_ld_map &&
298 dst[i].code == 0 &&
299 dst[i].dst_reg == 0 &&
300 dst[i].src_reg == 0 &&
301 dst[i].off == 0) {
302 was_ld_map = false;
303 dst[i].imm = 0;
304 } else {
305 was_ld_map = false;
306 }
307 }
308
309 psize = bpf_prog_insn_size(fp);
310 memset(&raw[psize], 0, raw_size - psize);
311 raw[psize++] = 0x80;
312
313 bsize = round_up(psize, SHA1_BLOCK_SIZE);
314 blocks = bsize / SHA1_BLOCK_SIZE;
315 todo = raw;
316 if (bsize - psize >= sizeof(__be64)) {
317 bits = (__be64 *)(todo + bsize - sizeof(__be64));
318 } else {
319 bits = (__be64 *)(todo + bsize + bits_offset);
320 blocks++;
321 }
322 *bits = cpu_to_be64((psize - 1) << 3);
323
324 while (blocks--) {
325 sha1_transform(digest, todo, ws);
326 todo += SHA1_BLOCK_SIZE;
327 }
328
329 result = (__force __be32 *)digest;
330 for (i = 0; i < SHA1_DIGEST_WORDS; i++)
331 result[i] = cpu_to_be32(digest[i]);
332 memcpy(fp->tag, result, sizeof(fp->tag));
333
334 vfree(raw);
335 return 0;
336 }
337
338 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
339 s32 end_new, s32 curr, const bool probe_pass)
340 {
341 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
342 s32 delta = end_new - end_old;
343 s64 imm = insn->imm;
344
345 if (curr < pos && curr + imm + 1 >= end_old)
346 imm += delta;
347 else if (curr >= end_new && curr + imm + 1 < end_new)
348 imm -= delta;
349 if (imm < imm_min || imm > imm_max)
350 return -ERANGE;
351 if (!probe_pass)
352 insn->imm = imm;
353 return 0;
354 }
355
356 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
357 s32 end_new, s32 curr, const bool probe_pass)
358 {
359 const s32 off_min = S16_MIN, off_max = S16_MAX;
360 s32 delta = end_new - end_old;
361 s32 off = insn->off;
362
363 if (curr < pos && curr + off + 1 >= end_old)
364 off += delta;
365 else if (curr >= end_new && curr + off + 1 < end_new)
366 off -= delta;
367 if (off < off_min || off > off_max)
368 return -ERANGE;
369 if (!probe_pass)
370 insn->off = off;
371 return 0;
372 }
373
374 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
375 s32 end_new, const bool probe_pass)
376 {
377 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
378 struct bpf_insn *insn = prog->insnsi;
379 int ret = 0;
380
381 for (i = 0; i < insn_cnt; i++, insn++) {
382 u8 code;
383
384 /* In the probing pass we still operate on the original,
385 * unpatched image in order to check overflows before we
386 * do any other adjustments. Therefore skip the patchlet.
387 */
388 if (probe_pass && i == pos) {
389 i = end_new;
390 insn = prog->insnsi + end_old;
391 }
392 code = insn->code;
393 if ((BPF_CLASS(code) != BPF_JMP &&
394 BPF_CLASS(code) != BPF_JMP32) ||
395 BPF_OP(code) == BPF_EXIT)
396 continue;
397 /* Adjust offset of jmps if we cross patch boundaries. */
398 if (BPF_OP(code) == BPF_CALL) {
399 if (insn->src_reg != BPF_PSEUDO_CALL)
400 continue;
401 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
402 end_new, i, probe_pass);
403 } else {
404 ret = bpf_adj_delta_to_off(insn, pos, end_old,
405 end_new, i, probe_pass);
406 }
407 if (ret)
408 break;
409 }
410
411 return ret;
412 }
413
414 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
415 {
416 struct bpf_line_info *linfo;
417 u32 i, nr_linfo;
418
419 nr_linfo = prog->aux->nr_linfo;
420 if (!nr_linfo || !delta)
421 return;
422
423 linfo = prog->aux->linfo;
424
425 for (i = 0; i < nr_linfo; i++)
426 if (off < linfo[i].insn_off)
427 break;
428
429 /* Push all off < linfo[i].insn_off by delta */
430 for (; i < nr_linfo; i++)
431 linfo[i].insn_off += delta;
432 }
433
434 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
435 const struct bpf_insn *patch, u32 len)
436 {
437 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
438 const u32 cnt_max = S16_MAX;
439 struct bpf_prog *prog_adj;
440 int err;
441
442 /* Since our patchlet doesn't expand the image, we're done. */
443 if (insn_delta == 0) {
444 memcpy(prog->insnsi + off, patch, sizeof(*patch));
445 return prog;
446 }
447
448 insn_adj_cnt = prog->len + insn_delta;
449
450 /* Reject anything that would potentially let the insn->off
451 * target overflow when we have excessive program expansions.
452 * We need to probe here before we do any reallocation where
453 * we afterwards may not fail anymore.
454 */
455 if (insn_adj_cnt > cnt_max &&
456 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
457 return ERR_PTR(err);
458
459 /* Several new instructions need to be inserted. Make room
460 * for them. Likely, there's no need for a new allocation as
461 * last page could have large enough tailroom.
462 */
463 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
464 GFP_USER);
465 if (!prog_adj)
466 return ERR_PTR(-ENOMEM);
467
468 prog_adj->len = insn_adj_cnt;
469
470 /* Patching happens in 3 steps:
471 *
472 * 1) Move over tail of insnsi from next instruction onwards,
473 * so we can patch the single target insn with one or more
474 * new ones (patching is always from 1 to n insns, n > 0).
475 * 2) Inject new instructions at the target location.
476 * 3) Adjust branch offsets if necessary.
477 */
478 insn_rest = insn_adj_cnt - off - len;
479
480 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
481 sizeof(*patch) * insn_rest);
482 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
483
484 /* We are guaranteed to not fail at this point, otherwise
485 * the ship has sailed to reverse to the original state. An
486 * overflow cannot happen at this point.
487 */
488 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
489
490 bpf_adj_linfo(prog_adj, off, insn_delta);
491
492 return prog_adj;
493 }
494
495 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
496 {
497 /* Branch offsets can't overflow when program is shrinking, no need
498 * to call bpf_adj_branches(..., true) here
499 */
500 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
501 sizeof(struct bpf_insn) * (prog->len - off - cnt));
502 prog->len -= cnt;
503
504 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
505 }
506
507 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
508 {
509 int i;
510
511 for (i = 0; i < fp->aux->func_cnt; i++)
512 bpf_prog_kallsyms_del(fp->aux->func[i]);
513 }
514
515 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
516 {
517 bpf_prog_kallsyms_del_subprogs(fp);
518 bpf_prog_kallsyms_del(fp);
519 }
520
521 #ifdef CONFIG_BPF_JIT
522 /* All BPF JIT sysctl knobs here. */
523 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
524 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
525 int bpf_jit_harden __read_mostly;
526 long bpf_jit_limit __read_mostly;
527
528 static void
529 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
530 {
531 const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog);
532 unsigned long addr = (unsigned long)hdr;
533
534 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
535
536 prog->aux->ksym.start = (unsigned long) prog->bpf_func;
537 prog->aux->ksym.end = addr + hdr->pages * PAGE_SIZE;
538 }
539
540 static void
541 bpf_prog_ksym_set_name(struct bpf_prog *prog)
542 {
543 char *sym = prog->aux->ksym.name;
544 const char *end = sym + KSYM_NAME_LEN;
545 const struct btf_type *type;
546 const char *func_name;
547
548 BUILD_BUG_ON(sizeof("bpf_prog_") +
549 sizeof(prog->tag) * 2 +
550 /* name has been null terminated.
551 * We should need +1 for the '_' preceding
552 * the name. However, the null character
553 * is double counted between the name and the
554 * sizeof("bpf_prog_") above, so we omit
555 * the +1 here.
556 */
557 sizeof(prog->aux->name) > KSYM_NAME_LEN);
558
559 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
560 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
561
562 /* prog->aux->name will be ignored if full btf name is available */
563 if (prog->aux->func_info_cnt) {
564 type = btf_type_by_id(prog->aux->btf,
565 prog->aux->func_info[prog->aux->func_idx].type_id);
566 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
567 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
568 return;
569 }
570
571 if (prog->aux->name[0])
572 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
573 else
574 *sym = 0;
575 }
576
577 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
578 {
579 return container_of(n, struct bpf_ksym, tnode)->start;
580 }
581
582 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
583 struct latch_tree_node *b)
584 {
585 return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
586 }
587
588 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
589 {
590 unsigned long val = (unsigned long)key;
591 const struct bpf_ksym *ksym;
592
593 ksym = container_of(n, struct bpf_ksym, tnode);
594
595 if (val < ksym->start)
596 return -1;
597 if (val >= ksym->end)
598 return 1;
599
600 return 0;
601 }
602
603 static const struct latch_tree_ops bpf_tree_ops = {
604 .less = bpf_tree_less,
605 .comp = bpf_tree_comp,
606 };
607
608 static DEFINE_SPINLOCK(bpf_lock);
609 static LIST_HEAD(bpf_kallsyms);
610 static struct latch_tree_root bpf_tree __cacheline_aligned;
611
612 void bpf_ksym_add(struct bpf_ksym *ksym)
613 {
614 spin_lock_bh(&bpf_lock);
615 WARN_ON_ONCE(!list_empty(&ksym->lnode));
616 list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
617 latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
618 spin_unlock_bh(&bpf_lock);
619 }
620
621 static void __bpf_ksym_del(struct bpf_ksym *ksym)
622 {
623 if (list_empty(&ksym->lnode))
624 return;
625
626 latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
627 list_del_rcu(&ksym->lnode);
628 }
629
630 void bpf_ksym_del(struct bpf_ksym *ksym)
631 {
632 spin_lock_bh(&bpf_lock);
633 __bpf_ksym_del(ksym);
634 spin_unlock_bh(&bpf_lock);
635 }
636
637 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
638 {
639 return fp->jited && !bpf_prog_was_classic(fp);
640 }
641
642 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
643 {
644 return list_empty(&fp->aux->ksym.lnode) ||
645 fp->aux->ksym.lnode.prev == LIST_POISON2;
646 }
647
648 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
649 {
650 if (!bpf_prog_kallsyms_candidate(fp) ||
651 !bpf_capable())
652 return;
653
654 bpf_prog_ksym_set_addr(fp);
655 bpf_prog_ksym_set_name(fp);
656 fp->aux->ksym.prog = true;
657
658 bpf_ksym_add(&fp->aux->ksym);
659 }
660
661 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
662 {
663 if (!bpf_prog_kallsyms_candidate(fp))
664 return;
665
666 bpf_ksym_del(&fp->aux->ksym);
667 }
668
669 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
670 {
671 struct latch_tree_node *n;
672
673 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
674 return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
675 }
676
677 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
678 unsigned long *off, char *sym)
679 {
680 struct bpf_ksym *ksym;
681 char *ret = NULL;
682
683 rcu_read_lock();
684 ksym = bpf_ksym_find(addr);
685 if (ksym) {
686 unsigned long symbol_start = ksym->start;
687 unsigned long symbol_end = ksym->end;
688
689 strncpy(sym, ksym->name, KSYM_NAME_LEN);
690
691 ret = sym;
692 if (size)
693 *size = symbol_end - symbol_start;
694 if (off)
695 *off = addr - symbol_start;
696 }
697 rcu_read_unlock();
698
699 return ret;
700 }
701
702 bool is_bpf_text_address(unsigned long addr)
703 {
704 bool ret;
705
706 rcu_read_lock();
707 ret = bpf_ksym_find(addr) != NULL;
708 rcu_read_unlock();
709
710 return ret;
711 }
712
713 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
714 {
715 struct bpf_ksym *ksym = bpf_ksym_find(addr);
716
717 return ksym && ksym->prog ?
718 container_of(ksym, struct bpf_prog_aux, ksym)->prog :
719 NULL;
720 }
721
722 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
723 {
724 const struct exception_table_entry *e = NULL;
725 struct bpf_prog *prog;
726
727 rcu_read_lock();
728 prog = bpf_prog_ksym_find(addr);
729 if (!prog)
730 goto out;
731 if (!prog->aux->num_exentries)
732 goto out;
733
734 e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
735 out:
736 rcu_read_unlock();
737 return e;
738 }
739
740 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
741 char *sym)
742 {
743 struct bpf_ksym *ksym;
744 unsigned int it = 0;
745 int ret = -ERANGE;
746
747 if (!bpf_jit_kallsyms_enabled())
748 return ret;
749
750 rcu_read_lock();
751 list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
752 if (it++ != symnum)
753 continue;
754
755 strncpy(sym, ksym->name, KSYM_NAME_LEN);
756
757 *value = ksym->start;
758 *type = BPF_SYM_ELF_TYPE;
759
760 ret = 0;
761 break;
762 }
763 rcu_read_unlock();
764
765 return ret;
766 }
767
768 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
769 struct bpf_jit_poke_descriptor *poke)
770 {
771 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
772 static const u32 poke_tab_max = 1024;
773 u32 slot = prog->aux->size_poke_tab;
774 u32 size = slot + 1;
775
776 if (size > poke_tab_max)
777 return -ENOSPC;
778 if (poke->tailcall_target || poke->tailcall_target_stable ||
779 poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
780 return -EINVAL;
781
782 switch (poke->reason) {
783 case BPF_POKE_REASON_TAIL_CALL:
784 if (!poke->tail_call.map)
785 return -EINVAL;
786 break;
787 default:
788 return -EINVAL;
789 }
790
791 tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
792 if (!tab)
793 return -ENOMEM;
794
795 memcpy(&tab[slot], poke, sizeof(*poke));
796 prog->aux->size_poke_tab = size;
797 prog->aux->poke_tab = tab;
798
799 return slot;
800 }
801
802 static atomic_long_t bpf_jit_current;
803
804 /* Can be overridden by an arch's JIT compiler if it has a custom,
805 * dedicated BPF backend memory area, or if neither of the two
806 * below apply.
807 */
808 u64 __weak bpf_jit_alloc_exec_limit(void)
809 {
810 #if defined(MODULES_VADDR)
811 return MODULES_END - MODULES_VADDR;
812 #else
813 return VMALLOC_END - VMALLOC_START;
814 #endif
815 }
816
817 static int __init bpf_jit_charge_init(void)
818 {
819 /* Only used as heuristic here to derive limit. */
820 bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
821 PAGE_SIZE), LONG_MAX);
822 return 0;
823 }
824 pure_initcall(bpf_jit_charge_init);
825
826 static int bpf_jit_charge_modmem(u32 pages)
827 {
828 if (atomic_long_add_return(pages, &bpf_jit_current) >
829 (bpf_jit_limit >> PAGE_SHIFT)) {
830 if (!capable(CAP_SYS_ADMIN)) {
831 atomic_long_sub(pages, &bpf_jit_current);
832 return -EPERM;
833 }
834 }
835
836 return 0;
837 }
838
839 static void bpf_jit_uncharge_modmem(u32 pages)
840 {
841 atomic_long_sub(pages, &bpf_jit_current);
842 }
843
844 void *__weak bpf_jit_alloc_exec(unsigned long size)
845 {
846 return module_alloc(size);
847 }
848
849 void __weak bpf_jit_free_exec(void *addr)
850 {
851 module_memfree(addr);
852 }
853
854 struct bpf_binary_header *
855 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
856 unsigned int alignment,
857 bpf_jit_fill_hole_t bpf_fill_ill_insns)
858 {
859 struct bpf_binary_header *hdr;
860 u32 size, hole, start, pages;
861
862 WARN_ON_ONCE(!is_power_of_2(alignment) ||
863 alignment > BPF_IMAGE_ALIGNMENT);
864
865 /* Most of BPF filters are really small, but if some of them
866 * fill a page, allow at least 128 extra bytes to insert a
867 * random section of illegal instructions.
868 */
869 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
870 pages = size / PAGE_SIZE;
871
872 if (bpf_jit_charge_modmem(pages))
873 return NULL;
874 hdr = bpf_jit_alloc_exec(size);
875 if (!hdr) {
876 bpf_jit_uncharge_modmem(pages);
877 return NULL;
878 }
879
880 /* Fill space with illegal/arch-dep instructions. */
881 bpf_fill_ill_insns(hdr, size);
882
883 hdr->pages = pages;
884 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
885 PAGE_SIZE - sizeof(*hdr));
886 start = (get_random_int() % hole) & ~(alignment - 1);
887
888 /* Leave a random number of instructions before BPF code. */
889 *image_ptr = &hdr->image[start];
890
891 return hdr;
892 }
893
894 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
895 {
896 u32 pages = hdr->pages;
897
898 bpf_jit_free_exec(hdr);
899 bpf_jit_uncharge_modmem(pages);
900 }
901
902 /* This symbol is only overridden by archs that have different
903 * requirements than the usual eBPF JITs, f.e. when they only
904 * implement cBPF JIT, do not set images read-only, etc.
905 */
906 void __weak bpf_jit_free(struct bpf_prog *fp)
907 {
908 if (fp->jited) {
909 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
910
911 bpf_jit_binary_free(hdr);
912
913 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
914 }
915
916 bpf_prog_unlock_free(fp);
917 }
918
919 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
920 const struct bpf_insn *insn, bool extra_pass,
921 u64 *func_addr, bool *func_addr_fixed)
922 {
923 s16 off = insn->off;
924 s32 imm = insn->imm;
925 u8 *addr;
926
927 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
928 if (!*func_addr_fixed) {
929 /* Place-holder address till the last pass has collected
930 * all addresses for JITed subprograms in which case we
931 * can pick them up from prog->aux.
932 */
933 if (!extra_pass)
934 addr = NULL;
935 else if (prog->aux->func &&
936 off >= 0 && off < prog->aux->func_cnt)
937 addr = (u8 *)prog->aux->func[off]->bpf_func;
938 else
939 return -EINVAL;
940 } else {
941 /* Address of a BPF helper call. Since part of the core
942 * kernel, it's always at a fixed location. __bpf_call_base
943 * and the helper with imm relative to it are both in core
944 * kernel.
945 */
946 addr = (u8 *)__bpf_call_base + imm;
947 }
948
949 *func_addr = (unsigned long)addr;
950 return 0;
951 }
952
953 static int bpf_jit_blind_insn(const struct bpf_insn *from,
954 const struct bpf_insn *aux,
955 struct bpf_insn *to_buff,
956 bool emit_zext)
957 {
958 struct bpf_insn *to = to_buff;
959 u32 imm_rnd = get_random_int();
960 s16 off;
961
962 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
963 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
964
965 /* Constraints on AX register:
966 *
967 * AX register is inaccessible from user space. It is mapped in
968 * all JITs, and used here for constant blinding rewrites. It is
969 * typically "stateless" meaning its contents are only valid within
970 * the executed instruction, but not across several instructions.
971 * There are a few exceptions however which are further detailed
972 * below.
973 *
974 * Constant blinding is only used by JITs, not in the interpreter.
975 * The interpreter uses AX in some occasions as a local temporary
976 * register e.g. in DIV or MOD instructions.
977 *
978 * In restricted circumstances, the verifier can also use the AX
979 * register for rewrites as long as they do not interfere with
980 * the above cases!
981 */
982 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
983 goto out;
984
985 if (from->imm == 0 &&
986 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
987 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
988 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
989 goto out;
990 }
991
992 switch (from->code) {
993 case BPF_ALU | BPF_ADD | BPF_K:
994 case BPF_ALU | BPF_SUB | BPF_K:
995 case BPF_ALU | BPF_AND | BPF_K:
996 case BPF_ALU | BPF_OR | BPF_K:
997 case BPF_ALU | BPF_XOR | BPF_K:
998 case BPF_ALU | BPF_MUL | BPF_K:
999 case BPF_ALU | BPF_MOV | BPF_K:
1000 case BPF_ALU | BPF_DIV | BPF_K:
1001 case BPF_ALU | BPF_MOD | BPF_K:
1002 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1003 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1004 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1005 break;
1006
1007 case BPF_ALU64 | BPF_ADD | BPF_K:
1008 case BPF_ALU64 | BPF_SUB | BPF_K:
1009 case BPF_ALU64 | BPF_AND | BPF_K:
1010 case BPF_ALU64 | BPF_OR | BPF_K:
1011 case BPF_ALU64 | BPF_XOR | BPF_K:
1012 case BPF_ALU64 | BPF_MUL | BPF_K:
1013 case BPF_ALU64 | BPF_MOV | BPF_K:
1014 case BPF_ALU64 | BPF_DIV | BPF_K:
1015 case BPF_ALU64 | BPF_MOD | BPF_K:
1016 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1017 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1018 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1019 break;
1020
1021 case BPF_JMP | BPF_JEQ | BPF_K:
1022 case BPF_JMP | BPF_JNE | BPF_K:
1023 case BPF_JMP | BPF_JGT | BPF_K:
1024 case BPF_JMP | BPF_JLT | BPF_K:
1025 case BPF_JMP | BPF_JGE | BPF_K:
1026 case BPF_JMP | BPF_JLE | BPF_K:
1027 case BPF_JMP | BPF_JSGT | BPF_K:
1028 case BPF_JMP | BPF_JSLT | BPF_K:
1029 case BPF_JMP | BPF_JSGE | BPF_K:
1030 case BPF_JMP | BPF_JSLE | BPF_K:
1031 case BPF_JMP | BPF_JSET | BPF_K:
1032 /* Accommodate for extra offset in case of a backjump. */
1033 off = from->off;
1034 if (off < 0)
1035 off -= 2;
1036 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1037 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1038 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1039 break;
1040
1041 case BPF_JMP32 | BPF_JEQ | BPF_K:
1042 case BPF_JMP32 | BPF_JNE | BPF_K:
1043 case BPF_JMP32 | BPF_JGT | BPF_K:
1044 case BPF_JMP32 | BPF_JLT | BPF_K:
1045 case BPF_JMP32 | BPF_JGE | BPF_K:
1046 case BPF_JMP32 | BPF_JLE | BPF_K:
1047 case BPF_JMP32 | BPF_JSGT | BPF_K:
1048 case BPF_JMP32 | BPF_JSLT | BPF_K:
1049 case BPF_JMP32 | BPF_JSGE | BPF_K:
1050 case BPF_JMP32 | BPF_JSLE | BPF_K:
1051 case BPF_JMP32 | BPF_JSET | BPF_K:
1052 /* Accommodate for extra offset in case of a backjump. */
1053 off = from->off;
1054 if (off < 0)
1055 off -= 2;
1056 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1057 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1058 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1059 off);
1060 break;
1061
1062 case BPF_LD | BPF_IMM | BPF_DW:
1063 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1064 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1065 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1066 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1067 break;
1068 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1069 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1070 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1071 if (emit_zext)
1072 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
1073 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1074 break;
1075
1076 case BPF_ST | BPF_MEM | BPF_DW:
1077 case BPF_ST | BPF_MEM | BPF_W:
1078 case BPF_ST | BPF_MEM | BPF_H:
1079 case BPF_ST | BPF_MEM | BPF_B:
1080 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1081 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1082 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1083 break;
1084 }
1085 out:
1086 return to - to_buff;
1087 }
1088
1089 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1090 gfp_t gfp_extra_flags)
1091 {
1092 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1093 struct bpf_prog *fp;
1094
1095 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1096 if (fp != NULL) {
1097 /* aux->prog still points to the fp_other one, so
1098 * when promoting the clone to the real program,
1099 * this still needs to be adapted.
1100 */
1101 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1102 }
1103
1104 return fp;
1105 }
1106
1107 static void bpf_prog_clone_free(struct bpf_prog *fp)
1108 {
1109 /* aux was stolen by the other clone, so we cannot free
1110 * it from this path! It will be freed eventually by the
1111 * other program on release.
1112 *
1113 * At this point, we don't need a deferred release since
1114 * clone is guaranteed to not be locked.
1115 */
1116 fp->aux = NULL;
1117 __bpf_prog_free(fp);
1118 }
1119
1120 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1121 {
1122 /* We have to repoint aux->prog to self, as we don't
1123 * know whether fp here is the clone or the original.
1124 */
1125 fp->aux->prog = fp;
1126 bpf_prog_clone_free(fp_other);
1127 }
1128
1129 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1130 {
1131 struct bpf_insn insn_buff[16], aux[2];
1132 struct bpf_prog *clone, *tmp;
1133 int insn_delta, insn_cnt;
1134 struct bpf_insn *insn;
1135 int i, rewritten;
1136
1137 if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1138 return prog;
1139
1140 clone = bpf_prog_clone_create(prog, GFP_USER);
1141 if (!clone)
1142 return ERR_PTR(-ENOMEM);
1143
1144 insn_cnt = clone->len;
1145 insn = clone->insnsi;
1146
1147 for (i = 0; i < insn_cnt; i++, insn++) {
1148 /* We temporarily need to hold the original ld64 insn
1149 * so that we can still access the first part in the
1150 * second blinding run.
1151 */
1152 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1153 insn[1].code == 0)
1154 memcpy(aux, insn, sizeof(aux));
1155
1156 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1157 clone->aux->verifier_zext);
1158 if (!rewritten)
1159 continue;
1160
1161 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1162 if (IS_ERR(tmp)) {
1163 /* Patching may have repointed aux->prog during
1164 * realloc from the original one, so we need to
1165 * fix it up here on error.
1166 */
1167 bpf_jit_prog_release_other(prog, clone);
1168 return tmp;
1169 }
1170
1171 clone = tmp;
1172 insn_delta = rewritten - 1;
1173
1174 /* Walk new program and skip insns we just inserted. */
1175 insn = clone->insnsi + i + insn_delta;
1176 insn_cnt += insn_delta;
1177 i += insn_delta;
1178 }
1179
1180 clone->blinded = 1;
1181 return clone;
1182 }
1183 #endif /* CONFIG_BPF_JIT */
1184
1185 /* Base function for offset calculation. Needs to go into .text section,
1186 * therefore keeping it non-static as well; will also be used by JITs
1187 * anyway later on, so do not let the compiler omit it. This also needs
1188 * to go into kallsyms for correlation from e.g. bpftool, so naming
1189 * must not change.
1190 */
1191 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1192 {
1193 return 0;
1194 }
1195 EXPORT_SYMBOL_GPL(__bpf_call_base);
1196
1197 /* All UAPI available opcodes. */
1198 #define BPF_INSN_MAP(INSN_2, INSN_3) \
1199 /* 32 bit ALU operations. */ \
1200 /* Register based. */ \
1201 INSN_3(ALU, ADD, X), \
1202 INSN_3(ALU, SUB, X), \
1203 INSN_3(ALU, AND, X), \
1204 INSN_3(ALU, OR, X), \
1205 INSN_3(ALU, LSH, X), \
1206 INSN_3(ALU, RSH, X), \
1207 INSN_3(ALU, XOR, X), \
1208 INSN_3(ALU, MUL, X), \
1209 INSN_3(ALU, MOV, X), \
1210 INSN_3(ALU, ARSH, X), \
1211 INSN_3(ALU, DIV, X), \
1212 INSN_3(ALU, MOD, X), \
1213 INSN_2(ALU, NEG), \
1214 INSN_3(ALU, END, TO_BE), \
1215 INSN_3(ALU, END, TO_LE), \
1216 /* Immediate based. */ \
1217 INSN_3(ALU, ADD, K), \
1218 INSN_3(ALU, SUB, K), \
1219 INSN_3(ALU, AND, K), \
1220 INSN_3(ALU, OR, K), \
1221 INSN_3(ALU, LSH, K), \
1222 INSN_3(ALU, RSH, K), \
1223 INSN_3(ALU, XOR, K), \
1224 INSN_3(ALU, MUL, K), \
1225 INSN_3(ALU, MOV, K), \
1226 INSN_3(ALU, ARSH, K), \
1227 INSN_3(ALU, DIV, K), \
1228 INSN_3(ALU, MOD, K), \
1229 /* 64 bit ALU operations. */ \
1230 /* Register based. */ \
1231 INSN_3(ALU64, ADD, X), \
1232 INSN_3(ALU64, SUB, X), \
1233 INSN_3(ALU64, AND, X), \
1234 INSN_3(ALU64, OR, X), \
1235 INSN_3(ALU64, LSH, X), \
1236 INSN_3(ALU64, RSH, X), \
1237 INSN_3(ALU64, XOR, X), \
1238 INSN_3(ALU64, MUL, X), \
1239 INSN_3(ALU64, MOV, X), \
1240 INSN_3(ALU64, ARSH, X), \
1241 INSN_3(ALU64, DIV, X), \
1242 INSN_3(ALU64, MOD, X), \
1243 INSN_2(ALU64, NEG), \
1244 /* Immediate based. */ \
1245 INSN_3(ALU64, ADD, K), \
1246 INSN_3(ALU64, SUB, K), \
1247 INSN_3(ALU64, AND, K), \
1248 INSN_3(ALU64, OR, K), \
1249 INSN_3(ALU64, LSH, K), \
1250 INSN_3(ALU64, RSH, K), \
1251 INSN_3(ALU64, XOR, K), \
1252 INSN_3(ALU64, MUL, K), \
1253 INSN_3(ALU64, MOV, K), \
1254 INSN_3(ALU64, ARSH, K), \
1255 INSN_3(ALU64, DIV, K), \
1256 INSN_3(ALU64, MOD, K), \
1257 /* Call instruction. */ \
1258 INSN_2(JMP, CALL), \
1259 /* Exit instruction. */ \
1260 INSN_2(JMP, EXIT), \
1261 /* 32-bit Jump instructions. */ \
1262 /* Register based. */ \
1263 INSN_3(JMP32, JEQ, X), \
1264 INSN_3(JMP32, JNE, X), \
1265 INSN_3(JMP32, JGT, X), \
1266 INSN_3(JMP32, JLT, X), \
1267 INSN_3(JMP32, JGE, X), \
1268 INSN_3(JMP32, JLE, X), \
1269 INSN_3(JMP32, JSGT, X), \
1270 INSN_3(JMP32, JSLT, X), \
1271 INSN_3(JMP32, JSGE, X), \
1272 INSN_3(JMP32, JSLE, X), \
1273 INSN_3(JMP32, JSET, X), \
1274 /* Immediate based. */ \
1275 INSN_3(JMP32, JEQ, K), \
1276 INSN_3(JMP32, JNE, K), \
1277 INSN_3(JMP32, JGT, K), \
1278 INSN_3(JMP32, JLT, K), \
1279 INSN_3(JMP32, JGE, K), \
1280 INSN_3(JMP32, JLE, K), \
1281 INSN_3(JMP32, JSGT, K), \
1282 INSN_3(JMP32, JSLT, K), \
1283 INSN_3(JMP32, JSGE, K), \
1284 INSN_3(JMP32, JSLE, K), \
1285 INSN_3(JMP32, JSET, K), \
1286 /* Jump instructions. */ \
1287 /* Register based. */ \
1288 INSN_3(JMP, JEQ, X), \
1289 INSN_3(JMP, JNE, X), \
1290 INSN_3(JMP, JGT, X), \
1291 INSN_3(JMP, JLT, X), \
1292 INSN_3(JMP, JGE, X), \
1293 INSN_3(JMP, JLE, X), \
1294 INSN_3(JMP, JSGT, X), \
1295 INSN_3(JMP, JSLT, X), \
1296 INSN_3(JMP, JSGE, X), \
1297 INSN_3(JMP, JSLE, X), \
1298 INSN_3(JMP, JSET, X), \
1299 /* Immediate based. */ \
1300 INSN_3(JMP, JEQ, K), \
1301 INSN_3(JMP, JNE, K), \
1302 INSN_3(JMP, JGT, K), \
1303 INSN_3(JMP, JLT, K), \
1304 INSN_3(JMP, JGE, K), \
1305 INSN_3(JMP, JLE, K), \
1306 INSN_3(JMP, JSGT, K), \
1307 INSN_3(JMP, JSLT, K), \
1308 INSN_3(JMP, JSGE, K), \
1309 INSN_3(JMP, JSLE, K), \
1310 INSN_3(JMP, JSET, K), \
1311 INSN_2(JMP, JA), \
1312 /* Store instructions. */ \
1313 /* Register based. */ \
1314 INSN_3(STX, MEM, B), \
1315 INSN_3(STX, MEM, H), \
1316 INSN_3(STX, MEM, W), \
1317 INSN_3(STX, MEM, DW), \
1318 INSN_3(STX, XADD, W), \
1319 INSN_3(STX, XADD, DW), \
1320 /* Immediate based. */ \
1321 INSN_3(ST, MEM, B), \
1322 INSN_3(ST, MEM, H), \
1323 INSN_3(ST, MEM, W), \
1324 INSN_3(ST, MEM, DW), \
1325 /* Load instructions. */ \
1326 /* Register based. */ \
1327 INSN_3(LDX, MEM, B), \
1328 INSN_3(LDX, MEM, H), \
1329 INSN_3(LDX, MEM, W), \
1330 INSN_3(LDX, MEM, DW), \
1331 /* Immediate based. */ \
1332 INSN_3(LD, IMM, DW)
1333
1334 bool bpf_opcode_in_insntable(u8 code)
1335 {
1336 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1337 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1338 static const bool public_insntable[256] = {
1339 [0 ... 255] = false,
1340 /* Now overwrite non-defaults ... */
1341 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1342 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1343 [BPF_LD | BPF_ABS | BPF_B] = true,
1344 [BPF_LD | BPF_ABS | BPF_H] = true,
1345 [BPF_LD | BPF_ABS | BPF_W] = true,
1346 [BPF_LD | BPF_IND | BPF_B] = true,
1347 [BPF_LD | BPF_IND | BPF_H] = true,
1348 [BPF_LD | BPF_IND | BPF_W] = true,
1349 };
1350 #undef BPF_INSN_3_TBL
1351 #undef BPF_INSN_2_TBL
1352 return public_insntable[code];
1353 }
1354
1355 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1356 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1357 {
1358 memset(dst, 0, size);
1359 return -EFAULT;
1360 }
1361
1362 /**
1363 * __bpf_prog_run - run eBPF program on a given context
1364 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1365 * @insn: is the array of eBPF instructions
1366 * @stack: is the eBPF storage stack
1367 *
1368 * Decode and execute eBPF instructions.
1369 */
1370 static u64 __no_fgcse ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1371 {
1372 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1373 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1374 static const void * const jumptable[256] __annotate_jump_table = {
1375 [0 ... 255] = &&default_label,
1376 /* Now overwrite non-defaults ... */
1377 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1378 /* Non-UAPI available opcodes. */
1379 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1380 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1381 [BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1382 [BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1383 [BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1384 [BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1385 };
1386 #undef BPF_INSN_3_LBL
1387 #undef BPF_INSN_2_LBL
1388 u32 tail_call_cnt = 0;
1389
1390 #define CONT ({ insn++; goto select_insn; })
1391 #define CONT_JMP ({ insn++; goto select_insn; })
1392
1393 select_insn:
1394 goto *jumptable[insn->code];
1395
1396 /* ALU */
1397 #define ALU(OPCODE, OP) \
1398 ALU64_##OPCODE##_X: \
1399 DST = DST OP SRC; \
1400 CONT; \
1401 ALU_##OPCODE##_X: \
1402 DST = (u32) DST OP (u32) SRC; \
1403 CONT; \
1404 ALU64_##OPCODE##_K: \
1405 DST = DST OP IMM; \
1406 CONT; \
1407 ALU_##OPCODE##_K: \
1408 DST = (u32) DST OP (u32) IMM; \
1409 CONT;
1410
1411 ALU(ADD, +)
1412 ALU(SUB, -)
1413 ALU(AND, &)
1414 ALU(OR, |)
1415 ALU(LSH, <<)
1416 ALU(RSH, >>)
1417 ALU(XOR, ^)
1418 ALU(MUL, *)
1419 #undef ALU
1420 ALU_NEG:
1421 DST = (u32) -DST;
1422 CONT;
1423 ALU64_NEG:
1424 DST = -DST;
1425 CONT;
1426 ALU_MOV_X:
1427 DST = (u32) SRC;
1428 CONT;
1429 ALU_MOV_K:
1430 DST = (u32) IMM;
1431 CONT;
1432 ALU64_MOV_X:
1433 DST = SRC;
1434 CONT;
1435 ALU64_MOV_K:
1436 DST = IMM;
1437 CONT;
1438 LD_IMM_DW:
1439 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1440 insn++;
1441 CONT;
1442 ALU_ARSH_X:
1443 DST = (u64) (u32) (((s32) DST) >> SRC);
1444 CONT;
1445 ALU_ARSH_K:
1446 DST = (u64) (u32) (((s32) DST) >> IMM);
1447 CONT;
1448 ALU64_ARSH_X:
1449 (*(s64 *) &DST) >>= SRC;
1450 CONT;
1451 ALU64_ARSH_K:
1452 (*(s64 *) &DST) >>= IMM;
1453 CONT;
1454 ALU64_MOD_X:
1455 div64_u64_rem(DST, SRC, &AX);
1456 DST = AX;
1457 CONT;
1458 ALU_MOD_X:
1459 AX = (u32) DST;
1460 DST = do_div(AX, (u32) SRC);
1461 CONT;
1462 ALU64_MOD_K:
1463 div64_u64_rem(DST, IMM, &AX);
1464 DST = AX;
1465 CONT;
1466 ALU_MOD_K:
1467 AX = (u32) DST;
1468 DST = do_div(AX, (u32) IMM);
1469 CONT;
1470 ALU64_DIV_X:
1471 DST = div64_u64(DST, SRC);
1472 CONT;
1473 ALU_DIV_X:
1474 AX = (u32) DST;
1475 do_div(AX, (u32) SRC);
1476 DST = (u32) AX;
1477 CONT;
1478 ALU64_DIV_K:
1479 DST = div64_u64(DST, IMM);
1480 CONT;
1481 ALU_DIV_K:
1482 AX = (u32) DST;
1483 do_div(AX, (u32) IMM);
1484 DST = (u32) AX;
1485 CONT;
1486 ALU_END_TO_BE:
1487 switch (IMM) {
1488 case 16:
1489 DST = (__force u16) cpu_to_be16(DST);
1490 break;
1491 case 32:
1492 DST = (__force u32) cpu_to_be32(DST);
1493 break;
1494 case 64:
1495 DST = (__force u64) cpu_to_be64(DST);
1496 break;
1497 }
1498 CONT;
1499 ALU_END_TO_LE:
1500 switch (IMM) {
1501 case 16:
1502 DST = (__force u16) cpu_to_le16(DST);
1503 break;
1504 case 32:
1505 DST = (__force u32) cpu_to_le32(DST);
1506 break;
1507 case 64:
1508 DST = (__force u64) cpu_to_le64(DST);
1509 break;
1510 }
1511 CONT;
1512
1513 /* CALL */
1514 JMP_CALL:
1515 /* Function call scratches BPF_R1-BPF_R5 registers,
1516 * preserves BPF_R6-BPF_R9, and stores return value
1517 * into BPF_R0.
1518 */
1519 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1520 BPF_R4, BPF_R5);
1521 CONT;
1522
1523 JMP_CALL_ARGS:
1524 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1525 BPF_R3, BPF_R4,
1526 BPF_R5,
1527 insn + insn->off + 1);
1528 CONT;
1529
1530 JMP_TAIL_CALL: {
1531 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1532 struct bpf_array *array = container_of(map, struct bpf_array, map);
1533 struct bpf_prog *prog;
1534 u32 index = BPF_R3;
1535
1536 if (unlikely(index >= array->map.max_entries))
1537 goto out;
1538 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1539 goto out;
1540
1541 tail_call_cnt++;
1542
1543 prog = READ_ONCE(array->ptrs[index]);
1544 if (!prog)
1545 goto out;
1546
1547 /* ARG1 at this point is guaranteed to point to CTX from
1548 * the verifier side due to the fact that the tail call is
1549 * handled like a helper, that is, bpf_tail_call_proto,
1550 * where arg1_type is ARG_PTR_TO_CTX.
1551 */
1552 insn = prog->insnsi;
1553 goto select_insn;
1554 out:
1555 CONT;
1556 }
1557 JMP_JA:
1558 insn += insn->off;
1559 CONT;
1560 JMP_EXIT:
1561 return BPF_R0;
1562 /* JMP */
1563 #define COND_JMP(SIGN, OPCODE, CMP_OP) \
1564 JMP_##OPCODE##_X: \
1565 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
1566 insn += insn->off; \
1567 CONT_JMP; \
1568 } \
1569 CONT; \
1570 JMP32_##OPCODE##_X: \
1571 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
1572 insn += insn->off; \
1573 CONT_JMP; \
1574 } \
1575 CONT; \
1576 JMP_##OPCODE##_K: \
1577 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
1578 insn += insn->off; \
1579 CONT_JMP; \
1580 } \
1581 CONT; \
1582 JMP32_##OPCODE##_K: \
1583 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
1584 insn += insn->off; \
1585 CONT_JMP; \
1586 } \
1587 CONT;
1588 COND_JMP(u, JEQ, ==)
1589 COND_JMP(u, JNE, !=)
1590 COND_JMP(u, JGT, >)
1591 COND_JMP(u, JLT, <)
1592 COND_JMP(u, JGE, >=)
1593 COND_JMP(u, JLE, <=)
1594 COND_JMP(u, JSET, &)
1595 COND_JMP(s, JSGT, >)
1596 COND_JMP(s, JSLT, <)
1597 COND_JMP(s, JSGE, >=)
1598 COND_JMP(s, JSLE, <=)
1599 #undef COND_JMP
1600 /* STX and ST and LDX*/
1601 #define LDST(SIZEOP, SIZE) \
1602 STX_MEM_##SIZEOP: \
1603 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1604 CONT; \
1605 ST_MEM_##SIZEOP: \
1606 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1607 CONT; \
1608 LDX_MEM_##SIZEOP: \
1609 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1610 CONT;
1611
1612 LDST(B, u8)
1613 LDST(H, u16)
1614 LDST(W, u32)
1615 LDST(DW, u64)
1616 #undef LDST
1617 #define LDX_PROBE(SIZEOP, SIZE) \
1618 LDX_PROBE_MEM_##SIZEOP: \
1619 bpf_probe_read_kernel(&DST, SIZE, (const void *)(long) (SRC + insn->off)); \
1620 CONT;
1621 LDX_PROBE(B, 1)
1622 LDX_PROBE(H, 2)
1623 LDX_PROBE(W, 4)
1624 LDX_PROBE(DW, 8)
1625 #undef LDX_PROBE
1626
1627 STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1628 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1629 (DST + insn->off));
1630 CONT;
1631 STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1632 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1633 (DST + insn->off));
1634 CONT;
1635
1636 default_label:
1637 /* If we ever reach this, we have a bug somewhere. Die hard here
1638 * instead of just returning 0; we could be somewhere in a subprog,
1639 * so execution could continue otherwise which we do /not/ want.
1640 *
1641 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1642 */
1643 pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1644 BUG_ON(1);
1645 return 0;
1646 }
1647
1648 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1649 #define DEFINE_BPF_PROG_RUN(stack_size) \
1650 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1651 { \
1652 u64 stack[stack_size / sizeof(u64)]; \
1653 u64 regs[MAX_BPF_EXT_REG]; \
1654 \
1655 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1656 ARG1 = (u64) (unsigned long) ctx; \
1657 return ___bpf_prog_run(regs, insn, stack); \
1658 }
1659
1660 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1661 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1662 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1663 const struct bpf_insn *insn) \
1664 { \
1665 u64 stack[stack_size / sizeof(u64)]; \
1666 u64 regs[MAX_BPF_EXT_REG]; \
1667 \
1668 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1669 BPF_R1 = r1; \
1670 BPF_R2 = r2; \
1671 BPF_R3 = r3; \
1672 BPF_R4 = r4; \
1673 BPF_R5 = r5; \
1674 return ___bpf_prog_run(regs, insn, stack); \
1675 }
1676
1677 #define EVAL1(FN, X) FN(X)
1678 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1679 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1680 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1681 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1682 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1683
1684 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1685 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1686 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1687
1688 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1689 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1690 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1691
1692 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1693
1694 static unsigned int (*interpreters[])(const void *ctx,
1695 const struct bpf_insn *insn) = {
1696 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1697 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1698 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1699 };
1700 #undef PROG_NAME_LIST
1701 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1702 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1703 const struct bpf_insn *insn) = {
1704 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1705 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1706 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1707 };
1708 #undef PROG_NAME_LIST
1709
1710 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1711 {
1712 stack_depth = max_t(u32, stack_depth, 1);
1713 insn->off = (s16) insn->imm;
1714 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1715 __bpf_call_base_args;
1716 insn->code = BPF_JMP | BPF_CALL_ARGS;
1717 }
1718
1719 #else
1720 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1721 const struct bpf_insn *insn)
1722 {
1723 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1724 * is not working properly, so warn about it!
1725 */
1726 WARN_ON_ONCE(1);
1727 return 0;
1728 }
1729 #endif
1730
1731 bool bpf_prog_array_compatible(struct bpf_array *array,
1732 const struct bpf_prog *fp)
1733 {
1734 if (fp->kprobe_override)
1735 return false;
1736
1737 if (!array->aux->type) {
1738 /* There's no owner yet where we could check for
1739 * compatibility.
1740 */
1741 array->aux->type = fp->type;
1742 array->aux->jited = fp->jited;
1743 return true;
1744 }
1745
1746 return array->aux->type == fp->type &&
1747 array->aux->jited == fp->jited;
1748 }
1749
1750 static int bpf_check_tail_call(const struct bpf_prog *fp)
1751 {
1752 struct bpf_prog_aux *aux = fp->aux;
1753 int i, ret = 0;
1754
1755 mutex_lock(&aux->used_maps_mutex);
1756 for (i = 0; i < aux->used_map_cnt; i++) {
1757 struct bpf_map *map = aux->used_maps[i];
1758 struct bpf_array *array;
1759
1760 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1761 continue;
1762
1763 array = container_of(map, struct bpf_array, map);
1764 if (!bpf_prog_array_compatible(array, fp)) {
1765 ret = -EINVAL;
1766 goto out;
1767 }
1768 }
1769
1770 out:
1771 mutex_unlock(&aux->used_maps_mutex);
1772 return ret;
1773 }
1774
1775 static void bpf_prog_select_func(struct bpf_prog *fp)
1776 {
1777 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1778 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1779
1780 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1781 #else
1782 fp->bpf_func = __bpf_prog_ret0_warn;
1783 #endif
1784 }
1785
1786 /**
1787 * bpf_prog_select_runtime - select exec runtime for BPF program
1788 * @fp: bpf_prog populated with internal BPF program
1789 * @err: pointer to error variable
1790 *
1791 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1792 * The BPF program will be executed via BPF_PROG_RUN() macro.
1793 */
1794 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1795 {
1796 /* In case of BPF to BPF calls, verifier did all the prep
1797 * work with regards to JITing, etc.
1798 */
1799 if (fp->bpf_func)
1800 goto finalize;
1801
1802 bpf_prog_select_func(fp);
1803
1804 /* eBPF JITs can rewrite the program in case constant
1805 * blinding is active. However, in case of error during
1806 * blinding, bpf_int_jit_compile() must always return a
1807 * valid program, which in this case would simply not
1808 * be JITed, but falls back to the interpreter.
1809 */
1810 if (!bpf_prog_is_dev_bound(fp->aux)) {
1811 *err = bpf_prog_alloc_jited_linfo(fp);
1812 if (*err)
1813 return fp;
1814
1815 fp = bpf_int_jit_compile(fp);
1816 if (!fp->jited) {
1817 bpf_prog_free_jited_linfo(fp);
1818 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1819 *err = -ENOTSUPP;
1820 return fp;
1821 #endif
1822 } else {
1823 bpf_prog_free_unused_jited_linfo(fp);
1824 }
1825 } else {
1826 *err = bpf_prog_offload_compile(fp);
1827 if (*err)
1828 return fp;
1829 }
1830
1831 finalize:
1832 bpf_prog_lock_ro(fp);
1833
1834 /* The tail call compatibility check can only be done at
1835 * this late stage as we need to determine, if we deal
1836 * with JITed or non JITed program concatenations and not
1837 * all eBPF JITs might immediately support all features.
1838 */
1839 *err = bpf_check_tail_call(fp);
1840
1841 return fp;
1842 }
1843 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1844
1845 static unsigned int __bpf_prog_ret1(const void *ctx,
1846 const struct bpf_insn *insn)
1847 {
1848 return 1;
1849 }
1850
1851 static struct bpf_prog_dummy {
1852 struct bpf_prog prog;
1853 } dummy_bpf_prog = {
1854 .prog = {
1855 .bpf_func = __bpf_prog_ret1,
1856 },
1857 };
1858
1859 /* to avoid allocating empty bpf_prog_array for cgroups that
1860 * don't have bpf program attached use one global 'empty_prog_array'
1861 * It will not be modified the caller of bpf_prog_array_alloc()
1862 * (since caller requested prog_cnt == 0)
1863 * that pointer should be 'freed' by bpf_prog_array_free()
1864 */
1865 static struct {
1866 struct bpf_prog_array hdr;
1867 struct bpf_prog *null_prog;
1868 } empty_prog_array = {
1869 .null_prog = NULL,
1870 };
1871
1872 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1873 {
1874 if (prog_cnt)
1875 return kzalloc(sizeof(struct bpf_prog_array) +
1876 sizeof(struct bpf_prog_array_item) *
1877 (prog_cnt + 1),
1878 flags);
1879
1880 return &empty_prog_array.hdr;
1881 }
1882
1883 void bpf_prog_array_free(struct bpf_prog_array *progs)
1884 {
1885 if (!progs || progs == &empty_prog_array.hdr)
1886 return;
1887 kfree_rcu(progs, rcu);
1888 }
1889
1890 int bpf_prog_array_length(struct bpf_prog_array *array)
1891 {
1892 struct bpf_prog_array_item *item;
1893 u32 cnt = 0;
1894
1895 for (item = array->items; item->prog; item++)
1896 if (item->prog != &dummy_bpf_prog.prog)
1897 cnt++;
1898 return cnt;
1899 }
1900
1901 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
1902 {
1903 struct bpf_prog_array_item *item;
1904
1905 for (item = array->items; item->prog; item++)
1906 if (item->prog != &dummy_bpf_prog.prog)
1907 return false;
1908 return true;
1909 }
1910
1911 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
1912 u32 *prog_ids,
1913 u32 request_cnt)
1914 {
1915 struct bpf_prog_array_item *item;
1916 int i = 0;
1917
1918 for (item = array->items; item->prog; item++) {
1919 if (item->prog == &dummy_bpf_prog.prog)
1920 continue;
1921 prog_ids[i] = item->prog->aux->id;
1922 if (++i == request_cnt) {
1923 item++;
1924 break;
1925 }
1926 }
1927
1928 return !!(item->prog);
1929 }
1930
1931 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
1932 __u32 __user *prog_ids, u32 cnt)
1933 {
1934 unsigned long err = 0;
1935 bool nospc;
1936 u32 *ids;
1937
1938 /* users of this function are doing:
1939 * cnt = bpf_prog_array_length();
1940 * if (cnt > 0)
1941 * bpf_prog_array_copy_to_user(..., cnt);
1942 * so below kcalloc doesn't need extra cnt > 0 check.
1943 */
1944 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1945 if (!ids)
1946 return -ENOMEM;
1947 nospc = bpf_prog_array_copy_core(array, ids, cnt);
1948 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1949 kfree(ids);
1950 if (err)
1951 return -EFAULT;
1952 if (nospc)
1953 return -ENOSPC;
1954 return 0;
1955 }
1956
1957 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
1958 struct bpf_prog *old_prog)
1959 {
1960 struct bpf_prog_array_item *item;
1961
1962 for (item = array->items; item->prog; item++)
1963 if (item->prog == old_prog) {
1964 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1965 break;
1966 }
1967 }
1968
1969 /**
1970 * bpf_prog_array_delete_safe_at() - Replaces the program at the given
1971 * index into the program array with
1972 * a dummy no-op program.
1973 * @array: a bpf_prog_array
1974 * @index: the index of the program to replace
1975 *
1976 * Skips over dummy programs, by not counting them, when calculating
1977 * the position of the program to replace.
1978 *
1979 * Return:
1980 * * 0 - Success
1981 * * -EINVAL - Invalid index value. Must be a non-negative integer.
1982 * * -ENOENT - Index out of range
1983 */
1984 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
1985 {
1986 return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
1987 }
1988
1989 /**
1990 * bpf_prog_array_update_at() - Updates the program at the given index
1991 * into the program array.
1992 * @array: a bpf_prog_array
1993 * @index: the index of the program to update
1994 * @prog: the program to insert into the array
1995 *
1996 * Skips over dummy programs, by not counting them, when calculating
1997 * the position of the program to update.
1998 *
1999 * Return:
2000 * * 0 - Success
2001 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2002 * * -ENOENT - Index out of range
2003 */
2004 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2005 struct bpf_prog *prog)
2006 {
2007 struct bpf_prog_array_item *item;
2008
2009 if (unlikely(index < 0))
2010 return -EINVAL;
2011
2012 for (item = array->items; item->prog; item++) {
2013 if (item->prog == &dummy_bpf_prog.prog)
2014 continue;
2015 if (!index) {
2016 WRITE_ONCE(item->prog, prog);
2017 return 0;
2018 }
2019 index--;
2020 }
2021 return -ENOENT;
2022 }
2023
2024 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2025 struct bpf_prog *exclude_prog,
2026 struct bpf_prog *include_prog,
2027 struct bpf_prog_array **new_array)
2028 {
2029 int new_prog_cnt, carry_prog_cnt = 0;
2030 struct bpf_prog_array_item *existing;
2031 struct bpf_prog_array *array;
2032 bool found_exclude = false;
2033 int new_prog_idx = 0;
2034
2035 /* Figure out how many existing progs we need to carry over to
2036 * the new array.
2037 */
2038 if (old_array) {
2039 existing = old_array->items;
2040 for (; existing->prog; existing++) {
2041 if (existing->prog == exclude_prog) {
2042 found_exclude = true;
2043 continue;
2044 }
2045 if (existing->prog != &dummy_bpf_prog.prog)
2046 carry_prog_cnt++;
2047 if (existing->prog == include_prog)
2048 return -EEXIST;
2049 }
2050 }
2051
2052 if (exclude_prog && !found_exclude)
2053 return -ENOENT;
2054
2055 /* How many progs (not NULL) will be in the new array? */
2056 new_prog_cnt = carry_prog_cnt;
2057 if (include_prog)
2058 new_prog_cnt += 1;
2059
2060 /* Do we have any prog (not NULL) in the new array? */
2061 if (!new_prog_cnt) {
2062 *new_array = NULL;
2063 return 0;
2064 }
2065
2066 /* +1 as the end of prog_array is marked with NULL */
2067 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2068 if (!array)
2069 return -ENOMEM;
2070
2071 /* Fill in the new prog array */
2072 if (carry_prog_cnt) {
2073 existing = old_array->items;
2074 for (; existing->prog; existing++)
2075 if (existing->prog != exclude_prog &&
2076 existing->prog != &dummy_bpf_prog.prog) {
2077 array->items[new_prog_idx++].prog =
2078 existing->prog;
2079 }
2080 }
2081 if (include_prog)
2082 array->items[new_prog_idx++].prog = include_prog;
2083 array->items[new_prog_idx].prog = NULL;
2084 *new_array = array;
2085 return 0;
2086 }
2087
2088 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2089 u32 *prog_ids, u32 request_cnt,
2090 u32 *prog_cnt)
2091 {
2092 u32 cnt = 0;
2093
2094 if (array)
2095 cnt = bpf_prog_array_length(array);
2096
2097 *prog_cnt = cnt;
2098
2099 /* return early if user requested only program count or nothing to copy */
2100 if (!request_cnt || !cnt)
2101 return 0;
2102
2103 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2104 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2105 : 0;
2106 }
2107
2108 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2109 struct bpf_map **used_maps, u32 len)
2110 {
2111 struct bpf_map *map;
2112 u32 i;
2113
2114 for (i = 0; i < len; i++) {
2115 map = used_maps[i];
2116 if (map->ops->map_poke_untrack)
2117 map->ops->map_poke_untrack(map, aux);
2118 bpf_map_put(map);
2119 }
2120 }
2121
2122 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2123 {
2124 __bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2125 kfree(aux->used_maps);
2126 }
2127
2128 static void bpf_prog_free_deferred(struct work_struct *work)
2129 {
2130 struct bpf_prog_aux *aux;
2131 int i;
2132
2133 aux = container_of(work, struct bpf_prog_aux, work);
2134 bpf_free_used_maps(aux);
2135 if (bpf_prog_is_dev_bound(aux))
2136 bpf_prog_offload_destroy(aux->prog);
2137 #ifdef CONFIG_PERF_EVENTS
2138 if (aux->prog->has_callchain_buf)
2139 put_callchain_buffers();
2140 #endif
2141 bpf_trampoline_put(aux->trampoline);
2142 for (i = 0; i < aux->func_cnt; i++)
2143 bpf_jit_free(aux->func[i]);
2144 if (aux->func_cnt) {
2145 kfree(aux->func);
2146 bpf_prog_unlock_free(aux->prog);
2147 } else {
2148 bpf_jit_free(aux->prog);
2149 }
2150 }
2151
2152 /* Free internal BPF program */
2153 void bpf_prog_free(struct bpf_prog *fp)
2154 {
2155 struct bpf_prog_aux *aux = fp->aux;
2156
2157 if (aux->linked_prog)
2158 bpf_prog_put(aux->linked_prog);
2159 INIT_WORK(&aux->work, bpf_prog_free_deferred);
2160 schedule_work(&aux->work);
2161 }
2162 EXPORT_SYMBOL_GPL(bpf_prog_free);
2163
2164 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2165 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2166
2167 void bpf_user_rnd_init_once(void)
2168 {
2169 prandom_init_once(&bpf_user_rnd_state);
2170 }
2171
2172 BPF_CALL_0(bpf_user_rnd_u32)
2173 {
2174 /* Should someone ever have the rather unwise idea to use some
2175 * of the registers passed into this function, then note that
2176 * this function is called from native eBPF and classic-to-eBPF
2177 * transformations. Register assignments from both sides are
2178 * different, f.e. classic always sets fn(ctx, A, X) here.
2179 */
2180 struct rnd_state *state;
2181 u32 res;
2182
2183 state = &get_cpu_var(bpf_user_rnd_state);
2184 res = prandom_u32_state(state);
2185 put_cpu_var(bpf_user_rnd_state);
2186
2187 return res;
2188 }
2189
2190 BPF_CALL_0(bpf_get_raw_cpu_id)
2191 {
2192 return raw_smp_processor_id();
2193 }
2194
2195 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2196 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2197 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2198 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2199 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2200 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2201 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2202 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2203 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2204 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2205
2206 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2207 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2208 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2209 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2210 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2211
2212 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2213 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2214 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2215 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2216 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2217 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2218 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2219 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2220
2221 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2222 {
2223 return NULL;
2224 }
2225
2226 u64 __weak
2227 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2228 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2229 {
2230 return -ENOTSUPP;
2231 }
2232 EXPORT_SYMBOL_GPL(bpf_event_output);
2233
2234 /* Always built-in helper functions. */
2235 const struct bpf_func_proto bpf_tail_call_proto = {
2236 .func = NULL,
2237 .gpl_only = false,
2238 .ret_type = RET_VOID,
2239 .arg1_type = ARG_PTR_TO_CTX,
2240 .arg2_type = ARG_CONST_MAP_PTR,
2241 .arg3_type = ARG_ANYTHING,
2242 };
2243
2244 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2245 * It is encouraged to implement bpf_int_jit_compile() instead, so that
2246 * eBPF and implicitly also cBPF can get JITed!
2247 */
2248 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2249 {
2250 return prog;
2251 }
2252
2253 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2254 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2255 */
2256 void __weak bpf_jit_compile(struct bpf_prog *prog)
2257 {
2258 }
2259
2260 bool __weak bpf_helper_changes_pkt_data(void *func)
2261 {
2262 return false;
2263 }
2264
2265 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2266 * analysis code and wants explicit zero extension inserted by verifier.
2267 * Otherwise, return FALSE.
2268 */
2269 bool __weak bpf_jit_needs_zext(void)
2270 {
2271 return false;
2272 }
2273
2274 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2275 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2276 */
2277 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2278 int len)
2279 {
2280 return -EFAULT;
2281 }
2282
2283 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2284 void *addr1, void *addr2)
2285 {
2286 return -ENOTSUPP;
2287 }
2288
2289 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2290 EXPORT_SYMBOL(bpf_stats_enabled_key);
2291
2292 /* All definitions of tracepoints related to BPF. */
2293 #define CREATE_TRACE_POINTS
2294 #include <linux/bpf_trace.h>
2295
2296 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2297 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
Ubuntu Hirsute Linux kernel mirror