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/objtool.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 <linux/bpf_verifier.h>
36 #include <linux/nodemask.h>
37
38 #include <asm/barrier.h>
39 #include <asm/unaligned.h>
40
41 /* Registers */
42 #define BPF_R0 regs[BPF_REG_0]
43 #define BPF_R1 regs[BPF_REG_1]
44 #define BPF_R2 regs[BPF_REG_2]
45 #define BPF_R3 regs[BPF_REG_3]
46 #define BPF_R4 regs[BPF_REG_4]
47 #define BPF_R5 regs[BPF_REG_5]
48 #define BPF_R6 regs[BPF_REG_6]
49 #define BPF_R7 regs[BPF_REG_7]
50 #define BPF_R8 regs[BPF_REG_8]
51 #define BPF_R9 regs[BPF_REG_9]
52 #define BPF_R10 regs[BPF_REG_10]
53
54 /* Named registers */
55 #define DST regs[insn->dst_reg]
56 #define SRC regs[insn->src_reg]
57 #define FP regs[BPF_REG_FP]
58 #define AX regs[BPF_REG_AX]
59 #define ARG1 regs[BPF_REG_ARG1]
60 #define CTX regs[BPF_REG_CTX]
61 #define IMM insn->imm
62
63 /* No hurry in this branch
64 *
65 * Exported for the bpf jit load helper.
66 */
bpf_internal_load_pointer_neg_helper(const struct sk_buff * skb,int k,unsigned int size)67 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
68 {
69 u8 *ptr = NULL;
70
71 if (k >= SKF_NET_OFF) {
72 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
73 } else if (k >= SKF_LL_OFF) {
74 if (unlikely(!skb_mac_header_was_set(skb)))
75 return NULL;
76 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
77 }
78 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
79 return ptr;
80
81 return NULL;
82 }
83
bpf_prog_alloc_no_stats(unsigned int size,gfp_t gfp_extra_flags)84 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
85 {
86 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
87 struct bpf_prog_aux *aux;
88 struct bpf_prog *fp;
89
90 size = round_up(size, PAGE_SIZE);
91 fp = __vmalloc(size, gfp_flags);
92 if (fp == NULL)
93 return NULL;
94
95 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT | gfp_extra_flags);
96 if (aux == NULL) {
97 vfree(fp);
98 return NULL;
99 }
100 fp->active = alloc_percpu_gfp(int, GFP_KERNEL_ACCOUNT | gfp_extra_flags);
101 if (!fp->active) {
102 vfree(fp);
103 kfree(aux);
104 return NULL;
105 }
106
107 fp->pages = size / PAGE_SIZE;
108 fp->aux = aux;
109 fp->aux->prog = fp;
110 fp->jit_requested = ebpf_jit_enabled();
111 fp->blinding_requested = bpf_jit_blinding_enabled(fp);
112 #ifdef CONFIG_CGROUP_BPF
113 aux->cgroup_atype = CGROUP_BPF_ATTACH_TYPE_INVALID;
114 #endif
115
116 INIT_LIST_HEAD_RCU(&fp->aux->ksym.lnode);
117 mutex_init(&fp->aux->used_maps_mutex);
118 mutex_init(&fp->aux->dst_mutex);
119
120 return fp;
121 }
122
bpf_prog_alloc(unsigned int size,gfp_t gfp_extra_flags)123 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
124 {
125 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
126 struct bpf_prog *prog;
127 int cpu;
128
129 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
130 if (!prog)
131 return NULL;
132
133 prog->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
134 if (!prog->stats) {
135 free_percpu(prog->active);
136 kfree(prog->aux);
137 vfree(prog);
138 return NULL;
139 }
140
141 for_each_possible_cpu(cpu) {
142 struct bpf_prog_stats *pstats;
143
144 pstats = per_cpu_ptr(prog->stats, cpu);
145 u64_stats_init(&pstats->syncp);
146 }
147 return prog;
148 }
149 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
150
bpf_prog_alloc_jited_linfo(struct bpf_prog * prog)151 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
152 {
153 if (!prog->aux->nr_linfo || !prog->jit_requested)
154 return 0;
155
156 prog->aux->jited_linfo = kvcalloc(prog->aux->nr_linfo,
157 sizeof(*prog->aux->jited_linfo),
158 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
159 if (!prog->aux->jited_linfo)
160 return -ENOMEM;
161
162 return 0;
163 }
164
bpf_prog_jit_attempt_done(struct bpf_prog * prog)165 void bpf_prog_jit_attempt_done(struct bpf_prog *prog)
166 {
167 if (prog->aux->jited_linfo &&
168 (!prog->jited || !prog->aux->jited_linfo[0])) {
169 kvfree(prog->aux->jited_linfo);
170 prog->aux->jited_linfo = NULL;
171 }
172
173 kfree(prog->aux->kfunc_tab);
174 prog->aux->kfunc_tab = NULL;
175 }
176
177 /* The jit engine is responsible to provide an array
178 * for insn_off to the jited_off mapping (insn_to_jit_off).
179 *
180 * The idx to this array is the insn_off. Hence, the insn_off
181 * here is relative to the prog itself instead of the main prog.
182 * This array has one entry for each xlated bpf insn.
183 *
184 * jited_off is the byte off to the end of the jited insn.
185 *
186 * Hence, with
187 * insn_start:
188 * The first bpf insn off of the prog. The insn off
189 * here is relative to the main prog.
190 * e.g. if prog is a subprog, insn_start > 0
191 * linfo_idx:
192 * The prog's idx to prog->aux->linfo and jited_linfo
193 *
194 * jited_linfo[linfo_idx] = prog->bpf_func
195 *
196 * For i > linfo_idx,
197 *
198 * jited_linfo[i] = prog->bpf_func +
199 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
200 */
bpf_prog_fill_jited_linfo(struct bpf_prog * prog,const u32 * insn_to_jit_off)201 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
202 const u32 *insn_to_jit_off)
203 {
204 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
205 const struct bpf_line_info *linfo;
206 void **jited_linfo;
207
208 if (!prog->aux->jited_linfo)
209 /* Userspace did not provide linfo */
210 return;
211
212 linfo_idx = prog->aux->linfo_idx;
213 linfo = &prog->aux->linfo[linfo_idx];
214 insn_start = linfo[0].insn_off;
215 insn_end = insn_start + prog->len;
216
217 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
218 jited_linfo[0] = prog->bpf_func;
219
220 nr_linfo = prog->aux->nr_linfo - linfo_idx;
221
222 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
223 /* The verifier ensures that linfo[i].insn_off is
224 * strictly increasing
225 */
226 jited_linfo[i] = prog->bpf_func +
227 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
228 }
229
bpf_prog_realloc(struct bpf_prog * fp_old,unsigned int size,gfp_t gfp_extra_flags)230 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
231 gfp_t gfp_extra_flags)
232 {
233 gfp_t gfp_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO | gfp_extra_flags;
234 struct bpf_prog *fp;
235 u32 pages;
236
237 size = round_up(size, PAGE_SIZE);
238 pages = size / PAGE_SIZE;
239 if (pages <= fp_old->pages)
240 return fp_old;
241
242 fp = __vmalloc(size, gfp_flags);
243 if (fp) {
244 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
245 fp->pages = pages;
246 fp->aux->prog = fp;
247
248 /* We keep fp->aux from fp_old around in the new
249 * reallocated structure.
250 */
251 fp_old->aux = NULL;
252 fp_old->stats = NULL;
253 fp_old->active = NULL;
254 __bpf_prog_free(fp_old);
255 }
256
257 return fp;
258 }
259
__bpf_prog_free(struct bpf_prog * fp)260 void __bpf_prog_free(struct bpf_prog *fp)
261 {
262 if (fp->aux) {
263 mutex_destroy(&fp->aux->used_maps_mutex);
264 mutex_destroy(&fp->aux->dst_mutex);
265 kfree(fp->aux->poke_tab);
266 kfree(fp->aux);
267 }
268 free_percpu(fp->stats);
269 free_percpu(fp->active);
270 vfree(fp);
271 }
272
bpf_prog_calc_tag(struct bpf_prog * fp)273 int bpf_prog_calc_tag(struct bpf_prog *fp)
274 {
275 const u32 bits_offset = SHA1_BLOCK_SIZE - sizeof(__be64);
276 u32 raw_size = bpf_prog_tag_scratch_size(fp);
277 u32 digest[SHA1_DIGEST_WORDS];
278 u32 ws[SHA1_WORKSPACE_WORDS];
279 u32 i, bsize, psize, blocks;
280 struct bpf_insn *dst;
281 bool was_ld_map;
282 u8 *raw, *todo;
283 __be32 *result;
284 __be64 *bits;
285
286 raw = vmalloc(raw_size);
287 if (!raw)
288 return -ENOMEM;
289
290 sha1_init(digest);
291 memset(ws, 0, sizeof(ws));
292
293 /* We need to take out the map fd for the digest calculation
294 * since they are unstable from user space side.
295 */
296 dst = (void *)raw;
297 for (i = 0, was_ld_map = false; i < fp->len; i++) {
298 dst[i] = fp->insnsi[i];
299 if (!was_ld_map &&
300 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
301 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
302 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
303 was_ld_map = true;
304 dst[i].imm = 0;
305 } else if (was_ld_map &&
306 dst[i].code == 0 &&
307 dst[i].dst_reg == 0 &&
308 dst[i].src_reg == 0 &&
309 dst[i].off == 0) {
310 was_ld_map = false;
311 dst[i].imm = 0;
312 } else {
313 was_ld_map = false;
314 }
315 }
316
317 psize = bpf_prog_insn_size(fp);
318 memset(&raw[psize], 0, raw_size - psize);
319 raw[psize++] = 0x80;
320
321 bsize = round_up(psize, SHA1_BLOCK_SIZE);
322 blocks = bsize / SHA1_BLOCK_SIZE;
323 todo = raw;
324 if (bsize - psize >= sizeof(__be64)) {
325 bits = (__be64 *)(todo + bsize - sizeof(__be64));
326 } else {
327 bits = (__be64 *)(todo + bsize + bits_offset);
328 blocks++;
329 }
330 *bits = cpu_to_be64((psize - 1) << 3);
331
332 while (blocks--) {
333 sha1_transform(digest, todo, ws);
334 todo += SHA1_BLOCK_SIZE;
335 }
336
337 result = (__force __be32 *)digest;
338 for (i = 0; i < SHA1_DIGEST_WORDS; i++)
339 result[i] = cpu_to_be32(digest[i]);
340 memcpy(fp->tag, result, sizeof(fp->tag));
341
342 vfree(raw);
343 return 0;
344 }
345
bpf_adj_delta_to_imm(struct bpf_insn * insn,u32 pos,s32 end_old,s32 end_new,s32 curr,const bool probe_pass)346 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
347 s32 end_new, s32 curr, const bool probe_pass)
348 {
349 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
350 s32 delta = end_new - end_old;
351 s64 imm = insn->imm;
352
353 if (curr < pos && curr + imm + 1 >= end_old)
354 imm += delta;
355 else if (curr >= end_new && curr + imm + 1 < end_new)
356 imm -= delta;
357 if (imm < imm_min || imm > imm_max)
358 return -ERANGE;
359 if (!probe_pass)
360 insn->imm = imm;
361 return 0;
362 }
363
bpf_adj_delta_to_off(struct bpf_insn * insn,u32 pos,s32 end_old,s32 end_new,s32 curr,const bool probe_pass)364 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
365 s32 end_new, s32 curr, const bool probe_pass)
366 {
367 const s32 off_min = S16_MIN, off_max = S16_MAX;
368 s32 delta = end_new - end_old;
369 s32 off = insn->off;
370
371 if (curr < pos && curr + off + 1 >= end_old)
372 off += delta;
373 else if (curr >= end_new && curr + off + 1 < end_new)
374 off -= delta;
375 if (off < off_min || off > off_max)
376 return -ERANGE;
377 if (!probe_pass)
378 insn->off = off;
379 return 0;
380 }
381
bpf_adj_branches(struct bpf_prog * prog,u32 pos,s32 end_old,s32 end_new,const bool probe_pass)382 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
383 s32 end_new, const bool probe_pass)
384 {
385 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
386 struct bpf_insn *insn = prog->insnsi;
387 int ret = 0;
388
389 for (i = 0; i < insn_cnt; i++, insn++) {
390 u8 code;
391
392 /* In the probing pass we still operate on the original,
393 * unpatched image in order to check overflows before we
394 * do any other adjustments. Therefore skip the patchlet.
395 */
396 if (probe_pass && i == pos) {
397 i = end_new;
398 insn = prog->insnsi + end_old;
399 }
400 if (bpf_pseudo_func(insn)) {
401 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
402 end_new, i, probe_pass);
403 if (ret)
404 return ret;
405 continue;
406 }
407 code = insn->code;
408 if ((BPF_CLASS(code) != BPF_JMP &&
409 BPF_CLASS(code) != BPF_JMP32) ||
410 BPF_OP(code) == BPF_EXIT)
411 continue;
412 /* Adjust offset of jmps if we cross patch boundaries. */
413 if (BPF_OP(code) == BPF_CALL) {
414 if (insn->src_reg != BPF_PSEUDO_CALL)
415 continue;
416 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
417 end_new, i, probe_pass);
418 } else {
419 ret = bpf_adj_delta_to_off(insn, pos, end_old,
420 end_new, i, probe_pass);
421 }
422 if (ret)
423 break;
424 }
425
426 return ret;
427 }
428
bpf_adj_linfo(struct bpf_prog * prog,u32 off,u32 delta)429 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
430 {
431 struct bpf_line_info *linfo;
432 u32 i, nr_linfo;
433
434 nr_linfo = prog->aux->nr_linfo;
435 if (!nr_linfo || !delta)
436 return;
437
438 linfo = prog->aux->linfo;
439
440 for (i = 0; i < nr_linfo; i++)
441 if (off < linfo[i].insn_off)
442 break;
443
444 /* Push all off < linfo[i].insn_off by delta */
445 for (; i < nr_linfo; i++)
446 linfo[i].insn_off += delta;
447 }
448
bpf_patch_insn_single(struct bpf_prog * prog,u32 off,const struct bpf_insn * patch,u32 len)449 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
450 const struct bpf_insn *patch, u32 len)
451 {
452 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
453 const u32 cnt_max = S16_MAX;
454 struct bpf_prog *prog_adj;
455 int err;
456
457 /* Since our patchlet doesn't expand the image, we're done. */
458 if (insn_delta == 0) {
459 memcpy(prog->insnsi + off, patch, sizeof(*patch));
460 return prog;
461 }
462
463 insn_adj_cnt = prog->len + insn_delta;
464
465 /* Reject anything that would potentially let the insn->off
466 * target overflow when we have excessive program expansions.
467 * We need to probe here before we do any reallocation where
468 * we afterwards may not fail anymore.
469 */
470 if (insn_adj_cnt > cnt_max &&
471 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
472 return ERR_PTR(err);
473
474 /* Several new instructions need to be inserted. Make room
475 * for them. Likely, there's no need for a new allocation as
476 * last page could have large enough tailroom.
477 */
478 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
479 GFP_USER);
480 if (!prog_adj)
481 return ERR_PTR(-ENOMEM);
482
483 prog_adj->len = insn_adj_cnt;
484
485 /* Patching happens in 3 steps:
486 *
487 * 1) Move over tail of insnsi from next instruction onwards,
488 * so we can patch the single target insn with one or more
489 * new ones (patching is always from 1 to n insns, n > 0).
490 * 2) Inject new instructions at the target location.
491 * 3) Adjust branch offsets if necessary.
492 */
493 insn_rest = insn_adj_cnt - off - len;
494
495 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
496 sizeof(*patch) * insn_rest);
497 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
498
499 /* We are guaranteed to not fail at this point, otherwise
500 * the ship has sailed to reverse to the original state. An
501 * overflow cannot happen at this point.
502 */
503 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
504
505 bpf_adj_linfo(prog_adj, off, insn_delta);
506
507 return prog_adj;
508 }
509
bpf_remove_insns(struct bpf_prog * prog,u32 off,u32 cnt)510 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
511 {
512 /* Branch offsets can't overflow when program is shrinking, no need
513 * to call bpf_adj_branches(..., true) here
514 */
515 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
516 sizeof(struct bpf_insn) * (prog->len - off - cnt));
517 prog->len -= cnt;
518
519 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
520 }
521
bpf_prog_kallsyms_del_subprogs(struct bpf_prog * fp)522 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
523 {
524 int i;
525
526 for (i = 0; i < fp->aux->func_cnt; i++)
527 bpf_prog_kallsyms_del(fp->aux->func[i]);
528 }
529
bpf_prog_kallsyms_del_all(struct bpf_prog * fp)530 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
531 {
532 bpf_prog_kallsyms_del_subprogs(fp);
533 bpf_prog_kallsyms_del(fp);
534 }
535
536 #ifdef CONFIG_BPF_JIT
537 /* All BPF JIT sysctl knobs here. */
538 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
539 int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
540 int bpf_jit_harden __read_mostly;
541 long bpf_jit_limit __read_mostly;
542 long bpf_jit_limit_max __read_mostly;
543
544 static void
bpf_prog_ksym_set_addr(struct bpf_prog * prog)545 bpf_prog_ksym_set_addr(struct bpf_prog *prog)
546 {
547 WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog));
548
549 prog->aux->ksym.start = (unsigned long) prog->bpf_func;
550 prog->aux->ksym.end = prog->aux->ksym.start + prog->jited_len;
551 }
552
553 static void
bpf_prog_ksym_set_name(struct bpf_prog * prog)554 bpf_prog_ksym_set_name(struct bpf_prog *prog)
555 {
556 char *sym = prog->aux->ksym.name;
557 const char *end = sym + KSYM_NAME_LEN;
558 const struct btf_type *type;
559 const char *func_name;
560
561 BUILD_BUG_ON(sizeof("bpf_prog_") +
562 sizeof(prog->tag) * 2 +
563 /* name has been null terminated.
564 * We should need +1 for the '_' preceding
565 * the name. However, the null character
566 * is double counted between the name and the
567 * sizeof("bpf_prog_") above, so we omit
568 * the +1 here.
569 */
570 sizeof(prog->aux->name) > KSYM_NAME_LEN);
571
572 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
573 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
574
575 /* prog->aux->name will be ignored if full btf name is available */
576 if (prog->aux->func_info_cnt) {
577 type = btf_type_by_id(prog->aux->btf,
578 prog->aux->func_info[prog->aux->func_idx].type_id);
579 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
580 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
581 return;
582 }
583
584 if (prog->aux->name[0])
585 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
586 else
587 *sym = 0;
588 }
589
bpf_get_ksym_start(struct latch_tree_node * n)590 static unsigned long bpf_get_ksym_start(struct latch_tree_node *n)
591 {
592 return container_of(n, struct bpf_ksym, tnode)->start;
593 }
594
bpf_tree_less(struct latch_tree_node * a,struct latch_tree_node * b)595 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
596 struct latch_tree_node *b)
597 {
598 return bpf_get_ksym_start(a) < bpf_get_ksym_start(b);
599 }
600
bpf_tree_comp(void * key,struct latch_tree_node * n)601 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
602 {
603 unsigned long val = (unsigned long)key;
604 const struct bpf_ksym *ksym;
605
606 ksym = container_of(n, struct bpf_ksym, tnode);
607
608 if (val < ksym->start)
609 return -1;
610 if (val >= ksym->end)
611 return 1;
612
613 return 0;
614 }
615
616 static const struct latch_tree_ops bpf_tree_ops = {
617 .less = bpf_tree_less,
618 .comp = bpf_tree_comp,
619 };
620
621 static DEFINE_SPINLOCK(bpf_lock);
622 static LIST_HEAD(bpf_kallsyms);
623 static struct latch_tree_root bpf_tree __cacheline_aligned;
624
bpf_ksym_add(struct bpf_ksym * ksym)625 void bpf_ksym_add(struct bpf_ksym *ksym)
626 {
627 spin_lock_bh(&bpf_lock);
628 WARN_ON_ONCE(!list_empty(&ksym->lnode));
629 list_add_tail_rcu(&ksym->lnode, &bpf_kallsyms);
630 latch_tree_insert(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
631 spin_unlock_bh(&bpf_lock);
632 }
633
__bpf_ksym_del(struct bpf_ksym * ksym)634 static void __bpf_ksym_del(struct bpf_ksym *ksym)
635 {
636 if (list_empty(&ksym->lnode))
637 return;
638
639 latch_tree_erase(&ksym->tnode, &bpf_tree, &bpf_tree_ops);
640 list_del_rcu(&ksym->lnode);
641 }
642
bpf_ksym_del(struct bpf_ksym * ksym)643 void bpf_ksym_del(struct bpf_ksym *ksym)
644 {
645 spin_lock_bh(&bpf_lock);
646 __bpf_ksym_del(ksym);
647 spin_unlock_bh(&bpf_lock);
648 }
649
bpf_prog_kallsyms_candidate(const struct bpf_prog * fp)650 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
651 {
652 return fp->jited && !bpf_prog_was_classic(fp);
653 }
654
bpf_prog_kallsyms_add(struct bpf_prog * fp)655 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
656 {
657 if (!bpf_prog_kallsyms_candidate(fp) ||
658 !bpf_capable())
659 return;
660
661 bpf_prog_ksym_set_addr(fp);
662 bpf_prog_ksym_set_name(fp);
663 fp->aux->ksym.prog = true;
664
665 bpf_ksym_add(&fp->aux->ksym);
666 }
667
bpf_prog_kallsyms_del(struct bpf_prog * fp)668 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
669 {
670 if (!bpf_prog_kallsyms_candidate(fp))
671 return;
672
673 bpf_ksym_del(&fp->aux->ksym);
674 }
675
bpf_ksym_find(unsigned long addr)676 static struct bpf_ksym *bpf_ksym_find(unsigned long addr)
677 {
678 struct latch_tree_node *n;
679
680 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
681 return n ? container_of(n, struct bpf_ksym, tnode) : NULL;
682 }
683
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)684 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
685 unsigned long *off, char *sym)
686 {
687 struct bpf_ksym *ksym;
688 char *ret = NULL;
689
690 rcu_read_lock();
691 ksym = bpf_ksym_find(addr);
692 if (ksym) {
693 unsigned long symbol_start = ksym->start;
694 unsigned long symbol_end = ksym->end;
695
696 strncpy(sym, ksym->name, KSYM_NAME_LEN);
697
698 ret = sym;
699 if (size)
700 *size = symbol_end - symbol_start;
701 if (off)
702 *off = addr - symbol_start;
703 }
704 rcu_read_unlock();
705
706 return ret;
707 }
708
is_bpf_text_address(unsigned long addr)709 bool is_bpf_text_address(unsigned long addr)
710 {
711 bool ret;
712
713 rcu_read_lock();
714 ret = bpf_ksym_find(addr) != NULL;
715 rcu_read_unlock();
716
717 return ret;
718 }
719
bpf_prog_ksym_find(unsigned long addr)720 static struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
721 {
722 struct bpf_ksym *ksym = bpf_ksym_find(addr);
723
724 return ksym && ksym->prog ?
725 container_of(ksym, struct bpf_prog_aux, ksym)->prog :
726 NULL;
727 }
728
search_bpf_extables(unsigned long addr)729 const struct exception_table_entry *search_bpf_extables(unsigned long addr)
730 {
731 const struct exception_table_entry *e = NULL;
732 struct bpf_prog *prog;
733
734 rcu_read_lock();
735 prog = bpf_prog_ksym_find(addr);
736 if (!prog)
737 goto out;
738 if (!prog->aux->num_exentries)
739 goto out;
740
741 e = search_extable(prog->aux->extable, prog->aux->num_exentries, addr);
742 out:
743 rcu_read_unlock();
744 return e;
745 }
746
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)747 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
748 char *sym)
749 {
750 struct bpf_ksym *ksym;
751 unsigned int it = 0;
752 int ret = -ERANGE;
753
754 if (!bpf_jit_kallsyms_enabled())
755 return ret;
756
757 rcu_read_lock();
758 list_for_each_entry_rcu(ksym, &bpf_kallsyms, lnode) {
759 if (it++ != symnum)
760 continue;
761
762 strncpy(sym, ksym->name, KSYM_NAME_LEN);
763
764 *value = ksym->start;
765 *type = BPF_SYM_ELF_TYPE;
766
767 ret = 0;
768 break;
769 }
770 rcu_read_unlock();
771
772 return ret;
773 }
774
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)775 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
776 struct bpf_jit_poke_descriptor *poke)
777 {
778 struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab;
779 static const u32 poke_tab_max = 1024;
780 u32 slot = prog->aux->size_poke_tab;
781 u32 size = slot + 1;
782
783 if (size > poke_tab_max)
784 return -ENOSPC;
785 if (poke->tailcall_target || poke->tailcall_target_stable ||
786 poke->tailcall_bypass || poke->adj_off || poke->bypass_addr)
787 return -EINVAL;
788
789 switch (poke->reason) {
790 case BPF_POKE_REASON_TAIL_CALL:
791 if (!poke->tail_call.map)
792 return -EINVAL;
793 break;
794 default:
795 return -EINVAL;
796 }
797
798 tab = krealloc(tab, size * sizeof(*poke), GFP_KERNEL);
799 if (!tab)
800 return -ENOMEM;
801
802 memcpy(&tab[slot], poke, sizeof(*poke));
803 prog->aux->size_poke_tab = size;
804 prog->aux->poke_tab = tab;
805
806 return slot;
807 }
808
809 /*
810 * BPF program pack allocator.
811 *
812 * Most BPF programs are pretty small. Allocating a hole page for each
813 * program is sometime a waste. Many small bpf program also adds pressure
814 * to instruction TLB. To solve this issue, we introduce a BPF program pack
815 * allocator. The prog_pack allocator uses HPAGE_PMD_SIZE page (2MB on x86)
816 * to host BPF programs.
817 */
818 #define BPF_PROG_CHUNK_SHIFT 6
819 #define BPF_PROG_CHUNK_SIZE (1 << BPF_PROG_CHUNK_SHIFT)
820 #define BPF_PROG_CHUNK_MASK (~(BPF_PROG_CHUNK_SIZE - 1))
821
822 struct bpf_prog_pack {
823 struct list_head list;
824 void *ptr;
825 unsigned long bitmap[];
826 };
827
bpf_jit_fill_hole_with_zero(void * area,unsigned int size)828 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size)
829 {
830 memset(area, 0, size);
831 }
832
833 #define BPF_PROG_SIZE_TO_NBITS(size) (round_up(size, BPF_PROG_CHUNK_SIZE) / BPF_PROG_CHUNK_SIZE)
834
835 static DEFINE_MUTEX(pack_mutex);
836 static LIST_HEAD(pack_list);
837
838 /* PMD_SIZE is not available in some special config, e.g. ARCH=arm with
839 * CONFIG_MMU=n. Use PAGE_SIZE in these cases.
840 */
841 #ifdef PMD_SIZE
842 #define BPF_PROG_PACK_SIZE (PMD_SIZE * num_possible_nodes())
843 #else
844 #define BPF_PROG_PACK_SIZE PAGE_SIZE
845 #endif
846
847 #define BPF_PROG_CHUNK_COUNT (BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE)
848
alloc_new_pack(bpf_jit_fill_hole_t bpf_fill_ill_insns)849 static struct bpf_prog_pack *alloc_new_pack(bpf_jit_fill_hole_t bpf_fill_ill_insns)
850 {
851 struct bpf_prog_pack *pack;
852
853 pack = kzalloc(struct_size(pack, bitmap, BITS_TO_LONGS(BPF_PROG_CHUNK_COUNT)),
854 GFP_KERNEL);
855 if (!pack)
856 return NULL;
857 pack->ptr = module_alloc(BPF_PROG_PACK_SIZE);
858 if (!pack->ptr) {
859 kfree(pack);
860 return NULL;
861 }
862 bpf_fill_ill_insns(pack->ptr, BPF_PROG_PACK_SIZE);
863 bitmap_zero(pack->bitmap, BPF_PROG_PACK_SIZE / BPF_PROG_CHUNK_SIZE);
864 list_add_tail(&pack->list, &pack_list);
865
866 set_vm_flush_reset_perms(pack->ptr);
867 set_memory_ro((unsigned long)pack->ptr, BPF_PROG_PACK_SIZE / PAGE_SIZE);
868 set_memory_x((unsigned long)pack->ptr, BPF_PROG_PACK_SIZE / PAGE_SIZE);
869 return pack;
870 }
871
bpf_prog_pack_alloc(u32 size,bpf_jit_fill_hole_t bpf_fill_ill_insns)872 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns)
873 {
874 unsigned int nbits = BPF_PROG_SIZE_TO_NBITS(size);
875 struct bpf_prog_pack *pack;
876 unsigned long pos;
877 void *ptr = NULL;
878
879 mutex_lock(&pack_mutex);
880 if (size > BPF_PROG_PACK_SIZE) {
881 size = round_up(size, PAGE_SIZE);
882 ptr = module_alloc(size);
883 if (ptr) {
884 bpf_fill_ill_insns(ptr, size);
885 set_vm_flush_reset_perms(ptr);
886 set_memory_ro((unsigned long)ptr, size / PAGE_SIZE);
887 set_memory_x((unsigned long)ptr, size / PAGE_SIZE);
888 }
889 goto out;
890 }
891 list_for_each_entry(pack, &pack_list, list) {
892 pos = bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
893 nbits, 0);
894 if (pos < BPF_PROG_CHUNK_COUNT)
895 goto found_free_area;
896 }
897
898 pack = alloc_new_pack(bpf_fill_ill_insns);
899 if (!pack)
900 goto out;
901
902 pos = 0;
903
904 found_free_area:
905 bitmap_set(pack->bitmap, pos, nbits);
906 ptr = (void *)(pack->ptr) + (pos << BPF_PROG_CHUNK_SHIFT);
907
908 out:
909 mutex_unlock(&pack_mutex);
910 return ptr;
911 }
912
bpf_prog_pack_free(struct bpf_binary_header * hdr)913 void bpf_prog_pack_free(struct bpf_binary_header *hdr)
914 {
915 struct bpf_prog_pack *pack = NULL, *tmp;
916 unsigned int nbits;
917 unsigned long pos;
918
919 mutex_lock(&pack_mutex);
920 if (hdr->size > BPF_PROG_PACK_SIZE) {
921 module_memfree(hdr);
922 goto out;
923 }
924
925 list_for_each_entry(tmp, &pack_list, list) {
926 if ((void *)hdr >= tmp->ptr && (tmp->ptr + BPF_PROG_PACK_SIZE) > (void *)hdr) {
927 pack = tmp;
928 break;
929 }
930 }
931
932 if (WARN_ONCE(!pack, "bpf_prog_pack bug\n"))
933 goto out;
934
935 nbits = BPF_PROG_SIZE_TO_NBITS(hdr->size);
936 pos = ((unsigned long)hdr - (unsigned long)pack->ptr) >> BPF_PROG_CHUNK_SHIFT;
937
938 WARN_ONCE(bpf_arch_text_invalidate(hdr, hdr->size),
939 "bpf_prog_pack bug: missing bpf_arch_text_invalidate?\n");
940
941 bitmap_clear(pack->bitmap, pos, nbits);
942 if (bitmap_find_next_zero_area(pack->bitmap, BPF_PROG_CHUNK_COUNT, 0,
943 BPF_PROG_CHUNK_COUNT, 0) == 0) {
944 list_del(&pack->list);
945 module_memfree(pack->ptr);
946 kfree(pack);
947 }
948 out:
949 mutex_unlock(&pack_mutex);
950 }
951
952 static atomic_long_t bpf_jit_current;
953
954 /* Can be overridden by an arch's JIT compiler if it has a custom,
955 * dedicated BPF backend memory area, or if neither of the two
956 * below apply.
957 */
bpf_jit_alloc_exec_limit(void)958 u64 __weak bpf_jit_alloc_exec_limit(void)
959 {
960 #if defined(MODULES_VADDR)
961 return MODULES_END - MODULES_VADDR;
962 #else
963 return VMALLOC_END - VMALLOC_START;
964 #endif
965 }
966
bpf_jit_charge_init(void)967 static int __init bpf_jit_charge_init(void)
968 {
969 /* Only used as heuristic here to derive limit. */
970 bpf_jit_limit_max = bpf_jit_alloc_exec_limit();
971 bpf_jit_limit = min_t(u64, round_up(bpf_jit_limit_max >> 2,
972 PAGE_SIZE), LONG_MAX);
973 return 0;
974 }
975 pure_initcall(bpf_jit_charge_init);
976
bpf_jit_charge_modmem(u32 size)977 int bpf_jit_charge_modmem(u32 size)
978 {
979 if (atomic_long_add_return(size, &bpf_jit_current) > READ_ONCE(bpf_jit_limit)) {
980 if (!bpf_capable()) {
981 atomic_long_sub(size, &bpf_jit_current);
982 return -EPERM;
983 }
984 }
985
986 return 0;
987 }
988
bpf_jit_uncharge_modmem(u32 size)989 void bpf_jit_uncharge_modmem(u32 size)
990 {
991 atomic_long_sub(size, &bpf_jit_current);
992 }
993
bpf_jit_alloc_exec(unsigned long size)994 void *__weak bpf_jit_alloc_exec(unsigned long size)
995 {
996 return module_alloc(size);
997 }
998
bpf_jit_free_exec(void * addr)999 void __weak bpf_jit_free_exec(void *addr)
1000 {
1001 module_memfree(addr);
1002 }
1003
1004 struct bpf_binary_header *
bpf_jit_binary_alloc(unsigned int proglen,u8 ** image_ptr,unsigned int alignment,bpf_jit_fill_hole_t bpf_fill_ill_insns)1005 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1006 unsigned int alignment,
1007 bpf_jit_fill_hole_t bpf_fill_ill_insns)
1008 {
1009 struct bpf_binary_header *hdr;
1010 u32 size, hole, start;
1011
1012 WARN_ON_ONCE(!is_power_of_2(alignment) ||
1013 alignment > BPF_IMAGE_ALIGNMENT);
1014
1015 /* Most of BPF filters are really small, but if some of them
1016 * fill a page, allow at least 128 extra bytes to insert a
1017 * random section of illegal instructions.
1018 */
1019 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
1020
1021 if (bpf_jit_charge_modmem(size))
1022 return NULL;
1023 hdr = bpf_jit_alloc_exec(size);
1024 if (!hdr) {
1025 bpf_jit_uncharge_modmem(size);
1026 return NULL;
1027 }
1028
1029 /* Fill space with illegal/arch-dep instructions. */
1030 bpf_fill_ill_insns(hdr, size);
1031
1032 hdr->size = size;
1033 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
1034 PAGE_SIZE - sizeof(*hdr));
1035 start = prandom_u32_max(hole) & ~(alignment - 1);
1036
1037 /* Leave a random number of instructions before BPF code. */
1038 *image_ptr = &hdr->image[start];
1039
1040 return hdr;
1041 }
1042
bpf_jit_binary_free(struct bpf_binary_header * hdr)1043 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
1044 {
1045 u32 size = hdr->size;
1046
1047 bpf_jit_free_exec(hdr);
1048 bpf_jit_uncharge_modmem(size);
1049 }
1050
1051 /* Allocate jit binary from bpf_prog_pack allocator.
1052 * Since the allocated memory is RO+X, the JIT engine cannot write directly
1053 * to the memory. To solve this problem, a RW buffer is also allocated at
1054 * as the same time. The JIT engine should calculate offsets based on the
1055 * RO memory address, but write JITed program to the RW buffer. Once the
1056 * JIT engine finishes, it calls bpf_jit_binary_pack_finalize, which copies
1057 * the JITed program to the RO memory.
1058 */
1059 struct bpf_binary_header *
bpf_jit_binary_pack_alloc(unsigned int proglen,u8 ** image_ptr,unsigned int alignment,struct bpf_binary_header ** rw_header,u8 ** rw_image,bpf_jit_fill_hole_t bpf_fill_ill_insns)1060 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **image_ptr,
1061 unsigned int alignment,
1062 struct bpf_binary_header **rw_header,
1063 u8 **rw_image,
1064 bpf_jit_fill_hole_t bpf_fill_ill_insns)
1065 {
1066 struct bpf_binary_header *ro_header;
1067 u32 size, hole, start;
1068
1069 WARN_ON_ONCE(!is_power_of_2(alignment) ||
1070 alignment > BPF_IMAGE_ALIGNMENT);
1071
1072 /* add 16 bytes for a random section of illegal instructions */
1073 size = round_up(proglen + sizeof(*ro_header) + 16, BPF_PROG_CHUNK_SIZE);
1074
1075 if (bpf_jit_charge_modmem(size))
1076 return NULL;
1077 ro_header = bpf_prog_pack_alloc(size, bpf_fill_ill_insns);
1078 if (!ro_header) {
1079 bpf_jit_uncharge_modmem(size);
1080 return NULL;
1081 }
1082
1083 *rw_header = kvmalloc(size, GFP_KERNEL);
1084 if (!*rw_header) {
1085 bpf_arch_text_copy(&ro_header->size, &size, sizeof(size));
1086 bpf_prog_pack_free(ro_header);
1087 bpf_jit_uncharge_modmem(size);
1088 return NULL;
1089 }
1090
1091 /* Fill space with illegal/arch-dep instructions. */
1092 bpf_fill_ill_insns(*rw_header, size);
1093 (*rw_header)->size = size;
1094
1095 hole = min_t(unsigned int, size - (proglen + sizeof(*ro_header)),
1096 BPF_PROG_CHUNK_SIZE - sizeof(*ro_header));
1097 start = prandom_u32_max(hole) & ~(alignment - 1);
1098
1099 *image_ptr = &ro_header->image[start];
1100 *rw_image = &(*rw_header)->image[start];
1101
1102 return ro_header;
1103 }
1104
1105 /* Copy JITed text from rw_header to its final location, the ro_header. */
bpf_jit_binary_pack_finalize(struct bpf_prog * prog,struct bpf_binary_header * ro_header,struct bpf_binary_header * rw_header)1106 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1107 struct bpf_binary_header *ro_header,
1108 struct bpf_binary_header *rw_header)
1109 {
1110 void *ptr;
1111
1112 ptr = bpf_arch_text_copy(ro_header, rw_header, rw_header->size);
1113
1114 kvfree(rw_header);
1115
1116 if (IS_ERR(ptr)) {
1117 bpf_prog_pack_free(ro_header);
1118 return PTR_ERR(ptr);
1119 }
1120 return 0;
1121 }
1122
1123 /* bpf_jit_binary_pack_free is called in two different scenarios:
1124 * 1) when the program is freed after;
1125 * 2) when the JIT engine fails (before bpf_jit_binary_pack_finalize).
1126 * For case 2), we need to free both the RO memory and the RW buffer.
1127 *
1128 * bpf_jit_binary_pack_free requires proper ro_header->size. However,
1129 * bpf_jit_binary_pack_alloc does not set it. Therefore, ro_header->size
1130 * must be set with either bpf_jit_binary_pack_finalize (normal path) or
1131 * bpf_arch_text_copy (when jit fails).
1132 */
bpf_jit_binary_pack_free(struct bpf_binary_header * ro_header,struct bpf_binary_header * rw_header)1133 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1134 struct bpf_binary_header *rw_header)
1135 {
1136 u32 size = ro_header->size;
1137
1138 bpf_prog_pack_free(ro_header);
1139 kvfree(rw_header);
1140 bpf_jit_uncharge_modmem(size);
1141 }
1142
1143 struct bpf_binary_header *
bpf_jit_binary_pack_hdr(const struct bpf_prog * fp)1144 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp)
1145 {
1146 unsigned long real_start = (unsigned long)fp->bpf_func;
1147 unsigned long addr;
1148
1149 addr = real_start & BPF_PROG_CHUNK_MASK;
1150 return (void *)addr;
1151 }
1152
1153 static inline struct bpf_binary_header *
bpf_jit_binary_hdr(const struct bpf_prog * fp)1154 bpf_jit_binary_hdr(const struct bpf_prog *fp)
1155 {
1156 unsigned long real_start = (unsigned long)fp->bpf_func;
1157 unsigned long addr;
1158
1159 addr = real_start & PAGE_MASK;
1160 return (void *)addr;
1161 }
1162
1163 /* This symbol is only overridden by archs that have different
1164 * requirements than the usual eBPF JITs, f.e. when they only
1165 * implement cBPF JIT, do not set images read-only, etc.
1166 */
bpf_jit_free(struct bpf_prog * fp)1167 void __weak bpf_jit_free(struct bpf_prog *fp)
1168 {
1169 if (fp->jited) {
1170 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
1171
1172 bpf_jit_binary_free(hdr);
1173 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
1174 }
1175
1176 bpf_prog_unlock_free(fp);
1177 }
1178
bpf_jit_get_func_addr(const struct bpf_prog * prog,const struct bpf_insn * insn,bool extra_pass,u64 * func_addr,bool * func_addr_fixed)1179 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1180 const struct bpf_insn *insn, bool extra_pass,
1181 u64 *func_addr, bool *func_addr_fixed)
1182 {
1183 s16 off = insn->off;
1184 s32 imm = insn->imm;
1185 u8 *addr;
1186
1187 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
1188 if (!*func_addr_fixed) {
1189 /* Place-holder address till the last pass has collected
1190 * all addresses for JITed subprograms in which case we
1191 * can pick them up from prog->aux.
1192 */
1193 if (!extra_pass)
1194 addr = NULL;
1195 else if (prog->aux->func &&
1196 off >= 0 && off < prog->aux->func_cnt)
1197 addr = (u8 *)prog->aux->func[off]->bpf_func;
1198 else
1199 return -EINVAL;
1200 } else {
1201 /* Address of a BPF helper call. Since part of the core
1202 * kernel, it's always at a fixed location. __bpf_call_base
1203 * and the helper with imm relative to it are both in core
1204 * kernel.
1205 */
1206 addr = (u8 *)__bpf_call_base + imm;
1207 }
1208
1209 *func_addr = (unsigned long)addr;
1210 return 0;
1211 }
1212
bpf_jit_blind_insn(const struct bpf_insn * from,const struct bpf_insn * aux,struct bpf_insn * to_buff,bool emit_zext)1213 static int bpf_jit_blind_insn(const struct bpf_insn *from,
1214 const struct bpf_insn *aux,
1215 struct bpf_insn *to_buff,
1216 bool emit_zext)
1217 {
1218 struct bpf_insn *to = to_buff;
1219 u32 imm_rnd = get_random_u32();
1220 s16 off;
1221
1222 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
1223 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
1224
1225 /* Constraints on AX register:
1226 *
1227 * AX register is inaccessible from user space. It is mapped in
1228 * all JITs, and used here for constant blinding rewrites. It is
1229 * typically "stateless" meaning its contents are only valid within
1230 * the executed instruction, but not across several instructions.
1231 * There are a few exceptions however which are further detailed
1232 * below.
1233 *
1234 * Constant blinding is only used by JITs, not in the interpreter.
1235 * The interpreter uses AX in some occasions as a local temporary
1236 * register e.g. in DIV or MOD instructions.
1237 *
1238 * In restricted circumstances, the verifier can also use the AX
1239 * register for rewrites as long as they do not interfere with
1240 * the above cases!
1241 */
1242 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
1243 goto out;
1244
1245 if (from->imm == 0 &&
1246 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
1247 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
1248 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
1249 goto out;
1250 }
1251
1252 switch (from->code) {
1253 case BPF_ALU | BPF_ADD | BPF_K:
1254 case BPF_ALU | BPF_SUB | BPF_K:
1255 case BPF_ALU | BPF_AND | BPF_K:
1256 case BPF_ALU | BPF_OR | BPF_K:
1257 case BPF_ALU | BPF_XOR | BPF_K:
1258 case BPF_ALU | BPF_MUL | BPF_K:
1259 case BPF_ALU | BPF_MOV | BPF_K:
1260 case BPF_ALU | BPF_DIV | BPF_K:
1261 case BPF_ALU | BPF_MOD | BPF_K:
1262 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1263 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1264 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
1265 break;
1266
1267 case BPF_ALU64 | BPF_ADD | BPF_K:
1268 case BPF_ALU64 | BPF_SUB | BPF_K:
1269 case BPF_ALU64 | BPF_AND | BPF_K:
1270 case BPF_ALU64 | BPF_OR | BPF_K:
1271 case BPF_ALU64 | BPF_XOR | BPF_K:
1272 case BPF_ALU64 | BPF_MUL | BPF_K:
1273 case BPF_ALU64 | BPF_MOV | BPF_K:
1274 case BPF_ALU64 | BPF_DIV | BPF_K:
1275 case BPF_ALU64 | BPF_MOD | BPF_K:
1276 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1277 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1278 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
1279 break;
1280
1281 case BPF_JMP | BPF_JEQ | BPF_K:
1282 case BPF_JMP | BPF_JNE | BPF_K:
1283 case BPF_JMP | BPF_JGT | BPF_K:
1284 case BPF_JMP | BPF_JLT | BPF_K:
1285 case BPF_JMP | BPF_JGE | BPF_K:
1286 case BPF_JMP | BPF_JLE | BPF_K:
1287 case BPF_JMP | BPF_JSGT | BPF_K:
1288 case BPF_JMP | BPF_JSLT | BPF_K:
1289 case BPF_JMP | BPF_JSGE | BPF_K:
1290 case BPF_JMP | BPF_JSLE | BPF_K:
1291 case BPF_JMP | BPF_JSET | BPF_K:
1292 /* Accommodate for extra offset in case of a backjump. */
1293 off = from->off;
1294 if (off < 0)
1295 off -= 2;
1296 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1297 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1298 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
1299 break;
1300
1301 case BPF_JMP32 | BPF_JEQ | BPF_K:
1302 case BPF_JMP32 | BPF_JNE | BPF_K:
1303 case BPF_JMP32 | BPF_JGT | BPF_K:
1304 case BPF_JMP32 | BPF_JLT | BPF_K:
1305 case BPF_JMP32 | BPF_JGE | BPF_K:
1306 case BPF_JMP32 | BPF_JLE | BPF_K:
1307 case BPF_JMP32 | BPF_JSGT | BPF_K:
1308 case BPF_JMP32 | BPF_JSLT | BPF_K:
1309 case BPF_JMP32 | BPF_JSGE | BPF_K:
1310 case BPF_JMP32 | BPF_JSLE | BPF_K:
1311 case BPF_JMP32 | BPF_JSET | BPF_K:
1312 /* Accommodate for extra offset in case of a backjump. */
1313 off = from->off;
1314 if (off < 0)
1315 off -= 2;
1316 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1317 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1318 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1319 off);
1320 break;
1321
1322 case BPF_LD | BPF_IMM | BPF_DW:
1323 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1324 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1325 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1326 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1327 break;
1328 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1329 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1330 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1331 if (emit_zext)
1332 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
1333 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1334 break;
1335
1336 case BPF_ST | BPF_MEM | BPF_DW:
1337 case BPF_ST | BPF_MEM | BPF_W:
1338 case BPF_ST | BPF_MEM | BPF_H:
1339 case BPF_ST | BPF_MEM | BPF_B:
1340 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1341 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1342 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1343 break;
1344 }
1345 out:
1346 return to - to_buff;
1347 }
1348
bpf_prog_clone_create(struct bpf_prog * fp_other,gfp_t gfp_extra_flags)1349 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1350 gfp_t gfp_extra_flags)
1351 {
1352 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1353 struct bpf_prog *fp;
1354
1355 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags);
1356 if (fp != NULL) {
1357 /* aux->prog still points to the fp_other one, so
1358 * when promoting the clone to the real program,
1359 * this still needs to be adapted.
1360 */
1361 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1362 }
1363
1364 return fp;
1365 }
1366
bpf_prog_clone_free(struct bpf_prog * fp)1367 static void bpf_prog_clone_free(struct bpf_prog *fp)
1368 {
1369 /* aux was stolen by the other clone, so we cannot free
1370 * it from this path! It will be freed eventually by the
1371 * other program on release.
1372 *
1373 * At this point, we don't need a deferred release since
1374 * clone is guaranteed to not be locked.
1375 */
1376 fp->aux = NULL;
1377 fp->stats = NULL;
1378 fp->active = NULL;
1379 __bpf_prog_free(fp);
1380 }
1381
bpf_jit_prog_release_other(struct bpf_prog * fp,struct bpf_prog * fp_other)1382 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1383 {
1384 /* We have to repoint aux->prog to self, as we don't
1385 * know whether fp here is the clone or the original.
1386 */
1387 fp->aux->prog = fp;
1388 bpf_prog_clone_free(fp_other);
1389 }
1390
bpf_jit_blind_constants(struct bpf_prog * prog)1391 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1392 {
1393 struct bpf_insn insn_buff[16], aux[2];
1394 struct bpf_prog *clone, *tmp;
1395 int insn_delta, insn_cnt;
1396 struct bpf_insn *insn;
1397 int i, rewritten;
1398
1399 if (!prog->blinding_requested || prog->blinded)
1400 return prog;
1401
1402 clone = bpf_prog_clone_create(prog, GFP_USER);
1403 if (!clone)
1404 return ERR_PTR(-ENOMEM);
1405
1406 insn_cnt = clone->len;
1407 insn = clone->insnsi;
1408
1409 for (i = 0; i < insn_cnt; i++, insn++) {
1410 if (bpf_pseudo_func(insn)) {
1411 /* ld_imm64 with an address of bpf subprog is not
1412 * a user controlled constant. Don't randomize it,
1413 * since it will conflict with jit_subprogs() logic.
1414 */
1415 insn++;
1416 i++;
1417 continue;
1418 }
1419
1420 /* We temporarily need to hold the original ld64 insn
1421 * so that we can still access the first part in the
1422 * second blinding run.
1423 */
1424 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1425 insn[1].code == 0)
1426 memcpy(aux, insn, sizeof(aux));
1427
1428 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1429 clone->aux->verifier_zext);
1430 if (!rewritten)
1431 continue;
1432
1433 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1434 if (IS_ERR(tmp)) {
1435 /* Patching may have repointed aux->prog during
1436 * realloc from the original one, so we need to
1437 * fix it up here on error.
1438 */
1439 bpf_jit_prog_release_other(prog, clone);
1440 return tmp;
1441 }
1442
1443 clone = tmp;
1444 insn_delta = rewritten - 1;
1445
1446 /* Walk new program and skip insns we just inserted. */
1447 insn = clone->insnsi + i + insn_delta;
1448 insn_cnt += insn_delta;
1449 i += insn_delta;
1450 }
1451
1452 clone->blinded = 1;
1453 return clone;
1454 }
1455 #endif /* CONFIG_BPF_JIT */
1456
1457 /* Base function for offset calculation. Needs to go into .text section,
1458 * therefore keeping it non-static as well; will also be used by JITs
1459 * anyway later on, so do not let the compiler omit it. This also needs
1460 * to go into kallsyms for correlation from e.g. bpftool, so naming
1461 * must not change.
1462 */
__bpf_call_base(u64 r1,u64 r2,u64 r3,u64 r4,u64 r5)1463 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1464 {
1465 return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(__bpf_call_base);
1468
1469 /* All UAPI available opcodes. */
1470 #define BPF_INSN_MAP(INSN_2, INSN_3) \
1471 /* 32 bit ALU operations. */ \
1472 /* Register based. */ \
1473 INSN_3(ALU, ADD, X), \
1474 INSN_3(ALU, SUB, X), \
1475 INSN_3(ALU, AND, X), \
1476 INSN_3(ALU, OR, X), \
1477 INSN_3(ALU, LSH, X), \
1478 INSN_3(ALU, RSH, X), \
1479 INSN_3(ALU, XOR, X), \
1480 INSN_3(ALU, MUL, X), \
1481 INSN_3(ALU, MOV, X), \
1482 INSN_3(ALU, ARSH, X), \
1483 INSN_3(ALU, DIV, X), \
1484 INSN_3(ALU, MOD, X), \
1485 INSN_2(ALU, NEG), \
1486 INSN_3(ALU, END, TO_BE), \
1487 INSN_3(ALU, END, TO_LE), \
1488 /* Immediate based. */ \
1489 INSN_3(ALU, ADD, K), \
1490 INSN_3(ALU, SUB, K), \
1491 INSN_3(ALU, AND, K), \
1492 INSN_3(ALU, OR, K), \
1493 INSN_3(ALU, LSH, K), \
1494 INSN_3(ALU, RSH, K), \
1495 INSN_3(ALU, XOR, K), \
1496 INSN_3(ALU, MUL, K), \
1497 INSN_3(ALU, MOV, K), \
1498 INSN_3(ALU, ARSH, K), \
1499 INSN_3(ALU, DIV, K), \
1500 INSN_3(ALU, MOD, K), \
1501 /* 64 bit ALU operations. */ \
1502 /* Register based. */ \
1503 INSN_3(ALU64, ADD, X), \
1504 INSN_3(ALU64, SUB, X), \
1505 INSN_3(ALU64, AND, X), \
1506 INSN_3(ALU64, OR, X), \
1507 INSN_3(ALU64, LSH, X), \
1508 INSN_3(ALU64, RSH, X), \
1509 INSN_3(ALU64, XOR, X), \
1510 INSN_3(ALU64, MUL, X), \
1511 INSN_3(ALU64, MOV, X), \
1512 INSN_3(ALU64, ARSH, X), \
1513 INSN_3(ALU64, DIV, X), \
1514 INSN_3(ALU64, MOD, X), \
1515 INSN_2(ALU64, NEG), \
1516 /* Immediate based. */ \
1517 INSN_3(ALU64, ADD, K), \
1518 INSN_3(ALU64, SUB, K), \
1519 INSN_3(ALU64, AND, K), \
1520 INSN_3(ALU64, OR, K), \
1521 INSN_3(ALU64, LSH, K), \
1522 INSN_3(ALU64, RSH, K), \
1523 INSN_3(ALU64, XOR, K), \
1524 INSN_3(ALU64, MUL, K), \
1525 INSN_3(ALU64, MOV, K), \
1526 INSN_3(ALU64, ARSH, K), \
1527 INSN_3(ALU64, DIV, K), \
1528 INSN_3(ALU64, MOD, K), \
1529 /* Call instruction. */ \
1530 INSN_2(JMP, CALL), \
1531 /* Exit instruction. */ \
1532 INSN_2(JMP, EXIT), \
1533 /* 32-bit Jump instructions. */ \
1534 /* Register based. */ \
1535 INSN_3(JMP32, JEQ, X), \
1536 INSN_3(JMP32, JNE, X), \
1537 INSN_3(JMP32, JGT, X), \
1538 INSN_3(JMP32, JLT, X), \
1539 INSN_3(JMP32, JGE, X), \
1540 INSN_3(JMP32, JLE, X), \
1541 INSN_3(JMP32, JSGT, X), \
1542 INSN_3(JMP32, JSLT, X), \
1543 INSN_3(JMP32, JSGE, X), \
1544 INSN_3(JMP32, JSLE, X), \
1545 INSN_3(JMP32, JSET, X), \
1546 /* Immediate based. */ \
1547 INSN_3(JMP32, JEQ, K), \
1548 INSN_3(JMP32, JNE, K), \
1549 INSN_3(JMP32, JGT, K), \
1550 INSN_3(JMP32, JLT, K), \
1551 INSN_3(JMP32, JGE, K), \
1552 INSN_3(JMP32, JLE, K), \
1553 INSN_3(JMP32, JSGT, K), \
1554 INSN_3(JMP32, JSLT, K), \
1555 INSN_3(JMP32, JSGE, K), \
1556 INSN_3(JMP32, JSLE, K), \
1557 INSN_3(JMP32, JSET, K), \
1558 /* Jump instructions. */ \
1559 /* Register based. */ \
1560 INSN_3(JMP, JEQ, X), \
1561 INSN_3(JMP, JNE, X), \
1562 INSN_3(JMP, JGT, X), \
1563 INSN_3(JMP, JLT, X), \
1564 INSN_3(JMP, JGE, X), \
1565 INSN_3(JMP, JLE, X), \
1566 INSN_3(JMP, JSGT, X), \
1567 INSN_3(JMP, JSLT, X), \
1568 INSN_3(JMP, JSGE, X), \
1569 INSN_3(JMP, JSLE, X), \
1570 INSN_3(JMP, JSET, X), \
1571 /* Immediate based. */ \
1572 INSN_3(JMP, JEQ, K), \
1573 INSN_3(JMP, JNE, K), \
1574 INSN_3(JMP, JGT, K), \
1575 INSN_3(JMP, JLT, K), \
1576 INSN_3(JMP, JGE, K), \
1577 INSN_3(JMP, JLE, K), \
1578 INSN_3(JMP, JSGT, K), \
1579 INSN_3(JMP, JSLT, K), \
1580 INSN_3(JMP, JSGE, K), \
1581 INSN_3(JMP, JSLE, K), \
1582 INSN_3(JMP, JSET, K), \
1583 INSN_2(JMP, JA), \
1584 /* Store instructions. */ \
1585 /* Register based. */ \
1586 INSN_3(STX, MEM, B), \
1587 INSN_3(STX, MEM, H), \
1588 INSN_3(STX, MEM, W), \
1589 INSN_3(STX, MEM, DW), \
1590 INSN_3(STX, ATOMIC, W), \
1591 INSN_3(STX, ATOMIC, DW), \
1592 /* Immediate based. */ \
1593 INSN_3(ST, MEM, B), \
1594 INSN_3(ST, MEM, H), \
1595 INSN_3(ST, MEM, W), \
1596 INSN_3(ST, MEM, DW), \
1597 /* Load instructions. */ \
1598 /* Register based. */ \
1599 INSN_3(LDX, MEM, B), \
1600 INSN_3(LDX, MEM, H), \
1601 INSN_3(LDX, MEM, W), \
1602 INSN_3(LDX, MEM, DW), \
1603 /* Immediate based. */ \
1604 INSN_3(LD, IMM, DW)
1605
bpf_opcode_in_insntable(u8 code)1606 bool bpf_opcode_in_insntable(u8 code)
1607 {
1608 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1609 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1610 static const bool public_insntable[256] = {
1611 [0 ... 255] = false,
1612 /* Now overwrite non-defaults ... */
1613 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1614 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1615 [BPF_LD | BPF_ABS | BPF_B] = true,
1616 [BPF_LD | BPF_ABS | BPF_H] = true,
1617 [BPF_LD | BPF_ABS | BPF_W] = true,
1618 [BPF_LD | BPF_IND | BPF_B] = true,
1619 [BPF_LD | BPF_IND | BPF_H] = true,
1620 [BPF_LD | BPF_IND | BPF_W] = true,
1621 };
1622 #undef BPF_INSN_3_TBL
1623 #undef BPF_INSN_2_TBL
1624 return public_insntable[code];
1625 }
1626
1627 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
bpf_probe_read_kernel(void * dst,u32 size,const void * unsafe_ptr)1628 u64 __weak bpf_probe_read_kernel(void *dst, u32 size, const void *unsafe_ptr)
1629 {
1630 memset(dst, 0, size);
1631 return -EFAULT;
1632 }
1633
1634 /**
1635 * ___bpf_prog_run - run eBPF program on a given context
1636 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1637 * @insn: is the array of eBPF instructions
1638 *
1639 * Decode and execute eBPF instructions.
1640 *
1641 * Return: whatever value is in %BPF_R0 at program exit
1642 */
___bpf_prog_run(u64 * regs,const struct bpf_insn * insn)1643 static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn)
1644 {
1645 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1646 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1647 static const void * const jumptable[256] __annotate_jump_table = {
1648 [0 ... 255] = &&default_label,
1649 /* Now overwrite non-defaults ... */
1650 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1651 /* Non-UAPI available opcodes. */
1652 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1653 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1654 [BPF_ST | BPF_NOSPEC] = &&ST_NOSPEC,
1655 [BPF_LDX | BPF_PROBE_MEM | BPF_B] = &&LDX_PROBE_MEM_B,
1656 [BPF_LDX | BPF_PROBE_MEM | BPF_H] = &&LDX_PROBE_MEM_H,
1657 [BPF_LDX | BPF_PROBE_MEM | BPF_W] = &&LDX_PROBE_MEM_W,
1658 [BPF_LDX | BPF_PROBE_MEM | BPF_DW] = &&LDX_PROBE_MEM_DW,
1659 };
1660 #undef BPF_INSN_3_LBL
1661 #undef BPF_INSN_2_LBL
1662 u32 tail_call_cnt = 0;
1663
1664 #define CONT ({ insn++; goto select_insn; })
1665 #define CONT_JMP ({ insn++; goto select_insn; })
1666
1667 select_insn:
1668 goto *jumptable[insn->code];
1669
1670 /* Explicitly mask the register-based shift amounts with 63 or 31
1671 * to avoid undefined behavior. Normally this won't affect the
1672 * generated code, for example, in case of native 64 bit archs such
1673 * as x86-64 or arm64, the compiler is optimizing the AND away for
1674 * the interpreter. In case of JITs, each of the JIT backends compiles
1675 * the BPF shift operations to machine instructions which produce
1676 * implementation-defined results in such a case; the resulting
1677 * contents of the register may be arbitrary, but program behaviour
1678 * as a whole remains defined. In other words, in case of JIT backends,
1679 * the AND must /not/ be added to the emitted LSH/RSH/ARSH translation.
1680 */
1681 /* ALU (shifts) */
1682 #define SHT(OPCODE, OP) \
1683 ALU64_##OPCODE##_X: \
1684 DST = DST OP (SRC & 63); \
1685 CONT; \
1686 ALU_##OPCODE##_X: \
1687 DST = (u32) DST OP ((u32) SRC & 31); \
1688 CONT; \
1689 ALU64_##OPCODE##_K: \
1690 DST = DST OP IMM; \
1691 CONT; \
1692 ALU_##OPCODE##_K: \
1693 DST = (u32) DST OP (u32) IMM; \
1694 CONT;
1695 /* ALU (rest) */
1696 #define ALU(OPCODE, OP) \
1697 ALU64_##OPCODE##_X: \
1698 DST = DST OP SRC; \
1699 CONT; \
1700 ALU_##OPCODE##_X: \
1701 DST = (u32) DST OP (u32) SRC; \
1702 CONT; \
1703 ALU64_##OPCODE##_K: \
1704 DST = DST OP IMM; \
1705 CONT; \
1706 ALU_##OPCODE##_K: \
1707 DST = (u32) DST OP (u32) IMM; \
1708 CONT;
1709 ALU(ADD, +)
1710 ALU(SUB, -)
1711 ALU(AND, &)
1712 ALU(OR, |)
1713 ALU(XOR, ^)
1714 ALU(MUL, *)
1715 SHT(LSH, <<)
1716 SHT(RSH, >>)
1717 #undef SHT
1718 #undef ALU
1719 ALU_NEG:
1720 DST = (u32) -DST;
1721 CONT;
1722 ALU64_NEG:
1723 DST = -DST;
1724 CONT;
1725 ALU_MOV_X:
1726 DST = (u32) SRC;
1727 CONT;
1728 ALU_MOV_K:
1729 DST = (u32) IMM;
1730 CONT;
1731 ALU64_MOV_X:
1732 DST = SRC;
1733 CONT;
1734 ALU64_MOV_K:
1735 DST = IMM;
1736 CONT;
1737 LD_IMM_DW:
1738 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1739 insn++;
1740 CONT;
1741 ALU_ARSH_X:
1742 DST = (u64) (u32) (((s32) DST) >> (SRC & 31));
1743 CONT;
1744 ALU_ARSH_K:
1745 DST = (u64) (u32) (((s32) DST) >> IMM);
1746 CONT;
1747 ALU64_ARSH_X:
1748 (*(s64 *) &DST) >>= (SRC & 63);
1749 CONT;
1750 ALU64_ARSH_K:
1751 (*(s64 *) &DST) >>= IMM;
1752 CONT;
1753 ALU64_MOD_X:
1754 div64_u64_rem(DST, SRC, &AX);
1755 DST = AX;
1756 CONT;
1757 ALU_MOD_X:
1758 AX = (u32) DST;
1759 DST = do_div(AX, (u32) SRC);
1760 CONT;
1761 ALU64_MOD_K:
1762 div64_u64_rem(DST, IMM, &AX);
1763 DST = AX;
1764 CONT;
1765 ALU_MOD_K:
1766 AX = (u32) DST;
1767 DST = do_div(AX, (u32) IMM);
1768 CONT;
1769 ALU64_DIV_X:
1770 DST = div64_u64(DST, SRC);
1771 CONT;
1772 ALU_DIV_X:
1773 AX = (u32) DST;
1774 do_div(AX, (u32) SRC);
1775 DST = (u32) AX;
1776 CONT;
1777 ALU64_DIV_K:
1778 DST = div64_u64(DST, IMM);
1779 CONT;
1780 ALU_DIV_K:
1781 AX = (u32) DST;
1782 do_div(AX, (u32) IMM);
1783 DST = (u32) AX;
1784 CONT;
1785 ALU_END_TO_BE:
1786 switch (IMM) {
1787 case 16:
1788 DST = (__force u16) cpu_to_be16(DST);
1789 break;
1790 case 32:
1791 DST = (__force u32) cpu_to_be32(DST);
1792 break;
1793 case 64:
1794 DST = (__force u64) cpu_to_be64(DST);
1795 break;
1796 }
1797 CONT;
1798 ALU_END_TO_LE:
1799 switch (IMM) {
1800 case 16:
1801 DST = (__force u16) cpu_to_le16(DST);
1802 break;
1803 case 32:
1804 DST = (__force u32) cpu_to_le32(DST);
1805 break;
1806 case 64:
1807 DST = (__force u64) cpu_to_le64(DST);
1808 break;
1809 }
1810 CONT;
1811
1812 /* CALL */
1813 JMP_CALL:
1814 /* Function call scratches BPF_R1-BPF_R5 registers,
1815 * preserves BPF_R6-BPF_R9, and stores return value
1816 * into BPF_R0.
1817 */
1818 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1819 BPF_R4, BPF_R5);
1820 CONT;
1821
1822 JMP_CALL_ARGS:
1823 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1824 BPF_R3, BPF_R4,
1825 BPF_R5,
1826 insn + insn->off + 1);
1827 CONT;
1828
1829 JMP_TAIL_CALL: {
1830 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1831 struct bpf_array *array = container_of(map, struct bpf_array, map);
1832 struct bpf_prog *prog;
1833 u32 index = BPF_R3;
1834
1835 if (unlikely(index >= array->map.max_entries))
1836 goto out;
1837
1838 if (unlikely(tail_call_cnt >= MAX_TAIL_CALL_CNT))
1839 goto out;
1840
1841 tail_call_cnt++;
1842
1843 prog = READ_ONCE(array->ptrs[index]);
1844 if (!prog)
1845 goto out;
1846
1847 /* ARG1 at this point is guaranteed to point to CTX from
1848 * the verifier side due to the fact that the tail call is
1849 * handled like a helper, that is, bpf_tail_call_proto,
1850 * where arg1_type is ARG_PTR_TO_CTX.
1851 */
1852 insn = prog->insnsi;
1853 goto select_insn;
1854 out:
1855 CONT;
1856 }
1857 JMP_JA:
1858 insn += insn->off;
1859 CONT;
1860 JMP_EXIT:
1861 return BPF_R0;
1862 /* JMP */
1863 #define COND_JMP(SIGN, OPCODE, CMP_OP) \
1864 JMP_##OPCODE##_X: \
1865 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
1866 insn += insn->off; \
1867 CONT_JMP; \
1868 } \
1869 CONT; \
1870 JMP32_##OPCODE##_X: \
1871 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
1872 insn += insn->off; \
1873 CONT_JMP; \
1874 } \
1875 CONT; \
1876 JMP_##OPCODE##_K: \
1877 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
1878 insn += insn->off; \
1879 CONT_JMP; \
1880 } \
1881 CONT; \
1882 JMP32_##OPCODE##_K: \
1883 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
1884 insn += insn->off; \
1885 CONT_JMP; \
1886 } \
1887 CONT;
1888 COND_JMP(u, JEQ, ==)
1889 COND_JMP(u, JNE, !=)
1890 COND_JMP(u, JGT, >)
1891 COND_JMP(u, JLT, <)
1892 COND_JMP(u, JGE, >=)
1893 COND_JMP(u, JLE, <=)
1894 COND_JMP(u, JSET, &)
1895 COND_JMP(s, JSGT, >)
1896 COND_JMP(s, JSLT, <)
1897 COND_JMP(s, JSGE, >=)
1898 COND_JMP(s, JSLE, <=)
1899 #undef COND_JMP
1900 /* ST, STX and LDX*/
1901 ST_NOSPEC:
1902 /* Speculation barrier for mitigating Speculative Store Bypass.
1903 * In case of arm64, we rely on the firmware mitigation as
1904 * controlled via the ssbd kernel parameter. Whenever the
1905 * mitigation is enabled, it works for all of the kernel code
1906 * with no need to provide any additional instructions here.
1907 * In case of x86, we use 'lfence' insn for mitigation. We
1908 * reuse preexisting logic from Spectre v1 mitigation that
1909 * happens to produce the required code on x86 for v4 as well.
1910 */
1911 #ifdef CONFIG_X86
1912 barrier_nospec();
1913 #endif
1914 CONT;
1915 #define LDST(SIZEOP, SIZE) \
1916 STX_MEM_##SIZEOP: \
1917 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1918 CONT; \
1919 ST_MEM_##SIZEOP: \
1920 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1921 CONT; \
1922 LDX_MEM_##SIZEOP: \
1923 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1924 CONT; \
1925 LDX_PROBE_MEM_##SIZEOP: \
1926 bpf_probe_read_kernel(&DST, sizeof(SIZE), \
1927 (const void *)(long) (SRC + insn->off)); \
1928 DST = *((SIZE *)&DST); \
1929 CONT;
1930
1931 LDST(B, u8)
1932 LDST(H, u16)
1933 LDST(W, u32)
1934 LDST(DW, u64)
1935 #undef LDST
1936
1937 #define ATOMIC_ALU_OP(BOP, KOP) \
1938 case BOP: \
1939 if (BPF_SIZE(insn->code) == BPF_W) \
1940 atomic_##KOP((u32) SRC, (atomic_t *)(unsigned long) \
1941 (DST + insn->off)); \
1942 else \
1943 atomic64_##KOP((u64) SRC, (atomic64_t *)(unsigned long) \
1944 (DST + insn->off)); \
1945 break; \
1946 case BOP | BPF_FETCH: \
1947 if (BPF_SIZE(insn->code) == BPF_W) \
1948 SRC = (u32) atomic_fetch_##KOP( \
1949 (u32) SRC, \
1950 (atomic_t *)(unsigned long) (DST + insn->off)); \
1951 else \
1952 SRC = (u64) atomic64_fetch_##KOP( \
1953 (u64) SRC, \
1954 (atomic64_t *)(unsigned long) (DST + insn->off)); \
1955 break;
1956
1957 STX_ATOMIC_DW:
1958 STX_ATOMIC_W:
1959 switch (IMM) {
1960 ATOMIC_ALU_OP(BPF_ADD, add)
1961 ATOMIC_ALU_OP(BPF_AND, and)
1962 ATOMIC_ALU_OP(BPF_OR, or)
1963 ATOMIC_ALU_OP(BPF_XOR, xor)
1964 #undef ATOMIC_ALU_OP
1965
1966 case BPF_XCHG:
1967 if (BPF_SIZE(insn->code) == BPF_W)
1968 SRC = (u32) atomic_xchg(
1969 (atomic_t *)(unsigned long) (DST + insn->off),
1970 (u32) SRC);
1971 else
1972 SRC = (u64) atomic64_xchg(
1973 (atomic64_t *)(unsigned long) (DST + insn->off),
1974 (u64) SRC);
1975 break;
1976 case BPF_CMPXCHG:
1977 if (BPF_SIZE(insn->code) == BPF_W)
1978 BPF_R0 = (u32) atomic_cmpxchg(
1979 (atomic_t *)(unsigned long) (DST + insn->off),
1980 (u32) BPF_R0, (u32) SRC);
1981 else
1982 BPF_R0 = (u64) atomic64_cmpxchg(
1983 (atomic64_t *)(unsigned long) (DST + insn->off),
1984 (u64) BPF_R0, (u64) SRC);
1985 break;
1986
1987 default:
1988 goto default_label;
1989 }
1990 CONT;
1991
1992 default_label:
1993 /* If we ever reach this, we have a bug somewhere. Die hard here
1994 * instead of just returning 0; we could be somewhere in a subprog,
1995 * so execution could continue otherwise which we do /not/ want.
1996 *
1997 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1998 */
1999 pr_warn("BPF interpreter: unknown opcode %02x (imm: 0x%x)\n",
2000 insn->code, insn->imm);
2001 BUG_ON(1);
2002 return 0;
2003 }
2004
2005 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
2006 #define DEFINE_BPF_PROG_RUN(stack_size) \
2007 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
2008 { \
2009 u64 stack[stack_size / sizeof(u64)]; \
2010 u64 regs[MAX_BPF_EXT_REG] = {}; \
2011 \
2012 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2013 ARG1 = (u64) (unsigned long) ctx; \
2014 return ___bpf_prog_run(regs, insn); \
2015 }
2016
2017 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
2018 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
2019 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
2020 const struct bpf_insn *insn) \
2021 { \
2022 u64 stack[stack_size / sizeof(u64)]; \
2023 u64 regs[MAX_BPF_EXT_REG]; \
2024 \
2025 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
2026 BPF_R1 = r1; \
2027 BPF_R2 = r2; \
2028 BPF_R3 = r3; \
2029 BPF_R4 = r4; \
2030 BPF_R5 = r5; \
2031 return ___bpf_prog_run(regs, insn); \
2032 }
2033
2034 #define EVAL1(FN, X) FN(X)
2035 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
2036 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
2037 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
2038 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
2039 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
2040
2041 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
2042 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
2043 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
2044
2045 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
2046 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
2047 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
2048
2049 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
2050
2051 static unsigned int (*interpreters[])(const void *ctx,
2052 const struct bpf_insn *insn) = {
2053 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2054 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2055 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2056 };
2057 #undef PROG_NAME_LIST
2058 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
2059 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
2060 const struct bpf_insn *insn) = {
2061 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
2062 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
2063 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
2064 };
2065 #undef PROG_NAME_LIST
2066
bpf_patch_call_args(struct bpf_insn * insn,u32 stack_depth)2067 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
2068 {
2069 stack_depth = max_t(u32, stack_depth, 1);
2070 insn->off = (s16) insn->imm;
2071 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
2072 __bpf_call_base_args;
2073 insn->code = BPF_JMP | BPF_CALL_ARGS;
2074 }
2075
2076 #else
__bpf_prog_ret0_warn(const void * ctx,const struct bpf_insn * insn)2077 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
2078 const struct bpf_insn *insn)
2079 {
2080 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
2081 * is not working properly, so warn about it!
2082 */
2083 WARN_ON_ONCE(1);
2084 return 0;
2085 }
2086 #endif
2087
bpf_prog_map_compatible(struct bpf_map * map,const struct bpf_prog * fp)2088 bool bpf_prog_map_compatible(struct bpf_map *map,
2089 const struct bpf_prog *fp)
2090 {
2091 enum bpf_prog_type prog_type = resolve_prog_type(fp);
2092 bool ret;
2093
2094 if (fp->kprobe_override)
2095 return false;
2096
2097 spin_lock(&map->owner.lock);
2098 if (!map->owner.type) {
2099 /* There's no owner yet where we could check for
2100 * compatibility.
2101 */
2102 map->owner.type = prog_type;
2103 map->owner.jited = fp->jited;
2104 map->owner.xdp_has_frags = fp->aux->xdp_has_frags;
2105 ret = true;
2106 } else {
2107 ret = map->owner.type == prog_type &&
2108 map->owner.jited == fp->jited &&
2109 map->owner.xdp_has_frags == fp->aux->xdp_has_frags;
2110 }
2111 spin_unlock(&map->owner.lock);
2112
2113 return ret;
2114 }
2115
bpf_check_tail_call(const struct bpf_prog * fp)2116 static int bpf_check_tail_call(const struct bpf_prog *fp)
2117 {
2118 struct bpf_prog_aux *aux = fp->aux;
2119 int i, ret = 0;
2120
2121 mutex_lock(&aux->used_maps_mutex);
2122 for (i = 0; i < aux->used_map_cnt; i++) {
2123 struct bpf_map *map = aux->used_maps[i];
2124
2125 if (!map_type_contains_progs(map))
2126 continue;
2127
2128 if (!bpf_prog_map_compatible(map, fp)) {
2129 ret = -EINVAL;
2130 goto out;
2131 }
2132 }
2133
2134 out:
2135 mutex_unlock(&aux->used_maps_mutex);
2136 return ret;
2137 }
2138
bpf_prog_select_func(struct bpf_prog * fp)2139 static void bpf_prog_select_func(struct bpf_prog *fp)
2140 {
2141 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
2142 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
2143
2144 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
2145 #else
2146 fp->bpf_func = __bpf_prog_ret0_warn;
2147 #endif
2148 }
2149
2150 /**
2151 * bpf_prog_select_runtime - select exec runtime for BPF program
2152 * @fp: bpf_prog populated with BPF program
2153 * @err: pointer to error variable
2154 *
2155 * Try to JIT eBPF program, if JIT is not available, use interpreter.
2156 * The BPF program will be executed via bpf_prog_run() function.
2157 *
2158 * Return: the &fp argument along with &err set to 0 for success or
2159 * a negative errno code on failure
2160 */
bpf_prog_select_runtime(struct bpf_prog * fp,int * err)2161 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
2162 {
2163 /* In case of BPF to BPF calls, verifier did all the prep
2164 * work with regards to JITing, etc.
2165 */
2166 bool jit_needed = false;
2167
2168 if (fp->bpf_func)
2169 goto finalize;
2170
2171 if (IS_ENABLED(CONFIG_BPF_JIT_ALWAYS_ON) ||
2172 bpf_prog_has_kfunc_call(fp))
2173 jit_needed = true;
2174
2175 bpf_prog_select_func(fp);
2176
2177 /* eBPF JITs can rewrite the program in case constant
2178 * blinding is active. However, in case of error during
2179 * blinding, bpf_int_jit_compile() must always return a
2180 * valid program, which in this case would simply not
2181 * be JITed, but falls back to the interpreter.
2182 */
2183 if (!bpf_prog_is_dev_bound(fp->aux)) {
2184 *err = bpf_prog_alloc_jited_linfo(fp);
2185 if (*err)
2186 return fp;
2187
2188 fp = bpf_int_jit_compile(fp);
2189 bpf_prog_jit_attempt_done(fp);
2190 if (!fp->jited && jit_needed) {
2191 *err = -ENOTSUPP;
2192 return fp;
2193 }
2194 } else {
2195 *err = bpf_prog_offload_compile(fp);
2196 if (*err)
2197 return fp;
2198 }
2199
2200 finalize:
2201 bpf_prog_lock_ro(fp);
2202
2203 /* The tail call compatibility check can only be done at
2204 * this late stage as we need to determine, if we deal
2205 * with JITed or non JITed program concatenations and not
2206 * all eBPF JITs might immediately support all features.
2207 */
2208 *err = bpf_check_tail_call(fp);
2209
2210 return fp;
2211 }
2212 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
2213
__bpf_prog_ret1(const void * ctx,const struct bpf_insn * insn)2214 static unsigned int __bpf_prog_ret1(const void *ctx,
2215 const struct bpf_insn *insn)
2216 {
2217 return 1;
2218 }
2219
2220 static struct bpf_prog_dummy {
2221 struct bpf_prog prog;
2222 } dummy_bpf_prog = {
2223 .prog = {
2224 .bpf_func = __bpf_prog_ret1,
2225 },
2226 };
2227
2228 struct bpf_empty_prog_array bpf_empty_prog_array = {
2229 .null_prog = NULL,
2230 };
2231 EXPORT_SYMBOL(bpf_empty_prog_array);
2232
bpf_prog_array_alloc(u32 prog_cnt,gfp_t flags)2233 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
2234 {
2235 if (prog_cnt)
2236 return kzalloc(sizeof(struct bpf_prog_array) +
2237 sizeof(struct bpf_prog_array_item) *
2238 (prog_cnt + 1),
2239 flags);
2240
2241 return &bpf_empty_prog_array.hdr;
2242 }
2243
bpf_prog_array_free(struct bpf_prog_array * progs)2244 void bpf_prog_array_free(struct bpf_prog_array *progs)
2245 {
2246 if (!progs || progs == &bpf_empty_prog_array.hdr)
2247 return;
2248 kfree_rcu(progs, rcu);
2249 }
2250
__bpf_prog_array_free_sleepable_cb(struct rcu_head * rcu)2251 static void __bpf_prog_array_free_sleepable_cb(struct rcu_head *rcu)
2252 {
2253 struct bpf_prog_array *progs;
2254
2255 progs = container_of(rcu, struct bpf_prog_array, rcu);
2256 kfree_rcu(progs, rcu);
2257 }
2258
bpf_prog_array_free_sleepable(struct bpf_prog_array * progs)2259 void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs)
2260 {
2261 if (!progs || progs == &bpf_empty_prog_array.hdr)
2262 return;
2263 call_rcu_tasks_trace(&progs->rcu, __bpf_prog_array_free_sleepable_cb);
2264 }
2265
bpf_prog_array_length(struct bpf_prog_array * array)2266 int bpf_prog_array_length(struct bpf_prog_array *array)
2267 {
2268 struct bpf_prog_array_item *item;
2269 u32 cnt = 0;
2270
2271 for (item = array->items; item->prog; item++)
2272 if (item->prog != &dummy_bpf_prog.prog)
2273 cnt++;
2274 return cnt;
2275 }
2276
bpf_prog_array_is_empty(struct bpf_prog_array * array)2277 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
2278 {
2279 struct bpf_prog_array_item *item;
2280
2281 for (item = array->items; item->prog; item++)
2282 if (item->prog != &dummy_bpf_prog.prog)
2283 return false;
2284 return true;
2285 }
2286
bpf_prog_array_copy_core(struct bpf_prog_array * array,u32 * prog_ids,u32 request_cnt)2287 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
2288 u32 *prog_ids,
2289 u32 request_cnt)
2290 {
2291 struct bpf_prog_array_item *item;
2292 int i = 0;
2293
2294 for (item = array->items; item->prog; item++) {
2295 if (item->prog == &dummy_bpf_prog.prog)
2296 continue;
2297 prog_ids[i] = item->prog->aux->id;
2298 if (++i == request_cnt) {
2299 item++;
2300 break;
2301 }
2302 }
2303
2304 return !!(item->prog);
2305 }
2306
bpf_prog_array_copy_to_user(struct bpf_prog_array * array,__u32 __user * prog_ids,u32 cnt)2307 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
2308 __u32 __user *prog_ids, u32 cnt)
2309 {
2310 unsigned long err = 0;
2311 bool nospc;
2312 u32 *ids;
2313
2314 /* users of this function are doing:
2315 * cnt = bpf_prog_array_length();
2316 * if (cnt > 0)
2317 * bpf_prog_array_copy_to_user(..., cnt);
2318 * so below kcalloc doesn't need extra cnt > 0 check.
2319 */
2320 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
2321 if (!ids)
2322 return -ENOMEM;
2323 nospc = bpf_prog_array_copy_core(array, ids, cnt);
2324 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
2325 kfree(ids);
2326 if (err)
2327 return -EFAULT;
2328 if (nospc)
2329 return -ENOSPC;
2330 return 0;
2331 }
2332
bpf_prog_array_delete_safe(struct bpf_prog_array * array,struct bpf_prog * old_prog)2333 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
2334 struct bpf_prog *old_prog)
2335 {
2336 struct bpf_prog_array_item *item;
2337
2338 for (item = array->items; item->prog; item++)
2339 if (item->prog == old_prog) {
2340 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
2341 break;
2342 }
2343 }
2344
2345 /**
2346 * bpf_prog_array_delete_safe_at() - Replaces the program at the given
2347 * index into the program array with
2348 * a dummy no-op program.
2349 * @array: a bpf_prog_array
2350 * @index: the index of the program to replace
2351 *
2352 * Skips over dummy programs, by not counting them, when calculating
2353 * the position of the program to replace.
2354 *
2355 * Return:
2356 * * 0 - Success
2357 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2358 * * -ENOENT - Index out of range
2359 */
bpf_prog_array_delete_safe_at(struct bpf_prog_array * array,int index)2360 int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index)
2361 {
2362 return bpf_prog_array_update_at(array, index, &dummy_bpf_prog.prog);
2363 }
2364
2365 /**
2366 * bpf_prog_array_update_at() - Updates the program at the given index
2367 * into the program array.
2368 * @array: a bpf_prog_array
2369 * @index: the index of the program to update
2370 * @prog: the program to insert into the array
2371 *
2372 * Skips over dummy programs, by not counting them, when calculating
2373 * the position of the program to update.
2374 *
2375 * Return:
2376 * * 0 - Success
2377 * * -EINVAL - Invalid index value. Must be a non-negative integer.
2378 * * -ENOENT - Index out of range
2379 */
bpf_prog_array_update_at(struct bpf_prog_array * array,int index,struct bpf_prog * prog)2380 int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2381 struct bpf_prog *prog)
2382 {
2383 struct bpf_prog_array_item *item;
2384
2385 if (unlikely(index < 0))
2386 return -EINVAL;
2387
2388 for (item = array->items; item->prog; item++) {
2389 if (item->prog == &dummy_bpf_prog.prog)
2390 continue;
2391 if (!index) {
2392 WRITE_ONCE(item->prog, prog);
2393 return 0;
2394 }
2395 index--;
2396 }
2397 return -ENOENT;
2398 }
2399
bpf_prog_array_copy(struct bpf_prog_array * old_array,struct bpf_prog * exclude_prog,struct bpf_prog * include_prog,u64 bpf_cookie,struct bpf_prog_array ** new_array)2400 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2401 struct bpf_prog *exclude_prog,
2402 struct bpf_prog *include_prog,
2403 u64 bpf_cookie,
2404 struct bpf_prog_array **new_array)
2405 {
2406 int new_prog_cnt, carry_prog_cnt = 0;
2407 struct bpf_prog_array_item *existing, *new;
2408 struct bpf_prog_array *array;
2409 bool found_exclude = false;
2410
2411 /* Figure out how many existing progs we need to carry over to
2412 * the new array.
2413 */
2414 if (old_array) {
2415 existing = old_array->items;
2416 for (; existing->prog; existing++) {
2417 if (existing->prog == exclude_prog) {
2418 found_exclude = true;
2419 continue;
2420 }
2421 if (existing->prog != &dummy_bpf_prog.prog)
2422 carry_prog_cnt++;
2423 if (existing->prog == include_prog)
2424 return -EEXIST;
2425 }
2426 }
2427
2428 if (exclude_prog && !found_exclude)
2429 return -ENOENT;
2430
2431 /* How many progs (not NULL) will be in the new array? */
2432 new_prog_cnt = carry_prog_cnt;
2433 if (include_prog)
2434 new_prog_cnt += 1;
2435
2436 /* Do we have any prog (not NULL) in the new array? */
2437 if (!new_prog_cnt) {
2438 *new_array = NULL;
2439 return 0;
2440 }
2441
2442 /* +1 as the end of prog_array is marked with NULL */
2443 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
2444 if (!array)
2445 return -ENOMEM;
2446 new = array->items;
2447
2448 /* Fill in the new prog array */
2449 if (carry_prog_cnt) {
2450 existing = old_array->items;
2451 for (; existing->prog; existing++) {
2452 if (existing->prog == exclude_prog ||
2453 existing->prog == &dummy_bpf_prog.prog)
2454 continue;
2455
2456 new->prog = existing->prog;
2457 new->bpf_cookie = existing->bpf_cookie;
2458 new++;
2459 }
2460 }
2461 if (include_prog) {
2462 new->prog = include_prog;
2463 new->bpf_cookie = bpf_cookie;
2464 new++;
2465 }
2466 new->prog = NULL;
2467 *new_array = array;
2468 return 0;
2469 }
2470
bpf_prog_array_copy_info(struct bpf_prog_array * array,u32 * prog_ids,u32 request_cnt,u32 * prog_cnt)2471 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2472 u32 *prog_ids, u32 request_cnt,
2473 u32 *prog_cnt)
2474 {
2475 u32 cnt = 0;
2476
2477 if (array)
2478 cnt = bpf_prog_array_length(array);
2479
2480 *prog_cnt = cnt;
2481
2482 /* return early if user requested only program count or nothing to copy */
2483 if (!request_cnt || !cnt)
2484 return 0;
2485
2486 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
2487 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
2488 : 0;
2489 }
2490
__bpf_free_used_maps(struct bpf_prog_aux * aux,struct bpf_map ** used_maps,u32 len)2491 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
2492 struct bpf_map **used_maps, u32 len)
2493 {
2494 struct bpf_map *map;
2495 u32 i;
2496
2497 for (i = 0; i < len; i++) {
2498 map = used_maps[i];
2499 if (map->ops->map_poke_untrack)
2500 map->ops->map_poke_untrack(map, aux);
2501 bpf_map_put(map);
2502 }
2503 }
2504
bpf_free_used_maps(struct bpf_prog_aux * aux)2505 static void bpf_free_used_maps(struct bpf_prog_aux *aux)
2506 {
2507 __bpf_free_used_maps(aux, aux->used_maps, aux->used_map_cnt);
2508 kfree(aux->used_maps);
2509 }
2510
__bpf_free_used_btfs(struct bpf_prog_aux * aux,struct btf_mod_pair * used_btfs,u32 len)2511 void __bpf_free_used_btfs(struct bpf_prog_aux *aux,
2512 struct btf_mod_pair *used_btfs, u32 len)
2513 {
2514 #ifdef CONFIG_BPF_SYSCALL
2515 struct btf_mod_pair *btf_mod;
2516 u32 i;
2517
2518 for (i = 0; i < len; i++) {
2519 btf_mod = &used_btfs[i];
2520 if (btf_mod->module)
2521 module_put(btf_mod->module);
2522 btf_put(btf_mod->btf);
2523 }
2524 #endif
2525 }
2526
bpf_free_used_btfs(struct bpf_prog_aux * aux)2527 static void bpf_free_used_btfs(struct bpf_prog_aux *aux)
2528 {
2529 __bpf_free_used_btfs(aux, aux->used_btfs, aux->used_btf_cnt);
2530 kfree(aux->used_btfs);
2531 }
2532
bpf_prog_free_deferred(struct work_struct * work)2533 static void bpf_prog_free_deferred(struct work_struct *work)
2534 {
2535 struct bpf_prog_aux *aux;
2536 int i;
2537
2538 aux = container_of(work, struct bpf_prog_aux, work);
2539 #ifdef CONFIG_BPF_SYSCALL
2540 bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
2541 #endif
2542 #ifdef CONFIG_CGROUP_BPF
2543 if (aux->cgroup_atype != CGROUP_BPF_ATTACH_TYPE_INVALID)
2544 bpf_cgroup_atype_put(aux->cgroup_atype);
2545 #endif
2546 bpf_free_used_maps(aux);
2547 bpf_free_used_btfs(aux);
2548 if (bpf_prog_is_dev_bound(aux))
2549 bpf_prog_offload_destroy(aux->prog);
2550 #ifdef CONFIG_PERF_EVENTS
2551 if (aux->prog->has_callchain_buf)
2552 put_callchain_buffers();
2553 #endif
2554 if (aux->dst_trampoline)
2555 bpf_trampoline_put(aux->dst_trampoline);
2556 for (i = 0; i < aux->func_cnt; i++) {
2557 /* We can just unlink the subprog poke descriptor table as
2558 * it was originally linked to the main program and is also
2559 * released along with it.
2560 */
2561 aux->func[i]->aux->poke_tab = NULL;
2562 bpf_jit_free(aux->func[i]);
2563 }
2564 if (aux->func_cnt) {
2565 kfree(aux->func);
2566 bpf_prog_unlock_free(aux->prog);
2567 } else {
2568 bpf_jit_free(aux->prog);
2569 }
2570 }
2571
bpf_prog_free(struct bpf_prog * fp)2572 void bpf_prog_free(struct bpf_prog *fp)
2573 {
2574 struct bpf_prog_aux *aux = fp->aux;
2575
2576 if (aux->dst_prog)
2577 bpf_prog_put(aux->dst_prog);
2578 INIT_WORK(&aux->work, bpf_prog_free_deferred);
2579 schedule_work(&aux->work);
2580 }
2581 EXPORT_SYMBOL_GPL(bpf_prog_free);
2582
2583 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2584 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2585
bpf_user_rnd_init_once(void)2586 void bpf_user_rnd_init_once(void)
2587 {
2588 prandom_init_once(&bpf_user_rnd_state);
2589 }
2590
BPF_CALL_0(bpf_user_rnd_u32)2591 BPF_CALL_0(bpf_user_rnd_u32)
2592 {
2593 /* Should someone ever have the rather unwise idea to use some
2594 * of the registers passed into this function, then note that
2595 * this function is called from native eBPF and classic-to-eBPF
2596 * transformations. Register assignments from both sides are
2597 * different, f.e. classic always sets fn(ctx, A, X) here.
2598 */
2599 struct rnd_state *state;
2600 u32 res;
2601
2602 state = &get_cpu_var(bpf_user_rnd_state);
2603 res = prandom_u32_state(state);
2604 put_cpu_var(bpf_user_rnd_state);
2605
2606 return res;
2607 }
2608
BPF_CALL_0(bpf_get_raw_cpu_id)2609 BPF_CALL_0(bpf_get_raw_cpu_id)
2610 {
2611 return raw_smp_processor_id();
2612 }
2613
2614 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2615 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2616 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2617 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2618 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2619 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2620 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2621 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto __weak;
2622 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2623 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2624 const struct bpf_func_proto bpf_jiffies64_proto __weak;
2625
2626 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2627 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2628 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2629 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2630 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto __weak;
2631 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto __weak;
2632 const struct bpf_func_proto bpf_ktime_get_tai_ns_proto __weak;
2633
2634 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2635 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2636 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2637 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2638 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto __weak;
2639 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2640 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
2641 const struct bpf_func_proto bpf_snprintf_btf_proto __weak;
2642 const struct bpf_func_proto bpf_seq_printf_btf_proto __weak;
2643 const struct bpf_func_proto bpf_set_retval_proto __weak;
2644 const struct bpf_func_proto bpf_get_retval_proto __weak;
2645
bpf_get_trace_printk_proto(void)2646 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2647 {
2648 return NULL;
2649 }
2650
bpf_get_trace_vprintk_proto(void)2651 const struct bpf_func_proto * __weak bpf_get_trace_vprintk_proto(void)
2652 {
2653 return NULL;
2654 }
2655
2656 u64 __weak
bpf_event_output(struct bpf_map * map,u64 flags,void * meta,u64 meta_size,void * ctx,u64 ctx_size,bpf_ctx_copy_t ctx_copy)2657 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2658 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2659 {
2660 return -ENOTSUPP;
2661 }
2662 EXPORT_SYMBOL_GPL(bpf_event_output);
2663
2664 /* Always built-in helper functions. */
2665 const struct bpf_func_proto bpf_tail_call_proto = {
2666 .func = NULL,
2667 .gpl_only = false,
2668 .ret_type = RET_VOID,
2669 .arg1_type = ARG_PTR_TO_CTX,
2670 .arg2_type = ARG_CONST_MAP_PTR,
2671 .arg3_type = ARG_ANYTHING,
2672 };
2673
2674 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2675 * It is encouraged to implement bpf_int_jit_compile() instead, so that
2676 * eBPF and implicitly also cBPF can get JITed!
2677 */
bpf_int_jit_compile(struct bpf_prog * prog)2678 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2679 {
2680 return prog;
2681 }
2682
2683 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2684 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2685 */
bpf_jit_compile(struct bpf_prog * prog)2686 void __weak bpf_jit_compile(struct bpf_prog *prog)
2687 {
2688 }
2689
bpf_helper_changes_pkt_data(void * func)2690 bool __weak bpf_helper_changes_pkt_data(void *func)
2691 {
2692 return false;
2693 }
2694
2695 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2696 * analysis code and wants explicit zero extension inserted by verifier.
2697 * Otherwise, return FALSE.
2698 *
2699 * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
2700 * you don't override this. JITs that don't want these extra insns can detect
2701 * them using insn_is_zext.
2702 */
bpf_jit_needs_zext(void)2703 bool __weak bpf_jit_needs_zext(void)
2704 {
2705 return false;
2706 }
2707
2708 /* Return TRUE if the JIT backend supports mixing bpf2bpf and tailcalls. */
bpf_jit_supports_subprog_tailcalls(void)2709 bool __weak bpf_jit_supports_subprog_tailcalls(void)
2710 {
2711 return false;
2712 }
2713
bpf_jit_supports_kfunc_call(void)2714 bool __weak bpf_jit_supports_kfunc_call(void)
2715 {
2716 return false;
2717 }
2718
2719 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2720 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2721 */
skb_copy_bits(const struct sk_buff * skb,int offset,void * to,int len)2722 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2723 int len)
2724 {
2725 return -EFAULT;
2726 }
2727
bpf_arch_text_poke(void * ip,enum bpf_text_poke_type t,void * addr1,void * addr2)2728 int __weak bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
2729 void *addr1, void *addr2)
2730 {
2731 return -ENOTSUPP;
2732 }
2733
bpf_arch_text_copy(void * dst,void * src,size_t len)2734 void * __weak bpf_arch_text_copy(void *dst, void *src, size_t len)
2735 {
2736 return ERR_PTR(-ENOTSUPP);
2737 }
2738
bpf_arch_text_invalidate(void * dst,size_t len)2739 int __weak bpf_arch_text_invalidate(void *dst, size_t len)
2740 {
2741 return -ENOTSUPP;
2742 }
2743
2744 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2745 EXPORT_SYMBOL(bpf_stats_enabled_key);
2746
2747 /* All definitions of tracepoints related to BPF. */
2748 #define CREATE_TRACE_POINTS
2749 #include <linux/bpf_trace.h>
2750
2751 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2752 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
2753