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
34 #include <asm/unaligned.h>
35
36 /* Registers */
37 #define BPF_R0 regs[BPF_REG_0]
38 #define BPF_R1 regs[BPF_REG_1]
39 #define BPF_R2 regs[BPF_REG_2]
40 #define BPF_R3 regs[BPF_REG_3]
41 #define BPF_R4 regs[BPF_REG_4]
42 #define BPF_R5 regs[BPF_REG_5]
43 #define BPF_R6 regs[BPF_REG_6]
44 #define BPF_R7 regs[BPF_REG_7]
45 #define BPF_R8 regs[BPF_REG_8]
46 #define BPF_R9 regs[BPF_REG_9]
47 #define BPF_R10 regs[BPF_REG_10]
48
49 /* Named registers */
50 #define DST regs[insn->dst_reg]
51 #define SRC regs[insn->src_reg]
52 #define FP regs[BPF_REG_FP]
53 #define AX regs[BPF_REG_AX]
54 #define ARG1 regs[BPF_REG_ARG1]
55 #define CTX regs[BPF_REG_CTX]
56 #define IMM insn->imm
57
58 /* No hurry in this branch
59 *
60 * Exported for the bpf jit load helper.
61 */
bpf_internal_load_pointer_neg_helper(const struct sk_buff * skb,int k,unsigned int size)62 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
63 {
64 u8 *ptr = NULL;
65
66 if (k >= SKF_NET_OFF)
67 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
68 else if (k >= SKF_LL_OFF)
69 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
70
71 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
72 return ptr;
73
74 return NULL;
75 }
76
bpf_prog_alloc_no_stats(unsigned int size,gfp_t gfp_extra_flags)77 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags)
78 {
79 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
80 struct bpf_prog_aux *aux;
81 struct bpf_prog *fp;
82
83 size = round_up(size, PAGE_SIZE);
84 fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
85 if (fp == NULL)
86 return NULL;
87
88 aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
89 if (aux == NULL) {
90 vfree(fp);
91 return NULL;
92 }
93
94 fp->pages = size / PAGE_SIZE;
95 fp->aux = aux;
96 fp->aux->prog = fp;
97 fp->jit_requested = ebpf_jit_enabled();
98
99 INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode);
100
101 return fp;
102 }
103
bpf_prog_alloc(unsigned int size,gfp_t gfp_extra_flags)104 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
105 {
106 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
107 struct bpf_prog *prog;
108 int cpu;
109
110 prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags);
111 if (!prog)
112 return NULL;
113
114 prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags);
115 if (!prog->aux->stats) {
116 kfree(prog->aux);
117 vfree(prog);
118 return NULL;
119 }
120
121 for_each_possible_cpu(cpu) {
122 struct bpf_prog_stats *pstats;
123
124 pstats = per_cpu_ptr(prog->aux->stats, cpu);
125 u64_stats_init(&pstats->syncp);
126 }
127 return prog;
128 }
129 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
130
bpf_prog_alloc_jited_linfo(struct bpf_prog * prog)131 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog)
132 {
133 if (!prog->aux->nr_linfo || !prog->jit_requested)
134 return 0;
135
136 prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo,
137 sizeof(*prog->aux->jited_linfo),
138 GFP_KERNEL | __GFP_NOWARN);
139 if (!prog->aux->jited_linfo)
140 return -ENOMEM;
141
142 return 0;
143 }
144
bpf_prog_free_jited_linfo(struct bpf_prog * prog)145 void bpf_prog_free_jited_linfo(struct bpf_prog *prog)
146 {
147 kfree(prog->aux->jited_linfo);
148 prog->aux->jited_linfo = NULL;
149 }
150
bpf_prog_free_unused_jited_linfo(struct bpf_prog * prog)151 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog)
152 {
153 if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0])
154 bpf_prog_free_jited_linfo(prog);
155 }
156
157 /* The jit engine is responsible to provide an array
158 * for insn_off to the jited_off mapping (insn_to_jit_off).
159 *
160 * The idx to this array is the insn_off. Hence, the insn_off
161 * here is relative to the prog itself instead of the main prog.
162 * This array has one entry for each xlated bpf insn.
163 *
164 * jited_off is the byte off to the last byte of the jited insn.
165 *
166 * Hence, with
167 * insn_start:
168 * The first bpf insn off of the prog. The insn off
169 * here is relative to the main prog.
170 * e.g. if prog is a subprog, insn_start > 0
171 * linfo_idx:
172 * The prog's idx to prog->aux->linfo and jited_linfo
173 *
174 * jited_linfo[linfo_idx] = prog->bpf_func
175 *
176 * For i > linfo_idx,
177 *
178 * jited_linfo[i] = prog->bpf_func +
179 * insn_to_jit_off[linfo[i].insn_off - insn_start - 1]
180 */
bpf_prog_fill_jited_linfo(struct bpf_prog * prog,const u32 * insn_to_jit_off)181 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
182 const u32 *insn_to_jit_off)
183 {
184 u32 linfo_idx, insn_start, insn_end, nr_linfo, i;
185 const struct bpf_line_info *linfo;
186 void **jited_linfo;
187
188 if (!prog->aux->jited_linfo)
189 /* Userspace did not provide linfo */
190 return;
191
192 linfo_idx = prog->aux->linfo_idx;
193 linfo = &prog->aux->linfo[linfo_idx];
194 insn_start = linfo[0].insn_off;
195 insn_end = insn_start + prog->len;
196
197 jited_linfo = &prog->aux->jited_linfo[linfo_idx];
198 jited_linfo[0] = prog->bpf_func;
199
200 nr_linfo = prog->aux->nr_linfo - linfo_idx;
201
202 for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++)
203 /* The verifier ensures that linfo[i].insn_off is
204 * strictly increasing
205 */
206 jited_linfo[i] = prog->bpf_func +
207 insn_to_jit_off[linfo[i].insn_off - insn_start - 1];
208 }
209
bpf_prog_free_linfo(struct bpf_prog * prog)210 void bpf_prog_free_linfo(struct bpf_prog *prog)
211 {
212 bpf_prog_free_jited_linfo(prog);
213 kvfree(prog->aux->linfo);
214 }
215
bpf_prog_realloc(struct bpf_prog * fp_old,unsigned int size,gfp_t gfp_extra_flags)216 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
217 gfp_t gfp_extra_flags)
218 {
219 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
220 struct bpf_prog *fp;
221 u32 pages, delta;
222 int ret;
223
224 BUG_ON(fp_old == NULL);
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, PAGE_KERNEL);
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
__bpf_prog_free(struct bpf_prog * fp)254 void __bpf_prog_free(struct bpf_prog *fp)
255 {
256 if (fp->aux) {
257 free_percpu(fp->aux->stats);
258 kfree(fp->aux);
259 }
260 vfree(fp);
261 }
262
bpf_prog_calc_tag(struct bpf_prog * fp)263 int bpf_prog_calc_tag(struct bpf_prog *fp)
264 {
265 const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64);
266 u32 raw_size = bpf_prog_tag_scratch_size(fp);
267 u32 digest[SHA_DIGEST_WORDS];
268 u32 ws[SHA_WORKSPACE_WORDS];
269 u32 i, bsize, psize, blocks;
270 struct bpf_insn *dst;
271 bool was_ld_map;
272 u8 *raw, *todo;
273 __be32 *result;
274 __be64 *bits;
275
276 raw = vmalloc(raw_size);
277 if (!raw)
278 return -ENOMEM;
279
280 sha_init(digest);
281 memset(ws, 0, sizeof(ws));
282
283 /* We need to take out the map fd for the digest calculation
284 * since they are unstable from user space side.
285 */
286 dst = (void *)raw;
287 for (i = 0, was_ld_map = false; i < fp->len; i++) {
288 dst[i] = fp->insnsi[i];
289 if (!was_ld_map &&
290 dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
291 (dst[i].src_reg == BPF_PSEUDO_MAP_FD ||
292 dst[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
293 was_ld_map = true;
294 dst[i].imm = 0;
295 } else if (was_ld_map &&
296 dst[i].code == 0 &&
297 dst[i].dst_reg == 0 &&
298 dst[i].src_reg == 0 &&
299 dst[i].off == 0) {
300 was_ld_map = false;
301 dst[i].imm = 0;
302 } else {
303 was_ld_map = false;
304 }
305 }
306
307 psize = bpf_prog_insn_size(fp);
308 memset(&raw[psize], 0, raw_size - psize);
309 raw[psize++] = 0x80;
310
311 bsize = round_up(psize, SHA_MESSAGE_BYTES);
312 blocks = bsize / SHA_MESSAGE_BYTES;
313 todo = raw;
314 if (bsize - psize >= sizeof(__be64)) {
315 bits = (__be64 *)(todo + bsize - sizeof(__be64));
316 } else {
317 bits = (__be64 *)(todo + bsize + bits_offset);
318 blocks++;
319 }
320 *bits = cpu_to_be64((psize - 1) << 3);
321
322 while (blocks--) {
323 sha_transform(digest, todo, ws);
324 todo += SHA_MESSAGE_BYTES;
325 }
326
327 result = (__force __be32 *)digest;
328 for (i = 0; i < SHA_DIGEST_WORDS; i++)
329 result[i] = cpu_to_be32(digest[i]);
330 memcpy(fp->tag, result, sizeof(fp->tag));
331
332 vfree(raw);
333 return 0;
334 }
335
bpf_adj_delta_to_imm(struct bpf_insn * insn,u32 pos,s32 end_old,s32 end_new,s32 curr,const bool probe_pass)336 static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old,
337 s32 end_new, s32 curr, const bool probe_pass)
338 {
339 const s64 imm_min = S32_MIN, imm_max = S32_MAX;
340 s32 delta = end_new - end_old;
341 s64 imm = insn->imm;
342
343 if (curr < pos && curr + imm + 1 >= end_old)
344 imm += delta;
345 else if (curr >= end_new && curr + imm + 1 < end_new)
346 imm -= delta;
347 if (imm < imm_min || imm > imm_max)
348 return -ERANGE;
349 if (!probe_pass)
350 insn->imm = imm;
351 return 0;
352 }
353
bpf_adj_delta_to_off(struct bpf_insn * insn,u32 pos,s32 end_old,s32 end_new,s32 curr,const bool probe_pass)354 static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old,
355 s32 end_new, s32 curr, const bool probe_pass)
356 {
357 const s32 off_min = S16_MIN, off_max = S16_MAX;
358 s32 delta = end_new - end_old;
359 s32 off = insn->off;
360
361 if (curr < pos && curr + off + 1 >= end_old)
362 off += delta;
363 else if (curr >= end_new && curr + off + 1 < end_new)
364 off -= delta;
365 if (off < off_min || off > off_max)
366 return -ERANGE;
367 if (!probe_pass)
368 insn->off = off;
369 return 0;
370 }
371
bpf_adj_branches(struct bpf_prog * prog,u32 pos,s32 end_old,s32 end_new,const bool probe_pass)372 static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old,
373 s32 end_new, const bool probe_pass)
374 {
375 u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0);
376 struct bpf_insn *insn = prog->insnsi;
377 int ret = 0;
378
379 for (i = 0; i < insn_cnt; i++, insn++) {
380 u8 code;
381
382 /* In the probing pass we still operate on the original,
383 * unpatched image in order to check overflows before we
384 * do any other adjustments. Therefore skip the patchlet.
385 */
386 if (probe_pass && i == pos) {
387 i = end_new;
388 insn = prog->insnsi + end_old;
389 }
390 code = insn->code;
391 if ((BPF_CLASS(code) != BPF_JMP &&
392 BPF_CLASS(code) != BPF_JMP32) ||
393 BPF_OP(code) == BPF_EXIT)
394 continue;
395 /* Adjust offset of jmps if we cross patch boundaries. */
396 if (BPF_OP(code) == BPF_CALL) {
397 if (insn->src_reg != BPF_PSEUDO_CALL)
398 continue;
399 ret = bpf_adj_delta_to_imm(insn, pos, end_old,
400 end_new, i, probe_pass);
401 } else {
402 ret = bpf_adj_delta_to_off(insn, pos, end_old,
403 end_new, i, probe_pass);
404 }
405 if (ret)
406 break;
407 }
408
409 return ret;
410 }
411
bpf_adj_linfo(struct bpf_prog * prog,u32 off,u32 delta)412 static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta)
413 {
414 struct bpf_line_info *linfo;
415 u32 i, nr_linfo;
416
417 nr_linfo = prog->aux->nr_linfo;
418 if (!nr_linfo || !delta)
419 return;
420
421 linfo = prog->aux->linfo;
422
423 for (i = 0; i < nr_linfo; i++)
424 if (off < linfo[i].insn_off)
425 break;
426
427 /* Push all off < linfo[i].insn_off by delta */
428 for (; i < nr_linfo; i++)
429 linfo[i].insn_off += delta;
430 }
431
bpf_patch_insn_single(struct bpf_prog * prog,u32 off,const struct bpf_insn * patch,u32 len)432 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
433 const struct bpf_insn *patch, u32 len)
434 {
435 u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
436 const u32 cnt_max = S16_MAX;
437 struct bpf_prog *prog_adj;
438 int err;
439
440 /* Since our patchlet doesn't expand the image, we're done. */
441 if (insn_delta == 0) {
442 memcpy(prog->insnsi + off, patch, sizeof(*patch));
443 return prog;
444 }
445
446 insn_adj_cnt = prog->len + insn_delta;
447
448 /* Reject anything that would potentially let the insn->off
449 * target overflow when we have excessive program expansions.
450 * We need to probe here before we do any reallocation where
451 * we afterwards may not fail anymore.
452 */
453 if (insn_adj_cnt > cnt_max &&
454 (err = bpf_adj_branches(prog, off, off + 1, off + len, true)))
455 return ERR_PTR(err);
456
457 /* Several new instructions need to be inserted. Make room
458 * for them. Likely, there's no need for a new allocation as
459 * last page could have large enough tailroom.
460 */
461 prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt),
462 GFP_USER);
463 if (!prog_adj)
464 return ERR_PTR(-ENOMEM);
465
466 prog_adj->len = insn_adj_cnt;
467
468 /* Patching happens in 3 steps:
469 *
470 * 1) Move over tail of insnsi from next instruction onwards,
471 * so we can patch the single target insn with one or more
472 * new ones (patching is always from 1 to n insns, n > 0).
473 * 2) Inject new instructions at the target location.
474 * 3) Adjust branch offsets if necessary.
475 */
476 insn_rest = insn_adj_cnt - off - len;
477
478 memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1,
479 sizeof(*patch) * insn_rest);
480 memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
481
482 /* We are guaranteed to not fail at this point, otherwise
483 * the ship has sailed to reverse to the original state. An
484 * overflow cannot happen at this point.
485 */
486 BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false));
487
488 bpf_adj_linfo(prog_adj, off, insn_delta);
489
490 return prog_adj;
491 }
492
bpf_remove_insns(struct bpf_prog * prog,u32 off,u32 cnt)493 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt)
494 {
495 /* Branch offsets can't overflow when program is shrinking, no need
496 * to call bpf_adj_branches(..., true) here
497 */
498 memmove(prog->insnsi + off, prog->insnsi + off + cnt,
499 sizeof(struct bpf_insn) * (prog->len - off - cnt));
500 prog->len -= cnt;
501
502 return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false));
503 }
504
bpf_prog_kallsyms_del_subprogs(struct bpf_prog * fp)505 static void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp)
506 {
507 int i;
508
509 for (i = 0; i < fp->aux->func_cnt; i++)
510 bpf_prog_kallsyms_del(fp->aux->func[i]);
511 }
512
bpf_prog_kallsyms_del_all(struct bpf_prog * fp)513 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp)
514 {
515 bpf_prog_kallsyms_del_subprogs(fp);
516 bpf_prog_kallsyms_del(fp);
517 }
518
519 #ifdef CONFIG_BPF_JIT
520 /* All BPF JIT sysctl knobs here. */
521 int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON);
522 int bpf_jit_harden __read_mostly;
523 int bpf_jit_kallsyms __read_mostly;
524 long bpf_jit_limit __read_mostly;
525
526 static __always_inline void
bpf_get_prog_addr_region(const struct bpf_prog * prog,unsigned long * symbol_start,unsigned long * symbol_end)527 bpf_get_prog_addr_region(const struct bpf_prog *prog,
528 unsigned long *symbol_start,
529 unsigned long *symbol_end)
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 *symbol_start = addr;
537 *symbol_end = addr + hdr->pages * PAGE_SIZE;
538 }
539
bpf_get_prog_name(const struct bpf_prog * prog,char * sym)540 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
541 {
542 const char *end = sym + KSYM_NAME_LEN;
543 const struct btf_type *type;
544 const char *func_name;
545
546 BUILD_BUG_ON(sizeof("bpf_prog_") +
547 sizeof(prog->tag) * 2 +
548 /* name has been null terminated.
549 * We should need +1 for the '_' preceding
550 * the name. However, the null character
551 * is double counted between the name and the
552 * sizeof("bpf_prog_") above, so we omit
553 * the +1 here.
554 */
555 sizeof(prog->aux->name) > KSYM_NAME_LEN);
556
557 sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_");
558 sym = bin2hex(sym, prog->tag, sizeof(prog->tag));
559
560 /* prog->aux->name will be ignored if full btf name is available */
561 if (prog->aux->func_info_cnt) {
562 type = btf_type_by_id(prog->aux->btf,
563 prog->aux->func_info[prog->aux->func_idx].type_id);
564 func_name = btf_name_by_offset(prog->aux->btf, type->name_off);
565 snprintf(sym, (size_t)(end - sym), "_%s", func_name);
566 return;
567 }
568
569 if (prog->aux->name[0])
570 snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name);
571 else
572 *sym = 0;
573 }
574
575 static __always_inline unsigned long
bpf_get_prog_addr_start(struct latch_tree_node * n)576 bpf_get_prog_addr_start(struct latch_tree_node *n)
577 {
578 unsigned long symbol_start, symbol_end;
579 const struct bpf_prog_aux *aux;
580
581 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
582 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
583
584 return symbol_start;
585 }
586
bpf_tree_less(struct latch_tree_node * a,struct latch_tree_node * b)587 static __always_inline bool bpf_tree_less(struct latch_tree_node *a,
588 struct latch_tree_node *b)
589 {
590 return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b);
591 }
592
bpf_tree_comp(void * key,struct latch_tree_node * n)593 static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n)
594 {
595 unsigned long val = (unsigned long)key;
596 unsigned long symbol_start, symbol_end;
597 const struct bpf_prog_aux *aux;
598
599 aux = container_of(n, struct bpf_prog_aux, ksym_tnode);
600 bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end);
601
602 if (val < symbol_start)
603 return -1;
604 if (val >= symbol_end)
605 return 1;
606
607 return 0;
608 }
609
610 static const struct latch_tree_ops bpf_tree_ops = {
611 .less = bpf_tree_less,
612 .comp = bpf_tree_comp,
613 };
614
615 static DEFINE_SPINLOCK(bpf_lock);
616 static LIST_HEAD(bpf_kallsyms);
617 static struct latch_tree_root bpf_tree __cacheline_aligned;
618
bpf_prog_ksym_node_add(struct bpf_prog_aux * aux)619 static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux)
620 {
621 WARN_ON_ONCE(!list_empty(&aux->ksym_lnode));
622 list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms);
623 latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
624 }
625
bpf_prog_ksym_node_del(struct bpf_prog_aux * aux)626 static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux)
627 {
628 if (list_empty(&aux->ksym_lnode))
629 return;
630
631 latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops);
632 list_del_rcu(&aux->ksym_lnode);
633 }
634
bpf_prog_kallsyms_candidate(const struct bpf_prog * fp)635 static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp)
636 {
637 return fp->jited && !bpf_prog_was_classic(fp);
638 }
639
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)640 static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
641 {
642 return list_empty(&fp->aux->ksym_lnode) ||
643 fp->aux->ksym_lnode.prev == LIST_POISON2;
644 }
645
bpf_prog_kallsyms_add(struct bpf_prog * fp)646 void bpf_prog_kallsyms_add(struct bpf_prog *fp)
647 {
648 if (!bpf_prog_kallsyms_candidate(fp) ||
649 !capable(CAP_SYS_ADMIN))
650 return;
651
652 spin_lock_bh(&bpf_lock);
653 bpf_prog_ksym_node_add(fp->aux);
654 spin_unlock_bh(&bpf_lock);
655 }
656
bpf_prog_kallsyms_del(struct bpf_prog * fp)657 void bpf_prog_kallsyms_del(struct bpf_prog *fp)
658 {
659 if (!bpf_prog_kallsyms_candidate(fp))
660 return;
661
662 spin_lock_bh(&bpf_lock);
663 bpf_prog_ksym_node_del(fp->aux);
664 spin_unlock_bh(&bpf_lock);
665 }
666
bpf_prog_kallsyms_find(unsigned long addr)667 static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr)
668 {
669 struct latch_tree_node *n;
670
671 if (!bpf_jit_kallsyms_enabled())
672 return NULL;
673
674 n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops);
675 return n ?
676 container_of(n, struct bpf_prog_aux, ksym_tnode)->prog :
677 NULL;
678 }
679
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)680 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
681 unsigned long *off, char *sym)
682 {
683 unsigned long symbol_start, symbol_end;
684 struct bpf_prog *prog;
685 char *ret = NULL;
686
687 rcu_read_lock();
688 prog = bpf_prog_kallsyms_find(addr);
689 if (prog) {
690 bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end);
691 bpf_get_prog_name(prog, sym);
692
693 ret = sym;
694 if (size)
695 *size = symbol_end - symbol_start;
696 if (off)
697 *off = addr - symbol_start;
698 }
699 rcu_read_unlock();
700
701 return ret;
702 }
703
is_bpf_text_address(unsigned long addr)704 bool is_bpf_text_address(unsigned long addr)
705 {
706 bool ret;
707
708 rcu_read_lock();
709 ret = bpf_prog_kallsyms_find(addr) != NULL;
710 rcu_read_unlock();
711
712 return ret;
713 }
714
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)715 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
716 char *sym)
717 {
718 struct bpf_prog_aux *aux;
719 unsigned int it = 0;
720 int ret = -ERANGE;
721
722 if (!bpf_jit_kallsyms_enabled())
723 return ret;
724
725 rcu_read_lock();
726 list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) {
727 if (it++ != symnum)
728 continue;
729
730 bpf_get_prog_name(aux->prog, sym);
731
732 *value = (unsigned long)aux->prog->bpf_func;
733 *type = BPF_SYM_ELF_TYPE;
734
735 ret = 0;
736 break;
737 }
738 rcu_read_unlock();
739
740 return ret;
741 }
742
743 static atomic_long_t bpf_jit_current;
744
745 /* Can be overridden by an arch's JIT compiler if it has a custom,
746 * dedicated BPF backend memory area, or if neither of the two
747 * below apply.
748 */
bpf_jit_alloc_exec_limit(void)749 u64 __weak bpf_jit_alloc_exec_limit(void)
750 {
751 #if defined(MODULES_VADDR)
752 return MODULES_END - MODULES_VADDR;
753 #else
754 return VMALLOC_END - VMALLOC_START;
755 #endif
756 }
757
bpf_jit_charge_init(void)758 static int __init bpf_jit_charge_init(void)
759 {
760 /* Only used as heuristic here to derive limit. */
761 bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2,
762 PAGE_SIZE), LONG_MAX);
763 return 0;
764 }
765 pure_initcall(bpf_jit_charge_init);
766
bpf_jit_charge_modmem(u32 pages)767 static int bpf_jit_charge_modmem(u32 pages)
768 {
769 if (atomic_long_add_return(pages, &bpf_jit_current) >
770 (bpf_jit_limit >> PAGE_SHIFT)) {
771 if (!capable(CAP_SYS_ADMIN)) {
772 atomic_long_sub(pages, &bpf_jit_current);
773 return -EPERM;
774 }
775 }
776
777 return 0;
778 }
779
bpf_jit_uncharge_modmem(u32 pages)780 static void bpf_jit_uncharge_modmem(u32 pages)
781 {
782 atomic_long_sub(pages, &bpf_jit_current);
783 }
784
bpf_jit_alloc_exec(unsigned long size)785 void *__weak bpf_jit_alloc_exec(unsigned long size)
786 {
787 return module_alloc(size);
788 }
789
bpf_jit_free_exec(void * addr)790 void __weak bpf_jit_free_exec(void *addr)
791 {
792 module_memfree(addr);
793 }
794
795 #if IS_ENABLED(CONFIG_BPF_JIT) && IS_ENABLED(CONFIG_CFI_CLANG)
arch_bpf_jit_check_func(const struct bpf_prog * prog)796 bool __weak arch_bpf_jit_check_func(const struct bpf_prog *prog)
797 {
798 return true;
799 }
800 EXPORT_SYMBOL(arch_bpf_jit_check_func);
801 #endif
802
803 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)804 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
805 unsigned int alignment,
806 bpf_jit_fill_hole_t bpf_fill_ill_insns)
807 {
808 struct bpf_binary_header *hdr;
809 u32 size, hole, start, pages;
810
811 /* Most of BPF filters are really small, but if some of them
812 * fill a page, allow at least 128 extra bytes to insert a
813 * random section of illegal instructions.
814 */
815 size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
816 pages = size / PAGE_SIZE;
817
818 if (bpf_jit_charge_modmem(pages))
819 return NULL;
820 hdr = bpf_jit_alloc_exec(size);
821 if (!hdr) {
822 bpf_jit_uncharge_modmem(pages);
823 return NULL;
824 }
825
826 /* Fill space with illegal/arch-dep instructions. */
827 bpf_fill_ill_insns(hdr, size);
828
829 bpf_jit_set_header_magic(hdr);
830 hdr->pages = pages;
831 hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
832 PAGE_SIZE - sizeof(*hdr));
833 start = (get_random_int() % hole) & ~(alignment - 1);
834
835 /* Leave a random number of instructions before BPF code. */
836 *image_ptr = &hdr->image[start];
837
838 return hdr;
839 }
840
bpf_jit_binary_free(struct bpf_binary_header * hdr)841 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
842 {
843 u32 pages = hdr->pages;
844
845 bpf_jit_free_exec(hdr);
846 bpf_jit_uncharge_modmem(pages);
847 }
848
849 /* This symbol is only overridden by archs that have different
850 * requirements than the usual eBPF JITs, f.e. when they only
851 * implement cBPF JIT, do not set images read-only, etc.
852 */
bpf_jit_free(struct bpf_prog * fp)853 void __weak bpf_jit_free(struct bpf_prog *fp)
854 {
855 if (fp->jited) {
856 struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp);
857
858 bpf_jit_binary_free(hdr);
859
860 WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
861 }
862
863 bpf_prog_unlock_free(fp);
864 }
865
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)866 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
867 const struct bpf_insn *insn, bool extra_pass,
868 u64 *func_addr, bool *func_addr_fixed)
869 {
870 s16 off = insn->off;
871 s32 imm = insn->imm;
872 u8 *addr;
873
874 *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL;
875 if (!*func_addr_fixed) {
876 /* Place-holder address till the last pass has collected
877 * all addresses for JITed subprograms in which case we
878 * can pick them up from prog->aux.
879 */
880 if (!extra_pass)
881 addr = NULL;
882 else if (prog->aux->func &&
883 off >= 0 && off < prog->aux->func_cnt)
884 addr = (u8 *)prog->aux->func[off]->bpf_func;
885 else
886 return -EINVAL;
887 } else {
888 /* Address of a BPF helper call. Since part of the core
889 * kernel, it's always at a fixed location. __bpf_call_base
890 * and the helper with imm relative to it are both in core
891 * kernel.
892 */
893 addr = (u8 *)__bpf_call_base + imm;
894 }
895
896 *func_addr = (unsigned long)addr;
897 return 0;
898 }
899
bpf_jit_blind_insn(const struct bpf_insn * from,const struct bpf_insn * aux,struct bpf_insn * to_buff,bool emit_zext)900 static int bpf_jit_blind_insn(const struct bpf_insn *from,
901 const struct bpf_insn *aux,
902 struct bpf_insn *to_buff,
903 bool emit_zext)
904 {
905 struct bpf_insn *to = to_buff;
906 u32 imm_rnd = get_random_int();
907 s16 off;
908
909 BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG);
910 BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG);
911
912 /* Constraints on AX register:
913 *
914 * AX register is inaccessible from user space. It is mapped in
915 * all JITs, and used here for constant blinding rewrites. It is
916 * typically "stateless" meaning its contents are only valid within
917 * the executed instruction, but not across several instructions.
918 * There are a few exceptions however which are further detailed
919 * below.
920 *
921 * Constant blinding is only used by JITs, not in the interpreter.
922 * The interpreter uses AX in some occasions as a local temporary
923 * register e.g. in DIV or MOD instructions.
924 *
925 * In restricted circumstances, the verifier can also use the AX
926 * register for rewrites as long as they do not interfere with
927 * the above cases!
928 */
929 if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX)
930 goto out;
931
932 if (from->imm == 0 &&
933 (from->code == (BPF_ALU | BPF_MOV | BPF_K) ||
934 from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) {
935 *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg);
936 goto out;
937 }
938
939 switch (from->code) {
940 case BPF_ALU | BPF_ADD | BPF_K:
941 case BPF_ALU | BPF_SUB | BPF_K:
942 case BPF_ALU | BPF_AND | BPF_K:
943 case BPF_ALU | BPF_OR | BPF_K:
944 case BPF_ALU | BPF_XOR | BPF_K:
945 case BPF_ALU | BPF_MUL | BPF_K:
946 case BPF_ALU | BPF_MOV | BPF_K:
947 case BPF_ALU | BPF_DIV | BPF_K:
948 case BPF_ALU | BPF_MOD | BPF_K:
949 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
950 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
951 *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX);
952 break;
953
954 case BPF_ALU64 | BPF_ADD | BPF_K:
955 case BPF_ALU64 | BPF_SUB | BPF_K:
956 case BPF_ALU64 | BPF_AND | BPF_K:
957 case BPF_ALU64 | BPF_OR | BPF_K:
958 case BPF_ALU64 | BPF_XOR | BPF_K:
959 case BPF_ALU64 | BPF_MUL | BPF_K:
960 case BPF_ALU64 | BPF_MOV | BPF_K:
961 case BPF_ALU64 | BPF_DIV | BPF_K:
962 case BPF_ALU64 | BPF_MOD | BPF_K:
963 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
964 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
965 *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX);
966 break;
967
968 case BPF_JMP | BPF_JEQ | BPF_K:
969 case BPF_JMP | BPF_JNE | BPF_K:
970 case BPF_JMP | BPF_JGT | BPF_K:
971 case BPF_JMP | BPF_JLT | BPF_K:
972 case BPF_JMP | BPF_JGE | BPF_K:
973 case BPF_JMP | BPF_JLE | BPF_K:
974 case BPF_JMP | BPF_JSGT | BPF_K:
975 case BPF_JMP | BPF_JSLT | BPF_K:
976 case BPF_JMP | BPF_JSGE | BPF_K:
977 case BPF_JMP | BPF_JSLE | BPF_K:
978 case BPF_JMP | BPF_JSET | BPF_K:
979 /* Accommodate for extra offset in case of a backjump. */
980 off = from->off;
981 if (off < 0)
982 off -= 2;
983 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
984 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
985 *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off);
986 break;
987
988 case BPF_JMP32 | BPF_JEQ | BPF_K:
989 case BPF_JMP32 | BPF_JNE | BPF_K:
990 case BPF_JMP32 | BPF_JGT | BPF_K:
991 case BPF_JMP32 | BPF_JLT | BPF_K:
992 case BPF_JMP32 | BPF_JGE | BPF_K:
993 case BPF_JMP32 | BPF_JLE | BPF_K:
994 case BPF_JMP32 | BPF_JSGT | BPF_K:
995 case BPF_JMP32 | BPF_JSLT | BPF_K:
996 case BPF_JMP32 | BPF_JSGE | BPF_K:
997 case BPF_JMP32 | BPF_JSLE | BPF_K:
998 case BPF_JMP32 | BPF_JSET | BPF_K:
999 /* Accommodate for extra offset in case of a backjump. */
1000 off = from->off;
1001 if (off < 0)
1002 off -= 2;
1003 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1004 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1005 *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX,
1006 off);
1007 break;
1008
1009 case BPF_LD | BPF_IMM | BPF_DW:
1010 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm);
1011 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1012 *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32);
1013 *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX);
1014 break;
1015 case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */
1016 *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm);
1017 *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1018 if (emit_zext)
1019 *to++ = BPF_ZEXT_REG(BPF_REG_AX);
1020 *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX);
1021 break;
1022
1023 case BPF_ST | BPF_MEM | BPF_DW:
1024 case BPF_ST | BPF_MEM | BPF_W:
1025 case BPF_ST | BPF_MEM | BPF_H:
1026 case BPF_ST | BPF_MEM | BPF_B:
1027 *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm);
1028 *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd);
1029 *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off);
1030 break;
1031 }
1032 out:
1033 return to - to_buff;
1034 }
1035
bpf_prog_clone_create(struct bpf_prog * fp_other,gfp_t gfp_extra_flags)1036 static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other,
1037 gfp_t gfp_extra_flags)
1038 {
1039 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags;
1040 struct bpf_prog *fp;
1041
1042 fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL);
1043 if (fp != NULL) {
1044 /* aux->prog still points to the fp_other one, so
1045 * when promoting the clone to the real program,
1046 * this still needs to be adapted.
1047 */
1048 memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE);
1049 }
1050
1051 return fp;
1052 }
1053
bpf_prog_clone_free(struct bpf_prog * fp)1054 static void bpf_prog_clone_free(struct bpf_prog *fp)
1055 {
1056 /* aux was stolen by the other clone, so we cannot free
1057 * it from this path! It will be freed eventually by the
1058 * other program on release.
1059 *
1060 * At this point, we don't need a deferred release since
1061 * clone is guaranteed to not be locked.
1062 */
1063 fp->aux = NULL;
1064 __bpf_prog_free(fp);
1065 }
1066
bpf_jit_prog_release_other(struct bpf_prog * fp,struct bpf_prog * fp_other)1067 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1068 {
1069 /* We have to repoint aux->prog to self, as we don't
1070 * know whether fp here is the clone or the original.
1071 */
1072 fp->aux->prog = fp;
1073 bpf_prog_clone_free(fp_other);
1074 }
1075
bpf_jit_blind_constants(struct bpf_prog * prog)1076 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog)
1077 {
1078 struct bpf_insn insn_buff[16], aux[2];
1079 struct bpf_prog *clone, *tmp;
1080 int insn_delta, insn_cnt;
1081 struct bpf_insn *insn;
1082 int i, rewritten;
1083
1084 if (!bpf_jit_blinding_enabled(prog) || prog->blinded)
1085 return prog;
1086
1087 clone = bpf_prog_clone_create(prog, GFP_USER);
1088 if (!clone)
1089 return ERR_PTR(-ENOMEM);
1090
1091 insn_cnt = clone->len;
1092 insn = clone->insnsi;
1093
1094 for (i = 0; i < insn_cnt; i++, insn++) {
1095 /* We temporarily need to hold the original ld64 insn
1096 * so that we can still access the first part in the
1097 * second blinding run.
1098 */
1099 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) &&
1100 insn[1].code == 0)
1101 memcpy(aux, insn, sizeof(aux));
1102
1103 rewritten = bpf_jit_blind_insn(insn, aux, insn_buff,
1104 clone->aux->verifier_zext);
1105 if (!rewritten)
1106 continue;
1107
1108 tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten);
1109 if (IS_ERR(tmp)) {
1110 /* Patching may have repointed aux->prog during
1111 * realloc from the original one, so we need to
1112 * fix it up here on error.
1113 */
1114 bpf_jit_prog_release_other(prog, clone);
1115 return tmp;
1116 }
1117
1118 clone = tmp;
1119 insn_delta = rewritten - 1;
1120
1121 /* Walk new program and skip insns we just inserted. */
1122 insn = clone->insnsi + i + insn_delta;
1123 insn_cnt += insn_delta;
1124 i += insn_delta;
1125 }
1126
1127 clone->blinded = 1;
1128 return clone;
1129 }
1130 #endif /* CONFIG_BPF_JIT */
1131
1132 /* Base function for offset calculation. Needs to go into .text section,
1133 * therefore keeping it non-static as well; will also be used by JITs
1134 * anyway later on, so do not let the compiler omit it. This also needs
1135 * to go into kallsyms for correlation from e.g. bpftool, so naming
1136 * must not change.
1137 */
__bpf_call_base(u64 r1,u64 r2,u64 r3,u64 r4,u64 r5)1138 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1139 {
1140 return 0;
1141 }
1142 EXPORT_SYMBOL_GPL(__bpf_call_base);
1143
1144 /* All UAPI available opcodes. */
1145 #define BPF_INSN_MAP(INSN_2, INSN_3) \
1146 /* 32 bit ALU operations. */ \
1147 /* Register based. */ \
1148 INSN_3(ALU, ADD, X), \
1149 INSN_3(ALU, SUB, X), \
1150 INSN_3(ALU, AND, X), \
1151 INSN_3(ALU, OR, X), \
1152 INSN_3(ALU, LSH, X), \
1153 INSN_3(ALU, RSH, X), \
1154 INSN_3(ALU, XOR, X), \
1155 INSN_3(ALU, MUL, X), \
1156 INSN_3(ALU, MOV, X), \
1157 INSN_3(ALU, ARSH, X), \
1158 INSN_3(ALU, DIV, X), \
1159 INSN_3(ALU, MOD, X), \
1160 INSN_2(ALU, NEG), \
1161 INSN_3(ALU, END, TO_BE), \
1162 INSN_3(ALU, END, TO_LE), \
1163 /* Immediate based. */ \
1164 INSN_3(ALU, ADD, K), \
1165 INSN_3(ALU, SUB, K), \
1166 INSN_3(ALU, AND, K), \
1167 INSN_3(ALU, OR, K), \
1168 INSN_3(ALU, LSH, K), \
1169 INSN_3(ALU, RSH, K), \
1170 INSN_3(ALU, XOR, K), \
1171 INSN_3(ALU, MUL, K), \
1172 INSN_3(ALU, MOV, K), \
1173 INSN_3(ALU, ARSH, K), \
1174 INSN_3(ALU, DIV, K), \
1175 INSN_3(ALU, MOD, K), \
1176 /* 64 bit ALU operations. */ \
1177 /* Register based. */ \
1178 INSN_3(ALU64, ADD, X), \
1179 INSN_3(ALU64, SUB, X), \
1180 INSN_3(ALU64, AND, X), \
1181 INSN_3(ALU64, OR, X), \
1182 INSN_3(ALU64, LSH, X), \
1183 INSN_3(ALU64, RSH, X), \
1184 INSN_3(ALU64, XOR, X), \
1185 INSN_3(ALU64, MUL, X), \
1186 INSN_3(ALU64, MOV, X), \
1187 INSN_3(ALU64, ARSH, X), \
1188 INSN_3(ALU64, DIV, X), \
1189 INSN_3(ALU64, MOD, X), \
1190 INSN_2(ALU64, NEG), \
1191 /* Immediate based. */ \
1192 INSN_3(ALU64, ADD, K), \
1193 INSN_3(ALU64, SUB, K), \
1194 INSN_3(ALU64, AND, K), \
1195 INSN_3(ALU64, OR, K), \
1196 INSN_3(ALU64, LSH, K), \
1197 INSN_3(ALU64, RSH, K), \
1198 INSN_3(ALU64, XOR, K), \
1199 INSN_3(ALU64, MUL, K), \
1200 INSN_3(ALU64, MOV, K), \
1201 INSN_3(ALU64, ARSH, K), \
1202 INSN_3(ALU64, DIV, K), \
1203 INSN_3(ALU64, MOD, K), \
1204 /* Call instruction. */ \
1205 INSN_2(JMP, CALL), \
1206 /* Exit instruction. */ \
1207 INSN_2(JMP, EXIT), \
1208 /* 32-bit Jump instructions. */ \
1209 /* Register based. */ \
1210 INSN_3(JMP32, JEQ, X), \
1211 INSN_3(JMP32, JNE, X), \
1212 INSN_3(JMP32, JGT, X), \
1213 INSN_3(JMP32, JLT, X), \
1214 INSN_3(JMP32, JGE, X), \
1215 INSN_3(JMP32, JLE, X), \
1216 INSN_3(JMP32, JSGT, X), \
1217 INSN_3(JMP32, JSLT, X), \
1218 INSN_3(JMP32, JSGE, X), \
1219 INSN_3(JMP32, JSLE, X), \
1220 INSN_3(JMP32, JSET, X), \
1221 /* Immediate based. */ \
1222 INSN_3(JMP32, JEQ, K), \
1223 INSN_3(JMP32, JNE, K), \
1224 INSN_3(JMP32, JGT, K), \
1225 INSN_3(JMP32, JLT, K), \
1226 INSN_3(JMP32, JGE, K), \
1227 INSN_3(JMP32, JLE, K), \
1228 INSN_3(JMP32, JSGT, K), \
1229 INSN_3(JMP32, JSLT, K), \
1230 INSN_3(JMP32, JSGE, K), \
1231 INSN_3(JMP32, JSLE, K), \
1232 INSN_3(JMP32, JSET, K), \
1233 /* Jump instructions. */ \
1234 /* Register based. */ \
1235 INSN_3(JMP, JEQ, X), \
1236 INSN_3(JMP, JNE, X), \
1237 INSN_3(JMP, JGT, X), \
1238 INSN_3(JMP, JLT, X), \
1239 INSN_3(JMP, JGE, X), \
1240 INSN_3(JMP, JLE, X), \
1241 INSN_3(JMP, JSGT, X), \
1242 INSN_3(JMP, JSLT, X), \
1243 INSN_3(JMP, JSGE, X), \
1244 INSN_3(JMP, JSLE, X), \
1245 INSN_3(JMP, JSET, X), \
1246 /* Immediate based. */ \
1247 INSN_3(JMP, JEQ, K), \
1248 INSN_3(JMP, JNE, K), \
1249 INSN_3(JMP, JGT, K), \
1250 INSN_3(JMP, JLT, K), \
1251 INSN_3(JMP, JGE, K), \
1252 INSN_3(JMP, JLE, K), \
1253 INSN_3(JMP, JSGT, K), \
1254 INSN_3(JMP, JSLT, K), \
1255 INSN_3(JMP, JSGE, K), \
1256 INSN_3(JMP, JSLE, K), \
1257 INSN_3(JMP, JSET, K), \
1258 INSN_2(JMP, JA), \
1259 /* Store instructions. */ \
1260 /* Register based. */ \
1261 INSN_3(STX, MEM, B), \
1262 INSN_3(STX, MEM, H), \
1263 INSN_3(STX, MEM, W), \
1264 INSN_3(STX, MEM, DW), \
1265 INSN_3(STX, XADD, W), \
1266 INSN_3(STX, XADD, DW), \
1267 /* Immediate based. */ \
1268 INSN_3(ST, MEM, B), \
1269 INSN_3(ST, MEM, H), \
1270 INSN_3(ST, MEM, W), \
1271 INSN_3(ST, MEM, DW), \
1272 /* Load instructions. */ \
1273 /* Register based. */ \
1274 INSN_3(LDX, MEM, B), \
1275 INSN_3(LDX, MEM, H), \
1276 INSN_3(LDX, MEM, W), \
1277 INSN_3(LDX, MEM, DW), \
1278 /* Immediate based. */ \
1279 INSN_3(LD, IMM, DW)
1280
bpf_opcode_in_insntable(u8 code)1281 bool bpf_opcode_in_insntable(u8 code)
1282 {
1283 #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true
1284 #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true
1285 static const bool public_insntable[256] = {
1286 [0 ... 255] = false,
1287 /* Now overwrite non-defaults ... */
1288 BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL),
1289 /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */
1290 [BPF_LD | BPF_ABS | BPF_B] = true,
1291 [BPF_LD | BPF_ABS | BPF_H] = true,
1292 [BPF_LD | BPF_ABS | BPF_W] = true,
1293 [BPF_LD | BPF_IND | BPF_B] = true,
1294 [BPF_LD | BPF_IND | BPF_H] = true,
1295 [BPF_LD | BPF_IND | BPF_W] = true,
1296 };
1297 #undef BPF_INSN_3_TBL
1298 #undef BPF_INSN_2_TBL
1299 return public_insntable[code];
1300 }
1301
1302 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1303 /**
1304 * __bpf_prog_run - run eBPF program on a given context
1305 * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers
1306 * @insn: is the array of eBPF instructions
1307 * @stack: is the eBPF storage stack
1308 *
1309 * Decode and execute eBPF instructions.
1310 */
___bpf_prog_run(u64 * regs,const struct bpf_insn * insn,u64 * stack)1311 static u64 __no_fgcse ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack)
1312 {
1313 #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y
1314 #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
1315 static const void * const jumptable[256] __annotate_jump_table = {
1316 [0 ... 255] = &&default_label,
1317 /* Now overwrite non-defaults ... */
1318 BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL),
1319 /* Non-UAPI available opcodes. */
1320 [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS,
1321 [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL,
1322 };
1323 #undef BPF_INSN_3_LBL
1324 #undef BPF_INSN_2_LBL
1325 u32 tail_call_cnt = 0;
1326
1327 #define CONT ({ insn++; goto select_insn; })
1328 #define CONT_JMP ({ insn++; goto select_insn; })
1329
1330 select_insn:
1331 goto *jumptable[insn->code];
1332
1333 /* ALU */
1334 #define ALU(OPCODE, OP) \
1335 ALU64_##OPCODE##_X: \
1336 DST = DST OP SRC; \
1337 CONT; \
1338 ALU_##OPCODE##_X: \
1339 DST = (u32) DST OP (u32) SRC; \
1340 CONT; \
1341 ALU64_##OPCODE##_K: \
1342 DST = DST OP IMM; \
1343 CONT; \
1344 ALU_##OPCODE##_K: \
1345 DST = (u32) DST OP (u32) IMM; \
1346 CONT;
1347
1348 ALU(ADD, +)
1349 ALU(SUB, -)
1350 ALU(AND, &)
1351 ALU(OR, |)
1352 ALU(LSH, <<)
1353 ALU(RSH, >>)
1354 ALU(XOR, ^)
1355 ALU(MUL, *)
1356 #undef ALU
1357 ALU_NEG:
1358 DST = (u32) -DST;
1359 CONT;
1360 ALU64_NEG:
1361 DST = -DST;
1362 CONT;
1363 ALU_MOV_X:
1364 DST = (u32) SRC;
1365 CONT;
1366 ALU_MOV_K:
1367 DST = (u32) IMM;
1368 CONT;
1369 ALU64_MOV_X:
1370 DST = SRC;
1371 CONT;
1372 ALU64_MOV_K:
1373 DST = IMM;
1374 CONT;
1375 LD_IMM_DW:
1376 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
1377 insn++;
1378 CONT;
1379 ALU_ARSH_X:
1380 DST = (u64) (u32) (((s32) DST) >> SRC);
1381 CONT;
1382 ALU_ARSH_K:
1383 DST = (u64) (u32) (((s32) DST) >> IMM);
1384 CONT;
1385 ALU64_ARSH_X:
1386 (*(s64 *) &DST) >>= SRC;
1387 CONT;
1388 ALU64_ARSH_K:
1389 (*(s64 *) &DST) >>= IMM;
1390 CONT;
1391 ALU64_MOD_X:
1392 div64_u64_rem(DST, SRC, &AX);
1393 DST = AX;
1394 CONT;
1395 ALU_MOD_X:
1396 AX = (u32) DST;
1397 DST = do_div(AX, (u32) SRC);
1398 CONT;
1399 ALU64_MOD_K:
1400 div64_u64_rem(DST, IMM, &AX);
1401 DST = AX;
1402 CONT;
1403 ALU_MOD_K:
1404 AX = (u32) DST;
1405 DST = do_div(AX, (u32) IMM);
1406 CONT;
1407 ALU64_DIV_X:
1408 DST = div64_u64(DST, SRC);
1409 CONT;
1410 ALU_DIV_X:
1411 AX = (u32) DST;
1412 do_div(AX, (u32) SRC);
1413 DST = (u32) AX;
1414 CONT;
1415 ALU64_DIV_K:
1416 DST = div64_u64(DST, IMM);
1417 CONT;
1418 ALU_DIV_K:
1419 AX = (u32) DST;
1420 do_div(AX, (u32) IMM);
1421 DST = (u32) AX;
1422 CONT;
1423 ALU_END_TO_BE:
1424 switch (IMM) {
1425 case 16:
1426 DST = (__force u16) cpu_to_be16(DST);
1427 break;
1428 case 32:
1429 DST = (__force u32) cpu_to_be32(DST);
1430 break;
1431 case 64:
1432 DST = (__force u64) cpu_to_be64(DST);
1433 break;
1434 }
1435 CONT;
1436 ALU_END_TO_LE:
1437 switch (IMM) {
1438 case 16:
1439 DST = (__force u16) cpu_to_le16(DST);
1440 break;
1441 case 32:
1442 DST = (__force u32) cpu_to_le32(DST);
1443 break;
1444 case 64:
1445 DST = (__force u64) cpu_to_le64(DST);
1446 break;
1447 }
1448 CONT;
1449
1450 /* CALL */
1451 JMP_CALL:
1452 /* Function call scratches BPF_R1-BPF_R5 registers,
1453 * preserves BPF_R6-BPF_R9, and stores return value
1454 * into BPF_R0.
1455 */
1456 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
1457 BPF_R4, BPF_R5);
1458 CONT;
1459
1460 JMP_CALL_ARGS:
1461 BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2,
1462 BPF_R3, BPF_R4,
1463 BPF_R5,
1464 insn + insn->off + 1);
1465 CONT;
1466
1467 JMP_TAIL_CALL: {
1468 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
1469 struct bpf_array *array = container_of(map, struct bpf_array, map);
1470 struct bpf_prog *prog;
1471 u32 index = BPF_R3;
1472
1473 if (unlikely(index >= array->map.max_entries))
1474 goto out;
1475 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
1476 goto out;
1477
1478 tail_call_cnt++;
1479
1480 prog = READ_ONCE(array->ptrs[index]);
1481 if (!prog)
1482 goto out;
1483
1484 /* ARG1 at this point is guaranteed to point to CTX from
1485 * the verifier side due to the fact that the tail call is
1486 * handeled like a helper, that is, bpf_tail_call_proto,
1487 * where arg1_type is ARG_PTR_TO_CTX.
1488 */
1489 insn = prog->insnsi;
1490 goto select_insn;
1491 out:
1492 CONT;
1493 }
1494 JMP_JA:
1495 insn += insn->off;
1496 CONT;
1497 JMP_EXIT:
1498 return BPF_R0;
1499 /* JMP */
1500 #define COND_JMP(SIGN, OPCODE, CMP_OP) \
1501 JMP_##OPCODE##_X: \
1502 if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \
1503 insn += insn->off; \
1504 CONT_JMP; \
1505 } \
1506 CONT; \
1507 JMP32_##OPCODE##_X: \
1508 if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \
1509 insn += insn->off; \
1510 CONT_JMP; \
1511 } \
1512 CONT; \
1513 JMP_##OPCODE##_K: \
1514 if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \
1515 insn += insn->off; \
1516 CONT_JMP; \
1517 } \
1518 CONT; \
1519 JMP32_##OPCODE##_K: \
1520 if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \
1521 insn += insn->off; \
1522 CONT_JMP; \
1523 } \
1524 CONT;
1525 COND_JMP(u, JEQ, ==)
1526 COND_JMP(u, JNE, !=)
1527 COND_JMP(u, JGT, >)
1528 COND_JMP(u, JLT, <)
1529 COND_JMP(u, JGE, >=)
1530 COND_JMP(u, JLE, <=)
1531 COND_JMP(u, JSET, &)
1532 COND_JMP(s, JSGT, >)
1533 COND_JMP(s, JSLT, <)
1534 COND_JMP(s, JSGE, >=)
1535 COND_JMP(s, JSLE, <=)
1536 #undef COND_JMP
1537 /* STX and ST and LDX*/
1538 #define LDST(SIZEOP, SIZE) \
1539 STX_MEM_##SIZEOP: \
1540 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
1541 CONT; \
1542 ST_MEM_##SIZEOP: \
1543 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
1544 CONT; \
1545 LDX_MEM_##SIZEOP: \
1546 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
1547 CONT;
1548
1549 LDST(B, u8)
1550 LDST(H, u16)
1551 LDST(W, u32)
1552 LDST(DW, u64)
1553 #undef LDST
1554 STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
1555 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
1556 (DST + insn->off));
1557 CONT;
1558 STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
1559 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
1560 (DST + insn->off));
1561 CONT;
1562
1563 default_label:
1564 /* If we ever reach this, we have a bug somewhere. Die hard here
1565 * instead of just returning 0; we could be somewhere in a subprog,
1566 * so execution could continue otherwise which we do /not/ want.
1567 *
1568 * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable().
1569 */
1570 pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code);
1571 BUG_ON(1);
1572 return 0;
1573 }
1574
1575 #define PROG_NAME(stack_size) __bpf_prog_run##stack_size
1576 #define DEFINE_BPF_PROG_RUN(stack_size) \
1577 static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \
1578 { \
1579 u64 stack[stack_size / sizeof(u64)]; \
1580 u64 regs[MAX_BPF_EXT_REG]; \
1581 \
1582 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1583 ARG1 = (u64) (unsigned long) ctx; \
1584 return ___bpf_prog_run(regs, insn, stack); \
1585 }
1586
1587 #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size
1588 #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \
1589 static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \
1590 const struct bpf_insn *insn) \
1591 { \
1592 u64 stack[stack_size / sizeof(u64)]; \
1593 u64 regs[MAX_BPF_EXT_REG]; \
1594 \
1595 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \
1596 BPF_R1 = r1; \
1597 BPF_R2 = r2; \
1598 BPF_R3 = r3; \
1599 BPF_R4 = r4; \
1600 BPF_R5 = r5; \
1601 return ___bpf_prog_run(regs, insn, stack); \
1602 }
1603
1604 #define EVAL1(FN, X) FN(X)
1605 #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y)
1606 #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y)
1607 #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y)
1608 #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y)
1609 #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y)
1610
1611 EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192);
1612 EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384);
1613 EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512);
1614
1615 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192);
1616 EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384);
1617 EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512);
1618
1619 #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size),
1620
1621 static unsigned int (*interpreters[])(const void *ctx,
1622 const struct bpf_insn *insn) = {
1623 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1624 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1625 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1626 };
1627 #undef PROG_NAME_LIST
1628 #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size),
1629 static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5,
1630 const struct bpf_insn *insn) = {
1631 EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192)
1632 EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384)
1633 EVAL4(PROG_NAME_LIST, 416, 448, 480, 512)
1634 };
1635 #undef PROG_NAME_LIST
1636
bpf_patch_call_args(struct bpf_insn * insn,u32 stack_depth)1637 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth)
1638 {
1639 stack_depth = max_t(u32, stack_depth, 1);
1640 insn->off = (s16) insn->imm;
1641 insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] -
1642 __bpf_call_base_args;
1643 insn->code = BPF_JMP | BPF_CALL_ARGS;
1644 }
1645
1646 #else
__bpf_prog_ret0_warn(const void * ctx,const struct bpf_insn * insn)1647 static unsigned int __bpf_prog_ret0_warn(const void *ctx,
1648 const struct bpf_insn *insn)
1649 {
1650 /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
1651 * is not working properly, so warn about it!
1652 */
1653 WARN_ON_ONCE(1);
1654 return 0;
1655 }
1656 #endif
1657
bpf_prog_array_compatible(struct bpf_array * array,const struct bpf_prog * fp)1658 bool bpf_prog_array_compatible(struct bpf_array *array,
1659 const struct bpf_prog *fp)
1660 {
1661 if (fp->kprobe_override)
1662 return false;
1663
1664 if (!array->owner_prog_type) {
1665 /* There's no owner yet where we could check for
1666 * compatibility.
1667 */
1668 array->owner_prog_type = fp->type;
1669 array->owner_jited = fp->jited;
1670
1671 return true;
1672 }
1673
1674 return array->owner_prog_type == fp->type &&
1675 array->owner_jited == fp->jited;
1676 }
1677
bpf_check_tail_call(const struct bpf_prog * fp)1678 static int bpf_check_tail_call(const struct bpf_prog *fp)
1679 {
1680 struct bpf_prog_aux *aux = fp->aux;
1681 int i;
1682
1683 for (i = 0; i < aux->used_map_cnt; i++) {
1684 struct bpf_map *map = aux->used_maps[i];
1685 struct bpf_array *array;
1686
1687 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1688 continue;
1689
1690 array = container_of(map, struct bpf_array, map);
1691 if (!bpf_prog_array_compatible(array, fp))
1692 return -EINVAL;
1693 }
1694
1695 return 0;
1696 }
1697
bpf_prog_select_func(struct bpf_prog * fp)1698 static void bpf_prog_select_func(struct bpf_prog *fp)
1699 {
1700 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1701 u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1);
1702
1703 fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1];
1704 #else
1705 fp->bpf_func = __bpf_prog_ret0_warn;
1706 #endif
1707 }
1708
1709 /**
1710 * bpf_prog_select_runtime - select exec runtime for BPF program
1711 * @fp: bpf_prog populated with internal BPF program
1712 * @err: pointer to error variable
1713 *
1714 * Try to JIT eBPF program, if JIT is not available, use interpreter.
1715 * The BPF program will be executed via BPF_PROG_RUN() macro.
1716 */
bpf_prog_select_runtime(struct bpf_prog * fp,int * err)1717 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
1718 {
1719 /* In case of BPF to BPF calls, verifier did all the prep
1720 * work with regards to JITing, etc.
1721 */
1722 if (fp->bpf_func)
1723 goto finalize;
1724
1725 bpf_prog_select_func(fp);
1726
1727 /* eBPF JITs can rewrite the program in case constant
1728 * blinding is active. However, in case of error during
1729 * blinding, bpf_int_jit_compile() must always return a
1730 * valid program, which in this case would simply not
1731 * be JITed, but falls back to the interpreter.
1732 */
1733 if (!bpf_prog_is_dev_bound(fp->aux)) {
1734 *err = bpf_prog_alloc_jited_linfo(fp);
1735 if (*err)
1736 return fp;
1737
1738 fp = bpf_int_jit_compile(fp);
1739 if (!fp->jited) {
1740 bpf_prog_free_jited_linfo(fp);
1741 #ifdef CONFIG_BPF_JIT_ALWAYS_ON
1742 *err = -ENOTSUPP;
1743 return fp;
1744 #endif
1745 } else {
1746 bpf_prog_free_unused_jited_linfo(fp);
1747 }
1748 } else {
1749 *err = bpf_prog_offload_compile(fp);
1750 if (*err)
1751 return fp;
1752 }
1753
1754 finalize:
1755 bpf_prog_lock_ro(fp);
1756
1757 /* The tail call compatibility check can only be done at
1758 * this late stage as we need to determine, if we deal
1759 * with JITed or non JITed program concatenations and not
1760 * all eBPF JITs might immediately support all features.
1761 */
1762 *err = bpf_check_tail_call(fp);
1763
1764 return fp;
1765 }
1766 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
1767
__bpf_prog_ret1(const void * ctx,const struct bpf_insn * insn)1768 static unsigned int __bpf_prog_ret1(const void *ctx,
1769 const struct bpf_insn *insn)
1770 {
1771 return 1;
1772 }
1773
1774 static struct bpf_prog_dummy {
1775 struct bpf_prog prog;
1776 } dummy_bpf_prog = {
1777 .prog = {
1778 .bpf_func = __bpf_prog_ret1,
1779 },
1780 };
1781
1782 /* to avoid allocating empty bpf_prog_array for cgroups that
1783 * don't have bpf program attached use one global 'empty_prog_array'
1784 * It will not be modified the caller of bpf_prog_array_alloc()
1785 * (since caller requested prog_cnt == 0)
1786 * that pointer should be 'freed' by bpf_prog_array_free()
1787 */
1788 static struct {
1789 struct bpf_prog_array hdr;
1790 struct bpf_prog *null_prog;
1791 } empty_prog_array = {
1792 .null_prog = NULL,
1793 };
1794
bpf_prog_array_alloc(u32 prog_cnt,gfp_t flags)1795 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags)
1796 {
1797 if (prog_cnt)
1798 return kzalloc(sizeof(struct bpf_prog_array) +
1799 sizeof(struct bpf_prog_array_item) *
1800 (prog_cnt + 1),
1801 flags);
1802
1803 return &empty_prog_array.hdr;
1804 }
1805
bpf_prog_array_free(struct bpf_prog_array * progs)1806 void bpf_prog_array_free(struct bpf_prog_array *progs)
1807 {
1808 if (!progs || progs == &empty_prog_array.hdr)
1809 return;
1810 kfree_rcu(progs, rcu);
1811 }
1812
bpf_prog_array_length(struct bpf_prog_array * array)1813 int bpf_prog_array_length(struct bpf_prog_array *array)
1814 {
1815 struct bpf_prog_array_item *item;
1816 u32 cnt = 0;
1817
1818 for (item = array->items; item->prog; item++)
1819 if (item->prog != &dummy_bpf_prog.prog)
1820 cnt++;
1821 return cnt;
1822 }
1823
bpf_prog_array_is_empty(struct bpf_prog_array * array)1824 bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
1825 {
1826 struct bpf_prog_array_item *item;
1827
1828 for (item = array->items; item->prog; item++)
1829 if (item->prog != &dummy_bpf_prog.prog)
1830 return false;
1831 return true;
1832 }
1833
bpf_prog_array_copy_core(struct bpf_prog_array * array,u32 * prog_ids,u32 request_cnt)1834 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
1835 u32 *prog_ids,
1836 u32 request_cnt)
1837 {
1838 struct bpf_prog_array_item *item;
1839 int i = 0;
1840
1841 for (item = array->items; item->prog; item++) {
1842 if (item->prog == &dummy_bpf_prog.prog)
1843 continue;
1844 prog_ids[i] = item->prog->aux->id;
1845 if (++i == request_cnt) {
1846 item++;
1847 break;
1848 }
1849 }
1850
1851 return !!(item->prog);
1852 }
1853
bpf_prog_array_copy_to_user(struct bpf_prog_array * array,__u32 __user * prog_ids,u32 cnt)1854 int bpf_prog_array_copy_to_user(struct bpf_prog_array *array,
1855 __u32 __user *prog_ids, u32 cnt)
1856 {
1857 unsigned long err = 0;
1858 bool nospc;
1859 u32 *ids;
1860
1861 /* users of this function are doing:
1862 * cnt = bpf_prog_array_length();
1863 * if (cnt > 0)
1864 * bpf_prog_array_copy_to_user(..., cnt);
1865 * so below kcalloc doesn't need extra cnt > 0 check.
1866 */
1867 ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN);
1868 if (!ids)
1869 return -ENOMEM;
1870 nospc = bpf_prog_array_copy_core(array, ids, cnt);
1871 err = copy_to_user(prog_ids, ids, cnt * sizeof(u32));
1872 kfree(ids);
1873 if (err)
1874 return -EFAULT;
1875 if (nospc)
1876 return -ENOSPC;
1877 return 0;
1878 }
1879
bpf_prog_array_delete_safe(struct bpf_prog_array * array,struct bpf_prog * old_prog)1880 void bpf_prog_array_delete_safe(struct bpf_prog_array *array,
1881 struct bpf_prog *old_prog)
1882 {
1883 struct bpf_prog_array_item *item;
1884
1885 for (item = array->items; item->prog; item++)
1886 if (item->prog == old_prog) {
1887 WRITE_ONCE(item->prog, &dummy_bpf_prog.prog);
1888 break;
1889 }
1890 }
1891
bpf_prog_array_copy(struct bpf_prog_array * old_array,struct bpf_prog * exclude_prog,struct bpf_prog * include_prog,struct bpf_prog_array ** new_array)1892 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
1893 struct bpf_prog *exclude_prog,
1894 struct bpf_prog *include_prog,
1895 struct bpf_prog_array **new_array)
1896 {
1897 int new_prog_cnt, carry_prog_cnt = 0;
1898 struct bpf_prog_array_item *existing;
1899 struct bpf_prog_array *array;
1900 bool found_exclude = false;
1901 int new_prog_idx = 0;
1902
1903 /* Figure out how many existing progs we need to carry over to
1904 * the new array.
1905 */
1906 if (old_array) {
1907 existing = old_array->items;
1908 for (; existing->prog; existing++) {
1909 if (existing->prog == exclude_prog) {
1910 found_exclude = true;
1911 continue;
1912 }
1913 if (existing->prog != &dummy_bpf_prog.prog)
1914 carry_prog_cnt++;
1915 if (existing->prog == include_prog)
1916 return -EEXIST;
1917 }
1918 }
1919
1920 if (exclude_prog && !found_exclude)
1921 return -ENOENT;
1922
1923 /* How many progs (not NULL) will be in the new array? */
1924 new_prog_cnt = carry_prog_cnt;
1925 if (include_prog)
1926 new_prog_cnt += 1;
1927
1928 /* Do we have any prog (not NULL) in the new array? */
1929 if (!new_prog_cnt) {
1930 *new_array = NULL;
1931 return 0;
1932 }
1933
1934 /* +1 as the end of prog_array is marked with NULL */
1935 array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL);
1936 if (!array)
1937 return -ENOMEM;
1938
1939 /* Fill in the new prog array */
1940 if (carry_prog_cnt) {
1941 existing = old_array->items;
1942 for (; existing->prog; existing++)
1943 if (existing->prog != exclude_prog &&
1944 existing->prog != &dummy_bpf_prog.prog) {
1945 array->items[new_prog_idx++].prog =
1946 existing->prog;
1947 }
1948 }
1949 if (include_prog)
1950 array->items[new_prog_idx++].prog = include_prog;
1951 array->items[new_prog_idx].prog = NULL;
1952 *new_array = array;
1953 return 0;
1954 }
1955
bpf_prog_array_copy_info(struct bpf_prog_array * array,u32 * prog_ids,u32 request_cnt,u32 * prog_cnt)1956 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
1957 u32 *prog_ids, u32 request_cnt,
1958 u32 *prog_cnt)
1959 {
1960 u32 cnt = 0;
1961
1962 if (array)
1963 cnt = bpf_prog_array_length(array);
1964
1965 *prog_cnt = cnt;
1966
1967 /* return early if user requested only program count or nothing to copy */
1968 if (!request_cnt || !cnt)
1969 return 0;
1970
1971 /* this function is called under trace/bpf_trace.c: bpf_event_mutex */
1972 return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC
1973 : 0;
1974 }
1975
bpf_prog_free_deferred(struct work_struct * work)1976 static void bpf_prog_free_deferred(struct work_struct *work)
1977 {
1978 struct bpf_prog_aux *aux;
1979 int i;
1980
1981 aux = container_of(work, struct bpf_prog_aux, work);
1982 if (bpf_prog_is_dev_bound(aux))
1983 bpf_prog_offload_destroy(aux->prog);
1984 #ifdef CONFIG_PERF_EVENTS
1985 if (aux->prog->has_callchain_buf)
1986 put_callchain_buffers();
1987 #endif
1988 for (i = 0; i < aux->func_cnt; i++)
1989 bpf_jit_free(aux->func[i]);
1990 if (aux->func_cnt) {
1991 kfree(aux->func);
1992 bpf_prog_unlock_free(aux->prog);
1993 } else {
1994 bpf_jit_free(aux->prog);
1995 }
1996 }
1997
1998 /* Free internal BPF program */
bpf_prog_free(struct bpf_prog * fp)1999 void bpf_prog_free(struct bpf_prog *fp)
2000 {
2001 struct bpf_prog_aux *aux = fp->aux;
2002
2003 INIT_WORK(&aux->work, bpf_prog_free_deferred);
2004 schedule_work(&aux->work);
2005 }
2006 EXPORT_SYMBOL_GPL(bpf_prog_free);
2007
2008 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
2009 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
2010
bpf_user_rnd_init_once(void)2011 void bpf_user_rnd_init_once(void)
2012 {
2013 prandom_init_once(&bpf_user_rnd_state);
2014 }
2015
BPF_CALL_0(bpf_user_rnd_u32)2016 BPF_CALL_0(bpf_user_rnd_u32)
2017 {
2018 /* Should someone ever have the rather unwise idea to use some
2019 * of the registers passed into this function, then note that
2020 * this function is called from native eBPF and classic-to-eBPF
2021 * transformations. Register assignments from both sides are
2022 * different, f.e. classic always sets fn(ctx, A, X) here.
2023 */
2024 struct rnd_state *state;
2025 u32 res;
2026
2027 state = &get_cpu_var(bpf_user_rnd_state);
2028 res = prandom_u32_state(state);
2029 put_cpu_var(bpf_user_rnd_state);
2030
2031 return res;
2032 }
2033
2034 /* Weak definitions of helper functions in case we don't have bpf syscall. */
2035 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
2036 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
2037 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
2038 const struct bpf_func_proto bpf_map_push_elem_proto __weak;
2039 const struct bpf_func_proto bpf_map_pop_elem_proto __weak;
2040 const struct bpf_func_proto bpf_map_peek_elem_proto __weak;
2041 const struct bpf_func_proto bpf_spin_lock_proto __weak;
2042 const struct bpf_func_proto bpf_spin_unlock_proto __weak;
2043
2044 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
2045 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
2046 const struct bpf_func_proto bpf_get_numa_node_id_proto __weak;
2047 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
2048
2049 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
2050 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
2051 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
2052 const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
2053 const struct bpf_func_proto bpf_get_local_storage_proto __weak;
2054
bpf_get_trace_printk_proto(void)2055 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
2056 {
2057 return NULL;
2058 }
2059
2060 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)2061 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2062 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
2063 {
2064 return -ENOTSUPP;
2065 }
2066 EXPORT_SYMBOL_GPL(bpf_event_output);
2067
2068 /* Always built-in helper functions. */
2069 const struct bpf_func_proto bpf_tail_call_proto = {
2070 .func = NULL,
2071 .gpl_only = false,
2072 .ret_type = RET_VOID,
2073 .arg1_type = ARG_PTR_TO_CTX,
2074 .arg2_type = ARG_CONST_MAP_PTR,
2075 .arg3_type = ARG_ANYTHING,
2076 };
2077
2078 /* Stub for JITs that only support cBPF. eBPF programs are interpreted.
2079 * It is encouraged to implement bpf_int_jit_compile() instead, so that
2080 * eBPF and implicitly also cBPF can get JITed!
2081 */
bpf_int_jit_compile(struct bpf_prog * prog)2082 struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog)
2083 {
2084 return prog;
2085 }
2086
2087 /* Stub for JITs that support eBPF. All cBPF code gets transformed into
2088 * eBPF by the kernel and is later compiled by bpf_int_jit_compile().
2089 */
bpf_jit_compile(struct bpf_prog * prog)2090 void __weak bpf_jit_compile(struct bpf_prog *prog)
2091 {
2092 }
2093
bpf_helper_changes_pkt_data(void * func)2094 bool __weak bpf_helper_changes_pkt_data(void *func)
2095 {
2096 return false;
2097 }
2098
2099 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
2100 * analysis code and wants explicit zero extension inserted by verifier.
2101 * Otherwise, return FALSE.
2102 */
bpf_jit_needs_zext(void)2103 bool __weak bpf_jit_needs_zext(void)
2104 {
2105 return false;
2106 }
2107
2108 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
2109 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
2110 */
skb_copy_bits(const struct sk_buff * skb,int offset,void * to,int len)2111 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
2112 int len)
2113 {
2114 return -EFAULT;
2115 }
2116
2117 DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
2118 EXPORT_SYMBOL(bpf_stats_enabled_key);
2119
2120 /* All definitions of tracepoints related to BPF. */
2121 #define CREATE_TRACE_POINTS
2122 #include <linux/bpf_trace.h>
2123
2124 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception);
2125 EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx);
2126