]>
Commit | Line | Data |
---|---|---|
f5bffecd AS |
1 | /* |
2 | * Linux Socket Filter - Kernel level socket filtering | |
3 | * | |
4 | * Based on the design of the Berkeley Packet Filter. The new | |
5 | * internal format has been designed by PLUMgrid: | |
6 | * | |
7 | * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com | |
8 | * | |
9 | * Authors: | |
10 | * | |
11 | * Jay Schulist <jschlst@samba.org> | |
12 | * Alexei Starovoitov <ast@plumgrid.com> | |
13 | * Daniel Borkmann <dborkman@redhat.com> | |
14 | * | |
15 | * This program is free software; you can redistribute it and/or | |
16 | * modify it under the terms of the GNU General Public License | |
17 | * as published by the Free Software Foundation; either version | |
18 | * 2 of the License, or (at your option) any later version. | |
19 | * | |
20 | * Andi Kleen - Fix a few bad bugs and races. | |
4df95ff4 | 21 | * Kris Katterjohn - Added many additional checks in bpf_check_classic() |
f5bffecd | 22 | */ |
738cbe72 | 23 | |
838e9690 | 24 | #include <uapi/linux/btf.h> |
f5bffecd AS |
25 | #include <linux/filter.h> |
26 | #include <linux/skbuff.h> | |
60a3b225 | 27 | #include <linux/vmalloc.h> |
738cbe72 DB |
28 | #include <linux/random.h> |
29 | #include <linux/moduleloader.h> | |
09756af4 | 30 | #include <linux/bpf.h> |
838e9690 | 31 | #include <linux/btf.h> |
39853cc0 | 32 | #include <linux/frame.h> |
74451e66 DB |
33 | #include <linux/rbtree_latch.h> |
34 | #include <linux/kallsyms.h> | |
35 | #include <linux/rcupdate.h> | |
c195651e | 36 | #include <linux/perf_event.h> |
f5bffecd | 37 | |
3324b584 DB |
38 | #include <asm/unaligned.h> |
39 | ||
f5bffecd AS |
40 | /* Registers */ |
41 | #define BPF_R0 regs[BPF_REG_0] | |
42 | #define BPF_R1 regs[BPF_REG_1] | |
43 | #define BPF_R2 regs[BPF_REG_2] | |
44 | #define BPF_R3 regs[BPF_REG_3] | |
45 | #define BPF_R4 regs[BPF_REG_4] | |
46 | #define BPF_R5 regs[BPF_REG_5] | |
47 | #define BPF_R6 regs[BPF_REG_6] | |
48 | #define BPF_R7 regs[BPF_REG_7] | |
49 | #define BPF_R8 regs[BPF_REG_8] | |
50 | #define BPF_R9 regs[BPF_REG_9] | |
51 | #define BPF_R10 regs[BPF_REG_10] | |
52 | ||
53 | /* Named registers */ | |
54 | #define DST regs[insn->dst_reg] | |
55 | #define SRC regs[insn->src_reg] | |
56 | #define FP regs[BPF_REG_FP] | |
144cd91c | 57 | #define AX regs[BPF_REG_AX] |
f5bffecd AS |
58 | #define ARG1 regs[BPF_REG_ARG1] |
59 | #define CTX regs[BPF_REG_CTX] | |
60 | #define IMM insn->imm | |
61 | ||
62 | /* No hurry in this branch | |
63 | * | |
64 | * Exported for the bpf jit load helper. | |
65 | */ | |
66 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size) | |
67 | { | |
68 | u8 *ptr = NULL; | |
69 | ||
70 | if (k >= SKF_NET_OFF) | |
71 | ptr = skb_network_header(skb) + k - SKF_NET_OFF; | |
72 | else if (k >= SKF_LL_OFF) | |
73 | ptr = skb_mac_header(skb) + k - SKF_LL_OFF; | |
3324b584 | 74 | |
f5bffecd AS |
75 | if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb)) |
76 | return ptr; | |
77 | ||
78 | return NULL; | |
79 | } | |
80 | ||
60a3b225 DB |
81 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags) |
82 | { | |
19809c2d | 83 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; |
09756af4 | 84 | struct bpf_prog_aux *aux; |
60a3b225 DB |
85 | struct bpf_prog *fp; |
86 | ||
87 | size = round_up(size, PAGE_SIZE); | |
88 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); | |
89 | if (fp == NULL) | |
90 | return NULL; | |
91 | ||
09756af4 AS |
92 | aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags); |
93 | if (aux == NULL) { | |
60a3b225 DB |
94 | vfree(fp); |
95 | return NULL; | |
96 | } | |
97 | ||
98 | fp->pages = size / PAGE_SIZE; | |
09756af4 | 99 | fp->aux = aux; |
e9d8afa9 | 100 | fp->aux->prog = fp; |
60b58afc | 101 | fp->jit_requested = ebpf_jit_enabled(); |
60a3b225 | 102 | |
74451e66 DB |
103 | INIT_LIST_HEAD_RCU(&fp->aux->ksym_lnode); |
104 | ||
60a3b225 DB |
105 | return fp; |
106 | } | |
107 | EXPORT_SYMBOL_GPL(bpf_prog_alloc); | |
108 | ||
c454a46b MKL |
109 | int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog) |
110 | { | |
111 | if (!prog->aux->nr_linfo || !prog->jit_requested) | |
112 | return 0; | |
113 | ||
114 | prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo, | |
115 | sizeof(*prog->aux->jited_linfo), | |
116 | GFP_KERNEL | __GFP_NOWARN); | |
117 | if (!prog->aux->jited_linfo) | |
118 | return -ENOMEM; | |
119 | ||
120 | return 0; | |
121 | } | |
122 | ||
123 | void bpf_prog_free_jited_linfo(struct bpf_prog *prog) | |
124 | { | |
125 | kfree(prog->aux->jited_linfo); | |
126 | prog->aux->jited_linfo = NULL; | |
127 | } | |
128 | ||
129 | void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog) | |
130 | { | |
131 | if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0]) | |
132 | bpf_prog_free_jited_linfo(prog); | |
133 | } | |
134 | ||
135 | /* The jit engine is responsible to provide an array | |
136 | * for insn_off to the jited_off mapping (insn_to_jit_off). | |
137 | * | |
138 | * The idx to this array is the insn_off. Hence, the insn_off | |
139 | * here is relative to the prog itself instead of the main prog. | |
140 | * This array has one entry for each xlated bpf insn. | |
141 | * | |
142 | * jited_off is the byte off to the last byte of the jited insn. | |
143 | * | |
144 | * Hence, with | |
145 | * insn_start: | |
146 | * The first bpf insn off of the prog. The insn off | |
147 | * here is relative to the main prog. | |
148 | * e.g. if prog is a subprog, insn_start > 0 | |
149 | * linfo_idx: | |
150 | * The prog's idx to prog->aux->linfo and jited_linfo | |
151 | * | |
152 | * jited_linfo[linfo_idx] = prog->bpf_func | |
153 | * | |
154 | * For i > linfo_idx, | |
155 | * | |
156 | * jited_linfo[i] = prog->bpf_func + | |
157 | * insn_to_jit_off[linfo[i].insn_off - insn_start - 1] | |
158 | */ | |
159 | void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, | |
160 | const u32 *insn_to_jit_off) | |
161 | { | |
162 | u32 linfo_idx, insn_start, insn_end, nr_linfo, i; | |
163 | const struct bpf_line_info *linfo; | |
164 | void **jited_linfo; | |
165 | ||
166 | if (!prog->aux->jited_linfo) | |
167 | /* Userspace did not provide linfo */ | |
168 | return; | |
169 | ||
170 | linfo_idx = prog->aux->linfo_idx; | |
171 | linfo = &prog->aux->linfo[linfo_idx]; | |
172 | insn_start = linfo[0].insn_off; | |
173 | insn_end = insn_start + prog->len; | |
174 | ||
175 | jited_linfo = &prog->aux->jited_linfo[linfo_idx]; | |
176 | jited_linfo[0] = prog->bpf_func; | |
177 | ||
178 | nr_linfo = prog->aux->nr_linfo - linfo_idx; | |
179 | ||
180 | for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++) | |
181 | /* The verifier ensures that linfo[i].insn_off is | |
182 | * strictly increasing | |
183 | */ | |
184 | jited_linfo[i] = prog->bpf_func + | |
185 | insn_to_jit_off[linfo[i].insn_off - insn_start - 1]; | |
186 | } | |
187 | ||
188 | void bpf_prog_free_linfo(struct bpf_prog *prog) | |
189 | { | |
190 | bpf_prog_free_jited_linfo(prog); | |
191 | kvfree(prog->aux->linfo); | |
192 | } | |
193 | ||
60a3b225 DB |
194 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, |
195 | gfp_t gfp_extra_flags) | |
196 | { | |
19809c2d | 197 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; |
60a3b225 | 198 | struct bpf_prog *fp; |
5ccb071e DB |
199 | u32 pages, delta; |
200 | int ret; | |
60a3b225 DB |
201 | |
202 | BUG_ON(fp_old == NULL); | |
203 | ||
204 | size = round_up(size, PAGE_SIZE); | |
5ccb071e DB |
205 | pages = size / PAGE_SIZE; |
206 | if (pages <= fp_old->pages) | |
60a3b225 DB |
207 | return fp_old; |
208 | ||
5ccb071e DB |
209 | delta = pages - fp_old->pages; |
210 | ret = __bpf_prog_charge(fp_old->aux->user, delta); | |
211 | if (ret) | |
212 | return NULL; | |
213 | ||
60a3b225 | 214 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
5ccb071e DB |
215 | if (fp == NULL) { |
216 | __bpf_prog_uncharge(fp_old->aux->user, delta); | |
217 | } else { | |
60a3b225 | 218 | memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE); |
5ccb071e | 219 | fp->pages = pages; |
e9d8afa9 | 220 | fp->aux->prog = fp; |
60a3b225 | 221 | |
09756af4 | 222 | /* We keep fp->aux from fp_old around in the new |
60a3b225 DB |
223 | * reallocated structure. |
224 | */ | |
09756af4 | 225 | fp_old->aux = NULL; |
60a3b225 DB |
226 | __bpf_prog_free(fp_old); |
227 | } | |
228 | ||
229 | return fp; | |
230 | } | |
60a3b225 DB |
231 | |
232 | void __bpf_prog_free(struct bpf_prog *fp) | |
233 | { | |
09756af4 | 234 | kfree(fp->aux); |
60a3b225 DB |
235 | vfree(fp); |
236 | } | |
60a3b225 | 237 | |
f1f7714e | 238 | int bpf_prog_calc_tag(struct bpf_prog *fp) |
7bd509e3 DB |
239 | { |
240 | const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64); | |
f1f7714e DB |
241 | u32 raw_size = bpf_prog_tag_scratch_size(fp); |
242 | u32 digest[SHA_DIGEST_WORDS]; | |
aafe6ae9 | 243 | u32 ws[SHA_WORKSPACE_WORDS]; |
7bd509e3 | 244 | u32 i, bsize, psize, blocks; |
aafe6ae9 | 245 | struct bpf_insn *dst; |
7bd509e3 | 246 | bool was_ld_map; |
aafe6ae9 | 247 | u8 *raw, *todo; |
7bd509e3 DB |
248 | __be32 *result; |
249 | __be64 *bits; | |
250 | ||
aafe6ae9 DB |
251 | raw = vmalloc(raw_size); |
252 | if (!raw) | |
253 | return -ENOMEM; | |
254 | ||
f1f7714e | 255 | sha_init(digest); |
7bd509e3 DB |
256 | memset(ws, 0, sizeof(ws)); |
257 | ||
258 | /* We need to take out the map fd for the digest calculation | |
259 | * since they are unstable from user space side. | |
260 | */ | |
aafe6ae9 | 261 | dst = (void *)raw; |
7bd509e3 DB |
262 | for (i = 0, was_ld_map = false; i < fp->len; i++) { |
263 | dst[i] = fp->insnsi[i]; | |
264 | if (!was_ld_map && | |
265 | dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
266 | dst[i].src_reg == BPF_PSEUDO_MAP_FD) { | |
267 | was_ld_map = true; | |
268 | dst[i].imm = 0; | |
269 | } else if (was_ld_map && | |
270 | dst[i].code == 0 && | |
271 | dst[i].dst_reg == 0 && | |
272 | dst[i].src_reg == 0 && | |
273 | dst[i].off == 0) { | |
274 | was_ld_map = false; | |
275 | dst[i].imm = 0; | |
276 | } else { | |
277 | was_ld_map = false; | |
278 | } | |
279 | } | |
280 | ||
aafe6ae9 DB |
281 | psize = bpf_prog_insn_size(fp); |
282 | memset(&raw[psize], 0, raw_size - psize); | |
7bd509e3 DB |
283 | raw[psize++] = 0x80; |
284 | ||
285 | bsize = round_up(psize, SHA_MESSAGE_BYTES); | |
286 | blocks = bsize / SHA_MESSAGE_BYTES; | |
aafe6ae9 | 287 | todo = raw; |
7bd509e3 DB |
288 | if (bsize - psize >= sizeof(__be64)) { |
289 | bits = (__be64 *)(todo + bsize - sizeof(__be64)); | |
290 | } else { | |
291 | bits = (__be64 *)(todo + bsize + bits_offset); | |
292 | blocks++; | |
293 | } | |
294 | *bits = cpu_to_be64((psize - 1) << 3); | |
295 | ||
296 | while (blocks--) { | |
f1f7714e | 297 | sha_transform(digest, todo, ws); |
7bd509e3 DB |
298 | todo += SHA_MESSAGE_BYTES; |
299 | } | |
300 | ||
f1f7714e | 301 | result = (__force __be32 *)digest; |
7bd509e3 | 302 | for (i = 0; i < SHA_DIGEST_WORDS; i++) |
f1f7714e DB |
303 | result[i] = cpu_to_be32(digest[i]); |
304 | memcpy(fp->tag, result, sizeof(fp->tag)); | |
aafe6ae9 DB |
305 | |
306 | vfree(raw); | |
307 | return 0; | |
7bd509e3 DB |
308 | } |
309 | ||
2cbd95a5 JK |
310 | static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old, |
311 | s32 end_new, u32 curr, const bool probe_pass) | |
c237ee5e | 312 | { |
050fad7c | 313 | const s64 imm_min = S32_MIN, imm_max = S32_MAX; |
2cbd95a5 | 314 | s32 delta = end_new - end_old; |
050fad7c DB |
315 | s64 imm = insn->imm; |
316 | ||
2cbd95a5 | 317 | if (curr < pos && curr + imm + 1 >= end_old) |
050fad7c | 318 | imm += delta; |
2cbd95a5 | 319 | else if (curr >= end_new && curr + imm + 1 < end_new) |
050fad7c DB |
320 | imm -= delta; |
321 | if (imm < imm_min || imm > imm_max) | |
322 | return -ERANGE; | |
323 | if (!probe_pass) | |
324 | insn->imm = imm; | |
325 | return 0; | |
326 | } | |
327 | ||
2cbd95a5 JK |
328 | static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old, |
329 | s32 end_new, u32 curr, const bool probe_pass) | |
050fad7c DB |
330 | { |
331 | const s32 off_min = S16_MIN, off_max = S16_MAX; | |
2cbd95a5 | 332 | s32 delta = end_new - end_old; |
050fad7c DB |
333 | s32 off = insn->off; |
334 | ||
2cbd95a5 | 335 | if (curr < pos && curr + off + 1 >= end_old) |
050fad7c | 336 | off += delta; |
2cbd95a5 | 337 | else if (curr >= end_new && curr + off + 1 < end_new) |
050fad7c DB |
338 | off -= delta; |
339 | if (off < off_min || off > off_max) | |
340 | return -ERANGE; | |
341 | if (!probe_pass) | |
342 | insn->off = off; | |
343 | return 0; | |
344 | } | |
345 | ||
2cbd95a5 JK |
346 | static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old, |
347 | s32 end_new, const bool probe_pass) | |
050fad7c | 348 | { |
2cbd95a5 | 349 | u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0); |
c237ee5e | 350 | struct bpf_insn *insn = prog->insnsi; |
050fad7c | 351 | int ret = 0; |
c237ee5e DB |
352 | |
353 | for (i = 0; i < insn_cnt; i++, insn++) { | |
050fad7c DB |
354 | u8 code; |
355 | ||
356 | /* In the probing pass we still operate on the original, | |
357 | * unpatched image in order to check overflows before we | |
358 | * do any other adjustments. Therefore skip the patchlet. | |
359 | */ | |
360 | if (probe_pass && i == pos) { | |
2cbd95a5 JK |
361 | i = end_new; |
362 | insn = prog->insnsi + end_old; | |
050fad7c | 363 | } |
1ea47e01 | 364 | code = insn->code; |
092ed096 JW |
365 | if ((BPF_CLASS(code) != BPF_JMP && |
366 | BPF_CLASS(code) != BPF_JMP32) || | |
050fad7c | 367 | BPF_OP(code) == BPF_EXIT) |
1ea47e01 | 368 | continue; |
050fad7c | 369 | /* Adjust offset of jmps if we cross patch boundaries. */ |
1ea47e01 | 370 | if (BPF_OP(code) == BPF_CALL) { |
050fad7c | 371 | if (insn->src_reg != BPF_PSEUDO_CALL) |
1ea47e01 | 372 | continue; |
2cbd95a5 JK |
373 | ret = bpf_adj_delta_to_imm(insn, pos, end_old, |
374 | end_new, i, probe_pass); | |
1ea47e01 | 375 | } else { |
2cbd95a5 JK |
376 | ret = bpf_adj_delta_to_off(insn, pos, end_old, |
377 | end_new, i, probe_pass); | |
1ea47e01 | 378 | } |
050fad7c DB |
379 | if (ret) |
380 | break; | |
c237ee5e | 381 | } |
050fad7c DB |
382 | |
383 | return ret; | |
c237ee5e DB |
384 | } |
385 | ||
c454a46b MKL |
386 | static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta) |
387 | { | |
388 | struct bpf_line_info *linfo; | |
389 | u32 i, nr_linfo; | |
390 | ||
391 | nr_linfo = prog->aux->nr_linfo; | |
392 | if (!nr_linfo || !delta) | |
393 | return; | |
394 | ||
395 | linfo = prog->aux->linfo; | |
396 | ||
397 | for (i = 0; i < nr_linfo; i++) | |
398 | if (off < linfo[i].insn_off) | |
399 | break; | |
400 | ||
401 | /* Push all off < linfo[i].insn_off by delta */ | |
402 | for (; i < nr_linfo; i++) | |
403 | linfo[i].insn_off += delta; | |
404 | } | |
405 | ||
c237ee5e DB |
406 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
407 | const struct bpf_insn *patch, u32 len) | |
408 | { | |
409 | u32 insn_adj_cnt, insn_rest, insn_delta = len - 1; | |
050fad7c | 410 | const u32 cnt_max = S16_MAX; |
c237ee5e DB |
411 | struct bpf_prog *prog_adj; |
412 | ||
413 | /* Since our patchlet doesn't expand the image, we're done. */ | |
414 | if (insn_delta == 0) { | |
415 | memcpy(prog->insnsi + off, patch, sizeof(*patch)); | |
416 | return prog; | |
417 | } | |
418 | ||
419 | insn_adj_cnt = prog->len + insn_delta; | |
420 | ||
050fad7c DB |
421 | /* Reject anything that would potentially let the insn->off |
422 | * target overflow when we have excessive program expansions. | |
423 | * We need to probe here before we do any reallocation where | |
424 | * we afterwards may not fail anymore. | |
425 | */ | |
426 | if (insn_adj_cnt > cnt_max && | |
2cbd95a5 | 427 | bpf_adj_branches(prog, off, off + 1, off + len, true)) |
050fad7c DB |
428 | return NULL; |
429 | ||
c237ee5e DB |
430 | /* Several new instructions need to be inserted. Make room |
431 | * for them. Likely, there's no need for a new allocation as | |
432 | * last page could have large enough tailroom. | |
433 | */ | |
434 | prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt), | |
435 | GFP_USER); | |
436 | if (!prog_adj) | |
437 | return NULL; | |
438 | ||
439 | prog_adj->len = insn_adj_cnt; | |
440 | ||
441 | /* Patching happens in 3 steps: | |
442 | * | |
443 | * 1) Move over tail of insnsi from next instruction onwards, | |
444 | * so we can patch the single target insn with one or more | |
445 | * new ones (patching is always from 1 to n insns, n > 0). | |
446 | * 2) Inject new instructions at the target location. | |
447 | * 3) Adjust branch offsets if necessary. | |
448 | */ | |
449 | insn_rest = insn_adj_cnt - off - len; | |
450 | ||
451 | memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1, | |
452 | sizeof(*patch) * insn_rest); | |
453 | memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len); | |
454 | ||
050fad7c DB |
455 | /* We are guaranteed to not fail at this point, otherwise |
456 | * the ship has sailed to reverse to the original state. An | |
457 | * overflow cannot happen at this point. | |
458 | */ | |
2cbd95a5 | 459 | BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false)); |
c237ee5e | 460 | |
c454a46b MKL |
461 | bpf_adj_linfo(prog_adj, off, insn_delta); |
462 | ||
c237ee5e DB |
463 | return prog_adj; |
464 | } | |
465 | ||
52875a04 JK |
466 | int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt) |
467 | { | |
468 | /* Branch offsets can't overflow when program is shrinking, no need | |
469 | * to call bpf_adj_branches(..., true) here | |
470 | */ | |
471 | memmove(prog->insnsi + off, prog->insnsi + off + cnt, | |
472 | sizeof(struct bpf_insn) * (prog->len - off - cnt)); | |
473 | prog->len -= cnt; | |
474 | ||
475 | return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false)); | |
476 | } | |
477 | ||
7d1982b4 DB |
478 | void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp) |
479 | { | |
480 | int i; | |
481 | ||
482 | for (i = 0; i < fp->aux->func_cnt; i++) | |
483 | bpf_prog_kallsyms_del(fp->aux->func[i]); | |
484 | } | |
485 | ||
486 | void bpf_prog_kallsyms_del_all(struct bpf_prog *fp) | |
487 | { | |
488 | bpf_prog_kallsyms_del_subprogs(fp); | |
489 | bpf_prog_kallsyms_del(fp); | |
490 | } | |
491 | ||
b954d834 | 492 | #ifdef CONFIG_BPF_JIT |
fa9dd599 DB |
493 | /* All BPF JIT sysctl knobs here. */ |
494 | int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON); | |
495 | int bpf_jit_harden __read_mostly; | |
496 | int bpf_jit_kallsyms __read_mostly; | |
fdadd049 | 497 | long bpf_jit_limit __read_mostly; |
fa9dd599 | 498 | |
74451e66 DB |
499 | static __always_inline void |
500 | bpf_get_prog_addr_region(const struct bpf_prog *prog, | |
501 | unsigned long *symbol_start, | |
502 | unsigned long *symbol_end) | |
503 | { | |
504 | const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog); | |
505 | unsigned long addr = (unsigned long)hdr; | |
506 | ||
507 | WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog)); | |
508 | ||
509 | *symbol_start = addr; | |
510 | *symbol_end = addr + hdr->pages * PAGE_SIZE; | |
511 | } | |
512 | ||
513 | static void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) | |
514 | { | |
368211fb | 515 | const char *end = sym + KSYM_NAME_LEN; |
838e9690 YS |
516 | const struct btf_type *type; |
517 | const char *func_name; | |
368211fb | 518 | |
74451e66 | 519 | BUILD_BUG_ON(sizeof("bpf_prog_") + |
368211fb MKL |
520 | sizeof(prog->tag) * 2 + |
521 | /* name has been null terminated. | |
522 | * We should need +1 for the '_' preceding | |
523 | * the name. However, the null character | |
524 | * is double counted between the name and the | |
525 | * sizeof("bpf_prog_") above, so we omit | |
526 | * the +1 here. | |
527 | */ | |
528 | sizeof(prog->aux->name) > KSYM_NAME_LEN); | |
74451e66 DB |
529 | |
530 | sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_"); | |
531 | sym = bin2hex(sym, prog->tag, sizeof(prog->tag)); | |
838e9690 YS |
532 | |
533 | /* prog->aux->name will be ignored if full btf name is available */ | |
7337224f | 534 | if (prog->aux->func_info_cnt) { |
ba64e7d8 YS |
535 | type = btf_type_by_id(prog->aux->btf, |
536 | prog->aux->func_info[prog->aux->func_idx].type_id); | |
838e9690 YS |
537 | func_name = btf_name_by_offset(prog->aux->btf, type->name_off); |
538 | snprintf(sym, (size_t)(end - sym), "_%s", func_name); | |
539 | return; | |
540 | } | |
541 | ||
368211fb MKL |
542 | if (prog->aux->name[0]) |
543 | snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name); | |
544 | else | |
545 | *sym = 0; | |
74451e66 DB |
546 | } |
547 | ||
548 | static __always_inline unsigned long | |
549 | bpf_get_prog_addr_start(struct latch_tree_node *n) | |
550 | { | |
551 | unsigned long symbol_start, symbol_end; | |
552 | const struct bpf_prog_aux *aux; | |
553 | ||
554 | aux = container_of(n, struct bpf_prog_aux, ksym_tnode); | |
555 | bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); | |
556 | ||
557 | return symbol_start; | |
558 | } | |
559 | ||
560 | static __always_inline bool bpf_tree_less(struct latch_tree_node *a, | |
561 | struct latch_tree_node *b) | |
562 | { | |
563 | return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b); | |
564 | } | |
565 | ||
566 | static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n) | |
567 | { | |
568 | unsigned long val = (unsigned long)key; | |
569 | unsigned long symbol_start, symbol_end; | |
570 | const struct bpf_prog_aux *aux; | |
571 | ||
572 | aux = container_of(n, struct bpf_prog_aux, ksym_tnode); | |
573 | bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); | |
574 | ||
575 | if (val < symbol_start) | |
576 | return -1; | |
577 | if (val >= symbol_end) | |
578 | return 1; | |
579 | ||
580 | return 0; | |
581 | } | |
582 | ||
583 | static const struct latch_tree_ops bpf_tree_ops = { | |
584 | .less = bpf_tree_less, | |
585 | .comp = bpf_tree_comp, | |
586 | }; | |
587 | ||
588 | static DEFINE_SPINLOCK(bpf_lock); | |
589 | static LIST_HEAD(bpf_kallsyms); | |
590 | static struct latch_tree_root bpf_tree __cacheline_aligned; | |
591 | ||
74451e66 DB |
592 | static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux) |
593 | { | |
594 | WARN_ON_ONCE(!list_empty(&aux->ksym_lnode)); | |
595 | list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms); | |
596 | latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); | |
597 | } | |
598 | ||
599 | static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux) | |
600 | { | |
601 | if (list_empty(&aux->ksym_lnode)) | |
602 | return; | |
603 | ||
604 | latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); | |
605 | list_del_rcu(&aux->ksym_lnode); | |
606 | } | |
607 | ||
608 | static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp) | |
609 | { | |
610 | return fp->jited && !bpf_prog_was_classic(fp); | |
611 | } | |
612 | ||
613 | static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) | |
614 | { | |
615 | return list_empty(&fp->aux->ksym_lnode) || | |
616 | fp->aux->ksym_lnode.prev == LIST_POISON2; | |
617 | } | |
618 | ||
619 | void bpf_prog_kallsyms_add(struct bpf_prog *fp) | |
620 | { | |
74451e66 DB |
621 | if (!bpf_prog_kallsyms_candidate(fp) || |
622 | !capable(CAP_SYS_ADMIN)) | |
623 | return; | |
624 | ||
d24f7c7f | 625 | spin_lock_bh(&bpf_lock); |
74451e66 | 626 | bpf_prog_ksym_node_add(fp->aux); |
d24f7c7f | 627 | spin_unlock_bh(&bpf_lock); |
74451e66 DB |
628 | } |
629 | ||
630 | void bpf_prog_kallsyms_del(struct bpf_prog *fp) | |
631 | { | |
74451e66 DB |
632 | if (!bpf_prog_kallsyms_candidate(fp)) |
633 | return; | |
634 | ||
d24f7c7f | 635 | spin_lock_bh(&bpf_lock); |
74451e66 | 636 | bpf_prog_ksym_node_del(fp->aux); |
d24f7c7f | 637 | spin_unlock_bh(&bpf_lock); |
74451e66 DB |
638 | } |
639 | ||
640 | static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr) | |
641 | { | |
642 | struct latch_tree_node *n; | |
643 | ||
644 | if (!bpf_jit_kallsyms_enabled()) | |
645 | return NULL; | |
646 | ||
647 | n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops); | |
648 | return n ? | |
649 | container_of(n, struct bpf_prog_aux, ksym_tnode)->prog : | |
650 | NULL; | |
651 | } | |
652 | ||
653 | const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, | |
654 | unsigned long *off, char *sym) | |
655 | { | |
656 | unsigned long symbol_start, symbol_end; | |
657 | struct bpf_prog *prog; | |
658 | char *ret = NULL; | |
659 | ||
660 | rcu_read_lock(); | |
661 | prog = bpf_prog_kallsyms_find(addr); | |
662 | if (prog) { | |
663 | bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end); | |
664 | bpf_get_prog_name(prog, sym); | |
665 | ||
666 | ret = sym; | |
667 | if (size) | |
668 | *size = symbol_end - symbol_start; | |
669 | if (off) | |
670 | *off = addr - symbol_start; | |
671 | } | |
672 | rcu_read_unlock(); | |
673 | ||
674 | return ret; | |
675 | } | |
676 | ||
677 | bool is_bpf_text_address(unsigned long addr) | |
678 | { | |
679 | bool ret; | |
680 | ||
681 | rcu_read_lock(); | |
682 | ret = bpf_prog_kallsyms_find(addr) != NULL; | |
683 | rcu_read_unlock(); | |
684 | ||
685 | return ret; | |
686 | } | |
687 | ||
688 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
689 | char *sym) | |
690 | { | |
74451e66 DB |
691 | struct bpf_prog_aux *aux; |
692 | unsigned int it = 0; | |
693 | int ret = -ERANGE; | |
694 | ||
695 | if (!bpf_jit_kallsyms_enabled()) | |
696 | return ret; | |
697 | ||
698 | rcu_read_lock(); | |
699 | list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) { | |
700 | if (it++ != symnum) | |
701 | continue; | |
702 | ||
74451e66 DB |
703 | bpf_get_prog_name(aux->prog, sym); |
704 | ||
df073470 | 705 | *value = (unsigned long)aux->prog->bpf_func; |
74451e66 DB |
706 | *type = BPF_SYM_ELF_TYPE; |
707 | ||
708 | ret = 0; | |
709 | break; | |
710 | } | |
711 | rcu_read_unlock(); | |
712 | ||
713 | return ret; | |
714 | } | |
715 | ||
ede95a63 DB |
716 | static atomic_long_t bpf_jit_current; |
717 | ||
fdadd049 DB |
718 | /* Can be overridden by an arch's JIT compiler if it has a custom, |
719 | * dedicated BPF backend memory area, or if neither of the two | |
720 | * below apply. | |
721 | */ | |
722 | u64 __weak bpf_jit_alloc_exec_limit(void) | |
723 | { | |
ede95a63 | 724 | #if defined(MODULES_VADDR) |
fdadd049 DB |
725 | return MODULES_END - MODULES_VADDR; |
726 | #else | |
727 | return VMALLOC_END - VMALLOC_START; | |
728 | #endif | |
729 | } | |
730 | ||
ede95a63 DB |
731 | static int __init bpf_jit_charge_init(void) |
732 | { | |
733 | /* Only used as heuristic here to derive limit. */ | |
fdadd049 DB |
734 | bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2, |
735 | PAGE_SIZE), LONG_MAX); | |
ede95a63 DB |
736 | return 0; |
737 | } | |
738 | pure_initcall(bpf_jit_charge_init); | |
ede95a63 DB |
739 | |
740 | static int bpf_jit_charge_modmem(u32 pages) | |
741 | { | |
742 | if (atomic_long_add_return(pages, &bpf_jit_current) > | |
743 | (bpf_jit_limit >> PAGE_SHIFT)) { | |
744 | if (!capable(CAP_SYS_ADMIN)) { | |
745 | atomic_long_sub(pages, &bpf_jit_current); | |
746 | return -EPERM; | |
747 | } | |
748 | } | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | static void bpf_jit_uncharge_modmem(u32 pages) | |
754 | { | |
755 | atomic_long_sub(pages, &bpf_jit_current); | |
756 | } | |
757 | ||
dc002bb6 AB |
758 | void *__weak bpf_jit_alloc_exec(unsigned long size) |
759 | { | |
760 | return module_alloc(size); | |
761 | } | |
762 | ||
763 | void __weak bpf_jit_free_exec(void *addr) | |
764 | { | |
765 | module_memfree(addr); | |
766 | } | |
767 | ||
738cbe72 DB |
768 | struct bpf_binary_header * |
769 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
770 | unsigned int alignment, | |
771 | bpf_jit_fill_hole_t bpf_fill_ill_insns) | |
772 | { | |
773 | struct bpf_binary_header *hdr; | |
ede95a63 | 774 | u32 size, hole, start, pages; |
738cbe72 DB |
775 | |
776 | /* Most of BPF filters are really small, but if some of them | |
777 | * fill a page, allow at least 128 extra bytes to insert a | |
778 | * random section of illegal instructions. | |
779 | */ | |
780 | size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); | |
ede95a63 DB |
781 | pages = size / PAGE_SIZE; |
782 | ||
783 | if (bpf_jit_charge_modmem(pages)) | |
784 | return NULL; | |
dc002bb6 | 785 | hdr = bpf_jit_alloc_exec(size); |
ede95a63 DB |
786 | if (!hdr) { |
787 | bpf_jit_uncharge_modmem(pages); | |
738cbe72 | 788 | return NULL; |
ede95a63 | 789 | } |
738cbe72 DB |
790 | |
791 | /* Fill space with illegal/arch-dep instructions. */ | |
792 | bpf_fill_ill_insns(hdr, size); | |
793 | ||
ede95a63 | 794 | hdr->pages = pages; |
738cbe72 DB |
795 | hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), |
796 | PAGE_SIZE - sizeof(*hdr)); | |
b7552e1b | 797 | start = (get_random_int() % hole) & ~(alignment - 1); |
738cbe72 DB |
798 | |
799 | /* Leave a random number of instructions before BPF code. */ | |
800 | *image_ptr = &hdr->image[start]; | |
801 | ||
802 | return hdr; | |
803 | } | |
804 | ||
805 | void bpf_jit_binary_free(struct bpf_binary_header *hdr) | |
806 | { | |
ede95a63 DB |
807 | u32 pages = hdr->pages; |
808 | ||
dc002bb6 | 809 | bpf_jit_free_exec(hdr); |
ede95a63 | 810 | bpf_jit_uncharge_modmem(pages); |
738cbe72 | 811 | } |
4f3446bb | 812 | |
74451e66 DB |
813 | /* This symbol is only overridden by archs that have different |
814 | * requirements than the usual eBPF JITs, f.e. when they only | |
815 | * implement cBPF JIT, do not set images read-only, etc. | |
816 | */ | |
817 | void __weak bpf_jit_free(struct bpf_prog *fp) | |
818 | { | |
819 | if (fp->jited) { | |
820 | struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp); | |
821 | ||
822 | bpf_jit_binary_unlock_ro(hdr); | |
823 | bpf_jit_binary_free(hdr); | |
824 | ||
825 | WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp)); | |
826 | } | |
827 | ||
828 | bpf_prog_unlock_free(fp); | |
829 | } | |
830 | ||
e2c95a61 DB |
831 | int bpf_jit_get_func_addr(const struct bpf_prog *prog, |
832 | const struct bpf_insn *insn, bool extra_pass, | |
833 | u64 *func_addr, bool *func_addr_fixed) | |
834 | { | |
835 | s16 off = insn->off; | |
836 | s32 imm = insn->imm; | |
837 | u8 *addr; | |
838 | ||
839 | *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL; | |
840 | if (!*func_addr_fixed) { | |
841 | /* Place-holder address till the last pass has collected | |
842 | * all addresses for JITed subprograms in which case we | |
843 | * can pick them up from prog->aux. | |
844 | */ | |
845 | if (!extra_pass) | |
846 | addr = NULL; | |
847 | else if (prog->aux->func && | |
848 | off >= 0 && off < prog->aux->func_cnt) | |
849 | addr = (u8 *)prog->aux->func[off]->bpf_func; | |
850 | else | |
851 | return -EINVAL; | |
852 | } else { | |
853 | /* Address of a BPF helper call. Since part of the core | |
854 | * kernel, it's always at a fixed location. __bpf_call_base | |
855 | * and the helper with imm relative to it are both in core | |
856 | * kernel. | |
857 | */ | |
858 | addr = (u8 *)__bpf_call_base + imm; | |
859 | } | |
860 | ||
861 | *func_addr = (unsigned long)addr; | |
862 | return 0; | |
863 | } | |
864 | ||
4f3446bb DB |
865 | static int bpf_jit_blind_insn(const struct bpf_insn *from, |
866 | const struct bpf_insn *aux, | |
867 | struct bpf_insn *to_buff) | |
868 | { | |
869 | struct bpf_insn *to = to_buff; | |
b7552e1b | 870 | u32 imm_rnd = get_random_int(); |
4f3446bb DB |
871 | s16 off; |
872 | ||
873 | BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG); | |
874 | BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG); | |
875 | ||
9b73bfdd DB |
876 | /* Constraints on AX register: |
877 | * | |
878 | * AX register is inaccessible from user space. It is mapped in | |
879 | * all JITs, and used here for constant blinding rewrites. It is | |
880 | * typically "stateless" meaning its contents are only valid within | |
881 | * the executed instruction, but not across several instructions. | |
882 | * There are a few exceptions however which are further detailed | |
883 | * below. | |
884 | * | |
885 | * Constant blinding is only used by JITs, not in the interpreter. | |
886 | * The interpreter uses AX in some occasions as a local temporary | |
887 | * register e.g. in DIV or MOD instructions. | |
888 | * | |
889 | * In restricted circumstances, the verifier can also use the AX | |
890 | * register for rewrites as long as they do not interfere with | |
891 | * the above cases! | |
892 | */ | |
893 | if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX) | |
894 | goto out; | |
895 | ||
4f3446bb DB |
896 | if (from->imm == 0 && |
897 | (from->code == (BPF_ALU | BPF_MOV | BPF_K) || | |
898 | from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) { | |
899 | *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg); | |
900 | goto out; | |
901 | } | |
902 | ||
903 | switch (from->code) { | |
904 | case BPF_ALU | BPF_ADD | BPF_K: | |
905 | case BPF_ALU | BPF_SUB | BPF_K: | |
906 | case BPF_ALU | BPF_AND | BPF_K: | |
907 | case BPF_ALU | BPF_OR | BPF_K: | |
908 | case BPF_ALU | BPF_XOR | BPF_K: | |
909 | case BPF_ALU | BPF_MUL | BPF_K: | |
910 | case BPF_ALU | BPF_MOV | BPF_K: | |
911 | case BPF_ALU | BPF_DIV | BPF_K: | |
912 | case BPF_ALU | BPF_MOD | BPF_K: | |
913 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
914 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
915 | *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX); | |
916 | break; | |
917 | ||
918 | case BPF_ALU64 | BPF_ADD | BPF_K: | |
919 | case BPF_ALU64 | BPF_SUB | BPF_K: | |
920 | case BPF_ALU64 | BPF_AND | BPF_K: | |
921 | case BPF_ALU64 | BPF_OR | BPF_K: | |
922 | case BPF_ALU64 | BPF_XOR | BPF_K: | |
923 | case BPF_ALU64 | BPF_MUL | BPF_K: | |
924 | case BPF_ALU64 | BPF_MOV | BPF_K: | |
925 | case BPF_ALU64 | BPF_DIV | BPF_K: | |
926 | case BPF_ALU64 | BPF_MOD | BPF_K: | |
927 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
928 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
929 | *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX); | |
930 | break; | |
931 | ||
932 | case BPF_JMP | BPF_JEQ | BPF_K: | |
933 | case BPF_JMP | BPF_JNE | BPF_K: | |
934 | case BPF_JMP | BPF_JGT | BPF_K: | |
92b31a9a | 935 | case BPF_JMP | BPF_JLT | BPF_K: |
4f3446bb | 936 | case BPF_JMP | BPF_JGE | BPF_K: |
92b31a9a | 937 | case BPF_JMP | BPF_JLE | BPF_K: |
4f3446bb | 938 | case BPF_JMP | BPF_JSGT | BPF_K: |
92b31a9a | 939 | case BPF_JMP | BPF_JSLT | BPF_K: |
4f3446bb | 940 | case BPF_JMP | BPF_JSGE | BPF_K: |
92b31a9a | 941 | case BPF_JMP | BPF_JSLE | BPF_K: |
4f3446bb DB |
942 | case BPF_JMP | BPF_JSET | BPF_K: |
943 | /* Accommodate for extra offset in case of a backjump. */ | |
944 | off = from->off; | |
945 | if (off < 0) | |
946 | off -= 2; | |
947 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
948 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
949 | *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off); | |
950 | break; | |
951 | ||
4f3446bb DB |
952 | case BPF_LD | BPF_IMM | BPF_DW: |
953 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm); | |
954 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
955 | *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
956 | *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX); | |
957 | break; | |
958 | case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */ | |
959 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm); | |
960 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
961 | *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX); | |
962 | break; | |
963 | ||
964 | case BPF_ST | BPF_MEM | BPF_DW: | |
965 | case BPF_ST | BPF_MEM | BPF_W: | |
966 | case BPF_ST | BPF_MEM | BPF_H: | |
967 | case BPF_ST | BPF_MEM | BPF_B: | |
968 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
969 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
970 | *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off); | |
971 | break; | |
972 | } | |
973 | out: | |
974 | return to - to_buff; | |
975 | } | |
976 | ||
977 | static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other, | |
978 | gfp_t gfp_extra_flags) | |
979 | { | |
19809c2d | 980 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; |
4f3446bb DB |
981 | struct bpf_prog *fp; |
982 | ||
983 | fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL); | |
984 | if (fp != NULL) { | |
4f3446bb DB |
985 | /* aux->prog still points to the fp_other one, so |
986 | * when promoting the clone to the real program, | |
987 | * this still needs to be adapted. | |
988 | */ | |
989 | memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE); | |
990 | } | |
991 | ||
992 | return fp; | |
993 | } | |
994 | ||
995 | static void bpf_prog_clone_free(struct bpf_prog *fp) | |
996 | { | |
997 | /* aux was stolen by the other clone, so we cannot free | |
998 | * it from this path! It will be freed eventually by the | |
999 | * other program on release. | |
1000 | * | |
1001 | * At this point, we don't need a deferred release since | |
1002 | * clone is guaranteed to not be locked. | |
1003 | */ | |
1004 | fp->aux = NULL; | |
1005 | __bpf_prog_free(fp); | |
1006 | } | |
1007 | ||
1008 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other) | |
1009 | { | |
1010 | /* We have to repoint aux->prog to self, as we don't | |
1011 | * know whether fp here is the clone or the original. | |
1012 | */ | |
1013 | fp->aux->prog = fp; | |
1014 | bpf_prog_clone_free(fp_other); | |
1015 | } | |
1016 | ||
1017 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog) | |
1018 | { | |
1019 | struct bpf_insn insn_buff[16], aux[2]; | |
1020 | struct bpf_prog *clone, *tmp; | |
1021 | int insn_delta, insn_cnt; | |
1022 | struct bpf_insn *insn; | |
1023 | int i, rewritten; | |
1024 | ||
1c2a088a | 1025 | if (!bpf_jit_blinding_enabled(prog) || prog->blinded) |
4f3446bb DB |
1026 | return prog; |
1027 | ||
1028 | clone = bpf_prog_clone_create(prog, GFP_USER); | |
1029 | if (!clone) | |
1030 | return ERR_PTR(-ENOMEM); | |
1031 | ||
1032 | insn_cnt = clone->len; | |
1033 | insn = clone->insnsi; | |
1034 | ||
1035 | for (i = 0; i < insn_cnt; i++, insn++) { | |
1036 | /* We temporarily need to hold the original ld64 insn | |
1037 | * so that we can still access the first part in the | |
1038 | * second blinding run. | |
1039 | */ | |
1040 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
1041 | insn[1].code == 0) | |
1042 | memcpy(aux, insn, sizeof(aux)); | |
1043 | ||
1044 | rewritten = bpf_jit_blind_insn(insn, aux, insn_buff); | |
1045 | if (!rewritten) | |
1046 | continue; | |
1047 | ||
1048 | tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten); | |
1049 | if (!tmp) { | |
1050 | /* Patching may have repointed aux->prog during | |
1051 | * realloc from the original one, so we need to | |
1052 | * fix it up here on error. | |
1053 | */ | |
1054 | bpf_jit_prog_release_other(prog, clone); | |
1055 | return ERR_PTR(-ENOMEM); | |
1056 | } | |
1057 | ||
1058 | clone = tmp; | |
1059 | insn_delta = rewritten - 1; | |
1060 | ||
1061 | /* Walk new program and skip insns we just inserted. */ | |
1062 | insn = clone->insnsi + i + insn_delta; | |
1063 | insn_cnt += insn_delta; | |
1064 | i += insn_delta; | |
1065 | } | |
1066 | ||
1c2a088a | 1067 | clone->blinded = 1; |
4f3446bb DB |
1068 | return clone; |
1069 | } | |
b954d834 | 1070 | #endif /* CONFIG_BPF_JIT */ |
738cbe72 | 1071 | |
f5bffecd AS |
1072 | /* Base function for offset calculation. Needs to go into .text section, |
1073 | * therefore keeping it non-static as well; will also be used by JITs | |
7105e828 DB |
1074 | * anyway later on, so do not let the compiler omit it. This also needs |
1075 | * to go into kallsyms for correlation from e.g. bpftool, so naming | |
1076 | * must not change. | |
f5bffecd AS |
1077 | */ |
1078 | noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
1079 | { | |
1080 | return 0; | |
1081 | } | |
4d9c5c53 | 1082 | EXPORT_SYMBOL_GPL(__bpf_call_base); |
f5bffecd | 1083 | |
5e581dad DB |
1084 | /* All UAPI available opcodes. */ |
1085 | #define BPF_INSN_MAP(INSN_2, INSN_3) \ | |
1086 | /* 32 bit ALU operations. */ \ | |
1087 | /* Register based. */ \ | |
2dc6b100 JW |
1088 | INSN_3(ALU, ADD, X), \ |
1089 | INSN_3(ALU, SUB, X), \ | |
1090 | INSN_3(ALU, AND, X), \ | |
1091 | INSN_3(ALU, OR, X), \ | |
1092 | INSN_3(ALU, LSH, X), \ | |
1093 | INSN_3(ALU, RSH, X), \ | |
1094 | INSN_3(ALU, XOR, X), \ | |
1095 | INSN_3(ALU, MUL, X), \ | |
1096 | INSN_3(ALU, MOV, X), \ | |
1097 | INSN_3(ALU, ARSH, X), \ | |
1098 | INSN_3(ALU, DIV, X), \ | |
1099 | INSN_3(ALU, MOD, X), \ | |
5e581dad DB |
1100 | INSN_2(ALU, NEG), \ |
1101 | INSN_3(ALU, END, TO_BE), \ | |
1102 | INSN_3(ALU, END, TO_LE), \ | |
1103 | /* Immediate based. */ \ | |
2dc6b100 JW |
1104 | INSN_3(ALU, ADD, K), \ |
1105 | INSN_3(ALU, SUB, K), \ | |
1106 | INSN_3(ALU, AND, K), \ | |
1107 | INSN_3(ALU, OR, K), \ | |
1108 | INSN_3(ALU, LSH, K), \ | |
1109 | INSN_3(ALU, RSH, K), \ | |
1110 | INSN_3(ALU, XOR, K), \ | |
1111 | INSN_3(ALU, MUL, K), \ | |
1112 | INSN_3(ALU, MOV, K), \ | |
1113 | INSN_3(ALU, ARSH, K), \ | |
1114 | INSN_3(ALU, DIV, K), \ | |
1115 | INSN_3(ALU, MOD, K), \ | |
5e581dad DB |
1116 | /* 64 bit ALU operations. */ \ |
1117 | /* Register based. */ \ | |
1118 | INSN_3(ALU64, ADD, X), \ | |
1119 | INSN_3(ALU64, SUB, X), \ | |
1120 | INSN_3(ALU64, AND, X), \ | |
1121 | INSN_3(ALU64, OR, X), \ | |
1122 | INSN_3(ALU64, LSH, X), \ | |
1123 | INSN_3(ALU64, RSH, X), \ | |
1124 | INSN_3(ALU64, XOR, X), \ | |
1125 | INSN_3(ALU64, MUL, X), \ | |
1126 | INSN_3(ALU64, MOV, X), \ | |
1127 | INSN_3(ALU64, ARSH, X), \ | |
1128 | INSN_3(ALU64, DIV, X), \ | |
1129 | INSN_3(ALU64, MOD, X), \ | |
1130 | INSN_2(ALU64, NEG), \ | |
1131 | /* Immediate based. */ \ | |
1132 | INSN_3(ALU64, ADD, K), \ | |
1133 | INSN_3(ALU64, SUB, K), \ | |
1134 | INSN_3(ALU64, AND, K), \ | |
1135 | INSN_3(ALU64, OR, K), \ | |
1136 | INSN_3(ALU64, LSH, K), \ | |
1137 | INSN_3(ALU64, RSH, K), \ | |
1138 | INSN_3(ALU64, XOR, K), \ | |
1139 | INSN_3(ALU64, MUL, K), \ | |
1140 | INSN_3(ALU64, MOV, K), \ | |
1141 | INSN_3(ALU64, ARSH, K), \ | |
1142 | INSN_3(ALU64, DIV, K), \ | |
1143 | INSN_3(ALU64, MOD, K), \ | |
1144 | /* Call instruction. */ \ | |
1145 | INSN_2(JMP, CALL), \ | |
1146 | /* Exit instruction. */ \ | |
1147 | INSN_2(JMP, EXIT), \ | |
1148 | /* Jump instructions. */ \ | |
1149 | /* Register based. */ \ | |
1150 | INSN_3(JMP, JEQ, X), \ | |
1151 | INSN_3(JMP, JNE, X), \ | |
1152 | INSN_3(JMP, JGT, X), \ | |
1153 | INSN_3(JMP, JLT, X), \ | |
1154 | INSN_3(JMP, JGE, X), \ | |
1155 | INSN_3(JMP, JLE, X), \ | |
1156 | INSN_3(JMP, JSGT, X), \ | |
1157 | INSN_3(JMP, JSLT, X), \ | |
1158 | INSN_3(JMP, JSGE, X), \ | |
1159 | INSN_3(JMP, JSLE, X), \ | |
1160 | INSN_3(JMP, JSET, X), \ | |
1161 | /* Immediate based. */ \ | |
1162 | INSN_3(JMP, JEQ, K), \ | |
1163 | INSN_3(JMP, JNE, K), \ | |
1164 | INSN_3(JMP, JGT, K), \ | |
1165 | INSN_3(JMP, JLT, K), \ | |
1166 | INSN_3(JMP, JGE, K), \ | |
1167 | INSN_3(JMP, JLE, K), \ | |
1168 | INSN_3(JMP, JSGT, K), \ | |
1169 | INSN_3(JMP, JSLT, K), \ | |
1170 | INSN_3(JMP, JSGE, K), \ | |
1171 | INSN_3(JMP, JSLE, K), \ | |
1172 | INSN_3(JMP, JSET, K), \ | |
1173 | INSN_2(JMP, JA), \ | |
1174 | /* Store instructions. */ \ | |
1175 | /* Register based. */ \ | |
1176 | INSN_3(STX, MEM, B), \ | |
1177 | INSN_3(STX, MEM, H), \ | |
1178 | INSN_3(STX, MEM, W), \ | |
1179 | INSN_3(STX, MEM, DW), \ | |
1180 | INSN_3(STX, XADD, W), \ | |
1181 | INSN_3(STX, XADD, DW), \ | |
1182 | /* Immediate based. */ \ | |
1183 | INSN_3(ST, MEM, B), \ | |
1184 | INSN_3(ST, MEM, H), \ | |
1185 | INSN_3(ST, MEM, W), \ | |
1186 | INSN_3(ST, MEM, DW), \ | |
1187 | /* Load instructions. */ \ | |
1188 | /* Register based. */ \ | |
1189 | INSN_3(LDX, MEM, B), \ | |
1190 | INSN_3(LDX, MEM, H), \ | |
1191 | INSN_3(LDX, MEM, W), \ | |
1192 | INSN_3(LDX, MEM, DW), \ | |
1193 | /* Immediate based. */ \ | |
e0cea7ce | 1194 | INSN_3(LD, IMM, DW) |
5e581dad DB |
1195 | |
1196 | bool bpf_opcode_in_insntable(u8 code) | |
1197 | { | |
1198 | #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true | |
1199 | #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true | |
1200 | static const bool public_insntable[256] = { | |
1201 | [0 ... 255] = false, | |
1202 | /* Now overwrite non-defaults ... */ | |
1203 | BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL), | |
e0cea7ce DB |
1204 | /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */ |
1205 | [BPF_LD | BPF_ABS | BPF_B] = true, | |
1206 | [BPF_LD | BPF_ABS | BPF_H] = true, | |
1207 | [BPF_LD | BPF_ABS | BPF_W] = true, | |
1208 | [BPF_LD | BPF_IND | BPF_B] = true, | |
1209 | [BPF_LD | BPF_IND | BPF_H] = true, | |
1210 | [BPF_LD | BPF_IND | BPF_W] = true, | |
5e581dad DB |
1211 | }; |
1212 | #undef BPF_INSN_3_TBL | |
1213 | #undef BPF_INSN_2_TBL | |
1214 | return public_insntable[code]; | |
1215 | } | |
1216 | ||
290af866 | 1217 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
f5bffecd | 1218 | /** |
7ae457c1 AS |
1219 | * __bpf_prog_run - run eBPF program on a given context |
1220 | * @ctx: is the data we are operating on | |
1221 | * @insn: is the array of eBPF instructions | |
f5bffecd | 1222 | * |
7ae457c1 | 1223 | * Decode and execute eBPF instructions. |
f5bffecd | 1224 | */ |
1ea47e01 | 1225 | static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack) |
f5bffecd | 1226 | { |
5e581dad DB |
1227 | #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y |
1228 | #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z | |
f5bffecd AS |
1229 | static const void *jumptable[256] = { |
1230 | [0 ... 255] = &&default_label, | |
1231 | /* Now overwrite non-defaults ... */ | |
5e581dad DB |
1232 | BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL), |
1233 | /* Non-UAPI available opcodes. */ | |
1ea47e01 | 1234 | [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS, |
71189fa9 | 1235 | [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL, |
f5bffecd | 1236 | }; |
5e581dad DB |
1237 | #undef BPF_INSN_3_LBL |
1238 | #undef BPF_INSN_2_LBL | |
04fd61ab | 1239 | u32 tail_call_cnt = 0; |
f5bffecd AS |
1240 | |
1241 | #define CONT ({ insn++; goto select_insn; }) | |
1242 | #define CONT_JMP ({ insn++; goto select_insn; }) | |
1243 | ||
f5bffecd AS |
1244 | select_insn: |
1245 | goto *jumptable[insn->code]; | |
1246 | ||
1247 | /* ALU */ | |
1248 | #define ALU(OPCODE, OP) \ | |
1249 | ALU64_##OPCODE##_X: \ | |
1250 | DST = DST OP SRC; \ | |
1251 | CONT; \ | |
1252 | ALU_##OPCODE##_X: \ | |
1253 | DST = (u32) DST OP (u32) SRC; \ | |
1254 | CONT; \ | |
1255 | ALU64_##OPCODE##_K: \ | |
1256 | DST = DST OP IMM; \ | |
1257 | CONT; \ | |
1258 | ALU_##OPCODE##_K: \ | |
1259 | DST = (u32) DST OP (u32) IMM; \ | |
1260 | CONT; | |
1261 | ||
1262 | ALU(ADD, +) | |
1263 | ALU(SUB, -) | |
1264 | ALU(AND, &) | |
1265 | ALU(OR, |) | |
1266 | ALU(LSH, <<) | |
1267 | ALU(RSH, >>) | |
1268 | ALU(XOR, ^) | |
1269 | ALU(MUL, *) | |
1270 | #undef ALU | |
1271 | ALU_NEG: | |
1272 | DST = (u32) -DST; | |
1273 | CONT; | |
1274 | ALU64_NEG: | |
1275 | DST = -DST; | |
1276 | CONT; | |
1277 | ALU_MOV_X: | |
1278 | DST = (u32) SRC; | |
1279 | CONT; | |
1280 | ALU_MOV_K: | |
1281 | DST = (u32) IMM; | |
1282 | CONT; | |
1283 | ALU64_MOV_X: | |
1284 | DST = SRC; | |
1285 | CONT; | |
1286 | ALU64_MOV_K: | |
1287 | DST = IMM; | |
1288 | CONT; | |
02ab695b AS |
1289 | LD_IMM_DW: |
1290 | DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32; | |
1291 | insn++; | |
1292 | CONT; | |
2dc6b100 JW |
1293 | ALU_ARSH_X: |
1294 | DST = (u64) (u32) ((*(s32 *) &DST) >> SRC); | |
1295 | CONT; | |
1296 | ALU_ARSH_K: | |
1297 | DST = (u64) (u32) ((*(s32 *) &DST) >> IMM); | |
1298 | CONT; | |
f5bffecd AS |
1299 | ALU64_ARSH_X: |
1300 | (*(s64 *) &DST) >>= SRC; | |
1301 | CONT; | |
1302 | ALU64_ARSH_K: | |
1303 | (*(s64 *) &DST) >>= IMM; | |
1304 | CONT; | |
1305 | ALU64_MOD_X: | |
144cd91c DB |
1306 | div64_u64_rem(DST, SRC, &AX); |
1307 | DST = AX; | |
f5bffecd AS |
1308 | CONT; |
1309 | ALU_MOD_X: | |
144cd91c DB |
1310 | AX = (u32) DST; |
1311 | DST = do_div(AX, (u32) SRC); | |
f5bffecd AS |
1312 | CONT; |
1313 | ALU64_MOD_K: | |
144cd91c DB |
1314 | div64_u64_rem(DST, IMM, &AX); |
1315 | DST = AX; | |
f5bffecd AS |
1316 | CONT; |
1317 | ALU_MOD_K: | |
144cd91c DB |
1318 | AX = (u32) DST; |
1319 | DST = do_div(AX, (u32) IMM); | |
f5bffecd AS |
1320 | CONT; |
1321 | ALU64_DIV_X: | |
876a7ae6 | 1322 | DST = div64_u64(DST, SRC); |
f5bffecd AS |
1323 | CONT; |
1324 | ALU_DIV_X: | |
144cd91c DB |
1325 | AX = (u32) DST; |
1326 | do_div(AX, (u32) SRC); | |
1327 | DST = (u32) AX; | |
f5bffecd AS |
1328 | CONT; |
1329 | ALU64_DIV_K: | |
876a7ae6 | 1330 | DST = div64_u64(DST, IMM); |
f5bffecd AS |
1331 | CONT; |
1332 | ALU_DIV_K: | |
144cd91c DB |
1333 | AX = (u32) DST; |
1334 | do_div(AX, (u32) IMM); | |
1335 | DST = (u32) AX; | |
f5bffecd AS |
1336 | CONT; |
1337 | ALU_END_TO_BE: | |
1338 | switch (IMM) { | |
1339 | case 16: | |
1340 | DST = (__force u16) cpu_to_be16(DST); | |
1341 | break; | |
1342 | case 32: | |
1343 | DST = (__force u32) cpu_to_be32(DST); | |
1344 | break; | |
1345 | case 64: | |
1346 | DST = (__force u64) cpu_to_be64(DST); | |
1347 | break; | |
1348 | } | |
1349 | CONT; | |
1350 | ALU_END_TO_LE: | |
1351 | switch (IMM) { | |
1352 | case 16: | |
1353 | DST = (__force u16) cpu_to_le16(DST); | |
1354 | break; | |
1355 | case 32: | |
1356 | DST = (__force u32) cpu_to_le32(DST); | |
1357 | break; | |
1358 | case 64: | |
1359 | DST = (__force u64) cpu_to_le64(DST); | |
1360 | break; | |
1361 | } | |
1362 | CONT; | |
1363 | ||
1364 | /* CALL */ | |
1365 | JMP_CALL: | |
1366 | /* Function call scratches BPF_R1-BPF_R5 registers, | |
1367 | * preserves BPF_R6-BPF_R9, and stores return value | |
1368 | * into BPF_R0. | |
1369 | */ | |
1370 | BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3, | |
1371 | BPF_R4, BPF_R5); | |
1372 | CONT; | |
1373 | ||
1ea47e01 AS |
1374 | JMP_CALL_ARGS: |
1375 | BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2, | |
1376 | BPF_R3, BPF_R4, | |
1377 | BPF_R5, | |
1378 | insn + insn->off + 1); | |
1379 | CONT; | |
1380 | ||
04fd61ab AS |
1381 | JMP_TAIL_CALL: { |
1382 | struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2; | |
1383 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
1384 | struct bpf_prog *prog; | |
90caccdd | 1385 | u32 index = BPF_R3; |
04fd61ab AS |
1386 | |
1387 | if (unlikely(index >= array->map.max_entries)) | |
1388 | goto out; | |
04fd61ab AS |
1389 | if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) |
1390 | goto out; | |
1391 | ||
1392 | tail_call_cnt++; | |
1393 | ||
2a36f0b9 | 1394 | prog = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 1395 | if (!prog) |
04fd61ab AS |
1396 | goto out; |
1397 | ||
c4675f93 DB |
1398 | /* ARG1 at this point is guaranteed to point to CTX from |
1399 | * the verifier side due to the fact that the tail call is | |
1400 | * handeled like a helper, that is, bpf_tail_call_proto, | |
1401 | * where arg1_type is ARG_PTR_TO_CTX. | |
1402 | */ | |
04fd61ab AS |
1403 | insn = prog->insnsi; |
1404 | goto select_insn; | |
1405 | out: | |
1406 | CONT; | |
1407 | } | |
f5bffecd AS |
1408 | /* JMP */ |
1409 | JMP_JA: | |
1410 | insn += insn->off; | |
1411 | CONT; | |
1412 | JMP_JEQ_X: | |
1413 | if (DST == SRC) { | |
1414 | insn += insn->off; | |
1415 | CONT_JMP; | |
1416 | } | |
1417 | CONT; | |
1418 | JMP_JEQ_K: | |
1419 | if (DST == IMM) { | |
1420 | insn += insn->off; | |
1421 | CONT_JMP; | |
1422 | } | |
1423 | CONT; | |
1424 | JMP_JNE_X: | |
1425 | if (DST != SRC) { | |
1426 | insn += insn->off; | |
1427 | CONT_JMP; | |
1428 | } | |
1429 | CONT; | |
1430 | JMP_JNE_K: | |
1431 | if (DST != IMM) { | |
1432 | insn += insn->off; | |
1433 | CONT_JMP; | |
1434 | } | |
1435 | CONT; | |
1436 | JMP_JGT_X: | |
1437 | if (DST > SRC) { | |
1438 | insn += insn->off; | |
1439 | CONT_JMP; | |
1440 | } | |
1441 | CONT; | |
1442 | JMP_JGT_K: | |
1443 | if (DST > IMM) { | |
1444 | insn += insn->off; | |
1445 | CONT_JMP; | |
1446 | } | |
1447 | CONT; | |
92b31a9a DB |
1448 | JMP_JLT_X: |
1449 | if (DST < SRC) { | |
1450 | insn += insn->off; | |
1451 | CONT_JMP; | |
1452 | } | |
1453 | CONT; | |
1454 | JMP_JLT_K: | |
1455 | if (DST < IMM) { | |
1456 | insn += insn->off; | |
1457 | CONT_JMP; | |
1458 | } | |
1459 | CONT; | |
f5bffecd AS |
1460 | JMP_JGE_X: |
1461 | if (DST >= SRC) { | |
1462 | insn += insn->off; | |
1463 | CONT_JMP; | |
1464 | } | |
1465 | CONT; | |
1466 | JMP_JGE_K: | |
1467 | if (DST >= IMM) { | |
1468 | insn += insn->off; | |
1469 | CONT_JMP; | |
1470 | } | |
1471 | CONT; | |
92b31a9a DB |
1472 | JMP_JLE_X: |
1473 | if (DST <= SRC) { | |
1474 | insn += insn->off; | |
1475 | CONT_JMP; | |
1476 | } | |
1477 | CONT; | |
1478 | JMP_JLE_K: | |
1479 | if (DST <= IMM) { | |
1480 | insn += insn->off; | |
1481 | CONT_JMP; | |
1482 | } | |
1483 | CONT; | |
f5bffecd AS |
1484 | JMP_JSGT_X: |
1485 | if (((s64) DST) > ((s64) SRC)) { | |
1486 | insn += insn->off; | |
1487 | CONT_JMP; | |
1488 | } | |
1489 | CONT; | |
1490 | JMP_JSGT_K: | |
1491 | if (((s64) DST) > ((s64) IMM)) { | |
1492 | insn += insn->off; | |
1493 | CONT_JMP; | |
1494 | } | |
1495 | CONT; | |
92b31a9a DB |
1496 | JMP_JSLT_X: |
1497 | if (((s64) DST) < ((s64) SRC)) { | |
1498 | insn += insn->off; | |
1499 | CONT_JMP; | |
1500 | } | |
1501 | CONT; | |
1502 | JMP_JSLT_K: | |
1503 | if (((s64) DST) < ((s64) IMM)) { | |
1504 | insn += insn->off; | |
1505 | CONT_JMP; | |
1506 | } | |
1507 | CONT; | |
f5bffecd AS |
1508 | JMP_JSGE_X: |
1509 | if (((s64) DST) >= ((s64) SRC)) { | |
1510 | insn += insn->off; | |
1511 | CONT_JMP; | |
1512 | } | |
1513 | CONT; | |
1514 | JMP_JSGE_K: | |
1515 | if (((s64) DST) >= ((s64) IMM)) { | |
1516 | insn += insn->off; | |
1517 | CONT_JMP; | |
1518 | } | |
1519 | CONT; | |
92b31a9a DB |
1520 | JMP_JSLE_X: |
1521 | if (((s64) DST) <= ((s64) SRC)) { | |
1522 | insn += insn->off; | |
1523 | CONT_JMP; | |
1524 | } | |
1525 | CONT; | |
1526 | JMP_JSLE_K: | |
1527 | if (((s64) DST) <= ((s64) IMM)) { | |
1528 | insn += insn->off; | |
1529 | CONT_JMP; | |
1530 | } | |
1531 | CONT; | |
f5bffecd AS |
1532 | JMP_JSET_X: |
1533 | if (DST & SRC) { | |
1534 | insn += insn->off; | |
1535 | CONT_JMP; | |
1536 | } | |
1537 | CONT; | |
1538 | JMP_JSET_K: | |
1539 | if (DST & IMM) { | |
1540 | insn += insn->off; | |
1541 | CONT_JMP; | |
1542 | } | |
1543 | CONT; | |
1544 | JMP_EXIT: | |
1545 | return BPF_R0; | |
1546 | ||
1547 | /* STX and ST and LDX*/ | |
1548 | #define LDST(SIZEOP, SIZE) \ | |
1549 | STX_MEM_##SIZEOP: \ | |
1550 | *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \ | |
1551 | CONT; \ | |
1552 | ST_MEM_##SIZEOP: \ | |
1553 | *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \ | |
1554 | CONT; \ | |
1555 | LDX_MEM_##SIZEOP: \ | |
1556 | DST = *(SIZE *)(unsigned long) (SRC + insn->off); \ | |
1557 | CONT; | |
1558 | ||
1559 | LDST(B, u8) | |
1560 | LDST(H, u16) | |
1561 | LDST(W, u32) | |
1562 | LDST(DW, u64) | |
1563 | #undef LDST | |
1564 | STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */ | |
1565 | atomic_add((u32) SRC, (atomic_t *)(unsigned long) | |
1566 | (DST + insn->off)); | |
1567 | CONT; | |
1568 | STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */ | |
1569 | atomic64_add((u64) SRC, (atomic64_t *)(unsigned long) | |
1570 | (DST + insn->off)); | |
1571 | CONT; | |
f5bffecd AS |
1572 | |
1573 | default_label: | |
5e581dad DB |
1574 | /* If we ever reach this, we have a bug somewhere. Die hard here |
1575 | * instead of just returning 0; we could be somewhere in a subprog, | |
1576 | * so execution could continue otherwise which we do /not/ want. | |
1577 | * | |
1578 | * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable(). | |
1579 | */ | |
1580 | pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code); | |
1581 | BUG_ON(1); | |
f5bffecd AS |
1582 | return 0; |
1583 | } | |
f696b8f4 AS |
1584 | STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */ |
1585 | ||
b870aa90 AS |
1586 | #define PROG_NAME(stack_size) __bpf_prog_run##stack_size |
1587 | #define DEFINE_BPF_PROG_RUN(stack_size) \ | |
1588 | static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \ | |
1589 | { \ | |
1590 | u64 stack[stack_size / sizeof(u64)]; \ | |
144cd91c | 1591 | u64 regs[MAX_BPF_EXT_REG]; \ |
b870aa90 AS |
1592 | \ |
1593 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ | |
1594 | ARG1 = (u64) (unsigned long) ctx; \ | |
1595 | return ___bpf_prog_run(regs, insn, stack); \ | |
f696b8f4 | 1596 | } |
f5bffecd | 1597 | |
1ea47e01 AS |
1598 | #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size |
1599 | #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \ | |
1600 | static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \ | |
1601 | const struct bpf_insn *insn) \ | |
1602 | { \ | |
1603 | u64 stack[stack_size / sizeof(u64)]; \ | |
144cd91c | 1604 | u64 regs[MAX_BPF_EXT_REG]; \ |
1ea47e01 AS |
1605 | \ |
1606 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ | |
1607 | BPF_R1 = r1; \ | |
1608 | BPF_R2 = r2; \ | |
1609 | BPF_R3 = r3; \ | |
1610 | BPF_R4 = r4; \ | |
1611 | BPF_R5 = r5; \ | |
1612 | return ___bpf_prog_run(regs, insn, stack); \ | |
1613 | } | |
1614 | ||
b870aa90 AS |
1615 | #define EVAL1(FN, X) FN(X) |
1616 | #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y) | |
1617 | #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y) | |
1618 | #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y) | |
1619 | #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y) | |
1620 | #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y) | |
1621 | ||
1622 | EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192); | |
1623 | EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384); | |
1624 | EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512); | |
1625 | ||
1ea47e01 AS |
1626 | EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192); |
1627 | EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384); | |
1628 | EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512); | |
1629 | ||
b870aa90 AS |
1630 | #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size), |
1631 | ||
1632 | static unsigned int (*interpreters[])(const void *ctx, | |
1633 | const struct bpf_insn *insn) = { | |
1634 | EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) | |
1635 | EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) | |
1636 | EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) | |
1637 | }; | |
1ea47e01 AS |
1638 | #undef PROG_NAME_LIST |
1639 | #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size), | |
1640 | static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, | |
1641 | const struct bpf_insn *insn) = { | |
1642 | EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) | |
1643 | EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) | |
1644 | EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) | |
1645 | }; | |
1646 | #undef PROG_NAME_LIST | |
1647 | ||
1648 | void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth) | |
1649 | { | |
1650 | stack_depth = max_t(u32, stack_depth, 1); | |
1651 | insn->off = (s16) insn->imm; | |
1652 | insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] - | |
1653 | __bpf_call_base_args; | |
1654 | insn->code = BPF_JMP | BPF_CALL_ARGS; | |
1655 | } | |
b870aa90 | 1656 | |
290af866 | 1657 | #else |
fa9dd599 DB |
1658 | static unsigned int __bpf_prog_ret0_warn(const void *ctx, |
1659 | const struct bpf_insn *insn) | |
290af866 | 1660 | { |
fa9dd599 DB |
1661 | /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON |
1662 | * is not working properly, so warn about it! | |
1663 | */ | |
1664 | WARN_ON_ONCE(1); | |
290af866 AS |
1665 | return 0; |
1666 | } | |
1667 | #endif | |
1668 | ||
3324b584 DB |
1669 | bool bpf_prog_array_compatible(struct bpf_array *array, |
1670 | const struct bpf_prog *fp) | |
04fd61ab | 1671 | { |
9802d865 JB |
1672 | if (fp->kprobe_override) |
1673 | return false; | |
1674 | ||
3324b584 DB |
1675 | if (!array->owner_prog_type) { |
1676 | /* There's no owner yet where we could check for | |
1677 | * compatibility. | |
1678 | */ | |
04fd61ab AS |
1679 | array->owner_prog_type = fp->type; |
1680 | array->owner_jited = fp->jited; | |
3324b584 DB |
1681 | |
1682 | return true; | |
04fd61ab | 1683 | } |
3324b584 DB |
1684 | |
1685 | return array->owner_prog_type == fp->type && | |
1686 | array->owner_jited == fp->jited; | |
04fd61ab AS |
1687 | } |
1688 | ||
3324b584 | 1689 | static int bpf_check_tail_call(const struct bpf_prog *fp) |
04fd61ab AS |
1690 | { |
1691 | struct bpf_prog_aux *aux = fp->aux; | |
1692 | int i; | |
1693 | ||
1694 | for (i = 0; i < aux->used_map_cnt; i++) { | |
3324b584 | 1695 | struct bpf_map *map = aux->used_maps[i]; |
04fd61ab | 1696 | struct bpf_array *array; |
04fd61ab | 1697 | |
04fd61ab AS |
1698 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) |
1699 | continue; | |
3324b584 | 1700 | |
04fd61ab AS |
1701 | array = container_of(map, struct bpf_array, map); |
1702 | if (!bpf_prog_array_compatible(array, fp)) | |
1703 | return -EINVAL; | |
1704 | } | |
1705 | ||
1706 | return 0; | |
1707 | } | |
1708 | ||
9facc336 DB |
1709 | static void bpf_prog_select_func(struct bpf_prog *fp) |
1710 | { | |
1711 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1712 | u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1); | |
1713 | ||
1714 | fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1]; | |
1715 | #else | |
1716 | fp->bpf_func = __bpf_prog_ret0_warn; | |
1717 | #endif | |
1718 | } | |
1719 | ||
f5bffecd | 1720 | /** |
3324b584 | 1721 | * bpf_prog_select_runtime - select exec runtime for BPF program |
7ae457c1 | 1722 | * @fp: bpf_prog populated with internal BPF program |
d1c55ab5 | 1723 | * @err: pointer to error variable |
f5bffecd | 1724 | * |
3324b584 DB |
1725 | * Try to JIT eBPF program, if JIT is not available, use interpreter. |
1726 | * The BPF program will be executed via BPF_PROG_RUN() macro. | |
f5bffecd | 1727 | */ |
d1c55ab5 | 1728 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err) |
f5bffecd | 1729 | { |
9facc336 DB |
1730 | /* In case of BPF to BPF calls, verifier did all the prep |
1731 | * work with regards to JITing, etc. | |
1732 | */ | |
1733 | if (fp->bpf_func) | |
1734 | goto finalize; | |
8007e40a | 1735 | |
9facc336 | 1736 | bpf_prog_select_func(fp); |
f5bffecd | 1737 | |
d1c55ab5 DB |
1738 | /* eBPF JITs can rewrite the program in case constant |
1739 | * blinding is active. However, in case of error during | |
1740 | * blinding, bpf_int_jit_compile() must always return a | |
1741 | * valid program, which in this case would simply not | |
1742 | * be JITed, but falls back to the interpreter. | |
1743 | */ | |
ab3f0063 | 1744 | if (!bpf_prog_is_dev_bound(fp->aux)) { |
c454a46b MKL |
1745 | *err = bpf_prog_alloc_jited_linfo(fp); |
1746 | if (*err) | |
1747 | return fp; | |
1748 | ||
ab3f0063 | 1749 | fp = bpf_int_jit_compile(fp); |
290af866 | 1750 | if (!fp->jited) { |
c454a46b MKL |
1751 | bpf_prog_free_jited_linfo(fp); |
1752 | #ifdef CONFIG_BPF_JIT_ALWAYS_ON | |
290af866 AS |
1753 | *err = -ENOTSUPP; |
1754 | return fp; | |
290af866 | 1755 | #endif |
c454a46b MKL |
1756 | } else { |
1757 | bpf_prog_free_unused_jited_linfo(fp); | |
1758 | } | |
ab3f0063 JK |
1759 | } else { |
1760 | *err = bpf_prog_offload_compile(fp); | |
1761 | if (*err) | |
1762 | return fp; | |
1763 | } | |
9facc336 DB |
1764 | |
1765 | finalize: | |
60a3b225 | 1766 | bpf_prog_lock_ro(fp); |
04fd61ab | 1767 | |
3324b584 DB |
1768 | /* The tail call compatibility check can only be done at |
1769 | * this late stage as we need to determine, if we deal | |
1770 | * with JITed or non JITed program concatenations and not | |
1771 | * all eBPF JITs might immediately support all features. | |
1772 | */ | |
d1c55ab5 | 1773 | *err = bpf_check_tail_call(fp); |
85782e03 | 1774 | |
d1c55ab5 | 1775 | return fp; |
f5bffecd | 1776 | } |
7ae457c1 | 1777 | EXPORT_SYMBOL_GPL(bpf_prog_select_runtime); |
f5bffecd | 1778 | |
e87c6bc3 YS |
1779 | static unsigned int __bpf_prog_ret1(const void *ctx, |
1780 | const struct bpf_insn *insn) | |
1781 | { | |
1782 | return 1; | |
1783 | } | |
1784 | ||
1785 | static struct bpf_prog_dummy { | |
1786 | struct bpf_prog prog; | |
1787 | } dummy_bpf_prog = { | |
1788 | .prog = { | |
1789 | .bpf_func = __bpf_prog_ret1, | |
1790 | }, | |
1791 | }; | |
1792 | ||
324bda9e AS |
1793 | /* to avoid allocating empty bpf_prog_array for cgroups that |
1794 | * don't have bpf program attached use one global 'empty_prog_array' | |
1795 | * It will not be modified the caller of bpf_prog_array_alloc() | |
1796 | * (since caller requested prog_cnt == 0) | |
1797 | * that pointer should be 'freed' by bpf_prog_array_free() | |
1798 | */ | |
1799 | static struct { | |
1800 | struct bpf_prog_array hdr; | |
1801 | struct bpf_prog *null_prog; | |
1802 | } empty_prog_array = { | |
1803 | .null_prog = NULL, | |
1804 | }; | |
1805 | ||
d29ab6e1 | 1806 | struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags) |
324bda9e AS |
1807 | { |
1808 | if (prog_cnt) | |
1809 | return kzalloc(sizeof(struct bpf_prog_array) + | |
394e40a2 RG |
1810 | sizeof(struct bpf_prog_array_item) * |
1811 | (prog_cnt + 1), | |
324bda9e AS |
1812 | flags); |
1813 | ||
1814 | return &empty_prog_array.hdr; | |
1815 | } | |
1816 | ||
1817 | void bpf_prog_array_free(struct bpf_prog_array __rcu *progs) | |
1818 | { | |
1819 | if (!progs || | |
1820 | progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr) | |
1821 | return; | |
1822 | kfree_rcu(progs, rcu); | |
1823 | } | |
1824 | ||
394e40a2 | 1825 | int bpf_prog_array_length(struct bpf_prog_array __rcu *array) |
468e2f64 | 1826 | { |
394e40a2 | 1827 | struct bpf_prog_array_item *item; |
468e2f64 AS |
1828 | u32 cnt = 0; |
1829 | ||
1830 | rcu_read_lock(); | |
394e40a2 RG |
1831 | item = rcu_dereference(array)->items; |
1832 | for (; item->prog; item++) | |
1833 | if (item->prog != &dummy_bpf_prog.prog) | |
c8c088ba | 1834 | cnt++; |
468e2f64 AS |
1835 | rcu_read_unlock(); |
1836 | return cnt; | |
1837 | } | |
1838 | ||
394e40a2 RG |
1839 | |
1840 | static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array, | |
3a38bb98 YS |
1841 | u32 *prog_ids, |
1842 | u32 request_cnt) | |
1843 | { | |
394e40a2 | 1844 | struct bpf_prog_array_item *item; |
3a38bb98 YS |
1845 | int i = 0; |
1846 | ||
965931e3 | 1847 | item = rcu_dereference_check(array, 1)->items; |
394e40a2 RG |
1848 | for (; item->prog; item++) { |
1849 | if (item->prog == &dummy_bpf_prog.prog) | |
3a38bb98 | 1850 | continue; |
394e40a2 | 1851 | prog_ids[i] = item->prog->aux->id; |
3a38bb98 | 1852 | if (++i == request_cnt) { |
394e40a2 | 1853 | item++; |
3a38bb98 YS |
1854 | break; |
1855 | } | |
1856 | } | |
1857 | ||
394e40a2 | 1858 | return !!(item->prog); |
3a38bb98 YS |
1859 | } |
1860 | ||
394e40a2 | 1861 | int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array, |
468e2f64 AS |
1862 | __u32 __user *prog_ids, u32 cnt) |
1863 | { | |
0911287c | 1864 | unsigned long err = 0; |
0911287c | 1865 | bool nospc; |
3a38bb98 | 1866 | u32 *ids; |
0911287c AS |
1867 | |
1868 | /* users of this function are doing: | |
1869 | * cnt = bpf_prog_array_length(); | |
1870 | * if (cnt > 0) | |
1871 | * bpf_prog_array_copy_to_user(..., cnt); | |
1872 | * so below kcalloc doesn't need extra cnt > 0 check, but | |
1873 | * bpf_prog_array_length() releases rcu lock and | |
1874 | * prog array could have been swapped with empty or larger array, | |
1875 | * so always copy 'cnt' prog_ids to the user. | |
1876 | * In a rare race the user will see zero prog_ids | |
1877 | */ | |
9c481b90 | 1878 | ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN); |
0911287c AS |
1879 | if (!ids) |
1880 | return -ENOMEM; | |
468e2f64 | 1881 | rcu_read_lock(); |
394e40a2 | 1882 | nospc = bpf_prog_array_copy_core(array, ids, cnt); |
468e2f64 | 1883 | rcu_read_unlock(); |
0911287c AS |
1884 | err = copy_to_user(prog_ids, ids, cnt * sizeof(u32)); |
1885 | kfree(ids); | |
1886 | if (err) | |
1887 | return -EFAULT; | |
1888 | if (nospc) | |
468e2f64 AS |
1889 | return -ENOSPC; |
1890 | return 0; | |
1891 | } | |
1892 | ||
394e40a2 | 1893 | void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array, |
e87c6bc3 YS |
1894 | struct bpf_prog *old_prog) |
1895 | { | |
394e40a2 | 1896 | struct bpf_prog_array_item *item = array->items; |
e87c6bc3 | 1897 | |
394e40a2 RG |
1898 | for (; item->prog; item++) |
1899 | if (item->prog == old_prog) { | |
1900 | WRITE_ONCE(item->prog, &dummy_bpf_prog.prog); | |
e87c6bc3 YS |
1901 | break; |
1902 | } | |
1903 | } | |
1904 | ||
1905 | int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array, | |
1906 | struct bpf_prog *exclude_prog, | |
1907 | struct bpf_prog *include_prog, | |
1908 | struct bpf_prog_array **new_array) | |
1909 | { | |
1910 | int new_prog_cnt, carry_prog_cnt = 0; | |
394e40a2 | 1911 | struct bpf_prog_array_item *existing; |
e87c6bc3 | 1912 | struct bpf_prog_array *array; |
170a7e3e | 1913 | bool found_exclude = false; |
e87c6bc3 YS |
1914 | int new_prog_idx = 0; |
1915 | ||
1916 | /* Figure out how many existing progs we need to carry over to | |
1917 | * the new array. | |
1918 | */ | |
1919 | if (old_array) { | |
394e40a2 RG |
1920 | existing = old_array->items; |
1921 | for (; existing->prog; existing++) { | |
1922 | if (existing->prog == exclude_prog) { | |
170a7e3e SY |
1923 | found_exclude = true; |
1924 | continue; | |
1925 | } | |
394e40a2 | 1926 | if (existing->prog != &dummy_bpf_prog.prog) |
e87c6bc3 | 1927 | carry_prog_cnt++; |
394e40a2 | 1928 | if (existing->prog == include_prog) |
e87c6bc3 YS |
1929 | return -EEXIST; |
1930 | } | |
1931 | } | |
1932 | ||
170a7e3e SY |
1933 | if (exclude_prog && !found_exclude) |
1934 | return -ENOENT; | |
1935 | ||
e87c6bc3 YS |
1936 | /* How many progs (not NULL) will be in the new array? */ |
1937 | new_prog_cnt = carry_prog_cnt; | |
1938 | if (include_prog) | |
1939 | new_prog_cnt += 1; | |
1940 | ||
1941 | /* Do we have any prog (not NULL) in the new array? */ | |
1942 | if (!new_prog_cnt) { | |
1943 | *new_array = NULL; | |
1944 | return 0; | |
1945 | } | |
1946 | ||
1947 | /* +1 as the end of prog_array is marked with NULL */ | |
1948 | array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL); | |
1949 | if (!array) | |
1950 | return -ENOMEM; | |
1951 | ||
1952 | /* Fill in the new prog array */ | |
1953 | if (carry_prog_cnt) { | |
394e40a2 RG |
1954 | existing = old_array->items; |
1955 | for (; existing->prog; existing++) | |
1956 | if (existing->prog != exclude_prog && | |
1957 | existing->prog != &dummy_bpf_prog.prog) { | |
1958 | array->items[new_prog_idx++].prog = | |
1959 | existing->prog; | |
1960 | } | |
e87c6bc3 YS |
1961 | } |
1962 | if (include_prog) | |
394e40a2 RG |
1963 | array->items[new_prog_idx++].prog = include_prog; |
1964 | array->items[new_prog_idx].prog = NULL; | |
e87c6bc3 YS |
1965 | *new_array = array; |
1966 | return 0; | |
1967 | } | |
1968 | ||
f371b304 | 1969 | int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array, |
3a38bb98 YS |
1970 | u32 *prog_ids, u32 request_cnt, |
1971 | u32 *prog_cnt) | |
f371b304 YS |
1972 | { |
1973 | u32 cnt = 0; | |
1974 | ||
1975 | if (array) | |
1976 | cnt = bpf_prog_array_length(array); | |
1977 | ||
3a38bb98 | 1978 | *prog_cnt = cnt; |
f371b304 YS |
1979 | |
1980 | /* return early if user requested only program count or nothing to copy */ | |
1981 | if (!request_cnt || !cnt) | |
1982 | return 0; | |
1983 | ||
3a38bb98 | 1984 | /* this function is called under trace/bpf_trace.c: bpf_event_mutex */ |
394e40a2 | 1985 | return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC |
3a38bb98 | 1986 | : 0; |
f371b304 YS |
1987 | } |
1988 | ||
60a3b225 DB |
1989 | static void bpf_prog_free_deferred(struct work_struct *work) |
1990 | { | |
09756af4 | 1991 | struct bpf_prog_aux *aux; |
1c2a088a | 1992 | int i; |
60a3b225 | 1993 | |
09756af4 | 1994 | aux = container_of(work, struct bpf_prog_aux, work); |
ab3f0063 JK |
1995 | if (bpf_prog_is_dev_bound(aux)) |
1996 | bpf_prog_offload_destroy(aux->prog); | |
c195651e YS |
1997 | #ifdef CONFIG_PERF_EVENTS |
1998 | if (aux->prog->has_callchain_buf) | |
1999 | put_callchain_buffers(); | |
2000 | #endif | |
1c2a088a AS |
2001 | for (i = 0; i < aux->func_cnt; i++) |
2002 | bpf_jit_free(aux->func[i]); | |
2003 | if (aux->func_cnt) { | |
2004 | kfree(aux->func); | |
2005 | bpf_prog_unlock_free(aux->prog); | |
2006 | } else { | |
2007 | bpf_jit_free(aux->prog); | |
2008 | } | |
60a3b225 DB |
2009 | } |
2010 | ||
2011 | /* Free internal BPF program */ | |
7ae457c1 | 2012 | void bpf_prog_free(struct bpf_prog *fp) |
f5bffecd | 2013 | { |
09756af4 | 2014 | struct bpf_prog_aux *aux = fp->aux; |
60a3b225 | 2015 | |
09756af4 | 2016 | INIT_WORK(&aux->work, bpf_prog_free_deferred); |
09756af4 | 2017 | schedule_work(&aux->work); |
f5bffecd | 2018 | } |
7ae457c1 | 2019 | EXPORT_SYMBOL_GPL(bpf_prog_free); |
f89b7755 | 2020 | |
3ad00405 DB |
2021 | /* RNG for unpriviledged user space with separated state from prandom_u32(). */ |
2022 | static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state); | |
2023 | ||
2024 | void bpf_user_rnd_init_once(void) | |
2025 | { | |
2026 | prandom_init_once(&bpf_user_rnd_state); | |
2027 | } | |
2028 | ||
f3694e00 | 2029 | BPF_CALL_0(bpf_user_rnd_u32) |
3ad00405 DB |
2030 | { |
2031 | /* Should someone ever have the rather unwise idea to use some | |
2032 | * of the registers passed into this function, then note that | |
2033 | * this function is called from native eBPF and classic-to-eBPF | |
2034 | * transformations. Register assignments from both sides are | |
2035 | * different, f.e. classic always sets fn(ctx, A, X) here. | |
2036 | */ | |
2037 | struct rnd_state *state; | |
2038 | u32 res; | |
2039 | ||
2040 | state = &get_cpu_var(bpf_user_rnd_state); | |
2041 | res = prandom_u32_state(state); | |
b761fe22 | 2042 | put_cpu_var(bpf_user_rnd_state); |
3ad00405 DB |
2043 | |
2044 | return res; | |
2045 | } | |
2046 | ||
3ba67dab DB |
2047 | /* Weak definitions of helper functions in case we don't have bpf syscall. */ |
2048 | const struct bpf_func_proto bpf_map_lookup_elem_proto __weak; | |
2049 | const struct bpf_func_proto bpf_map_update_elem_proto __weak; | |
2050 | const struct bpf_func_proto bpf_map_delete_elem_proto __weak; | |
f1a2e44a MV |
2051 | const struct bpf_func_proto bpf_map_push_elem_proto __weak; |
2052 | const struct bpf_func_proto bpf_map_pop_elem_proto __weak; | |
2053 | const struct bpf_func_proto bpf_map_peek_elem_proto __weak; | |
3ba67dab | 2054 | |
03e69b50 | 2055 | const struct bpf_func_proto bpf_get_prandom_u32_proto __weak; |
c04167ce | 2056 | const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak; |
2d0e30c3 | 2057 | const struct bpf_func_proto bpf_get_numa_node_id_proto __weak; |
17ca8cbf | 2058 | const struct bpf_func_proto bpf_ktime_get_ns_proto __weak; |
bd570ff9 | 2059 | |
ffeedafb AS |
2060 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak; |
2061 | const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak; | |
2062 | const struct bpf_func_proto bpf_get_current_comm_proto __weak; | |
bf6fa2c8 | 2063 | const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak; |
cd339431 | 2064 | const struct bpf_func_proto bpf_get_local_storage_proto __weak; |
bd570ff9 | 2065 | |
0756ea3e AS |
2066 | const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void) |
2067 | { | |
2068 | return NULL; | |
2069 | } | |
03e69b50 | 2070 | |
555c8a86 DB |
2071 | u64 __weak |
2072 | bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, | |
2073 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 2074 | { |
555c8a86 | 2075 | return -ENOTSUPP; |
bd570ff9 | 2076 | } |
6cb5fb38 | 2077 | EXPORT_SYMBOL_GPL(bpf_event_output); |
bd570ff9 | 2078 | |
3324b584 DB |
2079 | /* Always built-in helper functions. */ |
2080 | const struct bpf_func_proto bpf_tail_call_proto = { | |
2081 | .func = NULL, | |
2082 | .gpl_only = false, | |
2083 | .ret_type = RET_VOID, | |
2084 | .arg1_type = ARG_PTR_TO_CTX, | |
2085 | .arg2_type = ARG_CONST_MAP_PTR, | |
2086 | .arg3_type = ARG_ANYTHING, | |
2087 | }; | |
2088 | ||
9383191d DB |
2089 | /* Stub for JITs that only support cBPF. eBPF programs are interpreted. |
2090 | * It is encouraged to implement bpf_int_jit_compile() instead, so that | |
2091 | * eBPF and implicitly also cBPF can get JITed! | |
2092 | */ | |
d1c55ab5 | 2093 | struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog) |
3324b584 | 2094 | { |
d1c55ab5 | 2095 | return prog; |
3324b584 DB |
2096 | } |
2097 | ||
9383191d DB |
2098 | /* Stub for JITs that support eBPF. All cBPF code gets transformed into |
2099 | * eBPF by the kernel and is later compiled by bpf_int_jit_compile(). | |
2100 | */ | |
2101 | void __weak bpf_jit_compile(struct bpf_prog *prog) | |
2102 | { | |
2103 | } | |
2104 | ||
17bedab2 | 2105 | bool __weak bpf_helper_changes_pkt_data(void *func) |
969bf05e AS |
2106 | { |
2107 | return false; | |
2108 | } | |
2109 | ||
f89b7755 AS |
2110 | /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call |
2111 | * skb_copy_bits(), so provide a weak definition of it for NET-less config. | |
2112 | */ | |
2113 | int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to, | |
2114 | int len) | |
2115 | { | |
2116 | return -EFAULT; | |
2117 | } | |
a67edbf4 DB |
2118 | |
2119 | /* All definitions of tracepoints related to BPF. */ | |
2120 | #define CREATE_TRACE_POINTS | |
2121 | #include <linux/bpf_trace.h> | |
2122 | ||
2123 | EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception); |