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 | ||
492ecee8 | 81 | struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags) |
60a3b225 | 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 | } | |
492ecee8 AS |
107 | |
108 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags) | |
109 | { | |
110 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; | |
111 | struct bpf_prog *prog; | |
112 | ||
113 | prog = bpf_prog_alloc_no_stats(size, gfp_extra_flags); | |
114 | if (!prog) | |
115 | return NULL; | |
116 | ||
117 | prog->aux->stats = alloc_percpu_gfp(struct bpf_prog_stats, gfp_flags); | |
118 | if (!prog->aux->stats) { | |
119 | kfree(prog->aux); | |
120 | vfree(prog); | |
121 | return NULL; | |
122 | } | |
123 | ||
124 | u64_stats_init(&prog->aux->stats->syncp); | |
125 | return prog; | |
126 | } | |
60a3b225 DB |
127 | EXPORT_SYMBOL_GPL(bpf_prog_alloc); |
128 | ||
c454a46b MKL |
129 | int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog) |
130 | { | |
131 | if (!prog->aux->nr_linfo || !prog->jit_requested) | |
132 | return 0; | |
133 | ||
134 | prog->aux->jited_linfo = kcalloc(prog->aux->nr_linfo, | |
135 | sizeof(*prog->aux->jited_linfo), | |
136 | GFP_KERNEL | __GFP_NOWARN); | |
137 | if (!prog->aux->jited_linfo) | |
138 | return -ENOMEM; | |
139 | ||
140 | return 0; | |
141 | } | |
142 | ||
143 | void bpf_prog_free_jited_linfo(struct bpf_prog *prog) | |
144 | { | |
145 | kfree(prog->aux->jited_linfo); | |
146 | prog->aux->jited_linfo = NULL; | |
147 | } | |
148 | ||
149 | void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog) | |
150 | { | |
151 | if (prog->aux->jited_linfo && !prog->aux->jited_linfo[0]) | |
152 | bpf_prog_free_jited_linfo(prog); | |
153 | } | |
154 | ||
155 | /* The jit engine is responsible to provide an array | |
156 | * for insn_off to the jited_off mapping (insn_to_jit_off). | |
157 | * | |
158 | * The idx to this array is the insn_off. Hence, the insn_off | |
159 | * here is relative to the prog itself instead of the main prog. | |
160 | * This array has one entry for each xlated bpf insn. | |
161 | * | |
162 | * jited_off is the byte off to the last byte of the jited insn. | |
163 | * | |
164 | * Hence, with | |
165 | * insn_start: | |
166 | * The first bpf insn off of the prog. The insn off | |
167 | * here is relative to the main prog. | |
168 | * e.g. if prog is a subprog, insn_start > 0 | |
169 | * linfo_idx: | |
170 | * The prog's idx to prog->aux->linfo and jited_linfo | |
171 | * | |
172 | * jited_linfo[linfo_idx] = prog->bpf_func | |
173 | * | |
174 | * For i > linfo_idx, | |
175 | * | |
176 | * jited_linfo[i] = prog->bpf_func + | |
177 | * insn_to_jit_off[linfo[i].insn_off - insn_start - 1] | |
178 | */ | |
179 | void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, | |
180 | const u32 *insn_to_jit_off) | |
181 | { | |
182 | u32 linfo_idx, insn_start, insn_end, nr_linfo, i; | |
183 | const struct bpf_line_info *linfo; | |
184 | void **jited_linfo; | |
185 | ||
186 | if (!prog->aux->jited_linfo) | |
187 | /* Userspace did not provide linfo */ | |
188 | return; | |
189 | ||
190 | linfo_idx = prog->aux->linfo_idx; | |
191 | linfo = &prog->aux->linfo[linfo_idx]; | |
192 | insn_start = linfo[0].insn_off; | |
193 | insn_end = insn_start + prog->len; | |
194 | ||
195 | jited_linfo = &prog->aux->jited_linfo[linfo_idx]; | |
196 | jited_linfo[0] = prog->bpf_func; | |
197 | ||
198 | nr_linfo = prog->aux->nr_linfo - linfo_idx; | |
199 | ||
200 | for (i = 1; i < nr_linfo && linfo[i].insn_off < insn_end; i++) | |
201 | /* The verifier ensures that linfo[i].insn_off is | |
202 | * strictly increasing | |
203 | */ | |
204 | jited_linfo[i] = prog->bpf_func + | |
205 | insn_to_jit_off[linfo[i].insn_off - insn_start - 1]; | |
206 | } | |
207 | ||
208 | void bpf_prog_free_linfo(struct bpf_prog *prog) | |
209 | { | |
210 | bpf_prog_free_jited_linfo(prog); | |
211 | kvfree(prog->aux->linfo); | |
212 | } | |
213 | ||
60a3b225 DB |
214 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, |
215 | gfp_t gfp_extra_flags) | |
216 | { | |
19809c2d | 217 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; |
60a3b225 | 218 | struct bpf_prog *fp; |
5ccb071e DB |
219 | u32 pages, delta; |
220 | int ret; | |
60a3b225 DB |
221 | |
222 | BUG_ON(fp_old == NULL); | |
223 | ||
224 | size = round_up(size, PAGE_SIZE); | |
5ccb071e DB |
225 | pages = size / PAGE_SIZE; |
226 | if (pages <= fp_old->pages) | |
60a3b225 DB |
227 | return fp_old; |
228 | ||
5ccb071e DB |
229 | delta = pages - fp_old->pages; |
230 | ret = __bpf_prog_charge(fp_old->aux->user, delta); | |
231 | if (ret) | |
232 | return NULL; | |
233 | ||
60a3b225 | 234 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
5ccb071e DB |
235 | if (fp == NULL) { |
236 | __bpf_prog_uncharge(fp_old->aux->user, delta); | |
237 | } else { | |
60a3b225 | 238 | memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE); |
5ccb071e | 239 | fp->pages = pages; |
e9d8afa9 | 240 | fp->aux->prog = fp; |
60a3b225 | 241 | |
09756af4 | 242 | /* We keep fp->aux from fp_old around in the new |
60a3b225 DB |
243 | * reallocated structure. |
244 | */ | |
09756af4 | 245 | fp_old->aux = NULL; |
60a3b225 DB |
246 | __bpf_prog_free(fp_old); |
247 | } | |
248 | ||
249 | return fp; | |
250 | } | |
60a3b225 DB |
251 | |
252 | void __bpf_prog_free(struct bpf_prog *fp) | |
253 | { | |
492ecee8 AS |
254 | if (fp->aux) { |
255 | free_percpu(fp->aux->stats); | |
256 | kfree(fp->aux); | |
257 | } | |
60a3b225 DB |
258 | vfree(fp); |
259 | } | |
60a3b225 | 260 | |
f1f7714e | 261 | int bpf_prog_calc_tag(struct bpf_prog *fp) |
7bd509e3 DB |
262 | { |
263 | const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64); | |
f1f7714e DB |
264 | u32 raw_size = bpf_prog_tag_scratch_size(fp); |
265 | u32 digest[SHA_DIGEST_WORDS]; | |
aafe6ae9 | 266 | u32 ws[SHA_WORKSPACE_WORDS]; |
7bd509e3 | 267 | u32 i, bsize, psize, blocks; |
aafe6ae9 | 268 | struct bpf_insn *dst; |
7bd509e3 | 269 | bool was_ld_map; |
aafe6ae9 | 270 | u8 *raw, *todo; |
7bd509e3 DB |
271 | __be32 *result; |
272 | __be64 *bits; | |
273 | ||
aafe6ae9 DB |
274 | raw = vmalloc(raw_size); |
275 | if (!raw) | |
276 | return -ENOMEM; | |
277 | ||
f1f7714e | 278 | sha_init(digest); |
7bd509e3 DB |
279 | memset(ws, 0, sizeof(ws)); |
280 | ||
281 | /* We need to take out the map fd for the digest calculation | |
282 | * since they are unstable from user space side. | |
283 | */ | |
aafe6ae9 | 284 | dst = (void *)raw; |
7bd509e3 DB |
285 | for (i = 0, was_ld_map = false; i < fp->len; i++) { |
286 | dst[i] = fp->insnsi[i]; | |
287 | if (!was_ld_map && | |
288 | dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
289 | dst[i].src_reg == BPF_PSEUDO_MAP_FD) { | |
290 | was_ld_map = true; | |
291 | dst[i].imm = 0; | |
292 | } else if (was_ld_map && | |
293 | dst[i].code == 0 && | |
294 | dst[i].dst_reg == 0 && | |
295 | dst[i].src_reg == 0 && | |
296 | dst[i].off == 0) { | |
297 | was_ld_map = false; | |
298 | dst[i].imm = 0; | |
299 | } else { | |
300 | was_ld_map = false; | |
301 | } | |
302 | } | |
303 | ||
aafe6ae9 DB |
304 | psize = bpf_prog_insn_size(fp); |
305 | memset(&raw[psize], 0, raw_size - psize); | |
7bd509e3 DB |
306 | raw[psize++] = 0x80; |
307 | ||
308 | bsize = round_up(psize, SHA_MESSAGE_BYTES); | |
309 | blocks = bsize / SHA_MESSAGE_BYTES; | |
aafe6ae9 | 310 | todo = raw; |
7bd509e3 DB |
311 | if (bsize - psize >= sizeof(__be64)) { |
312 | bits = (__be64 *)(todo + bsize - sizeof(__be64)); | |
313 | } else { | |
314 | bits = (__be64 *)(todo + bsize + bits_offset); | |
315 | blocks++; | |
316 | } | |
317 | *bits = cpu_to_be64((psize - 1) << 3); | |
318 | ||
319 | while (blocks--) { | |
f1f7714e | 320 | sha_transform(digest, todo, ws); |
7bd509e3 DB |
321 | todo += SHA_MESSAGE_BYTES; |
322 | } | |
323 | ||
f1f7714e | 324 | result = (__force __be32 *)digest; |
7bd509e3 | 325 | for (i = 0; i < SHA_DIGEST_WORDS; i++) |
f1f7714e DB |
326 | result[i] = cpu_to_be32(digest[i]); |
327 | memcpy(fp->tag, result, sizeof(fp->tag)); | |
aafe6ae9 DB |
328 | |
329 | vfree(raw); | |
330 | return 0; | |
7bd509e3 DB |
331 | } |
332 | ||
2cbd95a5 JK |
333 | static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, s32 end_old, |
334 | s32 end_new, u32 curr, const bool probe_pass) | |
c237ee5e | 335 | { |
050fad7c | 336 | const s64 imm_min = S32_MIN, imm_max = S32_MAX; |
2cbd95a5 | 337 | s32 delta = end_new - end_old; |
050fad7c DB |
338 | s64 imm = insn->imm; |
339 | ||
2cbd95a5 | 340 | if (curr < pos && curr + imm + 1 >= end_old) |
050fad7c | 341 | imm += delta; |
2cbd95a5 | 342 | else if (curr >= end_new && curr + imm + 1 < end_new) |
050fad7c DB |
343 | imm -= delta; |
344 | if (imm < imm_min || imm > imm_max) | |
345 | return -ERANGE; | |
346 | if (!probe_pass) | |
347 | insn->imm = imm; | |
348 | return 0; | |
349 | } | |
350 | ||
2cbd95a5 JK |
351 | static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, s32 end_old, |
352 | s32 end_new, u32 curr, const bool probe_pass) | |
050fad7c DB |
353 | { |
354 | const s32 off_min = S16_MIN, off_max = S16_MAX; | |
2cbd95a5 | 355 | s32 delta = end_new - end_old; |
050fad7c DB |
356 | s32 off = insn->off; |
357 | ||
2cbd95a5 | 358 | if (curr < pos && curr + off + 1 >= end_old) |
050fad7c | 359 | off += delta; |
2cbd95a5 | 360 | else if (curr >= end_new && curr + off + 1 < end_new) |
050fad7c DB |
361 | off -= delta; |
362 | if (off < off_min || off > off_max) | |
363 | return -ERANGE; | |
364 | if (!probe_pass) | |
365 | insn->off = off; | |
366 | return 0; | |
367 | } | |
368 | ||
2cbd95a5 JK |
369 | static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, s32 end_old, |
370 | s32 end_new, const bool probe_pass) | |
050fad7c | 371 | { |
2cbd95a5 | 372 | u32 i, insn_cnt = prog->len + (probe_pass ? end_new - end_old : 0); |
c237ee5e | 373 | struct bpf_insn *insn = prog->insnsi; |
050fad7c | 374 | int ret = 0; |
c237ee5e DB |
375 | |
376 | for (i = 0; i < insn_cnt; i++, insn++) { | |
050fad7c DB |
377 | u8 code; |
378 | ||
379 | /* In the probing pass we still operate on the original, | |
380 | * unpatched image in order to check overflows before we | |
381 | * do any other adjustments. Therefore skip the patchlet. | |
382 | */ | |
383 | if (probe_pass && i == pos) { | |
2cbd95a5 JK |
384 | i = end_new; |
385 | insn = prog->insnsi + end_old; | |
050fad7c | 386 | } |
1ea47e01 | 387 | code = insn->code; |
092ed096 JW |
388 | if ((BPF_CLASS(code) != BPF_JMP && |
389 | BPF_CLASS(code) != BPF_JMP32) || | |
050fad7c | 390 | BPF_OP(code) == BPF_EXIT) |
1ea47e01 | 391 | continue; |
050fad7c | 392 | /* Adjust offset of jmps if we cross patch boundaries. */ |
1ea47e01 | 393 | if (BPF_OP(code) == BPF_CALL) { |
050fad7c | 394 | if (insn->src_reg != BPF_PSEUDO_CALL) |
1ea47e01 | 395 | continue; |
2cbd95a5 JK |
396 | ret = bpf_adj_delta_to_imm(insn, pos, end_old, |
397 | end_new, i, probe_pass); | |
1ea47e01 | 398 | } else { |
2cbd95a5 JK |
399 | ret = bpf_adj_delta_to_off(insn, pos, end_old, |
400 | end_new, i, probe_pass); | |
1ea47e01 | 401 | } |
050fad7c DB |
402 | if (ret) |
403 | break; | |
c237ee5e | 404 | } |
050fad7c DB |
405 | |
406 | return ret; | |
c237ee5e DB |
407 | } |
408 | ||
c454a46b MKL |
409 | static void bpf_adj_linfo(struct bpf_prog *prog, u32 off, u32 delta) |
410 | { | |
411 | struct bpf_line_info *linfo; | |
412 | u32 i, nr_linfo; | |
413 | ||
414 | nr_linfo = prog->aux->nr_linfo; | |
415 | if (!nr_linfo || !delta) | |
416 | return; | |
417 | ||
418 | linfo = prog->aux->linfo; | |
419 | ||
420 | for (i = 0; i < nr_linfo; i++) | |
421 | if (off < linfo[i].insn_off) | |
422 | break; | |
423 | ||
424 | /* Push all off < linfo[i].insn_off by delta */ | |
425 | for (; i < nr_linfo; i++) | |
426 | linfo[i].insn_off += delta; | |
427 | } | |
428 | ||
c237ee5e DB |
429 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
430 | const struct bpf_insn *patch, u32 len) | |
431 | { | |
432 | u32 insn_adj_cnt, insn_rest, insn_delta = len - 1; | |
050fad7c | 433 | const u32 cnt_max = S16_MAX; |
c237ee5e DB |
434 | struct bpf_prog *prog_adj; |
435 | ||
436 | /* Since our patchlet doesn't expand the image, we're done. */ | |
437 | if (insn_delta == 0) { | |
438 | memcpy(prog->insnsi + off, patch, sizeof(*patch)); | |
439 | return prog; | |
440 | } | |
441 | ||
442 | insn_adj_cnt = prog->len + insn_delta; | |
443 | ||
050fad7c DB |
444 | /* Reject anything that would potentially let the insn->off |
445 | * target overflow when we have excessive program expansions. | |
446 | * We need to probe here before we do any reallocation where | |
447 | * we afterwards may not fail anymore. | |
448 | */ | |
449 | if (insn_adj_cnt > cnt_max && | |
2cbd95a5 | 450 | bpf_adj_branches(prog, off, off + 1, off + len, true)) |
050fad7c DB |
451 | return NULL; |
452 | ||
c237ee5e DB |
453 | /* Several new instructions need to be inserted. Make room |
454 | * for them. Likely, there's no need for a new allocation as | |
455 | * last page could have large enough tailroom. | |
456 | */ | |
457 | prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt), | |
458 | GFP_USER); | |
459 | if (!prog_adj) | |
460 | return NULL; | |
461 | ||
462 | prog_adj->len = insn_adj_cnt; | |
463 | ||
464 | /* Patching happens in 3 steps: | |
465 | * | |
466 | * 1) Move over tail of insnsi from next instruction onwards, | |
467 | * so we can patch the single target insn with one or more | |
468 | * new ones (patching is always from 1 to n insns, n > 0). | |
469 | * 2) Inject new instructions at the target location. | |
470 | * 3) Adjust branch offsets if necessary. | |
471 | */ | |
472 | insn_rest = insn_adj_cnt - off - len; | |
473 | ||
474 | memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1, | |
475 | sizeof(*patch) * insn_rest); | |
476 | memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len); | |
477 | ||
050fad7c DB |
478 | /* We are guaranteed to not fail at this point, otherwise |
479 | * the ship has sailed to reverse to the original state. An | |
480 | * overflow cannot happen at this point. | |
481 | */ | |
2cbd95a5 | 482 | BUG_ON(bpf_adj_branches(prog_adj, off, off + 1, off + len, false)); |
c237ee5e | 483 | |
c454a46b MKL |
484 | bpf_adj_linfo(prog_adj, off, insn_delta); |
485 | ||
c237ee5e DB |
486 | return prog_adj; |
487 | } | |
488 | ||
52875a04 JK |
489 | int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt) |
490 | { | |
491 | /* Branch offsets can't overflow when program is shrinking, no need | |
492 | * to call bpf_adj_branches(..., true) here | |
493 | */ | |
494 | memmove(prog->insnsi + off, prog->insnsi + off + cnt, | |
495 | sizeof(struct bpf_insn) * (prog->len - off - cnt)); | |
496 | prog->len -= cnt; | |
497 | ||
498 | return WARN_ON_ONCE(bpf_adj_branches(prog, off, off + cnt, off, false)); | |
499 | } | |
500 | ||
7d1982b4 DB |
501 | void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp) |
502 | { | |
503 | int i; | |
504 | ||
505 | for (i = 0; i < fp->aux->func_cnt; i++) | |
506 | bpf_prog_kallsyms_del(fp->aux->func[i]); | |
507 | } | |
508 | ||
509 | void bpf_prog_kallsyms_del_all(struct bpf_prog *fp) | |
510 | { | |
511 | bpf_prog_kallsyms_del_subprogs(fp); | |
512 | bpf_prog_kallsyms_del(fp); | |
513 | } | |
514 | ||
b954d834 | 515 | #ifdef CONFIG_BPF_JIT |
fa9dd599 DB |
516 | /* All BPF JIT sysctl knobs here. */ |
517 | int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON); | |
518 | int bpf_jit_harden __read_mostly; | |
519 | int bpf_jit_kallsyms __read_mostly; | |
fdadd049 | 520 | long bpf_jit_limit __read_mostly; |
fa9dd599 | 521 | |
74451e66 DB |
522 | static __always_inline void |
523 | bpf_get_prog_addr_region(const struct bpf_prog *prog, | |
524 | unsigned long *symbol_start, | |
525 | unsigned long *symbol_end) | |
526 | { | |
527 | const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog); | |
528 | unsigned long addr = (unsigned long)hdr; | |
529 | ||
530 | WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog)); | |
531 | ||
532 | *symbol_start = addr; | |
533 | *symbol_end = addr + hdr->pages * PAGE_SIZE; | |
534 | } | |
535 | ||
536 | static void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) | |
537 | { | |
368211fb | 538 | const char *end = sym + KSYM_NAME_LEN; |
838e9690 YS |
539 | const struct btf_type *type; |
540 | const char *func_name; | |
368211fb | 541 | |
74451e66 | 542 | BUILD_BUG_ON(sizeof("bpf_prog_") + |
368211fb MKL |
543 | sizeof(prog->tag) * 2 + |
544 | /* name has been null terminated. | |
545 | * We should need +1 for the '_' preceding | |
546 | * the name. However, the null character | |
547 | * is double counted between the name and the | |
548 | * sizeof("bpf_prog_") above, so we omit | |
549 | * the +1 here. | |
550 | */ | |
551 | sizeof(prog->aux->name) > KSYM_NAME_LEN); | |
74451e66 DB |
552 | |
553 | sym += snprintf(sym, KSYM_NAME_LEN, "bpf_prog_"); | |
554 | sym = bin2hex(sym, prog->tag, sizeof(prog->tag)); | |
838e9690 YS |
555 | |
556 | /* prog->aux->name will be ignored if full btf name is available */ | |
7337224f | 557 | if (prog->aux->func_info_cnt) { |
ba64e7d8 YS |
558 | type = btf_type_by_id(prog->aux->btf, |
559 | prog->aux->func_info[prog->aux->func_idx].type_id); | |
838e9690 YS |
560 | func_name = btf_name_by_offset(prog->aux->btf, type->name_off); |
561 | snprintf(sym, (size_t)(end - sym), "_%s", func_name); | |
562 | return; | |
563 | } | |
564 | ||
368211fb MKL |
565 | if (prog->aux->name[0]) |
566 | snprintf(sym, (size_t)(end - sym), "_%s", prog->aux->name); | |
567 | else | |
568 | *sym = 0; | |
74451e66 DB |
569 | } |
570 | ||
571 | static __always_inline unsigned long | |
572 | bpf_get_prog_addr_start(struct latch_tree_node *n) | |
573 | { | |
574 | unsigned long symbol_start, symbol_end; | |
575 | const struct bpf_prog_aux *aux; | |
576 | ||
577 | aux = container_of(n, struct bpf_prog_aux, ksym_tnode); | |
578 | bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); | |
579 | ||
580 | return symbol_start; | |
581 | } | |
582 | ||
583 | static __always_inline bool bpf_tree_less(struct latch_tree_node *a, | |
584 | struct latch_tree_node *b) | |
585 | { | |
586 | return bpf_get_prog_addr_start(a) < bpf_get_prog_addr_start(b); | |
587 | } | |
588 | ||
589 | static __always_inline int bpf_tree_comp(void *key, struct latch_tree_node *n) | |
590 | { | |
591 | unsigned long val = (unsigned long)key; | |
592 | unsigned long symbol_start, symbol_end; | |
593 | const struct bpf_prog_aux *aux; | |
594 | ||
595 | aux = container_of(n, struct bpf_prog_aux, ksym_tnode); | |
596 | bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); | |
597 | ||
598 | if (val < symbol_start) | |
599 | return -1; | |
600 | if (val >= symbol_end) | |
601 | return 1; | |
602 | ||
603 | return 0; | |
604 | } | |
605 | ||
606 | static const struct latch_tree_ops bpf_tree_ops = { | |
607 | .less = bpf_tree_less, | |
608 | .comp = bpf_tree_comp, | |
609 | }; | |
610 | ||
611 | static DEFINE_SPINLOCK(bpf_lock); | |
612 | static LIST_HEAD(bpf_kallsyms); | |
613 | static struct latch_tree_root bpf_tree __cacheline_aligned; | |
614 | ||
74451e66 DB |
615 | static void bpf_prog_ksym_node_add(struct bpf_prog_aux *aux) |
616 | { | |
617 | WARN_ON_ONCE(!list_empty(&aux->ksym_lnode)); | |
618 | list_add_tail_rcu(&aux->ksym_lnode, &bpf_kallsyms); | |
619 | latch_tree_insert(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); | |
620 | } | |
621 | ||
622 | static void bpf_prog_ksym_node_del(struct bpf_prog_aux *aux) | |
623 | { | |
624 | if (list_empty(&aux->ksym_lnode)) | |
625 | return; | |
626 | ||
627 | latch_tree_erase(&aux->ksym_tnode, &bpf_tree, &bpf_tree_ops); | |
628 | list_del_rcu(&aux->ksym_lnode); | |
629 | } | |
630 | ||
631 | static bool bpf_prog_kallsyms_candidate(const struct bpf_prog *fp) | |
632 | { | |
633 | return fp->jited && !bpf_prog_was_classic(fp); | |
634 | } | |
635 | ||
636 | static bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) | |
637 | { | |
638 | return list_empty(&fp->aux->ksym_lnode) || | |
639 | fp->aux->ksym_lnode.prev == LIST_POISON2; | |
640 | } | |
641 | ||
642 | void bpf_prog_kallsyms_add(struct bpf_prog *fp) | |
643 | { | |
74451e66 DB |
644 | if (!bpf_prog_kallsyms_candidate(fp) || |
645 | !capable(CAP_SYS_ADMIN)) | |
646 | return; | |
647 | ||
d24f7c7f | 648 | spin_lock_bh(&bpf_lock); |
74451e66 | 649 | bpf_prog_ksym_node_add(fp->aux); |
d24f7c7f | 650 | spin_unlock_bh(&bpf_lock); |
74451e66 DB |
651 | } |
652 | ||
653 | void bpf_prog_kallsyms_del(struct bpf_prog *fp) | |
654 | { | |
74451e66 DB |
655 | if (!bpf_prog_kallsyms_candidate(fp)) |
656 | return; | |
657 | ||
d24f7c7f | 658 | spin_lock_bh(&bpf_lock); |
74451e66 | 659 | bpf_prog_ksym_node_del(fp->aux); |
d24f7c7f | 660 | spin_unlock_bh(&bpf_lock); |
74451e66 DB |
661 | } |
662 | ||
663 | static struct bpf_prog *bpf_prog_kallsyms_find(unsigned long addr) | |
664 | { | |
665 | struct latch_tree_node *n; | |
666 | ||
667 | if (!bpf_jit_kallsyms_enabled()) | |
668 | return NULL; | |
669 | ||
670 | n = latch_tree_find((void *)addr, &bpf_tree, &bpf_tree_ops); | |
671 | return n ? | |
672 | container_of(n, struct bpf_prog_aux, ksym_tnode)->prog : | |
673 | NULL; | |
674 | } | |
675 | ||
676 | const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, | |
677 | unsigned long *off, char *sym) | |
678 | { | |
679 | unsigned long symbol_start, symbol_end; | |
680 | struct bpf_prog *prog; | |
681 | char *ret = NULL; | |
682 | ||
683 | rcu_read_lock(); | |
684 | prog = bpf_prog_kallsyms_find(addr); | |
685 | if (prog) { | |
686 | bpf_get_prog_addr_region(prog, &symbol_start, &symbol_end); | |
687 | bpf_get_prog_name(prog, sym); | |
688 | ||
689 | ret = sym; | |
690 | if (size) | |
691 | *size = symbol_end - symbol_start; | |
692 | if (off) | |
693 | *off = addr - symbol_start; | |
694 | } | |
695 | rcu_read_unlock(); | |
696 | ||
697 | return ret; | |
698 | } | |
699 | ||
700 | bool is_bpf_text_address(unsigned long addr) | |
701 | { | |
702 | bool ret; | |
703 | ||
704 | rcu_read_lock(); | |
705 | ret = bpf_prog_kallsyms_find(addr) != NULL; | |
706 | rcu_read_unlock(); | |
707 | ||
708 | return ret; | |
709 | } | |
710 | ||
711 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
712 | char *sym) | |
713 | { | |
74451e66 DB |
714 | struct bpf_prog_aux *aux; |
715 | unsigned int it = 0; | |
716 | int ret = -ERANGE; | |
717 | ||
718 | if (!bpf_jit_kallsyms_enabled()) | |
719 | return ret; | |
720 | ||
721 | rcu_read_lock(); | |
722 | list_for_each_entry_rcu(aux, &bpf_kallsyms, ksym_lnode) { | |
723 | if (it++ != symnum) | |
724 | continue; | |
725 | ||
74451e66 DB |
726 | bpf_get_prog_name(aux->prog, sym); |
727 | ||
df073470 | 728 | *value = (unsigned long)aux->prog->bpf_func; |
74451e66 DB |
729 | *type = BPF_SYM_ELF_TYPE; |
730 | ||
731 | ret = 0; | |
732 | break; | |
733 | } | |
734 | rcu_read_unlock(); | |
735 | ||
736 | return ret; | |
737 | } | |
738 | ||
ede95a63 DB |
739 | static atomic_long_t bpf_jit_current; |
740 | ||
fdadd049 DB |
741 | /* Can be overridden by an arch's JIT compiler if it has a custom, |
742 | * dedicated BPF backend memory area, or if neither of the two | |
743 | * below apply. | |
744 | */ | |
745 | u64 __weak bpf_jit_alloc_exec_limit(void) | |
746 | { | |
ede95a63 | 747 | #if defined(MODULES_VADDR) |
fdadd049 DB |
748 | return MODULES_END - MODULES_VADDR; |
749 | #else | |
750 | return VMALLOC_END - VMALLOC_START; | |
751 | #endif | |
752 | } | |
753 | ||
ede95a63 DB |
754 | static int __init bpf_jit_charge_init(void) |
755 | { | |
756 | /* Only used as heuristic here to derive limit. */ | |
fdadd049 DB |
757 | bpf_jit_limit = min_t(u64, round_up(bpf_jit_alloc_exec_limit() >> 2, |
758 | PAGE_SIZE), LONG_MAX); | |
ede95a63 DB |
759 | return 0; |
760 | } | |
761 | pure_initcall(bpf_jit_charge_init); | |
ede95a63 DB |
762 | |
763 | static int bpf_jit_charge_modmem(u32 pages) | |
764 | { | |
765 | if (atomic_long_add_return(pages, &bpf_jit_current) > | |
766 | (bpf_jit_limit >> PAGE_SHIFT)) { | |
767 | if (!capable(CAP_SYS_ADMIN)) { | |
768 | atomic_long_sub(pages, &bpf_jit_current); | |
769 | return -EPERM; | |
770 | } | |
771 | } | |
772 | ||
773 | return 0; | |
774 | } | |
775 | ||
776 | static void bpf_jit_uncharge_modmem(u32 pages) | |
777 | { | |
778 | atomic_long_sub(pages, &bpf_jit_current); | |
779 | } | |
780 | ||
dc002bb6 AB |
781 | void *__weak bpf_jit_alloc_exec(unsigned long size) |
782 | { | |
783 | return module_alloc(size); | |
784 | } | |
785 | ||
786 | void __weak bpf_jit_free_exec(void *addr) | |
787 | { | |
788 | module_memfree(addr); | |
789 | } | |
790 | ||
738cbe72 DB |
791 | struct bpf_binary_header * |
792 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
793 | unsigned int alignment, | |
794 | bpf_jit_fill_hole_t bpf_fill_ill_insns) | |
795 | { | |
796 | struct bpf_binary_header *hdr; | |
ede95a63 | 797 | u32 size, hole, start, pages; |
738cbe72 DB |
798 | |
799 | /* Most of BPF filters are really small, but if some of them | |
800 | * fill a page, allow at least 128 extra bytes to insert a | |
801 | * random section of illegal instructions. | |
802 | */ | |
803 | size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); | |
ede95a63 DB |
804 | pages = size / PAGE_SIZE; |
805 | ||
806 | if (bpf_jit_charge_modmem(pages)) | |
807 | return NULL; | |
dc002bb6 | 808 | hdr = bpf_jit_alloc_exec(size); |
ede95a63 DB |
809 | if (!hdr) { |
810 | bpf_jit_uncharge_modmem(pages); | |
738cbe72 | 811 | return NULL; |
ede95a63 | 812 | } |
738cbe72 DB |
813 | |
814 | /* Fill space with illegal/arch-dep instructions. */ | |
815 | bpf_fill_ill_insns(hdr, size); | |
816 | ||
ede95a63 | 817 | hdr->pages = pages; |
738cbe72 DB |
818 | hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), |
819 | PAGE_SIZE - sizeof(*hdr)); | |
b7552e1b | 820 | start = (get_random_int() % hole) & ~(alignment - 1); |
738cbe72 DB |
821 | |
822 | /* Leave a random number of instructions before BPF code. */ | |
823 | *image_ptr = &hdr->image[start]; | |
824 | ||
825 | return hdr; | |
826 | } | |
827 | ||
828 | void bpf_jit_binary_free(struct bpf_binary_header *hdr) | |
829 | { | |
ede95a63 DB |
830 | u32 pages = hdr->pages; |
831 | ||
dc002bb6 | 832 | bpf_jit_free_exec(hdr); |
ede95a63 | 833 | bpf_jit_uncharge_modmem(pages); |
738cbe72 | 834 | } |
4f3446bb | 835 | |
74451e66 DB |
836 | /* This symbol is only overridden by archs that have different |
837 | * requirements than the usual eBPF JITs, f.e. when they only | |
838 | * implement cBPF JIT, do not set images read-only, etc. | |
839 | */ | |
840 | void __weak bpf_jit_free(struct bpf_prog *fp) | |
841 | { | |
842 | if (fp->jited) { | |
843 | struct bpf_binary_header *hdr = bpf_jit_binary_hdr(fp); | |
844 | ||
845 | bpf_jit_binary_unlock_ro(hdr); | |
846 | bpf_jit_binary_free(hdr); | |
847 | ||
848 | WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp)); | |
849 | } | |
850 | ||
851 | bpf_prog_unlock_free(fp); | |
852 | } | |
853 | ||
e2c95a61 DB |
854 | int bpf_jit_get_func_addr(const struct bpf_prog *prog, |
855 | const struct bpf_insn *insn, bool extra_pass, | |
856 | u64 *func_addr, bool *func_addr_fixed) | |
857 | { | |
858 | s16 off = insn->off; | |
859 | s32 imm = insn->imm; | |
860 | u8 *addr; | |
861 | ||
862 | *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL; | |
863 | if (!*func_addr_fixed) { | |
864 | /* Place-holder address till the last pass has collected | |
865 | * all addresses for JITed subprograms in which case we | |
866 | * can pick them up from prog->aux. | |
867 | */ | |
868 | if (!extra_pass) | |
869 | addr = NULL; | |
870 | else if (prog->aux->func && | |
871 | off >= 0 && off < prog->aux->func_cnt) | |
872 | addr = (u8 *)prog->aux->func[off]->bpf_func; | |
873 | else | |
874 | return -EINVAL; | |
875 | } else { | |
876 | /* Address of a BPF helper call. Since part of the core | |
877 | * kernel, it's always at a fixed location. __bpf_call_base | |
878 | * and the helper with imm relative to it are both in core | |
879 | * kernel. | |
880 | */ | |
881 | addr = (u8 *)__bpf_call_base + imm; | |
882 | } | |
883 | ||
884 | *func_addr = (unsigned long)addr; | |
885 | return 0; | |
886 | } | |
887 | ||
4f3446bb DB |
888 | static int bpf_jit_blind_insn(const struct bpf_insn *from, |
889 | const struct bpf_insn *aux, | |
890 | struct bpf_insn *to_buff) | |
891 | { | |
892 | struct bpf_insn *to = to_buff; | |
b7552e1b | 893 | u32 imm_rnd = get_random_int(); |
4f3446bb DB |
894 | s16 off; |
895 | ||
896 | BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG); | |
897 | BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG); | |
898 | ||
9b73bfdd DB |
899 | /* Constraints on AX register: |
900 | * | |
901 | * AX register is inaccessible from user space. It is mapped in | |
902 | * all JITs, and used here for constant blinding rewrites. It is | |
903 | * typically "stateless" meaning its contents are only valid within | |
904 | * the executed instruction, but not across several instructions. | |
905 | * There are a few exceptions however which are further detailed | |
906 | * below. | |
907 | * | |
908 | * Constant blinding is only used by JITs, not in the interpreter. | |
909 | * The interpreter uses AX in some occasions as a local temporary | |
910 | * register e.g. in DIV or MOD instructions. | |
911 | * | |
912 | * In restricted circumstances, the verifier can also use the AX | |
913 | * register for rewrites as long as they do not interfere with | |
914 | * the above cases! | |
915 | */ | |
916 | if (from->dst_reg == BPF_REG_AX || from->src_reg == BPF_REG_AX) | |
917 | goto out; | |
918 | ||
4f3446bb DB |
919 | if (from->imm == 0 && |
920 | (from->code == (BPF_ALU | BPF_MOV | BPF_K) || | |
921 | from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) { | |
922 | *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg); | |
923 | goto out; | |
924 | } | |
925 | ||
926 | switch (from->code) { | |
927 | case BPF_ALU | BPF_ADD | BPF_K: | |
928 | case BPF_ALU | BPF_SUB | BPF_K: | |
929 | case BPF_ALU | BPF_AND | BPF_K: | |
930 | case BPF_ALU | BPF_OR | BPF_K: | |
931 | case BPF_ALU | BPF_XOR | BPF_K: | |
932 | case BPF_ALU | BPF_MUL | BPF_K: | |
933 | case BPF_ALU | BPF_MOV | BPF_K: | |
934 | case BPF_ALU | BPF_DIV | BPF_K: | |
935 | case BPF_ALU | BPF_MOD | BPF_K: | |
936 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
937 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
938 | *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX); | |
939 | break; | |
940 | ||
941 | case BPF_ALU64 | BPF_ADD | BPF_K: | |
942 | case BPF_ALU64 | BPF_SUB | BPF_K: | |
943 | case BPF_ALU64 | BPF_AND | BPF_K: | |
944 | case BPF_ALU64 | BPF_OR | BPF_K: | |
945 | case BPF_ALU64 | BPF_XOR | BPF_K: | |
946 | case BPF_ALU64 | BPF_MUL | BPF_K: | |
947 | case BPF_ALU64 | BPF_MOV | BPF_K: | |
948 | case BPF_ALU64 | BPF_DIV | BPF_K: | |
949 | case BPF_ALU64 | BPF_MOD | BPF_K: | |
950 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
951 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
952 | *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX); | |
953 | break; | |
954 | ||
955 | case BPF_JMP | BPF_JEQ | BPF_K: | |
956 | case BPF_JMP | BPF_JNE | BPF_K: | |
957 | case BPF_JMP | BPF_JGT | BPF_K: | |
92b31a9a | 958 | case BPF_JMP | BPF_JLT | BPF_K: |
4f3446bb | 959 | case BPF_JMP | BPF_JGE | BPF_K: |
92b31a9a | 960 | case BPF_JMP | BPF_JLE | BPF_K: |
4f3446bb | 961 | case BPF_JMP | BPF_JSGT | BPF_K: |
92b31a9a | 962 | case BPF_JMP | BPF_JSLT | BPF_K: |
4f3446bb | 963 | case BPF_JMP | BPF_JSGE | BPF_K: |
92b31a9a | 964 | case BPF_JMP | BPF_JSLE | BPF_K: |
4f3446bb DB |
965 | case BPF_JMP | BPF_JSET | BPF_K: |
966 | /* Accommodate for extra offset in case of a backjump. */ | |
967 | off = from->off; | |
968 | if (off < 0) | |
969 | off -= 2; | |
970 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
971 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
972 | *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off); | |
973 | break; | |
974 | ||
a7b76c88 JW |
975 | case BPF_JMP32 | BPF_JEQ | BPF_K: |
976 | case BPF_JMP32 | BPF_JNE | BPF_K: | |
977 | case BPF_JMP32 | BPF_JGT | BPF_K: | |
978 | case BPF_JMP32 | BPF_JLT | BPF_K: | |
979 | case BPF_JMP32 | BPF_JGE | BPF_K: | |
980 | case BPF_JMP32 | BPF_JLE | BPF_K: | |
981 | case BPF_JMP32 | BPF_JSGT | BPF_K: | |
982 | case BPF_JMP32 | BPF_JSLT | BPF_K: | |
983 | case BPF_JMP32 | BPF_JSGE | BPF_K: | |
984 | case BPF_JMP32 | BPF_JSLE | BPF_K: | |
985 | case BPF_JMP32 | BPF_JSET | BPF_K: | |
986 | /* Accommodate for extra offset in case of a backjump. */ | |
987 | off = from->off; | |
988 | if (off < 0) | |
989 | off -= 2; | |
990 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
991 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
992 | *to++ = BPF_JMP32_REG(from->code, from->dst_reg, BPF_REG_AX, | |
993 | off); | |
994 | break; | |
995 | ||
4f3446bb DB |
996 | case BPF_LD | BPF_IMM | BPF_DW: |
997 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm); | |
998 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
999 | *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
1000 | *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX); | |
1001 | break; | |
1002 | case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */ | |
1003 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm); | |
1004 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
1005 | *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX); | |
1006 | break; | |
1007 | ||
1008 | case BPF_ST | BPF_MEM | BPF_DW: | |
1009 | case BPF_ST | BPF_MEM | BPF_W: | |
1010 | case BPF_ST | BPF_MEM | BPF_H: | |
1011 | case BPF_ST | BPF_MEM | BPF_B: | |
1012 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
1013 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
1014 | *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off); | |
1015 | break; | |
1016 | } | |
1017 | out: | |
1018 | return to - to_buff; | |
1019 | } | |
1020 | ||
1021 | static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other, | |
1022 | gfp_t gfp_extra_flags) | |
1023 | { | |
19809c2d | 1024 | gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | gfp_extra_flags; |
4f3446bb DB |
1025 | struct bpf_prog *fp; |
1026 | ||
1027 | fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL); | |
1028 | if (fp != NULL) { | |
4f3446bb DB |
1029 | /* aux->prog still points to the fp_other one, so |
1030 | * when promoting the clone to the real program, | |
1031 | * this still needs to be adapted. | |
1032 | */ | |
1033 | memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE); | |
1034 | } | |
1035 | ||
1036 | return fp; | |
1037 | } | |
1038 | ||
1039 | static void bpf_prog_clone_free(struct bpf_prog *fp) | |
1040 | { | |
1041 | /* aux was stolen by the other clone, so we cannot free | |
1042 | * it from this path! It will be freed eventually by the | |
1043 | * other program on release. | |
1044 | * | |
1045 | * At this point, we don't need a deferred release since | |
1046 | * clone is guaranteed to not be locked. | |
1047 | */ | |
1048 | fp->aux = NULL; | |
1049 | __bpf_prog_free(fp); | |
1050 | } | |
1051 | ||
1052 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other) | |
1053 | { | |
1054 | /* We have to repoint aux->prog to self, as we don't | |
1055 | * know whether fp here is the clone or the original. | |
1056 | */ | |
1057 | fp->aux->prog = fp; | |
1058 | bpf_prog_clone_free(fp_other); | |
1059 | } | |
1060 | ||
1061 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog) | |
1062 | { | |
1063 | struct bpf_insn insn_buff[16], aux[2]; | |
1064 | struct bpf_prog *clone, *tmp; | |
1065 | int insn_delta, insn_cnt; | |
1066 | struct bpf_insn *insn; | |
1067 | int i, rewritten; | |
1068 | ||
1c2a088a | 1069 | if (!bpf_jit_blinding_enabled(prog) || prog->blinded) |
4f3446bb DB |
1070 | return prog; |
1071 | ||
1072 | clone = bpf_prog_clone_create(prog, GFP_USER); | |
1073 | if (!clone) | |
1074 | return ERR_PTR(-ENOMEM); | |
1075 | ||
1076 | insn_cnt = clone->len; | |
1077 | insn = clone->insnsi; | |
1078 | ||
1079 | for (i = 0; i < insn_cnt; i++, insn++) { | |
1080 | /* We temporarily need to hold the original ld64 insn | |
1081 | * so that we can still access the first part in the | |
1082 | * second blinding run. | |
1083 | */ | |
1084 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
1085 | insn[1].code == 0) | |
1086 | memcpy(aux, insn, sizeof(aux)); | |
1087 | ||
1088 | rewritten = bpf_jit_blind_insn(insn, aux, insn_buff); | |
1089 | if (!rewritten) | |
1090 | continue; | |
1091 | ||
1092 | tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten); | |
1093 | if (!tmp) { | |
1094 | /* Patching may have repointed aux->prog during | |
1095 | * realloc from the original one, so we need to | |
1096 | * fix it up here on error. | |
1097 | */ | |
1098 | bpf_jit_prog_release_other(prog, clone); | |
1099 | return ERR_PTR(-ENOMEM); | |
1100 | } | |
1101 | ||
1102 | clone = tmp; | |
1103 | insn_delta = rewritten - 1; | |
1104 | ||
1105 | /* Walk new program and skip insns we just inserted. */ | |
1106 | insn = clone->insnsi + i + insn_delta; | |
1107 | insn_cnt += insn_delta; | |
1108 | i += insn_delta; | |
1109 | } | |
1110 | ||
1c2a088a | 1111 | clone->blinded = 1; |
4f3446bb DB |
1112 | return clone; |
1113 | } | |
b954d834 | 1114 | #endif /* CONFIG_BPF_JIT */ |
738cbe72 | 1115 | |
f5bffecd AS |
1116 | /* Base function for offset calculation. Needs to go into .text section, |
1117 | * therefore keeping it non-static as well; will also be used by JITs | |
7105e828 DB |
1118 | * anyway later on, so do not let the compiler omit it. This also needs |
1119 | * to go into kallsyms for correlation from e.g. bpftool, so naming | |
1120 | * must not change. | |
f5bffecd AS |
1121 | */ |
1122 | noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
1123 | { | |
1124 | return 0; | |
1125 | } | |
4d9c5c53 | 1126 | EXPORT_SYMBOL_GPL(__bpf_call_base); |
f5bffecd | 1127 | |
5e581dad DB |
1128 | /* All UAPI available opcodes. */ |
1129 | #define BPF_INSN_MAP(INSN_2, INSN_3) \ | |
1130 | /* 32 bit ALU operations. */ \ | |
1131 | /* Register based. */ \ | |
2dc6b100 JW |
1132 | INSN_3(ALU, ADD, X), \ |
1133 | INSN_3(ALU, SUB, X), \ | |
1134 | INSN_3(ALU, AND, X), \ | |
1135 | INSN_3(ALU, OR, X), \ | |
1136 | INSN_3(ALU, LSH, X), \ | |
1137 | INSN_3(ALU, RSH, X), \ | |
1138 | INSN_3(ALU, XOR, X), \ | |
1139 | INSN_3(ALU, MUL, X), \ | |
1140 | INSN_3(ALU, MOV, X), \ | |
1141 | INSN_3(ALU, ARSH, X), \ | |
1142 | INSN_3(ALU, DIV, X), \ | |
1143 | INSN_3(ALU, MOD, X), \ | |
5e581dad DB |
1144 | INSN_2(ALU, NEG), \ |
1145 | INSN_3(ALU, END, TO_BE), \ | |
1146 | INSN_3(ALU, END, TO_LE), \ | |
1147 | /* Immediate based. */ \ | |
2dc6b100 JW |
1148 | INSN_3(ALU, ADD, K), \ |
1149 | INSN_3(ALU, SUB, K), \ | |
1150 | INSN_3(ALU, AND, K), \ | |
1151 | INSN_3(ALU, OR, K), \ | |
1152 | INSN_3(ALU, LSH, K), \ | |
1153 | INSN_3(ALU, RSH, K), \ | |
1154 | INSN_3(ALU, XOR, K), \ | |
1155 | INSN_3(ALU, MUL, K), \ | |
1156 | INSN_3(ALU, MOV, K), \ | |
1157 | INSN_3(ALU, ARSH, K), \ | |
1158 | INSN_3(ALU, DIV, K), \ | |
1159 | INSN_3(ALU, MOD, K), \ | |
5e581dad DB |
1160 | /* 64 bit ALU operations. */ \ |
1161 | /* Register based. */ \ | |
1162 | INSN_3(ALU64, ADD, X), \ | |
1163 | INSN_3(ALU64, SUB, X), \ | |
1164 | INSN_3(ALU64, AND, X), \ | |
1165 | INSN_3(ALU64, OR, X), \ | |
1166 | INSN_3(ALU64, LSH, X), \ | |
1167 | INSN_3(ALU64, RSH, X), \ | |
1168 | INSN_3(ALU64, XOR, X), \ | |
1169 | INSN_3(ALU64, MUL, X), \ | |
1170 | INSN_3(ALU64, MOV, X), \ | |
1171 | INSN_3(ALU64, ARSH, X), \ | |
1172 | INSN_3(ALU64, DIV, X), \ | |
1173 | INSN_3(ALU64, MOD, X), \ | |
1174 | INSN_2(ALU64, NEG), \ | |
1175 | /* Immediate based. */ \ | |
1176 | INSN_3(ALU64, ADD, K), \ | |
1177 | INSN_3(ALU64, SUB, K), \ | |
1178 | INSN_3(ALU64, AND, K), \ | |
1179 | INSN_3(ALU64, OR, K), \ | |
1180 | INSN_3(ALU64, LSH, K), \ | |
1181 | INSN_3(ALU64, RSH, K), \ | |
1182 | INSN_3(ALU64, XOR, K), \ | |
1183 | INSN_3(ALU64, MUL, K), \ | |
1184 | INSN_3(ALU64, MOV, K), \ | |
1185 | INSN_3(ALU64, ARSH, K), \ | |
1186 | INSN_3(ALU64, DIV, K), \ | |
1187 | INSN_3(ALU64, MOD, K), \ | |
1188 | /* Call instruction. */ \ | |
1189 | INSN_2(JMP, CALL), \ | |
1190 | /* Exit instruction. */ \ | |
1191 | INSN_2(JMP, EXIT), \ | |
503a8865 JW |
1192 | /* 32-bit Jump instructions. */ \ |
1193 | /* Register based. */ \ | |
1194 | INSN_3(JMP32, JEQ, X), \ | |
1195 | INSN_3(JMP32, JNE, X), \ | |
1196 | INSN_3(JMP32, JGT, X), \ | |
1197 | INSN_3(JMP32, JLT, X), \ | |
1198 | INSN_3(JMP32, JGE, X), \ | |
1199 | INSN_3(JMP32, JLE, X), \ | |
1200 | INSN_3(JMP32, JSGT, X), \ | |
1201 | INSN_3(JMP32, JSLT, X), \ | |
1202 | INSN_3(JMP32, JSGE, X), \ | |
1203 | INSN_3(JMP32, JSLE, X), \ | |
1204 | INSN_3(JMP32, JSET, X), \ | |
1205 | /* Immediate based. */ \ | |
1206 | INSN_3(JMP32, JEQ, K), \ | |
1207 | INSN_3(JMP32, JNE, K), \ | |
1208 | INSN_3(JMP32, JGT, K), \ | |
1209 | INSN_3(JMP32, JLT, K), \ | |
1210 | INSN_3(JMP32, JGE, K), \ | |
1211 | INSN_3(JMP32, JLE, K), \ | |
1212 | INSN_3(JMP32, JSGT, K), \ | |
1213 | INSN_3(JMP32, JSLT, K), \ | |
1214 | INSN_3(JMP32, JSGE, K), \ | |
1215 | INSN_3(JMP32, JSLE, K), \ | |
1216 | INSN_3(JMP32, JSET, K), \ | |
5e581dad DB |
1217 | /* Jump instructions. */ \ |
1218 | /* Register based. */ \ | |
1219 | INSN_3(JMP, JEQ, X), \ | |
1220 | INSN_3(JMP, JNE, X), \ | |
1221 | INSN_3(JMP, JGT, X), \ | |
1222 | INSN_3(JMP, JLT, X), \ | |
1223 | INSN_3(JMP, JGE, X), \ | |
1224 | INSN_3(JMP, JLE, X), \ | |
1225 | INSN_3(JMP, JSGT, X), \ | |
1226 | INSN_3(JMP, JSLT, X), \ | |
1227 | INSN_3(JMP, JSGE, X), \ | |
1228 | INSN_3(JMP, JSLE, X), \ | |
1229 | INSN_3(JMP, JSET, X), \ | |
1230 | /* Immediate based. */ \ | |
1231 | INSN_3(JMP, JEQ, K), \ | |
1232 | INSN_3(JMP, JNE, K), \ | |
1233 | INSN_3(JMP, JGT, K), \ | |
1234 | INSN_3(JMP, JLT, K), \ | |
1235 | INSN_3(JMP, JGE, K), \ | |
1236 | INSN_3(JMP, JLE, K), \ | |
1237 | INSN_3(JMP, JSGT, K), \ | |
1238 | INSN_3(JMP, JSLT, K), \ | |
1239 | INSN_3(JMP, JSGE, K), \ | |
1240 | INSN_3(JMP, JSLE, K), \ | |
1241 | INSN_3(JMP, JSET, K), \ | |
1242 | INSN_2(JMP, JA), \ | |
1243 | /* Store instructions. */ \ | |
1244 | /* Register based. */ \ | |
1245 | INSN_3(STX, MEM, B), \ | |
1246 | INSN_3(STX, MEM, H), \ | |
1247 | INSN_3(STX, MEM, W), \ | |
1248 | INSN_3(STX, MEM, DW), \ | |
1249 | INSN_3(STX, XADD, W), \ | |
1250 | INSN_3(STX, XADD, DW), \ | |
1251 | /* Immediate based. */ \ | |
1252 | INSN_3(ST, MEM, B), \ | |
1253 | INSN_3(ST, MEM, H), \ | |
1254 | INSN_3(ST, MEM, W), \ | |
1255 | INSN_3(ST, MEM, DW), \ | |
1256 | /* Load instructions. */ \ | |
1257 | /* Register based. */ \ | |
1258 | INSN_3(LDX, MEM, B), \ | |
1259 | INSN_3(LDX, MEM, H), \ | |
1260 | INSN_3(LDX, MEM, W), \ | |
1261 | INSN_3(LDX, MEM, DW), \ | |
1262 | /* Immediate based. */ \ | |
e0cea7ce | 1263 | INSN_3(LD, IMM, DW) |
5e581dad DB |
1264 | |
1265 | bool bpf_opcode_in_insntable(u8 code) | |
1266 | { | |
1267 | #define BPF_INSN_2_TBL(x, y) [BPF_##x | BPF_##y] = true | |
1268 | #define BPF_INSN_3_TBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = true | |
1269 | static const bool public_insntable[256] = { | |
1270 | [0 ... 255] = false, | |
1271 | /* Now overwrite non-defaults ... */ | |
1272 | BPF_INSN_MAP(BPF_INSN_2_TBL, BPF_INSN_3_TBL), | |
e0cea7ce DB |
1273 | /* UAPI exposed, but rewritten opcodes. cBPF carry-over. */ |
1274 | [BPF_LD | BPF_ABS | BPF_B] = true, | |
1275 | [BPF_LD | BPF_ABS | BPF_H] = true, | |
1276 | [BPF_LD | BPF_ABS | BPF_W] = true, | |
1277 | [BPF_LD | BPF_IND | BPF_B] = true, | |
1278 | [BPF_LD | BPF_IND | BPF_H] = true, | |
1279 | [BPF_LD | BPF_IND | BPF_W] = true, | |
5e581dad DB |
1280 | }; |
1281 | #undef BPF_INSN_3_TBL | |
1282 | #undef BPF_INSN_2_TBL | |
1283 | return public_insntable[code]; | |
1284 | } | |
1285 | ||
290af866 | 1286 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
f5bffecd | 1287 | /** |
7ae457c1 | 1288 | * __bpf_prog_run - run eBPF program on a given context |
de1da68d | 1289 | * @regs: is the array of MAX_BPF_EXT_REG eBPF pseudo-registers |
7ae457c1 | 1290 | * @insn: is the array of eBPF instructions |
de1da68d | 1291 | * @stack: is the eBPF storage stack |
f5bffecd | 1292 | * |
7ae457c1 | 1293 | * Decode and execute eBPF instructions. |
f5bffecd | 1294 | */ |
1ea47e01 | 1295 | static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack) |
f5bffecd | 1296 | { |
5e581dad DB |
1297 | #define BPF_INSN_2_LBL(x, y) [BPF_##x | BPF_##y] = &&x##_##y |
1298 | #define BPF_INSN_3_LBL(x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z | |
f5bffecd AS |
1299 | static const void *jumptable[256] = { |
1300 | [0 ... 255] = &&default_label, | |
1301 | /* Now overwrite non-defaults ... */ | |
5e581dad DB |
1302 | BPF_INSN_MAP(BPF_INSN_2_LBL, BPF_INSN_3_LBL), |
1303 | /* Non-UAPI available opcodes. */ | |
1ea47e01 | 1304 | [BPF_JMP | BPF_CALL_ARGS] = &&JMP_CALL_ARGS, |
71189fa9 | 1305 | [BPF_JMP | BPF_TAIL_CALL] = &&JMP_TAIL_CALL, |
f5bffecd | 1306 | }; |
5e581dad DB |
1307 | #undef BPF_INSN_3_LBL |
1308 | #undef BPF_INSN_2_LBL | |
04fd61ab | 1309 | u32 tail_call_cnt = 0; |
f5bffecd AS |
1310 | |
1311 | #define CONT ({ insn++; goto select_insn; }) | |
1312 | #define CONT_JMP ({ insn++; goto select_insn; }) | |
1313 | ||
f5bffecd AS |
1314 | select_insn: |
1315 | goto *jumptable[insn->code]; | |
1316 | ||
1317 | /* ALU */ | |
1318 | #define ALU(OPCODE, OP) \ | |
1319 | ALU64_##OPCODE##_X: \ | |
1320 | DST = DST OP SRC; \ | |
1321 | CONT; \ | |
1322 | ALU_##OPCODE##_X: \ | |
1323 | DST = (u32) DST OP (u32) SRC; \ | |
1324 | CONT; \ | |
1325 | ALU64_##OPCODE##_K: \ | |
1326 | DST = DST OP IMM; \ | |
1327 | CONT; \ | |
1328 | ALU_##OPCODE##_K: \ | |
1329 | DST = (u32) DST OP (u32) IMM; \ | |
1330 | CONT; | |
1331 | ||
1332 | ALU(ADD, +) | |
1333 | ALU(SUB, -) | |
1334 | ALU(AND, &) | |
1335 | ALU(OR, |) | |
1336 | ALU(LSH, <<) | |
1337 | ALU(RSH, >>) | |
1338 | ALU(XOR, ^) | |
1339 | ALU(MUL, *) | |
1340 | #undef ALU | |
1341 | ALU_NEG: | |
1342 | DST = (u32) -DST; | |
1343 | CONT; | |
1344 | ALU64_NEG: | |
1345 | DST = -DST; | |
1346 | CONT; | |
1347 | ALU_MOV_X: | |
1348 | DST = (u32) SRC; | |
1349 | CONT; | |
1350 | ALU_MOV_K: | |
1351 | DST = (u32) IMM; | |
1352 | CONT; | |
1353 | ALU64_MOV_X: | |
1354 | DST = SRC; | |
1355 | CONT; | |
1356 | ALU64_MOV_K: | |
1357 | DST = IMM; | |
1358 | CONT; | |
02ab695b AS |
1359 | LD_IMM_DW: |
1360 | DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32; | |
1361 | insn++; | |
1362 | CONT; | |
2dc6b100 JW |
1363 | ALU_ARSH_X: |
1364 | DST = (u64) (u32) ((*(s32 *) &DST) >> SRC); | |
1365 | CONT; | |
1366 | ALU_ARSH_K: | |
1367 | DST = (u64) (u32) ((*(s32 *) &DST) >> IMM); | |
1368 | CONT; | |
f5bffecd AS |
1369 | ALU64_ARSH_X: |
1370 | (*(s64 *) &DST) >>= SRC; | |
1371 | CONT; | |
1372 | ALU64_ARSH_K: | |
1373 | (*(s64 *) &DST) >>= IMM; | |
1374 | CONT; | |
1375 | ALU64_MOD_X: | |
144cd91c DB |
1376 | div64_u64_rem(DST, SRC, &AX); |
1377 | DST = AX; | |
f5bffecd AS |
1378 | CONT; |
1379 | ALU_MOD_X: | |
144cd91c DB |
1380 | AX = (u32) DST; |
1381 | DST = do_div(AX, (u32) SRC); | |
f5bffecd AS |
1382 | CONT; |
1383 | ALU64_MOD_K: | |
144cd91c DB |
1384 | div64_u64_rem(DST, IMM, &AX); |
1385 | DST = AX; | |
f5bffecd AS |
1386 | CONT; |
1387 | ALU_MOD_K: | |
144cd91c DB |
1388 | AX = (u32) DST; |
1389 | DST = do_div(AX, (u32) IMM); | |
f5bffecd AS |
1390 | CONT; |
1391 | ALU64_DIV_X: | |
876a7ae6 | 1392 | DST = div64_u64(DST, SRC); |
f5bffecd AS |
1393 | CONT; |
1394 | ALU_DIV_X: | |
144cd91c DB |
1395 | AX = (u32) DST; |
1396 | do_div(AX, (u32) SRC); | |
1397 | DST = (u32) AX; | |
f5bffecd AS |
1398 | CONT; |
1399 | ALU64_DIV_K: | |
876a7ae6 | 1400 | DST = div64_u64(DST, IMM); |
f5bffecd AS |
1401 | CONT; |
1402 | ALU_DIV_K: | |
144cd91c DB |
1403 | AX = (u32) DST; |
1404 | do_div(AX, (u32) IMM); | |
1405 | DST = (u32) AX; | |
f5bffecd AS |
1406 | CONT; |
1407 | ALU_END_TO_BE: | |
1408 | switch (IMM) { | |
1409 | case 16: | |
1410 | DST = (__force u16) cpu_to_be16(DST); | |
1411 | break; | |
1412 | case 32: | |
1413 | DST = (__force u32) cpu_to_be32(DST); | |
1414 | break; | |
1415 | case 64: | |
1416 | DST = (__force u64) cpu_to_be64(DST); | |
1417 | break; | |
1418 | } | |
1419 | CONT; | |
1420 | ALU_END_TO_LE: | |
1421 | switch (IMM) { | |
1422 | case 16: | |
1423 | DST = (__force u16) cpu_to_le16(DST); | |
1424 | break; | |
1425 | case 32: | |
1426 | DST = (__force u32) cpu_to_le32(DST); | |
1427 | break; | |
1428 | case 64: | |
1429 | DST = (__force u64) cpu_to_le64(DST); | |
1430 | break; | |
1431 | } | |
1432 | CONT; | |
1433 | ||
1434 | /* CALL */ | |
1435 | JMP_CALL: | |
1436 | /* Function call scratches BPF_R1-BPF_R5 registers, | |
1437 | * preserves BPF_R6-BPF_R9, and stores return value | |
1438 | * into BPF_R0. | |
1439 | */ | |
1440 | BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3, | |
1441 | BPF_R4, BPF_R5); | |
1442 | CONT; | |
1443 | ||
1ea47e01 AS |
1444 | JMP_CALL_ARGS: |
1445 | BPF_R0 = (__bpf_call_base_args + insn->imm)(BPF_R1, BPF_R2, | |
1446 | BPF_R3, BPF_R4, | |
1447 | BPF_R5, | |
1448 | insn + insn->off + 1); | |
1449 | CONT; | |
1450 | ||
04fd61ab AS |
1451 | JMP_TAIL_CALL: { |
1452 | struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2; | |
1453 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
1454 | struct bpf_prog *prog; | |
90caccdd | 1455 | u32 index = BPF_R3; |
04fd61ab AS |
1456 | |
1457 | if (unlikely(index >= array->map.max_entries)) | |
1458 | goto out; | |
04fd61ab AS |
1459 | if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) |
1460 | goto out; | |
1461 | ||
1462 | tail_call_cnt++; | |
1463 | ||
2a36f0b9 | 1464 | prog = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 1465 | if (!prog) |
04fd61ab AS |
1466 | goto out; |
1467 | ||
c4675f93 DB |
1468 | /* ARG1 at this point is guaranteed to point to CTX from |
1469 | * the verifier side due to the fact that the tail call is | |
1470 | * handeled like a helper, that is, bpf_tail_call_proto, | |
1471 | * where arg1_type is ARG_PTR_TO_CTX. | |
1472 | */ | |
04fd61ab AS |
1473 | insn = prog->insnsi; |
1474 | goto select_insn; | |
1475 | out: | |
1476 | CONT; | |
1477 | } | |
f5bffecd AS |
1478 | JMP_JA: |
1479 | insn += insn->off; | |
1480 | CONT; | |
f5bffecd AS |
1481 | JMP_EXIT: |
1482 | return BPF_R0; | |
503a8865 JW |
1483 | /* JMP */ |
1484 | #define COND_JMP(SIGN, OPCODE, CMP_OP) \ | |
1485 | JMP_##OPCODE##_X: \ | |
1486 | if ((SIGN##64) DST CMP_OP (SIGN##64) SRC) { \ | |
1487 | insn += insn->off; \ | |
1488 | CONT_JMP; \ | |
1489 | } \ | |
1490 | CONT; \ | |
1491 | JMP32_##OPCODE##_X: \ | |
1492 | if ((SIGN##32) DST CMP_OP (SIGN##32) SRC) { \ | |
1493 | insn += insn->off; \ | |
1494 | CONT_JMP; \ | |
1495 | } \ | |
1496 | CONT; \ | |
1497 | JMP_##OPCODE##_K: \ | |
1498 | if ((SIGN##64) DST CMP_OP (SIGN##64) IMM) { \ | |
1499 | insn += insn->off; \ | |
1500 | CONT_JMP; \ | |
1501 | } \ | |
1502 | CONT; \ | |
1503 | JMP32_##OPCODE##_K: \ | |
1504 | if ((SIGN##32) DST CMP_OP (SIGN##32) IMM) { \ | |
1505 | insn += insn->off; \ | |
1506 | CONT_JMP; \ | |
1507 | } \ | |
1508 | CONT; | |
1509 | COND_JMP(u, JEQ, ==) | |
1510 | COND_JMP(u, JNE, !=) | |
1511 | COND_JMP(u, JGT, >) | |
1512 | COND_JMP(u, JLT, <) | |
1513 | COND_JMP(u, JGE, >=) | |
1514 | COND_JMP(u, JLE, <=) | |
1515 | COND_JMP(u, JSET, &) | |
1516 | COND_JMP(s, JSGT, >) | |
1517 | COND_JMP(s, JSLT, <) | |
1518 | COND_JMP(s, JSGE, >=) | |
1519 | COND_JMP(s, JSLE, <=) | |
1520 | #undef COND_JMP | |
f5bffecd AS |
1521 | /* STX and ST and LDX*/ |
1522 | #define LDST(SIZEOP, SIZE) \ | |
1523 | STX_MEM_##SIZEOP: \ | |
1524 | *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \ | |
1525 | CONT; \ | |
1526 | ST_MEM_##SIZEOP: \ | |
1527 | *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \ | |
1528 | CONT; \ | |
1529 | LDX_MEM_##SIZEOP: \ | |
1530 | DST = *(SIZE *)(unsigned long) (SRC + insn->off); \ | |
1531 | CONT; | |
1532 | ||
1533 | LDST(B, u8) | |
1534 | LDST(H, u16) | |
1535 | LDST(W, u32) | |
1536 | LDST(DW, u64) | |
1537 | #undef LDST | |
1538 | STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */ | |
1539 | atomic_add((u32) SRC, (atomic_t *)(unsigned long) | |
1540 | (DST + insn->off)); | |
1541 | CONT; | |
1542 | STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */ | |
1543 | atomic64_add((u64) SRC, (atomic64_t *)(unsigned long) | |
1544 | (DST + insn->off)); | |
1545 | CONT; | |
f5bffecd AS |
1546 | |
1547 | default_label: | |
5e581dad DB |
1548 | /* If we ever reach this, we have a bug somewhere. Die hard here |
1549 | * instead of just returning 0; we could be somewhere in a subprog, | |
1550 | * so execution could continue otherwise which we do /not/ want. | |
1551 | * | |
1552 | * Note, verifier whitelists all opcodes in bpf_opcode_in_insntable(). | |
1553 | */ | |
1554 | pr_warn("BPF interpreter: unknown opcode %02x\n", insn->code); | |
1555 | BUG_ON(1); | |
f5bffecd AS |
1556 | return 0; |
1557 | } | |
f696b8f4 AS |
1558 | STACK_FRAME_NON_STANDARD(___bpf_prog_run); /* jump table */ |
1559 | ||
b870aa90 AS |
1560 | #define PROG_NAME(stack_size) __bpf_prog_run##stack_size |
1561 | #define DEFINE_BPF_PROG_RUN(stack_size) \ | |
1562 | static unsigned int PROG_NAME(stack_size)(const void *ctx, const struct bpf_insn *insn) \ | |
1563 | { \ | |
1564 | u64 stack[stack_size / sizeof(u64)]; \ | |
144cd91c | 1565 | u64 regs[MAX_BPF_EXT_REG]; \ |
b870aa90 AS |
1566 | \ |
1567 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ | |
1568 | ARG1 = (u64) (unsigned long) ctx; \ | |
1569 | return ___bpf_prog_run(regs, insn, stack); \ | |
f696b8f4 | 1570 | } |
f5bffecd | 1571 | |
1ea47e01 AS |
1572 | #define PROG_NAME_ARGS(stack_size) __bpf_prog_run_args##stack_size |
1573 | #define DEFINE_BPF_PROG_RUN_ARGS(stack_size) \ | |
1574 | static u64 PROG_NAME_ARGS(stack_size)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, \ | |
1575 | const struct bpf_insn *insn) \ | |
1576 | { \ | |
1577 | u64 stack[stack_size / sizeof(u64)]; \ | |
144cd91c | 1578 | u64 regs[MAX_BPF_EXT_REG]; \ |
1ea47e01 AS |
1579 | \ |
1580 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; \ | |
1581 | BPF_R1 = r1; \ | |
1582 | BPF_R2 = r2; \ | |
1583 | BPF_R3 = r3; \ | |
1584 | BPF_R4 = r4; \ | |
1585 | BPF_R5 = r5; \ | |
1586 | return ___bpf_prog_run(regs, insn, stack); \ | |
1587 | } | |
1588 | ||
b870aa90 AS |
1589 | #define EVAL1(FN, X) FN(X) |
1590 | #define EVAL2(FN, X, Y...) FN(X) EVAL1(FN, Y) | |
1591 | #define EVAL3(FN, X, Y...) FN(X) EVAL2(FN, Y) | |
1592 | #define EVAL4(FN, X, Y...) FN(X) EVAL3(FN, Y) | |
1593 | #define EVAL5(FN, X, Y...) FN(X) EVAL4(FN, Y) | |
1594 | #define EVAL6(FN, X, Y...) FN(X) EVAL5(FN, Y) | |
1595 | ||
1596 | EVAL6(DEFINE_BPF_PROG_RUN, 32, 64, 96, 128, 160, 192); | |
1597 | EVAL6(DEFINE_BPF_PROG_RUN, 224, 256, 288, 320, 352, 384); | |
1598 | EVAL4(DEFINE_BPF_PROG_RUN, 416, 448, 480, 512); | |
1599 | ||
1ea47e01 AS |
1600 | EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 32, 64, 96, 128, 160, 192); |
1601 | EVAL6(DEFINE_BPF_PROG_RUN_ARGS, 224, 256, 288, 320, 352, 384); | |
1602 | EVAL4(DEFINE_BPF_PROG_RUN_ARGS, 416, 448, 480, 512); | |
1603 | ||
b870aa90 AS |
1604 | #define PROG_NAME_LIST(stack_size) PROG_NAME(stack_size), |
1605 | ||
1606 | static unsigned int (*interpreters[])(const void *ctx, | |
1607 | const struct bpf_insn *insn) = { | |
1608 | EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) | |
1609 | EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) | |
1610 | EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) | |
1611 | }; | |
1ea47e01 AS |
1612 | #undef PROG_NAME_LIST |
1613 | #define PROG_NAME_LIST(stack_size) PROG_NAME_ARGS(stack_size), | |
1614 | static u64 (*interpreters_args[])(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5, | |
1615 | const struct bpf_insn *insn) = { | |
1616 | EVAL6(PROG_NAME_LIST, 32, 64, 96, 128, 160, 192) | |
1617 | EVAL6(PROG_NAME_LIST, 224, 256, 288, 320, 352, 384) | |
1618 | EVAL4(PROG_NAME_LIST, 416, 448, 480, 512) | |
1619 | }; | |
1620 | #undef PROG_NAME_LIST | |
1621 | ||
1622 | void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth) | |
1623 | { | |
1624 | stack_depth = max_t(u32, stack_depth, 1); | |
1625 | insn->off = (s16) insn->imm; | |
1626 | insn->imm = interpreters_args[(round_up(stack_depth, 32) / 32) - 1] - | |
1627 | __bpf_call_base_args; | |
1628 | insn->code = BPF_JMP | BPF_CALL_ARGS; | |
1629 | } | |
b870aa90 | 1630 | |
290af866 | 1631 | #else |
fa9dd599 DB |
1632 | static unsigned int __bpf_prog_ret0_warn(const void *ctx, |
1633 | const struct bpf_insn *insn) | |
290af866 | 1634 | { |
fa9dd599 DB |
1635 | /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON |
1636 | * is not working properly, so warn about it! | |
1637 | */ | |
1638 | WARN_ON_ONCE(1); | |
290af866 AS |
1639 | return 0; |
1640 | } | |
1641 | #endif | |
1642 | ||
3324b584 DB |
1643 | bool bpf_prog_array_compatible(struct bpf_array *array, |
1644 | const struct bpf_prog *fp) | |
04fd61ab | 1645 | { |
9802d865 JB |
1646 | if (fp->kprobe_override) |
1647 | return false; | |
1648 | ||
3324b584 DB |
1649 | if (!array->owner_prog_type) { |
1650 | /* There's no owner yet where we could check for | |
1651 | * compatibility. | |
1652 | */ | |
04fd61ab AS |
1653 | array->owner_prog_type = fp->type; |
1654 | array->owner_jited = fp->jited; | |
3324b584 DB |
1655 | |
1656 | return true; | |
04fd61ab | 1657 | } |
3324b584 DB |
1658 | |
1659 | return array->owner_prog_type == fp->type && | |
1660 | array->owner_jited == fp->jited; | |
04fd61ab AS |
1661 | } |
1662 | ||
3324b584 | 1663 | static int bpf_check_tail_call(const struct bpf_prog *fp) |
04fd61ab AS |
1664 | { |
1665 | struct bpf_prog_aux *aux = fp->aux; | |
1666 | int i; | |
1667 | ||
1668 | for (i = 0; i < aux->used_map_cnt; i++) { | |
3324b584 | 1669 | struct bpf_map *map = aux->used_maps[i]; |
04fd61ab | 1670 | struct bpf_array *array; |
04fd61ab | 1671 | |
04fd61ab AS |
1672 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) |
1673 | continue; | |
3324b584 | 1674 | |
04fd61ab AS |
1675 | array = container_of(map, struct bpf_array, map); |
1676 | if (!bpf_prog_array_compatible(array, fp)) | |
1677 | return -EINVAL; | |
1678 | } | |
1679 | ||
1680 | return 0; | |
1681 | } | |
1682 | ||
9facc336 DB |
1683 | static void bpf_prog_select_func(struct bpf_prog *fp) |
1684 | { | |
1685 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1686 | u32 stack_depth = max_t(u32, fp->aux->stack_depth, 1); | |
1687 | ||
1688 | fp->bpf_func = interpreters[(round_up(stack_depth, 32) / 32) - 1]; | |
1689 | #else | |
1690 | fp->bpf_func = __bpf_prog_ret0_warn; | |
1691 | #endif | |
1692 | } | |
1693 | ||
f5bffecd | 1694 | /** |
3324b584 | 1695 | * bpf_prog_select_runtime - select exec runtime for BPF program |
7ae457c1 | 1696 | * @fp: bpf_prog populated with internal BPF program |
d1c55ab5 | 1697 | * @err: pointer to error variable |
f5bffecd | 1698 | * |
3324b584 DB |
1699 | * Try to JIT eBPF program, if JIT is not available, use interpreter. |
1700 | * The BPF program will be executed via BPF_PROG_RUN() macro. | |
f5bffecd | 1701 | */ |
d1c55ab5 | 1702 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err) |
f5bffecd | 1703 | { |
9facc336 DB |
1704 | /* In case of BPF to BPF calls, verifier did all the prep |
1705 | * work with regards to JITing, etc. | |
1706 | */ | |
1707 | if (fp->bpf_func) | |
1708 | goto finalize; | |
8007e40a | 1709 | |
9facc336 | 1710 | bpf_prog_select_func(fp); |
f5bffecd | 1711 | |
d1c55ab5 DB |
1712 | /* eBPF JITs can rewrite the program in case constant |
1713 | * blinding is active. However, in case of error during | |
1714 | * blinding, bpf_int_jit_compile() must always return a | |
1715 | * valid program, which in this case would simply not | |
1716 | * be JITed, but falls back to the interpreter. | |
1717 | */ | |
ab3f0063 | 1718 | if (!bpf_prog_is_dev_bound(fp->aux)) { |
c454a46b MKL |
1719 | *err = bpf_prog_alloc_jited_linfo(fp); |
1720 | if (*err) | |
1721 | return fp; | |
1722 | ||
ab3f0063 | 1723 | fp = bpf_int_jit_compile(fp); |
290af866 | 1724 | if (!fp->jited) { |
c454a46b MKL |
1725 | bpf_prog_free_jited_linfo(fp); |
1726 | #ifdef CONFIG_BPF_JIT_ALWAYS_ON | |
290af866 AS |
1727 | *err = -ENOTSUPP; |
1728 | return fp; | |
290af866 | 1729 | #endif |
c454a46b MKL |
1730 | } else { |
1731 | bpf_prog_free_unused_jited_linfo(fp); | |
1732 | } | |
ab3f0063 JK |
1733 | } else { |
1734 | *err = bpf_prog_offload_compile(fp); | |
1735 | if (*err) | |
1736 | return fp; | |
1737 | } | |
9facc336 DB |
1738 | |
1739 | finalize: | |
60a3b225 | 1740 | bpf_prog_lock_ro(fp); |
04fd61ab | 1741 | |
3324b584 DB |
1742 | /* The tail call compatibility check can only be done at |
1743 | * this late stage as we need to determine, if we deal | |
1744 | * with JITed or non JITed program concatenations and not | |
1745 | * all eBPF JITs might immediately support all features. | |
1746 | */ | |
d1c55ab5 | 1747 | *err = bpf_check_tail_call(fp); |
85782e03 | 1748 | |
d1c55ab5 | 1749 | return fp; |
f5bffecd | 1750 | } |
7ae457c1 | 1751 | EXPORT_SYMBOL_GPL(bpf_prog_select_runtime); |
f5bffecd | 1752 | |
e87c6bc3 YS |
1753 | static unsigned int __bpf_prog_ret1(const void *ctx, |
1754 | const struct bpf_insn *insn) | |
1755 | { | |
1756 | return 1; | |
1757 | } | |
1758 | ||
1759 | static struct bpf_prog_dummy { | |
1760 | struct bpf_prog prog; | |
1761 | } dummy_bpf_prog = { | |
1762 | .prog = { | |
1763 | .bpf_func = __bpf_prog_ret1, | |
1764 | }, | |
1765 | }; | |
1766 | ||
324bda9e AS |
1767 | /* to avoid allocating empty bpf_prog_array for cgroups that |
1768 | * don't have bpf program attached use one global 'empty_prog_array' | |
1769 | * It will not be modified the caller of bpf_prog_array_alloc() | |
1770 | * (since caller requested prog_cnt == 0) | |
1771 | * that pointer should be 'freed' by bpf_prog_array_free() | |
1772 | */ | |
1773 | static struct { | |
1774 | struct bpf_prog_array hdr; | |
1775 | struct bpf_prog *null_prog; | |
1776 | } empty_prog_array = { | |
1777 | .null_prog = NULL, | |
1778 | }; | |
1779 | ||
d29ab6e1 | 1780 | struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags) |
324bda9e AS |
1781 | { |
1782 | if (prog_cnt) | |
1783 | return kzalloc(sizeof(struct bpf_prog_array) + | |
394e40a2 RG |
1784 | sizeof(struct bpf_prog_array_item) * |
1785 | (prog_cnt + 1), | |
324bda9e AS |
1786 | flags); |
1787 | ||
1788 | return &empty_prog_array.hdr; | |
1789 | } | |
1790 | ||
1791 | void bpf_prog_array_free(struct bpf_prog_array __rcu *progs) | |
1792 | { | |
1793 | if (!progs || | |
1794 | progs == (struct bpf_prog_array __rcu *)&empty_prog_array.hdr) | |
1795 | return; | |
1796 | kfree_rcu(progs, rcu); | |
1797 | } | |
1798 | ||
394e40a2 | 1799 | int bpf_prog_array_length(struct bpf_prog_array __rcu *array) |
468e2f64 | 1800 | { |
394e40a2 | 1801 | struct bpf_prog_array_item *item; |
468e2f64 AS |
1802 | u32 cnt = 0; |
1803 | ||
1804 | rcu_read_lock(); | |
394e40a2 RG |
1805 | item = rcu_dereference(array)->items; |
1806 | for (; item->prog; item++) | |
1807 | if (item->prog != &dummy_bpf_prog.prog) | |
c8c088ba | 1808 | cnt++; |
468e2f64 AS |
1809 | rcu_read_unlock(); |
1810 | return cnt; | |
1811 | } | |
1812 | ||
394e40a2 RG |
1813 | |
1814 | static bool bpf_prog_array_copy_core(struct bpf_prog_array __rcu *array, | |
3a38bb98 YS |
1815 | u32 *prog_ids, |
1816 | u32 request_cnt) | |
1817 | { | |
394e40a2 | 1818 | struct bpf_prog_array_item *item; |
3a38bb98 YS |
1819 | int i = 0; |
1820 | ||
965931e3 | 1821 | item = rcu_dereference_check(array, 1)->items; |
394e40a2 RG |
1822 | for (; item->prog; item++) { |
1823 | if (item->prog == &dummy_bpf_prog.prog) | |
3a38bb98 | 1824 | continue; |
394e40a2 | 1825 | prog_ids[i] = item->prog->aux->id; |
3a38bb98 | 1826 | if (++i == request_cnt) { |
394e40a2 | 1827 | item++; |
3a38bb98 YS |
1828 | break; |
1829 | } | |
1830 | } | |
1831 | ||
394e40a2 | 1832 | return !!(item->prog); |
3a38bb98 YS |
1833 | } |
1834 | ||
394e40a2 | 1835 | int bpf_prog_array_copy_to_user(struct bpf_prog_array __rcu *array, |
468e2f64 AS |
1836 | __u32 __user *prog_ids, u32 cnt) |
1837 | { | |
0911287c | 1838 | unsigned long err = 0; |
0911287c | 1839 | bool nospc; |
3a38bb98 | 1840 | u32 *ids; |
0911287c AS |
1841 | |
1842 | /* users of this function are doing: | |
1843 | * cnt = bpf_prog_array_length(); | |
1844 | * if (cnt > 0) | |
1845 | * bpf_prog_array_copy_to_user(..., cnt); | |
1846 | * so below kcalloc doesn't need extra cnt > 0 check, but | |
1847 | * bpf_prog_array_length() releases rcu lock and | |
1848 | * prog array could have been swapped with empty or larger array, | |
1849 | * so always copy 'cnt' prog_ids to the user. | |
1850 | * In a rare race the user will see zero prog_ids | |
1851 | */ | |
9c481b90 | 1852 | ids = kcalloc(cnt, sizeof(u32), GFP_USER | __GFP_NOWARN); |
0911287c AS |
1853 | if (!ids) |
1854 | return -ENOMEM; | |
468e2f64 | 1855 | rcu_read_lock(); |
394e40a2 | 1856 | nospc = bpf_prog_array_copy_core(array, ids, cnt); |
468e2f64 | 1857 | rcu_read_unlock(); |
0911287c AS |
1858 | err = copy_to_user(prog_ids, ids, cnt * sizeof(u32)); |
1859 | kfree(ids); | |
1860 | if (err) | |
1861 | return -EFAULT; | |
1862 | if (nospc) | |
468e2f64 AS |
1863 | return -ENOSPC; |
1864 | return 0; | |
1865 | } | |
1866 | ||
394e40a2 | 1867 | void bpf_prog_array_delete_safe(struct bpf_prog_array __rcu *array, |
e87c6bc3 YS |
1868 | struct bpf_prog *old_prog) |
1869 | { | |
394e40a2 | 1870 | struct bpf_prog_array_item *item = array->items; |
e87c6bc3 | 1871 | |
394e40a2 RG |
1872 | for (; item->prog; item++) |
1873 | if (item->prog == old_prog) { | |
1874 | WRITE_ONCE(item->prog, &dummy_bpf_prog.prog); | |
e87c6bc3 YS |
1875 | break; |
1876 | } | |
1877 | } | |
1878 | ||
1879 | int bpf_prog_array_copy(struct bpf_prog_array __rcu *old_array, | |
1880 | struct bpf_prog *exclude_prog, | |
1881 | struct bpf_prog *include_prog, | |
1882 | struct bpf_prog_array **new_array) | |
1883 | { | |
1884 | int new_prog_cnt, carry_prog_cnt = 0; | |
394e40a2 | 1885 | struct bpf_prog_array_item *existing; |
e87c6bc3 | 1886 | struct bpf_prog_array *array; |
170a7e3e | 1887 | bool found_exclude = false; |
e87c6bc3 YS |
1888 | int new_prog_idx = 0; |
1889 | ||
1890 | /* Figure out how many existing progs we need to carry over to | |
1891 | * the new array. | |
1892 | */ | |
1893 | if (old_array) { | |
394e40a2 RG |
1894 | existing = old_array->items; |
1895 | for (; existing->prog; existing++) { | |
1896 | if (existing->prog == exclude_prog) { | |
170a7e3e SY |
1897 | found_exclude = true; |
1898 | continue; | |
1899 | } | |
394e40a2 | 1900 | if (existing->prog != &dummy_bpf_prog.prog) |
e87c6bc3 | 1901 | carry_prog_cnt++; |
394e40a2 | 1902 | if (existing->prog == include_prog) |
e87c6bc3 YS |
1903 | return -EEXIST; |
1904 | } | |
1905 | } | |
1906 | ||
170a7e3e SY |
1907 | if (exclude_prog && !found_exclude) |
1908 | return -ENOENT; | |
1909 | ||
e87c6bc3 YS |
1910 | /* How many progs (not NULL) will be in the new array? */ |
1911 | new_prog_cnt = carry_prog_cnt; | |
1912 | if (include_prog) | |
1913 | new_prog_cnt += 1; | |
1914 | ||
1915 | /* Do we have any prog (not NULL) in the new array? */ | |
1916 | if (!new_prog_cnt) { | |
1917 | *new_array = NULL; | |
1918 | return 0; | |
1919 | } | |
1920 | ||
1921 | /* +1 as the end of prog_array is marked with NULL */ | |
1922 | array = bpf_prog_array_alloc(new_prog_cnt + 1, GFP_KERNEL); | |
1923 | if (!array) | |
1924 | return -ENOMEM; | |
1925 | ||
1926 | /* Fill in the new prog array */ | |
1927 | if (carry_prog_cnt) { | |
394e40a2 RG |
1928 | existing = old_array->items; |
1929 | for (; existing->prog; existing++) | |
1930 | if (existing->prog != exclude_prog && | |
1931 | existing->prog != &dummy_bpf_prog.prog) { | |
1932 | array->items[new_prog_idx++].prog = | |
1933 | existing->prog; | |
1934 | } | |
e87c6bc3 YS |
1935 | } |
1936 | if (include_prog) | |
394e40a2 RG |
1937 | array->items[new_prog_idx++].prog = include_prog; |
1938 | array->items[new_prog_idx].prog = NULL; | |
e87c6bc3 YS |
1939 | *new_array = array; |
1940 | return 0; | |
1941 | } | |
1942 | ||
f371b304 | 1943 | int bpf_prog_array_copy_info(struct bpf_prog_array __rcu *array, |
3a38bb98 YS |
1944 | u32 *prog_ids, u32 request_cnt, |
1945 | u32 *prog_cnt) | |
f371b304 YS |
1946 | { |
1947 | u32 cnt = 0; | |
1948 | ||
1949 | if (array) | |
1950 | cnt = bpf_prog_array_length(array); | |
1951 | ||
3a38bb98 | 1952 | *prog_cnt = cnt; |
f371b304 YS |
1953 | |
1954 | /* return early if user requested only program count or nothing to copy */ | |
1955 | if (!request_cnt || !cnt) | |
1956 | return 0; | |
1957 | ||
3a38bb98 | 1958 | /* this function is called under trace/bpf_trace.c: bpf_event_mutex */ |
394e40a2 | 1959 | return bpf_prog_array_copy_core(array, prog_ids, request_cnt) ? -ENOSPC |
3a38bb98 | 1960 | : 0; |
f371b304 YS |
1961 | } |
1962 | ||
60a3b225 DB |
1963 | static void bpf_prog_free_deferred(struct work_struct *work) |
1964 | { | |
09756af4 | 1965 | struct bpf_prog_aux *aux; |
1c2a088a | 1966 | int i; |
60a3b225 | 1967 | |
09756af4 | 1968 | aux = container_of(work, struct bpf_prog_aux, work); |
ab3f0063 JK |
1969 | if (bpf_prog_is_dev_bound(aux)) |
1970 | bpf_prog_offload_destroy(aux->prog); | |
c195651e YS |
1971 | #ifdef CONFIG_PERF_EVENTS |
1972 | if (aux->prog->has_callchain_buf) | |
1973 | put_callchain_buffers(); | |
1974 | #endif | |
1c2a088a AS |
1975 | for (i = 0; i < aux->func_cnt; i++) |
1976 | bpf_jit_free(aux->func[i]); | |
1977 | if (aux->func_cnt) { | |
1978 | kfree(aux->func); | |
1979 | bpf_prog_unlock_free(aux->prog); | |
1980 | } else { | |
1981 | bpf_jit_free(aux->prog); | |
1982 | } | |
60a3b225 DB |
1983 | } |
1984 | ||
1985 | /* Free internal BPF program */ | |
7ae457c1 | 1986 | void bpf_prog_free(struct bpf_prog *fp) |
f5bffecd | 1987 | { |
09756af4 | 1988 | struct bpf_prog_aux *aux = fp->aux; |
60a3b225 | 1989 | |
09756af4 | 1990 | INIT_WORK(&aux->work, bpf_prog_free_deferred); |
09756af4 | 1991 | schedule_work(&aux->work); |
f5bffecd | 1992 | } |
7ae457c1 | 1993 | EXPORT_SYMBOL_GPL(bpf_prog_free); |
f89b7755 | 1994 | |
3ad00405 DB |
1995 | /* RNG for unpriviledged user space with separated state from prandom_u32(). */ |
1996 | static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state); | |
1997 | ||
1998 | void bpf_user_rnd_init_once(void) | |
1999 | { | |
2000 | prandom_init_once(&bpf_user_rnd_state); | |
2001 | } | |
2002 | ||
f3694e00 | 2003 | BPF_CALL_0(bpf_user_rnd_u32) |
3ad00405 DB |
2004 | { |
2005 | /* Should someone ever have the rather unwise idea to use some | |
2006 | * of the registers passed into this function, then note that | |
2007 | * this function is called from native eBPF and classic-to-eBPF | |
2008 | * transformations. Register assignments from both sides are | |
2009 | * different, f.e. classic always sets fn(ctx, A, X) here. | |
2010 | */ | |
2011 | struct rnd_state *state; | |
2012 | u32 res; | |
2013 | ||
2014 | state = &get_cpu_var(bpf_user_rnd_state); | |
2015 | res = prandom_u32_state(state); | |
b761fe22 | 2016 | put_cpu_var(bpf_user_rnd_state); |
3ad00405 DB |
2017 | |
2018 | return res; | |
2019 | } | |
2020 | ||
3ba67dab DB |
2021 | /* Weak definitions of helper functions in case we don't have bpf syscall. */ |
2022 | const struct bpf_func_proto bpf_map_lookup_elem_proto __weak; | |
2023 | const struct bpf_func_proto bpf_map_update_elem_proto __weak; | |
2024 | const struct bpf_func_proto bpf_map_delete_elem_proto __weak; | |
f1a2e44a MV |
2025 | const struct bpf_func_proto bpf_map_push_elem_proto __weak; |
2026 | const struct bpf_func_proto bpf_map_pop_elem_proto __weak; | |
2027 | const struct bpf_func_proto bpf_map_peek_elem_proto __weak; | |
d83525ca AS |
2028 | const struct bpf_func_proto bpf_spin_lock_proto __weak; |
2029 | const struct bpf_func_proto bpf_spin_unlock_proto __weak; | |
3ba67dab | 2030 | |
03e69b50 | 2031 | const struct bpf_func_proto bpf_get_prandom_u32_proto __weak; |
c04167ce | 2032 | const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak; |
2d0e30c3 | 2033 | const struct bpf_func_proto bpf_get_numa_node_id_proto __weak; |
17ca8cbf | 2034 | const struct bpf_func_proto bpf_ktime_get_ns_proto __weak; |
bd570ff9 | 2035 | |
ffeedafb AS |
2036 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak; |
2037 | const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak; | |
2038 | const struct bpf_func_proto bpf_get_current_comm_proto __weak; | |
bf6fa2c8 | 2039 | const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak; |
cd339431 | 2040 | const struct bpf_func_proto bpf_get_local_storage_proto __weak; |
bd570ff9 | 2041 | |
0756ea3e AS |
2042 | const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void) |
2043 | { | |
2044 | return NULL; | |
2045 | } | |
03e69b50 | 2046 | |
555c8a86 DB |
2047 | u64 __weak |
2048 | bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, | |
2049 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 2050 | { |
555c8a86 | 2051 | return -ENOTSUPP; |
bd570ff9 | 2052 | } |
6cb5fb38 | 2053 | EXPORT_SYMBOL_GPL(bpf_event_output); |
bd570ff9 | 2054 | |
3324b584 DB |
2055 | /* Always built-in helper functions. */ |
2056 | const struct bpf_func_proto bpf_tail_call_proto = { | |
2057 | .func = NULL, | |
2058 | .gpl_only = false, | |
2059 | .ret_type = RET_VOID, | |
2060 | .arg1_type = ARG_PTR_TO_CTX, | |
2061 | .arg2_type = ARG_CONST_MAP_PTR, | |
2062 | .arg3_type = ARG_ANYTHING, | |
2063 | }; | |
2064 | ||
9383191d DB |
2065 | /* Stub for JITs that only support cBPF. eBPF programs are interpreted. |
2066 | * It is encouraged to implement bpf_int_jit_compile() instead, so that | |
2067 | * eBPF and implicitly also cBPF can get JITed! | |
2068 | */ | |
d1c55ab5 | 2069 | struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog) |
3324b584 | 2070 | { |
d1c55ab5 | 2071 | return prog; |
3324b584 DB |
2072 | } |
2073 | ||
9383191d DB |
2074 | /* Stub for JITs that support eBPF. All cBPF code gets transformed into |
2075 | * eBPF by the kernel and is later compiled by bpf_int_jit_compile(). | |
2076 | */ | |
2077 | void __weak bpf_jit_compile(struct bpf_prog *prog) | |
2078 | { | |
2079 | } | |
2080 | ||
17bedab2 | 2081 | bool __weak bpf_helper_changes_pkt_data(void *func) |
969bf05e AS |
2082 | { |
2083 | return false; | |
2084 | } | |
2085 | ||
f89b7755 AS |
2086 | /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call |
2087 | * skb_copy_bits(), so provide a weak definition of it for NET-less config. | |
2088 | */ | |
2089 | int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to, | |
2090 | int len) | |
2091 | { | |
2092 | return -EFAULT; | |
2093 | } | |
a67edbf4 | 2094 | |
492ecee8 AS |
2095 | DEFINE_STATIC_KEY_FALSE(bpf_stats_enabled_key); |
2096 | EXPORT_SYMBOL(bpf_stats_enabled_key); | |
2097 | int sysctl_bpf_stats_enabled __read_mostly; | |
2098 | ||
a67edbf4 DB |
2099 | /* All definitions of tracepoints related to BPF. */ |
2100 | #define CREATE_TRACE_POINTS | |
2101 | #include <linux/bpf_trace.h> | |
2102 | ||
2103 | EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception); |