]>
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 | |
f5bffecd AS |
24 | #include <linux/filter.h> |
25 | #include <linux/skbuff.h> | |
60a3b225 | 26 | #include <linux/vmalloc.h> |
738cbe72 DB |
27 | #include <linux/random.h> |
28 | #include <linux/moduleloader.h> | |
09756af4 | 29 | #include <linux/bpf.h> |
39853cc0 | 30 | #include <linux/frame.h> |
f5bffecd | 31 | |
3324b584 DB |
32 | #include <asm/unaligned.h> |
33 | ||
f5bffecd AS |
34 | /* Registers */ |
35 | #define BPF_R0 regs[BPF_REG_0] | |
36 | #define BPF_R1 regs[BPF_REG_1] | |
37 | #define BPF_R2 regs[BPF_REG_2] | |
38 | #define BPF_R3 regs[BPF_REG_3] | |
39 | #define BPF_R4 regs[BPF_REG_4] | |
40 | #define BPF_R5 regs[BPF_REG_5] | |
41 | #define BPF_R6 regs[BPF_REG_6] | |
42 | #define BPF_R7 regs[BPF_REG_7] | |
43 | #define BPF_R8 regs[BPF_REG_8] | |
44 | #define BPF_R9 regs[BPF_REG_9] | |
45 | #define BPF_R10 regs[BPF_REG_10] | |
46 | ||
47 | /* Named registers */ | |
48 | #define DST regs[insn->dst_reg] | |
49 | #define SRC regs[insn->src_reg] | |
50 | #define FP regs[BPF_REG_FP] | |
51 | #define ARG1 regs[BPF_REG_ARG1] | |
52 | #define CTX regs[BPF_REG_CTX] | |
53 | #define IMM insn->imm | |
54 | ||
55 | /* No hurry in this branch | |
56 | * | |
57 | * Exported for the bpf jit load helper. | |
58 | */ | |
59 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size) | |
60 | { | |
61 | u8 *ptr = NULL; | |
62 | ||
63 | if (k >= SKF_NET_OFF) | |
64 | ptr = skb_network_header(skb) + k - SKF_NET_OFF; | |
65 | else if (k >= SKF_LL_OFF) | |
66 | ptr = skb_mac_header(skb) + k - SKF_LL_OFF; | |
3324b584 | 67 | |
f5bffecd AS |
68 | if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb)) |
69 | return ptr; | |
70 | ||
71 | return NULL; | |
72 | } | |
73 | ||
60a3b225 DB |
74 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags) |
75 | { | |
76 | gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO | | |
77 | gfp_extra_flags; | |
09756af4 | 78 | struct bpf_prog_aux *aux; |
60a3b225 DB |
79 | struct bpf_prog *fp; |
80 | ||
81 | size = round_up(size, PAGE_SIZE); | |
82 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); | |
83 | if (fp == NULL) | |
84 | return NULL; | |
85 | ||
a91263d5 DB |
86 | kmemcheck_annotate_bitfield(fp, meta); |
87 | ||
09756af4 AS |
88 | aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags); |
89 | if (aux == NULL) { | |
60a3b225 DB |
90 | vfree(fp); |
91 | return NULL; | |
92 | } | |
93 | ||
94 | fp->pages = size / PAGE_SIZE; | |
09756af4 | 95 | fp->aux = aux; |
e9d8afa9 | 96 | fp->aux->prog = fp; |
60a3b225 DB |
97 | |
98 | return fp; | |
99 | } | |
100 | EXPORT_SYMBOL_GPL(bpf_prog_alloc); | |
101 | ||
102 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, | |
103 | gfp_t gfp_extra_flags) | |
104 | { | |
105 | gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO | | |
106 | gfp_extra_flags; | |
107 | struct bpf_prog *fp; | |
5ccb071e DB |
108 | u32 pages, delta; |
109 | int ret; | |
60a3b225 DB |
110 | |
111 | BUG_ON(fp_old == NULL); | |
112 | ||
113 | size = round_up(size, PAGE_SIZE); | |
5ccb071e DB |
114 | pages = size / PAGE_SIZE; |
115 | if (pages <= fp_old->pages) | |
60a3b225 DB |
116 | return fp_old; |
117 | ||
5ccb071e DB |
118 | delta = pages - fp_old->pages; |
119 | ret = __bpf_prog_charge(fp_old->aux->user, delta); | |
120 | if (ret) | |
121 | return NULL; | |
122 | ||
60a3b225 | 123 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
5ccb071e DB |
124 | if (fp == NULL) { |
125 | __bpf_prog_uncharge(fp_old->aux->user, delta); | |
126 | } else { | |
a91263d5 DB |
127 | kmemcheck_annotate_bitfield(fp, meta); |
128 | ||
60a3b225 | 129 | memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE); |
5ccb071e | 130 | fp->pages = pages; |
e9d8afa9 | 131 | fp->aux->prog = fp; |
60a3b225 | 132 | |
09756af4 | 133 | /* We keep fp->aux from fp_old around in the new |
60a3b225 DB |
134 | * reallocated structure. |
135 | */ | |
09756af4 | 136 | fp_old->aux = NULL; |
60a3b225 DB |
137 | __bpf_prog_free(fp_old); |
138 | } | |
139 | ||
140 | return fp; | |
141 | } | |
60a3b225 DB |
142 | |
143 | void __bpf_prog_free(struct bpf_prog *fp) | |
144 | { | |
09756af4 | 145 | kfree(fp->aux); |
60a3b225 DB |
146 | vfree(fp); |
147 | } | |
60a3b225 | 148 | |
aafe6ae9 | 149 | int bpf_prog_calc_digest(struct bpf_prog *fp) |
7bd509e3 DB |
150 | { |
151 | const u32 bits_offset = SHA_MESSAGE_BYTES - sizeof(__be64); | |
aafe6ae9 DB |
152 | u32 raw_size = bpf_prog_digest_scratch_size(fp); |
153 | u32 ws[SHA_WORKSPACE_WORDS]; | |
7bd509e3 | 154 | u32 i, bsize, psize, blocks; |
aafe6ae9 | 155 | struct bpf_insn *dst; |
7bd509e3 | 156 | bool was_ld_map; |
aafe6ae9 | 157 | u8 *raw, *todo; |
7bd509e3 DB |
158 | __be32 *result; |
159 | __be64 *bits; | |
160 | ||
aafe6ae9 DB |
161 | raw = vmalloc(raw_size); |
162 | if (!raw) | |
163 | return -ENOMEM; | |
164 | ||
7bd509e3 DB |
165 | sha_init(fp->digest); |
166 | memset(ws, 0, sizeof(ws)); | |
167 | ||
168 | /* We need to take out the map fd for the digest calculation | |
169 | * since they are unstable from user space side. | |
170 | */ | |
aafe6ae9 | 171 | dst = (void *)raw; |
7bd509e3 DB |
172 | for (i = 0, was_ld_map = false; i < fp->len; i++) { |
173 | dst[i] = fp->insnsi[i]; | |
174 | if (!was_ld_map && | |
175 | dst[i].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
176 | dst[i].src_reg == BPF_PSEUDO_MAP_FD) { | |
177 | was_ld_map = true; | |
178 | dst[i].imm = 0; | |
179 | } else if (was_ld_map && | |
180 | dst[i].code == 0 && | |
181 | dst[i].dst_reg == 0 && | |
182 | dst[i].src_reg == 0 && | |
183 | dst[i].off == 0) { | |
184 | was_ld_map = false; | |
185 | dst[i].imm = 0; | |
186 | } else { | |
187 | was_ld_map = false; | |
188 | } | |
189 | } | |
190 | ||
aafe6ae9 DB |
191 | psize = bpf_prog_insn_size(fp); |
192 | memset(&raw[psize], 0, raw_size - psize); | |
7bd509e3 DB |
193 | raw[psize++] = 0x80; |
194 | ||
195 | bsize = round_up(psize, SHA_MESSAGE_BYTES); | |
196 | blocks = bsize / SHA_MESSAGE_BYTES; | |
aafe6ae9 | 197 | todo = raw; |
7bd509e3 DB |
198 | if (bsize - psize >= sizeof(__be64)) { |
199 | bits = (__be64 *)(todo + bsize - sizeof(__be64)); | |
200 | } else { | |
201 | bits = (__be64 *)(todo + bsize + bits_offset); | |
202 | blocks++; | |
203 | } | |
204 | *bits = cpu_to_be64((psize - 1) << 3); | |
205 | ||
206 | while (blocks--) { | |
207 | sha_transform(fp->digest, todo, ws); | |
208 | todo += SHA_MESSAGE_BYTES; | |
209 | } | |
210 | ||
211 | result = (__force __be32 *)fp->digest; | |
212 | for (i = 0; i < SHA_DIGEST_WORDS; i++) | |
213 | result[i] = cpu_to_be32(fp->digest[i]); | |
aafe6ae9 DB |
214 | |
215 | vfree(raw); | |
216 | return 0; | |
7bd509e3 DB |
217 | } |
218 | ||
c237ee5e DB |
219 | static bool bpf_is_jmp_and_has_target(const struct bpf_insn *insn) |
220 | { | |
221 | return BPF_CLASS(insn->code) == BPF_JMP && | |
222 | /* Call and Exit are both special jumps with no | |
223 | * target inside the BPF instruction image. | |
224 | */ | |
225 | BPF_OP(insn->code) != BPF_CALL && | |
226 | BPF_OP(insn->code) != BPF_EXIT; | |
227 | } | |
228 | ||
229 | static void bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta) | |
230 | { | |
231 | struct bpf_insn *insn = prog->insnsi; | |
232 | u32 i, insn_cnt = prog->len; | |
233 | ||
234 | for (i = 0; i < insn_cnt; i++, insn++) { | |
235 | if (!bpf_is_jmp_and_has_target(insn)) | |
236 | continue; | |
237 | ||
238 | /* Adjust offset of jmps if we cross boundaries. */ | |
239 | if (i < pos && i + insn->off + 1 > pos) | |
240 | insn->off += delta; | |
241 | else if (i > pos + delta && i + insn->off + 1 <= pos + delta) | |
242 | insn->off -= delta; | |
243 | } | |
244 | } | |
245 | ||
246 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, | |
247 | const struct bpf_insn *patch, u32 len) | |
248 | { | |
249 | u32 insn_adj_cnt, insn_rest, insn_delta = len - 1; | |
250 | struct bpf_prog *prog_adj; | |
251 | ||
252 | /* Since our patchlet doesn't expand the image, we're done. */ | |
253 | if (insn_delta == 0) { | |
254 | memcpy(prog->insnsi + off, patch, sizeof(*patch)); | |
255 | return prog; | |
256 | } | |
257 | ||
258 | insn_adj_cnt = prog->len + insn_delta; | |
259 | ||
260 | /* Several new instructions need to be inserted. Make room | |
261 | * for them. Likely, there's no need for a new allocation as | |
262 | * last page could have large enough tailroom. | |
263 | */ | |
264 | prog_adj = bpf_prog_realloc(prog, bpf_prog_size(insn_adj_cnt), | |
265 | GFP_USER); | |
266 | if (!prog_adj) | |
267 | return NULL; | |
268 | ||
269 | prog_adj->len = insn_adj_cnt; | |
270 | ||
271 | /* Patching happens in 3 steps: | |
272 | * | |
273 | * 1) Move over tail of insnsi from next instruction onwards, | |
274 | * so we can patch the single target insn with one or more | |
275 | * new ones (patching is always from 1 to n insns, n > 0). | |
276 | * 2) Inject new instructions at the target location. | |
277 | * 3) Adjust branch offsets if necessary. | |
278 | */ | |
279 | insn_rest = insn_adj_cnt - off - len; | |
280 | ||
281 | memmove(prog_adj->insnsi + off + len, prog_adj->insnsi + off + 1, | |
282 | sizeof(*patch) * insn_rest); | |
283 | memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len); | |
284 | ||
285 | bpf_adj_branches(prog_adj, off, insn_delta); | |
286 | ||
287 | return prog_adj; | |
288 | } | |
289 | ||
b954d834 | 290 | #ifdef CONFIG_BPF_JIT |
738cbe72 DB |
291 | struct bpf_binary_header * |
292 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
293 | unsigned int alignment, | |
294 | bpf_jit_fill_hole_t bpf_fill_ill_insns) | |
295 | { | |
296 | struct bpf_binary_header *hdr; | |
297 | unsigned int size, hole, start; | |
298 | ||
299 | /* Most of BPF filters are really small, but if some of them | |
300 | * fill a page, allow at least 128 extra bytes to insert a | |
301 | * random section of illegal instructions. | |
302 | */ | |
303 | size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); | |
304 | hdr = module_alloc(size); | |
305 | if (hdr == NULL) | |
306 | return NULL; | |
307 | ||
308 | /* Fill space with illegal/arch-dep instructions. */ | |
309 | bpf_fill_ill_insns(hdr, size); | |
310 | ||
311 | hdr->pages = size / PAGE_SIZE; | |
312 | hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), | |
313 | PAGE_SIZE - sizeof(*hdr)); | |
b7552e1b | 314 | start = (get_random_int() % hole) & ~(alignment - 1); |
738cbe72 DB |
315 | |
316 | /* Leave a random number of instructions before BPF code. */ | |
317 | *image_ptr = &hdr->image[start]; | |
318 | ||
319 | return hdr; | |
320 | } | |
321 | ||
322 | void bpf_jit_binary_free(struct bpf_binary_header *hdr) | |
323 | { | |
be1f221c | 324 | module_memfree(hdr); |
738cbe72 | 325 | } |
4f3446bb DB |
326 | |
327 | int bpf_jit_harden __read_mostly; | |
328 | ||
329 | static int bpf_jit_blind_insn(const struct bpf_insn *from, | |
330 | const struct bpf_insn *aux, | |
331 | struct bpf_insn *to_buff) | |
332 | { | |
333 | struct bpf_insn *to = to_buff; | |
b7552e1b | 334 | u32 imm_rnd = get_random_int(); |
4f3446bb DB |
335 | s16 off; |
336 | ||
337 | BUILD_BUG_ON(BPF_REG_AX + 1 != MAX_BPF_JIT_REG); | |
338 | BUILD_BUG_ON(MAX_BPF_REG + 1 != MAX_BPF_JIT_REG); | |
339 | ||
340 | if (from->imm == 0 && | |
341 | (from->code == (BPF_ALU | BPF_MOV | BPF_K) || | |
342 | from->code == (BPF_ALU64 | BPF_MOV | BPF_K))) { | |
343 | *to++ = BPF_ALU64_REG(BPF_XOR, from->dst_reg, from->dst_reg); | |
344 | goto out; | |
345 | } | |
346 | ||
347 | switch (from->code) { | |
348 | case BPF_ALU | BPF_ADD | BPF_K: | |
349 | case BPF_ALU | BPF_SUB | BPF_K: | |
350 | case BPF_ALU | BPF_AND | BPF_K: | |
351 | case BPF_ALU | BPF_OR | BPF_K: | |
352 | case BPF_ALU | BPF_XOR | BPF_K: | |
353 | case BPF_ALU | BPF_MUL | BPF_K: | |
354 | case BPF_ALU | BPF_MOV | BPF_K: | |
355 | case BPF_ALU | BPF_DIV | BPF_K: | |
356 | case BPF_ALU | BPF_MOD | BPF_K: | |
357 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
358 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
359 | *to++ = BPF_ALU32_REG(from->code, from->dst_reg, BPF_REG_AX); | |
360 | break; | |
361 | ||
362 | case BPF_ALU64 | BPF_ADD | BPF_K: | |
363 | case BPF_ALU64 | BPF_SUB | BPF_K: | |
364 | case BPF_ALU64 | BPF_AND | BPF_K: | |
365 | case BPF_ALU64 | BPF_OR | BPF_K: | |
366 | case BPF_ALU64 | BPF_XOR | BPF_K: | |
367 | case BPF_ALU64 | BPF_MUL | BPF_K: | |
368 | case BPF_ALU64 | BPF_MOV | BPF_K: | |
369 | case BPF_ALU64 | BPF_DIV | BPF_K: | |
370 | case BPF_ALU64 | BPF_MOD | BPF_K: | |
371 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
372 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
373 | *to++ = BPF_ALU64_REG(from->code, from->dst_reg, BPF_REG_AX); | |
374 | break; | |
375 | ||
376 | case BPF_JMP | BPF_JEQ | BPF_K: | |
377 | case BPF_JMP | BPF_JNE | BPF_K: | |
378 | case BPF_JMP | BPF_JGT | BPF_K: | |
379 | case BPF_JMP | BPF_JGE | BPF_K: | |
380 | case BPF_JMP | BPF_JSGT | BPF_K: | |
381 | case BPF_JMP | BPF_JSGE | BPF_K: | |
382 | case BPF_JMP | BPF_JSET | BPF_K: | |
383 | /* Accommodate for extra offset in case of a backjump. */ | |
384 | off = from->off; | |
385 | if (off < 0) | |
386 | off -= 2; | |
387 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
388 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
389 | *to++ = BPF_JMP_REG(from->code, from->dst_reg, BPF_REG_AX, off); | |
390 | break; | |
391 | ||
392 | case BPF_LD | BPF_ABS | BPF_W: | |
393 | case BPF_LD | BPF_ABS | BPF_H: | |
394 | case BPF_LD | BPF_ABS | BPF_B: | |
395 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
396 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
397 | *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0); | |
398 | break; | |
399 | ||
400 | case BPF_LD | BPF_IND | BPF_W: | |
401 | case BPF_LD | BPF_IND | BPF_H: | |
402 | case BPF_LD | BPF_IND | BPF_B: | |
403 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
404 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
405 | *to++ = BPF_ALU32_REG(BPF_ADD, BPF_REG_AX, from->src_reg); | |
406 | *to++ = BPF_LD_IND(from->code, BPF_REG_AX, 0); | |
407 | break; | |
408 | ||
409 | case BPF_LD | BPF_IMM | BPF_DW: | |
410 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[1].imm); | |
411 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
412 | *to++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
413 | *to++ = BPF_ALU64_REG(BPF_MOV, aux[0].dst_reg, BPF_REG_AX); | |
414 | break; | |
415 | case 0: /* Part 2 of BPF_LD | BPF_IMM | BPF_DW. */ | |
416 | *to++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ aux[0].imm); | |
417 | *to++ = BPF_ALU32_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
418 | *to++ = BPF_ALU64_REG(BPF_OR, aux[0].dst_reg, BPF_REG_AX); | |
419 | break; | |
420 | ||
421 | case BPF_ST | BPF_MEM | BPF_DW: | |
422 | case BPF_ST | BPF_MEM | BPF_W: | |
423 | case BPF_ST | BPF_MEM | BPF_H: | |
424 | case BPF_ST | BPF_MEM | BPF_B: | |
425 | *to++ = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, imm_rnd ^ from->imm); | |
426 | *to++ = BPF_ALU64_IMM(BPF_XOR, BPF_REG_AX, imm_rnd); | |
427 | *to++ = BPF_STX_MEM(from->code, from->dst_reg, BPF_REG_AX, from->off); | |
428 | break; | |
429 | } | |
430 | out: | |
431 | return to - to_buff; | |
432 | } | |
433 | ||
434 | static struct bpf_prog *bpf_prog_clone_create(struct bpf_prog *fp_other, | |
435 | gfp_t gfp_extra_flags) | |
436 | { | |
437 | gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO | | |
438 | gfp_extra_flags; | |
439 | struct bpf_prog *fp; | |
440 | ||
441 | fp = __vmalloc(fp_other->pages * PAGE_SIZE, gfp_flags, PAGE_KERNEL); | |
442 | if (fp != NULL) { | |
443 | kmemcheck_annotate_bitfield(fp, meta); | |
444 | ||
445 | /* aux->prog still points to the fp_other one, so | |
446 | * when promoting the clone to the real program, | |
447 | * this still needs to be adapted. | |
448 | */ | |
449 | memcpy(fp, fp_other, fp_other->pages * PAGE_SIZE); | |
450 | } | |
451 | ||
452 | return fp; | |
453 | } | |
454 | ||
455 | static void bpf_prog_clone_free(struct bpf_prog *fp) | |
456 | { | |
457 | /* aux was stolen by the other clone, so we cannot free | |
458 | * it from this path! It will be freed eventually by the | |
459 | * other program on release. | |
460 | * | |
461 | * At this point, we don't need a deferred release since | |
462 | * clone is guaranteed to not be locked. | |
463 | */ | |
464 | fp->aux = NULL; | |
465 | __bpf_prog_free(fp); | |
466 | } | |
467 | ||
468 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other) | |
469 | { | |
470 | /* We have to repoint aux->prog to self, as we don't | |
471 | * know whether fp here is the clone or the original. | |
472 | */ | |
473 | fp->aux->prog = fp; | |
474 | bpf_prog_clone_free(fp_other); | |
475 | } | |
476 | ||
477 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *prog) | |
478 | { | |
479 | struct bpf_insn insn_buff[16], aux[2]; | |
480 | struct bpf_prog *clone, *tmp; | |
481 | int insn_delta, insn_cnt; | |
482 | struct bpf_insn *insn; | |
483 | int i, rewritten; | |
484 | ||
485 | if (!bpf_jit_blinding_enabled()) | |
486 | return prog; | |
487 | ||
488 | clone = bpf_prog_clone_create(prog, GFP_USER); | |
489 | if (!clone) | |
490 | return ERR_PTR(-ENOMEM); | |
491 | ||
492 | insn_cnt = clone->len; | |
493 | insn = clone->insnsi; | |
494 | ||
495 | for (i = 0; i < insn_cnt; i++, insn++) { | |
496 | /* We temporarily need to hold the original ld64 insn | |
497 | * so that we can still access the first part in the | |
498 | * second blinding run. | |
499 | */ | |
500 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW) && | |
501 | insn[1].code == 0) | |
502 | memcpy(aux, insn, sizeof(aux)); | |
503 | ||
504 | rewritten = bpf_jit_blind_insn(insn, aux, insn_buff); | |
505 | if (!rewritten) | |
506 | continue; | |
507 | ||
508 | tmp = bpf_patch_insn_single(clone, i, insn_buff, rewritten); | |
509 | if (!tmp) { | |
510 | /* Patching may have repointed aux->prog during | |
511 | * realloc from the original one, so we need to | |
512 | * fix it up here on error. | |
513 | */ | |
514 | bpf_jit_prog_release_other(prog, clone); | |
515 | return ERR_PTR(-ENOMEM); | |
516 | } | |
517 | ||
518 | clone = tmp; | |
519 | insn_delta = rewritten - 1; | |
520 | ||
521 | /* Walk new program and skip insns we just inserted. */ | |
522 | insn = clone->insnsi + i + insn_delta; | |
523 | insn_cnt += insn_delta; | |
524 | i += insn_delta; | |
525 | } | |
526 | ||
527 | return clone; | |
528 | } | |
b954d834 | 529 | #endif /* CONFIG_BPF_JIT */ |
738cbe72 | 530 | |
f5bffecd AS |
531 | /* Base function for offset calculation. Needs to go into .text section, |
532 | * therefore keeping it non-static as well; will also be used by JITs | |
533 | * anyway later on, so do not let the compiler omit it. | |
534 | */ | |
535 | noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
536 | { | |
537 | return 0; | |
538 | } | |
4d9c5c53 | 539 | EXPORT_SYMBOL_GPL(__bpf_call_base); |
f5bffecd AS |
540 | |
541 | /** | |
7ae457c1 AS |
542 | * __bpf_prog_run - run eBPF program on a given context |
543 | * @ctx: is the data we are operating on | |
544 | * @insn: is the array of eBPF instructions | |
f5bffecd | 545 | * |
7ae457c1 | 546 | * Decode and execute eBPF instructions. |
f5bffecd | 547 | */ |
7ae457c1 | 548 | static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn) |
f5bffecd AS |
549 | { |
550 | u64 stack[MAX_BPF_STACK / sizeof(u64)]; | |
551 | u64 regs[MAX_BPF_REG], tmp; | |
552 | static const void *jumptable[256] = { | |
553 | [0 ... 255] = &&default_label, | |
554 | /* Now overwrite non-defaults ... */ | |
555 | /* 32 bit ALU operations */ | |
556 | [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X, | |
557 | [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K, | |
558 | [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X, | |
559 | [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K, | |
560 | [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X, | |
561 | [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K, | |
562 | [BPF_ALU | BPF_OR | BPF_X] = &&ALU_OR_X, | |
563 | [BPF_ALU | BPF_OR | BPF_K] = &&ALU_OR_K, | |
564 | [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X, | |
565 | [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K, | |
566 | [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X, | |
567 | [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K, | |
568 | [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X, | |
569 | [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K, | |
570 | [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X, | |
571 | [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K, | |
572 | [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X, | |
573 | [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K, | |
574 | [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X, | |
575 | [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K, | |
576 | [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X, | |
577 | [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K, | |
578 | [BPF_ALU | BPF_NEG] = &&ALU_NEG, | |
579 | [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE, | |
580 | [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE, | |
581 | /* 64 bit ALU operations */ | |
582 | [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X, | |
583 | [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K, | |
584 | [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X, | |
585 | [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K, | |
586 | [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X, | |
587 | [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K, | |
588 | [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X, | |
589 | [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K, | |
590 | [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X, | |
591 | [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K, | |
592 | [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X, | |
593 | [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K, | |
594 | [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X, | |
595 | [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K, | |
596 | [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X, | |
597 | [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K, | |
598 | [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X, | |
599 | [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K, | |
600 | [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X, | |
601 | [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K, | |
602 | [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X, | |
603 | [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K, | |
604 | [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X, | |
605 | [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K, | |
606 | [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG, | |
607 | /* Call instruction */ | |
608 | [BPF_JMP | BPF_CALL] = &&JMP_CALL, | |
04fd61ab | 609 | [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL, |
f5bffecd AS |
610 | /* Jumps */ |
611 | [BPF_JMP | BPF_JA] = &&JMP_JA, | |
612 | [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X, | |
613 | [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K, | |
614 | [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X, | |
615 | [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K, | |
616 | [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X, | |
617 | [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K, | |
618 | [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X, | |
619 | [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K, | |
620 | [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X, | |
621 | [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K, | |
622 | [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X, | |
623 | [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K, | |
624 | [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X, | |
625 | [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K, | |
626 | /* Program return */ | |
627 | [BPF_JMP | BPF_EXIT] = &&JMP_EXIT, | |
628 | /* Store instructions */ | |
629 | [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B, | |
630 | [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H, | |
631 | [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W, | |
632 | [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW, | |
633 | [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W, | |
634 | [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW, | |
635 | [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B, | |
636 | [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H, | |
637 | [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W, | |
638 | [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW, | |
639 | /* Load instructions */ | |
640 | [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B, | |
641 | [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H, | |
642 | [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W, | |
643 | [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW, | |
644 | [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W, | |
645 | [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H, | |
646 | [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B, | |
647 | [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W, | |
648 | [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H, | |
649 | [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B, | |
02ab695b | 650 | [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW, |
f5bffecd | 651 | }; |
04fd61ab | 652 | u32 tail_call_cnt = 0; |
f5bffecd AS |
653 | void *ptr; |
654 | int off; | |
655 | ||
656 | #define CONT ({ insn++; goto select_insn; }) | |
657 | #define CONT_JMP ({ insn++; goto select_insn; }) | |
658 | ||
659 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; | |
660 | ARG1 = (u64) (unsigned long) ctx; | |
661 | ||
f5bffecd AS |
662 | select_insn: |
663 | goto *jumptable[insn->code]; | |
664 | ||
665 | /* ALU */ | |
666 | #define ALU(OPCODE, OP) \ | |
667 | ALU64_##OPCODE##_X: \ | |
668 | DST = DST OP SRC; \ | |
669 | CONT; \ | |
670 | ALU_##OPCODE##_X: \ | |
671 | DST = (u32) DST OP (u32) SRC; \ | |
672 | CONT; \ | |
673 | ALU64_##OPCODE##_K: \ | |
674 | DST = DST OP IMM; \ | |
675 | CONT; \ | |
676 | ALU_##OPCODE##_K: \ | |
677 | DST = (u32) DST OP (u32) IMM; \ | |
678 | CONT; | |
679 | ||
680 | ALU(ADD, +) | |
681 | ALU(SUB, -) | |
682 | ALU(AND, &) | |
683 | ALU(OR, |) | |
684 | ALU(LSH, <<) | |
685 | ALU(RSH, >>) | |
686 | ALU(XOR, ^) | |
687 | ALU(MUL, *) | |
688 | #undef ALU | |
689 | ALU_NEG: | |
690 | DST = (u32) -DST; | |
691 | CONT; | |
692 | ALU64_NEG: | |
693 | DST = -DST; | |
694 | CONT; | |
695 | ALU_MOV_X: | |
696 | DST = (u32) SRC; | |
697 | CONT; | |
698 | ALU_MOV_K: | |
699 | DST = (u32) IMM; | |
700 | CONT; | |
701 | ALU64_MOV_X: | |
702 | DST = SRC; | |
703 | CONT; | |
704 | ALU64_MOV_K: | |
705 | DST = IMM; | |
706 | CONT; | |
02ab695b AS |
707 | LD_IMM_DW: |
708 | DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32; | |
709 | insn++; | |
710 | CONT; | |
f5bffecd AS |
711 | ALU64_ARSH_X: |
712 | (*(s64 *) &DST) >>= SRC; | |
713 | CONT; | |
714 | ALU64_ARSH_K: | |
715 | (*(s64 *) &DST) >>= IMM; | |
716 | CONT; | |
717 | ALU64_MOD_X: | |
718 | if (unlikely(SRC == 0)) | |
719 | return 0; | |
876a7ae6 AS |
720 | div64_u64_rem(DST, SRC, &tmp); |
721 | DST = tmp; | |
f5bffecd AS |
722 | CONT; |
723 | ALU_MOD_X: | |
724 | if (unlikely(SRC == 0)) | |
725 | return 0; | |
726 | tmp = (u32) DST; | |
727 | DST = do_div(tmp, (u32) SRC); | |
728 | CONT; | |
729 | ALU64_MOD_K: | |
876a7ae6 AS |
730 | div64_u64_rem(DST, IMM, &tmp); |
731 | DST = tmp; | |
f5bffecd AS |
732 | CONT; |
733 | ALU_MOD_K: | |
734 | tmp = (u32) DST; | |
735 | DST = do_div(tmp, (u32) IMM); | |
736 | CONT; | |
737 | ALU64_DIV_X: | |
738 | if (unlikely(SRC == 0)) | |
739 | return 0; | |
876a7ae6 | 740 | DST = div64_u64(DST, SRC); |
f5bffecd AS |
741 | CONT; |
742 | ALU_DIV_X: | |
743 | if (unlikely(SRC == 0)) | |
744 | return 0; | |
745 | tmp = (u32) DST; | |
746 | do_div(tmp, (u32) SRC); | |
747 | DST = (u32) tmp; | |
748 | CONT; | |
749 | ALU64_DIV_K: | |
876a7ae6 | 750 | DST = div64_u64(DST, IMM); |
f5bffecd AS |
751 | CONT; |
752 | ALU_DIV_K: | |
753 | tmp = (u32) DST; | |
754 | do_div(tmp, (u32) IMM); | |
755 | DST = (u32) tmp; | |
756 | CONT; | |
757 | ALU_END_TO_BE: | |
758 | switch (IMM) { | |
759 | case 16: | |
760 | DST = (__force u16) cpu_to_be16(DST); | |
761 | break; | |
762 | case 32: | |
763 | DST = (__force u32) cpu_to_be32(DST); | |
764 | break; | |
765 | case 64: | |
766 | DST = (__force u64) cpu_to_be64(DST); | |
767 | break; | |
768 | } | |
769 | CONT; | |
770 | ALU_END_TO_LE: | |
771 | switch (IMM) { | |
772 | case 16: | |
773 | DST = (__force u16) cpu_to_le16(DST); | |
774 | break; | |
775 | case 32: | |
776 | DST = (__force u32) cpu_to_le32(DST); | |
777 | break; | |
778 | case 64: | |
779 | DST = (__force u64) cpu_to_le64(DST); | |
780 | break; | |
781 | } | |
782 | CONT; | |
783 | ||
784 | /* CALL */ | |
785 | JMP_CALL: | |
786 | /* Function call scratches BPF_R1-BPF_R5 registers, | |
787 | * preserves BPF_R6-BPF_R9, and stores return value | |
788 | * into BPF_R0. | |
789 | */ | |
790 | BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3, | |
791 | BPF_R4, BPF_R5); | |
792 | CONT; | |
793 | ||
04fd61ab AS |
794 | JMP_TAIL_CALL: { |
795 | struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2; | |
796 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
797 | struct bpf_prog *prog; | |
798 | u64 index = BPF_R3; | |
799 | ||
800 | if (unlikely(index >= array->map.max_entries)) | |
801 | goto out; | |
04fd61ab AS |
802 | if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) |
803 | goto out; | |
804 | ||
805 | tail_call_cnt++; | |
806 | ||
2a36f0b9 | 807 | prog = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 808 | if (!prog) |
04fd61ab AS |
809 | goto out; |
810 | ||
c4675f93 DB |
811 | /* ARG1 at this point is guaranteed to point to CTX from |
812 | * the verifier side due to the fact that the tail call is | |
813 | * handeled like a helper, that is, bpf_tail_call_proto, | |
814 | * where arg1_type is ARG_PTR_TO_CTX. | |
815 | */ | |
04fd61ab AS |
816 | insn = prog->insnsi; |
817 | goto select_insn; | |
818 | out: | |
819 | CONT; | |
820 | } | |
f5bffecd AS |
821 | /* JMP */ |
822 | JMP_JA: | |
823 | insn += insn->off; | |
824 | CONT; | |
825 | JMP_JEQ_X: | |
826 | if (DST == SRC) { | |
827 | insn += insn->off; | |
828 | CONT_JMP; | |
829 | } | |
830 | CONT; | |
831 | JMP_JEQ_K: | |
832 | if (DST == IMM) { | |
833 | insn += insn->off; | |
834 | CONT_JMP; | |
835 | } | |
836 | CONT; | |
837 | JMP_JNE_X: | |
838 | if (DST != SRC) { | |
839 | insn += insn->off; | |
840 | CONT_JMP; | |
841 | } | |
842 | CONT; | |
843 | JMP_JNE_K: | |
844 | if (DST != IMM) { | |
845 | insn += insn->off; | |
846 | CONT_JMP; | |
847 | } | |
848 | CONT; | |
849 | JMP_JGT_X: | |
850 | if (DST > SRC) { | |
851 | insn += insn->off; | |
852 | CONT_JMP; | |
853 | } | |
854 | CONT; | |
855 | JMP_JGT_K: | |
856 | if (DST > IMM) { | |
857 | insn += insn->off; | |
858 | CONT_JMP; | |
859 | } | |
860 | CONT; | |
861 | JMP_JGE_X: | |
862 | if (DST >= SRC) { | |
863 | insn += insn->off; | |
864 | CONT_JMP; | |
865 | } | |
866 | CONT; | |
867 | JMP_JGE_K: | |
868 | if (DST >= IMM) { | |
869 | insn += insn->off; | |
870 | CONT_JMP; | |
871 | } | |
872 | CONT; | |
873 | JMP_JSGT_X: | |
874 | if (((s64) DST) > ((s64) SRC)) { | |
875 | insn += insn->off; | |
876 | CONT_JMP; | |
877 | } | |
878 | CONT; | |
879 | JMP_JSGT_K: | |
880 | if (((s64) DST) > ((s64) IMM)) { | |
881 | insn += insn->off; | |
882 | CONT_JMP; | |
883 | } | |
884 | CONT; | |
885 | JMP_JSGE_X: | |
886 | if (((s64) DST) >= ((s64) SRC)) { | |
887 | insn += insn->off; | |
888 | CONT_JMP; | |
889 | } | |
890 | CONT; | |
891 | JMP_JSGE_K: | |
892 | if (((s64) DST) >= ((s64) IMM)) { | |
893 | insn += insn->off; | |
894 | CONT_JMP; | |
895 | } | |
896 | CONT; | |
897 | JMP_JSET_X: | |
898 | if (DST & SRC) { | |
899 | insn += insn->off; | |
900 | CONT_JMP; | |
901 | } | |
902 | CONT; | |
903 | JMP_JSET_K: | |
904 | if (DST & IMM) { | |
905 | insn += insn->off; | |
906 | CONT_JMP; | |
907 | } | |
908 | CONT; | |
909 | JMP_EXIT: | |
910 | return BPF_R0; | |
911 | ||
912 | /* STX and ST and LDX*/ | |
913 | #define LDST(SIZEOP, SIZE) \ | |
914 | STX_MEM_##SIZEOP: \ | |
915 | *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \ | |
916 | CONT; \ | |
917 | ST_MEM_##SIZEOP: \ | |
918 | *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \ | |
919 | CONT; \ | |
920 | LDX_MEM_##SIZEOP: \ | |
921 | DST = *(SIZE *)(unsigned long) (SRC + insn->off); \ | |
922 | CONT; | |
923 | ||
924 | LDST(B, u8) | |
925 | LDST(H, u16) | |
926 | LDST(W, u32) | |
927 | LDST(DW, u64) | |
928 | #undef LDST | |
929 | STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */ | |
930 | atomic_add((u32) SRC, (atomic_t *)(unsigned long) | |
931 | (DST + insn->off)); | |
932 | CONT; | |
933 | STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */ | |
934 | atomic64_add((u64) SRC, (atomic64_t *)(unsigned long) | |
935 | (DST + insn->off)); | |
936 | CONT; | |
937 | LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */ | |
938 | off = IMM; | |
939 | load_word: | |
940 | /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are | |
941 | * only appearing in the programs where ctx == | |
942 | * skb. All programs keep 'ctx' in regs[BPF_REG_CTX] | |
8fb575ca | 943 | * == BPF_R6, bpf_convert_filter() saves it in BPF_R6, |
f5bffecd AS |
944 | * internal BPF verifier will check that BPF_R6 == |
945 | * ctx. | |
946 | * | |
947 | * BPF_ABS and BPF_IND are wrappers of function calls, | |
948 | * so they scratch BPF_R1-BPF_R5 registers, preserve | |
949 | * BPF_R6-BPF_R9, and store return value into BPF_R0. | |
950 | * | |
951 | * Implicit input: | |
952 | * ctx == skb == BPF_R6 == CTX | |
953 | * | |
954 | * Explicit input: | |
955 | * SRC == any register | |
956 | * IMM == 32-bit immediate | |
957 | * | |
958 | * Output: | |
959 | * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness | |
960 | */ | |
961 | ||
962 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp); | |
963 | if (likely(ptr != NULL)) { | |
964 | BPF_R0 = get_unaligned_be32(ptr); | |
965 | CONT; | |
966 | } | |
967 | ||
968 | return 0; | |
969 | LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */ | |
970 | off = IMM; | |
971 | load_half: | |
972 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp); | |
973 | if (likely(ptr != NULL)) { | |
974 | BPF_R0 = get_unaligned_be16(ptr); | |
975 | CONT; | |
976 | } | |
977 | ||
978 | return 0; | |
979 | LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */ | |
980 | off = IMM; | |
981 | load_byte: | |
982 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp); | |
983 | if (likely(ptr != NULL)) { | |
984 | BPF_R0 = *(u8 *)ptr; | |
985 | CONT; | |
986 | } | |
987 | ||
988 | return 0; | |
989 | LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */ | |
990 | off = IMM + SRC; | |
991 | goto load_word; | |
992 | LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */ | |
993 | off = IMM + SRC; | |
994 | goto load_half; | |
995 | LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */ | |
996 | off = IMM + SRC; | |
997 | goto load_byte; | |
998 | ||
999 | default_label: | |
1000 | /* If we ever reach this, we have a bug somewhere. */ | |
1001 | WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code); | |
1002 | return 0; | |
1003 | } | |
39853cc0 | 1004 | STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */ |
f5bffecd | 1005 | |
3324b584 DB |
1006 | bool bpf_prog_array_compatible(struct bpf_array *array, |
1007 | const struct bpf_prog *fp) | |
04fd61ab | 1008 | { |
3324b584 DB |
1009 | if (!array->owner_prog_type) { |
1010 | /* There's no owner yet where we could check for | |
1011 | * compatibility. | |
1012 | */ | |
04fd61ab AS |
1013 | array->owner_prog_type = fp->type; |
1014 | array->owner_jited = fp->jited; | |
3324b584 DB |
1015 | |
1016 | return true; | |
04fd61ab | 1017 | } |
3324b584 DB |
1018 | |
1019 | return array->owner_prog_type == fp->type && | |
1020 | array->owner_jited == fp->jited; | |
04fd61ab AS |
1021 | } |
1022 | ||
3324b584 | 1023 | static int bpf_check_tail_call(const struct bpf_prog *fp) |
04fd61ab AS |
1024 | { |
1025 | struct bpf_prog_aux *aux = fp->aux; | |
1026 | int i; | |
1027 | ||
1028 | for (i = 0; i < aux->used_map_cnt; i++) { | |
3324b584 | 1029 | struct bpf_map *map = aux->used_maps[i]; |
04fd61ab | 1030 | struct bpf_array *array; |
04fd61ab | 1031 | |
04fd61ab AS |
1032 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) |
1033 | continue; | |
3324b584 | 1034 | |
04fd61ab AS |
1035 | array = container_of(map, struct bpf_array, map); |
1036 | if (!bpf_prog_array_compatible(array, fp)) | |
1037 | return -EINVAL; | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
f5bffecd | 1043 | /** |
3324b584 | 1044 | * bpf_prog_select_runtime - select exec runtime for BPF program |
7ae457c1 | 1045 | * @fp: bpf_prog populated with internal BPF program |
d1c55ab5 | 1046 | * @err: pointer to error variable |
f5bffecd | 1047 | * |
3324b584 DB |
1048 | * Try to JIT eBPF program, if JIT is not available, use interpreter. |
1049 | * The BPF program will be executed via BPF_PROG_RUN() macro. | |
f5bffecd | 1050 | */ |
d1c55ab5 | 1051 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err) |
f5bffecd | 1052 | { |
7ae457c1 | 1053 | fp->bpf_func = (void *) __bpf_prog_run; |
f5bffecd | 1054 | |
d1c55ab5 DB |
1055 | /* eBPF JITs can rewrite the program in case constant |
1056 | * blinding is active. However, in case of error during | |
1057 | * blinding, bpf_int_jit_compile() must always return a | |
1058 | * valid program, which in this case would simply not | |
1059 | * be JITed, but falls back to the interpreter. | |
1060 | */ | |
1061 | fp = bpf_int_jit_compile(fp); | |
60a3b225 | 1062 | bpf_prog_lock_ro(fp); |
04fd61ab | 1063 | |
3324b584 DB |
1064 | /* The tail call compatibility check can only be done at |
1065 | * this late stage as we need to determine, if we deal | |
1066 | * with JITed or non JITed program concatenations and not | |
1067 | * all eBPF JITs might immediately support all features. | |
1068 | */ | |
d1c55ab5 DB |
1069 | *err = bpf_check_tail_call(fp); |
1070 | ||
1071 | return fp; | |
f5bffecd | 1072 | } |
7ae457c1 | 1073 | EXPORT_SYMBOL_GPL(bpf_prog_select_runtime); |
f5bffecd | 1074 | |
60a3b225 DB |
1075 | static void bpf_prog_free_deferred(struct work_struct *work) |
1076 | { | |
09756af4 | 1077 | struct bpf_prog_aux *aux; |
60a3b225 | 1078 | |
09756af4 AS |
1079 | aux = container_of(work, struct bpf_prog_aux, work); |
1080 | bpf_jit_free(aux->prog); | |
60a3b225 DB |
1081 | } |
1082 | ||
1083 | /* Free internal BPF program */ | |
7ae457c1 | 1084 | void bpf_prog_free(struct bpf_prog *fp) |
f5bffecd | 1085 | { |
09756af4 | 1086 | struct bpf_prog_aux *aux = fp->aux; |
60a3b225 | 1087 | |
09756af4 | 1088 | INIT_WORK(&aux->work, bpf_prog_free_deferred); |
09756af4 | 1089 | schedule_work(&aux->work); |
f5bffecd | 1090 | } |
7ae457c1 | 1091 | EXPORT_SYMBOL_GPL(bpf_prog_free); |
f89b7755 | 1092 | |
3ad00405 DB |
1093 | /* RNG for unpriviledged user space with separated state from prandom_u32(). */ |
1094 | static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state); | |
1095 | ||
1096 | void bpf_user_rnd_init_once(void) | |
1097 | { | |
1098 | prandom_init_once(&bpf_user_rnd_state); | |
1099 | } | |
1100 | ||
f3694e00 | 1101 | BPF_CALL_0(bpf_user_rnd_u32) |
3ad00405 DB |
1102 | { |
1103 | /* Should someone ever have the rather unwise idea to use some | |
1104 | * of the registers passed into this function, then note that | |
1105 | * this function is called from native eBPF and classic-to-eBPF | |
1106 | * transformations. Register assignments from both sides are | |
1107 | * different, f.e. classic always sets fn(ctx, A, X) here. | |
1108 | */ | |
1109 | struct rnd_state *state; | |
1110 | u32 res; | |
1111 | ||
1112 | state = &get_cpu_var(bpf_user_rnd_state); | |
1113 | res = prandom_u32_state(state); | |
b761fe22 | 1114 | put_cpu_var(bpf_user_rnd_state); |
3ad00405 DB |
1115 | |
1116 | return res; | |
1117 | } | |
1118 | ||
3ba67dab DB |
1119 | /* Weak definitions of helper functions in case we don't have bpf syscall. */ |
1120 | const struct bpf_func_proto bpf_map_lookup_elem_proto __weak; | |
1121 | const struct bpf_func_proto bpf_map_update_elem_proto __weak; | |
1122 | const struct bpf_func_proto bpf_map_delete_elem_proto __weak; | |
1123 | ||
03e69b50 | 1124 | const struct bpf_func_proto bpf_get_prandom_u32_proto __weak; |
c04167ce | 1125 | const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak; |
2d0e30c3 | 1126 | const struct bpf_func_proto bpf_get_numa_node_id_proto __weak; |
17ca8cbf | 1127 | const struct bpf_func_proto bpf_ktime_get_ns_proto __weak; |
bd570ff9 | 1128 | |
ffeedafb AS |
1129 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak; |
1130 | const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak; | |
1131 | const struct bpf_func_proto bpf_get_current_comm_proto __weak; | |
bd570ff9 | 1132 | |
0756ea3e AS |
1133 | const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void) |
1134 | { | |
1135 | return NULL; | |
1136 | } | |
03e69b50 | 1137 | |
555c8a86 DB |
1138 | u64 __weak |
1139 | bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, | |
1140 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 1141 | { |
555c8a86 | 1142 | return -ENOTSUPP; |
bd570ff9 DB |
1143 | } |
1144 | ||
3324b584 DB |
1145 | /* Always built-in helper functions. */ |
1146 | const struct bpf_func_proto bpf_tail_call_proto = { | |
1147 | .func = NULL, | |
1148 | .gpl_only = false, | |
1149 | .ret_type = RET_VOID, | |
1150 | .arg1_type = ARG_PTR_TO_CTX, | |
1151 | .arg2_type = ARG_CONST_MAP_PTR, | |
1152 | .arg3_type = ARG_ANYTHING, | |
1153 | }; | |
1154 | ||
1155 | /* For classic BPF JITs that don't implement bpf_int_jit_compile(). */ | |
d1c55ab5 | 1156 | struct bpf_prog * __weak bpf_int_jit_compile(struct bpf_prog *prog) |
3324b584 | 1157 | { |
d1c55ab5 | 1158 | return prog; |
3324b584 DB |
1159 | } |
1160 | ||
17bedab2 | 1161 | bool __weak bpf_helper_changes_pkt_data(void *func) |
969bf05e AS |
1162 | { |
1163 | return false; | |
1164 | } | |
1165 | ||
f89b7755 AS |
1166 | /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call |
1167 | * skb_copy_bits(), so provide a weak definition of it for NET-less config. | |
1168 | */ | |
1169 | int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to, | |
1170 | int len) | |
1171 | { | |
1172 | return -EFAULT; | |
1173 | } |