]>
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; | |
108 | ||
109 | BUG_ON(fp_old == NULL); | |
110 | ||
111 | size = round_up(size, PAGE_SIZE); | |
112 | if (size <= fp_old->pages * PAGE_SIZE) | |
113 | return fp_old; | |
114 | ||
115 | fp = __vmalloc(size, gfp_flags, PAGE_KERNEL); | |
116 | if (fp != NULL) { | |
a91263d5 DB |
117 | kmemcheck_annotate_bitfield(fp, meta); |
118 | ||
60a3b225 DB |
119 | memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE); |
120 | fp->pages = size / PAGE_SIZE; | |
e9d8afa9 | 121 | fp->aux->prog = fp; |
60a3b225 | 122 | |
09756af4 | 123 | /* We keep fp->aux from fp_old around in the new |
60a3b225 DB |
124 | * reallocated structure. |
125 | */ | |
09756af4 | 126 | fp_old->aux = NULL; |
60a3b225 DB |
127 | __bpf_prog_free(fp_old); |
128 | } | |
129 | ||
130 | return fp; | |
131 | } | |
60a3b225 DB |
132 | |
133 | void __bpf_prog_free(struct bpf_prog *fp) | |
134 | { | |
09756af4 | 135 | kfree(fp->aux); |
60a3b225 DB |
136 | vfree(fp); |
137 | } | |
60a3b225 | 138 | |
b954d834 | 139 | #ifdef CONFIG_BPF_JIT |
738cbe72 DB |
140 | struct bpf_binary_header * |
141 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
142 | unsigned int alignment, | |
143 | bpf_jit_fill_hole_t bpf_fill_ill_insns) | |
144 | { | |
145 | struct bpf_binary_header *hdr; | |
146 | unsigned int size, hole, start; | |
147 | ||
148 | /* Most of BPF filters are really small, but if some of them | |
149 | * fill a page, allow at least 128 extra bytes to insert a | |
150 | * random section of illegal instructions. | |
151 | */ | |
152 | size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); | |
153 | hdr = module_alloc(size); | |
154 | if (hdr == NULL) | |
155 | return NULL; | |
156 | ||
157 | /* Fill space with illegal/arch-dep instructions. */ | |
158 | bpf_fill_ill_insns(hdr, size); | |
159 | ||
160 | hdr->pages = size / PAGE_SIZE; | |
161 | hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), | |
162 | PAGE_SIZE - sizeof(*hdr)); | |
163 | start = (prandom_u32() % hole) & ~(alignment - 1); | |
164 | ||
165 | /* Leave a random number of instructions before BPF code. */ | |
166 | *image_ptr = &hdr->image[start]; | |
167 | ||
168 | return hdr; | |
169 | } | |
170 | ||
171 | void bpf_jit_binary_free(struct bpf_binary_header *hdr) | |
172 | { | |
be1f221c | 173 | module_memfree(hdr); |
738cbe72 | 174 | } |
b954d834 | 175 | #endif /* CONFIG_BPF_JIT */ |
738cbe72 | 176 | |
f5bffecd AS |
177 | /* Base function for offset calculation. Needs to go into .text section, |
178 | * therefore keeping it non-static as well; will also be used by JITs | |
179 | * anyway later on, so do not let the compiler omit it. | |
180 | */ | |
181 | noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
182 | { | |
183 | return 0; | |
184 | } | |
4d9c5c53 | 185 | EXPORT_SYMBOL_GPL(__bpf_call_base); |
f5bffecd AS |
186 | |
187 | /** | |
7ae457c1 AS |
188 | * __bpf_prog_run - run eBPF program on a given context |
189 | * @ctx: is the data we are operating on | |
190 | * @insn: is the array of eBPF instructions | |
f5bffecd | 191 | * |
7ae457c1 | 192 | * Decode and execute eBPF instructions. |
f5bffecd | 193 | */ |
7ae457c1 | 194 | static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn) |
f5bffecd AS |
195 | { |
196 | u64 stack[MAX_BPF_STACK / sizeof(u64)]; | |
197 | u64 regs[MAX_BPF_REG], tmp; | |
198 | static const void *jumptable[256] = { | |
199 | [0 ... 255] = &&default_label, | |
200 | /* Now overwrite non-defaults ... */ | |
201 | /* 32 bit ALU operations */ | |
202 | [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X, | |
203 | [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K, | |
204 | [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X, | |
205 | [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K, | |
206 | [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X, | |
207 | [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K, | |
208 | [BPF_ALU | BPF_OR | BPF_X] = &&ALU_OR_X, | |
209 | [BPF_ALU | BPF_OR | BPF_K] = &&ALU_OR_K, | |
210 | [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X, | |
211 | [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K, | |
212 | [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X, | |
213 | [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K, | |
214 | [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X, | |
215 | [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K, | |
216 | [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X, | |
217 | [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K, | |
218 | [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X, | |
219 | [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K, | |
220 | [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X, | |
221 | [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K, | |
222 | [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X, | |
223 | [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K, | |
224 | [BPF_ALU | BPF_NEG] = &&ALU_NEG, | |
225 | [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE, | |
226 | [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE, | |
227 | /* 64 bit ALU operations */ | |
228 | [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X, | |
229 | [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K, | |
230 | [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X, | |
231 | [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K, | |
232 | [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X, | |
233 | [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K, | |
234 | [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X, | |
235 | [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K, | |
236 | [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X, | |
237 | [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K, | |
238 | [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X, | |
239 | [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K, | |
240 | [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X, | |
241 | [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K, | |
242 | [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X, | |
243 | [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K, | |
244 | [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X, | |
245 | [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K, | |
246 | [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X, | |
247 | [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K, | |
248 | [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X, | |
249 | [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K, | |
250 | [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X, | |
251 | [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K, | |
252 | [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG, | |
253 | /* Call instruction */ | |
254 | [BPF_JMP | BPF_CALL] = &&JMP_CALL, | |
04fd61ab | 255 | [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL, |
f5bffecd AS |
256 | /* Jumps */ |
257 | [BPF_JMP | BPF_JA] = &&JMP_JA, | |
258 | [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X, | |
259 | [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K, | |
260 | [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X, | |
261 | [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K, | |
262 | [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X, | |
263 | [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K, | |
264 | [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X, | |
265 | [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K, | |
266 | [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X, | |
267 | [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K, | |
268 | [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X, | |
269 | [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K, | |
270 | [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X, | |
271 | [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K, | |
272 | /* Program return */ | |
273 | [BPF_JMP | BPF_EXIT] = &&JMP_EXIT, | |
274 | /* Store instructions */ | |
275 | [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B, | |
276 | [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H, | |
277 | [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W, | |
278 | [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW, | |
279 | [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W, | |
280 | [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW, | |
281 | [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B, | |
282 | [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H, | |
283 | [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W, | |
284 | [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW, | |
285 | /* Load instructions */ | |
286 | [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B, | |
287 | [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H, | |
288 | [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W, | |
289 | [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW, | |
290 | [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W, | |
291 | [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H, | |
292 | [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B, | |
293 | [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W, | |
294 | [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H, | |
295 | [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B, | |
02ab695b | 296 | [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW, |
f5bffecd | 297 | }; |
04fd61ab | 298 | u32 tail_call_cnt = 0; |
f5bffecd AS |
299 | void *ptr; |
300 | int off; | |
301 | ||
302 | #define CONT ({ insn++; goto select_insn; }) | |
303 | #define CONT_JMP ({ insn++; goto select_insn; }) | |
304 | ||
305 | FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; | |
306 | ARG1 = (u64) (unsigned long) ctx; | |
307 | ||
f5bffecd AS |
308 | select_insn: |
309 | goto *jumptable[insn->code]; | |
310 | ||
311 | /* ALU */ | |
312 | #define ALU(OPCODE, OP) \ | |
313 | ALU64_##OPCODE##_X: \ | |
314 | DST = DST OP SRC; \ | |
315 | CONT; \ | |
316 | ALU_##OPCODE##_X: \ | |
317 | DST = (u32) DST OP (u32) SRC; \ | |
318 | CONT; \ | |
319 | ALU64_##OPCODE##_K: \ | |
320 | DST = DST OP IMM; \ | |
321 | CONT; \ | |
322 | ALU_##OPCODE##_K: \ | |
323 | DST = (u32) DST OP (u32) IMM; \ | |
324 | CONT; | |
325 | ||
326 | ALU(ADD, +) | |
327 | ALU(SUB, -) | |
328 | ALU(AND, &) | |
329 | ALU(OR, |) | |
330 | ALU(LSH, <<) | |
331 | ALU(RSH, >>) | |
332 | ALU(XOR, ^) | |
333 | ALU(MUL, *) | |
334 | #undef ALU | |
335 | ALU_NEG: | |
336 | DST = (u32) -DST; | |
337 | CONT; | |
338 | ALU64_NEG: | |
339 | DST = -DST; | |
340 | CONT; | |
341 | ALU_MOV_X: | |
342 | DST = (u32) SRC; | |
343 | CONT; | |
344 | ALU_MOV_K: | |
345 | DST = (u32) IMM; | |
346 | CONT; | |
347 | ALU64_MOV_X: | |
348 | DST = SRC; | |
349 | CONT; | |
350 | ALU64_MOV_K: | |
351 | DST = IMM; | |
352 | CONT; | |
02ab695b AS |
353 | LD_IMM_DW: |
354 | DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32; | |
355 | insn++; | |
356 | CONT; | |
f5bffecd AS |
357 | ALU64_ARSH_X: |
358 | (*(s64 *) &DST) >>= SRC; | |
359 | CONT; | |
360 | ALU64_ARSH_K: | |
361 | (*(s64 *) &DST) >>= IMM; | |
362 | CONT; | |
363 | ALU64_MOD_X: | |
364 | if (unlikely(SRC == 0)) | |
365 | return 0; | |
876a7ae6 AS |
366 | div64_u64_rem(DST, SRC, &tmp); |
367 | DST = tmp; | |
f5bffecd AS |
368 | CONT; |
369 | ALU_MOD_X: | |
370 | if (unlikely(SRC == 0)) | |
371 | return 0; | |
372 | tmp = (u32) DST; | |
373 | DST = do_div(tmp, (u32) SRC); | |
374 | CONT; | |
375 | ALU64_MOD_K: | |
876a7ae6 AS |
376 | div64_u64_rem(DST, IMM, &tmp); |
377 | DST = tmp; | |
f5bffecd AS |
378 | CONT; |
379 | ALU_MOD_K: | |
380 | tmp = (u32) DST; | |
381 | DST = do_div(tmp, (u32) IMM); | |
382 | CONT; | |
383 | ALU64_DIV_X: | |
384 | if (unlikely(SRC == 0)) | |
385 | return 0; | |
876a7ae6 | 386 | DST = div64_u64(DST, SRC); |
f5bffecd AS |
387 | CONT; |
388 | ALU_DIV_X: | |
389 | if (unlikely(SRC == 0)) | |
390 | return 0; | |
391 | tmp = (u32) DST; | |
392 | do_div(tmp, (u32) SRC); | |
393 | DST = (u32) tmp; | |
394 | CONT; | |
395 | ALU64_DIV_K: | |
876a7ae6 | 396 | DST = div64_u64(DST, IMM); |
f5bffecd AS |
397 | CONT; |
398 | ALU_DIV_K: | |
399 | tmp = (u32) DST; | |
400 | do_div(tmp, (u32) IMM); | |
401 | DST = (u32) tmp; | |
402 | CONT; | |
403 | ALU_END_TO_BE: | |
404 | switch (IMM) { | |
405 | case 16: | |
406 | DST = (__force u16) cpu_to_be16(DST); | |
407 | break; | |
408 | case 32: | |
409 | DST = (__force u32) cpu_to_be32(DST); | |
410 | break; | |
411 | case 64: | |
412 | DST = (__force u64) cpu_to_be64(DST); | |
413 | break; | |
414 | } | |
415 | CONT; | |
416 | ALU_END_TO_LE: | |
417 | switch (IMM) { | |
418 | case 16: | |
419 | DST = (__force u16) cpu_to_le16(DST); | |
420 | break; | |
421 | case 32: | |
422 | DST = (__force u32) cpu_to_le32(DST); | |
423 | break; | |
424 | case 64: | |
425 | DST = (__force u64) cpu_to_le64(DST); | |
426 | break; | |
427 | } | |
428 | CONT; | |
429 | ||
430 | /* CALL */ | |
431 | JMP_CALL: | |
432 | /* Function call scratches BPF_R1-BPF_R5 registers, | |
433 | * preserves BPF_R6-BPF_R9, and stores return value | |
434 | * into BPF_R0. | |
435 | */ | |
436 | BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3, | |
437 | BPF_R4, BPF_R5); | |
438 | CONT; | |
439 | ||
04fd61ab AS |
440 | JMP_TAIL_CALL: { |
441 | struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2; | |
442 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
443 | struct bpf_prog *prog; | |
444 | u64 index = BPF_R3; | |
445 | ||
446 | if (unlikely(index >= array->map.max_entries)) | |
447 | goto out; | |
448 | ||
449 | if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) | |
450 | goto out; | |
451 | ||
452 | tail_call_cnt++; | |
453 | ||
2a36f0b9 | 454 | prog = READ_ONCE(array->ptrs[index]); |
04fd61ab AS |
455 | if (unlikely(!prog)) |
456 | goto out; | |
457 | ||
c4675f93 DB |
458 | /* ARG1 at this point is guaranteed to point to CTX from |
459 | * the verifier side due to the fact that the tail call is | |
460 | * handeled like a helper, that is, bpf_tail_call_proto, | |
461 | * where arg1_type is ARG_PTR_TO_CTX. | |
462 | */ | |
04fd61ab AS |
463 | insn = prog->insnsi; |
464 | goto select_insn; | |
465 | out: | |
466 | CONT; | |
467 | } | |
f5bffecd AS |
468 | /* JMP */ |
469 | JMP_JA: | |
470 | insn += insn->off; | |
471 | CONT; | |
472 | JMP_JEQ_X: | |
473 | if (DST == SRC) { | |
474 | insn += insn->off; | |
475 | CONT_JMP; | |
476 | } | |
477 | CONT; | |
478 | JMP_JEQ_K: | |
479 | if (DST == IMM) { | |
480 | insn += insn->off; | |
481 | CONT_JMP; | |
482 | } | |
483 | CONT; | |
484 | JMP_JNE_X: | |
485 | if (DST != SRC) { | |
486 | insn += insn->off; | |
487 | CONT_JMP; | |
488 | } | |
489 | CONT; | |
490 | JMP_JNE_K: | |
491 | if (DST != IMM) { | |
492 | insn += insn->off; | |
493 | CONT_JMP; | |
494 | } | |
495 | CONT; | |
496 | JMP_JGT_X: | |
497 | if (DST > SRC) { | |
498 | insn += insn->off; | |
499 | CONT_JMP; | |
500 | } | |
501 | CONT; | |
502 | JMP_JGT_K: | |
503 | if (DST > IMM) { | |
504 | insn += insn->off; | |
505 | CONT_JMP; | |
506 | } | |
507 | CONT; | |
508 | JMP_JGE_X: | |
509 | if (DST >= SRC) { | |
510 | insn += insn->off; | |
511 | CONT_JMP; | |
512 | } | |
513 | CONT; | |
514 | JMP_JGE_K: | |
515 | if (DST >= IMM) { | |
516 | insn += insn->off; | |
517 | CONT_JMP; | |
518 | } | |
519 | CONT; | |
520 | JMP_JSGT_X: | |
521 | if (((s64) DST) > ((s64) SRC)) { | |
522 | insn += insn->off; | |
523 | CONT_JMP; | |
524 | } | |
525 | CONT; | |
526 | JMP_JSGT_K: | |
527 | if (((s64) DST) > ((s64) IMM)) { | |
528 | insn += insn->off; | |
529 | CONT_JMP; | |
530 | } | |
531 | CONT; | |
532 | JMP_JSGE_X: | |
533 | if (((s64) DST) >= ((s64) SRC)) { | |
534 | insn += insn->off; | |
535 | CONT_JMP; | |
536 | } | |
537 | CONT; | |
538 | JMP_JSGE_K: | |
539 | if (((s64) DST) >= ((s64) IMM)) { | |
540 | insn += insn->off; | |
541 | CONT_JMP; | |
542 | } | |
543 | CONT; | |
544 | JMP_JSET_X: | |
545 | if (DST & SRC) { | |
546 | insn += insn->off; | |
547 | CONT_JMP; | |
548 | } | |
549 | CONT; | |
550 | JMP_JSET_K: | |
551 | if (DST & IMM) { | |
552 | insn += insn->off; | |
553 | CONT_JMP; | |
554 | } | |
555 | CONT; | |
556 | JMP_EXIT: | |
557 | return BPF_R0; | |
558 | ||
559 | /* STX and ST and LDX*/ | |
560 | #define LDST(SIZEOP, SIZE) \ | |
561 | STX_MEM_##SIZEOP: \ | |
562 | *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \ | |
563 | CONT; \ | |
564 | ST_MEM_##SIZEOP: \ | |
565 | *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \ | |
566 | CONT; \ | |
567 | LDX_MEM_##SIZEOP: \ | |
568 | DST = *(SIZE *)(unsigned long) (SRC + insn->off); \ | |
569 | CONT; | |
570 | ||
571 | LDST(B, u8) | |
572 | LDST(H, u16) | |
573 | LDST(W, u32) | |
574 | LDST(DW, u64) | |
575 | #undef LDST | |
576 | STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */ | |
577 | atomic_add((u32) SRC, (atomic_t *)(unsigned long) | |
578 | (DST + insn->off)); | |
579 | CONT; | |
580 | STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */ | |
581 | atomic64_add((u64) SRC, (atomic64_t *)(unsigned long) | |
582 | (DST + insn->off)); | |
583 | CONT; | |
584 | LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */ | |
585 | off = IMM; | |
586 | load_word: | |
587 | /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are | |
588 | * only appearing in the programs where ctx == | |
589 | * skb. All programs keep 'ctx' in regs[BPF_REG_CTX] | |
8fb575ca | 590 | * == BPF_R6, bpf_convert_filter() saves it in BPF_R6, |
f5bffecd AS |
591 | * internal BPF verifier will check that BPF_R6 == |
592 | * ctx. | |
593 | * | |
594 | * BPF_ABS and BPF_IND are wrappers of function calls, | |
595 | * so they scratch BPF_R1-BPF_R5 registers, preserve | |
596 | * BPF_R6-BPF_R9, and store return value into BPF_R0. | |
597 | * | |
598 | * Implicit input: | |
599 | * ctx == skb == BPF_R6 == CTX | |
600 | * | |
601 | * Explicit input: | |
602 | * SRC == any register | |
603 | * IMM == 32-bit immediate | |
604 | * | |
605 | * Output: | |
606 | * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness | |
607 | */ | |
608 | ||
609 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp); | |
610 | if (likely(ptr != NULL)) { | |
611 | BPF_R0 = get_unaligned_be32(ptr); | |
612 | CONT; | |
613 | } | |
614 | ||
615 | return 0; | |
616 | LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */ | |
617 | off = IMM; | |
618 | load_half: | |
619 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp); | |
620 | if (likely(ptr != NULL)) { | |
621 | BPF_R0 = get_unaligned_be16(ptr); | |
622 | CONT; | |
623 | } | |
624 | ||
625 | return 0; | |
626 | LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */ | |
627 | off = IMM; | |
628 | load_byte: | |
629 | ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp); | |
630 | if (likely(ptr != NULL)) { | |
631 | BPF_R0 = *(u8 *)ptr; | |
632 | CONT; | |
633 | } | |
634 | ||
635 | return 0; | |
636 | LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */ | |
637 | off = IMM + SRC; | |
638 | goto load_word; | |
639 | LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */ | |
640 | off = IMM + SRC; | |
641 | goto load_half; | |
642 | LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */ | |
643 | off = IMM + SRC; | |
644 | goto load_byte; | |
645 | ||
646 | default_label: | |
647 | /* If we ever reach this, we have a bug somewhere. */ | |
648 | WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code); | |
649 | return 0; | |
650 | } | |
39853cc0 | 651 | STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */ |
f5bffecd | 652 | |
3324b584 DB |
653 | bool bpf_prog_array_compatible(struct bpf_array *array, |
654 | const struct bpf_prog *fp) | |
04fd61ab | 655 | { |
3324b584 DB |
656 | if (!array->owner_prog_type) { |
657 | /* There's no owner yet where we could check for | |
658 | * compatibility. | |
659 | */ | |
04fd61ab AS |
660 | array->owner_prog_type = fp->type; |
661 | array->owner_jited = fp->jited; | |
3324b584 DB |
662 | |
663 | return true; | |
04fd61ab | 664 | } |
3324b584 DB |
665 | |
666 | return array->owner_prog_type == fp->type && | |
667 | array->owner_jited == fp->jited; | |
04fd61ab AS |
668 | } |
669 | ||
3324b584 | 670 | static int bpf_check_tail_call(const struct bpf_prog *fp) |
04fd61ab AS |
671 | { |
672 | struct bpf_prog_aux *aux = fp->aux; | |
673 | int i; | |
674 | ||
675 | for (i = 0; i < aux->used_map_cnt; i++) { | |
3324b584 | 676 | struct bpf_map *map = aux->used_maps[i]; |
04fd61ab | 677 | struct bpf_array *array; |
04fd61ab | 678 | |
04fd61ab AS |
679 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) |
680 | continue; | |
3324b584 | 681 | |
04fd61ab AS |
682 | array = container_of(map, struct bpf_array, map); |
683 | if (!bpf_prog_array_compatible(array, fp)) | |
684 | return -EINVAL; | |
685 | } | |
686 | ||
687 | return 0; | |
688 | } | |
689 | ||
f5bffecd | 690 | /** |
3324b584 | 691 | * bpf_prog_select_runtime - select exec runtime for BPF program |
7ae457c1 | 692 | * @fp: bpf_prog populated with internal BPF program |
f5bffecd | 693 | * |
3324b584 DB |
694 | * Try to JIT eBPF program, if JIT is not available, use interpreter. |
695 | * The BPF program will be executed via BPF_PROG_RUN() macro. | |
f5bffecd | 696 | */ |
04fd61ab | 697 | int bpf_prog_select_runtime(struct bpf_prog *fp) |
f5bffecd | 698 | { |
7ae457c1 | 699 | fp->bpf_func = (void *) __bpf_prog_run; |
f5bffecd | 700 | |
f5bffecd | 701 | bpf_int_jit_compile(fp); |
60a3b225 | 702 | bpf_prog_lock_ro(fp); |
04fd61ab | 703 | |
3324b584 DB |
704 | /* The tail call compatibility check can only be done at |
705 | * this late stage as we need to determine, if we deal | |
706 | * with JITed or non JITed program concatenations and not | |
707 | * all eBPF JITs might immediately support all features. | |
708 | */ | |
709 | return bpf_check_tail_call(fp); | |
f5bffecd | 710 | } |
7ae457c1 | 711 | EXPORT_SYMBOL_GPL(bpf_prog_select_runtime); |
f5bffecd | 712 | |
60a3b225 DB |
713 | static void bpf_prog_free_deferred(struct work_struct *work) |
714 | { | |
09756af4 | 715 | struct bpf_prog_aux *aux; |
60a3b225 | 716 | |
09756af4 AS |
717 | aux = container_of(work, struct bpf_prog_aux, work); |
718 | bpf_jit_free(aux->prog); | |
60a3b225 DB |
719 | } |
720 | ||
721 | /* Free internal BPF program */ | |
7ae457c1 | 722 | void bpf_prog_free(struct bpf_prog *fp) |
f5bffecd | 723 | { |
09756af4 | 724 | struct bpf_prog_aux *aux = fp->aux; |
60a3b225 | 725 | |
09756af4 | 726 | INIT_WORK(&aux->work, bpf_prog_free_deferred); |
09756af4 | 727 | schedule_work(&aux->work); |
f5bffecd | 728 | } |
7ae457c1 | 729 | EXPORT_SYMBOL_GPL(bpf_prog_free); |
f89b7755 | 730 | |
3ad00405 DB |
731 | /* RNG for unpriviledged user space with separated state from prandom_u32(). */ |
732 | static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state); | |
733 | ||
734 | void bpf_user_rnd_init_once(void) | |
735 | { | |
736 | prandom_init_once(&bpf_user_rnd_state); | |
737 | } | |
738 | ||
739 | u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
740 | { | |
741 | /* Should someone ever have the rather unwise idea to use some | |
742 | * of the registers passed into this function, then note that | |
743 | * this function is called from native eBPF and classic-to-eBPF | |
744 | * transformations. Register assignments from both sides are | |
745 | * different, f.e. classic always sets fn(ctx, A, X) here. | |
746 | */ | |
747 | struct rnd_state *state; | |
748 | u32 res; | |
749 | ||
750 | state = &get_cpu_var(bpf_user_rnd_state); | |
751 | res = prandom_u32_state(state); | |
752 | put_cpu_var(state); | |
753 | ||
754 | return res; | |
755 | } | |
756 | ||
3ba67dab DB |
757 | /* Weak definitions of helper functions in case we don't have bpf syscall. */ |
758 | const struct bpf_func_proto bpf_map_lookup_elem_proto __weak; | |
759 | const struct bpf_func_proto bpf_map_update_elem_proto __weak; | |
760 | const struct bpf_func_proto bpf_map_delete_elem_proto __weak; | |
761 | ||
03e69b50 | 762 | const struct bpf_func_proto bpf_get_prandom_u32_proto __weak; |
c04167ce | 763 | const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak; |
17ca8cbf | 764 | const struct bpf_func_proto bpf_ktime_get_ns_proto __weak; |
bd570ff9 | 765 | |
ffeedafb AS |
766 | const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak; |
767 | const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak; | |
768 | const struct bpf_func_proto bpf_get_current_comm_proto __weak; | |
bd570ff9 | 769 | |
0756ea3e AS |
770 | const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void) |
771 | { | |
772 | return NULL; | |
773 | } | |
03e69b50 | 774 | |
bd570ff9 DB |
775 | const struct bpf_func_proto * __weak bpf_get_event_output_proto(void) |
776 | { | |
777 | return NULL; | |
778 | } | |
779 | ||
3324b584 DB |
780 | /* Always built-in helper functions. */ |
781 | const struct bpf_func_proto bpf_tail_call_proto = { | |
782 | .func = NULL, | |
783 | .gpl_only = false, | |
784 | .ret_type = RET_VOID, | |
785 | .arg1_type = ARG_PTR_TO_CTX, | |
786 | .arg2_type = ARG_CONST_MAP_PTR, | |
787 | .arg3_type = ARG_ANYTHING, | |
788 | }; | |
789 | ||
790 | /* For classic BPF JITs that don't implement bpf_int_jit_compile(). */ | |
791 | void __weak bpf_int_jit_compile(struct bpf_prog *prog) | |
792 | { | |
793 | } | |
794 | ||
969bf05e AS |
795 | bool __weak bpf_helper_changes_skb_data(void *func) |
796 | { | |
797 | return false; | |
798 | } | |
799 | ||
f89b7755 AS |
800 | /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call |
801 | * skb_copy_bits(), so provide a weak definition of it for NET-less config. | |
802 | */ | |
803 | int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to, | |
804 | int len) | |
805 | { | |
806 | return -EFAULT; | |
807 | } |