]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux Socket Filter Data Structures | |
3 | */ | |
1da177e4 LT |
4 | #ifndef __LINUX_FILTER_H__ |
5 | #define __LINUX_FILTER_H__ | |
6 | ||
b954d834 DB |
7 | #include <stdarg.h> |
8 | ||
60063497 | 9 | #include <linux/atomic.h> |
4c355cdf | 10 | #include <linux/refcount.h> |
0c5fe1b4 | 11 | #include <linux/compat.h> |
9f12fbe6 | 12 | #include <linux/skbuff.h> |
b954d834 DB |
13 | #include <linux/linkage.h> |
14 | #include <linux/printk.h> | |
d45ed4a4 | 15 | #include <linux/workqueue.h> |
b13138ef | 16 | #include <linux/sched.h> |
4f3446bb | 17 | #include <linux/capability.h> |
7bd509e3 | 18 | #include <linux/cryptohash.h> |
4f3446bb | 19 | |
ff936a04 | 20 | #include <net/sch_generic.h> |
b954d834 | 21 | |
60a3b225 | 22 | #include <asm/cacheflush.h> |
b954d834 DB |
23 | |
24 | #include <uapi/linux/filter.h> | |
daedfb22 | 25 | #include <uapi/linux/bpf.h> |
60a3b225 DB |
26 | |
27 | struct sk_buff; | |
28 | struct sock; | |
29 | struct seccomp_data; | |
09756af4 | 30 | struct bpf_prog_aux; |
792d4b5c | 31 | |
30743837 DB |
32 | /* ArgX, context and stack frame pointer register positions. Note, |
33 | * Arg1, Arg2, Arg3, etc are used as argument mappings of function | |
34 | * calls in BPF_CALL instruction. | |
35 | */ | |
36 | #define BPF_REG_ARG1 BPF_REG_1 | |
37 | #define BPF_REG_ARG2 BPF_REG_2 | |
38 | #define BPF_REG_ARG3 BPF_REG_3 | |
39 | #define BPF_REG_ARG4 BPF_REG_4 | |
40 | #define BPF_REG_ARG5 BPF_REG_5 | |
41 | #define BPF_REG_CTX BPF_REG_6 | |
42 | #define BPF_REG_FP BPF_REG_10 | |
43 | ||
44 | /* Additional register mappings for converted user programs. */ | |
45 | #define BPF_REG_A BPF_REG_0 | |
46 | #define BPF_REG_X BPF_REG_7 | |
47 | #define BPF_REG_TMP BPF_REG_8 | |
bd4cf0ed | 48 | |
4f3446bb DB |
49 | /* Kernel hidden auxiliary/helper register for hardening step. |
50 | * Only used by eBPF JITs. It's nothing more than a temporary | |
51 | * register that JITs use internally, only that here it's part | |
52 | * of eBPF instructions that have been rewritten for blinding | |
53 | * constants. See JIT pre-step in bpf_jit_blind_constants(). | |
54 | */ | |
55 | #define BPF_REG_AX MAX_BPF_REG | |
56 | #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1) | |
57 | ||
74451e66 DB |
58 | /* As per nm, we expose JITed images as text (code) section for |
59 | * kallsyms. That way, tools like perf can find it to match | |
60 | * addresses. | |
61 | */ | |
62 | #define BPF_SYM_ELF_TYPE 't' | |
63 | ||
bd4cf0ed AS |
64 | /* BPF program can access up to 512 bytes of stack space. */ |
65 | #define MAX_BPF_STACK 512 | |
66 | ||
f1f7714e DB |
67 | #define BPF_TAG_SIZE 8 |
68 | ||
f8f6d679 DB |
69 | /* Helper macros for filter block array initializers. */ |
70 | ||
e430f34e | 71 | /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ |
f8f6d679 | 72 | |
e430f34e | 73 | #define BPF_ALU64_REG(OP, DST, SRC) \ |
2695fb55 | 74 | ((struct bpf_insn) { \ |
f8f6d679 | 75 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
76 | .dst_reg = DST, \ |
77 | .src_reg = SRC, \ | |
f8f6d679 DB |
78 | .off = 0, \ |
79 | .imm = 0 }) | |
80 | ||
e430f34e | 81 | #define BPF_ALU32_REG(OP, DST, SRC) \ |
2695fb55 | 82 | ((struct bpf_insn) { \ |
f8f6d679 | 83 | .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
84 | .dst_reg = DST, \ |
85 | .src_reg = SRC, \ | |
f8f6d679 DB |
86 | .off = 0, \ |
87 | .imm = 0 }) | |
88 | ||
e430f34e | 89 | /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ |
f8f6d679 | 90 | |
e430f34e | 91 | #define BPF_ALU64_IMM(OP, DST, IMM) \ |
2695fb55 | 92 | ((struct bpf_insn) { \ |
f8f6d679 | 93 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
94 | .dst_reg = DST, \ |
95 | .src_reg = 0, \ | |
f8f6d679 DB |
96 | .off = 0, \ |
97 | .imm = IMM }) | |
98 | ||
e430f34e | 99 | #define BPF_ALU32_IMM(OP, DST, IMM) \ |
2695fb55 | 100 | ((struct bpf_insn) { \ |
f8f6d679 | 101 | .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
102 | .dst_reg = DST, \ |
103 | .src_reg = 0, \ | |
f8f6d679 DB |
104 | .off = 0, \ |
105 | .imm = IMM }) | |
106 | ||
107 | /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ | |
108 | ||
e430f34e | 109 | #define BPF_ENDIAN(TYPE, DST, LEN) \ |
2695fb55 | 110 | ((struct bpf_insn) { \ |
f8f6d679 | 111 | .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ |
e430f34e AS |
112 | .dst_reg = DST, \ |
113 | .src_reg = 0, \ | |
f8f6d679 DB |
114 | .off = 0, \ |
115 | .imm = LEN }) | |
116 | ||
e430f34e | 117 | /* Short form of mov, dst_reg = src_reg */ |
f8f6d679 | 118 | |
e430f34e | 119 | #define BPF_MOV64_REG(DST, SRC) \ |
2695fb55 | 120 | ((struct bpf_insn) { \ |
f8f6d679 | 121 | .code = BPF_ALU64 | BPF_MOV | BPF_X, \ |
e430f34e AS |
122 | .dst_reg = DST, \ |
123 | .src_reg = SRC, \ | |
f8f6d679 DB |
124 | .off = 0, \ |
125 | .imm = 0 }) | |
126 | ||
e430f34e | 127 | #define BPF_MOV32_REG(DST, SRC) \ |
2695fb55 | 128 | ((struct bpf_insn) { \ |
f8f6d679 | 129 | .code = BPF_ALU | BPF_MOV | BPF_X, \ |
e430f34e AS |
130 | .dst_reg = DST, \ |
131 | .src_reg = SRC, \ | |
f8f6d679 DB |
132 | .off = 0, \ |
133 | .imm = 0 }) | |
134 | ||
e430f34e | 135 | /* Short form of mov, dst_reg = imm32 */ |
f8f6d679 | 136 | |
e430f34e | 137 | #define BPF_MOV64_IMM(DST, IMM) \ |
2695fb55 | 138 | ((struct bpf_insn) { \ |
f8f6d679 | 139 | .code = BPF_ALU64 | BPF_MOV | BPF_K, \ |
e430f34e AS |
140 | .dst_reg = DST, \ |
141 | .src_reg = 0, \ | |
f8f6d679 DB |
142 | .off = 0, \ |
143 | .imm = IMM }) | |
144 | ||
e430f34e | 145 | #define BPF_MOV32_IMM(DST, IMM) \ |
2695fb55 | 146 | ((struct bpf_insn) { \ |
f8f6d679 | 147 | .code = BPF_ALU | BPF_MOV | BPF_K, \ |
e430f34e AS |
148 | .dst_reg = DST, \ |
149 | .src_reg = 0, \ | |
f8f6d679 DB |
150 | .off = 0, \ |
151 | .imm = IMM }) | |
152 | ||
02ab695b AS |
153 | /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ |
154 | #define BPF_LD_IMM64(DST, IMM) \ | |
155 | BPF_LD_IMM64_RAW(DST, 0, IMM) | |
156 | ||
157 | #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ | |
158 | ((struct bpf_insn) { \ | |
159 | .code = BPF_LD | BPF_DW | BPF_IMM, \ | |
160 | .dst_reg = DST, \ | |
161 | .src_reg = SRC, \ | |
162 | .off = 0, \ | |
163 | .imm = (__u32) (IMM) }), \ | |
164 | ((struct bpf_insn) { \ | |
165 | .code = 0, /* zero is reserved opcode */ \ | |
166 | .dst_reg = 0, \ | |
167 | .src_reg = 0, \ | |
168 | .off = 0, \ | |
169 | .imm = ((__u64) (IMM)) >> 32 }) | |
170 | ||
0246e64d AS |
171 | /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ |
172 | #define BPF_LD_MAP_FD(DST, MAP_FD) \ | |
173 | BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) | |
174 | ||
e430f34e | 175 | /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ |
f8f6d679 | 176 | |
e430f34e | 177 | #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 178 | ((struct bpf_insn) { \ |
f8f6d679 | 179 | .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
180 | .dst_reg = DST, \ |
181 | .src_reg = SRC, \ | |
f8f6d679 DB |
182 | .off = 0, \ |
183 | .imm = IMM }) | |
184 | ||
e430f34e | 185 | #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 186 | ((struct bpf_insn) { \ |
f8f6d679 | 187 | .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
188 | .dst_reg = DST, \ |
189 | .src_reg = SRC, \ | |
f8f6d679 DB |
190 | .off = 0, \ |
191 | .imm = IMM }) | |
192 | ||
e430f34e | 193 | /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ |
f8f6d679 | 194 | |
e430f34e | 195 | #define BPF_LD_ABS(SIZE, IMM) \ |
2695fb55 | 196 | ((struct bpf_insn) { \ |
f8f6d679 | 197 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ |
e430f34e AS |
198 | .dst_reg = 0, \ |
199 | .src_reg = 0, \ | |
f8f6d679 | 200 | .off = 0, \ |
e430f34e | 201 | .imm = IMM }) |
f8f6d679 | 202 | |
e430f34e | 203 | /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ |
f8f6d679 | 204 | |
e430f34e | 205 | #define BPF_LD_IND(SIZE, SRC, IMM) \ |
2695fb55 | 206 | ((struct bpf_insn) { \ |
f8f6d679 | 207 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ |
e430f34e AS |
208 | .dst_reg = 0, \ |
209 | .src_reg = SRC, \ | |
f8f6d679 | 210 | .off = 0, \ |
e430f34e | 211 | .imm = IMM }) |
f8f6d679 | 212 | |
e430f34e | 213 | /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ |
f8f6d679 | 214 | |
e430f34e | 215 | #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ |
2695fb55 | 216 | ((struct bpf_insn) { \ |
f8f6d679 | 217 | .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
218 | .dst_reg = DST, \ |
219 | .src_reg = SRC, \ | |
f8f6d679 DB |
220 | .off = OFF, \ |
221 | .imm = 0 }) | |
222 | ||
e430f34e AS |
223 | /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ |
224 | ||
225 | #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ | |
2695fb55 | 226 | ((struct bpf_insn) { \ |
f8f6d679 | 227 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
228 | .dst_reg = DST, \ |
229 | .src_reg = SRC, \ | |
f8f6d679 DB |
230 | .off = OFF, \ |
231 | .imm = 0 }) | |
232 | ||
cffc642d MH |
233 | /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */ |
234 | ||
235 | #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \ | |
236 | ((struct bpf_insn) { \ | |
237 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \ | |
238 | .dst_reg = DST, \ | |
239 | .src_reg = SRC, \ | |
240 | .off = OFF, \ | |
241 | .imm = 0 }) | |
242 | ||
e430f34e AS |
243 | /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ |
244 | ||
245 | #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ | |
2695fb55 | 246 | ((struct bpf_insn) { \ |
e430f34e AS |
247 | .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ |
248 | .dst_reg = DST, \ | |
249 | .src_reg = 0, \ | |
250 | .off = OFF, \ | |
251 | .imm = IMM }) | |
252 | ||
253 | /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ | |
f8f6d679 | 254 | |
e430f34e | 255 | #define BPF_JMP_REG(OP, DST, SRC, OFF) \ |
2695fb55 | 256 | ((struct bpf_insn) { \ |
f8f6d679 | 257 | .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
258 | .dst_reg = DST, \ |
259 | .src_reg = SRC, \ | |
f8f6d679 DB |
260 | .off = OFF, \ |
261 | .imm = 0 }) | |
262 | ||
e430f34e | 263 | /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ |
f8f6d679 | 264 | |
e430f34e | 265 | #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ |
2695fb55 | 266 | ((struct bpf_insn) { \ |
f8f6d679 | 267 | .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
268 | .dst_reg = DST, \ |
269 | .src_reg = 0, \ | |
f8f6d679 DB |
270 | .off = OFF, \ |
271 | .imm = IMM }) | |
272 | ||
273 | /* Function call */ | |
274 | ||
275 | #define BPF_EMIT_CALL(FUNC) \ | |
2695fb55 | 276 | ((struct bpf_insn) { \ |
f8f6d679 | 277 | .code = BPF_JMP | BPF_CALL, \ |
e430f34e AS |
278 | .dst_reg = 0, \ |
279 | .src_reg = 0, \ | |
f8f6d679 DB |
280 | .off = 0, \ |
281 | .imm = ((FUNC) - __bpf_call_base) }) | |
282 | ||
283 | /* Raw code statement block */ | |
284 | ||
e430f34e | 285 | #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ |
2695fb55 | 286 | ((struct bpf_insn) { \ |
f8f6d679 | 287 | .code = CODE, \ |
e430f34e AS |
288 | .dst_reg = DST, \ |
289 | .src_reg = SRC, \ | |
f8f6d679 DB |
290 | .off = OFF, \ |
291 | .imm = IMM }) | |
292 | ||
293 | /* Program exit */ | |
294 | ||
295 | #define BPF_EXIT_INSN() \ | |
2695fb55 | 296 | ((struct bpf_insn) { \ |
f8f6d679 | 297 | .code = BPF_JMP | BPF_EXIT, \ |
e430f34e AS |
298 | .dst_reg = 0, \ |
299 | .src_reg = 0, \ | |
f8f6d679 DB |
300 | .off = 0, \ |
301 | .imm = 0 }) | |
302 | ||
a4afd37b DB |
303 | /* Internal classic blocks for direct assignment */ |
304 | ||
305 | #define __BPF_STMT(CODE, K) \ | |
306 | ((struct sock_filter) BPF_STMT(CODE, K)) | |
307 | ||
308 | #define __BPF_JUMP(CODE, K, JT, JF) \ | |
309 | ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) | |
310 | ||
f8f6d679 DB |
311 | #define bytes_to_bpf_size(bytes) \ |
312 | ({ \ | |
313 | int bpf_size = -EINVAL; \ | |
314 | \ | |
315 | if (bytes == sizeof(u8)) \ | |
316 | bpf_size = BPF_B; \ | |
317 | else if (bytes == sizeof(u16)) \ | |
318 | bpf_size = BPF_H; \ | |
319 | else if (bytes == sizeof(u32)) \ | |
320 | bpf_size = BPF_W; \ | |
321 | else if (bytes == sizeof(u64)) \ | |
322 | bpf_size = BPF_DW; \ | |
323 | \ | |
324 | bpf_size; \ | |
325 | }) | |
9739eef1 | 326 | |
f035a515 DB |
327 | #define BPF_SIZEOF(type) \ |
328 | ({ \ | |
329 | const int __size = bytes_to_bpf_size(sizeof(type)); \ | |
330 | BUILD_BUG_ON(__size < 0); \ | |
331 | __size; \ | |
332 | }) | |
333 | ||
334 | #define BPF_FIELD_SIZEOF(type, field) \ | |
335 | ({ \ | |
336 | const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ | |
337 | BUILD_BUG_ON(__size < 0); \ | |
338 | __size; \ | |
339 | }) | |
340 | ||
f3694e00 DB |
341 | #define __BPF_MAP_0(m, v, ...) v |
342 | #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) | |
343 | #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) | |
344 | #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) | |
345 | #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) | |
346 | #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) | |
347 | ||
348 | #define __BPF_REG_0(...) __BPF_PAD(5) | |
349 | #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) | |
350 | #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) | |
351 | #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) | |
352 | #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) | |
353 | #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) | |
354 | ||
355 | #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) | |
356 | #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) | |
357 | ||
358 | #define __BPF_CAST(t, a) \ | |
359 | (__force t) \ | |
360 | (__force \ | |
361 | typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ | |
362 | (unsigned long)0, (t)0))) a | |
363 | #define __BPF_V void | |
364 | #define __BPF_N | |
365 | ||
366 | #define __BPF_DECL_ARGS(t, a) t a | |
367 | #define __BPF_DECL_REGS(t, a) u64 a | |
368 | ||
369 | #define __BPF_PAD(n) \ | |
370 | __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ | |
371 | u64, __ur_3, u64, __ur_4, u64, __ur_5) | |
372 | ||
373 | #define BPF_CALL_x(x, name, ...) \ | |
374 | static __always_inline \ | |
375 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ | |
376 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ | |
377 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ | |
378 | { \ | |
379 | return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ | |
380 | } \ | |
381 | static __always_inline \ | |
382 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) | |
383 | ||
384 | #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) | |
385 | #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) | |
386 | #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) | |
387 | #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) | |
388 | #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) | |
389 | #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) | |
390 | ||
bd4cf0ed AS |
391 | #ifdef CONFIG_COMPAT |
392 | /* A struct sock_filter is architecture independent. */ | |
0c5fe1b4 WD |
393 | struct compat_sock_fprog { |
394 | u16 len; | |
bd4cf0ed | 395 | compat_uptr_t filter; /* struct sock_filter * */ |
0c5fe1b4 WD |
396 | }; |
397 | #endif | |
398 | ||
a3ea269b DB |
399 | struct sock_fprog_kern { |
400 | u16 len; | |
401 | struct sock_filter *filter; | |
402 | }; | |
403 | ||
738cbe72 DB |
404 | struct bpf_binary_header { |
405 | unsigned int pages; | |
406 | u8 image[]; | |
407 | }; | |
408 | ||
7ae457c1 | 409 | struct bpf_prog { |
286aad3c | 410 | u16 pages; /* Number of allocated pages */ |
a91263d5 DB |
411 | kmemcheck_bitfield_begin(meta); |
412 | u16 jited:1, /* Is our filter JIT'ed? */ | |
65869a47 | 413 | locked:1, /* Program image locked? */ |
c46646d0 | 414 | gpl_compatible:1, /* Is filter GPL compatible? */ |
ff936a04 | 415 | cb_access:1, /* Is control block accessed? */ |
e390b55d | 416 | dst_needed:1; /* Do we need dst entry? */ |
a91263d5 | 417 | kmemcheck_bitfield_end(meta); |
24701ece | 418 | enum bpf_prog_type type; /* Type of BPF program */ |
7bd509e3 | 419 | u32 len; /* Number of filter blocks */ |
f1f7714e | 420 | u8 tag[BPF_TAG_SIZE]; |
09756af4 | 421 | struct bpf_prog_aux *aux; /* Auxiliary fields */ |
24701ece | 422 | struct sock_fprog_kern *orig_prog; /* Original BPF program */ |
88575199 DB |
423 | unsigned int (*bpf_func)(const void *ctx, |
424 | const struct bpf_insn *insn); | |
60a3b225 | 425 | /* Instructions for interpreter */ |
d45ed4a4 | 426 | union { |
bd4cf0ed | 427 | struct sock_filter insns[0]; |
2695fb55 | 428 | struct bpf_insn insnsi[0]; |
d45ed4a4 | 429 | }; |
b715631f SH |
430 | }; |
431 | ||
7ae457c1 | 432 | struct sk_filter { |
4c355cdf | 433 | refcount_t refcnt; |
7ae457c1 AS |
434 | struct rcu_head rcu; |
435 | struct bpf_prog *prog; | |
436 | }; | |
437 | ||
438 | #define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi) | |
439 | ||
01dd194c DB |
440 | #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN |
441 | ||
db58ba45 AS |
442 | struct bpf_skb_data_end { |
443 | struct qdisc_skb_cb qdisc_cb; | |
444 | void *data_end; | |
445 | }; | |
446 | ||
6a773a15 BB |
447 | struct xdp_buff { |
448 | void *data; | |
449 | void *data_end; | |
17bedab2 | 450 | void *data_hard_start; |
6a773a15 BB |
451 | }; |
452 | ||
db58ba45 | 453 | /* compute the linear packet data range [data, data_end) which |
3a0af8fd | 454 | * will be accessed by cls_bpf, act_bpf and lwt programs |
db58ba45 AS |
455 | */ |
456 | static inline void bpf_compute_data_end(struct sk_buff *skb) | |
457 | { | |
458 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; | |
459 | ||
460 | BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb)); | |
461 | cb->data_end = skb->data + skb_headlen(skb); | |
462 | } | |
463 | ||
01dd194c DB |
464 | static inline u8 *bpf_skb_cb(struct sk_buff *skb) |
465 | { | |
466 | /* eBPF programs may read/write skb->cb[] area to transfer meta | |
467 | * data between tail calls. Since this also needs to work with | |
468 | * tc, that scratch memory is mapped to qdisc_skb_cb's data area. | |
469 | * | |
470 | * In some socket filter cases, the cb unfortunately needs to be | |
471 | * saved/restored so that protocol specific skb->cb[] data won't | |
472 | * be lost. In any case, due to unpriviledged eBPF programs | |
473 | * attached to sockets, we need to clear the bpf_skb_cb() area | |
474 | * to not leak previous contents to user space. | |
475 | */ | |
476 | BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN); | |
477 | BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != | |
478 | FIELD_SIZEOF(struct qdisc_skb_cb, data)); | |
479 | ||
480 | return qdisc_skb_cb(skb)->data; | |
481 | } | |
482 | ||
ff936a04 AS |
483 | static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, |
484 | struct sk_buff *skb) | |
485 | { | |
01dd194c DB |
486 | u8 *cb_data = bpf_skb_cb(skb); |
487 | u8 cb_saved[BPF_SKB_CB_LEN]; | |
ff936a04 AS |
488 | u32 res; |
489 | ||
ff936a04 | 490 | if (unlikely(prog->cb_access)) { |
01dd194c DB |
491 | memcpy(cb_saved, cb_data, sizeof(cb_saved)); |
492 | memset(cb_data, 0, sizeof(cb_saved)); | |
ff936a04 AS |
493 | } |
494 | ||
495 | res = BPF_PROG_RUN(prog, skb); | |
496 | ||
497 | if (unlikely(prog->cb_access)) | |
01dd194c | 498 | memcpy(cb_data, cb_saved, sizeof(cb_saved)); |
ff936a04 AS |
499 | |
500 | return res; | |
501 | } | |
502 | ||
503 | static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, | |
504 | struct sk_buff *skb) | |
505 | { | |
01dd194c | 506 | u8 *cb_data = bpf_skb_cb(skb); |
ff936a04 AS |
507 | |
508 | if (unlikely(prog->cb_access)) | |
01dd194c DB |
509 | memset(cb_data, 0, BPF_SKB_CB_LEN); |
510 | ||
ff936a04 AS |
511 | return BPF_PROG_RUN(prog, skb); |
512 | } | |
513 | ||
366cbf2f DB |
514 | static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, |
515 | struct xdp_buff *xdp) | |
6a773a15 | 516 | { |
366cbf2f DB |
517 | /* Caller needs to hold rcu_read_lock() (!), otherwise program |
518 | * can be released while still running, or map elements could be | |
519 | * freed early while still having concurrent users. XDP fastpath | |
520 | * already takes rcu_read_lock() when fetching the program, so | |
521 | * it's not necessary here anymore. | |
522 | */ | |
523 | return BPF_PROG_RUN(prog, xdp); | |
6a773a15 BB |
524 | } |
525 | ||
aafe6ae9 DB |
526 | static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) |
527 | { | |
528 | return prog->len * sizeof(struct bpf_insn); | |
529 | } | |
530 | ||
f1f7714e | 531 | static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) |
aafe6ae9 DB |
532 | { |
533 | return round_up(bpf_prog_insn_size(prog) + | |
534 | sizeof(__be64) + 1, SHA_MESSAGE_BYTES); | |
535 | } | |
536 | ||
7ae457c1 | 537 | static inline unsigned int bpf_prog_size(unsigned int proglen) |
b715631f | 538 | { |
7ae457c1 AS |
539 | return max(sizeof(struct bpf_prog), |
540 | offsetof(struct bpf_prog, insns[proglen])); | |
b715631f SH |
541 | } |
542 | ||
7b36f929 DB |
543 | static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) |
544 | { | |
545 | /* When classic BPF programs have been loaded and the arch | |
546 | * does not have a classic BPF JIT (anymore), they have been | |
547 | * converted via bpf_migrate_filter() to eBPF and thus always | |
548 | * have an unspec program type. | |
549 | */ | |
550 | return prog->type == BPF_PROG_TYPE_UNSPEC; | |
551 | } | |
552 | ||
009937e7 | 553 | #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) |
a3ea269b | 554 | |
9d876e79 | 555 | #ifdef CONFIG_ARCH_HAS_SET_MEMORY |
60a3b225 DB |
556 | static inline void bpf_prog_lock_ro(struct bpf_prog *fp) |
557 | { | |
65869a47 DB |
558 | fp->locked = 1; |
559 | WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages)); | |
60a3b225 DB |
560 | } |
561 | ||
562 | static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) | |
563 | { | |
65869a47 DB |
564 | if (fp->locked) { |
565 | WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages)); | |
566 | /* In case set_memory_rw() fails, we want to be the first | |
567 | * to crash here instead of some random place later on. | |
568 | */ | |
569 | fp->locked = 0; | |
570 | } | |
60a3b225 | 571 | } |
74451e66 | 572 | |
9d876e79 DB |
573 | static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
574 | { | |
65869a47 | 575 | WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages)); |
9d876e79 DB |
576 | } |
577 | ||
74451e66 DB |
578 | static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) |
579 | { | |
65869a47 | 580 | WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages)); |
74451e66 | 581 | } |
60a3b225 DB |
582 | #else |
583 | static inline void bpf_prog_lock_ro(struct bpf_prog *fp) | |
584 | { | |
585 | } | |
586 | ||
587 | static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) | |
588 | { | |
589 | } | |
74451e66 | 590 | |
9d876e79 DB |
591 | static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
592 | { | |
593 | } | |
594 | ||
74451e66 DB |
595 | static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) |
596 | { | |
597 | } | |
9d876e79 | 598 | #endif /* CONFIG_ARCH_HAS_SET_MEMORY */ |
60a3b225 | 599 | |
74451e66 DB |
600 | static inline struct bpf_binary_header * |
601 | bpf_jit_binary_hdr(const struct bpf_prog *fp) | |
602 | { | |
603 | unsigned long real_start = (unsigned long)fp->bpf_func; | |
604 | unsigned long addr = real_start & PAGE_MASK; | |
605 | ||
606 | return (void *)addr; | |
607 | } | |
608 | ||
f4979fce WB |
609 | int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); |
610 | static inline int sk_filter(struct sock *sk, struct sk_buff *skb) | |
611 | { | |
612 | return sk_filter_trim_cap(sk, skb, 1); | |
613 | } | |
bd4cf0ed | 614 | |
d1c55ab5 | 615 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); |
7ae457c1 | 616 | void bpf_prog_free(struct bpf_prog *fp); |
bd4cf0ed | 617 | |
60a3b225 DB |
618 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); |
619 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, | |
620 | gfp_t gfp_extra_flags); | |
621 | void __bpf_prog_free(struct bpf_prog *fp); | |
622 | ||
623 | static inline void bpf_prog_unlock_free(struct bpf_prog *fp) | |
624 | { | |
625 | bpf_prog_unlock_ro(fp); | |
626 | __bpf_prog_free(fp); | |
627 | } | |
628 | ||
ac67eb2c DB |
629 | typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, |
630 | unsigned int flen); | |
631 | ||
7ae457c1 | 632 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); |
ac67eb2c | 633 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, |
bab18991 | 634 | bpf_aux_classic_check_t trans, bool save_orig); |
7ae457c1 | 635 | void bpf_prog_destroy(struct bpf_prog *fp); |
a3ea269b | 636 | |
fbc907f0 | 637 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
89aa0758 | 638 | int sk_attach_bpf(u32 ufd, struct sock *sk); |
538950a1 CG |
639 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
640 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); | |
fbc907f0 | 641 | int sk_detach_filter(struct sock *sk); |
fbc907f0 DB |
642 | int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, |
643 | unsigned int len); | |
fbc907f0 | 644 | |
278571ba | 645 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); |
fbc907f0 | 646 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); |
0a14842f | 647 | |
62258278 | 648 | u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
d1c55ab5 DB |
649 | |
650 | struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); | |
9383191d | 651 | void bpf_jit_compile(struct bpf_prog *prog); |
17bedab2 | 652 | bool bpf_helper_changes_pkt_data(void *func); |
62258278 | 653 | |
c237ee5e DB |
654 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
655 | const struct bpf_insn *patch, u32 len); | |
6a773a15 | 656 | void bpf_warn_invalid_xdp_action(u32 act); |
c237ee5e | 657 | |
b954d834 | 658 | #ifdef CONFIG_BPF_JIT |
c94987e4 | 659 | extern int bpf_jit_enable; |
4f3446bb | 660 | extern int bpf_jit_harden; |
74451e66 | 661 | extern int bpf_jit_kallsyms; |
c94987e4 | 662 | |
b954d834 DB |
663 | typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); |
664 | ||
665 | struct bpf_binary_header * | |
666 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
667 | unsigned int alignment, | |
668 | bpf_jit_fill_hole_t bpf_fill_ill_insns); | |
669 | void bpf_jit_binary_free(struct bpf_binary_header *hdr); | |
670 | ||
b954d834 DB |
671 | void bpf_jit_free(struct bpf_prog *fp); |
672 | ||
4f3446bb DB |
673 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); |
674 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); | |
675 | ||
b954d834 DB |
676 | static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, |
677 | u32 pass, void *image) | |
678 | { | |
b13138ef DB |
679 | pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, |
680 | proglen, pass, image, current->comm, task_pid_nr(current)); | |
681 | ||
b954d834 DB |
682 | if (image) |
683 | print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, | |
684 | 16, 1, image, proglen, false); | |
685 | } | |
4f3446bb DB |
686 | |
687 | static inline bool bpf_jit_is_ebpf(void) | |
688 | { | |
689 | # ifdef CONFIG_HAVE_EBPF_JIT | |
690 | return true; | |
691 | # else | |
692 | return false; | |
693 | # endif | |
694 | } | |
695 | ||
81ed18ab AS |
696 | static inline bool ebpf_jit_enabled(void) |
697 | { | |
698 | return bpf_jit_enable && bpf_jit_is_ebpf(); | |
699 | } | |
700 | ||
74451e66 DB |
701 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
702 | { | |
703 | return fp->jited && bpf_jit_is_ebpf(); | |
704 | } | |
705 | ||
4f3446bb DB |
706 | static inline bool bpf_jit_blinding_enabled(void) |
707 | { | |
708 | /* These are the prerequisites, should someone ever have the | |
709 | * idea to call blinding outside of them, we make sure to | |
710 | * bail out. | |
711 | */ | |
712 | if (!bpf_jit_is_ebpf()) | |
713 | return false; | |
714 | if (!bpf_jit_enable) | |
715 | return false; | |
716 | if (!bpf_jit_harden) | |
717 | return false; | |
718 | if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) | |
719 | return false; | |
720 | ||
721 | return true; | |
722 | } | |
74451e66 DB |
723 | |
724 | static inline bool bpf_jit_kallsyms_enabled(void) | |
725 | { | |
726 | /* There are a couple of corner cases where kallsyms should | |
727 | * not be enabled f.e. on hardening. | |
728 | */ | |
729 | if (bpf_jit_harden) | |
730 | return false; | |
731 | if (!bpf_jit_kallsyms) | |
732 | return false; | |
733 | if (bpf_jit_kallsyms == 1) | |
734 | return true; | |
735 | ||
736 | return false; | |
737 | } | |
738 | ||
739 | const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, | |
740 | unsigned long *off, char *sym); | |
741 | bool is_bpf_text_address(unsigned long addr); | |
742 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
743 | char *sym); | |
744 | ||
745 | static inline const char * | |
746 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
747 | unsigned long *off, char **modname, char *sym) | |
748 | { | |
749 | const char *ret = __bpf_address_lookup(addr, size, off, sym); | |
750 | ||
751 | if (ret && modname) | |
752 | *modname = NULL; | |
753 | return ret; | |
754 | } | |
755 | ||
756 | void bpf_prog_kallsyms_add(struct bpf_prog *fp); | |
757 | void bpf_prog_kallsyms_del(struct bpf_prog *fp); | |
758 | ||
759 | #else /* CONFIG_BPF_JIT */ | |
760 | ||
81ed18ab AS |
761 | static inline bool ebpf_jit_enabled(void) |
762 | { | |
763 | return false; | |
764 | } | |
765 | ||
74451e66 DB |
766 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
767 | { | |
768 | return false; | |
769 | } | |
770 | ||
b954d834 DB |
771 | static inline void bpf_jit_free(struct bpf_prog *fp) |
772 | { | |
773 | bpf_prog_unlock_free(fp); | |
774 | } | |
74451e66 DB |
775 | |
776 | static inline bool bpf_jit_kallsyms_enabled(void) | |
777 | { | |
778 | return false; | |
779 | } | |
780 | ||
781 | static inline const char * | |
782 | __bpf_address_lookup(unsigned long addr, unsigned long *size, | |
783 | unsigned long *off, char *sym) | |
784 | { | |
785 | return NULL; | |
786 | } | |
787 | ||
788 | static inline bool is_bpf_text_address(unsigned long addr) | |
789 | { | |
790 | return false; | |
791 | } | |
792 | ||
793 | static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, | |
794 | char *type, char *sym) | |
795 | { | |
796 | return -ERANGE; | |
797 | } | |
798 | ||
799 | static inline const char * | |
800 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
801 | unsigned long *off, char **modname, char *sym) | |
802 | { | |
803 | return NULL; | |
804 | } | |
805 | ||
806 | static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) | |
807 | { | |
808 | } | |
809 | ||
810 | static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) | |
811 | { | |
812 | } | |
b954d834 DB |
813 | #endif /* CONFIG_BPF_JIT */ |
814 | ||
34805931 DB |
815 | #define BPF_ANC BIT(15) |
816 | ||
55795ef5 RV |
817 | static inline bool bpf_needs_clear_a(const struct sock_filter *first) |
818 | { | |
819 | switch (first->code) { | |
820 | case BPF_RET | BPF_K: | |
821 | case BPF_LD | BPF_W | BPF_LEN: | |
822 | return false; | |
823 | ||
824 | case BPF_LD | BPF_W | BPF_ABS: | |
825 | case BPF_LD | BPF_H | BPF_ABS: | |
826 | case BPF_LD | BPF_B | BPF_ABS: | |
827 | if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) | |
828 | return true; | |
829 | return false; | |
830 | ||
831 | default: | |
832 | return true; | |
833 | } | |
834 | } | |
835 | ||
34805931 DB |
836 | static inline u16 bpf_anc_helper(const struct sock_filter *ftest) |
837 | { | |
838 | BUG_ON(ftest->code & BPF_ANC); | |
839 | ||
840 | switch (ftest->code) { | |
841 | case BPF_LD | BPF_W | BPF_ABS: | |
842 | case BPF_LD | BPF_H | BPF_ABS: | |
843 | case BPF_LD | BPF_B | BPF_ABS: | |
844 | #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ | |
845 | return BPF_ANC | SKF_AD_##CODE | |
846 | switch (ftest->k) { | |
847 | BPF_ANCILLARY(PROTOCOL); | |
848 | BPF_ANCILLARY(PKTTYPE); | |
849 | BPF_ANCILLARY(IFINDEX); | |
850 | BPF_ANCILLARY(NLATTR); | |
851 | BPF_ANCILLARY(NLATTR_NEST); | |
852 | BPF_ANCILLARY(MARK); | |
853 | BPF_ANCILLARY(QUEUE); | |
854 | BPF_ANCILLARY(HATYPE); | |
855 | BPF_ANCILLARY(RXHASH); | |
856 | BPF_ANCILLARY(CPU); | |
857 | BPF_ANCILLARY(ALU_XOR_X); | |
858 | BPF_ANCILLARY(VLAN_TAG); | |
859 | BPF_ANCILLARY(VLAN_TAG_PRESENT); | |
860 | BPF_ANCILLARY(PAY_OFFSET); | |
861 | BPF_ANCILLARY(RANDOM); | |
27cd5452 | 862 | BPF_ANCILLARY(VLAN_TPID); |
34805931 DB |
863 | } |
864 | /* Fallthrough. */ | |
865 | default: | |
866 | return ftest->code; | |
867 | } | |
868 | } | |
869 | ||
9f12fbe6 ZSL |
870 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, |
871 | int k, unsigned int size); | |
872 | ||
873 | static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, | |
874 | unsigned int size, void *buffer) | |
875 | { | |
876 | if (k >= 0) | |
877 | return skb_header_pointer(skb, k, size, buffer); | |
878 | ||
879 | return bpf_internal_load_pointer_neg_helper(skb, k, size); | |
880 | } | |
881 | ||
ea02f941 MS |
882 | static inline int bpf_tell_extensions(void) |
883 | { | |
37692299 | 884 | return SKF_AD_MAX; |
ea02f941 MS |
885 | } |
886 | ||
1da177e4 | 887 | #endif /* __LINUX_FILTER_H__ */ |