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 | |
2d0bde57 LA |
22 | #ifdef CONFIG_ARCH_HAS_SET_MEMORY |
23 | #include <asm/set_memory.h> | |
24 | #endif | |
b954d834 DB |
25 | |
26 | #include <uapi/linux/filter.h> | |
daedfb22 | 27 | #include <uapi/linux/bpf.h> |
60a3b225 DB |
28 | |
29 | struct sk_buff; | |
30 | struct sock; | |
31 | struct seccomp_data; | |
09756af4 | 32 | struct bpf_prog_aux; |
792d4b5c | 33 | |
30743837 DB |
34 | /* ArgX, context and stack frame pointer register positions. Note, |
35 | * Arg1, Arg2, Arg3, etc are used as argument mappings of function | |
36 | * calls in BPF_CALL instruction. | |
37 | */ | |
38 | #define BPF_REG_ARG1 BPF_REG_1 | |
39 | #define BPF_REG_ARG2 BPF_REG_2 | |
40 | #define BPF_REG_ARG3 BPF_REG_3 | |
41 | #define BPF_REG_ARG4 BPF_REG_4 | |
42 | #define BPF_REG_ARG5 BPF_REG_5 | |
43 | #define BPF_REG_CTX BPF_REG_6 | |
44 | #define BPF_REG_FP BPF_REG_10 | |
45 | ||
46 | /* Additional register mappings for converted user programs. */ | |
47 | #define BPF_REG_A BPF_REG_0 | |
48 | #define BPF_REG_X BPF_REG_7 | |
49 | #define BPF_REG_TMP BPF_REG_8 | |
bd4cf0ed | 50 | |
4f3446bb DB |
51 | /* Kernel hidden auxiliary/helper register for hardening step. |
52 | * Only used by eBPF JITs. It's nothing more than a temporary | |
53 | * register that JITs use internally, only that here it's part | |
54 | * of eBPF instructions that have been rewritten for blinding | |
55 | * constants. See JIT pre-step in bpf_jit_blind_constants(). | |
56 | */ | |
57 | #define BPF_REG_AX MAX_BPF_REG | |
58 | #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1) | |
59 | ||
71189fa9 AS |
60 | /* unused opcode to mark special call to bpf_tail_call() helper */ |
61 | #define BPF_TAIL_CALL 0xf0 | |
62 | ||
74451e66 DB |
63 | /* As per nm, we expose JITed images as text (code) section for |
64 | * kallsyms. That way, tools like perf can find it to match | |
65 | * addresses. | |
66 | */ | |
67 | #define BPF_SYM_ELF_TYPE 't' | |
68 | ||
bd4cf0ed AS |
69 | /* BPF program can access up to 512 bytes of stack space. */ |
70 | #define MAX_BPF_STACK 512 | |
71 | ||
f8f6d679 DB |
72 | /* Helper macros for filter block array initializers. */ |
73 | ||
e430f34e | 74 | /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ |
f8f6d679 | 75 | |
e430f34e | 76 | #define BPF_ALU64_REG(OP, DST, SRC) \ |
2695fb55 | 77 | ((struct bpf_insn) { \ |
f8f6d679 | 78 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
79 | .dst_reg = DST, \ |
80 | .src_reg = SRC, \ | |
f8f6d679 DB |
81 | .off = 0, \ |
82 | .imm = 0 }) | |
83 | ||
e430f34e | 84 | #define BPF_ALU32_REG(OP, DST, SRC) \ |
2695fb55 | 85 | ((struct bpf_insn) { \ |
f8f6d679 | 86 | .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
87 | .dst_reg = DST, \ |
88 | .src_reg = SRC, \ | |
f8f6d679 DB |
89 | .off = 0, \ |
90 | .imm = 0 }) | |
91 | ||
e430f34e | 92 | /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ |
f8f6d679 | 93 | |
e430f34e | 94 | #define BPF_ALU64_IMM(OP, DST, IMM) \ |
2695fb55 | 95 | ((struct bpf_insn) { \ |
f8f6d679 | 96 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
97 | .dst_reg = DST, \ |
98 | .src_reg = 0, \ | |
f8f6d679 DB |
99 | .off = 0, \ |
100 | .imm = IMM }) | |
101 | ||
e430f34e | 102 | #define BPF_ALU32_IMM(OP, DST, IMM) \ |
2695fb55 | 103 | ((struct bpf_insn) { \ |
f8f6d679 | 104 | .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
105 | .dst_reg = DST, \ |
106 | .src_reg = 0, \ | |
f8f6d679 DB |
107 | .off = 0, \ |
108 | .imm = IMM }) | |
109 | ||
110 | /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ | |
111 | ||
e430f34e | 112 | #define BPF_ENDIAN(TYPE, DST, LEN) \ |
2695fb55 | 113 | ((struct bpf_insn) { \ |
f8f6d679 | 114 | .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ |
e430f34e AS |
115 | .dst_reg = DST, \ |
116 | .src_reg = 0, \ | |
f8f6d679 DB |
117 | .off = 0, \ |
118 | .imm = LEN }) | |
119 | ||
e430f34e | 120 | /* Short form of mov, dst_reg = src_reg */ |
f8f6d679 | 121 | |
e430f34e | 122 | #define BPF_MOV64_REG(DST, SRC) \ |
2695fb55 | 123 | ((struct bpf_insn) { \ |
f8f6d679 | 124 | .code = BPF_ALU64 | BPF_MOV | BPF_X, \ |
e430f34e AS |
125 | .dst_reg = DST, \ |
126 | .src_reg = SRC, \ | |
f8f6d679 DB |
127 | .off = 0, \ |
128 | .imm = 0 }) | |
129 | ||
e430f34e | 130 | #define BPF_MOV32_REG(DST, SRC) \ |
2695fb55 | 131 | ((struct bpf_insn) { \ |
f8f6d679 | 132 | .code = BPF_ALU | BPF_MOV | BPF_X, \ |
e430f34e AS |
133 | .dst_reg = DST, \ |
134 | .src_reg = SRC, \ | |
f8f6d679 DB |
135 | .off = 0, \ |
136 | .imm = 0 }) | |
137 | ||
e430f34e | 138 | /* Short form of mov, dst_reg = imm32 */ |
f8f6d679 | 139 | |
e430f34e | 140 | #define BPF_MOV64_IMM(DST, IMM) \ |
2695fb55 | 141 | ((struct bpf_insn) { \ |
f8f6d679 | 142 | .code = BPF_ALU64 | BPF_MOV | BPF_K, \ |
e430f34e AS |
143 | .dst_reg = DST, \ |
144 | .src_reg = 0, \ | |
f8f6d679 DB |
145 | .off = 0, \ |
146 | .imm = IMM }) | |
147 | ||
e430f34e | 148 | #define BPF_MOV32_IMM(DST, IMM) \ |
2695fb55 | 149 | ((struct bpf_insn) { \ |
f8f6d679 | 150 | .code = BPF_ALU | BPF_MOV | BPF_K, \ |
e430f34e AS |
151 | .dst_reg = DST, \ |
152 | .src_reg = 0, \ | |
f8f6d679 DB |
153 | .off = 0, \ |
154 | .imm = IMM }) | |
155 | ||
02ab695b AS |
156 | /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ |
157 | #define BPF_LD_IMM64(DST, IMM) \ | |
158 | BPF_LD_IMM64_RAW(DST, 0, IMM) | |
159 | ||
160 | #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ | |
161 | ((struct bpf_insn) { \ | |
162 | .code = BPF_LD | BPF_DW | BPF_IMM, \ | |
163 | .dst_reg = DST, \ | |
164 | .src_reg = SRC, \ | |
165 | .off = 0, \ | |
166 | .imm = (__u32) (IMM) }), \ | |
167 | ((struct bpf_insn) { \ | |
168 | .code = 0, /* zero is reserved opcode */ \ | |
169 | .dst_reg = 0, \ | |
170 | .src_reg = 0, \ | |
171 | .off = 0, \ | |
172 | .imm = ((__u64) (IMM)) >> 32 }) | |
173 | ||
0246e64d AS |
174 | /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ |
175 | #define BPF_LD_MAP_FD(DST, MAP_FD) \ | |
176 | BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) | |
177 | ||
e430f34e | 178 | /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ |
f8f6d679 | 179 | |
e430f34e | 180 | #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 181 | ((struct bpf_insn) { \ |
f8f6d679 | 182 | .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
183 | .dst_reg = DST, \ |
184 | .src_reg = SRC, \ | |
f8f6d679 DB |
185 | .off = 0, \ |
186 | .imm = IMM }) | |
187 | ||
e430f34e | 188 | #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 189 | ((struct bpf_insn) { \ |
f8f6d679 | 190 | .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
191 | .dst_reg = DST, \ |
192 | .src_reg = SRC, \ | |
f8f6d679 DB |
193 | .off = 0, \ |
194 | .imm = IMM }) | |
195 | ||
e430f34e | 196 | /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ |
f8f6d679 | 197 | |
e430f34e | 198 | #define BPF_LD_ABS(SIZE, IMM) \ |
2695fb55 | 199 | ((struct bpf_insn) { \ |
f8f6d679 | 200 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ |
e430f34e AS |
201 | .dst_reg = 0, \ |
202 | .src_reg = 0, \ | |
f8f6d679 | 203 | .off = 0, \ |
e430f34e | 204 | .imm = IMM }) |
f8f6d679 | 205 | |
e430f34e | 206 | /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ |
f8f6d679 | 207 | |
e430f34e | 208 | #define BPF_LD_IND(SIZE, SRC, IMM) \ |
2695fb55 | 209 | ((struct bpf_insn) { \ |
f8f6d679 | 210 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ |
e430f34e AS |
211 | .dst_reg = 0, \ |
212 | .src_reg = SRC, \ | |
f8f6d679 | 213 | .off = 0, \ |
e430f34e | 214 | .imm = IMM }) |
f8f6d679 | 215 | |
e430f34e | 216 | /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ |
f8f6d679 | 217 | |
e430f34e | 218 | #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ |
2695fb55 | 219 | ((struct bpf_insn) { \ |
f8f6d679 | 220 | .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
221 | .dst_reg = DST, \ |
222 | .src_reg = SRC, \ | |
f8f6d679 DB |
223 | .off = OFF, \ |
224 | .imm = 0 }) | |
225 | ||
e430f34e AS |
226 | /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ |
227 | ||
228 | #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ | |
2695fb55 | 229 | ((struct bpf_insn) { \ |
f8f6d679 | 230 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
231 | .dst_reg = DST, \ |
232 | .src_reg = SRC, \ | |
f8f6d679 DB |
233 | .off = OFF, \ |
234 | .imm = 0 }) | |
235 | ||
cffc642d MH |
236 | /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */ |
237 | ||
238 | #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \ | |
239 | ((struct bpf_insn) { \ | |
240 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \ | |
241 | .dst_reg = DST, \ | |
242 | .src_reg = SRC, \ | |
243 | .off = OFF, \ | |
244 | .imm = 0 }) | |
245 | ||
e430f34e AS |
246 | /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ |
247 | ||
248 | #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ | |
2695fb55 | 249 | ((struct bpf_insn) { \ |
e430f34e AS |
250 | .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ |
251 | .dst_reg = DST, \ | |
252 | .src_reg = 0, \ | |
253 | .off = OFF, \ | |
254 | .imm = IMM }) | |
255 | ||
256 | /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ | |
f8f6d679 | 257 | |
e430f34e | 258 | #define BPF_JMP_REG(OP, DST, SRC, OFF) \ |
2695fb55 | 259 | ((struct bpf_insn) { \ |
f8f6d679 | 260 | .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
261 | .dst_reg = DST, \ |
262 | .src_reg = SRC, \ | |
f8f6d679 DB |
263 | .off = OFF, \ |
264 | .imm = 0 }) | |
265 | ||
e430f34e | 266 | /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ |
f8f6d679 | 267 | |
e430f34e | 268 | #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ |
2695fb55 | 269 | ((struct bpf_insn) { \ |
f8f6d679 | 270 | .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
271 | .dst_reg = DST, \ |
272 | .src_reg = 0, \ | |
f8f6d679 DB |
273 | .off = OFF, \ |
274 | .imm = IMM }) | |
275 | ||
614d0d77 DB |
276 | /* Unconditional jumps, goto pc + off16 */ |
277 | ||
278 | #define BPF_JMP_A(OFF) \ | |
279 | ((struct bpf_insn) { \ | |
280 | .code = BPF_JMP | BPF_JA, \ | |
281 | .dst_reg = 0, \ | |
282 | .src_reg = 0, \ | |
283 | .off = OFF, \ | |
284 | .imm = 0 }) | |
285 | ||
f8f6d679 DB |
286 | /* Function call */ |
287 | ||
288 | #define BPF_EMIT_CALL(FUNC) \ | |
2695fb55 | 289 | ((struct bpf_insn) { \ |
f8f6d679 | 290 | .code = BPF_JMP | BPF_CALL, \ |
e430f34e AS |
291 | .dst_reg = 0, \ |
292 | .src_reg = 0, \ | |
f8f6d679 DB |
293 | .off = 0, \ |
294 | .imm = ((FUNC) - __bpf_call_base) }) | |
295 | ||
296 | /* Raw code statement block */ | |
297 | ||
e430f34e | 298 | #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ |
2695fb55 | 299 | ((struct bpf_insn) { \ |
f8f6d679 | 300 | .code = CODE, \ |
e430f34e AS |
301 | .dst_reg = DST, \ |
302 | .src_reg = SRC, \ | |
f8f6d679 DB |
303 | .off = OFF, \ |
304 | .imm = IMM }) | |
305 | ||
306 | /* Program exit */ | |
307 | ||
308 | #define BPF_EXIT_INSN() \ | |
2695fb55 | 309 | ((struct bpf_insn) { \ |
f8f6d679 | 310 | .code = BPF_JMP | BPF_EXIT, \ |
e430f34e AS |
311 | .dst_reg = 0, \ |
312 | .src_reg = 0, \ | |
f8f6d679 DB |
313 | .off = 0, \ |
314 | .imm = 0 }) | |
315 | ||
a4afd37b DB |
316 | /* Internal classic blocks for direct assignment */ |
317 | ||
318 | #define __BPF_STMT(CODE, K) \ | |
319 | ((struct sock_filter) BPF_STMT(CODE, K)) | |
320 | ||
321 | #define __BPF_JUMP(CODE, K, JT, JF) \ | |
322 | ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) | |
323 | ||
f8f6d679 DB |
324 | #define bytes_to_bpf_size(bytes) \ |
325 | ({ \ | |
326 | int bpf_size = -EINVAL; \ | |
327 | \ | |
328 | if (bytes == sizeof(u8)) \ | |
329 | bpf_size = BPF_B; \ | |
330 | else if (bytes == sizeof(u16)) \ | |
331 | bpf_size = BPF_H; \ | |
332 | else if (bytes == sizeof(u32)) \ | |
333 | bpf_size = BPF_W; \ | |
334 | else if (bytes == sizeof(u64)) \ | |
335 | bpf_size = BPF_DW; \ | |
336 | \ | |
337 | bpf_size; \ | |
338 | }) | |
9739eef1 | 339 | |
f035a515 DB |
340 | #define BPF_SIZEOF(type) \ |
341 | ({ \ | |
342 | const int __size = bytes_to_bpf_size(sizeof(type)); \ | |
343 | BUILD_BUG_ON(__size < 0); \ | |
344 | __size; \ | |
345 | }) | |
346 | ||
347 | #define BPF_FIELD_SIZEOF(type, field) \ | |
348 | ({ \ | |
349 | const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ | |
350 | BUILD_BUG_ON(__size < 0); \ | |
351 | __size; \ | |
352 | }) | |
353 | ||
f3694e00 DB |
354 | #define __BPF_MAP_0(m, v, ...) v |
355 | #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) | |
356 | #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) | |
357 | #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) | |
358 | #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) | |
359 | #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) | |
360 | ||
361 | #define __BPF_REG_0(...) __BPF_PAD(5) | |
362 | #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) | |
363 | #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) | |
364 | #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) | |
365 | #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) | |
366 | #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) | |
367 | ||
368 | #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) | |
369 | #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) | |
370 | ||
371 | #define __BPF_CAST(t, a) \ | |
372 | (__force t) \ | |
373 | (__force \ | |
374 | typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ | |
375 | (unsigned long)0, (t)0))) a | |
376 | #define __BPF_V void | |
377 | #define __BPF_N | |
378 | ||
379 | #define __BPF_DECL_ARGS(t, a) t a | |
380 | #define __BPF_DECL_REGS(t, a) u64 a | |
381 | ||
382 | #define __BPF_PAD(n) \ | |
383 | __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ | |
384 | u64, __ur_3, u64, __ur_4, u64, __ur_5) | |
385 | ||
386 | #define BPF_CALL_x(x, name, ...) \ | |
387 | static __always_inline \ | |
388 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ | |
389 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ | |
390 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ | |
391 | { \ | |
392 | return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ | |
393 | } \ | |
394 | static __always_inline \ | |
395 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) | |
396 | ||
397 | #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) | |
398 | #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) | |
399 | #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) | |
400 | #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) | |
401 | #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) | |
402 | #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) | |
403 | ||
bd4cf0ed AS |
404 | #ifdef CONFIG_COMPAT |
405 | /* A struct sock_filter is architecture independent. */ | |
0c5fe1b4 WD |
406 | struct compat_sock_fprog { |
407 | u16 len; | |
bd4cf0ed | 408 | compat_uptr_t filter; /* struct sock_filter * */ |
0c5fe1b4 WD |
409 | }; |
410 | #endif | |
411 | ||
a3ea269b DB |
412 | struct sock_fprog_kern { |
413 | u16 len; | |
414 | struct sock_filter *filter; | |
415 | }; | |
416 | ||
738cbe72 DB |
417 | struct bpf_binary_header { |
418 | unsigned int pages; | |
419 | u8 image[]; | |
420 | }; | |
421 | ||
7ae457c1 | 422 | struct bpf_prog { |
286aad3c | 423 | u16 pages; /* Number of allocated pages */ |
a91263d5 DB |
424 | kmemcheck_bitfield_begin(meta); |
425 | u16 jited:1, /* Is our filter JIT'ed? */ | |
65869a47 | 426 | locked:1, /* Program image locked? */ |
c46646d0 | 427 | gpl_compatible:1, /* Is filter GPL compatible? */ |
ff936a04 | 428 | cb_access:1, /* Is control block accessed? */ |
e390b55d | 429 | dst_needed:1; /* Do we need dst entry? */ |
a91263d5 | 430 | kmemcheck_bitfield_end(meta); |
24701ece | 431 | enum bpf_prog_type type; /* Type of BPF program */ |
7bd509e3 | 432 | u32 len; /* Number of filter blocks */ |
783d28dd | 433 | u32 jited_len; /* Size of jited insns in bytes */ |
f1f7714e | 434 | u8 tag[BPF_TAG_SIZE]; |
09756af4 | 435 | struct bpf_prog_aux *aux; /* Auxiliary fields */ |
24701ece | 436 | struct sock_fprog_kern *orig_prog; /* Original BPF program */ |
88575199 DB |
437 | unsigned int (*bpf_func)(const void *ctx, |
438 | const struct bpf_insn *insn); | |
60a3b225 | 439 | /* Instructions for interpreter */ |
d45ed4a4 | 440 | union { |
bd4cf0ed | 441 | struct sock_filter insns[0]; |
2695fb55 | 442 | struct bpf_insn insnsi[0]; |
d45ed4a4 | 443 | }; |
b715631f SH |
444 | }; |
445 | ||
7ae457c1 | 446 | struct sk_filter { |
4c355cdf | 447 | refcount_t refcnt; |
7ae457c1 AS |
448 | struct rcu_head rcu; |
449 | struct bpf_prog *prog; | |
450 | }; | |
451 | ||
452 | #define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi) | |
453 | ||
01dd194c DB |
454 | #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN |
455 | ||
db58ba45 AS |
456 | struct bpf_skb_data_end { |
457 | struct qdisc_skb_cb qdisc_cb; | |
458 | void *data_end; | |
459 | }; | |
460 | ||
6a773a15 BB |
461 | struct xdp_buff { |
462 | void *data; | |
463 | void *data_end; | |
17bedab2 | 464 | void *data_hard_start; |
6a773a15 BB |
465 | }; |
466 | ||
db58ba45 | 467 | /* compute the linear packet data range [data, data_end) which |
3a0af8fd | 468 | * will be accessed by cls_bpf, act_bpf and lwt programs |
db58ba45 AS |
469 | */ |
470 | static inline void bpf_compute_data_end(struct sk_buff *skb) | |
471 | { | |
472 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; | |
473 | ||
474 | BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb)); | |
475 | cb->data_end = skb->data + skb_headlen(skb); | |
476 | } | |
477 | ||
01dd194c DB |
478 | static inline u8 *bpf_skb_cb(struct sk_buff *skb) |
479 | { | |
480 | /* eBPF programs may read/write skb->cb[] area to transfer meta | |
481 | * data between tail calls. Since this also needs to work with | |
482 | * tc, that scratch memory is mapped to qdisc_skb_cb's data area. | |
483 | * | |
484 | * In some socket filter cases, the cb unfortunately needs to be | |
485 | * saved/restored so that protocol specific skb->cb[] data won't | |
486 | * be lost. In any case, due to unpriviledged eBPF programs | |
487 | * attached to sockets, we need to clear the bpf_skb_cb() area | |
488 | * to not leak previous contents to user space. | |
489 | */ | |
490 | BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN); | |
491 | BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != | |
492 | FIELD_SIZEOF(struct qdisc_skb_cb, data)); | |
493 | ||
494 | return qdisc_skb_cb(skb)->data; | |
495 | } | |
496 | ||
ff936a04 AS |
497 | static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, |
498 | struct sk_buff *skb) | |
499 | { | |
01dd194c DB |
500 | u8 *cb_data = bpf_skb_cb(skb); |
501 | u8 cb_saved[BPF_SKB_CB_LEN]; | |
ff936a04 AS |
502 | u32 res; |
503 | ||
ff936a04 | 504 | if (unlikely(prog->cb_access)) { |
01dd194c DB |
505 | memcpy(cb_saved, cb_data, sizeof(cb_saved)); |
506 | memset(cb_data, 0, sizeof(cb_saved)); | |
ff936a04 AS |
507 | } |
508 | ||
509 | res = BPF_PROG_RUN(prog, skb); | |
510 | ||
511 | if (unlikely(prog->cb_access)) | |
01dd194c | 512 | memcpy(cb_data, cb_saved, sizeof(cb_saved)); |
ff936a04 AS |
513 | |
514 | return res; | |
515 | } | |
516 | ||
517 | static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, | |
518 | struct sk_buff *skb) | |
519 | { | |
01dd194c | 520 | u8 *cb_data = bpf_skb_cb(skb); |
ff936a04 AS |
521 | |
522 | if (unlikely(prog->cb_access)) | |
01dd194c DB |
523 | memset(cb_data, 0, BPF_SKB_CB_LEN); |
524 | ||
ff936a04 AS |
525 | return BPF_PROG_RUN(prog, skb); |
526 | } | |
527 | ||
366cbf2f DB |
528 | static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, |
529 | struct xdp_buff *xdp) | |
6a773a15 | 530 | { |
366cbf2f DB |
531 | /* Caller needs to hold rcu_read_lock() (!), otherwise program |
532 | * can be released while still running, or map elements could be | |
533 | * freed early while still having concurrent users. XDP fastpath | |
534 | * already takes rcu_read_lock() when fetching the program, so | |
535 | * it's not necessary here anymore. | |
536 | */ | |
537 | return BPF_PROG_RUN(prog, xdp); | |
6a773a15 BB |
538 | } |
539 | ||
aafe6ae9 DB |
540 | static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) |
541 | { | |
542 | return prog->len * sizeof(struct bpf_insn); | |
543 | } | |
544 | ||
f1f7714e | 545 | static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) |
aafe6ae9 DB |
546 | { |
547 | return round_up(bpf_prog_insn_size(prog) + | |
548 | sizeof(__be64) + 1, SHA_MESSAGE_BYTES); | |
549 | } | |
550 | ||
7ae457c1 | 551 | static inline unsigned int bpf_prog_size(unsigned int proglen) |
b715631f | 552 | { |
7ae457c1 AS |
553 | return max(sizeof(struct bpf_prog), |
554 | offsetof(struct bpf_prog, insns[proglen])); | |
b715631f SH |
555 | } |
556 | ||
7b36f929 DB |
557 | static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) |
558 | { | |
559 | /* When classic BPF programs have been loaded and the arch | |
560 | * does not have a classic BPF JIT (anymore), they have been | |
561 | * converted via bpf_migrate_filter() to eBPF and thus always | |
562 | * have an unspec program type. | |
563 | */ | |
564 | return prog->type == BPF_PROG_TYPE_UNSPEC; | |
565 | } | |
566 | ||
009937e7 | 567 | #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) |
a3ea269b | 568 | |
9d876e79 | 569 | #ifdef CONFIG_ARCH_HAS_SET_MEMORY |
60a3b225 DB |
570 | static inline void bpf_prog_lock_ro(struct bpf_prog *fp) |
571 | { | |
65869a47 DB |
572 | fp->locked = 1; |
573 | WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages)); | |
60a3b225 DB |
574 | } |
575 | ||
576 | static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) | |
577 | { | |
65869a47 DB |
578 | if (fp->locked) { |
579 | WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages)); | |
580 | /* In case set_memory_rw() fails, we want to be the first | |
581 | * to crash here instead of some random place later on. | |
582 | */ | |
583 | fp->locked = 0; | |
584 | } | |
60a3b225 | 585 | } |
74451e66 | 586 | |
9d876e79 DB |
587 | static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
588 | { | |
65869a47 | 589 | WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages)); |
9d876e79 DB |
590 | } |
591 | ||
74451e66 DB |
592 | static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) |
593 | { | |
65869a47 | 594 | WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages)); |
74451e66 | 595 | } |
60a3b225 DB |
596 | #else |
597 | static inline void bpf_prog_lock_ro(struct bpf_prog *fp) | |
598 | { | |
599 | } | |
600 | ||
601 | static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) | |
602 | { | |
603 | } | |
74451e66 | 604 | |
9d876e79 DB |
605 | static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
606 | { | |
607 | } | |
608 | ||
74451e66 DB |
609 | static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) |
610 | { | |
611 | } | |
9d876e79 | 612 | #endif /* CONFIG_ARCH_HAS_SET_MEMORY */ |
60a3b225 | 613 | |
74451e66 DB |
614 | static inline struct bpf_binary_header * |
615 | bpf_jit_binary_hdr(const struct bpf_prog *fp) | |
616 | { | |
617 | unsigned long real_start = (unsigned long)fp->bpf_func; | |
618 | unsigned long addr = real_start & PAGE_MASK; | |
619 | ||
620 | return (void *)addr; | |
621 | } | |
622 | ||
f4979fce WB |
623 | int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); |
624 | static inline int sk_filter(struct sock *sk, struct sk_buff *skb) | |
625 | { | |
626 | return sk_filter_trim_cap(sk, skb, 1); | |
627 | } | |
bd4cf0ed | 628 | |
d1c55ab5 | 629 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); |
7ae457c1 | 630 | void bpf_prog_free(struct bpf_prog *fp); |
bd4cf0ed | 631 | |
60a3b225 DB |
632 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); |
633 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, | |
634 | gfp_t gfp_extra_flags); | |
635 | void __bpf_prog_free(struct bpf_prog *fp); | |
636 | ||
637 | static inline void bpf_prog_unlock_free(struct bpf_prog *fp) | |
638 | { | |
639 | bpf_prog_unlock_ro(fp); | |
640 | __bpf_prog_free(fp); | |
641 | } | |
642 | ||
ac67eb2c DB |
643 | typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, |
644 | unsigned int flen); | |
645 | ||
7ae457c1 | 646 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); |
ac67eb2c | 647 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, |
bab18991 | 648 | bpf_aux_classic_check_t trans, bool save_orig); |
7ae457c1 | 649 | void bpf_prog_destroy(struct bpf_prog *fp); |
a3ea269b | 650 | |
fbc907f0 | 651 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
89aa0758 | 652 | int sk_attach_bpf(u32 ufd, struct sock *sk); |
538950a1 CG |
653 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
654 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); | |
fbc907f0 | 655 | int sk_detach_filter(struct sock *sk); |
fbc907f0 DB |
656 | int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, |
657 | unsigned int len); | |
fbc907f0 | 658 | |
278571ba | 659 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); |
fbc907f0 | 660 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); |
0a14842f | 661 | |
62258278 | 662 | u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
d1c55ab5 DB |
663 | |
664 | struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); | |
9383191d | 665 | void bpf_jit_compile(struct bpf_prog *prog); |
17bedab2 | 666 | bool bpf_helper_changes_pkt_data(void *func); |
62258278 | 667 | |
c237ee5e DB |
668 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
669 | const struct bpf_insn *patch, u32 len); | |
6a773a15 | 670 | void bpf_warn_invalid_xdp_action(u32 act); |
c237ee5e | 671 | |
b954d834 | 672 | #ifdef CONFIG_BPF_JIT |
c94987e4 | 673 | extern int bpf_jit_enable; |
4f3446bb | 674 | extern int bpf_jit_harden; |
74451e66 | 675 | extern int bpf_jit_kallsyms; |
c94987e4 | 676 | |
b954d834 DB |
677 | typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); |
678 | ||
679 | struct bpf_binary_header * | |
680 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
681 | unsigned int alignment, | |
682 | bpf_jit_fill_hole_t bpf_fill_ill_insns); | |
683 | void bpf_jit_binary_free(struct bpf_binary_header *hdr); | |
684 | ||
b954d834 DB |
685 | void bpf_jit_free(struct bpf_prog *fp); |
686 | ||
4f3446bb DB |
687 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); |
688 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); | |
689 | ||
b954d834 DB |
690 | static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, |
691 | u32 pass, void *image) | |
692 | { | |
b13138ef DB |
693 | pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, |
694 | proglen, pass, image, current->comm, task_pid_nr(current)); | |
695 | ||
b954d834 DB |
696 | if (image) |
697 | print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, | |
698 | 16, 1, image, proglen, false); | |
699 | } | |
4f3446bb DB |
700 | |
701 | static inline bool bpf_jit_is_ebpf(void) | |
702 | { | |
703 | # ifdef CONFIG_HAVE_EBPF_JIT | |
704 | return true; | |
705 | # else | |
706 | return false; | |
707 | # endif | |
708 | } | |
709 | ||
81ed18ab AS |
710 | static inline bool ebpf_jit_enabled(void) |
711 | { | |
712 | return bpf_jit_enable && bpf_jit_is_ebpf(); | |
713 | } | |
714 | ||
74451e66 DB |
715 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
716 | { | |
717 | return fp->jited && bpf_jit_is_ebpf(); | |
718 | } | |
719 | ||
4f3446bb DB |
720 | static inline bool bpf_jit_blinding_enabled(void) |
721 | { | |
722 | /* These are the prerequisites, should someone ever have the | |
723 | * idea to call blinding outside of them, we make sure to | |
724 | * bail out. | |
725 | */ | |
726 | if (!bpf_jit_is_ebpf()) | |
727 | return false; | |
728 | if (!bpf_jit_enable) | |
729 | return false; | |
730 | if (!bpf_jit_harden) | |
731 | return false; | |
732 | if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) | |
733 | return false; | |
734 | ||
735 | return true; | |
736 | } | |
74451e66 DB |
737 | |
738 | static inline bool bpf_jit_kallsyms_enabled(void) | |
739 | { | |
740 | /* There are a couple of corner cases where kallsyms should | |
741 | * not be enabled f.e. on hardening. | |
742 | */ | |
743 | if (bpf_jit_harden) | |
744 | return false; | |
745 | if (!bpf_jit_kallsyms) | |
746 | return false; | |
747 | if (bpf_jit_kallsyms == 1) | |
748 | return true; | |
749 | ||
750 | return false; | |
751 | } | |
752 | ||
753 | const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, | |
754 | unsigned long *off, char *sym); | |
755 | bool is_bpf_text_address(unsigned long addr); | |
756 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
757 | char *sym); | |
758 | ||
759 | static inline const char * | |
760 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
761 | unsigned long *off, char **modname, char *sym) | |
762 | { | |
763 | const char *ret = __bpf_address_lookup(addr, size, off, sym); | |
764 | ||
765 | if (ret && modname) | |
766 | *modname = NULL; | |
767 | return ret; | |
768 | } | |
769 | ||
770 | void bpf_prog_kallsyms_add(struct bpf_prog *fp); | |
771 | void bpf_prog_kallsyms_del(struct bpf_prog *fp); | |
772 | ||
773 | #else /* CONFIG_BPF_JIT */ | |
774 | ||
81ed18ab AS |
775 | static inline bool ebpf_jit_enabled(void) |
776 | { | |
777 | return false; | |
778 | } | |
779 | ||
74451e66 DB |
780 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
781 | { | |
782 | return false; | |
783 | } | |
784 | ||
b954d834 DB |
785 | static inline void bpf_jit_free(struct bpf_prog *fp) |
786 | { | |
787 | bpf_prog_unlock_free(fp); | |
788 | } | |
74451e66 DB |
789 | |
790 | static inline bool bpf_jit_kallsyms_enabled(void) | |
791 | { | |
792 | return false; | |
793 | } | |
794 | ||
795 | static inline const char * | |
796 | __bpf_address_lookup(unsigned long addr, unsigned long *size, | |
797 | unsigned long *off, char *sym) | |
798 | { | |
799 | return NULL; | |
800 | } | |
801 | ||
802 | static inline bool is_bpf_text_address(unsigned long addr) | |
803 | { | |
804 | return false; | |
805 | } | |
806 | ||
807 | static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, | |
808 | char *type, char *sym) | |
809 | { | |
810 | return -ERANGE; | |
811 | } | |
812 | ||
813 | static inline const char * | |
814 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
815 | unsigned long *off, char **modname, char *sym) | |
816 | { | |
817 | return NULL; | |
818 | } | |
819 | ||
820 | static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) | |
821 | { | |
822 | } | |
823 | ||
824 | static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) | |
825 | { | |
826 | } | |
b954d834 DB |
827 | #endif /* CONFIG_BPF_JIT */ |
828 | ||
34805931 DB |
829 | #define BPF_ANC BIT(15) |
830 | ||
55795ef5 RV |
831 | static inline bool bpf_needs_clear_a(const struct sock_filter *first) |
832 | { | |
833 | switch (first->code) { | |
834 | case BPF_RET | BPF_K: | |
835 | case BPF_LD | BPF_W | BPF_LEN: | |
836 | return false; | |
837 | ||
838 | case BPF_LD | BPF_W | BPF_ABS: | |
839 | case BPF_LD | BPF_H | BPF_ABS: | |
840 | case BPF_LD | BPF_B | BPF_ABS: | |
841 | if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) | |
842 | return true; | |
843 | return false; | |
844 | ||
845 | default: | |
846 | return true; | |
847 | } | |
848 | } | |
849 | ||
34805931 DB |
850 | static inline u16 bpf_anc_helper(const struct sock_filter *ftest) |
851 | { | |
852 | BUG_ON(ftest->code & BPF_ANC); | |
853 | ||
854 | switch (ftest->code) { | |
855 | case BPF_LD | BPF_W | BPF_ABS: | |
856 | case BPF_LD | BPF_H | BPF_ABS: | |
857 | case BPF_LD | BPF_B | BPF_ABS: | |
858 | #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ | |
859 | return BPF_ANC | SKF_AD_##CODE | |
860 | switch (ftest->k) { | |
861 | BPF_ANCILLARY(PROTOCOL); | |
862 | BPF_ANCILLARY(PKTTYPE); | |
863 | BPF_ANCILLARY(IFINDEX); | |
864 | BPF_ANCILLARY(NLATTR); | |
865 | BPF_ANCILLARY(NLATTR_NEST); | |
866 | BPF_ANCILLARY(MARK); | |
867 | BPF_ANCILLARY(QUEUE); | |
868 | BPF_ANCILLARY(HATYPE); | |
869 | BPF_ANCILLARY(RXHASH); | |
870 | BPF_ANCILLARY(CPU); | |
871 | BPF_ANCILLARY(ALU_XOR_X); | |
872 | BPF_ANCILLARY(VLAN_TAG); | |
873 | BPF_ANCILLARY(VLAN_TAG_PRESENT); | |
874 | BPF_ANCILLARY(PAY_OFFSET); | |
875 | BPF_ANCILLARY(RANDOM); | |
27cd5452 | 876 | BPF_ANCILLARY(VLAN_TPID); |
34805931 DB |
877 | } |
878 | /* Fallthrough. */ | |
879 | default: | |
880 | return ftest->code; | |
881 | } | |
882 | } | |
883 | ||
9f12fbe6 ZSL |
884 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, |
885 | int k, unsigned int size); | |
886 | ||
887 | static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, | |
888 | unsigned int size, void *buffer) | |
889 | { | |
890 | if (k >= 0) | |
891 | return skb_header_pointer(skb, k, size, buffer); | |
892 | ||
893 | return bpf_internal_load_pointer_neg_helper(skb, k, size); | |
894 | } | |
895 | ||
ea02f941 MS |
896 | static inline int bpf_tell_extensions(void) |
897 | { | |
37692299 | 898 | return SKF_AD_MAX; |
ea02f941 MS |
899 | } |
900 | ||
40304b2a LB |
901 | struct bpf_sock_ops_kern { |
902 | struct sock *sk; | |
903 | u32 op; | |
904 | union { | |
905 | u32 reply; | |
906 | u32 replylong[4]; | |
907 | }; | |
908 | }; | |
909 | ||
1da177e4 | 910 | #endif /* __LINUX_FILTER_H__ */ |