1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Linux Socket Filter Data Structures
4 */
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
7
8 #include <linux/atomic.h>
9 #include <linux/refcount.h>
10 #include <linux/compat.h>
11 #include <linux/skbuff.h>
12 #include <linux/linkage.h>
13 #include <linux/printk.h>
14 #include <linux/workqueue.h>
15 #include <linux/sched.h>
16 #include <linux/capability.h>
17 #include <linux/set_memory.h>
18 #include <linux/kallsyms.h>
19 #include <linux/if_vlan.h>
20 #include <linux/vmalloc.h>
21 #include <linux/sockptr.h>
22 #include <crypto/sha1.h>
23 #include <linux/u64_stats_sync.h>
24
25 #include <net/sch_generic.h>
26
27 #include <asm/byteorder.h>
28 #include <uapi/linux/filter.h>
29 #include <uapi/linux/bpf.h>
30
31 struct sk_buff;
32 struct sock;
33 struct seccomp_data;
34 struct bpf_prog_aux;
35 struct xdp_rxq_info;
36 struct xdp_buff;
37 struct sock_reuseport;
38 struct ctl_table;
39 struct ctl_table_header;
40
41 /* ArgX, context and stack frame pointer register positions. Note,
42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
43 * calls in BPF_CALL instruction.
44 */
45 #define BPF_REG_ARG1 BPF_REG_1
46 #define BPF_REG_ARG2 BPF_REG_2
47 #define BPF_REG_ARG3 BPF_REG_3
48 #define BPF_REG_ARG4 BPF_REG_4
49 #define BPF_REG_ARG5 BPF_REG_5
50 #define BPF_REG_CTX BPF_REG_6
51 #define BPF_REG_FP BPF_REG_10
52
53 /* Additional register mappings for converted user programs. */
54 #define BPF_REG_A BPF_REG_0
55 #define BPF_REG_X BPF_REG_7
56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
59
60 /* Kernel hidden auxiliary/helper register. */
61 #define BPF_REG_AX MAX_BPF_REG
62 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
63 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
64
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL 0xf0
67
68 /* unused opcode to mark special load instruction. Same as BPF_ABS */
69 #define BPF_PROBE_MEM 0x20
70
71 /* unused opcode to mark call to interpreter with arguments */
72 #define BPF_CALL_ARGS 0xe0
73
74 /* unused opcode to mark speculation barrier for mitigating
75 * Speculative Store Bypass
76 */
77 #define BPF_NOSPEC 0xc0
78
79 /* As per nm, we expose JITed images as text (code) section for
80 * kallsyms. That way, tools like perf can find it to match
81 * addresses.
82 */
83 #define BPF_SYM_ELF_TYPE 't'
84
85 /* BPF program can access up to 512 bytes of stack space. */
86 #define MAX_BPF_STACK 512
87
88 /* Helper macros for filter block array initializers. */
89
90 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
91
92 #define BPF_ALU64_REG(OP, DST, SRC) \
93 ((struct bpf_insn) { \
94 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
95 .dst_reg = DST, \
96 .src_reg = SRC, \
97 .off = 0, \
98 .imm = 0 })
99
100 #define BPF_ALU32_REG(OP, DST, SRC) \
101 ((struct bpf_insn) { \
102 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
103 .dst_reg = DST, \
104 .src_reg = SRC, \
105 .off = 0, \
106 .imm = 0 })
107
108 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
109
110 #define BPF_ALU64_IMM(OP, DST, IMM) \
111 ((struct bpf_insn) { \
112 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
113 .dst_reg = DST, \
114 .src_reg = 0, \
115 .off = 0, \
116 .imm = IMM })
117
118 #define BPF_ALU32_IMM(OP, DST, IMM) \
119 ((struct bpf_insn) { \
120 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
121 .dst_reg = DST, \
122 .src_reg = 0, \
123 .off = 0, \
124 .imm = IMM })
125
126 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
127
128 #define BPF_ENDIAN(TYPE, DST, LEN) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
131 .dst_reg = DST, \
132 .src_reg = 0, \
133 .off = 0, \
134 .imm = LEN })
135
136 /* Short form of mov, dst_reg = src_reg */
137
138 #define BPF_MOV64_REG(DST, SRC) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
141 .dst_reg = DST, \
142 .src_reg = SRC, \
143 .off = 0, \
144 .imm = 0 })
145
146 #define BPF_MOV32_REG(DST, SRC) \
147 ((struct bpf_insn) { \
148 .code = BPF_ALU | BPF_MOV | BPF_X, \
149 .dst_reg = DST, \
150 .src_reg = SRC, \
151 .off = 0, \
152 .imm = 0 })
153
154 /* Short form of mov, dst_reg = imm32 */
155
156 #define BPF_MOV64_IMM(DST, IMM) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
159 .dst_reg = DST, \
160 .src_reg = 0, \
161 .off = 0, \
162 .imm = IMM })
163
164 #define BPF_MOV32_IMM(DST, IMM) \
165 ((struct bpf_insn) { \
166 .code = BPF_ALU | BPF_MOV | BPF_K, \
167 .dst_reg = DST, \
168 .src_reg = 0, \
169 .off = 0, \
170 .imm = IMM })
171
172 /* Special form of mov32, used for doing explicit zero extension on dst. */
173 #define BPF_ZEXT_REG(DST) \
174 ((struct bpf_insn) { \
175 .code = BPF_ALU | BPF_MOV | BPF_X, \
176 .dst_reg = DST, \
177 .src_reg = DST, \
178 .off = 0, \
179 .imm = 1 })
180
insn_is_zext(const struct bpf_insn * insn)181 static inline bool insn_is_zext(const struct bpf_insn *insn)
182 {
183 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
184 }
185
186 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
187 #define BPF_LD_IMM64(DST, IMM) \
188 BPF_LD_IMM64_RAW(DST, 0, IMM)
189
190 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
191 ((struct bpf_insn) { \
192 .code = BPF_LD | BPF_DW | BPF_IMM, \
193 .dst_reg = DST, \
194 .src_reg = SRC, \
195 .off = 0, \
196 .imm = (__u32) (IMM) }), \
197 ((struct bpf_insn) { \
198 .code = 0, /* zero is reserved opcode */ \
199 .dst_reg = 0, \
200 .src_reg = 0, \
201 .off = 0, \
202 .imm = ((__u64) (IMM)) >> 32 })
203
204 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
205 #define BPF_LD_MAP_FD(DST, MAP_FD) \
206 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
207
208 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
209
210 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
211 ((struct bpf_insn) { \
212 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
213 .dst_reg = DST, \
214 .src_reg = SRC, \
215 .off = 0, \
216 .imm = IMM })
217
218 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
219 ((struct bpf_insn) { \
220 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
221 .dst_reg = DST, \
222 .src_reg = SRC, \
223 .off = 0, \
224 .imm = IMM })
225
226 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
227
228 #define BPF_LD_ABS(SIZE, IMM) \
229 ((struct bpf_insn) { \
230 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
231 .dst_reg = 0, \
232 .src_reg = 0, \
233 .off = 0, \
234 .imm = IMM })
235
236 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
237
238 #define BPF_LD_IND(SIZE, SRC, IMM) \
239 ((struct bpf_insn) { \
240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
241 .dst_reg = 0, \
242 .src_reg = SRC, \
243 .off = 0, \
244 .imm = IMM })
245
246 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
247
248 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
249 ((struct bpf_insn) { \
250 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
251 .dst_reg = DST, \
252 .src_reg = SRC, \
253 .off = OFF, \
254 .imm = 0 })
255
256 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
257
258 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
259 ((struct bpf_insn) { \
260 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
261 .dst_reg = DST, \
262 .src_reg = SRC, \
263 .off = OFF, \
264 .imm = 0 })
265
266
267 /*
268 * Atomic operations:
269 *
270 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
271 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
272 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
273 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
274 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
275 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
276 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
277 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
278 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
279 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
280 */
281
282 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
283 ((struct bpf_insn) { \
284 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
285 .dst_reg = DST, \
286 .src_reg = SRC, \
287 .off = OFF, \
288 .imm = OP })
289
290 /* Legacy alias */
291 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
292
293 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
294
295 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
296 ((struct bpf_insn) { \
297 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
298 .dst_reg = DST, \
299 .src_reg = 0, \
300 .off = OFF, \
301 .imm = IMM })
302
303 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
304
305 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
306 ((struct bpf_insn) { \
307 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
308 .dst_reg = DST, \
309 .src_reg = SRC, \
310 .off = OFF, \
311 .imm = 0 })
312
313 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
314
315 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
316 ((struct bpf_insn) { \
317 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
318 .dst_reg = DST, \
319 .src_reg = 0, \
320 .off = OFF, \
321 .imm = IMM })
322
323 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
324
325 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
328 .dst_reg = DST, \
329 .src_reg = SRC, \
330 .off = OFF, \
331 .imm = 0 })
332
333 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
334
335 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
338 .dst_reg = DST, \
339 .src_reg = 0, \
340 .off = OFF, \
341 .imm = IMM })
342
343 /* Unconditional jumps, goto pc + off16 */
344
345 #define BPF_JMP_A(OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_JMP | BPF_JA, \
348 .dst_reg = 0, \
349 .src_reg = 0, \
350 .off = OFF, \
351 .imm = 0 })
352
353 /* Relative call */
354
355 #define BPF_CALL_REL(TGT) \
356 ((struct bpf_insn) { \
357 .code = BPF_JMP | BPF_CALL, \
358 .dst_reg = 0, \
359 .src_reg = BPF_PSEUDO_CALL, \
360 .off = 0, \
361 .imm = TGT })
362
363 /* Function call */
364
365 #define BPF_CAST_CALL(x) \
366 ((u64 (*)(u64, u64, u64, u64, u64))(x))
367
368 #define BPF_EMIT_CALL(FUNC) \
369 ((struct bpf_insn) { \
370 .code = BPF_JMP | BPF_CALL, \
371 .dst_reg = 0, \
372 .src_reg = 0, \
373 .off = 0, \
374 .imm = ((FUNC) - __bpf_call_base) })
375
376 /* Raw code statement block */
377
378 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
379 ((struct bpf_insn) { \
380 .code = CODE, \
381 .dst_reg = DST, \
382 .src_reg = SRC, \
383 .off = OFF, \
384 .imm = IMM })
385
386 /* Program exit */
387
388 #define BPF_EXIT_INSN() \
389 ((struct bpf_insn) { \
390 .code = BPF_JMP | BPF_EXIT, \
391 .dst_reg = 0, \
392 .src_reg = 0, \
393 .off = 0, \
394 .imm = 0 })
395
396 /* Speculation barrier */
397
398 #define BPF_ST_NOSPEC() \
399 ((struct bpf_insn) { \
400 .code = BPF_ST | BPF_NOSPEC, \
401 .dst_reg = 0, \
402 .src_reg = 0, \
403 .off = 0, \
404 .imm = 0 })
405
406 /* Internal classic blocks for direct assignment */
407
408 #define __BPF_STMT(CODE, K) \
409 ((struct sock_filter) BPF_STMT(CODE, K))
410
411 #define __BPF_JUMP(CODE, K, JT, JF) \
412 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
413
414 #define bytes_to_bpf_size(bytes) \
415 ({ \
416 int bpf_size = -EINVAL; \
417 \
418 if (bytes == sizeof(u8)) \
419 bpf_size = BPF_B; \
420 else if (bytes == sizeof(u16)) \
421 bpf_size = BPF_H; \
422 else if (bytes == sizeof(u32)) \
423 bpf_size = BPF_W; \
424 else if (bytes == sizeof(u64)) \
425 bpf_size = BPF_DW; \
426 \
427 bpf_size; \
428 })
429
430 #define bpf_size_to_bytes(bpf_size) \
431 ({ \
432 int bytes = -EINVAL; \
433 \
434 if (bpf_size == BPF_B) \
435 bytes = sizeof(u8); \
436 else if (bpf_size == BPF_H) \
437 bytes = sizeof(u16); \
438 else if (bpf_size == BPF_W) \
439 bytes = sizeof(u32); \
440 else if (bpf_size == BPF_DW) \
441 bytes = sizeof(u64); \
442 \
443 bytes; \
444 })
445
446 #define BPF_SIZEOF(type) \
447 ({ \
448 const int __size = bytes_to_bpf_size(sizeof(type)); \
449 BUILD_BUG_ON(__size < 0); \
450 __size; \
451 })
452
453 #define BPF_FIELD_SIZEOF(type, field) \
454 ({ \
455 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
456 BUILD_BUG_ON(__size < 0); \
457 __size; \
458 })
459
460 #define BPF_LDST_BYTES(insn) \
461 ({ \
462 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
463 WARN_ON(__size < 0); \
464 __size; \
465 })
466
467 #define __BPF_MAP_0(m, v, ...) v
468 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
469 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
470 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
471 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
472 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
473
474 #define __BPF_REG_0(...) __BPF_PAD(5)
475 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
476 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
477 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
478 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
479 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
480
481 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
482 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
483
484 #define __BPF_CAST(t, a) \
485 (__force t) \
486 (__force \
487 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
488 (unsigned long)0, (t)0))) a
489 #define __BPF_V void
490 #define __BPF_N
491
492 #define __BPF_DECL_ARGS(t, a) t a
493 #define __BPF_DECL_REGS(t, a) u64 a
494
495 #define __BPF_PAD(n) \
496 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
497 u64, __ur_3, u64, __ur_4, u64, __ur_5)
498
499 #define BPF_CALL_x(x, name, ...) \
500 static __always_inline \
501 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
502 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
503 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
504 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
505 { \
506 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
507 } \
508 static __always_inline \
509 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
510
511 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
512 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
513 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
514 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
515 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
516 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
517
518 #define bpf_ctx_range(TYPE, MEMBER) \
519 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
520 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
521 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
522 #if BITS_PER_LONG == 64
523 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
524 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
525 #else
526 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
527 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
528 #endif /* BITS_PER_LONG == 64 */
529
530 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
531 ({ \
532 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
533 *(PTR_SIZE) = (SIZE); \
534 offsetof(TYPE, MEMBER); \
535 })
536
537 /* A struct sock_filter is architecture independent. */
538 struct compat_sock_fprog {
539 u16 len;
540 compat_uptr_t filter; /* struct sock_filter * */
541 };
542
543 struct sock_fprog_kern {
544 u16 len;
545 struct sock_filter *filter;
546 };
547
548 /* Some arches need doubleword alignment for their instructions and/or data */
549 #define BPF_IMAGE_ALIGNMENT 8
550
551 struct bpf_binary_header {
552 u32 pages;
553 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
554 };
555
556 struct bpf_prog_stats {
557 u64 cnt;
558 u64 nsecs;
559 u64 misses;
560 struct u64_stats_sync syncp;
561 } __aligned(2 * sizeof(u64));
562
563 struct bpf_prog {
564 u16 pages; /* Number of allocated pages */
565 u16 jited:1, /* Is our filter JIT'ed? */
566 jit_requested:1,/* archs need to JIT the prog */
567 gpl_compatible:1, /* Is filter GPL compatible? */
568 cb_access:1, /* Is control block accessed? */
569 dst_needed:1, /* Do we need dst entry? */
570 blinded:1, /* Was blinded */
571 is_func:1, /* program is a bpf function */
572 kprobe_override:1, /* Do we override a kprobe? */
573 has_callchain_buf:1, /* callchain buffer allocated? */
574 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
575 call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */
576 call_get_func_ip:1; /* Do we call get_func_ip() */
577 enum bpf_prog_type type; /* Type of BPF program */
578 enum bpf_attach_type expected_attach_type; /* For some prog types */
579 u32 len; /* Number of filter blocks */
580 u32 jited_len; /* Size of jited insns in bytes */
581 u8 tag[BPF_TAG_SIZE];
582 struct bpf_prog_stats __percpu *stats;
583 int __percpu *active;
584 unsigned int (*bpf_func)(const void *ctx,
585 const struct bpf_insn *insn);
586 struct bpf_prog_aux *aux; /* Auxiliary fields */
587 struct sock_fprog_kern *orig_prog; /* Original BPF program */
588 /* Instructions for interpreter */
589 struct sock_filter insns[0];
590 struct bpf_insn insnsi[];
591 };
592
593 struct sk_filter {
594 refcount_t refcnt;
595 struct rcu_head rcu;
596 struct bpf_prog *prog;
597 };
598
599 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
600
601 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
602 const struct bpf_insn *insnsi,
603 unsigned int (*bpf_func)(const void *,
604 const struct bpf_insn *));
605
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)606 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
607 const void *ctx,
608 bpf_dispatcher_fn dfunc)
609 {
610 u32 ret;
611
612 cant_migrate();
613 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
614 struct bpf_prog_stats *stats;
615 u64 start = sched_clock();
616
617 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
618 stats = this_cpu_ptr(prog->stats);
619 u64_stats_update_begin(&stats->syncp);
620 stats->cnt++;
621 stats->nsecs += sched_clock() - start;
622 u64_stats_update_end(&stats->syncp);
623 } else {
624 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
625 }
626 return ret;
627 }
628
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)629 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
630 {
631 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
632 }
633
634 /*
635 * Use in preemptible and therefore migratable context to make sure that
636 * the execution of the BPF program runs on one CPU.
637 *
638 * This uses migrate_disable/enable() explicitly to document that the
639 * invocation of a BPF program does not require reentrancy protection
640 * against a BPF program which is invoked from a preempting task.
641 *
642 * For non RT enabled kernels migrate_disable/enable() maps to
643 * preempt_disable/enable(), i.e. it disables also preemption.
644 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)645 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
646 const void *ctx)
647 {
648 u32 ret;
649
650 migrate_disable();
651 ret = bpf_prog_run(prog, ctx);
652 migrate_enable();
653 return ret;
654 }
655
656 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
657
658 struct bpf_skb_data_end {
659 struct qdisc_skb_cb qdisc_cb;
660 void *data_meta;
661 void *data_end;
662 };
663
664 struct bpf_nh_params {
665 u32 nh_family;
666 union {
667 u32 ipv4_nh;
668 struct in6_addr ipv6_nh;
669 };
670 };
671
672 struct bpf_redirect_info {
673 u32 flags;
674 u32 tgt_index;
675 void *tgt_value;
676 struct bpf_map *map;
677 u32 map_id;
678 enum bpf_map_type map_type;
679 u32 kern_flags;
680 struct bpf_nh_params nh;
681 };
682
683 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
684
685 /* flags for bpf_redirect_info kern_flags */
686 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
687
688 /* Compute the linear packet data range [data, data_end) which
689 * will be accessed by various program types (cls_bpf, act_bpf,
690 * lwt, ...). Subsystems allowing direct data access must (!)
691 * ensure that cb[] area can be written to when BPF program is
692 * invoked (otherwise cb[] save/restore is necessary).
693 */
bpf_compute_data_pointers(struct sk_buff * skb)694 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
695 {
696 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
697
698 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
699 cb->data_meta = skb->data - skb_metadata_len(skb);
700 cb->data_end = skb->data + skb_headlen(skb);
701 }
702
703 /* Similar to bpf_compute_data_pointers(), except that save orginal
704 * data in cb->data and cb->meta_data for restore.
705 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)706 static inline void bpf_compute_and_save_data_end(
707 struct sk_buff *skb, void **saved_data_end)
708 {
709 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
710
711 *saved_data_end = cb->data_end;
712 cb->data_end = skb->data + skb_headlen(skb);
713 }
714
715 /* Restore data saved by bpf_compute_data_pointers(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)716 static inline void bpf_restore_data_end(
717 struct sk_buff *skb, void *saved_data_end)
718 {
719 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
720
721 cb->data_end = saved_data_end;
722 }
723
bpf_skb_cb(const struct sk_buff * skb)724 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
725 {
726 /* eBPF programs may read/write skb->cb[] area to transfer meta
727 * data between tail calls. Since this also needs to work with
728 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
729 *
730 * In some socket filter cases, the cb unfortunately needs to be
731 * saved/restored so that protocol specific skb->cb[] data won't
732 * be lost. In any case, due to unpriviledged eBPF programs
733 * attached to sockets, we need to clear the bpf_skb_cb() area
734 * to not leak previous contents to user space.
735 */
736 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
737 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
738 sizeof_field(struct qdisc_skb_cb, data));
739
740 return qdisc_skb_cb(skb)->data;
741 }
742
743 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)744 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
745 const void *ctx)
746 {
747 const struct sk_buff *skb = ctx;
748 u8 *cb_data = bpf_skb_cb(skb);
749 u8 cb_saved[BPF_SKB_CB_LEN];
750 u32 res;
751
752 if (unlikely(prog->cb_access)) {
753 memcpy(cb_saved, cb_data, sizeof(cb_saved));
754 memset(cb_data, 0, sizeof(cb_saved));
755 }
756
757 res = bpf_prog_run(prog, skb);
758
759 if (unlikely(prog->cb_access))
760 memcpy(cb_data, cb_saved, sizeof(cb_saved));
761
762 return res;
763 }
764
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)765 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
766 struct sk_buff *skb)
767 {
768 u32 res;
769
770 migrate_disable();
771 res = __bpf_prog_run_save_cb(prog, skb);
772 migrate_enable();
773 return res;
774 }
775
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)776 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
777 struct sk_buff *skb)
778 {
779 u8 *cb_data = bpf_skb_cb(skb);
780 u32 res;
781
782 if (unlikely(prog->cb_access))
783 memset(cb_data, 0, BPF_SKB_CB_LEN);
784
785 res = bpf_prog_run_pin_on_cpu(prog, skb);
786 return res;
787 }
788
789 DECLARE_BPF_DISPATCHER(xdp)
790
791 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
792
793 u32 xdp_master_redirect(struct xdp_buff *xdp);
794
bpf_prog_run_xdp(const struct bpf_prog * prog,struct xdp_buff * xdp)795 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
796 struct xdp_buff *xdp)
797 {
798 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus
799 * under local_bh_disable(), which provides the needed RCU protection
800 * for accessing map entries.
801 */
802 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
803
804 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) {
805 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev))
806 act = xdp_master_redirect(xdp);
807 }
808
809 return act;
810 }
811
812 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
813
bpf_prog_insn_size(const struct bpf_prog * prog)814 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
815 {
816 return prog->len * sizeof(struct bpf_insn);
817 }
818
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)819 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
820 {
821 return round_up(bpf_prog_insn_size(prog) +
822 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
823 }
824
bpf_prog_size(unsigned int proglen)825 static inline unsigned int bpf_prog_size(unsigned int proglen)
826 {
827 return max(sizeof(struct bpf_prog),
828 offsetof(struct bpf_prog, insns[proglen]));
829 }
830
bpf_prog_was_classic(const struct bpf_prog * prog)831 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
832 {
833 /* When classic BPF programs have been loaded and the arch
834 * does not have a classic BPF JIT (anymore), they have been
835 * converted via bpf_migrate_filter() to eBPF and thus always
836 * have an unspec program type.
837 */
838 return prog->type == BPF_PROG_TYPE_UNSPEC;
839 }
840
bpf_ctx_off_adjust_machine(u32 size)841 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
842 {
843 const u32 size_machine = sizeof(unsigned long);
844
845 if (size > size_machine && size % size_machine == 0)
846 size = size_machine;
847
848 return size;
849 }
850
851 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)852 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
853 {
854 return size <= size_default && (size & (size - 1)) == 0;
855 }
856
857 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)858 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
859 {
860 u8 access_off = off & (size_default - 1);
861
862 #ifdef __LITTLE_ENDIAN
863 return access_off;
864 #else
865 return size_default - (access_off + size);
866 #endif
867 }
868
869 #define bpf_ctx_wide_access_ok(off, size, type, field) \
870 (size == sizeof(__u64) && \
871 off >= offsetof(type, field) && \
872 off + sizeof(__u64) <= offsetofend(type, field) && \
873 off % sizeof(__u64) == 0)
874
875 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
876
bpf_prog_lock_ro(struct bpf_prog * fp)877 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
878 {
879 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
880 if (!fp->jited) {
881 set_vm_flush_reset_perms(fp);
882 set_memory_ro((unsigned long)fp, fp->pages);
883 }
884 #endif
885 }
886
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)887 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
888 {
889 set_vm_flush_reset_perms(hdr);
890 set_memory_ro((unsigned long)hdr, hdr->pages);
891 set_memory_x((unsigned long)hdr, hdr->pages);
892 }
893
894 static inline struct bpf_binary_header *
bpf_jit_binary_hdr(const struct bpf_prog * fp)895 bpf_jit_binary_hdr(const struct bpf_prog *fp)
896 {
897 unsigned long real_start = (unsigned long)fp->bpf_func;
898 unsigned long addr = real_start & PAGE_MASK;
899
900 return (void *)addr;
901 }
902
903 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)904 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
905 {
906 return sk_filter_trim_cap(sk, skb, 1);
907 }
908
909 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
910 void bpf_prog_free(struct bpf_prog *fp);
911
912 bool bpf_opcode_in_insntable(u8 code);
913
914 void bpf_prog_free_linfo(struct bpf_prog *prog);
915 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
916 const u32 *insn_to_jit_off);
917 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
918 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
919
920 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
921 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
922 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
923 gfp_t gfp_extra_flags);
924 void __bpf_prog_free(struct bpf_prog *fp);
925
bpf_prog_unlock_free(struct bpf_prog * fp)926 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
927 {
928 __bpf_prog_free(fp);
929 }
930
931 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
932 unsigned int flen);
933
934 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
935 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
936 bpf_aux_classic_check_t trans, bool save_orig);
937 void bpf_prog_destroy(struct bpf_prog *fp);
938
939 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
940 int sk_attach_bpf(u32 ufd, struct sock *sk);
941 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
942 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
943 void sk_reuseport_prog_free(struct bpf_prog *prog);
944 int sk_detach_filter(struct sock *sk);
945 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
946 unsigned int len);
947
948 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
949 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
950
951 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
952 #define __bpf_call_base_args \
953 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
954 (void *)__bpf_call_base)
955
956 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
957 void bpf_jit_compile(struct bpf_prog *prog);
958 bool bpf_jit_needs_zext(void);
959 bool bpf_jit_supports_kfunc_call(void);
960 bool bpf_helper_changes_pkt_data(void *func);
961
bpf_dump_raw_ok(const struct cred * cred)962 static inline bool bpf_dump_raw_ok(const struct cred *cred)
963 {
964 /* Reconstruction of call-sites is dependent on kallsyms,
965 * thus make dump the same restriction.
966 */
967 return kallsyms_show_value(cred);
968 }
969
970 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
971 const struct bpf_insn *patch, u32 len);
972 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
973
974 void bpf_clear_redirect_map(struct bpf_map *map);
975
xdp_return_frame_no_direct(void)976 static inline bool xdp_return_frame_no_direct(void)
977 {
978 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
979
980 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
981 }
982
xdp_set_return_frame_no_direct(void)983 static inline void xdp_set_return_frame_no_direct(void)
984 {
985 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
986
987 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
988 }
989
xdp_clear_return_frame_no_direct(void)990 static inline void xdp_clear_return_frame_no_direct(void)
991 {
992 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
993
994 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
995 }
996
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)997 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
998 unsigned int pktlen)
999 {
1000 unsigned int len;
1001
1002 if (unlikely(!(fwd->flags & IFF_UP)))
1003 return -ENETDOWN;
1004
1005 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1006 if (pktlen > len)
1007 return -EMSGSIZE;
1008
1009 return 0;
1010 }
1011
1012 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1013 * same cpu context. Further for best results no more than a single map
1014 * for the do_redirect/do_flush pair should be used. This limitation is
1015 * because we only track one map and force a flush when the map changes.
1016 * This does not appear to be a real limitation for existing software.
1017 */
1018 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1019 struct xdp_buff *xdp, struct bpf_prog *prog);
1020 int xdp_do_redirect(struct net_device *dev,
1021 struct xdp_buff *xdp,
1022 struct bpf_prog *prog);
1023 void xdp_do_flush(void);
1024
1025 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
1026 * it is no longer only flushing maps. Keep this define for compatibility
1027 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
1028 */
1029 #define xdp_do_flush_map xdp_do_flush
1030
1031 void bpf_warn_invalid_xdp_action(u32 act);
1032
1033 #ifdef CONFIG_INET
1034 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1035 struct bpf_prog *prog, struct sk_buff *skb,
1036 struct sock *migrating_sk,
1037 u32 hash);
1038 #else
1039 static inline struct sock *
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)1040 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1041 struct bpf_prog *prog, struct sk_buff *skb,
1042 struct sock *migrating_sk,
1043 u32 hash)
1044 {
1045 return NULL;
1046 }
1047 #endif
1048
1049 #ifdef CONFIG_BPF_JIT
1050 extern int bpf_jit_enable;
1051 extern int bpf_jit_harden;
1052 extern int bpf_jit_kallsyms;
1053 extern long bpf_jit_limit;
1054 extern long bpf_jit_limit_max;
1055
1056 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1057
1058 struct bpf_binary_header *
1059 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1060 unsigned int alignment,
1061 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1062 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1063 u64 bpf_jit_alloc_exec_limit(void);
1064 void *bpf_jit_alloc_exec(unsigned long size);
1065 void bpf_jit_free_exec(void *addr);
1066 void bpf_jit_free(struct bpf_prog *fp);
1067
1068 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1069 struct bpf_jit_poke_descriptor *poke);
1070
1071 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1072 const struct bpf_insn *insn, bool extra_pass,
1073 u64 *func_addr, bool *func_addr_fixed);
1074
1075 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1076 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1077
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1078 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1079 u32 pass, void *image)
1080 {
1081 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1082 proglen, pass, image, current->comm, task_pid_nr(current));
1083
1084 if (image)
1085 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1086 16, 1, image, proglen, false);
1087 }
1088
bpf_jit_is_ebpf(void)1089 static inline bool bpf_jit_is_ebpf(void)
1090 {
1091 # ifdef CONFIG_HAVE_EBPF_JIT
1092 return true;
1093 # else
1094 return false;
1095 # endif
1096 }
1097
ebpf_jit_enabled(void)1098 static inline bool ebpf_jit_enabled(void)
1099 {
1100 return bpf_jit_enable && bpf_jit_is_ebpf();
1101 }
1102
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1103 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1104 {
1105 return fp->jited && bpf_jit_is_ebpf();
1106 }
1107
bpf_jit_blinding_enabled(struct bpf_prog * prog)1108 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1109 {
1110 /* These are the prerequisites, should someone ever have the
1111 * idea to call blinding outside of them, we make sure to
1112 * bail out.
1113 */
1114 if (!bpf_jit_is_ebpf())
1115 return false;
1116 if (!prog->jit_requested)
1117 return false;
1118 if (!bpf_jit_harden)
1119 return false;
1120 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1121 return false;
1122
1123 return true;
1124 }
1125
bpf_jit_kallsyms_enabled(void)1126 static inline bool bpf_jit_kallsyms_enabled(void)
1127 {
1128 /* There are a couple of corner cases where kallsyms should
1129 * not be enabled f.e. on hardening.
1130 */
1131 if (bpf_jit_harden)
1132 return false;
1133 if (!bpf_jit_kallsyms)
1134 return false;
1135 if (bpf_jit_kallsyms == 1)
1136 return true;
1137
1138 return false;
1139 }
1140
1141 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1142 unsigned long *off, char *sym);
1143 bool is_bpf_text_address(unsigned long addr);
1144 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1145 char *sym);
1146
1147 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1148 bpf_address_lookup(unsigned long addr, unsigned long *size,
1149 unsigned long *off, char **modname, char *sym)
1150 {
1151 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1152
1153 if (ret && modname)
1154 *modname = NULL;
1155 return ret;
1156 }
1157
1158 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1159 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1160
1161 #else /* CONFIG_BPF_JIT */
1162
ebpf_jit_enabled(void)1163 static inline bool ebpf_jit_enabled(void)
1164 {
1165 return false;
1166 }
1167
bpf_jit_blinding_enabled(struct bpf_prog * prog)1168 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1169 {
1170 return false;
1171 }
1172
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1173 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1174 {
1175 return false;
1176 }
1177
1178 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1179 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1180 struct bpf_jit_poke_descriptor *poke)
1181 {
1182 return -ENOTSUPP;
1183 }
1184
bpf_jit_free(struct bpf_prog * fp)1185 static inline void bpf_jit_free(struct bpf_prog *fp)
1186 {
1187 bpf_prog_unlock_free(fp);
1188 }
1189
bpf_jit_kallsyms_enabled(void)1190 static inline bool bpf_jit_kallsyms_enabled(void)
1191 {
1192 return false;
1193 }
1194
1195 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1196 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1197 unsigned long *off, char *sym)
1198 {
1199 return NULL;
1200 }
1201
is_bpf_text_address(unsigned long addr)1202 static inline bool is_bpf_text_address(unsigned long addr)
1203 {
1204 return false;
1205 }
1206
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1207 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1208 char *type, char *sym)
1209 {
1210 return -ERANGE;
1211 }
1212
1213 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1214 bpf_address_lookup(unsigned long addr, unsigned long *size,
1215 unsigned long *off, char **modname, char *sym)
1216 {
1217 return NULL;
1218 }
1219
bpf_prog_kallsyms_add(struct bpf_prog * fp)1220 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1221 {
1222 }
1223
bpf_prog_kallsyms_del(struct bpf_prog * fp)1224 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1225 {
1226 }
1227
1228 #endif /* CONFIG_BPF_JIT */
1229
1230 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1231
1232 #define BPF_ANC BIT(15)
1233
bpf_needs_clear_a(const struct sock_filter * first)1234 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1235 {
1236 switch (first->code) {
1237 case BPF_RET | BPF_K:
1238 case BPF_LD | BPF_W | BPF_LEN:
1239 return false;
1240
1241 case BPF_LD | BPF_W | BPF_ABS:
1242 case BPF_LD | BPF_H | BPF_ABS:
1243 case BPF_LD | BPF_B | BPF_ABS:
1244 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1245 return true;
1246 return false;
1247
1248 default:
1249 return true;
1250 }
1251 }
1252
bpf_anc_helper(const struct sock_filter * ftest)1253 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1254 {
1255 BUG_ON(ftest->code & BPF_ANC);
1256
1257 switch (ftest->code) {
1258 case BPF_LD | BPF_W | BPF_ABS:
1259 case BPF_LD | BPF_H | BPF_ABS:
1260 case BPF_LD | BPF_B | BPF_ABS:
1261 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1262 return BPF_ANC | SKF_AD_##CODE
1263 switch (ftest->k) {
1264 BPF_ANCILLARY(PROTOCOL);
1265 BPF_ANCILLARY(PKTTYPE);
1266 BPF_ANCILLARY(IFINDEX);
1267 BPF_ANCILLARY(NLATTR);
1268 BPF_ANCILLARY(NLATTR_NEST);
1269 BPF_ANCILLARY(MARK);
1270 BPF_ANCILLARY(QUEUE);
1271 BPF_ANCILLARY(HATYPE);
1272 BPF_ANCILLARY(RXHASH);
1273 BPF_ANCILLARY(CPU);
1274 BPF_ANCILLARY(ALU_XOR_X);
1275 BPF_ANCILLARY(VLAN_TAG);
1276 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1277 BPF_ANCILLARY(PAY_OFFSET);
1278 BPF_ANCILLARY(RANDOM);
1279 BPF_ANCILLARY(VLAN_TPID);
1280 }
1281 fallthrough;
1282 default:
1283 return ftest->code;
1284 }
1285 }
1286
1287 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1288 int k, unsigned int size);
1289
bpf_tell_extensions(void)1290 static inline int bpf_tell_extensions(void)
1291 {
1292 return SKF_AD_MAX;
1293 }
1294
1295 struct bpf_sock_addr_kern {
1296 struct sock *sk;
1297 struct sockaddr *uaddr;
1298 /* Temporary "register" to make indirect stores to nested structures
1299 * defined above. We need three registers to make such a store, but
1300 * only two (src and dst) are available at convert_ctx_access time
1301 */
1302 u64 tmp_reg;
1303 void *t_ctx; /* Attach type specific context. */
1304 };
1305
1306 struct bpf_sock_ops_kern {
1307 struct sock *sk;
1308 union {
1309 u32 args[4];
1310 u32 reply;
1311 u32 replylong[4];
1312 };
1313 struct sk_buff *syn_skb;
1314 struct sk_buff *skb;
1315 void *skb_data_end;
1316 u8 op;
1317 u8 is_fullsock;
1318 u8 remaining_opt_len;
1319 u64 temp; /* temp and everything after is not
1320 * initialized to 0 before calling
1321 * the BPF program. New fields that
1322 * should be initialized to 0 should
1323 * be inserted before temp.
1324 * temp is scratch storage used by
1325 * sock_ops_convert_ctx_access
1326 * as temporary storage of a register.
1327 */
1328 };
1329
1330 struct bpf_sysctl_kern {
1331 struct ctl_table_header *head;
1332 struct ctl_table *table;
1333 void *cur_val;
1334 size_t cur_len;
1335 void *new_val;
1336 size_t new_len;
1337 int new_updated;
1338 int write;
1339 loff_t *ppos;
1340 /* Temporary "register" for indirect stores to ppos. */
1341 u64 tmp_reg;
1342 };
1343
1344 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1345 struct bpf_sockopt_buf {
1346 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1347 };
1348
1349 struct bpf_sockopt_kern {
1350 struct sock *sk;
1351 u8 *optval;
1352 u8 *optval_end;
1353 s32 level;
1354 s32 optname;
1355 s32 optlen;
1356 s32 retval;
1357 };
1358
1359 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1360
1361 struct bpf_sk_lookup_kern {
1362 u16 family;
1363 u16 protocol;
1364 __be16 sport;
1365 u16 dport;
1366 struct {
1367 __be32 saddr;
1368 __be32 daddr;
1369 } v4;
1370 struct {
1371 const struct in6_addr *saddr;
1372 const struct in6_addr *daddr;
1373 } v6;
1374 struct sock *selected_sk;
1375 bool no_reuseport;
1376 };
1377
1378 extern struct static_key_false bpf_sk_lookup_enabled;
1379
1380 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1381 *
1382 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1383 * SK_DROP. Their meaning is as follows:
1384 *
1385 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1386 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1387 * SK_DROP : terminate lookup with -ECONNREFUSED
1388 *
1389 * This macro aggregates return values and selected sockets from
1390 * multiple BPF programs according to following rules in order:
1391 *
1392 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1393 * macro result is SK_PASS and last ctx.selected_sk is used.
1394 * 2. If any program returned SK_DROP return value,
1395 * macro result is SK_DROP.
1396 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1397 *
1398 * Caller must ensure that the prog array is non-NULL, and that the
1399 * array as well as the programs it contains remain valid.
1400 */
1401 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1402 ({ \
1403 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1404 struct bpf_prog_array_item *_item; \
1405 struct sock *_selected_sk = NULL; \
1406 bool _no_reuseport = false; \
1407 struct bpf_prog *_prog; \
1408 bool _all_pass = true; \
1409 u32 _ret; \
1410 \
1411 migrate_disable(); \
1412 _item = &(array)->items[0]; \
1413 while ((_prog = READ_ONCE(_item->prog))) { \
1414 /* restore most recent selection */ \
1415 _ctx->selected_sk = _selected_sk; \
1416 _ctx->no_reuseport = _no_reuseport; \
1417 \
1418 _ret = func(_prog, _ctx); \
1419 if (_ret == SK_PASS && _ctx->selected_sk) { \
1420 /* remember last non-NULL socket */ \
1421 _selected_sk = _ctx->selected_sk; \
1422 _no_reuseport = _ctx->no_reuseport; \
1423 } else if (_ret == SK_DROP && _all_pass) { \
1424 _all_pass = false; \
1425 } \
1426 _item++; \
1427 } \
1428 _ctx->selected_sk = _selected_sk; \
1429 _ctx->no_reuseport = _no_reuseport; \
1430 migrate_enable(); \
1431 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1432 })
1433
bpf_sk_lookup_run_v4(struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,struct sock ** psk)1434 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1435 const __be32 saddr, const __be16 sport,
1436 const __be32 daddr, const u16 dport,
1437 struct sock **psk)
1438 {
1439 struct bpf_prog_array *run_array;
1440 struct sock *selected_sk = NULL;
1441 bool no_reuseport = false;
1442
1443 rcu_read_lock();
1444 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1445 if (run_array) {
1446 struct bpf_sk_lookup_kern ctx = {
1447 .family = AF_INET,
1448 .protocol = protocol,
1449 .v4.saddr = saddr,
1450 .v4.daddr = daddr,
1451 .sport = sport,
1452 .dport = dport,
1453 };
1454 u32 act;
1455
1456 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1457 if (act == SK_PASS) {
1458 selected_sk = ctx.selected_sk;
1459 no_reuseport = ctx.no_reuseport;
1460 } else {
1461 selected_sk = ERR_PTR(-ECONNREFUSED);
1462 }
1463 }
1464 rcu_read_unlock();
1465 *psk = selected_sk;
1466 return no_reuseport;
1467 }
1468
1469 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,struct sock ** psk)1470 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1471 const struct in6_addr *saddr,
1472 const __be16 sport,
1473 const struct in6_addr *daddr,
1474 const u16 dport,
1475 struct sock **psk)
1476 {
1477 struct bpf_prog_array *run_array;
1478 struct sock *selected_sk = NULL;
1479 bool no_reuseport = false;
1480
1481 rcu_read_lock();
1482 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1483 if (run_array) {
1484 struct bpf_sk_lookup_kern ctx = {
1485 .family = AF_INET6,
1486 .protocol = protocol,
1487 .v6.saddr = saddr,
1488 .v6.daddr = daddr,
1489 .sport = sport,
1490 .dport = dport,
1491 };
1492 u32 act;
1493
1494 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1495 if (act == SK_PASS) {
1496 selected_sk = ctx.selected_sk;
1497 no_reuseport = ctx.no_reuseport;
1498 } else {
1499 selected_sk = ERR_PTR(-ECONNREFUSED);
1500 }
1501 }
1502 rcu_read_unlock();
1503 *psk = selected_sk;
1504 return no_reuseport;
1505 }
1506 #endif /* IS_ENABLED(CONFIG_IPV6) */
1507
__bpf_xdp_redirect_map(struct bpf_map * map,u32 ifindex,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1508 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex,
1509 u64 flags, const u64 flag_mask,
1510 void *lookup_elem(struct bpf_map *map, u32 key))
1511 {
1512 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1513 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1514
1515 /* Lower bits of the flags are used as return code on lookup failure */
1516 if (unlikely(flags & ~(action_mask | flag_mask)))
1517 return XDP_ABORTED;
1518
1519 ri->tgt_value = lookup_elem(map, ifindex);
1520 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1521 /* If the lookup fails we want to clear out the state in the
1522 * redirect_info struct completely, so that if an eBPF program
1523 * performs multiple lookups, the last one always takes
1524 * precedence.
1525 */
1526 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1527 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1528 return flags & action_mask;
1529 }
1530
1531 ri->tgt_index = ifindex;
1532 ri->map_id = map->id;
1533 ri->map_type = map->map_type;
1534
1535 if (flags & BPF_F_BROADCAST) {
1536 WRITE_ONCE(ri->map, map);
1537 ri->flags = flags;
1538 } else {
1539 WRITE_ONCE(ri->map, NULL);
1540 ri->flags = 0;
1541 }
1542
1543 return XDP_REDIRECT;
1544 }
1545
1546 #endif /* __LINUX_FILTER_H__ */
1547
