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/bpf.h>
10 #include <linux/refcount.h>
11 #include <linux/compat.h>
12 #include <linux/skbuff.h>
13 #include <linux/linkage.h>
14 #include <linux/printk.h>
15 #include <linux/workqueue.h>
16 #include <linux/sched.h>
17 #include <linux/sched/clock.h>
18 #include <linux/capability.h>
19 #include <linux/set_memory.h>
20 #include <linux/kallsyms.h>
21 #include <linux/if_vlan.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sockptr.h>
24 #include <crypto/sha1.h>
25 #include <linux/u64_stats_sync.h>
26
27 #include <net/sch_generic.h>
28
29 #include <asm/byteorder.h>
30 #include <uapi/linux/filter.h>
31
32 struct sk_buff;
33 struct sock;
34 struct seccomp_data;
35 struct bpf_prog_aux;
36 struct xdp_rxq_info;
37 struct xdp_buff;
38 struct sock_reuseport;
39 struct ctl_table;
40 struct ctl_table_header;
41
42 /* ArgX, context and stack frame pointer register positions. Note,
43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
44 * calls in BPF_CALL instruction.
45 */
46 #define BPF_REG_ARG1 BPF_REG_1
47 #define BPF_REG_ARG2 BPF_REG_2
48 #define BPF_REG_ARG3 BPF_REG_3
49 #define BPF_REG_ARG4 BPF_REG_4
50 #define BPF_REG_ARG5 BPF_REG_5
51 #define BPF_REG_CTX BPF_REG_6
52 #define BPF_REG_FP BPF_REG_10
53
54 /* Additional register mappings for converted user programs. */
55 #define BPF_REG_A BPF_REG_0
56 #define BPF_REG_X BPF_REG_7
57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
60
61 /* Kernel hidden auxiliary/helper register. */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
65
66 /* unused opcode to mark special call to bpf_tail_call() helper */
67 #define BPF_TAIL_CALL 0xf0
68
69 /* unused opcode to mark special load instruction. Same as BPF_ABS */
70 #define BPF_PROBE_MEM 0x20
71
72 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
73 #define BPF_PROBE_MEMSX 0x40
74
75 /* unused opcode to mark call to interpreter with arguments */
76 #define BPF_CALL_ARGS 0xe0
77
78 /* unused opcode to mark speculation barrier for mitigating
79 * Speculative Store Bypass
80 */
81 #define BPF_NOSPEC 0xc0
82
83 /* As per nm, we expose JITed images as text (code) section for
84 * kallsyms. That way, tools like perf can find it to match
85 * addresses.
86 */
87 #define BPF_SYM_ELF_TYPE 't'
88
89 /* BPF program can access up to 512 bytes of stack space. */
90 #define MAX_BPF_STACK 512
91
92 /* Helper macros for filter block array initializers. */
93
94 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
95
96 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
97 ((struct bpf_insn) { \
98 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
99 .dst_reg = DST, \
100 .src_reg = SRC, \
101 .off = OFF, \
102 .imm = 0 })
103
104 #define BPF_ALU64_REG(OP, DST, SRC) \
105 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
106
107 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
108 ((struct bpf_insn) { \
109 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
110 .dst_reg = DST, \
111 .src_reg = SRC, \
112 .off = OFF, \
113 .imm = 0 })
114
115 #define BPF_ALU32_REG(OP, DST, SRC) \
116 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
117
118 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
119
120 #define BPF_ALU64_IMM(OP, DST, IMM) \
121 ((struct bpf_insn) { \
122 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
123 .dst_reg = DST, \
124 .src_reg = 0, \
125 .off = 0, \
126 .imm = IMM })
127
128 #define BPF_ALU32_IMM(OP, DST, IMM) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
131 .dst_reg = DST, \
132 .src_reg = 0, \
133 .off = 0, \
134 .imm = IMM })
135
136 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
137
138 #define BPF_ENDIAN(TYPE, DST, LEN) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
141 .dst_reg = DST, \
142 .src_reg = 0, \
143 .off = 0, \
144 .imm = LEN })
145
146 /* Short form of mov, dst_reg = src_reg */
147
148 #define BPF_MOV64_REG(DST, SRC) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
151 .dst_reg = DST, \
152 .src_reg = SRC, \
153 .off = 0, \
154 .imm = 0 })
155
156 #define BPF_MOV32_REG(DST, SRC) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU | BPF_MOV | BPF_X, \
159 .dst_reg = DST, \
160 .src_reg = SRC, \
161 .off = 0, \
162 .imm = 0 })
163
164 /* Short form of mov, dst_reg = imm32 */
165
166 #define BPF_MOV64_IMM(DST, IMM) \
167 ((struct bpf_insn) { \
168 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
169 .dst_reg = DST, \
170 .src_reg = 0, \
171 .off = 0, \
172 .imm = IMM })
173
174 #define BPF_MOV32_IMM(DST, IMM) \
175 ((struct bpf_insn) { \
176 .code = BPF_ALU | BPF_MOV | BPF_K, \
177 .dst_reg = DST, \
178 .src_reg = 0, \
179 .off = 0, \
180 .imm = IMM })
181
182 /* Special form of mov32, used for doing explicit zero extension on dst. */
183 #define BPF_ZEXT_REG(DST) \
184 ((struct bpf_insn) { \
185 .code = BPF_ALU | BPF_MOV | BPF_X, \
186 .dst_reg = DST, \
187 .src_reg = DST, \
188 .off = 0, \
189 .imm = 1 })
190
insn_is_zext(const struct bpf_insn * insn)191 static inline bool insn_is_zext(const struct bpf_insn *insn)
192 {
193 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
194 }
195
196 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
197 #define BPF_LD_IMM64(DST, IMM) \
198 BPF_LD_IMM64_RAW(DST, 0, IMM)
199
200 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
201 ((struct bpf_insn) { \
202 .code = BPF_LD | BPF_DW | BPF_IMM, \
203 .dst_reg = DST, \
204 .src_reg = SRC, \
205 .off = 0, \
206 .imm = (__u32) (IMM) }), \
207 ((struct bpf_insn) { \
208 .code = 0, /* zero is reserved opcode */ \
209 .dst_reg = 0, \
210 .src_reg = 0, \
211 .off = 0, \
212 .imm = ((__u64) (IMM)) >> 32 })
213
214 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
215 #define BPF_LD_MAP_FD(DST, MAP_FD) \
216 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
217
218 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
219
220 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
221 ((struct bpf_insn) { \
222 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
223 .dst_reg = DST, \
224 .src_reg = SRC, \
225 .off = 0, \
226 .imm = IMM })
227
228 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
229 ((struct bpf_insn) { \
230 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
231 .dst_reg = DST, \
232 .src_reg = SRC, \
233 .off = 0, \
234 .imm = IMM })
235
236 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
237
238 #define BPF_LD_ABS(SIZE, IMM) \
239 ((struct bpf_insn) { \
240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
241 .dst_reg = 0, \
242 .src_reg = 0, \
243 .off = 0, \
244 .imm = IMM })
245
246 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
247
248 #define BPF_LD_IND(SIZE, SRC, IMM) \
249 ((struct bpf_insn) { \
250 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
251 .dst_reg = 0, \
252 .src_reg = SRC, \
253 .off = 0, \
254 .imm = IMM })
255
256 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
257
258 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
259 ((struct bpf_insn) { \
260 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
261 .dst_reg = DST, \
262 .src_reg = SRC, \
263 .off = OFF, \
264 .imm = 0 })
265
266 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
267
268 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
269 ((struct bpf_insn) { \
270 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
271 .dst_reg = DST, \
272 .src_reg = SRC, \
273 .off = OFF, \
274 .imm = 0 })
275
276
277 /*
278 * Atomic operations:
279 *
280 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
281 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
282 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
283 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
284 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
285 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
286 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
287 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
288 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
289 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
290 */
291
292 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
293 ((struct bpf_insn) { \
294 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
295 .dst_reg = DST, \
296 .src_reg = SRC, \
297 .off = OFF, \
298 .imm = OP })
299
300 /* Legacy alias */
301 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
302
303 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
304
305 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
306 ((struct bpf_insn) { \
307 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
308 .dst_reg = DST, \
309 .src_reg = 0, \
310 .off = OFF, \
311 .imm = IMM })
312
313 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
314
315 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
316 ((struct bpf_insn) { \
317 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
318 .dst_reg = DST, \
319 .src_reg = SRC, \
320 .off = OFF, \
321 .imm = 0 })
322
323 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
324
325 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
328 .dst_reg = DST, \
329 .src_reg = 0, \
330 .off = OFF, \
331 .imm = IMM })
332
333 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
334
335 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
338 .dst_reg = DST, \
339 .src_reg = SRC, \
340 .off = OFF, \
341 .imm = 0 })
342
343 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
344
345 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
348 .dst_reg = DST, \
349 .src_reg = 0, \
350 .off = OFF, \
351 .imm = IMM })
352
353 /* Unconditional jumps, goto pc + off16 */
354
355 #define BPF_JMP_A(OFF) \
356 ((struct bpf_insn) { \
357 .code = BPF_JMP | BPF_JA, \
358 .dst_reg = 0, \
359 .src_reg = 0, \
360 .off = OFF, \
361 .imm = 0 })
362
363 /* Relative call */
364
365 #define BPF_CALL_REL(TGT) \
366 ((struct bpf_insn) { \
367 .code = BPF_JMP | BPF_CALL, \
368 .dst_reg = 0, \
369 .src_reg = BPF_PSEUDO_CALL, \
370 .off = 0, \
371 .imm = TGT })
372
373 /* Convert function address to BPF immediate */
374
375 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
376
377 #define BPF_EMIT_CALL(FUNC) \
378 ((struct bpf_insn) { \
379 .code = BPF_JMP | BPF_CALL, \
380 .dst_reg = 0, \
381 .src_reg = 0, \
382 .off = 0, \
383 .imm = BPF_CALL_IMM(FUNC) })
384
385 /* Raw code statement block */
386
387 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
388 ((struct bpf_insn) { \
389 .code = CODE, \
390 .dst_reg = DST, \
391 .src_reg = SRC, \
392 .off = OFF, \
393 .imm = IMM })
394
395 /* Program exit */
396
397 #define BPF_EXIT_INSN() \
398 ((struct bpf_insn) { \
399 .code = BPF_JMP | BPF_EXIT, \
400 .dst_reg = 0, \
401 .src_reg = 0, \
402 .off = 0, \
403 .imm = 0 })
404
405 /* Speculation barrier */
406
407 #define BPF_ST_NOSPEC() \
408 ((struct bpf_insn) { \
409 .code = BPF_ST | BPF_NOSPEC, \
410 .dst_reg = 0, \
411 .src_reg = 0, \
412 .off = 0, \
413 .imm = 0 })
414
415 /* Internal classic blocks for direct assignment */
416
417 #define __BPF_STMT(CODE, K) \
418 ((struct sock_filter) BPF_STMT(CODE, K))
419
420 #define __BPF_JUMP(CODE, K, JT, JF) \
421 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
422
423 #define bytes_to_bpf_size(bytes) \
424 ({ \
425 int bpf_size = -EINVAL; \
426 \
427 if (bytes == sizeof(u8)) \
428 bpf_size = BPF_B; \
429 else if (bytes == sizeof(u16)) \
430 bpf_size = BPF_H; \
431 else if (bytes == sizeof(u32)) \
432 bpf_size = BPF_W; \
433 else if (bytes == sizeof(u64)) \
434 bpf_size = BPF_DW; \
435 \
436 bpf_size; \
437 })
438
439 #define bpf_size_to_bytes(bpf_size) \
440 ({ \
441 int bytes = -EINVAL; \
442 \
443 if (bpf_size == BPF_B) \
444 bytes = sizeof(u8); \
445 else if (bpf_size == BPF_H) \
446 bytes = sizeof(u16); \
447 else if (bpf_size == BPF_W) \
448 bytes = sizeof(u32); \
449 else if (bpf_size == BPF_DW) \
450 bytes = sizeof(u64); \
451 \
452 bytes; \
453 })
454
455 #define BPF_SIZEOF(type) \
456 ({ \
457 const int __size = bytes_to_bpf_size(sizeof(type)); \
458 BUILD_BUG_ON(__size < 0); \
459 __size; \
460 })
461
462 #define BPF_FIELD_SIZEOF(type, field) \
463 ({ \
464 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
465 BUILD_BUG_ON(__size < 0); \
466 __size; \
467 })
468
469 #define BPF_LDST_BYTES(insn) \
470 ({ \
471 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
472 WARN_ON(__size < 0); \
473 __size; \
474 })
475
476 #define __BPF_MAP_0(m, v, ...) v
477 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
478 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
479 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
480 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
481 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
482
483 #define __BPF_REG_0(...) __BPF_PAD(5)
484 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
485 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
486 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
487 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
488 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
489
490 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
491 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
492
493 #define __BPF_CAST(t, a) \
494 (__force t) \
495 (__force \
496 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
497 (unsigned long)0, (t)0))) a
498 #define __BPF_V void
499 #define __BPF_N
500
501 #define __BPF_DECL_ARGS(t, a) t a
502 #define __BPF_DECL_REGS(t, a) u64 a
503
504 #define __BPF_PAD(n) \
505 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
506 u64, __ur_3, u64, __ur_4, u64, __ur_5)
507
508 #define BPF_CALL_x(x, name, ...) \
509 static __always_inline \
510 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
511 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
512 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
513 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
514 { \
515 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
516 } \
517 static __always_inline \
518 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
519
520 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
521 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
522 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
523 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
524 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
525 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
526
527 #define bpf_ctx_range(TYPE, MEMBER) \
528 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
529 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
530 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
531 #if BITS_PER_LONG == 64
532 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
533 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
534 #else
535 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
536 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
537 #endif /* BITS_PER_LONG == 64 */
538
539 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
540 ({ \
541 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
542 *(PTR_SIZE) = (SIZE); \
543 offsetof(TYPE, MEMBER); \
544 })
545
546 /* A struct sock_filter is architecture independent. */
547 struct compat_sock_fprog {
548 u16 len;
549 compat_uptr_t filter; /* struct sock_filter * */
550 };
551
552 struct sock_fprog_kern {
553 u16 len;
554 struct sock_filter *filter;
555 };
556
557 /* Some arches need doubleword alignment for their instructions and/or data */
558 #define BPF_IMAGE_ALIGNMENT 8
559
560 struct bpf_binary_header {
561 u32 size;
562 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
563 };
564
565 struct bpf_prog_stats {
566 u64_stats_t cnt;
567 u64_stats_t nsecs;
568 u64_stats_t misses;
569 struct u64_stats_sync syncp;
570 } __aligned(2 * sizeof(u64));
571
572 struct sk_filter {
573 refcount_t refcnt;
574 struct rcu_head rcu;
575 struct bpf_prog *prog;
576 };
577
578 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
579
580 extern struct mutex nf_conn_btf_access_lock;
581 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
582 const struct bpf_reg_state *reg,
583 int off, int size);
584
585 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
586 const struct bpf_insn *insnsi,
587 unsigned int (*bpf_func)(const void *,
588 const struct bpf_insn *));
589
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)590 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
591 const void *ctx,
592 bpf_dispatcher_fn dfunc)
593 {
594 u32 ret;
595
596 cant_migrate();
597 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
598 struct bpf_prog_stats *stats;
599 u64 start = sched_clock();
600 unsigned long flags;
601
602 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
603 stats = this_cpu_ptr(prog->stats);
604 flags = u64_stats_update_begin_irqsave(&stats->syncp);
605 u64_stats_inc(&stats->cnt);
606 u64_stats_add(&stats->nsecs, sched_clock() - start);
607 u64_stats_update_end_irqrestore(&stats->syncp, flags);
608 } else {
609 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
610 }
611 return ret;
612 }
613
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)614 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
615 {
616 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
617 }
618
619 /*
620 * Use in preemptible and therefore migratable context to make sure that
621 * the execution of the BPF program runs on one CPU.
622 *
623 * This uses migrate_disable/enable() explicitly to document that the
624 * invocation of a BPF program does not require reentrancy protection
625 * against a BPF program which is invoked from a preempting task.
626 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)627 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
628 const void *ctx)
629 {
630 u32 ret;
631
632 migrate_disable();
633 ret = bpf_prog_run(prog, ctx);
634 migrate_enable();
635 return ret;
636 }
637
638 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
639
640 struct bpf_skb_data_end {
641 struct qdisc_skb_cb qdisc_cb;
642 void *data_meta;
643 void *data_end;
644 };
645
646 struct bpf_nh_params {
647 u32 nh_family;
648 union {
649 u32 ipv4_nh;
650 struct in6_addr ipv6_nh;
651 };
652 };
653
654 struct bpf_redirect_info {
655 u64 tgt_index;
656 void *tgt_value;
657 struct bpf_map *map;
658 u32 flags;
659 u32 kern_flags;
660 u32 map_id;
661 enum bpf_map_type map_type;
662 struct bpf_nh_params nh;
663 };
664
665 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
666
667 /* flags for bpf_redirect_info kern_flags */
668 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
669
670 /* Compute the linear packet data range [data, data_end) which
671 * will be accessed by various program types (cls_bpf, act_bpf,
672 * lwt, ...). Subsystems allowing direct data access must (!)
673 * ensure that cb[] area can be written to when BPF program is
674 * invoked (otherwise cb[] save/restore is necessary).
675 */
bpf_compute_data_pointers(struct sk_buff * skb)676 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
677 {
678 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
679
680 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
681 cb->data_meta = skb->data - skb_metadata_len(skb);
682 cb->data_end = skb->data + skb_headlen(skb);
683 }
684
685 /* Similar to bpf_compute_data_pointers(), except that save orginal
686 * data in cb->data and cb->meta_data for restore.
687 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)688 static inline void bpf_compute_and_save_data_end(
689 struct sk_buff *skb, void **saved_data_end)
690 {
691 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
692
693 *saved_data_end = cb->data_end;
694 cb->data_end = skb->data + skb_headlen(skb);
695 }
696
697 /* Restore data saved by bpf_compute_data_pointers(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)698 static inline void bpf_restore_data_end(
699 struct sk_buff *skb, void *saved_data_end)
700 {
701 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
702
703 cb->data_end = saved_data_end;
704 }
705
bpf_skb_cb(const struct sk_buff * skb)706 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
707 {
708 /* eBPF programs may read/write skb->cb[] area to transfer meta
709 * data between tail calls. Since this also needs to work with
710 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
711 *
712 * In some socket filter cases, the cb unfortunately needs to be
713 * saved/restored so that protocol specific skb->cb[] data won't
714 * be lost. In any case, due to unpriviledged eBPF programs
715 * attached to sockets, we need to clear the bpf_skb_cb() area
716 * to not leak previous contents to user space.
717 */
718 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
719 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
720 sizeof_field(struct qdisc_skb_cb, data));
721
722 return qdisc_skb_cb(skb)->data;
723 }
724
725 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)726 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
727 const void *ctx)
728 {
729 const struct sk_buff *skb = ctx;
730 u8 *cb_data = bpf_skb_cb(skb);
731 u8 cb_saved[BPF_SKB_CB_LEN];
732 u32 res;
733
734 if (unlikely(prog->cb_access)) {
735 memcpy(cb_saved, cb_data, sizeof(cb_saved));
736 memset(cb_data, 0, sizeof(cb_saved));
737 }
738
739 res = bpf_prog_run(prog, skb);
740
741 if (unlikely(prog->cb_access))
742 memcpy(cb_data, cb_saved, sizeof(cb_saved));
743
744 return res;
745 }
746
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)747 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
748 struct sk_buff *skb)
749 {
750 u32 res;
751
752 migrate_disable();
753 res = __bpf_prog_run_save_cb(prog, skb);
754 migrate_enable();
755 return res;
756 }
757
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)758 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
759 struct sk_buff *skb)
760 {
761 u8 *cb_data = bpf_skb_cb(skb);
762 u32 res;
763
764 if (unlikely(prog->cb_access))
765 memset(cb_data, 0, BPF_SKB_CB_LEN);
766
767 res = bpf_prog_run_pin_on_cpu(prog, skb);
768 return res;
769 }
770
771 DECLARE_BPF_DISPATCHER(xdp)
772
773 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
774
775 u32 xdp_master_redirect(struct xdp_buff *xdp);
776
777 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
778
bpf_prog_insn_size(const struct bpf_prog * prog)779 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
780 {
781 return prog->len * sizeof(struct bpf_insn);
782 }
783
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)784 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
785 {
786 return round_up(bpf_prog_insn_size(prog) +
787 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
788 }
789
bpf_prog_size(unsigned int proglen)790 static inline unsigned int bpf_prog_size(unsigned int proglen)
791 {
792 return max(sizeof(struct bpf_prog),
793 offsetof(struct bpf_prog, insns[proglen]));
794 }
795
bpf_prog_was_classic(const struct bpf_prog * prog)796 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
797 {
798 /* When classic BPF programs have been loaded and the arch
799 * does not have a classic BPF JIT (anymore), they have been
800 * converted via bpf_migrate_filter() to eBPF and thus always
801 * have an unspec program type.
802 */
803 return prog->type == BPF_PROG_TYPE_UNSPEC;
804 }
805
bpf_ctx_off_adjust_machine(u32 size)806 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
807 {
808 const u32 size_machine = sizeof(unsigned long);
809
810 if (size > size_machine && size % size_machine == 0)
811 size = size_machine;
812
813 return size;
814 }
815
816 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)817 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
818 {
819 return size <= size_default && (size & (size - 1)) == 0;
820 }
821
822 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)823 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
824 {
825 u8 access_off = off & (size_default - 1);
826
827 #ifdef __LITTLE_ENDIAN
828 return access_off;
829 #else
830 return size_default - (access_off + size);
831 #endif
832 }
833
834 #define bpf_ctx_wide_access_ok(off, size, type, field) \
835 (size == sizeof(__u64) && \
836 off >= offsetof(type, field) && \
837 off + sizeof(__u64) <= offsetofend(type, field) && \
838 off % sizeof(__u64) == 0)
839
840 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
841
bpf_prog_lock_ro(struct bpf_prog * fp)842 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
843 {
844 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
845 if (!fp->jited) {
846 set_vm_flush_reset_perms(fp);
847 set_memory_ro((unsigned long)fp, fp->pages);
848 }
849 #endif
850 }
851
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)852 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
853 {
854 set_vm_flush_reset_perms(hdr);
855 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
856 }
857
858 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)859 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
860 {
861 return sk_filter_trim_cap(sk, skb, 1);
862 }
863
864 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
865 void bpf_prog_free(struct bpf_prog *fp);
866
867 bool bpf_opcode_in_insntable(u8 code);
868
869 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
870 const u32 *insn_to_jit_off);
871 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
872 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
873
874 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
875 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
876 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
877 gfp_t gfp_extra_flags);
878 void __bpf_prog_free(struct bpf_prog *fp);
879
bpf_prog_unlock_free(struct bpf_prog * fp)880 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
881 {
882 __bpf_prog_free(fp);
883 }
884
885 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
886 unsigned int flen);
887
888 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
889 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
890 bpf_aux_classic_check_t trans, bool save_orig);
891 void bpf_prog_destroy(struct bpf_prog *fp);
892
893 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
894 int sk_attach_bpf(u32 ufd, struct sock *sk);
895 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
896 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
897 void sk_reuseport_prog_free(struct bpf_prog *prog);
898 int sk_detach_filter(struct sock *sk);
899 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
900
901 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
902 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
903
904 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
905 #define __bpf_call_base_args \
906 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
907 (void *)__bpf_call_base)
908
909 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
910 void bpf_jit_compile(struct bpf_prog *prog);
911 bool bpf_jit_needs_zext(void);
912 bool bpf_jit_supports_subprog_tailcalls(void);
913 bool bpf_jit_supports_kfunc_call(void);
914 bool bpf_jit_supports_far_kfunc_call(void);
915 bool bpf_helper_changes_pkt_data(void *func);
916
bpf_dump_raw_ok(const struct cred * cred)917 static inline bool bpf_dump_raw_ok(const struct cred *cred)
918 {
919 /* Reconstruction of call-sites is dependent on kallsyms,
920 * thus make dump the same restriction.
921 */
922 return kallsyms_show_value(cred);
923 }
924
925 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
926 const struct bpf_insn *patch, u32 len);
927 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
928
929 void bpf_clear_redirect_map(struct bpf_map *map);
930
xdp_return_frame_no_direct(void)931 static inline bool xdp_return_frame_no_direct(void)
932 {
933 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
934
935 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
936 }
937
xdp_set_return_frame_no_direct(void)938 static inline void xdp_set_return_frame_no_direct(void)
939 {
940 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
941
942 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
943 }
944
xdp_clear_return_frame_no_direct(void)945 static inline void xdp_clear_return_frame_no_direct(void)
946 {
947 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
948
949 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
950 }
951
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)952 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
953 unsigned int pktlen)
954 {
955 unsigned int len;
956
957 if (unlikely(!(fwd->flags & IFF_UP)))
958 return -ENETDOWN;
959
960 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
961 if (pktlen > len)
962 return -EMSGSIZE;
963
964 return 0;
965 }
966
967 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
968 * same cpu context. Further for best results no more than a single map
969 * for the do_redirect/do_flush pair should be used. This limitation is
970 * because we only track one map and force a flush when the map changes.
971 * This does not appear to be a real limitation for existing software.
972 */
973 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
974 struct xdp_buff *xdp, struct bpf_prog *prog);
975 int xdp_do_redirect(struct net_device *dev,
976 struct xdp_buff *xdp,
977 struct bpf_prog *prog);
978 int xdp_do_redirect_frame(struct net_device *dev,
979 struct xdp_buff *xdp,
980 struct xdp_frame *xdpf,
981 struct bpf_prog *prog);
982 void xdp_do_flush(void);
983
984 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
985 * it is no longer only flushing maps. Keep this define for compatibility
986 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
987 */
988 #define xdp_do_flush_map xdp_do_flush
989
990 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act);
991
992 #ifdef CONFIG_INET
993 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
994 struct bpf_prog *prog, struct sk_buff *skb,
995 struct sock *migrating_sk,
996 u32 hash);
997 #else
998 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)999 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1000 struct bpf_prog *prog, struct sk_buff *skb,
1001 struct sock *migrating_sk,
1002 u32 hash)
1003 {
1004 return NULL;
1005 }
1006 #endif
1007
1008 #ifdef CONFIG_BPF_JIT
1009 extern int bpf_jit_enable;
1010 extern int bpf_jit_harden;
1011 extern int bpf_jit_kallsyms;
1012 extern long bpf_jit_limit;
1013 extern long bpf_jit_limit_max;
1014
1015 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1016
1017 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1018
1019 struct bpf_binary_header *
1020 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1021 unsigned int alignment,
1022 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1023 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1024 u64 bpf_jit_alloc_exec_limit(void);
1025 void *bpf_jit_alloc_exec(unsigned long size);
1026 void bpf_jit_free_exec(void *addr);
1027 void bpf_jit_free(struct bpf_prog *fp);
1028 struct bpf_binary_header *
1029 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1030
1031 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1032 void bpf_prog_pack_free(struct bpf_binary_header *hdr);
1033
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1034 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1035 {
1036 return list_empty(&fp->aux->ksym.lnode) ||
1037 fp->aux->ksym.lnode.prev == LIST_POISON2;
1038 }
1039
1040 struct bpf_binary_header *
1041 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1042 unsigned int alignment,
1043 struct bpf_binary_header **rw_hdr,
1044 u8 **rw_image,
1045 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1046 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1047 struct bpf_binary_header *ro_header,
1048 struct bpf_binary_header *rw_header);
1049 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1050 struct bpf_binary_header *rw_header);
1051
1052 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1053 struct bpf_jit_poke_descriptor *poke);
1054
1055 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1056 const struct bpf_insn *insn, bool extra_pass,
1057 u64 *func_addr, bool *func_addr_fixed);
1058
1059 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1060 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1061
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1062 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1063 u32 pass, void *image)
1064 {
1065 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1066 proglen, pass, image, current->comm, task_pid_nr(current));
1067
1068 if (image)
1069 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1070 16, 1, image, proglen, false);
1071 }
1072
bpf_jit_is_ebpf(void)1073 static inline bool bpf_jit_is_ebpf(void)
1074 {
1075 # ifdef CONFIG_HAVE_EBPF_JIT
1076 return true;
1077 # else
1078 return false;
1079 # endif
1080 }
1081
ebpf_jit_enabled(void)1082 static inline bool ebpf_jit_enabled(void)
1083 {
1084 return bpf_jit_enable && bpf_jit_is_ebpf();
1085 }
1086
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1087 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1088 {
1089 return fp->jited && bpf_jit_is_ebpf();
1090 }
1091
bpf_jit_blinding_enabled(struct bpf_prog * prog)1092 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1093 {
1094 /* These are the prerequisites, should someone ever have the
1095 * idea to call blinding outside of them, we make sure to
1096 * bail out.
1097 */
1098 if (!bpf_jit_is_ebpf())
1099 return false;
1100 if (!prog->jit_requested)
1101 return false;
1102 if (!bpf_jit_harden)
1103 return false;
1104 if (bpf_jit_harden == 1 && bpf_capable())
1105 return false;
1106
1107 return true;
1108 }
1109
bpf_jit_kallsyms_enabled(void)1110 static inline bool bpf_jit_kallsyms_enabled(void)
1111 {
1112 /* There are a couple of corner cases where kallsyms should
1113 * not be enabled f.e. on hardening.
1114 */
1115 if (bpf_jit_harden)
1116 return false;
1117 if (!bpf_jit_kallsyms)
1118 return false;
1119 if (bpf_jit_kallsyms == 1)
1120 return true;
1121
1122 return false;
1123 }
1124
1125 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1126 unsigned long *off, char *sym);
1127 bool is_bpf_text_address(unsigned long addr);
1128 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1129 char *sym);
1130
1131 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1132 bpf_address_lookup(unsigned long addr, unsigned long *size,
1133 unsigned long *off, char **modname, char *sym)
1134 {
1135 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1136
1137 if (ret && modname)
1138 *modname = NULL;
1139 return ret;
1140 }
1141
1142 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1143 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1144
1145 #else /* CONFIG_BPF_JIT */
1146
ebpf_jit_enabled(void)1147 static inline bool ebpf_jit_enabled(void)
1148 {
1149 return false;
1150 }
1151
bpf_jit_blinding_enabled(struct bpf_prog * prog)1152 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1153 {
1154 return false;
1155 }
1156
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1157 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1158 {
1159 return false;
1160 }
1161
1162 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1163 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1164 struct bpf_jit_poke_descriptor *poke)
1165 {
1166 return -ENOTSUPP;
1167 }
1168
bpf_jit_free(struct bpf_prog * fp)1169 static inline void bpf_jit_free(struct bpf_prog *fp)
1170 {
1171 bpf_prog_unlock_free(fp);
1172 }
1173
bpf_jit_kallsyms_enabled(void)1174 static inline bool bpf_jit_kallsyms_enabled(void)
1175 {
1176 return false;
1177 }
1178
1179 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1180 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1181 unsigned long *off, char *sym)
1182 {
1183 return NULL;
1184 }
1185
is_bpf_text_address(unsigned long addr)1186 static inline bool is_bpf_text_address(unsigned long addr)
1187 {
1188 return false;
1189 }
1190
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1191 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1192 char *type, char *sym)
1193 {
1194 return -ERANGE;
1195 }
1196
1197 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1198 bpf_address_lookup(unsigned long addr, unsigned long *size,
1199 unsigned long *off, char **modname, char *sym)
1200 {
1201 return NULL;
1202 }
1203
bpf_prog_kallsyms_add(struct bpf_prog * fp)1204 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1205 {
1206 }
1207
bpf_prog_kallsyms_del(struct bpf_prog * fp)1208 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1209 {
1210 }
1211
1212 #endif /* CONFIG_BPF_JIT */
1213
1214 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1215
1216 #define BPF_ANC BIT(15)
1217
bpf_needs_clear_a(const struct sock_filter * first)1218 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1219 {
1220 switch (first->code) {
1221 case BPF_RET | BPF_K:
1222 case BPF_LD | BPF_W | BPF_LEN:
1223 return false;
1224
1225 case BPF_LD | BPF_W | BPF_ABS:
1226 case BPF_LD | BPF_H | BPF_ABS:
1227 case BPF_LD | BPF_B | BPF_ABS:
1228 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1229 return true;
1230 return false;
1231
1232 default:
1233 return true;
1234 }
1235 }
1236
bpf_anc_helper(const struct sock_filter * ftest)1237 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1238 {
1239 BUG_ON(ftest->code & BPF_ANC);
1240
1241 switch (ftest->code) {
1242 case BPF_LD | BPF_W | BPF_ABS:
1243 case BPF_LD | BPF_H | BPF_ABS:
1244 case BPF_LD | BPF_B | BPF_ABS:
1245 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1246 return BPF_ANC | SKF_AD_##CODE
1247 switch (ftest->k) {
1248 BPF_ANCILLARY(PROTOCOL);
1249 BPF_ANCILLARY(PKTTYPE);
1250 BPF_ANCILLARY(IFINDEX);
1251 BPF_ANCILLARY(NLATTR);
1252 BPF_ANCILLARY(NLATTR_NEST);
1253 BPF_ANCILLARY(MARK);
1254 BPF_ANCILLARY(QUEUE);
1255 BPF_ANCILLARY(HATYPE);
1256 BPF_ANCILLARY(RXHASH);
1257 BPF_ANCILLARY(CPU);
1258 BPF_ANCILLARY(ALU_XOR_X);
1259 BPF_ANCILLARY(VLAN_TAG);
1260 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1261 BPF_ANCILLARY(PAY_OFFSET);
1262 BPF_ANCILLARY(RANDOM);
1263 BPF_ANCILLARY(VLAN_TPID);
1264 }
1265 fallthrough;
1266 default:
1267 return ftest->code;
1268 }
1269 }
1270
1271 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1272 int k, unsigned int size);
1273
bpf_tell_extensions(void)1274 static inline int bpf_tell_extensions(void)
1275 {
1276 return SKF_AD_MAX;
1277 }
1278
1279 struct bpf_sock_addr_kern {
1280 struct sock *sk;
1281 struct sockaddr *uaddr;
1282 /* Temporary "register" to make indirect stores to nested structures
1283 * defined above. We need three registers to make such a store, but
1284 * only two (src and dst) are available at convert_ctx_access time
1285 */
1286 u64 tmp_reg;
1287 void *t_ctx; /* Attach type specific context. */
1288 u32 uaddrlen;
1289 };
1290
1291 struct bpf_sock_ops_kern {
1292 struct sock *sk;
1293 union {
1294 u32 args[4];
1295 u32 reply;
1296 u32 replylong[4];
1297 };
1298 struct sk_buff *syn_skb;
1299 struct sk_buff *skb;
1300 void *skb_data_end;
1301 u8 op;
1302 u8 is_fullsock;
1303 u8 remaining_opt_len;
1304 u64 temp; /* temp and everything after is not
1305 * initialized to 0 before calling
1306 * the BPF program. New fields that
1307 * should be initialized to 0 should
1308 * be inserted before temp.
1309 * temp is scratch storage used by
1310 * sock_ops_convert_ctx_access
1311 * as temporary storage of a register.
1312 */
1313 };
1314
1315 struct bpf_sysctl_kern {
1316 struct ctl_table_header *head;
1317 struct ctl_table *table;
1318 void *cur_val;
1319 size_t cur_len;
1320 void *new_val;
1321 size_t new_len;
1322 int new_updated;
1323 int write;
1324 loff_t *ppos;
1325 /* Temporary "register" for indirect stores to ppos. */
1326 u64 tmp_reg;
1327 };
1328
1329 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1330 struct bpf_sockopt_buf {
1331 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1332 };
1333
1334 struct bpf_sockopt_kern {
1335 struct sock *sk;
1336 u8 *optval;
1337 u8 *optval_end;
1338 s32 level;
1339 s32 optname;
1340 s32 optlen;
1341 /* for retval in struct bpf_cg_run_ctx */
1342 struct task_struct *current_task;
1343 /* Temporary "register" for indirect stores to ppos. */
1344 u64 tmp_reg;
1345 };
1346
1347 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1348
1349 struct bpf_sk_lookup_kern {
1350 u16 family;
1351 u16 protocol;
1352 __be16 sport;
1353 u16 dport;
1354 struct {
1355 __be32 saddr;
1356 __be32 daddr;
1357 } v4;
1358 struct {
1359 const struct in6_addr *saddr;
1360 const struct in6_addr *daddr;
1361 } v6;
1362 struct sock *selected_sk;
1363 u32 ingress_ifindex;
1364 bool no_reuseport;
1365 };
1366
1367 extern struct static_key_false bpf_sk_lookup_enabled;
1368
1369 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1370 *
1371 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1372 * SK_DROP. Their meaning is as follows:
1373 *
1374 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1375 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1376 * SK_DROP : terminate lookup with -ECONNREFUSED
1377 *
1378 * This macro aggregates return values and selected sockets from
1379 * multiple BPF programs according to following rules in order:
1380 *
1381 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1382 * macro result is SK_PASS and last ctx.selected_sk is used.
1383 * 2. If any program returned SK_DROP return value,
1384 * macro result is SK_DROP.
1385 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1386 *
1387 * Caller must ensure that the prog array is non-NULL, and that the
1388 * array as well as the programs it contains remain valid.
1389 */
1390 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1391 ({ \
1392 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1393 struct bpf_prog_array_item *_item; \
1394 struct sock *_selected_sk = NULL; \
1395 bool _no_reuseport = false; \
1396 struct bpf_prog *_prog; \
1397 bool _all_pass = true; \
1398 u32 _ret; \
1399 \
1400 migrate_disable(); \
1401 _item = &(array)->items[0]; \
1402 while ((_prog = READ_ONCE(_item->prog))) { \
1403 /* restore most recent selection */ \
1404 _ctx->selected_sk = _selected_sk; \
1405 _ctx->no_reuseport = _no_reuseport; \
1406 \
1407 _ret = func(_prog, _ctx); \
1408 if (_ret == SK_PASS && _ctx->selected_sk) { \
1409 /* remember last non-NULL socket */ \
1410 _selected_sk = _ctx->selected_sk; \
1411 _no_reuseport = _ctx->no_reuseport; \
1412 } else if (_ret == SK_DROP && _all_pass) { \
1413 _all_pass = false; \
1414 } \
1415 _item++; \
1416 } \
1417 _ctx->selected_sk = _selected_sk; \
1418 _ctx->no_reuseport = _no_reuseport; \
1419 migrate_enable(); \
1420 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1421 })
1422
bpf_sk_lookup_run_v4(struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1423 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1424 const __be32 saddr, const __be16 sport,
1425 const __be32 daddr, const u16 dport,
1426 const int ifindex, struct sock **psk)
1427 {
1428 struct bpf_prog_array *run_array;
1429 struct sock *selected_sk = NULL;
1430 bool no_reuseport = false;
1431
1432 rcu_read_lock();
1433 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1434 if (run_array) {
1435 struct bpf_sk_lookup_kern ctx = {
1436 .family = AF_INET,
1437 .protocol = protocol,
1438 .v4.saddr = saddr,
1439 .v4.daddr = daddr,
1440 .sport = sport,
1441 .dport = dport,
1442 .ingress_ifindex = ifindex,
1443 };
1444 u32 act;
1445
1446 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1447 if (act == SK_PASS) {
1448 selected_sk = ctx.selected_sk;
1449 no_reuseport = ctx.no_reuseport;
1450 } else {
1451 selected_sk = ERR_PTR(-ECONNREFUSED);
1452 }
1453 }
1454 rcu_read_unlock();
1455 *psk = selected_sk;
1456 return no_reuseport;
1457 }
1458
1459 #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,const int ifindex,struct sock ** psk)1460 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1461 const struct in6_addr *saddr,
1462 const __be16 sport,
1463 const struct in6_addr *daddr,
1464 const u16 dport,
1465 const int ifindex, struct sock **psk)
1466 {
1467 struct bpf_prog_array *run_array;
1468 struct sock *selected_sk = NULL;
1469 bool no_reuseport = false;
1470
1471 rcu_read_lock();
1472 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1473 if (run_array) {
1474 struct bpf_sk_lookup_kern ctx = {
1475 .family = AF_INET6,
1476 .protocol = protocol,
1477 .v6.saddr = saddr,
1478 .v6.daddr = daddr,
1479 .sport = sport,
1480 .dport = dport,
1481 .ingress_ifindex = ifindex,
1482 };
1483 u32 act;
1484
1485 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1486 if (act == SK_PASS) {
1487 selected_sk = ctx.selected_sk;
1488 no_reuseport = ctx.no_reuseport;
1489 } else {
1490 selected_sk = ERR_PTR(-ECONNREFUSED);
1491 }
1492 }
1493 rcu_read_unlock();
1494 *psk = selected_sk;
1495 return no_reuseport;
1496 }
1497 #endif /* IS_ENABLED(CONFIG_IPV6) */
1498
__bpf_xdp_redirect_map(struct bpf_map * map,u64 index,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1499 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1500 u64 flags, const u64 flag_mask,
1501 void *lookup_elem(struct bpf_map *map, u32 key))
1502 {
1503 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1504 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1505
1506 /* Lower bits of the flags are used as return code on lookup failure */
1507 if (unlikely(flags & ~(action_mask | flag_mask)))
1508 return XDP_ABORTED;
1509
1510 ri->tgt_value = lookup_elem(map, index);
1511 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1512 /* If the lookup fails we want to clear out the state in the
1513 * redirect_info struct completely, so that if an eBPF program
1514 * performs multiple lookups, the last one always takes
1515 * precedence.
1516 */
1517 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1518 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1519 return flags & action_mask;
1520 }
1521
1522 ri->tgt_index = index;
1523 ri->map_id = map->id;
1524 ri->map_type = map->map_type;
1525
1526 if (flags & BPF_F_BROADCAST) {
1527 WRITE_ONCE(ri->map, map);
1528 ri->flags = flags;
1529 } else {
1530 WRITE_ONCE(ri->map, NULL);
1531 ri->flags = 0;
1532 }
1533
1534 return XDP_REDIRECT;
1535 }
1536
1537 #ifdef CONFIG_NET
1538 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1539 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1540 u32 len, u64 flags);
1541 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1542 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1543 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1544 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1545 void *buf, unsigned long len, bool flush);
1546 #else /* CONFIG_NET */
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1547 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1548 void *to, u32 len)
1549 {
1550 return -EOPNOTSUPP;
1551 }
1552
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1553 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1554 const void *from, u32 len, u64 flags)
1555 {
1556 return -EOPNOTSUPP;
1557 }
1558
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1559 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1560 void *buf, u32 len)
1561 {
1562 return -EOPNOTSUPP;
1563 }
1564
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1565 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1566 void *buf, u32 len)
1567 {
1568 return -EOPNOTSUPP;
1569 }
1570
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1571 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1572 {
1573 return NULL;
1574 }
1575
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1576 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1577 unsigned long len, bool flush)
1578 {
1579 }
1580 #endif /* CONFIG_NET */
1581
1582 #endif /* __LINUX_FILTER_H__ */
1583