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 special load instruction. Same as BPF_MSH */
76 #define BPF_PROBE_MEM32 0xa0
77
78 /* unused opcode to mark special atomic instruction */
79 #define BPF_PROBE_ATOMIC 0xe0
80
81 /* unused opcode to mark call to interpreter with arguments */
82 #define BPF_CALL_ARGS 0xe0
83
84 /* unused opcode to mark speculation barrier for mitigating
85 * Speculative Store Bypass
86 */
87 #define BPF_NOSPEC 0xc0
88
89 /* As per nm, we expose JITed images as text (code) section for
90 * kallsyms. That way, tools like perf can find it to match
91 * addresses.
92 */
93 #define BPF_SYM_ELF_TYPE 't'
94
95 /* BPF program can access up to 512 bytes of stack space. */
96 #define MAX_BPF_STACK 512
97
98 /* Helper macros for filter block array initializers. */
99
100 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
101
102 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
103 ((struct bpf_insn) { \
104 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
105 .dst_reg = DST, \
106 .src_reg = SRC, \
107 .off = OFF, \
108 .imm = 0 })
109
110 #define BPF_ALU64_REG(OP, DST, SRC) \
111 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
112
113 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
114 ((struct bpf_insn) { \
115 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
116 .dst_reg = DST, \
117 .src_reg = SRC, \
118 .off = OFF, \
119 .imm = 0 })
120
121 #define BPF_ALU32_REG(OP, DST, SRC) \
122 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
123
124 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
125
126 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \
127 ((struct bpf_insn) { \
128 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
129 .dst_reg = DST, \
130 .src_reg = 0, \
131 .off = OFF, \
132 .imm = IMM })
133 #define BPF_ALU64_IMM(OP, DST, IMM) \
134 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0)
135
136 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \
137 ((struct bpf_insn) { \
138 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
139 .dst_reg = DST, \
140 .src_reg = 0, \
141 .off = OFF, \
142 .imm = IMM })
143 #define BPF_ALU32_IMM(OP, DST, IMM) \
144 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0)
145
146 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
147
148 #define BPF_ENDIAN(TYPE, DST, LEN) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
151 .dst_reg = DST, \
152 .src_reg = 0, \
153 .off = 0, \
154 .imm = LEN })
155
156 /* Byte Swap, bswap16/32/64 */
157
158 #define BPF_BSWAP(DST, LEN) \
159 ((struct bpf_insn) { \
160 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \
161 .dst_reg = DST, \
162 .src_reg = 0, \
163 .off = 0, \
164 .imm = LEN })
165
166 /* Short form of mov, dst_reg = src_reg */
167
168 #define BPF_MOV64_REG(DST, SRC) \
169 ((struct bpf_insn) { \
170 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
171 .dst_reg = DST, \
172 .src_reg = SRC, \
173 .off = 0, \
174 .imm = 0 })
175
176 #define BPF_MOV32_REG(DST, SRC) \
177 ((struct bpf_insn) { \
178 .code = BPF_ALU | BPF_MOV | BPF_X, \
179 .dst_reg = DST, \
180 .src_reg = SRC, \
181 .off = 0, \
182 .imm = 0 })
183
184 /* Special (internal-only) form of mov, used to resolve per-CPU addrs:
185 * dst_reg = src_reg + <percpu_base_off>
186 * BPF_ADDR_PERCPU is used as a special insn->off value.
187 */
188 #define BPF_ADDR_PERCPU (-1)
189
190 #define BPF_MOV64_PERCPU_REG(DST, SRC) \
191 ((struct bpf_insn) { \
192 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
193 .dst_reg = DST, \
194 .src_reg = SRC, \
195 .off = BPF_ADDR_PERCPU, \
196 .imm = 0 })
197
198 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn)
199 {
200 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU;
201 }
202
203 /* Short form of mov, dst_reg = imm32 */
204
205 #define BPF_MOV64_IMM(DST, IMM) \
206 ((struct bpf_insn) { \
207 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
208 .dst_reg = DST, \
209 .src_reg = 0, \
210 .off = 0, \
211 .imm = IMM })
212
213 #define BPF_MOV32_IMM(DST, IMM) \
214 ((struct bpf_insn) { \
215 .code = BPF_ALU | BPF_MOV | BPF_K, \
216 .dst_reg = DST, \
217 .src_reg = 0, \
218 .off = 0, \
219 .imm = IMM })
220
221 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */
222
223 #define BPF_MOVSX64_REG(DST, SRC, OFF) \
224 ((struct bpf_insn) { \
225 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
226 .dst_reg = DST, \
227 .src_reg = SRC, \
228 .off = OFF, \
229 .imm = 0 })
230
231 #define BPF_MOVSX32_REG(DST, SRC, OFF) \
232 ((struct bpf_insn) { \
233 .code = BPF_ALU | BPF_MOV | BPF_X, \
234 .dst_reg = DST, \
235 .src_reg = SRC, \
236 .off = OFF, \
237 .imm = 0 })
238
239 /* Special form of mov32, used for doing explicit zero extension on dst. */
240 #define BPF_ZEXT_REG(DST) \
241 ((struct bpf_insn) { \
242 .code = BPF_ALU | BPF_MOV | BPF_X, \
243 .dst_reg = DST, \
244 .src_reg = DST, \
245 .off = 0, \
246 .imm = 1 })
247
248 static inline bool insn_is_zext(const struct bpf_insn *insn)
249 {
250 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
251 }
252
253 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers
254 * to pointers in user vma.
255 */
256 static inline bool insn_is_cast_user(const struct bpf_insn *insn)
257 {
258 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) &&
259 insn->off == BPF_ADDR_SPACE_CAST &&
260 insn->imm == 1U << 16;
261 }
262
263 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
264 #define BPF_LD_IMM64(DST, IMM) \
265 BPF_LD_IMM64_RAW(DST, 0, IMM)
266
267 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
268 ((struct bpf_insn) { \
269 .code = BPF_LD | BPF_DW | BPF_IMM, \
270 .dst_reg = DST, \
271 .src_reg = SRC, \
272 .off = 0, \
273 .imm = (__u32) (IMM) }), \
274 ((struct bpf_insn) { \
275 .code = 0, /* zero is reserved opcode */ \
276 .dst_reg = 0, \
277 .src_reg = 0, \
278 .off = 0, \
279 .imm = ((__u64) (IMM)) >> 32 })
280
281 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
282 #define BPF_LD_MAP_FD(DST, MAP_FD) \
283 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
284
285 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
286
287 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
288 ((struct bpf_insn) { \
289 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
290 .dst_reg = DST, \
291 .src_reg = SRC, \
292 .off = 0, \
293 .imm = IMM })
294
295 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
296 ((struct bpf_insn) { \
297 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
298 .dst_reg = DST, \
299 .src_reg = SRC, \
300 .off = 0, \
301 .imm = IMM })
302
303 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
304
305 #define BPF_LD_ABS(SIZE, IMM) \
306 ((struct bpf_insn) { \
307 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
308 .dst_reg = 0, \
309 .src_reg = 0, \
310 .off = 0, \
311 .imm = IMM })
312
313 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
314
315 #define BPF_LD_IND(SIZE, SRC, IMM) \
316 ((struct bpf_insn) { \
317 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
318 .dst_reg = 0, \
319 .src_reg = SRC, \
320 .off = 0, \
321 .imm = IMM })
322
323 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
324
325 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
328 .dst_reg = DST, \
329 .src_reg = SRC, \
330 .off = OFF, \
331 .imm = 0 })
332
333 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */
334
335 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \
338 .dst_reg = DST, \
339 .src_reg = SRC, \
340 .off = OFF, \
341 .imm = 0 })
342
343 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
344
345 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
348 .dst_reg = DST, \
349 .src_reg = SRC, \
350 .off = OFF, \
351 .imm = 0 })
352
353
354 /*
355 * Atomic operations:
356 *
357 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
358 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
359 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
360 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
361 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
362 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
363 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
364 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
365 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
366 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
367 */
368
369 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
370 ((struct bpf_insn) { \
371 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
372 .dst_reg = DST, \
373 .src_reg = SRC, \
374 .off = OFF, \
375 .imm = OP })
376
377 /* Legacy alias */
378 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
379
380 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
381
382 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
383 ((struct bpf_insn) { \
384 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
385 .dst_reg = DST, \
386 .src_reg = 0, \
387 .off = OFF, \
388 .imm = IMM })
389
390 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
391
392 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
393 ((struct bpf_insn) { \
394 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
395 .dst_reg = DST, \
396 .src_reg = SRC, \
397 .off = OFF, \
398 .imm = 0 })
399
400 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
401
402 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
403 ((struct bpf_insn) { \
404 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
405 .dst_reg = DST, \
406 .src_reg = 0, \
407 .off = OFF, \
408 .imm = IMM })
409
410 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
411
412 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
413 ((struct bpf_insn) { \
414 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
415 .dst_reg = DST, \
416 .src_reg = SRC, \
417 .off = OFF, \
418 .imm = 0 })
419
420 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
421
422 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
423 ((struct bpf_insn) { \
424 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
425 .dst_reg = DST, \
426 .src_reg = 0, \
427 .off = OFF, \
428 .imm = IMM })
429
430 /* Unconditional jumps, goto pc + off16 */
431
432 #define BPF_JMP_A(OFF) \
433 ((struct bpf_insn) { \
434 .code = BPF_JMP | BPF_JA, \
435 .dst_reg = 0, \
436 .src_reg = 0, \
437 .off = OFF, \
438 .imm = 0 })
439
440 /* Relative call */
441
442 #define BPF_CALL_REL(TGT) \
443 ((struct bpf_insn) { \
444 .code = BPF_JMP | BPF_CALL, \
445 .dst_reg = 0, \
446 .src_reg = BPF_PSEUDO_CALL, \
447 .off = 0, \
448 .imm = TGT })
449
450 /* Convert function address to BPF immediate */
451
452 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
453
454 #define BPF_EMIT_CALL(FUNC) \
455 ((struct bpf_insn) { \
456 .code = BPF_JMP | BPF_CALL, \
457 .dst_reg = 0, \
458 .src_reg = 0, \
459 .off = 0, \
460 .imm = BPF_CALL_IMM(FUNC) })
461
462 /* Raw code statement block */
463
464 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
465 ((struct bpf_insn) { \
466 .code = CODE, \
467 .dst_reg = DST, \
468 .src_reg = SRC, \
469 .off = OFF, \
470 .imm = IMM })
471
472 /* Program exit */
473
474 #define BPF_EXIT_INSN() \
475 ((struct bpf_insn) { \
476 .code = BPF_JMP | BPF_EXIT, \
477 .dst_reg = 0, \
478 .src_reg = 0, \
479 .off = 0, \
480 .imm = 0 })
481
482 /* Speculation barrier */
483
484 #define BPF_ST_NOSPEC() \
485 ((struct bpf_insn) { \
486 .code = BPF_ST | BPF_NOSPEC, \
487 .dst_reg = 0, \
488 .src_reg = 0, \
489 .off = 0, \
490 .imm = 0 })
491
492 /* Internal classic blocks for direct assignment */
493
494 #define __BPF_STMT(CODE, K) \
495 ((struct sock_filter) BPF_STMT(CODE, K))
496
497 #define __BPF_JUMP(CODE, K, JT, JF) \
498 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
499
500 #define bytes_to_bpf_size(bytes) \
501 ({ \
502 int bpf_size = -EINVAL; \
503 \
504 if (bytes == sizeof(u8)) \
505 bpf_size = BPF_B; \
506 else if (bytes == sizeof(u16)) \
507 bpf_size = BPF_H; \
508 else if (bytes == sizeof(u32)) \
509 bpf_size = BPF_W; \
510 else if (bytes == sizeof(u64)) \
511 bpf_size = BPF_DW; \
512 \
513 bpf_size; \
514 })
515
516 #define bpf_size_to_bytes(bpf_size) \
517 ({ \
518 int bytes = -EINVAL; \
519 \
520 if (bpf_size == BPF_B) \
521 bytes = sizeof(u8); \
522 else if (bpf_size == BPF_H) \
523 bytes = sizeof(u16); \
524 else if (bpf_size == BPF_W) \
525 bytes = sizeof(u32); \
526 else if (bpf_size == BPF_DW) \
527 bytes = sizeof(u64); \
528 \
529 bytes; \
530 })
531
532 #define BPF_SIZEOF(type) \
533 ({ \
534 const int __size = bytes_to_bpf_size(sizeof(type)); \
535 BUILD_BUG_ON(__size < 0); \
536 __size; \
537 })
538
539 #define BPF_FIELD_SIZEOF(type, field) \
540 ({ \
541 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
542 BUILD_BUG_ON(__size < 0); \
543 __size; \
544 })
545
546 #define BPF_LDST_BYTES(insn) \
547 ({ \
548 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
549 WARN_ON(__size < 0); \
550 __size; \
551 })
552
553 #define __BPF_MAP_0(m, v, ...) v
554 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
555 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
556 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
557 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
558 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
559
560 #define __BPF_REG_0(...) __BPF_PAD(5)
561 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
562 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
563 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
564 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
565 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
566
567 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
568 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
569
570 #define __BPF_CAST(t, a) \
571 (__force t) \
572 (__force \
573 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
574 (unsigned long)0, (t)0))) a
575 #define __BPF_V void
576 #define __BPF_N
577
578 #define __BPF_DECL_ARGS(t, a) t a
579 #define __BPF_DECL_REGS(t, a) u64 a
580
581 #define __BPF_PAD(n) \
582 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
583 u64, __ur_3, u64, __ur_4, u64, __ur_5)
584
585 #define BPF_CALL_x(x, attr, name, ...) \
586 static __always_inline \
587 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
588 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
589 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
590 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
591 { \
592 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
593 } \
594 static __always_inline \
595 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
596
597 #define __NOATTR
598 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
599 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
600 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
601 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
602 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
603 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
604
605 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
606
607 #define bpf_ctx_range(TYPE, MEMBER) \
608 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
609 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
610 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
611 #if BITS_PER_LONG == 64
612 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
613 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
614 #else
615 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
616 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
617 #endif /* BITS_PER_LONG == 64 */
618
619 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
620 ({ \
621 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
622 *(PTR_SIZE) = (SIZE); \
623 offsetof(TYPE, MEMBER); \
624 })
625
626 /* A struct sock_filter is architecture independent. */
627 struct compat_sock_fprog {
628 u16 len;
629 compat_uptr_t filter; /* struct sock_filter * */
630 };
631
632 struct sock_fprog_kern {
633 u16 len;
634 struct sock_filter *filter;
635 };
636
637 /* Some arches need doubleword alignment for their instructions and/or data */
638 #define BPF_IMAGE_ALIGNMENT 8
639
640 struct bpf_binary_header {
641 u32 size;
642 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
643 };
644
645 struct bpf_prog_stats {
646 u64_stats_t cnt;
647 u64_stats_t nsecs;
648 u64_stats_t misses;
649 struct u64_stats_sync syncp;
650 } __aligned(2 * sizeof(u64));
651
652 struct sk_filter {
653 refcount_t refcnt;
654 struct rcu_head rcu;
655 struct bpf_prog *prog;
656 };
657
658 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
659
660 extern struct mutex nf_conn_btf_access_lock;
661 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
662 const struct bpf_reg_state *reg,
663 int off, int size);
664
665 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
666 const struct bpf_insn *insnsi,
667 unsigned int (*bpf_func)(const void *,
668 const struct bpf_insn *));
669
670 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
671 const void *ctx,
672 bpf_dispatcher_fn dfunc)
673 {
674 u32 ret;
675
676 cant_migrate();
677 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
678 struct bpf_prog_stats *stats;
679 u64 duration, start = sched_clock();
680 unsigned long flags;
681
682 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
683
684 duration = sched_clock() - start;
685 stats = this_cpu_ptr(prog->stats);
686 flags = u64_stats_update_begin_irqsave(&stats->syncp);
687 u64_stats_inc(&stats->cnt);
688 u64_stats_add(&stats->nsecs, duration);
689 u64_stats_update_end_irqrestore(&stats->syncp, flags);
690 } else {
691 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
692 }
693 return ret;
694 }
695
696 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
697 {
698 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
699 }
700
701 /*
702 * Use in preemptible and therefore migratable context to make sure that
703 * the execution of the BPF program runs on one CPU.
704 *
705 * This uses migrate_disable/enable() explicitly to document that the
706 * invocation of a BPF program does not require reentrancy protection
707 * against a BPF program which is invoked from a preempting task.
708 */
709 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
710 const void *ctx)
711 {
712 u32 ret;
713
714 migrate_disable();
715 ret = bpf_prog_run(prog, ctx);
716 migrate_enable();
717 return ret;
718 }
719
720 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
721
722 struct bpf_skb_data_end {
723 struct qdisc_skb_cb qdisc_cb;
724 void *data_meta;
725 void *data_end;
726 };
727
728 struct bpf_nh_params {
729 u32 nh_family;
730 union {
731 u32 ipv4_nh;
732 struct in6_addr ipv6_nh;
733 };
734 };
735
736 /* flags for bpf_redirect_info kern_flags */
737 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
738 #define BPF_RI_F_RI_INIT BIT(1)
739 #define BPF_RI_F_CPU_MAP_INIT BIT(2)
740 #define BPF_RI_F_DEV_MAP_INIT BIT(3)
741 #define BPF_RI_F_XSK_MAP_INIT BIT(4)
742
743 struct bpf_redirect_info {
744 u64 tgt_index;
745 void *tgt_value;
746 struct bpf_map *map;
747 u32 flags;
748 u32 map_id;
749 enum bpf_map_type map_type;
750 struct bpf_nh_params nh;
751 u32 kern_flags;
752 };
753
754 struct bpf_net_context {
755 struct bpf_redirect_info ri;
756 struct list_head cpu_map_flush_list;
757 struct list_head dev_map_flush_list;
758 struct list_head xskmap_map_flush_list;
759 };
760
761 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
762 {
763 struct task_struct *tsk = current;
764
765 if (tsk->bpf_net_context != NULL)
766 return NULL;
767 bpf_net_ctx->ri.kern_flags = 0;
768
769 tsk->bpf_net_context = bpf_net_ctx;
770 return bpf_net_ctx;
771 }
772
773 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
774 {
775 if (bpf_net_ctx)
776 current->bpf_net_context = NULL;
777 }
778
779 static inline struct bpf_net_context *bpf_net_ctx_get(void)
780 {
781 return current->bpf_net_context;
782 }
783
784 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
785 {
786 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
787
788 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
789 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
790 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
791 }
792
793 return &bpf_net_ctx->ri;
794 }
795
796 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
797 {
798 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
799
800 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
801 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list);
802 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
803 }
804
805 return &bpf_net_ctx->cpu_map_flush_list;
806 }
807
808 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
809 {
810 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
811
812 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
813 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list);
814 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
815 }
816
817 return &bpf_net_ctx->dev_map_flush_list;
818 }
819
820 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
821 {
822 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
823
824 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
825 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list);
826 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
827 }
828
829 return &bpf_net_ctx->xskmap_map_flush_list;
830 }
831
832 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
833 struct list_head **lh_dev,
834 struct list_head **lh_xsk)
835 {
836 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
837 u32 kern_flags = bpf_net_ctx->ri.kern_flags;
838 struct list_head *lh;
839
840 *lh_map = *lh_dev = *lh_xsk = NULL;
841
842 if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
843 return;
844
845 lh = &bpf_net_ctx->dev_map_flush_list;
846 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh))
847 *lh_dev = lh;
848
849 lh = &bpf_net_ctx->cpu_map_flush_list;
850 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh))
851 *lh_map = lh;
852
853 lh = &bpf_net_ctx->xskmap_map_flush_list;
854 if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
855 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh))
856 *lh_xsk = lh;
857 }
858
859 /* Compute the linear packet data range [data, data_end) which
860 * will be accessed by various program types (cls_bpf, act_bpf,
861 * lwt, ...). Subsystems allowing direct data access must (!)
862 * ensure that cb[] area can be written to when BPF program is
863 * invoked (otherwise cb[] save/restore is necessary).
864 */
865 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
866 {
867 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
868
869 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
870 cb->data_meta = skb->data - skb_metadata_len(skb);
871 cb->data_end = skb->data + skb_headlen(skb);
872 }
873
874 /* Similar to bpf_compute_data_pointers(), except that save orginal
875 * data in cb->data and cb->meta_data for restore.
876 */
877 static inline void bpf_compute_and_save_data_end(
878 struct sk_buff *skb, void **saved_data_end)
879 {
880 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
881
882 *saved_data_end = cb->data_end;
883 cb->data_end = skb->data + skb_headlen(skb);
884 }
885
886 /* Restore data saved by bpf_compute_and_save_data_end(). */
887 static inline void bpf_restore_data_end(
888 struct sk_buff *skb, void *saved_data_end)
889 {
890 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
891
892 cb->data_end = saved_data_end;
893 }
894
895 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
896 {
897 /* eBPF programs may read/write skb->cb[] area to transfer meta
898 * data between tail calls. Since this also needs to work with
899 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
900 *
901 * In some socket filter cases, the cb unfortunately needs to be
902 * saved/restored so that protocol specific skb->cb[] data won't
903 * be lost. In any case, due to unpriviledged eBPF programs
904 * attached to sockets, we need to clear the bpf_skb_cb() area
905 * to not leak previous contents to user space.
906 */
907 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
908 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
909 sizeof_field(struct qdisc_skb_cb, data));
910
911 return qdisc_skb_cb(skb)->data;
912 }
913
914 /* Must be invoked with migration disabled */
915 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
916 const void *ctx)
917 {
918 const struct sk_buff *skb = ctx;
919 u8 *cb_data = bpf_skb_cb(skb);
920 u8 cb_saved[BPF_SKB_CB_LEN];
921 u32 res;
922
923 if (unlikely(prog->cb_access)) {
924 memcpy(cb_saved, cb_data, sizeof(cb_saved));
925 memset(cb_data, 0, sizeof(cb_saved));
926 }
927
928 res = bpf_prog_run(prog, skb);
929
930 if (unlikely(prog->cb_access))
931 memcpy(cb_data, cb_saved, sizeof(cb_saved));
932
933 return res;
934 }
935
936 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
937 struct sk_buff *skb)
938 {
939 u32 res;
940
941 migrate_disable();
942 res = __bpf_prog_run_save_cb(prog, skb);
943 migrate_enable();
944 return res;
945 }
946
947 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
948 struct sk_buff *skb)
949 {
950 u8 *cb_data = bpf_skb_cb(skb);
951 u32 res;
952
953 if (unlikely(prog->cb_access))
954 memset(cb_data, 0, BPF_SKB_CB_LEN);
955
956 res = bpf_prog_run_pin_on_cpu(prog, skb);
957 return res;
958 }
959
960 DECLARE_BPF_DISPATCHER(xdp)
961
962 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
963
964 u32 xdp_master_redirect(struct xdp_buff *xdp);
965
966 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
967
968 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
969 {
970 return prog->len * sizeof(struct bpf_insn);
971 }
972
973 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
974 {
975 return round_up(bpf_prog_insn_size(prog) +
976 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
977 }
978
979 static inline unsigned int bpf_prog_size(unsigned int proglen)
980 {
981 return max(sizeof(struct bpf_prog),
982 offsetof(struct bpf_prog, insns[proglen]));
983 }
984
985 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
986 {
987 /* When classic BPF programs have been loaded and the arch
988 * does not have a classic BPF JIT (anymore), they have been
989 * converted via bpf_migrate_filter() to eBPF and thus always
990 * have an unspec program type.
991 */
992 return prog->type == BPF_PROG_TYPE_UNSPEC;
993 }
994
995 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
996 {
997 const u32 size_machine = sizeof(unsigned long);
998
999 if (size > size_machine && size % size_machine == 0)
1000 size = size_machine;
1001
1002 return size;
1003 }
1004
1005 static inline bool
1006 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
1007 {
1008 return size <= size_default && (size & (size - 1)) == 0;
1009 }
1010
1011 static inline u8
1012 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
1013 {
1014 u8 access_off = off & (size_default - 1);
1015
1016 #ifdef __LITTLE_ENDIAN
1017 return access_off;
1018 #else
1019 return size_default - (access_off + size);
1020 #endif
1021 }
1022
1023 #define bpf_ctx_wide_access_ok(off, size, type, field) \
1024 (size == sizeof(__u64) && \
1025 off >= offsetof(type, field) && \
1026 off + sizeof(__u64) <= offsetofend(type, field) && \
1027 off % sizeof(__u64) == 0)
1028
1029 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1030
1031 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
1032 {
1033 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1034 if (!fp->jited) {
1035 set_vm_flush_reset_perms(fp);
1036 return set_memory_ro((unsigned long)fp, fp->pages);
1037 }
1038 #endif
1039 return 0;
1040 }
1041
1042 static inline int __must_check
1043 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
1044 {
1045 set_vm_flush_reset_perms(hdr);
1046 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
1047 }
1048
1049 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
1050 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
1051 {
1052 return sk_filter_trim_cap(sk, skb, 1);
1053 }
1054
1055 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
1056 void bpf_prog_free(struct bpf_prog *fp);
1057
1058 bool bpf_opcode_in_insntable(u8 code);
1059
1060 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
1061 const u32 *insn_to_jit_off);
1062 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
1063 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
1064
1065 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
1066 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
1067 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
1068 gfp_t gfp_extra_flags);
1069 void __bpf_prog_free(struct bpf_prog *fp);
1070
1071 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
1072 {
1073 __bpf_prog_free(fp);
1074 }
1075
1076 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
1077 unsigned int flen);
1078
1079 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
1080 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1081 bpf_aux_classic_check_t trans, bool save_orig);
1082 void bpf_prog_destroy(struct bpf_prog *fp);
1083
1084 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1085 int sk_attach_bpf(u32 ufd, struct sock *sk);
1086 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1087 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
1088 void sk_reuseport_prog_free(struct bpf_prog *prog);
1089 int sk_detach_filter(struct sock *sk);
1090 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
1091
1092 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
1093 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
1094
1095 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1096 #define __bpf_call_base_args \
1097 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
1098 (void *)__bpf_call_base)
1099
1100 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
1101 void bpf_jit_compile(struct bpf_prog *prog);
1102 bool bpf_jit_needs_zext(void);
1103 bool bpf_jit_inlines_helper_call(s32 imm);
1104 bool bpf_jit_supports_subprog_tailcalls(void);
1105 bool bpf_jit_supports_percpu_insn(void);
1106 bool bpf_jit_supports_kfunc_call(void);
1107 bool bpf_jit_supports_far_kfunc_call(void);
1108 bool bpf_jit_supports_exceptions(void);
1109 bool bpf_jit_supports_ptr_xchg(void);
1110 bool bpf_jit_supports_arena(void);
1111 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena);
1112 u64 bpf_arch_uaddress_limit(void);
1113 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie);
1114 bool bpf_helper_changes_pkt_data(void *func);
1115
1116 static inline bool bpf_dump_raw_ok(const struct cred *cred)
1117 {
1118 /* Reconstruction of call-sites is dependent on kallsyms,
1119 * thus make dump the same restriction.
1120 */
1121 return kallsyms_show_value(cred);
1122 }
1123
1124 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
1125 const struct bpf_insn *patch, u32 len);
1126 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
1127
1128 static inline bool xdp_return_frame_no_direct(void)
1129 {
1130 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1131
1132 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
1133 }
1134
1135 static inline void xdp_set_return_frame_no_direct(void)
1136 {
1137 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1138
1139 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
1140 }
1141
1142 static inline void xdp_clear_return_frame_no_direct(void)
1143 {
1144 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1145
1146 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
1147 }
1148
1149 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
1150 unsigned int pktlen)
1151 {
1152 unsigned int len;
1153
1154 if (unlikely(!(fwd->flags & IFF_UP)))
1155 return -ENETDOWN;
1156
1157 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1158 if (pktlen > len)
1159 return -EMSGSIZE;
1160
1161 return 0;
1162 }
1163
1164 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1165 * same cpu context. Further for best results no more than a single map
1166 * for the do_redirect/do_flush pair should be used. This limitation is
1167 * because we only track one map and force a flush when the map changes.
1168 * This does not appear to be a real limitation for existing software.
1169 */
1170 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1171 struct xdp_buff *xdp, struct bpf_prog *prog);
1172 int xdp_do_redirect(struct net_device *dev,
1173 struct xdp_buff *xdp,
1174 struct bpf_prog *prog);
1175 int xdp_do_redirect_frame(struct net_device *dev,
1176 struct xdp_buff *xdp,
1177 struct xdp_frame *xdpf,
1178 struct bpf_prog *prog);
1179 void xdp_do_flush(void);
1180
1181 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act);
1182
1183 #ifdef CONFIG_INET
1184 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1185 struct bpf_prog *prog, struct sk_buff *skb,
1186 struct sock *migrating_sk,
1187 u32 hash);
1188 #else
1189 static inline struct sock *
1190 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1191 struct bpf_prog *prog, struct sk_buff *skb,
1192 struct sock *migrating_sk,
1193 u32 hash)
1194 {
1195 return NULL;
1196 }
1197 #endif
1198
1199 #ifdef CONFIG_BPF_JIT
1200 extern int bpf_jit_enable;
1201 extern int bpf_jit_harden;
1202 extern int bpf_jit_kallsyms;
1203 extern long bpf_jit_limit;
1204 extern long bpf_jit_limit_max;
1205
1206 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1207
1208 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1209
1210 struct bpf_binary_header *
1211 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1212 unsigned int alignment,
1213 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1214 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1215 u64 bpf_jit_alloc_exec_limit(void);
1216 void *bpf_jit_alloc_exec(unsigned long size);
1217 void bpf_jit_free_exec(void *addr);
1218 void bpf_jit_free(struct bpf_prog *fp);
1219 struct bpf_binary_header *
1220 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1221
1222 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1223 void bpf_prog_pack_free(void *ptr, u32 size);
1224
1225 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1226 {
1227 return list_empty(&fp->aux->ksym.lnode) ||
1228 fp->aux->ksym.lnode.prev == LIST_POISON2;
1229 }
1230
1231 struct bpf_binary_header *
1232 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1233 unsigned int alignment,
1234 struct bpf_binary_header **rw_hdr,
1235 u8 **rw_image,
1236 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1237 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
1238 struct bpf_binary_header *rw_header);
1239 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1240 struct bpf_binary_header *rw_header);
1241
1242 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1243 struct bpf_jit_poke_descriptor *poke);
1244
1245 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1246 const struct bpf_insn *insn, bool extra_pass,
1247 u64 *func_addr, bool *func_addr_fixed);
1248
1249 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1250 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1251
1252 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1253 u32 pass, void *image)
1254 {
1255 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1256 proglen, pass, image, current->comm, task_pid_nr(current));
1257
1258 if (image)
1259 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1260 16, 1, image, proglen, false);
1261 }
1262
1263 static inline bool bpf_jit_is_ebpf(void)
1264 {
1265 # ifdef CONFIG_HAVE_EBPF_JIT
1266 return true;
1267 # else
1268 return false;
1269 # endif
1270 }
1271
1272 static inline bool ebpf_jit_enabled(void)
1273 {
1274 return bpf_jit_enable && bpf_jit_is_ebpf();
1275 }
1276
1277 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1278 {
1279 return fp->jited && bpf_jit_is_ebpf();
1280 }
1281
1282 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1283 {
1284 /* These are the prerequisites, should someone ever have the
1285 * idea to call blinding outside of them, we make sure to
1286 * bail out.
1287 */
1288 if (!bpf_jit_is_ebpf())
1289 return false;
1290 if (!prog->jit_requested)
1291 return false;
1292 if (!bpf_jit_harden)
1293 return false;
1294 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF))
1295 return false;
1296
1297 return true;
1298 }
1299
1300 static inline bool bpf_jit_kallsyms_enabled(void)
1301 {
1302 /* There are a couple of corner cases where kallsyms should
1303 * not be enabled f.e. on hardening.
1304 */
1305 if (bpf_jit_harden)
1306 return false;
1307 if (!bpf_jit_kallsyms)
1308 return false;
1309 if (bpf_jit_kallsyms == 1)
1310 return true;
1311
1312 return false;
1313 }
1314
1315 int __bpf_address_lookup(unsigned long addr, unsigned long *size,
1316 unsigned long *off, char *sym);
1317 bool is_bpf_text_address(unsigned long addr);
1318 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1319 char *sym);
1320 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
1321
1322 static inline int
1323 bpf_address_lookup(unsigned long addr, unsigned long *size,
1324 unsigned long *off, char **modname, char *sym)
1325 {
1326 int ret = __bpf_address_lookup(addr, size, off, sym);
1327
1328 if (ret && modname)
1329 *modname = NULL;
1330 return ret;
1331 }
1332
1333 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1334 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1335
1336 #else /* CONFIG_BPF_JIT */
1337
1338 static inline bool ebpf_jit_enabled(void)
1339 {
1340 return false;
1341 }
1342
1343 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1344 {
1345 return false;
1346 }
1347
1348 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1349 {
1350 return false;
1351 }
1352
1353 static inline int
1354 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1355 struct bpf_jit_poke_descriptor *poke)
1356 {
1357 return -ENOTSUPP;
1358 }
1359
1360 static inline void bpf_jit_free(struct bpf_prog *fp)
1361 {
1362 bpf_prog_unlock_free(fp);
1363 }
1364
1365 static inline bool bpf_jit_kallsyms_enabled(void)
1366 {
1367 return false;
1368 }
1369
1370 static inline int
1371 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1372 unsigned long *off, char *sym)
1373 {
1374 return 0;
1375 }
1376
1377 static inline bool is_bpf_text_address(unsigned long addr)
1378 {
1379 return false;
1380 }
1381
1382 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1383 char *type, char *sym)
1384 {
1385 return -ERANGE;
1386 }
1387
1388 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
1389 {
1390 return NULL;
1391 }
1392
1393 static inline int
1394 bpf_address_lookup(unsigned long addr, unsigned long *size,
1395 unsigned long *off, char **modname, char *sym)
1396 {
1397 return 0;
1398 }
1399
1400 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1401 {
1402 }
1403
1404 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1405 {
1406 }
1407
1408 #endif /* CONFIG_BPF_JIT */
1409
1410 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1411
1412 #define BPF_ANC BIT(15)
1413
1414 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1415 {
1416 switch (first->code) {
1417 case BPF_RET | BPF_K:
1418 case BPF_LD | BPF_W | BPF_LEN:
1419 return false;
1420
1421 case BPF_LD | BPF_W | BPF_ABS:
1422 case BPF_LD | BPF_H | BPF_ABS:
1423 case BPF_LD | BPF_B | BPF_ABS:
1424 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1425 return true;
1426 return false;
1427
1428 default:
1429 return true;
1430 }
1431 }
1432
1433 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1434 {
1435 BUG_ON(ftest->code & BPF_ANC);
1436
1437 switch (ftest->code) {
1438 case BPF_LD | BPF_W | BPF_ABS:
1439 case BPF_LD | BPF_H | BPF_ABS:
1440 case BPF_LD | BPF_B | BPF_ABS:
1441 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1442 return BPF_ANC | SKF_AD_##CODE
1443 switch (ftest->k) {
1444 BPF_ANCILLARY(PROTOCOL);
1445 BPF_ANCILLARY(PKTTYPE);
1446 BPF_ANCILLARY(IFINDEX);
1447 BPF_ANCILLARY(NLATTR);
1448 BPF_ANCILLARY(NLATTR_NEST);
1449 BPF_ANCILLARY(MARK);
1450 BPF_ANCILLARY(QUEUE);
1451 BPF_ANCILLARY(HATYPE);
1452 BPF_ANCILLARY(RXHASH);
1453 BPF_ANCILLARY(CPU);
1454 BPF_ANCILLARY(ALU_XOR_X);
1455 BPF_ANCILLARY(VLAN_TAG);
1456 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1457 BPF_ANCILLARY(PAY_OFFSET);
1458 BPF_ANCILLARY(RANDOM);
1459 BPF_ANCILLARY(VLAN_TPID);
1460 }
1461 fallthrough;
1462 default:
1463 return ftest->code;
1464 }
1465 }
1466
1467 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1468 int k, unsigned int size);
1469
1470 static inline int bpf_tell_extensions(void)
1471 {
1472 return SKF_AD_MAX;
1473 }
1474
1475 struct bpf_sock_addr_kern {
1476 struct sock *sk;
1477 struct sockaddr *uaddr;
1478 /* Temporary "register" to make indirect stores to nested structures
1479 * defined above. We need three registers to make such a store, but
1480 * only two (src and dst) are available at convert_ctx_access time
1481 */
1482 u64 tmp_reg;
1483 void *t_ctx; /* Attach type specific context. */
1484 u32 uaddrlen;
1485 };
1486
1487 struct bpf_sock_ops_kern {
1488 struct sock *sk;
1489 union {
1490 u32 args[4];
1491 u32 reply;
1492 u32 replylong[4];
1493 };
1494 struct sk_buff *syn_skb;
1495 struct sk_buff *skb;
1496 void *skb_data_end;
1497 u8 op;
1498 u8 is_fullsock;
1499 u8 remaining_opt_len;
1500 u64 temp; /* temp and everything after is not
1501 * initialized to 0 before calling
1502 * the BPF program. New fields that
1503 * should be initialized to 0 should
1504 * be inserted before temp.
1505 * temp is scratch storage used by
1506 * sock_ops_convert_ctx_access
1507 * as temporary storage of a register.
1508 */
1509 };
1510
1511 struct bpf_sysctl_kern {
1512 struct ctl_table_header *head;
1513 const struct ctl_table *table;
1514 void *cur_val;
1515 size_t cur_len;
1516 void *new_val;
1517 size_t new_len;
1518 int new_updated;
1519 int write;
1520 loff_t *ppos;
1521 /* Temporary "register" for indirect stores to ppos. */
1522 u64 tmp_reg;
1523 };
1524
1525 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1526 struct bpf_sockopt_buf {
1527 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1528 };
1529
1530 struct bpf_sockopt_kern {
1531 struct sock *sk;
1532 u8 *optval;
1533 u8 *optval_end;
1534 s32 level;
1535 s32 optname;
1536 s32 optlen;
1537 /* for retval in struct bpf_cg_run_ctx */
1538 struct task_struct *current_task;
1539 /* Temporary "register" for indirect stores to ppos. */
1540 u64 tmp_reg;
1541 };
1542
1543 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1544
1545 struct bpf_sk_lookup_kern {
1546 u16 family;
1547 u16 protocol;
1548 __be16 sport;
1549 u16 dport;
1550 struct {
1551 __be32 saddr;
1552 __be32 daddr;
1553 } v4;
1554 struct {
1555 const struct in6_addr *saddr;
1556 const struct in6_addr *daddr;
1557 } v6;
1558 struct sock *selected_sk;
1559 u32 ingress_ifindex;
1560 bool no_reuseport;
1561 };
1562
1563 extern struct static_key_false bpf_sk_lookup_enabled;
1564
1565 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1566 *
1567 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1568 * SK_DROP. Their meaning is as follows:
1569 *
1570 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1571 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1572 * SK_DROP : terminate lookup with -ECONNREFUSED
1573 *
1574 * This macro aggregates return values and selected sockets from
1575 * multiple BPF programs according to following rules in order:
1576 *
1577 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1578 * macro result is SK_PASS and last ctx.selected_sk is used.
1579 * 2. If any program returned SK_DROP return value,
1580 * macro result is SK_DROP.
1581 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1582 *
1583 * Caller must ensure that the prog array is non-NULL, and that the
1584 * array as well as the programs it contains remain valid.
1585 */
1586 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1587 ({ \
1588 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1589 struct bpf_prog_array_item *_item; \
1590 struct sock *_selected_sk = NULL; \
1591 bool _no_reuseport = false; \
1592 struct bpf_prog *_prog; \
1593 bool _all_pass = true; \
1594 u32 _ret; \
1595 \
1596 migrate_disable(); \
1597 _item = &(array)->items[0]; \
1598 while ((_prog = READ_ONCE(_item->prog))) { \
1599 /* restore most recent selection */ \
1600 _ctx->selected_sk = _selected_sk; \
1601 _ctx->no_reuseport = _no_reuseport; \
1602 \
1603 _ret = func(_prog, _ctx); \
1604 if (_ret == SK_PASS && _ctx->selected_sk) { \
1605 /* remember last non-NULL socket */ \
1606 _selected_sk = _ctx->selected_sk; \
1607 _no_reuseport = _ctx->no_reuseport; \
1608 } else if (_ret == SK_DROP && _all_pass) { \
1609 _all_pass = false; \
1610 } \
1611 _item++; \
1612 } \
1613 _ctx->selected_sk = _selected_sk; \
1614 _ctx->no_reuseport = _no_reuseport; \
1615 migrate_enable(); \
1616 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1617 })
1618
1619 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1620 const __be32 saddr, const __be16 sport,
1621 const __be32 daddr, const u16 dport,
1622 const int ifindex, struct sock **psk)
1623 {
1624 struct bpf_prog_array *run_array;
1625 struct sock *selected_sk = NULL;
1626 bool no_reuseport = false;
1627
1628 rcu_read_lock();
1629 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1630 if (run_array) {
1631 struct bpf_sk_lookup_kern ctx = {
1632 .family = AF_INET,
1633 .protocol = protocol,
1634 .v4.saddr = saddr,
1635 .v4.daddr = daddr,
1636 .sport = sport,
1637 .dport = dport,
1638 .ingress_ifindex = ifindex,
1639 };
1640 u32 act;
1641
1642 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1643 if (act == SK_PASS) {
1644 selected_sk = ctx.selected_sk;
1645 no_reuseport = ctx.no_reuseport;
1646 } else {
1647 selected_sk = ERR_PTR(-ECONNREFUSED);
1648 }
1649 }
1650 rcu_read_unlock();
1651 *psk = selected_sk;
1652 return no_reuseport;
1653 }
1654
1655 #if IS_ENABLED(CONFIG_IPV6)
1656 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1657 const struct in6_addr *saddr,
1658 const __be16 sport,
1659 const struct in6_addr *daddr,
1660 const u16 dport,
1661 const int ifindex, struct sock **psk)
1662 {
1663 struct bpf_prog_array *run_array;
1664 struct sock *selected_sk = NULL;
1665 bool no_reuseport = false;
1666
1667 rcu_read_lock();
1668 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1669 if (run_array) {
1670 struct bpf_sk_lookup_kern ctx = {
1671 .family = AF_INET6,
1672 .protocol = protocol,
1673 .v6.saddr = saddr,
1674 .v6.daddr = daddr,
1675 .sport = sport,
1676 .dport = dport,
1677 .ingress_ifindex = ifindex,
1678 };
1679 u32 act;
1680
1681 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1682 if (act == SK_PASS) {
1683 selected_sk = ctx.selected_sk;
1684 no_reuseport = ctx.no_reuseport;
1685 } else {
1686 selected_sk = ERR_PTR(-ECONNREFUSED);
1687 }
1688 }
1689 rcu_read_unlock();
1690 *psk = selected_sk;
1691 return no_reuseport;
1692 }
1693 #endif /* IS_ENABLED(CONFIG_IPV6) */
1694
1695 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1696 u64 flags, const u64 flag_mask,
1697 void *lookup_elem(struct bpf_map *map, u32 key))
1698 {
1699 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1700 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1701
1702 /* Lower bits of the flags are used as return code on lookup failure */
1703 if (unlikely(flags & ~(action_mask | flag_mask)))
1704 return XDP_ABORTED;
1705
1706 ri->tgt_value = lookup_elem(map, index);
1707 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1708 /* If the lookup fails we want to clear out the state in the
1709 * redirect_info struct completely, so that if an eBPF program
1710 * performs multiple lookups, the last one always takes
1711 * precedence.
1712 */
1713 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1714 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1715 return flags & action_mask;
1716 }
1717
1718 ri->tgt_index = index;
1719 ri->map_id = map->id;
1720 ri->map_type = map->map_type;
1721
1722 if (flags & BPF_F_BROADCAST) {
1723 WRITE_ONCE(ri->map, map);
1724 ri->flags = flags;
1725 } else {
1726 WRITE_ONCE(ri->map, NULL);
1727 ri->flags = 0;
1728 }
1729
1730 return XDP_REDIRECT;
1731 }
1732
1733 #ifdef CONFIG_NET
1734 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1735 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1736 u32 len, u64 flags);
1737 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1738 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1739 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1740 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1741 void *buf, unsigned long len, bool flush);
1742 #else /* CONFIG_NET */
1743 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1744 void *to, u32 len)
1745 {
1746 return -EOPNOTSUPP;
1747 }
1748
1749 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1750 const void *from, u32 len, u64 flags)
1751 {
1752 return -EOPNOTSUPP;
1753 }
1754
1755 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1756 void *buf, u32 len)
1757 {
1758 return -EOPNOTSUPP;
1759 }
1760
1761 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1762 void *buf, u32 len)
1763 {
1764 return -EOPNOTSUPP;
1765 }
1766
1767 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1768 {
1769 return NULL;
1770 }
1771
1772 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1773 unsigned long len, bool flush)
1774 {
1775 }
1776 #endif /* CONFIG_NET */
1777
1778 #endif /* __LINUX_FILTER_H__ */
1779
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.