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
32struct sk_buff;
33struct sock;
34struct seccomp_data;
35struct bpf_prog_aux;
36struct xdp_rxq_info;
37struct xdp_buff;
38struct sock_reuseport;
39struct ctl_table;
40struct 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
198static 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
248static 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 */
256static 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 * BPF_LOAD_ACQ dst_reg = smp_load_acquire(src_reg + off16)
368 * BPF_STORE_REL smp_store_release(dst_reg + off16, src_reg)
369 */
370
371#define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
372 ((struct bpf_insn) { \
373 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
374 .dst_reg = DST, \
375 .src_reg = SRC, \
376 .off = OFF, \
377 .imm = OP })
378
379/* Legacy alias */
380#define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
381
382/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
383
384#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
385 ((struct bpf_insn) { \
386 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
387 .dst_reg = DST, \
388 .src_reg = 0, \
389 .off = OFF, \
390 .imm = IMM })
391
392/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
393
394#define BPF_JMP_REG(OP, DST, SRC, OFF) \
395 ((struct bpf_insn) { \
396 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
397 .dst_reg = DST, \
398 .src_reg = SRC, \
399 .off = OFF, \
400 .imm = 0 })
401
402/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
403
404#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
405 ((struct bpf_insn) { \
406 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
407 .dst_reg = DST, \
408 .src_reg = 0, \
409 .off = OFF, \
410 .imm = IMM })
411
412/* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
413
414#define BPF_JMP32_REG(OP, DST, SRC, OFF) \
415 ((struct bpf_insn) { \
416 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
417 .dst_reg = DST, \
418 .src_reg = SRC, \
419 .off = OFF, \
420 .imm = 0 })
421
422/* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
423
424#define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
425 ((struct bpf_insn) { \
426 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
427 .dst_reg = DST, \
428 .src_reg = 0, \
429 .off = OFF, \
430 .imm = IMM })
431
432/* Unconditional jumps, goto pc + off16 */
433
434#define BPF_JMP_A(OFF) \
435 ((struct bpf_insn) { \
436 .code = BPF_JMP | BPF_JA, \
437 .dst_reg = 0, \
438 .src_reg = 0, \
439 .off = OFF, \
440 .imm = 0 })
441
442/* Unconditional jumps, gotol pc + imm32 */
443
444#define BPF_JMP32_A(IMM) \
445 ((struct bpf_insn) { \
446 .code = BPF_JMP32 | BPF_JA, \
447 .dst_reg = 0, \
448 .src_reg = 0, \
449 .off = 0, \
450 .imm = IMM })
451
452/* Relative call */
453
454#define BPF_CALL_REL(TGT) \
455 ((struct bpf_insn) { \
456 .code = BPF_JMP | BPF_CALL, \
457 .dst_reg = 0, \
458 .src_reg = BPF_PSEUDO_CALL, \
459 .off = 0, \
460 .imm = TGT })
461
462/* Convert function address to BPF immediate */
463
464#define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
465
466#define BPF_EMIT_CALL(FUNC) \
467 ((struct bpf_insn) { \
468 .code = BPF_JMP | BPF_CALL, \
469 .dst_reg = 0, \
470 .src_reg = 0, \
471 .off = 0, \
472 .imm = BPF_CALL_IMM(FUNC) })
473
474/* Kfunc call */
475
476#define BPF_CALL_KFUNC(OFF, IMM) \
477 ((struct bpf_insn) { \
478 .code = BPF_JMP | BPF_CALL, \
479 .dst_reg = 0, \
480 .src_reg = BPF_PSEUDO_KFUNC_CALL, \
481 .off = OFF, \
482 .imm = IMM })
483
484/* Raw code statement block */
485
486#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
487 ((struct bpf_insn) { \
488 .code = CODE, \
489 .dst_reg = DST, \
490 .src_reg = SRC, \
491 .off = OFF, \
492 .imm = IMM })
493
494/* Program exit */
495
496#define BPF_EXIT_INSN() \
497 ((struct bpf_insn) { \
498 .code = BPF_JMP | BPF_EXIT, \
499 .dst_reg = 0, \
500 .src_reg = 0, \
501 .off = 0, \
502 .imm = 0 })
503
504/* Speculation barrier */
505
506#define BPF_ST_NOSPEC() \
507 ((struct bpf_insn) { \
508 .code = BPF_ST | BPF_NOSPEC, \
509 .dst_reg = 0, \
510 .src_reg = 0, \
511 .off = 0, \
512 .imm = 0 })
513
514/* Internal classic blocks for direct assignment */
515
516#define __BPF_STMT(CODE, K) \
517 ((struct sock_filter) BPF_STMT(CODE, K))
518
519#define __BPF_JUMP(CODE, K, JT, JF) \
520 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
521
522#define bytes_to_bpf_size(bytes) \
523({ \
524 int bpf_size = -EINVAL; \
525 \
526 if (bytes == sizeof(u8)) \
527 bpf_size = BPF_B; \
528 else if (bytes == sizeof(u16)) \
529 bpf_size = BPF_H; \
530 else if (bytes == sizeof(u32)) \
531 bpf_size = BPF_W; \
532 else if (bytes == sizeof(u64)) \
533 bpf_size = BPF_DW; \
534 \
535 bpf_size; \
536})
537
538#define bpf_size_to_bytes(bpf_size) \
539({ \
540 int bytes = -EINVAL; \
541 \
542 if (bpf_size == BPF_B) \
543 bytes = sizeof(u8); \
544 else if (bpf_size == BPF_H) \
545 bytes = sizeof(u16); \
546 else if (bpf_size == BPF_W) \
547 bytes = sizeof(u32); \
548 else if (bpf_size == BPF_DW) \
549 bytes = sizeof(u64); \
550 \
551 bytes; \
552})
553
554#define BPF_SIZEOF(type) \
555 ({ \
556 const int __size = bytes_to_bpf_size(sizeof(type)); \
557 BUILD_BUG_ON(__size < 0); \
558 __size; \
559 })
560
561#define BPF_FIELD_SIZEOF(type, field) \
562 ({ \
563 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
564 BUILD_BUG_ON(__size < 0); \
565 __size; \
566 })
567
568#define BPF_LDST_BYTES(insn) \
569 ({ \
570 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
571 WARN_ON(__size < 0); \
572 __size; \
573 })
574
575#define __BPF_MAP_0(m, v, ...) v
576#define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
577#define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
578#define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
579#define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
580#define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
581
582#define __BPF_REG_0(...) __BPF_PAD(5)
583#define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
584#define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
585#define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
586#define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
587#define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
588
589#define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
590#define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
591
592#define __BPF_CAST(t, a) \
593 (__force t) \
594 (__force \
595 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
596 (unsigned long)0, (t)0))) a
597#define __BPF_V void
598#define __BPF_N
599
600#define __BPF_DECL_ARGS(t, a) t a
601#define __BPF_DECL_REGS(t, a) u64 a
602
603#define __BPF_PAD(n) \
604 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
605 u64, __ur_3, u64, __ur_4, u64, __ur_5)
606
607#define BPF_CALL_x(x, attr, name, ...) \
608 static __always_inline \
609 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
610 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
611 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
612 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
613 { \
614 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
615 } \
616 static __always_inline \
617 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
618
619#define __NOATTR
620#define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
621#define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
622#define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
623#define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
624#define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
625#define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
626
627#define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
628
629#define bpf_ctx_range(TYPE, MEMBER) \
630 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
631#define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
632 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
633#if BITS_PER_LONG == 64
634# define bpf_ctx_range_ptr(TYPE, MEMBER) \
635 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
636#else
637# define bpf_ctx_range_ptr(TYPE, MEMBER) \
638 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
639#endif /* BITS_PER_LONG == 64 */
640
641#define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
642 ({ \
643 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
644 *(PTR_SIZE) = (SIZE); \
645 offsetof(TYPE, MEMBER); \
646 })
647
648/* A struct sock_filter is architecture independent. */
649struct compat_sock_fprog {
650 u16 len;
651 compat_uptr_t filter; /* struct sock_filter * */
652};
653
654struct sock_fprog_kern {
655 u16 len;
656 struct sock_filter *filter;
657};
658
659/* Some arches need doubleword alignment for their instructions and/or data */
660#define BPF_IMAGE_ALIGNMENT 8
661
662struct bpf_binary_header {
663 u32 size;
664 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
665};
666
667struct bpf_prog_stats {
668 u64_stats_t cnt;
669 u64_stats_t nsecs;
670 u64_stats_t misses;
671 struct u64_stats_sync syncp;
672} __aligned(2 * sizeof(u64));
673
674struct bpf_timed_may_goto {
675 u64 count;
676 u64 timestamp;
677};
678
679struct sk_filter {
680 refcount_t refcnt;
681 struct rcu_head rcu;
682 struct bpf_prog *prog;
683};
684
685DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
686
687extern struct mutex nf_conn_btf_access_lock;
688extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
689 const struct bpf_reg_state *reg,
690 int off, int size);
691
692typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
693 const struct bpf_insn *insnsi,
694 unsigned int (*bpf_func)(const void *,
695 const struct bpf_insn *));
696
697static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
698 const void *ctx,
699 bpf_dispatcher_fn dfunc)
700{
701 u32 ret;
702
703 cant_migrate();
704 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
705 struct bpf_prog_stats *stats;
706 u64 duration, start = sched_clock();
707 unsigned long flags;
708
709 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
710
711 duration = sched_clock() - start;
712 stats = this_cpu_ptr(prog->stats);
713 flags = u64_stats_update_begin_irqsave(syncp: &stats->syncp);
714 u64_stats_inc(p: &stats->cnt);
715 u64_stats_add(p: &stats->nsecs, val: duration);
716 u64_stats_update_end_irqrestore(syncp: &stats->syncp, flags);
717 } else {
718 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
719 }
720 return ret;
721}
722
723static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
724{
725 return __bpf_prog_run(prog, ctx, dfunc: bpf_dispatcher_nop_func);
726}
727
728/*
729 * Use in preemptible and therefore migratable context to make sure that
730 * the execution of the BPF program runs on one CPU.
731 *
732 * This uses migrate_disable/enable() explicitly to document that the
733 * invocation of a BPF program does not require reentrancy protection
734 * against a BPF program which is invoked from a preempting task.
735 */
736static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
737 const void *ctx)
738{
739 u32 ret;
740
741 migrate_disable();
742 ret = bpf_prog_run(prog, ctx);
743 migrate_enable();
744 return ret;
745}
746
747#define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
748
749struct bpf_skb_data_end {
750 struct qdisc_skb_cb qdisc_cb;
751 void *data_meta;
752 void *data_end;
753};
754
755struct bpf_nh_params {
756 u32 nh_family;
757 union {
758 u32 ipv4_nh;
759 struct in6_addr ipv6_nh;
760 };
761};
762
763/* flags for bpf_redirect_info kern_flags */
764#define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
765#define BPF_RI_F_RI_INIT BIT(1)
766#define BPF_RI_F_CPU_MAP_INIT BIT(2)
767#define BPF_RI_F_DEV_MAP_INIT BIT(3)
768#define BPF_RI_F_XSK_MAP_INIT BIT(4)
769
770struct bpf_redirect_info {
771 u64 tgt_index;
772 void *tgt_value;
773 struct bpf_map *map;
774 u32 flags;
775 u32 map_id;
776 enum bpf_map_type map_type;
777 struct bpf_nh_params nh;
778 u32 kern_flags;
779};
780
781struct bpf_net_context {
782 struct bpf_redirect_info ri;
783 struct list_head cpu_map_flush_list;
784 struct list_head dev_map_flush_list;
785 struct list_head xskmap_map_flush_list;
786};
787
788static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
789{
790 struct task_struct *tsk = current;
791
792 if (tsk->bpf_net_context != NULL)
793 return NULL;
794 bpf_net_ctx->ri.kern_flags = 0;
795
796 tsk->bpf_net_context = bpf_net_ctx;
797 return bpf_net_ctx;
798}
799
800static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
801{
802 if (bpf_net_ctx)
803 current->bpf_net_context = NULL;
804}
805
806static inline struct bpf_net_context *bpf_net_ctx_get(void)
807{
808 return current->bpf_net_context;
809}
810
811static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
812{
813 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
814
815 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
816 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
817 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
818 }
819
820 return &bpf_net_ctx->ri;
821}
822
823static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
824{
825 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
826
827 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
828 INIT_LIST_HEAD(list: &bpf_net_ctx->cpu_map_flush_list);
829 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
830 }
831
832 return &bpf_net_ctx->cpu_map_flush_list;
833}
834
835static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
836{
837 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
838
839 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
840 INIT_LIST_HEAD(list: &bpf_net_ctx->dev_map_flush_list);
841 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
842 }
843
844 return &bpf_net_ctx->dev_map_flush_list;
845}
846
847static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
848{
849 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
850
851 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
852 INIT_LIST_HEAD(list: &bpf_net_ctx->xskmap_map_flush_list);
853 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
854 }
855
856 return &bpf_net_ctx->xskmap_map_flush_list;
857}
858
859static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
860 struct list_head **lh_dev,
861 struct list_head **lh_xsk)
862{
863 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
864 u32 kern_flags = bpf_net_ctx->ri.kern_flags;
865 struct list_head *lh;
866
867 *lh_map = *lh_dev = *lh_xsk = NULL;
868
869 if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
870 return;
871
872 lh = &bpf_net_ctx->dev_map_flush_list;
873 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(head: lh))
874 *lh_dev = lh;
875
876 lh = &bpf_net_ctx->cpu_map_flush_list;
877 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(head: lh))
878 *lh_map = lh;
879
880 lh = &bpf_net_ctx->xskmap_map_flush_list;
881 if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
882 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(head: lh))
883 *lh_xsk = lh;
884}
885
886/* Compute the linear packet data range [data, data_end) which
887 * will be accessed by various program types (cls_bpf, act_bpf,
888 * lwt, ...). Subsystems allowing direct data access must (!)
889 * ensure that cb[] area can be written to when BPF program is
890 * invoked (otherwise cb[] save/restore is necessary).
891 */
892static inline void bpf_compute_data_pointers(struct sk_buff *skb)
893{
894 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
895
896 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
897 cb->data_meta = skb->data - skb_metadata_len(skb);
898 cb->data_end = skb->data + skb_headlen(skb);
899}
900
901/* Similar to bpf_compute_data_pointers(), except that save orginal
902 * data in cb->data and cb->meta_data for restore.
903 */
904static inline void bpf_compute_and_save_data_end(
905 struct sk_buff *skb, void **saved_data_end)
906{
907 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
908
909 *saved_data_end = cb->data_end;
910 cb->data_end = skb->data + skb_headlen(skb);
911}
912
913/* Restore data saved by bpf_compute_and_save_data_end(). */
914static inline void bpf_restore_data_end(
915 struct sk_buff *skb, void *saved_data_end)
916{
917 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
918
919 cb->data_end = saved_data_end;
920}
921
922static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
923{
924 /* eBPF programs may read/write skb->cb[] area to transfer meta
925 * data between tail calls. Since this also needs to work with
926 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
927 *
928 * In some socket filter cases, the cb unfortunately needs to be
929 * saved/restored so that protocol specific skb->cb[] data won't
930 * be lost. In any case, due to unpriviledged eBPF programs
931 * attached to sockets, we need to clear the bpf_skb_cb() area
932 * to not leak previous contents to user space.
933 */
934 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
935 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
936 sizeof_field(struct qdisc_skb_cb, data));
937
938 return qdisc_skb_cb(skb)->data;
939}
940
941/* Must be invoked with migration disabled */
942static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
943 const void *ctx)
944{
945 const struct sk_buff *skb = ctx;
946 u8 *cb_data = bpf_skb_cb(skb);
947 u8 cb_saved[BPF_SKB_CB_LEN];
948 u32 res;
949
950 if (unlikely(prog->cb_access)) {
951 memcpy(cb_saved, cb_data, sizeof(cb_saved));
952 memset(cb_data, 0, sizeof(cb_saved));
953 }
954
955 res = bpf_prog_run(prog, ctx: skb);
956
957 if (unlikely(prog->cb_access))
958 memcpy(cb_data, cb_saved, sizeof(cb_saved));
959
960 return res;
961}
962
963static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
964 struct sk_buff *skb)
965{
966 u32 res;
967
968 migrate_disable();
969 res = __bpf_prog_run_save_cb(prog, ctx: skb);
970 migrate_enable();
971 return res;
972}
973
974static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
975 struct sk_buff *skb)
976{
977 u8 *cb_data = bpf_skb_cb(skb);
978 u32 res;
979
980 if (unlikely(prog->cb_access))
981 memset(cb_data, 0, BPF_SKB_CB_LEN);
982
983 res = bpf_prog_run_pin_on_cpu(prog, ctx: skb);
984 return res;
985}
986
987DECLARE_BPF_DISPATCHER(xdp)
988
989DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
990
991u32 xdp_master_redirect(struct xdp_buff *xdp);
992
993void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
994
995static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
996{
997 return prog->len * sizeof(struct bpf_insn);
998}
999
1000static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
1001{
1002 return round_up(bpf_prog_insn_size(prog) +
1003 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
1004}
1005
1006static inline unsigned int bpf_prog_size(unsigned int proglen)
1007{
1008 return max(sizeof(struct bpf_prog),
1009 offsetof(struct bpf_prog, insns[proglen]));
1010}
1011
1012static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
1013{
1014 /* When classic BPF programs have been loaded and the arch
1015 * does not have a classic BPF JIT (anymore), they have been
1016 * converted via bpf_migrate_filter() to eBPF and thus always
1017 * have an unspec program type.
1018 */
1019 return prog->type == BPF_PROG_TYPE_UNSPEC;
1020}
1021
1022static inline u32 bpf_ctx_off_adjust_machine(u32 size)
1023{
1024 const u32 size_machine = sizeof(unsigned long);
1025
1026 if (size > size_machine && size % size_machine == 0)
1027 size = size_machine;
1028
1029 return size;
1030}
1031
1032static inline bool
1033bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
1034{
1035 return size <= size_default && (size & (size - 1)) == 0;
1036}
1037
1038static inline u8
1039bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
1040{
1041 u8 access_off = off & (size_default - 1);
1042
1043#ifdef __LITTLE_ENDIAN
1044 return access_off;
1045#else
1046 return size_default - (access_off + size);
1047#endif
1048}
1049
1050#define bpf_ctx_wide_access_ok(off, size, type, field) \
1051 (size == sizeof(__u64) && \
1052 off >= offsetof(type, field) && \
1053 off + sizeof(__u64) <= offsetofend(type, field) && \
1054 off % sizeof(__u64) == 0)
1055
1056#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1057
1058static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
1059{
1060#ifndef CONFIG_BPF_JIT_ALWAYS_ON
1061 if (!fp->jited) {
1062 set_vm_flush_reset_perms(fp);
1063 return set_memory_ro((unsigned long)fp, fp->pages);
1064 }
1065#endif
1066 return 0;
1067}
1068
1069static inline int __must_check
1070bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
1071{
1072 set_vm_flush_reset_perms(hdr);
1073 return set_memory_rox(addr: (unsigned long)hdr, numpages: hdr->size >> PAGE_SHIFT);
1074}
1075
1076int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
1077static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
1078{
1079 return sk_filter_trim_cap(sk, skb, cap: 1);
1080}
1081
1082struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
1083void bpf_prog_free(struct bpf_prog *fp);
1084
1085bool bpf_opcode_in_insntable(u8 code);
1086
1087void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
1088 const u32 *insn_to_jit_off);
1089int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
1090void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
1091
1092struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
1093struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
1094struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
1095 gfp_t gfp_extra_flags);
1096void __bpf_prog_free(struct bpf_prog *fp);
1097
1098static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
1099{
1100 __bpf_prog_free(fp);
1101}
1102
1103typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
1104 unsigned int flen);
1105
1106int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
1107int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1108 bpf_aux_classic_check_t trans, bool save_orig);
1109void bpf_prog_destroy(struct bpf_prog *fp);
1110
1111int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1112int sk_attach_bpf(u32 ufd, struct sock *sk);
1113int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1114int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
1115void sk_reuseport_prog_free(struct bpf_prog *prog);
1116int sk_detach_filter(struct sock *sk);
1117int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
1118
1119bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
1120void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
1121
1122u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1123#define __bpf_call_base_args \
1124 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
1125 (void *)__bpf_call_base)
1126
1127struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
1128void bpf_jit_compile(struct bpf_prog *prog);
1129bool bpf_jit_needs_zext(void);
1130bool bpf_jit_inlines_helper_call(s32 imm);
1131bool bpf_jit_supports_subprog_tailcalls(void);
1132bool bpf_jit_supports_percpu_insn(void);
1133bool bpf_jit_supports_kfunc_call(void);
1134bool bpf_jit_supports_far_kfunc_call(void);
1135bool bpf_jit_supports_exceptions(void);
1136bool bpf_jit_supports_ptr_xchg(void);
1137bool bpf_jit_supports_arena(void);
1138bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena);
1139bool bpf_jit_supports_private_stack(void);
1140bool bpf_jit_supports_timed_may_goto(void);
1141u64 bpf_arch_uaddress_limit(void);
1142void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie);
1143u64 arch_bpf_timed_may_goto(void);
1144u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *);
1145bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id);
1146
1147static inline bool bpf_dump_raw_ok(const struct cred *cred)
1148{
1149 /* Reconstruction of call-sites is dependent on kallsyms,
1150 * thus make dump the same restriction.
1151 */
1152 return kallsyms_show_value(cred);
1153}
1154
1155struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
1156 const struct bpf_insn *patch, u32 len);
1157int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
1158
1159static inline bool xdp_return_frame_no_direct(void)
1160{
1161 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1162
1163 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
1164}
1165
1166static inline void xdp_set_return_frame_no_direct(void)
1167{
1168 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1169
1170 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
1171}
1172
1173static inline void xdp_clear_return_frame_no_direct(void)
1174{
1175 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1176
1177 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
1178}
1179
1180static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
1181 unsigned int pktlen)
1182{
1183 unsigned int len;
1184
1185 if (unlikely(!(fwd->flags & IFF_UP)))
1186 return -ENETDOWN;
1187
1188 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1189 if (pktlen > len)
1190 return -EMSGSIZE;
1191
1192 return 0;
1193}
1194
1195/* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1196 * same cpu context. Further for best results no more than a single map
1197 * for the do_redirect/do_flush pair should be used. This limitation is
1198 * because we only track one map and force a flush when the map changes.
1199 * This does not appear to be a real limitation for existing software.
1200 */
1201int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1202 struct xdp_buff *xdp, const struct bpf_prog *prog);
1203int xdp_do_redirect(struct net_device *dev,
1204 struct xdp_buff *xdp,
1205 const struct bpf_prog *prog);
1206int xdp_do_redirect_frame(struct net_device *dev,
1207 struct xdp_buff *xdp,
1208 struct xdp_frame *xdpf,
1209 const struct bpf_prog *prog);
1210void xdp_do_flush(void);
1211
1212void bpf_warn_invalid_xdp_action(const struct net_device *dev,
1213 const struct bpf_prog *prog, u32 act);
1214
1215#ifdef CONFIG_INET
1216struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1217 struct bpf_prog *prog, struct sk_buff *skb,
1218 struct sock *migrating_sk,
1219 u32 hash);
1220#else
1221static inline struct sock *
1222bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1223 struct bpf_prog *prog, struct sk_buff *skb,
1224 struct sock *migrating_sk,
1225 u32 hash)
1226{
1227 return NULL;
1228}
1229#endif
1230
1231#ifdef CONFIG_BPF_JIT
1232extern int bpf_jit_enable;
1233extern int bpf_jit_harden;
1234extern int bpf_jit_kallsyms;
1235extern long bpf_jit_limit;
1236extern long bpf_jit_limit_max;
1237
1238typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1239
1240void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1241
1242struct bpf_binary_header *
1243bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1244 unsigned int alignment,
1245 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1246void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1247u64 bpf_jit_alloc_exec_limit(void);
1248void *bpf_jit_alloc_exec(unsigned long size);
1249void bpf_jit_free_exec(void *addr);
1250void bpf_jit_free(struct bpf_prog *fp);
1251struct bpf_binary_header *
1252bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1253
1254void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1255void bpf_prog_pack_free(void *ptr, u32 size);
1256
1257static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1258{
1259 return list_empty(head: &fp->aux->ksym.lnode) ||
1260 fp->aux->ksym.lnode.prev == LIST_POISON2;
1261}
1262
1263struct bpf_binary_header *
1264bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1265 unsigned int alignment,
1266 struct bpf_binary_header **rw_hdr,
1267 u8 **rw_image,
1268 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1269int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
1270 struct bpf_binary_header *rw_header);
1271void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1272 struct bpf_binary_header *rw_header);
1273
1274int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1275 struct bpf_jit_poke_descriptor *poke);
1276
1277int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1278 const struct bpf_insn *insn, bool extra_pass,
1279 u64 *func_addr, bool *func_addr_fixed);
1280
1281struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1282void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1283
1284static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1285 u32 pass, void *image)
1286{
1287 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1288 proglen, pass, image, current->comm, task_pid_nr(current));
1289
1290 if (image)
1291 print_hex_dump(KERN_ERR, prefix_str: "JIT code: ", prefix_type: DUMP_PREFIX_OFFSET,
1292 rowsize: 16, groupsize: 1, buf: image, len: proglen, ascii: false);
1293}
1294
1295static inline bool bpf_jit_is_ebpf(void)
1296{
1297# ifdef CONFIG_HAVE_EBPF_JIT
1298 return true;
1299# else
1300 return false;
1301# endif
1302}
1303
1304static inline bool ebpf_jit_enabled(void)
1305{
1306 return bpf_jit_enable && bpf_jit_is_ebpf();
1307}
1308
1309static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1310{
1311 return fp->jited && bpf_jit_is_ebpf();
1312}
1313
1314static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1315{
1316 /* These are the prerequisites, should someone ever have the
1317 * idea to call blinding outside of them, we make sure to
1318 * bail out.
1319 */
1320 if (!bpf_jit_is_ebpf())
1321 return false;
1322 if (!prog->jit_requested)
1323 return false;
1324 if (!bpf_jit_harden)
1325 return false;
1326 if (bpf_jit_harden == 1 && bpf_token_capable(token: prog->aux->token, CAP_BPF))
1327 return false;
1328
1329 return true;
1330}
1331
1332static inline bool bpf_jit_kallsyms_enabled(void)
1333{
1334 /* There are a couple of corner cases where kallsyms should
1335 * not be enabled f.e. on hardening.
1336 */
1337 if (bpf_jit_harden)
1338 return false;
1339 if (!bpf_jit_kallsyms)
1340 return false;
1341 if (bpf_jit_kallsyms == 1)
1342 return true;
1343
1344 return false;
1345}
1346
1347int __bpf_address_lookup(unsigned long addr, unsigned long *size,
1348 unsigned long *off, char *sym);
1349bool is_bpf_text_address(unsigned long addr);
1350int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1351 char *sym);
1352struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
1353
1354static inline int
1355bpf_address_lookup(unsigned long addr, unsigned long *size,
1356 unsigned long *off, char **modname, char *sym)
1357{
1358 int ret = __bpf_address_lookup(addr, size, off, sym);
1359
1360 if (ret && modname)
1361 *modname = NULL;
1362 return ret;
1363}
1364
1365void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1366void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1367
1368#else /* CONFIG_BPF_JIT */
1369
1370static inline bool ebpf_jit_enabled(void)
1371{
1372 return false;
1373}
1374
1375static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1376{
1377 return false;
1378}
1379
1380static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1381{
1382 return false;
1383}
1384
1385static inline int
1386bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1387 struct bpf_jit_poke_descriptor *poke)
1388{
1389 return -ENOTSUPP;
1390}
1391
1392static inline void bpf_jit_free(struct bpf_prog *fp)
1393{
1394 bpf_prog_unlock_free(fp);
1395}
1396
1397static inline bool bpf_jit_kallsyms_enabled(void)
1398{
1399 return false;
1400}
1401
1402static inline int
1403__bpf_address_lookup(unsigned long addr, unsigned long *size,
1404 unsigned long *off, char *sym)
1405{
1406 return 0;
1407}
1408
1409static inline bool is_bpf_text_address(unsigned long addr)
1410{
1411 return false;
1412}
1413
1414static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1415 char *type, char *sym)
1416{
1417 return -ERANGE;
1418}
1419
1420static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
1421{
1422 return NULL;
1423}
1424
1425static inline int
1426bpf_address_lookup(unsigned long addr, unsigned long *size,
1427 unsigned long *off, char **modname, char *sym)
1428{
1429 return 0;
1430}
1431
1432static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1433{
1434}
1435
1436static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1437{
1438}
1439
1440#endif /* CONFIG_BPF_JIT */
1441
1442void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1443
1444#define BPF_ANC BIT(15)
1445
1446static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1447{
1448 switch (first->code) {
1449 case BPF_RET | BPF_K:
1450 case BPF_LD | BPF_W | BPF_LEN:
1451 return false;
1452
1453 case BPF_LD | BPF_W | BPF_ABS:
1454 case BPF_LD | BPF_H | BPF_ABS:
1455 case BPF_LD | BPF_B | BPF_ABS:
1456 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1457 return true;
1458 return false;
1459
1460 default:
1461 return true;
1462 }
1463}
1464
1465static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1466{
1467 BUG_ON(ftest->code & BPF_ANC);
1468
1469 switch (ftest->code) {
1470 case BPF_LD | BPF_W | BPF_ABS:
1471 case BPF_LD | BPF_H | BPF_ABS:
1472 case BPF_LD | BPF_B | BPF_ABS:
1473#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1474 return BPF_ANC | SKF_AD_##CODE
1475 switch (ftest->k) {
1476 BPF_ANCILLARY(PROTOCOL);
1477 BPF_ANCILLARY(PKTTYPE);
1478 BPF_ANCILLARY(IFINDEX);
1479 BPF_ANCILLARY(NLATTR);
1480 BPF_ANCILLARY(NLATTR_NEST);
1481 BPF_ANCILLARY(MARK);
1482 BPF_ANCILLARY(QUEUE);
1483 BPF_ANCILLARY(HATYPE);
1484 BPF_ANCILLARY(RXHASH);
1485 BPF_ANCILLARY(CPU);
1486 BPF_ANCILLARY(ALU_XOR_X);
1487 BPF_ANCILLARY(VLAN_TAG);
1488 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1489 BPF_ANCILLARY(PAY_OFFSET);
1490 BPF_ANCILLARY(RANDOM);
1491 BPF_ANCILLARY(VLAN_TPID);
1492 }
1493 fallthrough;
1494 default:
1495 return ftest->code;
1496 }
1497}
1498
1499void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1500 int k, unsigned int size);
1501
1502static inline int bpf_tell_extensions(void)
1503{
1504 return SKF_AD_MAX;
1505}
1506
1507struct bpf_sock_addr_kern {
1508 struct sock *sk;
1509 struct sockaddr *uaddr;
1510 /* Temporary "register" to make indirect stores to nested structures
1511 * defined above. We need three registers to make such a store, but
1512 * only two (src and dst) are available at convert_ctx_access time
1513 */
1514 u64 tmp_reg;
1515 void *t_ctx; /* Attach type specific context. */
1516 u32 uaddrlen;
1517};
1518
1519struct bpf_sock_ops_kern {
1520 struct sock *sk;
1521 union {
1522 u32 args[4];
1523 u32 reply;
1524 u32 replylong[4];
1525 };
1526 struct sk_buff *syn_skb;
1527 struct sk_buff *skb;
1528 void *skb_data_end;
1529 u8 op;
1530 u8 is_fullsock;
1531 u8 is_locked_tcp_sock;
1532 u8 remaining_opt_len;
1533 u64 temp; /* temp and everything after is not
1534 * initialized to 0 before calling
1535 * the BPF program. New fields that
1536 * should be initialized to 0 should
1537 * be inserted before temp.
1538 * temp is scratch storage used by
1539 * sock_ops_convert_ctx_access
1540 * as temporary storage of a register.
1541 */
1542};
1543
1544struct bpf_sysctl_kern {
1545 struct ctl_table_header *head;
1546 const struct ctl_table *table;
1547 void *cur_val;
1548 size_t cur_len;
1549 void *new_val;
1550 size_t new_len;
1551 int new_updated;
1552 int write;
1553 loff_t *ppos;
1554 /* Temporary "register" for indirect stores to ppos. */
1555 u64 tmp_reg;
1556};
1557
1558#define BPF_SOCKOPT_KERN_BUF_SIZE 32
1559struct bpf_sockopt_buf {
1560 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1561};
1562
1563struct bpf_sockopt_kern {
1564 struct sock *sk;
1565 u8 *optval;
1566 u8 *optval_end;
1567 s32 level;
1568 s32 optname;
1569 s32 optlen;
1570 /* for retval in struct bpf_cg_run_ctx */
1571 struct task_struct *current_task;
1572 /* Temporary "register" for indirect stores to ppos. */
1573 u64 tmp_reg;
1574};
1575
1576int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1577
1578struct bpf_sk_lookup_kern {
1579 u16 family;
1580 u16 protocol;
1581 __be16 sport;
1582 u16 dport;
1583 struct {
1584 __be32 saddr;
1585 __be32 daddr;
1586 } v4;
1587 struct {
1588 const struct in6_addr *saddr;
1589 const struct in6_addr *daddr;
1590 } v6;
1591 struct sock *selected_sk;
1592 u32 ingress_ifindex;
1593 bool no_reuseport;
1594};
1595
1596extern struct static_key_false bpf_sk_lookup_enabled;
1597
1598/* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1599 *
1600 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1601 * SK_DROP. Their meaning is as follows:
1602 *
1603 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1604 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1605 * SK_DROP : terminate lookup with -ECONNREFUSED
1606 *
1607 * This macro aggregates return values and selected sockets from
1608 * multiple BPF programs according to following rules in order:
1609 *
1610 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1611 * macro result is SK_PASS and last ctx.selected_sk is used.
1612 * 2. If any program returned SK_DROP return value,
1613 * macro result is SK_DROP.
1614 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1615 *
1616 * Caller must ensure that the prog array is non-NULL, and that the
1617 * array as well as the programs it contains remain valid.
1618 */
1619#define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1620 ({ \
1621 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1622 struct bpf_prog_array_item *_item; \
1623 struct sock *_selected_sk = NULL; \
1624 bool _no_reuseport = false; \
1625 struct bpf_prog *_prog; \
1626 bool _all_pass = true; \
1627 u32 _ret; \
1628 \
1629 migrate_disable(); \
1630 _item = &(array)->items[0]; \
1631 while ((_prog = READ_ONCE(_item->prog))) { \
1632 /* restore most recent selection */ \
1633 _ctx->selected_sk = _selected_sk; \
1634 _ctx->no_reuseport = _no_reuseport; \
1635 \
1636 _ret = func(_prog, _ctx); \
1637 if (_ret == SK_PASS && _ctx->selected_sk) { \
1638 /* remember last non-NULL socket */ \
1639 _selected_sk = _ctx->selected_sk; \
1640 _no_reuseport = _ctx->no_reuseport; \
1641 } else if (_ret == SK_DROP && _all_pass) { \
1642 _all_pass = false; \
1643 } \
1644 _item++; \
1645 } \
1646 _ctx->selected_sk = _selected_sk; \
1647 _ctx->no_reuseport = _no_reuseport; \
1648 migrate_enable(); \
1649 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1650 })
1651
1652static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol,
1653 const __be32 saddr, const __be16 sport,
1654 const __be32 daddr, const u16 dport,
1655 const int ifindex, struct sock **psk)
1656{
1657 struct bpf_prog_array *run_array;
1658 struct sock *selected_sk = NULL;
1659 bool no_reuseport = false;
1660
1661 rcu_read_lock();
1662 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1663 if (run_array) {
1664 struct bpf_sk_lookup_kern ctx = {
1665 .family = AF_INET,
1666 .protocol = protocol,
1667 .v4.saddr = saddr,
1668 .v4.daddr = daddr,
1669 .sport = sport,
1670 .dport = dport,
1671 .ingress_ifindex = ifindex,
1672 };
1673 u32 act;
1674
1675 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1676 if (act == SK_PASS) {
1677 selected_sk = ctx.selected_sk;
1678 no_reuseport = ctx.no_reuseport;
1679 } else {
1680 selected_sk = ERR_PTR(error: -ECONNREFUSED);
1681 }
1682 }
1683 rcu_read_unlock();
1684 *psk = selected_sk;
1685 return no_reuseport;
1686}
1687
1688#if IS_ENABLED(CONFIG_IPV6)
1689static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol,
1690 const struct in6_addr *saddr,
1691 const __be16 sport,
1692 const struct in6_addr *daddr,
1693 const u16 dport,
1694 const int ifindex, struct sock **psk)
1695{
1696 struct bpf_prog_array *run_array;
1697 struct sock *selected_sk = NULL;
1698 bool no_reuseport = false;
1699
1700 rcu_read_lock();
1701 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1702 if (run_array) {
1703 struct bpf_sk_lookup_kern ctx = {
1704 .family = AF_INET6,
1705 .protocol = protocol,
1706 .v6.saddr = saddr,
1707 .v6.daddr = daddr,
1708 .sport = sport,
1709 .dport = dport,
1710 .ingress_ifindex = ifindex,
1711 };
1712 u32 act;
1713
1714 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1715 if (act == SK_PASS) {
1716 selected_sk = ctx.selected_sk;
1717 no_reuseport = ctx.no_reuseport;
1718 } else {
1719 selected_sk = ERR_PTR(error: -ECONNREFUSED);
1720 }
1721 }
1722 rcu_read_unlock();
1723 *psk = selected_sk;
1724 return no_reuseport;
1725}
1726#endif /* IS_ENABLED(CONFIG_IPV6) */
1727
1728static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1729 u64 flags, const u64 flag_mask,
1730 void *lookup_elem(struct bpf_map *map, u32 key))
1731{
1732 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1733 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1734
1735 /* Lower bits of the flags are used as return code on lookup failure */
1736 if (unlikely(flags & ~(action_mask | flag_mask)))
1737 return XDP_ABORTED;
1738
1739 ri->tgt_value = lookup_elem(map, index);
1740 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1741 /* If the lookup fails we want to clear out the state in the
1742 * redirect_info struct completely, so that if an eBPF program
1743 * performs multiple lookups, the last one always takes
1744 * precedence.
1745 */
1746 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1747 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1748 return flags & action_mask;
1749 }
1750
1751 ri->tgt_index = index;
1752 ri->map_id = map->id;
1753 ri->map_type = map->map_type;
1754
1755 if (flags & BPF_F_BROADCAST) {
1756 WRITE_ONCE(ri->map, map);
1757 ri->flags = flags;
1758 } else {
1759 WRITE_ONCE(ri->map, NULL);
1760 ri->flags = 0;
1761 }
1762
1763 return XDP_REDIRECT;
1764}
1765
1766#ifdef CONFIG_NET
1767int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1768int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1769 u32 len, u64 flags);
1770int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1771int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1772void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1773void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1774 void *buf, unsigned long len, bool flush);
1775#else /* CONFIG_NET */
1776static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1777 void *to, u32 len)
1778{
1779 return -EOPNOTSUPP;
1780}
1781
1782static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1783 const void *from, u32 len, u64 flags)
1784{
1785 return -EOPNOTSUPP;
1786}
1787
1788static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1789 void *buf, u32 len)
1790{
1791 return -EOPNOTSUPP;
1792}
1793
1794static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1795 void *buf, u32 len)
1796{
1797 return -EOPNOTSUPP;
1798}
1799
1800static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1801{
1802 return NULL;
1803}
1804
1805static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1806 unsigned long len, bool flush)
1807{
1808}
1809#endif /* CONFIG_NET */
1810
1811#endif /* __LINUX_FILTER_H__ */
1812

source code of linux/include/linux/filter.h

Morty Proxy This is a proxified and sanitized view of the page, visit original site.