1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Linux Socket Filter Data Structures
4 */
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
7
8 #include <linux/atomic.h>
9 #include <linux/bpf.h>
10 #include <linux/refcount.h>
11 #include <linux/compat.h>
12 #include <linux/skbuff.h>
13 #include <linux/linkage.h>
14 #include <linux/printk.h>
15 #include <linux/workqueue.h>
16 #include <linux/sched.h>
17 #include <linux/sched/clock.h>
18 #include <linux/capability.h>
19 #include <linux/set_memory.h>
20 #include <linux/kallsyms.h>
21 #include <linux/if_vlan.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sockptr.h>
24 #include <crypto/sha1.h>
25 #include <linux/u64_stats_sync.h>
26
27 #include <net/sch_generic.h>
28
29 #include <asm/byteorder.h>
30 #include <uapi/linux/filter.h>
31
32 struct sk_buff;
33 struct sock;
34 struct seccomp_data;
35 struct bpf_prog_aux;
36 struct xdp_rxq_info;
37 struct xdp_buff;
38 struct sock_reuseport;
39 struct ctl_table;
40 struct ctl_table_header;
41
42 /* ArgX, context and stack frame pointer register positions. Note,
43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
44 * calls in BPF_CALL instruction.
45 */
46 #define BPF_REG_ARG1 BPF_REG_1
47 #define BPF_REG_ARG2 BPF_REG_2
48 #define BPF_REG_ARG3 BPF_REG_3
49 #define BPF_REG_ARG4 BPF_REG_4
50 #define BPF_REG_ARG5 BPF_REG_5
51 #define BPF_REG_CTX BPF_REG_6
52 #define BPF_REG_FP BPF_REG_10
53
54 /* Additional register mappings for converted user programs. */
55 #define BPF_REG_A BPF_REG_0
56 #define BPF_REG_X BPF_REG_7
57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
60
61 /* Kernel hidden auxiliary/helper register. */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
65
66 /* unused opcode to mark special call to bpf_tail_call() helper */
67 #define BPF_TAIL_CALL 0xf0
68
69 /* unused opcode to mark special load instruction. Same as BPF_ABS */
70 #define BPF_PROBE_MEM 0x20
71
72 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
73 #define BPF_PROBE_MEMSX 0x40
74
75 /* unused opcode to mark call to interpreter with arguments */
76 #define BPF_CALL_ARGS 0xe0
77
78 /* unused opcode to mark speculation barrier for mitigating
79 * Speculative Store Bypass
80 */
81 #define BPF_NOSPEC 0xc0
82
83 /* As per nm, we expose JITed images as text (code) section for
84 * kallsyms. That way, tools like perf can find it to match
85 * addresses.
86 */
87 #define BPF_SYM_ELF_TYPE 't'
88
89 /* BPF program can access up to 512 bytes of stack space. */
90 #define MAX_BPF_STACK 512
91
92 /* Helper macros for filter block array initializers. */
93
94 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
95
96 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
97 ((struct bpf_insn) { \
98 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
99 .dst_reg = DST, \
100 .src_reg = SRC, \
101 .off = OFF, \
102 .imm = 0 })
103
104 #define BPF_ALU64_REG(OP, DST, SRC) \
105 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
106
107 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
108 ((struct bpf_insn) { \
109 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
110 .dst_reg = DST, \
111 .src_reg = SRC, \
112 .off = OFF, \
113 .imm = 0 })
114
115 #define BPF_ALU32_REG(OP, DST, SRC) \
116 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
117
118 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
119
120 #define BPF_ALU64_IMM(OP, DST, IMM) \
121 ((struct bpf_insn) { \
122 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
123 .dst_reg = DST, \
124 .src_reg = 0, \
125 .off = 0, \
126 .imm = IMM })
127
128 #define BPF_ALU32_IMM(OP, DST, IMM) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
131 .dst_reg = DST, \
132 .src_reg = 0, \
133 .off = 0, \
134 .imm = IMM })
135
136 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
137
138 #define BPF_ENDIAN(TYPE, DST, LEN) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
141 .dst_reg = DST, \
142 .src_reg = 0, \
143 .off = 0, \
144 .imm = LEN })
145
146 /* Short form of mov, dst_reg = src_reg */
147
148 #define BPF_MOV64_REG(DST, SRC) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
151 .dst_reg = DST, \
152 .src_reg = SRC, \
153 .off = 0, \
154 .imm = 0 })
155
156 #define BPF_MOV32_REG(DST, SRC) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU | BPF_MOV | BPF_X, \
159 .dst_reg = DST, \
160 .src_reg = SRC, \
161 .off = 0, \
162 .imm = 0 })
163
164 /* Short form of mov, dst_reg = imm32 */
165
166 #define BPF_MOV64_IMM(DST, IMM) \
167 ((struct bpf_insn) { \
168 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
169 .dst_reg = DST, \
170 .src_reg = 0, \
171 .off = 0, \
172 .imm = IMM })
173
174 #define BPF_MOV32_IMM(DST, IMM) \
175 ((struct bpf_insn) { \
176 .code = BPF_ALU | BPF_MOV | BPF_K, \
177 .dst_reg = DST, \
178 .src_reg = 0, \
179 .off = 0, \
180 .imm = IMM })
181
182 /* Special form of mov32, used for doing explicit zero extension on dst. */
183 #define BPF_ZEXT_REG(DST) \
184 ((struct bpf_insn) { \
185 .code = BPF_ALU | BPF_MOV | BPF_X, \
186 .dst_reg = DST, \
187 .src_reg = DST, \
188 .off = 0, \
189 .imm = 1 })
190
insn_is_zext(const struct bpf_insn * insn)191 static inline bool insn_is_zext(const struct bpf_insn *insn)
192 {
193 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
194 }
195
196 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
197 #define BPF_LD_IMM64(DST, IMM) \
198 BPF_LD_IMM64_RAW(DST, 0, IMM)
199
200 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
201 ((struct bpf_insn) { \
202 .code = BPF_LD | BPF_DW | BPF_IMM, \
203 .dst_reg = DST, \
204 .src_reg = SRC, \
205 .off = 0, \
206 .imm = (__u32) (IMM) }), \
207 ((struct bpf_insn) { \
208 .code = 0, /* zero is reserved opcode */ \
209 .dst_reg = 0, \
210 .src_reg = 0, \
211 .off = 0, \
212 .imm = ((__u64) (IMM)) >> 32 })
213
214 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
215 #define BPF_LD_MAP_FD(DST, MAP_FD) \
216 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
217
218 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
219
220 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
221 ((struct bpf_insn) { \
222 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
223 .dst_reg = DST, \
224 .src_reg = SRC, \
225 .off = 0, \
226 .imm = IMM })
227
228 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
229 ((struct bpf_insn) { \
230 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
231 .dst_reg = DST, \
232 .src_reg = SRC, \
233 .off = 0, \
234 .imm = IMM })
235
236 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
237
238 #define BPF_LD_ABS(SIZE, IMM) \
239 ((struct bpf_insn) { \
240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
241 .dst_reg = 0, \
242 .src_reg = 0, \
243 .off = 0, \
244 .imm = IMM })
245
246 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
247
248 #define BPF_LD_IND(SIZE, SRC, IMM) \
249 ((struct bpf_insn) { \
250 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
251 .dst_reg = 0, \
252 .src_reg = SRC, \
253 .off = 0, \
254 .imm = IMM })
255
256 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
257
258 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
259 ((struct bpf_insn) { \
260 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
261 .dst_reg = DST, \
262 .src_reg = SRC, \
263 .off = OFF, \
264 .imm = 0 })
265
266 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
267
268 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
269 ((struct bpf_insn) { \
270 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
271 .dst_reg = DST, \
272 .src_reg = SRC, \
273 .off = OFF, \
274 .imm = 0 })
275
276
277 /*
278 * Atomic operations:
279 *
280 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
281 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
282 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
283 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
284 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
285 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
286 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
287 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
288 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
289 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
290 */
291
292 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
293 ((struct bpf_insn) { \
294 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
295 .dst_reg = DST, \
296 .src_reg = SRC, \
297 .off = OFF, \
298 .imm = OP })
299
300 /* Legacy alias */
301 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
302
303 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
304
305 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
306 ((struct bpf_insn) { \
307 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
308 .dst_reg = DST, \
309 .src_reg = 0, \
310 .off = OFF, \
311 .imm = IMM })
312
313 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
314
315 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
316 ((struct bpf_insn) { \
317 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
318 .dst_reg = DST, \
319 .src_reg = SRC, \
320 .off = OFF, \
321 .imm = 0 })
322
323 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
324
325 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
328 .dst_reg = DST, \
329 .src_reg = 0, \
330 .off = OFF, \
331 .imm = IMM })
332
333 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
334
335 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
338 .dst_reg = DST, \
339 .src_reg = SRC, \
340 .off = OFF, \
341 .imm = 0 })
342
343 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
344
345 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
348 .dst_reg = DST, \
349 .src_reg = 0, \
350 .off = OFF, \
351 .imm = IMM })
352
353 /* Unconditional jumps, goto pc + off16 */
354
355 #define BPF_JMP_A(OFF) \
356 ((struct bpf_insn) { \
357 .code = BPF_JMP | BPF_JA, \
358 .dst_reg = 0, \
359 .src_reg = 0, \
360 .off = OFF, \
361 .imm = 0 })
362
363 /* Relative call */
364
365 #define BPF_CALL_REL(TGT) \
366 ((struct bpf_insn) { \
367 .code = BPF_JMP | BPF_CALL, \
368 .dst_reg = 0, \
369 .src_reg = BPF_PSEUDO_CALL, \
370 .off = 0, \
371 .imm = TGT })
372
373 /* Convert function address to BPF immediate */
374
375 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
376
377 #define BPF_EMIT_CALL(FUNC) \
378 ((struct bpf_insn) { \
379 .code = BPF_JMP | BPF_CALL, \
380 .dst_reg = 0, \
381 .src_reg = 0, \
382 .off = 0, \
383 .imm = BPF_CALL_IMM(FUNC) })
384
385 /* Raw code statement block */
386
387 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
388 ((struct bpf_insn) { \
389 .code = CODE, \
390 .dst_reg = DST, \
391 .src_reg = SRC, \
392 .off = OFF, \
393 .imm = IMM })
394
395 /* Program exit */
396
397 #define BPF_EXIT_INSN() \
398 ((struct bpf_insn) { \
399 .code = BPF_JMP | BPF_EXIT, \
400 .dst_reg = 0, \
401 .src_reg = 0, \
402 .off = 0, \
403 .imm = 0 })
404
405 /* Speculation barrier */
406
407 #define BPF_ST_NOSPEC() \
408 ((struct bpf_insn) { \
409 .code = BPF_ST | BPF_NOSPEC, \
410 .dst_reg = 0, \
411 .src_reg = 0, \
412 .off = 0, \
413 .imm = 0 })
414
415 /* Internal classic blocks for direct assignment */
416
417 #define __BPF_STMT(CODE, K) \
418 ((struct sock_filter) BPF_STMT(CODE, K))
419
420 #define __BPF_JUMP(CODE, K, JT, JF) \
421 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
422
423 #define bytes_to_bpf_size(bytes) \
424 ({ \
425 int bpf_size = -EINVAL; \
426 \
427 if (bytes == sizeof(u8)) \
428 bpf_size = BPF_B; \
429 else if (bytes == sizeof(u16)) \
430 bpf_size = BPF_H; \
431 else if (bytes == sizeof(u32)) \
432 bpf_size = BPF_W; \
433 else if (bytes == sizeof(u64)) \
434 bpf_size = BPF_DW; \
435 \
436 bpf_size; \
437 })
438
439 #define bpf_size_to_bytes(bpf_size) \
440 ({ \
441 int bytes = -EINVAL; \
442 \
443 if (bpf_size == BPF_B) \
444 bytes = sizeof(u8); \
445 else if (bpf_size == BPF_H) \
446 bytes = sizeof(u16); \
447 else if (bpf_size == BPF_W) \
448 bytes = sizeof(u32); \
449 else if (bpf_size == BPF_DW) \
450 bytes = sizeof(u64); \
451 \
452 bytes; \
453 })
454
455 #define BPF_SIZEOF(type) \
456 ({ \
457 const int __size = bytes_to_bpf_size(sizeof(type)); \
458 BUILD_BUG_ON(__size < 0); \
459 __size; \
460 })
461
462 #define BPF_FIELD_SIZEOF(type, field) \
463 ({ \
464 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
465 BUILD_BUG_ON(__size < 0); \
466 __size; \
467 })
468
469 #define BPF_LDST_BYTES(insn) \
470 ({ \
471 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
472 WARN_ON(__size < 0); \
473 __size; \
474 })
475
476 #define __BPF_MAP_0(m, v, ...) v
477 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
478 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
479 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
480 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
481 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
482
483 #define __BPF_REG_0(...) __BPF_PAD(5)
484 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
485 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
486 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
487 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
488 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
489
490 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
491 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
492
493 #define __BPF_CAST(t, a) \
494 (__force t) \
495 (__force \
496 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
497 (unsigned long)0, (t)0))) a
498 #define __BPF_V void
499 #define __BPF_N
500
501 #define __BPF_DECL_ARGS(t, a) t a
502 #define __BPF_DECL_REGS(t, a) u64 a
503
504 #define __BPF_PAD(n) \
505 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
506 u64, __ur_3, u64, __ur_4, u64, __ur_5)
507
508 #define BPF_CALL_x(x, attr, name, ...) \
509 static __always_inline \
510 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
511 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
512 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
513 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
514 { \
515 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
516 } \
517 static __always_inline \
518 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
519
520 #define __NOATTR
521 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
522 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
523 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
524 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
525 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
526 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
527
528 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
529
530 #define bpf_ctx_range(TYPE, MEMBER) \
531 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
532 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
533 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
534 #if BITS_PER_LONG == 64
535 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
536 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
537 #else
538 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
539 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
540 #endif /* BITS_PER_LONG == 64 */
541
542 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
543 ({ \
544 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
545 *(PTR_SIZE) = (SIZE); \
546 offsetof(TYPE, MEMBER); \
547 })
548
549 /* A struct sock_filter is architecture independent. */
550 struct compat_sock_fprog {
551 u16 len;
552 compat_uptr_t filter; /* struct sock_filter * */
553 };
554
555 struct sock_fprog_kern {
556 u16 len;
557 struct sock_filter *filter;
558 };
559
560 /* Some arches need doubleword alignment for their instructions and/or data */
561 #define BPF_IMAGE_ALIGNMENT 8
562
563 struct bpf_binary_header {
564 u32 size;
565 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
566 };
567
568 struct bpf_prog_stats {
569 u64_stats_t cnt;
570 u64_stats_t nsecs;
571 u64_stats_t misses;
572 struct u64_stats_sync syncp;
573 } __aligned(2 * sizeof(u64));
574
575 struct sk_filter {
576 refcount_t refcnt;
577 struct rcu_head rcu;
578 struct bpf_prog *prog;
579 };
580
581 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
582
583 extern struct mutex nf_conn_btf_access_lock;
584 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
585 const struct bpf_reg_state *reg,
586 int off, int size);
587
588 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
589 const struct bpf_insn *insnsi,
590 unsigned int (*bpf_func)(const void *,
591 const struct bpf_insn *));
592
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)593 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
594 const void *ctx,
595 bpf_dispatcher_fn dfunc)
596 {
597 u32 ret;
598
599 cant_migrate();
600 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
601 struct bpf_prog_stats *stats;
602 u64 start = sched_clock();
603 unsigned long flags;
604
605 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
606 stats = this_cpu_ptr(prog->stats);
607 flags = u64_stats_update_begin_irqsave(&stats->syncp);
608 u64_stats_inc(&stats->cnt);
609 u64_stats_add(&stats->nsecs, sched_clock() - start);
610 u64_stats_update_end_irqrestore(&stats->syncp, flags);
611 } else {
612 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
613 }
614 return ret;
615 }
616
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)617 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
618 {
619 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
620 }
621
622 /*
623 * Use in preemptible and therefore migratable context to make sure that
624 * the execution of the BPF program runs on one CPU.
625 *
626 * This uses migrate_disable/enable() explicitly to document that the
627 * invocation of a BPF program does not require reentrancy protection
628 * against a BPF program which is invoked from a preempting task.
629 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)630 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
631 const void *ctx)
632 {
633 u32 ret;
634
635 migrate_disable();
636 ret = bpf_prog_run(prog, ctx);
637 migrate_enable();
638 return ret;
639 }
640
641 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
642
643 struct bpf_skb_data_end {
644 struct qdisc_skb_cb qdisc_cb;
645 void *data_meta;
646 void *data_end;
647 };
648
649 struct bpf_nh_params {
650 u32 nh_family;
651 union {
652 u32 ipv4_nh;
653 struct in6_addr ipv6_nh;
654 };
655 };
656
657 struct bpf_redirect_info {
658 u64 tgt_index;
659 void *tgt_value;
660 struct bpf_map *map;
661 u32 flags;
662 u32 kern_flags;
663 u32 map_id;
664 enum bpf_map_type map_type;
665 struct bpf_nh_params nh;
666 };
667
668 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
669
670 /* flags for bpf_redirect_info kern_flags */
671 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
672
673 /* Compute the linear packet data range [data, data_end) which
674 * will be accessed by various program types (cls_bpf, act_bpf,
675 * lwt, ...). Subsystems allowing direct data access must (!)
676 * ensure that cb[] area can be written to when BPF program is
677 * invoked (otherwise cb[] save/restore is necessary).
678 */
bpf_compute_data_pointers(struct sk_buff * skb)679 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
680 {
681 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
682
683 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
684 cb->data_meta = skb->data - skb_metadata_len(skb);
685 cb->data_end = skb->data + skb_headlen(skb);
686 }
687
688 /* Similar to bpf_compute_data_pointers(), except that save orginal
689 * data in cb->data and cb->meta_data for restore.
690 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)691 static inline void bpf_compute_and_save_data_end(
692 struct sk_buff *skb, void **saved_data_end)
693 {
694 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
695
696 *saved_data_end = cb->data_end;
697 cb->data_end = skb->data + skb_headlen(skb);
698 }
699
700 /* Restore data saved by bpf_compute_data_pointers(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)701 static inline void bpf_restore_data_end(
702 struct sk_buff *skb, void *saved_data_end)
703 {
704 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
705
706 cb->data_end = saved_data_end;
707 }
708
bpf_skb_cb(const struct sk_buff * skb)709 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
710 {
711 /* eBPF programs may read/write skb->cb[] area to transfer meta
712 * data between tail calls. Since this also needs to work with
713 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
714 *
715 * In some socket filter cases, the cb unfortunately needs to be
716 * saved/restored so that protocol specific skb->cb[] data won't
717 * be lost. In any case, due to unpriviledged eBPF programs
718 * attached to sockets, we need to clear the bpf_skb_cb() area
719 * to not leak previous contents to user space.
720 */
721 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
722 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
723 sizeof_field(struct qdisc_skb_cb, data));
724
725 return qdisc_skb_cb(skb)->data;
726 }
727
728 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)729 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
730 const void *ctx)
731 {
732 const struct sk_buff *skb = ctx;
733 u8 *cb_data = bpf_skb_cb(skb);
734 u8 cb_saved[BPF_SKB_CB_LEN];
735 u32 res;
736
737 if (unlikely(prog->cb_access)) {
738 memcpy(cb_saved, cb_data, sizeof(cb_saved));
739 memset(cb_data, 0, sizeof(cb_saved));
740 }
741
742 res = bpf_prog_run(prog, skb);
743
744 if (unlikely(prog->cb_access))
745 memcpy(cb_data, cb_saved, sizeof(cb_saved));
746
747 return res;
748 }
749
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)750 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
751 struct sk_buff *skb)
752 {
753 u32 res;
754
755 migrate_disable();
756 res = __bpf_prog_run_save_cb(prog, skb);
757 migrate_enable();
758 return res;
759 }
760
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)761 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
762 struct sk_buff *skb)
763 {
764 u8 *cb_data = bpf_skb_cb(skb);
765 u32 res;
766
767 if (unlikely(prog->cb_access))
768 memset(cb_data, 0, BPF_SKB_CB_LEN);
769
770 res = bpf_prog_run_pin_on_cpu(prog, skb);
771 return res;
772 }
773
774 DECLARE_BPF_DISPATCHER(xdp)
775
776 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
777
778 u32 xdp_master_redirect(struct xdp_buff *xdp);
779
780 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
781
bpf_prog_insn_size(const struct bpf_prog * prog)782 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
783 {
784 return prog->len * sizeof(struct bpf_insn);
785 }
786
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)787 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
788 {
789 return round_up(bpf_prog_insn_size(prog) +
790 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
791 }
792
bpf_prog_size(unsigned int proglen)793 static inline unsigned int bpf_prog_size(unsigned int proglen)
794 {
795 return max(sizeof(struct bpf_prog),
796 offsetof(struct bpf_prog, insns[proglen]));
797 }
798
bpf_prog_was_classic(const struct bpf_prog * prog)799 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
800 {
801 /* When classic BPF programs have been loaded and the arch
802 * does not have a classic BPF JIT (anymore), they have been
803 * converted via bpf_migrate_filter() to eBPF and thus always
804 * have an unspec program type.
805 */
806 return prog->type == BPF_PROG_TYPE_UNSPEC;
807 }
808
bpf_ctx_off_adjust_machine(u32 size)809 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
810 {
811 const u32 size_machine = sizeof(unsigned long);
812
813 if (size > size_machine && size % size_machine == 0)
814 size = size_machine;
815
816 return size;
817 }
818
819 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)820 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
821 {
822 return size <= size_default && (size & (size - 1)) == 0;
823 }
824
825 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)826 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
827 {
828 u8 access_off = off & (size_default - 1);
829
830 #ifdef __LITTLE_ENDIAN
831 return access_off;
832 #else
833 return size_default - (access_off + size);
834 #endif
835 }
836
837 #define bpf_ctx_wide_access_ok(off, size, type, field) \
838 (size == sizeof(__u64) && \
839 off >= offsetof(type, field) && \
840 off + sizeof(__u64) <= offsetofend(type, field) && \
841 off % sizeof(__u64) == 0)
842
843 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
844
bpf_prog_lock_ro(struct bpf_prog * fp)845 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
846 {
847 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
848 if (!fp->jited) {
849 set_vm_flush_reset_perms(fp);
850 set_memory_ro((unsigned long)fp, fp->pages);
851 }
852 #endif
853 }
854
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)855 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
856 {
857 set_vm_flush_reset_perms(hdr);
858 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
859 }
860
861 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)862 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
863 {
864 return sk_filter_trim_cap(sk, skb, 1);
865 }
866
867 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
868 void bpf_prog_free(struct bpf_prog *fp);
869
870 bool bpf_opcode_in_insntable(u8 code);
871
872 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
873 const u32 *insn_to_jit_off);
874 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
875 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
876
877 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
878 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
879 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
880 gfp_t gfp_extra_flags);
881 void __bpf_prog_free(struct bpf_prog *fp);
882
bpf_prog_unlock_free(struct bpf_prog * fp)883 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
884 {
885 __bpf_prog_free(fp);
886 }
887
888 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
889 unsigned int flen);
890
891 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
892 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
893 bpf_aux_classic_check_t trans, bool save_orig);
894 void bpf_prog_destroy(struct bpf_prog *fp);
895
896 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
897 int sk_attach_bpf(u32 ufd, struct sock *sk);
898 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
899 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
900 void sk_reuseport_prog_free(struct bpf_prog *prog);
901 int sk_detach_filter(struct sock *sk);
902 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
903
904 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
905 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
906
907 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
908 #define __bpf_call_base_args \
909 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
910 (void *)__bpf_call_base)
911
912 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
913 void bpf_jit_compile(struct bpf_prog *prog);
914 bool bpf_jit_needs_zext(void);
915 bool bpf_jit_supports_subprog_tailcalls(void);
916 bool bpf_jit_supports_kfunc_call(void);
917 bool bpf_jit_supports_far_kfunc_call(void);
918 bool bpf_helper_changes_pkt_data(void *func);
919
bpf_dump_raw_ok(const struct cred * cred)920 static inline bool bpf_dump_raw_ok(const struct cred *cred)
921 {
922 /* Reconstruction of call-sites is dependent on kallsyms,
923 * thus make dump the same restriction.
924 */
925 return kallsyms_show_value(cred);
926 }
927
928 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
929 const struct bpf_insn *patch, u32 len);
930 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
931
932 void bpf_clear_redirect_map(struct bpf_map *map);
933
xdp_return_frame_no_direct(void)934 static inline bool xdp_return_frame_no_direct(void)
935 {
936 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
937
938 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
939 }
940
xdp_set_return_frame_no_direct(void)941 static inline void xdp_set_return_frame_no_direct(void)
942 {
943 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
944
945 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
946 }
947
xdp_clear_return_frame_no_direct(void)948 static inline void xdp_clear_return_frame_no_direct(void)
949 {
950 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
951
952 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
953 }
954
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)955 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
956 unsigned int pktlen)
957 {
958 unsigned int len;
959
960 if (unlikely(!(fwd->flags & IFF_UP)))
961 return -ENETDOWN;
962
963 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
964 if (pktlen > len)
965 return -EMSGSIZE;
966
967 return 0;
968 }
969
970 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
971 * same cpu context. Further for best results no more than a single map
972 * for the do_redirect/do_flush pair should be used. This limitation is
973 * because we only track one map and force a flush when the map changes.
974 * This does not appear to be a real limitation for existing software.
975 */
976 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
977 struct xdp_buff *xdp, struct bpf_prog *prog);
978 int xdp_do_redirect(struct net_device *dev,
979 struct xdp_buff *xdp,
980 struct bpf_prog *prog);
981 int xdp_do_redirect_frame(struct net_device *dev,
982 struct xdp_buff *xdp,
983 struct xdp_frame *xdpf,
984 struct bpf_prog *prog);
985 void xdp_do_flush(void);
986
987 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
988 * it is no longer only flushing maps. Keep this define for compatibility
989 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
990 */
991 #define xdp_do_flush_map xdp_do_flush
992
993 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act);
994
995 #ifdef CONFIG_INET
996 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
997 struct bpf_prog *prog, struct sk_buff *skb,
998 struct sock *migrating_sk,
999 u32 hash);
1000 #else
1001 static inline struct sock *
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)1002 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1003 struct bpf_prog *prog, struct sk_buff *skb,
1004 struct sock *migrating_sk,
1005 u32 hash)
1006 {
1007 return NULL;
1008 }
1009 #endif
1010
1011 #ifdef CONFIG_BPF_JIT
1012 extern int bpf_jit_enable;
1013 extern int bpf_jit_harden;
1014 extern int bpf_jit_kallsyms;
1015 extern long bpf_jit_limit;
1016 extern long bpf_jit_limit_max;
1017
1018 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1019
1020 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1021
1022 struct bpf_binary_header *
1023 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1024 unsigned int alignment,
1025 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1026 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1027 u64 bpf_jit_alloc_exec_limit(void);
1028 void *bpf_jit_alloc_exec(unsigned long size);
1029 void bpf_jit_free_exec(void *addr);
1030 void bpf_jit_free(struct bpf_prog *fp);
1031 struct bpf_binary_header *
1032 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1033
1034 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1035 void bpf_prog_pack_free(struct bpf_binary_header *hdr);
1036
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1037 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1038 {
1039 return list_empty(&fp->aux->ksym.lnode) ||
1040 fp->aux->ksym.lnode.prev == LIST_POISON2;
1041 }
1042
1043 struct bpf_binary_header *
1044 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1045 unsigned int alignment,
1046 struct bpf_binary_header **rw_hdr,
1047 u8 **rw_image,
1048 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1049 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1050 struct bpf_binary_header *ro_header,
1051 struct bpf_binary_header *rw_header);
1052 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1053 struct bpf_binary_header *rw_header);
1054
1055 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1056 struct bpf_jit_poke_descriptor *poke);
1057
1058 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1059 const struct bpf_insn *insn, bool extra_pass,
1060 u64 *func_addr, bool *func_addr_fixed);
1061
1062 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1063 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1064
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1065 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1066 u32 pass, void *image)
1067 {
1068 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1069 proglen, pass, image, current->comm, task_pid_nr(current));
1070
1071 if (image)
1072 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1073 16, 1, image, proglen, false);
1074 }
1075
bpf_jit_is_ebpf(void)1076 static inline bool bpf_jit_is_ebpf(void)
1077 {
1078 # ifdef CONFIG_HAVE_EBPF_JIT
1079 return true;
1080 # else
1081 return false;
1082 # endif
1083 }
1084
ebpf_jit_enabled(void)1085 static inline bool ebpf_jit_enabled(void)
1086 {
1087 return bpf_jit_enable && bpf_jit_is_ebpf();
1088 }
1089
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1090 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1091 {
1092 return fp->jited && bpf_jit_is_ebpf();
1093 }
1094
bpf_jit_blinding_enabled(struct bpf_prog * prog)1095 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1096 {
1097 /* These are the prerequisites, should someone ever have the
1098 * idea to call blinding outside of them, we make sure to
1099 * bail out.
1100 */
1101 if (!bpf_jit_is_ebpf())
1102 return false;
1103 if (!prog->jit_requested)
1104 return false;
1105 if (!bpf_jit_harden)
1106 return false;
1107 if (bpf_jit_harden == 1 && bpf_capable())
1108 return false;
1109
1110 return true;
1111 }
1112
bpf_jit_kallsyms_enabled(void)1113 static inline bool bpf_jit_kallsyms_enabled(void)
1114 {
1115 /* There are a couple of corner cases where kallsyms should
1116 * not be enabled f.e. on hardening.
1117 */
1118 if (bpf_jit_harden)
1119 return false;
1120 if (!bpf_jit_kallsyms)
1121 return false;
1122 if (bpf_jit_kallsyms == 1)
1123 return true;
1124
1125 return false;
1126 }
1127
1128 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1129 unsigned long *off, char *sym);
1130 bool is_bpf_text_address(unsigned long addr);
1131 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1132 char *sym);
1133
1134 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1135 bpf_address_lookup(unsigned long addr, unsigned long *size,
1136 unsigned long *off, char **modname, char *sym)
1137 {
1138 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1139
1140 if (ret && modname)
1141 *modname = NULL;
1142 return ret;
1143 }
1144
1145 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1146 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1147
1148 #else /* CONFIG_BPF_JIT */
1149
ebpf_jit_enabled(void)1150 static inline bool ebpf_jit_enabled(void)
1151 {
1152 return false;
1153 }
1154
bpf_jit_blinding_enabled(struct bpf_prog * prog)1155 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1156 {
1157 return false;
1158 }
1159
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1160 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1161 {
1162 return false;
1163 }
1164
1165 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1166 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1167 struct bpf_jit_poke_descriptor *poke)
1168 {
1169 return -ENOTSUPP;
1170 }
1171
bpf_jit_free(struct bpf_prog * fp)1172 static inline void bpf_jit_free(struct bpf_prog *fp)
1173 {
1174 bpf_prog_unlock_free(fp);
1175 }
1176
bpf_jit_kallsyms_enabled(void)1177 static inline bool bpf_jit_kallsyms_enabled(void)
1178 {
1179 return false;
1180 }
1181
1182 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1183 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1184 unsigned long *off, char *sym)
1185 {
1186 return NULL;
1187 }
1188
is_bpf_text_address(unsigned long addr)1189 static inline bool is_bpf_text_address(unsigned long addr)
1190 {
1191 return false;
1192 }
1193
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1194 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1195 char *type, char *sym)
1196 {
1197 return -ERANGE;
1198 }
1199
1200 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1201 bpf_address_lookup(unsigned long addr, unsigned long *size,
1202 unsigned long *off, char **modname, char *sym)
1203 {
1204 return NULL;
1205 }
1206
bpf_prog_kallsyms_add(struct bpf_prog * fp)1207 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1208 {
1209 }
1210
bpf_prog_kallsyms_del(struct bpf_prog * fp)1211 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1212 {
1213 }
1214
1215 #endif /* CONFIG_BPF_JIT */
1216
1217 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1218
1219 #define BPF_ANC BIT(15)
1220
bpf_needs_clear_a(const struct sock_filter * first)1221 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1222 {
1223 switch (first->code) {
1224 case BPF_RET | BPF_K:
1225 case BPF_LD | BPF_W | BPF_LEN:
1226 return false;
1227
1228 case BPF_LD | BPF_W | BPF_ABS:
1229 case BPF_LD | BPF_H | BPF_ABS:
1230 case BPF_LD | BPF_B | BPF_ABS:
1231 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1232 return true;
1233 return false;
1234
1235 default:
1236 return true;
1237 }
1238 }
1239
bpf_anc_helper(const struct sock_filter * ftest)1240 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1241 {
1242 BUG_ON(ftest->code & BPF_ANC);
1243
1244 switch (ftest->code) {
1245 case BPF_LD | BPF_W | BPF_ABS:
1246 case BPF_LD | BPF_H | BPF_ABS:
1247 case BPF_LD | BPF_B | BPF_ABS:
1248 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1249 return BPF_ANC | SKF_AD_##CODE
1250 switch (ftest->k) {
1251 BPF_ANCILLARY(PROTOCOL);
1252 BPF_ANCILLARY(PKTTYPE);
1253 BPF_ANCILLARY(IFINDEX);
1254 BPF_ANCILLARY(NLATTR);
1255 BPF_ANCILLARY(NLATTR_NEST);
1256 BPF_ANCILLARY(MARK);
1257 BPF_ANCILLARY(QUEUE);
1258 BPF_ANCILLARY(HATYPE);
1259 BPF_ANCILLARY(RXHASH);
1260 BPF_ANCILLARY(CPU);
1261 BPF_ANCILLARY(ALU_XOR_X);
1262 BPF_ANCILLARY(VLAN_TAG);
1263 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1264 BPF_ANCILLARY(PAY_OFFSET);
1265 BPF_ANCILLARY(RANDOM);
1266 BPF_ANCILLARY(VLAN_TPID);
1267 }
1268 fallthrough;
1269 default:
1270 return ftest->code;
1271 }
1272 }
1273
1274 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1275 int k, unsigned int size);
1276
bpf_tell_extensions(void)1277 static inline int bpf_tell_extensions(void)
1278 {
1279 return SKF_AD_MAX;
1280 }
1281
1282 struct bpf_sock_addr_kern {
1283 struct sock *sk;
1284 struct sockaddr *uaddr;
1285 /* Temporary "register" to make indirect stores to nested structures
1286 * defined above. We need three registers to make such a store, but
1287 * only two (src and dst) are available at convert_ctx_access time
1288 */
1289 u64 tmp_reg;
1290 void *t_ctx; /* Attach type specific context. */
1291 u32 uaddrlen;
1292 };
1293
1294 struct bpf_sock_ops_kern {
1295 struct sock *sk;
1296 union {
1297 u32 args[4];
1298 u32 reply;
1299 u32 replylong[4];
1300 };
1301 struct sk_buff *syn_skb;
1302 struct sk_buff *skb;
1303 void *skb_data_end;
1304 u8 op;
1305 u8 is_fullsock;
1306 u8 remaining_opt_len;
1307 u64 temp; /* temp and everything after is not
1308 * initialized to 0 before calling
1309 * the BPF program. New fields that
1310 * should be initialized to 0 should
1311 * be inserted before temp.
1312 * temp is scratch storage used by
1313 * sock_ops_convert_ctx_access
1314 * as temporary storage of a register.
1315 */
1316 };
1317
1318 struct bpf_sysctl_kern {
1319 struct ctl_table_header *head;
1320 struct ctl_table *table;
1321 void *cur_val;
1322 size_t cur_len;
1323 void *new_val;
1324 size_t new_len;
1325 int new_updated;
1326 int write;
1327 loff_t *ppos;
1328 /* Temporary "register" for indirect stores to ppos. */
1329 u64 tmp_reg;
1330 };
1331
1332 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1333 struct bpf_sockopt_buf {
1334 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1335 };
1336
1337 struct bpf_sockopt_kern {
1338 struct sock *sk;
1339 u8 *optval;
1340 u8 *optval_end;
1341 s32 level;
1342 s32 optname;
1343 s32 optlen;
1344 /* for retval in struct bpf_cg_run_ctx */
1345 struct task_struct *current_task;
1346 /* Temporary "register" for indirect stores to ppos. */
1347 u64 tmp_reg;
1348 };
1349
1350 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1351
1352 struct bpf_sk_lookup_kern {
1353 u16 family;
1354 u16 protocol;
1355 __be16 sport;
1356 u16 dport;
1357 struct {
1358 __be32 saddr;
1359 __be32 daddr;
1360 } v4;
1361 struct {
1362 const struct in6_addr *saddr;
1363 const struct in6_addr *daddr;
1364 } v6;
1365 struct sock *selected_sk;
1366 u32 ingress_ifindex;
1367 bool no_reuseport;
1368 };
1369
1370 extern struct static_key_false bpf_sk_lookup_enabled;
1371
1372 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1373 *
1374 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1375 * SK_DROP. Their meaning is as follows:
1376 *
1377 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1378 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1379 * SK_DROP : terminate lookup with -ECONNREFUSED
1380 *
1381 * This macro aggregates return values and selected sockets from
1382 * multiple BPF programs according to following rules in order:
1383 *
1384 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1385 * macro result is SK_PASS and last ctx.selected_sk is used.
1386 * 2. If any program returned SK_DROP return value,
1387 * macro result is SK_DROP.
1388 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1389 *
1390 * Caller must ensure that the prog array is non-NULL, and that the
1391 * array as well as the programs it contains remain valid.
1392 */
1393 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1394 ({ \
1395 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1396 struct bpf_prog_array_item *_item; \
1397 struct sock *_selected_sk = NULL; \
1398 bool _no_reuseport = false; \
1399 struct bpf_prog *_prog; \
1400 bool _all_pass = true; \
1401 u32 _ret; \
1402 \
1403 migrate_disable(); \
1404 _item = &(array)->items[0]; \
1405 while ((_prog = READ_ONCE(_item->prog))) { \
1406 /* restore most recent selection */ \
1407 _ctx->selected_sk = _selected_sk; \
1408 _ctx->no_reuseport = _no_reuseport; \
1409 \
1410 _ret = func(_prog, _ctx); \
1411 if (_ret == SK_PASS && _ctx->selected_sk) { \
1412 /* remember last non-NULL socket */ \
1413 _selected_sk = _ctx->selected_sk; \
1414 _no_reuseport = _ctx->no_reuseport; \
1415 } else if (_ret == SK_DROP && _all_pass) { \
1416 _all_pass = false; \
1417 } \
1418 _item++; \
1419 } \
1420 _ctx->selected_sk = _selected_sk; \
1421 _ctx->no_reuseport = _no_reuseport; \
1422 migrate_enable(); \
1423 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1424 })
1425
bpf_sk_lookup_run_v4(struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1426 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1427 const __be32 saddr, const __be16 sport,
1428 const __be32 daddr, const u16 dport,
1429 const int ifindex, struct sock **psk)
1430 {
1431 struct bpf_prog_array *run_array;
1432 struct sock *selected_sk = NULL;
1433 bool no_reuseport = false;
1434
1435 rcu_read_lock();
1436 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1437 if (run_array) {
1438 struct bpf_sk_lookup_kern ctx = {
1439 .family = AF_INET,
1440 .protocol = protocol,
1441 .v4.saddr = saddr,
1442 .v4.daddr = daddr,
1443 .sport = sport,
1444 .dport = dport,
1445 .ingress_ifindex = ifindex,
1446 };
1447 u32 act;
1448
1449 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1450 if (act == SK_PASS) {
1451 selected_sk = ctx.selected_sk;
1452 no_reuseport = ctx.no_reuseport;
1453 } else {
1454 selected_sk = ERR_PTR(-ECONNREFUSED);
1455 }
1456 }
1457 rcu_read_unlock();
1458 *psk = selected_sk;
1459 return no_reuseport;
1460 }
1461
1462 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,const int ifindex,struct sock ** psk)1463 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1464 const struct in6_addr *saddr,
1465 const __be16 sport,
1466 const struct in6_addr *daddr,
1467 const u16 dport,
1468 const int ifindex, struct sock **psk)
1469 {
1470 struct bpf_prog_array *run_array;
1471 struct sock *selected_sk = NULL;
1472 bool no_reuseport = false;
1473
1474 rcu_read_lock();
1475 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1476 if (run_array) {
1477 struct bpf_sk_lookup_kern ctx = {
1478 .family = AF_INET6,
1479 .protocol = protocol,
1480 .v6.saddr = saddr,
1481 .v6.daddr = daddr,
1482 .sport = sport,
1483 .dport = dport,
1484 .ingress_ifindex = ifindex,
1485 };
1486 u32 act;
1487
1488 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1489 if (act == SK_PASS) {
1490 selected_sk = ctx.selected_sk;
1491 no_reuseport = ctx.no_reuseport;
1492 } else {
1493 selected_sk = ERR_PTR(-ECONNREFUSED);
1494 }
1495 }
1496 rcu_read_unlock();
1497 *psk = selected_sk;
1498 return no_reuseport;
1499 }
1500 #endif /* IS_ENABLED(CONFIG_IPV6) */
1501
__bpf_xdp_redirect_map(struct bpf_map * map,u64 index,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1502 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1503 u64 flags, const u64 flag_mask,
1504 void *lookup_elem(struct bpf_map *map, u32 key))
1505 {
1506 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1507 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1508
1509 /* Lower bits of the flags are used as return code on lookup failure */
1510 if (unlikely(flags & ~(action_mask | flag_mask)))
1511 return XDP_ABORTED;
1512
1513 ri->tgt_value = lookup_elem(map, index);
1514 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1515 /* If the lookup fails we want to clear out the state in the
1516 * redirect_info struct completely, so that if an eBPF program
1517 * performs multiple lookups, the last one always takes
1518 * precedence.
1519 */
1520 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1521 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1522 return flags & action_mask;
1523 }
1524
1525 ri->tgt_index = index;
1526 ri->map_id = map->id;
1527 ri->map_type = map->map_type;
1528
1529 if (flags & BPF_F_BROADCAST) {
1530 WRITE_ONCE(ri->map, map);
1531 ri->flags = flags;
1532 } else {
1533 WRITE_ONCE(ri->map, NULL);
1534 ri->flags = 0;
1535 }
1536
1537 return XDP_REDIRECT;
1538 }
1539
1540 #ifdef CONFIG_NET
1541 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1542 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1543 u32 len, u64 flags);
1544 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1545 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1546 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1547 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1548 void *buf, unsigned long len, bool flush);
1549 #else /* CONFIG_NET */
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1550 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1551 void *to, u32 len)
1552 {
1553 return -EOPNOTSUPP;
1554 }
1555
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1556 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1557 const void *from, u32 len, u64 flags)
1558 {
1559 return -EOPNOTSUPP;
1560 }
1561
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1562 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1563 void *buf, u32 len)
1564 {
1565 return -EOPNOTSUPP;
1566 }
1567
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1568 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1569 void *buf, u32 len)
1570 {
1571 return -EOPNOTSUPP;
1572 }
1573
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1574 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1575 {
1576 return NULL;
1577 }
1578
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1579 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1580 unsigned long len, bool flush)
1581 {
1582 }
1583 #endif /* CONFIG_NET */
1584
1585 #endif /* __LINUX_FILTER_H__ */
1586