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