]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - include/linux/filter.h
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
dax: Fix unlock mismatch with updated API
[mirror_ubuntu-eoan-kernel.git] / include / linux / filter.h
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 <stdarg.h>
9
10 #include <linux/atomic.h>
11 #include <linux/refcount.h>
12 #include <linux/compat.h>
13 #include <linux/skbuff.h>
14 #include <linux/linkage.h>
15 #include <linux/printk.h>
16 #include <linux/workqueue.h>
17 #include <linux/sched.h>
18 #include <linux/capability.h>
19 #include <linux/cryptohash.h>
20 #include <linux/set_memory.h>
21 #include <linux/kallsyms.h>
22 #include <linux/if_vlan.h>
23
24 #include <net/sch_generic.h>
25
26 #include <uapi/linux/filter.h>
27 #include <uapi/linux/bpf.h>
28
29 struct sk_buff;
30 struct sock;
31 struct seccomp_data;
32 struct bpf_prog_aux;
33 struct xdp_rxq_info;
34 struct xdp_buff;
35 struct sock_reuseport;
36
37 /* ArgX, context and stack frame pointer register positions. Note,
38 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
39 * calls in BPF_CALL instruction.
40 */
41 #define BPF_REG_ARG1 BPF_REG_1
42 #define BPF_REG_ARG2 BPF_REG_2
43 #define BPF_REG_ARG3 BPF_REG_3
44 #define BPF_REG_ARG4 BPF_REG_4
45 #define BPF_REG_ARG5 BPF_REG_5
46 #define BPF_REG_CTX BPF_REG_6
47 #define BPF_REG_FP BPF_REG_10
48
49 /* Additional register mappings for converted user programs. */
50 #define BPF_REG_A BPF_REG_0
51 #define BPF_REG_X BPF_REG_7
52 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
53 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
54 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
55
56 /* Kernel hidden auxiliary/helper register for hardening step.
57 * Only used by eBPF JITs. It's nothing more than a temporary
58 * register that JITs use internally, only that here it's part
59 * of eBPF instructions that have been rewritten for blinding
60 * constants. See JIT pre-step in bpf_jit_blind_constants().
61 */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
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 call to interpreter with arguments */
69 #define BPF_CALL_ARGS 0xe0
70
71 /* As per nm, we expose JITed images as text (code) section for
72 * kallsyms. That way, tools like perf can find it to match
73 * addresses.
74 */
75 #define BPF_SYM_ELF_TYPE 't'
76
77 /* BPF program can access up to 512 bytes of stack space. */
78 #define MAX_BPF_STACK 512
79
80 /* Helper macros for filter block array initializers. */
81
82 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
83
84 #define BPF_ALU64_REG(OP, DST, SRC) \
85 ((struct bpf_insn) { \
86 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
87 .dst_reg = DST, \
88 .src_reg = SRC, \
89 .off = 0, \
90 .imm = 0 })
91
92 #define BPF_ALU32_REG(OP, DST, SRC) \
93 ((struct bpf_insn) { \
94 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
95 .dst_reg = DST, \
96 .src_reg = SRC, \
97 .off = 0, \
98 .imm = 0 })
99
100 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
101
102 #define BPF_ALU64_IMM(OP, DST, IMM) \
103 ((struct bpf_insn) { \
104 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
105 .dst_reg = DST, \
106 .src_reg = 0, \
107 .off = 0, \
108 .imm = IMM })
109
110 #define BPF_ALU32_IMM(OP, DST, IMM) \
111 ((struct bpf_insn) { \
112 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
113 .dst_reg = DST, \
114 .src_reg = 0, \
115 .off = 0, \
116 .imm = IMM })
117
118 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
119
120 #define BPF_ENDIAN(TYPE, DST, LEN) \
121 ((struct bpf_insn) { \
122 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
123 .dst_reg = DST, \
124 .src_reg = 0, \
125 .off = 0, \
126 .imm = LEN })
127
128 /* Short form of mov, dst_reg = src_reg */
129
130 #define BPF_MOV64_REG(DST, SRC) \
131 ((struct bpf_insn) { \
132 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
133 .dst_reg = DST, \
134 .src_reg = SRC, \
135 .off = 0, \
136 .imm = 0 })
137
138 #define BPF_MOV32_REG(DST, SRC) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU | BPF_MOV | BPF_X, \
141 .dst_reg = DST, \
142 .src_reg = SRC, \
143 .off = 0, \
144 .imm = 0 })
145
146 /* Short form of mov, dst_reg = imm32 */
147
148 #define BPF_MOV64_IMM(DST, IMM) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
151 .dst_reg = DST, \
152 .src_reg = 0, \
153 .off = 0, \
154 .imm = IMM })
155
156 #define BPF_MOV32_IMM(DST, IMM) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU | BPF_MOV | BPF_K, \
159 .dst_reg = DST, \
160 .src_reg = 0, \
161 .off = 0, \
162 .imm = IMM })
163
164 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
165 #define BPF_LD_IMM64(DST, IMM) \
166 BPF_LD_IMM64_RAW(DST, 0, IMM)
167
168 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
169 ((struct bpf_insn) { \
170 .code = BPF_LD | BPF_DW | BPF_IMM, \
171 .dst_reg = DST, \
172 .src_reg = SRC, \
173 .off = 0, \
174 .imm = (__u32) (IMM) }), \
175 ((struct bpf_insn) { \
176 .code = 0, /* zero is reserved opcode */ \
177 .dst_reg = 0, \
178 .src_reg = 0, \
179 .off = 0, \
180 .imm = ((__u64) (IMM)) >> 32 })
181
182 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
183 #define BPF_LD_MAP_FD(DST, MAP_FD) \
184 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
185
186 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
187
188 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
189 ((struct bpf_insn) { \
190 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
191 .dst_reg = DST, \
192 .src_reg = SRC, \
193 .off = 0, \
194 .imm = IMM })
195
196 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
197 ((struct bpf_insn) { \
198 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
199 .dst_reg = DST, \
200 .src_reg = SRC, \
201 .off = 0, \
202 .imm = IMM })
203
204 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
205
206 #define BPF_LD_ABS(SIZE, IMM) \
207 ((struct bpf_insn) { \
208 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
209 .dst_reg = 0, \
210 .src_reg = 0, \
211 .off = 0, \
212 .imm = IMM })
213
214 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
215
216 #define BPF_LD_IND(SIZE, SRC, IMM) \
217 ((struct bpf_insn) { \
218 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
219 .dst_reg = 0, \
220 .src_reg = SRC, \
221 .off = 0, \
222 .imm = IMM })
223
224 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
225
226 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
227 ((struct bpf_insn) { \
228 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
229 .dst_reg = DST, \
230 .src_reg = SRC, \
231 .off = OFF, \
232 .imm = 0 })
233
234 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
235
236 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
237 ((struct bpf_insn) { \
238 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
239 .dst_reg = DST, \
240 .src_reg = SRC, \
241 .off = OFF, \
242 .imm = 0 })
243
244 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
245
246 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
247 ((struct bpf_insn) { \
248 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
249 .dst_reg = DST, \
250 .src_reg = SRC, \
251 .off = OFF, \
252 .imm = 0 })
253
254 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
255
256 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
257 ((struct bpf_insn) { \
258 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
259 .dst_reg = DST, \
260 .src_reg = 0, \
261 .off = OFF, \
262 .imm = IMM })
263
264 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
265
266 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
267 ((struct bpf_insn) { \
268 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
269 .dst_reg = DST, \
270 .src_reg = SRC, \
271 .off = OFF, \
272 .imm = 0 })
273
274 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
275
276 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
277 ((struct bpf_insn) { \
278 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
279 .dst_reg = DST, \
280 .src_reg = 0, \
281 .off = OFF, \
282 .imm = IMM })
283
284 /* Unconditional jumps, goto pc + off16 */
285
286 #define BPF_JMP_A(OFF) \
287 ((struct bpf_insn) { \
288 .code = BPF_JMP | BPF_JA, \
289 .dst_reg = 0, \
290 .src_reg = 0, \
291 .off = OFF, \
292 .imm = 0 })
293
294 /* Relative call */
295
296 #define BPF_CALL_REL(TGT) \
297 ((struct bpf_insn) { \
298 .code = BPF_JMP | BPF_CALL, \
299 .dst_reg = 0, \
300 .src_reg = BPF_PSEUDO_CALL, \
301 .off = 0, \
302 .imm = TGT })
303
304 /* Function call */
305
306 #define BPF_CAST_CALL(x) \
307 ((u64 (*)(u64, u64, u64, u64, u64))(x))
308
309 #define BPF_EMIT_CALL(FUNC) \
310 ((struct bpf_insn) { \
311 .code = BPF_JMP | BPF_CALL, \
312 .dst_reg = 0, \
313 .src_reg = 0, \
314 .off = 0, \
315 .imm = ((FUNC) - __bpf_call_base) })
316
317 /* Raw code statement block */
318
319 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
320 ((struct bpf_insn) { \
321 .code = CODE, \
322 .dst_reg = DST, \
323 .src_reg = SRC, \
324 .off = OFF, \
325 .imm = IMM })
326
327 /* Program exit */
328
329 #define BPF_EXIT_INSN() \
330 ((struct bpf_insn) { \
331 .code = BPF_JMP | BPF_EXIT, \
332 .dst_reg = 0, \
333 .src_reg = 0, \
334 .off = 0, \
335 .imm = 0 })
336
337 /* Internal classic blocks for direct assignment */
338
339 #define __BPF_STMT(CODE, K) \
340 ((struct sock_filter) BPF_STMT(CODE, K))
341
342 #define __BPF_JUMP(CODE, K, JT, JF) \
343 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
344
345 #define bytes_to_bpf_size(bytes) \
346 ({ \
347 int bpf_size = -EINVAL; \
348 \
349 if (bytes == sizeof(u8)) \
350 bpf_size = BPF_B; \
351 else if (bytes == sizeof(u16)) \
352 bpf_size = BPF_H; \
353 else if (bytes == sizeof(u32)) \
354 bpf_size = BPF_W; \
355 else if (bytes == sizeof(u64)) \
356 bpf_size = BPF_DW; \
357 \
358 bpf_size; \
359 })
360
361 #define bpf_size_to_bytes(bpf_size) \
362 ({ \
363 int bytes = -EINVAL; \
364 \
365 if (bpf_size == BPF_B) \
366 bytes = sizeof(u8); \
367 else if (bpf_size == BPF_H) \
368 bytes = sizeof(u16); \
369 else if (bpf_size == BPF_W) \
370 bytes = sizeof(u32); \
371 else if (bpf_size == BPF_DW) \
372 bytes = sizeof(u64); \
373 \
374 bytes; \
375 })
376
377 #define BPF_SIZEOF(type) \
378 ({ \
379 const int __size = bytes_to_bpf_size(sizeof(type)); \
380 BUILD_BUG_ON(__size < 0); \
381 __size; \
382 })
383
384 #define BPF_FIELD_SIZEOF(type, field) \
385 ({ \
386 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
387 BUILD_BUG_ON(__size < 0); \
388 __size; \
389 })
390
391 #define BPF_LDST_BYTES(insn) \
392 ({ \
393 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
394 WARN_ON(__size < 0); \
395 __size; \
396 })
397
398 #define __BPF_MAP_0(m, v, ...) v
399 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
400 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
401 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
402 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
403 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
404
405 #define __BPF_REG_0(...) __BPF_PAD(5)
406 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
407 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
408 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
409 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
410 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
411
412 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
413 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
414
415 #define __BPF_CAST(t, a) \
416 (__force t) \
417 (__force \
418 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
419 (unsigned long)0, (t)0))) a
420 #define __BPF_V void
421 #define __BPF_N
422
423 #define __BPF_DECL_ARGS(t, a) t a
424 #define __BPF_DECL_REGS(t, a) u64 a
425
426 #define __BPF_PAD(n) \
427 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
428 u64, __ur_3, u64, __ur_4, u64, __ur_5)
429
430 #define BPF_CALL_x(x, name, ...) \
431 static __always_inline \
432 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
433 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
434 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
435 { \
436 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
437 } \
438 static __always_inline \
439 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
440
441 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
442 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
443 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
444 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
445 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
446 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
447
448 #define bpf_ctx_range(TYPE, MEMBER) \
449 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
450 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
451 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
452
453 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
454 ({ \
455 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
456 *(PTR_SIZE) = (SIZE); \
457 offsetof(TYPE, MEMBER); \
458 })
459
460 #ifdef CONFIG_COMPAT
461 /* A struct sock_filter is architecture independent. */
462 struct compat_sock_fprog {
463 u16 len;
464 compat_uptr_t filter; /* struct sock_filter * */
465 };
466 #endif
467
468 struct sock_fprog_kern {
469 u16 len;
470 struct sock_filter *filter;
471 };
472
473 struct bpf_binary_header {
474 u32 pages;
475 /* Some arches need word alignment for their instructions */
476 u8 image[] __aligned(4);
477 };
478
479 struct bpf_prog {
480 u16 pages; /* Number of allocated pages */
481 u16 jited:1, /* Is our filter JIT'ed? */
482 jit_requested:1,/* archs need to JIT the prog */
483 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */
484 gpl_compatible:1, /* Is filter GPL compatible? */
485 cb_access:1, /* Is control block accessed? */
486 dst_needed:1, /* Do we need dst entry? */
487 blinded:1, /* Was blinded */
488 is_func:1, /* program is a bpf function */
489 kprobe_override:1, /* Do we override a kprobe? */
490 has_callchain_buf:1; /* callchain buffer allocated? */
491 enum bpf_prog_type type; /* Type of BPF program */
492 enum bpf_attach_type expected_attach_type; /* For some prog types */
493 u32 len; /* Number of filter blocks */
494 u32 jited_len; /* Size of jited insns in bytes */
495 u8 tag[BPF_TAG_SIZE];
496 struct bpf_prog_aux *aux; /* Auxiliary fields */
497 struct sock_fprog_kern *orig_prog; /* Original BPF program */
498 unsigned int (*bpf_func)(const void *ctx,
499 const struct bpf_insn *insn);
500 /* Instructions for interpreter */
501 union {
502 struct sock_filter insns[0];
503 struct bpf_insn insnsi[0];
504 };
505 };
506
507 struct sk_filter {
508 refcount_t refcnt;
509 struct rcu_head rcu;
510 struct bpf_prog *prog;
511 };
512
513 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
514
515 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
516
517 struct bpf_skb_data_end {
518 struct qdisc_skb_cb qdisc_cb;
519 void *data_meta;
520 void *data_end;
521 };
522
523 struct bpf_redirect_info {
524 u32 ifindex;
525 u32 flags;
526 struct bpf_map *map;
527 struct bpf_map *map_to_flush;
528 u32 kern_flags;
529 };
530
531 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
532
533 /* flags for bpf_redirect_info kern_flags */
534 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
535
536 /* Compute the linear packet data range [data, data_end) which
537 * will be accessed by various program types (cls_bpf, act_bpf,
538 * lwt, ...). Subsystems allowing direct data access must (!)
539 * ensure that cb[] area can be written to when BPF program is
540 * invoked (otherwise cb[] save/restore is necessary).
541 */
542 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
543 {
544 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
545
546 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
547 cb->data_meta = skb->data - skb_metadata_len(skb);
548 cb->data_end = skb->data + skb_headlen(skb);
549 }
550
551 /* Similar to bpf_compute_data_pointers(), except that save orginal
552 * data in cb->data and cb->meta_data for restore.
553 */
554 static inline void bpf_compute_and_save_data_end(
555 struct sk_buff *skb, void **saved_data_end)
556 {
557 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
558
559 *saved_data_end = cb->data_end;
560 cb->data_end = skb->data + skb_headlen(skb);
561 }
562
563 /* Restore data saved by bpf_compute_data_pointers(). */
564 static inline void bpf_restore_data_end(
565 struct sk_buff *skb, void *saved_data_end)
566 {
567 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
568
569 cb->data_end = saved_data_end;
570 }
571
572 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
573 {
574 /* eBPF programs may read/write skb->cb[] area to transfer meta
575 * data between tail calls. Since this also needs to work with
576 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
577 *
578 * In some socket filter cases, the cb unfortunately needs to be
579 * saved/restored so that protocol specific skb->cb[] data won't
580 * be lost. In any case, due to unpriviledged eBPF programs
581 * attached to sockets, we need to clear the bpf_skb_cb() area
582 * to not leak previous contents to user space.
583 */
584 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
585 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
586 FIELD_SIZEOF(struct qdisc_skb_cb, data));
587
588 return qdisc_skb_cb(skb)->data;
589 }
590
591 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
592 struct sk_buff *skb)
593 {
594 u8 *cb_data = bpf_skb_cb(skb);
595 u8 cb_saved[BPF_SKB_CB_LEN];
596 u32 res;
597
598 if (unlikely(prog->cb_access)) {
599 memcpy(cb_saved, cb_data, sizeof(cb_saved));
600 memset(cb_data, 0, sizeof(cb_saved));
601 }
602
603 res = BPF_PROG_RUN(prog, skb);
604
605 if (unlikely(prog->cb_access))
606 memcpy(cb_data, cb_saved, sizeof(cb_saved));
607
608 return res;
609 }
610
611 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
612 struct sk_buff *skb)
613 {
614 u8 *cb_data = bpf_skb_cb(skb);
615
616 if (unlikely(prog->cb_access))
617 memset(cb_data, 0, BPF_SKB_CB_LEN);
618
619 return BPF_PROG_RUN(prog, skb);
620 }
621
622 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
623 struct xdp_buff *xdp)
624 {
625 /* Caller needs to hold rcu_read_lock() (!), otherwise program
626 * can be released while still running, or map elements could be
627 * freed early while still having concurrent users. XDP fastpath
628 * already takes rcu_read_lock() when fetching the program, so
629 * it's not necessary here anymore.
630 */
631 return BPF_PROG_RUN(prog, xdp);
632 }
633
634 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
635 {
636 return prog->len * sizeof(struct bpf_insn);
637 }
638
639 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
640 {
641 return round_up(bpf_prog_insn_size(prog) +
642 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
643 }
644
645 static inline unsigned int bpf_prog_size(unsigned int proglen)
646 {
647 return max(sizeof(struct bpf_prog),
648 offsetof(struct bpf_prog, insns[proglen]));
649 }
650
651 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
652 {
653 /* When classic BPF programs have been loaded and the arch
654 * does not have a classic BPF JIT (anymore), they have been
655 * converted via bpf_migrate_filter() to eBPF and thus always
656 * have an unspec program type.
657 */
658 return prog->type == BPF_PROG_TYPE_UNSPEC;
659 }
660
661 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
662 {
663 const u32 size_machine = sizeof(unsigned long);
664
665 if (size > size_machine && size % size_machine == 0)
666 size = size_machine;
667
668 return size;
669 }
670
671 static inline bool bpf_ctx_narrow_align_ok(u32 off, u32 size_access,
672 u32 size_default)
673 {
674 size_default = bpf_ctx_off_adjust_machine(size_default);
675 size_access = bpf_ctx_off_adjust_machine(size_access);
676
677 #ifdef __LITTLE_ENDIAN
678 return (off & (size_default - 1)) == 0;
679 #else
680 return (off & (size_default - 1)) + size_access == size_default;
681 #endif
682 }
683
684 static inline bool
685 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
686 {
687 return bpf_ctx_narrow_align_ok(off, size, size_default) &&
688 size <= size_default && (size & (size - 1)) == 0;
689 }
690
691 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
692
693 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
694 {
695 fp->undo_set_mem = 1;
696 set_memory_ro((unsigned long)fp, fp->pages);
697 }
698
699 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
700 {
701 if (fp->undo_set_mem)
702 set_memory_rw((unsigned long)fp, fp->pages);
703 }
704
705 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
706 {
707 set_memory_ro((unsigned long)hdr, hdr->pages);
708 }
709
710 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
711 {
712 set_memory_rw((unsigned long)hdr, hdr->pages);
713 }
714
715 static inline struct bpf_binary_header *
716 bpf_jit_binary_hdr(const struct bpf_prog *fp)
717 {
718 unsigned long real_start = (unsigned long)fp->bpf_func;
719 unsigned long addr = real_start & PAGE_MASK;
720
721 return (void *)addr;
722 }
723
724 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
725 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
726 {
727 return sk_filter_trim_cap(sk, skb, 1);
728 }
729
730 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
731 void bpf_prog_free(struct bpf_prog *fp);
732
733 bool bpf_opcode_in_insntable(u8 code);
734
735 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
736 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
737 gfp_t gfp_extra_flags);
738 void __bpf_prog_free(struct bpf_prog *fp);
739
740 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
741 {
742 bpf_prog_unlock_ro(fp);
743 __bpf_prog_free(fp);
744 }
745
746 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
747 unsigned int flen);
748
749 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
750 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
751 bpf_aux_classic_check_t trans, bool save_orig);
752 void bpf_prog_destroy(struct bpf_prog *fp);
753
754 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
755 int sk_attach_bpf(u32 ufd, struct sock *sk);
756 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
757 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
758 void sk_reuseport_prog_free(struct bpf_prog *prog);
759 int sk_detach_filter(struct sock *sk);
760 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
761 unsigned int len);
762
763 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
764 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
765
766 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
767 #define __bpf_call_base_args \
768 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
769 __bpf_call_base)
770
771 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
772 void bpf_jit_compile(struct bpf_prog *prog);
773 bool bpf_helper_changes_pkt_data(void *func);
774
775 static inline bool bpf_dump_raw_ok(void)
776 {
777 /* Reconstruction of call-sites is dependent on kallsyms,
778 * thus make dump the same restriction.
779 */
780 return kallsyms_show_value() == 1;
781 }
782
783 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
784 const struct bpf_insn *patch, u32 len);
785
786 void bpf_clear_redirect_map(struct bpf_map *map);
787
788 static inline bool xdp_return_frame_no_direct(void)
789 {
790 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
791
792 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
793 }
794
795 static inline void xdp_set_return_frame_no_direct(void)
796 {
797 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
798
799 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
800 }
801
802 static inline void xdp_clear_return_frame_no_direct(void)
803 {
804 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
805
806 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
807 }
808
809 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
810 unsigned int pktlen)
811 {
812 unsigned int len;
813
814 if (unlikely(!(fwd->flags & IFF_UP)))
815 return -ENETDOWN;
816
817 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
818 if (pktlen > len)
819 return -EMSGSIZE;
820
821 return 0;
822 }
823
824 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
825 * same cpu context. Further for best results no more than a single map
826 * for the do_redirect/do_flush pair should be used. This limitation is
827 * because we only track one map and force a flush when the map changes.
828 * This does not appear to be a real limitation for existing software.
829 */
830 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
831 struct xdp_buff *xdp, struct bpf_prog *prog);
832 int xdp_do_redirect(struct net_device *dev,
833 struct xdp_buff *xdp,
834 struct bpf_prog *prog);
835 void xdp_do_flush_map(void);
836
837 void bpf_warn_invalid_xdp_action(u32 act);
838
839 #ifdef CONFIG_INET
840 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
841 struct bpf_prog *prog, struct sk_buff *skb,
842 u32 hash);
843 #else
844 static inline struct sock *
845 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
846 struct bpf_prog *prog, struct sk_buff *skb,
847 u32 hash)
848 {
849 return NULL;
850 }
851 #endif
852
853 #ifdef CONFIG_BPF_JIT
854 extern int bpf_jit_enable;
855 extern int bpf_jit_harden;
856 extern int bpf_jit_kallsyms;
857 extern int bpf_jit_limit;
858
859 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
860
861 struct bpf_binary_header *
862 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
863 unsigned int alignment,
864 bpf_jit_fill_hole_t bpf_fill_ill_insns);
865 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
866
867 void bpf_jit_free(struct bpf_prog *fp);
868
869 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
870 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
871
872 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
873 u32 pass, void *image)
874 {
875 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
876 proglen, pass, image, current->comm, task_pid_nr(current));
877
878 if (image)
879 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
880 16, 1, image, proglen, false);
881 }
882
883 static inline bool bpf_jit_is_ebpf(void)
884 {
885 # ifdef CONFIG_HAVE_EBPF_JIT
886 return true;
887 # else
888 return false;
889 # endif
890 }
891
892 static inline bool ebpf_jit_enabled(void)
893 {
894 return bpf_jit_enable && bpf_jit_is_ebpf();
895 }
896
897 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
898 {
899 return fp->jited && bpf_jit_is_ebpf();
900 }
901
902 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
903 {
904 /* These are the prerequisites, should someone ever have the
905 * idea to call blinding outside of them, we make sure to
906 * bail out.
907 */
908 if (!bpf_jit_is_ebpf())
909 return false;
910 if (!prog->jit_requested)
911 return false;
912 if (!bpf_jit_harden)
913 return false;
914 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
915 return false;
916
917 return true;
918 }
919
920 static inline bool bpf_jit_kallsyms_enabled(void)
921 {
922 /* There are a couple of corner cases where kallsyms should
923 * not be enabled f.e. on hardening.
924 */
925 if (bpf_jit_harden)
926 return false;
927 if (!bpf_jit_kallsyms)
928 return false;
929 if (bpf_jit_kallsyms == 1)
930 return true;
931
932 return false;
933 }
934
935 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
936 unsigned long *off, char *sym);
937 bool is_bpf_text_address(unsigned long addr);
938 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
939 char *sym);
940
941 static inline const char *
942 bpf_address_lookup(unsigned long addr, unsigned long *size,
943 unsigned long *off, char **modname, char *sym)
944 {
945 const char *ret = __bpf_address_lookup(addr, size, off, sym);
946
947 if (ret && modname)
948 *modname = NULL;
949 return ret;
950 }
951
952 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
953 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
954
955 #else /* CONFIG_BPF_JIT */
956
957 static inline bool ebpf_jit_enabled(void)
958 {
959 return false;
960 }
961
962 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
963 {
964 return false;
965 }
966
967 static inline void bpf_jit_free(struct bpf_prog *fp)
968 {
969 bpf_prog_unlock_free(fp);
970 }
971
972 static inline bool bpf_jit_kallsyms_enabled(void)
973 {
974 return false;
975 }
976
977 static inline const char *
978 __bpf_address_lookup(unsigned long addr, unsigned long *size,
979 unsigned long *off, char *sym)
980 {
981 return NULL;
982 }
983
984 static inline bool is_bpf_text_address(unsigned long addr)
985 {
986 return false;
987 }
988
989 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
990 char *type, char *sym)
991 {
992 return -ERANGE;
993 }
994
995 static inline const char *
996 bpf_address_lookup(unsigned long addr, unsigned long *size,
997 unsigned long *off, char **modname, char *sym)
998 {
999 return NULL;
1000 }
1001
1002 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1003 {
1004 }
1005
1006 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1007 {
1008 }
1009 #endif /* CONFIG_BPF_JIT */
1010
1011 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
1012 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1013
1014 #define BPF_ANC BIT(15)
1015
1016 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1017 {
1018 switch (first->code) {
1019 case BPF_RET | BPF_K:
1020 case BPF_LD | BPF_W | BPF_LEN:
1021 return false;
1022
1023 case BPF_LD | BPF_W | BPF_ABS:
1024 case BPF_LD | BPF_H | BPF_ABS:
1025 case BPF_LD | BPF_B | BPF_ABS:
1026 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1027 return true;
1028 return false;
1029
1030 default:
1031 return true;
1032 }
1033 }
1034
1035 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1036 {
1037 BUG_ON(ftest->code & BPF_ANC);
1038
1039 switch (ftest->code) {
1040 case BPF_LD | BPF_W | BPF_ABS:
1041 case BPF_LD | BPF_H | BPF_ABS:
1042 case BPF_LD | BPF_B | BPF_ABS:
1043 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1044 return BPF_ANC | SKF_AD_##CODE
1045 switch (ftest->k) {
1046 BPF_ANCILLARY(PROTOCOL);
1047 BPF_ANCILLARY(PKTTYPE);
1048 BPF_ANCILLARY(IFINDEX);
1049 BPF_ANCILLARY(NLATTR);
1050 BPF_ANCILLARY(NLATTR_NEST);
1051 BPF_ANCILLARY(MARK);
1052 BPF_ANCILLARY(QUEUE);
1053 BPF_ANCILLARY(HATYPE);
1054 BPF_ANCILLARY(RXHASH);
1055 BPF_ANCILLARY(CPU);
1056 BPF_ANCILLARY(ALU_XOR_X);
1057 BPF_ANCILLARY(VLAN_TAG);
1058 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1059 BPF_ANCILLARY(PAY_OFFSET);
1060 BPF_ANCILLARY(RANDOM);
1061 BPF_ANCILLARY(VLAN_TPID);
1062 }
1063 /* Fallthrough. */
1064 default:
1065 return ftest->code;
1066 }
1067 }
1068
1069 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1070 int k, unsigned int size);
1071
1072 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1073 unsigned int size, void *buffer)
1074 {
1075 if (k >= 0)
1076 return skb_header_pointer(skb, k, size, buffer);
1077
1078 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1079 }
1080
1081 static inline int bpf_tell_extensions(void)
1082 {
1083 return SKF_AD_MAX;
1084 }
1085
1086 struct bpf_sock_addr_kern {
1087 struct sock *sk;
1088 struct sockaddr *uaddr;
1089 /* Temporary "register" to make indirect stores to nested structures
1090 * defined above. We need three registers to make such a store, but
1091 * only two (src and dst) are available at convert_ctx_access time
1092 */
1093 u64 tmp_reg;
1094 void *t_ctx; /* Attach type specific context. */
1095 };
1096
1097 struct bpf_sock_ops_kern {
1098 struct sock *sk;
1099 u32 op;
1100 union {
1101 u32 args[4];
1102 u32 reply;
1103 u32 replylong[4];
1104 };
1105 u32 is_fullsock;
1106 u64 temp; /* temp and everything after is not
1107 * initialized to 0 before calling
1108 * the BPF program. New fields that
1109 * should be initialized to 0 should
1110 * be inserted before temp.
1111 * temp is scratch storage used by
1112 * sock_ops_convert_ctx_access
1113 * as temporary storage of a register.
1114 */
1115 };
1116
1117 #endif /* __LINUX_FILTER_H__ */
Ubuntu Eoan Linux kernel mirror