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 <linux/static_call.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_PARAMS MAX_BPF_REG
63 #define BPF_REG_AX (MAX_BPF_REG + 1)
64 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 2)
65 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
66
67 /* unused opcode to mark special call to bpf_tail_call() helper */
68 #define BPF_TAIL_CALL 0xf0
69
70 /* unused opcode to mark special load instruction. Same as BPF_ABS */
71 #define BPF_PROBE_MEM 0x20
72
73 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
74 #define BPF_PROBE_MEMSX 0x40
75
76 /* unused opcode to mark special load instruction. Same as BPF_MSH */
77 #define BPF_PROBE_MEM32 0xa0
78
79 /* unused opcode to mark special atomic instruction */
80 #define BPF_PROBE_ATOMIC 0xe0
81
82 /* unused opcode to mark special ldsx instruction. Same as BPF_NOSPEC */
83 #define BPF_PROBE_MEM32SX 0xc0
84
85 /* unused opcode to mark call to interpreter with arguments */
86 #define BPF_CALL_ARGS 0xe0
87
88 /* unused opcode to mark speculation barrier for mitigating
89 * Spectre v1 and v4
90 */
91 #define BPF_NOSPEC 0xc0
92
93 /* As per nm, we expose JITed images as text (code) section for
94 * kallsyms. That way, tools like perf can find it to match
95 * addresses.
96 */
97 #define BPF_SYM_ELF_TYPE 't'
98
99 /* BPF program can access up to 512 bytes of stack space. */
100 #define MAX_BPF_STACK 512
101
102 /* Helper macros for filter block array initializers. */
103
104 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
105
106 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
107 ((struct bpf_insn) { \
108 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
109 .dst_reg = DST, \
110 .src_reg = SRC, \
111 .off = OFF, \
112 .imm = 0 })
113
114 #define BPF_ALU64_REG(OP, DST, SRC) \
115 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
116
117 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
118 ((struct bpf_insn) { \
119 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
120 .dst_reg = DST, \
121 .src_reg = SRC, \
122 .off = OFF, \
123 .imm = 0 })
124
125 #define BPF_ALU32_REG(OP, DST, SRC) \
126 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
127
128 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
129
130 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \
131 ((struct bpf_insn) { \
132 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
133 .dst_reg = DST, \
134 .src_reg = 0, \
135 .off = OFF, \
136 .imm = IMM })
137 #define BPF_ALU64_IMM(OP, DST, IMM) \
138 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0)
139
140 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \
141 ((struct bpf_insn) { \
142 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
143 .dst_reg = DST, \
144 .src_reg = 0, \
145 .off = OFF, \
146 .imm = IMM })
147 #define BPF_ALU32_IMM(OP, DST, IMM) \
148 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0)
149
150 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
151
152 #define BPF_ENDIAN(TYPE, DST, LEN) \
153 ((struct bpf_insn) { \
154 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
155 .dst_reg = DST, \
156 .src_reg = 0, \
157 .off = 0, \
158 .imm = LEN })
159
160 /* Byte Swap, bswap16/32/64 */
161
162 #define BPF_BSWAP(DST, LEN) \
163 ((struct bpf_insn) { \
164 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \
165 .dst_reg = DST, \
166 .src_reg = 0, \
167 .off = 0, \
168 .imm = LEN })
169
170 /* Short form of mov, dst_reg = src_reg */
171
172 #define BPF_MOV64_REG(DST, SRC) \
173 ((struct bpf_insn) { \
174 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
175 .dst_reg = DST, \
176 .src_reg = SRC, \
177 .off = 0, \
178 .imm = 0 })
179
180 #define BPF_MOV32_REG(DST, SRC) \
181 ((struct bpf_insn) { \
182 .code = BPF_ALU | BPF_MOV | BPF_X, \
183 .dst_reg = DST, \
184 .src_reg = SRC, \
185 .off = 0, \
186 .imm = 0 })
187
188 /* Special (internal-only) form of mov, used to resolve per-CPU addrs:
189 * dst_reg = src_reg + <percpu_base_off>
190 * BPF_ADDR_PERCPU is used as a special insn->off value.
191 */
192 #define BPF_ADDR_PERCPU (-1)
193
194 #define BPF_MOV64_PERCPU_REG(DST, SRC) \
195 ((struct bpf_insn) { \
196 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
197 .dst_reg = DST, \
198 .src_reg = SRC, \
199 .off = BPF_ADDR_PERCPU, \
200 .imm = 0 })
201
insn_is_mov_percpu_addr(const struct bpf_insn * insn)202 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn)
203 {
204 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU;
205 }
206
207 /* Short form of mov, dst_reg = imm32 */
208
209 #define BPF_MOV64_IMM(DST, IMM) \
210 ((struct bpf_insn) { \
211 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
212 .dst_reg = DST, \
213 .src_reg = 0, \
214 .off = 0, \
215 .imm = IMM })
216
217 #define BPF_MOV32_IMM(DST, IMM) \
218 ((struct bpf_insn) { \
219 .code = BPF_ALU | BPF_MOV | BPF_K, \
220 .dst_reg = DST, \
221 .src_reg = 0, \
222 .off = 0, \
223 .imm = IMM })
224
225 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */
226
227 #define BPF_MOVSX64_REG(DST, SRC, OFF) \
228 ((struct bpf_insn) { \
229 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
230 .dst_reg = DST, \
231 .src_reg = SRC, \
232 .off = OFF, \
233 .imm = 0 })
234
235 #define BPF_MOVSX32_REG(DST, SRC, OFF) \
236 ((struct bpf_insn) { \
237 .code = BPF_ALU | BPF_MOV | BPF_X, \
238 .dst_reg = DST, \
239 .src_reg = SRC, \
240 .off = OFF, \
241 .imm = 0 })
242
243 /* Special form of mov32, used for doing explicit zero extension on dst. */
244 #define BPF_ZEXT_REG(DST) \
245 ((struct bpf_insn) { \
246 .code = BPF_ALU | BPF_MOV | BPF_X, \
247 .dst_reg = DST, \
248 .src_reg = DST, \
249 .off = 0, \
250 .imm = 1 })
251
insn_is_zext(const struct bpf_insn * insn)252 static inline bool insn_is_zext(const struct bpf_insn *insn)
253 {
254 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
255 }
256
257 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers
258 * to pointers in user vma.
259 */
insn_is_cast_user(const struct bpf_insn * insn)260 static inline bool insn_is_cast_user(const struct bpf_insn *insn)
261 {
262 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) &&
263 insn->off == BPF_ADDR_SPACE_CAST &&
264 insn->imm == 1U << 16;
265 }
266
267 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
268 #define BPF_LD_IMM64(DST, IMM) \
269 BPF_LD_IMM64_RAW(DST, 0, IMM)
270
271 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
272 ((struct bpf_insn) { \
273 .code = BPF_LD | BPF_DW | BPF_IMM, \
274 .dst_reg = DST, \
275 .src_reg = SRC, \
276 .off = 0, \
277 .imm = (__u32) (IMM) }), \
278 ((struct bpf_insn) { \
279 .code = 0, /* zero is reserved opcode */ \
280 .dst_reg = 0, \
281 .src_reg = 0, \
282 .off = 0, \
283 .imm = ((__u64) (IMM)) >> 32 })
284
285 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
286 #define BPF_LD_MAP_FD(DST, MAP_FD) \
287 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
288
289 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
290
291 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
292 ((struct bpf_insn) { \
293 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
294 .dst_reg = DST, \
295 .src_reg = SRC, \
296 .off = 0, \
297 .imm = IMM })
298
299 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
300 ((struct bpf_insn) { \
301 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
302 .dst_reg = DST, \
303 .src_reg = SRC, \
304 .off = 0, \
305 .imm = IMM })
306
307 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
308
309 #define BPF_LD_ABS(SIZE, IMM) \
310 ((struct bpf_insn) { \
311 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
312 .dst_reg = 0, \
313 .src_reg = 0, \
314 .off = 0, \
315 .imm = IMM })
316
317 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
318
319 #define BPF_LD_IND(SIZE, SRC, IMM) \
320 ((struct bpf_insn) { \
321 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
322 .dst_reg = 0, \
323 .src_reg = SRC, \
324 .off = 0, \
325 .imm = IMM })
326
327 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
328
329 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
330 ((struct bpf_insn) { \
331 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
332 .dst_reg = DST, \
333 .src_reg = SRC, \
334 .off = OFF, \
335 .imm = 0 })
336
337 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */
338
339 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \
340 ((struct bpf_insn) { \
341 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \
342 .dst_reg = DST, \
343 .src_reg = SRC, \
344 .off = OFF, \
345 .imm = 0 })
346
347 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
348
349 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
350 ((struct bpf_insn) { \
351 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
352 .dst_reg = DST, \
353 .src_reg = SRC, \
354 .off = OFF, \
355 .imm = 0 })
356
357
358 /*
359 * Atomic operations:
360 *
361 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
362 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
363 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
364 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
365 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
366 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
367 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
368 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
369 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
370 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
371 * BPF_LOAD_ACQ dst_reg = smp_load_acquire(src_reg + off16)
372 * BPF_STORE_REL smp_store_release(dst_reg + off16, src_reg)
373 */
374
375 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
376 ((struct bpf_insn) { \
377 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
378 .dst_reg = DST, \
379 .src_reg = SRC, \
380 .off = OFF, \
381 .imm = OP })
382
383 /* Legacy alias */
384 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
385
386 /*
387 * Given a BPF_ATOMIC instruction @atomic_insn, return true if it is an
388 * atomic load or store, and false if it is a read-modify-write instruction.
389 */
390 static inline bool
bpf_atomic_is_load_store(const struct bpf_insn * atomic_insn)391 bpf_atomic_is_load_store(const struct bpf_insn *atomic_insn)
392 {
393 switch (atomic_insn->imm) {
394 case BPF_LOAD_ACQ:
395 case BPF_STORE_REL:
396 return true;
397 default:
398 return false;
399 }
400 }
401
402 /*
403 * A load-acquire is the only BPF_STX class instruction that reads into
404 * dst_reg from src_reg + off16, i.e. it has the operand roles of a BPF_LDX.
405 * Unlike bpf_atomic_is_load_store(), @insn is not assumed to be a BPF_ATOMIC
406 * instruction here, so that callers which walk all instruction classes can
407 * use this directly.
408 */
bpf_atomic_is_load_acq(const struct bpf_insn * insn)409 static inline bool bpf_atomic_is_load_acq(const struct bpf_insn *insn)
410 {
411 return BPF_CLASS(insn->code) == BPF_STX &&
412 (BPF_MODE(insn->code) == BPF_ATOMIC ||
413 BPF_MODE(insn->code) == BPF_PROBE_ATOMIC) &&
414 insn->imm == BPF_LOAD_ACQ;
415 }
416
417 /*
418 * Given an instruction @insn, return the number of the BPF register that a
419 * BPF_ATOMIC reads the value at its memory operand into, or -1 if there is
420 * no such register. That is the register a BPF_PROBE_ATOMIC has to clear when
421 * the access faults. Like bpf_atomic_is_load_acq(), @insn is not assumed to
422 * be a BPF_ATOMIC here.
423 */
bpf_atomic_load_reg(const struct bpf_insn * insn)424 static inline int bpf_atomic_load_reg(const struct bpf_insn *insn)
425 {
426 if (BPF_CLASS(insn->code) != BPF_STX ||
427 (BPF_MODE(insn->code) != BPF_ATOMIC &&
428 BPF_MODE(insn->code) != BPF_PROBE_ATOMIC))
429 return -1;
430
431 switch (insn->imm) {
432 case BPF_LOAD_ACQ:
433 return insn->dst_reg;
434 case BPF_CMPXCHG:
435 return BPF_REG_0;
436 default:
437 return (insn->imm & BPF_FETCH) ? insn->src_reg : -1;
438 }
439 }
440
441 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
442
443 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
444 ((struct bpf_insn) { \
445 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
446 .dst_reg = DST, \
447 .src_reg = 0, \
448 .off = OFF, \
449 .imm = IMM })
450
451 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
452
453 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
454 ((struct bpf_insn) { \
455 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
456 .dst_reg = DST, \
457 .src_reg = SRC, \
458 .off = OFF, \
459 .imm = 0 })
460
461 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
462
463 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
464 ((struct bpf_insn) { \
465 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
466 .dst_reg = DST, \
467 .src_reg = 0, \
468 .off = OFF, \
469 .imm = IMM })
470
471 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
472
473 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
474 ((struct bpf_insn) { \
475 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
476 .dst_reg = DST, \
477 .src_reg = SRC, \
478 .off = OFF, \
479 .imm = 0 })
480
481 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
482
483 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
484 ((struct bpf_insn) { \
485 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
486 .dst_reg = DST, \
487 .src_reg = 0, \
488 .off = OFF, \
489 .imm = IMM })
490
491 /* Unconditional jumps, goto pc + off16 */
492
493 #define BPF_JMP_A(OFF) \
494 ((struct bpf_insn) { \
495 .code = BPF_JMP | BPF_JA, \
496 .dst_reg = 0, \
497 .src_reg = 0, \
498 .off = OFF, \
499 .imm = 0 })
500
501 /* Unconditional jumps, gotol pc + imm32 */
502
503 #define BPF_JMP32_A(IMM) \
504 ((struct bpf_insn) { \
505 .code = BPF_JMP32 | BPF_JA, \
506 .dst_reg = 0, \
507 .src_reg = 0, \
508 .off = 0, \
509 .imm = IMM })
510
511 /* Relative call */
512
513 #define BPF_CALL_REL(TGT) \
514 ((struct bpf_insn) { \
515 .code = BPF_JMP | BPF_CALL, \
516 .dst_reg = 0, \
517 .src_reg = BPF_PSEUDO_CALL, \
518 .off = 0, \
519 .imm = TGT })
520
521 /* Convert function address to BPF immediate */
522
523 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
524
525 #define BPF_EMIT_CALL(FUNC) \
526 ((struct bpf_insn) { \
527 .code = BPF_JMP | BPF_CALL, \
528 .dst_reg = 0, \
529 .src_reg = 0, \
530 .off = 0, \
531 .imm = BPF_CALL_IMM(FUNC) })
532
533 /* Kfunc call */
534
535 #define BPF_CALL_KFUNC(OFF, IMM) \
536 ((struct bpf_insn) { \
537 .code = BPF_JMP | BPF_CALL, \
538 .dst_reg = 0, \
539 .src_reg = BPF_PSEUDO_KFUNC_CALL, \
540 .off = OFF, \
541 .imm = IMM })
542
543 /* Raw code statement block */
544
545 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
546 ((struct bpf_insn) { \
547 .code = CODE, \
548 .dst_reg = DST, \
549 .src_reg = SRC, \
550 .off = OFF, \
551 .imm = IMM })
552
553 /* Program exit */
554
555 #define BPF_EXIT_INSN() \
556 ((struct bpf_insn) { \
557 .code = BPF_JMP | BPF_EXIT, \
558 .dst_reg = 0, \
559 .src_reg = 0, \
560 .off = 0, \
561 .imm = 0 })
562
563 /* Speculation barrier */
564
565 #define BPF_ST_NOSPEC() \
566 ((struct bpf_insn) { \
567 .code = BPF_ST | BPF_NOSPEC, \
568 .dst_reg = 0, \
569 .src_reg = 0, \
570 .off = 0, \
571 .imm = 0 })
572
573 /* Internal classic blocks for direct assignment */
574
575 #define __BPF_STMT(CODE, K) \
576 ((struct sock_filter) BPF_STMT(CODE, K))
577
578 #define __BPF_JUMP(CODE, K, JT, JF) \
579 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
580
581 #define bytes_to_bpf_size(bytes) \
582 ({ \
583 int bpf_size = -EINVAL; \
584 \
585 if (bytes == sizeof(u8)) \
586 bpf_size = BPF_B; \
587 else if (bytes == sizeof(u16)) \
588 bpf_size = BPF_H; \
589 else if (bytes == sizeof(u32)) \
590 bpf_size = BPF_W; \
591 else if (bytes == sizeof(u64)) \
592 bpf_size = BPF_DW; \
593 \
594 bpf_size; \
595 })
596
597 #define bpf_size_to_bytes(bpf_size) \
598 ({ \
599 int bytes = -EINVAL; \
600 \
601 if (bpf_size == BPF_B) \
602 bytes = sizeof(u8); \
603 else if (bpf_size == BPF_H) \
604 bytes = sizeof(u16); \
605 else if (bpf_size == BPF_W) \
606 bytes = sizeof(u32); \
607 else if (bpf_size == BPF_DW) \
608 bytes = sizeof(u64); \
609 \
610 bytes; \
611 })
612
613 #define BPF_SIZEOF(type) \
614 ({ \
615 const int __size = bytes_to_bpf_size(sizeof(type)); \
616 BUILD_BUG_ON(__size < 0); \
617 __size; \
618 })
619
620 #define BPF_FIELD_SIZEOF(type, field) \
621 ({ \
622 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
623 BUILD_BUG_ON(__size < 0); \
624 __size; \
625 })
626
627 #define BPF_LDST_BYTES(insn) \
628 ({ \
629 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
630 WARN_ON(__size < 0); \
631 __size; \
632 })
633
634 #define __BPF_MAP_0(m, v, ...) v
635 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
636 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
637 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
638 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
639 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
640
641 #define __BPF_REG_0(...) __BPF_PAD(5)
642 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
643 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
644 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
645 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
646 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
647
648 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
649 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
650
651 #define __BPF_CAST(t, a) \
652 (__force t) \
653 (__force \
654 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
655 (unsigned long)0, (t)0))) a
656 #define __BPF_V void
657 #define __BPF_N
658
659 #define __BPF_DECL_ARGS(t, a) t a
660 #define __BPF_DECL_REGS(t, a) u64 a
661
662 #define __BPF_PAD(n) \
663 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
664 u64, __ur_3, u64, __ur_4, u64, __ur_5)
665
666 #define BPF_CALL_x(x, attr, name, ...) \
667 static __always_inline \
668 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
669 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
670 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
671 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
672 { \
673 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
674 } \
675 static __always_inline \
676 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
677
678 #define __NOATTR
679 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
680 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
681 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
682 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
683 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
684 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
685
686 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
687
688 #define bpf_ctx_range(TYPE, MEMBER) \
689 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
690 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
691 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
692 #if BITS_PER_LONG == 64
693 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
694 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
695 #else
696 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
697 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
698 #endif /* BITS_PER_LONG == 64 */
699
700 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
701 ({ \
702 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
703 *(PTR_SIZE) = (SIZE); \
704 offsetof(TYPE, MEMBER); \
705 })
706
707 /* A struct sock_filter is architecture independent. */
708 struct compat_sock_fprog {
709 u16 len;
710 compat_uptr_t filter; /* struct sock_filter * */
711 };
712
713 struct sock_fprog_kern {
714 u16 len;
715 struct sock_filter *filter;
716 };
717
718 /* Some arches need doubleword alignment for their instructions and/or data */
719 #define BPF_IMAGE_ALIGNMENT 8
720
721 struct bpf_binary_header {
722 u32 size;
723 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
724 };
725
726 struct bpf_prog_stats {
727 u64_stats_t cnt;
728 u64_stats_t nsecs;
729 u64_stats_t misses;
730 struct u64_stats_sync syncp;
731 } __aligned(2 * sizeof(u64));
732
733 struct bpf_timed_may_goto {
734 u64 count;
735 u64 timestamp;
736 };
737
738 struct sk_filter {
739 refcount_t refcnt;
740 struct rcu_head rcu;
741 struct bpf_prog *prog;
742 };
743
744 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
745
746 extern struct mutex nf_conn_btf_access_lock;
747 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
748 const struct bpf_reg_state *reg,
749 int off, int size);
750
751 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
752 const struct bpf_insn *insnsi,
753 unsigned int (*bpf_func)(const void *,
754 const struct bpf_insn *));
755
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)756 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
757 const void *ctx,
758 bpf_dispatcher_fn dfunc)
759 {
760 u32 ret;
761
762 cant_migrate();
763 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
764 struct bpf_prog_stats *stats;
765 u64 duration, start = sched_clock();
766 unsigned long flags;
767
768 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
769
770 duration = sched_clock() - start;
771 if (likely(prog->stats)) {
772 stats = this_cpu_ptr(prog->stats);
773 flags = u64_stats_update_begin_irqsave(&stats->syncp);
774 u64_stats_inc(&stats->cnt);
775 u64_stats_add(&stats->nsecs, duration);
776 u64_stats_update_end_irqrestore(&stats->syncp, flags);
777 }
778 } else {
779 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
780 }
781 return ret;
782 }
783
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)784 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
785 {
786 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
787 }
788
789 /*
790 * Use in preemptible and therefore migratable context to make sure that
791 * the execution of the BPF program runs on one CPU.
792 *
793 * This uses migrate_disable/enable() explicitly to document that the
794 * invocation of a BPF program does not require reentrancy protection
795 * against a BPF program which is invoked from a preempting task.
796 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)797 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
798 const void *ctx)
799 {
800 u32 ret;
801
802 migrate_disable();
803 ret = bpf_prog_run(prog, ctx);
804 migrate_enable();
805 return ret;
806 }
807
is_stack_arg_ldx(const struct bpf_insn * insn)808 static inline bool is_stack_arg_ldx(const struct bpf_insn *insn)
809 {
810 return insn->code == (BPF_LDX | BPF_MEM | BPF_DW) &&
811 insn->src_reg == BPF_REG_PARAMS &&
812 insn->off > 0 && insn->off % 8 == 0;
813 }
814
is_stack_arg_st(const struct bpf_insn * insn)815 static inline bool is_stack_arg_st(const struct bpf_insn *insn)
816 {
817 return insn->code == (BPF_ST | BPF_MEM | BPF_DW) &&
818 insn->dst_reg == BPF_REG_PARAMS &&
819 insn->off < 0 && insn->off % 8 == 0;
820 }
821
is_stack_arg_stx(const struct bpf_insn * insn)822 static inline bool is_stack_arg_stx(const struct bpf_insn *insn)
823 {
824 return insn->code == (BPF_STX | BPF_MEM | BPF_DW) &&
825 insn->dst_reg == BPF_REG_PARAMS &&
826 insn->off < 0 && insn->off % 8 == 0;
827 }
828
829 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
830
831 struct bpf_skb_data_end {
832 struct qdisc_skb_cb qdisc_cb;
833 void *data_meta;
834 void *data_end;
835 };
836
837 struct bpf_nh_params {
838 u32 nh_family;
839 union {
840 u32 ipv4_nh;
841 struct in6_addr ipv6_nh;
842 };
843 };
844
845 /* flags for bpf_redirect_info kern_flags */
846 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
847 #define BPF_RI_F_RI_INIT BIT(1)
848 #define BPF_RI_F_CPU_MAP_INIT BIT(2)
849 #define BPF_RI_F_DEV_MAP_INIT BIT(3)
850 #define BPF_RI_F_XSK_MAP_INIT BIT(4)
851
852 struct bpf_redirect_info {
853 u64 tgt_index;
854 void *tgt_value;
855 struct bpf_map *map;
856 u32 flags;
857 u32 map_id;
858 enum bpf_map_type map_type;
859 struct bpf_nh_params nh;
860 u32 kern_flags;
861 };
862
863 struct bpf_net_context {
864 struct bpf_redirect_info ri;
865 struct list_head cpu_map_flush_list;
866 struct list_head dev_map_flush_list;
867 struct list_head xskmap_map_flush_list;
868 };
869
bpf_net_ctx_set(struct bpf_net_context * bpf_net_ctx)870 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
871 {
872 struct task_struct *tsk = current;
873
874 if (tsk->bpf_net_context != NULL)
875 return NULL;
876 bpf_net_ctx->ri.kern_flags = 0;
877
878 tsk->bpf_net_context = bpf_net_ctx;
879 return bpf_net_ctx;
880 }
881
bpf_net_ctx_clear(struct bpf_net_context * bpf_net_ctx)882 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
883 {
884 if (bpf_net_ctx)
885 current->bpf_net_context = NULL;
886 }
887
bpf_net_ctx_get(void)888 static inline struct bpf_net_context *bpf_net_ctx_get(void)
889 {
890 return current->bpf_net_context;
891 }
892
bpf_net_ctx_get_ri(void)893 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
894 {
895 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
896
897 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
898 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
899 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
900 }
901
902 return &bpf_net_ctx->ri;
903 }
904
bpf_net_ctx_get_cpu_map_flush_list(void)905 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
906 {
907 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
908
909 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
910 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list);
911 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
912 }
913
914 return &bpf_net_ctx->cpu_map_flush_list;
915 }
916
bpf_net_ctx_get_dev_flush_list(void)917 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
918 {
919 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
920
921 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
922 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list);
923 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
924 }
925
926 return &bpf_net_ctx->dev_map_flush_list;
927 }
928
bpf_net_ctx_get_xskmap_flush_list(void)929 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
930 {
931 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
932
933 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
934 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list);
935 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
936 }
937
938 return &bpf_net_ctx->xskmap_map_flush_list;
939 }
940
bpf_net_ctx_get_all_used_flush_lists(struct list_head ** lh_map,struct list_head ** lh_dev,struct list_head ** lh_xsk)941 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
942 struct list_head **lh_dev,
943 struct list_head **lh_xsk)
944 {
945 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
946 u32 kern_flags = bpf_net_ctx->ri.kern_flags;
947 struct list_head *lh;
948
949 *lh_map = *lh_dev = *lh_xsk = NULL;
950
951 if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
952 return;
953
954 lh = &bpf_net_ctx->dev_map_flush_list;
955 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh))
956 *lh_dev = lh;
957
958 lh = &bpf_net_ctx->cpu_map_flush_list;
959 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh))
960 *lh_map = lh;
961
962 lh = &bpf_net_ctx->xskmap_map_flush_list;
963 if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
964 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh))
965 *lh_xsk = lh;
966 }
967
968 /* Compute the linear packet data range [data, data_end) which
969 * will be accessed by various program types (cls_bpf, act_bpf,
970 * lwt, ...). Subsystems allowing direct data access must (!)
971 * ensure that cb[] area can be written to when BPF program is
972 * invoked (otherwise cb[] save/restore is necessary).
973 */
bpf_compute_data_pointers(struct sk_buff * skb)974 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
975 {
976 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
977
978 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
979 cb->data_meta = skb->data - skb_metadata_len(skb);
980 cb->data_end = skb->data + skb_headlen(skb);
981 }
982
bpf_prog_run_data_pointers(const struct bpf_prog * prog,struct sk_buff * skb)983 static inline int bpf_prog_run_data_pointers(
984 const struct bpf_prog *prog,
985 struct sk_buff *skb)
986 {
987 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
988 void *save_data_meta, *save_data_end;
989 int res;
990
991 save_data_meta = cb->data_meta;
992 save_data_end = cb->data_end;
993
994 bpf_compute_data_pointers(skb);
995 res = bpf_prog_run(prog, skb);
996
997 cb->data_meta = save_data_meta;
998 cb->data_end = save_data_end;
999
1000 return res;
1001 }
1002
1003 /* Similar to bpf_compute_data_pointers(), except that save orginal
1004 * data in cb->data and cb->meta_data for restore.
1005 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)1006 static inline void bpf_compute_and_save_data_end(
1007 struct sk_buff *skb, void **saved_data_end)
1008 {
1009 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
1010
1011 *saved_data_end = cb->data_end;
1012 cb->data_end = skb->data + skb_headlen(skb);
1013 }
1014
1015 /* Restore data saved by bpf_compute_and_save_data_end(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)1016 static inline void bpf_restore_data_end(
1017 struct sk_buff *skb, void *saved_data_end)
1018 {
1019 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
1020
1021 cb->data_end = saved_data_end;
1022 }
1023
bpf_skb_cb(const struct sk_buff * skb)1024 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
1025 {
1026 /* eBPF programs may read/write skb->cb[] area to transfer meta
1027 * data between tail calls. Since this also needs to work with
1028 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
1029 *
1030 * In some socket filter cases, the cb unfortunately needs to be
1031 * saved/restored so that protocol specific skb->cb[] data won't
1032 * be lost. In any case, due to unpriviledged eBPF programs
1033 * attached to sockets, we need to clear the bpf_skb_cb() area
1034 * to not leak previous contents to user space.
1035 */
1036 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
1037 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
1038 sizeof_field(struct qdisc_skb_cb, data));
1039
1040 return qdisc_skb_cb(skb)->data;
1041 }
1042
1043 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)1044 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
1045 const void *ctx)
1046 {
1047 const struct sk_buff *skb = ctx;
1048 u8 *cb_data = bpf_skb_cb(skb);
1049 u8 cb_saved[BPF_SKB_CB_LEN];
1050 u32 res;
1051
1052 if (unlikely(prog->cb_access)) {
1053 memcpy(cb_saved, cb_data, sizeof(cb_saved));
1054 memset(cb_data, 0, sizeof(cb_saved));
1055 }
1056
1057 res = bpf_prog_run(prog, skb);
1058
1059 if (unlikely(prog->cb_access))
1060 memcpy(cb_data, cb_saved, sizeof(cb_saved));
1061
1062 return res;
1063 }
1064
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)1065 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
1066 struct sk_buff *skb)
1067 {
1068 u32 res;
1069
1070 migrate_disable();
1071 res = __bpf_prog_run_save_cb(prog, skb);
1072 migrate_enable();
1073 return res;
1074 }
1075
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)1076 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
1077 struct sk_buff *skb)
1078 {
1079 u8 *cb_data = bpf_skb_cb(skb);
1080 u32 res;
1081
1082 if (unlikely(prog->cb_access))
1083 memset(cb_data, 0, BPF_SKB_CB_LEN);
1084
1085 res = bpf_prog_run_pin_on_cpu(prog, skb);
1086 return res;
1087 }
1088
1089 DECLARE_BPF_DISPATCHER(xdp)
1090
1091 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
1092
1093 u32 xdp_master_redirect(struct xdp_buff *xdp);
1094
1095 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
1096
bpf_prog_insn_size(const struct bpf_prog * prog)1097 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
1098 {
1099 return prog->len * sizeof(struct bpf_insn);
1100 }
1101
bpf_prog_size(unsigned int proglen)1102 static inline unsigned int bpf_prog_size(unsigned int proglen)
1103 {
1104 return max(sizeof(struct bpf_prog),
1105 offsetof(struct bpf_prog, insns[proglen]));
1106 }
1107
bpf_prog_was_classic(const struct bpf_prog * prog)1108 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
1109 {
1110 /* When classic BPF programs have been loaded and the arch
1111 * does not have a classic BPF JIT (anymore), they have been
1112 * converted via bpf_migrate_filter() to eBPF and thus always
1113 * have an unspec program type.
1114 */
1115 return prog->type == BPF_PROG_TYPE_UNSPEC;
1116 }
1117
bpf_ctx_off_adjust_machine(u32 size)1118 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
1119 {
1120 const u32 size_machine = sizeof(unsigned long);
1121
1122 if (size > size_machine && size % size_machine == 0)
1123 size = size_machine;
1124
1125 return size;
1126 }
1127
1128 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)1129 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
1130 {
1131 return size <= size_default && (size & (size - 1)) == 0;
1132 }
1133
1134 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)1135 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
1136 {
1137 u8 access_off = off & (size_default - 1);
1138
1139 #ifdef __LITTLE_ENDIAN
1140 return access_off;
1141 #else
1142 return size_default - (access_off + size);
1143 #endif
1144 }
1145
1146 #define bpf_ctx_wide_access_ok(off, size, type, field) \
1147 (size == sizeof(__u64) && \
1148 off >= offsetof(type, field) && \
1149 off + sizeof(__u64) <= offsetofend(type, field) && \
1150 off % sizeof(__u64) == 0)
1151
1152 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1153
bpf_prog_lock_ro(struct bpf_prog * fp)1154 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
1155 {
1156 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1157 if (!fp->jited) {
1158 set_vm_flush_reset_perms(fp);
1159 return set_memory_ro((unsigned long)fp, fp->pages);
1160 }
1161 #endif
1162 return 0;
1163 }
1164
1165 static inline int __must_check
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)1166 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
1167 {
1168 set_vm_flush_reset_perms(hdr);
1169 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
1170 }
1171
1172 enum skb_drop_reason
1173 sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
1174
sk_filter(struct sock * sk,struct sk_buff * skb)1175 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
1176 {
1177 enum skb_drop_reason drop_reason;
1178
1179 drop_reason = sk_filter_trim_cap(sk, skb, 1);
1180 return drop_reason ? -EPERM : 0;
1181 }
1182
1183 static inline enum skb_drop_reason
sk_filter_reason(struct sock * sk,struct sk_buff * skb)1184 sk_filter_reason(struct sock *sk, struct sk_buff *skb)
1185 {
1186 return sk_filter_trim_cap(sk, skb, 1);
1187 }
1188
1189 struct bpf_prog *__bpf_prog_select_runtime(struct bpf_verifier_env *env, struct bpf_prog *fp,
1190 int *err);
1191 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
1192 void bpf_prog_free(struct bpf_prog *fp);
1193
1194 bool bpf_opcode_in_insntable(u8 code);
1195
1196 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
1197 const u32 *insn_to_jit_off);
1198 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
1199 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
1200
1201 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
1202 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
1203 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
1204 gfp_t gfp_extra_flags);
1205 void __bpf_prog_free(struct bpf_prog *fp);
1206
bpf_prog_unlock_free(struct bpf_prog * fp)1207 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
1208 {
1209 __bpf_prog_free(fp);
1210 }
1211
1212 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
1213 unsigned int flen);
1214
1215 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
1216 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1217 bpf_aux_classic_check_t trans, bool save_orig);
1218 void bpf_prog_destroy(struct bpf_prog *fp);
1219
1220 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1221 int sk_attach_bpf(u32 ufd, struct sock *sk);
1222 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1223 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
1224 void sk_reuseport_prog_free(struct bpf_prog *prog);
1225 int sk_detach_filter(struct sock *sk);
1226 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
1227
1228 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
1229 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
1230
1231 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1232
1233 struct bpf_prog *bpf_int_jit_compile(struct bpf_verifier_env *env, struct bpf_prog *prog);
1234 void bpf_jit_compile(struct bpf_prog *prog);
1235 bool bpf_jit_needs_zext(void);
1236 bool bpf_jit_inlines_helper_call(s32 imm);
1237 bool bpf_jit_supports_subprog_tailcalls(void);
1238 bool bpf_jit_supports_percpu_insn(void);
1239 bool bpf_jit_supports_kfunc_call(void);
1240 bool bpf_jit_supports_stack_args(void);
1241 bool bpf_jit_supports_arena_args(void);
1242 bool bpf_jit_supports_far_kfunc_call(void);
1243 bool bpf_jit_supports_exceptions(void);
1244 bool bpf_jit_supports_ptr_xchg(void);
1245 bool bpf_jit_supports_arena(void);
1246 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena);
1247 bool bpf_jit_supports_private_stack(void);
1248 bool bpf_jit_supports_timed_may_goto(void);
1249 bool bpf_jit_supports_fsession(void);
1250 u64 bpf_arch_uaddress_limit(void);
1251 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie);
1252 u64 arch_bpf_timed_may_goto(void);
1253 u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *);
1254 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id);
1255
bpf_dump_raw_ok(const struct cred * cred)1256 static inline bool bpf_dump_raw_ok(const struct cred *cred)
1257 {
1258 /* Reconstruction of call-sites is dependent on kallsyms,
1259 * thus make dump the same restriction.
1260 */
1261 return kallsyms_show_value(cred);
1262 }
1263
1264 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
1265 const struct bpf_insn *patch, u32 len);
1266
1267 #ifdef CONFIG_BPF_SYSCALL
1268 struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
1269 const struct bpf_insn *patch, u32 len);
1270 #else
bpf_patch_insn_data(struct bpf_verifier_env * env,u32 off,const struct bpf_insn * patch,u32 len)1271 static inline struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
1272 const struct bpf_insn *patch, u32 len)
1273 {
1274 return ERR_PTR(-ENOTSUPP);
1275 }
1276 #endif /* CONFIG_BPF_SYSCALL */
1277
1278 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
1279
xdp_return_frame_no_direct(void)1280 static inline bool xdp_return_frame_no_direct(void)
1281 {
1282 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1283
1284 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
1285 }
1286
xdp_set_return_frame_no_direct(void)1287 static inline void xdp_set_return_frame_no_direct(void)
1288 {
1289 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1290
1291 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
1292 }
1293
xdp_clear_return_frame_no_direct(void)1294 static inline void xdp_clear_return_frame_no_direct(void)
1295 {
1296 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1297
1298 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
1299 }
1300
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)1301 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
1302 unsigned int pktlen)
1303 {
1304 unsigned int len;
1305
1306 if (unlikely(!(fwd->flags & IFF_UP)))
1307 return -ENETDOWN;
1308
1309 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1310 if (pktlen > len)
1311 return -EMSGSIZE;
1312
1313 return 0;
1314 }
1315
1316 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1317 * same cpu context. Further for best results no more than a single map
1318 * for the do_redirect/do_flush pair should be used. This limitation is
1319 * because we only track one map and force a flush when the map changes.
1320 * This does not appear to be a real limitation for existing software.
1321 */
1322 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1323 struct xdp_buff *xdp, const struct bpf_prog *prog);
1324 int xdp_do_redirect(struct net_device *dev,
1325 struct xdp_buff *xdp,
1326 const struct bpf_prog *prog);
1327 int xdp_do_redirect_frame(struct net_device *dev,
1328 struct xdp_buff *xdp,
1329 struct xdp_frame *xdpf,
1330 const struct bpf_prog *prog);
1331 void xdp_do_flush(void);
1332
1333 void bpf_warn_invalid_xdp_action(const struct net_device *dev,
1334 const struct bpf_prog *prog, u32 act);
1335
1336 #ifdef CONFIG_INET
1337 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1338 struct bpf_prog *prog, struct sk_buff *skb,
1339 struct sock *migrating_sk,
1340 u32 hash);
1341 #else
1342 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)1343 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1344 struct bpf_prog *prog, struct sk_buff *skb,
1345 struct sock *migrating_sk,
1346 u32 hash)
1347 {
1348 return NULL;
1349 }
1350 #endif
1351
1352 #ifdef CONFIG_BPF_JIT
1353 extern int bpf_jit_enable;
1354 extern int bpf_jit_harden;
1355 extern int bpf_jit_kallsyms;
1356 extern long bpf_jit_limit;
1357 extern long bpf_jit_limit_max;
1358
1359 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1360
1361 /*
1362 * Flush the indirect branch predictors before reusing JIT memory, so that
1363 * indirect jumps into a newly written program don't reuse predictions left
1364 * behind by an old program that occupied the same space.
1365 */
1366 void bpf_arch_pred_flush(void);
1367 DECLARE_STATIC_CALL(bpf_arch_pred_flush, bpf_arch_pred_flush);
1368 DECLARE_STATIC_KEY_FALSE(bpf_pred_flush_enabled);
1369
1370 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1371
1372 struct bpf_binary_header *
1373 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1374 unsigned int alignment,
1375 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1376 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1377 u64 bpf_jit_alloc_exec_limit(void);
1378 void *bpf_jit_alloc_exec(unsigned long size);
1379 void *bpf_jit_alloc_exec_rw(unsigned long size);
1380 void bpf_jit_free_exec(void *addr);
1381 void bpf_jit_free(struct bpf_prog *fp);
1382 struct bpf_binary_header *
1383 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1384
1385 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns, bool was_classic);
1386 void bpf_prog_pack_free(void *ptr, u32 size);
1387
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1388 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1389 {
1390 return list_empty(&fp->aux->ksym.lnode) ||
1391 fp->aux->ksym.lnode.prev == LIST_POISON2;
1392 }
1393
1394 struct bpf_binary_header *
1395 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1396 unsigned int alignment,
1397 struct bpf_binary_header **rw_hdr,
1398 u8 **rw_image,
1399 bpf_jit_fill_hole_t bpf_fill_ill_insns,
1400 bool was_classic);
1401 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
1402 struct bpf_binary_header *rw_header);
1403 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1404 struct bpf_binary_header *rw_header);
1405
1406 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1407 struct bpf_jit_poke_descriptor *poke);
1408
1409 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1410 const struct bpf_insn *insn, bool extra_pass,
1411 u64 *func_addr, bool *func_addr_fixed);
1412
1413 const char *bpf_jit_get_prog_name(struct bpf_prog *prog);
1414
1415 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog);
1416 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1417
bpf_prog_need_blind(const struct bpf_prog * prog)1418 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog)
1419 {
1420 return prog->blinding_requested && !prog->blinded;
1421 }
1422
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1423 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1424 u32 pass, void *image)
1425 {
1426 pr_err("flen=%u proglen=%u pass=%u image=%p from=%s pid=%d\n", flen,
1427 proglen, pass, image, current->comm, task_pid_nr(current));
1428
1429 if (image)
1430 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1431 16, 1, image, proglen, false);
1432 }
1433
bpf_jit_is_ebpf(void)1434 static inline bool bpf_jit_is_ebpf(void)
1435 {
1436 # ifdef CONFIG_HAVE_EBPF_JIT
1437 return true;
1438 # else
1439 return false;
1440 # endif
1441 }
1442
ebpf_jit_enabled(void)1443 static inline bool ebpf_jit_enabled(void)
1444 {
1445 return bpf_jit_enable && bpf_jit_is_ebpf();
1446 }
1447
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1448 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1449 {
1450 return fp->jited && bpf_jit_is_ebpf();
1451 }
1452
bpf_jit_blinding_enabled(struct bpf_prog * prog)1453 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1454 {
1455 /* These are the prerequisites, should someone ever have the
1456 * idea to call blinding outside of them, we make sure to
1457 * bail out.
1458 */
1459 if (!bpf_jit_is_ebpf())
1460 return false;
1461 if (!prog->jit_requested)
1462 return false;
1463 if (!bpf_jit_harden)
1464 return false;
1465 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF))
1466 return false;
1467
1468 return true;
1469 }
1470
bpf_jit_kallsyms_enabled(void)1471 static inline bool bpf_jit_kallsyms_enabled(void)
1472 {
1473 /* There are a couple of corner cases where kallsyms should
1474 * not be enabled f.e. on hardening.
1475 */
1476 if (bpf_jit_harden)
1477 return false;
1478 if (!bpf_jit_kallsyms)
1479 return false;
1480 if (bpf_jit_kallsyms == 1)
1481 return true;
1482
1483 return false;
1484 }
1485
1486 int bpf_address_lookup(unsigned long addr, unsigned long *size,
1487 unsigned long *off, char *sym);
1488 bool is_bpf_text_address(unsigned long addr);
1489 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1490 char *sym);
1491 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
1492
1493 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1494 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1495
1496 #else /* CONFIG_BPF_JIT */
1497
ebpf_jit_enabled(void)1498 static inline bool ebpf_jit_enabled(void)
1499 {
1500 return false;
1501 }
1502
bpf_jit_blinding_enabled(struct bpf_prog * prog)1503 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1504 {
1505 return false;
1506 }
1507
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1508 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1509 {
1510 return false;
1511 }
1512
1513 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1514 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1515 struct bpf_jit_poke_descriptor *poke)
1516 {
1517 return -ENOTSUPP;
1518 }
1519
bpf_jit_free(struct bpf_prog * fp)1520 static inline void bpf_jit_free(struct bpf_prog *fp)
1521 {
1522 bpf_prog_unlock_free(fp);
1523 }
1524
bpf_jit_kallsyms_enabled(void)1525 static inline bool bpf_jit_kallsyms_enabled(void)
1526 {
1527 return false;
1528 }
1529
1530 static inline int
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1531 bpf_address_lookup(unsigned long addr, unsigned long *size,
1532 unsigned long *off, char *sym)
1533 {
1534 return 0;
1535 }
1536
is_bpf_text_address(unsigned long addr)1537 static inline bool is_bpf_text_address(unsigned long addr)
1538 {
1539 return false;
1540 }
1541
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1542 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1543 char *type, char *sym)
1544 {
1545 return -ERANGE;
1546 }
1547
bpf_prog_ksym_find(unsigned long addr)1548 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
1549 {
1550 return NULL;
1551 }
1552
bpf_prog_kallsyms_add(struct bpf_prog * fp)1553 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1554 {
1555 }
1556
bpf_prog_kallsyms_del(struct bpf_prog * fp)1557 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1558 {
1559 }
1560
bpf_prog_need_blind(const struct bpf_prog * prog)1561 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog)
1562 {
1563 return false;
1564 }
1565
1566 static inline
bpf_jit_blind_constants(struct bpf_verifier_env * env,struct bpf_prog * prog)1567 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog)
1568 {
1569 return prog;
1570 }
1571
bpf_jit_prog_release_other(struct bpf_prog * fp,struct bpf_prog * fp_other)1572 static inline void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1573 {
1574 }
1575 #endif /* CONFIG_BPF_JIT */
1576
1577 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1578
1579 #define BPF_ANC BIT(15)
1580
bpf_needs_clear_a(const struct sock_filter * first)1581 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1582 {
1583 switch (first->code) {
1584 case BPF_RET | BPF_K:
1585 case BPF_LD | BPF_W | BPF_LEN:
1586 return false;
1587
1588 case BPF_LD | BPF_W | BPF_ABS:
1589 case BPF_LD | BPF_H | BPF_ABS:
1590 case BPF_LD | BPF_B | BPF_ABS:
1591 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1592 return true;
1593 return false;
1594
1595 default:
1596 return true;
1597 }
1598 }
1599
bpf_anc_helper(const struct sock_filter * ftest)1600 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1601 {
1602 BUG_ON(ftest->code & BPF_ANC);
1603
1604 switch (ftest->code) {
1605 case BPF_LD | BPF_W | BPF_ABS:
1606 case BPF_LD | BPF_H | BPF_ABS:
1607 case BPF_LD | BPF_B | BPF_ABS:
1608 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1609 return BPF_ANC | SKF_AD_##CODE
1610 switch (ftest->k) {
1611 BPF_ANCILLARY(PROTOCOL);
1612 BPF_ANCILLARY(PKTTYPE);
1613 BPF_ANCILLARY(IFINDEX);
1614 BPF_ANCILLARY(NLATTR);
1615 BPF_ANCILLARY(NLATTR_NEST);
1616 BPF_ANCILLARY(MARK);
1617 BPF_ANCILLARY(QUEUE);
1618 BPF_ANCILLARY(HATYPE);
1619 BPF_ANCILLARY(RXHASH);
1620 BPF_ANCILLARY(CPU);
1621 BPF_ANCILLARY(ALU_XOR_X);
1622 BPF_ANCILLARY(VLAN_TAG);
1623 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1624 BPF_ANCILLARY(PAY_OFFSET);
1625 BPF_ANCILLARY(RANDOM);
1626 BPF_ANCILLARY(VLAN_TPID);
1627 }
1628 fallthrough;
1629 default:
1630 return ftest->code;
1631 }
1632 }
1633
1634 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1635 int k, unsigned int size);
1636
bpf_tell_extensions(void)1637 static inline int bpf_tell_extensions(void)
1638 {
1639 return SKF_AD_MAX;
1640 }
1641
1642 struct bpf_sock_addr_kern {
1643 struct sock *sk;
1644 struct sockaddr_unsized *uaddr;
1645 /* Temporary "register" to make indirect stores to nested structures
1646 * defined above. We need three registers to make such a store, but
1647 * only two (src and dst) are available at convert_ctx_access time
1648 */
1649 u64 tmp_reg;
1650 void *t_ctx; /* Attach type specific context. */
1651 u32 uaddrlen;
1652 };
1653
1654 struct bpf_sock_ops_kern {
1655 struct sock *sk;
1656 union {
1657 u32 args[4];
1658 u32 reply;
1659 u32 replylong[4];
1660 };
1661 struct sk_buff *syn_skb;
1662 struct sk_buff *skb;
1663 void *skb_data_end;
1664 u8 op;
1665 u8 is_fullsock;
1666 u8 is_locked_tcp_sock;
1667 u8 remaining_opt_len;
1668 u64 temp; /* temp and everything after is not
1669 * initialized to 0 before calling
1670 * the BPF program. New fields that
1671 * should be initialized to 0 should
1672 * be inserted before temp.
1673 * temp is scratch storage used by
1674 * sock_ops_convert_ctx_access
1675 * as temporary storage of a register.
1676 */
1677 };
1678
1679 struct bpf_sysctl_kern {
1680 struct ctl_table_header *head;
1681 const struct ctl_table *table;
1682 void *cur_val;
1683 size_t cur_len;
1684 void *new_val;
1685 size_t new_len;
1686 int new_updated;
1687 int write;
1688 loff_t *ppos;
1689 /* Temporary "register" for indirect stores to ppos. */
1690 u64 tmp_reg;
1691 };
1692
1693 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1694 struct bpf_sockopt_buf {
1695 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1696 };
1697
1698 struct bpf_sockopt_kern {
1699 struct sock *sk;
1700 u8 *optval;
1701 u8 *optval_end;
1702 s32 level;
1703 s32 optname;
1704 s32 optlen;
1705 /* for retval in struct bpf_cg_run_ctx */
1706 struct task_struct *current_task;
1707 /* Temporary "register" for indirect stores to ppos. */
1708 u64 tmp_reg;
1709 };
1710
1711 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1712
1713 struct bpf_sk_lookup_kern {
1714 u16 family;
1715 u16 protocol;
1716 __be16 sport;
1717 u16 dport;
1718 struct {
1719 __be32 saddr;
1720 __be32 daddr;
1721 } v4;
1722 struct {
1723 const struct in6_addr *saddr;
1724 const struct in6_addr *daddr;
1725 } v6;
1726 struct sock *selected_sk;
1727 u32 ingress_ifindex;
1728 bool no_reuseport;
1729 };
1730
1731 extern struct static_key_false bpf_sk_lookup_enabled;
1732
1733 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1734 *
1735 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1736 * SK_DROP. Their meaning is as follows:
1737 *
1738 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1739 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1740 * SK_DROP : terminate lookup with -ECONNREFUSED
1741 *
1742 * This macro aggregates return values and selected sockets from
1743 * multiple BPF programs according to following rules in order:
1744 *
1745 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1746 * macro result is SK_PASS and last ctx.selected_sk is used.
1747 * 2. If any program returned SK_DROP return value,
1748 * macro result is SK_DROP.
1749 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1750 *
1751 * Caller must ensure that the prog array is non-NULL, and that the
1752 * array as well as the programs it contains remain valid.
1753 */
1754 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1755 ({ \
1756 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1757 struct bpf_prog_array_item *_item; \
1758 struct sock *_selected_sk = NULL; \
1759 bool _no_reuseport = false; \
1760 struct bpf_prog *_prog; \
1761 bool _all_pass = true; \
1762 u32 _ret; \
1763 \
1764 migrate_disable(); \
1765 _item = &(array)->items[0]; \
1766 while ((_prog = READ_ONCE(_item->prog))) { \
1767 /* restore most recent selection */ \
1768 _ctx->selected_sk = _selected_sk; \
1769 _ctx->no_reuseport = _no_reuseport; \
1770 \
1771 _ret = func(_prog, _ctx); \
1772 if (_ret == SK_PASS && _ctx->selected_sk) { \
1773 /* remember last non-NULL socket */ \
1774 _selected_sk = _ctx->selected_sk; \
1775 _no_reuseport = _ctx->no_reuseport; \
1776 } else if (_ret == SK_DROP && _all_pass) { \
1777 _all_pass = false; \
1778 } \
1779 _item++; \
1780 } \
1781 _ctx->selected_sk = _selected_sk; \
1782 _ctx->no_reuseport = _no_reuseport; \
1783 migrate_enable(); \
1784 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1785 })
1786
bpf_sk_lookup_run_v4(const struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1787 static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol,
1788 const __be32 saddr, const __be16 sport,
1789 const __be32 daddr, const u16 dport,
1790 const int ifindex, struct sock **psk)
1791 {
1792 struct bpf_prog_array *run_array;
1793 struct sock *selected_sk = NULL;
1794 bool no_reuseport = false;
1795
1796 rcu_read_lock();
1797 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1798 if (run_array) {
1799 struct bpf_sk_lookup_kern ctx = {
1800 .family = AF_INET,
1801 .protocol = protocol,
1802 .v4.saddr = saddr,
1803 .v4.daddr = daddr,
1804 .sport = sport,
1805 .dport = dport,
1806 .ingress_ifindex = ifindex,
1807 };
1808 u32 act;
1809
1810 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1811 if (act == SK_PASS) {
1812 selected_sk = ctx.selected_sk;
1813 no_reuseport = ctx.no_reuseport;
1814 } else {
1815 selected_sk = ERR_PTR(-ECONNREFUSED);
1816 }
1817 }
1818 rcu_read_unlock();
1819 *psk = selected_sk;
1820 return no_reuseport;
1821 }
1822
1823 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(const 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)1824 static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol,
1825 const struct in6_addr *saddr,
1826 const __be16 sport,
1827 const struct in6_addr *daddr,
1828 const u16 dport,
1829 const int ifindex, struct sock **psk)
1830 {
1831 struct bpf_prog_array *run_array;
1832 struct sock *selected_sk = NULL;
1833 bool no_reuseport = false;
1834
1835 rcu_read_lock();
1836 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1837 if (run_array) {
1838 struct bpf_sk_lookup_kern ctx = {
1839 .family = AF_INET6,
1840 .protocol = protocol,
1841 .v6.saddr = saddr,
1842 .v6.daddr = daddr,
1843 .sport = sport,
1844 .dport = dport,
1845 .ingress_ifindex = ifindex,
1846 };
1847 u32 act;
1848
1849 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1850 if (act == SK_PASS) {
1851 selected_sk = ctx.selected_sk;
1852 no_reuseport = ctx.no_reuseport;
1853 } else {
1854 selected_sk = ERR_PTR(-ECONNREFUSED);
1855 }
1856 }
1857 rcu_read_unlock();
1858 *psk = selected_sk;
1859 return no_reuseport;
1860 }
1861 #endif /* IS_ENABLED(CONFIG_IPV6) */
1862
__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))1863 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1864 u64 flags, const u64 flag_mask,
1865 void *lookup_elem(struct bpf_map *map, u32 key))
1866 {
1867 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1868 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1869
1870 /* Lower bits of the flags are used as return code on lookup failure */
1871 if (unlikely(flags & ~(action_mask | flag_mask)))
1872 return XDP_ABORTED;
1873
1874 ri->tgt_value = lookup_elem(map, index);
1875 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1876 /* If the lookup fails we want to clear out the state in the
1877 * redirect_info struct completely, so that if an eBPF program
1878 * performs multiple lookups, the last one always takes
1879 * precedence.
1880 */
1881 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1882 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1883 return flags & action_mask;
1884 }
1885
1886 ri->tgt_index = index;
1887 ri->map_id = map->id;
1888 ri->map_type = map->map_type;
1889
1890 if (flags & BPF_F_BROADCAST) {
1891 WRITE_ONCE(ri->map, map);
1892 ri->flags = flags;
1893 } else {
1894 WRITE_ONCE(ri->map, NULL);
1895 ri->flags = 0;
1896 }
1897
1898 return XDP_REDIRECT;
1899 }
1900
1901 #ifdef CONFIG_NET
1902 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1903 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1904 u32 len, u64 flags);
1905 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1906 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1907 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1908 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1909 void *buf, unsigned long len, bool flush);
1910 int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset,
1911 const void *from, u32 len, u64 flags);
1912 void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset);
1913 #else /* CONFIG_NET */
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1914 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1915 void *to, u32 len)
1916 {
1917 return -EOPNOTSUPP;
1918 }
1919
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1920 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1921 const void *from, u32 len, u64 flags)
1922 {
1923 return -EOPNOTSUPP;
1924 }
1925
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1926 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1927 void *buf, u32 len)
1928 {
1929 return -EOPNOTSUPP;
1930 }
1931
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1932 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1933 void *buf, u32 len)
1934 {
1935 return -EOPNOTSUPP;
1936 }
1937
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1938 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1939 {
1940 return NULL;
1941 }
1942
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1943 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1944 unsigned long len, bool flush)
1945 {
1946 }
1947
__bpf_skb_meta_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1948 static inline int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset,
1949 const void *from, u32 len,
1950 u64 flags)
1951 {
1952 return -EOPNOTSUPP;
1953 }
1954
bpf_skb_meta_pointer(struct sk_buff * skb,u32 offset)1955 static inline void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset)
1956 {
1957 return ERR_PTR(-EOPNOTSUPP);
1958 }
1959 #endif /* CONFIG_NET */
1960
1961 #endif /* __LINUX_FILTER_H__ */
1962