bpf: Add access to snd_cwnd and others in sock_ops
[linux-2.6-block.git] / include / linux / filter.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
1da177e4
LT
2/*
3 * Linux Socket Filter Data Structures
4 */
1da177e4
LT
5#ifndef __LINUX_FILTER_H__
6#define __LINUX_FILTER_H__
7
b954d834
DB
8#include <stdarg.h>
9
60063497 10#include <linux/atomic.h>
4c355cdf 11#include <linux/refcount.h>
0c5fe1b4 12#include <linux/compat.h>
9f12fbe6 13#include <linux/skbuff.h>
b954d834
DB
14#include <linux/linkage.h>
15#include <linux/printk.h>
d45ed4a4 16#include <linux/workqueue.h>
b13138ef 17#include <linux/sched.h>
4f3446bb 18#include <linux/capability.h>
7bd509e3 19#include <linux/cryptohash.h>
820a0b24 20#include <linux/set_memory.h>
4f3446bb 21
ff936a04 22#include <net/sch_generic.h>
b954d834 23
b954d834 24#include <uapi/linux/filter.h>
daedfb22 25#include <uapi/linux/bpf.h>
60a3b225
DB
26
27struct sk_buff;
28struct sock;
29struct seccomp_data;
09756af4 30struct bpf_prog_aux;
792d4b5c 31
30743837
DB
32/* ArgX, context and stack frame pointer register positions. Note,
33 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
34 * calls in BPF_CALL instruction.
35 */
36#define BPF_REG_ARG1 BPF_REG_1
37#define BPF_REG_ARG2 BPF_REG_2
38#define BPF_REG_ARG3 BPF_REG_3
39#define BPF_REG_ARG4 BPF_REG_4
40#define BPF_REG_ARG5 BPF_REG_5
41#define BPF_REG_CTX BPF_REG_6
42#define BPF_REG_FP BPF_REG_10
43
44/* Additional register mappings for converted user programs. */
45#define BPF_REG_A BPF_REG_0
46#define BPF_REG_X BPF_REG_7
47#define BPF_REG_TMP BPF_REG_8
bd4cf0ed 48
4f3446bb
DB
49/* Kernel hidden auxiliary/helper register for hardening step.
50 * Only used by eBPF JITs. It's nothing more than a temporary
51 * register that JITs use internally, only that here it's part
52 * of eBPF instructions that have been rewritten for blinding
53 * constants. See JIT pre-step in bpf_jit_blind_constants().
54 */
55#define BPF_REG_AX MAX_BPF_REG
56#define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
57
71189fa9
AS
58/* unused opcode to mark special call to bpf_tail_call() helper */
59#define BPF_TAIL_CALL 0xf0
60
74451e66
DB
61/* As per nm, we expose JITed images as text (code) section for
62 * kallsyms. That way, tools like perf can find it to match
63 * addresses.
64 */
65#define BPF_SYM_ELF_TYPE 't'
66
bd4cf0ed
AS
67/* BPF program can access up to 512 bytes of stack space. */
68#define MAX_BPF_STACK 512
69
f8f6d679
DB
70/* Helper macros for filter block array initializers. */
71
e430f34e 72/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
f8f6d679 73
e430f34e 74#define BPF_ALU64_REG(OP, DST, SRC) \
2695fb55 75 ((struct bpf_insn) { \
f8f6d679 76 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
e430f34e
AS
77 .dst_reg = DST, \
78 .src_reg = SRC, \
f8f6d679
DB
79 .off = 0, \
80 .imm = 0 })
81
e430f34e 82#define BPF_ALU32_REG(OP, DST, SRC) \
2695fb55 83 ((struct bpf_insn) { \
f8f6d679 84 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
e430f34e
AS
85 .dst_reg = DST, \
86 .src_reg = SRC, \
f8f6d679
DB
87 .off = 0, \
88 .imm = 0 })
89
e430f34e 90/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
f8f6d679 91
e430f34e 92#define BPF_ALU64_IMM(OP, DST, IMM) \
2695fb55 93 ((struct bpf_insn) { \
f8f6d679 94 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
e430f34e
AS
95 .dst_reg = DST, \
96 .src_reg = 0, \
f8f6d679
DB
97 .off = 0, \
98 .imm = IMM })
99
e430f34e 100#define BPF_ALU32_IMM(OP, DST, IMM) \
2695fb55 101 ((struct bpf_insn) { \
f8f6d679 102 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
e430f34e
AS
103 .dst_reg = DST, \
104 .src_reg = 0, \
f8f6d679
DB
105 .off = 0, \
106 .imm = IMM })
107
108/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
109
e430f34e 110#define BPF_ENDIAN(TYPE, DST, LEN) \
2695fb55 111 ((struct bpf_insn) { \
f8f6d679 112 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
e430f34e
AS
113 .dst_reg = DST, \
114 .src_reg = 0, \
f8f6d679
DB
115 .off = 0, \
116 .imm = LEN })
117
e430f34e 118/* Short form of mov, dst_reg = src_reg */
f8f6d679 119
e430f34e 120#define BPF_MOV64_REG(DST, SRC) \
2695fb55 121 ((struct bpf_insn) { \
f8f6d679 122 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
e430f34e
AS
123 .dst_reg = DST, \
124 .src_reg = SRC, \
f8f6d679
DB
125 .off = 0, \
126 .imm = 0 })
127
e430f34e 128#define BPF_MOV32_REG(DST, SRC) \
2695fb55 129 ((struct bpf_insn) { \
f8f6d679 130 .code = BPF_ALU | BPF_MOV | BPF_X, \
e430f34e
AS
131 .dst_reg = DST, \
132 .src_reg = SRC, \
f8f6d679
DB
133 .off = 0, \
134 .imm = 0 })
135
e430f34e 136/* Short form of mov, dst_reg = imm32 */
f8f6d679 137
e430f34e 138#define BPF_MOV64_IMM(DST, IMM) \
2695fb55 139 ((struct bpf_insn) { \
f8f6d679 140 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
e430f34e
AS
141 .dst_reg = DST, \
142 .src_reg = 0, \
f8f6d679
DB
143 .off = 0, \
144 .imm = IMM })
145
e430f34e 146#define BPF_MOV32_IMM(DST, IMM) \
2695fb55 147 ((struct bpf_insn) { \
f8f6d679 148 .code = BPF_ALU | BPF_MOV | BPF_K, \
e430f34e
AS
149 .dst_reg = DST, \
150 .src_reg = 0, \
f8f6d679
DB
151 .off = 0, \
152 .imm = IMM })
153
02ab695b
AS
154/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
155#define BPF_LD_IMM64(DST, IMM) \
156 BPF_LD_IMM64_RAW(DST, 0, IMM)
157
158#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
159 ((struct bpf_insn) { \
160 .code = BPF_LD | BPF_DW | BPF_IMM, \
161 .dst_reg = DST, \
162 .src_reg = SRC, \
163 .off = 0, \
164 .imm = (__u32) (IMM) }), \
165 ((struct bpf_insn) { \
166 .code = 0, /* zero is reserved opcode */ \
167 .dst_reg = 0, \
168 .src_reg = 0, \
169 .off = 0, \
170 .imm = ((__u64) (IMM)) >> 32 })
171
0246e64d
AS
172/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
173#define BPF_LD_MAP_FD(DST, MAP_FD) \
174 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
175
e430f34e 176/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
f8f6d679 177
e430f34e 178#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
2695fb55 179 ((struct bpf_insn) { \
f8f6d679 180 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
e430f34e
AS
181 .dst_reg = DST, \
182 .src_reg = SRC, \
f8f6d679
DB
183 .off = 0, \
184 .imm = IMM })
185
e430f34e 186#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
2695fb55 187 ((struct bpf_insn) { \
f8f6d679 188 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
e430f34e
AS
189 .dst_reg = DST, \
190 .src_reg = SRC, \
f8f6d679
DB
191 .off = 0, \
192 .imm = IMM })
193
e430f34e 194/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
f8f6d679 195
e430f34e 196#define BPF_LD_ABS(SIZE, IMM) \
2695fb55 197 ((struct bpf_insn) { \
f8f6d679 198 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
e430f34e
AS
199 .dst_reg = 0, \
200 .src_reg = 0, \
f8f6d679 201 .off = 0, \
e430f34e 202 .imm = IMM })
f8f6d679 203
e430f34e 204/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
f8f6d679 205
e430f34e 206#define BPF_LD_IND(SIZE, SRC, IMM) \
2695fb55 207 ((struct bpf_insn) { \
f8f6d679 208 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
e430f34e
AS
209 .dst_reg = 0, \
210 .src_reg = SRC, \
f8f6d679 211 .off = 0, \
e430f34e 212 .imm = IMM })
f8f6d679 213
e430f34e 214/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
f8f6d679 215
e430f34e 216#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
2695fb55 217 ((struct bpf_insn) { \
f8f6d679 218 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
e430f34e
AS
219 .dst_reg = DST, \
220 .src_reg = SRC, \
f8f6d679
DB
221 .off = OFF, \
222 .imm = 0 })
223
e430f34e
AS
224/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
225
226#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
2695fb55 227 ((struct bpf_insn) { \
f8f6d679 228 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
e430f34e
AS
229 .dst_reg = DST, \
230 .src_reg = SRC, \
f8f6d679
DB
231 .off = OFF, \
232 .imm = 0 })
233
cffc642d
MH
234/* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
235
236#define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
237 ((struct bpf_insn) { \
238 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
239 .dst_reg = DST, \
240 .src_reg = SRC, \
241 .off = OFF, \
242 .imm = 0 })
243
e430f34e
AS
244/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
245
246#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
2695fb55 247 ((struct bpf_insn) { \
e430f34e
AS
248 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
249 .dst_reg = DST, \
250 .src_reg = 0, \
251 .off = OFF, \
252 .imm = IMM })
253
254/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
f8f6d679 255
e430f34e 256#define BPF_JMP_REG(OP, DST, SRC, OFF) \
2695fb55 257 ((struct bpf_insn) { \
f8f6d679 258 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
e430f34e
AS
259 .dst_reg = DST, \
260 .src_reg = SRC, \
f8f6d679
DB
261 .off = OFF, \
262 .imm = 0 })
263
e430f34e 264/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
f8f6d679 265
e430f34e 266#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
2695fb55 267 ((struct bpf_insn) { \
f8f6d679 268 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
e430f34e
AS
269 .dst_reg = DST, \
270 .src_reg = 0, \
f8f6d679
DB
271 .off = OFF, \
272 .imm = IMM })
273
614d0d77
DB
274/* Unconditional jumps, goto pc + off16 */
275
276#define BPF_JMP_A(OFF) \
277 ((struct bpf_insn) { \
278 .code = BPF_JMP | BPF_JA, \
279 .dst_reg = 0, \
280 .src_reg = 0, \
281 .off = OFF, \
282 .imm = 0 })
283
f8f6d679
DB
284/* Function call */
285
286#define BPF_EMIT_CALL(FUNC) \
2695fb55 287 ((struct bpf_insn) { \
f8f6d679 288 .code = BPF_JMP | BPF_CALL, \
e430f34e
AS
289 .dst_reg = 0, \
290 .src_reg = 0, \
f8f6d679
DB
291 .off = 0, \
292 .imm = ((FUNC) - __bpf_call_base) })
293
294/* Raw code statement block */
295
e430f34e 296#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
2695fb55 297 ((struct bpf_insn) { \
f8f6d679 298 .code = CODE, \
e430f34e
AS
299 .dst_reg = DST, \
300 .src_reg = SRC, \
f8f6d679
DB
301 .off = OFF, \
302 .imm = IMM })
303
304/* Program exit */
305
306#define BPF_EXIT_INSN() \
2695fb55 307 ((struct bpf_insn) { \
f8f6d679 308 .code = BPF_JMP | BPF_EXIT, \
e430f34e
AS
309 .dst_reg = 0, \
310 .src_reg = 0, \
f8f6d679
DB
311 .off = 0, \
312 .imm = 0 })
313
a4afd37b
DB
314/* Internal classic blocks for direct assignment */
315
316#define __BPF_STMT(CODE, K) \
317 ((struct sock_filter) BPF_STMT(CODE, K))
318
319#define __BPF_JUMP(CODE, K, JT, JF) \
320 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
321
f8f6d679
DB
322#define bytes_to_bpf_size(bytes) \
323({ \
324 int bpf_size = -EINVAL; \
325 \
326 if (bytes == sizeof(u8)) \
327 bpf_size = BPF_B; \
328 else if (bytes == sizeof(u16)) \
329 bpf_size = BPF_H; \
330 else if (bytes == sizeof(u32)) \
331 bpf_size = BPF_W; \
332 else if (bytes == sizeof(u64)) \
333 bpf_size = BPF_DW; \
334 \
335 bpf_size; \
336})
9739eef1 337
f96da094
DB
338#define bpf_size_to_bytes(bpf_size) \
339({ \
340 int bytes = -EINVAL; \
341 \
342 if (bpf_size == BPF_B) \
343 bytes = sizeof(u8); \
344 else if (bpf_size == BPF_H) \
345 bytes = sizeof(u16); \
346 else if (bpf_size == BPF_W) \
347 bytes = sizeof(u32); \
348 else if (bpf_size == BPF_DW) \
349 bytes = sizeof(u64); \
350 \
351 bytes; \
352})
353
f035a515
DB
354#define BPF_SIZEOF(type) \
355 ({ \
356 const int __size = bytes_to_bpf_size(sizeof(type)); \
357 BUILD_BUG_ON(__size < 0); \
358 __size; \
359 })
360
361#define BPF_FIELD_SIZEOF(type, field) \
362 ({ \
363 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
364 BUILD_BUG_ON(__size < 0); \
365 __size; \
366 })
367
f96da094
DB
368#define BPF_LDST_BYTES(insn) \
369 ({ \
370 const int __size = bpf_size_to_bytes(BPF_SIZE(insn->code)); \
371 WARN_ON(__size < 0); \
372 __size; \
373 })
374
f3694e00
DB
375#define __BPF_MAP_0(m, v, ...) v
376#define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
377#define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
378#define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
379#define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
380#define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
381
382#define __BPF_REG_0(...) __BPF_PAD(5)
383#define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
384#define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
385#define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
386#define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
387#define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
388
389#define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
390#define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
391
392#define __BPF_CAST(t, a) \
393 (__force t) \
394 (__force \
395 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
396 (unsigned long)0, (t)0))) a
397#define __BPF_V void
398#define __BPF_N
399
400#define __BPF_DECL_ARGS(t, a) t a
401#define __BPF_DECL_REGS(t, a) u64 a
402
403#define __BPF_PAD(n) \
404 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
405 u64, __ur_3, u64, __ur_4, u64, __ur_5)
406
407#define BPF_CALL_x(x, name, ...) \
408 static __always_inline \
409 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
410 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
411 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
412 { \
413 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
414 } \
415 static __always_inline \
416 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
417
418#define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
419#define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
420#define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
421#define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
422#define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
423#define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
424
f96da094
DB
425#define bpf_ctx_range(TYPE, MEMBER) \
426 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
427#define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
428 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
429
430#define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
431 ({ \
432 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
433 *(PTR_SIZE) = (SIZE); \
434 offsetof(TYPE, MEMBER); \
435 })
436
bd4cf0ed
AS
437#ifdef CONFIG_COMPAT
438/* A struct sock_filter is architecture independent. */
0c5fe1b4
WD
439struct compat_sock_fprog {
440 u16 len;
bd4cf0ed 441 compat_uptr_t filter; /* struct sock_filter * */
0c5fe1b4
WD
442};
443#endif
444
a3ea269b
DB
445struct sock_fprog_kern {
446 u16 len;
447 struct sock_filter *filter;
448};
449
738cbe72
DB
450struct bpf_binary_header {
451 unsigned int pages;
452 u8 image[];
453};
454
7ae457c1 455struct bpf_prog {
286aad3c 456 u16 pages; /* Number of allocated pages */
a91263d5 457 u16 jited:1, /* Is our filter JIT'ed? */
65869a47 458 locked:1, /* Program image locked? */
c46646d0 459 gpl_compatible:1, /* Is filter GPL compatible? */
ff936a04 460 cb_access:1, /* Is control block accessed? */
f3edacbd 461 dst_needed:1; /* Do we need dst entry? */
24701ece 462 enum bpf_prog_type type; /* Type of BPF program */
7bd509e3 463 u32 len; /* Number of filter blocks */
783d28dd 464 u32 jited_len; /* Size of jited insns in bytes */
f1f7714e 465 u8 tag[BPF_TAG_SIZE];
09756af4 466 struct bpf_prog_aux *aux; /* Auxiliary fields */
24701ece 467 struct sock_fprog_kern *orig_prog; /* Original BPF program */
88575199
DB
468 unsigned int (*bpf_func)(const void *ctx,
469 const struct bpf_insn *insn);
60a3b225 470 /* Instructions for interpreter */
d45ed4a4 471 union {
bd4cf0ed 472 struct sock_filter insns[0];
2695fb55 473 struct bpf_insn insnsi[0];
d45ed4a4 474 };
b715631f
SH
475};
476
7ae457c1 477struct sk_filter {
4c355cdf 478 refcount_t refcnt;
7ae457c1
AS
479 struct rcu_head rcu;
480 struct bpf_prog *prog;
481};
482
324bda9e 483#define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
7ae457c1 484
01dd194c
DB
485#define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
486
db58ba45
AS
487struct bpf_skb_data_end {
488 struct qdisc_skb_cb qdisc_cb;
de8f3a83 489 void *data_meta;
db58ba45
AS
490 void *data_end;
491};
492
6a773a15
BB
493struct xdp_buff {
494 void *data;
495 void *data_end;
de8f3a83 496 void *data_meta;
17bedab2 497 void *data_hard_start;
6a773a15
BB
498};
499
6aaae2b6
DB
500/* Compute the linear packet data range [data, data_end) which
501 * will be accessed by various program types (cls_bpf, act_bpf,
502 * lwt, ...). Subsystems allowing direct data access must (!)
503 * ensure that cb[] area can be written to when BPF program is
504 * invoked (otherwise cb[] save/restore is necessary).
db58ba45 505 */
6aaae2b6 506static inline void bpf_compute_data_pointers(struct sk_buff *skb)
db58ba45
AS
507{
508 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
509
510 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
de8f3a83
DB
511 cb->data_meta = skb->data - skb_metadata_len(skb);
512 cb->data_end = skb->data + skb_headlen(skb);
db58ba45
AS
513}
514
01dd194c
DB
515static inline u8 *bpf_skb_cb(struct sk_buff *skb)
516{
517 /* eBPF programs may read/write skb->cb[] area to transfer meta
518 * data between tail calls. Since this also needs to work with
519 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
520 *
521 * In some socket filter cases, the cb unfortunately needs to be
522 * saved/restored so that protocol specific skb->cb[] data won't
523 * be lost. In any case, due to unpriviledged eBPF programs
524 * attached to sockets, we need to clear the bpf_skb_cb() area
525 * to not leak previous contents to user space.
526 */
527 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
528 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
529 FIELD_SIZEOF(struct qdisc_skb_cb, data));
530
531 return qdisc_skb_cb(skb)->data;
532}
533
ff936a04
AS
534static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
535 struct sk_buff *skb)
536{
01dd194c
DB
537 u8 *cb_data = bpf_skb_cb(skb);
538 u8 cb_saved[BPF_SKB_CB_LEN];
ff936a04
AS
539 u32 res;
540
ff936a04 541 if (unlikely(prog->cb_access)) {
01dd194c
DB
542 memcpy(cb_saved, cb_data, sizeof(cb_saved));
543 memset(cb_data, 0, sizeof(cb_saved));
ff936a04
AS
544 }
545
546 res = BPF_PROG_RUN(prog, skb);
547
548 if (unlikely(prog->cb_access))
01dd194c 549 memcpy(cb_data, cb_saved, sizeof(cb_saved));
ff936a04
AS
550
551 return res;
552}
553
554static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
555 struct sk_buff *skb)
556{
01dd194c 557 u8 *cb_data = bpf_skb_cb(skb);
ff936a04
AS
558
559 if (unlikely(prog->cb_access))
01dd194c
DB
560 memset(cb_data, 0, BPF_SKB_CB_LEN);
561
ff936a04
AS
562 return BPF_PROG_RUN(prog, skb);
563}
564
366cbf2f
DB
565static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
566 struct xdp_buff *xdp)
6a773a15 567{
366cbf2f
DB
568 /* Caller needs to hold rcu_read_lock() (!), otherwise program
569 * can be released while still running, or map elements could be
570 * freed early while still having concurrent users. XDP fastpath
571 * already takes rcu_read_lock() when fetching the program, so
572 * it's not necessary here anymore.
573 */
574 return BPF_PROG_RUN(prog, xdp);
6a773a15
BB
575}
576
aafe6ae9
DB
577static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
578{
579 return prog->len * sizeof(struct bpf_insn);
580}
581
f1f7714e 582static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
aafe6ae9
DB
583{
584 return round_up(bpf_prog_insn_size(prog) +
585 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
586}
587
7ae457c1 588static inline unsigned int bpf_prog_size(unsigned int proglen)
b715631f 589{
7ae457c1
AS
590 return max(sizeof(struct bpf_prog),
591 offsetof(struct bpf_prog, insns[proglen]));
b715631f
SH
592}
593
7b36f929
DB
594static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
595{
596 /* When classic BPF programs have been loaded and the arch
597 * does not have a classic BPF JIT (anymore), they have been
598 * converted via bpf_migrate_filter() to eBPF and thus always
599 * have an unspec program type.
600 */
601 return prog->type == BPF_PROG_TYPE_UNSPEC;
602}
603
f96da094
DB
604static inline bool
605bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
606{
607 bool off_ok;
608#ifdef __LITTLE_ENDIAN
609 off_ok = (off & (size_default - 1)) == 0;
610#else
611 off_ok = (off & (size_default - 1)) + size == size_default;
612#endif
613 return off_ok && size <= size_default && (size & (size - 1)) == 0;
614}
615
009937e7 616#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
a3ea269b 617
9d876e79 618#ifdef CONFIG_ARCH_HAS_SET_MEMORY
60a3b225
DB
619static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
620{
65869a47
DB
621 fp->locked = 1;
622 WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
60a3b225
DB
623}
624
625static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
626{
65869a47
DB
627 if (fp->locked) {
628 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
629 /* In case set_memory_rw() fails, we want to be the first
630 * to crash here instead of some random place later on.
631 */
632 fp->locked = 0;
633 }
60a3b225 634}
74451e66 635
9d876e79
DB
636static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
637{
65869a47 638 WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
9d876e79
DB
639}
640
74451e66
DB
641static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
642{
65869a47 643 WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
74451e66 644}
60a3b225
DB
645#else
646static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
647{
648}
649
650static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
651{
652}
74451e66 653
9d876e79
DB
654static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
655{
656}
657
74451e66
DB
658static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
659{
660}
9d876e79 661#endif /* CONFIG_ARCH_HAS_SET_MEMORY */
60a3b225 662
74451e66
DB
663static inline struct bpf_binary_header *
664bpf_jit_binary_hdr(const struct bpf_prog *fp)
665{
666 unsigned long real_start = (unsigned long)fp->bpf_func;
667 unsigned long addr = real_start & PAGE_MASK;
668
669 return (void *)addr;
670}
671
f4979fce
WB
672int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
673static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
674{
675 return sk_filter_trim_cap(sk, skb, 1);
676}
bd4cf0ed 677
d1c55ab5 678struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
7ae457c1 679void bpf_prog_free(struct bpf_prog *fp);
bd4cf0ed 680
60a3b225
DB
681struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
682struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
683 gfp_t gfp_extra_flags);
684void __bpf_prog_free(struct bpf_prog *fp);
685
686static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
687{
688 bpf_prog_unlock_ro(fp);
689 __bpf_prog_free(fp);
690}
691
ac67eb2c
DB
692typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
693 unsigned int flen);
694
7ae457c1 695int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
ac67eb2c 696int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
bab18991 697 bpf_aux_classic_check_t trans, bool save_orig);
7ae457c1 698void bpf_prog_destroy(struct bpf_prog *fp);
a3ea269b 699
fbc907f0 700int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
89aa0758 701int sk_attach_bpf(u32 ufd, struct sock *sk);
538950a1
CG
702int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
703int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
fbc907f0 704int sk_detach_filter(struct sock *sk);
fbc907f0
DB
705int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
706 unsigned int len);
fbc907f0 707
278571ba 708bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
fbc907f0 709void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
0a14842f 710
62258278 711u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
d1c55ab5
DB
712
713struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
9383191d 714void bpf_jit_compile(struct bpf_prog *prog);
17bedab2 715bool bpf_helper_changes_pkt_data(void *func);
62258278 716
c237ee5e
DB
717struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
718 const struct bpf_insn *patch, u32 len);
814abfab 719
11393cc9
JF
720/* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
721 * same cpu context. Further for best results no more than a single map
722 * for the do_redirect/do_flush pair should be used. This limitation is
723 * because we only track one map and force a flush when the map changes.
2ddf71e2 724 * This does not appear to be a real limitation for existing software.
11393cc9 725 */
2facaad6
JDB
726int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
727 struct bpf_prog *prog);
5acaee0a
JF
728int xdp_do_redirect(struct net_device *dev,
729 struct xdp_buff *xdp,
730 struct bpf_prog *prog);
11393cc9 731void xdp_do_flush_map(void);
814abfab 732
de8f3a83
DB
733/* Drivers not supporting XDP metadata can use this helper, which
734 * rejects any room expansion for metadata as a result.
735 */
736static __always_inline void
737xdp_set_data_meta_invalid(struct xdp_buff *xdp)
738{
739 xdp->data_meta = xdp->data + 1;
740}
741
742static __always_inline bool
743xdp_data_meta_unsupported(const struct xdp_buff *xdp)
744{
745 return unlikely(xdp->data_meta > xdp->data);
746}
747
6a773a15 748void bpf_warn_invalid_xdp_action(u32 act);
c237ee5e 749
34f79502 750struct sock *do_sk_redirect_map(struct sk_buff *skb);
174a79ff 751
b954d834 752#ifdef CONFIG_BPF_JIT
c94987e4 753extern int bpf_jit_enable;
4f3446bb 754extern int bpf_jit_harden;
74451e66 755extern int bpf_jit_kallsyms;
c94987e4 756
b954d834
DB
757typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
758
759struct bpf_binary_header *
760bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
761 unsigned int alignment,
762 bpf_jit_fill_hole_t bpf_fill_ill_insns);
763void bpf_jit_binary_free(struct bpf_binary_header *hdr);
764
b954d834
DB
765void bpf_jit_free(struct bpf_prog *fp);
766
4f3446bb
DB
767struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
768void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
769
b954d834
DB
770static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
771 u32 pass, void *image)
772{
b13138ef
DB
773 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
774 proglen, pass, image, current->comm, task_pid_nr(current));
775
b954d834
DB
776 if (image)
777 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
778 16, 1, image, proglen, false);
779}
4f3446bb
DB
780
781static inline bool bpf_jit_is_ebpf(void)
782{
783# ifdef CONFIG_HAVE_EBPF_JIT
784 return true;
785# else
786 return false;
787# endif
788}
789
81ed18ab
AS
790static inline bool ebpf_jit_enabled(void)
791{
792 return bpf_jit_enable && bpf_jit_is_ebpf();
793}
794
74451e66
DB
795static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
796{
797 return fp->jited && bpf_jit_is_ebpf();
798}
799
4f3446bb
DB
800static inline bool bpf_jit_blinding_enabled(void)
801{
802 /* These are the prerequisites, should someone ever have the
803 * idea to call blinding outside of them, we make sure to
804 * bail out.
805 */
806 if (!bpf_jit_is_ebpf())
807 return false;
808 if (!bpf_jit_enable)
809 return false;
810 if (!bpf_jit_harden)
811 return false;
812 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
813 return false;
814
815 return true;
816}
74451e66
DB
817
818static inline bool bpf_jit_kallsyms_enabled(void)
819{
820 /* There are a couple of corner cases where kallsyms should
821 * not be enabled f.e. on hardening.
822 */
823 if (bpf_jit_harden)
824 return false;
825 if (!bpf_jit_kallsyms)
826 return false;
827 if (bpf_jit_kallsyms == 1)
828 return true;
829
830 return false;
831}
832
833const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
834 unsigned long *off, char *sym);
835bool is_bpf_text_address(unsigned long addr);
836int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
837 char *sym);
838
839static inline const char *
840bpf_address_lookup(unsigned long addr, unsigned long *size,
841 unsigned long *off, char **modname, char *sym)
842{
843 const char *ret = __bpf_address_lookup(addr, size, off, sym);
844
845 if (ret && modname)
846 *modname = NULL;
847 return ret;
848}
849
850void bpf_prog_kallsyms_add(struct bpf_prog *fp);
851void bpf_prog_kallsyms_del(struct bpf_prog *fp);
852
853#else /* CONFIG_BPF_JIT */
854
81ed18ab
AS
855static inline bool ebpf_jit_enabled(void)
856{
857 return false;
858}
859
74451e66
DB
860static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
861{
862 return false;
863}
864
b954d834
DB
865static inline void bpf_jit_free(struct bpf_prog *fp)
866{
867 bpf_prog_unlock_free(fp);
868}
74451e66
DB
869
870static inline bool bpf_jit_kallsyms_enabled(void)
871{
872 return false;
873}
874
875static inline const char *
876__bpf_address_lookup(unsigned long addr, unsigned long *size,
877 unsigned long *off, char *sym)
878{
879 return NULL;
880}
881
882static inline bool is_bpf_text_address(unsigned long addr)
883{
884 return false;
885}
886
887static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
888 char *type, char *sym)
889{
890 return -ERANGE;
891}
892
893static inline const char *
894bpf_address_lookup(unsigned long addr, unsigned long *size,
895 unsigned long *off, char **modname, char *sym)
896{
897 return NULL;
898}
899
900static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
901{
902}
903
904static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
905{
906}
b954d834
DB
907#endif /* CONFIG_BPF_JIT */
908
34805931
DB
909#define BPF_ANC BIT(15)
910
55795ef5
RV
911static inline bool bpf_needs_clear_a(const struct sock_filter *first)
912{
913 switch (first->code) {
914 case BPF_RET | BPF_K:
915 case BPF_LD | BPF_W | BPF_LEN:
916 return false;
917
918 case BPF_LD | BPF_W | BPF_ABS:
919 case BPF_LD | BPF_H | BPF_ABS:
920 case BPF_LD | BPF_B | BPF_ABS:
921 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
922 return true;
923 return false;
924
925 default:
926 return true;
927 }
928}
929
34805931
DB
930static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
931{
932 BUG_ON(ftest->code & BPF_ANC);
933
934 switch (ftest->code) {
935 case BPF_LD | BPF_W | BPF_ABS:
936 case BPF_LD | BPF_H | BPF_ABS:
937 case BPF_LD | BPF_B | BPF_ABS:
938#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
939 return BPF_ANC | SKF_AD_##CODE
940 switch (ftest->k) {
941 BPF_ANCILLARY(PROTOCOL);
942 BPF_ANCILLARY(PKTTYPE);
943 BPF_ANCILLARY(IFINDEX);
944 BPF_ANCILLARY(NLATTR);
945 BPF_ANCILLARY(NLATTR_NEST);
946 BPF_ANCILLARY(MARK);
947 BPF_ANCILLARY(QUEUE);
948 BPF_ANCILLARY(HATYPE);
949 BPF_ANCILLARY(RXHASH);
950 BPF_ANCILLARY(CPU);
951 BPF_ANCILLARY(ALU_XOR_X);
952 BPF_ANCILLARY(VLAN_TAG);
953 BPF_ANCILLARY(VLAN_TAG_PRESENT);
954 BPF_ANCILLARY(PAY_OFFSET);
955 BPF_ANCILLARY(RANDOM);
27cd5452 956 BPF_ANCILLARY(VLAN_TPID);
34805931
DB
957 }
958 /* Fallthrough. */
959 default:
960 return ftest->code;
961 }
962}
963
9f12fbe6
ZSL
964void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
965 int k, unsigned int size);
966
967static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
968 unsigned int size, void *buffer)
969{
970 if (k >= 0)
971 return skb_header_pointer(skb, k, size, buffer);
972
973 return bpf_internal_load_pointer_neg_helper(skb, k, size);
974}
975
ea02f941
MS
976static inline int bpf_tell_extensions(void)
977{
37692299 978 return SKF_AD_MAX;
ea02f941
MS
979}
980
40304b2a
LB
981struct bpf_sock_ops_kern {
982 struct sock *sk;
983 u32 op;
984 union {
985 u32 reply;
986 u32 replylong[4];
987 };
f19397a5 988 u32 is_fullsock;
40304b2a
LB
989};
990
1da177e4 991#endif /* __LINUX_FILTER_H__ */