]>
Commit | Line | Data |
---|---|---|
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> |
820a0b24 | 19 | #include <linux/set_memory.h> |
7105e828 | 20 | #include <linux/kallsyms.h> |
6d5fc195 | 21 | #include <linux/if_vlan.h> |
d53d2f78 | 22 | #include <linux/vmalloc.h> |
b1ea9ff6 | 23 | #include <linux/sockptr.h> |
a24d22b2 | 24 | #include <crypto/sha1.h> |
700d4796 | 25 | #include <linux/u64_stats_sync.h> |
4f3446bb | 26 | |
ff936a04 | 27 | #include <net/sch_generic.h> |
b954d834 | 28 | |
d9b8aada | 29 | #include <asm/byteorder.h> |
b954d834 | 30 | #include <uapi/linux/filter.h> |
daedfb22 | 31 | #include <uapi/linux/bpf.h> |
60a3b225 DB |
32 | |
33 | struct sk_buff; | |
34 | struct sock; | |
35 | struct seccomp_data; | |
09756af4 | 36 | struct bpf_prog_aux; |
297dd12c | 37 | struct xdp_rxq_info; |
106ca27f | 38 | struct xdp_buff; |
2dbb9b9e | 39 | struct sock_reuseport; |
7b146ceb AI |
40 | struct ctl_table; |
41 | struct ctl_table_header; | |
792d4b5c | 42 | |
30743837 DB |
43 | /* ArgX, context and stack frame pointer register positions. Note, |
44 | * Arg1, Arg2, Arg3, etc are used as argument mappings of function | |
45 | * calls in BPF_CALL instruction. | |
46 | */ | |
47 | #define BPF_REG_ARG1 BPF_REG_1 | |
48 | #define BPF_REG_ARG2 BPF_REG_2 | |
49 | #define BPF_REG_ARG3 BPF_REG_3 | |
50 | #define BPF_REG_ARG4 BPF_REG_4 | |
51 | #define BPF_REG_ARG5 BPF_REG_5 | |
52 | #define BPF_REG_CTX BPF_REG_6 | |
53 | #define BPF_REG_FP BPF_REG_10 | |
54 | ||
55 | /* Additional register mappings for converted user programs. */ | |
56 | #define BPF_REG_A BPF_REG_0 | |
57 | #define BPF_REG_X BPF_REG_7 | |
e0cea7ce DB |
58 | #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ |
59 | #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ | |
60 | #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ | |
bd4cf0ed | 61 | |
9b73bfdd | 62 | /* Kernel hidden auxiliary/helper register. */ |
4f3446bb | 63 | #define BPF_REG_AX MAX_BPF_REG |
144cd91c DB |
64 | #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) |
65 | #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG | |
4f3446bb | 66 | |
71189fa9 AS |
67 | /* unused opcode to mark special call to bpf_tail_call() helper */ |
68 | #define BPF_TAIL_CALL 0xf0 | |
69 | ||
2a02759e AS |
70 | /* unused opcode to mark special load instruction. Same as BPF_ABS */ |
71 | #define BPF_PROBE_MEM 0x20 | |
72 | ||
1ea47e01 AS |
73 | /* unused opcode to mark call to interpreter with arguments */ |
74 | #define BPF_CALL_ARGS 0xe0 | |
75 | ||
74451e66 DB |
76 | /* As per nm, we expose JITed images as text (code) section for |
77 | * kallsyms. That way, tools like perf can find it to match | |
78 | * addresses. | |
79 | */ | |
80 | #define BPF_SYM_ELF_TYPE 't' | |
81 | ||
bd4cf0ed AS |
82 | /* BPF program can access up to 512 bytes of stack space. */ |
83 | #define MAX_BPF_STACK 512 | |
84 | ||
f8f6d679 DB |
85 | /* Helper macros for filter block array initializers. */ |
86 | ||
e430f34e | 87 | /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ |
f8f6d679 | 88 | |
e430f34e | 89 | #define BPF_ALU64_REG(OP, DST, SRC) \ |
2695fb55 | 90 | ((struct bpf_insn) { \ |
f8f6d679 | 91 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
92 | .dst_reg = DST, \ |
93 | .src_reg = SRC, \ | |
f8f6d679 DB |
94 | .off = 0, \ |
95 | .imm = 0 }) | |
96 | ||
e430f34e | 97 | #define BPF_ALU32_REG(OP, DST, SRC) \ |
2695fb55 | 98 | ((struct bpf_insn) { \ |
f8f6d679 | 99 | .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
100 | .dst_reg = DST, \ |
101 | .src_reg = SRC, \ | |
f8f6d679 DB |
102 | .off = 0, \ |
103 | .imm = 0 }) | |
104 | ||
e430f34e | 105 | /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ |
f8f6d679 | 106 | |
e430f34e | 107 | #define BPF_ALU64_IMM(OP, DST, IMM) \ |
2695fb55 | 108 | ((struct bpf_insn) { \ |
f8f6d679 | 109 | .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
110 | .dst_reg = DST, \ |
111 | .src_reg = 0, \ | |
f8f6d679 DB |
112 | .off = 0, \ |
113 | .imm = IMM }) | |
114 | ||
e430f34e | 115 | #define BPF_ALU32_IMM(OP, DST, IMM) \ |
2695fb55 | 116 | ((struct bpf_insn) { \ |
f8f6d679 | 117 | .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
118 | .dst_reg = DST, \ |
119 | .src_reg = 0, \ | |
f8f6d679 DB |
120 | .off = 0, \ |
121 | .imm = IMM }) | |
122 | ||
123 | /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ | |
124 | ||
e430f34e | 125 | #define BPF_ENDIAN(TYPE, DST, LEN) \ |
2695fb55 | 126 | ((struct bpf_insn) { \ |
f8f6d679 | 127 | .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ |
e430f34e AS |
128 | .dst_reg = DST, \ |
129 | .src_reg = 0, \ | |
f8f6d679 DB |
130 | .off = 0, \ |
131 | .imm = LEN }) | |
132 | ||
e430f34e | 133 | /* Short form of mov, dst_reg = src_reg */ |
f8f6d679 | 134 | |
e430f34e | 135 | #define BPF_MOV64_REG(DST, SRC) \ |
2695fb55 | 136 | ((struct bpf_insn) { \ |
f8f6d679 | 137 | .code = BPF_ALU64 | BPF_MOV | BPF_X, \ |
e430f34e AS |
138 | .dst_reg = DST, \ |
139 | .src_reg = SRC, \ | |
f8f6d679 DB |
140 | .off = 0, \ |
141 | .imm = 0 }) | |
142 | ||
e430f34e | 143 | #define BPF_MOV32_REG(DST, SRC) \ |
2695fb55 | 144 | ((struct bpf_insn) { \ |
f8f6d679 | 145 | .code = BPF_ALU | BPF_MOV | BPF_X, \ |
e430f34e AS |
146 | .dst_reg = DST, \ |
147 | .src_reg = SRC, \ | |
f8f6d679 DB |
148 | .off = 0, \ |
149 | .imm = 0 }) | |
150 | ||
e430f34e | 151 | /* Short form of mov, dst_reg = imm32 */ |
f8f6d679 | 152 | |
e430f34e | 153 | #define BPF_MOV64_IMM(DST, IMM) \ |
2695fb55 | 154 | ((struct bpf_insn) { \ |
f8f6d679 | 155 | .code = BPF_ALU64 | BPF_MOV | BPF_K, \ |
e430f34e AS |
156 | .dst_reg = DST, \ |
157 | .src_reg = 0, \ | |
f8f6d679 DB |
158 | .off = 0, \ |
159 | .imm = IMM }) | |
160 | ||
e430f34e | 161 | #define BPF_MOV32_IMM(DST, IMM) \ |
2695fb55 | 162 | ((struct bpf_insn) { \ |
f8f6d679 | 163 | .code = BPF_ALU | BPF_MOV | BPF_K, \ |
e430f34e AS |
164 | .dst_reg = DST, \ |
165 | .src_reg = 0, \ | |
f8f6d679 DB |
166 | .off = 0, \ |
167 | .imm = IMM }) | |
168 | ||
7d134041 JW |
169 | /* Special form of mov32, used for doing explicit zero extension on dst. */ |
170 | #define BPF_ZEXT_REG(DST) \ | |
171 | ((struct bpf_insn) { \ | |
172 | .code = BPF_ALU | BPF_MOV | BPF_X, \ | |
173 | .dst_reg = DST, \ | |
174 | .src_reg = DST, \ | |
175 | .off = 0, \ | |
176 | .imm = 1 }) | |
177 | ||
178 | static inline bool insn_is_zext(const struct bpf_insn *insn) | |
179 | { | |
180 | return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; | |
181 | } | |
182 | ||
02ab695b AS |
183 | /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ |
184 | #define BPF_LD_IMM64(DST, IMM) \ | |
185 | BPF_LD_IMM64_RAW(DST, 0, IMM) | |
186 | ||
187 | #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ | |
188 | ((struct bpf_insn) { \ | |
189 | .code = BPF_LD | BPF_DW | BPF_IMM, \ | |
190 | .dst_reg = DST, \ | |
191 | .src_reg = SRC, \ | |
192 | .off = 0, \ | |
193 | .imm = (__u32) (IMM) }), \ | |
194 | ((struct bpf_insn) { \ | |
195 | .code = 0, /* zero is reserved opcode */ \ | |
196 | .dst_reg = 0, \ | |
197 | .src_reg = 0, \ | |
198 | .off = 0, \ | |
199 | .imm = ((__u64) (IMM)) >> 32 }) | |
200 | ||
0246e64d AS |
201 | /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ |
202 | #define BPF_LD_MAP_FD(DST, MAP_FD) \ | |
203 | BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) | |
204 | ||
e430f34e | 205 | /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ |
f8f6d679 | 206 | |
e430f34e | 207 | #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 208 | ((struct bpf_insn) { \ |
f8f6d679 | 209 | .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
210 | .dst_reg = DST, \ |
211 | .src_reg = SRC, \ | |
f8f6d679 DB |
212 | .off = 0, \ |
213 | .imm = IMM }) | |
214 | ||
e430f34e | 215 | #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ |
2695fb55 | 216 | ((struct bpf_insn) { \ |
f8f6d679 | 217 | .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ |
e430f34e AS |
218 | .dst_reg = DST, \ |
219 | .src_reg = SRC, \ | |
f8f6d679 DB |
220 | .off = 0, \ |
221 | .imm = IMM }) | |
222 | ||
e430f34e | 223 | /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ |
f8f6d679 | 224 | |
e430f34e | 225 | #define BPF_LD_ABS(SIZE, IMM) \ |
2695fb55 | 226 | ((struct bpf_insn) { \ |
f8f6d679 | 227 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ |
e430f34e AS |
228 | .dst_reg = 0, \ |
229 | .src_reg = 0, \ | |
f8f6d679 | 230 | .off = 0, \ |
e430f34e | 231 | .imm = IMM }) |
f8f6d679 | 232 | |
e430f34e | 233 | /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ |
f8f6d679 | 234 | |
e430f34e | 235 | #define BPF_LD_IND(SIZE, SRC, IMM) \ |
2695fb55 | 236 | ((struct bpf_insn) { \ |
f8f6d679 | 237 | .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ |
e430f34e AS |
238 | .dst_reg = 0, \ |
239 | .src_reg = SRC, \ | |
f8f6d679 | 240 | .off = 0, \ |
e430f34e | 241 | .imm = IMM }) |
f8f6d679 | 242 | |
e430f34e | 243 | /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ |
f8f6d679 | 244 | |
e430f34e | 245 | #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ |
2695fb55 | 246 | ((struct bpf_insn) { \ |
f8f6d679 | 247 | .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
248 | .dst_reg = DST, \ |
249 | .src_reg = SRC, \ | |
f8f6d679 DB |
250 | .off = OFF, \ |
251 | .imm = 0 }) | |
252 | ||
e430f34e AS |
253 | /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ |
254 | ||
255 | #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ | |
2695fb55 | 256 | ((struct bpf_insn) { \ |
f8f6d679 | 257 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ |
e430f34e AS |
258 | .dst_reg = DST, \ |
259 | .src_reg = SRC, \ | |
f8f6d679 DB |
260 | .off = OFF, \ |
261 | .imm = 0 }) | |
262 | ||
cffc642d | 263 | |
91c960b0 BJ |
264 | /* |
265 | * Atomic operations: | |
266 | * | |
267 | * BPF_ADD *(uint *) (dst_reg + off16) += src_reg | |
981f94c3 BJ |
268 | * BPF_AND *(uint *) (dst_reg + off16) &= src_reg |
269 | * BPF_OR *(uint *) (dst_reg + off16) |= src_reg | |
270 | * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg | |
5ca419f2 | 271 | * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); |
981f94c3 BJ |
272 | * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); |
273 | * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); | |
274 | * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); | |
5ffa2550 BJ |
275 | * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) |
276 | * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) | |
91c960b0 BJ |
277 | */ |
278 | ||
279 | #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ | |
cffc642d | 280 | ((struct bpf_insn) { \ |
91c960b0 | 281 | .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ |
cffc642d MH |
282 | .dst_reg = DST, \ |
283 | .src_reg = SRC, \ | |
284 | .off = OFF, \ | |
91c960b0 BJ |
285 | .imm = OP }) |
286 | ||
287 | /* Legacy alias */ | |
288 | #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) | |
cffc642d | 289 | |
e430f34e AS |
290 | /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ |
291 | ||
292 | #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ | |
2695fb55 | 293 | ((struct bpf_insn) { \ |
e430f34e AS |
294 | .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ |
295 | .dst_reg = DST, \ | |
296 | .src_reg = 0, \ | |
297 | .off = OFF, \ | |
298 | .imm = IMM }) | |
299 | ||
300 | /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ | |
f8f6d679 | 301 | |
e430f34e | 302 | #define BPF_JMP_REG(OP, DST, SRC, OFF) \ |
2695fb55 | 303 | ((struct bpf_insn) { \ |
f8f6d679 | 304 | .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ |
e430f34e AS |
305 | .dst_reg = DST, \ |
306 | .src_reg = SRC, \ | |
f8f6d679 DB |
307 | .off = OFF, \ |
308 | .imm = 0 }) | |
309 | ||
e430f34e | 310 | /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ |
f8f6d679 | 311 | |
e430f34e | 312 | #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ |
2695fb55 | 313 | ((struct bpf_insn) { \ |
f8f6d679 | 314 | .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ |
e430f34e AS |
315 | .dst_reg = DST, \ |
316 | .src_reg = 0, \ | |
f8f6d679 | 317 | .off = OFF, \ |
a7b76c88 JW |
318 | .imm = IMM }) |
319 | ||
320 | /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ | |
321 | ||
322 | #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ | |
323 | ((struct bpf_insn) { \ | |
324 | .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ | |
325 | .dst_reg = DST, \ | |
326 | .src_reg = SRC, \ | |
327 | .off = OFF, \ | |
328 | .imm = 0 }) | |
329 | ||
330 | /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ | |
331 | ||
332 | #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ | |
333 | ((struct bpf_insn) { \ | |
334 | .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ | |
335 | .dst_reg = DST, \ | |
336 | .src_reg = 0, \ | |
337 | .off = OFF, \ | |
f8f6d679 DB |
338 | .imm = IMM }) |
339 | ||
614d0d77 DB |
340 | /* Unconditional jumps, goto pc + off16 */ |
341 | ||
342 | #define BPF_JMP_A(OFF) \ | |
343 | ((struct bpf_insn) { \ | |
344 | .code = BPF_JMP | BPF_JA, \ | |
345 | .dst_reg = 0, \ | |
346 | .src_reg = 0, \ | |
347 | .off = OFF, \ | |
348 | .imm = 0 }) | |
349 | ||
06be0864 DB |
350 | /* Relative call */ |
351 | ||
352 | #define BPF_CALL_REL(TGT) \ | |
353 | ((struct bpf_insn) { \ | |
354 | .code = BPF_JMP | BPF_CALL, \ | |
355 | .dst_reg = 0, \ | |
356 | .src_reg = BPF_PSEUDO_CALL, \ | |
357 | .off = 0, \ | |
358 | .imm = TGT }) | |
359 | ||
f8f6d679 DB |
360 | /* Function call */ |
361 | ||
09772d92 DB |
362 | #define BPF_CAST_CALL(x) \ |
363 | ((u64 (*)(u64, u64, u64, u64, u64))(x)) | |
364 | ||
f8f6d679 | 365 | #define BPF_EMIT_CALL(FUNC) \ |
2695fb55 | 366 | ((struct bpf_insn) { \ |
f8f6d679 | 367 | .code = BPF_JMP | BPF_CALL, \ |
e430f34e AS |
368 | .dst_reg = 0, \ |
369 | .src_reg = 0, \ | |
f8f6d679 DB |
370 | .off = 0, \ |
371 | .imm = ((FUNC) - __bpf_call_base) }) | |
372 | ||
373 | /* Raw code statement block */ | |
374 | ||
e430f34e | 375 | #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ |
2695fb55 | 376 | ((struct bpf_insn) { \ |
f8f6d679 | 377 | .code = CODE, \ |
e430f34e AS |
378 | .dst_reg = DST, \ |
379 | .src_reg = SRC, \ | |
f8f6d679 DB |
380 | .off = OFF, \ |
381 | .imm = IMM }) | |
382 | ||
383 | /* Program exit */ | |
384 | ||
385 | #define BPF_EXIT_INSN() \ | |
2695fb55 | 386 | ((struct bpf_insn) { \ |
f8f6d679 | 387 | .code = BPF_JMP | BPF_EXIT, \ |
e430f34e AS |
388 | .dst_reg = 0, \ |
389 | .src_reg = 0, \ | |
f8f6d679 DB |
390 | .off = 0, \ |
391 | .imm = 0 }) | |
392 | ||
a4afd37b DB |
393 | /* Internal classic blocks for direct assignment */ |
394 | ||
395 | #define __BPF_STMT(CODE, K) \ | |
396 | ((struct sock_filter) BPF_STMT(CODE, K)) | |
397 | ||
398 | #define __BPF_JUMP(CODE, K, JT, JF) \ | |
399 | ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) | |
400 | ||
f8f6d679 DB |
401 | #define bytes_to_bpf_size(bytes) \ |
402 | ({ \ | |
403 | int bpf_size = -EINVAL; \ | |
404 | \ | |
405 | if (bytes == sizeof(u8)) \ | |
406 | bpf_size = BPF_B; \ | |
407 | else if (bytes == sizeof(u16)) \ | |
408 | bpf_size = BPF_H; \ | |
409 | else if (bytes == sizeof(u32)) \ | |
410 | bpf_size = BPF_W; \ | |
411 | else if (bytes == sizeof(u64)) \ | |
412 | bpf_size = BPF_DW; \ | |
413 | \ | |
414 | bpf_size; \ | |
415 | }) | |
9739eef1 | 416 | |
f96da094 DB |
417 | #define bpf_size_to_bytes(bpf_size) \ |
418 | ({ \ | |
419 | int bytes = -EINVAL; \ | |
420 | \ | |
421 | if (bpf_size == BPF_B) \ | |
422 | bytes = sizeof(u8); \ | |
423 | else if (bpf_size == BPF_H) \ | |
424 | bytes = sizeof(u16); \ | |
425 | else if (bpf_size == BPF_W) \ | |
426 | bytes = sizeof(u32); \ | |
427 | else if (bpf_size == BPF_DW) \ | |
428 | bytes = sizeof(u64); \ | |
429 | \ | |
430 | bytes; \ | |
431 | }) | |
432 | ||
f035a515 DB |
433 | #define BPF_SIZEOF(type) \ |
434 | ({ \ | |
435 | const int __size = bytes_to_bpf_size(sizeof(type)); \ | |
436 | BUILD_BUG_ON(__size < 0); \ | |
437 | __size; \ | |
438 | }) | |
439 | ||
440 | #define BPF_FIELD_SIZEOF(type, field) \ | |
441 | ({ \ | |
c593642c | 442 | const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ |
f035a515 DB |
443 | BUILD_BUG_ON(__size < 0); \ |
444 | __size; \ | |
445 | }) | |
446 | ||
f96da094 DB |
447 | #define BPF_LDST_BYTES(insn) \ |
448 | ({ \ | |
e59ac634 | 449 | const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ |
f96da094 DB |
450 | WARN_ON(__size < 0); \ |
451 | __size; \ | |
452 | }) | |
453 | ||
f3694e00 DB |
454 | #define __BPF_MAP_0(m, v, ...) v |
455 | #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) | |
456 | #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) | |
457 | #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) | |
458 | #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) | |
459 | #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) | |
460 | ||
461 | #define __BPF_REG_0(...) __BPF_PAD(5) | |
462 | #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) | |
463 | #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) | |
464 | #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) | |
465 | #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) | |
466 | #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) | |
467 | ||
468 | #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) | |
469 | #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) | |
470 | ||
471 | #define __BPF_CAST(t, a) \ | |
472 | (__force t) \ | |
473 | (__force \ | |
474 | typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ | |
475 | (unsigned long)0, (t)0))) a | |
476 | #define __BPF_V void | |
477 | #define __BPF_N | |
478 | ||
479 | #define __BPF_DECL_ARGS(t, a) t a | |
480 | #define __BPF_DECL_REGS(t, a) u64 a | |
481 | ||
482 | #define __BPF_PAD(n) \ | |
483 | __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ | |
484 | u64, __ur_3, u64, __ur_4, u64, __ur_5) | |
485 | ||
486 | #define BPF_CALL_x(x, name, ...) \ | |
487 | static __always_inline \ | |
488 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ | |
7c6a469e | 489 | typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ |
f3694e00 DB |
490 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ |
491 | u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ | |
492 | { \ | |
7c6a469e | 493 | return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ |
f3694e00 DB |
494 | } \ |
495 | static __always_inline \ | |
496 | u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) | |
497 | ||
498 | #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) | |
499 | #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) | |
500 | #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) | |
501 | #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) | |
502 | #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) | |
503 | #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) | |
504 | ||
f96da094 DB |
505 | #define bpf_ctx_range(TYPE, MEMBER) \ |
506 | offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 | |
507 | #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ | |
508 | offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 | |
b7df9ada DB |
509 | #if BITS_PER_LONG == 64 |
510 | # define bpf_ctx_range_ptr(TYPE, MEMBER) \ | |
511 | offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 | |
512 | #else | |
513 | # define bpf_ctx_range_ptr(TYPE, MEMBER) \ | |
514 | offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 | |
515 | #endif /* BITS_PER_LONG == 64 */ | |
f96da094 DB |
516 | |
517 | #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ | |
518 | ({ \ | |
c593642c | 519 | BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ |
f96da094 DB |
520 | *(PTR_SIZE) = (SIZE); \ |
521 | offsetof(TYPE, MEMBER); \ | |
522 | }) | |
523 | ||
bd4cf0ed | 524 | /* A struct sock_filter is architecture independent. */ |
0c5fe1b4 WD |
525 | struct compat_sock_fprog { |
526 | u16 len; | |
bd4cf0ed | 527 | compat_uptr_t filter; /* struct sock_filter * */ |
0c5fe1b4 | 528 | }; |
0c5fe1b4 | 529 | |
a3ea269b DB |
530 | struct sock_fprog_kern { |
531 | u16 len; | |
532 | struct sock_filter *filter; | |
533 | }; | |
534 | ||
b7b3fc8d IL |
535 | /* Some arches need doubleword alignment for their instructions and/or data */ |
536 | #define BPF_IMAGE_ALIGNMENT 8 | |
537 | ||
738cbe72 | 538 | struct bpf_binary_header { |
85782e03 | 539 | u32 pages; |
b7b3fc8d | 540 | u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); |
738cbe72 DB |
541 | }; |
542 | ||
700d4796 AS |
543 | struct bpf_prog_stats { |
544 | u64 cnt; | |
545 | u64 nsecs; | |
9ed9e9ba | 546 | u64 misses; |
700d4796 AS |
547 | struct u64_stats_sync syncp; |
548 | } __aligned(2 * sizeof(u64)); | |
549 | ||
7ae457c1 | 550 | struct bpf_prog { |
286aad3c | 551 | u16 pages; /* Number of allocated pages */ |
a91263d5 | 552 | u16 jited:1, /* Is our filter JIT'ed? */ |
60b58afc | 553 | jit_requested:1,/* archs need to JIT the prog */ |
c46646d0 | 554 | gpl_compatible:1, /* Is filter GPL compatible? */ |
ff936a04 | 555 | cb_access:1, /* Is control block accessed? */ |
9802d865 | 556 | dst_needed:1, /* Do we need dst entry? */ |
1c2a088a AS |
557 | blinded:1, /* Was blinded */ |
558 | is_func:1, /* program is a bpf function */ | |
c195651e | 559 | kprobe_override:1, /* Do we override a kprobe? */ |
5cf1e914 | 560 | has_callchain_buf:1, /* callchain buffer allocated? */ |
5d99cb2c SL |
561 | enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */ |
562 | call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */ | |
24701ece | 563 | enum bpf_prog_type type; /* Type of BPF program */ |
5e43f899 | 564 | enum bpf_attach_type expected_attach_type; /* For some prog types */ |
7bd509e3 | 565 | u32 len; /* Number of filter blocks */ |
783d28dd | 566 | u32 jited_len; /* Size of jited insns in bytes */ |
f1f7714e | 567 | u8 tag[BPF_TAG_SIZE]; |
700d4796 | 568 | struct bpf_prog_stats __percpu *stats; |
ca06f55b | 569 | int __percpu *active; |
88575199 DB |
570 | unsigned int (*bpf_func)(const void *ctx, |
571 | const struct bpf_insn *insn); | |
700d4796 AS |
572 | struct bpf_prog_aux *aux; /* Auxiliary fields */ |
573 | struct sock_fprog_kern *orig_prog; /* Original BPF program */ | |
60a3b225 | 574 | /* Instructions for interpreter */ |
d26c0cc5 AB |
575 | struct sock_filter insns[0]; |
576 | struct bpf_insn insnsi[]; | |
b715631f SH |
577 | }; |
578 | ||
7ae457c1 | 579 | struct sk_filter { |
4c355cdf | 580 | refcount_t refcnt; |
7ae457c1 AS |
581 | struct rcu_head rcu; |
582 | struct bpf_prog *prog; | |
583 | }; | |
584 | ||
492ecee8 AS |
585 | DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); |
586 | ||
7e6897f9 | 587 | #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \ |
343a3e8b | 588 | u32 __ret; \ |
37e1d920 | 589 | cant_migrate(); \ |
7e6897f9 | 590 | if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ |
343a3e8b AB |
591 | struct bpf_prog_stats *__stats; \ |
592 | u64 __start = sched_clock(); \ | |
593 | __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ | |
700d4796 | 594 | __stats = this_cpu_ptr(prog->stats); \ |
343a3e8b AB |
595 | u64_stats_update_begin(&__stats->syncp); \ |
596 | __stats->cnt++; \ | |
597 | __stats->nsecs += sched_clock() - __start; \ | |
598 | u64_stats_update_end(&__stats->syncp); \ | |
7e6897f9 | 599 | } else { \ |
343a3e8b | 600 | __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \ |
7e6897f9 | 601 | } \ |
343a3e8b | 602 | __ret; }) |
7ae457c1 | 603 | |
3c58482a | 604 | #define BPF_PROG_RUN(prog, ctx) \ |
6a64037d | 605 | __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func) |
3c58482a TG |
606 | |
607 | /* | |
608 | * Use in preemptible and therefore migratable context to make sure that | |
609 | * the execution of the BPF program runs on one CPU. | |
610 | * | |
611 | * This uses migrate_disable/enable() explicitly to document that the | |
612 | * invocation of a BPF program does not require reentrancy protection | |
613 | * against a BPF program which is invoked from a preempting task. | |
614 | * | |
615 | * For non RT enabled kernels migrate_disable/enable() maps to | |
616 | * preempt_disable/enable(), i.e. it disables also preemption. | |
617 | */ | |
618 | static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, | |
619 | const void *ctx) | |
620 | { | |
621 | u32 ret; | |
622 | ||
623 | migrate_disable(); | |
6a64037d | 624 | ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func); |
3c58482a TG |
625 | migrate_enable(); |
626 | return ret; | |
627 | } | |
7e6897f9 | 628 | |
01dd194c DB |
629 | #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN |
630 | ||
db58ba45 AS |
631 | struct bpf_skb_data_end { |
632 | struct qdisc_skb_cb qdisc_cb; | |
de8f3a83 | 633 | void *data_meta; |
db58ba45 AS |
634 | void *data_end; |
635 | }; | |
636 | ||
ba452c9e THJ |
637 | struct bpf_nh_params { |
638 | u32 nh_family; | |
639 | union { | |
640 | u32 ipv4_nh; | |
641 | struct in6_addr ipv6_nh; | |
642 | }; | |
643 | }; | |
644 | ||
0b19cc0a | 645 | struct bpf_redirect_info { |
0b19cc0a | 646 | u32 flags; |
4b55cf29 | 647 | u32 tgt_index; |
43e74c02 | 648 | void *tgt_value; |
ee75aef2 BT |
649 | u32 map_id; |
650 | enum bpf_map_type map_type; | |
2539650f | 651 | u32 kern_flags; |
ba452c9e | 652 | struct bpf_nh_params nh; |
0b19cc0a TM |
653 | }; |
654 | ||
655 | DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); | |
656 | ||
2539650f TM |
657 | /* flags for bpf_redirect_info kern_flags */ |
658 | #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ | |
659 | ||
6aaae2b6 DB |
660 | /* Compute the linear packet data range [data, data_end) which |
661 | * will be accessed by various program types (cls_bpf, act_bpf, | |
662 | * lwt, ...). Subsystems allowing direct data access must (!) | |
663 | * ensure that cb[] area can be written to when BPF program is | |
664 | * invoked (otherwise cb[] save/restore is necessary). | |
db58ba45 | 665 | */ |
6aaae2b6 | 666 | static inline void bpf_compute_data_pointers(struct sk_buff *skb) |
db58ba45 AS |
667 | { |
668 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; | |
669 | ||
c593642c | 670 | BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); |
de8f3a83 DB |
671 | cb->data_meta = skb->data - skb_metadata_len(skb); |
672 | cb->data_end = skb->data + skb_headlen(skb); | |
db58ba45 AS |
673 | } |
674 | ||
b39b5f41 SL |
675 | /* Similar to bpf_compute_data_pointers(), except that save orginal |
676 | * data in cb->data and cb->meta_data for restore. | |
677 | */ | |
678 | static inline void bpf_compute_and_save_data_end( | |
679 | struct sk_buff *skb, void **saved_data_end) | |
680 | { | |
681 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; | |
682 | ||
683 | *saved_data_end = cb->data_end; | |
684 | cb->data_end = skb->data + skb_headlen(skb); | |
685 | } | |
686 | ||
687 | /* Restore data saved by bpf_compute_data_pointers(). */ | |
688 | static inline void bpf_restore_data_end( | |
689 | struct sk_buff *skb, void *saved_data_end) | |
690 | { | |
691 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; | |
692 | ||
693 | cb->data_end = saved_data_end; | |
694 | } | |
695 | ||
01dd194c DB |
696 | static inline u8 *bpf_skb_cb(struct sk_buff *skb) |
697 | { | |
698 | /* eBPF programs may read/write skb->cb[] area to transfer meta | |
699 | * data between tail calls. Since this also needs to work with | |
700 | * tc, that scratch memory is mapped to qdisc_skb_cb's data area. | |
701 | * | |
702 | * In some socket filter cases, the cb unfortunately needs to be | |
703 | * saved/restored so that protocol specific skb->cb[] data won't | |
704 | * be lost. In any case, due to unpriviledged eBPF programs | |
705 | * attached to sockets, we need to clear the bpf_skb_cb() area | |
706 | * to not leak previous contents to user space. | |
707 | */ | |
c593642c PB |
708 | BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); |
709 | BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != | |
710 | sizeof_field(struct qdisc_skb_cb, data)); | |
01dd194c DB |
711 | |
712 | return qdisc_skb_cb(skb)->data; | |
713 | } | |
714 | ||
2a916f2f | 715 | /* Must be invoked with migration disabled */ |
6cab5e90 AS |
716 | static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, |
717 | struct sk_buff *skb) | |
ff936a04 | 718 | { |
01dd194c DB |
719 | u8 *cb_data = bpf_skb_cb(skb); |
720 | u8 cb_saved[BPF_SKB_CB_LEN]; | |
ff936a04 AS |
721 | u32 res; |
722 | ||
ff936a04 | 723 | if (unlikely(prog->cb_access)) { |
01dd194c DB |
724 | memcpy(cb_saved, cb_data, sizeof(cb_saved)); |
725 | memset(cb_data, 0, sizeof(cb_saved)); | |
ff936a04 AS |
726 | } |
727 | ||
728 | res = BPF_PROG_RUN(prog, skb); | |
729 | ||
730 | if (unlikely(prog->cb_access)) | |
01dd194c | 731 | memcpy(cb_data, cb_saved, sizeof(cb_saved)); |
ff936a04 AS |
732 | |
733 | return res; | |
734 | } | |
735 | ||
6cab5e90 AS |
736 | static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, |
737 | struct sk_buff *skb) | |
738 | { | |
739 | u32 res; | |
740 | ||
2a916f2f | 741 | migrate_disable(); |
6cab5e90 | 742 | res = __bpf_prog_run_save_cb(prog, skb); |
2a916f2f | 743 | migrate_enable(); |
6cab5e90 AS |
744 | return res; |
745 | } | |
746 | ||
ff936a04 AS |
747 | static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, |
748 | struct sk_buff *skb) | |
749 | { | |
01dd194c | 750 | u8 *cb_data = bpf_skb_cb(skb); |
6cab5e90 | 751 | u32 res; |
ff936a04 AS |
752 | |
753 | if (unlikely(prog->cb_access)) | |
01dd194c DB |
754 | memset(cb_data, 0, BPF_SKB_CB_LEN); |
755 | ||
3d9f773c | 756 | res = bpf_prog_run_pin_on_cpu(prog, skb); |
6cab5e90 | 757 | return res; |
ff936a04 AS |
758 | } |
759 | ||
6a64037d | 760 | DECLARE_BPF_DISPATCHER(xdp) |
7e6897f9 | 761 | |
366cbf2f DB |
762 | static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, |
763 | struct xdp_buff *xdp) | |
6a773a15 | 764 | { |
366cbf2f DB |
765 | /* Caller needs to hold rcu_read_lock() (!), otherwise program |
766 | * can be released while still running, or map elements could be | |
767 | * freed early while still having concurrent users. XDP fastpath | |
768 | * already takes rcu_read_lock() when fetching the program, so | |
769 | * it's not necessary here anymore. | |
770 | */ | |
6a64037d | 771 | return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); |
6a773a15 BB |
772 | } |
773 | ||
7e6897f9 BT |
774 | void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); |
775 | ||
aafe6ae9 DB |
776 | static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) |
777 | { | |
778 | return prog->len * sizeof(struct bpf_insn); | |
779 | } | |
780 | ||
f1f7714e | 781 | static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) |
aafe6ae9 DB |
782 | { |
783 | return round_up(bpf_prog_insn_size(prog) + | |
6b0b0fa2 | 784 | sizeof(__be64) + 1, SHA1_BLOCK_SIZE); |
aafe6ae9 DB |
785 | } |
786 | ||
7ae457c1 | 787 | static inline unsigned int bpf_prog_size(unsigned int proglen) |
b715631f | 788 | { |
7ae457c1 AS |
789 | return max(sizeof(struct bpf_prog), |
790 | offsetof(struct bpf_prog, insns[proglen])); | |
b715631f SH |
791 | } |
792 | ||
7b36f929 DB |
793 | static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) |
794 | { | |
795 | /* When classic BPF programs have been loaded and the arch | |
796 | * does not have a classic BPF JIT (anymore), they have been | |
797 | * converted via bpf_migrate_filter() to eBPF and thus always | |
798 | * have an unspec program type. | |
799 | */ | |
800 | return prog->type == BPF_PROG_TYPE_UNSPEC; | |
801 | } | |
802 | ||
bc23105c DB |
803 | static inline u32 bpf_ctx_off_adjust_machine(u32 size) |
804 | { | |
805 | const u32 size_machine = sizeof(unsigned long); | |
806 | ||
807 | if (size > size_machine && size % size_machine == 0) | |
808 | size = size_machine; | |
809 | ||
810 | return size; | |
811 | } | |
812 | ||
bc23105c DB |
813 | static inline bool |
814 | bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) | |
815 | { | |
46f53a65 | 816 | return size <= size_default && (size & (size - 1)) == 0; |
f96da094 DB |
817 | } |
818 | ||
d9b8aada | 819 | static inline u8 |
d895a0f1 | 820 | bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) |
d9b8aada | 821 | { |
d895a0f1 | 822 | u8 access_off = off & (size_default - 1); |
d9b8aada IL |
823 | |
824 | #ifdef __LITTLE_ENDIAN | |
d895a0f1 | 825 | return access_off; |
d9b8aada | 826 | #else |
d895a0f1 | 827 | return size_default - (access_off + size); |
d9b8aada IL |
828 | #endif |
829 | } | |
830 | ||
b4399546 | 831 | #define bpf_ctx_wide_access_ok(off, size, type, field) \ |
600c70ba SF |
832 | (size == sizeof(__u64) && \ |
833 | off >= offsetof(type, field) && \ | |
834 | off + sizeof(__u64) <= offsetofend(type, field) && \ | |
835 | off % sizeof(__u64) == 0) | |
836 | ||
009937e7 | 837 | #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) |
a3ea269b | 838 | |
60a3b225 DB |
839 | static inline void bpf_prog_lock_ro(struct bpf_prog *fp) |
840 | { | |
e1608f3f DB |
841 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
842 | if (!fp->jited) { | |
843 | set_vm_flush_reset_perms(fp); | |
844 | set_memory_ro((unsigned long)fp, fp->pages); | |
845 | } | |
846 | #endif | |
60a3b225 DB |
847 | } |
848 | ||
9d876e79 DB |
849 | static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
850 | { | |
d53d2f78 | 851 | set_vm_flush_reset_perms(hdr); |
85782e03 | 852 | set_memory_ro((unsigned long)hdr, hdr->pages); |
f2c65fb3 | 853 | set_memory_x((unsigned long)hdr, hdr->pages); |
9d876e79 DB |
854 | } |
855 | ||
74451e66 DB |
856 | static inline struct bpf_binary_header * |
857 | bpf_jit_binary_hdr(const struct bpf_prog *fp) | |
858 | { | |
859 | unsigned long real_start = (unsigned long)fp->bpf_func; | |
860 | unsigned long addr = real_start & PAGE_MASK; | |
861 | ||
862 | return (void *)addr; | |
863 | } | |
864 | ||
f4979fce WB |
865 | int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); |
866 | static inline int sk_filter(struct sock *sk, struct sk_buff *skb) | |
867 | { | |
868 | return sk_filter_trim_cap(sk, skb, 1); | |
869 | } | |
bd4cf0ed | 870 | |
d1c55ab5 | 871 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); |
7ae457c1 | 872 | void bpf_prog_free(struct bpf_prog *fp); |
bd4cf0ed | 873 | |
5e581dad DB |
874 | bool bpf_opcode_in_insntable(u8 code); |
875 | ||
c454a46b MKL |
876 | void bpf_prog_free_linfo(struct bpf_prog *prog); |
877 | void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, | |
878 | const u32 *insn_to_jit_off); | |
879 | int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); | |
e16301fb | 880 | void bpf_prog_jit_attempt_done(struct bpf_prog *prog); |
c454a46b | 881 | |
60a3b225 | 882 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); |
492ecee8 | 883 | struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); |
60a3b225 DB |
884 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, |
885 | gfp_t gfp_extra_flags); | |
886 | void __bpf_prog_free(struct bpf_prog *fp); | |
887 | ||
888 | static inline void bpf_prog_unlock_free(struct bpf_prog *fp) | |
889 | { | |
60a3b225 DB |
890 | __bpf_prog_free(fp); |
891 | } | |
892 | ||
ac67eb2c DB |
893 | typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, |
894 | unsigned int flen); | |
895 | ||
7ae457c1 | 896 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); |
ac67eb2c | 897 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, |
bab18991 | 898 | bpf_aux_classic_check_t trans, bool save_orig); |
7ae457c1 | 899 | void bpf_prog_destroy(struct bpf_prog *fp); |
a3ea269b | 900 | |
fbc907f0 | 901 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
89aa0758 | 902 | int sk_attach_bpf(u32 ufd, struct sock *sk); |
538950a1 CG |
903 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
904 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); | |
8217ca65 | 905 | void sk_reuseport_prog_free(struct bpf_prog *prog); |
fbc907f0 | 906 | int sk_detach_filter(struct sock *sk); |
fbc907f0 DB |
907 | int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, |
908 | unsigned int len); | |
fbc907f0 | 909 | |
278571ba | 910 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); |
fbc907f0 | 911 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); |
0a14842f | 912 | |
62258278 | 913 | u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
1ea47e01 AS |
914 | #define __bpf_call_base_args \ |
915 | ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ | |
6943c2b0 | 916 | (void *)__bpf_call_base) |
d1c55ab5 DB |
917 | |
918 | struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); | |
9383191d | 919 | void bpf_jit_compile(struct bpf_prog *prog); |
a4b1d3c1 | 920 | bool bpf_jit_needs_zext(void); |
e6ac2450 | 921 | bool bpf_jit_supports_kfunc_call(void); |
17bedab2 | 922 | bool bpf_helper_changes_pkt_data(void *func); |
62258278 | 923 | |
63960260 | 924 | static inline bool bpf_dump_raw_ok(const struct cred *cred) |
7105e828 DB |
925 | { |
926 | /* Reconstruction of call-sites is dependent on kallsyms, | |
927 | * thus make dump the same restriction. | |
928 | */ | |
63960260 | 929 | return kallsyms_show_value(cred); |
7105e828 DB |
930 | } |
931 | ||
c237ee5e DB |
932 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
933 | const struct bpf_insn *patch, u32 len); | |
52875a04 | 934 | int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); |
814abfab | 935 | |
f6069b9a DB |
936 | void bpf_clear_redirect_map(struct bpf_map *map); |
937 | ||
2539650f TM |
938 | static inline bool xdp_return_frame_no_direct(void) |
939 | { | |
940 | struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); | |
941 | ||
942 | return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; | |
943 | } | |
944 | ||
945 | static inline void xdp_set_return_frame_no_direct(void) | |
946 | { | |
947 | struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); | |
948 | ||
949 | ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; | |
950 | } | |
951 | ||
952 | static inline void xdp_clear_return_frame_no_direct(void) | |
953 | { | |
954 | struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); | |
955 | ||
956 | ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; | |
957 | } | |
958 | ||
d8d7218a TM |
959 | static inline int xdp_ok_fwd_dev(const struct net_device *fwd, |
960 | unsigned int pktlen) | |
6d5fc195 TM |
961 | { |
962 | unsigned int len; | |
963 | ||
964 | if (unlikely(!(fwd->flags & IFF_UP))) | |
965 | return -ENETDOWN; | |
966 | ||
967 | len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; | |
d8d7218a | 968 | if (pktlen > len) |
6d5fc195 TM |
969 | return -EMSGSIZE; |
970 | ||
971 | return 0; | |
972 | } | |
973 | ||
1d233886 | 974 | /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the |
11393cc9 JF |
975 | * same cpu context. Further for best results no more than a single map |
976 | * for the do_redirect/do_flush pair should be used. This limitation is | |
977 | * because we only track one map and force a flush when the map changes. | |
2ddf71e2 | 978 | * This does not appear to be a real limitation for existing software. |
11393cc9 | 979 | */ |
2facaad6 | 980 | int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, |
02671e23 | 981 | struct xdp_buff *xdp, struct bpf_prog *prog); |
5acaee0a JF |
982 | int xdp_do_redirect(struct net_device *dev, |
983 | struct xdp_buff *xdp, | |
984 | struct bpf_prog *prog); | |
1d233886 THJ |
985 | void xdp_do_flush(void); |
986 | ||
987 | /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as | |
988 | * it is no longer only flushing maps. Keep this define for compatibility | |
989 | * until all drivers are updated - do not use xdp_do_flush_map() in new code! | |
990 | */ | |
991 | #define xdp_do_flush_map xdp_do_flush | |
814abfab | 992 | |
6a773a15 | 993 | void bpf_warn_invalid_xdp_action(u32 act); |
c237ee5e | 994 | |
2dbb9b9e MKL |
995 | #ifdef CONFIG_INET |
996 | struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, | |
997 | struct bpf_prog *prog, struct sk_buff *skb, | |
998 | u32 hash); | |
999 | #else | |
1000 | static inline struct sock * | |
1001 | bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, | |
1002 | struct bpf_prog *prog, struct sk_buff *skb, | |
1003 | u32 hash) | |
1004 | { | |
1005 | return NULL; | |
1006 | } | |
1007 | #endif | |
1008 | ||
b954d834 | 1009 | #ifdef CONFIG_BPF_JIT |
c94987e4 | 1010 | extern int bpf_jit_enable; |
4f3446bb | 1011 | extern int bpf_jit_harden; |
74451e66 | 1012 | extern int bpf_jit_kallsyms; |
fdadd049 | 1013 | extern long bpf_jit_limit; |
c94987e4 | 1014 | |
b954d834 DB |
1015 | typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); |
1016 | ||
1017 | struct bpf_binary_header * | |
1018 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, | |
1019 | unsigned int alignment, | |
1020 | bpf_jit_fill_hole_t bpf_fill_ill_insns); | |
1021 | void bpf_jit_binary_free(struct bpf_binary_header *hdr); | |
116bfa96 VK |
1022 | u64 bpf_jit_alloc_exec_limit(void); |
1023 | void *bpf_jit_alloc_exec(unsigned long size); | |
1024 | void bpf_jit_free_exec(void *addr); | |
b954d834 DB |
1025 | void bpf_jit_free(struct bpf_prog *fp); |
1026 | ||
a66886fe DB |
1027 | int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, |
1028 | struct bpf_jit_poke_descriptor *poke); | |
1029 | ||
e2c95a61 DB |
1030 | int bpf_jit_get_func_addr(const struct bpf_prog *prog, |
1031 | const struct bpf_insn *insn, bool extra_pass, | |
1032 | u64 *func_addr, bool *func_addr_fixed); | |
1033 | ||
4f3446bb DB |
1034 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); |
1035 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); | |
1036 | ||
b954d834 DB |
1037 | static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, |
1038 | u32 pass, void *image) | |
1039 | { | |
b13138ef DB |
1040 | pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, |
1041 | proglen, pass, image, current->comm, task_pid_nr(current)); | |
1042 | ||
b954d834 DB |
1043 | if (image) |
1044 | print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, | |
1045 | 16, 1, image, proglen, false); | |
1046 | } | |
4f3446bb DB |
1047 | |
1048 | static inline bool bpf_jit_is_ebpf(void) | |
1049 | { | |
1050 | # ifdef CONFIG_HAVE_EBPF_JIT | |
1051 | return true; | |
1052 | # else | |
1053 | return false; | |
1054 | # endif | |
1055 | } | |
1056 | ||
81ed18ab AS |
1057 | static inline bool ebpf_jit_enabled(void) |
1058 | { | |
1059 | return bpf_jit_enable && bpf_jit_is_ebpf(); | |
1060 | } | |
1061 | ||
74451e66 DB |
1062 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
1063 | { | |
1064 | return fp->jited && bpf_jit_is_ebpf(); | |
1065 | } | |
1066 | ||
60b58afc | 1067 | static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) |
4f3446bb DB |
1068 | { |
1069 | /* These are the prerequisites, should someone ever have the | |
1070 | * idea to call blinding outside of them, we make sure to | |
1071 | * bail out. | |
1072 | */ | |
1073 | if (!bpf_jit_is_ebpf()) | |
1074 | return false; | |
60b58afc | 1075 | if (!prog->jit_requested) |
4f3446bb DB |
1076 | return false; |
1077 | if (!bpf_jit_harden) | |
1078 | return false; | |
1079 | if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) | |
1080 | return false; | |
1081 | ||
1082 | return true; | |
1083 | } | |
74451e66 DB |
1084 | |
1085 | static inline bool bpf_jit_kallsyms_enabled(void) | |
1086 | { | |
1087 | /* There are a couple of corner cases where kallsyms should | |
1088 | * not be enabled f.e. on hardening. | |
1089 | */ | |
1090 | if (bpf_jit_harden) | |
1091 | return false; | |
1092 | if (!bpf_jit_kallsyms) | |
1093 | return false; | |
1094 | if (bpf_jit_kallsyms == 1) | |
1095 | return true; | |
1096 | ||
1097 | return false; | |
1098 | } | |
1099 | ||
1100 | const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, | |
1101 | unsigned long *off, char *sym); | |
1102 | bool is_bpf_text_address(unsigned long addr); | |
1103 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
1104 | char *sym); | |
1105 | ||
1106 | static inline const char * | |
1107 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
1108 | unsigned long *off, char **modname, char *sym) | |
1109 | { | |
1110 | const char *ret = __bpf_address_lookup(addr, size, off, sym); | |
1111 | ||
1112 | if (ret && modname) | |
1113 | *modname = NULL; | |
1114 | return ret; | |
1115 | } | |
1116 | ||
1117 | void bpf_prog_kallsyms_add(struct bpf_prog *fp); | |
1118 | void bpf_prog_kallsyms_del(struct bpf_prog *fp); | |
1119 | ||
1120 | #else /* CONFIG_BPF_JIT */ | |
1121 | ||
81ed18ab AS |
1122 | static inline bool ebpf_jit_enabled(void) |
1123 | { | |
1124 | return false; | |
1125 | } | |
1126 | ||
b8cd76ca DB |
1127 | static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) |
1128 | { | |
1129 | return false; | |
1130 | } | |
1131 | ||
74451e66 DB |
1132 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
1133 | { | |
1134 | return false; | |
1135 | } | |
1136 | ||
a66886fe DB |
1137 | static inline int |
1138 | bpf_jit_add_poke_descriptor(struct bpf_prog *prog, | |
1139 | struct bpf_jit_poke_descriptor *poke) | |
1140 | { | |
1141 | return -ENOTSUPP; | |
1142 | } | |
1143 | ||
b954d834 DB |
1144 | static inline void bpf_jit_free(struct bpf_prog *fp) |
1145 | { | |
1146 | bpf_prog_unlock_free(fp); | |
1147 | } | |
74451e66 DB |
1148 | |
1149 | static inline bool bpf_jit_kallsyms_enabled(void) | |
1150 | { | |
1151 | return false; | |
1152 | } | |
1153 | ||
1154 | static inline const char * | |
1155 | __bpf_address_lookup(unsigned long addr, unsigned long *size, | |
1156 | unsigned long *off, char *sym) | |
1157 | { | |
1158 | return NULL; | |
1159 | } | |
1160 | ||
1161 | static inline bool is_bpf_text_address(unsigned long addr) | |
1162 | { | |
1163 | return false; | |
1164 | } | |
1165 | ||
1166 | static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, | |
1167 | char *type, char *sym) | |
1168 | { | |
1169 | return -ERANGE; | |
1170 | } | |
1171 | ||
1172 | static inline const char * | |
1173 | bpf_address_lookup(unsigned long addr, unsigned long *size, | |
1174 | unsigned long *off, char **modname, char *sym) | |
1175 | { | |
1176 | return NULL; | |
1177 | } | |
1178 | ||
1179 | static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) | |
1180 | { | |
1181 | } | |
1182 | ||
1183 | static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) | |
1184 | { | |
1185 | } | |
6ee52e2a | 1186 | |
b954d834 DB |
1187 | #endif /* CONFIG_BPF_JIT */ |
1188 | ||
7d1982b4 DB |
1189 | void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); |
1190 | ||
34805931 DB |
1191 | #define BPF_ANC BIT(15) |
1192 | ||
55795ef5 RV |
1193 | static inline bool bpf_needs_clear_a(const struct sock_filter *first) |
1194 | { | |
1195 | switch (first->code) { | |
1196 | case BPF_RET | BPF_K: | |
1197 | case BPF_LD | BPF_W | BPF_LEN: | |
1198 | return false; | |
1199 | ||
1200 | case BPF_LD | BPF_W | BPF_ABS: | |
1201 | case BPF_LD | BPF_H | BPF_ABS: | |
1202 | case BPF_LD | BPF_B | BPF_ABS: | |
1203 | if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) | |
1204 | return true; | |
1205 | return false; | |
1206 | ||
1207 | default: | |
1208 | return true; | |
1209 | } | |
1210 | } | |
1211 | ||
34805931 DB |
1212 | static inline u16 bpf_anc_helper(const struct sock_filter *ftest) |
1213 | { | |
1214 | BUG_ON(ftest->code & BPF_ANC); | |
1215 | ||
1216 | switch (ftest->code) { | |
1217 | case BPF_LD | BPF_W | BPF_ABS: | |
1218 | case BPF_LD | BPF_H | BPF_ABS: | |
1219 | case BPF_LD | BPF_B | BPF_ABS: | |
1220 | #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ | |
1221 | return BPF_ANC | SKF_AD_##CODE | |
1222 | switch (ftest->k) { | |
1223 | BPF_ANCILLARY(PROTOCOL); | |
1224 | BPF_ANCILLARY(PKTTYPE); | |
1225 | BPF_ANCILLARY(IFINDEX); | |
1226 | BPF_ANCILLARY(NLATTR); | |
1227 | BPF_ANCILLARY(NLATTR_NEST); | |
1228 | BPF_ANCILLARY(MARK); | |
1229 | BPF_ANCILLARY(QUEUE); | |
1230 | BPF_ANCILLARY(HATYPE); | |
1231 | BPF_ANCILLARY(RXHASH); | |
1232 | BPF_ANCILLARY(CPU); | |
1233 | BPF_ANCILLARY(ALU_XOR_X); | |
1234 | BPF_ANCILLARY(VLAN_TAG); | |
1235 | BPF_ANCILLARY(VLAN_TAG_PRESENT); | |
1236 | BPF_ANCILLARY(PAY_OFFSET); | |
1237 | BPF_ANCILLARY(RANDOM); | |
27cd5452 | 1238 | BPF_ANCILLARY(VLAN_TPID); |
34805931 | 1239 | } |
df561f66 | 1240 | fallthrough; |
34805931 DB |
1241 | default: |
1242 | return ftest->code; | |
1243 | } | |
1244 | } | |
1245 | ||
9f12fbe6 ZSL |
1246 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, |
1247 | int k, unsigned int size); | |
1248 | ||
ea02f941 MS |
1249 | static inline int bpf_tell_extensions(void) |
1250 | { | |
37692299 | 1251 | return SKF_AD_MAX; |
ea02f941 MS |
1252 | } |
1253 | ||
4fbac77d AI |
1254 | struct bpf_sock_addr_kern { |
1255 | struct sock *sk; | |
1256 | struct sockaddr *uaddr; | |
1257 | /* Temporary "register" to make indirect stores to nested structures | |
1258 | * defined above. We need three registers to make such a store, but | |
1259 | * only two (src and dst) are available at convert_ctx_access time | |
1260 | */ | |
1261 | u64 tmp_reg; | |
1cedee13 | 1262 | void *t_ctx; /* Attach type specific context. */ |
4fbac77d AI |
1263 | }; |
1264 | ||
40304b2a LB |
1265 | struct bpf_sock_ops_kern { |
1266 | struct sock *sk; | |
40304b2a | 1267 | union { |
de525be2 | 1268 | u32 args[4]; |
40304b2a LB |
1269 | u32 reply; |
1270 | u32 replylong[4]; | |
1271 | }; | |
0813a841 MKL |
1272 | struct sk_buff *syn_skb; |
1273 | struct sk_buff *skb; | |
1274 | void *skb_data_end; | |
c9985d09 MKL |
1275 | u8 op; |
1276 | u8 is_fullsock; | |
0813a841 | 1277 | u8 remaining_opt_len; |
b73042b8 LB |
1278 | u64 temp; /* temp and everything after is not |
1279 | * initialized to 0 before calling | |
1280 | * the BPF program. New fields that | |
1281 | * should be initialized to 0 should | |
1282 | * be inserted before temp. | |
1283 | * temp is scratch storage used by | |
1284 | * sock_ops_convert_ctx_access | |
1285 | * as temporary storage of a register. | |
1286 | */ | |
40304b2a LB |
1287 | }; |
1288 | ||
7b146ceb AI |
1289 | struct bpf_sysctl_kern { |
1290 | struct ctl_table_header *head; | |
1291 | struct ctl_table *table; | |
1d11b301 AI |
1292 | void *cur_val; |
1293 | size_t cur_len; | |
4e63acdf AI |
1294 | void *new_val; |
1295 | size_t new_len; | |
1296 | int new_updated; | |
7b146ceb | 1297 | int write; |
e1550bfe AI |
1298 | loff_t *ppos; |
1299 | /* Temporary "register" for indirect stores to ppos. */ | |
1300 | u64 tmp_reg; | |
7b146ceb AI |
1301 | }; |
1302 | ||
20f2505f SF |
1303 | #define BPF_SOCKOPT_KERN_BUF_SIZE 32 |
1304 | struct bpf_sockopt_buf { | |
1305 | u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; | |
1306 | }; | |
1307 | ||
0d01da6a SF |
1308 | struct bpf_sockopt_kern { |
1309 | struct sock *sk; | |
1310 | u8 *optval; | |
1311 | u8 *optval_end; | |
1312 | s32 level; | |
1313 | s32 optname; | |
1314 | s32 optlen; | |
1315 | s32 retval; | |
1316 | }; | |
1317 | ||
b1ea9ff6 | 1318 | int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); |
4d295e54 | 1319 | |
e9ddbb77 JS |
1320 | struct bpf_sk_lookup_kern { |
1321 | u16 family; | |
1322 | u16 protocol; | |
d66423fb LB |
1323 | __be16 sport; |
1324 | u16 dport; | |
e9ddbb77 JS |
1325 | struct { |
1326 | __be32 saddr; | |
1327 | __be32 daddr; | |
1328 | } v4; | |
1329 | struct { | |
1330 | const struct in6_addr *saddr; | |
1331 | const struct in6_addr *daddr; | |
1332 | } v6; | |
e9ddbb77 JS |
1333 | struct sock *selected_sk; |
1334 | bool no_reuseport; | |
1335 | }; | |
1336 | ||
1559b4aa JS |
1337 | extern struct static_key_false bpf_sk_lookup_enabled; |
1338 | ||
1339 | /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. | |
1340 | * | |
1341 | * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and | |
1342 | * SK_DROP. Their meaning is as follows: | |
1343 | * | |
1344 | * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result | |
1345 | * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup | |
1346 | * SK_DROP : terminate lookup with -ECONNREFUSED | |
1347 | * | |
1348 | * This macro aggregates return values and selected sockets from | |
1349 | * multiple BPF programs according to following rules in order: | |
1350 | * | |
1351 | * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, | |
1352 | * macro result is SK_PASS and last ctx.selected_sk is used. | |
1353 | * 2. If any program returned SK_DROP return value, | |
1354 | * macro result is SK_DROP. | |
1355 | * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. | |
1356 | * | |
1357 | * Caller must ensure that the prog array is non-NULL, and that the | |
1358 | * array as well as the programs it contains remain valid. | |
1359 | */ | |
1360 | #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ | |
1361 | ({ \ | |
1362 | struct bpf_sk_lookup_kern *_ctx = &(ctx); \ | |
1363 | struct bpf_prog_array_item *_item; \ | |
1364 | struct sock *_selected_sk = NULL; \ | |
1365 | bool _no_reuseport = false; \ | |
1366 | struct bpf_prog *_prog; \ | |
1367 | bool _all_pass = true; \ | |
1368 | u32 _ret; \ | |
1369 | \ | |
1370 | migrate_disable(); \ | |
1371 | _item = &(array)->items[0]; \ | |
1372 | while ((_prog = READ_ONCE(_item->prog))) { \ | |
1373 | /* restore most recent selection */ \ | |
1374 | _ctx->selected_sk = _selected_sk; \ | |
1375 | _ctx->no_reuseport = _no_reuseport; \ | |
1376 | \ | |
1377 | _ret = func(_prog, _ctx); \ | |
1378 | if (_ret == SK_PASS && _ctx->selected_sk) { \ | |
1379 | /* remember last non-NULL socket */ \ | |
1380 | _selected_sk = _ctx->selected_sk; \ | |
1381 | _no_reuseport = _ctx->no_reuseport; \ | |
1382 | } else if (_ret == SK_DROP && _all_pass) { \ | |
1383 | _all_pass = false; \ | |
1384 | } \ | |
1385 | _item++; \ | |
1386 | } \ | |
1387 | _ctx->selected_sk = _selected_sk; \ | |
1388 | _ctx->no_reuseport = _no_reuseport; \ | |
1389 | migrate_enable(); \ | |
1390 | _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ | |
1391 | }) | |
1392 | ||
1393 | static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, | |
1394 | const __be32 saddr, const __be16 sport, | |
1395 | const __be32 daddr, const u16 dport, | |
1396 | struct sock **psk) | |
1397 | { | |
1398 | struct bpf_prog_array *run_array; | |
1399 | struct sock *selected_sk = NULL; | |
1400 | bool no_reuseport = false; | |
1401 | ||
1402 | rcu_read_lock(); | |
1403 | run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); | |
1404 | if (run_array) { | |
1405 | struct bpf_sk_lookup_kern ctx = { | |
1406 | .family = AF_INET, | |
1407 | .protocol = protocol, | |
1408 | .v4.saddr = saddr, | |
1409 | .v4.daddr = daddr, | |
1410 | .sport = sport, | |
1411 | .dport = dport, | |
1412 | }; | |
1413 | u32 act; | |
1414 | ||
1415 | act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); | |
1416 | if (act == SK_PASS) { | |
1417 | selected_sk = ctx.selected_sk; | |
1418 | no_reuseport = ctx.no_reuseport; | |
1419 | } else { | |
1420 | selected_sk = ERR_PTR(-ECONNREFUSED); | |
1421 | } | |
1422 | } | |
1423 | rcu_read_unlock(); | |
1424 | *psk = selected_sk; | |
1425 | return no_reuseport; | |
1426 | } | |
1427 | ||
1122702f JS |
1428 | #if IS_ENABLED(CONFIG_IPV6) |
1429 | static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, | |
1430 | const struct in6_addr *saddr, | |
1431 | const __be16 sport, | |
1432 | const struct in6_addr *daddr, | |
1433 | const u16 dport, | |
1434 | struct sock **psk) | |
1435 | { | |
1436 | struct bpf_prog_array *run_array; | |
1437 | struct sock *selected_sk = NULL; | |
1438 | bool no_reuseport = false; | |
1439 | ||
1440 | rcu_read_lock(); | |
1441 | run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); | |
1442 | if (run_array) { | |
1443 | struct bpf_sk_lookup_kern ctx = { | |
1444 | .family = AF_INET6, | |
1445 | .protocol = protocol, | |
1446 | .v6.saddr = saddr, | |
1447 | .v6.daddr = daddr, | |
1448 | .sport = sport, | |
1449 | .dport = dport, | |
1450 | }; | |
1451 | u32 act; | |
1452 | ||
1453 | act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN); | |
1454 | if (act == SK_PASS) { | |
1455 | selected_sk = ctx.selected_sk; | |
1456 | no_reuseport = ctx.no_reuseport; | |
1457 | } else { | |
1458 | selected_sk = ERR_PTR(-ECONNREFUSED); | |
1459 | } | |
1460 | } | |
1461 | rcu_read_unlock(); | |
1462 | *psk = selected_sk; | |
1463 | return no_reuseport; | |
1464 | } | |
1465 | #endif /* IS_ENABLED(CONFIG_IPV6) */ | |
1466 | ||
e6a4750f BT |
1467 | static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, u64 flags, |
1468 | void *lookup_elem(struct bpf_map *map, u32 key)) | |
1469 | { | |
1470 | struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); | |
1471 | ||
1472 | /* Lower bits of the flags are used as return code on lookup failure */ | |
1473 | if (unlikely(flags > XDP_TX)) | |
1474 | return XDP_ABORTED; | |
1475 | ||
1476 | ri->tgt_value = lookup_elem(map, ifindex); | |
1477 | if (unlikely(!ri->tgt_value)) { | |
1478 | /* If the lookup fails we want to clear out the state in the | |
1479 | * redirect_info struct completely, so that if an eBPF program | |
1480 | * performs multiple lookups, the last one always takes | |
1481 | * precedence. | |
1482 | */ | |
ee75aef2 BT |
1483 | ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ |
1484 | ri->map_type = BPF_MAP_TYPE_UNSPEC; | |
e6a4750f BT |
1485 | return flags; |
1486 | } | |
1487 | ||
e6a4750f | 1488 | ri->tgt_index = ifindex; |
ee75aef2 BT |
1489 | ri->map_id = map->id; |
1490 | ri->map_type = map->map_type; | |
e6a4750f BT |
1491 | |
1492 | return XDP_REDIRECT; | |
1493 | } | |
1494 | ||
1da177e4 | 1495 | #endif /* __LINUX_FILTER_H__ */ |