1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Linux Socket Filter Data Structures
4  */
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
7 
8 #include <linux/atomic.h>
9 #include <linux/bpf.h>
10 #include <linux/refcount.h>
11 #include <linux/compat.h>
12 #include <linux/skbuff.h>
13 #include <linux/linkage.h>
14 #include <linux/printk.h>
15 #include <linux/workqueue.h>
16 #include <linux/sched.h>
17 #include <linux/capability.h>
18 #include <linux/set_memory.h>
19 #include <linux/kallsyms.h>
20 #include <linux/if_vlan.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sockptr.h>
23 #include <crypto/sha1.h>
24 #include <linux/u64_stats_sync.h>
25 
26 #include <net/sch_generic.h>
27 
28 #include <asm/byteorder.h>
29 #include <uapi/linux/filter.h>
30 
31 struct sk_buff;
32 struct sock;
33 struct seccomp_data;
34 struct bpf_prog_aux;
35 struct xdp_rxq_info;
36 struct xdp_buff;
37 struct sock_reuseport;
38 struct ctl_table;
39 struct ctl_table_header;
40 
41 /* ArgX, context and stack frame pointer register positions. Note,
42  * Arg1, Arg2, Arg3, etc are used as argument mappings of function
43  * calls in BPF_CALL instruction.
44  */
45 #define BPF_REG_ARG1	BPF_REG_1
46 #define BPF_REG_ARG2	BPF_REG_2
47 #define BPF_REG_ARG3	BPF_REG_3
48 #define BPF_REG_ARG4	BPF_REG_4
49 #define BPF_REG_ARG5	BPF_REG_5
50 #define BPF_REG_CTX	BPF_REG_6
51 #define BPF_REG_FP	BPF_REG_10
52 
53 /* Additional register mappings for converted user programs. */
54 #define BPF_REG_A	BPF_REG_0
55 #define BPF_REG_X	BPF_REG_7
56 #define BPF_REG_TMP	BPF_REG_2	/* scratch reg */
57 #define BPF_REG_D	BPF_REG_8	/* data, callee-saved */
58 #define BPF_REG_H	BPF_REG_9	/* hlen, callee-saved */
59 
60 /* Kernel hidden auxiliary/helper register. */
61 #define BPF_REG_AX		MAX_BPF_REG
62 #define MAX_BPF_EXT_REG		(MAX_BPF_REG + 1)
63 #define MAX_BPF_JIT_REG		MAX_BPF_EXT_REG
64 
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL	0xf0
67 
68 /* unused opcode to mark special load instruction. Same as BPF_ABS */
69 #define BPF_PROBE_MEM	0x20
70 
71 /* unused opcode to mark call to interpreter with arguments */
72 #define BPF_CALL_ARGS	0xe0
73 
74 /* unused opcode to mark speculation barrier for mitigating
75  * Speculative Store Bypass
76  */
77 #define BPF_NOSPEC	0xc0
78 
79 /* As per nm, we expose JITed images as text (code) section for
80  * kallsyms. That way, tools like perf can find it to match
81  * addresses.
82  */
83 #define BPF_SYM_ELF_TYPE	't'
84 
85 /* BPF program can access up to 512 bytes of stack space. */
86 #define MAX_BPF_STACK	512
87 
88 /* Helper macros for filter block array initializers. */
89 
90 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
91 
92 #define BPF_ALU64_REG(OP, DST, SRC)				\
93 	((struct bpf_insn) {					\
94 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,	\
95 		.dst_reg = DST,					\
96 		.src_reg = SRC,					\
97 		.off   = 0,					\
98 		.imm   = 0 })
99 
100 #define BPF_ALU32_REG(OP, DST, SRC)				\
101 	((struct bpf_insn) {					\
102 		.code  = BPF_ALU | BPF_OP(OP) | BPF_X,		\
103 		.dst_reg = DST,					\
104 		.src_reg = SRC,					\
105 		.off   = 0,					\
106 		.imm   = 0 })
107 
108 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
109 
110 #define BPF_ALU64_IMM(OP, DST, IMM)				\
111 	((struct bpf_insn) {					\
112 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,	\
113 		.dst_reg = DST,					\
114 		.src_reg = 0,					\
115 		.off   = 0,					\
116 		.imm   = IMM })
117 
118 #define BPF_ALU32_IMM(OP, DST, IMM)				\
119 	((struct bpf_insn) {					\
120 		.code  = BPF_ALU | BPF_OP(OP) | BPF_K,		\
121 		.dst_reg = DST,					\
122 		.src_reg = 0,					\
123 		.off   = 0,					\
124 		.imm   = IMM })
125 
126 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
127 
128 #define BPF_ENDIAN(TYPE, DST, LEN)				\
129 	((struct bpf_insn) {					\
130 		.code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),	\
131 		.dst_reg = DST,					\
132 		.src_reg = 0,					\
133 		.off   = 0,					\
134 		.imm   = LEN })
135 
136 /* Short form of mov, dst_reg = src_reg */
137 
138 #define BPF_MOV64_REG(DST, SRC)					\
139 	((struct bpf_insn) {					\
140 		.code  = BPF_ALU64 | BPF_MOV | BPF_X,		\
141 		.dst_reg = DST,					\
142 		.src_reg = SRC,					\
143 		.off   = 0,					\
144 		.imm   = 0 })
145 
146 #define BPF_MOV32_REG(DST, SRC)					\
147 	((struct bpf_insn) {					\
148 		.code  = BPF_ALU | BPF_MOV | BPF_X,		\
149 		.dst_reg = DST,					\
150 		.src_reg = SRC,					\
151 		.off   = 0,					\
152 		.imm   = 0 })
153 
154 /* Short form of mov, dst_reg = imm32 */
155 
156 #define BPF_MOV64_IMM(DST, IMM)					\
157 	((struct bpf_insn) {					\
158 		.code  = BPF_ALU64 | BPF_MOV | BPF_K,		\
159 		.dst_reg = DST,					\
160 		.src_reg = 0,					\
161 		.off   = 0,					\
162 		.imm   = IMM })
163 
164 #define BPF_MOV32_IMM(DST, IMM)					\
165 	((struct bpf_insn) {					\
166 		.code  = BPF_ALU | BPF_MOV | BPF_K,		\
167 		.dst_reg = DST,					\
168 		.src_reg = 0,					\
169 		.off   = 0,					\
170 		.imm   = IMM })
171 
172 /* Special form of mov32, used for doing explicit zero extension on dst. */
173 #define BPF_ZEXT_REG(DST)					\
174 	((struct bpf_insn) {					\
175 		.code  = BPF_ALU | BPF_MOV | BPF_X,		\
176 		.dst_reg = DST,					\
177 		.src_reg = DST,					\
178 		.off   = 0,					\
179 		.imm   = 1 })
180 
insn_is_zext(const struct bpf_insn * insn)181 static inline bool insn_is_zext(const struct bpf_insn *insn)
182 {
183 	return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
184 }
185 
186 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
187 #define BPF_LD_IMM64(DST, IMM)					\
188 	BPF_LD_IMM64_RAW(DST, 0, IMM)
189 
190 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)				\
191 	((struct bpf_insn) {					\
192 		.code  = BPF_LD | BPF_DW | BPF_IMM,		\
193 		.dst_reg = DST,					\
194 		.src_reg = SRC,					\
195 		.off   = 0,					\
196 		.imm   = (__u32) (IMM) }),			\
197 	((struct bpf_insn) {					\
198 		.code  = 0, /* zero is reserved opcode */	\
199 		.dst_reg = 0,					\
200 		.src_reg = 0,					\
201 		.off   = 0,					\
202 		.imm   = ((__u64) (IMM)) >> 32 })
203 
204 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
205 #define BPF_LD_MAP_FD(DST, MAP_FD)				\
206 	BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
207 
208 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
209 
210 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)			\
211 	((struct bpf_insn) {					\
212 		.code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),	\
213 		.dst_reg = DST,					\
214 		.src_reg = SRC,					\
215 		.off   = 0,					\
216 		.imm   = IMM })
217 
218 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)			\
219 	((struct bpf_insn) {					\
220 		.code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),	\
221 		.dst_reg = DST,					\
222 		.src_reg = SRC,					\
223 		.off   = 0,					\
224 		.imm   = IMM })
225 
226 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
227 
228 #define BPF_LD_ABS(SIZE, IMM)					\
229 	((struct bpf_insn) {					\
230 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,	\
231 		.dst_reg = 0,					\
232 		.src_reg = 0,					\
233 		.off   = 0,					\
234 		.imm   = IMM })
235 
236 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
237 
238 #define BPF_LD_IND(SIZE, SRC, IMM)				\
239 	((struct bpf_insn) {					\
240 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,	\
241 		.dst_reg = 0,					\
242 		.src_reg = SRC,					\
243 		.off   = 0,					\
244 		.imm   = IMM })
245 
246 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
247 
248 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)			\
249 	((struct bpf_insn) {					\
250 		.code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,	\
251 		.dst_reg = DST,					\
252 		.src_reg = SRC,					\
253 		.off   = OFF,					\
254 		.imm   = 0 })
255 
256 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
257 
258 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)			\
259 	((struct bpf_insn) {					\
260 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,	\
261 		.dst_reg = DST,					\
262 		.src_reg = SRC,					\
263 		.off   = OFF,					\
264 		.imm   = 0 })
265 
266 
267 /*
268  * Atomic operations:
269  *
270  *   BPF_ADD                  *(uint *) (dst_reg + off16) += src_reg
271  *   BPF_AND                  *(uint *) (dst_reg + off16) &= src_reg
272  *   BPF_OR                   *(uint *) (dst_reg + off16) |= src_reg
273  *   BPF_XOR                  *(uint *) (dst_reg + off16) ^= src_reg
274  *   BPF_ADD | BPF_FETCH      src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
275  *   BPF_AND | BPF_FETCH      src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
276  *   BPF_OR | BPF_FETCH       src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
277  *   BPF_XOR | BPF_FETCH      src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
278  *   BPF_XCHG                 src_reg = atomic_xchg(dst_reg + off16, src_reg)
279  *   BPF_CMPXCHG              r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
280  */
281 
282 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF)			\
283 	((struct bpf_insn) {					\
284 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC,	\
285 		.dst_reg = DST,					\
286 		.src_reg = SRC,					\
287 		.off   = OFF,					\
288 		.imm   = OP })
289 
290 /* Legacy alias */
291 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
292 
293 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
294 
295 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)				\
296 	((struct bpf_insn) {					\
297 		.code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,	\
298 		.dst_reg = DST,					\
299 		.src_reg = 0,					\
300 		.off   = OFF,					\
301 		.imm   = IMM })
302 
303 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
304 
305 #define BPF_JMP_REG(OP, DST, SRC, OFF)				\
306 	((struct bpf_insn) {					\
307 		.code  = BPF_JMP | BPF_OP(OP) | BPF_X,		\
308 		.dst_reg = DST,					\
309 		.src_reg = SRC,					\
310 		.off   = OFF,					\
311 		.imm   = 0 })
312 
313 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
314 
315 #define BPF_JMP_IMM(OP, DST, IMM, OFF)				\
316 	((struct bpf_insn) {					\
317 		.code  = BPF_JMP | BPF_OP(OP) | BPF_K,		\
318 		.dst_reg = DST,					\
319 		.src_reg = 0,					\
320 		.off   = OFF,					\
321 		.imm   = IMM })
322 
323 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
324 
325 #define BPF_JMP32_REG(OP, DST, SRC, OFF)			\
326 	((struct bpf_insn) {					\
327 		.code  = BPF_JMP32 | BPF_OP(OP) | BPF_X,	\
328 		.dst_reg = DST,					\
329 		.src_reg = SRC,					\
330 		.off   = OFF,					\
331 		.imm   = 0 })
332 
333 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
334 
335 #define BPF_JMP32_IMM(OP, DST, IMM, OFF)			\
336 	((struct bpf_insn) {					\
337 		.code  = BPF_JMP32 | BPF_OP(OP) | BPF_K,	\
338 		.dst_reg = DST,					\
339 		.src_reg = 0,					\
340 		.off   = OFF,					\
341 		.imm   = IMM })
342 
343 /* Unconditional jumps, goto pc + off16 */
344 
345 #define BPF_JMP_A(OFF)						\
346 	((struct bpf_insn) {					\
347 		.code  = BPF_JMP | BPF_JA,			\
348 		.dst_reg = 0,					\
349 		.src_reg = 0,					\
350 		.off   = OFF,					\
351 		.imm   = 0 })
352 
353 /* Relative call */
354 
355 #define BPF_CALL_REL(TGT)					\
356 	((struct bpf_insn) {					\
357 		.code  = BPF_JMP | BPF_CALL,			\
358 		.dst_reg = 0,					\
359 		.src_reg = BPF_PSEUDO_CALL,			\
360 		.off   = 0,					\
361 		.imm   = TGT })
362 
363 /* Convert function address to BPF immediate */
364 
365 #define BPF_CALL_IMM(x)	((void *)(x) - (void *)__bpf_call_base)
366 
367 #define BPF_EMIT_CALL(FUNC)					\
368 	((struct bpf_insn) {					\
369 		.code  = BPF_JMP | BPF_CALL,			\
370 		.dst_reg = 0,					\
371 		.src_reg = 0,					\
372 		.off   = 0,					\
373 		.imm   = BPF_CALL_IMM(FUNC) })
374 
375 /* Raw code statement block */
376 
377 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)			\
378 	((struct bpf_insn) {					\
379 		.code  = CODE,					\
380 		.dst_reg = DST,					\
381 		.src_reg = SRC,					\
382 		.off   = OFF,					\
383 		.imm   = IMM })
384 
385 /* Program exit */
386 
387 #define BPF_EXIT_INSN()						\
388 	((struct bpf_insn) {					\
389 		.code  = BPF_JMP | BPF_EXIT,			\
390 		.dst_reg = 0,					\
391 		.src_reg = 0,					\
392 		.off   = 0,					\
393 		.imm   = 0 })
394 
395 /* Speculation barrier */
396 
397 #define BPF_ST_NOSPEC()						\
398 	((struct bpf_insn) {					\
399 		.code  = BPF_ST | BPF_NOSPEC,			\
400 		.dst_reg = 0,					\
401 		.src_reg = 0,					\
402 		.off   = 0,					\
403 		.imm   = 0 })
404 
405 /* Internal classic blocks for direct assignment */
406 
407 #define __BPF_STMT(CODE, K)					\
408 	((struct sock_filter) BPF_STMT(CODE, K))
409 
410 #define __BPF_JUMP(CODE, K, JT, JF)				\
411 	((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
412 
413 #define bytes_to_bpf_size(bytes)				\
414 ({								\
415 	int bpf_size = -EINVAL;					\
416 								\
417 	if (bytes == sizeof(u8))				\
418 		bpf_size = BPF_B;				\
419 	else if (bytes == sizeof(u16))				\
420 		bpf_size = BPF_H;				\
421 	else if (bytes == sizeof(u32))				\
422 		bpf_size = BPF_W;				\
423 	else if (bytes == sizeof(u64))				\
424 		bpf_size = BPF_DW;				\
425 								\
426 	bpf_size;						\
427 })
428 
429 #define bpf_size_to_bytes(bpf_size)				\
430 ({								\
431 	int bytes = -EINVAL;					\
432 								\
433 	if (bpf_size == BPF_B)					\
434 		bytes = sizeof(u8);				\
435 	else if (bpf_size == BPF_H)				\
436 		bytes = sizeof(u16);				\
437 	else if (bpf_size == BPF_W)				\
438 		bytes = sizeof(u32);				\
439 	else if (bpf_size == BPF_DW)				\
440 		bytes = sizeof(u64);				\
441 								\
442 	bytes;							\
443 })
444 
445 #define BPF_SIZEOF(type)					\
446 	({							\
447 		const int __size = bytes_to_bpf_size(sizeof(type)); \
448 		BUILD_BUG_ON(__size < 0);			\
449 		__size;						\
450 	})
451 
452 #define BPF_FIELD_SIZEOF(type, field)				\
453 	({							\
454 		const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
455 		BUILD_BUG_ON(__size < 0);			\
456 		__size;						\
457 	})
458 
459 #define BPF_LDST_BYTES(insn)					\
460 	({							\
461 		const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
462 		WARN_ON(__size < 0);				\
463 		__size;						\
464 	})
465 
466 #define __BPF_MAP_0(m, v, ...) v
467 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
468 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
469 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
470 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
471 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
472 
473 #define __BPF_REG_0(...) __BPF_PAD(5)
474 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
475 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
476 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
477 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
478 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
479 
480 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
481 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
482 
483 #define __BPF_CAST(t, a)						       \
484 	(__force t)							       \
485 	(__force							       \
486 	 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
487 				      (unsigned long)0, (t)0))) a
488 #define __BPF_V void
489 #define __BPF_N
490 
491 #define __BPF_DECL_ARGS(t, a) t   a
492 #define __BPF_DECL_REGS(t, a) u64 a
493 
494 #define __BPF_PAD(n)							       \
495 	__BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
496 		  u64, __ur_3, u64, __ur_4, u64, __ur_5)
497 
498 #define BPF_CALL_x(x, name, ...)					       \
499 	static __always_inline						       \
500 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
501 	typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
502 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));	       \
503 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))	       \
504 	{								       \
505 		return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
506 	}								       \
507 	static __always_inline						       \
508 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
509 
510 #define BPF_CALL_0(name, ...)	BPF_CALL_x(0, name, __VA_ARGS__)
511 #define BPF_CALL_1(name, ...)	BPF_CALL_x(1, name, __VA_ARGS__)
512 #define BPF_CALL_2(name, ...)	BPF_CALL_x(2, name, __VA_ARGS__)
513 #define BPF_CALL_3(name, ...)	BPF_CALL_x(3, name, __VA_ARGS__)
514 #define BPF_CALL_4(name, ...)	BPF_CALL_x(4, name, __VA_ARGS__)
515 #define BPF_CALL_5(name, ...)	BPF_CALL_x(5, name, __VA_ARGS__)
516 
517 #define bpf_ctx_range(TYPE, MEMBER)						\
518 	offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
519 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)				\
520 	offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
521 #if BITS_PER_LONG == 64
522 # define bpf_ctx_range_ptr(TYPE, MEMBER)					\
523 	offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
524 #else
525 # define bpf_ctx_range_ptr(TYPE, MEMBER)					\
526 	offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
527 #endif /* BITS_PER_LONG == 64 */
528 
529 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)				\
530 	({									\
531 		BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE));		\
532 		*(PTR_SIZE) = (SIZE);						\
533 		offsetof(TYPE, MEMBER);						\
534 	})
535 
536 /* A struct sock_filter is architecture independent. */
537 struct compat_sock_fprog {
538 	u16		len;
539 	compat_uptr_t	filter;	/* struct sock_filter * */
540 };
541 
542 struct sock_fprog_kern {
543 	u16			len;
544 	struct sock_filter	*filter;
545 };
546 
547 /* Some arches need doubleword alignment for their instructions and/or data */
548 #define BPF_IMAGE_ALIGNMENT 8
549 
550 struct bpf_binary_header {
551 	u32 size;
552 	u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
553 };
554 
555 struct bpf_prog_stats {
556 	u64_stats_t cnt;
557 	u64_stats_t nsecs;
558 	u64_stats_t misses;
559 	struct u64_stats_sync syncp;
560 } __aligned(2 * sizeof(u64));
561 
562 struct bpf_prog {
563 	u16			pages;		/* Number of allocated pages */
564 	u16			jited:1,	/* Is our filter JIT'ed? */
565 				jit_requested:1,/* archs need to JIT the prog */
566 				gpl_compatible:1, /* Is filter GPL compatible? */
567 				cb_access:1,	/* Is control block accessed? */
568 				dst_needed:1,	/* Do we need dst entry? */
569 				blinding_requested:1, /* needs constant blinding */
570 				blinded:1,	/* Was blinded */
571 				is_func:1,	/* program is a bpf function */
572 				kprobe_override:1, /* Do we override a kprobe? */
573 				has_callchain_buf:1, /* callchain buffer allocated? */
574 				enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
575 				call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */
576 				call_get_func_ip:1, /* Do we call get_func_ip() */
577 				tstamp_type_access:1; /* Accessed __sk_buff->tstamp_type */
578 	enum bpf_prog_type	type;		/* Type of BPF program */
579 	enum bpf_attach_type	expected_attach_type; /* For some prog types */
580 	u32			len;		/* Number of filter blocks */
581 	u32			jited_len;	/* Size of jited insns in bytes */
582 	u8			tag[BPF_TAG_SIZE];
583 	struct bpf_prog_stats __percpu *stats;
584 	int __percpu		*active;
585 	unsigned int		(*bpf_func)(const void *ctx,
586 					    const struct bpf_insn *insn);
587 	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
588 	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
589 	/* Instructions for interpreter */
590 	union {
591 		DECLARE_FLEX_ARRAY(struct sock_filter, insns);
592 		DECLARE_FLEX_ARRAY(struct bpf_insn, insnsi);
593 	};
594 };
595 
596 struct sk_filter {
597 	refcount_t	refcnt;
598 	struct rcu_head	rcu;
599 	struct bpf_prog	*prog;
600 };
601 
602 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
603 
604 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
605 					  const struct bpf_insn *insnsi,
606 					  unsigned int (*bpf_func)(const void *,
607 								   const struct bpf_insn *));
608 
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)609 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
610 					  const void *ctx,
611 					  bpf_dispatcher_fn dfunc)
612 {
613 	u32 ret;
614 
615 	cant_migrate();
616 	if (static_branch_unlikely(&bpf_stats_enabled_key)) {
617 		struct bpf_prog_stats *stats;
618 		u64 start = sched_clock();
619 		unsigned long flags;
620 
621 		ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
622 		stats = this_cpu_ptr(prog->stats);
623 		flags = u64_stats_update_begin_irqsave(&stats->syncp);
624 		u64_stats_inc(&stats->cnt);
625 		u64_stats_add(&stats->nsecs, sched_clock() - start);
626 		u64_stats_update_end_irqrestore(&stats->syncp, flags);
627 	} else {
628 		ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
629 	}
630 	return ret;
631 }
632 
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)633 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
634 {
635 	return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
636 }
637 
638 /*
639  * Use in preemptible and therefore migratable context to make sure that
640  * the execution of the BPF program runs on one CPU.
641  *
642  * This uses migrate_disable/enable() explicitly to document that the
643  * invocation of a BPF program does not require reentrancy protection
644  * against a BPF program which is invoked from a preempting task.
645  */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)646 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
647 					  const void *ctx)
648 {
649 	u32 ret;
650 
651 	migrate_disable();
652 	ret = bpf_prog_run(prog, ctx);
653 	migrate_enable();
654 	return ret;
655 }
656 
657 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
658 
659 struct bpf_skb_data_end {
660 	struct qdisc_skb_cb qdisc_cb;
661 	void *data_meta;
662 	void *data_end;
663 };
664 
665 struct bpf_nh_params {
666 	u32 nh_family;
667 	union {
668 		u32 ipv4_nh;
669 		struct in6_addr ipv6_nh;
670 	};
671 };
672 
673 struct bpf_redirect_info {
674 	u32 flags;
675 	u32 tgt_index;
676 	void *tgt_value;
677 	struct bpf_map *map;
678 	u32 map_id;
679 	enum bpf_map_type map_type;
680 	u32 kern_flags;
681 	struct bpf_nh_params nh;
682 };
683 
684 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
685 
686 /* flags for bpf_redirect_info kern_flags */
687 #define BPF_RI_F_RF_NO_DIRECT	BIT(0)	/* no napi_direct on return_frame */
688 
689 /* Compute the linear packet data range [data, data_end) which
690  * will be accessed by various program types (cls_bpf, act_bpf,
691  * lwt, ...). Subsystems allowing direct data access must (!)
692  * ensure that cb[] area can be written to when BPF program is
693  * invoked (otherwise cb[] save/restore is necessary).
694  */
bpf_compute_data_pointers(struct sk_buff * skb)695 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
696 {
697 	struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
698 
699 	BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
700 	cb->data_meta = skb->data - skb_metadata_len(skb);
701 	cb->data_end  = skb->data + skb_headlen(skb);
702 }
703 
704 /* Similar to bpf_compute_data_pointers(), except that save orginal
705  * data in cb->data and cb->meta_data for restore.
706  */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)707 static inline void bpf_compute_and_save_data_end(
708 	struct sk_buff *skb, void **saved_data_end)
709 {
710 	struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
711 
712 	*saved_data_end = cb->data_end;
713 	cb->data_end  = skb->data + skb_headlen(skb);
714 }
715 
716 /* Restore data saved by bpf_compute_data_pointers(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)717 static inline void bpf_restore_data_end(
718 	struct sk_buff *skb, void *saved_data_end)
719 {
720 	struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
721 
722 	cb->data_end = saved_data_end;
723 }
724 
bpf_skb_cb(const struct sk_buff * skb)725 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
726 {
727 	/* eBPF programs may read/write skb->cb[] area to transfer meta
728 	 * data between tail calls. Since this also needs to work with
729 	 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
730 	 *
731 	 * In some socket filter cases, the cb unfortunately needs to be
732 	 * saved/restored so that protocol specific skb->cb[] data won't
733 	 * be lost. In any case, due to unpriviledged eBPF programs
734 	 * attached to sockets, we need to clear the bpf_skb_cb() area
735 	 * to not leak previous contents to user space.
736 	 */
737 	BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
738 	BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
739 		     sizeof_field(struct qdisc_skb_cb, data));
740 
741 	return qdisc_skb_cb(skb)->data;
742 }
743 
744 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)745 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
746 					 const void *ctx)
747 {
748 	const struct sk_buff *skb = ctx;
749 	u8 *cb_data = bpf_skb_cb(skb);
750 	u8 cb_saved[BPF_SKB_CB_LEN];
751 	u32 res;
752 
753 	if (unlikely(prog->cb_access)) {
754 		memcpy(cb_saved, cb_data, sizeof(cb_saved));
755 		memset(cb_data, 0, sizeof(cb_saved));
756 	}
757 
758 	res = bpf_prog_run(prog, skb);
759 
760 	if (unlikely(prog->cb_access))
761 		memcpy(cb_data, cb_saved, sizeof(cb_saved));
762 
763 	return res;
764 }
765 
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)766 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
767 				       struct sk_buff *skb)
768 {
769 	u32 res;
770 
771 	migrate_disable();
772 	res = __bpf_prog_run_save_cb(prog, skb);
773 	migrate_enable();
774 	return res;
775 }
776 
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)777 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
778 					struct sk_buff *skb)
779 {
780 	u8 *cb_data = bpf_skb_cb(skb);
781 	u32 res;
782 
783 	if (unlikely(prog->cb_access))
784 		memset(cb_data, 0, BPF_SKB_CB_LEN);
785 
786 	res = bpf_prog_run_pin_on_cpu(prog, skb);
787 	return res;
788 }
789 
790 DECLARE_BPF_DISPATCHER(xdp)
791 
792 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
793 
794 u32 xdp_master_redirect(struct xdp_buff *xdp);
795 
bpf_prog_run_xdp(const struct bpf_prog * prog,struct xdp_buff * xdp)796 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
797 					    struct xdp_buff *xdp)
798 {
799 	/* Driver XDP hooks are invoked within a single NAPI poll cycle and thus
800 	 * under local_bh_disable(), which provides the needed RCU protection
801 	 * for accessing map entries.
802 	 */
803 	u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
804 
805 	if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) {
806 		if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev))
807 			act = xdp_master_redirect(xdp);
808 	}
809 
810 	return act;
811 }
812 
813 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
814 
bpf_prog_insn_size(const struct bpf_prog * prog)815 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
816 {
817 	return prog->len * sizeof(struct bpf_insn);
818 }
819 
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)820 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
821 {
822 	return round_up(bpf_prog_insn_size(prog) +
823 			sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
824 }
825 
bpf_prog_size(unsigned int proglen)826 static inline unsigned int bpf_prog_size(unsigned int proglen)
827 {
828 	return max(sizeof(struct bpf_prog),
829 		   offsetof(struct bpf_prog, insns[proglen]));
830 }
831 
bpf_prog_was_classic(const struct bpf_prog * prog)832 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
833 {
834 	/* When classic BPF programs have been loaded and the arch
835 	 * does not have a classic BPF JIT (anymore), they have been
836 	 * converted via bpf_migrate_filter() to eBPF and thus always
837 	 * have an unspec program type.
838 	 */
839 	return prog->type == BPF_PROG_TYPE_UNSPEC;
840 }
841 
bpf_ctx_off_adjust_machine(u32 size)842 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
843 {
844 	const u32 size_machine = sizeof(unsigned long);
845 
846 	if (size > size_machine && size % size_machine == 0)
847 		size = size_machine;
848 
849 	return size;
850 }
851 
852 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)853 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
854 {
855 	return size <= size_default && (size & (size - 1)) == 0;
856 }
857 
858 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)859 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
860 {
861 	u8 access_off = off & (size_default - 1);
862 
863 #ifdef __LITTLE_ENDIAN
864 	return access_off;
865 #else
866 	return size_default - (access_off + size);
867 #endif
868 }
869 
870 #define bpf_ctx_wide_access_ok(off, size, type, field)			\
871 	(size == sizeof(__u64) &&					\
872 	off >= offsetof(type, field) &&					\
873 	off + sizeof(__u64) <= offsetofend(type, field) &&		\
874 	off % sizeof(__u64) == 0)
875 
876 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
877 
bpf_prog_lock_ro(struct bpf_prog * fp)878 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
879 {
880 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
881 	if (!fp->jited) {
882 		set_vm_flush_reset_perms(fp);
883 		set_memory_ro((unsigned long)fp, fp->pages);
884 	}
885 #endif
886 }
887 
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)888 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
889 {
890 	set_vm_flush_reset_perms(hdr);
891 	set_memory_ro((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
892 	set_memory_x((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
893 }
894 
895 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)896 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
897 {
898 	return sk_filter_trim_cap(sk, skb, 1);
899 }
900 
901 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
902 void bpf_prog_free(struct bpf_prog *fp);
903 
904 bool bpf_opcode_in_insntable(u8 code);
905 
906 void bpf_prog_free_linfo(struct bpf_prog *prog);
907 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
908 			       const u32 *insn_to_jit_off);
909 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
910 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
911 
912 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
913 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
914 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
915 				  gfp_t gfp_extra_flags);
916 void __bpf_prog_free(struct bpf_prog *fp);
917 
bpf_prog_unlock_free(struct bpf_prog * fp)918 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
919 {
920 	__bpf_prog_free(fp);
921 }
922 
923 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
924 				       unsigned int flen);
925 
926 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
927 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
928 			      bpf_aux_classic_check_t trans, bool save_orig);
929 void bpf_prog_destroy(struct bpf_prog *fp);
930 
931 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
932 int sk_attach_bpf(u32 ufd, struct sock *sk);
933 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
934 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
935 void sk_reuseport_prog_free(struct bpf_prog *prog);
936 int sk_detach_filter(struct sock *sk);
937 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
938 		  unsigned int len);
939 
940 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
941 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
942 
943 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
944 #define __bpf_call_base_args \
945 	((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
946 	 (void *)__bpf_call_base)
947 
948 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
949 void bpf_jit_compile(struct bpf_prog *prog);
950 bool bpf_jit_needs_zext(void);
951 bool bpf_jit_supports_subprog_tailcalls(void);
952 bool bpf_jit_supports_kfunc_call(void);
953 bool bpf_helper_changes_pkt_data(void *func);
954 
bpf_dump_raw_ok(const struct cred * cred)955 static inline bool bpf_dump_raw_ok(const struct cred *cred)
956 {
957 	/* Reconstruction of call-sites is dependent on kallsyms,
958 	 * thus make dump the same restriction.
959 	 */
960 	return kallsyms_show_value(cred);
961 }
962 
963 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
964 				       const struct bpf_insn *patch, u32 len);
965 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
966 
967 void bpf_clear_redirect_map(struct bpf_map *map);
968 
xdp_return_frame_no_direct(void)969 static inline bool xdp_return_frame_no_direct(void)
970 {
971 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
972 
973 	return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
974 }
975 
xdp_set_return_frame_no_direct(void)976 static inline void xdp_set_return_frame_no_direct(void)
977 {
978 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
979 
980 	ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
981 }
982 
xdp_clear_return_frame_no_direct(void)983 static inline void xdp_clear_return_frame_no_direct(void)
984 {
985 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
986 
987 	ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
988 }
989 
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)990 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
991 				 unsigned int pktlen)
992 {
993 	unsigned int len;
994 
995 	if (unlikely(!(fwd->flags & IFF_UP)))
996 		return -ENETDOWN;
997 
998 	len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
999 	if (pktlen > len)
1000 		return -EMSGSIZE;
1001 
1002 	return 0;
1003 }
1004 
1005 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1006  * same cpu context. Further for best results no more than a single map
1007  * for the do_redirect/do_flush pair should be used. This limitation is
1008  * because we only track one map and force a flush when the map changes.
1009  * This does not appear to be a real limitation for existing software.
1010  */
1011 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1012 			    struct xdp_buff *xdp, struct bpf_prog *prog);
1013 int xdp_do_redirect(struct net_device *dev,
1014 		    struct xdp_buff *xdp,
1015 		    struct bpf_prog *prog);
1016 int xdp_do_redirect_frame(struct net_device *dev,
1017 			  struct xdp_buff *xdp,
1018 			  struct xdp_frame *xdpf,
1019 			  struct bpf_prog *prog);
1020 void xdp_do_flush(void);
1021 
1022 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
1023  * it is no longer only flushing maps. Keep this define for compatibility
1024  * until all drivers are updated - do not use xdp_do_flush_map() in new code!
1025  */
1026 #define xdp_do_flush_map xdp_do_flush
1027 
1028 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act);
1029 
1030 #ifdef CONFIG_INET
1031 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1032 				  struct bpf_prog *prog, struct sk_buff *skb,
1033 				  struct sock *migrating_sk,
1034 				  u32 hash);
1035 #else
1036 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)1037 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1038 		     struct bpf_prog *prog, struct sk_buff *skb,
1039 		     struct sock *migrating_sk,
1040 		     u32 hash)
1041 {
1042 	return NULL;
1043 }
1044 #endif
1045 
1046 #ifdef CONFIG_BPF_JIT
1047 extern int bpf_jit_enable;
1048 extern int bpf_jit_harden;
1049 extern int bpf_jit_kallsyms;
1050 extern long bpf_jit_limit;
1051 extern long bpf_jit_limit_max;
1052 
1053 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1054 
1055 struct bpf_binary_header *
1056 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1057 		     unsigned int alignment,
1058 		     bpf_jit_fill_hole_t bpf_fill_ill_insns);
1059 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1060 u64 bpf_jit_alloc_exec_limit(void);
1061 void *bpf_jit_alloc_exec(unsigned long size);
1062 void bpf_jit_free_exec(void *addr);
1063 void bpf_jit_free(struct bpf_prog *fp);
1064 struct bpf_binary_header *
1065 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1066 
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1067 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1068 {
1069 	return list_empty(&fp->aux->ksym.lnode) ||
1070 	       fp->aux->ksym.lnode.prev == LIST_POISON2;
1071 }
1072 
1073 struct bpf_binary_header *
1074 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1075 			  unsigned int alignment,
1076 			  struct bpf_binary_header **rw_hdr,
1077 			  u8 **rw_image,
1078 			  bpf_jit_fill_hole_t bpf_fill_ill_insns);
1079 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1080 				 struct bpf_binary_header *ro_header,
1081 				 struct bpf_binary_header *rw_header);
1082 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1083 			      struct bpf_binary_header *rw_header);
1084 
1085 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1086 				struct bpf_jit_poke_descriptor *poke);
1087 
1088 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1089 			  const struct bpf_insn *insn, bool extra_pass,
1090 			  u64 *func_addr, bool *func_addr_fixed);
1091 
1092 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1093 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1094 
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1095 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1096 				u32 pass, void *image)
1097 {
1098 	pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1099 	       proglen, pass, image, current->comm, task_pid_nr(current));
1100 
1101 	if (image)
1102 		print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1103 			       16, 1, image, proglen, false);
1104 }
1105 
bpf_jit_is_ebpf(void)1106 static inline bool bpf_jit_is_ebpf(void)
1107 {
1108 # ifdef CONFIG_HAVE_EBPF_JIT
1109 	return true;
1110 # else
1111 	return false;
1112 # endif
1113 }
1114 
ebpf_jit_enabled(void)1115 static inline bool ebpf_jit_enabled(void)
1116 {
1117 	return bpf_jit_enable && bpf_jit_is_ebpf();
1118 }
1119 
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1120 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1121 {
1122 	return fp->jited && bpf_jit_is_ebpf();
1123 }
1124 
bpf_jit_blinding_enabled(struct bpf_prog * prog)1125 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1126 {
1127 	/* These are the prerequisites, should someone ever have the
1128 	 * idea to call blinding outside of them, we make sure to
1129 	 * bail out.
1130 	 */
1131 	if (!bpf_jit_is_ebpf())
1132 		return false;
1133 	if (!prog->jit_requested)
1134 		return false;
1135 	if (!bpf_jit_harden)
1136 		return false;
1137 	if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1138 		return false;
1139 
1140 	return true;
1141 }
1142 
bpf_jit_kallsyms_enabled(void)1143 static inline bool bpf_jit_kallsyms_enabled(void)
1144 {
1145 	/* There are a couple of corner cases where kallsyms should
1146 	 * not be enabled f.e. on hardening.
1147 	 */
1148 	if (bpf_jit_harden)
1149 		return false;
1150 	if (!bpf_jit_kallsyms)
1151 		return false;
1152 	if (bpf_jit_kallsyms == 1)
1153 		return true;
1154 
1155 	return false;
1156 }
1157 
1158 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1159 				 unsigned long *off, char *sym);
1160 bool is_bpf_text_address(unsigned long addr);
1161 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1162 		    char *sym);
1163 
1164 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1165 bpf_address_lookup(unsigned long addr, unsigned long *size,
1166 		   unsigned long *off, char **modname, char *sym)
1167 {
1168 	const char *ret = __bpf_address_lookup(addr, size, off, sym);
1169 
1170 	if (ret && modname)
1171 		*modname = NULL;
1172 	return ret;
1173 }
1174 
1175 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1176 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1177 
1178 #else /* CONFIG_BPF_JIT */
1179 
ebpf_jit_enabled(void)1180 static inline bool ebpf_jit_enabled(void)
1181 {
1182 	return false;
1183 }
1184 
bpf_jit_blinding_enabled(struct bpf_prog * prog)1185 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1186 {
1187 	return false;
1188 }
1189 
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1190 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1191 {
1192 	return false;
1193 }
1194 
1195 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1196 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1197 			    struct bpf_jit_poke_descriptor *poke)
1198 {
1199 	return -ENOTSUPP;
1200 }
1201 
bpf_jit_free(struct bpf_prog * fp)1202 static inline void bpf_jit_free(struct bpf_prog *fp)
1203 {
1204 	bpf_prog_unlock_free(fp);
1205 }
1206 
bpf_jit_kallsyms_enabled(void)1207 static inline bool bpf_jit_kallsyms_enabled(void)
1208 {
1209 	return false;
1210 }
1211 
1212 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1213 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1214 		     unsigned long *off, char *sym)
1215 {
1216 	return NULL;
1217 }
1218 
is_bpf_text_address(unsigned long addr)1219 static inline bool is_bpf_text_address(unsigned long addr)
1220 {
1221 	return false;
1222 }
1223 
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1224 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1225 				  char *type, char *sym)
1226 {
1227 	return -ERANGE;
1228 }
1229 
1230 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1231 bpf_address_lookup(unsigned long addr, unsigned long *size,
1232 		   unsigned long *off, char **modname, char *sym)
1233 {
1234 	return NULL;
1235 }
1236 
bpf_prog_kallsyms_add(struct bpf_prog * fp)1237 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1238 {
1239 }
1240 
bpf_prog_kallsyms_del(struct bpf_prog * fp)1241 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1242 {
1243 }
1244 
1245 #endif /* CONFIG_BPF_JIT */
1246 
1247 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1248 
1249 #define BPF_ANC		BIT(15)
1250 
bpf_needs_clear_a(const struct sock_filter * first)1251 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1252 {
1253 	switch (first->code) {
1254 	case BPF_RET | BPF_K:
1255 	case BPF_LD | BPF_W | BPF_LEN:
1256 		return false;
1257 
1258 	case BPF_LD | BPF_W | BPF_ABS:
1259 	case BPF_LD | BPF_H | BPF_ABS:
1260 	case BPF_LD | BPF_B | BPF_ABS:
1261 		if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1262 			return true;
1263 		return false;
1264 
1265 	default:
1266 		return true;
1267 	}
1268 }
1269 
bpf_anc_helper(const struct sock_filter * ftest)1270 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1271 {
1272 	BUG_ON(ftest->code & BPF_ANC);
1273 
1274 	switch (ftest->code) {
1275 	case BPF_LD | BPF_W | BPF_ABS:
1276 	case BPF_LD | BPF_H | BPF_ABS:
1277 	case BPF_LD | BPF_B | BPF_ABS:
1278 #define BPF_ANCILLARY(CODE)	case SKF_AD_OFF + SKF_AD_##CODE:	\
1279 				return BPF_ANC | SKF_AD_##CODE
1280 		switch (ftest->k) {
1281 		BPF_ANCILLARY(PROTOCOL);
1282 		BPF_ANCILLARY(PKTTYPE);
1283 		BPF_ANCILLARY(IFINDEX);
1284 		BPF_ANCILLARY(NLATTR);
1285 		BPF_ANCILLARY(NLATTR_NEST);
1286 		BPF_ANCILLARY(MARK);
1287 		BPF_ANCILLARY(QUEUE);
1288 		BPF_ANCILLARY(HATYPE);
1289 		BPF_ANCILLARY(RXHASH);
1290 		BPF_ANCILLARY(CPU);
1291 		BPF_ANCILLARY(ALU_XOR_X);
1292 		BPF_ANCILLARY(VLAN_TAG);
1293 		BPF_ANCILLARY(VLAN_TAG_PRESENT);
1294 		BPF_ANCILLARY(PAY_OFFSET);
1295 		BPF_ANCILLARY(RANDOM);
1296 		BPF_ANCILLARY(VLAN_TPID);
1297 		}
1298 		fallthrough;
1299 	default:
1300 		return ftest->code;
1301 	}
1302 }
1303 
1304 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1305 					   int k, unsigned int size);
1306 
bpf_tell_extensions(void)1307 static inline int bpf_tell_extensions(void)
1308 {
1309 	return SKF_AD_MAX;
1310 }
1311 
1312 struct bpf_sock_addr_kern {
1313 	struct sock *sk;
1314 	struct sockaddr *uaddr;
1315 	/* Temporary "register" to make indirect stores to nested structures
1316 	 * defined above. We need three registers to make such a store, but
1317 	 * only two (src and dst) are available at convert_ctx_access time
1318 	 */
1319 	u64 tmp_reg;
1320 	void *t_ctx;	/* Attach type specific context. */
1321 };
1322 
1323 struct bpf_sock_ops_kern {
1324 	struct	sock *sk;
1325 	union {
1326 		u32 args[4];
1327 		u32 reply;
1328 		u32 replylong[4];
1329 	};
1330 	struct sk_buff	*syn_skb;
1331 	struct sk_buff	*skb;
1332 	void	*skb_data_end;
1333 	u8	op;
1334 	u8	is_fullsock;
1335 	u8	remaining_opt_len;
1336 	u64	temp;			/* temp and everything after is not
1337 					 * initialized to 0 before calling
1338 					 * the BPF program. New fields that
1339 					 * should be initialized to 0 should
1340 					 * be inserted before temp.
1341 					 * temp is scratch storage used by
1342 					 * sock_ops_convert_ctx_access
1343 					 * as temporary storage of a register.
1344 					 */
1345 };
1346 
1347 struct bpf_sysctl_kern {
1348 	struct ctl_table_header *head;
1349 	struct ctl_table *table;
1350 	void *cur_val;
1351 	size_t cur_len;
1352 	void *new_val;
1353 	size_t new_len;
1354 	int new_updated;
1355 	int write;
1356 	loff_t *ppos;
1357 	/* Temporary "register" for indirect stores to ppos. */
1358 	u64 tmp_reg;
1359 };
1360 
1361 #define BPF_SOCKOPT_KERN_BUF_SIZE	32
1362 struct bpf_sockopt_buf {
1363 	u8		data[BPF_SOCKOPT_KERN_BUF_SIZE];
1364 };
1365 
1366 struct bpf_sockopt_kern {
1367 	struct sock	*sk;
1368 	u8		*optval;
1369 	u8		*optval_end;
1370 	s32		level;
1371 	s32		optname;
1372 	s32		optlen;
1373 	/* for retval in struct bpf_cg_run_ctx */
1374 	struct task_struct *current_task;
1375 	/* Temporary "register" for indirect stores to ppos. */
1376 	u64		tmp_reg;
1377 };
1378 
1379 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1380 
1381 struct bpf_sk_lookup_kern {
1382 	u16		family;
1383 	u16		protocol;
1384 	__be16		sport;
1385 	u16		dport;
1386 	struct {
1387 		__be32 saddr;
1388 		__be32 daddr;
1389 	} v4;
1390 	struct {
1391 		const struct in6_addr *saddr;
1392 		const struct in6_addr *daddr;
1393 	} v6;
1394 	struct sock	*selected_sk;
1395 	u32		ingress_ifindex;
1396 	bool		no_reuseport;
1397 };
1398 
1399 extern struct static_key_false bpf_sk_lookup_enabled;
1400 
1401 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1402  *
1403  * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1404  * SK_DROP. Their meaning is as follows:
1405  *
1406  *  SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1407  *  SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1408  *  SK_DROP                           : terminate lookup with -ECONNREFUSED
1409  *
1410  * This macro aggregates return values and selected sockets from
1411  * multiple BPF programs according to following rules in order:
1412  *
1413  *  1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1414  *     macro result is SK_PASS and last ctx.selected_sk is used.
1415  *  2. If any program returned SK_DROP return value,
1416  *     macro result is SK_DROP.
1417  *  3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1418  *
1419  * Caller must ensure that the prog array is non-NULL, and that the
1420  * array as well as the programs it contains remain valid.
1421  */
1422 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func)			\
1423 	({								\
1424 		struct bpf_sk_lookup_kern *_ctx = &(ctx);		\
1425 		struct bpf_prog_array_item *_item;			\
1426 		struct sock *_selected_sk = NULL;			\
1427 		bool _no_reuseport = false;				\
1428 		struct bpf_prog *_prog;					\
1429 		bool _all_pass = true;					\
1430 		u32 _ret;						\
1431 									\
1432 		migrate_disable();					\
1433 		_item = &(array)->items[0];				\
1434 		while ((_prog = READ_ONCE(_item->prog))) {		\
1435 			/* restore most recent selection */		\
1436 			_ctx->selected_sk = _selected_sk;		\
1437 			_ctx->no_reuseport = _no_reuseport;		\
1438 									\
1439 			_ret = func(_prog, _ctx);			\
1440 			if (_ret == SK_PASS && _ctx->selected_sk) {	\
1441 				/* remember last non-NULL socket */	\
1442 				_selected_sk = _ctx->selected_sk;	\
1443 				_no_reuseport = _ctx->no_reuseport;	\
1444 			} else if (_ret == SK_DROP && _all_pass) {	\
1445 				_all_pass = false;			\
1446 			}						\
1447 			_item++;					\
1448 		}							\
1449 		_ctx->selected_sk = _selected_sk;			\
1450 		_ctx->no_reuseport = _no_reuseport;			\
1451 		migrate_enable();					\
1452 		_all_pass || _selected_sk ? SK_PASS : SK_DROP;		\
1453 	 })
1454 
bpf_sk_lookup_run_v4(struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1455 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1456 					const __be32 saddr, const __be16 sport,
1457 					const __be32 daddr, const u16 dport,
1458 					const int ifindex, struct sock **psk)
1459 {
1460 	struct bpf_prog_array *run_array;
1461 	struct sock *selected_sk = NULL;
1462 	bool no_reuseport = false;
1463 
1464 	rcu_read_lock();
1465 	run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1466 	if (run_array) {
1467 		struct bpf_sk_lookup_kern ctx = {
1468 			.family		= AF_INET,
1469 			.protocol	= protocol,
1470 			.v4.saddr	= saddr,
1471 			.v4.daddr	= daddr,
1472 			.sport		= sport,
1473 			.dport		= dport,
1474 			.ingress_ifindex	= ifindex,
1475 		};
1476 		u32 act;
1477 
1478 		act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1479 		if (act == SK_PASS) {
1480 			selected_sk = ctx.selected_sk;
1481 			no_reuseport = ctx.no_reuseport;
1482 		} else {
1483 			selected_sk = ERR_PTR(-ECONNREFUSED);
1484 		}
1485 	}
1486 	rcu_read_unlock();
1487 	*psk = selected_sk;
1488 	return no_reuseport;
1489 }
1490 
1491 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,const int ifindex,struct sock ** psk)1492 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1493 					const struct in6_addr *saddr,
1494 					const __be16 sport,
1495 					const struct in6_addr *daddr,
1496 					const u16 dport,
1497 					const int ifindex, struct sock **psk)
1498 {
1499 	struct bpf_prog_array *run_array;
1500 	struct sock *selected_sk = NULL;
1501 	bool no_reuseport = false;
1502 
1503 	rcu_read_lock();
1504 	run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1505 	if (run_array) {
1506 		struct bpf_sk_lookup_kern ctx = {
1507 			.family		= AF_INET6,
1508 			.protocol	= protocol,
1509 			.v6.saddr	= saddr,
1510 			.v6.daddr	= daddr,
1511 			.sport		= sport,
1512 			.dport		= dport,
1513 			.ingress_ifindex	= ifindex,
1514 		};
1515 		u32 act;
1516 
1517 		act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1518 		if (act == SK_PASS) {
1519 			selected_sk = ctx.selected_sk;
1520 			no_reuseport = ctx.no_reuseport;
1521 		} else {
1522 			selected_sk = ERR_PTR(-ECONNREFUSED);
1523 		}
1524 	}
1525 	rcu_read_unlock();
1526 	*psk = selected_sk;
1527 	return no_reuseport;
1528 }
1529 #endif /* IS_ENABLED(CONFIG_IPV6) */
1530 
__bpf_xdp_redirect_map(struct bpf_map * map,u32 ifindex,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1531 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex,
1532 						  u64 flags, const u64 flag_mask,
1533 						  void *lookup_elem(struct bpf_map *map, u32 key))
1534 {
1535 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1536 	const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1537 
1538 	/* Lower bits of the flags are used as return code on lookup failure */
1539 	if (unlikely(flags & ~(action_mask | flag_mask)))
1540 		return XDP_ABORTED;
1541 
1542 	ri->tgt_value = lookup_elem(map, ifindex);
1543 	if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1544 		/* If the lookup fails we want to clear out the state in the
1545 		 * redirect_info struct completely, so that if an eBPF program
1546 		 * performs multiple lookups, the last one always takes
1547 		 * precedence.
1548 		 */
1549 		ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1550 		ri->map_type = BPF_MAP_TYPE_UNSPEC;
1551 		return flags & action_mask;
1552 	}
1553 
1554 	ri->tgt_index = ifindex;
1555 	ri->map_id = map->id;
1556 	ri->map_type = map->map_type;
1557 
1558 	if (flags & BPF_F_BROADCAST) {
1559 		WRITE_ONCE(ri->map, map);
1560 		ri->flags = flags;
1561 	} else {
1562 		WRITE_ONCE(ri->map, NULL);
1563 		ri->flags = 0;
1564 	}
1565 
1566 	return XDP_REDIRECT;
1567 }
1568 
1569 #endif /* __LINUX_FILTER_H__ */
1570