blob: 1f09c521adfe23be913bb5780438bb1ae4ce1cd5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Linux Socket Filter Data Structures
3 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004#ifndef __LINUX_FILTER_H__
5#define __LINUX_FILTER_H__
6
Daniel Borkmannb954d832014-09-10 15:01:02 +02007#include <stdarg.h>
8
Arun Sharma600634972011-07-26 16:09:06 -07009#include <linux/atomic.h>
Will Drewry0c5fe1b2012-04-12 16:47:53 -050010#include <linux/compat.h>
Zi Shen Lim9f12fbe2014-07-03 07:56:54 -070011#include <linux/skbuff.h>
Daniel Borkmannb954d832014-09-10 15:01:02 +020012#include <linux/linkage.h>
13#include <linux/printk.h>
Alexei Starovoitovd45ed4a2013-10-04 00:14:06 -070014#include <linux/workqueue.h>
Daniel Borkmannb13138e2015-07-30 12:42:49 +020015#include <linux/sched.h>
Daniel Borkmann4f3446b2016-05-13 19:08:32 +020016#include <linux/capability.h>
17
Alexei Starovoitovff936a02015-10-07 10:55:41 -070018#include <net/sch_generic.h>
Daniel Borkmannb954d832014-09-10 15:01:02 +020019
Daniel Borkmann60a3b222014-09-02 22:53:44 +020020#include <asm/cacheflush.h>
Daniel Borkmannb954d832014-09-10 15:01:02 +020021
22#include <uapi/linux/filter.h>
Alexei Starovoitovdaedfb22014-09-04 22:17:18 -070023#include <uapi/linux/bpf.h>
Daniel Borkmann60a3b222014-09-02 22:53:44 +020024
25struct sk_buff;
26struct sock;
27struct seccomp_data;
Alexei Starovoitov09756af2014-09-26 00:17:00 -070028struct bpf_prog_aux;
Heiko Carstens792d4b52011-05-22 07:08:11 +000029
Daniel Borkmann30743832014-05-01 18:34:19 +020030/* ArgX, context and stack frame pointer register positions. Note,
31 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
32 * calls in BPF_CALL instruction.
33 */
34#define BPF_REG_ARG1 BPF_REG_1
35#define BPF_REG_ARG2 BPF_REG_2
36#define BPF_REG_ARG3 BPF_REG_3
37#define BPF_REG_ARG4 BPF_REG_4
38#define BPF_REG_ARG5 BPF_REG_5
39#define BPF_REG_CTX BPF_REG_6
40#define BPF_REG_FP BPF_REG_10
41
42/* Additional register mappings for converted user programs. */
43#define BPF_REG_A BPF_REG_0
44#define BPF_REG_X BPF_REG_7
45#define BPF_REG_TMP BPF_REG_8
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +010046
Daniel Borkmann4f3446b2016-05-13 19:08:32 +020047/* Kernel hidden auxiliary/helper register for hardening step.
48 * Only used by eBPF JITs. It's nothing more than a temporary
49 * register that JITs use internally, only that here it's part
50 * of eBPF instructions that have been rewritten for blinding
51 * constants. See JIT pre-step in bpf_jit_blind_constants().
52 */
53#define BPF_REG_AX MAX_BPF_REG
54#define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
55
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +010056/* BPF program can access up to 512 bytes of stack space. */
57#define MAX_BPF_STACK 512
58
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020059/* Helper macros for filter block array initializers. */
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070060
Alexei Starovoitove430f342014-06-06 14:46:06 -070061/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070062
Alexei Starovoitove430f342014-06-06 14:46:06 -070063#define BPF_ALU64_REG(OP, DST, SRC) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -070064 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020065 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
Alexei Starovoitove430f342014-06-06 14:46:06 -070066 .dst_reg = DST, \
67 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020068 .off = 0, \
69 .imm = 0 })
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070070
Alexei Starovoitove430f342014-06-06 14:46:06 -070071#define BPF_ALU32_REG(OP, DST, SRC) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -070072 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020073 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
Alexei Starovoitove430f342014-06-06 14:46:06 -070074 .dst_reg = DST, \
75 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020076 .off = 0, \
77 .imm = 0 })
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070078
Alexei Starovoitove430f342014-06-06 14:46:06 -070079/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070080
Alexei Starovoitove430f342014-06-06 14:46:06 -070081#define BPF_ALU64_IMM(OP, DST, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -070082 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020083 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
Alexei Starovoitove430f342014-06-06 14:46:06 -070084 .dst_reg = DST, \
85 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020086 .off = 0, \
87 .imm = IMM })
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070088
Alexei Starovoitove430f342014-06-06 14:46:06 -070089#define BPF_ALU32_IMM(OP, DST, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -070090 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020091 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
Alexei Starovoitove430f342014-06-06 14:46:06 -070092 .dst_reg = DST, \
93 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020094 .off = 0, \
95 .imm = IMM })
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070096
Daniel Borkmannf8f6d672014-05-29 10:22:51 +020097/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
Alexei Starovoitov9739eef2014-05-08 14:10:51 -070098
Alexei Starovoitove430f342014-06-06 14:46:06 -070099#define BPF_ENDIAN(TYPE, DST, LEN) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700100 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200101 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700102 .dst_reg = DST, \
103 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200104 .off = 0, \
105 .imm = LEN })
106
Alexei Starovoitove430f342014-06-06 14:46:06 -0700107/* Short form of mov, dst_reg = src_reg */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200108
Alexei Starovoitove430f342014-06-06 14:46:06 -0700109#define BPF_MOV64_REG(DST, SRC) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700110 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200111 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700112 .dst_reg = DST, \
113 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200114 .off = 0, \
115 .imm = 0 })
116
Alexei Starovoitove430f342014-06-06 14:46:06 -0700117#define BPF_MOV32_REG(DST, SRC) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700118 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200119 .code = BPF_ALU | BPF_MOV | BPF_X, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700120 .dst_reg = DST, \
121 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200122 .off = 0, \
123 .imm = 0 })
124
Alexei Starovoitove430f342014-06-06 14:46:06 -0700125/* Short form of mov, dst_reg = imm32 */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200126
Alexei Starovoitove430f342014-06-06 14:46:06 -0700127#define BPF_MOV64_IMM(DST, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700128 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200129 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700130 .dst_reg = DST, \
131 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200132 .off = 0, \
133 .imm = IMM })
134
Alexei Starovoitove430f342014-06-06 14:46:06 -0700135#define BPF_MOV32_IMM(DST, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700136 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200137 .code = BPF_ALU | BPF_MOV | BPF_K, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700138 .dst_reg = DST, \
139 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200140 .off = 0, \
141 .imm = IMM })
142
Alexei Starovoitov02ab6952014-09-04 22:17:17 -0700143/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
144#define BPF_LD_IMM64(DST, IMM) \
145 BPF_LD_IMM64_RAW(DST, 0, IMM)
146
147#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
148 ((struct bpf_insn) { \
149 .code = BPF_LD | BPF_DW | BPF_IMM, \
150 .dst_reg = DST, \
151 .src_reg = SRC, \
152 .off = 0, \
153 .imm = (__u32) (IMM) }), \
154 ((struct bpf_insn) { \
155 .code = 0, /* zero is reserved opcode */ \
156 .dst_reg = 0, \
157 .src_reg = 0, \
158 .off = 0, \
159 .imm = ((__u64) (IMM)) >> 32 })
160
Alexei Starovoitov0246e642014-09-26 00:17:04 -0700161/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
162#define BPF_LD_MAP_FD(DST, MAP_FD) \
163 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
164
Alexei Starovoitove430f342014-06-06 14:46:06 -0700165/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200166
Alexei Starovoitove430f342014-06-06 14:46:06 -0700167#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700168 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200169 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700170 .dst_reg = DST, \
171 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200172 .off = 0, \
173 .imm = IMM })
174
Alexei Starovoitove430f342014-06-06 14:46:06 -0700175#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700176 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200177 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700178 .dst_reg = DST, \
179 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200180 .off = 0, \
181 .imm = IMM })
182
Alexei Starovoitove430f342014-06-06 14:46:06 -0700183/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200184
Alexei Starovoitove430f342014-06-06 14:46:06 -0700185#define BPF_LD_ABS(SIZE, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700186 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200187 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700188 .dst_reg = 0, \
189 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200190 .off = 0, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700191 .imm = IMM })
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200192
Alexei Starovoitove430f342014-06-06 14:46:06 -0700193/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200194
Alexei Starovoitove430f342014-06-06 14:46:06 -0700195#define BPF_LD_IND(SIZE, SRC, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700196 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200197 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700198 .dst_reg = 0, \
199 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200200 .off = 0, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700201 .imm = IMM })
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200202
Alexei Starovoitove430f342014-06-06 14:46:06 -0700203/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200204
Alexei Starovoitove430f342014-06-06 14:46:06 -0700205#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700206 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200207 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700208 .dst_reg = DST, \
209 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200210 .off = OFF, \
211 .imm = 0 })
212
Alexei Starovoitove430f342014-06-06 14:46:06 -0700213/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
214
215#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700216 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200217 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700218 .dst_reg = DST, \
219 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200220 .off = OFF, \
221 .imm = 0 })
222
Michael Holzheucffc6422015-05-11 22:22:44 -0700223/* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
224
225#define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
226 ((struct bpf_insn) { \
227 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
228 .dst_reg = DST, \
229 .src_reg = SRC, \
230 .off = OFF, \
231 .imm = 0 })
232
Alexei Starovoitove430f342014-06-06 14:46:06 -0700233/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200234
Alexei Starovoitove430f342014-06-06 14:46:06 -0700235#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700236 ((struct bpf_insn) { \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700237 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
238 .dst_reg = DST, \
239 .src_reg = 0, \
240 .off = OFF, \
241 .imm = IMM })
242
243/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
244
245#define BPF_JMP_REG(OP, DST, SRC, OFF) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700246 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200247 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700248 .dst_reg = DST, \
249 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200250 .off = OFF, \
251 .imm = 0 })
252
Alexei Starovoitove430f342014-06-06 14:46:06 -0700253/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200254
Alexei Starovoitove430f342014-06-06 14:46:06 -0700255#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700256 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200257 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700258 .dst_reg = DST, \
259 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200260 .off = OFF, \
261 .imm = IMM })
262
263/* Function call */
264
265#define BPF_EMIT_CALL(FUNC) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700266 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200267 .code = BPF_JMP | BPF_CALL, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700268 .dst_reg = 0, \
269 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200270 .off = 0, \
271 .imm = ((FUNC) - __bpf_call_base) })
272
273/* Raw code statement block */
274
Alexei Starovoitove430f342014-06-06 14:46:06 -0700275#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700276 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200277 .code = CODE, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700278 .dst_reg = DST, \
279 .src_reg = SRC, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200280 .off = OFF, \
281 .imm = IMM })
282
283/* Program exit */
284
285#define BPF_EXIT_INSN() \
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700286 ((struct bpf_insn) { \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200287 .code = BPF_JMP | BPF_EXIT, \
Alexei Starovoitove430f342014-06-06 14:46:06 -0700288 .dst_reg = 0, \
289 .src_reg = 0, \
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200290 .off = 0, \
291 .imm = 0 })
292
Daniel Borkmanna4afd37b2015-05-13 13:12:43 +0200293/* Internal classic blocks for direct assignment */
294
295#define __BPF_STMT(CODE, K) \
296 ((struct sock_filter) BPF_STMT(CODE, K))
297
298#define __BPF_JUMP(CODE, K, JT, JF) \
299 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
300
Daniel Borkmannf8f6d672014-05-29 10:22:51 +0200301#define bytes_to_bpf_size(bytes) \
302({ \
303 int bpf_size = -EINVAL; \
304 \
305 if (bytes == sizeof(u8)) \
306 bpf_size = BPF_B; \
307 else if (bytes == sizeof(u16)) \
308 bpf_size = BPF_H; \
309 else if (bytes == sizeof(u32)) \
310 bpf_size = BPF_W; \
311 else if (bytes == sizeof(u64)) \
312 bpf_size = BPF_DW; \
313 \
314 bpf_size; \
315})
Alexei Starovoitov9739eef2014-05-08 14:10:51 -0700316
Daniel Borkmannf035a512016-09-09 02:45:29 +0200317#define BPF_SIZEOF(type) \
318 ({ \
319 const int __size = bytes_to_bpf_size(sizeof(type)); \
320 BUILD_BUG_ON(__size < 0); \
321 __size; \
322 })
323
324#define BPF_FIELD_SIZEOF(type, field) \
325 ({ \
326 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
327 BUILD_BUG_ON(__size < 0); \
328 __size; \
329 })
330
Daniel Borkmannf3694e02016-09-09 02:45:31 +0200331#define __BPF_MAP_0(m, v, ...) v
332#define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
333#define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
334#define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
335#define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
336#define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
337
338#define __BPF_REG_0(...) __BPF_PAD(5)
339#define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
340#define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
341#define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
342#define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
343#define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
344
345#define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
346#define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
347
348#define __BPF_CAST(t, a) \
349 (__force t) \
350 (__force \
351 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
352 (unsigned long)0, (t)0))) a
353#define __BPF_V void
354#define __BPF_N
355
356#define __BPF_DECL_ARGS(t, a) t a
357#define __BPF_DECL_REGS(t, a) u64 a
358
359#define __BPF_PAD(n) \
360 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
361 u64, __ur_3, u64, __ur_4, u64, __ur_5)
362
363#define BPF_CALL_x(x, name, ...) \
364 static __always_inline \
365 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
366 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
367 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
368 { \
369 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
370 } \
371 static __always_inline \
372 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
373
374#define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
375#define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
376#define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
377#define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
378#define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
379#define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
380
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100381#ifdef CONFIG_COMPAT
382/* A struct sock_filter is architecture independent. */
Will Drewry0c5fe1b2012-04-12 16:47:53 -0500383struct compat_sock_fprog {
384 u16 len;
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100385 compat_uptr_t filter; /* struct sock_filter * */
Will Drewry0c5fe1b2012-04-12 16:47:53 -0500386};
387#endif
388
Daniel Borkmanna3ea2692014-03-28 18:58:19 +0100389struct sock_fprog_kern {
390 u16 len;
391 struct sock_filter *filter;
392};
393
Daniel Borkmann738cbe72014-09-08 08:04:47 +0200394struct bpf_binary_header {
395 unsigned int pages;
396 u8 image[];
397};
398
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700399struct bpf_prog {
Daniel Borkmann286aad32014-09-08 08:04:49 +0200400 u16 pages; /* Number of allocated pages */
Daniel Borkmanna91263d2015-09-30 01:41:50 +0200401 kmemcheck_bitfield_begin(meta);
402 u16 jited:1, /* Is our filter JIT'ed? */
Daniel Borkmannc46646d2015-09-30 01:41:51 +0200403 gpl_compatible:1, /* Is filter GPL compatible? */
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700404 cb_access:1, /* Is control block accessed? */
Daniel Borkmannc46646d2015-09-30 01:41:51 +0200405 dst_needed:1; /* Do we need dst entry? */
Daniel Borkmanna91263d2015-09-30 01:41:50 +0200406 kmemcheck_bitfield_end(meta);
Daniel Borkmann286aad32014-09-08 08:04:49 +0200407 u32 len; /* Number of filter blocks */
Daniel Borkmann24701ec2015-03-01 12:31:47 +0100408 enum bpf_prog_type type; /* Type of BPF program */
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700409 struct bpf_prog_aux *aux; /* Auxiliary fields */
Daniel Borkmann24701ec2015-03-01 12:31:47 +0100410 struct sock_fprog_kern *orig_prog; /* Original BPF program */
Eric Dumazet0a148422011-04-20 09:27:32 +0000411 unsigned int (*bpf_func)(const struct sk_buff *skb,
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700412 const struct bpf_insn *filter);
Daniel Borkmann60a3b222014-09-02 22:53:44 +0200413 /* Instructions for interpreter */
Alexei Starovoitovd45ed4a2013-10-04 00:14:06 -0700414 union {
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100415 struct sock_filter insns[0];
Alexei Starovoitov2695fb52014-07-24 16:38:21 -0700416 struct bpf_insn insnsi[0];
Alexei Starovoitovd45ed4a2013-10-04 00:14:06 -0700417 };
Stephen Hemmingerb7156312008-04-10 01:33:47 -0700418};
419
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700420struct sk_filter {
421 atomic_t refcnt;
422 struct rcu_head rcu;
423 struct bpf_prog *prog;
424};
425
426#define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
427
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100428#define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
429
Alexei Starovoitovdb58ba42016-05-05 19:49:12 -0700430struct bpf_skb_data_end {
431 struct qdisc_skb_cb qdisc_cb;
432 void *data_end;
433};
434
Brenden Blanco6a773a12016-07-19 12:16:47 -0700435struct xdp_buff {
436 void *data;
437 void *data_end;
438};
439
Alexei Starovoitovdb58ba42016-05-05 19:49:12 -0700440/* compute the linear packet data range [data, data_end) which
441 * will be accessed by cls_bpf and act_bpf programs
442 */
443static inline void bpf_compute_data_end(struct sk_buff *skb)
444{
445 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
446
447 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
448 cb->data_end = skb->data + skb_headlen(skb);
449}
450
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100451static inline u8 *bpf_skb_cb(struct sk_buff *skb)
452{
453 /* eBPF programs may read/write skb->cb[] area to transfer meta
454 * data between tail calls. Since this also needs to work with
455 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
456 *
457 * In some socket filter cases, the cb unfortunately needs to be
458 * saved/restored so that protocol specific skb->cb[] data won't
459 * be lost. In any case, due to unpriviledged eBPF programs
460 * attached to sockets, we need to clear the bpf_skb_cb() area
461 * to not leak previous contents to user space.
462 */
463 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
464 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
465 FIELD_SIZEOF(struct qdisc_skb_cb, data));
466
467 return qdisc_skb_cb(skb)->data;
468}
469
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700470static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
471 struct sk_buff *skb)
472{
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100473 u8 *cb_data = bpf_skb_cb(skb);
474 u8 cb_saved[BPF_SKB_CB_LEN];
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700475 u32 res;
476
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700477 if (unlikely(prog->cb_access)) {
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100478 memcpy(cb_saved, cb_data, sizeof(cb_saved));
479 memset(cb_data, 0, sizeof(cb_saved));
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700480 }
481
482 res = BPF_PROG_RUN(prog, skb);
483
484 if (unlikely(prog->cb_access))
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100485 memcpy(cb_data, cb_saved, sizeof(cb_saved));
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700486
487 return res;
488}
489
490static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
491 struct sk_buff *skb)
492{
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100493 u8 *cb_data = bpf_skb_cb(skb);
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700494
495 if (unlikely(prog->cb_access))
Daniel Borkmann01dd1942016-01-06 22:32:16 +0100496 memset(cb_data, 0, BPF_SKB_CB_LEN);
497
Alexei Starovoitovff936a02015-10-07 10:55:41 -0700498 return BPF_PROG_RUN(prog, skb);
499}
500
Brenden Blanco6a773a12016-07-19 12:16:47 -0700501static inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
502 struct xdp_buff *xdp)
503{
504 u32 ret;
505
506 rcu_read_lock();
507 ret = BPF_PROG_RUN(prog, (void *)xdp);
508 rcu_read_unlock();
509
510 return ret;
511}
512
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700513static inline unsigned int bpf_prog_size(unsigned int proglen)
Stephen Hemmingerb7156312008-04-10 01:33:47 -0700514{
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700515 return max(sizeof(struct bpf_prog),
516 offsetof(struct bpf_prog, insns[proglen]));
Stephen Hemmingerb7156312008-04-10 01:33:47 -0700517}
518
Daniel Borkmann7b36f922015-07-30 12:42:47 +0200519static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
520{
521 /* When classic BPF programs have been loaded and the arch
522 * does not have a classic BPF JIT (anymore), they have been
523 * converted via bpf_migrate_filter() to eBPF and thus always
524 * have an unspec program type.
525 */
526 return prog->type == BPF_PROG_TYPE_UNSPEC;
527}
528
Alexei Starovoitov009937e2014-07-30 20:34:13 -0700529#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
Daniel Borkmanna3ea2692014-03-28 18:58:19 +0100530
Daniel Borkmann60a3b222014-09-02 22:53:44 +0200531#ifdef CONFIG_DEBUG_SET_MODULE_RONX
532static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
533{
534 set_memory_ro((unsigned long)fp, fp->pages);
535}
536
537static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
538{
539 set_memory_rw((unsigned long)fp, fp->pages);
540}
541#else
542static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
543{
544}
545
546static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
547{
548}
549#endif /* CONFIG_DEBUG_SET_MODULE_RONX */
550
Willem de Bruijnf4979fc2016-07-12 18:18:56 -0400551int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
552static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
553{
554 return sk_filter_trim_cap(sk, skb, 1);
555}
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100556
Daniel Borkmannd1c55ab2016-05-13 19:08:31 +0200557struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700558void bpf_prog_free(struct bpf_prog *fp);
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100559
Daniel Borkmann60a3b222014-09-02 22:53:44 +0200560struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
561struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
562 gfp_t gfp_extra_flags);
563void __bpf_prog_free(struct bpf_prog *fp);
564
565static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
566{
567 bpf_prog_unlock_ro(fp);
568 __bpf_prog_free(fp);
569}
570
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200571typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
572 unsigned int flen);
573
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700574int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200575int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
Daniel Borkmannbab18992015-10-02 15:17:33 +0200576 bpf_aux_classic_check_t trans, bool save_orig);
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700577void bpf_prog_destroy(struct bpf_prog *fp);
Daniel Borkmanna3ea2692014-03-28 18:58:19 +0100578
Daniel Borkmannfbc907f2014-03-28 18:58:20 +0100579int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
Alexei Starovoitov89aa0752014-12-01 15:06:35 -0800580int sk_attach_bpf(u32 ufd, struct sock *sk);
Craig Gallek538950a2016-01-04 17:41:47 -0500581int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
582int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
Daniel Borkmannfbc907f2014-03-28 18:58:20 +0100583int sk_detach_filter(struct sock *sk);
Daniel Borkmannfbc907f2014-03-28 18:58:20 +0100584int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
585 unsigned int len);
Daniel Borkmannfbc907f2014-03-28 18:58:20 +0100586
Alexei Starovoitov278571b2014-07-30 20:34:12 -0700587bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
Daniel Borkmannfbc907f2014-03-28 18:58:20 +0100588void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
Eric Dumazet0a148422011-04-20 09:27:32 +0000589
Alexei Starovoitov62258272014-05-13 19:50:46 -0700590u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
Daniel Borkmannd1c55ab2016-05-13 19:08:31 +0200591
592struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
Alexei Starovoitov4e10df92015-07-20 20:34:18 -0700593bool bpf_helper_changes_skb_data(void *func);
Alexei Starovoitov62258272014-05-13 19:50:46 -0700594
Daniel Borkmannc237ee52016-05-13 19:08:30 +0200595struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
596 const struct bpf_insn *patch, u32 len);
Brenden Blanco6a773a12016-07-19 12:16:47 -0700597void bpf_warn_invalid_xdp_action(u32 act);
Daniel Borkmannc237ee52016-05-13 19:08:30 +0200598
Daniel Borkmannb954d832014-09-10 15:01:02 +0200599#ifdef CONFIG_BPF_JIT
Daniel Borkmannc94987e2016-05-13 19:08:27 +0200600extern int bpf_jit_enable;
Daniel Borkmann4f3446b2016-05-13 19:08:32 +0200601extern int bpf_jit_harden;
Daniel Borkmannc94987e2016-05-13 19:08:27 +0200602
Daniel Borkmannb954d832014-09-10 15:01:02 +0200603typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
604
605struct bpf_binary_header *
606bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
607 unsigned int alignment,
608 bpf_jit_fill_hole_t bpf_fill_ill_insns);
609void bpf_jit_binary_free(struct bpf_binary_header *hdr);
610
611void bpf_jit_compile(struct bpf_prog *fp);
612void bpf_jit_free(struct bpf_prog *fp);
613
Daniel Borkmann4f3446b2016-05-13 19:08:32 +0200614struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
615void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
616
Daniel Borkmannb954d832014-09-10 15:01:02 +0200617static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
618 u32 pass, void *image)
619{
Daniel Borkmannb13138e2015-07-30 12:42:49 +0200620 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
621 proglen, pass, image, current->comm, task_pid_nr(current));
622
Daniel Borkmannb954d832014-09-10 15:01:02 +0200623 if (image)
624 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
625 16, 1, image, proglen, false);
626}
Daniel Borkmann4f3446b2016-05-13 19:08:32 +0200627
628static inline bool bpf_jit_is_ebpf(void)
629{
630# ifdef CONFIG_HAVE_EBPF_JIT
631 return true;
632# else
633 return false;
634# endif
635}
636
637static inline bool bpf_jit_blinding_enabled(void)
638{
639 /* These are the prerequisites, should someone ever have the
640 * idea to call blinding outside of them, we make sure to
641 * bail out.
642 */
643 if (!bpf_jit_is_ebpf())
644 return false;
645 if (!bpf_jit_enable)
646 return false;
647 if (!bpf_jit_harden)
648 return false;
649 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
650 return false;
651
652 return true;
653}
Daniel Borkmannb954d832014-09-10 15:01:02 +0200654#else
655static inline void bpf_jit_compile(struct bpf_prog *fp)
656{
657}
658
659static inline void bpf_jit_free(struct bpf_prog *fp)
660{
661 bpf_prog_unlock_free(fp);
662}
663#endif /* CONFIG_BPF_JIT */
664
Daniel Borkmann34805932014-05-29 10:22:50 +0200665#define BPF_ANC BIT(15)
666
Rabin Vincent55795ef2016-01-05 16:23:07 +0100667static inline bool bpf_needs_clear_a(const struct sock_filter *first)
668{
669 switch (first->code) {
670 case BPF_RET | BPF_K:
671 case BPF_LD | BPF_W | BPF_LEN:
672 return false;
673
674 case BPF_LD | BPF_W | BPF_ABS:
675 case BPF_LD | BPF_H | BPF_ABS:
676 case BPF_LD | BPF_B | BPF_ABS:
677 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
678 return true;
679 return false;
680
681 default:
682 return true;
683 }
684}
685
Daniel Borkmann34805932014-05-29 10:22:50 +0200686static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
687{
688 BUG_ON(ftest->code & BPF_ANC);
689
690 switch (ftest->code) {
691 case BPF_LD | BPF_W | BPF_ABS:
692 case BPF_LD | BPF_H | BPF_ABS:
693 case BPF_LD | BPF_B | BPF_ABS:
694#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
695 return BPF_ANC | SKF_AD_##CODE
696 switch (ftest->k) {
697 BPF_ANCILLARY(PROTOCOL);
698 BPF_ANCILLARY(PKTTYPE);
699 BPF_ANCILLARY(IFINDEX);
700 BPF_ANCILLARY(NLATTR);
701 BPF_ANCILLARY(NLATTR_NEST);
702 BPF_ANCILLARY(MARK);
703 BPF_ANCILLARY(QUEUE);
704 BPF_ANCILLARY(HATYPE);
705 BPF_ANCILLARY(RXHASH);
706 BPF_ANCILLARY(CPU);
707 BPF_ANCILLARY(ALU_XOR_X);
708 BPF_ANCILLARY(VLAN_TAG);
709 BPF_ANCILLARY(VLAN_TAG_PRESENT);
710 BPF_ANCILLARY(PAY_OFFSET);
711 BPF_ANCILLARY(RANDOM);
Michal Sekletar27cd5452015-03-24 14:48:41 +0100712 BPF_ANCILLARY(VLAN_TPID);
Daniel Borkmann34805932014-05-29 10:22:50 +0200713 }
714 /* Fallthrough. */
715 default:
716 return ftest->code;
717 }
718}
719
Zi Shen Lim9f12fbe2014-07-03 07:56:54 -0700720void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
721 int k, unsigned int size);
722
723static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
724 unsigned int size, void *buffer)
725{
726 if (k >= 0)
727 return skb_header_pointer(skb, k, size, buffer);
728
729 return bpf_internal_load_pointer_neg_helper(skb, k, size);
730}
731
Michal Sekletarea02f942014-01-17 17:09:45 +0100732static inline int bpf_tell_extensions(void)
733{
Daniel Borkmann37692292014-01-21 00:19:37 +0100734 return SKF_AD_MAX;
Michal Sekletarea02f942014-01-17 17:09:45 +0100735}
736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737#endif /* __LINUX_FILTER_H__ */