aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorEric Dumazet <eric.dumazet@gmail.com>2011-04-20 05:27:32 -0400
committerDavid S. Miller <davem@davemloft.net>2011-04-28 02:05:08 -0400
commit0a14842f5a3c0e88a1e59fac5c3025db39721f74 (patch)
tree4d0832c4c9ced2503e2d899eb56952f87511d4ab
parentb678027cb77b079bc8e5b94172995d173bdb494b (diff)
net: filter: Just In Time compiler for x86-64
In order to speedup packet filtering, here is an implementation of a JIT compiler for x86_64 It is disabled by default, and must be enabled by the admin. echo 1 >/proc/sys/net/core/bpf_jit_enable It uses module_alloc() and module_free() to get memory in the 2GB text kernel range since we call helpers functions from the generated code. EAX : BPF A accumulator EBX : BPF X accumulator RDI : pointer to skb (first argument given to JIT function) RBP : frame pointer (even if CONFIG_FRAME_POINTER=n) r9d : skb->len - skb->data_len (headlen) r8 : skb->data To get a trace of generated code, use : echo 2 >/proc/sys/net/core/bpf_jit_enable Example of generated code : # tcpdump -p -n -s 0 -i eth1 host 192.168.20.0/24 flen=18 proglen=147 pass=3 image=ffffffffa00b5000 JIT code: ffffffffa00b5000: 55 48 89 e5 48 83 ec 60 48 89 5d f8 44 8b 4f 60 JIT code: ffffffffa00b5010: 44 2b 4f 64 4c 8b 87 b8 00 00 00 be 0c 00 00 00 JIT code: ffffffffa00b5020: e8 24 7b f7 e0 3d 00 08 00 00 75 28 be 1a 00 00 JIT code: ffffffffa00b5030: 00 e8 fe 7a f7 e0 24 00 3d 00 14 a8 c0 74 49 be JIT code: ffffffffa00b5040: 1e 00 00 00 e8 eb 7a f7 e0 24 00 3d 00 14 a8 c0 JIT code: ffffffffa00b5050: 74 36 eb 3b 3d 06 08 00 00 74 07 3d 35 80 00 00 JIT code: ffffffffa00b5060: 75 2d be 1c 00 00 00 e8 c8 7a f7 e0 24 00 3d 00 JIT code: ffffffffa00b5070: 14 a8 c0 74 13 be 26 00 00 00 e8 b5 7a f7 e0 24 JIT code: ffffffffa00b5080: 00 3d 00 14 a8 c0 75 07 b8 ff ff 00 00 eb 02 31 JIT code: ffffffffa00b5090: c0 c9 c3 BPF program is 144 bytes long, so native program is almost same size ;) (000) ldh [12] (001) jeq #0x800 jt 2 jf 8 (002) ld [26] (003) and #0xffffff00 (004) jeq #0xc0a81400 jt 16 jf 5 (005) ld [30] (006) and #0xffffff00 (007) jeq #0xc0a81400 jt 16 jf 17 (008) jeq #0x806 jt 10 jf 9 (009) jeq #0x8035 jt 10 jf 17 (010) ld [28] (011) and #0xffffff00 (012) jeq #0xc0a81400 jt 16 jf 13 (013) ld [38] (014) and #0xffffff00 (015) jeq #0xc0a81400 jt 16 jf 17 (016) ret #65535 (017) ret #0 Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Ben Hutchings <bhutchings@solarflare.com> Cc: Hagen Paul Pfeifer <hagen@jauu.net> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat
-rw-r--r--Documentation/sysctl/net.txt11
-rw-r--r--MAINTAINERS1
-rw-r--r--arch/x86/Kbuild1
-rw-r--r--arch/x86/Kconfig1
-rw-r--r--arch/x86/net/Makefile4
-rw-r--r--arch/x86/net/bpf_jit.S140
-rw-r--r--arch/x86/net/bpf_jit_comp.c654
-rw-r--r--include/linux/filter.h76
-rw-r--r--include/linux/netdevice.h1
-rw-r--r--include/linux/skbuff.h2
-rw-r--r--net/Kconfig13
-rw-r--r--net/core/filter.c65
-rw-r--r--net/core/sysctl_net_core.c9
-rw-r--r--net/packet/af_packet.c2
14 files changed, 918 insertions, 62 deletions
diff --git a/Documentation/sysctl/net.txt b/Documentation/sysctl/net.txt
index cbd05ffc606b..3201a7097e4d 100644
--- a/Documentation/sysctl/net.txt
+++ b/Documentation/sysctl/net.txt
@@ -32,6 +32,17 @@ Table : Subdirectories in /proc/sys/net
321. /proc/sys/net/core - Network core options 321. /proc/sys/net/core - Network core options
33------------------------------------------------------- 33-------------------------------------------------------
34 34
35bpf_jit_enable
36--------------
37
38This enables Berkeley Packet Filter Just in Time compiler.
39Currently supported on x86_64 architecture, bpf_jit provides a framework
40to speed packet filtering, the one used by tcpdump/libpcap for example.
41Values :
42 0 - disable the JIT (default value)
43 1 - enable the JIT
44 2 - enable the JIT and ask the compiler to emit traces on kernel log.
45
35rmem_default 46rmem_default
36------------ 47------------
37 48
diff --git a/MAINTAINERS b/MAINTAINERS
index b5266ad50167..17c0917a26ea 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -4372,6 +4372,7 @@ S: Maintained
4372F: net/ipv4/ 4372F: net/ipv4/
4373F: net/ipv6/ 4373F: net/ipv6/
4374F: include/net/ip* 4374F: include/net/ip*
4375F: arch/x86/net/*
4375 4376
4376NETWORKING [LABELED] (NetLabel, CIPSO, Labeled IPsec, SECMARK) 4377NETWORKING [LABELED] (NetLabel, CIPSO, Labeled IPsec, SECMARK)
4377M: Paul Moore <paul.moore@hp.com> 4378M: Paul Moore <paul.moore@hp.com>
diff --git a/arch/x86/Kbuild b/arch/x86/Kbuild
index 0e103236b754..0e9dec6cadd1 100644
--- a/arch/x86/Kbuild
+++ b/arch/x86/Kbuild
@@ -15,3 +15,4 @@ obj-y += vdso/
15obj-$(CONFIG_IA32_EMULATION) += ia32/ 15obj-$(CONFIG_IA32_EMULATION) += ia32/
16 16
17obj-y += platform/ 17obj-y += platform/
18obj-y += net/
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index cc6c53a95bfd..855a1bdc437d 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -72,6 +72,7 @@ config X86
72 select IRQ_FORCED_THREADING 72 select IRQ_FORCED_THREADING
73 select USE_GENERIC_SMP_HELPERS if SMP 73 select USE_GENERIC_SMP_HELPERS if SMP
74 select ARCH_NO_SYSDEV_OPS 74 select ARCH_NO_SYSDEV_OPS
75 select HAVE_BPF_JIT if X86_64
75 76
76config INSTRUCTION_DECODER 77config INSTRUCTION_DECODER
77 def_bool (KPROBES || PERF_EVENTS) 78 def_bool (KPROBES || PERF_EVENTS)
diff --git a/arch/x86/net/Makefile b/arch/x86/net/Makefile
new file mode 100644
index 000000000000..90568c33ddb0
--- /dev/null
+++ b/arch/x86/net/Makefile
@@ -0,0 +1,4 @@
1#
2# Arch-specific network modules
3#
4obj-$(CONFIG_BPF_JIT) += bpf_jit.o bpf_jit_comp.o
diff --git a/arch/x86/net/bpf_jit.S b/arch/x86/net/bpf_jit.S
new file mode 100644
index 000000000000..66870223f8c5
--- /dev/null
+++ b/arch/x86/net/bpf_jit.S
@@ -0,0 +1,140 @@
1/* bpf_jit.S : BPF JIT helper functions
2 *
3 * Copyright (C) 2011 Eric Dumazet (eric.dumazet@gmail.com)
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; version 2
8 * of the License.
9 */
10#include <linux/linkage.h>
11#include <asm/dwarf2.h>
12
13/*
14 * Calling convention :
15 * rdi : skb pointer
16 * esi : offset of byte(s) to fetch in skb (can be scratched)
17 * r8 : copy of skb->data
18 * r9d : hlen = skb->len - skb->data_len
19 */
20#define SKBDATA %r8
21
22sk_load_word_ind:
23 .globl sk_load_word_ind
24
25 add %ebx,%esi /* offset += X */
26# test %esi,%esi /* if (offset < 0) goto bpf_error; */
27 js bpf_error
28
29sk_load_word:
30 .globl sk_load_word
31
32 mov %r9d,%eax # hlen
33 sub %esi,%eax # hlen - offset
34 cmp $3,%eax
35 jle bpf_slow_path_word
36 mov (SKBDATA,%rsi),%eax
37 bswap %eax /* ntohl() */
38 ret
39
40
41sk_load_half_ind:
42 .globl sk_load_half_ind
43
44 add %ebx,%esi /* offset += X */
45 js bpf_error
46
47sk_load_half:
48 .globl sk_load_half
49
50 mov %r9d,%eax
51 sub %esi,%eax # hlen - offset
52 cmp $1,%eax
53 jle bpf_slow_path_half
54 movzwl (SKBDATA,%rsi),%eax
55 rol $8,%ax # ntohs()
56 ret
57
58sk_load_byte_ind:
59 .globl sk_load_byte_ind
60 add %ebx,%esi /* offset += X */
61 js bpf_error
62
63sk_load_byte:
64 .globl sk_load_byte
65
66 cmp %esi,%r9d /* if (offset >= hlen) goto bpf_slow_path_byte */
67 jle bpf_slow_path_byte
68 movzbl (SKBDATA,%rsi),%eax
69 ret
70
71/**
72 * sk_load_byte_msh - BPF_S_LDX_B_MSH helper
73 *
74 * Implements BPF_S_LDX_B_MSH : ldxb 4*([offset]&0xf)
75 * Must preserve A accumulator (%eax)
76 * Inputs : %esi is the offset value, already known positive
77 */
78ENTRY(sk_load_byte_msh)
79 CFI_STARTPROC
80 cmp %esi,%r9d /* if (offset >= hlen) goto bpf_slow_path_byte_msh */
81 jle bpf_slow_path_byte_msh
82 movzbl (SKBDATA,%rsi),%ebx
83 and $15,%bl
84 shl $2,%bl
85 ret
86 CFI_ENDPROC
87ENDPROC(sk_load_byte_msh)
88
89bpf_error:
90# force a return 0 from jit handler
91 xor %eax,%eax
92 mov -8(%rbp),%rbx
93 leaveq
94 ret
95
96/* rsi contains offset and can be scratched */
97#define bpf_slow_path_common(LEN) \
98 push %rdi; /* save skb */ \
99 push %r9; \
100 push SKBDATA; \
101/* rsi already has offset */ \
102 mov $LEN,%ecx; /* len */ \
103 lea -12(%rbp),%rdx; \
104 call skb_copy_bits; \
105 test %eax,%eax; \
106 pop SKBDATA; \
107 pop %r9; \
108 pop %rdi
109
110
111bpf_slow_path_word:
112 bpf_slow_path_common(4)
113 js bpf_error
114 mov -12(%rbp),%eax
115 bswap %eax
116 ret
117
118bpf_slow_path_half:
119 bpf_slow_path_common(2)
120 js bpf_error
121 mov -12(%rbp),%ax
122 rol $8,%ax
123 movzwl %ax,%eax
124 ret
125
126bpf_slow_path_byte:
127 bpf_slow_path_common(1)
128 js bpf_error
129 movzbl -12(%rbp),%eax
130 ret
131
132bpf_slow_path_byte_msh:
133 xchg %eax,%ebx /* dont lose A , X is about to be scratched */
134 bpf_slow_path_common(1)
135 js bpf_error
136 movzbl -12(%rbp),%eax
137 and $15,%al
138 shl $2,%al
139 xchg %eax,%ebx
140 ret
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
new file mode 100644
index 000000000000..bfab3fa10edc
--- /dev/null
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -0,0 +1,654 @@
1/* bpf_jit_comp.c : BPF JIT compiler
2 *
3 * Copyright (C) 2011 Eric Dumazet (eric.dumazet@gmail.com)
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; version 2
8 * of the License.
9 */
10#include <linux/moduleloader.h>
11#include <asm/cacheflush.h>
12#include <linux/netdevice.h>
13#include <linux/filter.h>
14
15/*
16 * Conventions :
17 * EAX : BPF A accumulator
18 * EBX : BPF X accumulator
19 * RDI : pointer to skb (first argument given to JIT function)
20 * RBP : frame pointer (even if CONFIG_FRAME_POINTER=n)
21 * ECX,EDX,ESI : scratch registers
22 * r9d : skb->len - skb->data_len (headlen)
23 * r8 : skb->data
24 * -8(RBP) : saved RBX value
25 * -16(RBP)..-80(RBP) : BPF_MEMWORDS values
26 */
27int bpf_jit_enable __read_mostly;
28
29/*
30 * assembly code in arch/x86/net/bpf_jit.S
31 */
32extern u8 sk_load_word[], sk_load_half[], sk_load_byte[], sk_load_byte_msh[];
33extern u8 sk_load_word_ind[], sk_load_half_ind[], sk_load_byte_ind[];
34
35static inline u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
36{
37 if (len == 1)
38 *ptr = bytes;
39 else if (len == 2)
40 *(u16 *)ptr = bytes;
41 else {
42 *(u32 *)ptr = bytes;
43 barrier();
44 }
45 return ptr + len;
46}
47
48#define EMIT(bytes, len) do { prog = emit_code(prog, bytes, len); } while (0)
49
50#define EMIT1(b1) EMIT(b1, 1)
51#define EMIT2(b1, b2) EMIT((b1) + ((b2) << 8), 2)
52#define EMIT3(b1, b2, b3) EMIT((b1) + ((b2) << 8) + ((b3) << 16), 3)
53#define EMIT4(b1, b2, b3, b4) EMIT((b1) + ((b2) << 8) + ((b3) << 16) + ((b4) << 24), 4)
54#define EMIT1_off32(b1, off) do { EMIT1(b1); EMIT(off, 4);} while (0)
55
56#define CLEAR_A() EMIT2(0x31, 0xc0) /* xor %eax,%eax */
57#define CLEAR_X() EMIT2(0x31, 0xdb) /* xor %ebx,%ebx */
58
59static inline bool is_imm8(int value)
60{
61 return value <= 127 && value >= -128;
62}
63
64static inline bool is_near(int offset)
65{
66 return offset <= 127 && offset >= -128;
67}
68
69#define EMIT_JMP(offset) \
70do { \
71 if (offset) { \
72 if (is_near(offset)) \
73 EMIT2(0xeb, offset); /* jmp .+off8 */ \
74 else \
75 EMIT1_off32(0xe9, offset); /* jmp .+off32 */ \
76 } \
77} while (0)
78
79/* list of x86 cond jumps opcodes (. + s8)
80 * Add 0x10 (and an extra 0x0f) to generate far jumps (. + s32)
81 */
82#define X86_JB 0x72
83#define X86_JAE 0x73
84#define X86_JE 0x74
85#define X86_JNE 0x75
86#define X86_JBE 0x76
87#define X86_JA 0x77
88
89#define EMIT_COND_JMP(op, offset) \
90do { \
91 if (is_near(offset)) \
92 EMIT2(op, offset); /* jxx .+off8 */ \
93 else { \
94 EMIT2(0x0f, op + 0x10); \
95 EMIT(offset, 4); /* jxx .+off32 */ \
96 } \
97} while (0)
98
99#define COND_SEL(CODE, TOP, FOP) \
100 case CODE: \
101 t_op = TOP; \
102 f_op = FOP; \
103 goto cond_branch
104
105
106#define SEEN_DATAREF 1 /* might call external helpers */
107#define SEEN_XREG 2 /* ebx is used */
108#define SEEN_MEM 4 /* use mem[] for temporary storage */
109
110static inline void bpf_flush_icache(void *start, void *end)
111{
112 mm_segment_t old_fs = get_fs();
113
114 set_fs(KERNEL_DS);
115 smp_wmb();
116 flush_icache_range((unsigned long)start, (unsigned long)end);
117 set_fs(old_fs);
118}
119
120
121void bpf_jit_compile(struct sk_filter *fp)
122{
123 u8 temp[64];
124 u8 *prog;
125 unsigned int proglen, oldproglen = 0;
126 int ilen, i;
127 int t_offset, f_offset;
128 u8 t_op, f_op, seen = 0, pass;
129 u8 *image = NULL;
130 u8 *func;
131 int pc_ret0 = -1; /* bpf index of first RET #0 instruction (if any) */
132 unsigned int cleanup_addr; /* epilogue code offset */
133 unsigned int *addrs;
134 const struct sock_filter *filter = fp->insns;
135 int flen = fp->len;
136
137 if (!bpf_jit_enable)
138 return;
139
140 addrs = kmalloc(flen * sizeof(*addrs), GFP_KERNEL);
141 if (addrs == NULL)
142 return;
143
144 /* Before first pass, make a rough estimation of addrs[]
145 * each bpf instruction is translated to less than 64 bytes
146 */
147 for (proglen = 0, i = 0; i < flen; i++) {
148 proglen += 64;
149 addrs[i] = proglen;
150 }
151 cleanup_addr = proglen; /* epilogue address */
152
153 for (pass = 0; pass < 10; pass++) {
154 /* no prologue/epilogue for trivial filters (RET something) */
155 proglen = 0;
156 prog = temp;
157
158 if (seen) {
159 EMIT4(0x55, 0x48, 0x89, 0xe5); /* push %rbp; mov %rsp,%rbp */
160 EMIT4(0x48, 0x83, 0xec, 96); /* subq $96,%rsp */
161 /* note : must save %rbx in case bpf_error is hit */
162 if (seen & (SEEN_XREG | SEEN_DATAREF))
163 EMIT4(0x48, 0x89, 0x5d, 0xf8); /* mov %rbx, -8(%rbp) */
164 if (seen & SEEN_XREG)
165 CLEAR_X(); /* make sure we dont leek kernel memory */
166
167 /*
168 * If this filter needs to access skb data,
169 * loads r9 and r8 with :
170 * r9 = skb->len - skb->data_len
171 * r8 = skb->data
172 */
173 if (seen & SEEN_DATAREF) {
174 if (offsetof(struct sk_buff, len) <= 127)
175 /* mov off8(%rdi),%r9d */
176 EMIT4(0x44, 0x8b, 0x4f, offsetof(struct sk_buff, len));
177 else {
178 /* mov off32(%rdi),%r9d */
179 EMIT3(0x44, 0x8b, 0x8f);
180 EMIT(offsetof(struct sk_buff, len), 4);
181 }
182 if (is_imm8(offsetof(struct sk_buff, data_len)))
183 /* sub off8(%rdi),%r9d */
184 EMIT4(0x44, 0x2b, 0x4f, offsetof(struct sk_buff, data_len));
185 else {
186 EMIT3(0x44, 0x2b, 0x8f);
187 EMIT(offsetof(struct sk_buff, data_len), 4);
188 }
189
190 if (is_imm8(offsetof(struct sk_buff, data)))
191 /* mov off8(%rdi),%r8 */
192 EMIT4(0x4c, 0x8b, 0x47, offsetof(struct sk_buff, data));
193 else {
194 /* mov off32(%rdi),%r8 */
195 EMIT3(0x4c, 0x8b, 0x87);
196 EMIT(offsetof(struct sk_buff, data), 4);
197 }
198 }
199 }
200
201 switch (filter[0].code) {
202 case BPF_S_RET_K:
203 case BPF_S_LD_W_LEN:
204 case BPF_S_ANC_PROTOCOL:
205 case BPF_S_ANC_IFINDEX:
206 case BPF_S_ANC_MARK:
207 case BPF_S_ANC_RXHASH:
208 case BPF_S_ANC_CPU:
209 case BPF_S_ANC_QUEUE:
210 case BPF_S_LD_W_ABS:
211 case BPF_S_LD_H_ABS:
212 case BPF_S_LD_B_ABS:
213 /* first instruction sets A register (or is RET 'constant') */
214 break;
215 default:
216 /* make sure we dont leak kernel information to user */
217 CLEAR_A(); /* A = 0 */
218 }
219
220 for (i = 0; i < flen; i++) {
221 unsigned int K = filter[i].k;
222
223 switch (filter[i].code) {
224 case BPF_S_ALU_ADD_X: /* A += X; */
225 seen |= SEEN_XREG;
226 EMIT2(0x01, 0xd8); /* add %ebx,%eax */
227 break;
228 case BPF_S_ALU_ADD_K: /* A += K; */
229 if (!K)
230 break;
231 if (is_imm8(K))
232 EMIT3(0x83, 0xc0, K); /* add imm8,%eax */
233 else
234 EMIT1_off32(0x05, K); /* add imm32,%eax */
235 break;
236 case BPF_S_ALU_SUB_X: /* A -= X; */
237 seen |= SEEN_XREG;
238 EMIT2(0x29, 0xd8); /* sub %ebx,%eax */
239 break;
240 case BPF_S_ALU_SUB_K: /* A -= K */
241 if (!K)
242 break;
243 if (is_imm8(K))
244 EMIT3(0x83, 0xe8, K); /* sub imm8,%eax */
245 else
246 EMIT1_off32(0x2d, K); /* sub imm32,%eax */
247 break;
248 case BPF_S_ALU_MUL_X: /* A *= X; */
249 seen |= SEEN_XREG;
250 EMIT3(0x0f, 0xaf, 0xc3); /* imul %ebx,%eax */
251 break;
252 case BPF_S_ALU_MUL_K: /* A *= K */
253 if (is_imm8(K))
254 EMIT3(0x6b, 0xc0, K); /* imul imm8,%eax,%eax */
255 else {
256 EMIT2(0x69, 0xc0); /* imul imm32,%eax */
257 EMIT(K, 4);
258 }
259 break;
260 case BPF_S_ALU_DIV_X: /* A /= X; */
261 seen |= SEEN_XREG;
262 EMIT2(0x85, 0xdb); /* test %ebx,%ebx */
263 if (pc_ret0 != -1)
264 EMIT_COND_JMP(X86_JE, addrs[pc_ret0] - (addrs[i] - 4));
265 else {
266 EMIT_COND_JMP(X86_JNE, 2 + 5);
267 CLEAR_A();
268 EMIT1_off32(0xe9, cleanup_addr - (addrs[i] - 4)); /* jmp .+off32 */
269 }
270 EMIT4(0x31, 0xd2, 0xf7, 0xf3); /* xor %edx,%edx; div %ebx */
271 break;
272 case BPF_S_ALU_DIV_K: /* A = reciprocal_divide(A, K); */
273 EMIT3(0x48, 0x69, 0xc0); /* imul imm32,%rax,%rax */
274 EMIT(K, 4);
275 EMIT4(0x48, 0xc1, 0xe8, 0x20); /* shr $0x20,%rax */
276 break;
277 case BPF_S_ALU_AND_X:
278 seen |= SEEN_XREG;
279 EMIT2(0x21, 0xd8); /* and %ebx,%eax */
280 break;
281 case BPF_S_ALU_AND_K:
282 if (K >= 0xFFFFFF00) {
283 EMIT2(0x24, K & 0xFF); /* and imm8,%al */
284 } else if (K >= 0xFFFF0000) {
285 EMIT2(0x66, 0x25); /* and imm16,%ax */
286 EMIT2(K, 2);
287 } else {
288 EMIT1_off32(0x25, K); /* and imm32,%eax */
289 }
290 break;
291 case BPF_S_ALU_OR_X:
292 seen |= SEEN_XREG;
293 EMIT2(0x09, 0xd8); /* or %ebx,%eax */
294 break;
295 case BPF_S_ALU_OR_K:
296 if (is_imm8(K))
297 EMIT3(0x83, 0xc8, K); /* or imm8,%eax */
298 else
299 EMIT1_off32(0x0d, K); /* or imm32,%eax */
300 break;
301 case BPF_S_ALU_LSH_X: /* A <<= X; */
302 seen |= SEEN_XREG;
303 EMIT4(0x89, 0xd9, 0xd3, 0xe0); /* mov %ebx,%ecx; shl %cl,%eax */
304 break;
305 case BPF_S_ALU_LSH_K:
306 if (K == 0)
307 break;
308 else if (K == 1)
309 EMIT2(0xd1, 0xe0); /* shl %eax */
310 else
311 EMIT3(0xc1, 0xe0, K);
312 break;
313 case BPF_S_ALU_RSH_X: /* A >>= X; */
314 seen |= SEEN_XREG;
315 EMIT4(0x89, 0xd9, 0xd3, 0xe8); /* mov %ebx,%ecx; shr %cl,%eax */
316 break;
317 case BPF_S_ALU_RSH_K: /* A >>= K; */
318 if (K == 0)
319 break;
320 else if (K == 1)
321 EMIT2(0xd1, 0xe8); /* shr %eax */
322 else
323 EMIT3(0xc1, 0xe8, K);
324 break;
325 case BPF_S_ALU_NEG:
326 EMIT2(0xf7, 0xd8); /* neg %eax */
327 break;
328 case BPF_S_RET_K:
329 if (!K) {
330 if (pc_ret0 == -1)
331 pc_ret0 = i;
332 CLEAR_A();
333 } else {
334 EMIT1_off32(0xb8, K); /* mov $imm32,%eax */
335 }
336 /* fallinto */
337 case BPF_S_RET_A:
338 if (seen) {
339 if (i != flen - 1) {
340 EMIT_JMP(cleanup_addr - addrs[i]);
341 break;
342 }
343 if (seen & SEEN_XREG)
344 EMIT4(0x48, 0x8b, 0x5d, 0xf8); /* mov -8(%rbp),%rbx */
345 EMIT1(0xc9); /* leaveq */
346 }
347 EMIT1(0xc3); /* ret */
348 break;
349 case BPF_S_MISC_TAX: /* X = A */
350 seen |= SEEN_XREG;
351 EMIT2(0x89, 0xc3); /* mov %eax,%ebx */
352 break;
353 case BPF_S_MISC_TXA: /* A = X */
354 seen |= SEEN_XREG;
355 EMIT2(0x89, 0xd8); /* mov %ebx,%eax */
356 break;
357 case BPF_S_LD_IMM: /* A = K */
358 if (!K)
359 CLEAR_A();
360 else
361 EMIT1_off32(0xb8, K); /* mov $imm32,%eax */
362 break;
363 case BPF_S_LDX_IMM: /* X = K */
364 seen |= SEEN_XREG;
365 if (!K)
366 CLEAR_X();
367 else
368 EMIT1_off32(0xbb, K); /* mov $imm32,%ebx */
369 break;
370 case BPF_S_LD_MEM: /* A = mem[K] : mov off8(%rbp),%eax */
371 seen |= SEEN_MEM;
372 EMIT3(0x8b, 0x45, 0xf0 - K*4);
373 break;
374 case BPF_S_LDX_MEM: /* X = mem[K] : mov off8(%rbp),%ebx */
375 seen |= SEEN_XREG | SEEN_MEM;
376 EMIT3(0x8b, 0x5d, 0xf0 - K*4);
377 break;
378 case BPF_S_ST: /* mem[K] = A : mov %eax,off8(%rbp) */
379 seen |= SEEN_MEM;
380 EMIT3(0x89, 0x45, 0xf0 - K*4);
381 break;
382 case BPF_S_STX: /* mem[K] = X : mov %ebx,off8(%rbp) */
383 seen |= SEEN_XREG | SEEN_MEM;
384 EMIT3(0x89, 0x5d, 0xf0 - K*4);
385 break;
386 case BPF_S_LD_W_LEN: /* A = skb->len; */
387 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
388 if (is_imm8(offsetof(struct sk_buff, len)))
389 /* mov off8(%rdi),%eax */
390 EMIT3(0x8b, 0x47, offsetof(struct sk_buff, len));
391 else {
392 EMIT2(0x8b, 0x87);
393 EMIT(offsetof(struct sk_buff, len), 4);
394 }
395 break;
396 case BPF_S_LDX_W_LEN: /* X = skb->len; */
397 seen |= SEEN_XREG;
398 if (is_imm8(offsetof(struct sk_buff, len)))
399 /* mov off8(%rdi),%ebx */
400 EMIT3(0x8b, 0x5f, offsetof(struct sk_buff, len));
401 else {
402 EMIT2(0x8b, 0x9f);
403 EMIT(offsetof(struct sk_buff, len), 4);
404 }
405 break;
406 case BPF_S_ANC_PROTOCOL: /* A = ntohs(skb->protocol); */
407 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
408 if (is_imm8(offsetof(struct sk_buff, protocol))) {
409 /* movzwl off8(%rdi),%eax */
410 EMIT4(0x0f, 0xb7, 0x47, offsetof(struct sk_buff, protocol));
411 } else {
412 EMIT3(0x0f, 0xb7, 0x87); /* movzwl off32(%rdi),%eax */
413 EMIT(offsetof(struct sk_buff, protocol), 4);
414 }
415 EMIT2(0x86, 0xc4); /* ntohs() : xchg %al,%ah */
416 break;
417 case BPF_S_ANC_IFINDEX:
418 if (is_imm8(offsetof(struct sk_buff, dev))) {
419 /* movq off8(%rdi),%rax */
420 EMIT4(0x48, 0x8b, 0x47, offsetof(struct sk_buff, dev));
421 } else {
422 EMIT3(0x48, 0x8b, 0x87); /* movq off32(%rdi),%rax */
423 EMIT(offsetof(struct sk_buff, dev), 4);
424 }
425 EMIT3(0x48, 0x85, 0xc0); /* test %rax,%rax */
426 EMIT_COND_JMP(X86_JE, cleanup_addr - (addrs[i] - 6));
427 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
428 EMIT2(0x8b, 0x80); /* mov off32(%rax),%eax */
429 EMIT(offsetof(struct net_device, ifindex), 4);
430 break;
431 case BPF_S_ANC_MARK:
432 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
433 if (is_imm8(offsetof(struct sk_buff, mark))) {
434 /* mov off8(%rdi),%eax */
435 EMIT3(0x8b, 0x47, offsetof(struct sk_buff, mark));
436 } else {
437 EMIT2(0x8b, 0x87);
438 EMIT(offsetof(struct sk_buff, mark), 4);
439 }
440 break;
441 case BPF_S_ANC_RXHASH:
442 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, rxhash) != 4);
443 if (is_imm8(offsetof(struct sk_buff, rxhash))) {
444 /* mov off8(%rdi),%eax */
445 EMIT3(0x8b, 0x47, offsetof(struct sk_buff, rxhash));
446 } else {
447 EMIT2(0x8b, 0x87);
448 EMIT(offsetof(struct sk_buff, rxhash), 4);
449 }
450 break;
451 case BPF_S_ANC_QUEUE:
452 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
453 if (is_imm8(offsetof(struct sk_buff, queue_mapping))) {
454 /* movzwl off8(%rdi),%eax */
455 EMIT4(0x0f, 0xb7, 0x47, offsetof(struct sk_buff, queue_mapping));
456 } else {
457 EMIT3(0x0f, 0xb7, 0x87); /* movzwl off32(%rdi),%eax */
458 EMIT(offsetof(struct sk_buff, queue_mapping), 4);
459 }
460 break;
461 case BPF_S_ANC_CPU:
462#ifdef CONFIG_SMP
463 EMIT4(0x65, 0x8b, 0x04, 0x25); /* mov %gs:off32,%eax */
464 EMIT((u32)(unsigned long)&cpu_number, 4); /* A = smp_processor_id(); */
465#else
466 CLEAR_A();
467#endif
468 break;
469 case BPF_S_LD_W_ABS:
470 func = sk_load_word;
471common_load: seen |= SEEN_DATAREF;
472 if ((int)K < 0)
473 goto out;
474 t_offset = func - (image + addrs[i]);
475 EMIT1_off32(0xbe, K); /* mov imm32,%esi */
476 EMIT1_off32(0xe8, t_offset); /* call */
477 break;
478 case BPF_S_LD_H_ABS:
479 func = sk_load_half;
480 goto common_load;
481 case BPF_S_LD_B_ABS:
482 func = sk_load_byte;
483 goto common_load;
484 case BPF_S_LDX_B_MSH:
485 if ((int)K < 0) {
486 if (pc_ret0 != -1) {
487 EMIT_JMP(addrs[pc_ret0] - addrs[i]);
488 break;
489 }
490 CLEAR_A();
491 EMIT_JMP(cleanup_addr - addrs[i]);
492 break;
493 }
494 seen |= SEEN_DATAREF | SEEN_XREG;
495 t_offset = sk_load_byte_msh - (image + addrs[i]);
496 EMIT1_off32(0xbe, K); /* mov imm32,%esi */
497 EMIT1_off32(0xe8, t_offset); /* call sk_load_byte_msh */
498 break;
499 case BPF_S_LD_W_IND:
500 func = sk_load_word_ind;
501common_load_ind: seen |= SEEN_DATAREF | SEEN_XREG;
502 t_offset = func - (image + addrs[i]);
503 EMIT1_off32(0xbe, K); /* mov imm32,%esi */
504 EMIT1_off32(0xe8, t_offset); /* call sk_load_xxx_ind */
505 break;
506 case BPF_S_LD_H_IND:
507 func = sk_load_half_ind;
508 goto common_load_ind;
509 case BPF_S_LD_B_IND:
510 func = sk_load_byte_ind;
511 goto common_load_ind;
512 case BPF_S_JMP_JA:
513 t_offset = addrs[i + K] - addrs[i];
514 EMIT_JMP(t_offset);
515 break;
516 COND_SEL(BPF_S_JMP_JGT_K, X86_JA, X86_JBE);
517 COND_SEL(BPF_S_JMP_JGE_K, X86_JAE, X86_JB);
518 COND_SEL(BPF_S_JMP_JEQ_K, X86_JE, X86_JNE);
519 COND_SEL(BPF_S_JMP_JSET_K,X86_JNE, X86_JE);
520 COND_SEL(BPF_S_JMP_JGT_X, X86_JA, X86_JBE);
521 COND_SEL(BPF_S_JMP_JGE_X, X86_JAE, X86_JB);
522 COND_SEL(BPF_S_JMP_JEQ_X, X86_JE, X86_JNE);
523 COND_SEL(BPF_S_JMP_JSET_X,X86_JNE, X86_JE);
524
525cond_branch: f_offset = addrs[i + filter[i].jf] - addrs[i];
526 t_offset = addrs[i + filter[i].jt] - addrs[i];
527
528 /* same targets, can avoid doing the test :) */
529 if (filter[i].jt == filter[i].jf) {
530 EMIT_JMP(t_offset);
531 break;
532 }
533
534 switch (filter[i].code) {
535 case BPF_S_JMP_JGT_X:
536 case BPF_S_JMP_JGE_X:
537 case BPF_S_JMP_JEQ_X:
538 seen |= SEEN_XREG;
539 EMIT2(0x39, 0xd8); /* cmp %ebx,%eax */
540 break;
541 case BPF_S_JMP_JSET_X:
542 seen |= SEEN_XREG;
543 EMIT2(0x85, 0xd8); /* test %ebx,%eax */
544 break;
545 case BPF_S_JMP_JEQ_K:
546 if (K == 0) {
547 EMIT2(0x85, 0xc0); /* test %eax,%eax */
548 break;
549 }
550 case BPF_S_JMP_JGT_K:
551 case BPF_S_JMP_JGE_K:
552 if (K <= 127)
553 EMIT3(0x83, 0xf8, K); /* cmp imm8,%eax */
554 else
555 EMIT1_off32(0x3d, K); /* cmp imm32,%eax */
556 break;
557 case BPF_S_JMP_JSET_K:
558 if (K <= 0xFF)
559 EMIT2(0xa8, K); /* test imm8,%al */
560 else if (!(K & 0xFFFF00FF))
561 EMIT3(0xf6, 0xc4, K >> 8); /* test imm8,%ah */
562 else if (K <= 0xFFFF) {
563 EMIT2(0x66, 0xa9); /* test imm16,%ax */
564 EMIT(K, 2);
565 } else {
566 EMIT1_off32(0xa9, K); /* test imm32,%eax */
567 }
568 break;
569 }
570 if (filter[i].jt != 0) {
571 if (filter[i].jf)
572 t_offset += is_near(f_offset) ? 2 : 6;
573 EMIT_COND_JMP(t_op, t_offset);
574 if (filter[i].jf)
575 EMIT_JMP(f_offset);
576 break;
577 }
578 EMIT_COND_JMP(f_op, f_offset);
579 break;
580 default:
581 /* hmm, too complex filter, give up with jit compiler */
582 goto out;
583 }
584 ilen = prog - temp;
585 if (image) {
586 if (unlikely(proglen + ilen > oldproglen)) {
587 pr_err("bpb_jit_compile fatal error\n");
588 kfree(addrs);
589 module_free(NULL, image);
590 return;
591 }
592 memcpy(image + proglen, temp, ilen);
593 }
594 proglen += ilen;
595 addrs[i] = proglen;
596 prog = temp;
597 }
598 /* last bpf instruction is always a RET :
599 * use it to give the cleanup instruction(s) addr
600 */
601 cleanup_addr = proglen - 1; /* ret */
602 if (seen)
603 cleanup_addr -= 1; /* leaveq */
604 if (seen & SEEN_XREG)
605 cleanup_addr -= 4; /* mov -8(%rbp),%rbx */
606
607 if (image) {
608 WARN_ON(proglen != oldproglen);
609 break;
610 }
611 if (proglen == oldproglen) {
612 image = module_alloc(max_t(unsigned int,
613 proglen,
614 sizeof(struct work_struct)));
615 if (!image)
616 goto out;
617 }
618 oldproglen = proglen;
619 }
620 if (bpf_jit_enable > 1)
621 pr_err("flen=%d proglen=%u pass=%d image=%p\n",
622 flen, proglen, pass, image);
623
624 if (image) {
625 if (bpf_jit_enable > 1)
626 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_ADDRESS,
627 16, 1, image, proglen, false);
628
629 bpf_flush_icache(image, image + proglen);
630
631 fp->bpf_func = (void *)image;
632 }
633out:
634 kfree(addrs);
635 return;
636}
637
638static void jit_free_defer(struct work_struct *arg)
639{
640 module_free(NULL, arg);
641}
642
643/* run from softirq, we must use a work_struct to call
644 * module_free() from process context
645 */
646void bpf_jit_free(struct sk_filter *fp)
647{
648 if (fp->bpf_func != sk_run_filter) {
649 struct work_struct *work = (struct work_struct *)fp->bpf_func;
650
651 INIT_WORK(work, jit_free_defer);
652 schedule_work(work);
653 }
654}
diff --git a/include/linux/filter.h b/include/linux/filter.h
index 45266b75409a..4609b85e559d 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -135,6 +135,8 @@ struct sk_filter
135{ 135{
136 atomic_t refcnt; 136 atomic_t refcnt;
137 unsigned int len; /* Number of filter blocks */ 137 unsigned int len; /* Number of filter blocks */
138 unsigned int (*bpf_func)(const struct sk_buff *skb,
139 const struct sock_filter *filter);
138 struct rcu_head rcu; 140 struct rcu_head rcu;
139 struct sock_filter insns[0]; 141 struct sock_filter insns[0];
140}; 142};
@@ -153,6 +155,80 @@ extern unsigned int sk_run_filter(const struct sk_buff *skb,
153extern int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 155extern int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
154extern int sk_detach_filter(struct sock *sk); 156extern int sk_detach_filter(struct sock *sk);
155extern int sk_chk_filter(struct sock_filter *filter, int flen); 157extern int sk_chk_filter(struct sock_filter *filter, int flen);
158
159#ifdef CONFIG_BPF_JIT
160extern void bpf_jit_compile(struct sk_filter *fp);
161extern void bpf_jit_free(struct sk_filter *fp);
162#define SK_RUN_FILTER(FILTER, SKB) (*FILTER->bpf_func)(SKB, FILTER->insns)
163#else
164static inline void bpf_jit_compile(struct sk_filter *fp)
165{
166}
167static inline void bpf_jit_free(struct sk_filter *fp)
168{
169}
170#define SK_RUN_FILTER(FILTER, SKB) sk_run_filter(SKB, FILTER->insns)
171#endif
172
173enum {
174 BPF_S_RET_K = 1,
175 BPF_S_RET_A,
176 BPF_S_ALU_ADD_K,
177 BPF_S_ALU_ADD_X,
178 BPF_S_ALU_SUB_K,
179 BPF_S_ALU_SUB_X,
180 BPF_S_ALU_MUL_K,
181 BPF_S_ALU_MUL_X,
182 BPF_S_ALU_DIV_X,
183 BPF_S_ALU_AND_K,
184 BPF_S_ALU_AND_X,
185 BPF_S_ALU_OR_K,
186 BPF_S_ALU_OR_X,
187 BPF_S_ALU_LSH_K,
188 BPF_S_ALU_LSH_X,
189 BPF_S_ALU_RSH_K,
190 BPF_S_ALU_RSH_X,
191 BPF_S_ALU_NEG,
192 BPF_S_LD_W_ABS,
193 BPF_S_LD_H_ABS,
194 BPF_S_LD_B_ABS,
195 BPF_S_LD_W_LEN,
196 BPF_S_LD_W_IND,
197 BPF_S_LD_H_IND,
198 BPF_S_LD_B_IND,
199 BPF_S_LD_IMM,
200 BPF_S_LDX_W_LEN,
201 BPF_S_LDX_B_MSH,
202 BPF_S_LDX_IMM,
203 BPF_S_MISC_TAX,
204 BPF_S_MISC_TXA,
205 BPF_S_ALU_DIV_K,
206 BPF_S_LD_MEM,
207 BPF_S_LDX_MEM,
208 BPF_S_ST,
209 BPF_S_STX,
210 BPF_S_JMP_JA,
211 BPF_S_JMP_JEQ_K,
212 BPF_S_JMP_JEQ_X,
213 BPF_S_JMP_JGE_K,
214 BPF_S_JMP_JGE_X,
215 BPF_S_JMP_JGT_K,
216 BPF_S_JMP_JGT_X,
217 BPF_S_JMP_JSET_K,
218 BPF_S_JMP_JSET_X,
219 /* Ancillary data */
220 BPF_S_ANC_PROTOCOL,
221 BPF_S_ANC_PKTTYPE,
222 BPF_S_ANC_IFINDEX,
223 BPF_S_ANC_NLATTR,
224 BPF_S_ANC_NLATTR_NEST,
225 BPF_S_ANC_MARK,
226 BPF_S_ANC_QUEUE,
227 BPF_S_ANC_HATYPE,
228 BPF_S_ANC_RXHASH,
229 BPF_S_ANC_CPU,
230};
231
156#endif /* __KERNEL__ */ 232#endif /* __KERNEL__ */
157 233
158#endif /* __LINUX_FILTER_H__ */ 234#endif /* __LINUX_FILTER_H__ */
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index cb8178ab3c52..364bcf212f71 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2514,6 +2514,7 @@ extern struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
2514extern int netdev_max_backlog; 2514extern int netdev_max_backlog;
2515extern int netdev_tstamp_prequeue; 2515extern int netdev_tstamp_prequeue;
2516extern int weight_p; 2516extern int weight_p;
2517extern int bpf_jit_enable;
2517extern int netdev_set_master(struct net_device *dev, struct net_device *master); 2518extern int netdev_set_master(struct net_device *dev, struct net_device *master);
2518extern int netdev_set_bond_master(struct net_device *dev, 2519extern int netdev_set_bond_master(struct net_device *dev,
2519 struct net_device *master); 2520 struct net_device *master);
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index d0ae90af0b40..79aafbbf430a 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -391,8 +391,8 @@ struct sk_buff {
391 391
392 __u32 rxhash; 392 __u32 rxhash;
393 393
394 __u16 queue_mapping;
394 kmemcheck_bitfield_begin(flags2); 395 kmemcheck_bitfield_begin(flags2);
395 __u16 queue_mapping:16;
396#ifdef CONFIG_IPV6_NDISC_NODETYPE 396#ifdef CONFIG_IPV6_NDISC_NODETYPE
397 __u8 ndisc_nodetype:2; 397 __u8 ndisc_nodetype:2;
398#endif 398#endif
diff --git a/net/Kconfig b/net/Kconfig
index 79cabf1ee68b..745fb02d2fda 100644
--- a/net/Kconfig
+++ b/net/Kconfig
@@ -232,6 +232,19 @@ config XPS
232 depends on SMP && SYSFS && USE_GENERIC_SMP_HELPERS 232 depends on SMP && SYSFS && USE_GENERIC_SMP_HELPERS
233 default y 233 default y
234 234
235config HAVE_BPF_JIT
236 bool
237
238config BPF_JIT
239 bool "enable BPF Just In Time compiler"
240 depends on HAVE_BPF_JIT
241 ---help---
242 Berkeley Packet Filter filtering capabilities are normally handled
243 by an interpreter. This option allows kernel to generate a native
244 code when filter is loaded in memory. This should speedup
245 packet sniffing (libpcap/tcpdump). Note : Admin should enable
246 this feature changing /proc/sys/net/core/bpf_jit_enable
247
235menu "Network testing" 248menu "Network testing"
236 249
237config NET_PKTGEN 250config NET_PKTGEN
diff --git a/net/core/filter.c b/net/core/filter.c
index afb8afb066bb..0eb8c4466eaa 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -39,65 +39,6 @@
39#include <linux/filter.h> 39#include <linux/filter.h>
40#include <linux/reciprocal_div.h> 40#include <linux/reciprocal_div.h>
41 41
42enum {
43 BPF_S_RET_K = 1,
44 BPF_S_RET_A,
45 BPF_S_ALU_ADD_K,
46 BPF_S_ALU_ADD_X,
47 BPF_S_ALU_SUB_K,
48 BPF_S_ALU_SUB_X,
49 BPF_S_ALU_MUL_K,
50 BPF_S_ALU_MUL_X,
51 BPF_S_ALU_DIV_X,
52 BPF_S_ALU_AND_K,
53 BPF_S_ALU_AND_X,
54 BPF_S_ALU_OR_K,
55 BPF_S_ALU_OR_X,
56 BPF_S_ALU_LSH_K,
57 BPF_S_ALU_LSH_X,
58 BPF_S_ALU_RSH_K,
59 BPF_S_ALU_RSH_X,
60 BPF_S_ALU_NEG,
61 BPF_S_LD_W_ABS,
62 BPF_S_LD_H_ABS,
63 BPF_S_LD_B_ABS,
64 BPF_S_LD_W_LEN,
65 BPF_S_LD_W_IND,
66 BPF_S_LD_H_IND,
67 BPF_S_LD_B_IND,
68 BPF_S_LD_IMM,
69 BPF_S_LDX_W_LEN,
70 BPF_S_LDX_B_MSH,
71 BPF_S_LDX_IMM,
72 BPF_S_MISC_TAX,
73 BPF_S_MISC_TXA,
74 BPF_S_ALU_DIV_K,
75 BPF_S_LD_MEM,
76 BPF_S_LDX_MEM,
77 BPF_S_ST,
78 BPF_S_STX,
79 BPF_S_JMP_JA,
80 BPF_S_JMP_JEQ_K,
81 BPF_S_JMP_JEQ_X,
82 BPF_S_JMP_JGE_K,
83 BPF_S_JMP_JGE_X,
84 BPF_S_JMP_JGT_K,
85 BPF_S_JMP_JGT_X,
86 BPF_S_JMP_JSET_K,
87 BPF_S_JMP_JSET_X,
88 /* Ancillary data */
89 BPF_S_ANC_PROTOCOL,
90 BPF_S_ANC_PKTTYPE,
91 BPF_S_ANC_IFINDEX,
92 BPF_S_ANC_NLATTR,
93 BPF_S_ANC_NLATTR_NEST,
94 BPF_S_ANC_MARK,
95 BPF_S_ANC_QUEUE,
96 BPF_S_ANC_HATYPE,
97 BPF_S_ANC_RXHASH,
98 BPF_S_ANC_CPU,
99};
100
101/* No hurry in this branch */ 42/* No hurry in this branch */
102static void *__load_pointer(const struct sk_buff *skb, int k, unsigned int size) 43static void *__load_pointer(const struct sk_buff *skb, int k, unsigned int size)
103{ 44{
@@ -145,7 +86,7 @@ int sk_filter(struct sock *sk, struct sk_buff *skb)
145 rcu_read_lock(); 86 rcu_read_lock();
146 filter = rcu_dereference(sk->sk_filter); 87 filter = rcu_dereference(sk->sk_filter);
147 if (filter) { 88 if (filter) {
148 unsigned int pkt_len = sk_run_filter(skb, filter->insns); 89 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
149 90
150 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM; 91 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
151 } 92 }
@@ -638,6 +579,7 @@ void sk_filter_release_rcu(struct rcu_head *rcu)
638{ 579{
639 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 580 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
640 581
582 bpf_jit_free(fp);
641 kfree(fp); 583 kfree(fp);
642} 584}
643EXPORT_SYMBOL(sk_filter_release_rcu); 585EXPORT_SYMBOL(sk_filter_release_rcu);
@@ -672,6 +614,7 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
672 614
673 atomic_set(&fp->refcnt, 1); 615 atomic_set(&fp->refcnt, 1);
674 fp->len = fprog->len; 616 fp->len = fprog->len;
617 fp->bpf_func = sk_run_filter;
675 618
676 err = sk_chk_filter(fp->insns, fp->len); 619 err = sk_chk_filter(fp->insns, fp->len);
677 if (err) { 620 if (err) {
@@ -679,6 +622,8 @@ int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
679 return err; 622 return err;
680 } 623 }
681 624
625 bpf_jit_compile(fp);
626
682 old_fp = rcu_dereference_protected(sk->sk_filter, 627 old_fp = rcu_dereference_protected(sk->sk_filter,
683 sock_owned_by_user(sk)); 628 sock_owned_by_user(sk));
684 rcu_assign_pointer(sk->sk_filter, fp); 629 rcu_assign_pointer(sk->sk_filter, fp);
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 385b6095fdc4..a829e3f60aeb 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -122,6 +122,15 @@ static struct ctl_table net_core_table[] = {
122 .mode = 0644, 122 .mode = 0644,
123 .proc_handler = proc_dointvec 123 .proc_handler = proc_dointvec
124 }, 124 },
125#ifdef CONFIG_BPF_JIT
126 {
127 .procname = "bpf_jit_enable",
128 .data = &bpf_jit_enable,
129 .maxlen = sizeof(int),
130 .mode = 0644,
131 .proc_handler = proc_dointvec
132 },
133#endif
125 { 134 {
126 .procname = "netdev_tstamp_prequeue", 135 .procname = "netdev_tstamp_prequeue",
127 .data = &netdev_tstamp_prequeue, 136 .data = &netdev_tstamp_prequeue,
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index b5362e96022b..549527bca87a 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -538,7 +538,7 @@ static inline unsigned int run_filter(const struct sk_buff *skb,
538 rcu_read_lock(); 538 rcu_read_lock();
539 filter = rcu_dereference(sk->sk_filter); 539 filter = rcu_dereference(sk->sk_filter);
540 if (filter != NULL) 540 if (filter != NULL)
541 res = sk_run_filter(skb, filter->insns); 541 res = SK_RUN_FILTER(filter, skb);
542 rcu_read_unlock(); 542 rcu_read_unlock();
543 543
544 return res; 544 return res;
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-02 12:39:08 -0500