Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux Socket Filter - Kernel level socket filtering | |
3 | * | |
bd4cf0ed AS |
4 | * Based on the design of the Berkeley Packet Filter. The new |
5 | * internal format has been designed by PLUMgrid: | |
1da177e4 | 6 | * |
bd4cf0ed AS |
7 | * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com |
8 | * | |
9 | * Authors: | |
10 | * | |
11 | * Jay Schulist <jschlst@samba.org> | |
12 | * Alexei Starovoitov <ast@plumgrid.com> | |
13 | * Daniel Borkmann <dborkman@redhat.com> | |
1da177e4 LT |
14 | * |
15 | * This program is free software; you can redistribute it and/or | |
16 | * modify it under the terms of the GNU General Public License | |
17 | * as published by the Free Software Foundation; either version | |
18 | * 2 of the License, or (at your option) any later version. | |
19 | * | |
20 | * Andi Kleen - Fix a few bad bugs and races. | |
4df95ff4 | 21 | * Kris Katterjohn - Added many additional checks in bpf_check_classic() |
1da177e4 LT |
22 | */ |
23 | ||
24 | #include <linux/module.h> | |
25 | #include <linux/types.h> | |
1da177e4 LT |
26 | #include <linux/mm.h> |
27 | #include <linux/fcntl.h> | |
28 | #include <linux/socket.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/inet.h> | |
31 | #include <linux/netdevice.h> | |
32 | #include <linux/if_packet.h> | |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
1da177e4 LT |
34 | #include <net/ip.h> |
35 | #include <net/protocol.h> | |
4738c1db | 36 | #include <net/netlink.h> |
1da177e4 LT |
37 | #include <linux/skbuff.h> |
38 | #include <net/sock.h> | |
10b89ee4 | 39 | #include <net/flow_dissector.h> |
1da177e4 LT |
40 | #include <linux/errno.h> |
41 | #include <linux/timer.h> | |
1da177e4 | 42 | #include <asm/uaccess.h> |
40daafc8 | 43 | #include <asm/unaligned.h> |
1da177e4 | 44 | #include <linux/filter.h> |
86e4ca66 | 45 | #include <linux/ratelimit.h> |
46b325c7 | 46 | #include <linux/seccomp.h> |
f3335031 | 47 | #include <linux/if_vlan.h> |
89aa0758 | 48 | #include <linux/bpf.h> |
d691f9e8 | 49 | #include <net/sch_generic.h> |
8d20aabe | 50 | #include <net/cls_cgroup.h> |
d3aa45ce | 51 | #include <net/dst_metadata.h> |
c46646d0 | 52 | #include <net/dst.h> |
538950a1 | 53 | #include <net/sock_reuseport.h> |
1da177e4 | 54 | |
43db6d65 SH |
55 | /** |
56 | * sk_filter - run a packet through a socket filter | |
57 | * @sk: sock associated with &sk_buff | |
58 | * @skb: buffer to filter | |
43db6d65 | 59 | * |
ff936a04 AS |
60 | * Run the eBPF program and then cut skb->data to correct size returned by |
61 | * the program. If pkt_len is 0 we toss packet. If skb->len is smaller | |
43db6d65 | 62 | * than pkt_len we keep whole skb->data. This is the socket level |
ff936a04 | 63 | * wrapper to BPF_PROG_RUN. It returns 0 if the packet should |
43db6d65 SH |
64 | * be accepted or -EPERM if the packet should be tossed. |
65 | * | |
66 | */ | |
67 | int sk_filter(struct sock *sk, struct sk_buff *skb) | |
68 | { | |
69 | int err; | |
70 | struct sk_filter *filter; | |
71 | ||
c93bdd0e MG |
72 | /* |
73 | * If the skb was allocated from pfmemalloc reserves, only | |
74 | * allow SOCK_MEMALLOC sockets to use it as this socket is | |
75 | * helping free memory | |
76 | */ | |
77 | if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) | |
78 | return -ENOMEM; | |
79 | ||
43db6d65 SH |
80 | err = security_sock_rcv_skb(sk, skb); |
81 | if (err) | |
82 | return err; | |
83 | ||
80f8f102 ED |
84 | rcu_read_lock(); |
85 | filter = rcu_dereference(sk->sk_filter); | |
43db6d65 | 86 | if (filter) { |
ff936a04 | 87 | unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb); |
0d7da9dd | 88 | |
43db6d65 SH |
89 | err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM; |
90 | } | |
80f8f102 | 91 | rcu_read_unlock(); |
43db6d65 SH |
92 | |
93 | return err; | |
94 | } | |
95 | EXPORT_SYMBOL(sk_filter); | |
96 | ||
30743837 | 97 | static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 98 | { |
56193d1b | 99 | return skb_get_poff((struct sk_buff *)(unsigned long) ctx); |
bd4cf0ed AS |
100 | } |
101 | ||
30743837 | 102 | static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 103 | { |
eb9672f4 | 104 | struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx; |
bd4cf0ed AS |
105 | struct nlattr *nla; |
106 | ||
107 | if (skb_is_nonlinear(skb)) | |
108 | return 0; | |
109 | ||
05ab8f26 MK |
110 | if (skb->len < sizeof(struct nlattr)) |
111 | return 0; | |
112 | ||
30743837 | 113 | if (a > skb->len - sizeof(struct nlattr)) |
bd4cf0ed AS |
114 | return 0; |
115 | ||
30743837 | 116 | nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); |
bd4cf0ed AS |
117 | if (nla) |
118 | return (void *) nla - (void *) skb->data; | |
119 | ||
120 | return 0; | |
121 | } | |
122 | ||
30743837 | 123 | static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 124 | { |
eb9672f4 | 125 | struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx; |
bd4cf0ed AS |
126 | struct nlattr *nla; |
127 | ||
128 | if (skb_is_nonlinear(skb)) | |
129 | return 0; | |
130 | ||
05ab8f26 MK |
131 | if (skb->len < sizeof(struct nlattr)) |
132 | return 0; | |
133 | ||
30743837 | 134 | if (a > skb->len - sizeof(struct nlattr)) |
bd4cf0ed AS |
135 | return 0; |
136 | ||
30743837 DB |
137 | nla = (struct nlattr *) &skb->data[a]; |
138 | if (nla->nla_len > skb->len - a) | |
bd4cf0ed AS |
139 | return 0; |
140 | ||
30743837 | 141 | nla = nla_find_nested(nla, x); |
bd4cf0ed AS |
142 | if (nla) |
143 | return (void *) nla - (void *) skb->data; | |
144 | ||
145 | return 0; | |
146 | } | |
147 | ||
30743837 | 148 | static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed AS |
149 | { |
150 | return raw_smp_processor_id(); | |
151 | } | |
152 | ||
9bac3d6d AS |
153 | static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, |
154 | struct bpf_insn *insn_buf) | |
155 | { | |
156 | struct bpf_insn *insn = insn_buf; | |
157 | ||
158 | switch (skb_field) { | |
159 | case SKF_AD_MARK: | |
160 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); | |
161 | ||
162 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
163 | offsetof(struct sk_buff, mark)); | |
164 | break; | |
165 | ||
166 | case SKF_AD_PKTTYPE: | |
167 | *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); | |
168 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); | |
169 | #ifdef __BIG_ENDIAN_BITFIELD | |
170 | *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); | |
171 | #endif | |
172 | break; | |
173 | ||
174 | case SKF_AD_QUEUE: | |
175 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); | |
176 | ||
177 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
178 | offsetof(struct sk_buff, queue_mapping)); | |
179 | break; | |
c2497395 | 180 | |
c2497395 AS |
181 | case SKF_AD_VLAN_TAG: |
182 | case SKF_AD_VLAN_TAG_PRESENT: | |
183 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); | |
184 | BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); | |
185 | ||
186 | /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ | |
187 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
188 | offsetof(struct sk_buff, vlan_tci)); | |
189 | if (skb_field == SKF_AD_VLAN_TAG) { | |
190 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, | |
191 | ~VLAN_TAG_PRESENT); | |
192 | } else { | |
193 | /* dst_reg >>= 12 */ | |
194 | *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12); | |
195 | /* dst_reg &= 1 */ | |
196 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); | |
197 | } | |
198 | break; | |
9bac3d6d AS |
199 | } |
200 | ||
201 | return insn - insn_buf; | |
202 | } | |
203 | ||
bd4cf0ed | 204 | static bool convert_bpf_extensions(struct sock_filter *fp, |
2695fb55 | 205 | struct bpf_insn **insnp) |
bd4cf0ed | 206 | { |
2695fb55 | 207 | struct bpf_insn *insn = *insnp; |
9bac3d6d | 208 | u32 cnt; |
bd4cf0ed AS |
209 | |
210 | switch (fp->k) { | |
211 | case SKF_AD_OFF + SKF_AD_PROTOCOL: | |
0b8c707d DB |
212 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); |
213 | ||
214 | /* A = *(u16 *) (CTX + offsetof(protocol)) */ | |
215 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, | |
216 | offsetof(struct sk_buff, protocol)); | |
217 | /* A = ntohs(A) [emitting a nop or swap16] */ | |
218 | *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); | |
bd4cf0ed AS |
219 | break; |
220 | ||
221 | case SKF_AD_OFF + SKF_AD_PKTTYPE: | |
9bac3d6d AS |
222 | cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); |
223 | insn += cnt - 1; | |
bd4cf0ed AS |
224 | break; |
225 | ||
226 | case SKF_AD_OFF + SKF_AD_IFINDEX: | |
227 | case SKF_AD_OFF + SKF_AD_HATYPE: | |
bd4cf0ed AS |
228 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); |
229 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); | |
f8f6d679 DB |
230 | BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0); |
231 | ||
232 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)), | |
233 | BPF_REG_TMP, BPF_REG_CTX, | |
234 | offsetof(struct sk_buff, dev)); | |
235 | /* if (tmp != 0) goto pc + 1 */ | |
236 | *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); | |
237 | *insn++ = BPF_EXIT_INSN(); | |
238 | if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) | |
239 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, | |
240 | offsetof(struct net_device, ifindex)); | |
241 | else | |
242 | *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, | |
243 | offsetof(struct net_device, type)); | |
bd4cf0ed AS |
244 | break; |
245 | ||
246 | case SKF_AD_OFF + SKF_AD_MARK: | |
9bac3d6d AS |
247 | cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); |
248 | insn += cnt - 1; | |
bd4cf0ed AS |
249 | break; |
250 | ||
251 | case SKF_AD_OFF + SKF_AD_RXHASH: | |
252 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
253 | ||
9739eef1 AS |
254 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, |
255 | offsetof(struct sk_buff, hash)); | |
bd4cf0ed AS |
256 | break; |
257 | ||
258 | case SKF_AD_OFF + SKF_AD_QUEUE: | |
9bac3d6d AS |
259 | cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); |
260 | insn += cnt - 1; | |
bd4cf0ed AS |
261 | break; |
262 | ||
263 | case SKF_AD_OFF + SKF_AD_VLAN_TAG: | |
c2497395 AS |
264 | cnt = convert_skb_access(SKF_AD_VLAN_TAG, |
265 | BPF_REG_A, BPF_REG_CTX, insn); | |
266 | insn += cnt - 1; | |
267 | break; | |
bd4cf0ed | 268 | |
c2497395 AS |
269 | case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: |
270 | cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, | |
271 | BPF_REG_A, BPF_REG_CTX, insn); | |
272 | insn += cnt - 1; | |
bd4cf0ed AS |
273 | break; |
274 | ||
27cd5452 MS |
275 | case SKF_AD_OFF + SKF_AD_VLAN_TPID: |
276 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); | |
277 | ||
278 | /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ | |
279 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, | |
280 | offsetof(struct sk_buff, vlan_proto)); | |
281 | /* A = ntohs(A) [emitting a nop or swap16] */ | |
282 | *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); | |
283 | break; | |
284 | ||
bd4cf0ed AS |
285 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: |
286 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
287 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
288 | case SKF_AD_OFF + SKF_AD_CPU: | |
4cd3675e | 289 | case SKF_AD_OFF + SKF_AD_RANDOM: |
e430f34e | 290 | /* arg1 = CTX */ |
f8f6d679 | 291 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); |
bd4cf0ed | 292 | /* arg2 = A */ |
f8f6d679 | 293 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); |
bd4cf0ed | 294 | /* arg3 = X */ |
f8f6d679 | 295 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); |
e430f34e | 296 | /* Emit call(arg1=CTX, arg2=A, arg3=X) */ |
bd4cf0ed AS |
297 | switch (fp->k) { |
298 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
f8f6d679 | 299 | *insn = BPF_EMIT_CALL(__skb_get_pay_offset); |
bd4cf0ed AS |
300 | break; |
301 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
f8f6d679 | 302 | *insn = BPF_EMIT_CALL(__skb_get_nlattr); |
bd4cf0ed AS |
303 | break; |
304 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
f8f6d679 | 305 | *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest); |
bd4cf0ed AS |
306 | break; |
307 | case SKF_AD_OFF + SKF_AD_CPU: | |
f8f6d679 | 308 | *insn = BPF_EMIT_CALL(__get_raw_cpu_id); |
bd4cf0ed | 309 | break; |
4cd3675e | 310 | case SKF_AD_OFF + SKF_AD_RANDOM: |
3ad00405 DB |
311 | *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); |
312 | bpf_user_rnd_init_once(); | |
4cd3675e | 313 | break; |
bd4cf0ed AS |
314 | } |
315 | break; | |
316 | ||
317 | case SKF_AD_OFF + SKF_AD_ALU_XOR_X: | |
9739eef1 AS |
318 | /* A ^= X */ |
319 | *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); | |
bd4cf0ed AS |
320 | break; |
321 | ||
322 | default: | |
323 | /* This is just a dummy call to avoid letting the compiler | |
324 | * evict __bpf_call_base() as an optimization. Placed here | |
325 | * where no-one bothers. | |
326 | */ | |
327 | BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); | |
328 | return false; | |
329 | } | |
330 | ||
331 | *insnp = insn; | |
332 | return true; | |
333 | } | |
334 | ||
335 | /** | |
8fb575ca | 336 | * bpf_convert_filter - convert filter program |
bd4cf0ed AS |
337 | * @prog: the user passed filter program |
338 | * @len: the length of the user passed filter program | |
339 | * @new_prog: buffer where converted program will be stored | |
340 | * @new_len: pointer to store length of converted program | |
341 | * | |
342 | * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style. | |
343 | * Conversion workflow: | |
344 | * | |
345 | * 1) First pass for calculating the new program length: | |
8fb575ca | 346 | * bpf_convert_filter(old_prog, old_len, NULL, &new_len) |
bd4cf0ed AS |
347 | * |
348 | * 2) 2nd pass to remap in two passes: 1st pass finds new | |
349 | * jump offsets, 2nd pass remapping: | |
2695fb55 | 350 | * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len); |
8fb575ca | 351 | * bpf_convert_filter(old_prog, old_len, new_prog, &new_len); |
bd4cf0ed | 352 | */ |
d9e12f42 NS |
353 | static int bpf_convert_filter(struct sock_filter *prog, int len, |
354 | struct bpf_insn *new_prog, int *new_len) | |
bd4cf0ed AS |
355 | { |
356 | int new_flen = 0, pass = 0, target, i; | |
2695fb55 | 357 | struct bpf_insn *new_insn; |
bd4cf0ed AS |
358 | struct sock_filter *fp; |
359 | int *addrs = NULL; | |
360 | u8 bpf_src; | |
361 | ||
362 | BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); | |
30743837 | 363 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); |
bd4cf0ed | 364 | |
6f9a093b | 365 | if (len <= 0 || len > BPF_MAXINSNS) |
bd4cf0ed AS |
366 | return -EINVAL; |
367 | ||
368 | if (new_prog) { | |
658da937 DB |
369 | addrs = kcalloc(len, sizeof(*addrs), |
370 | GFP_KERNEL | __GFP_NOWARN); | |
bd4cf0ed AS |
371 | if (!addrs) |
372 | return -ENOMEM; | |
373 | } | |
374 | ||
375 | do_pass: | |
376 | new_insn = new_prog; | |
377 | fp = prog; | |
378 | ||
8b614aeb DB |
379 | /* Classic BPF related prologue emission. */ |
380 | if (new_insn) { | |
381 | /* Classic BPF expects A and X to be reset first. These need | |
382 | * to be guaranteed to be the first two instructions. | |
383 | */ | |
384 | *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); | |
385 | *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); | |
386 | ||
387 | /* All programs must keep CTX in callee saved BPF_REG_CTX. | |
388 | * In eBPF case it's done by the compiler, here we need to | |
389 | * do this ourself. Initial CTX is present in BPF_REG_ARG1. | |
390 | */ | |
391 | *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); | |
392 | } else { | |
393 | new_insn += 3; | |
394 | } | |
bd4cf0ed AS |
395 | |
396 | for (i = 0; i < len; fp++, i++) { | |
2695fb55 AS |
397 | struct bpf_insn tmp_insns[6] = { }; |
398 | struct bpf_insn *insn = tmp_insns; | |
bd4cf0ed AS |
399 | |
400 | if (addrs) | |
401 | addrs[i] = new_insn - new_prog; | |
402 | ||
403 | switch (fp->code) { | |
404 | /* All arithmetic insns and skb loads map as-is. */ | |
405 | case BPF_ALU | BPF_ADD | BPF_X: | |
406 | case BPF_ALU | BPF_ADD | BPF_K: | |
407 | case BPF_ALU | BPF_SUB | BPF_X: | |
408 | case BPF_ALU | BPF_SUB | BPF_K: | |
409 | case BPF_ALU | BPF_AND | BPF_X: | |
410 | case BPF_ALU | BPF_AND | BPF_K: | |
411 | case BPF_ALU | BPF_OR | BPF_X: | |
412 | case BPF_ALU | BPF_OR | BPF_K: | |
413 | case BPF_ALU | BPF_LSH | BPF_X: | |
414 | case BPF_ALU | BPF_LSH | BPF_K: | |
415 | case BPF_ALU | BPF_RSH | BPF_X: | |
416 | case BPF_ALU | BPF_RSH | BPF_K: | |
417 | case BPF_ALU | BPF_XOR | BPF_X: | |
418 | case BPF_ALU | BPF_XOR | BPF_K: | |
419 | case BPF_ALU | BPF_MUL | BPF_X: | |
420 | case BPF_ALU | BPF_MUL | BPF_K: | |
421 | case BPF_ALU | BPF_DIV | BPF_X: | |
422 | case BPF_ALU | BPF_DIV | BPF_K: | |
423 | case BPF_ALU | BPF_MOD | BPF_X: | |
424 | case BPF_ALU | BPF_MOD | BPF_K: | |
425 | case BPF_ALU | BPF_NEG: | |
426 | case BPF_LD | BPF_ABS | BPF_W: | |
427 | case BPF_LD | BPF_ABS | BPF_H: | |
428 | case BPF_LD | BPF_ABS | BPF_B: | |
429 | case BPF_LD | BPF_IND | BPF_W: | |
430 | case BPF_LD | BPF_IND | BPF_H: | |
431 | case BPF_LD | BPF_IND | BPF_B: | |
432 | /* Check for overloaded BPF extension and | |
433 | * directly convert it if found, otherwise | |
434 | * just move on with mapping. | |
435 | */ | |
436 | if (BPF_CLASS(fp->code) == BPF_LD && | |
437 | BPF_MODE(fp->code) == BPF_ABS && | |
438 | convert_bpf_extensions(fp, &insn)) | |
439 | break; | |
440 | ||
f8f6d679 | 441 | *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); |
bd4cf0ed AS |
442 | break; |
443 | ||
f8f6d679 DB |
444 | /* Jump transformation cannot use BPF block macros |
445 | * everywhere as offset calculation and target updates | |
446 | * require a bit more work than the rest, i.e. jump | |
447 | * opcodes map as-is, but offsets need adjustment. | |
448 | */ | |
449 | ||
450 | #define BPF_EMIT_JMP \ | |
bd4cf0ed AS |
451 | do { \ |
452 | if (target >= len || target < 0) \ | |
453 | goto err; \ | |
454 | insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ | |
455 | /* Adjust pc relative offset for 2nd or 3rd insn. */ \ | |
456 | insn->off -= insn - tmp_insns; \ | |
457 | } while (0) | |
458 | ||
f8f6d679 DB |
459 | case BPF_JMP | BPF_JA: |
460 | target = i + fp->k + 1; | |
461 | insn->code = fp->code; | |
462 | BPF_EMIT_JMP; | |
bd4cf0ed AS |
463 | break; |
464 | ||
465 | case BPF_JMP | BPF_JEQ | BPF_K: | |
466 | case BPF_JMP | BPF_JEQ | BPF_X: | |
467 | case BPF_JMP | BPF_JSET | BPF_K: | |
468 | case BPF_JMP | BPF_JSET | BPF_X: | |
469 | case BPF_JMP | BPF_JGT | BPF_K: | |
470 | case BPF_JMP | BPF_JGT | BPF_X: | |
471 | case BPF_JMP | BPF_JGE | BPF_K: | |
472 | case BPF_JMP | BPF_JGE | BPF_X: | |
473 | if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { | |
474 | /* BPF immediates are signed, zero extend | |
475 | * immediate into tmp register and use it | |
476 | * in compare insn. | |
477 | */ | |
f8f6d679 | 478 | *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); |
bd4cf0ed | 479 | |
e430f34e AS |
480 | insn->dst_reg = BPF_REG_A; |
481 | insn->src_reg = BPF_REG_TMP; | |
bd4cf0ed AS |
482 | bpf_src = BPF_X; |
483 | } else { | |
e430f34e | 484 | insn->dst_reg = BPF_REG_A; |
bd4cf0ed AS |
485 | insn->imm = fp->k; |
486 | bpf_src = BPF_SRC(fp->code); | |
19539ce7 | 487 | insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; |
1da177e4 | 488 | } |
bd4cf0ed AS |
489 | |
490 | /* Common case where 'jump_false' is next insn. */ | |
491 | if (fp->jf == 0) { | |
492 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
493 | target = i + fp->jt + 1; | |
f8f6d679 | 494 | BPF_EMIT_JMP; |
bd4cf0ed | 495 | break; |
1da177e4 | 496 | } |
bd4cf0ed AS |
497 | |
498 | /* Convert JEQ into JNE when 'jump_true' is next insn. */ | |
499 | if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) { | |
500 | insn->code = BPF_JMP | BPF_JNE | bpf_src; | |
501 | target = i + fp->jf + 1; | |
f8f6d679 | 502 | BPF_EMIT_JMP; |
bd4cf0ed | 503 | break; |
0b05b2a4 | 504 | } |
bd4cf0ed AS |
505 | |
506 | /* Other jumps are mapped into two insns: Jxx and JA. */ | |
507 | target = i + fp->jt + 1; | |
508 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
f8f6d679 | 509 | BPF_EMIT_JMP; |
bd4cf0ed AS |
510 | insn++; |
511 | ||
512 | insn->code = BPF_JMP | BPF_JA; | |
513 | target = i + fp->jf + 1; | |
f8f6d679 | 514 | BPF_EMIT_JMP; |
bd4cf0ed AS |
515 | break; |
516 | ||
517 | /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ | |
518 | case BPF_LDX | BPF_MSH | BPF_B: | |
9739eef1 | 519 | /* tmp = A */ |
f8f6d679 | 520 | *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A); |
1268e253 | 521 | /* A = BPF_R0 = *(u8 *) (skb->data + K) */ |
f8f6d679 | 522 | *insn++ = BPF_LD_ABS(BPF_B, fp->k); |
9739eef1 | 523 | /* A &= 0xf */ |
f8f6d679 | 524 | *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); |
9739eef1 | 525 | /* A <<= 2 */ |
f8f6d679 | 526 | *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); |
9739eef1 | 527 | /* X = A */ |
f8f6d679 | 528 | *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
9739eef1 | 529 | /* A = tmp */ |
f8f6d679 | 530 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); |
bd4cf0ed AS |
531 | break; |
532 | ||
6205b9cf DB |
533 | /* RET_K is remaped into 2 insns. RET_A case doesn't need an |
534 | * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. | |
535 | */ | |
bd4cf0ed AS |
536 | case BPF_RET | BPF_A: |
537 | case BPF_RET | BPF_K: | |
6205b9cf DB |
538 | if (BPF_RVAL(fp->code) == BPF_K) |
539 | *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, | |
540 | 0, fp->k); | |
9739eef1 | 541 | *insn = BPF_EXIT_INSN(); |
bd4cf0ed AS |
542 | break; |
543 | ||
544 | /* Store to stack. */ | |
545 | case BPF_ST: | |
546 | case BPF_STX: | |
f8f6d679 DB |
547 | *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == |
548 | BPF_ST ? BPF_REG_A : BPF_REG_X, | |
549 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
550 | break; |
551 | ||
552 | /* Load from stack. */ | |
553 | case BPF_LD | BPF_MEM: | |
554 | case BPF_LDX | BPF_MEM: | |
f8f6d679 DB |
555 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
556 | BPF_REG_A : BPF_REG_X, BPF_REG_FP, | |
557 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
558 | break; |
559 | ||
560 | /* A = K or X = K */ | |
561 | case BPF_LD | BPF_IMM: | |
562 | case BPF_LDX | BPF_IMM: | |
f8f6d679 DB |
563 | *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? |
564 | BPF_REG_A : BPF_REG_X, fp->k); | |
bd4cf0ed AS |
565 | break; |
566 | ||
567 | /* X = A */ | |
568 | case BPF_MISC | BPF_TAX: | |
f8f6d679 | 569 | *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
bd4cf0ed AS |
570 | break; |
571 | ||
572 | /* A = X */ | |
573 | case BPF_MISC | BPF_TXA: | |
f8f6d679 | 574 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); |
bd4cf0ed AS |
575 | break; |
576 | ||
577 | /* A = skb->len or X = skb->len */ | |
578 | case BPF_LD | BPF_W | BPF_LEN: | |
579 | case BPF_LDX | BPF_W | BPF_LEN: | |
f8f6d679 DB |
580 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
581 | BPF_REG_A : BPF_REG_X, BPF_REG_CTX, | |
582 | offsetof(struct sk_buff, len)); | |
bd4cf0ed AS |
583 | break; |
584 | ||
f8f6d679 | 585 | /* Access seccomp_data fields. */ |
bd4cf0ed | 586 | case BPF_LDX | BPF_ABS | BPF_W: |
9739eef1 AS |
587 | /* A = *(u32 *) (ctx + K) */ |
588 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); | |
bd4cf0ed AS |
589 | break; |
590 | ||
ca9f1fd2 | 591 | /* Unknown instruction. */ |
1da177e4 | 592 | default: |
bd4cf0ed | 593 | goto err; |
1da177e4 | 594 | } |
bd4cf0ed AS |
595 | |
596 | insn++; | |
597 | if (new_prog) | |
598 | memcpy(new_insn, tmp_insns, | |
599 | sizeof(*insn) * (insn - tmp_insns)); | |
bd4cf0ed | 600 | new_insn += insn - tmp_insns; |
1da177e4 LT |
601 | } |
602 | ||
bd4cf0ed AS |
603 | if (!new_prog) { |
604 | /* Only calculating new length. */ | |
605 | *new_len = new_insn - new_prog; | |
606 | return 0; | |
607 | } | |
608 | ||
609 | pass++; | |
610 | if (new_flen != new_insn - new_prog) { | |
611 | new_flen = new_insn - new_prog; | |
612 | if (pass > 2) | |
613 | goto err; | |
bd4cf0ed AS |
614 | goto do_pass; |
615 | } | |
616 | ||
617 | kfree(addrs); | |
618 | BUG_ON(*new_len != new_flen); | |
1da177e4 | 619 | return 0; |
bd4cf0ed AS |
620 | err: |
621 | kfree(addrs); | |
622 | return -EINVAL; | |
1da177e4 LT |
623 | } |
624 | ||
bd4cf0ed | 625 | /* Security: |
bd4cf0ed | 626 | * |
2d5311e4 | 627 | * As we dont want to clear mem[] array for each packet going through |
8ea6e345 | 628 | * __bpf_prog_run(), we check that filter loaded by user never try to read |
2d5311e4 | 629 | * a cell if not previously written, and we check all branches to be sure |
25985edc | 630 | * a malicious user doesn't try to abuse us. |
2d5311e4 | 631 | */ |
ec31a05c | 632 | static int check_load_and_stores(const struct sock_filter *filter, int flen) |
2d5311e4 | 633 | { |
34805931 | 634 | u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ |
2d5311e4 ED |
635 | int pc, ret = 0; |
636 | ||
637 | BUILD_BUG_ON(BPF_MEMWORDS > 16); | |
34805931 | 638 | |
99e72a0f | 639 | masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); |
2d5311e4 ED |
640 | if (!masks) |
641 | return -ENOMEM; | |
34805931 | 642 | |
2d5311e4 ED |
643 | memset(masks, 0xff, flen * sizeof(*masks)); |
644 | ||
645 | for (pc = 0; pc < flen; pc++) { | |
646 | memvalid &= masks[pc]; | |
647 | ||
648 | switch (filter[pc].code) { | |
34805931 DB |
649 | case BPF_ST: |
650 | case BPF_STX: | |
2d5311e4 ED |
651 | memvalid |= (1 << filter[pc].k); |
652 | break; | |
34805931 DB |
653 | case BPF_LD | BPF_MEM: |
654 | case BPF_LDX | BPF_MEM: | |
2d5311e4 ED |
655 | if (!(memvalid & (1 << filter[pc].k))) { |
656 | ret = -EINVAL; | |
657 | goto error; | |
658 | } | |
659 | break; | |
34805931 DB |
660 | case BPF_JMP | BPF_JA: |
661 | /* A jump must set masks on target */ | |
2d5311e4 ED |
662 | masks[pc + 1 + filter[pc].k] &= memvalid; |
663 | memvalid = ~0; | |
664 | break; | |
34805931 DB |
665 | case BPF_JMP | BPF_JEQ | BPF_K: |
666 | case BPF_JMP | BPF_JEQ | BPF_X: | |
667 | case BPF_JMP | BPF_JGE | BPF_K: | |
668 | case BPF_JMP | BPF_JGE | BPF_X: | |
669 | case BPF_JMP | BPF_JGT | BPF_K: | |
670 | case BPF_JMP | BPF_JGT | BPF_X: | |
671 | case BPF_JMP | BPF_JSET | BPF_K: | |
672 | case BPF_JMP | BPF_JSET | BPF_X: | |
673 | /* A jump must set masks on targets */ | |
2d5311e4 ED |
674 | masks[pc + 1 + filter[pc].jt] &= memvalid; |
675 | masks[pc + 1 + filter[pc].jf] &= memvalid; | |
676 | memvalid = ~0; | |
677 | break; | |
678 | } | |
679 | } | |
680 | error: | |
681 | kfree(masks); | |
682 | return ret; | |
683 | } | |
684 | ||
34805931 DB |
685 | static bool chk_code_allowed(u16 code_to_probe) |
686 | { | |
687 | static const bool codes[] = { | |
688 | /* 32 bit ALU operations */ | |
689 | [BPF_ALU | BPF_ADD | BPF_K] = true, | |
690 | [BPF_ALU | BPF_ADD | BPF_X] = true, | |
691 | [BPF_ALU | BPF_SUB | BPF_K] = true, | |
692 | [BPF_ALU | BPF_SUB | BPF_X] = true, | |
693 | [BPF_ALU | BPF_MUL | BPF_K] = true, | |
694 | [BPF_ALU | BPF_MUL | BPF_X] = true, | |
695 | [BPF_ALU | BPF_DIV | BPF_K] = true, | |
696 | [BPF_ALU | BPF_DIV | BPF_X] = true, | |
697 | [BPF_ALU | BPF_MOD | BPF_K] = true, | |
698 | [BPF_ALU | BPF_MOD | BPF_X] = true, | |
699 | [BPF_ALU | BPF_AND | BPF_K] = true, | |
700 | [BPF_ALU | BPF_AND | BPF_X] = true, | |
701 | [BPF_ALU | BPF_OR | BPF_K] = true, | |
702 | [BPF_ALU | BPF_OR | BPF_X] = true, | |
703 | [BPF_ALU | BPF_XOR | BPF_K] = true, | |
704 | [BPF_ALU | BPF_XOR | BPF_X] = true, | |
705 | [BPF_ALU | BPF_LSH | BPF_K] = true, | |
706 | [BPF_ALU | BPF_LSH | BPF_X] = true, | |
707 | [BPF_ALU | BPF_RSH | BPF_K] = true, | |
708 | [BPF_ALU | BPF_RSH | BPF_X] = true, | |
709 | [BPF_ALU | BPF_NEG] = true, | |
710 | /* Load instructions */ | |
711 | [BPF_LD | BPF_W | BPF_ABS] = true, | |
712 | [BPF_LD | BPF_H | BPF_ABS] = true, | |
713 | [BPF_LD | BPF_B | BPF_ABS] = true, | |
714 | [BPF_LD | BPF_W | BPF_LEN] = true, | |
715 | [BPF_LD | BPF_W | BPF_IND] = true, | |
716 | [BPF_LD | BPF_H | BPF_IND] = true, | |
717 | [BPF_LD | BPF_B | BPF_IND] = true, | |
718 | [BPF_LD | BPF_IMM] = true, | |
719 | [BPF_LD | BPF_MEM] = true, | |
720 | [BPF_LDX | BPF_W | BPF_LEN] = true, | |
721 | [BPF_LDX | BPF_B | BPF_MSH] = true, | |
722 | [BPF_LDX | BPF_IMM] = true, | |
723 | [BPF_LDX | BPF_MEM] = true, | |
724 | /* Store instructions */ | |
725 | [BPF_ST] = true, | |
726 | [BPF_STX] = true, | |
727 | /* Misc instructions */ | |
728 | [BPF_MISC | BPF_TAX] = true, | |
729 | [BPF_MISC | BPF_TXA] = true, | |
730 | /* Return instructions */ | |
731 | [BPF_RET | BPF_K] = true, | |
732 | [BPF_RET | BPF_A] = true, | |
733 | /* Jump instructions */ | |
734 | [BPF_JMP | BPF_JA] = true, | |
735 | [BPF_JMP | BPF_JEQ | BPF_K] = true, | |
736 | [BPF_JMP | BPF_JEQ | BPF_X] = true, | |
737 | [BPF_JMP | BPF_JGE | BPF_K] = true, | |
738 | [BPF_JMP | BPF_JGE | BPF_X] = true, | |
739 | [BPF_JMP | BPF_JGT | BPF_K] = true, | |
740 | [BPF_JMP | BPF_JGT | BPF_X] = true, | |
741 | [BPF_JMP | BPF_JSET | BPF_K] = true, | |
742 | [BPF_JMP | BPF_JSET | BPF_X] = true, | |
743 | }; | |
744 | ||
745 | if (code_to_probe >= ARRAY_SIZE(codes)) | |
746 | return false; | |
747 | ||
748 | return codes[code_to_probe]; | |
749 | } | |
750 | ||
1da177e4 | 751 | /** |
4df95ff4 | 752 | * bpf_check_classic - verify socket filter code |
1da177e4 LT |
753 | * @filter: filter to verify |
754 | * @flen: length of filter | |
755 | * | |
756 | * Check the user's filter code. If we let some ugly | |
757 | * filter code slip through kaboom! The filter must contain | |
93699863 KK |
758 | * no references or jumps that are out of range, no illegal |
759 | * instructions, and must end with a RET instruction. | |
1da177e4 | 760 | * |
7b11f69f KK |
761 | * All jumps are forward as they are not signed. |
762 | * | |
763 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
1da177e4 | 764 | */ |
d9e12f42 NS |
765 | static int bpf_check_classic(const struct sock_filter *filter, |
766 | unsigned int flen) | |
1da177e4 | 767 | { |
aa1113d9 | 768 | bool anc_found; |
34805931 | 769 | int pc; |
1da177e4 | 770 | |
1b93ae64 | 771 | if (flen == 0 || flen > BPF_MAXINSNS) |
1da177e4 LT |
772 | return -EINVAL; |
773 | ||
34805931 | 774 | /* Check the filter code now */ |
1da177e4 | 775 | for (pc = 0; pc < flen; pc++) { |
ec31a05c | 776 | const struct sock_filter *ftest = &filter[pc]; |
93699863 | 777 | |
34805931 DB |
778 | /* May we actually operate on this code? */ |
779 | if (!chk_code_allowed(ftest->code)) | |
cba328fc | 780 | return -EINVAL; |
34805931 | 781 | |
93699863 | 782 | /* Some instructions need special checks */ |
34805931 DB |
783 | switch (ftest->code) { |
784 | case BPF_ALU | BPF_DIV | BPF_K: | |
785 | case BPF_ALU | BPF_MOD | BPF_K: | |
786 | /* Check for division by zero */ | |
b6069a95 ED |
787 | if (ftest->k == 0) |
788 | return -EINVAL; | |
789 | break; | |
229394e8 RV |
790 | case BPF_ALU | BPF_LSH | BPF_K: |
791 | case BPF_ALU | BPF_RSH | BPF_K: | |
792 | if (ftest->k >= 32) | |
793 | return -EINVAL; | |
794 | break; | |
34805931 DB |
795 | case BPF_LD | BPF_MEM: |
796 | case BPF_LDX | BPF_MEM: | |
797 | case BPF_ST: | |
798 | case BPF_STX: | |
799 | /* Check for invalid memory addresses */ | |
93699863 KK |
800 | if (ftest->k >= BPF_MEMWORDS) |
801 | return -EINVAL; | |
802 | break; | |
34805931 DB |
803 | case BPF_JMP | BPF_JA: |
804 | /* Note, the large ftest->k might cause loops. | |
93699863 KK |
805 | * Compare this with conditional jumps below, |
806 | * where offsets are limited. --ANK (981016) | |
807 | */ | |
34805931 | 808 | if (ftest->k >= (unsigned int)(flen - pc - 1)) |
93699863 | 809 | return -EINVAL; |
01f2f3f6 | 810 | break; |
34805931 DB |
811 | case BPF_JMP | BPF_JEQ | BPF_K: |
812 | case BPF_JMP | BPF_JEQ | BPF_X: | |
813 | case BPF_JMP | BPF_JGE | BPF_K: | |
814 | case BPF_JMP | BPF_JGE | BPF_X: | |
815 | case BPF_JMP | BPF_JGT | BPF_K: | |
816 | case BPF_JMP | BPF_JGT | BPF_X: | |
817 | case BPF_JMP | BPF_JSET | BPF_K: | |
818 | case BPF_JMP | BPF_JSET | BPF_X: | |
819 | /* Both conditionals must be safe */ | |
e35bedf3 | 820 | if (pc + ftest->jt + 1 >= flen || |
93699863 KK |
821 | pc + ftest->jf + 1 >= flen) |
822 | return -EINVAL; | |
cba328fc | 823 | break; |
34805931 DB |
824 | case BPF_LD | BPF_W | BPF_ABS: |
825 | case BPF_LD | BPF_H | BPF_ABS: | |
826 | case BPF_LD | BPF_B | BPF_ABS: | |
aa1113d9 | 827 | anc_found = false; |
34805931 DB |
828 | if (bpf_anc_helper(ftest) & BPF_ANC) |
829 | anc_found = true; | |
830 | /* Ancillary operation unknown or unsupported */ | |
aa1113d9 DB |
831 | if (anc_found == false && ftest->k >= SKF_AD_OFF) |
832 | return -EINVAL; | |
01f2f3f6 HPP |
833 | } |
834 | } | |
93699863 | 835 | |
34805931 | 836 | /* Last instruction must be a RET code */ |
01f2f3f6 | 837 | switch (filter[flen - 1].code) { |
34805931 DB |
838 | case BPF_RET | BPF_K: |
839 | case BPF_RET | BPF_A: | |
2d5311e4 | 840 | return check_load_and_stores(filter, flen); |
cba328fc | 841 | } |
34805931 | 842 | |
cba328fc | 843 | return -EINVAL; |
1da177e4 LT |
844 | } |
845 | ||
7ae457c1 AS |
846 | static int bpf_prog_store_orig_filter(struct bpf_prog *fp, |
847 | const struct sock_fprog *fprog) | |
a3ea269b | 848 | { |
009937e7 | 849 | unsigned int fsize = bpf_classic_proglen(fprog); |
a3ea269b DB |
850 | struct sock_fprog_kern *fkprog; |
851 | ||
852 | fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); | |
853 | if (!fp->orig_prog) | |
854 | return -ENOMEM; | |
855 | ||
856 | fkprog = fp->orig_prog; | |
857 | fkprog->len = fprog->len; | |
658da937 DB |
858 | |
859 | fkprog->filter = kmemdup(fp->insns, fsize, | |
860 | GFP_KERNEL | __GFP_NOWARN); | |
a3ea269b DB |
861 | if (!fkprog->filter) { |
862 | kfree(fp->orig_prog); | |
863 | return -ENOMEM; | |
864 | } | |
865 | ||
866 | return 0; | |
867 | } | |
868 | ||
7ae457c1 | 869 | static void bpf_release_orig_filter(struct bpf_prog *fp) |
a3ea269b DB |
870 | { |
871 | struct sock_fprog_kern *fprog = fp->orig_prog; | |
872 | ||
873 | if (fprog) { | |
874 | kfree(fprog->filter); | |
875 | kfree(fprog); | |
876 | } | |
877 | } | |
878 | ||
7ae457c1 AS |
879 | static void __bpf_prog_release(struct bpf_prog *prog) |
880 | { | |
24701ece | 881 | if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { |
89aa0758 AS |
882 | bpf_prog_put(prog); |
883 | } else { | |
884 | bpf_release_orig_filter(prog); | |
885 | bpf_prog_free(prog); | |
886 | } | |
7ae457c1 AS |
887 | } |
888 | ||
34c5bd66 PN |
889 | static void __sk_filter_release(struct sk_filter *fp) |
890 | { | |
7ae457c1 AS |
891 | __bpf_prog_release(fp->prog); |
892 | kfree(fp); | |
34c5bd66 PN |
893 | } |
894 | ||
47e958ea | 895 | /** |
46bcf14f | 896 | * sk_filter_release_rcu - Release a socket filter by rcu_head |
47e958ea PE |
897 | * @rcu: rcu_head that contains the sk_filter to free |
898 | */ | |
fbc907f0 | 899 | static void sk_filter_release_rcu(struct rcu_head *rcu) |
47e958ea PE |
900 | { |
901 | struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); | |
902 | ||
34c5bd66 | 903 | __sk_filter_release(fp); |
47e958ea | 904 | } |
fbc907f0 DB |
905 | |
906 | /** | |
907 | * sk_filter_release - release a socket filter | |
908 | * @fp: filter to remove | |
909 | * | |
910 | * Remove a filter from a socket and release its resources. | |
911 | */ | |
912 | static void sk_filter_release(struct sk_filter *fp) | |
913 | { | |
914 | if (atomic_dec_and_test(&fp->refcnt)) | |
915 | call_rcu(&fp->rcu, sk_filter_release_rcu); | |
916 | } | |
917 | ||
918 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) | |
919 | { | |
7ae457c1 | 920 | u32 filter_size = bpf_prog_size(fp->prog->len); |
fbc907f0 | 921 | |
278571ba AS |
922 | atomic_sub(filter_size, &sk->sk_omem_alloc); |
923 | sk_filter_release(fp); | |
fbc907f0 | 924 | } |
47e958ea | 925 | |
278571ba AS |
926 | /* try to charge the socket memory if there is space available |
927 | * return true on success | |
928 | */ | |
929 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) | |
bd4cf0ed | 930 | { |
7ae457c1 | 931 | u32 filter_size = bpf_prog_size(fp->prog->len); |
278571ba AS |
932 | |
933 | /* same check as in sock_kmalloc() */ | |
934 | if (filter_size <= sysctl_optmem_max && | |
935 | atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { | |
936 | atomic_inc(&fp->refcnt); | |
937 | atomic_add(filter_size, &sk->sk_omem_alloc); | |
938 | return true; | |
bd4cf0ed | 939 | } |
278571ba | 940 | return false; |
bd4cf0ed AS |
941 | } |
942 | ||
7ae457c1 | 943 | static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) |
bd4cf0ed AS |
944 | { |
945 | struct sock_filter *old_prog; | |
7ae457c1 | 946 | struct bpf_prog *old_fp; |
34805931 | 947 | int err, new_len, old_len = fp->len; |
bd4cf0ed AS |
948 | |
949 | /* We are free to overwrite insns et al right here as it | |
950 | * won't be used at this point in time anymore internally | |
951 | * after the migration to the internal BPF instruction | |
952 | * representation. | |
953 | */ | |
954 | BUILD_BUG_ON(sizeof(struct sock_filter) != | |
2695fb55 | 955 | sizeof(struct bpf_insn)); |
bd4cf0ed | 956 | |
bd4cf0ed AS |
957 | /* Conversion cannot happen on overlapping memory areas, |
958 | * so we need to keep the user BPF around until the 2nd | |
959 | * pass. At this time, the user BPF is stored in fp->insns. | |
960 | */ | |
961 | old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), | |
658da937 | 962 | GFP_KERNEL | __GFP_NOWARN); |
bd4cf0ed AS |
963 | if (!old_prog) { |
964 | err = -ENOMEM; | |
965 | goto out_err; | |
966 | } | |
967 | ||
968 | /* 1st pass: calculate the new program length. */ | |
8fb575ca | 969 | err = bpf_convert_filter(old_prog, old_len, NULL, &new_len); |
bd4cf0ed AS |
970 | if (err) |
971 | goto out_err_free; | |
972 | ||
973 | /* Expand fp for appending the new filter representation. */ | |
974 | old_fp = fp; | |
60a3b225 | 975 | fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); |
bd4cf0ed AS |
976 | if (!fp) { |
977 | /* The old_fp is still around in case we couldn't | |
978 | * allocate new memory, so uncharge on that one. | |
979 | */ | |
980 | fp = old_fp; | |
981 | err = -ENOMEM; | |
982 | goto out_err_free; | |
983 | } | |
984 | ||
bd4cf0ed AS |
985 | fp->len = new_len; |
986 | ||
2695fb55 | 987 | /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ |
8fb575ca | 988 | err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len); |
bd4cf0ed | 989 | if (err) |
8fb575ca | 990 | /* 2nd bpf_convert_filter() can fail only if it fails |
bd4cf0ed AS |
991 | * to allocate memory, remapping must succeed. Note, |
992 | * that at this time old_fp has already been released | |
278571ba | 993 | * by krealloc(). |
bd4cf0ed AS |
994 | */ |
995 | goto out_err_free; | |
996 | ||
d1c55ab5 DB |
997 | /* We are guaranteed to never error here with cBPF to eBPF |
998 | * transitions, since there's no issue with type compatibility | |
999 | * checks on program arrays. | |
1000 | */ | |
1001 | fp = bpf_prog_select_runtime(fp, &err); | |
5fe821a9 | 1002 | |
bd4cf0ed AS |
1003 | kfree(old_prog); |
1004 | return fp; | |
1005 | ||
1006 | out_err_free: | |
1007 | kfree(old_prog); | |
1008 | out_err: | |
7ae457c1 | 1009 | __bpf_prog_release(fp); |
bd4cf0ed AS |
1010 | return ERR_PTR(err); |
1011 | } | |
1012 | ||
ac67eb2c DB |
1013 | static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, |
1014 | bpf_aux_classic_check_t trans) | |
302d6637 JP |
1015 | { |
1016 | int err; | |
1017 | ||
bd4cf0ed | 1018 | fp->bpf_func = NULL; |
a91263d5 | 1019 | fp->jited = 0; |
302d6637 | 1020 | |
4df95ff4 | 1021 | err = bpf_check_classic(fp->insns, fp->len); |
418c96ac | 1022 | if (err) { |
7ae457c1 | 1023 | __bpf_prog_release(fp); |
bd4cf0ed | 1024 | return ERR_PTR(err); |
418c96ac | 1025 | } |
302d6637 | 1026 | |
4ae92bc7 NS |
1027 | /* There might be additional checks and transformations |
1028 | * needed on classic filters, f.e. in case of seccomp. | |
1029 | */ | |
1030 | if (trans) { | |
1031 | err = trans(fp->insns, fp->len); | |
1032 | if (err) { | |
1033 | __bpf_prog_release(fp); | |
1034 | return ERR_PTR(err); | |
1035 | } | |
1036 | } | |
1037 | ||
bd4cf0ed AS |
1038 | /* Probe if we can JIT compile the filter and if so, do |
1039 | * the compilation of the filter. | |
1040 | */ | |
302d6637 | 1041 | bpf_jit_compile(fp); |
bd4cf0ed AS |
1042 | |
1043 | /* JIT compiler couldn't process this filter, so do the | |
1044 | * internal BPF translation for the optimized interpreter. | |
1045 | */ | |
5fe821a9 | 1046 | if (!fp->jited) |
7ae457c1 | 1047 | fp = bpf_migrate_filter(fp); |
bd4cf0ed AS |
1048 | |
1049 | return fp; | |
302d6637 JP |
1050 | } |
1051 | ||
1052 | /** | |
7ae457c1 | 1053 | * bpf_prog_create - create an unattached filter |
c6c4b97c | 1054 | * @pfp: the unattached filter that is created |
677a9fd3 | 1055 | * @fprog: the filter program |
302d6637 | 1056 | * |
c6c4b97c | 1057 | * Create a filter independent of any socket. We first run some |
302d6637 JP |
1058 | * sanity checks on it to make sure it does not explode on us later. |
1059 | * If an error occurs or there is insufficient memory for the filter | |
1060 | * a negative errno code is returned. On success the return is zero. | |
1061 | */ | |
7ae457c1 | 1062 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) |
302d6637 | 1063 | { |
009937e7 | 1064 | unsigned int fsize = bpf_classic_proglen(fprog); |
7ae457c1 | 1065 | struct bpf_prog *fp; |
302d6637 JP |
1066 | |
1067 | /* Make sure new filter is there and in the right amounts. */ | |
1068 | if (fprog->filter == NULL) | |
1069 | return -EINVAL; | |
1070 | ||
60a3b225 | 1071 | fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); |
302d6637 JP |
1072 | if (!fp) |
1073 | return -ENOMEM; | |
a3ea269b | 1074 | |
302d6637 JP |
1075 | memcpy(fp->insns, fprog->filter, fsize); |
1076 | ||
302d6637 | 1077 | fp->len = fprog->len; |
a3ea269b DB |
1078 | /* Since unattached filters are not copied back to user |
1079 | * space through sk_get_filter(), we do not need to hold | |
1080 | * a copy here, and can spare us the work. | |
1081 | */ | |
1082 | fp->orig_prog = NULL; | |
302d6637 | 1083 | |
7ae457c1 | 1084 | /* bpf_prepare_filter() already takes care of freeing |
bd4cf0ed AS |
1085 | * memory in case something goes wrong. |
1086 | */ | |
4ae92bc7 | 1087 | fp = bpf_prepare_filter(fp, NULL); |
bd4cf0ed AS |
1088 | if (IS_ERR(fp)) |
1089 | return PTR_ERR(fp); | |
302d6637 JP |
1090 | |
1091 | *pfp = fp; | |
1092 | return 0; | |
302d6637 | 1093 | } |
7ae457c1 | 1094 | EXPORT_SYMBOL_GPL(bpf_prog_create); |
302d6637 | 1095 | |
ac67eb2c DB |
1096 | /** |
1097 | * bpf_prog_create_from_user - create an unattached filter from user buffer | |
1098 | * @pfp: the unattached filter that is created | |
1099 | * @fprog: the filter program | |
1100 | * @trans: post-classic verifier transformation handler | |
bab18991 | 1101 | * @save_orig: save classic BPF program |
ac67eb2c DB |
1102 | * |
1103 | * This function effectively does the same as bpf_prog_create(), only | |
1104 | * that it builds up its insns buffer from user space provided buffer. | |
1105 | * It also allows for passing a bpf_aux_classic_check_t handler. | |
1106 | */ | |
1107 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, | |
bab18991 | 1108 | bpf_aux_classic_check_t trans, bool save_orig) |
ac67eb2c DB |
1109 | { |
1110 | unsigned int fsize = bpf_classic_proglen(fprog); | |
1111 | struct bpf_prog *fp; | |
bab18991 | 1112 | int err; |
ac67eb2c DB |
1113 | |
1114 | /* Make sure new filter is there and in the right amounts. */ | |
1115 | if (fprog->filter == NULL) | |
1116 | return -EINVAL; | |
1117 | ||
1118 | fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); | |
1119 | if (!fp) | |
1120 | return -ENOMEM; | |
1121 | ||
1122 | if (copy_from_user(fp->insns, fprog->filter, fsize)) { | |
1123 | __bpf_prog_free(fp); | |
1124 | return -EFAULT; | |
1125 | } | |
1126 | ||
1127 | fp->len = fprog->len; | |
ac67eb2c DB |
1128 | fp->orig_prog = NULL; |
1129 | ||
bab18991 DB |
1130 | if (save_orig) { |
1131 | err = bpf_prog_store_orig_filter(fp, fprog); | |
1132 | if (err) { | |
1133 | __bpf_prog_free(fp); | |
1134 | return -ENOMEM; | |
1135 | } | |
1136 | } | |
1137 | ||
ac67eb2c DB |
1138 | /* bpf_prepare_filter() already takes care of freeing |
1139 | * memory in case something goes wrong. | |
1140 | */ | |
1141 | fp = bpf_prepare_filter(fp, trans); | |
1142 | if (IS_ERR(fp)) | |
1143 | return PTR_ERR(fp); | |
1144 | ||
1145 | *pfp = fp; | |
1146 | return 0; | |
1147 | } | |
2ea273d7 | 1148 | EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); |
ac67eb2c | 1149 | |
7ae457c1 | 1150 | void bpf_prog_destroy(struct bpf_prog *fp) |
302d6637 | 1151 | { |
7ae457c1 | 1152 | __bpf_prog_release(fp); |
302d6637 | 1153 | } |
7ae457c1 | 1154 | EXPORT_SYMBOL_GPL(bpf_prog_destroy); |
302d6637 | 1155 | |
8ced425e | 1156 | static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) |
49b31e57 DB |
1157 | { |
1158 | struct sk_filter *fp, *old_fp; | |
1159 | ||
1160 | fp = kmalloc(sizeof(*fp), GFP_KERNEL); | |
1161 | if (!fp) | |
1162 | return -ENOMEM; | |
1163 | ||
1164 | fp->prog = prog; | |
1165 | atomic_set(&fp->refcnt, 0); | |
1166 | ||
1167 | if (!sk_filter_charge(sk, fp)) { | |
1168 | kfree(fp); | |
1169 | return -ENOMEM; | |
1170 | } | |
1171 | ||
8ced425e HFS |
1172 | old_fp = rcu_dereference_protected(sk->sk_filter, |
1173 | lockdep_sock_is_held(sk)); | |
49b31e57 | 1174 | rcu_assign_pointer(sk->sk_filter, fp); |
8ced425e | 1175 | |
49b31e57 DB |
1176 | if (old_fp) |
1177 | sk_filter_uncharge(sk, old_fp); | |
1178 | ||
1179 | return 0; | |
1180 | } | |
1181 | ||
538950a1 CG |
1182 | static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk) |
1183 | { | |
1184 | struct bpf_prog *old_prog; | |
1185 | int err; | |
1186 | ||
1187 | if (bpf_prog_size(prog->len) > sysctl_optmem_max) | |
1188 | return -ENOMEM; | |
1189 | ||
fa463497 | 1190 | if (sk_unhashed(sk) && sk->sk_reuseport) { |
538950a1 CG |
1191 | err = reuseport_alloc(sk); |
1192 | if (err) | |
1193 | return err; | |
1194 | } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) { | |
1195 | /* The socket wasn't bound with SO_REUSEPORT */ | |
1196 | return -EINVAL; | |
1197 | } | |
1198 | ||
1199 | old_prog = reuseport_attach_prog(sk, prog); | |
1200 | if (old_prog) | |
1201 | bpf_prog_destroy(old_prog); | |
1202 | ||
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | static | |
1207 | struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) | |
1da177e4 | 1208 | { |
009937e7 | 1209 | unsigned int fsize = bpf_classic_proglen(fprog); |
7ae457c1 AS |
1210 | unsigned int bpf_fsize = bpf_prog_size(fprog->len); |
1211 | struct bpf_prog *prog; | |
1da177e4 LT |
1212 | int err; |
1213 | ||
d59577b6 | 1214 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
538950a1 | 1215 | return ERR_PTR(-EPERM); |
d59577b6 | 1216 | |
1da177e4 | 1217 | /* Make sure new filter is there and in the right amounts. */ |
e35bedf3 | 1218 | if (fprog->filter == NULL) |
538950a1 | 1219 | return ERR_PTR(-EINVAL); |
1da177e4 | 1220 | |
60a3b225 | 1221 | prog = bpf_prog_alloc(bpf_fsize, 0); |
7ae457c1 | 1222 | if (!prog) |
538950a1 | 1223 | return ERR_PTR(-ENOMEM); |
a3ea269b | 1224 | |
7ae457c1 | 1225 | if (copy_from_user(prog->insns, fprog->filter, fsize)) { |
c0d1379a | 1226 | __bpf_prog_free(prog); |
538950a1 | 1227 | return ERR_PTR(-EFAULT); |
1da177e4 LT |
1228 | } |
1229 | ||
7ae457c1 | 1230 | prog->len = fprog->len; |
1da177e4 | 1231 | |
7ae457c1 | 1232 | err = bpf_prog_store_orig_filter(prog, fprog); |
a3ea269b | 1233 | if (err) { |
c0d1379a | 1234 | __bpf_prog_free(prog); |
538950a1 | 1235 | return ERR_PTR(-ENOMEM); |
a3ea269b DB |
1236 | } |
1237 | ||
7ae457c1 | 1238 | /* bpf_prepare_filter() already takes care of freeing |
bd4cf0ed AS |
1239 | * memory in case something goes wrong. |
1240 | */ | |
538950a1 CG |
1241 | return bpf_prepare_filter(prog, NULL); |
1242 | } | |
1243 | ||
1244 | /** | |
1245 | * sk_attach_filter - attach a socket filter | |
1246 | * @fprog: the filter program | |
1247 | * @sk: the socket to use | |
1248 | * | |
1249 | * Attach the user's filter code. We first run some sanity checks on | |
1250 | * it to make sure it does not explode on us later. If an error | |
1251 | * occurs or there is insufficient memory for the filter a negative | |
1252 | * errno code is returned. On success the return is zero. | |
1253 | */ | |
8ced425e | 1254 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) |
538950a1 CG |
1255 | { |
1256 | struct bpf_prog *prog = __get_filter(fprog, sk); | |
1257 | int err; | |
1258 | ||
7ae457c1 AS |
1259 | if (IS_ERR(prog)) |
1260 | return PTR_ERR(prog); | |
1261 | ||
8ced425e | 1262 | err = __sk_attach_prog(prog, sk); |
49b31e57 | 1263 | if (err < 0) { |
7ae457c1 | 1264 | __bpf_prog_release(prog); |
49b31e57 | 1265 | return err; |
278571ba AS |
1266 | } |
1267 | ||
d3904b73 | 1268 | return 0; |
1da177e4 | 1269 | } |
8ced425e | 1270 | EXPORT_SYMBOL_GPL(sk_attach_filter); |
1da177e4 | 1271 | |
538950a1 | 1272 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) |
89aa0758 | 1273 | { |
538950a1 | 1274 | struct bpf_prog *prog = __get_filter(fprog, sk); |
49b31e57 | 1275 | int err; |
89aa0758 | 1276 | |
538950a1 CG |
1277 | if (IS_ERR(prog)) |
1278 | return PTR_ERR(prog); | |
1279 | ||
1280 | err = __reuseport_attach_prog(prog, sk); | |
1281 | if (err < 0) { | |
1282 | __bpf_prog_release(prog); | |
1283 | return err; | |
1284 | } | |
1285 | ||
1286 | return 0; | |
1287 | } | |
1288 | ||
1289 | static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) | |
1290 | { | |
1291 | struct bpf_prog *prog; | |
1292 | ||
89aa0758 | 1293 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
538950a1 | 1294 | return ERR_PTR(-EPERM); |
89aa0758 AS |
1295 | |
1296 | prog = bpf_prog_get(ufd); | |
198bf1b0 | 1297 | if (IS_ERR(prog)) |
538950a1 | 1298 | return prog; |
89aa0758 | 1299 | |
24701ece | 1300 | if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) { |
89aa0758 | 1301 | bpf_prog_put(prog); |
538950a1 | 1302 | return ERR_PTR(-EINVAL); |
89aa0758 AS |
1303 | } |
1304 | ||
538950a1 CG |
1305 | return prog; |
1306 | } | |
1307 | ||
1308 | int sk_attach_bpf(u32 ufd, struct sock *sk) | |
1309 | { | |
1310 | struct bpf_prog *prog = __get_bpf(ufd, sk); | |
1311 | int err; | |
1312 | ||
1313 | if (IS_ERR(prog)) | |
1314 | return PTR_ERR(prog); | |
1315 | ||
8ced425e | 1316 | err = __sk_attach_prog(prog, sk); |
49b31e57 | 1317 | if (err < 0) { |
89aa0758 | 1318 | bpf_prog_put(prog); |
49b31e57 | 1319 | return err; |
89aa0758 AS |
1320 | } |
1321 | ||
89aa0758 AS |
1322 | return 0; |
1323 | } | |
1324 | ||
538950a1 CG |
1325 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) |
1326 | { | |
1327 | struct bpf_prog *prog = __get_bpf(ufd, sk); | |
1328 | int err; | |
1329 | ||
1330 | if (IS_ERR(prog)) | |
1331 | return PTR_ERR(prog); | |
1332 | ||
1333 | err = __reuseport_attach_prog(prog, sk); | |
1334 | if (err < 0) { | |
1335 | bpf_prog_put(prog); | |
1336 | return err; | |
1337 | } | |
1338 | ||
1339 | return 0; | |
1340 | } | |
1341 | ||
21cafc1d DB |
1342 | struct bpf_scratchpad { |
1343 | union { | |
1344 | __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; | |
1345 | u8 buff[MAX_BPF_STACK]; | |
1346 | }; | |
1347 | }; | |
1348 | ||
1349 | static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); | |
91bc4822 | 1350 | |
db58ba45 AS |
1351 | static inline int bpf_try_make_writable(struct sk_buff *skb, |
1352 | unsigned int write_len) | |
1353 | { | |
1354 | int err; | |
1355 | ||
1356 | if (!skb_cloned(skb)) | |
1357 | return 0; | |
1358 | if (skb_clone_writable(skb, write_len)) | |
1359 | return 0; | |
1360 | err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
1361 | if (!err) | |
1362 | bpf_compute_data_end(skb); | |
1363 | return err; | |
1364 | } | |
1365 | ||
91bc4822 | 1366 | static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags) |
608cd71a | 1367 | { |
21cafc1d | 1368 | struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); |
608cd71a | 1369 | struct sk_buff *skb = (struct sk_buff *) (long) r1; |
a166151c | 1370 | int offset = (int) r2; |
608cd71a AS |
1371 | void *from = (void *) (long) r3; |
1372 | unsigned int len = (unsigned int) r4; | |
608cd71a AS |
1373 | void *ptr; |
1374 | ||
8afd54c8 | 1375 | if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) |
781c53bc DB |
1376 | return -EINVAL; |
1377 | ||
608cd71a AS |
1378 | /* bpf verifier guarantees that: |
1379 | * 'from' pointer points to bpf program stack | |
1380 | * 'len' bytes of it were initialized | |
1381 | * 'len' > 0 | |
1382 | * 'skb' is a valid pointer to 'struct sk_buff' | |
1383 | * | |
1384 | * so check for invalid 'offset' and too large 'len' | |
1385 | */ | |
21cafc1d | 1386 | if (unlikely((u32) offset > 0xffff || len > sizeof(sp->buff))) |
608cd71a | 1387 | return -EFAULT; |
db58ba45 | 1388 | if (unlikely(bpf_try_make_writable(skb, offset + len))) |
608cd71a AS |
1389 | return -EFAULT; |
1390 | ||
21cafc1d | 1391 | ptr = skb_header_pointer(skb, offset, len, sp->buff); |
608cd71a AS |
1392 | if (unlikely(!ptr)) |
1393 | return -EFAULT; | |
1394 | ||
781c53bc | 1395 | if (flags & BPF_F_RECOMPUTE_CSUM) |
91bc4822 | 1396 | skb_postpull_rcsum(skb, ptr, len); |
608cd71a AS |
1397 | |
1398 | memcpy(ptr, from, len); | |
1399 | ||
21cafc1d | 1400 | if (ptr == sp->buff) |
608cd71a AS |
1401 | /* skb_store_bits cannot return -EFAULT here */ |
1402 | skb_store_bits(skb, offset, ptr, len); | |
1403 | ||
781c53bc | 1404 | if (flags & BPF_F_RECOMPUTE_CSUM) |
f8ffad69 | 1405 | skb_postpush_rcsum(skb, ptr, len); |
8afd54c8 DB |
1406 | if (flags & BPF_F_INVALIDATE_HASH) |
1407 | skb_clear_hash(skb); | |
f8ffad69 | 1408 | |
608cd71a AS |
1409 | return 0; |
1410 | } | |
1411 | ||
577c50aa | 1412 | static const struct bpf_func_proto bpf_skb_store_bytes_proto = { |
608cd71a AS |
1413 | .func = bpf_skb_store_bytes, |
1414 | .gpl_only = false, | |
1415 | .ret_type = RET_INTEGER, | |
1416 | .arg1_type = ARG_PTR_TO_CTX, | |
1417 | .arg2_type = ARG_ANYTHING, | |
1418 | .arg3_type = ARG_PTR_TO_STACK, | |
1419 | .arg4_type = ARG_CONST_STACK_SIZE, | |
91bc4822 AS |
1420 | .arg5_type = ARG_ANYTHING, |
1421 | }; | |
1422 | ||
05c74e5e DB |
1423 | static u64 bpf_skb_load_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1424 | { | |
1425 | const struct sk_buff *skb = (const struct sk_buff *)(unsigned long) r1; | |
1426 | int offset = (int) r2; | |
1427 | void *to = (void *)(unsigned long) r3; | |
1428 | unsigned int len = (unsigned int) r4; | |
1429 | void *ptr; | |
1430 | ||
074f528e DB |
1431 | if (unlikely((u32) offset > 0xffff)) |
1432 | goto err_clear; | |
05c74e5e DB |
1433 | |
1434 | ptr = skb_header_pointer(skb, offset, len, to); | |
1435 | if (unlikely(!ptr)) | |
074f528e | 1436 | goto err_clear; |
05c74e5e DB |
1437 | if (ptr != to) |
1438 | memcpy(to, ptr, len); | |
1439 | ||
1440 | return 0; | |
074f528e DB |
1441 | err_clear: |
1442 | memset(to, 0, len); | |
1443 | return -EFAULT; | |
05c74e5e DB |
1444 | } |
1445 | ||
577c50aa | 1446 | static const struct bpf_func_proto bpf_skb_load_bytes_proto = { |
05c74e5e DB |
1447 | .func = bpf_skb_load_bytes, |
1448 | .gpl_only = false, | |
1449 | .ret_type = RET_INTEGER, | |
1450 | .arg1_type = ARG_PTR_TO_CTX, | |
1451 | .arg2_type = ARG_ANYTHING, | |
074f528e | 1452 | .arg3_type = ARG_PTR_TO_RAW_STACK, |
05c74e5e DB |
1453 | .arg4_type = ARG_CONST_STACK_SIZE, |
1454 | }; | |
1455 | ||
a166151c | 1456 | static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags) |
91bc4822 AS |
1457 | { |
1458 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
a166151c | 1459 | int offset = (int) r2; |
91bc4822 AS |
1460 | __sum16 sum, *ptr; |
1461 | ||
781c53bc DB |
1462 | if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) |
1463 | return -EINVAL; | |
a166151c | 1464 | if (unlikely((u32) offset > 0xffff)) |
91bc4822 | 1465 | return -EFAULT; |
db58ba45 | 1466 | if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum)))) |
91bc4822 AS |
1467 | return -EFAULT; |
1468 | ||
1469 | ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum); | |
1470 | if (unlikely(!ptr)) | |
1471 | return -EFAULT; | |
1472 | ||
781c53bc | 1473 | switch (flags & BPF_F_HDR_FIELD_MASK) { |
8050c0f0 DB |
1474 | case 0: |
1475 | if (unlikely(from != 0)) | |
1476 | return -EINVAL; | |
1477 | ||
1478 | csum_replace_by_diff(ptr, to); | |
1479 | break; | |
91bc4822 AS |
1480 | case 2: |
1481 | csum_replace2(ptr, from, to); | |
1482 | break; | |
1483 | case 4: | |
1484 | csum_replace4(ptr, from, to); | |
1485 | break; | |
1486 | default: | |
1487 | return -EINVAL; | |
1488 | } | |
1489 | ||
1490 | if (ptr == &sum) | |
1491 | /* skb_store_bits guaranteed to not return -EFAULT here */ | |
1492 | skb_store_bits(skb, offset, ptr, sizeof(sum)); | |
1493 | ||
1494 | return 0; | |
1495 | } | |
1496 | ||
577c50aa | 1497 | static const struct bpf_func_proto bpf_l3_csum_replace_proto = { |
91bc4822 AS |
1498 | .func = bpf_l3_csum_replace, |
1499 | .gpl_only = false, | |
1500 | .ret_type = RET_INTEGER, | |
1501 | .arg1_type = ARG_PTR_TO_CTX, | |
1502 | .arg2_type = ARG_ANYTHING, | |
1503 | .arg3_type = ARG_ANYTHING, | |
1504 | .arg4_type = ARG_ANYTHING, | |
1505 | .arg5_type = ARG_ANYTHING, | |
1506 | }; | |
1507 | ||
a166151c | 1508 | static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags) |
91bc4822 AS |
1509 | { |
1510 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
781c53bc | 1511 | bool is_pseudo = flags & BPF_F_PSEUDO_HDR; |
2f72959a | 1512 | bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; |
a166151c | 1513 | int offset = (int) r2; |
91bc4822 AS |
1514 | __sum16 sum, *ptr; |
1515 | ||
2f72959a DB |
1516 | if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR | |
1517 | BPF_F_HDR_FIELD_MASK))) | |
781c53bc | 1518 | return -EINVAL; |
a166151c | 1519 | if (unlikely((u32) offset > 0xffff)) |
91bc4822 | 1520 | return -EFAULT; |
db58ba45 | 1521 | if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum)))) |
91bc4822 AS |
1522 | return -EFAULT; |
1523 | ||
1524 | ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum); | |
1525 | if (unlikely(!ptr)) | |
1526 | return -EFAULT; | |
2f72959a DB |
1527 | if (is_mmzero && !*ptr) |
1528 | return 0; | |
91bc4822 | 1529 | |
781c53bc | 1530 | switch (flags & BPF_F_HDR_FIELD_MASK) { |
7d672345 DB |
1531 | case 0: |
1532 | if (unlikely(from != 0)) | |
1533 | return -EINVAL; | |
1534 | ||
1535 | inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); | |
1536 | break; | |
91bc4822 AS |
1537 | case 2: |
1538 | inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); | |
1539 | break; | |
1540 | case 4: | |
1541 | inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); | |
1542 | break; | |
1543 | default: | |
1544 | return -EINVAL; | |
1545 | } | |
1546 | ||
2f72959a DB |
1547 | if (is_mmzero && !*ptr) |
1548 | *ptr = CSUM_MANGLED_0; | |
91bc4822 AS |
1549 | if (ptr == &sum) |
1550 | /* skb_store_bits guaranteed to not return -EFAULT here */ | |
1551 | skb_store_bits(skb, offset, ptr, sizeof(sum)); | |
1552 | ||
1553 | return 0; | |
1554 | } | |
1555 | ||
577c50aa | 1556 | static const struct bpf_func_proto bpf_l4_csum_replace_proto = { |
91bc4822 AS |
1557 | .func = bpf_l4_csum_replace, |
1558 | .gpl_only = false, | |
1559 | .ret_type = RET_INTEGER, | |
1560 | .arg1_type = ARG_PTR_TO_CTX, | |
1561 | .arg2_type = ARG_ANYTHING, | |
1562 | .arg3_type = ARG_ANYTHING, | |
1563 | .arg4_type = ARG_ANYTHING, | |
1564 | .arg5_type = ARG_ANYTHING, | |
608cd71a AS |
1565 | }; |
1566 | ||
7d672345 DB |
1567 | static u64 bpf_csum_diff(u64 r1, u64 from_size, u64 r3, u64 to_size, u64 seed) |
1568 | { | |
21cafc1d | 1569 | struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); |
7d672345 DB |
1570 | u64 diff_size = from_size + to_size; |
1571 | __be32 *from = (__be32 *) (long) r1; | |
1572 | __be32 *to = (__be32 *) (long) r3; | |
1573 | int i, j = 0; | |
1574 | ||
1575 | /* This is quite flexible, some examples: | |
1576 | * | |
1577 | * from_size == 0, to_size > 0, seed := csum --> pushing data | |
1578 | * from_size > 0, to_size == 0, seed := csum --> pulling data | |
1579 | * from_size > 0, to_size > 0, seed := 0 --> diffing data | |
1580 | * | |
1581 | * Even for diffing, from_size and to_size don't need to be equal. | |
1582 | */ | |
1583 | if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || | |
1584 | diff_size > sizeof(sp->diff))) | |
1585 | return -EINVAL; | |
1586 | ||
1587 | for (i = 0; i < from_size / sizeof(__be32); i++, j++) | |
1588 | sp->diff[j] = ~from[i]; | |
1589 | for (i = 0; i < to_size / sizeof(__be32); i++, j++) | |
1590 | sp->diff[j] = to[i]; | |
1591 | ||
1592 | return csum_partial(sp->diff, diff_size, seed); | |
1593 | } | |
1594 | ||
577c50aa | 1595 | static const struct bpf_func_proto bpf_csum_diff_proto = { |
7d672345 DB |
1596 | .func = bpf_csum_diff, |
1597 | .gpl_only = false, | |
1598 | .ret_type = RET_INTEGER, | |
1599 | .arg1_type = ARG_PTR_TO_STACK, | |
1600 | .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO, | |
1601 | .arg3_type = ARG_PTR_TO_STACK, | |
1602 | .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO, | |
1603 | .arg5_type = ARG_ANYTHING, | |
1604 | }; | |
1605 | ||
3896d655 AS |
1606 | static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5) |
1607 | { | |
1608 | struct sk_buff *skb = (struct sk_buff *) (long) r1, *skb2; | |
1609 | struct net_device *dev; | |
1610 | ||
781c53bc DB |
1611 | if (unlikely(flags & ~(BPF_F_INGRESS))) |
1612 | return -EINVAL; | |
1613 | ||
3896d655 AS |
1614 | dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); |
1615 | if (unlikely(!dev)) | |
1616 | return -EINVAL; | |
1617 | ||
3896d655 AS |
1618 | skb2 = skb_clone(skb, GFP_ATOMIC); |
1619 | if (unlikely(!skb2)) | |
1620 | return -ENOMEM; | |
1621 | ||
781c53bc | 1622 | if (flags & BPF_F_INGRESS) { |
f8ffad69 DB |
1623 | if (skb_at_tc_ingress(skb2)) |
1624 | skb_postpush_rcsum(skb2, skb_mac_header(skb2), | |
1625 | skb2->mac_len); | |
3896d655 | 1626 | return dev_forward_skb(dev, skb2); |
f8ffad69 | 1627 | } |
3896d655 AS |
1628 | |
1629 | skb2->dev = dev; | |
1630 | return dev_queue_xmit(skb2); | |
1631 | } | |
1632 | ||
577c50aa | 1633 | static const struct bpf_func_proto bpf_clone_redirect_proto = { |
3896d655 AS |
1634 | .func = bpf_clone_redirect, |
1635 | .gpl_only = false, | |
1636 | .ret_type = RET_INTEGER, | |
1637 | .arg1_type = ARG_PTR_TO_CTX, | |
1638 | .arg2_type = ARG_ANYTHING, | |
1639 | .arg3_type = ARG_ANYTHING, | |
1640 | }; | |
1641 | ||
27b29f63 AS |
1642 | struct redirect_info { |
1643 | u32 ifindex; | |
1644 | u32 flags; | |
1645 | }; | |
1646 | ||
1647 | static DEFINE_PER_CPU(struct redirect_info, redirect_info); | |
781c53bc | 1648 | |
27b29f63 AS |
1649 | static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5) |
1650 | { | |
1651 | struct redirect_info *ri = this_cpu_ptr(&redirect_info); | |
1652 | ||
781c53bc DB |
1653 | if (unlikely(flags & ~(BPF_F_INGRESS))) |
1654 | return TC_ACT_SHOT; | |
1655 | ||
27b29f63 AS |
1656 | ri->ifindex = ifindex; |
1657 | ri->flags = flags; | |
781c53bc | 1658 | |
27b29f63 AS |
1659 | return TC_ACT_REDIRECT; |
1660 | } | |
1661 | ||
1662 | int skb_do_redirect(struct sk_buff *skb) | |
1663 | { | |
1664 | struct redirect_info *ri = this_cpu_ptr(&redirect_info); | |
1665 | struct net_device *dev; | |
1666 | ||
1667 | dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex); | |
1668 | ri->ifindex = 0; | |
1669 | if (unlikely(!dev)) { | |
1670 | kfree_skb(skb); | |
1671 | return -EINVAL; | |
1672 | } | |
1673 | ||
781c53bc | 1674 | if (ri->flags & BPF_F_INGRESS) { |
f8ffad69 DB |
1675 | if (skb_at_tc_ingress(skb)) |
1676 | skb_postpush_rcsum(skb, skb_mac_header(skb), | |
1677 | skb->mac_len); | |
27b29f63 | 1678 | return dev_forward_skb(dev, skb); |
f8ffad69 | 1679 | } |
27b29f63 AS |
1680 | |
1681 | skb->dev = dev; | |
1682 | return dev_queue_xmit(skb); | |
1683 | } | |
1684 | ||
577c50aa | 1685 | static const struct bpf_func_proto bpf_redirect_proto = { |
27b29f63 AS |
1686 | .func = bpf_redirect, |
1687 | .gpl_only = false, | |
1688 | .ret_type = RET_INTEGER, | |
1689 | .arg1_type = ARG_ANYTHING, | |
1690 | .arg2_type = ARG_ANYTHING, | |
1691 | }; | |
1692 | ||
8d20aabe DB |
1693 | static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1694 | { | |
1695 | return task_get_classid((struct sk_buff *) (unsigned long) r1); | |
1696 | } | |
1697 | ||
1698 | static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { | |
1699 | .func = bpf_get_cgroup_classid, | |
1700 | .gpl_only = false, | |
1701 | .ret_type = RET_INTEGER, | |
1702 | .arg1_type = ARG_PTR_TO_CTX, | |
1703 | }; | |
1704 | ||
c46646d0 DB |
1705 | static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1706 | { | |
808c1b69 | 1707 | return dst_tclassid((struct sk_buff *) (unsigned long) r1); |
c46646d0 DB |
1708 | } |
1709 | ||
1710 | static const struct bpf_func_proto bpf_get_route_realm_proto = { | |
1711 | .func = bpf_get_route_realm, | |
1712 | .gpl_only = false, | |
1713 | .ret_type = RET_INTEGER, | |
1714 | .arg1_type = ARG_PTR_TO_CTX, | |
1715 | }; | |
1716 | ||
4e10df9a AS |
1717 | static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5) |
1718 | { | |
1719 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1720 | __be16 vlan_proto = (__force __be16) r2; | |
db58ba45 | 1721 | int ret; |
4e10df9a AS |
1722 | |
1723 | if (unlikely(vlan_proto != htons(ETH_P_8021Q) && | |
1724 | vlan_proto != htons(ETH_P_8021AD))) | |
1725 | vlan_proto = htons(ETH_P_8021Q); | |
1726 | ||
db58ba45 AS |
1727 | ret = skb_vlan_push(skb, vlan_proto, vlan_tci); |
1728 | bpf_compute_data_end(skb); | |
1729 | return ret; | |
4e10df9a AS |
1730 | } |
1731 | ||
1732 | const struct bpf_func_proto bpf_skb_vlan_push_proto = { | |
1733 | .func = bpf_skb_vlan_push, | |
1734 | .gpl_only = false, | |
1735 | .ret_type = RET_INTEGER, | |
1736 | .arg1_type = ARG_PTR_TO_CTX, | |
1737 | .arg2_type = ARG_ANYTHING, | |
1738 | .arg3_type = ARG_ANYTHING, | |
1739 | }; | |
4d9c5c53 | 1740 | EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto); |
4e10df9a AS |
1741 | |
1742 | static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
1743 | { | |
1744 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
db58ba45 | 1745 | int ret; |
4e10df9a | 1746 | |
db58ba45 AS |
1747 | ret = skb_vlan_pop(skb); |
1748 | bpf_compute_data_end(skb); | |
1749 | return ret; | |
4e10df9a AS |
1750 | } |
1751 | ||
1752 | const struct bpf_func_proto bpf_skb_vlan_pop_proto = { | |
1753 | .func = bpf_skb_vlan_pop, | |
1754 | .gpl_only = false, | |
1755 | .ret_type = RET_INTEGER, | |
1756 | .arg1_type = ARG_PTR_TO_CTX, | |
1757 | }; | |
4d9c5c53 | 1758 | EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto); |
4e10df9a AS |
1759 | |
1760 | bool bpf_helper_changes_skb_data(void *func) | |
1761 | { | |
1762 | if (func == bpf_skb_vlan_push) | |
1763 | return true; | |
1764 | if (func == bpf_skb_vlan_pop) | |
1765 | return true; | |
3697649f DB |
1766 | if (func == bpf_skb_store_bytes) |
1767 | return true; | |
1768 | if (func == bpf_l3_csum_replace) | |
1769 | return true; | |
1770 | if (func == bpf_l4_csum_replace) | |
1771 | return true; | |
1772 | ||
4e10df9a AS |
1773 | return false; |
1774 | } | |
1775 | ||
c6c33454 DB |
1776 | static unsigned short bpf_tunnel_key_af(u64 flags) |
1777 | { | |
1778 | return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; | |
1779 | } | |
1780 | ||
d3aa45ce AS |
1781 | static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5) |
1782 | { | |
1783 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1784 | struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2; | |
c6c33454 DB |
1785 | const struct ip_tunnel_info *info = skb_tunnel_info(skb); |
1786 | u8 compat[sizeof(struct bpf_tunnel_key)]; | |
074f528e DB |
1787 | void *to_orig = to; |
1788 | int err; | |
d3aa45ce | 1789 | |
074f528e DB |
1790 | if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { |
1791 | err = -EINVAL; | |
1792 | goto err_clear; | |
1793 | } | |
1794 | if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { | |
1795 | err = -EPROTO; | |
1796 | goto err_clear; | |
1797 | } | |
c6c33454 | 1798 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) { |
074f528e | 1799 | err = -EINVAL; |
c6c33454 | 1800 | switch (size) { |
4018ab18 | 1801 | case offsetof(struct bpf_tunnel_key, tunnel_label): |
c0e760c9 | 1802 | case offsetof(struct bpf_tunnel_key, tunnel_ext): |
4018ab18 | 1803 | goto set_compat; |
c6c33454 DB |
1804 | case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): |
1805 | /* Fixup deprecated structure layouts here, so we have | |
1806 | * a common path later on. | |
1807 | */ | |
1808 | if (ip_tunnel_info_af(info) != AF_INET) | |
074f528e | 1809 | goto err_clear; |
4018ab18 | 1810 | set_compat: |
c6c33454 DB |
1811 | to = (struct bpf_tunnel_key *)compat; |
1812 | break; | |
1813 | default: | |
074f528e | 1814 | goto err_clear; |
c6c33454 DB |
1815 | } |
1816 | } | |
d3aa45ce AS |
1817 | |
1818 | to->tunnel_id = be64_to_cpu(info->key.tun_id); | |
c6c33454 DB |
1819 | to->tunnel_tos = info->key.tos; |
1820 | to->tunnel_ttl = info->key.ttl; | |
1821 | ||
4018ab18 | 1822 | if (flags & BPF_F_TUNINFO_IPV6) { |
c6c33454 DB |
1823 | memcpy(to->remote_ipv6, &info->key.u.ipv6.src, |
1824 | sizeof(to->remote_ipv6)); | |
4018ab18 DB |
1825 | to->tunnel_label = be32_to_cpu(info->key.label); |
1826 | } else { | |
c6c33454 | 1827 | to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); |
4018ab18 | 1828 | } |
c6c33454 DB |
1829 | |
1830 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) | |
074f528e | 1831 | memcpy(to_orig, to, size); |
d3aa45ce AS |
1832 | |
1833 | return 0; | |
074f528e DB |
1834 | err_clear: |
1835 | memset(to_orig, 0, size); | |
1836 | return err; | |
d3aa45ce AS |
1837 | } |
1838 | ||
577c50aa | 1839 | static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { |
d3aa45ce AS |
1840 | .func = bpf_skb_get_tunnel_key, |
1841 | .gpl_only = false, | |
1842 | .ret_type = RET_INTEGER, | |
1843 | .arg1_type = ARG_PTR_TO_CTX, | |
074f528e | 1844 | .arg2_type = ARG_PTR_TO_RAW_STACK, |
d3aa45ce AS |
1845 | .arg3_type = ARG_CONST_STACK_SIZE, |
1846 | .arg4_type = ARG_ANYTHING, | |
1847 | }; | |
1848 | ||
14ca0751 DB |
1849 | static u64 bpf_skb_get_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5) |
1850 | { | |
1851 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1852 | u8 *to = (u8 *) (long) r2; | |
1853 | const struct ip_tunnel_info *info = skb_tunnel_info(skb); | |
074f528e | 1854 | int err; |
14ca0751 DB |
1855 | |
1856 | if (unlikely(!info || | |
074f528e DB |
1857 | !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { |
1858 | err = -ENOENT; | |
1859 | goto err_clear; | |
1860 | } | |
1861 | if (unlikely(size < info->options_len)) { | |
1862 | err = -ENOMEM; | |
1863 | goto err_clear; | |
1864 | } | |
14ca0751 DB |
1865 | |
1866 | ip_tunnel_info_opts_get(to, info); | |
074f528e DB |
1867 | if (size > info->options_len) |
1868 | memset(to + info->options_len, 0, size - info->options_len); | |
14ca0751 DB |
1869 | |
1870 | return info->options_len; | |
074f528e DB |
1871 | err_clear: |
1872 | memset(to, 0, size); | |
1873 | return err; | |
14ca0751 DB |
1874 | } |
1875 | ||
1876 | static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { | |
1877 | .func = bpf_skb_get_tunnel_opt, | |
1878 | .gpl_only = false, | |
1879 | .ret_type = RET_INTEGER, | |
1880 | .arg1_type = ARG_PTR_TO_CTX, | |
074f528e | 1881 | .arg2_type = ARG_PTR_TO_RAW_STACK, |
14ca0751 DB |
1882 | .arg3_type = ARG_CONST_STACK_SIZE, |
1883 | }; | |
1884 | ||
d3aa45ce AS |
1885 | static struct metadata_dst __percpu *md_dst; |
1886 | ||
1887 | static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5) | |
1888 | { | |
1889 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1890 | struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2; | |
1891 | struct metadata_dst *md = this_cpu_ptr(md_dst); | |
c6c33454 | 1892 | u8 compat[sizeof(struct bpf_tunnel_key)]; |
d3aa45ce AS |
1893 | struct ip_tunnel_info *info; |
1894 | ||
22080870 DB |
1895 | if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | |
1896 | BPF_F_DONT_FRAGMENT))) | |
d3aa45ce | 1897 | return -EINVAL; |
c6c33454 DB |
1898 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) { |
1899 | switch (size) { | |
4018ab18 | 1900 | case offsetof(struct bpf_tunnel_key, tunnel_label): |
c0e760c9 | 1901 | case offsetof(struct bpf_tunnel_key, tunnel_ext): |
c6c33454 DB |
1902 | case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): |
1903 | /* Fixup deprecated structure layouts here, so we have | |
1904 | * a common path later on. | |
1905 | */ | |
1906 | memcpy(compat, from, size); | |
1907 | memset(compat + size, 0, sizeof(compat) - size); | |
1908 | from = (struct bpf_tunnel_key *)compat; | |
1909 | break; | |
1910 | default: | |
1911 | return -EINVAL; | |
1912 | } | |
1913 | } | |
c0e760c9 DB |
1914 | if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || |
1915 | from->tunnel_ext)) | |
4018ab18 | 1916 | return -EINVAL; |
d3aa45ce AS |
1917 | |
1918 | skb_dst_drop(skb); | |
1919 | dst_hold((struct dst_entry *) md); | |
1920 | skb_dst_set(skb, (struct dst_entry *) md); | |
1921 | ||
1922 | info = &md->u.tun_info; | |
1923 | info->mode = IP_TUNNEL_INFO_TX; | |
c6c33454 | 1924 | |
db3c6139 | 1925 | info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; |
22080870 DB |
1926 | if (flags & BPF_F_DONT_FRAGMENT) |
1927 | info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; | |
1928 | ||
d3aa45ce | 1929 | info->key.tun_id = cpu_to_be64(from->tunnel_id); |
c6c33454 DB |
1930 | info->key.tos = from->tunnel_tos; |
1931 | info->key.ttl = from->tunnel_ttl; | |
1932 | ||
1933 | if (flags & BPF_F_TUNINFO_IPV6) { | |
1934 | info->mode |= IP_TUNNEL_INFO_IPV6; | |
1935 | memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, | |
1936 | sizeof(from->remote_ipv6)); | |
4018ab18 DB |
1937 | info->key.label = cpu_to_be32(from->tunnel_label) & |
1938 | IPV6_FLOWLABEL_MASK; | |
c6c33454 DB |
1939 | } else { |
1940 | info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); | |
2da897e5 DB |
1941 | if (flags & BPF_F_ZERO_CSUM_TX) |
1942 | info->key.tun_flags &= ~TUNNEL_CSUM; | |
c6c33454 | 1943 | } |
d3aa45ce AS |
1944 | |
1945 | return 0; | |
1946 | } | |
1947 | ||
577c50aa | 1948 | static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { |
d3aa45ce AS |
1949 | .func = bpf_skb_set_tunnel_key, |
1950 | .gpl_only = false, | |
1951 | .ret_type = RET_INTEGER, | |
1952 | .arg1_type = ARG_PTR_TO_CTX, | |
1953 | .arg2_type = ARG_PTR_TO_STACK, | |
1954 | .arg3_type = ARG_CONST_STACK_SIZE, | |
1955 | .arg4_type = ARG_ANYTHING, | |
1956 | }; | |
1957 | ||
14ca0751 DB |
1958 | static u64 bpf_skb_set_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5) |
1959 | { | |
1960 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1961 | u8 *from = (u8 *) (long) r2; | |
1962 | struct ip_tunnel_info *info = skb_tunnel_info(skb); | |
1963 | const struct metadata_dst *md = this_cpu_ptr(md_dst); | |
1964 | ||
1965 | if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) | |
1966 | return -EINVAL; | |
fca5fdf6 | 1967 | if (unlikely(size > IP_TUNNEL_OPTS_MAX)) |
14ca0751 DB |
1968 | return -ENOMEM; |
1969 | ||
1970 | ip_tunnel_info_opts_set(info, from, size); | |
1971 | ||
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { | |
1976 | .func = bpf_skb_set_tunnel_opt, | |
1977 | .gpl_only = false, | |
1978 | .ret_type = RET_INTEGER, | |
1979 | .arg1_type = ARG_PTR_TO_CTX, | |
1980 | .arg2_type = ARG_PTR_TO_STACK, | |
1981 | .arg3_type = ARG_CONST_STACK_SIZE, | |
1982 | }; | |
1983 | ||
1984 | static const struct bpf_func_proto * | |
1985 | bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) | |
d3aa45ce AS |
1986 | { |
1987 | if (!md_dst) { | |
14ca0751 DB |
1988 | /* Race is not possible, since it's called from verifier |
1989 | * that is holding verifier mutex. | |
d3aa45ce | 1990 | */ |
fca5fdf6 | 1991 | md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, |
14ca0751 | 1992 | GFP_KERNEL); |
d3aa45ce AS |
1993 | if (!md_dst) |
1994 | return NULL; | |
1995 | } | |
14ca0751 DB |
1996 | |
1997 | switch (which) { | |
1998 | case BPF_FUNC_skb_set_tunnel_key: | |
1999 | return &bpf_skb_set_tunnel_key_proto; | |
2000 | case BPF_FUNC_skb_set_tunnel_opt: | |
2001 | return &bpf_skb_set_tunnel_opt_proto; | |
2002 | default: | |
2003 | return NULL; | |
2004 | } | |
d3aa45ce AS |
2005 | } |
2006 | ||
d4052c4a DB |
2007 | static const struct bpf_func_proto * |
2008 | sk_filter_func_proto(enum bpf_func_id func_id) | |
89aa0758 AS |
2009 | { |
2010 | switch (func_id) { | |
2011 | case BPF_FUNC_map_lookup_elem: | |
2012 | return &bpf_map_lookup_elem_proto; | |
2013 | case BPF_FUNC_map_update_elem: | |
2014 | return &bpf_map_update_elem_proto; | |
2015 | case BPF_FUNC_map_delete_elem: | |
2016 | return &bpf_map_delete_elem_proto; | |
03e69b50 DB |
2017 | case BPF_FUNC_get_prandom_u32: |
2018 | return &bpf_get_prandom_u32_proto; | |
c04167ce DB |
2019 | case BPF_FUNC_get_smp_processor_id: |
2020 | return &bpf_get_smp_processor_id_proto; | |
04fd61ab AS |
2021 | case BPF_FUNC_tail_call: |
2022 | return &bpf_tail_call_proto; | |
17ca8cbf DB |
2023 | case BPF_FUNC_ktime_get_ns: |
2024 | return &bpf_ktime_get_ns_proto; | |
0756ea3e | 2025 | case BPF_FUNC_trace_printk: |
1be7f75d AS |
2026 | if (capable(CAP_SYS_ADMIN)) |
2027 | return bpf_get_trace_printk_proto(); | |
89aa0758 AS |
2028 | default: |
2029 | return NULL; | |
2030 | } | |
2031 | } | |
2032 | ||
608cd71a AS |
2033 | static const struct bpf_func_proto * |
2034 | tc_cls_act_func_proto(enum bpf_func_id func_id) | |
2035 | { | |
2036 | switch (func_id) { | |
2037 | case BPF_FUNC_skb_store_bytes: | |
2038 | return &bpf_skb_store_bytes_proto; | |
05c74e5e DB |
2039 | case BPF_FUNC_skb_load_bytes: |
2040 | return &bpf_skb_load_bytes_proto; | |
7d672345 DB |
2041 | case BPF_FUNC_csum_diff: |
2042 | return &bpf_csum_diff_proto; | |
91bc4822 AS |
2043 | case BPF_FUNC_l3_csum_replace: |
2044 | return &bpf_l3_csum_replace_proto; | |
2045 | case BPF_FUNC_l4_csum_replace: | |
2046 | return &bpf_l4_csum_replace_proto; | |
3896d655 AS |
2047 | case BPF_FUNC_clone_redirect: |
2048 | return &bpf_clone_redirect_proto; | |
8d20aabe DB |
2049 | case BPF_FUNC_get_cgroup_classid: |
2050 | return &bpf_get_cgroup_classid_proto; | |
4e10df9a AS |
2051 | case BPF_FUNC_skb_vlan_push: |
2052 | return &bpf_skb_vlan_push_proto; | |
2053 | case BPF_FUNC_skb_vlan_pop: | |
2054 | return &bpf_skb_vlan_pop_proto; | |
d3aa45ce AS |
2055 | case BPF_FUNC_skb_get_tunnel_key: |
2056 | return &bpf_skb_get_tunnel_key_proto; | |
2057 | case BPF_FUNC_skb_set_tunnel_key: | |
14ca0751 DB |
2058 | return bpf_get_skb_set_tunnel_proto(func_id); |
2059 | case BPF_FUNC_skb_get_tunnel_opt: | |
2060 | return &bpf_skb_get_tunnel_opt_proto; | |
2061 | case BPF_FUNC_skb_set_tunnel_opt: | |
2062 | return bpf_get_skb_set_tunnel_proto(func_id); | |
27b29f63 AS |
2063 | case BPF_FUNC_redirect: |
2064 | return &bpf_redirect_proto; | |
c46646d0 DB |
2065 | case BPF_FUNC_get_route_realm: |
2066 | return &bpf_get_route_realm_proto; | |
bd570ff9 DB |
2067 | case BPF_FUNC_perf_event_output: |
2068 | return bpf_get_event_output_proto(); | |
608cd71a AS |
2069 | default: |
2070 | return sk_filter_func_proto(func_id); | |
2071 | } | |
2072 | } | |
2073 | ||
d691f9e8 | 2074 | static bool __is_valid_access(int off, int size, enum bpf_access_type type) |
89aa0758 | 2075 | { |
9bac3d6d AS |
2076 | if (off < 0 || off >= sizeof(struct __sk_buff)) |
2077 | return false; | |
4936e352 | 2078 | /* The verifier guarantees that size > 0. */ |
9bac3d6d AS |
2079 | if (off % size != 0) |
2080 | return false; | |
4936e352 | 2081 | if (size != sizeof(__u32)) |
9bac3d6d AS |
2082 | return false; |
2083 | ||
2084 | return true; | |
2085 | } | |
2086 | ||
d691f9e8 | 2087 | static bool sk_filter_is_valid_access(int off, int size, |
19de99f7 AS |
2088 | enum bpf_access_type type, |
2089 | enum bpf_reg_type *reg_type) | |
d691f9e8 | 2090 | { |
db58ba45 AS |
2091 | switch (off) { |
2092 | case offsetof(struct __sk_buff, tc_classid): | |
2093 | case offsetof(struct __sk_buff, data): | |
2094 | case offsetof(struct __sk_buff, data_end): | |
045efa82 | 2095 | return false; |
db58ba45 | 2096 | } |
045efa82 | 2097 | |
d691f9e8 AS |
2098 | if (type == BPF_WRITE) { |
2099 | switch (off) { | |
2100 | case offsetof(struct __sk_buff, cb[0]) ... | |
4936e352 | 2101 | offsetof(struct __sk_buff, cb[4]): |
d691f9e8 AS |
2102 | break; |
2103 | default: | |
2104 | return false; | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | return __is_valid_access(off, size, type); | |
2109 | } | |
2110 | ||
2111 | static bool tc_cls_act_is_valid_access(int off, int size, | |
19de99f7 AS |
2112 | enum bpf_access_type type, |
2113 | enum bpf_reg_type *reg_type) | |
d691f9e8 AS |
2114 | { |
2115 | if (type == BPF_WRITE) { | |
2116 | switch (off) { | |
2117 | case offsetof(struct __sk_buff, mark): | |
2118 | case offsetof(struct __sk_buff, tc_index): | |
754f1e6a | 2119 | case offsetof(struct __sk_buff, priority): |
d691f9e8 | 2120 | case offsetof(struct __sk_buff, cb[0]) ... |
09c37a2c DB |
2121 | offsetof(struct __sk_buff, cb[4]): |
2122 | case offsetof(struct __sk_buff, tc_classid): | |
d691f9e8 AS |
2123 | break; |
2124 | default: | |
2125 | return false; | |
2126 | } | |
2127 | } | |
19de99f7 AS |
2128 | |
2129 | switch (off) { | |
2130 | case offsetof(struct __sk_buff, data): | |
2131 | *reg_type = PTR_TO_PACKET; | |
2132 | break; | |
2133 | case offsetof(struct __sk_buff, data_end): | |
2134 | *reg_type = PTR_TO_PACKET_END; | |
2135 | break; | |
2136 | } | |
2137 | ||
d691f9e8 AS |
2138 | return __is_valid_access(off, size, type); |
2139 | } | |
2140 | ||
2141 | static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg, | |
2142 | int src_reg, int ctx_off, | |
ff936a04 AS |
2143 | struct bpf_insn *insn_buf, |
2144 | struct bpf_prog *prog) | |
9bac3d6d AS |
2145 | { |
2146 | struct bpf_insn *insn = insn_buf; | |
2147 | ||
2148 | switch (ctx_off) { | |
2149 | case offsetof(struct __sk_buff, len): | |
2150 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4); | |
2151 | ||
2152 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2153 | offsetof(struct sk_buff, len)); | |
2154 | break; | |
2155 | ||
0b8c707d DB |
2156 | case offsetof(struct __sk_buff, protocol): |
2157 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); | |
2158 | ||
2159 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2160 | offsetof(struct sk_buff, protocol)); | |
2161 | break; | |
2162 | ||
27cd5452 MS |
2163 | case offsetof(struct __sk_buff, vlan_proto): |
2164 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); | |
2165 | ||
2166 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2167 | offsetof(struct sk_buff, vlan_proto)); | |
2168 | break; | |
2169 | ||
bcad5718 DB |
2170 | case offsetof(struct __sk_buff, priority): |
2171 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4); | |
2172 | ||
754f1e6a DB |
2173 | if (type == BPF_WRITE) |
2174 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, | |
2175 | offsetof(struct sk_buff, priority)); | |
2176 | else | |
2177 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2178 | offsetof(struct sk_buff, priority)); | |
bcad5718 DB |
2179 | break; |
2180 | ||
37e82c2f AS |
2181 | case offsetof(struct __sk_buff, ingress_ifindex): |
2182 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4); | |
2183 | ||
2184 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2185 | offsetof(struct sk_buff, skb_iif)); | |
2186 | break; | |
2187 | ||
2188 | case offsetof(struct __sk_buff, ifindex): | |
2189 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); | |
2190 | ||
2191 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)), | |
2192 | dst_reg, src_reg, | |
2193 | offsetof(struct sk_buff, dev)); | |
2194 | *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1); | |
2195 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg, | |
2196 | offsetof(struct net_device, ifindex)); | |
2197 | break; | |
2198 | ||
ba7591d8 DB |
2199 | case offsetof(struct __sk_buff, hash): |
2200 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
2201 | ||
2202 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2203 | offsetof(struct sk_buff, hash)); | |
2204 | break; | |
2205 | ||
9bac3d6d | 2206 | case offsetof(struct __sk_buff, mark): |
d691f9e8 AS |
2207 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); |
2208 | ||
2209 | if (type == BPF_WRITE) | |
2210 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, | |
2211 | offsetof(struct sk_buff, mark)); | |
2212 | else | |
2213 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2214 | offsetof(struct sk_buff, mark)); | |
2215 | break; | |
9bac3d6d AS |
2216 | |
2217 | case offsetof(struct __sk_buff, pkt_type): | |
2218 | return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn); | |
2219 | ||
2220 | case offsetof(struct __sk_buff, queue_mapping): | |
2221 | return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn); | |
c2497395 | 2222 | |
c2497395 AS |
2223 | case offsetof(struct __sk_buff, vlan_present): |
2224 | return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, | |
2225 | dst_reg, src_reg, insn); | |
2226 | ||
2227 | case offsetof(struct __sk_buff, vlan_tci): | |
2228 | return convert_skb_access(SKF_AD_VLAN_TAG, | |
2229 | dst_reg, src_reg, insn); | |
d691f9e8 AS |
2230 | |
2231 | case offsetof(struct __sk_buff, cb[0]) ... | |
2232 | offsetof(struct __sk_buff, cb[4]): | |
2233 | BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20); | |
2234 | ||
ff936a04 | 2235 | prog->cb_access = 1; |
d691f9e8 AS |
2236 | ctx_off -= offsetof(struct __sk_buff, cb[0]); |
2237 | ctx_off += offsetof(struct sk_buff, cb); | |
2238 | ctx_off += offsetof(struct qdisc_skb_cb, data); | |
2239 | if (type == BPF_WRITE) | |
2240 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off); | |
2241 | else | |
2242 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off); | |
2243 | break; | |
2244 | ||
045efa82 DB |
2245 | case offsetof(struct __sk_buff, tc_classid): |
2246 | ctx_off -= offsetof(struct __sk_buff, tc_classid); | |
2247 | ctx_off += offsetof(struct sk_buff, cb); | |
2248 | ctx_off += offsetof(struct qdisc_skb_cb, tc_classid); | |
09c37a2c DB |
2249 | if (type == BPF_WRITE) |
2250 | *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off); | |
2251 | else | |
2252 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, ctx_off); | |
045efa82 DB |
2253 | break; |
2254 | ||
db58ba45 AS |
2255 | case offsetof(struct __sk_buff, data): |
2256 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, data)), | |
2257 | dst_reg, src_reg, | |
2258 | offsetof(struct sk_buff, data)); | |
2259 | break; | |
2260 | ||
2261 | case offsetof(struct __sk_buff, data_end): | |
2262 | ctx_off -= offsetof(struct __sk_buff, data_end); | |
2263 | ctx_off += offsetof(struct sk_buff, cb); | |
2264 | ctx_off += offsetof(struct bpf_skb_data_end, data_end); | |
2265 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(sizeof(void *)), | |
2266 | dst_reg, src_reg, ctx_off); | |
2267 | break; | |
2268 | ||
d691f9e8 AS |
2269 | case offsetof(struct __sk_buff, tc_index): |
2270 | #ifdef CONFIG_NET_SCHED | |
2271 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2); | |
2272 | ||
2273 | if (type == BPF_WRITE) | |
2274 | *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, | |
2275 | offsetof(struct sk_buff, tc_index)); | |
2276 | else | |
2277 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2278 | offsetof(struct sk_buff, tc_index)); | |
2279 | break; | |
2280 | #else | |
2281 | if (type == BPF_WRITE) | |
2282 | *insn++ = BPF_MOV64_REG(dst_reg, dst_reg); | |
2283 | else | |
2284 | *insn++ = BPF_MOV64_IMM(dst_reg, 0); | |
2285 | break; | |
2286 | #endif | |
9bac3d6d AS |
2287 | } |
2288 | ||
2289 | return insn - insn_buf; | |
89aa0758 AS |
2290 | } |
2291 | ||
d4052c4a | 2292 | static const struct bpf_verifier_ops sk_filter_ops = { |
4936e352 DB |
2293 | .get_func_proto = sk_filter_func_proto, |
2294 | .is_valid_access = sk_filter_is_valid_access, | |
2295 | .convert_ctx_access = bpf_net_convert_ctx_access, | |
89aa0758 AS |
2296 | }; |
2297 | ||
608cd71a | 2298 | static const struct bpf_verifier_ops tc_cls_act_ops = { |
4936e352 DB |
2299 | .get_func_proto = tc_cls_act_func_proto, |
2300 | .is_valid_access = tc_cls_act_is_valid_access, | |
2301 | .convert_ctx_access = bpf_net_convert_ctx_access, | |
608cd71a AS |
2302 | }; |
2303 | ||
d4052c4a | 2304 | static struct bpf_prog_type_list sk_filter_type __read_mostly = { |
4936e352 DB |
2305 | .ops = &sk_filter_ops, |
2306 | .type = BPF_PROG_TYPE_SOCKET_FILTER, | |
89aa0758 AS |
2307 | }; |
2308 | ||
96be4325 | 2309 | static struct bpf_prog_type_list sched_cls_type __read_mostly = { |
4936e352 DB |
2310 | .ops = &tc_cls_act_ops, |
2311 | .type = BPF_PROG_TYPE_SCHED_CLS, | |
96be4325 DB |
2312 | }; |
2313 | ||
94caee8c | 2314 | static struct bpf_prog_type_list sched_act_type __read_mostly = { |
4936e352 DB |
2315 | .ops = &tc_cls_act_ops, |
2316 | .type = BPF_PROG_TYPE_SCHED_ACT, | |
94caee8c DB |
2317 | }; |
2318 | ||
d4052c4a | 2319 | static int __init register_sk_filter_ops(void) |
89aa0758 | 2320 | { |
d4052c4a | 2321 | bpf_register_prog_type(&sk_filter_type); |
96be4325 | 2322 | bpf_register_prog_type(&sched_cls_type); |
94caee8c | 2323 | bpf_register_prog_type(&sched_act_type); |
96be4325 | 2324 | |
89aa0758 AS |
2325 | return 0; |
2326 | } | |
d4052c4a DB |
2327 | late_initcall(register_sk_filter_ops); |
2328 | ||
8ced425e | 2329 | int sk_detach_filter(struct sock *sk) |
55b33325 PE |
2330 | { |
2331 | int ret = -ENOENT; | |
2332 | struct sk_filter *filter; | |
2333 | ||
d59577b6 VB |
2334 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
2335 | return -EPERM; | |
2336 | ||
8ced425e HFS |
2337 | filter = rcu_dereference_protected(sk->sk_filter, |
2338 | lockdep_sock_is_held(sk)); | |
55b33325 | 2339 | if (filter) { |
a9b3cd7f | 2340 | RCU_INIT_POINTER(sk->sk_filter, NULL); |
46bcf14f | 2341 | sk_filter_uncharge(sk, filter); |
55b33325 PE |
2342 | ret = 0; |
2343 | } | |
a3ea269b | 2344 | |
55b33325 PE |
2345 | return ret; |
2346 | } | |
8ced425e | 2347 | EXPORT_SYMBOL_GPL(sk_detach_filter); |
a8fc9277 | 2348 | |
a3ea269b DB |
2349 | int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, |
2350 | unsigned int len) | |
a8fc9277 | 2351 | { |
a3ea269b | 2352 | struct sock_fprog_kern *fprog; |
a8fc9277 | 2353 | struct sk_filter *filter; |
a3ea269b | 2354 | int ret = 0; |
a8fc9277 PE |
2355 | |
2356 | lock_sock(sk); | |
2357 | filter = rcu_dereference_protected(sk->sk_filter, | |
8ced425e | 2358 | lockdep_sock_is_held(sk)); |
a8fc9277 PE |
2359 | if (!filter) |
2360 | goto out; | |
a3ea269b DB |
2361 | |
2362 | /* We're copying the filter that has been originally attached, | |
93d08b69 DB |
2363 | * so no conversion/decode needed anymore. eBPF programs that |
2364 | * have no original program cannot be dumped through this. | |
a3ea269b | 2365 | */ |
93d08b69 | 2366 | ret = -EACCES; |
7ae457c1 | 2367 | fprog = filter->prog->orig_prog; |
93d08b69 DB |
2368 | if (!fprog) |
2369 | goto out; | |
a3ea269b DB |
2370 | |
2371 | ret = fprog->len; | |
a8fc9277 | 2372 | if (!len) |
a3ea269b | 2373 | /* User space only enquires number of filter blocks. */ |
a8fc9277 | 2374 | goto out; |
a3ea269b | 2375 | |
a8fc9277 | 2376 | ret = -EINVAL; |
a3ea269b | 2377 | if (len < fprog->len) |
a8fc9277 PE |
2378 | goto out; |
2379 | ||
2380 | ret = -EFAULT; | |
009937e7 | 2381 | if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) |
a3ea269b | 2382 | goto out; |
a8fc9277 | 2383 | |
a3ea269b DB |
2384 | /* Instead of bytes, the API requests to return the number |
2385 | * of filter blocks. | |
2386 | */ | |
2387 | ret = fprog->len; | |
a8fc9277 PE |
2388 | out: |
2389 | release_sock(sk); | |
2390 | return ret; | |
2391 | } |