]>
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 | 55 | /** |
f4979fce | 56 | * sk_filter_trim_cap - run a packet through a socket filter |
43db6d65 SH |
57 | * @sk: sock associated with &sk_buff |
58 | * @skb: buffer to filter | |
f4979fce | 59 | * @cap: limit on how short the eBPF program may trim the packet |
43db6d65 | 60 | * |
ff936a04 AS |
61 | * Run the eBPF program and then cut skb->data to correct size returned by |
62 | * the program. If pkt_len is 0 we toss packet. If skb->len is smaller | |
43db6d65 | 63 | * than pkt_len we keep whole skb->data. This is the socket level |
ff936a04 | 64 | * wrapper to BPF_PROG_RUN. It returns 0 if the packet should |
43db6d65 SH |
65 | * be accepted or -EPERM if the packet should be tossed. |
66 | * | |
67 | */ | |
f4979fce | 68 | int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) |
43db6d65 SH |
69 | { |
70 | int err; | |
71 | struct sk_filter *filter; | |
72 | ||
c93bdd0e MG |
73 | /* |
74 | * If the skb was allocated from pfmemalloc reserves, only | |
75 | * allow SOCK_MEMALLOC sockets to use it as this socket is | |
76 | * helping free memory | |
77 | */ | |
78 | if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) | |
79 | return -ENOMEM; | |
80 | ||
43db6d65 SH |
81 | err = security_sock_rcv_skb(sk, skb); |
82 | if (err) | |
83 | return err; | |
84 | ||
80f8f102 ED |
85 | rcu_read_lock(); |
86 | filter = rcu_dereference(sk->sk_filter); | |
43db6d65 | 87 | if (filter) { |
ff936a04 | 88 | unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb); |
f4979fce | 89 | err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; |
43db6d65 | 90 | } |
80f8f102 | 91 | rcu_read_unlock(); |
43db6d65 SH |
92 | |
93 | return err; | |
94 | } | |
f4979fce | 95 | EXPORT_SYMBOL(sk_filter_trim_cap); |
43db6d65 | 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 | ||
80b48c44 DB |
153 | static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = { |
154 | .func = __get_raw_cpu_id, | |
155 | .gpl_only = false, | |
156 | .ret_type = RET_INTEGER, | |
157 | }; | |
158 | ||
9bac3d6d AS |
159 | static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, |
160 | struct bpf_insn *insn_buf) | |
161 | { | |
162 | struct bpf_insn *insn = insn_buf; | |
163 | ||
164 | switch (skb_field) { | |
165 | case SKF_AD_MARK: | |
166 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); | |
167 | ||
168 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
169 | offsetof(struct sk_buff, mark)); | |
170 | break; | |
171 | ||
172 | case SKF_AD_PKTTYPE: | |
173 | *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); | |
174 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); | |
175 | #ifdef __BIG_ENDIAN_BITFIELD | |
176 | *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); | |
177 | #endif | |
178 | break; | |
179 | ||
180 | case SKF_AD_QUEUE: | |
181 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); | |
182 | ||
183 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
184 | offsetof(struct sk_buff, queue_mapping)); | |
185 | break; | |
c2497395 | 186 | |
c2497395 AS |
187 | case SKF_AD_VLAN_TAG: |
188 | case SKF_AD_VLAN_TAG_PRESENT: | |
189 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); | |
190 | BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); | |
191 | ||
192 | /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ | |
193 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
194 | offsetof(struct sk_buff, vlan_tci)); | |
195 | if (skb_field == SKF_AD_VLAN_TAG) { | |
196 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, | |
197 | ~VLAN_TAG_PRESENT); | |
198 | } else { | |
199 | /* dst_reg >>= 12 */ | |
200 | *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12); | |
201 | /* dst_reg &= 1 */ | |
202 | *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); | |
203 | } | |
204 | break; | |
9bac3d6d AS |
205 | } |
206 | ||
207 | return insn - insn_buf; | |
208 | } | |
209 | ||
bd4cf0ed | 210 | static bool convert_bpf_extensions(struct sock_filter *fp, |
2695fb55 | 211 | struct bpf_insn **insnp) |
bd4cf0ed | 212 | { |
2695fb55 | 213 | struct bpf_insn *insn = *insnp; |
9bac3d6d | 214 | u32 cnt; |
bd4cf0ed AS |
215 | |
216 | switch (fp->k) { | |
217 | case SKF_AD_OFF + SKF_AD_PROTOCOL: | |
0b8c707d DB |
218 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); |
219 | ||
220 | /* A = *(u16 *) (CTX + offsetof(protocol)) */ | |
221 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, | |
222 | offsetof(struct sk_buff, protocol)); | |
223 | /* A = ntohs(A) [emitting a nop or swap16] */ | |
224 | *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); | |
bd4cf0ed AS |
225 | break; |
226 | ||
227 | case SKF_AD_OFF + SKF_AD_PKTTYPE: | |
9bac3d6d AS |
228 | cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); |
229 | insn += cnt - 1; | |
bd4cf0ed AS |
230 | break; |
231 | ||
232 | case SKF_AD_OFF + SKF_AD_IFINDEX: | |
233 | case SKF_AD_OFF + SKF_AD_HATYPE: | |
bd4cf0ed AS |
234 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); |
235 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); | |
f8f6d679 DB |
236 | BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0); |
237 | ||
238 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)), | |
239 | BPF_REG_TMP, BPF_REG_CTX, | |
240 | offsetof(struct sk_buff, dev)); | |
241 | /* if (tmp != 0) goto pc + 1 */ | |
242 | *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); | |
243 | *insn++ = BPF_EXIT_INSN(); | |
244 | if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) | |
245 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, | |
246 | offsetof(struct net_device, ifindex)); | |
247 | else | |
248 | *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, | |
249 | offsetof(struct net_device, type)); | |
bd4cf0ed AS |
250 | break; |
251 | ||
252 | case SKF_AD_OFF + SKF_AD_MARK: | |
9bac3d6d AS |
253 | cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); |
254 | insn += cnt - 1; | |
bd4cf0ed AS |
255 | break; |
256 | ||
257 | case SKF_AD_OFF + SKF_AD_RXHASH: | |
258 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
259 | ||
9739eef1 AS |
260 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, |
261 | offsetof(struct sk_buff, hash)); | |
bd4cf0ed AS |
262 | break; |
263 | ||
264 | case SKF_AD_OFF + SKF_AD_QUEUE: | |
9bac3d6d AS |
265 | cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); |
266 | insn += cnt - 1; | |
bd4cf0ed AS |
267 | break; |
268 | ||
269 | case SKF_AD_OFF + SKF_AD_VLAN_TAG: | |
c2497395 AS |
270 | cnt = convert_skb_access(SKF_AD_VLAN_TAG, |
271 | BPF_REG_A, BPF_REG_CTX, insn); | |
272 | insn += cnt - 1; | |
273 | break; | |
bd4cf0ed | 274 | |
c2497395 AS |
275 | case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: |
276 | cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, | |
277 | BPF_REG_A, BPF_REG_CTX, insn); | |
278 | insn += cnt - 1; | |
bd4cf0ed AS |
279 | break; |
280 | ||
27cd5452 MS |
281 | case SKF_AD_OFF + SKF_AD_VLAN_TPID: |
282 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); | |
283 | ||
284 | /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ | |
285 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, | |
286 | offsetof(struct sk_buff, vlan_proto)); | |
287 | /* A = ntohs(A) [emitting a nop or swap16] */ | |
288 | *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); | |
289 | break; | |
290 | ||
bd4cf0ed AS |
291 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: |
292 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
293 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
294 | case SKF_AD_OFF + SKF_AD_CPU: | |
4cd3675e | 295 | case SKF_AD_OFF + SKF_AD_RANDOM: |
e430f34e | 296 | /* arg1 = CTX */ |
f8f6d679 | 297 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); |
bd4cf0ed | 298 | /* arg2 = A */ |
f8f6d679 | 299 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); |
bd4cf0ed | 300 | /* arg3 = X */ |
f8f6d679 | 301 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); |
e430f34e | 302 | /* Emit call(arg1=CTX, arg2=A, arg3=X) */ |
bd4cf0ed AS |
303 | switch (fp->k) { |
304 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
f8f6d679 | 305 | *insn = BPF_EMIT_CALL(__skb_get_pay_offset); |
bd4cf0ed AS |
306 | break; |
307 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
f8f6d679 | 308 | *insn = BPF_EMIT_CALL(__skb_get_nlattr); |
bd4cf0ed AS |
309 | break; |
310 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
f8f6d679 | 311 | *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest); |
bd4cf0ed AS |
312 | break; |
313 | case SKF_AD_OFF + SKF_AD_CPU: | |
f8f6d679 | 314 | *insn = BPF_EMIT_CALL(__get_raw_cpu_id); |
bd4cf0ed | 315 | break; |
4cd3675e | 316 | case SKF_AD_OFF + SKF_AD_RANDOM: |
3ad00405 DB |
317 | *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); |
318 | bpf_user_rnd_init_once(); | |
4cd3675e | 319 | break; |
bd4cf0ed AS |
320 | } |
321 | break; | |
322 | ||
323 | case SKF_AD_OFF + SKF_AD_ALU_XOR_X: | |
9739eef1 AS |
324 | /* A ^= X */ |
325 | *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); | |
bd4cf0ed AS |
326 | break; |
327 | ||
328 | default: | |
329 | /* This is just a dummy call to avoid letting the compiler | |
330 | * evict __bpf_call_base() as an optimization. Placed here | |
331 | * where no-one bothers. | |
332 | */ | |
333 | BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); | |
334 | return false; | |
335 | } | |
336 | ||
337 | *insnp = insn; | |
338 | return true; | |
339 | } | |
340 | ||
341 | /** | |
8fb575ca | 342 | * bpf_convert_filter - convert filter program |
bd4cf0ed AS |
343 | * @prog: the user passed filter program |
344 | * @len: the length of the user passed filter program | |
345 | * @new_prog: buffer where converted program will be stored | |
346 | * @new_len: pointer to store length of converted program | |
347 | * | |
348 | * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style. | |
349 | * Conversion workflow: | |
350 | * | |
351 | * 1) First pass for calculating the new program length: | |
8fb575ca | 352 | * bpf_convert_filter(old_prog, old_len, NULL, &new_len) |
bd4cf0ed AS |
353 | * |
354 | * 2) 2nd pass to remap in two passes: 1st pass finds new | |
355 | * jump offsets, 2nd pass remapping: | |
2695fb55 | 356 | * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len); |
8fb575ca | 357 | * bpf_convert_filter(old_prog, old_len, new_prog, &new_len); |
bd4cf0ed | 358 | */ |
d9e12f42 NS |
359 | static int bpf_convert_filter(struct sock_filter *prog, int len, |
360 | struct bpf_insn *new_prog, int *new_len) | |
bd4cf0ed AS |
361 | { |
362 | int new_flen = 0, pass = 0, target, i; | |
2695fb55 | 363 | struct bpf_insn *new_insn; |
bd4cf0ed AS |
364 | struct sock_filter *fp; |
365 | int *addrs = NULL; | |
366 | u8 bpf_src; | |
367 | ||
368 | BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); | |
30743837 | 369 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); |
bd4cf0ed | 370 | |
6f9a093b | 371 | if (len <= 0 || len > BPF_MAXINSNS) |
bd4cf0ed AS |
372 | return -EINVAL; |
373 | ||
374 | if (new_prog) { | |
658da937 DB |
375 | addrs = kcalloc(len, sizeof(*addrs), |
376 | GFP_KERNEL | __GFP_NOWARN); | |
bd4cf0ed AS |
377 | if (!addrs) |
378 | return -ENOMEM; | |
379 | } | |
380 | ||
381 | do_pass: | |
382 | new_insn = new_prog; | |
383 | fp = prog; | |
384 | ||
8b614aeb DB |
385 | /* Classic BPF related prologue emission. */ |
386 | if (new_insn) { | |
387 | /* Classic BPF expects A and X to be reset first. These need | |
388 | * to be guaranteed to be the first two instructions. | |
389 | */ | |
390 | *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); | |
391 | *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); | |
392 | ||
393 | /* All programs must keep CTX in callee saved BPF_REG_CTX. | |
394 | * In eBPF case it's done by the compiler, here we need to | |
395 | * do this ourself. Initial CTX is present in BPF_REG_ARG1. | |
396 | */ | |
397 | *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); | |
398 | } else { | |
399 | new_insn += 3; | |
400 | } | |
bd4cf0ed AS |
401 | |
402 | for (i = 0; i < len; fp++, i++) { | |
2695fb55 AS |
403 | struct bpf_insn tmp_insns[6] = { }; |
404 | struct bpf_insn *insn = tmp_insns; | |
bd4cf0ed AS |
405 | |
406 | if (addrs) | |
407 | addrs[i] = new_insn - new_prog; | |
408 | ||
409 | switch (fp->code) { | |
410 | /* All arithmetic insns and skb loads map as-is. */ | |
411 | case BPF_ALU | BPF_ADD | BPF_X: | |
412 | case BPF_ALU | BPF_ADD | BPF_K: | |
413 | case BPF_ALU | BPF_SUB | BPF_X: | |
414 | case BPF_ALU | BPF_SUB | BPF_K: | |
415 | case BPF_ALU | BPF_AND | BPF_X: | |
416 | case BPF_ALU | BPF_AND | BPF_K: | |
417 | case BPF_ALU | BPF_OR | BPF_X: | |
418 | case BPF_ALU | BPF_OR | BPF_K: | |
419 | case BPF_ALU | BPF_LSH | BPF_X: | |
420 | case BPF_ALU | BPF_LSH | BPF_K: | |
421 | case BPF_ALU | BPF_RSH | BPF_X: | |
422 | case BPF_ALU | BPF_RSH | BPF_K: | |
423 | case BPF_ALU | BPF_XOR | BPF_X: | |
424 | case BPF_ALU | BPF_XOR | BPF_K: | |
425 | case BPF_ALU | BPF_MUL | BPF_X: | |
426 | case BPF_ALU | BPF_MUL | BPF_K: | |
427 | case BPF_ALU | BPF_DIV | BPF_X: | |
428 | case BPF_ALU | BPF_DIV | BPF_K: | |
429 | case BPF_ALU | BPF_MOD | BPF_X: | |
430 | case BPF_ALU | BPF_MOD | BPF_K: | |
431 | case BPF_ALU | BPF_NEG: | |
432 | case BPF_LD | BPF_ABS | BPF_W: | |
433 | case BPF_LD | BPF_ABS | BPF_H: | |
434 | case BPF_LD | BPF_ABS | BPF_B: | |
435 | case BPF_LD | BPF_IND | BPF_W: | |
436 | case BPF_LD | BPF_IND | BPF_H: | |
437 | case BPF_LD | BPF_IND | BPF_B: | |
438 | /* Check for overloaded BPF extension and | |
439 | * directly convert it if found, otherwise | |
440 | * just move on with mapping. | |
441 | */ | |
442 | if (BPF_CLASS(fp->code) == BPF_LD && | |
443 | BPF_MODE(fp->code) == BPF_ABS && | |
444 | convert_bpf_extensions(fp, &insn)) | |
445 | break; | |
446 | ||
f8f6d679 | 447 | *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); |
bd4cf0ed AS |
448 | break; |
449 | ||
f8f6d679 DB |
450 | /* Jump transformation cannot use BPF block macros |
451 | * everywhere as offset calculation and target updates | |
452 | * require a bit more work than the rest, i.e. jump | |
453 | * opcodes map as-is, but offsets need adjustment. | |
454 | */ | |
455 | ||
456 | #define BPF_EMIT_JMP \ | |
bd4cf0ed AS |
457 | do { \ |
458 | if (target >= len || target < 0) \ | |
459 | goto err; \ | |
460 | insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ | |
461 | /* Adjust pc relative offset for 2nd or 3rd insn. */ \ | |
462 | insn->off -= insn - tmp_insns; \ | |
463 | } while (0) | |
464 | ||
f8f6d679 DB |
465 | case BPF_JMP | BPF_JA: |
466 | target = i + fp->k + 1; | |
467 | insn->code = fp->code; | |
468 | BPF_EMIT_JMP; | |
bd4cf0ed AS |
469 | break; |
470 | ||
471 | case BPF_JMP | BPF_JEQ | BPF_K: | |
472 | case BPF_JMP | BPF_JEQ | BPF_X: | |
473 | case BPF_JMP | BPF_JSET | BPF_K: | |
474 | case BPF_JMP | BPF_JSET | BPF_X: | |
475 | case BPF_JMP | BPF_JGT | BPF_K: | |
476 | case BPF_JMP | BPF_JGT | BPF_X: | |
477 | case BPF_JMP | BPF_JGE | BPF_K: | |
478 | case BPF_JMP | BPF_JGE | BPF_X: | |
479 | if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { | |
480 | /* BPF immediates are signed, zero extend | |
481 | * immediate into tmp register and use it | |
482 | * in compare insn. | |
483 | */ | |
f8f6d679 | 484 | *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); |
bd4cf0ed | 485 | |
e430f34e AS |
486 | insn->dst_reg = BPF_REG_A; |
487 | insn->src_reg = BPF_REG_TMP; | |
bd4cf0ed AS |
488 | bpf_src = BPF_X; |
489 | } else { | |
e430f34e | 490 | insn->dst_reg = BPF_REG_A; |
bd4cf0ed AS |
491 | insn->imm = fp->k; |
492 | bpf_src = BPF_SRC(fp->code); | |
19539ce7 | 493 | insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; |
1da177e4 | 494 | } |
bd4cf0ed AS |
495 | |
496 | /* Common case where 'jump_false' is next insn. */ | |
497 | if (fp->jf == 0) { | |
498 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
499 | target = i + fp->jt + 1; | |
f8f6d679 | 500 | BPF_EMIT_JMP; |
bd4cf0ed | 501 | break; |
1da177e4 | 502 | } |
bd4cf0ed AS |
503 | |
504 | /* Convert JEQ into JNE when 'jump_true' is next insn. */ | |
505 | if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) { | |
506 | insn->code = BPF_JMP | BPF_JNE | bpf_src; | |
507 | target = i + fp->jf + 1; | |
f8f6d679 | 508 | BPF_EMIT_JMP; |
bd4cf0ed | 509 | break; |
0b05b2a4 | 510 | } |
bd4cf0ed AS |
511 | |
512 | /* Other jumps are mapped into two insns: Jxx and JA. */ | |
513 | target = i + fp->jt + 1; | |
514 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
f8f6d679 | 515 | BPF_EMIT_JMP; |
bd4cf0ed AS |
516 | insn++; |
517 | ||
518 | insn->code = BPF_JMP | BPF_JA; | |
519 | target = i + fp->jf + 1; | |
f8f6d679 | 520 | BPF_EMIT_JMP; |
bd4cf0ed AS |
521 | break; |
522 | ||
523 | /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ | |
524 | case BPF_LDX | BPF_MSH | BPF_B: | |
9739eef1 | 525 | /* tmp = A */ |
f8f6d679 | 526 | *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A); |
1268e253 | 527 | /* A = BPF_R0 = *(u8 *) (skb->data + K) */ |
f8f6d679 | 528 | *insn++ = BPF_LD_ABS(BPF_B, fp->k); |
9739eef1 | 529 | /* A &= 0xf */ |
f8f6d679 | 530 | *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); |
9739eef1 | 531 | /* A <<= 2 */ |
f8f6d679 | 532 | *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); |
9739eef1 | 533 | /* X = A */ |
f8f6d679 | 534 | *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
9739eef1 | 535 | /* A = tmp */ |
f8f6d679 | 536 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); |
bd4cf0ed AS |
537 | break; |
538 | ||
6205b9cf DB |
539 | /* RET_K is remaped into 2 insns. RET_A case doesn't need an |
540 | * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. | |
541 | */ | |
bd4cf0ed AS |
542 | case BPF_RET | BPF_A: |
543 | case BPF_RET | BPF_K: | |
6205b9cf DB |
544 | if (BPF_RVAL(fp->code) == BPF_K) |
545 | *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, | |
546 | 0, fp->k); | |
9739eef1 | 547 | *insn = BPF_EXIT_INSN(); |
bd4cf0ed AS |
548 | break; |
549 | ||
550 | /* Store to stack. */ | |
551 | case BPF_ST: | |
552 | case BPF_STX: | |
f8f6d679 DB |
553 | *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == |
554 | BPF_ST ? BPF_REG_A : BPF_REG_X, | |
555 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
556 | break; |
557 | ||
558 | /* Load from stack. */ | |
559 | case BPF_LD | BPF_MEM: | |
560 | case BPF_LDX | BPF_MEM: | |
f8f6d679 DB |
561 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
562 | BPF_REG_A : BPF_REG_X, BPF_REG_FP, | |
563 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
564 | break; |
565 | ||
566 | /* A = K or X = K */ | |
567 | case BPF_LD | BPF_IMM: | |
568 | case BPF_LDX | BPF_IMM: | |
f8f6d679 DB |
569 | *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? |
570 | BPF_REG_A : BPF_REG_X, fp->k); | |
bd4cf0ed AS |
571 | break; |
572 | ||
573 | /* X = A */ | |
574 | case BPF_MISC | BPF_TAX: | |
f8f6d679 | 575 | *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
bd4cf0ed AS |
576 | break; |
577 | ||
578 | /* A = X */ | |
579 | case BPF_MISC | BPF_TXA: | |
f8f6d679 | 580 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); |
bd4cf0ed AS |
581 | break; |
582 | ||
583 | /* A = skb->len or X = skb->len */ | |
584 | case BPF_LD | BPF_W | BPF_LEN: | |
585 | case BPF_LDX | BPF_W | BPF_LEN: | |
f8f6d679 DB |
586 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
587 | BPF_REG_A : BPF_REG_X, BPF_REG_CTX, | |
588 | offsetof(struct sk_buff, len)); | |
bd4cf0ed AS |
589 | break; |
590 | ||
f8f6d679 | 591 | /* Access seccomp_data fields. */ |
bd4cf0ed | 592 | case BPF_LDX | BPF_ABS | BPF_W: |
9739eef1 AS |
593 | /* A = *(u32 *) (ctx + K) */ |
594 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); | |
bd4cf0ed AS |
595 | break; |
596 | ||
ca9f1fd2 | 597 | /* Unknown instruction. */ |
1da177e4 | 598 | default: |
bd4cf0ed | 599 | goto err; |
1da177e4 | 600 | } |
bd4cf0ed AS |
601 | |
602 | insn++; | |
603 | if (new_prog) | |
604 | memcpy(new_insn, tmp_insns, | |
605 | sizeof(*insn) * (insn - tmp_insns)); | |
bd4cf0ed | 606 | new_insn += insn - tmp_insns; |
1da177e4 LT |
607 | } |
608 | ||
bd4cf0ed AS |
609 | if (!new_prog) { |
610 | /* Only calculating new length. */ | |
611 | *new_len = new_insn - new_prog; | |
612 | return 0; | |
613 | } | |
614 | ||
615 | pass++; | |
616 | if (new_flen != new_insn - new_prog) { | |
617 | new_flen = new_insn - new_prog; | |
618 | if (pass > 2) | |
619 | goto err; | |
bd4cf0ed AS |
620 | goto do_pass; |
621 | } | |
622 | ||
623 | kfree(addrs); | |
624 | BUG_ON(*new_len != new_flen); | |
1da177e4 | 625 | return 0; |
bd4cf0ed AS |
626 | err: |
627 | kfree(addrs); | |
628 | return -EINVAL; | |
1da177e4 LT |
629 | } |
630 | ||
bd4cf0ed | 631 | /* Security: |
bd4cf0ed | 632 | * |
2d5311e4 | 633 | * As we dont want to clear mem[] array for each packet going through |
8ea6e345 | 634 | * __bpf_prog_run(), we check that filter loaded by user never try to read |
2d5311e4 | 635 | * a cell if not previously written, and we check all branches to be sure |
25985edc | 636 | * a malicious user doesn't try to abuse us. |
2d5311e4 | 637 | */ |
ec31a05c | 638 | static int check_load_and_stores(const struct sock_filter *filter, int flen) |
2d5311e4 | 639 | { |
34805931 | 640 | u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ |
2d5311e4 ED |
641 | int pc, ret = 0; |
642 | ||
643 | BUILD_BUG_ON(BPF_MEMWORDS > 16); | |
34805931 | 644 | |
99e72a0f | 645 | masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); |
2d5311e4 ED |
646 | if (!masks) |
647 | return -ENOMEM; | |
34805931 | 648 | |
2d5311e4 ED |
649 | memset(masks, 0xff, flen * sizeof(*masks)); |
650 | ||
651 | for (pc = 0; pc < flen; pc++) { | |
652 | memvalid &= masks[pc]; | |
653 | ||
654 | switch (filter[pc].code) { | |
34805931 DB |
655 | case BPF_ST: |
656 | case BPF_STX: | |
2d5311e4 ED |
657 | memvalid |= (1 << filter[pc].k); |
658 | break; | |
34805931 DB |
659 | case BPF_LD | BPF_MEM: |
660 | case BPF_LDX | BPF_MEM: | |
2d5311e4 ED |
661 | if (!(memvalid & (1 << filter[pc].k))) { |
662 | ret = -EINVAL; | |
663 | goto error; | |
664 | } | |
665 | break; | |
34805931 DB |
666 | case BPF_JMP | BPF_JA: |
667 | /* A jump must set masks on target */ | |
2d5311e4 ED |
668 | masks[pc + 1 + filter[pc].k] &= memvalid; |
669 | memvalid = ~0; | |
670 | break; | |
34805931 DB |
671 | case BPF_JMP | BPF_JEQ | BPF_K: |
672 | case BPF_JMP | BPF_JEQ | BPF_X: | |
673 | case BPF_JMP | BPF_JGE | BPF_K: | |
674 | case BPF_JMP | BPF_JGE | BPF_X: | |
675 | case BPF_JMP | BPF_JGT | BPF_K: | |
676 | case BPF_JMP | BPF_JGT | BPF_X: | |
677 | case BPF_JMP | BPF_JSET | BPF_K: | |
678 | case BPF_JMP | BPF_JSET | BPF_X: | |
679 | /* A jump must set masks on targets */ | |
2d5311e4 ED |
680 | masks[pc + 1 + filter[pc].jt] &= memvalid; |
681 | masks[pc + 1 + filter[pc].jf] &= memvalid; | |
682 | memvalid = ~0; | |
683 | break; | |
684 | } | |
685 | } | |
686 | error: | |
687 | kfree(masks); | |
688 | return ret; | |
689 | } | |
690 | ||
34805931 DB |
691 | static bool chk_code_allowed(u16 code_to_probe) |
692 | { | |
693 | static const bool codes[] = { | |
694 | /* 32 bit ALU operations */ | |
695 | [BPF_ALU | BPF_ADD | BPF_K] = true, | |
696 | [BPF_ALU | BPF_ADD | BPF_X] = true, | |
697 | [BPF_ALU | BPF_SUB | BPF_K] = true, | |
698 | [BPF_ALU | BPF_SUB | BPF_X] = true, | |
699 | [BPF_ALU | BPF_MUL | BPF_K] = true, | |
700 | [BPF_ALU | BPF_MUL | BPF_X] = true, | |
701 | [BPF_ALU | BPF_DIV | BPF_K] = true, | |
702 | [BPF_ALU | BPF_DIV | BPF_X] = true, | |
703 | [BPF_ALU | BPF_MOD | BPF_K] = true, | |
704 | [BPF_ALU | BPF_MOD | BPF_X] = true, | |
705 | [BPF_ALU | BPF_AND | BPF_K] = true, | |
706 | [BPF_ALU | BPF_AND | BPF_X] = true, | |
707 | [BPF_ALU | BPF_OR | BPF_K] = true, | |
708 | [BPF_ALU | BPF_OR | BPF_X] = true, | |
709 | [BPF_ALU | BPF_XOR | BPF_K] = true, | |
710 | [BPF_ALU | BPF_XOR | BPF_X] = true, | |
711 | [BPF_ALU | BPF_LSH | BPF_K] = true, | |
712 | [BPF_ALU | BPF_LSH | BPF_X] = true, | |
713 | [BPF_ALU | BPF_RSH | BPF_K] = true, | |
714 | [BPF_ALU | BPF_RSH | BPF_X] = true, | |
715 | [BPF_ALU | BPF_NEG] = true, | |
716 | /* Load instructions */ | |
717 | [BPF_LD | BPF_W | BPF_ABS] = true, | |
718 | [BPF_LD | BPF_H | BPF_ABS] = true, | |
719 | [BPF_LD | BPF_B | BPF_ABS] = true, | |
720 | [BPF_LD | BPF_W | BPF_LEN] = true, | |
721 | [BPF_LD | BPF_W | BPF_IND] = true, | |
722 | [BPF_LD | BPF_H | BPF_IND] = true, | |
723 | [BPF_LD | BPF_B | BPF_IND] = true, | |
724 | [BPF_LD | BPF_IMM] = true, | |
725 | [BPF_LD | BPF_MEM] = true, | |
726 | [BPF_LDX | BPF_W | BPF_LEN] = true, | |
727 | [BPF_LDX | BPF_B | BPF_MSH] = true, | |
728 | [BPF_LDX | BPF_IMM] = true, | |
729 | [BPF_LDX | BPF_MEM] = true, | |
730 | /* Store instructions */ | |
731 | [BPF_ST] = true, | |
732 | [BPF_STX] = true, | |
733 | /* Misc instructions */ | |
734 | [BPF_MISC | BPF_TAX] = true, | |
735 | [BPF_MISC | BPF_TXA] = true, | |
736 | /* Return instructions */ | |
737 | [BPF_RET | BPF_K] = true, | |
738 | [BPF_RET | BPF_A] = true, | |
739 | /* Jump instructions */ | |
740 | [BPF_JMP | BPF_JA] = true, | |
741 | [BPF_JMP | BPF_JEQ | BPF_K] = true, | |
742 | [BPF_JMP | BPF_JEQ | BPF_X] = true, | |
743 | [BPF_JMP | BPF_JGE | BPF_K] = true, | |
744 | [BPF_JMP | BPF_JGE | BPF_X] = true, | |
745 | [BPF_JMP | BPF_JGT | BPF_K] = true, | |
746 | [BPF_JMP | BPF_JGT | BPF_X] = true, | |
747 | [BPF_JMP | BPF_JSET | BPF_K] = true, | |
748 | [BPF_JMP | BPF_JSET | BPF_X] = true, | |
749 | }; | |
750 | ||
751 | if (code_to_probe >= ARRAY_SIZE(codes)) | |
752 | return false; | |
753 | ||
754 | return codes[code_to_probe]; | |
755 | } | |
756 | ||
f7bd9e36 DB |
757 | static bool bpf_check_basics_ok(const struct sock_filter *filter, |
758 | unsigned int flen) | |
759 | { | |
760 | if (filter == NULL) | |
761 | return false; | |
762 | if (flen == 0 || flen > BPF_MAXINSNS) | |
763 | return false; | |
764 | ||
765 | return true; | |
766 | } | |
767 | ||
1da177e4 | 768 | /** |
4df95ff4 | 769 | * bpf_check_classic - verify socket filter code |
1da177e4 LT |
770 | * @filter: filter to verify |
771 | * @flen: length of filter | |
772 | * | |
773 | * Check the user's filter code. If we let some ugly | |
774 | * filter code slip through kaboom! The filter must contain | |
93699863 KK |
775 | * no references or jumps that are out of range, no illegal |
776 | * instructions, and must end with a RET instruction. | |
1da177e4 | 777 | * |
7b11f69f KK |
778 | * All jumps are forward as they are not signed. |
779 | * | |
780 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
1da177e4 | 781 | */ |
d9e12f42 NS |
782 | static int bpf_check_classic(const struct sock_filter *filter, |
783 | unsigned int flen) | |
1da177e4 | 784 | { |
aa1113d9 | 785 | bool anc_found; |
34805931 | 786 | int pc; |
1da177e4 | 787 | |
34805931 | 788 | /* Check the filter code now */ |
1da177e4 | 789 | for (pc = 0; pc < flen; pc++) { |
ec31a05c | 790 | const struct sock_filter *ftest = &filter[pc]; |
93699863 | 791 | |
34805931 DB |
792 | /* May we actually operate on this code? */ |
793 | if (!chk_code_allowed(ftest->code)) | |
cba328fc | 794 | return -EINVAL; |
34805931 | 795 | |
93699863 | 796 | /* Some instructions need special checks */ |
34805931 DB |
797 | switch (ftest->code) { |
798 | case BPF_ALU | BPF_DIV | BPF_K: | |
799 | case BPF_ALU | BPF_MOD | BPF_K: | |
800 | /* Check for division by zero */ | |
b6069a95 ED |
801 | if (ftest->k == 0) |
802 | return -EINVAL; | |
803 | break; | |
229394e8 RV |
804 | case BPF_ALU | BPF_LSH | BPF_K: |
805 | case BPF_ALU | BPF_RSH | BPF_K: | |
806 | if (ftest->k >= 32) | |
807 | return -EINVAL; | |
808 | break; | |
34805931 DB |
809 | case BPF_LD | BPF_MEM: |
810 | case BPF_LDX | BPF_MEM: | |
811 | case BPF_ST: | |
812 | case BPF_STX: | |
813 | /* Check for invalid memory addresses */ | |
93699863 KK |
814 | if (ftest->k >= BPF_MEMWORDS) |
815 | return -EINVAL; | |
816 | break; | |
34805931 DB |
817 | case BPF_JMP | BPF_JA: |
818 | /* Note, the large ftest->k might cause loops. | |
93699863 KK |
819 | * Compare this with conditional jumps below, |
820 | * where offsets are limited. --ANK (981016) | |
821 | */ | |
34805931 | 822 | if (ftest->k >= (unsigned int)(flen - pc - 1)) |
93699863 | 823 | return -EINVAL; |
01f2f3f6 | 824 | break; |
34805931 DB |
825 | case BPF_JMP | BPF_JEQ | BPF_K: |
826 | case BPF_JMP | BPF_JEQ | BPF_X: | |
827 | case BPF_JMP | BPF_JGE | BPF_K: | |
828 | case BPF_JMP | BPF_JGE | BPF_X: | |
829 | case BPF_JMP | BPF_JGT | BPF_K: | |
830 | case BPF_JMP | BPF_JGT | BPF_X: | |
831 | case BPF_JMP | BPF_JSET | BPF_K: | |
832 | case BPF_JMP | BPF_JSET | BPF_X: | |
833 | /* Both conditionals must be safe */ | |
e35bedf3 | 834 | if (pc + ftest->jt + 1 >= flen || |
93699863 KK |
835 | pc + ftest->jf + 1 >= flen) |
836 | return -EINVAL; | |
cba328fc | 837 | break; |
34805931 DB |
838 | case BPF_LD | BPF_W | BPF_ABS: |
839 | case BPF_LD | BPF_H | BPF_ABS: | |
840 | case BPF_LD | BPF_B | BPF_ABS: | |
aa1113d9 | 841 | anc_found = false; |
34805931 DB |
842 | if (bpf_anc_helper(ftest) & BPF_ANC) |
843 | anc_found = true; | |
844 | /* Ancillary operation unknown or unsupported */ | |
aa1113d9 DB |
845 | if (anc_found == false && ftest->k >= SKF_AD_OFF) |
846 | return -EINVAL; | |
01f2f3f6 HPP |
847 | } |
848 | } | |
93699863 | 849 | |
34805931 | 850 | /* Last instruction must be a RET code */ |
01f2f3f6 | 851 | switch (filter[flen - 1].code) { |
34805931 DB |
852 | case BPF_RET | BPF_K: |
853 | case BPF_RET | BPF_A: | |
2d5311e4 | 854 | return check_load_and_stores(filter, flen); |
cba328fc | 855 | } |
34805931 | 856 | |
cba328fc | 857 | return -EINVAL; |
1da177e4 LT |
858 | } |
859 | ||
7ae457c1 AS |
860 | static int bpf_prog_store_orig_filter(struct bpf_prog *fp, |
861 | const struct sock_fprog *fprog) | |
a3ea269b | 862 | { |
009937e7 | 863 | unsigned int fsize = bpf_classic_proglen(fprog); |
a3ea269b DB |
864 | struct sock_fprog_kern *fkprog; |
865 | ||
866 | fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); | |
867 | if (!fp->orig_prog) | |
868 | return -ENOMEM; | |
869 | ||
870 | fkprog = fp->orig_prog; | |
871 | fkprog->len = fprog->len; | |
658da937 DB |
872 | |
873 | fkprog->filter = kmemdup(fp->insns, fsize, | |
874 | GFP_KERNEL | __GFP_NOWARN); | |
a3ea269b DB |
875 | if (!fkprog->filter) { |
876 | kfree(fp->orig_prog); | |
877 | return -ENOMEM; | |
878 | } | |
879 | ||
880 | return 0; | |
881 | } | |
882 | ||
7ae457c1 | 883 | static void bpf_release_orig_filter(struct bpf_prog *fp) |
a3ea269b DB |
884 | { |
885 | struct sock_fprog_kern *fprog = fp->orig_prog; | |
886 | ||
887 | if (fprog) { | |
888 | kfree(fprog->filter); | |
889 | kfree(fprog); | |
890 | } | |
891 | } | |
892 | ||
7ae457c1 AS |
893 | static void __bpf_prog_release(struct bpf_prog *prog) |
894 | { | |
24701ece | 895 | if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { |
89aa0758 AS |
896 | bpf_prog_put(prog); |
897 | } else { | |
898 | bpf_release_orig_filter(prog); | |
899 | bpf_prog_free(prog); | |
900 | } | |
7ae457c1 AS |
901 | } |
902 | ||
34c5bd66 PN |
903 | static void __sk_filter_release(struct sk_filter *fp) |
904 | { | |
7ae457c1 AS |
905 | __bpf_prog_release(fp->prog); |
906 | kfree(fp); | |
34c5bd66 PN |
907 | } |
908 | ||
47e958ea | 909 | /** |
46bcf14f | 910 | * sk_filter_release_rcu - Release a socket filter by rcu_head |
47e958ea PE |
911 | * @rcu: rcu_head that contains the sk_filter to free |
912 | */ | |
fbc907f0 | 913 | static void sk_filter_release_rcu(struct rcu_head *rcu) |
47e958ea PE |
914 | { |
915 | struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); | |
916 | ||
34c5bd66 | 917 | __sk_filter_release(fp); |
47e958ea | 918 | } |
fbc907f0 DB |
919 | |
920 | /** | |
921 | * sk_filter_release - release a socket filter | |
922 | * @fp: filter to remove | |
923 | * | |
924 | * Remove a filter from a socket and release its resources. | |
925 | */ | |
926 | static void sk_filter_release(struct sk_filter *fp) | |
927 | { | |
928 | if (atomic_dec_and_test(&fp->refcnt)) | |
929 | call_rcu(&fp->rcu, sk_filter_release_rcu); | |
930 | } | |
931 | ||
932 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) | |
933 | { | |
7ae457c1 | 934 | u32 filter_size = bpf_prog_size(fp->prog->len); |
fbc907f0 | 935 | |
278571ba AS |
936 | atomic_sub(filter_size, &sk->sk_omem_alloc); |
937 | sk_filter_release(fp); | |
fbc907f0 | 938 | } |
47e958ea | 939 | |
278571ba AS |
940 | /* try to charge the socket memory if there is space available |
941 | * return true on success | |
942 | */ | |
943 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) | |
bd4cf0ed | 944 | { |
7ae457c1 | 945 | u32 filter_size = bpf_prog_size(fp->prog->len); |
278571ba AS |
946 | |
947 | /* same check as in sock_kmalloc() */ | |
948 | if (filter_size <= sysctl_optmem_max && | |
949 | atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { | |
950 | atomic_inc(&fp->refcnt); | |
951 | atomic_add(filter_size, &sk->sk_omem_alloc); | |
952 | return true; | |
bd4cf0ed | 953 | } |
278571ba | 954 | return false; |
bd4cf0ed AS |
955 | } |
956 | ||
7ae457c1 | 957 | static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) |
bd4cf0ed AS |
958 | { |
959 | struct sock_filter *old_prog; | |
7ae457c1 | 960 | struct bpf_prog *old_fp; |
34805931 | 961 | int err, new_len, old_len = fp->len; |
bd4cf0ed AS |
962 | |
963 | /* We are free to overwrite insns et al right here as it | |
964 | * won't be used at this point in time anymore internally | |
965 | * after the migration to the internal BPF instruction | |
966 | * representation. | |
967 | */ | |
968 | BUILD_BUG_ON(sizeof(struct sock_filter) != | |
2695fb55 | 969 | sizeof(struct bpf_insn)); |
bd4cf0ed | 970 | |
bd4cf0ed AS |
971 | /* Conversion cannot happen on overlapping memory areas, |
972 | * so we need to keep the user BPF around until the 2nd | |
973 | * pass. At this time, the user BPF is stored in fp->insns. | |
974 | */ | |
975 | old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), | |
658da937 | 976 | GFP_KERNEL | __GFP_NOWARN); |
bd4cf0ed AS |
977 | if (!old_prog) { |
978 | err = -ENOMEM; | |
979 | goto out_err; | |
980 | } | |
981 | ||
982 | /* 1st pass: calculate the new program length. */ | |
8fb575ca | 983 | err = bpf_convert_filter(old_prog, old_len, NULL, &new_len); |
bd4cf0ed AS |
984 | if (err) |
985 | goto out_err_free; | |
986 | ||
987 | /* Expand fp for appending the new filter representation. */ | |
988 | old_fp = fp; | |
60a3b225 | 989 | fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); |
bd4cf0ed AS |
990 | if (!fp) { |
991 | /* The old_fp is still around in case we couldn't | |
992 | * allocate new memory, so uncharge on that one. | |
993 | */ | |
994 | fp = old_fp; | |
995 | err = -ENOMEM; | |
996 | goto out_err_free; | |
997 | } | |
998 | ||
bd4cf0ed AS |
999 | fp->len = new_len; |
1000 | ||
2695fb55 | 1001 | /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ |
8fb575ca | 1002 | err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len); |
bd4cf0ed | 1003 | if (err) |
8fb575ca | 1004 | /* 2nd bpf_convert_filter() can fail only if it fails |
bd4cf0ed AS |
1005 | * to allocate memory, remapping must succeed. Note, |
1006 | * that at this time old_fp has already been released | |
278571ba | 1007 | * by krealloc(). |
bd4cf0ed AS |
1008 | */ |
1009 | goto out_err_free; | |
1010 | ||
d1c55ab5 DB |
1011 | /* We are guaranteed to never error here with cBPF to eBPF |
1012 | * transitions, since there's no issue with type compatibility | |
1013 | * checks on program arrays. | |
1014 | */ | |
1015 | fp = bpf_prog_select_runtime(fp, &err); | |
5fe821a9 | 1016 | |
bd4cf0ed AS |
1017 | kfree(old_prog); |
1018 | return fp; | |
1019 | ||
1020 | out_err_free: | |
1021 | kfree(old_prog); | |
1022 | out_err: | |
7ae457c1 | 1023 | __bpf_prog_release(fp); |
bd4cf0ed AS |
1024 | return ERR_PTR(err); |
1025 | } | |
1026 | ||
ac67eb2c DB |
1027 | static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, |
1028 | bpf_aux_classic_check_t trans) | |
302d6637 JP |
1029 | { |
1030 | int err; | |
1031 | ||
bd4cf0ed | 1032 | fp->bpf_func = NULL; |
a91263d5 | 1033 | fp->jited = 0; |
302d6637 | 1034 | |
4df95ff4 | 1035 | err = bpf_check_classic(fp->insns, fp->len); |
418c96ac | 1036 | if (err) { |
7ae457c1 | 1037 | __bpf_prog_release(fp); |
bd4cf0ed | 1038 | return ERR_PTR(err); |
418c96ac | 1039 | } |
302d6637 | 1040 | |
4ae92bc7 NS |
1041 | /* There might be additional checks and transformations |
1042 | * needed on classic filters, f.e. in case of seccomp. | |
1043 | */ | |
1044 | if (trans) { | |
1045 | err = trans(fp->insns, fp->len); | |
1046 | if (err) { | |
1047 | __bpf_prog_release(fp); | |
1048 | return ERR_PTR(err); | |
1049 | } | |
1050 | } | |
1051 | ||
bd4cf0ed AS |
1052 | /* Probe if we can JIT compile the filter and if so, do |
1053 | * the compilation of the filter. | |
1054 | */ | |
302d6637 | 1055 | bpf_jit_compile(fp); |
bd4cf0ed AS |
1056 | |
1057 | /* JIT compiler couldn't process this filter, so do the | |
1058 | * internal BPF translation for the optimized interpreter. | |
1059 | */ | |
5fe821a9 | 1060 | if (!fp->jited) |
7ae457c1 | 1061 | fp = bpf_migrate_filter(fp); |
bd4cf0ed AS |
1062 | |
1063 | return fp; | |
302d6637 JP |
1064 | } |
1065 | ||
1066 | /** | |
7ae457c1 | 1067 | * bpf_prog_create - create an unattached filter |
c6c4b97c | 1068 | * @pfp: the unattached filter that is created |
677a9fd3 | 1069 | * @fprog: the filter program |
302d6637 | 1070 | * |
c6c4b97c | 1071 | * Create a filter independent of any socket. We first run some |
302d6637 JP |
1072 | * sanity checks on it to make sure it does not explode on us later. |
1073 | * If an error occurs or there is insufficient memory for the filter | |
1074 | * a negative errno code is returned. On success the return is zero. | |
1075 | */ | |
7ae457c1 | 1076 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) |
302d6637 | 1077 | { |
009937e7 | 1078 | unsigned int fsize = bpf_classic_proglen(fprog); |
7ae457c1 | 1079 | struct bpf_prog *fp; |
302d6637 JP |
1080 | |
1081 | /* Make sure new filter is there and in the right amounts. */ | |
f7bd9e36 | 1082 | if (!bpf_check_basics_ok(fprog->filter, fprog->len)) |
302d6637 JP |
1083 | return -EINVAL; |
1084 | ||
60a3b225 | 1085 | fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); |
302d6637 JP |
1086 | if (!fp) |
1087 | return -ENOMEM; | |
a3ea269b | 1088 | |
302d6637 JP |
1089 | memcpy(fp->insns, fprog->filter, fsize); |
1090 | ||
302d6637 | 1091 | fp->len = fprog->len; |
a3ea269b DB |
1092 | /* Since unattached filters are not copied back to user |
1093 | * space through sk_get_filter(), we do not need to hold | |
1094 | * a copy here, and can spare us the work. | |
1095 | */ | |
1096 | fp->orig_prog = NULL; | |
302d6637 | 1097 | |
7ae457c1 | 1098 | /* bpf_prepare_filter() already takes care of freeing |
bd4cf0ed AS |
1099 | * memory in case something goes wrong. |
1100 | */ | |
4ae92bc7 | 1101 | fp = bpf_prepare_filter(fp, NULL); |
bd4cf0ed AS |
1102 | if (IS_ERR(fp)) |
1103 | return PTR_ERR(fp); | |
302d6637 JP |
1104 | |
1105 | *pfp = fp; | |
1106 | return 0; | |
302d6637 | 1107 | } |
7ae457c1 | 1108 | EXPORT_SYMBOL_GPL(bpf_prog_create); |
302d6637 | 1109 | |
ac67eb2c DB |
1110 | /** |
1111 | * bpf_prog_create_from_user - create an unattached filter from user buffer | |
1112 | * @pfp: the unattached filter that is created | |
1113 | * @fprog: the filter program | |
1114 | * @trans: post-classic verifier transformation handler | |
bab18991 | 1115 | * @save_orig: save classic BPF program |
ac67eb2c DB |
1116 | * |
1117 | * This function effectively does the same as bpf_prog_create(), only | |
1118 | * that it builds up its insns buffer from user space provided buffer. | |
1119 | * It also allows for passing a bpf_aux_classic_check_t handler. | |
1120 | */ | |
1121 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, | |
bab18991 | 1122 | bpf_aux_classic_check_t trans, bool save_orig) |
ac67eb2c DB |
1123 | { |
1124 | unsigned int fsize = bpf_classic_proglen(fprog); | |
1125 | struct bpf_prog *fp; | |
bab18991 | 1126 | int err; |
ac67eb2c DB |
1127 | |
1128 | /* Make sure new filter is there and in the right amounts. */ | |
f7bd9e36 | 1129 | if (!bpf_check_basics_ok(fprog->filter, fprog->len)) |
ac67eb2c DB |
1130 | return -EINVAL; |
1131 | ||
1132 | fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); | |
1133 | if (!fp) | |
1134 | return -ENOMEM; | |
1135 | ||
1136 | if (copy_from_user(fp->insns, fprog->filter, fsize)) { | |
1137 | __bpf_prog_free(fp); | |
1138 | return -EFAULT; | |
1139 | } | |
1140 | ||
1141 | fp->len = fprog->len; | |
ac67eb2c DB |
1142 | fp->orig_prog = NULL; |
1143 | ||
bab18991 DB |
1144 | if (save_orig) { |
1145 | err = bpf_prog_store_orig_filter(fp, fprog); | |
1146 | if (err) { | |
1147 | __bpf_prog_free(fp); | |
1148 | return -ENOMEM; | |
1149 | } | |
1150 | } | |
1151 | ||
ac67eb2c DB |
1152 | /* bpf_prepare_filter() already takes care of freeing |
1153 | * memory in case something goes wrong. | |
1154 | */ | |
1155 | fp = bpf_prepare_filter(fp, trans); | |
1156 | if (IS_ERR(fp)) | |
1157 | return PTR_ERR(fp); | |
1158 | ||
1159 | *pfp = fp; | |
1160 | return 0; | |
1161 | } | |
2ea273d7 | 1162 | EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); |
ac67eb2c | 1163 | |
7ae457c1 | 1164 | void bpf_prog_destroy(struct bpf_prog *fp) |
302d6637 | 1165 | { |
7ae457c1 | 1166 | __bpf_prog_release(fp); |
302d6637 | 1167 | } |
7ae457c1 | 1168 | EXPORT_SYMBOL_GPL(bpf_prog_destroy); |
302d6637 | 1169 | |
8ced425e | 1170 | static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) |
49b31e57 DB |
1171 | { |
1172 | struct sk_filter *fp, *old_fp; | |
1173 | ||
1174 | fp = kmalloc(sizeof(*fp), GFP_KERNEL); | |
1175 | if (!fp) | |
1176 | return -ENOMEM; | |
1177 | ||
1178 | fp->prog = prog; | |
1179 | atomic_set(&fp->refcnt, 0); | |
1180 | ||
1181 | if (!sk_filter_charge(sk, fp)) { | |
1182 | kfree(fp); | |
1183 | return -ENOMEM; | |
1184 | } | |
1185 | ||
8ced425e HFS |
1186 | old_fp = rcu_dereference_protected(sk->sk_filter, |
1187 | lockdep_sock_is_held(sk)); | |
49b31e57 | 1188 | rcu_assign_pointer(sk->sk_filter, fp); |
8ced425e | 1189 | |
49b31e57 DB |
1190 | if (old_fp) |
1191 | sk_filter_uncharge(sk, old_fp); | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
538950a1 CG |
1196 | static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk) |
1197 | { | |
1198 | struct bpf_prog *old_prog; | |
1199 | int err; | |
1200 | ||
1201 | if (bpf_prog_size(prog->len) > sysctl_optmem_max) | |
1202 | return -ENOMEM; | |
1203 | ||
fa463497 | 1204 | if (sk_unhashed(sk) && sk->sk_reuseport) { |
538950a1 CG |
1205 | err = reuseport_alloc(sk); |
1206 | if (err) | |
1207 | return err; | |
1208 | } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) { | |
1209 | /* The socket wasn't bound with SO_REUSEPORT */ | |
1210 | return -EINVAL; | |
1211 | } | |
1212 | ||
1213 | old_prog = reuseport_attach_prog(sk, prog); | |
1214 | if (old_prog) | |
1215 | bpf_prog_destroy(old_prog); | |
1216 | ||
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | static | |
1221 | struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) | |
1da177e4 | 1222 | { |
009937e7 | 1223 | unsigned int fsize = bpf_classic_proglen(fprog); |
7ae457c1 | 1224 | struct bpf_prog *prog; |
1da177e4 LT |
1225 | int err; |
1226 | ||
d59577b6 | 1227 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
538950a1 | 1228 | return ERR_PTR(-EPERM); |
d59577b6 | 1229 | |
1da177e4 | 1230 | /* Make sure new filter is there and in the right amounts. */ |
f7bd9e36 | 1231 | if (!bpf_check_basics_ok(fprog->filter, fprog->len)) |
538950a1 | 1232 | return ERR_PTR(-EINVAL); |
1da177e4 | 1233 | |
f7bd9e36 | 1234 | prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); |
7ae457c1 | 1235 | if (!prog) |
538950a1 | 1236 | return ERR_PTR(-ENOMEM); |
a3ea269b | 1237 | |
7ae457c1 | 1238 | if (copy_from_user(prog->insns, fprog->filter, fsize)) { |
c0d1379a | 1239 | __bpf_prog_free(prog); |
538950a1 | 1240 | return ERR_PTR(-EFAULT); |
1da177e4 LT |
1241 | } |
1242 | ||
7ae457c1 | 1243 | prog->len = fprog->len; |
1da177e4 | 1244 | |
7ae457c1 | 1245 | err = bpf_prog_store_orig_filter(prog, fprog); |
a3ea269b | 1246 | if (err) { |
c0d1379a | 1247 | __bpf_prog_free(prog); |
538950a1 | 1248 | return ERR_PTR(-ENOMEM); |
a3ea269b DB |
1249 | } |
1250 | ||
7ae457c1 | 1251 | /* bpf_prepare_filter() already takes care of freeing |
bd4cf0ed AS |
1252 | * memory in case something goes wrong. |
1253 | */ | |
538950a1 CG |
1254 | return bpf_prepare_filter(prog, NULL); |
1255 | } | |
1256 | ||
1257 | /** | |
1258 | * sk_attach_filter - attach a socket filter | |
1259 | * @fprog: the filter program | |
1260 | * @sk: the socket to use | |
1261 | * | |
1262 | * Attach the user's filter code. We first run some sanity checks on | |
1263 | * it to make sure it does not explode on us later. If an error | |
1264 | * occurs or there is insufficient memory for the filter a negative | |
1265 | * errno code is returned. On success the return is zero. | |
1266 | */ | |
8ced425e | 1267 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) |
538950a1 CG |
1268 | { |
1269 | struct bpf_prog *prog = __get_filter(fprog, sk); | |
1270 | int err; | |
1271 | ||
7ae457c1 AS |
1272 | if (IS_ERR(prog)) |
1273 | return PTR_ERR(prog); | |
1274 | ||
8ced425e | 1275 | err = __sk_attach_prog(prog, sk); |
49b31e57 | 1276 | if (err < 0) { |
7ae457c1 | 1277 | __bpf_prog_release(prog); |
49b31e57 | 1278 | return err; |
278571ba AS |
1279 | } |
1280 | ||
d3904b73 | 1281 | return 0; |
1da177e4 | 1282 | } |
8ced425e | 1283 | EXPORT_SYMBOL_GPL(sk_attach_filter); |
1da177e4 | 1284 | |
538950a1 | 1285 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) |
89aa0758 | 1286 | { |
538950a1 | 1287 | struct bpf_prog *prog = __get_filter(fprog, sk); |
49b31e57 | 1288 | int err; |
89aa0758 | 1289 | |
538950a1 CG |
1290 | if (IS_ERR(prog)) |
1291 | return PTR_ERR(prog); | |
1292 | ||
1293 | err = __reuseport_attach_prog(prog, sk); | |
1294 | if (err < 0) { | |
1295 | __bpf_prog_release(prog); | |
1296 | return err; | |
1297 | } | |
1298 | ||
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) | |
1303 | { | |
89aa0758 | 1304 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
538950a1 | 1305 | return ERR_PTR(-EPERM); |
89aa0758 | 1306 | |
113214be | 1307 | return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); |
538950a1 CG |
1308 | } |
1309 | ||
1310 | int sk_attach_bpf(u32 ufd, struct sock *sk) | |
1311 | { | |
1312 | struct bpf_prog *prog = __get_bpf(ufd, sk); | |
1313 | int err; | |
1314 | ||
1315 | if (IS_ERR(prog)) | |
1316 | return PTR_ERR(prog); | |
1317 | ||
8ced425e | 1318 | err = __sk_attach_prog(prog, sk); |
49b31e57 | 1319 | if (err < 0) { |
89aa0758 | 1320 | bpf_prog_put(prog); |
49b31e57 | 1321 | return err; |
89aa0758 AS |
1322 | } |
1323 | ||
89aa0758 AS |
1324 | return 0; |
1325 | } | |
1326 | ||
538950a1 CG |
1327 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) |
1328 | { | |
1329 | struct bpf_prog *prog = __get_bpf(ufd, sk); | |
1330 | int err; | |
1331 | ||
1332 | if (IS_ERR(prog)) | |
1333 | return PTR_ERR(prog); | |
1334 | ||
1335 | err = __reuseport_attach_prog(prog, sk); | |
1336 | if (err < 0) { | |
1337 | bpf_prog_put(prog); | |
1338 | return err; | |
1339 | } | |
1340 | ||
1341 | return 0; | |
1342 | } | |
1343 | ||
21cafc1d DB |
1344 | struct bpf_scratchpad { |
1345 | union { | |
1346 | __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; | |
1347 | u8 buff[MAX_BPF_STACK]; | |
1348 | }; | |
1349 | }; | |
1350 | ||
1351 | static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); | |
91bc4822 | 1352 | |
5293efe6 DB |
1353 | static inline int __bpf_try_make_writable(struct sk_buff *skb, |
1354 | unsigned int write_len) | |
1355 | { | |
1356 | return skb_ensure_writable(skb, write_len); | |
1357 | } | |
1358 | ||
db58ba45 AS |
1359 | static inline int bpf_try_make_writable(struct sk_buff *skb, |
1360 | unsigned int write_len) | |
1361 | { | |
5293efe6 | 1362 | int err = __bpf_try_make_writable(skb, write_len); |
db58ba45 | 1363 | |
0ed661d5 | 1364 | bpf_compute_data_end(skb); |
db58ba45 AS |
1365 | return err; |
1366 | } | |
1367 | ||
a2bfe6bf DB |
1368 | static inline void bpf_push_mac_rcsum(struct sk_buff *skb) |
1369 | { | |
1370 | if (skb_at_tc_ingress(skb)) | |
1371 | skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); | |
1372 | } | |
1373 | ||
8065694e DB |
1374 | static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) |
1375 | { | |
1376 | if (skb_at_tc_ingress(skb)) | |
1377 | skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); | |
1378 | } | |
1379 | ||
91bc4822 | 1380 | static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags) |
608cd71a AS |
1381 | { |
1382 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
0ed661d5 | 1383 | unsigned int offset = (unsigned int) r2; |
608cd71a AS |
1384 | void *from = (void *) (long) r3; |
1385 | unsigned int len = (unsigned int) r4; | |
608cd71a AS |
1386 | void *ptr; |
1387 | ||
8afd54c8 | 1388 | if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) |
781c53bc | 1389 | return -EINVAL; |
0ed661d5 | 1390 | if (unlikely(offset > 0xffff)) |
608cd71a | 1391 | return -EFAULT; |
db58ba45 | 1392 | if (unlikely(bpf_try_make_writable(skb, offset + len))) |
608cd71a AS |
1393 | return -EFAULT; |
1394 | ||
0ed661d5 | 1395 | ptr = skb->data + offset; |
781c53bc | 1396 | if (flags & BPF_F_RECOMPUTE_CSUM) |
479ffccc | 1397 | __skb_postpull_rcsum(skb, ptr, len, offset); |
608cd71a AS |
1398 | |
1399 | memcpy(ptr, from, len); | |
1400 | ||
781c53bc | 1401 | if (flags & BPF_F_RECOMPUTE_CSUM) |
479ffccc | 1402 | __skb_postpush_rcsum(skb, ptr, len, offset); |
8afd54c8 DB |
1403 | if (flags & BPF_F_INVALIDATE_HASH) |
1404 | skb_clear_hash(skb); | |
f8ffad69 | 1405 | |
608cd71a AS |
1406 | return 0; |
1407 | } | |
1408 | ||
577c50aa | 1409 | static const struct bpf_func_proto bpf_skb_store_bytes_proto = { |
608cd71a AS |
1410 | .func = bpf_skb_store_bytes, |
1411 | .gpl_only = false, | |
1412 | .ret_type = RET_INTEGER, | |
1413 | .arg1_type = ARG_PTR_TO_CTX, | |
1414 | .arg2_type = ARG_ANYTHING, | |
1415 | .arg3_type = ARG_PTR_TO_STACK, | |
1416 | .arg4_type = ARG_CONST_STACK_SIZE, | |
91bc4822 AS |
1417 | .arg5_type = ARG_ANYTHING, |
1418 | }; | |
1419 | ||
05c74e5e DB |
1420 | static u64 bpf_skb_load_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1421 | { | |
1422 | const struct sk_buff *skb = (const struct sk_buff *)(unsigned long) r1; | |
0ed661d5 | 1423 | unsigned int offset = (unsigned int) r2; |
05c74e5e DB |
1424 | void *to = (void *)(unsigned long) r3; |
1425 | unsigned int len = (unsigned int) r4; | |
1426 | void *ptr; | |
1427 | ||
0ed661d5 | 1428 | if (unlikely(offset > 0xffff)) |
074f528e | 1429 | goto err_clear; |
05c74e5e DB |
1430 | |
1431 | ptr = skb_header_pointer(skb, offset, len, to); | |
1432 | if (unlikely(!ptr)) | |
074f528e | 1433 | goto err_clear; |
05c74e5e DB |
1434 | if (ptr != to) |
1435 | memcpy(to, ptr, len); | |
1436 | ||
1437 | return 0; | |
074f528e DB |
1438 | err_clear: |
1439 | memset(to, 0, len); | |
1440 | return -EFAULT; | |
05c74e5e DB |
1441 | } |
1442 | ||
577c50aa | 1443 | static const struct bpf_func_proto bpf_skb_load_bytes_proto = { |
05c74e5e DB |
1444 | .func = bpf_skb_load_bytes, |
1445 | .gpl_only = false, | |
1446 | .ret_type = RET_INTEGER, | |
1447 | .arg1_type = ARG_PTR_TO_CTX, | |
1448 | .arg2_type = ARG_ANYTHING, | |
074f528e | 1449 | .arg3_type = ARG_PTR_TO_RAW_STACK, |
05c74e5e DB |
1450 | .arg4_type = ARG_CONST_STACK_SIZE, |
1451 | }; | |
1452 | ||
a166151c | 1453 | static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags) |
91bc4822 AS |
1454 | { |
1455 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
0ed661d5 DB |
1456 | unsigned int offset = (unsigned int) r2; |
1457 | __sum16 *ptr; | |
91bc4822 | 1458 | |
781c53bc DB |
1459 | if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) |
1460 | return -EINVAL; | |
0ed661d5 | 1461 | if (unlikely(offset > 0xffff || offset & 1)) |
91bc4822 | 1462 | return -EFAULT; |
0ed661d5 | 1463 | if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) |
91bc4822 AS |
1464 | return -EFAULT; |
1465 | ||
0ed661d5 | 1466 | ptr = (__sum16 *)(skb->data + offset); |
781c53bc | 1467 | switch (flags & BPF_F_HDR_FIELD_MASK) { |
8050c0f0 DB |
1468 | case 0: |
1469 | if (unlikely(from != 0)) | |
1470 | return -EINVAL; | |
1471 | ||
1472 | csum_replace_by_diff(ptr, to); | |
1473 | break; | |
91bc4822 AS |
1474 | case 2: |
1475 | csum_replace2(ptr, from, to); | |
1476 | break; | |
1477 | case 4: | |
1478 | csum_replace4(ptr, from, to); | |
1479 | break; | |
1480 | default: | |
1481 | return -EINVAL; | |
1482 | } | |
1483 | ||
91bc4822 AS |
1484 | return 0; |
1485 | } | |
1486 | ||
577c50aa | 1487 | static const struct bpf_func_proto bpf_l3_csum_replace_proto = { |
91bc4822 AS |
1488 | .func = bpf_l3_csum_replace, |
1489 | .gpl_only = false, | |
1490 | .ret_type = RET_INTEGER, | |
1491 | .arg1_type = ARG_PTR_TO_CTX, | |
1492 | .arg2_type = ARG_ANYTHING, | |
1493 | .arg3_type = ARG_ANYTHING, | |
1494 | .arg4_type = ARG_ANYTHING, | |
1495 | .arg5_type = ARG_ANYTHING, | |
1496 | }; | |
1497 | ||
a166151c | 1498 | static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags) |
91bc4822 AS |
1499 | { |
1500 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
781c53bc | 1501 | bool is_pseudo = flags & BPF_F_PSEUDO_HDR; |
2f72959a | 1502 | bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; |
0ed661d5 DB |
1503 | unsigned int offset = (unsigned int) r2; |
1504 | __sum16 *ptr; | |
91bc4822 | 1505 | |
2f72959a DB |
1506 | if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR | |
1507 | BPF_F_HDR_FIELD_MASK))) | |
781c53bc | 1508 | return -EINVAL; |
0ed661d5 | 1509 | if (unlikely(offset > 0xffff || offset & 1)) |
91bc4822 | 1510 | return -EFAULT; |
0ed661d5 | 1511 | if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) |
91bc4822 AS |
1512 | return -EFAULT; |
1513 | ||
0ed661d5 | 1514 | ptr = (__sum16 *)(skb->data + offset); |
2f72959a DB |
1515 | if (is_mmzero && !*ptr) |
1516 | return 0; | |
91bc4822 | 1517 | |
781c53bc | 1518 | switch (flags & BPF_F_HDR_FIELD_MASK) { |
7d672345 DB |
1519 | case 0: |
1520 | if (unlikely(from != 0)) | |
1521 | return -EINVAL; | |
1522 | ||
1523 | inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); | |
1524 | break; | |
91bc4822 AS |
1525 | case 2: |
1526 | inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); | |
1527 | break; | |
1528 | case 4: | |
1529 | inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); | |
1530 | break; | |
1531 | default: | |
1532 | return -EINVAL; | |
1533 | } | |
1534 | ||
2f72959a DB |
1535 | if (is_mmzero && !*ptr) |
1536 | *ptr = CSUM_MANGLED_0; | |
91bc4822 AS |
1537 | return 0; |
1538 | } | |
1539 | ||
577c50aa | 1540 | static const struct bpf_func_proto bpf_l4_csum_replace_proto = { |
91bc4822 AS |
1541 | .func = bpf_l4_csum_replace, |
1542 | .gpl_only = false, | |
1543 | .ret_type = RET_INTEGER, | |
1544 | .arg1_type = ARG_PTR_TO_CTX, | |
1545 | .arg2_type = ARG_ANYTHING, | |
1546 | .arg3_type = ARG_ANYTHING, | |
1547 | .arg4_type = ARG_ANYTHING, | |
1548 | .arg5_type = ARG_ANYTHING, | |
608cd71a AS |
1549 | }; |
1550 | ||
7d672345 DB |
1551 | static u64 bpf_csum_diff(u64 r1, u64 from_size, u64 r3, u64 to_size, u64 seed) |
1552 | { | |
21cafc1d | 1553 | struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); |
7d672345 DB |
1554 | u64 diff_size = from_size + to_size; |
1555 | __be32 *from = (__be32 *) (long) r1; | |
1556 | __be32 *to = (__be32 *) (long) r3; | |
1557 | int i, j = 0; | |
1558 | ||
1559 | /* This is quite flexible, some examples: | |
1560 | * | |
1561 | * from_size == 0, to_size > 0, seed := csum --> pushing data | |
1562 | * from_size > 0, to_size == 0, seed := csum --> pulling data | |
1563 | * from_size > 0, to_size > 0, seed := 0 --> diffing data | |
1564 | * | |
1565 | * Even for diffing, from_size and to_size don't need to be equal. | |
1566 | */ | |
1567 | if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || | |
1568 | diff_size > sizeof(sp->diff))) | |
1569 | return -EINVAL; | |
1570 | ||
1571 | for (i = 0; i < from_size / sizeof(__be32); i++, j++) | |
1572 | sp->diff[j] = ~from[i]; | |
1573 | for (i = 0; i < to_size / sizeof(__be32); i++, j++) | |
1574 | sp->diff[j] = to[i]; | |
1575 | ||
1576 | return csum_partial(sp->diff, diff_size, seed); | |
1577 | } | |
1578 | ||
577c50aa | 1579 | static const struct bpf_func_proto bpf_csum_diff_proto = { |
7d672345 DB |
1580 | .func = bpf_csum_diff, |
1581 | .gpl_only = false, | |
1582 | .ret_type = RET_INTEGER, | |
1583 | .arg1_type = ARG_PTR_TO_STACK, | |
1584 | .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO, | |
1585 | .arg3_type = ARG_PTR_TO_STACK, | |
1586 | .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO, | |
1587 | .arg5_type = ARG_ANYTHING, | |
1588 | }; | |
1589 | ||
a70b506e DB |
1590 | static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) |
1591 | { | |
a70b506e DB |
1592 | return dev_forward_skb(dev, skb); |
1593 | } | |
1594 | ||
1595 | static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) | |
1596 | { | |
1597 | int ret; | |
1598 | ||
1599 | if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) { | |
1600 | net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); | |
1601 | kfree_skb(skb); | |
1602 | return -ENETDOWN; | |
1603 | } | |
1604 | ||
1605 | skb->dev = dev; | |
1606 | ||
1607 | __this_cpu_inc(xmit_recursion); | |
1608 | ret = dev_queue_xmit(skb); | |
1609 | __this_cpu_dec(xmit_recursion); | |
1610 | ||
1611 | return ret; | |
1612 | } | |
1613 | ||
3896d655 AS |
1614 | static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5) |
1615 | { | |
a70b506e | 1616 | struct sk_buff *skb = (struct sk_buff *) (long) r1; |
3896d655 AS |
1617 | struct net_device *dev; |
1618 | ||
781c53bc DB |
1619 | if (unlikely(flags & ~(BPF_F_INGRESS))) |
1620 | return -EINVAL; | |
1621 | ||
3896d655 AS |
1622 | dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); |
1623 | if (unlikely(!dev)) | |
1624 | return -EINVAL; | |
1625 | ||
a70b506e DB |
1626 | skb = skb_clone(skb, GFP_ATOMIC); |
1627 | if (unlikely(!skb)) | |
3896d655 AS |
1628 | return -ENOMEM; |
1629 | ||
a2bfe6bf DB |
1630 | bpf_push_mac_rcsum(skb); |
1631 | ||
a70b506e DB |
1632 | return flags & BPF_F_INGRESS ? |
1633 | __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); | |
3896d655 AS |
1634 | } |
1635 | ||
577c50aa | 1636 | static const struct bpf_func_proto bpf_clone_redirect_proto = { |
3896d655 AS |
1637 | .func = bpf_clone_redirect, |
1638 | .gpl_only = false, | |
1639 | .ret_type = RET_INTEGER, | |
1640 | .arg1_type = ARG_PTR_TO_CTX, | |
1641 | .arg2_type = ARG_ANYTHING, | |
1642 | .arg3_type = ARG_ANYTHING, | |
1643 | }; | |
1644 | ||
27b29f63 AS |
1645 | struct redirect_info { |
1646 | u32 ifindex; | |
1647 | u32 flags; | |
1648 | }; | |
1649 | ||
1650 | static DEFINE_PER_CPU(struct redirect_info, redirect_info); | |
781c53bc | 1651 | |
27b29f63 AS |
1652 | static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5) |
1653 | { | |
1654 | struct redirect_info *ri = this_cpu_ptr(&redirect_info); | |
1655 | ||
781c53bc DB |
1656 | if (unlikely(flags & ~(BPF_F_INGRESS))) |
1657 | return TC_ACT_SHOT; | |
1658 | ||
27b29f63 AS |
1659 | ri->ifindex = ifindex; |
1660 | ri->flags = flags; | |
781c53bc | 1661 | |
27b29f63 AS |
1662 | return TC_ACT_REDIRECT; |
1663 | } | |
1664 | ||
1665 | int skb_do_redirect(struct sk_buff *skb) | |
1666 | { | |
1667 | struct redirect_info *ri = this_cpu_ptr(&redirect_info); | |
1668 | struct net_device *dev; | |
1669 | ||
1670 | dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex); | |
1671 | ri->ifindex = 0; | |
1672 | if (unlikely(!dev)) { | |
1673 | kfree_skb(skb); | |
1674 | return -EINVAL; | |
1675 | } | |
1676 | ||
a2bfe6bf DB |
1677 | bpf_push_mac_rcsum(skb); |
1678 | ||
a70b506e DB |
1679 | return ri->flags & BPF_F_INGRESS ? |
1680 | __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); | |
27b29f63 AS |
1681 | } |
1682 | ||
577c50aa | 1683 | static const struct bpf_func_proto bpf_redirect_proto = { |
27b29f63 AS |
1684 | .func = bpf_redirect, |
1685 | .gpl_only = false, | |
1686 | .ret_type = RET_INTEGER, | |
1687 | .arg1_type = ARG_ANYTHING, | |
1688 | .arg2_type = ARG_ANYTHING, | |
1689 | }; | |
1690 | ||
8d20aabe DB |
1691 | static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1692 | { | |
1693 | return task_get_classid((struct sk_buff *) (unsigned long) r1); | |
1694 | } | |
1695 | ||
1696 | static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { | |
1697 | .func = bpf_get_cgroup_classid, | |
1698 | .gpl_only = false, | |
1699 | .ret_type = RET_INTEGER, | |
1700 | .arg1_type = ARG_PTR_TO_CTX, | |
1701 | }; | |
1702 | ||
c46646d0 DB |
1703 | static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1704 | { | |
808c1b69 | 1705 | return dst_tclassid((struct sk_buff *) (unsigned long) r1); |
c46646d0 DB |
1706 | } |
1707 | ||
1708 | static const struct bpf_func_proto bpf_get_route_realm_proto = { | |
1709 | .func = bpf_get_route_realm, | |
1710 | .gpl_only = false, | |
1711 | .ret_type = RET_INTEGER, | |
1712 | .arg1_type = ARG_PTR_TO_CTX, | |
1713 | }; | |
1714 | ||
13c5c240 DB |
1715 | static u64 bpf_get_hash_recalc(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1716 | { | |
1717 | /* If skb_clear_hash() was called due to mangling, we can | |
1718 | * trigger SW recalculation here. Later access to hash | |
1719 | * can then use the inline skb->hash via context directly | |
1720 | * instead of calling this helper again. | |
1721 | */ | |
1722 | return skb_get_hash((struct sk_buff *) (unsigned long) r1); | |
1723 | } | |
1724 | ||
1725 | static const struct bpf_func_proto bpf_get_hash_recalc_proto = { | |
1726 | .func = bpf_get_hash_recalc, | |
1727 | .gpl_only = false, | |
1728 | .ret_type = RET_INTEGER, | |
1729 | .arg1_type = ARG_PTR_TO_CTX, | |
1730 | }; | |
1731 | ||
4e10df9a AS |
1732 | static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5) |
1733 | { | |
1734 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1735 | __be16 vlan_proto = (__force __be16) r2; | |
db58ba45 | 1736 | int ret; |
4e10df9a AS |
1737 | |
1738 | if (unlikely(vlan_proto != htons(ETH_P_8021Q) && | |
1739 | vlan_proto != htons(ETH_P_8021AD))) | |
1740 | vlan_proto = htons(ETH_P_8021Q); | |
1741 | ||
8065694e | 1742 | bpf_push_mac_rcsum(skb); |
db58ba45 | 1743 | ret = skb_vlan_push(skb, vlan_proto, vlan_tci); |
8065694e DB |
1744 | bpf_pull_mac_rcsum(skb); |
1745 | ||
db58ba45 AS |
1746 | bpf_compute_data_end(skb); |
1747 | return ret; | |
4e10df9a AS |
1748 | } |
1749 | ||
1750 | const struct bpf_func_proto bpf_skb_vlan_push_proto = { | |
1751 | .func = bpf_skb_vlan_push, | |
1752 | .gpl_only = false, | |
1753 | .ret_type = RET_INTEGER, | |
1754 | .arg1_type = ARG_PTR_TO_CTX, | |
1755 | .arg2_type = ARG_ANYTHING, | |
1756 | .arg3_type = ARG_ANYTHING, | |
1757 | }; | |
4d9c5c53 | 1758 | EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto); |
4e10df9a AS |
1759 | |
1760 | static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
1761 | { | |
1762 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
db58ba45 | 1763 | int ret; |
4e10df9a | 1764 | |
8065694e | 1765 | bpf_push_mac_rcsum(skb); |
db58ba45 | 1766 | ret = skb_vlan_pop(skb); |
8065694e DB |
1767 | bpf_pull_mac_rcsum(skb); |
1768 | ||
db58ba45 AS |
1769 | bpf_compute_data_end(skb); |
1770 | return ret; | |
4e10df9a AS |
1771 | } |
1772 | ||
1773 | const struct bpf_func_proto bpf_skb_vlan_pop_proto = { | |
1774 | .func = bpf_skb_vlan_pop, | |
1775 | .gpl_only = false, | |
1776 | .ret_type = RET_INTEGER, | |
1777 | .arg1_type = ARG_PTR_TO_CTX, | |
1778 | }; | |
4d9c5c53 | 1779 | EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto); |
4e10df9a | 1780 | |
6578171a DB |
1781 | static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) |
1782 | { | |
1783 | /* Caller already did skb_cow() with len as headroom, | |
1784 | * so no need to do it here. | |
1785 | */ | |
1786 | skb_push(skb, len); | |
1787 | memmove(skb->data, skb->data + len, off); | |
1788 | memset(skb->data + off, 0, len); | |
1789 | ||
1790 | /* No skb_postpush_rcsum(skb, skb->data + off, len) | |
1791 | * needed here as it does not change the skb->csum | |
1792 | * result for checksum complete when summing over | |
1793 | * zeroed blocks. | |
1794 | */ | |
1795 | return 0; | |
1796 | } | |
1797 | ||
1798 | static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) | |
1799 | { | |
1800 | /* skb_ensure_writable() is not needed here, as we're | |
1801 | * already working on an uncloned skb. | |
1802 | */ | |
1803 | if (unlikely(!pskb_may_pull(skb, off + len))) | |
1804 | return -ENOMEM; | |
1805 | ||
1806 | skb_postpull_rcsum(skb, skb->data + off, len); | |
1807 | memmove(skb->data + len, skb->data, off); | |
1808 | __skb_pull(skb, len); | |
1809 | ||
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) | |
1814 | { | |
1815 | bool trans_same = skb->transport_header == skb->network_header; | |
1816 | int ret; | |
1817 | ||
1818 | /* There's no need for __skb_push()/__skb_pull() pair to | |
1819 | * get to the start of the mac header as we're guaranteed | |
1820 | * to always start from here under eBPF. | |
1821 | */ | |
1822 | ret = bpf_skb_generic_push(skb, off, len); | |
1823 | if (likely(!ret)) { | |
1824 | skb->mac_header -= len; | |
1825 | skb->network_header -= len; | |
1826 | if (trans_same) | |
1827 | skb->transport_header = skb->network_header; | |
1828 | } | |
1829 | ||
1830 | return ret; | |
1831 | } | |
1832 | ||
1833 | static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) | |
1834 | { | |
1835 | bool trans_same = skb->transport_header == skb->network_header; | |
1836 | int ret; | |
1837 | ||
1838 | /* Same here, __skb_push()/__skb_pull() pair not needed. */ | |
1839 | ret = bpf_skb_generic_pop(skb, off, len); | |
1840 | if (likely(!ret)) { | |
1841 | skb->mac_header += len; | |
1842 | skb->network_header += len; | |
1843 | if (trans_same) | |
1844 | skb->transport_header = skb->network_header; | |
1845 | } | |
1846 | ||
1847 | return ret; | |
1848 | } | |
1849 | ||
1850 | static int bpf_skb_proto_4_to_6(struct sk_buff *skb) | |
1851 | { | |
1852 | const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); | |
1853 | u32 off = skb->network_header - skb->mac_header; | |
1854 | int ret; | |
1855 | ||
1856 | ret = skb_cow(skb, len_diff); | |
1857 | if (unlikely(ret < 0)) | |
1858 | return ret; | |
1859 | ||
1860 | ret = bpf_skb_net_hdr_push(skb, off, len_diff); | |
1861 | if (unlikely(ret < 0)) | |
1862 | return ret; | |
1863 | ||
1864 | if (skb_is_gso(skb)) { | |
1865 | /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV4 needs to | |
1866 | * be changed into SKB_GSO_TCPV6. | |
1867 | */ | |
1868 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
1869 | skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4; | |
1870 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6; | |
1871 | } | |
1872 | ||
1873 | /* Due to IPv6 header, MSS needs to be downgraded. */ | |
1874 | skb_shinfo(skb)->gso_size -= len_diff; | |
1875 | /* Header must be checked, and gso_segs recomputed. */ | |
1876 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1877 | skb_shinfo(skb)->gso_segs = 0; | |
1878 | } | |
1879 | ||
1880 | skb->protocol = htons(ETH_P_IPV6); | |
1881 | skb_clear_hash(skb); | |
1882 | ||
1883 | return 0; | |
1884 | } | |
1885 | ||
1886 | static int bpf_skb_proto_6_to_4(struct sk_buff *skb) | |
1887 | { | |
1888 | const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); | |
1889 | u32 off = skb->network_header - skb->mac_header; | |
1890 | int ret; | |
1891 | ||
1892 | ret = skb_unclone(skb, GFP_ATOMIC); | |
1893 | if (unlikely(ret < 0)) | |
1894 | return ret; | |
1895 | ||
1896 | ret = bpf_skb_net_hdr_pop(skb, off, len_diff); | |
1897 | if (unlikely(ret < 0)) | |
1898 | return ret; | |
1899 | ||
1900 | if (skb_is_gso(skb)) { | |
1901 | /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV6 needs to | |
1902 | * be changed into SKB_GSO_TCPV4. | |
1903 | */ | |
1904 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | |
1905 | skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6; | |
1906 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4; | |
1907 | } | |
1908 | ||
1909 | /* Due to IPv4 header, MSS can be upgraded. */ | |
1910 | skb_shinfo(skb)->gso_size += len_diff; | |
1911 | /* Header must be checked, and gso_segs recomputed. */ | |
1912 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1913 | skb_shinfo(skb)->gso_segs = 0; | |
1914 | } | |
1915 | ||
1916 | skb->protocol = htons(ETH_P_IP); | |
1917 | skb_clear_hash(skb); | |
1918 | ||
1919 | return 0; | |
1920 | } | |
1921 | ||
1922 | static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) | |
1923 | { | |
1924 | __be16 from_proto = skb->protocol; | |
1925 | ||
1926 | if (from_proto == htons(ETH_P_IP) && | |
1927 | to_proto == htons(ETH_P_IPV6)) | |
1928 | return bpf_skb_proto_4_to_6(skb); | |
1929 | ||
1930 | if (from_proto == htons(ETH_P_IPV6) && | |
1931 | to_proto == htons(ETH_P_IP)) | |
1932 | return bpf_skb_proto_6_to_4(skb); | |
1933 | ||
1934 | return -ENOTSUPP; | |
1935 | } | |
1936 | ||
1937 | static u64 bpf_skb_change_proto(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5) | |
1938 | { | |
1939 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1940 | __be16 proto = (__force __be16) r2; | |
1941 | int ret; | |
1942 | ||
1943 | if (unlikely(flags)) | |
1944 | return -EINVAL; | |
1945 | ||
1946 | /* General idea is that this helper does the basic groundwork | |
1947 | * needed for changing the protocol, and eBPF program fills the | |
1948 | * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() | |
1949 | * and other helpers, rather than passing a raw buffer here. | |
1950 | * | |
1951 | * The rationale is to keep this minimal and without a need to | |
1952 | * deal with raw packet data. F.e. even if we would pass buffers | |
1953 | * here, the program still needs to call the bpf_lX_csum_replace() | |
1954 | * helpers anyway. Plus, this way we keep also separation of | |
1955 | * concerns, since f.e. bpf_skb_store_bytes() should only take | |
1956 | * care of stores. | |
1957 | * | |
1958 | * Currently, additional options and extension header space are | |
1959 | * not supported, but flags register is reserved so we can adapt | |
1960 | * that. For offloads, we mark packet as dodgy, so that headers | |
1961 | * need to be verified first. | |
1962 | */ | |
1963 | ret = bpf_skb_proto_xlat(skb, proto); | |
1964 | bpf_compute_data_end(skb); | |
1965 | return ret; | |
1966 | } | |
1967 | ||
1968 | static const struct bpf_func_proto bpf_skb_change_proto_proto = { | |
1969 | .func = bpf_skb_change_proto, | |
1970 | .gpl_only = false, | |
1971 | .ret_type = RET_INTEGER, | |
1972 | .arg1_type = ARG_PTR_TO_CTX, | |
1973 | .arg2_type = ARG_ANYTHING, | |
1974 | .arg3_type = ARG_ANYTHING, | |
1975 | }; | |
1976 | ||
d2485c42 DB |
1977 | static u64 bpf_skb_change_type(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
1978 | { | |
1979 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
1980 | u32 pkt_type = r2; | |
1981 | ||
1982 | /* We only allow a restricted subset to be changed for now. */ | |
45c7fffa DB |
1983 | if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || |
1984 | !skb_pkt_type_ok(pkt_type))) | |
d2485c42 DB |
1985 | return -EINVAL; |
1986 | ||
1987 | skb->pkt_type = pkt_type; | |
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | static const struct bpf_func_proto bpf_skb_change_type_proto = { | |
1992 | .func = bpf_skb_change_type, | |
1993 | .gpl_only = false, | |
1994 | .ret_type = RET_INTEGER, | |
1995 | .arg1_type = ARG_PTR_TO_CTX, | |
1996 | .arg2_type = ARG_ANYTHING, | |
1997 | }; | |
1998 | ||
5293efe6 DB |
1999 | static u32 __bpf_skb_min_len(const struct sk_buff *skb) |
2000 | { | |
2001 | u32 min_len = skb_network_offset(skb); | |
2002 | ||
2003 | if (skb_transport_header_was_set(skb)) | |
2004 | min_len = skb_transport_offset(skb); | |
2005 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
2006 | min_len = skb_checksum_start_offset(skb) + | |
2007 | skb->csum_offset + sizeof(__sum16); | |
2008 | return min_len; | |
2009 | } | |
2010 | ||
2011 | static u32 __bpf_skb_max_len(const struct sk_buff *skb) | |
2012 | { | |
6088b582 | 2013 | return skb->dev->mtu + skb->dev->hard_header_len; |
5293efe6 DB |
2014 | } |
2015 | ||
2016 | static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) | |
2017 | { | |
2018 | unsigned int old_len = skb->len; | |
2019 | int ret; | |
2020 | ||
2021 | ret = __skb_grow_rcsum(skb, new_len); | |
2022 | if (!ret) | |
2023 | memset(skb->data + old_len, 0, new_len - old_len); | |
2024 | return ret; | |
2025 | } | |
2026 | ||
2027 | static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) | |
2028 | { | |
2029 | return __skb_trim_rcsum(skb, new_len); | |
2030 | } | |
2031 | ||
2032 | static u64 bpf_skb_change_tail(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5) | |
2033 | { | |
2034 | struct sk_buff *skb = (struct sk_buff *)(long) r1; | |
2035 | u32 max_len = __bpf_skb_max_len(skb); | |
2036 | u32 min_len = __bpf_skb_min_len(skb); | |
2037 | u32 new_len = (u32) r2; | |
2038 | int ret; | |
2039 | ||
2040 | if (unlikely(flags || new_len > max_len || new_len < min_len)) | |
2041 | return -EINVAL; | |
2042 | if (skb->encapsulation) | |
2043 | return -ENOTSUPP; | |
2044 | ||
2045 | /* The basic idea of this helper is that it's performing the | |
2046 | * needed work to either grow or trim an skb, and eBPF program | |
2047 | * rewrites the rest via helpers like bpf_skb_store_bytes(), | |
2048 | * bpf_lX_csum_replace() and others rather than passing a raw | |
2049 | * buffer here. This one is a slow path helper and intended | |
2050 | * for replies with control messages. | |
2051 | * | |
2052 | * Like in bpf_skb_change_proto(), we want to keep this rather | |
2053 | * minimal and without protocol specifics so that we are able | |
2054 | * to separate concerns as in bpf_skb_store_bytes() should only | |
2055 | * be the one responsible for writing buffers. | |
2056 | * | |
2057 | * It's really expected to be a slow path operation here for | |
2058 | * control message replies, so we're implicitly linearizing, | |
2059 | * uncloning and drop offloads from the skb by this. | |
2060 | */ | |
2061 | ret = __bpf_try_make_writable(skb, skb->len); | |
2062 | if (!ret) { | |
2063 | if (new_len > skb->len) | |
2064 | ret = bpf_skb_grow_rcsum(skb, new_len); | |
2065 | else if (new_len < skb->len) | |
2066 | ret = bpf_skb_trim_rcsum(skb, new_len); | |
2067 | if (!ret && skb_is_gso(skb)) | |
2068 | skb_gso_reset(skb); | |
2069 | } | |
2070 | ||
2071 | bpf_compute_data_end(skb); | |
2072 | return ret; | |
2073 | } | |
2074 | ||
2075 | static const struct bpf_func_proto bpf_skb_change_tail_proto = { | |
2076 | .func = bpf_skb_change_tail, | |
2077 | .gpl_only = false, | |
2078 | .ret_type = RET_INTEGER, | |
2079 | .arg1_type = ARG_PTR_TO_CTX, | |
2080 | .arg2_type = ARG_ANYTHING, | |
2081 | .arg3_type = ARG_ANYTHING, | |
2082 | }; | |
2083 | ||
4e10df9a AS |
2084 | bool bpf_helper_changes_skb_data(void *func) |
2085 | { | |
2086 | if (func == bpf_skb_vlan_push) | |
2087 | return true; | |
2088 | if (func == bpf_skb_vlan_pop) | |
2089 | return true; | |
3697649f DB |
2090 | if (func == bpf_skb_store_bytes) |
2091 | return true; | |
6578171a DB |
2092 | if (func == bpf_skb_change_proto) |
2093 | return true; | |
5293efe6 DB |
2094 | if (func == bpf_skb_change_tail) |
2095 | return true; | |
3697649f DB |
2096 | if (func == bpf_l3_csum_replace) |
2097 | return true; | |
2098 | if (func == bpf_l4_csum_replace) | |
2099 | return true; | |
2100 | ||
4e10df9a AS |
2101 | return false; |
2102 | } | |
2103 | ||
555c8a86 | 2104 | static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, |
aa7145c1 | 2105 | unsigned long off, unsigned long len) |
555c8a86 | 2106 | { |
aa7145c1 | 2107 | void *ptr = skb_header_pointer(skb, off, len, dst_buff); |
555c8a86 DB |
2108 | |
2109 | if (unlikely(!ptr)) | |
2110 | return len; | |
2111 | if (ptr != dst_buff) | |
2112 | memcpy(dst_buff, ptr, len); | |
2113 | ||
2114 | return 0; | |
2115 | } | |
2116 | ||
2117 | static u64 bpf_skb_event_output(u64 r1, u64 r2, u64 flags, u64 r4, | |
2118 | u64 meta_size) | |
2119 | { | |
2120 | struct sk_buff *skb = (struct sk_buff *)(long) r1; | |
2121 | struct bpf_map *map = (struct bpf_map *)(long) r2; | |
2122 | u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; | |
2123 | void *meta = (void *)(long) r4; | |
2124 | ||
2125 | if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) | |
2126 | return -EINVAL; | |
2127 | if (unlikely(skb_size > skb->len)) | |
2128 | return -EFAULT; | |
2129 | ||
2130 | return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, | |
2131 | bpf_skb_copy); | |
2132 | } | |
2133 | ||
2134 | static const struct bpf_func_proto bpf_skb_event_output_proto = { | |
2135 | .func = bpf_skb_event_output, | |
2136 | .gpl_only = true, | |
2137 | .ret_type = RET_INTEGER, | |
2138 | .arg1_type = ARG_PTR_TO_CTX, | |
2139 | .arg2_type = ARG_CONST_MAP_PTR, | |
2140 | .arg3_type = ARG_ANYTHING, | |
2141 | .arg4_type = ARG_PTR_TO_STACK, | |
2142 | .arg5_type = ARG_CONST_STACK_SIZE, | |
2143 | }; | |
2144 | ||
c6c33454 DB |
2145 | static unsigned short bpf_tunnel_key_af(u64 flags) |
2146 | { | |
2147 | return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; | |
2148 | } | |
2149 | ||
d3aa45ce AS |
2150 | static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5) |
2151 | { | |
2152 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
2153 | struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2; | |
c6c33454 DB |
2154 | const struct ip_tunnel_info *info = skb_tunnel_info(skb); |
2155 | u8 compat[sizeof(struct bpf_tunnel_key)]; | |
074f528e DB |
2156 | void *to_orig = to; |
2157 | int err; | |
d3aa45ce | 2158 | |
074f528e DB |
2159 | if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { |
2160 | err = -EINVAL; | |
2161 | goto err_clear; | |
2162 | } | |
2163 | if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { | |
2164 | err = -EPROTO; | |
2165 | goto err_clear; | |
2166 | } | |
c6c33454 | 2167 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) { |
074f528e | 2168 | err = -EINVAL; |
c6c33454 | 2169 | switch (size) { |
4018ab18 | 2170 | case offsetof(struct bpf_tunnel_key, tunnel_label): |
c0e760c9 | 2171 | case offsetof(struct bpf_tunnel_key, tunnel_ext): |
4018ab18 | 2172 | goto set_compat; |
c6c33454 DB |
2173 | case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): |
2174 | /* Fixup deprecated structure layouts here, so we have | |
2175 | * a common path later on. | |
2176 | */ | |
2177 | if (ip_tunnel_info_af(info) != AF_INET) | |
074f528e | 2178 | goto err_clear; |
4018ab18 | 2179 | set_compat: |
c6c33454 DB |
2180 | to = (struct bpf_tunnel_key *)compat; |
2181 | break; | |
2182 | default: | |
074f528e | 2183 | goto err_clear; |
c6c33454 DB |
2184 | } |
2185 | } | |
d3aa45ce AS |
2186 | |
2187 | to->tunnel_id = be64_to_cpu(info->key.tun_id); | |
c6c33454 DB |
2188 | to->tunnel_tos = info->key.tos; |
2189 | to->tunnel_ttl = info->key.ttl; | |
2190 | ||
4018ab18 | 2191 | if (flags & BPF_F_TUNINFO_IPV6) { |
c6c33454 DB |
2192 | memcpy(to->remote_ipv6, &info->key.u.ipv6.src, |
2193 | sizeof(to->remote_ipv6)); | |
4018ab18 DB |
2194 | to->tunnel_label = be32_to_cpu(info->key.label); |
2195 | } else { | |
c6c33454 | 2196 | to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); |
4018ab18 | 2197 | } |
c6c33454 DB |
2198 | |
2199 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) | |
074f528e | 2200 | memcpy(to_orig, to, size); |
d3aa45ce AS |
2201 | |
2202 | return 0; | |
074f528e DB |
2203 | err_clear: |
2204 | memset(to_orig, 0, size); | |
2205 | return err; | |
d3aa45ce AS |
2206 | } |
2207 | ||
577c50aa | 2208 | static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { |
d3aa45ce AS |
2209 | .func = bpf_skb_get_tunnel_key, |
2210 | .gpl_only = false, | |
2211 | .ret_type = RET_INTEGER, | |
2212 | .arg1_type = ARG_PTR_TO_CTX, | |
074f528e | 2213 | .arg2_type = ARG_PTR_TO_RAW_STACK, |
d3aa45ce AS |
2214 | .arg3_type = ARG_CONST_STACK_SIZE, |
2215 | .arg4_type = ARG_ANYTHING, | |
2216 | }; | |
2217 | ||
14ca0751 DB |
2218 | static u64 bpf_skb_get_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5) |
2219 | { | |
2220 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
2221 | u8 *to = (u8 *) (long) r2; | |
2222 | const struct ip_tunnel_info *info = skb_tunnel_info(skb); | |
074f528e | 2223 | int err; |
14ca0751 DB |
2224 | |
2225 | if (unlikely(!info || | |
074f528e DB |
2226 | !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { |
2227 | err = -ENOENT; | |
2228 | goto err_clear; | |
2229 | } | |
2230 | if (unlikely(size < info->options_len)) { | |
2231 | err = -ENOMEM; | |
2232 | goto err_clear; | |
2233 | } | |
14ca0751 DB |
2234 | |
2235 | ip_tunnel_info_opts_get(to, info); | |
074f528e DB |
2236 | if (size > info->options_len) |
2237 | memset(to + info->options_len, 0, size - info->options_len); | |
14ca0751 DB |
2238 | |
2239 | return info->options_len; | |
074f528e DB |
2240 | err_clear: |
2241 | memset(to, 0, size); | |
2242 | return err; | |
14ca0751 DB |
2243 | } |
2244 | ||
2245 | static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { | |
2246 | .func = bpf_skb_get_tunnel_opt, | |
2247 | .gpl_only = false, | |
2248 | .ret_type = RET_INTEGER, | |
2249 | .arg1_type = ARG_PTR_TO_CTX, | |
074f528e | 2250 | .arg2_type = ARG_PTR_TO_RAW_STACK, |
14ca0751 DB |
2251 | .arg3_type = ARG_CONST_STACK_SIZE, |
2252 | }; | |
2253 | ||
d3aa45ce AS |
2254 | static struct metadata_dst __percpu *md_dst; |
2255 | ||
2256 | static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5) | |
2257 | { | |
2258 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
2259 | struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2; | |
2260 | struct metadata_dst *md = this_cpu_ptr(md_dst); | |
c6c33454 | 2261 | u8 compat[sizeof(struct bpf_tunnel_key)]; |
d3aa45ce AS |
2262 | struct ip_tunnel_info *info; |
2263 | ||
22080870 DB |
2264 | if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | |
2265 | BPF_F_DONT_FRAGMENT))) | |
d3aa45ce | 2266 | return -EINVAL; |
c6c33454 DB |
2267 | if (unlikely(size != sizeof(struct bpf_tunnel_key))) { |
2268 | switch (size) { | |
4018ab18 | 2269 | case offsetof(struct bpf_tunnel_key, tunnel_label): |
c0e760c9 | 2270 | case offsetof(struct bpf_tunnel_key, tunnel_ext): |
c6c33454 DB |
2271 | case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): |
2272 | /* Fixup deprecated structure layouts here, so we have | |
2273 | * a common path later on. | |
2274 | */ | |
2275 | memcpy(compat, from, size); | |
2276 | memset(compat + size, 0, sizeof(compat) - size); | |
2277 | from = (struct bpf_tunnel_key *)compat; | |
2278 | break; | |
2279 | default: | |
2280 | return -EINVAL; | |
2281 | } | |
2282 | } | |
c0e760c9 DB |
2283 | if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || |
2284 | from->tunnel_ext)) | |
4018ab18 | 2285 | return -EINVAL; |
d3aa45ce AS |
2286 | |
2287 | skb_dst_drop(skb); | |
2288 | dst_hold((struct dst_entry *) md); | |
2289 | skb_dst_set(skb, (struct dst_entry *) md); | |
2290 | ||
2291 | info = &md->u.tun_info; | |
2292 | info->mode = IP_TUNNEL_INFO_TX; | |
c6c33454 | 2293 | |
db3c6139 | 2294 | info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; |
22080870 DB |
2295 | if (flags & BPF_F_DONT_FRAGMENT) |
2296 | info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; | |
2297 | ||
d3aa45ce | 2298 | info->key.tun_id = cpu_to_be64(from->tunnel_id); |
c6c33454 DB |
2299 | info->key.tos = from->tunnel_tos; |
2300 | info->key.ttl = from->tunnel_ttl; | |
2301 | ||
2302 | if (flags & BPF_F_TUNINFO_IPV6) { | |
2303 | info->mode |= IP_TUNNEL_INFO_IPV6; | |
2304 | memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, | |
2305 | sizeof(from->remote_ipv6)); | |
4018ab18 DB |
2306 | info->key.label = cpu_to_be32(from->tunnel_label) & |
2307 | IPV6_FLOWLABEL_MASK; | |
c6c33454 DB |
2308 | } else { |
2309 | info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); | |
2da897e5 DB |
2310 | if (flags & BPF_F_ZERO_CSUM_TX) |
2311 | info->key.tun_flags &= ~TUNNEL_CSUM; | |
c6c33454 | 2312 | } |
d3aa45ce AS |
2313 | |
2314 | return 0; | |
2315 | } | |
2316 | ||
577c50aa | 2317 | static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { |
d3aa45ce AS |
2318 | .func = bpf_skb_set_tunnel_key, |
2319 | .gpl_only = false, | |
2320 | .ret_type = RET_INTEGER, | |
2321 | .arg1_type = ARG_PTR_TO_CTX, | |
2322 | .arg2_type = ARG_PTR_TO_STACK, | |
2323 | .arg3_type = ARG_CONST_STACK_SIZE, | |
2324 | .arg4_type = ARG_ANYTHING, | |
2325 | }; | |
2326 | ||
14ca0751 DB |
2327 | static u64 bpf_skb_set_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5) |
2328 | { | |
2329 | struct sk_buff *skb = (struct sk_buff *) (long) r1; | |
2330 | u8 *from = (u8 *) (long) r2; | |
2331 | struct ip_tunnel_info *info = skb_tunnel_info(skb); | |
2332 | const struct metadata_dst *md = this_cpu_ptr(md_dst); | |
2333 | ||
2334 | if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) | |
2335 | return -EINVAL; | |
fca5fdf6 | 2336 | if (unlikely(size > IP_TUNNEL_OPTS_MAX)) |
14ca0751 DB |
2337 | return -ENOMEM; |
2338 | ||
2339 | ip_tunnel_info_opts_set(info, from, size); | |
2340 | ||
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { | |
2345 | .func = bpf_skb_set_tunnel_opt, | |
2346 | .gpl_only = false, | |
2347 | .ret_type = RET_INTEGER, | |
2348 | .arg1_type = ARG_PTR_TO_CTX, | |
2349 | .arg2_type = ARG_PTR_TO_STACK, | |
2350 | .arg3_type = ARG_CONST_STACK_SIZE, | |
2351 | }; | |
2352 | ||
2353 | static const struct bpf_func_proto * | |
2354 | bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) | |
d3aa45ce AS |
2355 | { |
2356 | if (!md_dst) { | |
14ca0751 DB |
2357 | /* Race is not possible, since it's called from verifier |
2358 | * that is holding verifier mutex. | |
d3aa45ce | 2359 | */ |
fca5fdf6 | 2360 | md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, |
14ca0751 | 2361 | GFP_KERNEL); |
d3aa45ce AS |
2362 | if (!md_dst) |
2363 | return NULL; | |
2364 | } | |
14ca0751 DB |
2365 | |
2366 | switch (which) { | |
2367 | case BPF_FUNC_skb_set_tunnel_key: | |
2368 | return &bpf_skb_set_tunnel_key_proto; | |
2369 | case BPF_FUNC_skb_set_tunnel_opt: | |
2370 | return &bpf_skb_set_tunnel_opt_proto; | |
2371 | default: | |
2372 | return NULL; | |
2373 | } | |
d3aa45ce AS |
2374 | } |
2375 | ||
747ea55e | 2376 | static u64 bpf_skb_under_cgroup(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
4a482f34 MKL |
2377 | { |
2378 | struct sk_buff *skb = (struct sk_buff *)(long)r1; | |
2379 | struct bpf_map *map = (struct bpf_map *)(long)r2; | |
2380 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
2381 | struct cgroup *cgrp; | |
2382 | struct sock *sk; | |
2383 | u32 i = (u32)r3; | |
2384 | ||
2385 | sk = skb->sk; | |
2386 | if (!sk || !sk_fullsock(sk)) | |
2387 | return -ENOENT; | |
2388 | ||
2389 | if (unlikely(i >= array->map.max_entries)) | |
2390 | return -E2BIG; | |
2391 | ||
2392 | cgrp = READ_ONCE(array->ptrs[i]); | |
2393 | if (unlikely(!cgrp)) | |
2394 | return -EAGAIN; | |
2395 | ||
54fd9c2d | 2396 | return sk_under_cgroup_hierarchy(sk, cgrp); |
4a482f34 MKL |
2397 | } |
2398 | ||
747ea55e DB |
2399 | static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { |
2400 | .func = bpf_skb_under_cgroup, | |
4a482f34 MKL |
2401 | .gpl_only = false, |
2402 | .ret_type = RET_INTEGER, | |
2403 | .arg1_type = ARG_PTR_TO_CTX, | |
2404 | .arg2_type = ARG_CONST_MAP_PTR, | |
2405 | .arg3_type = ARG_ANYTHING, | |
2406 | }; | |
4a482f34 | 2407 | |
4de16969 DB |
2408 | static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, |
2409 | unsigned long off, unsigned long len) | |
2410 | { | |
2411 | memcpy(dst_buff, src_buff + off, len); | |
2412 | return 0; | |
2413 | } | |
2414 | ||
2415 | static u64 bpf_xdp_event_output(u64 r1, u64 r2, u64 flags, u64 r4, | |
2416 | u64 meta_size) | |
2417 | { | |
2418 | struct xdp_buff *xdp = (struct xdp_buff *)(long) r1; | |
2419 | struct bpf_map *map = (struct bpf_map *)(long) r2; | |
2420 | u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; | |
2421 | void *meta = (void *)(long) r4; | |
2422 | ||
2423 | if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) | |
2424 | return -EINVAL; | |
2425 | if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data))) | |
2426 | return -EFAULT; | |
2427 | ||
2428 | return bpf_event_output(map, flags, meta, meta_size, xdp, xdp_size, | |
2429 | bpf_xdp_copy); | |
2430 | } | |
2431 | ||
2432 | static const struct bpf_func_proto bpf_xdp_event_output_proto = { | |
2433 | .func = bpf_xdp_event_output, | |
2434 | .gpl_only = true, | |
2435 | .ret_type = RET_INTEGER, | |
2436 | .arg1_type = ARG_PTR_TO_CTX, | |
2437 | .arg2_type = ARG_CONST_MAP_PTR, | |
2438 | .arg3_type = ARG_ANYTHING, | |
2439 | .arg4_type = ARG_PTR_TO_STACK, | |
2440 | .arg5_type = ARG_CONST_STACK_SIZE, | |
2441 | }; | |
2442 | ||
d4052c4a DB |
2443 | static const struct bpf_func_proto * |
2444 | sk_filter_func_proto(enum bpf_func_id func_id) | |
89aa0758 AS |
2445 | { |
2446 | switch (func_id) { | |
2447 | case BPF_FUNC_map_lookup_elem: | |
2448 | return &bpf_map_lookup_elem_proto; | |
2449 | case BPF_FUNC_map_update_elem: | |
2450 | return &bpf_map_update_elem_proto; | |
2451 | case BPF_FUNC_map_delete_elem: | |
2452 | return &bpf_map_delete_elem_proto; | |
03e69b50 DB |
2453 | case BPF_FUNC_get_prandom_u32: |
2454 | return &bpf_get_prandom_u32_proto; | |
c04167ce | 2455 | case BPF_FUNC_get_smp_processor_id: |
80b48c44 | 2456 | return &bpf_get_raw_smp_processor_id_proto; |
04fd61ab AS |
2457 | case BPF_FUNC_tail_call: |
2458 | return &bpf_tail_call_proto; | |
17ca8cbf DB |
2459 | case BPF_FUNC_ktime_get_ns: |
2460 | return &bpf_ktime_get_ns_proto; | |
0756ea3e | 2461 | case BPF_FUNC_trace_printk: |
1be7f75d AS |
2462 | if (capable(CAP_SYS_ADMIN)) |
2463 | return bpf_get_trace_printk_proto(); | |
89aa0758 AS |
2464 | default: |
2465 | return NULL; | |
2466 | } | |
2467 | } | |
2468 | ||
608cd71a AS |
2469 | static const struct bpf_func_proto * |
2470 | tc_cls_act_func_proto(enum bpf_func_id func_id) | |
2471 | { | |
2472 | switch (func_id) { | |
2473 | case BPF_FUNC_skb_store_bytes: | |
2474 | return &bpf_skb_store_bytes_proto; | |
05c74e5e DB |
2475 | case BPF_FUNC_skb_load_bytes: |
2476 | return &bpf_skb_load_bytes_proto; | |
7d672345 DB |
2477 | case BPF_FUNC_csum_diff: |
2478 | return &bpf_csum_diff_proto; | |
91bc4822 AS |
2479 | case BPF_FUNC_l3_csum_replace: |
2480 | return &bpf_l3_csum_replace_proto; | |
2481 | case BPF_FUNC_l4_csum_replace: | |
2482 | return &bpf_l4_csum_replace_proto; | |
3896d655 AS |
2483 | case BPF_FUNC_clone_redirect: |
2484 | return &bpf_clone_redirect_proto; | |
8d20aabe DB |
2485 | case BPF_FUNC_get_cgroup_classid: |
2486 | return &bpf_get_cgroup_classid_proto; | |
4e10df9a AS |
2487 | case BPF_FUNC_skb_vlan_push: |
2488 | return &bpf_skb_vlan_push_proto; | |
2489 | case BPF_FUNC_skb_vlan_pop: | |
2490 | return &bpf_skb_vlan_pop_proto; | |
6578171a DB |
2491 | case BPF_FUNC_skb_change_proto: |
2492 | return &bpf_skb_change_proto_proto; | |
d2485c42 DB |
2493 | case BPF_FUNC_skb_change_type: |
2494 | return &bpf_skb_change_type_proto; | |
5293efe6 DB |
2495 | case BPF_FUNC_skb_change_tail: |
2496 | return &bpf_skb_change_tail_proto; | |
d3aa45ce AS |
2497 | case BPF_FUNC_skb_get_tunnel_key: |
2498 | return &bpf_skb_get_tunnel_key_proto; | |
2499 | case BPF_FUNC_skb_set_tunnel_key: | |
14ca0751 DB |
2500 | return bpf_get_skb_set_tunnel_proto(func_id); |
2501 | case BPF_FUNC_skb_get_tunnel_opt: | |
2502 | return &bpf_skb_get_tunnel_opt_proto; | |
2503 | case BPF_FUNC_skb_set_tunnel_opt: | |
2504 | return bpf_get_skb_set_tunnel_proto(func_id); | |
27b29f63 AS |
2505 | case BPF_FUNC_redirect: |
2506 | return &bpf_redirect_proto; | |
c46646d0 DB |
2507 | case BPF_FUNC_get_route_realm: |
2508 | return &bpf_get_route_realm_proto; | |
13c5c240 DB |
2509 | case BPF_FUNC_get_hash_recalc: |
2510 | return &bpf_get_hash_recalc_proto; | |
bd570ff9 | 2511 | case BPF_FUNC_perf_event_output: |
555c8a86 | 2512 | return &bpf_skb_event_output_proto; |
80b48c44 DB |
2513 | case BPF_FUNC_get_smp_processor_id: |
2514 | return &bpf_get_smp_processor_id_proto; | |
747ea55e DB |
2515 | case BPF_FUNC_skb_under_cgroup: |
2516 | return &bpf_skb_under_cgroup_proto; | |
608cd71a AS |
2517 | default: |
2518 | return sk_filter_func_proto(func_id); | |
2519 | } | |
2520 | } | |
2521 | ||
6a773a15 BB |
2522 | static const struct bpf_func_proto * |
2523 | xdp_func_proto(enum bpf_func_id func_id) | |
2524 | { | |
4de16969 DB |
2525 | switch (func_id) { |
2526 | case BPF_FUNC_perf_event_output: | |
2527 | return &bpf_xdp_event_output_proto; | |
2528 | default: | |
2529 | return sk_filter_func_proto(func_id); | |
2530 | } | |
6a773a15 BB |
2531 | } |
2532 | ||
d691f9e8 | 2533 | static bool __is_valid_access(int off, int size, enum bpf_access_type type) |
89aa0758 | 2534 | { |
9bac3d6d AS |
2535 | if (off < 0 || off >= sizeof(struct __sk_buff)) |
2536 | return false; | |
4936e352 | 2537 | /* The verifier guarantees that size > 0. */ |
9bac3d6d AS |
2538 | if (off % size != 0) |
2539 | return false; | |
4936e352 | 2540 | if (size != sizeof(__u32)) |
9bac3d6d AS |
2541 | return false; |
2542 | ||
2543 | return true; | |
2544 | } | |
2545 | ||
d691f9e8 | 2546 | static bool sk_filter_is_valid_access(int off, int size, |
19de99f7 AS |
2547 | enum bpf_access_type type, |
2548 | enum bpf_reg_type *reg_type) | |
d691f9e8 | 2549 | { |
db58ba45 AS |
2550 | switch (off) { |
2551 | case offsetof(struct __sk_buff, tc_classid): | |
2552 | case offsetof(struct __sk_buff, data): | |
2553 | case offsetof(struct __sk_buff, data_end): | |
045efa82 | 2554 | return false; |
db58ba45 | 2555 | } |
045efa82 | 2556 | |
d691f9e8 AS |
2557 | if (type == BPF_WRITE) { |
2558 | switch (off) { | |
2559 | case offsetof(struct __sk_buff, cb[0]) ... | |
4936e352 | 2560 | offsetof(struct __sk_buff, cb[4]): |
d691f9e8 AS |
2561 | break; |
2562 | default: | |
2563 | return false; | |
2564 | } | |
2565 | } | |
2566 | ||
2567 | return __is_valid_access(off, size, type); | |
2568 | } | |
2569 | ||
2570 | static bool tc_cls_act_is_valid_access(int off, int size, | |
19de99f7 AS |
2571 | enum bpf_access_type type, |
2572 | enum bpf_reg_type *reg_type) | |
d691f9e8 AS |
2573 | { |
2574 | if (type == BPF_WRITE) { | |
2575 | switch (off) { | |
2576 | case offsetof(struct __sk_buff, mark): | |
2577 | case offsetof(struct __sk_buff, tc_index): | |
754f1e6a | 2578 | case offsetof(struct __sk_buff, priority): |
d691f9e8 | 2579 | case offsetof(struct __sk_buff, cb[0]) ... |
09c37a2c DB |
2580 | offsetof(struct __sk_buff, cb[4]): |
2581 | case offsetof(struct __sk_buff, tc_classid): | |
d691f9e8 AS |
2582 | break; |
2583 | default: | |
2584 | return false; | |
2585 | } | |
2586 | } | |
19de99f7 AS |
2587 | |
2588 | switch (off) { | |
2589 | case offsetof(struct __sk_buff, data): | |
2590 | *reg_type = PTR_TO_PACKET; | |
2591 | break; | |
2592 | case offsetof(struct __sk_buff, data_end): | |
2593 | *reg_type = PTR_TO_PACKET_END; | |
2594 | break; | |
2595 | } | |
2596 | ||
d691f9e8 AS |
2597 | return __is_valid_access(off, size, type); |
2598 | } | |
2599 | ||
6a773a15 BB |
2600 | static bool __is_valid_xdp_access(int off, int size, |
2601 | enum bpf_access_type type) | |
2602 | { | |
2603 | if (off < 0 || off >= sizeof(struct xdp_md)) | |
2604 | return false; | |
2605 | if (off % size != 0) | |
2606 | return false; | |
6088b582 | 2607 | if (size != sizeof(__u32)) |
6a773a15 BB |
2608 | return false; |
2609 | ||
2610 | return true; | |
2611 | } | |
2612 | ||
2613 | static bool xdp_is_valid_access(int off, int size, | |
2614 | enum bpf_access_type type, | |
2615 | enum bpf_reg_type *reg_type) | |
2616 | { | |
2617 | if (type == BPF_WRITE) | |
2618 | return false; | |
2619 | ||
2620 | switch (off) { | |
2621 | case offsetof(struct xdp_md, data): | |
2622 | *reg_type = PTR_TO_PACKET; | |
2623 | break; | |
2624 | case offsetof(struct xdp_md, data_end): | |
2625 | *reg_type = PTR_TO_PACKET_END; | |
2626 | break; | |
2627 | } | |
2628 | ||
2629 | return __is_valid_xdp_access(off, size, type); | |
2630 | } | |
2631 | ||
2632 | void bpf_warn_invalid_xdp_action(u32 act) | |
2633 | { | |
2634 | WARN_ONCE(1, "Illegal XDP return value %u, expect packet loss\n", act); | |
2635 | } | |
2636 | EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); | |
2637 | ||
d691f9e8 AS |
2638 | static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg, |
2639 | int src_reg, int ctx_off, | |
ff936a04 AS |
2640 | struct bpf_insn *insn_buf, |
2641 | struct bpf_prog *prog) | |
9bac3d6d AS |
2642 | { |
2643 | struct bpf_insn *insn = insn_buf; | |
2644 | ||
2645 | switch (ctx_off) { | |
2646 | case offsetof(struct __sk_buff, len): | |
2647 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4); | |
2648 | ||
2649 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2650 | offsetof(struct sk_buff, len)); | |
2651 | break; | |
2652 | ||
0b8c707d DB |
2653 | case offsetof(struct __sk_buff, protocol): |
2654 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); | |
2655 | ||
2656 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2657 | offsetof(struct sk_buff, protocol)); | |
2658 | break; | |
2659 | ||
27cd5452 MS |
2660 | case offsetof(struct __sk_buff, vlan_proto): |
2661 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); | |
2662 | ||
2663 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2664 | offsetof(struct sk_buff, vlan_proto)); | |
2665 | break; | |
2666 | ||
bcad5718 DB |
2667 | case offsetof(struct __sk_buff, priority): |
2668 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4); | |
2669 | ||
754f1e6a DB |
2670 | if (type == BPF_WRITE) |
2671 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, | |
2672 | offsetof(struct sk_buff, priority)); | |
2673 | else | |
2674 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2675 | offsetof(struct sk_buff, priority)); | |
bcad5718 DB |
2676 | break; |
2677 | ||
37e82c2f AS |
2678 | case offsetof(struct __sk_buff, ingress_ifindex): |
2679 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4); | |
2680 | ||
2681 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2682 | offsetof(struct sk_buff, skb_iif)); | |
2683 | break; | |
2684 | ||
2685 | case offsetof(struct __sk_buff, ifindex): | |
2686 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); | |
2687 | ||
2688 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)), | |
2689 | dst_reg, src_reg, | |
2690 | offsetof(struct sk_buff, dev)); | |
2691 | *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1); | |
2692 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg, | |
2693 | offsetof(struct net_device, ifindex)); | |
2694 | break; | |
2695 | ||
ba7591d8 DB |
2696 | case offsetof(struct __sk_buff, hash): |
2697 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
2698 | ||
2699 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2700 | offsetof(struct sk_buff, hash)); | |
2701 | break; | |
2702 | ||
9bac3d6d | 2703 | case offsetof(struct __sk_buff, mark): |
d691f9e8 AS |
2704 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); |
2705 | ||
2706 | if (type == BPF_WRITE) | |
2707 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, | |
2708 | offsetof(struct sk_buff, mark)); | |
2709 | else | |
2710 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, | |
2711 | offsetof(struct sk_buff, mark)); | |
2712 | break; | |
9bac3d6d AS |
2713 | |
2714 | case offsetof(struct __sk_buff, pkt_type): | |
2715 | return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn); | |
2716 | ||
2717 | case offsetof(struct __sk_buff, queue_mapping): | |
2718 | return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn); | |
c2497395 | 2719 | |
c2497395 AS |
2720 | case offsetof(struct __sk_buff, vlan_present): |
2721 | return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, | |
2722 | dst_reg, src_reg, insn); | |
2723 | ||
2724 | case offsetof(struct __sk_buff, vlan_tci): | |
2725 | return convert_skb_access(SKF_AD_VLAN_TAG, | |
2726 | dst_reg, src_reg, insn); | |
d691f9e8 AS |
2727 | |
2728 | case offsetof(struct __sk_buff, cb[0]) ... | |
6088b582 | 2729 | offsetof(struct __sk_buff, cb[4]): |
d691f9e8 AS |
2730 | BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20); |
2731 | ||
ff936a04 | 2732 | prog->cb_access = 1; |
d691f9e8 AS |
2733 | ctx_off -= offsetof(struct __sk_buff, cb[0]); |
2734 | ctx_off += offsetof(struct sk_buff, cb); | |
2735 | ctx_off += offsetof(struct qdisc_skb_cb, data); | |
2736 | if (type == BPF_WRITE) | |
2737 | *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off); | |
2738 | else | |
2739 | *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off); | |
2740 | break; | |
2741 | ||
045efa82 DB |
2742 | case offsetof(struct __sk_buff, tc_classid): |
2743 | ctx_off -= offsetof(struct __sk_buff, tc_classid); | |
2744 | ctx_off += offsetof(struct sk_buff, cb); | |
2745 | ctx_off += offsetof(struct qdisc_skb_cb, tc_classid); | |
09c37a2c DB |
2746 | if (type == BPF_WRITE) |
2747 | *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off); | |
2748 | else | |
2749 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, ctx_off); | |
045efa82 DB |
2750 | break; |
2751 | ||
db58ba45 AS |
2752 | case offsetof(struct __sk_buff, data): |
2753 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, data)), | |
2754 | dst_reg, src_reg, | |
2755 | offsetof(struct sk_buff, data)); | |
2756 | break; | |
2757 | ||
2758 | case offsetof(struct __sk_buff, data_end): | |
2759 | ctx_off -= offsetof(struct __sk_buff, data_end); | |
2760 | ctx_off += offsetof(struct sk_buff, cb); | |
2761 | ctx_off += offsetof(struct bpf_skb_data_end, data_end); | |
2762 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(sizeof(void *)), | |
2763 | dst_reg, src_reg, ctx_off); | |
2764 | break; | |
2765 | ||
d691f9e8 AS |
2766 | case offsetof(struct __sk_buff, tc_index): |
2767 | #ifdef CONFIG_NET_SCHED | |
2768 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2); | |
2769 | ||
2770 | if (type == BPF_WRITE) | |
2771 | *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, | |
2772 | offsetof(struct sk_buff, tc_index)); | |
2773 | else | |
2774 | *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, | |
2775 | offsetof(struct sk_buff, tc_index)); | |
2776 | break; | |
2777 | #else | |
2778 | if (type == BPF_WRITE) | |
2779 | *insn++ = BPF_MOV64_REG(dst_reg, dst_reg); | |
2780 | else | |
2781 | *insn++ = BPF_MOV64_IMM(dst_reg, 0); | |
2782 | break; | |
2783 | #endif | |
9bac3d6d AS |
2784 | } |
2785 | ||
2786 | return insn - insn_buf; | |
89aa0758 AS |
2787 | } |
2788 | ||
6a773a15 BB |
2789 | static u32 xdp_convert_ctx_access(enum bpf_access_type type, int dst_reg, |
2790 | int src_reg, int ctx_off, | |
2791 | struct bpf_insn *insn_buf, | |
2792 | struct bpf_prog *prog) | |
2793 | { | |
2794 | struct bpf_insn *insn = insn_buf; | |
2795 | ||
2796 | switch (ctx_off) { | |
2797 | case offsetof(struct xdp_md, data): | |
2798 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct xdp_buff, data)), | |
2799 | dst_reg, src_reg, | |
2800 | offsetof(struct xdp_buff, data)); | |
2801 | break; | |
2802 | case offsetof(struct xdp_md, data_end): | |
2803 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct xdp_buff, data_end)), | |
2804 | dst_reg, src_reg, | |
2805 | offsetof(struct xdp_buff, data_end)); | |
2806 | break; | |
2807 | } | |
2808 | ||
2809 | return insn - insn_buf; | |
2810 | } | |
2811 | ||
d4052c4a | 2812 | static const struct bpf_verifier_ops sk_filter_ops = { |
4936e352 DB |
2813 | .get_func_proto = sk_filter_func_proto, |
2814 | .is_valid_access = sk_filter_is_valid_access, | |
2815 | .convert_ctx_access = bpf_net_convert_ctx_access, | |
89aa0758 AS |
2816 | }; |
2817 | ||
608cd71a | 2818 | static const struct bpf_verifier_ops tc_cls_act_ops = { |
4936e352 DB |
2819 | .get_func_proto = tc_cls_act_func_proto, |
2820 | .is_valid_access = tc_cls_act_is_valid_access, | |
2821 | .convert_ctx_access = bpf_net_convert_ctx_access, | |
608cd71a AS |
2822 | }; |
2823 | ||
6a773a15 BB |
2824 | static const struct bpf_verifier_ops xdp_ops = { |
2825 | .get_func_proto = xdp_func_proto, | |
2826 | .is_valid_access = xdp_is_valid_access, | |
2827 | .convert_ctx_access = xdp_convert_ctx_access, | |
2828 | }; | |
2829 | ||
d4052c4a | 2830 | static struct bpf_prog_type_list sk_filter_type __read_mostly = { |
4936e352 DB |
2831 | .ops = &sk_filter_ops, |
2832 | .type = BPF_PROG_TYPE_SOCKET_FILTER, | |
89aa0758 AS |
2833 | }; |
2834 | ||
96be4325 | 2835 | static struct bpf_prog_type_list sched_cls_type __read_mostly = { |
4936e352 DB |
2836 | .ops = &tc_cls_act_ops, |
2837 | .type = BPF_PROG_TYPE_SCHED_CLS, | |
96be4325 DB |
2838 | }; |
2839 | ||
94caee8c | 2840 | static struct bpf_prog_type_list sched_act_type __read_mostly = { |
4936e352 DB |
2841 | .ops = &tc_cls_act_ops, |
2842 | .type = BPF_PROG_TYPE_SCHED_ACT, | |
94caee8c DB |
2843 | }; |
2844 | ||
6a773a15 BB |
2845 | static struct bpf_prog_type_list xdp_type __read_mostly = { |
2846 | .ops = &xdp_ops, | |
2847 | .type = BPF_PROG_TYPE_XDP, | |
2848 | }; | |
2849 | ||
d4052c4a | 2850 | static int __init register_sk_filter_ops(void) |
89aa0758 | 2851 | { |
d4052c4a | 2852 | bpf_register_prog_type(&sk_filter_type); |
96be4325 | 2853 | bpf_register_prog_type(&sched_cls_type); |
94caee8c | 2854 | bpf_register_prog_type(&sched_act_type); |
6a773a15 | 2855 | bpf_register_prog_type(&xdp_type); |
96be4325 | 2856 | |
89aa0758 AS |
2857 | return 0; |
2858 | } | |
d4052c4a DB |
2859 | late_initcall(register_sk_filter_ops); |
2860 | ||
8ced425e | 2861 | int sk_detach_filter(struct sock *sk) |
55b33325 PE |
2862 | { |
2863 | int ret = -ENOENT; | |
2864 | struct sk_filter *filter; | |
2865 | ||
d59577b6 VB |
2866 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
2867 | return -EPERM; | |
2868 | ||
8ced425e HFS |
2869 | filter = rcu_dereference_protected(sk->sk_filter, |
2870 | lockdep_sock_is_held(sk)); | |
55b33325 | 2871 | if (filter) { |
a9b3cd7f | 2872 | RCU_INIT_POINTER(sk->sk_filter, NULL); |
46bcf14f | 2873 | sk_filter_uncharge(sk, filter); |
55b33325 PE |
2874 | ret = 0; |
2875 | } | |
a3ea269b | 2876 | |
55b33325 PE |
2877 | return ret; |
2878 | } | |
8ced425e | 2879 | EXPORT_SYMBOL_GPL(sk_detach_filter); |
a8fc9277 | 2880 | |
a3ea269b DB |
2881 | int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, |
2882 | unsigned int len) | |
a8fc9277 | 2883 | { |
a3ea269b | 2884 | struct sock_fprog_kern *fprog; |
a8fc9277 | 2885 | struct sk_filter *filter; |
a3ea269b | 2886 | int ret = 0; |
a8fc9277 PE |
2887 | |
2888 | lock_sock(sk); | |
2889 | filter = rcu_dereference_protected(sk->sk_filter, | |
8ced425e | 2890 | lockdep_sock_is_held(sk)); |
a8fc9277 PE |
2891 | if (!filter) |
2892 | goto out; | |
a3ea269b DB |
2893 | |
2894 | /* We're copying the filter that has been originally attached, | |
93d08b69 DB |
2895 | * so no conversion/decode needed anymore. eBPF programs that |
2896 | * have no original program cannot be dumped through this. | |
a3ea269b | 2897 | */ |
93d08b69 | 2898 | ret = -EACCES; |
7ae457c1 | 2899 | fprog = filter->prog->orig_prog; |
93d08b69 DB |
2900 | if (!fprog) |
2901 | goto out; | |
a3ea269b DB |
2902 | |
2903 | ret = fprog->len; | |
a8fc9277 | 2904 | if (!len) |
a3ea269b | 2905 | /* User space only enquires number of filter blocks. */ |
a8fc9277 | 2906 | goto out; |
a3ea269b | 2907 | |
a8fc9277 | 2908 | ret = -EINVAL; |
a3ea269b | 2909 | if (len < fprog->len) |
a8fc9277 PE |
2910 | goto out; |
2911 | ||
2912 | ret = -EFAULT; | |
009937e7 | 2913 | if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) |
a3ea269b | 2914 | goto out; |
a8fc9277 | 2915 | |
a3ea269b DB |
2916 | /* Instead of bytes, the API requests to return the number |
2917 | * of filter blocks. | |
2918 | */ | |
2919 | ret = fprog->len; | |
a8fc9277 PE |
2920 | out: |
2921 | release_sock(sk); | |
2922 | return ret; | |
2923 | } |