]> git.proxmox.com Git - ovs.git/blame - datapath/flow_netlink.c
projects / ovs.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tunnel: Add support for matching on OAM packets.
[ovs.git] / datapath / flow_netlink.c
CommitLineData
a097c0b2
PS		 1/*
2 * Copyright (c) 2007-2013 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
0a0857df
JP		 19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
a097c0b2
PS		 21#include "flow.h"
22#include "datapath.h"
23#include <linux/uaccess.h>
24#include <linux/netdevice.h>
25#include <linux/etherdevice.h>
26#include <linux/if_ether.h>
27#include <linux/if_vlan.h>
28#include <net/llc_pdu.h>
29#include <linux/kernel.h>
30#include <linux/jhash.h>
31#include <linux/jiffies.h>
32#include <linux/llc.h>
33#include <linux/module.h>
34#include <linux/in.h>
35#include <linux/rcupdate.h>
36#include <linux/if_arp.h>
37#include <linux/ip.h>
38#include <linux/ipv6.h>
39#include <linux/sctp.h>
40#include <linux/tcp.h>
41#include <linux/udp.h>
42#include <linux/icmp.h>
43#include <linux/icmpv6.h>
44#include <linux/rculist.h>
45#include <net/ip.h>
29c71cfa 46#include <net/ip_tunnels.h>
a097c0b2
PS		 47#include <net/ipv6.h>
48#include <net/ndisc.h>
49
50#include "flow_netlink.h"
51
52static void update_range__(struct sw_flow_match *match,
53 size_t offset, size_t size, bool is_mask)
54{
55 struct sw_flow_key_range *range = NULL;
56 size_t start = rounddown(offset, sizeof(long));
57 size_t end = roundup(offset + size, sizeof(long));
58
59 if (!is_mask)
60 range = &match->range;
61 else if (match->mask)
62 range = &match->mask->range;
63
64 if (!range)
65 return;
66
67 if (range->start == range->end) {
68 range->start = start;
69 range->end = end;
70 return;
71 }
72
73 if (range->start > start)
74 range->start = start;
75
76 if (range->end < end)
77 range->end = end;
78}
79
80#define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
81 do { \
82 update_range__(match, offsetof(struct sw_flow_key, field), \
83 sizeof((match)->key->field), is_mask); \
84 if (is_mask) { \
85 if ((match)->mask) \
86 (match)->mask->key.field = value; \
87 } else { \
88 (match)->key->field = value; \
89 } \
90 } while (0)
91
92#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
93 do { \
94 update_range__(match, offsetof(struct sw_flow_key, field), \
95 len, is_mask); \
96 if (is_mask) { \
97 if ((match)->mask) \
98 memcpy(&(match)->mask->key.field, value_p, len);\
99 } else { \
100 memcpy(&(match)->key->field, value_p, len); \
101 } \
102 } while (0)
103
104static u16 range_n_bytes(const struct sw_flow_key_range *range)
105{
106 return range->end - range->start;
107}
108
109static bool match_validate(const struct sw_flow_match *match,
110 u64 key_attrs, u64 mask_attrs)
111{
112 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
113 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
114
115 /* The following mask attributes allowed only if they
116 * pass the validation tests. */
117 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
118 | (1ULL << OVS_KEY_ATTR_IPV6)
119 | (1ULL << OVS_KEY_ATTR_TCP)
dc235f7f 120 | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
a097c0b2
PS		 121 | (1ULL << OVS_KEY_ATTR_UDP)
122 | (1ULL << OVS_KEY_ATTR_SCTP)
123 | (1ULL << OVS_KEY_ATTR_ICMP)
124 | (1ULL << OVS_KEY_ATTR_ICMPV6)
125 | (1ULL << OVS_KEY_ATTR_ARP)
126 | (1ULL << OVS_KEY_ATTR_ND));
127
128 /* Always allowed mask fields. */
129 mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
130 | (1ULL << OVS_KEY_ATTR_IN_PORT)
131 | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
132
133 /* Check key attributes. */
134 if (match->key->eth.type == htons(ETH_P_ARP)
135 || match->key->eth.type == htons(ETH_P_RARP)) {
136 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
137 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
138 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
139 }
140
141 if (match->key->eth.type == htons(ETH_P_IP)) {
142 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
143 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
144 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
145
146 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
147 if (match->key->ip.proto == IPPROTO_UDP) {
148 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
149 if (match->mask && (match->mask->key.ip.proto == 0xff))
150 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
151 }
152
153 if (match->key->ip.proto == IPPROTO_SCTP) {
154 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
155 if (match->mask && (match->mask->key.ip.proto == 0xff))
156 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
157 }
158
159 if (match->key->ip.proto == IPPROTO_TCP) {
160 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 161 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
162 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 163 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 164 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
165 }
a097c0b2
PS		 166 }
167
168 if (match->key->ip.proto == IPPROTO_ICMP) {
169 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
170 if (match->mask && (match->mask->key.ip.proto == 0xff))
171 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
172 }
173 }
174 }
175
176 if (match->key->eth.type == htons(ETH_P_IPV6)) {
177 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
178 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
179 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
180
181 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
182 if (match->key->ip.proto == IPPROTO_UDP) {
183 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
184 if (match->mask && (match->mask->key.ip.proto == 0xff))
185 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
186 }
187
188 if (match->key->ip.proto == IPPROTO_SCTP) {
189 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
190 if (match->mask && (match->mask->key.ip.proto == 0xff))
191 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
192 }
193
194 if (match->key->ip.proto == IPPROTO_TCP) {
195 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 196 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
197 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 198 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR		 199 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
200 }
a097c0b2
PS		 201 }
202
203 if (match->key->ip.proto == IPPROTO_ICMPV6) {
204 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
205 if (match->mask && (match->mask->key.ip.proto == 0xff))
206 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
207
708fb4c5 208 if (match->key->tp.src ==
a097c0b2 209 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
708fb4c5 210 match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
a097c0b2 211 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
708fb4c5 212 if (match->mask && (match->mask->key.tp.src == htons(0xffff)))
a097c0b2
PS		 213 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
214 }
215 }
216 }
217 }
218
219 if ((key_attrs & key_expected) != key_expected) {
220 /* Key attributes check failed. */
221 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
29dd5cb7 222 (unsigned long long)key_attrs, (unsigned long long)key_expected);
a097c0b2
PS		 223 return false;
224 }
225
226 if ((mask_attrs & mask_allowed) != mask_attrs) {
227 /* Mask attributes check failed. */
228 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
29dd5cb7 229 (unsigned long long)mask_attrs, (unsigned long long)mask_allowed);
a097c0b2
PS		 230 return false;
231 }
232
233 return true;
234}
235
236/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
237static const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
238 [OVS_KEY_ATTR_ENCAP] = -1,
239 [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
240 [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
241 [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
242 [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
243 [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
244 [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
245 [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
246 [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
247 [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
dc235f7f 248 [OVS_KEY_ATTR_TCP_FLAGS] = sizeof(__be16),
a097c0b2
PS		 249 [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
250 [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
251 [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
252 [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
253 [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
254 [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
7804df20 255 [OVS_KEY_ATTR_DP_HASH] = sizeof(u32),
a6059080 256 [OVS_KEY_ATTR_RECIRC_ID] = sizeof(u32),
a097c0b2
PS		 257 [OVS_KEY_ATTR_TUNNEL] = -1,
258};
259
260static bool is_all_zero(const u8 *fp, size_t size)
261{
262 int i;
263
264 if (!fp)
265 return false;
266
267 for (i = 0; i < size; i++)
268 if (fp[i])
269 return false;
270
271 return true;
272}
273
274static int __parse_flow_nlattrs(const struct nlattr *attr,
275 const struct nlattr *a[],
276 u64 *attrsp, bool nz)
277{
278 const struct nlattr *nla;
279 u64 attrs;
280 int rem;
281
282 attrs = *attrsp;
283 nla_for_each_nested(nla, attr, rem) {
284 u16 type = nla_type(nla);
285 int expected_len;
286
287 if (type > OVS_KEY_ATTR_MAX) {
288 OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
289 type, OVS_KEY_ATTR_MAX);
290 return -EINVAL;
291 }
292
293 if (attrs & (1ULL << type)) {
294 OVS_NLERR("Duplicate key attribute (type %d).\n", type);
295 return -EINVAL;
296 }
297
298 expected_len = ovs_key_lens[type];
299 if (nla_len(nla) != expected_len && expected_len != -1) {
300 OVS_NLERR("Key attribute has unexpected length (type=%d"
301 ", length=%d, expected=%d).\n", type,
302 nla_len(nla), expected_len);
303 return -EINVAL;
304 }
305
306 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
307 attrs |= 1ULL << type;
308 a[type] = nla;
309 }
310 }
311 if (rem) {
312 OVS_NLERR("Message has %d unknown bytes.\n", rem);
313 return -EINVAL;
314 }
315
316 *attrsp = attrs;
317 return 0;
318}
319
320static int parse_flow_mask_nlattrs(const struct nlattr *attr,
321 const struct nlattr *a[], u64 *attrsp)
322{
323 return __parse_flow_nlattrs(attr, a, attrsp, true);
324}
325
326static int parse_flow_nlattrs(const struct nlattr *attr,
327 const struct nlattr *a[], u64 *attrsp)
328{
329 return __parse_flow_nlattrs(attr, a, attrsp, false);
330}
331
332static int ipv4_tun_from_nlattr(const struct nlattr *attr,
333 struct sw_flow_match *match, bool is_mask)
334{
335 struct nlattr *a;
336 int rem;
337 bool ttl = false;
338 __be16 tun_flags = 0;
339
340 nla_for_each_nested(a, attr, rem) {
341 int type = nla_type(a);
342 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
343 [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
344 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
345 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
346 [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
347 [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
348 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
349 [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
94872594 350 [OVS_TUNNEL_KEY_ATTR_OAM] = 0,
a097c0b2
PS		 351 };
352
353 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
354 OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
355 type, OVS_TUNNEL_KEY_ATTR_MAX);
356 return -EINVAL;
357 }
358
359 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
360 OVS_NLERR("IPv4 tunnel attribute type has unexpected "
361 " length (type=%d, length=%d, expected=%d).\n",
362 type, nla_len(a), ovs_tunnel_key_lens[type]);
363 return -EINVAL;
364 }
365
366 switch (type) {
367 case OVS_TUNNEL_KEY_ATTR_ID:
368 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
369 nla_get_be64(a), is_mask);
370 tun_flags |= TUNNEL_KEY;
371 break;
372 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
373 SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
374 nla_get_be32(a), is_mask);
375 break;
376 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
377 SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
378 nla_get_be32(a), is_mask);
379 break;
380 case OVS_TUNNEL_KEY_ATTR_TOS:
381 SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
382 nla_get_u8(a), is_mask);
383 break;
384 case OVS_TUNNEL_KEY_ATTR_TTL:
385 SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
386 nla_get_u8(a), is_mask);
387 ttl = true;
388 break;
389 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
390 tun_flags |= TUNNEL_DONT_FRAGMENT;
391 break;
392 case OVS_TUNNEL_KEY_ATTR_CSUM:
393 tun_flags |= TUNNEL_CSUM;
394 break;
94872594
JG		 395 case OVS_TUNNEL_KEY_ATTR_OAM:
396 tun_flags |= TUNNEL_OAM;
397 break;
a097c0b2
PS		 398 default:
399 return -EINVAL;
400 }
401 }
402
403 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
404
405 if (rem > 0) {
406 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
407 return -EINVAL;
408 }
409
410 if (!is_mask) {
411 if (!match->key->tun_key.ipv4_dst) {
412 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
413 return -EINVAL;
414 }
415
416 if (!ttl) {
417 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
418 return -EINVAL;
419 }
420 }
421
422 return 0;
423}
424
425static int ipv4_tun_to_nlattr(struct sk_buff *skb,
426 const struct ovs_key_ipv4_tunnel *tun_key,
427 const struct ovs_key_ipv4_tunnel *output)
428{
429 struct nlattr *nla;
430
431 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
432 if (!nla)
433 return -EMSGSIZE;
434
435 if (output->tun_flags & TUNNEL_KEY &&
436 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
437 return -EMSGSIZE;
438 if (output->ipv4_src &&
439 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
440 return -EMSGSIZE;
441 if (output->ipv4_dst &&
442 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
443 return -EMSGSIZE;
444 if (output->ipv4_tos &&
445 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
446 return -EMSGSIZE;
447 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
448 return -EMSGSIZE;
449 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
450 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
451 return -EMSGSIZE;
452 if ((output->tun_flags & TUNNEL_CSUM) &&
453 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
454 return -EMSGSIZE;
94872594
JG		 455 if ((output->tun_flags & TUNNEL_OAM) &&
456 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
457 return -EMSGSIZE;
a097c0b2
PS		 458
459 nla_nest_end(skb, nla);
460 return 0;
461}
462
463
464static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
465 const struct nlattr **a, bool is_mask)
466{
7804df20
AZ		 467 if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
468 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
469
470 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
471 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
472 }
473
a6059080
AZ		 474 if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
475 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
476
477 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
478 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
479 }
480
a097c0b2
PS		 481 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
482 SW_FLOW_KEY_PUT(match, phy.priority,
483 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
484 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
485 }
486
487 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
488 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
489
490 if (is_mask)
491 in_port = 0xffffffff; /* Always exact match in_port. */
492 else if (in_port >= DP_MAX_PORTS)
493 return -EINVAL;
494
495 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
496 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
497 } else if (!is_mask) {
498 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
499 }
500
501 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
502 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
503
504 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
505 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
506 }
507 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
508 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
509 is_mask))
510 return -EINVAL;
511 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
512 }
513 return 0;
514}
515
df65fec1
JR		 516static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
517 const struct nlattr **a, bool is_mask)
a097c0b2
PS		 518{
519 int err;
520 u64 orig_attrs = attrs;
521
522 err = metadata_from_nlattrs(match, &attrs, a, is_mask);
523 if (err)
524 return err;
525
526 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
527 const struct ovs_key_ethernet *eth_key;
528
529 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
530 SW_FLOW_KEY_MEMCPY(match, eth.src,
531 eth_key->eth_src, ETH_ALEN, is_mask);
532 SW_FLOW_KEY_MEMCPY(match, eth.dst,
533 eth_key->eth_dst, ETH_ALEN, is_mask);
534 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
535 }
536
537 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
538 __be16 tci;
539
540 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
541 if (!(tci & htons(VLAN_TAG_PRESENT))) {
542 if (is_mask)
543 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
544 else
545 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
546
547 return -EINVAL;
548 }
549
550 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
551 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
552 } else if (!is_mask)
553 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
554
555 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
556 __be16 eth_type;
557
558 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
559 if (is_mask) {
560 /* Always exact match EtherType. */
561 eth_type = htons(0xffff);
562 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
563 OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
564 ntohs(eth_type), ETH_P_802_3_MIN);
565 return -EINVAL;
566 }
567
568 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
569 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
570 } else if (!is_mask) {
571 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
572 }
573
574 if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
575 const struct ovs_key_ipv4 *ipv4_key;
576
577 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
578 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
579 OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
580 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
581 return -EINVAL;
582 }
583 SW_FLOW_KEY_PUT(match, ip.proto,
584 ipv4_key->ipv4_proto, is_mask);
585 SW_FLOW_KEY_PUT(match, ip.tos,
586 ipv4_key->ipv4_tos, is_mask);
587 SW_FLOW_KEY_PUT(match, ip.ttl,
588 ipv4_key->ipv4_ttl, is_mask);
589 SW_FLOW_KEY_PUT(match, ip.frag,
590 ipv4_key->ipv4_frag, is_mask);
591 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
592 ipv4_key->ipv4_src, is_mask);
593 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
594 ipv4_key->ipv4_dst, is_mask);
595 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
596 }
597
598 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
599 const struct ovs_key_ipv6 *ipv6_key;
600
601 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
602 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
603 OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
604 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
605 return -EINVAL;
606 }
607 SW_FLOW_KEY_PUT(match, ipv6.label,
608 ipv6_key->ipv6_label, is_mask);
609 SW_FLOW_KEY_PUT(match, ip.proto,
610 ipv6_key->ipv6_proto, is_mask);
611 SW_FLOW_KEY_PUT(match, ip.tos,
612 ipv6_key->ipv6_tclass, is_mask);
613 SW_FLOW_KEY_PUT(match, ip.ttl,
614 ipv6_key->ipv6_hlimit, is_mask);
615 SW_FLOW_KEY_PUT(match, ip.frag,
616 ipv6_key->ipv6_frag, is_mask);
617 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
618 ipv6_key->ipv6_src,
619 sizeof(match->key->ipv6.addr.src),
620 is_mask);
621 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
622 ipv6_key->ipv6_dst,
623 sizeof(match->key->ipv6.addr.dst),
624 is_mask);
625
626 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
627 }
628
629 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
630 const struct ovs_key_arp *arp_key;
631
632 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
633 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
634 OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
635 arp_key->arp_op);
636 return -EINVAL;
637 }
638
639 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
640 arp_key->arp_sip, is_mask);
641 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
642 arp_key->arp_tip, is_mask);
643 SW_FLOW_KEY_PUT(match, ip.proto,
644 ntohs(arp_key->arp_op), is_mask);
645 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
646 arp_key->arp_sha, ETH_ALEN, is_mask);
647 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
648 arp_key->arp_tha, ETH_ALEN, is_mask);
649
650 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
651 }
652
653 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
654 const struct ovs_key_tcp *tcp_key;
655
656 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
708fb4c5
JR		 657 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
658 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
a097c0b2
PS		 659 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
660 }
661
dc235f7f
JR		 662 if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
663 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
708fb4c5 664 SW_FLOW_KEY_PUT(match, tp.flags,
dc235f7f
JR		 665 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
666 is_mask);
667 } else {
708fb4c5 668 SW_FLOW_KEY_PUT(match, tp.flags,
dc235f7f
JR		 669 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
670 is_mask);
671 }
672 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
673 }
674
a097c0b2
PS		 675 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
676 const struct ovs_key_udp *udp_key;
677
678 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
708fb4c5
JR		 679 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
680 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
a097c0b2
PS		 681 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
682 }
683
684 if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
685 const struct ovs_key_sctp *sctp_key;
686
687 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
708fb4c5
JR		 688 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
689 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
a097c0b2
PS		 690 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
691 }
692
693 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
694 const struct ovs_key_icmp *icmp_key;
695
696 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
708fb4c5 697 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 698 htons(icmp_key->icmp_type), is_mask);
708fb4c5 699 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS		 700 htons(icmp_key->icmp_code), is_mask);
701 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
702 }
703
704 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
705 const struct ovs_key_icmpv6 *icmpv6_key;
706
707 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
708fb4c5 708 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 709 htons(icmpv6_key->icmpv6_type), is_mask);
708fb4c5 710 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS		 711 htons(icmpv6_key->icmpv6_code), is_mask);
712 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
713 }
714
715 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
716 const struct ovs_key_nd *nd_key;
717
718 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
719 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
720 nd_key->nd_target,
721 sizeof(match->key->ipv6.nd.target),
722 is_mask);
723 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
724 nd_key->nd_sll, ETH_ALEN, is_mask);
725 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
726 nd_key->nd_tll, ETH_ALEN, is_mask);
727 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
728 }
729
730 if (attrs != 0)
731 return -EINVAL;
732
733 return 0;
734}
735
736static void sw_flow_mask_set(struct sw_flow_mask *mask,
737 struct sw_flow_key_range *range, u8 val)
738{
739 u8 *m = (u8 *)&mask->key + range->start;
740
741 mask->range = *range;
742 memset(m, val, range_n_bytes(range));
743}
744
745/**
746 * ovs_nla_get_match - parses Netlink attributes into a flow key and
747 * mask. In case the 'mask' is NULL, the flow is treated as exact match
748 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
749 * does not include any don't care bit.
750 * @match: receives the extracted flow match information.
751 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
752 * sequence. The fields should of the packet that triggered the creation
753 * of this flow.
754 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
755 * attribute specifies the mask field of the wildcarded flow.
756 */
757int ovs_nla_get_match(struct sw_flow_match *match,
758 const struct nlattr *key,
759 const struct nlattr *mask)
760{
761 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
762 const struct nlattr *encap;
763 u64 key_attrs = 0;
764 u64 mask_attrs = 0;
765 bool encap_valid = false;
766 int err;
767
768 err = parse_flow_nlattrs(key, a, &key_attrs);
769 if (err)
770 return err;
771
772 if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
773 (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
774 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
775 __be16 tci;
776
777 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
778 (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
779 OVS_NLERR("Invalid Vlan frame.\n");
780 return -EINVAL;
781 }
782
783 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
784 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
785 encap = a[OVS_KEY_ATTR_ENCAP];
786 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
787 encap_valid = true;
788
789 if (tci & htons(VLAN_TAG_PRESENT)) {
790 err = parse_flow_nlattrs(encap, a, &key_attrs);
791 if (err)
792 return err;
793 } else if (!tci) {
794 /* Corner case for truncated 802.1Q header. */
795 if (nla_len(encap)) {
796 OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
797 return -EINVAL;
798 }
799 } else {
800 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
801 return -EINVAL;
802 }
803 }
804
df65fec1 805 err = ovs_key_from_nlattrs(match, key_attrs, a, false);
a097c0b2
PS		 806 if (err)
807 return err;
808
809 if (mask) {
810 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
811 if (err)
812 return err;
813
814 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
815 __be16 eth_type = 0;
816 __be16 tci = 0;
817
818 if (!encap_valid) {
819 OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
820 return -EINVAL;
821 }
822
823 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
824 if (a[OVS_KEY_ATTR_ETHERTYPE])
825 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
826
827 if (eth_type == htons(0xffff)) {
828 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
829 encap = a[OVS_KEY_ATTR_ENCAP];
830 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
831 } else {
832 OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
833 ntohs(eth_type));
834 return -EINVAL;
835 }
836
837 if (a[OVS_KEY_ATTR_VLAN])
838 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
839
840 if (!(tci & htons(VLAN_TAG_PRESENT))) {
841 OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
842 return -EINVAL;
843 }
844 }
845
df65fec1 846 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
a097c0b2
PS		 847 if (err)
848 return err;
849 } else {
850 /* Populate exact match flow's key mask. */
851 if (match->mask)
852 sw_flow_mask_set(match->mask, &match->range, 0xff);
853 }
854
855 if (!match_validate(match, key_attrs, mask_attrs))
856 return -EINVAL;
857
858 return 0;
859}
860
861/**
862 * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
863 * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
864 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
865 * sequence.
866 *
867 * This parses a series of Netlink attributes that form a flow key, which must
868 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
869 * get the metadata, that is, the parts of the flow key that cannot be
870 * extracted from the packet itself.
871 */
872
873int ovs_nla_get_flow_metadata(struct sw_flow *flow,
874 const struct nlattr *attr)
875{
876 struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
877 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
878 u64 attrs = 0;
879 int err;
880 struct sw_flow_match match;
881
882 flow->key.phy.in_port = DP_MAX_PORTS;
883 flow->key.phy.priority = 0;
884 flow->key.phy.skb_mark = 0;
7804df20 885 flow->key.ovs_flow_hash = 0;
a6059080 886 flow->key.recirc_id = 0;
a097c0b2
PS		 887 memset(tun_key, 0, sizeof(flow->key.tun_key));
888
889 err = parse_flow_nlattrs(attr, a, &attrs);
890 if (err)
891 return -EINVAL;
892
893 memset(&match, 0, sizeof(match));
894 match.key = &flow->key;
895
896 err = metadata_from_nlattrs(&match, &attrs, a, false);
897 if (err)
898 return err;
899
900 return 0;
901}
902
903int ovs_nla_put_flow(const struct sw_flow_key *swkey,
904 const struct sw_flow_key *output, struct sk_buff *skb)
905{
906 struct ovs_key_ethernet *eth_key;
907 struct nlattr *nla, *encap;
908 bool is_mask = (swkey != output);
909
7804df20
AZ		 910 if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
911 goto nla_put_failure;
912
a6059080
AZ		 913 if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
914 goto nla_put_failure;
915
a097c0b2
PS		 916 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
917 goto nla_put_failure;
918
919 if ((swkey->tun_key.ipv4_dst || is_mask) &&
920 ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
921 goto nla_put_failure;
922
923 if (swkey->phy.in_port == DP_MAX_PORTS) {
924 if (is_mask && (output->phy.in_port == 0xffff))
925 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
926 goto nla_put_failure;
927 } else {
928 u16 upper_u16;
929 upper_u16 = !is_mask ? 0 : 0xffff;
930
931 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
932 (upper_u16 << 16) | output->phy.in_port))
933 goto nla_put_failure;
934 }
935
936 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
937 goto nla_put_failure;
938
939 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
940 if (!nla)
941 goto nla_put_failure;
942
943 eth_key = nla_data(nla);
982a47ec
JP		 944 ether_addr_copy(eth_key->eth_src, output->eth.src);
945 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
a097c0b2
PS		 946
947 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
948 __be16 eth_type;
949 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
950 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
951 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
952 goto nla_put_failure;
953 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
954 if (!swkey->eth.tci)
955 goto unencap;
956 } else
957 encap = NULL;
958
959 if (swkey->eth.type == htons(ETH_P_802_2)) {
960 /*
961 * Ethertype 802.2 is represented in the netlink with omitted
962 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
963 * 0xffff in the mask attribute. Ethertype can also
964 * be wildcarded.
965 */
966 if (is_mask && output->eth.type)
967 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
968 output->eth.type))
969 goto nla_put_failure;
970 goto unencap;
971 }
972
973 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
974 goto nla_put_failure;
975
976 if (swkey->eth.type == htons(ETH_P_IP)) {
977 struct ovs_key_ipv4 *ipv4_key;
978
979 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
980 if (!nla)
981 goto nla_put_failure;
982 ipv4_key = nla_data(nla);
983 ipv4_key->ipv4_src = output->ipv4.addr.src;
984 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
985 ipv4_key->ipv4_proto = output->ip.proto;
986 ipv4_key->ipv4_tos = output->ip.tos;
987 ipv4_key->ipv4_ttl = output->ip.ttl;
988 ipv4_key->ipv4_frag = output->ip.frag;
989 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
990 struct ovs_key_ipv6 *ipv6_key;
991
992 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
993 if (!nla)
994 goto nla_put_failure;
995 ipv6_key = nla_data(nla);
996 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
997 sizeof(ipv6_key->ipv6_src));
998 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
999 sizeof(ipv6_key->ipv6_dst));
1000 ipv6_key->ipv6_label = output->ipv6.label;
1001 ipv6_key->ipv6_proto = output->ip.proto;
1002 ipv6_key->ipv6_tclass = output->ip.tos;
1003 ipv6_key->ipv6_hlimit = output->ip.ttl;
1004 ipv6_key->ipv6_frag = output->ip.frag;
1005 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1006 swkey->eth.type == htons(ETH_P_RARP)) {
1007 struct ovs_key_arp *arp_key;
1008
1009 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1010 if (!nla)
1011 goto nla_put_failure;
1012 arp_key = nla_data(nla);
1013 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1014 arp_key->arp_sip = output->ipv4.addr.src;
1015 arp_key->arp_tip = output->ipv4.addr.dst;
1016 arp_key->arp_op = htons(output->ip.proto);
982a47ec
JP		 1017 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1018 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
a097c0b2
PS		 1019 }
1020
1021 if ((swkey->eth.type == htons(ETH_P_IP) ||
1022 swkey->eth.type == htons(ETH_P_IPV6)) &&
1023 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1024
1025 if (swkey->ip.proto == IPPROTO_TCP) {
1026 struct ovs_key_tcp *tcp_key;
1027
1028 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1029 if (!nla)
1030 goto nla_put_failure;
1031 tcp_key = nla_data(nla);
708fb4c5
JR		 1032 tcp_key->tcp_src = output->tp.src;
1033 tcp_key->tcp_dst = output->tp.dst;
1034 if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1035 output->tp.flags))
1036 goto nla_put_failure;
a097c0b2
PS		 1037 } else if (swkey->ip.proto == IPPROTO_UDP) {
1038 struct ovs_key_udp *udp_key;
1039
1040 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1041 if (!nla)
1042 goto nla_put_failure;
1043 udp_key = nla_data(nla);
708fb4c5
JR		 1044 udp_key->udp_src = output->tp.src;
1045 udp_key->udp_dst = output->tp.dst;
a097c0b2
PS		 1046 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1047 struct ovs_key_sctp *sctp_key;
1048
1049 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1050 if (!nla)
1051 goto nla_put_failure;
1052 sctp_key = nla_data(nla);
708fb4c5
JR		 1053 sctp_key->sctp_src = output->tp.src;
1054 sctp_key->sctp_dst = output->tp.dst;
a097c0b2
PS		 1055 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1056 swkey->ip.proto == IPPROTO_ICMP) {
1057 struct ovs_key_icmp *icmp_key;
1058
1059 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1060 if (!nla)
1061 goto nla_put_failure;
1062 icmp_key = nla_data(nla);
708fb4c5
JR		 1063 icmp_key->icmp_type = ntohs(output->tp.src);
1064 icmp_key->icmp_code = ntohs(output->tp.dst);
a097c0b2
PS		 1065 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1066 swkey->ip.proto == IPPROTO_ICMPV6) {
1067 struct ovs_key_icmpv6 *icmpv6_key;
1068
1069 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1070 sizeof(*icmpv6_key));
1071 if (!nla)
1072 goto nla_put_failure;
1073 icmpv6_key = nla_data(nla);
708fb4c5
JR		 1074 icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1075 icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
a097c0b2
PS		 1076
1077 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1078 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1079 struct ovs_key_nd *nd_key;
1080
1081 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1082 if (!nla)
1083 goto nla_put_failure;
1084 nd_key = nla_data(nla);
1085 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1086 sizeof(nd_key->nd_target));
982a47ec
JP		 1087 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1088 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
a097c0b2
PS		 1089 }
1090 }
1091 }
1092
1093unencap:
1094 if (encap)
1095 nla_nest_end(skb, encap);
1096
1097 return 0;
1098
1099nla_put_failure:
1100 return -EMSGSIZE;
1101}
1102
1103#define MAX_ACTIONS_BUFSIZE (32 * 1024)
1104
1105struct sw_flow_actions *ovs_nla_alloc_flow_actions(int size)
1106{
1107 struct sw_flow_actions *sfa;
1108
1109 if (size > MAX_ACTIONS_BUFSIZE)
1110 return ERR_PTR(-EINVAL);
1111
1112 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1113 if (!sfa)
1114 return ERR_PTR(-ENOMEM);
1115
1116 sfa->actions_len = 0;
1117 return sfa;
1118}
1119
1120/* RCU callback used by ovs_nla_free_flow_actions. */
1121static void rcu_free_acts_callback(struct rcu_head *rcu)
1122{
1123 struct sw_flow_actions *sf_acts = container_of(rcu,
1124 struct sw_flow_actions, rcu);
1125 kfree(sf_acts);
1126}
1127
1128/* Schedules 'sf_acts' to be freed after the next RCU grace period.
1129 * The caller must hold rcu_read_lock for this to be sensible. */
1130void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1131{
1132 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
1133}
1134
1135static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
1136 int attr_len)
1137{
1138
1139 struct sw_flow_actions *acts;
1140 int new_acts_size;
1141 int req_size = NLA_ALIGN(attr_len);
1142 int next_offset = offsetof(struct sw_flow_actions, actions) +
1143 (*sfa)->actions_len;
1144
1145 if (req_size <= (ksize(*sfa) - next_offset))
1146 goto out;
1147
1148 new_acts_size = ksize(*sfa) * 2;
1149
1150 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1151 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1152 return ERR_PTR(-EMSGSIZE);
1153 new_acts_size = MAX_ACTIONS_BUFSIZE;
1154 }
1155
1156 acts = ovs_nla_alloc_flow_actions(new_acts_size);
1157 if (IS_ERR(acts))
1158 return (void *)acts;
1159
1160 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1161 acts->actions_len = (*sfa)->actions_len;
1162 kfree(*sfa);
1163 *sfa = acts;
1164
1165out:
1166 (*sfa)->actions_len += req_size;
1167 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1168}
1169
f0cd669f
JG		 1170static struct nlattr *__add_action(struct sw_flow_actions **sfa, int attrtype,
1171 void *data, int len)
a097c0b2
PS		 1172{
1173 struct nlattr *a;
1174
1175 a = reserve_sfa_size(sfa, nla_attr_size(len));
1176 if (IS_ERR(a))
f0cd669f 1177 return a;
a097c0b2
PS		 1178
1179 a->nla_type = attrtype;
1180 a->nla_len = nla_attr_size(len);
1181
1182 if (data)
1183 memcpy(nla_data(a), data, len);
1184 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1185
f0cd669f
JG		 1186 return a;
1187}
1188
1189static int add_action(struct sw_flow_actions **sfa, int attrtype,
1190 void *data, int len)
1191{
1192 struct nlattr *a;
1193
1194 a = __add_action(sfa, attrtype, data, len);
1195 if (IS_ERR(a))
1196 return PTR_ERR(a);
1197
a097c0b2
PS		 1198 return 0;
1199}
1200
1201static inline int add_nested_action_start(struct sw_flow_actions **sfa,
1202 int attrtype)
1203{
1204 int used = (*sfa)->actions_len;
1205 int err;
1206
1207 err = add_action(sfa, attrtype, NULL, 0);
1208 if (err)
1209 return err;
1210
1211 return used;
1212}
1213
1214static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1215 int st_offset)
1216{
1217 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1218 st_offset);
1219
1220 a->nla_len = sfa->actions_len - st_offset;
1221}
1222
1223static int validate_and_copy_sample(const struct nlattr *attr,
1224 const struct sw_flow_key *key, int depth,
1225 struct sw_flow_actions **sfa)
1226{
1227 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1228 const struct nlattr *probability, *actions;
1229 const struct nlattr *a;
1230 int rem, start, err, st_acts;
1231
1232 memset(attrs, 0, sizeof(attrs));
1233 nla_for_each_nested(a, attr, rem) {
1234 int type = nla_type(a);
1235 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1236 return -EINVAL;
1237 attrs[type] = a;
1238 }
1239 if (rem)
1240 return -EINVAL;
1241
1242 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1243 if (!probability || nla_len(probability) != sizeof(u32))
1244 return -EINVAL;
1245
1246 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1247 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1248 return -EINVAL;
1249
1250 /* validation done, copy sample action. */
1251 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
1252 if (start < 0)
1253 return start;
1254 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1255 nla_data(probability), sizeof(u32));
1256 if (err)
1257 return err;
1258 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
1259 if (st_acts < 0)
1260 return st_acts;
1261
1262 err = ovs_nla_copy_actions(actions, key, depth + 1, sfa);
1263 if (err)
1264 return err;
1265
1266 add_nested_action_end(*sfa, st_acts);
1267 add_nested_action_end(*sfa, start);
1268
1269 return 0;
1270}
1271
1272static int validate_tp_port(const struct sw_flow_key *flow_key)
1273{
708fb4c5
JR		 1274 if ((flow_key->eth.type == htons(ETH_P_IP) ||
1275 flow_key->eth.type == htons(ETH_P_IPV6)) &&
1276 (flow_key->tp.src || flow_key->tp.dst))
1277 return 0;
a097c0b2
PS		 1278
1279 return -EINVAL;
1280}
1281
1282void ovs_match_init(struct sw_flow_match *match,
1283 struct sw_flow_key *key,
1284 struct sw_flow_mask *mask)
1285{
1286 memset(match, 0, sizeof(*match));
1287 match->key = key;
1288 match->mask = mask;
1289
1290 memset(key, 0, sizeof(*key));
1291
1292 if (mask) {
1293 memset(&mask->key, 0, sizeof(mask->key));
1294 mask->range.start = mask->range.end = 0;
1295 }
1296}
1297
1298static int validate_and_copy_set_tun(const struct nlattr *attr,
1299 struct sw_flow_actions **sfa)
1300{
1301 struct sw_flow_match match;
1302 struct sw_flow_key key;
f0cd669f
JG		 1303 struct ovs_tunnel_info *tun_info;
1304 struct nlattr *a;
a097c0b2
PS		 1305 int err, start;
1306
1307 ovs_match_init(&match, &key, NULL);
1308 err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
1309 if (err)
1310 return err;
1311
1312 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
1313 if (start < 0)
1314 return start;
1315
f0cd669f
JG		 1316 a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
1317 sizeof(*tun_info));
1318 if (IS_ERR(a))
1319 return PTR_ERR(a);
1320
1321 tun_info = nla_data(a);
1322 tun_info->tunnel = key.tun_key;
1323
a097c0b2
PS		 1324 add_nested_action_end(*sfa, start);
1325
1326 return err;
1327}
1328
1329static int validate_set(const struct nlattr *a,
1330 const struct sw_flow_key *flow_key,
1331 struct sw_flow_actions **sfa,
1332 bool *set_tun)
1333{
1334 const struct nlattr *ovs_key = nla_data(a);
1335 int key_type = nla_type(ovs_key);
1336
1337 /* There can be only one key in a action */
1338 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1339 return -EINVAL;
1340
1341 if (key_type > OVS_KEY_ATTR_MAX ||
1342 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
1343 ovs_key_lens[key_type] != -1))
1344 return -EINVAL;
1345
1346 switch (key_type) {
1347 const struct ovs_key_ipv4 *ipv4_key;
1348 const struct ovs_key_ipv6 *ipv6_key;
1349 int err;
1350
1351 case OVS_KEY_ATTR_PRIORITY:
1352 case OVS_KEY_ATTR_SKB_MARK:
1353 case OVS_KEY_ATTR_ETHERNET:
1354 break;
1355
1356 case OVS_KEY_ATTR_TUNNEL:
1357 *set_tun = true;
1358 err = validate_and_copy_set_tun(a, sfa);
1359 if (err)
1360 return err;
1361 break;
1362
1363 case OVS_KEY_ATTR_IPV4:
1364 if (flow_key->eth.type != htons(ETH_P_IP))
1365 return -EINVAL;
1366
1367 if (!flow_key->ip.proto)
1368 return -EINVAL;
1369
1370 ipv4_key = nla_data(ovs_key);
1371 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1372 return -EINVAL;
1373
1374 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1375 return -EINVAL;
1376
1377 break;
1378
1379 case OVS_KEY_ATTR_IPV6:
1380 if (flow_key->eth.type != htons(ETH_P_IPV6))
1381 return -EINVAL;
1382
1383 if (!flow_key->ip.proto)
1384 return -EINVAL;
1385
1386 ipv6_key = nla_data(ovs_key);
1387 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1388 return -EINVAL;
1389
1390 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1391 return -EINVAL;
1392
1393 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1394 return -EINVAL;
1395
1396 break;
1397
1398 case OVS_KEY_ATTR_TCP:
1399 if (flow_key->ip.proto != IPPROTO_TCP)
1400 return -EINVAL;
1401
1402 return validate_tp_port(flow_key);
1403
1404 case OVS_KEY_ATTR_UDP:
1405 if (flow_key->ip.proto != IPPROTO_UDP)
1406 return -EINVAL;
1407
1408 return validate_tp_port(flow_key);
1409
1410 case OVS_KEY_ATTR_SCTP:
1411 if (flow_key->ip.proto != IPPROTO_SCTP)
1412 return -EINVAL;
1413
1414 return validate_tp_port(flow_key);
1415
1416 default:
1417 return -EINVAL;
1418 }
1419
1420 return 0;
1421}
1422
1423static int validate_userspace(const struct nlattr *attr)
1424{
1425 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
1426 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
1427 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
1428 };
1429 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
1430 int error;
1431
1432 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
1433 attr, userspace_policy);
1434 if (error)
1435 return error;
1436
1437 if (!a[OVS_USERSPACE_ATTR_PID] ||
1438 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
1439 return -EINVAL;
1440
1441 return 0;
1442}
1443
1444static int copy_action(const struct nlattr *from,
1445 struct sw_flow_actions **sfa)
1446{
1447 int totlen = NLA_ALIGN(from->nla_len);
1448 struct nlattr *to;
1449
1450 to = reserve_sfa_size(sfa, from->nla_len);
1451 if (IS_ERR(to))
1452 return PTR_ERR(to);
1453
1454 memcpy(to, from, totlen);
1455 return 0;
1456}
1457
1458int ovs_nla_copy_actions(const struct nlattr *attr,
1459 const struct sw_flow_key *key,
1460 int depth,
1461 struct sw_flow_actions **sfa)
1462{
1463 const struct nlattr *a;
1464 int rem, err;
1465
1466 if (depth >= SAMPLE_ACTION_DEPTH)
1467 return -EOVERFLOW;
1468
1469 nla_for_each_nested(a, attr, rem) {
1470 /* Expected argument lengths, (u32)-1 for variable length. */
1471 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
1472 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
a6059080 1473 [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
a097c0b2
PS		 1474 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
1475 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
1476 [OVS_ACTION_ATTR_POP_VLAN] = 0,
1477 [OVS_ACTION_ATTR_SET] = (u32)-1,
7804df20
AZ		 1478 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
1479 [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
a097c0b2
PS		 1480 };
1481 const struct ovs_action_push_vlan *vlan;
1482 int type = nla_type(a);
1483 bool skip_copy;
1484
1485 if (type > OVS_ACTION_ATTR_MAX ||
1486 (action_lens[type] != nla_len(a) &&
1487 action_lens[type] != (u32)-1))
1488 return -EINVAL;
1489
1490 skip_copy = false;
1491 switch (type) {
1492 case OVS_ACTION_ATTR_UNSPEC:
1493 return -EINVAL;
1494
1495 case OVS_ACTION_ATTR_USERSPACE:
1496 err = validate_userspace(a);
1497 if (err)
1498 return err;
1499 break;
1500
1501 case OVS_ACTION_ATTR_OUTPUT:
1502 if (nla_get_u32(a) >= DP_MAX_PORTS)
1503 return -EINVAL;
1504 break;
1505
7804df20
AZ		 1506 case OVS_ACTION_ATTR_HASH: {
1507 const struct ovs_action_hash *act_hash = nla_data(a);
1508
1509 switch (act_hash->hash_alg) {
1510 case OVS_HASH_ALG_L4:
1511 break;
1512 default:
1513 return -EINVAL;
1514 }
1515
1516 break;
1517 }
a097c0b2
PS		 1518
1519 case OVS_ACTION_ATTR_POP_VLAN:
1520 break;
1521
1522 case OVS_ACTION_ATTR_PUSH_VLAN:
1523 vlan = nla_data(a);
1524 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
1525 return -EINVAL;
1526 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
1527 return -EINVAL;
1528 break;
1529
a6059080
AZ		 1530 case OVS_ACTION_ATTR_RECIRC:
1531 break;
1532
a097c0b2
PS		 1533 case OVS_ACTION_ATTR_SET:
1534 err = validate_set(a, key, sfa, &skip_copy);
1535 if (err)
1536 return err;
1537 break;
1538
1539 case OVS_ACTION_ATTR_SAMPLE:
1540 err = validate_and_copy_sample(a, key, depth, sfa);
1541 if (err)
1542 return err;
1543 skip_copy = true;
1544 break;
1545
1546 default:
1547 return -EINVAL;
1548 }
1549 if (!skip_copy) {
1550 err = copy_action(a, sfa);
1551 if (err)
1552 return err;
1553 }
1554 }
1555
1556 if (rem > 0)
1557 return -EINVAL;
1558
1559 return 0;
1560}
1561
1562static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1563{
1564 const struct nlattr *a;
1565 struct nlattr *start;
1566 int err = 0, rem;
1567
1568 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1569 if (!start)
1570 return -EMSGSIZE;
1571
1572 nla_for_each_nested(a, attr, rem) {
1573 int type = nla_type(a);
1574 struct nlattr *st_sample;
1575
1576 switch (type) {
1577 case OVS_SAMPLE_ATTR_PROBABILITY:
1578 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
1579 sizeof(u32), nla_data(a)))
1580 return -EMSGSIZE;
1581 break;
1582 case OVS_SAMPLE_ATTR_ACTIONS:
1583 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1584 if (!st_sample)
1585 return -EMSGSIZE;
1586 err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
1587 if (err)
1588 return err;
1589 nla_nest_end(skb, st_sample);
1590 break;
1591 }
1592 }
1593
1594 nla_nest_end(skb, start);
1595 return err;
1596}
1597
1598static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1599{
1600 const struct nlattr *ovs_key = nla_data(a);
1601 int key_type = nla_type(ovs_key);
1602 struct nlattr *start;
1603 int err;
1604
1605 switch (key_type) {
f0cd669f
JG		 1606 case OVS_KEY_ATTR_TUNNEL_INFO: {
1607 struct ovs_tunnel_info *tun_info = nla_data(ovs_key);
1608
a097c0b2
PS		 1609 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1610 if (!start)
1611 return -EMSGSIZE;
1612
f0cd669f
JG		 1613 err = ipv4_tun_to_nlattr(skb, &tun_info->tunnel,
1614 &tun_info->tunnel);
a097c0b2
PS		 1615 if (err)
1616 return err;
1617 nla_nest_end(skb, start);
1618 break;
f0cd669f 1619 }
a097c0b2
PS		 1620 default:
1621 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1622 return -EMSGSIZE;
1623 break;
1624 }
1625
1626 return 0;
1627}
1628
1629int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
1630{
1631 const struct nlattr *a;
1632 int rem, err;
1633
1634 nla_for_each_attr(a, attr, len, rem) {
1635 int type = nla_type(a);
1636
1637 switch (type) {
1638 case OVS_ACTION_ATTR_SET:
1639 err = set_action_to_attr(a, skb);
1640 if (err)
1641 return err;
1642 break;
1643
1644 case OVS_ACTION_ATTR_SAMPLE:
1645 err = sample_action_to_attr(a, skb);
1646 if (err)
1647 return err;
1648 break;
1649 default:
1650 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1651 return -EMSGSIZE;
1652 break;
1653 }
1654 }
1655
1656 return 0;
1657}
openvswitch packages for PVE