projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ip_tunnels: use tos and ttl fields also for IPv6
[deliverable/linux.git] / net / openvswitch / flow_netlink.c
CommitLineData
e6445719 1/*
971427f3 2 * Copyright (c) 2007-2014 Nicira, Inc.
e6445719
PS		 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
2235ad1c
JP		 19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
e6445719
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>
f5796684 45#include <net/geneve.h>
e6445719
PS		 46#include <net/ip.h>
47#include <net/ipv6.h>
48#include <net/ndisc.h>
25cd9ba0 49#include <net/mpls.h>
614732ea 50#include <net/vxlan.h>
e6445719
PS		 51
52#include "flow_netlink.h"
53
81bfe3c3
TG		 54struct ovs_len_tbl {
55 int len;
56 const struct ovs_len_tbl *next;
57};
58
59#define OVS_ATTR_NESTED -1
60
a85311bf
PS		 61static void update_range(struct sw_flow_match *match,
62 size_t offset, size_t size, bool is_mask)
e6445719 63{
a85311bf 64 struct sw_flow_key_range *range;
e6445719
PS		 65 size_t start = rounddown(offset, sizeof(long));
66 size_t end = roundup(offset + size, sizeof(long));
67
68 if (!is_mask)
69 range = &match->range;
a85311bf 70 else
e6445719
PS		 71 range = &match->mask->range;
72
e6445719
PS		 73 if (range->start == range->end) {
74 range->start = start;
75 range->end = end;
76 return;
77 }
78
79 if (range->start > start)
80 range->start = start;
81
82 if (range->end < end)
83 range->end = end;
84}
85
86#define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
87 do { \
a85311bf
PS		 88 update_range(match, offsetof(struct sw_flow_key, field), \
89 sizeof((match)->key->field), is_mask); \
90 if (is_mask) \
91 (match)->mask->key.field = value; \
92 else \
e6445719 93 (match)->key->field = value; \
e6445719
PS		 94 } while (0)
95
f5796684
JG		 96#define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask) \
97 do { \
a85311bf 98 update_range(match, offset, len, is_mask); \
f5796684
JG		 99 if (is_mask) \
100 memcpy((u8 *)&(match)->mask->key + offset, value_p, \
a85311bf 101 len); \
f5796684
JG		 102 else \
103 memcpy((u8 *)(match)->key + offset, value_p, len); \
e6445719
PS		 104 } while (0)
105
f5796684
JG		 106#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
107 SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
108 value_p, len, is_mask)
109
a85311bf
PS		 110#define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask) \
111 do { \
112 update_range(match, offsetof(struct sw_flow_key, field), \
113 sizeof((match)->key->field), is_mask); \
114 if (is_mask) \
115 memset((u8 *)&(match)->mask->key.field, value, \
116 sizeof((match)->mask->key.field)); \
117 else \
f47de068
PS		 118 memset((u8 *)&(match)->key->field, value, \
119 sizeof((match)->key->field)); \
f47de068 120 } while (0)
e6445719
PS		 121
122static bool match_validate(const struct sw_flow_match *match,
05da5898 123 u64 key_attrs, u64 mask_attrs, bool log)
e6445719
PS		 124{
125 u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
126 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
127
128 /* The following mask attributes allowed only if they
129 * pass the validation tests. */
130 mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
131 | (1 << OVS_KEY_ATTR_IPV6)
132 | (1 << OVS_KEY_ATTR_TCP)
5eb26b15 133 | (1 << OVS_KEY_ATTR_TCP_FLAGS)
e6445719
PS		 134 | (1 << OVS_KEY_ATTR_UDP)
135 | (1 << OVS_KEY_ATTR_SCTP)
136 | (1 << OVS_KEY_ATTR_ICMP)
137 | (1 << OVS_KEY_ATTR_ICMPV6)
138 | (1 << OVS_KEY_ATTR_ARP)
25cd9ba0
SH		 139 | (1 << OVS_KEY_ATTR_ND)
140 | (1 << OVS_KEY_ATTR_MPLS));
e6445719
PS		 141
142 /* Always allowed mask fields. */
143 mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
144 | (1 << OVS_KEY_ATTR_IN_PORT)
145 | (1 << OVS_KEY_ATTR_ETHERTYPE));
146
147 /* Check key attributes. */
148 if (match->key->eth.type == htons(ETH_P_ARP)
149 || match->key->eth.type == htons(ETH_P_RARP)) {
150 key_expected |= 1 << OVS_KEY_ATTR_ARP;
f2a01517 151 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
e6445719
PS		 152 mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
153 }
154
25cd9ba0
SH		 155 if (eth_p_mpls(match->key->eth.type)) {
156 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
157 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
158 mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
159 }
160
e6445719
PS		 161 if (match->key->eth.type == htons(ETH_P_IP)) {
162 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
163 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
164 mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
165
166 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
167 if (match->key->ip.proto == IPPROTO_UDP) {
168 key_expected |= 1 << OVS_KEY_ATTR_UDP;
169 if (match->mask && (match->mask->key.ip.proto == 0xff))
170 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
171 }
172
173 if (match->key->ip.proto == IPPROTO_SCTP) {
174 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
175 if (match->mask && (match->mask->key.ip.proto == 0xff))
176 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
177 }
178
179 if (match->key->ip.proto == IPPROTO_TCP) {
180 key_expected |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 181 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
182 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
e6445719 183 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 184 mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
185 }
e6445719
PS		 186 }
187
188 if (match->key->ip.proto == IPPROTO_ICMP) {
189 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
190 if (match->mask && (match->mask->key.ip.proto == 0xff))
191 mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
192 }
193 }
194 }
195
196 if (match->key->eth.type == htons(ETH_P_IPV6)) {
197 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
198 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
199 mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
200
201 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
202 if (match->key->ip.proto == IPPROTO_UDP) {
203 key_expected |= 1 << OVS_KEY_ATTR_UDP;
204 if (match->mask && (match->mask->key.ip.proto == 0xff))
205 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
206 }
207
208 if (match->key->ip.proto == IPPROTO_SCTP) {
209 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
210 if (match->mask && (match->mask->key.ip.proto == 0xff))
211 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
212 }
213
214 if (match->key->ip.proto == IPPROTO_TCP) {
215 key_expected |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 216 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
217 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
e6445719 218 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 219 mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
220 }
e6445719
PS		 221 }
222
223 if (match->key->ip.proto == IPPROTO_ICMPV6) {
224 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
225 if (match->mask && (match->mask->key.ip.proto == 0xff))
226 mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
227
1139e241 228 if (match->key->tp.src ==
e6445719 229 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1139e241 230 match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
e6445719 231 key_expected |= 1 << OVS_KEY_ATTR_ND;
f2a01517 232 if (match->mask && (match->mask->key.tp.src == htons(0xff)))
e6445719
PS		 233 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
234 }
235 }
236 }
237 }
238
239 if ((key_attrs & key_expected) != key_expected) {
240 /* Key attributes check failed. */
05da5898
JR		 241 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
242 (unsigned long long)key_attrs,
243 (unsigned long long)key_expected);
e6445719
PS		 244 return false;
245 }
246
247 if ((mask_attrs & mask_allowed) != mask_attrs) {
248 /* Mask attributes check failed. */
05da5898
JR		 249 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
250 (unsigned long long)mask_attrs,
251 (unsigned long long)mask_allowed);
e6445719
PS		 252 return false;
253 }
254
255 return true;
256}
257
8f0aad6f
WZ		 258size_t ovs_tun_key_attr_size(void)
259{
260 /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
261 * updating this function.
262 */
263 return nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
264 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
265 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
266 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
267 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
268 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
269 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
270 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_OAM */
271 + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
1dd144cf
TG		 272 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
273 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
274 */
8f0aad6f
WZ		 275 + nla_total_size(2) /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
276 + nla_total_size(2); /* OVS_TUNNEL_KEY_ATTR_TP_DST */
277}
278
41af73e9
JS		 279size_t ovs_key_attr_size(void)
280{
281 /* Whenever adding new OVS_KEY_ FIELDS, we should consider
282 * updating this function.
283 */
284 BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
285
286 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
287 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
8f0aad6f 288 + ovs_tun_key_attr_size()
41af73e9
JS		 289 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
290 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
291 + nla_total_size(4) /* OVS_KEY_ATTR_DP_HASH */
292 + nla_total_size(4) /* OVS_KEY_ATTR_RECIRC_ID */
293 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
294 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
295 + nla_total_size(4) /* OVS_KEY_ATTR_VLAN */
296 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
297 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
298 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
299 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
300 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
301}
302
81bfe3c3
TG		 303static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
304 [OVS_TUNNEL_KEY_ATTR_ID] = { .len = sizeof(u64) },
305 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = { .len = sizeof(u32) },
306 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = { .len = sizeof(u32) },
307 [OVS_TUNNEL_KEY_ATTR_TOS] = { .len = 1 },
308 [OVS_TUNNEL_KEY_ATTR_TTL] = { .len = 1 },
309 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
310 [OVS_TUNNEL_KEY_ATTR_CSUM] = { .len = 0 },
311 [OVS_TUNNEL_KEY_ATTR_TP_SRC] = { .len = sizeof(u16) },
312 [OVS_TUNNEL_KEY_ATTR_TP_DST] = { .len = sizeof(u16) },
313 [OVS_TUNNEL_KEY_ATTR_OAM] = { .len = 0 },
314 [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = { .len = OVS_ATTR_NESTED },
1dd144cf 315 [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS] = { .len = OVS_ATTR_NESTED },
81bfe3c3
TG		 316};
317
e6445719 318/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
81bfe3c3
TG		 319static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
320 [OVS_KEY_ATTR_ENCAP] = { .len = OVS_ATTR_NESTED },
321 [OVS_KEY_ATTR_PRIORITY] = { .len = sizeof(u32) },
322 [OVS_KEY_ATTR_IN_PORT] = { .len = sizeof(u32) },
323 [OVS_KEY_ATTR_SKB_MARK] = { .len = sizeof(u32) },
324 [OVS_KEY_ATTR_ETHERNET] = { .len = sizeof(struct ovs_key_ethernet) },
325 [OVS_KEY_ATTR_VLAN] = { .len = sizeof(__be16) },
326 [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
327 [OVS_KEY_ATTR_IPV4] = { .len = sizeof(struct ovs_key_ipv4) },
328 [OVS_KEY_ATTR_IPV6] = { .len = sizeof(struct ovs_key_ipv6) },
329 [OVS_KEY_ATTR_TCP] = { .len = sizeof(struct ovs_key_tcp) },
330 [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
331 [OVS_KEY_ATTR_UDP] = { .len = sizeof(struct ovs_key_udp) },
332 [OVS_KEY_ATTR_SCTP] = { .len = sizeof(struct ovs_key_sctp) },
333 [OVS_KEY_ATTR_ICMP] = { .len = sizeof(struct ovs_key_icmp) },
334 [OVS_KEY_ATTR_ICMPV6] = { .len = sizeof(struct ovs_key_icmpv6) },
335 [OVS_KEY_ATTR_ARP] = { .len = sizeof(struct ovs_key_arp) },
336 [OVS_KEY_ATTR_ND] = { .len = sizeof(struct ovs_key_nd) },
337 [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
338 [OVS_KEY_ATTR_DP_HASH] = { .len = sizeof(u32) },
339 [OVS_KEY_ATTR_TUNNEL] = { .len = OVS_ATTR_NESTED,
340 .next = ovs_tunnel_key_lens, },
341 [OVS_KEY_ATTR_MPLS] = { .len = sizeof(struct ovs_key_mpls) },
e6445719
PS		 342};
343
344static bool is_all_zero(const u8 *fp, size_t size)
345{
346 int i;
347
348 if (!fp)
349 return false;
350
351 for (i = 0; i < size; i++)
352 if (fp[i])
353 return false;
354
355 return true;
356}
357
358static int __parse_flow_nlattrs(const struct nlattr *attr,
359 const struct nlattr *a[],
05da5898 360 u64 *attrsp, bool log, bool nz)
e6445719
PS		 361{
362 const struct nlattr *nla;
363 u64 attrs;
364 int rem;
365
366 attrs = *attrsp;
367 nla_for_each_nested(nla, attr, rem) {
368 u16 type = nla_type(nla);
369 int expected_len;
370
371 if (type > OVS_KEY_ATTR_MAX) {
05da5898 372 OVS_NLERR(log, "Key type %d is out of range max %d",
e6445719
PS		 373 type, OVS_KEY_ATTR_MAX);
374 return -EINVAL;
375 }
376
377 if (attrs & (1 << type)) {
05da5898 378 OVS_NLERR(log, "Duplicate key (type %d).", type);
e6445719
PS		 379 return -EINVAL;
380 }
381
81bfe3c3
TG		 382 expected_len = ovs_key_lens[type].len;
383 if (nla_len(nla) != expected_len && expected_len != OVS_ATTR_NESTED) {
05da5898
JR		 384 OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
385 type, nla_len(nla), expected_len);
e6445719
PS		 386 return -EINVAL;
387 }
388
389 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
390 attrs |= 1 << type;
391 a[type] = nla;
392 }
393 }
394 if (rem) {
05da5898 395 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
e6445719
PS		 396 return -EINVAL;
397 }
398
399 *attrsp = attrs;
400 return 0;
401}
402
403static int parse_flow_mask_nlattrs(const struct nlattr *attr,
05da5898
JR		 404 const struct nlattr *a[], u64 *attrsp,
405 bool log)
e6445719 406{
05da5898 407 return __parse_flow_nlattrs(attr, a, attrsp, log, true);
e6445719
PS		 408}
409
410static int parse_flow_nlattrs(const struct nlattr *attr,
05da5898
JR		 411 const struct nlattr *a[], u64 *attrsp,
412 bool log)
e6445719 413{
05da5898
JR		 414 return __parse_flow_nlattrs(attr, a, attrsp, log, false);
415}
416
417static int genev_tun_opt_from_nlattr(const struct nlattr *a,
418 struct sw_flow_match *match, bool is_mask,
419 bool log)
420{
421 unsigned long opt_key_offset;
422
423 if (nla_len(a) > sizeof(match->key->tun_opts)) {
424 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
425 nla_len(a), sizeof(match->key->tun_opts));
426 return -EINVAL;
427 }
428
429 if (nla_len(a) % 4 != 0) {
430 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
431 nla_len(a));
432 return -EINVAL;
433 }
434
435 /* We need to record the length of the options passed
436 * down, otherwise packets with the same format but
437 * additional options will be silently matched.
438 */
439 if (!is_mask) {
440 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
441 false);
442 } else {
443 /* This is somewhat unusual because it looks at
444 * both the key and mask while parsing the
445 * attributes (and by extension assumes the key
446 * is parsed first). Normally, we would verify
447 * that each is the correct length and that the
448 * attributes line up in the validate function.
449 * However, that is difficult because this is
450 * variable length and we won't have the
451 * information later.
452 */
453 if (match->key->tun_opts_len != nla_len(a)) {
454 OVS_NLERR(log, "Geneve option len %d != mask len %d",
455 match->key->tun_opts_len, nla_len(a));
456 return -EINVAL;
457 }
458
459 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
460 }
461
d91641d9 462 opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
05da5898
JR		 463 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
464 nla_len(a), is_mask);
465 return 0;
e6445719
PS		 466}
467
1dd144cf
TG		 468static const struct nla_policy vxlan_opt_policy[OVS_VXLAN_EXT_MAX + 1] = {
469 [OVS_VXLAN_EXT_GBP] = { .type = NLA_U32 },
470};
471
472static int vxlan_tun_opt_from_nlattr(const struct nlattr *a,
473 struct sw_flow_match *match, bool is_mask,
474 bool log)
475{
476 struct nlattr *tb[OVS_VXLAN_EXT_MAX+1];
477 unsigned long opt_key_offset;
614732ea 478 struct vxlan_metadata opts;
1dd144cf
TG		 479 int err;
480
481 BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
482
483 err = nla_parse_nested(tb, OVS_VXLAN_EXT_MAX, a, vxlan_opt_policy);
484 if (err < 0)
485 return err;
486
487 memset(&opts, 0, sizeof(opts));
488
489 if (tb[OVS_VXLAN_EXT_GBP])
490 opts.gbp = nla_get_u32(tb[OVS_VXLAN_EXT_GBP]);
491
492 if (!is_mask)
493 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
494 else
495 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
496
497 opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
498 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
499 is_mask);
500 return 0;
501}
502
e6445719 503static int ipv4_tun_from_nlattr(const struct nlattr *attr,
05da5898
JR		 504 struct sw_flow_match *match, bool is_mask,
505 bool log)
e6445719
PS		 506{
507 struct nlattr *a;
508 int rem;
509 bool ttl = false;
510 __be16 tun_flags = 0;
1dd144cf 511 int opts_type = 0;
e6445719
PS		 512
513 nla_for_each_nested(a, attr, rem) {
514 int type = nla_type(a);
05da5898
JR		 515 int err;
516
e6445719 517 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
05da5898
JR		 518 OVS_NLERR(log, "Tunnel attr %d out of range max %d",
519 type, OVS_TUNNEL_KEY_ATTR_MAX);
e6445719
PS		 520 return -EINVAL;
521 }
522
81bfe3c3
TG		 523 if (ovs_tunnel_key_lens[type].len != nla_len(a) &&
524 ovs_tunnel_key_lens[type].len != OVS_ATTR_NESTED) {
05da5898 525 OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
81bfe3c3 526 type, nla_len(a), ovs_tunnel_key_lens[type].len);
e6445719
PS		 527 return -EINVAL;
528 }
529
530 switch (type) {
531 case OVS_TUNNEL_KEY_ATTR_ID:
532 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
533 nla_get_be64(a), is_mask);
534 tun_flags |= TUNNEL_KEY;
535 break;
536 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
c1ea5d67 537 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
67b61f6c 538 nla_get_in_addr(a), is_mask);
e6445719
PS		 539 break;
540 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
c1ea5d67 541 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
67b61f6c 542 nla_get_in_addr(a), is_mask);
e6445719
PS		 543 break;
544 case OVS_TUNNEL_KEY_ATTR_TOS:
7c383fb2 545 SW_FLOW_KEY_PUT(match, tun_key.tos,
e6445719
PS		 546 nla_get_u8(a), is_mask);
547 break;
548 case OVS_TUNNEL_KEY_ATTR_TTL:
7c383fb2 549 SW_FLOW_KEY_PUT(match, tun_key.ttl,
e6445719
PS		 550 nla_get_u8(a), is_mask);
551 ttl = true;
552 break;
553 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
554 tun_flags |= TUNNEL_DONT_FRAGMENT;
555 break;
556 case OVS_TUNNEL_KEY_ATTR_CSUM:
557 tun_flags |= TUNNEL_CSUM;
558 break;
8f0aad6f
WZ		 559 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
560 SW_FLOW_KEY_PUT(match, tun_key.tp_src,
561 nla_get_be16(a), is_mask);
562 break;
563 case OVS_TUNNEL_KEY_ATTR_TP_DST:
564 SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
565 nla_get_be16(a), is_mask);
566 break;
67fa0341
JG		 567 case OVS_TUNNEL_KEY_ATTR_OAM:
568 tun_flags |= TUNNEL_OAM;
569 break;
f5796684 570 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1dd144cf
TG		 571 if (opts_type) {
572 OVS_NLERR(log, "Multiple metadata blocks provided");
573 return -EINVAL;
574 }
575
05da5898
JR		 576 err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
577 if (err)
578 return err;
f5796684 579
1dd144cf
TG		 580 tun_flags |= TUNNEL_GENEVE_OPT;
581 opts_type = type;
582 break;
583 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
584 if (opts_type) {
585 OVS_NLERR(log, "Multiple metadata blocks provided");
586 return -EINVAL;
587 }
588
589 err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
590 if (err)
591 return err;
592
593 tun_flags |= TUNNEL_VXLAN_OPT;
594 opts_type = type;
f5796684 595 break;
e6445719 596 default:
05da5898 597 OVS_NLERR(log, "Unknown IPv4 tunnel attribute %d",
f5796684 598 type);
e6445719
PS		 599 return -EINVAL;
600 }
601 }
602
603 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
604
605 if (rem > 0) {
05da5898
JR		 606 OVS_NLERR(log, "IPv4 tunnel attribute has %d unknown bytes.",
607 rem);
e6445719
PS		 608 return -EINVAL;
609 }
610
611 if (!is_mask) {
c1ea5d67 612 if (!match->key->tun_key.u.ipv4.dst) {
05da5898 613 OVS_NLERR(log, "IPv4 tunnel dst address is zero");
e6445719
PS		 614 return -EINVAL;
615 }
616
617 if (!ttl) {
05da5898 618 OVS_NLERR(log, "IPv4 tunnel TTL not specified.");
e6445719
PS		 619 return -EINVAL;
620 }
621 }
622
1dd144cf
TG		 623 return opts_type;
624}
625
626static int vxlan_opt_to_nlattr(struct sk_buff *skb,
627 const void *tun_opts, int swkey_tun_opts_len)
628{
614732ea 629 const struct vxlan_metadata *opts = tun_opts;
1dd144cf
TG		 630 struct nlattr *nla;
631
632 nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
633 if (!nla)
634 return -EMSGSIZE;
635
636 if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
637 return -EMSGSIZE;
638
639 nla_nest_end(skb, nla);
e6445719
PS		 640 return 0;
641}
642
f5796684 643static int __ipv4_tun_to_nlattr(struct sk_buff *skb,
1d8fff90 644 const struct ip_tunnel_key *output,
d91641d9 645 const void *tun_opts, int swkey_tun_opts_len)
e6445719 646{
e6445719
PS		 647 if (output->tun_flags & TUNNEL_KEY &&
648 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
649 return -EMSGSIZE;
c1ea5d67 650 if (output->u.ipv4.src &&
930345ea 651 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
c1ea5d67 652 output->u.ipv4.src))
e6445719 653 return -EMSGSIZE;
c1ea5d67 654 if (output->u.ipv4.dst &&
930345ea 655 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
c1ea5d67 656 output->u.ipv4.dst))
e6445719 657 return -EMSGSIZE;
7c383fb2
JB		 658 if (output->tos &&
659 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
e6445719 660 return -EMSGSIZE;
7c383fb2 661 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
e6445719
PS		 662 return -EMSGSIZE;
663 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
67fa0341 664 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
e6445719
PS		 665 return -EMSGSIZE;
666 if ((output->tun_flags & TUNNEL_CSUM) &&
67fa0341
JG		 667 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
668 return -EMSGSIZE;
8f0aad6f
WZ		 669 if (output->tp_src &&
670 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
671 return -EMSGSIZE;
672 if (output->tp_dst &&
673 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
674 return -EMSGSIZE;
67fa0341
JG		 675 if ((output->tun_flags & TUNNEL_OAM) &&
676 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
e6445719 677 return -EMSGSIZE;
1dd144cf
TG		 678 if (tun_opts) {
679 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
680 nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
681 swkey_tun_opts_len, tun_opts))
682 return -EMSGSIZE;
683 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
684 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
685 return -EMSGSIZE;
686 }
e6445719 687
e6445719
PS		 688 return 0;
689}
690
f5796684 691static int ipv4_tun_to_nlattr(struct sk_buff *skb,
1d8fff90 692 const struct ip_tunnel_key *output,
d91641d9 693 const void *tun_opts, int swkey_tun_opts_len)
f5796684
JG		 694{
695 struct nlattr *nla;
696 int err;
697
698 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
699 if (!nla)
700 return -EMSGSIZE;
701
702 err = __ipv4_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len);
703 if (err)
704 return err;
705
706 nla_nest_end(skb, nla);
707 return 0;
708}
709
8f0aad6f 710int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
1d8fff90 711 const struct ip_tunnel_info *egress_tun_info)
8f0aad6f 712{
1d8fff90 713 return __ipv4_tun_to_nlattr(skb, &egress_tun_info->key,
8f0aad6f
WZ		 714 egress_tun_info->options,
715 egress_tun_info->options_len);
716}
717
e6445719 718static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs,
05da5898
JR		 719 const struct nlattr **a, bool is_mask,
720 bool log)
e6445719 721{
971427f3
AZ		 722 if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
723 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
724
725 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
726 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
727 }
728
729 if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
730 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
731
732 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
733 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
734 }
735
e6445719
PS		 736 if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
737 SW_FLOW_KEY_PUT(match, phy.priority,
738 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
739 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
740 }
741
742 if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
743 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
744
426cda5c 745 if (is_mask) {
e6445719 746 in_port = 0xffffffff; /* Always exact match in_port. */
426cda5c 747 } else if (in_port >= DP_MAX_PORTS) {
05da5898 748 OVS_NLERR(log, "Port %d exceeds max allowable %d",
426cda5c 749 in_port, DP_MAX_PORTS);
e6445719 750 return -EINVAL;
426cda5c 751 }
e6445719
PS		 752
753 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
754 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
755 } else if (!is_mask) {
756 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
757 }
758
759 if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
760 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
761
762 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
763 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
764 }
765 if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
766 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1dd144cf 767 is_mask, log) < 0)
e6445719
PS		 768 return -EINVAL;
769 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
770 }
771 return 0;
772}
773
23dabf88 774static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs,
05da5898
JR		 775 const struct nlattr **a, bool is_mask,
776 bool log)
e6445719
PS		 777{
778 int err;
e6445719 779
05da5898 780 err = metadata_from_nlattrs(match, &attrs, a, is_mask, log);
e6445719
PS		 781 if (err)
782 return err;
783
784 if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
785 const struct ovs_key_ethernet *eth_key;
786
787 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
788 SW_FLOW_KEY_MEMCPY(match, eth.src,
789 eth_key->eth_src, ETH_ALEN, is_mask);
790 SW_FLOW_KEY_MEMCPY(match, eth.dst,
791 eth_key->eth_dst, ETH_ALEN, is_mask);
792 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
793 }
794
795 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
796 __be16 tci;
797
798 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
799 if (!(tci & htons(VLAN_TAG_PRESENT))) {
800 if (is_mask)
05da5898 801 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
e6445719 802 else
05da5898 803 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
e6445719
PS		 804
805 return -EINVAL;
806 }
807
808 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
809 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
a85311bf 810 }
e6445719
PS		 811
812 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
813 __be16 eth_type;
814
815 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
816 if (is_mask) {
817 /* Always exact match EtherType. */
818 eth_type = htons(0xffff);
6713fc9b 819 } else if (!eth_proto_is_802_3(eth_type)) {
05da5898
JR		 820 OVS_NLERR(log, "EtherType %x is less than min %x",
821 ntohs(eth_type), ETH_P_802_3_MIN);
e6445719
PS		 822 return -EINVAL;
823 }
824
825 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
826 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
827 } else if (!is_mask) {
828 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
829 }
830
831 if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
832 const struct ovs_key_ipv4 *ipv4_key;
833
834 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
835 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
05da5898
JR		 836 OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
837 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
e6445719
PS		 838 return -EINVAL;
839 }
840 SW_FLOW_KEY_PUT(match, ip.proto,
841 ipv4_key->ipv4_proto, is_mask);
842 SW_FLOW_KEY_PUT(match, ip.tos,
843 ipv4_key->ipv4_tos, is_mask);
844 SW_FLOW_KEY_PUT(match, ip.ttl,
845 ipv4_key->ipv4_ttl, is_mask);
846 SW_FLOW_KEY_PUT(match, ip.frag,
847 ipv4_key->ipv4_frag, is_mask);
848 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
849 ipv4_key->ipv4_src, is_mask);
850 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
851 ipv4_key->ipv4_dst, is_mask);
852 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
853 }
854
855 if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
856 const struct ovs_key_ipv6 *ipv6_key;
857
858 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
859 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
05da5898
JR		 860 OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
861 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
e6445719
PS		 862 return -EINVAL;
863 }
fecaef85 864
d3052bb5 865 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
14591433 866 OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
fecaef85
JR		 867 ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
868 return -EINVAL;
869 }
870
e6445719
PS		 871 SW_FLOW_KEY_PUT(match, ipv6.label,
872 ipv6_key->ipv6_label, is_mask);
873 SW_FLOW_KEY_PUT(match, ip.proto,
874 ipv6_key->ipv6_proto, is_mask);
875 SW_FLOW_KEY_PUT(match, ip.tos,
876 ipv6_key->ipv6_tclass, is_mask);
877 SW_FLOW_KEY_PUT(match, ip.ttl,
878 ipv6_key->ipv6_hlimit, is_mask);
879 SW_FLOW_KEY_PUT(match, ip.frag,
880 ipv6_key->ipv6_frag, is_mask);
881 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
882 ipv6_key->ipv6_src,
883 sizeof(match->key->ipv6.addr.src),
884 is_mask);
885 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
886 ipv6_key->ipv6_dst,
887 sizeof(match->key->ipv6.addr.dst),
888 is_mask);
889
890 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
891 }
892
893 if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
894 const struct ovs_key_arp *arp_key;
895
896 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
897 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
05da5898 898 OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
e6445719
PS		 899 arp_key->arp_op);
900 return -EINVAL;
901 }
902
903 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
904 arp_key->arp_sip, is_mask);
905 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
906 arp_key->arp_tip, is_mask);
907 SW_FLOW_KEY_PUT(match, ip.proto,
908 ntohs(arp_key->arp_op), is_mask);
909 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
910 arp_key->arp_sha, ETH_ALEN, is_mask);
911 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
912 arp_key->arp_tha, ETH_ALEN, is_mask);
913
914 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
915 }
916
25cd9ba0
SH		 917 if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
918 const struct ovs_key_mpls *mpls_key;
919
920 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
921 SW_FLOW_KEY_PUT(match, mpls.top_lse,
922 mpls_key->mpls_lse, is_mask);
923
924 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
925 }
926
e6445719
PS		 927 if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
928 const struct ovs_key_tcp *tcp_key;
929
930 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1139e241
JR		 931 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
932 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
e6445719
PS		 933 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
934 }
935
5eb26b15 936 if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1b760fb9
JS		 937 SW_FLOW_KEY_PUT(match, tp.flags,
938 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
939 is_mask);
5eb26b15
JR		 940 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
941 }
942
e6445719
PS		 943 if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
944 const struct ovs_key_udp *udp_key;
945
946 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1139e241
JR		 947 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
948 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
e6445719
PS		 949 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
950 }
951
952 if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
953 const struct ovs_key_sctp *sctp_key;
954
955 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1139e241
JR		 956 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
957 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
e6445719
PS		 958 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
959 }
960
961 if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
962 const struct ovs_key_icmp *icmp_key;
963
964 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1139e241 965 SW_FLOW_KEY_PUT(match, tp.src,
e6445719 966 htons(icmp_key->icmp_type), is_mask);
1139e241 967 SW_FLOW_KEY_PUT(match, tp.dst,
e6445719
PS		 968 htons(icmp_key->icmp_code), is_mask);
969 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
970 }
971
972 if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
973 const struct ovs_key_icmpv6 *icmpv6_key;
974
975 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1139e241 976 SW_FLOW_KEY_PUT(match, tp.src,
e6445719 977 htons(icmpv6_key->icmpv6_type), is_mask);
1139e241 978 SW_FLOW_KEY_PUT(match, tp.dst,
e6445719
PS		 979 htons(icmpv6_key->icmpv6_code), is_mask);
980 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
981 }
982
983 if (attrs & (1 << OVS_KEY_ATTR_ND)) {
984 const struct ovs_key_nd *nd_key;
985
986 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
987 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
988 nd_key->nd_target,
989 sizeof(match->key->ipv6.nd.target),
990 is_mask);
991 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
992 nd_key->nd_sll, ETH_ALEN, is_mask);
993 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
994 nd_key->nd_tll, ETH_ALEN, is_mask);
995 attrs &= ~(1 << OVS_KEY_ATTR_ND);
996 }
997
426cda5c 998 if (attrs != 0) {
05da5898 999 OVS_NLERR(log, "Unknown key attributes %llx",
426cda5c 1000 (unsigned long long)attrs);
e6445719 1001 return -EINVAL;
426cda5c 1002 }
e6445719
PS		 1003
1004 return 0;
1005}
1006
81bfe3c3
TG		 1007static void nlattr_set(struct nlattr *attr, u8 val,
1008 const struct ovs_len_tbl *tbl)
e6445719 1009{
f47de068
PS		 1010 struct nlattr *nla;
1011 int rem;
e6445719 1012
f47de068
PS		 1013 /* The nlattr stream should already have been validated */
1014 nla_for_each_nested(nla, attr, rem) {
81bfe3c3
TG		 1015 if (tbl && tbl[nla_type(nla)].len == OVS_ATTR_NESTED)
1016 nlattr_set(nla, val, tbl[nla_type(nla)].next);
f47de068
PS		 1017 else
1018 memset(nla_data(nla), val, nla_len(nla));
1019 }
1020}
1021
1022static void mask_set_nlattr(struct nlattr *attr, u8 val)
1023{
81bfe3c3 1024 nlattr_set(attr, val, ovs_key_lens);
e6445719
PS		 1025}
1026
1027/**
1028 * ovs_nla_get_match - parses Netlink attributes into a flow key and
1029 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1030 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1031 * does not include any don't care bit.
1032 * @match: receives the extracted flow match information.
1033 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1034 * sequence. The fields should of the packet that triggered the creation
1035 * of this flow.
1036 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1037 * attribute specifies the mask field of the wildcarded flow.
05da5898
JR		 1038 * @log: Boolean to allow kernel error logging. Normally true, but when
1039 * probing for feature compatibility this should be passed in as false to
1040 * suppress unnecessary error logging.
e6445719
PS		 1041 */
1042int ovs_nla_get_match(struct sw_flow_match *match,
a85311bf 1043 const struct nlattr *nla_key,
05da5898
JR		 1044 const struct nlattr *nla_mask,
1045 bool log)
e6445719
PS		 1046{
1047 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1048 const struct nlattr *encap;
f47de068 1049 struct nlattr *newmask = NULL;
e6445719
PS		 1050 u64 key_attrs = 0;
1051 u64 mask_attrs = 0;
1052 bool encap_valid = false;
1053 int err;
1054
05da5898 1055 err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
e6445719
PS		 1056 if (err)
1057 return err;
1058
1059 if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1060 (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1061 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1062 __be16 tci;
1063
1064 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1065 (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
05da5898 1066 OVS_NLERR(log, "Invalid Vlan frame.");
e6445719
PS		 1067 return -EINVAL;
1068 }
1069
1070 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1071 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1072 encap = a[OVS_KEY_ATTR_ENCAP];
1073 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1074 encap_valid = true;
1075
1076 if (tci & htons(VLAN_TAG_PRESENT)) {
05da5898 1077 err = parse_flow_nlattrs(encap, a, &key_attrs, log);
e6445719
PS		 1078 if (err)
1079 return err;
1080 } else if (!tci) {
1081 /* Corner case for truncated 802.1Q header. */
1082 if (nla_len(encap)) {
05da5898 1083 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
e6445719
PS		 1084 return -EINVAL;
1085 }
1086 } else {
05da5898 1087 OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
e6445719
PS		 1088 return -EINVAL;
1089 }
1090 }
1091
05da5898 1092 err = ovs_key_from_nlattrs(match, key_attrs, a, false, log);
e6445719
PS		 1093 if (err)
1094 return err;
1095
a85311bf
PS		 1096 if (match->mask) {
1097 if (!nla_mask) {
1098 /* Create an exact match mask. We need to set to 0xff
1099 * all the 'match->mask' fields that have been touched
1100 * in 'match->key'. We cannot simply memset
1101 * 'match->mask', because padding bytes and fields not
1102 * specified in 'match->key' should be left to 0.
1103 * Instead, we use a stream of netlink attributes,
1104 * copied from 'key' and set to 0xff.
1105 * ovs_key_from_nlattrs() will take care of filling
1106 * 'match->mask' appropriately.
1107 */
1108 newmask = kmemdup(nla_key,
1109 nla_total_size(nla_len(nla_key)),
1110 GFP_KERNEL);
1111 if (!newmask)
1112 return -ENOMEM;
f47de068 1113
a85311bf 1114 mask_set_nlattr(newmask, 0xff);
f47de068 1115
a85311bf
PS		 1116 /* The userspace does not send tunnel attributes that
1117 * are 0, but we should not wildcard them nonetheless.
1118 */
c1ea5d67 1119 if (match->key->tun_key.u.ipv4.dst)
a85311bf
PS		 1120 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1121 0xff, true);
f47de068 1122
a85311bf
PS		 1123 nla_mask = newmask;
1124 }
f47de068 1125
05da5898 1126 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
e6445719 1127 if (err)
f47de068 1128 goto free_newmask;
e6445719 1129
a85311bf
PS		 1130 /* Always match on tci. */
1131 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1132
f47de068 1133 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
e6445719
PS		 1134 __be16 eth_type = 0;
1135 __be16 tci = 0;
1136
1137 if (!encap_valid) {
05da5898 1138 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
f47de068
PS		 1139 err = -EINVAL;
1140 goto free_newmask;
e6445719
PS		 1141 }
1142
1143 mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1144 if (a[OVS_KEY_ATTR_ETHERTYPE])
1145 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1146
1147 if (eth_type == htons(0xffff)) {
1148 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1149 encap = a[OVS_KEY_ATTR_ENCAP];
05da5898
JR		 1150 err = parse_flow_mask_nlattrs(encap, a,
1151 &mask_attrs, log);
f47de068
PS		 1152 if (err)
1153 goto free_newmask;
e6445719 1154 } else {
05da5898
JR		 1155 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1156 ntohs(eth_type));
f47de068
PS		 1157 err = -EINVAL;
1158 goto free_newmask;
e6445719
PS		 1159 }
1160
1161 if (a[OVS_KEY_ATTR_VLAN])
1162 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1163
1164 if (!(tci & htons(VLAN_TAG_PRESENT))) {
05da5898
JR		 1165 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1166 ntohs(tci));
f47de068
PS		 1167 err = -EINVAL;
1168 goto free_newmask;
e6445719
PS		 1169 }
1170 }
1171
05da5898 1172 err = ovs_key_from_nlattrs(match, mask_attrs, a, true, log);
e6445719 1173 if (err)
f47de068 1174 goto free_newmask;
e6445719
PS		 1175 }
1176
05da5898 1177 if (!match_validate(match, key_attrs, mask_attrs, log))
f47de068 1178 err = -EINVAL;
e6445719 1179
f47de068
PS		 1180free_newmask:
1181 kfree(newmask);
1182 return err;
e6445719
PS		 1183}
1184
74ed7ab9
JS		 1185static size_t get_ufid_len(const struct nlattr *attr, bool log)
1186{
1187 size_t len;
1188
1189 if (!attr)
1190 return 0;
1191
1192 len = nla_len(attr);
1193 if (len < 1 || len > MAX_UFID_LENGTH) {
1194 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1195 nla_len(attr), MAX_UFID_LENGTH);
1196 return 0;
1197 }
1198
1199 return len;
1200}
1201
1202/* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1203 * or false otherwise.
1204 */
1205bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1206 bool log)
1207{
1208 sfid->ufid_len = get_ufid_len(attr, log);
1209 if (sfid->ufid_len)
1210 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1211
1212 return sfid->ufid_len;
1213}
1214
1215int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1216 const struct sw_flow_key *key, bool log)
1217{
1218 struct sw_flow_key *new_key;
1219
1220 if (ovs_nla_get_ufid(sfid, ufid, log))
1221 return 0;
1222
1223 /* If UFID was not provided, use unmasked key. */
1224 new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1225 if (!new_key)
1226 return -ENOMEM;
1227 memcpy(new_key, key, sizeof(*key));
1228 sfid->unmasked_key = new_key;
1229
1230 return 0;
1231}
1232
1233u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1234{
1235 return attr ? nla_get_u32(attr) : 0;
1236}
1237
e6445719
PS		 1238/**
1239 * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
83c8df26 1240 * @key: Receives extracted in_port, priority, tun_key and skb_mark.
e6445719
PS		 1241 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1242 * sequence.
05da5898
JR		 1243 * @log: Boolean to allow kernel error logging. Normally true, but when
1244 * probing for feature compatibility this should be passed in as false to
1245 * suppress unnecessary error logging.
e6445719
PS		 1246 *
1247 * This parses a series of Netlink attributes that form a flow key, which must
1248 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1249 * get the metadata, that is, the parts of the flow key that cannot be
1250 * extracted from the packet itself.
1251 */
1252
83c8df26 1253int ovs_nla_get_flow_metadata(const struct nlattr *attr,
05da5898
JR		 1254 struct sw_flow_key *key,
1255 bool log)
e6445719 1256{
e6445719 1257 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
83c8df26 1258 struct sw_flow_match match;
e6445719
PS		 1259 u64 attrs = 0;
1260 int err;
e6445719 1261
05da5898 1262 err = parse_flow_nlattrs(attr, a, &attrs, log);
e6445719
PS		 1263 if (err)
1264 return -EINVAL;
1265
1266 memset(&match, 0, sizeof(match));
83c8df26 1267 match.key = key;
e6445719 1268
83c8df26 1269 key->phy.in_port = DP_MAX_PORTS;
e6445719 1270
05da5898 1271 return metadata_from_nlattrs(&match, &attrs, a, false, log);
e6445719
PS		 1272}
1273
5b4237bb
JS		 1274static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1275 const struct sw_flow_key *output, bool is_mask,
1276 struct sk_buff *skb)
e6445719
PS		 1277{
1278 struct ovs_key_ethernet *eth_key;
1279 struct nlattr *nla, *encap;
e6445719 1280
971427f3
AZ		 1281 if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1282 goto nla_put_failure;
1283
1284 if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1285 goto nla_put_failure;
1286
e6445719
PS		 1287 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1288 goto nla_put_failure;
1289
c1ea5d67 1290 if ((swkey->tun_key.u.ipv4.dst || is_mask)) {
d91641d9 1291 const void *opts = NULL;
f5796684
JG		 1292
1293 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
d91641d9 1294 opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
f5796684
JG		 1295
1296 if (ipv4_tun_to_nlattr(skb, &output->tun_key, opts,
1297 swkey->tun_opts_len))
1298 goto nla_put_failure;
1299 }
e6445719
PS		 1300
1301 if (swkey->phy.in_port == DP_MAX_PORTS) {
1302 if (is_mask && (output->phy.in_port == 0xffff))
1303 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1304 goto nla_put_failure;
1305 } else {
1306 u16 upper_u16;
1307 upper_u16 = !is_mask ? 0 : 0xffff;
1308
1309 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1310 (upper_u16 << 16) | output->phy.in_port))
1311 goto nla_put_failure;
1312 }
1313
1314 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1315 goto nla_put_failure;
1316
1317 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1318 if (!nla)
1319 goto nla_put_failure;
1320
1321 eth_key = nla_data(nla);
8c63ff09
JP		 1322 ether_addr_copy(eth_key->eth_src, output->eth.src);
1323 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
e6445719
PS		 1324
1325 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1326 __be16 eth_type;
1327 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1328 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1329 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1330 goto nla_put_failure;
1331 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1332 if (!swkey->eth.tci)
1333 goto unencap;
1334 } else
1335 encap = NULL;
1336
1337 if (swkey->eth.type == htons(ETH_P_802_2)) {
1338 /*
1339 * Ethertype 802.2 is represented in the netlink with omitted
1340 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1341 * 0xffff in the mask attribute. Ethertype can also
1342 * be wildcarded.
1343 */
1344 if (is_mask && output->eth.type)
1345 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1346 output->eth.type))
1347 goto nla_put_failure;
1348 goto unencap;
1349 }
1350
1351 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1352 goto nla_put_failure;
1353
1354 if (swkey->eth.type == htons(ETH_P_IP)) {
1355 struct ovs_key_ipv4 *ipv4_key;
1356
1357 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1358 if (!nla)
1359 goto nla_put_failure;
1360 ipv4_key = nla_data(nla);
1361 ipv4_key->ipv4_src = output->ipv4.addr.src;
1362 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1363 ipv4_key->ipv4_proto = output->ip.proto;
1364 ipv4_key->ipv4_tos = output->ip.tos;
1365 ipv4_key->ipv4_ttl = output->ip.ttl;
1366 ipv4_key->ipv4_frag = output->ip.frag;
1367 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1368 struct ovs_key_ipv6 *ipv6_key;
1369
1370 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1371 if (!nla)
1372 goto nla_put_failure;
1373 ipv6_key = nla_data(nla);
1374 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1375 sizeof(ipv6_key->ipv6_src));
1376 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1377 sizeof(ipv6_key->ipv6_dst));
1378 ipv6_key->ipv6_label = output->ipv6.label;
1379 ipv6_key->ipv6_proto = output->ip.proto;
1380 ipv6_key->ipv6_tclass = output->ip.tos;
1381 ipv6_key->ipv6_hlimit = output->ip.ttl;
1382 ipv6_key->ipv6_frag = output->ip.frag;
1383 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1384 swkey->eth.type == htons(ETH_P_RARP)) {
1385 struct ovs_key_arp *arp_key;
1386
1387 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1388 if (!nla)
1389 goto nla_put_failure;
1390 arp_key = nla_data(nla);
1391 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1392 arp_key->arp_sip = output->ipv4.addr.src;
1393 arp_key->arp_tip = output->ipv4.addr.dst;
1394 arp_key->arp_op = htons(output->ip.proto);
8c63ff09
JP		 1395 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1396 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
25cd9ba0
SH		 1397 } else if (eth_p_mpls(swkey->eth.type)) {
1398 struct ovs_key_mpls *mpls_key;
1399
1400 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1401 if (!nla)
1402 goto nla_put_failure;
1403 mpls_key = nla_data(nla);
1404 mpls_key->mpls_lse = output->mpls.top_lse;
e6445719
PS		 1405 }
1406
1407 if ((swkey->eth.type == htons(ETH_P_IP) ||
1408 swkey->eth.type == htons(ETH_P_IPV6)) &&
1409 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1410
1411 if (swkey->ip.proto == IPPROTO_TCP) {
1412 struct ovs_key_tcp *tcp_key;
1413
1414 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1415 if (!nla)
1416 goto nla_put_failure;
1417 tcp_key = nla_data(nla);
1139e241
JR		 1418 tcp_key->tcp_src = output->tp.src;
1419 tcp_key->tcp_dst = output->tp.dst;
1420 if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1421 output->tp.flags))
1422 goto nla_put_failure;
e6445719
PS		 1423 } else if (swkey->ip.proto == IPPROTO_UDP) {
1424 struct ovs_key_udp *udp_key;
1425
1426 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1427 if (!nla)
1428 goto nla_put_failure;
1429 udp_key = nla_data(nla);
1139e241
JR		 1430 udp_key->udp_src = output->tp.src;
1431 udp_key->udp_dst = output->tp.dst;
e6445719
PS		 1432 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1433 struct ovs_key_sctp *sctp_key;
1434
1435 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1436 if (!nla)
1437 goto nla_put_failure;
1438 sctp_key = nla_data(nla);
1139e241
JR		 1439 sctp_key->sctp_src = output->tp.src;
1440 sctp_key->sctp_dst = output->tp.dst;
e6445719
PS		 1441 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1442 swkey->ip.proto == IPPROTO_ICMP) {
1443 struct ovs_key_icmp *icmp_key;
1444
1445 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1446 if (!nla)
1447 goto nla_put_failure;
1448 icmp_key = nla_data(nla);
1139e241
JR		 1449 icmp_key->icmp_type = ntohs(output->tp.src);
1450 icmp_key->icmp_code = ntohs(output->tp.dst);
e6445719
PS		 1451 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1452 swkey->ip.proto == IPPROTO_ICMPV6) {
1453 struct ovs_key_icmpv6 *icmpv6_key;
1454
1455 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1456 sizeof(*icmpv6_key));
1457 if (!nla)
1458 goto nla_put_failure;
1459 icmpv6_key = nla_data(nla);
1139e241
JR		 1460 icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1461 icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
e6445719
PS		 1462
1463 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1464 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1465 struct ovs_key_nd *nd_key;
1466
1467 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1468 if (!nla)
1469 goto nla_put_failure;
1470 nd_key = nla_data(nla);
1471 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1472 sizeof(nd_key->nd_target));
8c63ff09
JP		 1473 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1474 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
e6445719
PS		 1475 }
1476 }
1477 }
1478
1479unencap:
1480 if (encap)
1481 nla_nest_end(skb, encap);
1482
1483 return 0;
1484
1485nla_put_failure:
1486 return -EMSGSIZE;
1487}
1488
5b4237bb
JS		 1489int ovs_nla_put_key(const struct sw_flow_key *swkey,
1490 const struct sw_flow_key *output, int attr, bool is_mask,
1491 struct sk_buff *skb)
1492{
1493 int err;
1494 struct nlattr *nla;
1495
1496 nla = nla_nest_start(skb, attr);
1497 if (!nla)
1498 return -EMSGSIZE;
1499 err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1500 if (err)
1501 return err;
1502 nla_nest_end(skb, nla);
1503
1504 return 0;
1505}
1506
1507/* Called with ovs_mutex or RCU read lock. */
74ed7ab9
JS		 1508int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1509{
1510 if (ovs_identifier_is_ufid(&flow->id))
1511 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1512 flow->id.ufid);
1513
1514 return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1515 OVS_FLOW_ATTR_KEY, false, skb);
1516}
1517
1518/* Called with ovs_mutex or RCU read lock. */
1519int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
5b4237bb 1520{
26ad0b83 1521 return ovs_nla_put_key(&flow->key, &flow->key,
5b4237bb
JS		 1522 OVS_FLOW_ATTR_KEY, false, skb);
1523}
1524
1525/* Called with ovs_mutex or RCU read lock. */
1526int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1527{
1528 return ovs_nla_put_key(&flow->key, &flow->mask->key,
1529 OVS_FLOW_ATTR_MASK, true, skb);
1530}
1531
e6445719
PS		 1532#define MAX_ACTIONS_BUFSIZE (32 * 1024)
1533
05da5898 1534static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
e6445719
PS		 1535{
1536 struct sw_flow_actions *sfa;
1537
426cda5c 1538 if (size > MAX_ACTIONS_BUFSIZE) {
05da5898 1539 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
e6445719 1540 return ERR_PTR(-EINVAL);
426cda5c 1541 }
e6445719
PS		 1542
1543 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1544 if (!sfa)
1545 return ERR_PTR(-ENOMEM);
1546
1547 sfa->actions_len = 0;
1548 return sfa;
1549}
1550
34ae932a
TG		 1551static void ovs_nla_free_set_action(const struct nlattr *a)
1552{
1553 const struct nlattr *ovs_key = nla_data(a);
1554 struct ovs_tunnel_info *ovs_tun;
1555
1556 switch (nla_type(ovs_key)) {
1557 case OVS_KEY_ATTR_TUNNEL_INFO:
1558 ovs_tun = nla_data(ovs_key);
1559 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1560 break;
1561 }
1562}
1563
1564void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1565{
1566 const struct nlattr *a;
1567 int rem;
1568
1569 if (!sf_acts)
1570 return;
1571
1572 nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1573 switch (nla_type(a)) {
1574 case OVS_ACTION_ATTR_SET:
1575 ovs_nla_free_set_action(a);
1576 break;
1577 }
1578 }
1579
1580 kfree(sf_acts);
1581}
1582
1583static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1584{
1585 ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1586}
1587
e6445719
PS		 1588/* Schedules 'sf_acts' to be freed after the next RCU grace period.
1589 * The caller must hold rcu_read_lock for this to be sensible. */
34ae932a 1590void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
e6445719 1591{
34ae932a 1592 call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
e6445719
PS		 1593}
1594
1595static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
05da5898 1596 int attr_len, bool log)
e6445719
PS		 1597{
1598
1599 struct sw_flow_actions *acts;
1600 int new_acts_size;
1601 int req_size = NLA_ALIGN(attr_len);
1602 int next_offset = offsetof(struct sw_flow_actions, actions) +
1603 (*sfa)->actions_len;
1604
1605 if (req_size <= (ksize(*sfa) - next_offset))
1606 goto out;
1607
1608 new_acts_size = ksize(*sfa) * 2;
1609
1610 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1611 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1612 return ERR_PTR(-EMSGSIZE);
1613 new_acts_size = MAX_ACTIONS_BUFSIZE;
1614 }
1615
05da5898 1616 acts = nla_alloc_flow_actions(new_acts_size, log);
e6445719
PS		 1617 if (IS_ERR(acts))
1618 return (void *)acts;
1619
1620 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1621 acts->actions_len = (*sfa)->actions_len;
1622 kfree(*sfa);
1623 *sfa = acts;
1624
1625out:
1626 (*sfa)->actions_len += req_size;
1627 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1628}
1629
f0b128c1 1630static struct nlattr *__add_action(struct sw_flow_actions **sfa,
05da5898 1631 int attrtype, void *data, int len, bool log)
e6445719
PS		 1632{
1633 struct nlattr *a;
1634
05da5898 1635 a = reserve_sfa_size(sfa, nla_attr_size(len), log);
e6445719 1636 if (IS_ERR(a))
f0b128c1 1637 return a;
e6445719
PS		 1638
1639 a->nla_type = attrtype;
1640 a->nla_len = nla_attr_size(len);
1641
1642 if (data)
1643 memcpy(nla_data(a), data, len);
1644 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1645
f0b128c1
JG		 1646 return a;
1647}
1648
1649static int add_action(struct sw_flow_actions **sfa, int attrtype,
05da5898 1650 void *data, int len, bool log)
f0b128c1
JG		 1651{
1652 struct nlattr *a;
1653
05da5898 1654 a = __add_action(sfa, attrtype, data, len, log);
f0b128c1 1655
f35423c1 1656 return PTR_ERR_OR_ZERO(a);
e6445719
PS		 1657}
1658
1659static inline int add_nested_action_start(struct sw_flow_actions **sfa,
05da5898 1660 int attrtype, bool log)
e6445719
PS		 1661{
1662 int used = (*sfa)->actions_len;
1663 int err;
1664
05da5898 1665 err = add_action(sfa, attrtype, NULL, 0, log);
e6445719
PS		 1666 if (err)
1667 return err;
1668
1669 return used;
1670}
1671
1672static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1673 int st_offset)
1674{
1675 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1676 st_offset);
1677
1678 a->nla_len = sfa->actions_len - st_offset;
1679}
1680
2fdb957d 1681static int __ovs_nla_copy_actions(const struct nlattr *attr,
25cd9ba0
SH		 1682 const struct sw_flow_key *key,
1683 int depth, struct sw_flow_actions **sfa,
05da5898 1684 __be16 eth_type, __be16 vlan_tci, bool log);
25cd9ba0 1685
e6445719
PS		 1686static int validate_and_copy_sample(const struct nlattr *attr,
1687 const struct sw_flow_key *key, int depth,
25cd9ba0 1688 struct sw_flow_actions **sfa,
05da5898 1689 __be16 eth_type, __be16 vlan_tci, bool log)
e6445719
PS		 1690{
1691 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1692 const struct nlattr *probability, *actions;
1693 const struct nlattr *a;
1694 int rem, start, err, st_acts;
1695
1696 memset(attrs, 0, sizeof(attrs));
1697 nla_for_each_nested(a, attr, rem) {
1698 int type = nla_type(a);
1699 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1700 return -EINVAL;
1701 attrs[type] = a;
1702 }
1703 if (rem)
1704 return -EINVAL;
1705
1706 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1707 if (!probability || nla_len(probability) != sizeof(u32))
1708 return -EINVAL;
1709
1710 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1711 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1712 return -EINVAL;
1713
1714 /* validation done, copy sample action. */
05da5898 1715 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
e6445719
PS		 1716 if (start < 0)
1717 return start;
1718 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
05da5898 1719 nla_data(probability), sizeof(u32), log);
e6445719
PS		 1720 if (err)
1721 return err;
05da5898 1722 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
e6445719
PS		 1723 if (st_acts < 0)
1724 return st_acts;
1725
2fdb957d 1726 err = __ovs_nla_copy_actions(actions, key, depth + 1, sfa,
05da5898 1727 eth_type, vlan_tci, log);
e6445719
PS		 1728 if (err)
1729 return err;
1730
1731 add_nested_action_end(*sfa, st_acts);
1732 add_nested_action_end(*sfa, start);
1733
1734 return 0;
1735}
1736
e6445719
PS		 1737void ovs_match_init(struct sw_flow_match *match,
1738 struct sw_flow_key *key,
1739 struct sw_flow_mask *mask)
1740{
1741 memset(match, 0, sizeof(*match));
1742 match->key = key;
1743 match->mask = mask;
1744
1745 memset(key, 0, sizeof(*key));
1746
1747 if (mask) {
1748 memset(&mask->key, 0, sizeof(mask->key));
1749 mask->range.start = mask->range.end = 0;
1750 }
1751}
1752
d91641d9
TG		 1753static int validate_geneve_opts(struct sw_flow_key *key)
1754{
1755 struct geneve_opt *option;
1756 int opts_len = key->tun_opts_len;
1757 bool crit_opt = false;
1758
1759 option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1760 while (opts_len > 0) {
1761 int len;
1762
1763 if (opts_len < sizeof(*option))
1764 return -EINVAL;
1765
1766 len = sizeof(*option) + option->length * 4;
1767 if (len > opts_len)
1768 return -EINVAL;
1769
1770 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1771
1772 option = (struct geneve_opt *)((u8 *)option + len);
1773 opts_len -= len;
1774 };
1775
1776 key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1777
1778 return 0;
1779}
1780
e6445719 1781static int validate_and_copy_set_tun(const struct nlattr *attr,
05da5898 1782 struct sw_flow_actions **sfa, bool log)
e6445719
PS		 1783{
1784 struct sw_flow_match match;
1785 struct sw_flow_key key;
34ae932a 1786 struct metadata_dst *tun_dst;
1d8fff90 1787 struct ip_tunnel_info *tun_info;
34ae932a 1788 struct ovs_tunnel_info *ovs_tun;
f0b128c1 1789 struct nlattr *a;
13101602 1790 int err = 0, start, opts_type;
e6445719
PS		 1791
1792 ovs_match_init(&match, &key, NULL);
1dd144cf
TG		 1793 opts_type = ipv4_tun_from_nlattr(nla_data(attr), &match, false, log);
1794 if (opts_type < 0)
1795 return opts_type;
e6445719 1796
f5796684 1797 if (key.tun_opts_len) {
1dd144cf
TG		 1798 switch (opts_type) {
1799 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1800 err = validate_geneve_opts(&key);
1801 if (err < 0)
1802 return err;
1803 break;
1804 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1805 break;
1806 }
f5796684
JG		 1807 };
1808
05da5898 1809 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
e6445719
PS		 1810 if (start < 0)
1811 return start;
1812
34ae932a
TG		 1813 tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1814 if (!tun_dst)
1815 return -ENOMEM;
1816
f0b128c1 1817 a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
34ae932a
TG		 1818 sizeof(*ovs_tun), log);
1819 if (IS_ERR(a)) {
1820 dst_release((struct dst_entry *)tun_dst);
f0b128c1 1821 return PTR_ERR(a);
34ae932a
TG		 1822 }
1823
1824 ovs_tun = nla_data(a);
1825 ovs_tun->tun_dst = tun_dst;
f0b128c1 1826
34ae932a
TG		 1827 tun_info = &tun_dst->u.tun_info;
1828 tun_info->mode = IP_TUNNEL_INFO_TX;
1d8fff90 1829 tun_info->key = key.tun_key;
f5796684
JG		 1830 tun_info->options_len = key.tun_opts_len;
1831
1832 if (tun_info->options_len) {
1833 /* We need to store the options in the action itself since
1834 * everything else will go away after flow setup. We can append
1835 * it to tun_info and then point there.
1836 */
d91641d9
TG		 1837 memcpy((tun_info + 1),
1838 TUN_METADATA_OPTS(&key, key.tun_opts_len), key.tun_opts_len);
1839 tun_info->options = (tun_info + 1);
f5796684
JG		 1840 } else {
1841 tun_info->options = NULL;
1842 }
f0b128c1 1843
e6445719
PS		 1844 add_nested_action_end(*sfa, start);
1845
1846 return err;
1847}
1848
83d2b9ba
JR		 1849/* Return false if there are any non-masked bits set.
1850 * Mask follows data immediately, before any netlink padding.
1851 */
1852static bool validate_masked(u8 *data, int len)
1853{
1854 u8 *mask = data + len;
1855
1856 while (len--)
1857 if (*data++ & ~*mask++)
1858 return false;
1859
1860 return true;
1861}
1862
e6445719
PS		 1863static int validate_set(const struct nlattr *a,
1864 const struct sw_flow_key *flow_key,
1865 struct sw_flow_actions **sfa,
83d2b9ba 1866 bool *skip_copy, __be16 eth_type, bool masked, bool log)
e6445719
PS		 1867{
1868 const struct nlattr *ovs_key = nla_data(a);
1869 int key_type = nla_type(ovs_key);
83d2b9ba 1870 size_t key_len;
e6445719
PS		 1871
1872 /* There can be only one key in a action */
1873 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
1874 return -EINVAL;
1875
83d2b9ba
JR		 1876 key_len = nla_len(ovs_key);
1877 if (masked)
1878 key_len /= 2;
1879
e6445719 1880 if (key_type > OVS_KEY_ATTR_MAX ||
83d2b9ba 1881 (ovs_key_lens[key_type].len != key_len &&
81bfe3c3 1882 ovs_key_lens[key_type].len != OVS_ATTR_NESTED))
e6445719
PS		 1883 return -EINVAL;
1884
83d2b9ba
JR		 1885 if (masked && !validate_masked(nla_data(ovs_key), key_len))
1886 return -EINVAL;
1887
e6445719
PS		 1888 switch (key_type) {
1889 const struct ovs_key_ipv4 *ipv4_key;
1890 const struct ovs_key_ipv6 *ipv6_key;
1891 int err;
1892
1893 case OVS_KEY_ATTR_PRIORITY:
1894 case OVS_KEY_ATTR_SKB_MARK:
1895 case OVS_KEY_ATTR_ETHERNET:
1896 break;
1897
1898 case OVS_KEY_ATTR_TUNNEL:
25cd9ba0
SH		 1899 if (eth_p_mpls(eth_type))
1900 return -EINVAL;
1901
83d2b9ba
JR		 1902 if (masked)
1903 return -EINVAL; /* Masked tunnel set not supported. */
1904
1905 *skip_copy = true;
05da5898 1906 err = validate_and_copy_set_tun(a, sfa, log);
e6445719
PS		 1907 if (err)
1908 return err;
1909 break;
1910
1911 case OVS_KEY_ATTR_IPV4:
25cd9ba0 1912 if (eth_type != htons(ETH_P_IP))
e6445719
PS		 1913 return -EINVAL;
1914
e6445719 1915 ipv4_key = nla_data(ovs_key);
e6445719 1916
83d2b9ba
JR		 1917 if (masked) {
1918 const struct ovs_key_ipv4 *mask = ipv4_key + 1;
e6445719 1919
83d2b9ba
JR		 1920 /* Non-writeable fields. */
1921 if (mask->ipv4_proto || mask->ipv4_frag)
1922 return -EINVAL;
1923 } else {
1924 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
1925 return -EINVAL;
1926
1927 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
1928 return -EINVAL;
1929 }
e6445719
PS		 1930 break;
1931
1932 case OVS_KEY_ATTR_IPV6:
25cd9ba0 1933 if (eth_type != htons(ETH_P_IPV6))
e6445719
PS		 1934 return -EINVAL;
1935
e6445719 1936 ipv6_key = nla_data(ovs_key);
e6445719 1937
83d2b9ba
JR		 1938 if (masked) {
1939 const struct ovs_key_ipv6 *mask = ipv6_key + 1;
1940
1941 /* Non-writeable fields. */
1942 if (mask->ipv6_proto || mask->ipv6_frag)
1943 return -EINVAL;
1944
1945 /* Invalid bits in the flow label mask? */
1946 if (ntohl(mask->ipv6_label) & 0xFFF00000)
1947 return -EINVAL;
1948 } else {
1949 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
1950 return -EINVAL;
e6445719 1951
83d2b9ba
JR		 1952 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
1953 return -EINVAL;
1954 }
e6445719
PS		 1955 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
1956 return -EINVAL;
1957
1958 break;
1959
1960 case OVS_KEY_ATTR_TCP:
83d2b9ba
JR		 1961 if ((eth_type != htons(ETH_P_IP) &&
1962 eth_type != htons(ETH_P_IPV6)) ||
1963 flow_key->ip.proto != IPPROTO_TCP)
e6445719
PS		 1964 return -EINVAL;
1965
83d2b9ba 1966 break;
e6445719
PS		 1967
1968 case OVS_KEY_ATTR_UDP:
83d2b9ba
JR		 1969 if ((eth_type != htons(ETH_P_IP) &&
1970 eth_type != htons(ETH_P_IPV6)) ||
1971 flow_key->ip.proto != IPPROTO_UDP)
e6445719
PS		 1972 return -EINVAL;
1973
83d2b9ba 1974 break;
25cd9ba0
SH		 1975
1976 case OVS_KEY_ATTR_MPLS:
1977 if (!eth_p_mpls(eth_type))
1978 return -EINVAL;
1979 break;
e6445719
PS		 1980
1981 case OVS_KEY_ATTR_SCTP:
83d2b9ba
JR		 1982 if ((eth_type != htons(ETH_P_IP) &&
1983 eth_type != htons(ETH_P_IPV6)) ||
1984 flow_key->ip.proto != IPPROTO_SCTP)
e6445719
PS		 1985 return -EINVAL;
1986
83d2b9ba 1987 break;
e6445719
PS		 1988
1989 default:
1990 return -EINVAL;
1991 }
1992
83d2b9ba
JR		 1993 /* Convert non-masked non-tunnel set actions to masked set actions. */
1994 if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
1995 int start, len = key_len * 2;
1996 struct nlattr *at;
1997
1998 *skip_copy = true;
1999
2000 start = add_nested_action_start(sfa,
2001 OVS_ACTION_ATTR_SET_TO_MASKED,
2002 log);
2003 if (start < 0)
2004 return start;
2005
2006 at = __add_action(sfa, key_type, NULL, len, log);
2007 if (IS_ERR(at))
2008 return PTR_ERR(at);
2009
2010 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2011 memset(nla_data(at) + key_len, 0xff, key_len); /* Mask. */
2012 /* Clear non-writeable bits from otherwise writeable fields. */
2013 if (key_type == OVS_KEY_ATTR_IPV6) {
2014 struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2015
2016 mask->ipv6_label &= htonl(0x000FFFFF);
2017 }
2018 add_nested_action_end(*sfa, start);
2019 }
2020
e6445719
PS		 2021 return 0;
2022}
2023
2024static int validate_userspace(const struct nlattr *attr)
2025{
2026 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2027 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2028 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
8f0aad6f 2029 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
e6445719
PS		 2030 };
2031 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2032 int error;
2033
2034 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2035 attr, userspace_policy);
2036 if (error)
2037 return error;
2038
2039 if (!a[OVS_USERSPACE_ATTR_PID] ||
2040 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2041 return -EINVAL;
2042
2043 return 0;
2044}
2045
2046static int copy_action(const struct nlattr *from,
05da5898 2047 struct sw_flow_actions **sfa, bool log)
e6445719
PS		 2048{
2049 int totlen = NLA_ALIGN(from->nla_len);
2050 struct nlattr *to;
2051
05da5898 2052 to = reserve_sfa_size(sfa, from->nla_len, log);
e6445719
PS		 2053 if (IS_ERR(to))
2054 return PTR_ERR(to);
2055
2056 memcpy(to, from, totlen);
2057 return 0;
2058}
2059
2fdb957d 2060static int __ovs_nla_copy_actions(const struct nlattr *attr,
25cd9ba0
SH		 2061 const struct sw_flow_key *key,
2062 int depth, struct sw_flow_actions **sfa,
05da5898 2063 __be16 eth_type, __be16 vlan_tci, bool log)
e6445719
PS		 2064{
2065 const struct nlattr *a;
2066 int rem, err;
2067
2068 if (depth >= SAMPLE_ACTION_DEPTH)
2069 return -EOVERFLOW;
2070
2071 nla_for_each_nested(a, attr, rem) {
2072 /* Expected argument lengths, (u32)-1 for variable length. */
2073 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2074 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
971427f3 2075 [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
e6445719 2076 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
25cd9ba0
SH		 2077 [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2078 [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
e6445719
PS		 2079 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2080 [OVS_ACTION_ATTR_POP_VLAN] = 0,
2081 [OVS_ACTION_ATTR_SET] = (u32)-1,
83d2b9ba 2082 [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
971427f3
AZ		 2083 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
2084 [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash)
e6445719
PS		 2085 };
2086 const struct ovs_action_push_vlan *vlan;
2087 int type = nla_type(a);
2088 bool skip_copy;
2089
2090 if (type > OVS_ACTION_ATTR_MAX ||
2091 (action_lens[type] != nla_len(a) &&
2092 action_lens[type] != (u32)-1))
2093 return -EINVAL;
2094
2095 skip_copy = false;
2096 switch (type) {
2097 case OVS_ACTION_ATTR_UNSPEC:
2098 return -EINVAL;
2099
2100 case OVS_ACTION_ATTR_USERSPACE:
2101 err = validate_userspace(a);
2102 if (err)
2103 return err;
2104 break;
2105
2106 case OVS_ACTION_ATTR_OUTPUT:
2107 if (nla_get_u32(a) >= DP_MAX_PORTS)
2108 return -EINVAL;
2109 break;
2110
971427f3
AZ		 2111 case OVS_ACTION_ATTR_HASH: {
2112 const struct ovs_action_hash *act_hash = nla_data(a);
2113
2114 switch (act_hash->hash_alg) {
2115 case OVS_HASH_ALG_L4:
2116 break;
2117 default:
2118 return -EINVAL;
2119 }
2120
2121 break;
2122 }
e6445719
PS		 2123
2124 case OVS_ACTION_ATTR_POP_VLAN:
25cd9ba0 2125 vlan_tci = htons(0);
e6445719
PS		 2126 break;
2127
2128 case OVS_ACTION_ATTR_PUSH_VLAN:
2129 vlan = nla_data(a);
2130 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2131 return -EINVAL;
2132 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2133 return -EINVAL;
25cd9ba0 2134 vlan_tci = vlan->vlan_tci;
e6445719
PS		 2135 break;
2136
971427f3
AZ		 2137 case OVS_ACTION_ATTR_RECIRC:
2138 break;
2139
25cd9ba0
SH		 2140 case OVS_ACTION_ATTR_PUSH_MPLS: {
2141 const struct ovs_action_push_mpls *mpls = nla_data(a);
2142
25cd9ba0
SH		 2143 if (!eth_p_mpls(mpls->mpls_ethertype))
2144 return -EINVAL;
2145 /* Prohibit push MPLS other than to a white list
2146 * for packets that have a known tag order.
2147 */
2148 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2149 (eth_type != htons(ETH_P_IP) &&
2150 eth_type != htons(ETH_P_IPV6) &&
2151 eth_type != htons(ETH_P_ARP) &&
2152 eth_type != htons(ETH_P_RARP) &&
2153 !eth_p_mpls(eth_type)))
2154 return -EINVAL;
2155 eth_type = mpls->mpls_ethertype;
2156 break;
2157 }
2158
2159 case OVS_ACTION_ATTR_POP_MPLS:
2160 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2161 !eth_p_mpls(eth_type))
2162 return -EINVAL;
2163
2164 /* Disallow subsequent L2.5+ set and mpls_pop actions
2165 * as there is no check here to ensure that the new
2166 * eth_type is valid and thus set actions could
2167 * write off the end of the packet or otherwise
2168 * corrupt it.
2169 *
2170 * Support for these actions is planned using packet
2171 * recirculation.
2172 */
2173 eth_type = htons(0);
2174 break;
2175
e6445719 2176 case OVS_ACTION_ATTR_SET:
25cd9ba0 2177 err = validate_set(a, key, sfa,
83d2b9ba
JR		 2178 &skip_copy, eth_type, false, log);
2179 if (err)
2180 return err;
2181 break;
2182
2183 case OVS_ACTION_ATTR_SET_MASKED:
2184 err = validate_set(a, key, sfa,
2185 &skip_copy, eth_type, true, log);
e6445719
PS		 2186 if (err)
2187 return err;
2188 break;
2189
2190 case OVS_ACTION_ATTR_SAMPLE:
25cd9ba0 2191 err = validate_and_copy_sample(a, key, depth, sfa,
05da5898 2192 eth_type, vlan_tci, log);
e6445719
PS		 2193 if (err)
2194 return err;
2195 skip_copy = true;
2196 break;
2197
2198 default:
05da5898 2199 OVS_NLERR(log, "Unknown Action type %d", type);
e6445719
PS		 2200 return -EINVAL;
2201 }
2202 if (!skip_copy) {
05da5898 2203 err = copy_action(a, sfa, log);
e6445719
PS		 2204 if (err)
2205 return err;
2206 }
2207 }
2208
2209 if (rem > 0)
2210 return -EINVAL;
2211
2212 return 0;
2213}
2214
83d2b9ba 2215/* 'key' must be the masked key. */
25cd9ba0
SH		 2216int ovs_nla_copy_actions(const struct nlattr *attr,
2217 const struct sw_flow_key *key,
05da5898 2218 struct sw_flow_actions **sfa, bool log)
25cd9ba0 2219{
2fdb957d
PS		 2220 int err;
2221
05da5898 2222 *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2fdb957d
PS		 2223 if (IS_ERR(*sfa))
2224 return PTR_ERR(*sfa);
2225
2226 err = __ovs_nla_copy_actions(attr, key, 0, sfa, key->eth.type,
05da5898 2227 key->eth.tci, log);
2fdb957d 2228 if (err)
34ae932a 2229 ovs_nla_free_flow_actions(*sfa);
2fdb957d
PS		 2230
2231 return err;
25cd9ba0
SH		 2232}
2233
e6445719
PS		 2234static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2235{
2236 const struct nlattr *a;
2237 struct nlattr *start;
2238 int err = 0, rem;
2239
2240 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2241 if (!start)
2242 return -EMSGSIZE;
2243
2244 nla_for_each_nested(a, attr, rem) {
2245 int type = nla_type(a);
2246 struct nlattr *st_sample;
2247
2248 switch (type) {
2249 case OVS_SAMPLE_ATTR_PROBABILITY:
2250 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2251 sizeof(u32), nla_data(a)))
2252 return -EMSGSIZE;
2253 break;
2254 case OVS_SAMPLE_ATTR_ACTIONS:
2255 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2256 if (!st_sample)
2257 return -EMSGSIZE;
2258 err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2259 if (err)
2260 return err;
2261 nla_nest_end(skb, st_sample);
2262 break;
2263 }
2264 }
2265
2266 nla_nest_end(skb, start);
2267 return err;
2268}
2269
2270static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2271{
2272 const struct nlattr *ovs_key = nla_data(a);
2273 int key_type = nla_type(ovs_key);
2274 struct nlattr *start;
2275 int err;
2276
2277 switch (key_type) {
f0b128c1 2278 case OVS_KEY_ATTR_TUNNEL_INFO: {
34ae932a
TG		 2279 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2280 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
f0b128c1 2281
e6445719
PS		 2282 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2283 if (!start)
2284 return -EMSGSIZE;
2285
1d8fff90 2286 err = ipv4_tun_to_nlattr(skb, &tun_info->key,
f5796684
JG		 2287 tun_info->options_len ?
2288 tun_info->options : NULL,
2289 tun_info->options_len);
e6445719
PS		 2290 if (err)
2291 return err;
2292 nla_nest_end(skb, start);
2293 break;
f0b128c1 2294 }
e6445719
PS		 2295 default:
2296 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2297 return -EMSGSIZE;
2298 break;
2299 }
2300
2301 return 0;
2302}
2303
83d2b9ba
JR		 2304static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2305 struct sk_buff *skb)
2306{
2307 const struct nlattr *ovs_key = nla_data(a);
f4f8e738 2308 struct nlattr *nla;
83d2b9ba
JR		 2309 size_t key_len = nla_len(ovs_key) / 2;
2310
2311 /* Revert the conversion we did from a non-masked set action to
2312 * masked set action.
2313 */
f4f8e738
JS		 2314 nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2315 if (!nla)
83d2b9ba
JR		 2316 return -EMSGSIZE;
2317
f4f8e738
JS		 2318 if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2319 return -EMSGSIZE;
2320
2321 nla_nest_end(skb, nla);
83d2b9ba
JR		 2322 return 0;
2323}
2324
e6445719
PS		 2325int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2326{
2327 const struct nlattr *a;
2328 int rem, err;
2329
2330 nla_for_each_attr(a, attr, len, rem) {
2331 int type = nla_type(a);
2332
2333 switch (type) {
2334 case OVS_ACTION_ATTR_SET:
2335 err = set_action_to_attr(a, skb);
2336 if (err)
2337 return err;
2338 break;
2339
83d2b9ba
JR		 2340 case OVS_ACTION_ATTR_SET_TO_MASKED:
2341 err = masked_set_action_to_set_action_attr(a, skb);
2342 if (err)
2343 return err;
2344 break;
2345
e6445719
PS		 2346 case OVS_ACTION_ATTR_SAMPLE:
2347 err = sample_action_to_attr(a, skb);
2348 if (err)
2349 return err;
2350 break;
2351 default:
2352 if (nla_put(skb, type, nla_len(a), nla_data(a)))
2353 return -EMSGSIZE;
2354 break;
2355 }
2356 }
2357
2358 return 0;
2359}
Linux kernel - Deliverables
This page took 0.234266 seconds and 5 git commands to generate.