]> git.proxmox.com Git - mirror_ovs.git/blob - lib/tc.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
lib/tc: add single mpls match offload support
[mirror_ovs.git] / lib / tc.c
1 /*
2 * Copyright (c) 2009-2017 Nicira, Inc.
3 * Copyright (c) 2016 Mellanox Technologies, Ltd.
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18 #include <config.h>
19 #include "tc.h"
20
21 #include <errno.h>
22 #include <linux/if_ether.h>
23 #include <linux/rtnetlink.h>
24 #include <linux/tc_act/tc_csum.h>
25 #include <linux/tc_act/tc_gact.h>
26 #include <linux/tc_act/tc_mirred.h>
27 #include <linux/tc_act/tc_pedit.h>
28 #include <linux/tc_act/tc_tunnel_key.h>
29 #include <linux/tc_act/tc_vlan.h>
30 #include <linux/gen_stats.h>
31 #include <net/if.h>
32 #include <unistd.h>
33
34 #include "byte-order.h"
35 #include "netlink-socket.h"
36 #include "netlink.h"
37 #include "openvswitch/ofpbuf.h"
38 #include "openvswitch/util.h"
39 #include "openvswitch/vlog.h"
40 #include "packets.h"
41 #include "timeval.h"
42 #include "unaligned.h"
43
44 #define MAX_PEDIT_OFFSETS 32
45
46 #ifndef TCM_IFINDEX_MAGIC_BLOCK
47 #define TCM_IFINDEX_MAGIC_BLOCK (0xFFFFFFFFU)
48 #endif
49
50 #if TCA_MAX < 14
51 #define TCA_INGRESS_BLOCK 13
52 #endif
53
54 VLOG_DEFINE_THIS_MODULE(tc);
55
56 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
57
58 enum tc_offload_policy {
59 TC_POLICY_NONE,
60 TC_POLICY_SKIP_SW,
61 TC_POLICY_SKIP_HW
62 };
63
64 static enum tc_offload_policy tc_policy = TC_POLICY_NONE;
65
66 struct tc_pedit_key_ex {
67 enum pedit_header_type htype;
68 enum pedit_cmd cmd;
69 };
70
71 struct flower_key_to_pedit {
72 enum pedit_header_type htype;
73 int offset;
74 int flower_offset;
75 int size;
76 };
77
78 static struct flower_key_to_pedit flower_pedit_map[] = {
79 {
80 TCA_PEDIT_KEY_EX_HDR_TYPE_IP4,
81 12,
82 offsetof(struct tc_flower_key, ipv4.ipv4_src),
83 MEMBER_SIZEOF(struct tc_flower_key, ipv4.ipv4_src)
84 }, {
85 TCA_PEDIT_KEY_EX_HDR_TYPE_IP4,
86 16,
87 offsetof(struct tc_flower_key, ipv4.ipv4_dst),
88 MEMBER_SIZEOF(struct tc_flower_key, ipv4.ipv4_dst)
89 }, {
90 TCA_PEDIT_KEY_EX_HDR_TYPE_IP4,
91 8,
92 offsetof(struct tc_flower_key, ipv4.rewrite_ttl),
93 MEMBER_SIZEOF(struct tc_flower_key, ipv4.rewrite_ttl)
94 }, {
95 TCA_PEDIT_KEY_EX_HDR_TYPE_IP6,
96 8,
97 offsetof(struct tc_flower_key, ipv6.ipv6_src),
98 MEMBER_SIZEOF(struct tc_flower_key, ipv6.ipv6_src)
99 }, {
100 TCA_PEDIT_KEY_EX_HDR_TYPE_IP6,
101 24,
102 offsetof(struct tc_flower_key, ipv6.ipv6_dst),
103 MEMBER_SIZEOF(struct tc_flower_key, ipv6.ipv6_dst)
104 }, {
105 TCA_PEDIT_KEY_EX_HDR_TYPE_ETH,
106 6,
107 offsetof(struct tc_flower_key, src_mac),
108 MEMBER_SIZEOF(struct tc_flower_key, src_mac)
109 }, {
110 TCA_PEDIT_KEY_EX_HDR_TYPE_ETH,
111 0,
112 offsetof(struct tc_flower_key, dst_mac),
113 MEMBER_SIZEOF(struct tc_flower_key, dst_mac)
114 }, {
115 TCA_PEDIT_KEY_EX_HDR_TYPE_ETH,
116 12,
117 offsetof(struct tc_flower_key, eth_type),
118 MEMBER_SIZEOF(struct tc_flower_key, eth_type)
119 }, {
120 TCA_PEDIT_KEY_EX_HDR_TYPE_TCP,
121 0,
122 offsetof(struct tc_flower_key, tcp_src),
123 MEMBER_SIZEOF(struct tc_flower_key, tcp_src)
124 }, {
125 TCA_PEDIT_KEY_EX_HDR_TYPE_TCP,
126 2,
127 offsetof(struct tc_flower_key, tcp_dst),
128 MEMBER_SIZEOF(struct tc_flower_key, tcp_dst)
129 }, {
130 TCA_PEDIT_KEY_EX_HDR_TYPE_UDP,
131 0,
132 offsetof(struct tc_flower_key, udp_src),
133 MEMBER_SIZEOF(struct tc_flower_key, udp_src)
134 }, {
135 TCA_PEDIT_KEY_EX_HDR_TYPE_UDP,
136 2,
137 offsetof(struct tc_flower_key, udp_dst),
138 MEMBER_SIZEOF(struct tc_flower_key, udp_dst)
139 },
140 };
141
142 static inline int
143 csum_update_flag(struct tc_flower *flower,
144 enum pedit_header_type htype);
145
146 struct tcmsg *
147 tc_make_request(int ifindex, int type, unsigned int flags,
148 struct ofpbuf *request)
149 {
150 struct tcmsg *tcmsg;
151
152 ofpbuf_init(request, 512);
153 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
154 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
155 tcmsg->tcm_family = AF_UNSPEC;
156 tcmsg->tcm_ifindex = ifindex;
157 /* Caller should fill in tcmsg->tcm_handle. */
158 /* Caller should fill in tcmsg->tcm_parent. */
159
160 return tcmsg;
161 }
162
163 int
164 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
165 {
166 int error = nl_transact(NETLINK_ROUTE, request, replyp);
167 ofpbuf_uninit(request);
168 return error;
169 }
170
171 /* Adds or deletes a root ingress qdisc on device with specified ifindex.
172 *
173 * This function is equivalent to running the following when 'add' is true:
174 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
175 *
176 * This function is equivalent to running the following when 'add' is false:
177 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
178 *
179 * Where dev <devname> is the device with specified ifindex name.
180 *
181 * The configuration and stats may be seen with the following command:
182 * /sbin/tc -s qdisc show dev <devname>
183 *
184 * If block_id is greater than 0, then the ingress qdisc is added to a block.
185 * In this case, it is equivalent to running (when 'add' is true):
186 * /sbin/tc qdisc add dev <devname> ingress_block <block_id> ingress
187 *
188 * Returns 0 if successful, otherwise a positive errno value.
189 */
190 int
191 tc_add_del_ingress_qdisc(int ifindex, bool add, uint32_t block_id)
192 {
193 struct ofpbuf request;
194 struct tcmsg *tcmsg;
195 int error;
196 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
197 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
198
199 tcmsg = tc_make_request(ifindex, type, flags, &request);
200 tcmsg->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
201 tcmsg->tcm_parent = TC_H_INGRESS;
202 nl_msg_put_string(&request, TCA_KIND, "ingress");
203 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
204 if (block_id) {
205 nl_msg_put_u32(&request, TCA_INGRESS_BLOCK, block_id);
206 }
207
208 error = tc_transact(&request, NULL);
209 if (error) {
210 /* If we're deleting the qdisc, don't worry about some of the
211 * error conditions. */
212 if (!add && (error == ENOENT || error == EINVAL)) {
213 return 0;
214 }
215 return error;
216 }
217
218 return 0;
219 }
220
221 static const struct nl_policy tca_policy[] = {
222 [TCA_KIND] = { .type = NL_A_STRING, .optional = false, },
223 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false, },
224 [TCA_STATS] = { .type = NL_A_UNSPEC,
225 .min_len = sizeof(struct tc_stats), .optional = true, },
226 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = true, },
227 };
228
229 static const struct nl_policy tca_flower_policy[] = {
230 [TCA_FLOWER_CLASSID] = { .type = NL_A_U32, .optional = true, },
231 [TCA_FLOWER_INDEV] = { .type = NL_A_STRING, .max_len = IFNAMSIZ,
232 .optional = true, },
233 [TCA_FLOWER_KEY_ETH_SRC] = { .type = NL_A_UNSPEC,
234 .min_len = ETH_ALEN, .optional = true, },
235 [TCA_FLOWER_KEY_ETH_DST] = { .type = NL_A_UNSPEC,
236 .min_len = ETH_ALEN, .optional = true, },
237 [TCA_FLOWER_KEY_ETH_SRC_MASK] = { .type = NL_A_UNSPEC,
238 .min_len = ETH_ALEN,
239 .optional = true, },
240 [TCA_FLOWER_KEY_ETH_DST_MASK] = { .type = NL_A_UNSPEC,
241 .min_len = ETH_ALEN,
242 .optional = true, },
243 [TCA_FLOWER_KEY_ETH_TYPE] = { .type = NL_A_U16, .optional = false, },
244 [TCA_FLOWER_FLAGS] = { .type = NL_A_U32, .optional = false, },
245 [TCA_FLOWER_ACT] = { .type = NL_A_NESTED, .optional = false, },
246 [TCA_FLOWER_KEY_IP_PROTO] = { .type = NL_A_U8, .optional = true, },
247 [TCA_FLOWER_KEY_IPV4_SRC] = { .type = NL_A_U32, .optional = true, },
248 [TCA_FLOWER_KEY_IPV4_DST] = {.type = NL_A_U32, .optional = true, },
249 [TCA_FLOWER_KEY_IPV4_SRC_MASK] = { .type = NL_A_U32, .optional = true, },
250 [TCA_FLOWER_KEY_IPV4_DST_MASK] = { .type = NL_A_U32, .optional = true, },
251 [TCA_FLOWER_KEY_IPV6_SRC] = { .type = NL_A_UNSPEC,
252 .min_len = sizeof(struct in6_addr),
253 .optional = true, },
254 [TCA_FLOWER_KEY_IPV6_DST] = { .type = NL_A_UNSPEC,
255 .min_len = sizeof(struct in6_addr),
256 .optional = true, },
257 [TCA_FLOWER_KEY_IPV6_SRC_MASK] = { .type = NL_A_UNSPEC,
258 .min_len = sizeof(struct in6_addr),
259 .optional = true, },
260 [TCA_FLOWER_KEY_IPV6_DST_MASK] = { .type = NL_A_UNSPEC,
261 .min_len = sizeof(struct in6_addr),
262 .optional = true, },
263 [TCA_FLOWER_KEY_TCP_SRC] = { .type = NL_A_U16, .optional = true, },
264 [TCA_FLOWER_KEY_TCP_DST] = { .type = NL_A_U16, .optional = true, },
265 [TCA_FLOWER_KEY_TCP_SRC_MASK] = { .type = NL_A_U16, .optional = true, },
266 [TCA_FLOWER_KEY_TCP_DST_MASK] = { .type = NL_A_U16, .optional = true, },
267 [TCA_FLOWER_KEY_UDP_SRC] = { .type = NL_A_U16, .optional = true, },
268 [TCA_FLOWER_KEY_UDP_DST] = { .type = NL_A_U16, .optional = true, },
269 [TCA_FLOWER_KEY_UDP_SRC_MASK] = { .type = NL_A_U16, .optional = true, },
270 [TCA_FLOWER_KEY_UDP_DST_MASK] = { .type = NL_A_U16, .optional = true, },
271 [TCA_FLOWER_KEY_SCTP_SRC] = { .type = NL_A_U16, .optional = true, },
272 [TCA_FLOWER_KEY_SCTP_DST] = { .type = NL_A_U16, .optional = true, },
273 [TCA_FLOWER_KEY_SCTP_SRC_MASK] = { .type = NL_A_U16, .optional = true, },
274 [TCA_FLOWER_KEY_SCTP_DST_MASK] = { .type = NL_A_U16, .optional = true, },
275 [TCA_FLOWER_KEY_MPLS_TTL] = { .type = NL_A_U8, .optional = true, },
276 [TCA_FLOWER_KEY_MPLS_TC] = { .type = NL_A_U8, .optional = true, },
277 [TCA_FLOWER_KEY_MPLS_BOS] = { .type = NL_A_U8, .optional = true, },
278 [TCA_FLOWER_KEY_MPLS_LABEL] = { .type = NL_A_U32, .optional = true, },
279 [TCA_FLOWER_KEY_VLAN_ID] = { .type = NL_A_U16, .optional = true, },
280 [TCA_FLOWER_KEY_VLAN_PRIO] = { .type = NL_A_U8, .optional = true, },
281 [TCA_FLOWER_KEY_VLAN_ETH_TYPE] = { .type = NL_A_U16, .optional = true, },
282 [TCA_FLOWER_KEY_ENC_KEY_ID] = { .type = NL_A_U32, .optional = true, },
283 [TCA_FLOWER_KEY_ENC_IPV4_SRC] = { .type = NL_A_U32, .optional = true, },
284 [TCA_FLOWER_KEY_ENC_IPV4_DST] = { .type = NL_A_U32, .optional = true, },
285 [TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK] = { .type = NL_A_U32,
286 .optional = true, },
287 [TCA_FLOWER_KEY_ENC_IPV4_DST_MASK] = { .type = NL_A_U32,
288 .optional = true, },
289 [TCA_FLOWER_KEY_ENC_IPV6_SRC] = { .type = NL_A_UNSPEC,
290 .min_len = sizeof(struct in6_addr),
291 .optional = true, },
292 [TCA_FLOWER_KEY_ENC_IPV6_DST] = { .type = NL_A_UNSPEC,
293 .min_len = sizeof(struct in6_addr),
294 .optional = true, },
295 [TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK] = { .type = NL_A_UNSPEC,
296 .min_len = sizeof(struct in6_addr),
297 .optional = true, },
298 [TCA_FLOWER_KEY_ENC_IPV6_DST_MASK] = { .type = NL_A_UNSPEC,
299 .min_len = sizeof(struct in6_addr),
300 .optional = true, },
301 [TCA_FLOWER_KEY_ENC_UDP_DST_PORT] = { .type = NL_A_U16,
302 .optional = true, },
303 [TCA_FLOWER_KEY_FLAGS] = { .type = NL_A_BE32, .optional = true, },
304 [TCA_FLOWER_KEY_FLAGS_MASK] = { .type = NL_A_BE32, .optional = true, },
305 [TCA_FLOWER_KEY_IP_TTL] = { .type = NL_A_U8,
306 .optional = true, },
307 [TCA_FLOWER_KEY_IP_TTL_MASK] = { .type = NL_A_U8,
308 .optional = true, },
309 [TCA_FLOWER_KEY_IP_TOS] = { .type = NL_A_U8,
310 .optional = true, },
311 [TCA_FLOWER_KEY_IP_TOS_MASK] = { .type = NL_A_U8,
312 .optional = true, },
313 [TCA_FLOWER_KEY_TCP_FLAGS] = { .type = NL_A_U16,
314 .optional = true, },
315 [TCA_FLOWER_KEY_TCP_FLAGS_MASK] = { .type = NL_A_U16,
316 .optional = true, },
317 [TCA_FLOWER_KEY_CVLAN_ID] = { .type = NL_A_U16, .optional = true, },
318 [TCA_FLOWER_KEY_CVLAN_PRIO] = { .type = NL_A_U8, .optional = true, },
319 [TCA_FLOWER_KEY_CVLAN_ETH_TYPE] = { .type = NL_A_U16, .optional = true, },
320 [TCA_FLOWER_KEY_ENC_IP_TOS] = { .type = NL_A_U8,
321 .optional = true, },
322 [TCA_FLOWER_KEY_ENC_IP_TOS_MASK] = { .type = NL_A_U8,
323 .optional = true, },
324 [TCA_FLOWER_KEY_ENC_IP_TTL] = { .type = NL_A_U8,
325 .optional = true, },
326 [TCA_FLOWER_KEY_ENC_IP_TTL_MASK] = { .type = NL_A_U8,
327 .optional = true, },
328 };
329
330 static void
331 nl_parse_flower_eth(struct nlattr **attrs, struct tc_flower *flower)
332 {
333 const struct eth_addr *eth;
334
335 if (attrs[TCA_FLOWER_KEY_ETH_SRC_MASK]) {
336 eth = nl_attr_get_unspec(attrs[TCA_FLOWER_KEY_ETH_SRC], ETH_ALEN);
337 memcpy(&flower->key.src_mac, eth, sizeof flower->key.src_mac);
338
339 eth = nl_attr_get_unspec(attrs[TCA_FLOWER_KEY_ETH_SRC_MASK], ETH_ALEN);
340 memcpy(&flower->mask.src_mac, eth, sizeof flower->mask.src_mac);
341 }
342 if (attrs[TCA_FLOWER_KEY_ETH_DST_MASK]) {
343 eth = nl_attr_get_unspec(attrs[TCA_FLOWER_KEY_ETH_DST], ETH_ALEN);
344 memcpy(&flower->key.dst_mac, eth, sizeof flower->key.dst_mac);
345
346 eth = nl_attr_get_unspec(attrs[TCA_FLOWER_KEY_ETH_DST_MASK], ETH_ALEN);
347 memcpy(&flower->mask.dst_mac, eth, sizeof flower->mask.dst_mac);
348 }
349 }
350
351 static void
352 nl_parse_flower_mpls(struct nlattr **attrs, struct tc_flower *flower)
353 {
354 uint8_t ttl, tc, bos;
355 uint32_t label;
356
357 if (!eth_type_mpls(flower->key.eth_type)) {
358 return;
359 }
360
361 flower->key.encap_eth_type[0] =
362 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_ETH_TYPE]);
363 flower->key.mpls_lse = 0;
364 flower->mask.mpls_lse = 0;
365
366 if (attrs[TCA_FLOWER_KEY_MPLS_TTL]) {
367 ttl = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_MPLS_TTL]);
368 set_mpls_lse_ttl(&flower->key.mpls_lse, ttl);
369 set_mpls_lse_ttl(&flower->mask.mpls_lse, 0xff);
370 }
371
372 if (attrs[TCA_FLOWER_KEY_MPLS_BOS]) {
373 bos = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_MPLS_BOS]);
374 set_mpls_lse_bos(&flower->key.mpls_lse, bos);
375 set_mpls_lse_ttl(&flower->mask.mpls_lse, 0xff);
376 }
377
378 if (attrs[TCA_FLOWER_KEY_MPLS_TC]) {
379 tc = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_MPLS_TC]);
380 set_mpls_lse_tc(&flower->key.mpls_lse, tc);
381 set_mpls_lse_tc(&flower->mask.mpls_lse, 0xff);
382 }
383
384 if (attrs[TCA_FLOWER_KEY_MPLS_LABEL]) {
385 label = nl_attr_get_u32(attrs[TCA_FLOWER_KEY_MPLS_LABEL]);
386 set_mpls_lse_label(&flower->key.mpls_lse, htonl(label));
387 set_mpls_lse_label(&flower->mask.mpls_lse, OVS_BE32_MAX);
388 }
389 }
390
391 static void
392 nl_parse_flower_vlan(struct nlattr **attrs, struct tc_flower *flower)
393 {
394 ovs_be16 encap_ethtype;
395
396 if (!eth_type_vlan(flower->key.eth_type)) {
397 return;
398 }
399
400 flower->key.encap_eth_type[0] =
401 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_ETH_TYPE]);
402
403 if (attrs[TCA_FLOWER_KEY_VLAN_ID]) {
404 flower->key.vlan_id[0] =
405 nl_attr_get_u16(attrs[TCA_FLOWER_KEY_VLAN_ID]);
406 }
407 if (attrs[TCA_FLOWER_KEY_VLAN_PRIO]) {
408 flower->key.vlan_prio[0] =
409 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_VLAN_PRIO]);
410 }
411
412 if (!attrs[TCA_FLOWER_KEY_VLAN_ETH_TYPE]) {
413 return;
414 }
415
416 encap_ethtype = nl_attr_get_be16(attrs[TCA_FLOWER_KEY_VLAN_ETH_TYPE]);
417 if (!eth_type_vlan(encap_ethtype)) {
418 return;
419 }
420
421 flower->key.encap_eth_type[1] = flower->key.encap_eth_type[0];
422 flower->key.encap_eth_type[0] = encap_ethtype;
423
424 if (attrs[TCA_FLOWER_KEY_CVLAN_ID]) {
425 flower->key.vlan_id[1] =
426 nl_attr_get_u16(attrs[TCA_FLOWER_KEY_CVLAN_ID]);
427 }
428 if (attrs[TCA_FLOWER_KEY_CVLAN_PRIO]) {
429 flower->key.vlan_prio[1] =
430 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_CVLAN_PRIO]);
431 }
432 }
433
434 static void
435 nl_parse_flower_tunnel(struct nlattr **attrs, struct tc_flower *flower)
436 {
437 if (attrs[TCA_FLOWER_KEY_ENC_KEY_ID]) {
438 ovs_be32 id = nl_attr_get_be32(attrs[TCA_FLOWER_KEY_ENC_KEY_ID]);
439
440 flower->key.tunnel.id = be32_to_be64(id);
441 }
442 if (attrs[TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK]) {
443 flower->key.tunnel.ipv4.ipv4_src =
444 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_ENC_IPV4_SRC]);
445 }
446 if (attrs[TCA_FLOWER_KEY_ENC_IPV4_DST_MASK]) {
447 flower->key.tunnel.ipv4.ipv4_dst =
448 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_ENC_IPV4_DST]);
449 }
450 if (attrs[TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK]) {
451 flower->key.tunnel.ipv6.ipv6_src =
452 nl_attr_get_in6_addr(attrs[TCA_FLOWER_KEY_ENC_IPV6_SRC]);
453 }
454 if (attrs[TCA_FLOWER_KEY_ENC_IPV6_DST_MASK]) {
455 flower->key.tunnel.ipv6.ipv6_dst =
456 nl_attr_get_in6_addr(attrs[TCA_FLOWER_KEY_ENC_IPV6_DST]);
457 }
458 if (attrs[TCA_FLOWER_KEY_ENC_UDP_DST_PORT]) {
459 flower->key.tunnel.tp_dst =
460 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_ENC_UDP_DST_PORT]);
461 }
462 if (attrs[TCA_FLOWER_KEY_ENC_IP_TOS_MASK]) {
463 flower->key.tunnel.tos =
464 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_ENC_IP_TOS]);
465 flower->mask.tunnel.tos =
466 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_ENC_IP_TOS_MASK]);
467 }
468 if (attrs[TCA_FLOWER_KEY_ENC_IP_TTL_MASK]) {
469 flower->key.tunnel.ttl =
470 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_ENC_IP_TTL]);
471 flower->mask.tunnel.ttl =
472 nl_attr_get_u8(attrs[TCA_FLOWER_KEY_ENC_IP_TTL_MASK]);
473 }
474 }
475
476 static void
477 nl_parse_flower_ip(struct nlattr **attrs, struct tc_flower *flower) {
478 uint8_t ip_proto = 0;
479 struct tc_flower_key *key = &flower->key;
480 struct tc_flower_key *mask = &flower->mask;
481
482 if (attrs[TCA_FLOWER_KEY_IP_PROTO]) {
483 ip_proto = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_IP_PROTO]);
484 key->ip_proto = ip_proto;
485 mask->ip_proto = UINT8_MAX;
486 }
487
488 if (attrs[TCA_FLOWER_KEY_FLAGS_MASK]) {
489 key->flags = ntohl(nl_attr_get_be32(attrs[TCA_FLOWER_KEY_FLAGS]));
490 mask->flags =
491 ntohl(nl_attr_get_be32(attrs[TCA_FLOWER_KEY_FLAGS_MASK]));
492 }
493
494 if (attrs[TCA_FLOWER_KEY_IPV4_SRC_MASK]) {
495 key->ipv4.ipv4_src =
496 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_IPV4_SRC]);
497 mask->ipv4.ipv4_src =
498 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_IPV4_SRC_MASK]);
499 }
500 if (attrs[TCA_FLOWER_KEY_IPV4_DST_MASK]) {
501 key->ipv4.ipv4_dst =
502 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_IPV4_DST]);
503 mask->ipv4.ipv4_dst =
504 nl_attr_get_be32(attrs[TCA_FLOWER_KEY_IPV4_DST_MASK]);
505 }
506 if (attrs[TCA_FLOWER_KEY_IPV6_SRC_MASK]) {
507 struct nlattr *attr = attrs[TCA_FLOWER_KEY_IPV6_SRC];
508 struct nlattr *attr_mask = attrs[TCA_FLOWER_KEY_IPV6_SRC_MASK];
509
510 key->ipv6.ipv6_src = nl_attr_get_in6_addr(attr);
511 mask->ipv6.ipv6_src = nl_attr_get_in6_addr(attr_mask);
512 }
513 if (attrs[TCA_FLOWER_KEY_IPV6_DST_MASK]) {
514 struct nlattr *attr = attrs[TCA_FLOWER_KEY_IPV6_DST];
515 struct nlattr *attr_mask = attrs[TCA_FLOWER_KEY_IPV6_DST_MASK];
516
517 key->ipv6.ipv6_dst = nl_attr_get_in6_addr(attr);
518 mask->ipv6.ipv6_dst = nl_attr_get_in6_addr(attr_mask);
519 }
520
521 if (ip_proto == IPPROTO_TCP) {
522 if (attrs[TCA_FLOWER_KEY_TCP_SRC_MASK]) {
523 key->tcp_src =
524 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_SRC]);
525 mask->tcp_src =
526 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_SRC_MASK]);
527 }
528 if (attrs[TCA_FLOWER_KEY_TCP_DST_MASK]) {
529 key->tcp_dst =
530 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_DST]);
531 mask->tcp_dst =
532 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_DST_MASK]);
533 }
534 if (attrs[TCA_FLOWER_KEY_TCP_FLAGS_MASK]) {
535 key->tcp_flags =
536 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_FLAGS]);
537 mask->tcp_flags =
538 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_TCP_FLAGS_MASK]);
539 }
540 } else if (ip_proto == IPPROTO_UDP) {
541 if (attrs[TCA_FLOWER_KEY_UDP_SRC_MASK]) {
542 key->udp_src = nl_attr_get_be16(attrs[TCA_FLOWER_KEY_UDP_SRC]);
543 mask->udp_src =
544 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_UDP_SRC_MASK]);
545 }
546 if (attrs[TCA_FLOWER_KEY_UDP_DST_MASK]) {
547 key->udp_dst = nl_attr_get_be16(attrs[TCA_FLOWER_KEY_UDP_DST]);
548 mask->udp_dst =
549 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_UDP_DST_MASK]);
550 }
551 } else if (ip_proto == IPPROTO_SCTP) {
552 if (attrs[TCA_FLOWER_KEY_SCTP_SRC_MASK]) {
553 key->sctp_src = nl_attr_get_be16(attrs[TCA_FLOWER_KEY_SCTP_SRC]);
554 mask->sctp_src =
555 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_SCTP_SRC_MASK]);
556 }
557 if (attrs[TCA_FLOWER_KEY_SCTP_DST_MASK]) {
558 key->sctp_dst = nl_attr_get_be16(attrs[TCA_FLOWER_KEY_SCTP_DST]);
559 mask->sctp_dst =
560 nl_attr_get_be16(attrs[TCA_FLOWER_KEY_SCTP_DST_MASK]);
561 }
562 }
563
564 if (attrs[TCA_FLOWER_KEY_IP_TTL_MASK]) {
565 key->ip_ttl = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_IP_TTL]);
566 mask->ip_ttl = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_IP_TTL_MASK]);
567 }
568
569 if (attrs[TCA_FLOWER_KEY_IP_TOS_MASK]) {
570 key->ip_tos = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_IP_TOS]);
571 mask->ip_tos = nl_attr_get_u8(attrs[TCA_FLOWER_KEY_IP_TOS_MASK]);
572 }
573 }
574
575 static enum tc_offloaded_state
576 nl_get_flower_offloaded_state(struct nlattr **attrs)
577 {
578 uint32_t flower_flags = 0;
579
580 if (attrs[TCA_FLOWER_FLAGS]) {
581 flower_flags = nl_attr_get_u32(attrs[TCA_FLOWER_FLAGS]);
582 if (flower_flags & TCA_CLS_FLAGS_NOT_IN_HW) {
583 return TC_OFFLOADED_STATE_NOT_IN_HW;
584 } else if (flower_flags & TCA_CLS_FLAGS_IN_HW) {
585 return TC_OFFLOADED_STATE_IN_HW;
586 }
587 }
588 return TC_OFFLOADED_STATE_UNDEFINED;
589 }
590
591 static void
592 nl_parse_flower_flags(struct nlattr **attrs, struct tc_flower *flower)
593 {
594 flower->offloaded_state = nl_get_flower_offloaded_state(attrs);
595 }
596
597 static const struct nl_policy pedit_policy[] = {
598 [TCA_PEDIT_PARMS_EX] = { .type = NL_A_UNSPEC,
599 .min_len = sizeof(struct tc_pedit),
600 .optional = false, },
601 [TCA_PEDIT_KEYS_EX] = { .type = NL_A_NESTED,
602 .optional = false, },
603 };
604
605 static int
606 nl_parse_act_pedit(struct nlattr *options, struct tc_flower *flower)
607 {
608 struct tc_action *action;
609 struct nlattr *pe_attrs[ARRAY_SIZE(pedit_policy)];
610 const struct tc_pedit *pe;
611 const struct tc_pedit_key *keys;
612 const struct nlattr *nla, *keys_ex, *ex_type;
613 const void *keys_attr;
614 char *rewrite_key = (void *) &flower->rewrite.key;
615 char *rewrite_mask = (void *) &flower->rewrite.mask;
616 size_t keys_ex_size, left;
617 int type, i = 0, err;
618
619 if (!nl_parse_nested(options, pedit_policy, pe_attrs,
620 ARRAY_SIZE(pedit_policy))) {
621 VLOG_ERR_RL(&error_rl, "failed to parse pedit action options");
622 return EPROTO;
623 }
624
625 pe = nl_attr_get_unspec(pe_attrs[TCA_PEDIT_PARMS_EX], sizeof *pe);
626 keys = pe->keys;
627 keys_attr = pe_attrs[TCA_PEDIT_KEYS_EX];
628 keys_ex = nl_attr_get(keys_attr);
629 keys_ex_size = nl_attr_get_size(keys_attr);
630
631 NL_ATTR_FOR_EACH (nla, left, keys_ex, keys_ex_size) {
632 if (i >= pe->nkeys) {
633 break;
634 }
635
636 if (nl_attr_type(nla) != TCA_PEDIT_KEY_EX) {
637 VLOG_ERR_RL(&error_rl, "unable to parse legacy pedit type: %d",
638 nl_attr_type(nla));
639 return EOPNOTSUPP;
640 }
641
642 ex_type = nl_attr_find_nested(nla, TCA_PEDIT_KEY_EX_HTYPE);
643 type = nl_attr_get_u16(ex_type);
644
645 err = csum_update_flag(flower, type);
646 if (err) {
647 return err;
648 }
649
650 for (int j = 0; j < ARRAY_SIZE(flower_pedit_map); j++) {
651 struct flower_key_to_pedit *m = &flower_pedit_map[j];
652 int flower_off = m->flower_offset;
653 int sz = m->size;
654 int mf = m->offset;
655
656 if (m->htype != type) {
657 continue;
658 }
659
660 /* check overlap between current pedit key, which is always
661 * 4 bytes (range [off, off + 3]), and a map entry in
662 * flower_pedit_map (range [mf, mf + sz - 1]) */
663 if ((keys->off >= mf && keys->off < mf + sz)
664 || (keys->off + 3 >= mf && keys->off + 3 < mf + sz)) {
665 int diff = flower_off + (keys->off - mf);
666 uint32_t *dst = (void *) (rewrite_key + diff);
667 uint32_t *dst_m = (void *) (rewrite_mask + diff);
668 uint32_t mask = ~(keys->mask);
669 uint32_t zero_bits;
670
671 if (keys->off < mf) {
672 zero_bits = 8 * (mf - keys->off);
673 mask &= UINT32_MAX << zero_bits;
674 } else if (keys->off + 4 > mf + m->size) {
675 zero_bits = 8 * (keys->off + 4 - mf - m->size);
676 mask &= UINT32_MAX >> zero_bits;
677 }
678
679 *dst_m |= mask;
680 *dst |= keys->val & mask;
681 }
682 }
683
684 keys++;
685 i++;
686 }
687
688 action = &flower->actions[flower->action_count++];
689 action->type = TC_ACT_PEDIT;
690
691 return 0;
692 }
693
694 static const struct nl_policy tunnel_key_policy[] = {
695 [TCA_TUNNEL_KEY_PARMS] = { .type = NL_A_UNSPEC,
696 .min_len = sizeof(struct tc_tunnel_key),
697 .optional = false, },
698 [TCA_TUNNEL_KEY_ENC_IPV4_SRC] = { .type = NL_A_U32, .optional = true, },
699 [TCA_TUNNEL_KEY_ENC_IPV4_DST] = { .type = NL_A_U32, .optional = true, },
700 [TCA_TUNNEL_KEY_ENC_IPV6_SRC] = { .type = NL_A_UNSPEC,
701 .min_len = sizeof(struct in6_addr),
702 .optional = true, },
703 [TCA_TUNNEL_KEY_ENC_IPV6_DST] = { .type = NL_A_UNSPEC,
704 .min_len = sizeof(struct in6_addr),
705 .optional = true, },
706 [TCA_TUNNEL_KEY_ENC_KEY_ID] = { .type = NL_A_U32, .optional = true, },
707 [TCA_TUNNEL_KEY_ENC_DST_PORT] = { .type = NL_A_U16, .optional = true, },
708 [TCA_TUNNEL_KEY_ENC_TOS] = { .type = NL_A_U8, .optional = true, },
709 [TCA_TUNNEL_KEY_ENC_TTL] = { .type = NL_A_U8, .optional = true, },
710 };
711
712 static int
713 nl_parse_act_tunnel_key(struct nlattr *options, struct tc_flower *flower)
714 {
715 struct nlattr *tun_attrs[ARRAY_SIZE(tunnel_key_policy)];
716 const struct nlattr *tun_parms;
717 const struct tc_tunnel_key *tun;
718 struct tc_action *action;
719
720 if (!nl_parse_nested(options, tunnel_key_policy, tun_attrs,
721 ARRAY_SIZE(tunnel_key_policy))) {
722 VLOG_ERR_RL(&error_rl, "failed to parse tunnel_key action options");
723 return EPROTO;
724 }
725
726 tun_parms = tun_attrs[TCA_TUNNEL_KEY_PARMS];
727 tun = nl_attr_get_unspec(tun_parms, sizeof *tun);
728 if (tun->t_action == TCA_TUNNEL_KEY_ACT_SET) {
729 struct nlattr *id = tun_attrs[TCA_TUNNEL_KEY_ENC_KEY_ID];
730 struct nlattr *dst_port = tun_attrs[TCA_TUNNEL_KEY_ENC_DST_PORT];
731 struct nlattr *ipv4_src = tun_attrs[TCA_TUNNEL_KEY_ENC_IPV4_SRC];
732 struct nlattr *ipv4_dst = tun_attrs[TCA_TUNNEL_KEY_ENC_IPV4_DST];
733 struct nlattr *ipv6_src = tun_attrs[TCA_TUNNEL_KEY_ENC_IPV6_SRC];
734 struct nlattr *ipv6_dst = tun_attrs[TCA_TUNNEL_KEY_ENC_IPV6_DST];
735 struct nlattr *tos = tun_attrs[TCA_TUNNEL_KEY_ENC_TOS];
736 struct nlattr *ttl = tun_attrs[TCA_TUNNEL_KEY_ENC_TTL];
737
738 action = &flower->actions[flower->action_count++];
739 action->type = TC_ACT_ENCAP;
740 action->encap.ipv4.ipv4_src = ipv4_src ? nl_attr_get_be32(ipv4_src) : 0;
741 action->encap.ipv4.ipv4_dst = ipv4_dst ? nl_attr_get_be32(ipv4_dst) : 0;
742 if (ipv6_src) {
743 action->encap.ipv6.ipv6_src = nl_attr_get_in6_addr(ipv6_src);
744 }
745 if (ipv6_dst) {
746 action->encap.ipv6.ipv6_dst = nl_attr_get_in6_addr(ipv6_dst);
747 }
748 action->encap.id = id ? be32_to_be64(nl_attr_get_be32(id)) : 0;
749 action->encap.tp_dst = dst_port ? nl_attr_get_be16(dst_port) : 0;
750 action->encap.tos = tos ? nl_attr_get_u8(tos) : 0;
751 action->encap.ttl = ttl ? nl_attr_get_u8(ttl) : 0;
752 } else if (tun->t_action == TCA_TUNNEL_KEY_ACT_RELEASE) {
753 flower->tunnel = true;
754 } else {
755 VLOG_ERR_RL(&error_rl, "unknown tunnel actions: %d, %d",
756 tun->action, tun->t_action);
757 return EINVAL;
758 }
759 return 0;
760 }
761
762 static const struct nl_policy gact_policy[] = {
763 [TCA_GACT_PARMS] = { .type = NL_A_UNSPEC,
764 .min_len = sizeof(struct tc_gact),
765 .optional = false, },
766 [TCA_GACT_TM] = { .type = NL_A_UNSPEC,
767 .min_len = sizeof(struct tcf_t),
768 .optional = false, },
769 };
770
771 static int
772 get_user_hz(void)
773 {
774 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
775 static int user_hz = 100;
776
777 if (ovsthread_once_start(&once)) {
778 user_hz = sysconf(_SC_CLK_TCK);
779 ovsthread_once_done(&once);
780 }
781
782 return user_hz;
783 }
784
785 static void
786 nl_parse_tcf(const struct tcf_t *tm, struct tc_flower *flower)
787 {
788 flower->lastused = time_msec() - (tm->lastuse * 1000 / get_user_hz());
789 }
790
791 static int
792 nl_parse_act_drop(struct nlattr *options, struct tc_flower *flower)
793 {
794 struct nlattr *gact_attrs[ARRAY_SIZE(gact_policy)];
795 const struct tc_gact *p;
796 struct nlattr *gact_parms;
797 const struct tcf_t *tm;
798
799 if (!nl_parse_nested(options, gact_policy, gact_attrs,
800 ARRAY_SIZE(gact_policy))) {
801 VLOG_ERR_RL(&error_rl, "failed to parse gact action options");
802 return EPROTO;
803 }
804
805 gact_parms = gact_attrs[TCA_GACT_PARMS];
806 p = nl_attr_get_unspec(gact_parms, sizeof *p);
807
808 if (p->action != TC_ACT_SHOT) {
809 VLOG_ERR_RL(&error_rl, "unknown gact action: %d", p->action);
810 return EINVAL;
811 }
812
813 tm = nl_attr_get_unspec(gact_attrs[TCA_GACT_TM], sizeof *tm);
814 nl_parse_tcf(tm, flower);
815
816 return 0;
817 }
818
819 static const struct nl_policy mirred_policy[] = {
820 [TCA_MIRRED_PARMS] = { .type = NL_A_UNSPEC,
821 .min_len = sizeof(struct tc_mirred),
822 .optional = false, },
823 [TCA_MIRRED_TM] = { .type = NL_A_UNSPEC,
824 .min_len = sizeof(struct tcf_t),
825 .optional = false, },
826 };
827
828 static int
829 nl_parse_act_mirred(struct nlattr *options, struct tc_flower *flower)
830 {
831
832 struct nlattr *mirred_attrs[ARRAY_SIZE(mirred_policy)];
833 const struct tc_mirred *m;
834 const struct nlattr *mirred_parms;
835 const struct tcf_t *tm;
836 struct nlattr *mirred_tm;
837 struct tc_action *action;
838
839 if (!nl_parse_nested(options, mirred_policy, mirred_attrs,
840 ARRAY_SIZE(mirred_policy))) {
841 VLOG_ERR_RL(&error_rl, "failed to parse mirred action options");
842 return EPROTO;
843 }
844
845 mirred_parms = mirred_attrs[TCA_MIRRED_PARMS];
846 m = nl_attr_get_unspec(mirred_parms, sizeof *m);
847
848 if (m->eaction != TCA_EGRESS_REDIR && m->eaction != TCA_EGRESS_MIRROR) {
849 VLOG_ERR_RL(&error_rl, "unknown mirred action: %d, %d, %d",
850 m->action, m->eaction, m->ifindex);
851 return EINVAL;
852 }
853
854 action = &flower->actions[flower->action_count++];
855 action->ifindex_out = m->ifindex;
856 action->type = TC_ACT_OUTPUT;
857
858 mirred_tm = mirred_attrs[TCA_MIRRED_TM];
859 tm = nl_attr_get_unspec(mirred_tm, sizeof *tm);
860 nl_parse_tcf(tm, flower);
861
862 return 0;
863 }
864
865 static const struct nl_policy vlan_policy[] = {
866 [TCA_VLAN_PARMS] = { .type = NL_A_UNSPEC,
867 .min_len = sizeof(struct tc_vlan),
868 .optional = false, },
869 [TCA_VLAN_PUSH_VLAN_ID] = { .type = NL_A_U16, .optional = true, },
870 [TCA_VLAN_PUSH_VLAN_PROTOCOL] = { .type = NL_A_U16, .optional = true, },
871 [TCA_VLAN_PUSH_VLAN_PRIORITY] = { .type = NL_A_U8, .optional = true, },
872 };
873
874 static int
875 nl_parse_act_vlan(struct nlattr *options, struct tc_flower *flower)
876 {
877 struct nlattr *vlan_attrs[ARRAY_SIZE(vlan_policy)];
878 const struct tc_vlan *v;
879 const struct nlattr *vlan_parms;
880 struct tc_action *action;
881
882 if (!nl_parse_nested(options, vlan_policy, vlan_attrs,
883 ARRAY_SIZE(vlan_policy))) {
884 VLOG_ERR_RL(&error_rl, "failed to parse vlan action options");
885 return EPROTO;
886 }
887
888 action = &flower->actions[flower->action_count++];
889 vlan_parms = vlan_attrs[TCA_VLAN_PARMS];
890 v = nl_attr_get_unspec(vlan_parms, sizeof *v);
891 if (v->v_action == TCA_VLAN_ACT_PUSH) {
892 struct nlattr *vlan_tpid = vlan_attrs[TCA_VLAN_PUSH_VLAN_PROTOCOL];
893 struct nlattr *vlan_id = vlan_attrs[TCA_VLAN_PUSH_VLAN_ID];
894 struct nlattr *vlan_prio = vlan_attrs[TCA_VLAN_PUSH_VLAN_PRIORITY];
895
896 action->vlan.vlan_push_tpid = nl_attr_get_be16(vlan_tpid);
897 action->vlan.vlan_push_id = nl_attr_get_u16(vlan_id);
898 action->vlan.vlan_push_prio = vlan_prio ? nl_attr_get_u8(vlan_prio) : 0;
899 action->type = TC_ACT_VLAN_PUSH;
900 } else if (v->v_action == TCA_VLAN_ACT_POP) {
901 action->type = TC_ACT_VLAN_POP;
902 } else {
903 VLOG_ERR_RL(&error_rl, "unknown vlan action: %d, %d",
904 v->action, v->v_action);
905 return EINVAL;
906 }
907 return 0;
908 }
909
910 static const struct nl_policy csum_policy[] = {
911 [TCA_CSUM_PARMS] = { .type = NL_A_UNSPEC,
912 .min_len = sizeof(struct tc_csum),
913 .optional = false, },
914 };
915
916 static int
917 nl_parse_act_csum(struct nlattr *options, struct tc_flower *flower)
918 {
919 struct nlattr *csum_attrs[ARRAY_SIZE(csum_policy)];
920 const struct tc_csum *c;
921 const struct nlattr *csum_parms;
922
923 if (!nl_parse_nested(options, csum_policy, csum_attrs,
924 ARRAY_SIZE(csum_policy))) {
925 VLOG_ERR_RL(&error_rl, "failed to parse csum action options");
926 return EPROTO;
927 }
928
929 csum_parms = csum_attrs[TCA_CSUM_PARMS];
930 c = nl_attr_get_unspec(csum_parms, sizeof *c);
931
932 /* sanity checks */
933 if (c->update_flags != flower->csum_update_flags) {
934 VLOG_WARN_RL(&error_rl,
935 "expected different act csum flags: 0x%x != 0x%x",
936 flower->csum_update_flags, c->update_flags);
937 return EINVAL;
938 }
939 flower->csum_update_flags = 0; /* so we know csum was handled */
940
941 if (flower->needs_full_ip_proto_mask
942 && flower->mask.ip_proto != UINT8_MAX) {
943 VLOG_WARN_RL(&error_rl, "expected full matching on flower ip_proto");
944 return EINVAL;
945 }
946
947 return 0;
948 }
949
950 static const struct nl_policy act_policy[] = {
951 [TCA_ACT_KIND] = { .type = NL_A_STRING, .optional = false, },
952 [TCA_ACT_COOKIE] = { .type = NL_A_UNSPEC, .optional = true, },
953 [TCA_ACT_OPTIONS] = { .type = NL_A_NESTED, .optional = false, },
954 [TCA_ACT_STATS] = { .type = NL_A_NESTED, .optional = false, },
955 };
956
957 static const struct nl_policy stats_policy[] = {
958 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC,
959 .min_len = sizeof(struct gnet_stats_basic),
960 .optional = false, },
961 };
962
963 static int
964 nl_parse_single_action(struct nlattr *action, struct tc_flower *flower)
965 {
966 struct nlattr *act_options;
967 struct nlattr *act_stats;
968 struct nlattr *act_cookie;
969 const char *act_kind;
970 struct nlattr *action_attrs[ARRAY_SIZE(act_policy)];
971 struct nlattr *stats_attrs[ARRAY_SIZE(stats_policy)];
972 struct ovs_flow_stats *stats = &flower->stats;
973 const struct gnet_stats_basic *bs;
974 int err = 0;
975
976 if (!nl_parse_nested(action, act_policy, action_attrs,
977 ARRAY_SIZE(act_policy))) {
978 VLOG_ERR_RL(&error_rl, "failed to parse single action options");
979 return EPROTO;
980 }
981
982 act_kind = nl_attr_get_string(action_attrs[TCA_ACT_KIND]);
983 act_options = action_attrs[TCA_ACT_OPTIONS];
984 act_cookie = action_attrs[TCA_ACT_COOKIE];
985
986 if (!strcmp(act_kind, "gact")) {
987 err = nl_parse_act_drop(act_options, flower);
988 } else if (!strcmp(act_kind, "mirred")) {
989 err = nl_parse_act_mirred(act_options, flower);
990 } else if (!strcmp(act_kind, "vlan")) {
991 err = nl_parse_act_vlan(act_options, flower);
992 } else if (!strcmp(act_kind, "tunnel_key")) {
993 err = nl_parse_act_tunnel_key(act_options, flower);
994 } else if (!strcmp(act_kind, "pedit")) {
995 err = nl_parse_act_pedit(act_options, flower);
996 } else if (!strcmp(act_kind, "csum")) {
997 nl_parse_act_csum(act_options, flower);
998 } else {
999 VLOG_ERR_RL(&error_rl, "unknown tc action kind: %s", act_kind);
1000 err = EINVAL;
1001 }
1002
1003 if (err) {
1004 return err;
1005 }
1006
1007 if (act_cookie) {
1008 flower->act_cookie.data = nl_attr_get(act_cookie);
1009 flower->act_cookie.len = nl_attr_get_size(act_cookie);
1010 }
1011
1012 act_stats = action_attrs[TCA_ACT_STATS];
1013
1014 if (!nl_parse_nested(act_stats, stats_policy, stats_attrs,
1015 ARRAY_SIZE(stats_policy))) {
1016 VLOG_ERR_RL(&error_rl, "failed to parse action stats policy");
1017 return EPROTO;
1018 }
1019
1020 bs = nl_attr_get_unspec(stats_attrs[TCA_STATS_BASIC], sizeof *bs);
1021 put_32aligned_u64(&stats->n_packets, bs->packets);
1022 put_32aligned_u64(&stats->n_bytes, bs->bytes);
1023
1024 return 0;
1025 }
1026
1027 #define TCA_ACT_MIN_PRIO 1
1028
1029 static int
1030 nl_parse_flower_actions(struct nlattr **attrs, struct tc_flower *flower)
1031 {
1032 const struct nlattr *actions = attrs[TCA_FLOWER_ACT];
1033 static struct nl_policy actions_orders_policy[TCA_ACT_MAX_PRIO + 1] = {};
1034 struct nlattr *actions_orders[ARRAY_SIZE(actions_orders_policy)];
1035 const int max_size = ARRAY_SIZE(actions_orders_policy);
1036
1037 for (int i = TCA_ACT_MIN_PRIO; i < max_size; i++) {
1038 actions_orders_policy[i].type = NL_A_NESTED;
1039 actions_orders_policy[i].optional = true;
1040 }
1041
1042 if (!nl_parse_nested(actions, actions_orders_policy, actions_orders,
1043 ARRAY_SIZE(actions_orders_policy))) {
1044 VLOG_ERR_RL(&error_rl, "failed to parse flower order of actions");
1045 return EPROTO;
1046 }
1047
1048 for (int i = TCA_ACT_MIN_PRIO; i < max_size; i++) {
1049 if (actions_orders[i]) {
1050 int err;
1051
1052 if (flower->action_count >= TCA_ACT_MAX_PRIO) {
1053 VLOG_DBG_RL(&error_rl, "Can only support %d actions", flower->action_count);
1054 return EOPNOTSUPP;
1055 }
1056 err = nl_parse_single_action(actions_orders[i], flower);
1057
1058 if (err) {
1059 return err;
1060 }
1061 }
1062 }
1063
1064 if (flower->csum_update_flags) {
1065 VLOG_WARN_RL(&error_rl,
1066 "expected act csum with flags: 0x%x",
1067 flower->csum_update_flags);
1068 return EINVAL;
1069 }
1070
1071 return 0;
1072 }
1073
1074 static int
1075 nl_parse_flower_options(struct nlattr *nl_options, struct tc_flower *flower)
1076 {
1077 struct nlattr *attrs[ARRAY_SIZE(tca_flower_policy)];
1078
1079 if (!nl_parse_nested(nl_options, tca_flower_policy,
1080 attrs, ARRAY_SIZE(tca_flower_policy))) {
1081 VLOG_ERR_RL(&error_rl, "failed to parse flower classifier options");
1082 return EPROTO;
1083 }
1084
1085 nl_parse_flower_eth(attrs, flower);
1086 nl_parse_flower_mpls(attrs, flower);
1087 nl_parse_flower_vlan(attrs, flower);
1088 nl_parse_flower_ip(attrs, flower);
1089 nl_parse_flower_tunnel(attrs, flower);
1090 nl_parse_flower_flags(attrs, flower);
1091 return nl_parse_flower_actions(attrs, flower);
1092 }
1093
1094 int
1095 parse_netlink_to_tc_flower(struct ofpbuf *reply, struct tc_flower *flower)
1096 {
1097 struct tcmsg *tc;
1098 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
1099 const char *kind;
1100
1101 if (NLMSG_HDRLEN + sizeof *tc > reply->size) {
1102 return EPROTO;
1103 }
1104
1105 memset(flower, 0, sizeof *flower);
1106
1107 tc = ofpbuf_at_assert(reply, NLMSG_HDRLEN, sizeof *tc);
1108 flower->handle = tc->tcm_handle;
1109 flower->key.eth_type = (OVS_FORCE ovs_be16) tc_get_minor(tc->tcm_info);
1110 flower->mask.eth_type = OVS_BE16_MAX;
1111 flower->prio = tc_get_major(tc->tcm_info);
1112
1113 if (!flower->handle) {
1114 return EAGAIN;
1115 }
1116
1117 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof *tc,
1118 tca_policy, ta, ARRAY_SIZE(ta))) {
1119 VLOG_ERR_RL(&error_rl, "failed to parse tca policy");
1120 return EPROTO;
1121 }
1122
1123 kind = nl_attr_get_string(ta[TCA_KIND]);
1124 if (strcmp(kind, "flower")) {
1125 VLOG_DBG_ONCE("Unsupported filter: %s", kind);
1126 return EPROTO;
1127 }
1128
1129 return nl_parse_flower_options(ta[TCA_OPTIONS], flower);
1130 }
1131
1132 int
1133 tc_dump_flower_start(int ifindex, struct nl_dump *dump, uint32_t block_id)
1134 {
1135 struct ofpbuf request;
1136 struct tcmsg *tcmsg;
1137 int index;
1138
1139 index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex;
1140 tcmsg = tc_make_request(index, RTM_GETTFILTER, NLM_F_DUMP, &request);
1141 tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT;
1142 tcmsg->tcm_info = TC_H_UNSPEC;
1143 tcmsg->tcm_handle = 0;
1144
1145 nl_dump_start(dump, NETLINK_ROUTE, &request);
1146 ofpbuf_uninit(&request);
1147
1148 return 0;
1149 }
1150
1151 int
1152 tc_flush(int ifindex, uint32_t block_id)
1153 {
1154 struct ofpbuf request;
1155 struct tcmsg *tcmsg;
1156 int index;
1157
1158 index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex;
1159 tcmsg = tc_make_request(index, RTM_DELTFILTER, NLM_F_ACK, &request);
1160 tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT;
1161 tcmsg->tcm_info = TC_H_UNSPEC;
1162
1163 return tc_transact(&request, NULL);
1164 }
1165
1166 int
1167 tc_del_filter(int ifindex, int prio, int handle, uint32_t block_id)
1168 {
1169 struct ofpbuf request;
1170 struct tcmsg *tcmsg;
1171 struct ofpbuf *reply;
1172 int error;
1173 int index;
1174
1175 index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex;
1176 tcmsg = tc_make_request(index, RTM_DELTFILTER, NLM_F_ECHO, &request);
1177 tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT;
1178 tcmsg->tcm_info = tc_make_handle(prio, 0);
1179 tcmsg->tcm_handle = handle;
1180
1181 error = tc_transact(&request, &reply);
1182 if (!error) {
1183 ofpbuf_delete(reply);
1184 }
1185 return error;
1186 }
1187
1188 int
1189 tc_get_flower(int ifindex, int prio, int handle, struct tc_flower *flower,
1190 uint32_t block_id)
1191 {
1192 struct ofpbuf request;
1193 struct tcmsg *tcmsg;
1194 struct ofpbuf *reply;
1195 int error;
1196 int index;
1197
1198 index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex;
1199 tcmsg = tc_make_request(index, RTM_GETTFILTER, NLM_F_ECHO, &request);
1200 tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT;
1201 tcmsg->tcm_info = tc_make_handle(prio, 0);
1202 tcmsg->tcm_handle = handle;
1203
1204 error = tc_transact(&request, &reply);
1205 if (error) {
1206 return error;
1207 }
1208
1209 error = parse_netlink_to_tc_flower(reply, flower);
1210 ofpbuf_delete(reply);
1211 return error;
1212 }
1213
1214 static int
1215 tc_get_tc_cls_policy(enum tc_offload_policy policy)
1216 {
1217 if (policy == TC_POLICY_SKIP_HW) {
1218 return TCA_CLS_FLAGS_SKIP_HW;
1219 } else if (policy == TC_POLICY_SKIP_SW) {
1220 return TCA_CLS_FLAGS_SKIP_SW;
1221 }
1222
1223 return 0;
1224 }
1225
1226 static void
1227 nl_msg_put_act_csum(struct ofpbuf *request, uint32_t flags)
1228 {
1229 size_t offset;
1230
1231 nl_msg_put_string(request, TCA_ACT_KIND, "csum");
1232 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1233 {
1234 struct tc_csum parm = { .action = TC_ACT_PIPE,
1235 .update_flags = flags };
1236
1237 nl_msg_put_unspec(request, TCA_CSUM_PARMS, &parm, sizeof parm);
1238 }
1239 nl_msg_end_nested(request, offset);
1240 }
1241
1242 static void
1243 nl_msg_put_act_pedit(struct ofpbuf *request, struct tc_pedit *parm,
1244 struct tc_pedit_key_ex *ex)
1245 {
1246 size_t ksize = sizeof *parm + parm->nkeys * sizeof(struct tc_pedit_key);
1247 size_t offset, offset_keys_ex, offset_key;
1248 int i;
1249
1250 nl_msg_put_string(request, TCA_ACT_KIND, "pedit");
1251 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1252 {
1253 parm->action = TC_ACT_PIPE;
1254
1255 nl_msg_put_unspec(request, TCA_PEDIT_PARMS_EX, parm, ksize);
1256 offset_keys_ex = nl_msg_start_nested(request, TCA_PEDIT_KEYS_EX);
1257 for (i = 0; i < parm->nkeys; i++, ex++) {
1258 offset_key = nl_msg_start_nested(request, TCA_PEDIT_KEY_EX);
1259 nl_msg_put_u16(request, TCA_PEDIT_KEY_EX_HTYPE, ex->htype);
1260 nl_msg_put_u16(request, TCA_PEDIT_KEY_EX_CMD, ex->cmd);
1261 nl_msg_end_nested(request, offset_key);
1262 }
1263 nl_msg_end_nested(request, offset_keys_ex);
1264 }
1265 nl_msg_end_nested(request, offset);
1266 }
1267
1268 static void
1269 nl_msg_put_act_push_vlan(struct ofpbuf *request, ovs_be16 tpid,
1270 uint16_t vid, uint8_t prio)
1271 {
1272 size_t offset;
1273
1274 nl_msg_put_string(request, TCA_ACT_KIND, "vlan");
1275 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1276 {
1277 struct tc_vlan parm = { .action = TC_ACT_PIPE,
1278 .v_action = TCA_VLAN_ACT_PUSH };
1279
1280 nl_msg_put_unspec(request, TCA_VLAN_PARMS, &parm, sizeof parm);
1281 nl_msg_put_be16(request, TCA_VLAN_PUSH_VLAN_PROTOCOL, tpid);
1282 nl_msg_put_u16(request, TCA_VLAN_PUSH_VLAN_ID, vid);
1283 nl_msg_put_u8(request, TCA_VLAN_PUSH_VLAN_PRIORITY, prio);
1284 }
1285 nl_msg_end_nested(request, offset);
1286 }
1287
1288 static void
1289 nl_msg_put_act_pop_vlan(struct ofpbuf *request)
1290 {
1291 size_t offset;
1292
1293 nl_msg_put_string(request, TCA_ACT_KIND, "vlan");
1294 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1295 {
1296 struct tc_vlan parm = { .action = TC_ACT_PIPE,
1297 .v_action = TCA_VLAN_ACT_POP };
1298
1299 nl_msg_put_unspec(request, TCA_VLAN_PARMS, &parm, sizeof parm);
1300 }
1301 nl_msg_end_nested(request, offset);
1302 }
1303
1304 static void
1305 nl_msg_put_act_tunnel_key_release(struct ofpbuf *request)
1306 {
1307 size_t offset;
1308
1309 nl_msg_put_string(request, TCA_ACT_KIND, "tunnel_key");
1310 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1311 {
1312 struct tc_tunnel_key tun = { .action = TC_ACT_PIPE,
1313 .t_action = TCA_TUNNEL_KEY_ACT_RELEASE };
1314
1315 nl_msg_put_unspec(request, TCA_TUNNEL_KEY_PARMS, &tun, sizeof tun);
1316 }
1317 nl_msg_end_nested(request, offset);
1318 }
1319
1320 static void
1321 nl_msg_put_act_tunnel_key_set(struct ofpbuf *request, ovs_be64 id,
1322 ovs_be32 ipv4_src, ovs_be32 ipv4_dst,
1323 struct in6_addr *ipv6_src,
1324 struct in6_addr *ipv6_dst,
1325 ovs_be16 tp_dst,
1326 uint8_t tos, uint8_t ttl)
1327 {
1328 size_t offset;
1329
1330 nl_msg_put_string(request, TCA_ACT_KIND, "tunnel_key");
1331 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1332 {
1333 struct tc_tunnel_key tun = { .action = TC_ACT_PIPE,
1334 .t_action = TCA_TUNNEL_KEY_ACT_SET };
1335
1336 nl_msg_put_unspec(request, TCA_TUNNEL_KEY_PARMS, &tun, sizeof tun);
1337
1338 ovs_be32 id32 = be64_to_be32(id);
1339 nl_msg_put_be32(request, TCA_TUNNEL_KEY_ENC_KEY_ID, id32);
1340 if (ipv4_dst) {
1341 nl_msg_put_be32(request, TCA_TUNNEL_KEY_ENC_IPV4_SRC, ipv4_src);
1342 nl_msg_put_be32(request, TCA_TUNNEL_KEY_ENC_IPV4_DST, ipv4_dst);
1343 } else if (!is_all_zeros(ipv6_dst, sizeof *ipv6_dst)) {
1344 nl_msg_put_in6_addr(request, TCA_TUNNEL_KEY_ENC_IPV6_DST,
1345 ipv6_dst);
1346 nl_msg_put_in6_addr(request, TCA_TUNNEL_KEY_ENC_IPV6_SRC,
1347 ipv6_src);
1348 }
1349 if (tos) {
1350 nl_msg_put_u8(request, TCA_TUNNEL_KEY_ENC_TOS, tos);
1351 }
1352 if (ttl) {
1353 nl_msg_put_u8(request, TCA_TUNNEL_KEY_ENC_TTL, ttl);
1354 }
1355 nl_msg_put_be16(request, TCA_TUNNEL_KEY_ENC_DST_PORT, tp_dst);
1356 }
1357 nl_msg_end_nested(request, offset);
1358 }
1359
1360 static void
1361 nl_msg_put_act_drop(struct ofpbuf *request)
1362 {
1363 size_t offset;
1364
1365 nl_msg_put_string(request, TCA_ACT_KIND, "gact");
1366 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1367 {
1368 struct tc_gact p = { .action = TC_ACT_SHOT };
1369
1370 nl_msg_put_unspec(request, TCA_GACT_PARMS, &p, sizeof p);
1371 }
1372 nl_msg_end_nested(request, offset);
1373 }
1374
1375 static void
1376 nl_msg_put_act_mirred(struct ofpbuf *request, int ifindex, int action,
1377 int eaction)
1378 {
1379 size_t offset;
1380
1381 nl_msg_put_string(request, TCA_ACT_KIND, "mirred");
1382 offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS);
1383 {
1384 struct tc_mirred m = { .action = action,
1385 .eaction = eaction,
1386 .ifindex = ifindex };
1387
1388 nl_msg_put_unspec(request, TCA_MIRRED_PARMS, &m, sizeof m);
1389 }
1390 nl_msg_end_nested(request, offset);
1391 }
1392
1393 static inline void
1394 nl_msg_put_act_cookie(struct ofpbuf *request, struct tc_cookie *ck) {
1395 if (ck->len) {
1396 nl_msg_put_unspec(request, TCA_ACT_COOKIE, ck->data, ck->len);
1397 }
1398 }
1399
1400 /* Given flower, a key_to_pedit map entry, calculates the rest,
1401 * where:
1402 *
1403 * mask, data - pointers of where read the first word of flower->key/mask.
1404 * current_offset - which offset to use for the first pedit action.
1405 * cnt - max pedits actions to use.
1406 * first_word_mask/last_word_mask - the mask to use for the first/last read
1407 * (as we read entire words). */
1408 static void
1409 calc_offsets(struct tc_flower *flower, struct flower_key_to_pedit *m,
1410 int *cur_offset, int *cnt, uint32_t *last_word_mask,
1411 uint32_t *first_word_mask, uint32_t **mask, uint32_t **data)
1412 {
1413 int start_offset, max_offset, total_size;
1414 int diff, right_zero_bits, left_zero_bits;
1415 char *rewrite_key = (void *) &flower->rewrite.key;
1416 char *rewrite_mask = (void *) &flower->rewrite.mask;
1417
1418 max_offset = m->offset + m->size;
1419 start_offset = ROUND_DOWN(m->offset, 4);
1420 diff = m->offset - start_offset;
1421 total_size = max_offset - start_offset;
1422 right_zero_bits = 8 * (4 - (max_offset % 4));
1423 left_zero_bits = 8 * (m->offset - start_offset);
1424
1425 *cur_offset = start_offset;
1426 *cnt = (total_size / 4) + (total_size % 4 ? 1 : 0);
1427 *last_word_mask = UINT32_MAX >> right_zero_bits;
1428 *first_word_mask = UINT32_MAX << left_zero_bits;
1429 *data = (void *) (rewrite_key + m->flower_offset - diff);
1430 *mask = (void *) (rewrite_mask + m->flower_offset - diff);
1431 }
1432
1433 static inline int
1434 csum_update_flag(struct tc_flower *flower,
1435 enum pedit_header_type htype) {
1436 /* Explictily specifiy the csum flags so HW can return EOPNOTSUPP
1437 * if it doesn't support a checksum recalculation of some headers.
1438 * And since OVS allows a flow such as
1439 * eth(dst=<mac>),eth_type(0x0800) actions=set(ipv4(src=<new_ip>))
1440 * we need to force a more specific flow as this can, for example,
1441 * need a recalculation of icmp checksum if the packet that passes
1442 * is ICMPv6 and tcp checksum if its tcp. */
1443
1444 switch (htype) {
1445 case TCA_PEDIT_KEY_EX_HDR_TYPE_IP4:
1446 flower->csum_update_flags |= TCA_CSUM_UPDATE_FLAG_IPV4HDR;
1447 /* Fall through. */
1448 case TCA_PEDIT_KEY_EX_HDR_TYPE_IP6:
1449 case TCA_PEDIT_KEY_EX_HDR_TYPE_TCP:
1450 case TCA_PEDIT_KEY_EX_HDR_TYPE_UDP:
1451 if (flower->key.ip_proto == IPPROTO_TCP) {
1452 flower->needs_full_ip_proto_mask = true;
1453 flower->csum_update_flags |= TCA_CSUM_UPDATE_FLAG_TCP;
1454 } else if (flower->key.ip_proto == IPPROTO_UDP) {
1455 flower->needs_full_ip_proto_mask = true;
1456 flower->csum_update_flags |= TCA_CSUM_UPDATE_FLAG_UDP;
1457 } else if (flower->key.ip_proto == IPPROTO_ICMP) {
1458 flower->needs_full_ip_proto_mask = true;
1459 } else if (flower->key.ip_proto == IPPROTO_ICMPV6) {
1460 flower->needs_full_ip_proto_mask = true;
1461 flower->csum_update_flags |= TCA_CSUM_UPDATE_FLAG_ICMP;
1462 } else {
1463 VLOG_WARN_RL(&error_rl,
1464 "can't offload rewrite of IP/IPV6 with ip_proto: %d",
1465 flower->key.ip_proto);
1466 break;
1467 }
1468 /* Fall through. */
1469 case TCA_PEDIT_KEY_EX_HDR_TYPE_ETH:
1470 return 0; /* success */
1471
1472 case TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK:
1473 case __PEDIT_HDR_TYPE_MAX:
1474 default:
1475 break;
1476 }
1477
1478 return EOPNOTSUPP;
1479 }
1480
1481 static int
1482 nl_msg_put_flower_rewrite_pedits(struct ofpbuf *request,
1483 struct tc_flower *flower)
1484 {
1485 struct {
1486 struct tc_pedit sel;
1487 struct tc_pedit_key keys[MAX_PEDIT_OFFSETS];
1488 struct tc_pedit_key_ex keys_ex[MAX_PEDIT_OFFSETS];
1489 } sel = {
1490 .sel = {
1491 .nkeys = 0
1492 }
1493 };
1494 int i, j, err;
1495
1496 for (i = 0; i < ARRAY_SIZE(flower_pedit_map); i++) {
1497 struct flower_key_to_pedit *m = &flower_pedit_map[i];
1498 struct tc_pedit_key *pedit_key = NULL;
1499 struct tc_pedit_key_ex *pedit_key_ex = NULL;
1500 uint32_t *mask, *data, first_word_mask, last_word_mask;
1501 int cnt = 0, cur_offset = 0;
1502
1503 if (!m->size) {
1504 continue;
1505 }
1506
1507 calc_offsets(flower, m, &cur_offset, &cnt, &last_word_mask,
1508 &first_word_mask, &mask, &data);
1509
1510 for (j = 0; j < cnt; j++, mask++, data++, cur_offset += 4) {
1511 uint32_t mask_word = *mask;
1512
1513 if (j == 0) {
1514 mask_word &= first_word_mask;
1515 }
1516 if (j == cnt - 1) {
1517 mask_word &= last_word_mask;
1518 }
1519 if (!mask_word) {
1520 continue;
1521 }
1522 if (sel.sel.nkeys == MAX_PEDIT_OFFSETS) {
1523 VLOG_WARN_RL(&error_rl, "reached too many pedit offsets: %d",
1524 MAX_PEDIT_OFFSETS);
1525 return EOPNOTSUPP;
1526 }
1527
1528 pedit_key = &sel.keys[sel.sel.nkeys];
1529 pedit_key_ex = &sel.keys_ex[sel.sel.nkeys];
1530 pedit_key_ex->cmd = TCA_PEDIT_KEY_EX_CMD_SET;
1531 pedit_key_ex->htype = m->htype;
1532 pedit_key->off = cur_offset;
1533 pedit_key->mask = ~mask_word;
1534 pedit_key->val = *data & mask_word;
1535 sel.sel.nkeys++;
1536
1537 err = csum_update_flag(flower, m->htype);
1538 if (err) {
1539 return err;
1540 }
1541
1542 if (flower->needs_full_ip_proto_mask) {
1543 flower->mask.ip_proto = UINT8_MAX;
1544 }
1545 }
1546 }
1547 nl_msg_put_act_pedit(request, &sel.sel, sel.keys_ex);
1548
1549 return 0;
1550 }
1551
1552 static int
1553 nl_msg_put_flower_acts(struct ofpbuf *request, struct tc_flower *flower)
1554 {
1555 size_t offset;
1556 size_t act_offset;
1557 uint16_t act_index = 1;
1558 struct tc_action *action;
1559 int i, ifindex = 0;
1560
1561 offset = nl_msg_start_nested(request, TCA_FLOWER_ACT);
1562 {
1563 int error;
1564
1565 if (flower->tunnel) {
1566 act_offset = nl_msg_start_nested(request, act_index++);
1567 nl_msg_put_act_tunnel_key_release(request);
1568 nl_msg_end_nested(request, act_offset);
1569 }
1570
1571 action = flower->actions;
1572 for (i = 0; i < flower->action_count; i++, action++) {
1573 switch (action->type) {
1574 case TC_ACT_PEDIT: {
1575 act_offset = nl_msg_start_nested(request, act_index++);
1576 error = nl_msg_put_flower_rewrite_pedits(request, flower);
1577 if (error) {
1578 return error;
1579 }
1580 nl_msg_end_nested(request, act_offset);
1581
1582 if (flower->csum_update_flags) {
1583 act_offset = nl_msg_start_nested(request, act_index++);
1584 nl_msg_put_act_csum(request, flower->csum_update_flags);
1585 nl_msg_end_nested(request, act_offset);
1586 }
1587 }
1588 break;
1589 case TC_ACT_ENCAP: {
1590 act_offset = nl_msg_start_nested(request, act_index++);
1591 nl_msg_put_act_tunnel_key_set(request, action->encap.id,
1592 action->encap.ipv4.ipv4_src,
1593 action->encap.ipv4.ipv4_dst,
1594 &action->encap.ipv6.ipv6_src,
1595 &action->encap.ipv6.ipv6_dst,
1596 action->encap.tp_dst,
1597 action->encap.tos,
1598 action->encap.ttl);
1599 nl_msg_end_nested(request, act_offset);
1600 }
1601 break;
1602 case TC_ACT_VLAN_POP: {
1603 act_offset = nl_msg_start_nested(request, act_index++);
1604 nl_msg_put_act_pop_vlan(request);
1605 nl_msg_end_nested(request, act_offset);
1606 }
1607 break;
1608 case TC_ACT_VLAN_PUSH: {
1609 act_offset = nl_msg_start_nested(request, act_index++);
1610 nl_msg_put_act_push_vlan(request,
1611 action->vlan.vlan_push_tpid,
1612 action->vlan.vlan_push_id,
1613 action->vlan.vlan_push_prio);
1614 nl_msg_end_nested(request, act_offset);
1615 }
1616 break;
1617 case TC_ACT_OUTPUT: {
1618 ifindex = action->ifindex_out;
1619 if (ifindex < 1) {
1620 VLOG_ERR_RL(&error_rl, "%s: invalid ifindex: %d, type: %d",
1621 __func__, ifindex, action->type);
1622 return EINVAL;
1623 }
1624 act_offset = nl_msg_start_nested(request, act_index++);
1625 if (i == flower->action_count - 1) {
1626 nl_msg_put_act_mirred(request, ifindex, TC_ACT_STOLEN,
1627 TCA_EGRESS_REDIR);
1628 } else {
1629 nl_msg_put_act_mirred(request, ifindex, TC_ACT_PIPE,
1630 TCA_EGRESS_MIRROR);
1631 }
1632 nl_msg_put_act_cookie(request, &flower->act_cookie);
1633 nl_msg_end_nested(request, act_offset);
1634 }
1635 break;
1636 }
1637 }
1638 }
1639 if (!ifindex) {
1640 act_offset = nl_msg_start_nested(request, act_index++);
1641 nl_msg_put_act_drop(request);
1642 nl_msg_put_act_cookie(request, &flower->act_cookie);
1643 nl_msg_end_nested(request, act_offset);
1644 }
1645 nl_msg_end_nested(request, offset);
1646
1647 return 0;
1648 }
1649
1650 static void
1651 nl_msg_put_masked_value(struct ofpbuf *request, uint16_t type,
1652 uint16_t mask_type, const void *data,
1653 const void *mask_data, size_t len)
1654 {
1655 if (mask_type != TCA_FLOWER_UNSPEC) {
1656 if (is_all_zeros(mask_data, len)) {
1657 return;
1658 }
1659 nl_msg_put_unspec(request, mask_type, mask_data, len);
1660 }
1661 nl_msg_put_unspec(request, type, data, len);
1662 }
1663
1664 static void
1665 nl_msg_put_flower_tunnel(struct ofpbuf *request, struct tc_flower *flower)
1666 {
1667 ovs_be32 ipv4_src = flower->key.tunnel.ipv4.ipv4_src;
1668 ovs_be32 ipv4_dst = flower->key.tunnel.ipv4.ipv4_dst;
1669 struct in6_addr *ipv6_src = &flower->key.tunnel.ipv6.ipv6_src;
1670 struct in6_addr *ipv6_dst = &flower->key.tunnel.ipv6.ipv6_dst;
1671 ovs_be16 tp_dst = flower->key.tunnel.tp_dst;
1672 ovs_be32 id = be64_to_be32(flower->key.tunnel.id);
1673 uint8_t tos = flower->key.tunnel.tos;
1674 uint8_t ttl = flower->key.tunnel.ttl;
1675 uint8_t tos_mask = flower->mask.tunnel.tos;
1676 uint8_t ttl_mask = flower->mask.tunnel.ttl;
1677
1678 if (ipv4_dst) {
1679 nl_msg_put_be32(request, TCA_FLOWER_KEY_ENC_IPV4_SRC, ipv4_src);
1680 nl_msg_put_be32(request, TCA_FLOWER_KEY_ENC_IPV4_DST, ipv4_dst);
1681 } else if (!is_all_zeros(ipv6_dst, sizeof *ipv6_dst)) {
1682 nl_msg_put_in6_addr(request, TCA_FLOWER_KEY_ENC_IPV6_SRC, ipv6_src);
1683 nl_msg_put_in6_addr(request, TCA_FLOWER_KEY_ENC_IPV6_DST, ipv6_dst);
1684 }
1685 if (tos_mask) {
1686 nl_msg_put_u8(request, TCA_FLOWER_KEY_ENC_IP_TOS, tos);
1687 nl_msg_put_u8(request, TCA_FLOWER_KEY_ENC_IP_TOS_MASK, tos_mask);
1688 }
1689 if (ttl_mask) {
1690 nl_msg_put_u8(request, TCA_FLOWER_KEY_ENC_IP_TTL, ttl);
1691 nl_msg_put_u8(request, TCA_FLOWER_KEY_ENC_IP_TTL_MASK, ttl_mask);
1692 }
1693 nl_msg_put_be16(request, TCA_FLOWER_KEY_ENC_UDP_DST_PORT, tp_dst);
1694 nl_msg_put_be32(request, TCA_FLOWER_KEY_ENC_KEY_ID, id);
1695 }
1696
1697 #define FLOWER_PUT_MASKED_VALUE(member, type) \
1698 nl_msg_put_masked_value(request, type, type##_MASK, &flower->key.member, \
1699 &flower->mask.member, sizeof flower->key.member)
1700
1701 static int
1702 nl_msg_put_flower_options(struct ofpbuf *request, struct tc_flower *flower)
1703 {
1704
1705 uint16_t host_eth_type = ntohs(flower->key.eth_type);
1706 bool is_vlan = eth_type_vlan(flower->key.eth_type);
1707 bool is_qinq = is_vlan && eth_type_vlan(flower->key.encap_eth_type[0]);
1708 bool is_mpls = eth_type_mpls(flower->key.eth_type);
1709 int err;
1710
1711 /* need to parse acts first as some acts require changing the matching
1712 * see csum_update_flag() */
1713 err = nl_msg_put_flower_acts(request, flower);
1714 if (err) {
1715 return err;
1716 }
1717
1718 if (is_vlan) {
1719 if (is_qinq) {
1720 host_eth_type = ntohs(flower->key.encap_eth_type[1]);
1721 } else {
1722 host_eth_type = ntohs(flower->key.encap_eth_type[0]);
1723 }
1724 }
1725
1726 if (is_mpls) {
1727 host_eth_type = ntohs(flower->key.encap_eth_type[0]);
1728 }
1729
1730 FLOWER_PUT_MASKED_VALUE(dst_mac, TCA_FLOWER_KEY_ETH_DST);
1731 FLOWER_PUT_MASKED_VALUE(src_mac, TCA_FLOWER_KEY_ETH_SRC);
1732
1733 if (host_eth_type == ETH_P_IP || host_eth_type == ETH_P_IPV6) {
1734 FLOWER_PUT_MASKED_VALUE(ip_ttl, TCA_FLOWER_KEY_IP_TTL);
1735 FLOWER_PUT_MASKED_VALUE(ip_tos, TCA_FLOWER_KEY_IP_TOS);
1736
1737 if (flower->mask.ip_proto && flower->key.ip_proto) {
1738 nl_msg_put_u8(request, TCA_FLOWER_KEY_IP_PROTO,
1739 flower->key.ip_proto);
1740 }
1741
1742 if (flower->mask.flags) {
1743 nl_msg_put_be32(request, TCA_FLOWER_KEY_FLAGS,
1744 htonl(flower->key.flags));
1745 nl_msg_put_be32(request, TCA_FLOWER_KEY_FLAGS_MASK,
1746 htonl(flower->mask.flags));
1747 }
1748
1749 if (flower->key.ip_proto == IPPROTO_UDP) {
1750 FLOWER_PUT_MASKED_VALUE(udp_src, TCA_FLOWER_KEY_UDP_SRC);
1751 FLOWER_PUT_MASKED_VALUE(udp_dst, TCA_FLOWER_KEY_UDP_DST);
1752 } else if (flower->key.ip_proto == IPPROTO_TCP) {
1753 FLOWER_PUT_MASKED_VALUE(tcp_src, TCA_FLOWER_KEY_TCP_SRC);
1754 FLOWER_PUT_MASKED_VALUE(tcp_dst, TCA_FLOWER_KEY_TCP_DST);
1755 FLOWER_PUT_MASKED_VALUE(tcp_flags, TCA_FLOWER_KEY_TCP_FLAGS);
1756 } else if (flower->key.ip_proto == IPPROTO_SCTP) {
1757 FLOWER_PUT_MASKED_VALUE(sctp_src, TCA_FLOWER_KEY_SCTP_SRC);
1758 FLOWER_PUT_MASKED_VALUE(sctp_dst, TCA_FLOWER_KEY_SCTP_DST);
1759 }
1760 }
1761
1762 if (host_eth_type == ETH_P_IP) {
1763 FLOWER_PUT_MASKED_VALUE(ipv4.ipv4_src, TCA_FLOWER_KEY_IPV4_SRC);
1764 FLOWER_PUT_MASKED_VALUE(ipv4.ipv4_dst, TCA_FLOWER_KEY_IPV4_DST);
1765 } else if (host_eth_type == ETH_P_IPV6) {
1766 FLOWER_PUT_MASKED_VALUE(ipv6.ipv6_src, TCA_FLOWER_KEY_IPV6_SRC);
1767 FLOWER_PUT_MASKED_VALUE(ipv6.ipv6_dst, TCA_FLOWER_KEY_IPV6_DST);
1768 }
1769
1770 nl_msg_put_be16(request, TCA_FLOWER_KEY_ETH_TYPE, flower->key.eth_type);
1771
1772 if (is_mpls) {
1773 if (mpls_lse_to_ttl(flower->mask.mpls_lse)) {
1774 nl_msg_put_u8(request, TCA_FLOWER_KEY_MPLS_TTL,
1775 mpls_lse_to_ttl(flower->key.mpls_lse));
1776 }
1777 if (mpls_lse_to_tc(flower->mask.mpls_lse)) {
1778 nl_msg_put_u8(request, TCA_FLOWER_KEY_MPLS_TC,
1779 mpls_lse_to_tc(flower->key.mpls_lse));
1780 }
1781 if (mpls_lse_to_bos(flower->mask.mpls_lse)) {
1782 nl_msg_put_u8(request, TCA_FLOWER_KEY_MPLS_BOS,
1783 mpls_lse_to_bos(flower->key.mpls_lse));
1784 }
1785 if (mpls_lse_to_label(flower->mask.mpls_lse)) {
1786 nl_msg_put_u32(request, TCA_FLOWER_KEY_MPLS_LABEL,
1787 mpls_lse_to_label(flower->key.mpls_lse));
1788 }
1789 }
1790
1791 if (is_vlan) {
1792 if (flower->key.vlan_id[0] || flower->key.vlan_prio[0]) {
1793 nl_msg_put_u16(request, TCA_FLOWER_KEY_VLAN_ID,
1794 flower->key.vlan_id[0]);
1795 nl_msg_put_u8(request, TCA_FLOWER_KEY_VLAN_PRIO,
1796 flower->key.vlan_prio[0]);
1797 }
1798 if (flower->key.encap_eth_type[0]) {
1799 nl_msg_put_be16(request, TCA_FLOWER_KEY_VLAN_ETH_TYPE,
1800 flower->key.encap_eth_type[0]);
1801 }
1802
1803 if (is_qinq) {
1804 if (flower->key.vlan_id[1] || flower->key.vlan_prio[1]) {
1805 nl_msg_put_u16(request, TCA_FLOWER_KEY_CVLAN_ID,
1806 flower->key.vlan_id[1]);
1807 nl_msg_put_u8(request, TCA_FLOWER_KEY_CVLAN_PRIO,
1808 flower->key.vlan_prio[1]);
1809 }
1810 if (flower->key.encap_eth_type[1]) {
1811 nl_msg_put_be16(request, TCA_FLOWER_KEY_CVLAN_ETH_TYPE,
1812 flower->key.encap_eth_type[1]);
1813 }
1814 }
1815 }
1816
1817 nl_msg_put_u32(request, TCA_FLOWER_FLAGS, tc_get_tc_cls_policy(tc_policy));
1818
1819 if (flower->tunnel) {
1820 nl_msg_put_flower_tunnel(request, flower);
1821 }
1822
1823 return 0;
1824 }
1825
1826 int
1827 tc_replace_flower(int ifindex, uint16_t prio, uint32_t handle,
1828 struct tc_flower *flower, uint32_t block_id)
1829 {
1830 struct ofpbuf request;
1831 struct tcmsg *tcmsg;
1832 struct ofpbuf *reply;
1833 int error = 0;
1834 size_t basic_offset;
1835 uint16_t eth_type = (OVS_FORCE uint16_t) flower->key.eth_type;
1836 int index;
1837
1838 index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex;
1839 tcmsg = tc_make_request(index, RTM_NEWTFILTER, NLM_F_CREATE | NLM_F_ECHO,
1840 &request);
1841 tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT;
1842 tcmsg->tcm_info = tc_make_handle(prio, eth_type);
1843 tcmsg->tcm_handle = handle;
1844
1845 nl_msg_put_string(&request, TCA_KIND, "flower");
1846 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
1847 {
1848 error = nl_msg_put_flower_options(&request, flower);
1849
1850 if (error) {
1851 ofpbuf_uninit(&request);
1852 return error;
1853 }
1854 }
1855 nl_msg_end_nested(&request, basic_offset);
1856
1857 error = tc_transact(&request, &reply);
1858 if (!error) {
1859 struct tcmsg *tc =
1860 ofpbuf_at_assert(reply, NLMSG_HDRLEN, sizeof *tc);
1861
1862 flower->prio = tc_get_major(tc->tcm_info);
1863 flower->handle = tc->tcm_handle;
1864 ofpbuf_delete(reply);
1865 }
1866
1867 return error;
1868 }
1869
1870 void
1871 tc_set_policy(const char *policy)
1872 {
1873 if (!policy) {
1874 return;
1875 }
1876
1877 if (!strcmp(policy, "skip_sw")) {
1878 tc_policy = TC_POLICY_SKIP_SW;
1879 } else if (!strcmp(policy, "skip_hw")) {
1880 tc_policy = TC_POLICY_SKIP_HW;
1881 } else if (!strcmp(policy, "none")) {
1882 tc_policy = TC_POLICY_NONE;
1883 } else {
1884 VLOG_WARN("tc: Invalid policy '%s'", policy);
1885 return;
1886 }
1887
1888 VLOG_INFO("tc: Using policy '%s'", policy);
1889 }
openvswitch mirror