]>
Commit | Line | Data |
---|---|---|
96fba48f | 1 | /* |
4ea96698 | 2 | * Copyright (c) 2008-2018 Nicira, Inc. |
96fba48f BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
9fe3b9a2 | 18 | |
93451a0a | 19 | #include "dpif-netlink.h" |
96fba48f | 20 | |
96fba48f BP |
21 | #include <ctype.h> |
22 | #include <errno.h> | |
23 | #include <fcntl.h> | |
24 | #include <inttypes.h> | |
25 | #include <net/if.h> | |
b90fa799 | 26 | #include <linux/types.h> |
aae51f53 | 27 | #include <linux/pkt_sched.h> |
8522ba09 | 28 | #include <poll.h> |
96fba48f | 29 | #include <stdlib.h> |
8522ba09 | 30 | #include <strings.h> |
50f80534 | 31 | #include <sys/epoll.h> |
10dcf8de | 32 | #include <sys/stat.h> |
96fba48f BP |
33 | #include <unistd.h> |
34 | ||
773cd538 | 35 | #include "bitmap.h" |
c4e08753 | 36 | #include "dpif-netlink-rtnl.h" |
0d71302e | 37 | #include "dpif-provider.h" |
1579cf67 | 38 | #include "fat-rwlock.h" |
0d71302e | 39 | #include "flow.h" |
032aa6a3 | 40 | #include "netdev-linux.h" |
b6cabb8f | 41 | #include "netdev-offload.h" |
0d71302e | 42 | #include "netdev-provider.h" |
c3827f61 | 43 | #include "netdev-vport.h" |
0d71302e | 44 | #include "netdev.h" |
c11c9f4a | 45 | #include "netlink-conntrack.h" |
45c8d3a1 | 46 | #include "netlink-notifier.h" |
982b8810 | 47 | #include "netlink-socket.h" |
856081f6 | 48 | #include "netlink.h" |
bfda5239 | 49 | #include "netnsid.h" |
feebdea2 | 50 | #include "odp-util.h" |
0d71302e BP |
51 | #include "openvswitch/dynamic-string.h" |
52 | #include "openvswitch/flow.h" | |
1f161318 | 53 | #include "openvswitch/hmap.h" |
0d71302e | 54 | #include "openvswitch/match.h" |
64c96779 | 55 | #include "openvswitch/ofpbuf.h" |
fd016ae3 | 56 | #include "openvswitch/poll-loop.h" |
ee89ea7b | 57 | #include "openvswitch/shash.h" |
92d0d515 | 58 | #include "openvswitch/thread.h" |
0d71302e BP |
59 | #include "openvswitch/vlog.h" |
60 | #include "packets.h" | |
61 | #include "random.h" | |
b3c01ed3 | 62 | #include "sset.h" |
14b4d2f9 | 63 | #include "timeval.h" |
d6569377 | 64 | #include "unaligned.h" |
96fba48f | 65 | #include "util.h" |
5136ce49 | 66 | |
93451a0a | 67 | VLOG_DEFINE_THIS_MODULE(dpif_netlink); |
09cac43f | 68 | #ifdef _WIN32 |
da467899 | 69 | #include "wmi.h" |
09cac43f NR |
70 | enum { WINDOWS = 1 }; |
71 | #else | |
72 | enum { WINDOWS = 0 }; | |
73 | #endif | |
95b1d73a | 74 | enum { MAX_PORTS = USHRT_MAX }; |
773cd538 | 75 | |
24b019f8 JP |
76 | /* This ethtool flag was introduced in Linux 2.6.24, so it might be |
77 | * missing if we have old headers. */ | |
78 | #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */ | |
79 | ||
f2280b41 | 80 | #define FLOW_DUMP_MAX_BATCH 50 |
8b668ee3 | 81 | #define OPERATE_MAX_OPS 50 |
f2280b41 | 82 | |
69c51582 MC |
83 | #ifndef EPOLLEXCLUSIVE |
84 | #define EPOLLEXCLUSIVE (1u << 28) | |
85 | #endif | |
86 | ||
93451a0a | 87 | struct dpif_netlink_dp { |
aaff4b55 BP |
88 | /* Generic Netlink header. */ |
89 | uint8_t cmd; | |
d6569377 | 90 | |
df2c07f4 | 91 | /* struct ovs_header. */ |
254f2dc8 | 92 | int dp_ifindex; |
d6569377 BP |
93 | |
94 | /* Attributes. */ | |
df2c07f4 | 95 | const char *name; /* OVS_DP_ATTR_NAME. */ |
fcd5d230 | 96 | const uint32_t *upcall_pid; /* OVS_DP_ATTR_UPCALL_PID. */ |
b7fd5e38 | 97 | uint32_t user_features; /* OVS_DP_ATTR_USER_FEATURES */ |
6a54dedc BP |
98 | const struct ovs_dp_stats *stats; /* OVS_DP_ATTR_STATS. */ |
99 | const struct ovs_dp_megaflow_stats *megaflow_stats; | |
847108dc | 100 | /* OVS_DP_ATTR_MEGAFLOW_STATS.*/ |
d6569377 BP |
101 | }; |
102 | ||
93451a0a AS |
103 | static void dpif_netlink_dp_init(struct dpif_netlink_dp *); |
104 | static int dpif_netlink_dp_from_ofpbuf(struct dpif_netlink_dp *, | |
105 | const struct ofpbuf *); | |
106 | static void dpif_netlink_dp_dump_start(struct nl_dump *); | |
107 | static int dpif_netlink_dp_transact(const struct dpif_netlink_dp *request, | |
108 | struct dpif_netlink_dp *reply, | |
109 | struct ofpbuf **bufp); | |
110 | static int dpif_netlink_dp_get(const struct dpif *, | |
111 | struct dpif_netlink_dp *reply, | |
112 | struct ofpbuf **bufp); | |
b2ae4069 PB |
113 | static int |
114 | dpif_netlink_set_features(struct dpif *dpif_, uint32_t new_features); | |
93451a0a AS |
115 | |
116 | struct dpif_netlink_flow { | |
37a1300c BP |
117 | /* Generic Netlink header. */ |
118 | uint8_t cmd; | |
d6569377 | 119 | |
df2c07f4 | 120 | /* struct ovs_header. */ |
d6569377 | 121 | unsigned int nlmsg_flags; |
254f2dc8 | 122 | int dp_ifindex; |
d6569377 BP |
123 | |
124 | /* Attributes. | |
125 | * | |
0e70cdcb BP |
126 | * The 'stats' member points to 64-bit data that might only be aligned on |
127 | * 32-bit boundaries, so get_unaligned_u64() should be used to access its | |
128 | * values. | |
d2a23af2 | 129 | * |
df2c07f4 | 130 | * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in |
d2a23af2 | 131 | * the Netlink version of the command, even if actions_len is zero. */ |
df2c07f4 | 132 | const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */ |
d6569377 | 133 | size_t key_len; |
e6cc0bab AZ |
134 | const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */ |
135 | size_t mask_len; | |
df2c07f4 | 136 | const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */ |
d6569377 | 137 | size_t actions_len; |
70e5ed6f JS |
138 | ovs_u128 ufid; /* OVS_FLOW_ATTR_FLOW_ID. */ |
139 | bool ufid_present; /* Is there a UFID? */ | |
140 | bool ufid_terse; /* Skip serializing key/mask/acts? */ | |
df2c07f4 JP |
141 | const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */ |
142 | const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */ | |
0e70cdcb | 143 | const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */ |
df2c07f4 | 144 | bool clear; /* OVS_FLOW_ATTR_CLEAR. */ |
43f9ac0a | 145 | bool probe; /* OVS_FLOW_ATTR_PROBE. */ |
d6569377 BP |
146 | }; |
147 | ||
93451a0a AS |
148 | static void dpif_netlink_flow_init(struct dpif_netlink_flow *); |
149 | static int dpif_netlink_flow_from_ofpbuf(struct dpif_netlink_flow *, | |
150 | const struct ofpbuf *); | |
151 | static void dpif_netlink_flow_to_ofpbuf(const struct dpif_netlink_flow *, | |
152 | struct ofpbuf *); | |
153 | static int dpif_netlink_flow_transact(struct dpif_netlink_flow *request, | |
154 | struct dpif_netlink_flow *reply, | |
155 | struct ofpbuf **bufp); | |
156 | static void dpif_netlink_flow_get_stats(const struct dpif_netlink_flow *, | |
157 | struct dpif_flow_stats *); | |
7a5e0ee7 | 158 | static void dpif_netlink_flow_to_dpif_flow(struct dpif_flow *, |
93451a0a | 159 | const struct dpif_netlink_flow *); |
d6569377 | 160 | |
989fd548 | 161 | /* One of the dpif channels between the kernel and userspace. */ |
fe3d61b3 | 162 | struct dpif_channel { |
14b4d2f9 | 163 | struct nl_sock *sock; /* Netlink socket. */ |
14b4d2f9 | 164 | long long int last_poll; /* Last time this channel was polled. */ |
fe3d61b3 BP |
165 | }; |
166 | ||
09cac43f NR |
167 | #ifdef _WIN32 |
168 | #define VPORT_SOCK_POOL_SIZE 1 | |
169 | /* On Windows, there is no native support for epoll. There are equivalent | |
170 | * interfaces though, that are not used currently. For simpicity, a pool of | |
171 | * netlink sockets is used. Each socket is represented by 'struct | |
172 | * dpif_windows_vport_sock'. Since it is a pool, multiple OVS ports may be | |
173 | * sharing the same socket. In the future, we can add a reference count and | |
174 | * such fields. */ | |
175 | struct dpif_windows_vport_sock { | |
176 | struct nl_sock *nl_sock; /* netlink socket. */ | |
177 | }; | |
178 | #endif | |
179 | ||
1579cf67 | 180 | struct dpif_handler { |
1579cf67 AW |
181 | struct epoll_event *epoll_events; |
182 | int epoll_fd; /* epoll fd that includes channel socks. */ | |
183 | int n_events; /* Num events returned by epoll_wait(). */ | |
184 | int event_offset; /* Offset into 'epoll_events'. */ | |
09cac43f NR |
185 | |
186 | #ifdef _WIN32 | |
187 | /* Pool of sockets. */ | |
188 | struct dpif_windows_vport_sock *vport_sock_pool; | |
189 | size_t last_used_pool_idx; /* Index to aid in allocating a | |
190 | socket in the pool to a port. */ | |
191 | #endif | |
1579cf67 | 192 | }; |
14b4d2f9 | 193 | |
96fba48f | 194 | /* Datapath interface for the openvswitch Linux kernel module. */ |
93451a0a | 195 | struct dpif_netlink { |
96fba48f | 196 | struct dpif dpif; |
254f2dc8 | 197 | int dp_ifindex; |
dcdcad68 | 198 | uint32_t user_features; |
e9e28be3 | 199 | |
b063d9f0 | 200 | /* Upcall messages. */ |
1579cf67 AW |
201 | struct fat_rwlock upcall_lock; |
202 | struct dpif_handler *handlers; | |
203 | uint32_t n_handlers; /* Num of upcall handlers. */ | |
69c51582 | 204 | struct dpif_channel *channels; /* Array of channels for each port. */ |
1579cf67 AW |
205 | int uc_array_size; /* Size of 'handler->channels' and */ |
206 | /* 'handler->epoll_events'. */ | |
982b8810 | 207 | |
e9e28be3 | 208 | /* Change notification. */ |
e4516b20 | 209 | struct nl_sock *port_notifier; /* vport multicast group subscriber. */ |
61eae437 | 210 | bool refresh_channels; |
96fba48f BP |
211 | }; |
212 | ||
93451a0a | 213 | static void report_loss(struct dpif_netlink *, struct dpif_channel *, |
9b00386b | 214 | uint32_t ch_idx, uint32_t handler_id); |
1579cf67 | 215 | |
96fba48f BP |
216 | static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); |
217 | ||
e4516b20 BP |
218 | /* Generic Netlink family numbers for OVS. |
219 | * | |
93451a0a | 220 | * Initialized by dpif_netlink_init(). */ |
df2c07f4 JP |
221 | static int ovs_datapath_family; |
222 | static int ovs_vport_family; | |
223 | static int ovs_flow_family; | |
224 | static int ovs_packet_family; | |
80738e5f | 225 | static int ovs_meter_family; |
906ff9d2 | 226 | static int ovs_ct_limit_family; |
982b8810 | 227 | |
e4516b20 BP |
228 | /* Generic Netlink multicast groups for OVS. |
229 | * | |
93451a0a | 230 | * Initialized by dpif_netlink_init(). */ |
e4516b20 | 231 | static unsigned int ovs_vport_mcgroup; |
982b8810 | 232 | |
921c370a EG |
233 | /* If true, tunnel devices are created using OVS compat/genetlink. |
234 | * If false, tunnel devices are created with rtnetlink and using light weight | |
235 | * tunnels. If we fail to create the tunnel the rtnetlink+LWT, then we fallback | |
236 | * to using the compat interface. */ | |
237 | static bool ovs_tunnels_out_of_tree = true; | |
238 | ||
93451a0a AS |
239 | static int dpif_netlink_init(void); |
240 | static int open_dpif(const struct dpif_netlink_dp *, struct dpif **); | |
241 | static uint32_t dpif_netlink_port_get_pid(const struct dpif *, | |
769b5034 | 242 | odp_port_t port_no); |
09cac43f | 243 | static void dpif_netlink_handler_uninit(struct dpif_handler *handler); |
93451a0a AS |
244 | static int dpif_netlink_refresh_channels(struct dpif_netlink *, |
245 | uint32_t n_handlers); | |
246 | static void dpif_netlink_vport_to_ofpbuf(const struct dpif_netlink_vport *, | |
247 | struct ofpbuf *); | |
248 | static int dpif_netlink_vport_from_ofpbuf(struct dpif_netlink_vport *, | |
249 | const struct ofpbuf *); | |
921c370a EG |
250 | static int dpif_netlink_port_query__(const struct dpif_netlink *dpif, |
251 | odp_port_t port_no, const char *port_name, | |
252 | struct dpif_port *dpif_port); | |
f0fef760 | 253 | |
d240e46a | 254 | static int |
622ea8fd | 255 | create_nl_sock(struct dpif_netlink *dpif OVS_UNUSED, struct nl_sock **sockp) |
d240e46a AGS |
256 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
257 | { | |
258 | #ifndef _WIN32 | |
622ea8fd | 259 | return nl_sock_create(NETLINK_GENERIC, sockp); |
d240e46a AGS |
260 | #else |
261 | /* Pick netlink sockets to use in a round-robin fashion from each | |
262 | * handler's pool of sockets. */ | |
263 | struct dpif_handler *handler = &dpif->handlers[0]; | |
264 | struct dpif_windows_vport_sock *sock_pool = handler->vport_sock_pool; | |
265 | size_t index = handler->last_used_pool_idx; | |
266 | ||
267 | /* A pool of sockets is allocated when the handler is initialized. */ | |
268 | if (sock_pool == NULL) { | |
622ea8fd | 269 | *sockp = NULL; |
d240e46a AGS |
270 | return EINVAL; |
271 | } | |
272 | ||
273 | ovs_assert(index < VPORT_SOCK_POOL_SIZE); | |
622ea8fd BP |
274 | *sockp = sock_pool[index].nl_sock; |
275 | ovs_assert(*sockp); | |
d240e46a AGS |
276 | index = (index == VPORT_SOCK_POOL_SIZE - 1) ? 0 : index + 1; |
277 | handler->last_used_pool_idx = index; | |
278 | return 0; | |
279 | #endif | |
280 | } | |
281 | ||
282 | static void | |
622ea8fd | 283 | close_nl_sock(struct nl_sock *sock) |
d240e46a AGS |
284 | { |
285 | #ifndef _WIN32 | |
622ea8fd | 286 | nl_sock_destroy(sock); |
d240e46a AGS |
287 | #endif |
288 | } | |
289 | ||
93451a0a AS |
290 | static struct dpif_netlink * |
291 | dpif_netlink_cast(const struct dpif *dpif) | |
96fba48f | 292 | { |
93451a0a AS |
293 | dpif_assert_class(dpif, &dpif_netlink_class); |
294 | return CONTAINER_OF(dpif, struct dpif_netlink, dpif); | |
96fba48f BP |
295 | } |
296 | ||
d3d22744 | 297 | static int |
93451a0a AS |
298 | dpif_netlink_enumerate(struct sset *all_dps, |
299 | const struct dpif_class *dpif_class OVS_UNUSED) | |
d3d22744 | 300 | { |
aaff4b55 | 301 | struct nl_dump dump; |
d57695d7 JS |
302 | uint64_t reply_stub[NL_DUMP_BUFSIZE / 8]; |
303 | struct ofpbuf msg, buf; | |
aaff4b55 | 304 | int error; |
982b8810 | 305 | |
93451a0a | 306 | error = dpif_netlink_init(); |
aaff4b55 BP |
307 | if (error) { |
308 | return error; | |
982b8810 | 309 | } |
d3d22744 | 310 | |
d57695d7 | 311 | ofpbuf_use_stub(&buf, reply_stub, sizeof reply_stub); |
93451a0a | 312 | dpif_netlink_dp_dump_start(&dump); |
d57695d7 | 313 | while (nl_dump_next(&dump, &msg, &buf)) { |
93451a0a | 314 | struct dpif_netlink_dp dp; |
d6569377 | 315 | |
93451a0a | 316 | if (!dpif_netlink_dp_from_ofpbuf(&dp, &msg)) { |
d0c23a1a | 317 | sset_add(all_dps, dp.name); |
d3d22744 BP |
318 | } |
319 | } | |
d57695d7 | 320 | ofpbuf_uninit(&buf); |
aaff4b55 | 321 | return nl_dump_done(&dump); |
d3d22744 BP |
322 | } |
323 | ||
96fba48f | 324 | static int |
93451a0a AS |
325 | dpif_netlink_open(const struct dpif_class *class OVS_UNUSED, const char *name, |
326 | bool create, struct dpif **dpifp) | |
96fba48f | 327 | { |
93451a0a | 328 | struct dpif_netlink_dp dp_request, dp; |
c19e6535 | 329 | struct ofpbuf *buf; |
ea36840f | 330 | uint32_t upcall_pid; |
c19e6535 | 331 | int error; |
96fba48f | 332 | |
93451a0a | 333 | error = dpif_netlink_init(); |
982b8810 BP |
334 | if (error) { |
335 | return error; | |
336 | } | |
337 | ||
982b8810 | 338 | /* Create or look up datapath. */ |
93451a0a | 339 | dpif_netlink_dp_init(&dp_request); |
dcdcad68 PB |
340 | upcall_pid = 0; |
341 | dp_request.upcall_pid = &upcall_pid; | |
342 | dp_request.name = name; | |
343 | ||
ea36840f BP |
344 | if (create) { |
345 | dp_request.cmd = OVS_DP_CMD_NEW; | |
ea36840f | 346 | } else { |
dcdcad68 PB |
347 | dp_request.cmd = OVS_DP_CMD_GET; |
348 | ||
349 | error = dpif_netlink_dp_transact(&dp_request, &dp, &buf); | |
350 | if (error) { | |
351 | return error; | |
352 | } | |
353 | dp_request.user_features = dp.user_features; | |
354 | ofpbuf_delete(buf); | |
355 | ||
b7fd5e38 TG |
356 | /* Use OVS_DP_CMD_SET to report user features */ |
357 | dp_request.cmd = OVS_DP_CMD_SET; | |
ea36840f | 358 | } |
dcdcad68 | 359 | |
b7fd5e38 | 360 | dp_request.user_features |= OVS_DP_F_UNALIGNED; |
1579cf67 | 361 | dp_request.user_features |= OVS_DP_F_VPORT_PIDS; |
93451a0a | 362 | error = dpif_netlink_dp_transact(&dp_request, &dp, &buf); |
982b8810 BP |
363 | if (error) { |
364 | return error; | |
c19e6535 | 365 | } |
254f2dc8 | 366 | |
e4516b20 | 367 | error = open_dpif(&dp, dpifp); |
b2ae4069 | 368 | dpif_netlink_set_features(*dpifp, OVS_DP_F_TC_RECIRC_SHARING); |
8f4a4df5 | 369 | ofpbuf_delete(buf); |
b2ae4069 | 370 | |
e4516b20 | 371 | return error; |
c19e6535 BP |
372 | } |
373 | ||
e4516b20 | 374 | static int |
93451a0a | 375 | open_dpif(const struct dpif_netlink_dp *dp, struct dpif **dpifp) |
c19e6535 | 376 | { |
93451a0a | 377 | struct dpif_netlink *dpif; |
c19e6535 | 378 | |
17411ecf | 379 | dpif = xzalloc(sizeof *dpif); |
e4516b20 | 380 | dpif->port_notifier = NULL; |
1579cf67 | 381 | fat_rwlock_init(&dpif->upcall_lock); |
c19e6535 | 382 | |
93451a0a | 383 | dpif_init(&dpif->dpif, &dpif_netlink_class, dp->name, |
254f2dc8 | 384 | dp->dp_ifindex, dp->dp_ifindex); |
c19e6535 | 385 | |
254f2dc8 | 386 | dpif->dp_ifindex = dp->dp_ifindex; |
dcdcad68 | 387 | dpif->user_features = dp->user_features; |
c19e6535 | 388 | *dpifp = &dpif->dpif; |
e4516b20 BP |
389 | |
390 | return 0; | |
96fba48f BP |
391 | } |
392 | ||
09cac43f NR |
393 | #ifdef _WIN32 |
394 | static void | |
395 | vport_delete_sock_pool(struct dpif_handler *handler) | |
396 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
397 | { | |
398 | if (handler->vport_sock_pool) { | |
399 | uint32_t i; | |
400 | struct dpif_windows_vport_sock *sock_pool = | |
401 | handler->vport_sock_pool; | |
402 | ||
403 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
404 | if (sock_pool[i].nl_sock) { | |
405 | nl_sock_unsubscribe_packets(sock_pool[i].nl_sock); | |
406 | nl_sock_destroy(sock_pool[i].nl_sock); | |
407 | sock_pool[i].nl_sock = NULL; | |
408 | } | |
409 | } | |
410 | ||
411 | free(handler->vport_sock_pool); | |
412 | handler->vport_sock_pool = NULL; | |
413 | } | |
414 | } | |
415 | ||
416 | static int | |
417 | vport_create_sock_pool(struct dpif_handler *handler) | |
418 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
419 | { | |
420 | struct dpif_windows_vport_sock *sock_pool; | |
421 | size_t i; | |
422 | int error = 0; | |
423 | ||
424 | sock_pool = xzalloc(VPORT_SOCK_POOL_SIZE * sizeof *sock_pool); | |
425 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
426 | error = nl_sock_create(NETLINK_GENERIC, &sock_pool[i].nl_sock); | |
427 | if (error) { | |
428 | goto error; | |
429 | } | |
430 | ||
431 | /* Enable the netlink socket to receive packets. This is equivalent to | |
432 | * calling nl_sock_join_mcgroup() to receive events. */ | |
433 | error = nl_sock_subscribe_packets(sock_pool[i].nl_sock); | |
434 | if (error) { | |
435 | goto error; | |
436 | } | |
437 | } | |
438 | ||
439 | handler->vport_sock_pool = sock_pool; | |
440 | handler->last_used_pool_idx = 0; | |
441 | return 0; | |
442 | ||
443 | error: | |
444 | vport_delete_sock_pool(handler); | |
445 | return error; | |
446 | } | |
09cac43f NR |
447 | #endif /* _WIN32 */ |
448 | ||
69c51582 MC |
449 | /* Given the port number 'port_idx', extracts the pid of netlink socket |
450 | * associated to the port and assigns it to 'upcall_pid'. */ | |
1579cf67 | 451 | static bool |
69c51582 MC |
452 | vport_get_pid(struct dpif_netlink *dpif, uint32_t port_idx, |
453 | uint32_t *upcall_pid) | |
1579cf67 | 454 | { |
1579cf67 | 455 | /* Since the nl_sock can only be assigned in either all |
69c51582 | 456 | * or none "dpif" channels, the following check |
1579cf67 | 457 | * would suffice. */ |
69c51582 | 458 | if (!dpif->channels[port_idx].sock) { |
1579cf67 AW |
459 | return false; |
460 | } | |
09cac43f | 461 | ovs_assert(!WINDOWS || dpif->n_handlers <= 1); |
1579cf67 | 462 | |
69c51582 | 463 | *upcall_pid = nl_sock_pid(dpif->channels[port_idx].sock); |
989fd548 | 464 | |
1579cf67 | 465 | return true; |
989fd548 JP |
466 | } |
467 | ||
468 | static int | |
69c51582 | 469 | vport_add_channel(struct dpif_netlink *dpif, odp_port_t port_no, |
622ea8fd | 470 | struct nl_sock *sock) |
989fd548 JP |
471 | { |
472 | struct epoll_event event; | |
4e022ec0 | 473 | uint32_t port_idx = odp_to_u32(port_no); |
69c51582 | 474 | size_t i; |
1579cf67 | 475 | int error; |
989fd548 | 476 | |
1579cf67 | 477 | if (dpif->handlers == NULL) { |
622ea8fd | 478 | close_nl_sock(sock); |
989fd548 JP |
479 | return 0; |
480 | } | |
481 | ||
1579cf67 AW |
482 | /* We assume that the datapath densely chooses port numbers, which can |
483 | * therefore be used as an index into 'channels' and 'epoll_events' of | |
69c51582 | 484 | * 'dpif'. */ |
4e022ec0 AW |
485 | if (port_idx >= dpif->uc_array_size) { |
486 | uint32_t new_size = port_idx + 1; | |
989fd548 | 487 | |
12d76859 | 488 | if (new_size > MAX_PORTS) { |
989fd548 JP |
489 | VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big", |
490 | dpif_name(&dpif->dpif), port_no); | |
491 | return EFBIG; | |
492 | } | |
493 | ||
69c51582 MC |
494 | dpif->channels = xrealloc(dpif->channels, |
495 | new_size * sizeof *dpif->channels); | |
1579cf67 | 496 | |
69c51582 MC |
497 | for (i = dpif->uc_array_size; i < new_size; i++) { |
498 | dpif->channels[i].sock = NULL; | |
499 | } | |
1579cf67 | 500 | |
69c51582 MC |
501 | for (i = 0; i < dpif->n_handlers; i++) { |
502 | struct dpif_handler *handler = &dpif->handlers[i]; | |
1579cf67 AW |
503 | |
504 | handler->epoll_events = xrealloc(handler->epoll_events, | |
505 | new_size * sizeof *handler->epoll_events); | |
989fd548 | 506 | |
1579cf67 | 507 | } |
989fd548 JP |
508 | dpif->uc_array_size = new_size; |
509 | } | |
510 | ||
511 | memset(&event, 0, sizeof event); | |
69c51582 | 512 | event.events = EPOLLIN | EPOLLEXCLUSIVE; |
4e022ec0 | 513 | event.data.u32 = port_idx; |
989fd548 | 514 | |
1579cf67 AW |
515 | for (i = 0; i < dpif->n_handlers; i++) { |
516 | struct dpif_handler *handler = &dpif->handlers[i]; | |
517 | ||
09cac43f | 518 | #ifndef _WIN32 |
622ea8fd | 519 | if (epoll_ctl(handler->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), |
1579cf67 AW |
520 | &event) < 0) { |
521 | error = errno; | |
522 | goto error; | |
523 | } | |
93451a0a | 524 | #endif |
1579cf67 | 525 | } |
622ea8fd | 526 | dpif->channels[port_idx].sock = sock; |
69c51582 | 527 | dpif->channels[port_idx].last_poll = LLONG_MIN; |
989fd548 JP |
528 | |
529 | return 0; | |
1579cf67 AW |
530 | |
531 | error: | |
09cac43f | 532 | #ifndef _WIN32 |
69c51582 MC |
533 | while (i--) { |
534 | epoll_ctl(dpif->handlers[i].epoll_fd, EPOLL_CTL_DEL, | |
622ea8fd | 535 | nl_sock_fd(sock), NULL); |
1579cf67 | 536 | } |
69c51582 MC |
537 | #endif |
538 | dpif->channels[port_idx].sock = NULL; | |
1579cf67 AW |
539 | |
540 | return error; | |
989fd548 JP |
541 | } |
542 | ||
543 | static void | |
93451a0a | 544 | vport_del_channels(struct dpif_netlink *dpif, odp_port_t port_no) |
989fd548 | 545 | { |
4e022ec0 | 546 | uint32_t port_idx = odp_to_u32(port_no); |
1579cf67 | 547 | size_t i; |
989fd548 | 548 | |
69c51582 MC |
549 | if (!dpif->handlers || port_idx >= dpif->uc_array_size |
550 | || !dpif->channels[port_idx].sock) { | |
989fd548 JP |
551 | return; |
552 | } | |
553 | ||
1579cf67 AW |
554 | for (i = 0; i < dpif->n_handlers; i++) { |
555 | struct dpif_handler *handler = &dpif->handlers[i]; | |
09cac43f | 556 | #ifndef _WIN32 |
1579cf67 | 557 | epoll_ctl(handler->epoll_fd, EPOLL_CTL_DEL, |
69c51582 | 558 | nl_sock_fd(dpif->channels[port_idx].sock), NULL); |
09cac43f | 559 | #endif |
1579cf67 AW |
560 | handler->event_offset = handler->n_events = 0; |
561 | } | |
69c51582 MC |
562 | #ifndef _WIN32 |
563 | nl_sock_destroy(dpif->channels[port_idx].sock); | |
564 | #endif | |
565 | dpif->channels[port_idx].sock = NULL; | |
1579cf67 AW |
566 | } |
567 | ||
568 | static void | |
93451a0a AS |
569 | destroy_all_channels(struct dpif_netlink *dpif) |
570 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
1579cf67 AW |
571 | { |
572 | unsigned int i; | |
573 | ||
574 | if (!dpif->handlers) { | |
575 | return; | |
576 | } | |
577 | ||
578 | for (i = 0; i < dpif->uc_array_size; i++ ) { | |
93451a0a | 579 | struct dpif_netlink_vport vport_request; |
1579cf67 AW |
580 | uint32_t upcall_pids = 0; |
581 | ||
69c51582 | 582 | if (!dpif->channels[i].sock) { |
1579cf67 AW |
583 | continue; |
584 | } | |
585 | ||
586 | /* Turn off upcalls. */ | |
93451a0a | 587 | dpif_netlink_vport_init(&vport_request); |
1579cf67 AW |
588 | vport_request.cmd = OVS_VPORT_CMD_SET; |
589 | vport_request.dp_ifindex = dpif->dp_ifindex; | |
590 | vport_request.port_no = u32_to_odp(i); | |
a78f446a | 591 | vport_request.n_upcall_pids = 1; |
1579cf67 | 592 | vport_request.upcall_pids = &upcall_pids; |
93451a0a | 593 | dpif_netlink_vport_transact(&vport_request, NULL, NULL); |
1579cf67 AW |
594 | |
595 | vport_del_channels(dpif, u32_to_odp(i)); | |
596 | } | |
597 | ||
598 | for (i = 0; i < dpif->n_handlers; i++) { | |
599 | struct dpif_handler *handler = &dpif->handlers[i]; | |
600 | ||
09cac43f | 601 | dpif_netlink_handler_uninit(handler); |
1579cf67 | 602 | free(handler->epoll_events); |
1579cf67 | 603 | } |
69c51582 | 604 | free(dpif->channels); |
1579cf67 AW |
605 | free(dpif->handlers); |
606 | dpif->handlers = NULL; | |
69c51582 | 607 | dpif->channels = NULL; |
1579cf67 AW |
608 | dpif->n_handlers = 0; |
609 | dpif->uc_array_size = 0; | |
17411ecf JG |
610 | } |
611 | ||
96fba48f | 612 | static void |
93451a0a | 613 | dpif_netlink_close(struct dpif *dpif_) |
96fba48f | 614 | { |
93451a0a | 615 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
c7178a0b | 616 | |
e4516b20 | 617 | nl_sock_destroy(dpif->port_notifier); |
1579cf67 AW |
618 | |
619 | fat_rwlock_wrlock(&dpif->upcall_lock); | |
620 | destroy_all_channels(dpif); | |
621 | fat_rwlock_unlock(&dpif->upcall_lock); | |
622 | ||
623 | fat_rwlock_destroy(&dpif->upcall_lock); | |
96fba48f BP |
624 | free(dpif); |
625 | } | |
626 | ||
627 | static int | |
93451a0a | 628 | dpif_netlink_destroy(struct dpif *dpif_) |
96fba48f | 629 | { |
93451a0a AS |
630 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
631 | struct dpif_netlink_dp dp; | |
d6569377 | 632 | |
93451a0a | 633 | dpif_netlink_dp_init(&dp); |
df2c07f4 | 634 | dp.cmd = OVS_DP_CMD_DEL; |
254f2dc8 | 635 | dp.dp_ifindex = dpif->dp_ifindex; |
93451a0a | 636 | return dpif_netlink_dp_transact(&dp, NULL, NULL); |
96fba48f BP |
637 | } |
638 | ||
a36de779 | 639 | static bool |
93451a0a | 640 | dpif_netlink_run(struct dpif *dpif_) |
61eae437 | 641 | { |
93451a0a | 642 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1579cf67 | 643 | |
61eae437 BP |
644 | if (dpif->refresh_channels) { |
645 | dpif->refresh_channels = false; | |
1579cf67 | 646 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 647 | dpif_netlink_refresh_channels(dpif, dpif->n_handlers); |
1579cf67 | 648 | fat_rwlock_unlock(&dpif->upcall_lock); |
61eae437 | 649 | } |
a36de779 | 650 | return false; |
61eae437 BP |
651 | } |
652 | ||
96fba48f | 653 | static int |
93451a0a | 654 | dpif_netlink_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) |
96fba48f | 655 | { |
93451a0a | 656 | struct dpif_netlink_dp dp; |
d6569377 BP |
657 | struct ofpbuf *buf; |
658 | int error; | |
659 | ||
93451a0a | 660 | error = dpif_netlink_dp_get(dpif_, &dp, &buf); |
d6569377 | 661 | if (!error) { |
6a54dedc BP |
662 | memset(stats, 0, sizeof *stats); |
663 | ||
664 | if (dp.stats) { | |
665 | stats->n_hit = get_32aligned_u64(&dp.stats->n_hit); | |
666 | stats->n_missed = get_32aligned_u64(&dp.stats->n_missed); | |
667 | stats->n_lost = get_32aligned_u64(&dp.stats->n_lost); | |
668 | stats->n_flows = get_32aligned_u64(&dp.stats->n_flows); | |
669 | } | |
670 | ||
671 | if (dp.megaflow_stats) { | |
672 | stats->n_masks = dp.megaflow_stats->n_masks; | |
673 | stats->n_mask_hit = get_32aligned_u64( | |
674 | &dp.megaflow_stats->n_mask_hit); | |
675 | } else { | |
676 | stats->n_masks = UINT32_MAX; | |
677 | stats->n_mask_hit = UINT64_MAX; | |
678 | } | |
d6569377 BP |
679 | ofpbuf_delete(buf); |
680 | } | |
681 | return error; | |
96fba48f BP |
682 | } |
683 | ||
dcdcad68 PB |
684 | static int |
685 | dpif_netlink_set_features(struct dpif *dpif_, uint32_t new_features) | |
686 | { | |
687 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); | |
688 | struct dpif_netlink_dp request, reply; | |
689 | struct ofpbuf *bufp; | |
690 | int error; | |
691 | ||
692 | dpif_netlink_dp_init(&request); | |
693 | request.cmd = OVS_DP_CMD_SET; | |
25a2af4f | 694 | request.name = dpif_->base_name; |
dcdcad68 PB |
695 | request.dp_ifindex = dpif->dp_ifindex; |
696 | request.user_features = dpif->user_features | new_features; | |
697 | ||
698 | error = dpif_netlink_dp_transact(&request, &reply, &bufp); | |
699 | if (!error) { | |
700 | dpif->user_features = reply.user_features; | |
701 | ofpbuf_delete(bufp); | |
702 | if (!(dpif->user_features & new_features)) { | |
703 | return -EOPNOTSUPP; | |
704 | } | |
705 | } | |
706 | ||
707 | return error; | |
708 | } | |
709 | ||
b9ad7294 | 710 | static const char * |
93451a0a | 711 | get_vport_type(const struct dpif_netlink_vport *vport) |
b9ad7294 EJ |
712 | { |
713 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
714 | ||
715 | switch (vport->type) { | |
5ed51209 JS |
716 | case OVS_VPORT_TYPE_NETDEV: { |
717 | const char *type = netdev_get_type_from_name(vport->name); | |
718 | ||
719 | return type ? type : "system"; | |
720 | } | |
b9ad7294 EJ |
721 | |
722 | case OVS_VPORT_TYPE_INTERNAL: | |
723 | return "internal"; | |
724 | ||
c1fc1411 JG |
725 | case OVS_VPORT_TYPE_GENEVE: |
726 | return "geneve"; | |
727 | ||
b9ad7294 EJ |
728 | case OVS_VPORT_TYPE_GRE: |
729 | return "gre"; | |
730 | ||
b9ad7294 EJ |
731 | case OVS_VPORT_TYPE_VXLAN: |
732 | return "vxlan"; | |
733 | ||
a6ae068b LJ |
734 | case OVS_VPORT_TYPE_LISP: |
735 | return "lisp"; | |
736 | ||
4237026e PS |
737 | case OVS_VPORT_TYPE_STT: |
738 | return "stt"; | |
739 | ||
c387d817 | 740 | case OVS_VPORT_TYPE_ERSPAN: |
98514eea WT |
741 | return "erspan"; |
742 | ||
c387d817 | 743 | case OVS_VPORT_TYPE_IP6ERSPAN: |
3b10ceee GR |
744 | return "ip6erspan"; |
745 | ||
c387d817 | 746 | case OVS_VPORT_TYPE_IP6GRE: |
3b10ceee | 747 | return "ip6gre"; |
c387d817 | 748 | |
3c6d05a0 WT |
749 | case OVS_VPORT_TYPE_GTPU: |
750 | return "gtpu"; | |
751 | ||
ebe0e518 MV |
752 | case OVS_VPORT_TYPE_BAREUDP: |
753 | return "bareudp"; | |
754 | ||
b9ad7294 EJ |
755 | case OVS_VPORT_TYPE_UNSPEC: |
756 | case __OVS_VPORT_TYPE_MAX: | |
757 | break; | |
758 | } | |
759 | ||
760 | VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", | |
761 | vport->dp_ifindex, vport->name, (unsigned int) vport->type); | |
762 | return "unknown"; | |
763 | } | |
764 | ||
c4e08753 | 765 | enum ovs_vport_type |
20c57607 | 766 | netdev_to_ovs_vport_type(const char *type) |
c060c4cf | 767 | { |
c060c4cf EJ |
768 | if (!strcmp(type, "tap") || !strcmp(type, "system")) { |
769 | return OVS_VPORT_TYPE_NETDEV; | |
770 | } else if (!strcmp(type, "internal")) { | |
771 | return OVS_VPORT_TYPE_INTERNAL; | |
4237026e PS |
772 | } else if (strstr(type, "stt")) { |
773 | return OVS_VPORT_TYPE_STT; | |
c1fc1411 JG |
774 | } else if (!strcmp(type, "geneve")) { |
775 | return OVS_VPORT_TYPE_GENEVE; | |
c060c4cf EJ |
776 | } else if (!strcmp(type, "vxlan")) { |
777 | return OVS_VPORT_TYPE_VXLAN; | |
a6ae068b LJ |
778 | } else if (!strcmp(type, "lisp")) { |
779 | return OVS_VPORT_TYPE_LISP; | |
7dc18ae9 WT |
780 | } else if (!strcmp(type, "erspan")) { |
781 | return OVS_VPORT_TYPE_ERSPAN; | |
782 | } else if (!strcmp(type, "ip6erspan")) { | |
783 | return OVS_VPORT_TYPE_IP6ERSPAN; | |
3b10ceee GR |
784 | } else if (!strcmp(type, "ip6gre")) { |
785 | return OVS_VPORT_TYPE_IP6GRE; | |
1c385f49 GR |
786 | } else if (!strcmp(type, "gre")) { |
787 | return OVS_VPORT_TYPE_GRE; | |
3c6d05a0 WT |
788 | } else if (!strcmp(type, "gtpu")) { |
789 | return OVS_VPORT_TYPE_GTPU; | |
ebe0e518 MV |
790 | } else if (!strcmp(type, "bareudp")) { |
791 | return OVS_VPORT_TYPE_BAREUDP; | |
c060c4cf EJ |
792 | } else { |
793 | return OVS_VPORT_TYPE_UNSPEC; | |
794 | } | |
795 | } | |
796 | ||
96fba48f | 797 | static int |
20c57607 EG |
798 | dpif_netlink_port_add__(struct dpif_netlink *dpif, const char *name, |
799 | enum ovs_vport_type type, | |
800 | struct ofpbuf *options, | |
93451a0a | 801 | odp_port_t *port_nop) |
b90de034 | 802 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 803 | { |
93451a0a | 804 | struct dpif_netlink_vport request, reply; |
c19e6535 | 805 | struct ofpbuf *buf; |
622ea8fd | 806 | struct nl_sock *sock = NULL; |
790a4372 | 807 | uint32_t upcall_pids = 0; |
1579cf67 | 808 | int error = 0; |
96fba48f | 809 | |
1579cf67 | 810 | if (dpif->handlers) { |
622ea8fd | 811 | error = create_nl_sock(dpif, &sock); |
713a45db | 812 | if (error) { |
989fd548 JP |
813 | return error; |
814 | } | |
815 | } | |
816 | ||
93451a0a | 817 | dpif_netlink_vport_init(&request); |
df2c07f4 | 818 | request.cmd = OVS_VPORT_CMD_NEW; |
254f2dc8 | 819 | request.dp_ifindex = dpif->dp_ifindex; |
20c57607 EG |
820 | request.type = type; |
821 | request.name = name; | |
822 | ||
823 | request.port_no = *port_nop; | |
622ea8fd BP |
824 | if (sock) { |
825 | upcall_pids = nl_sock_pid(sock); | |
790a4372 | 826 | } |
69c51582 MC |
827 | request.n_upcall_pids = 1; |
828 | request.upcall_pids = &upcall_pids; | |
20c57607 EG |
829 | |
830 | if (options) { | |
831 | request.options = options->data; | |
832 | request.options_len = options->size; | |
833 | } | |
834 | ||
835 | error = dpif_netlink_vport_transact(&request, &reply, &buf); | |
836 | if (!error) { | |
837 | *port_nop = reply.port_no; | |
838 | } else { | |
839 | if (error == EBUSY && *port_nop != ODPP_NONE) { | |
840 | VLOG_INFO("%s: requested port %"PRIu32" is in use", | |
841 | dpif_name(&dpif->dpif), *port_nop); | |
842 | } | |
843 | ||
622ea8fd | 844 | close_nl_sock(sock); |
20c57607 EG |
845 | goto exit; |
846 | } | |
847 | ||
622ea8fd | 848 | error = vport_add_channel(dpif, *port_nop, sock); |
69c51582 MC |
849 | if (error) { |
850 | VLOG_INFO("%s: could not add channel for port %s", | |
851 | dpif_name(&dpif->dpif), name); | |
852 | ||
853 | /* Delete the port. */ | |
854 | dpif_netlink_vport_init(&request); | |
855 | request.cmd = OVS_VPORT_CMD_DEL; | |
856 | request.dp_ifindex = dpif->dp_ifindex; | |
857 | request.port_no = *port_nop; | |
858 | dpif_netlink_vport_transact(&request, NULL, NULL); | |
622ea8fd | 859 | close_nl_sock(sock); |
69c51582 | 860 | goto exit; |
20c57607 | 861 | } |
20c57607 EG |
862 | |
863 | exit: | |
864 | ofpbuf_delete(buf); | |
20c57607 EG |
865 | |
866 | return error; | |
867 | } | |
868 | ||
869 | static int | |
870 | dpif_netlink_port_add_compat(struct dpif_netlink *dpif, struct netdev *netdev, | |
871 | odp_port_t *port_nop) | |
872 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
873 | { | |
874 | const struct netdev_tunnel_config *tnl_cfg; | |
875 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; | |
876 | const char *type = netdev_get_type(netdev); | |
877 | uint64_t options_stub[64 / 8]; | |
878 | enum ovs_vport_type ovs_type; | |
879 | struct ofpbuf options; | |
880 | const char *name; | |
881 | ||
882 | name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); | |
883 | ||
884 | ovs_type = netdev_to_ovs_vport_type(netdev_get_type(netdev)); | |
885 | if (ovs_type == OVS_VPORT_TYPE_UNSPEC) { | |
c283069c BP |
886 | VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has " |
887 | "unsupported type `%s'", | |
9b00386b | 888 | dpif_name(&dpif->dpif), name, type); |
c283069c BP |
889 | return EINVAL; |
890 | } | |
c3827f61 | 891 | |
20c57607 | 892 | if (ovs_type == OVS_VPORT_TYPE_NETDEV) { |
93451a0a | 893 | #ifdef _WIN32 |
09cac43f | 894 | /* XXX : Map appropiate Windows handle */ |
93451a0a | 895 | #else |
24b019f8 | 896 | netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false); |
93451a0a | 897 | #endif |
24b019f8 JP |
898 | } |
899 | ||
da467899 | 900 | #ifdef _WIN32 |
20c57607 | 901 | if (ovs_type == OVS_VPORT_TYPE_INTERNAL) { |
da467899 AS |
902 | if (!create_wmi_port(name)){ |
903 | VLOG_ERR("Could not create wmi internal port with name:%s", name); | |
da467899 AS |
904 | return EINVAL; |
905 | }; | |
906 | } | |
907 | #endif | |
908 | ||
26508d9a | 909 | tnl_cfg = netdev_get_tunnel_config(netdev); |
526df7d8 | 910 | if (tnl_cfg && (tnl_cfg->dst_port != 0 || tnl_cfg->exts)) { |
26508d9a | 911 | ofpbuf_use_stack(&options, options_stub, sizeof options_stub); |
526df7d8 TG |
912 | if (tnl_cfg->dst_port) { |
913 | nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT, | |
914 | ntohs(tnl_cfg->dst_port)); | |
915 | } | |
916 | if (tnl_cfg->exts) { | |
917 | size_t ext_ofs; | |
918 | int i; | |
919 | ||
920 | ext_ofs = nl_msg_start_nested(&options, OVS_TUNNEL_ATTR_EXTENSION); | |
921 | for (i = 0; i < 32; i++) { | |
922 | if (tnl_cfg->exts & (1 << i)) { | |
923 | nl_msg_put_flag(&options, i); | |
924 | } | |
925 | } | |
926 | nl_msg_end_nested(&options, ext_ofs); | |
927 | } | |
20c57607 EG |
928 | return dpif_netlink_port_add__(dpif, name, ovs_type, &options, |
929 | port_nop); | |
2510ba7c | 930 | } else { |
20c57607 | 931 | return dpif_netlink_port_add__(dpif, name, ovs_type, NULL, port_nop); |
78a2d59c | 932 | } |
c3827f61 | 933 | |
20c57607 | 934 | } |
989fd548 | 935 | |
921c370a | 936 | static int |
c4e08753 EG |
937 | dpif_netlink_rtnl_port_create_and_add(struct dpif_netlink *dpif, |
938 | struct netdev *netdev, | |
939 | odp_port_t *port_nop) | |
940 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
941 | { | |
942 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
943 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; | |
944 | const char *name; | |
945 | int error; | |
989fd548 | 946 | |
c4e08753 EG |
947 | error = dpif_netlink_rtnl_port_create(netdev); |
948 | if (error) { | |
949 | if (error != EOPNOTSUPP) { | |
d52ef4eb | 950 | VLOG_WARN_RL(&rl, "Failed to create %s with rtnetlink: %s", |
c4e08753 EG |
951 | netdev_get_name(netdev), ovs_strerror(error)); |
952 | } | |
953 | return error; | |
954 | } | |
1579cf67 | 955 | |
c4e08753 EG |
956 | name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
957 | error = dpif_netlink_port_add__(dpif, name, OVS_VPORT_TYPE_NETDEV, NULL, | |
958 | port_nop); | |
c37cb3ee | 959 | if (error) { |
c4e08753 EG |
960 | dpif_netlink_rtnl_port_destroy(name, netdev_get_type(netdev)); |
961 | } | |
962 | return error; | |
963 | } | |
96fba48f BP |
964 | |
965 | static int | |
93451a0a AS |
966 | dpif_netlink_port_add(struct dpif *dpif_, struct netdev *netdev, |
967 | odp_port_t *port_nop) | |
9fafa796 | 968 | { |
93451a0a | 969 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
921c370a | 970 | int error = EOPNOTSUPP; |
9fafa796 | 971 | |
1579cf67 | 972 | fat_rwlock_wrlock(&dpif->upcall_lock); |
921c370a EG |
973 | if (!ovs_tunnels_out_of_tree) { |
974 | error = dpif_netlink_rtnl_port_create_and_add(dpif, netdev, port_nop); | |
975 | } | |
c37cb3ee | 976 | if (error) { |
921c370a EG |
977 | error = dpif_netlink_port_add_compat(dpif, netdev, port_nop); |
978 | } | |
1579cf67 | 979 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
980 | |
981 | return error; | |
982 | } | |
983 | ||
984 | static int | |
93451a0a | 985 | dpif_netlink_port_del__(struct dpif_netlink *dpif, odp_port_t port_no) |
b90de034 | 986 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 987 | { |
93451a0a | 988 | struct dpif_netlink_vport vport; |
921c370a | 989 | struct dpif_port dpif_port; |
773cd538 | 990 | int error; |
c19e6535 | 991 | |
921c370a EG |
992 | error = dpif_netlink_port_query__(dpif, port_no, NULL, &dpif_port); |
993 | if (error) { | |
994 | return error; | |
995 | } | |
996 | ||
93451a0a | 997 | dpif_netlink_vport_init(&vport); |
df2c07f4 | 998 | vport.cmd = OVS_VPORT_CMD_DEL; |
254f2dc8 | 999 | vport.dp_ifindex = dpif->dp_ifindex; |
c19e6535 | 1000 | vport.port_no = port_no; |
da467899 | 1001 | #ifdef _WIN32 |
921c370a EG |
1002 | if (!strcmp(dpif_port.type, "internal")) { |
1003 | if (!delete_wmi_port(dpif_port.name)) { | |
da467899 | 1004 | VLOG_ERR("Could not delete wmi port with name: %s", |
921c370a | 1005 | dpif_port.name); |
da467899 AS |
1006 | }; |
1007 | } | |
1008 | #endif | |
93451a0a | 1009 | error = dpif_netlink_vport_transact(&vport, NULL, NULL); |
773cd538 | 1010 | |
1579cf67 | 1011 | vport_del_channels(dpif, port_no); |
989fd548 | 1012 | |
921c370a EG |
1013 | if (!error && !ovs_tunnels_out_of_tree) { |
1014 | error = dpif_netlink_rtnl_port_destroy(dpif_port.name, dpif_port.type); | |
1015 | if (error == EOPNOTSUPP) { | |
1016 | error = 0; | |
1017 | } | |
1018 | } | |
1019 | ||
1020 | dpif_port_destroy(&dpif_port); | |
1021 | ||
773cd538 | 1022 | return error; |
c3827f61 | 1023 | } |
3abc4a1a | 1024 | |
9fafa796 | 1025 | static int |
93451a0a | 1026 | dpif_netlink_port_del(struct dpif *dpif_, odp_port_t port_no) |
9fafa796 | 1027 | { |
93451a0a | 1028 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
1029 | int error; |
1030 | ||
1579cf67 | 1031 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 1032 | error = dpif_netlink_port_del__(dpif, port_no); |
1579cf67 | 1033 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
1034 | |
1035 | return error; | |
1036 | } | |
1037 | ||
c3827f61 | 1038 | static int |
93451a0a AS |
1039 | dpif_netlink_port_query__(const struct dpif_netlink *dpif, odp_port_t port_no, |
1040 | const char *port_name, struct dpif_port *dpif_port) | |
c3827f61 | 1041 | { |
93451a0a AS |
1042 | struct dpif_netlink_vport request; |
1043 | struct dpif_netlink_vport reply; | |
c19e6535 | 1044 | struct ofpbuf *buf; |
4c738a8d BP |
1045 | int error; |
1046 | ||
93451a0a | 1047 | dpif_netlink_vport_init(&request); |
df2c07f4 | 1048 | request.cmd = OVS_VPORT_CMD_GET; |
9b00386b | 1049 | request.dp_ifindex = dpif->dp_ifindex; |
c19e6535 BP |
1050 | request.port_no = port_no; |
1051 | request.name = port_name; | |
4c738a8d | 1052 | |
93451a0a | 1053 | error = dpif_netlink_vport_transact(&request, &reply, &buf); |
c19e6535 | 1054 | if (!error) { |
33db1592 BP |
1055 | if (reply.dp_ifindex != request.dp_ifindex) { |
1056 | /* A query by name reported that 'port_name' is in some datapath | |
1057 | * other than 'dpif', but the caller wants to know about 'dpif'. */ | |
1058 | error = ENODEV; | |
4afba28d | 1059 | } else if (dpif_port) { |
33db1592 | 1060 | dpif_port->name = xstrdup(reply.name); |
b9ad7294 | 1061 | dpif_port->type = xstrdup(get_vport_type(&reply)); |
33db1592 BP |
1062 | dpif_port->port_no = reply.port_no; |
1063 | } | |
c19e6535 | 1064 | ofpbuf_delete(buf); |
3abc4a1a | 1065 | } |
c19e6535 | 1066 | return error; |
96fba48f BP |
1067 | } |
1068 | ||
1069 | static int | |
93451a0a AS |
1070 | dpif_netlink_port_query_by_number(const struct dpif *dpif_, odp_port_t port_no, |
1071 | struct dpif_port *dpif_port) | |
96fba48f | 1072 | { |
93451a0a | 1073 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9b00386b | 1074 | |
93451a0a | 1075 | return dpif_netlink_port_query__(dpif, port_no, NULL, dpif_port); |
96fba48f BP |
1076 | } |
1077 | ||
1078 | static int | |
93451a0a | 1079 | dpif_netlink_port_query_by_name(const struct dpif *dpif_, const char *devname, |
4c738a8d | 1080 | struct dpif_port *dpif_port) |
96fba48f | 1081 | { |
93451a0a | 1082 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9b00386b | 1083 | |
93451a0a | 1084 | return dpif_netlink_port_query__(dpif, 0, devname, dpif_port); |
96fba48f BP |
1085 | } |
1086 | ||
98403001 | 1087 | static uint32_t |
93451a0a | 1088 | dpif_netlink_port_get_pid__(const struct dpif_netlink *dpif, |
769b5034 | 1089 | odp_port_t port_no) |
b90de034 | 1090 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
98403001 | 1091 | { |
4e022ec0 | 1092 | uint32_t port_idx = odp_to_u32(port_no); |
9fafa796 | 1093 | uint32_t pid = 0; |
98403001 | 1094 | |
f8fc5489 | 1095 | if (dpif->handlers && dpif->uc_array_size > 0) { |
4e022ec0 | 1096 | /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s |
989fd548 | 1097 | * channel, since it is not heavily loaded. */ |
4e022ec0 | 1098 | uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx; |
1579cf67 | 1099 | |
17f2748d AW |
1100 | /* Needs to check in case the socket pointer is changed in between |
1101 | * the holding of upcall_lock. A known case happens when the main | |
1102 | * thread deletes the vport while the handler thread is handling | |
1103 | * the upcall from that port. */ | |
69c51582 MC |
1104 | if (dpif->channels[idx].sock) { |
1105 | pid = nl_sock_pid(dpif->channels[idx].sock); | |
17f2748d | 1106 | } |
98403001 | 1107 | } |
9fafa796 BP |
1108 | |
1109 | return pid; | |
98403001 BP |
1110 | } |
1111 | ||
b90de034 | 1112 | static uint32_t |
769b5034 | 1113 | dpif_netlink_port_get_pid(const struct dpif *dpif_, odp_port_t port_no) |
b90de034 | 1114 | { |
93451a0a | 1115 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
b90de034 AW |
1116 | uint32_t ret; |
1117 | ||
1118 | fat_rwlock_rdlock(&dpif->upcall_lock); | |
769b5034 | 1119 | ret = dpif_netlink_port_get_pid__(dpif, port_no); |
b90de034 AW |
1120 | fat_rwlock_unlock(&dpif->upcall_lock); |
1121 | ||
1122 | return ret; | |
1123 | } | |
1124 | ||
96fba48f | 1125 | static int |
93451a0a | 1126 | dpif_netlink_flow_flush(struct dpif *dpif_) |
96fba48f | 1127 | { |
8842fdf1 | 1128 | const char *dpif_type_str = dpif_normalize_type(dpif_type(dpif_)); |
93451a0a AS |
1129 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1130 | struct dpif_netlink_flow flow; | |
37a1300c | 1131 | |
93451a0a | 1132 | dpif_netlink_flow_init(&flow); |
df2c07f4 | 1133 | flow.cmd = OVS_FLOW_CMD_DEL; |
254f2dc8 | 1134 | flow.dp_ifindex = dpif->dp_ifindex; |
f7dde6df PB |
1135 | |
1136 | if (netdev_is_flow_api_enabled()) { | |
8842fdf1 | 1137 | netdev_ports_flow_flush(dpif_type_str); |
f7dde6df PB |
1138 | } |
1139 | ||
93451a0a | 1140 | return dpif_netlink_flow_transact(&flow, NULL, NULL); |
96fba48f BP |
1141 | } |
1142 | ||
93451a0a | 1143 | struct dpif_netlink_port_state { |
f0fef760 | 1144 | struct nl_dump dump; |
d57695d7 | 1145 | struct ofpbuf buf; |
c19e6535 BP |
1146 | }; |
1147 | ||
222837c4 | 1148 | static void |
93451a0a AS |
1149 | dpif_netlink_port_dump_start__(const struct dpif_netlink *dpif, |
1150 | struct nl_dump *dump) | |
96fba48f | 1151 | { |
93451a0a | 1152 | struct dpif_netlink_vport request; |
f0fef760 BP |
1153 | struct ofpbuf *buf; |
1154 | ||
93451a0a | 1155 | dpif_netlink_vport_init(&request); |
067f1e23 | 1156 | request.cmd = OVS_VPORT_CMD_GET; |
254f2dc8 | 1157 | request.dp_ifindex = dpif->dp_ifindex; |
f0fef760 BP |
1158 | |
1159 | buf = ofpbuf_new(1024); | |
93451a0a | 1160 | dpif_netlink_vport_to_ofpbuf(&request, buf); |
222837c4 | 1161 | nl_dump_start(dump, NETLINK_GENERIC, buf); |
f0fef760 | 1162 | ofpbuf_delete(buf); |
222837c4 BP |
1163 | } |
1164 | ||
1165 | static int | |
93451a0a | 1166 | dpif_netlink_port_dump_start(const struct dpif *dpif_, void **statep) |
222837c4 | 1167 | { |
93451a0a AS |
1168 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1169 | struct dpif_netlink_port_state *state; | |
222837c4 BP |
1170 | |
1171 | *statep = state = xmalloc(sizeof *state); | |
93451a0a | 1172 | dpif_netlink_port_dump_start__(dpif, &state->dump); |
f0fef760 | 1173 | |
d57695d7 | 1174 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); |
b0ec0f27 BP |
1175 | return 0; |
1176 | } | |
1177 | ||
7c1ef244 | 1178 | static int |
93451a0a AS |
1179 | dpif_netlink_port_dump_next__(const struct dpif_netlink *dpif, |
1180 | struct nl_dump *dump, | |
1181 | struct dpif_netlink_vport *vport, | |
1182 | struct ofpbuf *buffer) | |
222837c4 | 1183 | { |
222837c4 BP |
1184 | struct ofpbuf buf; |
1185 | int error; | |
1186 | ||
d57695d7 | 1187 | if (!nl_dump_next(dump, &buf, buffer)) { |
222837c4 BP |
1188 | return EOF; |
1189 | } | |
1190 | ||
93451a0a | 1191 | error = dpif_netlink_vport_from_ofpbuf(vport, &buf); |
222837c4 BP |
1192 | if (error) { |
1193 | VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)", | |
1194 | dpif_name(&dpif->dpif), ovs_strerror(error)); | |
1195 | } | |
1196 | return error; | |
1197 | } | |
1198 | ||
b0ec0f27 | 1199 | static int |
93451a0a AS |
1200 | dpif_netlink_port_dump_next(const struct dpif *dpif_, void *state_, |
1201 | struct dpif_port *dpif_port) | |
b0ec0f27 | 1202 | { |
93451a0a AS |
1203 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1204 | struct dpif_netlink_port_state *state = state_; | |
1205 | struct dpif_netlink_vport vport; | |
96fba48f BP |
1206 | int error; |
1207 | ||
93451a0a AS |
1208 | error = dpif_netlink_port_dump_next__(dpif, &state->dump, &vport, |
1209 | &state->buf); | |
c3827f61 | 1210 | if (error) { |
f0fef760 | 1211 | return error; |
c3827f61 | 1212 | } |
ebc56baa | 1213 | dpif_port->name = CONST_CAST(char *, vport.name); |
b9ad7294 | 1214 | dpif_port->type = CONST_CAST(char *, get_vport_type(&vport)); |
f0fef760 BP |
1215 | dpif_port->port_no = vport.port_no; |
1216 | return 0; | |
b0ec0f27 BP |
1217 | } |
1218 | ||
1219 | static int | |
93451a0a | 1220 | dpif_netlink_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_) |
b0ec0f27 | 1221 | { |
93451a0a | 1222 | struct dpif_netlink_port_state *state = state_; |
f0fef760 | 1223 | int error = nl_dump_done(&state->dump); |
8522b383 | 1224 | |
d57695d7 | 1225 | ofpbuf_uninit(&state->buf); |
b0ec0f27 | 1226 | free(state); |
f0fef760 | 1227 | return error; |
96fba48f BP |
1228 | } |
1229 | ||
e9e28be3 | 1230 | static int |
93451a0a | 1231 | dpif_netlink_port_poll(const struct dpif *dpif_, char **devnamep) |
e9e28be3 | 1232 | { |
93451a0a | 1233 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
e9e28be3 | 1234 | |
e4516b20 BP |
1235 | /* Lazily create the Netlink socket to listen for notifications. */ |
1236 | if (!dpif->port_notifier) { | |
1237 | struct nl_sock *sock; | |
1238 | int error; | |
1239 | ||
1240 | error = nl_sock_create(NETLINK_GENERIC, &sock); | |
1241 | if (error) { | |
1242 | return error; | |
1243 | } | |
1244 | ||
1245 | error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup); | |
1246 | if (error) { | |
1247 | nl_sock_destroy(sock); | |
1248 | return error; | |
1249 | } | |
1250 | dpif->port_notifier = sock; | |
1251 | ||
1252 | /* We have no idea of the current state so report that everything | |
1253 | * changed. */ | |
1254 | return ENOBUFS; | |
1255 | } | |
1256 | ||
1257 | for (;;) { | |
1258 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1259 | uint64_t buf_stub[4096 / 8]; | |
1260 | struct ofpbuf buf; | |
1261 | int error; | |
1262 | ||
1263 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
a86bd14e | 1264 | error = nl_sock_recv(dpif->port_notifier, &buf, NULL, false); |
e4516b20 | 1265 | if (!error) { |
93451a0a | 1266 | struct dpif_netlink_vport vport; |
e4516b20 | 1267 | |
93451a0a | 1268 | error = dpif_netlink_vport_from_ofpbuf(&vport, &buf); |
e4516b20 BP |
1269 | if (!error) { |
1270 | if (vport.dp_ifindex == dpif->dp_ifindex | |
1271 | && (vport.cmd == OVS_VPORT_CMD_NEW | |
1272 | || vport.cmd == OVS_VPORT_CMD_DEL | |
1273 | || vport.cmd == OVS_VPORT_CMD_SET)) { | |
1274 | VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8, | |
1275 | dpif->dpif.full_name, vport.name, vport.cmd); | |
1579cf67 | 1276 | if (vport.cmd == OVS_VPORT_CMD_DEL && dpif->handlers) { |
61eae437 BP |
1277 | dpif->refresh_channels = true; |
1278 | } | |
e4516b20 | 1279 | *devnamep = xstrdup(vport.name); |
59e0c910 | 1280 | ofpbuf_uninit(&buf); |
e4516b20 | 1281 | return 0; |
e4516b20 BP |
1282 | } |
1283 | } | |
59e0c910 BP |
1284 | } else if (error != EAGAIN) { |
1285 | VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", | |
1286 | ovs_strerror(error)); | |
1287 | nl_sock_drain(dpif->port_notifier); | |
1288 | error = ENOBUFS; | |
e4516b20 BP |
1289 | } |
1290 | ||
59e0c910 BP |
1291 | ofpbuf_uninit(&buf); |
1292 | if (error) { | |
1293 | return error; | |
1294 | } | |
e9e28be3 | 1295 | } |
e9e28be3 BP |
1296 | } |
1297 | ||
1298 | static void | |
93451a0a | 1299 | dpif_netlink_port_poll_wait(const struct dpif *dpif_) |
e9e28be3 | 1300 | { |
93451a0a | 1301 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
e4516b20 BP |
1302 | |
1303 | if (dpif->port_notifier) { | |
1304 | nl_sock_wait(dpif->port_notifier, POLLIN); | |
1305 | } else { | |
e9e28be3 | 1306 | poll_immediate_wake(); |
e9e28be3 BP |
1307 | } |
1308 | } | |
1309 | ||
6fe09f8c | 1310 | static void |
70e5ed6f JS |
1311 | dpif_netlink_flow_init_ufid(struct dpif_netlink_flow *request, |
1312 | const ovs_u128 *ufid, bool terse) | |
1313 | { | |
1314 | if (ufid) { | |
1315 | request->ufid = *ufid; | |
1316 | request->ufid_present = true; | |
1317 | } else { | |
1318 | request->ufid_present = false; | |
1319 | } | |
1320 | request->ufid_terse = terse; | |
1321 | } | |
1322 | ||
1323 | static void | |
1324 | dpif_netlink_init_flow_get__(const struct dpif_netlink *dpif, | |
1325 | const struct nlattr *key, size_t key_len, | |
1326 | const ovs_u128 *ufid, bool terse, | |
1327 | struct dpif_netlink_flow *request) | |
96fba48f | 1328 | { |
93451a0a | 1329 | dpif_netlink_flow_init(request); |
6fe09f8c JS |
1330 | request->cmd = OVS_FLOW_CMD_GET; |
1331 | request->dp_ifindex = dpif->dp_ifindex; | |
1332 | request->key = key; | |
1333 | request->key_len = key_len; | |
70e5ed6f JS |
1334 | dpif_netlink_flow_init_ufid(request, ufid, terse); |
1335 | } | |
1336 | ||
1337 | static void | |
1338 | dpif_netlink_init_flow_get(const struct dpif_netlink *dpif, | |
1339 | const struct dpif_flow_get *get, | |
1340 | struct dpif_netlink_flow *request) | |
1341 | { | |
1342 | dpif_netlink_init_flow_get__(dpif, get->key, get->key_len, get->ufid, | |
1343 | false, request); | |
30053024 BP |
1344 | } |
1345 | ||
1346 | static int | |
70e5ed6f JS |
1347 | dpif_netlink_flow_get__(const struct dpif_netlink *dpif, |
1348 | const struct nlattr *key, size_t key_len, | |
1349 | const ovs_u128 *ufid, bool terse, | |
1350 | struct dpif_netlink_flow *reply, struct ofpbuf **bufp) | |
30053024 | 1351 | { |
93451a0a | 1352 | struct dpif_netlink_flow request; |
30053024 | 1353 | |
70e5ed6f | 1354 | dpif_netlink_init_flow_get__(dpif, key, key_len, ufid, terse, &request); |
93451a0a | 1355 | return dpif_netlink_flow_transact(&request, reply, bufp); |
96fba48f BP |
1356 | } |
1357 | ||
70e5ed6f JS |
1358 | static int |
1359 | dpif_netlink_flow_get(const struct dpif_netlink *dpif, | |
1360 | const struct dpif_netlink_flow *flow, | |
1361 | struct dpif_netlink_flow *reply, struct ofpbuf **bufp) | |
1362 | { | |
1363 | return dpif_netlink_flow_get__(dpif, flow->key, flow->key_len, | |
1364 | flow->ufid_present ? &flow->ufid : NULL, | |
1365 | false, reply, bufp); | |
1366 | } | |
1367 | ||
6bc60024 | 1368 | static void |
93451a0a AS |
1369 | dpif_netlink_init_flow_put(struct dpif_netlink *dpif, |
1370 | const struct dpif_flow_put *put, | |
1371 | struct dpif_netlink_flow *request) | |
6bc60024 | 1372 | { |
d64e176c | 1373 | static const struct nlattr dummy_action; |
6bc60024 | 1374 | |
93451a0a | 1375 | dpif_netlink_flow_init(request); |
89625d1e | 1376 | request->cmd = (put->flags & DPIF_FP_CREATE |
6bc60024 BP |
1377 | ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET); |
1378 | request->dp_ifindex = dpif->dp_ifindex; | |
89625d1e BP |
1379 | request->key = put->key; |
1380 | request->key_len = put->key_len; | |
e6cc0bab AZ |
1381 | request->mask = put->mask; |
1382 | request->mask_len = put->mask_len; | |
70e5ed6f JS |
1383 | dpif_netlink_flow_init_ufid(request, put->ufid, false); |
1384 | ||
6bc60024 | 1385 | /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ |
d64e176c BP |
1386 | request->actions = (put->actions |
1387 | ? put->actions | |
1388 | : CONST_CAST(struct nlattr *, &dummy_action)); | |
89625d1e BP |
1389 | request->actions_len = put->actions_len; |
1390 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
6bc60024 BP |
1391 | request->clear = true; |
1392 | } | |
43f9ac0a JR |
1393 | if (put->flags & DPIF_FP_PROBE) { |
1394 | request->probe = true; | |
1395 | } | |
89625d1e | 1396 | request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; |
6bc60024 BP |
1397 | } |
1398 | ||
b99d3cee | 1399 | static void |
70e5ed6f JS |
1400 | dpif_netlink_init_flow_del__(struct dpif_netlink *dpif, |
1401 | const struct nlattr *key, size_t key_len, | |
1402 | const ovs_u128 *ufid, bool terse, | |
1403 | struct dpif_netlink_flow *request) | |
96fba48f | 1404 | { |
93451a0a | 1405 | dpif_netlink_flow_init(request); |
b99d3cee BP |
1406 | request->cmd = OVS_FLOW_CMD_DEL; |
1407 | request->dp_ifindex = dpif->dp_ifindex; | |
70e5ed6f JS |
1408 | request->key = key; |
1409 | request->key_len = key_len; | |
1410 | dpif_netlink_flow_init_ufid(request, ufid, terse); | |
1411 | } | |
1412 | ||
1413 | static void | |
1414 | dpif_netlink_init_flow_del(struct dpif_netlink *dpif, | |
1415 | const struct dpif_flow_del *del, | |
1416 | struct dpif_netlink_flow *request) | |
1417 | { | |
37382aa6 AS |
1418 | dpif_netlink_init_flow_del__(dpif, del->key, del->key_len, |
1419 | del->ufid, del->terse, request); | |
70e5ed6f JS |
1420 | } |
1421 | ||
93451a0a | 1422 | struct dpif_netlink_flow_dump { |
ac64794a BP |
1423 | struct dpif_flow_dump up; |
1424 | struct nl_dump nl_dump; | |
d2ad7ef1 | 1425 | atomic_int status; |
f2280b41 PB |
1426 | struct netdev_flow_dump **netdev_dumps; |
1427 | int netdev_dumps_num; /* Number of netdev_flow_dumps */ | |
1428 | struct ovs_mutex netdev_lock; /* Guards the following. */ | |
1429 | int netdev_current_dump OVS_GUARDED; /* Shared current dump */ | |
a692410a | 1430 | struct dpif_flow_dump_types types; /* Type of dump */ |
e723fd32 JS |
1431 | }; |
1432 | ||
93451a0a AS |
1433 | static struct dpif_netlink_flow_dump * |
1434 | dpif_netlink_flow_dump_cast(struct dpif_flow_dump *dump) | |
e723fd32 | 1435 | { |
93451a0a | 1436 | return CONTAINER_OF(dump, struct dpif_netlink_flow_dump, up); |
e723fd32 JS |
1437 | } |
1438 | ||
f2280b41 PB |
1439 | static void |
1440 | start_netdev_dump(const struct dpif *dpif_, | |
1441 | struct dpif_netlink_flow_dump *dump) | |
1442 | { | |
1443 | ovs_mutex_init(&dump->netdev_lock); | |
1444 | ||
a692410a | 1445 | if (!(dump->types.netdev_flows)) { |
f2280b41 PB |
1446 | dump->netdev_dumps_num = 0; |
1447 | dump->netdev_dumps = NULL; | |
1448 | return; | |
1449 | } | |
1450 | ||
1451 | ovs_mutex_lock(&dump->netdev_lock); | |
1452 | dump->netdev_current_dump = 0; | |
1453 | dump->netdev_dumps | |
8842fdf1 | 1454 | = netdev_ports_flow_dump_create(dpif_normalize_type(dpif_type(dpif_)), |
19153657 VB |
1455 | &dump->netdev_dumps_num, |
1456 | dump->up.terse); | |
f2280b41 PB |
1457 | ovs_mutex_unlock(&dump->netdev_lock); |
1458 | } | |
1459 | ||
a692410a GT |
1460 | static void |
1461 | dpif_netlink_populate_flow_dump_types(struct dpif_netlink_flow_dump *dump, | |
1462 | struct dpif_flow_dump_types *types) | |
1463 | { | |
1464 | if (!types) { | |
1465 | dump->types.ovs_flows = true; | |
1466 | dump->types.netdev_flows = true; | |
1467 | } else { | |
1468 | memcpy(&dump->types, types, sizeof *types); | |
494a7455 | 1469 | } |
7e8b7199 PB |
1470 | } |
1471 | ||
ac64794a | 1472 | static struct dpif_flow_dump * |
7e8b7199 | 1473 | dpif_netlink_flow_dump_create(const struct dpif *dpif_, bool terse, |
a692410a | 1474 | struct dpif_flow_dump_types *types) |
96fba48f | 1475 | { |
93451a0a AS |
1476 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1477 | struct dpif_netlink_flow_dump *dump; | |
1478 | struct dpif_netlink_flow request; | |
37a1300c BP |
1479 | struct ofpbuf *buf; |
1480 | ||
ac64794a BP |
1481 | dump = xmalloc(sizeof *dump); |
1482 | dpif_flow_dump_init(&dump->up, dpif_); | |
37a1300c | 1483 | |
a692410a | 1484 | dpif_netlink_populate_flow_dump_types(dump, types); |
37a1300c | 1485 | |
a692410a | 1486 | if (dump->types.ovs_flows) { |
7e8b7199 PB |
1487 | dpif_netlink_flow_init(&request); |
1488 | request.cmd = OVS_FLOW_CMD_GET; | |
1489 | request.dp_ifindex = dpif->dp_ifindex; | |
1490 | request.ufid_present = false; | |
1491 | request.ufid_terse = terse; | |
1492 | ||
1493 | buf = ofpbuf_new(1024); | |
1494 | dpif_netlink_flow_to_ofpbuf(&request, buf); | |
1495 | nl_dump_start(&dump->nl_dump, NETLINK_GENERIC, buf); | |
1496 | ofpbuf_delete(buf); | |
1497 | } | |
ac64794a | 1498 | atomic_init(&dump->status, 0); |
64bb477f | 1499 | dump->up.terse = terse; |
30053024 | 1500 | |
f2280b41 PB |
1501 | start_netdev_dump(dpif_, dump); |
1502 | ||
ac64794a | 1503 | return &dump->up; |
704a1e09 BP |
1504 | } |
1505 | ||
1506 | static int | |
93451a0a | 1507 | dpif_netlink_flow_dump_destroy(struct dpif_flow_dump *dump_) |
704a1e09 | 1508 | { |
93451a0a | 1509 | struct dpif_netlink_flow_dump *dump = dpif_netlink_flow_dump_cast(dump_); |
7e8b7199 | 1510 | unsigned int nl_status = 0; |
ac64794a | 1511 | int dump_status; |
96fba48f | 1512 | |
a692410a | 1513 | if (dump->types.ovs_flows) { |
7e8b7199 PB |
1514 | nl_status = nl_dump_done(&dump->nl_dump); |
1515 | } | |
1516 | ||
f2280b41 PB |
1517 | for (int i = 0; i < dump->netdev_dumps_num; i++) { |
1518 | int err = netdev_flow_dump_destroy(dump->netdev_dumps[i]); | |
1519 | ||
1520 | if (err != 0 && err != EOPNOTSUPP) { | |
1521 | VLOG_ERR("failed dumping netdev: %s", ovs_strerror(err)); | |
1522 | } | |
1523 | } | |
1524 | ||
1525 | free(dump->netdev_dumps); | |
1526 | ovs_mutex_destroy(&dump->netdev_lock); | |
1527 | ||
7424fc44 JR |
1528 | /* No other thread has access to 'dump' at this point. */ |
1529 | atomic_read_relaxed(&dump->status, &dump_status); | |
ac64794a BP |
1530 | free(dump); |
1531 | return dump_status ? dump_status : nl_status; | |
1532 | } | |
feebdea2 | 1533 | |
93451a0a | 1534 | struct dpif_netlink_flow_dump_thread { |
ac64794a | 1535 | struct dpif_flow_dump_thread up; |
93451a0a AS |
1536 | struct dpif_netlink_flow_dump *dump; |
1537 | struct dpif_netlink_flow flow; | |
ac64794a BP |
1538 | struct dpif_flow_stats stats; |
1539 | struct ofpbuf nl_flows; /* Always used to store flows. */ | |
1540 | struct ofpbuf *nl_actions; /* Used if kernel does not supply actions. */ | |
f2280b41 PB |
1541 | int netdev_dump_idx; /* This thread current netdev dump index */ |
1542 | bool netdev_done; /* If we are finished dumping netdevs */ | |
1543 | ||
1544 | /* (Key/Mask/Actions) Buffers for netdev dumping */ | |
1545 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; | |
1546 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
1547 | struct odputil_keybuf actbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
1548 | }; |
1549 | ||
93451a0a AS |
1550 | static struct dpif_netlink_flow_dump_thread * |
1551 | dpif_netlink_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
ac64794a | 1552 | { |
93451a0a | 1553 | return CONTAINER_OF(thread, struct dpif_netlink_flow_dump_thread, up); |
ac64794a BP |
1554 | } |
1555 | ||
1556 | static struct dpif_flow_dump_thread * | |
93451a0a | 1557 | dpif_netlink_flow_dump_thread_create(struct dpif_flow_dump *dump_) |
ac64794a | 1558 | { |
93451a0a AS |
1559 | struct dpif_netlink_flow_dump *dump = dpif_netlink_flow_dump_cast(dump_); |
1560 | struct dpif_netlink_flow_dump_thread *thread; | |
ac64794a BP |
1561 | |
1562 | thread = xmalloc(sizeof *thread); | |
1563 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
1564 | thread->dump = dump; | |
1565 | ofpbuf_init(&thread->nl_flows, NL_DUMP_BUFSIZE); | |
1566 | thread->nl_actions = NULL; | |
f2280b41 PB |
1567 | thread->netdev_dump_idx = 0; |
1568 | thread->netdev_done = !(thread->netdev_dump_idx < dump->netdev_dumps_num); | |
ac64794a BP |
1569 | |
1570 | return &thread->up; | |
1571 | } | |
1572 | ||
1573 | static void | |
93451a0a | 1574 | dpif_netlink_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) |
ac64794a | 1575 | { |
93451a0a AS |
1576 | struct dpif_netlink_flow_dump_thread *thread |
1577 | = dpif_netlink_flow_dump_thread_cast(thread_); | |
ac64794a BP |
1578 | |
1579 | ofpbuf_uninit(&thread->nl_flows); | |
1580 | ofpbuf_delete(thread->nl_actions); | |
1581 | free(thread); | |
1582 | } | |
1583 | ||
1584 | static void | |
7a5e0ee7 | 1585 | dpif_netlink_flow_to_dpif_flow(struct dpif_flow *dpif_flow, |
7fe98598 | 1586 | const struct dpif_netlink_flow *datapath_flow) |
ac64794a | 1587 | { |
7fe98598 NR |
1588 | dpif_flow->key = datapath_flow->key; |
1589 | dpif_flow->key_len = datapath_flow->key_len; | |
1590 | dpif_flow->mask = datapath_flow->mask; | |
1591 | dpif_flow->mask_len = datapath_flow->mask_len; | |
1592 | dpif_flow->actions = datapath_flow->actions; | |
1593 | dpif_flow->actions_len = datapath_flow->actions_len; | |
70e5ed6f | 1594 | dpif_flow->ufid_present = datapath_flow->ufid_present; |
ec97c2df | 1595 | dpif_flow->pmd_id = PMD_ID_NULL; |
70e5ed6f JS |
1596 | if (datapath_flow->ufid_present) { |
1597 | dpif_flow->ufid = datapath_flow->ufid; | |
1598 | } else { | |
1599 | ovs_assert(datapath_flow->key && datapath_flow->key_len); | |
7a5e0ee7 IM |
1600 | odp_flow_key_hash(datapath_flow->key, datapath_flow->key_len, |
1601 | &dpif_flow->ufid); | |
70e5ed6f | 1602 | } |
7fe98598 | 1603 | dpif_netlink_flow_get_stats(datapath_flow, &dpif_flow->stats); |
d63ca532 GT |
1604 | dpif_flow->attrs.offloaded = false; |
1605 | dpif_flow->attrs.dp_layer = "ovs"; | |
d7b55c5c | 1606 | dpif_flow->attrs.dp_extra_info = NULL; |
ac64794a BP |
1607 | } |
1608 | ||
f2280b41 PB |
1609 | /* The design is such that all threads are working together on the first dump |
1610 | * to the last, in order (at first they all on dump 0). | |
1611 | * When the first thread finds that the given dump is finished, | |
1612 | * they all move to the next. If two or more threads find the same dump | |
1613 | * is finished at the same time, the first one will advance the shared | |
1614 | * netdev_current_dump and the others will catch up. */ | |
1615 | static void | |
1616 | dpif_netlink_advance_netdev_dump(struct dpif_netlink_flow_dump_thread *thread) | |
1617 | { | |
1618 | struct dpif_netlink_flow_dump *dump = thread->dump; | |
1619 | ||
1620 | ovs_mutex_lock(&dump->netdev_lock); | |
1621 | /* if we haven't finished (dumped everything) */ | |
1622 | if (dump->netdev_current_dump < dump->netdev_dumps_num) { | |
1623 | /* if we are the first to find that current dump is finished | |
1624 | * advance it. */ | |
1625 | if (thread->netdev_dump_idx == dump->netdev_current_dump) { | |
1626 | thread->netdev_dump_idx = ++dump->netdev_current_dump; | |
1627 | /* did we just finish the last dump? done. */ | |
1628 | if (dump->netdev_current_dump == dump->netdev_dumps_num) { | |
1629 | thread->netdev_done = true; | |
1630 | } | |
1631 | } else { | |
1632 | /* otherwise, we are behind, catch up */ | |
1633 | thread->netdev_dump_idx = dump->netdev_current_dump; | |
1634 | } | |
1635 | } else { | |
1636 | /* some other thread finished */ | |
1637 | thread->netdev_done = true; | |
1638 | } | |
1639 | ovs_mutex_unlock(&dump->netdev_lock); | |
1640 | } | |
1641 | ||
1642 | static int | |
1643 | dpif_netlink_netdev_match_to_dpif_flow(struct match *match, | |
1644 | struct ofpbuf *key_buf, | |
1645 | struct ofpbuf *mask_buf, | |
1646 | struct nlattr *actions, | |
1647 | struct dpif_flow_stats *stats, | |
d63ca532 | 1648 | struct dpif_flow_attrs *attrs, |
f2280b41 PB |
1649 | ovs_u128 *ufid, |
1650 | struct dpif_flow *flow, | |
19153657 | 1651 | bool terse) |
f2280b41 | 1652 | { |
f2280b41 PB |
1653 | memset(flow, 0, sizeof *flow); |
1654 | ||
19153657 VB |
1655 | if (!terse) { |
1656 | struct odp_flow_key_parms odp_parms = { | |
1657 | .flow = &match->flow, | |
1658 | .mask = &match->wc.masks, | |
1659 | .support = { | |
1660 | .max_vlan_headers = 2, | |
1661 | .recirc = true, | |
1662 | .ct_state = true, | |
1663 | .ct_zone = true, | |
1664 | .ct_mark = true, | |
1665 | .ct_label = true, | |
1666 | }, | |
1667 | }; | |
1668 | size_t offset; | |
1669 | ||
1670 | /* Key */ | |
1671 | offset = key_buf->size; | |
1672 | flow->key = ofpbuf_tail(key_buf); | |
1673 | odp_flow_key_from_flow(&odp_parms, key_buf); | |
1674 | flow->key_len = key_buf->size - offset; | |
1675 | ||
1676 | /* Mask */ | |
1677 | offset = mask_buf->size; | |
1678 | flow->mask = ofpbuf_tail(mask_buf); | |
1679 | odp_parms.key_buf = key_buf; | |
1680 | odp_flow_key_from_mask(&odp_parms, mask_buf); | |
1681 | flow->mask_len = mask_buf->size - offset; | |
1682 | ||
1683 | /* Actions */ | |
1684 | flow->actions = nl_attr_get(actions); | |
1685 | flow->actions_len = nl_attr_get_size(actions); | |
1686 | } | |
f2280b41 PB |
1687 | |
1688 | /* Stats */ | |
1689 | memcpy(&flow->stats, stats, sizeof *stats); | |
1690 | ||
1691 | /* UFID */ | |
1692 | flow->ufid_present = true; | |
1693 | flow->ufid = *ufid; | |
1694 | ||
1695 | flow->pmd_id = PMD_ID_NULL; | |
4742003c | 1696 | |
d63ca532 | 1697 | memcpy(&flow->attrs, attrs, sizeof *attrs); |
4742003c | 1698 | |
f2280b41 PB |
1699 | return 0; |
1700 | } | |
1701 | ||
ac64794a | 1702 | static int |
93451a0a AS |
1703 | dpif_netlink_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
1704 | struct dpif_flow *flows, int max_flows) | |
ac64794a | 1705 | { |
93451a0a AS |
1706 | struct dpif_netlink_flow_dump_thread *thread |
1707 | = dpif_netlink_flow_dump_thread_cast(thread_); | |
1708 | struct dpif_netlink_flow_dump *dump = thread->dump; | |
1709 | struct dpif_netlink *dpif = dpif_netlink_cast(thread->up.dpif); | |
ac64794a BP |
1710 | int n_flows; |
1711 | ||
1712 | ofpbuf_delete(thread->nl_actions); | |
1713 | thread->nl_actions = NULL; | |
1714 | ||
1715 | n_flows = 0; | |
f2280b41 PB |
1716 | max_flows = MIN(max_flows, FLOW_DUMP_MAX_BATCH); |
1717 | ||
1718 | while (!thread->netdev_done && n_flows < max_flows) { | |
1719 | struct odputil_keybuf *maskbuf = &thread->maskbuf[n_flows]; | |
1720 | struct odputil_keybuf *keybuf = &thread->keybuf[n_flows]; | |
1721 | struct odputil_keybuf *actbuf = &thread->actbuf[n_flows]; | |
1722 | struct ofpbuf key, mask, act; | |
1723 | struct dpif_flow *f = &flows[n_flows]; | |
1724 | int cur = thread->netdev_dump_idx; | |
1725 | struct netdev_flow_dump *netdev_dump = dump->netdev_dumps[cur]; | |
1726 | struct match match; | |
1727 | struct nlattr *actions; | |
1728 | struct dpif_flow_stats stats; | |
d63ca532 | 1729 | struct dpif_flow_attrs attrs; |
f2280b41 PB |
1730 | ovs_u128 ufid; |
1731 | bool has_next; | |
1732 | ||
1733 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); | |
1734 | ofpbuf_use_stack(&act, actbuf, sizeof *actbuf); | |
1735 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
1736 | has_next = netdev_flow_dump_next(netdev_dump, &match, | |
d63ca532 | 1737 | &actions, &stats, &attrs, |
f2280b41 PB |
1738 | &ufid, |
1739 | &thread->nl_flows, | |
1740 | &act); | |
1741 | if (has_next) { | |
1742 | dpif_netlink_netdev_match_to_dpif_flow(&match, | |
1743 | &key, &mask, | |
1744 | actions, | |
1745 | &stats, | |
d63ca532 | 1746 | &attrs, |
f2280b41 PB |
1747 | &ufid, |
1748 | f, | |
1749 | dump->up.terse); | |
1750 | n_flows++; | |
1751 | } else { | |
1752 | dpif_netlink_advance_netdev_dump(thread); | |
1753 | } | |
1754 | } | |
1755 | ||
a692410a | 1756 | if (!(dump->types.ovs_flows)) { |
7e8b7199 PB |
1757 | return n_flows; |
1758 | } | |
1759 | ||
ac64794a | 1760 | while (!n_flows |
6fd6ed71 | 1761 | || (n_flows < max_flows && thread->nl_flows.size)) { |
7fe98598 | 1762 | struct dpif_netlink_flow datapath_flow; |
ac64794a BP |
1763 | struct ofpbuf nl_flow; |
1764 | int error; | |
1765 | ||
1766 | /* Try to grab another flow. */ | |
1767 | if (!nl_dump_next(&dump->nl_dump, &nl_flow, &thread->nl_flows)) { | |
1768 | break; | |
feebdea2 | 1769 | } |
30053024 | 1770 | |
ac64794a | 1771 | /* Convert the flow to our output format. */ |
7fe98598 | 1772 | error = dpif_netlink_flow_from_ofpbuf(&datapath_flow, &nl_flow); |
30053024 | 1773 | if (error) { |
7424fc44 | 1774 | atomic_store_relaxed(&dump->status, error); |
ac64794a | 1775 | break; |
feebdea2 | 1776 | } |
30053024 | 1777 | |
64bb477f JS |
1778 | if (dump->up.terse || datapath_flow.actions) { |
1779 | /* Common case: we don't want actions, or the flow includes | |
1780 | * actions. */ | |
7a5e0ee7 | 1781 | dpif_netlink_flow_to_dpif_flow(&flows[n_flows++], &datapath_flow); |
ac64794a BP |
1782 | } else { |
1783 | /* Rare case: the flow does not include actions. Retrieve this | |
1784 | * individual flow again to get the actions. */ | |
70e5ed6f | 1785 | error = dpif_netlink_flow_get(dpif, &datapath_flow, |
7fe98598 | 1786 | &datapath_flow, &thread->nl_actions); |
30053024 BP |
1787 | if (error == ENOENT) { |
1788 | VLOG_DBG("dumped flow disappeared on get"); | |
ac64794a | 1789 | continue; |
30053024 | 1790 | } else if (error) { |
10a89ef0 BP |
1791 | VLOG_WARN("error fetching dumped flow: %s", |
1792 | ovs_strerror(error)); | |
7424fc44 | 1793 | atomic_store_relaxed(&dump->status, error); |
ac64794a | 1794 | break; |
30053024 | 1795 | } |
30053024 | 1796 | |
ac64794a BP |
1797 | /* Save this flow. Then exit, because we only have one buffer to |
1798 | * handle this case. */ | |
7a5e0ee7 | 1799 | dpif_netlink_flow_to_dpif_flow(&flows[n_flows++], &datapath_flow); |
ac64794a BP |
1800 | break; |
1801 | } | |
feebdea2 | 1802 | } |
ac64794a | 1803 | return n_flows; |
96fba48f BP |
1804 | } |
1805 | ||
eabe7c68 | 1806 | static void |
93451a0a AS |
1807 | dpif_netlink_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, |
1808 | struct ofpbuf *buf) | |
96fba48f | 1809 | { |
89625d1e | 1810 | struct ovs_header *k_exec; |
758c456d | 1811 | size_t key_ofs; |
f7cd0081 | 1812 | |
eabe7c68 | 1813 | ofpbuf_prealloc_tailroom(buf, (64 |
cf62fa4c | 1814 | + dp_packet_size(d_exec->packet) |
758c456d | 1815 | + ODP_KEY_METADATA_SIZE |
eabe7c68 | 1816 | + d_exec->actions_len)); |
f7cd0081 | 1817 | |
df2c07f4 | 1818 | nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, |
69685a88 | 1819 | OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION); |
f7cd0081 | 1820 | |
89625d1e BP |
1821 | k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec); |
1822 | k_exec->dp_ifindex = dp_ifindex; | |
f7cd0081 | 1823 | |
89625d1e | 1824 | nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, |
cf62fa4c PS |
1825 | dp_packet_data(d_exec->packet), |
1826 | dp_packet_size(d_exec->packet)); | |
758c456d JR |
1827 | |
1828 | key_ofs = nl_msg_start_nested(buf, OVS_PACKET_ATTR_KEY); | |
beb75a40 | 1829 | odp_key_from_dp_packet(buf, d_exec->packet); |
758c456d JR |
1830 | nl_msg_end_nested(buf, key_ofs); |
1831 | ||
89625d1e BP |
1832 | nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, |
1833 | d_exec->actions, d_exec->actions_len); | |
43f9ac0a | 1834 | if (d_exec->probe) { |
2e460098 | 1835 | nl_msg_put_flag(buf, OVS_PACKET_ATTR_PROBE); |
43f9ac0a | 1836 | } |
27130224 AZ |
1837 | if (d_exec->mtu) { |
1838 | nl_msg_put_u16(buf, OVS_PACKET_ATTR_MRU, d_exec->mtu); | |
1839 | } | |
0442bfb1 TZ |
1840 | |
1841 | if (d_exec->hash) { | |
1842 | nl_msg_put_u64(buf, OVS_PACKET_ATTR_HASH, d_exec->hash); | |
1843 | } | |
6bc60024 BP |
1844 | } |
1845 | ||
0f3358ea BP |
1846 | /* Executes, against 'dpif', up to the first 'n_ops' operations in 'ops'. |
1847 | * Returns the number actually executed (at least 1, if 'n_ops' is | |
1848 | * positive). */ | |
1849 | static size_t | |
93451a0a AS |
1850 | dpif_netlink_operate__(struct dpif_netlink *dpif, |
1851 | struct dpif_op **ops, size_t n_ops) | |
6bc60024 | 1852 | { |
eabe7c68 BP |
1853 | struct op_auxdata { |
1854 | struct nl_transaction txn; | |
72d32ac0 | 1855 | |
eabe7c68 BP |
1856 | struct ofpbuf request; |
1857 | uint64_t request_stub[1024 / 8]; | |
72d32ac0 BP |
1858 | |
1859 | struct ofpbuf reply; | |
1860 | uint64_t reply_stub[1024 / 8]; | |
8b668ee3 | 1861 | } auxes[OPERATE_MAX_OPS]; |
eabe7c68 | 1862 | |
8b668ee3 | 1863 | struct nl_transaction *txnsp[OPERATE_MAX_OPS]; |
6bc60024 BP |
1864 | size_t i; |
1865 | ||
8b668ee3 | 1866 | n_ops = MIN(n_ops, OPERATE_MAX_OPS); |
6bc60024 | 1867 | for (i = 0; i < n_ops; i++) { |
eabe7c68 | 1868 | struct op_auxdata *aux = &auxes[i]; |
c2b565b5 | 1869 | struct dpif_op *op = ops[i]; |
b99d3cee BP |
1870 | struct dpif_flow_put *put; |
1871 | struct dpif_flow_del *del; | |
6fe09f8c | 1872 | struct dpif_flow_get *get; |
93451a0a | 1873 | struct dpif_netlink_flow flow; |
eabe7c68 BP |
1874 | |
1875 | ofpbuf_use_stub(&aux->request, | |
1876 | aux->request_stub, sizeof aux->request_stub); | |
1877 | aux->txn.request = &aux->request; | |
b99d3cee | 1878 | |
72d32ac0 BP |
1879 | ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub); |
1880 | aux->txn.reply = NULL; | |
1881 | ||
b99d3cee BP |
1882 | switch (op->type) { |
1883 | case DPIF_OP_FLOW_PUT: | |
fa37affa | 1884 | put = &op->flow_put; |
93451a0a | 1885 | dpif_netlink_init_flow_put(dpif, put, &flow); |
6bc60024 | 1886 | if (put->stats) { |
eabe7c68 | 1887 | flow.nlmsg_flags |= NLM_F_ECHO; |
72d32ac0 | 1888 | aux->txn.reply = &aux->reply; |
6bc60024 | 1889 | } |
93451a0a | 1890 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
b99d3cee BP |
1891 | break; |
1892 | ||
1893 | case DPIF_OP_FLOW_DEL: | |
fa37affa | 1894 | del = &op->flow_del; |
93451a0a | 1895 | dpif_netlink_init_flow_del(dpif, del, &flow); |
b99d3cee | 1896 | if (del->stats) { |
eabe7c68 | 1897 | flow.nlmsg_flags |= NLM_F_ECHO; |
72d32ac0 | 1898 | aux->txn.reply = &aux->reply; |
b99d3cee | 1899 | } |
93451a0a | 1900 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
b99d3cee | 1901 | break; |
6bc60024 | 1902 | |
b99d3cee | 1903 | case DPIF_OP_EXECUTE: |
0f3358ea BP |
1904 | /* Can't execute a packet that won't fit in a Netlink attribute. */ |
1905 | if (OVS_UNLIKELY(nl_attr_oversized( | |
fa37affa | 1906 | dp_packet_size(op->execute.packet)))) { |
0f3358ea BP |
1907 | /* Report an error immediately if this is the first operation. |
1908 | * Otherwise the easiest thing to do is to postpone to the next | |
1909 | * call (when this will be the first operation). */ | |
1910 | if (i == 0) { | |
1911 | VLOG_ERR_RL(&error_rl, | |
1912 | "dropping oversized %"PRIu32"-byte packet", | |
fa37affa | 1913 | dp_packet_size(op->execute.packet)); |
0f3358ea BP |
1914 | op->error = ENOBUFS; |
1915 | return 1; | |
1916 | } | |
1917 | n_ops = i; | |
1918 | } else { | |
fa37affa | 1919 | dpif_netlink_encode_execute(dpif->dp_ifindex, &op->execute, |
0f3358ea BP |
1920 | &aux->request); |
1921 | } | |
b99d3cee BP |
1922 | break; |
1923 | ||
6fe09f8c | 1924 | case DPIF_OP_FLOW_GET: |
fa37affa | 1925 | get = &op->flow_get; |
70e5ed6f | 1926 | dpif_netlink_init_flow_get(dpif, get, &flow); |
6fe09f8c | 1927 | aux->txn.reply = get->buffer; |
93451a0a | 1928 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
6fe09f8c JS |
1929 | break; |
1930 | ||
b99d3cee | 1931 | default: |
428b2edd | 1932 | OVS_NOT_REACHED(); |
6bc60024 BP |
1933 | } |
1934 | } | |
1935 | ||
6bc60024 | 1936 | for (i = 0; i < n_ops; i++) { |
eabe7c68 | 1937 | txnsp[i] = &auxes[i].txn; |
6bc60024 | 1938 | } |
a88b4e04 | 1939 | nl_transact_multiple(NETLINK_GENERIC, txnsp, n_ops); |
6bc60024 | 1940 | |
6bc60024 | 1941 | for (i = 0; i < n_ops; i++) { |
72d32ac0 | 1942 | struct op_auxdata *aux = &auxes[i]; |
eabe7c68 | 1943 | struct nl_transaction *txn = &auxes[i].txn; |
c2b565b5 | 1944 | struct dpif_op *op = ops[i]; |
b99d3cee BP |
1945 | struct dpif_flow_put *put; |
1946 | struct dpif_flow_del *del; | |
6fe09f8c | 1947 | struct dpif_flow_get *get; |
6bc60024 | 1948 | |
b99d3cee | 1949 | op->error = txn->error; |
6bc60024 | 1950 | |
b99d3cee BP |
1951 | switch (op->type) { |
1952 | case DPIF_OP_FLOW_PUT: | |
fa37affa | 1953 | put = &op->flow_put; |
cfceb2b5 | 1954 | if (put->stats) { |
b99d3cee | 1955 | if (!op->error) { |
93451a0a | 1956 | struct dpif_netlink_flow reply; |
cfceb2b5 | 1957 | |
93451a0a AS |
1958 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, |
1959 | txn->reply); | |
cfceb2b5 | 1960 | if (!op->error) { |
93451a0a | 1961 | dpif_netlink_flow_get_stats(&reply, put->stats); |
cfceb2b5 BP |
1962 | } |
1963 | } | |
6bc60024 | 1964 | } |
b99d3cee BP |
1965 | break; |
1966 | ||
1967 | case DPIF_OP_FLOW_DEL: | |
fa37affa | 1968 | del = &op->flow_del; |
cfceb2b5 | 1969 | if (del->stats) { |
b99d3cee | 1970 | if (!op->error) { |
93451a0a | 1971 | struct dpif_netlink_flow reply; |
cfceb2b5 | 1972 | |
93451a0a AS |
1973 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, |
1974 | txn->reply); | |
cfceb2b5 | 1975 | if (!op->error) { |
93451a0a | 1976 | dpif_netlink_flow_get_stats(&reply, del->stats); |
cfceb2b5 BP |
1977 | } |
1978 | } | |
b99d3cee BP |
1979 | } |
1980 | break; | |
1981 | ||
1982 | case DPIF_OP_EXECUTE: | |
1983 | break; | |
1984 | ||
6fe09f8c | 1985 | case DPIF_OP_FLOW_GET: |
fa37affa | 1986 | get = &op->flow_get; |
6fe09f8c | 1987 | if (!op->error) { |
93451a0a | 1988 | struct dpif_netlink_flow reply; |
6fe09f8c | 1989 | |
93451a0a | 1990 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, txn->reply); |
6fe09f8c | 1991 | if (!op->error) { |
7a5e0ee7 | 1992 | dpif_netlink_flow_to_dpif_flow(get->flow, &reply); |
6fe09f8c JS |
1993 | } |
1994 | } | |
1995 | break; | |
1996 | ||
b99d3cee | 1997 | default: |
428b2edd | 1998 | OVS_NOT_REACHED(); |
6bc60024 BP |
1999 | } |
2000 | ||
72d32ac0 BP |
2001 | ofpbuf_uninit(&aux->request); |
2002 | ofpbuf_uninit(&aux->reply); | |
6bc60024 | 2003 | } |
0f3358ea BP |
2004 | |
2005 | return n_ops; | |
eabe7c68 BP |
2006 | } |
2007 | ||
6c343984 PB |
2008 | static int |
2009 | parse_flow_get(struct dpif_netlink *dpif, struct dpif_flow_get *get) | |
2010 | { | |
8842fdf1 | 2011 | const char *dpif_type_str = dpif_normalize_type(dpif_type(&dpif->dpif)); |
6c343984 PB |
2012 | struct dpif_flow *dpif_flow = get->flow; |
2013 | struct match match; | |
2014 | struct nlattr *actions; | |
2015 | struct dpif_flow_stats stats; | |
d63ca532 | 2016 | struct dpif_flow_attrs attrs; |
6c343984 PB |
2017 | struct ofpbuf buf; |
2018 | uint64_t act_buf[1024 / 8]; | |
2019 | struct odputil_keybuf maskbuf; | |
2020 | struct odputil_keybuf keybuf; | |
2021 | struct odputil_keybuf actbuf; | |
2022 | struct ofpbuf key, mask, act; | |
2023 | int err; | |
2024 | ||
2025 | ofpbuf_use_stack(&buf, &act_buf, sizeof act_buf); | |
8842fdf1 IM |
2026 | err = netdev_ports_flow_get(dpif_type_str, &match, &actions, get->ufid, |
2027 | &stats, &attrs, &buf); | |
6c343984 PB |
2028 | if (err) { |
2029 | return err; | |
2030 | } | |
2031 | ||
2032 | VLOG_DBG("found flow from netdev, translating to dpif flow"); | |
2033 | ||
2034 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2035 | ofpbuf_use_stack(&act, &actbuf, sizeof actbuf); | |
2036 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); | |
2037 | dpif_netlink_netdev_match_to_dpif_flow(&match, &key, &mask, actions, | |
d63ca532 | 2038 | &stats, &attrs, |
6c343984 PB |
2039 | (ovs_u128 *) get->ufid, |
2040 | dpif_flow, | |
2041 | false); | |
2042 | ofpbuf_put(get->buffer, nl_attr_get(actions), nl_attr_get_size(actions)); | |
2043 | dpif_flow->actions = ofpbuf_at(get->buffer, 0, 0); | |
2044 | dpif_flow->actions_len = nl_attr_get_size(actions); | |
2045 | ||
2046 | return 0; | |
2047 | } | |
2048 | ||
8b668ee3 PB |
2049 | static int |
2050 | parse_flow_put(struct dpif_netlink *dpif, struct dpif_flow_put *put) | |
2051 | { | |
8842fdf1 | 2052 | const char *dpif_type_str = dpif_normalize_type(dpif_type(&dpif->dpif)); |
8b668ee3 PB |
2053 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
2054 | struct match match; | |
2055 | odp_port_t in_port; | |
2056 | const struct nlattr *nla; | |
2057 | size_t left; | |
8b668ee3 PB |
2058 | struct netdev *dev; |
2059 | struct offload_info info; | |
2060 | ovs_be16 dst_port = 0; | |
d9677a1f | 2061 | uint8_t csum_on = false; |
8b668ee3 PB |
2062 | int err; |
2063 | ||
2064 | if (put->flags & DPIF_FP_PROBE) { | |
2065 | return EOPNOTSUPP; | |
2066 | } | |
2067 | ||
2068 | err = parse_key_and_mask_to_match(put->key, put->key_len, put->mask, | |
2069 | put->mask_len, &match); | |
2070 | if (err) { | |
2071 | return err; | |
2072 | } | |
2073 | ||
8b668ee3 | 2074 | in_port = match.flow.in_port.odp_port; |
8842fdf1 | 2075 | dev = netdev_ports_get(in_port, dpif_type_str); |
8b668ee3 PB |
2076 | if (!dev) { |
2077 | return EOPNOTSUPP; | |
2078 | } | |
2079 | ||
00a0a011 | 2080 | /* Get tunnel dst port */ |
8b668ee3 PB |
2081 | NL_ATTR_FOR_EACH(nla, left, put->actions, put->actions_len) { |
2082 | if (nl_attr_type(nla) == OVS_ACTION_ATTR_OUTPUT) { | |
2083 | const struct netdev_tunnel_config *tnl_cfg; | |
2084 | struct netdev *outdev; | |
2085 | odp_port_t out_port; | |
2086 | ||
8b668ee3 | 2087 | out_port = nl_attr_get_odp_port(nla); |
8842fdf1 | 2088 | outdev = netdev_ports_get(out_port, dpif_type_str); |
8b668ee3 PB |
2089 | if (!outdev) { |
2090 | err = EOPNOTSUPP; | |
2091 | goto out; | |
2092 | } | |
2093 | tnl_cfg = netdev_get_tunnel_config(outdev); | |
2094 | if (tnl_cfg && tnl_cfg->dst_port != 0) { | |
2095 | dst_port = tnl_cfg->dst_port; | |
2096 | } | |
d9677a1f EB |
2097 | if (tnl_cfg) { |
2098 | csum_on = tnl_cfg->csum; | |
2099 | } | |
8b668ee3 PB |
2100 | netdev_close(outdev); |
2101 | } | |
2102 | } | |
2103 | ||
8b668ee3 | 2104 | info.tp_dst_port = dst_port; |
d9677a1f | 2105 | info.tunnel_csum_on = csum_on; |
b2ae4069 PB |
2106 | info.recirc_id_shared_with_tc = (dpif->user_features |
2107 | & OVS_DP_F_TC_RECIRC_SHARING); | |
65b84d4a | 2108 | info.tc_modify_flow_deleted = false; |
8b668ee3 PB |
2109 | err = netdev_flow_put(dev, &match, |
2110 | CONST_CAST(struct nlattr *, put->actions), | |
2111 | put->actions_len, | |
2112 | CONST_CAST(ovs_u128 *, put->ufid), | |
2113 | &info, put->stats); | |
2114 | ||
2115 | if (!err) { | |
2116 | if (put->flags & DPIF_FP_MODIFY) { | |
2117 | struct dpif_op *opp; | |
2118 | struct dpif_op op; | |
2119 | ||
2120 | op.type = DPIF_OP_FLOW_DEL; | |
fa37affa BP |
2121 | op.flow_del.key = put->key; |
2122 | op.flow_del.key_len = put->key_len; | |
2123 | op.flow_del.ufid = put->ufid; | |
2124 | op.flow_del.pmd_id = put->pmd_id; | |
2125 | op.flow_del.stats = NULL; | |
2126 | op.flow_del.terse = false; | |
8b668ee3 PB |
2127 | |
2128 | opp = &op; | |
2129 | dpif_netlink_operate__(dpif, &opp, 1); | |
2130 | } | |
2131 | ||
2132 | VLOG_DBG("added flow"); | |
2133 | } else if (err != EEXIST) { | |
738c785f | 2134 | struct netdev *oor_netdev = NULL; |
1028cb71 | 2135 | enum vlog_level level; |
738c785f SB |
2136 | if (err == ENOSPC && netdev_is_offload_rebalance_policy_enabled()) { |
2137 | /* | |
2138 | * We need to set OOR on the input netdev (i.e, 'dev') for the | |
2139 | * flow. But if the flow has a tunnel attribute (i.e, decap action, | |
2140 | * with a virtual device like a VxLAN interface as its in-port), | |
2141 | * then lookup and set OOR on the underlying tunnel (real) netdev. | |
2142 | */ | |
2143 | oor_netdev = flow_get_tunnel_netdev(&match.flow.tunnel); | |
2144 | if (!oor_netdev) { | |
2145 | /* Not a 'tunnel' flow */ | |
2146 | oor_netdev = dev; | |
2147 | } | |
2148 | netdev_set_hw_info(oor_netdev, HW_INFO_TYPE_OOR, true); | |
2149 | } | |
1028cb71 | 2150 | level = (err == ENOSPC || err == EOPNOTSUPP) ? VLL_DBG : VLL_ERR; |
2151 | VLOG_RL(&rl, level, "failed to offload flow: %s: %s", | |
2152 | ovs_strerror(err), | |
2153 | (oor_netdev ? oor_netdev->name : dev->name)); | |
8b668ee3 PB |
2154 | } |
2155 | ||
2156 | out: | |
2157 | if (err && err != EEXIST && (put->flags & DPIF_FP_MODIFY)) { | |
2158 | /* Modified rule can't be offloaded, try and delete from HW */ | |
65b84d4a | 2159 | int del_err = 0; |
2160 | ||
2161 | if (!info.tc_modify_flow_deleted) { | |
2162 | del_err = netdev_flow_del(dev, put->ufid, put->stats); | |
2163 | } | |
8b668ee3 PB |
2164 | |
2165 | if (!del_err) { | |
2166 | /* Delete from hw success, so old flow was offloaded. | |
2167 | * Change flags to create the flow in kernel */ | |
2168 | put->flags &= ~DPIF_FP_MODIFY; | |
2169 | put->flags |= DPIF_FP_CREATE; | |
2170 | } else if (del_err != ENOENT) { | |
2171 | VLOG_ERR_RL(&rl, "failed to delete offloaded flow: %s", | |
2172 | ovs_strerror(del_err)); | |
2173 | /* stop proccesing the flow in kernel */ | |
2174 | err = 0; | |
2175 | } | |
2176 | } | |
2177 | ||
2178 | netdev_close(dev); | |
2179 | ||
2180 | return err; | |
2181 | } | |
2182 | ||
8b668ee3 PB |
2183 | static int |
2184 | try_send_to_netdev(struct dpif_netlink *dpif, struct dpif_op *op) | |
eabe7c68 | 2185 | { |
8b668ee3 | 2186 | int err = EOPNOTSUPP; |
9b00386b | 2187 | |
8b668ee3 PB |
2188 | switch (op->type) { |
2189 | case DPIF_OP_FLOW_PUT: { | |
fa37affa | 2190 | struct dpif_flow_put *put = &op->flow_put; |
8b668ee3 PB |
2191 | |
2192 | if (!put->ufid) { | |
2193 | break; | |
2194 | } | |
3cd99886 | 2195 | |
8b668ee3 | 2196 | err = parse_flow_put(dpif, put); |
f7392b44 | 2197 | log_flow_put_message(&dpif->dpif, &this_module, put, 0); |
8b668ee3 PB |
2198 | break; |
2199 | } | |
0335a89c | 2200 | case DPIF_OP_FLOW_DEL: { |
fa37affa | 2201 | struct dpif_flow_del *del = &op->flow_del; |
0335a89c PB |
2202 | |
2203 | if (!del->ufid) { | |
2204 | break; | |
2205 | } | |
3cd99886 | 2206 | |
8842fdf1 IM |
2207 | err = netdev_ports_flow_del( |
2208 | dpif_normalize_type(dpif_type(&dpif->dpif)), | |
2209 | del->ufid, | |
2210 | del->stats); | |
f7392b44 | 2211 | log_flow_del_message(&dpif->dpif, &this_module, del, 0); |
0335a89c PB |
2212 | break; |
2213 | } | |
6c343984 | 2214 | case DPIF_OP_FLOW_GET: { |
fa37affa | 2215 | struct dpif_flow_get *get = &op->flow_get; |
6c343984 | 2216 | |
fa37affa | 2217 | if (!op->flow_get.ufid) { |
6c343984 PB |
2218 | break; |
2219 | } | |
3cd99886 | 2220 | |
6c343984 | 2221 | err = parse_flow_get(dpif, get); |
f7392b44 | 2222 | log_flow_get_message(&dpif->dpif, &this_module, get, 0); |
6c343984 PB |
2223 | break; |
2224 | } | |
8b668ee3 PB |
2225 | case DPIF_OP_EXECUTE: |
2226 | default: | |
2227 | break; | |
2228 | } | |
2229 | ||
2230 | return err; | |
2231 | } | |
2232 | ||
2233 | static void | |
2234 | dpif_netlink_operate_chunks(struct dpif_netlink *dpif, struct dpif_op **ops, | |
2235 | size_t n_ops) | |
2236 | { | |
eabe7c68 | 2237 | while (n_ops > 0) { |
0f3358ea | 2238 | size_t chunk = dpif_netlink_operate__(dpif, ops, n_ops); |
8b668ee3 | 2239 | |
eabe7c68 BP |
2240 | ops += chunk; |
2241 | n_ops -= chunk; | |
2242 | } | |
6bc60024 BP |
2243 | } |
2244 | ||
8b668ee3 | 2245 | static void |
57924fc9 SB |
2246 | dpif_netlink_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops, |
2247 | enum dpif_offload_type offload_type) | |
8b668ee3 PB |
2248 | { |
2249 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); | |
2250 | struct dpif_op *new_ops[OPERATE_MAX_OPS]; | |
2251 | int count = 0; | |
2252 | int i = 0; | |
2253 | int err = 0; | |
2254 | ||
57924fc9 SB |
2255 | if (offload_type == DPIF_OFFLOAD_ALWAYS && !netdev_is_flow_api_enabled()) { |
2256 | VLOG_DBG("Invalid offload_type: %d", offload_type); | |
2257 | return; | |
2258 | } | |
2259 | ||
2260 | if (offload_type != DPIF_OFFLOAD_NEVER && netdev_is_flow_api_enabled()) { | |
8b668ee3 PB |
2261 | while (n_ops > 0) { |
2262 | count = 0; | |
2263 | ||
2264 | while (n_ops > 0 && count < OPERATE_MAX_OPS) { | |
2265 | struct dpif_op *op = ops[i++]; | |
2266 | ||
2267 | err = try_send_to_netdev(dpif, op); | |
2268 | if (err && err != EEXIST) { | |
57924fc9 SB |
2269 | if (offload_type == DPIF_OFFLOAD_ALWAYS) { |
2270 | /* We got an error while offloading an op. Since | |
2271 | * OFFLOAD_ALWAYS is specified, we stop further | |
2272 | * processing and return to the caller without | |
2273 | * invoking kernel datapath as fallback. But the | |
2274 | * interface requires us to process all n_ops; so | |
2275 | * return the same error in the remaining ops too. | |
2276 | */ | |
2277 | op->error = err; | |
2278 | n_ops--; | |
2279 | while (n_ops > 0) { | |
2280 | op = ops[i++]; | |
2281 | op->error = err; | |
2282 | n_ops--; | |
2283 | } | |
2284 | return; | |
2285 | } | |
8b668ee3 PB |
2286 | new_ops[count++] = op; |
2287 | } else { | |
2288 | op->error = err; | |
2289 | } | |
2290 | ||
2291 | n_ops--; | |
2292 | } | |
2293 | ||
2294 | dpif_netlink_operate_chunks(dpif, new_ops, count); | |
2295 | } | |
57924fc9 | 2296 | } else if (offload_type != DPIF_OFFLOAD_ALWAYS) { |
8b668ee3 PB |
2297 | dpif_netlink_operate_chunks(dpif, ops, n_ops); |
2298 | } | |
2299 | } | |
2300 | ||
09cac43f NR |
2301 | #if _WIN32 |
2302 | static void | |
2303 | dpif_netlink_handler_uninit(struct dpif_handler *handler) | |
2304 | { | |
2305 | vport_delete_sock_pool(handler); | |
2306 | } | |
2307 | ||
2308 | static int | |
2309 | dpif_netlink_handler_init(struct dpif_handler *handler) | |
2310 | { | |
2311 | return vport_create_sock_pool(handler); | |
2312 | } | |
2313 | #else | |
2314 | ||
2315 | static int | |
2316 | dpif_netlink_handler_init(struct dpif_handler *handler) | |
2317 | { | |
2318 | handler->epoll_fd = epoll_create(10); | |
2319 | return handler->epoll_fd < 0 ? errno : 0; | |
2320 | } | |
2321 | ||
2322 | static void | |
2323 | dpif_netlink_handler_uninit(struct dpif_handler *handler) | |
2324 | { | |
2325 | close(handler->epoll_fd); | |
2326 | } | |
2327 | #endif | |
2328 | ||
1579cf67 AW |
2329 | /* Synchronizes 'channels' in 'dpif->handlers' with the set of vports |
2330 | * currently in 'dpif' in the kernel, by adding a new set of channels for | |
2331 | * any kernel vport that lacks one and deleting any channels that have no | |
2332 | * backing kernel vports. */ | |
96fba48f | 2333 | static int |
93451a0a | 2334 | dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) |
b90de034 | 2335 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 2336 | { |
8381a3d3 | 2337 | unsigned long int *keep_channels; |
93451a0a | 2338 | struct dpif_netlink_vport vport; |
8381a3d3 BP |
2339 | size_t keep_channels_nbits; |
2340 | struct nl_dump dump; | |
d57695d7 JS |
2341 | uint64_t reply_stub[NL_DUMP_BUFSIZE / 8]; |
2342 | struct ofpbuf buf; | |
8381a3d3 BP |
2343 | int retval = 0; |
2344 | size_t i; | |
982b8810 | 2345 | |
09cac43f NR |
2346 | ovs_assert(!WINDOWS || n_handlers <= 1); |
2347 | ovs_assert(!WINDOWS || dpif->n_handlers <= 1); | |
2348 | ||
1579cf67 AW |
2349 | if (dpif->n_handlers != n_handlers) { |
2350 | destroy_all_channels(dpif); | |
2351 | dpif->handlers = xzalloc(n_handlers * sizeof *dpif->handlers); | |
2352 | for (i = 0; i < n_handlers; i++) { | |
09cac43f | 2353 | int error; |
1579cf67 AW |
2354 | struct dpif_handler *handler = &dpif->handlers[i]; |
2355 | ||
09cac43f NR |
2356 | error = dpif_netlink_handler_init(handler); |
2357 | if (error) { | |
1579cf67 AW |
2358 | size_t j; |
2359 | ||
2360 | for (j = 0; j < i; j++) { | |
aa5c0216 | 2361 | struct dpif_handler *tmp = &dpif->handlers[j]; |
09cac43f | 2362 | dpif_netlink_handler_uninit(tmp); |
1579cf67 AW |
2363 | } |
2364 | free(dpif->handlers); | |
2365 | dpif->handlers = NULL; | |
2366 | ||
09cac43f | 2367 | return error; |
1579cf67 | 2368 | } |
8381a3d3 | 2369 | } |
1579cf67 AW |
2370 | dpif->n_handlers = n_handlers; |
2371 | } | |
2372 | ||
2373 | for (i = 0; i < n_handlers; i++) { | |
2374 | struct dpif_handler *handler = &dpif->handlers[i]; | |
2375 | ||
2376 | handler->event_offset = handler->n_events = 0; | |
17411ecf | 2377 | } |
b063d9f0 | 2378 | |
8381a3d3 BP |
2379 | keep_channels_nbits = dpif->uc_array_size; |
2380 | keep_channels = bitmap_allocate(keep_channels_nbits); | |
982b8810 | 2381 | |
d57695d7 | 2382 | ofpbuf_use_stub(&buf, reply_stub, sizeof reply_stub); |
93451a0a AS |
2383 | dpif_netlink_port_dump_start__(dpif, &dump); |
2384 | while (!dpif_netlink_port_dump_next__(dpif, &dump, &vport, &buf)) { | |
8381a3d3 | 2385 | uint32_t port_no = odp_to_u32(vport.port_no); |
69c51582 | 2386 | uint32_t upcall_pid; |
8381a3d3 | 2387 | int error; |
50f80534 | 2388 | |
1579cf67 | 2389 | if (port_no >= dpif->uc_array_size |
69c51582 | 2390 | || !vport_get_pid(dpif, port_no, &upcall_pid)) { |
622ea8fd BP |
2391 | struct nl_sock *sock; |
2392 | error = create_nl_sock(dpif, &sock); | |
1579cf67 | 2393 | |
d240e46a | 2394 | if (error) { |
1579cf67 AW |
2395 | goto error; |
2396 | } | |
2397 | ||
622ea8fd | 2398 | error = vport_add_channel(dpif, vport.port_no, sock); |
b063d9f0 | 2399 | if (error) { |
1579cf67 | 2400 | VLOG_INFO("%s: could not add channels for port %s", |
9b00386b | 2401 | dpif_name(&dpif->dpif), vport.name); |
622ea8fd | 2402 | nl_sock_destroy(sock); |
8381a3d3 BP |
2403 | retval = error; |
2404 | goto error; | |
982b8810 | 2405 | } |
622ea8fd | 2406 | upcall_pid = nl_sock_pid(sock); |
8381a3d3 | 2407 | } |
50f80534 | 2408 | |
8381a3d3 | 2409 | /* Configure the vport to deliver misses to 'sock'. */ |
1579cf67 | 2410 | if (vport.upcall_pids[0] == 0 |
69c51582 MC |
2411 | || vport.n_upcall_pids != 1 |
2412 | || upcall_pid != vport.upcall_pids[0]) { | |
93451a0a | 2413 | struct dpif_netlink_vport vport_request; |
989fd548 | 2414 | |
93451a0a | 2415 | dpif_netlink_vport_init(&vport_request); |
989fd548 JP |
2416 | vport_request.cmd = OVS_VPORT_CMD_SET; |
2417 | vport_request.dp_ifindex = dpif->dp_ifindex; | |
8381a3d3 | 2418 | vport_request.port_no = vport.port_no; |
69c51582 MC |
2419 | vport_request.n_upcall_pids = 1; |
2420 | vport_request.upcall_pids = &upcall_pid; | |
93451a0a | 2421 | error = dpif_netlink_vport_transact(&vport_request, NULL, NULL); |
1579cf67 | 2422 | if (error) { |
989fd548 JP |
2423 | VLOG_WARN_RL(&error_rl, |
2424 | "%s: failed to set upcall pid on port: %s", | |
10a89ef0 | 2425 | dpif_name(&dpif->dpif), ovs_strerror(error)); |
989fd548 | 2426 | |
8381a3d3 BP |
2427 | if (error != ENODEV && error != ENOENT) { |
2428 | retval = error; | |
989fd548 | 2429 | } else { |
8381a3d3 BP |
2430 | /* The vport isn't really there, even though the dump says |
2431 | * it is. Probably we just hit a race after a port | |
2432 | * disappeared. */ | |
989fd548 | 2433 | } |
8381a3d3 | 2434 | goto error; |
50f80534 | 2435 | } |
8381a3d3 | 2436 | } |
14b4d2f9 | 2437 | |
8381a3d3 BP |
2438 | if (port_no < keep_channels_nbits) { |
2439 | bitmap_set1(keep_channels, port_no); | |
2440 | } | |
2441 | continue; | |
2442 | ||
2443 | error: | |
1579cf67 | 2444 | vport_del_channels(dpif, vport.port_no); |
982b8810 | 2445 | } |
8381a3d3 | 2446 | nl_dump_done(&dump); |
d57695d7 | 2447 | ofpbuf_uninit(&buf); |
b063d9f0 | 2448 | |
8381a3d3 BP |
2449 | /* Discard any saved channels that we didn't reuse. */ |
2450 | for (i = 0; i < keep_channels_nbits; i++) { | |
2451 | if (!bitmap_is_set(keep_channels, i)) { | |
1579cf67 | 2452 | vport_del_channels(dpif, u32_to_odp(i)); |
8381a3d3 BP |
2453 | } |
2454 | } | |
2455 | free(keep_channels); | |
2456 | ||
2457 | return retval; | |
2458 | } | |
2459 | ||
2460 | static int | |
93451a0a | 2461 | dpif_netlink_recv_set__(struct dpif_netlink *dpif, bool enable) |
b90de034 | 2462 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
8381a3d3 | 2463 | { |
1579cf67 | 2464 | if ((dpif->handlers != NULL) == enable) { |
8381a3d3 BP |
2465 | return 0; |
2466 | } else if (!enable) { | |
1579cf67 | 2467 | destroy_all_channels(dpif); |
8381a3d3 BP |
2468 | return 0; |
2469 | } else { | |
93451a0a | 2470 | return dpif_netlink_refresh_channels(dpif, 1); |
8381a3d3 | 2471 | } |
96fba48f BP |
2472 | } |
2473 | ||
9fafa796 | 2474 | static int |
93451a0a | 2475 | dpif_netlink_recv_set(struct dpif *dpif_, bool enable) |
9fafa796 | 2476 | { |
93451a0a | 2477 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
2478 | int error; |
2479 | ||
1579cf67 | 2480 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 2481 | error = dpif_netlink_recv_set__(dpif, enable); |
1579cf67 | 2482 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
2483 | |
2484 | return error; | |
2485 | } | |
2486 | ||
1954e6bb | 2487 | static int |
93451a0a | 2488 | dpif_netlink_handlers_set(struct dpif *dpif_, uint32_t n_handlers) |
1954e6bb | 2489 | { |
93451a0a | 2490 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1579cf67 AW |
2491 | int error = 0; |
2492 | ||
09cac43f NR |
2493 | #ifdef _WIN32 |
2494 | /* Multiple upcall handlers will be supported once kernel datapath supports | |
2495 | * it. */ | |
2496 | if (n_handlers > 1) { | |
2497 | return error; | |
2498 | } | |
2499 | #endif | |
2500 | ||
1579cf67 AW |
2501 | fat_rwlock_wrlock(&dpif->upcall_lock); |
2502 | if (dpif->handlers) { | |
93451a0a | 2503 | error = dpif_netlink_refresh_channels(dpif, n_handlers); |
1579cf67 AW |
2504 | } |
2505 | fat_rwlock_unlock(&dpif->upcall_lock); | |
2506 | ||
2507 | return error; | |
1954e6bb AW |
2508 | } |
2509 | ||
aae51f53 | 2510 | static int |
93451a0a | 2511 | dpif_netlink_queue_to_priority(const struct dpif *dpif OVS_UNUSED, |
aae51f53 BP |
2512 | uint32_t queue_id, uint32_t *priority) |
2513 | { | |
2514 | if (queue_id < 0xf000) { | |
17ee3c1f | 2515 | *priority = TC_H_MAKE(1 << 16, queue_id + 1); |
aae51f53 BP |
2516 | return 0; |
2517 | } else { | |
2518 | return EINVAL; | |
2519 | } | |
2520 | } | |
2521 | ||
96fba48f | 2522 | static int |
7a5e0ee7 IM |
2523 | parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, |
2524 | int *dp_ifindex) | |
856081f6 | 2525 | { |
df2c07f4 | 2526 | static const struct nl_policy ovs_packet_policy[] = { |
856081f6 | 2527 | /* Always present. */ |
df2c07f4 | 2528 | [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, |
856081f6 | 2529 | .min_len = ETH_HEADER_LEN }, |
df2c07f4 | 2530 | [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, |
856081f6 | 2531 | |
df2c07f4 | 2532 | /* OVS_PACKET_CMD_ACTION only. */ |
e995e3df | 2533 | [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true }, |
8b7ea2d4 | 2534 | [OVS_PACKET_ATTR_EGRESS_TUN_KEY] = { .type = NL_A_NESTED, .optional = true }, |
7321bda3 | 2535 | [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, |
0442bfb1 TZ |
2536 | [OVS_PACKET_ATTR_MRU] = { .type = NL_A_U16, .optional = true }, |
2537 | [OVS_PACKET_ATTR_HASH] = { .type = NL_A_U64, .optional = true } | |
856081f6 BP |
2538 | }; |
2539 | ||
0a2869d5 BP |
2540 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
2541 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
2542 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
2543 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
982b8810 | 2544 | |
0a2869d5 | 2545 | struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)]; |
df2c07f4 JP |
2546 | if (!nlmsg || !genl || !ovs_header |
2547 | || nlmsg->nlmsg_type != ovs_packet_family | |
2548 | || !nl_policy_parse(&b, 0, ovs_packet_policy, a, | |
2549 | ARRAY_SIZE(ovs_packet_policy))) { | |
856081f6 BP |
2550 | return EINVAL; |
2551 | } | |
2552 | ||
0a2869d5 BP |
2553 | int type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS |
2554 | : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION | |
2555 | : -1); | |
aaff4b55 BP |
2556 | if (type < 0) { |
2557 | return EINVAL; | |
2558 | } | |
82272ede | 2559 | |
877c9270 | 2560 | /* (Re)set ALL fields of '*upcall' on successful return. */ |
aaff4b55 | 2561 | upcall->type = type; |
ebc56baa BP |
2562 | upcall->key = CONST_CAST(struct nlattr *, |
2563 | nl_attr_get(a[OVS_PACKET_ATTR_KEY])); | |
df2c07f4 | 2564 | upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]); |
7a5e0ee7 | 2565 | odp_flow_key_hash(upcall->key, upcall->key_len, &upcall->ufid); |
e995e3df | 2566 | upcall->userdata = a[OVS_PACKET_ATTR_USERDATA]; |
8b7ea2d4 | 2567 | upcall->out_tun_key = a[OVS_PACKET_ATTR_EGRESS_TUN_KEY]; |
7321bda3 | 2568 | upcall->actions = a[OVS_PACKET_ATTR_ACTIONS]; |
27130224 | 2569 | upcall->mru = a[OVS_PACKET_ATTR_MRU]; |
0442bfb1 | 2570 | upcall->hash = a[OVS_PACKET_ATTR_HASH]; |
da546e07 JR |
2571 | |
2572 | /* Allow overwriting the netlink attribute header without reallocating. */ | |
cf62fa4c | 2573 | dp_packet_use_stub(&upcall->packet, |
da546e07 JR |
2574 | CONST_CAST(struct nlattr *, |
2575 | nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1, | |
2576 | nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) + | |
2577 | sizeof(struct nlattr)); | |
cf62fa4c PS |
2578 | dp_packet_set_data(&upcall->packet, |
2579 | (char *)dp_packet_data(&upcall->packet) + sizeof(struct nlattr)); | |
2580 | dp_packet_set_size(&upcall->packet, nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET])); | |
da546e07 | 2581 | |
2482b0b0 JS |
2582 | if (nl_attr_find__(upcall->key, upcall->key_len, OVS_KEY_ATTR_ETHERNET)) { |
2583 | /* Ethernet frame */ | |
2584 | upcall->packet.packet_type = htonl(PT_ETH); | |
2585 | } else { | |
2586 | /* Non-Ethernet packet. Get the Ethertype from the NL attributes */ | |
2587 | ovs_be16 ethertype = 0; | |
2588 | const struct nlattr *et_nla = nl_attr_find__(upcall->key, | |
2589 | upcall->key_len, | |
2590 | OVS_KEY_ATTR_ETHERTYPE); | |
2591 | if (et_nla) { | |
2592 | ethertype = nl_attr_get_be16(et_nla); | |
2593 | } | |
2594 | upcall->packet.packet_type = PACKET_TYPE_BE(OFPHTN_ETHERTYPE, | |
2595 | ntohs(ethertype)); | |
2596 | dp_packet_set_l3(&upcall->packet, dp_packet_data(&upcall->packet)); | |
2597 | } | |
2598 | ||
df2c07f4 | 2599 | *dp_ifindex = ovs_header->dp_ifindex; |
982b8810 | 2600 | |
856081f6 BP |
2601 | return 0; |
2602 | } | |
2603 | ||
09cac43f NR |
2604 | #ifdef _WIN32 |
2605 | #define PACKET_RECV_BATCH_SIZE 50 | |
2606 | static int | |
2607 | dpif_netlink_recv_windows(struct dpif_netlink *dpif, uint32_t handler_id, | |
2608 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
2609 | OVS_REQ_RDLOCK(dpif->upcall_lock) | |
2610 | { | |
2611 | struct dpif_handler *handler; | |
2612 | int read_tries = 0; | |
2613 | struct dpif_windows_vport_sock *sock_pool; | |
2614 | uint32_t i; | |
2615 | ||
2616 | if (!dpif->handlers) { | |
2617 | return EAGAIN; | |
2618 | } | |
2619 | ||
2620 | /* Only one handler is supported currently. */ | |
2621 | if (handler_id >= 1) { | |
2622 | return EAGAIN; | |
2623 | } | |
2624 | ||
2625 | if (handler_id >= dpif->n_handlers) { | |
2626 | return EAGAIN; | |
2627 | } | |
2628 | ||
2629 | handler = &dpif->handlers[handler_id]; | |
2630 | sock_pool = handler->vport_sock_pool; | |
2631 | ||
2632 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
2633 | for (;;) { | |
2634 | int dp_ifindex; | |
2635 | int error; | |
2636 | ||
2637 | if (++read_tries > PACKET_RECV_BATCH_SIZE) { | |
2638 | return EAGAIN; | |
2639 | } | |
2640 | ||
a86bd14e | 2641 | error = nl_sock_recv(sock_pool[i].nl_sock, buf, NULL, false); |
09cac43f NR |
2642 | if (error == ENOBUFS) { |
2643 | /* ENOBUFS typically means that we've received so many | |
2644 | * packets that the buffer overflowed. Try again | |
2645 | * immediately because there's almost certainly a packet | |
2646 | * waiting for us. */ | |
2647 | /* XXX: report_loss(dpif, ch, idx, handler_id); */ | |
2648 | continue; | |
2649 | } | |
2650 | ||
2651 | /* XXX: ch->last_poll = time_msec(); */ | |
2652 | if (error) { | |
2653 | if (error == EAGAIN) { | |
2654 | break; | |
2655 | } | |
2656 | return error; | |
2657 | } | |
2658 | ||
7a5e0ee7 | 2659 | error = parse_odp_packet(buf, upcall, &dp_ifindex); |
09cac43f NR |
2660 | if (!error && dp_ifindex == dpif->dp_ifindex) { |
2661 | return 0; | |
2662 | } else if (error) { | |
2663 | return error; | |
2664 | } | |
2665 | } | |
2666 | } | |
2667 | ||
2668 | return EAGAIN; | |
2669 | } | |
2670 | #else | |
856081f6 | 2671 | static int |
93451a0a AS |
2672 | dpif_netlink_recv__(struct dpif_netlink *dpif, uint32_t handler_id, |
2673 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
b90de034 | 2674 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
96fba48f | 2675 | { |
1579cf67 | 2676 | struct dpif_handler *handler; |
17411ecf | 2677 | int read_tries = 0; |
96fba48f | 2678 | |
1579cf67 AW |
2679 | if (!dpif->handlers || handler_id >= dpif->n_handlers) { |
2680 | return EAGAIN; | |
982b8810 BP |
2681 | } |
2682 | ||
1579cf67 AW |
2683 | handler = &dpif->handlers[handler_id]; |
2684 | if (handler->event_offset >= handler->n_events) { | |
8522ba09 | 2685 | int retval; |
989fd548 | 2686 | |
1579cf67 | 2687 | handler->event_offset = handler->n_events = 0; |
f6d1465c | 2688 | |
8522ba09 | 2689 | do { |
1579cf67 | 2690 | retval = epoll_wait(handler->epoll_fd, handler->epoll_events, |
989fd548 | 2691 | dpif->uc_array_size, 0); |
8522ba09 | 2692 | } while (retval < 0 && errno == EINTR); |
09cac43f | 2693 | |
8522ba09 BP |
2694 | if (retval < 0) { |
2695 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
10a89ef0 | 2696 | VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", ovs_strerror(errno)); |
989fd548 | 2697 | } else if (retval > 0) { |
1579cf67 | 2698 | handler->n_events = retval; |
8522ba09 | 2699 | } |
8522ba09 BP |
2700 | } |
2701 | ||
1579cf67 AW |
2702 | while (handler->event_offset < handler->n_events) { |
2703 | int idx = handler->epoll_events[handler->event_offset].data.u32; | |
69c51582 | 2704 | struct dpif_channel *ch = &dpif->channels[idx]; |
8522ba09 | 2705 | |
1579cf67 | 2706 | handler->event_offset++; |
17411ecf | 2707 | |
f6d1465c | 2708 | for (;;) { |
8522ba09 | 2709 | int dp_ifindex; |
f6d1465c | 2710 | int error; |
17411ecf | 2711 | |
f6d1465c BP |
2712 | if (++read_tries > 50) { |
2713 | return EAGAIN; | |
2714 | } | |
17411ecf | 2715 | |
a86bd14e | 2716 | error = nl_sock_recv(ch->sock, buf, NULL, false); |
14b4d2f9 BP |
2717 | if (error == ENOBUFS) { |
2718 | /* ENOBUFS typically means that we've received so many | |
2719 | * packets that the buffer overflowed. Try again | |
2720 | * immediately because there's almost certainly a packet | |
2721 | * waiting for us. */ | |
9b00386b | 2722 | report_loss(dpif, ch, idx, handler_id); |
14b4d2f9 BP |
2723 | continue; |
2724 | } | |
2725 | ||
2726 | ch->last_poll = time_msec(); | |
72d32ac0 | 2727 | if (error) { |
72d32ac0 BP |
2728 | if (error == EAGAIN) { |
2729 | break; | |
2730 | } | |
f6d1465c BP |
2731 | return error; |
2732 | } | |
17411ecf | 2733 | |
7a5e0ee7 | 2734 | error = parse_odp_packet(buf, upcall, &dp_ifindex); |
a12b3ead | 2735 | if (!error && dp_ifindex == dpif->dp_ifindex) { |
f6d1465c | 2736 | return 0; |
989fd548 | 2737 | } else if (error) { |
f6d1465c | 2738 | return error; |
17411ecf | 2739 | } |
982b8810 | 2740 | } |
50f80534 | 2741 | } |
982b8810 BP |
2742 | |
2743 | return EAGAIN; | |
96fba48f | 2744 | } |
09cac43f | 2745 | #endif |
96fba48f | 2746 | |
9fafa796 | 2747 | static int |
93451a0a AS |
2748 | dpif_netlink_recv(struct dpif *dpif_, uint32_t handler_id, |
2749 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
9fafa796 | 2750 | { |
93451a0a | 2751 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
2752 | int error; |
2753 | ||
1579cf67 | 2754 | fat_rwlock_rdlock(&dpif->upcall_lock); |
09cac43f NR |
2755 | #ifdef _WIN32 |
2756 | error = dpif_netlink_recv_windows(dpif, handler_id, upcall, buf); | |
2757 | #else | |
93451a0a | 2758 | error = dpif_netlink_recv__(dpif, handler_id, upcall, buf); |
09cac43f | 2759 | #endif |
1579cf67 | 2760 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
2761 | |
2762 | return error; | |
2763 | } | |
2764 | ||
96fba48f | 2765 | static void |
93451a0a | 2766 | dpif_netlink_recv_wait__(struct dpif_netlink *dpif, uint32_t handler_id) |
b90de034 | 2767 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
96fba48f | 2768 | { |
93451a0a | 2769 | #ifdef _WIN32 |
09cac43f NR |
2770 | uint32_t i; |
2771 | struct dpif_windows_vport_sock *sock_pool = | |
2772 | dpif->handlers[handler_id].vport_sock_pool; | |
2773 | ||
2774 | /* Only one handler is supported currently. */ | |
2775 | if (handler_id >= 1) { | |
2776 | return; | |
2777 | } | |
2778 | ||
2779 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
2780 | nl_sock_wait(sock_pool[i].nl_sock, POLLIN); | |
2781 | } | |
93451a0a | 2782 | #else |
1579cf67 AW |
2783 | if (dpif->handlers && handler_id < dpif->n_handlers) { |
2784 | struct dpif_handler *handler = &dpif->handlers[handler_id]; | |
2785 | ||
2786 | poll_fd_wait(handler->epoll_fd, POLLIN); | |
17411ecf | 2787 | } |
93451a0a | 2788 | #endif |
96fba48f BP |
2789 | } |
2790 | ||
1ba530f4 | 2791 | static void |
93451a0a | 2792 | dpif_netlink_recv_wait(struct dpif *dpif_, uint32_t handler_id) |
1ba530f4 | 2793 | { |
93451a0a | 2794 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
17411ecf | 2795 | |
b90de034 | 2796 | fat_rwlock_rdlock(&dpif->upcall_lock); |
93451a0a | 2797 | dpif_netlink_recv_wait__(dpif, handler_id); |
b90de034 AW |
2798 | fat_rwlock_unlock(&dpif->upcall_lock); |
2799 | } | |
2800 | ||
2801 | static void | |
93451a0a | 2802 | dpif_netlink_recv_purge__(struct dpif_netlink *dpif) |
b90de034 AW |
2803 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
2804 | { | |
1579cf67 | 2805 | if (dpif->handlers) { |
69c51582 | 2806 | size_t i; |
1579cf67 | 2807 | |
69c51582 MC |
2808 | if (!dpif->channels[0].sock) { |
2809 | return; | |
2810 | } | |
1579cf67 | 2811 | for (i = 0; i < dpif->uc_array_size; i++ ) { |
1ba530f4 | 2812 | |
69c51582 | 2813 | nl_sock_drain(dpif->channels[i].sock); |
989fd548 | 2814 | } |
1ba530f4 | 2815 | } |
b90de034 AW |
2816 | } |
2817 | ||
2818 | static void | |
93451a0a | 2819 | dpif_netlink_recv_purge(struct dpif *dpif_) |
b90de034 | 2820 | { |
93451a0a | 2821 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
b90de034 AW |
2822 | |
2823 | fat_rwlock_wrlock(&dpif->upcall_lock); | |
93451a0a | 2824 | dpif_netlink_recv_purge__(dpif); |
1579cf67 | 2825 | fat_rwlock_unlock(&dpif->upcall_lock); |
1ba530f4 BP |
2826 | } |
2827 | ||
b5cbbcf6 AZ |
2828 | static char * |
2829 | dpif_netlink_get_datapath_version(void) | |
2830 | { | |
2831 | char *version_str = NULL; | |
2832 | ||
2833 | #ifdef __linux__ | |
2834 | ||
2835 | #define MAX_VERSION_STR_SIZE 80 | |
2836 | #define LINUX_DATAPATH_VERSION_FILE "/sys/module/openvswitch/version" | |
2837 | FILE *f; | |
2838 | ||
2839 | f = fopen(LINUX_DATAPATH_VERSION_FILE, "r"); | |
2840 | if (f) { | |
2841 | char *newline; | |
2842 | char version[MAX_VERSION_STR_SIZE]; | |
2843 | ||
2844 | if (fgets(version, MAX_VERSION_STR_SIZE, f)) { | |
2845 | newline = strchr(version, '\n'); | |
2846 | if (newline) { | |
2847 | *newline = '\0'; | |
2848 | } | |
2849 | version_str = xstrdup(version); | |
2850 | } | |
2851 | fclose(f); | |
2852 | } | |
2853 | #endif | |
2854 | ||
2855 | return version_str; | |
2856 | } | |
2857 | ||
c11c9f4a DDP |
2858 | struct dpif_netlink_ct_dump_state { |
2859 | struct ct_dpif_dump_state up; | |
2860 | struct nl_ct_dump_state *nl_ct_dump; | |
2861 | }; | |
2862 | ||
2863 | static int | |
2864 | dpif_netlink_ct_dump_start(struct dpif *dpif OVS_UNUSED, | |
2865 | struct ct_dpif_dump_state **dump_, | |
ded30c74 | 2866 | const uint16_t *zone, int *ptot_bkts) |
c11c9f4a DDP |
2867 | { |
2868 | struct dpif_netlink_ct_dump_state *dump; | |
2869 | int err; | |
2870 | ||
2871 | dump = xzalloc(sizeof *dump); | |
ded30c74 | 2872 | err = nl_ct_dump_start(&dump->nl_ct_dump, zone, ptot_bkts); |
c11c9f4a DDP |
2873 | if (err) { |
2874 | free(dump); | |
2875 | return err; | |
2876 | } | |
2877 | ||
2878 | *dump_ = &dump->up; | |
2879 | ||
2880 | return 0; | |
2881 | } | |
2882 | ||
2883 | static int | |
2884 | dpif_netlink_ct_dump_next(struct dpif *dpif OVS_UNUSED, | |
2885 | struct ct_dpif_dump_state *dump_, | |
2886 | struct ct_dpif_entry *entry) | |
2887 | { | |
2888 | struct dpif_netlink_ct_dump_state *dump; | |
2889 | ||
2890 | INIT_CONTAINER(dump, dump_, up); | |
2891 | ||
2892 | return nl_ct_dump_next(dump->nl_ct_dump, entry); | |
2893 | } | |
2894 | ||
2895 | static int | |
2896 | dpif_netlink_ct_dump_done(struct dpif *dpif OVS_UNUSED, | |
2897 | struct ct_dpif_dump_state *dump_) | |
2898 | { | |
2899 | struct dpif_netlink_ct_dump_state *dump; | |
c11c9f4a DDP |
2900 | |
2901 | INIT_CONTAINER(dump, dump_, up); | |
2902 | ||
1f161318 | 2903 | int err = nl_ct_dump_done(dump->nl_ct_dump); |
c11c9f4a DDP |
2904 | free(dump); |
2905 | return err; | |
2906 | } | |
15eabc97 DDP |
2907 | |
2908 | static int | |
817a7657 YHW |
2909 | dpif_netlink_ct_flush(struct dpif *dpif OVS_UNUSED, const uint16_t *zone, |
2910 | const struct ct_dpif_tuple *tuple) | |
15eabc97 | 2911 | { |
817a7657 YHW |
2912 | if (tuple) { |
2913 | return nl_ct_flush_tuple(tuple, zone ? *zone : 0); | |
2914 | } else if (zone) { | |
15eabc97 DDP |
2915 | return nl_ct_flush_zone(*zone); |
2916 | } else { | |
2917 | return nl_ct_flush(); | |
2918 | } | |
2919 | } | |
c11c9f4a | 2920 | |
906ff9d2 YHW |
2921 | static int |
2922 | dpif_netlink_ct_set_limits(struct dpif *dpif OVS_UNUSED, | |
2923 | const uint32_t *default_limits, | |
2924 | const struct ovs_list *zone_limits) | |
2925 | { | |
2926 | struct ovs_zone_limit req_zone_limit; | |
2927 | ||
2928 | if (ovs_ct_limit_family < 0) { | |
2929 | return EOPNOTSUPP; | |
2930 | } | |
2931 | ||
2932 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
2933 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
2934 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_SET, | |
2935 | OVS_CT_LIMIT_VERSION); | |
2936 | ||
2937 | struct ovs_header *ovs_header; | |
2938 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
2939 | ovs_header->dp_ifindex = 0; | |
2940 | ||
2941 | size_t opt_offset; | |
2942 | opt_offset = nl_msg_start_nested(request, OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
2943 | if (default_limits) { | |
2944 | req_zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE; | |
2945 | req_zone_limit.limit = *default_limits; | |
2946 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2947 | } | |
2948 | ||
2949 | if (!ovs_list_is_empty(zone_limits)) { | |
2950 | struct ct_dpif_zone_limit *zone_limit; | |
2951 | ||
2952 | LIST_FOR_EACH (zone_limit, node, zone_limits) { | |
2953 | req_zone_limit.zone_id = zone_limit->zone; | |
2954 | req_zone_limit.limit = zone_limit->limit; | |
2955 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2956 | } | |
2957 | } | |
2958 | nl_msg_end_nested(request, opt_offset); | |
2959 | ||
2960 | int err = nl_transact(NETLINK_GENERIC, request, NULL); | |
c225ce22 | 2961 | ofpbuf_delete(request); |
906ff9d2 YHW |
2962 | return err; |
2963 | } | |
2964 | ||
2965 | static int | |
2966 | dpif_netlink_zone_limits_from_ofpbuf(const struct ofpbuf *buf, | |
2967 | uint32_t *default_limit, | |
2968 | struct ovs_list *zone_limits) | |
2969 | { | |
2970 | static const struct nl_policy ovs_ct_limit_policy[] = { | |
2971 | [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NL_A_NESTED, | |
2972 | .optional = true }, | |
2973 | }; | |
2974 | ||
2975 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); | |
2976 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
2977 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
2978 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
2979 | ||
2980 | struct nlattr *attr[ARRAY_SIZE(ovs_ct_limit_policy)]; | |
2981 | ||
2982 | if (!nlmsg || !genl || !ovs_header | |
2983 | || nlmsg->nlmsg_type != ovs_ct_limit_family | |
2984 | || !nl_policy_parse(&b, 0, ovs_ct_limit_policy, attr, | |
2985 | ARRAY_SIZE(ovs_ct_limit_policy))) { | |
2986 | return EINVAL; | |
2987 | } | |
2988 | ||
2989 | ||
2990 | if (!attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) { | |
2991 | return EINVAL; | |
2992 | } | |
2993 | ||
2994 | int rem = NLA_ALIGN( | |
2995 | nl_attr_get_size(attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT])); | |
2996 | const struct ovs_zone_limit *zone_limit = | |
2997 | nl_attr_get(attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]); | |
2998 | ||
2999 | while (rem >= sizeof *zone_limit) { | |
3000 | if (zone_limit->zone_id == OVS_ZONE_LIMIT_DEFAULT_ZONE) { | |
3001 | *default_limit = zone_limit->limit; | |
3002 | } else if (zone_limit->zone_id < OVS_ZONE_LIMIT_DEFAULT_ZONE || | |
3003 | zone_limit->zone_id > UINT16_MAX) { | |
3004 | } else { | |
3005 | ct_dpif_push_zone_limit(zone_limits, zone_limit->zone_id, | |
3006 | zone_limit->limit, zone_limit->count); | |
3007 | } | |
3008 | rem -= NLA_ALIGN(sizeof *zone_limit); | |
3009 | zone_limit = ALIGNED_CAST(struct ovs_zone_limit *, | |
3010 | (unsigned char *) zone_limit + NLA_ALIGN(sizeof *zone_limit)); | |
3011 | } | |
3012 | return 0; | |
3013 | } | |
3014 | ||
3015 | static int | |
3016 | dpif_netlink_ct_get_limits(struct dpif *dpif OVS_UNUSED, | |
3017 | uint32_t *default_limit, | |
3018 | const struct ovs_list *zone_limits_request, | |
3019 | struct ovs_list *zone_limits_reply) | |
3020 | { | |
3021 | if (ovs_ct_limit_family < 0) { | |
3022 | return EOPNOTSUPP; | |
3023 | } | |
3024 | ||
3025 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
3026 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
3027 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_GET, | |
3028 | OVS_CT_LIMIT_VERSION); | |
3029 | ||
3030 | struct ovs_header *ovs_header; | |
3031 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
3032 | ovs_header->dp_ifindex = 0; | |
3033 | ||
3034 | if (!ovs_list_is_empty(zone_limits_request)) { | |
3035 | size_t opt_offset = nl_msg_start_nested(request, | |
3036 | OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
3037 | ||
3038 | struct ovs_zone_limit req_zone_limit; | |
3039 | req_zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE; | |
3040 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
3041 | ||
3042 | struct ct_dpif_zone_limit *zone_limit; | |
3043 | LIST_FOR_EACH (zone_limit, node, zone_limits_request) { | |
3044 | req_zone_limit.zone_id = zone_limit->zone; | |
3045 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
3046 | } | |
3047 | ||
3048 | nl_msg_end_nested(request, opt_offset); | |
3049 | } | |
3050 | ||
3051 | struct ofpbuf *reply; | |
3052 | int err = nl_transact(NETLINK_GENERIC, request, &reply); | |
3053 | if (err) { | |
3054 | goto out; | |
3055 | } | |
3056 | ||
3057 | err = dpif_netlink_zone_limits_from_ofpbuf(reply, default_limit, | |
3058 | zone_limits_reply); | |
3059 | ||
3060 | out: | |
c225ce22 YS |
3061 | ofpbuf_delete(request); |
3062 | ofpbuf_delete(reply); | |
906ff9d2 YHW |
3063 | return err; |
3064 | } | |
3065 | ||
3066 | static int | |
3067 | dpif_netlink_ct_del_limits(struct dpif *dpif OVS_UNUSED, | |
3068 | const struct ovs_list *zone_limits) | |
3069 | { | |
3070 | if (ovs_ct_limit_family < 0) { | |
3071 | return EOPNOTSUPP; | |
3072 | } | |
3073 | ||
3074 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
3075 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
3076 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_DEL, | |
3077 | OVS_CT_LIMIT_VERSION); | |
3078 | ||
3079 | struct ovs_header *ovs_header; | |
3080 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
3081 | ovs_header->dp_ifindex = 0; | |
3082 | ||
3083 | if (!ovs_list_is_empty(zone_limits)) { | |
3084 | size_t opt_offset = | |
3085 | nl_msg_start_nested(request, OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
3086 | ||
3087 | struct ct_dpif_zone_limit *zone_limit; | |
3088 | LIST_FOR_EACH (zone_limit, node, zone_limits) { | |
3089 | struct ovs_zone_limit req_zone_limit; | |
3090 | req_zone_limit.zone_id = zone_limit->zone; | |
3091 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
3092 | } | |
3093 | nl_msg_end_nested(request, opt_offset); | |
3094 | } | |
3095 | ||
3096 | int err = nl_transact(NETLINK_GENERIC, request, NULL); | |
3097 | ||
c225ce22 | 3098 | ofpbuf_delete(request); |
906ff9d2 YHW |
3099 | return err; |
3100 | } | |
1f161318 YHW |
3101 | |
3102 | #define NL_TP_NAME_PREFIX "ovs_tp_" | |
3103 | ||
3104 | struct dpif_netlink_timeout_policy_protocol { | |
3105 | uint16_t l3num; | |
3106 | uint8_t l4num; | |
3107 | }; | |
3108 | ||
3109 | enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol { | |
3110 | DPIF_NL_TP_AF_INET_TCP, | |
3111 | DPIF_NL_TP_AF_INET_UDP, | |
3112 | DPIF_NL_TP_AF_INET_ICMP, | |
3113 | DPIF_NL_TP_AF_INET6_TCP, | |
3114 | DPIF_NL_TP_AF_INET6_UDP, | |
3115 | DPIF_NL_TP_AF_INET6_ICMPV6, | |
3116 | DPIF_NL_TP_MAX | |
3117 | }; | |
3118 | ||
3119 | #define DPIF_NL_ALL_TP ((1UL << DPIF_NL_TP_MAX) - 1) | |
3120 | ||
3121 | ||
3122 | static struct dpif_netlink_timeout_policy_protocol tp_protos[] = { | |
3123 | [DPIF_NL_TP_AF_INET_TCP] = { .l3num = AF_INET, .l4num = IPPROTO_TCP }, | |
3124 | [DPIF_NL_TP_AF_INET_UDP] = { .l3num = AF_INET, .l4num = IPPROTO_UDP }, | |
3125 | [DPIF_NL_TP_AF_INET_ICMP] = { .l3num = AF_INET, .l4num = IPPROTO_ICMP }, | |
3126 | [DPIF_NL_TP_AF_INET6_TCP] = { .l3num = AF_INET6, .l4num = IPPROTO_TCP }, | |
3127 | [DPIF_NL_TP_AF_INET6_UDP] = { .l3num = AF_INET6, .l4num = IPPROTO_UDP }, | |
3128 | [DPIF_NL_TP_AF_INET6_ICMPV6] = { .l3num = AF_INET6, | |
3129 | .l4num = IPPROTO_ICMPV6 }, | |
3130 | }; | |
3131 | ||
3132 | static void | |
3133 | dpif_netlink_format_tp_name(uint32_t id, uint16_t l3num, uint8_t l4num, | |
187bb41f | 3134 | char **tp_name) |
1f161318 | 3135 | { |
187bb41f YHW |
3136 | struct ds ds = DS_EMPTY_INITIALIZER; |
3137 | ds_put_format(&ds, "%s%"PRIu32"_", NL_TP_NAME_PREFIX, id); | |
3138 | ct_dpif_format_ipproto(&ds, l4num); | |
1f161318 YHW |
3139 | |
3140 | if (l3num == AF_INET) { | |
187bb41f | 3141 | ds_put_cstr(&ds, "4"); |
1f161318 | 3142 | } else if (l3num == AF_INET6 && l4num != IPPROTO_ICMPV6) { |
187bb41f | 3143 | ds_put_cstr(&ds, "6"); |
1f161318 YHW |
3144 | } |
3145 | ||
187bb41f YHW |
3146 | ovs_assert(ds.length < CTNL_TIMEOUT_NAME_MAX); |
3147 | ||
3148 | *tp_name = ds_steal_cstr(&ds); | |
3149 | } | |
3150 | ||
3151 | static int | |
3152 | dpif_netlink_ct_get_timeout_policy_name(struct dpif *dpif OVS_UNUSED, | |
3153 | uint32_t tp_id, uint16_t dl_type, | |
3154 | uint8_t nw_proto, char **tp_name, | |
3155 | bool *is_generic) | |
3156 | { | |
3157 | dpif_netlink_format_tp_name(tp_id, | |
3158 | dl_type == ETH_TYPE_IP ? AF_INET : AF_INET6, | |
3159 | nw_proto, tp_name); | |
3160 | *is_generic = false; | |
3161 | return 0; | |
1f161318 YHW |
3162 | } |
3163 | ||
3164 | #define CT_DPIF_NL_TP_TCP_MAPPINGS \ | |
3165 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT, SYN_SENT) \ | |
3166 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_RECV, SYN_RECV) \ | |
3167 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, ESTABLISHED, ESTABLISHED) \ | |
3168 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, FIN_WAIT, FIN_WAIT) \ | |
3169 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE_WAIT, CLOSE_WAIT) \ | |
3170 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, LAST_ACK, LAST_ACK) \ | |
3171 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, TIME_WAIT, TIME_WAIT) \ | |
3172 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE, CLOSE) \ | |
3173 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT2, SYN_SENT2) \ | |
3174 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, RETRANSMIT, RETRANS) \ | |
3175 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, UNACK, UNACK) | |
3176 | ||
3177 | #define CT_DPIF_NL_TP_UDP_MAPPINGS \ | |
3178 | CT_DPIF_NL_TP_MAPPING(UDP, UDP, SINGLE, UNREPLIED) \ | |
3179 | CT_DPIF_NL_TP_MAPPING(UDP, UDP, MULTIPLE, REPLIED) | |
3180 | ||
3181 | #define CT_DPIF_NL_TP_ICMP_MAPPINGS \ | |
3182 | CT_DPIF_NL_TP_MAPPING(ICMP, ICMP, FIRST, TIMEOUT) | |
3183 | ||
3184 | #define CT_DPIF_NL_TP_ICMPV6_MAPPINGS \ | |
3185 | CT_DPIF_NL_TP_MAPPING(ICMP, ICMPV6, FIRST, TIMEOUT) | |
3186 | ||
3187 | ||
3188 | #define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2) \ | |
3189 | if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) { \ | |
3190 | nl_tp->present |= 1 << CTA_TIMEOUT_##PROTO2##_##ATTR2; \ | |
3191 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2] = \ | |
3192 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]; \ | |
3193 | } | |
3194 | ||
3195 | static void | |
3196 | dpif_netlink_get_nl_tp_tcp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3197 | struct nl_ct_timeout_policy *nl_tp) | |
3198 | { | |
3199 | CT_DPIF_NL_TP_TCP_MAPPINGS | |
3200 | } | |
3201 | ||
3202 | static void | |
3203 | dpif_netlink_get_nl_tp_udp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3204 | struct nl_ct_timeout_policy *nl_tp) | |
3205 | { | |
3206 | CT_DPIF_NL_TP_UDP_MAPPINGS | |
3207 | } | |
3208 | ||
3209 | static void | |
3210 | dpif_netlink_get_nl_tp_icmp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3211 | struct nl_ct_timeout_policy *nl_tp) | |
3212 | { | |
3213 | CT_DPIF_NL_TP_ICMP_MAPPINGS | |
3214 | } | |
3215 | ||
3216 | static void | |
3217 | dpif_netlink_get_nl_tp_icmpv6_attrs(const struct ct_dpif_timeout_policy *tp, | |
3218 | struct nl_ct_timeout_policy *nl_tp) | |
3219 | { | |
3220 | CT_DPIF_NL_TP_ICMPV6_MAPPINGS | |
3221 | } | |
3222 | ||
3223 | #undef CT_DPIF_NL_TP_MAPPING | |
3224 | ||
3225 | static void | |
3226 | dpif_netlink_get_nl_tp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3227 | uint8_t l4num, struct nl_ct_timeout_policy *nl_tp) | |
3228 | { | |
3229 | nl_tp->present = 0; | |
3230 | ||
3231 | if (l4num == IPPROTO_TCP) { | |
3232 | dpif_netlink_get_nl_tp_tcp_attrs(tp, nl_tp); | |
3233 | } else if (l4num == IPPROTO_UDP) { | |
3234 | dpif_netlink_get_nl_tp_udp_attrs(tp, nl_tp); | |
3235 | } else if (l4num == IPPROTO_ICMP) { | |
3236 | dpif_netlink_get_nl_tp_icmp_attrs(tp, nl_tp); | |
3237 | } else if (l4num == IPPROTO_ICMPV6) { | |
3238 | dpif_netlink_get_nl_tp_icmpv6_attrs(tp, nl_tp); | |
3239 | } | |
3240 | } | |
3241 | ||
3242 | #define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2) \ | |
3243 | if (nl_tp->present & (1 << CTA_TIMEOUT_##PROTO2##_##ATTR2)) { \ | |
3244 | if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) { \ | |
3245 | if (tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] != \ | |
3246 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]) { \ | |
3247 | VLOG_WARN_RL(&error_rl, "Inconsistent timeout policy %s " \ | |
3248 | "attribute %s=%"PRIu32" while %s=%"PRIu32, \ | |
3249 | nl_tp->name, "CTA_TIMEOUT_"#PROTO2"_"#ATTR2, \ | |
3250 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2], \ | |
3251 | "CT_DPIF_TP_ATTR_"#PROTO1"_"#ATTR1, \ | |
3252 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]); \ | |
3253 | } \ | |
3254 | } else { \ | |
3255 | tp->present |= 1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1; \ | |
3256 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] = \ | |
3257 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]; \ | |
3258 | } \ | |
3259 | } | |
3260 | ||
3261 | static void | |
3262 | dpif_netlink_set_ct_dpif_tp_tcp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3263 | struct ct_dpif_timeout_policy *tp) | |
3264 | { | |
3265 | CT_DPIF_NL_TP_TCP_MAPPINGS | |
3266 | } | |
3267 | ||
3268 | static void | |
3269 | dpif_netlink_set_ct_dpif_tp_udp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3270 | struct ct_dpif_timeout_policy *tp) | |
3271 | { | |
3272 | CT_DPIF_NL_TP_UDP_MAPPINGS | |
3273 | } | |
3274 | ||
3275 | static void | |
3276 | dpif_netlink_set_ct_dpif_tp_icmp_attrs( | |
3277 | const struct nl_ct_timeout_policy *nl_tp, | |
3278 | struct ct_dpif_timeout_policy *tp) | |
3279 | { | |
3280 | CT_DPIF_NL_TP_ICMP_MAPPINGS | |
3281 | } | |
3282 | ||
3283 | static void | |
3284 | dpif_netlink_set_ct_dpif_tp_icmpv6_attrs( | |
3285 | const struct nl_ct_timeout_policy *nl_tp, | |
3286 | struct ct_dpif_timeout_policy *tp) | |
3287 | { | |
3288 | CT_DPIF_NL_TP_ICMPV6_MAPPINGS | |
3289 | } | |
3290 | ||
3291 | #undef CT_DPIF_NL_TP_MAPPING | |
3292 | ||
3293 | static void | |
3294 | dpif_netlink_set_ct_dpif_tp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3295 | struct ct_dpif_timeout_policy *tp) | |
3296 | { | |
3297 | if (nl_tp->l4num == IPPROTO_TCP) { | |
3298 | dpif_netlink_set_ct_dpif_tp_tcp_attrs(nl_tp, tp); | |
3299 | } else if (nl_tp->l4num == IPPROTO_UDP) { | |
3300 | dpif_netlink_set_ct_dpif_tp_udp_attrs(nl_tp, tp); | |
3301 | } else if (nl_tp->l4num == IPPROTO_ICMP) { | |
3302 | dpif_netlink_set_ct_dpif_tp_icmp_attrs(nl_tp, tp); | |
3303 | } else if (nl_tp->l4num == IPPROTO_ICMPV6) { | |
3304 | dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(nl_tp, tp); | |
3305 | } | |
3306 | } | |
3307 | ||
3308 | #ifdef _WIN32 | |
3309 | static int | |
3310 | dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3311 | const struct ct_dpif_timeout_policy *tp) | |
3312 | { | |
3313 | return EOPNOTSUPP; | |
3314 | } | |
3315 | ||
3316 | static int | |
3317 | dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3318 | uint32_t tp_id, | |
3319 | struct ct_dpif_timeout_policy *tp) | |
3320 | { | |
3321 | return EOPNOTSUPP; | |
3322 | } | |
3323 | ||
3324 | static int | |
3325 | dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3326 | uint32_t tp_id) | |
3327 | { | |
3328 | return EOPNOTSUPP; | |
3329 | } | |
3330 | ||
3331 | static int | |
3332 | dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED, | |
3333 | void **statep) | |
3334 | { | |
3335 | return EOPNOTSUPP; | |
3336 | } | |
3337 | ||
3338 | static int | |
3339 | dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED, | |
3340 | void *state, | |
3341 | struct ct_dpif_timeout_policy **tp) | |
3342 | { | |
3343 | return EOPNOTSUPP; | |
3344 | } | |
3345 | ||
3346 | static int | |
3347 | dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED, | |
3348 | void *state) | |
3349 | { | |
3350 | return EOPNOTSUPP; | |
3351 | } | |
3352 | #else | |
3353 | static int | |
3354 | dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3355 | const struct ct_dpif_timeout_policy *tp) | |
3356 | { | |
1f161318 YHW |
3357 | int err = 0; |
3358 | ||
3359 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f YHW |
3360 | struct nl_ct_timeout_policy nl_tp; |
3361 | char *nl_tp_name; | |
3362 | ||
1f161318 YHW |
3363 | dpif_netlink_format_tp_name(tp->id, tp_protos[i].l3num, |
3364 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f YHW |
3365 | ovs_strlcpy(nl_tp.name, nl_tp_name, sizeof nl_tp.name); |
3366 | free(nl_tp_name); | |
3367 | ||
1f161318 YHW |
3368 | nl_tp.l3num = tp_protos[i].l3num; |
3369 | nl_tp.l4num = tp_protos[i].l4num; | |
3370 | dpif_netlink_get_nl_tp_attrs(tp, tp_protos[i].l4num, &nl_tp); | |
3371 | err = nl_ct_set_timeout_policy(&nl_tp); | |
3372 | if (err) { | |
3373 | VLOG_WARN_RL(&error_rl, "failed to add timeout policy %s (%s)", | |
3374 | nl_tp.name, ovs_strerror(err)); | |
3375 | goto out; | |
3376 | } | |
3377 | } | |
3378 | ||
3379 | out: | |
1f161318 YHW |
3380 | return err; |
3381 | } | |
3382 | ||
3383 | static int | |
3384 | dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3385 | uint32_t tp_id, | |
3386 | struct ct_dpif_timeout_policy *tp) | |
3387 | { | |
1f161318 YHW |
3388 | int err = 0; |
3389 | ||
3390 | tp->id = tp_id; | |
3391 | tp->present = 0; | |
3392 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f YHW |
3393 | struct nl_ct_timeout_policy nl_tp; |
3394 | char *nl_tp_name; | |
3395 | ||
1f161318 YHW |
3396 | dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num, |
3397 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f | 3398 | err = nl_ct_get_timeout_policy(nl_tp_name, &nl_tp); |
1f161318 YHW |
3399 | |
3400 | if (err) { | |
3401 | VLOG_WARN_RL(&error_rl, "failed to get timeout policy %s (%s)", | |
187bb41f YHW |
3402 | nl_tp_name, ovs_strerror(err)); |
3403 | free(nl_tp_name); | |
1f161318 YHW |
3404 | goto out; |
3405 | } | |
187bb41f | 3406 | free(nl_tp_name); |
1f161318 YHW |
3407 | dpif_netlink_set_ct_dpif_tp_attrs(&nl_tp, tp); |
3408 | } | |
3409 | ||
3410 | out: | |
1f161318 YHW |
3411 | return err; |
3412 | } | |
3413 | ||
3414 | /* Returns 0 if all the sub timeout policies are deleted or not exist in the | |
3415 | * kernel. Returns 1 if any sub timeout policy deletion failed. */ | |
3416 | static int | |
3417 | dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3418 | uint32_t tp_id) | |
3419 | { | |
1f161318 YHW |
3420 | int ret = 0; |
3421 | ||
3422 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f | 3423 | char *nl_tp_name; |
1f161318 YHW |
3424 | dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num, |
3425 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f | 3426 | int err = nl_ct_del_timeout_policy(nl_tp_name); |
1f161318 YHW |
3427 | if (err == ENOENT) { |
3428 | err = 0; | |
3429 | } | |
3430 | if (err) { | |
3431 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(6, 6); | |
3432 | VLOG_INFO_RL(&rl, "failed to delete timeout policy %s (%s)", | |
187bb41f | 3433 | nl_tp_name, ovs_strerror(err)); |
1f161318 YHW |
3434 | ret = 1; |
3435 | } | |
187bb41f | 3436 | free(nl_tp_name); |
1f161318 YHW |
3437 | } |
3438 | ||
1f161318 YHW |
3439 | return ret; |
3440 | } | |
3441 | ||
3442 | struct dpif_netlink_ct_timeout_policy_dump_state { | |
3443 | struct nl_ct_timeout_policy_dump_state *nl_dump_state; | |
3444 | struct hmap tp_dump_map; | |
3445 | }; | |
3446 | ||
3447 | struct dpif_netlink_tp_dump_node { | |
3448 | struct hmap_node hmap_node; /* node in tp_dump_map. */ | |
3449 | struct ct_dpif_timeout_policy *tp; | |
3450 | uint32_t l3_l4_present; | |
3451 | }; | |
3452 | ||
3453 | static struct dpif_netlink_tp_dump_node * | |
3454 | get_dpif_netlink_tp_dump_node_by_tp_id(uint32_t tp_id, | |
3455 | struct hmap *tp_dump_map) | |
3456 | { | |
3457 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3458 | ||
3459 | HMAP_FOR_EACH_WITH_HASH (tp_dump_node, hmap_node, hash_int(tp_id, 0), | |
3460 | tp_dump_map) { | |
3461 | if (tp_dump_node->tp->id == tp_id) { | |
3462 | return tp_dump_node; | |
3463 | } | |
3464 | } | |
3465 | return NULL; | |
3466 | } | |
3467 | ||
3468 | static void | |
3469 | update_dpif_netlink_tp_dump_node( | |
3470 | const struct nl_ct_timeout_policy *nl_tp, | |
3471 | struct dpif_netlink_tp_dump_node *tp_dump_node) | |
3472 | { | |
3473 | dpif_netlink_set_ct_dpif_tp_attrs(nl_tp, tp_dump_node->tp); | |
3474 | for (int i = 0; i < DPIF_NL_TP_MAX; ++i) { | |
3475 | if (nl_tp->l3num == tp_protos[i].l3num && | |
3476 | nl_tp->l4num == tp_protos[i].l4num) { | |
3477 | tp_dump_node->l3_l4_present |= 1 << i; | |
3478 | break; | |
3479 | } | |
3480 | } | |
3481 | } | |
3482 | ||
3483 | static int | |
3484 | dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED, | |
3485 | void **statep) | |
3486 | { | |
3487 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state; | |
3488 | ||
3489 | *statep = dump_state = xzalloc(sizeof *dump_state); | |
3490 | int err = nl_ct_timeout_policy_dump_start(&dump_state->nl_dump_state); | |
3491 | if (err) { | |
3492 | free(dump_state); | |
3493 | return err; | |
3494 | } | |
3495 | hmap_init(&dump_state->tp_dump_map); | |
3496 | return 0; | |
3497 | } | |
3498 | ||
3499 | static void | |
3500 | get_and_cleanup_tp_dump_node(struct hmap *hmap, | |
3501 | struct dpif_netlink_tp_dump_node *tp_dump_node, | |
3502 | struct ct_dpif_timeout_policy *tp) | |
3503 | { | |
3504 | hmap_remove(hmap, &tp_dump_node->hmap_node); | |
3505 | *tp = *tp_dump_node->tp; | |
3506 | free(tp_dump_node->tp); | |
3507 | free(tp_dump_node); | |
3508 | } | |
3509 | ||
3510 | static int | |
3511 | dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED, | |
3512 | void *state, | |
3513 | struct ct_dpif_timeout_policy *tp) | |
3514 | { | |
3515 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state; | |
3516 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3517 | int err; | |
3518 | ||
3519 | /* Dumps all the timeout policies in the kernel. */ | |
3520 | do { | |
3521 | struct nl_ct_timeout_policy nl_tp; | |
3522 | uint32_t tp_id; | |
3523 | ||
3524 | err = nl_ct_timeout_policy_dump_next(dump_state->nl_dump_state, | |
3525 | &nl_tp); | |
3526 | if (err) { | |
3527 | break; | |
3528 | } | |
3529 | ||
3530 | /* We only interest in OVS installed timeout policies. */ | |
3531 | if (!ovs_scan(nl_tp.name, NL_TP_NAME_PREFIX"%"PRIu32, &tp_id)) { | |
3532 | continue; | |
3533 | } | |
3534 | ||
3535 | tp_dump_node = get_dpif_netlink_tp_dump_node_by_tp_id( | |
3536 | tp_id, &dump_state->tp_dump_map); | |
3537 | if (!tp_dump_node) { | |
3538 | tp_dump_node = xzalloc(sizeof *tp_dump_node); | |
3539 | tp_dump_node->tp = xzalloc(sizeof *tp_dump_node->tp); | |
3540 | tp_dump_node->tp->id = tp_id; | |
3541 | hmap_insert(&dump_state->tp_dump_map, &tp_dump_node->hmap_node, | |
3542 | hash_int(tp_id, 0)); | |
3543 | } | |
3544 | ||
3545 | update_dpif_netlink_tp_dump_node(&nl_tp, tp_dump_node); | |
3546 | ||
3547 | /* Returns one ct_dpif_timeout_policy if we gather all the L3/L4 | |
3548 | * sub-pieces. */ | |
3549 | if (tp_dump_node->l3_l4_present == DPIF_NL_ALL_TP) { | |
3550 | get_and_cleanup_tp_dump_node(&dump_state->tp_dump_map, | |
3551 | tp_dump_node, tp); | |
3552 | break; | |
3553 | } | |
3554 | } while (true); | |
3555 | ||
3556 | /* Dump the incomplete timeout policies. */ | |
3557 | if (err == EOF) { | |
3558 | if (!hmap_is_empty(&dump_state->tp_dump_map)) { | |
3559 | struct hmap_node *hmap_node = hmap_first(&dump_state->tp_dump_map); | |
3560 | tp_dump_node = CONTAINER_OF(hmap_node, | |
3561 | struct dpif_netlink_tp_dump_node, | |
3562 | hmap_node); | |
3563 | get_and_cleanup_tp_dump_node(&dump_state->tp_dump_map, | |
3564 | tp_dump_node, tp); | |
3565 | return 0; | |
3566 | } | |
3567 | } | |
3568 | ||
3569 | return err; | |
3570 | } | |
3571 | ||
3572 | static int | |
3573 | dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED, | |
3574 | void *state) | |
3575 | { | |
3576 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state; | |
3577 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3578 | ||
3579 | int err = nl_ct_timeout_policy_dump_done(dump_state->nl_dump_state); | |
3580 | HMAP_FOR_EACH_POP (tp_dump_node, hmap_node, &dump_state->tp_dump_map) { | |
3581 | free(tp_dump_node->tp); | |
3582 | free(tp_dump_node); | |
3583 | } | |
3584 | hmap_destroy(&dump_state->tp_dump_map); | |
3585 | free(dump_state); | |
3586 | return err; | |
3587 | } | |
3588 | #endif | |
3589 | ||
5dddf960 JR |
3590 | \f |
3591 | /* Meters */ | |
80738e5f AZ |
3592 | |
3593 | /* Set of supported meter flags */ | |
3594 | #define DP_SUPPORTED_METER_FLAGS_MASK \ | |
3595 | (OFPMF13_STATS | OFPMF13_PKTPS | OFPMF13_KBPS | OFPMF13_BURST) | |
3596 | ||
92d0d515 JP |
3597 | /* Meter support was introduced in Linux 4.15. In some versions of |
3598 | * Linux 4.15, 4.16, and 4.17, there was a bug that never set the id | |
3599 | * when the meter was created, so all meters essentially had an id of | |
3600 | * zero. Check for that condition and disable meters on those kernels. */ | |
3601 | static bool probe_broken_meters(struct dpif *); | |
3602 | ||
5dddf960 | 3603 | static void |
80738e5f AZ |
3604 | dpif_netlink_meter_init(struct dpif_netlink *dpif, struct ofpbuf *buf, |
3605 | void *stub, size_t size, uint32_t command) | |
3606 | { | |
3607 | ofpbuf_use_stub(buf, stub, size); | |
3608 | ||
3609 | nl_msg_put_genlmsghdr(buf, 0, ovs_meter_family, NLM_F_REQUEST | NLM_F_ECHO, | |
3610 | command, OVS_METER_VERSION); | |
3611 | ||
3612 | struct ovs_header *ovs_header; | |
3613 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); | |
3614 | ovs_header->dp_ifindex = dpif->dp_ifindex; | |
3615 | } | |
3616 | ||
3617 | /* Execute meter 'request' in the kernel datapath. If the command | |
3618 | * fails, returns a positive errno value. Otherwise, stores the reply | |
3619 | * in '*replyp', parses the policy according to 'reply_policy' into the | |
3620 | * array of Netlink attribute in 'a', and returns 0. On success, the | |
3621 | * caller is responsible for calling ofpbuf_delete() on '*replyp' | |
3622 | * ('replyp' will contain pointers into 'a'). */ | |
3623 | static int | |
3624 | dpif_netlink_meter_transact(struct ofpbuf *request, struct ofpbuf **replyp, | |
3625 | const struct nl_policy *reply_policy, | |
3626 | struct nlattr **a, size_t size_a) | |
3627 | { | |
3628 | int error = nl_transact(NETLINK_GENERIC, request, replyp); | |
3629 | ofpbuf_uninit(request); | |
3630 | ||
3631 | if (error) { | |
3632 | return error; | |
3633 | } | |
3634 | ||
3635 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(*replyp, sizeof *nlmsg); | |
3636 | struct genlmsghdr *genl = ofpbuf_try_pull(*replyp, sizeof *genl); | |
3637 | struct ovs_header *ovs_header = ofpbuf_try_pull(*replyp, | |
3638 | sizeof *ovs_header); | |
3639 | if (!nlmsg || !genl || !ovs_header | |
3640 | || nlmsg->nlmsg_type != ovs_meter_family | |
3641 | || !nl_policy_parse(*replyp, 0, reply_policy, a, size_a)) { | |
3642 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3643 | VLOG_DBG_RL(&rl, | |
3644 | "Kernel module response to meter tranaction is invalid"); | |
3645 | return EINVAL; | |
3646 | } | |
3647 | return 0; | |
3648 | } | |
3649 | ||
3650 | static void | |
3651 | dpif_netlink_meter_get_features(const struct dpif *dpif_, | |
5dddf960 JR |
3652 | struct ofputil_meter_features *features) |
3653 | { | |
92d0d515 JP |
3654 | if (probe_broken_meters(CONST_CAST(struct dpif *, dpif_))) { |
3655 | features = NULL; | |
3656 | return; | |
3657 | } | |
3658 | ||
80738e5f AZ |
3659 | struct ofpbuf buf, *msg; |
3660 | uint64_t stub[1024 / 8]; | |
3661 | ||
3662 | static const struct nl_policy ovs_meter_features_policy[] = { | |
3663 | [OVS_METER_ATTR_MAX_METERS] = { .type = NL_A_U32 }, | |
3664 | [OVS_METER_ATTR_MAX_BANDS] = { .type = NL_A_U32 }, | |
3665 | [OVS_METER_ATTR_BANDS] = { .type = NL_A_NESTED, .optional = true }, | |
3666 | }; | |
3667 | struct nlattr *a[ARRAY_SIZE(ovs_meter_features_policy)]; | |
3668 | ||
3669 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); | |
3670 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, | |
3671 | OVS_METER_CMD_FEATURES); | |
3672 | if (dpif_netlink_meter_transact(&buf, &msg, ovs_meter_features_policy, a, | |
3673 | ARRAY_SIZE(ovs_meter_features_policy))) { | |
3674 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3675 | VLOG_INFO_RL(&rl, | |
3676 | "dpif_netlink_meter_transact OVS_METER_CMD_FEATURES failed"); | |
3677 | return; | |
3678 | } | |
3679 | ||
3680 | features->max_meters = nl_attr_get_u32(a[OVS_METER_ATTR_MAX_METERS]); | |
3681 | features->max_bands = nl_attr_get_u32(a[OVS_METER_ATTR_MAX_BANDS]); | |
3682 | ||
3683 | /* Bands is a nested attribute of zero or more nested | |
3684 | * band attributes. */ | |
3685 | if (a[OVS_METER_ATTR_BANDS]) { | |
3686 | const struct nlattr *nla; | |
3687 | size_t left; | |
3688 | ||
3689 | NL_NESTED_FOR_EACH (nla, left, a[OVS_METER_ATTR_BANDS]) { | |
3690 | const struct nlattr *band_nla; | |
3691 | size_t band_left; | |
3692 | ||
3693 | NL_NESTED_FOR_EACH (band_nla, band_left, nla) { | |
3694 | if (nl_attr_type(band_nla) == OVS_BAND_ATTR_TYPE) { | |
3695 | if (nl_attr_get_size(band_nla) == sizeof(uint32_t)) { | |
3696 | switch (nl_attr_get_u32(band_nla)) { | |
3697 | case OVS_METER_BAND_TYPE_DROP: | |
3698 | features->band_types |= 1 << OFPMBT13_DROP; | |
3699 | break; | |
3700 | } | |
3701 | } | |
3702 | } | |
3703 | } | |
3704 | } | |
3705 | } | |
3706 | features->capabilities = DP_SUPPORTED_METER_FLAGS_MASK; | |
3707 | ||
3708 | ofpbuf_delete(msg); | |
5dddf960 JR |
3709 | } |
3710 | ||
3711 | static int | |
60ebc04d JP |
3712 | dpif_netlink_meter_set__(struct dpif *dpif_, ofproto_meter_id meter_id, |
3713 | struct ofputil_meter_config *config) | |
5dddf960 | 3714 | { |
80738e5f AZ |
3715 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
3716 | struct ofpbuf buf, *msg; | |
3717 | uint64_t stub[1024 / 8]; | |
3718 | ||
3719 | static const struct nl_policy ovs_meter_set_response_policy[] = { | |
3720 | [OVS_METER_ATTR_ID] = { .type = NL_A_U32 }, | |
3721 | }; | |
3722 | struct nlattr *a[ARRAY_SIZE(ovs_meter_set_response_policy)]; | |
3723 | ||
3724 | if (config->flags & ~DP_SUPPORTED_METER_FLAGS_MASK) { | |
3725 | return EBADF; /* Unsupported flags set */ | |
3726 | } | |
3727 | ||
3728 | for (size_t i = 0; i < config->n_bands; i++) { | |
3729 | switch (config->bands[i].type) { | |
3730 | case OFPMBT13_DROP: | |
3731 | break; | |
3732 | default: | |
3733 | return ENODEV; /* Unsupported band type */ | |
3734 | } | |
3735 | } | |
3736 | ||
3737 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, OVS_METER_CMD_SET); | |
3738 | ||
8101f03f JP |
3739 | nl_msg_put_u32(&buf, OVS_METER_ATTR_ID, meter_id.uint32); |
3740 | ||
80738e5f AZ |
3741 | if (config->flags & OFPMF13_KBPS) { |
3742 | nl_msg_put_flag(&buf, OVS_METER_ATTR_KBPS); | |
3743 | } | |
3744 | ||
3745 | size_t bands_offset = nl_msg_start_nested(&buf, OVS_METER_ATTR_BANDS); | |
3746 | /* Bands */ | |
3747 | for (size_t i = 0; i < config->n_bands; ++i) { | |
3748 | struct ofputil_meter_band * band = &config->bands[i]; | |
3749 | uint32_t band_type; | |
3750 | ||
3751 | size_t band_offset = nl_msg_start_nested(&buf, OVS_BAND_ATTR_UNSPEC); | |
3752 | ||
3753 | switch (band->type) { | |
3754 | case OFPMBT13_DROP: | |
3755 | band_type = OVS_METER_BAND_TYPE_DROP; | |
3756 | break; | |
3757 | default: | |
3758 | band_type = OVS_METER_BAND_TYPE_UNSPEC; | |
3759 | } | |
3760 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_TYPE, band_type); | |
3761 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_RATE, band->rate); | |
3762 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_BURST, | |
3763 | config->flags & OFPMF13_BURST ? | |
3764 | band->burst_size : band->rate); | |
3765 | nl_msg_end_nested(&buf, band_offset); | |
3766 | } | |
3767 | nl_msg_end_nested(&buf, bands_offset); | |
3768 | ||
3769 | int error = dpif_netlink_meter_transact(&buf, &msg, | |
3770 | ovs_meter_set_response_policy, a, | |
3771 | ARRAY_SIZE(ovs_meter_set_response_policy)); | |
3772 | if (error) { | |
3773 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3774 | VLOG_INFO_RL(&rl, | |
3775 | "dpif_netlink_meter_transact OVS_METER_CMD_SET failed"); | |
3776 | return error; | |
3777 | } | |
3778 | ||
8101f03f JP |
3779 | if (nl_attr_get_u32(a[OVS_METER_ATTR_ID]) != meter_id.uint32) { |
3780 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3781 | VLOG_INFO_RL(&rl, | |
3782 | "Kernel returned a different meter id than requested"); | |
3783 | } | |
80738e5f AZ |
3784 | ofpbuf_delete(msg); |
3785 | return 0; | |
5dddf960 JR |
3786 | } |
3787 | ||
60ebc04d JP |
3788 | static int |
3789 | dpif_netlink_meter_set(struct dpif *dpif_, ofproto_meter_id meter_id, | |
3790 | struct ofputil_meter_config *config) | |
3791 | { | |
3792 | if (probe_broken_meters(dpif_)) { | |
3793 | return ENOMEM; | |
3794 | } | |
3795 | ||
3796 | return dpif_netlink_meter_set__(dpif_, meter_id, config); | |
3797 | } | |
3798 | ||
80738e5f AZ |
3799 | /* Retrieve statistics and/or delete meter 'meter_id'. Statistics are |
3800 | * stored in 'stats', if it is not null. If 'command' is | |
3801 | * OVS_METER_CMD_DEL, the meter is deleted and statistics are optionally | |
3802 | * retrieved. If 'command' is OVS_METER_CMD_GET, then statistics are | |
3803 | * simply retrieved. */ | |
5dddf960 | 3804 | static int |
80738e5f AZ |
3805 | dpif_netlink_meter_get_stats(const struct dpif *dpif_, |
3806 | ofproto_meter_id meter_id, | |
3807 | struct ofputil_meter_stats *stats, | |
3808 | uint16_t max_bands, | |
3809 | enum ovs_meter_cmd command) | |
5dddf960 | 3810 | { |
80738e5f AZ |
3811 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
3812 | struct ofpbuf buf, *msg; | |
3813 | uint64_t stub[1024 / 8]; | |
3814 | ||
3815 | static const struct nl_policy ovs_meter_stats_policy[] = { | |
3816 | [OVS_METER_ATTR_ID] = { .type = NL_A_U32, .optional = true}, | |
3817 | [OVS_METER_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), | |
3818 | .optional = true}, | |
3819 | [OVS_METER_ATTR_BANDS] = { .type = NL_A_NESTED, .optional = true }, | |
3820 | }; | |
3821 | struct nlattr *a[ARRAY_SIZE(ovs_meter_stats_policy)]; | |
3822 | ||
3823 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, command); | |
3824 | ||
3825 | nl_msg_put_u32(&buf, OVS_METER_ATTR_ID, meter_id.uint32); | |
3826 | ||
3827 | int error = dpif_netlink_meter_transact(&buf, &msg, | |
3828 | ovs_meter_stats_policy, a, | |
3829 | ARRAY_SIZE(ovs_meter_stats_policy)); | |
3830 | if (error) { | |
3831 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3832 | VLOG_INFO_RL(&rl, "dpif_netlink_meter_transact %s failed", | |
3833 | command == OVS_METER_CMD_GET ? "get" : "del"); | |
3834 | return error; | |
3835 | } | |
3836 | ||
3837 | if (stats | |
3838 | && a[OVS_METER_ATTR_ID] | |
3839 | && a[OVS_METER_ATTR_STATS] | |
3840 | && nl_attr_get_u32(a[OVS_METER_ATTR_ID]) == meter_id.uint32) { | |
3841 | /* return stats */ | |
3842 | const struct ovs_flow_stats *stat; | |
3843 | const struct nlattr *nla; | |
3844 | size_t left; | |
3845 | ||
3846 | stat = nl_attr_get(a[OVS_METER_ATTR_STATS]); | |
3847 | stats->packet_in_count = get_32aligned_u64(&stat->n_packets); | |
3848 | stats->byte_in_count = get_32aligned_u64(&stat->n_bytes); | |
3849 | ||
3850 | if (a[OVS_METER_ATTR_BANDS]) { | |
3851 | size_t n_bands = 0; | |
3852 | NL_NESTED_FOR_EACH (nla, left, a[OVS_METER_ATTR_BANDS]) { | |
3853 | const struct nlattr *band_nla; | |
3854 | band_nla = nl_attr_find_nested(nla, OVS_BAND_ATTR_STATS); | |
3855 | if (band_nla && nl_attr_get_size(band_nla) \ | |
3856 | == sizeof(struct ovs_flow_stats)) { | |
3857 | stat = nl_attr_get(band_nla); | |
3858 | ||
3859 | if (n_bands < max_bands) { | |
3860 | stats->bands[n_bands].packet_count | |
3861 | = get_32aligned_u64(&stat->n_packets); | |
3862 | stats->bands[n_bands].byte_count | |
3863 | = get_32aligned_u64(&stat->n_bytes); | |
3864 | ++n_bands; | |
3865 | } | |
3866 | } else { | |
3867 | stats->bands[n_bands].packet_count = 0; | |
3868 | stats->bands[n_bands].byte_count = 0; | |
3869 | ++n_bands; | |
3870 | } | |
3871 | } | |
3872 | stats->n_bands = n_bands; | |
3873 | } else { | |
3874 | /* For a non-existent meter, return 0 stats. */ | |
3875 | stats->n_bands = 0; | |
3876 | } | |
3877 | } | |
3878 | ||
3879 | ofpbuf_delete(msg); | |
3880 | return error; | |
5dddf960 JR |
3881 | } |
3882 | ||
3883 | static int | |
80738e5f AZ |
3884 | dpif_netlink_meter_get(const struct dpif *dpif, ofproto_meter_id meter_id, |
3885 | struct ofputil_meter_stats *stats, uint16_t max_bands) | |
5dddf960 | 3886 | { |
80738e5f AZ |
3887 | return dpif_netlink_meter_get_stats(dpif, meter_id, stats, max_bands, |
3888 | OVS_METER_CMD_GET); | |
3889 | } | |
3890 | ||
3891 | static int | |
3892 | dpif_netlink_meter_del(struct dpif *dpif, ofproto_meter_id meter_id, | |
3893 | struct ofputil_meter_stats *stats, uint16_t max_bands) | |
3894 | { | |
3895 | return dpif_netlink_meter_get_stats(dpif, meter_id, stats, max_bands, | |
3896 | OVS_METER_CMD_DEL); | |
5dddf960 JR |
3897 | } |
3898 | ||
92d0d515 JP |
3899 | static bool |
3900 | probe_broken_meters__(struct dpif *dpif) | |
3901 | { | |
3902 | /* This test is destructive if a probe occurs while ovs-vswitchd is | |
3903 | * running (e.g., an ovs-dpctl meter command is called), so choose a | |
3904 | * random high meter id to make this less likely to occur. */ | |
3905 | ofproto_meter_id id1 = { 54545401 }; | |
3906 | ofproto_meter_id id2 = { 54545402 }; | |
3907 | struct ofputil_meter_band band = {OFPMBT13_DROP, 0, 1, 0}; | |
3908 | struct ofputil_meter_config config1 = { 1, OFPMF13_KBPS, 1, &band}; | |
3909 | struct ofputil_meter_config config2 = { 2, OFPMF13_KBPS, 1, &band}; | |
3910 | ||
3911 | /* Try adding two meters and make sure that they both come back with | |
60ebc04d JP |
3912 | * the proper meter id. Use the "__" version so that we don't cause |
3913 | * a recurve deadlock. */ | |
3914 | dpif_netlink_meter_set__(dpif, id1, &config1); | |
3915 | dpif_netlink_meter_set__(dpif, id2, &config2); | |
92d0d515 JP |
3916 | |
3917 | if (dpif_netlink_meter_get(dpif, id1, NULL, 0) | |
3918 | || dpif_netlink_meter_get(dpif, id2, NULL, 0)) { | |
3919 | VLOG_INFO("The kernel module has a broken meter implementation."); | |
3920 | return true; | |
3921 | } | |
3922 | ||
3923 | dpif_netlink_meter_del(dpif, id1, NULL, 0); | |
3924 | dpif_netlink_meter_del(dpif, id2, NULL, 0); | |
3925 | ||
3926 | return false; | |
3927 | } | |
3928 | ||
3929 | static bool | |
3930 | probe_broken_meters(struct dpif *dpif) | |
3931 | { | |
3932 | /* This is a once-only test because currently OVS only has at most a single | |
3933 | * Netlink capable datapath on any given platform. */ | |
3934 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3935 | ||
3936 | static bool broken_meters = false; | |
3937 | if (ovsthread_once_start(&once)) { | |
3938 | broken_meters = probe_broken_meters__(dpif); | |
3939 | ovsthread_once_done(&once); | |
3940 | } | |
3941 | return broken_meters; | |
3942 | } | |
5dddf960 | 3943 | \f |
93451a0a | 3944 | const struct dpif_class dpif_netlink_class = { |
1a6f1e2a | 3945 | "system", |
f87c1357 | 3946 | false, /* cleanup_required */ |
c8973eb6 | 3947 | NULL, /* init */ |
93451a0a | 3948 | dpif_netlink_enumerate, |
0aeaabc8 | 3949 | NULL, |
93451a0a AS |
3950 | dpif_netlink_open, |
3951 | dpif_netlink_close, | |
3952 | dpif_netlink_destroy, | |
3953 | dpif_netlink_run, | |
e4516b20 | 3954 | NULL, /* wait */ |
93451a0a | 3955 | dpif_netlink_get_stats, |
dcdcad68 | 3956 | dpif_netlink_set_features, |
93451a0a AS |
3957 | dpif_netlink_port_add, |
3958 | dpif_netlink_port_del, | |
91364d18 | 3959 | NULL, /* port_set_config */ |
93451a0a AS |
3960 | dpif_netlink_port_query_by_number, |
3961 | dpif_netlink_port_query_by_name, | |
3962 | dpif_netlink_port_get_pid, | |
3963 | dpif_netlink_port_dump_start, | |
3964 | dpif_netlink_port_dump_next, | |
3965 | dpif_netlink_port_dump_done, | |
3966 | dpif_netlink_port_poll, | |
3967 | dpif_netlink_port_poll_wait, | |
3968 | dpif_netlink_flow_flush, | |
3969 | dpif_netlink_flow_dump_create, | |
3970 | dpif_netlink_flow_dump_destroy, | |
3971 | dpif_netlink_flow_dump_thread_create, | |
3972 | dpif_netlink_flow_dump_thread_destroy, | |
3973 | dpif_netlink_flow_dump_next, | |
3974 | dpif_netlink_operate, | |
3975 | dpif_netlink_recv_set, | |
3976 | dpif_netlink_handlers_set, | |
d4f6865c | 3977 | NULL, /* set_config */ |
93451a0a AS |
3978 | dpif_netlink_queue_to_priority, |
3979 | dpif_netlink_recv, | |
3980 | dpif_netlink_recv_wait, | |
3981 | dpif_netlink_recv_purge, | |
e4e74c3a | 3982 | NULL, /* register_dp_purge_cb */ |
6b31e073 RW |
3983 | NULL, /* register_upcall_cb */ |
3984 | NULL, /* enable_upcall */ | |
3985 | NULL, /* disable_upcall */ | |
b5cbbcf6 | 3986 | dpif_netlink_get_datapath_version, /* get_datapath_version */ |
c11c9f4a DDP |
3987 | dpif_netlink_ct_dump_start, |
3988 | dpif_netlink_ct_dump_next, | |
3989 | dpif_netlink_ct_dump_done, | |
5dddf960 | 3990 | dpif_netlink_ct_flush, |
c92339ad DB |
3991 | NULL, /* ct_set_maxconns */ |
3992 | NULL, /* ct_get_maxconns */ | |
875075b3 | 3993 | NULL, /* ct_get_nconns */ |
64207120 DB |
3994 | NULL, /* ct_set_tcp_seq_chk */ |
3995 | NULL, /* ct_get_tcp_seq_chk */ | |
906ff9d2 YHW |
3996 | dpif_netlink_ct_set_limits, |
3997 | dpif_netlink_ct_get_limits, | |
3998 | dpif_netlink_ct_del_limits, | |
1f161318 YHW |
3999 | dpif_netlink_ct_set_timeout_policy, |
4000 | dpif_netlink_ct_get_timeout_policy, | |
4001 | dpif_netlink_ct_del_timeout_policy, | |
4002 | dpif_netlink_ct_timeout_policy_dump_start, | |
4003 | dpif_netlink_ct_timeout_policy_dump_next, | |
4004 | dpif_netlink_ct_timeout_policy_dump_done, | |
187bb41f | 4005 | dpif_netlink_ct_get_timeout_policy_name, |
4ea96698 DB |
4006 | NULL, /* ipf_set_enabled */ |
4007 | NULL, /* ipf_set_min_frag */ | |
4008 | NULL, /* ipf_set_max_nfrags */ | |
4009 | NULL, /* ipf_get_status */ | |
4010 | NULL, /* ipf_dump_start */ | |
4011 | NULL, /* ipf_dump_next */ | |
4012 | NULL, /* ipf_dump_done */ | |
5dddf960 JR |
4013 | dpif_netlink_meter_get_features, |
4014 | dpif_netlink_meter_set, | |
4015 | dpif_netlink_meter_get, | |
4016 | dpif_netlink_meter_del, | |
9df65060 VDA |
4017 | NULL, /* bond_add */ |
4018 | NULL, /* bond_del */ | |
4019 | NULL, /* bond_stats_get */ | |
96fba48f | 4020 | }; |
93451a0a | 4021 | |
96fba48f | 4022 | static int |
93451a0a | 4023 | dpif_netlink_init(void) |
96fba48f | 4024 | { |
eb8ed438 BP |
4025 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
4026 | static int error; | |
982b8810 | 4027 | |
eb8ed438 | 4028 | if (ovsthread_once_start(&once)) { |
df2c07f4 JP |
4029 | error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY, |
4030 | &ovs_datapath_family); | |
37a1300c | 4031 | if (error) { |
e0e2410d | 4032 | VLOG_INFO("Generic Netlink family '%s' does not exist. " |
cae7529c CL |
4033 | "The Open vSwitch kernel module is probably not loaded.", |
4034 | OVS_DATAPATH_FAMILY); | |
37a1300c | 4035 | } |
f0fef760 | 4036 | if (!error) { |
df2c07f4 | 4037 | error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family); |
f0fef760 | 4038 | } |
37a1300c | 4039 | if (!error) { |
df2c07f4 | 4040 | error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family); |
37a1300c | 4041 | } |
aaff4b55 | 4042 | if (!error) { |
df2c07f4 JP |
4043 | error = nl_lookup_genl_family(OVS_PACKET_FAMILY, |
4044 | &ovs_packet_family); | |
aaff4b55 | 4045 | } |
c7178a0b EJ |
4046 | if (!error) { |
4047 | error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP, | |
b3dcb73c | 4048 | &ovs_vport_mcgroup); |
c7178a0b | 4049 | } |
80738e5f AZ |
4050 | if (!error) { |
4051 | if (nl_lookup_genl_family(OVS_METER_FAMILY, &ovs_meter_family)) { | |
4052 | VLOG_INFO("The kernel module does not support meters."); | |
4053 | } | |
4054 | } | |
906ff9d2 YHW |
4055 | if (nl_lookup_genl_family(OVS_CT_LIMIT_FAMILY, |
4056 | &ovs_ct_limit_family) < 0) { | |
4057 | VLOG_INFO("Generic Netlink family '%s' does not exist. " | |
4058 | "Please update the Open vSwitch kernel module to enable " | |
4059 | "the conntrack limit feature.", OVS_CT_LIMIT_FAMILY); | |
4060 | } | |
eb8ed438 | 4061 | |
921c370a EG |
4062 | ovs_tunnels_out_of_tree = dpif_netlink_rtnl_probe_oot_tunnels(); |
4063 | ||
eb8ed438 | 4064 | ovsthread_once_done(&once); |
982b8810 BP |
4065 | } |
4066 | ||
4067 | return error; | |
96fba48f BP |
4068 | } |
4069 | ||
c19e6535 | 4070 | bool |
93451a0a | 4071 | dpif_netlink_is_internal_device(const char *name) |
9fe3b9a2 | 4072 | { |
93451a0a | 4073 | struct dpif_netlink_vport reply; |
c19e6535 | 4074 | struct ofpbuf *buf; |
9fe3b9a2 | 4075 | int error; |
96fba48f | 4076 | |
93451a0a | 4077 | error = dpif_netlink_vport_get(name, &reply, &buf); |
c19e6535 BP |
4078 | if (!error) { |
4079 | ofpbuf_delete(buf); | |
141d9ce4 | 4080 | } else if (error != ENODEV && error != ENOENT) { |
c19e6535 | 4081 | VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)", |
10a89ef0 | 4082 | name, ovs_strerror(error)); |
96fba48f BP |
4083 | } |
4084 | ||
df2c07f4 | 4085 | return reply.type == OVS_VPORT_TYPE_INTERNAL; |
96fba48f | 4086 | } |
e0467f6d | 4087 | |
df2c07f4 | 4088 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
c19e6535 BP |
4089 | * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a |
4090 | * positive errno value. | |
4091 | * | |
4092 | * 'vport' will contain pointers into 'buf', so the caller should not free | |
4093 | * 'buf' while 'vport' is still in use. */ | |
4094 | static int | |
93451a0a | 4095 | dpif_netlink_vport_from_ofpbuf(struct dpif_netlink_vport *vport, |
c19e6535 BP |
4096 | const struct ofpbuf *buf) |
4097 | { | |
df2c07f4 JP |
4098 | static const struct nl_policy ovs_vport_policy[] = { |
4099 | [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 }, | |
4100 | [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 }, | |
4101 | [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, | |
1579cf67 | 4102 | [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_UNSPEC }, |
f7df9823 | 4103 | [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats), |
c19e6535 | 4104 | .optional = true }, |
df2c07f4 | 4105 | [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, |
bfda5239 | 4106 | [OVS_VPORT_ATTR_NETNSID] = { .type = NL_A_U32, .optional = true }, |
c19e6535 BP |
4107 | }; |
4108 | ||
93451a0a | 4109 | dpif_netlink_vport_init(vport); |
c19e6535 | 4110 | |
0a2869d5 BP |
4111 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4112 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4113 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4114 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4115 | ||
4116 | struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)]; | |
df2c07f4 JP |
4117 | if (!nlmsg || !genl || !ovs_header |
4118 | || nlmsg->nlmsg_type != ovs_vport_family | |
4119 | || !nl_policy_parse(&b, 0, ovs_vport_policy, a, | |
4120 | ARRAY_SIZE(ovs_vport_policy))) { | |
c19e6535 BP |
4121 | return EINVAL; |
4122 | } | |
c19e6535 | 4123 | |
f0fef760 | 4124 | vport->cmd = genl->cmd; |
df2c07f4 | 4125 | vport->dp_ifindex = ovs_header->dp_ifindex; |
4e022ec0 | 4126 | vport->port_no = nl_attr_get_odp_port(a[OVS_VPORT_ATTR_PORT_NO]); |
df2c07f4 JP |
4127 | vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]); |
4128 | vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]); | |
b063d9f0 | 4129 | if (a[OVS_VPORT_ATTR_UPCALL_PID]) { |
1579cf67 AW |
4130 | vport->n_upcall_pids = nl_attr_get_size(a[OVS_VPORT_ATTR_UPCALL_PID]) |
4131 | / (sizeof *vport->upcall_pids); | |
4132 | vport->upcall_pids = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]); | |
4133 | ||
b063d9f0 | 4134 | } |
df2c07f4 JP |
4135 | if (a[OVS_VPORT_ATTR_STATS]) { |
4136 | vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); | |
4137 | } | |
df2c07f4 JP |
4138 | if (a[OVS_VPORT_ATTR_OPTIONS]) { |
4139 | vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]); | |
4140 | vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]); | |
c19e6535 | 4141 | } |
bfda5239 FL |
4142 | if (a[OVS_VPORT_ATTR_NETNSID]) { |
4143 | netnsid_set(&vport->netnsid, | |
4144 | nl_attr_get_u32(a[OVS_VPORT_ATTR_NETNSID])); | |
4145 | } else { | |
4146 | netnsid_set_local(&vport->netnsid); | |
4147 | } | |
c19e6535 BP |
4148 | return 0; |
4149 | } | |
4150 | ||
df2c07f4 | 4151 | /* Appends to 'buf' (which must initially be empty) a "struct ovs_header" |
c19e6535 BP |
4152 | * followed by Netlink attributes corresponding to 'vport'. */ |
4153 | static void | |
93451a0a AS |
4154 | dpif_netlink_vport_to_ofpbuf(const struct dpif_netlink_vport *vport, |
4155 | struct ofpbuf *buf) | |
c19e6535 | 4156 | { |
df2c07f4 | 4157 | struct ovs_header *ovs_header; |
f0fef760 | 4158 | |
df2c07f4 | 4159 | nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO, |
69685a88 | 4160 | vport->cmd, OVS_VPORT_VERSION); |
c19e6535 | 4161 | |
df2c07f4 JP |
4162 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4163 | ovs_header->dp_ifindex = vport->dp_ifindex; | |
c19e6535 | 4164 | |
4e022ec0 AW |
4165 | if (vport->port_no != ODPP_NONE) { |
4166 | nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no); | |
c19e6535 BP |
4167 | } |
4168 | ||
df2c07f4 JP |
4169 | if (vport->type != OVS_VPORT_TYPE_UNSPEC) { |
4170 | nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type); | |
c19e6535 BP |
4171 | } |
4172 | ||
4173 | if (vport->name) { | |
df2c07f4 | 4174 | nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name); |
c19e6535 BP |
4175 | } |
4176 | ||
1579cf67 AW |
4177 | if (vport->upcall_pids) { |
4178 | nl_msg_put_unspec(buf, OVS_VPORT_ATTR_UPCALL_PID, | |
4179 | vport->upcall_pids, | |
4180 | vport->n_upcall_pids * sizeof *vport->upcall_pids); | |
a24a6574 | 4181 | } |
b063d9f0 | 4182 | |
c19e6535 | 4183 | if (vport->stats) { |
df2c07f4 | 4184 | nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS, |
c19e6535 BP |
4185 | vport->stats, sizeof *vport->stats); |
4186 | } | |
4187 | ||
c19e6535 | 4188 | if (vport->options) { |
df2c07f4 | 4189 | nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, |
c19e6535 BP |
4190 | vport->options, vport->options_len); |
4191 | } | |
c19e6535 BP |
4192 | } |
4193 | ||
4194 | /* Clears 'vport' to "empty" values. */ | |
4195 | void | |
93451a0a | 4196 | dpif_netlink_vport_init(struct dpif_netlink_vport *vport) |
c19e6535 BP |
4197 | { |
4198 | memset(vport, 0, sizeof *vport); | |
4e022ec0 | 4199 | vport->port_no = ODPP_NONE; |
c19e6535 BP |
4200 | } |
4201 | ||
4202 | /* Executes 'request' in the kernel datapath. If the command fails, returns a | |
4203 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4204 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
df2c07f4 | 4205 | * result of the command is expected to be an ovs_vport also, which is decoded |
c19e6535 BP |
4206 | * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the |
4207 | * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
4208 | int | |
93451a0a AS |
4209 | dpif_netlink_vport_transact(const struct dpif_netlink_vport *request, |
4210 | struct dpif_netlink_vport *reply, | |
4211 | struct ofpbuf **bufp) | |
c19e6535 | 4212 | { |
f0fef760 | 4213 | struct ofpbuf *request_buf; |
c19e6535 BP |
4214 | int error; |
4215 | ||
cb22974d | 4216 | ovs_assert((reply != NULL) == (bufp != NULL)); |
c19e6535 | 4217 | |
93451a0a | 4218 | error = dpif_netlink_init(); |
42bb6c72 BP |
4219 | if (error) { |
4220 | if (reply) { | |
4221 | *bufp = NULL; | |
93451a0a | 4222 | dpif_netlink_vport_init(reply); |
42bb6c72 BP |
4223 | } |
4224 | return error; | |
4225 | } | |
4226 | ||
f0fef760 | 4227 | request_buf = ofpbuf_new(1024); |
93451a0a | 4228 | dpif_netlink_vport_to_ofpbuf(request, request_buf); |
a88b4e04 | 4229 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
f0fef760 | 4230 | ofpbuf_delete(request_buf); |
c19e6535 | 4231 | |
f0fef760 BP |
4232 | if (reply) { |
4233 | if (!error) { | |
93451a0a | 4234 | error = dpif_netlink_vport_from_ofpbuf(reply, *bufp); |
f0fef760 | 4235 | } |
c19e6535 | 4236 | if (error) { |
93451a0a | 4237 | dpif_netlink_vport_init(reply); |
f0fef760 BP |
4238 | ofpbuf_delete(*bufp); |
4239 | *bufp = NULL; | |
c19e6535 | 4240 | } |
c19e6535 BP |
4241 | } |
4242 | return error; | |
4243 | } | |
4244 | ||
4245 | /* Obtains information about the kernel vport named 'name' and stores it into | |
4246 | * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no | |
4247 | * longer needed ('reply' will contain pointers into '*bufp'). */ | |
4248 | int | |
93451a0a AS |
4249 | dpif_netlink_vport_get(const char *name, struct dpif_netlink_vport *reply, |
4250 | struct ofpbuf **bufp) | |
c19e6535 | 4251 | { |
93451a0a | 4252 | struct dpif_netlink_vport request; |
c19e6535 | 4253 | |
93451a0a | 4254 | dpif_netlink_vport_init(&request); |
df2c07f4 | 4255 | request.cmd = OVS_VPORT_CMD_GET; |
c19e6535 BP |
4256 | request.name = name; |
4257 | ||
93451a0a | 4258 | return dpif_netlink_vport_transact(&request, reply, bufp); |
c19e6535 | 4259 | } |
93451a0a | 4260 | |
df2c07f4 | 4261 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
aaff4b55 BP |
4262 | * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a |
4263 | * positive errno value. | |
d6569377 BP |
4264 | * |
4265 | * 'dp' will contain pointers into 'buf', so the caller should not free 'buf' | |
4266 | * while 'dp' is still in use. */ | |
4267 | static int | |
93451a0a | 4268 | dpif_netlink_dp_from_ofpbuf(struct dpif_netlink_dp *dp, const struct ofpbuf *buf) |
d6569377 | 4269 | { |
df2c07f4 JP |
4270 | static const struct nl_policy ovs_datapath_policy[] = { |
4271 | [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, | |
f7df9823 | 4272 | [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats), |
d6569377 | 4273 | .optional = true }, |
847108dc AZ |
4274 | [OVS_DP_ATTR_MEGAFLOW_STATS] = { |
4275 | NL_POLICY_FOR(struct ovs_dp_megaflow_stats), | |
4276 | .optional = true }, | |
dcdcad68 PB |
4277 | [OVS_DP_ATTR_USER_FEATURES] = { |
4278 | .type = NL_A_U32, | |
4279 | .optional = true }, | |
d6569377 BP |
4280 | }; |
4281 | ||
93451a0a | 4282 | dpif_netlink_dp_init(dp); |
d6569377 | 4283 | |
0a2869d5 BP |
4284 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4285 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4286 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4287 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4288 | ||
4289 | struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; | |
df2c07f4 JP |
4290 | if (!nlmsg || !genl || !ovs_header |
4291 | || nlmsg->nlmsg_type != ovs_datapath_family | |
4292 | || !nl_policy_parse(&b, 0, ovs_datapath_policy, a, | |
4293 | ARRAY_SIZE(ovs_datapath_policy))) { | |
d6569377 BP |
4294 | return EINVAL; |
4295 | } | |
d6569377 | 4296 | |
aaff4b55 | 4297 | dp->cmd = genl->cmd; |
df2c07f4 JP |
4298 | dp->dp_ifindex = ovs_header->dp_ifindex; |
4299 | dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]); | |
4300 | if (a[OVS_DP_ATTR_STATS]) { | |
6a54dedc | 4301 | dp->stats = nl_attr_get(a[OVS_DP_ATTR_STATS]); |
d6569377 | 4302 | } |
982b8810 | 4303 | |
847108dc | 4304 | if (a[OVS_DP_ATTR_MEGAFLOW_STATS]) { |
6a54dedc | 4305 | dp->megaflow_stats = nl_attr_get(a[OVS_DP_ATTR_MEGAFLOW_STATS]); |
847108dc AZ |
4306 | } |
4307 | ||
dcdcad68 PB |
4308 | if (a[OVS_DP_ATTR_USER_FEATURES]) { |
4309 | dp->user_features = nl_attr_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); | |
4310 | } | |
4311 | ||
d6569377 BP |
4312 | return 0; |
4313 | } | |
4314 | ||
aaff4b55 | 4315 | /* Appends to 'buf' the Generic Netlink message described by 'dp'. */ |
d6569377 | 4316 | static void |
93451a0a | 4317 | dpif_netlink_dp_to_ofpbuf(const struct dpif_netlink_dp *dp, struct ofpbuf *buf) |
d6569377 | 4318 | { |
df2c07f4 | 4319 | struct ovs_header *ovs_header; |
d6569377 | 4320 | |
df2c07f4 | 4321 | nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family, |
69685a88 JG |
4322 | NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, |
4323 | OVS_DATAPATH_VERSION); | |
aaff4b55 | 4324 | |
df2c07f4 JP |
4325 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4326 | ovs_header->dp_ifindex = dp->dp_ifindex; | |
d6569377 BP |
4327 | |
4328 | if (dp->name) { | |
df2c07f4 | 4329 | nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name); |
d6569377 BP |
4330 | } |
4331 | ||
a24a6574 BP |
4332 | if (dp->upcall_pid) { |
4333 | nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid); | |
4334 | } | |
b063d9f0 | 4335 | |
b7fd5e38 TG |
4336 | if (dp->user_features) { |
4337 | nl_msg_put_u32(buf, OVS_DP_ATTR_USER_FEATURES, dp->user_features); | |
4338 | } | |
4339 | ||
df2c07f4 | 4340 | /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ |
d6569377 BP |
4341 | } |
4342 | ||
4343 | /* Clears 'dp' to "empty" values. */ | |
d3d8f1f7 | 4344 | static void |
93451a0a | 4345 | dpif_netlink_dp_init(struct dpif_netlink_dp *dp) |
d6569377 BP |
4346 | { |
4347 | memset(dp, 0, sizeof *dp); | |
d6569377 BP |
4348 | } |
4349 | ||
aaff4b55 | 4350 | static void |
93451a0a | 4351 | dpif_netlink_dp_dump_start(struct nl_dump *dump) |
aaff4b55 | 4352 | { |
93451a0a | 4353 | struct dpif_netlink_dp request; |
aaff4b55 BP |
4354 | struct ofpbuf *buf; |
4355 | ||
93451a0a | 4356 | dpif_netlink_dp_init(&request); |
df2c07f4 | 4357 | request.cmd = OVS_DP_CMD_GET; |
aaff4b55 BP |
4358 | |
4359 | buf = ofpbuf_new(1024); | |
93451a0a | 4360 | dpif_netlink_dp_to_ofpbuf(&request, buf); |
a88b4e04 | 4361 | nl_dump_start(dump, NETLINK_GENERIC, buf); |
aaff4b55 BP |
4362 | ofpbuf_delete(buf); |
4363 | } | |
4364 | ||
d6569377 BP |
4365 | /* Executes 'request' in the kernel datapath. If the command fails, returns a |
4366 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4367 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
aaff4b55 BP |
4368 | * result of the command is expected to be of the same form, which is decoded |
4369 | * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the | |
4370 | * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4371 | static int |
93451a0a AS |
4372 | dpif_netlink_dp_transact(const struct dpif_netlink_dp *request, |
4373 | struct dpif_netlink_dp *reply, struct ofpbuf **bufp) | |
d6569377 | 4374 | { |
aaff4b55 | 4375 | struct ofpbuf *request_buf; |
d6569377 | 4376 | int error; |
d6569377 | 4377 | |
cb22974d | 4378 | ovs_assert((reply != NULL) == (bufp != NULL)); |
d6569377 | 4379 | |
aaff4b55 | 4380 | request_buf = ofpbuf_new(1024); |
93451a0a | 4381 | dpif_netlink_dp_to_ofpbuf(request, request_buf); |
a88b4e04 | 4382 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
aaff4b55 | 4383 | ofpbuf_delete(request_buf); |
d6569377 | 4384 | |
aaff4b55 | 4385 | if (reply) { |
93451a0a | 4386 | dpif_netlink_dp_init(reply); |
aaff4b55 | 4387 | if (!error) { |
93451a0a | 4388 | error = dpif_netlink_dp_from_ofpbuf(reply, *bufp); |
aaff4b55 | 4389 | } |
d6569377 | 4390 | if (error) { |
aaff4b55 BP |
4391 | ofpbuf_delete(*bufp); |
4392 | *bufp = NULL; | |
d6569377 | 4393 | } |
d6569377 BP |
4394 | } |
4395 | return error; | |
4396 | } | |
4397 | ||
4398 | /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. | |
4399 | * The caller must free '*bufp' when the reply is no longer needed ('reply' | |
4400 | * will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4401 | static int |
93451a0a AS |
4402 | dpif_netlink_dp_get(const struct dpif *dpif_, struct dpif_netlink_dp *reply, |
4403 | struct ofpbuf **bufp) | |
d6569377 | 4404 | { |
93451a0a AS |
4405 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
4406 | struct dpif_netlink_dp request; | |
d6569377 | 4407 | |
93451a0a | 4408 | dpif_netlink_dp_init(&request); |
df2c07f4 | 4409 | request.cmd = OVS_DP_CMD_GET; |
254f2dc8 | 4410 | request.dp_ifindex = dpif->dp_ifindex; |
d6569377 | 4411 | |
93451a0a | 4412 | return dpif_netlink_dp_transact(&request, reply, bufp); |
d6569377 | 4413 | } |
93451a0a | 4414 | |
df2c07f4 | 4415 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
37a1300c | 4416 | * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a |
d6569377 BP |
4417 | * positive errno value. |
4418 | * | |
4419 | * 'flow' will contain pointers into 'buf', so the caller should not free 'buf' | |
4420 | * while 'flow' is still in use. */ | |
4421 | static int | |
93451a0a AS |
4422 | dpif_netlink_flow_from_ofpbuf(struct dpif_netlink_flow *flow, |
4423 | const struct ofpbuf *buf) | |
d6569377 | 4424 | { |
70e5ed6f JS |
4425 | static const struct nl_policy ovs_flow_policy[__OVS_FLOW_ATTR_MAX] = { |
4426 | [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED, .optional = true }, | |
e6cc0bab | 4427 | [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true }, |
df2c07f4 | 4428 | [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, |
f7df9823 | 4429 | [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), |
d6569377 | 4430 | .optional = true }, |
df2c07f4 JP |
4431 | [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, |
4432 | [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, | |
ab79d262 | 4433 | [OVS_FLOW_ATTR_UFID] = { .type = NL_A_U128, .optional = true }, |
df2c07f4 | 4434 | /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */ |
43f9ac0a | 4435 | /* The kernel never uses OVS_FLOW_ATTR_PROBE. */ |
70e5ed6f | 4436 | /* The kernel never uses OVS_FLOW_ATTR_UFID_FLAGS. */ |
d6569377 BP |
4437 | }; |
4438 | ||
93451a0a | 4439 | dpif_netlink_flow_init(flow); |
d6569377 | 4440 | |
0a2869d5 BP |
4441 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4442 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4443 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4444 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4445 | ||
4446 | struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)]; | |
df2c07f4 JP |
4447 | if (!nlmsg || !genl || !ovs_header |
4448 | || nlmsg->nlmsg_type != ovs_flow_family | |
4449 | || !nl_policy_parse(&b, 0, ovs_flow_policy, a, | |
4450 | ARRAY_SIZE(ovs_flow_policy))) { | |
d6569377 BP |
4451 | return EINVAL; |
4452 | } | |
70e5ed6f JS |
4453 | if (!a[OVS_FLOW_ATTR_KEY] && !a[OVS_FLOW_ATTR_UFID]) { |
4454 | return EINVAL; | |
4455 | } | |
d6569377 | 4456 | |
37a1300c | 4457 | flow->nlmsg_flags = nlmsg->nlmsg_flags; |
df2c07f4 | 4458 | flow->dp_ifindex = ovs_header->dp_ifindex; |
70e5ed6f JS |
4459 | if (a[OVS_FLOW_ATTR_KEY]) { |
4460 | flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]); | |
4461 | flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]); | |
4462 | } | |
e6cc0bab | 4463 | |
70e5ed6f | 4464 | if (a[OVS_FLOW_ATTR_UFID]) { |
ab79d262 | 4465 | flow->ufid = nl_attr_get_u128(a[OVS_FLOW_ATTR_UFID]); |
70e5ed6f JS |
4466 | flow->ufid_present = true; |
4467 | } | |
e6cc0bab AZ |
4468 | if (a[OVS_FLOW_ATTR_MASK]) { |
4469 | flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]); | |
4470 | flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]); | |
4471 | } | |
df2c07f4 JP |
4472 | if (a[OVS_FLOW_ATTR_ACTIONS]) { |
4473 | flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]); | |
4474 | flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]); | |
d6569377 | 4475 | } |
df2c07f4 JP |
4476 | if (a[OVS_FLOW_ATTR_STATS]) { |
4477 | flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]); | |
d6569377 | 4478 | } |
df2c07f4 JP |
4479 | if (a[OVS_FLOW_ATTR_TCP_FLAGS]) { |
4480 | flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]); | |
d6569377 | 4481 | } |
df2c07f4 JP |
4482 | if (a[OVS_FLOW_ATTR_USED]) { |
4483 | flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]); | |
9e980142 | 4484 | } |
d6569377 BP |
4485 | return 0; |
4486 | } | |
4487 | ||
beb75a40 JS |
4488 | |
4489 | /* | |
a8a3eee4 JS |
4490 | * If PACKET_TYPE attribute is present in 'data', it filters PACKET_TYPE out. |
4491 | * If the flow is not Ethernet, the OVS_KEY_ATTR_PACKET_TYPE is converted to | |
4492 | * OVS_KEY_ATTR_ETHERTYPE. Puts 'data' to 'buf'. | |
beb75a40 JS |
4493 | */ |
4494 | static void | |
4495 | put_exclude_packet_type(struct ofpbuf *buf, uint16_t type, | |
4496 | const struct nlattr *data, uint16_t data_len) | |
4497 | { | |
4498 | const struct nlattr *packet_type; | |
4499 | ||
4500 | packet_type = nl_attr_find__(data, data_len, OVS_KEY_ATTR_PACKET_TYPE); | |
4501 | ||
4502 | if (packet_type) { | |
4503 | /* exclude PACKET_TYPE Netlink attribute. */ | |
4504 | ovs_assert(NLA_ALIGN(packet_type->nla_len) == NL_A_U32_SIZE); | |
4505 | size_t packet_type_len = NL_A_U32_SIZE; | |
4506 | size_t first_chunk_size = (uint8_t *)packet_type - (uint8_t *)data; | |
4507 | size_t second_chunk_size = data_len - first_chunk_size | |
4508 | - packet_type_len; | |
beb75a40 | 4509 | struct nlattr *next_attr = nl_attr_next(packet_type); |
1ca5b61b | 4510 | size_t ofs; |
beb75a40 | 4511 | |
1ca5b61b JS |
4512 | ofs = nl_msg_start_nested(buf, type); |
4513 | nl_msg_put(buf, data, first_chunk_size); | |
4514 | nl_msg_put(buf, next_attr, second_chunk_size); | |
a8a3eee4 JS |
4515 | if (!nl_attr_find__(data, data_len, OVS_KEY_ATTR_ETHERNET)) { |
4516 | ovs_be16 pt = pt_ns_type_be(nl_attr_get_be32(packet_type)); | |
4517 | const struct nlattr *nla; | |
4518 | ||
7c5793e6 | 4519 | nla = nl_attr_find(buf, ofs + NLA_HDRLEN, OVS_KEY_ATTR_ETHERTYPE); |
a8a3eee4 JS |
4520 | if (nla) { |
4521 | ovs_be16 *ethertype; | |
4522 | ||
4523 | ethertype = CONST_CAST(ovs_be16 *, nl_attr_get(nla)); | |
4524 | *ethertype = pt; | |
4525 | } else { | |
4526 | nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, pt); | |
4527 | } | |
4528 | } | |
1ca5b61b | 4529 | nl_msg_end_nested(buf, ofs); |
beb75a40 JS |
4530 | } else { |
4531 | nl_msg_put_unspec(buf, type, data, data_len); | |
4532 | } | |
4533 | } | |
4534 | ||
df2c07f4 | 4535 | /* Appends to 'buf' (which must initially be empty) a "struct ovs_header" |
d6569377 BP |
4536 | * followed by Netlink attributes corresponding to 'flow'. */ |
4537 | static void | |
93451a0a AS |
4538 | dpif_netlink_flow_to_ofpbuf(const struct dpif_netlink_flow *flow, |
4539 | struct ofpbuf *buf) | |
d6569377 | 4540 | { |
df2c07f4 | 4541 | struct ovs_header *ovs_header; |
d6569377 | 4542 | |
df2c07f4 | 4543 | nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, |
30b44744 | 4544 | NLM_F_REQUEST | flow->nlmsg_flags, |
69685a88 | 4545 | flow->cmd, OVS_FLOW_VERSION); |
37a1300c | 4546 | |
df2c07f4 JP |
4547 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4548 | ovs_header->dp_ifindex = flow->dp_ifindex; | |
d6569377 | 4549 | |
70e5ed6f | 4550 | if (flow->ufid_present) { |
ab79d262 | 4551 | nl_msg_put_u128(buf, OVS_FLOW_ATTR_UFID, flow->ufid); |
70e5ed6f JS |
4552 | } |
4553 | if (flow->ufid_terse) { | |
4554 | nl_msg_put_u32(buf, OVS_FLOW_ATTR_UFID_FLAGS, | |
4555 | OVS_UFID_F_OMIT_KEY | OVS_UFID_F_OMIT_MASK | |
4556 | | OVS_UFID_F_OMIT_ACTIONS); | |
4557 | } | |
64bb477f JS |
4558 | if (!flow->ufid_terse || !flow->ufid_present) { |
4559 | if (flow->key_len) { | |
beb75a40 JS |
4560 | put_exclude_packet_type(buf, OVS_FLOW_ATTR_KEY, flow->key, |
4561 | flow->key_len); | |
64bb477f | 4562 | } |
64bb477f | 4563 | if (flow->mask_len) { |
beb75a40 JS |
4564 | put_exclude_packet_type(buf, OVS_FLOW_ATTR_MASK, flow->mask, |
4565 | flow->mask_len); | |
64bb477f JS |
4566 | } |
4567 | if (flow->actions || flow->actions_len) { | |
4568 | nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS, | |
4569 | flow->actions, flow->actions_len); | |
4570 | } | |
d6569377 BP |
4571 | } |
4572 | ||
4573 | /* We never need to send these to the kernel. */ | |
cb22974d BP |
4574 | ovs_assert(!flow->stats); |
4575 | ovs_assert(!flow->tcp_flags); | |
4576 | ovs_assert(!flow->used); | |
d6569377 BP |
4577 | |
4578 | if (flow->clear) { | |
df2c07f4 | 4579 | nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR); |
d6569377 | 4580 | } |
43f9ac0a JR |
4581 | if (flow->probe) { |
4582 | nl_msg_put_flag(buf, OVS_FLOW_ATTR_PROBE); | |
4583 | } | |
d6569377 BP |
4584 | } |
4585 | ||
4586 | /* Clears 'flow' to "empty" values. */ | |
d3d8f1f7 | 4587 | static void |
93451a0a | 4588 | dpif_netlink_flow_init(struct dpif_netlink_flow *flow) |
d6569377 BP |
4589 | { |
4590 | memset(flow, 0, sizeof *flow); | |
4591 | } | |
4592 | ||
4593 | /* Executes 'request' in the kernel datapath. If the command fails, returns a | |
4594 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4595 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
37a1300c BP |
4596 | * result of the command is expected to be a flow also, which is decoded and |
4597 | * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply | |
4598 | * is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4599 | static int |
93451a0a AS |
4600 | dpif_netlink_flow_transact(struct dpif_netlink_flow *request, |
4601 | struct dpif_netlink_flow *reply, | |
4602 | struct ofpbuf **bufp) | |
d6569377 | 4603 | { |
37a1300c | 4604 | struct ofpbuf *request_buf; |
d6569377 | 4605 | int error; |
d6569377 | 4606 | |
cb22974d | 4607 | ovs_assert((reply != NULL) == (bufp != NULL)); |
d6569377 | 4608 | |
30b44744 BP |
4609 | if (reply) { |
4610 | request->nlmsg_flags |= NLM_F_ECHO; | |
4611 | } | |
4612 | ||
37a1300c | 4613 | request_buf = ofpbuf_new(1024); |
93451a0a | 4614 | dpif_netlink_flow_to_ofpbuf(request, request_buf); |
a88b4e04 | 4615 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
37a1300c | 4616 | ofpbuf_delete(request_buf); |
d6569377 | 4617 | |
37a1300c BP |
4618 | if (reply) { |
4619 | if (!error) { | |
93451a0a | 4620 | error = dpif_netlink_flow_from_ofpbuf(reply, *bufp); |
37a1300c | 4621 | } |
d6569377 | 4622 | if (error) { |
93451a0a | 4623 | dpif_netlink_flow_init(reply); |
37a1300c BP |
4624 | ofpbuf_delete(*bufp); |
4625 | *bufp = NULL; | |
d6569377 | 4626 | } |
d6569377 BP |
4627 | } |
4628 | return error; | |
4629 | } | |
4630 | ||
4631 | static void | |
93451a0a AS |
4632 | dpif_netlink_flow_get_stats(const struct dpif_netlink_flow *flow, |
4633 | struct dpif_flow_stats *stats) | |
d6569377 BP |
4634 | { |
4635 | if (flow->stats) { | |
6a54dedc BP |
4636 | stats->n_packets = get_32aligned_u64(&flow->stats->n_packets); |
4637 | stats->n_bytes = get_32aligned_u64(&flow->stats->n_bytes); | |
d6569377 BP |
4638 | } else { |
4639 | stats->n_packets = 0; | |
4640 | stats->n_bytes = 0; | |
4641 | } | |
0e70cdcb | 4642 | stats->used = flow->used ? get_32aligned_u64(flow->used) : 0; |
d6569377 BP |
4643 | stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; |
4644 | } | |
e0467f6d | 4645 | |
14b4d2f9 BP |
4646 | /* Logs information about a packet that was recently lost in 'ch' (in |
4647 | * 'dpif_'). */ | |
4648 | static void | |
93451a0a | 4649 | report_loss(struct dpif_netlink *dpif, struct dpif_channel *ch, uint32_t ch_idx, |
1579cf67 | 4650 | uint32_t handler_id) |
14b4d2f9 | 4651 | { |
14b4d2f9 | 4652 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
14b4d2f9 BP |
4653 | struct ds s; |
4654 | ||
8d675c5a | 4655 | if (VLOG_DROP_WARN(&rl)) { |
14b4d2f9 BP |
4656 | return; |
4657 | } | |
4658 | ||
4659 | ds_init(&s); | |
4660 | if (ch->last_poll != LLONG_MIN) { | |
4661 | ds_put_format(&s, " (last polled %lld ms ago)", | |
4662 | time_msec() - ch->last_poll); | |
4663 | } | |
14b4d2f9 | 4664 | |
1579cf67 | 4665 | VLOG_WARN("%s: lost packet on port channel %u of handler %u", |
9b00386b | 4666 | dpif_name(&dpif->dpif), ch_idx, handler_id); |
14b4d2f9 BP |
4667 | ds_destroy(&s); |
4668 | } |