]>
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); | |
113 | ||
114 | struct dpif_netlink_flow { | |
37a1300c BP |
115 | /* Generic Netlink header. */ |
116 | uint8_t cmd; | |
d6569377 | 117 | |
df2c07f4 | 118 | /* struct ovs_header. */ |
d6569377 | 119 | unsigned int nlmsg_flags; |
254f2dc8 | 120 | int dp_ifindex; |
d6569377 BP |
121 | |
122 | /* Attributes. | |
123 | * | |
0e70cdcb BP |
124 | * The 'stats' member points to 64-bit data that might only be aligned on |
125 | * 32-bit boundaries, so get_unaligned_u64() should be used to access its | |
126 | * values. | |
d2a23af2 | 127 | * |
df2c07f4 | 128 | * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in |
d2a23af2 | 129 | * the Netlink version of the command, even if actions_len is zero. */ |
df2c07f4 | 130 | const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */ |
d6569377 | 131 | size_t key_len; |
e6cc0bab AZ |
132 | const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */ |
133 | size_t mask_len; | |
df2c07f4 | 134 | const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */ |
d6569377 | 135 | size_t actions_len; |
70e5ed6f JS |
136 | ovs_u128 ufid; /* OVS_FLOW_ATTR_FLOW_ID. */ |
137 | bool ufid_present; /* Is there a UFID? */ | |
138 | bool ufid_terse; /* Skip serializing key/mask/acts? */ | |
df2c07f4 JP |
139 | const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */ |
140 | const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */ | |
0e70cdcb | 141 | const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */ |
df2c07f4 | 142 | bool clear; /* OVS_FLOW_ATTR_CLEAR. */ |
43f9ac0a | 143 | bool probe; /* OVS_FLOW_ATTR_PROBE. */ |
d6569377 BP |
144 | }; |
145 | ||
93451a0a AS |
146 | static void dpif_netlink_flow_init(struct dpif_netlink_flow *); |
147 | static int dpif_netlink_flow_from_ofpbuf(struct dpif_netlink_flow *, | |
148 | const struct ofpbuf *); | |
149 | static void dpif_netlink_flow_to_ofpbuf(const struct dpif_netlink_flow *, | |
150 | struct ofpbuf *); | |
151 | static int dpif_netlink_flow_transact(struct dpif_netlink_flow *request, | |
152 | struct dpif_netlink_flow *reply, | |
153 | struct ofpbuf **bufp); | |
154 | static void dpif_netlink_flow_get_stats(const struct dpif_netlink_flow *, | |
155 | struct dpif_flow_stats *); | |
7af12bd7 | 156 | static void dpif_netlink_flow_to_dpif_flow(struct dpif *, struct dpif_flow *, |
93451a0a | 157 | const struct dpif_netlink_flow *); |
d6569377 | 158 | |
989fd548 | 159 | /* One of the dpif channels between the kernel and userspace. */ |
fe3d61b3 | 160 | struct dpif_channel { |
14b4d2f9 | 161 | struct nl_sock *sock; /* Netlink socket. */ |
14b4d2f9 | 162 | long long int last_poll; /* Last time this channel was polled. */ |
fe3d61b3 BP |
163 | }; |
164 | ||
09cac43f NR |
165 | #ifdef _WIN32 |
166 | #define VPORT_SOCK_POOL_SIZE 1 | |
167 | /* On Windows, there is no native support for epoll. There are equivalent | |
168 | * interfaces though, that are not used currently. For simpicity, a pool of | |
169 | * netlink sockets is used. Each socket is represented by 'struct | |
170 | * dpif_windows_vport_sock'. Since it is a pool, multiple OVS ports may be | |
171 | * sharing the same socket. In the future, we can add a reference count and | |
172 | * such fields. */ | |
173 | struct dpif_windows_vport_sock { | |
174 | struct nl_sock *nl_sock; /* netlink socket. */ | |
175 | }; | |
176 | #endif | |
177 | ||
1579cf67 | 178 | struct dpif_handler { |
1579cf67 AW |
179 | struct epoll_event *epoll_events; |
180 | int epoll_fd; /* epoll fd that includes channel socks. */ | |
181 | int n_events; /* Num events returned by epoll_wait(). */ | |
182 | int event_offset; /* Offset into 'epoll_events'. */ | |
09cac43f NR |
183 | |
184 | #ifdef _WIN32 | |
185 | /* Pool of sockets. */ | |
186 | struct dpif_windows_vport_sock *vport_sock_pool; | |
187 | size_t last_used_pool_idx; /* Index to aid in allocating a | |
188 | socket in the pool to a port. */ | |
189 | #endif | |
1579cf67 | 190 | }; |
14b4d2f9 | 191 | |
96fba48f | 192 | /* Datapath interface for the openvswitch Linux kernel module. */ |
93451a0a | 193 | struct dpif_netlink { |
96fba48f | 194 | struct dpif dpif; |
254f2dc8 | 195 | int dp_ifindex; |
e9e28be3 | 196 | |
b063d9f0 | 197 | /* Upcall messages. */ |
1579cf67 AW |
198 | struct fat_rwlock upcall_lock; |
199 | struct dpif_handler *handlers; | |
200 | uint32_t n_handlers; /* Num of upcall handlers. */ | |
69c51582 | 201 | struct dpif_channel *channels; /* Array of channels for each port. */ |
1579cf67 AW |
202 | int uc_array_size; /* Size of 'handler->channels' and */ |
203 | /* 'handler->epoll_events'. */ | |
982b8810 | 204 | |
e9e28be3 | 205 | /* Change notification. */ |
e4516b20 | 206 | struct nl_sock *port_notifier; /* vport multicast group subscriber. */ |
61eae437 | 207 | bool refresh_channels; |
96fba48f BP |
208 | }; |
209 | ||
93451a0a | 210 | static void report_loss(struct dpif_netlink *, struct dpif_channel *, |
9b00386b | 211 | uint32_t ch_idx, uint32_t handler_id); |
1579cf67 | 212 | |
96fba48f BP |
213 | static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); |
214 | ||
e4516b20 BP |
215 | /* Generic Netlink family numbers for OVS. |
216 | * | |
93451a0a | 217 | * Initialized by dpif_netlink_init(). */ |
df2c07f4 JP |
218 | static int ovs_datapath_family; |
219 | static int ovs_vport_family; | |
220 | static int ovs_flow_family; | |
221 | static int ovs_packet_family; | |
80738e5f | 222 | static int ovs_meter_family; |
906ff9d2 | 223 | static int ovs_ct_limit_family; |
982b8810 | 224 | |
e4516b20 BP |
225 | /* Generic Netlink multicast groups for OVS. |
226 | * | |
93451a0a | 227 | * Initialized by dpif_netlink_init(). */ |
e4516b20 | 228 | static unsigned int ovs_vport_mcgroup; |
982b8810 | 229 | |
921c370a EG |
230 | /* If true, tunnel devices are created using OVS compat/genetlink. |
231 | * If false, tunnel devices are created with rtnetlink and using light weight | |
232 | * tunnels. If we fail to create the tunnel the rtnetlink+LWT, then we fallback | |
233 | * to using the compat interface. */ | |
234 | static bool ovs_tunnels_out_of_tree = true; | |
235 | ||
93451a0a AS |
236 | static int dpif_netlink_init(void); |
237 | static int open_dpif(const struct dpif_netlink_dp *, struct dpif **); | |
238 | static uint32_t dpif_netlink_port_get_pid(const struct dpif *, | |
769b5034 | 239 | odp_port_t port_no); |
09cac43f | 240 | static void dpif_netlink_handler_uninit(struct dpif_handler *handler); |
93451a0a AS |
241 | static int dpif_netlink_refresh_channels(struct dpif_netlink *, |
242 | uint32_t n_handlers); | |
243 | static void dpif_netlink_vport_to_ofpbuf(const struct dpif_netlink_vport *, | |
244 | struct ofpbuf *); | |
245 | static int dpif_netlink_vport_from_ofpbuf(struct dpif_netlink_vport *, | |
246 | const struct ofpbuf *); | |
921c370a EG |
247 | static int dpif_netlink_port_query__(const struct dpif_netlink *dpif, |
248 | odp_port_t port_no, const char *port_name, | |
249 | struct dpif_port *dpif_port); | |
f0fef760 | 250 | |
d240e46a | 251 | static int |
622ea8fd | 252 | create_nl_sock(struct dpif_netlink *dpif OVS_UNUSED, struct nl_sock **sockp) |
d240e46a AGS |
253 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
254 | { | |
255 | #ifndef _WIN32 | |
622ea8fd | 256 | return nl_sock_create(NETLINK_GENERIC, sockp); |
d240e46a AGS |
257 | #else |
258 | /* Pick netlink sockets to use in a round-robin fashion from each | |
259 | * handler's pool of sockets. */ | |
260 | struct dpif_handler *handler = &dpif->handlers[0]; | |
261 | struct dpif_windows_vport_sock *sock_pool = handler->vport_sock_pool; | |
262 | size_t index = handler->last_used_pool_idx; | |
263 | ||
264 | /* A pool of sockets is allocated when the handler is initialized. */ | |
265 | if (sock_pool == NULL) { | |
622ea8fd | 266 | *sockp = NULL; |
d240e46a AGS |
267 | return EINVAL; |
268 | } | |
269 | ||
270 | ovs_assert(index < VPORT_SOCK_POOL_SIZE); | |
622ea8fd BP |
271 | *sockp = sock_pool[index].nl_sock; |
272 | ovs_assert(*sockp); | |
d240e46a AGS |
273 | index = (index == VPORT_SOCK_POOL_SIZE - 1) ? 0 : index + 1; |
274 | handler->last_used_pool_idx = index; | |
275 | return 0; | |
276 | #endif | |
277 | } | |
278 | ||
279 | static void | |
622ea8fd | 280 | close_nl_sock(struct nl_sock *sock) |
d240e46a AGS |
281 | { |
282 | #ifndef _WIN32 | |
622ea8fd | 283 | nl_sock_destroy(sock); |
d240e46a AGS |
284 | #endif |
285 | } | |
286 | ||
93451a0a AS |
287 | static struct dpif_netlink * |
288 | dpif_netlink_cast(const struct dpif *dpif) | |
96fba48f | 289 | { |
93451a0a AS |
290 | dpif_assert_class(dpif, &dpif_netlink_class); |
291 | return CONTAINER_OF(dpif, struct dpif_netlink, dpif); | |
96fba48f BP |
292 | } |
293 | ||
d3d22744 | 294 | static int |
93451a0a AS |
295 | dpif_netlink_enumerate(struct sset *all_dps, |
296 | const struct dpif_class *dpif_class OVS_UNUSED) | |
d3d22744 | 297 | { |
aaff4b55 | 298 | struct nl_dump dump; |
d57695d7 JS |
299 | uint64_t reply_stub[NL_DUMP_BUFSIZE / 8]; |
300 | struct ofpbuf msg, buf; | |
aaff4b55 | 301 | int error; |
982b8810 | 302 | |
93451a0a | 303 | error = dpif_netlink_init(); |
aaff4b55 BP |
304 | if (error) { |
305 | return error; | |
982b8810 | 306 | } |
d3d22744 | 307 | |
d57695d7 | 308 | ofpbuf_use_stub(&buf, reply_stub, sizeof reply_stub); |
93451a0a | 309 | dpif_netlink_dp_dump_start(&dump); |
d57695d7 | 310 | while (nl_dump_next(&dump, &msg, &buf)) { |
93451a0a | 311 | struct dpif_netlink_dp dp; |
d6569377 | 312 | |
93451a0a | 313 | if (!dpif_netlink_dp_from_ofpbuf(&dp, &msg)) { |
d0c23a1a | 314 | sset_add(all_dps, dp.name); |
d3d22744 BP |
315 | } |
316 | } | |
d57695d7 | 317 | ofpbuf_uninit(&buf); |
aaff4b55 | 318 | return nl_dump_done(&dump); |
d3d22744 BP |
319 | } |
320 | ||
96fba48f | 321 | static int |
93451a0a AS |
322 | dpif_netlink_open(const struct dpif_class *class OVS_UNUSED, const char *name, |
323 | bool create, struct dpif **dpifp) | |
96fba48f | 324 | { |
93451a0a | 325 | struct dpif_netlink_dp dp_request, dp; |
c19e6535 | 326 | struct ofpbuf *buf; |
ea36840f | 327 | uint32_t upcall_pid; |
c19e6535 | 328 | int error; |
96fba48f | 329 | |
93451a0a | 330 | error = dpif_netlink_init(); |
982b8810 BP |
331 | if (error) { |
332 | return error; | |
333 | } | |
334 | ||
982b8810 | 335 | /* Create or look up datapath. */ |
93451a0a | 336 | dpif_netlink_dp_init(&dp_request); |
ea36840f BP |
337 | if (create) { |
338 | dp_request.cmd = OVS_DP_CMD_NEW; | |
339 | upcall_pid = 0; | |
340 | dp_request.upcall_pid = &upcall_pid; | |
341 | } else { | |
b7fd5e38 TG |
342 | /* Use OVS_DP_CMD_SET to report user features */ |
343 | dp_request.cmd = OVS_DP_CMD_SET; | |
ea36840f | 344 | } |
254f2dc8 | 345 | dp_request.name = name; |
b7fd5e38 | 346 | dp_request.user_features |= OVS_DP_F_UNALIGNED; |
1579cf67 | 347 | dp_request.user_features |= OVS_DP_F_VPORT_PIDS; |
93451a0a | 348 | error = dpif_netlink_dp_transact(&dp_request, &dp, &buf); |
982b8810 BP |
349 | if (error) { |
350 | return error; | |
c19e6535 | 351 | } |
254f2dc8 | 352 | |
e4516b20 | 353 | error = open_dpif(&dp, dpifp); |
8f4a4df5 | 354 | ofpbuf_delete(buf); |
e4516b20 | 355 | return error; |
c19e6535 BP |
356 | } |
357 | ||
e4516b20 | 358 | static int |
93451a0a | 359 | open_dpif(const struct dpif_netlink_dp *dp, struct dpif **dpifp) |
c19e6535 | 360 | { |
93451a0a | 361 | struct dpif_netlink *dpif; |
c19e6535 | 362 | |
17411ecf | 363 | dpif = xzalloc(sizeof *dpif); |
e4516b20 | 364 | dpif->port_notifier = NULL; |
1579cf67 | 365 | fat_rwlock_init(&dpif->upcall_lock); |
c19e6535 | 366 | |
93451a0a | 367 | dpif_init(&dpif->dpif, &dpif_netlink_class, dp->name, |
254f2dc8 | 368 | dp->dp_ifindex, dp->dp_ifindex); |
c19e6535 | 369 | |
254f2dc8 | 370 | dpif->dp_ifindex = dp->dp_ifindex; |
c19e6535 | 371 | *dpifp = &dpif->dpif; |
e4516b20 BP |
372 | |
373 | return 0; | |
96fba48f BP |
374 | } |
375 | ||
09cac43f NR |
376 | #ifdef _WIN32 |
377 | static void | |
378 | vport_delete_sock_pool(struct dpif_handler *handler) | |
379 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
380 | { | |
381 | if (handler->vport_sock_pool) { | |
382 | uint32_t i; | |
383 | struct dpif_windows_vport_sock *sock_pool = | |
384 | handler->vport_sock_pool; | |
385 | ||
386 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
387 | if (sock_pool[i].nl_sock) { | |
388 | nl_sock_unsubscribe_packets(sock_pool[i].nl_sock); | |
389 | nl_sock_destroy(sock_pool[i].nl_sock); | |
390 | sock_pool[i].nl_sock = NULL; | |
391 | } | |
392 | } | |
393 | ||
394 | free(handler->vport_sock_pool); | |
395 | handler->vport_sock_pool = NULL; | |
396 | } | |
397 | } | |
398 | ||
399 | static int | |
400 | vport_create_sock_pool(struct dpif_handler *handler) | |
401 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
402 | { | |
403 | struct dpif_windows_vport_sock *sock_pool; | |
404 | size_t i; | |
405 | int error = 0; | |
406 | ||
407 | sock_pool = xzalloc(VPORT_SOCK_POOL_SIZE * sizeof *sock_pool); | |
408 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
409 | error = nl_sock_create(NETLINK_GENERIC, &sock_pool[i].nl_sock); | |
410 | if (error) { | |
411 | goto error; | |
412 | } | |
413 | ||
414 | /* Enable the netlink socket to receive packets. This is equivalent to | |
415 | * calling nl_sock_join_mcgroup() to receive events. */ | |
416 | error = nl_sock_subscribe_packets(sock_pool[i].nl_sock); | |
417 | if (error) { | |
418 | goto error; | |
419 | } | |
420 | } | |
421 | ||
422 | handler->vport_sock_pool = sock_pool; | |
423 | handler->last_used_pool_idx = 0; | |
424 | return 0; | |
425 | ||
426 | error: | |
427 | vport_delete_sock_pool(handler); | |
428 | return error; | |
429 | } | |
09cac43f NR |
430 | #endif /* _WIN32 */ |
431 | ||
69c51582 MC |
432 | /* Given the port number 'port_idx', extracts the pid of netlink socket |
433 | * associated to the port and assigns it to 'upcall_pid'. */ | |
1579cf67 | 434 | static bool |
69c51582 MC |
435 | vport_get_pid(struct dpif_netlink *dpif, uint32_t port_idx, |
436 | uint32_t *upcall_pid) | |
1579cf67 | 437 | { |
1579cf67 | 438 | /* Since the nl_sock can only be assigned in either all |
69c51582 | 439 | * or none "dpif" channels, the following check |
1579cf67 | 440 | * would suffice. */ |
69c51582 | 441 | if (!dpif->channels[port_idx].sock) { |
1579cf67 AW |
442 | return false; |
443 | } | |
09cac43f | 444 | ovs_assert(!WINDOWS || dpif->n_handlers <= 1); |
1579cf67 | 445 | |
69c51582 | 446 | *upcall_pid = nl_sock_pid(dpif->channels[port_idx].sock); |
989fd548 | 447 | |
1579cf67 | 448 | return true; |
989fd548 JP |
449 | } |
450 | ||
451 | static int | |
69c51582 | 452 | vport_add_channel(struct dpif_netlink *dpif, odp_port_t port_no, |
622ea8fd | 453 | struct nl_sock *sock) |
989fd548 JP |
454 | { |
455 | struct epoll_event event; | |
4e022ec0 | 456 | uint32_t port_idx = odp_to_u32(port_no); |
69c51582 | 457 | size_t i; |
1579cf67 | 458 | int error; |
989fd548 | 459 | |
1579cf67 | 460 | if (dpif->handlers == NULL) { |
622ea8fd | 461 | close_nl_sock(sock); |
989fd548 JP |
462 | return 0; |
463 | } | |
464 | ||
1579cf67 AW |
465 | /* We assume that the datapath densely chooses port numbers, which can |
466 | * therefore be used as an index into 'channels' and 'epoll_events' of | |
69c51582 | 467 | * 'dpif'. */ |
4e022ec0 AW |
468 | if (port_idx >= dpif->uc_array_size) { |
469 | uint32_t new_size = port_idx + 1; | |
989fd548 | 470 | |
12d76859 | 471 | if (new_size > MAX_PORTS) { |
989fd548 JP |
472 | VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big", |
473 | dpif_name(&dpif->dpif), port_no); | |
474 | return EFBIG; | |
475 | } | |
476 | ||
69c51582 MC |
477 | dpif->channels = xrealloc(dpif->channels, |
478 | new_size * sizeof *dpif->channels); | |
1579cf67 | 479 | |
69c51582 MC |
480 | for (i = dpif->uc_array_size; i < new_size; i++) { |
481 | dpif->channels[i].sock = NULL; | |
482 | } | |
1579cf67 | 483 | |
69c51582 MC |
484 | for (i = 0; i < dpif->n_handlers; i++) { |
485 | struct dpif_handler *handler = &dpif->handlers[i]; | |
1579cf67 AW |
486 | |
487 | handler->epoll_events = xrealloc(handler->epoll_events, | |
488 | new_size * sizeof *handler->epoll_events); | |
989fd548 | 489 | |
1579cf67 | 490 | } |
989fd548 JP |
491 | dpif->uc_array_size = new_size; |
492 | } | |
493 | ||
494 | memset(&event, 0, sizeof event); | |
69c51582 | 495 | event.events = EPOLLIN | EPOLLEXCLUSIVE; |
4e022ec0 | 496 | event.data.u32 = port_idx; |
989fd548 | 497 | |
1579cf67 AW |
498 | for (i = 0; i < dpif->n_handlers; i++) { |
499 | struct dpif_handler *handler = &dpif->handlers[i]; | |
500 | ||
09cac43f | 501 | #ifndef _WIN32 |
622ea8fd | 502 | if (epoll_ctl(handler->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), |
1579cf67 AW |
503 | &event) < 0) { |
504 | error = errno; | |
505 | goto error; | |
506 | } | |
93451a0a | 507 | #endif |
1579cf67 | 508 | } |
622ea8fd | 509 | dpif->channels[port_idx].sock = sock; |
69c51582 | 510 | dpif->channels[port_idx].last_poll = LLONG_MIN; |
989fd548 JP |
511 | |
512 | return 0; | |
1579cf67 AW |
513 | |
514 | error: | |
09cac43f | 515 | #ifndef _WIN32 |
69c51582 MC |
516 | while (i--) { |
517 | epoll_ctl(dpif->handlers[i].epoll_fd, EPOLL_CTL_DEL, | |
622ea8fd | 518 | nl_sock_fd(sock), NULL); |
1579cf67 | 519 | } |
69c51582 MC |
520 | #endif |
521 | dpif->channels[port_idx].sock = NULL; | |
1579cf67 AW |
522 | |
523 | return error; | |
989fd548 JP |
524 | } |
525 | ||
526 | static void | |
93451a0a | 527 | vport_del_channels(struct dpif_netlink *dpif, odp_port_t port_no) |
989fd548 | 528 | { |
4e022ec0 | 529 | uint32_t port_idx = odp_to_u32(port_no); |
1579cf67 | 530 | size_t i; |
989fd548 | 531 | |
69c51582 MC |
532 | if (!dpif->handlers || port_idx >= dpif->uc_array_size |
533 | || !dpif->channels[port_idx].sock) { | |
989fd548 JP |
534 | return; |
535 | } | |
536 | ||
1579cf67 AW |
537 | for (i = 0; i < dpif->n_handlers; i++) { |
538 | struct dpif_handler *handler = &dpif->handlers[i]; | |
09cac43f | 539 | #ifndef _WIN32 |
1579cf67 | 540 | epoll_ctl(handler->epoll_fd, EPOLL_CTL_DEL, |
69c51582 | 541 | nl_sock_fd(dpif->channels[port_idx].sock), NULL); |
09cac43f | 542 | #endif |
1579cf67 AW |
543 | handler->event_offset = handler->n_events = 0; |
544 | } | |
69c51582 MC |
545 | #ifndef _WIN32 |
546 | nl_sock_destroy(dpif->channels[port_idx].sock); | |
547 | #endif | |
548 | dpif->channels[port_idx].sock = NULL; | |
1579cf67 AW |
549 | } |
550 | ||
551 | static void | |
93451a0a AS |
552 | destroy_all_channels(struct dpif_netlink *dpif) |
553 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
1579cf67 AW |
554 | { |
555 | unsigned int i; | |
556 | ||
557 | if (!dpif->handlers) { | |
558 | return; | |
559 | } | |
560 | ||
561 | for (i = 0; i < dpif->uc_array_size; i++ ) { | |
93451a0a | 562 | struct dpif_netlink_vport vport_request; |
1579cf67 AW |
563 | uint32_t upcall_pids = 0; |
564 | ||
69c51582 | 565 | if (!dpif->channels[i].sock) { |
1579cf67 AW |
566 | continue; |
567 | } | |
568 | ||
569 | /* Turn off upcalls. */ | |
93451a0a | 570 | dpif_netlink_vport_init(&vport_request); |
1579cf67 AW |
571 | vport_request.cmd = OVS_VPORT_CMD_SET; |
572 | vport_request.dp_ifindex = dpif->dp_ifindex; | |
573 | vport_request.port_no = u32_to_odp(i); | |
a78f446a | 574 | vport_request.n_upcall_pids = 1; |
1579cf67 | 575 | vport_request.upcall_pids = &upcall_pids; |
93451a0a | 576 | dpif_netlink_vport_transact(&vport_request, NULL, NULL); |
1579cf67 AW |
577 | |
578 | vport_del_channels(dpif, u32_to_odp(i)); | |
579 | } | |
580 | ||
581 | for (i = 0; i < dpif->n_handlers; i++) { | |
582 | struct dpif_handler *handler = &dpif->handlers[i]; | |
583 | ||
09cac43f | 584 | dpif_netlink_handler_uninit(handler); |
1579cf67 | 585 | free(handler->epoll_events); |
1579cf67 | 586 | } |
69c51582 | 587 | free(dpif->channels); |
1579cf67 AW |
588 | free(dpif->handlers); |
589 | dpif->handlers = NULL; | |
69c51582 | 590 | dpif->channels = NULL; |
1579cf67 AW |
591 | dpif->n_handlers = 0; |
592 | dpif->uc_array_size = 0; | |
17411ecf JG |
593 | } |
594 | ||
96fba48f | 595 | static void |
93451a0a | 596 | dpif_netlink_close(struct dpif *dpif_) |
96fba48f | 597 | { |
93451a0a | 598 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
c7178a0b | 599 | |
e4516b20 | 600 | nl_sock_destroy(dpif->port_notifier); |
1579cf67 AW |
601 | |
602 | fat_rwlock_wrlock(&dpif->upcall_lock); | |
603 | destroy_all_channels(dpif); | |
604 | fat_rwlock_unlock(&dpif->upcall_lock); | |
605 | ||
606 | fat_rwlock_destroy(&dpif->upcall_lock); | |
96fba48f BP |
607 | free(dpif); |
608 | } | |
609 | ||
610 | static int | |
93451a0a | 611 | dpif_netlink_destroy(struct dpif *dpif_) |
96fba48f | 612 | { |
93451a0a AS |
613 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
614 | struct dpif_netlink_dp dp; | |
d6569377 | 615 | |
93451a0a | 616 | dpif_netlink_dp_init(&dp); |
df2c07f4 | 617 | dp.cmd = OVS_DP_CMD_DEL; |
254f2dc8 | 618 | dp.dp_ifindex = dpif->dp_ifindex; |
93451a0a | 619 | return dpif_netlink_dp_transact(&dp, NULL, NULL); |
96fba48f BP |
620 | } |
621 | ||
a36de779 | 622 | static bool |
93451a0a | 623 | dpif_netlink_run(struct dpif *dpif_) |
61eae437 | 624 | { |
93451a0a | 625 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1579cf67 | 626 | |
61eae437 BP |
627 | if (dpif->refresh_channels) { |
628 | dpif->refresh_channels = false; | |
1579cf67 | 629 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 630 | dpif_netlink_refresh_channels(dpif, dpif->n_handlers); |
1579cf67 | 631 | fat_rwlock_unlock(&dpif->upcall_lock); |
61eae437 | 632 | } |
a36de779 | 633 | return false; |
61eae437 BP |
634 | } |
635 | ||
96fba48f | 636 | static int |
93451a0a | 637 | dpif_netlink_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats) |
96fba48f | 638 | { |
93451a0a | 639 | struct dpif_netlink_dp dp; |
d6569377 BP |
640 | struct ofpbuf *buf; |
641 | int error; | |
642 | ||
93451a0a | 643 | error = dpif_netlink_dp_get(dpif_, &dp, &buf); |
d6569377 | 644 | if (!error) { |
6a54dedc BP |
645 | memset(stats, 0, sizeof *stats); |
646 | ||
647 | if (dp.stats) { | |
648 | stats->n_hit = get_32aligned_u64(&dp.stats->n_hit); | |
649 | stats->n_missed = get_32aligned_u64(&dp.stats->n_missed); | |
650 | stats->n_lost = get_32aligned_u64(&dp.stats->n_lost); | |
651 | stats->n_flows = get_32aligned_u64(&dp.stats->n_flows); | |
652 | } | |
653 | ||
654 | if (dp.megaflow_stats) { | |
655 | stats->n_masks = dp.megaflow_stats->n_masks; | |
656 | stats->n_mask_hit = get_32aligned_u64( | |
657 | &dp.megaflow_stats->n_mask_hit); | |
658 | } else { | |
659 | stats->n_masks = UINT32_MAX; | |
660 | stats->n_mask_hit = UINT64_MAX; | |
661 | } | |
d6569377 BP |
662 | ofpbuf_delete(buf); |
663 | } | |
664 | return error; | |
96fba48f BP |
665 | } |
666 | ||
b9ad7294 | 667 | static const char * |
93451a0a | 668 | get_vport_type(const struct dpif_netlink_vport *vport) |
b9ad7294 EJ |
669 | { |
670 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
671 | ||
672 | switch (vport->type) { | |
5ed51209 JS |
673 | case OVS_VPORT_TYPE_NETDEV: { |
674 | const char *type = netdev_get_type_from_name(vport->name); | |
675 | ||
676 | return type ? type : "system"; | |
677 | } | |
b9ad7294 EJ |
678 | |
679 | case OVS_VPORT_TYPE_INTERNAL: | |
680 | return "internal"; | |
681 | ||
c1fc1411 JG |
682 | case OVS_VPORT_TYPE_GENEVE: |
683 | return "geneve"; | |
684 | ||
b9ad7294 EJ |
685 | case OVS_VPORT_TYPE_GRE: |
686 | return "gre"; | |
687 | ||
b9ad7294 EJ |
688 | case OVS_VPORT_TYPE_VXLAN: |
689 | return "vxlan"; | |
690 | ||
a6ae068b LJ |
691 | case OVS_VPORT_TYPE_LISP: |
692 | return "lisp"; | |
693 | ||
4237026e PS |
694 | case OVS_VPORT_TYPE_STT: |
695 | return "stt"; | |
696 | ||
c387d817 | 697 | case OVS_VPORT_TYPE_ERSPAN: |
98514eea WT |
698 | return "erspan"; |
699 | ||
c387d817 | 700 | case OVS_VPORT_TYPE_IP6ERSPAN: |
3b10ceee GR |
701 | return "ip6erspan"; |
702 | ||
c387d817 | 703 | case OVS_VPORT_TYPE_IP6GRE: |
3b10ceee | 704 | return "ip6gre"; |
c387d817 | 705 | |
b9ad7294 EJ |
706 | case OVS_VPORT_TYPE_UNSPEC: |
707 | case __OVS_VPORT_TYPE_MAX: | |
708 | break; | |
709 | } | |
710 | ||
711 | VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", | |
712 | vport->dp_ifindex, vport->name, (unsigned int) vport->type); | |
713 | return "unknown"; | |
714 | } | |
715 | ||
c4e08753 | 716 | enum ovs_vport_type |
20c57607 | 717 | netdev_to_ovs_vport_type(const char *type) |
c060c4cf | 718 | { |
c060c4cf EJ |
719 | if (!strcmp(type, "tap") || !strcmp(type, "system")) { |
720 | return OVS_VPORT_TYPE_NETDEV; | |
721 | } else if (!strcmp(type, "internal")) { | |
722 | return OVS_VPORT_TYPE_INTERNAL; | |
4237026e PS |
723 | } else if (strstr(type, "stt")) { |
724 | return OVS_VPORT_TYPE_STT; | |
c1fc1411 JG |
725 | } else if (!strcmp(type, "geneve")) { |
726 | return OVS_VPORT_TYPE_GENEVE; | |
c060c4cf EJ |
727 | } else if (!strcmp(type, "vxlan")) { |
728 | return OVS_VPORT_TYPE_VXLAN; | |
a6ae068b LJ |
729 | } else if (!strcmp(type, "lisp")) { |
730 | return OVS_VPORT_TYPE_LISP; | |
7dc18ae9 WT |
731 | } else if (!strcmp(type, "erspan")) { |
732 | return OVS_VPORT_TYPE_ERSPAN; | |
733 | } else if (!strcmp(type, "ip6erspan")) { | |
734 | return OVS_VPORT_TYPE_IP6ERSPAN; | |
3b10ceee GR |
735 | } else if (!strcmp(type, "ip6gre")) { |
736 | return OVS_VPORT_TYPE_IP6GRE; | |
1c385f49 GR |
737 | } else if (!strcmp(type, "gre")) { |
738 | return OVS_VPORT_TYPE_GRE; | |
c060c4cf EJ |
739 | } else { |
740 | return OVS_VPORT_TYPE_UNSPEC; | |
741 | } | |
742 | } | |
743 | ||
96fba48f | 744 | static int |
20c57607 EG |
745 | dpif_netlink_port_add__(struct dpif_netlink *dpif, const char *name, |
746 | enum ovs_vport_type type, | |
747 | struct ofpbuf *options, | |
93451a0a | 748 | odp_port_t *port_nop) |
b90de034 | 749 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 750 | { |
93451a0a | 751 | struct dpif_netlink_vport request, reply; |
c19e6535 | 752 | struct ofpbuf *buf; |
622ea8fd | 753 | struct nl_sock *sock = NULL; |
790a4372 | 754 | uint32_t upcall_pids = 0; |
1579cf67 | 755 | int error = 0; |
96fba48f | 756 | |
1579cf67 | 757 | if (dpif->handlers) { |
622ea8fd | 758 | error = create_nl_sock(dpif, &sock); |
713a45db | 759 | if (error) { |
989fd548 JP |
760 | return error; |
761 | } | |
762 | } | |
763 | ||
93451a0a | 764 | dpif_netlink_vport_init(&request); |
df2c07f4 | 765 | request.cmd = OVS_VPORT_CMD_NEW; |
254f2dc8 | 766 | request.dp_ifindex = dpif->dp_ifindex; |
20c57607 EG |
767 | request.type = type; |
768 | request.name = name; | |
769 | ||
770 | request.port_no = *port_nop; | |
622ea8fd BP |
771 | if (sock) { |
772 | upcall_pids = nl_sock_pid(sock); | |
790a4372 | 773 | } |
69c51582 MC |
774 | request.n_upcall_pids = 1; |
775 | request.upcall_pids = &upcall_pids; | |
20c57607 EG |
776 | |
777 | if (options) { | |
778 | request.options = options->data; | |
779 | request.options_len = options->size; | |
780 | } | |
781 | ||
782 | error = dpif_netlink_vport_transact(&request, &reply, &buf); | |
783 | if (!error) { | |
784 | *port_nop = reply.port_no; | |
785 | } else { | |
786 | if (error == EBUSY && *port_nop != ODPP_NONE) { | |
787 | VLOG_INFO("%s: requested port %"PRIu32" is in use", | |
788 | dpif_name(&dpif->dpif), *port_nop); | |
789 | } | |
790 | ||
622ea8fd | 791 | close_nl_sock(sock); |
20c57607 EG |
792 | goto exit; |
793 | } | |
794 | ||
622ea8fd | 795 | error = vport_add_channel(dpif, *port_nop, sock); |
69c51582 MC |
796 | if (error) { |
797 | VLOG_INFO("%s: could not add channel for port %s", | |
798 | dpif_name(&dpif->dpif), name); | |
799 | ||
800 | /* Delete the port. */ | |
801 | dpif_netlink_vport_init(&request); | |
802 | request.cmd = OVS_VPORT_CMD_DEL; | |
803 | request.dp_ifindex = dpif->dp_ifindex; | |
804 | request.port_no = *port_nop; | |
805 | dpif_netlink_vport_transact(&request, NULL, NULL); | |
622ea8fd | 806 | close_nl_sock(sock); |
69c51582 | 807 | goto exit; |
20c57607 | 808 | } |
20c57607 EG |
809 | |
810 | exit: | |
811 | ofpbuf_delete(buf); | |
20c57607 EG |
812 | |
813 | return error; | |
814 | } | |
815 | ||
816 | static int | |
817 | dpif_netlink_port_add_compat(struct dpif_netlink *dpif, struct netdev *netdev, | |
818 | odp_port_t *port_nop) | |
819 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
820 | { | |
821 | const struct netdev_tunnel_config *tnl_cfg; | |
822 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; | |
823 | const char *type = netdev_get_type(netdev); | |
824 | uint64_t options_stub[64 / 8]; | |
825 | enum ovs_vport_type ovs_type; | |
826 | struct ofpbuf options; | |
827 | const char *name; | |
828 | ||
829 | name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); | |
830 | ||
831 | ovs_type = netdev_to_ovs_vport_type(netdev_get_type(netdev)); | |
832 | if (ovs_type == OVS_VPORT_TYPE_UNSPEC) { | |
c283069c BP |
833 | VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has " |
834 | "unsupported type `%s'", | |
9b00386b | 835 | dpif_name(&dpif->dpif), name, type); |
c283069c BP |
836 | return EINVAL; |
837 | } | |
c3827f61 | 838 | |
20c57607 | 839 | if (ovs_type == OVS_VPORT_TYPE_NETDEV) { |
93451a0a | 840 | #ifdef _WIN32 |
09cac43f | 841 | /* XXX : Map appropiate Windows handle */ |
93451a0a | 842 | #else |
24b019f8 | 843 | netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false); |
93451a0a | 844 | #endif |
24b019f8 JP |
845 | } |
846 | ||
da467899 | 847 | #ifdef _WIN32 |
20c57607 | 848 | if (ovs_type == OVS_VPORT_TYPE_INTERNAL) { |
da467899 AS |
849 | if (!create_wmi_port(name)){ |
850 | VLOG_ERR("Could not create wmi internal port with name:%s", name); | |
da467899 AS |
851 | return EINVAL; |
852 | }; | |
853 | } | |
854 | #endif | |
855 | ||
26508d9a | 856 | tnl_cfg = netdev_get_tunnel_config(netdev); |
526df7d8 | 857 | if (tnl_cfg && (tnl_cfg->dst_port != 0 || tnl_cfg->exts)) { |
26508d9a | 858 | ofpbuf_use_stack(&options, options_stub, sizeof options_stub); |
526df7d8 TG |
859 | if (tnl_cfg->dst_port) { |
860 | nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT, | |
861 | ntohs(tnl_cfg->dst_port)); | |
862 | } | |
863 | if (tnl_cfg->exts) { | |
864 | size_t ext_ofs; | |
865 | int i; | |
866 | ||
867 | ext_ofs = nl_msg_start_nested(&options, OVS_TUNNEL_ATTR_EXTENSION); | |
868 | for (i = 0; i < 32; i++) { | |
869 | if (tnl_cfg->exts & (1 << i)) { | |
870 | nl_msg_put_flag(&options, i); | |
871 | } | |
872 | } | |
873 | nl_msg_end_nested(&options, ext_ofs); | |
874 | } | |
20c57607 EG |
875 | return dpif_netlink_port_add__(dpif, name, ovs_type, &options, |
876 | port_nop); | |
2510ba7c | 877 | } else { |
20c57607 | 878 | return dpif_netlink_port_add__(dpif, name, ovs_type, NULL, port_nop); |
78a2d59c | 879 | } |
c3827f61 | 880 | |
20c57607 | 881 | } |
989fd548 | 882 | |
921c370a | 883 | static int |
c4e08753 EG |
884 | dpif_netlink_rtnl_port_create_and_add(struct dpif_netlink *dpif, |
885 | struct netdev *netdev, | |
886 | odp_port_t *port_nop) | |
887 | OVS_REQ_WRLOCK(dpif->upcall_lock) | |
888 | { | |
889 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
890 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; | |
891 | const char *name; | |
892 | int error; | |
989fd548 | 893 | |
c4e08753 EG |
894 | error = dpif_netlink_rtnl_port_create(netdev); |
895 | if (error) { | |
896 | if (error != EOPNOTSUPP) { | |
d52ef4eb | 897 | VLOG_WARN_RL(&rl, "Failed to create %s with rtnetlink: %s", |
c4e08753 EG |
898 | netdev_get_name(netdev), ovs_strerror(error)); |
899 | } | |
900 | return error; | |
901 | } | |
1579cf67 | 902 | |
c4e08753 EG |
903 | name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
904 | error = dpif_netlink_port_add__(dpif, name, OVS_VPORT_TYPE_NETDEV, NULL, | |
905 | port_nop); | |
c37cb3ee | 906 | if (error) { |
c4e08753 EG |
907 | dpif_netlink_rtnl_port_destroy(name, netdev_get_type(netdev)); |
908 | } | |
909 | return error; | |
910 | } | |
96fba48f BP |
911 | |
912 | static int | |
93451a0a AS |
913 | dpif_netlink_port_add(struct dpif *dpif_, struct netdev *netdev, |
914 | odp_port_t *port_nop) | |
9fafa796 | 915 | { |
93451a0a | 916 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
921c370a | 917 | int error = EOPNOTSUPP; |
9fafa796 | 918 | |
1579cf67 | 919 | fat_rwlock_wrlock(&dpif->upcall_lock); |
921c370a EG |
920 | if (!ovs_tunnels_out_of_tree) { |
921 | error = dpif_netlink_rtnl_port_create_and_add(dpif, netdev, port_nop); | |
922 | } | |
c37cb3ee | 923 | if (error) { |
921c370a EG |
924 | error = dpif_netlink_port_add_compat(dpif, netdev, port_nop); |
925 | } | |
1579cf67 | 926 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
927 | |
928 | return error; | |
929 | } | |
930 | ||
931 | static int | |
93451a0a | 932 | dpif_netlink_port_del__(struct dpif_netlink *dpif, odp_port_t port_no) |
b90de034 | 933 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 934 | { |
93451a0a | 935 | struct dpif_netlink_vport vport; |
921c370a | 936 | struct dpif_port dpif_port; |
773cd538 | 937 | int error; |
c19e6535 | 938 | |
921c370a EG |
939 | error = dpif_netlink_port_query__(dpif, port_no, NULL, &dpif_port); |
940 | if (error) { | |
941 | return error; | |
942 | } | |
943 | ||
93451a0a | 944 | dpif_netlink_vport_init(&vport); |
df2c07f4 | 945 | vport.cmd = OVS_VPORT_CMD_DEL; |
254f2dc8 | 946 | vport.dp_ifindex = dpif->dp_ifindex; |
c19e6535 | 947 | vport.port_no = port_no; |
da467899 | 948 | #ifdef _WIN32 |
921c370a EG |
949 | if (!strcmp(dpif_port.type, "internal")) { |
950 | if (!delete_wmi_port(dpif_port.name)) { | |
da467899 | 951 | VLOG_ERR("Could not delete wmi port with name: %s", |
921c370a | 952 | dpif_port.name); |
da467899 AS |
953 | }; |
954 | } | |
955 | #endif | |
93451a0a | 956 | error = dpif_netlink_vport_transact(&vport, NULL, NULL); |
773cd538 | 957 | |
1579cf67 | 958 | vport_del_channels(dpif, port_no); |
989fd548 | 959 | |
921c370a EG |
960 | if (!error && !ovs_tunnels_out_of_tree) { |
961 | error = dpif_netlink_rtnl_port_destroy(dpif_port.name, dpif_port.type); | |
962 | if (error == EOPNOTSUPP) { | |
963 | error = 0; | |
964 | } | |
965 | } | |
966 | ||
967 | dpif_port_destroy(&dpif_port); | |
968 | ||
773cd538 | 969 | return error; |
c3827f61 | 970 | } |
3abc4a1a | 971 | |
9fafa796 | 972 | static int |
93451a0a | 973 | dpif_netlink_port_del(struct dpif *dpif_, odp_port_t port_no) |
9fafa796 | 974 | { |
93451a0a | 975 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
976 | int error; |
977 | ||
1579cf67 | 978 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 979 | error = dpif_netlink_port_del__(dpif, port_no); |
1579cf67 | 980 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
981 | |
982 | return error; | |
983 | } | |
984 | ||
c3827f61 | 985 | static int |
93451a0a AS |
986 | dpif_netlink_port_query__(const struct dpif_netlink *dpif, odp_port_t port_no, |
987 | const char *port_name, struct dpif_port *dpif_port) | |
c3827f61 | 988 | { |
93451a0a AS |
989 | struct dpif_netlink_vport request; |
990 | struct dpif_netlink_vport reply; | |
c19e6535 | 991 | struct ofpbuf *buf; |
4c738a8d BP |
992 | int error; |
993 | ||
93451a0a | 994 | dpif_netlink_vport_init(&request); |
df2c07f4 | 995 | request.cmd = OVS_VPORT_CMD_GET; |
9b00386b | 996 | request.dp_ifindex = dpif->dp_ifindex; |
c19e6535 BP |
997 | request.port_no = port_no; |
998 | request.name = port_name; | |
4c738a8d | 999 | |
93451a0a | 1000 | error = dpif_netlink_vport_transact(&request, &reply, &buf); |
c19e6535 | 1001 | if (!error) { |
33db1592 BP |
1002 | if (reply.dp_ifindex != request.dp_ifindex) { |
1003 | /* A query by name reported that 'port_name' is in some datapath | |
1004 | * other than 'dpif', but the caller wants to know about 'dpif'. */ | |
1005 | error = ENODEV; | |
4afba28d | 1006 | } else if (dpif_port) { |
33db1592 | 1007 | dpif_port->name = xstrdup(reply.name); |
b9ad7294 | 1008 | dpif_port->type = xstrdup(get_vport_type(&reply)); |
33db1592 BP |
1009 | dpif_port->port_no = reply.port_no; |
1010 | } | |
c19e6535 | 1011 | ofpbuf_delete(buf); |
3abc4a1a | 1012 | } |
c19e6535 | 1013 | return error; |
96fba48f BP |
1014 | } |
1015 | ||
1016 | static int | |
93451a0a AS |
1017 | dpif_netlink_port_query_by_number(const struct dpif *dpif_, odp_port_t port_no, |
1018 | struct dpif_port *dpif_port) | |
96fba48f | 1019 | { |
93451a0a | 1020 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9b00386b | 1021 | |
93451a0a | 1022 | return dpif_netlink_port_query__(dpif, port_no, NULL, dpif_port); |
96fba48f BP |
1023 | } |
1024 | ||
1025 | static int | |
93451a0a | 1026 | dpif_netlink_port_query_by_name(const struct dpif *dpif_, const char *devname, |
4c738a8d | 1027 | struct dpif_port *dpif_port) |
96fba48f | 1028 | { |
93451a0a | 1029 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9b00386b | 1030 | |
93451a0a | 1031 | return dpif_netlink_port_query__(dpif, 0, devname, dpif_port); |
96fba48f BP |
1032 | } |
1033 | ||
98403001 | 1034 | static uint32_t |
93451a0a | 1035 | dpif_netlink_port_get_pid__(const struct dpif_netlink *dpif, |
769b5034 | 1036 | odp_port_t port_no) |
b90de034 | 1037 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
98403001 | 1038 | { |
4e022ec0 | 1039 | uint32_t port_idx = odp_to_u32(port_no); |
9fafa796 | 1040 | uint32_t pid = 0; |
98403001 | 1041 | |
f8fc5489 | 1042 | if (dpif->handlers && dpif->uc_array_size > 0) { |
4e022ec0 | 1043 | /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s |
989fd548 | 1044 | * channel, since it is not heavily loaded. */ |
4e022ec0 | 1045 | uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx; |
1579cf67 | 1046 | |
17f2748d AW |
1047 | /* Needs to check in case the socket pointer is changed in between |
1048 | * the holding of upcall_lock. A known case happens when the main | |
1049 | * thread deletes the vport while the handler thread is handling | |
1050 | * the upcall from that port. */ | |
69c51582 MC |
1051 | if (dpif->channels[idx].sock) { |
1052 | pid = nl_sock_pid(dpif->channels[idx].sock); | |
17f2748d | 1053 | } |
98403001 | 1054 | } |
9fafa796 BP |
1055 | |
1056 | return pid; | |
98403001 BP |
1057 | } |
1058 | ||
b90de034 | 1059 | static uint32_t |
769b5034 | 1060 | dpif_netlink_port_get_pid(const struct dpif *dpif_, odp_port_t port_no) |
b90de034 | 1061 | { |
93451a0a | 1062 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
b90de034 AW |
1063 | uint32_t ret; |
1064 | ||
1065 | fat_rwlock_rdlock(&dpif->upcall_lock); | |
769b5034 | 1066 | ret = dpif_netlink_port_get_pid__(dpif, port_no); |
b90de034 AW |
1067 | fat_rwlock_unlock(&dpif->upcall_lock); |
1068 | ||
1069 | return ret; | |
1070 | } | |
1071 | ||
96fba48f | 1072 | static int |
93451a0a | 1073 | dpif_netlink_flow_flush(struct dpif *dpif_) |
96fba48f | 1074 | { |
93451a0a AS |
1075 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1076 | struct dpif_netlink_flow flow; | |
37a1300c | 1077 | |
93451a0a | 1078 | dpif_netlink_flow_init(&flow); |
df2c07f4 | 1079 | flow.cmd = OVS_FLOW_CMD_DEL; |
254f2dc8 | 1080 | flow.dp_ifindex = dpif->dp_ifindex; |
f7dde6df PB |
1081 | |
1082 | if (netdev_is_flow_api_enabled()) { | |
dfaf79dd | 1083 | netdev_ports_flow_flush(dpif_->dpif_class); |
f7dde6df PB |
1084 | } |
1085 | ||
93451a0a | 1086 | return dpif_netlink_flow_transact(&flow, NULL, NULL); |
96fba48f BP |
1087 | } |
1088 | ||
93451a0a | 1089 | struct dpif_netlink_port_state { |
f0fef760 | 1090 | struct nl_dump dump; |
d57695d7 | 1091 | struct ofpbuf buf; |
c19e6535 BP |
1092 | }; |
1093 | ||
222837c4 | 1094 | static void |
93451a0a AS |
1095 | dpif_netlink_port_dump_start__(const struct dpif_netlink *dpif, |
1096 | struct nl_dump *dump) | |
96fba48f | 1097 | { |
93451a0a | 1098 | struct dpif_netlink_vport request; |
f0fef760 BP |
1099 | struct ofpbuf *buf; |
1100 | ||
93451a0a | 1101 | dpif_netlink_vport_init(&request); |
067f1e23 | 1102 | request.cmd = OVS_VPORT_CMD_GET; |
254f2dc8 | 1103 | request.dp_ifindex = dpif->dp_ifindex; |
f0fef760 BP |
1104 | |
1105 | buf = ofpbuf_new(1024); | |
93451a0a | 1106 | dpif_netlink_vport_to_ofpbuf(&request, buf); |
222837c4 | 1107 | nl_dump_start(dump, NETLINK_GENERIC, buf); |
f0fef760 | 1108 | ofpbuf_delete(buf); |
222837c4 BP |
1109 | } |
1110 | ||
1111 | static int | |
93451a0a | 1112 | dpif_netlink_port_dump_start(const struct dpif *dpif_, void **statep) |
222837c4 | 1113 | { |
93451a0a AS |
1114 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1115 | struct dpif_netlink_port_state *state; | |
222837c4 BP |
1116 | |
1117 | *statep = state = xmalloc(sizeof *state); | |
93451a0a | 1118 | dpif_netlink_port_dump_start__(dpif, &state->dump); |
f0fef760 | 1119 | |
d57695d7 | 1120 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); |
b0ec0f27 BP |
1121 | return 0; |
1122 | } | |
1123 | ||
7c1ef244 | 1124 | static int |
93451a0a AS |
1125 | dpif_netlink_port_dump_next__(const struct dpif_netlink *dpif, |
1126 | struct nl_dump *dump, | |
1127 | struct dpif_netlink_vport *vport, | |
1128 | struct ofpbuf *buffer) | |
222837c4 | 1129 | { |
222837c4 BP |
1130 | struct ofpbuf buf; |
1131 | int error; | |
1132 | ||
d57695d7 | 1133 | if (!nl_dump_next(dump, &buf, buffer)) { |
222837c4 BP |
1134 | return EOF; |
1135 | } | |
1136 | ||
93451a0a | 1137 | error = dpif_netlink_vport_from_ofpbuf(vport, &buf); |
222837c4 BP |
1138 | if (error) { |
1139 | VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)", | |
1140 | dpif_name(&dpif->dpif), ovs_strerror(error)); | |
1141 | } | |
1142 | return error; | |
1143 | } | |
1144 | ||
b0ec0f27 | 1145 | static int |
93451a0a AS |
1146 | dpif_netlink_port_dump_next(const struct dpif *dpif_, void *state_, |
1147 | struct dpif_port *dpif_port) | |
b0ec0f27 | 1148 | { |
93451a0a AS |
1149 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1150 | struct dpif_netlink_port_state *state = state_; | |
1151 | struct dpif_netlink_vport vport; | |
96fba48f BP |
1152 | int error; |
1153 | ||
93451a0a AS |
1154 | error = dpif_netlink_port_dump_next__(dpif, &state->dump, &vport, |
1155 | &state->buf); | |
c3827f61 | 1156 | if (error) { |
f0fef760 | 1157 | return error; |
c3827f61 | 1158 | } |
ebc56baa | 1159 | dpif_port->name = CONST_CAST(char *, vport.name); |
b9ad7294 | 1160 | dpif_port->type = CONST_CAST(char *, get_vport_type(&vport)); |
f0fef760 BP |
1161 | dpif_port->port_no = vport.port_no; |
1162 | return 0; | |
b0ec0f27 BP |
1163 | } |
1164 | ||
1165 | static int | |
93451a0a | 1166 | dpif_netlink_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_) |
b0ec0f27 | 1167 | { |
93451a0a | 1168 | struct dpif_netlink_port_state *state = state_; |
f0fef760 | 1169 | int error = nl_dump_done(&state->dump); |
8522b383 | 1170 | |
d57695d7 | 1171 | ofpbuf_uninit(&state->buf); |
b0ec0f27 | 1172 | free(state); |
f0fef760 | 1173 | return error; |
96fba48f BP |
1174 | } |
1175 | ||
e9e28be3 | 1176 | static int |
93451a0a | 1177 | dpif_netlink_port_poll(const struct dpif *dpif_, char **devnamep) |
e9e28be3 | 1178 | { |
93451a0a | 1179 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
e9e28be3 | 1180 | |
e4516b20 BP |
1181 | /* Lazily create the Netlink socket to listen for notifications. */ |
1182 | if (!dpif->port_notifier) { | |
1183 | struct nl_sock *sock; | |
1184 | int error; | |
1185 | ||
1186 | error = nl_sock_create(NETLINK_GENERIC, &sock); | |
1187 | if (error) { | |
1188 | return error; | |
1189 | } | |
1190 | ||
1191 | error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup); | |
1192 | if (error) { | |
1193 | nl_sock_destroy(sock); | |
1194 | return error; | |
1195 | } | |
1196 | dpif->port_notifier = sock; | |
1197 | ||
1198 | /* We have no idea of the current state so report that everything | |
1199 | * changed. */ | |
1200 | return ENOBUFS; | |
1201 | } | |
1202 | ||
1203 | for (;;) { | |
1204 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1205 | uint64_t buf_stub[4096 / 8]; | |
1206 | struct ofpbuf buf; | |
1207 | int error; | |
1208 | ||
1209 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
a86bd14e | 1210 | error = nl_sock_recv(dpif->port_notifier, &buf, NULL, false); |
e4516b20 | 1211 | if (!error) { |
93451a0a | 1212 | struct dpif_netlink_vport vport; |
e4516b20 | 1213 | |
93451a0a | 1214 | error = dpif_netlink_vport_from_ofpbuf(&vport, &buf); |
e4516b20 BP |
1215 | if (!error) { |
1216 | if (vport.dp_ifindex == dpif->dp_ifindex | |
1217 | && (vport.cmd == OVS_VPORT_CMD_NEW | |
1218 | || vport.cmd == OVS_VPORT_CMD_DEL | |
1219 | || vport.cmd == OVS_VPORT_CMD_SET)) { | |
1220 | VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8, | |
1221 | dpif->dpif.full_name, vport.name, vport.cmd); | |
1579cf67 | 1222 | if (vport.cmd == OVS_VPORT_CMD_DEL && dpif->handlers) { |
61eae437 BP |
1223 | dpif->refresh_channels = true; |
1224 | } | |
e4516b20 | 1225 | *devnamep = xstrdup(vport.name); |
59e0c910 | 1226 | ofpbuf_uninit(&buf); |
e4516b20 | 1227 | return 0; |
e4516b20 BP |
1228 | } |
1229 | } | |
59e0c910 BP |
1230 | } else if (error != EAGAIN) { |
1231 | VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", | |
1232 | ovs_strerror(error)); | |
1233 | nl_sock_drain(dpif->port_notifier); | |
1234 | error = ENOBUFS; | |
e4516b20 BP |
1235 | } |
1236 | ||
59e0c910 BP |
1237 | ofpbuf_uninit(&buf); |
1238 | if (error) { | |
1239 | return error; | |
1240 | } | |
e9e28be3 | 1241 | } |
e9e28be3 BP |
1242 | } |
1243 | ||
1244 | static void | |
93451a0a | 1245 | dpif_netlink_port_poll_wait(const struct dpif *dpif_) |
e9e28be3 | 1246 | { |
93451a0a | 1247 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
e4516b20 BP |
1248 | |
1249 | if (dpif->port_notifier) { | |
1250 | nl_sock_wait(dpif->port_notifier, POLLIN); | |
1251 | } else { | |
e9e28be3 | 1252 | poll_immediate_wake(); |
e9e28be3 BP |
1253 | } |
1254 | } | |
1255 | ||
6fe09f8c | 1256 | static void |
70e5ed6f JS |
1257 | dpif_netlink_flow_init_ufid(struct dpif_netlink_flow *request, |
1258 | const ovs_u128 *ufid, bool terse) | |
1259 | { | |
1260 | if (ufid) { | |
1261 | request->ufid = *ufid; | |
1262 | request->ufid_present = true; | |
1263 | } else { | |
1264 | request->ufid_present = false; | |
1265 | } | |
1266 | request->ufid_terse = terse; | |
1267 | } | |
1268 | ||
1269 | static void | |
1270 | dpif_netlink_init_flow_get__(const struct dpif_netlink *dpif, | |
1271 | const struct nlattr *key, size_t key_len, | |
1272 | const ovs_u128 *ufid, bool terse, | |
1273 | struct dpif_netlink_flow *request) | |
96fba48f | 1274 | { |
93451a0a | 1275 | dpif_netlink_flow_init(request); |
6fe09f8c JS |
1276 | request->cmd = OVS_FLOW_CMD_GET; |
1277 | request->dp_ifindex = dpif->dp_ifindex; | |
1278 | request->key = key; | |
1279 | request->key_len = key_len; | |
70e5ed6f JS |
1280 | dpif_netlink_flow_init_ufid(request, ufid, terse); |
1281 | } | |
1282 | ||
1283 | static void | |
1284 | dpif_netlink_init_flow_get(const struct dpif_netlink *dpif, | |
1285 | const struct dpif_flow_get *get, | |
1286 | struct dpif_netlink_flow *request) | |
1287 | { | |
1288 | dpif_netlink_init_flow_get__(dpif, get->key, get->key_len, get->ufid, | |
1289 | false, request); | |
30053024 BP |
1290 | } |
1291 | ||
1292 | static int | |
70e5ed6f JS |
1293 | dpif_netlink_flow_get__(const struct dpif_netlink *dpif, |
1294 | const struct nlattr *key, size_t key_len, | |
1295 | const ovs_u128 *ufid, bool terse, | |
1296 | struct dpif_netlink_flow *reply, struct ofpbuf **bufp) | |
30053024 | 1297 | { |
93451a0a | 1298 | struct dpif_netlink_flow request; |
30053024 | 1299 | |
70e5ed6f | 1300 | dpif_netlink_init_flow_get__(dpif, key, key_len, ufid, terse, &request); |
93451a0a | 1301 | return dpif_netlink_flow_transact(&request, reply, bufp); |
96fba48f BP |
1302 | } |
1303 | ||
70e5ed6f JS |
1304 | static int |
1305 | dpif_netlink_flow_get(const struct dpif_netlink *dpif, | |
1306 | const struct dpif_netlink_flow *flow, | |
1307 | struct dpif_netlink_flow *reply, struct ofpbuf **bufp) | |
1308 | { | |
1309 | return dpif_netlink_flow_get__(dpif, flow->key, flow->key_len, | |
1310 | flow->ufid_present ? &flow->ufid : NULL, | |
1311 | false, reply, bufp); | |
1312 | } | |
1313 | ||
6bc60024 | 1314 | static void |
93451a0a AS |
1315 | dpif_netlink_init_flow_put(struct dpif_netlink *dpif, |
1316 | const struct dpif_flow_put *put, | |
1317 | struct dpif_netlink_flow *request) | |
6bc60024 | 1318 | { |
d64e176c | 1319 | static const struct nlattr dummy_action; |
6bc60024 | 1320 | |
93451a0a | 1321 | dpif_netlink_flow_init(request); |
89625d1e | 1322 | request->cmd = (put->flags & DPIF_FP_CREATE |
6bc60024 BP |
1323 | ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET); |
1324 | request->dp_ifindex = dpif->dp_ifindex; | |
89625d1e BP |
1325 | request->key = put->key; |
1326 | request->key_len = put->key_len; | |
e6cc0bab AZ |
1327 | request->mask = put->mask; |
1328 | request->mask_len = put->mask_len; | |
70e5ed6f JS |
1329 | dpif_netlink_flow_init_ufid(request, put->ufid, false); |
1330 | ||
6bc60024 | 1331 | /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */ |
d64e176c BP |
1332 | request->actions = (put->actions |
1333 | ? put->actions | |
1334 | : CONST_CAST(struct nlattr *, &dummy_action)); | |
89625d1e BP |
1335 | request->actions_len = put->actions_len; |
1336 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
6bc60024 BP |
1337 | request->clear = true; |
1338 | } | |
43f9ac0a JR |
1339 | if (put->flags & DPIF_FP_PROBE) { |
1340 | request->probe = true; | |
1341 | } | |
89625d1e | 1342 | request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; |
6bc60024 BP |
1343 | } |
1344 | ||
b99d3cee | 1345 | static void |
70e5ed6f JS |
1346 | dpif_netlink_init_flow_del__(struct dpif_netlink *dpif, |
1347 | const struct nlattr *key, size_t key_len, | |
1348 | const ovs_u128 *ufid, bool terse, | |
1349 | struct dpif_netlink_flow *request) | |
96fba48f | 1350 | { |
93451a0a | 1351 | dpif_netlink_flow_init(request); |
b99d3cee BP |
1352 | request->cmd = OVS_FLOW_CMD_DEL; |
1353 | request->dp_ifindex = dpif->dp_ifindex; | |
70e5ed6f JS |
1354 | request->key = key; |
1355 | request->key_len = key_len; | |
1356 | dpif_netlink_flow_init_ufid(request, ufid, terse); | |
1357 | } | |
1358 | ||
1359 | static void | |
1360 | dpif_netlink_init_flow_del(struct dpif_netlink *dpif, | |
1361 | const struct dpif_flow_del *del, | |
1362 | struct dpif_netlink_flow *request) | |
1363 | { | |
37382aa6 AS |
1364 | dpif_netlink_init_flow_del__(dpif, del->key, del->key_len, |
1365 | del->ufid, del->terse, request); | |
70e5ed6f JS |
1366 | } |
1367 | ||
93451a0a | 1368 | struct dpif_netlink_flow_dump { |
ac64794a BP |
1369 | struct dpif_flow_dump up; |
1370 | struct nl_dump nl_dump; | |
d2ad7ef1 | 1371 | atomic_int status; |
f2280b41 PB |
1372 | struct netdev_flow_dump **netdev_dumps; |
1373 | int netdev_dumps_num; /* Number of netdev_flow_dumps */ | |
1374 | struct ovs_mutex netdev_lock; /* Guards the following. */ | |
1375 | int netdev_current_dump OVS_GUARDED; /* Shared current dump */ | |
a692410a | 1376 | struct dpif_flow_dump_types types; /* Type of dump */ |
e723fd32 JS |
1377 | }; |
1378 | ||
93451a0a AS |
1379 | static struct dpif_netlink_flow_dump * |
1380 | dpif_netlink_flow_dump_cast(struct dpif_flow_dump *dump) | |
e723fd32 | 1381 | { |
93451a0a | 1382 | return CONTAINER_OF(dump, struct dpif_netlink_flow_dump, up); |
e723fd32 JS |
1383 | } |
1384 | ||
f2280b41 PB |
1385 | static void |
1386 | start_netdev_dump(const struct dpif *dpif_, | |
1387 | struct dpif_netlink_flow_dump *dump) | |
1388 | { | |
1389 | ovs_mutex_init(&dump->netdev_lock); | |
1390 | ||
a692410a | 1391 | if (!(dump->types.netdev_flows)) { |
f2280b41 PB |
1392 | dump->netdev_dumps_num = 0; |
1393 | dump->netdev_dumps = NULL; | |
1394 | return; | |
1395 | } | |
1396 | ||
1397 | ovs_mutex_lock(&dump->netdev_lock); | |
1398 | dump->netdev_current_dump = 0; | |
1399 | dump->netdev_dumps | |
dfaf79dd | 1400 | = netdev_ports_flow_dump_create(dpif_->dpif_class, |
f2280b41 PB |
1401 | &dump->netdev_dumps_num); |
1402 | ovs_mutex_unlock(&dump->netdev_lock); | |
1403 | } | |
1404 | ||
a692410a GT |
1405 | static void |
1406 | dpif_netlink_populate_flow_dump_types(struct dpif_netlink_flow_dump *dump, | |
1407 | struct dpif_flow_dump_types *types) | |
1408 | { | |
1409 | if (!types) { | |
1410 | dump->types.ovs_flows = true; | |
1411 | dump->types.netdev_flows = true; | |
1412 | } else { | |
1413 | memcpy(&dump->types, types, sizeof *types); | |
494a7455 | 1414 | } |
7e8b7199 PB |
1415 | } |
1416 | ||
ac64794a | 1417 | static struct dpif_flow_dump * |
7e8b7199 | 1418 | dpif_netlink_flow_dump_create(const struct dpif *dpif_, bool terse, |
a692410a | 1419 | struct dpif_flow_dump_types *types) |
96fba48f | 1420 | { |
93451a0a AS |
1421 | const struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1422 | struct dpif_netlink_flow_dump *dump; | |
1423 | struct dpif_netlink_flow request; | |
37a1300c BP |
1424 | struct ofpbuf *buf; |
1425 | ||
ac64794a BP |
1426 | dump = xmalloc(sizeof *dump); |
1427 | dpif_flow_dump_init(&dump->up, dpif_); | |
37a1300c | 1428 | |
a692410a | 1429 | dpif_netlink_populate_flow_dump_types(dump, types); |
37a1300c | 1430 | |
a692410a | 1431 | if (dump->types.ovs_flows) { |
7e8b7199 PB |
1432 | dpif_netlink_flow_init(&request); |
1433 | request.cmd = OVS_FLOW_CMD_GET; | |
1434 | request.dp_ifindex = dpif->dp_ifindex; | |
1435 | request.ufid_present = false; | |
1436 | request.ufid_terse = terse; | |
1437 | ||
1438 | buf = ofpbuf_new(1024); | |
1439 | dpif_netlink_flow_to_ofpbuf(&request, buf); | |
1440 | nl_dump_start(&dump->nl_dump, NETLINK_GENERIC, buf); | |
1441 | ofpbuf_delete(buf); | |
1442 | } | |
ac64794a | 1443 | atomic_init(&dump->status, 0); |
64bb477f | 1444 | dump->up.terse = terse; |
30053024 | 1445 | |
f2280b41 PB |
1446 | start_netdev_dump(dpif_, dump); |
1447 | ||
ac64794a | 1448 | return &dump->up; |
704a1e09 BP |
1449 | } |
1450 | ||
1451 | static int | |
93451a0a | 1452 | dpif_netlink_flow_dump_destroy(struct dpif_flow_dump *dump_) |
704a1e09 | 1453 | { |
93451a0a | 1454 | struct dpif_netlink_flow_dump *dump = dpif_netlink_flow_dump_cast(dump_); |
7e8b7199 | 1455 | unsigned int nl_status = 0; |
ac64794a | 1456 | int dump_status; |
96fba48f | 1457 | |
a692410a | 1458 | if (dump->types.ovs_flows) { |
7e8b7199 PB |
1459 | nl_status = nl_dump_done(&dump->nl_dump); |
1460 | } | |
1461 | ||
f2280b41 PB |
1462 | for (int i = 0; i < dump->netdev_dumps_num; i++) { |
1463 | int err = netdev_flow_dump_destroy(dump->netdev_dumps[i]); | |
1464 | ||
1465 | if (err != 0 && err != EOPNOTSUPP) { | |
1466 | VLOG_ERR("failed dumping netdev: %s", ovs_strerror(err)); | |
1467 | } | |
1468 | } | |
1469 | ||
1470 | free(dump->netdev_dumps); | |
1471 | ovs_mutex_destroy(&dump->netdev_lock); | |
1472 | ||
7424fc44 JR |
1473 | /* No other thread has access to 'dump' at this point. */ |
1474 | atomic_read_relaxed(&dump->status, &dump_status); | |
ac64794a BP |
1475 | free(dump); |
1476 | return dump_status ? dump_status : nl_status; | |
1477 | } | |
feebdea2 | 1478 | |
93451a0a | 1479 | struct dpif_netlink_flow_dump_thread { |
ac64794a | 1480 | struct dpif_flow_dump_thread up; |
93451a0a AS |
1481 | struct dpif_netlink_flow_dump *dump; |
1482 | struct dpif_netlink_flow flow; | |
ac64794a BP |
1483 | struct dpif_flow_stats stats; |
1484 | struct ofpbuf nl_flows; /* Always used to store flows. */ | |
1485 | struct ofpbuf *nl_actions; /* Used if kernel does not supply actions. */ | |
f2280b41 PB |
1486 | int netdev_dump_idx; /* This thread current netdev dump index */ |
1487 | bool netdev_done; /* If we are finished dumping netdevs */ | |
1488 | ||
1489 | /* (Key/Mask/Actions) Buffers for netdev dumping */ | |
1490 | struct odputil_keybuf keybuf[FLOW_DUMP_MAX_BATCH]; | |
1491 | struct odputil_keybuf maskbuf[FLOW_DUMP_MAX_BATCH]; | |
1492 | struct odputil_keybuf actbuf[FLOW_DUMP_MAX_BATCH]; | |
ac64794a BP |
1493 | }; |
1494 | ||
93451a0a AS |
1495 | static struct dpif_netlink_flow_dump_thread * |
1496 | dpif_netlink_flow_dump_thread_cast(struct dpif_flow_dump_thread *thread) | |
ac64794a | 1497 | { |
93451a0a | 1498 | return CONTAINER_OF(thread, struct dpif_netlink_flow_dump_thread, up); |
ac64794a BP |
1499 | } |
1500 | ||
1501 | static struct dpif_flow_dump_thread * | |
93451a0a | 1502 | dpif_netlink_flow_dump_thread_create(struct dpif_flow_dump *dump_) |
ac64794a | 1503 | { |
93451a0a AS |
1504 | struct dpif_netlink_flow_dump *dump = dpif_netlink_flow_dump_cast(dump_); |
1505 | struct dpif_netlink_flow_dump_thread *thread; | |
ac64794a BP |
1506 | |
1507 | thread = xmalloc(sizeof *thread); | |
1508 | dpif_flow_dump_thread_init(&thread->up, &dump->up); | |
1509 | thread->dump = dump; | |
1510 | ofpbuf_init(&thread->nl_flows, NL_DUMP_BUFSIZE); | |
1511 | thread->nl_actions = NULL; | |
f2280b41 PB |
1512 | thread->netdev_dump_idx = 0; |
1513 | thread->netdev_done = !(thread->netdev_dump_idx < dump->netdev_dumps_num); | |
ac64794a BP |
1514 | |
1515 | return &thread->up; | |
1516 | } | |
1517 | ||
1518 | static void | |
93451a0a | 1519 | dpif_netlink_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread_) |
ac64794a | 1520 | { |
93451a0a AS |
1521 | struct dpif_netlink_flow_dump_thread *thread |
1522 | = dpif_netlink_flow_dump_thread_cast(thread_); | |
ac64794a BP |
1523 | |
1524 | ofpbuf_uninit(&thread->nl_flows); | |
1525 | ofpbuf_delete(thread->nl_actions); | |
1526 | free(thread); | |
1527 | } | |
1528 | ||
1529 | static void | |
7af12bd7 | 1530 | dpif_netlink_flow_to_dpif_flow(struct dpif *dpif, struct dpif_flow *dpif_flow, |
7fe98598 | 1531 | const struct dpif_netlink_flow *datapath_flow) |
ac64794a | 1532 | { |
7fe98598 NR |
1533 | dpif_flow->key = datapath_flow->key; |
1534 | dpif_flow->key_len = datapath_flow->key_len; | |
1535 | dpif_flow->mask = datapath_flow->mask; | |
1536 | dpif_flow->mask_len = datapath_flow->mask_len; | |
1537 | dpif_flow->actions = datapath_flow->actions; | |
1538 | dpif_flow->actions_len = datapath_flow->actions_len; | |
70e5ed6f | 1539 | dpif_flow->ufid_present = datapath_flow->ufid_present; |
ec97c2df | 1540 | dpif_flow->pmd_id = PMD_ID_NULL; |
70e5ed6f JS |
1541 | if (datapath_flow->ufid_present) { |
1542 | dpif_flow->ufid = datapath_flow->ufid; | |
1543 | } else { | |
1544 | ovs_assert(datapath_flow->key && datapath_flow->key_len); | |
1545 | dpif_flow_hash(dpif, datapath_flow->key, datapath_flow->key_len, | |
1546 | &dpif_flow->ufid); | |
1547 | } | |
7fe98598 | 1548 | dpif_netlink_flow_get_stats(datapath_flow, &dpif_flow->stats); |
d63ca532 GT |
1549 | dpif_flow->attrs.offloaded = false; |
1550 | dpif_flow->attrs.dp_layer = "ovs"; | |
ac64794a BP |
1551 | } |
1552 | ||
f2280b41 PB |
1553 | /* The design is such that all threads are working together on the first dump |
1554 | * to the last, in order (at first they all on dump 0). | |
1555 | * When the first thread finds that the given dump is finished, | |
1556 | * they all move to the next. If two or more threads find the same dump | |
1557 | * is finished at the same time, the first one will advance the shared | |
1558 | * netdev_current_dump and the others will catch up. */ | |
1559 | static void | |
1560 | dpif_netlink_advance_netdev_dump(struct dpif_netlink_flow_dump_thread *thread) | |
1561 | { | |
1562 | struct dpif_netlink_flow_dump *dump = thread->dump; | |
1563 | ||
1564 | ovs_mutex_lock(&dump->netdev_lock); | |
1565 | /* if we haven't finished (dumped everything) */ | |
1566 | if (dump->netdev_current_dump < dump->netdev_dumps_num) { | |
1567 | /* if we are the first to find that current dump is finished | |
1568 | * advance it. */ | |
1569 | if (thread->netdev_dump_idx == dump->netdev_current_dump) { | |
1570 | thread->netdev_dump_idx = ++dump->netdev_current_dump; | |
1571 | /* did we just finish the last dump? done. */ | |
1572 | if (dump->netdev_current_dump == dump->netdev_dumps_num) { | |
1573 | thread->netdev_done = true; | |
1574 | } | |
1575 | } else { | |
1576 | /* otherwise, we are behind, catch up */ | |
1577 | thread->netdev_dump_idx = dump->netdev_current_dump; | |
1578 | } | |
1579 | } else { | |
1580 | /* some other thread finished */ | |
1581 | thread->netdev_done = true; | |
1582 | } | |
1583 | ovs_mutex_unlock(&dump->netdev_lock); | |
1584 | } | |
1585 | ||
1586 | static int | |
1587 | dpif_netlink_netdev_match_to_dpif_flow(struct match *match, | |
1588 | struct ofpbuf *key_buf, | |
1589 | struct ofpbuf *mask_buf, | |
1590 | struct nlattr *actions, | |
1591 | struct dpif_flow_stats *stats, | |
d63ca532 | 1592 | struct dpif_flow_attrs *attrs, |
f2280b41 PB |
1593 | ovs_u128 *ufid, |
1594 | struct dpif_flow *flow, | |
1595 | bool terse OVS_UNUSED) | |
1596 | { | |
1597 | ||
1598 | struct odp_flow_key_parms odp_parms = { | |
1599 | .flow = &match->flow, | |
1600 | .mask = &match->wc.masks, | |
1601 | .support = { | |
f9885dc5 | 1602 | .max_vlan_headers = 2, |
f2280b41 PB |
1603 | }, |
1604 | }; | |
1605 | size_t offset; | |
1606 | ||
1607 | memset(flow, 0, sizeof *flow); | |
1608 | ||
1609 | /* Key */ | |
1610 | offset = key_buf->size; | |
1611 | flow->key = ofpbuf_tail(key_buf); | |
1612 | odp_flow_key_from_flow(&odp_parms, key_buf); | |
1613 | flow->key_len = key_buf->size - offset; | |
1614 | ||
1615 | /* Mask */ | |
1616 | offset = mask_buf->size; | |
1617 | flow->mask = ofpbuf_tail(mask_buf); | |
1618 | odp_parms.key_buf = key_buf; | |
1619 | odp_flow_key_from_mask(&odp_parms, mask_buf); | |
1620 | flow->mask_len = mask_buf->size - offset; | |
1621 | ||
1622 | /* Actions */ | |
1623 | flow->actions = nl_attr_get(actions); | |
1624 | flow->actions_len = nl_attr_get_size(actions); | |
1625 | ||
1626 | /* Stats */ | |
1627 | memcpy(&flow->stats, stats, sizeof *stats); | |
1628 | ||
1629 | /* UFID */ | |
1630 | flow->ufid_present = true; | |
1631 | flow->ufid = *ufid; | |
1632 | ||
1633 | flow->pmd_id = PMD_ID_NULL; | |
4742003c | 1634 | |
d63ca532 | 1635 | memcpy(&flow->attrs, attrs, sizeof *attrs); |
4742003c | 1636 | |
f2280b41 PB |
1637 | return 0; |
1638 | } | |
1639 | ||
ac64794a | 1640 | static int |
93451a0a AS |
1641 | dpif_netlink_flow_dump_next(struct dpif_flow_dump_thread *thread_, |
1642 | struct dpif_flow *flows, int max_flows) | |
ac64794a | 1643 | { |
93451a0a AS |
1644 | struct dpif_netlink_flow_dump_thread *thread |
1645 | = dpif_netlink_flow_dump_thread_cast(thread_); | |
1646 | struct dpif_netlink_flow_dump *dump = thread->dump; | |
1647 | struct dpif_netlink *dpif = dpif_netlink_cast(thread->up.dpif); | |
ac64794a BP |
1648 | int n_flows; |
1649 | ||
1650 | ofpbuf_delete(thread->nl_actions); | |
1651 | thread->nl_actions = NULL; | |
1652 | ||
1653 | n_flows = 0; | |
f2280b41 PB |
1654 | max_flows = MIN(max_flows, FLOW_DUMP_MAX_BATCH); |
1655 | ||
1656 | while (!thread->netdev_done && n_flows < max_flows) { | |
1657 | struct odputil_keybuf *maskbuf = &thread->maskbuf[n_flows]; | |
1658 | struct odputil_keybuf *keybuf = &thread->keybuf[n_flows]; | |
1659 | struct odputil_keybuf *actbuf = &thread->actbuf[n_flows]; | |
1660 | struct ofpbuf key, mask, act; | |
1661 | struct dpif_flow *f = &flows[n_flows]; | |
1662 | int cur = thread->netdev_dump_idx; | |
1663 | struct netdev_flow_dump *netdev_dump = dump->netdev_dumps[cur]; | |
1664 | struct match match; | |
1665 | struct nlattr *actions; | |
1666 | struct dpif_flow_stats stats; | |
d63ca532 | 1667 | struct dpif_flow_attrs attrs; |
f2280b41 PB |
1668 | ovs_u128 ufid; |
1669 | bool has_next; | |
1670 | ||
1671 | ofpbuf_use_stack(&key, keybuf, sizeof *keybuf); | |
1672 | ofpbuf_use_stack(&act, actbuf, sizeof *actbuf); | |
1673 | ofpbuf_use_stack(&mask, maskbuf, sizeof *maskbuf); | |
1674 | has_next = netdev_flow_dump_next(netdev_dump, &match, | |
d63ca532 | 1675 | &actions, &stats, &attrs, |
f2280b41 PB |
1676 | &ufid, |
1677 | &thread->nl_flows, | |
1678 | &act); | |
1679 | if (has_next) { | |
1680 | dpif_netlink_netdev_match_to_dpif_flow(&match, | |
1681 | &key, &mask, | |
1682 | actions, | |
1683 | &stats, | |
d63ca532 | 1684 | &attrs, |
f2280b41 PB |
1685 | &ufid, |
1686 | f, | |
1687 | dump->up.terse); | |
1688 | n_flows++; | |
1689 | } else { | |
1690 | dpif_netlink_advance_netdev_dump(thread); | |
1691 | } | |
1692 | } | |
1693 | ||
a692410a | 1694 | if (!(dump->types.ovs_flows)) { |
7e8b7199 PB |
1695 | return n_flows; |
1696 | } | |
1697 | ||
ac64794a | 1698 | while (!n_flows |
6fd6ed71 | 1699 | || (n_flows < max_flows && thread->nl_flows.size)) { |
7fe98598 | 1700 | struct dpif_netlink_flow datapath_flow; |
ac64794a BP |
1701 | struct ofpbuf nl_flow; |
1702 | int error; | |
1703 | ||
1704 | /* Try to grab another flow. */ | |
1705 | if (!nl_dump_next(&dump->nl_dump, &nl_flow, &thread->nl_flows)) { | |
1706 | break; | |
feebdea2 | 1707 | } |
30053024 | 1708 | |
ac64794a | 1709 | /* Convert the flow to our output format. */ |
7fe98598 | 1710 | error = dpif_netlink_flow_from_ofpbuf(&datapath_flow, &nl_flow); |
30053024 | 1711 | if (error) { |
7424fc44 | 1712 | atomic_store_relaxed(&dump->status, error); |
ac64794a | 1713 | break; |
feebdea2 | 1714 | } |
30053024 | 1715 | |
64bb477f JS |
1716 | if (dump->up.terse || datapath_flow.actions) { |
1717 | /* Common case: we don't want actions, or the flow includes | |
1718 | * actions. */ | |
7af12bd7 JS |
1719 | dpif_netlink_flow_to_dpif_flow(&dpif->dpif, &flows[n_flows++], |
1720 | &datapath_flow); | |
ac64794a BP |
1721 | } else { |
1722 | /* Rare case: the flow does not include actions. Retrieve this | |
1723 | * individual flow again to get the actions. */ | |
70e5ed6f | 1724 | error = dpif_netlink_flow_get(dpif, &datapath_flow, |
7fe98598 | 1725 | &datapath_flow, &thread->nl_actions); |
30053024 BP |
1726 | if (error == ENOENT) { |
1727 | VLOG_DBG("dumped flow disappeared on get"); | |
ac64794a | 1728 | continue; |
30053024 | 1729 | } else if (error) { |
10a89ef0 BP |
1730 | VLOG_WARN("error fetching dumped flow: %s", |
1731 | ovs_strerror(error)); | |
7424fc44 | 1732 | atomic_store_relaxed(&dump->status, error); |
ac64794a | 1733 | break; |
30053024 | 1734 | } |
30053024 | 1735 | |
ac64794a BP |
1736 | /* Save this flow. Then exit, because we only have one buffer to |
1737 | * handle this case. */ | |
7af12bd7 JS |
1738 | dpif_netlink_flow_to_dpif_flow(&dpif->dpif, &flows[n_flows++], |
1739 | &datapath_flow); | |
ac64794a BP |
1740 | break; |
1741 | } | |
feebdea2 | 1742 | } |
ac64794a | 1743 | return n_flows; |
96fba48f BP |
1744 | } |
1745 | ||
eabe7c68 | 1746 | static void |
93451a0a AS |
1747 | dpif_netlink_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec, |
1748 | struct ofpbuf *buf) | |
96fba48f | 1749 | { |
89625d1e | 1750 | struct ovs_header *k_exec; |
758c456d | 1751 | size_t key_ofs; |
f7cd0081 | 1752 | |
eabe7c68 | 1753 | ofpbuf_prealloc_tailroom(buf, (64 |
cf62fa4c | 1754 | + dp_packet_size(d_exec->packet) |
758c456d | 1755 | + ODP_KEY_METADATA_SIZE |
eabe7c68 | 1756 | + d_exec->actions_len)); |
f7cd0081 | 1757 | |
df2c07f4 | 1758 | nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST, |
69685a88 | 1759 | OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION); |
f7cd0081 | 1760 | |
89625d1e BP |
1761 | k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec); |
1762 | k_exec->dp_ifindex = dp_ifindex; | |
f7cd0081 | 1763 | |
89625d1e | 1764 | nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, |
cf62fa4c PS |
1765 | dp_packet_data(d_exec->packet), |
1766 | dp_packet_size(d_exec->packet)); | |
758c456d JR |
1767 | |
1768 | key_ofs = nl_msg_start_nested(buf, OVS_PACKET_ATTR_KEY); | |
beb75a40 | 1769 | odp_key_from_dp_packet(buf, d_exec->packet); |
758c456d JR |
1770 | nl_msg_end_nested(buf, key_ofs); |
1771 | ||
89625d1e BP |
1772 | nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, |
1773 | d_exec->actions, d_exec->actions_len); | |
43f9ac0a | 1774 | if (d_exec->probe) { |
2e460098 | 1775 | nl_msg_put_flag(buf, OVS_PACKET_ATTR_PROBE); |
43f9ac0a | 1776 | } |
27130224 AZ |
1777 | if (d_exec->mtu) { |
1778 | nl_msg_put_u16(buf, OVS_PACKET_ATTR_MRU, d_exec->mtu); | |
1779 | } | |
0442bfb1 TZ |
1780 | |
1781 | if (d_exec->hash) { | |
1782 | nl_msg_put_u64(buf, OVS_PACKET_ATTR_HASH, d_exec->hash); | |
1783 | } | |
6bc60024 BP |
1784 | } |
1785 | ||
0f3358ea BP |
1786 | /* Executes, against 'dpif', up to the first 'n_ops' operations in 'ops'. |
1787 | * Returns the number actually executed (at least 1, if 'n_ops' is | |
1788 | * positive). */ | |
1789 | static size_t | |
93451a0a AS |
1790 | dpif_netlink_operate__(struct dpif_netlink *dpif, |
1791 | struct dpif_op **ops, size_t n_ops) | |
6bc60024 | 1792 | { |
eabe7c68 BP |
1793 | struct op_auxdata { |
1794 | struct nl_transaction txn; | |
72d32ac0 | 1795 | |
eabe7c68 BP |
1796 | struct ofpbuf request; |
1797 | uint64_t request_stub[1024 / 8]; | |
72d32ac0 BP |
1798 | |
1799 | struct ofpbuf reply; | |
1800 | uint64_t reply_stub[1024 / 8]; | |
8b668ee3 | 1801 | } auxes[OPERATE_MAX_OPS]; |
eabe7c68 | 1802 | |
8b668ee3 | 1803 | struct nl_transaction *txnsp[OPERATE_MAX_OPS]; |
6bc60024 BP |
1804 | size_t i; |
1805 | ||
8b668ee3 | 1806 | n_ops = MIN(n_ops, OPERATE_MAX_OPS); |
6bc60024 | 1807 | for (i = 0; i < n_ops; i++) { |
eabe7c68 | 1808 | struct op_auxdata *aux = &auxes[i]; |
c2b565b5 | 1809 | struct dpif_op *op = ops[i]; |
b99d3cee BP |
1810 | struct dpif_flow_put *put; |
1811 | struct dpif_flow_del *del; | |
6fe09f8c | 1812 | struct dpif_flow_get *get; |
93451a0a | 1813 | struct dpif_netlink_flow flow; |
eabe7c68 BP |
1814 | |
1815 | ofpbuf_use_stub(&aux->request, | |
1816 | aux->request_stub, sizeof aux->request_stub); | |
1817 | aux->txn.request = &aux->request; | |
b99d3cee | 1818 | |
72d32ac0 BP |
1819 | ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub); |
1820 | aux->txn.reply = NULL; | |
1821 | ||
b99d3cee BP |
1822 | switch (op->type) { |
1823 | case DPIF_OP_FLOW_PUT: | |
fa37affa | 1824 | put = &op->flow_put; |
93451a0a | 1825 | dpif_netlink_init_flow_put(dpif, put, &flow); |
6bc60024 | 1826 | if (put->stats) { |
eabe7c68 | 1827 | flow.nlmsg_flags |= NLM_F_ECHO; |
72d32ac0 | 1828 | aux->txn.reply = &aux->reply; |
6bc60024 | 1829 | } |
93451a0a | 1830 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
b99d3cee BP |
1831 | break; |
1832 | ||
1833 | case DPIF_OP_FLOW_DEL: | |
fa37affa | 1834 | del = &op->flow_del; |
93451a0a | 1835 | dpif_netlink_init_flow_del(dpif, del, &flow); |
b99d3cee | 1836 | if (del->stats) { |
eabe7c68 | 1837 | flow.nlmsg_flags |= NLM_F_ECHO; |
72d32ac0 | 1838 | aux->txn.reply = &aux->reply; |
b99d3cee | 1839 | } |
93451a0a | 1840 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
b99d3cee | 1841 | break; |
6bc60024 | 1842 | |
b99d3cee | 1843 | case DPIF_OP_EXECUTE: |
0f3358ea BP |
1844 | /* Can't execute a packet that won't fit in a Netlink attribute. */ |
1845 | if (OVS_UNLIKELY(nl_attr_oversized( | |
fa37affa | 1846 | dp_packet_size(op->execute.packet)))) { |
0f3358ea BP |
1847 | /* Report an error immediately if this is the first operation. |
1848 | * Otherwise the easiest thing to do is to postpone to the next | |
1849 | * call (when this will be the first operation). */ | |
1850 | if (i == 0) { | |
1851 | VLOG_ERR_RL(&error_rl, | |
1852 | "dropping oversized %"PRIu32"-byte packet", | |
fa37affa | 1853 | dp_packet_size(op->execute.packet)); |
0f3358ea BP |
1854 | op->error = ENOBUFS; |
1855 | return 1; | |
1856 | } | |
1857 | n_ops = i; | |
1858 | } else { | |
fa37affa | 1859 | dpif_netlink_encode_execute(dpif->dp_ifindex, &op->execute, |
0f3358ea BP |
1860 | &aux->request); |
1861 | } | |
b99d3cee BP |
1862 | break; |
1863 | ||
6fe09f8c | 1864 | case DPIF_OP_FLOW_GET: |
fa37affa | 1865 | get = &op->flow_get; |
70e5ed6f | 1866 | dpif_netlink_init_flow_get(dpif, get, &flow); |
6fe09f8c | 1867 | aux->txn.reply = get->buffer; |
93451a0a | 1868 | dpif_netlink_flow_to_ofpbuf(&flow, &aux->request); |
6fe09f8c JS |
1869 | break; |
1870 | ||
b99d3cee | 1871 | default: |
428b2edd | 1872 | OVS_NOT_REACHED(); |
6bc60024 BP |
1873 | } |
1874 | } | |
1875 | ||
6bc60024 | 1876 | for (i = 0; i < n_ops; i++) { |
eabe7c68 | 1877 | txnsp[i] = &auxes[i].txn; |
6bc60024 | 1878 | } |
a88b4e04 | 1879 | nl_transact_multiple(NETLINK_GENERIC, txnsp, n_ops); |
6bc60024 | 1880 | |
6bc60024 | 1881 | for (i = 0; i < n_ops; i++) { |
72d32ac0 | 1882 | struct op_auxdata *aux = &auxes[i]; |
eabe7c68 | 1883 | struct nl_transaction *txn = &auxes[i].txn; |
c2b565b5 | 1884 | struct dpif_op *op = ops[i]; |
b99d3cee BP |
1885 | struct dpif_flow_put *put; |
1886 | struct dpif_flow_del *del; | |
6fe09f8c | 1887 | struct dpif_flow_get *get; |
6bc60024 | 1888 | |
b99d3cee | 1889 | op->error = txn->error; |
6bc60024 | 1890 | |
b99d3cee BP |
1891 | switch (op->type) { |
1892 | case DPIF_OP_FLOW_PUT: | |
fa37affa | 1893 | put = &op->flow_put; |
cfceb2b5 | 1894 | if (put->stats) { |
b99d3cee | 1895 | if (!op->error) { |
93451a0a | 1896 | struct dpif_netlink_flow reply; |
cfceb2b5 | 1897 | |
93451a0a AS |
1898 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, |
1899 | txn->reply); | |
cfceb2b5 | 1900 | if (!op->error) { |
93451a0a | 1901 | dpif_netlink_flow_get_stats(&reply, put->stats); |
cfceb2b5 BP |
1902 | } |
1903 | } | |
6bc60024 | 1904 | } |
b99d3cee BP |
1905 | break; |
1906 | ||
1907 | case DPIF_OP_FLOW_DEL: | |
fa37affa | 1908 | del = &op->flow_del; |
cfceb2b5 | 1909 | if (del->stats) { |
b99d3cee | 1910 | if (!op->error) { |
93451a0a | 1911 | struct dpif_netlink_flow reply; |
cfceb2b5 | 1912 | |
93451a0a AS |
1913 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, |
1914 | txn->reply); | |
cfceb2b5 | 1915 | if (!op->error) { |
93451a0a | 1916 | dpif_netlink_flow_get_stats(&reply, del->stats); |
cfceb2b5 BP |
1917 | } |
1918 | } | |
b99d3cee BP |
1919 | } |
1920 | break; | |
1921 | ||
1922 | case DPIF_OP_EXECUTE: | |
1923 | break; | |
1924 | ||
6fe09f8c | 1925 | case DPIF_OP_FLOW_GET: |
fa37affa | 1926 | get = &op->flow_get; |
6fe09f8c | 1927 | if (!op->error) { |
93451a0a | 1928 | struct dpif_netlink_flow reply; |
6fe09f8c | 1929 | |
93451a0a | 1930 | op->error = dpif_netlink_flow_from_ofpbuf(&reply, txn->reply); |
6fe09f8c | 1931 | if (!op->error) { |
7af12bd7 JS |
1932 | dpif_netlink_flow_to_dpif_flow(&dpif->dpif, get->flow, |
1933 | &reply); | |
6fe09f8c JS |
1934 | } |
1935 | } | |
1936 | break; | |
1937 | ||
b99d3cee | 1938 | default: |
428b2edd | 1939 | OVS_NOT_REACHED(); |
6bc60024 BP |
1940 | } |
1941 | ||
72d32ac0 BP |
1942 | ofpbuf_uninit(&aux->request); |
1943 | ofpbuf_uninit(&aux->reply); | |
6bc60024 | 1944 | } |
0f3358ea BP |
1945 | |
1946 | return n_ops; | |
eabe7c68 BP |
1947 | } |
1948 | ||
6c343984 PB |
1949 | static int |
1950 | parse_flow_get(struct dpif_netlink *dpif, struct dpif_flow_get *get) | |
1951 | { | |
1952 | struct dpif_flow *dpif_flow = get->flow; | |
1953 | struct match match; | |
1954 | struct nlattr *actions; | |
1955 | struct dpif_flow_stats stats; | |
d63ca532 | 1956 | struct dpif_flow_attrs attrs; |
6c343984 PB |
1957 | struct ofpbuf buf; |
1958 | uint64_t act_buf[1024 / 8]; | |
1959 | struct odputil_keybuf maskbuf; | |
1960 | struct odputil_keybuf keybuf; | |
1961 | struct odputil_keybuf actbuf; | |
1962 | struct ofpbuf key, mask, act; | |
1963 | int err; | |
1964 | ||
1965 | ofpbuf_use_stack(&buf, &act_buf, sizeof act_buf); | |
dfaf79dd | 1966 | err = netdev_ports_flow_get(dpif->dpif.dpif_class, &match, |
d63ca532 | 1967 | &actions, get->ufid, &stats, &attrs, &buf); |
6c343984 PB |
1968 | if (err) { |
1969 | return err; | |
1970 | } | |
1971 | ||
1972 | VLOG_DBG("found flow from netdev, translating to dpif flow"); | |
1973 | ||
1974 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
1975 | ofpbuf_use_stack(&act, &actbuf, sizeof actbuf); | |
1976 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); | |
1977 | dpif_netlink_netdev_match_to_dpif_flow(&match, &key, &mask, actions, | |
d63ca532 | 1978 | &stats, &attrs, |
6c343984 PB |
1979 | (ovs_u128 *) get->ufid, |
1980 | dpif_flow, | |
1981 | false); | |
1982 | ofpbuf_put(get->buffer, nl_attr_get(actions), nl_attr_get_size(actions)); | |
1983 | dpif_flow->actions = ofpbuf_at(get->buffer, 0, 0); | |
1984 | dpif_flow->actions_len = nl_attr_get_size(actions); | |
1985 | ||
1986 | return 0; | |
1987 | } | |
1988 | ||
8b668ee3 PB |
1989 | static int |
1990 | parse_flow_put(struct dpif_netlink *dpif, struct dpif_flow_put *put) | |
1991 | { | |
1992 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
dfaf79dd | 1993 | const struct dpif_class *dpif_class = dpif->dpif.dpif_class; |
8b668ee3 PB |
1994 | struct match match; |
1995 | odp_port_t in_port; | |
1996 | const struct nlattr *nla; | |
1997 | size_t left; | |
8b668ee3 PB |
1998 | struct netdev *dev; |
1999 | struct offload_info info; | |
2000 | ovs_be16 dst_port = 0; | |
d9677a1f | 2001 | uint8_t csum_on = false; |
8b668ee3 PB |
2002 | int err; |
2003 | ||
2004 | if (put->flags & DPIF_FP_PROBE) { | |
2005 | return EOPNOTSUPP; | |
2006 | } | |
2007 | ||
2008 | err = parse_key_and_mask_to_match(put->key, put->key_len, put->mask, | |
2009 | put->mask_len, &match); | |
2010 | if (err) { | |
2011 | return err; | |
2012 | } | |
2013 | ||
8b668ee3 | 2014 | in_port = match.flow.in_port.odp_port; |
dfaf79dd | 2015 | dev = netdev_ports_get(in_port, dpif_class); |
8b668ee3 PB |
2016 | if (!dev) { |
2017 | return EOPNOTSUPP; | |
2018 | } | |
2019 | ||
00a0a011 | 2020 | /* Get tunnel dst port */ |
8b668ee3 PB |
2021 | NL_ATTR_FOR_EACH(nla, left, put->actions, put->actions_len) { |
2022 | if (nl_attr_type(nla) == OVS_ACTION_ATTR_OUTPUT) { | |
2023 | const struct netdev_tunnel_config *tnl_cfg; | |
2024 | struct netdev *outdev; | |
2025 | odp_port_t out_port; | |
2026 | ||
8b668ee3 | 2027 | out_port = nl_attr_get_odp_port(nla); |
dfaf79dd | 2028 | outdev = netdev_ports_get(out_port, dpif_class); |
8b668ee3 PB |
2029 | if (!outdev) { |
2030 | err = EOPNOTSUPP; | |
2031 | goto out; | |
2032 | } | |
2033 | tnl_cfg = netdev_get_tunnel_config(outdev); | |
2034 | if (tnl_cfg && tnl_cfg->dst_port != 0) { | |
2035 | dst_port = tnl_cfg->dst_port; | |
2036 | } | |
d9677a1f EB |
2037 | if (tnl_cfg) { |
2038 | csum_on = tnl_cfg->csum; | |
2039 | } | |
8b668ee3 PB |
2040 | netdev_close(outdev); |
2041 | } | |
2042 | } | |
2043 | ||
dfaf79dd | 2044 | info.dpif_class = dpif_class; |
8b668ee3 | 2045 | info.tp_dst_port = dst_port; |
d9677a1f | 2046 | info.tunnel_csum_on = csum_on; |
8b668ee3 PB |
2047 | err = netdev_flow_put(dev, &match, |
2048 | CONST_CAST(struct nlattr *, put->actions), | |
2049 | put->actions_len, | |
2050 | CONST_CAST(ovs_u128 *, put->ufid), | |
2051 | &info, put->stats); | |
2052 | ||
2053 | if (!err) { | |
2054 | if (put->flags & DPIF_FP_MODIFY) { | |
2055 | struct dpif_op *opp; | |
2056 | struct dpif_op op; | |
2057 | ||
2058 | op.type = DPIF_OP_FLOW_DEL; | |
fa37affa BP |
2059 | op.flow_del.key = put->key; |
2060 | op.flow_del.key_len = put->key_len; | |
2061 | op.flow_del.ufid = put->ufid; | |
2062 | op.flow_del.pmd_id = put->pmd_id; | |
2063 | op.flow_del.stats = NULL; | |
2064 | op.flow_del.terse = false; | |
8b668ee3 PB |
2065 | |
2066 | opp = &op; | |
2067 | dpif_netlink_operate__(dpif, &opp, 1); | |
2068 | } | |
2069 | ||
2070 | VLOG_DBG("added flow"); | |
2071 | } else if (err != EEXIST) { | |
738c785f | 2072 | struct netdev *oor_netdev = NULL; |
1028cb71 | 2073 | enum vlog_level level; |
738c785f SB |
2074 | if (err == ENOSPC && netdev_is_offload_rebalance_policy_enabled()) { |
2075 | /* | |
2076 | * We need to set OOR on the input netdev (i.e, 'dev') for the | |
2077 | * flow. But if the flow has a tunnel attribute (i.e, decap action, | |
2078 | * with a virtual device like a VxLAN interface as its in-port), | |
2079 | * then lookup and set OOR on the underlying tunnel (real) netdev. | |
2080 | */ | |
2081 | oor_netdev = flow_get_tunnel_netdev(&match.flow.tunnel); | |
2082 | if (!oor_netdev) { | |
2083 | /* Not a 'tunnel' flow */ | |
2084 | oor_netdev = dev; | |
2085 | } | |
2086 | netdev_set_hw_info(oor_netdev, HW_INFO_TYPE_OOR, true); | |
2087 | } | |
1028cb71 | 2088 | level = (err == ENOSPC || err == EOPNOTSUPP) ? VLL_DBG : VLL_ERR; |
2089 | VLOG_RL(&rl, level, "failed to offload flow: %s: %s", | |
2090 | ovs_strerror(err), | |
2091 | (oor_netdev ? oor_netdev->name : dev->name)); | |
8b668ee3 PB |
2092 | } |
2093 | ||
2094 | out: | |
2095 | if (err && err != EEXIST && (put->flags & DPIF_FP_MODIFY)) { | |
2096 | /* Modified rule can't be offloaded, try and delete from HW */ | |
2097 | int del_err = netdev_flow_del(dev, put->ufid, put->stats); | |
2098 | ||
2099 | if (!del_err) { | |
2100 | /* Delete from hw success, so old flow was offloaded. | |
2101 | * Change flags to create the flow in kernel */ | |
2102 | put->flags &= ~DPIF_FP_MODIFY; | |
2103 | put->flags |= DPIF_FP_CREATE; | |
2104 | } else if (del_err != ENOENT) { | |
2105 | VLOG_ERR_RL(&rl, "failed to delete offloaded flow: %s", | |
2106 | ovs_strerror(del_err)); | |
2107 | /* stop proccesing the flow in kernel */ | |
2108 | err = 0; | |
2109 | } | |
2110 | } | |
2111 | ||
2112 | netdev_close(dev); | |
2113 | ||
2114 | return err; | |
2115 | } | |
2116 | ||
8b668ee3 PB |
2117 | static int |
2118 | try_send_to_netdev(struct dpif_netlink *dpif, struct dpif_op *op) | |
eabe7c68 | 2119 | { |
8b668ee3 | 2120 | int err = EOPNOTSUPP; |
9b00386b | 2121 | |
8b668ee3 PB |
2122 | switch (op->type) { |
2123 | case DPIF_OP_FLOW_PUT: { | |
fa37affa | 2124 | struct dpif_flow_put *put = &op->flow_put; |
8b668ee3 PB |
2125 | |
2126 | if (!put->ufid) { | |
2127 | break; | |
2128 | } | |
3cd99886 RD |
2129 | |
2130 | log_flow_put_message(&dpif->dpif, &this_module, put, 0); | |
8b668ee3 PB |
2131 | err = parse_flow_put(dpif, put); |
2132 | break; | |
2133 | } | |
0335a89c | 2134 | case DPIF_OP_FLOW_DEL: { |
fa37affa | 2135 | struct dpif_flow_del *del = &op->flow_del; |
0335a89c PB |
2136 | |
2137 | if (!del->ufid) { | |
2138 | break; | |
2139 | } | |
3cd99886 RD |
2140 | |
2141 | log_flow_del_message(&dpif->dpif, &this_module, del, 0); | |
dfaf79dd | 2142 | err = netdev_ports_flow_del(dpif->dpif.dpif_class, del->ufid, |
0335a89c PB |
2143 | del->stats); |
2144 | break; | |
2145 | } | |
6c343984 | 2146 | case DPIF_OP_FLOW_GET: { |
fa37affa | 2147 | struct dpif_flow_get *get = &op->flow_get; |
6c343984 | 2148 | |
fa37affa | 2149 | if (!op->flow_get.ufid) { |
6c343984 PB |
2150 | break; |
2151 | } | |
3cd99886 RD |
2152 | |
2153 | log_flow_get_message(&dpif->dpif, &this_module, get, 0); | |
6c343984 PB |
2154 | err = parse_flow_get(dpif, get); |
2155 | break; | |
2156 | } | |
8b668ee3 PB |
2157 | case DPIF_OP_EXECUTE: |
2158 | default: | |
2159 | break; | |
2160 | } | |
2161 | ||
2162 | return err; | |
2163 | } | |
2164 | ||
2165 | static void | |
2166 | dpif_netlink_operate_chunks(struct dpif_netlink *dpif, struct dpif_op **ops, | |
2167 | size_t n_ops) | |
2168 | { | |
eabe7c68 | 2169 | while (n_ops > 0) { |
0f3358ea | 2170 | size_t chunk = dpif_netlink_operate__(dpif, ops, n_ops); |
8b668ee3 | 2171 | |
eabe7c68 BP |
2172 | ops += chunk; |
2173 | n_ops -= chunk; | |
2174 | } | |
6bc60024 BP |
2175 | } |
2176 | ||
8b668ee3 | 2177 | static void |
57924fc9 SB |
2178 | dpif_netlink_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops, |
2179 | enum dpif_offload_type offload_type) | |
8b668ee3 PB |
2180 | { |
2181 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); | |
2182 | struct dpif_op *new_ops[OPERATE_MAX_OPS]; | |
2183 | int count = 0; | |
2184 | int i = 0; | |
2185 | int err = 0; | |
2186 | ||
57924fc9 SB |
2187 | if (offload_type == DPIF_OFFLOAD_ALWAYS && !netdev_is_flow_api_enabled()) { |
2188 | VLOG_DBG("Invalid offload_type: %d", offload_type); | |
2189 | return; | |
2190 | } | |
2191 | ||
2192 | if (offload_type != DPIF_OFFLOAD_NEVER && netdev_is_flow_api_enabled()) { | |
8b668ee3 PB |
2193 | while (n_ops > 0) { |
2194 | count = 0; | |
2195 | ||
2196 | while (n_ops > 0 && count < OPERATE_MAX_OPS) { | |
2197 | struct dpif_op *op = ops[i++]; | |
2198 | ||
2199 | err = try_send_to_netdev(dpif, op); | |
2200 | if (err && err != EEXIST) { | |
57924fc9 SB |
2201 | if (offload_type == DPIF_OFFLOAD_ALWAYS) { |
2202 | /* We got an error while offloading an op. Since | |
2203 | * OFFLOAD_ALWAYS is specified, we stop further | |
2204 | * processing and return to the caller without | |
2205 | * invoking kernel datapath as fallback. But the | |
2206 | * interface requires us to process all n_ops; so | |
2207 | * return the same error in the remaining ops too. | |
2208 | */ | |
2209 | op->error = err; | |
2210 | n_ops--; | |
2211 | while (n_ops > 0) { | |
2212 | op = ops[i++]; | |
2213 | op->error = err; | |
2214 | n_ops--; | |
2215 | } | |
2216 | return; | |
2217 | } | |
8b668ee3 PB |
2218 | new_ops[count++] = op; |
2219 | } else { | |
2220 | op->error = err; | |
2221 | } | |
2222 | ||
2223 | n_ops--; | |
2224 | } | |
2225 | ||
2226 | dpif_netlink_operate_chunks(dpif, new_ops, count); | |
2227 | } | |
57924fc9 | 2228 | } else if (offload_type != DPIF_OFFLOAD_ALWAYS) { |
8b668ee3 PB |
2229 | dpif_netlink_operate_chunks(dpif, ops, n_ops); |
2230 | } | |
2231 | } | |
2232 | ||
09cac43f NR |
2233 | #if _WIN32 |
2234 | static void | |
2235 | dpif_netlink_handler_uninit(struct dpif_handler *handler) | |
2236 | { | |
2237 | vport_delete_sock_pool(handler); | |
2238 | } | |
2239 | ||
2240 | static int | |
2241 | dpif_netlink_handler_init(struct dpif_handler *handler) | |
2242 | { | |
2243 | return vport_create_sock_pool(handler); | |
2244 | } | |
2245 | #else | |
2246 | ||
2247 | static int | |
2248 | dpif_netlink_handler_init(struct dpif_handler *handler) | |
2249 | { | |
2250 | handler->epoll_fd = epoll_create(10); | |
2251 | return handler->epoll_fd < 0 ? errno : 0; | |
2252 | } | |
2253 | ||
2254 | static void | |
2255 | dpif_netlink_handler_uninit(struct dpif_handler *handler) | |
2256 | { | |
2257 | close(handler->epoll_fd); | |
2258 | } | |
2259 | #endif | |
2260 | ||
1579cf67 AW |
2261 | /* Synchronizes 'channels' in 'dpif->handlers' with the set of vports |
2262 | * currently in 'dpif' in the kernel, by adding a new set of channels for | |
2263 | * any kernel vport that lacks one and deleting any channels that have no | |
2264 | * backing kernel vports. */ | |
96fba48f | 2265 | static int |
93451a0a | 2266 | dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) |
b90de034 | 2267 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
96fba48f | 2268 | { |
8381a3d3 | 2269 | unsigned long int *keep_channels; |
93451a0a | 2270 | struct dpif_netlink_vport vport; |
8381a3d3 BP |
2271 | size_t keep_channels_nbits; |
2272 | struct nl_dump dump; | |
d57695d7 JS |
2273 | uint64_t reply_stub[NL_DUMP_BUFSIZE / 8]; |
2274 | struct ofpbuf buf; | |
8381a3d3 BP |
2275 | int retval = 0; |
2276 | size_t i; | |
982b8810 | 2277 | |
09cac43f NR |
2278 | ovs_assert(!WINDOWS || n_handlers <= 1); |
2279 | ovs_assert(!WINDOWS || dpif->n_handlers <= 1); | |
2280 | ||
1579cf67 AW |
2281 | if (dpif->n_handlers != n_handlers) { |
2282 | destroy_all_channels(dpif); | |
2283 | dpif->handlers = xzalloc(n_handlers * sizeof *dpif->handlers); | |
2284 | for (i = 0; i < n_handlers; i++) { | |
09cac43f | 2285 | int error; |
1579cf67 AW |
2286 | struct dpif_handler *handler = &dpif->handlers[i]; |
2287 | ||
09cac43f NR |
2288 | error = dpif_netlink_handler_init(handler); |
2289 | if (error) { | |
1579cf67 AW |
2290 | size_t j; |
2291 | ||
2292 | for (j = 0; j < i; j++) { | |
aa5c0216 | 2293 | struct dpif_handler *tmp = &dpif->handlers[j]; |
09cac43f | 2294 | dpif_netlink_handler_uninit(tmp); |
1579cf67 AW |
2295 | } |
2296 | free(dpif->handlers); | |
2297 | dpif->handlers = NULL; | |
2298 | ||
09cac43f | 2299 | return error; |
1579cf67 | 2300 | } |
8381a3d3 | 2301 | } |
1579cf67 AW |
2302 | dpif->n_handlers = n_handlers; |
2303 | } | |
2304 | ||
2305 | for (i = 0; i < n_handlers; i++) { | |
2306 | struct dpif_handler *handler = &dpif->handlers[i]; | |
2307 | ||
2308 | handler->event_offset = handler->n_events = 0; | |
17411ecf | 2309 | } |
b063d9f0 | 2310 | |
8381a3d3 BP |
2311 | keep_channels_nbits = dpif->uc_array_size; |
2312 | keep_channels = bitmap_allocate(keep_channels_nbits); | |
982b8810 | 2313 | |
d57695d7 | 2314 | ofpbuf_use_stub(&buf, reply_stub, sizeof reply_stub); |
93451a0a AS |
2315 | dpif_netlink_port_dump_start__(dpif, &dump); |
2316 | while (!dpif_netlink_port_dump_next__(dpif, &dump, &vport, &buf)) { | |
8381a3d3 | 2317 | uint32_t port_no = odp_to_u32(vport.port_no); |
69c51582 | 2318 | uint32_t upcall_pid; |
8381a3d3 | 2319 | int error; |
50f80534 | 2320 | |
1579cf67 | 2321 | if (port_no >= dpif->uc_array_size |
69c51582 | 2322 | || !vport_get_pid(dpif, port_no, &upcall_pid)) { |
622ea8fd BP |
2323 | struct nl_sock *sock; |
2324 | error = create_nl_sock(dpif, &sock); | |
1579cf67 | 2325 | |
d240e46a | 2326 | if (error) { |
1579cf67 AW |
2327 | goto error; |
2328 | } | |
2329 | ||
622ea8fd | 2330 | error = vport_add_channel(dpif, vport.port_no, sock); |
b063d9f0 | 2331 | if (error) { |
1579cf67 | 2332 | VLOG_INFO("%s: could not add channels for port %s", |
9b00386b | 2333 | dpif_name(&dpif->dpif), vport.name); |
622ea8fd | 2334 | nl_sock_destroy(sock); |
8381a3d3 BP |
2335 | retval = error; |
2336 | goto error; | |
982b8810 | 2337 | } |
622ea8fd | 2338 | upcall_pid = nl_sock_pid(sock); |
8381a3d3 | 2339 | } |
50f80534 | 2340 | |
8381a3d3 | 2341 | /* Configure the vport to deliver misses to 'sock'. */ |
1579cf67 | 2342 | if (vport.upcall_pids[0] == 0 |
69c51582 MC |
2343 | || vport.n_upcall_pids != 1 |
2344 | || upcall_pid != vport.upcall_pids[0]) { | |
93451a0a | 2345 | struct dpif_netlink_vport vport_request; |
989fd548 | 2346 | |
93451a0a | 2347 | dpif_netlink_vport_init(&vport_request); |
989fd548 JP |
2348 | vport_request.cmd = OVS_VPORT_CMD_SET; |
2349 | vport_request.dp_ifindex = dpif->dp_ifindex; | |
8381a3d3 | 2350 | vport_request.port_no = vport.port_no; |
69c51582 MC |
2351 | vport_request.n_upcall_pids = 1; |
2352 | vport_request.upcall_pids = &upcall_pid; | |
93451a0a | 2353 | error = dpif_netlink_vport_transact(&vport_request, NULL, NULL); |
1579cf67 | 2354 | if (error) { |
989fd548 JP |
2355 | VLOG_WARN_RL(&error_rl, |
2356 | "%s: failed to set upcall pid on port: %s", | |
10a89ef0 | 2357 | dpif_name(&dpif->dpif), ovs_strerror(error)); |
989fd548 | 2358 | |
8381a3d3 BP |
2359 | if (error != ENODEV && error != ENOENT) { |
2360 | retval = error; | |
989fd548 | 2361 | } else { |
8381a3d3 BP |
2362 | /* The vport isn't really there, even though the dump says |
2363 | * it is. Probably we just hit a race after a port | |
2364 | * disappeared. */ | |
989fd548 | 2365 | } |
8381a3d3 | 2366 | goto error; |
50f80534 | 2367 | } |
8381a3d3 | 2368 | } |
14b4d2f9 | 2369 | |
8381a3d3 BP |
2370 | if (port_no < keep_channels_nbits) { |
2371 | bitmap_set1(keep_channels, port_no); | |
2372 | } | |
2373 | continue; | |
2374 | ||
2375 | error: | |
1579cf67 | 2376 | vport_del_channels(dpif, vport.port_no); |
982b8810 | 2377 | } |
8381a3d3 | 2378 | nl_dump_done(&dump); |
d57695d7 | 2379 | ofpbuf_uninit(&buf); |
b063d9f0 | 2380 | |
8381a3d3 BP |
2381 | /* Discard any saved channels that we didn't reuse. */ |
2382 | for (i = 0; i < keep_channels_nbits; i++) { | |
2383 | if (!bitmap_is_set(keep_channels, i)) { | |
1579cf67 | 2384 | vport_del_channels(dpif, u32_to_odp(i)); |
8381a3d3 BP |
2385 | } |
2386 | } | |
2387 | free(keep_channels); | |
2388 | ||
2389 | return retval; | |
2390 | } | |
2391 | ||
2392 | static int | |
93451a0a | 2393 | dpif_netlink_recv_set__(struct dpif_netlink *dpif, bool enable) |
b90de034 | 2394 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
8381a3d3 | 2395 | { |
1579cf67 | 2396 | if ((dpif->handlers != NULL) == enable) { |
8381a3d3 BP |
2397 | return 0; |
2398 | } else if (!enable) { | |
1579cf67 | 2399 | destroy_all_channels(dpif); |
8381a3d3 BP |
2400 | return 0; |
2401 | } else { | |
93451a0a | 2402 | return dpif_netlink_refresh_channels(dpif, 1); |
8381a3d3 | 2403 | } |
96fba48f BP |
2404 | } |
2405 | ||
9fafa796 | 2406 | static int |
93451a0a | 2407 | dpif_netlink_recv_set(struct dpif *dpif_, bool enable) |
9fafa796 | 2408 | { |
93451a0a | 2409 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
2410 | int error; |
2411 | ||
1579cf67 | 2412 | fat_rwlock_wrlock(&dpif->upcall_lock); |
93451a0a | 2413 | error = dpif_netlink_recv_set__(dpif, enable); |
1579cf67 | 2414 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
2415 | |
2416 | return error; | |
2417 | } | |
2418 | ||
1954e6bb | 2419 | static int |
93451a0a | 2420 | dpif_netlink_handlers_set(struct dpif *dpif_, uint32_t n_handlers) |
1954e6bb | 2421 | { |
93451a0a | 2422 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
1579cf67 AW |
2423 | int error = 0; |
2424 | ||
09cac43f NR |
2425 | #ifdef _WIN32 |
2426 | /* Multiple upcall handlers will be supported once kernel datapath supports | |
2427 | * it. */ | |
2428 | if (n_handlers > 1) { | |
2429 | return error; | |
2430 | } | |
2431 | #endif | |
2432 | ||
1579cf67 AW |
2433 | fat_rwlock_wrlock(&dpif->upcall_lock); |
2434 | if (dpif->handlers) { | |
93451a0a | 2435 | error = dpif_netlink_refresh_channels(dpif, n_handlers); |
1579cf67 AW |
2436 | } |
2437 | fat_rwlock_unlock(&dpif->upcall_lock); | |
2438 | ||
2439 | return error; | |
1954e6bb AW |
2440 | } |
2441 | ||
aae51f53 | 2442 | static int |
93451a0a | 2443 | dpif_netlink_queue_to_priority(const struct dpif *dpif OVS_UNUSED, |
aae51f53 BP |
2444 | uint32_t queue_id, uint32_t *priority) |
2445 | { | |
2446 | if (queue_id < 0xf000) { | |
17ee3c1f | 2447 | *priority = TC_H_MAKE(1 << 16, queue_id + 1); |
aae51f53 BP |
2448 | return 0; |
2449 | } else { | |
2450 | return EINVAL; | |
2451 | } | |
2452 | } | |
2453 | ||
96fba48f | 2454 | static int |
7af12bd7 JS |
2455 | parse_odp_packet(const struct dpif_netlink *dpif, struct ofpbuf *buf, |
2456 | struct dpif_upcall *upcall, int *dp_ifindex) | |
856081f6 | 2457 | { |
df2c07f4 | 2458 | static const struct nl_policy ovs_packet_policy[] = { |
856081f6 | 2459 | /* Always present. */ |
df2c07f4 | 2460 | [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC, |
856081f6 | 2461 | .min_len = ETH_HEADER_LEN }, |
df2c07f4 | 2462 | [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED }, |
856081f6 | 2463 | |
df2c07f4 | 2464 | /* OVS_PACKET_CMD_ACTION only. */ |
e995e3df | 2465 | [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true }, |
8b7ea2d4 | 2466 | [OVS_PACKET_ATTR_EGRESS_TUN_KEY] = { .type = NL_A_NESTED, .optional = true }, |
7321bda3 | 2467 | [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, |
0442bfb1 TZ |
2468 | [OVS_PACKET_ATTR_MRU] = { .type = NL_A_U16, .optional = true }, |
2469 | [OVS_PACKET_ATTR_HASH] = { .type = NL_A_U64, .optional = true } | |
856081f6 BP |
2470 | }; |
2471 | ||
0a2869d5 BP |
2472 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
2473 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
2474 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
2475 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
982b8810 | 2476 | |
0a2869d5 | 2477 | struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)]; |
df2c07f4 JP |
2478 | if (!nlmsg || !genl || !ovs_header |
2479 | || nlmsg->nlmsg_type != ovs_packet_family | |
2480 | || !nl_policy_parse(&b, 0, ovs_packet_policy, a, | |
2481 | ARRAY_SIZE(ovs_packet_policy))) { | |
856081f6 BP |
2482 | return EINVAL; |
2483 | } | |
2484 | ||
0a2869d5 BP |
2485 | int type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS |
2486 | : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION | |
2487 | : -1); | |
aaff4b55 BP |
2488 | if (type < 0) { |
2489 | return EINVAL; | |
2490 | } | |
82272ede | 2491 | |
877c9270 | 2492 | /* (Re)set ALL fields of '*upcall' on successful return. */ |
aaff4b55 | 2493 | upcall->type = type; |
ebc56baa BP |
2494 | upcall->key = CONST_CAST(struct nlattr *, |
2495 | nl_attr_get(a[OVS_PACKET_ATTR_KEY])); | |
df2c07f4 | 2496 | upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]); |
7af12bd7 | 2497 | dpif_flow_hash(&dpif->dpif, upcall->key, upcall->key_len, &upcall->ufid); |
e995e3df | 2498 | upcall->userdata = a[OVS_PACKET_ATTR_USERDATA]; |
8b7ea2d4 | 2499 | upcall->out_tun_key = a[OVS_PACKET_ATTR_EGRESS_TUN_KEY]; |
7321bda3 | 2500 | upcall->actions = a[OVS_PACKET_ATTR_ACTIONS]; |
27130224 | 2501 | upcall->mru = a[OVS_PACKET_ATTR_MRU]; |
0442bfb1 | 2502 | upcall->hash = a[OVS_PACKET_ATTR_HASH]; |
da546e07 JR |
2503 | |
2504 | /* Allow overwriting the netlink attribute header without reallocating. */ | |
cf62fa4c | 2505 | dp_packet_use_stub(&upcall->packet, |
da546e07 JR |
2506 | CONST_CAST(struct nlattr *, |
2507 | nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1, | |
2508 | nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) + | |
2509 | sizeof(struct nlattr)); | |
cf62fa4c PS |
2510 | dp_packet_set_data(&upcall->packet, |
2511 | (char *)dp_packet_data(&upcall->packet) + sizeof(struct nlattr)); | |
2512 | dp_packet_set_size(&upcall->packet, nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET])); | |
da546e07 | 2513 | |
2482b0b0 JS |
2514 | if (nl_attr_find__(upcall->key, upcall->key_len, OVS_KEY_ATTR_ETHERNET)) { |
2515 | /* Ethernet frame */ | |
2516 | upcall->packet.packet_type = htonl(PT_ETH); | |
2517 | } else { | |
2518 | /* Non-Ethernet packet. Get the Ethertype from the NL attributes */ | |
2519 | ovs_be16 ethertype = 0; | |
2520 | const struct nlattr *et_nla = nl_attr_find__(upcall->key, | |
2521 | upcall->key_len, | |
2522 | OVS_KEY_ATTR_ETHERTYPE); | |
2523 | if (et_nla) { | |
2524 | ethertype = nl_attr_get_be16(et_nla); | |
2525 | } | |
2526 | upcall->packet.packet_type = PACKET_TYPE_BE(OFPHTN_ETHERTYPE, | |
2527 | ntohs(ethertype)); | |
2528 | dp_packet_set_l3(&upcall->packet, dp_packet_data(&upcall->packet)); | |
2529 | } | |
2530 | ||
df2c07f4 | 2531 | *dp_ifindex = ovs_header->dp_ifindex; |
982b8810 | 2532 | |
856081f6 BP |
2533 | return 0; |
2534 | } | |
2535 | ||
09cac43f NR |
2536 | #ifdef _WIN32 |
2537 | #define PACKET_RECV_BATCH_SIZE 50 | |
2538 | static int | |
2539 | dpif_netlink_recv_windows(struct dpif_netlink *dpif, uint32_t handler_id, | |
2540 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
2541 | OVS_REQ_RDLOCK(dpif->upcall_lock) | |
2542 | { | |
2543 | struct dpif_handler *handler; | |
2544 | int read_tries = 0; | |
2545 | struct dpif_windows_vport_sock *sock_pool; | |
2546 | uint32_t i; | |
2547 | ||
2548 | if (!dpif->handlers) { | |
2549 | return EAGAIN; | |
2550 | } | |
2551 | ||
2552 | /* Only one handler is supported currently. */ | |
2553 | if (handler_id >= 1) { | |
2554 | return EAGAIN; | |
2555 | } | |
2556 | ||
2557 | if (handler_id >= dpif->n_handlers) { | |
2558 | return EAGAIN; | |
2559 | } | |
2560 | ||
2561 | handler = &dpif->handlers[handler_id]; | |
2562 | sock_pool = handler->vport_sock_pool; | |
2563 | ||
2564 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
2565 | for (;;) { | |
2566 | int dp_ifindex; | |
2567 | int error; | |
2568 | ||
2569 | if (++read_tries > PACKET_RECV_BATCH_SIZE) { | |
2570 | return EAGAIN; | |
2571 | } | |
2572 | ||
a86bd14e | 2573 | error = nl_sock_recv(sock_pool[i].nl_sock, buf, NULL, false); |
09cac43f NR |
2574 | if (error == ENOBUFS) { |
2575 | /* ENOBUFS typically means that we've received so many | |
2576 | * packets that the buffer overflowed. Try again | |
2577 | * immediately because there's almost certainly a packet | |
2578 | * waiting for us. */ | |
2579 | /* XXX: report_loss(dpif, ch, idx, handler_id); */ | |
2580 | continue; | |
2581 | } | |
2582 | ||
2583 | /* XXX: ch->last_poll = time_msec(); */ | |
2584 | if (error) { | |
2585 | if (error == EAGAIN) { | |
2586 | break; | |
2587 | } | |
2588 | return error; | |
2589 | } | |
2590 | ||
27edb4aa | 2591 | error = parse_odp_packet(dpif, buf, upcall, &dp_ifindex); |
09cac43f NR |
2592 | if (!error && dp_ifindex == dpif->dp_ifindex) { |
2593 | return 0; | |
2594 | } else if (error) { | |
2595 | return error; | |
2596 | } | |
2597 | } | |
2598 | } | |
2599 | ||
2600 | return EAGAIN; | |
2601 | } | |
2602 | #else | |
856081f6 | 2603 | static int |
93451a0a AS |
2604 | dpif_netlink_recv__(struct dpif_netlink *dpif, uint32_t handler_id, |
2605 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
b90de034 | 2606 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
96fba48f | 2607 | { |
1579cf67 | 2608 | struct dpif_handler *handler; |
17411ecf | 2609 | int read_tries = 0; |
96fba48f | 2610 | |
1579cf67 AW |
2611 | if (!dpif->handlers || handler_id >= dpif->n_handlers) { |
2612 | return EAGAIN; | |
982b8810 BP |
2613 | } |
2614 | ||
1579cf67 AW |
2615 | handler = &dpif->handlers[handler_id]; |
2616 | if (handler->event_offset >= handler->n_events) { | |
8522ba09 | 2617 | int retval; |
989fd548 | 2618 | |
1579cf67 | 2619 | handler->event_offset = handler->n_events = 0; |
f6d1465c | 2620 | |
8522ba09 | 2621 | do { |
1579cf67 | 2622 | retval = epoll_wait(handler->epoll_fd, handler->epoll_events, |
989fd548 | 2623 | dpif->uc_array_size, 0); |
8522ba09 | 2624 | } while (retval < 0 && errno == EINTR); |
09cac43f | 2625 | |
8522ba09 BP |
2626 | if (retval < 0) { |
2627 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
10a89ef0 | 2628 | VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", ovs_strerror(errno)); |
989fd548 | 2629 | } else if (retval > 0) { |
1579cf67 | 2630 | handler->n_events = retval; |
8522ba09 | 2631 | } |
8522ba09 BP |
2632 | } |
2633 | ||
1579cf67 AW |
2634 | while (handler->event_offset < handler->n_events) { |
2635 | int idx = handler->epoll_events[handler->event_offset].data.u32; | |
69c51582 | 2636 | struct dpif_channel *ch = &dpif->channels[idx]; |
8522ba09 | 2637 | |
1579cf67 | 2638 | handler->event_offset++; |
17411ecf | 2639 | |
f6d1465c | 2640 | for (;;) { |
8522ba09 | 2641 | int dp_ifindex; |
f6d1465c | 2642 | int error; |
17411ecf | 2643 | |
f6d1465c BP |
2644 | if (++read_tries > 50) { |
2645 | return EAGAIN; | |
2646 | } | |
17411ecf | 2647 | |
a86bd14e | 2648 | error = nl_sock_recv(ch->sock, buf, NULL, false); |
14b4d2f9 BP |
2649 | if (error == ENOBUFS) { |
2650 | /* ENOBUFS typically means that we've received so many | |
2651 | * packets that the buffer overflowed. Try again | |
2652 | * immediately because there's almost certainly a packet | |
2653 | * waiting for us. */ | |
9b00386b | 2654 | report_loss(dpif, ch, idx, handler_id); |
14b4d2f9 BP |
2655 | continue; |
2656 | } | |
2657 | ||
2658 | ch->last_poll = time_msec(); | |
72d32ac0 | 2659 | if (error) { |
72d32ac0 BP |
2660 | if (error == EAGAIN) { |
2661 | break; | |
2662 | } | |
f6d1465c BP |
2663 | return error; |
2664 | } | |
17411ecf | 2665 | |
7af12bd7 | 2666 | error = parse_odp_packet(dpif, buf, upcall, &dp_ifindex); |
a12b3ead | 2667 | if (!error && dp_ifindex == dpif->dp_ifindex) { |
f6d1465c | 2668 | return 0; |
989fd548 | 2669 | } else if (error) { |
f6d1465c | 2670 | return error; |
17411ecf | 2671 | } |
982b8810 | 2672 | } |
50f80534 | 2673 | } |
982b8810 BP |
2674 | |
2675 | return EAGAIN; | |
96fba48f | 2676 | } |
09cac43f | 2677 | #endif |
96fba48f | 2678 | |
9fafa796 | 2679 | static int |
93451a0a AS |
2680 | dpif_netlink_recv(struct dpif *dpif_, uint32_t handler_id, |
2681 | struct dpif_upcall *upcall, struct ofpbuf *buf) | |
9fafa796 | 2682 | { |
93451a0a | 2683 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
9fafa796 BP |
2684 | int error; |
2685 | ||
1579cf67 | 2686 | fat_rwlock_rdlock(&dpif->upcall_lock); |
09cac43f NR |
2687 | #ifdef _WIN32 |
2688 | error = dpif_netlink_recv_windows(dpif, handler_id, upcall, buf); | |
2689 | #else | |
93451a0a | 2690 | error = dpif_netlink_recv__(dpif, handler_id, upcall, buf); |
09cac43f | 2691 | #endif |
1579cf67 | 2692 | fat_rwlock_unlock(&dpif->upcall_lock); |
9fafa796 BP |
2693 | |
2694 | return error; | |
2695 | } | |
2696 | ||
96fba48f | 2697 | static void |
93451a0a | 2698 | dpif_netlink_recv_wait__(struct dpif_netlink *dpif, uint32_t handler_id) |
b90de034 | 2699 | OVS_REQ_RDLOCK(dpif->upcall_lock) |
96fba48f | 2700 | { |
93451a0a | 2701 | #ifdef _WIN32 |
09cac43f NR |
2702 | uint32_t i; |
2703 | struct dpif_windows_vport_sock *sock_pool = | |
2704 | dpif->handlers[handler_id].vport_sock_pool; | |
2705 | ||
2706 | /* Only one handler is supported currently. */ | |
2707 | if (handler_id >= 1) { | |
2708 | return; | |
2709 | } | |
2710 | ||
2711 | for (i = 0; i < VPORT_SOCK_POOL_SIZE; i++) { | |
2712 | nl_sock_wait(sock_pool[i].nl_sock, POLLIN); | |
2713 | } | |
93451a0a | 2714 | #else |
1579cf67 AW |
2715 | if (dpif->handlers && handler_id < dpif->n_handlers) { |
2716 | struct dpif_handler *handler = &dpif->handlers[handler_id]; | |
2717 | ||
2718 | poll_fd_wait(handler->epoll_fd, POLLIN); | |
17411ecf | 2719 | } |
93451a0a | 2720 | #endif |
96fba48f BP |
2721 | } |
2722 | ||
1ba530f4 | 2723 | static void |
93451a0a | 2724 | dpif_netlink_recv_wait(struct dpif *dpif_, uint32_t handler_id) |
1ba530f4 | 2725 | { |
93451a0a | 2726 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
17411ecf | 2727 | |
b90de034 | 2728 | fat_rwlock_rdlock(&dpif->upcall_lock); |
93451a0a | 2729 | dpif_netlink_recv_wait__(dpif, handler_id); |
b90de034 AW |
2730 | fat_rwlock_unlock(&dpif->upcall_lock); |
2731 | } | |
2732 | ||
2733 | static void | |
93451a0a | 2734 | dpif_netlink_recv_purge__(struct dpif_netlink *dpif) |
b90de034 AW |
2735 | OVS_REQ_WRLOCK(dpif->upcall_lock) |
2736 | { | |
1579cf67 | 2737 | if (dpif->handlers) { |
69c51582 | 2738 | size_t i; |
1579cf67 | 2739 | |
69c51582 MC |
2740 | if (!dpif->channels[0].sock) { |
2741 | return; | |
2742 | } | |
1579cf67 | 2743 | for (i = 0; i < dpif->uc_array_size; i++ ) { |
1ba530f4 | 2744 | |
69c51582 | 2745 | nl_sock_drain(dpif->channels[i].sock); |
989fd548 | 2746 | } |
1ba530f4 | 2747 | } |
b90de034 AW |
2748 | } |
2749 | ||
2750 | static void | |
93451a0a | 2751 | dpif_netlink_recv_purge(struct dpif *dpif_) |
b90de034 | 2752 | { |
93451a0a | 2753 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
b90de034 AW |
2754 | |
2755 | fat_rwlock_wrlock(&dpif->upcall_lock); | |
93451a0a | 2756 | dpif_netlink_recv_purge__(dpif); |
1579cf67 | 2757 | fat_rwlock_unlock(&dpif->upcall_lock); |
1ba530f4 BP |
2758 | } |
2759 | ||
b5cbbcf6 AZ |
2760 | static char * |
2761 | dpif_netlink_get_datapath_version(void) | |
2762 | { | |
2763 | char *version_str = NULL; | |
2764 | ||
2765 | #ifdef __linux__ | |
2766 | ||
2767 | #define MAX_VERSION_STR_SIZE 80 | |
2768 | #define LINUX_DATAPATH_VERSION_FILE "/sys/module/openvswitch/version" | |
2769 | FILE *f; | |
2770 | ||
2771 | f = fopen(LINUX_DATAPATH_VERSION_FILE, "r"); | |
2772 | if (f) { | |
2773 | char *newline; | |
2774 | char version[MAX_VERSION_STR_SIZE]; | |
2775 | ||
2776 | if (fgets(version, MAX_VERSION_STR_SIZE, f)) { | |
2777 | newline = strchr(version, '\n'); | |
2778 | if (newline) { | |
2779 | *newline = '\0'; | |
2780 | } | |
2781 | version_str = xstrdup(version); | |
2782 | } | |
2783 | fclose(f); | |
2784 | } | |
2785 | #endif | |
2786 | ||
2787 | return version_str; | |
2788 | } | |
2789 | ||
c11c9f4a DDP |
2790 | struct dpif_netlink_ct_dump_state { |
2791 | struct ct_dpif_dump_state up; | |
2792 | struct nl_ct_dump_state *nl_ct_dump; | |
2793 | }; | |
2794 | ||
2795 | static int | |
2796 | dpif_netlink_ct_dump_start(struct dpif *dpif OVS_UNUSED, | |
2797 | struct ct_dpif_dump_state **dump_, | |
ded30c74 | 2798 | const uint16_t *zone, int *ptot_bkts) |
c11c9f4a DDP |
2799 | { |
2800 | struct dpif_netlink_ct_dump_state *dump; | |
2801 | int err; | |
2802 | ||
2803 | dump = xzalloc(sizeof *dump); | |
ded30c74 | 2804 | err = nl_ct_dump_start(&dump->nl_ct_dump, zone, ptot_bkts); |
c11c9f4a DDP |
2805 | if (err) { |
2806 | free(dump); | |
2807 | return err; | |
2808 | } | |
2809 | ||
2810 | *dump_ = &dump->up; | |
2811 | ||
2812 | return 0; | |
2813 | } | |
2814 | ||
2815 | static int | |
2816 | dpif_netlink_ct_dump_next(struct dpif *dpif OVS_UNUSED, | |
2817 | struct ct_dpif_dump_state *dump_, | |
2818 | struct ct_dpif_entry *entry) | |
2819 | { | |
2820 | struct dpif_netlink_ct_dump_state *dump; | |
2821 | ||
2822 | INIT_CONTAINER(dump, dump_, up); | |
2823 | ||
2824 | return nl_ct_dump_next(dump->nl_ct_dump, entry); | |
2825 | } | |
2826 | ||
2827 | static int | |
2828 | dpif_netlink_ct_dump_done(struct dpif *dpif OVS_UNUSED, | |
2829 | struct ct_dpif_dump_state *dump_) | |
2830 | { | |
2831 | struct dpif_netlink_ct_dump_state *dump; | |
c11c9f4a DDP |
2832 | |
2833 | INIT_CONTAINER(dump, dump_, up); | |
2834 | ||
1f161318 | 2835 | int err = nl_ct_dump_done(dump->nl_ct_dump); |
c11c9f4a DDP |
2836 | free(dump); |
2837 | return err; | |
2838 | } | |
15eabc97 DDP |
2839 | |
2840 | static int | |
817a7657 YHW |
2841 | dpif_netlink_ct_flush(struct dpif *dpif OVS_UNUSED, const uint16_t *zone, |
2842 | const struct ct_dpif_tuple *tuple) | |
15eabc97 | 2843 | { |
817a7657 YHW |
2844 | if (tuple) { |
2845 | return nl_ct_flush_tuple(tuple, zone ? *zone : 0); | |
2846 | } else if (zone) { | |
15eabc97 DDP |
2847 | return nl_ct_flush_zone(*zone); |
2848 | } else { | |
2849 | return nl_ct_flush(); | |
2850 | } | |
2851 | } | |
c11c9f4a | 2852 | |
906ff9d2 YHW |
2853 | static int |
2854 | dpif_netlink_ct_set_limits(struct dpif *dpif OVS_UNUSED, | |
2855 | const uint32_t *default_limits, | |
2856 | const struct ovs_list *zone_limits) | |
2857 | { | |
2858 | struct ovs_zone_limit req_zone_limit; | |
2859 | ||
2860 | if (ovs_ct_limit_family < 0) { | |
2861 | return EOPNOTSUPP; | |
2862 | } | |
2863 | ||
2864 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
2865 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
2866 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_SET, | |
2867 | OVS_CT_LIMIT_VERSION); | |
2868 | ||
2869 | struct ovs_header *ovs_header; | |
2870 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
2871 | ovs_header->dp_ifindex = 0; | |
2872 | ||
2873 | size_t opt_offset; | |
2874 | opt_offset = nl_msg_start_nested(request, OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
2875 | if (default_limits) { | |
2876 | req_zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE; | |
2877 | req_zone_limit.limit = *default_limits; | |
2878 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2879 | } | |
2880 | ||
2881 | if (!ovs_list_is_empty(zone_limits)) { | |
2882 | struct ct_dpif_zone_limit *zone_limit; | |
2883 | ||
2884 | LIST_FOR_EACH (zone_limit, node, zone_limits) { | |
2885 | req_zone_limit.zone_id = zone_limit->zone; | |
2886 | req_zone_limit.limit = zone_limit->limit; | |
2887 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2888 | } | |
2889 | } | |
2890 | nl_msg_end_nested(request, opt_offset); | |
2891 | ||
2892 | int err = nl_transact(NETLINK_GENERIC, request, NULL); | |
c225ce22 | 2893 | ofpbuf_delete(request); |
906ff9d2 YHW |
2894 | return err; |
2895 | } | |
2896 | ||
2897 | static int | |
2898 | dpif_netlink_zone_limits_from_ofpbuf(const struct ofpbuf *buf, | |
2899 | uint32_t *default_limit, | |
2900 | struct ovs_list *zone_limits) | |
2901 | { | |
2902 | static const struct nl_policy ovs_ct_limit_policy[] = { | |
2903 | [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NL_A_NESTED, | |
2904 | .optional = true }, | |
2905 | }; | |
2906 | ||
2907 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); | |
2908 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
2909 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
2910 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
2911 | ||
2912 | struct nlattr *attr[ARRAY_SIZE(ovs_ct_limit_policy)]; | |
2913 | ||
2914 | if (!nlmsg || !genl || !ovs_header | |
2915 | || nlmsg->nlmsg_type != ovs_ct_limit_family | |
2916 | || !nl_policy_parse(&b, 0, ovs_ct_limit_policy, attr, | |
2917 | ARRAY_SIZE(ovs_ct_limit_policy))) { | |
2918 | return EINVAL; | |
2919 | } | |
2920 | ||
2921 | ||
2922 | if (!attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) { | |
2923 | return EINVAL; | |
2924 | } | |
2925 | ||
2926 | int rem = NLA_ALIGN( | |
2927 | nl_attr_get_size(attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT])); | |
2928 | const struct ovs_zone_limit *zone_limit = | |
2929 | nl_attr_get(attr[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]); | |
2930 | ||
2931 | while (rem >= sizeof *zone_limit) { | |
2932 | if (zone_limit->zone_id == OVS_ZONE_LIMIT_DEFAULT_ZONE) { | |
2933 | *default_limit = zone_limit->limit; | |
2934 | } else if (zone_limit->zone_id < OVS_ZONE_LIMIT_DEFAULT_ZONE || | |
2935 | zone_limit->zone_id > UINT16_MAX) { | |
2936 | } else { | |
2937 | ct_dpif_push_zone_limit(zone_limits, zone_limit->zone_id, | |
2938 | zone_limit->limit, zone_limit->count); | |
2939 | } | |
2940 | rem -= NLA_ALIGN(sizeof *zone_limit); | |
2941 | zone_limit = ALIGNED_CAST(struct ovs_zone_limit *, | |
2942 | (unsigned char *) zone_limit + NLA_ALIGN(sizeof *zone_limit)); | |
2943 | } | |
2944 | return 0; | |
2945 | } | |
2946 | ||
2947 | static int | |
2948 | dpif_netlink_ct_get_limits(struct dpif *dpif OVS_UNUSED, | |
2949 | uint32_t *default_limit, | |
2950 | const struct ovs_list *zone_limits_request, | |
2951 | struct ovs_list *zone_limits_reply) | |
2952 | { | |
2953 | if (ovs_ct_limit_family < 0) { | |
2954 | return EOPNOTSUPP; | |
2955 | } | |
2956 | ||
2957 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
2958 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
2959 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_GET, | |
2960 | OVS_CT_LIMIT_VERSION); | |
2961 | ||
2962 | struct ovs_header *ovs_header; | |
2963 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
2964 | ovs_header->dp_ifindex = 0; | |
2965 | ||
2966 | if (!ovs_list_is_empty(zone_limits_request)) { | |
2967 | size_t opt_offset = nl_msg_start_nested(request, | |
2968 | OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
2969 | ||
2970 | struct ovs_zone_limit req_zone_limit; | |
2971 | req_zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE; | |
2972 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2973 | ||
2974 | struct ct_dpif_zone_limit *zone_limit; | |
2975 | LIST_FOR_EACH (zone_limit, node, zone_limits_request) { | |
2976 | req_zone_limit.zone_id = zone_limit->zone; | |
2977 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
2978 | } | |
2979 | ||
2980 | nl_msg_end_nested(request, opt_offset); | |
2981 | } | |
2982 | ||
2983 | struct ofpbuf *reply; | |
2984 | int err = nl_transact(NETLINK_GENERIC, request, &reply); | |
2985 | if (err) { | |
2986 | goto out; | |
2987 | } | |
2988 | ||
2989 | err = dpif_netlink_zone_limits_from_ofpbuf(reply, default_limit, | |
2990 | zone_limits_reply); | |
2991 | ||
2992 | out: | |
c225ce22 YS |
2993 | ofpbuf_delete(request); |
2994 | ofpbuf_delete(reply); | |
906ff9d2 YHW |
2995 | return err; |
2996 | } | |
2997 | ||
2998 | static int | |
2999 | dpif_netlink_ct_del_limits(struct dpif *dpif OVS_UNUSED, | |
3000 | const struct ovs_list *zone_limits) | |
3001 | { | |
3002 | if (ovs_ct_limit_family < 0) { | |
3003 | return EOPNOTSUPP; | |
3004 | } | |
3005 | ||
3006 | struct ofpbuf *request = ofpbuf_new(NL_DUMP_BUFSIZE); | |
3007 | nl_msg_put_genlmsghdr(request, 0, ovs_ct_limit_family, | |
3008 | NLM_F_REQUEST | NLM_F_ECHO, OVS_CT_LIMIT_CMD_DEL, | |
3009 | OVS_CT_LIMIT_VERSION); | |
3010 | ||
3011 | struct ovs_header *ovs_header; | |
3012 | ovs_header = ofpbuf_put_uninit(request, sizeof *ovs_header); | |
3013 | ovs_header->dp_ifindex = 0; | |
3014 | ||
3015 | if (!ovs_list_is_empty(zone_limits)) { | |
3016 | size_t opt_offset = | |
3017 | nl_msg_start_nested(request, OVS_CT_LIMIT_ATTR_ZONE_LIMIT); | |
3018 | ||
3019 | struct ct_dpif_zone_limit *zone_limit; | |
3020 | LIST_FOR_EACH (zone_limit, node, zone_limits) { | |
3021 | struct ovs_zone_limit req_zone_limit; | |
3022 | req_zone_limit.zone_id = zone_limit->zone; | |
3023 | nl_msg_put(request, &req_zone_limit, sizeof req_zone_limit); | |
3024 | } | |
3025 | nl_msg_end_nested(request, opt_offset); | |
3026 | } | |
3027 | ||
3028 | int err = nl_transact(NETLINK_GENERIC, request, NULL); | |
3029 | ||
c225ce22 | 3030 | ofpbuf_delete(request); |
906ff9d2 YHW |
3031 | return err; |
3032 | } | |
1f161318 YHW |
3033 | |
3034 | #define NL_TP_NAME_PREFIX "ovs_tp_" | |
3035 | ||
3036 | struct dpif_netlink_timeout_policy_protocol { | |
3037 | uint16_t l3num; | |
3038 | uint8_t l4num; | |
3039 | }; | |
3040 | ||
3041 | enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol { | |
3042 | DPIF_NL_TP_AF_INET_TCP, | |
3043 | DPIF_NL_TP_AF_INET_UDP, | |
3044 | DPIF_NL_TP_AF_INET_ICMP, | |
3045 | DPIF_NL_TP_AF_INET6_TCP, | |
3046 | DPIF_NL_TP_AF_INET6_UDP, | |
3047 | DPIF_NL_TP_AF_INET6_ICMPV6, | |
3048 | DPIF_NL_TP_MAX | |
3049 | }; | |
3050 | ||
3051 | #define DPIF_NL_ALL_TP ((1UL << DPIF_NL_TP_MAX) - 1) | |
3052 | ||
3053 | ||
3054 | static struct dpif_netlink_timeout_policy_protocol tp_protos[] = { | |
3055 | [DPIF_NL_TP_AF_INET_TCP] = { .l3num = AF_INET, .l4num = IPPROTO_TCP }, | |
3056 | [DPIF_NL_TP_AF_INET_UDP] = { .l3num = AF_INET, .l4num = IPPROTO_UDP }, | |
3057 | [DPIF_NL_TP_AF_INET_ICMP] = { .l3num = AF_INET, .l4num = IPPROTO_ICMP }, | |
3058 | [DPIF_NL_TP_AF_INET6_TCP] = { .l3num = AF_INET6, .l4num = IPPROTO_TCP }, | |
3059 | [DPIF_NL_TP_AF_INET6_UDP] = { .l3num = AF_INET6, .l4num = IPPROTO_UDP }, | |
3060 | [DPIF_NL_TP_AF_INET6_ICMPV6] = { .l3num = AF_INET6, | |
3061 | .l4num = IPPROTO_ICMPV6 }, | |
3062 | }; | |
3063 | ||
3064 | static void | |
3065 | dpif_netlink_format_tp_name(uint32_t id, uint16_t l3num, uint8_t l4num, | |
187bb41f | 3066 | char **tp_name) |
1f161318 | 3067 | { |
187bb41f YHW |
3068 | struct ds ds = DS_EMPTY_INITIALIZER; |
3069 | ds_put_format(&ds, "%s%"PRIu32"_", NL_TP_NAME_PREFIX, id); | |
3070 | ct_dpif_format_ipproto(&ds, l4num); | |
1f161318 YHW |
3071 | |
3072 | if (l3num == AF_INET) { | |
187bb41f | 3073 | ds_put_cstr(&ds, "4"); |
1f161318 | 3074 | } else if (l3num == AF_INET6 && l4num != IPPROTO_ICMPV6) { |
187bb41f | 3075 | ds_put_cstr(&ds, "6"); |
1f161318 YHW |
3076 | } |
3077 | ||
187bb41f YHW |
3078 | ovs_assert(ds.length < CTNL_TIMEOUT_NAME_MAX); |
3079 | ||
3080 | *tp_name = ds_steal_cstr(&ds); | |
3081 | } | |
3082 | ||
3083 | static int | |
3084 | dpif_netlink_ct_get_timeout_policy_name(struct dpif *dpif OVS_UNUSED, | |
3085 | uint32_t tp_id, uint16_t dl_type, | |
3086 | uint8_t nw_proto, char **tp_name, | |
3087 | bool *is_generic) | |
3088 | { | |
3089 | dpif_netlink_format_tp_name(tp_id, | |
3090 | dl_type == ETH_TYPE_IP ? AF_INET : AF_INET6, | |
3091 | nw_proto, tp_name); | |
3092 | *is_generic = false; | |
3093 | return 0; | |
1f161318 YHW |
3094 | } |
3095 | ||
3096 | #define CT_DPIF_NL_TP_TCP_MAPPINGS \ | |
3097 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT, SYN_SENT) \ | |
3098 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_RECV, SYN_RECV) \ | |
3099 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, ESTABLISHED, ESTABLISHED) \ | |
3100 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, FIN_WAIT, FIN_WAIT) \ | |
3101 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE_WAIT, CLOSE_WAIT) \ | |
3102 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, LAST_ACK, LAST_ACK) \ | |
3103 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, TIME_WAIT, TIME_WAIT) \ | |
3104 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE, CLOSE) \ | |
3105 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT2, SYN_SENT2) \ | |
3106 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, RETRANSMIT, RETRANS) \ | |
3107 | CT_DPIF_NL_TP_MAPPING(TCP, TCP, UNACK, UNACK) | |
3108 | ||
3109 | #define CT_DPIF_NL_TP_UDP_MAPPINGS \ | |
3110 | CT_DPIF_NL_TP_MAPPING(UDP, UDP, SINGLE, UNREPLIED) \ | |
3111 | CT_DPIF_NL_TP_MAPPING(UDP, UDP, MULTIPLE, REPLIED) | |
3112 | ||
3113 | #define CT_DPIF_NL_TP_ICMP_MAPPINGS \ | |
3114 | CT_DPIF_NL_TP_MAPPING(ICMP, ICMP, FIRST, TIMEOUT) | |
3115 | ||
3116 | #define CT_DPIF_NL_TP_ICMPV6_MAPPINGS \ | |
3117 | CT_DPIF_NL_TP_MAPPING(ICMP, ICMPV6, FIRST, TIMEOUT) | |
3118 | ||
3119 | ||
3120 | #define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2) \ | |
3121 | if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) { \ | |
3122 | nl_tp->present |= 1 << CTA_TIMEOUT_##PROTO2##_##ATTR2; \ | |
3123 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2] = \ | |
3124 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]; \ | |
3125 | } | |
3126 | ||
3127 | static void | |
3128 | dpif_netlink_get_nl_tp_tcp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3129 | struct nl_ct_timeout_policy *nl_tp) | |
3130 | { | |
3131 | CT_DPIF_NL_TP_TCP_MAPPINGS | |
3132 | } | |
3133 | ||
3134 | static void | |
3135 | dpif_netlink_get_nl_tp_udp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3136 | struct nl_ct_timeout_policy *nl_tp) | |
3137 | { | |
3138 | CT_DPIF_NL_TP_UDP_MAPPINGS | |
3139 | } | |
3140 | ||
3141 | static void | |
3142 | dpif_netlink_get_nl_tp_icmp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3143 | struct nl_ct_timeout_policy *nl_tp) | |
3144 | { | |
3145 | CT_DPIF_NL_TP_ICMP_MAPPINGS | |
3146 | } | |
3147 | ||
3148 | static void | |
3149 | dpif_netlink_get_nl_tp_icmpv6_attrs(const struct ct_dpif_timeout_policy *tp, | |
3150 | struct nl_ct_timeout_policy *nl_tp) | |
3151 | { | |
3152 | CT_DPIF_NL_TP_ICMPV6_MAPPINGS | |
3153 | } | |
3154 | ||
3155 | #undef CT_DPIF_NL_TP_MAPPING | |
3156 | ||
3157 | static void | |
3158 | dpif_netlink_get_nl_tp_attrs(const struct ct_dpif_timeout_policy *tp, | |
3159 | uint8_t l4num, struct nl_ct_timeout_policy *nl_tp) | |
3160 | { | |
3161 | nl_tp->present = 0; | |
3162 | ||
3163 | if (l4num == IPPROTO_TCP) { | |
3164 | dpif_netlink_get_nl_tp_tcp_attrs(tp, nl_tp); | |
3165 | } else if (l4num == IPPROTO_UDP) { | |
3166 | dpif_netlink_get_nl_tp_udp_attrs(tp, nl_tp); | |
3167 | } else if (l4num == IPPROTO_ICMP) { | |
3168 | dpif_netlink_get_nl_tp_icmp_attrs(tp, nl_tp); | |
3169 | } else if (l4num == IPPROTO_ICMPV6) { | |
3170 | dpif_netlink_get_nl_tp_icmpv6_attrs(tp, nl_tp); | |
3171 | } | |
3172 | } | |
3173 | ||
3174 | #define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2) \ | |
3175 | if (nl_tp->present & (1 << CTA_TIMEOUT_##PROTO2##_##ATTR2)) { \ | |
3176 | if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) { \ | |
3177 | if (tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] != \ | |
3178 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]) { \ | |
3179 | VLOG_WARN_RL(&error_rl, "Inconsistent timeout policy %s " \ | |
3180 | "attribute %s=%"PRIu32" while %s=%"PRIu32, \ | |
3181 | nl_tp->name, "CTA_TIMEOUT_"#PROTO2"_"#ATTR2, \ | |
3182 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2], \ | |
3183 | "CT_DPIF_TP_ATTR_"#PROTO1"_"#ATTR1, \ | |
3184 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]); \ | |
3185 | } \ | |
3186 | } else { \ | |
3187 | tp->present |= 1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1; \ | |
3188 | tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] = \ | |
3189 | nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]; \ | |
3190 | } \ | |
3191 | } | |
3192 | ||
3193 | static void | |
3194 | dpif_netlink_set_ct_dpif_tp_tcp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3195 | struct ct_dpif_timeout_policy *tp) | |
3196 | { | |
3197 | CT_DPIF_NL_TP_TCP_MAPPINGS | |
3198 | } | |
3199 | ||
3200 | static void | |
3201 | dpif_netlink_set_ct_dpif_tp_udp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3202 | struct ct_dpif_timeout_policy *tp) | |
3203 | { | |
3204 | CT_DPIF_NL_TP_UDP_MAPPINGS | |
3205 | } | |
3206 | ||
3207 | static void | |
3208 | dpif_netlink_set_ct_dpif_tp_icmp_attrs( | |
3209 | const struct nl_ct_timeout_policy *nl_tp, | |
3210 | struct ct_dpif_timeout_policy *tp) | |
3211 | { | |
3212 | CT_DPIF_NL_TP_ICMP_MAPPINGS | |
3213 | } | |
3214 | ||
3215 | static void | |
3216 | dpif_netlink_set_ct_dpif_tp_icmpv6_attrs( | |
3217 | const struct nl_ct_timeout_policy *nl_tp, | |
3218 | struct ct_dpif_timeout_policy *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_set_ct_dpif_tp_attrs(const struct nl_ct_timeout_policy *nl_tp, | |
3227 | struct ct_dpif_timeout_policy *tp) | |
3228 | { | |
3229 | if (nl_tp->l4num == IPPROTO_TCP) { | |
3230 | dpif_netlink_set_ct_dpif_tp_tcp_attrs(nl_tp, tp); | |
3231 | } else if (nl_tp->l4num == IPPROTO_UDP) { | |
3232 | dpif_netlink_set_ct_dpif_tp_udp_attrs(nl_tp, tp); | |
3233 | } else if (nl_tp->l4num == IPPROTO_ICMP) { | |
3234 | dpif_netlink_set_ct_dpif_tp_icmp_attrs(nl_tp, tp); | |
3235 | } else if (nl_tp->l4num == IPPROTO_ICMPV6) { | |
3236 | dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(nl_tp, tp); | |
3237 | } | |
3238 | } | |
3239 | ||
3240 | #ifdef _WIN32 | |
3241 | static int | |
3242 | dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3243 | const struct ct_dpif_timeout_policy *tp) | |
3244 | { | |
3245 | return EOPNOTSUPP; | |
3246 | } | |
3247 | ||
3248 | static int | |
3249 | dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3250 | uint32_t tp_id, | |
3251 | struct ct_dpif_timeout_policy *tp) | |
3252 | { | |
3253 | return EOPNOTSUPP; | |
3254 | } | |
3255 | ||
3256 | static int | |
3257 | dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3258 | uint32_t tp_id) | |
3259 | { | |
3260 | return EOPNOTSUPP; | |
3261 | } | |
3262 | ||
3263 | static int | |
3264 | dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED, | |
3265 | void **statep) | |
3266 | { | |
3267 | return EOPNOTSUPP; | |
3268 | } | |
3269 | ||
3270 | static int | |
3271 | dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED, | |
3272 | void *state, | |
3273 | struct ct_dpif_timeout_policy **tp) | |
3274 | { | |
3275 | return EOPNOTSUPP; | |
3276 | } | |
3277 | ||
3278 | static int | |
3279 | dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED, | |
3280 | void *state) | |
3281 | { | |
3282 | return EOPNOTSUPP; | |
3283 | } | |
3284 | #else | |
3285 | static int | |
3286 | dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3287 | const struct ct_dpif_timeout_policy *tp) | |
3288 | { | |
1f161318 YHW |
3289 | int err = 0; |
3290 | ||
3291 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f YHW |
3292 | struct nl_ct_timeout_policy nl_tp; |
3293 | char *nl_tp_name; | |
3294 | ||
1f161318 YHW |
3295 | dpif_netlink_format_tp_name(tp->id, tp_protos[i].l3num, |
3296 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f YHW |
3297 | ovs_strlcpy(nl_tp.name, nl_tp_name, sizeof nl_tp.name); |
3298 | free(nl_tp_name); | |
3299 | ||
1f161318 YHW |
3300 | nl_tp.l3num = tp_protos[i].l3num; |
3301 | nl_tp.l4num = tp_protos[i].l4num; | |
3302 | dpif_netlink_get_nl_tp_attrs(tp, tp_protos[i].l4num, &nl_tp); | |
3303 | err = nl_ct_set_timeout_policy(&nl_tp); | |
3304 | if (err) { | |
3305 | VLOG_WARN_RL(&error_rl, "failed to add timeout policy %s (%s)", | |
3306 | nl_tp.name, ovs_strerror(err)); | |
3307 | goto out; | |
3308 | } | |
3309 | } | |
3310 | ||
3311 | out: | |
1f161318 YHW |
3312 | return err; |
3313 | } | |
3314 | ||
3315 | static int | |
3316 | dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3317 | uint32_t tp_id, | |
3318 | struct ct_dpif_timeout_policy *tp) | |
3319 | { | |
1f161318 YHW |
3320 | int err = 0; |
3321 | ||
3322 | tp->id = tp_id; | |
3323 | tp->present = 0; | |
3324 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f YHW |
3325 | struct nl_ct_timeout_policy nl_tp; |
3326 | char *nl_tp_name; | |
3327 | ||
1f161318 YHW |
3328 | dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num, |
3329 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f | 3330 | err = nl_ct_get_timeout_policy(nl_tp_name, &nl_tp); |
1f161318 YHW |
3331 | |
3332 | if (err) { | |
3333 | VLOG_WARN_RL(&error_rl, "failed to get timeout policy %s (%s)", | |
187bb41f YHW |
3334 | nl_tp_name, ovs_strerror(err)); |
3335 | free(nl_tp_name); | |
1f161318 YHW |
3336 | goto out; |
3337 | } | |
187bb41f | 3338 | free(nl_tp_name); |
1f161318 YHW |
3339 | dpif_netlink_set_ct_dpif_tp_attrs(&nl_tp, tp); |
3340 | } | |
3341 | ||
3342 | out: | |
1f161318 YHW |
3343 | return err; |
3344 | } | |
3345 | ||
3346 | /* Returns 0 if all the sub timeout policies are deleted or not exist in the | |
3347 | * kernel. Returns 1 if any sub timeout policy deletion failed. */ | |
3348 | static int | |
3349 | dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED, | |
3350 | uint32_t tp_id) | |
3351 | { | |
1f161318 YHW |
3352 | int ret = 0; |
3353 | ||
3354 | for (int i = 0; i < ARRAY_SIZE(tp_protos); ++i) { | |
187bb41f | 3355 | char *nl_tp_name; |
1f161318 YHW |
3356 | dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num, |
3357 | tp_protos[i].l4num, &nl_tp_name); | |
187bb41f | 3358 | int err = nl_ct_del_timeout_policy(nl_tp_name); |
1f161318 YHW |
3359 | if (err == ENOENT) { |
3360 | err = 0; | |
3361 | } | |
3362 | if (err) { | |
3363 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(6, 6); | |
3364 | VLOG_INFO_RL(&rl, "failed to delete timeout policy %s (%s)", | |
187bb41f | 3365 | nl_tp_name, ovs_strerror(err)); |
1f161318 YHW |
3366 | ret = 1; |
3367 | } | |
187bb41f | 3368 | free(nl_tp_name); |
1f161318 YHW |
3369 | } |
3370 | ||
1f161318 YHW |
3371 | return ret; |
3372 | } | |
3373 | ||
3374 | struct dpif_netlink_ct_timeout_policy_dump_state { | |
3375 | struct nl_ct_timeout_policy_dump_state *nl_dump_state; | |
3376 | struct hmap tp_dump_map; | |
3377 | }; | |
3378 | ||
3379 | struct dpif_netlink_tp_dump_node { | |
3380 | struct hmap_node hmap_node; /* node in tp_dump_map. */ | |
3381 | struct ct_dpif_timeout_policy *tp; | |
3382 | uint32_t l3_l4_present; | |
3383 | }; | |
3384 | ||
3385 | static struct dpif_netlink_tp_dump_node * | |
3386 | get_dpif_netlink_tp_dump_node_by_tp_id(uint32_t tp_id, | |
3387 | struct hmap *tp_dump_map) | |
3388 | { | |
3389 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3390 | ||
3391 | HMAP_FOR_EACH_WITH_HASH (tp_dump_node, hmap_node, hash_int(tp_id, 0), | |
3392 | tp_dump_map) { | |
3393 | if (tp_dump_node->tp->id == tp_id) { | |
3394 | return tp_dump_node; | |
3395 | } | |
3396 | } | |
3397 | return NULL; | |
3398 | } | |
3399 | ||
3400 | static void | |
3401 | update_dpif_netlink_tp_dump_node( | |
3402 | const struct nl_ct_timeout_policy *nl_tp, | |
3403 | struct dpif_netlink_tp_dump_node *tp_dump_node) | |
3404 | { | |
3405 | dpif_netlink_set_ct_dpif_tp_attrs(nl_tp, tp_dump_node->tp); | |
3406 | for (int i = 0; i < DPIF_NL_TP_MAX; ++i) { | |
3407 | if (nl_tp->l3num == tp_protos[i].l3num && | |
3408 | nl_tp->l4num == tp_protos[i].l4num) { | |
3409 | tp_dump_node->l3_l4_present |= 1 << i; | |
3410 | break; | |
3411 | } | |
3412 | } | |
3413 | } | |
3414 | ||
3415 | static int | |
3416 | dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED, | |
3417 | void **statep) | |
3418 | { | |
3419 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state; | |
3420 | ||
3421 | *statep = dump_state = xzalloc(sizeof *dump_state); | |
3422 | int err = nl_ct_timeout_policy_dump_start(&dump_state->nl_dump_state); | |
3423 | if (err) { | |
3424 | free(dump_state); | |
3425 | return err; | |
3426 | } | |
3427 | hmap_init(&dump_state->tp_dump_map); | |
3428 | return 0; | |
3429 | } | |
3430 | ||
3431 | static void | |
3432 | get_and_cleanup_tp_dump_node(struct hmap *hmap, | |
3433 | struct dpif_netlink_tp_dump_node *tp_dump_node, | |
3434 | struct ct_dpif_timeout_policy *tp) | |
3435 | { | |
3436 | hmap_remove(hmap, &tp_dump_node->hmap_node); | |
3437 | *tp = *tp_dump_node->tp; | |
3438 | free(tp_dump_node->tp); | |
3439 | free(tp_dump_node); | |
3440 | } | |
3441 | ||
3442 | static int | |
3443 | dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED, | |
3444 | void *state, | |
3445 | struct ct_dpif_timeout_policy *tp) | |
3446 | { | |
3447 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state; | |
3448 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3449 | int err; | |
3450 | ||
3451 | /* Dumps all the timeout policies in the kernel. */ | |
3452 | do { | |
3453 | struct nl_ct_timeout_policy nl_tp; | |
3454 | uint32_t tp_id; | |
3455 | ||
3456 | err = nl_ct_timeout_policy_dump_next(dump_state->nl_dump_state, | |
3457 | &nl_tp); | |
3458 | if (err) { | |
3459 | break; | |
3460 | } | |
3461 | ||
3462 | /* We only interest in OVS installed timeout policies. */ | |
3463 | if (!ovs_scan(nl_tp.name, NL_TP_NAME_PREFIX"%"PRIu32, &tp_id)) { | |
3464 | continue; | |
3465 | } | |
3466 | ||
3467 | tp_dump_node = get_dpif_netlink_tp_dump_node_by_tp_id( | |
3468 | tp_id, &dump_state->tp_dump_map); | |
3469 | if (!tp_dump_node) { | |
3470 | tp_dump_node = xzalloc(sizeof *tp_dump_node); | |
3471 | tp_dump_node->tp = xzalloc(sizeof *tp_dump_node->tp); | |
3472 | tp_dump_node->tp->id = tp_id; | |
3473 | hmap_insert(&dump_state->tp_dump_map, &tp_dump_node->hmap_node, | |
3474 | hash_int(tp_id, 0)); | |
3475 | } | |
3476 | ||
3477 | update_dpif_netlink_tp_dump_node(&nl_tp, tp_dump_node); | |
3478 | ||
3479 | /* Returns one ct_dpif_timeout_policy if we gather all the L3/L4 | |
3480 | * sub-pieces. */ | |
3481 | if (tp_dump_node->l3_l4_present == DPIF_NL_ALL_TP) { | |
3482 | get_and_cleanup_tp_dump_node(&dump_state->tp_dump_map, | |
3483 | tp_dump_node, tp); | |
3484 | break; | |
3485 | } | |
3486 | } while (true); | |
3487 | ||
3488 | /* Dump the incomplete timeout policies. */ | |
3489 | if (err == EOF) { | |
3490 | if (!hmap_is_empty(&dump_state->tp_dump_map)) { | |
3491 | struct hmap_node *hmap_node = hmap_first(&dump_state->tp_dump_map); | |
3492 | tp_dump_node = CONTAINER_OF(hmap_node, | |
3493 | struct dpif_netlink_tp_dump_node, | |
3494 | hmap_node); | |
3495 | get_and_cleanup_tp_dump_node(&dump_state->tp_dump_map, | |
3496 | tp_dump_node, tp); | |
3497 | return 0; | |
3498 | } | |
3499 | } | |
3500 | ||
3501 | return err; | |
3502 | } | |
3503 | ||
3504 | static int | |
3505 | dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED, | |
3506 | void *state) | |
3507 | { | |
3508 | struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state; | |
3509 | struct dpif_netlink_tp_dump_node *tp_dump_node; | |
3510 | ||
3511 | int err = nl_ct_timeout_policy_dump_done(dump_state->nl_dump_state); | |
3512 | HMAP_FOR_EACH_POP (tp_dump_node, hmap_node, &dump_state->tp_dump_map) { | |
3513 | free(tp_dump_node->tp); | |
3514 | free(tp_dump_node); | |
3515 | } | |
3516 | hmap_destroy(&dump_state->tp_dump_map); | |
3517 | free(dump_state); | |
3518 | return err; | |
3519 | } | |
3520 | #endif | |
3521 | ||
5dddf960 JR |
3522 | \f |
3523 | /* Meters */ | |
80738e5f AZ |
3524 | |
3525 | /* Set of supported meter flags */ | |
3526 | #define DP_SUPPORTED_METER_FLAGS_MASK \ | |
3527 | (OFPMF13_STATS | OFPMF13_PKTPS | OFPMF13_KBPS | OFPMF13_BURST) | |
3528 | ||
92d0d515 JP |
3529 | /* Meter support was introduced in Linux 4.15. In some versions of |
3530 | * Linux 4.15, 4.16, and 4.17, there was a bug that never set the id | |
3531 | * when the meter was created, so all meters essentially had an id of | |
3532 | * zero. Check for that condition and disable meters on those kernels. */ | |
3533 | static bool probe_broken_meters(struct dpif *); | |
3534 | ||
5dddf960 | 3535 | static void |
80738e5f AZ |
3536 | dpif_netlink_meter_init(struct dpif_netlink *dpif, struct ofpbuf *buf, |
3537 | void *stub, size_t size, uint32_t command) | |
3538 | { | |
3539 | ofpbuf_use_stub(buf, stub, size); | |
3540 | ||
3541 | nl_msg_put_genlmsghdr(buf, 0, ovs_meter_family, NLM_F_REQUEST | NLM_F_ECHO, | |
3542 | command, OVS_METER_VERSION); | |
3543 | ||
3544 | struct ovs_header *ovs_header; | |
3545 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); | |
3546 | ovs_header->dp_ifindex = dpif->dp_ifindex; | |
3547 | } | |
3548 | ||
3549 | /* Execute meter 'request' in the kernel datapath. If the command | |
3550 | * fails, returns a positive errno value. Otherwise, stores the reply | |
3551 | * in '*replyp', parses the policy according to 'reply_policy' into the | |
3552 | * array of Netlink attribute in 'a', and returns 0. On success, the | |
3553 | * caller is responsible for calling ofpbuf_delete() on '*replyp' | |
3554 | * ('replyp' will contain pointers into 'a'). */ | |
3555 | static int | |
3556 | dpif_netlink_meter_transact(struct ofpbuf *request, struct ofpbuf **replyp, | |
3557 | const struct nl_policy *reply_policy, | |
3558 | struct nlattr **a, size_t size_a) | |
3559 | { | |
3560 | int error = nl_transact(NETLINK_GENERIC, request, replyp); | |
3561 | ofpbuf_uninit(request); | |
3562 | ||
3563 | if (error) { | |
3564 | return error; | |
3565 | } | |
3566 | ||
3567 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(*replyp, sizeof *nlmsg); | |
3568 | struct genlmsghdr *genl = ofpbuf_try_pull(*replyp, sizeof *genl); | |
3569 | struct ovs_header *ovs_header = ofpbuf_try_pull(*replyp, | |
3570 | sizeof *ovs_header); | |
3571 | if (!nlmsg || !genl || !ovs_header | |
3572 | || nlmsg->nlmsg_type != ovs_meter_family | |
3573 | || !nl_policy_parse(*replyp, 0, reply_policy, a, size_a)) { | |
3574 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3575 | VLOG_DBG_RL(&rl, | |
3576 | "Kernel module response to meter tranaction is invalid"); | |
3577 | return EINVAL; | |
3578 | } | |
3579 | return 0; | |
3580 | } | |
3581 | ||
3582 | static void | |
3583 | dpif_netlink_meter_get_features(const struct dpif *dpif_, | |
5dddf960 JR |
3584 | struct ofputil_meter_features *features) |
3585 | { | |
92d0d515 JP |
3586 | if (probe_broken_meters(CONST_CAST(struct dpif *, dpif_))) { |
3587 | features = NULL; | |
3588 | return; | |
3589 | } | |
3590 | ||
80738e5f AZ |
3591 | struct ofpbuf buf, *msg; |
3592 | uint64_t stub[1024 / 8]; | |
3593 | ||
3594 | static const struct nl_policy ovs_meter_features_policy[] = { | |
3595 | [OVS_METER_ATTR_MAX_METERS] = { .type = NL_A_U32 }, | |
3596 | [OVS_METER_ATTR_MAX_BANDS] = { .type = NL_A_U32 }, | |
3597 | [OVS_METER_ATTR_BANDS] = { .type = NL_A_NESTED, .optional = true }, | |
3598 | }; | |
3599 | struct nlattr *a[ARRAY_SIZE(ovs_meter_features_policy)]; | |
3600 | ||
3601 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); | |
3602 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, | |
3603 | OVS_METER_CMD_FEATURES); | |
3604 | if (dpif_netlink_meter_transact(&buf, &msg, ovs_meter_features_policy, a, | |
3605 | ARRAY_SIZE(ovs_meter_features_policy))) { | |
3606 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3607 | VLOG_INFO_RL(&rl, | |
3608 | "dpif_netlink_meter_transact OVS_METER_CMD_FEATURES failed"); | |
3609 | return; | |
3610 | } | |
3611 | ||
3612 | features->max_meters = nl_attr_get_u32(a[OVS_METER_ATTR_MAX_METERS]); | |
3613 | features->max_bands = nl_attr_get_u32(a[OVS_METER_ATTR_MAX_BANDS]); | |
3614 | ||
3615 | /* Bands is a nested attribute of zero or more nested | |
3616 | * band attributes. */ | |
3617 | if (a[OVS_METER_ATTR_BANDS]) { | |
3618 | const struct nlattr *nla; | |
3619 | size_t left; | |
3620 | ||
3621 | NL_NESTED_FOR_EACH (nla, left, a[OVS_METER_ATTR_BANDS]) { | |
3622 | const struct nlattr *band_nla; | |
3623 | size_t band_left; | |
3624 | ||
3625 | NL_NESTED_FOR_EACH (band_nla, band_left, nla) { | |
3626 | if (nl_attr_type(band_nla) == OVS_BAND_ATTR_TYPE) { | |
3627 | if (nl_attr_get_size(band_nla) == sizeof(uint32_t)) { | |
3628 | switch (nl_attr_get_u32(band_nla)) { | |
3629 | case OVS_METER_BAND_TYPE_DROP: | |
3630 | features->band_types |= 1 << OFPMBT13_DROP; | |
3631 | break; | |
3632 | } | |
3633 | } | |
3634 | } | |
3635 | } | |
3636 | } | |
3637 | } | |
3638 | features->capabilities = DP_SUPPORTED_METER_FLAGS_MASK; | |
3639 | ||
3640 | ofpbuf_delete(msg); | |
5dddf960 JR |
3641 | } |
3642 | ||
3643 | static int | |
60ebc04d JP |
3644 | dpif_netlink_meter_set__(struct dpif *dpif_, ofproto_meter_id meter_id, |
3645 | struct ofputil_meter_config *config) | |
5dddf960 | 3646 | { |
80738e5f AZ |
3647 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
3648 | struct ofpbuf buf, *msg; | |
3649 | uint64_t stub[1024 / 8]; | |
3650 | ||
3651 | static const struct nl_policy ovs_meter_set_response_policy[] = { | |
3652 | [OVS_METER_ATTR_ID] = { .type = NL_A_U32 }, | |
3653 | }; | |
3654 | struct nlattr *a[ARRAY_SIZE(ovs_meter_set_response_policy)]; | |
3655 | ||
3656 | if (config->flags & ~DP_SUPPORTED_METER_FLAGS_MASK) { | |
3657 | return EBADF; /* Unsupported flags set */ | |
3658 | } | |
3659 | ||
3660 | for (size_t i = 0; i < config->n_bands; i++) { | |
3661 | switch (config->bands[i].type) { | |
3662 | case OFPMBT13_DROP: | |
3663 | break; | |
3664 | default: | |
3665 | return ENODEV; /* Unsupported band type */ | |
3666 | } | |
3667 | } | |
3668 | ||
3669 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, OVS_METER_CMD_SET); | |
3670 | ||
8101f03f JP |
3671 | nl_msg_put_u32(&buf, OVS_METER_ATTR_ID, meter_id.uint32); |
3672 | ||
80738e5f AZ |
3673 | if (config->flags & OFPMF13_KBPS) { |
3674 | nl_msg_put_flag(&buf, OVS_METER_ATTR_KBPS); | |
3675 | } | |
3676 | ||
3677 | size_t bands_offset = nl_msg_start_nested(&buf, OVS_METER_ATTR_BANDS); | |
3678 | /* Bands */ | |
3679 | for (size_t i = 0; i < config->n_bands; ++i) { | |
3680 | struct ofputil_meter_band * band = &config->bands[i]; | |
3681 | uint32_t band_type; | |
3682 | ||
3683 | size_t band_offset = nl_msg_start_nested(&buf, OVS_BAND_ATTR_UNSPEC); | |
3684 | ||
3685 | switch (band->type) { | |
3686 | case OFPMBT13_DROP: | |
3687 | band_type = OVS_METER_BAND_TYPE_DROP; | |
3688 | break; | |
3689 | default: | |
3690 | band_type = OVS_METER_BAND_TYPE_UNSPEC; | |
3691 | } | |
3692 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_TYPE, band_type); | |
3693 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_RATE, band->rate); | |
3694 | nl_msg_put_u32(&buf, OVS_BAND_ATTR_BURST, | |
3695 | config->flags & OFPMF13_BURST ? | |
3696 | band->burst_size : band->rate); | |
3697 | nl_msg_end_nested(&buf, band_offset); | |
3698 | } | |
3699 | nl_msg_end_nested(&buf, bands_offset); | |
3700 | ||
3701 | int error = dpif_netlink_meter_transact(&buf, &msg, | |
3702 | ovs_meter_set_response_policy, a, | |
3703 | ARRAY_SIZE(ovs_meter_set_response_policy)); | |
3704 | if (error) { | |
3705 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3706 | VLOG_INFO_RL(&rl, | |
3707 | "dpif_netlink_meter_transact OVS_METER_CMD_SET failed"); | |
3708 | return error; | |
3709 | } | |
3710 | ||
8101f03f JP |
3711 | if (nl_attr_get_u32(a[OVS_METER_ATTR_ID]) != meter_id.uint32) { |
3712 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3713 | VLOG_INFO_RL(&rl, | |
3714 | "Kernel returned a different meter id than requested"); | |
3715 | } | |
80738e5f AZ |
3716 | ofpbuf_delete(msg); |
3717 | return 0; | |
5dddf960 JR |
3718 | } |
3719 | ||
60ebc04d JP |
3720 | static int |
3721 | dpif_netlink_meter_set(struct dpif *dpif_, ofproto_meter_id meter_id, | |
3722 | struct ofputil_meter_config *config) | |
3723 | { | |
3724 | if (probe_broken_meters(dpif_)) { | |
3725 | return ENOMEM; | |
3726 | } | |
3727 | ||
3728 | return dpif_netlink_meter_set__(dpif_, meter_id, config); | |
3729 | } | |
3730 | ||
80738e5f AZ |
3731 | /* Retrieve statistics and/or delete meter 'meter_id'. Statistics are |
3732 | * stored in 'stats', if it is not null. If 'command' is | |
3733 | * OVS_METER_CMD_DEL, the meter is deleted and statistics are optionally | |
3734 | * retrieved. If 'command' is OVS_METER_CMD_GET, then statistics are | |
3735 | * simply retrieved. */ | |
5dddf960 | 3736 | static int |
80738e5f AZ |
3737 | dpif_netlink_meter_get_stats(const struct dpif *dpif_, |
3738 | ofproto_meter_id meter_id, | |
3739 | struct ofputil_meter_stats *stats, | |
3740 | uint16_t max_bands, | |
3741 | enum ovs_meter_cmd command) | |
5dddf960 | 3742 | { |
80738e5f AZ |
3743 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
3744 | struct ofpbuf buf, *msg; | |
3745 | uint64_t stub[1024 / 8]; | |
3746 | ||
3747 | static const struct nl_policy ovs_meter_stats_policy[] = { | |
3748 | [OVS_METER_ATTR_ID] = { .type = NL_A_U32, .optional = true}, | |
3749 | [OVS_METER_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), | |
3750 | .optional = true}, | |
3751 | [OVS_METER_ATTR_BANDS] = { .type = NL_A_NESTED, .optional = true }, | |
3752 | }; | |
3753 | struct nlattr *a[ARRAY_SIZE(ovs_meter_stats_policy)]; | |
3754 | ||
3755 | dpif_netlink_meter_init(dpif, &buf, stub, sizeof stub, command); | |
3756 | ||
3757 | nl_msg_put_u32(&buf, OVS_METER_ATTR_ID, meter_id.uint32); | |
3758 | ||
3759 | int error = dpif_netlink_meter_transact(&buf, &msg, | |
3760 | ovs_meter_stats_policy, a, | |
3761 | ARRAY_SIZE(ovs_meter_stats_policy)); | |
3762 | if (error) { | |
3763 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3764 | VLOG_INFO_RL(&rl, "dpif_netlink_meter_transact %s failed", | |
3765 | command == OVS_METER_CMD_GET ? "get" : "del"); | |
3766 | return error; | |
3767 | } | |
3768 | ||
3769 | if (stats | |
3770 | && a[OVS_METER_ATTR_ID] | |
3771 | && a[OVS_METER_ATTR_STATS] | |
3772 | && nl_attr_get_u32(a[OVS_METER_ATTR_ID]) == meter_id.uint32) { | |
3773 | /* return stats */ | |
3774 | const struct ovs_flow_stats *stat; | |
3775 | const struct nlattr *nla; | |
3776 | size_t left; | |
3777 | ||
3778 | stat = nl_attr_get(a[OVS_METER_ATTR_STATS]); | |
3779 | stats->packet_in_count = get_32aligned_u64(&stat->n_packets); | |
3780 | stats->byte_in_count = get_32aligned_u64(&stat->n_bytes); | |
3781 | ||
3782 | if (a[OVS_METER_ATTR_BANDS]) { | |
3783 | size_t n_bands = 0; | |
3784 | NL_NESTED_FOR_EACH (nla, left, a[OVS_METER_ATTR_BANDS]) { | |
3785 | const struct nlattr *band_nla; | |
3786 | band_nla = nl_attr_find_nested(nla, OVS_BAND_ATTR_STATS); | |
3787 | if (band_nla && nl_attr_get_size(band_nla) \ | |
3788 | == sizeof(struct ovs_flow_stats)) { | |
3789 | stat = nl_attr_get(band_nla); | |
3790 | ||
3791 | if (n_bands < max_bands) { | |
3792 | stats->bands[n_bands].packet_count | |
3793 | = get_32aligned_u64(&stat->n_packets); | |
3794 | stats->bands[n_bands].byte_count | |
3795 | = get_32aligned_u64(&stat->n_bytes); | |
3796 | ++n_bands; | |
3797 | } | |
3798 | } else { | |
3799 | stats->bands[n_bands].packet_count = 0; | |
3800 | stats->bands[n_bands].byte_count = 0; | |
3801 | ++n_bands; | |
3802 | } | |
3803 | } | |
3804 | stats->n_bands = n_bands; | |
3805 | } else { | |
3806 | /* For a non-existent meter, return 0 stats. */ | |
3807 | stats->n_bands = 0; | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | ofpbuf_delete(msg); | |
3812 | return error; | |
5dddf960 JR |
3813 | } |
3814 | ||
3815 | static int | |
80738e5f AZ |
3816 | dpif_netlink_meter_get(const struct dpif *dpif, ofproto_meter_id meter_id, |
3817 | struct ofputil_meter_stats *stats, uint16_t max_bands) | |
5dddf960 | 3818 | { |
80738e5f AZ |
3819 | return dpif_netlink_meter_get_stats(dpif, meter_id, stats, max_bands, |
3820 | OVS_METER_CMD_GET); | |
3821 | } | |
3822 | ||
3823 | static int | |
3824 | dpif_netlink_meter_del(struct dpif *dpif, ofproto_meter_id meter_id, | |
3825 | struct ofputil_meter_stats *stats, uint16_t max_bands) | |
3826 | { | |
3827 | return dpif_netlink_meter_get_stats(dpif, meter_id, stats, max_bands, | |
3828 | OVS_METER_CMD_DEL); | |
5dddf960 JR |
3829 | } |
3830 | ||
92d0d515 JP |
3831 | static bool |
3832 | probe_broken_meters__(struct dpif *dpif) | |
3833 | { | |
3834 | /* This test is destructive if a probe occurs while ovs-vswitchd is | |
3835 | * running (e.g., an ovs-dpctl meter command is called), so choose a | |
3836 | * random high meter id to make this less likely to occur. */ | |
3837 | ofproto_meter_id id1 = { 54545401 }; | |
3838 | ofproto_meter_id id2 = { 54545402 }; | |
3839 | struct ofputil_meter_band band = {OFPMBT13_DROP, 0, 1, 0}; | |
3840 | struct ofputil_meter_config config1 = { 1, OFPMF13_KBPS, 1, &band}; | |
3841 | struct ofputil_meter_config config2 = { 2, OFPMF13_KBPS, 1, &band}; | |
3842 | ||
3843 | /* Try adding two meters and make sure that they both come back with | |
60ebc04d JP |
3844 | * the proper meter id. Use the "__" version so that we don't cause |
3845 | * a recurve deadlock. */ | |
3846 | dpif_netlink_meter_set__(dpif, id1, &config1); | |
3847 | dpif_netlink_meter_set__(dpif, id2, &config2); | |
92d0d515 JP |
3848 | |
3849 | if (dpif_netlink_meter_get(dpif, id1, NULL, 0) | |
3850 | || dpif_netlink_meter_get(dpif, id2, NULL, 0)) { | |
3851 | VLOG_INFO("The kernel module has a broken meter implementation."); | |
3852 | return true; | |
3853 | } | |
3854 | ||
3855 | dpif_netlink_meter_del(dpif, id1, NULL, 0); | |
3856 | dpif_netlink_meter_del(dpif, id2, NULL, 0); | |
3857 | ||
3858 | return false; | |
3859 | } | |
3860 | ||
3861 | static bool | |
3862 | probe_broken_meters(struct dpif *dpif) | |
3863 | { | |
3864 | /* This is a once-only test because currently OVS only has at most a single | |
3865 | * Netlink capable datapath on any given platform. */ | |
3866 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3867 | ||
3868 | static bool broken_meters = false; | |
3869 | if (ovsthread_once_start(&once)) { | |
3870 | broken_meters = probe_broken_meters__(dpif); | |
3871 | ovsthread_once_done(&once); | |
3872 | } | |
3873 | return broken_meters; | |
3874 | } | |
5dddf960 | 3875 | \f |
93451a0a | 3876 | const struct dpif_class dpif_netlink_class = { |
1a6f1e2a | 3877 | "system", |
f87c1357 | 3878 | false, /* cleanup_required */ |
c8973eb6 | 3879 | NULL, /* init */ |
93451a0a | 3880 | dpif_netlink_enumerate, |
0aeaabc8 | 3881 | NULL, |
93451a0a AS |
3882 | dpif_netlink_open, |
3883 | dpif_netlink_close, | |
3884 | dpif_netlink_destroy, | |
3885 | dpif_netlink_run, | |
e4516b20 | 3886 | NULL, /* wait */ |
93451a0a AS |
3887 | dpif_netlink_get_stats, |
3888 | dpif_netlink_port_add, | |
3889 | dpif_netlink_port_del, | |
91364d18 | 3890 | NULL, /* port_set_config */ |
93451a0a AS |
3891 | dpif_netlink_port_query_by_number, |
3892 | dpif_netlink_port_query_by_name, | |
3893 | dpif_netlink_port_get_pid, | |
3894 | dpif_netlink_port_dump_start, | |
3895 | dpif_netlink_port_dump_next, | |
3896 | dpif_netlink_port_dump_done, | |
3897 | dpif_netlink_port_poll, | |
3898 | dpif_netlink_port_poll_wait, | |
3899 | dpif_netlink_flow_flush, | |
3900 | dpif_netlink_flow_dump_create, | |
3901 | dpif_netlink_flow_dump_destroy, | |
3902 | dpif_netlink_flow_dump_thread_create, | |
3903 | dpif_netlink_flow_dump_thread_destroy, | |
3904 | dpif_netlink_flow_dump_next, | |
3905 | dpif_netlink_operate, | |
3906 | dpif_netlink_recv_set, | |
3907 | dpif_netlink_handlers_set, | |
d4f6865c | 3908 | NULL, /* set_config */ |
93451a0a AS |
3909 | dpif_netlink_queue_to_priority, |
3910 | dpif_netlink_recv, | |
3911 | dpif_netlink_recv_wait, | |
3912 | dpif_netlink_recv_purge, | |
e4e74c3a | 3913 | NULL, /* register_dp_purge_cb */ |
6b31e073 RW |
3914 | NULL, /* register_upcall_cb */ |
3915 | NULL, /* enable_upcall */ | |
3916 | NULL, /* disable_upcall */ | |
b5cbbcf6 | 3917 | dpif_netlink_get_datapath_version, /* get_datapath_version */ |
c11c9f4a DDP |
3918 | dpif_netlink_ct_dump_start, |
3919 | dpif_netlink_ct_dump_next, | |
3920 | dpif_netlink_ct_dump_done, | |
5dddf960 | 3921 | dpif_netlink_ct_flush, |
c92339ad DB |
3922 | NULL, /* ct_set_maxconns */ |
3923 | NULL, /* ct_get_maxconns */ | |
875075b3 | 3924 | NULL, /* ct_get_nconns */ |
64207120 DB |
3925 | NULL, /* ct_set_tcp_seq_chk */ |
3926 | NULL, /* ct_get_tcp_seq_chk */ | |
906ff9d2 YHW |
3927 | dpif_netlink_ct_set_limits, |
3928 | dpif_netlink_ct_get_limits, | |
3929 | dpif_netlink_ct_del_limits, | |
1f161318 YHW |
3930 | dpif_netlink_ct_set_timeout_policy, |
3931 | dpif_netlink_ct_get_timeout_policy, | |
3932 | dpif_netlink_ct_del_timeout_policy, | |
3933 | dpif_netlink_ct_timeout_policy_dump_start, | |
3934 | dpif_netlink_ct_timeout_policy_dump_next, | |
3935 | dpif_netlink_ct_timeout_policy_dump_done, | |
187bb41f | 3936 | dpif_netlink_ct_get_timeout_policy_name, |
4ea96698 DB |
3937 | NULL, /* ipf_set_enabled */ |
3938 | NULL, /* ipf_set_min_frag */ | |
3939 | NULL, /* ipf_set_max_nfrags */ | |
3940 | NULL, /* ipf_get_status */ | |
3941 | NULL, /* ipf_dump_start */ | |
3942 | NULL, /* ipf_dump_next */ | |
3943 | NULL, /* ipf_dump_done */ | |
5dddf960 JR |
3944 | dpif_netlink_meter_get_features, |
3945 | dpif_netlink_meter_set, | |
3946 | dpif_netlink_meter_get, | |
3947 | dpif_netlink_meter_del, | |
96fba48f | 3948 | }; |
93451a0a | 3949 | |
96fba48f | 3950 | static int |
93451a0a | 3951 | dpif_netlink_init(void) |
96fba48f | 3952 | { |
eb8ed438 BP |
3953 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3954 | static int error; | |
982b8810 | 3955 | |
eb8ed438 | 3956 | if (ovsthread_once_start(&once)) { |
df2c07f4 JP |
3957 | error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY, |
3958 | &ovs_datapath_family); | |
37a1300c | 3959 | if (error) { |
e0e2410d | 3960 | VLOG_INFO("Generic Netlink family '%s' does not exist. " |
cae7529c CL |
3961 | "The Open vSwitch kernel module is probably not loaded.", |
3962 | OVS_DATAPATH_FAMILY); | |
37a1300c | 3963 | } |
f0fef760 | 3964 | if (!error) { |
df2c07f4 | 3965 | error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family); |
f0fef760 | 3966 | } |
37a1300c | 3967 | if (!error) { |
df2c07f4 | 3968 | error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family); |
37a1300c | 3969 | } |
aaff4b55 | 3970 | if (!error) { |
df2c07f4 JP |
3971 | error = nl_lookup_genl_family(OVS_PACKET_FAMILY, |
3972 | &ovs_packet_family); | |
aaff4b55 | 3973 | } |
c7178a0b EJ |
3974 | if (!error) { |
3975 | error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP, | |
b3dcb73c | 3976 | &ovs_vport_mcgroup); |
c7178a0b | 3977 | } |
80738e5f AZ |
3978 | if (!error) { |
3979 | if (nl_lookup_genl_family(OVS_METER_FAMILY, &ovs_meter_family)) { | |
3980 | VLOG_INFO("The kernel module does not support meters."); | |
3981 | } | |
3982 | } | |
906ff9d2 YHW |
3983 | if (nl_lookup_genl_family(OVS_CT_LIMIT_FAMILY, |
3984 | &ovs_ct_limit_family) < 0) { | |
3985 | VLOG_INFO("Generic Netlink family '%s' does not exist. " | |
3986 | "Please update the Open vSwitch kernel module to enable " | |
3987 | "the conntrack limit feature.", OVS_CT_LIMIT_FAMILY); | |
3988 | } | |
eb8ed438 | 3989 | |
921c370a EG |
3990 | ovs_tunnels_out_of_tree = dpif_netlink_rtnl_probe_oot_tunnels(); |
3991 | ||
eb8ed438 | 3992 | ovsthread_once_done(&once); |
982b8810 BP |
3993 | } |
3994 | ||
3995 | return error; | |
96fba48f BP |
3996 | } |
3997 | ||
c19e6535 | 3998 | bool |
93451a0a | 3999 | dpif_netlink_is_internal_device(const char *name) |
9fe3b9a2 | 4000 | { |
93451a0a | 4001 | struct dpif_netlink_vport reply; |
c19e6535 | 4002 | struct ofpbuf *buf; |
9fe3b9a2 | 4003 | int error; |
96fba48f | 4004 | |
93451a0a | 4005 | error = dpif_netlink_vport_get(name, &reply, &buf); |
c19e6535 BP |
4006 | if (!error) { |
4007 | ofpbuf_delete(buf); | |
141d9ce4 | 4008 | } else if (error != ENODEV && error != ENOENT) { |
c19e6535 | 4009 | VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)", |
10a89ef0 | 4010 | name, ovs_strerror(error)); |
96fba48f BP |
4011 | } |
4012 | ||
df2c07f4 | 4013 | return reply.type == OVS_VPORT_TYPE_INTERNAL; |
96fba48f | 4014 | } |
e0467f6d | 4015 | |
df2c07f4 | 4016 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
c19e6535 BP |
4017 | * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a |
4018 | * positive errno value. | |
4019 | * | |
4020 | * 'vport' will contain pointers into 'buf', so the caller should not free | |
4021 | * 'buf' while 'vport' is still in use. */ | |
4022 | static int | |
93451a0a | 4023 | dpif_netlink_vport_from_ofpbuf(struct dpif_netlink_vport *vport, |
c19e6535 BP |
4024 | const struct ofpbuf *buf) |
4025 | { | |
df2c07f4 JP |
4026 | static const struct nl_policy ovs_vport_policy[] = { |
4027 | [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 }, | |
4028 | [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 }, | |
4029 | [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, | |
1579cf67 | 4030 | [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_UNSPEC }, |
f7df9823 | 4031 | [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats), |
c19e6535 | 4032 | .optional = true }, |
df2c07f4 | 4033 | [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, |
bfda5239 | 4034 | [OVS_VPORT_ATTR_NETNSID] = { .type = NL_A_U32, .optional = true }, |
c19e6535 BP |
4035 | }; |
4036 | ||
93451a0a | 4037 | dpif_netlink_vport_init(vport); |
c19e6535 | 4038 | |
0a2869d5 BP |
4039 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4040 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4041 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4042 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4043 | ||
4044 | struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)]; | |
df2c07f4 JP |
4045 | if (!nlmsg || !genl || !ovs_header |
4046 | || nlmsg->nlmsg_type != ovs_vport_family | |
4047 | || !nl_policy_parse(&b, 0, ovs_vport_policy, a, | |
4048 | ARRAY_SIZE(ovs_vport_policy))) { | |
c19e6535 BP |
4049 | return EINVAL; |
4050 | } | |
c19e6535 | 4051 | |
f0fef760 | 4052 | vport->cmd = genl->cmd; |
df2c07f4 | 4053 | vport->dp_ifindex = ovs_header->dp_ifindex; |
4e022ec0 | 4054 | vport->port_no = nl_attr_get_odp_port(a[OVS_VPORT_ATTR_PORT_NO]); |
df2c07f4 JP |
4055 | vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]); |
4056 | vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]); | |
b063d9f0 | 4057 | if (a[OVS_VPORT_ATTR_UPCALL_PID]) { |
1579cf67 AW |
4058 | vport->n_upcall_pids = nl_attr_get_size(a[OVS_VPORT_ATTR_UPCALL_PID]) |
4059 | / (sizeof *vport->upcall_pids); | |
4060 | vport->upcall_pids = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]); | |
4061 | ||
b063d9f0 | 4062 | } |
df2c07f4 JP |
4063 | if (a[OVS_VPORT_ATTR_STATS]) { |
4064 | vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]); | |
4065 | } | |
df2c07f4 JP |
4066 | if (a[OVS_VPORT_ATTR_OPTIONS]) { |
4067 | vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]); | |
4068 | vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]); | |
c19e6535 | 4069 | } |
bfda5239 FL |
4070 | if (a[OVS_VPORT_ATTR_NETNSID]) { |
4071 | netnsid_set(&vport->netnsid, | |
4072 | nl_attr_get_u32(a[OVS_VPORT_ATTR_NETNSID])); | |
4073 | } else { | |
4074 | netnsid_set_local(&vport->netnsid); | |
4075 | } | |
c19e6535 BP |
4076 | return 0; |
4077 | } | |
4078 | ||
df2c07f4 | 4079 | /* Appends to 'buf' (which must initially be empty) a "struct ovs_header" |
c19e6535 BP |
4080 | * followed by Netlink attributes corresponding to 'vport'. */ |
4081 | static void | |
93451a0a AS |
4082 | dpif_netlink_vport_to_ofpbuf(const struct dpif_netlink_vport *vport, |
4083 | struct ofpbuf *buf) | |
c19e6535 | 4084 | { |
df2c07f4 | 4085 | struct ovs_header *ovs_header; |
f0fef760 | 4086 | |
df2c07f4 | 4087 | nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO, |
69685a88 | 4088 | vport->cmd, OVS_VPORT_VERSION); |
c19e6535 | 4089 | |
df2c07f4 JP |
4090 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4091 | ovs_header->dp_ifindex = vport->dp_ifindex; | |
c19e6535 | 4092 | |
4e022ec0 AW |
4093 | if (vport->port_no != ODPP_NONE) { |
4094 | nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no); | |
c19e6535 BP |
4095 | } |
4096 | ||
df2c07f4 JP |
4097 | if (vport->type != OVS_VPORT_TYPE_UNSPEC) { |
4098 | nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type); | |
c19e6535 BP |
4099 | } |
4100 | ||
4101 | if (vport->name) { | |
df2c07f4 | 4102 | nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name); |
c19e6535 BP |
4103 | } |
4104 | ||
1579cf67 AW |
4105 | if (vport->upcall_pids) { |
4106 | nl_msg_put_unspec(buf, OVS_VPORT_ATTR_UPCALL_PID, | |
4107 | vport->upcall_pids, | |
4108 | vport->n_upcall_pids * sizeof *vport->upcall_pids); | |
a24a6574 | 4109 | } |
b063d9f0 | 4110 | |
c19e6535 | 4111 | if (vport->stats) { |
df2c07f4 | 4112 | nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS, |
c19e6535 BP |
4113 | vport->stats, sizeof *vport->stats); |
4114 | } | |
4115 | ||
c19e6535 | 4116 | if (vport->options) { |
df2c07f4 | 4117 | nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS, |
c19e6535 BP |
4118 | vport->options, vport->options_len); |
4119 | } | |
c19e6535 BP |
4120 | } |
4121 | ||
4122 | /* Clears 'vport' to "empty" values. */ | |
4123 | void | |
93451a0a | 4124 | dpif_netlink_vport_init(struct dpif_netlink_vport *vport) |
c19e6535 BP |
4125 | { |
4126 | memset(vport, 0, sizeof *vport); | |
4e022ec0 | 4127 | vport->port_no = ODPP_NONE; |
c19e6535 BP |
4128 | } |
4129 | ||
4130 | /* Executes 'request' in the kernel datapath. If the command fails, returns a | |
4131 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4132 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
df2c07f4 | 4133 | * result of the command is expected to be an ovs_vport also, which is decoded |
c19e6535 BP |
4134 | * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the |
4135 | * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
4136 | int | |
93451a0a AS |
4137 | dpif_netlink_vport_transact(const struct dpif_netlink_vport *request, |
4138 | struct dpif_netlink_vport *reply, | |
4139 | struct ofpbuf **bufp) | |
c19e6535 | 4140 | { |
f0fef760 | 4141 | struct ofpbuf *request_buf; |
c19e6535 BP |
4142 | int error; |
4143 | ||
cb22974d | 4144 | ovs_assert((reply != NULL) == (bufp != NULL)); |
c19e6535 | 4145 | |
93451a0a | 4146 | error = dpif_netlink_init(); |
42bb6c72 BP |
4147 | if (error) { |
4148 | if (reply) { | |
4149 | *bufp = NULL; | |
93451a0a | 4150 | dpif_netlink_vport_init(reply); |
42bb6c72 BP |
4151 | } |
4152 | return error; | |
4153 | } | |
4154 | ||
f0fef760 | 4155 | request_buf = ofpbuf_new(1024); |
93451a0a | 4156 | dpif_netlink_vport_to_ofpbuf(request, request_buf); |
a88b4e04 | 4157 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
f0fef760 | 4158 | ofpbuf_delete(request_buf); |
c19e6535 | 4159 | |
f0fef760 BP |
4160 | if (reply) { |
4161 | if (!error) { | |
93451a0a | 4162 | error = dpif_netlink_vport_from_ofpbuf(reply, *bufp); |
f0fef760 | 4163 | } |
c19e6535 | 4164 | if (error) { |
93451a0a | 4165 | dpif_netlink_vport_init(reply); |
f0fef760 BP |
4166 | ofpbuf_delete(*bufp); |
4167 | *bufp = NULL; | |
c19e6535 | 4168 | } |
c19e6535 BP |
4169 | } |
4170 | return error; | |
4171 | } | |
4172 | ||
4173 | /* Obtains information about the kernel vport named 'name' and stores it into | |
4174 | * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no | |
4175 | * longer needed ('reply' will contain pointers into '*bufp'). */ | |
4176 | int | |
93451a0a AS |
4177 | dpif_netlink_vport_get(const char *name, struct dpif_netlink_vport *reply, |
4178 | struct ofpbuf **bufp) | |
c19e6535 | 4179 | { |
93451a0a | 4180 | struct dpif_netlink_vport request; |
c19e6535 | 4181 | |
93451a0a | 4182 | dpif_netlink_vport_init(&request); |
df2c07f4 | 4183 | request.cmd = OVS_VPORT_CMD_GET; |
c19e6535 BP |
4184 | request.name = name; |
4185 | ||
93451a0a | 4186 | return dpif_netlink_vport_transact(&request, reply, bufp); |
c19e6535 | 4187 | } |
93451a0a | 4188 | |
df2c07f4 | 4189 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
aaff4b55 BP |
4190 | * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a |
4191 | * positive errno value. | |
d6569377 BP |
4192 | * |
4193 | * 'dp' will contain pointers into 'buf', so the caller should not free 'buf' | |
4194 | * while 'dp' is still in use. */ | |
4195 | static int | |
93451a0a | 4196 | dpif_netlink_dp_from_ofpbuf(struct dpif_netlink_dp *dp, const struct ofpbuf *buf) |
d6569377 | 4197 | { |
df2c07f4 JP |
4198 | static const struct nl_policy ovs_datapath_policy[] = { |
4199 | [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, | |
f7df9823 | 4200 | [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats), |
d6569377 | 4201 | .optional = true }, |
847108dc AZ |
4202 | [OVS_DP_ATTR_MEGAFLOW_STATS] = { |
4203 | NL_POLICY_FOR(struct ovs_dp_megaflow_stats), | |
4204 | .optional = true }, | |
d6569377 BP |
4205 | }; |
4206 | ||
93451a0a | 4207 | dpif_netlink_dp_init(dp); |
d6569377 | 4208 | |
0a2869d5 BP |
4209 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4210 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4211 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4212 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4213 | ||
4214 | struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)]; | |
df2c07f4 JP |
4215 | if (!nlmsg || !genl || !ovs_header |
4216 | || nlmsg->nlmsg_type != ovs_datapath_family | |
4217 | || !nl_policy_parse(&b, 0, ovs_datapath_policy, a, | |
4218 | ARRAY_SIZE(ovs_datapath_policy))) { | |
d6569377 BP |
4219 | return EINVAL; |
4220 | } | |
d6569377 | 4221 | |
aaff4b55 | 4222 | dp->cmd = genl->cmd; |
df2c07f4 JP |
4223 | dp->dp_ifindex = ovs_header->dp_ifindex; |
4224 | dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]); | |
4225 | if (a[OVS_DP_ATTR_STATS]) { | |
6a54dedc | 4226 | dp->stats = nl_attr_get(a[OVS_DP_ATTR_STATS]); |
d6569377 | 4227 | } |
982b8810 | 4228 | |
847108dc | 4229 | if (a[OVS_DP_ATTR_MEGAFLOW_STATS]) { |
6a54dedc | 4230 | dp->megaflow_stats = nl_attr_get(a[OVS_DP_ATTR_MEGAFLOW_STATS]); |
847108dc AZ |
4231 | } |
4232 | ||
d6569377 BP |
4233 | return 0; |
4234 | } | |
4235 | ||
aaff4b55 | 4236 | /* Appends to 'buf' the Generic Netlink message described by 'dp'. */ |
d6569377 | 4237 | static void |
93451a0a | 4238 | dpif_netlink_dp_to_ofpbuf(const struct dpif_netlink_dp *dp, struct ofpbuf *buf) |
d6569377 | 4239 | { |
df2c07f4 | 4240 | struct ovs_header *ovs_header; |
d6569377 | 4241 | |
df2c07f4 | 4242 | nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family, |
69685a88 JG |
4243 | NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, |
4244 | OVS_DATAPATH_VERSION); | |
aaff4b55 | 4245 | |
df2c07f4 JP |
4246 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4247 | ovs_header->dp_ifindex = dp->dp_ifindex; | |
d6569377 BP |
4248 | |
4249 | if (dp->name) { | |
df2c07f4 | 4250 | nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name); |
d6569377 BP |
4251 | } |
4252 | ||
a24a6574 BP |
4253 | if (dp->upcall_pid) { |
4254 | nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid); | |
4255 | } | |
b063d9f0 | 4256 | |
b7fd5e38 TG |
4257 | if (dp->user_features) { |
4258 | nl_msg_put_u32(buf, OVS_DP_ATTR_USER_FEATURES, dp->user_features); | |
4259 | } | |
4260 | ||
df2c07f4 | 4261 | /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */ |
d6569377 BP |
4262 | } |
4263 | ||
4264 | /* Clears 'dp' to "empty" values. */ | |
d3d8f1f7 | 4265 | static void |
93451a0a | 4266 | dpif_netlink_dp_init(struct dpif_netlink_dp *dp) |
d6569377 BP |
4267 | { |
4268 | memset(dp, 0, sizeof *dp); | |
d6569377 BP |
4269 | } |
4270 | ||
aaff4b55 | 4271 | static void |
93451a0a | 4272 | dpif_netlink_dp_dump_start(struct nl_dump *dump) |
aaff4b55 | 4273 | { |
93451a0a | 4274 | struct dpif_netlink_dp request; |
aaff4b55 BP |
4275 | struct ofpbuf *buf; |
4276 | ||
93451a0a | 4277 | dpif_netlink_dp_init(&request); |
df2c07f4 | 4278 | request.cmd = OVS_DP_CMD_GET; |
aaff4b55 BP |
4279 | |
4280 | buf = ofpbuf_new(1024); | |
93451a0a | 4281 | dpif_netlink_dp_to_ofpbuf(&request, buf); |
a88b4e04 | 4282 | nl_dump_start(dump, NETLINK_GENERIC, buf); |
aaff4b55 BP |
4283 | ofpbuf_delete(buf); |
4284 | } | |
4285 | ||
d6569377 BP |
4286 | /* Executes 'request' in the kernel datapath. If the command fails, returns a |
4287 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4288 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
aaff4b55 BP |
4289 | * result of the command is expected to be of the same form, which is decoded |
4290 | * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the | |
4291 | * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4292 | static int |
93451a0a AS |
4293 | dpif_netlink_dp_transact(const struct dpif_netlink_dp *request, |
4294 | struct dpif_netlink_dp *reply, struct ofpbuf **bufp) | |
d6569377 | 4295 | { |
aaff4b55 | 4296 | struct ofpbuf *request_buf; |
d6569377 | 4297 | int error; |
d6569377 | 4298 | |
cb22974d | 4299 | ovs_assert((reply != NULL) == (bufp != NULL)); |
d6569377 | 4300 | |
aaff4b55 | 4301 | request_buf = ofpbuf_new(1024); |
93451a0a | 4302 | dpif_netlink_dp_to_ofpbuf(request, request_buf); |
a88b4e04 | 4303 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
aaff4b55 | 4304 | ofpbuf_delete(request_buf); |
d6569377 | 4305 | |
aaff4b55 | 4306 | if (reply) { |
93451a0a | 4307 | dpif_netlink_dp_init(reply); |
aaff4b55 | 4308 | if (!error) { |
93451a0a | 4309 | error = dpif_netlink_dp_from_ofpbuf(reply, *bufp); |
aaff4b55 | 4310 | } |
d6569377 | 4311 | if (error) { |
aaff4b55 BP |
4312 | ofpbuf_delete(*bufp); |
4313 | *bufp = NULL; | |
d6569377 | 4314 | } |
d6569377 BP |
4315 | } |
4316 | return error; | |
4317 | } | |
4318 | ||
4319 | /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. | |
4320 | * The caller must free '*bufp' when the reply is no longer needed ('reply' | |
4321 | * will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4322 | static int |
93451a0a AS |
4323 | dpif_netlink_dp_get(const struct dpif *dpif_, struct dpif_netlink_dp *reply, |
4324 | struct ofpbuf **bufp) | |
d6569377 | 4325 | { |
93451a0a AS |
4326 | struct dpif_netlink *dpif = dpif_netlink_cast(dpif_); |
4327 | struct dpif_netlink_dp request; | |
d6569377 | 4328 | |
93451a0a | 4329 | dpif_netlink_dp_init(&request); |
df2c07f4 | 4330 | request.cmd = OVS_DP_CMD_GET; |
254f2dc8 | 4331 | request.dp_ifindex = dpif->dp_ifindex; |
d6569377 | 4332 | |
93451a0a | 4333 | return dpif_netlink_dp_transact(&request, reply, bufp); |
d6569377 | 4334 | } |
93451a0a | 4335 | |
df2c07f4 | 4336 | /* Parses the contents of 'buf', which contains a "struct ovs_header" followed |
37a1300c | 4337 | * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a |
d6569377 BP |
4338 | * positive errno value. |
4339 | * | |
4340 | * 'flow' will contain pointers into 'buf', so the caller should not free 'buf' | |
4341 | * while 'flow' is still in use. */ | |
4342 | static int | |
93451a0a AS |
4343 | dpif_netlink_flow_from_ofpbuf(struct dpif_netlink_flow *flow, |
4344 | const struct ofpbuf *buf) | |
d6569377 | 4345 | { |
70e5ed6f JS |
4346 | static const struct nl_policy ovs_flow_policy[__OVS_FLOW_ATTR_MAX] = { |
4347 | [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED, .optional = true }, | |
e6cc0bab | 4348 | [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true }, |
df2c07f4 | 4349 | [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, |
f7df9823 | 4350 | [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats), |
d6569377 | 4351 | .optional = true }, |
df2c07f4 JP |
4352 | [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, |
4353 | [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, | |
ab79d262 | 4354 | [OVS_FLOW_ATTR_UFID] = { .type = NL_A_U128, .optional = true }, |
df2c07f4 | 4355 | /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */ |
43f9ac0a | 4356 | /* The kernel never uses OVS_FLOW_ATTR_PROBE. */ |
70e5ed6f | 4357 | /* The kernel never uses OVS_FLOW_ATTR_UFID_FLAGS. */ |
d6569377 BP |
4358 | }; |
4359 | ||
93451a0a | 4360 | dpif_netlink_flow_init(flow); |
d6569377 | 4361 | |
0a2869d5 BP |
4362 | struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size); |
4363 | struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); | |
4364 | struct genlmsghdr *genl = ofpbuf_try_pull(&b, sizeof *genl); | |
4365 | struct ovs_header *ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header); | |
4366 | ||
4367 | struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)]; | |
df2c07f4 JP |
4368 | if (!nlmsg || !genl || !ovs_header |
4369 | || nlmsg->nlmsg_type != ovs_flow_family | |
4370 | || !nl_policy_parse(&b, 0, ovs_flow_policy, a, | |
4371 | ARRAY_SIZE(ovs_flow_policy))) { | |
d6569377 BP |
4372 | return EINVAL; |
4373 | } | |
70e5ed6f JS |
4374 | if (!a[OVS_FLOW_ATTR_KEY] && !a[OVS_FLOW_ATTR_UFID]) { |
4375 | return EINVAL; | |
4376 | } | |
d6569377 | 4377 | |
37a1300c | 4378 | flow->nlmsg_flags = nlmsg->nlmsg_flags; |
df2c07f4 | 4379 | flow->dp_ifindex = ovs_header->dp_ifindex; |
70e5ed6f JS |
4380 | if (a[OVS_FLOW_ATTR_KEY]) { |
4381 | flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]); | |
4382 | flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]); | |
4383 | } | |
e6cc0bab | 4384 | |
70e5ed6f | 4385 | if (a[OVS_FLOW_ATTR_UFID]) { |
ab79d262 | 4386 | flow->ufid = nl_attr_get_u128(a[OVS_FLOW_ATTR_UFID]); |
70e5ed6f JS |
4387 | flow->ufid_present = true; |
4388 | } | |
e6cc0bab AZ |
4389 | if (a[OVS_FLOW_ATTR_MASK]) { |
4390 | flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]); | |
4391 | flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]); | |
4392 | } | |
df2c07f4 JP |
4393 | if (a[OVS_FLOW_ATTR_ACTIONS]) { |
4394 | flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]); | |
4395 | flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]); | |
d6569377 | 4396 | } |
df2c07f4 JP |
4397 | if (a[OVS_FLOW_ATTR_STATS]) { |
4398 | flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]); | |
d6569377 | 4399 | } |
df2c07f4 JP |
4400 | if (a[OVS_FLOW_ATTR_TCP_FLAGS]) { |
4401 | flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]); | |
d6569377 | 4402 | } |
df2c07f4 JP |
4403 | if (a[OVS_FLOW_ATTR_USED]) { |
4404 | flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]); | |
9e980142 | 4405 | } |
d6569377 BP |
4406 | return 0; |
4407 | } | |
4408 | ||
beb75a40 JS |
4409 | |
4410 | /* | |
a8a3eee4 JS |
4411 | * If PACKET_TYPE attribute is present in 'data', it filters PACKET_TYPE out. |
4412 | * If the flow is not Ethernet, the OVS_KEY_ATTR_PACKET_TYPE is converted to | |
4413 | * OVS_KEY_ATTR_ETHERTYPE. Puts 'data' to 'buf'. | |
beb75a40 JS |
4414 | */ |
4415 | static void | |
4416 | put_exclude_packet_type(struct ofpbuf *buf, uint16_t type, | |
4417 | const struct nlattr *data, uint16_t data_len) | |
4418 | { | |
4419 | const struct nlattr *packet_type; | |
4420 | ||
4421 | packet_type = nl_attr_find__(data, data_len, OVS_KEY_ATTR_PACKET_TYPE); | |
4422 | ||
4423 | if (packet_type) { | |
4424 | /* exclude PACKET_TYPE Netlink attribute. */ | |
4425 | ovs_assert(NLA_ALIGN(packet_type->nla_len) == NL_A_U32_SIZE); | |
4426 | size_t packet_type_len = NL_A_U32_SIZE; | |
4427 | size_t first_chunk_size = (uint8_t *)packet_type - (uint8_t *)data; | |
4428 | size_t second_chunk_size = data_len - first_chunk_size | |
4429 | - packet_type_len; | |
beb75a40 | 4430 | struct nlattr *next_attr = nl_attr_next(packet_type); |
1ca5b61b | 4431 | size_t ofs; |
beb75a40 | 4432 | |
1ca5b61b JS |
4433 | ofs = nl_msg_start_nested(buf, type); |
4434 | nl_msg_put(buf, data, first_chunk_size); | |
4435 | nl_msg_put(buf, next_attr, second_chunk_size); | |
a8a3eee4 JS |
4436 | if (!nl_attr_find__(data, data_len, OVS_KEY_ATTR_ETHERNET)) { |
4437 | ovs_be16 pt = pt_ns_type_be(nl_attr_get_be32(packet_type)); | |
4438 | const struct nlattr *nla; | |
4439 | ||
7c5793e6 | 4440 | nla = nl_attr_find(buf, ofs + NLA_HDRLEN, OVS_KEY_ATTR_ETHERTYPE); |
a8a3eee4 JS |
4441 | if (nla) { |
4442 | ovs_be16 *ethertype; | |
4443 | ||
4444 | ethertype = CONST_CAST(ovs_be16 *, nl_attr_get(nla)); | |
4445 | *ethertype = pt; | |
4446 | } else { | |
4447 | nl_msg_put_be16(buf, OVS_KEY_ATTR_ETHERTYPE, pt); | |
4448 | } | |
4449 | } | |
1ca5b61b | 4450 | nl_msg_end_nested(buf, ofs); |
beb75a40 JS |
4451 | } else { |
4452 | nl_msg_put_unspec(buf, type, data, data_len); | |
4453 | } | |
4454 | } | |
4455 | ||
df2c07f4 | 4456 | /* Appends to 'buf' (which must initially be empty) a "struct ovs_header" |
d6569377 BP |
4457 | * followed by Netlink attributes corresponding to 'flow'. */ |
4458 | static void | |
93451a0a AS |
4459 | dpif_netlink_flow_to_ofpbuf(const struct dpif_netlink_flow *flow, |
4460 | struct ofpbuf *buf) | |
d6569377 | 4461 | { |
df2c07f4 | 4462 | struct ovs_header *ovs_header; |
d6569377 | 4463 | |
df2c07f4 | 4464 | nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family, |
30b44744 | 4465 | NLM_F_REQUEST | flow->nlmsg_flags, |
69685a88 | 4466 | flow->cmd, OVS_FLOW_VERSION); |
37a1300c | 4467 | |
df2c07f4 JP |
4468 | ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header); |
4469 | ovs_header->dp_ifindex = flow->dp_ifindex; | |
d6569377 | 4470 | |
70e5ed6f | 4471 | if (flow->ufid_present) { |
ab79d262 | 4472 | nl_msg_put_u128(buf, OVS_FLOW_ATTR_UFID, flow->ufid); |
70e5ed6f JS |
4473 | } |
4474 | if (flow->ufid_terse) { | |
4475 | nl_msg_put_u32(buf, OVS_FLOW_ATTR_UFID_FLAGS, | |
4476 | OVS_UFID_F_OMIT_KEY | OVS_UFID_F_OMIT_MASK | |
4477 | | OVS_UFID_F_OMIT_ACTIONS); | |
4478 | } | |
64bb477f JS |
4479 | if (!flow->ufid_terse || !flow->ufid_present) { |
4480 | if (flow->key_len) { | |
beb75a40 JS |
4481 | put_exclude_packet_type(buf, OVS_FLOW_ATTR_KEY, flow->key, |
4482 | flow->key_len); | |
64bb477f | 4483 | } |
64bb477f | 4484 | if (flow->mask_len) { |
beb75a40 JS |
4485 | put_exclude_packet_type(buf, OVS_FLOW_ATTR_MASK, flow->mask, |
4486 | flow->mask_len); | |
64bb477f JS |
4487 | } |
4488 | if (flow->actions || flow->actions_len) { | |
4489 | nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS, | |
4490 | flow->actions, flow->actions_len); | |
4491 | } | |
d6569377 BP |
4492 | } |
4493 | ||
4494 | /* We never need to send these to the kernel. */ | |
cb22974d BP |
4495 | ovs_assert(!flow->stats); |
4496 | ovs_assert(!flow->tcp_flags); | |
4497 | ovs_assert(!flow->used); | |
d6569377 BP |
4498 | |
4499 | if (flow->clear) { | |
df2c07f4 | 4500 | nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR); |
d6569377 | 4501 | } |
43f9ac0a JR |
4502 | if (flow->probe) { |
4503 | nl_msg_put_flag(buf, OVS_FLOW_ATTR_PROBE); | |
4504 | } | |
d6569377 BP |
4505 | } |
4506 | ||
4507 | /* Clears 'flow' to "empty" values. */ | |
d3d8f1f7 | 4508 | static void |
93451a0a | 4509 | dpif_netlink_flow_init(struct dpif_netlink_flow *flow) |
d6569377 BP |
4510 | { |
4511 | memset(flow, 0, sizeof *flow); | |
4512 | } | |
4513 | ||
4514 | /* Executes 'request' in the kernel datapath. If the command fails, returns a | |
4515 | * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 | |
4516 | * without doing anything else. If 'reply' and 'bufp' are nonnull, then the | |
37a1300c BP |
4517 | * result of the command is expected to be a flow also, which is decoded and |
4518 | * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply | |
4519 | * is no longer needed ('reply' will contain pointers into '*bufp'). */ | |
d3d8f1f7 | 4520 | static int |
93451a0a AS |
4521 | dpif_netlink_flow_transact(struct dpif_netlink_flow *request, |
4522 | struct dpif_netlink_flow *reply, | |
4523 | struct ofpbuf **bufp) | |
d6569377 | 4524 | { |
37a1300c | 4525 | struct ofpbuf *request_buf; |
d6569377 | 4526 | int error; |
d6569377 | 4527 | |
cb22974d | 4528 | ovs_assert((reply != NULL) == (bufp != NULL)); |
d6569377 | 4529 | |
30b44744 BP |
4530 | if (reply) { |
4531 | request->nlmsg_flags |= NLM_F_ECHO; | |
4532 | } | |
4533 | ||
37a1300c | 4534 | request_buf = ofpbuf_new(1024); |
93451a0a | 4535 | dpif_netlink_flow_to_ofpbuf(request, request_buf); |
a88b4e04 | 4536 | error = nl_transact(NETLINK_GENERIC, request_buf, bufp); |
37a1300c | 4537 | ofpbuf_delete(request_buf); |
d6569377 | 4538 | |
37a1300c BP |
4539 | if (reply) { |
4540 | if (!error) { | |
93451a0a | 4541 | error = dpif_netlink_flow_from_ofpbuf(reply, *bufp); |
37a1300c | 4542 | } |
d6569377 | 4543 | if (error) { |
93451a0a | 4544 | dpif_netlink_flow_init(reply); |
37a1300c BP |
4545 | ofpbuf_delete(*bufp); |
4546 | *bufp = NULL; | |
d6569377 | 4547 | } |
d6569377 BP |
4548 | } |
4549 | return error; | |
4550 | } | |
4551 | ||
4552 | static void | |
93451a0a AS |
4553 | dpif_netlink_flow_get_stats(const struct dpif_netlink_flow *flow, |
4554 | struct dpif_flow_stats *stats) | |
d6569377 BP |
4555 | { |
4556 | if (flow->stats) { | |
6a54dedc BP |
4557 | stats->n_packets = get_32aligned_u64(&flow->stats->n_packets); |
4558 | stats->n_bytes = get_32aligned_u64(&flow->stats->n_bytes); | |
d6569377 BP |
4559 | } else { |
4560 | stats->n_packets = 0; | |
4561 | stats->n_bytes = 0; | |
4562 | } | |
0e70cdcb | 4563 | stats->used = flow->used ? get_32aligned_u64(flow->used) : 0; |
d6569377 BP |
4564 | stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; |
4565 | } | |
e0467f6d | 4566 | |
14b4d2f9 BP |
4567 | /* Logs information about a packet that was recently lost in 'ch' (in |
4568 | * 'dpif_'). */ | |
4569 | static void | |
93451a0a | 4570 | report_loss(struct dpif_netlink *dpif, struct dpif_channel *ch, uint32_t ch_idx, |
1579cf67 | 4571 | uint32_t handler_id) |
14b4d2f9 | 4572 | { |
14b4d2f9 | 4573 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); |
14b4d2f9 BP |
4574 | struct ds s; |
4575 | ||
8d675c5a | 4576 | if (VLOG_DROP_WARN(&rl)) { |
14b4d2f9 BP |
4577 | return; |
4578 | } | |
4579 | ||
4580 | ds_init(&s); | |
4581 | if (ch->last_poll != LLONG_MIN) { | |
4582 | ds_put_format(&s, " (last polled %lld ms ago)", | |
4583 | time_msec() - ch->last_poll); | |
4584 | } | |
14b4d2f9 | 4585 | |
1579cf67 | 4586 | VLOG_WARN("%s: lost packet on port channel %u of handler %u", |
9b00386b | 4587 | dpif_name(&dpif->dpif), ch_idx, handler_id); |
14b4d2f9 BP |
4588 | ds_destroy(&s); |
4589 | } |