]>
Commit | Line | Data |
---|---|---|
83b03fe0 | 1 | /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc. |
e1ec7dd4 EJ |
2 | * |
3 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
4 | * you may not use this file except in compliance with the License. | |
5 | * You may obtain a copy of the License at: | |
6 | * | |
7 | * http://www.apache.org/licenses/LICENSE-2.0 | |
8 | * | |
9 | * Unless required by applicable law or agreed to in writing, software | |
10 | * distributed under the License is distributed on an "AS IS" BASIS, | |
11 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
12 | * See the License for the specific language governing permissions and | |
13 | * limitations under the License. */ | |
14 | ||
15 | #include <config.h> | |
16 | #include "ofproto-dpif-upcall.h" | |
17 | ||
18 | #include <errno.h> | |
19 | #include <stdbool.h> | |
20 | #include <inttypes.h> | |
21 | ||
0fb7792a | 22 | #include "connmgr.h" |
e1ec7dd4 | 23 | #include "coverage.h" |
9fce0584 | 24 | #include "cmap.h" |
e1ec7dd4 | 25 | #include "dpif.h" |
3e8a2ad1 | 26 | #include "openvswitch/dynamic-string.h" |
e1ec7dd4 | 27 | #include "fail-open.h" |
05067881 | 28 | #include "guarded-list.h" |
e1ec7dd4 | 29 | #include "latch.h" |
b19bab5b | 30 | #include "openvswitch/list.h" |
e1ec7dd4 | 31 | #include "netlink.h" |
64c96779 | 32 | #include "openvswitch/ofpbuf.h" |
10e57640 EJ |
33 | #include "ofproto-dpif-ipfix.h" |
34 | #include "ofproto-dpif-sflow.h" | |
e79a6c83 | 35 | #include "ofproto-dpif-xlate.h" |
901a517e | 36 | #include "ofproto-dpif-xlate-cache.h" |
0f2ea848 | 37 | #include "ovs-rcu.h" |
e1ec7dd4 EJ |
38 | #include "packets.h" |
39 | #include "poll-loop.h" | |
e22d52ee EJ |
40 | #include "seq.h" |
41 | #include "unixctl.h" | |
e6211adc | 42 | #include "openvswitch/vlog.h" |
e1ec7dd4 EJ |
43 | |
44 | #define MAX_QUEUE_LENGTH 512 | |
6b31e073 | 45 | #define UPCALL_MAX_BATCH 64 |
e79a6c83 | 46 | #define REVALIDATE_MAX_BATCH 50 |
e1ec7dd4 EJ |
47 | |
48 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif_upcall); | |
49 | ||
ec47af51 JS |
50 | COVERAGE_DEFINE(dumped_duplicate_flow); |
51 | COVERAGE_DEFINE(dumped_new_flow); | |
23597df0 JS |
52 | COVERAGE_DEFINE(handler_duplicate_upcall); |
53 | COVERAGE_DEFINE(upcall_ukey_contention); | |
67f08985 | 54 | COVERAGE_DEFINE(upcall_ukey_replace); |
3b62a9d3 | 55 | COVERAGE_DEFINE(revalidate_missed_dp_flow); |
73a3c475 | 56 | |
9a159f74 AW |
57 | /* A thread that reads upcalls from dpif, forwards each upcall's packet, |
58 | * and possibly sets up a kernel flow as a cache. */ | |
e1ec7dd4 EJ |
59 | struct handler { |
60 | struct udpif *udpif; /* Parent udpif. */ | |
61 | pthread_t thread; /* Thread ID. */ | |
9a159f74 | 62 | uint32_t handler_id; /* Handler id. */ |
e1ec7dd4 EJ |
63 | }; |
64 | ||
b8d3daeb | 65 | /* In the absence of a multiple-writer multiple-reader datastructure for |
dcf5840f JS |
66 | * storing udpif_keys ("ukeys"), we use a large number of cmaps, each with its |
67 | * own lock for writing. */ | |
b8d3daeb JS |
68 | #define N_UMAPS 512 /* per udpif. */ |
69 | struct umap { | |
70 | struct ovs_mutex mutex; /* Take for writing to the following. */ | |
71 | struct cmap cmap; /* Datapath flow keys. */ | |
72 | }; | |
73 | ||
7d170098 | 74 | /* A thread that processes datapath flows, updates OpenFlow statistics, and |
dcf5840f JS |
75 | * updates or removes them if necessary. |
76 | * | |
77 | * Revalidator threads operate in two phases: "dump" and "sweep". In between | |
78 | * each phase, all revalidators sync up so that all revalidator threads are | |
79 | * either in one phase or the other, but not a combination. | |
80 | * | |
81 | * During the dump phase, revalidators fetch flows from the datapath and | |
82 | * attribute the statistics to OpenFlow rules. Each datapath flow has a | |
83 | * corresponding ukey which caches the most recently seen statistics. If | |
84 | * a flow needs to be deleted (for example, because it is unused over a | |
85 | * period of time), revalidator threads may delete the flow during the | |
86 | * dump phase. The datapath is not guaranteed to reliably dump all flows | |
87 | * from the datapath, and there is no mapping between datapath flows to | |
88 | * revalidators, so a particular flow may be handled by zero or more | |
89 | * revalidators during a single dump phase. To avoid duplicate attribution | |
90 | * of statistics, ukeys are never deleted during this phase. | |
91 | * | |
92 | * During the sweep phase, each revalidator takes ownership of a different | |
93 | * slice of umaps and sweeps through all ukeys in those umaps to figure out | |
94 | * whether they need to be deleted. During this phase, revalidators may | |
95 | * fetch individual flows which were not dumped during the dump phase to | |
96 | * validate them and attribute statistics. | |
97 | */ | |
e79a6c83 EJ |
98 | struct revalidator { |
99 | struct udpif *udpif; /* Parent udpif. */ | |
e79a6c83 | 100 | pthread_t thread; /* Thread ID. */ |
8ba0a522 | 101 | unsigned int id; /* ovsthread_id_self(). */ |
e79a6c83 EJ |
102 | }; |
103 | ||
e1ec7dd4 EJ |
104 | /* An upcall handler for ofproto_dpif. |
105 | * | |
9a159f74 AW |
106 | * udpif keeps records of two kind of logically separate units: |
107 | * | |
108 | * upcall handling | |
109 | * --------------- | |
110 | * | |
111 | * - An array of 'struct handler's for upcall handling and flow | |
112 | * installation. | |
e79a6c83 | 113 | * |
9a159f74 AW |
114 | * flow revalidation |
115 | * ----------------- | |
116 | * | |
7d170098 EJ |
117 | * - Revalidation threads which read the datapath flow table and maintains |
118 | * them. | |
119 | */ | |
e1ec7dd4 | 120 | struct udpif { |
ca6ba700 | 121 | struct ovs_list list_node; /* In all_udpifs list. */ |
e22d52ee | 122 | |
e1ec7dd4 EJ |
123 | struct dpif *dpif; /* Datapath handle. */ |
124 | struct dpif_backer *backer; /* Opaque dpif_backer pointer. */ | |
125 | ||
10e57640 | 126 | struct handler *handlers; /* Upcall handlers. */ |
e1ec7dd4 EJ |
127 | size_t n_handlers; |
128 | ||
e79a6c83 EJ |
129 | struct revalidator *revalidators; /* Flow revalidators. */ |
130 | size_t n_revalidators; | |
131 | ||
e79a6c83 EJ |
132 | struct latch exit_latch; /* Tells child threads to exit. */ |
133 | ||
7d170098 EJ |
134 | /* Revalidation. */ |
135 | struct seq *reval_seq; /* Incremented to force revalidation. */ | |
7d170098 | 136 | bool reval_exit; /* Set by leader on 'exit_latch. */ |
d8043da7 | 137 | struct ovs_barrier reval_barrier; /* Barrier used by revalidators. */ |
ac64794a | 138 | struct dpif_flow_dump *dump; /* DPIF flow dump state. */ |
e79a6c83 | 139 | long long int dump_duration; /* Duration of the last flow dump. */ |
7d170098 | 140 | struct seq *dump_seq; /* Increments each dump iteration. */ |
64bb477f | 141 | atomic_bool enable_ufid; /* If true, skip dumping flow attrs. */ |
7d170098 | 142 | |
dba82d38 AW |
143 | /* These variables provide a mechanism for the main thread to pause |
144 | * all revalidation without having to completely shut the threads down. | |
145 | * 'pause_latch' is shared between the main thread and the lead | |
146 | * revalidator thread, so when it is desirable to halt revalidation, the | |
147 | * main thread will set the latch. 'pause' and 'pause_barrier' are shared | |
148 | * by revalidator threads. The lead revalidator will set 'pause' when it | |
149 | * observes the latch has been set, and this will cause all revalidator | |
150 | * threads to wait on 'pause_barrier' at the beginning of the next | |
151 | * revalidation round. */ | |
152 | bool pause; /* Set by leader on 'pause_latch. */ | |
153 | struct latch pause_latch; /* Set to force revalidators pause. */ | |
154 | struct ovs_barrier pause_barrier; /* Barrier used to pause all */ | |
155 | /* revalidators by main thread. */ | |
156 | ||
b8d3daeb | 157 | /* There are 'N_UMAPS' maps containing 'struct udpif_key' elements. |
7d170098 EJ |
158 | * |
159 | * During the flow dump phase, revalidators insert into these with a random | |
160 | * distribution. During the garbage collection phase, each revalidator | |
b8d3daeb JS |
161 | * takes care of garbage collecting a slice of these maps. */ |
162 | struct umap *ukeys; | |
e1ec7dd4 | 163 | |
e79a6c83 EJ |
164 | /* Datapath flow statistics. */ |
165 | unsigned int max_n_flows; | |
166 | unsigned int avg_n_flows; | |
e1ec7dd4 | 167 | |
e79a6c83 | 168 | /* Following fields are accessed and modified by different threads. */ |
e79a6c83 | 169 | atomic_uint flow_limit; /* Datapath flow hard limit. */ |
64ca9472 JS |
170 | |
171 | /* n_flows_mutex prevents multiple threads updating these concurrently. */ | |
b482e960 | 172 | atomic_uint n_flows; /* Number of flows in the datapath. */ |
64ca9472 JS |
173 | atomic_llong n_flows_timestamp; /* Last time n_flows was updated. */ |
174 | struct ovs_mutex n_flows_mutex; | |
27f57736 JS |
175 | |
176 | /* Following fields are accessed and modified only from the main thread. */ | |
177 | struct unixctl_conn **conns; /* Connections waiting on dump_seq. */ | |
178 | uint64_t conn_seq; /* Corresponds to 'dump_seq' when | |
179 | conns[n_conns-1] was stored. */ | |
180 | size_t n_conns; /* Number of connections waiting. */ | |
e1ec7dd4 EJ |
181 | }; |
182 | ||
10e57640 EJ |
183 | enum upcall_type { |
184 | BAD_UPCALL, /* Some kind of bug somewhere. */ | |
185 | MISS_UPCALL, /* A flow miss. */ | |
186 | SFLOW_UPCALL, /* sFlow sample. */ | |
187 | FLOW_SAMPLE_UPCALL, /* Per-flow sampling. */ | |
188 | IPFIX_UPCALL /* Per-bridge sampling. */ | |
189 | }; | |
190 | ||
43b2f131 EJ |
191 | enum reval_result { |
192 | UKEY_KEEP, | |
193 | UKEY_DELETE, | |
194 | UKEY_MODIFY | |
195 | }; | |
196 | ||
10e57640 | 197 | struct upcall { |
cc377352 | 198 | struct ofproto_dpif *ofproto; /* Parent ofproto. */ |
e672ff9b JR |
199 | const struct recirc_id_node *recirc; /* Recirculation context. */ |
200 | bool have_recirc_ref; /* Reference held on recirc ctx? */ | |
a0bab870 | 201 | |
cc377352 EJ |
202 | /* The flow and packet are only required to be constant when using |
203 | * dpif-netdev. If a modification is absolutely necessary, a const cast | |
204 | * may be used with other datapaths. */ | |
205 | const struct flow *flow; /* Parsed representation of the packet. */ | |
7af12bd7 | 206 | const ovs_u128 *ufid; /* Unique identifier for 'flow'. */ |
bd5131ba | 207 | unsigned pmd_id; /* Datapath poll mode driver id. */ |
cf62fa4c | 208 | const struct dp_packet *packet; /* Packet associated with this upcall. */ |
cc377352 | 209 | ofp_port_t in_port; /* OpenFlow in port, or OFPP_NONE. */ |
27130224 AZ |
210 | uint16_t mru; /* If !0, Maximum receive unit of |
211 | fragmented IP packet */ | |
a0bab870 | 212 | |
cc377352 EJ |
213 | enum dpif_upcall_type type; /* Datapath type of the upcall. */ |
214 | const struct nlattr *userdata; /* Userdata for DPIF_UC_ACTION Upcalls. */ | |
7321bda3 | 215 | const struct nlattr *actions; /* Flow actions in DPIF_UC_ACTION Upcalls. */ |
cc377352 EJ |
216 | |
217 | bool xout_initialized; /* True if 'xout' must be uninitialized. */ | |
218 | struct xlate_out xout; /* Result of xlate_actions(). */ | |
1520ef4f | 219 | struct ofpbuf odp_actions; /* Datapath actions from xlate_actions(). */ |
49a73e0c | 220 | struct flow_wildcards wc; /* Dependencies that megaflow must match. */ |
2338727d | 221 | struct ofpbuf put_actions; /* Actions 'put' in the fastpath. */ |
cc377352 | 222 | |
dcc2c6cd JR |
223 | struct dpif_ipfix *ipfix; /* IPFIX pointer or NULL. */ |
224 | struct dpif_sflow *sflow; /* SFlow pointer or NULL. */ | |
a0bab870 | 225 | |
23597df0 JS |
226 | struct udpif_key *ukey; /* Revalidator flow cache. */ |
227 | bool ukey_persists; /* Set true to keep 'ukey' beyond the | |
228 | lifetime of this upcall. */ | |
229 | ||
230 | uint64_t dump_seq; /* udpif->dump_seq at translation time. */ | |
231 | uint64_t reval_seq; /* udpif->reval_seq at translation time. */ | |
232 | ||
cc377352 EJ |
233 | /* Not used by the upcall callback interface. */ |
234 | const struct nlattr *key; /* Datapath flow key. */ | |
235 | size_t key_len; /* Datapath flow key length. */ | |
8b7ea2d4 | 236 | const struct nlattr *out_tun_key; /* Datapath output tunnel key. */ |
1520ef4f BP |
237 | |
238 | uint64_t odp_actions_stub[1024 / 8]; /* Stub for odp_actions. */ | |
10e57640 EJ |
239 | }; |
240 | ||
54ebeff4 JS |
241 | /* Ukeys must transition through these states using transition_ukey(). */ |
242 | enum ukey_state { | |
243 | UKEY_CREATED = 0, | |
244 | UKEY_VISIBLE, /* Ukey is in umap, datapath flow install is queued. */ | |
245 | UKEY_OPERATIONAL, /* Ukey is in umap, datapath flow is installed. */ | |
246 | UKEY_EVICTING, /* Ukey is in umap, datapath flow delete is queued. */ | |
247 | UKEY_EVICTED, /* Ukey is in umap, datapath flow is deleted. */ | |
248 | UKEY_DELETED, /* Ukey removed from umap, ukey free is deferred. */ | |
249 | }; | |
250 | #define N_UKEY_STATES (UKEY_DELETED + 1) | |
251 | ||
e79a6c83 EJ |
252 | /* 'udpif_key's are responsible for tracking the little bit of state udpif |
253 | * needs to do flow expiration which can't be pulled directly from the | |
23597df0 JS |
254 | * datapath. They may be created by any handler or revalidator thread at any |
255 | * time, and read by any revalidator during the dump phase. They are however | |
256 | * each owned by a single revalidator which takes care of destroying them | |
257 | * during the garbage-collection phase. | |
7d170098 | 258 | * |
b8d3daeb JS |
259 | * The mutex within the ukey protects some members of the ukey. The ukey |
260 | * itself is protected by RCU and is held within a umap in the parent udpif. | |
261 | * Adding or removing a ukey from a umap is only safe when holding the | |
262 | * corresponding umap lock. */ | |
e79a6c83 | 263 | struct udpif_key { |
9fce0584 | 264 | struct cmap_node cmap_node; /* In parent revalidator 'ukeys' map. */ |
e79a6c83 | 265 | |
7d170098 EJ |
266 | /* These elements are read only once created, and therefore aren't |
267 | * protected by a mutex. */ | |
268 | const struct nlattr *key; /* Datapath flow key. */ | |
e79a6c83 | 269 | size_t key_len; /* Length of 'key'. */ |
bc2df54d JS |
270 | const struct nlattr *mask; /* Datapath flow mask. */ |
271 | size_t mask_len; /* Length of 'mask'. */ | |
7af12bd7 | 272 | ovs_u128 ufid; /* Unique flow identifier. */ |
70e5ed6f | 273 | bool ufid_present; /* True if 'ufid' is in datapath. */ |
9fce0584 | 274 | uint32_t hash; /* Pre-computed hash for 'key'. */ |
bd5131ba | 275 | unsigned pmd_id; /* Datapath poll mode driver id. */ |
e79a6c83 | 276 | |
7d170098 EJ |
277 | struct ovs_mutex mutex; /* Guards the following. */ |
278 | struct dpif_flow_stats stats OVS_GUARDED; /* Last known stats.*/ | |
279 | long long int created OVS_GUARDED; /* Estimate of creation time. */ | |
efa08531 | 280 | uint64_t dump_seq OVS_GUARDED; /* Tracks udpif->dump_seq. */ |
23597df0 | 281 | uint64_t reval_seq OVS_GUARDED; /* Tracks udpif->reval_seq. */ |
54ebeff4 JS |
282 | enum ukey_state state OVS_GUARDED; /* Tracks ukey lifetime. */ |
283 | ||
b7637498 EJ |
284 | /* Datapath flow actions as nlattrs. Protected by RCU. Read with |
285 | * ukey_get_actions(), and write with ukey_set_actions(). */ | |
286 | OVSRCU_TYPE(struct ofpbuf *) actions; | |
7d170098 EJ |
287 | |
288 | struct xlate_cache *xcache OVS_GUARDED; /* Cache for xlate entries that | |
289 | * are affected by this ukey. | |
290 | * Used for stats and learning.*/ | |
02334943 | 291 | union { |
bc2df54d JS |
292 | struct odputil_keybuf buf; |
293 | struct nlattr nla; | |
294 | } keybuf, maskbuf; | |
e672ff9b | 295 | |
fbf5d6ec JR |
296 | uint32_t key_recirc_id; /* Non-zero if reference is held by the ukey. */ |
297 | struct recirc_refs recircs; /* Action recirc IDs with references held. */ | |
e79a6c83 EJ |
298 | }; |
299 | ||
6dad4d44 JS |
300 | /* Datapath operation with optional ukey attached. */ |
301 | struct ukey_op { | |
302 | struct udpif_key *ukey; | |
303 | struct dpif_flow_stats stats; /* Stats for 'op'. */ | |
304 | struct dpif_op dop; /* Flow operation. */ | |
305 | }; | |
306 | ||
e1ec7dd4 | 307 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
55951e15 | 308 | static struct ovs_list all_udpifs = OVS_LIST_INITIALIZER(&all_udpifs); |
e1ec7dd4 | 309 | |
cc377352 EJ |
310 | static size_t recv_upcalls(struct handler *); |
311 | static int process_upcall(struct udpif *, struct upcall *, | |
49a73e0c | 312 | struct ofpbuf *odp_actions, struct flow_wildcards *); |
6b31e073 | 313 | static void handle_upcalls(struct udpif *, struct upcall *, size_t n_upcalls); |
1f867548 AW |
314 | static void udpif_stop_threads(struct udpif *); |
315 | static void udpif_start_threads(struct udpif *, size_t n_handlers, | |
316 | size_t n_revalidators); | |
dba82d38 AW |
317 | static void udpif_pause_revalidators(struct udpif *); |
318 | static void udpif_resume_revalidators(struct udpif *); | |
10e57640 | 319 | static void *udpif_upcall_handler(void *); |
e79a6c83 | 320 | static void *udpif_revalidator(void *); |
0e2a9f6f | 321 | static unsigned long udpif_get_n_flows(struct udpif *); |
7d170098 | 322 | static void revalidate(struct revalidator *); |
dba82d38 | 323 | static void revalidator_pause(struct revalidator *); |
e79a6c83 | 324 | static void revalidator_sweep(struct revalidator *); |
e96a5c24 | 325 | static void revalidator_purge(struct revalidator *); |
e22d52ee EJ |
326 | static void upcall_unixctl_show(struct unixctl_conn *conn, int argc, |
327 | const char *argv[], void *aux); | |
e79a6c83 EJ |
328 | static void upcall_unixctl_disable_megaflows(struct unixctl_conn *, int argc, |
329 | const char *argv[], void *aux); | |
330 | static void upcall_unixctl_enable_megaflows(struct unixctl_conn *, int argc, | |
331 | const char *argv[], void *aux); | |
64bb477f JS |
332 | static void upcall_unixctl_disable_ufid(struct unixctl_conn *, int argc, |
333 | const char *argv[], void *aux); | |
334 | static void upcall_unixctl_enable_ufid(struct unixctl_conn *, int argc, | |
335 | const char *argv[], void *aux); | |
94b8c324 JS |
336 | static void upcall_unixctl_set_flow_limit(struct unixctl_conn *conn, int argc, |
337 | const char *argv[], void *aux); | |
27f57736 JS |
338 | static void upcall_unixctl_dump_wait(struct unixctl_conn *conn, int argc, |
339 | const char *argv[], void *aux); | |
98bb4286 JS |
340 | static void upcall_unixctl_purge(struct unixctl_conn *conn, int argc, |
341 | const char *argv[], void *aux); | |
7d170098 | 342 | |
49a73e0c BP |
343 | static struct udpif_key *ukey_create_from_upcall(struct upcall *, |
344 | struct flow_wildcards *); | |
64bb477f JS |
345 | static int ukey_create_from_dpif_flow(const struct udpif *, |
346 | const struct dpif_flow *, | |
347 | struct udpif_key **); | |
b7637498 EJ |
348 | static void ukey_get_actions(struct udpif_key *, const struct nlattr **actions, |
349 | size_t *size); | |
54ebeff4 JS |
350 | static bool ukey_install__(struct udpif *, struct udpif_key *ukey) |
351 | OVS_TRY_LOCK(true, ukey->mutex); | |
23597df0 | 352 | static bool ukey_install(struct udpif *udpif, struct udpif_key *ukey); |
54ebeff4 JS |
353 | static void transition_ukey(struct udpif_key *ukey, enum ukey_state dst) |
354 | OVS_REQUIRES(ukey->mutex); | |
7af12bd7 | 355 | static struct udpif_key *ukey_lookup(struct udpif *udpif, |
5f2ccb1c IM |
356 | const ovs_u128 *ufid, |
357 | const unsigned pmd_id); | |
23597df0 | 358 | static int ukey_acquire(struct udpif *, const struct dpif_flow *, |
64bb477f | 359 | struct udpif_key **result, int *error); |
9fce0584 | 360 | static void ukey_delete__(struct udpif_key *); |
b8d3daeb | 361 | static void ukey_delete(struct umap *, struct udpif_key *); |
cc377352 EJ |
362 | static enum upcall_type classify_upcall(enum dpif_upcall_type type, |
363 | const struct nlattr *userdata); | |
364 | ||
f673dcd8 JS |
365 | static void put_op_init(struct ukey_op *op, struct udpif_key *ukey, |
366 | enum dpif_flow_put_flags flags); | |
54ebeff4 JS |
367 | static void delete_op_init(struct udpif *udpif, struct ukey_op *op, |
368 | struct udpif_key *ukey); | |
f673dcd8 | 369 | |
cc377352 | 370 | static int upcall_receive(struct upcall *, const struct dpif_backer *, |
cf62fa4c | 371 | const struct dp_packet *packet, enum dpif_upcall_type, |
7af12bd7 | 372 | const struct nlattr *userdata, const struct flow *, |
27130224 | 373 | const unsigned int mru, |
bd5131ba | 374 | const ovs_u128 *ufid, const unsigned pmd_id); |
cc377352 | 375 | static void upcall_uninit(struct upcall *); |
e79a6c83 | 376 | |
623540e4 | 377 | static upcall_callback upcall_cb; |
e4e74c3a | 378 | static dp_purge_callback dp_purge_cb; |
623540e4 | 379 | |
e79a6c83 | 380 | static atomic_bool enable_megaflows = ATOMIC_VAR_INIT(true); |
70f07728 | 381 | static atomic_bool enable_ufid = ATOMIC_VAR_INIT(true); |
e1ec7dd4 | 382 | |
0fc1f5c0 HH |
383 | void |
384 | udpif_init(void) | |
e1ec7dd4 | 385 | { |
e22d52ee | 386 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
e22d52ee EJ |
387 | if (ovsthread_once_start(&once)) { |
388 | unixctl_command_register("upcall/show", "", 0, 0, upcall_unixctl_show, | |
389 | NULL); | |
e79a6c83 EJ |
390 | unixctl_command_register("upcall/disable-megaflows", "", 0, 0, |
391 | upcall_unixctl_disable_megaflows, NULL); | |
392 | unixctl_command_register("upcall/enable-megaflows", "", 0, 0, | |
393 | upcall_unixctl_enable_megaflows, NULL); | |
64bb477f JS |
394 | unixctl_command_register("upcall/disable-ufid", "", 0, 0, |
395 | upcall_unixctl_disable_ufid, NULL); | |
396 | unixctl_command_register("upcall/enable-ufid", "", 0, 0, | |
397 | upcall_unixctl_enable_ufid, NULL); | |
31b418cb | 398 | unixctl_command_register("upcall/set-flow-limit", "flow-limit-number", |
399 | 1, 1, upcall_unixctl_set_flow_limit, NULL); | |
27f57736 JS |
400 | unixctl_command_register("revalidator/wait", "", 0, 0, |
401 | upcall_unixctl_dump_wait, NULL); | |
98bb4286 JS |
402 | unixctl_command_register("revalidator/purge", "", 0, 0, |
403 | upcall_unixctl_purge, NULL); | |
e22d52ee EJ |
404 | ovsthread_once_done(&once); |
405 | } | |
0fc1f5c0 HH |
406 | } |
407 | ||
408 | struct udpif * | |
409 | udpif_create(struct dpif_backer *backer, struct dpif *dpif) | |
410 | { | |
411 | struct udpif *udpif = xzalloc(sizeof *udpif); | |
e22d52ee | 412 | |
e1ec7dd4 EJ |
413 | udpif->dpif = dpif; |
414 | udpif->backer = backer; | |
e79a6c83 | 415 | atomic_init(&udpif->flow_limit, MIN(ofproto_flow_limit, 10000)); |
d7285d74 | 416 | udpif->reval_seq = seq_create(); |
e79a6c83 | 417 | udpif->dump_seq = seq_create(); |
e1ec7dd4 | 418 | latch_init(&udpif->exit_latch); |
dba82d38 | 419 | latch_init(&udpif->pause_latch); |
417e7e66 | 420 | ovs_list_push_back(&all_udpifs, &udpif->list_node); |
64bb477f | 421 | atomic_init(&udpif->enable_ufid, false); |
64ca9472 JS |
422 | atomic_init(&udpif->n_flows, 0); |
423 | atomic_init(&udpif->n_flows_timestamp, LLONG_MIN); | |
424 | ovs_mutex_init(&udpif->n_flows_mutex); | |
b8d3daeb JS |
425 | udpif->ukeys = xmalloc(N_UMAPS * sizeof *udpif->ukeys); |
426 | for (int i = 0; i < N_UMAPS; i++) { | |
427 | cmap_init(&udpif->ukeys[i].cmap); | |
428 | ovs_mutex_init(&udpif->ukeys[i].mutex); | |
429 | } | |
e1ec7dd4 | 430 | |
623540e4 | 431 | dpif_register_upcall_cb(dpif, upcall_cb, udpif); |
e4e74c3a | 432 | dpif_register_dp_purge_cb(dpif, dp_purge_cb, udpif); |
6b31e073 | 433 | |
e1ec7dd4 EJ |
434 | return udpif; |
435 | } | |
436 | ||
27f57736 JS |
437 | void |
438 | udpif_run(struct udpif *udpif) | |
439 | { | |
440 | if (udpif->conns && udpif->conn_seq != seq_read(udpif->dump_seq)) { | |
441 | int i; | |
442 | ||
443 | for (i = 0; i < udpif->n_conns; i++) { | |
444 | unixctl_command_reply(udpif->conns[i], NULL); | |
445 | } | |
446 | free(udpif->conns); | |
447 | udpif->conns = NULL; | |
448 | udpif->n_conns = 0; | |
449 | } | |
450 | } | |
451 | ||
e1ec7dd4 EJ |
452 | void |
453 | udpif_destroy(struct udpif *udpif) | |
454 | { | |
1f867548 | 455 | udpif_stop_threads(udpif); |
e1ec7dd4 | 456 | |
b803e6ac JS |
457 | dpif_register_dp_purge_cb(udpif->dpif, NULL, udpif); |
458 | dpif_register_upcall_cb(udpif->dpif, NULL, udpif); | |
459 | ||
b8d3daeb JS |
460 | for (int i = 0; i < N_UMAPS; i++) { |
461 | cmap_destroy(&udpif->ukeys[i].cmap); | |
462 | ovs_mutex_destroy(&udpif->ukeys[i].mutex); | |
463 | } | |
464 | free(udpif->ukeys); | |
465 | udpif->ukeys = NULL; | |
466 | ||
417e7e66 | 467 | ovs_list_remove(&udpif->list_node); |
e1ec7dd4 | 468 | latch_destroy(&udpif->exit_latch); |
dba82d38 | 469 | latch_destroy(&udpif->pause_latch); |
d7285d74 | 470 | seq_destroy(udpif->reval_seq); |
e79a6c83 | 471 | seq_destroy(udpif->dump_seq); |
64ca9472 | 472 | ovs_mutex_destroy(&udpif->n_flows_mutex); |
e1ec7dd4 EJ |
473 | free(udpif); |
474 | } | |
475 | ||
1f867548 AW |
476 | /* Stops the handler and revalidator threads, must be enclosed in |
477 | * ovsrcu quiescent state unless when destroying udpif. */ | |
478 | static void | |
479 | udpif_stop_threads(struct udpif *udpif) | |
e1ec7dd4 | 480 | { |
3aadc5bb | 481 | if (udpif && (udpif->n_handlers != 0 || udpif->n_revalidators != 0)) { |
e1ec7dd4 EJ |
482 | size_t i; |
483 | ||
484 | latch_set(&udpif->exit_latch); | |
485 | ||
e1ec7dd4 EJ |
486 | for (i = 0; i < udpif->n_handlers; i++) { |
487 | struct handler *handler = &udpif->handlers[i]; | |
488 | ||
e79a6c83 EJ |
489 | xpthread_join(handler->thread, NULL); |
490 | } | |
491 | ||
492 | for (i = 0; i < udpif->n_revalidators; i++) { | |
7d170098 | 493 | xpthread_join(udpif->revalidators[i].thread, NULL); |
e1ec7dd4 EJ |
494 | } |
495 | ||
6b31e073 RW |
496 | dpif_disable_upcall(udpif->dpif); |
497 | ||
e79a6c83 EJ |
498 | for (i = 0; i < udpif->n_revalidators; i++) { |
499 | struct revalidator *revalidator = &udpif->revalidators[i]; | |
e79a6c83 | 500 | |
e96a5c24 JS |
501 | /* Delete ukeys, and delete all flows from the datapath to prevent |
502 | * double-counting stats. */ | |
503 | revalidator_purge(revalidator); | |
e79a6c83 EJ |
504 | } |
505 | ||
e1ec7dd4 EJ |
506 | latch_poll(&udpif->exit_latch); |
507 | ||
d8043da7 | 508 | ovs_barrier_destroy(&udpif->reval_barrier); |
dba82d38 | 509 | ovs_barrier_destroy(&udpif->pause_barrier); |
7d170098 | 510 | |
e79a6c83 EJ |
511 | free(udpif->revalidators); |
512 | udpif->revalidators = NULL; | |
513 | udpif->n_revalidators = 0; | |
514 | ||
e1ec7dd4 EJ |
515 | free(udpif->handlers); |
516 | udpif->handlers = NULL; | |
517 | udpif->n_handlers = 0; | |
518 | } | |
1f867548 | 519 | } |
e1ec7dd4 | 520 | |
1f867548 AW |
521 | /* Starts the handler and revalidator threads, must be enclosed in |
522 | * ovsrcu quiescent state. */ | |
523 | static void | |
524 | udpif_start_threads(struct udpif *udpif, size_t n_handlers, | |
525 | size_t n_revalidators) | |
526 | { | |
6f12bda3 | 527 | if (udpif && n_handlers && n_revalidators) { |
e1ec7dd4 | 528 | size_t i; |
8e1ffd75 | 529 | bool enable_ufid; |
e1ec7dd4 EJ |
530 | |
531 | udpif->n_handlers = n_handlers; | |
e79a6c83 EJ |
532 | udpif->n_revalidators = n_revalidators; |
533 | ||
e1ec7dd4 EJ |
534 | udpif->handlers = xzalloc(udpif->n_handlers * sizeof *udpif->handlers); |
535 | for (i = 0; i < udpif->n_handlers; i++) { | |
536 | struct handler *handler = &udpif->handlers[i]; | |
537 | ||
538 | handler->udpif = udpif; | |
9a159f74 | 539 | handler->handler_id = i; |
8ba0a522 BP |
540 | handler->thread = ovs_thread_create( |
541 | "handler", udpif_upcall_handler, handler); | |
e1ec7dd4 | 542 | } |
e1ec7dd4 | 543 | |
8e1ffd75 JS |
544 | enable_ufid = ofproto_dpif_get_enable_ufid(udpif->backer); |
545 | atomic_init(&udpif->enable_ufid, enable_ufid); | |
6b31e073 RW |
546 | dpif_enable_upcall(udpif->dpif); |
547 | ||
d8043da7 | 548 | ovs_barrier_init(&udpif->reval_barrier, udpif->n_revalidators); |
dba82d38 | 549 | ovs_barrier_init(&udpif->pause_barrier, udpif->n_revalidators + 1); |
7d170098 | 550 | udpif->reval_exit = false; |
dba82d38 | 551 | udpif->pause = false; |
e79a6c83 EJ |
552 | udpif->revalidators = xzalloc(udpif->n_revalidators |
553 | * sizeof *udpif->revalidators); | |
554 | for (i = 0; i < udpif->n_revalidators; i++) { | |
555 | struct revalidator *revalidator = &udpif->revalidators[i]; | |
556 | ||
557 | revalidator->udpif = udpif; | |
8ba0a522 BP |
558 | revalidator->thread = ovs_thread_create( |
559 | "revalidator", udpif_revalidator, revalidator); | |
e79a6c83 | 560 | } |
e1ec7dd4 | 561 | } |
1f867548 | 562 | } |
0f2ea848 | 563 | |
dba82d38 AW |
564 | /* Pauses all revalidators. Should only be called by the main thread. |
565 | * When function returns, all revalidators are paused and will proceed | |
566 | * only after udpif_resume_revalidators() is called. */ | |
567 | static void | |
568 | udpif_pause_revalidators(struct udpif *udpif) | |
569 | { | |
d997f23f ZK |
570 | if (ofproto_dpif_backer_enabled(udpif->backer)) { |
571 | latch_set(&udpif->pause_latch); | |
572 | ovs_barrier_block(&udpif->pause_barrier); | |
573 | } | |
dba82d38 AW |
574 | } |
575 | ||
576 | /* Resumes the pausing of revalidators. Should only be called by the | |
577 | * main thread. */ | |
578 | static void | |
579 | udpif_resume_revalidators(struct udpif *udpif) | |
580 | { | |
d997f23f ZK |
581 | if (ofproto_dpif_backer_enabled(udpif->backer)) { |
582 | latch_poll(&udpif->pause_latch); | |
583 | ovs_barrier_block(&udpif->pause_barrier); | |
584 | } | |
dba82d38 AW |
585 | } |
586 | ||
1f867548 AW |
587 | /* Tells 'udpif' how many threads it should use to handle upcalls. |
588 | * 'n_handlers' and 'n_revalidators' can never be zero. 'udpif''s | |
589 | * datapath handle must have packet reception enabled before starting | |
590 | * threads. */ | |
591 | void | |
592 | udpif_set_threads(struct udpif *udpif, size_t n_handlers, | |
593 | size_t n_revalidators) | |
594 | { | |
3aadc5bb | 595 | ovs_assert(udpif); |
1f867548 AW |
596 | ovs_assert(n_handlers && n_revalidators); |
597 | ||
598 | ovsrcu_quiesce_start(); | |
3aadc5bb AW |
599 | if (udpif->n_handlers != n_handlers |
600 | || udpif->n_revalidators != n_revalidators) { | |
601 | udpif_stop_threads(udpif); | |
602 | } | |
1f867548 | 603 | |
3aadc5bb | 604 | if (!udpif->handlers && !udpif->revalidators) { |
380fffec AW |
605 | int error; |
606 | ||
607 | error = dpif_handlers_set(udpif->dpif, n_handlers); | |
608 | if (error) { | |
609 | VLOG_ERR("failed to configure handlers in dpif %s: %s", | |
610 | dpif_name(udpif->dpif), ovs_strerror(error)); | |
611 | return; | |
612 | } | |
613 | ||
3aadc5bb AW |
614 | udpif_start_threads(udpif, n_handlers, n_revalidators); |
615 | } | |
0f2ea848 | 616 | ovsrcu_quiesce_end(); |
e1ec7dd4 EJ |
617 | } |
618 | ||
3f142f59 BP |
619 | /* Waits for all ongoing upcall translations to complete. This ensures that |
620 | * there are no transient references to any removed ofprotos (or other | |
621 | * objects). In particular, this should be called after an ofproto is removed | |
622 | * (e.g. via xlate_remove_ofproto()) but before it is destroyed. */ | |
623 | void | |
624 | udpif_synchronize(struct udpif *udpif) | |
625 | { | |
626 | /* This is stronger than necessary. It would be sufficient to ensure | |
627 | * (somehow) that each handler and revalidator thread had passed through | |
628 | * its main loop once. */ | |
629 | size_t n_handlers = udpif->n_handlers; | |
630 | size_t n_revalidators = udpif->n_revalidators; | |
1f867548 AW |
631 | |
632 | ovsrcu_quiesce_start(); | |
633 | udpif_stop_threads(udpif); | |
634 | udpif_start_threads(udpif, n_handlers, n_revalidators); | |
635 | ovsrcu_quiesce_end(); | |
3f142f59 BP |
636 | } |
637 | ||
e1ec7dd4 EJ |
638 | /* Notifies 'udpif' that something changed which may render previous |
639 | * xlate_actions() results invalid. */ | |
640 | void | |
641 | udpif_revalidate(struct udpif *udpif) | |
642 | { | |
d7285d74 | 643 | seq_change(udpif->reval_seq); |
e79a6c83 | 644 | } |
05067881 | 645 | |
e79a6c83 EJ |
646 | /* Returns a seq which increments every time 'udpif' pulls stats from the |
647 | * datapath. Callers can use this to get a sense of when might be a good time | |
648 | * to do periodic work which relies on relatively up to date statistics. */ | |
649 | struct seq * | |
650 | udpif_dump_seq(struct udpif *udpif) | |
651 | { | |
652 | return udpif->dump_seq; | |
e1ec7dd4 EJ |
653 | } |
654 | ||
1c030aa5 EJ |
655 | void |
656 | udpif_get_memory_usage(struct udpif *udpif, struct simap *usage) | |
657 | { | |
658 | size_t i; | |
659 | ||
1c030aa5 | 660 | simap_increase(usage, "handlers", udpif->n_handlers); |
e79a6c83 EJ |
661 | |
662 | simap_increase(usage, "revalidators", udpif->n_revalidators); | |
b8d3daeb | 663 | for (i = 0; i < N_UMAPS; i++) { |
9fce0584 | 664 | simap_increase(usage, "udpif keys", cmap_count(&udpif->ukeys[i].cmap)); |
e79a6c83 | 665 | } |
1c030aa5 EJ |
666 | } |
667 | ||
1b5b5071 | 668 | /* Remove flows from a single datapath. */ |
e79a6c83 | 669 | void |
1b5b5071 AZ |
670 | udpif_flush(struct udpif *udpif) |
671 | { | |
672 | size_t n_handlers, n_revalidators; | |
673 | ||
674 | n_handlers = udpif->n_handlers; | |
675 | n_revalidators = udpif->n_revalidators; | |
676 | ||
1f867548 AW |
677 | ovsrcu_quiesce_start(); |
678 | ||
679 | udpif_stop_threads(udpif); | |
1b5b5071 | 680 | dpif_flow_flush(udpif->dpif); |
1f867548 AW |
681 | udpif_start_threads(udpif, n_handlers, n_revalidators); |
682 | ||
683 | ovsrcu_quiesce_end(); | |
1b5b5071 AZ |
684 | } |
685 | ||
686 | /* Removes all flows from all datapaths. */ | |
687 | static void | |
688 | udpif_flush_all_datapaths(void) | |
e79a6c83 EJ |
689 | { |
690 | struct udpif *udpif; | |
691 | ||
692 | LIST_FOR_EACH (udpif, list_node, &all_udpifs) { | |
1b5b5071 | 693 | udpif_flush(udpif); |
e79a6c83 EJ |
694 | } |
695 | } | |
1b5b5071 | 696 | |
70f07728 JS |
697 | static bool |
698 | udpif_use_ufid(struct udpif *udpif) | |
699 | { | |
700 | bool enable; | |
701 | ||
702 | atomic_read_relaxed(&enable_ufid, &enable); | |
703 | return enable && ofproto_dpif_get_enable_ufid(udpif->backer); | |
704 | } | |
705 | ||
e79a6c83 | 706 | \f |
0e2a9f6f | 707 | static unsigned long |
64ca9472 | 708 | udpif_get_n_flows(struct udpif *udpif) |
e1ec7dd4 | 709 | { |
64ca9472 | 710 | long long int time, now; |
0e2a9f6f | 711 | unsigned long flow_count; |
64ca9472 JS |
712 | |
713 | now = time_msec(); | |
b482e960 | 714 | atomic_read_relaxed(&udpif->n_flows_timestamp, &time); |
64ca9472 JS |
715 | if (time < now - 100 && !ovs_mutex_trylock(&udpif->n_flows_mutex)) { |
716 | struct dpif_dp_stats stats; | |
717 | ||
b482e960 | 718 | atomic_store_relaxed(&udpif->n_flows_timestamp, now); |
64ca9472 JS |
719 | dpif_get_dp_stats(udpif->dpif, &stats); |
720 | flow_count = stats.n_flows; | |
b482e960 | 721 | atomic_store_relaxed(&udpif->n_flows, flow_count); |
64ca9472 JS |
722 | ovs_mutex_unlock(&udpif->n_flows_mutex); |
723 | } else { | |
b482e960 | 724 | atomic_read_relaxed(&udpif->n_flows, &flow_count); |
64ca9472 JS |
725 | } |
726 | return flow_count; | |
e79a6c83 | 727 | } |
e1ec7dd4 | 728 | |
a0bab870 | 729 | /* The upcall handler thread tries to read a batch of UPCALL_MAX_BATCH |
9a159f74 AW |
730 | * upcalls from dpif, processes the batch and installs corresponding flows |
731 | * in dpif. */ | |
e1ec7dd4 | 732 | static void * |
10e57640 | 733 | udpif_upcall_handler(void *arg) |
e1ec7dd4 | 734 | { |
e1ec7dd4 | 735 | struct handler *handler = arg; |
9a159f74 | 736 | struct udpif *udpif = handler->udpif; |
e1ec7dd4 | 737 | |
61057e88 | 738 | while (!latch_is_set(&handler->udpif->exit_latch)) { |
23597df0 JS |
739 | if (recv_upcalls(handler)) { |
740 | poll_immediate_wake(); | |
741 | } else { | |
9a159f74 AW |
742 | dpif_recv_wait(udpif->dpif, handler->handler_id); |
743 | latch_wait(&udpif->exit_latch); | |
e1ec7dd4 | 744 | } |
23597df0 | 745 | poll_block(); |
e1ec7dd4 | 746 | } |
61057e88 BP |
747 | |
748 | return NULL; | |
e1ec7dd4 | 749 | } |
e79a6c83 | 750 | |
cc377352 EJ |
751 | static size_t |
752 | recv_upcalls(struct handler *handler) | |
753 | { | |
754 | struct udpif *udpif = handler->udpif; | |
755 | uint64_t recv_stubs[UPCALL_MAX_BATCH][512 / 8]; | |
756 | struct ofpbuf recv_bufs[UPCALL_MAX_BATCH]; | |
a6f4ad08 | 757 | struct dpif_upcall dupcalls[UPCALL_MAX_BATCH]; |
cc377352 | 758 | struct upcall upcalls[UPCALL_MAX_BATCH]; |
ff601a08 | 759 | struct flow flows[UPCALL_MAX_BATCH]; |
cc377352 EJ |
760 | size_t n_upcalls, i; |
761 | ||
762 | n_upcalls = 0; | |
763 | while (n_upcalls < UPCALL_MAX_BATCH) { | |
764 | struct ofpbuf *recv_buf = &recv_bufs[n_upcalls]; | |
a6f4ad08 | 765 | struct dpif_upcall *dupcall = &dupcalls[n_upcalls]; |
cc377352 | 766 | struct upcall *upcall = &upcalls[n_upcalls]; |
ff601a08 | 767 | struct flow *flow = &flows[n_upcalls]; |
27130224 | 768 | unsigned int mru; |
cc377352 EJ |
769 | int error; |
770 | ||
7174c145 | 771 | ofpbuf_use_stub(recv_buf, recv_stubs[n_upcalls], |
cc377352 | 772 | sizeof recv_stubs[n_upcalls]); |
a6f4ad08 | 773 | if (dpif_recv(udpif->dpif, handler->handler_id, dupcall, recv_buf)) { |
cc377352 EJ |
774 | ofpbuf_uninit(recv_buf); |
775 | break; | |
776 | } | |
777 | ||
ff601a08 | 778 | if (odp_flow_key_to_flow(dupcall->key, dupcall->key_len, flow) |
cc377352 EJ |
779 | == ODP_FIT_ERROR) { |
780 | goto free_dupcall; | |
781 | } | |
782 | ||
27130224 AZ |
783 | if (dupcall->mru) { |
784 | mru = nl_attr_get_u16(dupcall->mru); | |
785 | } else { | |
786 | mru = 0; | |
787 | } | |
788 | ||
a6f4ad08 | 789 | error = upcall_receive(upcall, udpif->backer, &dupcall->packet, |
27130224 | 790 | dupcall->type, dupcall->userdata, flow, mru, |
1c1e46ed | 791 | &dupcall->ufid, PMD_ID_NULL); |
cc377352 EJ |
792 | if (error) { |
793 | if (error == ENODEV) { | |
794 | /* Received packet on datapath port for which we couldn't | |
795 | * associate an ofproto. This can happen if a port is removed | |
796 | * while traffic is being received. Print a rate-limited | |
797 | * message in case it happens frequently. */ | |
a6f4ad08 | 798 | dpif_flow_put(udpif->dpif, DPIF_FP_CREATE, dupcall->key, |
70e5ed6f | 799 | dupcall->key_len, NULL, 0, NULL, 0, |
1c1e46ed | 800 | &dupcall->ufid, PMD_ID_NULL, NULL); |
cc377352 | 801 | VLOG_INFO_RL(&rl, "received packet on unassociated datapath " |
ff601a08 | 802 | "port %"PRIu32, flow->in_port.odp_port); |
cc377352 EJ |
803 | } |
804 | goto free_dupcall; | |
805 | } | |
806 | ||
a6f4ad08 AW |
807 | upcall->key = dupcall->key; |
808 | upcall->key_len = dupcall->key_len; | |
7af12bd7 | 809 | upcall->ufid = &dupcall->ufid; |
cc377352 | 810 | |
8b7ea2d4 | 811 | upcall->out_tun_key = dupcall->out_tun_key; |
0a1017cb | 812 | upcall->actions = dupcall->actions; |
8b7ea2d4 | 813 | |
cf62fa4c PS |
814 | pkt_metadata_from_flow(&dupcall->packet.md, flow); |
815 | flow_extract(&dupcall->packet, flow); | |
cc377352 | 816 | |
1520ef4f BP |
817 | error = process_upcall(udpif, upcall, |
818 | &upcall->odp_actions, &upcall->wc); | |
cc377352 EJ |
819 | if (error) { |
820 | goto cleanup; | |
821 | } | |
822 | ||
823 | n_upcalls++; | |
824 | continue; | |
825 | ||
826 | cleanup: | |
827 | upcall_uninit(upcall); | |
828 | free_dupcall: | |
cf62fa4c | 829 | dp_packet_uninit(&dupcall->packet); |
cc377352 EJ |
830 | ofpbuf_uninit(recv_buf); |
831 | } | |
832 | ||
833 | if (n_upcalls) { | |
834 | handle_upcalls(handler->udpif, upcalls, n_upcalls); | |
835 | for (i = 0; i < n_upcalls; i++) { | |
cf62fa4c | 836 | dp_packet_uninit(&dupcalls[i].packet); |
cc377352 EJ |
837 | ofpbuf_uninit(&recv_bufs[i]); |
838 | upcall_uninit(&upcalls[i]); | |
839 | } | |
840 | } | |
841 | ||
842 | return n_upcalls; | |
843 | } | |
844 | ||
e79a6c83 EJ |
845 | static void * |
846 | udpif_revalidator(void *arg) | |
e1ec7dd4 | 847 | { |
7d170098 | 848 | /* Used by all revalidators. */ |
e79a6c83 | 849 | struct revalidator *revalidator = arg; |
7d170098 EJ |
850 | struct udpif *udpif = revalidator->udpif; |
851 | bool leader = revalidator == &udpif->revalidators[0]; | |
852 | ||
853 | /* Used only by the leader. */ | |
854 | long long int start_time = 0; | |
855 | uint64_t last_reval_seq = 0; | |
7d170098 | 856 | size_t n_flows = 0; |
e1ec7dd4 | 857 | |
8ba0a522 | 858 | revalidator->id = ovsthread_id_self(); |
e79a6c83 | 859 | for (;;) { |
7d170098 EJ |
860 | if (leader) { |
861 | uint64_t reval_seq; | |
e79a6c83 | 862 | |
e672ff9b JR |
863 | recirc_run(); /* Recirculation cleanup. */ |
864 | ||
7d170098 | 865 | reval_seq = seq_read(udpif->reval_seq); |
7d170098 | 866 | last_reval_seq = reval_seq; |
e79a6c83 | 867 | |
7d170098 EJ |
868 | n_flows = udpif_get_n_flows(udpif); |
869 | udpif->max_n_flows = MAX(n_flows, udpif->max_n_flows); | |
870 | udpif->avg_n_flows = (udpif->avg_n_flows + n_flows) / 2; | |
871 | ||
dba82d38 AW |
872 | /* Only the leader checks the pause latch to prevent a race where |
873 | * some threads think it's false and proceed to block on | |
874 | * reval_barrier and others think it's true and block indefinitely | |
875 | * on the pause_barrier */ | |
876 | udpif->pause = latch_is_set(&udpif->pause_latch); | |
877 | ||
7d170098 EJ |
878 | /* Only the leader checks the exit latch to prevent a race where |
879 | * some threads think it's true and exit and others think it's | |
880 | * false and block indefinitely on the reval_barrier */ | |
881 | udpif->reval_exit = latch_is_set(&udpif->exit_latch); | |
882 | ||
883 | start_time = time_msec(); | |
884 | if (!udpif->reval_exit) { | |
64bb477f JS |
885 | bool terse_dump; |
886 | ||
70f07728 | 887 | terse_dump = udpif_use_ufid(udpif); |
64bb477f | 888 | udpif->dump = dpif_flow_dump_create(udpif->dpif, terse_dump); |
e79a6c83 EJ |
889 | } |
890 | } | |
891 | ||
7d170098 | 892 | /* Wait for the leader to start the flow dump. */ |
d8043da7 | 893 | ovs_barrier_block(&udpif->reval_barrier); |
dba82d38 AW |
894 | if (udpif->pause) { |
895 | revalidator_pause(revalidator); | |
896 | } | |
897 | ||
7d170098 EJ |
898 | if (udpif->reval_exit) { |
899 | break; | |
e79a6c83 | 900 | } |
7d170098 EJ |
901 | revalidate(revalidator); |
902 | ||
903 | /* Wait for all flows to have been dumped before we garbage collect. */ | |
d8043da7 | 904 | ovs_barrier_block(&udpif->reval_barrier); |
7d170098 EJ |
905 | revalidator_sweep(revalidator); |
906 | ||
907 | /* Wait for all revalidators to finish garbage collection. */ | |
d8043da7 | 908 | ovs_barrier_block(&udpif->reval_barrier); |
7d170098 EJ |
909 | |
910 | if (leader) { | |
b482e960 | 911 | unsigned int flow_limit; |
7d170098 EJ |
912 | long long int duration; |
913 | ||
b482e960 JR |
914 | atomic_read_relaxed(&udpif->flow_limit, &flow_limit); |
915 | ||
ac64794a | 916 | dpif_flow_dump_destroy(udpif->dump); |
7d170098 EJ |
917 | seq_change(udpif->dump_seq); |
918 | ||
919 | duration = MAX(time_msec() - start_time, 1); | |
7d170098 EJ |
920 | udpif->dump_duration = duration; |
921 | if (duration > 2000) { | |
922 | flow_limit /= duration / 1000; | |
923 | } else if (duration > 1300) { | |
924 | flow_limit = flow_limit * 3 / 4; | |
925 | } else if (duration < 1000 && n_flows > 2000 | |
926 | && flow_limit < n_flows * 1000 / duration) { | |
927 | flow_limit += 1000; | |
928 | } | |
929 | flow_limit = MIN(ofproto_flow_limit, MAX(flow_limit, 1000)); | |
b482e960 | 930 | atomic_store_relaxed(&udpif->flow_limit, flow_limit); |
e79a6c83 | 931 | |
7d170098 EJ |
932 | if (duration > 2000) { |
933 | VLOG_INFO("Spent an unreasonably long %lldms dumping flows", | |
934 | duration); | |
935 | } | |
e79a6c83 | 936 | |
7d170098 EJ |
937 | poll_timer_wait_until(start_time + MIN(ofproto_max_idle, 500)); |
938 | seq_wait(udpif->reval_seq, last_reval_seq); | |
939 | latch_wait(&udpif->exit_latch); | |
dba82d38 | 940 | latch_wait(&udpif->pause_latch); |
7d170098 | 941 | poll_block(); |
70d0cd06 JR |
942 | |
943 | if (!latch_is_set(&udpif->pause_latch) && | |
944 | !latch_is_set(&udpif->exit_latch)) { | |
945 | long long int now = time_msec(); | |
946 | /* Block again if we are woken up within 5ms of the last start | |
947 | * time. */ | |
948 | start_time += 5; | |
949 | ||
950 | if (now < start_time) { | |
951 | poll_timer_wait_until(start_time); | |
952 | latch_wait(&udpif->exit_latch); | |
953 | latch_wait(&udpif->pause_latch); | |
954 | poll_block(); | |
955 | } | |
956 | } | |
e79a6c83 EJ |
957 | } |
958 | } | |
959 | ||
960 | return NULL; | |
961 | } | |
962 | \f | |
e1ec7dd4 | 963 | static enum upcall_type |
cc377352 | 964 | classify_upcall(enum dpif_upcall_type type, const struct nlattr *userdata) |
e1ec7dd4 | 965 | { |
e1ec7dd4 EJ |
966 | union user_action_cookie cookie; |
967 | size_t userdata_len; | |
968 | ||
969 | /* First look at the upcall type. */ | |
cc377352 | 970 | switch (type) { |
e1ec7dd4 EJ |
971 | case DPIF_UC_ACTION: |
972 | break; | |
973 | ||
974 | case DPIF_UC_MISS: | |
975 | return MISS_UPCALL; | |
976 | ||
977 | case DPIF_N_UC_TYPES: | |
978 | default: | |
cc377352 | 979 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, type); |
e1ec7dd4 EJ |
980 | return BAD_UPCALL; |
981 | } | |
982 | ||
983 | /* "action" upcalls need a closer look. */ | |
cc377352 | 984 | if (!userdata) { |
e1ec7dd4 EJ |
985 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); |
986 | return BAD_UPCALL; | |
987 | } | |
cc377352 | 988 | userdata_len = nl_attr_get_size(userdata); |
e1ec7dd4 EJ |
989 | if (userdata_len < sizeof cookie.type |
990 | || userdata_len > sizeof cookie) { | |
34582733 | 991 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %"PRIuSIZE, |
e1ec7dd4 EJ |
992 | userdata_len); |
993 | return BAD_UPCALL; | |
994 | } | |
995 | memset(&cookie, 0, sizeof cookie); | |
cc377352 | 996 | memcpy(&cookie, nl_attr_get(userdata), userdata_len); |
f5790bf6 | 997 | if (userdata_len == MAX(8, sizeof cookie.sflow) |
e1ec7dd4 EJ |
998 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { |
999 | return SFLOW_UPCALL; | |
f5790bf6 | 1000 | } else if (userdata_len == MAX(8, sizeof cookie.slow_path) |
e1ec7dd4 EJ |
1001 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { |
1002 | return MISS_UPCALL; | |
f5790bf6 | 1003 | } else if (userdata_len == MAX(8, sizeof cookie.flow_sample) |
e1ec7dd4 EJ |
1004 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { |
1005 | return FLOW_SAMPLE_UPCALL; | |
f5790bf6 | 1006 | } else if (userdata_len == MAX(8, sizeof cookie.ipfix) |
e1ec7dd4 EJ |
1007 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { |
1008 | return IPFIX_UPCALL; | |
1009 | } else { | |
1010 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
34582733 | 1011 | " and size %"PRIuSIZE, cookie.type, userdata_len); |
e1ec7dd4 EJ |
1012 | return BAD_UPCALL; |
1013 | } | |
1014 | } | |
1015 | ||
e79a6c83 EJ |
1016 | /* Calculates slow path actions for 'xout'. 'buf' must statically be |
1017 | * initialized with at least 128 bytes of space. */ | |
1018 | static void | |
1019 | compose_slow_path(struct udpif *udpif, struct xlate_out *xout, | |
cc377352 | 1020 | const struct flow *flow, odp_port_t odp_in_port, |
9a159f74 | 1021 | struct ofpbuf *buf) |
e79a6c83 EJ |
1022 | { |
1023 | union user_action_cookie cookie; | |
1024 | odp_port_t port; | |
1025 | uint32_t pid; | |
1026 | ||
1027 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
1028 | cookie.slow_path.unused = 0; | |
1029 | cookie.slow_path.reason = xout->slow; | |
1030 | ||
1031 | port = xout->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP) | |
1032 | ? ODPP_NONE | |
1033 | : odp_in_port; | |
9a159f74 | 1034 | pid = dpif_port_get_pid(udpif->dpif, port, flow_hash_5tuple(flow, 0)); |
7321bda3 NM |
1035 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, |
1036 | ODPP_NONE, false, buf); | |
e79a6c83 EJ |
1037 | } |
1038 | ||
3d76b86c AW |
1039 | /* If there is no error, the upcall must be destroyed with upcall_uninit() |
1040 | * before quiescing, as the referred objects are guaranteed to exist only | |
1041 | * until the calling thread quiesces. Otherwise, do not call upcall_uninit() | |
1042 | * since the 'upcall->put_actions' remains uninitialized. */ | |
cc377352 EJ |
1043 | static int |
1044 | upcall_receive(struct upcall *upcall, const struct dpif_backer *backer, | |
cf62fa4c | 1045 | const struct dp_packet *packet, enum dpif_upcall_type type, |
7af12bd7 | 1046 | const struct nlattr *userdata, const struct flow *flow, |
27130224 | 1047 | const unsigned int mru, |
bd5131ba | 1048 | const ovs_u128 *ufid, const unsigned pmd_id) |
cc377352 EJ |
1049 | { |
1050 | int error; | |
1051 | ||
5c476ea3 JR |
1052 | error = xlate_lookup(backer, flow, &upcall->ofproto, &upcall->ipfix, |
1053 | &upcall->sflow, NULL, &upcall->in_port); | |
cc377352 EJ |
1054 | if (error) { |
1055 | return error; | |
1056 | } | |
1057 | ||
e672ff9b JR |
1058 | upcall->recirc = NULL; |
1059 | upcall->have_recirc_ref = false; | |
cc377352 EJ |
1060 | upcall->flow = flow; |
1061 | upcall->packet = packet; | |
7af12bd7 | 1062 | upcall->ufid = ufid; |
1c1e46ed | 1063 | upcall->pmd_id = pmd_id; |
cc377352 EJ |
1064 | upcall->type = type; |
1065 | upcall->userdata = userdata; | |
1520ef4f BP |
1066 | ofpbuf_use_stub(&upcall->odp_actions, upcall->odp_actions_stub, |
1067 | sizeof upcall->odp_actions_stub); | |
cc377352 EJ |
1068 | ofpbuf_init(&upcall->put_actions, 0); |
1069 | ||
1070 | upcall->xout_initialized = false; | |
23597df0 | 1071 | upcall->ukey_persists = false; |
cc377352 | 1072 | |
23597df0 | 1073 | upcall->ukey = NULL; |
cc377352 EJ |
1074 | upcall->key = NULL; |
1075 | upcall->key_len = 0; | |
27130224 | 1076 | upcall->mru = mru; |
cc377352 | 1077 | |
8b7ea2d4 | 1078 | upcall->out_tun_key = NULL; |
7321bda3 | 1079 | upcall->actions = NULL; |
8b7ea2d4 | 1080 | |
cc377352 EJ |
1081 | return 0; |
1082 | } | |
1083 | ||
a0bab870 | 1084 | static void |
cc377352 | 1085 | upcall_xlate(struct udpif *udpif, struct upcall *upcall, |
49a73e0c | 1086 | struct ofpbuf *odp_actions, struct flow_wildcards *wc) |
e1ec7dd4 | 1087 | { |
cc377352 | 1088 | struct dpif_flow_stats stats; |
691d39b2 | 1089 | struct xlate_in xin; |
a0bab870 | 1090 | |
cc377352 | 1091 | stats.n_packets = 1; |
cf62fa4c | 1092 | stats.n_bytes = dp_packet_size(upcall->packet); |
cc377352 EJ |
1093 | stats.used = time_msec(); |
1094 | stats.tcp_flags = ntohs(upcall->flow->tcp_flags); | |
a0bab870 | 1095 | |
1f4a8933 JR |
1096 | xlate_in_init(&xin, upcall->ofproto, |
1097 | ofproto_dpif_get_tables_version(upcall->ofproto), | |
1098 | upcall->flow, upcall->in_port, NULL, | |
1520ef4f | 1099 | stats.tcp_flags, upcall->packet, wc, odp_actions); |
a0bab870 | 1100 | |
cc377352 EJ |
1101 | if (upcall->type == DPIF_UC_MISS) { |
1102 | xin.resubmit_stats = &stats; | |
e672ff9b | 1103 | |
1d361a81 | 1104 | if (xin.frozen_state) { |
e672ff9b JR |
1105 | /* We may install a datapath flow only if we get a reference to the |
1106 | * recirculation context (otherwise we could have recirculation | |
1107 | * upcalls using recirculation ID for which no context can be | |
1108 | * found). We may still execute the flow's actions even if we | |
1109 | * don't install the flow. */ | |
1d361a81 | 1110 | upcall->recirc = recirc_id_node_from_state(xin.frozen_state); |
29b1ea3f | 1111 | upcall->have_recirc_ref = recirc_id_node_try_ref_rcu(upcall->recirc); |
e672ff9b | 1112 | } |
a0bab870 | 1113 | } else { |
e672ff9b JR |
1114 | /* For non-miss upcalls, we are either executing actions (one of which |
1115 | * is an userspace action) for an upcall, in which case the stats have | |
1116 | * already been taken care of, or there's a flow in the datapath which | |
1117 | * this packet was accounted to. Presumably the revalidators will deal | |
a0bab870 | 1118 | * with pushing its stats eventually. */ |
e1ec7dd4 EJ |
1119 | } |
1120 | ||
23597df0 JS |
1121 | upcall->dump_seq = seq_read(udpif->dump_seq); |
1122 | upcall->reval_seq = seq_read(udpif->reval_seq); | |
d1d7816b | 1123 | |
a0bab870 | 1124 | xlate_actions(&xin, &upcall->xout); |
d1d7816b JG |
1125 | if (wc) { |
1126 | /* Convert the input port wildcard from OFP to ODP format. There's no | |
1127 | * real way to do this for arbitrary bitmasks since the numbering spaces | |
1128 | * aren't the same. However, flow translation always exact matches the | |
1129 | * whole thing, so we can do the same here. */ | |
1130 | WC_MASK_FIELD(wc, in_port.odp_port); | |
1131 | } | |
1132 | ||
cc377352 EJ |
1133 | upcall->xout_initialized = true; |
1134 | ||
cc377352 EJ |
1135 | if (!upcall->xout.slow) { |
1136 | ofpbuf_use_const(&upcall->put_actions, | |
1520ef4f | 1137 | odp_actions->data, odp_actions->size); |
cc377352 | 1138 | } else { |
fff1b9c0 | 1139 | /* upcall->put_actions already initialized by upcall_receive(). */ |
cc377352 EJ |
1140 | compose_slow_path(udpif, &upcall->xout, upcall->flow, |
1141 | upcall->flow->in_port.odp_port, | |
1142 | &upcall->put_actions); | |
1143 | } | |
23597df0 | 1144 | |
7cde8208 JR |
1145 | /* This function is also called for slow-pathed flows. As we are only |
1146 | * going to create new datapath flows for actual datapath misses, there is | |
1147 | * no point in creating a ukey otherwise. */ | |
1148 | if (upcall->type == DPIF_UC_MISS) { | |
49a73e0c | 1149 | upcall->ukey = ukey_create_from_upcall(upcall, wc); |
7cde8208 | 1150 | } |
e1ec7dd4 EJ |
1151 | } |
1152 | ||
3eed53e9 | 1153 | static void |
cc377352 | 1154 | upcall_uninit(struct upcall *upcall) |
6b31e073 | 1155 | { |
cc377352 EJ |
1156 | if (upcall) { |
1157 | if (upcall->xout_initialized) { | |
1158 | xlate_out_uninit(&upcall->xout); | |
1159 | } | |
1520ef4f | 1160 | ofpbuf_uninit(&upcall->odp_actions); |
cc377352 | 1161 | ofpbuf_uninit(&upcall->put_actions); |
e672ff9b JR |
1162 | if (upcall->ukey) { |
1163 | if (!upcall->ukey_persists) { | |
1164 | ukey_delete__(upcall->ukey); | |
1165 | } | |
1166 | } else if (upcall->have_recirc_ref) { | |
1167 | /* The reference was transferred to the ukey if one was created. */ | |
1168 | recirc_id_node_unref(upcall->recirc); | |
23597df0 | 1169 | } |
cc377352 | 1170 | } |
6b31e073 RW |
1171 | } |
1172 | ||
b4c63252 JS |
1173 | /* If there are less flows than the limit, and this is a miss upcall which |
1174 | * | |
1175 | * - Has no recirc_id, OR | |
1176 | * - Has a recirc_id and we can get a reference on the recirc ctx, | |
1177 | * | |
1178 | * Then we should install the flow (true). Otherwise, return false. */ | |
1179 | static bool | |
1180 | should_install_flow(struct udpif *udpif, struct upcall *upcall) | |
1181 | { | |
1182 | unsigned int flow_limit; | |
1183 | ||
1184 | if (upcall->type != DPIF_UC_MISS) { | |
1185 | return false; | |
1186 | } else if (upcall->recirc && !upcall->have_recirc_ref) { | |
5221d53e | 1187 | VLOG_DBG_RL(&rl, "upcall: no reference for recirc flow"); |
b4c63252 JS |
1188 | return false; |
1189 | } | |
1190 | ||
1191 | atomic_read_relaxed(&udpif->flow_limit, &flow_limit); | |
1192 | if (udpif_get_n_flows(udpif) >= flow_limit) { | |
1193 | VLOG_WARN_RL(&rl, "upcall: datapath flow limit reached"); | |
1194 | return false; | |
1195 | } | |
1196 | ||
1197 | return true; | |
1198 | } | |
1199 | ||
623540e4 | 1200 | static int |
cf62fa4c | 1201 | upcall_cb(const struct dp_packet *packet, const struct flow *flow, ovs_u128 *ufid, |
bd5131ba | 1202 | unsigned pmd_id, enum dpif_upcall_type type, |
1c1e46ed AW |
1203 | const struct nlattr *userdata, struct ofpbuf *actions, |
1204 | struct flow_wildcards *wc, struct ofpbuf *put_actions, void *aux) | |
6b31e073 | 1205 | { |
9cec8274 | 1206 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); |
623540e4 | 1207 | struct udpif *udpif = aux; |
623540e4 EJ |
1208 | struct upcall upcall; |
1209 | bool megaflow; | |
1210 | int error; | |
6b31e073 | 1211 | |
b482e960 | 1212 | atomic_read_relaxed(&enable_megaflows, &megaflow); |
b482e960 | 1213 | |
623540e4 | 1214 | error = upcall_receive(&upcall, udpif->backer, packet, type, userdata, |
27130224 | 1215 | flow, 0, ufid, pmd_id); |
623540e4 | 1216 | if (error) { |
3d76b86c | 1217 | return error; |
6b31e073 | 1218 | } |
6b31e073 | 1219 | |
49a73e0c | 1220 | error = process_upcall(udpif, &upcall, actions, wc); |
623540e4 EJ |
1221 | if (error) { |
1222 | goto out; | |
1223 | } | |
cc377352 | 1224 | |
623540e4 | 1225 | if (upcall.xout.slow && put_actions) { |
6fd6ed71 PS |
1226 | ofpbuf_put(put_actions, upcall.put_actions.data, |
1227 | upcall.put_actions.size); | |
623540e4 | 1228 | } |
cc377352 | 1229 | |
49a73e0c BP |
1230 | if (OVS_UNLIKELY(!megaflow)) { |
1231 | flow_wildcards_init_for_packet(wc, flow); | |
623540e4 | 1232 | } |
9a159f74 | 1233 | |
b4c63252 | 1234 | if (!should_install_flow(udpif, &upcall)) { |
23597df0 | 1235 | error = ENOSPC; |
e672ff9b | 1236 | goto out; |
6b31e073 | 1237 | } |
623540e4 | 1238 | |
e672ff9b | 1239 | if (upcall.ukey && !ukey_install(udpif, upcall.ukey)) { |
9cec8274 | 1240 | VLOG_WARN_RL(&rl, "upcall_cb failure: ukey installation fails"); |
e672ff9b JR |
1241 | error = ENOSPC; |
1242 | } | |
623540e4 | 1243 | out: |
23597df0 JS |
1244 | if (!error) { |
1245 | upcall.ukey_persists = true; | |
1246 | } | |
623540e4 EJ |
1247 | upcall_uninit(&upcall); |
1248 | return error; | |
6b31e073 | 1249 | } |
10e57640 | 1250 | |
3eed53e9 | 1251 | static int |
cc377352 | 1252 | process_upcall(struct udpif *udpif, struct upcall *upcall, |
49a73e0c | 1253 | struct ofpbuf *odp_actions, struct flow_wildcards *wc) |
6b31e073 | 1254 | { |
cc377352 | 1255 | const struct nlattr *userdata = upcall->userdata; |
cf62fa4c | 1256 | const struct dp_packet *packet = upcall->packet; |
cc377352 | 1257 | const struct flow *flow = upcall->flow; |
04a19fb8 | 1258 | |
cc377352 EJ |
1259 | switch (classify_upcall(upcall->type, userdata)) { |
1260 | case MISS_UPCALL: | |
49a73e0c | 1261 | upcall_xlate(udpif, upcall, odp_actions, wc); |
cc377352 | 1262 | return 0; |
10e57640 | 1263 | |
6b31e073 | 1264 | case SFLOW_UPCALL: |
cc377352 | 1265 | if (upcall->sflow) { |
6b31e073 | 1266 | union user_action_cookie cookie; |
7321bda3 | 1267 | const struct nlattr *actions; |
b7637498 | 1268 | size_t actions_len = 0; |
7321bda3 NM |
1269 | struct dpif_sflow_actions sflow_actions; |
1270 | memset(&sflow_actions, 0, sizeof sflow_actions); | |
6b31e073 | 1271 | memset(&cookie, 0, sizeof cookie); |
cc377352 | 1272 | memcpy(&cookie, nl_attr_get(userdata), sizeof cookie.sflow); |
7321bda3 NM |
1273 | if (upcall->actions) { |
1274 | /* Actions were passed up from datapath. */ | |
1275 | actions = nl_attr_get(upcall->actions); | |
1276 | actions_len = nl_attr_get_size(upcall->actions); | |
1277 | if (actions && actions_len) { | |
1278 | dpif_sflow_read_actions(flow, actions, actions_len, | |
1279 | &sflow_actions); | |
1280 | } | |
1281 | } | |
1282 | if (actions_len == 0) { | |
1283 | /* Lookup actions in userspace cache. */ | |
5f2ccb1c IM |
1284 | struct udpif_key *ukey = ukey_lookup(udpif, upcall->ufid, |
1285 | upcall->pmd_id); | |
7321bda3 | 1286 | if (ukey) { |
b7637498 | 1287 | ukey_get_actions(ukey, &actions, &actions_len); |
7321bda3 NM |
1288 | dpif_sflow_read_actions(flow, actions, actions_len, |
1289 | &sflow_actions); | |
1290 | } | |
1291 | } | |
cc377352 | 1292 | dpif_sflow_received(upcall->sflow, packet, flow, |
7321bda3 NM |
1293 | flow->in_port.odp_port, &cookie, |
1294 | actions_len > 0 ? &sflow_actions : NULL); | |
6b31e073 RW |
1295 | } |
1296 | break; | |
cc377352 | 1297 | |
6b31e073 | 1298 | case IPFIX_UPCALL: |
cc377352 | 1299 | if (upcall->ipfix) { |
8b7ea2d4 WZ |
1300 | union user_action_cookie cookie; |
1301 | struct flow_tnl output_tunnel_key; | |
1302 | ||
1303 | memset(&cookie, 0, sizeof cookie); | |
1304 | memcpy(&cookie, nl_attr_get(userdata), sizeof cookie.ipfix); | |
1305 | ||
1306 | if (upcall->out_tun_key) { | |
8d8ab6c2 | 1307 | odp_tun_key_from_attr(upcall->out_tun_key, &output_tunnel_key); |
8b7ea2d4 WZ |
1308 | } |
1309 | dpif_ipfix_bridge_sample(upcall->ipfix, packet, flow, | |
1310 | flow->in_port.odp_port, | |
1311 | cookie.ipfix.output_odp_port, | |
1312 | upcall->out_tun_key ? | |
1313 | &output_tunnel_key : NULL); | |
6b31e073 RW |
1314 | } |
1315 | break; | |
cc377352 | 1316 | |
6b31e073 | 1317 | case FLOW_SAMPLE_UPCALL: |
cc377352 | 1318 | if (upcall->ipfix) { |
6b31e073 | 1319 | union user_action_cookie cookie; |
f69f713b | 1320 | struct flow_tnl output_tunnel_key; |
6b31e073 RW |
1321 | |
1322 | memset(&cookie, 0, sizeof cookie); | |
cc377352 | 1323 | memcpy(&cookie, nl_attr_get(userdata), sizeof cookie.flow_sample); |
6b31e073 | 1324 | |
f69f713b | 1325 | if (upcall->out_tun_key) { |
8d8ab6c2 | 1326 | odp_tun_key_from_attr(upcall->out_tun_key, &output_tunnel_key); |
f69f713b BY |
1327 | } |
1328 | ||
6b31e073 RW |
1329 | /* The flow reflects exactly the contents of the packet. |
1330 | * Sample the packet using it. */ | |
cc377352 | 1331 | dpif_ipfix_flow_sample(upcall->ipfix, packet, flow, |
f69f713b BY |
1332 | &cookie, flow->in_port.odp_port, |
1333 | upcall->out_tun_key ? | |
1334 | &output_tunnel_key : NULL); | |
e1ec7dd4 | 1335 | } |
6b31e073 | 1336 | break; |
cc377352 | 1337 | |
6b31e073 RW |
1338 | case BAD_UPCALL: |
1339 | break; | |
6b31e073 | 1340 | } |
10e57640 | 1341 | |
cc377352 | 1342 | return EAGAIN; |
9a159f74 AW |
1343 | } |
1344 | ||
1345 | static void | |
6b31e073 | 1346 | handle_upcalls(struct udpif *udpif, struct upcall *upcalls, |
a0bab870 | 1347 | size_t n_upcalls) |
9a159f74 | 1348 | { |
a0bab870 | 1349 | struct dpif_op *opsp[UPCALL_MAX_BATCH * 2]; |
6dad4d44 | 1350 | struct ukey_op ops[UPCALL_MAX_BATCH * 2]; |
23597df0 | 1351 | size_t n_ops, n_opsp, i; |
9a159f74 | 1352 | |
a0bab870 | 1353 | /* Handle the packets individually in order of arrival. |
04a19fb8 BP |
1354 | * |
1355 | * - For SLOW_CFM, SLOW_LACP, SLOW_STP, and SLOW_BFD, translation is what | |
1356 | * processes received packets for these protocols. | |
1357 | * | |
1358 | * - For SLOW_CONTROLLER, translation sends the packet to the OpenFlow | |
1359 | * controller. | |
1360 | * | |
1361 | * The loop fills 'ops' with an array of operations to execute in the | |
1362 | * datapath. */ | |
1363 | n_ops = 0; | |
9a159f74 AW |
1364 | for (i = 0; i < n_upcalls; i++) { |
1365 | struct upcall *upcall = &upcalls[i]; | |
cf62fa4c | 1366 | const struct dp_packet *packet = upcall->packet; |
6dad4d44 | 1367 | struct ukey_op *op; |
d02c42bf | 1368 | |
b4c63252 | 1369 | if (should_install_flow(udpif, upcall)) { |
bc2df54d | 1370 | struct udpif_key *ukey = upcall->ukey; |
d02c42bf | 1371 | |
54ebeff4 | 1372 | if (ukey_install(udpif, ukey)) { |
f3e8c44e JS |
1373 | upcall->ukey_persists = true; |
1374 | put_op_init(&ops[n_ops++], ukey, DPIF_FP_CREATE); | |
1375 | } | |
e79a6c83 EJ |
1376 | } |
1377 | ||
1520ef4f | 1378 | if (upcall->odp_actions.size) { |
04a19fb8 | 1379 | op = &ops[n_ops++]; |
23597df0 | 1380 | op->ukey = NULL; |
6dad4d44 | 1381 | op->dop.type = DPIF_OP_EXECUTE; |
cf62fa4c | 1382 | op->dop.u.execute.packet = CONST_CAST(struct dp_packet *, packet); |
1cceb31b | 1383 | op->dop.u.execute.flow = upcall->flow; |
a0bab870 | 1384 | odp_key_to_pkt_metadata(upcall->key, upcall->key_len, |
cf62fa4c | 1385 | &op->dop.u.execute.packet->md); |
1520ef4f BP |
1386 | op->dop.u.execute.actions = upcall->odp_actions.data; |
1387 | op->dop.u.execute.actions_len = upcall->odp_actions.size; | |
6dad4d44 JS |
1388 | op->dop.u.execute.needs_help = (upcall->xout.slow & SLOW_ACTION) != 0; |
1389 | op->dop.u.execute.probe = false; | |
27130224 | 1390 | op->dop.u.execute.mtu = upcall->mru; |
04a19fb8 | 1391 | } |
e1ec7dd4 | 1392 | } |
e1ec7dd4 | 1393 | |
54ebeff4 | 1394 | /* Execute batch. */ |
23597df0 | 1395 | n_opsp = 0; |
da546e07 | 1396 | for (i = 0; i < n_ops; i++) { |
23597df0 JS |
1397 | opsp[n_opsp++] = &ops[i].dop; |
1398 | } | |
1399 | dpif_operate(udpif->dpif, opsp, n_opsp); | |
1400 | for (i = 0; i < n_ops; i++) { | |
54ebeff4 JS |
1401 | struct udpif_key *ukey = ops[i].ukey; |
1402 | ||
1403 | if (ukey) { | |
1404 | ovs_mutex_lock(&ukey->mutex); | |
1405 | if (ops[i].dop.error) { | |
1406 | transition_ukey(ukey, UKEY_EVICTED); | |
1407 | } else { | |
1408 | transition_ukey(ukey, UKEY_OPERATIONAL); | |
1409 | } | |
1410 | ovs_mutex_unlock(&ukey->mutex); | |
23597df0 | 1411 | } |
da546e07 | 1412 | } |
e79a6c83 EJ |
1413 | } |
1414 | ||
7af12bd7 | 1415 | static uint32_t |
5f2ccb1c | 1416 | get_ukey_hash(const ovs_u128 *ufid, const unsigned pmd_id) |
7af12bd7 | 1417 | { |
5f2ccb1c | 1418 | return hash_2words(ufid->u32[0], pmd_id); |
7af12bd7 JS |
1419 | } |
1420 | ||
e79a6c83 | 1421 | static struct udpif_key * |
5f2ccb1c | 1422 | ukey_lookup(struct udpif *udpif, const ovs_u128 *ufid, const unsigned pmd_id) |
e79a6c83 EJ |
1423 | { |
1424 | struct udpif_key *ukey; | |
5f2ccb1c | 1425 | int idx = get_ukey_hash(ufid, pmd_id) % N_UMAPS; |
7af12bd7 | 1426 | struct cmap *cmap = &udpif->ukeys[idx].cmap; |
e79a6c83 | 1427 | |
5f2ccb1c IM |
1428 | CMAP_FOR_EACH_WITH_HASH (ukey, cmap_node, |
1429 | get_ukey_hash(ufid, pmd_id), cmap) { | |
2ff8484b | 1430 | if (ovs_u128_equals(ukey->ufid, *ufid)) { |
e79a6c83 EJ |
1431 | return ukey; |
1432 | } | |
1433 | } | |
1434 | return NULL; | |
1435 | } | |
1436 | ||
b7637498 EJ |
1437 | /* Provides safe lockless access of RCU protected 'ukey->actions'. Callers may |
1438 | * alternatively access the field directly if they take 'ukey->mutex'. */ | |
1439 | static void | |
1440 | ukey_get_actions(struct udpif_key *ukey, const struct nlattr **actions, size_t *size) | |
1441 | { | |
1442 | const struct ofpbuf *buf = ovsrcu_get(struct ofpbuf *, &ukey->actions); | |
1443 | *actions = buf->data; | |
1444 | *size = buf->size; | |
1445 | } | |
1446 | ||
1447 | static void | |
1448 | ukey_set_actions(struct udpif_key *ukey, const struct ofpbuf *actions) | |
1449 | { | |
1450 | ovsrcu_postpone(ofpbuf_delete, | |
1451 | ovsrcu_get_protected(struct ofpbuf *, &ukey->actions)); | |
1452 | ovsrcu_set(&ukey->actions, ofpbuf_clone(actions)); | |
1453 | } | |
1454 | ||
13bb6ed0 | 1455 | static struct udpif_key * |
7af12bd7 | 1456 | ukey_create__(const struct nlattr *key, size_t key_len, |
bc2df54d | 1457 | const struct nlattr *mask, size_t mask_len, |
70e5ed6f | 1458 | bool ufid_present, const ovs_u128 *ufid, |
bd5131ba | 1459 | const unsigned pmd_id, const struct ofpbuf *actions, |
e672ff9b | 1460 | uint64_t dump_seq, uint64_t reval_seq, long long int used, |
fbf5d6ec | 1461 | uint32_t key_recirc_id, struct xlate_out *xout) |
feca8bd7 | 1462 | OVS_NO_THREAD_SAFETY_ANALYSIS |
13bb6ed0 | 1463 | { |
fbf5d6ec | 1464 | struct udpif_key *ukey = xmalloc(sizeof *ukey); |
13bb6ed0 | 1465 | |
bc2df54d JS |
1466 | memcpy(&ukey->keybuf, key, key_len); |
1467 | ukey->key = &ukey->keybuf.nla; | |
1468 | ukey->key_len = key_len; | |
1469 | memcpy(&ukey->maskbuf, mask, mask_len); | |
1470 | ukey->mask = &ukey->maskbuf.nla; | |
1471 | ukey->mask_len = mask_len; | |
70e5ed6f | 1472 | ukey->ufid_present = ufid_present; |
7af12bd7 | 1473 | ukey->ufid = *ufid; |
1c1e46ed | 1474 | ukey->pmd_id = pmd_id; |
5f2ccb1c | 1475 | ukey->hash = get_ukey_hash(&ukey->ufid, pmd_id); |
b7637498 EJ |
1476 | |
1477 | ovsrcu_init(&ukey->actions, NULL); | |
1478 | ukey_set_actions(ukey, actions); | |
23597df0 JS |
1479 | |
1480 | ovs_mutex_init(&ukey->mutex); | |
1481 | ukey->dump_seq = dump_seq; | |
1482 | ukey->reval_seq = reval_seq; | |
54ebeff4 | 1483 | ukey->state = UKEY_CREATED; |
23597df0 | 1484 | ukey->created = time_msec(); |
13bb6ed0 | 1485 | memset(&ukey->stats, 0, sizeof ukey->stats); |
23597df0 | 1486 | ukey->stats.used = used; |
b256dc52 | 1487 | ukey->xcache = NULL; |
13bb6ed0 | 1488 | |
fbf5d6ec JR |
1489 | ukey->key_recirc_id = key_recirc_id; |
1490 | recirc_refs_init(&ukey->recircs); | |
1491 | if (xout) { | |
1492 | /* Take ownership of the action recirc id references. */ | |
1493 | recirc_refs_swap(&ukey->recircs, &xout->recircs); | |
e672ff9b | 1494 | } |
e672ff9b | 1495 | |
13bb6ed0 JS |
1496 | return ukey; |
1497 | } | |
1498 | ||
23597df0 | 1499 | static struct udpif_key * |
49a73e0c | 1500 | ukey_create_from_upcall(struct upcall *upcall, struct flow_wildcards *wc) |
23597df0 | 1501 | { |
bc2df54d JS |
1502 | struct odputil_keybuf keystub, maskstub; |
1503 | struct ofpbuf keybuf, maskbuf; | |
2494ccd7 | 1504 | bool megaflow; |
5262eea1 JG |
1505 | struct odp_flow_key_parms odp_parms = { |
1506 | .flow = upcall->flow, | |
49a73e0c | 1507 | .mask = &wc->masks, |
5262eea1 | 1508 | }; |
bc2df54d | 1509 | |
2494ccd7 | 1510 | odp_parms.support = ofproto_dpif_get_support(upcall->ofproto)->odp; |
bc2df54d JS |
1511 | if (upcall->key_len) { |
1512 | ofpbuf_use_const(&keybuf, upcall->key, upcall->key_len); | |
1513 | } else { | |
1514 | /* dpif-netdev doesn't provide a netlink-formatted flow key in the | |
1515 | * upcall, so convert the upcall's flow here. */ | |
1516 | ofpbuf_use_stack(&keybuf, &keystub, sizeof keystub); | |
5262eea1 | 1517 | odp_flow_key_from_flow(&odp_parms, &keybuf); |
bc2df54d JS |
1518 | } |
1519 | ||
1520 | atomic_read_relaxed(&enable_megaflows, &megaflow); | |
bc2df54d JS |
1521 | ofpbuf_use_stack(&maskbuf, &maskstub, sizeof maskstub); |
1522 | if (megaflow) { | |
ec1f6f32 | 1523 | odp_parms.key_buf = &keybuf; |
5262eea1 | 1524 | odp_flow_key_from_mask(&odp_parms, &maskbuf); |
bc2df54d JS |
1525 | } |
1526 | ||
6fd6ed71 | 1527 | return ukey_create__(keybuf.data, keybuf.size, maskbuf.data, maskbuf.size, |
1c1e46ed AW |
1528 | true, upcall->ufid, upcall->pmd_id, |
1529 | &upcall->put_actions, upcall->dump_seq, | |
e672ff9b | 1530 | upcall->reval_seq, 0, |
fbf5d6ec | 1531 | upcall->have_recirc_ref ? upcall->recirc->id : 0, |
e672ff9b | 1532 | &upcall->xout); |
23597df0 JS |
1533 | } |
1534 | ||
64bb477f | 1535 | static int |
23597df0 | 1536 | ukey_create_from_dpif_flow(const struct udpif *udpif, |
64bb477f JS |
1537 | const struct dpif_flow *flow, |
1538 | struct udpif_key **ukey) | |
23597df0 | 1539 | { |
64bb477f | 1540 | struct dpif_flow full_flow; |
bc2df54d | 1541 | struct ofpbuf actions; |
23597df0 | 1542 | uint64_t dump_seq, reval_seq; |
64bb477f | 1543 | uint64_t stub[DPIF_FLOW_BUFSIZE / 8]; |
e672ff9b JR |
1544 | const struct nlattr *a; |
1545 | unsigned int left; | |
64bb477f | 1546 | |
e672ff9b | 1547 | if (!flow->key_len || !flow->actions_len) { |
64bb477f JS |
1548 | struct ofpbuf buf; |
1549 | int err; | |
1550 | ||
e672ff9b JR |
1551 | /* If the key or actions were not provided by the datapath, fetch the |
1552 | * full flow. */ | |
64bb477f | 1553 | ofpbuf_use_stack(&buf, &stub, sizeof stub); |
af50de80 JS |
1554 | err = dpif_flow_get(udpif->dpif, flow->key, flow->key_len, |
1555 | flow->ufid_present ? &flow->ufid : NULL, | |
1c1e46ed | 1556 | flow->pmd_id, &buf, &full_flow); |
64bb477f JS |
1557 | if (err) { |
1558 | return err; | |
1559 | } | |
1560 | flow = &full_flow; | |
1561 | } | |
e672ff9b JR |
1562 | |
1563 | /* Check the flow actions for recirculation action. As recirculation | |
1564 | * relies on OVS userspace internal state, we need to delete all old | |
994fcc5a JR |
1565 | * datapath flows with either a non-zero recirc_id in the key, or any |
1566 | * recirculation actions upon OVS restart. */ | |
1567 | NL_ATTR_FOR_EACH_UNSAFE (a, left, flow->key, flow->key_len) { | |
1568 | if (nl_attr_type(a) == OVS_KEY_ATTR_RECIRC_ID | |
1569 | && nl_attr_get_u32(a) != 0) { | |
1570 | return EINVAL; | |
1571 | } | |
1572 | } | |
e672ff9b JR |
1573 | NL_ATTR_FOR_EACH_UNSAFE (a, left, flow->actions, flow->actions_len) { |
1574 | if (nl_attr_type(a) == OVS_ACTION_ATTR_RECIRC) { | |
1575 | return EINVAL; | |
1576 | } | |
1577 | } | |
1578 | ||
23597df0 JS |
1579 | dump_seq = seq_read(udpif->dump_seq); |
1580 | reval_seq = seq_read(udpif->reval_seq); | |
bc2df54d | 1581 | ofpbuf_use_const(&actions, &flow->actions, flow->actions_len); |
64bb477f JS |
1582 | *ukey = ukey_create__(flow->key, flow->key_len, |
1583 | flow->mask, flow->mask_len, flow->ufid_present, | |
1c1e46ed | 1584 | &flow->ufid, flow->pmd_id, &actions, dump_seq, |
fbf5d6ec | 1585 | reval_seq, flow->stats.used, 0, NULL); |
1c1e46ed | 1586 | |
64bb477f | 1587 | return 0; |
23597df0 JS |
1588 | } |
1589 | ||
67f08985 JS |
1590 | static bool |
1591 | try_ukey_replace(struct umap *umap, struct udpif_key *old_ukey, | |
1592 | struct udpif_key *new_ukey) | |
1593 | OVS_REQUIRES(umap->mutex) | |
1594 | OVS_TRY_LOCK(true, new_ukey->mutex) | |
1595 | { | |
1596 | bool replaced = false; | |
1597 | ||
1598 | if (!ovs_mutex_trylock(&old_ukey->mutex)) { | |
1599 | if (old_ukey->state == UKEY_EVICTED) { | |
1600 | /* The flow was deleted during the current revalidator dump, | |
1601 | * but its ukey won't be fully cleaned up until the sweep phase. | |
1602 | * In the mean time, we are receiving upcalls for this traffic. | |
1603 | * Expedite the (new) flow install by replacing the ukey. */ | |
1604 | ovs_mutex_lock(&new_ukey->mutex); | |
1605 | cmap_replace(&umap->cmap, &old_ukey->cmap_node, | |
1606 | &new_ukey->cmap_node, new_ukey->hash); | |
1607 | ovsrcu_postpone(ukey_delete__, old_ukey); | |
1608 | transition_ukey(old_ukey, UKEY_DELETED); | |
1609 | transition_ukey(new_ukey, UKEY_VISIBLE); | |
1610 | replaced = true; | |
1611 | } | |
1612 | ovs_mutex_unlock(&old_ukey->mutex); | |
1613 | } | |
1614 | ||
1615 | if (replaced) { | |
1616 | COVERAGE_INC(upcall_ukey_replace); | |
1617 | } else { | |
1618 | COVERAGE_INC(handler_duplicate_upcall); | |
1619 | } | |
1620 | return replaced; | |
1621 | } | |
1622 | ||
23597df0 JS |
1623 | /* Attempts to insert a ukey into the shared ukey maps. |
1624 | * | |
1625 | * On success, returns true, installs the ukey and returns it in a locked | |
1626 | * state. Otherwise, returns false. */ | |
1627 | static bool | |
54ebeff4 | 1628 | ukey_install__(struct udpif *udpif, struct udpif_key *new_ukey) |
23597df0 JS |
1629 | OVS_TRY_LOCK(true, new_ukey->mutex) |
1630 | { | |
1631 | struct umap *umap; | |
1632 | struct udpif_key *old_ukey; | |
1633 | uint32_t idx; | |
1634 | bool locked = false; | |
1635 | ||
1636 | idx = new_ukey->hash % N_UMAPS; | |
1637 | umap = &udpif->ukeys[idx]; | |
1638 | ovs_mutex_lock(&umap->mutex); | |
5f2ccb1c | 1639 | old_ukey = ukey_lookup(udpif, &new_ukey->ufid, new_ukey->pmd_id); |
23597df0 JS |
1640 | if (old_ukey) { |
1641 | /* Uncommon case: A ukey is already installed with the same UFID. */ | |
1642 | if (old_ukey->key_len == new_ukey->key_len | |
1643 | && !memcmp(old_ukey->key, new_ukey->key, new_ukey->key_len)) { | |
67f08985 | 1644 | locked = try_ukey_replace(umap, old_ukey, new_ukey); |
23597df0 JS |
1645 | } else { |
1646 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1647 | ||
70e5ed6f JS |
1648 | odp_format_ufid(&old_ukey->ufid, &ds); |
1649 | ds_put_cstr(&ds, " "); | |
23597df0 JS |
1650 | odp_flow_key_format(old_ukey->key, old_ukey->key_len, &ds); |
1651 | ds_put_cstr(&ds, "\n"); | |
70e5ed6f JS |
1652 | odp_format_ufid(&new_ukey->ufid, &ds); |
1653 | ds_put_cstr(&ds, " "); | |
23597df0 JS |
1654 | odp_flow_key_format(new_ukey->key, new_ukey->key_len, &ds); |
1655 | ||
1656 | VLOG_WARN_RL(&rl, "Conflicting ukey for flows:\n%s", ds_cstr(&ds)); | |
1657 | ds_destroy(&ds); | |
1658 | } | |
1659 | } else { | |
1660 | ovs_mutex_lock(&new_ukey->mutex); | |
1661 | cmap_insert(&umap->cmap, &new_ukey->cmap_node, new_ukey->hash); | |
54ebeff4 | 1662 | transition_ukey(new_ukey, UKEY_VISIBLE); |
23597df0 JS |
1663 | locked = true; |
1664 | } | |
1665 | ovs_mutex_unlock(&umap->mutex); | |
1666 | ||
1667 | return locked; | |
1668 | } | |
1669 | ||
1670 | static void | |
54ebeff4 JS |
1671 | transition_ukey(struct udpif_key *ukey, enum ukey_state dst) |
1672 | OVS_REQUIRES(ukey->mutex) | |
23597df0 | 1673 | { |
54ebeff4 JS |
1674 | ovs_assert(dst >= ukey->state); |
1675 | if (ukey->state == dst) { | |
1676 | return; | |
1677 | } | |
1678 | ||
1679 | /* Valid state transitions: | |
1680 | * UKEY_CREATED -> UKEY_VISIBLE | |
1681 | * Ukey is now visible in the umap. | |
1682 | * UKEY_VISIBLE -> UKEY_OPERATIONAL | |
1683 | * A handler has installed the flow, and the flow is in the datapath. | |
1684 | * UKEY_VISIBLE -> UKEY_EVICTING | |
1685 | * A handler installs the flow, then revalidator sweeps the ukey before | |
1686 | * the flow is dumped. Most likely the flow was installed; start trying | |
1687 | * to delete it. | |
1688 | * UKEY_VISIBLE -> UKEY_EVICTED | |
1689 | * A handler attempts to install the flow, but the datapath rejects it. | |
1690 | * Consider that the datapath has already destroyed it. | |
1691 | * UKEY_OPERATIONAL -> UKEY_EVICTING | |
1692 | * A revalidator decides to evict the datapath flow. | |
1693 | * UKEY_EVICTING -> UKEY_EVICTED | |
1694 | * A revalidator has evicted the datapath flow. | |
1695 | * UKEY_EVICTED -> UKEY_DELETED | |
1696 | * A revalidator has removed the ukey from the umap and is deleting it. | |
1697 | */ | |
1698 | if (ukey->state == dst - 1 || (ukey->state == UKEY_VISIBLE && | |
1699 | dst < UKEY_DELETED)) { | |
1700 | ukey->state = dst; | |
1701 | } else { | |
1702 | struct ds ds = DS_EMPTY_INITIALIZER; | |
23597df0 | 1703 | |
54ebeff4 JS |
1704 | odp_format_ufid(&ukey->ufid, &ds); |
1705 | VLOG_WARN_RL(&rl, "Invalid state transition for ukey %s: %d -> %d", | |
1706 | ds_cstr(&ds), ukey->state, dst); | |
1707 | ds_destroy(&ds); | |
23597df0 | 1708 | } |
23597df0 JS |
1709 | } |
1710 | ||
1711 | static bool | |
1712 | ukey_install(struct udpif *udpif, struct udpif_key *ukey) | |
1713 | { | |
54ebeff4 JS |
1714 | bool installed; |
1715 | ||
1716 | installed = ukey_install__(udpif, ukey); | |
1717 | if (installed) { | |
1718 | ovs_mutex_unlock(&ukey->mutex); | |
1719 | } | |
1720 | ||
1721 | return installed; | |
23597df0 JS |
1722 | } |
1723 | ||
1724 | /* Searches for a ukey in 'udpif->ukeys' that matches 'flow' and attempts to | |
1725 | * lock the ukey. If the ukey does not exist, create it. | |
7d170098 | 1726 | * |
64bb477f JS |
1727 | * Returns 0 on success, setting *result to the matching ukey and returning it |
1728 | * in a locked state. Otherwise, returns an errno and clears *result. EBUSY | |
1729 | * indicates that another thread is handling this flow. Other errors indicate | |
1730 | * an unexpected condition creating a new ukey. | |
1731 | * | |
1732 | * *error is an output parameter provided to appease the threadsafety analyser, | |
1733 | * and its value matches the return value. */ | |
23597df0 JS |
1734 | static int |
1735 | ukey_acquire(struct udpif *udpif, const struct dpif_flow *flow, | |
64bb477f JS |
1736 | struct udpif_key **result, int *error) |
1737 | OVS_TRY_LOCK(0, (*result)->mutex) | |
7d170098 | 1738 | { |
feca8bd7 | 1739 | struct udpif_key *ukey; |
64bb477f | 1740 | int retval; |
feca8bd7 | 1741 | |
5f2ccb1c | 1742 | ukey = ukey_lookup(udpif, &flow->ufid, flow->pmd_id); |
23597df0 | 1743 | if (ukey) { |
64bb477f | 1744 | retval = ovs_mutex_trylock(&ukey->mutex); |
23597df0 | 1745 | } else { |
23597df0 JS |
1746 | /* Usually we try to avoid installing flows from revalidator threads, |
1747 | * because locking on a umap may cause handler threads to block. | |
1748 | * However there are certain cases, like when ovs-vswitchd is | |
1749 | * restarted, where it is desirable to handle flows that exist in the | |
1750 | * datapath gracefully (ie, don't just clear the datapath). */ | |
64bb477f JS |
1751 | bool install; |
1752 | ||
1753 | retval = ukey_create_from_dpif_flow(udpif, flow, &ukey); | |
1754 | if (retval) { | |
1755 | goto done; | |
1756 | } | |
54ebeff4 | 1757 | install = ukey_install__(udpif, ukey); |
64bb477f | 1758 | if (install) { |
64bb477f | 1759 | retval = 0; |
23597df0 JS |
1760 | } else { |
1761 | ukey_delete__(ukey); | |
64bb477f | 1762 | retval = EBUSY; |
23597df0 | 1763 | } |
7d170098 | 1764 | } |
7d170098 | 1765 | |
64bb477f JS |
1766 | done: |
1767 | *error = retval; | |
1768 | if (retval) { | |
feca8bd7 | 1769 | *result = NULL; |
64bb477f JS |
1770 | } else { |
1771 | *result = ukey; | |
feca8bd7 | 1772 | } |
64bb477f | 1773 | return retval; |
7d170098 EJ |
1774 | } |
1775 | ||
e79a6c83 | 1776 | static void |
9fce0584 | 1777 | ukey_delete__(struct udpif_key *ukey) |
7d170098 | 1778 | OVS_NO_THREAD_SAFETY_ANALYSIS |
e79a6c83 | 1779 | { |
23597df0 | 1780 | if (ukey) { |
fbf5d6ec JR |
1781 | if (ukey->key_recirc_id) { |
1782 | recirc_free_id(ukey->key_recirc_id); | |
e672ff9b | 1783 | } |
fbf5d6ec | 1784 | recirc_refs_unref(&ukey->recircs); |
23597df0 | 1785 | xlate_cache_delete(ukey->xcache); |
b7637498 | 1786 | ofpbuf_delete(ovsrcu_get(struct ofpbuf *, &ukey->actions)); |
23597df0 JS |
1787 | ovs_mutex_destroy(&ukey->mutex); |
1788 | free(ukey); | |
1789 | } | |
e79a6c83 EJ |
1790 | } |
1791 | ||
9fce0584 | 1792 | static void |
b8d3daeb JS |
1793 | ukey_delete(struct umap *umap, struct udpif_key *ukey) |
1794 | OVS_REQUIRES(umap->mutex) | |
9fce0584 | 1795 | { |
54ebeff4 | 1796 | ovs_mutex_lock(&ukey->mutex); |
b8d3daeb | 1797 | cmap_remove(&umap->cmap, &ukey->cmap_node, ukey->hash); |
9fce0584 | 1798 | ovsrcu_postpone(ukey_delete__, ukey); |
54ebeff4 JS |
1799 | transition_ukey(ukey, UKEY_DELETED); |
1800 | ovs_mutex_unlock(&ukey->mutex); | |
9fce0584 JS |
1801 | } |
1802 | ||
698ffe36 | 1803 | static bool |
49fae772 JS |
1804 | should_revalidate(const struct udpif *udpif, uint64_t packets, |
1805 | long long int used) | |
698ffe36 JS |
1806 | { |
1807 | long long int metric, now, duration; | |
1808 | ||
95beec19 JS |
1809 | if (!used) { |
1810 | /* Always revalidate the first time a flow is dumped. */ | |
1811 | return true; | |
1812 | } | |
1813 | ||
49fae772 JS |
1814 | if (udpif->dump_duration < 200) { |
1815 | /* We are likely to handle full revalidation for the flows. */ | |
1816 | return true; | |
1817 | } | |
1818 | ||
698ffe36 JS |
1819 | /* Calculate the mean time between seeing these packets. If this |
1820 | * exceeds the threshold, then delete the flow rather than performing | |
1821 | * costly revalidation for flows that aren't being hit frequently. | |
1822 | * | |
1823 | * This is targeted at situations where the dump_duration is high (~1s), | |
1824 | * and revalidation is triggered by a call to udpif_revalidate(). In | |
1825 | * these situations, revalidation of all flows causes fluctuations in the | |
1826 | * flow_limit due to the interaction with the dump_duration and max_idle. | |
1827 | * This tends to result in deletion of low-throughput flows anyway, so | |
1828 | * skip the revalidation and just delete those flows. */ | |
1829 | packets = MAX(packets, 1); | |
1830 | now = MAX(used, time_msec()); | |
1831 | duration = now - used; | |
1832 | metric = duration / packets; | |
1833 | ||
49fae772 JS |
1834 | if (metric < 200) { |
1835 | /* The flow is receiving more than ~5pps, so keep it. */ | |
1836 | return true; | |
698ffe36 | 1837 | } |
49fae772 | 1838 | return false; |
698ffe36 JS |
1839 | } |
1840 | ||
c1c5c122 JS |
1841 | struct reval_context { |
1842 | /* Optional output parameters */ | |
1843 | struct flow_wildcards *wc; | |
1844 | struct ofpbuf *odp_actions; | |
1845 | struct netflow **netflow; | |
1846 | struct xlate_cache *xcache; | |
1847 | ||
1848 | /* Required output parameters */ | |
1849 | struct xlate_out xout; | |
1850 | struct flow flow; | |
1851 | }; | |
1852 | ||
dd0dc9ed | 1853 | /* Translates 'key' into a flow, populating 'ctx' as it goes along. |
c1c5c122 JS |
1854 | * |
1855 | * Returns 0 on success, otherwise a positive errno value. | |
1856 | * | |
1857 | * The caller is responsible for uninitializing ctx->xout on success. | |
1858 | */ | |
1859 | static int | |
dd0dc9ed JS |
1860 | xlate_key(struct udpif *udpif, const struct nlattr *key, unsigned int len, |
1861 | const struct dpif_flow_stats *push, struct reval_context *ctx) | |
c1c5c122 JS |
1862 | { |
1863 | struct ofproto_dpif *ofproto; | |
1864 | ofp_port_t ofp_in_port; | |
1865 | struct xlate_in xin; | |
1866 | int error; | |
1867 | ||
dd0dc9ed | 1868 | if (odp_flow_key_to_flow(key, len, &ctx->flow) == ODP_FIT_ERROR) { |
c1c5c122 JS |
1869 | return EINVAL; |
1870 | } | |
1871 | ||
1872 | error = xlate_lookup(udpif->backer, &ctx->flow, &ofproto, NULL, NULL, | |
1873 | ctx->netflow, &ofp_in_port); | |
1874 | if (error) { | |
1875 | return error; | |
1876 | } | |
1877 | ||
1878 | xlate_in_init(&xin, ofproto, ofproto_dpif_get_tables_version(ofproto), | |
1879 | &ctx->flow, ofp_in_port, NULL, push->tcp_flags, | |
1880 | NULL, ctx->wc, ctx->odp_actions); | |
1881 | if (push->n_packets) { | |
1882 | xin.resubmit_stats = push; | |
1883 | xin.allow_side_effects = true; | |
1884 | } | |
1885 | xin.xcache = ctx->xcache; | |
1886 | xlate_actions(&xin, &ctx->xout); | |
1887 | ||
1888 | return 0; | |
1889 | } | |
1890 | ||
dd0dc9ed JS |
1891 | static int |
1892 | xlate_ukey(struct udpif *udpif, const struct udpif_key *ukey, | |
fbf803b6 | 1893 | uint16_t tcp_flags, struct reval_context *ctx) |
dd0dc9ed | 1894 | { |
fbf803b6 JS |
1895 | struct dpif_flow_stats push = { |
1896 | .tcp_flags = tcp_flags, | |
1897 | }; | |
1898 | return xlate_key(udpif, ukey->key, ukey->key_len, &push, ctx); | |
1899 | } | |
1900 | ||
1901 | static int | |
1902 | populate_xcache(struct udpif *udpif, struct udpif_key *ukey, | |
1903 | uint16_t tcp_flags) | |
1904 | OVS_REQUIRES(ukey->mutex) | |
1905 | { | |
1906 | struct reval_context ctx = { | |
1907 | .odp_actions = NULL, | |
1908 | .netflow = NULL, | |
1909 | .wc = NULL, | |
1910 | }; | |
1911 | int error; | |
1912 | ||
1913 | ovs_assert(!ukey->xcache); | |
1914 | ukey->xcache = ctx.xcache = xlate_cache_new(); | |
1915 | error = xlate_ukey(udpif, ukey, tcp_flags, &ctx); | |
1916 | if (error) { | |
1917 | return error; | |
1918 | } | |
1919 | xlate_out_uninit(&ctx.xout); | |
1920 | ||
1921 | return 0; | |
dd0dc9ed JS |
1922 | } |
1923 | ||
43b2f131 | 1924 | static enum reval_result |
e2b0b03d | 1925 | revalidate_ukey__(struct udpif *udpif, const struct udpif_key *ukey, |
fbf803b6 | 1926 | uint16_t tcp_flags, struct ofpbuf *odp_actions, |
e2b0b03d | 1927 | struct recirc_refs *recircs, struct xlate_cache *xcache) |
e79a6c83 | 1928 | { |
c1c5c122 | 1929 | struct xlate_out *xoutp; |
42f3baca | 1930 | struct netflow *netflow; |
6448a693 | 1931 | struct flow_wildcards dp_mask, wc; |
43b2f131 | 1932 | enum reval_result result; |
c1c5c122 JS |
1933 | struct reval_context ctx = { |
1934 | .odp_actions = odp_actions, | |
1935 | .netflow = &netflow, | |
e2b0b03d JS |
1936 | .xcache = xcache, |
1937 | .wc = &wc, | |
c1c5c122 | 1938 | }; |
e79a6c83 | 1939 | |
43b2f131 | 1940 | result = UKEY_DELETE; |
e79a6c83 | 1941 | xoutp = NULL; |
42f3baca | 1942 | netflow = NULL; |
e79a6c83 | 1943 | |
fbf803b6 | 1944 | if (xlate_ukey(udpif, ukey, tcp_flags, &ctx)) { |
cc377352 EJ |
1945 | goto exit; |
1946 | } | |
c1c5c122 | 1947 | xoutp = &ctx.xout; |
ddeca9a4 | 1948 | |
c1c5c122 | 1949 | if (xoutp->slow) { |
43b2f131 | 1950 | ofpbuf_clear(odp_actions); |
c1c5c122 | 1951 | compose_slow_path(udpif, xoutp, &ctx.flow, ctx.flow.in_port.odp_port, |
43b2f131 | 1952 | odp_actions); |
e79a6c83 EJ |
1953 | } |
1954 | ||
c1c5c122 | 1955 | if (odp_flow_key_to_mask(ukey->mask, ukey->mask_len, &dp_mask, &ctx.flow) |
6448a693 | 1956 | == ODP_FIT_ERROR) { |
e79a6c83 EJ |
1957 | goto exit; |
1958 | } | |
1959 | ||
6448a693 JR |
1960 | /* Do not modify if any bit is wildcarded by the installed datapath flow, |
1961 | * but not the newly revalidated wildcard mask (wc), i.e., if revalidation | |
1962 | * tells that the datapath flow is now too generic and must be narrowed | |
1963 | * down. Note that we do not know if the datapath has ignored any of the | |
1964 | * wildcarded bits, so we may be overtly conservative here. */ | |
c1c5c122 | 1965 | if (flow_wildcards_has_extra(&dp_mask, ctx.wc)) { |
6448a693 | 1966 | goto exit; |
e79a6c83 | 1967 | } |
bc2df54d | 1968 | |
43b2f131 EJ |
1969 | if (!ofpbuf_equal(odp_actions, |
1970 | ovsrcu_get(struct ofpbuf *, &ukey->actions))) { | |
1971 | /* The datapath mask was OK, but the actions seem to have changed. | |
1972 | * Let's modify it in place. */ | |
1973 | result = UKEY_MODIFY; | |
fbf5d6ec JR |
1974 | /* Transfer recirc action ID references to the caller. */ |
1975 | recirc_refs_swap(recircs, &xoutp->recircs); | |
43b2f131 EJ |
1976 | goto exit; |
1977 | } | |
1978 | ||
1979 | result = UKEY_KEEP; | |
e79a6c83 EJ |
1980 | |
1981 | exit: | |
43b2f131 | 1982 | if (netflow && result == UKEY_DELETE) { |
c1c5c122 | 1983 | netflow_flow_clear(netflow, &ctx.flow); |
42f3baca | 1984 | } |
e79a6c83 | 1985 | xlate_out_uninit(xoutp); |
43b2f131 | 1986 | return result; |
e79a6c83 EJ |
1987 | } |
1988 | ||
95beec19 JS |
1989 | /* Verifies that the datapath actions of 'ukey' are still correct, and pushes |
1990 | * 'stats' for it. | |
1991 | * | |
1992 | * Returns a recommended action for 'ukey', options include: | |
1993 | * UKEY_DELETE The ukey should be deleted. | |
1994 | * UKEY_KEEP The ukey is fine as is. | |
1995 | * UKEY_MODIFY The ukey's actions should be changed but is otherwise | |
1996 | * fine. Callers should change the actions to those found | |
1997 | * in the caller supplied 'odp_actions' buffer. The | |
1998 | * recirculation references can be found in 'recircs' and | |
1999 | * must be handled by the caller. | |
2000 | * | |
2001 | * If the result is UKEY_MODIFY, then references to all recirc_ids used by the | |
2002 | * new flow will be held within 'recircs' (which may be none). | |
2003 | * | |
2004 | * The caller is responsible for both initializing 'recircs' prior this call, | |
2005 | * and ensuring any references are eventually freed. | |
2006 | */ | |
2007 | static enum reval_result | |
2008 | revalidate_ukey(struct udpif *udpif, struct udpif_key *ukey, | |
2009 | const struct dpif_flow_stats *stats, | |
2010 | struct ofpbuf *odp_actions, uint64_t reval_seq, | |
2011 | struct recirc_refs *recircs) | |
2012 | OVS_REQUIRES(ukey->mutex) | |
2013 | { | |
2014 | bool need_revalidate = ukey->reval_seq != reval_seq; | |
2015 | enum reval_result result = UKEY_DELETE; | |
2016 | struct dpif_flow_stats push; | |
2017 | ||
2018 | ofpbuf_clear(odp_actions); | |
2019 | ||
2020 | push.used = stats->used; | |
2021 | push.tcp_flags = stats->tcp_flags; | |
2022 | push.n_packets = (stats->n_packets > ukey->stats.n_packets | |
2023 | ? stats->n_packets - ukey->stats.n_packets | |
2024 | : 0); | |
2025 | push.n_bytes = (stats->n_bytes > ukey->stats.n_bytes | |
2026 | ? stats->n_bytes - ukey->stats.n_bytes | |
2027 | : 0); | |
2028 | ||
e2b0b03d JS |
2029 | if (need_revalidate) { |
2030 | if (should_revalidate(udpif, push.n_packets, ukey->stats.used)) { | |
2031 | if (!ukey->xcache) { | |
2032 | ukey->xcache = xlate_cache_new(); | |
2033 | } else { | |
2034 | xlate_cache_clear(ukey->xcache); | |
2035 | } | |
2036 | result = revalidate_ukey__(udpif, ukey, push.tcp_flags, | |
2037 | odp_actions, recircs, ukey->xcache); | |
2038 | } /* else delete; too expensive to revalidate */ | |
2039 | } else if (!push.n_packets || ukey->xcache | |
2040 | || !populate_xcache(udpif, ukey, push.tcp_flags)) { | |
2041 | result = UKEY_KEEP; | |
95beec19 JS |
2042 | } |
2043 | ||
fbf803b6 | 2044 | /* Stats for deleted flows will be attributed upon flow deletion. Skip. */ |
95beec19 | 2045 | if (result != UKEY_DELETE) { |
fbf803b6 JS |
2046 | xlate_push_stats(ukey->xcache, &push); |
2047 | ukey->stats = *stats; | |
95beec19 JS |
2048 | ukey->reval_seq = reval_seq; |
2049 | } | |
e2b0b03d | 2050 | |
95beec19 JS |
2051 | return result; |
2052 | } | |
2053 | ||
64bb477f | 2054 | static void |
8e1ffd75 JS |
2055 | delete_op_init__(struct udpif *udpif, struct ukey_op *op, |
2056 | const struct dpif_flow *flow) | |
64bb477f | 2057 | { |
4c438b67 | 2058 | op->ukey = NULL; |
64bb477f JS |
2059 | op->dop.type = DPIF_OP_FLOW_DEL; |
2060 | op->dop.u.flow_del.key = flow->key; | |
2061 | op->dop.u.flow_del.key_len = flow->key_len; | |
2062 | op->dop.u.flow_del.ufid = flow->ufid_present ? &flow->ufid : NULL; | |
1c1e46ed | 2063 | op->dop.u.flow_del.pmd_id = flow->pmd_id; |
64bb477f | 2064 | op->dop.u.flow_del.stats = &op->stats; |
70f07728 | 2065 | op->dop.u.flow_del.terse = udpif_use_ufid(udpif); |
64bb477f JS |
2066 | } |
2067 | ||
e79a6c83 | 2068 | static void |
8e1ffd75 | 2069 | delete_op_init(struct udpif *udpif, struct ukey_op *op, struct udpif_key *ukey) |
13bb6ed0 JS |
2070 | { |
2071 | op->ukey = ukey; | |
6dad4d44 | 2072 | op->dop.type = DPIF_OP_FLOW_DEL; |
bc2df54d JS |
2073 | op->dop.u.flow_del.key = ukey->key; |
2074 | op->dop.u.flow_del.key_len = ukey->key_len; | |
70e5ed6f | 2075 | op->dop.u.flow_del.ufid = ukey->ufid_present ? &ukey->ufid : NULL; |
1c1e46ed | 2076 | op->dop.u.flow_del.pmd_id = ukey->pmd_id; |
6dad4d44 | 2077 | op->dop.u.flow_del.stats = &op->stats; |
70f07728 | 2078 | op->dop.u.flow_del.terse = udpif_use_ufid(udpif); |
13bb6ed0 JS |
2079 | } |
2080 | ||
43b2f131 | 2081 | static void |
f673dcd8 JS |
2082 | put_op_init(struct ukey_op *op, struct udpif_key *ukey, |
2083 | enum dpif_flow_put_flags flags) | |
43b2f131 EJ |
2084 | { |
2085 | op->ukey = ukey; | |
2086 | op->dop.type = DPIF_OP_FLOW_PUT; | |
f673dcd8 | 2087 | op->dop.u.flow_put.flags = flags; |
43b2f131 EJ |
2088 | op->dop.u.flow_put.key = ukey->key; |
2089 | op->dop.u.flow_put.key_len = ukey->key_len; | |
2090 | op->dop.u.flow_put.mask = ukey->mask; | |
2091 | op->dop.u.flow_put.mask_len = ukey->mask_len; | |
4106e01a | 2092 | op->dop.u.flow_put.ufid = ukey->ufid_present ? &ukey->ufid : NULL; |
43b2f131 EJ |
2093 | op->dop.u.flow_put.pmd_id = ukey->pmd_id; |
2094 | op->dop.u.flow_put.stats = NULL; | |
2095 | ukey_get_actions(ukey, &op->dop.u.flow_put.actions, | |
2096 | &op->dop.u.flow_put.actions_len); | |
2097 | } | |
2098 | ||
dcf5840f JS |
2099 | /* Executes datapath operations 'ops' and attributes stats retrieved from the |
2100 | * datapath as part of those operations. */ | |
13bb6ed0 | 2101 | static void |
dcf5840f | 2102 | push_dp_ops(struct udpif *udpif, struct ukey_op *ops, size_t n_ops) |
e79a6c83 | 2103 | { |
13bb6ed0 JS |
2104 | struct dpif_op *opsp[REVALIDATE_MAX_BATCH]; |
2105 | size_t i; | |
e79a6c83 | 2106 | |
13bb6ed0 JS |
2107 | ovs_assert(n_ops <= REVALIDATE_MAX_BATCH); |
2108 | for (i = 0; i < n_ops; i++) { | |
6dad4d44 | 2109 | opsp[i] = &ops[i].dop; |
13bb6ed0 JS |
2110 | } |
2111 | dpif_operate(udpif->dpif, opsp, n_ops); | |
2112 | ||
2113 | for (i = 0; i < n_ops; i++) { | |
6dad4d44 | 2114 | struct ukey_op *op = &ops[i]; |
13bb6ed0 JS |
2115 | struct dpif_flow_stats *push, *stats, push_buf; |
2116 | ||
6dad4d44 | 2117 | stats = op->dop.u.flow_del.stats; |
5e73c322 JS |
2118 | push = &push_buf; |
2119 | ||
43b2f131 EJ |
2120 | if (op->dop.type != DPIF_OP_FLOW_DEL) { |
2121 | /* Only deleted flows need their stats pushed. */ | |
2122 | continue; | |
2123 | } | |
2124 | ||
e83c9357 AW |
2125 | if (op->dop.error) { |
2126 | /* flow_del error, 'stats' is unusable. */ | |
2127 | continue; | |
2128 | } | |
2129 | ||
64bb477f JS |
2130 | if (op->ukey) { |
2131 | ovs_mutex_lock(&op->ukey->mutex); | |
54ebeff4 | 2132 | transition_ukey(op->ukey, UKEY_EVICTED); |
64bb477f JS |
2133 | push->used = MAX(stats->used, op->ukey->stats.used); |
2134 | push->tcp_flags = stats->tcp_flags | op->ukey->stats.tcp_flags; | |
2135 | push->n_packets = stats->n_packets - op->ukey->stats.n_packets; | |
2136 | push->n_bytes = stats->n_bytes - op->ukey->stats.n_bytes; | |
2137 | ovs_mutex_unlock(&op->ukey->mutex); | |
2138 | } else { | |
2139 | push = stats; | |
2140 | } | |
13bb6ed0 JS |
2141 | |
2142 | if (push->n_packets || netflow_exists()) { | |
64bb477f JS |
2143 | const struct nlattr *key = op->dop.u.flow_del.key; |
2144 | size_t key_len = op->dop.u.flow_del.key_len; | |
13bb6ed0 | 2145 | struct netflow *netflow; |
dd0dc9ed JS |
2146 | struct reval_context ctx = { |
2147 | .netflow = &netflow, | |
2148 | }; | |
5e73c322 | 2149 | int error; |
b256dc52 | 2150 | |
64bb477f JS |
2151 | if (op->ukey) { |
2152 | ovs_mutex_lock(&op->ukey->mutex); | |
2153 | if (op->ukey->xcache) { | |
2154 | xlate_push_stats(op->ukey->xcache, push); | |
2155 | ovs_mutex_unlock(&op->ukey->mutex); | |
2156 | continue; | |
2157 | } | |
7d170098 | 2158 | ovs_mutex_unlock(&op->ukey->mutex); |
64bb477f JS |
2159 | key = op->ukey->key; |
2160 | key_len = op->ukey->key_len; | |
b256dc52 | 2161 | } |
13bb6ed0 | 2162 | |
dd0dc9ed JS |
2163 | error = xlate_key(udpif, key, key_len, push, &ctx); |
2164 | if (error) { | |
2165 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
13bb6ed0 | 2166 | |
dd0dc9ed JS |
2167 | VLOG_WARN_RL(&rl, "xlate_actions failed (%s)!", |
2168 | xlate_strerror(error)); | |
2169 | } else { | |
2170 | xlate_out_uninit(&ctx.xout); | |
13bb6ed0 | 2171 | if (netflow) { |
dd0dc9ed | 2172 | netflow_flow_clear(netflow, &ctx.flow); |
13bb6ed0 JS |
2173 | } |
2174 | } | |
2175 | } | |
2176 | } | |
7d170098 | 2177 | } |
13bb6ed0 | 2178 | |
dcf5840f JS |
2179 | /* Executes datapath operations 'ops', attributes stats retrieved from the |
2180 | * datapath, and deletes ukeys corresponding to deleted flows. */ | |
7d170098 | 2181 | static void |
6dad4d44 JS |
2182 | push_ukey_ops(struct udpif *udpif, struct umap *umap, |
2183 | struct ukey_op *ops, size_t n_ops) | |
7d170098 EJ |
2184 | { |
2185 | int i; | |
13bb6ed0 | 2186 | |
dcf5840f | 2187 | push_dp_ops(udpif, ops, n_ops); |
b8d3daeb | 2188 | ovs_mutex_lock(&umap->mutex); |
7d170098 | 2189 | for (i = 0; i < n_ops; i++) { |
c56eba3b JS |
2190 | if (ops[i].dop.type == DPIF_OP_FLOW_DEL) { |
2191 | ukey_delete(umap, ops[i].ukey); | |
2192 | } | |
13bb6ed0 | 2193 | } |
b8d3daeb | 2194 | ovs_mutex_unlock(&umap->mutex); |
13bb6ed0 JS |
2195 | } |
2196 | ||
64bb477f JS |
2197 | static void |
2198 | log_unexpected_flow(const struct dpif_flow *flow, int error) | |
2199 | { | |
2200 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 60); | |
2201 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2202 | ||
2203 | ds_put_format(&ds, "Failed to acquire udpif_key corresponding to " | |
2204 | "unexpected flow (%s): ", ovs_strerror(error)); | |
2205 | odp_format_ufid(&flow->ufid, &ds); | |
2206 | VLOG_WARN_RL(&rl, "%s", ds_cstr(&ds)); | |
2f22698f | 2207 | ds_destroy(&ds); |
64bb477f JS |
2208 | } |
2209 | ||
fbf5d6ec JR |
2210 | static void |
2211 | reval_op_init(struct ukey_op *op, enum reval_result result, | |
2212 | struct udpif *udpif, struct udpif_key *ukey, | |
2213 | struct recirc_refs *recircs, struct ofpbuf *odp_actions) | |
54ebeff4 | 2214 | OVS_REQUIRES(ukey->mutex) |
fbf5d6ec JR |
2215 | { |
2216 | if (result == UKEY_DELETE) { | |
2217 | delete_op_init(udpif, op, ukey); | |
54ebeff4 | 2218 | transition_ukey(ukey, UKEY_EVICTING); |
fbf5d6ec JR |
2219 | } else if (result == UKEY_MODIFY) { |
2220 | /* Store the new recircs. */ | |
2221 | recirc_refs_swap(&ukey->recircs, recircs); | |
2222 | /* Release old recircs. */ | |
2223 | recirc_refs_unref(recircs); | |
2224 | /* ukey->key_recirc_id remains, as the key is the same as before. */ | |
2225 | ||
2226 | ukey_set_actions(ukey, odp_actions); | |
f673dcd8 | 2227 | put_op_init(op, ukey, DPIF_FP_MODIFY); |
fbf5d6ec JR |
2228 | } |
2229 | } | |
2230 | ||
13bb6ed0 | 2231 | static void |
7d170098 | 2232 | revalidate(struct revalidator *revalidator) |
13bb6ed0 | 2233 | { |
43b2f131 EJ |
2234 | uint64_t odp_actions_stub[1024 / 8]; |
2235 | struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub); | |
2236 | ||
13bb6ed0 | 2237 | struct udpif *udpif = revalidator->udpif; |
ac64794a | 2238 | struct dpif_flow_dump_thread *dump_thread; |
23597df0 | 2239 | uint64_t dump_seq, reval_seq; |
e79a6c83 | 2240 | unsigned int flow_limit; |
e79a6c83 | 2241 | |
efa08531 | 2242 | dump_seq = seq_read(udpif->dump_seq); |
23597df0 | 2243 | reval_seq = seq_read(udpif->reval_seq); |
b482e960 | 2244 | atomic_read_relaxed(&udpif->flow_limit, &flow_limit); |
ac64794a BP |
2245 | dump_thread = dpif_flow_dump_thread_create(udpif->dump); |
2246 | for (;;) { | |
6dad4d44 | 2247 | struct ukey_op ops[REVALIDATE_MAX_BATCH]; |
ac64794a | 2248 | int n_ops = 0; |
e79a6c83 | 2249 | |
ac64794a BP |
2250 | struct dpif_flow flows[REVALIDATE_MAX_BATCH]; |
2251 | const struct dpif_flow *f; | |
2252 | int n_dumped; | |
7d170098 | 2253 | |
ac64794a BP |
2254 | long long int max_idle; |
2255 | long long int now; | |
2256 | size_t n_dp_flows; | |
2257 | bool kill_them_all; | |
e79a6c83 | 2258 | |
ac64794a BP |
2259 | n_dumped = dpif_flow_dump_next(dump_thread, flows, ARRAY_SIZE(flows)); |
2260 | if (!n_dumped) { | |
2261 | break; | |
73a3c475 JS |
2262 | } |
2263 | ||
ac64794a BP |
2264 | now = time_msec(); |
2265 | ||
2266 | /* In normal operation we want to keep flows around until they have | |
2267 | * been idle for 'ofproto_max_idle' milliseconds. However: | |
2268 | * | |
2269 | * - If the number of datapath flows climbs above 'flow_limit', | |
2270 | * drop that down to 100 ms to try to bring the flows down to | |
2271 | * the limit. | |
2272 | * | |
2273 | * - If the number of datapath flows climbs above twice | |
2274 | * 'flow_limit', delete all the datapath flows as an emergency | |
2275 | * measure. (We reassess this condition for the next batch of | |
2276 | * datapath flows, so we will recover before all the flows are | |
2277 | * gone.) */ | |
2278 | n_dp_flows = udpif_get_n_flows(udpif); | |
2279 | kill_them_all = n_dp_flows > flow_limit * 2; | |
2280 | max_idle = n_dp_flows > flow_limit ? 100 : ofproto_max_idle; | |
2281 | ||
2282 | for (f = flows; f < &flows[n_dumped]; f++) { | |
2283 | long long int used = f->stats.used; | |
fbf5d6ec | 2284 | struct recirc_refs recircs = RECIRC_REFS_EMPTY_INITIALIZER; |
43b2f131 | 2285 | enum reval_result result; |
feca8bd7 | 2286 | struct udpif_key *ukey; |
43b2f131 | 2287 | bool already_dumped; |
64bb477f | 2288 | int error; |
acaa8dac | 2289 | |
64bb477f JS |
2290 | if (ukey_acquire(udpif, f, &ukey, &error)) { |
2291 | if (error == EBUSY) { | |
2292 | /* Another thread is processing this flow, so don't bother | |
2293 | * processing it.*/ | |
2294 | COVERAGE_INC(upcall_ukey_contention); | |
2295 | } else { | |
2296 | log_unexpected_flow(f, error); | |
c744eb04 | 2297 | if (error != ENOENT) { |
8e1ffd75 | 2298 | delete_op_init__(udpif, &ops[n_ops++], f); |
c744eb04 | 2299 | } |
64bb477f | 2300 | } |
acaa8dac JS |
2301 | continue; |
2302 | } | |
2303 | ||
efa08531 | 2304 | already_dumped = ukey->dump_seq == dump_seq; |
acaa8dac | 2305 | if (already_dumped) { |
ec47af51 JS |
2306 | /* The flow has already been handled during this flow dump |
2307 | * operation. Skip it. */ | |
2308 | if (ukey->xcache) { | |
2309 | COVERAGE_INC(dumped_duplicate_flow); | |
2310 | } else { | |
2311 | COVERAGE_INC(dumped_new_flow); | |
2312 | } | |
acaa8dac JS |
2313 | ovs_mutex_unlock(&ukey->mutex); |
2314 | continue; | |
2315 | } | |
2316 | ||
54ebeff4 JS |
2317 | /* The flow is now confirmed to be in the datapath. */ |
2318 | transition_ukey(ukey, UKEY_OPERATIONAL); | |
2319 | ||
acaa8dac JS |
2320 | if (!used) { |
2321 | used = ukey->created; | |
2322 | } | |
ac64794a | 2323 | if (kill_them_all || (used && used < now - max_idle)) { |
43b2f131 | 2324 | result = UKEY_DELETE; |
ac64794a | 2325 | } else { |
43b2f131 | 2326 | result = revalidate_ukey(udpif, ukey, &f->stats, &odp_actions, |
fbf5d6ec | 2327 | reval_seq, &recircs); |
ac64794a | 2328 | } |
efa08531 | 2329 | ukey->dump_seq = dump_seq; |
e79a6c83 | 2330 | |
fbf5d6ec JR |
2331 | if (result != UKEY_KEEP) { |
2332 | /* Takes ownership of 'recircs'. */ | |
2333 | reval_op_init(&ops[n_ops++], result, udpif, ukey, &recircs, | |
2334 | &odp_actions); | |
ac64794a | 2335 | } |
acaa8dac | 2336 | ovs_mutex_unlock(&ukey->mutex); |
7d170098 | 2337 | } |
ad3415c0 | 2338 | |
ac64794a | 2339 | if (n_ops) { |
dcf5840f JS |
2340 | /* Push datapath ops but defer ukey deletion to 'sweep' phase. */ |
2341 | push_dp_ops(udpif, ops, n_ops); | |
7d170098 | 2342 | } |
9fce0584 | 2343 | ovsrcu_quiesce(); |
e79a6c83 | 2344 | } |
ac64794a | 2345 | dpif_flow_dump_thread_destroy(dump_thread); |
43b2f131 | 2346 | ofpbuf_uninit(&odp_actions); |
3b62a9d3 JS |
2347 | } |
2348 | ||
dba82d38 AW |
2349 | /* Pauses the 'revalidator', can only proceed after main thread |
2350 | * calls udpif_resume_revalidators(). */ | |
2351 | static void | |
2352 | revalidator_pause(struct revalidator *revalidator) | |
2353 | { | |
2354 | /* The first block is for sync'ing the pause with main thread. */ | |
2355 | ovs_barrier_block(&revalidator->udpif->pause_barrier); | |
2356 | /* The second block is for pausing until main thread resumes. */ | |
2357 | ovs_barrier_block(&revalidator->udpif->pause_barrier); | |
2358 | } | |
2359 | ||
e79a6c83 | 2360 | static void |
e96a5c24 | 2361 | revalidator_sweep__(struct revalidator *revalidator, bool purge) |
e79a6c83 | 2362 | { |
b8d3daeb | 2363 | struct udpif *udpif; |
23597df0 | 2364 | uint64_t dump_seq, reval_seq; |
b8d3daeb | 2365 | int slice; |
e4b79342 | 2366 | |
b8d3daeb JS |
2367 | udpif = revalidator->udpif; |
2368 | dump_seq = seq_read(udpif->dump_seq); | |
23597df0 | 2369 | reval_seq = seq_read(udpif->reval_seq); |
b8d3daeb JS |
2370 | slice = revalidator - udpif->revalidators; |
2371 | ovs_assert(slice < udpif->n_revalidators); | |
2372 | ||
2373 | for (int i = slice; i < N_UMAPS; i += udpif->n_revalidators) { | |
43b2f131 EJ |
2374 | uint64_t odp_actions_stub[1024 / 8]; |
2375 | struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub); | |
2376 | ||
6dad4d44 | 2377 | struct ukey_op ops[REVALIDATE_MAX_BATCH]; |
b8d3daeb JS |
2378 | struct udpif_key *ukey; |
2379 | struct umap *umap = &udpif->ukeys[i]; | |
2380 | size_t n_ops = 0; | |
e79a6c83 | 2381 | |
b8d3daeb | 2382 | CMAP_FOR_EACH(ukey, cmap_node, &umap->cmap) { |
54ebeff4 | 2383 | enum ukey_state ukey_state; |
a2606936 | 2384 | |
23597df0 JS |
2385 | /* Handler threads could be holding a ukey lock while it installs a |
2386 | * new flow, so don't hang around waiting for access to it. */ | |
2387 | if (ovs_mutex_trylock(&ukey->mutex)) { | |
2388 | continue; | |
2389 | } | |
54ebeff4 JS |
2390 | ukey_state = ukey->state; |
2391 | if (ukey_state == UKEY_OPERATIONAL | |
2392 | || (ukey_state == UKEY_VISIBLE && purge)) { | |
cebfec69 JS |
2393 | struct recirc_refs recircs = RECIRC_REFS_EMPTY_INITIALIZER; |
2394 | bool seq_mismatch = (ukey->dump_seq != dump_seq | |
2395 | && ukey->reval_seq != reval_seq); | |
2396 | enum reval_result result; | |
2397 | ||
2398 | if (purge) { | |
2399 | result = UKEY_DELETE; | |
2400 | } else if (!seq_mismatch) { | |
2401 | result = UKEY_KEEP; | |
2402 | } else { | |
2403 | struct dpif_flow_stats stats; | |
2404 | COVERAGE_INC(revalidate_missed_dp_flow); | |
2405 | memset(&stats, 0, sizeof stats); | |
2406 | result = revalidate_ukey(udpif, ukey, &stats, &odp_actions, | |
2407 | reval_seq, &recircs); | |
2408 | } | |
2409 | if (result != UKEY_KEEP) { | |
2410 | /* Clears 'recircs' if filled by revalidate_ukey(). */ | |
2411 | reval_op_init(&ops[n_ops++], result, udpif, ukey, &recircs, | |
2412 | &odp_actions); | |
2413 | } | |
43b2f131 | 2414 | } |
fbf5d6ec | 2415 | ovs_mutex_unlock(&ukey->mutex); |
43b2f131 | 2416 | |
54ebeff4 | 2417 | if (ukey_state == UKEY_EVICTED) { |
dcf5840f JS |
2418 | /* The common flow deletion case involves deletion of the flow |
2419 | * during the dump phase and ukey deletion here. */ | |
b8d3daeb JS |
2420 | ovs_mutex_lock(&umap->mutex); |
2421 | ukey_delete(umap, ukey); | |
2422 | ovs_mutex_unlock(&umap->mutex); | |
e4b79342 | 2423 | } |
cebfec69 JS |
2424 | |
2425 | if (n_ops == REVALIDATE_MAX_BATCH) { | |
dcf5840f JS |
2426 | /* Update/delete missed flows and clean up corresponding ukeys |
2427 | * if necessary. */ | |
cebfec69 JS |
2428 | push_ukey_ops(udpif, umap, ops, n_ops); |
2429 | n_ops = 0; | |
2430 | } | |
e79a6c83 | 2431 | } |
e4b79342 | 2432 | |
b8d3daeb | 2433 | if (n_ops) { |
6dad4d44 | 2434 | push_ukey_ops(udpif, umap, ops, n_ops); |
b8d3daeb | 2435 | } |
43b2f131 EJ |
2436 | |
2437 | ofpbuf_uninit(&odp_actions); | |
b8d3daeb | 2438 | ovsrcu_quiesce(); |
e4b79342 | 2439 | } |
e1ec7dd4 | 2440 | } |
e96a5c24 JS |
2441 | |
2442 | static void | |
2443 | revalidator_sweep(struct revalidator *revalidator) | |
2444 | { | |
2445 | revalidator_sweep__(revalidator, false); | |
2446 | } | |
2447 | ||
2448 | static void | |
2449 | revalidator_purge(struct revalidator *revalidator) | |
2450 | { | |
2451 | revalidator_sweep__(revalidator, true); | |
2452 | } | |
e4e74c3a AW |
2453 | |
2454 | /* In reaction to dpif purge, purges all 'ukey's with same 'pmd_id'. */ | |
2455 | static void | |
2456 | dp_purge_cb(void *aux, unsigned pmd_id) | |
54ebeff4 | 2457 | OVS_NO_THREAD_SAFETY_ANALYSIS |
e4e74c3a AW |
2458 | { |
2459 | struct udpif *udpif = aux; | |
2460 | size_t i; | |
2461 | ||
2462 | udpif_pause_revalidators(udpif); | |
2463 | for (i = 0; i < N_UMAPS; i++) { | |
2464 | struct ukey_op ops[REVALIDATE_MAX_BATCH]; | |
2465 | struct udpif_key *ukey; | |
2466 | struct umap *umap = &udpif->ukeys[i]; | |
2467 | size_t n_ops = 0; | |
2468 | ||
2469 | CMAP_FOR_EACH(ukey, cmap_node, &umap->cmap) { | |
54ebeff4 | 2470 | if (ukey->pmd_id == pmd_id) { |
e4e74c3a | 2471 | delete_op_init(udpif, &ops[n_ops++], ukey); |
54ebeff4 JS |
2472 | transition_ukey(ukey, UKEY_EVICTING); |
2473 | ||
e4e74c3a AW |
2474 | if (n_ops == REVALIDATE_MAX_BATCH) { |
2475 | push_ukey_ops(udpif, umap, ops, n_ops); | |
2476 | n_ops = 0; | |
2477 | } | |
2478 | } | |
2479 | } | |
2480 | ||
2481 | if (n_ops) { | |
2482 | push_ukey_ops(udpif, umap, ops, n_ops); | |
2483 | } | |
2484 | ||
2485 | ovsrcu_quiesce(); | |
2486 | } | |
2487 | udpif_resume_revalidators(udpif); | |
2488 | } | |
e22d52ee EJ |
2489 | \f |
2490 | static void | |
2491 | upcall_unixctl_show(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2492 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
2493 | { | |
2494 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2495 | struct udpif *udpif; | |
2496 | ||
2497 | LIST_FOR_EACH (udpif, list_node, &all_udpifs) { | |
e79a6c83 | 2498 | unsigned int flow_limit; |
64bb477f | 2499 | bool ufid_enabled; |
e22d52ee EJ |
2500 | size_t i; |
2501 | ||
b482e960 | 2502 | atomic_read_relaxed(&udpif->flow_limit, &flow_limit); |
70f07728 | 2503 | ufid_enabled = udpif_use_ufid(udpif); |
e79a6c83 | 2504 | |
e22d52ee | 2505 | ds_put_format(&ds, "%s:\n", dpif_name(udpif->dpif)); |
0e2a9f6f | 2506 | ds_put_format(&ds, "\tflows : (current %lu)" |
e79a6c83 EJ |
2507 | " (avg %u) (max %u) (limit %u)\n", udpif_get_n_flows(udpif), |
2508 | udpif->avg_n_flows, udpif->max_n_flows, flow_limit); | |
e79a6c83 | 2509 | ds_put_format(&ds, "\tdump duration : %lldms\n", udpif->dump_duration); |
64bb477f JS |
2510 | ds_put_format(&ds, "\tufid enabled : "); |
2511 | if (ufid_enabled) { | |
2512 | ds_put_format(&ds, "true\n"); | |
2513 | } else { | |
2514 | ds_put_format(&ds, "false\n"); | |
2515 | } | |
e79a6c83 | 2516 | ds_put_char(&ds, '\n'); |
b8d3daeb | 2517 | |
e79a6c83 EJ |
2518 | for (i = 0; i < n_revalidators; i++) { |
2519 | struct revalidator *revalidator = &udpif->revalidators[i]; | |
b8d3daeb | 2520 | int j, elements = 0; |
e79a6c83 | 2521 | |
b8d3daeb JS |
2522 | for (j = i; j < N_UMAPS; j += n_revalidators) { |
2523 | elements += cmap_count(&udpif->ukeys[j].cmap); | |
2524 | } | |
2525 | ds_put_format(&ds, "\t%u: (keys %d)\n", revalidator->id, elements); | |
e79a6c83 | 2526 | } |
e22d52ee EJ |
2527 | } |
2528 | ||
2529 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
2530 | ds_destroy(&ds); | |
2531 | } | |
e79a6c83 EJ |
2532 | |
2533 | /* Disable using the megaflows. | |
2534 | * | |
2535 | * This command is only needed for advanced debugging, so it's not | |
2536 | * documented in the man page. */ | |
2537 | static void | |
2538 | upcall_unixctl_disable_megaflows(struct unixctl_conn *conn, | |
2539 | int argc OVS_UNUSED, | |
2540 | const char *argv[] OVS_UNUSED, | |
2541 | void *aux OVS_UNUSED) | |
2542 | { | |
b482e960 | 2543 | atomic_store_relaxed(&enable_megaflows, false); |
1b5b5071 | 2544 | udpif_flush_all_datapaths(); |
e79a6c83 EJ |
2545 | unixctl_command_reply(conn, "megaflows disabled"); |
2546 | } | |
2547 | ||
2548 | /* Re-enable using megaflows. | |
2549 | * | |
2550 | * This command is only needed for advanced debugging, so it's not | |
2551 | * documented in the man page. */ | |
2552 | static void | |
2553 | upcall_unixctl_enable_megaflows(struct unixctl_conn *conn, | |
2554 | int argc OVS_UNUSED, | |
2555 | const char *argv[] OVS_UNUSED, | |
2556 | void *aux OVS_UNUSED) | |
2557 | { | |
b482e960 | 2558 | atomic_store_relaxed(&enable_megaflows, true); |
1b5b5071 | 2559 | udpif_flush_all_datapaths(); |
e79a6c83 EJ |
2560 | unixctl_command_reply(conn, "megaflows enabled"); |
2561 | } | |
94b8c324 | 2562 | |
64bb477f JS |
2563 | /* Disable skipping flow attributes during flow dump. |
2564 | * | |
2565 | * This command is only needed for advanced debugging, so it's not | |
2566 | * documented in the man page. */ | |
2567 | static void | |
2568 | upcall_unixctl_disable_ufid(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2569 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
2570 | { | |
70f07728 | 2571 | atomic_store_relaxed(&enable_ufid, false); |
64bb477f JS |
2572 | unixctl_command_reply(conn, "Datapath dumping tersely using UFID disabled"); |
2573 | } | |
2574 | ||
2575 | /* Re-enable skipping flow attributes during flow dump. | |
2576 | * | |
2577 | * This command is only needed for advanced debugging, so it's not documented | |
2578 | * in the man page. */ | |
2579 | static void | |
2580 | upcall_unixctl_enable_ufid(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2581 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
2582 | { | |
70f07728 JS |
2583 | atomic_store_relaxed(&enable_ufid, true); |
2584 | unixctl_command_reply(conn, "Datapath dumping tersely using UFID enabled " | |
2585 | "for supported datapaths"); | |
64bb477f JS |
2586 | } |
2587 | ||
94b8c324 JS |
2588 | /* Set the flow limit. |
2589 | * | |
2590 | * This command is only needed for advanced debugging, so it's not | |
2591 | * documented in the man page. */ | |
2592 | static void | |
2593 | upcall_unixctl_set_flow_limit(struct unixctl_conn *conn, | |
2594 | int argc OVS_UNUSED, | |
2595 | const char *argv[] OVS_UNUSED, | |
2596 | void *aux OVS_UNUSED) | |
2597 | { | |
2598 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2599 | struct udpif *udpif; | |
2600 | unsigned int flow_limit = atoi(argv[1]); | |
2601 | ||
2602 | LIST_FOR_EACH (udpif, list_node, &all_udpifs) { | |
b482e960 | 2603 | atomic_store_relaxed(&udpif->flow_limit, flow_limit); |
94b8c324 JS |
2604 | } |
2605 | ds_put_format(&ds, "set flow_limit to %u\n", flow_limit); | |
2606 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
2607 | ds_destroy(&ds); | |
2608 | } | |
27f57736 JS |
2609 | |
2610 | static void | |
2611 | upcall_unixctl_dump_wait(struct unixctl_conn *conn, | |
2612 | int argc OVS_UNUSED, | |
2613 | const char *argv[] OVS_UNUSED, | |
2614 | void *aux OVS_UNUSED) | |
2615 | { | |
417e7e66 | 2616 | if (ovs_list_is_singleton(&all_udpifs)) { |
d72eff6c | 2617 | struct udpif *udpif = NULL; |
27f57736 JS |
2618 | size_t len; |
2619 | ||
417e7e66 | 2620 | udpif = OBJECT_CONTAINING(ovs_list_front(&all_udpifs), udpif, list_node); |
27f57736 JS |
2621 | len = (udpif->n_conns + 1) * sizeof *udpif->conns; |
2622 | udpif->conn_seq = seq_read(udpif->dump_seq); | |
2623 | udpif->conns = xrealloc(udpif->conns, len); | |
2624 | udpif->conns[udpif->n_conns++] = conn; | |
2625 | } else { | |
2626 | unixctl_command_reply_error(conn, "can't wait on multiple udpifs."); | |
2627 | } | |
2628 | } | |
98bb4286 JS |
2629 | |
2630 | static void | |
2631 | upcall_unixctl_purge(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2632 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
2633 | { | |
2634 | struct udpif *udpif; | |
2635 | ||
2636 | LIST_FOR_EACH (udpif, list_node, &all_udpifs) { | |
2637 | int n; | |
2638 | ||
2639 | for (n = 0; n < udpif->n_revalidators; n++) { | |
2640 | revalidator_purge(&udpif->revalidators[n]); | |
2641 | } | |
2642 | } | |
2643 | unixctl_command_reply(conn, ""); | |
2644 | } |