]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e09ee259 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
9583bc14 | 19 | #include "ofproto/ofproto-dpif.h" |
5bee6e26 | 20 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
21 | |
22 | #include <errno.h> | |
23 | ||
ccc09689 | 24 | #include "bfd.h" |
abe529af | 25 | #include "bond.h" |
daff3353 | 26 | #include "bundle.h" |
abe529af BP |
27 | #include "byte-order.h" |
28 | #include "connmgr.h" | |
29 | #include "coverage.h" | |
30 | #include "cfm.h" | |
31 | #include "dpif.h" | |
32 | #include "dynamic-string.h" | |
33 | #include "fail-open.h" | |
34 | #include "hmapx.h" | |
35 | #include "lacp.h" | |
75a75043 | 36 | #include "learn.h" |
abe529af | 37 | #include "mac-learning.h" |
816fd533 | 38 | #include "meta-flow.h" |
abe529af | 39 | #include "multipath.h" |
0a740f48 | 40 | #include "netdev-vport.h" |
abe529af BP |
41 | #include "netdev.h" |
42 | #include "netlink.h" | |
43 | #include "nx-match.h" | |
44 | #include "odp-util.h" | |
1ac7c9bd | 45 | #include "odp-execute.h" |
abe529af BP |
46 | #include "ofp-util.h" |
47 | #include "ofpbuf.h" | |
f25d0cf3 | 48 | #include "ofp-actions.h" |
31a19d69 | 49 | #include "ofp-parse.h" |
abe529af | 50 | #include "ofp-print.h" |
9d6ac44e | 51 | #include "ofproto-dpif-governor.h" |
29089a54 | 52 | #include "ofproto-dpif-ipfix.h" |
ec7ceaed | 53 | #include "ofproto-dpif-mirror.h" |
bae473fe | 54 | #include "ofproto-dpif-sflow.h" |
9583bc14 | 55 | #include "ofproto-dpif-xlate.h" |
abe529af | 56 | #include "poll-loop.h" |
0d085684 | 57 | #include "simap.h" |
27022416 | 58 | #include "smap.h" |
abe529af | 59 | #include "timer.h" |
b9ad7294 | 60 | #include "tunnel.h" |
6c1491fb | 61 | #include "unaligned.h" |
abe529af BP |
62 | #include "unixctl.h" |
63 | #include "vlan-bitmap.h" | |
64 | #include "vlog.h" | |
65 | ||
66 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
67 | ||
abe529af | 68 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af | 69 | COVERAGE_DEFINE(facet_changed_rule); |
abe529af BP |
70 | COVERAGE_DEFINE(facet_revalidate); |
71 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 72 | COVERAGE_DEFINE(facet_suppress); |
9b1a48c2 | 73 | COVERAGE_DEFINE(subfacet_install_fail); |
ada3a58d | 74 | COVERAGE_DEFINE(packet_in_overflow); |
3d9c5e58 | 75 | COVERAGE_DEFINE(flow_mod_overflow); |
abe529af | 76 | |
46c88433 EJ |
77 | /* Number of implemented OpenFlow tables. */ |
78 | enum { N_TABLES = 255 }; | |
79 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
80 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
81 | ||
a088a1ff | 82 | struct flow_miss; |
ac35f9c8 | 83 | struct facet; |
abe529af | 84 | |
29901626 | 85 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
bcd2633a JP |
86 | const struct flow *, |
87 | struct flow_wildcards *wc); | |
abe529af | 88 | |
7e741ffb | 89 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
b0f7b9b5 | 90 | |
46c88433 EJ |
91 | struct ofbundle { |
92 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
93 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
94 | void *aux; /* Key supplied by ofproto's client. */ | |
95 | char *name; /* Identifier for log messages. */ | |
96 | ||
97 | /* Configuration. */ | |
98 | struct list ports; /* Contains "struct ofport"s. */ | |
99 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
100 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
101 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
102 | * NULL if all VLANs are trunked. */ | |
103 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
104 | struct bond *bond; /* Nonnull iff more than one port. */ | |
105 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
106 | ||
107 | /* Status. */ | |
108 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
109 | }; | |
110 | ||
abe529af | 111 | static void bundle_remove(struct ofport *); |
7bde8dd8 | 112 | static void bundle_update(struct ofbundle *); |
abe529af BP |
113 | static void bundle_destroy(struct ofbundle *); |
114 | static void bundle_del_port(struct ofport_dpif *); | |
115 | static void bundle_run(struct ofbundle *); | |
116 | static void bundle_wait(struct ofbundle *); | |
33158a18 | 117 | |
21f7563c JP |
118 | static void stp_run(struct ofproto_dpif *ofproto); |
119 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
120 | static int set_stp_port(struct ofport *, |
121 | const struct ofproto_port_stp_settings *); | |
21f7563c | 122 | |
6a7e895f BP |
123 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, |
124 | enum slow_path_reason, | |
125 | uint64_t *stub, size_t stub_size, | |
126 | const struct nlattr **actionsp, | |
127 | size_t *actions_lenp); | |
128 | ||
129 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
130 | * states: | |
131 | * | |
132 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
133 | * case just after the subfacet is created, just before the subfacet is | |
134 | * destroyed, or if the datapath returns an error when we try to install a | |
135 | * subfacet. | |
136 | * | |
137 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
138 | * | |
139 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
140 | * ofproto_dpif is installed in the datapath. | |
141 | */ | |
142 | enum subfacet_path { | |
143 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
144 | SF_FAST_PATH, /* Full actions are installed. */ | |
145 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
146 | }; | |
147 | ||
5f5fbd17 BP |
148 | /* A dpif flow and actions associated with a facet. |
149 | * | |
150 | * See also the large comment on struct facet. */ | |
151 | struct subfacet { | |
152 | /* Owners. */ | |
153 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
154 | struct list list_node; /* In struct facet's 'facets' list. */ | |
155 | struct facet *facet; /* Owning facet. */ | |
04d08d54 | 156 | struct dpif_backer *backer; /* Owning backer. */ |
5f5fbd17 | 157 | |
5f5fbd17 BP |
158 | enum odp_key_fitness key_fitness; |
159 | struct nlattr *key; | |
160 | int key_len; | |
161 | ||
162 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 163 | long long int created; /* Time created. */ |
5f5fbd17 BP |
164 | |
165 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
166 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
167 | ||
6a7e895f | 168 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 BP |
169 | }; |
170 | ||
1d85f9e5 JP |
171 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
172 | ||
a088a1ff | 173 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 174 | long long int now); |
04d08d54 | 175 | static struct subfacet *subfacet_find(struct dpif_backer *, |
acf60855 | 176 | const struct nlattr *key, size_t key_len, |
9566abf9 | 177 | uint32_t key_hash); |
5f5fbd17 BP |
178 | static void subfacet_destroy(struct subfacet *); |
179 | static void subfacet_destroy__(struct subfacet *); | |
04d08d54 | 180 | static void subfacet_destroy_batch(struct dpif_backer *, |
1d85f9e5 | 181 | struct subfacet **, int n); |
5f5fbd17 BP |
182 | static void subfacet_reset_dp_stats(struct subfacet *, |
183 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
184 | static void subfacet_update_stats(struct subfacet *, |
185 | const struct dpif_flow_stats *); | |
5f5fbd17 | 186 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
187 | const struct ofpbuf *odp_actions, |
188 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
189 | static void subfacet_uninstall(struct subfacet *); |
190 | ||
bcd2633a | 191 | /* A unique, non-overlapping instantiation of an OpenFlow flow. |
b0f7b9b5 BP |
192 | * |
193 | * A facet associates a "struct flow", which represents the Open vSwitch | |
bcd2633a JP |
194 | * userspace idea of an exact-match flow, with one or more subfacets. |
195 | * While the facet is created based on an exact-match flow, it is stored | |
196 | * within the ofproto based on the wildcards that could be expressed | |
197 | * based on the flow table and other configuration. (See the 'wc' | |
198 | * description in "struct xlate_out" for more details.) | |
b0f7b9b5 | 199 | * |
bcd2633a JP |
200 | * Each subfacet tracks the datapath's idea of the flow equivalent to |
201 | * the facet. When the kernel module (or other dpif implementation) and | |
202 | * Open vSwitch userspace agree on the definition of a flow key, there | |
203 | * is exactly one subfacet per facet. If the dpif implementation | |
204 | * supports more-specific flow matching than userspace, however, a facet | |
205 | * can have more than one subfacet. Examples include the dpif | |
206 | * implementation not supporting the same wildcards as userspace or some | |
207 | * distinction in flow that userspace simply doesn't understand. | |
208 | * | |
209 | * Flow expiration works in terms of subfacets, so a facet must have at | |
210 | * least one subfacet or it will never expire, leaking memory. */ | |
abe529af | 211 | struct facet { |
b0f7b9b5 BP |
212 | /* Owners. */ |
213 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
b7e2f6b3 | 214 | struct ofproto_dpif *ofproto; |
b0f7b9b5 BP |
215 | |
216 | /* Owned data. */ | |
217 | struct list subfacets; | |
abe529af BP |
218 | long long int used; /* Time last used; time created if not used. */ |
219 | ||
b0f7b9b5 | 220 | /* Key. */ |
bcd2633a JP |
221 | struct flow flow; /* Flow of the creating subfacet. */ |
222 | struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */ | |
b0f7b9b5 | 223 | |
abe529af BP |
224 | /* These statistics: |
225 | * | |
226 | * - Do include packets and bytes sent "by hand", e.g. with | |
227 | * dpif_execute(). | |
228 | * | |
229 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 230 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 231 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
232 | * |
233 | * - Do not include packets or bytes that can be obtained from the | |
234 | * datapath for any existing subfacet. | |
abe529af BP |
235 | */ |
236 | uint64_t packet_count; /* Number of packets received. */ | |
237 | uint64_t byte_count; /* Number of bytes received. */ | |
238 | ||
b0f7b9b5 | 239 | /* Resubmit statistics. */ |
9d24de3b JP |
240 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
241 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
242 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 243 | |
b0f7b9b5 | 244 | /* Accounting. */ |
907a4c5e | 245 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 246 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 247 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 248 | |
bbafd73b | 249 | struct xlate_out xout; |
b7e2f6b3 | 250 | bool fail_open; /* Facet matched the fail open rule. */ |
4dff9097 | 251 | |
26cd7e34 BP |
252 | /* Storage for a single subfacet, to reduce malloc() time and space |
253 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
254 | * case has exactly one subfacet. However, 'one_subfacet' may not |
255 | * always be valid, since it could have been removed after newer | |
256 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 257 | struct subfacet one_subfacet; |
6cf474d7 EJ |
258 | |
259 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
260 | }; |
261 | ||
bcd2633a JP |
262 | static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *, |
263 | struct xlate_out *, | |
264 | struct dpif_flow_stats *); | |
15baa734 | 265 | static void facet_remove(struct facet *); |
abe529af BP |
266 | static void facet_free(struct facet *); |
267 | ||
bcd2633a | 268 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); |
abe529af | 269 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
bcd2633a | 270 | const struct flow *); |
5bf64ade | 271 | static bool facet_revalidate(struct facet *); |
6814e51f | 272 | static bool facet_check_consistency(struct facet *); |
abe529af | 273 | |
15baa734 | 274 | static void facet_flush_stats(struct facet *); |
abe529af | 275 | |
bbb5d219 | 276 | static void facet_reset_counters(struct facet *); |
9dfb1f78 | 277 | static void facet_push_stats(struct facet *, bool may_learn); |
3de9590b BP |
278 | static void facet_learn(struct facet *); |
279 | static void facet_account(struct facet *); | |
8844e035 | 280 | static void push_all_stats(void); |
abe529af BP |
281 | |
282 | static bool facet_is_controller_flow(struct facet *); | |
283 | ||
46c88433 EJ |
284 | struct ofport_dpif { |
285 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ | |
286 | struct ofport up; | |
287 | ||
288 | odp_port_t odp_port; | |
289 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
290 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
291 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
292 | struct bfd *bfd; /* BFD, if any. */ | |
46c88433 EJ |
293 | bool may_enable; /* May be enabled in bonds. */ |
294 | bool is_tunnel; /* This port is a tunnel. */ | |
295 | long long int carrier_seq; /* Carrier status changes. */ | |
296 | struct ofport_dpif *peer; /* Peer if patch port. */ | |
297 | ||
298 | /* Spanning tree. */ | |
299 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
300 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
301 | long long int stp_state_entered; | |
302 | ||
55954f6e EJ |
303 | /* Queue to DSCP mapping. */ |
304 | struct ofproto_port_queue *qdscp; | |
305 | size_t n_qdscp; | |
46c88433 EJ |
306 | |
307 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
308 | * | |
309 | * This is deprecated. It is only for compatibility with broken device | |
310 | * drivers in old versions of Linux that do not properly support VLANs when | |
311 | * VLAN devices are not used. When broken device drivers are no longer in | |
312 | * widespread use, we will delete these interfaces. */ | |
313 | ofp_port_t realdev_ofp_port; | |
314 | int vlandev_vid; | |
315 | }; | |
316 | ||
52a90c29 BP |
317 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
318 | * | |
319 | * This is deprecated. It is only for compatibility with broken device drivers | |
320 | * in old versions of Linux that do not properly support VLANs when VLAN | |
321 | * devices are not used. When broken device drivers are no longer in | |
322 | * widespread use, we will delete these interfaces. */ | |
323 | struct vlan_splinter { | |
324 | struct hmap_node realdev_vid_node; | |
325 | struct hmap_node vlandev_node; | |
4e022ec0 AW |
326 | ofp_port_t realdev_ofp_port; |
327 | ofp_port_t vlandev_ofp_port; | |
52a90c29 BP |
328 | int vid; |
329 | }; | |
330 | ||
52a90c29 | 331 | static void vsp_remove(struct ofport_dpif *); |
4e022ec0 | 332 | static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid); |
52a90c29 | 333 | |
46c88433 EJ |
334 | static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
335 | ofp_port_t); | |
336 | ||
4e022ec0 | 337 | static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *, |
46c88433 | 338 | odp_port_t); |
e1b1d06a | 339 | |
abe529af BP |
340 | static struct ofport_dpif * |
341 | ofport_dpif_cast(const struct ofport *ofport) | |
342 | { | |
abe529af BP |
343 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
344 | } | |
345 | ||
346 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 347 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 348 | static void port_wait(struct ofport_dpif *); |
8aee94b6 | 349 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 350 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
6cbbf4fa | 351 | static void ofport_update_peer(struct ofport_dpif *); |
8fa4d1d0 | 352 | static void run_fast_rl(void); |
3d9c5e58 | 353 | static int run_fast(struct ofproto *); |
abe529af | 354 | |
7ee20df1 BP |
355 | struct dpif_completion { |
356 | struct list list_node; | |
357 | struct ofoperation *op; | |
358 | }; | |
359 | ||
cdc3ab65 EJ |
360 | /* Reasons that we might need to revalidate every facet, and corresponding |
361 | * coverage counters. | |
362 | * | |
363 | * A value of 0 means that there is no need to revalidate. | |
364 | * | |
365 | * It would be nice to have some cleaner way to integrate with coverage | |
366 | * counters, but with only a few reasons I guess this is good enough for | |
367 | * now. */ | |
368 | enum revalidate_reason { | |
369 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
370 | REV_STP, /* Spanning tree protocol port status change. */ | |
4a1b8f30 | 371 | REV_BOND, /* Bonding changed. */ |
cdc3ab65 EJ |
372 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ |
373 | REV_FLOW_TABLE, /* Flow table changed. */ | |
30618594 | 374 | REV_MAC_LEARNING, /* Mac learning changed. */ |
cdc3ab65 EJ |
375 | REV_INCONSISTENCY /* Facet self-check failed. */ |
376 | }; | |
3c4a309c BP |
377 | COVERAGE_DEFINE(rev_reconfigure); |
378 | COVERAGE_DEFINE(rev_stp); | |
4a1b8f30 | 379 | COVERAGE_DEFINE(rev_bond); |
3c4a309c BP |
380 | COVERAGE_DEFINE(rev_port_toggled); |
381 | COVERAGE_DEFINE(rev_flow_table); | |
30618594 | 382 | COVERAGE_DEFINE(rev_mac_learning); |
3c4a309c BP |
383 | COVERAGE_DEFINE(rev_inconsistency); |
384 | ||
8f73d537 EJ |
385 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
386 | * These are datapath flows which have no associated ofproto, if they did we | |
387 | * would use facets. */ | |
388 | struct drop_key { | |
389 | struct hmap_node hmap_node; | |
390 | struct nlattr *key; | |
391 | size_t key_len; | |
392 | }; | |
393 | ||
cdc3ab65 EJ |
394 | struct avg_subfacet_rates { |
395 | double add_rate; /* Moving average of new flows created per minute. */ | |
396 | double del_rate; /* Moving average of flows deleted per minute. */ | |
397 | }; | |
398 | ||
399 | /* All datapaths of a given type share a single dpif backer instance. */ | |
400 | struct dpif_backer { | |
401 | char *type; | |
402 | int refcount; | |
403 | struct dpif *dpif; | |
404 | struct timer next_expiration; | |
8449c4d6 EJ |
405 | |
406 | struct ovs_rwlock odp_to_ofport_lock; | |
407 | struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */ | |
cdc3ab65 EJ |
408 | |
409 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ | |
410 | ||
411 | /* Facet revalidation flags applying to facets which use this backer. */ | |
412 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
cdc3ab65 EJ |
413 | |
414 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
415 | bool recv_set_enable; /* Enables or disables receiving packets. */ | |
416 | ||
417 | struct hmap subfacets; | |
418 | struct governor *governor; | |
419 | ||
420 | /* Subfacet statistics. | |
421 | * | |
422 | * These keep track of the total number of subfacets added and deleted and | |
423 | * flow life span. They are useful for computing the flow rates stats | |
424 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
425 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
426 | * performance in new situations. */ | |
427 | long long int created; /* Time when it is created. */ | |
428 | unsigned max_n_subfacet; /* Maximum number of flows */ | |
429 | unsigned avg_n_subfacet; /* Average number of flows. */ | |
430 | long long int avg_subfacet_life; /* Average life span of subfacets. */ | |
431 | ||
432 | /* The average number of subfacets... */ | |
433 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
434 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
435 | struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */ | |
436 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
437 | ||
438 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
439 | unsigned subfacet_add_count; | |
440 | unsigned subfacet_del_count; | |
441 | ||
442 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
443 | unsigned long long int total_subfacet_add_count; | |
444 | unsigned long long int total_subfacet_del_count; | |
445 | }; | |
446 | ||
acf60855 JP |
447 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ |
448 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
449 | ||
8f73d537 | 450 | static void drop_key_clear(struct dpif_backer *); |
dc54ef36 | 451 | static void update_moving_averages(struct dpif_backer *backer); |
735d7efb | 452 | |
46c88433 EJ |
453 | struct ofproto_dpif { |
454 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
455 | struct ofproto up; | |
456 | struct dpif_backer *backer; | |
457 | ||
458 | /* Special OpenFlow rules. */ | |
459 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
460 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
461 | struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ | |
462 | ||
463 | /* Bridging. */ | |
464 | struct netflow *netflow; | |
465 | struct dpif_sflow *sflow; | |
466 | struct dpif_ipfix *ipfix; | |
467 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
468 | struct mac_learning *ml; | |
469 | bool has_bonded_bundles; | |
470 | struct mbridge *mbridge; | |
471 | ||
472 | /* Facets. */ | |
473 | struct classifier facets; /* Contains 'struct facet's. */ | |
474 | long long int consistency_rl; | |
475 | ||
46c88433 EJ |
476 | /* Support for debugging async flow mods. */ |
477 | struct list completions; | |
478 | ||
479 | struct netdev_stats stats; /* To account packets generated and consumed in | |
480 | * userspace. */ | |
481 | ||
482 | /* Spanning tree. */ | |
483 | struct stp *stp; | |
484 | long long int stp_last_tick; | |
485 | ||
486 | /* VLAN splinters. */ | |
13501305 EJ |
487 | struct ovs_mutex vsp_mutex; |
488 | struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */ | |
489 | struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */ | |
46c88433 EJ |
490 | |
491 | /* Ports. */ | |
492 | struct sset ports; /* Set of standard port names. */ | |
493 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
494 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ | |
495 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
496 | ||
497 | /* Per ofproto's dpif stats. */ | |
498 | uint64_t n_hit; | |
499 | uint64_t n_missed; | |
3d9c5e58 EJ |
500 | |
501 | /* Work queues. */ | |
502 | struct ovs_mutex flow_mod_mutex; | |
503 | struct list flow_mods OVS_GUARDED; | |
504 | size_t n_flow_mods OVS_GUARDED; | |
ada3a58d EJ |
505 | |
506 | struct ovs_mutex pin_mutex; | |
507 | struct list pins OVS_GUARDED; | |
508 | size_t n_pins OVS_GUARDED; | |
46c88433 EJ |
509 | }; |
510 | ||
7ee20df1 BP |
511 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
512 | * for debugging the asynchronous flow_mod implementation.) */ | |
513 | static bool clogged; | |
514 | ||
6b35e630 JP |
515 | /* By default, flows in the datapath are wildcarded (megaflows). They |
516 | * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ | |
517 | static bool enable_megaflows = true; | |
518 | ||
b44a10b7 BP |
519 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
520 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
521 | ||
abe529af BP |
522 | static void ofproto_dpif_unixctl_init(void); |
523 | ||
46c88433 EJ |
524 | static inline struct ofproto_dpif * |
525 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
526 | { | |
527 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
528 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
529 | } | |
530 | ||
531 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, | |
532 | ofp_port_t ofp_port); | |
ade6ad9c EJ |
533 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
534 | const struct ofpbuf *packet, struct ds *); | |
46c88433 | 535 | |
501f8d1f BP |
536 | /* Upcalls. */ |
537 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 538 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
539 | |
540 | /* Flow expiration. */ | |
acf60855 | 541 | static int expire(struct dpif_backer *); |
abe529af | 542 | |
6fca1ffb BP |
543 | /* NetFlow. */ |
544 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
545 | ||
abe529af | 546 | /* Utilities. */ |
52a90c29 | 547 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
9583bc14 | 548 | |
abe529af BP |
549 | /* Global variables. */ |
550 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
551 | |
552 | /* Initial mappings of port to bridge mappings. */ | |
553 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
46c88433 | 554 | |
3d9c5e58 EJ |
555 | /* Executes and takes ownership of 'fm'. */ |
556 | void | |
46c88433 EJ |
557 | ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, |
558 | struct ofputil_flow_mod *fm) | |
559 | { | |
3d9c5e58 EJ |
560 | ovs_mutex_lock(&ofproto->flow_mod_mutex); |
561 | if (ofproto->n_flow_mods > 1024) { | |
562 | ovs_mutex_unlock(&ofproto->flow_mod_mutex); | |
563 | COVERAGE_INC(flow_mod_overflow); | |
564 | free(fm->ofpacts); | |
565 | free(fm); | |
566 | return; | |
567 | } | |
568 | ||
569 | list_push_back(&ofproto->flow_mods, &fm->list_node); | |
570 | ofproto->n_flow_mods++; | |
571 | ovs_mutex_unlock(&ofproto->flow_mod_mutex); | |
46c88433 EJ |
572 | } |
573 | ||
574 | void | |
575 | ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, | |
576 | struct ofputil_packet_in *pin) | |
577 | { | |
ada3a58d EJ |
578 | ovs_mutex_lock(&ofproto->pin_mutex); |
579 | if (ofproto->n_pins > 1024) { | |
580 | ovs_mutex_unlock(&ofproto->pin_mutex); | |
581 | COVERAGE_INC(packet_in_overflow); | |
582 | free(CONST_CAST(void *, pin->packet)); | |
583 | free(pin); | |
584 | return; | |
585 | } | |
586 | ||
587 | list_push_back(&ofproto->pins, &pin->list_node); | |
588 | ofproto->n_pins++; | |
589 | ovs_mutex_unlock(&ofproto->pin_mutex); | |
46c88433 | 590 | } |
abe529af BP |
591 | \f |
592 | /* Factory functions. */ | |
593 | ||
b0408fca | 594 | static void |
acf60855 | 595 | init(const struct shash *iface_hints) |
b0408fca | 596 | { |
acf60855 JP |
597 | struct shash_node *node; |
598 | ||
599 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
600 | SHASH_FOR_EACH(node, iface_hints) { | |
601 | const struct iface_hint *orig_hint = node->data; | |
602 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
603 | ||
604 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
605 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
606 | new_hint->ofp_port = orig_hint->ofp_port; | |
607 | ||
608 | shash_add(&init_ofp_ports, node->name, new_hint); | |
609 | } | |
b0408fca JP |
610 | } |
611 | ||
abe529af BP |
612 | static void |
613 | enumerate_types(struct sset *types) | |
614 | { | |
615 | dp_enumerate_types(types); | |
616 | } | |
617 | ||
618 | static int | |
619 | enumerate_names(const char *type, struct sset *names) | |
620 | { | |
acf60855 JP |
621 | struct ofproto_dpif *ofproto; |
622 | ||
623 | sset_clear(names); | |
624 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
625 | if (strcmp(type, ofproto->up.type)) { | |
626 | continue; | |
627 | } | |
628 | sset_add(names, ofproto->up.name); | |
629 | } | |
630 | ||
631 | return 0; | |
abe529af BP |
632 | } |
633 | ||
634 | static int | |
635 | del(const char *type, const char *name) | |
636 | { | |
637 | struct dpif *dpif; | |
638 | int error; | |
639 | ||
640 | error = dpif_open(name, type, &dpif); | |
641 | if (!error) { | |
642 | error = dpif_delete(dpif); | |
643 | dpif_close(dpif); | |
644 | } | |
645 | return error; | |
646 | } | |
647 | \f | |
0aeaabc8 JP |
648 | static const char * |
649 | port_open_type(const char *datapath_type, const char *port_type) | |
650 | { | |
651 | return dpif_port_open_type(datapath_type, port_type); | |
652 | } | |
653 | ||
acf60855 JP |
654 | /* Type functions. */ |
655 | ||
36beb9be BP |
656 | static void process_dpif_port_changes(struct dpif_backer *); |
657 | static void process_dpif_all_ports_changed(struct dpif_backer *); | |
658 | static void process_dpif_port_change(struct dpif_backer *, | |
659 | const char *devname); | |
660 | static void process_dpif_port_error(struct dpif_backer *, int error); | |
661 | ||
476cb42a BP |
662 | static struct ofproto_dpif * |
663 | lookup_ofproto_dpif_by_port_name(const char *name) | |
664 | { | |
665 | struct ofproto_dpif *ofproto; | |
666 | ||
667 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
668 | if (sset_contains(&ofproto->ports, name)) { | |
669 | return ofproto; | |
670 | } | |
671 | } | |
672 | ||
673 | return NULL; | |
674 | } | |
675 | ||
acf60855 JP |
676 | static int |
677 | type_run(const char *type) | |
678 | { | |
e20b5746 | 679 | static long long int push_timer = LLONG_MIN; |
acf60855 | 680 | struct dpif_backer *backer; |
acf60855 JP |
681 | |
682 | backer = shash_find_data(&all_dpif_backers, type); | |
683 | if (!backer) { | |
684 | /* This is not necessarily a problem, since backers are only | |
685 | * created on demand. */ | |
686 | return 0; | |
687 | } | |
688 | ||
689 | dpif_run(backer->dpif); | |
690 | ||
e20b5746 EJ |
691 | /* The most natural place to push facet statistics is when they're pulled |
692 | * from the datapath. However, when there are many flows in the datapath, | |
693 | * this expensive operation can occur so frequently, that it reduces our | |
694 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
695 | * here instead. */ | |
696 | if (time_msec() > push_timer) { | |
697 | push_timer = time_msec() + 2000; | |
698 | push_all_stats(); | |
699 | } | |
700 | ||
40358701 GS |
701 | /* If vswitchd started with other_config:flow_restore_wait set as "true", |
702 | * and the configuration has now changed to "false", enable receiving | |
703 | * packets from the datapath. */ | |
704 | if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { | |
36beb9be BP |
705 | int error; |
706 | ||
40358701 GS |
707 | backer->recv_set_enable = true; |
708 | ||
709 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); | |
710 | if (error) { | |
711 | VLOG_ERR("Failed to enable receiving packets in dpif."); | |
712 | return error; | |
713 | } | |
714 | dpif_flow_flush(backer->dpif); | |
715 | backer->need_revalidate = REV_RECONFIGURE; | |
716 | } | |
717 | ||
a1616063 | 718 | if (backer->need_revalidate) { |
2cc3c58e | 719 | struct ofproto_dpif *ofproto; |
a614d823 KM |
720 | struct simap_node *node; |
721 | struct simap tmp_backers; | |
722 | ||
723 | /* Handle tunnel garbage collection. */ | |
724 | simap_init(&tmp_backers); | |
725 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
726 | ||
727 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
728 | struct ofport_dpif *iter; | |
729 | ||
730 | if (backer != ofproto->backer) { | |
731 | continue; | |
732 | } | |
733 | ||
734 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
3aa30359 | 735 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
a614d823 KM |
736 | const char *dp_port; |
737 | ||
42943cde | 738 | if (!iter->is_tunnel) { |
a614d823 KM |
739 | continue; |
740 | } | |
741 | ||
3aa30359 BP |
742 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev, |
743 | namebuf, sizeof namebuf); | |
a614d823 KM |
744 | node = simap_find(&tmp_backers, dp_port); |
745 | if (node) { | |
746 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
747 | simap_delete(&tmp_backers, node); | |
748 | node = simap_find(&backer->tnl_backers, dp_port); | |
749 | } else { | |
750 | node = simap_find(&backer->tnl_backers, dp_port); | |
751 | if (!node) { | |
4e022ec0 | 752 | odp_port_t odp_port = ODPP_NONE; |
a614d823 KM |
753 | |
754 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
755 | &odp_port)) { | |
4e022ec0 AW |
756 | simap_put(&backer->tnl_backers, dp_port, |
757 | odp_to_u32(odp_port)); | |
a614d823 KM |
758 | node = simap_find(&backer->tnl_backers, dp_port); |
759 | } | |
760 | } | |
761 | } | |
762 | ||
4e022ec0 | 763 | iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE; |
42943cde EJ |
764 | if (tnl_port_reconfigure(iter, iter->up.netdev, |
765 | iter->odp_port)) { | |
a614d823 KM |
766 | backer->need_revalidate = REV_RECONFIGURE; |
767 | } | |
768 | } | |
769 | } | |
770 | ||
771 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
4e022ec0 | 772 | dpif_port_del(backer->dpif, u32_to_odp(node->data)); |
a614d823 KM |
773 | } |
774 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
775 | |
776 | switch (backer->need_revalidate) { | |
777 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
778 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
4a1b8f30 | 779 | case REV_BOND: COVERAGE_INC(rev_bond); break; |
2cc3c58e EJ |
780 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; |
781 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
30618594 | 782 | case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break; |
2cc3c58e EJ |
783 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; |
784 | } | |
f728af2e BP |
785 | backer->need_revalidate = 0; |
786 | ||
a1616063 EJ |
787 | /* Clear the drop_keys in case we should now be accepting some |
788 | * formerly dropped flows. */ | |
789 | drop_key_clear(backer); | |
790 | ||
2cc3c58e | 791 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 792 | struct facet *facet, *next; |
a1616063 | 793 | struct ofport_dpif *ofport; |
bcd2633a | 794 | struct cls_cursor cursor; |
a1616063 | 795 | struct ofbundle *bundle; |
2cc3c58e EJ |
796 | |
797 | if (ofproto->backer != backer) { | |
798 | continue; | |
799 | } | |
800 | ||
dc24a00f | 801 | ovs_rwlock_wrlock(&xlate_rwlock); |
89a8a7f0 | 802 | xlate_ofproto_set(ofproto, ofproto->up.name, |
ec89fc6f EJ |
803 | ofproto->backer->dpif, ofproto->miss_rule, |
804 | ofproto->no_packet_in_rule, ofproto->ml, | |
9d189a50 EJ |
805 | ofproto->stp, ofproto->mbridge, |
806 | ofproto->sflow, ofproto->ipfix, | |
807 | ofproto->up.frag_handling, | |
a1616063 EJ |
808 | ofproto->up.forward_bpdu, |
809 | connmgr_has_in_band(ofproto->up.connmgr), | |
9d189a50 | 810 | ofproto->netflow != NULL); |
46c88433 | 811 | |
a1616063 EJ |
812 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
813 | xlate_bundle_set(ofproto, bundle, bundle->name, | |
814 | bundle->vlan_mode, bundle->vlan, | |
815 | bundle->trunks, bundle->use_priority_tags, | |
816 | bundle->bond, bundle->lacp, | |
817 | bundle->floodable); | |
818 | } | |
819 | ||
820 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
9d189a50 EJ |
821 | int stp_port = ofport->stp_port |
822 | ? stp_port_no(ofport->stp_port) | |
823 | : 0; | |
a1616063 EJ |
824 | xlate_ofport_set(ofproto, ofport->bundle, ofport, |
825 | ofport->up.ofp_port, ofport->odp_port, | |
826 | ofport->up.netdev, ofport->cfm, | |
9d189a50 | 827 | ofport->bfd, ofport->peer, stp_port, |
55954f6e | 828 | ofport->qdscp, ofport->n_qdscp, |
9d189a50 EJ |
829 | ofport->up.pp.config, ofport->is_tunnel, |
830 | ofport->may_enable); | |
46c88433 | 831 | } |
dc24a00f | 832 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 833 | |
0b4f2078 EJ |
834 | /* Only ofproto-dpif cares about the facet classifier so we just |
835 | * lock cls_cursor_init() to appease the thread safety analysis. */ | |
836 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
bcd2633a | 837 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
0b4f2078 | 838 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 839 | CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { |
a1616063 EJ |
840 | facet_revalidate(facet); |
841 | run_fast_rl(); | |
2cc3c58e EJ |
842 | } |
843 | } | |
2cc3c58e EJ |
844 | } |
845 | ||
40358701 GS |
846 | if (!backer->recv_set_enable) { |
847 | /* Wake up before a max of 1000ms. */ | |
848 | timer_set_duration(&backer->next_expiration, 1000); | |
849 | } else if (timer_expired(&backer->next_expiration)) { | |
acf60855 JP |
850 | int delay = expire(backer); |
851 | timer_set_duration(&backer->next_expiration, delay); | |
852 | } | |
853 | ||
36beb9be | 854 | process_dpif_port_changes(backer); |
acf60855 | 855 | |
04d08d54 EJ |
856 | if (backer->governor) { |
857 | size_t n_subfacets; | |
858 | ||
859 | governor_run(backer->governor); | |
860 | ||
861 | /* If the governor has shrunk to its minimum size and the number of | |
862 | * subfacets has dwindled, then drop the governor entirely. | |
863 | * | |
864 | * For hysteresis, the number of subfacets to drop the governor is | |
865 | * smaller than the number needed to trigger its creation. */ | |
866 | n_subfacets = hmap_count(&backer->subfacets); | |
867 | if (n_subfacets * 4 < flow_eviction_threshold | |
868 | && governor_is_idle(backer->governor)) { | |
869 | governor_destroy(backer->governor); | |
870 | backer->governor = NULL; | |
871 | } | |
872 | } | |
873 | ||
acf60855 JP |
874 | return 0; |
875 | } | |
876 | ||
36beb9be BP |
877 | /* Check for and handle port changes in 'backer''s dpif. */ |
878 | static void | |
879 | process_dpif_port_changes(struct dpif_backer *backer) | |
880 | { | |
881 | for (;;) { | |
882 | char *devname; | |
883 | int error; | |
884 | ||
885 | error = dpif_port_poll(backer->dpif, &devname); | |
886 | switch (error) { | |
887 | case EAGAIN: | |
888 | return; | |
889 | ||
890 | case ENOBUFS: | |
891 | process_dpif_all_ports_changed(backer); | |
892 | break; | |
893 | ||
894 | case 0: | |
895 | process_dpif_port_change(backer, devname); | |
896 | free(devname); | |
897 | break; | |
898 | ||
899 | default: | |
900 | process_dpif_port_error(backer, error); | |
901 | break; | |
902 | } | |
903 | } | |
904 | } | |
905 | ||
906 | static void | |
907 | process_dpif_all_ports_changed(struct dpif_backer *backer) | |
908 | { | |
909 | struct ofproto_dpif *ofproto; | |
910 | struct dpif_port dpif_port; | |
911 | struct dpif_port_dump dump; | |
912 | struct sset devnames; | |
913 | const char *devname; | |
914 | ||
915 | sset_init(&devnames); | |
916 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
917 | if (ofproto->backer == backer) { | |
918 | struct ofport *ofport; | |
919 | ||
920 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
921 | sset_add(&devnames, netdev_get_name(ofport->netdev)); | |
922 | } | |
923 | } | |
924 | } | |
925 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) { | |
926 | sset_add(&devnames, dpif_port.name); | |
927 | } | |
928 | ||
929 | SSET_FOR_EACH (devname, &devnames) { | |
930 | process_dpif_port_change(backer, devname); | |
931 | } | |
932 | sset_destroy(&devnames); | |
933 | } | |
934 | ||
935 | static void | |
936 | process_dpif_port_change(struct dpif_backer *backer, const char *devname) | |
937 | { | |
938 | struct ofproto_dpif *ofproto; | |
939 | struct dpif_port port; | |
940 | ||
941 | /* Don't report on the datapath's device. */ | |
942 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
943 | return; | |
944 | } | |
945 | ||
946 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
947 | &all_ofproto_dpifs) { | |
948 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { | |
949 | return; | |
950 | } | |
951 | } | |
952 | ||
953 | ofproto = lookup_ofproto_dpif_by_port_name(devname); | |
954 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { | |
955 | /* The port was removed. If we know the datapath, | |
956 | * report it through poll_set(). If we don't, it may be | |
957 | * notifying us of a removal we initiated, so ignore it. | |
958 | * If there's a pending ENOBUFS, let it stand, since | |
959 | * everything will be reevaluated. */ | |
960 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
961 | sset_add(&ofproto->port_poll_set, devname); | |
962 | ofproto->port_poll_errno = 0; | |
963 | } | |
964 | } else if (!ofproto) { | |
965 | /* The port was added, but we don't know with which | |
966 | * ofproto we should associate it. Delete it. */ | |
967 | dpif_port_del(backer->dpif, port.port_no); | |
74cc3969 BP |
968 | } else { |
969 | struct ofport_dpif *ofport; | |
970 | ||
971 | ofport = ofport_dpif_cast(shash_find_data( | |
972 | &ofproto->up.port_by_name, devname)); | |
973 | if (ofport | |
974 | && ofport->odp_port != port.port_no | |
975 | && !odp_port_to_ofport(backer, port.port_no)) | |
976 | { | |
977 | /* 'ofport''s datapath port number has changed from | |
978 | * 'ofport->odp_port' to 'port.port_no'. Update our internal data | |
979 | * structures to match. */ | |
8449c4d6 | 980 | ovs_rwlock_wrlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
981 | hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node); |
982 | ofport->odp_port = port.port_no; | |
983 | hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node, | |
984 | hash_odp_port(port.port_no)); | |
8449c4d6 | 985 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
986 | backer->need_revalidate = REV_RECONFIGURE; |
987 | } | |
36beb9be BP |
988 | } |
989 | dpif_port_destroy(&port); | |
990 | } | |
991 | ||
992 | /* Propagate 'error' to all ofprotos based on 'backer'. */ | |
993 | static void | |
994 | process_dpif_port_error(struct dpif_backer *backer, int error) | |
995 | { | |
996 | struct ofproto_dpif *ofproto; | |
997 | ||
998 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
999 | if (ofproto->backer == backer) { | |
1000 | sset_clear(&ofproto->port_poll_set); | |
1001 | ofproto->port_poll_errno = error; | |
1002 | } | |
1003 | } | |
1004 | } | |
1005 | ||
acf60855 | 1006 | static int |
8fa4d1d0 | 1007 | dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) |
acf60855 | 1008 | { |
acf60855 JP |
1009 | unsigned int work; |
1010 | ||
40358701 GS |
1011 | /* If recv_set_enable is false, we should not handle upcalls. */ |
1012 | if (!backer->recv_set_enable) { | |
1013 | return 0; | |
1014 | } | |
1015 | ||
acf60855 JP |
1016 | /* Handle one or more batches of upcalls, until there's nothing left to do |
1017 | * or until we do a fixed total amount of work. | |
1018 | * | |
1019 | * We do work in batches because it can be much cheaper to set up a number | |
1020 | * of flows and fire off their patches all at once. We do multiple batches | |
1021 | * because in some cases handling a packet can cause another packet to be | |
1022 | * queued almost immediately as part of the return flow. Both | |
1023 | * optimizations can make major improvements on some benchmarks and | |
1024 | * presumably for real traffic as well. */ | |
1025 | work = 0; | |
8fa4d1d0 EJ |
1026 | while (work < max_batch) { |
1027 | int retval = handle_upcalls(backer, max_batch - work); | |
acf60855 JP |
1028 | if (retval <= 0) { |
1029 | return -retval; | |
1030 | } | |
1031 | work += retval; | |
1032 | } | |
1033 | ||
1034 | return 0; | |
1035 | } | |
1036 | ||
8fa4d1d0 EJ |
1037 | static int |
1038 | type_run_fast(const char *type) | |
1039 | { | |
1040 | struct dpif_backer *backer; | |
1041 | ||
1042 | backer = shash_find_data(&all_dpif_backers, type); | |
1043 | if (!backer) { | |
1044 | /* This is not necessarily a problem, since backers are only | |
1045 | * created on demand. */ | |
1046 | return 0; | |
1047 | } | |
1048 | ||
1049 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1050 | } | |
1051 | ||
1052 | static void | |
1053 | run_fast_rl(void) | |
1054 | { | |
1055 | static long long int port_rl = LLONG_MIN; | |
1056 | static unsigned int backer_rl = 0; | |
1057 | ||
1058 | if (time_msec() >= port_rl) { | |
1059 | struct ofproto_dpif *ofproto; | |
8fa4d1d0 EJ |
1060 | |
1061 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3d9c5e58 | 1062 | run_fast(&ofproto->up); |
8fa4d1d0 EJ |
1063 | } |
1064 | port_rl = time_msec() + 200; | |
1065 | } | |
1066 | ||
1067 | /* XXX: We have to be careful not to do too much work in this function. If | |
1068 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1069 | * performance improves signifcantly, but at a cost. It's possible for the | |
1070 | * number of flows in the datapath to increase without bound, and for poll | |
1071 | * loops to take 10s of seconds. The correct solution to this problem, | |
1072 | * long term, is to separate flow miss handling into it's own thread so it | |
1073 | * isn't affected by revalidations, and expirations. Until then, this is | |
1074 | * the best we can do. */ | |
1075 | if (++backer_rl >= 10) { | |
1076 | struct shash_node *node; | |
1077 | ||
1078 | backer_rl = 0; | |
1079 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1080 | dpif_backer_run_fast(node->data, 1); | |
1081 | } | |
1082 | } | |
1083 | } | |
1084 | ||
acf60855 JP |
1085 | static void |
1086 | type_wait(const char *type) | |
1087 | { | |
1088 | struct dpif_backer *backer; | |
1089 | ||
1090 | backer = shash_find_data(&all_dpif_backers, type); | |
1091 | if (!backer) { | |
1092 | /* This is not necessarily a problem, since backers are only | |
1093 | * created on demand. */ | |
1094 | return; | |
1095 | } | |
1096 | ||
04d08d54 EJ |
1097 | if (backer->governor) { |
1098 | governor_wait(backer->governor); | |
1099 | } | |
1100 | ||
acf60855 JP |
1101 | timer_wait(&backer->next_expiration); |
1102 | } | |
1103 | \f | |
abe529af BP |
1104 | /* Basic life-cycle. */ |
1105 | ||
c57b2226 BP |
1106 | static int add_internal_flows(struct ofproto_dpif *); |
1107 | ||
abe529af BP |
1108 | static struct ofproto * |
1109 | alloc(void) | |
1110 | { | |
1111 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1112 | return &ofproto->up; | |
1113 | } | |
1114 | ||
1115 | static void | |
1116 | dealloc(struct ofproto *ofproto_) | |
1117 | { | |
1118 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1119 | free(ofproto); | |
1120 | } | |
1121 | ||
acf60855 JP |
1122 | static void |
1123 | close_dpif_backer(struct dpif_backer *backer) | |
1124 | { | |
1125 | struct shash_node *node; | |
1126 | ||
cb22974d | 1127 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1128 | |
1129 | if (--backer->refcount) { | |
1130 | return; | |
1131 | } | |
1132 | ||
8f73d537 EJ |
1133 | drop_key_clear(backer); |
1134 | hmap_destroy(&backer->drop_keys); | |
1135 | ||
7d82ab2e | 1136 | simap_destroy(&backer->tnl_backers); |
8449c4d6 | 1137 | ovs_rwlock_destroy(&backer->odp_to_ofport_lock); |
acf60855 JP |
1138 | hmap_destroy(&backer->odp_to_ofport_map); |
1139 | node = shash_find(&all_dpif_backers, backer->type); | |
1140 | free(backer->type); | |
1141 | shash_delete(&all_dpif_backers, node); | |
1142 | dpif_close(backer->dpif); | |
1143 | ||
04d08d54 EJ |
1144 | ovs_assert(hmap_is_empty(&backer->subfacets)); |
1145 | hmap_destroy(&backer->subfacets); | |
1146 | governor_destroy(backer->governor); | |
1147 | ||
acf60855 JP |
1148 | free(backer); |
1149 | } | |
1150 | ||
1151 | /* Datapath port slated for removal from datapath. */ | |
1152 | struct odp_garbage { | |
1153 | struct list list_node; | |
4e022ec0 | 1154 | odp_port_t odp_port; |
acf60855 JP |
1155 | }; |
1156 | ||
1157 | static int | |
1158 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1159 | { | |
1160 | struct dpif_backer *backer; | |
1161 | struct dpif_port_dump port_dump; | |
1162 | struct dpif_port port; | |
1163 | struct shash_node *node; | |
1164 | struct list garbage_list; | |
1165 | struct odp_garbage *garbage, *next; | |
1166 | struct sset names; | |
1167 | char *backer_name; | |
1168 | const char *name; | |
1169 | int error; | |
1170 | ||
1171 | backer = shash_find_data(&all_dpif_backers, type); | |
1172 | if (backer) { | |
1173 | backer->refcount++; | |
1174 | *backerp = backer; | |
1175 | return 0; | |
1176 | } | |
1177 | ||
1178 | backer_name = xasprintf("ovs-%s", type); | |
1179 | ||
1180 | /* Remove any existing datapaths, since we assume we're the only | |
1181 | * userspace controlling the datapath. */ | |
1182 | sset_init(&names); | |
1183 | dp_enumerate_names(type, &names); | |
1184 | SSET_FOR_EACH(name, &names) { | |
1185 | struct dpif *old_dpif; | |
1186 | ||
1187 | /* Don't remove our backer if it exists. */ | |
1188 | if (!strcmp(name, backer_name)) { | |
1189 | continue; | |
1190 | } | |
1191 | ||
1192 | if (dpif_open(name, type, &old_dpif)) { | |
1193 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1194 | } else { | |
1195 | dpif_delete(old_dpif); | |
1196 | dpif_close(old_dpif); | |
1197 | } | |
1198 | } | |
1199 | sset_destroy(&names); | |
1200 | ||
1201 | backer = xmalloc(sizeof *backer); | |
1202 | ||
1203 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1204 | free(backer_name); | |
1205 | if (error) { | |
1206 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
10a89ef0 | 1207 | ovs_strerror(error)); |
4c1b1289 | 1208 | free(backer); |
acf60855 JP |
1209 | return error; |
1210 | } | |
1211 | ||
1212 | backer->type = xstrdup(type); | |
04d08d54 | 1213 | backer->governor = NULL; |
acf60855 JP |
1214 | backer->refcount = 1; |
1215 | hmap_init(&backer->odp_to_ofport_map); | |
8449c4d6 | 1216 | ovs_rwlock_init(&backer->odp_to_ofport_lock); |
8f73d537 | 1217 | hmap_init(&backer->drop_keys); |
04d08d54 | 1218 | hmap_init(&backer->subfacets); |
acf60855 | 1219 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1220 | backer->need_revalidate = 0; |
7d82ab2e | 1221 | simap_init(&backer->tnl_backers); |
40358701 | 1222 | backer->recv_set_enable = !ofproto_get_flow_restore_wait(); |
acf60855 JP |
1223 | *backerp = backer; |
1224 | ||
40358701 GS |
1225 | if (backer->recv_set_enable) { |
1226 | dpif_flow_flush(backer->dpif); | |
1227 | } | |
acf60855 JP |
1228 | |
1229 | /* Loop through the ports already on the datapath and remove any | |
1230 | * that we don't need anymore. */ | |
1231 | list_init(&garbage_list); | |
1232 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1233 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1234 | node = shash_find(&init_ofp_ports, port.name); | |
1235 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1236 | garbage = xmalloc(sizeof *garbage); | |
1237 | garbage->odp_port = port.port_no; | |
1238 | list_push_front(&garbage_list, &garbage->list_node); | |
1239 | } | |
1240 | } | |
1241 | dpif_port_dump_done(&port_dump); | |
1242 | ||
1243 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1244 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1245 | list_remove(&garbage->list_node); | |
1246 | free(garbage); | |
1247 | } | |
1248 | ||
1249 | shash_add(&all_dpif_backers, type, backer); | |
1250 | ||
40358701 | 1251 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); |
acf60855 JP |
1252 | if (error) { |
1253 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
10a89ef0 | 1254 | type, ovs_strerror(error)); |
acf60855 JP |
1255 | close_dpif_backer(backer); |
1256 | return error; | |
1257 | } | |
1258 | ||
dc54ef36 EJ |
1259 | backer->max_n_subfacet = 0; |
1260 | backer->created = time_msec(); | |
1261 | backer->last_minute = backer->created; | |
1262 | memset(&backer->hourly, 0, sizeof backer->hourly); | |
1263 | memset(&backer->daily, 0, sizeof backer->daily); | |
1264 | memset(&backer->lifetime, 0, sizeof backer->lifetime); | |
1265 | backer->subfacet_add_count = 0; | |
1266 | backer->subfacet_del_count = 0; | |
1267 | backer->total_subfacet_add_count = 0; | |
1268 | backer->total_subfacet_del_count = 0; | |
1269 | backer->avg_n_subfacet = 0; | |
1270 | backer->avg_subfacet_life = 0; | |
1271 | ||
acf60855 JP |
1272 | return error; |
1273 | } | |
1274 | ||
abe529af | 1275 | static int |
0f5f95a9 | 1276 | construct(struct ofproto *ofproto_) |
abe529af BP |
1277 | { |
1278 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1279 | struct shash_node *node, *next; |
4e022ec0 | 1280 | odp_port_t max_ports; |
abe529af | 1281 | int error; |
abe529af | 1282 | |
acf60855 | 1283 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1284 | if (error) { |
abe529af BP |
1285 | return error; |
1286 | } | |
1287 | ||
acf60855 | 1288 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
4e022ec0 AW |
1289 | ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports), |
1290 | ofp_to_u16(OFPP_MAX)))); | |
91858960 | 1291 | |
abe529af BP |
1292 | ofproto->netflow = NULL; |
1293 | ofproto->sflow = NULL; | |
29089a54 | 1294 | ofproto->ipfix = NULL; |
21f7563c | 1295 | ofproto->stp = NULL; |
abe529af | 1296 | hmap_init(&ofproto->bundles); |
e764773c | 1297 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
ec7ceaed | 1298 | ofproto->mbridge = mbridge_create(); |
abe529af | 1299 | ofproto->has_bonded_bundles = false; |
13501305 | 1300 | ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL); |
abe529af | 1301 | |
bcd2633a | 1302 | classifier_init(&ofproto->facets); |
4d2a0f39 | 1303 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 | 1304 | |
7ee20df1 BP |
1305 | list_init(&ofproto->completions); |
1306 | ||
3d9c5e58 EJ |
1307 | ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL); |
1308 | ovs_mutex_lock(&ofproto->flow_mod_mutex); | |
1309 | list_init(&ofproto->flow_mods); | |
1310 | ofproto->n_flow_mods = 0; | |
1311 | ovs_mutex_unlock(&ofproto->flow_mod_mutex); | |
1312 | ||
ada3a58d EJ |
1313 | ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL); |
1314 | ovs_mutex_lock(&ofproto->pin_mutex); | |
1315 | list_init(&ofproto->pins); | |
1316 | ofproto->n_pins = 0; | |
1317 | ovs_mutex_unlock(&ofproto->pin_mutex); | |
1318 | ||
abe529af BP |
1319 | ofproto_dpif_unixctl_init(); |
1320 | ||
52a90c29 BP |
1321 | hmap_init(&ofproto->vlandev_map); |
1322 | hmap_init(&ofproto->realdev_vid_map); | |
1323 | ||
acf60855 | 1324 | sset_init(&ofproto->ports); |
0a740f48 | 1325 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1326 | sset_init(&ofproto->port_poll_set); |
1327 | ofproto->port_poll_errno = 0; | |
1328 | ||
1329 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1330 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1331 | |
1332 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1333 | /* Check if the datapath already has this port. */ | |
1334 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1335 | sset_add(&ofproto->ports, node->name); | |
1336 | } | |
1337 | ||
1338 | free(iface_hint->br_name); | |
1339 | free(iface_hint->br_type); | |
4f9e08a5 | 1340 | free(iface_hint); |
acf60855 JP |
1341 | shash_delete(&init_ofp_ports, node); |
1342 | } | |
1343 | } | |
e1b1d06a | 1344 | |
b44a10b7 BP |
1345 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1346 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1347 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1348 | |
1349 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1350 | error = add_internal_flows(ofproto); |
1351 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1352 | ||
735d7efb AZ |
1353 | ofproto->n_hit = 0; |
1354 | ofproto->n_missed = 0; | |
1355 | ||
c57b2226 BP |
1356 | return error; |
1357 | } | |
1358 | ||
1359 | static int | |
1360 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1361 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1362 | { |
1363 | struct ofputil_flow_mod fm; | |
1364 | int error; | |
1365 | ||
81a76618 BP |
1366 | match_init_catchall(&fm.match); |
1367 | fm.priority = 0; | |
1368 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1369 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1370 | fm.cookie = htonll(0); |
1371 | fm.cookie_mask = htonll(0); | |
23342857 | 1372 | fm.modify_cookie = false; |
c57b2226 BP |
1373 | fm.table_id = TBL_INTERNAL; |
1374 | fm.command = OFPFC_ADD; | |
1375 | fm.idle_timeout = 0; | |
1376 | fm.hard_timeout = 0; | |
1377 | fm.buffer_id = 0; | |
1378 | fm.out_port = 0; | |
1379 | fm.flags = 0; | |
f25d0cf3 BP |
1380 | fm.ofpacts = ofpacts->data; |
1381 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1382 | |
1383 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1384 | if (error) { | |
1385 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1386 | id, ofperr_to_string(error)); | |
1387 | return error; | |
1388 | } | |
1389 | ||
9583bc14 EJ |
1390 | *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, |
1391 | TBL_INTERNAL); | |
cb22974d | 1392 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1393 | |
abe529af BP |
1394 | return 0; |
1395 | } | |
1396 | ||
c57b2226 BP |
1397 | static int |
1398 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1399 | { | |
f25d0cf3 BP |
1400 | struct ofpact_controller *controller; |
1401 | uint64_t ofpacts_stub[128 / 8]; | |
1402 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1403 | int error; |
1404 | int id; | |
1405 | ||
f25d0cf3 | 1406 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1407 | id = 1; |
1408 | ||
f25d0cf3 BP |
1409 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1410 | controller->max_len = UINT16_MAX; | |
1411 | controller->controller_id = 0; | |
1412 | controller->reason = OFPR_NO_MATCH; | |
1413 | ofpact_pad(&ofpacts); | |
1414 | ||
1415 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1416 | if (error) { |
1417 | return error; | |
1418 | } | |
1419 | ||
f25d0cf3 BP |
1420 | ofpbuf_clear(&ofpacts); |
1421 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 | 1422 | &ofproto->no_packet_in_rule); |
7fd51d39 BP |
1423 | if (error) { |
1424 | return error; | |
1425 | } | |
1426 | ||
1427 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1428 | &ofproto->drop_frags_rule); | |
c57b2226 BP |
1429 | return error; |
1430 | } | |
1431 | ||
7ee20df1 BP |
1432 | static void |
1433 | complete_operations(struct ofproto_dpif *ofproto) | |
1434 | { | |
1435 | struct dpif_completion *c, *next; | |
1436 | ||
1437 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1438 | ofoperation_complete(c->op, 0); | |
1439 | list_remove(&c->list_node); | |
1440 | free(c); | |
1441 | } | |
1442 | } | |
1443 | ||
abe529af BP |
1444 | static void |
1445 | destruct(struct ofproto *ofproto_) | |
1446 | { | |
1447 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1448 | struct rule_dpif *rule, *next_rule; |
ada3a58d | 1449 | struct ofputil_flow_mod *pin, *next_pin; |
3d9c5e58 | 1450 | struct ofputil_flow_mod *fm, *next_fm; |
d0918789 | 1451 | struct oftable *table; |
abe529af | 1452 | |
875b94ed | 1453 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1454 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1455 | xlate_remove_ofproto(ofproto); |
dc24a00f | 1456 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 1457 | |
b44a10b7 | 1458 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1459 | complete_operations(ofproto); |
1460 | ||
0697b5c3 BP |
1461 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1462 | struct cls_cursor cursor; | |
1463 | ||
0b4f2078 | 1464 | ovs_rwlock_wrlock(&table->cls.rwlock); |
d0918789 | 1465 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 | 1466 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
0b4f2078 | 1467 | ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up); |
0697b5c3 | 1468 | } |
0b4f2078 | 1469 | ovs_rwlock_unlock(&table->cls.rwlock); |
7ee20df1 BP |
1470 | } |
1471 | ||
3d9c5e58 EJ |
1472 | ovs_mutex_lock(&ofproto->flow_mod_mutex); |
1473 | LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) { | |
1474 | list_remove(&fm->list_node); | |
1475 | ofproto->n_flow_mods--; | |
1476 | free(fm->ofpacts); | |
1477 | free(fm); | |
1478 | } | |
1479 | ovs_mutex_unlock(&ofproto->flow_mod_mutex); | |
1480 | ovs_mutex_destroy(&ofproto->flow_mod_mutex); | |
1481 | ||
ada3a58d EJ |
1482 | ovs_mutex_lock(&ofproto->pin_mutex); |
1483 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) { | |
1484 | list_remove(&pin->list_node); | |
1485 | ofproto->n_pins--; | |
1486 | free(pin->ofpacts); | |
1487 | free(pin); | |
1488 | } | |
1489 | ovs_mutex_unlock(&ofproto->pin_mutex); | |
1490 | ovs_mutex_destroy(&ofproto->pin_mutex); | |
1491 | ||
ec7ceaed | 1492 | mbridge_unref(ofproto->mbridge); |
abe529af BP |
1493 | |
1494 | netflow_destroy(ofproto->netflow); | |
9723bcce | 1495 | dpif_sflow_unref(ofproto->sflow); |
abe529af | 1496 | hmap_destroy(&ofproto->bundles); |
5d989517 | 1497 | mac_learning_unref(ofproto->ml); |
abe529af | 1498 | |
bcd2633a | 1499 | classifier_destroy(&ofproto->facets); |
abe529af | 1500 | |
52a90c29 BP |
1501 | hmap_destroy(&ofproto->vlandev_map); |
1502 | hmap_destroy(&ofproto->realdev_vid_map); | |
1503 | ||
acf60855 | 1504 | sset_destroy(&ofproto->ports); |
0a740f48 | 1505 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1506 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1507 | |
13501305 EJ |
1508 | ovs_mutex_destroy(&ofproto->vsp_mutex); |
1509 | ||
acf60855 | 1510 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1511 | } |
1512 | ||
1513 | static int | |
5fcc0d00 | 1514 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1515 | { |
1516 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ada3a58d EJ |
1517 | struct ofputil_packet_in *pin, *next_pin; |
1518 | struct ofputil_flow_mod *fm, *next_fm; | |
1519 | struct list flow_mods, pins; | |
0aa66d6e | 1520 | struct ofport_dpif *ofport; |
abe529af | 1521 | |
40358701 GS |
1522 | /* Do not perform any periodic activity required by 'ofproto' while |
1523 | * waiting for flow restore to complete. */ | |
1524 | if (ofproto_get_flow_restore_wait()) { | |
1525 | return 0; | |
1526 | } | |
1527 | ||
3d9c5e58 EJ |
1528 | ovs_mutex_lock(&ofproto->flow_mod_mutex); |
1529 | if (ofproto->n_flow_mods) { | |
1530 | flow_mods = ofproto->flow_mods; | |
1531 | list_moved(&flow_mods); | |
1532 | list_init(&ofproto->flow_mods); | |
1533 | ofproto->n_flow_mods = 0; | |
1534 | } else { | |
1535 | list_init(&flow_mods); | |
1536 | } | |
1537 | ovs_mutex_unlock(&ofproto->flow_mod_mutex); | |
1538 | ||
ada3a58d | 1539 | LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) { |
3d9c5e58 EJ |
1540 | int error = ofproto_flow_mod(&ofproto->up, fm); |
1541 | if (error && !VLOG_DROP_WARN(&rl)) { | |
1542 | VLOG_WARN("learning action failed to modify flow table (%s)", | |
1543 | ofperr_get_name(error)); | |
1544 | } | |
1545 | ||
1546 | list_remove(&fm->list_node); | |
1547 | free(fm->ofpacts); | |
1548 | free(fm); | |
1549 | } | |
1550 | ||
ada3a58d EJ |
1551 | ovs_mutex_lock(&ofproto->pin_mutex); |
1552 | if (ofproto->n_pins) { | |
1553 | pins = ofproto->pins; | |
1554 | list_moved(&pins); | |
1555 | list_init(&ofproto->pins); | |
1556 | ofproto->n_pins = 0; | |
1557 | } else { | |
1558 | list_init(&pins); | |
1559 | } | |
1560 | ovs_mutex_unlock(&ofproto->pin_mutex); | |
1561 | ||
1562 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { | |
1563 | connmgr_send_packet_in(ofproto->up.connmgr, pin); | |
1564 | list_remove(&pin->list_node); | |
1565 | free(CONST_CAST(void *, pin->packet)); | |
1566 | free(pin); | |
1567 | } | |
1568 | ||
0aa66d6e EJ |
1569 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1570 | port_run_fast(ofport); | |
1571 | } | |
1572 | ||
5fcc0d00 BP |
1573 | return 0; |
1574 | } | |
1575 | ||
1576 | static int | |
1577 | run(struct ofproto *ofproto_) | |
1578 | { | |
1579 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1580 | struct ofport_dpif *ofport; | |
1581 | struct ofbundle *bundle; | |
1582 | int error; | |
1583 | ||
1584 | if (!clogged) { | |
1585 | complete_operations(ofproto); | |
1586 | } | |
5fcc0d00 | 1587 | |
ec7ceaed EJ |
1588 | if (mbridge_need_revalidate(ofproto->mbridge)) { |
1589 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
509c0149 | 1590 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 1591 | mac_learning_flush(ofproto->ml); |
509c0149 | 1592 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
ec7ceaed EJ |
1593 | } |
1594 | ||
40358701 GS |
1595 | /* Do not perform any periodic activity below required by 'ofproto' while |
1596 | * waiting for flow restore to complete. */ | |
1597 | if (ofproto_get_flow_restore_wait()) { | |
1598 | return 0; | |
1599 | } | |
1600 | ||
5fcc0d00 BP |
1601 | error = run_fast(ofproto_); |
1602 | if (error) { | |
1603 | return error; | |
abe529af BP |
1604 | } |
1605 | ||
abe529af | 1606 | if (ofproto->netflow) { |
6fca1ffb BP |
1607 | if (netflow_run(ofproto->netflow)) { |
1608 | send_netflow_active_timeouts(ofproto); | |
1609 | } | |
abe529af BP |
1610 | } |
1611 | if (ofproto->sflow) { | |
bae473fe | 1612 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1613 | } |
1614 | ||
1615 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1616 | port_run(ofport); | |
1617 | } | |
1618 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1619 | bundle_run(bundle); | |
1620 | } | |
1621 | ||
21f7563c | 1622 | stp_run(ofproto); |
509c0149 | 1623 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 EJ |
1624 | if (mac_learning_run(ofproto->ml)) { |
1625 | ofproto->backer->need_revalidate = REV_MAC_LEARNING; | |
1626 | } | |
509c0149 | 1627 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 1628 | |
6814e51f | 1629 | /* Check the consistency of a random facet, to aid debugging. */ |
0b4f2078 | 1630 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
4d2a0f39 | 1631 | if (time_msec() >= ofproto->consistency_rl |
bcd2633a | 1632 | && !classifier_is_empty(&ofproto->facets) |
2cc3c58e | 1633 | && !ofproto->backer->need_revalidate) { |
bcd2633a JP |
1634 | struct cls_table *table; |
1635 | struct cls_rule *cr; | |
6814e51f BP |
1636 | struct facet *facet; |
1637 | ||
4d2a0f39 EJ |
1638 | ofproto->consistency_rl = time_msec() + 250; |
1639 | ||
bcd2633a JP |
1640 | table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables), |
1641 | struct cls_table, hmap_node); | |
1642 | cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, | |
1643 | hmap_node); | |
1644 | facet = CONTAINER_OF(cr, struct facet, cr); | |
1645 | ||
a1616063 EJ |
1646 | if (!facet_check_consistency(facet)) { |
1647 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
6814e51f BP |
1648 | } |
1649 | } | |
0b4f2078 | 1650 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 1651 | |
abe529af BP |
1652 | return 0; |
1653 | } | |
1654 | ||
1655 | static void | |
1656 | wait(struct ofproto *ofproto_) | |
1657 | { | |
1658 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1659 | struct ofport_dpif *ofport; | |
1660 | struct ofbundle *bundle; | |
1661 | ||
7ee20df1 BP |
1662 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1663 | poll_immediate_wake(); | |
1664 | } | |
1665 | ||
40358701 GS |
1666 | if (ofproto_get_flow_restore_wait()) { |
1667 | return; | |
1668 | } | |
1669 | ||
acf60855 JP |
1670 | dpif_wait(ofproto->backer->dpif); |
1671 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1672 | if (ofproto->sflow) { |
bae473fe | 1673 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1674 | } |
abe529af BP |
1675 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1676 | port_wait(ofport); | |
1677 | } | |
1678 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1679 | bundle_wait(bundle); | |
1680 | } | |
6fca1ffb BP |
1681 | if (ofproto->netflow) { |
1682 | netflow_wait(ofproto->netflow); | |
1683 | } | |
509c0149 | 1684 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
1c313b88 | 1685 | mac_learning_wait(ofproto->ml); |
509c0149 | 1686 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c | 1687 | stp_wait(ofproto); |
2cc3c58e | 1688 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1689 | /* Shouldn't happen, but if it does just go around again. */ |
1690 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1691 | poll_immediate_wake(); | |
abe529af BP |
1692 | } |
1693 | } | |
1694 | ||
0d085684 BP |
1695 | static void |
1696 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1697 | { | |
1698 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bcd2633a | 1699 | struct cls_cursor cursor; |
04d08d54 EJ |
1700 | size_t n_subfacets = 0; |
1701 | struct facet *facet; | |
0d085684 | 1702 | |
0b4f2078 | 1703 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a | 1704 | simap_increase(usage, "facets", classifier_count(&ofproto->facets)); |
0b4f2078 | 1705 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 1706 | |
0b4f2078 | 1707 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
1708 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
1709 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
04d08d54 EJ |
1710 | n_subfacets += list_size(&facet->subfacets); |
1711 | } | |
0b4f2078 | 1712 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
04d08d54 | 1713 | simap_increase(usage, "subfacets", n_subfacets); |
0d085684 BP |
1714 | } |
1715 | ||
abe529af BP |
1716 | static void |
1717 | flush(struct ofproto *ofproto_) | |
1718 | { | |
1719 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1720 | struct subfacet *subfacet, *next_subfacet; |
1721 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1722 | int n_batch; | |
b0f7b9b5 | 1723 | |
acf60855 JP |
1724 | n_batch = 0; |
1725 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
04d08d54 | 1726 | &ofproto->backer->subfacets) { |
b7e2f6b3 | 1727 | if (subfacet->facet->ofproto != ofproto) { |
04d08d54 EJ |
1728 | continue; |
1729 | } | |
1730 | ||
acf60855 JP |
1731 | if (subfacet->path != SF_NOT_INSTALLED) { |
1732 | batch[n_batch++] = subfacet; | |
1733 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
04d08d54 | 1734 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 JP |
1735 | n_batch = 0; |
1736 | } | |
1737 | } else { | |
1738 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1739 | } |
abe529af | 1740 | } |
acf60855 JP |
1741 | |
1742 | if (n_batch > 0) { | |
04d08d54 | 1743 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 | 1744 | } |
abe529af BP |
1745 | } |
1746 | ||
6c1491fb BP |
1747 | static void |
1748 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1749 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1750 | { |
1751 | *arp_match_ip = true; | |
9e1fd49b BP |
1752 | *actions = (OFPUTIL_A_OUTPUT | |
1753 | OFPUTIL_A_SET_VLAN_VID | | |
1754 | OFPUTIL_A_SET_VLAN_PCP | | |
1755 | OFPUTIL_A_STRIP_VLAN | | |
1756 | OFPUTIL_A_SET_DL_SRC | | |
1757 | OFPUTIL_A_SET_DL_DST | | |
1758 | OFPUTIL_A_SET_NW_SRC | | |
1759 | OFPUTIL_A_SET_NW_DST | | |
1760 | OFPUTIL_A_SET_NW_TOS | | |
1761 | OFPUTIL_A_SET_TP_SRC | | |
1762 | OFPUTIL_A_SET_TP_DST | | |
1763 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1764 | } |
1765 | ||
1766 | static void | |
307975da | 1767 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1768 | { |
1769 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1770 | struct dpif_dp_stats s; |
7fd51d39 | 1771 | uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; |
7e741ffb | 1772 | uint64_t n_lookup; |
6c1491fb BP |
1773 | |
1774 | strcpy(ots->name, "classifier"); | |
1775 | ||
acf60855 | 1776 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1777 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1778 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
7fd51d39 | 1779 | rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); |
acf60855 | 1780 | |
7fd51d39 | 1781 | n_lookup = s.n_hit + s.n_missed - n_dropped_frags; |
7e741ffb JG |
1782 | ots->lookup_count = htonll(n_lookup); |
1783 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1784 | } |
1785 | ||
abe529af BP |
1786 | static struct ofport * |
1787 | port_alloc(void) | |
1788 | { | |
1789 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1790 | return &port->up; | |
1791 | } | |
1792 | ||
1793 | static void | |
1794 | port_dealloc(struct ofport *port_) | |
1795 | { | |
1796 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1797 | free(port); | |
1798 | } | |
1799 | ||
1800 | static int | |
1801 | port_construct(struct ofport *port_) | |
1802 | { | |
1803 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1804 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1805 | const struct netdev *netdev = port->up.netdev; |
3aa30359 | 1806 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
e1b1d06a JP |
1807 | struct dpif_port dpif_port; |
1808 | int error; | |
abe529af | 1809 | |
2cc3c58e | 1810 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1811 | port->bundle = NULL; |
1812 | port->cfm = NULL; | |
ccc09689 | 1813 | port->bfd = NULL; |
d5ffa7f2 | 1814 | port->may_enable = true; |
21f7563c JP |
1815 | port->stp_port = NULL; |
1816 | port->stp_state = STP_DISABLED; | |
42943cde | 1817 | port->is_tunnel = false; |
6cbbf4fa | 1818 | port->peer = NULL; |
55954f6e EJ |
1819 | port->qdscp = NULL; |
1820 | port->n_qdscp = 0; | |
52a90c29 BP |
1821 | port->realdev_ofp_port = 0; |
1822 | port->vlandev_vid = 0; | |
b9ad7294 | 1823 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1824 | |
b9ad7294 | 1825 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1826 | /* By bailing out here, we don't submit the port to the sFlow module |
1827 | * to be considered for counter polling export. This is correct | |
1828 | * because the patch port represents an interface that sFlow considers | |
1829 | * to be "internal" to the switch as a whole, and therefore not an | |
1830 | * candidate for counter polling. */ | |
4e022ec0 | 1831 | port->odp_port = ODPP_NONE; |
6cbbf4fa | 1832 | ofport_update_peer(port); |
0a740f48 EJ |
1833 | return 0; |
1834 | } | |
1835 | ||
acf60855 | 1836 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
3aa30359 BP |
1837 | netdev_vport_get_dpif_port(netdev, namebuf, |
1838 | sizeof namebuf), | |
e1b1d06a JP |
1839 | &dpif_port); |
1840 | if (error) { | |
1841 | return error; | |
1842 | } | |
1843 | ||
1844 | port->odp_port = dpif_port.port_no; | |
1845 | ||
b9ad7294 | 1846 | if (netdev_get_tunnel_config(netdev)) { |
42943cde EJ |
1847 | tnl_port_add(port, port->up.netdev, port->odp_port); |
1848 | port->is_tunnel = true; | |
b9ad7294 EJ |
1849 | } else { |
1850 | /* Sanity-check that a mapping doesn't already exist. This | |
1851 | * shouldn't happen for non-tunnel ports. */ | |
1852 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1853 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1854 | dpif_port.name); | |
da78d43d | 1855 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1856 | return EBUSY; |
1857 | } | |
e1b1d06a | 1858 | |
8449c4d6 | 1859 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1860 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
f9c0c3ec | 1861 | hash_odp_port(port->odp_port)); |
8449c4d6 | 1862 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1863 | } |
da78d43d | 1864 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1865 | |
abe529af | 1866 | if (ofproto->sflow) { |
e1b1d06a | 1867 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1868 | } |
1869 | ||
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | static void | |
1874 | port_destruct(struct ofport *port_) | |
1875 | { | |
1876 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
02f8d646 | 1878 | const char *devname = netdev_get_name(port->up.netdev); |
3aa30359 BP |
1879 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1880 | const char *dp_port_name; | |
abe529af | 1881 | |
7894e7da | 1882 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1883 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1884 | xlate_ofport_remove(port); |
dc24a00f | 1885 | ovs_rwlock_unlock(&xlate_rwlock); |
7894e7da | 1886 | |
3aa30359 BP |
1887 | dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, |
1888 | sizeof namebuf); | |
a614d823 | 1889 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1890 | /* The underlying device is still there, so delete it. This |
1891 | * happens when the ofproto is being destroyed, since the caller | |
1892 | * assumes that removal of attached ports will happen as part of | |
1893 | * destruction. */ | |
42943cde | 1894 | if (!port->is_tunnel) { |
a614d823 KM |
1895 | dpif_port_del(ofproto->backer->dpif, port->odp_port); |
1896 | } | |
acf60855 JP |
1897 | } |
1898 | ||
6cbbf4fa EJ |
1899 | if (port->peer) { |
1900 | port->peer->peer = NULL; | |
1901 | port->peer = NULL; | |
1902 | } | |
1903 | ||
42943cde | 1904 | if (port->odp_port != ODPP_NONE && !port->is_tunnel) { |
8449c4d6 | 1905 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 | 1906 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
8449c4d6 | 1907 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 EJ |
1908 | } |
1909 | ||
42943cde | 1910 | tnl_port_del(port); |
02f8d646 | 1911 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1912 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
abe529af | 1913 | bundle_remove(port_); |
a5610457 | 1914 | set_cfm(port_, NULL); |
8aee94b6 | 1915 | set_bfd(port_, NULL); |
abe529af | 1916 | if (ofproto->sflow) { |
bae473fe | 1917 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1918 | } |
8b36f51e | 1919 | |
55954f6e | 1920 | free(port->qdscp); |
abe529af BP |
1921 | } |
1922 | ||
1923 | static void | |
1924 | port_modified(struct ofport *port_) | |
1925 | { | |
1926 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1927 | ||
1928 | if (port->bundle && port->bundle->bond) { | |
1929 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1930 | } | |
9d46b444 EJ |
1931 | |
1932 | if (port->cfm) { | |
1933 | cfm_set_netdev(port->cfm, port->up.netdev); | |
1934 | } | |
f1e78bbd | 1935 | |
42943cde EJ |
1936 | if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, |
1937 | port->odp_port)) { | |
7894e7da EJ |
1938 | ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = |
1939 | REV_RECONFIGURE; | |
f1e78bbd | 1940 | } |
6cbbf4fa EJ |
1941 | |
1942 | ofport_update_peer(port); | |
abe529af BP |
1943 | } |
1944 | ||
1945 | static void | |
9e1fd49b | 1946 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1947 | { |
1948 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1949 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1950 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1951 | |
9e1fd49b | 1952 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1953 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1954 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1955 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1956 | |
9e1fd49b | 1957 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1958 | bundle_update(port->bundle); |
1959 | } | |
abe529af BP |
1960 | } |
1961 | } | |
1962 | ||
1963 | static int | |
1964 | set_sflow(struct ofproto *ofproto_, | |
1965 | const struct ofproto_sflow_options *sflow_options) | |
1966 | { | |
1967 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1968 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1969 | |
abe529af | 1970 | if (sflow_options) { |
bae473fe | 1971 | if (!ds) { |
abe529af BP |
1972 | struct ofport_dpif *ofport; |
1973 | ||
4213f19d | 1974 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1975 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1976 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1977 | } |
2cc3c58e | 1978 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1979 | } |
bae473fe | 1980 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1981 | } else { |
6ff686f2 | 1982 | if (ds) { |
9723bcce | 1983 | dpif_sflow_unref(ds); |
2cc3c58e | 1984 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1985 | ofproto->sflow = NULL; |
1986 | } | |
abe529af BP |
1987 | } |
1988 | return 0; | |
1989 | } | |
1990 | ||
29089a54 RL |
1991 | static int |
1992 | set_ipfix( | |
1993 | struct ofproto *ofproto_, | |
1994 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1995 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1996 | size_t n_flow_exporters_options) | |
1997 | { | |
1998 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1999 | struct dpif_ipfix *di = ofproto->ipfix; | |
2000 | ||
2001 | if (bridge_exporter_options || flow_exporters_options) { | |
2002 | if (!di) { | |
2003 | di = ofproto->ipfix = dpif_ipfix_create(); | |
2004 | } | |
2005 | dpif_ipfix_set_options( | |
2006 | di, bridge_exporter_options, flow_exporters_options, | |
2007 | n_flow_exporters_options); | |
2008 | } else { | |
2009 | if (di) { | |
d857c8aa | 2010 | dpif_ipfix_unref(di); |
29089a54 RL |
2011 | ofproto->ipfix = NULL; |
2012 | } | |
2013 | } | |
2014 | return 0; | |
2015 | } | |
2016 | ||
abe529af | 2017 | static int |
a5610457 | 2018 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
2019 | { |
2020 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2021 | int error; | |
2022 | ||
a5610457 | 2023 | if (!s) { |
abe529af BP |
2024 | error = 0; |
2025 | } else { | |
2026 | if (!ofport->cfm) { | |
8c977421 EJ |
2027 | struct ofproto_dpif *ofproto; |
2028 | ||
2029 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 2030 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 2031 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
2032 | } |
2033 | ||
a5610457 | 2034 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
2035 | return 0; |
2036 | } | |
2037 | ||
2038 | error = EINVAL; | |
2039 | } | |
8e9a73fa | 2040 | cfm_unref(ofport->cfm); |
abe529af BP |
2041 | ofport->cfm = NULL; |
2042 | return error; | |
2043 | } | |
2044 | ||
9a9e3786 BP |
2045 | static bool |
2046 | get_cfm_status(const struct ofport *ofport_, | |
2047 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
2048 | { |
2049 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2050 | ||
2051 | if (ofport->cfm) { | |
9a9e3786 BP |
2052 | status->faults = cfm_get_fault(ofport->cfm); |
2053 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
2054 | status->health = cfm_get_health(ofport->cfm); | |
2055 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
2056 | return true; | |
1de11730 | 2057 | } else { |
9a9e3786 | 2058 | return false; |
1de11730 EJ |
2059 | } |
2060 | } | |
ccc09689 EJ |
2061 | |
2062 | static int | |
2063 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
2064 | { | |
2065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
2066 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2067 | struct bfd *old; | |
2068 | ||
2069 | old = ofport->bfd; | |
2070 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
2071 | if (ofport->bfd != old) { | |
2072 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2073 | } | |
2074 | ||
2075 | return 0; | |
2076 | } | |
2077 | ||
2078 | static int | |
2079 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
2080 | { | |
2081 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2082 | ||
2083 | if (ofport->bfd) { | |
2084 | bfd_get_status(ofport->bfd, smap); | |
2085 | return 0; | |
2086 | } else { | |
2087 | return ENOENT; | |
2088 | } | |
2089 | } | |
abe529af | 2090 | \f |
21f7563c JP |
2091 | /* Spanning Tree. */ |
2092 | ||
2093 | static void | |
2094 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2095 | { | |
2096 | struct ofproto_dpif *ofproto = ofproto_; | |
2097 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2098 | struct ofport_dpif *ofport; | |
2099 | ||
2100 | ofport = stp_port_get_aux(sp); | |
2101 | if (!ofport) { | |
2102 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2103 | ofproto->up.name, port_num); | |
2104 | } else { | |
2105 | struct eth_header *eth = pkt->l2; | |
2106 | ||
2107 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2108 | if (eth_addr_is_zero(eth->eth_src)) { | |
2109 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2110 | "with unknown MAC", ofproto->up.name, port_num); | |
2111 | } else { | |
97d6520b | 2112 | send_packet(ofport, pkt); |
21f7563c JP |
2113 | } |
2114 | } | |
2115 | ofpbuf_delete(pkt); | |
2116 | } | |
2117 | ||
2118 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2119 | static int | |
2120 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2121 | { | |
2122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2123 | ||
2124 | /* Only revalidate flows if the configuration changed. */ | |
2125 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2126 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2127 | } |
2128 | ||
2129 | if (s) { | |
2130 | if (!ofproto->stp) { | |
2131 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2132 | send_bpdu_cb, ofproto); | |
2133 | ofproto->stp_last_tick = time_msec(); | |
2134 | } | |
2135 | ||
2136 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2137 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2138 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2139 | stp_set_max_age(ofproto->stp, s->max_age); | |
2140 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2141 | } else { | |
851bf71d EJ |
2142 | struct ofport *ofport; |
2143 | ||
2144 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2145 | set_stp_port(ofport, NULL); | |
2146 | } | |
2147 | ||
bd54dbcd | 2148 | stp_unref(ofproto->stp); |
21f7563c JP |
2149 | ofproto->stp = NULL; |
2150 | } | |
2151 | ||
2152 | return 0; | |
2153 | } | |
2154 | ||
2155 | static int | |
2156 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2157 | { | |
2158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2159 | ||
2160 | if (ofproto->stp) { | |
2161 | s->enabled = true; | |
2162 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2163 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2164 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2165 | } else { | |
2166 | s->enabled = false; | |
2167 | } | |
2168 | ||
2169 | return 0; | |
2170 | } | |
2171 | ||
2172 | static void | |
2173 | update_stp_port_state(struct ofport_dpif *ofport) | |
2174 | { | |
2175 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2176 | enum stp_state state; | |
2177 | ||
2178 | /* Figure out new state. */ | |
2179 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2180 | : STP_DISABLED; | |
2181 | ||
2182 | /* Update state. */ | |
2183 | if (ofport->stp_state != state) { | |
9e1fd49b | 2184 | enum ofputil_port_state of_state; |
21f7563c JP |
2185 | bool fwd_change; |
2186 | ||
2187 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2188 | netdev_get_name(ofport->up.netdev), | |
2189 | stp_state_name(ofport->stp_state), | |
2190 | stp_state_name(state)); | |
2191 | if (stp_learn_in_state(ofport->stp_state) | |
2192 | != stp_learn_in_state(state)) { | |
2193 | /* xxx Learning action flows should also be flushed. */ | |
509c0149 | 2194 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2195 | mac_learning_flush(ofproto->ml); |
509c0149 | 2196 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c JP |
2197 | } |
2198 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2199 | != stp_forward_in_state(state); | |
2200 | ||
2cc3c58e | 2201 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2202 | ofport->stp_state = state; |
2203 | ofport->stp_state_entered = time_msec(); | |
2204 | ||
b308140a | 2205 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2206 | bundle_update(ofport->bundle); |
2207 | } | |
2208 | ||
2209 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2210 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2211 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2212 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2213 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2214 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2215 | : 0); | |
21f7563c JP |
2216 | ofproto_port_set_state(&ofport->up, of_state); |
2217 | } | |
2218 | } | |
2219 | ||
2220 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2221 | * caller is responsible for assigning STP port numbers and ensuring | |
2222 | * there are no duplicates. */ | |
2223 | static int | |
2224 | set_stp_port(struct ofport *ofport_, | |
2225 | const struct ofproto_port_stp_settings *s) | |
2226 | { | |
2227 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2228 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2229 | struct stp_port *sp = ofport->stp_port; | |
2230 | ||
2231 | if (!s || !s->enable) { | |
2232 | if (sp) { | |
2233 | ofport->stp_port = NULL; | |
2234 | stp_port_disable(sp); | |
ecd12731 | 2235 | update_stp_port_state(ofport); |
21f7563c JP |
2236 | } |
2237 | return 0; | |
2238 | } else if (sp && stp_port_no(sp) != s->port_num | |
2239 | && ofport == stp_port_get_aux(sp)) { | |
2240 | /* The port-id changed, so disable the old one if it's not | |
2241 | * already in use by another port. */ | |
2242 | stp_port_disable(sp); | |
2243 | } | |
2244 | ||
2245 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2246 | stp_port_enable(sp); | |
2247 | ||
2248 | stp_port_set_aux(sp, ofport); | |
2249 | stp_port_set_priority(sp, s->priority); | |
2250 | stp_port_set_path_cost(sp, s->path_cost); | |
2251 | ||
2252 | update_stp_port_state(ofport); | |
2253 | ||
2254 | return 0; | |
2255 | } | |
2256 | ||
2257 | static int | |
2258 | get_stp_port_status(struct ofport *ofport_, | |
2259 | struct ofproto_port_stp_status *s) | |
2260 | { | |
2261 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2262 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2263 | struct stp_port *sp = ofport->stp_port; | |
2264 | ||
2265 | if (!ofproto->stp || !sp) { | |
2266 | s->enabled = false; | |
2267 | return 0; | |
2268 | } | |
2269 | ||
2270 | s->enabled = true; | |
2271 | s->port_id = stp_port_get_id(sp); | |
2272 | s->state = stp_port_get_state(sp); | |
2273 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2274 | s->role = stp_port_get_role(sp); | |
80740385 | 2275 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2276 | |
2277 | return 0; | |
2278 | } | |
2279 | ||
2280 | static void | |
2281 | stp_run(struct ofproto_dpif *ofproto) | |
2282 | { | |
2283 | if (ofproto->stp) { | |
2284 | long long int now = time_msec(); | |
2285 | long long int elapsed = now - ofproto->stp_last_tick; | |
2286 | struct stp_port *sp; | |
2287 | ||
2288 | if (elapsed > 0) { | |
2289 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2290 | ofproto->stp_last_tick = now; | |
2291 | } | |
2292 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2293 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2294 | ||
2295 | if (ofport) { | |
2296 | update_stp_port_state(ofport); | |
2297 | } | |
2298 | } | |
6ae50723 EJ |
2299 | |
2300 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
509c0149 | 2301 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2302 | mac_learning_flush(ofproto->ml); |
509c0149 | 2303 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
6ae50723 | 2304 | } |
21f7563c JP |
2305 | } |
2306 | } | |
2307 | ||
2308 | static void | |
2309 | stp_wait(struct ofproto_dpif *ofproto) | |
2310 | { | |
2311 | if (ofproto->stp) { | |
2312 | poll_timer_wait(1000); | |
2313 | } | |
2314 | } | |
21f7563c | 2315 | \f |
8b36f51e | 2316 | static int |
55954f6e | 2317 | set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp, |
8b36f51e EJ |
2318 | size_t n_qdscp) |
2319 | { | |
2320 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2321 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
8b36f51e | 2322 | |
55954f6e EJ |
2323 | if (ofport->n_qdscp != n_qdscp |
2324 | || (n_qdscp && memcmp(ofport->qdscp, qdscp, | |
2325 | n_qdscp * sizeof *qdscp))) { | |
2cc3c58e | 2326 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
55954f6e EJ |
2327 | free(ofport->qdscp); |
2328 | ofport->qdscp = n_qdscp | |
2329 | ? xmemdup(qdscp, n_qdscp * sizeof *qdscp) | |
2330 | : NULL; | |
2331 | ofport->n_qdscp = n_qdscp; | |
8b36f51e EJ |
2332 | } |
2333 | ||
8b36f51e EJ |
2334 | return 0; |
2335 | } | |
2336 | \f | |
abe529af BP |
2337 | /* Bundles. */ |
2338 | ||
b44a10b7 BP |
2339 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2340 | * ofproto to revalidate every flow. | |
2341 | * | |
2342 | * Normally MAC learning entries are removed only from the ofproto associated | |
2343 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2344 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2345 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2346 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2347 | * with the host from which it migrated. */ | |
abe529af | 2348 | static void |
b44a10b7 | 2349 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2350 | { |
2351 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2352 | struct mac_learning *ml = ofproto->ml; | |
2353 | struct mac_entry *mac, *next_mac; | |
2354 | ||
2cc3c58e | 2355 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
509c0149 | 2356 | ovs_rwlock_wrlock(&ml->rwlock); |
abe529af BP |
2357 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2358 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2359 | if (all_ofprotos) { |
2360 | struct ofproto_dpif *o; | |
2361 | ||
2362 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2363 | if (o != ofproto) { | |
2364 | struct mac_entry *e; | |
2365 | ||
509c0149 | 2366 | ovs_rwlock_wrlock(&o->ml->rwlock); |
30618594 | 2367 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan); |
b44a10b7 | 2368 | if (e) { |
b44a10b7 BP |
2369 | mac_learning_expire(o->ml, e); |
2370 | } | |
509c0149 | 2371 | ovs_rwlock_unlock(&o->ml->rwlock); |
b44a10b7 BP |
2372 | } |
2373 | } | |
2374 | } | |
2375 | ||
abe529af BP |
2376 | mac_learning_expire(ml, mac); |
2377 | } | |
2378 | } | |
509c0149 | 2379 | ovs_rwlock_unlock(&ml->rwlock); |
abe529af BP |
2380 | } |
2381 | ||
2382 | static struct ofbundle * | |
2383 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2384 | { | |
2385 | struct ofbundle *bundle; | |
2386 | ||
2387 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2388 | &ofproto->bundles) { | |
2389 | if (bundle->aux == aux) { | |
2390 | return bundle; | |
2391 | } | |
2392 | } | |
2393 | return NULL; | |
2394 | } | |
2395 | ||
7bde8dd8 JP |
2396 | static void |
2397 | bundle_update(struct ofbundle *bundle) | |
2398 | { | |
2399 | struct ofport_dpif *port; | |
2400 | ||
2401 | bundle->floodable = true; | |
2402 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2403 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2404 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2405 | bundle->floodable = false; |
2406 | break; | |
2407 | } | |
2408 | } | |
2409 | } | |
2410 | ||
abe529af BP |
2411 | static void |
2412 | bundle_del_port(struct ofport_dpif *port) | |
2413 | { | |
2414 | struct ofbundle *bundle = port->bundle; | |
2415 | ||
2cc3c58e | 2416 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2417 | |
abe529af BP |
2418 | list_remove(&port->bundle_node); |
2419 | port->bundle = NULL; | |
2420 | ||
2421 | if (bundle->lacp) { | |
2422 | lacp_slave_unregister(bundle->lacp, port); | |
2423 | } | |
2424 | if (bundle->bond) { | |
2425 | bond_slave_unregister(bundle->bond, port); | |
2426 | } | |
2427 | ||
7bde8dd8 | 2428 | bundle_update(bundle); |
abe529af BP |
2429 | } |
2430 | ||
2431 | static bool | |
4e022ec0 | 2432 | bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, |
df53d41c | 2433 | struct lacp_slave_settings *lacp) |
abe529af BP |
2434 | { |
2435 | struct ofport_dpif *port; | |
2436 | ||
2437 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2438 | if (!port) { | |
2439 | return false; | |
2440 | } | |
2441 | ||
2442 | if (port->bundle != bundle) { | |
2cc3c58e | 2443 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2444 | if (port->bundle) { |
2445 | bundle_del_port(port); | |
2446 | } | |
2447 | ||
2448 | port->bundle = bundle; | |
2449 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2450 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2451 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2452 | bundle->floodable = false; |
2453 | } | |
2454 | } | |
2455 | if (lacp) { | |
2cc3c58e | 2456 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2457 | lacp_slave_register(bundle->lacp, port, lacp); |
2458 | } | |
2459 | ||
2460 | return true; | |
2461 | } | |
2462 | ||
2463 | static void | |
2464 | bundle_destroy(struct ofbundle *bundle) | |
2465 | { | |
2466 | struct ofproto_dpif *ofproto; | |
2467 | struct ofport_dpif *port, *next_port; | |
abe529af BP |
2468 | |
2469 | if (!bundle) { | |
2470 | return; | |
2471 | } | |
2472 | ||
2473 | ofproto = bundle->ofproto; | |
ec7ceaed | 2474 | mbridge_unregister_bundle(ofproto->mbridge, bundle->aux); |
abe529af | 2475 | |
dc24a00f | 2476 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 2477 | xlate_bundle_remove(bundle); |
dc24a00f | 2478 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 2479 | |
abe529af BP |
2480 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { |
2481 | bundle_del_port(port); | |
2482 | } | |
2483 | ||
b44a10b7 | 2484 | bundle_flush_macs(bundle, true); |
abe529af BP |
2485 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2486 | free(bundle->name); | |
2487 | free(bundle->trunks); | |
91779071 | 2488 | lacp_unref(bundle->lacp); |
03366a2d | 2489 | bond_unref(bundle->bond); |
abe529af BP |
2490 | free(bundle); |
2491 | } | |
2492 | ||
2493 | static int | |
2494 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2495 | const struct ofproto_bundle_settings *s) | |
2496 | { | |
2497 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2498 | bool need_flush = false; | |
abe529af BP |
2499 | struct ofport_dpif *port; |
2500 | struct ofbundle *bundle; | |
ecac4ebf BP |
2501 | unsigned long *trunks; |
2502 | int vlan; | |
abe529af BP |
2503 | size_t i; |
2504 | bool ok; | |
2505 | ||
2506 | if (!s) { | |
2507 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2508 | return 0; | |
2509 | } | |
2510 | ||
cb22974d BP |
2511 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2512 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2513 | |
2514 | bundle = bundle_lookup(ofproto, aux); | |
2515 | if (!bundle) { | |
2516 | bundle = xmalloc(sizeof *bundle); | |
2517 | ||
2518 | bundle->ofproto = ofproto; | |
2519 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2520 | hash_pointer(aux, 0)); | |
2521 | bundle->aux = aux; | |
2522 | bundle->name = NULL; | |
2523 | ||
2524 | list_init(&bundle->ports); | |
ecac4ebf | 2525 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2526 | bundle->vlan = -1; |
2527 | bundle->trunks = NULL; | |
5e9ceccd | 2528 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2529 | bundle->lacp = NULL; |
2530 | bundle->bond = NULL; | |
2531 | ||
2532 | bundle->floodable = true; | |
ec7ceaed | 2533 | mbridge_register_bundle(ofproto->mbridge, bundle); |
abe529af BP |
2534 | } |
2535 | ||
2536 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2537 | free(bundle->name); | |
2538 | bundle->name = xstrdup(s->name); | |
2539 | } | |
2540 | ||
2541 | /* LACP. */ | |
2542 | if (s->lacp) { | |
2543 | if (!bundle->lacp) { | |
2cc3c58e | 2544 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2545 | bundle->lacp = lacp_create(); |
2546 | } | |
2547 | lacp_configure(bundle->lacp, s->lacp); | |
2548 | } else { | |
91779071 | 2549 | lacp_unref(bundle->lacp); |
abe529af BP |
2550 | bundle->lacp = NULL; |
2551 | } | |
2552 | ||
2553 | /* Update set of ports. */ | |
2554 | ok = true; | |
2555 | for (i = 0; i < s->n_slaves; i++) { | |
2556 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2557 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2558 | ok = false; |
2559 | } | |
2560 | } | |
2561 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2562 | struct ofport_dpif *next_port; | |
2563 | ||
2564 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2565 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2566 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2567 | goto found; |
2568 | } | |
2569 | } | |
2570 | ||
2571 | bundle_del_port(port); | |
2572 | found: ; | |
2573 | } | |
2574 | } | |
cb22974d | 2575 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2576 | |
2577 | if (list_is_empty(&bundle->ports)) { | |
2578 | bundle_destroy(bundle); | |
2579 | return EINVAL; | |
2580 | } | |
2581 | ||
ecac4ebf | 2582 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2583 | if (s->vlan_mode != bundle->vlan_mode |
2584 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2585 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2586 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2587 | need_flush = true; |
2588 | } | |
2589 | ||
abe529af | 2590 | /* Set VLAN tag. */ |
ecac4ebf BP |
2591 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2592 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2593 | : 0); | |
2594 | if (vlan != bundle->vlan) { | |
2595 | bundle->vlan = vlan; | |
abe529af BP |
2596 | need_flush = true; |
2597 | } | |
2598 | ||
2599 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2600 | switch (s->vlan_mode) { |
2601 | case PORT_VLAN_ACCESS: | |
2602 | trunks = NULL; | |
2603 | break; | |
2604 | ||
2605 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2606 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2607 | break; |
2608 | ||
2609 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2610 | case PORT_VLAN_NATIVE_TAGGED: | |
2611 | if (vlan != 0 && (!s->trunks | |
2612 | || !bitmap_is_set(s->trunks, vlan) | |
2613 | || bitmap_is_set(s->trunks, 0))) { | |
2614 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2615 | if (s->trunks) { | |
2616 | trunks = bitmap_clone(s->trunks, 4096); | |
2617 | } else { | |
2618 | trunks = bitmap_allocate1(4096); | |
2619 | } | |
2620 | bitmap_set1(trunks, vlan); | |
2621 | bitmap_set0(trunks, 0); | |
2622 | } else { | |
ebc56baa | 2623 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2624 | } |
2625 | break; | |
2626 | ||
2627 | default: | |
2628 | NOT_REACHED(); | |
2629 | } | |
abe529af BP |
2630 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2631 | free(bundle->trunks); | |
ecac4ebf BP |
2632 | if (trunks == s->trunks) { |
2633 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2634 | } else { | |
2635 | bundle->trunks = trunks; | |
2636 | trunks = NULL; | |
2637 | } | |
abe529af BP |
2638 | need_flush = true; |
2639 | } | |
ecac4ebf BP |
2640 | if (trunks != s->trunks) { |
2641 | free(trunks); | |
2642 | } | |
abe529af BP |
2643 | |
2644 | /* Bonding. */ | |
2645 | if (!list_is_short(&bundle->ports)) { | |
2646 | bundle->ofproto->has_bonded_bundles = true; | |
2647 | if (bundle->bond) { | |
2648 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2649 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2650 | } |
2651 | } else { | |
2652 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2653 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2654 | } |
2655 | ||
2656 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2657 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2658 | } |
2659 | } else { | |
03366a2d | 2660 | bond_unref(bundle->bond); |
abe529af BP |
2661 | bundle->bond = NULL; |
2662 | } | |
2663 | ||
2664 | /* If we changed something that would affect MAC learning, un-learn | |
2665 | * everything on this port and force flow revalidation. */ | |
2666 | if (need_flush) { | |
b44a10b7 | 2667 | bundle_flush_macs(bundle, false); |
abe529af BP |
2668 | } |
2669 | ||
2670 | return 0; | |
2671 | } | |
2672 | ||
2673 | static void | |
2674 | bundle_remove(struct ofport *port_) | |
2675 | { | |
2676 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2677 | struct ofbundle *bundle = port->bundle; | |
2678 | ||
2679 | if (bundle) { | |
2680 | bundle_del_port(port); | |
2681 | if (list_is_empty(&bundle->ports)) { | |
2682 | bundle_destroy(bundle); | |
2683 | } else if (list_is_short(&bundle->ports)) { | |
03366a2d | 2684 | bond_unref(bundle->bond); |
abe529af BP |
2685 | bundle->bond = NULL; |
2686 | } | |
2687 | } | |
2688 | } | |
2689 | ||
2690 | static void | |
5f877369 | 2691 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2692 | { |
2693 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2694 | struct ofport_dpif *port = port_; | |
2695 | uint8_t ea[ETH_ADDR_LEN]; | |
2696 | int error; | |
2697 | ||
2698 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2699 | if (!error) { | |
abe529af | 2700 | struct ofpbuf packet; |
5f877369 | 2701 | void *packet_pdu; |
abe529af BP |
2702 | |
2703 | ofpbuf_init(&packet, 0); | |
2704 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2705 | pdu_size); |
2706 | memcpy(packet_pdu, pdu, pdu_size); | |
2707 | ||
97d6520b | 2708 | send_packet(port, &packet); |
abe529af BP |
2709 | ofpbuf_uninit(&packet); |
2710 | } else { | |
2711 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2712 | "%s (%s)", port->bundle->name, | |
10a89ef0 | 2713 | netdev_get_name(port->up.netdev), ovs_strerror(error)); |
abe529af BP |
2714 | } |
2715 | } | |
2716 | ||
2717 | static void | |
2718 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2719 | { | |
2720 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2721 | int error, n_packets, n_errors; | |
2722 | struct mac_entry *e; | |
2723 | ||
2724 | error = n_packets = n_errors = 0; | |
509c0149 | 2725 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
2726 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
2727 | if (e->port.p != bundle) { | |
ea131871 JG |
2728 | struct ofpbuf *learning_packet; |
2729 | struct ofport_dpif *port; | |
4dd1e3ca | 2730 | void *port_void; |
ea131871 JG |
2731 | int ret; |
2732 | ||
4dd1e3ca BP |
2733 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2734 | * struct ofport_dpif more effective. */ | |
2735 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2736 | e->mac, e->vlan, | |
2737 | &port_void); | |
2738 | port = port_void; | |
97d6520b | 2739 | ret = send_packet(port, learning_packet); |
ea131871 | 2740 | ofpbuf_delete(learning_packet); |
abe529af BP |
2741 | if (ret) { |
2742 | error = ret; | |
2743 | n_errors++; | |
2744 | } | |
2745 | n_packets++; | |
2746 | } | |
2747 | } | |
509c0149 | 2748 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2749 | |
2750 | if (n_errors) { | |
2751 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2752 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2753 | "packets, last error was: %s", | |
10a89ef0 | 2754 | bundle->name, n_errors, n_packets, ovs_strerror(error)); |
abe529af BP |
2755 | } else { |
2756 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2757 | bundle->name, n_packets); | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | static void | |
2762 | bundle_run(struct ofbundle *bundle) | |
2763 | { | |
2764 | if (bundle->lacp) { | |
2765 | lacp_run(bundle->lacp, send_pdu_cb); | |
2766 | } | |
2767 | if (bundle->bond) { | |
2768 | struct ofport_dpif *port; | |
2769 | ||
2770 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2771 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2772 | } |
2773 | ||
4a1b8f30 EJ |
2774 | if (bond_run(bundle->bond, lacp_status(bundle->lacp))) { |
2775 | bundle->ofproto->backer->need_revalidate = REV_BOND; | |
2776 | } | |
2777 | ||
abe529af BP |
2778 | if (bond_should_send_learning_packets(bundle->bond)) { |
2779 | bundle_send_learning_packets(bundle); | |
2780 | } | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | static void | |
2785 | bundle_wait(struct ofbundle *bundle) | |
2786 | { | |
2787 | if (bundle->lacp) { | |
2788 | lacp_wait(bundle->lacp); | |
2789 | } | |
2790 | if (bundle->bond) { | |
2791 | bond_wait(bundle->bond); | |
2792 | } | |
2793 | } | |
2794 | \f | |
2795 | /* Mirrors. */ | |
2796 | ||
2797 | static int | |
ec7ceaed EJ |
2798 | mirror_set__(struct ofproto *ofproto_, void *aux, |
2799 | const struct ofproto_mirror_settings *s) | |
abe529af BP |
2800 | { |
2801 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ec7ceaed EJ |
2802 | struct ofbundle **srcs, **dsts; |
2803 | int error; | |
2804 | size_t i; | |
abe529af | 2805 | |
abe529af | 2806 | if (!s) { |
ec7ceaed | 2807 | mirror_destroy(ofproto->mbridge, aux); |
abe529af BP |
2808 | return 0; |
2809 | } | |
2810 | ||
ec7ceaed EJ |
2811 | srcs = xmalloc(s->n_srcs * sizeof *srcs); |
2812 | dsts = xmalloc(s->n_dsts * sizeof *dsts); | |
abe529af | 2813 | |
ec7ceaed EJ |
2814 | for (i = 0; i < s->n_srcs; i++) { |
2815 | srcs[i] = bundle_lookup(ofproto, s->srcs[i]); | |
abe529af BP |
2816 | } |
2817 | ||
ec7ceaed EJ |
2818 | for (i = 0; i < s->n_dsts; i++) { |
2819 | dsts[i] = bundle_lookup(ofproto, s->dsts[i]); | |
abe529af BP |
2820 | } |
2821 | ||
ec7ceaed EJ |
2822 | error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts, |
2823 | s->n_dsts, s->src_vlans, | |
2824 | bundle_lookup(ofproto, s->out_bundle), s->out_vlan); | |
2825 | free(srcs); | |
2826 | free(dsts); | |
2827 | return error; | |
abe529af BP |
2828 | } |
2829 | ||
9d24de3b | 2830 | static int |
ec7ceaed EJ |
2831 | mirror_get_stats__(struct ofproto *ofproto, void *aux, |
2832 | uint64_t *packets, uint64_t *bytes) | |
9d24de3b | 2833 | { |
8844e035 | 2834 | push_all_stats(); |
ec7ceaed EJ |
2835 | return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets, |
2836 | bytes); | |
9d24de3b JP |
2837 | } |
2838 | ||
abe529af BP |
2839 | static int |
2840 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2841 | { | |
2842 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2843 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
abe529af | 2844 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { |
30618594 | 2845 | mac_learning_flush(ofproto->ml); |
abe529af | 2846 | } |
509c0149 | 2847 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2848 | return 0; |
2849 | } | |
2850 | ||
2851 | static bool | |
b4affc74 | 2852 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2853 | { |
2854 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2855 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
ec7ceaed | 2856 | return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0; |
abe529af | 2857 | } |
8402c74b SS |
2858 | |
2859 | static void | |
b53055f4 | 2860 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2861 | { |
2862 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2863 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2864 | } |
e764773c BP |
2865 | |
2866 | static void | |
c4069512 BP |
2867 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2868 | size_t max_entries) | |
e764773c BP |
2869 | { |
2870 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2871 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
e764773c | 2872 | mac_learning_set_idle_time(ofproto->ml, idle_time); |
c4069512 | 2873 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
509c0149 | 2874 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
e764773c | 2875 | } |
abe529af BP |
2876 | \f |
2877 | /* Ports. */ | |
2878 | ||
46c88433 | 2879 | static struct ofport_dpif * |
4e022ec0 | 2880 | get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
abe529af | 2881 | { |
7df6a8bd BP |
2882 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2883 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2884 | } |
2885 | ||
46c88433 | 2886 | static struct ofport_dpif * |
4e022ec0 | 2887 | get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) |
abe529af | 2888 | { |
7c33b188 JR |
2889 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2890 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2891 | } |
2892 | ||
2893 | static void | |
e1b1d06a JP |
2894 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2895 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2896 | struct dpif_port *dpif_port) |
2897 | { | |
2898 | ofproto_port->name = dpif_port->name; | |
2899 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2900 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2901 | } |
2902 | ||
6cbbf4fa EJ |
2903 | static void |
2904 | ofport_update_peer(struct ofport_dpif *ofport) | |
0a740f48 EJ |
2905 | { |
2906 | const struct ofproto_dpif *ofproto; | |
6cbbf4fa | 2907 | struct dpif_backer *backer; |
161b6042 | 2908 | char *peer_name; |
6cbbf4fa EJ |
2909 | |
2910 | if (!netdev_vport_is_patch(ofport->up.netdev)) { | |
2911 | return; | |
2912 | } | |
2913 | ||
2914 | backer = ofproto_dpif_cast(ofport->up.ofproto)->backer; | |
7894e7da | 2915 | backer->need_revalidate = REV_RECONFIGURE; |
0a740f48 | 2916 | |
6cbbf4fa EJ |
2917 | if (ofport->peer) { |
2918 | ofport->peer->peer = NULL; | |
2919 | ofport->peer = NULL; | |
2920 | } | |
2921 | ||
2922 | peer_name = netdev_vport_patch_peer(ofport->up.netdev); | |
2923 | if (!peer_name) { | |
2924 | return; | |
0a740f48 EJ |
2925 | } |
2926 | ||
2927 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6cbbf4fa EJ |
2928 | struct ofport *peer_ofport; |
2929 | struct ofport_dpif *peer; | |
161b6042 | 2930 | char *peer_peer; |
0a740f48 | 2931 | |
462abef2 EJ |
2932 | if (ofproto->backer != backer) { |
2933 | continue; | |
2934 | } | |
2935 | ||
6cbbf4fa EJ |
2936 | peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name); |
2937 | if (!peer_ofport) { | |
2938 | continue; | |
2939 | } | |
2940 | ||
2941 | peer = ofport_dpif_cast(peer_ofport); | |
2942 | peer_peer = netdev_vport_patch_peer(peer->up.netdev); | |
2943 | if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev), | |
2944 | peer_peer)) { | |
2945 | ofport->peer = peer; | |
2946 | ofport->peer->peer = ofport; | |
0a740f48 | 2947 | } |
161b6042 | 2948 | free(peer_peer); |
6cbbf4fa | 2949 | |
161b6042 | 2950 | break; |
0a740f48 | 2951 | } |
161b6042 | 2952 | free(peer_name); |
0a740f48 EJ |
2953 | } |
2954 | ||
0aa66d6e EJ |
2955 | static void |
2956 | port_run_fast(struct ofport_dpif *ofport) | |
2957 | { | |
2958 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2959 | struct ofpbuf packet; | |
2960 | ||
2961 | ofpbuf_init(&packet, 0); | |
2962 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2963 | send_packet(ofport, &packet); | |
2964 | ofpbuf_uninit(&packet); | |
2965 | } | |
ccc09689 EJ |
2966 | |
2967 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
2968 | struct ofpbuf packet; | |
2969 | ||
2970 | ofpbuf_init(&packet, 0); | |
2971 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
2972 | send_packet(ofport, &packet); | |
2973 | ofpbuf_uninit(&packet); | |
2974 | } | |
0aa66d6e EJ |
2975 | } |
2976 | ||
abe529af BP |
2977 | static void |
2978 | port_run(struct ofport_dpif *ofport) | |
2979 | { | |
3e5b3fdb EJ |
2980 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2981 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2982 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2983 | ||
3e5b3fdb EJ |
2984 | ofport->carrier_seq = carrier_seq; |
2985 | ||
0aa66d6e | 2986 | port_run_fast(ofport); |
b9ad7294 | 2987 | |
abe529af | 2988 | if (ofport->cfm) { |
4653c558 EJ |
2989 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2990 | ||
abe529af | 2991 | cfm_run(ofport->cfm); |
4653c558 EJ |
2992 | enable = enable && !cfm_get_fault(ofport->cfm); |
2993 | ||
2994 | if (cfm_opup >= 0) { | |
2995 | enable = enable && cfm_opup; | |
2996 | } | |
abe529af | 2997 | } |
015e08bc | 2998 | |
ccc09689 EJ |
2999 | if (ofport->bfd) { |
3000 | bfd_run(ofport->bfd); | |
3001 | enable = enable && bfd_forwarding(ofport->bfd); | |
3002 | } | |
3003 | ||
015e08bc EJ |
3004 | if (ofport->bundle) { |
3005 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
3006 | if (carrier_changed) { |
3007 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3008 | } | |
015e08bc EJ |
3009 | } |
3010 | ||
daff3353 EJ |
3011 | if (ofport->may_enable != enable) { |
3012 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
83341407 | 3013 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3014 | } |
3015 | ||
015e08bc | 3016 | ofport->may_enable = enable; |
abe529af BP |
3017 | } |
3018 | ||
3019 | static void | |
3020 | port_wait(struct ofport_dpif *ofport) | |
3021 | { | |
3022 | if (ofport->cfm) { | |
3023 | cfm_wait(ofport->cfm); | |
3024 | } | |
ccc09689 EJ |
3025 | |
3026 | if (ofport->bfd) { | |
3027 | bfd_wait(ofport->bfd); | |
3028 | } | |
abe529af BP |
3029 | } |
3030 | ||
3031 | static int | |
3032 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3033 | struct ofproto_port *ofproto_port) | |
3034 | { | |
3035 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3036 | struct dpif_port dpif_port; | |
3037 | int error; | |
3038 | ||
0a740f48 EJ |
3039 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3040 | const char *type = netdev_get_type_from_name(devname); | |
3041 | ||
3042 | /* We may be called before ofproto->up.port_by_name is populated with | |
3043 | * the appropriate ofport. For this reason, we must get the name and | |
3044 | * type from the netdev layer directly. */ | |
3045 | if (type) { | |
3046 | const struct ofport *ofport; | |
3047 | ||
3048 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3049 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3050 | ofproto_port->name = xstrdup(devname); | |
3051 | ofproto_port->type = xstrdup(type); | |
3052 | return 0; | |
3053 | } | |
3054 | return ENODEV; | |
3055 | } | |
3056 | ||
acf60855 JP |
3057 | if (!sset_contains(&ofproto->ports, devname)) { |
3058 | return ENODEV; | |
3059 | } | |
3060 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3061 | devname, &dpif_port); | |
abe529af | 3062 | if (!error) { |
e1b1d06a | 3063 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3064 | } |
3065 | return error; | |
3066 | } | |
3067 | ||
3068 | static int | |
e1b1d06a | 3069 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3070 | { |
3071 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3072 | const char *devname = netdev_get_name(netdev); |
3aa30359 BP |
3073 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
3074 | const char *dp_port_name; | |
abe529af | 3075 | |
0a740f48 EJ |
3076 | if (netdev_vport_is_patch(netdev)) { |
3077 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3078 | return 0; | |
3079 | } | |
3080 | ||
3aa30359 | 3081 | dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
b9ad7294 | 3082 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
4e022ec0 | 3083 | odp_port_t port_no = ODPP_NONE; |
7d82ab2e KM |
3084 | int error; |
3085 | ||
3086 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3087 | if (error) { |
3088 | return error; | |
3089 | } | |
7d82ab2e | 3090 | if (netdev_get_tunnel_config(netdev)) { |
4e022ec0 AW |
3091 | simap_put(&ofproto->backer->tnl_backers, |
3092 | dp_port_name, odp_to_u32(port_no)); | |
7d82ab2e | 3093 | } |
acf60855 | 3094 | } |
b9ad7294 EJ |
3095 | |
3096 | if (netdev_get_tunnel_config(netdev)) { | |
3097 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3098 | } else { |
3099 | sset_add(&ofproto->ports, devname); | |
3100 | } | |
3101 | return 0; | |
3102 | } | |
3103 | ||
abe529af | 3104 | static int |
4e022ec0 | 3105 | port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) |
abe529af BP |
3106 | { |
3107 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3108 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3109 | int error = 0; |
abe529af | 3110 | |
b9ad7294 EJ |
3111 | if (!ofport) { |
3112 | return 0; | |
e1b1d06a | 3113 | } |
b9ad7294 EJ |
3114 | |
3115 | sset_find_and_delete(&ofproto->ghost_ports, | |
3116 | netdev_get_name(ofport->up.netdev)); | |
a614d823 | 3117 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
42943cde | 3118 | if (!ofport->is_tunnel) { |
b9ad7294 EJ |
3119 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3120 | if (!error) { | |
abe529af BP |
3121 | /* The caller is going to close ofport->up.netdev. If this is a |
3122 | * bonded port, then the bond is using that netdev, so remove it | |
3123 | * from the bond. The client will need to reconfigure everything | |
3124 | * after deleting ports, so then the slave will get re-added. */ | |
3125 | bundle_remove(&ofport->up); | |
3126 | } | |
3127 | } | |
3128 | return error; | |
3129 | } | |
3130 | ||
6527c598 PS |
3131 | static int |
3132 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3133 | { | |
3134 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3135 | int error; | |
3136 | ||
8844e035 EJ |
3137 | push_all_stats(); |
3138 | ||
6527c598 PS |
3139 | error = netdev_get_stats(ofport->up.netdev, stats); |
3140 | ||
ee382d89 | 3141 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3142 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3143 | ||
3144 | /* ofproto->stats.tx_packets represents packets that we created | |
3145 | * internally and sent to some port (e.g. packets sent with | |
3146 | * send_packet()). Account for them as if they had come from | |
3147 | * OFPP_LOCAL and got forwarded. */ | |
3148 | ||
3149 | if (stats->rx_packets != UINT64_MAX) { | |
3150 | stats->rx_packets += ofproto->stats.tx_packets; | |
3151 | } | |
3152 | ||
3153 | if (stats->rx_bytes != UINT64_MAX) { | |
3154 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3155 | } | |
3156 | ||
3157 | /* ofproto->stats.rx_packets represents packets that were received on | |
3158 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3159 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3160 | |
3161 | if (stats->tx_packets != UINT64_MAX) { | |
3162 | stats->tx_packets += ofproto->stats.rx_packets; | |
3163 | } | |
3164 | ||
3165 | if (stats->tx_bytes != UINT64_MAX) { | |
3166 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3167 | } | |
3168 | } | |
3169 | ||
3170 | return error; | |
3171 | } | |
3172 | ||
abe529af | 3173 | struct port_dump_state { |
acf60855 JP |
3174 | uint32_t bucket; |
3175 | uint32_t offset; | |
0a740f48 | 3176 | bool ghost; |
da78d43d BP |
3177 | |
3178 | struct ofproto_port port; | |
3179 | bool has_port; | |
abe529af BP |
3180 | }; |
3181 | ||
3182 | static int | |
acf60855 | 3183 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3184 | { |
0a740f48 | 3185 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3186 | return 0; |
3187 | } | |
3188 | ||
3189 | static int | |
b9ad7294 | 3190 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3191 | struct ofproto_port *port) |
3192 | { | |
e1b1d06a | 3193 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3194 | struct port_dump_state *state = state_; |
0a740f48 | 3195 | const struct sset *sset; |
acf60855 | 3196 | struct sset_node *node; |
abe529af | 3197 | |
da78d43d BP |
3198 | if (state->has_port) { |
3199 | ofproto_port_destroy(&state->port); | |
3200 | state->has_port = false; | |
3201 | } | |
0a740f48 EJ |
3202 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3203 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3204 | int error; |
3205 | ||
da78d43d BP |
3206 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3207 | if (!error) { | |
3208 | *port = state->port; | |
3209 | state->has_port = true; | |
3210 | return 0; | |
3211 | } else if (error != ENODEV) { | |
acf60855 JP |
3212 | return error; |
3213 | } | |
abe529af | 3214 | } |
acf60855 | 3215 | |
0a740f48 EJ |
3216 | if (!state->ghost) { |
3217 | state->ghost = true; | |
3218 | state->bucket = 0; | |
3219 | state->offset = 0; | |
3220 | return port_dump_next(ofproto_, state_, port); | |
3221 | } | |
3222 | ||
acf60855 | 3223 | return EOF; |
abe529af BP |
3224 | } |
3225 | ||
3226 | static int | |
3227 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3228 | { | |
3229 | struct port_dump_state *state = state_; | |
3230 | ||
da78d43d BP |
3231 | if (state->has_port) { |
3232 | ofproto_port_destroy(&state->port); | |
3233 | } | |
abe529af BP |
3234 | free(state); |
3235 | return 0; | |
3236 | } | |
3237 | ||
3238 | static int | |
3239 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3240 | { | |
3241 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3242 | |
3243 | if (ofproto->port_poll_errno) { | |
3244 | int error = ofproto->port_poll_errno; | |
3245 | ofproto->port_poll_errno = 0; | |
3246 | return error; | |
3247 | } | |
3248 | ||
3249 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3250 | return EAGAIN; | |
3251 | } | |
3252 | ||
3253 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3254 | return 0; | |
abe529af BP |
3255 | } |
3256 | ||
3257 | static void | |
3258 | port_poll_wait(const struct ofproto *ofproto_) | |
3259 | { | |
3260 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3261 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3262 | } |
3263 | ||
3264 | static int | |
3265 | port_is_lacp_current(const struct ofport *ofport_) | |
3266 | { | |
3267 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3268 | return (ofport->bundle && ofport->bundle->lacp | |
3269 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3270 | : -1); | |
3271 | } | |
3272 | \f | |
3273 | /* Upcall handling. */ | |
3274 | ||
501f8d1f BP |
3275 | /* Flow miss batching. |
3276 | * | |
3277 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3278 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3279 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3280 | * more packets, plus possibly installing the flow in the dpif. | |
3281 | * | |
3282 | * So far we only batch the operations that affect flow setup time the most. | |
3283 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3284 | struct flow_miss { | |
3285 | struct hmap_node hmap_node; | |
acf60855 | 3286 | struct ofproto_dpif *ofproto; |
501f8d1f | 3287 | struct flow flow; |
b0f7b9b5 | 3288 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3289 | const struct nlattr *key; |
3290 | size_t key_len; | |
3291 | struct list packets; | |
6a7e895f | 3292 | enum dpif_upcall_type upcall_type; |
501f8d1f BP |
3293 | }; |
3294 | ||
3295 | struct flow_miss_op { | |
c2b565b5 | 3296 | struct dpif_op dpif_op; |
bbafd73b EJ |
3297 | |
3298 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3299 | struct xlate_out xout; | |
3300 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
9b1a48c2 | 3301 | |
d445cc16 JP |
3302 | struct ofpbuf mask; /* Flow mask for "put" ops. */ |
3303 | struct odputil_keybuf maskbuf; | |
3304 | ||
9b1a48c2 JP |
3305 | /* If this is a "put" op, then a pointer to the subfacet that should |
3306 | * be marked as uninstalled if the operation fails. */ | |
3307 | struct subfacet *subfacet; | |
501f8d1f BP |
3308 | }; |
3309 | ||
62cd7072 BP |
3310 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3311 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3312 | * configurations. */ |
62cd7072 | 3313 | static void |
a39edbd4 | 3314 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3315 | const struct flow *flow) |
62cd7072 BP |
3316 | { |
3317 | struct ofputil_packet_in pin; | |
3318 | ||
3e3252fa EJ |
3319 | pin.packet = packet->data; |
3320 | pin.packet_len = packet->size; | |
62cd7072 | 3321 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3322 | pin.controller_id = 0; |
54834960 EJ |
3323 | |
3324 | pin.table_id = 0; | |
3325 | pin.cookie = 0; | |
3326 | ||
62cd7072 | 3327 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3328 | |
3329 | flow_get_metadata(flow, &pin.fmd); | |
3330 | ||
d8653c38 | 3331 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3332 | } |
3333 | ||
501f8d1f | 3334 | static struct flow_miss * |
ddbc5954 JP |
3335 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3336 | const struct flow *flow, uint32_t hash) | |
abe529af | 3337 | { |
501f8d1f | 3338 | struct flow_miss *miss; |
abe529af | 3339 | |
501f8d1f | 3340 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3341 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3342 | return miss; |
3343 | } | |
3344 | } | |
abe529af | 3345 | |
b23cdad9 | 3346 | return NULL; |
501f8d1f | 3347 | } |
abe529af | 3348 | |
9d6ac44e BP |
3349 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3350 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3351 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3352 | * returned op->subfacet, and if anything needs to be freed after processing | |
3353 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3354 | static void |
9d6ac44e BP |
3355 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3356 | struct flow_miss_op *op) | |
501f8d1f | 3357 | { |
4e022ec0 AW |
3358 | if (miss->flow.in_port.ofp_port |
3359 | != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port, | |
47b0deae | 3360 | miss->flow.vlan_tci)) { |
9d6ac44e BP |
3361 | /* This packet was received on a VLAN splinter port. We |
3362 | * added a VLAN to the packet to make the packet resemble | |
3363 | * the flow, but the actions were composed assuming that | |
3364 | * the packet contained no VLAN. So, we must remove the | |
3365 | * VLAN header from the packet before trying to execute the | |
3366 | * actions. */ | |
3367 | eth_pop_vlan(packet); | |
3368 | } | |
3369 | ||
9b1a48c2 | 3370 | op->subfacet = NULL; |
bbafd73b | 3371 | op->xout_garbage = false; |
9d6ac44e BP |
3372 | op->dpif_op.type = DPIF_OP_EXECUTE; |
3373 | op->dpif_op.u.execute.key = miss->key; | |
3374 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3375 | op->dpif_op.u.execute.packet = packet; | |
d445cc16 | 3376 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
9d6ac44e BP |
3377 | } |
3378 | ||
3379 | /* Helper for handle_flow_miss_without_facet() and | |
3380 | * handle_flow_miss_with_facet(). */ | |
3381 | static void | |
b7e2f6b3 EJ |
3382 | handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet, |
3383 | const struct flow *flow, bool fail_open) | |
9d6ac44e | 3384 | { |
b7e2f6b3 | 3385 | if (fail_open) { |
9d6ac44e BP |
3386 | /* |
3387 | * Extra-special case for fail-open mode. | |
3388 | * | |
3389 | * We are in fail-open mode and the packet matched the fail-open | |
3390 | * rule, but we are connected to a controller too. We should send | |
3391 | * the packet up to the controller in the hope that it will try to | |
3392 | * set up a flow and thereby allow us to exit fail-open. | |
3393 | * | |
3394 | * See the top-level comment in fail-open.c for more information. | |
3395 | */ | |
3396 | send_packet_in_miss(ofproto, packet, flow); | |
3397 | } | |
3398 | } | |
3399 | ||
3400 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
bcd2633a JP |
3401 | * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace |
3402 | * and (usually) installing a datapath flow. The answer is usually "yes" (a | |
3403 | * return value of true). However, for short flows the cost of bookkeeping is | |
3404 | * much higher than the benefits, so when the datapath holds a large number of | |
3405 | * flows we impose some heuristics to decide which flows are likely to be worth | |
3406 | * tracking. */ | |
9d6ac44e | 3407 | static bool |
bcd2633a | 3408 | flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc) |
9d6ac44e | 3409 | { |
04d08d54 | 3410 | struct dpif_backer *backer = miss->ofproto->backer; |
bcd2633a | 3411 | uint32_t hash; |
04d08d54 | 3412 | |
7155fa52 JS |
3413 | switch (flow_miss_model) { |
3414 | case OFPROTO_HANDLE_MISS_AUTO: | |
3415 | break; | |
3416 | case OFPROTO_HANDLE_MISS_WITH_FACETS: | |
3417 | return true; | |
3418 | case OFPROTO_HANDLE_MISS_WITHOUT_FACETS: | |
3419 | return false; | |
3420 | } | |
3421 | ||
04d08d54 | 3422 | if (!backer->governor) { |
9d6ac44e BP |
3423 | size_t n_subfacets; |
3424 | ||
04d08d54 | 3425 | n_subfacets = hmap_count(&backer->subfacets); |
380f49c4 | 3426 | if (n_subfacets * 2 <= flow_eviction_threshold) { |
9d6ac44e BP |
3427 | return true; |
3428 | } | |
3429 | ||
c985ec94 | 3430 | backer->governor = governor_create(); |
9d6ac44e BP |
3431 | } |
3432 | ||
bcd2633a | 3433 | hash = flow_hash_in_wildcards(&miss->flow, wc, 0); |
04d08d54 | 3434 | return governor_should_install_flow(backer->governor, hash, |
9d6ac44e BP |
3435 | list_size(&miss->packets)); |
3436 | } | |
3437 | ||
bcd2633a JP |
3438 | /* Handles 'miss' without creating a facet or subfacet or creating any datapath |
3439 | * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'. | |
3440 | * May add an "execute" operation to 'ops' and increment '*n_ops'. */ | |
9d6ac44e | 3441 | static void |
bcd2633a JP |
3442 | handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout, |
3443 | struct flow_miss *miss, | |
9d6ac44e BP |
3444 | struct flow_miss_op *ops, size_t *n_ops) |
3445 | { | |
530a1d91 | 3446 | struct ofpbuf *packet; |
2b459b83 | 3447 | |
9d6ac44e | 3448 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
abe529af | 3449 | |
9d6ac44e | 3450 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3451 | |
b7e2f6b3 EJ |
3452 | handle_flow_miss_common(miss->ofproto, packet, &miss->flow, |
3453 | rule->up.cr.priority == FAIL_OPEN_PRIORITY); | |
7820e253 | 3454 | |
bcd2633a JP |
3455 | if (xout->slow) { |
3456 | struct xlate_in xin; | |
abe529af | 3457 | |
47b0deae | 3458 | xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet); |
bcd2633a JP |
3459 | xlate_actions_for_side_effects(&xin); |
3460 | } | |
abe529af | 3461 | |
bcd2633a JP |
3462 | if (xout->odp_actions.size) { |
3463 | struct flow_miss_op *op = &ops[*n_ops]; | |
9d6ac44e BP |
3464 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
3465 | ||
3466 | init_flow_miss_execute_op(miss, packet, op); | |
bcd2633a | 3467 | xlate_out_copy(&op->xout, xout); |
bbafd73b EJ |
3468 | execute->actions = op->xout.odp_actions.data; |
3469 | execute->actions_len = op->xout.odp_actions.size; | |
3470 | op->xout_garbage = true; | |
9d6ac44e BP |
3471 | |
3472 | (*n_ops)++; | |
9d6ac44e | 3473 | } |
abe529af | 3474 | } |
9d6ac44e BP |
3475 | } |
3476 | ||
3477 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3478 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3479 | * | |
3480 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3481 | * This is really important only for new facets: if we just called time_msec() | |
3482 | * here, then the new subfacet or its packets could look (occasionally) as | |
3483 | * though it was used some time after the facet was used. That can make a | |
3484 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
bcd2633a JP |
3485 | * e.g. NetFlow statistics. |
3486 | * | |
3487 | * If non-null, 'stats' will be folded into 'facet'. */ | |
9d6ac44e BP |
3488 | static void |
3489 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
bcd2633a | 3490 | long long int now, struct dpif_flow_stats *stats, |
9d6ac44e BP |
3491 | struct flow_miss_op *ops, size_t *n_ops) |
3492 | { | |
6a7e895f | 3493 | enum subfacet_path want_path; |
9d6ac44e BP |
3494 | struct subfacet *subfacet; |
3495 | struct ofpbuf *packet; | |
abe529af | 3496 | |
bcd2633a | 3497 | want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
b0f7b9b5 | 3498 | |
530a1d91 | 3499 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3500 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 3501 | |
b7e2f6b3 EJ |
3502 | handle_flow_miss_common(miss->ofproto, packet, &miss->flow, |
3503 | facet->fail_open); | |
501f8d1f | 3504 | |
4dff9097 | 3505 | if (want_path != SF_FAST_PATH) { |
b7e2f6b3 | 3506 | struct rule_dpif *rule; |
bbafd73b | 3507 | struct xlate_in xin; |
454a77e5 | 3508 | |
b7e2f6b3 EJ |
3509 | rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); |
3510 | xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet); | |
bbafd73b | 3511 | xlate_actions_for_side_effects(&xin); |
501f8d1f | 3512 | } |
67d91f78 | 3513 | |
bbafd73b | 3514 | if (facet->xout.odp_actions.size) { |
9d6ac44e | 3515 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
8338659a | 3516 | |
9d6ac44e | 3517 | init_flow_miss_execute_op(miss, packet, op); |
bbafd73b EJ |
3518 | execute->actions = facet->xout.odp_actions.data, |
3519 | execute->actions_len = facet->xout.odp_actions.size; | |
9d6ac44e | 3520 | (*n_ops)++; |
5fe20d5d | 3521 | } |
501f8d1f BP |
3522 | } |
3523 | ||
a41d7bf2 JP |
3524 | /* Don't install the flow if it's the result of the "userspace" |
3525 | * action for an already installed facet. This can occur when a | |
3526 | * datapath flow with wildcards has a "userspace" action and flows | |
3527 | * sent to userspace result in a different subfacet, which will then | |
3528 | * be rejected as overlapping by the datapath. */ | |
3529 | if (miss->upcall_type == DPIF_UC_ACTION | |
3530 | && !list_is_empty(&facet->subfacets)) { | |
3531 | if (stats) { | |
3532 | facet->used = MAX(facet->used, stats->used); | |
3533 | facet->packet_count += stats->n_packets; | |
3534 | facet->byte_count += stats->n_bytes; | |
3535 | facet->tcp_flags |= stats->tcp_flags; | |
3536 | } | |
3537 | return; | |
3538 | } | |
3539 | ||
3540 | subfacet = subfacet_create(facet, miss, now); | |
3541 | if (stats) { | |
3542 | subfacet_update_stats(subfacet, stats); | |
3543 | } | |
3544 | ||
24ce4b08 | 3545 | if (subfacet->path != want_path) { |
501f8d1f | 3546 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3547 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3548 | |
c84451a6 EJ |
3549 | subfacet->path = want_path; |
3550 | ||
d445cc16 | 3551 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
6b35e630 JP |
3552 | if (enable_megaflows) { |
3553 | odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks, | |
3554 | &miss->flow, UINT32_MAX); | |
3555 | } | |
d445cc16 | 3556 | |
bbafd73b | 3557 | op->xout_garbage = false; |
c2b565b5 | 3558 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
9b1a48c2 | 3559 | op->subfacet = subfacet; |
0f74f625 | 3560 | put->flags = DPIF_FP_CREATE; |
501f8d1f BP |
3561 | put->key = miss->key; |
3562 | put->key_len = miss->key_len; | |
d445cc16 JP |
3563 | put->mask = op->mask.data; |
3564 | put->mask_len = op->mask.size; | |
3565 | ||
6a7e895f | 3566 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3567 | put->actions = facet->xout.odp_actions.data; |
3568 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3569 | } else { |
b7e2f6b3 | 3570 | compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow, |
bbafd73b | 3571 | op->slow_stub, sizeof op->slow_stub, |
6a7e895f BP |
3572 | &put->actions, &put->actions_len); |
3573 | } | |
501f8d1f BP |
3574 | put->stats = NULL; |
3575 | } | |
3576 | } | |
3577 | ||
acf60855 JP |
3578 | /* Handles flow miss 'miss'. May add any required datapath operations |
3579 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3580 | static void |
acf60855 JP |
3581 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3582 | size_t *n_ops) | |
9d6ac44e | 3583 | { |
acf60855 | 3584 | struct ofproto_dpif *ofproto = miss->ofproto; |
bcd2633a JP |
3585 | struct dpif_flow_stats stats__; |
3586 | struct dpif_flow_stats *stats = &stats__; | |
3587 | struct ofpbuf *packet; | |
9d6ac44e | 3588 | struct facet *facet; |
459b16a1 | 3589 | long long int now; |
9d6ac44e | 3590 | |
bcd2633a JP |
3591 | now = time_msec(); |
3592 | memset(stats, 0, sizeof *stats); | |
3593 | stats->used = now; | |
3594 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
3595 | stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow); | |
3596 | stats->n_bytes += packet->size; | |
3597 | stats->n_packets++; | |
3598 | } | |
9d6ac44e | 3599 | |
bcd2633a | 3600 | facet = facet_lookup_valid(ofproto, &miss->flow); |
9d6ac44e | 3601 | if (!facet) { |
bcd2633a JP |
3602 | struct flow_wildcards wc; |
3603 | struct rule_dpif *rule; | |
3604 | struct xlate_out xout; | |
3605 | struct xlate_in xin; | |
3606 | ||
3607 | flow_wildcards_init_catchall(&wc); | |
3608 | rule = rule_dpif_lookup(ofproto, &miss->flow, &wc); | |
3609 | rule_credit_stats(rule, stats); | |
3610 | ||
47b0deae EJ |
3611 | xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags, |
3612 | NULL); | |
bcd2633a JP |
3613 | xin.resubmit_stats = stats; |
3614 | xin.may_learn = true; | |
3615 | xlate_actions(&xin, &xout); | |
3616 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); | |
3617 | ||
ba33dd03 EJ |
3618 | /* There does not exist a bijection between 'struct flow' and datapath |
3619 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3620 | * assumption used throughout the facet and subfacet handling code. | |
3621 | * Since we have to handle these misses in userspace anyway, we simply | |
ec9f40dc | 3622 | * skip facet creation, avoiding the problem altogether. */ |
ba33dd03 | 3623 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE |
bcd2633a JP |
3624 | || !flow_miss_should_make_facet(miss, &xout.wc)) { |
3625 | handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops); | |
9d6ac44e BP |
3626 | return; |
3627 | } | |
3628 | ||
bcd2633a JP |
3629 | facet = facet_create(miss, rule, &xout, stats); |
3630 | stats = NULL; | |
9d6ac44e | 3631 | } |
bcd2633a | 3632 | handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops); |
9d6ac44e BP |
3633 | } |
3634 | ||
8f73d537 EJ |
3635 | static struct drop_key * |
3636 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3637 | size_t key_len) | |
3638 | { | |
3639 | struct drop_key *drop_key; | |
3640 | ||
3641 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3642 | &backer->drop_keys) { | |
3643 | if (drop_key->key_len == key_len | |
3644 | && !memcmp(drop_key->key, key, key_len)) { | |
3645 | return drop_key; | |
3646 | } | |
3647 | } | |
3648 | return NULL; | |
3649 | } | |
3650 | ||
3651 | static void | |
3652 | drop_key_clear(struct dpif_backer *backer) | |
3653 | { | |
3654 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3655 | struct drop_key *drop_key, *next; | |
3656 | ||
3657 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3658 | int error; | |
3659 | ||
3660 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3661 | NULL); | |
3662 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3663 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3664 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
10a89ef0 BP |
3665 | VLOG_WARN("Failed to delete drop key (%s) (%s)", |
3666 | ovs_strerror(error), ds_cstr(&ds)); | |
8f73d537 EJ |
3667 | ds_destroy(&ds); |
3668 | } | |
3669 | ||
3670 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3671 | free(drop_key->key); | |
3672 | free(drop_key); | |
3673 | } | |
3674 | } | |
3675 | ||
501f8d1f | 3676 | static void |
acf60855 | 3677 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3678 | size_t n_upcalls) |
3679 | { | |
3680 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3681 | struct flow_miss *miss; |
3682 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3683 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3684 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3685 | struct hmap todo; |
b23cdad9 | 3686 | int n_misses; |
501f8d1f BP |
3687 | size_t n_ops; |
3688 | size_t i; | |
3689 | ||
3690 | if (!n_upcalls) { | |
3691 | return; | |
3692 | } | |
3693 | ||
3694 | /* Construct the to-do list. | |
3695 | * | |
3696 | * This just amounts to extracting the flow from each packet and sticking | |
3697 | * the packets that have the same flow in the same "flow_miss" structure so | |
3698 | * that we can process them together. */ | |
3699 | hmap_init(&todo); | |
b23cdad9 | 3700 | n_misses = 0; |
501f8d1f | 3701 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3702 | struct flow_miss *miss = &misses[n_misses]; |
3703 | struct flow_miss *existing_miss; | |
acf60855 | 3704 | struct ofproto_dpif *ofproto; |
4e022ec0 | 3705 | odp_port_t odp_in_port; |
1d446463 | 3706 | struct flow flow; |
b23cdad9 | 3707 | uint32_t hash; |
e09ee259 | 3708 | int error; |
501f8d1f | 3709 | |
8449c4d6 EJ |
3710 | error = xlate_receive(backer, upcall->packet, upcall->key, |
3711 | upcall->key_len, &flow, &miss->key_fitness, | |
3712 | &ofproto, &odp_in_port); | |
e09ee259 | 3713 | if (error == ENODEV) { |
8f73d537 EJ |
3714 | struct drop_key *drop_key; |
3715 | ||
c3385516 AW |
3716 | /* Received packet on datapath port for which we couldn't |
3717 | * associate an ofproto. This can happen if a port is removed | |
3718 | * while traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
3719 | * in case it happens frequently. Install a drop flow so |
3720 | * that future packets of the flow are inexpensively dropped | |
3721 | * in the kernel. */ | |
c3385516 AW |
3722 | VLOG_INFO_RL(&rl, "received packet on unassociated datapath port " |
3723 | "%"PRIu32, odp_in_port); | |
8f73d537 EJ |
3724 | |
3725 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
3726 | if (!drop_key) { | |
92e02bc6 AZ |
3727 | int ret; |
3728 | ret = dpif_flow_put(backer->dpif, | |
3729 | DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3730 | upcall->key, upcall->key_len, | |
3731 | NULL, 0, NULL, 0, NULL); | |
3732 | ||
3733 | if (!ret) { | |
3734 | drop_key = xmalloc(sizeof *drop_key); | |
3735 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
3736 | drop_key->key_len = upcall->key_len; | |
3737 | ||
3738 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3739 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3740 | } | |
8f73d537 EJ |
3741 | } |
3742 | continue; | |
acf60855 | 3743 | } |
e09ee259 | 3744 | if (error) { |
b0f7b9b5 BP |
3745 | continue; |
3746 | } | |
735d7efb AZ |
3747 | |
3748 | ofproto->n_missed++; | |
72e8bf28 | 3749 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
4e022ec0 | 3750 | &flow.tunnel, &flow.in_port, &miss->flow); |
501f8d1f | 3751 | |
501f8d1f | 3752 | /* Add other packets to a to-do list. */ |
b23cdad9 | 3753 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 3754 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
3755 | if (!existing_miss) { |
3756 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 3757 | miss->ofproto = ofproto; |
b23cdad9 BP |
3758 | miss->key = upcall->key; |
3759 | miss->key_len = upcall->key_len; | |
6a7e895f | 3760 | miss->upcall_type = upcall->type; |
b23cdad9 BP |
3761 | list_init(&miss->packets); |
3762 | ||
3763 | n_misses++; | |
3764 | } else { | |
3765 | miss = existing_miss; | |
3766 | } | |
501f8d1f BP |
3767 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3768 | } | |
3769 | ||
3770 | /* Process each element in the to-do list, constructing the set of | |
3771 | * operations to batch. */ | |
3772 | n_ops = 0; | |
33bb0caa | 3773 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 3774 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3775 | } |
cb22974d | 3776 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3777 | |
3778 | /* Execute batch. */ | |
3779 | for (i = 0; i < n_ops; i++) { | |
3780 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3781 | } | |
acf60855 | 3782 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 3783 | |
501f8d1f | 3784 | for (i = 0; i < n_ops; i++) { |
9b1a48c2 JP |
3785 | if (dpif_ops[i]->error != 0 |
3786 | && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT | |
3787 | && flow_miss_ops[i].subfacet) { | |
3788 | struct subfacet *subfacet = flow_miss_ops[i].subfacet; | |
3789 | ||
3790 | COVERAGE_INC(subfacet_install_fail); | |
3791 | ||
cfd5c9eb JP |
3792 | /* Zero-out subfacet counters when installation failed, but |
3793 | * datapath reported hits. This should not happen and | |
3794 | * indicates a bug, since if the datapath flow exists, we | |
3795 | * should not be attempting to create a new subfacet. A | |
3796 | * buggy datapath could trigger this, so just zero out the | |
3797 | * counters and log an error. */ | |
3798 | if (subfacet->dp_packet_count || subfacet->dp_byte_count) { | |
3799 | VLOG_ERR_RL(&rl, "failed to install subfacet for which " | |
3800 | "datapath reported hits"); | |
3801 | subfacet->dp_packet_count = subfacet->dp_byte_count = 0; | |
3802 | } | |
3803 | ||
9b1a48c2 JP |
3804 | subfacet->path = SF_NOT_INSTALLED; |
3805 | } | |
3806 | ||
3807 | /* Free memory. */ | |
bbafd73b EJ |
3808 | if (flow_miss_ops[i].xout_garbage) { |
3809 | xlate_out_uninit(&flow_miss_ops[i].xout); | |
3810 | } | |
501f8d1f | 3811 | } |
33bb0caa | 3812 | hmap_destroy(&todo); |
abe529af BP |
3813 | } |
3814 | ||
29089a54 RL |
3815 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, |
3816 | IPFIX_UPCALL } | |
6a7e895f BP |
3817 | classify_upcall(const struct dpif_upcall *upcall) |
3818 | { | |
29089a54 | 3819 | size_t userdata_len; |
6a7e895f BP |
3820 | union user_action_cookie cookie; |
3821 | ||
3822 | /* First look at the upcall type. */ | |
3823 | switch (upcall->type) { | |
3824 | case DPIF_UC_ACTION: | |
3825 | break; | |
3826 | ||
3827 | case DPIF_UC_MISS: | |
3828 | return MISS_UPCALL; | |
3829 | ||
3830 | case DPIF_N_UC_TYPES: | |
3831 | default: | |
3832 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3833 | return BAD_UPCALL; | |
3834 | } | |
3835 | ||
3836 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
3837 | if (!upcall->userdata) { |
3838 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
3839 | return BAD_UPCALL; | |
3840 | } | |
29089a54 RL |
3841 | userdata_len = nl_attr_get_size(upcall->userdata); |
3842 | if (userdata_len < sizeof cookie.type | |
3843 | || userdata_len > sizeof cookie) { | |
e995e3df | 3844 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", |
29089a54 | 3845 | userdata_len); |
e995e3df BP |
3846 | return BAD_UPCALL; |
3847 | } | |
29089a54 RL |
3848 | memset(&cookie, 0, sizeof cookie); |
3849 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
3850 | if (userdata_len == sizeof cookie.sflow | |
3851 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
6a7e895f | 3852 | return SFLOW_UPCALL; |
29089a54 RL |
3853 | } else if (userdata_len == sizeof cookie.slow_path |
3854 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
6a7e895f | 3855 | return MISS_UPCALL; |
29089a54 RL |
3856 | } else if (userdata_len == sizeof cookie.flow_sample |
3857 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
3858 | return FLOW_SAMPLE_UPCALL; | |
3859 | } else if (userdata_len == sizeof cookie.ipfix | |
3860 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
3861 | return IPFIX_UPCALL; | |
3862 | } else { | |
3863 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
3864 | " and size %zu", cookie.type, userdata_len); | |
6a7e895f BP |
3865 | return BAD_UPCALL; |
3866 | } | |
3867 | } | |
3868 | ||
abe529af | 3869 | static void |
acf60855 | 3870 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 3871 | const struct dpif_upcall *upcall) |
abe529af | 3872 | { |
acf60855 | 3873 | struct ofproto_dpif *ofproto; |
1673e0e4 | 3874 | union user_action_cookie cookie; |
e84173dc | 3875 | struct flow flow; |
4e022ec0 | 3876 | odp_port_t odp_in_port; |
abe529af | 3877 | |
8449c4d6 EJ |
3878 | if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
3879 | &flow, NULL, &ofproto, &odp_in_port) | |
e09ee259 | 3880 | || !ofproto->sflow) { |
e84173dc BP |
3881 | return; |
3882 | } | |
3883 | ||
29089a54 RL |
3884 | memset(&cookie, 0, sizeof cookie); |
3885 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
e1b1d06a JP |
3886 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
3887 | odp_in_port, &cookie); | |
6ff686f2 PS |
3888 | } |
3889 | ||
29089a54 RL |
3890 | static void |
3891 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
3892 | const struct dpif_upcall *upcall) | |
3893 | { | |
3894 | struct ofproto_dpif *ofproto; | |
3895 | union user_action_cookie cookie; | |
3896 | struct flow flow; | |
3897 | ||
8449c4d6 EJ |
3898 | if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
3899 | &flow, NULL, &ofproto, NULL) | |
29089a54 RL |
3900 | || !ofproto->ipfix) { |
3901 | return; | |
3902 | } | |
3903 | ||
3904 | memset(&cookie, 0, sizeof cookie); | |
3905 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
3906 | ||
3907 | /* The flow reflects exactly the contents of the packet. Sample | |
3908 | * the packet using it. */ | |
3909 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
3910 | cookie.flow_sample.collector_set_id, | |
3911 | cookie.flow_sample.probability, | |
3912 | cookie.flow_sample.obs_domain_id, | |
3913 | cookie.flow_sample.obs_point_id); | |
3914 | } | |
3915 | ||
3916 | static void | |
3917 | handle_ipfix_upcall(struct dpif_backer *backer, | |
3918 | const struct dpif_upcall *upcall) | |
3919 | { | |
3920 | struct ofproto_dpif *ofproto; | |
3921 | struct flow flow; | |
3922 | ||
8449c4d6 EJ |
3923 | if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
3924 | &flow, NULL, &ofproto, NULL) | |
29089a54 RL |
3925 | || !ofproto->ipfix) { |
3926 | return; | |
3927 | } | |
3928 | ||
3929 | /* The flow reflects exactly the contents of the packet. Sample | |
3930 | * the packet using it. */ | |
3931 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
3932 | } | |
3933 | ||
9b16c439 | 3934 | static int |
acf60855 | 3935 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 3936 | { |
9b16c439 | 3937 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3938 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3939 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3940 | int n_processed; | |
9b16c439 BP |
3941 | int n_misses; |
3942 | int i; | |
abe529af | 3943 | |
cb22974d | 3944 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3945 | |
9b16c439 | 3946 | n_misses = 0; |
90a7c55e | 3947 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3948 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3949 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3950 | int error; |
3951 | ||
90a7c55e BP |
3952 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3953 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 3954 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 3955 | if (error) { |
90a7c55e | 3956 | ofpbuf_uninit(buf); |
9b16c439 BP |
3957 | break; |
3958 | } | |
3959 | ||
6a7e895f BP |
3960 | switch (classify_upcall(upcall)) { |
3961 | case MISS_UPCALL: | |
9b16c439 BP |
3962 | /* Handle it later. */ |
3963 | n_misses++; | |
3964 | break; | |
3965 | ||
6a7e895f | 3966 | case SFLOW_UPCALL: |
acf60855 | 3967 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
3968 | ofpbuf_uninit(buf); |
3969 | break; | |
3970 | ||
29089a54 RL |
3971 | case FLOW_SAMPLE_UPCALL: |
3972 | handle_flow_sample_upcall(backer, upcall); | |
3973 | ofpbuf_uninit(buf); | |
3974 | break; | |
3975 | ||
3976 | case IPFIX_UPCALL: | |
3977 | handle_ipfix_upcall(backer, upcall); | |
3978 | ofpbuf_uninit(buf); | |
3979 | break; | |
3980 | ||
6a7e895f BP |
3981 | case BAD_UPCALL: |
3982 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3983 | break; |
3984 | } | |
abe529af | 3985 | } |
9b16c439 | 3986 | |
6a7e895f | 3987 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 3988 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
3989 | for (i = 0; i < n_misses; i++) { |
3990 | ofpbuf_uninit(&miss_bufs[i]); | |
3991 | } | |
9b16c439 | 3992 | |
90a7c55e | 3993 | return n_processed; |
abe529af BP |
3994 | } |
3995 | \f | |
3996 | /* Flow expiration. */ | |
3997 | ||
04d08d54 | 3998 | static int subfacet_max_idle(const struct dpif_backer *); |
acf60855 | 3999 | static void update_stats(struct dpif_backer *); |
abe529af | 4000 | static void rule_expire(struct rule_dpif *); |
04d08d54 | 4001 | static void expire_subfacets(struct dpif_backer *, int dp_max_idle); |
abe529af BP |
4002 | |
4003 | /* This function is called periodically by run(). Its job is to collect | |
4004 | * updates for the flows that have been installed into the datapath, most | |
4005 | * importantly when they last were used, and then use that information to | |
4006 | * expire flows that have not been used recently. | |
4007 | * | |
4008 | * Returns the number of milliseconds after which it should be called again. */ | |
4009 | static int | |
acf60855 | 4010 | expire(struct dpif_backer *backer) |
abe529af | 4011 | { |
acf60855 | 4012 | struct ofproto_dpif *ofproto; |
dc54ef36 | 4013 | size_t n_subfacets; |
04d08d54 | 4014 | int max_idle; |
abe529af | 4015 | |
8f73d537 EJ |
4016 | /* Periodically clear out the drop keys in an effort to keep them |
4017 | * relatively few. */ | |
4018 | drop_key_clear(backer); | |
4019 | ||
acf60855 JP |
4020 | /* Update stats for each flow in the backer. */ |
4021 | update_stats(backer); | |
abe529af | 4022 | |
04d08d54 EJ |
4023 | n_subfacets = hmap_count(&backer->subfacets); |
4024 | if (n_subfacets) { | |
9fc0165a | 4025 | struct subfacet *subfacet; |
04d08d54 | 4026 | long long int total, now; |
abe529af | 4027 | |
04d08d54 EJ |
4028 | total = 0; |
4029 | now = time_msec(); | |
4030 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { | |
4031 | total += now - subfacet->created; | |
acf60855 | 4032 | } |
04d08d54 EJ |
4033 | backer->avg_subfacet_life += total / n_subfacets; |
4034 | } | |
4035 | backer->avg_subfacet_life /= 2; | |
0697b5c3 | 4036 | |
04d08d54 EJ |
4037 | backer->avg_n_subfacet += n_subfacets; |
4038 | backer->avg_n_subfacet /= 2; | |
4039 | ||
4040 | backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); | |
655ab909 | 4041 | |
04d08d54 EJ |
4042 | max_idle = subfacet_max_idle(backer); |
4043 | expire_subfacets(backer, max_idle); | |
acf60855 | 4044 | |
04d08d54 EJ |
4045 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
4046 | struct rule *rule, *next_rule; | |
4047 | ||
4048 | if (ofproto->backer != backer) { | |
4049 | continue; | |
4050 | } | |
acf60855 JP |
4051 | |
4052 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4053 | * has passed. */ | |
83b16b98 | 4054 | ovs_mutex_lock(&ofproto->up.expirable_mutex); |
e503cc19 SH |
4055 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4056 | &ofproto->up.expirable) { | |
4057 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4058 | } |
83b16b98 | 4059 | ovs_mutex_unlock(&ofproto->up.expirable_mutex); |
abe529af | 4060 | |
acf60855 JP |
4061 | /* All outstanding data in existing flows has been accounted, so it's a |
4062 | * good time to do bond rebalancing. */ | |
4063 | if (ofproto->has_bonded_bundles) { | |
4064 | struct ofbundle *bundle; | |
abe529af | 4065 | |
acf60855 JP |
4066 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4067 | if (bundle->bond) { | |
4a1b8f30 | 4068 | bond_rebalance(bundle->bond); |
acf60855 | 4069 | } |
abe529af BP |
4070 | } |
4071 | } | |
4072 | } | |
4073 | ||
acf60855 | 4074 | return MIN(max_idle, 1000); |
abe529af BP |
4075 | } |
4076 | ||
a218c879 BP |
4077 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4078 | * as 'subfacet''s statistics. */ | |
4079 | static void | |
4080 | update_subfacet_stats(struct subfacet *subfacet, | |
4081 | const struct dpif_flow_stats *stats) | |
4082 | { | |
4083 | struct facet *facet = subfacet->facet; | |
9dfb1f78 EJ |
4084 | struct dpif_flow_stats diff; |
4085 | ||
4086 | diff.tcp_flags = stats->tcp_flags; | |
4087 | diff.used = stats->used; | |
a218c879 BP |
4088 | |
4089 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
9dfb1f78 | 4090 | diff.n_packets = stats->n_packets - subfacet->dp_packet_count; |
a218c879 BP |
4091 | } else { |
4092 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
9dfb1f78 | 4093 | diff.n_packets = 0; |
a218c879 BP |
4094 | } |
4095 | ||
4096 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
9dfb1f78 | 4097 | diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; |
a218c879 BP |
4098 | } else { |
4099 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
9dfb1f78 | 4100 | diff.n_bytes = 0; |
a218c879 BP |
4101 | } |
4102 | ||
b7e2f6b3 | 4103 | facet->ofproto->n_hit += diff.n_packets; |
a218c879 BP |
4104 | subfacet->dp_packet_count = stats->n_packets; |
4105 | subfacet->dp_byte_count = stats->n_bytes; | |
9dfb1f78 | 4106 | subfacet_update_stats(subfacet, &diff); |
a218c879 | 4107 | |
a218c879 BP |
4108 | if (facet->accounted_bytes < facet->byte_count) { |
4109 | facet_learn(facet); | |
4110 | facet_account(facet); | |
4111 | facet->accounted_bytes = facet->byte_count; | |
4112 | } | |
a218c879 BP |
4113 | } |
4114 | ||
4115 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4116 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4117 | static void | |
04d08d54 | 4118 | delete_unexpected_flow(struct dpif_backer *backer, |
a218c879 BP |
4119 | const struct nlattr *key, size_t key_len) |
4120 | { | |
4121 | if (!VLOG_DROP_WARN(&rl)) { | |
4122 | struct ds s; | |
4123 | ||
4124 | ds_init(&s); | |
4125 | odp_flow_key_format(key, key_len, &s); | |
04d08d54 | 4126 | VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); |
a218c879 BP |
4127 | ds_destroy(&s); |
4128 | } | |
4129 | ||
4130 | COVERAGE_INC(facet_unexpected); | |
04d08d54 | 4131 | dpif_flow_del(backer->dpif, key, key_len, NULL); |
a218c879 BP |
4132 | } |
4133 | ||
abe529af BP |
4134 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4135 | * | |
4136 | * This function also pushes statistics updates to rules which each facet | |
4137 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4138 | * facet changes the rule it resubmits into at some time in between | |
4139 | * update_stats() runs, it is possible that statistics accrued to the | |
4140 | * old rule will be incorrectly attributed to the new rule. This could be | |
4141 | * avoided by calling update_stats() whenever rules are created or | |
4142 | * deleted. However, the performance impact of making so many calls to the | |
4143 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4144 | * |
4145 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4146 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4147 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4148 | * the hit or miss counts of br1. | |
abe529af BP |
4149 | */ |
4150 | static void | |
acf60855 | 4151 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4152 | { |
4153 | const struct dpif_flow_stats *stats; | |
4154 | struct dpif_flow_dump dump; | |
61fb711d JP |
4155 | const struct nlattr *key, *mask; |
4156 | size_t key_len, mask_len; | |
abe529af | 4157 | |
acf60855 | 4158 | dpif_flow_dump_start(&dump, backer->dpif); |
e6cc0bab | 4159 | while (dpif_flow_dump_next(&dump, &key, &key_len, |
61fb711d | 4160 | &mask, &mask_len, NULL, NULL, &stats)) { |
b0f7b9b5 | 4161 | struct subfacet *subfacet; |
acf60855 | 4162 | uint32_t key_hash; |
abe529af | 4163 | |
acf60855 | 4164 | key_hash = odp_flow_key_hash(key, key_len); |
04d08d54 | 4165 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
6a7e895f BP |
4166 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4167 | case SF_FAST_PATH: | |
a218c879 | 4168 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4169 | break; |
4170 | ||
4171 | case SF_SLOW_PATH: | |
4172 | /* Stats are updated per-packet. */ | |
4173 | break; | |
4174 | ||
4175 | case SF_NOT_INSTALLED: | |
4176 | default: | |
04d08d54 | 4177 | delete_unexpected_flow(backer, key, key_len); |
6a7e895f | 4178 | break; |
abe529af | 4179 | } |
8fa4d1d0 | 4180 | run_fast_rl(); |
abe529af BP |
4181 | } |
4182 | dpif_flow_dump_done(&dump); | |
3af56aef | 4183 | |
dc54ef36 | 4184 | update_moving_averages(backer); |
abe529af BP |
4185 | } |
4186 | ||
4187 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4188 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4189 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4190 | * fold its statistic into its rule. */ | |
abe529af | 4191 | static int |
04d08d54 | 4192 | subfacet_max_idle(const struct dpif_backer *backer) |
abe529af BP |
4193 | { |
4194 | /* | |
4195 | * Idle time histogram. | |
4196 | * | |
b0f7b9b5 BP |
4197 | * Most of the time a switch has a relatively small number of subfacets. |
4198 | * When this is the case we might as well keep statistics for all of them | |
4199 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4200 | * well. |
4201 | * | |
b0f7b9b5 | 4202 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4203 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4204 | * significant. However, with a large number of subfacets it is likely |
4205 | * that only a few of them are "heavy hitters" that consume a large amount | |
4206 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4207 | * the kernel and maintaining in userspaces; other subfacets we can | |
4208 | * discard. | |
abe529af BP |
4209 | * |
4210 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4211 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4212 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4213 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4214 | * the most-recently-used 1% of subfacets (but at least |
380f49c4 | 4215 | * flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4216 | * the most-recently-used bucket of subfacets is kept, so actually an |
4217 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4218 | * (though the next run will delete most of those unless they receive |
4219 | * additional data). | |
abe529af | 4220 | * |
b0f7b9b5 BP |
4221 | * This requires a second pass through the subfacets, in addition to the |
4222 | * pass made by update_stats(), because the former function never looks at | |
4223 | * uninstallable subfacets. | |
abe529af BP |
4224 | */ |
4225 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4226 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4227 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4228 | int total, subtotal, bucket; |
b0f7b9b5 | 4229 | struct subfacet *subfacet; |
abe529af BP |
4230 | long long int now; |
4231 | int i; | |
4232 | ||
04d08d54 | 4233 | total = hmap_count(&backer->subfacets); |
380f49c4 | 4234 | if (total <= flow_eviction_threshold) { |
abe529af BP |
4235 | return N_BUCKETS * BUCKET_WIDTH; |
4236 | } | |
4237 | ||
4238 | /* Build histogram. */ | |
4239 | now = time_msec(); | |
04d08d54 | 4240 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { |
b0f7b9b5 | 4241 | long long int idle = now - subfacet->used; |
abe529af BP |
4242 | int bucket = (idle <= 0 ? 0 |
4243 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4244 | : (unsigned int) idle / BUCKET_WIDTH); | |
4245 | buckets[bucket]++; | |
4246 | } | |
4247 | ||
4248 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4249 | subtotal = bucket = 0; |
4250 | do { | |
4251 | subtotal += buckets[bucket++]; | |
084f5290 | 4252 | } while (bucket < N_BUCKETS && |
380f49c4 | 4253 | subtotal < MAX(flow_eviction_threshold, total / 100)); |
abe529af BP |
4254 | |
4255 | if (VLOG_IS_DBG_ENABLED()) { | |
4256 | struct ds s; | |
4257 | ||
4258 | ds_init(&s); | |
4259 | ds_put_cstr(&s, "keep"); | |
4260 | for (i = 0; i < N_BUCKETS; i++) { | |
4261 | if (i == bucket) { | |
4262 | ds_put_cstr(&s, ", drop"); | |
4263 | } | |
4264 | if (buckets[i]) { | |
4265 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4266 | } | |
4267 | } | |
04d08d54 | 4268 | VLOG_INFO("%s (msec:count)", ds_cstr(&s)); |
abe529af BP |
4269 | ds_destroy(&s); |
4270 | } | |
4271 | ||
4272 | return bucket * BUCKET_WIDTH; | |
4273 | } | |
4274 | ||
abe529af | 4275 | static void |
04d08d54 | 4276 | expire_subfacets(struct dpif_backer *backer, int dp_max_idle) |
abe529af | 4277 | { |
625b0720 BP |
4278 | /* Cutoff time for most flows. */ |
4279 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4280 | ||
4281 | /* We really want to keep flows for special protocols around, so use a more | |
4282 | * conservative cutoff. */ | |
4283 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4284 | |
b0f7b9b5 | 4285 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4286 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4287 | int n_batch; |
abe529af | 4288 | |
b99d3cee | 4289 | n_batch = 0; |
b0f7b9b5 | 4290 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
04d08d54 | 4291 | &backer->subfacets) { |
625b0720 BP |
4292 | long long int cutoff; |
4293 | ||
bbafd73b EJ |
4294 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
4295 | | SLOW_STP) | |
625b0720 BP |
4296 | ? special_cutoff |
4297 | : normal_cutoff); | |
b0f7b9b5 | 4298 | if (subfacet->used < cutoff) { |
6a7e895f | 4299 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4300 | batch[n_batch++] = subfacet; |
1d85f9e5 | 4301 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
04d08d54 | 4302 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee BP |
4303 | n_batch = 0; |
4304 | } | |
4305 | } else { | |
4306 | subfacet_destroy(subfacet); | |
4307 | } | |
abe529af BP |
4308 | } |
4309 | } | |
b99d3cee BP |
4310 | |
4311 | if (n_batch > 0) { | |
04d08d54 | 4312 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee | 4313 | } |
abe529af BP |
4314 | } |
4315 | ||
4316 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4317 | * then delete it entirely. */ | |
4318 | static void | |
4319 | rule_expire(struct rule_dpif *rule) | |
4320 | { | |
a3779dbc | 4321 | uint16_t idle_timeout, hard_timeout; |
abe529af BP |
4322 | long long int now; |
4323 | uint8_t reason; | |
4324 | ||
e2a3d183 BP |
4325 | if (rule->up.pending) { |
4326 | /* We'll have to expire it later. */ | |
4327 | return; | |
4328 | } | |
4329 | ||
a3779dbc EJ |
4330 | ovs_mutex_lock(&rule->up.timeout_mutex); |
4331 | hard_timeout = rule->up.hard_timeout; | |
4332 | idle_timeout = rule->up.idle_timeout; | |
4333 | ovs_mutex_unlock(&rule->up.timeout_mutex); | |
4334 | ||
abe529af BP |
4335 | /* Has 'rule' expired? */ |
4336 | now = time_msec(); | |
a3779dbc | 4337 | if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) { |
abe529af | 4338 | reason = OFPRR_HARD_TIMEOUT; |
a3779dbc | 4339 | } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) { |
abe529af BP |
4340 | reason = OFPRR_IDLE_TIMEOUT; |
4341 | } else { | |
4342 | return; | |
4343 | } | |
4344 | ||
4345 | COVERAGE_INC(ofproto_dpif_expired); | |
4346 | ||
abe529af BP |
4347 | /* Get rid of the rule. */ |
4348 | ofproto_rule_expire(&rule->up, reason); | |
4349 | } | |
4350 | \f | |
4351 | /* Facets. */ | |
4352 | ||
4dff9097 | 4353 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
4354 | * |
4355 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 4356 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 4357 | * |
bcd2633a JP |
4358 | * 'rule' and 'xout' must have been created based on 'miss'. |
4359 | * | |
4360 | * 'facet'' statistics are initialized based on 'stats'. | |
2b459b83 | 4361 | * |
b0f7b9b5 BP |
4362 | * The facet will initially have no subfacets. The caller should create (at |
4363 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4364 | static struct facet * |
bcd2633a JP |
4365 | facet_create(const struct flow_miss *miss, struct rule_dpif *rule, |
4366 | struct xlate_out *xout, struct dpif_flow_stats *stats) | |
abe529af | 4367 | { |
4dff9097 | 4368 | struct ofproto_dpif *ofproto = miss->ofproto; |
abe529af | 4369 | struct facet *facet; |
bcd2633a | 4370 | struct match match; |
abe529af BP |
4371 | |
4372 | facet = xzalloc(sizeof *facet); | |
b7e2f6b3 | 4373 | facet->ofproto = miss->ofproto; |
bcd2633a JP |
4374 | facet->packet_count = facet->prev_packet_count = stats->n_packets; |
4375 | facet->byte_count = facet->prev_byte_count = stats->n_bytes; | |
4376 | facet->tcp_flags = stats->tcp_flags; | |
4377 | facet->used = stats->used; | |
4dff9097 | 4378 | facet->flow = miss->flow; |
4dff9097 EJ |
4379 | facet->learn_rl = time_msec() + 500; |
4380 | ||
b0f7b9b5 | 4381 | list_init(&facet->subfacets); |
abe529af BP |
4382 | netflow_flow_init(&facet->nf_flow); |
4383 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4384 | ||
bcd2633a JP |
4385 | xlate_out_copy(&facet->xout, xout); |
4386 | ||
4387 | match_init(&match, &facet->flow, &facet->xout.wc); | |
4388 | cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); | |
0b4f2078 | 4389 | ovs_rwlock_wrlock(&ofproto->facets.rwlock); |
bcd2633a | 4390 | classifier_insert(&ofproto->facets, &facet->cr); |
0b4f2078 | 4391 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 4392 | |
bbafd73b | 4393 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; |
b7e2f6b3 | 4394 | facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; |
6cf474d7 | 4395 | |
abe529af BP |
4396 | return facet; |
4397 | } | |
4398 | ||
4399 | static void | |
4400 | facet_free(struct facet *facet) | |
4401 | { | |
4dff9097 | 4402 | if (facet) { |
bbafd73b | 4403 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
4404 | free(facet); |
4405 | } | |
abe529af BP |
4406 | } |
4407 | ||
3d9e05f8 | 4408 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4409 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4410 | static bool |
4411 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4412 | const struct nlattr *odp_actions, size_t actions_len, | |
4413 | struct ofpbuf *packet) | |
4414 | { | |
4415 | struct odputil_keybuf keybuf; | |
4416 | struct ofpbuf key; | |
4417 | int error; | |
4418 | ||
6ff686f2 | 4419 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a | 4420 | odp_flow_key_from_flow(&key, flow, |
4e022ec0 | 4421 | ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port)); |
80e5eed9 | 4422 | |
acf60855 | 4423 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4424 | odp_actions, actions_len, packet); |
6ff686f2 | 4425 | return !error; |
abe529af BP |
4426 | } |
4427 | ||
bcd2633a | 4428 | /* Remove 'facet' from its ofproto and free up the associated memory: |
abe529af BP |
4429 | * |
4430 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4431 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4432 | * |
4433 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4434 | */ | |
4435 | static void | |
15baa734 | 4436 | facet_remove(struct facet *facet) |
abe529af | 4437 | { |
b0f7b9b5 BP |
4438 | struct subfacet *subfacet, *next_subfacet; |
4439 | ||
cb22974d | 4440 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4441 | |
4442 | /* First uninstall all of the subfacets to get final statistics. */ | |
4443 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4444 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4445 | } |
4446 | ||
4447 | /* Flush the final stats to the rule. | |
4448 | * | |
4449 | * This might require us to have at least one subfacet around so that we | |
4450 | * can use its actions for accounting in facet_account(), which is why we | |
4451 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4452 | facet_flush_stats(facet); |
551a2f6c BP |
4453 | |
4454 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4455 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4456 | &facet->subfacets) { | |
15baa734 | 4457 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4458 | } |
0b4f2078 | 4459 | ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock); |
b7e2f6b3 | 4460 | classifier_remove(&facet->ofproto->facets, &facet->cr); |
0b4f2078 | 4461 | ovs_rwlock_unlock(&facet->ofproto->facets.rwlock); |
bcd2633a | 4462 | cls_rule_destroy(&facet->cr); |
abe529af BP |
4463 | facet_free(facet); |
4464 | } | |
4465 | ||
3de9590b BP |
4466 | /* Feed information from 'facet' back into the learning table to keep it in |
4467 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4468 | static void |
3de9590b | 4469 | facet_learn(struct facet *facet) |
abe529af | 4470 | { |
62bd7349 | 4471 | long long int now = time_msec(); |
abe529af | 4472 | |
bbafd73b | 4473 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
4474 | return; |
4475 | } | |
4476 | ||
62bd7349 | 4477 | facet->learn_rl = now + 500; |
6cf474d7 | 4478 | |
bbafd73b EJ |
4479 | if (!facet->xout.has_learn |
4480 | && !facet->xout.has_normal | |
4481 | && (!facet->xout.has_fin_timeout | |
3de9590b | 4482 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
4483 | return; |
4484 | } | |
abe529af | 4485 | |
9dfb1f78 | 4486 | facet_push_stats(facet, true); |
3de9590b BP |
4487 | } |
4488 | ||
4489 | static void | |
4490 | facet_account(struct facet *facet) | |
4491 | { | |
3de9590b BP |
4492 | const struct nlattr *a; |
4493 | unsigned int left; | |
4494 | ovs_be16 vlan_tci; | |
4495 | uint64_t n_bytes; | |
abe529af | 4496 | |
b7e2f6b3 | 4497 | if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { |
abe529af BP |
4498 | return; |
4499 | } | |
3de9590b | 4500 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4501 | |
4502 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4503 | * as a basis. We also need to track the actual VLAN on which the packet | |
4504 | * is going to be sent to ensure that it matches the one passed to | |
4505 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4506 | * hash bucket.) |
4507 | * | |
4508 | * We use the actions from an arbitrary subfacet because they should all | |
4509 | * be equally valid for our purpose. */ | |
d78be13b | 4510 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
4511 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
4512 | facet->xout.odp_actions.size) { | |
fea393b1 | 4513 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4514 | struct ofport_dpif *port; |
abe529af | 4515 | |
d78be13b | 4516 | switch (nl_attr_type(a)) { |
df2c07f4 | 4517 | case OVS_ACTION_ATTR_OUTPUT: |
b7e2f6b3 | 4518 | port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); |
abe529af | 4519 | if (port && port->bundle && port->bundle->bond) { |
d78be13b | 4520 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4521 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4522 | } |
d78be13b BP |
4523 | break; |
4524 | ||
fea393b1 BP |
4525 | case OVS_ACTION_ATTR_POP_VLAN: |
4526 | vlan_tci = htons(0); | |
d78be13b BP |
4527 | break; |
4528 | ||
fea393b1 BP |
4529 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4530 | vlan = nl_attr_get(a); | |
4531 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4532 | break; |
abe529af BP |
4533 | } |
4534 | } | |
4535 | } | |
4536 | ||
abe529af BP |
4537 | /* Returns true if the only action for 'facet' is to send to the controller. |
4538 | * (We don't report NetFlow expiration messages for such facets because they | |
4539 | * are just part of the control logic for the network, not real traffic). */ | |
4540 | static bool | |
4541 | facet_is_controller_flow(struct facet *facet) | |
4542 | { | |
f25d0cf3 | 4543 | if (facet) { |
b7e2f6b3 EJ |
4544 | struct ofproto_dpif *ofproto = facet->ofproto; |
4545 | const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow, | |
4546 | NULL); | |
4547 | const struct ofpact *ofpacts = rule->up.ofpacts; | |
4548 | size_t ofpacts_len = rule->up.ofpacts_len; | |
f25d0cf3 | 4549 | |
dd30ff28 BP |
4550 | if (ofpacts_len > 0 && |
4551 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4552 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4553 | return true; | |
4554 | } | |
4555 | } | |
4556 | return false; | |
abe529af BP |
4557 | } |
4558 | ||
4559 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4560 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4561 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4562 | * its packet and byte counts before this function is called. */ | |
4563 | static void | |
15baa734 | 4564 | facet_flush_stats(struct facet *facet) |
abe529af | 4565 | { |
b7e2f6b3 | 4566 | struct ofproto_dpif *ofproto = facet->ofproto; |
b0f7b9b5 BP |
4567 | struct subfacet *subfacet; |
4568 | ||
4569 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4570 | ovs_assert(!subfacet->dp_byte_count); |
4571 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4572 | } |
abe529af | 4573 | |
9dfb1f78 | 4574 | facet_push_stats(facet, false); |
3de9590b BP |
4575 | if (facet->accounted_bytes < facet->byte_count) { |
4576 | facet_account(facet); | |
4577 | facet->accounted_bytes = facet->byte_count; | |
4578 | } | |
abe529af BP |
4579 | |
4580 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4581 | struct ofexpired expired; | |
4582 | expired.flow = facet->flow; | |
4583 | expired.packet_count = facet->packet_count; | |
4584 | expired.byte_count = facet->byte_count; | |
4585 | expired.used = facet->used; | |
4586 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4587 | } | |
4588 | ||
abe529af BP |
4589 | /* Reset counters to prevent double counting if 'facet' ever gets |
4590 | * reinstalled. */ | |
bbb5d219 | 4591 | facet_reset_counters(facet); |
abe529af BP |
4592 | |
4593 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4594 | facet->tcp_flags = 0; |
abe529af BP |
4595 | } |
4596 | ||
bcd2633a JP |
4597 | /* Searches 'ofproto''s table of facets for one which would be responsible for |
4598 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4599 | * |
abe529af BP |
4600 | * The returned facet might need revalidation; use facet_lookup_valid() |
4601 | * instead if that is important. */ | |
4602 | static struct facet * | |
bcd2633a | 4603 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4604 | { |
0b4f2078 EJ |
4605 | struct cls_rule *cr; |
4606 | ||
4607 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
4608 | cr = classifier_lookup(&ofproto->facets, flow, NULL); | |
4609 | ovs_rwlock_unlock(&ofproto->facets.rwlock); | |
bcd2633a | 4610 | return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL; |
abe529af BP |
4611 | } |
4612 | ||
bcd2633a JP |
4613 | /* Searches 'ofproto''s table of facets for one capable that covers |
4614 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4615 | * |
abe529af BP |
4616 | * The returned facet is guaranteed to be valid. */ |
4617 | static struct facet * | |
bcd2633a | 4618 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4619 | { |
c57b2226 | 4620 | struct facet *facet; |
abe529af | 4621 | |
bcd2633a | 4622 | facet = facet_find(ofproto, flow); |
abe529af | 4623 | if (facet |
a1616063 | 4624 | && ofproto->backer->need_revalidate |
5bf64ade | 4625 | && !facet_revalidate(facet)) { |
0305ce1f | 4626 | return NULL; |
abe529af BP |
4627 | } |
4628 | ||
4629 | return facet; | |
4630 | } | |
4631 | ||
6814e51f BP |
4632 | static bool |
4633 | facet_check_consistency(struct facet *facet) | |
4634 | { | |
4635 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4636 | ||
bbafd73b EJ |
4637 | struct xlate_out xout; |
4638 | struct xlate_in xin; | |
050ac423 | 4639 | |
6814e51f | 4640 | struct rule_dpif *rule; |
b7e2f6b3 | 4641 | bool ok, fail_open; |
6814e51f BP |
4642 | |
4643 | /* Check the datapath actions for consistency. */ | |
b7e2f6b3 EJ |
4644 | rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); |
4645 | xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL); | |
bbafd73b | 4646 | xlate_actions(&xin, &xout); |
6814e51f | 4647 | |
b7e2f6b3 | 4648 | fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; |
bbafd73b | 4649 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
b7e2f6b3 EJ |
4650 | && facet->xout.slow == xout.slow |
4651 | && facet->fail_open == fail_open; | |
4dff9097 EJ |
4652 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4653 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4654 | |
4dff9097 EJ |
4655 | flow_format(&s, &facet->flow); |
4656 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4657 | |
bbafd73b | 4658 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4659 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4660 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4661 | facet->xout.odp_actions.size); | |
9616614b | 4662 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4663 | format_odp_actions(&s, xout.odp_actions.data, |
4664 | xout.odp_actions.size); | |
c0a71f4e | 4665 | ds_put_char(&s, ')'); |
6814e51f | 4666 | } |
4dff9097 | 4667 | |
bbafd73b EJ |
4668 | if (facet->xout.slow != xout.slow) { |
4669 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4670 | } |
4671 | ||
b7e2f6b3 EJ |
4672 | if (facet->fail_open != fail_open) { |
4673 | ds_put_format(&s, " fail open incorrect. should be %s", | |
4674 | fail_open ? "true" : "false"); | |
4675 | } | |
9616614b | 4676 | ds_destroy(&s); |
6814e51f | 4677 | } |
bbafd73b | 4678 | xlate_out_uninit(&xout); |
6814e51f BP |
4679 | |
4680 | return ok; | |
4681 | } | |
4682 | ||
15baa734 | 4683 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4684 | * |
4685 | * - If the rule found is different from 'facet''s current rule, moves | |
4686 | * 'facet' to the new rule and recompiles its actions. | |
4687 | * | |
4688 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4689 | * where it is and recompiles its actions anyway. |
4690 | * | |
4691 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
8449c4d6 | 4692 | * xlate_receive(), 'facet' is removed. |
5bf64ade EJ |
4693 | * |
4694 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
4695 | static bool | |
15baa734 | 4696 | facet_revalidate(struct facet *facet) |
abe529af | 4697 | { |
b7e2f6b3 | 4698 | struct ofproto_dpif *ofproto = facet->ofproto; |
abe529af | 4699 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4700 | struct subfacet *subfacet; |
bcd2633a | 4701 | struct flow_wildcards wc; |
bbafd73b EJ |
4702 | struct xlate_out xout; |
4703 | struct xlate_in xin; | |
abe529af BP |
4704 | |
4705 | COVERAGE_INC(facet_revalidate); | |
4706 | ||
f231418e EJ |
4707 | /* Check that child subfacets still correspond to this facet. Tunnel |
4708 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4709 | * change. */ | |
4710 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4711 | struct ofproto_dpif *recv_ofproto; | |
4712 | struct flow recv_flow; | |
4713 | int error; | |
4714 | ||
8449c4d6 EJ |
4715 | error = xlate_receive(ofproto->backer, NULL, subfacet->key, |
4716 | subfacet->key_len, &recv_flow, NULL, | |
4717 | &recv_ofproto, NULL); | |
f231418e EJ |
4718 | if (error |
4719 | || recv_ofproto != ofproto | |
bcd2633a | 4720 | || facet != facet_find(ofproto, &recv_flow)) { |
f231418e | 4721 | facet_remove(facet); |
5bf64ade | 4722 | return false; |
f231418e EJ |
4723 | } |
4724 | } | |
4725 | ||
bcd2633a JP |
4726 | flow_wildcards_init_catchall(&wc); |
4727 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc); | |
abe529af | 4728 | |
df2c07f4 | 4729 | /* Calculate new datapath actions. |
abe529af BP |
4730 | * |
4731 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4732 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4733 | * around to properly compose it. */ | |
47b0deae | 4734 | xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); |
bbafd73b | 4735 | xlate_actions(&xin, &xout); |
bcd2633a | 4736 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); |
4dff9097 EJ |
4737 | |
4738 | /* A facet's slow path reason should only change under dramatic | |
4739 | * circumstances. Rather than try to update everything, it's simpler to | |
bcd2633a JP |
4740 | * remove the facet and start over. |
4741 | * | |
4742 | * More importantly, if a facet's wildcards change, it will be relatively | |
4743 | * difficult to figure out if its subfacets still belong to it, and if not | |
4744 | * which facet they may belong to. Again, to avoid the complexity, we | |
4745 | * simply give up instead. */ | |
4746 | if (facet->xout.slow != xout.slow | |
4747 | || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { | |
4dff9097 | 4748 | facet_remove(facet); |
bbafd73b | 4749 | xlate_out_uninit(&xout); |
4dff9097 EJ |
4750 | return false; |
4751 | } | |
6a7e895f | 4752 | |
bbafd73b | 4753 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
4754 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
4755 | if (subfacet->path == SF_FAST_PATH) { | |
4756 | struct dpif_flow_stats stats; | |
b95fc6ba | 4757 | |
bbafd73b | 4758 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 4759 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 4760 | } |
abe529af | 4761 | } |
b95fc6ba | 4762 | |
15baa734 | 4763 | facet_flush_stats(facet); |
4dff9097 | 4764 | |
bbafd73b EJ |
4765 | ofpbuf_clear(&facet->xout.odp_actions); |
4766 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
4767 | xout.odp_actions.size); | |
abe529af BP |
4768 | } |
4769 | ||
4770 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
4771 | facet->xout.slow = xout.slow; |
4772 | facet->xout.has_learn = xout.has_learn; | |
4773 | facet->xout.has_normal = xout.has_normal; | |
4774 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
4775 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
4776 | facet->xout.mirrors = xout.mirrors; | |
4777 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
b7e2f6b3 EJ |
4778 | facet->used = MAX(facet->used, new_rule->up.created); |
4779 | facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY; | |
5bf64ade | 4780 | |
bbafd73b | 4781 | xlate_out_uninit(&xout); |
5bf64ade | 4782 | return true; |
abe529af BP |
4783 | } |
4784 | ||
bbb5d219 EJ |
4785 | static void |
4786 | facet_reset_counters(struct facet *facet) | |
4787 | { | |
4788 | facet->packet_count = 0; | |
4789 | facet->byte_count = 0; | |
9d24de3b JP |
4790 | facet->prev_packet_count = 0; |
4791 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4792 | facet->accounted_bytes = 0; |
4793 | } | |
4794 | ||
abe529af | 4795 | static void |
9dfb1f78 | 4796 | facet_push_stats(struct facet *facet, bool may_learn) |
abe529af | 4797 | { |
112bc5f4 | 4798 | struct dpif_flow_stats stats; |
abe529af | 4799 | |
cb22974d BP |
4800 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4801 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4802 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4803 | |
112bc5f4 BP |
4804 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4805 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4806 | stats.used = facet->used; | |
9dfb1f78 EJ |
4807 | stats.tcp_flags = facet->tcp_flags; |
4808 | ||
4809 | if (may_learn || stats.n_packets || facet->used > facet->prev_used) { | |
b7e2f6b3 | 4810 | struct ofproto_dpif *ofproto = facet->ofproto; |
9dfb1f78 | 4811 | struct ofport_dpif *in_port; |
b7e2f6b3 | 4812 | struct rule_dpif *rule; |
9dfb1f78 | 4813 | struct xlate_in xin; |
abe529af | 4814 | |
9d24de3b JP |
4815 | facet->prev_packet_count = facet->packet_count; |
4816 | facet->prev_byte_count = facet->byte_count; | |
4817 | facet->prev_used = facet->used; | |
abe529af | 4818 | |
4e022ec0 | 4819 | in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port); |
42943cde | 4820 | if (in_port && in_port->is_tunnel) { |
9dfb1f78 EJ |
4821 | netdev_vport_inc_rx(in_port->up.netdev, &stats); |
4822 | } | |
9d24de3b | 4823 | |
b7e2f6b3 EJ |
4824 | rule = rule_dpif_lookup(ofproto, &facet->flow, NULL); |
4825 | rule_credit_stats(rule, &stats); | |
9dfb1f78 EJ |
4826 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, |
4827 | facet->used); | |
4828 | netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); | |
ec7ceaed EJ |
4829 | mirror_update_stats(ofproto->mbridge, facet->xout.mirrors, |
4830 | stats.n_packets, stats.n_bytes); | |
9dfb1f78 | 4831 | |
b7e2f6b3 EJ |
4832 | xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags, |
4833 | NULL); | |
9dfb1f78 EJ |
4834 | xin.resubmit_stats = &stats; |
4835 | xin.may_learn = may_learn; | |
4836 | xlate_actions_for_side_effects(&xin); | |
abe529af BP |
4837 | } |
4838 | } | |
4839 | ||
8844e035 | 4840 | static void |
4464589e | 4841 | push_all_stats__(bool run_fast) |
8844e035 EJ |
4842 | { |
4843 | static long long int rl = LLONG_MIN; | |
4844 | struct ofproto_dpif *ofproto; | |
4845 | ||
4846 | if (time_msec() < rl) { | |
4847 | return; | |
4848 | } | |
4849 | ||
4850 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
bcd2633a | 4851 | struct cls_cursor cursor; |
8844e035 EJ |
4852 | struct facet *facet; |
4853 | ||
0b4f2078 | 4854 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
4855 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
4856 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
9dfb1f78 | 4857 | facet_push_stats(facet, false); |
4464589e EJ |
4858 | if (run_fast) { |
4859 | run_fast_rl(); | |
4860 | } | |
8844e035 | 4861 | } |
0b4f2078 | 4862 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
8844e035 EJ |
4863 | } |
4864 | ||
4865 | rl = time_msec() + 100; | |
4866 | } | |
4867 | ||
4464589e EJ |
4868 | static void |
4869 | push_all_stats(void) | |
4870 | { | |
4871 | push_all_stats__(true); | |
4872 | } | |
4873 | ||
9583bc14 | 4874 | void |
112bc5f4 | 4875 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4876 | { |
9fbe253e | 4877 | ovs_mutex_lock(&rule->stats_mutex); |
112bc5f4 BP |
4878 | rule->packet_count += stats->n_packets; |
4879 | rule->byte_count += stats->n_bytes; | |
4880 | ofproto_rule_update_used(&rule->up, stats->used); | |
9fbe253e | 4881 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af | 4882 | } |
abe529af | 4883 | \f |
b0f7b9b5 BP |
4884 | /* Subfacets. */ |
4885 | ||
4886 | static struct subfacet * | |
04d08d54 EJ |
4887 | subfacet_find(struct dpif_backer *backer, const struct nlattr *key, |
4888 | size_t key_len, uint32_t key_hash) | |
b0f7b9b5 BP |
4889 | { |
4890 | struct subfacet *subfacet; | |
4891 | ||
4892 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
04d08d54 | 4893 | &backer->subfacets) { |
9566abf9 EJ |
4894 | if (subfacet->key_len == key_len |
4895 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
4896 | return subfacet; |
4897 | } | |
4898 | } | |
4899 | ||
4900 | return NULL; | |
4901 | } | |
4902 | ||
4903 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
4904 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
4905 | * existing subfacet if there is one, otherwise creates and returns a | |
4dff9097 | 4906 | * new subfacet. */ |
b0f7b9b5 | 4907 | static struct subfacet * |
a088a1ff JP |
4908 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
4909 | long long int now) | |
b0f7b9b5 | 4910 | { |
04d08d54 | 4911 | struct dpif_backer *backer = miss->ofproto->backer; |
a088a1ff JP |
4912 | enum odp_key_fitness key_fitness = miss->key_fitness; |
4913 | const struct nlattr *key = miss->key; | |
4914 | size_t key_len = miss->key_len; | |
4915 | uint32_t key_hash; | |
b0f7b9b5 BP |
4916 | struct subfacet *subfacet; |
4917 | ||
a088a1ff JP |
4918 | key_hash = odp_flow_key_hash(key, key_len); |
4919 | ||
3b145dd7 BP |
4920 | if (list_is_empty(&facet->subfacets)) { |
4921 | subfacet = &facet->one_subfacet; | |
4922 | } else { | |
04d08d54 | 4923 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
3b145dd7 BP |
4924 | if (subfacet) { |
4925 | if (subfacet->facet == facet) { | |
4926 | return subfacet; | |
4927 | } | |
4928 | ||
4929 | /* This shouldn't happen. */ | |
4930 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
4931 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
4932 | } |
4933 | ||
3b145dd7 | 4934 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
4935 | } |
4936 | ||
04d08d54 | 4937 | hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); |
b0f7b9b5 BP |
4938 | list_push_back(&facet->subfacets, &subfacet->list_node); |
4939 | subfacet->facet = facet; | |
b0f7b9b5 | 4940 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
4941 | subfacet->key = xmemdup(key, key_len); |
4942 | subfacet->key_len = key_len; | |
459b16a1 | 4943 | subfacet->used = now; |
655ab909 | 4944 | subfacet->created = now; |
26cd7e34 BP |
4945 | subfacet->dp_packet_count = 0; |
4946 | subfacet->dp_byte_count = 0; | |
6a7e895f | 4947 | subfacet->path = SF_NOT_INSTALLED; |
04d08d54 | 4948 | subfacet->backer = backer; |
b0f7b9b5 | 4949 | |
04d08d54 | 4950 | backer->subfacet_add_count++; |
b0f7b9b5 BP |
4951 | return subfacet; |
4952 | } | |
4953 | ||
b0f7b9b5 BP |
4954 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
4955 | * its facet within 'ofproto', and frees it. */ | |
4956 | static void | |
15baa734 | 4957 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4958 | { |
15baa734 | 4959 | struct facet *facet = subfacet->facet; |
b7e2f6b3 | 4960 | struct ofproto_dpif *ofproto = facet->ofproto; |
15baa734 | 4961 | |
655ab909 | 4962 | /* Update ofproto stats before uninstall the subfacet. */ |
dc54ef36 | 4963 | ofproto->backer->subfacet_del_count++; |
655ab909 | 4964 | |
15baa734 | 4965 | subfacet_uninstall(subfacet); |
04d08d54 | 4966 | hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); |
b0f7b9b5 BP |
4967 | list_remove(&subfacet->list_node); |
4968 | free(subfacet->key); | |
26cd7e34 BP |
4969 | if (subfacet != &facet->one_subfacet) { |
4970 | free(subfacet); | |
4971 | } | |
b0f7b9b5 BP |
4972 | } |
4973 | ||
4974 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4975 | * last remaining subfacet in its facet destroys the facet too. */ | |
4976 | static void | |
15baa734 | 4977 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4978 | { |
4979 | struct facet *facet = subfacet->facet; | |
4980 | ||
551a2f6c BP |
4981 | if (list_is_singleton(&facet->subfacets)) { |
4982 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4983 | facet_remove(facet); |
551a2f6c | 4984 | } else { |
15baa734 | 4985 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4986 | } |
4987 | } | |
4988 | ||
1d85f9e5 | 4989 | static void |
04d08d54 | 4990 | subfacet_destroy_batch(struct dpif_backer *backer, |
1d85f9e5 JP |
4991 | struct subfacet **subfacets, int n) |
4992 | { | |
1d85f9e5 JP |
4993 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
4994 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
4995 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
4996 | int i; | |
4997 | ||
4998 | for (i = 0; i < n; i++) { | |
4999 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
5000 | ops[i].u.flow_del.key = subfacets[i]->key; |
5001 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
5002 | ops[i].u.flow_del.stats = &stats[i]; |
5003 | opsp[i] = &ops[i]; | |
5004 | } | |
5005 | ||
04d08d54 | 5006 | dpif_operate(backer->dpif, opsp, n); |
1d85f9e5 JP |
5007 | for (i = 0; i < n; i++) { |
5008 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5009 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5010 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 5011 | run_fast_rl(); |
1d85f9e5 JP |
5012 | } |
5013 | } | |
5014 | ||
b0f7b9b5 BP |
5015 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5016 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5017 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5018 | * since 'subfacet' was last updated. | |
5019 | * | |
5020 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5021 | static int | |
4dff9097 EJ |
5022 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
5023 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 5024 | { |
15baa734 | 5025 | struct facet *facet = subfacet->facet; |
bbafd73b | 5026 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
5027 | const struct nlattr *actions = odp_actions->data; |
5028 | size_t actions_len = odp_actions->size; | |
d445cc16 JP |
5029 | struct odputil_keybuf maskbuf; |
5030 | struct ofpbuf mask; | |
4dff9097 | 5031 | |
6a7e895f | 5032 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 5033 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
5034 | int ret; |
5035 | ||
0f74f625 AZ |
5036 | flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE |
5037 | : DPIF_FP_MODIFY; | |
b0f7b9b5 BP |
5038 | if (stats) { |
5039 | flags |= DPIF_FP_ZERO_STATS; | |
5040 | } | |
5041 | ||
6a7e895f | 5042 | if (path == SF_SLOW_PATH) { |
b7e2f6b3 | 5043 | compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
5044 | slow_path_stub, sizeof slow_path_stub, |
5045 | &actions, &actions_len); | |
5046 | } | |
5047 | ||
d445cc16 | 5048 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6b35e630 JP |
5049 | if (enable_megaflows) { |
5050 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
5051 | &facet->flow, UINT32_MAX); | |
5052 | } | |
d445cc16 JP |
5053 | |
5054 | ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, | |
5055 | subfacet->key_len, mask.data, mask.size, | |
e6cc0bab | 5056 | actions, actions_len, stats); |
b0f7b9b5 BP |
5057 | |
5058 | if (stats) { | |
5059 | subfacet_reset_dp_stats(subfacet, stats); | |
5060 | } | |
5061 | ||
9b1a48c2 JP |
5062 | if (ret) { |
5063 | COVERAGE_INC(subfacet_install_fail); | |
5064 | } else { | |
6a7e895f BP |
5065 | subfacet->path = path; |
5066 | } | |
b0f7b9b5 BP |
5067 | return ret; |
5068 | } | |
5069 | ||
5070 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
5071 | static void | |
15baa734 | 5072 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5073 | { |
6a7e895f | 5074 | if (subfacet->path != SF_NOT_INSTALLED) { |
b7e2f6b3 | 5075 | struct ofproto_dpif *ofproto = subfacet->facet->ofproto; |
b0f7b9b5 | 5076 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5077 | int error; |
5078 | ||
9566abf9 EJ |
5079 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5080 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5081 | subfacet_reset_dp_stats(subfacet, &stats); |
5082 | if (!error) { | |
15baa734 | 5083 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5084 | } |
6a7e895f | 5085 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5086 | } else { |
cb22974d BP |
5087 | ovs_assert(subfacet->dp_packet_count == 0); |
5088 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5089 | } |
5090 | } | |
5091 | ||
5092 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5093 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5094 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5095 | * was reset in the datapath. 'stats' will be modified to include only | |
5096 | * statistics new since 'subfacet' was last updated. */ | |
5097 | static void | |
5098 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5099 | struct dpif_flow_stats *stats) | |
5100 | { | |
5101 | if (stats | |
5102 | && subfacet->dp_packet_count <= stats->n_packets | |
5103 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5104 | stats->n_packets -= subfacet->dp_packet_count; | |
5105 | stats->n_bytes -= subfacet->dp_byte_count; | |
5106 | } | |
5107 | ||
5108 | subfacet->dp_packet_count = 0; | |
5109 | subfacet->dp_byte_count = 0; | |
5110 | } | |
5111 | ||
b0f7b9b5 BP |
5112 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. |
5113 | * | |
5114 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5115 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5116 | * represents a packet that was sent by hand or if it represents statistics | |
5117 | * that have been cleared out of the datapath. */ | |
5118 | static void | |
15baa734 | 5119 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5120 | const struct dpif_flow_stats *stats) |
5121 | { | |
5122 | if (stats->n_packets || stats->used > subfacet->used) { | |
5123 | struct facet *facet = subfacet->facet; | |
5124 | ||
9dfb1f78 EJ |
5125 | subfacet->used = MAX(subfacet->used, stats->used); |
5126 | facet->used = MAX(facet->used, stats->used); | |
b0f7b9b5 BP |
5127 | facet->packet_count += stats->n_packets; |
5128 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5129 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5130 | } |
5131 | } | |
5132 | \f | |
abe529af BP |
5133 | /* Rules. */ |
5134 | ||
bcd2633a JP |
5135 | /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets |
5136 | * the fields that were relevant as part of the lookup. */ | |
abe529af | 5137 | static struct rule_dpif * |
bcd2633a JP |
5138 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
5139 | struct flow_wildcards *wc) | |
c57b2226 | 5140 | { |
ec89fc6f | 5141 | struct ofport_dpif *port; |
c57b2226 BP |
5142 | struct rule_dpif *rule; |
5143 | ||
9583bc14 | 5144 | rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0); |
c57b2226 BP |
5145 | if (rule) { |
5146 | return rule; | |
5147 | } | |
ec89fc6f EJ |
5148 | port = get_ofp_port(ofproto, flow->in_port.ofp_port); |
5149 | if (!port) { | |
5150 | VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16, | |
5151 | flow->in_port.ofp_port); | |
5152 | } | |
c57b2226 | 5153 | |
ec89fc6f EJ |
5154 | return choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule, |
5155 | ofproto->no_packet_in_rule); | |
c57b2226 BP |
5156 | } |
5157 | ||
9583bc14 EJ |
5158 | struct rule_dpif * |
5159 | rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, | |
5160 | const struct flow *flow, struct flow_wildcards *wc, | |
5161 | uint8_t table_id) | |
abe529af | 5162 | { |
7257b535 BP |
5163 | struct cls_rule *cls_rule; |
5164 | struct classifier *cls; | |
7fd51d39 | 5165 | bool frag; |
7257b535 | 5166 | |
9cdaaebe BP |
5167 | if (table_id >= N_TABLES) { |
5168 | return NULL; | |
5169 | } | |
5170 | ||
1dd35f8a | 5171 | if (wc) { |
7431e171 | 5172 | memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type); |
1dd35f8a JP |
5173 | wc->masks.nw_frag |= FLOW_NW_FRAG_MASK; |
5174 | } | |
5175 | ||
d0918789 | 5176 | cls = &ofproto->up.tables[table_id].cls; |
7fd51d39 BP |
5177 | frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; |
5178 | if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
5179 | /* We must pretend that transport ports are unavailable. */ | |
7257b535 BP |
5180 | struct flow ofpc_normal_flow = *flow; |
5181 | ofpc_normal_flow.tp_src = htons(0); | |
5182 | ofpc_normal_flow.tp_dst = htons(0); | |
0b4f2078 | 5183 | ovs_rwlock_rdlock(&cls->rwlock); |
bcd2633a | 5184 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); |
0b4f2078 | 5185 | ovs_rwlock_unlock(&cls->rwlock); |
7fd51d39 BP |
5186 | } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { |
5187 | cls_rule = &ofproto->drop_frags_rule->up.cr; | |
bcd2633a JP |
5188 | if (wc) { |
5189 | flow_wildcards_init_exact(wc); | |
5190 | } | |
7257b535 | 5191 | } else { |
0b4f2078 | 5192 | ovs_rwlock_rdlock(&cls->rwlock); |
bcd2633a | 5193 | cls_rule = classifier_lookup(cls, flow, wc); |
0b4f2078 | 5194 | ovs_rwlock_unlock(&cls->rwlock); |
7257b535 BP |
5195 | } |
5196 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5197 | } |
5198 | ||
ec89fc6f EJ |
5199 | /* Given a port configuration (specified as zero if there's no port), chooses |
5200 | * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a | |
5201 | * flow table miss. */ | |
9583bc14 | 5202 | struct rule_dpif * |
ec89fc6f EJ |
5203 | choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule, |
5204 | struct rule_dpif *no_packet_in_rule) | |
c376f9a3 | 5205 | { |
ec89fc6f | 5206 | return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; |
c376f9a3 IY |
5207 | } |
5208 | ||
7ee20df1 BP |
5209 | static void |
5210 | complete_operation(struct rule_dpif *rule) | |
5211 | { | |
5212 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5213 | ||
a1616063 | 5214 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
7ee20df1 BP |
5215 | if (clogged) { |
5216 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5217 | c->op = rule->up.pending; | |
5218 | list_push_back(&ofproto->completions, &c->list_node); | |
5219 | } else { | |
5220 | ofoperation_complete(rule->up.pending, 0); | |
5221 | } | |
5222 | } | |
5223 | ||
abe529af BP |
5224 | static struct rule * |
5225 | rule_alloc(void) | |
5226 | { | |
5227 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5228 | return &rule->up; | |
5229 | } | |
5230 | ||
5231 | static void | |
5232 | rule_dealloc(struct rule *rule_) | |
5233 | { | |
5234 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5235 | free(rule); | |
5236 | } | |
5237 | ||
90bf1e07 | 5238 | static enum ofperr |
abe529af BP |
5239 | rule_construct(struct rule *rule_) |
5240 | { | |
5241 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
9fbe253e EJ |
5242 | ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL); |
5243 | ovs_mutex_lock(&rule->stats_mutex); | |
abe529af BP |
5244 | rule->packet_count = 0; |
5245 | rule->byte_count = 0; | |
9fbe253e | 5246 | ovs_mutex_unlock(&rule->stats_mutex); |
7ee20df1 | 5247 | complete_operation(rule); |
abe529af BP |
5248 | return 0; |
5249 | } | |
5250 | ||
5251 | static void | |
9fbe253e | 5252 | rule_destruct(struct rule *rule_) |
abe529af | 5253 | { |
9fbe253e EJ |
5254 | struct rule_dpif *rule = rule_dpif_cast(rule_); |
5255 | complete_operation(rule); | |
5256 | ovs_mutex_destroy(&rule->stats_mutex); | |
abe529af BP |
5257 | } |
5258 | ||
5259 | static void | |
5260 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5261 | { | |
5262 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 5263 | |
4464589e EJ |
5264 | /* push_all_stats() can handle flow misses which, when using the learn |
5265 | * action, can cause rules to be added and deleted. This can corrupt our | |
5266 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5267 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5268 | push_all_stats__(false); | |
bf1e8ff9 | 5269 | |
abe529af BP |
5270 | /* Start from historical data for 'rule' itself that are no longer tracked |
5271 | * in facets. This counts, for example, facets that have expired. */ | |
9fbe253e | 5272 | ovs_mutex_lock(&rule->stats_mutex); |
abe529af BP |
5273 | *packets = rule->packet_count; |
5274 | *bytes = rule->byte_count; | |
9fbe253e | 5275 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af BP |
5276 | } |
5277 | ||
0a740f48 EJ |
5278 | static void |
5279 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5280 | struct ofpbuf *packet) | |
abe529af | 5281 | { |
abe529af | 5282 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
112bc5f4 | 5283 | struct dpif_flow_stats stats; |
bbafd73b EJ |
5284 | struct xlate_out xout; |
5285 | struct xlate_in xin; | |
abe529af | 5286 | |
a7752d4a | 5287 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5288 | rule_credit_stats(rule, &stats); |
5289 | ||
47b0deae | 5290 | xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); |
bbafd73b EJ |
5291 | xin.resubmit_stats = &stats; |
5292 | xlate_actions(&xin, &xout); | |
112bc5f4 | 5293 | |
bbafd73b EJ |
5294 | execute_odp_actions(ofproto, flow, xout.odp_actions.data, |
5295 | xout.odp_actions.size, packet); | |
112bc5f4 | 5296 | |
bbafd73b | 5297 | xlate_out_uninit(&xout); |
0a740f48 | 5298 | } |
5bf0e941 | 5299 | |
0a740f48 EJ |
5300 | static enum ofperr |
5301 | rule_execute(struct rule *rule, const struct flow *flow, | |
5302 | struct ofpbuf *packet) | |
5303 | { | |
5304 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5305 | ofpbuf_delete(packet); | |
5bf0e941 | 5306 | return 0; |
abe529af BP |
5307 | } |
5308 | ||
7ee20df1 BP |
5309 | static void |
5310 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5311 | { |
5312 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5313 | |
5314 | complete_operation(rule); | |
abe529af BP |
5315 | } |
5316 | \f | |
97d6520b | 5317 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5318 | * May modify 'packet'. |
abe529af BP |
5319 | * Returns 0 if successful, otherwise a positive errno value. */ |
5320 | static int | |
52a90c29 | 5321 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5322 | { |
79e15b78 | 5323 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5324 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 | 5325 | struct ofpbuf key, odp_actions; |
0f49659a | 5326 | struct dpif_flow_stats stats; |
80e5eed9 | 5327 | struct odputil_keybuf keybuf; |
0f49659a | 5328 | struct ofpact_output output; |
bbafd73b EJ |
5329 | struct xlate_out xout; |
5330 | struct xlate_in xin; | |
80e5eed9 | 5331 | struct flow flow; |
4e022ec0 | 5332 | union flow_in_port in_port_; |
abe529af BP |
5333 | int error; |
5334 | ||
b9ad7294 | 5335 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
0f49659a | 5336 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
b9ad7294 | 5337 | |
0f49659a | 5338 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ |
4e022ec0 AW |
5339 | in_port_.ofp_port = OFPP_NONE; |
5340 | flow_extract(packet, 0, 0, NULL, &in_port_, &flow); | |
0f49659a EJ |
5341 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, |
5342 | OFPP_LOCAL)); | |
5343 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 | 5344 | |
0f49659a EJ |
5345 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); |
5346 | output.port = ofport->up.ofp_port; | |
5347 | output.max_len = 0; | |
b9ad7294 | 5348 | |
47b0deae | 5349 | xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet); |
bbafd73b EJ |
5350 | xin.ofpacts_len = sizeof output; |
5351 | xin.ofpacts = &output.ofpact; | |
5352 | xin.resubmit_stats = &stats; | |
5353 | xlate_actions(&xin, &xout); | |
6ff686f2 | 5354 | |
acf60855 | 5355 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 | 5356 | key.data, key.size, |
bbafd73b | 5357 | xout.odp_actions.data, xout.odp_actions.size, |
abe529af | 5358 | packet); |
bbafd73b | 5359 | xlate_out_uninit(&xout); |
abe529af BP |
5360 | |
5361 | if (error) { | |
0f49659a EJ |
5362 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", |
5363 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
10a89ef0 | 5364 | ovs_strerror(error)); |
abe529af | 5365 | } |
79e15b78 EJ |
5366 | |
5367 | ofproto->stats.tx_packets++; | |
5368 | ofproto->stats.tx_bytes += packet->size; | |
abe529af BP |
5369 | return error; |
5370 | } | |
abe529af | 5371 | |
6a7e895f BP |
5372 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5373 | * The action will state 'slow' as the reason that the action is in the slow | |
5374 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5375 | * dump-flows" output to see why a flow is in the slow path.) | |
5376 | * | |
5377 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5378 | * 'stub_size' must be large enough for the action. | |
5379 | * | |
5380 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5381 | * respectively. */ | |
5382 | static void | |
5383 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5384 | enum slow_path_reason slow, | |
5385 | uint64_t *stub, size_t stub_size, | |
5386 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5387 | { | |
5388 | union user_action_cookie cookie; | |
5389 | struct ofpbuf buf; | |
5390 | ||
5391 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5392 | cookie.slow_path.unused = 0; | |
5393 | cookie.slow_path.reason = slow; | |
5394 | ||
5395 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5396 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
4e022ec0 AW |
5397 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, |
5398 | ODPP_NONE); | |
29089a54 | 5399 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5400 | } else { |
89a8a7f0 EJ |
5401 | odp_port_t odp_port; |
5402 | uint32_t pid; | |
5403 | ||
5404 | odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port); | |
5405 | pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port); | |
5406 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); | |
625b0720 | 5407 | } |
6a7e895f BP |
5408 | *actionsp = buf.data; |
5409 | *actions_lenp = buf.size; | |
5410 | } | |
9583bc14 EJ |
5411 | \f |
5412 | static bool | |
5413 | set_frag_handling(struct ofproto *ofproto_, | |
5414 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5415 | { |
5416 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 5417 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 5418 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
5419 | return true; |
5420 | } else { | |
5421 | return false; | |
5422 | } | |
abe529af BP |
5423 | } |
5424 | ||
90bf1e07 | 5425 | static enum ofperr |
abe529af BP |
5426 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
5427 | const struct flow *flow, | |
f25d0cf3 | 5428 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
5429 | { |
5430 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
548de4dd BP |
5431 | struct odputil_keybuf keybuf; |
5432 | struct dpif_flow_stats stats; | |
bbafd73b EJ |
5433 | struct xlate_out xout; |
5434 | struct xlate_in xin; | |
548de4dd | 5435 | struct ofpbuf key; |
112bc5f4 | 5436 | |
80e5eed9 | 5437 | |
548de4dd | 5438 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a | 5439 | odp_flow_key_from_flow(&key, flow, |
4e022ec0 AW |
5440 | ofp_port_to_odp_port(ofproto, |
5441 | flow->in_port.ofp_port)); | |
050ac423 | 5442 | |
548de4dd | 5443 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 5444 | |
47b0deae | 5445 | xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet); |
bbafd73b EJ |
5446 | xin.resubmit_stats = &stats; |
5447 | xin.ofpacts_len = ofpacts_len; | |
5448 | xin.ofpacts = ofpacts; | |
2284188b | 5449 | |
bbafd73b | 5450 | xlate_actions(&xin, &xout); |
acf60855 | 5451 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
bbafd73b EJ |
5452 | xout.odp_actions.data, xout.odp_actions.size, packet); |
5453 | xlate_out_uninit(&xout); | |
2284188b | 5454 | |
548de4dd | 5455 | return 0; |
abe529af | 5456 | } |
6fca1ffb BP |
5457 | \f |
5458 | /* NetFlow. */ | |
5459 | ||
5460 | static int | |
5461 | set_netflow(struct ofproto *ofproto_, | |
5462 | const struct netflow_options *netflow_options) | |
5463 | { | |
5464 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5465 | ||
5466 | if (netflow_options) { | |
5467 | if (!ofproto->netflow) { | |
5468 | ofproto->netflow = netflow_create(); | |
7bab1576 | 5469 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6fca1ffb BP |
5470 | } |
5471 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7bab1576 EJ |
5472 | } else if (ofproto->netflow) { |
5473 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
6fca1ffb BP |
5474 | netflow_destroy(ofproto->netflow); |
5475 | ofproto->netflow = NULL; | |
6fca1ffb | 5476 | } |
7bab1576 EJ |
5477 | |
5478 | return 0; | |
6fca1ffb | 5479 | } |
abe529af BP |
5480 | |
5481 | static void | |
5482 | get_netflow_ids(const struct ofproto *ofproto_, | |
5483 | uint8_t *engine_type, uint8_t *engine_id) | |
5484 | { | |
5485 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5486 | ||
acf60855 | 5487 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 5488 | } |
6fca1ffb BP |
5489 | |
5490 | static void | |
5491 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
5492 | { | |
5493 | if (!facet_is_controller_flow(facet) && | |
5494 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 5495 | struct subfacet *subfacet; |
6fca1ffb BP |
5496 | struct ofexpired expired; |
5497 | ||
b0f7b9b5 | 5498 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 5499 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 5500 | struct dpif_flow_stats stats; |
6fca1ffb | 5501 | |
bcd2633a JP |
5502 | subfacet_install(subfacet, &facet->xout.odp_actions, |
5503 | &stats); | |
15baa734 | 5504 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5505 | } |
6fca1ffb BP |
5506 | } |
5507 | ||
5508 | expired.flow = facet->flow; | |
5509 | expired.packet_count = facet->packet_count; | |
5510 | expired.byte_count = facet->byte_count; | |
5511 | expired.used = facet->used; | |
5512 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
5513 | } | |
5514 | } | |
5515 | ||
5516 | static void | |
5517 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
5518 | { | |
bcd2633a | 5519 | struct cls_cursor cursor; |
6fca1ffb BP |
5520 | struct facet *facet; |
5521 | ||
0b4f2078 | 5522 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5523 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5524 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6fca1ffb BP |
5525 | send_active_timeout(ofproto, facet); |
5526 | } | |
0b4f2078 | 5527 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6fca1ffb | 5528 | } |
abe529af BP |
5529 | \f |
5530 | static struct ofproto_dpif * | |
5531 | ofproto_dpif_lookup(const char *name) | |
5532 | { | |
b44a10b7 BP |
5533 | struct ofproto_dpif *ofproto; |
5534 | ||
5535 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
5536 | hash_string(name, 0), &all_ofproto_dpifs) { | |
5537 | if (!strcmp(ofproto->up.name, name)) { | |
5538 | return ofproto; | |
5539 | } | |
5540 | } | |
5541 | return NULL; | |
abe529af BP |
5542 | } |
5543 | ||
f0a3aa2e | 5544 | static void |
96e466a3 | 5545 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 5546 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 5547 | { |
490df1ef | 5548 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 5549 | |
96e466a3 EJ |
5550 | if (argc > 1) { |
5551 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5552 | if (!ofproto) { | |
bde9f75d | 5553 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
5554 | return; |
5555 | } | |
509c0149 | 5556 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5557 | mac_learning_flush(ofproto->ml); |
509c0149 | 5558 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 EJ |
5559 | } else { |
5560 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
509c0149 | 5561 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5562 | mac_learning_flush(ofproto->ml); |
509c0149 | 5563 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 | 5564 | } |
f0a3aa2e | 5565 | } |
f0a3aa2e | 5566 | |
bde9f75d | 5567 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
5568 | } |
5569 | ||
46c88433 EJ |
5570 | static struct ofport_dpif * |
5571 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5572 | { | |
5573 | return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif, | |
5574 | bundle_node); | |
5575 | } | |
5576 | ||
abe529af | 5577 | static void |
0e15264f BP |
5578 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5579 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
5580 | { |
5581 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5582 | const struct ofproto_dpif *ofproto; | |
5583 | const struct mac_entry *e; | |
5584 | ||
0e15264f | 5585 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 5586 | if (!ofproto) { |
bde9f75d | 5587 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
5588 | return; |
5589 | } | |
5590 | ||
5591 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
509c0149 | 5592 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
5593 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
5594 | struct ofbundle *bundle = e->port.p; | |
ad28062f BP |
5595 | char name[OFP_MAX_PORT_NAME_LEN]; |
5596 | ||
5597 | ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port, | |
5598 | name, sizeof name); | |
5599 | ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n", | |
5600 | name, e->vlan, ETH_ADDR_ARGS(e->mac), | |
e764773c | 5601 | mac_entry_age(ofproto->ml, e)); |
abe529af | 5602 | } |
509c0149 | 5603 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
bde9f75d | 5604 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
5605 | ds_destroy(&ds); |
5606 | } | |
5607 | ||
6a6455e5 | 5608 | struct trace_ctx { |
bbafd73b EJ |
5609 | struct xlate_out xout; |
5610 | struct xlate_in xin; | |
abe529af BP |
5611 | struct flow flow; |
5612 | struct ds *result; | |
5613 | }; | |
5614 | ||
5615 | static void | |
4d0acc70 | 5616 | trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) |
abe529af BP |
5617 | { |
5618 | ds_put_char_multiple(result, '\t', level); | |
5619 | if (!rule) { | |
5620 | ds_put_cstr(result, "No match\n"); | |
5621 | return; | |
5622 | } | |
5623 | ||
29901626 | 5624 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
4d0acc70 | 5625 | rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie)); |
79feb7df | 5626 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5627 | ds_put_char(result, '\n'); |
5628 | ||
5629 | ds_put_char_multiple(result, '\t', level); | |
5630 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 5631 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
5632 | ds_put_char(result, '\n'); |
5633 | } | |
5634 | ||
5635 | static void | |
5636 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 5637 | struct trace_ctx *trace) |
abe529af BP |
5638 | { |
5639 | ds_put_char_multiple(result, '\t', level); | |
5640 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 5641 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
5642 | ds_put_cstr(result, "unchanged"); |
5643 | } else { | |
bbafd73b EJ |
5644 | flow_format(result, &trace->xin.flow); |
5645 | trace->flow = trace->xin.flow; | |
abe529af BP |
5646 | } |
5647 | ds_put_char(result, '\n'); | |
5648 | } | |
5649 | ||
eb9e1c26 EJ |
5650 | static void |
5651 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 5652 | struct trace_ctx *trace) |
eb9e1c26 EJ |
5653 | { |
5654 | size_t i; | |
5655 | ||
5656 | ds_put_char_multiple(result, '\t', level); | |
5657 | ds_put_format(result, "%s:", title); | |
5658 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5659 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5660 | } | |
5661 | ds_put_char(result, '\n'); | |
5662 | } | |
5663 | ||
1ed8d352 EJ |
5664 | static void |
5665 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 5666 | struct trace_ctx *trace) |
1ed8d352 | 5667 | { |
bbafd73b | 5668 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
5669 | |
5670 | ds_put_char_multiple(result, '\t', level); | |
5671 | ds_put_format(result, "%s: ", title); | |
5672 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
5673 | ds_put_char(result, '\n'); | |
5674 | } | |
5675 | ||
abe529af | 5676 | static void |
4d0acc70 | 5677 | trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse) |
abe529af | 5678 | { |
4d0acc70 | 5679 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
abe529af BP |
5680 | struct ds *result = trace->result; |
5681 | ||
5682 | ds_put_char(result, '\n'); | |
4d0acc70 EJ |
5683 | trace_format_flow(result, recurse + 1, "Resubmitted flow", trace); |
5684 | trace_format_regs(result, recurse + 1, "Resubmitted regs", trace); | |
5685 | trace_format_odp(result, recurse + 1, "Resubmitted odp", trace); | |
5686 | trace_format_rule(result, recurse + 1, rule); | |
abe529af BP |
5687 | } |
5688 | ||
479df176 | 5689 | static void |
4d0acc70 | 5690 | trace_report(struct xlate_in *xin, const char *s, int recurse) |
479df176 | 5691 | { |
4d0acc70 | 5692 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
479df176 BP |
5693 | struct ds *result = trace->result; |
5694 | ||
4d0acc70 | 5695 | ds_put_char_multiple(result, '\t', recurse); |
479df176 BP |
5696 | ds_put_cstr(result, s); |
5697 | ds_put_char(result, '\n'); | |
5698 | } | |
5699 | ||
abe529af | 5700 | static void |
0e15264f | 5701 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
5702 | void *aux OVS_UNUSED) |
5703 | { | |
50aa28fd | 5704 | const struct dpif_backer *backer; |
abe529af | 5705 | struct ofproto_dpif *ofproto; |
ea5e3887 | 5706 | struct ofpbuf odp_key, odp_mask; |
876b0e1c | 5707 | struct ofpbuf *packet; |
abe529af BP |
5708 | struct ds result; |
5709 | struct flow flow; | |
abe529af BP |
5710 | char *s; |
5711 | ||
876b0e1c | 5712 | packet = NULL; |
50aa28fd | 5713 | backer = NULL; |
abe529af | 5714 | ds_init(&result); |
50aa28fd | 5715 | ofpbuf_init(&odp_key, 0); |
ea5e3887 | 5716 | ofpbuf_init(&odp_mask, 0); |
abe529af | 5717 | |
50aa28fd AW |
5718 | /* Handle "-generate" or a hex string as the last argument. */ |
5719 | if (!strcmp(argv[argc - 1], "-generate")) { | |
5720 | packet = ofpbuf_new(0); | |
5721 | argc--; | |
5722 | } else { | |
5723 | const char *error = eth_from_hex(argv[argc - 1], &packet); | |
5724 | if (!error) { | |
5725 | argc--; | |
5726 | } else if (argc == 4) { | |
5727 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
5728 | unixctl_command_reply_error(conn, error); | |
5729 | goto exit; | |
5730 | } | |
e84173dc | 5731 | } |
876b0e1c | 5732 | |
50aa28fd AW |
5733 | /* Parse the flow and determine whether a datapath or |
5734 | * bridge is specified. If function odp_flow_key_from_string() | |
5735 | * returns 0, the flow is a odp_flow. If function | |
5736 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
ea5e3887 | 5737 | if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) { |
50aa28fd AW |
5738 | /* If the odp_flow is the second argument, |
5739 | * the datapath name is the first argument. */ | |
5740 | if (argc == 3) { | |
5741 | const char *dp_type; | |
5742 | if (!strncmp(argv[1], "ovs-", 4)) { | |
5743 | dp_type = argv[1] + 4; | |
5744 | } else { | |
5745 | dp_type = argv[1]; | |
31a19d69 | 5746 | } |
50aa28fd AW |
5747 | backer = shash_find_data(&all_dpif_backers, dp_type); |
5748 | if (!backer) { | |
5749 | unixctl_command_reply_error(conn, "Cannot find datapath " | |
5750 | "of this name"); | |
31a19d69 BP |
5751 | goto exit; |
5752 | } | |
5753 | } else { | |
50aa28fd AW |
5754 | /* No datapath name specified, so there should be only one |
5755 | * datapath. */ | |
5756 | struct shash_node *node; | |
5757 | if (shash_count(&all_dpif_backers) != 1) { | |
5758 | unixctl_command_reply_error(conn, "Must specify datapath " | |
5759 | "name, there is more than one type of datapath"); | |
31a19d69 BP |
5760 | goto exit; |
5761 | } | |
50aa28fd AW |
5762 | node = shash_first(&all_dpif_backers); |
5763 | backer = node->data; | |
876b0e1c | 5764 | } |
8b3b8dd1 | 5765 | |
8449c4d6 EJ |
5766 | if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow, |
5767 | NULL, &ofproto, NULL)) { | |
50aa28fd AW |
5768 | unixctl_command_reply_error(conn, "Invalid datapath flow"); |
5769 | goto exit; | |
8b3b8dd1 | 5770 | } |
50aa28fd AW |
5771 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
5772 | } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { | |
5773 | if (argc != 3) { | |
5774 | unixctl_command_reply_error(conn, "Must specify bridge name"); | |
876b0e1c BP |
5775 | goto exit; |
5776 | } | |
5777 | ||
50aa28fd AW |
5778 | ofproto = ofproto_dpif_lookup(argv[1]); |
5779 | if (!ofproto) { | |
5780 | unixctl_command_reply_error(conn, "Unknown bridge name"); | |
5781 | goto exit; | |
5782 | } | |
876b0e1c | 5783 | } else { |
50aa28fd | 5784 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
abe529af BP |
5785 | goto exit; |
5786 | } | |
5787 | ||
50aa28fd AW |
5788 | /* Generate a packet, if requested. */ |
5789 | if (packet) { | |
5790 | if (!packet->size) { | |
5791 | flow_compose(packet, &flow); | |
5792 | } else { | |
4e022ec0 AW |
5793 | union flow_in_port in_port_; |
5794 | ||
5795 | in_port_ = flow.in_port; | |
50aa28fd AW |
5796 | ds_put_cstr(&result, "Packet: "); |
5797 | s = ofp_packet_to_string(packet->data, packet->size); | |
5798 | ds_put_cstr(&result, s); | |
5799 | free(s); | |
5800 | ||
5801 | /* Use the metadata from the flow and the packet argument | |
5802 | * to reconstruct the flow. */ | |
5803 | flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, | |
4e022ec0 | 5804 | &in_port_, &flow); |
50aa28fd AW |
5805 | } |
5806 | } | |
5807 | ||
47b0deae | 5808 | ofproto_trace(ofproto, &flow, packet, &result); |
6a6455e5 EJ |
5809 | unixctl_command_reply(conn, ds_cstr(&result)); |
5810 | ||
5811 | exit: | |
5812 | ds_destroy(&result); | |
5813 | ofpbuf_delete(packet); | |
5814 | ofpbuf_uninit(&odp_key); | |
ea5e3887 | 5815 | ofpbuf_uninit(&odp_mask); |
6a6455e5 EJ |
5816 | } |
5817 | ||
ade6ad9c | 5818 | static void |
6a6455e5 | 5819 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, |
47b0deae | 5820 | const struct ofpbuf *packet, struct ds *ds) |
6a6455e5 EJ |
5821 | { |
5822 | struct rule_dpif *rule; | |
5823 | ||
5824 | ds_put_cstr(ds, "Flow: "); | |
5825 | flow_format(ds, flow); | |
5826 | ds_put_char(ds, '\n'); | |
abe529af | 5827 | |
bcd2633a | 5828 | rule = rule_dpif_lookup(ofproto, flow, NULL); |
c57b2226 | 5829 | |
4d0acc70 | 5830 | trace_format_rule(ds, 0, rule); |
c57b2226 BP |
5831 | if (rule == ofproto->miss_rule) { |
5832 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
5833 | } else if (rule == ofproto->no_packet_in_rule) { | |
5834 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
5835 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7fd51d39 BP |
5836 | } else if (rule == ofproto->drop_frags_rule) { |
5837 | ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " | |
5838 | "and the fragment handling mode is \"drop\".\n"); | |
c57b2226 BP |
5839 | } |
5840 | ||
abe529af | 5841 | if (rule) { |
050ac423 BP |
5842 | uint64_t odp_actions_stub[1024 / 8]; |
5843 | struct ofpbuf odp_actions; | |
6a6455e5 | 5844 | struct trace_ctx trace; |
bcd2633a | 5845 | struct match match; |
0e553d9c | 5846 | uint8_t tcp_flags; |
abe529af | 5847 | |
6a6455e5 EJ |
5848 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
5849 | trace.result = ds; | |
5850 | trace.flow = *flow; | |
050ac423 BP |
5851 | ofpbuf_use_stub(&odp_actions, |
5852 | odp_actions_stub, sizeof odp_actions_stub); | |
47b0deae | 5853 | xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); |
bbafd73b EJ |
5854 | trace.xin.resubmit_hook = trace_resubmit; |
5855 | trace.xin.report_hook = trace_report; | |
bcd2633a | 5856 | |
bbafd73b | 5857 | xlate_actions(&trace.xin, &trace.xout); |
abe529af | 5858 | |
6a6455e5 EJ |
5859 | ds_put_char(ds, '\n'); |
5860 | trace_format_flow(ds, 0, "Final flow", &trace); | |
bcd2633a JP |
5861 | |
5862 | match_init(&match, flow, &trace.xout.wc); | |
5863 | ds_put_cstr(ds, "Relevant fields: "); | |
5864 | match_format(&match, ds, OFP_DEFAULT_PRIORITY); | |
5865 | ds_put_char(ds, '\n'); | |
5866 | ||
6a6455e5 | 5867 | ds_put_cstr(ds, "Datapath actions: "); |
bbafd73b EJ |
5868 | format_odp_actions(ds, trace.xout.odp_actions.data, |
5869 | trace.xout.odp_actions.size); | |
876b0e1c | 5870 | |
bbafd73b | 5871 | if (trace.xout.slow) { |
6a7e895f BP |
5872 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
5873 | "slow path because it:"); | |
bbafd73b | 5874 | switch (trace.xout.slow) { |
98f0520f EJ |
5875 | case SLOW_CFM: |
5876 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
5877 | break; | |
5878 | case SLOW_LACP: | |
5879 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
5880 | break; | |
5881 | case SLOW_STP: | |
5882 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
5883 | break; | |
5884 | case SLOW_BFD: | |
5885 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
5886 | break; | |
5887 | case SLOW_CONTROLLER: | |
5888 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
5889 | "to the OpenFlow controller."); | |
5890 | break; | |
5891 | case __SLOW_MAX: | |
5892 | NOT_REACHED(); | |
6a7e895f | 5893 | } |
876b0e1c | 5894 | } |
bbafd73b EJ |
5895 | |
5896 | xlate_out_uninit(&trace.xout); | |
abe529af | 5897 | } |
abe529af BP |
5898 | } |
5899 | ||
7ee20df1 | 5900 | static void |
0e15264f BP |
5901 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
5902 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
5903 | { |
5904 | clogged = true; | |
bde9f75d | 5905 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
5906 | } |
5907 | ||
5908 | static void | |
0e15264f BP |
5909 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
5910 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
5911 | { |
5912 | clogged = false; | |
bde9f75d | 5913 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
5914 | } |
5915 | ||
6814e51f BP |
5916 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
5917 | * 'reply' describing the results. */ | |
5918 | static void | |
5919 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
5920 | { | |
bcd2633a | 5921 | struct cls_cursor cursor; |
6814e51f BP |
5922 | struct facet *facet; |
5923 | int errors; | |
5924 | ||
5925 | errors = 0; | |
0b4f2078 | 5926 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5927 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5928 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6814e51f BP |
5929 | if (!facet_check_consistency(facet)) { |
5930 | errors++; | |
5931 | } | |
5932 | } | |
0b4f2078 | 5933 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 5934 | if (errors) { |
2cc3c58e | 5935 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
5936 | } |
5937 | ||
5938 | if (errors) { | |
5939 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
5940 | ofproto->up.name, errors); | |
5941 | } else { | |
5942 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
5943 | } | |
5944 | } | |
5945 | ||
5946 | static void | |
5947 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
5948 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
5949 | { | |
5950 | struct ds reply = DS_EMPTY_INITIALIZER; | |
5951 | struct ofproto_dpif *ofproto; | |
5952 | ||
5953 | if (argc > 1) { | |
5954 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5955 | if (!ofproto) { | |
bde9f75d EJ |
5956 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
5957 | "ofproto/list for help)"); | |
6814e51f BP |
5958 | return; |
5959 | } | |
5960 | ofproto_dpif_self_check__(ofproto, &reply); | |
5961 | } else { | |
5962 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5963 | ofproto_dpif_self_check__(ofproto, &reply); | |
5964 | } | |
5965 | } | |
5966 | ||
bde9f75d | 5967 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
5968 | ds_destroy(&reply); |
5969 | } | |
5970 | ||
27022416 JP |
5971 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
5972 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
5973 | * to destroy 'ofproto_shash' and free the returned value. */ | |
5974 | static const struct shash_node ** | |
5975 | get_ofprotos(struct shash *ofproto_shash) | |
5976 | { | |
5977 | const struct ofproto_dpif *ofproto; | |
5978 | ||
5979 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5980 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
5981 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
5982 | } | |
5983 | ||
5984 | return shash_sort(ofproto_shash); | |
5985 | } | |
5986 | ||
5987 | static void | |
5988 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
5989 | const char *argv[] OVS_UNUSED, | |
5990 | void *aux OVS_UNUSED) | |
5991 | { | |
5992 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5993 | struct shash ofproto_shash; | |
5994 | const struct shash_node **sorted_ofprotos; | |
5995 | int i; | |
5996 | ||
5997 | shash_init(&ofproto_shash); | |
5998 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
5999 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
6000 | const struct shash_node *node = sorted_ofprotos[i]; | |
6001 | ds_put_format(&ds, "%s\n", node->name); | |
6002 | } | |
6003 | ||
6004 | shash_destroy(&ofproto_shash); | |
6005 | free(sorted_ofprotos); | |
6006 | ||
6007 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6008 | ds_destroy(&ds); | |
6009 | } | |
6010 | ||
6011 | static void | |
dc54ef36 EJ |
6012 | show_dp_rates(struct ds *ds, const char *heading, |
6013 | const struct avg_subfacet_rates *rates) | |
6014 | { | |
6015 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
6016 | heading, rates->add_rate, rates->del_rate); | |
6017 | } | |
6018 | ||
6019 | static void | |
6020 | dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) | |
27022416 | 6021 | { |
dc54ef36 EJ |
6022 | const struct shash_node **ofprotos; |
6023 | struct ofproto_dpif *ofproto; | |
6024 | struct shash ofproto_shash; | |
09672174 | 6025 | uint64_t n_hit, n_missed; |
dc54ef36 | 6026 | long long int minutes; |
09672174 | 6027 | size_t i; |
655ab909 | 6028 | |
04d08d54 | 6029 | n_hit = n_missed = 0; |
dc54ef36 EJ |
6030 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
6031 | if (ofproto->backer == backer) { | |
dc54ef36 EJ |
6032 | n_missed += ofproto->n_missed; |
6033 | n_hit += ofproto->n_hit; | |
6034 | } | |
6035 | } | |
655ab909 | 6036 | |
dc54ef36 EJ |
6037 | ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
6038 | dpif_name(backer->dpif), n_hit, n_missed); | |
6039 | ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," | |
04d08d54 | 6040 | " life span: %lldms\n", hmap_count(&backer->subfacets), |
dc54ef36 EJ |
6041 | backer->avg_n_subfacet, backer->max_n_subfacet, |
6042 | backer->avg_subfacet_life); | |
6043 | ||
6044 | minutes = (time_msec() - backer->created) / (1000 * 60); | |
655ab909 | 6045 | if (minutes >= 60) { |
dc54ef36 | 6046 | show_dp_rates(ds, "\thourly avg:", &backer->hourly); |
655ab909 AZ |
6047 | } |
6048 | if (minutes >= 60 * 24) { | |
dc54ef36 | 6049 | show_dp_rates(ds, "\tdaily avg:", &backer->daily); |
655ab909 | 6050 | } |
dc54ef36 | 6051 | show_dp_rates(ds, "\toverall avg:", &backer->lifetime); |
27022416 | 6052 | |
dc54ef36 EJ |
6053 | shash_init(&ofproto_shash); |
6054 | ofprotos = get_ofprotos(&ofproto_shash); | |
6055 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
6056 | struct ofproto_dpif *ofproto = ofprotos[i]->data; | |
6057 | const struct shash_node **ports; | |
6058 | size_t j; | |
0a740f48 | 6059 | |
dc54ef36 EJ |
6060 | if (ofproto->backer != backer) { |
6061 | continue; | |
0a740f48 | 6062 | } |
27022416 | 6063 | |
dc54ef36 EJ |
6064 | ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
6065 | ofproto->up.name, ofproto->n_hit, ofproto->n_missed); | |
6066 | ||
6067 | ports = shash_sort(&ofproto->up.port_by_name); | |
6068 | for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { | |
6069 | const struct shash_node *node = ports[j]; | |
6070 | struct ofport *ofport = node->data; | |
6071 | struct smap config; | |
4e022ec0 | 6072 | odp_port_t odp_port; |
27022416 | 6073 | |
dc54ef36 EJ |
6074 | ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), |
6075 | ofport->ofp_port); | |
27022416 | 6076 | |
dc54ef36 | 6077 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); |
4e022ec0 | 6078 | if (odp_port != ODPP_NONE) { |
dc54ef36 EJ |
6079 | ds_put_format(ds, "%"PRIu32":", odp_port); |
6080 | } else { | |
6081 | ds_put_cstr(ds, "none:"); | |
6082 | } | |
27022416 | 6083 | |
dc54ef36 | 6084 | ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); |
27022416 | 6085 | |
dc54ef36 EJ |
6086 | smap_init(&config); |
6087 | if (!netdev_get_config(ofport->netdev, &config)) { | |
6088 | const struct smap_node **nodes; | |
6089 | size_t i; | |
27022416 | 6090 | |
dc54ef36 EJ |
6091 | nodes = smap_sort(&config); |
6092 | for (i = 0; i < smap_count(&config); i++) { | |
6093 | const struct smap_node *node = nodes[i]; | |
6094 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
6095 | node->key, node->value); | |
6096 | } | |
6097 | free(nodes); | |
27022416 | 6098 | } |
dc54ef36 EJ |
6099 | smap_destroy(&config); |
6100 | ||
27022416 | 6101 | ds_put_char(ds, ')'); |
dc54ef36 | 6102 | ds_put_char(ds, '\n'); |
27022416 | 6103 | } |
dc54ef36 | 6104 | free(ports); |
27022416 | 6105 | } |
dc54ef36 EJ |
6106 | shash_destroy(&ofproto_shash); |
6107 | free(ofprotos); | |
27022416 JP |
6108 | } |
6109 | ||
6110 | static void | |
dc54ef36 EJ |
6111 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6112 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
27022416 JP |
6113 | { |
6114 | struct ds ds = DS_EMPTY_INITIALIZER; | |
dc54ef36 EJ |
6115 | const struct shash_node **backers; |
6116 | int i; | |
27022416 | 6117 | |
dc54ef36 EJ |
6118 | backers = shash_sort(&all_dpif_backers); |
6119 | for (i = 0; i < shash_count(&all_dpif_backers); i++) { | |
6120 | dpif_show_backer(backers[i]->data, &ds); | |
27022416 | 6121 | } |
dc54ef36 | 6122 | free(backers); |
27022416 JP |
6123 | |
6124 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6125 | ds_destroy(&ds); | |
6126 | } | |
6127 | ||
bcd2633a JP |
6128 | /* Dump the megaflow (facet) cache. This is useful to check the |
6129 | * correctness of flow wildcarding, since the same mechanism is used for | |
6130 | * both xlate caching and kernel wildcarding. | |
6131 | * | |
6132 | * It's important to note that in the output the flow description uses | |
6133 | * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports. | |
6134 | * | |
6135 | * This command is only needed for advanced debugging, so it's not | |
6136 | * documented in the man page. */ | |
6137 | static void | |
6138 | ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn, | |
6139 | int argc OVS_UNUSED, const char *argv[], | |
6140 | void *aux OVS_UNUSED) | |
6141 | { | |
6142 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6143 | const struct ofproto_dpif *ofproto; | |
6144 | long long int now = time_msec(); | |
6145 | struct cls_cursor cursor; | |
6146 | struct facet *facet; | |
6147 | ||
6148 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6149 | if (!ofproto) { | |
6150 | unixctl_command_reply_error(conn, "no such bridge"); | |
6151 | return; | |
6152 | } | |
6153 | ||
0b4f2078 | 6154 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
6155 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
6156 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6157 | cls_rule_format(&facet->cr, &ds); | |
6158 | ds_put_cstr(&ds, ", "); | |
7ee9e841 | 6159 | ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets)); |
bcd2633a JP |
6160 | ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); |
6161 | ds_put_cstr(&ds, "Datapath actions: "); | |
8ecafa3e JP |
6162 | if (facet->xout.slow) { |
6163 | uint64_t slow_path_stub[128 / 8]; | |
6164 | const struct nlattr *actions; | |
6165 | size_t actions_len; | |
6166 | ||
6167 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, | |
6168 | slow_path_stub, sizeof slow_path_stub, | |
6169 | &actions, &actions_len); | |
6170 | format_odp_actions(&ds, actions, actions_len); | |
6171 | } else { | |
6172 | format_odp_actions(&ds, facet->xout.odp_actions.data, | |
6173 | facet->xout.odp_actions.size); | |
6174 | } | |
bcd2633a JP |
6175 | ds_put_cstr(&ds, "\n"); |
6176 | } | |
0b4f2078 | 6177 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a JP |
6178 | |
6179 | ds_chomp(&ds, '\n'); | |
6180 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6181 | ds_destroy(&ds); | |
6182 | } | |
6183 | ||
6b35e630 JP |
6184 | /* Disable using the megaflows. |
6185 | * | |
6186 | * This command is only needed for advanced debugging, so it's not | |
6187 | * documented in the man page. */ | |
6188 | static void | |
6189 | ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn, | |
6190 | int argc OVS_UNUSED, | |
6191 | const char *argv[] OVS_UNUSED, | |
6192 | void *aux OVS_UNUSED) | |
6193 | { | |
6194 | struct ofproto_dpif *ofproto; | |
6195 | ||
6196 | enable_megaflows = false; | |
6197 | ||
6198 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6199 | flush(&ofproto->up); | |
6200 | } | |
6201 | ||
6202 | unixctl_command_reply(conn, "megaflows disabled"); | |
6203 | } | |
6204 | ||
6205 | /* Re-enable using megaflows. | |
6206 | * | |
6207 | * This command is only needed for advanced debugging, so it's not | |
6208 | * documented in the man page. */ | |
6209 | static void | |
6210 | ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn, | |
6211 | int argc OVS_UNUSED, | |
6212 | const char *argv[] OVS_UNUSED, | |
6213 | void *aux OVS_UNUSED) | |
6214 | { | |
6215 | struct ofproto_dpif *ofproto; | |
6216 | ||
6217 | enable_megaflows = true; | |
6218 | ||
6219 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6220 | flush(&ofproto->up); | |
6221 | } | |
6222 | ||
6223 | unixctl_command_reply(conn, "megaflows enabled"); | |
6224 | } | |
6225 | ||
27022416 JP |
6226 | static void |
6227 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
6228 | int argc OVS_UNUSED, const char *argv[], | |
6229 | void *aux OVS_UNUSED) | |
6230 | { | |
6231 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6232 | const struct ofproto_dpif *ofproto; | |
6233 | struct subfacet *subfacet; | |
6234 | ||
6235 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6236 | if (!ofproto) { | |
6237 | unixctl_command_reply_error(conn, "no such bridge"); | |
6238 | return; | |
6239 | } | |
6240 | ||
af37354d EJ |
6241 | update_stats(ofproto->backer); |
6242 | ||
04d08d54 | 6243 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { |
4dff9097 | 6244 | struct facet *facet = subfacet->facet; |
64f9cb76 JP |
6245 | struct odputil_keybuf maskbuf; |
6246 | struct ofpbuf mask; | |
4dff9097 | 6247 | |
b7e2f6b3 | 6248 | if (facet->ofproto != ofproto) { |
04d08d54 EJ |
6249 | continue; |
6250 | } | |
6251 | ||
64f9cb76 JP |
6252 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6253 | if (enable_megaflows) { | |
6254 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
6255 | &facet->flow, UINT32_MAX); | |
6256 | } | |
6257 | ||
6258 | odp_flow_format(subfacet->key, subfacet->key_len, | |
041e7168 | 6259 | mask.data, mask.size, &ds, false); |
27022416 JP |
6260 | |
6261 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
6262 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
6263 | if (subfacet->used) { | |
6264 | ds_put_format(&ds, "%.3fs", | |
6265 | (time_msec() - subfacet->used) / 1000.0); | |
6266 | } else { | |
6267 | ds_put_format(&ds, "never"); | |
6268 | } | |
6269 | if (subfacet->facet->tcp_flags) { | |
6270 | ds_put_cstr(&ds, ", flags:"); | |
6271 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
6272 | } | |
6273 | ||
6274 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 6275 | if (facet->xout.slow) { |
f2245da3 JP |
6276 | uint64_t slow_path_stub[128 / 8]; |
6277 | const struct nlattr *actions; | |
6278 | size_t actions_len; | |
6279 | ||
bbafd73b | 6280 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
6281 | slow_path_stub, sizeof slow_path_stub, |
6282 | &actions, &actions_len); | |
6283 | format_odp_actions(&ds, actions, actions_len); | |
6284 | } else { | |
bbafd73b EJ |
6285 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
6286 | facet->xout.odp_actions.size); | |
f2245da3 | 6287 | } |
27022416 JP |
6288 | ds_put_char(&ds, '\n'); |
6289 | } | |
6290 | ||
6291 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6292 | ds_destroy(&ds); | |
6293 | } | |
6294 | ||
6295 | static void | |
6296 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
6297 | int argc OVS_UNUSED, const char *argv[], | |
6298 | void *aux OVS_UNUSED) | |
6299 | { | |
6300 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6301 | struct ofproto_dpif *ofproto; | |
6302 | ||
6303 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6304 | if (!ofproto) { | |
6305 | unixctl_command_reply_error(conn, "no such bridge"); | |
6306 | return; | |
6307 | } | |
6308 | ||
6309 | flush(&ofproto->up); | |
6310 | ||
6311 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6312 | ds_destroy(&ds); | |
6313 | } | |
6314 | ||
abe529af BP |
6315 | static void |
6316 | ofproto_dpif_unixctl_init(void) | |
6317 | { | |
6318 | static bool registered; | |
6319 | if (registered) { | |
6320 | return; | |
6321 | } | |
6322 | registered = true; | |
6323 | ||
0e15264f BP |
6324 | unixctl_command_register( |
6325 | "ofproto/trace", | |
50aa28fd AW |
6326 | "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", |
6327 | 1, 3, ofproto_unixctl_trace, NULL); | |
96e466a3 | 6328 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6329 | ofproto_unixctl_fdb_flush, NULL); |
6330 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6331 | ofproto_unixctl_fdb_show, NULL); | |
6332 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6333 | ofproto_dpif_clog, NULL); | |
6334 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6335 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6336 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6337 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
6338 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
6339 | ofproto_unixctl_dpif_dump_dps, NULL); | |
dc54ef36 EJ |
6340 | unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show, |
6341 | NULL); | |
27022416 JP |
6342 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, |
6343 | ofproto_unixctl_dpif_dump_flows, NULL); | |
6344 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
6345 | ofproto_unixctl_dpif_del_flows, NULL); | |
bcd2633a JP |
6346 | unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1, |
6347 | ofproto_unixctl_dpif_dump_megaflows, NULL); | |
6b35e630 JP |
6348 | unixctl_command_register("dpif/disable-megaflows", "", 0, 0, |
6349 | ofproto_unixctl_dpif_disable_megaflows, NULL); | |
6350 | unixctl_command_register("dpif/enable-megaflows", "", 0, 0, | |
6351 | ofproto_unixctl_dpif_enable_megaflows, NULL); | |
abe529af BP |
6352 | } |
6353 | \f | |
52a90c29 BP |
6354 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6355 | * | |
6356 | * This is deprecated. It is only for compatibility with broken device drivers | |
6357 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6358 | * devices are not used. When broken device drivers are no longer in | |
6359 | * widespread use, we will delete these interfaces. */ | |
6360 | ||
6361 | static int | |
4e022ec0 | 6362 | set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6363 | { |
6364 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6365 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6366 | ||
6367 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6368 | && vid == ofport->vlandev_vid) { | |
6369 | return 0; | |
6370 | } | |
6371 | ||
2cc3c58e | 6372 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
6373 | |
6374 | if (ofport->realdev_ofp_port) { | |
6375 | vsp_remove(ofport); | |
6376 | } | |
6377 | if (realdev_ofp_port && ofport->bundle) { | |
6378 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6379 | * themselves be part of a bundle. */ | |
6380 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6381 | } | |
6382 | ||
6383 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6384 | ofport->vlandev_vid = vid; | |
6385 | ||
6386 | if (realdev_ofp_port) { | |
6387 | vsp_add(ofport, realdev_ofp_port, vid); | |
6388 | } | |
6389 | ||
6390 | return 0; | |
6391 | } | |
6392 | ||
6393 | static uint32_t | |
4e022ec0 | 6394 | hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid) |
52a90c29 | 6395 | { |
4e022ec0 | 6396 | return hash_2words(ofp_to_u16(realdev_ofp_port), vid); |
52a90c29 BP |
6397 | } |
6398 | ||
46c88433 EJ |
6399 | bool |
6400 | ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto) | |
13501305 | 6401 | OVS_EXCLUDED(ofproto->vsp_mutex) |
46c88433 | 6402 | { |
13501305 EJ |
6403 | bool ret; |
6404 | ||
6405 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
6406 | ret = !hmap_is_empty(&ofproto->realdev_vid_map); | |
6407 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
6408 | return ret; | |
46c88433 EJ |
6409 | } |
6410 | ||
13501305 EJ |
6411 | static ofp_port_t |
6412 | vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto, | |
6413 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
6414 | OVS_REQUIRES(ofproto->vsp_mutex) | |
52a90c29 BP |
6415 | { |
6416 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
52a90c29 BP |
6417 | int vid = vlan_tci_to_vid(vlan_tci); |
6418 | const struct vlan_splinter *vsp; | |
6419 | ||
6420 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6421 | hash_realdev_vid(realdev_ofp_port, vid), | |
6422 | &ofproto->realdev_vid_map) { | |
6423 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6424 | && vsp->vid == vid) { | |
deea1200 | 6425 | return vsp->vlandev_ofp_port; |
52a90c29 BP |
6426 | } |
6427 | } | |
6428 | } | |
deea1200 | 6429 | return realdev_ofp_port; |
52a90c29 BP |
6430 | } |
6431 | ||
13501305 EJ |
6432 | /* Returns the OFP port number of the Linux VLAN device that corresponds to |
6433 | * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in | |
6434 | * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and | |
6435 | * 'vlan_tci' 9, it would return the port number of eth0.9. | |
6436 | * | |
6437 | * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this | |
6438 | * function just returns its 'realdev_ofp_port' argument. */ | |
6439 | ofp_port_t | |
6440 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6441 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
6442 | OVS_EXCLUDED(ofproto->vsp_mutex) | |
6443 | { | |
6444 | ofp_port_t ret; | |
6445 | ||
6446 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
6447 | ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci); | |
6448 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
6449 | return ret; | |
6450 | } | |
6451 | ||
52a90c29 | 6452 | static struct vlan_splinter * |
4e022ec0 | 6453 | vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) |
52a90c29 BP |
6454 | { |
6455 | struct vlan_splinter *vsp; | |
6456 | ||
4e022ec0 | 6457 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, |
f9c0c3ec | 6458 | hash_ofp_port(vlandev_ofp_port), |
52a90c29 BP |
6459 | &ofproto->vlandev_map) { |
6460 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6461 | return vsp; | |
6462 | } | |
6463 | } | |
6464 | ||
6465 | return NULL; | |
6466 | } | |
6467 | ||
40e05935 BP |
6468 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
6469 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
6470 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
6471 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
6472 | * eth0 and store 9 in '*vid'. | |
6473 | * | |
6474 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
6475 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
6476 | * always does.*/ | |
4e022ec0 | 6477 | static ofp_port_t |
52a90c29 | 6478 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, |
4e022ec0 | 6479 | ofp_port_t vlandev_ofp_port, int *vid) |
bd3950dd | 6480 | OVS_REQUIRES(ofproto->vsp_mutex) |
52a90c29 BP |
6481 | { |
6482 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6483 | const struct vlan_splinter *vsp; | |
6484 | ||
6485 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6486 | if (vsp) { | |
6487 | if (vid) { | |
6488 | *vid = vsp->vid; | |
6489 | } | |
6490 | return vsp->realdev_ofp_port; | |
6491 | } | |
6492 | } | |
6493 | return 0; | |
6494 | } | |
6495 | ||
b98d8985 BP |
6496 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
6497 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
6498 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
6499 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
6500 | * always the case unless VLAN splinters are enabled), returns false without | |
6501 | * making any changes. */ | |
8449c4d6 | 6502 | bool |
b98d8985 | 6503 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) |
13501305 | 6504 | OVS_EXCLUDED(ofproto->vsp_mutex) |
b98d8985 | 6505 | { |
4e022ec0 | 6506 | ofp_port_t realdev; |
b98d8985 BP |
6507 | int vid; |
6508 | ||
13501305 | 6509 | ovs_mutex_lock(&ofproto->vsp_mutex); |
4e022ec0 | 6510 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid); |
13501305 | 6511 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
b98d8985 BP |
6512 | if (!realdev) { |
6513 | return false; | |
6514 | } | |
6515 | ||
6516 | /* Cause the flow to be processed as if it came in on the real device with | |
6517 | * the VLAN device's VLAN ID. */ | |
4e022ec0 | 6518 | flow->in_port.ofp_port = realdev; |
b98d8985 BP |
6519 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); |
6520 | return true; | |
6521 | } | |
6522 | ||
52a90c29 BP |
6523 | static void |
6524 | vsp_remove(struct ofport_dpif *port) | |
6525 | { | |
6526 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6527 | struct vlan_splinter *vsp; | |
6528 | ||
13501305 | 6529 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 BP |
6530 | vsp = vlandev_find(ofproto, port->up.ofp_port); |
6531 | if (vsp) { | |
6532 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6533 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6534 | free(vsp); | |
6535 | ||
6536 | port->realdev_ofp_port = 0; | |
6537 | } else { | |
6538 | VLOG_ERR("missing vlan device record"); | |
6539 | } | |
13501305 | 6540 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 BP |
6541 | } |
6542 | ||
6543 | static void | |
4e022ec0 | 6544 | vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6545 | { |
6546 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6547 | ||
13501305 | 6548 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 | 6549 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) |
13501305 | 6550 | && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid)) |
52a90c29 BP |
6551 | == realdev_ofp_port)) { |
6552 | struct vlan_splinter *vsp; | |
6553 | ||
6554 | vsp = xmalloc(sizeof *vsp); | |
52a90c29 BP |
6555 | vsp->realdev_ofp_port = realdev_ofp_port; |
6556 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6557 | vsp->vid = vid; | |
6558 | ||
6559 | port->realdev_ofp_port = realdev_ofp_port; | |
13501305 EJ |
6560 | |
6561 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6562 | hash_ofp_port(port->up.ofp_port)); | |
6563 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6564 | hash_realdev_vid(realdev_ofp_port, vid)); | |
52a90c29 BP |
6565 | } else { |
6566 | VLOG_ERR("duplicate vlan device record"); | |
6567 | } | |
13501305 | 6568 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 | 6569 | } |
e1b1d06a | 6570 | |
46c88433 | 6571 | static odp_port_t |
4e022ec0 | 6572 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
e1b1d06a JP |
6573 | { |
6574 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
4e022ec0 | 6575 | return ofport ? ofport->odp_port : ODPP_NONE; |
e1b1d06a JP |
6576 | } |
6577 | ||
8449c4d6 | 6578 | struct ofport_dpif * |
4e022ec0 | 6579 | odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port) |
e1b1d06a JP |
6580 | { |
6581 | struct ofport_dpif *port; | |
6582 | ||
8449c4d6 | 6583 | ovs_rwlock_rdlock(&backer->odp_to_ofport_lock); |
f9c0c3ec | 6584 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port), |
acf60855 | 6585 | &backer->odp_to_ofport_map) { |
e1b1d06a | 6586 | if (port->odp_port == odp_port) { |
8449c4d6 | 6587 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 | 6588 | return port; |
e1b1d06a JP |
6589 | } |
6590 | } | |
6591 | ||
8449c4d6 | 6592 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 JP |
6593 | return NULL; |
6594 | } | |
6595 | ||
4e022ec0 AW |
6596 | static ofp_port_t |
6597 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) | |
acf60855 JP |
6598 | { |
6599 | struct ofport_dpif *port; | |
6600 | ||
6601 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 6602 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
6603 | return port->up.ofp_port; |
6604 | } else { | |
6605 | return OFPP_NONE; | |
6606 | } | |
e1b1d06a | 6607 | } |
655ab909 | 6608 | |
655ab909 AZ |
6609 | /* Compute exponentially weighted moving average, adding 'new' as the newest, |
6610 | * most heavily weighted element. 'base' designates the rate of decay: after | |
6611 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
6612 | * (about .37). */ | |
6613 | static void | |
6614 | exp_mavg(double *avg, int base, double new) | |
6615 | { | |
6616 | *avg = (*avg * (base - 1) + new) / base; | |
6617 | } | |
6618 | ||
6619 | static void | |
dc54ef36 | 6620 | update_moving_averages(struct dpif_backer *backer) |
655ab909 AZ |
6621 | { |
6622 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
dc54ef36 EJ |
6623 | long long int minutes = (time_msec() - backer->created) / min_ms; |
6624 | ||
6625 | if (minutes > 0) { | |
6626 | backer->lifetime.add_rate = (double) backer->total_subfacet_add_count | |
6627 | / minutes; | |
6628 | backer->lifetime.del_rate = (double) backer->total_subfacet_del_count | |
6629 | / minutes; | |
6630 | } else { | |
6631 | backer->lifetime.add_rate = 0.0; | |
6632 | backer->lifetime.del_rate = 0.0; | |
6633 | } | |
655ab909 AZ |
6634 | |
6635 | /* Update hourly averages on the minute boundaries. */ | |
dc54ef36 EJ |
6636 | if (time_msec() - backer->last_minute >= min_ms) { |
6637 | exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count); | |
6638 | exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count); | |
655ab909 AZ |
6639 | |
6640 | /* Update daily averages on the hour boundaries. */ | |
dc54ef36 EJ |
6641 | if ((backer->last_minute - backer->created) / min_ms % 60 == 59) { |
6642 | exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate); | |
6643 | exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate); | |
655ab909 AZ |
6644 | } |
6645 | ||
dc54ef36 EJ |
6646 | backer->total_subfacet_add_count += backer->subfacet_add_count; |
6647 | backer->total_subfacet_del_count += backer->subfacet_del_count; | |
6648 | backer->subfacet_add_count = 0; | |
6649 | backer->subfacet_del_count = 0; | |
6650 | backer->last_minute += min_ms; | |
655ab909 AZ |
6651 | } |
6652 | } | |
e1b1d06a | 6653 | |
abe529af | 6654 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 6655 | init, |
abe529af BP |
6656 | enumerate_types, |
6657 | enumerate_names, | |
6658 | del, | |
0aeaabc8 | 6659 | port_open_type, |
acf60855 JP |
6660 | type_run, |
6661 | type_run_fast, | |
6662 | type_wait, | |
abe529af BP |
6663 | alloc, |
6664 | construct, | |
6665 | destruct, | |
6666 | dealloc, | |
6667 | run, | |
5fcc0d00 | 6668 | run_fast, |
abe529af | 6669 | wait, |
0d085684 | 6670 | get_memory_usage, |
abe529af | 6671 | flush, |
6c1491fb BP |
6672 | get_features, |
6673 | get_tables, | |
abe529af BP |
6674 | port_alloc, |
6675 | port_construct, | |
6676 | port_destruct, | |
6677 | port_dealloc, | |
6678 | port_modified, | |
6679 | port_reconfigured, | |
6680 | port_query_by_name, | |
6681 | port_add, | |
6682 | port_del, | |
6527c598 | 6683 | port_get_stats, |
abe529af BP |
6684 | port_dump_start, |
6685 | port_dump_next, | |
6686 | port_dump_done, | |
6687 | port_poll, | |
6688 | port_poll_wait, | |
6689 | port_is_lacp_current, | |
0ab6decf | 6690 | NULL, /* rule_choose_table */ |
abe529af BP |
6691 | rule_alloc, |
6692 | rule_construct, | |
6693 | rule_destruct, | |
6694 | rule_dealloc, | |
abe529af BP |
6695 | rule_get_stats, |
6696 | rule_execute, | |
6697 | rule_modify_actions, | |
7257b535 | 6698 | set_frag_handling, |
abe529af BP |
6699 | packet_out, |
6700 | set_netflow, | |
6701 | get_netflow_ids, | |
6702 | set_sflow, | |
29089a54 | 6703 | set_ipfix, |
abe529af | 6704 | set_cfm, |
9a9e3786 | 6705 | get_cfm_status, |
ccc09689 EJ |
6706 | set_bfd, |
6707 | get_bfd_status, | |
21f7563c JP |
6708 | set_stp, |
6709 | get_stp_status, | |
6710 | set_stp_port, | |
6711 | get_stp_port_status, | |
8b36f51e | 6712 | set_queues, |
abe529af BP |
6713 | bundle_set, |
6714 | bundle_remove, | |
ec7ceaed EJ |
6715 | mirror_set__, |
6716 | mirror_get_stats__, | |
abe529af BP |
6717 | set_flood_vlans, |
6718 | is_mirror_output_bundle, | |
8402c74b | 6719 | forward_bpdu_changed, |
c4069512 | 6720 | set_mac_table_config, |
52a90c29 | 6721 | set_realdev, |
9cae45dc JR |
6722 | NULL, /* meter_get_features */ |
6723 | NULL, /* meter_set */ | |
6724 | NULL, /* meter_get */ | |
6725 | NULL, /* meter_del */ | |
abe529af | 6726 | }; |