]>
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" | |
05067881 | 34 | #include "guarded-list.h" |
abe529af BP |
35 | #include "hmapx.h" |
36 | #include "lacp.h" | |
75a75043 | 37 | #include "learn.h" |
abe529af | 38 | #include "mac-learning.h" |
816fd533 | 39 | #include "meta-flow.h" |
abe529af | 40 | #include "multipath.h" |
0a740f48 | 41 | #include "netdev-vport.h" |
abe529af BP |
42 | #include "netdev.h" |
43 | #include "netlink.h" | |
44 | #include "nx-match.h" | |
45 | #include "odp-util.h" | |
1ac7c9bd | 46 | #include "odp-execute.h" |
abe529af BP |
47 | #include "ofp-util.h" |
48 | #include "ofpbuf.h" | |
f25d0cf3 | 49 | #include "ofp-actions.h" |
31a19d69 | 50 | #include "ofp-parse.h" |
abe529af | 51 | #include "ofp-print.h" |
9d6ac44e | 52 | #include "ofproto-dpif-governor.h" |
29089a54 | 53 | #include "ofproto-dpif-ipfix.h" |
ec7ceaed | 54 | #include "ofproto-dpif-mirror.h" |
635c5db9 | 55 | #include "ofproto-dpif-monitor.h" |
bae473fe | 56 | #include "ofproto-dpif-sflow.h" |
e1ec7dd4 | 57 | #include "ofproto-dpif-upcall.h" |
9583bc14 | 58 | #include "ofproto-dpif-xlate.h" |
abe529af | 59 | #include "poll-loop.h" |
0d085684 | 60 | #include "simap.h" |
27022416 | 61 | #include "smap.h" |
abe529af | 62 | #include "timer.h" |
b9ad7294 | 63 | #include "tunnel.h" |
6c1491fb | 64 | #include "unaligned.h" |
abe529af BP |
65 | #include "unixctl.h" |
66 | #include "vlan-bitmap.h" | |
67 | #include "vlog.h" | |
68 | ||
69 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
70 | ||
abe529af | 71 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
72 | COVERAGE_DEFINE(facet_revalidate); |
73 | COVERAGE_DEFINE(facet_unexpected); | |
264ac358 AW |
74 | COVERAGE_DEFINE(facet_create); |
75 | COVERAGE_DEFINE(facet_remove); | |
76 | COVERAGE_DEFINE(subfacet_create); | |
77 | COVERAGE_DEFINE(subfacet_destroy); | |
9b1a48c2 | 78 | COVERAGE_DEFINE(subfacet_install_fail); |
ada3a58d | 79 | COVERAGE_DEFINE(packet_in_overflow); |
abe529af | 80 | |
46c88433 EJ |
81 | /* Number of implemented OpenFlow tables. */ |
82 | enum { N_TABLES = 255 }; | |
83 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
84 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
85 | ||
a088a1ff | 86 | struct flow_miss; |
ac35f9c8 | 87 | struct facet; |
abe529af | 88 | |
70742c7f EJ |
89 | struct rule_dpif { |
90 | struct rule up; | |
91 | ||
92 | /* These statistics: | |
93 | * | |
94 | * - Do include packets and bytes from facets that have been deleted or | |
95 | * whose own statistics have been folded into the rule. | |
96 | * | |
97 | * - Do include packets and bytes sent "by hand" that were accounted to | |
98 | * the rule without any facet being involved (this is a rare corner | |
99 | * case in rule_execute()). | |
100 | * | |
101 | * - Do not include packet or bytes that can be obtained from any facet's | |
102 | * packet_count or byte_count member or that can be obtained from the | |
103 | * datapath by, e.g., dpif_flow_get() for any subfacet. | |
104 | */ | |
105 | struct ovs_mutex stats_mutex; | |
106 | uint64_t packet_count OVS_GUARDED; /* Number of packets received. */ | |
107 | uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */ | |
108 | }; | |
109 | ||
7e741ffb | 110 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
70742c7f | 111 | static struct rule_dpif *rule_dpif_cast(const struct rule *); |
b0f7b9b5 | 112 | |
46c88433 EJ |
113 | struct ofbundle { |
114 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
115 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
116 | void *aux; /* Key supplied by ofproto's client. */ | |
117 | char *name; /* Identifier for log messages. */ | |
118 | ||
119 | /* Configuration. */ | |
120 | struct list ports; /* Contains "struct ofport"s. */ | |
121 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
122 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
123 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
124 | * NULL if all VLANs are trunked. */ | |
125 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
126 | struct bond *bond; /* Nonnull iff more than one port. */ | |
127 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
128 | ||
129 | /* Status. */ | |
130 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
131 | }; | |
132 | ||
abe529af | 133 | static void bundle_remove(struct ofport *); |
7bde8dd8 | 134 | static void bundle_update(struct ofbundle *); |
abe529af BP |
135 | static void bundle_destroy(struct ofbundle *); |
136 | static void bundle_del_port(struct ofport_dpif *); | |
137 | static void bundle_run(struct ofbundle *); | |
138 | static void bundle_wait(struct ofbundle *); | |
33158a18 | 139 | |
21f7563c JP |
140 | static void stp_run(struct ofproto_dpif *ofproto); |
141 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
142 | static int set_stp_port(struct ofport *, |
143 | const struct ofproto_port_stp_settings *); | |
21f7563c | 144 | |
6a7e895f BP |
145 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, |
146 | enum slow_path_reason, | |
147 | uint64_t *stub, size_t stub_size, | |
148 | const struct nlattr **actionsp, | |
149 | size_t *actions_lenp); | |
150 | ||
151 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
152 | * states: | |
153 | * | |
154 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
155 | * case just after the subfacet is created, just before the subfacet is | |
156 | * destroyed, or if the datapath returns an error when we try to install a | |
157 | * subfacet. | |
158 | * | |
159 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
160 | * | |
161 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
162 | * ofproto_dpif is installed in the datapath. | |
163 | */ | |
164 | enum subfacet_path { | |
165 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
166 | SF_FAST_PATH, /* Full actions are installed. */ | |
167 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
168 | }; | |
169 | ||
5f5fbd17 BP |
170 | /* A dpif flow and actions associated with a facet. |
171 | * | |
172 | * See also the large comment on struct facet. */ | |
173 | struct subfacet { | |
174 | /* Owners. */ | |
175 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
176 | struct list list_node; /* In struct facet's 'facets' list. */ | |
177 | struct facet *facet; /* Owning facet. */ | |
04d08d54 | 178 | struct dpif_backer *backer; /* Owning backer. */ |
5f5fbd17 | 179 | |
5f5fbd17 BP |
180 | struct nlattr *key; |
181 | int key_len; | |
182 | ||
183 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 184 | long long int created; /* Time created. */ |
5f5fbd17 BP |
185 | |
186 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
187 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
188 | ||
6a7e895f | 189 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 BP |
190 | }; |
191 | ||
1d85f9e5 JP |
192 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
193 | ||
e1ec7dd4 | 194 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *); |
04d08d54 | 195 | static struct subfacet *subfacet_find(struct dpif_backer *, |
acf60855 | 196 | const struct nlattr *key, size_t key_len, |
9566abf9 | 197 | uint32_t key_hash); |
5f5fbd17 BP |
198 | static void subfacet_destroy(struct subfacet *); |
199 | static void subfacet_destroy__(struct subfacet *); | |
04d08d54 | 200 | static void subfacet_destroy_batch(struct dpif_backer *, |
1d85f9e5 | 201 | struct subfacet **, int n); |
5f5fbd17 BP |
202 | static void subfacet_reset_dp_stats(struct subfacet *, |
203 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
204 | static void subfacet_update_stats(struct subfacet *, |
205 | const struct dpif_flow_stats *); | |
5f5fbd17 | 206 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
207 | const struct ofpbuf *odp_actions, |
208 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
209 | static void subfacet_uninstall(struct subfacet *); |
210 | ||
bcd2633a | 211 | /* A unique, non-overlapping instantiation of an OpenFlow flow. |
b0f7b9b5 BP |
212 | * |
213 | * A facet associates a "struct flow", which represents the Open vSwitch | |
bcd2633a JP |
214 | * userspace idea of an exact-match flow, with one or more subfacets. |
215 | * While the facet is created based on an exact-match flow, it is stored | |
216 | * within the ofproto based on the wildcards that could be expressed | |
217 | * based on the flow table and other configuration. (See the 'wc' | |
218 | * description in "struct xlate_out" for more details.) | |
b0f7b9b5 | 219 | * |
bcd2633a JP |
220 | * Each subfacet tracks the datapath's idea of the flow equivalent to |
221 | * the facet. When the kernel module (or other dpif implementation) and | |
222 | * Open vSwitch userspace agree on the definition of a flow key, there | |
223 | * is exactly one subfacet per facet. If the dpif implementation | |
224 | * supports more-specific flow matching than userspace, however, a facet | |
225 | * can have more than one subfacet. Examples include the dpif | |
226 | * implementation not supporting the same wildcards as userspace or some | |
227 | * distinction in flow that userspace simply doesn't understand. | |
228 | * | |
229 | * Flow expiration works in terms of subfacets, so a facet must have at | |
230 | * least one subfacet or it will never expire, leaking memory. */ | |
abe529af | 231 | struct facet { |
74432d57 | 232 | /* Owner. */ |
b7e2f6b3 | 233 | struct ofproto_dpif *ofproto; |
b0f7b9b5 BP |
234 | |
235 | /* Owned data. */ | |
236 | struct list subfacets; | |
abe529af BP |
237 | long long int used; /* Time last used; time created if not used. */ |
238 | ||
b0f7b9b5 | 239 | /* Key. */ |
bcd2633a JP |
240 | struct flow flow; /* Flow of the creating subfacet. */ |
241 | struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */ | |
b0f7b9b5 | 242 | |
abe529af BP |
243 | /* These statistics: |
244 | * | |
245 | * - Do include packets and bytes sent "by hand", e.g. with | |
246 | * dpif_execute(). | |
247 | * | |
248 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 249 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 250 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
251 | * |
252 | * - Do not include packets or bytes that can be obtained from the | |
253 | * datapath for any existing subfacet. | |
abe529af BP |
254 | */ |
255 | uint64_t packet_count; /* Number of packets received. */ | |
256 | uint64_t byte_count; /* Number of bytes received. */ | |
257 | ||
b0f7b9b5 | 258 | /* Resubmit statistics. */ |
9d24de3b JP |
259 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
260 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
261 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 262 | |
b0f7b9b5 | 263 | /* Accounting. */ |
907a4c5e | 264 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 265 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 266 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 267 | |
bbafd73b | 268 | struct xlate_out xout; |
4dff9097 | 269 | |
26cd7e34 BP |
270 | /* Storage for a single subfacet, to reduce malloc() time and space |
271 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
272 | * case has exactly one subfacet. However, 'one_subfacet' may not |
273 | * always be valid, since it could have been removed after newer | |
274 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 275 | struct subfacet one_subfacet; |
6cf474d7 EJ |
276 | |
277 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
278 | }; |
279 | ||
e1ec7dd4 | 280 | static struct facet *facet_create(const struct flow_miss *); |
15baa734 | 281 | static void facet_remove(struct facet *); |
abe529af BP |
282 | static void facet_free(struct facet *); |
283 | ||
bcd2633a | 284 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); |
abe529af | 285 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
bcd2633a | 286 | const struct flow *); |
5bf64ade | 287 | static bool facet_revalidate(struct facet *); |
6814e51f | 288 | static bool facet_check_consistency(struct facet *); |
abe529af | 289 | |
15baa734 | 290 | static void facet_flush_stats(struct facet *); |
abe529af | 291 | |
bbb5d219 | 292 | static void facet_reset_counters(struct facet *); |
e1ec7dd4 EJ |
293 | static void flow_push_stats(struct ofproto_dpif *, struct flow *, |
294 | struct dpif_flow_stats *, bool may_learn); | |
9dfb1f78 | 295 | static void facet_push_stats(struct facet *, bool may_learn); |
3de9590b BP |
296 | static void facet_learn(struct facet *); |
297 | static void facet_account(struct facet *); | |
8844e035 | 298 | static void push_all_stats(void); |
abe529af BP |
299 | |
300 | static bool facet_is_controller_flow(struct facet *); | |
301 | ||
46c88433 EJ |
302 | struct ofport_dpif { |
303 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ | |
304 | struct ofport up; | |
305 | ||
306 | odp_port_t odp_port; | |
307 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
308 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
309 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
310 | struct bfd *bfd; /* BFD, if any. */ | |
46c88433 EJ |
311 | bool may_enable; /* May be enabled in bonds. */ |
312 | bool is_tunnel; /* This port is a tunnel. */ | |
313 | long long int carrier_seq; /* Carrier status changes. */ | |
314 | struct ofport_dpif *peer; /* Peer if patch port. */ | |
315 | ||
316 | /* Spanning tree. */ | |
317 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
318 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
319 | long long int stp_state_entered; | |
320 | ||
55954f6e EJ |
321 | /* Queue to DSCP mapping. */ |
322 | struct ofproto_port_queue *qdscp; | |
323 | size_t n_qdscp; | |
46c88433 EJ |
324 | |
325 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
326 | * | |
327 | * This is deprecated. It is only for compatibility with broken device | |
328 | * drivers in old versions of Linux that do not properly support VLANs when | |
329 | * VLAN devices are not used. When broken device drivers are no longer in | |
330 | * widespread use, we will delete these interfaces. */ | |
331 | ofp_port_t realdev_ofp_port; | |
332 | int vlandev_vid; | |
333 | }; | |
334 | ||
52a90c29 BP |
335 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
336 | * | |
337 | * This is deprecated. It is only for compatibility with broken device drivers | |
338 | * in old versions of Linux that do not properly support VLANs when VLAN | |
339 | * devices are not used. When broken device drivers are no longer in | |
340 | * widespread use, we will delete these interfaces. */ | |
341 | struct vlan_splinter { | |
342 | struct hmap_node realdev_vid_node; | |
343 | struct hmap_node vlandev_node; | |
4e022ec0 AW |
344 | ofp_port_t realdev_ofp_port; |
345 | ofp_port_t vlandev_ofp_port; | |
52a90c29 BP |
346 | int vid; |
347 | }; | |
348 | ||
52a90c29 | 349 | static void vsp_remove(struct ofport_dpif *); |
4e022ec0 | 350 | static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid); |
52a90c29 | 351 | |
46c88433 EJ |
352 | static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
353 | ofp_port_t); | |
354 | ||
4e022ec0 | 355 | static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *, |
46c88433 | 356 | odp_port_t); |
e1b1d06a | 357 | |
abe529af BP |
358 | static struct ofport_dpif * |
359 | ofport_dpif_cast(const struct ofport *ofport) | |
360 | { | |
abe529af BP |
361 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
362 | } | |
363 | ||
364 | static void port_run(struct ofport_dpif *); | |
8aee94b6 | 365 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 366 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
6cbbf4fa | 367 | static void ofport_update_peer(struct ofport_dpif *); |
8fa4d1d0 | 368 | static void run_fast_rl(void); |
3d9c5e58 | 369 | static int run_fast(struct ofproto *); |
abe529af | 370 | |
7ee20df1 BP |
371 | struct dpif_completion { |
372 | struct list list_node; | |
373 | struct ofoperation *op; | |
374 | }; | |
375 | ||
cdc3ab65 EJ |
376 | /* Reasons that we might need to revalidate every facet, and corresponding |
377 | * coverage counters. | |
378 | * | |
379 | * A value of 0 means that there is no need to revalidate. | |
380 | * | |
381 | * It would be nice to have some cleaner way to integrate with coverage | |
382 | * counters, but with only a few reasons I guess this is good enough for | |
383 | * now. */ | |
384 | enum revalidate_reason { | |
385 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
386 | REV_STP, /* Spanning tree protocol port status change. */ | |
4a1b8f30 | 387 | REV_BOND, /* Bonding changed. */ |
cdc3ab65 EJ |
388 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ |
389 | REV_FLOW_TABLE, /* Flow table changed. */ | |
30618594 | 390 | REV_MAC_LEARNING, /* Mac learning changed. */ |
cdc3ab65 EJ |
391 | REV_INCONSISTENCY /* Facet self-check failed. */ |
392 | }; | |
3c4a309c BP |
393 | COVERAGE_DEFINE(rev_reconfigure); |
394 | COVERAGE_DEFINE(rev_stp); | |
4a1b8f30 | 395 | COVERAGE_DEFINE(rev_bond); |
3c4a309c BP |
396 | COVERAGE_DEFINE(rev_port_toggled); |
397 | COVERAGE_DEFINE(rev_flow_table); | |
30618594 | 398 | COVERAGE_DEFINE(rev_mac_learning); |
3c4a309c BP |
399 | COVERAGE_DEFINE(rev_inconsistency); |
400 | ||
cdc3ab65 EJ |
401 | struct avg_subfacet_rates { |
402 | double add_rate; /* Moving average of new flows created per minute. */ | |
403 | double del_rate; /* Moving average of flows deleted per minute. */ | |
404 | }; | |
405 | ||
406 | /* All datapaths of a given type share a single dpif backer instance. */ | |
407 | struct dpif_backer { | |
408 | char *type; | |
409 | int refcount; | |
410 | struct dpif *dpif; | |
e1ec7dd4 | 411 | struct udpif *udpif; |
cdc3ab65 | 412 | struct timer next_expiration; |
8449c4d6 EJ |
413 | |
414 | struct ovs_rwlock odp_to_ofport_lock; | |
415 | struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */ | |
cdc3ab65 EJ |
416 | |
417 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ | |
418 | ||
419 | /* Facet revalidation flags applying to facets which use this backer. */ | |
420 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
cdc3ab65 EJ |
421 | |
422 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
423 | bool recv_set_enable; /* Enables or disables receiving packets. */ | |
424 | ||
425 | struct hmap subfacets; | |
426 | struct governor *governor; | |
427 | ||
428 | /* Subfacet statistics. | |
429 | * | |
430 | * These keep track of the total number of subfacets added and deleted and | |
431 | * flow life span. They are useful for computing the flow rates stats | |
432 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
433 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
434 | * performance in new situations. */ | |
435 | long long int created; /* Time when it is created. */ | |
436 | unsigned max_n_subfacet; /* Maximum number of flows */ | |
437 | unsigned avg_n_subfacet; /* Average number of flows. */ | |
438 | long long int avg_subfacet_life; /* Average life span of subfacets. */ | |
439 | ||
448a4b2f AW |
440 | /* Number of upcall handling threads. */ |
441 | unsigned int n_handler_threads; | |
cdc3ab65 EJ |
442 | }; |
443 | ||
acf60855 JP |
444 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ |
445 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
446 | ||
8f73d537 | 447 | static void drop_key_clear(struct dpif_backer *); |
735d7efb | 448 | |
46c88433 EJ |
449 | struct ofproto_dpif { |
450 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
451 | struct ofproto up; | |
452 | struct dpif_backer *backer; | |
453 | ||
454 | /* Special OpenFlow rules. */ | |
455 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
456 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
457 | struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ | |
458 | ||
459 | /* Bridging. */ | |
460 | struct netflow *netflow; | |
461 | struct dpif_sflow *sflow; | |
462 | struct dpif_ipfix *ipfix; | |
463 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
464 | struct mac_learning *ml; | |
465 | bool has_bonded_bundles; | |
466 | struct mbridge *mbridge; | |
467 | ||
468 | /* Facets. */ | |
469 | struct classifier facets; /* Contains 'struct facet's. */ | |
470 | long long int consistency_rl; | |
471 | ||
0da3c61b AW |
472 | struct ovs_mutex stats_mutex; |
473 | struct netdev_stats stats OVS_GUARDED; /* To account packets generated and | |
474 | * consumed in userspace. */ | |
46c88433 EJ |
475 | |
476 | /* Spanning tree. */ | |
477 | struct stp *stp; | |
478 | long long int stp_last_tick; | |
479 | ||
480 | /* VLAN splinters. */ | |
13501305 EJ |
481 | struct ovs_mutex vsp_mutex; |
482 | struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */ | |
483 | struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */ | |
46c88433 EJ |
484 | |
485 | /* Ports. */ | |
486 | struct sset ports; /* Set of standard port names. */ | |
487 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
488 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ | |
489 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
490 | ||
491 | /* Per ofproto's dpif stats. */ | |
492 | uint64_t n_hit; | |
493 | uint64_t n_missed; | |
3d9c5e58 EJ |
494 | |
495 | /* Work queues. */ | |
05067881 | 496 | struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */ |
46c88433 EJ |
497 | }; |
498 | ||
6b35e630 JP |
499 | /* By default, flows in the datapath are wildcarded (megaflows). They |
500 | * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ | |
501 | static bool enable_megaflows = true; | |
502 | ||
b44a10b7 BP |
503 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
504 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
505 | ||
abe529af BP |
506 | static void ofproto_dpif_unixctl_init(void); |
507 | ||
46c88433 EJ |
508 | static inline struct ofproto_dpif * |
509 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
510 | { | |
511 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
512 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
513 | } | |
514 | ||
515 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, | |
516 | ofp_port_t ofp_port); | |
ade6ad9c EJ |
517 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
518 | const struct ofpbuf *packet, struct ds *); | |
46c88433 | 519 | |
501f8d1f | 520 | /* Upcalls. */ |
e1ec7dd4 | 521 | static void handle_upcalls(struct dpif_backer *); |
abe529af BP |
522 | |
523 | /* Flow expiration. */ | |
acf60855 | 524 | static int expire(struct dpif_backer *); |
abe529af | 525 | |
6fca1ffb BP |
526 | /* NetFlow. */ |
527 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
528 | ||
abe529af BP |
529 | /* Global variables. */ |
530 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
531 | |
532 | /* Initial mappings of port to bridge mappings. */ | |
533 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
46c88433 | 534 | |
1b63b91e BP |
535 | /* Executes 'fm'. The caller retains ownership of 'fm' and everything in |
536 | * it. */ | |
3d9c5e58 | 537 | void |
46c88433 EJ |
538 | ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, |
539 | struct ofputil_flow_mod *fm) | |
540 | { | |
1b63b91e | 541 | ofproto_flow_mod(&ofproto->up, fm); |
46c88433 EJ |
542 | } |
543 | ||
d5ff77e2 YT |
544 | /* Appends 'pin' to the queue of "packet ins" to be sent to the controller. |
545 | * Takes ownership of 'pin' and pin->packet. */ | |
46c88433 EJ |
546 | void |
547 | ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, | |
548 | struct ofputil_packet_in *pin) | |
549 | { | |
05067881 | 550 | if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) { |
ada3a58d EJ |
551 | COVERAGE_INC(packet_in_overflow); |
552 | free(CONST_CAST(void *, pin->packet)); | |
553 | free(pin); | |
ada3a58d | 554 | } |
46c88433 | 555 | } |
abe529af BP |
556 | \f |
557 | /* Factory functions. */ | |
558 | ||
b0408fca | 559 | static void |
acf60855 | 560 | init(const struct shash *iface_hints) |
b0408fca | 561 | { |
acf60855 JP |
562 | struct shash_node *node; |
563 | ||
564 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
565 | SHASH_FOR_EACH(node, iface_hints) { | |
566 | const struct iface_hint *orig_hint = node->data; | |
567 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
568 | ||
569 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
570 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
571 | new_hint->ofp_port = orig_hint->ofp_port; | |
572 | ||
573 | shash_add(&init_ofp_ports, node->name, new_hint); | |
574 | } | |
b0408fca JP |
575 | } |
576 | ||
abe529af BP |
577 | static void |
578 | enumerate_types(struct sset *types) | |
579 | { | |
580 | dp_enumerate_types(types); | |
581 | } | |
582 | ||
583 | static int | |
584 | enumerate_names(const char *type, struct sset *names) | |
585 | { | |
acf60855 JP |
586 | struct ofproto_dpif *ofproto; |
587 | ||
588 | sset_clear(names); | |
589 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
590 | if (strcmp(type, ofproto->up.type)) { | |
591 | continue; | |
592 | } | |
593 | sset_add(names, ofproto->up.name); | |
594 | } | |
595 | ||
596 | return 0; | |
abe529af BP |
597 | } |
598 | ||
599 | static int | |
600 | del(const char *type, const char *name) | |
601 | { | |
602 | struct dpif *dpif; | |
603 | int error; | |
604 | ||
605 | error = dpif_open(name, type, &dpif); | |
606 | if (!error) { | |
607 | error = dpif_delete(dpif); | |
608 | dpif_close(dpif); | |
609 | } | |
610 | return error; | |
611 | } | |
612 | \f | |
0aeaabc8 JP |
613 | static const char * |
614 | port_open_type(const char *datapath_type, const char *port_type) | |
615 | { | |
616 | return dpif_port_open_type(datapath_type, port_type); | |
617 | } | |
618 | ||
acf60855 JP |
619 | /* Type functions. */ |
620 | ||
36beb9be BP |
621 | static void process_dpif_port_changes(struct dpif_backer *); |
622 | static void process_dpif_all_ports_changed(struct dpif_backer *); | |
623 | static void process_dpif_port_change(struct dpif_backer *, | |
624 | const char *devname); | |
625 | static void process_dpif_port_error(struct dpif_backer *, int error); | |
626 | ||
476cb42a BP |
627 | static struct ofproto_dpif * |
628 | lookup_ofproto_dpif_by_port_name(const char *name) | |
629 | { | |
630 | struct ofproto_dpif *ofproto; | |
631 | ||
632 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
633 | if (sset_contains(&ofproto->ports, name)) { | |
634 | return ofproto; | |
635 | } | |
636 | } | |
637 | ||
638 | return NULL; | |
639 | } | |
640 | ||
acf60855 JP |
641 | static int |
642 | type_run(const char *type) | |
643 | { | |
e20b5746 | 644 | static long long int push_timer = LLONG_MIN; |
acf60855 | 645 | struct dpif_backer *backer; |
acf60855 JP |
646 | |
647 | backer = shash_find_data(&all_dpif_backers, type); | |
648 | if (!backer) { | |
649 | /* This is not necessarily a problem, since backers are only | |
650 | * created on demand. */ | |
651 | return 0; | |
652 | } | |
653 | ||
654 | dpif_run(backer->dpif); | |
655 | ||
e20b5746 EJ |
656 | /* The most natural place to push facet statistics is when they're pulled |
657 | * from the datapath. However, when there are many flows in the datapath, | |
658 | * this expensive operation can occur so frequently, that it reduces our | |
659 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
660 | * here instead. */ | |
661 | if (time_msec() > push_timer) { | |
662 | push_timer = time_msec() + 2000; | |
663 | push_all_stats(); | |
664 | } | |
665 | ||
40358701 GS |
666 | /* If vswitchd started with other_config:flow_restore_wait set as "true", |
667 | * and the configuration has now changed to "false", enable receiving | |
668 | * packets from the datapath. */ | |
669 | if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { | |
36beb9be BP |
670 | int error; |
671 | ||
40358701 GS |
672 | backer->recv_set_enable = true; |
673 | ||
674 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); | |
675 | if (error) { | |
e1ec7dd4 | 676 | udpif_recv_set(backer->udpif, 0, false); |
40358701 GS |
677 | VLOG_ERR("Failed to enable receiving packets in dpif."); |
678 | return error; | |
679 | } | |
448a4b2f AW |
680 | udpif_recv_set(backer->udpif, n_handler_threads, |
681 | backer->recv_set_enable); | |
40358701 GS |
682 | dpif_flow_flush(backer->dpif); |
683 | backer->need_revalidate = REV_RECONFIGURE; | |
684 | } | |
685 | ||
448a4b2f AW |
686 | /* If the n_handler_threads is reconfigured, call udpif_recv_set() |
687 | * to reset the handler threads. */ | |
688 | if (backer->n_handler_threads != n_handler_threads) { | |
689 | udpif_recv_set(backer->udpif, n_handler_threads, | |
690 | backer->recv_set_enable); | |
691 | backer->n_handler_threads = n_handler_threads; | |
692 | } | |
693 | ||
a1616063 | 694 | if (backer->need_revalidate) { |
2cc3c58e | 695 | struct ofproto_dpif *ofproto; |
a614d823 KM |
696 | struct simap_node *node; |
697 | struct simap tmp_backers; | |
698 | ||
699 | /* Handle tunnel garbage collection. */ | |
700 | simap_init(&tmp_backers); | |
701 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
702 | ||
703 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
704 | struct ofport_dpif *iter; | |
705 | ||
706 | if (backer != ofproto->backer) { | |
707 | continue; | |
708 | } | |
709 | ||
710 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
3aa30359 | 711 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
a614d823 KM |
712 | const char *dp_port; |
713 | ||
42943cde | 714 | if (!iter->is_tunnel) { |
a614d823 KM |
715 | continue; |
716 | } | |
717 | ||
3aa30359 BP |
718 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev, |
719 | namebuf, sizeof namebuf); | |
a614d823 KM |
720 | node = simap_find(&tmp_backers, dp_port); |
721 | if (node) { | |
722 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
723 | simap_delete(&tmp_backers, node); | |
724 | node = simap_find(&backer->tnl_backers, dp_port); | |
725 | } else { | |
726 | node = simap_find(&backer->tnl_backers, dp_port); | |
727 | if (!node) { | |
4e022ec0 | 728 | odp_port_t odp_port = ODPP_NONE; |
a614d823 KM |
729 | |
730 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
731 | &odp_port)) { | |
4e022ec0 AW |
732 | simap_put(&backer->tnl_backers, dp_port, |
733 | odp_to_u32(odp_port)); | |
a614d823 KM |
734 | node = simap_find(&backer->tnl_backers, dp_port); |
735 | } | |
736 | } | |
737 | } | |
738 | ||
4e022ec0 | 739 | iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE; |
42943cde EJ |
740 | if (tnl_port_reconfigure(iter, iter->up.netdev, |
741 | iter->odp_port)) { | |
a614d823 KM |
742 | backer->need_revalidate = REV_RECONFIGURE; |
743 | } | |
744 | } | |
745 | } | |
746 | ||
747 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
4e022ec0 | 748 | dpif_port_del(backer->dpif, u32_to_odp(node->data)); |
a614d823 KM |
749 | } |
750 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
751 | |
752 | switch (backer->need_revalidate) { | |
753 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
754 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
4a1b8f30 | 755 | case REV_BOND: COVERAGE_INC(rev_bond); break; |
2cc3c58e EJ |
756 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; |
757 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
30618594 | 758 | case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break; |
2cc3c58e EJ |
759 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; |
760 | } | |
f728af2e BP |
761 | backer->need_revalidate = 0; |
762 | ||
a1616063 EJ |
763 | /* Clear the drop_keys in case we should now be accepting some |
764 | * formerly dropped flows. */ | |
765 | drop_key_clear(backer); | |
766 | ||
2cc3c58e | 767 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 768 | struct facet *facet, *next; |
a1616063 | 769 | struct ofport_dpif *ofport; |
bcd2633a | 770 | struct cls_cursor cursor; |
a1616063 | 771 | struct ofbundle *bundle; |
2cc3c58e EJ |
772 | |
773 | if (ofproto->backer != backer) { | |
774 | continue; | |
775 | } | |
776 | ||
dc24a00f | 777 | ovs_rwlock_wrlock(&xlate_rwlock); |
89a8a7f0 | 778 | xlate_ofproto_set(ofproto, ofproto->up.name, |
ec89fc6f EJ |
779 | ofproto->backer->dpif, ofproto->miss_rule, |
780 | ofproto->no_packet_in_rule, ofproto->ml, | |
9d189a50 EJ |
781 | ofproto->stp, ofproto->mbridge, |
782 | ofproto->sflow, ofproto->ipfix, | |
783 | ofproto->up.frag_handling, | |
a1616063 EJ |
784 | ofproto->up.forward_bpdu, |
785 | connmgr_has_in_band(ofproto->up.connmgr), | |
9d189a50 | 786 | ofproto->netflow != NULL); |
46c88433 | 787 | |
a1616063 EJ |
788 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
789 | xlate_bundle_set(ofproto, bundle, bundle->name, | |
790 | bundle->vlan_mode, bundle->vlan, | |
791 | bundle->trunks, bundle->use_priority_tags, | |
792 | bundle->bond, bundle->lacp, | |
793 | bundle->floodable); | |
794 | } | |
795 | ||
796 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
9d189a50 EJ |
797 | int stp_port = ofport->stp_port |
798 | ? stp_port_no(ofport->stp_port) | |
92cf817b | 799 | : -1; |
a1616063 EJ |
800 | xlate_ofport_set(ofproto, ofport->bundle, ofport, |
801 | ofport->up.ofp_port, ofport->odp_port, | |
802 | ofport->up.netdev, ofport->cfm, | |
9d189a50 | 803 | ofport->bfd, ofport->peer, stp_port, |
55954f6e | 804 | ofport->qdscp, ofport->n_qdscp, |
9d189a50 EJ |
805 | ofport->up.pp.config, ofport->is_tunnel, |
806 | ofport->may_enable); | |
46c88433 | 807 | } |
dc24a00f | 808 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 809 | |
0b4f2078 EJ |
810 | /* Only ofproto-dpif cares about the facet classifier so we just |
811 | * lock cls_cursor_init() to appease the thread safety analysis. */ | |
812 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
bcd2633a | 813 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
0b4f2078 | 814 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 815 | CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { |
a1616063 EJ |
816 | facet_revalidate(facet); |
817 | run_fast_rl(); | |
2cc3c58e EJ |
818 | } |
819 | } | |
e1ec7dd4 EJ |
820 | |
821 | udpif_revalidate(backer->udpif); | |
2cc3c58e EJ |
822 | } |
823 | ||
40358701 GS |
824 | if (!backer->recv_set_enable) { |
825 | /* Wake up before a max of 1000ms. */ | |
826 | timer_set_duration(&backer->next_expiration, 1000); | |
827 | } else if (timer_expired(&backer->next_expiration)) { | |
acf60855 JP |
828 | int delay = expire(backer); |
829 | timer_set_duration(&backer->next_expiration, delay); | |
830 | } | |
831 | ||
36beb9be | 832 | process_dpif_port_changes(backer); |
acf60855 | 833 | |
04d08d54 EJ |
834 | if (backer->governor) { |
835 | size_t n_subfacets; | |
836 | ||
837 | governor_run(backer->governor); | |
838 | ||
839 | /* If the governor has shrunk to its minimum size and the number of | |
840 | * subfacets has dwindled, then drop the governor entirely. | |
841 | * | |
842 | * For hysteresis, the number of subfacets to drop the governor is | |
843 | * smaller than the number needed to trigger its creation. */ | |
844 | n_subfacets = hmap_count(&backer->subfacets); | |
845 | if (n_subfacets * 4 < flow_eviction_threshold | |
846 | && governor_is_idle(backer->governor)) { | |
847 | governor_destroy(backer->governor); | |
848 | backer->governor = NULL; | |
849 | } | |
850 | } | |
851 | ||
acf60855 JP |
852 | return 0; |
853 | } | |
854 | ||
36beb9be BP |
855 | /* Check for and handle port changes in 'backer''s dpif. */ |
856 | static void | |
857 | process_dpif_port_changes(struct dpif_backer *backer) | |
858 | { | |
859 | for (;;) { | |
860 | char *devname; | |
861 | int error; | |
862 | ||
863 | error = dpif_port_poll(backer->dpif, &devname); | |
864 | switch (error) { | |
865 | case EAGAIN: | |
866 | return; | |
867 | ||
868 | case ENOBUFS: | |
869 | process_dpif_all_ports_changed(backer); | |
870 | break; | |
871 | ||
872 | case 0: | |
873 | process_dpif_port_change(backer, devname); | |
874 | free(devname); | |
875 | break; | |
876 | ||
877 | default: | |
878 | process_dpif_port_error(backer, error); | |
879 | break; | |
880 | } | |
881 | } | |
882 | } | |
883 | ||
884 | static void | |
885 | process_dpif_all_ports_changed(struct dpif_backer *backer) | |
886 | { | |
887 | struct ofproto_dpif *ofproto; | |
888 | struct dpif_port dpif_port; | |
889 | struct dpif_port_dump dump; | |
890 | struct sset devnames; | |
891 | const char *devname; | |
892 | ||
893 | sset_init(&devnames); | |
894 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
895 | if (ofproto->backer == backer) { | |
896 | struct ofport *ofport; | |
897 | ||
898 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
899 | sset_add(&devnames, netdev_get_name(ofport->netdev)); | |
900 | } | |
901 | } | |
902 | } | |
903 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) { | |
904 | sset_add(&devnames, dpif_port.name); | |
905 | } | |
906 | ||
907 | SSET_FOR_EACH (devname, &devnames) { | |
908 | process_dpif_port_change(backer, devname); | |
909 | } | |
910 | sset_destroy(&devnames); | |
911 | } | |
912 | ||
913 | static void | |
914 | process_dpif_port_change(struct dpif_backer *backer, const char *devname) | |
915 | { | |
916 | struct ofproto_dpif *ofproto; | |
917 | struct dpif_port port; | |
918 | ||
919 | /* Don't report on the datapath's device. */ | |
920 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
921 | return; | |
922 | } | |
923 | ||
924 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
925 | &all_ofproto_dpifs) { | |
926 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { | |
927 | return; | |
928 | } | |
929 | } | |
930 | ||
931 | ofproto = lookup_ofproto_dpif_by_port_name(devname); | |
932 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { | |
933 | /* The port was removed. If we know the datapath, | |
934 | * report it through poll_set(). If we don't, it may be | |
935 | * notifying us of a removal we initiated, so ignore it. | |
936 | * If there's a pending ENOBUFS, let it stand, since | |
937 | * everything will be reevaluated. */ | |
938 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
939 | sset_add(&ofproto->port_poll_set, devname); | |
940 | ofproto->port_poll_errno = 0; | |
941 | } | |
942 | } else if (!ofproto) { | |
943 | /* The port was added, but we don't know with which | |
944 | * ofproto we should associate it. Delete it. */ | |
945 | dpif_port_del(backer->dpif, port.port_no); | |
74cc3969 BP |
946 | } else { |
947 | struct ofport_dpif *ofport; | |
948 | ||
949 | ofport = ofport_dpif_cast(shash_find_data( | |
950 | &ofproto->up.port_by_name, devname)); | |
951 | if (ofport | |
952 | && ofport->odp_port != port.port_no | |
953 | && !odp_port_to_ofport(backer, port.port_no)) | |
954 | { | |
955 | /* 'ofport''s datapath port number has changed from | |
956 | * 'ofport->odp_port' to 'port.port_no'. Update our internal data | |
957 | * structures to match. */ | |
8449c4d6 | 958 | ovs_rwlock_wrlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
959 | hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node); |
960 | ofport->odp_port = port.port_no; | |
961 | hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node, | |
962 | hash_odp_port(port.port_no)); | |
8449c4d6 | 963 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
964 | backer->need_revalidate = REV_RECONFIGURE; |
965 | } | |
36beb9be BP |
966 | } |
967 | dpif_port_destroy(&port); | |
968 | } | |
969 | ||
970 | /* Propagate 'error' to all ofprotos based on 'backer'. */ | |
971 | static void | |
972 | process_dpif_port_error(struct dpif_backer *backer, int error) | |
973 | { | |
974 | struct ofproto_dpif *ofproto; | |
975 | ||
976 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
977 | if (ofproto->backer == backer) { | |
978 | sset_clear(&ofproto->port_poll_set); | |
979 | ofproto->port_poll_errno = error; | |
980 | } | |
981 | } | |
982 | } | |
983 | ||
acf60855 | 984 | static int |
e1ec7dd4 | 985 | dpif_backer_run_fast(struct dpif_backer *backer) |
acf60855 | 986 | { |
e1ec7dd4 | 987 | handle_upcalls(backer); |
acf60855 JP |
988 | |
989 | return 0; | |
990 | } | |
991 | ||
8fa4d1d0 EJ |
992 | static int |
993 | type_run_fast(const char *type) | |
994 | { | |
995 | struct dpif_backer *backer; | |
996 | ||
997 | backer = shash_find_data(&all_dpif_backers, type); | |
998 | if (!backer) { | |
999 | /* This is not necessarily a problem, since backers are only | |
1000 | * created on demand. */ | |
1001 | return 0; | |
1002 | } | |
1003 | ||
e1ec7dd4 | 1004 | return dpif_backer_run_fast(backer); |
8fa4d1d0 EJ |
1005 | } |
1006 | ||
1007 | static void | |
1008 | run_fast_rl(void) | |
1009 | { | |
1010 | static long long int port_rl = LLONG_MIN; | |
8fa4d1d0 EJ |
1011 | |
1012 | if (time_msec() >= port_rl) { | |
1013 | struct ofproto_dpif *ofproto; | |
8fa4d1d0 EJ |
1014 | |
1015 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3d9c5e58 | 1016 | run_fast(&ofproto->up); |
8fa4d1d0 EJ |
1017 | } |
1018 | port_rl = time_msec() + 200; | |
1019 | } | |
8fa4d1d0 EJ |
1020 | } |
1021 | ||
acf60855 JP |
1022 | static void |
1023 | type_wait(const char *type) | |
1024 | { | |
1025 | struct dpif_backer *backer; | |
1026 | ||
1027 | backer = shash_find_data(&all_dpif_backers, type); | |
1028 | if (!backer) { | |
1029 | /* This is not necessarily a problem, since backers are only | |
1030 | * created on demand. */ | |
1031 | return; | |
1032 | } | |
1033 | ||
04d08d54 EJ |
1034 | if (backer->governor) { |
1035 | governor_wait(backer->governor); | |
1036 | } | |
1037 | ||
acf60855 | 1038 | timer_wait(&backer->next_expiration); |
c0365fc8 YT |
1039 | dpif_wait(backer->dpif); |
1040 | udpif_wait(backer->udpif); | |
acf60855 JP |
1041 | } |
1042 | \f | |
abe529af BP |
1043 | /* Basic life-cycle. */ |
1044 | ||
c57b2226 BP |
1045 | static int add_internal_flows(struct ofproto_dpif *); |
1046 | ||
abe529af BP |
1047 | static struct ofproto * |
1048 | alloc(void) | |
1049 | { | |
1050 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1051 | return &ofproto->up; | |
1052 | } | |
1053 | ||
1054 | static void | |
1055 | dealloc(struct ofproto *ofproto_) | |
1056 | { | |
1057 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1058 | free(ofproto); | |
1059 | } | |
1060 | ||
acf60855 JP |
1061 | static void |
1062 | close_dpif_backer(struct dpif_backer *backer) | |
1063 | { | |
cb22974d | 1064 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1065 | |
1066 | if (--backer->refcount) { | |
1067 | return; | |
1068 | } | |
1069 | ||
8f73d537 EJ |
1070 | drop_key_clear(backer); |
1071 | hmap_destroy(&backer->drop_keys); | |
1072 | ||
b395cf1f BP |
1073 | udpif_destroy(backer->udpif); |
1074 | ||
7d82ab2e | 1075 | simap_destroy(&backer->tnl_backers); |
8449c4d6 | 1076 | ovs_rwlock_destroy(&backer->odp_to_ofport_lock); |
acf60855 | 1077 | hmap_destroy(&backer->odp_to_ofport_map); |
4f7cc3c7 | 1078 | shash_find_and_delete(&all_dpif_backers, backer->type); |
acf60855 | 1079 | free(backer->type); |
acf60855 JP |
1080 | dpif_close(backer->dpif); |
1081 | ||
04d08d54 EJ |
1082 | ovs_assert(hmap_is_empty(&backer->subfacets)); |
1083 | hmap_destroy(&backer->subfacets); | |
1084 | governor_destroy(backer->governor); | |
1085 | ||
acf60855 JP |
1086 | free(backer); |
1087 | } | |
1088 | ||
1089 | /* Datapath port slated for removal from datapath. */ | |
1090 | struct odp_garbage { | |
1091 | struct list list_node; | |
4e022ec0 | 1092 | odp_port_t odp_port; |
acf60855 JP |
1093 | }; |
1094 | ||
1095 | static int | |
1096 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1097 | { | |
1098 | struct dpif_backer *backer; | |
1099 | struct dpif_port_dump port_dump; | |
1100 | struct dpif_port port; | |
1101 | struct shash_node *node; | |
1102 | struct list garbage_list; | |
1103 | struct odp_garbage *garbage, *next; | |
1104 | struct sset names; | |
1105 | char *backer_name; | |
1106 | const char *name; | |
1107 | int error; | |
1108 | ||
1109 | backer = shash_find_data(&all_dpif_backers, type); | |
1110 | if (backer) { | |
1111 | backer->refcount++; | |
1112 | *backerp = backer; | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | backer_name = xasprintf("ovs-%s", type); | |
1117 | ||
1118 | /* Remove any existing datapaths, since we assume we're the only | |
1119 | * userspace controlling the datapath. */ | |
1120 | sset_init(&names); | |
1121 | dp_enumerate_names(type, &names); | |
1122 | SSET_FOR_EACH(name, &names) { | |
1123 | struct dpif *old_dpif; | |
1124 | ||
1125 | /* Don't remove our backer if it exists. */ | |
1126 | if (!strcmp(name, backer_name)) { | |
1127 | continue; | |
1128 | } | |
1129 | ||
1130 | if (dpif_open(name, type, &old_dpif)) { | |
1131 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1132 | } else { | |
1133 | dpif_delete(old_dpif); | |
1134 | dpif_close(old_dpif); | |
1135 | } | |
1136 | } | |
1137 | sset_destroy(&names); | |
1138 | ||
1139 | backer = xmalloc(sizeof *backer); | |
1140 | ||
1141 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1142 | free(backer_name); | |
1143 | if (error) { | |
1144 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
10a89ef0 | 1145 | ovs_strerror(error)); |
4c1b1289 | 1146 | free(backer); |
acf60855 JP |
1147 | return error; |
1148 | } | |
e1ec7dd4 | 1149 | backer->udpif = udpif_create(backer, backer->dpif); |
acf60855 JP |
1150 | |
1151 | backer->type = xstrdup(type); | |
04d08d54 | 1152 | backer->governor = NULL; |
acf60855 JP |
1153 | backer->refcount = 1; |
1154 | hmap_init(&backer->odp_to_ofport_map); | |
8449c4d6 | 1155 | ovs_rwlock_init(&backer->odp_to_ofport_lock); |
8f73d537 | 1156 | hmap_init(&backer->drop_keys); |
04d08d54 | 1157 | hmap_init(&backer->subfacets); |
acf60855 | 1158 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1159 | backer->need_revalidate = 0; |
7d82ab2e | 1160 | simap_init(&backer->tnl_backers); |
40358701 | 1161 | backer->recv_set_enable = !ofproto_get_flow_restore_wait(); |
acf60855 JP |
1162 | *backerp = backer; |
1163 | ||
40358701 GS |
1164 | if (backer->recv_set_enable) { |
1165 | dpif_flow_flush(backer->dpif); | |
1166 | } | |
acf60855 JP |
1167 | |
1168 | /* Loop through the ports already on the datapath and remove any | |
1169 | * that we don't need anymore. */ | |
1170 | list_init(&garbage_list); | |
1171 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1172 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1173 | node = shash_find(&init_ofp_ports, port.name); | |
1174 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1175 | garbage = xmalloc(sizeof *garbage); | |
1176 | garbage->odp_port = port.port_no; | |
1177 | list_push_front(&garbage_list, &garbage->list_node); | |
1178 | } | |
1179 | } | |
1180 | dpif_port_dump_done(&port_dump); | |
1181 | ||
1182 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1183 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1184 | list_remove(&garbage->list_node); | |
1185 | free(garbage); | |
1186 | } | |
1187 | ||
1188 | shash_add(&all_dpif_backers, type, backer); | |
1189 | ||
40358701 | 1190 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); |
acf60855 JP |
1191 | if (error) { |
1192 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
10a89ef0 | 1193 | type, ovs_strerror(error)); |
acf60855 JP |
1194 | close_dpif_backer(backer); |
1195 | return error; | |
1196 | } | |
448a4b2f AW |
1197 | udpif_recv_set(backer->udpif, n_handler_threads, |
1198 | backer->recv_set_enable); | |
1199 | backer->n_handler_threads = n_handler_threads; | |
acf60855 | 1200 | |
dc54ef36 EJ |
1201 | backer->max_n_subfacet = 0; |
1202 | backer->created = time_msec(); | |
dc54ef36 EJ |
1203 | backer->avg_n_subfacet = 0; |
1204 | backer->avg_subfacet_life = 0; | |
1205 | ||
acf60855 JP |
1206 | return error; |
1207 | } | |
1208 | ||
abe529af | 1209 | static int |
0f5f95a9 | 1210 | construct(struct ofproto *ofproto_) |
abe529af BP |
1211 | { |
1212 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1213 | struct shash_node *node, *next; |
1dd16b9a | 1214 | uint32_t max_ports; |
abe529af | 1215 | int error; |
abe529af | 1216 | |
acf60855 | 1217 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1218 | if (error) { |
abe529af BP |
1219 | return error; |
1220 | } | |
1221 | ||
acf60855 | 1222 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
1dd16b9a | 1223 | ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX))); |
91858960 | 1224 | |
abe529af BP |
1225 | ofproto->netflow = NULL; |
1226 | ofproto->sflow = NULL; | |
29089a54 | 1227 | ofproto->ipfix = NULL; |
21f7563c | 1228 | ofproto->stp = NULL; |
abe529af | 1229 | hmap_init(&ofproto->bundles); |
e764773c | 1230 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
ec7ceaed | 1231 | ofproto->mbridge = mbridge_create(); |
abe529af | 1232 | ofproto->has_bonded_bundles = false; |
0da3c61b | 1233 | ovs_mutex_init(&ofproto->stats_mutex); |
834d6caf | 1234 | ovs_mutex_init(&ofproto->vsp_mutex); |
abe529af | 1235 | |
bcd2633a | 1236 | classifier_init(&ofproto->facets); |
4d2a0f39 | 1237 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 | 1238 | |
05067881 | 1239 | guarded_list_init(&ofproto->pins); |
ada3a58d | 1240 | |
abe529af BP |
1241 | ofproto_dpif_unixctl_init(); |
1242 | ||
52a90c29 BP |
1243 | hmap_init(&ofproto->vlandev_map); |
1244 | hmap_init(&ofproto->realdev_vid_map); | |
1245 | ||
acf60855 | 1246 | sset_init(&ofproto->ports); |
0a740f48 | 1247 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1248 | sset_init(&ofproto->port_poll_set); |
1249 | ofproto->port_poll_errno = 0; | |
1250 | ||
1251 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1252 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1253 | |
1254 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1255 | /* Check if the datapath already has this port. */ | |
1256 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1257 | sset_add(&ofproto->ports, node->name); | |
1258 | } | |
1259 | ||
1260 | free(iface_hint->br_name); | |
1261 | free(iface_hint->br_type); | |
4f9e08a5 | 1262 | free(iface_hint); |
acf60855 JP |
1263 | shash_delete(&init_ofp_ports, node); |
1264 | } | |
1265 | } | |
e1b1d06a | 1266 | |
b44a10b7 BP |
1267 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1268 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1269 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1270 | |
1271 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1272 | error = add_internal_flows(ofproto); |
1273 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1274 | ||
735d7efb AZ |
1275 | ofproto->n_hit = 0; |
1276 | ofproto->n_missed = 0; | |
1277 | ||
c57b2226 BP |
1278 | return error; |
1279 | } | |
1280 | ||
1281 | static int | |
1282 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1283 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1284 | { |
1285 | struct ofputil_flow_mod fm; | |
1286 | int error; | |
1287 | ||
81a76618 BP |
1288 | match_init_catchall(&fm.match); |
1289 | fm.priority = 0; | |
1290 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1291 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1292 | fm.cookie = htonll(0); |
1293 | fm.cookie_mask = htonll(0); | |
23342857 | 1294 | fm.modify_cookie = false; |
c57b2226 BP |
1295 | fm.table_id = TBL_INTERNAL; |
1296 | fm.command = OFPFC_ADD; | |
1297 | fm.idle_timeout = 0; | |
1298 | fm.hard_timeout = 0; | |
1299 | fm.buffer_id = 0; | |
1300 | fm.out_port = 0; | |
1301 | fm.flags = 0; | |
f25d0cf3 BP |
1302 | fm.ofpacts = ofpacts->data; |
1303 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1304 | |
1305 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1306 | if (error) { | |
1307 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1308 | id, ofperr_to_string(error)); | |
1309 | return error; | |
1310 | } | |
1311 | ||
ad3efdcb EJ |
1312 | if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL, |
1313 | rulep)) { | |
a2143702 | 1314 | rule_dpif_unref(*rulep); |
ad3efdcb EJ |
1315 | } else { |
1316 | NOT_REACHED(); | |
1317 | } | |
0f5f95a9 | 1318 | |
abe529af BP |
1319 | return 0; |
1320 | } | |
1321 | ||
c57b2226 BP |
1322 | static int |
1323 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1324 | { | |
f25d0cf3 BP |
1325 | struct ofpact_controller *controller; |
1326 | uint64_t ofpacts_stub[128 / 8]; | |
1327 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1328 | int error; |
1329 | int id; | |
1330 | ||
f25d0cf3 | 1331 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1332 | id = 1; |
1333 | ||
f25d0cf3 BP |
1334 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1335 | controller->max_len = UINT16_MAX; | |
1336 | controller->controller_id = 0; | |
1337 | controller->reason = OFPR_NO_MATCH; | |
1338 | ofpact_pad(&ofpacts); | |
1339 | ||
1340 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1341 | if (error) { |
1342 | return error; | |
1343 | } | |
1344 | ||
f25d0cf3 BP |
1345 | ofpbuf_clear(&ofpacts); |
1346 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 | 1347 | &ofproto->no_packet_in_rule); |
7fd51d39 BP |
1348 | if (error) { |
1349 | return error; | |
1350 | } | |
1351 | ||
1352 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1353 | &ofproto->drop_frags_rule); | |
c57b2226 BP |
1354 | return error; |
1355 | } | |
1356 | ||
abe529af BP |
1357 | static void |
1358 | destruct(struct ofproto *ofproto_) | |
1359 | { | |
1360 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1361 | struct rule_dpif *rule, *next_rule; |
24623c91 | 1362 | struct ofputil_packet_in *pin, *next_pin; |
d6aa7ad6 EJ |
1363 | struct facet *facet, *next_facet; |
1364 | struct cls_cursor cursor; | |
d0918789 | 1365 | struct oftable *table; |
1b63b91e | 1366 | struct list pins; |
abe529af | 1367 | |
d6aa7ad6 EJ |
1368 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
1369 | cls_cursor_init(&cursor, &ofproto->facets, NULL); | |
1370 | ovs_rwlock_unlock(&ofproto->facets.rwlock); | |
1371 | CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) { | |
1372 | facet_remove(facet); | |
1373 | } | |
1374 | ||
875b94ed | 1375 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1376 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1377 | xlate_remove_ofproto(ofproto); |
dc24a00f | 1378 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 1379 | |
05067881 BP |
1380 | /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a |
1381 | * use-after-free error. */ | |
1382 | udpif_revalidate(ofproto->backer->udpif); | |
1fe409d5 | 1383 | |
b44a10b7 | 1384 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 | 1385 | |
0697b5c3 BP |
1386 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1387 | struct cls_cursor cursor; | |
1388 | ||
15aaf599 | 1389 | ovs_rwlock_rdlock(&table->cls.rwlock); |
d0918789 | 1390 | cls_cursor_init(&cursor, &table->cls, NULL); |
15aaf599 | 1391 | ovs_rwlock_unlock(&table->cls.rwlock); |
0697b5c3 | 1392 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
8b81d1ef | 1393 | ofproto_rule_delete(&ofproto->up, &rule->up); |
0697b5c3 | 1394 | } |
7ee20df1 BP |
1395 | } |
1396 | ||
05067881 BP |
1397 | guarded_list_pop_all(&ofproto->pins, &pins); |
1398 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { | |
ada3a58d | 1399 | list_remove(&pin->list_node); |
24623c91 | 1400 | free(CONST_CAST(void *, pin->packet)); |
ada3a58d EJ |
1401 | free(pin); |
1402 | } | |
05067881 | 1403 | guarded_list_destroy(&ofproto->pins); |
ada3a58d | 1404 | |
ec7ceaed | 1405 | mbridge_unref(ofproto->mbridge); |
abe529af BP |
1406 | |
1407 | netflow_destroy(ofproto->netflow); | |
9723bcce | 1408 | dpif_sflow_unref(ofproto->sflow); |
abe529af | 1409 | hmap_destroy(&ofproto->bundles); |
5d989517 | 1410 | mac_learning_unref(ofproto->ml); |
abe529af | 1411 | |
bcd2633a | 1412 | classifier_destroy(&ofproto->facets); |
abe529af | 1413 | |
52a90c29 BP |
1414 | hmap_destroy(&ofproto->vlandev_map); |
1415 | hmap_destroy(&ofproto->realdev_vid_map); | |
1416 | ||
acf60855 | 1417 | sset_destroy(&ofproto->ports); |
0a740f48 | 1418 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1419 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1420 | |
0da3c61b | 1421 | ovs_mutex_destroy(&ofproto->stats_mutex); |
13501305 EJ |
1422 | ovs_mutex_destroy(&ofproto->vsp_mutex); |
1423 | ||
acf60855 | 1424 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1425 | } |
1426 | ||
1427 | static int | |
5fcc0d00 | 1428 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1429 | { |
1430 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ada3a58d | 1431 | struct ofputil_packet_in *pin, *next_pin; |
1b63b91e | 1432 | struct list pins; |
abe529af | 1433 | |
40358701 GS |
1434 | /* Do not perform any periodic activity required by 'ofproto' while |
1435 | * waiting for flow restore to complete. */ | |
1436 | if (ofproto_get_flow_restore_wait()) { | |
1437 | return 0; | |
1438 | } | |
1439 | ||
05067881 | 1440 | guarded_list_pop_all(&ofproto->pins, &pins); |
ada3a58d EJ |
1441 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { |
1442 | connmgr_send_packet_in(ofproto->up.connmgr, pin); | |
1443 | list_remove(&pin->list_node); | |
1444 | free(CONST_CAST(void *, pin->packet)); | |
1445 | free(pin); | |
1446 | } | |
1447 | ||
635c5db9 | 1448 | ofproto_dpif_monitor_run_fast(); |
5fcc0d00 BP |
1449 | return 0; |
1450 | } | |
1451 | ||
1452 | static int | |
1453 | run(struct ofproto *ofproto_) | |
1454 | { | |
1455 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1456 | struct ofport_dpif *ofport; | |
1457 | struct ofbundle *bundle; | |
1458 | int error; | |
1459 | ||
ec7ceaed EJ |
1460 | if (mbridge_need_revalidate(ofproto->mbridge)) { |
1461 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
509c0149 | 1462 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 1463 | mac_learning_flush(ofproto->ml); |
509c0149 | 1464 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
ec7ceaed EJ |
1465 | } |
1466 | ||
40358701 GS |
1467 | /* Do not perform any periodic activity below required by 'ofproto' while |
1468 | * waiting for flow restore to complete. */ | |
1469 | if (ofproto_get_flow_restore_wait()) { | |
1470 | return 0; | |
1471 | } | |
1472 | ||
5fcc0d00 BP |
1473 | error = run_fast(ofproto_); |
1474 | if (error) { | |
1475 | return error; | |
abe529af BP |
1476 | } |
1477 | ||
abe529af | 1478 | if (ofproto->netflow) { |
6fca1ffb BP |
1479 | if (netflow_run(ofproto->netflow)) { |
1480 | send_netflow_active_timeouts(ofproto); | |
1481 | } | |
abe529af BP |
1482 | } |
1483 | if (ofproto->sflow) { | |
bae473fe | 1484 | dpif_sflow_run(ofproto->sflow); |
abe529af | 1485 | } |
978427a5 RL |
1486 | if (ofproto->ipfix) { |
1487 | dpif_ipfix_run(ofproto->ipfix); | |
1488 | } | |
abe529af | 1489 | |
635c5db9 AW |
1490 | ofproto_dpif_monitor_run_fast(); |
1491 | ofproto_dpif_monitor_run(); | |
1492 | ||
abe529af BP |
1493 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1494 | port_run(ofport); | |
1495 | } | |
1496 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1497 | bundle_run(bundle); | |
1498 | } | |
1499 | ||
21f7563c | 1500 | stp_run(ofproto); |
509c0149 | 1501 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 EJ |
1502 | if (mac_learning_run(ofproto->ml)) { |
1503 | ofproto->backer->need_revalidate = REV_MAC_LEARNING; | |
1504 | } | |
509c0149 | 1505 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 1506 | |
6814e51f | 1507 | /* Check the consistency of a random facet, to aid debugging. */ |
0b4f2078 | 1508 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
4d2a0f39 | 1509 | if (time_msec() >= ofproto->consistency_rl |
bcd2633a | 1510 | && !classifier_is_empty(&ofproto->facets) |
2cc3c58e | 1511 | && !ofproto->backer->need_revalidate) { |
bcd2633a JP |
1512 | struct cls_table *table; |
1513 | struct cls_rule *cr; | |
6814e51f BP |
1514 | struct facet *facet; |
1515 | ||
4d2a0f39 EJ |
1516 | ofproto->consistency_rl = time_msec() + 250; |
1517 | ||
bcd2633a JP |
1518 | table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables), |
1519 | struct cls_table, hmap_node); | |
1520 | cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, | |
1521 | hmap_node); | |
1522 | facet = CONTAINER_OF(cr, struct facet, cr); | |
1523 | ||
a1616063 EJ |
1524 | if (!facet_check_consistency(facet)) { |
1525 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
6814e51f BP |
1526 | } |
1527 | } | |
0b4f2078 | 1528 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 1529 | |
abe529af BP |
1530 | return 0; |
1531 | } | |
1532 | ||
1533 | static void | |
1534 | wait(struct ofproto *ofproto_) | |
1535 | { | |
1536 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af BP |
1537 | struct ofbundle *bundle; |
1538 | ||
40358701 GS |
1539 | if (ofproto_get_flow_restore_wait()) { |
1540 | return; | |
1541 | } | |
1542 | ||
abe529af | 1543 | if (ofproto->sflow) { |
bae473fe | 1544 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1545 | } |
978427a5 RL |
1546 | if (ofproto->ipfix) { |
1547 | dpif_ipfix_wait(ofproto->ipfix); | |
1548 | } | |
635c5db9 | 1549 | ofproto_dpif_monitor_wait(); |
abe529af BP |
1550 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
1551 | bundle_wait(bundle); | |
1552 | } | |
6fca1ffb BP |
1553 | if (ofproto->netflow) { |
1554 | netflow_wait(ofproto->netflow); | |
1555 | } | |
509c0149 | 1556 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
1c313b88 | 1557 | mac_learning_wait(ofproto->ml); |
509c0149 | 1558 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c | 1559 | stp_wait(ofproto); |
2cc3c58e | 1560 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1561 | /* Shouldn't happen, but if it does just go around again. */ |
1562 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1563 | poll_immediate_wake(); | |
abe529af BP |
1564 | } |
1565 | } | |
1566 | ||
0d085684 BP |
1567 | static void |
1568 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1569 | { | |
1570 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bcd2633a | 1571 | struct cls_cursor cursor; |
04d08d54 EJ |
1572 | size_t n_subfacets = 0; |
1573 | struct facet *facet; | |
0d085684 | 1574 | |
0b4f2078 | 1575 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a | 1576 | simap_increase(usage, "facets", classifier_count(&ofproto->facets)); |
0b4f2078 | 1577 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 1578 | |
0b4f2078 | 1579 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
1580 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
1581 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
04d08d54 EJ |
1582 | n_subfacets += list_size(&facet->subfacets); |
1583 | } | |
0b4f2078 | 1584 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
04d08d54 | 1585 | simap_increase(usage, "subfacets", n_subfacets); |
0d085684 BP |
1586 | } |
1587 | ||
abe529af BP |
1588 | static void |
1589 | flush(struct ofproto *ofproto_) | |
1590 | { | |
1591 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1592 | struct subfacet *subfacet, *next_subfacet; |
1593 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1594 | int n_batch; | |
b0f7b9b5 | 1595 | |
acf60855 JP |
1596 | n_batch = 0; |
1597 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
04d08d54 | 1598 | &ofproto->backer->subfacets) { |
b7e2f6b3 | 1599 | if (subfacet->facet->ofproto != ofproto) { |
04d08d54 EJ |
1600 | continue; |
1601 | } | |
1602 | ||
acf60855 JP |
1603 | if (subfacet->path != SF_NOT_INSTALLED) { |
1604 | batch[n_batch++] = subfacet; | |
1605 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
04d08d54 | 1606 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 JP |
1607 | n_batch = 0; |
1608 | } | |
1609 | } else { | |
1610 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1611 | } |
abe529af | 1612 | } |
acf60855 JP |
1613 | |
1614 | if (n_batch > 0) { | |
04d08d54 | 1615 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 | 1616 | } |
abe529af BP |
1617 | } |
1618 | ||
6c1491fb BP |
1619 | static void |
1620 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1621 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1622 | { |
1623 | *arp_match_ip = true; | |
9e1fd49b BP |
1624 | *actions = (OFPUTIL_A_OUTPUT | |
1625 | OFPUTIL_A_SET_VLAN_VID | | |
1626 | OFPUTIL_A_SET_VLAN_PCP | | |
1627 | OFPUTIL_A_STRIP_VLAN | | |
1628 | OFPUTIL_A_SET_DL_SRC | | |
1629 | OFPUTIL_A_SET_DL_DST | | |
1630 | OFPUTIL_A_SET_NW_SRC | | |
1631 | OFPUTIL_A_SET_NW_DST | | |
1632 | OFPUTIL_A_SET_NW_TOS | | |
1633 | OFPUTIL_A_SET_TP_SRC | | |
1634 | OFPUTIL_A_SET_TP_DST | | |
1635 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1636 | } |
1637 | ||
1638 | static void | |
307975da | 1639 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1640 | { |
1641 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1642 | struct dpif_dp_stats s; |
7fd51d39 | 1643 | uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; |
7e741ffb | 1644 | uint64_t n_lookup; |
6c1491fb BP |
1645 | |
1646 | strcpy(ots->name, "classifier"); | |
1647 | ||
acf60855 | 1648 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1649 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1650 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
7fd51d39 | 1651 | rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); |
acf60855 | 1652 | |
7fd51d39 | 1653 | n_lookup = s.n_hit + s.n_missed - n_dropped_frags; |
7e741ffb JG |
1654 | ots->lookup_count = htonll(n_lookup); |
1655 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1656 | } |
1657 | ||
abe529af BP |
1658 | static struct ofport * |
1659 | port_alloc(void) | |
1660 | { | |
1661 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1662 | return &port->up; | |
1663 | } | |
1664 | ||
1665 | static void | |
1666 | port_dealloc(struct ofport *port_) | |
1667 | { | |
1668 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1669 | free(port); | |
1670 | } | |
1671 | ||
1672 | static int | |
1673 | port_construct(struct ofport *port_) | |
1674 | { | |
1675 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1676 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1677 | const struct netdev *netdev = port->up.netdev; |
3aa30359 | 1678 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
e1b1d06a JP |
1679 | struct dpif_port dpif_port; |
1680 | int error; | |
abe529af | 1681 | |
2cc3c58e | 1682 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1683 | port->bundle = NULL; |
1684 | port->cfm = NULL; | |
ccc09689 | 1685 | port->bfd = NULL; |
d5ffa7f2 | 1686 | port->may_enable = true; |
21f7563c JP |
1687 | port->stp_port = NULL; |
1688 | port->stp_state = STP_DISABLED; | |
42943cde | 1689 | port->is_tunnel = false; |
6cbbf4fa | 1690 | port->peer = NULL; |
55954f6e EJ |
1691 | port->qdscp = NULL; |
1692 | port->n_qdscp = 0; | |
52a90c29 BP |
1693 | port->realdev_ofp_port = 0; |
1694 | port->vlandev_vid = 0; | |
b9ad7294 | 1695 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1696 | |
b9ad7294 | 1697 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1698 | /* By bailing out here, we don't submit the port to the sFlow module |
1699 | * to be considered for counter polling export. This is correct | |
1700 | * because the patch port represents an interface that sFlow considers | |
1701 | * to be "internal" to the switch as a whole, and therefore not an | |
1702 | * candidate for counter polling. */ | |
4e022ec0 | 1703 | port->odp_port = ODPP_NONE; |
6cbbf4fa | 1704 | ofport_update_peer(port); |
0a740f48 EJ |
1705 | return 0; |
1706 | } | |
1707 | ||
acf60855 | 1708 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
3aa30359 BP |
1709 | netdev_vport_get_dpif_port(netdev, namebuf, |
1710 | sizeof namebuf), | |
e1b1d06a JP |
1711 | &dpif_port); |
1712 | if (error) { | |
1713 | return error; | |
1714 | } | |
1715 | ||
1716 | port->odp_port = dpif_port.port_no; | |
1717 | ||
b9ad7294 | 1718 | if (netdev_get_tunnel_config(netdev)) { |
42943cde EJ |
1719 | tnl_port_add(port, port->up.netdev, port->odp_port); |
1720 | port->is_tunnel = true; | |
b9ad7294 EJ |
1721 | } else { |
1722 | /* Sanity-check that a mapping doesn't already exist. This | |
1723 | * shouldn't happen for non-tunnel ports. */ | |
1724 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1725 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1726 | dpif_port.name); | |
da78d43d | 1727 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1728 | return EBUSY; |
1729 | } | |
e1b1d06a | 1730 | |
8449c4d6 | 1731 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1732 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
f9c0c3ec | 1733 | hash_odp_port(port->odp_port)); |
8449c4d6 | 1734 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1735 | } |
da78d43d | 1736 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1737 | |
abe529af | 1738 | if (ofproto->sflow) { |
e1b1d06a | 1739 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1740 | } |
1741 | ||
1742 | return 0; | |
1743 | } | |
1744 | ||
1745 | static void | |
1746 | port_destruct(struct ofport *port_) | |
1747 | { | |
1748 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1749 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
02f8d646 | 1750 | const char *devname = netdev_get_name(port->up.netdev); |
3aa30359 BP |
1751 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1752 | const char *dp_port_name; | |
abe529af | 1753 | |
7894e7da | 1754 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1755 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1756 | xlate_ofport_remove(port); |
dc24a00f | 1757 | ovs_rwlock_unlock(&xlate_rwlock); |
7894e7da | 1758 | |
3aa30359 BP |
1759 | dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, |
1760 | sizeof namebuf); | |
a614d823 | 1761 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1762 | /* The underlying device is still there, so delete it. This |
1763 | * happens when the ofproto is being destroyed, since the caller | |
1764 | * assumes that removal of attached ports will happen as part of | |
1765 | * destruction. */ | |
42943cde | 1766 | if (!port->is_tunnel) { |
a614d823 KM |
1767 | dpif_port_del(ofproto->backer->dpif, port->odp_port); |
1768 | } | |
acf60855 JP |
1769 | } |
1770 | ||
6cbbf4fa EJ |
1771 | if (port->peer) { |
1772 | port->peer->peer = NULL; | |
1773 | port->peer = NULL; | |
1774 | } | |
1775 | ||
42943cde | 1776 | if (port->odp_port != ODPP_NONE && !port->is_tunnel) { |
8449c4d6 | 1777 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 | 1778 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
8449c4d6 | 1779 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 EJ |
1780 | } |
1781 | ||
42943cde | 1782 | tnl_port_del(port); |
02f8d646 | 1783 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1784 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
abe529af | 1785 | bundle_remove(port_); |
a5610457 | 1786 | set_cfm(port_, NULL); |
8aee94b6 | 1787 | set_bfd(port_, NULL); |
abe529af | 1788 | if (ofproto->sflow) { |
bae473fe | 1789 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1790 | } |
8b36f51e | 1791 | |
55954f6e | 1792 | free(port->qdscp); |
abe529af BP |
1793 | } |
1794 | ||
1795 | static void | |
1796 | port_modified(struct ofport *port_) | |
1797 | { | |
1798 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1799 | ||
1800 | if (port->bundle && port->bundle->bond) { | |
1801 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1802 | } | |
9d46b444 EJ |
1803 | |
1804 | if (port->cfm) { | |
1805 | cfm_set_netdev(port->cfm, port->up.netdev); | |
1806 | } | |
f1e78bbd | 1807 | |
c1c4e8c7 AW |
1808 | if (port->bfd) { |
1809 | bfd_set_netdev(port->bfd, port->up.netdev); | |
1810 | } | |
1811 | ||
635c5db9 AW |
1812 | ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm, |
1813 | port->up.pp.hw_addr); | |
1814 | ||
42943cde EJ |
1815 | if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, |
1816 | port->odp_port)) { | |
7894e7da EJ |
1817 | ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = |
1818 | REV_RECONFIGURE; | |
f1e78bbd | 1819 | } |
6cbbf4fa EJ |
1820 | |
1821 | ofport_update_peer(port); | |
abe529af BP |
1822 | } |
1823 | ||
1824 | static void | |
9e1fd49b | 1825 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1826 | { |
1827 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1828 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1829 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1830 | |
9e1fd49b | 1831 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1832 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1833 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1834 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1835 | |
9e1fd49b | 1836 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1837 | bundle_update(port->bundle); |
1838 | } | |
abe529af BP |
1839 | } |
1840 | } | |
1841 | ||
1842 | static int | |
1843 | set_sflow(struct ofproto *ofproto_, | |
1844 | const struct ofproto_sflow_options *sflow_options) | |
1845 | { | |
1846 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1847 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1848 | |
abe529af | 1849 | if (sflow_options) { |
bae473fe | 1850 | if (!ds) { |
abe529af BP |
1851 | struct ofport_dpif *ofport; |
1852 | ||
4213f19d | 1853 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1854 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1855 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1856 | } |
2cc3c58e | 1857 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1858 | } |
bae473fe | 1859 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1860 | } else { |
6ff686f2 | 1861 | if (ds) { |
9723bcce | 1862 | dpif_sflow_unref(ds); |
2cc3c58e | 1863 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1864 | ofproto->sflow = NULL; |
1865 | } | |
abe529af BP |
1866 | } |
1867 | return 0; | |
1868 | } | |
1869 | ||
29089a54 RL |
1870 | static int |
1871 | set_ipfix( | |
1872 | struct ofproto *ofproto_, | |
1873 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1874 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1875 | size_t n_flow_exporters_options) | |
1876 | { | |
1877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1878 | struct dpif_ipfix *di = ofproto->ipfix; | |
978427a5 | 1879 | bool has_options = bridge_exporter_options || flow_exporters_options; |
29089a54 | 1880 | |
978427a5 RL |
1881 | if (has_options && !di) { |
1882 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1883 | } | |
1884 | ||
1885 | if (di) { | |
1886 | /* Call set_options in any case to cleanly flush the flow | |
1887 | * caches in the last exporters that are to be destroyed. */ | |
29089a54 RL |
1888 | dpif_ipfix_set_options( |
1889 | di, bridge_exporter_options, flow_exporters_options, | |
1890 | n_flow_exporters_options); | |
978427a5 RL |
1891 | |
1892 | if (!has_options) { | |
d857c8aa | 1893 | dpif_ipfix_unref(di); |
29089a54 RL |
1894 | ofproto->ipfix = NULL; |
1895 | } | |
1896 | } | |
978427a5 | 1897 | |
29089a54 RL |
1898 | return 0; |
1899 | } | |
1900 | ||
abe529af | 1901 | static int |
a5610457 | 1902 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1903 | { |
1904 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
635c5db9 | 1905 | int error = 0; |
abe529af | 1906 | |
635c5db9 | 1907 | if (s) { |
abe529af | 1908 | if (!ofport->cfm) { |
8c977421 EJ |
1909 | struct ofproto_dpif *ofproto; |
1910 | ||
1911 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1912 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 1913 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
1914 | } |
1915 | ||
a5610457 | 1916 | if (cfm_configure(ofport->cfm, s)) { |
635c5db9 AW |
1917 | error = 0; |
1918 | goto out; | |
abe529af BP |
1919 | } |
1920 | ||
1921 | error = EINVAL; | |
1922 | } | |
8e9a73fa | 1923 | cfm_unref(ofport->cfm); |
abe529af | 1924 | ofport->cfm = NULL; |
635c5db9 AW |
1925 | out: |
1926 | ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, | |
1927 | ofport->up.pp.hw_addr); | |
abe529af BP |
1928 | return error; |
1929 | } | |
1930 | ||
9a9e3786 BP |
1931 | static bool |
1932 | get_cfm_status(const struct ofport *ofport_, | |
1933 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1934 | { |
1935 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1936 | ||
1937 | if (ofport->cfm) { | |
9a9e3786 BP |
1938 | status->faults = cfm_get_fault(ofport->cfm); |
1939 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1940 | status->health = cfm_get_health(ofport->cfm); | |
1941 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1942 | return true; | |
1de11730 | 1943 | } else { |
9a9e3786 | 1944 | return false; |
1de11730 EJ |
1945 | } |
1946 | } | |
ccc09689 EJ |
1947 | |
1948 | static int | |
1949 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
1950 | { | |
1951 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
1952 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1953 | struct bfd *old; | |
1954 | ||
1955 | old = ofport->bfd; | |
c1c4e8c7 AW |
1956 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), |
1957 | cfg, ofport->up.netdev); | |
ccc09689 EJ |
1958 | if (ofport->bfd != old) { |
1959 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1960 | } | |
635c5db9 AW |
1961 | ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, |
1962 | ofport->up.pp.hw_addr); | |
ccc09689 EJ |
1963 | return 0; |
1964 | } | |
1965 | ||
1966 | static int | |
1967 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
1968 | { | |
1969 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1970 | ||
1971 | if (ofport->bfd) { | |
1972 | bfd_get_status(ofport->bfd, smap); | |
1973 | return 0; | |
1974 | } else { | |
1975 | return ENOENT; | |
1976 | } | |
1977 | } | |
abe529af | 1978 | \f |
21f7563c JP |
1979 | /* Spanning Tree. */ |
1980 | ||
1981 | static void | |
1982 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1983 | { | |
1984 | struct ofproto_dpif *ofproto = ofproto_; | |
1985 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1986 | struct ofport_dpif *ofport; | |
1987 | ||
1988 | ofport = stp_port_get_aux(sp); | |
1989 | if (!ofport) { | |
1990 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1991 | ofproto->up.name, port_num); | |
1992 | } else { | |
1993 | struct eth_header *eth = pkt->l2; | |
1994 | ||
1995 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1996 | if (eth_addr_is_zero(eth->eth_src)) { | |
1997 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1998 | "with unknown MAC", ofproto->up.name, port_num); | |
1999 | } else { | |
91d6cd12 | 2000 | ofproto_dpif_send_packet(ofport, pkt); |
21f7563c JP |
2001 | } |
2002 | } | |
2003 | ofpbuf_delete(pkt); | |
2004 | } | |
2005 | ||
2006 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2007 | static int | |
2008 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2009 | { | |
2010 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2011 | ||
2012 | /* Only revalidate flows if the configuration changed. */ | |
2013 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2014 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2015 | } |
2016 | ||
2017 | if (s) { | |
2018 | if (!ofproto->stp) { | |
2019 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2020 | send_bpdu_cb, ofproto); | |
2021 | ofproto->stp_last_tick = time_msec(); | |
2022 | } | |
2023 | ||
2024 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2025 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2026 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2027 | stp_set_max_age(ofproto->stp, s->max_age); | |
2028 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2029 | } else { | |
851bf71d EJ |
2030 | struct ofport *ofport; |
2031 | ||
2032 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2033 | set_stp_port(ofport, NULL); | |
2034 | } | |
2035 | ||
bd54dbcd | 2036 | stp_unref(ofproto->stp); |
21f7563c JP |
2037 | ofproto->stp = NULL; |
2038 | } | |
2039 | ||
2040 | return 0; | |
2041 | } | |
2042 | ||
2043 | static int | |
2044 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2045 | { | |
2046 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2047 | ||
2048 | if (ofproto->stp) { | |
2049 | s->enabled = true; | |
2050 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2051 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2052 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2053 | } else { | |
2054 | s->enabled = false; | |
2055 | } | |
2056 | ||
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | static void | |
2061 | update_stp_port_state(struct ofport_dpif *ofport) | |
2062 | { | |
2063 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2064 | enum stp_state state; | |
2065 | ||
2066 | /* Figure out new state. */ | |
2067 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2068 | : STP_DISABLED; | |
2069 | ||
2070 | /* Update state. */ | |
2071 | if (ofport->stp_state != state) { | |
9e1fd49b | 2072 | enum ofputil_port_state of_state; |
21f7563c JP |
2073 | bool fwd_change; |
2074 | ||
2075 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2076 | netdev_get_name(ofport->up.netdev), | |
2077 | stp_state_name(ofport->stp_state), | |
2078 | stp_state_name(state)); | |
2079 | if (stp_learn_in_state(ofport->stp_state) | |
2080 | != stp_learn_in_state(state)) { | |
2081 | /* xxx Learning action flows should also be flushed. */ | |
509c0149 | 2082 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2083 | mac_learning_flush(ofproto->ml); |
509c0149 | 2084 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c JP |
2085 | } |
2086 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2087 | != stp_forward_in_state(state); | |
2088 | ||
2cc3c58e | 2089 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2090 | ofport->stp_state = state; |
2091 | ofport->stp_state_entered = time_msec(); | |
2092 | ||
b308140a | 2093 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2094 | bundle_update(ofport->bundle); |
2095 | } | |
2096 | ||
2097 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2098 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2099 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2100 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2101 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2102 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2103 | : 0); | |
21f7563c JP |
2104 | ofproto_port_set_state(&ofport->up, of_state); |
2105 | } | |
2106 | } | |
2107 | ||
2108 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2109 | * caller is responsible for assigning STP port numbers and ensuring | |
2110 | * there are no duplicates. */ | |
2111 | static int | |
2112 | set_stp_port(struct ofport *ofport_, | |
2113 | const struct ofproto_port_stp_settings *s) | |
2114 | { | |
2115 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2116 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2117 | struct stp_port *sp = ofport->stp_port; | |
2118 | ||
2119 | if (!s || !s->enable) { | |
2120 | if (sp) { | |
2121 | ofport->stp_port = NULL; | |
2122 | stp_port_disable(sp); | |
ecd12731 | 2123 | update_stp_port_state(ofport); |
21f7563c JP |
2124 | } |
2125 | return 0; | |
2126 | } else if (sp && stp_port_no(sp) != s->port_num | |
2127 | && ofport == stp_port_get_aux(sp)) { | |
2128 | /* The port-id changed, so disable the old one if it's not | |
2129 | * already in use by another port. */ | |
2130 | stp_port_disable(sp); | |
2131 | } | |
2132 | ||
2133 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2134 | stp_port_enable(sp); | |
2135 | ||
2136 | stp_port_set_aux(sp, ofport); | |
2137 | stp_port_set_priority(sp, s->priority); | |
2138 | stp_port_set_path_cost(sp, s->path_cost); | |
2139 | ||
2140 | update_stp_port_state(ofport); | |
2141 | ||
2142 | return 0; | |
2143 | } | |
2144 | ||
2145 | static int | |
2146 | get_stp_port_status(struct ofport *ofport_, | |
2147 | struct ofproto_port_stp_status *s) | |
2148 | { | |
2149 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2150 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2151 | struct stp_port *sp = ofport->stp_port; | |
2152 | ||
2153 | if (!ofproto->stp || !sp) { | |
2154 | s->enabled = false; | |
2155 | return 0; | |
2156 | } | |
2157 | ||
2158 | s->enabled = true; | |
2159 | s->port_id = stp_port_get_id(sp); | |
2160 | s->state = stp_port_get_state(sp); | |
2161 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2162 | s->role = stp_port_get_role(sp); | |
80740385 | 2163 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2164 | |
2165 | return 0; | |
2166 | } | |
2167 | ||
2168 | static void | |
2169 | stp_run(struct ofproto_dpif *ofproto) | |
2170 | { | |
2171 | if (ofproto->stp) { | |
2172 | long long int now = time_msec(); | |
2173 | long long int elapsed = now - ofproto->stp_last_tick; | |
2174 | struct stp_port *sp; | |
2175 | ||
2176 | if (elapsed > 0) { | |
2177 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2178 | ofproto->stp_last_tick = now; | |
2179 | } | |
2180 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2181 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2182 | ||
2183 | if (ofport) { | |
2184 | update_stp_port_state(ofport); | |
2185 | } | |
2186 | } | |
6ae50723 EJ |
2187 | |
2188 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
509c0149 | 2189 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2190 | mac_learning_flush(ofproto->ml); |
509c0149 | 2191 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
6ae50723 | 2192 | } |
21f7563c JP |
2193 | } |
2194 | } | |
2195 | ||
2196 | static void | |
2197 | stp_wait(struct ofproto_dpif *ofproto) | |
2198 | { | |
2199 | if (ofproto->stp) { | |
2200 | poll_timer_wait(1000); | |
2201 | } | |
2202 | } | |
21f7563c | 2203 | \f |
8b36f51e | 2204 | static int |
55954f6e | 2205 | set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp, |
8b36f51e EJ |
2206 | size_t n_qdscp) |
2207 | { | |
2208 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2209 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
8b36f51e | 2210 | |
55954f6e EJ |
2211 | if (ofport->n_qdscp != n_qdscp |
2212 | || (n_qdscp && memcmp(ofport->qdscp, qdscp, | |
2213 | n_qdscp * sizeof *qdscp))) { | |
2cc3c58e | 2214 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
55954f6e EJ |
2215 | free(ofport->qdscp); |
2216 | ofport->qdscp = n_qdscp | |
2217 | ? xmemdup(qdscp, n_qdscp * sizeof *qdscp) | |
2218 | : NULL; | |
2219 | ofport->n_qdscp = n_qdscp; | |
8b36f51e EJ |
2220 | } |
2221 | ||
8b36f51e EJ |
2222 | return 0; |
2223 | } | |
2224 | \f | |
abe529af BP |
2225 | /* Bundles. */ |
2226 | ||
b44a10b7 BP |
2227 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2228 | * ofproto to revalidate every flow. | |
2229 | * | |
2230 | * Normally MAC learning entries are removed only from the ofproto associated | |
2231 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2232 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2233 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2234 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2235 | * with the host from which it migrated. */ | |
abe529af | 2236 | static void |
b44a10b7 | 2237 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2238 | { |
2239 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2240 | struct mac_learning *ml = ofproto->ml; | |
2241 | struct mac_entry *mac, *next_mac; | |
2242 | ||
2cc3c58e | 2243 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
509c0149 | 2244 | ovs_rwlock_wrlock(&ml->rwlock); |
abe529af BP |
2245 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2246 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2247 | if (all_ofprotos) { |
2248 | struct ofproto_dpif *o; | |
2249 | ||
2250 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2251 | if (o != ofproto) { | |
2252 | struct mac_entry *e; | |
2253 | ||
509c0149 | 2254 | ovs_rwlock_wrlock(&o->ml->rwlock); |
30618594 | 2255 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan); |
b44a10b7 | 2256 | if (e) { |
b44a10b7 BP |
2257 | mac_learning_expire(o->ml, e); |
2258 | } | |
509c0149 | 2259 | ovs_rwlock_unlock(&o->ml->rwlock); |
b44a10b7 BP |
2260 | } |
2261 | } | |
2262 | } | |
2263 | ||
abe529af BP |
2264 | mac_learning_expire(ml, mac); |
2265 | } | |
2266 | } | |
509c0149 | 2267 | ovs_rwlock_unlock(&ml->rwlock); |
abe529af BP |
2268 | } |
2269 | ||
2270 | static struct ofbundle * | |
2271 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2272 | { | |
2273 | struct ofbundle *bundle; | |
2274 | ||
2275 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2276 | &ofproto->bundles) { | |
2277 | if (bundle->aux == aux) { | |
2278 | return bundle; | |
2279 | } | |
2280 | } | |
2281 | return NULL; | |
2282 | } | |
2283 | ||
7bde8dd8 JP |
2284 | static void |
2285 | bundle_update(struct ofbundle *bundle) | |
2286 | { | |
2287 | struct ofport_dpif *port; | |
2288 | ||
2289 | bundle->floodable = true; | |
2290 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2291 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2292 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2293 | bundle->floodable = false; |
2294 | break; | |
2295 | } | |
2296 | } | |
2297 | } | |
2298 | ||
abe529af BP |
2299 | static void |
2300 | bundle_del_port(struct ofport_dpif *port) | |
2301 | { | |
2302 | struct ofbundle *bundle = port->bundle; | |
2303 | ||
2cc3c58e | 2304 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2305 | |
abe529af BP |
2306 | list_remove(&port->bundle_node); |
2307 | port->bundle = NULL; | |
2308 | ||
2309 | if (bundle->lacp) { | |
2310 | lacp_slave_unregister(bundle->lacp, port); | |
2311 | } | |
2312 | if (bundle->bond) { | |
2313 | bond_slave_unregister(bundle->bond, port); | |
2314 | } | |
2315 | ||
7bde8dd8 | 2316 | bundle_update(bundle); |
abe529af BP |
2317 | } |
2318 | ||
2319 | static bool | |
4e022ec0 | 2320 | bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, |
df53d41c | 2321 | struct lacp_slave_settings *lacp) |
abe529af BP |
2322 | { |
2323 | struct ofport_dpif *port; | |
2324 | ||
2325 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2326 | if (!port) { | |
2327 | return false; | |
2328 | } | |
2329 | ||
2330 | if (port->bundle != bundle) { | |
2cc3c58e | 2331 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 2332 | if (port->bundle) { |
e019a574 | 2333 | bundle_remove(&port->up); |
abe529af BP |
2334 | } |
2335 | ||
2336 | port->bundle = bundle; | |
2337 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2338 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2339 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2340 | bundle->floodable = false; |
2341 | } | |
2342 | } | |
2343 | if (lacp) { | |
2cc3c58e | 2344 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2345 | lacp_slave_register(bundle->lacp, port, lacp); |
2346 | } | |
2347 | ||
2348 | return true; | |
2349 | } | |
2350 | ||
2351 | static void | |
2352 | bundle_destroy(struct ofbundle *bundle) | |
2353 | { | |
2354 | struct ofproto_dpif *ofproto; | |
2355 | struct ofport_dpif *port, *next_port; | |
abe529af BP |
2356 | |
2357 | if (!bundle) { | |
2358 | return; | |
2359 | } | |
2360 | ||
2361 | ofproto = bundle->ofproto; | |
ec7ceaed | 2362 | mbridge_unregister_bundle(ofproto->mbridge, bundle->aux); |
abe529af | 2363 | |
dc24a00f | 2364 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 2365 | xlate_bundle_remove(bundle); |
dc24a00f | 2366 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 2367 | |
abe529af BP |
2368 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { |
2369 | bundle_del_port(port); | |
2370 | } | |
2371 | ||
b44a10b7 | 2372 | bundle_flush_macs(bundle, true); |
abe529af BP |
2373 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2374 | free(bundle->name); | |
2375 | free(bundle->trunks); | |
91779071 | 2376 | lacp_unref(bundle->lacp); |
03366a2d | 2377 | bond_unref(bundle->bond); |
abe529af BP |
2378 | free(bundle); |
2379 | } | |
2380 | ||
2381 | static int | |
2382 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2383 | const struct ofproto_bundle_settings *s) | |
2384 | { | |
2385 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2386 | bool need_flush = false; | |
abe529af BP |
2387 | struct ofport_dpif *port; |
2388 | struct ofbundle *bundle; | |
ecac4ebf BP |
2389 | unsigned long *trunks; |
2390 | int vlan; | |
abe529af BP |
2391 | size_t i; |
2392 | bool ok; | |
2393 | ||
2394 | if (!s) { | |
2395 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2396 | return 0; | |
2397 | } | |
2398 | ||
cb22974d BP |
2399 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2400 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2401 | |
2402 | bundle = bundle_lookup(ofproto, aux); | |
2403 | if (!bundle) { | |
2404 | bundle = xmalloc(sizeof *bundle); | |
2405 | ||
2406 | bundle->ofproto = ofproto; | |
2407 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2408 | hash_pointer(aux, 0)); | |
2409 | bundle->aux = aux; | |
2410 | bundle->name = NULL; | |
2411 | ||
2412 | list_init(&bundle->ports); | |
ecac4ebf | 2413 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2414 | bundle->vlan = -1; |
2415 | bundle->trunks = NULL; | |
5e9ceccd | 2416 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2417 | bundle->lacp = NULL; |
2418 | bundle->bond = NULL; | |
2419 | ||
2420 | bundle->floodable = true; | |
ec7ceaed | 2421 | mbridge_register_bundle(ofproto->mbridge, bundle); |
abe529af BP |
2422 | } |
2423 | ||
2424 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2425 | free(bundle->name); | |
2426 | bundle->name = xstrdup(s->name); | |
2427 | } | |
2428 | ||
2429 | /* LACP. */ | |
2430 | if (s->lacp) { | |
2431 | if (!bundle->lacp) { | |
2cc3c58e | 2432 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2433 | bundle->lacp = lacp_create(); |
2434 | } | |
2435 | lacp_configure(bundle->lacp, s->lacp); | |
2436 | } else { | |
91779071 | 2437 | lacp_unref(bundle->lacp); |
abe529af BP |
2438 | bundle->lacp = NULL; |
2439 | } | |
2440 | ||
2441 | /* Update set of ports. */ | |
2442 | ok = true; | |
2443 | for (i = 0; i < s->n_slaves; i++) { | |
2444 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2445 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2446 | ok = false; |
2447 | } | |
2448 | } | |
2449 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2450 | struct ofport_dpif *next_port; | |
2451 | ||
2452 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2453 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2454 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2455 | goto found; |
2456 | } | |
2457 | } | |
2458 | ||
2459 | bundle_del_port(port); | |
2460 | found: ; | |
2461 | } | |
2462 | } | |
cb22974d | 2463 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2464 | |
2465 | if (list_is_empty(&bundle->ports)) { | |
2466 | bundle_destroy(bundle); | |
2467 | return EINVAL; | |
2468 | } | |
2469 | ||
ecac4ebf | 2470 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2471 | if (s->vlan_mode != bundle->vlan_mode |
2472 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2473 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2474 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2475 | need_flush = true; |
2476 | } | |
2477 | ||
abe529af | 2478 | /* Set VLAN tag. */ |
ecac4ebf BP |
2479 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2480 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2481 | : 0); | |
2482 | if (vlan != bundle->vlan) { | |
2483 | bundle->vlan = vlan; | |
abe529af BP |
2484 | need_flush = true; |
2485 | } | |
2486 | ||
2487 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2488 | switch (s->vlan_mode) { |
2489 | case PORT_VLAN_ACCESS: | |
2490 | trunks = NULL; | |
2491 | break; | |
2492 | ||
2493 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2494 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2495 | break; |
2496 | ||
2497 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2498 | case PORT_VLAN_NATIVE_TAGGED: | |
2499 | if (vlan != 0 && (!s->trunks | |
2500 | || !bitmap_is_set(s->trunks, vlan) | |
2501 | || bitmap_is_set(s->trunks, 0))) { | |
2502 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2503 | if (s->trunks) { | |
2504 | trunks = bitmap_clone(s->trunks, 4096); | |
2505 | } else { | |
2506 | trunks = bitmap_allocate1(4096); | |
2507 | } | |
2508 | bitmap_set1(trunks, vlan); | |
2509 | bitmap_set0(trunks, 0); | |
2510 | } else { | |
ebc56baa | 2511 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2512 | } |
2513 | break; | |
2514 | ||
2515 | default: | |
2516 | NOT_REACHED(); | |
2517 | } | |
abe529af BP |
2518 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2519 | free(bundle->trunks); | |
ecac4ebf BP |
2520 | if (trunks == s->trunks) { |
2521 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2522 | } else { | |
2523 | bundle->trunks = trunks; | |
2524 | trunks = NULL; | |
2525 | } | |
abe529af BP |
2526 | need_flush = true; |
2527 | } | |
ecac4ebf BP |
2528 | if (trunks != s->trunks) { |
2529 | free(trunks); | |
2530 | } | |
abe529af BP |
2531 | |
2532 | /* Bonding. */ | |
2533 | if (!list_is_short(&bundle->ports)) { | |
2534 | bundle->ofproto->has_bonded_bundles = true; | |
2535 | if (bundle->bond) { | |
2536 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2537 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2538 | } |
2539 | } else { | |
2540 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2541 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2542 | } |
2543 | ||
2544 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2545 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2546 | } |
2547 | } else { | |
03366a2d | 2548 | bond_unref(bundle->bond); |
abe529af BP |
2549 | bundle->bond = NULL; |
2550 | } | |
2551 | ||
2552 | /* If we changed something that would affect MAC learning, un-learn | |
2553 | * everything on this port and force flow revalidation. */ | |
2554 | if (need_flush) { | |
b44a10b7 | 2555 | bundle_flush_macs(bundle, false); |
abe529af BP |
2556 | } |
2557 | ||
2558 | return 0; | |
2559 | } | |
2560 | ||
2561 | static void | |
2562 | bundle_remove(struct ofport *port_) | |
2563 | { | |
2564 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2565 | struct ofbundle *bundle = port->bundle; | |
2566 | ||
2567 | if (bundle) { | |
2568 | bundle_del_port(port); | |
2569 | if (list_is_empty(&bundle->ports)) { | |
2570 | bundle_destroy(bundle); | |
2571 | } else if (list_is_short(&bundle->ports)) { | |
03366a2d | 2572 | bond_unref(bundle->bond); |
abe529af BP |
2573 | bundle->bond = NULL; |
2574 | } | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | static void | |
5f877369 | 2579 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2580 | { |
2581 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2582 | struct ofport_dpif *port = port_; | |
2583 | uint8_t ea[ETH_ADDR_LEN]; | |
2584 | int error; | |
2585 | ||
2586 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2587 | if (!error) { | |
abe529af | 2588 | struct ofpbuf packet; |
5f877369 | 2589 | void *packet_pdu; |
abe529af BP |
2590 | |
2591 | ofpbuf_init(&packet, 0); | |
2592 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2593 | pdu_size); |
2594 | memcpy(packet_pdu, pdu, pdu_size); | |
2595 | ||
91d6cd12 | 2596 | ofproto_dpif_send_packet(port, &packet); |
abe529af BP |
2597 | ofpbuf_uninit(&packet); |
2598 | } else { | |
2599 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2600 | "%s (%s)", port->bundle->name, | |
10a89ef0 | 2601 | netdev_get_name(port->up.netdev), ovs_strerror(error)); |
abe529af BP |
2602 | } |
2603 | } | |
2604 | ||
2605 | static void | |
2606 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2607 | { | |
2608 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
0165217e | 2609 | struct ofpbuf *learning_packet; |
abe529af BP |
2610 | int error, n_packets, n_errors; |
2611 | struct mac_entry *e; | |
0165217e | 2612 | struct list packets; |
abe529af | 2613 | |
0165217e | 2614 | list_init(&packets); |
509c0149 | 2615 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
2616 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
2617 | if (e->port.p != bundle) { | |
4dd1e3ca | 2618 | void *port_void; |
ea131871 | 2619 | |
4dd1e3ca BP |
2620 | learning_packet = bond_compose_learning_packet(bundle->bond, |
2621 | e->mac, e->vlan, | |
2622 | &port_void); | |
0165217e EJ |
2623 | learning_packet->private_p = port_void; |
2624 | list_push_back(&packets, &learning_packet->list_node); | |
abe529af BP |
2625 | } |
2626 | } | |
509c0149 | 2627 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 2628 | |
0165217e EJ |
2629 | error = n_packets = n_errors = 0; |
2630 | LIST_FOR_EACH (learning_packet, list_node, &packets) { | |
2631 | int ret; | |
2632 | ||
91d6cd12 | 2633 | ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet); |
0165217e EJ |
2634 | if (ret) { |
2635 | error = ret; | |
2636 | n_errors++; | |
2637 | } | |
2638 | n_packets++; | |
2639 | } | |
2640 | ofpbuf_list_delete(&packets); | |
2641 | ||
abe529af BP |
2642 | if (n_errors) { |
2643 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2644 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2645 | "packets, last error was: %s", | |
10a89ef0 | 2646 | bundle->name, n_errors, n_packets, ovs_strerror(error)); |
abe529af BP |
2647 | } else { |
2648 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2649 | bundle->name, n_packets); | |
2650 | } | |
2651 | } | |
2652 | ||
2653 | static void | |
2654 | bundle_run(struct ofbundle *bundle) | |
2655 | { | |
2656 | if (bundle->lacp) { | |
2657 | lacp_run(bundle->lacp, send_pdu_cb); | |
2658 | } | |
2659 | if (bundle->bond) { | |
2660 | struct ofport_dpif *port; | |
2661 | ||
2662 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2663 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2664 | } |
2665 | ||
4a1b8f30 EJ |
2666 | if (bond_run(bundle->bond, lacp_status(bundle->lacp))) { |
2667 | bundle->ofproto->backer->need_revalidate = REV_BOND; | |
2668 | } | |
2669 | ||
abe529af BP |
2670 | if (bond_should_send_learning_packets(bundle->bond)) { |
2671 | bundle_send_learning_packets(bundle); | |
2672 | } | |
2673 | } | |
2674 | } | |
2675 | ||
2676 | static void | |
2677 | bundle_wait(struct ofbundle *bundle) | |
2678 | { | |
2679 | if (bundle->lacp) { | |
2680 | lacp_wait(bundle->lacp); | |
2681 | } | |
2682 | if (bundle->bond) { | |
2683 | bond_wait(bundle->bond); | |
2684 | } | |
2685 | } | |
2686 | \f | |
2687 | /* Mirrors. */ | |
2688 | ||
2689 | static int | |
ec7ceaed EJ |
2690 | mirror_set__(struct ofproto *ofproto_, void *aux, |
2691 | const struct ofproto_mirror_settings *s) | |
abe529af BP |
2692 | { |
2693 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ec7ceaed EJ |
2694 | struct ofbundle **srcs, **dsts; |
2695 | int error; | |
2696 | size_t i; | |
abe529af | 2697 | |
abe529af | 2698 | if (!s) { |
ec7ceaed | 2699 | mirror_destroy(ofproto->mbridge, aux); |
abe529af BP |
2700 | return 0; |
2701 | } | |
2702 | ||
ec7ceaed EJ |
2703 | srcs = xmalloc(s->n_srcs * sizeof *srcs); |
2704 | dsts = xmalloc(s->n_dsts * sizeof *dsts); | |
abe529af | 2705 | |
ec7ceaed EJ |
2706 | for (i = 0; i < s->n_srcs; i++) { |
2707 | srcs[i] = bundle_lookup(ofproto, s->srcs[i]); | |
abe529af BP |
2708 | } |
2709 | ||
ec7ceaed EJ |
2710 | for (i = 0; i < s->n_dsts; i++) { |
2711 | dsts[i] = bundle_lookup(ofproto, s->dsts[i]); | |
abe529af BP |
2712 | } |
2713 | ||
ec7ceaed EJ |
2714 | error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts, |
2715 | s->n_dsts, s->src_vlans, | |
2716 | bundle_lookup(ofproto, s->out_bundle), s->out_vlan); | |
2717 | free(srcs); | |
2718 | free(dsts); | |
2719 | return error; | |
abe529af BP |
2720 | } |
2721 | ||
9d24de3b | 2722 | static int |
ec7ceaed EJ |
2723 | mirror_get_stats__(struct ofproto *ofproto, void *aux, |
2724 | uint64_t *packets, uint64_t *bytes) | |
9d24de3b | 2725 | { |
8844e035 | 2726 | push_all_stats(); |
ec7ceaed EJ |
2727 | return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets, |
2728 | bytes); | |
9d24de3b JP |
2729 | } |
2730 | ||
abe529af BP |
2731 | static int |
2732 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2733 | { | |
2734 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2735 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
abe529af | 2736 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { |
30618594 | 2737 | mac_learning_flush(ofproto->ml); |
abe529af | 2738 | } |
509c0149 | 2739 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2740 | return 0; |
2741 | } | |
2742 | ||
2743 | static bool | |
b4affc74 | 2744 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2745 | { |
2746 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2747 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
ec7ceaed | 2748 | return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0; |
abe529af | 2749 | } |
8402c74b SS |
2750 | |
2751 | static void | |
b53055f4 | 2752 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2753 | { |
2754 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2755 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2756 | } |
e764773c BP |
2757 | |
2758 | static void | |
c4069512 BP |
2759 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2760 | size_t max_entries) | |
e764773c BP |
2761 | { |
2762 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2763 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
e764773c | 2764 | mac_learning_set_idle_time(ofproto->ml, idle_time); |
c4069512 | 2765 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
509c0149 | 2766 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
e764773c | 2767 | } |
abe529af BP |
2768 | \f |
2769 | /* Ports. */ | |
2770 | ||
46c88433 | 2771 | static struct ofport_dpif * |
4e022ec0 | 2772 | get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
abe529af | 2773 | { |
7df6a8bd BP |
2774 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2775 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2776 | } |
2777 | ||
46c88433 | 2778 | static struct ofport_dpif * |
4e022ec0 | 2779 | get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) |
abe529af | 2780 | { |
7c33b188 JR |
2781 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2782 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2783 | } |
2784 | ||
2785 | static void | |
e1b1d06a JP |
2786 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2787 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2788 | struct dpif_port *dpif_port) |
2789 | { | |
2790 | ofproto_port->name = dpif_port->name; | |
2791 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2792 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2793 | } |
2794 | ||
6cbbf4fa EJ |
2795 | static void |
2796 | ofport_update_peer(struct ofport_dpif *ofport) | |
0a740f48 EJ |
2797 | { |
2798 | const struct ofproto_dpif *ofproto; | |
6cbbf4fa | 2799 | struct dpif_backer *backer; |
161b6042 | 2800 | char *peer_name; |
6cbbf4fa EJ |
2801 | |
2802 | if (!netdev_vport_is_patch(ofport->up.netdev)) { | |
2803 | return; | |
2804 | } | |
2805 | ||
2806 | backer = ofproto_dpif_cast(ofport->up.ofproto)->backer; | |
7894e7da | 2807 | backer->need_revalidate = REV_RECONFIGURE; |
0a740f48 | 2808 | |
6cbbf4fa EJ |
2809 | if (ofport->peer) { |
2810 | ofport->peer->peer = NULL; | |
2811 | ofport->peer = NULL; | |
2812 | } | |
2813 | ||
2814 | peer_name = netdev_vport_patch_peer(ofport->up.netdev); | |
2815 | if (!peer_name) { | |
2816 | return; | |
0a740f48 EJ |
2817 | } |
2818 | ||
2819 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6cbbf4fa EJ |
2820 | struct ofport *peer_ofport; |
2821 | struct ofport_dpif *peer; | |
161b6042 | 2822 | char *peer_peer; |
0a740f48 | 2823 | |
462abef2 EJ |
2824 | if (ofproto->backer != backer) { |
2825 | continue; | |
2826 | } | |
2827 | ||
6cbbf4fa EJ |
2828 | peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name); |
2829 | if (!peer_ofport) { | |
2830 | continue; | |
2831 | } | |
2832 | ||
2833 | peer = ofport_dpif_cast(peer_ofport); | |
2834 | peer_peer = netdev_vport_patch_peer(peer->up.netdev); | |
2835 | if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev), | |
2836 | peer_peer)) { | |
2837 | ofport->peer = peer; | |
2838 | ofport->peer->peer = ofport; | |
0a740f48 | 2839 | } |
161b6042 | 2840 | free(peer_peer); |
6cbbf4fa | 2841 | |
161b6042 | 2842 | break; |
0a740f48 | 2843 | } |
161b6042 | 2844 | free(peer_name); |
0a740f48 EJ |
2845 | } |
2846 | ||
abe529af BP |
2847 | static void |
2848 | port_run(struct ofport_dpif *ofport) | |
2849 | { | |
3e5b3fdb EJ |
2850 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2851 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc | 2852 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2d344ba5 AW |
2853 | bool cfm_enable = false; |
2854 | bool bfd_enable = false; | |
015e08bc | 2855 | |
3e5b3fdb EJ |
2856 | ofport->carrier_seq = carrier_seq; |
2857 | ||
abe529af | 2858 | if (ofport->cfm) { |
4653c558 EJ |
2859 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2860 | ||
2d344ba5 | 2861 | cfm_enable = !cfm_get_fault(ofport->cfm); |
4653c558 EJ |
2862 | |
2863 | if (cfm_opup >= 0) { | |
2d344ba5 | 2864 | cfm_enable = cfm_enable && cfm_opup; |
4653c558 | 2865 | } |
abe529af | 2866 | } |
015e08bc | 2867 | |
ccc09689 | 2868 | if (ofport->bfd) { |
2d344ba5 AW |
2869 | bfd_enable = bfd_forwarding(ofport->bfd); |
2870 | } | |
2871 | ||
2872 | if (ofport->bfd || ofport->cfm) { | |
2873 | enable = enable && (cfm_enable || bfd_enable); | |
ccc09689 EJ |
2874 | } |
2875 | ||
015e08bc EJ |
2876 | if (ofport->bundle) { |
2877 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2878 | if (carrier_changed) { |
2879 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2880 | } | |
015e08bc EJ |
2881 | } |
2882 | ||
daff3353 EJ |
2883 | if (ofport->may_enable != enable) { |
2884 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
83341407 | 2885 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
2886 | } |
2887 | ||
015e08bc | 2888 | ofport->may_enable = enable; |
abe529af BP |
2889 | } |
2890 | ||
abe529af BP |
2891 | static int |
2892 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2893 | struct ofproto_port *ofproto_port) | |
2894 | { | |
2895 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2896 | struct dpif_port dpif_port; | |
2897 | int error; | |
2898 | ||
0a740f48 EJ |
2899 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
2900 | const char *type = netdev_get_type_from_name(devname); | |
2901 | ||
2902 | /* We may be called before ofproto->up.port_by_name is populated with | |
2903 | * the appropriate ofport. For this reason, we must get the name and | |
2904 | * type from the netdev layer directly. */ | |
2905 | if (type) { | |
2906 | const struct ofport *ofport; | |
2907 | ||
2908 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
2909 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
2910 | ofproto_port->name = xstrdup(devname); | |
2911 | ofproto_port->type = xstrdup(type); | |
2912 | return 0; | |
2913 | } | |
2914 | return ENODEV; | |
2915 | } | |
2916 | ||
acf60855 JP |
2917 | if (!sset_contains(&ofproto->ports, devname)) { |
2918 | return ENODEV; | |
2919 | } | |
2920 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
2921 | devname, &dpif_port); | |
abe529af | 2922 | if (!error) { |
e1b1d06a | 2923 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
2924 | } |
2925 | return error; | |
2926 | } | |
2927 | ||
2928 | static int | |
e1b1d06a | 2929 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
2930 | { |
2931 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 2932 | const char *devname = netdev_get_name(netdev); |
3aa30359 BP |
2933 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
2934 | const char *dp_port_name; | |
abe529af | 2935 | |
0a740f48 EJ |
2936 | if (netdev_vport_is_patch(netdev)) { |
2937 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
2938 | return 0; | |
2939 | } | |
2940 | ||
3aa30359 | 2941 | dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
b9ad7294 | 2942 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
4e022ec0 | 2943 | odp_port_t port_no = ODPP_NONE; |
7d82ab2e KM |
2944 | int error; |
2945 | ||
2946 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
2947 | if (error) { |
2948 | return error; | |
2949 | } | |
7d82ab2e | 2950 | if (netdev_get_tunnel_config(netdev)) { |
4e022ec0 AW |
2951 | simap_put(&ofproto->backer->tnl_backers, |
2952 | dp_port_name, odp_to_u32(port_no)); | |
7d82ab2e | 2953 | } |
acf60855 | 2954 | } |
b9ad7294 EJ |
2955 | |
2956 | if (netdev_get_tunnel_config(netdev)) { | |
2957 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
2958 | } else { |
2959 | sset_add(&ofproto->ports, devname); | |
2960 | } | |
2961 | return 0; | |
2962 | } | |
2963 | ||
abe529af | 2964 | static int |
4e022ec0 | 2965 | port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) |
abe529af BP |
2966 | { |
2967 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 2968 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 2969 | int error = 0; |
abe529af | 2970 | |
b9ad7294 EJ |
2971 | if (!ofport) { |
2972 | return 0; | |
e1b1d06a | 2973 | } |
b9ad7294 EJ |
2974 | |
2975 | sset_find_and_delete(&ofproto->ghost_ports, | |
2976 | netdev_get_name(ofport->up.netdev)); | |
a614d823 | 2977 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8911c674 | 2978 | if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) { |
b9ad7294 EJ |
2979 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
2980 | if (!error) { | |
abe529af BP |
2981 | /* The caller is going to close ofport->up.netdev. If this is a |
2982 | * bonded port, then the bond is using that netdev, so remove it | |
2983 | * from the bond. The client will need to reconfigure everything | |
2984 | * after deleting ports, so then the slave will get re-added. */ | |
2985 | bundle_remove(&ofport->up); | |
2986 | } | |
2987 | } | |
2988 | return error; | |
2989 | } | |
2990 | ||
6527c598 PS |
2991 | static int |
2992 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2993 | { | |
2994 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2995 | int error; | |
2996 | ||
8844e035 EJ |
2997 | push_all_stats(); |
2998 | ||
6527c598 PS |
2999 | error = netdev_get_stats(ofport->up.netdev, stats); |
3000 | ||
ee382d89 | 3001 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3002 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3003 | ||
0da3c61b | 3004 | ovs_mutex_lock(&ofproto->stats_mutex); |
6527c598 PS |
3005 | /* ofproto->stats.tx_packets represents packets that we created |
3006 | * internally and sent to some port (e.g. packets sent with | |
91d6cd12 AW |
3007 | * ofproto_dpif_send_packet()). Account for them as if they had |
3008 | * come from OFPP_LOCAL and got forwarded. */ | |
6527c598 PS |
3009 | |
3010 | if (stats->rx_packets != UINT64_MAX) { | |
3011 | stats->rx_packets += ofproto->stats.tx_packets; | |
3012 | } | |
3013 | ||
3014 | if (stats->rx_bytes != UINT64_MAX) { | |
3015 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3016 | } | |
3017 | ||
3018 | /* ofproto->stats.rx_packets represents packets that were received on | |
3019 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3020 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3021 | |
3022 | if (stats->tx_packets != UINT64_MAX) { | |
3023 | stats->tx_packets += ofproto->stats.rx_packets; | |
3024 | } | |
3025 | ||
3026 | if (stats->tx_bytes != UINT64_MAX) { | |
3027 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3028 | } | |
0da3c61b | 3029 | ovs_mutex_unlock(&ofproto->stats_mutex); |
6527c598 PS |
3030 | } |
3031 | ||
3032 | return error; | |
3033 | } | |
3034 | ||
abe529af | 3035 | struct port_dump_state { |
acf60855 JP |
3036 | uint32_t bucket; |
3037 | uint32_t offset; | |
0a740f48 | 3038 | bool ghost; |
da78d43d BP |
3039 | |
3040 | struct ofproto_port port; | |
3041 | bool has_port; | |
abe529af BP |
3042 | }; |
3043 | ||
3044 | static int | |
acf60855 | 3045 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3046 | { |
0a740f48 | 3047 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3048 | return 0; |
3049 | } | |
3050 | ||
3051 | static int | |
b9ad7294 | 3052 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3053 | struct ofproto_port *port) |
3054 | { | |
e1b1d06a | 3055 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3056 | struct port_dump_state *state = state_; |
0a740f48 | 3057 | const struct sset *sset; |
acf60855 | 3058 | struct sset_node *node; |
abe529af | 3059 | |
da78d43d BP |
3060 | if (state->has_port) { |
3061 | ofproto_port_destroy(&state->port); | |
3062 | state->has_port = false; | |
3063 | } | |
0a740f48 EJ |
3064 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3065 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3066 | int error; |
3067 | ||
da78d43d BP |
3068 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3069 | if (!error) { | |
3070 | *port = state->port; | |
3071 | state->has_port = true; | |
3072 | return 0; | |
3073 | } else if (error != ENODEV) { | |
acf60855 JP |
3074 | return error; |
3075 | } | |
abe529af | 3076 | } |
acf60855 | 3077 | |
0a740f48 EJ |
3078 | if (!state->ghost) { |
3079 | state->ghost = true; | |
3080 | state->bucket = 0; | |
3081 | state->offset = 0; | |
3082 | return port_dump_next(ofproto_, state_, port); | |
3083 | } | |
3084 | ||
acf60855 | 3085 | return EOF; |
abe529af BP |
3086 | } |
3087 | ||
3088 | static int | |
3089 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3090 | { | |
3091 | struct port_dump_state *state = state_; | |
3092 | ||
da78d43d BP |
3093 | if (state->has_port) { |
3094 | ofproto_port_destroy(&state->port); | |
3095 | } | |
abe529af BP |
3096 | free(state); |
3097 | return 0; | |
3098 | } | |
3099 | ||
3100 | static int | |
3101 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3102 | { | |
3103 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3104 | |
3105 | if (ofproto->port_poll_errno) { | |
3106 | int error = ofproto->port_poll_errno; | |
3107 | ofproto->port_poll_errno = 0; | |
3108 | return error; | |
3109 | } | |
3110 | ||
3111 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3112 | return EAGAIN; | |
3113 | } | |
3114 | ||
3115 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3116 | return 0; | |
abe529af BP |
3117 | } |
3118 | ||
3119 | static void | |
3120 | port_poll_wait(const struct ofproto *ofproto_) | |
3121 | { | |
3122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3123 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3124 | } |
3125 | ||
3126 | static int | |
3127 | port_is_lacp_current(const struct ofport *ofport_) | |
3128 | { | |
3129 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3130 | return (ofport->bundle && ofport->bundle->lacp | |
3131 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3132 | : -1); | |
3133 | } | |
3134 | \f | |
3135 | /* Upcall handling. */ | |
3136 | ||
501f8d1f | 3137 | struct flow_miss_op { |
c2b565b5 | 3138 | struct dpif_op dpif_op; |
bbafd73b EJ |
3139 | |
3140 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3141 | struct xlate_out xout; | |
3142 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
9b1a48c2 | 3143 | |
d445cc16 JP |
3144 | struct ofpbuf mask; /* Flow mask for "put" ops. */ |
3145 | struct odputil_keybuf maskbuf; | |
3146 | ||
9b1a48c2 JP |
3147 | /* If this is a "put" op, then a pointer to the subfacet that should |
3148 | * be marked as uninstalled if the operation fails. */ | |
3149 | struct subfacet *subfacet; | |
501f8d1f BP |
3150 | }; |
3151 | ||
9d6ac44e | 3152 | /* Figures out whether a flow that missed in 'ofproto', whose details are in |
bcd2633a JP |
3153 | * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace |
3154 | * and (usually) installing a datapath flow. The answer is usually "yes" (a | |
3155 | * return value of true). However, for short flows the cost of bookkeeping is | |
3156 | * much higher than the benefits, so when the datapath holds a large number of | |
3157 | * flows we impose some heuristics to decide which flows are likely to be worth | |
3158 | * tracking. */ | |
9d6ac44e | 3159 | static bool |
e1ec7dd4 | 3160 | flow_miss_should_make_facet(struct flow_miss *miss) |
9d6ac44e | 3161 | { |
04d08d54 | 3162 | struct dpif_backer *backer = miss->ofproto->backer; |
bcd2633a | 3163 | uint32_t hash; |
04d08d54 | 3164 | |
7155fa52 JS |
3165 | switch (flow_miss_model) { |
3166 | case OFPROTO_HANDLE_MISS_AUTO: | |
3167 | break; | |
3168 | case OFPROTO_HANDLE_MISS_WITH_FACETS: | |
3169 | return true; | |
3170 | case OFPROTO_HANDLE_MISS_WITHOUT_FACETS: | |
3171 | return false; | |
3172 | } | |
3173 | ||
04d08d54 | 3174 | if (!backer->governor) { |
9d6ac44e BP |
3175 | size_t n_subfacets; |
3176 | ||
04d08d54 | 3177 | n_subfacets = hmap_count(&backer->subfacets); |
380f49c4 | 3178 | if (n_subfacets * 2 <= flow_eviction_threshold) { |
9d6ac44e BP |
3179 | return true; |
3180 | } | |
3181 | ||
c985ec94 | 3182 | backer->governor = governor_create(); |
9d6ac44e BP |
3183 | } |
3184 | ||
e1ec7dd4 | 3185 | hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0); |
04d08d54 | 3186 | return governor_should_install_flow(backer->governor, hash, |
15e34f24 | 3187 | miss->stats.n_packets); |
9d6ac44e BP |
3188 | } |
3189 | ||
9d6ac44e | 3190 | /* Handles 'miss', which matches 'facet'. May add any required datapath |
459b16a1 BP |
3191 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3192 | * | |
e1ec7dd4 EJ |
3193 | * All of the packets in 'miss' are considered to have arrived at time |
3194 | * 'miss->stats.used'. This is really important only for new facets: if we | |
3195 | * just called time_msec() here, then the new subfacet or its packets could | |
3196 | * look (occasionally) as though it was used some time after the facet was | |
3197 | * used. That can make a one-packet flow look like it has a nonzero duration, | |
3198 | * which looks odd in e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3199 | static void |
3200 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
3201 | struct flow_miss_op *ops, size_t *n_ops) | |
3202 | { | |
6a7e895f | 3203 | enum subfacet_path want_path; |
9d6ac44e | 3204 | struct subfacet *subfacet; |
67d91f78 | 3205 | |
e1ec7dd4 EJ |
3206 | facet->packet_count += miss->stats.n_packets; |
3207 | facet->prev_packet_count += miss->stats.n_packets; | |
3208 | facet->byte_count += miss->stats.n_bytes; | |
3209 | facet->prev_byte_count += miss->stats.n_bytes; | |
501f8d1f | 3210 | |
e1ec7dd4 | 3211 | want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
501f8d1f | 3212 | |
a41d7bf2 JP |
3213 | /* Don't install the flow if it's the result of the "userspace" |
3214 | * action for an already installed facet. This can occur when a | |
3215 | * datapath flow with wildcards has a "userspace" action and flows | |
3216 | * sent to userspace result in a different subfacet, which will then | |
3217 | * be rejected as overlapping by the datapath. */ | |
3218 | if (miss->upcall_type == DPIF_UC_ACTION | |
3219 | && !list_is_empty(&facet->subfacets)) { | |
a41d7bf2 JP |
3220 | return; |
3221 | } | |
3222 | ||
e1ec7dd4 | 3223 | subfacet = subfacet_create(facet, miss); |
24ce4b08 | 3224 | if (subfacet->path != want_path) { |
501f8d1f | 3225 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3226 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3227 | |
c84451a6 EJ |
3228 | subfacet->path = want_path; |
3229 | ||
d445cc16 | 3230 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
6b35e630 JP |
3231 | if (enable_megaflows) { |
3232 | odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks, | |
3233 | &miss->flow, UINT32_MAX); | |
3234 | } | |
d445cc16 | 3235 | |
bbafd73b | 3236 | op->xout_garbage = false; |
c2b565b5 | 3237 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
9b1a48c2 | 3238 | op->subfacet = subfacet; |
0f74f625 | 3239 | put->flags = DPIF_FP_CREATE; |
501f8d1f BP |
3240 | put->key = miss->key; |
3241 | put->key_len = miss->key_len; | |
d445cc16 JP |
3242 | put->mask = op->mask.data; |
3243 | put->mask_len = op->mask.size; | |
3244 | ||
6a7e895f | 3245 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3246 | put->actions = facet->xout.odp_actions.data; |
3247 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3248 | } else { |
b7e2f6b3 | 3249 | compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow, |
bbafd73b | 3250 | op->slow_stub, sizeof op->slow_stub, |
6a7e895f BP |
3251 | &put->actions, &put->actions_len); |
3252 | } | |
501f8d1f BP |
3253 | put->stats = NULL; |
3254 | } | |
3255 | } | |
3256 | ||
acf60855 JP |
3257 | /* Handles flow miss 'miss'. May add any required datapath operations |
3258 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3259 | static void |
acf60855 JP |
3260 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3261 | size_t *n_ops) | |
9d6ac44e BP |
3262 | { |
3263 | struct facet *facet; | |
9d6ac44e | 3264 | |
15e34f24 | 3265 | miss->ofproto->n_missed += miss->stats.n_packets; |
9d6ac44e | 3266 | |
e1ec7dd4 | 3267 | facet = facet_lookup_valid(miss->ofproto, &miss->flow); |
9d6ac44e | 3268 | if (!facet) { |
ba33dd03 EJ |
3269 | /* There does not exist a bijection between 'struct flow' and datapath |
3270 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3271 | * assumption used throughout the facet and subfacet handling code. | |
3272 | * Since we have to handle these misses in userspace anyway, we simply | |
ec9f40dc | 3273 | * skip facet creation, avoiding the problem altogether. */ |
ba33dd03 | 3274 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE |
e1ec7dd4 | 3275 | || !flow_miss_should_make_facet(miss)) { |
9d6ac44e BP |
3276 | return; |
3277 | } | |
3278 | ||
e1ec7dd4 | 3279 | facet = facet_create(miss); |
9d6ac44e | 3280 | } |
e1ec7dd4 | 3281 | handle_flow_miss_with_facet(miss, facet, ops, n_ops); |
9d6ac44e BP |
3282 | } |
3283 | ||
8f73d537 EJ |
3284 | static struct drop_key * |
3285 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3286 | size_t key_len) | |
3287 | { | |
3288 | struct drop_key *drop_key; | |
3289 | ||
3290 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3291 | &backer->drop_keys) { | |
3292 | if (drop_key->key_len == key_len | |
3293 | && !memcmp(drop_key->key, key, key_len)) { | |
3294 | return drop_key; | |
3295 | } | |
3296 | } | |
3297 | return NULL; | |
3298 | } | |
3299 | ||
3300 | static void | |
3301 | drop_key_clear(struct dpif_backer *backer) | |
3302 | { | |
3303 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3304 | struct drop_key *drop_key, *next; | |
3305 | ||
3306 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3307 | int error; | |
3308 | ||
3309 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3310 | NULL); | |
3311 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3312 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3313 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
10a89ef0 BP |
3314 | VLOG_WARN("Failed to delete drop key (%s) (%s)", |
3315 | ovs_strerror(error), ds_cstr(&ds)); | |
8f73d537 EJ |
3316 | ds_destroy(&ds); |
3317 | } | |
3318 | ||
3319 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
e1ec7dd4 | 3320 | drop_key_destroy(drop_key); |
8f73d537 | 3321 | } |
e1ec7dd4 EJ |
3322 | |
3323 | udpif_drop_key_clear(backer->udpif); | |
8f73d537 EJ |
3324 | } |
3325 | ||
501f8d1f | 3326 | static void |
e1ec7dd4 | 3327 | handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb) |
501f8d1f | 3328 | { |
e1ec7dd4 EJ |
3329 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH]; |
3330 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH]; | |
b23cdad9 | 3331 | struct flow_miss *miss; |
e1ec7dd4 | 3332 | size_t n_ops, i; |
501f8d1f BP |
3333 | |
3334 | /* Process each element in the to-do list, constructing the set of | |
3335 | * operations to batch. */ | |
3336 | n_ops = 0; | |
e1ec7dd4 | 3337 | HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) { |
acf60855 | 3338 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3339 | } |
cb22974d | 3340 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3341 | |
3342 | /* Execute batch. */ | |
3343 | for (i = 0; i < n_ops; i++) { | |
3344 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3345 | } | |
acf60855 | 3346 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 3347 | |
501f8d1f | 3348 | for (i = 0; i < n_ops; i++) { |
9b1a48c2 JP |
3349 | if (dpif_ops[i]->error != 0 |
3350 | && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT | |
3351 | && flow_miss_ops[i].subfacet) { | |
3352 | struct subfacet *subfacet = flow_miss_ops[i].subfacet; | |
3353 | ||
3354 | COVERAGE_INC(subfacet_install_fail); | |
3355 | ||
cfd5c9eb JP |
3356 | /* Zero-out subfacet counters when installation failed, but |
3357 | * datapath reported hits. This should not happen and | |
3358 | * indicates a bug, since if the datapath flow exists, we | |
3359 | * should not be attempting to create a new subfacet. A | |
3360 | * buggy datapath could trigger this, so just zero out the | |
3361 | * counters and log an error. */ | |
3362 | if (subfacet->dp_packet_count || subfacet->dp_byte_count) { | |
3363 | VLOG_ERR_RL(&rl, "failed to install subfacet for which " | |
3364 | "datapath reported hits"); | |
3365 | subfacet->dp_packet_count = subfacet->dp_byte_count = 0; | |
3366 | } | |
3367 | ||
9b1a48c2 JP |
3368 | subfacet->path = SF_NOT_INSTALLED; |
3369 | } | |
6a7e895f BP |
3370 | } |
3371 | } | |
3372 | ||
e1ec7dd4 EJ |
3373 | static void |
3374 | handle_upcalls(struct dpif_backer *backer) | |
6ff686f2 | 3375 | { |
e1ec7dd4 | 3376 | struct flow_miss_batch *fmb; |
90a7c55e | 3377 | int n_processed; |
abe529af | 3378 | |
e1ec7dd4 EJ |
3379 | for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) { |
3380 | struct drop_key *drop_key = drop_key_next(backer->udpif); | |
3381 | if (!drop_key) { | |
3382 | break; | |
3383 | } | |
3384 | ||
3385 | if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) { | |
3386 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3387 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3388 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3389 | drop_key->key, drop_key->key_len, | |
3390 | NULL, 0, NULL, 0, NULL); | |
3391 | } else { | |
3392 | drop_key_destroy(drop_key); | |
3393 | } | |
90a7c55e | 3394 | } |
9b16c439 | 3395 | |
e1ec7dd4 EJ |
3396 | fmb = flow_miss_batch_next(backer->udpif); |
3397 | if (fmb) { | |
3398 | handle_flow_misses(backer, fmb); | |
3399 | flow_miss_batch_destroy(fmb); | |
3400 | } | |
abe529af BP |
3401 | } |
3402 | \f | |
3403 | /* Flow expiration. */ | |
3404 | ||
04d08d54 | 3405 | static int subfacet_max_idle(const struct dpif_backer *); |
acf60855 | 3406 | static void update_stats(struct dpif_backer *); |
15aaf599 | 3407 | static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex); |
04d08d54 | 3408 | static void expire_subfacets(struct dpif_backer *, int dp_max_idle); |
abe529af BP |
3409 | |
3410 | /* This function is called periodically by run(). Its job is to collect | |
3411 | * updates for the flows that have been installed into the datapath, most | |
3412 | * importantly when they last were used, and then use that information to | |
3413 | * expire flows that have not been used recently. | |
3414 | * | |
3415 | * Returns the number of milliseconds after which it should be called again. */ | |
3416 | static int | |
acf60855 | 3417 | expire(struct dpif_backer *backer) |
abe529af | 3418 | { |
acf60855 | 3419 | struct ofproto_dpif *ofproto; |
dc54ef36 | 3420 | size_t n_subfacets; |
04d08d54 | 3421 | int max_idle; |
abe529af | 3422 | |
8f73d537 EJ |
3423 | /* Periodically clear out the drop keys in an effort to keep them |
3424 | * relatively few. */ | |
3425 | drop_key_clear(backer); | |
3426 | ||
acf60855 JP |
3427 | /* Update stats for each flow in the backer. */ |
3428 | update_stats(backer); | |
abe529af | 3429 | |
04d08d54 EJ |
3430 | n_subfacets = hmap_count(&backer->subfacets); |
3431 | if (n_subfacets) { | |
9fc0165a | 3432 | struct subfacet *subfacet; |
04d08d54 | 3433 | long long int total, now; |
abe529af | 3434 | |
04d08d54 EJ |
3435 | total = 0; |
3436 | now = time_msec(); | |
3437 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { | |
3438 | total += now - subfacet->created; | |
acf60855 | 3439 | } |
04d08d54 EJ |
3440 | backer->avg_subfacet_life += total / n_subfacets; |
3441 | } | |
3442 | backer->avg_subfacet_life /= 2; | |
0697b5c3 | 3443 | |
04d08d54 EJ |
3444 | backer->avg_n_subfacet += n_subfacets; |
3445 | backer->avg_n_subfacet /= 2; | |
3446 | ||
3447 | backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); | |
655ab909 | 3448 | |
04d08d54 EJ |
3449 | max_idle = subfacet_max_idle(backer); |
3450 | expire_subfacets(backer, max_idle); | |
acf60855 | 3451 | |
04d08d54 EJ |
3452 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
3453 | struct rule *rule, *next_rule; | |
3454 | ||
3455 | if (ofproto->backer != backer) { | |
3456 | continue; | |
3457 | } | |
acf60855 JP |
3458 | |
3459 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
3460 | * has passed. */ | |
abe7b10f | 3461 | ovs_mutex_lock(&ofproto_mutex); |
e503cc19 SH |
3462 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
3463 | &ofproto->up.expirable) { | |
3464 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 3465 | } |
abe7b10f | 3466 | ovs_mutex_unlock(&ofproto_mutex); |
abe529af | 3467 | |
acf60855 JP |
3468 | /* All outstanding data in existing flows has been accounted, so it's a |
3469 | * good time to do bond rebalancing. */ | |
3470 | if (ofproto->has_bonded_bundles) { | |
3471 | struct ofbundle *bundle; | |
abe529af | 3472 | |
acf60855 JP |
3473 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
3474 | if (bundle->bond) { | |
4a1b8f30 | 3475 | bond_rebalance(bundle->bond); |
acf60855 | 3476 | } |
abe529af BP |
3477 | } |
3478 | } | |
3479 | } | |
3480 | ||
acf60855 | 3481 | return MIN(max_idle, 1000); |
abe529af BP |
3482 | } |
3483 | ||
a218c879 BP |
3484 | /* Updates flow table statistics given that the datapath just reported 'stats' |
3485 | * as 'subfacet''s statistics. */ | |
3486 | static void | |
3487 | update_subfacet_stats(struct subfacet *subfacet, | |
3488 | const struct dpif_flow_stats *stats) | |
3489 | { | |
3490 | struct facet *facet = subfacet->facet; | |
9dfb1f78 EJ |
3491 | struct dpif_flow_stats diff; |
3492 | ||
3493 | diff.tcp_flags = stats->tcp_flags; | |
3494 | diff.used = stats->used; | |
a218c879 BP |
3495 | |
3496 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
9dfb1f78 | 3497 | diff.n_packets = stats->n_packets - subfacet->dp_packet_count; |
a218c879 BP |
3498 | } else { |
3499 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
9dfb1f78 | 3500 | diff.n_packets = 0; |
a218c879 BP |
3501 | } |
3502 | ||
3503 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
9dfb1f78 | 3504 | diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; |
a218c879 BP |
3505 | } else { |
3506 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
9dfb1f78 | 3507 | diff.n_bytes = 0; |
a218c879 BP |
3508 | } |
3509 | ||
b7e2f6b3 | 3510 | facet->ofproto->n_hit += diff.n_packets; |
a218c879 BP |
3511 | subfacet->dp_packet_count = stats->n_packets; |
3512 | subfacet->dp_byte_count = stats->n_bytes; | |
9dfb1f78 | 3513 | subfacet_update_stats(subfacet, &diff); |
a218c879 | 3514 | |
a218c879 BP |
3515 | if (facet->accounted_bytes < facet->byte_count) { |
3516 | facet_learn(facet); | |
3517 | facet_account(facet); | |
3518 | facet->accounted_bytes = facet->byte_count; | |
3519 | } | |
a218c879 BP |
3520 | } |
3521 | ||
3522 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
3523 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
3524 | static void | |
04d08d54 | 3525 | delete_unexpected_flow(struct dpif_backer *backer, |
a218c879 BP |
3526 | const struct nlattr *key, size_t key_len) |
3527 | { | |
3528 | if (!VLOG_DROP_WARN(&rl)) { | |
3529 | struct ds s; | |
3530 | ||
3531 | ds_init(&s); | |
3532 | odp_flow_key_format(key, key_len, &s); | |
04d08d54 | 3533 | VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); |
a218c879 BP |
3534 | ds_destroy(&s); |
3535 | } | |
3536 | ||
3537 | COVERAGE_INC(facet_unexpected); | |
04d08d54 | 3538 | dpif_flow_del(backer->dpif, key, key_len, NULL); |
a218c879 BP |
3539 | } |
3540 | ||
abe529af BP |
3541 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
3542 | * | |
3543 | * This function also pushes statistics updates to rules which each facet | |
3544 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3545 | * facet changes the rule it resubmits into at some time in between | |
3546 | * update_stats() runs, it is possible that statistics accrued to the | |
3547 | * old rule will be incorrectly attributed to the new rule. This could be | |
3548 | * avoided by calling update_stats() whenever rules are created or | |
3549 | * deleted. However, the performance impact of making so many calls to the | |
3550 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
3551 | * |
3552 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
3553 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
3554 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
3555 | * the hit or miss counts of br1. | |
abe529af BP |
3556 | */ |
3557 | static void | |
acf60855 | 3558 | update_stats(struct dpif_backer *backer) |
abe529af BP |
3559 | { |
3560 | const struct dpif_flow_stats *stats; | |
3561 | struct dpif_flow_dump dump; | |
61fb711d JP |
3562 | const struct nlattr *key, *mask; |
3563 | size_t key_len, mask_len; | |
abe529af | 3564 | |
acf60855 | 3565 | dpif_flow_dump_start(&dump, backer->dpif); |
e6cc0bab | 3566 | while (dpif_flow_dump_next(&dump, &key, &key_len, |
61fb711d | 3567 | &mask, &mask_len, NULL, NULL, &stats)) { |
b0f7b9b5 | 3568 | struct subfacet *subfacet; |
acf60855 | 3569 | uint32_t key_hash; |
abe529af | 3570 | |
acf60855 | 3571 | key_hash = odp_flow_key_hash(key, key_len); |
04d08d54 | 3572 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
6a7e895f BP |
3573 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
3574 | case SF_FAST_PATH: | |
a218c879 | 3575 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
3576 | break; |
3577 | ||
3578 | case SF_SLOW_PATH: | |
3579 | /* Stats are updated per-packet. */ | |
3580 | break; | |
3581 | ||
3582 | case SF_NOT_INSTALLED: | |
3583 | default: | |
04d08d54 | 3584 | delete_unexpected_flow(backer, key, key_len); |
6a7e895f | 3585 | break; |
abe529af | 3586 | } |
8fa4d1d0 | 3587 | run_fast_rl(); |
abe529af BP |
3588 | } |
3589 | dpif_flow_dump_done(&dump); | |
3590 | } | |
3591 | ||
3592 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3593 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3594 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3595 | * fold its statistic into its rule. */ | |
abe529af | 3596 | static int |
04d08d54 | 3597 | subfacet_max_idle(const struct dpif_backer *backer) |
abe529af BP |
3598 | { |
3599 | /* | |
3600 | * Idle time histogram. | |
3601 | * | |
b0f7b9b5 BP |
3602 | * Most of the time a switch has a relatively small number of subfacets. |
3603 | * When this is the case we might as well keep statistics for all of them | |
3604 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3605 | * well. |
3606 | * | |
b0f7b9b5 | 3607 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3608 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3609 | * significant. However, with a large number of subfacets it is likely |
3610 | * that only a few of them are "heavy hitters" that consume a large amount | |
3611 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3612 | * the kernel and maintaining in userspaces; other subfacets we can | |
3613 | * discard. | |
abe529af BP |
3614 | * |
3615 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3616 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3617 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3618 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3619 | * the most-recently-used 1% of subfacets (but at least |
380f49c4 | 3620 | * flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3621 | * the most-recently-used bucket of subfacets is kept, so actually an |
3622 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3623 | * (though the next run will delete most of those unless they receive |
3624 | * additional data). | |
abe529af | 3625 | * |
b0f7b9b5 BP |
3626 | * This requires a second pass through the subfacets, in addition to the |
3627 | * pass made by update_stats(), because the former function never looks at | |
3628 | * uninstallable subfacets. | |
abe529af | 3629 | */ |
31ef9f51 | 3630 | enum { BUCKET_WIDTH = 100 }; |
abe529af BP |
3631 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; |
3632 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3633 | int total, subtotal, bucket; |
b0f7b9b5 | 3634 | struct subfacet *subfacet; |
abe529af BP |
3635 | long long int now; |
3636 | int i; | |
3637 | ||
04d08d54 | 3638 | total = hmap_count(&backer->subfacets); |
380f49c4 | 3639 | if (total <= flow_eviction_threshold) { |
abe529af BP |
3640 | return N_BUCKETS * BUCKET_WIDTH; |
3641 | } | |
3642 | ||
3643 | /* Build histogram. */ | |
3644 | now = time_msec(); | |
04d08d54 | 3645 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { |
b0f7b9b5 | 3646 | long long int idle = now - subfacet->used; |
abe529af BP |
3647 | int bucket = (idle <= 0 ? 0 |
3648 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3649 | : (unsigned int) idle / BUCKET_WIDTH); | |
3650 | buckets[bucket]++; | |
3651 | } | |
3652 | ||
3653 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3654 | subtotal = bucket = 0; |
3655 | do { | |
3656 | subtotal += buckets[bucket++]; | |
084f5290 | 3657 | } while (bucket < N_BUCKETS && |
380f49c4 | 3658 | subtotal < MAX(flow_eviction_threshold, total / 100)); |
abe529af BP |
3659 | |
3660 | if (VLOG_IS_DBG_ENABLED()) { | |
3661 | struct ds s; | |
3662 | ||
3663 | ds_init(&s); | |
3664 | ds_put_cstr(&s, "keep"); | |
3665 | for (i = 0; i < N_BUCKETS; i++) { | |
3666 | if (i == bucket) { | |
3667 | ds_put_cstr(&s, ", drop"); | |
3668 | } | |
3669 | if (buckets[i]) { | |
3670 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3671 | } | |
3672 | } | |
04d08d54 | 3673 | VLOG_INFO("%s (msec:count)", ds_cstr(&s)); |
abe529af BP |
3674 | ds_destroy(&s); |
3675 | } | |
3676 | ||
3677 | return bucket * BUCKET_WIDTH; | |
3678 | } | |
3679 | ||
abe529af | 3680 | static void |
04d08d54 | 3681 | expire_subfacets(struct dpif_backer *backer, int dp_max_idle) |
abe529af | 3682 | { |
625b0720 BP |
3683 | /* Cutoff time for most flows. */ |
3684 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
3685 | ||
3686 | /* We really want to keep flows for special protocols around, so use a more | |
3687 | * conservative cutoff. */ | |
3688 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 3689 | |
b0f7b9b5 | 3690 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 3691 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 3692 | int n_batch; |
abe529af | 3693 | |
b99d3cee | 3694 | n_batch = 0; |
b0f7b9b5 | 3695 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
04d08d54 | 3696 | &backer->subfacets) { |
625b0720 BP |
3697 | long long int cutoff; |
3698 | ||
bbafd73b EJ |
3699 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
3700 | | SLOW_STP) | |
625b0720 BP |
3701 | ? special_cutoff |
3702 | : normal_cutoff); | |
b0f7b9b5 | 3703 | if (subfacet->used < cutoff) { |
6a7e895f | 3704 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 3705 | batch[n_batch++] = subfacet; |
1d85f9e5 | 3706 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
04d08d54 | 3707 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee BP |
3708 | n_batch = 0; |
3709 | } | |
3710 | } else { | |
3711 | subfacet_destroy(subfacet); | |
3712 | } | |
abe529af BP |
3713 | } |
3714 | } | |
b99d3cee BP |
3715 | |
3716 | if (n_batch > 0) { | |
04d08d54 | 3717 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee | 3718 | } |
abe529af BP |
3719 | } |
3720 | ||
3721 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3722 | * then delete it entirely. */ | |
3723 | static void | |
3724 | rule_expire(struct rule_dpif *rule) | |
15aaf599 | 3725 | OVS_REQUIRES(ofproto_mutex) |
abe529af | 3726 | { |
a3779dbc | 3727 | uint16_t idle_timeout, hard_timeout; |
15aaf599 BP |
3728 | long long int now = time_msec(); |
3729 | int reason; | |
abe529af | 3730 | |
35232d32 | 3731 | ovs_assert(!rule->up.pending); |
e2a3d183 | 3732 | |
15aaf599 | 3733 | /* Has 'rule' expired? */ |
d10976b7 | 3734 | ovs_mutex_lock(&rule->up.mutex); |
a3779dbc EJ |
3735 | hard_timeout = rule->up.hard_timeout; |
3736 | idle_timeout = rule->up.idle_timeout; | |
a3779dbc | 3737 | if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) { |
abe529af | 3738 | reason = OFPRR_HARD_TIMEOUT; |
a3779dbc | 3739 | } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) { |
abe529af BP |
3740 | reason = OFPRR_IDLE_TIMEOUT; |
3741 | } else { | |
15aaf599 | 3742 | reason = -1; |
abe529af | 3743 | } |
15aaf599 | 3744 | ovs_mutex_unlock(&rule->up.mutex); |
abe529af | 3745 | |
15aaf599 BP |
3746 | if (reason >= 0) { |
3747 | COVERAGE_INC(ofproto_dpif_expired); | |
3748 | ofproto_rule_expire(&rule->up, reason); | |
3749 | } | |
abe529af BP |
3750 | } |
3751 | \f | |
3752 | /* Facets. */ | |
3753 | ||
4dff9097 | 3754 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
3755 | * |
3756 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 3757 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 3758 | * |
bcd2633a JP |
3759 | * 'rule' and 'xout' must have been created based on 'miss'. |
3760 | * | |
3761 | * 'facet'' statistics are initialized based on 'stats'. | |
2b459b83 | 3762 | * |
b0f7b9b5 BP |
3763 | * The facet will initially have no subfacets. The caller should create (at |
3764 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3765 | static struct facet * |
e1ec7dd4 | 3766 | facet_create(const struct flow_miss *miss) |
abe529af | 3767 | { |
4dff9097 | 3768 | struct ofproto_dpif *ofproto = miss->ofproto; |
abe529af | 3769 | struct facet *facet; |
bcd2633a | 3770 | struct match match; |
abe529af | 3771 | |
264ac358 | 3772 | COVERAGE_INC(facet_create); |
abe529af | 3773 | facet = xzalloc(sizeof *facet); |
b7e2f6b3 | 3774 | facet->ofproto = miss->ofproto; |
e1ec7dd4 | 3775 | facet->used = miss->stats.used; |
4dff9097 | 3776 | facet->flow = miss->flow; |
4dff9097 EJ |
3777 | facet->learn_rl = time_msec() + 500; |
3778 | ||
b0f7b9b5 | 3779 | list_init(&facet->subfacets); |
abe529af BP |
3780 | netflow_flow_init(&facet->nf_flow); |
3781 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3782 | ||
e1ec7dd4 | 3783 | xlate_out_copy(&facet->xout, &miss->xout); |
bcd2633a JP |
3784 | |
3785 | match_init(&match, &facet->flow, &facet->xout.wc); | |
3786 | cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); | |
0b4f2078 | 3787 | ovs_rwlock_wrlock(&ofproto->facets.rwlock); |
bcd2633a | 3788 | classifier_insert(&ofproto->facets, &facet->cr); |
0b4f2078 | 3789 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 3790 | |
bbafd73b | 3791 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; |
abe529af BP |
3792 | return facet; |
3793 | } | |
3794 | ||
3795 | static void | |
3796 | facet_free(struct facet *facet) | |
3797 | { | |
4dff9097 | 3798 | if (facet) { |
bbafd73b | 3799 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
3800 | free(facet); |
3801 | } | |
abe529af BP |
3802 | } |
3803 | ||
c4af8a90 AW |
3804 | /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'. |
3805 | * 'flow' must reflect the data in 'packet'. */ | |
419460f4 AW |
3806 | int |
3807 | ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto, | |
3808 | const struct flow *flow, | |
3809 | struct rule_dpif *rule, | |
3810 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
3811 | struct ofpbuf *packet) | |
3d9e05f8 BP |
3812 | { |
3813 | struct odputil_keybuf keybuf; | |
39d007ab BP |
3814 | struct dpif_flow_stats stats; |
3815 | struct xlate_out xout; | |
3816 | struct xlate_in xin; | |
3817 | ofp_port_t in_port; | |
3d9e05f8 | 3818 | struct ofpbuf key; |
39d007ab BP |
3819 | int error; |
3820 | ||
3821 | ovs_assert((rule != NULL) != (ofpacts != NULL)); | |
3822 | ||
3823 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); | |
3824 | if (rule) { | |
3825 | rule_dpif_credit_stats(rule, &stats); | |
3826 | } | |
3827 | ||
3828 | xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); | |
3829 | xin.ofpacts = ofpacts; | |
3830 | xin.ofpacts_len = ofpacts_len; | |
3831 | xin.resubmit_stats = &stats; | |
3832 | xlate_actions(&xin, &xout); | |
3d9e05f8 | 3833 | |
6ff686f2 | 3834 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
39d007ab BP |
3835 | in_port = flow->in_port.ofp_port; |
3836 | if (in_port == OFPP_NONE) { | |
3837 | in_port = OFPP_LOCAL; | |
3838 | } | |
3839 | odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port)); | |
3840 | ||
3841 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, | |
7fd91025 BP |
3842 | xout.odp_actions.data, xout.odp_actions.size, packet, |
3843 | (xout.slow & SLOW_ACTION) != 0); | |
39d007ab | 3844 | xlate_out_uninit(&xout); |
80e5eed9 | 3845 | |
39d007ab | 3846 | return error; |
abe529af BP |
3847 | } |
3848 | ||
bcd2633a | 3849 | /* Remove 'facet' from its ofproto and free up the associated memory: |
abe529af BP |
3850 | * |
3851 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3852 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3853 | * |
3854 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3855 | */ | |
3856 | static void | |
15baa734 | 3857 | facet_remove(struct facet *facet) |
abe529af | 3858 | { |
b0f7b9b5 BP |
3859 | struct subfacet *subfacet, *next_subfacet; |
3860 | ||
264ac358 | 3861 | COVERAGE_INC(facet_remove); |
cb22974d | 3862 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
3863 | |
3864 | /* First uninstall all of the subfacets to get final statistics. */ | |
3865 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3866 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3867 | } |
3868 | ||
3869 | /* Flush the final stats to the rule. | |
3870 | * | |
3871 | * This might require us to have at least one subfacet around so that we | |
3872 | * can use its actions for accounting in facet_account(), which is why we | |
3873 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3874 | facet_flush_stats(facet); |
551a2f6c BP |
3875 | |
3876 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3877 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3878 | &facet->subfacets) { | |
15baa734 | 3879 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3880 | } |
0b4f2078 | 3881 | ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock); |
b7e2f6b3 | 3882 | classifier_remove(&facet->ofproto->facets, &facet->cr); |
0b4f2078 | 3883 | ovs_rwlock_unlock(&facet->ofproto->facets.rwlock); |
bcd2633a | 3884 | cls_rule_destroy(&facet->cr); |
abe529af BP |
3885 | facet_free(facet); |
3886 | } | |
3887 | ||
3de9590b BP |
3888 | /* Feed information from 'facet' back into the learning table to keep it in |
3889 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3890 | static void |
3de9590b | 3891 | facet_learn(struct facet *facet) |
abe529af | 3892 | { |
62bd7349 | 3893 | long long int now = time_msec(); |
abe529af | 3894 | |
bbafd73b | 3895 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
3896 | return; |
3897 | } | |
3898 | ||
62bd7349 | 3899 | facet->learn_rl = now + 500; |
6cf474d7 | 3900 | |
bbafd73b EJ |
3901 | if (!facet->xout.has_learn |
3902 | && !facet->xout.has_normal | |
3903 | && (!facet->xout.has_fin_timeout | |
3de9590b | 3904 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
3905 | return; |
3906 | } | |
abe529af | 3907 | |
9dfb1f78 | 3908 | facet_push_stats(facet, true); |
3de9590b BP |
3909 | } |
3910 | ||
3911 | static void | |
3912 | facet_account(struct facet *facet) | |
3913 | { | |
3de9590b BP |
3914 | const struct nlattr *a; |
3915 | unsigned int left; | |
3916 | ovs_be16 vlan_tci; | |
3917 | uint64_t n_bytes; | |
abe529af | 3918 | |
b7e2f6b3 | 3919 | if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { |
abe529af BP |
3920 | return; |
3921 | } | |
3de9590b | 3922 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3923 | |
3924 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3925 | * as a basis. We also need to track the actual VLAN on which the packet | |
3926 | * is going to be sent to ensure that it matches the one passed to | |
3927 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3928 | * hash bucket.) |
3929 | * | |
3930 | * We use the actions from an arbitrary subfacet because they should all | |
3931 | * be equally valid for our purpose. */ | |
d78be13b | 3932 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
3933 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
3934 | facet->xout.odp_actions.size) { | |
fea393b1 | 3935 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3936 | struct ofport_dpif *port; |
abe529af | 3937 | |
d78be13b | 3938 | switch (nl_attr_type(a)) { |
df2c07f4 | 3939 | case OVS_ACTION_ATTR_OUTPUT: |
b7e2f6b3 | 3940 | port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); |
abe529af | 3941 | if (port && port->bundle && port->bundle->bond) { |
d78be13b | 3942 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3943 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3944 | } |
d78be13b BP |
3945 | break; |
3946 | ||
fea393b1 BP |
3947 | case OVS_ACTION_ATTR_POP_VLAN: |
3948 | vlan_tci = htons(0); | |
d78be13b BP |
3949 | break; |
3950 | ||
fea393b1 BP |
3951 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3952 | vlan = nl_attr_get(a); | |
3953 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3954 | break; |
abe529af BP |
3955 | } |
3956 | } | |
3957 | } | |
3958 | ||
abe529af BP |
3959 | /* Returns true if the only action for 'facet' is to send to the controller. |
3960 | * (We don't report NetFlow expiration messages for such facets because they | |
3961 | * are just part of the control logic for the network, not real traffic). */ | |
3962 | static bool | |
3963 | facet_is_controller_flow(struct facet *facet) | |
3964 | { | |
f25d0cf3 | 3965 | if (facet) { |
b7e2f6b3 | 3966 | struct ofproto_dpif *ofproto = facet->ofproto; |
ad3efdcb | 3967 | const struct ofpact *ofpacts; |
15aaf599 | 3968 | struct rule_actions *actions; |
ad3efdcb EJ |
3969 | struct rule_dpif *rule; |
3970 | size_t ofpacts_len; | |
3971 | bool is_controller; | |
3972 | ||
3973 | rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule); | |
15aaf599 BP |
3974 | actions = rule_dpif_get_actions(rule); |
3975 | rule_dpif_unref(rule); | |
3976 | ||
3977 | ofpacts_len = actions->ofpacts_len; | |
3978 | ofpacts = actions->ofpacts; | |
ad3efdcb EJ |
3979 | is_controller = ofpacts_len > 0 |
3980 | && ofpacts->type == OFPACT_CONTROLLER | |
3981 | && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len); | |
15aaf599 | 3982 | rule_actions_unref(actions); |
6f00e29b | 3983 | |
ad3efdcb | 3984 | return is_controller; |
f25d0cf3 BP |
3985 | } |
3986 | return false; | |
abe529af BP |
3987 | } |
3988 | ||
3989 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3990 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3991 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3992 | * its packet and byte counts before this function is called. */ | |
3993 | static void | |
15baa734 | 3994 | facet_flush_stats(struct facet *facet) |
abe529af | 3995 | { |
b7e2f6b3 | 3996 | struct ofproto_dpif *ofproto = facet->ofproto; |
b0f7b9b5 BP |
3997 | struct subfacet *subfacet; |
3998 | ||
3999 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4000 | ovs_assert(!subfacet->dp_byte_count); |
4001 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4002 | } |
abe529af | 4003 | |
9dfb1f78 | 4004 | facet_push_stats(facet, false); |
3de9590b BP |
4005 | if (facet->accounted_bytes < facet->byte_count) { |
4006 | facet_account(facet); | |
4007 | facet->accounted_bytes = facet->byte_count; | |
4008 | } | |
abe529af BP |
4009 | |
4010 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4011 | struct ofexpired expired; | |
4012 | expired.flow = facet->flow; | |
4013 | expired.packet_count = facet->packet_count; | |
4014 | expired.byte_count = facet->byte_count; | |
4015 | expired.used = facet->used; | |
4016 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4017 | } | |
4018 | ||
abe529af BP |
4019 | /* Reset counters to prevent double counting if 'facet' ever gets |
4020 | * reinstalled. */ | |
bbb5d219 | 4021 | facet_reset_counters(facet); |
abe529af BP |
4022 | |
4023 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4024 | facet->tcp_flags = 0; |
abe529af BP |
4025 | } |
4026 | ||
bcd2633a JP |
4027 | /* Searches 'ofproto''s table of facets for one which would be responsible for |
4028 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4029 | * |
abe529af BP |
4030 | * The returned facet might need revalidation; use facet_lookup_valid() |
4031 | * instead if that is important. */ | |
4032 | static struct facet * | |
bcd2633a | 4033 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4034 | { |
0b4f2078 EJ |
4035 | struct cls_rule *cr; |
4036 | ||
4037 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
4038 | cr = classifier_lookup(&ofproto->facets, flow, NULL); | |
4039 | ovs_rwlock_unlock(&ofproto->facets.rwlock); | |
bcd2633a | 4040 | return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL; |
abe529af BP |
4041 | } |
4042 | ||
bcd2633a JP |
4043 | /* Searches 'ofproto''s table of facets for one capable that covers |
4044 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4045 | * |
abe529af BP |
4046 | * The returned facet is guaranteed to be valid. */ |
4047 | static struct facet * | |
bcd2633a | 4048 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4049 | { |
c57b2226 | 4050 | struct facet *facet; |
abe529af | 4051 | |
bcd2633a | 4052 | facet = facet_find(ofproto, flow); |
abe529af | 4053 | if (facet |
a1616063 | 4054 | && ofproto->backer->need_revalidate |
5bf64ade | 4055 | && !facet_revalidate(facet)) { |
0305ce1f | 4056 | return NULL; |
abe529af BP |
4057 | } |
4058 | ||
4059 | return facet; | |
4060 | } | |
4061 | ||
6814e51f BP |
4062 | static bool |
4063 | facet_check_consistency(struct facet *facet) | |
4064 | { | |
4065 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4066 | ||
bbafd73b EJ |
4067 | struct xlate_out xout; |
4068 | struct xlate_in xin; | |
e1ec7dd4 | 4069 | bool ok; |
6814e51f BP |
4070 | |
4071 | /* Check the datapath actions for consistency. */ | |
10c44245 | 4072 | xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL); |
bbafd73b | 4073 | xlate_actions(&xin, &xout); |
6814e51f | 4074 | |
bbafd73b | 4075 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
e1ec7dd4 | 4076 | && facet->xout.slow == xout.slow; |
4dff9097 EJ |
4077 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4078 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4079 | |
4dff9097 EJ |
4080 | flow_format(&s, &facet->flow); |
4081 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4082 | |
bbafd73b | 4083 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4084 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4085 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4086 | facet->xout.odp_actions.size); | |
9616614b | 4087 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4088 | format_odp_actions(&s, xout.odp_actions.data, |
4089 | xout.odp_actions.size); | |
c0a71f4e | 4090 | ds_put_char(&s, ')'); |
6814e51f | 4091 | } |
4dff9097 | 4092 | |
bbafd73b EJ |
4093 | if (facet->xout.slow != xout.slow) { |
4094 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4095 | } |
4096 | ||
9616614b | 4097 | ds_destroy(&s); |
6814e51f | 4098 | } |
bbafd73b | 4099 | xlate_out_uninit(&xout); |
6814e51f BP |
4100 | |
4101 | return ok; | |
4102 | } | |
4103 | ||
15baa734 | 4104 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4105 | * |
4106 | * - If the rule found is different from 'facet''s current rule, moves | |
4107 | * 'facet' to the new rule and recompiles its actions. | |
4108 | * | |
4109 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4110 | * where it is and recompiles its actions anyway. |
4111 | * | |
4112 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
8449c4d6 | 4113 | * xlate_receive(), 'facet' is removed. |
5bf64ade EJ |
4114 | * |
4115 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
4116 | static bool | |
15baa734 | 4117 | facet_revalidate(struct facet *facet) |
abe529af | 4118 | { |
b7e2f6b3 | 4119 | struct ofproto_dpif *ofproto = facet->ofproto; |
abe529af | 4120 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4121 | struct subfacet *subfacet; |
bcd2633a | 4122 | struct flow_wildcards wc; |
bbafd73b EJ |
4123 | struct xlate_out xout; |
4124 | struct xlate_in xin; | |
abe529af BP |
4125 | |
4126 | COVERAGE_INC(facet_revalidate); | |
4127 | ||
f231418e EJ |
4128 | /* Check that child subfacets still correspond to this facet. Tunnel |
4129 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4130 | * change. */ | |
4131 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4132 | struct ofproto_dpif *recv_ofproto; | |
4133 | struct flow recv_flow; | |
4134 | int error; | |
4135 | ||
8449c4d6 EJ |
4136 | error = xlate_receive(ofproto->backer, NULL, subfacet->key, |
4137 | subfacet->key_len, &recv_flow, NULL, | |
4138 | &recv_ofproto, NULL); | |
f231418e EJ |
4139 | if (error |
4140 | || recv_ofproto != ofproto | |
bcd2633a | 4141 | || facet != facet_find(ofproto, &recv_flow)) { |
f231418e | 4142 | facet_remove(facet); |
5bf64ade | 4143 | return false; |
f231418e EJ |
4144 | } |
4145 | } | |
4146 | ||
bcd2633a | 4147 | flow_wildcards_init_catchall(&wc); |
ad3efdcb | 4148 | rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule); |
abe529af | 4149 | |
df2c07f4 | 4150 | /* Calculate new datapath actions. |
abe529af BP |
4151 | * |
4152 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4153 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4154 | * around to properly compose it. */ | |
47b0deae | 4155 | xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); |
bbafd73b | 4156 | xlate_actions(&xin, &xout); |
bcd2633a | 4157 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); |
4dff9097 EJ |
4158 | |
4159 | /* A facet's slow path reason should only change under dramatic | |
4160 | * circumstances. Rather than try to update everything, it's simpler to | |
bcd2633a JP |
4161 | * remove the facet and start over. |
4162 | * | |
4163 | * More importantly, if a facet's wildcards change, it will be relatively | |
4164 | * difficult to figure out if its subfacets still belong to it, and if not | |
4165 | * which facet they may belong to. Again, to avoid the complexity, we | |
4166 | * simply give up instead. */ | |
4167 | if (facet->xout.slow != xout.slow | |
4168 | || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { | |
4dff9097 | 4169 | facet_remove(facet); |
bbafd73b | 4170 | xlate_out_uninit(&xout); |
a2143702 | 4171 | rule_dpif_unref(new_rule); |
4dff9097 EJ |
4172 | return false; |
4173 | } | |
6a7e895f | 4174 | |
bbafd73b | 4175 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
4176 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
4177 | if (subfacet->path == SF_FAST_PATH) { | |
4178 | struct dpif_flow_stats stats; | |
b95fc6ba | 4179 | |
bbafd73b | 4180 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 4181 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 4182 | } |
abe529af | 4183 | } |
b95fc6ba | 4184 | |
15baa734 | 4185 | facet_flush_stats(facet); |
4dff9097 | 4186 | |
bbafd73b EJ |
4187 | ofpbuf_clear(&facet->xout.odp_actions); |
4188 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
4189 | xout.odp_actions.size); | |
abe529af BP |
4190 | } |
4191 | ||
4192 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
4193 | facet->xout.slow = xout.slow; |
4194 | facet->xout.has_learn = xout.has_learn; | |
4195 | facet->xout.has_normal = xout.has_normal; | |
4196 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
4197 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
4198 | facet->xout.mirrors = xout.mirrors; | |
4199 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
15aaf599 BP |
4200 | |
4201 | ovs_mutex_lock(&new_rule->up.mutex); | |
b7e2f6b3 | 4202 | facet->used = MAX(facet->used, new_rule->up.created); |
15aaf599 | 4203 | ovs_mutex_unlock(&new_rule->up.mutex); |
5bf64ade | 4204 | |
bbafd73b | 4205 | xlate_out_uninit(&xout); |
a2143702 | 4206 | rule_dpif_unref(new_rule); |
5bf64ade | 4207 | return true; |
abe529af BP |
4208 | } |
4209 | ||
bbb5d219 EJ |
4210 | static void |
4211 | facet_reset_counters(struct facet *facet) | |
4212 | { | |
4213 | facet->packet_count = 0; | |
4214 | facet->byte_count = 0; | |
9d24de3b JP |
4215 | facet->prev_packet_count = 0; |
4216 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4217 | facet->accounted_bytes = 0; |
4218 | } | |
4219 | ||
e1ec7dd4 EJ |
4220 | static void |
4221 | flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow, | |
4222 | struct dpif_flow_stats *stats, bool may_learn) | |
4223 | { | |
4224 | struct ofport_dpif *in_port; | |
e1ec7dd4 EJ |
4225 | struct xlate_in xin; |
4226 | ||
4227 | in_port = get_ofp_port(ofproto, flow->in_port.ofp_port); | |
4228 | if (in_port && in_port->is_tunnel) { | |
4229 | netdev_vport_inc_rx(in_port->up.netdev, stats); | |
4230 | } | |
4231 | ||
10c44245 | 4232 | xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL); |
e1ec7dd4 EJ |
4233 | xin.resubmit_stats = stats; |
4234 | xin.may_learn = may_learn; | |
4235 | xlate_actions_for_side_effects(&xin); | |
e1ec7dd4 EJ |
4236 | } |
4237 | ||
abe529af | 4238 | static void |
9dfb1f78 | 4239 | facet_push_stats(struct facet *facet, bool may_learn) |
abe529af | 4240 | { |
112bc5f4 | 4241 | struct dpif_flow_stats stats; |
abe529af | 4242 | |
cb22974d BP |
4243 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4244 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4245 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4246 | |
112bc5f4 BP |
4247 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4248 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4249 | stats.used = facet->used; | |
9dfb1f78 EJ |
4250 | stats.tcp_flags = facet->tcp_flags; |
4251 | ||
4252 | if (may_learn || stats.n_packets || facet->used > facet->prev_used) { | |
9d24de3b JP |
4253 | facet->prev_packet_count = facet->packet_count; |
4254 | facet->prev_byte_count = facet->byte_count; | |
4255 | facet->prev_used = facet->used; | |
abe529af | 4256 | |
e1ec7dd4 | 4257 | netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow, |
9dfb1f78 EJ |
4258 | facet->used); |
4259 | netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); | |
e1ec7dd4 | 4260 | mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors, |
ec7ceaed | 4261 | stats.n_packets, stats.n_bytes); |
e1ec7dd4 | 4262 | flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn); |
abe529af BP |
4263 | } |
4264 | } | |
4265 | ||
8844e035 | 4266 | static void |
4464589e | 4267 | push_all_stats__(bool run_fast) |
8844e035 EJ |
4268 | { |
4269 | static long long int rl = LLONG_MIN; | |
4270 | struct ofproto_dpif *ofproto; | |
4271 | ||
4272 | if (time_msec() < rl) { | |
4273 | return; | |
4274 | } | |
4275 | ||
4276 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
bcd2633a | 4277 | struct cls_cursor cursor; |
8844e035 EJ |
4278 | struct facet *facet; |
4279 | ||
0b4f2078 | 4280 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
4281 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
4282 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
9dfb1f78 | 4283 | facet_push_stats(facet, false); |
4464589e EJ |
4284 | if (run_fast) { |
4285 | run_fast_rl(); | |
4286 | } | |
8844e035 | 4287 | } |
0b4f2078 | 4288 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
8844e035 EJ |
4289 | } |
4290 | ||
4291 | rl = time_msec() + 100; | |
4292 | } | |
4293 | ||
4464589e EJ |
4294 | static void |
4295 | push_all_stats(void) | |
4296 | { | |
4297 | push_all_stats__(true); | |
4298 | } | |
4299 | ||
9583bc14 | 4300 | void |
70742c7f EJ |
4301 | rule_dpif_credit_stats(struct rule_dpif *rule, |
4302 | const struct dpif_flow_stats *stats) | |
abe529af | 4303 | { |
9fbe253e | 4304 | ovs_mutex_lock(&rule->stats_mutex); |
112bc5f4 BP |
4305 | rule->packet_count += stats->n_packets; |
4306 | rule->byte_count += stats->n_bytes; | |
448c2fa8 | 4307 | rule->up.used = MAX(rule->up.used, stats->used); |
9fbe253e | 4308 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af | 4309 | } |
70742c7f EJ |
4310 | |
4311 | bool | |
4312 | rule_dpif_fail_open(const struct rule_dpif *rule) | |
4313 | { | |
4314 | return rule->up.cr.priority == FAIL_OPEN_PRIORITY; | |
4315 | } | |
4316 | ||
4317 | ovs_be64 | |
4318 | rule_dpif_get_flow_cookie(const struct rule_dpif *rule) | |
15aaf599 | 4319 | OVS_REQUIRES(rule->up.mutex) |
70742c7f EJ |
4320 | { |
4321 | return rule->up.flow_cookie; | |
4322 | } | |
4323 | ||
4324 | void | |
4325 | rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout, | |
4326 | uint16_t hard_timeout) | |
4327 | { | |
4328 | ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout); | |
4329 | } | |
4330 | ||
6f00e29b BP |
4331 | /* Returns 'rule''s actions. The caller owns a reference on the returned |
4332 | * actions and must eventually release it (with rule_actions_unref()) to avoid | |
4333 | * a memory leak. */ | |
4334 | struct rule_actions * | |
4335 | rule_dpif_get_actions(const struct rule_dpif *rule) | |
70742c7f | 4336 | { |
15aaf599 | 4337 | return rule_get_actions(&rule->up); |
70742c7f | 4338 | } |
abe529af | 4339 | \f |
b0f7b9b5 BP |
4340 | /* Subfacets. */ |
4341 | ||
4342 | static struct subfacet * | |
04d08d54 EJ |
4343 | subfacet_find(struct dpif_backer *backer, const struct nlattr *key, |
4344 | size_t key_len, uint32_t key_hash) | |
b0f7b9b5 BP |
4345 | { |
4346 | struct subfacet *subfacet; | |
4347 | ||
4348 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
04d08d54 | 4349 | &backer->subfacets) { |
9566abf9 EJ |
4350 | if (subfacet->key_len == key_len |
4351 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
4352 | return subfacet; |
4353 | } | |
4354 | } | |
4355 | ||
4356 | return NULL; | |
4357 | } | |
4358 | ||
4359 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
4360 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
4361 | * existing subfacet if there is one, otherwise creates and returns a | |
4dff9097 | 4362 | * new subfacet. */ |
b0f7b9b5 | 4363 | static struct subfacet * |
e1ec7dd4 | 4364 | subfacet_create(struct facet *facet, struct flow_miss *miss) |
b0f7b9b5 | 4365 | { |
04d08d54 | 4366 | struct dpif_backer *backer = miss->ofproto->backer; |
a088a1ff JP |
4367 | const struct nlattr *key = miss->key; |
4368 | size_t key_len = miss->key_len; | |
4369 | uint32_t key_hash; | |
b0f7b9b5 BP |
4370 | struct subfacet *subfacet; |
4371 | ||
a088a1ff JP |
4372 | key_hash = odp_flow_key_hash(key, key_len); |
4373 | ||
3b145dd7 BP |
4374 | if (list_is_empty(&facet->subfacets)) { |
4375 | subfacet = &facet->one_subfacet; | |
4376 | } else { | |
04d08d54 | 4377 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
3b145dd7 BP |
4378 | if (subfacet) { |
4379 | if (subfacet->facet == facet) { | |
4380 | return subfacet; | |
4381 | } | |
4382 | ||
4383 | /* This shouldn't happen. */ | |
4384 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
4385 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
4386 | } |
4387 | ||
3b145dd7 | 4388 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
4389 | } |
4390 | ||
264ac358 | 4391 | COVERAGE_INC(subfacet_create); |
04d08d54 | 4392 | hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); |
b0f7b9b5 BP |
4393 | list_push_back(&facet->subfacets, &subfacet->list_node); |
4394 | subfacet->facet = facet; | |
9566abf9 EJ |
4395 | subfacet->key = xmemdup(key, key_len); |
4396 | subfacet->key_len = key_len; | |
e1ec7dd4 EJ |
4397 | subfacet->used = miss->stats.used; |
4398 | subfacet->created = subfacet->used; | |
26cd7e34 BP |
4399 | subfacet->dp_packet_count = 0; |
4400 | subfacet->dp_byte_count = 0; | |
6a7e895f | 4401 | subfacet->path = SF_NOT_INSTALLED; |
04d08d54 | 4402 | subfacet->backer = backer; |
b0f7b9b5 BP |
4403 | |
4404 | return subfacet; | |
4405 | } | |
4406 | ||
b0f7b9b5 BP |
4407 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
4408 | * its facet within 'ofproto', and frees it. */ | |
4409 | static void | |
15baa734 | 4410 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4411 | { |
15baa734 | 4412 | struct facet *facet = subfacet->facet; |
655ab909 | 4413 | |
264ac358 | 4414 | COVERAGE_INC(subfacet_destroy); |
15baa734 | 4415 | subfacet_uninstall(subfacet); |
04d08d54 | 4416 | hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); |
b0f7b9b5 BP |
4417 | list_remove(&subfacet->list_node); |
4418 | free(subfacet->key); | |
26cd7e34 BP |
4419 | if (subfacet != &facet->one_subfacet) { |
4420 | free(subfacet); | |
4421 | } | |
b0f7b9b5 BP |
4422 | } |
4423 | ||
4424 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4425 | * last remaining subfacet in its facet destroys the facet too. */ | |
4426 | static void | |
15baa734 | 4427 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4428 | { |
4429 | struct facet *facet = subfacet->facet; | |
4430 | ||
551a2f6c BP |
4431 | if (list_is_singleton(&facet->subfacets)) { |
4432 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4433 | facet_remove(facet); |
551a2f6c | 4434 | } else { |
15baa734 | 4435 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4436 | } |
4437 | } | |
4438 | ||
1d85f9e5 | 4439 | static void |
04d08d54 | 4440 | subfacet_destroy_batch(struct dpif_backer *backer, |
1d85f9e5 JP |
4441 | struct subfacet **subfacets, int n) |
4442 | { | |
1d85f9e5 JP |
4443 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
4444 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
4445 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
4446 | int i; | |
4447 | ||
4448 | for (i = 0; i < n; i++) { | |
4449 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
4450 | ops[i].u.flow_del.key = subfacets[i]->key; |
4451 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
4452 | ops[i].u.flow_del.stats = &stats[i]; |
4453 | opsp[i] = &ops[i]; | |
4454 | } | |
4455 | ||
04d08d54 | 4456 | dpif_operate(backer->dpif, opsp, n); |
1d85f9e5 JP |
4457 | for (i = 0; i < n; i++) { |
4458 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
4459 | subfacets[i]->path = SF_NOT_INSTALLED; | |
4460 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 4461 | run_fast_rl(); |
1d85f9e5 JP |
4462 | } |
4463 | } | |
4464 | ||
b0f7b9b5 BP |
4465 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4466 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4467 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4468 | * since 'subfacet' was last updated. | |
4469 | * | |
4470 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4471 | static int | |
4dff9097 EJ |
4472 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
4473 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 4474 | { |
15baa734 | 4475 | struct facet *facet = subfacet->facet; |
bbafd73b | 4476 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
4477 | const struct nlattr *actions = odp_actions->data; |
4478 | size_t actions_len = odp_actions->size; | |
d445cc16 JP |
4479 | struct odputil_keybuf maskbuf; |
4480 | struct ofpbuf mask; | |
4dff9097 | 4481 | |
6a7e895f | 4482 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 4483 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
4484 | int ret; |
4485 | ||
0f74f625 AZ |
4486 | flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE |
4487 | : DPIF_FP_MODIFY; | |
b0f7b9b5 BP |
4488 | if (stats) { |
4489 | flags |= DPIF_FP_ZERO_STATS; | |
4490 | } | |
4491 | ||
6a7e895f | 4492 | if (path == SF_SLOW_PATH) { |
b7e2f6b3 | 4493 | compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
4494 | slow_path_stub, sizeof slow_path_stub, |
4495 | &actions, &actions_len); | |
4496 | } | |
4497 | ||
d445cc16 | 4498 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6b35e630 JP |
4499 | if (enable_megaflows) { |
4500 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
4501 | &facet->flow, UINT32_MAX); | |
4502 | } | |
d445cc16 JP |
4503 | |
4504 | ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, | |
4505 | subfacet->key_len, mask.data, mask.size, | |
e6cc0bab | 4506 | actions, actions_len, stats); |
b0f7b9b5 BP |
4507 | |
4508 | if (stats) { | |
4509 | subfacet_reset_dp_stats(subfacet, stats); | |
4510 | } | |
4511 | ||
9b1a48c2 JP |
4512 | if (ret) { |
4513 | COVERAGE_INC(subfacet_install_fail); | |
4514 | } else { | |
6a7e895f BP |
4515 | subfacet->path = path; |
4516 | } | |
b0f7b9b5 BP |
4517 | return ret; |
4518 | } | |
4519 | ||
4520 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
4521 | static void | |
15baa734 | 4522 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 4523 | { |
6a7e895f | 4524 | if (subfacet->path != SF_NOT_INSTALLED) { |
b7e2f6b3 | 4525 | struct ofproto_dpif *ofproto = subfacet->facet->ofproto; |
b0f7b9b5 | 4526 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
4527 | int error; |
4528 | ||
9566abf9 EJ |
4529 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
4530 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
4531 | subfacet_reset_dp_stats(subfacet, &stats); |
4532 | if (!error) { | |
15baa734 | 4533 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 4534 | } |
6a7e895f | 4535 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 4536 | } else { |
cb22974d BP |
4537 | ovs_assert(subfacet->dp_packet_count == 0); |
4538 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
4539 | } |
4540 | } | |
4541 | ||
4542 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4543 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4544 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4545 | * was reset in the datapath. 'stats' will be modified to include only | |
4546 | * statistics new since 'subfacet' was last updated. */ | |
4547 | static void | |
4548 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4549 | struct dpif_flow_stats *stats) | |
4550 | { | |
4551 | if (stats | |
4552 | && subfacet->dp_packet_count <= stats->n_packets | |
4553 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4554 | stats->n_packets -= subfacet->dp_packet_count; | |
4555 | stats->n_bytes -= subfacet->dp_byte_count; | |
4556 | } | |
4557 | ||
4558 | subfacet->dp_packet_count = 0; | |
4559 | subfacet->dp_byte_count = 0; | |
4560 | } | |
4561 | ||
b0f7b9b5 BP |
4562 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. |
4563 | * | |
4564 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4565 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4566 | * represents a packet that was sent by hand or if it represents statistics | |
4567 | * that have been cleared out of the datapath. */ | |
4568 | static void | |
15baa734 | 4569 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4570 | const struct dpif_flow_stats *stats) |
4571 | { | |
4572 | if (stats->n_packets || stats->used > subfacet->used) { | |
4573 | struct facet *facet = subfacet->facet; | |
4574 | ||
9dfb1f78 EJ |
4575 | subfacet->used = MAX(subfacet->used, stats->used); |
4576 | facet->used = MAX(facet->used, stats->used); | |
b0f7b9b5 BP |
4577 | facet->packet_count += stats->n_packets; |
4578 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4579 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4580 | } |
4581 | } | |
4582 | \f | |
abe529af BP |
4583 | /* Rules. */ |
4584 | ||
bcd2633a JP |
4585 | /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets |
4586 | * the fields that were relevant as part of the lookup. */ | |
ad3efdcb | 4587 | void |
bcd2633a | 4588 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
ad3efdcb | 4589 | struct flow_wildcards *wc, struct rule_dpif **rule) |
c57b2226 | 4590 | { |
ec89fc6f | 4591 | struct ofport_dpif *port; |
c57b2226 | 4592 | |
ad3efdcb EJ |
4593 | if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) { |
4594 | return; | |
c57b2226 | 4595 | } |
ec89fc6f EJ |
4596 | port = get_ofp_port(ofproto, flow->in_port.ofp_port); |
4597 | if (!port) { | |
4598 | VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16, | |
4599 | flow->in_port.ofp_port); | |
4600 | } | |
c57b2226 | 4601 | |
70742c7f EJ |
4602 | choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule, |
4603 | ofproto->no_packet_in_rule, rule); | |
c57b2226 BP |
4604 | } |
4605 | ||
ad3efdcb | 4606 | bool |
9583bc14 EJ |
4607 | rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, |
4608 | const struct flow *flow, struct flow_wildcards *wc, | |
ad3efdcb | 4609 | uint8_t table_id, struct rule_dpif **rule) |
abe529af | 4610 | { |
49a0e0eb | 4611 | const struct cls_rule *cls_rule; |
7257b535 | 4612 | struct classifier *cls; |
7fd51d39 | 4613 | bool frag; |
7257b535 | 4614 | |
ad3efdcb | 4615 | *rule = NULL; |
9cdaaebe | 4616 | if (table_id >= N_TABLES) { |
ad3efdcb | 4617 | return false; |
9cdaaebe BP |
4618 | } |
4619 | ||
1dd35f8a | 4620 | if (wc) { |
7431e171 | 4621 | memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type); |
1dd35f8a JP |
4622 | wc->masks.nw_frag |= FLOW_NW_FRAG_MASK; |
4623 | } | |
4624 | ||
d0918789 | 4625 | cls = &ofproto->up.tables[table_id].cls; |
ad3efdcb | 4626 | ovs_rwlock_rdlock(&cls->rwlock); |
7fd51d39 BP |
4627 | frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; |
4628 | if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
4629 | /* We must pretend that transport ports are unavailable. */ | |
7257b535 BP |
4630 | struct flow ofpc_normal_flow = *flow; |
4631 | ofpc_normal_flow.tp_src = htons(0); | |
4632 | ofpc_normal_flow.tp_dst = htons(0); | |
bcd2633a | 4633 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); |
7fd51d39 BP |
4634 | } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { |
4635 | cls_rule = &ofproto->drop_frags_rule->up.cr; | |
bcd2633a JP |
4636 | if (wc) { |
4637 | flow_wildcards_init_exact(wc); | |
4638 | } | |
7257b535 | 4639 | } else { |
bcd2633a | 4640 | cls_rule = classifier_lookup(cls, flow, wc); |
7257b535 | 4641 | } |
ad3efdcb EJ |
4642 | |
4643 | *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
a2143702 | 4644 | rule_dpif_ref(*rule); |
ad3efdcb EJ |
4645 | ovs_rwlock_unlock(&cls->rwlock); |
4646 | ||
4647 | return *rule != NULL; | |
abe529af BP |
4648 | } |
4649 | ||
ec89fc6f EJ |
4650 | /* Given a port configuration (specified as zero if there's no port), chooses |
4651 | * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a | |
4652 | * flow table miss. */ | |
70742c7f | 4653 | void |
ec89fc6f | 4654 | choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule, |
70742c7f | 4655 | struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule) |
c376f9a3 | 4656 | { |
70742c7f | 4657 | *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; |
a2143702 BP |
4658 | rule_dpif_ref(*rule); |
4659 | } | |
4660 | ||
4661 | void | |
4662 | rule_dpif_ref(struct rule_dpif *rule) | |
4663 | { | |
4664 | if (rule) { | |
4665 | ofproto_rule_ref(&rule->up); | |
4666 | } | |
c376f9a3 IY |
4667 | } |
4668 | ||
ad3efdcb | 4669 | void |
a2143702 | 4670 | rule_dpif_unref(struct rule_dpif *rule) |
ad3efdcb EJ |
4671 | { |
4672 | if (rule) { | |
a2143702 | 4673 | ofproto_rule_unref(&rule->up); |
ad3efdcb EJ |
4674 | } |
4675 | } | |
4676 | ||
7ee20df1 BP |
4677 | static void |
4678 | complete_operation(struct rule_dpif *rule) | |
15aaf599 | 4679 | OVS_REQUIRES(ofproto_mutex) |
7ee20df1 BP |
4680 | { |
4681 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4682 | ||
a1616063 | 4683 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
35232d32 | 4684 | ofoperation_complete(rule->up.pending, 0); |
7ee20df1 BP |
4685 | } |
4686 | ||
70742c7f EJ |
4687 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) |
4688 | { | |
4689 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
4690 | } | |
4691 | ||
abe529af BP |
4692 | static struct rule * |
4693 | rule_alloc(void) | |
4694 | { | |
4695 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4696 | return &rule->up; | |
4697 | } | |
4698 | ||
4699 | static void | |
4700 | rule_dealloc(struct rule *rule_) | |
4701 | { | |
4702 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4703 | free(rule); | |
4704 | } | |
4705 | ||
90bf1e07 | 4706 | static enum ofperr |
abe529af BP |
4707 | rule_construct(struct rule *rule_) |
4708 | { | |
4709 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
834d6caf | 4710 | ovs_mutex_init(&rule->stats_mutex); |
9fbe253e | 4711 | ovs_mutex_lock(&rule->stats_mutex); |
abe529af BP |
4712 | rule->packet_count = 0; |
4713 | rule->byte_count = 0; | |
9fbe253e | 4714 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af BP |
4715 | return 0; |
4716 | } | |
4717 | ||
4718 | static void | |
8037acb4 | 4719 | rule_insert(struct rule *rule_) |
15aaf599 | 4720 | OVS_REQUIRES(ofproto_mutex) |
abe529af | 4721 | { |
9fbe253e EJ |
4722 | struct rule_dpif *rule = rule_dpif_cast(rule_); |
4723 | complete_operation(rule); | |
8037acb4 BP |
4724 | } |
4725 | ||
4726 | static void | |
4727 | rule_delete(struct rule *rule_) | |
15aaf599 | 4728 | OVS_REQUIRES(ofproto_mutex) |
8037acb4 BP |
4729 | { |
4730 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4731 | complete_operation(rule); | |
4732 | } | |
4733 | ||
4734 | static void | |
4735 | rule_destruct(struct rule *rule_) | |
4736 | { | |
4737 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
9fbe253e | 4738 | ovs_mutex_destroy(&rule->stats_mutex); |
abe529af BP |
4739 | } |
4740 | ||
4741 | static void | |
4742 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4743 | { | |
4744 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 4745 | |
4464589e EJ |
4746 | /* push_all_stats() can handle flow misses which, when using the learn |
4747 | * action, can cause rules to be added and deleted. This can corrupt our | |
4748 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
4749 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
4750 | push_all_stats__(false); | |
bf1e8ff9 | 4751 | |
abe529af BP |
4752 | /* Start from historical data for 'rule' itself that are no longer tracked |
4753 | * in facets. This counts, for example, facets that have expired. */ | |
9fbe253e | 4754 | ovs_mutex_lock(&rule->stats_mutex); |
abe529af BP |
4755 | *packets = rule->packet_count; |
4756 | *bytes = rule->byte_count; | |
9fbe253e | 4757 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af BP |
4758 | } |
4759 | ||
0a740f48 EJ |
4760 | static void |
4761 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
4762 | struct ofpbuf *packet) | |
abe529af | 4763 | { |
419460f4 AW |
4764 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
4765 | ||
4766 | ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet); | |
0a740f48 | 4767 | } |
5bf0e941 | 4768 | |
0a740f48 EJ |
4769 | static enum ofperr |
4770 | rule_execute(struct rule *rule, const struct flow *flow, | |
4771 | struct ofpbuf *packet) | |
4772 | { | |
4773 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
4774 | ofpbuf_delete(packet); | |
5bf0e941 | 4775 | return 0; |
abe529af BP |
4776 | } |
4777 | ||
7ee20df1 | 4778 | static void |
b277c7cc | 4779 | rule_modify_actions(struct rule *rule_, bool reset_counters) |
15aaf599 | 4780 | OVS_REQUIRES(ofproto_mutex) |
abe529af BP |
4781 | { |
4782 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 | 4783 | |
b277c7cc BP |
4784 | if (reset_counters) { |
4785 | ovs_mutex_lock(&rule->stats_mutex); | |
4786 | rule->packet_count = 0; | |
4787 | rule->byte_count = 0; | |
4788 | ovs_mutex_unlock(&rule->stats_mutex); | |
4789 | } | |
4790 | ||
7ee20df1 | 4791 | complete_operation(rule); |
abe529af BP |
4792 | } |
4793 | \f | |
97d6520b | 4794 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4795 | * May modify 'packet'. |
abe529af | 4796 | * Returns 0 if successful, otherwise a positive errno value. */ |
91d6cd12 AW |
4797 | int |
4798 | ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) | |
abe529af | 4799 | { |
79e15b78 | 4800 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
abe529af BP |
4801 | int error; |
4802 | ||
91d6cd12 | 4803 | error = xlate_send_packet(ofport, packet); |
79e15b78 | 4804 | |
0da3c61b | 4805 | ovs_mutex_lock(&ofproto->stats_mutex); |
79e15b78 EJ |
4806 | ofproto->stats.tx_packets++; |
4807 | ofproto->stats.tx_bytes += packet->size; | |
0da3c61b | 4808 | ovs_mutex_unlock(&ofproto->stats_mutex); |
abe529af BP |
4809 | return error; |
4810 | } | |
abe529af | 4811 | |
6a7e895f BP |
4812 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
4813 | * The action will state 'slow' as the reason that the action is in the slow | |
4814 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
4815 | * dump-flows" output to see why a flow is in the slow path.) | |
4816 | * | |
4817 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
4818 | * 'stub_size' must be large enough for the action. | |
4819 | * | |
4820 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
4821 | * respectively. */ | |
4822 | static void | |
4823 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
4824 | enum slow_path_reason slow, | |
4825 | uint64_t *stub, size_t stub_size, | |
4826 | const struct nlattr **actionsp, size_t *actions_lenp) | |
4827 | { | |
4828 | union user_action_cookie cookie; | |
4829 | struct ofpbuf buf; | |
4830 | ||
4831 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
4832 | cookie.slow_path.unused = 0; | |
4833 | cookie.slow_path.reason = slow; | |
4834 | ||
4835 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 4836 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
4e022ec0 AW |
4837 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, |
4838 | ODPP_NONE); | |
29089a54 | 4839 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 4840 | } else { |
89a8a7f0 EJ |
4841 | odp_port_t odp_port; |
4842 | uint32_t pid; | |
4843 | ||
4844 | odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port); | |
4845 | pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port); | |
4846 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); | |
625b0720 | 4847 | } |
6a7e895f BP |
4848 | *actionsp = buf.data; |
4849 | *actions_lenp = buf.size; | |
4850 | } | |
9583bc14 EJ |
4851 | \f |
4852 | static bool | |
4853 | set_frag_handling(struct ofproto *ofproto_, | |
4854 | enum ofp_config_flags frag_handling) | |
abe529af BP |
4855 | { |
4856 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 4857 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 4858 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
4859 | return true; |
4860 | } else { | |
4861 | return false; | |
4862 | } | |
abe529af BP |
4863 | } |
4864 | ||
90bf1e07 | 4865 | static enum ofperr |
419460f4 | 4866 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
abe529af | 4867 | const struct flow *flow, |
f25d0cf3 | 4868 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af | 4869 | { |
419460f4 AW |
4870 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
4871 | ||
4872 | ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts, | |
4873 | ofpacts_len, packet); | |
548de4dd | 4874 | return 0; |
abe529af | 4875 | } |
6fca1ffb BP |
4876 | \f |
4877 | /* NetFlow. */ | |
4878 | ||
4879 | static int | |
4880 | set_netflow(struct ofproto *ofproto_, | |
4881 | const struct netflow_options *netflow_options) | |
4882 | { | |
4883 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4884 | ||
4885 | if (netflow_options) { | |
4886 | if (!ofproto->netflow) { | |
4887 | ofproto->netflow = netflow_create(); | |
7bab1576 | 4888 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6fca1ffb BP |
4889 | } |
4890 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7bab1576 EJ |
4891 | } else if (ofproto->netflow) { |
4892 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
6fca1ffb BP |
4893 | netflow_destroy(ofproto->netflow); |
4894 | ofproto->netflow = NULL; | |
6fca1ffb | 4895 | } |
7bab1576 EJ |
4896 | |
4897 | return 0; | |
6fca1ffb | 4898 | } |
abe529af BP |
4899 | |
4900 | static void | |
4901 | get_netflow_ids(const struct ofproto *ofproto_, | |
4902 | uint8_t *engine_type, uint8_t *engine_id) | |
4903 | { | |
4904 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4905 | ||
acf60855 | 4906 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 4907 | } |
6fca1ffb BP |
4908 | |
4909 | static void | |
4910 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
4911 | { | |
4912 | if (!facet_is_controller_flow(facet) && | |
4913 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 4914 | struct subfacet *subfacet; |
6fca1ffb BP |
4915 | struct ofexpired expired; |
4916 | ||
b0f7b9b5 | 4917 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4918 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 4919 | struct dpif_flow_stats stats; |
6fca1ffb | 4920 | |
bcd2633a JP |
4921 | subfacet_install(subfacet, &facet->xout.odp_actions, |
4922 | &stats); | |
15baa734 | 4923 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 4924 | } |
6fca1ffb BP |
4925 | } |
4926 | ||
4927 | expired.flow = facet->flow; | |
4928 | expired.packet_count = facet->packet_count; | |
4929 | expired.byte_count = facet->byte_count; | |
4930 | expired.used = facet->used; | |
4931 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4932 | } | |
4933 | } | |
4934 | ||
4935 | static void | |
4936 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
4937 | { | |
bcd2633a | 4938 | struct cls_cursor cursor; |
6fca1ffb BP |
4939 | struct facet *facet; |
4940 | ||
0b4f2078 | 4941 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
4942 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
4943 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6fca1ffb BP |
4944 | send_active_timeout(ofproto, facet); |
4945 | } | |
0b4f2078 | 4946 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6fca1ffb | 4947 | } |
abe529af BP |
4948 | \f |
4949 | static struct ofproto_dpif * | |
4950 | ofproto_dpif_lookup(const char *name) | |
4951 | { | |
b44a10b7 BP |
4952 | struct ofproto_dpif *ofproto; |
4953 | ||
4954 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
4955 | hash_string(name, 0), &all_ofproto_dpifs) { | |
4956 | if (!strcmp(ofproto->up.name, name)) { | |
4957 | return ofproto; | |
4958 | } | |
4959 | } | |
4960 | return NULL; | |
abe529af BP |
4961 | } |
4962 | ||
f0a3aa2e | 4963 | static void |
96e466a3 | 4964 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 4965 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 4966 | { |
490df1ef | 4967 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 4968 | |
96e466a3 EJ |
4969 | if (argc > 1) { |
4970 | ofproto = ofproto_dpif_lookup(argv[1]); | |
4971 | if (!ofproto) { | |
bde9f75d | 4972 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
4973 | return; |
4974 | } | |
509c0149 | 4975 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 4976 | mac_learning_flush(ofproto->ml); |
509c0149 | 4977 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 EJ |
4978 | } else { |
4979 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
509c0149 | 4980 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 4981 | mac_learning_flush(ofproto->ml); |
509c0149 | 4982 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 | 4983 | } |
f0a3aa2e | 4984 | } |
f0a3aa2e | 4985 | |
bde9f75d | 4986 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
4987 | } |
4988 | ||
46c88433 EJ |
4989 | static struct ofport_dpif * |
4990 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
4991 | { | |
4992 | return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif, | |
4993 | bundle_node); | |
4994 | } | |
4995 | ||
abe529af | 4996 | static void |
0e15264f BP |
4997 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
4998 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
4999 | { |
5000 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5001 | const struct ofproto_dpif *ofproto; | |
5002 | const struct mac_entry *e; | |
5003 | ||
0e15264f | 5004 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 5005 | if (!ofproto) { |
bde9f75d | 5006 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
5007 | return; |
5008 | } | |
5009 | ||
5010 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
509c0149 | 5011 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
5012 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
5013 | struct ofbundle *bundle = e->port.p; | |
ad28062f BP |
5014 | char name[OFP_MAX_PORT_NAME_LEN]; |
5015 | ||
5016 | ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port, | |
5017 | name, sizeof name); | |
5018 | ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n", | |
5019 | name, e->vlan, ETH_ADDR_ARGS(e->mac), | |
e764773c | 5020 | mac_entry_age(ofproto->ml, e)); |
abe529af | 5021 | } |
509c0149 | 5022 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
bde9f75d | 5023 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
5024 | ds_destroy(&ds); |
5025 | } | |
5026 | ||
6a6455e5 | 5027 | struct trace_ctx { |
bbafd73b EJ |
5028 | struct xlate_out xout; |
5029 | struct xlate_in xin; | |
abe529af BP |
5030 | struct flow flow; |
5031 | struct ds *result; | |
5032 | }; | |
5033 | ||
5034 | static void | |
4d0acc70 | 5035 | trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) |
abe529af | 5036 | { |
15aaf599 BP |
5037 | struct rule_actions *actions; |
5038 | ovs_be64 cookie; | |
5039 | ||
abe529af BP |
5040 | ds_put_char_multiple(result, '\t', level); |
5041 | if (!rule) { | |
5042 | ds_put_cstr(result, "No match\n"); | |
5043 | return; | |
5044 | } | |
5045 | ||
15aaf599 BP |
5046 | ovs_mutex_lock(&rule->up.mutex); |
5047 | cookie = rule->up.flow_cookie; | |
5048 | ovs_mutex_unlock(&rule->up.mutex); | |
5049 | ||
29901626 | 5050 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
15aaf599 | 5051 | rule ? rule->up.table_id : 0, ntohll(cookie)); |
79feb7df | 5052 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5053 | ds_put_char(result, '\n'); |
5054 | ||
15aaf599 BP |
5055 | actions = rule_dpif_get_actions(rule); |
5056 | ||
abe529af BP |
5057 | ds_put_char_multiple(result, '\t', level); |
5058 | ds_put_cstr(result, "OpenFlow "); | |
15aaf599 | 5059 | ofpacts_format(actions->ofpacts, actions->ofpacts_len, result); |
abe529af | 5060 | ds_put_char(result, '\n'); |
15aaf599 BP |
5061 | |
5062 | rule_actions_unref(actions); | |
abe529af BP |
5063 | } |
5064 | ||
5065 | static void | |
5066 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 5067 | struct trace_ctx *trace) |
abe529af BP |
5068 | { |
5069 | ds_put_char_multiple(result, '\t', level); | |
5070 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 5071 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
5072 | ds_put_cstr(result, "unchanged"); |
5073 | } else { | |
bbafd73b EJ |
5074 | flow_format(result, &trace->xin.flow); |
5075 | trace->flow = trace->xin.flow; | |
abe529af BP |
5076 | } |
5077 | ds_put_char(result, '\n'); | |
5078 | } | |
5079 | ||
eb9e1c26 EJ |
5080 | static void |
5081 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 5082 | struct trace_ctx *trace) |
eb9e1c26 EJ |
5083 | { |
5084 | size_t i; | |
5085 | ||
5086 | ds_put_char_multiple(result, '\t', level); | |
5087 | ds_put_format(result, "%s:", title); | |
5088 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5089 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5090 | } | |
5091 | ds_put_char(result, '\n'); | |
5092 | } | |
5093 | ||
1ed8d352 EJ |
5094 | static void |
5095 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 5096 | struct trace_ctx *trace) |
1ed8d352 | 5097 | { |
bbafd73b | 5098 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
5099 | |
5100 | ds_put_char_multiple(result, '\t', level); | |
5101 | ds_put_format(result, "%s: ", title); | |
5102 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
5103 | ds_put_char(result, '\n'); | |
5104 | } | |
5105 | ||
abe529af | 5106 | static void |
4d0acc70 | 5107 | trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse) |
abe529af | 5108 | { |
4d0acc70 | 5109 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
abe529af BP |
5110 | struct ds *result = trace->result; |
5111 | ||
5112 | ds_put_char(result, '\n'); | |
4d0acc70 EJ |
5113 | trace_format_flow(result, recurse + 1, "Resubmitted flow", trace); |
5114 | trace_format_regs(result, recurse + 1, "Resubmitted regs", trace); | |
5115 | trace_format_odp(result, recurse + 1, "Resubmitted odp", trace); | |
5116 | trace_format_rule(result, recurse + 1, rule); | |
abe529af BP |
5117 | } |
5118 | ||
479df176 | 5119 | static void |
4d0acc70 | 5120 | trace_report(struct xlate_in *xin, const char *s, int recurse) |
479df176 | 5121 | { |
4d0acc70 | 5122 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
479df176 BP |
5123 | struct ds *result = trace->result; |
5124 | ||
4d0acc70 | 5125 | ds_put_char_multiple(result, '\t', recurse); |
479df176 BP |
5126 | ds_put_cstr(result, s); |
5127 | ds_put_char(result, '\n'); | |
5128 | } | |
5129 | ||
abe529af | 5130 | static void |
0e15264f | 5131 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
5132 | void *aux OVS_UNUSED) |
5133 | { | |
0a37839c | 5134 | const struct dpif_backer *backer = NULL; |
abe529af | 5135 | struct ofproto_dpif *ofproto; |
ea5e3887 | 5136 | struct ofpbuf odp_key, odp_mask; |
876b0e1c | 5137 | struct ofpbuf *packet; |
abe529af BP |
5138 | struct ds result; |
5139 | struct flow flow; | |
0a37839c | 5140 | struct simap port_names; |
abe529af BP |
5141 | char *s; |
5142 | ||
876b0e1c | 5143 | packet = NULL; |
50aa28fd | 5144 | backer = NULL; |
abe529af | 5145 | ds_init(&result); |
50aa28fd | 5146 | ofpbuf_init(&odp_key, 0); |
ea5e3887 | 5147 | ofpbuf_init(&odp_mask, 0); |
0a37839c | 5148 | simap_init(&port_names); |
abe529af | 5149 | |
50aa28fd AW |
5150 | /* Handle "-generate" or a hex string as the last argument. */ |
5151 | if (!strcmp(argv[argc - 1], "-generate")) { | |
5152 | packet = ofpbuf_new(0); | |
5153 | argc--; | |
5154 | } else { | |
5155 | const char *error = eth_from_hex(argv[argc - 1], &packet); | |
5156 | if (!error) { | |
5157 | argc--; | |
5158 | } else if (argc == 4) { | |
5159 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
5160 | unixctl_command_reply_error(conn, error); | |
5161 | goto exit; | |
5162 | } | |
e84173dc | 5163 | } |
876b0e1c | 5164 | |
0a37839c GS |
5165 | /* odp_flow can have its in_port specified as a name instead of port no. |
5166 | * We do not yet know whether a given flow is a odp_flow or a br_flow. | |
5167 | * But, to know whether a flow is odp_flow through odp_flow_from_string(), | |
5168 | * we need to create a simap of name to port no. */ | |
5169 | if (argc == 3) { | |
5170 | const char *dp_type; | |
5171 | if (!strncmp(argv[1], "ovs-", 4)) { | |
5172 | dp_type = argv[1] + 4; | |
5173 | } else { | |
5174 | dp_type = argv[1]; | |
5175 | } | |
5176 | backer = shash_find_data(&all_dpif_backers, dp_type); | |
5177 | } else { | |
5178 | struct shash_node *node; | |
5179 | if (shash_count(&all_dpif_backers) == 1) { | |
5180 | node = shash_first(&all_dpif_backers); | |
5181 | backer = node->data; | |
5182 | } | |
5183 | } | |
5184 | if (backer && backer->dpif) { | |
5185 | struct dpif_port dpif_port; | |
5186 | struct dpif_port_dump port_dump; | |
5187 | DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) { | |
5188 | simap_put(&port_names, dpif_port.name, | |
5189 | odp_to_u32(dpif_port.port_no)); | |
5190 | } | |
5191 | } | |
5192 | ||
50aa28fd AW |
5193 | /* Parse the flow and determine whether a datapath or |
5194 | * bridge is specified. If function odp_flow_key_from_string() | |
5195 | * returns 0, the flow is a odp_flow. If function | |
5196 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
0a37839c GS |
5197 | if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) { |
5198 | if (!backer) { | |
5199 | unixctl_command_reply_error(conn, "Cannot find the datapath"); | |
5200 | goto exit; | |
876b0e1c | 5201 | } |
8b3b8dd1 | 5202 | |
8449c4d6 EJ |
5203 | if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow, |
5204 | NULL, &ofproto, NULL)) { | |
50aa28fd AW |
5205 | unixctl_command_reply_error(conn, "Invalid datapath flow"); |
5206 | goto exit; | |
8b3b8dd1 | 5207 | } |
50aa28fd AW |
5208 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
5209 | } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { | |
5210 | if (argc != 3) { | |
5211 | unixctl_command_reply_error(conn, "Must specify bridge name"); | |
876b0e1c BP |
5212 | goto exit; |
5213 | } | |
5214 | ||
50aa28fd AW |
5215 | ofproto = ofproto_dpif_lookup(argv[1]); |
5216 | if (!ofproto) { | |
5217 | unixctl_command_reply_error(conn, "Unknown bridge name"); | |
5218 | goto exit; | |
5219 | } | |
876b0e1c | 5220 | } else { |
50aa28fd | 5221 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
abe529af BP |
5222 | goto exit; |
5223 | } | |
5224 | ||
50aa28fd AW |
5225 | /* Generate a packet, if requested. */ |
5226 | if (packet) { | |
5227 | if (!packet->size) { | |
5228 | flow_compose(packet, &flow); | |
5229 | } else { | |
4e022ec0 AW |
5230 | union flow_in_port in_port_; |
5231 | ||
5232 | in_port_ = flow.in_port; | |
50aa28fd AW |
5233 | ds_put_cstr(&result, "Packet: "); |
5234 | s = ofp_packet_to_string(packet->data, packet->size); | |
5235 | ds_put_cstr(&result, s); | |
5236 | free(s); | |
5237 | ||
5238 | /* Use the metadata from the flow and the packet argument | |
5239 | * to reconstruct the flow. */ | |
1362e248 | 5240 | flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL, |
4e022ec0 | 5241 | &in_port_, &flow); |
50aa28fd AW |
5242 | } |
5243 | } | |
5244 | ||
47b0deae | 5245 | ofproto_trace(ofproto, &flow, packet, &result); |
6a6455e5 EJ |
5246 | unixctl_command_reply(conn, ds_cstr(&result)); |
5247 | ||
5248 | exit: | |
5249 | ds_destroy(&result); | |
5250 | ofpbuf_delete(packet); | |
5251 | ofpbuf_uninit(&odp_key); | |
ea5e3887 | 5252 | ofpbuf_uninit(&odp_mask); |
0a37839c | 5253 | simap_destroy(&port_names); |
6a6455e5 EJ |
5254 | } |
5255 | ||
ade6ad9c | 5256 | static void |
6a6455e5 | 5257 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, |
47b0deae | 5258 | const struct ofpbuf *packet, struct ds *ds) |
6a6455e5 EJ |
5259 | { |
5260 | struct rule_dpif *rule; | |
083f277e | 5261 | struct flow_wildcards wc; |
6a6455e5 EJ |
5262 | |
5263 | ds_put_cstr(ds, "Flow: "); | |
5264 | flow_format(ds, flow); | |
5265 | ds_put_char(ds, '\n'); | |
abe529af | 5266 | |
083f277e JG |
5267 | flow_wildcards_init_catchall(&wc); |
5268 | rule_dpif_lookup(ofproto, flow, &wc, &rule); | |
c57b2226 | 5269 | |
4d0acc70 | 5270 | trace_format_rule(ds, 0, rule); |
c57b2226 BP |
5271 | if (rule == ofproto->miss_rule) { |
5272 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
5273 | } else if (rule == ofproto->no_packet_in_rule) { | |
5274 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
5275 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7fd51d39 BP |
5276 | } else if (rule == ofproto->drop_frags_rule) { |
5277 | ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " | |
5278 | "and the fragment handling mode is \"drop\".\n"); | |
c57b2226 BP |
5279 | } |
5280 | ||
abe529af | 5281 | if (rule) { |
050ac423 BP |
5282 | uint64_t odp_actions_stub[1024 / 8]; |
5283 | struct ofpbuf odp_actions; | |
6a6455e5 | 5284 | struct trace_ctx trace; |
bcd2633a | 5285 | struct match match; |
0e553d9c | 5286 | uint8_t tcp_flags; |
abe529af | 5287 | |
6a6455e5 EJ |
5288 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
5289 | trace.result = ds; | |
5290 | trace.flow = *flow; | |
050ac423 BP |
5291 | ofpbuf_use_stub(&odp_actions, |
5292 | odp_actions_stub, sizeof odp_actions_stub); | |
47b0deae | 5293 | xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); |
bbafd73b EJ |
5294 | trace.xin.resubmit_hook = trace_resubmit; |
5295 | trace.xin.report_hook = trace_report; | |
bcd2633a | 5296 | |
bbafd73b | 5297 | xlate_actions(&trace.xin, &trace.xout); |
083f277e | 5298 | flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc); |
abe529af | 5299 | |
6a6455e5 EJ |
5300 | ds_put_char(ds, '\n'); |
5301 | trace_format_flow(ds, 0, "Final flow", &trace); | |
bcd2633a JP |
5302 | |
5303 | match_init(&match, flow, &trace.xout.wc); | |
5304 | ds_put_cstr(ds, "Relevant fields: "); | |
5305 | match_format(&match, ds, OFP_DEFAULT_PRIORITY); | |
5306 | ds_put_char(ds, '\n'); | |
5307 | ||
6a6455e5 | 5308 | ds_put_cstr(ds, "Datapath actions: "); |
bbafd73b EJ |
5309 | format_odp_actions(ds, trace.xout.odp_actions.data, |
5310 | trace.xout.odp_actions.size); | |
876b0e1c | 5311 | |
bbafd73b | 5312 | if (trace.xout.slow) { |
04594cd5 BP |
5313 | enum slow_path_reason slow; |
5314 | ||
6a7e895f BP |
5315 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
5316 | "slow path because it:"); | |
04594cd5 BP |
5317 | |
5318 | slow = trace.xout.slow; | |
5319 | while (slow) { | |
5320 | enum slow_path_reason bit = rightmost_1bit(slow); | |
5321 | ||
5322 | ds_put_format(ds, "\n\t- %s.", | |
5323 | slow_path_reason_to_explanation(bit)); | |
5324 | ||
5325 | slow &= ~bit; | |
6a7e895f | 5326 | } |
876b0e1c | 5327 | } |
bbafd73b EJ |
5328 | |
5329 | xlate_out_uninit(&trace.xout); | |
abe529af | 5330 | } |
ad3efdcb | 5331 | |
a2143702 | 5332 | rule_dpif_unref(rule); |
abe529af BP |
5333 | } |
5334 | ||
6814e51f BP |
5335 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
5336 | * 'reply' describing the results. */ | |
5337 | static void | |
5338 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
5339 | { | |
bcd2633a | 5340 | struct cls_cursor cursor; |
6814e51f BP |
5341 | struct facet *facet; |
5342 | int errors; | |
5343 | ||
5344 | errors = 0; | |
0b4f2078 | 5345 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5346 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5347 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6814e51f BP |
5348 | if (!facet_check_consistency(facet)) { |
5349 | errors++; | |
5350 | } | |
5351 | } | |
0b4f2078 | 5352 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 5353 | if (errors) { |
2cc3c58e | 5354 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
5355 | } |
5356 | ||
5357 | if (errors) { | |
5358 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
5359 | ofproto->up.name, errors); | |
5360 | } else { | |
5361 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
5362 | } | |
5363 | } | |
5364 | ||
5365 | static void | |
5366 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
5367 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
5368 | { | |
5369 | struct ds reply = DS_EMPTY_INITIALIZER; | |
5370 | struct ofproto_dpif *ofproto; | |
5371 | ||
5372 | if (argc > 1) { | |
5373 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5374 | if (!ofproto) { | |
bde9f75d EJ |
5375 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
5376 | "ofproto/list for help)"); | |
6814e51f BP |
5377 | return; |
5378 | } | |
5379 | ofproto_dpif_self_check__(ofproto, &reply); | |
5380 | } else { | |
5381 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5382 | ofproto_dpif_self_check__(ofproto, &reply); | |
5383 | } | |
5384 | } | |
5385 | ||
bde9f75d | 5386 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
5387 | ds_destroy(&reply); |
5388 | } | |
5389 | ||
27022416 JP |
5390 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
5391 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
5392 | * to destroy 'ofproto_shash' and free the returned value. */ | |
5393 | static const struct shash_node ** | |
5394 | get_ofprotos(struct shash *ofproto_shash) | |
5395 | { | |
5396 | const struct ofproto_dpif *ofproto; | |
5397 | ||
5398 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5399 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
5400 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
5401 | } | |
5402 | ||
5403 | return shash_sort(ofproto_shash); | |
5404 | } | |
5405 | ||
5406 | static void | |
5407 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
5408 | const char *argv[] OVS_UNUSED, | |
5409 | void *aux OVS_UNUSED) | |
5410 | { | |
5411 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5412 | struct shash ofproto_shash; | |
5413 | const struct shash_node **sorted_ofprotos; | |
5414 | int i; | |
5415 | ||
5416 | shash_init(&ofproto_shash); | |
5417 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
5418 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5419 | const struct shash_node *node = sorted_ofprotos[i]; | |
5420 | ds_put_format(&ds, "%s\n", node->name); | |
5421 | } | |
5422 | ||
5423 | shash_destroy(&ofproto_shash); | |
5424 | free(sorted_ofprotos); | |
5425 | ||
5426 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5427 | ds_destroy(&ds); | |
5428 | } | |
5429 | ||
dc54ef36 EJ |
5430 | static void |
5431 | dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) | |
27022416 | 5432 | { |
dc54ef36 EJ |
5433 | const struct shash_node **ofprotos; |
5434 | struct ofproto_dpif *ofproto; | |
5435 | struct shash ofproto_shash; | |
09672174 | 5436 | uint64_t n_hit, n_missed; |
09672174 | 5437 | size_t i; |
655ab909 | 5438 | |
04d08d54 | 5439 | n_hit = n_missed = 0; |
dc54ef36 EJ |
5440 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
5441 | if (ofproto->backer == backer) { | |
dc54ef36 EJ |
5442 | n_missed += ofproto->n_missed; |
5443 | n_hit += ofproto->n_hit; | |
5444 | } | |
5445 | } | |
655ab909 | 5446 | |
dc54ef36 EJ |
5447 | ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
5448 | dpif_name(backer->dpif), n_hit, n_missed); | |
5449 | ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," | |
04d08d54 | 5450 | " life span: %lldms\n", hmap_count(&backer->subfacets), |
dc54ef36 EJ |
5451 | backer->avg_n_subfacet, backer->max_n_subfacet, |
5452 | backer->avg_subfacet_life); | |
5453 | ||
dc54ef36 EJ |
5454 | shash_init(&ofproto_shash); |
5455 | ofprotos = get_ofprotos(&ofproto_shash); | |
5456 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5457 | struct ofproto_dpif *ofproto = ofprotos[i]->data; | |
5458 | const struct shash_node **ports; | |
5459 | size_t j; | |
0a740f48 | 5460 | |
dc54ef36 EJ |
5461 | if (ofproto->backer != backer) { |
5462 | continue; | |
0a740f48 | 5463 | } |
27022416 | 5464 | |
dc54ef36 EJ |
5465 | ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
5466 | ofproto->up.name, ofproto->n_hit, ofproto->n_missed); | |
5467 | ||
5468 | ports = shash_sort(&ofproto->up.port_by_name); | |
5469 | for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { | |
5470 | const struct shash_node *node = ports[j]; | |
5471 | struct ofport *ofport = node->data; | |
5472 | struct smap config; | |
4e022ec0 | 5473 | odp_port_t odp_port; |
27022416 | 5474 | |
dc54ef36 EJ |
5475 | ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), |
5476 | ofport->ofp_port); | |
27022416 | 5477 | |
dc54ef36 | 5478 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); |
4e022ec0 | 5479 | if (odp_port != ODPP_NONE) { |
dc54ef36 EJ |
5480 | ds_put_format(ds, "%"PRIu32":", odp_port); |
5481 | } else { | |
5482 | ds_put_cstr(ds, "none:"); | |
5483 | } | |
27022416 | 5484 | |
dc54ef36 | 5485 | ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); |
27022416 | 5486 | |
dc54ef36 EJ |
5487 | smap_init(&config); |
5488 | if (!netdev_get_config(ofport->netdev, &config)) { | |
5489 | const struct smap_node **nodes; | |
5490 | size_t i; | |
27022416 | 5491 | |
dc54ef36 EJ |
5492 | nodes = smap_sort(&config); |
5493 | for (i = 0; i < smap_count(&config); i++) { | |
5494 | const struct smap_node *node = nodes[i]; | |
5495 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
5496 | node->key, node->value); | |
5497 | } | |
5498 | free(nodes); | |
27022416 | 5499 | } |
dc54ef36 EJ |
5500 | smap_destroy(&config); |
5501 | ||
27022416 | 5502 | ds_put_char(ds, ')'); |
dc54ef36 | 5503 | ds_put_char(ds, '\n'); |
27022416 | 5504 | } |
dc54ef36 | 5505 | free(ports); |
27022416 | 5506 | } |
dc54ef36 EJ |
5507 | shash_destroy(&ofproto_shash); |
5508 | free(ofprotos); | |
27022416 JP |
5509 | } |
5510 | ||
5511 | static void | |
dc54ef36 EJ |
5512 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5513 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
27022416 JP |
5514 | { |
5515 | struct ds ds = DS_EMPTY_INITIALIZER; | |
dc54ef36 EJ |
5516 | const struct shash_node **backers; |
5517 | int i; | |
27022416 | 5518 | |
dc54ef36 EJ |
5519 | backers = shash_sort(&all_dpif_backers); |
5520 | for (i = 0; i < shash_count(&all_dpif_backers); i++) { | |
5521 | dpif_show_backer(backers[i]->data, &ds); | |
27022416 | 5522 | } |
dc54ef36 | 5523 | free(backers); |
27022416 JP |
5524 | |
5525 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5526 | ds_destroy(&ds); | |
5527 | } | |
5528 | ||
bcd2633a JP |
5529 | /* Dump the megaflow (facet) cache. This is useful to check the |
5530 | * correctness of flow wildcarding, since the same mechanism is used for | |
5531 | * both xlate caching and kernel wildcarding. | |
5532 | * | |
5533 | * It's important to note that in the output the flow description uses | |
5534 | * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports. | |
5535 | * | |
5536 | * This command is only needed for advanced debugging, so it's not | |
5537 | * documented in the man page. */ | |
5538 | static void | |
5539 | ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn, | |
5540 | int argc OVS_UNUSED, const char *argv[], | |
5541 | void *aux OVS_UNUSED) | |
5542 | { | |
5543 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5544 | const struct ofproto_dpif *ofproto; | |
5545 | long long int now = time_msec(); | |
5546 | struct cls_cursor cursor; | |
5547 | struct facet *facet; | |
5548 | ||
5549 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5550 | if (!ofproto) { | |
5551 | unixctl_command_reply_error(conn, "no such bridge"); | |
5552 | return; | |
5553 | } | |
5554 | ||
0b4f2078 | 5555 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5556 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5557 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
5558 | cls_rule_format(&facet->cr, &ds); | |
5559 | ds_put_cstr(&ds, ", "); | |
7ee9e841 | 5560 | ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets)); |
bcd2633a JP |
5561 | ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); |
5562 | ds_put_cstr(&ds, "Datapath actions: "); | |
8ecafa3e JP |
5563 | if (facet->xout.slow) { |
5564 | uint64_t slow_path_stub[128 / 8]; | |
5565 | const struct nlattr *actions; | |
5566 | size_t actions_len; | |
5567 | ||
5568 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, | |
5569 | slow_path_stub, sizeof slow_path_stub, | |
5570 | &actions, &actions_len); | |
5571 | format_odp_actions(&ds, actions, actions_len); | |
5572 | } else { | |
5573 | format_odp_actions(&ds, facet->xout.odp_actions.data, | |
5574 | facet->xout.odp_actions.size); | |
5575 | } | |
bcd2633a JP |
5576 | ds_put_cstr(&ds, "\n"); |
5577 | } | |
0b4f2078 | 5578 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5579 | |
5580 | ds_chomp(&ds, '\n'); | |
5581 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5582 | ds_destroy(&ds); | |
5583 | } | |
5584 | ||
6b35e630 JP |
5585 | /* Disable using the megaflows. |
5586 | * | |
5587 | * This command is only needed for advanced debugging, so it's not | |
5588 | * documented in the man page. */ | |
5589 | static void | |
5590 | ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn, | |
5591 | int argc OVS_UNUSED, | |
5592 | const char *argv[] OVS_UNUSED, | |
5593 | void *aux OVS_UNUSED) | |
5594 | { | |
5595 | struct ofproto_dpif *ofproto; | |
5596 | ||
5597 | enable_megaflows = false; | |
5598 | ||
5599 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5600 | flush(&ofproto->up); | |
5601 | } | |
5602 | ||
5603 | unixctl_command_reply(conn, "megaflows disabled"); | |
5604 | } | |
5605 | ||
5606 | /* Re-enable using megaflows. | |
5607 | * | |
5608 | * This command is only needed for advanced debugging, so it's not | |
5609 | * documented in the man page. */ | |
5610 | static void | |
5611 | ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn, | |
5612 | int argc OVS_UNUSED, | |
5613 | const char *argv[] OVS_UNUSED, | |
5614 | void *aux OVS_UNUSED) | |
5615 | { | |
5616 | struct ofproto_dpif *ofproto; | |
5617 | ||
5618 | enable_megaflows = true; | |
5619 | ||
5620 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5621 | flush(&ofproto->up); | |
5622 | } | |
5623 | ||
5624 | unixctl_command_reply(conn, "megaflows enabled"); | |
5625 | } | |
5626 | ||
27022416 JP |
5627 | static void |
5628 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
5629 | int argc OVS_UNUSED, const char *argv[], | |
5630 | void *aux OVS_UNUSED) | |
5631 | { | |
5632 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5633 | const struct ofproto_dpif *ofproto; | |
5634 | struct subfacet *subfacet; | |
5635 | ||
5636 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5637 | if (!ofproto) { | |
5638 | unixctl_command_reply_error(conn, "no such bridge"); | |
5639 | return; | |
5640 | } | |
5641 | ||
af37354d EJ |
5642 | update_stats(ofproto->backer); |
5643 | ||
04d08d54 | 5644 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { |
4dff9097 | 5645 | struct facet *facet = subfacet->facet; |
64f9cb76 JP |
5646 | struct odputil_keybuf maskbuf; |
5647 | struct ofpbuf mask; | |
4dff9097 | 5648 | |
b7e2f6b3 | 5649 | if (facet->ofproto != ofproto) { |
04d08d54 EJ |
5650 | continue; |
5651 | } | |
5652 | ||
64f9cb76 JP |
5653 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
5654 | if (enable_megaflows) { | |
5655 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
5656 | &facet->flow, UINT32_MAX); | |
5657 | } | |
5658 | ||
5659 | odp_flow_format(subfacet->key, subfacet->key_len, | |
0a37839c | 5660 | mask.data, mask.size, NULL, &ds, false); |
27022416 JP |
5661 | |
5662 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
5663 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
5664 | if (subfacet->used) { | |
5665 | ds_put_format(&ds, "%.3fs", | |
5666 | (time_msec() - subfacet->used) / 1000.0); | |
5667 | } else { | |
5668 | ds_put_format(&ds, "never"); | |
5669 | } | |
5670 | if (subfacet->facet->tcp_flags) { | |
5671 | ds_put_cstr(&ds, ", flags:"); | |
5672 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
5673 | } | |
5674 | ||
5675 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 5676 | if (facet->xout.slow) { |
f2245da3 JP |
5677 | uint64_t slow_path_stub[128 / 8]; |
5678 | const struct nlattr *actions; | |
5679 | size_t actions_len; | |
5680 | ||
bbafd73b | 5681 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
5682 | slow_path_stub, sizeof slow_path_stub, |
5683 | &actions, &actions_len); | |
5684 | format_odp_actions(&ds, actions, actions_len); | |
5685 | } else { | |
bbafd73b EJ |
5686 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
5687 | facet->xout.odp_actions.size); | |
f2245da3 | 5688 | } |
27022416 JP |
5689 | ds_put_char(&ds, '\n'); |
5690 | } | |
5691 | ||
5692 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5693 | ds_destroy(&ds); | |
5694 | } | |
5695 | ||
5696 | static void | |
5697 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
5698 | int argc OVS_UNUSED, const char *argv[], | |
5699 | void *aux OVS_UNUSED) | |
5700 | { | |
5701 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5702 | struct ofproto_dpif *ofproto; | |
5703 | ||
5704 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5705 | if (!ofproto) { | |
5706 | unixctl_command_reply_error(conn, "no such bridge"); | |
5707 | return; | |
5708 | } | |
5709 | ||
5710 | flush(&ofproto->up); | |
5711 | ||
5712 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5713 | ds_destroy(&ds); | |
5714 | } | |
5715 | ||
abe529af BP |
5716 | static void |
5717 | ofproto_dpif_unixctl_init(void) | |
5718 | { | |
5719 | static bool registered; | |
5720 | if (registered) { | |
5721 | return; | |
5722 | } | |
5723 | registered = true; | |
5724 | ||
0e15264f BP |
5725 | unixctl_command_register( |
5726 | "ofproto/trace", | |
50aa28fd AW |
5727 | "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", |
5728 | 1, 3, ofproto_unixctl_trace, NULL); | |
96e466a3 | 5729 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
5730 | ofproto_unixctl_fdb_flush, NULL); |
5731 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
5732 | ofproto_unixctl_fdb_show, NULL); | |
6814e51f BP |
5733 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
5734 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
5735 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
5736 | ofproto_unixctl_dpif_dump_dps, NULL); | |
dc54ef36 EJ |
5737 | unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show, |
5738 | NULL); | |
27022416 JP |
5739 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, |
5740 | ofproto_unixctl_dpif_dump_flows, NULL); | |
5741 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
5742 | ofproto_unixctl_dpif_del_flows, NULL); | |
bcd2633a JP |
5743 | unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1, |
5744 | ofproto_unixctl_dpif_dump_megaflows, NULL); | |
6b35e630 JP |
5745 | unixctl_command_register("dpif/disable-megaflows", "", 0, 0, |
5746 | ofproto_unixctl_dpif_disable_megaflows, NULL); | |
5747 | unixctl_command_register("dpif/enable-megaflows", "", 0, 0, | |
5748 | ofproto_unixctl_dpif_enable_megaflows, NULL); | |
abe529af BP |
5749 | } |
5750 | \f | |
52a90c29 BP |
5751 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
5752 | * | |
5753 | * This is deprecated. It is only for compatibility with broken device drivers | |
5754 | * in old versions of Linux that do not properly support VLANs when VLAN | |
5755 | * devices are not used. When broken device drivers are no longer in | |
5756 | * widespread use, we will delete these interfaces. */ | |
5757 | ||
5758 | static int | |
4e022ec0 | 5759 | set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
5760 | { |
5761 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
5762 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
5763 | ||
5764 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
5765 | && vid == ofport->vlandev_vid) { | |
5766 | return 0; | |
5767 | } | |
5768 | ||
2cc3c58e | 5769 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
5770 | |
5771 | if (ofport->realdev_ofp_port) { | |
5772 | vsp_remove(ofport); | |
5773 | } | |
5774 | if (realdev_ofp_port && ofport->bundle) { | |
5775 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
5776 | * themselves be part of a bundle. */ | |
5777 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
5778 | } | |
5779 | ||
5780 | ofport->realdev_ofp_port = realdev_ofp_port; | |
5781 | ofport->vlandev_vid = vid; | |
5782 | ||
5783 | if (realdev_ofp_port) { | |
5784 | vsp_add(ofport, realdev_ofp_port, vid); | |
5785 | } | |
5786 | ||
5787 | return 0; | |
5788 | } | |
5789 | ||
5790 | static uint32_t | |
4e022ec0 | 5791 | hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid) |
52a90c29 | 5792 | { |
4e022ec0 | 5793 | return hash_2words(ofp_to_u16(realdev_ofp_port), vid); |
52a90c29 BP |
5794 | } |
5795 | ||
46c88433 EJ |
5796 | bool |
5797 | ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto) | |
13501305 | 5798 | OVS_EXCLUDED(ofproto->vsp_mutex) |
46c88433 | 5799 | { |
13501305 EJ |
5800 | bool ret; |
5801 | ||
5802 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
5803 | ret = !hmap_is_empty(&ofproto->realdev_vid_map); | |
5804 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
5805 | return ret; | |
46c88433 EJ |
5806 | } |
5807 | ||
13501305 EJ |
5808 | static ofp_port_t |
5809 | vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto, | |
5810 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
5811 | OVS_REQUIRES(ofproto->vsp_mutex) | |
52a90c29 BP |
5812 | { |
5813 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
52a90c29 BP |
5814 | int vid = vlan_tci_to_vid(vlan_tci); |
5815 | const struct vlan_splinter *vsp; | |
5816 | ||
5817 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
5818 | hash_realdev_vid(realdev_ofp_port, vid), | |
5819 | &ofproto->realdev_vid_map) { | |
5820 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
5821 | && vsp->vid == vid) { | |
deea1200 | 5822 | return vsp->vlandev_ofp_port; |
52a90c29 BP |
5823 | } |
5824 | } | |
5825 | } | |
deea1200 | 5826 | return realdev_ofp_port; |
52a90c29 BP |
5827 | } |
5828 | ||
13501305 EJ |
5829 | /* Returns the OFP port number of the Linux VLAN device that corresponds to |
5830 | * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in | |
5831 | * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and | |
5832 | * 'vlan_tci' 9, it would return the port number of eth0.9. | |
5833 | * | |
5834 | * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this | |
5835 | * function just returns its 'realdev_ofp_port' argument. */ | |
5836 | ofp_port_t | |
5837 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
5838 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
5839 | OVS_EXCLUDED(ofproto->vsp_mutex) | |
5840 | { | |
5841 | ofp_port_t ret; | |
5842 | ||
5843 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
5844 | ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci); | |
5845 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
5846 | return ret; | |
5847 | } | |
5848 | ||
52a90c29 | 5849 | static struct vlan_splinter * |
4e022ec0 | 5850 | vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) |
52a90c29 BP |
5851 | { |
5852 | struct vlan_splinter *vsp; | |
5853 | ||
4e022ec0 | 5854 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, |
f9c0c3ec | 5855 | hash_ofp_port(vlandev_ofp_port), |
52a90c29 BP |
5856 | &ofproto->vlandev_map) { |
5857 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
5858 | return vsp; | |
5859 | } | |
5860 | } | |
5861 | ||
5862 | return NULL; | |
5863 | } | |
5864 | ||
40e05935 BP |
5865 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
5866 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
5867 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
5868 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
5869 | * eth0 and store 9 in '*vid'. | |
5870 | * | |
5871 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
5872 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
5873 | * always does.*/ | |
4e022ec0 | 5874 | static ofp_port_t |
52a90c29 | 5875 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, |
4e022ec0 | 5876 | ofp_port_t vlandev_ofp_port, int *vid) |
bd3950dd | 5877 | OVS_REQUIRES(ofproto->vsp_mutex) |
52a90c29 BP |
5878 | { |
5879 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
5880 | const struct vlan_splinter *vsp; | |
5881 | ||
5882 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
5883 | if (vsp) { | |
5884 | if (vid) { | |
5885 | *vid = vsp->vid; | |
5886 | } | |
5887 | return vsp->realdev_ofp_port; | |
5888 | } | |
5889 | } | |
5890 | return 0; | |
5891 | } | |
5892 | ||
b98d8985 BP |
5893 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
5894 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
5895 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
5896 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
5897 | * always the case unless VLAN splinters are enabled), returns false without | |
5898 | * making any changes. */ | |
8449c4d6 | 5899 | bool |
b98d8985 | 5900 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) |
13501305 | 5901 | OVS_EXCLUDED(ofproto->vsp_mutex) |
b98d8985 | 5902 | { |
4e022ec0 | 5903 | ofp_port_t realdev; |
b98d8985 BP |
5904 | int vid; |
5905 | ||
13501305 | 5906 | ovs_mutex_lock(&ofproto->vsp_mutex); |
4e022ec0 | 5907 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid); |
13501305 | 5908 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
b98d8985 BP |
5909 | if (!realdev) { |
5910 | return false; | |
5911 | } | |
5912 | ||
5913 | /* Cause the flow to be processed as if it came in on the real device with | |
5914 | * the VLAN device's VLAN ID. */ | |
4e022ec0 | 5915 | flow->in_port.ofp_port = realdev; |
b98d8985 BP |
5916 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); |
5917 | return true; | |
5918 | } | |
5919 | ||
52a90c29 BP |
5920 | static void |
5921 | vsp_remove(struct ofport_dpif *port) | |
5922 | { | |
5923 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
5924 | struct vlan_splinter *vsp; | |
5925 | ||
13501305 | 5926 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 BP |
5927 | vsp = vlandev_find(ofproto, port->up.ofp_port); |
5928 | if (vsp) { | |
5929 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
5930 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
5931 | free(vsp); | |
5932 | ||
5933 | port->realdev_ofp_port = 0; | |
5934 | } else { | |
5935 | VLOG_ERR("missing vlan device record"); | |
5936 | } | |
13501305 | 5937 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 BP |
5938 | } |
5939 | ||
5940 | static void | |
4e022ec0 | 5941 | vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
5942 | { |
5943 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
5944 | ||
13501305 | 5945 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 | 5946 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) |
13501305 | 5947 | && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid)) |
52a90c29 BP |
5948 | == realdev_ofp_port)) { |
5949 | struct vlan_splinter *vsp; | |
5950 | ||
5951 | vsp = xmalloc(sizeof *vsp); | |
52a90c29 BP |
5952 | vsp->realdev_ofp_port = realdev_ofp_port; |
5953 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
5954 | vsp->vid = vid; | |
5955 | ||
5956 | port->realdev_ofp_port = realdev_ofp_port; | |
13501305 EJ |
5957 | |
5958 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
5959 | hash_ofp_port(port->up.ofp_port)); | |
5960 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
5961 | hash_realdev_vid(realdev_ofp_port, vid)); | |
52a90c29 BP |
5962 | } else { |
5963 | VLOG_ERR("duplicate vlan device record"); | |
5964 | } | |
13501305 | 5965 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 | 5966 | } |
e1b1d06a | 5967 | |
46c88433 | 5968 | static odp_port_t |
4e022ec0 | 5969 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
e1b1d06a JP |
5970 | { |
5971 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
4e022ec0 | 5972 | return ofport ? ofport->odp_port : ODPP_NONE; |
e1b1d06a JP |
5973 | } |
5974 | ||
8449c4d6 | 5975 | struct ofport_dpif * |
4e022ec0 | 5976 | odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port) |
e1b1d06a JP |
5977 | { |
5978 | struct ofport_dpif *port; | |
5979 | ||
8449c4d6 | 5980 | ovs_rwlock_rdlock(&backer->odp_to_ofport_lock); |
f9c0c3ec | 5981 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port), |
acf60855 | 5982 | &backer->odp_to_ofport_map) { |
e1b1d06a | 5983 | if (port->odp_port == odp_port) { |
8449c4d6 | 5984 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 | 5985 | return port; |
e1b1d06a JP |
5986 | } |
5987 | } | |
5988 | ||
8449c4d6 | 5989 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 JP |
5990 | return NULL; |
5991 | } | |
5992 | ||
4e022ec0 AW |
5993 | static ofp_port_t |
5994 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) | |
acf60855 JP |
5995 | { |
5996 | struct ofport_dpif *port; | |
5997 | ||
5998 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 5999 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
6000 | return port->up.ofp_port; |
6001 | } else { | |
6002 | return OFPP_NONE; | |
6003 | } | |
e1b1d06a | 6004 | } |
655ab909 | 6005 | |
abe529af | 6006 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 6007 | init, |
abe529af BP |
6008 | enumerate_types, |
6009 | enumerate_names, | |
6010 | del, | |
0aeaabc8 | 6011 | port_open_type, |
acf60855 JP |
6012 | type_run, |
6013 | type_run_fast, | |
6014 | type_wait, | |
abe529af BP |
6015 | alloc, |
6016 | construct, | |
6017 | destruct, | |
6018 | dealloc, | |
6019 | run, | |
5fcc0d00 | 6020 | run_fast, |
abe529af | 6021 | wait, |
0d085684 | 6022 | get_memory_usage, |
abe529af | 6023 | flush, |
6c1491fb BP |
6024 | get_features, |
6025 | get_tables, | |
abe529af BP |
6026 | port_alloc, |
6027 | port_construct, | |
6028 | port_destruct, | |
6029 | port_dealloc, | |
6030 | port_modified, | |
6031 | port_reconfigured, | |
6032 | port_query_by_name, | |
6033 | port_add, | |
6034 | port_del, | |
6527c598 | 6035 | port_get_stats, |
abe529af BP |
6036 | port_dump_start, |
6037 | port_dump_next, | |
6038 | port_dump_done, | |
6039 | port_poll, | |
6040 | port_poll_wait, | |
6041 | port_is_lacp_current, | |
0ab6decf | 6042 | NULL, /* rule_choose_table */ |
abe529af BP |
6043 | rule_alloc, |
6044 | rule_construct, | |
8037acb4 BP |
6045 | rule_insert, |
6046 | rule_delete, | |
abe529af BP |
6047 | rule_destruct, |
6048 | rule_dealloc, | |
abe529af BP |
6049 | rule_get_stats, |
6050 | rule_execute, | |
6051 | rule_modify_actions, | |
7257b535 | 6052 | set_frag_handling, |
abe529af BP |
6053 | packet_out, |
6054 | set_netflow, | |
6055 | get_netflow_ids, | |
6056 | set_sflow, | |
29089a54 | 6057 | set_ipfix, |
abe529af | 6058 | set_cfm, |
9a9e3786 | 6059 | get_cfm_status, |
ccc09689 EJ |
6060 | set_bfd, |
6061 | get_bfd_status, | |
21f7563c JP |
6062 | set_stp, |
6063 | get_stp_status, | |
6064 | set_stp_port, | |
6065 | get_stp_port_status, | |
8b36f51e | 6066 | set_queues, |
abe529af BP |
6067 | bundle_set, |
6068 | bundle_remove, | |
ec7ceaed EJ |
6069 | mirror_set__, |
6070 | mirror_get_stats__, | |
abe529af BP |
6071 | set_flood_vlans, |
6072 | is_mirror_output_bundle, | |
8402c74b | 6073 | forward_bpdu_changed, |
c4069512 | 6074 | set_mac_table_config, |
52a90c29 | 6075 | set_realdev, |
9cae45dc JR |
6076 | NULL, /* meter_get_features */ |
6077 | NULL, /* meter_set */ | |
6078 | NULL, /* meter_get */ | |
6079 | NULL, /* meter_del */ | |
7395c052 NZ |
6080 | NULL, /* group_alloc */ |
6081 | NULL, /* group_construct */ | |
6082 | NULL, /* group_destruct */ | |
6083 | NULL, /* group_dealloc */ | |
6084 | NULL, /* group_modify */ | |
6085 | NULL, /* group_get_stats */ | |
abe529af | 6086 | }; |