]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e09ee259 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
9583bc14 | 19 | #include "ofproto/ofproto-dpif.h" |
5bee6e26 | 20 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
21 | |
22 | #include <errno.h> | |
23 | ||
ccc09689 | 24 | #include "bfd.h" |
abe529af | 25 | #include "bond.h" |
daff3353 | 26 | #include "bundle.h" |
abe529af BP |
27 | #include "byte-order.h" |
28 | #include "connmgr.h" | |
29 | #include "coverage.h" | |
30 | #include "cfm.h" | |
31 | #include "dpif.h" | |
32 | #include "dynamic-string.h" | |
33 | #include "fail-open.h" | |
34 | #include "hmapx.h" | |
35 | #include "lacp.h" | |
75a75043 | 36 | #include "learn.h" |
abe529af | 37 | #include "mac-learning.h" |
816fd533 | 38 | #include "meta-flow.h" |
abe529af | 39 | #include "multipath.h" |
0a740f48 | 40 | #include "netdev-vport.h" |
abe529af BP |
41 | #include "netdev.h" |
42 | #include "netlink.h" | |
43 | #include "nx-match.h" | |
44 | #include "odp-util.h" | |
1ac7c9bd | 45 | #include "odp-execute.h" |
abe529af BP |
46 | #include "ofp-util.h" |
47 | #include "ofpbuf.h" | |
f25d0cf3 | 48 | #include "ofp-actions.h" |
31a19d69 | 49 | #include "ofp-parse.h" |
abe529af | 50 | #include "ofp-print.h" |
9d6ac44e | 51 | #include "ofproto-dpif-governor.h" |
29089a54 | 52 | #include "ofproto-dpif-ipfix.h" |
ec7ceaed | 53 | #include "ofproto-dpif-mirror.h" |
bae473fe | 54 | #include "ofproto-dpif-sflow.h" |
9583bc14 | 55 | #include "ofproto-dpif-xlate.h" |
abe529af | 56 | #include "poll-loop.h" |
0d085684 | 57 | #include "simap.h" |
27022416 | 58 | #include "smap.h" |
abe529af | 59 | #include "timer.h" |
b9ad7294 | 60 | #include "tunnel.h" |
6c1491fb | 61 | #include "unaligned.h" |
abe529af BP |
62 | #include "unixctl.h" |
63 | #include "vlan-bitmap.h" | |
64 | #include "vlog.h" | |
65 | ||
66 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
67 | ||
abe529af | 68 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af | 69 | COVERAGE_DEFINE(facet_changed_rule); |
abe529af BP |
70 | COVERAGE_DEFINE(facet_revalidate); |
71 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 72 | COVERAGE_DEFINE(facet_suppress); |
9b1a48c2 | 73 | COVERAGE_DEFINE(subfacet_install_fail); |
abe529af | 74 | |
46c88433 EJ |
75 | /* Number of implemented OpenFlow tables. */ |
76 | enum { N_TABLES = 255 }; | |
77 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
78 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
79 | ||
a088a1ff | 80 | struct flow_miss; |
ac35f9c8 | 81 | struct facet; |
abe529af | 82 | |
29901626 | 83 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
bcd2633a JP |
84 | const struct flow *, |
85 | struct flow_wildcards *wc); | |
abe529af | 86 | |
7e741ffb | 87 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
b0f7b9b5 | 88 | |
46c88433 EJ |
89 | struct ofbundle { |
90 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
91 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
92 | void *aux; /* Key supplied by ofproto's client. */ | |
93 | char *name; /* Identifier for log messages. */ | |
94 | ||
95 | /* Configuration. */ | |
96 | struct list ports; /* Contains "struct ofport"s. */ | |
97 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
98 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
99 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
100 | * NULL if all VLANs are trunked. */ | |
101 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
102 | struct bond *bond; /* Nonnull iff more than one port. */ | |
103 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
104 | ||
105 | /* Status. */ | |
106 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
107 | }; | |
108 | ||
abe529af | 109 | static void bundle_remove(struct ofport *); |
7bde8dd8 | 110 | static void bundle_update(struct ofbundle *); |
abe529af BP |
111 | static void bundle_destroy(struct ofbundle *); |
112 | static void bundle_del_port(struct ofport_dpif *); | |
113 | static void bundle_run(struct ofbundle *); | |
114 | static void bundle_wait(struct ofbundle *); | |
33158a18 | 115 | |
21f7563c JP |
116 | static void stp_run(struct ofproto_dpif *ofproto); |
117 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
118 | static int set_stp_port(struct ofport *, |
119 | const struct ofproto_port_stp_settings *); | |
21f7563c | 120 | |
6a7e895f BP |
121 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, |
122 | enum slow_path_reason, | |
123 | uint64_t *stub, size_t stub_size, | |
124 | const struct nlattr **actionsp, | |
125 | size_t *actions_lenp); | |
126 | ||
127 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
128 | * states: | |
129 | * | |
130 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
131 | * case just after the subfacet is created, just before the subfacet is | |
132 | * destroyed, or if the datapath returns an error when we try to install a | |
133 | * subfacet. | |
134 | * | |
135 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
136 | * | |
137 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
138 | * ofproto_dpif is installed in the datapath. | |
139 | */ | |
140 | enum subfacet_path { | |
141 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
142 | SF_FAST_PATH, /* Full actions are installed. */ | |
143 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
144 | }; | |
145 | ||
5f5fbd17 BP |
146 | /* A dpif flow and actions associated with a facet. |
147 | * | |
148 | * See also the large comment on struct facet. */ | |
149 | struct subfacet { | |
150 | /* Owners. */ | |
151 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
152 | struct list list_node; /* In struct facet's 'facets' list. */ | |
153 | struct facet *facet; /* Owning facet. */ | |
04d08d54 | 154 | struct dpif_backer *backer; /* Owning backer. */ |
5f5fbd17 | 155 | |
5f5fbd17 BP |
156 | enum odp_key_fitness key_fitness; |
157 | struct nlattr *key; | |
158 | int key_len; | |
159 | ||
160 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 161 | long long int created; /* Time created. */ |
5f5fbd17 BP |
162 | |
163 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
164 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
165 | ||
6a7e895f | 166 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 BP |
167 | }; |
168 | ||
1d85f9e5 JP |
169 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
170 | ||
a088a1ff | 171 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 172 | long long int now); |
04d08d54 | 173 | static struct subfacet *subfacet_find(struct dpif_backer *, |
acf60855 | 174 | const struct nlattr *key, size_t key_len, |
9566abf9 | 175 | uint32_t key_hash); |
5f5fbd17 BP |
176 | static void subfacet_destroy(struct subfacet *); |
177 | static void subfacet_destroy__(struct subfacet *); | |
04d08d54 | 178 | static void subfacet_destroy_batch(struct dpif_backer *, |
1d85f9e5 | 179 | struct subfacet **, int n); |
5f5fbd17 BP |
180 | static void subfacet_reset_dp_stats(struct subfacet *, |
181 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
182 | static void subfacet_update_stats(struct subfacet *, |
183 | const struct dpif_flow_stats *); | |
5f5fbd17 | 184 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
185 | const struct ofpbuf *odp_actions, |
186 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
187 | static void subfacet_uninstall(struct subfacet *); |
188 | ||
bcd2633a | 189 | /* A unique, non-overlapping instantiation of an OpenFlow flow. |
b0f7b9b5 BP |
190 | * |
191 | * A facet associates a "struct flow", which represents the Open vSwitch | |
bcd2633a JP |
192 | * userspace idea of an exact-match flow, with one or more subfacets. |
193 | * While the facet is created based on an exact-match flow, it is stored | |
194 | * within the ofproto based on the wildcards that could be expressed | |
195 | * based on the flow table and other configuration. (See the 'wc' | |
196 | * description in "struct xlate_out" for more details.) | |
b0f7b9b5 | 197 | * |
bcd2633a JP |
198 | * Each subfacet tracks the datapath's idea of the flow equivalent to |
199 | * the facet. When the kernel module (or other dpif implementation) and | |
200 | * Open vSwitch userspace agree on the definition of a flow key, there | |
201 | * is exactly one subfacet per facet. If the dpif implementation | |
202 | * supports more-specific flow matching than userspace, however, a facet | |
203 | * can have more than one subfacet. Examples include the dpif | |
204 | * implementation not supporting the same wildcards as userspace or some | |
205 | * distinction in flow that userspace simply doesn't understand. | |
206 | * | |
207 | * Flow expiration works in terms of subfacets, so a facet must have at | |
208 | * least one subfacet or it will never expire, leaking memory. */ | |
abe529af | 209 | struct facet { |
b0f7b9b5 BP |
210 | /* Owners. */ |
211 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
b7e2f6b3 | 212 | struct ofproto_dpif *ofproto; |
b0f7b9b5 BP |
213 | |
214 | /* Owned data. */ | |
215 | struct list subfacets; | |
abe529af BP |
216 | long long int used; /* Time last used; time created if not used. */ |
217 | ||
b0f7b9b5 | 218 | /* Key. */ |
bcd2633a JP |
219 | struct flow flow; /* Flow of the creating subfacet. */ |
220 | struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */ | |
b0f7b9b5 | 221 | |
abe529af BP |
222 | /* These statistics: |
223 | * | |
224 | * - Do include packets and bytes sent "by hand", e.g. with | |
225 | * dpif_execute(). | |
226 | * | |
227 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 228 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 229 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
230 | * |
231 | * - Do not include packets or bytes that can be obtained from the | |
232 | * datapath for any existing subfacet. | |
abe529af BP |
233 | */ |
234 | uint64_t packet_count; /* Number of packets received. */ | |
235 | uint64_t byte_count; /* Number of bytes received. */ | |
236 | ||
b0f7b9b5 | 237 | /* Resubmit statistics. */ |
9d24de3b JP |
238 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
239 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
240 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 241 | |
b0f7b9b5 | 242 | /* Accounting. */ |
907a4c5e | 243 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 244 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 245 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 246 | |
bbafd73b | 247 | struct xlate_out xout; |
b7e2f6b3 | 248 | bool fail_open; /* Facet matched the fail open rule. */ |
4dff9097 | 249 | |
26cd7e34 BP |
250 | /* Storage for a single subfacet, to reduce malloc() time and space |
251 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
252 | * case has exactly one subfacet. However, 'one_subfacet' may not |
253 | * always be valid, since it could have been removed after newer | |
254 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 255 | struct subfacet one_subfacet; |
6cf474d7 EJ |
256 | |
257 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
258 | }; |
259 | ||
bcd2633a JP |
260 | static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *, |
261 | struct xlate_out *, | |
262 | struct dpif_flow_stats *); | |
15baa734 | 263 | static void facet_remove(struct facet *); |
abe529af BP |
264 | static void facet_free(struct facet *); |
265 | ||
bcd2633a | 266 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); |
abe529af | 267 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
bcd2633a | 268 | const struct flow *); |
5bf64ade | 269 | static bool facet_revalidate(struct facet *); |
6814e51f | 270 | static bool facet_check_consistency(struct facet *); |
abe529af | 271 | |
15baa734 | 272 | static void facet_flush_stats(struct facet *); |
abe529af | 273 | |
bbb5d219 | 274 | static void facet_reset_counters(struct facet *); |
9dfb1f78 | 275 | static void facet_push_stats(struct facet *, bool may_learn); |
3de9590b BP |
276 | static void facet_learn(struct facet *); |
277 | static void facet_account(struct facet *); | |
8844e035 | 278 | static void push_all_stats(void); |
abe529af BP |
279 | |
280 | static bool facet_is_controller_flow(struct facet *); | |
281 | ||
46c88433 EJ |
282 | struct ofport_dpif { |
283 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ | |
284 | struct ofport up; | |
285 | ||
286 | odp_port_t odp_port; | |
287 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
288 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
289 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
290 | struct bfd *bfd; /* BFD, if any. */ | |
46c88433 EJ |
291 | bool may_enable; /* May be enabled in bonds. */ |
292 | bool is_tunnel; /* This port is a tunnel. */ | |
293 | long long int carrier_seq; /* Carrier status changes. */ | |
294 | struct ofport_dpif *peer; /* Peer if patch port. */ | |
295 | ||
296 | /* Spanning tree. */ | |
297 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
298 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
299 | long long int stp_state_entered; | |
300 | ||
55954f6e EJ |
301 | /* Queue to DSCP mapping. */ |
302 | struct ofproto_port_queue *qdscp; | |
303 | size_t n_qdscp; | |
46c88433 EJ |
304 | |
305 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
306 | * | |
307 | * This is deprecated. It is only for compatibility with broken device | |
308 | * drivers in old versions of Linux that do not properly support VLANs when | |
309 | * VLAN devices are not used. When broken device drivers are no longer in | |
310 | * widespread use, we will delete these interfaces. */ | |
311 | ofp_port_t realdev_ofp_port; | |
312 | int vlandev_vid; | |
313 | }; | |
314 | ||
52a90c29 BP |
315 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
316 | * | |
317 | * This is deprecated. It is only for compatibility with broken device drivers | |
318 | * in old versions of Linux that do not properly support VLANs when VLAN | |
319 | * devices are not used. When broken device drivers are no longer in | |
320 | * widespread use, we will delete these interfaces. */ | |
321 | struct vlan_splinter { | |
322 | struct hmap_node realdev_vid_node; | |
323 | struct hmap_node vlandev_node; | |
4e022ec0 AW |
324 | ofp_port_t realdev_ofp_port; |
325 | ofp_port_t vlandev_ofp_port; | |
52a90c29 BP |
326 | int vid; |
327 | }; | |
328 | ||
b98d8985 | 329 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 | 330 | static void vsp_remove(struct ofport_dpif *); |
4e022ec0 | 331 | static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid); |
52a90c29 | 332 | |
46c88433 EJ |
333 | static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
334 | ofp_port_t); | |
335 | ||
4e022ec0 | 336 | static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *, |
46c88433 | 337 | odp_port_t); |
e1b1d06a | 338 | |
abe529af BP |
339 | static struct ofport_dpif * |
340 | ofport_dpif_cast(const struct ofport *ofport) | |
341 | { | |
abe529af BP |
342 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
343 | } | |
344 | ||
345 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 346 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 347 | static void port_wait(struct ofport_dpif *); |
8aee94b6 | 348 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 349 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
6cbbf4fa | 350 | static void ofport_update_peer(struct ofport_dpif *); |
8fa4d1d0 | 351 | static void run_fast_rl(void); |
abe529af | 352 | |
7ee20df1 BP |
353 | struct dpif_completion { |
354 | struct list list_node; | |
355 | struct ofoperation *op; | |
356 | }; | |
357 | ||
cdc3ab65 EJ |
358 | /* Reasons that we might need to revalidate every facet, and corresponding |
359 | * coverage counters. | |
360 | * | |
361 | * A value of 0 means that there is no need to revalidate. | |
362 | * | |
363 | * It would be nice to have some cleaner way to integrate with coverage | |
364 | * counters, but with only a few reasons I guess this is good enough for | |
365 | * now. */ | |
366 | enum revalidate_reason { | |
367 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
368 | REV_STP, /* Spanning tree protocol port status change. */ | |
4a1b8f30 | 369 | REV_BOND, /* Bonding changed. */ |
cdc3ab65 EJ |
370 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ |
371 | REV_FLOW_TABLE, /* Flow table changed. */ | |
30618594 | 372 | REV_MAC_LEARNING, /* Mac learning changed. */ |
cdc3ab65 EJ |
373 | REV_INCONSISTENCY /* Facet self-check failed. */ |
374 | }; | |
3c4a309c BP |
375 | COVERAGE_DEFINE(rev_reconfigure); |
376 | COVERAGE_DEFINE(rev_stp); | |
4a1b8f30 | 377 | COVERAGE_DEFINE(rev_bond); |
3c4a309c BP |
378 | COVERAGE_DEFINE(rev_port_toggled); |
379 | COVERAGE_DEFINE(rev_flow_table); | |
30618594 | 380 | COVERAGE_DEFINE(rev_mac_learning); |
3c4a309c BP |
381 | COVERAGE_DEFINE(rev_inconsistency); |
382 | ||
8f73d537 EJ |
383 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
384 | * These are datapath flows which have no associated ofproto, if they did we | |
385 | * would use facets. */ | |
386 | struct drop_key { | |
387 | struct hmap_node hmap_node; | |
388 | struct nlattr *key; | |
389 | size_t key_len; | |
390 | }; | |
391 | ||
cdc3ab65 EJ |
392 | struct avg_subfacet_rates { |
393 | double add_rate; /* Moving average of new flows created per minute. */ | |
394 | double del_rate; /* Moving average of flows deleted per minute. */ | |
395 | }; | |
396 | ||
397 | /* All datapaths of a given type share a single dpif backer instance. */ | |
398 | struct dpif_backer { | |
399 | char *type; | |
400 | int refcount; | |
401 | struct dpif *dpif; | |
402 | struct timer next_expiration; | |
403 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
404 | ||
405 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ | |
406 | ||
407 | /* Facet revalidation flags applying to facets which use this backer. */ | |
408 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
cdc3ab65 EJ |
409 | |
410 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
411 | bool recv_set_enable; /* Enables or disables receiving packets. */ | |
412 | ||
413 | struct hmap subfacets; | |
414 | struct governor *governor; | |
415 | ||
416 | /* Subfacet statistics. | |
417 | * | |
418 | * These keep track of the total number of subfacets added and deleted and | |
419 | * flow life span. They are useful for computing the flow rates stats | |
420 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
421 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
422 | * performance in new situations. */ | |
423 | long long int created; /* Time when it is created. */ | |
424 | unsigned max_n_subfacet; /* Maximum number of flows */ | |
425 | unsigned avg_n_subfacet; /* Average number of flows. */ | |
426 | long long int avg_subfacet_life; /* Average life span of subfacets. */ | |
427 | ||
428 | /* The average number of subfacets... */ | |
429 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
430 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
431 | struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */ | |
432 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
433 | ||
434 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
435 | unsigned subfacet_add_count; | |
436 | unsigned subfacet_del_count; | |
437 | ||
438 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
439 | unsigned long long int total_subfacet_add_count; | |
440 | unsigned long long int total_subfacet_del_count; | |
441 | }; | |
442 | ||
acf60855 JP |
443 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ |
444 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
445 | ||
8f73d537 | 446 | static void drop_key_clear(struct dpif_backer *); |
acf60855 | 447 | static struct ofport_dpif * |
4e022ec0 | 448 | odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port); |
dc54ef36 | 449 | static void update_moving_averages(struct dpif_backer *backer); |
735d7efb | 450 | |
46c88433 EJ |
451 | struct ofproto_dpif { |
452 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
453 | struct ofproto up; | |
454 | struct dpif_backer *backer; | |
455 | ||
456 | /* Special OpenFlow rules. */ | |
457 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
458 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
459 | struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ | |
460 | ||
461 | /* Bridging. */ | |
462 | struct netflow *netflow; | |
463 | struct dpif_sflow *sflow; | |
464 | struct dpif_ipfix *ipfix; | |
465 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
466 | struct mac_learning *ml; | |
467 | bool has_bonded_bundles; | |
468 | struct mbridge *mbridge; | |
469 | ||
470 | /* Facets. */ | |
471 | struct classifier facets; /* Contains 'struct facet's. */ | |
472 | long long int consistency_rl; | |
473 | ||
46c88433 EJ |
474 | /* Support for debugging async flow mods. */ |
475 | struct list completions; | |
476 | ||
477 | struct netdev_stats stats; /* To account packets generated and consumed in | |
478 | * userspace. */ | |
479 | ||
480 | /* Spanning tree. */ | |
481 | struct stp *stp; | |
482 | long long int stp_last_tick; | |
483 | ||
484 | /* VLAN splinters. */ | |
485 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
486 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
487 | ||
488 | /* Ports. */ | |
489 | struct sset ports; /* Set of standard port names. */ | |
490 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
491 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ | |
492 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
493 | ||
494 | /* Per ofproto's dpif stats. */ | |
495 | uint64_t n_hit; | |
496 | uint64_t n_missed; | |
497 | }; | |
498 | ||
7ee20df1 BP |
499 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
500 | * for debugging the asynchronous flow_mod implementation.) */ | |
501 | static bool clogged; | |
502 | ||
6b35e630 JP |
503 | /* By default, flows in the datapath are wildcarded (megaflows). They |
504 | * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ | |
505 | static bool enable_megaflows = true; | |
506 | ||
b44a10b7 BP |
507 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
508 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
509 | ||
abe529af BP |
510 | static void ofproto_dpif_unixctl_init(void); |
511 | ||
46c88433 EJ |
512 | static inline struct ofproto_dpif * |
513 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
514 | { | |
515 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
516 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
517 | } | |
518 | ||
519 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, | |
520 | ofp_port_t ofp_port); | |
521 | ||
501f8d1f BP |
522 | /* Upcalls. */ |
523 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 524 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
525 | |
526 | /* Flow expiration. */ | |
acf60855 | 527 | static int expire(struct dpif_backer *); |
abe529af | 528 | |
6fca1ffb BP |
529 | /* NetFlow. */ |
530 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
531 | ||
abe529af | 532 | /* Utilities. */ |
52a90c29 | 533 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
9583bc14 | 534 | |
abe529af BP |
535 | /* Global variables. */ |
536 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
537 | |
538 | /* Initial mappings of port to bridge mappings. */ | |
539 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
46c88433 EJ |
540 | |
541 | int | |
542 | ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, | |
543 | struct ofputil_flow_mod *fm) | |
544 | { | |
545 | return ofproto_flow_mod(&ofproto->up, fm); | |
546 | } | |
547 | ||
548 | void | |
549 | ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, | |
550 | struct ofputil_packet_in *pin) | |
551 | { | |
552 | connmgr_send_packet_in(ofproto->up.connmgr, pin); | |
553 | } | |
abe529af BP |
554 | \f |
555 | /* Factory functions. */ | |
556 | ||
b0408fca | 557 | static void |
acf60855 | 558 | init(const struct shash *iface_hints) |
b0408fca | 559 | { |
acf60855 JP |
560 | struct shash_node *node; |
561 | ||
562 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
563 | SHASH_FOR_EACH(node, iface_hints) { | |
564 | const struct iface_hint *orig_hint = node->data; | |
565 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
566 | ||
567 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
568 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
569 | new_hint->ofp_port = orig_hint->ofp_port; | |
570 | ||
571 | shash_add(&init_ofp_ports, node->name, new_hint); | |
572 | } | |
b0408fca JP |
573 | } |
574 | ||
abe529af BP |
575 | static void |
576 | enumerate_types(struct sset *types) | |
577 | { | |
578 | dp_enumerate_types(types); | |
579 | } | |
580 | ||
581 | static int | |
582 | enumerate_names(const char *type, struct sset *names) | |
583 | { | |
acf60855 JP |
584 | struct ofproto_dpif *ofproto; |
585 | ||
586 | sset_clear(names); | |
587 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
588 | if (strcmp(type, ofproto->up.type)) { | |
589 | continue; | |
590 | } | |
591 | sset_add(names, ofproto->up.name); | |
592 | } | |
593 | ||
594 | return 0; | |
abe529af BP |
595 | } |
596 | ||
597 | static int | |
598 | del(const char *type, const char *name) | |
599 | { | |
600 | struct dpif *dpif; | |
601 | int error; | |
602 | ||
603 | error = dpif_open(name, type, &dpif); | |
604 | if (!error) { | |
605 | error = dpif_delete(dpif); | |
606 | dpif_close(dpif); | |
607 | } | |
608 | return error; | |
609 | } | |
610 | \f | |
0aeaabc8 JP |
611 | static const char * |
612 | port_open_type(const char *datapath_type, const char *port_type) | |
613 | { | |
614 | return dpif_port_open_type(datapath_type, port_type); | |
615 | } | |
616 | ||
acf60855 JP |
617 | /* Type functions. */ |
618 | ||
36beb9be BP |
619 | static void process_dpif_port_changes(struct dpif_backer *); |
620 | static void process_dpif_all_ports_changed(struct dpif_backer *); | |
621 | static void process_dpif_port_change(struct dpif_backer *, | |
622 | const char *devname); | |
623 | static void process_dpif_port_error(struct dpif_backer *, int error); | |
624 | ||
476cb42a BP |
625 | static struct ofproto_dpif * |
626 | lookup_ofproto_dpif_by_port_name(const char *name) | |
627 | { | |
628 | struct ofproto_dpif *ofproto; | |
629 | ||
630 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
631 | if (sset_contains(&ofproto->ports, name)) { | |
632 | return ofproto; | |
633 | } | |
634 | } | |
635 | ||
636 | return NULL; | |
637 | } | |
638 | ||
acf60855 JP |
639 | static int |
640 | type_run(const char *type) | |
641 | { | |
e20b5746 | 642 | static long long int push_timer = LLONG_MIN; |
acf60855 | 643 | struct dpif_backer *backer; |
acf60855 JP |
644 | |
645 | backer = shash_find_data(&all_dpif_backers, type); | |
646 | if (!backer) { | |
647 | /* This is not necessarily a problem, since backers are only | |
648 | * created on demand. */ | |
649 | return 0; | |
650 | } | |
651 | ||
652 | dpif_run(backer->dpif); | |
653 | ||
e20b5746 EJ |
654 | /* The most natural place to push facet statistics is when they're pulled |
655 | * from the datapath. However, when there are many flows in the datapath, | |
656 | * this expensive operation can occur so frequently, that it reduces our | |
657 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
658 | * here instead. */ | |
659 | if (time_msec() > push_timer) { | |
660 | push_timer = time_msec() + 2000; | |
661 | push_all_stats(); | |
662 | } | |
663 | ||
40358701 GS |
664 | /* If vswitchd started with other_config:flow_restore_wait set as "true", |
665 | * and the configuration has now changed to "false", enable receiving | |
666 | * packets from the datapath. */ | |
667 | if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { | |
36beb9be BP |
668 | int error; |
669 | ||
40358701 GS |
670 | backer->recv_set_enable = true; |
671 | ||
672 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); | |
673 | if (error) { | |
674 | VLOG_ERR("Failed to enable receiving packets in dpif."); | |
675 | return error; | |
676 | } | |
677 | dpif_flow_flush(backer->dpif); | |
678 | backer->need_revalidate = REV_RECONFIGURE; | |
679 | } | |
680 | ||
a1616063 | 681 | if (backer->need_revalidate) { |
2cc3c58e | 682 | struct ofproto_dpif *ofproto; |
a614d823 KM |
683 | struct simap_node *node; |
684 | struct simap tmp_backers; | |
685 | ||
686 | /* Handle tunnel garbage collection. */ | |
687 | simap_init(&tmp_backers); | |
688 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
689 | ||
690 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
691 | struct ofport_dpif *iter; | |
692 | ||
693 | if (backer != ofproto->backer) { | |
694 | continue; | |
695 | } | |
696 | ||
697 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
3aa30359 | 698 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
a614d823 KM |
699 | const char *dp_port; |
700 | ||
42943cde | 701 | if (!iter->is_tunnel) { |
a614d823 KM |
702 | continue; |
703 | } | |
704 | ||
3aa30359 BP |
705 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev, |
706 | namebuf, sizeof namebuf); | |
a614d823 KM |
707 | node = simap_find(&tmp_backers, dp_port); |
708 | if (node) { | |
709 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
710 | simap_delete(&tmp_backers, node); | |
711 | node = simap_find(&backer->tnl_backers, dp_port); | |
712 | } else { | |
713 | node = simap_find(&backer->tnl_backers, dp_port); | |
714 | if (!node) { | |
4e022ec0 | 715 | odp_port_t odp_port = ODPP_NONE; |
a614d823 KM |
716 | |
717 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
718 | &odp_port)) { | |
4e022ec0 AW |
719 | simap_put(&backer->tnl_backers, dp_port, |
720 | odp_to_u32(odp_port)); | |
a614d823 KM |
721 | node = simap_find(&backer->tnl_backers, dp_port); |
722 | } | |
723 | } | |
724 | } | |
725 | ||
4e022ec0 | 726 | iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE; |
42943cde EJ |
727 | if (tnl_port_reconfigure(iter, iter->up.netdev, |
728 | iter->odp_port)) { | |
a614d823 KM |
729 | backer->need_revalidate = REV_RECONFIGURE; |
730 | } | |
731 | } | |
732 | } | |
733 | ||
734 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
4e022ec0 | 735 | dpif_port_del(backer->dpif, u32_to_odp(node->data)); |
a614d823 KM |
736 | } |
737 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
738 | |
739 | switch (backer->need_revalidate) { | |
740 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
741 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
4a1b8f30 | 742 | case REV_BOND: COVERAGE_INC(rev_bond); break; |
2cc3c58e EJ |
743 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; |
744 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
30618594 | 745 | case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break; |
2cc3c58e EJ |
746 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; |
747 | } | |
f728af2e BP |
748 | backer->need_revalidate = 0; |
749 | ||
a1616063 EJ |
750 | /* Clear the drop_keys in case we should now be accepting some |
751 | * formerly dropped flows. */ | |
752 | drop_key_clear(backer); | |
753 | ||
2cc3c58e | 754 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 755 | struct facet *facet, *next; |
a1616063 | 756 | struct ofport_dpif *ofport; |
bcd2633a | 757 | struct cls_cursor cursor; |
a1616063 | 758 | struct ofbundle *bundle; |
2cc3c58e EJ |
759 | |
760 | if (ofproto->backer != backer) { | |
761 | continue; | |
762 | } | |
763 | ||
a1616063 | 764 | xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml, |
9d189a50 EJ |
765 | ofproto->stp, ofproto->mbridge, |
766 | ofproto->sflow, ofproto->ipfix, | |
767 | ofproto->up.frag_handling, | |
a1616063 EJ |
768 | ofproto->up.forward_bpdu, |
769 | connmgr_has_in_band(ofproto->up.connmgr), | |
9d189a50 | 770 | ofproto->netflow != NULL); |
46c88433 | 771 | |
a1616063 EJ |
772 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
773 | xlate_bundle_set(ofproto, bundle, bundle->name, | |
774 | bundle->vlan_mode, bundle->vlan, | |
775 | bundle->trunks, bundle->use_priority_tags, | |
776 | bundle->bond, bundle->lacp, | |
777 | bundle->floodable); | |
778 | } | |
779 | ||
780 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
9d189a50 EJ |
781 | int stp_port = ofport->stp_port |
782 | ? stp_port_no(ofport->stp_port) | |
783 | : 0; | |
a1616063 EJ |
784 | xlate_ofport_set(ofproto, ofport->bundle, ofport, |
785 | ofport->up.ofp_port, ofport->odp_port, | |
786 | ofport->up.netdev, ofport->cfm, | |
9d189a50 | 787 | ofport->bfd, ofport->peer, stp_port, |
55954f6e | 788 | ofport->qdscp, ofport->n_qdscp, |
9d189a50 EJ |
789 | ofport->up.pp.config, ofport->is_tunnel, |
790 | ofport->may_enable); | |
46c88433 EJ |
791 | } |
792 | ||
bcd2633a JP |
793 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
794 | CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { | |
a1616063 EJ |
795 | facet_revalidate(facet); |
796 | run_fast_rl(); | |
2cc3c58e EJ |
797 | } |
798 | } | |
2cc3c58e EJ |
799 | } |
800 | ||
40358701 GS |
801 | if (!backer->recv_set_enable) { |
802 | /* Wake up before a max of 1000ms. */ | |
803 | timer_set_duration(&backer->next_expiration, 1000); | |
804 | } else if (timer_expired(&backer->next_expiration)) { | |
acf60855 JP |
805 | int delay = expire(backer); |
806 | timer_set_duration(&backer->next_expiration, delay); | |
807 | } | |
808 | ||
36beb9be | 809 | process_dpif_port_changes(backer); |
acf60855 | 810 | |
04d08d54 EJ |
811 | if (backer->governor) { |
812 | size_t n_subfacets; | |
813 | ||
814 | governor_run(backer->governor); | |
815 | ||
816 | /* If the governor has shrunk to its minimum size and the number of | |
817 | * subfacets has dwindled, then drop the governor entirely. | |
818 | * | |
819 | * For hysteresis, the number of subfacets to drop the governor is | |
820 | * smaller than the number needed to trigger its creation. */ | |
821 | n_subfacets = hmap_count(&backer->subfacets); | |
822 | if (n_subfacets * 4 < flow_eviction_threshold | |
823 | && governor_is_idle(backer->governor)) { | |
824 | governor_destroy(backer->governor); | |
825 | backer->governor = NULL; | |
826 | } | |
827 | } | |
828 | ||
acf60855 JP |
829 | return 0; |
830 | } | |
831 | ||
36beb9be BP |
832 | /* Check for and handle port changes in 'backer''s dpif. */ |
833 | static void | |
834 | process_dpif_port_changes(struct dpif_backer *backer) | |
835 | { | |
836 | for (;;) { | |
837 | char *devname; | |
838 | int error; | |
839 | ||
840 | error = dpif_port_poll(backer->dpif, &devname); | |
841 | switch (error) { | |
842 | case EAGAIN: | |
843 | return; | |
844 | ||
845 | case ENOBUFS: | |
846 | process_dpif_all_ports_changed(backer); | |
847 | break; | |
848 | ||
849 | case 0: | |
850 | process_dpif_port_change(backer, devname); | |
851 | free(devname); | |
852 | break; | |
853 | ||
854 | default: | |
855 | process_dpif_port_error(backer, error); | |
856 | break; | |
857 | } | |
858 | } | |
859 | } | |
860 | ||
861 | static void | |
862 | process_dpif_all_ports_changed(struct dpif_backer *backer) | |
863 | { | |
864 | struct ofproto_dpif *ofproto; | |
865 | struct dpif_port dpif_port; | |
866 | struct dpif_port_dump dump; | |
867 | struct sset devnames; | |
868 | const char *devname; | |
869 | ||
870 | sset_init(&devnames); | |
871 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
872 | if (ofproto->backer == backer) { | |
873 | struct ofport *ofport; | |
874 | ||
875 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
876 | sset_add(&devnames, netdev_get_name(ofport->netdev)); | |
877 | } | |
878 | } | |
879 | } | |
880 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) { | |
881 | sset_add(&devnames, dpif_port.name); | |
882 | } | |
883 | ||
884 | SSET_FOR_EACH (devname, &devnames) { | |
885 | process_dpif_port_change(backer, devname); | |
886 | } | |
887 | sset_destroy(&devnames); | |
888 | } | |
889 | ||
890 | static void | |
891 | process_dpif_port_change(struct dpif_backer *backer, const char *devname) | |
892 | { | |
893 | struct ofproto_dpif *ofproto; | |
894 | struct dpif_port port; | |
895 | ||
896 | /* Don't report on the datapath's device. */ | |
897 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
898 | return; | |
899 | } | |
900 | ||
901 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
902 | &all_ofproto_dpifs) { | |
903 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { | |
904 | return; | |
905 | } | |
906 | } | |
907 | ||
908 | ofproto = lookup_ofproto_dpif_by_port_name(devname); | |
909 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { | |
910 | /* The port was removed. If we know the datapath, | |
911 | * report it through poll_set(). If we don't, it may be | |
912 | * notifying us of a removal we initiated, so ignore it. | |
913 | * If there's a pending ENOBUFS, let it stand, since | |
914 | * everything will be reevaluated. */ | |
915 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
916 | sset_add(&ofproto->port_poll_set, devname); | |
917 | ofproto->port_poll_errno = 0; | |
918 | } | |
919 | } else if (!ofproto) { | |
920 | /* The port was added, but we don't know with which | |
921 | * ofproto we should associate it. Delete it. */ | |
922 | dpif_port_del(backer->dpif, port.port_no); | |
74cc3969 BP |
923 | } else { |
924 | struct ofport_dpif *ofport; | |
925 | ||
926 | ofport = ofport_dpif_cast(shash_find_data( | |
927 | &ofproto->up.port_by_name, devname)); | |
928 | if (ofport | |
929 | && ofport->odp_port != port.port_no | |
930 | && !odp_port_to_ofport(backer, port.port_no)) | |
931 | { | |
932 | /* 'ofport''s datapath port number has changed from | |
933 | * 'ofport->odp_port' to 'port.port_no'. Update our internal data | |
934 | * structures to match. */ | |
935 | hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node); | |
936 | ofport->odp_port = port.port_no; | |
937 | hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node, | |
938 | hash_odp_port(port.port_no)); | |
939 | backer->need_revalidate = REV_RECONFIGURE; | |
940 | } | |
36beb9be BP |
941 | } |
942 | dpif_port_destroy(&port); | |
943 | } | |
944 | ||
945 | /* Propagate 'error' to all ofprotos based on 'backer'. */ | |
946 | static void | |
947 | process_dpif_port_error(struct dpif_backer *backer, int error) | |
948 | { | |
949 | struct ofproto_dpif *ofproto; | |
950 | ||
951 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
952 | if (ofproto->backer == backer) { | |
953 | sset_clear(&ofproto->port_poll_set); | |
954 | ofproto->port_poll_errno = error; | |
955 | } | |
956 | } | |
957 | } | |
958 | ||
acf60855 | 959 | static int |
8fa4d1d0 | 960 | dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) |
acf60855 | 961 | { |
acf60855 JP |
962 | unsigned int work; |
963 | ||
40358701 GS |
964 | /* If recv_set_enable is false, we should not handle upcalls. */ |
965 | if (!backer->recv_set_enable) { | |
966 | return 0; | |
967 | } | |
968 | ||
acf60855 JP |
969 | /* Handle one or more batches of upcalls, until there's nothing left to do |
970 | * or until we do a fixed total amount of work. | |
971 | * | |
972 | * We do work in batches because it can be much cheaper to set up a number | |
973 | * of flows and fire off their patches all at once. We do multiple batches | |
974 | * because in some cases handling a packet can cause another packet to be | |
975 | * queued almost immediately as part of the return flow. Both | |
976 | * optimizations can make major improvements on some benchmarks and | |
977 | * presumably for real traffic as well. */ | |
978 | work = 0; | |
8fa4d1d0 EJ |
979 | while (work < max_batch) { |
980 | int retval = handle_upcalls(backer, max_batch - work); | |
acf60855 JP |
981 | if (retval <= 0) { |
982 | return -retval; | |
983 | } | |
984 | work += retval; | |
985 | } | |
986 | ||
987 | return 0; | |
988 | } | |
989 | ||
8fa4d1d0 EJ |
990 | static int |
991 | type_run_fast(const char *type) | |
992 | { | |
993 | struct dpif_backer *backer; | |
994 | ||
995 | backer = shash_find_data(&all_dpif_backers, type); | |
996 | if (!backer) { | |
997 | /* This is not necessarily a problem, since backers are only | |
998 | * created on demand. */ | |
999 | return 0; | |
1000 | } | |
1001 | ||
1002 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1003 | } | |
1004 | ||
1005 | static void | |
1006 | run_fast_rl(void) | |
1007 | { | |
1008 | static long long int port_rl = LLONG_MIN; | |
1009 | static unsigned int backer_rl = 0; | |
1010 | ||
1011 | if (time_msec() >= port_rl) { | |
1012 | struct ofproto_dpif *ofproto; | |
1013 | struct ofport_dpif *ofport; | |
1014 | ||
1015 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1016 | ||
1017 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1018 | port_run_fast(ofport); | |
1019 | } | |
1020 | } | |
1021 | port_rl = time_msec() + 200; | |
1022 | } | |
1023 | ||
1024 | /* XXX: We have to be careful not to do too much work in this function. If | |
1025 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1026 | * performance improves signifcantly, but at a cost. It's possible for the | |
1027 | * number of flows in the datapath to increase without bound, and for poll | |
1028 | * loops to take 10s of seconds. The correct solution to this problem, | |
1029 | * long term, is to separate flow miss handling into it's own thread so it | |
1030 | * isn't affected by revalidations, and expirations. Until then, this is | |
1031 | * the best we can do. */ | |
1032 | if (++backer_rl >= 10) { | |
1033 | struct shash_node *node; | |
1034 | ||
1035 | backer_rl = 0; | |
1036 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1037 | dpif_backer_run_fast(node->data, 1); | |
1038 | } | |
1039 | } | |
1040 | } | |
1041 | ||
acf60855 JP |
1042 | static void |
1043 | type_wait(const char *type) | |
1044 | { | |
1045 | struct dpif_backer *backer; | |
1046 | ||
1047 | backer = shash_find_data(&all_dpif_backers, type); | |
1048 | if (!backer) { | |
1049 | /* This is not necessarily a problem, since backers are only | |
1050 | * created on demand. */ | |
1051 | return; | |
1052 | } | |
1053 | ||
04d08d54 EJ |
1054 | if (backer->governor) { |
1055 | governor_wait(backer->governor); | |
1056 | } | |
1057 | ||
acf60855 JP |
1058 | timer_wait(&backer->next_expiration); |
1059 | } | |
1060 | \f | |
abe529af BP |
1061 | /* Basic life-cycle. */ |
1062 | ||
c57b2226 BP |
1063 | static int add_internal_flows(struct ofproto_dpif *); |
1064 | ||
abe529af BP |
1065 | static struct ofproto * |
1066 | alloc(void) | |
1067 | { | |
1068 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1069 | return &ofproto->up; | |
1070 | } | |
1071 | ||
1072 | static void | |
1073 | dealloc(struct ofproto *ofproto_) | |
1074 | { | |
1075 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1076 | free(ofproto); | |
1077 | } | |
1078 | ||
acf60855 JP |
1079 | static void |
1080 | close_dpif_backer(struct dpif_backer *backer) | |
1081 | { | |
1082 | struct shash_node *node; | |
1083 | ||
cb22974d | 1084 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1085 | |
1086 | if (--backer->refcount) { | |
1087 | return; | |
1088 | } | |
1089 | ||
8f73d537 EJ |
1090 | drop_key_clear(backer); |
1091 | hmap_destroy(&backer->drop_keys); | |
1092 | ||
7d82ab2e | 1093 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1094 | hmap_destroy(&backer->odp_to_ofport_map); |
1095 | node = shash_find(&all_dpif_backers, backer->type); | |
1096 | free(backer->type); | |
1097 | shash_delete(&all_dpif_backers, node); | |
1098 | dpif_close(backer->dpif); | |
1099 | ||
04d08d54 EJ |
1100 | ovs_assert(hmap_is_empty(&backer->subfacets)); |
1101 | hmap_destroy(&backer->subfacets); | |
1102 | governor_destroy(backer->governor); | |
1103 | ||
acf60855 JP |
1104 | free(backer); |
1105 | } | |
1106 | ||
1107 | /* Datapath port slated for removal from datapath. */ | |
1108 | struct odp_garbage { | |
1109 | struct list list_node; | |
4e022ec0 | 1110 | odp_port_t odp_port; |
acf60855 JP |
1111 | }; |
1112 | ||
1113 | static int | |
1114 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1115 | { | |
1116 | struct dpif_backer *backer; | |
1117 | struct dpif_port_dump port_dump; | |
1118 | struct dpif_port port; | |
1119 | struct shash_node *node; | |
1120 | struct list garbage_list; | |
1121 | struct odp_garbage *garbage, *next; | |
1122 | struct sset names; | |
1123 | char *backer_name; | |
1124 | const char *name; | |
1125 | int error; | |
1126 | ||
1127 | backer = shash_find_data(&all_dpif_backers, type); | |
1128 | if (backer) { | |
1129 | backer->refcount++; | |
1130 | *backerp = backer; | |
1131 | return 0; | |
1132 | } | |
1133 | ||
1134 | backer_name = xasprintf("ovs-%s", type); | |
1135 | ||
1136 | /* Remove any existing datapaths, since we assume we're the only | |
1137 | * userspace controlling the datapath. */ | |
1138 | sset_init(&names); | |
1139 | dp_enumerate_names(type, &names); | |
1140 | SSET_FOR_EACH(name, &names) { | |
1141 | struct dpif *old_dpif; | |
1142 | ||
1143 | /* Don't remove our backer if it exists. */ | |
1144 | if (!strcmp(name, backer_name)) { | |
1145 | continue; | |
1146 | } | |
1147 | ||
1148 | if (dpif_open(name, type, &old_dpif)) { | |
1149 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1150 | } else { | |
1151 | dpif_delete(old_dpif); | |
1152 | dpif_close(old_dpif); | |
1153 | } | |
1154 | } | |
1155 | sset_destroy(&names); | |
1156 | ||
1157 | backer = xmalloc(sizeof *backer); | |
1158 | ||
1159 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1160 | free(backer_name); | |
1161 | if (error) { | |
1162 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
10a89ef0 | 1163 | ovs_strerror(error)); |
4c1b1289 | 1164 | free(backer); |
acf60855 JP |
1165 | return error; |
1166 | } | |
1167 | ||
1168 | backer->type = xstrdup(type); | |
04d08d54 | 1169 | backer->governor = NULL; |
acf60855 JP |
1170 | backer->refcount = 1; |
1171 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1172 | hmap_init(&backer->drop_keys); |
04d08d54 | 1173 | hmap_init(&backer->subfacets); |
acf60855 | 1174 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1175 | backer->need_revalidate = 0; |
7d82ab2e | 1176 | simap_init(&backer->tnl_backers); |
40358701 | 1177 | backer->recv_set_enable = !ofproto_get_flow_restore_wait(); |
acf60855 JP |
1178 | *backerp = backer; |
1179 | ||
40358701 GS |
1180 | if (backer->recv_set_enable) { |
1181 | dpif_flow_flush(backer->dpif); | |
1182 | } | |
acf60855 JP |
1183 | |
1184 | /* Loop through the ports already on the datapath and remove any | |
1185 | * that we don't need anymore. */ | |
1186 | list_init(&garbage_list); | |
1187 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1188 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1189 | node = shash_find(&init_ofp_ports, port.name); | |
1190 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1191 | garbage = xmalloc(sizeof *garbage); | |
1192 | garbage->odp_port = port.port_no; | |
1193 | list_push_front(&garbage_list, &garbage->list_node); | |
1194 | } | |
1195 | } | |
1196 | dpif_port_dump_done(&port_dump); | |
1197 | ||
1198 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1199 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1200 | list_remove(&garbage->list_node); | |
1201 | free(garbage); | |
1202 | } | |
1203 | ||
1204 | shash_add(&all_dpif_backers, type, backer); | |
1205 | ||
40358701 | 1206 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); |
acf60855 JP |
1207 | if (error) { |
1208 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
10a89ef0 | 1209 | type, ovs_strerror(error)); |
acf60855 JP |
1210 | close_dpif_backer(backer); |
1211 | return error; | |
1212 | } | |
1213 | ||
dc54ef36 EJ |
1214 | backer->max_n_subfacet = 0; |
1215 | backer->created = time_msec(); | |
1216 | backer->last_minute = backer->created; | |
1217 | memset(&backer->hourly, 0, sizeof backer->hourly); | |
1218 | memset(&backer->daily, 0, sizeof backer->daily); | |
1219 | memset(&backer->lifetime, 0, sizeof backer->lifetime); | |
1220 | backer->subfacet_add_count = 0; | |
1221 | backer->subfacet_del_count = 0; | |
1222 | backer->total_subfacet_add_count = 0; | |
1223 | backer->total_subfacet_del_count = 0; | |
1224 | backer->avg_n_subfacet = 0; | |
1225 | backer->avg_subfacet_life = 0; | |
1226 | ||
acf60855 JP |
1227 | return error; |
1228 | } | |
1229 | ||
abe529af | 1230 | static int |
0f5f95a9 | 1231 | construct(struct ofproto *ofproto_) |
abe529af BP |
1232 | { |
1233 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1234 | struct shash_node *node, *next; |
4e022ec0 | 1235 | odp_port_t max_ports; |
abe529af | 1236 | int error; |
abe529af | 1237 | |
acf60855 | 1238 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1239 | if (error) { |
abe529af BP |
1240 | return error; |
1241 | } | |
1242 | ||
acf60855 | 1243 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
4e022ec0 AW |
1244 | ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports), |
1245 | ofp_to_u16(OFPP_MAX)))); | |
91858960 | 1246 | |
abe529af BP |
1247 | ofproto->netflow = NULL; |
1248 | ofproto->sflow = NULL; | |
29089a54 | 1249 | ofproto->ipfix = NULL; |
21f7563c | 1250 | ofproto->stp = NULL; |
abe529af | 1251 | hmap_init(&ofproto->bundles); |
e764773c | 1252 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
ec7ceaed | 1253 | ofproto->mbridge = mbridge_create(); |
abe529af BP |
1254 | ofproto->has_bonded_bundles = false; |
1255 | ||
bcd2633a | 1256 | classifier_init(&ofproto->facets); |
4d2a0f39 | 1257 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 | 1258 | |
7ee20df1 BP |
1259 | list_init(&ofproto->completions); |
1260 | ||
abe529af BP |
1261 | ofproto_dpif_unixctl_init(); |
1262 | ||
52a90c29 BP |
1263 | hmap_init(&ofproto->vlandev_map); |
1264 | hmap_init(&ofproto->realdev_vid_map); | |
1265 | ||
acf60855 | 1266 | sset_init(&ofproto->ports); |
0a740f48 | 1267 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1268 | sset_init(&ofproto->port_poll_set); |
1269 | ofproto->port_poll_errno = 0; | |
1270 | ||
1271 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1272 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1273 | |
1274 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1275 | /* Check if the datapath already has this port. */ | |
1276 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1277 | sset_add(&ofproto->ports, node->name); | |
1278 | } | |
1279 | ||
1280 | free(iface_hint->br_name); | |
1281 | free(iface_hint->br_type); | |
4f9e08a5 | 1282 | free(iface_hint); |
acf60855 JP |
1283 | shash_delete(&init_ofp_ports, node); |
1284 | } | |
1285 | } | |
e1b1d06a | 1286 | |
b44a10b7 BP |
1287 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1288 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1289 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1290 | |
1291 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1292 | error = add_internal_flows(ofproto); |
1293 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1294 | ||
735d7efb AZ |
1295 | ofproto->n_hit = 0; |
1296 | ofproto->n_missed = 0; | |
1297 | ||
c57b2226 BP |
1298 | return error; |
1299 | } | |
1300 | ||
1301 | static int | |
1302 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1303 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1304 | { |
1305 | struct ofputil_flow_mod fm; | |
1306 | int error; | |
1307 | ||
81a76618 BP |
1308 | match_init_catchall(&fm.match); |
1309 | fm.priority = 0; | |
1310 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1311 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1312 | fm.cookie = htonll(0); |
1313 | fm.cookie_mask = htonll(0); | |
23342857 | 1314 | fm.modify_cookie = false; |
c57b2226 BP |
1315 | fm.table_id = TBL_INTERNAL; |
1316 | fm.command = OFPFC_ADD; | |
1317 | fm.idle_timeout = 0; | |
1318 | fm.hard_timeout = 0; | |
1319 | fm.buffer_id = 0; | |
1320 | fm.out_port = 0; | |
1321 | fm.flags = 0; | |
f25d0cf3 BP |
1322 | fm.ofpacts = ofpacts->data; |
1323 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1324 | |
1325 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1326 | if (error) { | |
1327 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1328 | id, ofperr_to_string(error)); | |
1329 | return error; | |
1330 | } | |
1331 | ||
9583bc14 EJ |
1332 | *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, |
1333 | TBL_INTERNAL); | |
cb22974d | 1334 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1335 | |
abe529af BP |
1336 | return 0; |
1337 | } | |
1338 | ||
c57b2226 BP |
1339 | static int |
1340 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1341 | { | |
f25d0cf3 BP |
1342 | struct ofpact_controller *controller; |
1343 | uint64_t ofpacts_stub[128 / 8]; | |
1344 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1345 | int error; |
1346 | int id; | |
1347 | ||
f25d0cf3 | 1348 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1349 | id = 1; |
1350 | ||
f25d0cf3 BP |
1351 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1352 | controller->max_len = UINT16_MAX; | |
1353 | controller->controller_id = 0; | |
1354 | controller->reason = OFPR_NO_MATCH; | |
1355 | ofpact_pad(&ofpacts); | |
1356 | ||
1357 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1358 | if (error) { |
1359 | return error; | |
1360 | } | |
1361 | ||
f25d0cf3 BP |
1362 | ofpbuf_clear(&ofpacts); |
1363 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 | 1364 | &ofproto->no_packet_in_rule); |
7fd51d39 BP |
1365 | if (error) { |
1366 | return error; | |
1367 | } | |
1368 | ||
1369 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1370 | &ofproto->drop_frags_rule); | |
c57b2226 BP |
1371 | return error; |
1372 | } | |
1373 | ||
7ee20df1 BP |
1374 | static void |
1375 | complete_operations(struct ofproto_dpif *ofproto) | |
1376 | { | |
1377 | struct dpif_completion *c, *next; | |
1378 | ||
1379 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1380 | ofoperation_complete(c->op, 0); | |
1381 | list_remove(&c->list_node); | |
1382 | free(c); | |
1383 | } | |
1384 | } | |
1385 | ||
abe529af BP |
1386 | static void |
1387 | destruct(struct ofproto *ofproto_) | |
1388 | { | |
1389 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1390 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1391 | struct oftable *table; |
abe529af | 1392 | |
875b94ed | 1393 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
46c88433 EJ |
1394 | xlate_remove_ofproto(ofproto); |
1395 | ||
b44a10b7 | 1396 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1397 | complete_operations(ofproto); |
1398 | ||
0697b5c3 BP |
1399 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1400 | struct cls_cursor cursor; | |
1401 | ||
d0918789 | 1402 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1403 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1404 | ofproto_rule_destroy(&rule->up); | |
1405 | } | |
7ee20df1 BP |
1406 | } |
1407 | ||
ec7ceaed | 1408 | mbridge_unref(ofproto->mbridge); |
abe529af BP |
1409 | |
1410 | netflow_destroy(ofproto->netflow); | |
9723bcce | 1411 | dpif_sflow_unref(ofproto->sflow); |
abe529af | 1412 | hmap_destroy(&ofproto->bundles); |
5d989517 | 1413 | mac_learning_unref(ofproto->ml); |
abe529af | 1414 | |
bcd2633a | 1415 | classifier_destroy(&ofproto->facets); |
abe529af | 1416 | |
52a90c29 BP |
1417 | hmap_destroy(&ofproto->vlandev_map); |
1418 | hmap_destroy(&ofproto->realdev_vid_map); | |
1419 | ||
acf60855 | 1420 | sset_destroy(&ofproto->ports); |
0a740f48 | 1421 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1422 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 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_); | |
0aa66d6e | 1431 | struct ofport_dpif *ofport; |
abe529af | 1432 | |
40358701 GS |
1433 | /* Do not perform any periodic activity required by 'ofproto' while |
1434 | * waiting for flow restore to complete. */ | |
1435 | if (ofproto_get_flow_restore_wait()) { | |
1436 | return 0; | |
1437 | } | |
1438 | ||
0aa66d6e EJ |
1439 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1440 | port_run_fast(ofport); | |
1441 | } | |
1442 | ||
5fcc0d00 BP |
1443 | return 0; |
1444 | } | |
1445 | ||
1446 | static int | |
1447 | run(struct ofproto *ofproto_) | |
1448 | { | |
1449 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1450 | struct ofport_dpif *ofport; | |
1451 | struct ofbundle *bundle; | |
1452 | int error; | |
1453 | ||
1454 | if (!clogged) { | |
1455 | complete_operations(ofproto); | |
1456 | } | |
5fcc0d00 | 1457 | |
ec7ceaed EJ |
1458 | if (mbridge_need_revalidate(ofproto->mbridge)) { |
1459 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
509c0149 | 1460 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 1461 | mac_learning_flush(ofproto->ml); |
509c0149 | 1462 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
ec7ceaed EJ |
1463 | } |
1464 | ||
40358701 GS |
1465 | /* Do not perform any periodic activity below required by 'ofproto' while |
1466 | * waiting for flow restore to complete. */ | |
1467 | if (ofproto_get_flow_restore_wait()) { | |
1468 | return 0; | |
1469 | } | |
1470 | ||
5fcc0d00 BP |
1471 | error = run_fast(ofproto_); |
1472 | if (error) { | |
1473 | return error; | |
abe529af BP |
1474 | } |
1475 | ||
abe529af | 1476 | if (ofproto->netflow) { |
6fca1ffb BP |
1477 | if (netflow_run(ofproto->netflow)) { |
1478 | send_netflow_active_timeouts(ofproto); | |
1479 | } | |
abe529af BP |
1480 | } |
1481 | if (ofproto->sflow) { | |
bae473fe | 1482 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1483 | } |
1484 | ||
1485 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1486 | port_run(ofport); | |
1487 | } | |
1488 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1489 | bundle_run(bundle); | |
1490 | } | |
1491 | ||
21f7563c | 1492 | stp_run(ofproto); |
509c0149 | 1493 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 EJ |
1494 | if (mac_learning_run(ofproto->ml)) { |
1495 | ofproto->backer->need_revalidate = REV_MAC_LEARNING; | |
1496 | } | |
509c0149 | 1497 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 1498 | |
6814e51f | 1499 | /* Check the consistency of a random facet, to aid debugging. */ |
4d2a0f39 | 1500 | if (time_msec() >= ofproto->consistency_rl |
bcd2633a | 1501 | && !classifier_is_empty(&ofproto->facets) |
2cc3c58e | 1502 | && !ofproto->backer->need_revalidate) { |
bcd2633a JP |
1503 | struct cls_table *table; |
1504 | struct cls_rule *cr; | |
6814e51f BP |
1505 | struct facet *facet; |
1506 | ||
4d2a0f39 EJ |
1507 | ofproto->consistency_rl = time_msec() + 250; |
1508 | ||
bcd2633a JP |
1509 | table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables), |
1510 | struct cls_table, hmap_node); | |
1511 | cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, | |
1512 | hmap_node); | |
1513 | facet = CONTAINER_OF(cr, struct facet, cr); | |
1514 | ||
a1616063 EJ |
1515 | if (!facet_check_consistency(facet)) { |
1516 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
6814e51f BP |
1517 | } |
1518 | } | |
1519 | ||
abe529af BP |
1520 | return 0; |
1521 | } | |
1522 | ||
1523 | static void | |
1524 | wait(struct ofproto *ofproto_) | |
1525 | { | |
1526 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1527 | struct ofport_dpif *ofport; | |
1528 | struct ofbundle *bundle; | |
1529 | ||
7ee20df1 BP |
1530 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1531 | poll_immediate_wake(); | |
1532 | } | |
1533 | ||
40358701 GS |
1534 | if (ofproto_get_flow_restore_wait()) { |
1535 | return; | |
1536 | } | |
1537 | ||
acf60855 JP |
1538 | dpif_wait(ofproto->backer->dpif); |
1539 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1540 | if (ofproto->sflow) { |
bae473fe | 1541 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1542 | } |
abe529af BP |
1543 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1544 | port_wait(ofport); | |
1545 | } | |
1546 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1547 | bundle_wait(bundle); | |
1548 | } | |
6fca1ffb BP |
1549 | if (ofproto->netflow) { |
1550 | netflow_wait(ofproto->netflow); | |
1551 | } | |
509c0149 | 1552 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
1c313b88 | 1553 | mac_learning_wait(ofproto->ml); |
509c0149 | 1554 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c | 1555 | stp_wait(ofproto); |
2cc3c58e | 1556 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1557 | /* Shouldn't happen, but if it does just go around again. */ |
1558 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1559 | poll_immediate_wake(); | |
abe529af BP |
1560 | } |
1561 | } | |
1562 | ||
0d085684 BP |
1563 | static void |
1564 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1565 | { | |
1566 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bcd2633a | 1567 | struct cls_cursor cursor; |
04d08d54 EJ |
1568 | size_t n_subfacets = 0; |
1569 | struct facet *facet; | |
0d085684 | 1570 | |
bcd2633a JP |
1571 | simap_increase(usage, "facets", classifier_count(&ofproto->facets)); |
1572 | ||
1573 | cls_cursor_init(&cursor, &ofproto->facets, NULL); | |
1574 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
04d08d54 EJ |
1575 | n_subfacets += list_size(&facet->subfacets); |
1576 | } | |
1577 | simap_increase(usage, "subfacets", n_subfacets); | |
0d085684 BP |
1578 | } |
1579 | ||
abe529af BP |
1580 | static void |
1581 | flush(struct ofproto *ofproto_) | |
1582 | { | |
1583 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1584 | struct subfacet *subfacet, *next_subfacet; |
1585 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1586 | int n_batch; | |
b0f7b9b5 | 1587 | |
acf60855 JP |
1588 | n_batch = 0; |
1589 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
04d08d54 | 1590 | &ofproto->backer->subfacets) { |
b7e2f6b3 | 1591 | if (subfacet->facet->ofproto != ofproto) { |
04d08d54 EJ |
1592 | continue; |
1593 | } | |
1594 | ||
acf60855 JP |
1595 | if (subfacet->path != SF_NOT_INSTALLED) { |
1596 | batch[n_batch++] = subfacet; | |
1597 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
04d08d54 | 1598 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 JP |
1599 | n_batch = 0; |
1600 | } | |
1601 | } else { | |
1602 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1603 | } |
abe529af | 1604 | } |
acf60855 JP |
1605 | |
1606 | if (n_batch > 0) { | |
04d08d54 | 1607 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 | 1608 | } |
abe529af BP |
1609 | } |
1610 | ||
6c1491fb BP |
1611 | static void |
1612 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1613 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1614 | { |
1615 | *arp_match_ip = true; | |
9e1fd49b BP |
1616 | *actions = (OFPUTIL_A_OUTPUT | |
1617 | OFPUTIL_A_SET_VLAN_VID | | |
1618 | OFPUTIL_A_SET_VLAN_PCP | | |
1619 | OFPUTIL_A_STRIP_VLAN | | |
1620 | OFPUTIL_A_SET_DL_SRC | | |
1621 | OFPUTIL_A_SET_DL_DST | | |
1622 | OFPUTIL_A_SET_NW_SRC | | |
1623 | OFPUTIL_A_SET_NW_DST | | |
1624 | OFPUTIL_A_SET_NW_TOS | | |
1625 | OFPUTIL_A_SET_TP_SRC | | |
1626 | OFPUTIL_A_SET_TP_DST | | |
1627 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1628 | } |
1629 | ||
1630 | static void | |
307975da | 1631 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1632 | { |
1633 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1634 | struct dpif_dp_stats s; |
7fd51d39 | 1635 | uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; |
7e741ffb | 1636 | uint64_t n_lookup; |
6c1491fb BP |
1637 | |
1638 | strcpy(ots->name, "classifier"); | |
1639 | ||
acf60855 | 1640 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1641 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1642 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
7fd51d39 | 1643 | rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); |
acf60855 | 1644 | |
7fd51d39 | 1645 | n_lookup = s.n_hit + s.n_missed - n_dropped_frags; |
7e741ffb JG |
1646 | ots->lookup_count = htonll(n_lookup); |
1647 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1648 | } |
1649 | ||
abe529af BP |
1650 | static struct ofport * |
1651 | port_alloc(void) | |
1652 | { | |
1653 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1654 | return &port->up; | |
1655 | } | |
1656 | ||
1657 | static void | |
1658 | port_dealloc(struct ofport *port_) | |
1659 | { | |
1660 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1661 | free(port); | |
1662 | } | |
1663 | ||
1664 | static int | |
1665 | port_construct(struct ofport *port_) | |
1666 | { | |
1667 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1668 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1669 | const struct netdev *netdev = port->up.netdev; |
3aa30359 | 1670 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
e1b1d06a JP |
1671 | struct dpif_port dpif_port; |
1672 | int error; | |
abe529af | 1673 | |
2cc3c58e | 1674 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1675 | port->bundle = NULL; |
1676 | port->cfm = NULL; | |
ccc09689 | 1677 | port->bfd = NULL; |
d5ffa7f2 | 1678 | port->may_enable = true; |
21f7563c JP |
1679 | port->stp_port = NULL; |
1680 | port->stp_state = STP_DISABLED; | |
42943cde | 1681 | port->is_tunnel = false; |
6cbbf4fa | 1682 | port->peer = NULL; |
55954f6e EJ |
1683 | port->qdscp = NULL; |
1684 | port->n_qdscp = 0; | |
52a90c29 BP |
1685 | port->realdev_ofp_port = 0; |
1686 | port->vlandev_vid = 0; | |
b9ad7294 | 1687 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1688 | |
b9ad7294 | 1689 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1690 | /* By bailing out here, we don't submit the port to the sFlow module |
1691 | * to be considered for counter polling export. This is correct | |
1692 | * because the patch port represents an interface that sFlow considers | |
1693 | * to be "internal" to the switch as a whole, and therefore not an | |
1694 | * candidate for counter polling. */ | |
4e022ec0 | 1695 | port->odp_port = ODPP_NONE; |
6cbbf4fa | 1696 | ofport_update_peer(port); |
0a740f48 EJ |
1697 | return 0; |
1698 | } | |
1699 | ||
acf60855 | 1700 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
3aa30359 BP |
1701 | netdev_vport_get_dpif_port(netdev, namebuf, |
1702 | sizeof namebuf), | |
e1b1d06a JP |
1703 | &dpif_port); |
1704 | if (error) { | |
1705 | return error; | |
1706 | } | |
1707 | ||
1708 | port->odp_port = dpif_port.port_no; | |
1709 | ||
b9ad7294 | 1710 | if (netdev_get_tunnel_config(netdev)) { |
42943cde EJ |
1711 | tnl_port_add(port, port->up.netdev, port->odp_port); |
1712 | port->is_tunnel = true; | |
b9ad7294 EJ |
1713 | } else { |
1714 | /* Sanity-check that a mapping doesn't already exist. This | |
1715 | * shouldn't happen for non-tunnel ports. */ | |
1716 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1717 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1718 | dpif_port.name); | |
da78d43d | 1719 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1720 | return EBUSY; |
1721 | } | |
e1b1d06a | 1722 | |
b9ad7294 | 1723 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
f9c0c3ec | 1724 | hash_odp_port(port->odp_port)); |
b9ad7294 | 1725 | } |
da78d43d | 1726 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1727 | |
abe529af | 1728 | if (ofproto->sflow) { |
e1b1d06a | 1729 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1730 | } |
1731 | ||
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | static void | |
1736 | port_destruct(struct ofport *port_) | |
1737 | { | |
1738 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1739 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
02f8d646 | 1740 | const char *devname = netdev_get_name(port->up.netdev); |
3aa30359 BP |
1741 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1742 | const char *dp_port_name; | |
abe529af | 1743 | |
7894e7da | 1744 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
46c88433 | 1745 | xlate_ofport_remove(port); |
7894e7da | 1746 | |
3aa30359 BP |
1747 | dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, |
1748 | sizeof namebuf); | |
a614d823 | 1749 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1750 | /* The underlying device is still there, so delete it. This |
1751 | * happens when the ofproto is being destroyed, since the caller | |
1752 | * assumes that removal of attached ports will happen as part of | |
1753 | * destruction. */ | |
42943cde | 1754 | if (!port->is_tunnel) { |
a614d823 KM |
1755 | dpif_port_del(ofproto->backer->dpif, port->odp_port); |
1756 | } | |
acf60855 JP |
1757 | } |
1758 | ||
6cbbf4fa EJ |
1759 | if (port->peer) { |
1760 | port->peer->peer = NULL; | |
1761 | port->peer = NULL; | |
1762 | } | |
1763 | ||
42943cde | 1764 | if (port->odp_port != ODPP_NONE && !port->is_tunnel) { |
0a740f48 EJ |
1765 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1766 | } | |
1767 | ||
42943cde | 1768 | tnl_port_del(port); |
02f8d646 | 1769 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1770 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
abe529af | 1771 | bundle_remove(port_); |
a5610457 | 1772 | set_cfm(port_, NULL); |
8aee94b6 | 1773 | set_bfd(port_, NULL); |
abe529af | 1774 | if (ofproto->sflow) { |
bae473fe | 1775 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1776 | } |
8b36f51e | 1777 | |
55954f6e | 1778 | free(port->qdscp); |
abe529af BP |
1779 | } |
1780 | ||
1781 | static void | |
1782 | port_modified(struct ofport *port_) | |
1783 | { | |
1784 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1785 | ||
1786 | if (port->bundle && port->bundle->bond) { | |
1787 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1788 | } | |
9d46b444 EJ |
1789 | |
1790 | if (port->cfm) { | |
1791 | cfm_set_netdev(port->cfm, port->up.netdev); | |
1792 | } | |
f1e78bbd | 1793 | |
42943cde EJ |
1794 | if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, |
1795 | port->odp_port)) { | |
7894e7da EJ |
1796 | ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = |
1797 | REV_RECONFIGURE; | |
f1e78bbd | 1798 | } |
6cbbf4fa EJ |
1799 | |
1800 | ofport_update_peer(port); | |
abe529af BP |
1801 | } |
1802 | ||
1803 | static void | |
9e1fd49b | 1804 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1805 | { |
1806 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1807 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1808 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1809 | |
9e1fd49b | 1810 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1811 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1812 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1813 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1814 | |
9e1fd49b | 1815 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1816 | bundle_update(port->bundle); |
1817 | } | |
abe529af BP |
1818 | } |
1819 | } | |
1820 | ||
1821 | static int | |
1822 | set_sflow(struct ofproto *ofproto_, | |
1823 | const struct ofproto_sflow_options *sflow_options) | |
1824 | { | |
1825 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1826 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1827 | |
abe529af | 1828 | if (sflow_options) { |
bae473fe | 1829 | if (!ds) { |
abe529af BP |
1830 | struct ofport_dpif *ofport; |
1831 | ||
4213f19d | 1832 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1833 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1834 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1835 | } |
2cc3c58e | 1836 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1837 | } |
bae473fe | 1838 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1839 | } else { |
6ff686f2 | 1840 | if (ds) { |
9723bcce | 1841 | dpif_sflow_unref(ds); |
2cc3c58e | 1842 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1843 | ofproto->sflow = NULL; |
1844 | } | |
abe529af BP |
1845 | } |
1846 | return 0; | |
1847 | } | |
1848 | ||
29089a54 RL |
1849 | static int |
1850 | set_ipfix( | |
1851 | struct ofproto *ofproto_, | |
1852 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1853 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1854 | size_t n_flow_exporters_options) | |
1855 | { | |
1856 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1857 | struct dpif_ipfix *di = ofproto->ipfix; | |
1858 | ||
1859 | if (bridge_exporter_options || flow_exporters_options) { | |
1860 | if (!di) { | |
1861 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1862 | } | |
1863 | dpif_ipfix_set_options( | |
1864 | di, bridge_exporter_options, flow_exporters_options, | |
1865 | n_flow_exporters_options); | |
1866 | } else { | |
1867 | if (di) { | |
d857c8aa | 1868 | dpif_ipfix_unref(di); |
29089a54 RL |
1869 | ofproto->ipfix = NULL; |
1870 | } | |
1871 | } | |
1872 | return 0; | |
1873 | } | |
1874 | ||
abe529af | 1875 | static int |
a5610457 | 1876 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1877 | { |
1878 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1879 | int error; | |
1880 | ||
a5610457 | 1881 | if (!s) { |
abe529af BP |
1882 | error = 0; |
1883 | } else { | |
1884 | if (!ofport->cfm) { | |
8c977421 EJ |
1885 | struct ofproto_dpif *ofproto; |
1886 | ||
1887 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1888 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 1889 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
1890 | } |
1891 | ||
a5610457 | 1892 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1893 | return 0; |
1894 | } | |
1895 | ||
1896 | error = EINVAL; | |
1897 | } | |
8e9a73fa | 1898 | cfm_unref(ofport->cfm); |
abe529af BP |
1899 | ofport->cfm = NULL; |
1900 | return error; | |
1901 | } | |
1902 | ||
9a9e3786 BP |
1903 | static bool |
1904 | get_cfm_status(const struct ofport *ofport_, | |
1905 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1906 | { |
1907 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1908 | ||
1909 | if (ofport->cfm) { | |
9a9e3786 BP |
1910 | status->faults = cfm_get_fault(ofport->cfm); |
1911 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1912 | status->health = cfm_get_health(ofport->cfm); | |
1913 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1914 | return true; | |
1de11730 | 1915 | } else { |
9a9e3786 | 1916 | return false; |
1de11730 EJ |
1917 | } |
1918 | } | |
ccc09689 EJ |
1919 | |
1920 | static int | |
1921 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
1922 | { | |
1923 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
1924 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1925 | struct bfd *old; | |
1926 | ||
1927 | old = ofport->bfd; | |
1928 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
1929 | if (ofport->bfd != old) { | |
1930 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1931 | } | |
1932 | ||
1933 | return 0; | |
1934 | } | |
1935 | ||
1936 | static int | |
1937 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
1938 | { | |
1939 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1940 | ||
1941 | if (ofport->bfd) { | |
1942 | bfd_get_status(ofport->bfd, smap); | |
1943 | return 0; | |
1944 | } else { | |
1945 | return ENOENT; | |
1946 | } | |
1947 | } | |
abe529af | 1948 | \f |
21f7563c JP |
1949 | /* Spanning Tree. */ |
1950 | ||
1951 | static void | |
1952 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1953 | { | |
1954 | struct ofproto_dpif *ofproto = ofproto_; | |
1955 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1956 | struct ofport_dpif *ofport; | |
1957 | ||
1958 | ofport = stp_port_get_aux(sp); | |
1959 | if (!ofport) { | |
1960 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1961 | ofproto->up.name, port_num); | |
1962 | } else { | |
1963 | struct eth_header *eth = pkt->l2; | |
1964 | ||
1965 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1966 | if (eth_addr_is_zero(eth->eth_src)) { | |
1967 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1968 | "with unknown MAC", ofproto->up.name, port_num); | |
1969 | } else { | |
97d6520b | 1970 | send_packet(ofport, pkt); |
21f7563c JP |
1971 | } |
1972 | } | |
1973 | ofpbuf_delete(pkt); | |
1974 | } | |
1975 | ||
1976 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1977 | static int | |
1978 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1979 | { | |
1980 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1981 | ||
1982 | /* Only revalidate flows if the configuration changed. */ | |
1983 | if (!s != !ofproto->stp) { | |
2cc3c58e | 1984 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
1985 | } |
1986 | ||
1987 | if (s) { | |
1988 | if (!ofproto->stp) { | |
1989 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1990 | send_bpdu_cb, ofproto); | |
1991 | ofproto->stp_last_tick = time_msec(); | |
1992 | } | |
1993 | ||
1994 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1995 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1996 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1997 | stp_set_max_age(ofproto->stp, s->max_age); | |
1998 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1999 | } else { | |
851bf71d EJ |
2000 | struct ofport *ofport; |
2001 | ||
2002 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2003 | set_stp_port(ofport, NULL); | |
2004 | } | |
2005 | ||
bd54dbcd | 2006 | stp_unref(ofproto->stp); |
21f7563c JP |
2007 | ofproto->stp = NULL; |
2008 | } | |
2009 | ||
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | static int | |
2014 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2015 | { | |
2016 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2017 | ||
2018 | if (ofproto->stp) { | |
2019 | s->enabled = true; | |
2020 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2021 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2022 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2023 | } else { | |
2024 | s->enabled = false; | |
2025 | } | |
2026 | ||
2027 | return 0; | |
2028 | } | |
2029 | ||
2030 | static void | |
2031 | update_stp_port_state(struct ofport_dpif *ofport) | |
2032 | { | |
2033 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2034 | enum stp_state state; | |
2035 | ||
2036 | /* Figure out new state. */ | |
2037 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2038 | : STP_DISABLED; | |
2039 | ||
2040 | /* Update state. */ | |
2041 | if (ofport->stp_state != state) { | |
9e1fd49b | 2042 | enum ofputil_port_state of_state; |
21f7563c JP |
2043 | bool fwd_change; |
2044 | ||
2045 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2046 | netdev_get_name(ofport->up.netdev), | |
2047 | stp_state_name(ofport->stp_state), | |
2048 | stp_state_name(state)); | |
2049 | if (stp_learn_in_state(ofport->stp_state) | |
2050 | != stp_learn_in_state(state)) { | |
2051 | /* xxx Learning action flows should also be flushed. */ | |
509c0149 | 2052 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2053 | mac_learning_flush(ofproto->ml); |
509c0149 | 2054 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c JP |
2055 | } |
2056 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2057 | != stp_forward_in_state(state); | |
2058 | ||
2cc3c58e | 2059 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2060 | ofport->stp_state = state; |
2061 | ofport->stp_state_entered = time_msec(); | |
2062 | ||
b308140a | 2063 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2064 | bundle_update(ofport->bundle); |
2065 | } | |
2066 | ||
2067 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2068 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2069 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2070 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2071 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2072 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2073 | : 0); | |
21f7563c JP |
2074 | ofproto_port_set_state(&ofport->up, of_state); |
2075 | } | |
2076 | } | |
2077 | ||
2078 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2079 | * caller is responsible for assigning STP port numbers and ensuring | |
2080 | * there are no duplicates. */ | |
2081 | static int | |
2082 | set_stp_port(struct ofport *ofport_, | |
2083 | const struct ofproto_port_stp_settings *s) | |
2084 | { | |
2085 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2086 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2087 | struct stp_port *sp = ofport->stp_port; | |
2088 | ||
2089 | if (!s || !s->enable) { | |
2090 | if (sp) { | |
2091 | ofport->stp_port = NULL; | |
2092 | stp_port_disable(sp); | |
ecd12731 | 2093 | update_stp_port_state(ofport); |
21f7563c JP |
2094 | } |
2095 | return 0; | |
2096 | } else if (sp && stp_port_no(sp) != s->port_num | |
2097 | && ofport == stp_port_get_aux(sp)) { | |
2098 | /* The port-id changed, so disable the old one if it's not | |
2099 | * already in use by another port. */ | |
2100 | stp_port_disable(sp); | |
2101 | } | |
2102 | ||
2103 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2104 | stp_port_enable(sp); | |
2105 | ||
2106 | stp_port_set_aux(sp, ofport); | |
2107 | stp_port_set_priority(sp, s->priority); | |
2108 | stp_port_set_path_cost(sp, s->path_cost); | |
2109 | ||
2110 | update_stp_port_state(ofport); | |
2111 | ||
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | static int | |
2116 | get_stp_port_status(struct ofport *ofport_, | |
2117 | struct ofproto_port_stp_status *s) | |
2118 | { | |
2119 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2120 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2121 | struct stp_port *sp = ofport->stp_port; | |
2122 | ||
2123 | if (!ofproto->stp || !sp) { | |
2124 | s->enabled = false; | |
2125 | return 0; | |
2126 | } | |
2127 | ||
2128 | s->enabled = true; | |
2129 | s->port_id = stp_port_get_id(sp); | |
2130 | s->state = stp_port_get_state(sp); | |
2131 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2132 | s->role = stp_port_get_role(sp); | |
80740385 | 2133 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2134 | |
2135 | return 0; | |
2136 | } | |
2137 | ||
2138 | static void | |
2139 | stp_run(struct ofproto_dpif *ofproto) | |
2140 | { | |
2141 | if (ofproto->stp) { | |
2142 | long long int now = time_msec(); | |
2143 | long long int elapsed = now - ofproto->stp_last_tick; | |
2144 | struct stp_port *sp; | |
2145 | ||
2146 | if (elapsed > 0) { | |
2147 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2148 | ofproto->stp_last_tick = now; | |
2149 | } | |
2150 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2151 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2152 | ||
2153 | if (ofport) { | |
2154 | update_stp_port_state(ofport); | |
2155 | } | |
2156 | } | |
6ae50723 EJ |
2157 | |
2158 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
509c0149 | 2159 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2160 | mac_learning_flush(ofproto->ml); |
509c0149 | 2161 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
6ae50723 | 2162 | } |
21f7563c JP |
2163 | } |
2164 | } | |
2165 | ||
2166 | static void | |
2167 | stp_wait(struct ofproto_dpif *ofproto) | |
2168 | { | |
2169 | if (ofproto->stp) { | |
2170 | poll_timer_wait(1000); | |
2171 | } | |
2172 | } | |
21f7563c | 2173 | \f |
cdc3ab65 EJ |
2174 | int |
2175 | ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto, | |
2176 | uint32_t queue_id, uint32_t *priority) | |
2177 | { | |
2178 | return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority); | |
2179 | } | |
2180 | ||
8b36f51e | 2181 | static int |
55954f6e | 2182 | set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp, |
8b36f51e EJ |
2183 | size_t n_qdscp) |
2184 | { | |
2185 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2186 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
8b36f51e | 2187 | |
55954f6e EJ |
2188 | if (ofport->n_qdscp != n_qdscp |
2189 | || (n_qdscp && memcmp(ofport->qdscp, qdscp, | |
2190 | n_qdscp * sizeof *qdscp))) { | |
2cc3c58e | 2191 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
55954f6e EJ |
2192 | free(ofport->qdscp); |
2193 | ofport->qdscp = n_qdscp | |
2194 | ? xmemdup(qdscp, n_qdscp * sizeof *qdscp) | |
2195 | : NULL; | |
2196 | ofport->n_qdscp = n_qdscp; | |
8b36f51e EJ |
2197 | } |
2198 | ||
8b36f51e EJ |
2199 | return 0; |
2200 | } | |
2201 | \f | |
abe529af BP |
2202 | /* Bundles. */ |
2203 | ||
b44a10b7 BP |
2204 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2205 | * ofproto to revalidate every flow. | |
2206 | * | |
2207 | * Normally MAC learning entries are removed only from the ofproto associated | |
2208 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2209 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2210 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2211 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2212 | * with the host from which it migrated. */ | |
abe529af | 2213 | static void |
b44a10b7 | 2214 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2215 | { |
2216 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2217 | struct mac_learning *ml = ofproto->ml; | |
2218 | struct mac_entry *mac, *next_mac; | |
2219 | ||
2cc3c58e | 2220 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
509c0149 | 2221 | ovs_rwlock_wrlock(&ml->rwlock); |
abe529af BP |
2222 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2223 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2224 | if (all_ofprotos) { |
2225 | struct ofproto_dpif *o; | |
2226 | ||
2227 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2228 | if (o != ofproto) { | |
2229 | struct mac_entry *e; | |
2230 | ||
509c0149 | 2231 | ovs_rwlock_wrlock(&o->ml->rwlock); |
30618594 | 2232 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan); |
b44a10b7 | 2233 | if (e) { |
b44a10b7 BP |
2234 | mac_learning_expire(o->ml, e); |
2235 | } | |
509c0149 | 2236 | ovs_rwlock_unlock(&o->ml->rwlock); |
b44a10b7 BP |
2237 | } |
2238 | } | |
2239 | } | |
2240 | ||
abe529af BP |
2241 | mac_learning_expire(ml, mac); |
2242 | } | |
2243 | } | |
509c0149 | 2244 | ovs_rwlock_unlock(&ml->rwlock); |
abe529af BP |
2245 | } |
2246 | ||
2247 | static struct ofbundle * | |
2248 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2249 | { | |
2250 | struct ofbundle *bundle; | |
2251 | ||
2252 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2253 | &ofproto->bundles) { | |
2254 | if (bundle->aux == aux) { | |
2255 | return bundle; | |
2256 | } | |
2257 | } | |
2258 | return NULL; | |
2259 | } | |
2260 | ||
7bde8dd8 JP |
2261 | static void |
2262 | bundle_update(struct ofbundle *bundle) | |
2263 | { | |
2264 | struct ofport_dpif *port; | |
2265 | ||
2266 | bundle->floodable = true; | |
2267 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2268 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2269 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2270 | bundle->floodable = false; |
2271 | break; | |
2272 | } | |
2273 | } | |
2274 | } | |
2275 | ||
abe529af BP |
2276 | static void |
2277 | bundle_del_port(struct ofport_dpif *port) | |
2278 | { | |
2279 | struct ofbundle *bundle = port->bundle; | |
2280 | ||
2cc3c58e | 2281 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2282 | |
abe529af BP |
2283 | list_remove(&port->bundle_node); |
2284 | port->bundle = NULL; | |
2285 | ||
2286 | if (bundle->lacp) { | |
2287 | lacp_slave_unregister(bundle->lacp, port); | |
2288 | } | |
2289 | if (bundle->bond) { | |
2290 | bond_slave_unregister(bundle->bond, port); | |
2291 | } | |
2292 | ||
7bde8dd8 | 2293 | bundle_update(bundle); |
abe529af BP |
2294 | } |
2295 | ||
2296 | static bool | |
4e022ec0 | 2297 | bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, |
df53d41c | 2298 | struct lacp_slave_settings *lacp) |
abe529af BP |
2299 | { |
2300 | struct ofport_dpif *port; | |
2301 | ||
2302 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2303 | if (!port) { | |
2304 | return false; | |
2305 | } | |
2306 | ||
2307 | if (port->bundle != bundle) { | |
2cc3c58e | 2308 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2309 | if (port->bundle) { |
2310 | bundle_del_port(port); | |
2311 | } | |
2312 | ||
2313 | port->bundle = bundle; | |
2314 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2315 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2316 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2317 | bundle->floodable = false; |
2318 | } | |
2319 | } | |
2320 | if (lacp) { | |
2cc3c58e | 2321 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2322 | lacp_slave_register(bundle->lacp, port, lacp); |
2323 | } | |
2324 | ||
2325 | return true; | |
2326 | } | |
2327 | ||
2328 | static void | |
2329 | bundle_destroy(struct ofbundle *bundle) | |
2330 | { | |
2331 | struct ofproto_dpif *ofproto; | |
2332 | struct ofport_dpif *port, *next_port; | |
abe529af BP |
2333 | |
2334 | if (!bundle) { | |
2335 | return; | |
2336 | } | |
2337 | ||
2338 | ofproto = bundle->ofproto; | |
ec7ceaed | 2339 | mbridge_unregister_bundle(ofproto->mbridge, bundle->aux); |
abe529af | 2340 | |
46c88433 EJ |
2341 | xlate_bundle_remove(bundle); |
2342 | ||
abe529af BP |
2343 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { |
2344 | bundle_del_port(port); | |
2345 | } | |
2346 | ||
b44a10b7 | 2347 | bundle_flush_macs(bundle, true); |
abe529af BP |
2348 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2349 | free(bundle->name); | |
2350 | free(bundle->trunks); | |
91779071 | 2351 | lacp_unref(bundle->lacp); |
03366a2d | 2352 | bond_unref(bundle->bond); |
abe529af BP |
2353 | free(bundle); |
2354 | } | |
2355 | ||
2356 | static int | |
2357 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2358 | const struct ofproto_bundle_settings *s) | |
2359 | { | |
2360 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2361 | bool need_flush = false; | |
abe529af BP |
2362 | struct ofport_dpif *port; |
2363 | struct ofbundle *bundle; | |
ecac4ebf BP |
2364 | unsigned long *trunks; |
2365 | int vlan; | |
abe529af BP |
2366 | size_t i; |
2367 | bool ok; | |
2368 | ||
2369 | if (!s) { | |
2370 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2371 | return 0; | |
2372 | } | |
2373 | ||
cb22974d BP |
2374 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2375 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2376 | |
2377 | bundle = bundle_lookup(ofproto, aux); | |
2378 | if (!bundle) { | |
2379 | bundle = xmalloc(sizeof *bundle); | |
2380 | ||
2381 | bundle->ofproto = ofproto; | |
2382 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2383 | hash_pointer(aux, 0)); | |
2384 | bundle->aux = aux; | |
2385 | bundle->name = NULL; | |
2386 | ||
2387 | list_init(&bundle->ports); | |
ecac4ebf | 2388 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2389 | bundle->vlan = -1; |
2390 | bundle->trunks = NULL; | |
5e9ceccd | 2391 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2392 | bundle->lacp = NULL; |
2393 | bundle->bond = NULL; | |
2394 | ||
2395 | bundle->floodable = true; | |
ec7ceaed | 2396 | mbridge_register_bundle(ofproto->mbridge, bundle); |
abe529af BP |
2397 | } |
2398 | ||
2399 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2400 | free(bundle->name); | |
2401 | bundle->name = xstrdup(s->name); | |
2402 | } | |
2403 | ||
2404 | /* LACP. */ | |
2405 | if (s->lacp) { | |
2406 | if (!bundle->lacp) { | |
2cc3c58e | 2407 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2408 | bundle->lacp = lacp_create(); |
2409 | } | |
2410 | lacp_configure(bundle->lacp, s->lacp); | |
2411 | } else { | |
91779071 | 2412 | lacp_unref(bundle->lacp); |
abe529af BP |
2413 | bundle->lacp = NULL; |
2414 | } | |
2415 | ||
2416 | /* Update set of ports. */ | |
2417 | ok = true; | |
2418 | for (i = 0; i < s->n_slaves; i++) { | |
2419 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2420 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2421 | ok = false; |
2422 | } | |
2423 | } | |
2424 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2425 | struct ofport_dpif *next_port; | |
2426 | ||
2427 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2428 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2429 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2430 | goto found; |
2431 | } | |
2432 | } | |
2433 | ||
2434 | bundle_del_port(port); | |
2435 | found: ; | |
2436 | } | |
2437 | } | |
cb22974d | 2438 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2439 | |
2440 | if (list_is_empty(&bundle->ports)) { | |
2441 | bundle_destroy(bundle); | |
2442 | return EINVAL; | |
2443 | } | |
2444 | ||
ecac4ebf | 2445 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2446 | if (s->vlan_mode != bundle->vlan_mode |
2447 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2448 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2449 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2450 | need_flush = true; |
2451 | } | |
2452 | ||
abe529af | 2453 | /* Set VLAN tag. */ |
ecac4ebf BP |
2454 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2455 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2456 | : 0); | |
2457 | if (vlan != bundle->vlan) { | |
2458 | bundle->vlan = vlan; | |
abe529af BP |
2459 | need_flush = true; |
2460 | } | |
2461 | ||
2462 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2463 | switch (s->vlan_mode) { |
2464 | case PORT_VLAN_ACCESS: | |
2465 | trunks = NULL; | |
2466 | break; | |
2467 | ||
2468 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2469 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2470 | break; |
2471 | ||
2472 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2473 | case PORT_VLAN_NATIVE_TAGGED: | |
2474 | if (vlan != 0 && (!s->trunks | |
2475 | || !bitmap_is_set(s->trunks, vlan) | |
2476 | || bitmap_is_set(s->trunks, 0))) { | |
2477 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2478 | if (s->trunks) { | |
2479 | trunks = bitmap_clone(s->trunks, 4096); | |
2480 | } else { | |
2481 | trunks = bitmap_allocate1(4096); | |
2482 | } | |
2483 | bitmap_set1(trunks, vlan); | |
2484 | bitmap_set0(trunks, 0); | |
2485 | } else { | |
ebc56baa | 2486 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2487 | } |
2488 | break; | |
2489 | ||
2490 | default: | |
2491 | NOT_REACHED(); | |
2492 | } | |
abe529af BP |
2493 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2494 | free(bundle->trunks); | |
ecac4ebf BP |
2495 | if (trunks == s->trunks) { |
2496 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2497 | } else { | |
2498 | bundle->trunks = trunks; | |
2499 | trunks = NULL; | |
2500 | } | |
abe529af BP |
2501 | need_flush = true; |
2502 | } | |
ecac4ebf BP |
2503 | if (trunks != s->trunks) { |
2504 | free(trunks); | |
2505 | } | |
abe529af BP |
2506 | |
2507 | /* Bonding. */ | |
2508 | if (!list_is_short(&bundle->ports)) { | |
2509 | bundle->ofproto->has_bonded_bundles = true; | |
2510 | if (bundle->bond) { | |
2511 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2512 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2513 | } |
2514 | } else { | |
2515 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2516 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2517 | } |
2518 | ||
2519 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2520 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2521 | } |
2522 | } else { | |
03366a2d | 2523 | bond_unref(bundle->bond); |
abe529af BP |
2524 | bundle->bond = NULL; |
2525 | } | |
2526 | ||
2527 | /* If we changed something that would affect MAC learning, un-learn | |
2528 | * everything on this port and force flow revalidation. */ | |
2529 | if (need_flush) { | |
b44a10b7 | 2530 | bundle_flush_macs(bundle, false); |
abe529af BP |
2531 | } |
2532 | ||
2533 | return 0; | |
2534 | } | |
2535 | ||
2536 | static void | |
2537 | bundle_remove(struct ofport *port_) | |
2538 | { | |
2539 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2540 | struct ofbundle *bundle = port->bundle; | |
2541 | ||
2542 | if (bundle) { | |
2543 | bundle_del_port(port); | |
2544 | if (list_is_empty(&bundle->ports)) { | |
2545 | bundle_destroy(bundle); | |
2546 | } else if (list_is_short(&bundle->ports)) { | |
03366a2d | 2547 | bond_unref(bundle->bond); |
abe529af BP |
2548 | bundle->bond = NULL; |
2549 | } | |
2550 | } | |
2551 | } | |
2552 | ||
2553 | static void | |
5f877369 | 2554 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2555 | { |
2556 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2557 | struct ofport_dpif *port = port_; | |
2558 | uint8_t ea[ETH_ADDR_LEN]; | |
2559 | int error; | |
2560 | ||
2561 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2562 | if (!error) { | |
abe529af | 2563 | struct ofpbuf packet; |
5f877369 | 2564 | void *packet_pdu; |
abe529af BP |
2565 | |
2566 | ofpbuf_init(&packet, 0); | |
2567 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2568 | pdu_size); |
2569 | memcpy(packet_pdu, pdu, pdu_size); | |
2570 | ||
97d6520b | 2571 | send_packet(port, &packet); |
abe529af BP |
2572 | ofpbuf_uninit(&packet); |
2573 | } else { | |
2574 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2575 | "%s (%s)", port->bundle->name, | |
10a89ef0 | 2576 | netdev_get_name(port->up.netdev), ovs_strerror(error)); |
abe529af BP |
2577 | } |
2578 | } | |
2579 | ||
2580 | static void | |
2581 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2582 | { | |
2583 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2584 | int error, n_packets, n_errors; | |
2585 | struct mac_entry *e; | |
2586 | ||
2587 | error = n_packets = n_errors = 0; | |
509c0149 | 2588 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
2589 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
2590 | if (e->port.p != bundle) { | |
ea131871 JG |
2591 | struct ofpbuf *learning_packet; |
2592 | struct ofport_dpif *port; | |
4dd1e3ca | 2593 | void *port_void; |
ea131871 JG |
2594 | int ret; |
2595 | ||
4dd1e3ca BP |
2596 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2597 | * struct ofport_dpif more effective. */ | |
2598 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2599 | e->mac, e->vlan, | |
2600 | &port_void); | |
2601 | port = port_void; | |
97d6520b | 2602 | ret = send_packet(port, learning_packet); |
ea131871 | 2603 | ofpbuf_delete(learning_packet); |
abe529af BP |
2604 | if (ret) { |
2605 | error = ret; | |
2606 | n_errors++; | |
2607 | } | |
2608 | n_packets++; | |
2609 | } | |
2610 | } | |
509c0149 | 2611 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2612 | |
2613 | if (n_errors) { | |
2614 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2615 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2616 | "packets, last error was: %s", | |
10a89ef0 | 2617 | bundle->name, n_errors, n_packets, ovs_strerror(error)); |
abe529af BP |
2618 | } else { |
2619 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2620 | bundle->name, n_packets); | |
2621 | } | |
2622 | } | |
2623 | ||
2624 | static void | |
2625 | bundle_run(struct ofbundle *bundle) | |
2626 | { | |
2627 | if (bundle->lacp) { | |
2628 | lacp_run(bundle->lacp, send_pdu_cb); | |
2629 | } | |
2630 | if (bundle->bond) { | |
2631 | struct ofport_dpif *port; | |
2632 | ||
2633 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2634 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2635 | } |
2636 | ||
4a1b8f30 EJ |
2637 | if (bond_run(bundle->bond, lacp_status(bundle->lacp))) { |
2638 | bundle->ofproto->backer->need_revalidate = REV_BOND; | |
2639 | } | |
2640 | ||
abe529af BP |
2641 | if (bond_should_send_learning_packets(bundle->bond)) { |
2642 | bundle_send_learning_packets(bundle); | |
2643 | } | |
2644 | } | |
2645 | } | |
2646 | ||
2647 | static void | |
2648 | bundle_wait(struct ofbundle *bundle) | |
2649 | { | |
2650 | if (bundle->lacp) { | |
2651 | lacp_wait(bundle->lacp); | |
2652 | } | |
2653 | if (bundle->bond) { | |
2654 | bond_wait(bundle->bond); | |
2655 | } | |
2656 | } | |
2657 | \f | |
2658 | /* Mirrors. */ | |
2659 | ||
2660 | static int | |
ec7ceaed EJ |
2661 | mirror_set__(struct ofproto *ofproto_, void *aux, |
2662 | const struct ofproto_mirror_settings *s) | |
abe529af BP |
2663 | { |
2664 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ec7ceaed EJ |
2665 | struct ofbundle **srcs, **dsts; |
2666 | int error; | |
2667 | size_t i; | |
abe529af | 2668 | |
abe529af | 2669 | if (!s) { |
ec7ceaed | 2670 | mirror_destroy(ofproto->mbridge, aux); |
abe529af BP |
2671 | return 0; |
2672 | } | |
2673 | ||
ec7ceaed EJ |
2674 | srcs = xmalloc(s->n_srcs * sizeof *srcs); |
2675 | dsts = xmalloc(s->n_dsts * sizeof *dsts); | |
abe529af | 2676 | |
ec7ceaed EJ |
2677 | for (i = 0; i < s->n_srcs; i++) { |
2678 | srcs[i] = bundle_lookup(ofproto, s->srcs[i]); | |
abe529af BP |
2679 | } |
2680 | ||
ec7ceaed EJ |
2681 | for (i = 0; i < s->n_dsts; i++) { |
2682 | dsts[i] = bundle_lookup(ofproto, s->dsts[i]); | |
abe529af BP |
2683 | } |
2684 | ||
ec7ceaed EJ |
2685 | error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts, |
2686 | s->n_dsts, s->src_vlans, | |
2687 | bundle_lookup(ofproto, s->out_bundle), s->out_vlan); | |
2688 | free(srcs); | |
2689 | free(dsts); | |
2690 | return error; | |
abe529af BP |
2691 | } |
2692 | ||
9d24de3b | 2693 | static int |
ec7ceaed EJ |
2694 | mirror_get_stats__(struct ofproto *ofproto, void *aux, |
2695 | uint64_t *packets, uint64_t *bytes) | |
9d24de3b | 2696 | { |
8844e035 | 2697 | push_all_stats(); |
ec7ceaed EJ |
2698 | return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets, |
2699 | bytes); | |
9d24de3b JP |
2700 | } |
2701 | ||
abe529af BP |
2702 | static int |
2703 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2704 | { | |
2705 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2706 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
abe529af | 2707 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { |
30618594 | 2708 | mac_learning_flush(ofproto->ml); |
abe529af | 2709 | } |
509c0149 | 2710 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2711 | return 0; |
2712 | } | |
2713 | ||
2714 | static bool | |
b4affc74 | 2715 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2716 | { |
2717 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2718 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
ec7ceaed | 2719 | return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0; |
abe529af | 2720 | } |
8402c74b SS |
2721 | |
2722 | static void | |
b53055f4 | 2723 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2724 | { |
2725 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2726 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2727 | } |
e764773c BP |
2728 | |
2729 | static void | |
c4069512 BP |
2730 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2731 | size_t max_entries) | |
e764773c BP |
2732 | { |
2733 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2734 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
e764773c | 2735 | mac_learning_set_idle_time(ofproto->ml, idle_time); |
c4069512 | 2736 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
509c0149 | 2737 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
e764773c | 2738 | } |
abe529af BP |
2739 | \f |
2740 | /* Ports. */ | |
2741 | ||
46c88433 | 2742 | static struct ofport_dpif * |
4e022ec0 | 2743 | get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
abe529af | 2744 | { |
7df6a8bd BP |
2745 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2746 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2747 | } |
2748 | ||
46c88433 | 2749 | static struct ofport_dpif * |
4e022ec0 | 2750 | get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) |
abe529af | 2751 | { |
7c33b188 JR |
2752 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2753 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2754 | } |
2755 | ||
2756 | static void | |
e1b1d06a JP |
2757 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2758 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2759 | struct dpif_port *dpif_port) |
2760 | { | |
2761 | ofproto_port->name = dpif_port->name; | |
2762 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2763 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2764 | } |
2765 | ||
6cbbf4fa EJ |
2766 | static void |
2767 | ofport_update_peer(struct ofport_dpif *ofport) | |
0a740f48 EJ |
2768 | { |
2769 | const struct ofproto_dpif *ofproto; | |
6cbbf4fa EJ |
2770 | struct dpif_backer *backer; |
2771 | const char *peer_name; | |
2772 | ||
2773 | if (!netdev_vport_is_patch(ofport->up.netdev)) { | |
2774 | return; | |
2775 | } | |
2776 | ||
2777 | backer = ofproto_dpif_cast(ofport->up.ofproto)->backer; | |
7894e7da | 2778 | backer->need_revalidate = REV_RECONFIGURE; |
0a740f48 | 2779 | |
6cbbf4fa EJ |
2780 | if (ofport->peer) { |
2781 | ofport->peer->peer = NULL; | |
2782 | ofport->peer = NULL; | |
2783 | } | |
2784 | ||
2785 | peer_name = netdev_vport_patch_peer(ofport->up.netdev); | |
2786 | if (!peer_name) { | |
2787 | return; | |
0a740f48 EJ |
2788 | } |
2789 | ||
2790 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6cbbf4fa EJ |
2791 | struct ofport *peer_ofport; |
2792 | struct ofport_dpif *peer; | |
2793 | const char *peer_peer; | |
0a740f48 | 2794 | |
462abef2 EJ |
2795 | if (ofproto->backer != backer) { |
2796 | continue; | |
2797 | } | |
2798 | ||
6cbbf4fa EJ |
2799 | peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name); |
2800 | if (!peer_ofport) { | |
2801 | continue; | |
2802 | } | |
2803 | ||
2804 | peer = ofport_dpif_cast(peer_ofport); | |
2805 | peer_peer = netdev_vport_patch_peer(peer->up.netdev); | |
2806 | if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev), | |
2807 | peer_peer)) { | |
2808 | ofport->peer = peer; | |
2809 | ofport->peer->peer = ofport; | |
0a740f48 | 2810 | } |
6cbbf4fa EJ |
2811 | |
2812 | return; | |
0a740f48 | 2813 | } |
0a740f48 EJ |
2814 | } |
2815 | ||
0aa66d6e EJ |
2816 | static void |
2817 | port_run_fast(struct ofport_dpif *ofport) | |
2818 | { | |
2819 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2820 | struct ofpbuf packet; | |
2821 | ||
2822 | ofpbuf_init(&packet, 0); | |
2823 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2824 | send_packet(ofport, &packet); | |
2825 | ofpbuf_uninit(&packet); | |
2826 | } | |
ccc09689 EJ |
2827 | |
2828 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
2829 | struct ofpbuf packet; | |
2830 | ||
2831 | ofpbuf_init(&packet, 0); | |
2832 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
2833 | send_packet(ofport, &packet); | |
2834 | ofpbuf_uninit(&packet); | |
2835 | } | |
0aa66d6e EJ |
2836 | } |
2837 | ||
abe529af BP |
2838 | static void |
2839 | port_run(struct ofport_dpif *ofport) | |
2840 | { | |
3e5b3fdb EJ |
2841 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2842 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2843 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2844 | ||
3e5b3fdb EJ |
2845 | ofport->carrier_seq = carrier_seq; |
2846 | ||
0aa66d6e | 2847 | port_run_fast(ofport); |
b9ad7294 | 2848 | |
abe529af | 2849 | if (ofport->cfm) { |
4653c558 EJ |
2850 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2851 | ||
abe529af | 2852 | cfm_run(ofport->cfm); |
4653c558 EJ |
2853 | enable = enable && !cfm_get_fault(ofport->cfm); |
2854 | ||
2855 | if (cfm_opup >= 0) { | |
2856 | enable = enable && cfm_opup; | |
2857 | } | |
abe529af | 2858 | } |
015e08bc | 2859 | |
ccc09689 EJ |
2860 | if (ofport->bfd) { |
2861 | bfd_run(ofport->bfd); | |
2862 | enable = enable && bfd_forwarding(ofport->bfd); | |
2863 | } | |
2864 | ||
015e08bc EJ |
2865 | if (ofport->bundle) { |
2866 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2867 | if (carrier_changed) { |
2868 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2869 | } | |
015e08bc EJ |
2870 | } |
2871 | ||
daff3353 EJ |
2872 | if (ofport->may_enable != enable) { |
2873 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
83341407 | 2874 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
2875 | } |
2876 | ||
015e08bc | 2877 | ofport->may_enable = enable; |
abe529af BP |
2878 | } |
2879 | ||
2880 | static void | |
2881 | port_wait(struct ofport_dpif *ofport) | |
2882 | { | |
2883 | if (ofport->cfm) { | |
2884 | cfm_wait(ofport->cfm); | |
2885 | } | |
ccc09689 EJ |
2886 | |
2887 | if (ofport->bfd) { | |
2888 | bfd_wait(ofport->bfd); | |
2889 | } | |
abe529af BP |
2890 | } |
2891 | ||
2892 | static int | |
2893 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2894 | struct ofproto_port *ofproto_port) | |
2895 | { | |
2896 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2897 | struct dpif_port dpif_port; | |
2898 | int error; | |
2899 | ||
0a740f48 EJ |
2900 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
2901 | const char *type = netdev_get_type_from_name(devname); | |
2902 | ||
2903 | /* We may be called before ofproto->up.port_by_name is populated with | |
2904 | * the appropriate ofport. For this reason, we must get the name and | |
2905 | * type from the netdev layer directly. */ | |
2906 | if (type) { | |
2907 | const struct ofport *ofport; | |
2908 | ||
2909 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
2910 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
2911 | ofproto_port->name = xstrdup(devname); | |
2912 | ofproto_port->type = xstrdup(type); | |
2913 | return 0; | |
2914 | } | |
2915 | return ENODEV; | |
2916 | } | |
2917 | ||
acf60855 JP |
2918 | if (!sset_contains(&ofproto->ports, devname)) { |
2919 | return ENODEV; | |
2920 | } | |
2921 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
2922 | devname, &dpif_port); | |
abe529af | 2923 | if (!error) { |
e1b1d06a | 2924 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
2925 | } |
2926 | return error; | |
2927 | } | |
2928 | ||
2929 | static int | |
e1b1d06a | 2930 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
2931 | { |
2932 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 2933 | const char *devname = netdev_get_name(netdev); |
3aa30359 BP |
2934 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
2935 | const char *dp_port_name; | |
abe529af | 2936 | |
0a740f48 EJ |
2937 | if (netdev_vport_is_patch(netdev)) { |
2938 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
2939 | return 0; | |
2940 | } | |
2941 | ||
3aa30359 | 2942 | dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
b9ad7294 | 2943 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
4e022ec0 | 2944 | odp_port_t port_no = ODPP_NONE; |
7d82ab2e KM |
2945 | int error; |
2946 | ||
2947 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
2948 | if (error) { |
2949 | return error; | |
2950 | } | |
7d82ab2e | 2951 | if (netdev_get_tunnel_config(netdev)) { |
4e022ec0 AW |
2952 | simap_put(&ofproto->backer->tnl_backers, |
2953 | dp_port_name, odp_to_u32(port_no)); | |
7d82ab2e | 2954 | } |
acf60855 | 2955 | } |
b9ad7294 EJ |
2956 | |
2957 | if (netdev_get_tunnel_config(netdev)) { | |
2958 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
2959 | } else { |
2960 | sset_add(&ofproto->ports, devname); | |
2961 | } | |
2962 | return 0; | |
2963 | } | |
2964 | ||
abe529af | 2965 | static int |
4e022ec0 | 2966 | port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) |
abe529af BP |
2967 | { |
2968 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 2969 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 2970 | int error = 0; |
abe529af | 2971 | |
b9ad7294 EJ |
2972 | if (!ofport) { |
2973 | return 0; | |
e1b1d06a | 2974 | } |
b9ad7294 EJ |
2975 | |
2976 | sset_find_and_delete(&ofproto->ghost_ports, | |
2977 | netdev_get_name(ofport->up.netdev)); | |
a614d823 | 2978 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
42943cde | 2979 | if (!ofport->is_tunnel) { |
b9ad7294 EJ |
2980 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
2981 | if (!error) { | |
abe529af BP |
2982 | /* The caller is going to close ofport->up.netdev. If this is a |
2983 | * bonded port, then the bond is using that netdev, so remove it | |
2984 | * from the bond. The client will need to reconfigure everything | |
2985 | * after deleting ports, so then the slave will get re-added. */ | |
2986 | bundle_remove(&ofport->up); | |
2987 | } | |
2988 | } | |
2989 | return error; | |
2990 | } | |
2991 | ||
6527c598 PS |
2992 | static int |
2993 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2994 | { | |
2995 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2996 | int error; | |
2997 | ||
8844e035 EJ |
2998 | push_all_stats(); |
2999 | ||
6527c598 PS |
3000 | error = netdev_get_stats(ofport->up.netdev, stats); |
3001 | ||
ee382d89 | 3002 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3003 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3004 | ||
3005 | /* ofproto->stats.tx_packets represents packets that we created | |
3006 | * internally and sent to some port (e.g. packets sent with | |
3007 | * send_packet()). Account for them as if they had come from | |
3008 | * OFPP_LOCAL and got forwarded. */ | |
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 | } | |
3029 | } | |
3030 | ||
3031 | return error; | |
3032 | } | |
3033 | ||
abe529af | 3034 | struct port_dump_state { |
acf60855 JP |
3035 | uint32_t bucket; |
3036 | uint32_t offset; | |
0a740f48 | 3037 | bool ghost; |
da78d43d BP |
3038 | |
3039 | struct ofproto_port port; | |
3040 | bool has_port; | |
abe529af BP |
3041 | }; |
3042 | ||
3043 | static int | |
acf60855 | 3044 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3045 | { |
0a740f48 | 3046 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3047 | return 0; |
3048 | } | |
3049 | ||
3050 | static int | |
b9ad7294 | 3051 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3052 | struct ofproto_port *port) |
3053 | { | |
e1b1d06a | 3054 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3055 | struct port_dump_state *state = state_; |
0a740f48 | 3056 | const struct sset *sset; |
acf60855 | 3057 | struct sset_node *node; |
abe529af | 3058 | |
da78d43d BP |
3059 | if (state->has_port) { |
3060 | ofproto_port_destroy(&state->port); | |
3061 | state->has_port = false; | |
3062 | } | |
0a740f48 EJ |
3063 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3064 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3065 | int error; |
3066 | ||
da78d43d BP |
3067 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3068 | if (!error) { | |
3069 | *port = state->port; | |
3070 | state->has_port = true; | |
3071 | return 0; | |
3072 | } else if (error != ENODEV) { | |
acf60855 JP |
3073 | return error; |
3074 | } | |
abe529af | 3075 | } |
acf60855 | 3076 | |
0a740f48 EJ |
3077 | if (!state->ghost) { |
3078 | state->ghost = true; | |
3079 | state->bucket = 0; | |
3080 | state->offset = 0; | |
3081 | return port_dump_next(ofproto_, state_, port); | |
3082 | } | |
3083 | ||
acf60855 | 3084 | return EOF; |
abe529af BP |
3085 | } |
3086 | ||
3087 | static int | |
3088 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3089 | { | |
3090 | struct port_dump_state *state = state_; | |
3091 | ||
da78d43d BP |
3092 | if (state->has_port) { |
3093 | ofproto_port_destroy(&state->port); | |
3094 | } | |
abe529af BP |
3095 | free(state); |
3096 | return 0; | |
3097 | } | |
3098 | ||
3099 | static int | |
3100 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3101 | { | |
3102 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3103 | |
3104 | if (ofproto->port_poll_errno) { | |
3105 | int error = ofproto->port_poll_errno; | |
3106 | ofproto->port_poll_errno = 0; | |
3107 | return error; | |
3108 | } | |
3109 | ||
3110 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3111 | return EAGAIN; | |
3112 | } | |
3113 | ||
3114 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3115 | return 0; | |
abe529af BP |
3116 | } |
3117 | ||
3118 | static void | |
3119 | port_poll_wait(const struct ofproto *ofproto_) | |
3120 | { | |
3121 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3122 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3123 | } |
3124 | ||
3125 | static int | |
3126 | port_is_lacp_current(const struct ofport *ofport_) | |
3127 | { | |
3128 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3129 | return (ofport->bundle && ofport->bundle->lacp | |
3130 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3131 | : -1); | |
3132 | } | |
3133 | \f | |
3134 | /* Upcall handling. */ | |
3135 | ||
501f8d1f BP |
3136 | /* Flow miss batching. |
3137 | * | |
3138 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3139 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3140 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3141 | * more packets, plus possibly installing the flow in the dpif. | |
3142 | * | |
3143 | * So far we only batch the operations that affect flow setup time the most. | |
3144 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3145 | struct flow_miss { | |
3146 | struct hmap_node hmap_node; | |
acf60855 | 3147 | struct ofproto_dpif *ofproto; |
501f8d1f | 3148 | struct flow flow; |
b0f7b9b5 | 3149 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3150 | const struct nlattr *key; |
3151 | size_t key_len; | |
3152 | struct list packets; | |
6a7e895f | 3153 | enum dpif_upcall_type upcall_type; |
501f8d1f BP |
3154 | }; |
3155 | ||
3156 | struct flow_miss_op { | |
c2b565b5 | 3157 | struct dpif_op dpif_op; |
bbafd73b EJ |
3158 | |
3159 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3160 | struct xlate_out xout; | |
3161 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
9b1a48c2 | 3162 | |
d445cc16 JP |
3163 | struct ofpbuf mask; /* Flow mask for "put" ops. */ |
3164 | struct odputil_keybuf maskbuf; | |
3165 | ||
9b1a48c2 JP |
3166 | /* If this is a "put" op, then a pointer to the subfacet that should |
3167 | * be marked as uninstalled if the operation fails. */ | |
3168 | struct subfacet *subfacet; | |
501f8d1f BP |
3169 | }; |
3170 | ||
62cd7072 BP |
3171 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3172 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3173 | * configurations. */ |
62cd7072 | 3174 | static void |
a39edbd4 | 3175 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3176 | const struct flow *flow) |
62cd7072 BP |
3177 | { |
3178 | struct ofputil_packet_in pin; | |
3179 | ||
3e3252fa EJ |
3180 | pin.packet = packet->data; |
3181 | pin.packet_len = packet->size; | |
62cd7072 | 3182 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3183 | pin.controller_id = 0; |
54834960 EJ |
3184 | |
3185 | pin.table_id = 0; | |
3186 | pin.cookie = 0; | |
3187 | ||
62cd7072 | 3188 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3189 | |
3190 | flow_get_metadata(flow, &pin.fmd); | |
3191 | ||
d8653c38 | 3192 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3193 | } |
3194 | ||
501f8d1f | 3195 | static struct flow_miss * |
ddbc5954 JP |
3196 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3197 | const struct flow *flow, uint32_t hash) | |
abe529af | 3198 | { |
501f8d1f | 3199 | struct flow_miss *miss; |
abe529af | 3200 | |
501f8d1f | 3201 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3202 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3203 | return miss; |
3204 | } | |
3205 | } | |
abe529af | 3206 | |
b23cdad9 | 3207 | return NULL; |
501f8d1f | 3208 | } |
abe529af | 3209 | |
9d6ac44e BP |
3210 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3211 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3212 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3213 | * returned op->subfacet, and if anything needs to be freed after processing | |
3214 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3215 | static void |
9d6ac44e BP |
3216 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3217 | struct flow_miss_op *op) | |
501f8d1f | 3218 | { |
4e022ec0 AW |
3219 | if (miss->flow.in_port.ofp_port |
3220 | != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port, | |
47b0deae | 3221 | miss->flow.vlan_tci)) { |
9d6ac44e BP |
3222 | /* This packet was received on a VLAN splinter port. We |
3223 | * added a VLAN to the packet to make the packet resemble | |
3224 | * the flow, but the actions were composed assuming that | |
3225 | * the packet contained no VLAN. So, we must remove the | |
3226 | * VLAN header from the packet before trying to execute the | |
3227 | * actions. */ | |
3228 | eth_pop_vlan(packet); | |
3229 | } | |
3230 | ||
9b1a48c2 | 3231 | op->subfacet = NULL; |
bbafd73b | 3232 | op->xout_garbage = false; |
9d6ac44e BP |
3233 | op->dpif_op.type = DPIF_OP_EXECUTE; |
3234 | op->dpif_op.u.execute.key = miss->key; | |
3235 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3236 | op->dpif_op.u.execute.packet = packet; | |
d445cc16 | 3237 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
9d6ac44e BP |
3238 | } |
3239 | ||
3240 | /* Helper for handle_flow_miss_without_facet() and | |
3241 | * handle_flow_miss_with_facet(). */ | |
3242 | static void | |
b7e2f6b3 EJ |
3243 | handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet, |
3244 | const struct flow *flow, bool fail_open) | |
9d6ac44e | 3245 | { |
b7e2f6b3 | 3246 | if (fail_open) { |
9d6ac44e BP |
3247 | /* |
3248 | * Extra-special case for fail-open mode. | |
3249 | * | |
3250 | * We are in fail-open mode and the packet matched the fail-open | |
3251 | * rule, but we are connected to a controller too. We should send | |
3252 | * the packet up to the controller in the hope that it will try to | |
3253 | * set up a flow and thereby allow us to exit fail-open. | |
3254 | * | |
3255 | * See the top-level comment in fail-open.c for more information. | |
3256 | */ | |
3257 | send_packet_in_miss(ofproto, packet, flow); | |
3258 | } | |
3259 | } | |
3260 | ||
3261 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
bcd2633a JP |
3262 | * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace |
3263 | * and (usually) installing a datapath flow. The answer is usually "yes" (a | |
3264 | * return value of true). However, for short flows the cost of bookkeeping is | |
3265 | * much higher than the benefits, so when the datapath holds a large number of | |
3266 | * flows we impose some heuristics to decide which flows are likely to be worth | |
3267 | * tracking. */ | |
9d6ac44e | 3268 | static bool |
bcd2633a | 3269 | flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc) |
9d6ac44e | 3270 | { |
04d08d54 | 3271 | struct dpif_backer *backer = miss->ofproto->backer; |
bcd2633a | 3272 | uint32_t hash; |
04d08d54 | 3273 | |
7155fa52 JS |
3274 | switch (flow_miss_model) { |
3275 | case OFPROTO_HANDLE_MISS_AUTO: | |
3276 | break; | |
3277 | case OFPROTO_HANDLE_MISS_WITH_FACETS: | |
3278 | return true; | |
3279 | case OFPROTO_HANDLE_MISS_WITHOUT_FACETS: | |
3280 | return false; | |
3281 | } | |
3282 | ||
04d08d54 | 3283 | if (!backer->governor) { |
9d6ac44e BP |
3284 | size_t n_subfacets; |
3285 | ||
04d08d54 | 3286 | n_subfacets = hmap_count(&backer->subfacets); |
380f49c4 | 3287 | if (n_subfacets * 2 <= flow_eviction_threshold) { |
9d6ac44e BP |
3288 | return true; |
3289 | } | |
3290 | ||
c985ec94 | 3291 | backer->governor = governor_create(); |
9d6ac44e BP |
3292 | } |
3293 | ||
bcd2633a | 3294 | hash = flow_hash_in_wildcards(&miss->flow, wc, 0); |
04d08d54 | 3295 | return governor_should_install_flow(backer->governor, hash, |
9d6ac44e BP |
3296 | list_size(&miss->packets)); |
3297 | } | |
3298 | ||
bcd2633a JP |
3299 | /* Handles 'miss' without creating a facet or subfacet or creating any datapath |
3300 | * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'. | |
3301 | * May add an "execute" operation to 'ops' and increment '*n_ops'. */ | |
9d6ac44e | 3302 | static void |
bcd2633a JP |
3303 | handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout, |
3304 | struct flow_miss *miss, | |
9d6ac44e BP |
3305 | struct flow_miss_op *ops, size_t *n_ops) |
3306 | { | |
530a1d91 | 3307 | struct ofpbuf *packet; |
2b459b83 | 3308 | |
9d6ac44e | 3309 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
abe529af | 3310 | |
9d6ac44e | 3311 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3312 | |
b7e2f6b3 EJ |
3313 | handle_flow_miss_common(miss->ofproto, packet, &miss->flow, |
3314 | rule->up.cr.priority == FAIL_OPEN_PRIORITY); | |
7820e253 | 3315 | |
bcd2633a JP |
3316 | if (xout->slow) { |
3317 | struct xlate_in xin; | |
abe529af | 3318 | |
47b0deae | 3319 | xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet); |
bcd2633a JP |
3320 | xlate_actions_for_side_effects(&xin); |
3321 | } | |
abe529af | 3322 | |
bcd2633a JP |
3323 | if (xout->odp_actions.size) { |
3324 | struct flow_miss_op *op = &ops[*n_ops]; | |
9d6ac44e BP |
3325 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
3326 | ||
3327 | init_flow_miss_execute_op(miss, packet, op); | |
bcd2633a | 3328 | xlate_out_copy(&op->xout, xout); |
bbafd73b EJ |
3329 | execute->actions = op->xout.odp_actions.data; |
3330 | execute->actions_len = op->xout.odp_actions.size; | |
3331 | op->xout_garbage = true; | |
9d6ac44e BP |
3332 | |
3333 | (*n_ops)++; | |
9d6ac44e | 3334 | } |
abe529af | 3335 | } |
9d6ac44e BP |
3336 | } |
3337 | ||
3338 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3339 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3340 | * | |
3341 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3342 | * This is really important only for new facets: if we just called time_msec() | |
3343 | * here, then the new subfacet or its packets could look (occasionally) as | |
3344 | * though it was used some time after the facet was used. That can make a | |
3345 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
bcd2633a JP |
3346 | * e.g. NetFlow statistics. |
3347 | * | |
3348 | * If non-null, 'stats' will be folded into 'facet'. */ | |
9d6ac44e BP |
3349 | static void |
3350 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
bcd2633a | 3351 | long long int now, struct dpif_flow_stats *stats, |
9d6ac44e BP |
3352 | struct flow_miss_op *ops, size_t *n_ops) |
3353 | { | |
6a7e895f | 3354 | enum subfacet_path want_path; |
9d6ac44e BP |
3355 | struct subfacet *subfacet; |
3356 | struct ofpbuf *packet; | |
abe529af | 3357 | |
bcd2633a | 3358 | want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
b0f7b9b5 | 3359 | |
530a1d91 | 3360 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3361 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 3362 | |
b7e2f6b3 EJ |
3363 | handle_flow_miss_common(miss->ofproto, packet, &miss->flow, |
3364 | facet->fail_open); | |
501f8d1f | 3365 | |
4dff9097 | 3366 | if (want_path != SF_FAST_PATH) { |
b7e2f6b3 | 3367 | struct rule_dpif *rule; |
bbafd73b | 3368 | struct xlate_in xin; |
454a77e5 | 3369 | |
b7e2f6b3 EJ |
3370 | rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); |
3371 | xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet); | |
bbafd73b | 3372 | xlate_actions_for_side_effects(&xin); |
501f8d1f | 3373 | } |
67d91f78 | 3374 | |
bbafd73b | 3375 | if (facet->xout.odp_actions.size) { |
9d6ac44e | 3376 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
8338659a | 3377 | |
9d6ac44e | 3378 | init_flow_miss_execute_op(miss, packet, op); |
bbafd73b EJ |
3379 | execute->actions = facet->xout.odp_actions.data, |
3380 | execute->actions_len = facet->xout.odp_actions.size; | |
9d6ac44e | 3381 | (*n_ops)++; |
5fe20d5d | 3382 | } |
501f8d1f BP |
3383 | } |
3384 | ||
a41d7bf2 JP |
3385 | /* Don't install the flow if it's the result of the "userspace" |
3386 | * action for an already installed facet. This can occur when a | |
3387 | * datapath flow with wildcards has a "userspace" action and flows | |
3388 | * sent to userspace result in a different subfacet, which will then | |
3389 | * be rejected as overlapping by the datapath. */ | |
3390 | if (miss->upcall_type == DPIF_UC_ACTION | |
3391 | && !list_is_empty(&facet->subfacets)) { | |
3392 | if (stats) { | |
3393 | facet->used = MAX(facet->used, stats->used); | |
3394 | facet->packet_count += stats->n_packets; | |
3395 | facet->byte_count += stats->n_bytes; | |
3396 | facet->tcp_flags |= stats->tcp_flags; | |
3397 | } | |
3398 | return; | |
3399 | } | |
3400 | ||
3401 | subfacet = subfacet_create(facet, miss, now); | |
3402 | if (stats) { | |
3403 | subfacet_update_stats(subfacet, stats); | |
3404 | } | |
3405 | ||
6a7e895f | 3406 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { |
501f8d1f | 3407 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3408 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3409 | |
c84451a6 EJ |
3410 | subfacet->path = want_path; |
3411 | ||
d445cc16 | 3412 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
6b35e630 JP |
3413 | if (enable_megaflows) { |
3414 | odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks, | |
3415 | &miss->flow, UINT32_MAX); | |
3416 | } | |
d445cc16 | 3417 | |
bbafd73b | 3418 | op->xout_garbage = false; |
c2b565b5 | 3419 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
9b1a48c2 | 3420 | op->subfacet = subfacet; |
0f74f625 | 3421 | put->flags = DPIF_FP_CREATE; |
501f8d1f BP |
3422 | put->key = miss->key; |
3423 | put->key_len = miss->key_len; | |
d445cc16 JP |
3424 | put->mask = op->mask.data; |
3425 | put->mask_len = op->mask.size; | |
3426 | ||
6a7e895f | 3427 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3428 | put->actions = facet->xout.odp_actions.data; |
3429 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3430 | } else { |
b7e2f6b3 | 3431 | compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow, |
bbafd73b | 3432 | op->slow_stub, sizeof op->slow_stub, |
6a7e895f BP |
3433 | &put->actions, &put->actions_len); |
3434 | } | |
501f8d1f BP |
3435 | put->stats = NULL; |
3436 | } | |
3437 | } | |
3438 | ||
acf60855 JP |
3439 | /* Handles flow miss 'miss'. May add any required datapath operations |
3440 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3441 | static void |
acf60855 JP |
3442 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3443 | size_t *n_ops) | |
9d6ac44e | 3444 | { |
acf60855 | 3445 | struct ofproto_dpif *ofproto = miss->ofproto; |
bcd2633a JP |
3446 | struct dpif_flow_stats stats__; |
3447 | struct dpif_flow_stats *stats = &stats__; | |
3448 | struct ofpbuf *packet; | |
9d6ac44e | 3449 | struct facet *facet; |
459b16a1 | 3450 | long long int now; |
9d6ac44e | 3451 | |
bcd2633a JP |
3452 | now = time_msec(); |
3453 | memset(stats, 0, sizeof *stats); | |
3454 | stats->used = now; | |
3455 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
3456 | stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow); | |
3457 | stats->n_bytes += packet->size; | |
3458 | stats->n_packets++; | |
3459 | } | |
9d6ac44e | 3460 | |
bcd2633a | 3461 | facet = facet_lookup_valid(ofproto, &miss->flow); |
9d6ac44e | 3462 | if (!facet) { |
bcd2633a JP |
3463 | struct flow_wildcards wc; |
3464 | struct rule_dpif *rule; | |
3465 | struct xlate_out xout; | |
3466 | struct xlate_in xin; | |
3467 | ||
3468 | flow_wildcards_init_catchall(&wc); | |
3469 | rule = rule_dpif_lookup(ofproto, &miss->flow, &wc); | |
3470 | rule_credit_stats(rule, stats); | |
3471 | ||
47b0deae EJ |
3472 | xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags, |
3473 | NULL); | |
bcd2633a JP |
3474 | xin.resubmit_stats = stats; |
3475 | xin.may_learn = true; | |
3476 | xlate_actions(&xin, &xout); | |
3477 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); | |
3478 | ||
ba33dd03 EJ |
3479 | /* There does not exist a bijection between 'struct flow' and datapath |
3480 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3481 | * assumption used throughout the facet and subfacet handling code. | |
3482 | * Since we have to handle these misses in userspace anyway, we simply | |
ec9f40dc | 3483 | * skip facet creation, avoiding the problem altogether. */ |
ba33dd03 | 3484 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE |
bcd2633a JP |
3485 | || !flow_miss_should_make_facet(miss, &xout.wc)) { |
3486 | handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops); | |
9d6ac44e BP |
3487 | return; |
3488 | } | |
3489 | ||
bcd2633a JP |
3490 | facet = facet_create(miss, rule, &xout, stats); |
3491 | stats = NULL; | |
9d6ac44e | 3492 | } |
bcd2633a | 3493 | handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops); |
9d6ac44e BP |
3494 | } |
3495 | ||
8f73d537 EJ |
3496 | static struct drop_key * |
3497 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3498 | size_t key_len) | |
3499 | { | |
3500 | struct drop_key *drop_key; | |
3501 | ||
3502 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3503 | &backer->drop_keys) { | |
3504 | if (drop_key->key_len == key_len | |
3505 | && !memcmp(drop_key->key, key, key_len)) { | |
3506 | return drop_key; | |
3507 | } | |
3508 | } | |
3509 | return NULL; | |
3510 | } | |
3511 | ||
3512 | static void | |
3513 | drop_key_clear(struct dpif_backer *backer) | |
3514 | { | |
3515 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3516 | struct drop_key *drop_key, *next; | |
3517 | ||
3518 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3519 | int error; | |
3520 | ||
3521 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3522 | NULL); | |
3523 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3524 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3525 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
10a89ef0 BP |
3526 | VLOG_WARN("Failed to delete drop key (%s) (%s)", |
3527 | ovs_strerror(error), ds_cstr(&ds)); | |
8f73d537 EJ |
3528 | ds_destroy(&ds); |
3529 | } | |
3530 | ||
3531 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3532 | free(drop_key->key); | |
3533 | free(drop_key); | |
3534 | } | |
3535 | } | |
3536 | ||
e09ee259 EJ |
3537 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3538 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3539 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3540 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3541 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3542 | * 'packet' ingressed. | |
e2a6ca36 | 3543 | * |
e09ee259 EJ |
3544 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3545 | * 'flow''s in_port to OFPP_NONE. | |
3546 | * | |
3547 | * This function does post-processing on data returned from | |
3548 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3549 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3550 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3551 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3552 | * a VLAN header onto 'packet' (if it is nonnull). | |
3553 | * | |
b9ad7294 EJ |
3554 | * Similarly, this function also includes some logic to help with tunnels. It |
3555 | * may modify 'flow' as necessary to make the tunneling implementation | |
3556 | * transparent to the upcall processing logic. | |
3557 | * | |
e09ee259 EJ |
3558 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3559 | * or some other positive errno if there are other problems. */ | |
3560 | static int | |
3561 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3562 | const struct nlattr *key, size_t key_len, | |
3563 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
4e022ec0 | 3564 | struct ofproto_dpif **ofproto, odp_port_t *odp_in_port) |
e84173dc | 3565 | { |
e09ee259 EJ |
3566 | const struct ofport_dpif *port; |
3567 | enum odp_key_fitness fitness; | |
b9ad7294 | 3568 | int error = ENODEV; |
e09ee259 EJ |
3569 | |
3570 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3571 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3572 | error = EINVAL; |
3573 | goto exit; | |
3574 | } | |
3575 | ||
e09ee259 | 3576 | if (odp_in_port) { |
4e022ec0 | 3577 | *odp_in_port = flow->in_port.odp_port; |
e09ee259 EJ |
3578 | } |
3579 | ||
3abe1835 | 3580 | port = (tnl_port_should_receive(flow) |
e1939dbf | 3581 | ? tnl_port_receive(flow) |
4e022ec0 AW |
3582 | : odp_port_to_ofport(backer, flow->in_port.odp_port)); |
3583 | flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE; | |
3abe1835 BP |
3584 | if (!port) { |
3585 | goto exit; | |
3586 | } | |
3587 | ||
3588 | /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port | |
3589 | * it's theoretically possible that we'll receive an ofport belonging to an | |
3590 | * entirely different datapath. In practice, this can't happen because no | |
3591 | * platforms has two separate datapaths which each support tunneling. */ | |
3592 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3593 | ||
3594 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3595 | if (packet) { | |
3596 | /* Make the packet resemble the flow, so that it gets sent to | |
3597 | * an OpenFlow controller properly, so that it looks correct | |
3598 | * for sFlow, and so that flow_extract() will get the correct | |
3599 | * vlan_tci if it is called on 'packet'. | |
3600 | * | |
3601 | * The allocated space inside 'packet' probably also contains | |
3602 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3603 | * a struct dpif_upcall (see the large comment on that | |
3604 | * structure definition), so pushing data on 'packet' is in | |
3605 | * general not a good idea since it could overwrite 'key' or | |
3606 | * free it as a side effect. However, it's OK in this special | |
3607 | * case because we know that 'packet' is inside a Netlink | |
3608 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3609 | * "struct nlattr", which is fine since we don't need that | |
3610 | * header anymore. */ | |
3611 | eth_push_vlan(packet, flow->vlan_tci); | |
3612 | } | |
3613 | /* We can't reproduce 'key' from 'flow'. */ | |
3614 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 | 3615 | } |
e09ee259 | 3616 | error = 0; |
52a90c29 | 3617 | |
b9ad7294 EJ |
3618 | if (ofproto) { |
3619 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3620 | } | |
3621 | ||
e09ee259 EJ |
3622 | exit: |
3623 | if (fitnessp) { | |
3624 | *fitnessp = fitness; | |
3625 | } | |
3626 | return error; | |
e84173dc BP |
3627 | } |
3628 | ||
501f8d1f | 3629 | static void |
acf60855 | 3630 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3631 | size_t n_upcalls) |
3632 | { | |
3633 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3634 | struct flow_miss *miss; |
3635 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3636 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3637 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3638 | struct hmap todo; |
b23cdad9 | 3639 | int n_misses; |
501f8d1f BP |
3640 | size_t n_ops; |
3641 | size_t i; | |
3642 | ||
3643 | if (!n_upcalls) { | |
3644 | return; | |
3645 | } | |
3646 | ||
3647 | /* Construct the to-do list. | |
3648 | * | |
3649 | * This just amounts to extracting the flow from each packet and sticking | |
3650 | * the packets that have the same flow in the same "flow_miss" structure so | |
3651 | * that we can process them together. */ | |
3652 | hmap_init(&todo); | |
b23cdad9 | 3653 | n_misses = 0; |
501f8d1f | 3654 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3655 | struct flow_miss *miss = &misses[n_misses]; |
3656 | struct flow_miss *existing_miss; | |
acf60855 | 3657 | struct ofproto_dpif *ofproto; |
4e022ec0 | 3658 | odp_port_t odp_in_port; |
1d446463 | 3659 | struct flow flow; |
b23cdad9 | 3660 | uint32_t hash; |
e09ee259 | 3661 | int error; |
501f8d1f | 3662 | |
e09ee259 EJ |
3663 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
3664 | upcall->key_len, &flow, &miss->key_fitness, | |
47b0deae | 3665 | &ofproto, &odp_in_port); |
e09ee259 | 3666 | if (error == ENODEV) { |
8f73d537 EJ |
3667 | struct drop_key *drop_key; |
3668 | ||
c3385516 AW |
3669 | /* Received packet on datapath port for which we couldn't |
3670 | * associate an ofproto. This can happen if a port is removed | |
3671 | * while traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
3672 | * in case it happens frequently. Install a drop flow so |
3673 | * that future packets of the flow are inexpensively dropped | |
3674 | * in the kernel. */ | |
c3385516 AW |
3675 | VLOG_INFO_RL(&rl, "received packet on unassociated datapath port " |
3676 | "%"PRIu32, odp_in_port); | |
8f73d537 EJ |
3677 | |
3678 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
3679 | if (!drop_key) { | |
92e02bc6 AZ |
3680 | int ret; |
3681 | ret = dpif_flow_put(backer->dpif, | |
3682 | DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3683 | upcall->key, upcall->key_len, | |
3684 | NULL, 0, NULL, 0, NULL); | |
3685 | ||
3686 | if (!ret) { | |
3687 | drop_key = xmalloc(sizeof *drop_key); | |
3688 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
3689 | drop_key->key_len = upcall->key_len; | |
3690 | ||
3691 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3692 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3693 | } | |
8f73d537 EJ |
3694 | } |
3695 | continue; | |
acf60855 | 3696 | } |
e09ee259 | 3697 | if (error) { |
b0f7b9b5 BP |
3698 | continue; |
3699 | } | |
735d7efb AZ |
3700 | |
3701 | ofproto->n_missed++; | |
72e8bf28 | 3702 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
4e022ec0 | 3703 | &flow.tunnel, &flow.in_port, &miss->flow); |
501f8d1f | 3704 | |
501f8d1f | 3705 | /* Add other packets to a to-do list. */ |
b23cdad9 | 3706 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 3707 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
3708 | if (!existing_miss) { |
3709 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 3710 | miss->ofproto = ofproto; |
b23cdad9 BP |
3711 | miss->key = upcall->key; |
3712 | miss->key_len = upcall->key_len; | |
6a7e895f | 3713 | miss->upcall_type = upcall->type; |
b23cdad9 BP |
3714 | list_init(&miss->packets); |
3715 | ||
3716 | n_misses++; | |
3717 | } else { | |
3718 | miss = existing_miss; | |
3719 | } | |
501f8d1f BP |
3720 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3721 | } | |
3722 | ||
3723 | /* Process each element in the to-do list, constructing the set of | |
3724 | * operations to batch. */ | |
3725 | n_ops = 0; | |
33bb0caa | 3726 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 3727 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3728 | } |
cb22974d | 3729 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3730 | |
3731 | /* Execute batch. */ | |
3732 | for (i = 0; i < n_ops; i++) { | |
3733 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3734 | } | |
acf60855 | 3735 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 3736 | |
501f8d1f | 3737 | for (i = 0; i < n_ops; i++) { |
9b1a48c2 JP |
3738 | if (dpif_ops[i]->error != 0 |
3739 | && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT | |
3740 | && flow_miss_ops[i].subfacet) { | |
3741 | struct subfacet *subfacet = flow_miss_ops[i].subfacet; | |
3742 | ||
3743 | COVERAGE_INC(subfacet_install_fail); | |
3744 | ||
cfd5c9eb JP |
3745 | /* Zero-out subfacet counters when installation failed, but |
3746 | * datapath reported hits. This should not happen and | |
3747 | * indicates a bug, since if the datapath flow exists, we | |
3748 | * should not be attempting to create a new subfacet. A | |
3749 | * buggy datapath could trigger this, so just zero out the | |
3750 | * counters and log an error. */ | |
3751 | if (subfacet->dp_packet_count || subfacet->dp_byte_count) { | |
3752 | VLOG_ERR_RL(&rl, "failed to install subfacet for which " | |
3753 | "datapath reported hits"); | |
3754 | subfacet->dp_packet_count = subfacet->dp_byte_count = 0; | |
3755 | } | |
3756 | ||
9b1a48c2 JP |
3757 | subfacet->path = SF_NOT_INSTALLED; |
3758 | } | |
3759 | ||
3760 | /* Free memory. */ | |
bbafd73b EJ |
3761 | if (flow_miss_ops[i].xout_garbage) { |
3762 | xlate_out_uninit(&flow_miss_ops[i].xout); | |
3763 | } | |
501f8d1f | 3764 | } |
33bb0caa | 3765 | hmap_destroy(&todo); |
abe529af BP |
3766 | } |
3767 | ||
29089a54 RL |
3768 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, |
3769 | IPFIX_UPCALL } | |
6a7e895f BP |
3770 | classify_upcall(const struct dpif_upcall *upcall) |
3771 | { | |
29089a54 | 3772 | size_t userdata_len; |
6a7e895f BP |
3773 | union user_action_cookie cookie; |
3774 | ||
3775 | /* First look at the upcall type. */ | |
3776 | switch (upcall->type) { | |
3777 | case DPIF_UC_ACTION: | |
3778 | break; | |
3779 | ||
3780 | case DPIF_UC_MISS: | |
3781 | return MISS_UPCALL; | |
3782 | ||
3783 | case DPIF_N_UC_TYPES: | |
3784 | default: | |
3785 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3786 | return BAD_UPCALL; | |
3787 | } | |
3788 | ||
3789 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
3790 | if (!upcall->userdata) { |
3791 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
3792 | return BAD_UPCALL; | |
3793 | } | |
29089a54 RL |
3794 | userdata_len = nl_attr_get_size(upcall->userdata); |
3795 | if (userdata_len < sizeof cookie.type | |
3796 | || userdata_len > sizeof cookie) { | |
e995e3df | 3797 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", |
29089a54 | 3798 | userdata_len); |
e995e3df BP |
3799 | return BAD_UPCALL; |
3800 | } | |
29089a54 RL |
3801 | memset(&cookie, 0, sizeof cookie); |
3802 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
3803 | if (userdata_len == sizeof cookie.sflow | |
3804 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
6a7e895f | 3805 | return SFLOW_UPCALL; |
29089a54 RL |
3806 | } else if (userdata_len == sizeof cookie.slow_path |
3807 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
6a7e895f | 3808 | return MISS_UPCALL; |
29089a54 RL |
3809 | } else if (userdata_len == sizeof cookie.flow_sample |
3810 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
3811 | return FLOW_SAMPLE_UPCALL; | |
3812 | } else if (userdata_len == sizeof cookie.ipfix | |
3813 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
3814 | return IPFIX_UPCALL; | |
3815 | } else { | |
3816 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
3817 | " and size %zu", cookie.type, userdata_len); | |
6a7e895f BP |
3818 | return BAD_UPCALL; |
3819 | } | |
3820 | } | |
3821 | ||
abe529af | 3822 | static void |
acf60855 | 3823 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 3824 | const struct dpif_upcall *upcall) |
abe529af | 3825 | { |
acf60855 | 3826 | struct ofproto_dpif *ofproto; |
1673e0e4 | 3827 | union user_action_cookie cookie; |
e84173dc | 3828 | struct flow flow; |
4e022ec0 | 3829 | odp_port_t odp_in_port; |
abe529af | 3830 | |
e09ee259 | 3831 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
47b0deae | 3832 | &flow, NULL, &ofproto, &odp_in_port) |
e09ee259 | 3833 | || !ofproto->sflow) { |
e84173dc BP |
3834 | return; |
3835 | } | |
3836 | ||
29089a54 RL |
3837 | memset(&cookie, 0, sizeof cookie); |
3838 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
e1b1d06a JP |
3839 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
3840 | odp_in_port, &cookie); | |
6ff686f2 PS |
3841 | } |
3842 | ||
29089a54 RL |
3843 | static void |
3844 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
3845 | const struct dpif_upcall *upcall) | |
3846 | { | |
3847 | struct ofproto_dpif *ofproto; | |
3848 | union user_action_cookie cookie; | |
3849 | struct flow flow; | |
3850 | ||
3851 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
47b0deae | 3852 | &flow, NULL, &ofproto, NULL) |
29089a54 RL |
3853 | || !ofproto->ipfix) { |
3854 | return; | |
3855 | } | |
3856 | ||
3857 | memset(&cookie, 0, sizeof cookie); | |
3858 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
3859 | ||
3860 | /* The flow reflects exactly the contents of the packet. Sample | |
3861 | * the packet using it. */ | |
3862 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
3863 | cookie.flow_sample.collector_set_id, | |
3864 | cookie.flow_sample.probability, | |
3865 | cookie.flow_sample.obs_domain_id, | |
3866 | cookie.flow_sample.obs_point_id); | |
3867 | } | |
3868 | ||
3869 | static void | |
3870 | handle_ipfix_upcall(struct dpif_backer *backer, | |
3871 | const struct dpif_upcall *upcall) | |
3872 | { | |
3873 | struct ofproto_dpif *ofproto; | |
3874 | struct flow flow; | |
3875 | ||
3876 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
47b0deae | 3877 | &flow, NULL, &ofproto, NULL) |
29089a54 RL |
3878 | || !ofproto->ipfix) { |
3879 | return; | |
3880 | } | |
3881 | ||
3882 | /* The flow reflects exactly the contents of the packet. Sample | |
3883 | * the packet using it. */ | |
3884 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
3885 | } | |
3886 | ||
9b16c439 | 3887 | static int |
acf60855 | 3888 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 3889 | { |
9b16c439 | 3890 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3891 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3892 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3893 | int n_processed; | |
9b16c439 BP |
3894 | int n_misses; |
3895 | int i; | |
abe529af | 3896 | |
cb22974d | 3897 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3898 | |
9b16c439 | 3899 | n_misses = 0; |
90a7c55e | 3900 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3901 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3902 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3903 | int error; |
3904 | ||
90a7c55e BP |
3905 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3906 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 3907 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 3908 | if (error) { |
90a7c55e | 3909 | ofpbuf_uninit(buf); |
9b16c439 BP |
3910 | break; |
3911 | } | |
3912 | ||
6a7e895f BP |
3913 | switch (classify_upcall(upcall)) { |
3914 | case MISS_UPCALL: | |
9b16c439 BP |
3915 | /* Handle it later. */ |
3916 | n_misses++; | |
3917 | break; | |
3918 | ||
6a7e895f | 3919 | case SFLOW_UPCALL: |
acf60855 | 3920 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
3921 | ofpbuf_uninit(buf); |
3922 | break; | |
3923 | ||
29089a54 RL |
3924 | case FLOW_SAMPLE_UPCALL: |
3925 | handle_flow_sample_upcall(backer, upcall); | |
3926 | ofpbuf_uninit(buf); | |
3927 | break; | |
3928 | ||
3929 | case IPFIX_UPCALL: | |
3930 | handle_ipfix_upcall(backer, upcall); | |
3931 | ofpbuf_uninit(buf); | |
3932 | break; | |
3933 | ||
6a7e895f BP |
3934 | case BAD_UPCALL: |
3935 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3936 | break; |
3937 | } | |
abe529af | 3938 | } |
9b16c439 | 3939 | |
6a7e895f | 3940 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 3941 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
3942 | for (i = 0; i < n_misses; i++) { |
3943 | ofpbuf_uninit(&miss_bufs[i]); | |
3944 | } | |
9b16c439 | 3945 | |
90a7c55e | 3946 | return n_processed; |
abe529af BP |
3947 | } |
3948 | \f | |
3949 | /* Flow expiration. */ | |
3950 | ||
04d08d54 | 3951 | static int subfacet_max_idle(const struct dpif_backer *); |
acf60855 | 3952 | static void update_stats(struct dpif_backer *); |
abe529af | 3953 | static void rule_expire(struct rule_dpif *); |
04d08d54 | 3954 | static void expire_subfacets(struct dpif_backer *, int dp_max_idle); |
abe529af BP |
3955 | |
3956 | /* This function is called periodically by run(). Its job is to collect | |
3957 | * updates for the flows that have been installed into the datapath, most | |
3958 | * importantly when they last were used, and then use that information to | |
3959 | * expire flows that have not been used recently. | |
3960 | * | |
3961 | * Returns the number of milliseconds after which it should be called again. */ | |
3962 | static int | |
acf60855 | 3963 | expire(struct dpif_backer *backer) |
abe529af | 3964 | { |
acf60855 | 3965 | struct ofproto_dpif *ofproto; |
dc54ef36 | 3966 | size_t n_subfacets; |
04d08d54 | 3967 | int max_idle; |
abe529af | 3968 | |
8f73d537 EJ |
3969 | /* Periodically clear out the drop keys in an effort to keep them |
3970 | * relatively few. */ | |
3971 | drop_key_clear(backer); | |
3972 | ||
acf60855 JP |
3973 | /* Update stats for each flow in the backer. */ |
3974 | update_stats(backer); | |
abe529af | 3975 | |
04d08d54 EJ |
3976 | n_subfacets = hmap_count(&backer->subfacets); |
3977 | if (n_subfacets) { | |
9fc0165a | 3978 | struct subfacet *subfacet; |
04d08d54 | 3979 | long long int total, now; |
abe529af | 3980 | |
04d08d54 EJ |
3981 | total = 0; |
3982 | now = time_msec(); | |
3983 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { | |
3984 | total += now - subfacet->created; | |
acf60855 | 3985 | } |
04d08d54 EJ |
3986 | backer->avg_subfacet_life += total / n_subfacets; |
3987 | } | |
3988 | backer->avg_subfacet_life /= 2; | |
0697b5c3 | 3989 | |
04d08d54 EJ |
3990 | backer->avg_n_subfacet += n_subfacets; |
3991 | backer->avg_n_subfacet /= 2; | |
3992 | ||
3993 | backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); | |
655ab909 | 3994 | |
04d08d54 EJ |
3995 | max_idle = subfacet_max_idle(backer); |
3996 | expire_subfacets(backer, max_idle); | |
acf60855 | 3997 | |
04d08d54 EJ |
3998 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
3999 | struct rule *rule, *next_rule; | |
4000 | ||
4001 | if (ofproto->backer != backer) { | |
4002 | continue; | |
4003 | } | |
acf60855 JP |
4004 | |
4005 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4006 | * has passed. */ | |
e503cc19 SH |
4007 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4008 | &ofproto->up.expirable) { | |
4009 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4010 | } |
abe529af | 4011 | |
acf60855 JP |
4012 | /* All outstanding data in existing flows has been accounted, so it's a |
4013 | * good time to do bond rebalancing. */ | |
4014 | if (ofproto->has_bonded_bundles) { | |
4015 | struct ofbundle *bundle; | |
abe529af | 4016 | |
acf60855 JP |
4017 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4018 | if (bundle->bond) { | |
4a1b8f30 | 4019 | bond_rebalance(bundle->bond); |
acf60855 | 4020 | } |
abe529af BP |
4021 | } |
4022 | } | |
4023 | } | |
4024 | ||
acf60855 | 4025 | return MIN(max_idle, 1000); |
abe529af BP |
4026 | } |
4027 | ||
a218c879 BP |
4028 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4029 | * as 'subfacet''s statistics. */ | |
4030 | static void | |
4031 | update_subfacet_stats(struct subfacet *subfacet, | |
4032 | const struct dpif_flow_stats *stats) | |
4033 | { | |
4034 | struct facet *facet = subfacet->facet; | |
9dfb1f78 EJ |
4035 | struct dpif_flow_stats diff; |
4036 | ||
4037 | diff.tcp_flags = stats->tcp_flags; | |
4038 | diff.used = stats->used; | |
a218c879 BP |
4039 | |
4040 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
9dfb1f78 | 4041 | diff.n_packets = stats->n_packets - subfacet->dp_packet_count; |
a218c879 BP |
4042 | } else { |
4043 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
9dfb1f78 | 4044 | diff.n_packets = 0; |
a218c879 BP |
4045 | } |
4046 | ||
4047 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
9dfb1f78 | 4048 | diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; |
a218c879 BP |
4049 | } else { |
4050 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
9dfb1f78 | 4051 | diff.n_bytes = 0; |
a218c879 BP |
4052 | } |
4053 | ||
b7e2f6b3 | 4054 | facet->ofproto->n_hit += diff.n_packets; |
a218c879 BP |
4055 | subfacet->dp_packet_count = stats->n_packets; |
4056 | subfacet->dp_byte_count = stats->n_bytes; | |
9dfb1f78 | 4057 | subfacet_update_stats(subfacet, &diff); |
a218c879 | 4058 | |
a218c879 BP |
4059 | if (facet->accounted_bytes < facet->byte_count) { |
4060 | facet_learn(facet); | |
4061 | facet_account(facet); | |
4062 | facet->accounted_bytes = facet->byte_count; | |
4063 | } | |
a218c879 BP |
4064 | } |
4065 | ||
4066 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4067 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4068 | static void | |
04d08d54 | 4069 | delete_unexpected_flow(struct dpif_backer *backer, |
a218c879 BP |
4070 | const struct nlattr *key, size_t key_len) |
4071 | { | |
4072 | if (!VLOG_DROP_WARN(&rl)) { | |
4073 | struct ds s; | |
4074 | ||
4075 | ds_init(&s); | |
4076 | odp_flow_key_format(key, key_len, &s); | |
04d08d54 | 4077 | VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); |
a218c879 BP |
4078 | ds_destroy(&s); |
4079 | } | |
4080 | ||
4081 | COVERAGE_INC(facet_unexpected); | |
04d08d54 | 4082 | dpif_flow_del(backer->dpif, key, key_len, NULL); |
a218c879 BP |
4083 | } |
4084 | ||
abe529af BP |
4085 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4086 | * | |
4087 | * This function also pushes statistics updates to rules which each facet | |
4088 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4089 | * facet changes the rule it resubmits into at some time in between | |
4090 | * update_stats() runs, it is possible that statistics accrued to the | |
4091 | * old rule will be incorrectly attributed to the new rule. This could be | |
4092 | * avoided by calling update_stats() whenever rules are created or | |
4093 | * deleted. However, the performance impact of making so many calls to the | |
4094 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4095 | * |
4096 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4097 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4098 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4099 | * the hit or miss counts of br1. | |
abe529af BP |
4100 | */ |
4101 | static void | |
acf60855 | 4102 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4103 | { |
4104 | const struct dpif_flow_stats *stats; | |
4105 | struct dpif_flow_dump dump; | |
61fb711d JP |
4106 | const struct nlattr *key, *mask; |
4107 | size_t key_len, mask_len; | |
abe529af | 4108 | |
acf60855 | 4109 | dpif_flow_dump_start(&dump, backer->dpif); |
e6cc0bab | 4110 | while (dpif_flow_dump_next(&dump, &key, &key_len, |
61fb711d | 4111 | &mask, &mask_len, NULL, NULL, &stats)) { |
b0f7b9b5 | 4112 | struct subfacet *subfacet; |
acf60855 | 4113 | uint32_t key_hash; |
abe529af | 4114 | |
acf60855 | 4115 | key_hash = odp_flow_key_hash(key, key_len); |
04d08d54 | 4116 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
6a7e895f BP |
4117 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4118 | case SF_FAST_PATH: | |
a218c879 | 4119 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4120 | break; |
4121 | ||
4122 | case SF_SLOW_PATH: | |
4123 | /* Stats are updated per-packet. */ | |
4124 | break; | |
4125 | ||
4126 | case SF_NOT_INSTALLED: | |
4127 | default: | |
04d08d54 | 4128 | delete_unexpected_flow(backer, key, key_len); |
6a7e895f | 4129 | break; |
abe529af | 4130 | } |
8fa4d1d0 | 4131 | run_fast_rl(); |
abe529af BP |
4132 | } |
4133 | dpif_flow_dump_done(&dump); | |
3af56aef | 4134 | |
dc54ef36 | 4135 | update_moving_averages(backer); |
abe529af BP |
4136 | } |
4137 | ||
4138 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4139 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4140 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4141 | * fold its statistic into its rule. */ | |
abe529af | 4142 | static int |
04d08d54 | 4143 | subfacet_max_idle(const struct dpif_backer *backer) |
abe529af BP |
4144 | { |
4145 | /* | |
4146 | * Idle time histogram. | |
4147 | * | |
b0f7b9b5 BP |
4148 | * Most of the time a switch has a relatively small number of subfacets. |
4149 | * When this is the case we might as well keep statistics for all of them | |
4150 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4151 | * well. |
4152 | * | |
b0f7b9b5 | 4153 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4154 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4155 | * significant. However, with a large number of subfacets it is likely |
4156 | * that only a few of them are "heavy hitters" that consume a large amount | |
4157 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4158 | * the kernel and maintaining in userspaces; other subfacets we can | |
4159 | * discard. | |
abe529af BP |
4160 | * |
4161 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4162 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4163 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4164 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4165 | * the most-recently-used 1% of subfacets (but at least |
380f49c4 | 4166 | * flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4167 | * the most-recently-used bucket of subfacets is kept, so actually an |
4168 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4169 | * (though the next run will delete most of those unless they receive |
4170 | * additional data). | |
abe529af | 4171 | * |
b0f7b9b5 BP |
4172 | * This requires a second pass through the subfacets, in addition to the |
4173 | * pass made by update_stats(), because the former function never looks at | |
4174 | * uninstallable subfacets. | |
abe529af BP |
4175 | */ |
4176 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4177 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4178 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4179 | int total, subtotal, bucket; |
b0f7b9b5 | 4180 | struct subfacet *subfacet; |
abe529af BP |
4181 | long long int now; |
4182 | int i; | |
4183 | ||
04d08d54 | 4184 | total = hmap_count(&backer->subfacets); |
380f49c4 | 4185 | if (total <= flow_eviction_threshold) { |
abe529af BP |
4186 | return N_BUCKETS * BUCKET_WIDTH; |
4187 | } | |
4188 | ||
4189 | /* Build histogram. */ | |
4190 | now = time_msec(); | |
04d08d54 | 4191 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { |
b0f7b9b5 | 4192 | long long int idle = now - subfacet->used; |
abe529af BP |
4193 | int bucket = (idle <= 0 ? 0 |
4194 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4195 | : (unsigned int) idle / BUCKET_WIDTH); | |
4196 | buckets[bucket]++; | |
4197 | } | |
4198 | ||
4199 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4200 | subtotal = bucket = 0; |
4201 | do { | |
4202 | subtotal += buckets[bucket++]; | |
084f5290 | 4203 | } while (bucket < N_BUCKETS && |
380f49c4 | 4204 | subtotal < MAX(flow_eviction_threshold, total / 100)); |
abe529af BP |
4205 | |
4206 | if (VLOG_IS_DBG_ENABLED()) { | |
4207 | struct ds s; | |
4208 | ||
4209 | ds_init(&s); | |
4210 | ds_put_cstr(&s, "keep"); | |
4211 | for (i = 0; i < N_BUCKETS; i++) { | |
4212 | if (i == bucket) { | |
4213 | ds_put_cstr(&s, ", drop"); | |
4214 | } | |
4215 | if (buckets[i]) { | |
4216 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4217 | } | |
4218 | } | |
04d08d54 | 4219 | VLOG_INFO("%s (msec:count)", ds_cstr(&s)); |
abe529af BP |
4220 | ds_destroy(&s); |
4221 | } | |
4222 | ||
4223 | return bucket * BUCKET_WIDTH; | |
4224 | } | |
4225 | ||
abe529af | 4226 | static void |
04d08d54 | 4227 | expire_subfacets(struct dpif_backer *backer, int dp_max_idle) |
abe529af | 4228 | { |
625b0720 BP |
4229 | /* Cutoff time for most flows. */ |
4230 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4231 | ||
4232 | /* We really want to keep flows for special protocols around, so use a more | |
4233 | * conservative cutoff. */ | |
4234 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4235 | |
b0f7b9b5 | 4236 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4237 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4238 | int n_batch; |
abe529af | 4239 | |
b99d3cee | 4240 | n_batch = 0; |
b0f7b9b5 | 4241 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
04d08d54 | 4242 | &backer->subfacets) { |
625b0720 BP |
4243 | long long int cutoff; |
4244 | ||
bbafd73b EJ |
4245 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
4246 | | SLOW_STP) | |
625b0720 BP |
4247 | ? special_cutoff |
4248 | : normal_cutoff); | |
b0f7b9b5 | 4249 | if (subfacet->used < cutoff) { |
6a7e895f | 4250 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4251 | batch[n_batch++] = subfacet; |
1d85f9e5 | 4252 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
04d08d54 | 4253 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee BP |
4254 | n_batch = 0; |
4255 | } | |
4256 | } else { | |
4257 | subfacet_destroy(subfacet); | |
4258 | } | |
abe529af BP |
4259 | } |
4260 | } | |
b99d3cee BP |
4261 | |
4262 | if (n_batch > 0) { | |
04d08d54 | 4263 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee | 4264 | } |
abe529af BP |
4265 | } |
4266 | ||
4267 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4268 | * then delete it entirely. */ | |
4269 | static void | |
4270 | rule_expire(struct rule_dpif *rule) | |
4271 | { | |
abe529af BP |
4272 | long long int now; |
4273 | uint8_t reason; | |
4274 | ||
e2a3d183 BP |
4275 | if (rule->up.pending) { |
4276 | /* We'll have to expire it later. */ | |
4277 | return; | |
4278 | } | |
4279 | ||
abe529af BP |
4280 | /* Has 'rule' expired? */ |
4281 | now = time_msec(); | |
4282 | if (rule->up.hard_timeout | |
308881af | 4283 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4284 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4285 | } else if (rule->up.idle_timeout |
1745cd08 | 4286 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4287 | reason = OFPRR_IDLE_TIMEOUT; |
4288 | } else { | |
4289 | return; | |
4290 | } | |
4291 | ||
4292 | COVERAGE_INC(ofproto_dpif_expired); | |
4293 | ||
abe529af BP |
4294 | /* Get rid of the rule. */ |
4295 | ofproto_rule_expire(&rule->up, reason); | |
4296 | } | |
4297 | \f | |
4298 | /* Facets. */ | |
4299 | ||
4dff9097 | 4300 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
4301 | * |
4302 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 4303 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 4304 | * |
bcd2633a JP |
4305 | * 'rule' and 'xout' must have been created based on 'miss'. |
4306 | * | |
4307 | * 'facet'' statistics are initialized based on 'stats'. | |
2b459b83 | 4308 | * |
b0f7b9b5 BP |
4309 | * The facet will initially have no subfacets. The caller should create (at |
4310 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4311 | static struct facet * |
bcd2633a JP |
4312 | facet_create(const struct flow_miss *miss, struct rule_dpif *rule, |
4313 | struct xlate_out *xout, struct dpif_flow_stats *stats) | |
abe529af | 4314 | { |
4dff9097 | 4315 | struct ofproto_dpif *ofproto = miss->ofproto; |
abe529af | 4316 | struct facet *facet; |
bcd2633a | 4317 | struct match match; |
abe529af BP |
4318 | |
4319 | facet = xzalloc(sizeof *facet); | |
b7e2f6b3 | 4320 | facet->ofproto = miss->ofproto; |
bcd2633a JP |
4321 | facet->packet_count = facet->prev_packet_count = stats->n_packets; |
4322 | facet->byte_count = facet->prev_byte_count = stats->n_bytes; | |
4323 | facet->tcp_flags = stats->tcp_flags; | |
4324 | facet->used = stats->used; | |
4dff9097 | 4325 | facet->flow = miss->flow; |
4dff9097 EJ |
4326 | facet->learn_rl = time_msec() + 500; |
4327 | ||
b0f7b9b5 | 4328 | list_init(&facet->subfacets); |
abe529af BP |
4329 | netflow_flow_init(&facet->nf_flow); |
4330 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4331 | ||
bcd2633a JP |
4332 | xlate_out_copy(&facet->xout, xout); |
4333 | ||
4334 | match_init(&match, &facet->flow, &facet->xout.wc); | |
4335 | cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); | |
4336 | classifier_insert(&ofproto->facets, &facet->cr); | |
4337 | ||
bbafd73b | 4338 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; |
b7e2f6b3 | 4339 | facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; |
6cf474d7 | 4340 | |
abe529af BP |
4341 | return facet; |
4342 | } | |
4343 | ||
4344 | static void | |
4345 | facet_free(struct facet *facet) | |
4346 | { | |
4dff9097 | 4347 | if (facet) { |
bbafd73b | 4348 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
4349 | free(facet); |
4350 | } | |
abe529af BP |
4351 | } |
4352 | ||
3d9e05f8 | 4353 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4354 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4355 | static bool |
4356 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4357 | const struct nlattr *odp_actions, size_t actions_len, | |
4358 | struct ofpbuf *packet) | |
4359 | { | |
4360 | struct odputil_keybuf keybuf; | |
4361 | struct ofpbuf key; | |
4362 | int error; | |
4363 | ||
6ff686f2 | 4364 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a | 4365 | odp_flow_key_from_flow(&key, flow, |
4e022ec0 | 4366 | ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port)); |
80e5eed9 | 4367 | |
acf60855 | 4368 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4369 | odp_actions, actions_len, packet); |
6ff686f2 | 4370 | return !error; |
abe529af BP |
4371 | } |
4372 | ||
bcd2633a | 4373 | /* Remove 'facet' from its ofproto and free up the associated memory: |
abe529af BP |
4374 | * |
4375 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4376 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4377 | * |
4378 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4379 | */ | |
4380 | static void | |
15baa734 | 4381 | facet_remove(struct facet *facet) |
abe529af | 4382 | { |
b0f7b9b5 BP |
4383 | struct subfacet *subfacet, *next_subfacet; |
4384 | ||
cb22974d | 4385 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4386 | |
4387 | /* First uninstall all of the subfacets to get final statistics. */ | |
4388 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4389 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4390 | } |
4391 | ||
4392 | /* Flush the final stats to the rule. | |
4393 | * | |
4394 | * This might require us to have at least one subfacet around so that we | |
4395 | * can use its actions for accounting in facet_account(), which is why we | |
4396 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4397 | facet_flush_stats(facet); |
551a2f6c BP |
4398 | |
4399 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4400 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4401 | &facet->subfacets) { | |
15baa734 | 4402 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4403 | } |
b7e2f6b3 | 4404 | classifier_remove(&facet->ofproto->facets, &facet->cr); |
bcd2633a | 4405 | cls_rule_destroy(&facet->cr); |
abe529af BP |
4406 | facet_free(facet); |
4407 | } | |
4408 | ||
3de9590b BP |
4409 | /* Feed information from 'facet' back into the learning table to keep it in |
4410 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4411 | static void |
3de9590b | 4412 | facet_learn(struct facet *facet) |
abe529af | 4413 | { |
62bd7349 | 4414 | long long int now = time_msec(); |
abe529af | 4415 | |
bbafd73b | 4416 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
4417 | return; |
4418 | } | |
4419 | ||
62bd7349 | 4420 | facet->learn_rl = now + 500; |
6cf474d7 | 4421 | |
bbafd73b EJ |
4422 | if (!facet->xout.has_learn |
4423 | && !facet->xout.has_normal | |
4424 | && (!facet->xout.has_fin_timeout | |
3de9590b | 4425 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
4426 | return; |
4427 | } | |
abe529af | 4428 | |
9dfb1f78 | 4429 | facet_push_stats(facet, true); |
3de9590b BP |
4430 | } |
4431 | ||
4432 | static void | |
4433 | facet_account(struct facet *facet) | |
4434 | { | |
3de9590b BP |
4435 | const struct nlattr *a; |
4436 | unsigned int left; | |
4437 | ovs_be16 vlan_tci; | |
4438 | uint64_t n_bytes; | |
abe529af | 4439 | |
b7e2f6b3 | 4440 | if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { |
abe529af BP |
4441 | return; |
4442 | } | |
3de9590b | 4443 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4444 | |
4445 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4446 | * as a basis. We also need to track the actual VLAN on which the packet | |
4447 | * is going to be sent to ensure that it matches the one passed to | |
4448 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4449 | * hash bucket.) |
4450 | * | |
4451 | * We use the actions from an arbitrary subfacet because they should all | |
4452 | * be equally valid for our purpose. */ | |
d78be13b | 4453 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
4454 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
4455 | facet->xout.odp_actions.size) { | |
fea393b1 | 4456 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4457 | struct ofport_dpif *port; |
abe529af | 4458 | |
d78be13b | 4459 | switch (nl_attr_type(a)) { |
df2c07f4 | 4460 | case OVS_ACTION_ATTR_OUTPUT: |
b7e2f6b3 | 4461 | port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); |
abe529af | 4462 | if (port && port->bundle && port->bundle->bond) { |
d78be13b | 4463 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4464 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4465 | } |
d78be13b BP |
4466 | break; |
4467 | ||
fea393b1 BP |
4468 | case OVS_ACTION_ATTR_POP_VLAN: |
4469 | vlan_tci = htons(0); | |
d78be13b BP |
4470 | break; |
4471 | ||
fea393b1 BP |
4472 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4473 | vlan = nl_attr_get(a); | |
4474 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4475 | break; |
abe529af BP |
4476 | } |
4477 | } | |
4478 | } | |
4479 | ||
abe529af BP |
4480 | /* Returns true if the only action for 'facet' is to send to the controller. |
4481 | * (We don't report NetFlow expiration messages for such facets because they | |
4482 | * are just part of the control logic for the network, not real traffic). */ | |
4483 | static bool | |
4484 | facet_is_controller_flow(struct facet *facet) | |
4485 | { | |
f25d0cf3 | 4486 | if (facet) { |
b7e2f6b3 EJ |
4487 | struct ofproto_dpif *ofproto = facet->ofproto; |
4488 | const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow, | |
4489 | NULL); | |
4490 | const struct ofpact *ofpacts = rule->up.ofpacts; | |
4491 | size_t ofpacts_len = rule->up.ofpacts_len; | |
f25d0cf3 | 4492 | |
dd30ff28 BP |
4493 | if (ofpacts_len > 0 && |
4494 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4495 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4496 | return true; | |
4497 | } | |
4498 | } | |
4499 | return false; | |
abe529af BP |
4500 | } |
4501 | ||
4502 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4503 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4504 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4505 | * its packet and byte counts before this function is called. */ | |
4506 | static void | |
15baa734 | 4507 | facet_flush_stats(struct facet *facet) |
abe529af | 4508 | { |
b7e2f6b3 | 4509 | struct ofproto_dpif *ofproto = facet->ofproto; |
b0f7b9b5 BP |
4510 | struct subfacet *subfacet; |
4511 | ||
4512 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4513 | ovs_assert(!subfacet->dp_byte_count); |
4514 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4515 | } |
abe529af | 4516 | |
9dfb1f78 | 4517 | facet_push_stats(facet, false); |
3de9590b BP |
4518 | if (facet->accounted_bytes < facet->byte_count) { |
4519 | facet_account(facet); | |
4520 | facet->accounted_bytes = facet->byte_count; | |
4521 | } | |
abe529af BP |
4522 | |
4523 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4524 | struct ofexpired expired; | |
4525 | expired.flow = facet->flow; | |
4526 | expired.packet_count = facet->packet_count; | |
4527 | expired.byte_count = facet->byte_count; | |
4528 | expired.used = facet->used; | |
4529 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4530 | } | |
4531 | ||
abe529af BP |
4532 | /* Reset counters to prevent double counting if 'facet' ever gets |
4533 | * reinstalled. */ | |
bbb5d219 | 4534 | facet_reset_counters(facet); |
abe529af BP |
4535 | |
4536 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4537 | facet->tcp_flags = 0; |
abe529af BP |
4538 | } |
4539 | ||
bcd2633a JP |
4540 | /* Searches 'ofproto''s table of facets for one which would be responsible for |
4541 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4542 | * |
abe529af BP |
4543 | * The returned facet might need revalidation; use facet_lookup_valid() |
4544 | * instead if that is important. */ | |
4545 | static struct facet * | |
bcd2633a | 4546 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4547 | { |
bcd2633a JP |
4548 | struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL); |
4549 | return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL; | |
abe529af BP |
4550 | } |
4551 | ||
bcd2633a JP |
4552 | /* Searches 'ofproto''s table of facets for one capable that covers |
4553 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4554 | * |
abe529af BP |
4555 | * The returned facet is guaranteed to be valid. */ |
4556 | static struct facet * | |
bcd2633a | 4557 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4558 | { |
c57b2226 | 4559 | struct facet *facet; |
abe529af | 4560 | |
bcd2633a | 4561 | facet = facet_find(ofproto, flow); |
abe529af | 4562 | if (facet |
a1616063 | 4563 | && ofproto->backer->need_revalidate |
5bf64ade | 4564 | && !facet_revalidate(facet)) { |
0305ce1f | 4565 | return NULL; |
abe529af BP |
4566 | } |
4567 | ||
4568 | return facet; | |
4569 | } | |
4570 | ||
6814e51f BP |
4571 | static bool |
4572 | facet_check_consistency(struct facet *facet) | |
4573 | { | |
4574 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4575 | ||
bbafd73b EJ |
4576 | struct xlate_out xout; |
4577 | struct xlate_in xin; | |
050ac423 | 4578 | |
6814e51f | 4579 | struct rule_dpif *rule; |
b7e2f6b3 | 4580 | bool ok, fail_open; |
6814e51f BP |
4581 | |
4582 | /* Check the datapath actions for consistency. */ | |
b7e2f6b3 EJ |
4583 | rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); |
4584 | xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL); | |
bbafd73b | 4585 | xlate_actions(&xin, &xout); |
6814e51f | 4586 | |
b7e2f6b3 | 4587 | fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; |
bbafd73b | 4588 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
b7e2f6b3 EJ |
4589 | && facet->xout.slow == xout.slow |
4590 | && facet->fail_open == fail_open; | |
4dff9097 EJ |
4591 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4592 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4593 | |
4dff9097 EJ |
4594 | flow_format(&s, &facet->flow); |
4595 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4596 | |
bbafd73b | 4597 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4598 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4599 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4600 | facet->xout.odp_actions.size); | |
9616614b | 4601 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4602 | format_odp_actions(&s, xout.odp_actions.data, |
4603 | xout.odp_actions.size); | |
c0a71f4e | 4604 | ds_put_char(&s, ')'); |
6814e51f | 4605 | } |
4dff9097 | 4606 | |
bbafd73b EJ |
4607 | if (facet->xout.slow != xout.slow) { |
4608 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4609 | } |
4610 | ||
b7e2f6b3 EJ |
4611 | if (facet->fail_open != fail_open) { |
4612 | ds_put_format(&s, " fail open incorrect. should be %s", | |
4613 | fail_open ? "true" : "false"); | |
4614 | } | |
9616614b | 4615 | ds_destroy(&s); |
6814e51f | 4616 | } |
bbafd73b | 4617 | xlate_out_uninit(&xout); |
6814e51f BP |
4618 | |
4619 | return ok; | |
4620 | } | |
4621 | ||
15baa734 | 4622 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4623 | * |
4624 | * - If the rule found is different from 'facet''s current rule, moves | |
4625 | * 'facet' to the new rule and recompiles its actions. | |
4626 | * | |
4627 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4628 | * where it is and recompiles its actions anyway. |
4629 | * | |
4630 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
5bf64ade EJ |
4631 | * ofproto_receive(), 'facet' is removed. |
4632 | * | |
4633 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
4634 | static bool | |
15baa734 | 4635 | facet_revalidate(struct facet *facet) |
abe529af | 4636 | { |
b7e2f6b3 | 4637 | struct ofproto_dpif *ofproto = facet->ofproto; |
abe529af | 4638 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4639 | struct subfacet *subfacet; |
bcd2633a | 4640 | struct flow_wildcards wc; |
bbafd73b EJ |
4641 | struct xlate_out xout; |
4642 | struct xlate_in xin; | |
abe529af BP |
4643 | |
4644 | COVERAGE_INC(facet_revalidate); | |
4645 | ||
f231418e EJ |
4646 | /* Check that child subfacets still correspond to this facet. Tunnel |
4647 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4648 | * change. */ | |
4649 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4650 | struct ofproto_dpif *recv_ofproto; | |
4651 | struct flow recv_flow; | |
4652 | int error; | |
4653 | ||
4654 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
4655 | subfacet->key_len, &recv_flow, NULL, | |
47b0deae | 4656 | &recv_ofproto, NULL); |
f231418e EJ |
4657 | if (error |
4658 | || recv_ofproto != ofproto | |
bcd2633a | 4659 | || facet != facet_find(ofproto, &recv_flow)) { |
f231418e | 4660 | facet_remove(facet); |
5bf64ade | 4661 | return false; |
f231418e EJ |
4662 | } |
4663 | } | |
4664 | ||
bcd2633a JP |
4665 | flow_wildcards_init_catchall(&wc); |
4666 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc); | |
abe529af | 4667 | |
df2c07f4 | 4668 | /* Calculate new datapath actions. |
abe529af BP |
4669 | * |
4670 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4671 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4672 | * around to properly compose it. */ | |
47b0deae | 4673 | xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); |
bbafd73b | 4674 | xlate_actions(&xin, &xout); |
bcd2633a | 4675 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); |
4dff9097 EJ |
4676 | |
4677 | /* A facet's slow path reason should only change under dramatic | |
4678 | * circumstances. Rather than try to update everything, it's simpler to | |
bcd2633a JP |
4679 | * remove the facet and start over. |
4680 | * | |
4681 | * More importantly, if a facet's wildcards change, it will be relatively | |
4682 | * difficult to figure out if its subfacets still belong to it, and if not | |
4683 | * which facet they may belong to. Again, to avoid the complexity, we | |
4684 | * simply give up instead. */ | |
4685 | if (facet->xout.slow != xout.slow | |
4686 | || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { | |
4dff9097 | 4687 | facet_remove(facet); |
bbafd73b | 4688 | xlate_out_uninit(&xout); |
4dff9097 EJ |
4689 | return false; |
4690 | } | |
6a7e895f | 4691 | |
bbafd73b | 4692 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
4693 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
4694 | if (subfacet->path == SF_FAST_PATH) { | |
4695 | struct dpif_flow_stats stats; | |
b95fc6ba | 4696 | |
bbafd73b | 4697 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 4698 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 4699 | } |
abe529af | 4700 | } |
b95fc6ba | 4701 | |
15baa734 | 4702 | facet_flush_stats(facet); |
4dff9097 | 4703 | |
bbafd73b EJ |
4704 | ofpbuf_clear(&facet->xout.odp_actions); |
4705 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
4706 | xout.odp_actions.size); | |
abe529af BP |
4707 | } |
4708 | ||
4709 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
4710 | facet->xout.slow = xout.slow; |
4711 | facet->xout.has_learn = xout.has_learn; | |
4712 | facet->xout.has_normal = xout.has_normal; | |
4713 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
4714 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
4715 | facet->xout.mirrors = xout.mirrors; | |
4716 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
b7e2f6b3 EJ |
4717 | facet->used = MAX(facet->used, new_rule->up.created); |
4718 | facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY; | |
5bf64ade | 4719 | |
bbafd73b | 4720 | xlate_out_uninit(&xout); |
5bf64ade | 4721 | return true; |
abe529af BP |
4722 | } |
4723 | ||
bbb5d219 EJ |
4724 | static void |
4725 | facet_reset_counters(struct facet *facet) | |
4726 | { | |
4727 | facet->packet_count = 0; | |
4728 | facet->byte_count = 0; | |
9d24de3b JP |
4729 | facet->prev_packet_count = 0; |
4730 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4731 | facet->accounted_bytes = 0; |
4732 | } | |
4733 | ||
abe529af | 4734 | static void |
9dfb1f78 | 4735 | facet_push_stats(struct facet *facet, bool may_learn) |
abe529af | 4736 | { |
112bc5f4 | 4737 | struct dpif_flow_stats stats; |
abe529af | 4738 | |
cb22974d BP |
4739 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4740 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4741 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4742 | |
112bc5f4 BP |
4743 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4744 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4745 | stats.used = facet->used; | |
9dfb1f78 EJ |
4746 | stats.tcp_flags = facet->tcp_flags; |
4747 | ||
4748 | if (may_learn || stats.n_packets || facet->used > facet->prev_used) { | |
b7e2f6b3 | 4749 | struct ofproto_dpif *ofproto = facet->ofproto; |
9dfb1f78 | 4750 | struct ofport_dpif *in_port; |
b7e2f6b3 | 4751 | struct rule_dpif *rule; |
9dfb1f78 | 4752 | struct xlate_in xin; |
abe529af | 4753 | |
9d24de3b JP |
4754 | facet->prev_packet_count = facet->packet_count; |
4755 | facet->prev_byte_count = facet->byte_count; | |
4756 | facet->prev_used = facet->used; | |
abe529af | 4757 | |
4e022ec0 | 4758 | in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port); |
42943cde | 4759 | if (in_port && in_port->is_tunnel) { |
9dfb1f78 EJ |
4760 | netdev_vport_inc_rx(in_port->up.netdev, &stats); |
4761 | } | |
9d24de3b | 4762 | |
b7e2f6b3 EJ |
4763 | rule = rule_dpif_lookup(ofproto, &facet->flow, NULL); |
4764 | rule_credit_stats(rule, &stats); | |
9dfb1f78 EJ |
4765 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, |
4766 | facet->used); | |
4767 | netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); | |
ec7ceaed EJ |
4768 | mirror_update_stats(ofproto->mbridge, facet->xout.mirrors, |
4769 | stats.n_packets, stats.n_bytes); | |
9dfb1f78 | 4770 | |
b7e2f6b3 EJ |
4771 | xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags, |
4772 | NULL); | |
9dfb1f78 EJ |
4773 | xin.resubmit_stats = &stats; |
4774 | xin.may_learn = may_learn; | |
4775 | xlate_actions_for_side_effects(&xin); | |
abe529af BP |
4776 | } |
4777 | } | |
4778 | ||
8844e035 | 4779 | static void |
4464589e | 4780 | push_all_stats__(bool run_fast) |
8844e035 EJ |
4781 | { |
4782 | static long long int rl = LLONG_MIN; | |
4783 | struct ofproto_dpif *ofproto; | |
4784 | ||
4785 | if (time_msec() < rl) { | |
4786 | return; | |
4787 | } | |
4788 | ||
4789 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
bcd2633a | 4790 | struct cls_cursor cursor; |
8844e035 EJ |
4791 | struct facet *facet; |
4792 | ||
bcd2633a JP |
4793 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
4794 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
9dfb1f78 | 4795 | facet_push_stats(facet, false); |
4464589e EJ |
4796 | if (run_fast) { |
4797 | run_fast_rl(); | |
4798 | } | |
8844e035 EJ |
4799 | } |
4800 | } | |
4801 | ||
4802 | rl = time_msec() + 100; | |
4803 | } | |
4804 | ||
4464589e EJ |
4805 | static void |
4806 | push_all_stats(void) | |
4807 | { | |
4808 | push_all_stats__(true); | |
4809 | } | |
4810 | ||
9583bc14 | 4811 | void |
112bc5f4 | 4812 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4813 | { |
112bc5f4 BP |
4814 | rule->packet_count += stats->n_packets; |
4815 | rule->byte_count += stats->n_bytes; | |
4816 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af | 4817 | } |
abe529af | 4818 | \f |
b0f7b9b5 BP |
4819 | /* Subfacets. */ |
4820 | ||
4821 | static struct subfacet * | |
04d08d54 EJ |
4822 | subfacet_find(struct dpif_backer *backer, const struct nlattr *key, |
4823 | size_t key_len, uint32_t key_hash) | |
b0f7b9b5 BP |
4824 | { |
4825 | struct subfacet *subfacet; | |
4826 | ||
4827 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
04d08d54 | 4828 | &backer->subfacets) { |
9566abf9 EJ |
4829 | if (subfacet->key_len == key_len |
4830 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
4831 | return subfacet; |
4832 | } | |
4833 | } | |
4834 | ||
4835 | return NULL; | |
4836 | } | |
4837 | ||
4838 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
4839 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
4840 | * existing subfacet if there is one, otherwise creates and returns a | |
4dff9097 | 4841 | * new subfacet. */ |
b0f7b9b5 | 4842 | static struct subfacet * |
a088a1ff JP |
4843 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
4844 | long long int now) | |
b0f7b9b5 | 4845 | { |
04d08d54 | 4846 | struct dpif_backer *backer = miss->ofproto->backer; |
a088a1ff JP |
4847 | enum odp_key_fitness key_fitness = miss->key_fitness; |
4848 | const struct nlattr *key = miss->key; | |
4849 | size_t key_len = miss->key_len; | |
4850 | uint32_t key_hash; | |
b0f7b9b5 BP |
4851 | struct subfacet *subfacet; |
4852 | ||
a088a1ff JP |
4853 | key_hash = odp_flow_key_hash(key, key_len); |
4854 | ||
3b145dd7 BP |
4855 | if (list_is_empty(&facet->subfacets)) { |
4856 | subfacet = &facet->one_subfacet; | |
4857 | } else { | |
04d08d54 | 4858 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
3b145dd7 BP |
4859 | if (subfacet) { |
4860 | if (subfacet->facet == facet) { | |
4861 | return subfacet; | |
4862 | } | |
4863 | ||
4864 | /* This shouldn't happen. */ | |
4865 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
4866 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
4867 | } |
4868 | ||
3b145dd7 | 4869 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
4870 | } |
4871 | ||
04d08d54 | 4872 | hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); |
b0f7b9b5 BP |
4873 | list_push_back(&facet->subfacets, &subfacet->list_node); |
4874 | subfacet->facet = facet; | |
b0f7b9b5 | 4875 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
4876 | subfacet->key = xmemdup(key, key_len); |
4877 | subfacet->key_len = key_len; | |
459b16a1 | 4878 | subfacet->used = now; |
655ab909 | 4879 | subfacet->created = now; |
26cd7e34 BP |
4880 | subfacet->dp_packet_count = 0; |
4881 | subfacet->dp_byte_count = 0; | |
6a7e895f | 4882 | subfacet->path = SF_NOT_INSTALLED; |
04d08d54 | 4883 | subfacet->backer = backer; |
b0f7b9b5 | 4884 | |
04d08d54 | 4885 | backer->subfacet_add_count++; |
b0f7b9b5 BP |
4886 | return subfacet; |
4887 | } | |
4888 | ||
b0f7b9b5 BP |
4889 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
4890 | * its facet within 'ofproto', and frees it. */ | |
4891 | static void | |
15baa734 | 4892 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4893 | { |
15baa734 | 4894 | struct facet *facet = subfacet->facet; |
b7e2f6b3 | 4895 | struct ofproto_dpif *ofproto = facet->ofproto; |
15baa734 | 4896 | |
655ab909 | 4897 | /* Update ofproto stats before uninstall the subfacet. */ |
dc54ef36 | 4898 | ofproto->backer->subfacet_del_count++; |
655ab909 | 4899 | |
15baa734 | 4900 | subfacet_uninstall(subfacet); |
04d08d54 | 4901 | hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); |
b0f7b9b5 BP |
4902 | list_remove(&subfacet->list_node); |
4903 | free(subfacet->key); | |
26cd7e34 BP |
4904 | if (subfacet != &facet->one_subfacet) { |
4905 | free(subfacet); | |
4906 | } | |
b0f7b9b5 BP |
4907 | } |
4908 | ||
4909 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4910 | * last remaining subfacet in its facet destroys the facet too. */ | |
4911 | static void | |
15baa734 | 4912 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4913 | { |
4914 | struct facet *facet = subfacet->facet; | |
4915 | ||
551a2f6c BP |
4916 | if (list_is_singleton(&facet->subfacets)) { |
4917 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4918 | facet_remove(facet); |
551a2f6c | 4919 | } else { |
15baa734 | 4920 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4921 | } |
4922 | } | |
4923 | ||
1d85f9e5 | 4924 | static void |
04d08d54 | 4925 | subfacet_destroy_batch(struct dpif_backer *backer, |
1d85f9e5 JP |
4926 | struct subfacet **subfacets, int n) |
4927 | { | |
1d85f9e5 JP |
4928 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
4929 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
4930 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
4931 | int i; | |
4932 | ||
4933 | for (i = 0; i < n; i++) { | |
4934 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
4935 | ops[i].u.flow_del.key = subfacets[i]->key; |
4936 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
4937 | ops[i].u.flow_del.stats = &stats[i]; |
4938 | opsp[i] = &ops[i]; | |
4939 | } | |
4940 | ||
04d08d54 | 4941 | dpif_operate(backer->dpif, opsp, n); |
1d85f9e5 JP |
4942 | for (i = 0; i < n; i++) { |
4943 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
4944 | subfacets[i]->path = SF_NOT_INSTALLED; | |
4945 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 4946 | run_fast_rl(); |
1d85f9e5 JP |
4947 | } |
4948 | } | |
4949 | ||
b0f7b9b5 BP |
4950 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4951 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4952 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4953 | * since 'subfacet' was last updated. | |
4954 | * | |
4955 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4956 | static int | |
4dff9097 EJ |
4957 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
4958 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 4959 | { |
15baa734 | 4960 | struct facet *facet = subfacet->facet; |
bbafd73b | 4961 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
4962 | const struct nlattr *actions = odp_actions->data; |
4963 | size_t actions_len = odp_actions->size; | |
d445cc16 JP |
4964 | struct odputil_keybuf maskbuf; |
4965 | struct ofpbuf mask; | |
4dff9097 | 4966 | |
6a7e895f | 4967 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 4968 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
4969 | int ret; |
4970 | ||
0f74f625 AZ |
4971 | flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE |
4972 | : DPIF_FP_MODIFY; | |
b0f7b9b5 BP |
4973 | if (stats) { |
4974 | flags |= DPIF_FP_ZERO_STATS; | |
4975 | } | |
4976 | ||
6a7e895f | 4977 | if (path == SF_SLOW_PATH) { |
b7e2f6b3 | 4978 | compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
4979 | slow_path_stub, sizeof slow_path_stub, |
4980 | &actions, &actions_len); | |
4981 | } | |
4982 | ||
d445cc16 | 4983 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6b35e630 JP |
4984 | if (enable_megaflows) { |
4985 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
4986 | &facet->flow, UINT32_MAX); | |
4987 | } | |
d445cc16 JP |
4988 | |
4989 | ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, | |
4990 | subfacet->key_len, mask.data, mask.size, | |
e6cc0bab | 4991 | actions, actions_len, stats); |
b0f7b9b5 BP |
4992 | |
4993 | if (stats) { | |
4994 | subfacet_reset_dp_stats(subfacet, stats); | |
4995 | } | |
4996 | ||
9b1a48c2 JP |
4997 | if (ret) { |
4998 | COVERAGE_INC(subfacet_install_fail); | |
4999 | } else { | |
6a7e895f BP |
5000 | subfacet->path = path; |
5001 | } | |
b0f7b9b5 BP |
5002 | return ret; |
5003 | } | |
5004 | ||
5005 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
5006 | static void | |
15baa734 | 5007 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5008 | { |
6a7e895f | 5009 | if (subfacet->path != SF_NOT_INSTALLED) { |
b7e2f6b3 | 5010 | struct ofproto_dpif *ofproto = subfacet->facet->ofproto; |
b0f7b9b5 | 5011 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5012 | int error; |
5013 | ||
9566abf9 EJ |
5014 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5015 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5016 | subfacet_reset_dp_stats(subfacet, &stats); |
5017 | if (!error) { | |
15baa734 | 5018 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5019 | } |
6a7e895f | 5020 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5021 | } else { |
cb22974d BP |
5022 | ovs_assert(subfacet->dp_packet_count == 0); |
5023 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5024 | } |
5025 | } | |
5026 | ||
5027 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5028 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5029 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5030 | * was reset in the datapath. 'stats' will be modified to include only | |
5031 | * statistics new since 'subfacet' was last updated. */ | |
5032 | static void | |
5033 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5034 | struct dpif_flow_stats *stats) | |
5035 | { | |
5036 | if (stats | |
5037 | && subfacet->dp_packet_count <= stats->n_packets | |
5038 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5039 | stats->n_packets -= subfacet->dp_packet_count; | |
5040 | stats->n_bytes -= subfacet->dp_byte_count; | |
5041 | } | |
5042 | ||
5043 | subfacet->dp_packet_count = 0; | |
5044 | subfacet->dp_byte_count = 0; | |
5045 | } | |
5046 | ||
b0f7b9b5 BP |
5047 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. |
5048 | * | |
5049 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5050 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5051 | * represents a packet that was sent by hand or if it represents statistics | |
5052 | * that have been cleared out of the datapath. */ | |
5053 | static void | |
15baa734 | 5054 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5055 | const struct dpif_flow_stats *stats) |
5056 | { | |
5057 | if (stats->n_packets || stats->used > subfacet->used) { | |
5058 | struct facet *facet = subfacet->facet; | |
5059 | ||
9dfb1f78 EJ |
5060 | subfacet->used = MAX(subfacet->used, stats->used); |
5061 | facet->used = MAX(facet->used, stats->used); | |
b0f7b9b5 BP |
5062 | facet->packet_count += stats->n_packets; |
5063 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5064 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5065 | } |
5066 | } | |
5067 | \f | |
abe529af BP |
5068 | /* Rules. */ |
5069 | ||
bcd2633a JP |
5070 | /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets |
5071 | * the fields that were relevant as part of the lookup. */ | |
abe529af | 5072 | static struct rule_dpif * |
bcd2633a JP |
5073 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
5074 | struct flow_wildcards *wc) | |
c57b2226 | 5075 | { |
c57b2226 BP |
5076 | struct rule_dpif *rule; |
5077 | ||
9583bc14 | 5078 | rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0); |
c57b2226 BP |
5079 | if (rule) { |
5080 | return rule; | |
5081 | } | |
5082 | ||
c376f9a3 | 5083 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5084 | } |
5085 | ||
9583bc14 EJ |
5086 | struct rule_dpif * |
5087 | rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, | |
5088 | const struct flow *flow, struct flow_wildcards *wc, | |
5089 | uint8_t table_id) | |
abe529af | 5090 | { |
7257b535 BP |
5091 | struct cls_rule *cls_rule; |
5092 | struct classifier *cls; | |
7fd51d39 | 5093 | bool frag; |
7257b535 | 5094 | |
9cdaaebe BP |
5095 | if (table_id >= N_TABLES) { |
5096 | return NULL; | |
5097 | } | |
5098 | ||
1dd35f8a | 5099 | if (wc) { |
7431e171 | 5100 | memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type); |
1dd35f8a JP |
5101 | wc->masks.nw_frag |= FLOW_NW_FRAG_MASK; |
5102 | } | |
5103 | ||
d0918789 | 5104 | cls = &ofproto->up.tables[table_id].cls; |
7fd51d39 BP |
5105 | frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; |
5106 | if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
5107 | /* We must pretend that transport ports are unavailable. */ | |
7257b535 BP |
5108 | struct flow ofpc_normal_flow = *flow; |
5109 | ofpc_normal_flow.tp_src = htons(0); | |
5110 | ofpc_normal_flow.tp_dst = htons(0); | |
bcd2633a | 5111 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); |
7fd51d39 BP |
5112 | } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { |
5113 | cls_rule = &ofproto->drop_frags_rule->up.cr; | |
bcd2633a JP |
5114 | if (wc) { |
5115 | flow_wildcards_init_exact(wc); | |
5116 | } | |
7257b535 | 5117 | } else { |
bcd2633a | 5118 | cls_rule = classifier_lookup(cls, flow, wc); |
7257b535 BP |
5119 | } |
5120 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5121 | } |
5122 | ||
9583bc14 | 5123 | struct rule_dpif * |
c376f9a3 IY |
5124 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) |
5125 | { | |
5126 | struct ofport_dpif *port; | |
5127 | ||
4e022ec0 | 5128 | port = get_ofp_port(ofproto, flow->in_port.ofp_port); |
c376f9a3 | 5129 | if (!port) { |
4e022ec0 AW |
5130 | VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16, |
5131 | flow->in_port.ofp_port); | |
c376f9a3 IY |
5132 | return ofproto->miss_rule; |
5133 | } | |
5134 | ||
5135 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5136 | return ofproto->no_packet_in_rule; | |
5137 | } | |
5138 | return ofproto->miss_rule; | |
5139 | } | |
5140 | ||
7ee20df1 BP |
5141 | static void |
5142 | complete_operation(struct rule_dpif *rule) | |
5143 | { | |
5144 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5145 | ||
a1616063 | 5146 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
7ee20df1 BP |
5147 | if (clogged) { |
5148 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5149 | c->op = rule->up.pending; | |
5150 | list_push_back(&ofproto->completions, &c->list_node); | |
5151 | } else { | |
5152 | ofoperation_complete(rule->up.pending, 0); | |
5153 | } | |
5154 | } | |
5155 | ||
abe529af BP |
5156 | static struct rule * |
5157 | rule_alloc(void) | |
5158 | { | |
5159 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5160 | return &rule->up; | |
5161 | } | |
5162 | ||
5163 | static void | |
5164 | rule_dealloc(struct rule *rule_) | |
5165 | { | |
5166 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5167 | free(rule); | |
5168 | } | |
5169 | ||
90bf1e07 | 5170 | static enum ofperr |
abe529af BP |
5171 | rule_construct(struct rule *rule_) |
5172 | { | |
5173 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5174 | rule->packet_count = 0; |
5175 | rule->byte_count = 0; | |
7ee20df1 | 5176 | complete_operation(rule); |
abe529af BP |
5177 | return 0; |
5178 | } | |
5179 | ||
5180 | static void | |
b7e2f6b3 | 5181 | rule_destruct(struct rule *rule) |
abe529af | 5182 | { |
b7e2f6b3 | 5183 | complete_operation(rule_dpif_cast(rule)); |
abe529af BP |
5184 | } |
5185 | ||
5186 | static void | |
5187 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5188 | { | |
5189 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 5190 | |
4464589e EJ |
5191 | /* push_all_stats() can handle flow misses which, when using the learn |
5192 | * action, can cause rules to be added and deleted. This can corrupt our | |
5193 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5194 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5195 | push_all_stats__(false); | |
bf1e8ff9 | 5196 | |
abe529af BP |
5197 | /* Start from historical data for 'rule' itself that are no longer tracked |
5198 | * in facets. This counts, for example, facets that have expired. */ | |
5199 | *packets = rule->packet_count; | |
5200 | *bytes = rule->byte_count; | |
abe529af BP |
5201 | } |
5202 | ||
0a740f48 EJ |
5203 | static void |
5204 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5205 | struct ofpbuf *packet) | |
abe529af | 5206 | { |
abe529af | 5207 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
112bc5f4 | 5208 | struct dpif_flow_stats stats; |
bbafd73b EJ |
5209 | struct xlate_out xout; |
5210 | struct xlate_in xin; | |
abe529af | 5211 | |
a7752d4a | 5212 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5213 | rule_credit_stats(rule, &stats); |
5214 | ||
47b0deae | 5215 | xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); |
bbafd73b EJ |
5216 | xin.resubmit_stats = &stats; |
5217 | xlate_actions(&xin, &xout); | |
112bc5f4 | 5218 | |
bbafd73b EJ |
5219 | execute_odp_actions(ofproto, flow, xout.odp_actions.data, |
5220 | xout.odp_actions.size, packet); | |
112bc5f4 | 5221 | |
bbafd73b | 5222 | xlate_out_uninit(&xout); |
0a740f48 | 5223 | } |
5bf0e941 | 5224 | |
0a740f48 EJ |
5225 | static enum ofperr |
5226 | rule_execute(struct rule *rule, const struct flow *flow, | |
5227 | struct ofpbuf *packet) | |
5228 | { | |
5229 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5230 | ofpbuf_delete(packet); | |
5bf0e941 | 5231 | return 0; |
abe529af BP |
5232 | } |
5233 | ||
7ee20df1 BP |
5234 | static void |
5235 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5236 | { |
5237 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5238 | |
5239 | complete_operation(rule); | |
abe529af BP |
5240 | } |
5241 | \f | |
97d6520b | 5242 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5243 | * May modify 'packet'. |
abe529af BP |
5244 | * Returns 0 if successful, otherwise a positive errno value. */ |
5245 | static int | |
52a90c29 | 5246 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5247 | { |
79e15b78 | 5248 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5249 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 | 5250 | struct ofpbuf key, odp_actions; |
0f49659a | 5251 | struct dpif_flow_stats stats; |
80e5eed9 | 5252 | struct odputil_keybuf keybuf; |
0f49659a | 5253 | struct ofpact_output output; |
bbafd73b EJ |
5254 | struct xlate_out xout; |
5255 | struct xlate_in xin; | |
80e5eed9 | 5256 | struct flow flow; |
4e022ec0 | 5257 | union flow_in_port in_port_; |
abe529af BP |
5258 | int error; |
5259 | ||
b9ad7294 | 5260 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
0f49659a | 5261 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
b9ad7294 | 5262 | |
0f49659a | 5263 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ |
4e022ec0 AW |
5264 | in_port_.ofp_port = OFPP_NONE; |
5265 | flow_extract(packet, 0, 0, NULL, &in_port_, &flow); | |
0f49659a EJ |
5266 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, |
5267 | OFPP_LOCAL)); | |
5268 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 | 5269 | |
0f49659a EJ |
5270 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); |
5271 | output.port = ofport->up.ofp_port; | |
5272 | output.max_len = 0; | |
b9ad7294 | 5273 | |
47b0deae | 5274 | xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet); |
bbafd73b EJ |
5275 | xin.ofpacts_len = sizeof output; |
5276 | xin.ofpacts = &output.ofpact; | |
5277 | xin.resubmit_stats = &stats; | |
5278 | xlate_actions(&xin, &xout); | |
6ff686f2 | 5279 | |
acf60855 | 5280 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 | 5281 | key.data, key.size, |
bbafd73b | 5282 | xout.odp_actions.data, xout.odp_actions.size, |
abe529af | 5283 | packet); |
bbafd73b | 5284 | xlate_out_uninit(&xout); |
abe529af BP |
5285 | |
5286 | if (error) { | |
0f49659a EJ |
5287 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", |
5288 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
10a89ef0 | 5289 | ovs_strerror(error)); |
abe529af | 5290 | } |
79e15b78 EJ |
5291 | |
5292 | ofproto->stats.tx_packets++; | |
5293 | ofproto->stats.tx_bytes += packet->size; | |
abe529af BP |
5294 | return error; |
5295 | } | |
abe529af | 5296 | |
6a7e895f BP |
5297 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5298 | * The action will state 'slow' as the reason that the action is in the slow | |
5299 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5300 | * dump-flows" output to see why a flow is in the slow path.) | |
5301 | * | |
5302 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5303 | * 'stub_size' must be large enough for the action. | |
5304 | * | |
5305 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5306 | * respectively. */ | |
5307 | static void | |
5308 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5309 | enum slow_path_reason slow, | |
5310 | uint64_t *stub, size_t stub_size, | |
5311 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5312 | { | |
5313 | union user_action_cookie cookie; | |
5314 | struct ofpbuf buf; | |
5315 | ||
5316 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5317 | cookie.slow_path.unused = 0; | |
5318 | cookie.slow_path.reason = slow; | |
5319 | ||
5320 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5321 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
4e022ec0 AW |
5322 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, |
5323 | ODPP_NONE); | |
29089a54 | 5324 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5325 | } else { |
29089a54 RL |
5326 | put_userspace_action(ofproto, &buf, flow, &cookie, |
5327 | sizeof cookie.slow_path); | |
625b0720 | 5328 | } |
6a7e895f BP |
5329 | *actionsp = buf.data; |
5330 | *actions_lenp = buf.size; | |
5331 | } | |
5332 | ||
9583bc14 | 5333 | size_t |
98403001 BP |
5334 | put_userspace_action(const struct ofproto_dpif *ofproto, |
5335 | struct ofpbuf *odp_actions, | |
5336 | const struct flow *flow, | |
29089a54 RL |
5337 | const union user_action_cookie *cookie, |
5338 | const size_t cookie_size) | |
98403001 | 5339 | { |
98403001 BP |
5340 | uint32_t pid; |
5341 | ||
acf60855 | 5342 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
4e022ec0 AW |
5343 | ofp_port_to_odp_port(ofproto, |
5344 | flow->in_port.ofp_port)); | |
98403001 | 5345 | |
29089a54 RL |
5346 | return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); |
5347 | } | |
9583bc14 EJ |
5348 | \f |
5349 | static bool | |
5350 | set_frag_handling(struct ofproto *ofproto_, | |
5351 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5352 | { |
5353 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 5354 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 5355 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
5356 | return true; |
5357 | } else { | |
5358 | return false; | |
5359 | } | |
abe529af BP |
5360 | } |
5361 | ||
90bf1e07 | 5362 | static enum ofperr |
abe529af BP |
5363 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
5364 | const struct flow *flow, | |
f25d0cf3 | 5365 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
5366 | { |
5367 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
548de4dd BP |
5368 | struct odputil_keybuf keybuf; |
5369 | struct dpif_flow_stats stats; | |
bbafd73b EJ |
5370 | struct xlate_out xout; |
5371 | struct xlate_in xin; | |
548de4dd | 5372 | struct ofpbuf key; |
112bc5f4 | 5373 | |
80e5eed9 | 5374 | |
548de4dd | 5375 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a | 5376 | odp_flow_key_from_flow(&key, flow, |
4e022ec0 AW |
5377 | ofp_port_to_odp_port(ofproto, |
5378 | flow->in_port.ofp_port)); | |
050ac423 | 5379 | |
548de4dd | 5380 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 5381 | |
47b0deae | 5382 | xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet); |
bbafd73b EJ |
5383 | xin.resubmit_stats = &stats; |
5384 | xin.ofpacts_len = ofpacts_len; | |
5385 | xin.ofpacts = ofpacts; | |
2284188b | 5386 | |
bbafd73b | 5387 | xlate_actions(&xin, &xout); |
acf60855 | 5388 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
bbafd73b EJ |
5389 | xout.odp_actions.data, xout.odp_actions.size, packet); |
5390 | xlate_out_uninit(&xout); | |
2284188b | 5391 | |
548de4dd | 5392 | return 0; |
abe529af | 5393 | } |
6fca1ffb BP |
5394 | \f |
5395 | /* NetFlow. */ | |
5396 | ||
5397 | static int | |
5398 | set_netflow(struct ofproto *ofproto_, | |
5399 | const struct netflow_options *netflow_options) | |
5400 | { | |
5401 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5402 | ||
5403 | if (netflow_options) { | |
5404 | if (!ofproto->netflow) { | |
5405 | ofproto->netflow = netflow_create(); | |
7bab1576 | 5406 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6fca1ffb BP |
5407 | } |
5408 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7bab1576 EJ |
5409 | } else if (ofproto->netflow) { |
5410 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
6fca1ffb BP |
5411 | netflow_destroy(ofproto->netflow); |
5412 | ofproto->netflow = NULL; | |
6fca1ffb | 5413 | } |
7bab1576 EJ |
5414 | |
5415 | return 0; | |
6fca1ffb | 5416 | } |
abe529af BP |
5417 | |
5418 | static void | |
5419 | get_netflow_ids(const struct ofproto *ofproto_, | |
5420 | uint8_t *engine_type, uint8_t *engine_id) | |
5421 | { | |
5422 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5423 | ||
acf60855 | 5424 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 5425 | } |
6fca1ffb BP |
5426 | |
5427 | static void | |
5428 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
5429 | { | |
5430 | if (!facet_is_controller_flow(facet) && | |
5431 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 5432 | struct subfacet *subfacet; |
6fca1ffb BP |
5433 | struct ofexpired expired; |
5434 | ||
b0f7b9b5 | 5435 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 5436 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 5437 | struct dpif_flow_stats stats; |
6fca1ffb | 5438 | |
bcd2633a JP |
5439 | subfacet_install(subfacet, &facet->xout.odp_actions, |
5440 | &stats); | |
15baa734 | 5441 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5442 | } |
6fca1ffb BP |
5443 | } |
5444 | ||
5445 | expired.flow = facet->flow; | |
5446 | expired.packet_count = facet->packet_count; | |
5447 | expired.byte_count = facet->byte_count; | |
5448 | expired.used = facet->used; | |
5449 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
5450 | } | |
5451 | } | |
5452 | ||
5453 | static void | |
5454 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
5455 | { | |
bcd2633a | 5456 | struct cls_cursor cursor; |
6fca1ffb BP |
5457 | struct facet *facet; |
5458 | ||
bcd2633a JP |
5459 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5460 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6fca1ffb BP |
5461 | send_active_timeout(ofproto, facet); |
5462 | } | |
5463 | } | |
abe529af BP |
5464 | \f |
5465 | static struct ofproto_dpif * | |
5466 | ofproto_dpif_lookup(const char *name) | |
5467 | { | |
b44a10b7 BP |
5468 | struct ofproto_dpif *ofproto; |
5469 | ||
5470 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
5471 | hash_string(name, 0), &all_ofproto_dpifs) { | |
5472 | if (!strcmp(ofproto->up.name, name)) { | |
5473 | return ofproto; | |
5474 | } | |
5475 | } | |
5476 | return NULL; | |
abe529af BP |
5477 | } |
5478 | ||
f0a3aa2e | 5479 | static void |
96e466a3 | 5480 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 5481 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 5482 | { |
490df1ef | 5483 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 5484 | |
96e466a3 EJ |
5485 | if (argc > 1) { |
5486 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5487 | if (!ofproto) { | |
bde9f75d | 5488 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
5489 | return; |
5490 | } | |
509c0149 | 5491 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5492 | mac_learning_flush(ofproto->ml); |
509c0149 | 5493 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 EJ |
5494 | } else { |
5495 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
509c0149 | 5496 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5497 | mac_learning_flush(ofproto->ml); |
509c0149 | 5498 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 | 5499 | } |
f0a3aa2e | 5500 | } |
f0a3aa2e | 5501 | |
bde9f75d | 5502 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
5503 | } |
5504 | ||
46c88433 EJ |
5505 | static struct ofport_dpif * |
5506 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5507 | { | |
5508 | return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif, | |
5509 | bundle_node); | |
5510 | } | |
5511 | ||
abe529af | 5512 | static void |
0e15264f BP |
5513 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5514 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
5515 | { |
5516 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5517 | const struct ofproto_dpif *ofproto; | |
5518 | const struct mac_entry *e; | |
5519 | ||
0e15264f | 5520 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 5521 | if (!ofproto) { |
bde9f75d | 5522 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
5523 | return; |
5524 | } | |
5525 | ||
5526 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
509c0149 | 5527 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
5528 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
5529 | struct ofbundle *bundle = e->port.p; | |
ad28062f BP |
5530 | char name[OFP_MAX_PORT_NAME_LEN]; |
5531 | ||
5532 | ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port, | |
5533 | name, sizeof name); | |
5534 | ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n", | |
5535 | name, e->vlan, ETH_ADDR_ARGS(e->mac), | |
e764773c | 5536 | mac_entry_age(ofproto->ml, e)); |
abe529af | 5537 | } |
509c0149 | 5538 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
bde9f75d | 5539 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
5540 | ds_destroy(&ds); |
5541 | } | |
5542 | ||
6a6455e5 | 5543 | struct trace_ctx { |
bbafd73b EJ |
5544 | struct xlate_out xout; |
5545 | struct xlate_in xin; | |
abe529af BP |
5546 | struct flow flow; |
5547 | struct ds *result; | |
5548 | }; | |
5549 | ||
5550 | static void | |
4d0acc70 | 5551 | trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) |
abe529af BP |
5552 | { |
5553 | ds_put_char_multiple(result, '\t', level); | |
5554 | if (!rule) { | |
5555 | ds_put_cstr(result, "No match\n"); | |
5556 | return; | |
5557 | } | |
5558 | ||
29901626 | 5559 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
4d0acc70 | 5560 | rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie)); |
79feb7df | 5561 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5562 | ds_put_char(result, '\n'); |
5563 | ||
5564 | ds_put_char_multiple(result, '\t', level); | |
5565 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 5566 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
5567 | ds_put_char(result, '\n'); |
5568 | } | |
5569 | ||
5570 | static void | |
5571 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 5572 | struct trace_ctx *trace) |
abe529af BP |
5573 | { |
5574 | ds_put_char_multiple(result, '\t', level); | |
5575 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 5576 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
5577 | ds_put_cstr(result, "unchanged"); |
5578 | } else { | |
bbafd73b EJ |
5579 | flow_format(result, &trace->xin.flow); |
5580 | trace->flow = trace->xin.flow; | |
abe529af BP |
5581 | } |
5582 | ds_put_char(result, '\n'); | |
5583 | } | |
5584 | ||
eb9e1c26 EJ |
5585 | static void |
5586 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 5587 | struct trace_ctx *trace) |
eb9e1c26 EJ |
5588 | { |
5589 | size_t i; | |
5590 | ||
5591 | ds_put_char_multiple(result, '\t', level); | |
5592 | ds_put_format(result, "%s:", title); | |
5593 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5594 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5595 | } | |
5596 | ds_put_char(result, '\n'); | |
5597 | } | |
5598 | ||
1ed8d352 EJ |
5599 | static void |
5600 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 5601 | struct trace_ctx *trace) |
1ed8d352 | 5602 | { |
bbafd73b | 5603 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
5604 | |
5605 | ds_put_char_multiple(result, '\t', level); | |
5606 | ds_put_format(result, "%s: ", title); | |
5607 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
5608 | ds_put_char(result, '\n'); | |
5609 | } | |
5610 | ||
abe529af | 5611 | static void |
4d0acc70 | 5612 | trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse) |
abe529af | 5613 | { |
4d0acc70 | 5614 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
abe529af BP |
5615 | struct ds *result = trace->result; |
5616 | ||
5617 | ds_put_char(result, '\n'); | |
4d0acc70 EJ |
5618 | trace_format_flow(result, recurse + 1, "Resubmitted flow", trace); |
5619 | trace_format_regs(result, recurse + 1, "Resubmitted regs", trace); | |
5620 | trace_format_odp(result, recurse + 1, "Resubmitted odp", trace); | |
5621 | trace_format_rule(result, recurse + 1, rule); | |
abe529af BP |
5622 | } |
5623 | ||
479df176 | 5624 | static void |
4d0acc70 | 5625 | trace_report(struct xlate_in *xin, const char *s, int recurse) |
479df176 | 5626 | { |
4d0acc70 | 5627 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
479df176 BP |
5628 | struct ds *result = trace->result; |
5629 | ||
4d0acc70 | 5630 | ds_put_char_multiple(result, '\t', recurse); |
479df176 BP |
5631 | ds_put_cstr(result, s); |
5632 | ds_put_char(result, '\n'); | |
5633 | } | |
5634 | ||
abe529af | 5635 | static void |
0e15264f | 5636 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
5637 | void *aux OVS_UNUSED) |
5638 | { | |
50aa28fd | 5639 | const struct dpif_backer *backer; |
abe529af | 5640 | struct ofproto_dpif *ofproto; |
ea5e3887 | 5641 | struct ofpbuf odp_key, odp_mask; |
876b0e1c | 5642 | struct ofpbuf *packet; |
abe529af BP |
5643 | struct ds result; |
5644 | struct flow flow; | |
abe529af BP |
5645 | char *s; |
5646 | ||
876b0e1c | 5647 | packet = NULL; |
50aa28fd | 5648 | backer = NULL; |
abe529af | 5649 | ds_init(&result); |
50aa28fd | 5650 | ofpbuf_init(&odp_key, 0); |
ea5e3887 | 5651 | ofpbuf_init(&odp_mask, 0); |
abe529af | 5652 | |
50aa28fd AW |
5653 | /* Handle "-generate" or a hex string as the last argument. */ |
5654 | if (!strcmp(argv[argc - 1], "-generate")) { | |
5655 | packet = ofpbuf_new(0); | |
5656 | argc--; | |
5657 | } else { | |
5658 | const char *error = eth_from_hex(argv[argc - 1], &packet); | |
5659 | if (!error) { | |
5660 | argc--; | |
5661 | } else if (argc == 4) { | |
5662 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
5663 | unixctl_command_reply_error(conn, error); | |
5664 | goto exit; | |
5665 | } | |
e84173dc | 5666 | } |
876b0e1c | 5667 | |
50aa28fd AW |
5668 | /* Parse the flow and determine whether a datapath or |
5669 | * bridge is specified. If function odp_flow_key_from_string() | |
5670 | * returns 0, the flow is a odp_flow. If function | |
5671 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
ea5e3887 | 5672 | if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) { |
50aa28fd AW |
5673 | /* If the odp_flow is the second argument, |
5674 | * the datapath name is the first argument. */ | |
5675 | if (argc == 3) { | |
5676 | const char *dp_type; | |
5677 | if (!strncmp(argv[1], "ovs-", 4)) { | |
5678 | dp_type = argv[1] + 4; | |
5679 | } else { | |
5680 | dp_type = argv[1]; | |
31a19d69 | 5681 | } |
50aa28fd AW |
5682 | backer = shash_find_data(&all_dpif_backers, dp_type); |
5683 | if (!backer) { | |
5684 | unixctl_command_reply_error(conn, "Cannot find datapath " | |
5685 | "of this name"); | |
31a19d69 BP |
5686 | goto exit; |
5687 | } | |
5688 | } else { | |
50aa28fd AW |
5689 | /* No datapath name specified, so there should be only one |
5690 | * datapath. */ | |
5691 | struct shash_node *node; | |
5692 | if (shash_count(&all_dpif_backers) != 1) { | |
5693 | unixctl_command_reply_error(conn, "Must specify datapath " | |
5694 | "name, there is more than one type of datapath"); | |
31a19d69 BP |
5695 | goto exit; |
5696 | } | |
50aa28fd AW |
5697 | node = shash_first(&all_dpif_backers); |
5698 | backer = node->data; | |
876b0e1c | 5699 | } |
8b3b8dd1 | 5700 | |
50aa28fd AW |
5701 | /* Extract the ofproto_dpif object from the ofproto_receive() |
5702 | * function. */ | |
5703 | if (ofproto_receive(backer, NULL, odp_key.data, | |
47b0deae | 5704 | odp_key.size, &flow, NULL, &ofproto, NULL)) { |
50aa28fd AW |
5705 | unixctl_command_reply_error(conn, "Invalid datapath flow"); |
5706 | goto exit; | |
8b3b8dd1 | 5707 | } |
50aa28fd AW |
5708 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
5709 | } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { | |
5710 | if (argc != 3) { | |
5711 | unixctl_command_reply_error(conn, "Must specify bridge name"); | |
876b0e1c BP |
5712 | goto exit; |
5713 | } | |
5714 | ||
50aa28fd AW |
5715 | ofproto = ofproto_dpif_lookup(argv[1]); |
5716 | if (!ofproto) { | |
5717 | unixctl_command_reply_error(conn, "Unknown bridge name"); | |
5718 | goto exit; | |
5719 | } | |
876b0e1c | 5720 | } else { |
50aa28fd | 5721 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
abe529af BP |
5722 | goto exit; |
5723 | } | |
5724 | ||
50aa28fd AW |
5725 | /* Generate a packet, if requested. */ |
5726 | if (packet) { | |
5727 | if (!packet->size) { | |
5728 | flow_compose(packet, &flow); | |
5729 | } else { | |
4e022ec0 AW |
5730 | union flow_in_port in_port_; |
5731 | ||
5732 | in_port_ = flow.in_port; | |
50aa28fd AW |
5733 | ds_put_cstr(&result, "Packet: "); |
5734 | s = ofp_packet_to_string(packet->data, packet->size); | |
5735 | ds_put_cstr(&result, s); | |
5736 | free(s); | |
5737 | ||
5738 | /* Use the metadata from the flow and the packet argument | |
5739 | * to reconstruct the flow. */ | |
5740 | flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, | |
4e022ec0 | 5741 | &in_port_, &flow); |
50aa28fd AW |
5742 | } |
5743 | } | |
5744 | ||
47b0deae | 5745 | ofproto_trace(ofproto, &flow, packet, &result); |
6a6455e5 EJ |
5746 | unixctl_command_reply(conn, ds_cstr(&result)); |
5747 | ||
5748 | exit: | |
5749 | ds_destroy(&result); | |
5750 | ofpbuf_delete(packet); | |
5751 | ofpbuf_uninit(&odp_key); | |
ea5e3887 | 5752 | ofpbuf_uninit(&odp_mask); |
6a6455e5 EJ |
5753 | } |
5754 | ||
9583bc14 | 5755 | void |
6a6455e5 | 5756 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, |
47b0deae | 5757 | const struct ofpbuf *packet, struct ds *ds) |
6a6455e5 EJ |
5758 | { |
5759 | struct rule_dpif *rule; | |
5760 | ||
5761 | ds_put_cstr(ds, "Flow: "); | |
5762 | flow_format(ds, flow); | |
5763 | ds_put_char(ds, '\n'); | |
abe529af | 5764 | |
bcd2633a | 5765 | rule = rule_dpif_lookup(ofproto, flow, NULL); |
c57b2226 | 5766 | |
4d0acc70 | 5767 | trace_format_rule(ds, 0, rule); |
c57b2226 BP |
5768 | if (rule == ofproto->miss_rule) { |
5769 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
5770 | } else if (rule == ofproto->no_packet_in_rule) { | |
5771 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
5772 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7fd51d39 BP |
5773 | } else if (rule == ofproto->drop_frags_rule) { |
5774 | ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " | |
5775 | "and the fragment handling mode is \"drop\".\n"); | |
c57b2226 BP |
5776 | } |
5777 | ||
abe529af | 5778 | if (rule) { |
050ac423 BP |
5779 | uint64_t odp_actions_stub[1024 / 8]; |
5780 | struct ofpbuf odp_actions; | |
6a6455e5 | 5781 | struct trace_ctx trace; |
bcd2633a | 5782 | struct match match; |
0e553d9c | 5783 | uint8_t tcp_flags; |
abe529af | 5784 | |
6a6455e5 EJ |
5785 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
5786 | trace.result = ds; | |
5787 | trace.flow = *flow; | |
050ac423 BP |
5788 | ofpbuf_use_stub(&odp_actions, |
5789 | odp_actions_stub, sizeof odp_actions_stub); | |
47b0deae | 5790 | xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); |
bbafd73b EJ |
5791 | trace.xin.resubmit_hook = trace_resubmit; |
5792 | trace.xin.report_hook = trace_report; | |
bcd2633a | 5793 | |
bbafd73b | 5794 | xlate_actions(&trace.xin, &trace.xout); |
abe529af | 5795 | |
6a6455e5 EJ |
5796 | ds_put_char(ds, '\n'); |
5797 | trace_format_flow(ds, 0, "Final flow", &trace); | |
bcd2633a JP |
5798 | |
5799 | match_init(&match, flow, &trace.xout.wc); | |
5800 | ds_put_cstr(ds, "Relevant fields: "); | |
5801 | match_format(&match, ds, OFP_DEFAULT_PRIORITY); | |
5802 | ds_put_char(ds, '\n'); | |
5803 | ||
6a6455e5 | 5804 | ds_put_cstr(ds, "Datapath actions: "); |
bbafd73b EJ |
5805 | format_odp_actions(ds, trace.xout.odp_actions.data, |
5806 | trace.xout.odp_actions.size); | |
876b0e1c | 5807 | |
bbafd73b | 5808 | if (trace.xout.slow) { |
6a7e895f BP |
5809 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
5810 | "slow path because it:"); | |
bbafd73b | 5811 | switch (trace.xout.slow) { |
98f0520f EJ |
5812 | case SLOW_CFM: |
5813 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
5814 | break; | |
5815 | case SLOW_LACP: | |
5816 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
5817 | break; | |
5818 | case SLOW_STP: | |
5819 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
5820 | break; | |
5821 | case SLOW_BFD: | |
5822 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
5823 | break; | |
5824 | case SLOW_CONTROLLER: | |
5825 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
5826 | "to the OpenFlow controller."); | |
5827 | break; | |
5828 | case __SLOW_MAX: | |
5829 | NOT_REACHED(); | |
6a7e895f | 5830 | } |
876b0e1c | 5831 | } |
bbafd73b EJ |
5832 | |
5833 | xlate_out_uninit(&trace.xout); | |
abe529af | 5834 | } |
abe529af BP |
5835 | } |
5836 | ||
7ee20df1 | 5837 | static void |
0e15264f BP |
5838 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
5839 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
5840 | { |
5841 | clogged = true; | |
bde9f75d | 5842 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
5843 | } |
5844 | ||
5845 | static void | |
0e15264f BP |
5846 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
5847 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
5848 | { |
5849 | clogged = false; | |
bde9f75d | 5850 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
5851 | } |
5852 | ||
6814e51f BP |
5853 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
5854 | * 'reply' describing the results. */ | |
5855 | static void | |
5856 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
5857 | { | |
bcd2633a | 5858 | struct cls_cursor cursor; |
6814e51f BP |
5859 | struct facet *facet; |
5860 | int errors; | |
5861 | ||
5862 | errors = 0; | |
bcd2633a JP |
5863 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5864 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6814e51f BP |
5865 | if (!facet_check_consistency(facet)) { |
5866 | errors++; | |
5867 | } | |
5868 | } | |
5869 | if (errors) { | |
2cc3c58e | 5870 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
5871 | } |
5872 | ||
5873 | if (errors) { | |
5874 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
5875 | ofproto->up.name, errors); | |
5876 | } else { | |
5877 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
5878 | } | |
5879 | } | |
5880 | ||
5881 | static void | |
5882 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
5883 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
5884 | { | |
5885 | struct ds reply = DS_EMPTY_INITIALIZER; | |
5886 | struct ofproto_dpif *ofproto; | |
5887 | ||
5888 | if (argc > 1) { | |
5889 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5890 | if (!ofproto) { | |
bde9f75d EJ |
5891 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
5892 | "ofproto/list for help)"); | |
6814e51f BP |
5893 | return; |
5894 | } | |
5895 | ofproto_dpif_self_check__(ofproto, &reply); | |
5896 | } else { | |
5897 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5898 | ofproto_dpif_self_check__(ofproto, &reply); | |
5899 | } | |
5900 | } | |
5901 | ||
bde9f75d | 5902 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
5903 | ds_destroy(&reply); |
5904 | } | |
5905 | ||
27022416 JP |
5906 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
5907 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
5908 | * to destroy 'ofproto_shash' and free the returned value. */ | |
5909 | static const struct shash_node ** | |
5910 | get_ofprotos(struct shash *ofproto_shash) | |
5911 | { | |
5912 | const struct ofproto_dpif *ofproto; | |
5913 | ||
5914 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5915 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
5916 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
5917 | } | |
5918 | ||
5919 | return shash_sort(ofproto_shash); | |
5920 | } | |
5921 | ||
5922 | static void | |
5923 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
5924 | const char *argv[] OVS_UNUSED, | |
5925 | void *aux OVS_UNUSED) | |
5926 | { | |
5927 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5928 | struct shash ofproto_shash; | |
5929 | const struct shash_node **sorted_ofprotos; | |
5930 | int i; | |
5931 | ||
5932 | shash_init(&ofproto_shash); | |
5933 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
5934 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5935 | const struct shash_node *node = sorted_ofprotos[i]; | |
5936 | ds_put_format(&ds, "%s\n", node->name); | |
5937 | } | |
5938 | ||
5939 | shash_destroy(&ofproto_shash); | |
5940 | free(sorted_ofprotos); | |
5941 | ||
5942 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5943 | ds_destroy(&ds); | |
5944 | } | |
5945 | ||
5946 | static void | |
dc54ef36 EJ |
5947 | show_dp_rates(struct ds *ds, const char *heading, |
5948 | const struct avg_subfacet_rates *rates) | |
5949 | { | |
5950 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
5951 | heading, rates->add_rate, rates->del_rate); | |
5952 | } | |
5953 | ||
5954 | static void | |
5955 | dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) | |
27022416 | 5956 | { |
dc54ef36 EJ |
5957 | const struct shash_node **ofprotos; |
5958 | struct ofproto_dpif *ofproto; | |
5959 | struct shash ofproto_shash; | |
09672174 | 5960 | uint64_t n_hit, n_missed; |
dc54ef36 | 5961 | long long int minutes; |
09672174 | 5962 | size_t i; |
655ab909 | 5963 | |
04d08d54 | 5964 | n_hit = n_missed = 0; |
dc54ef36 EJ |
5965 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
5966 | if (ofproto->backer == backer) { | |
dc54ef36 EJ |
5967 | n_missed += ofproto->n_missed; |
5968 | n_hit += ofproto->n_hit; | |
5969 | } | |
5970 | } | |
655ab909 | 5971 | |
dc54ef36 EJ |
5972 | ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
5973 | dpif_name(backer->dpif), n_hit, n_missed); | |
5974 | ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," | |
04d08d54 | 5975 | " life span: %lldms\n", hmap_count(&backer->subfacets), |
dc54ef36 EJ |
5976 | backer->avg_n_subfacet, backer->max_n_subfacet, |
5977 | backer->avg_subfacet_life); | |
5978 | ||
5979 | minutes = (time_msec() - backer->created) / (1000 * 60); | |
655ab909 | 5980 | if (minutes >= 60) { |
dc54ef36 | 5981 | show_dp_rates(ds, "\thourly avg:", &backer->hourly); |
655ab909 AZ |
5982 | } |
5983 | if (minutes >= 60 * 24) { | |
dc54ef36 | 5984 | show_dp_rates(ds, "\tdaily avg:", &backer->daily); |
655ab909 | 5985 | } |
dc54ef36 | 5986 | show_dp_rates(ds, "\toverall avg:", &backer->lifetime); |
27022416 | 5987 | |
dc54ef36 EJ |
5988 | shash_init(&ofproto_shash); |
5989 | ofprotos = get_ofprotos(&ofproto_shash); | |
5990 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5991 | struct ofproto_dpif *ofproto = ofprotos[i]->data; | |
5992 | const struct shash_node **ports; | |
5993 | size_t j; | |
0a740f48 | 5994 | |
dc54ef36 EJ |
5995 | if (ofproto->backer != backer) { |
5996 | continue; | |
0a740f48 | 5997 | } |
27022416 | 5998 | |
dc54ef36 EJ |
5999 | ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
6000 | ofproto->up.name, ofproto->n_hit, ofproto->n_missed); | |
6001 | ||
6002 | ports = shash_sort(&ofproto->up.port_by_name); | |
6003 | for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { | |
6004 | const struct shash_node *node = ports[j]; | |
6005 | struct ofport *ofport = node->data; | |
6006 | struct smap config; | |
4e022ec0 | 6007 | odp_port_t odp_port; |
27022416 | 6008 | |
dc54ef36 EJ |
6009 | ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), |
6010 | ofport->ofp_port); | |
27022416 | 6011 | |
dc54ef36 | 6012 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); |
4e022ec0 | 6013 | if (odp_port != ODPP_NONE) { |
dc54ef36 EJ |
6014 | ds_put_format(ds, "%"PRIu32":", odp_port); |
6015 | } else { | |
6016 | ds_put_cstr(ds, "none:"); | |
6017 | } | |
27022416 | 6018 | |
dc54ef36 | 6019 | ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); |
27022416 | 6020 | |
dc54ef36 EJ |
6021 | smap_init(&config); |
6022 | if (!netdev_get_config(ofport->netdev, &config)) { | |
6023 | const struct smap_node **nodes; | |
6024 | size_t i; | |
27022416 | 6025 | |
dc54ef36 EJ |
6026 | nodes = smap_sort(&config); |
6027 | for (i = 0; i < smap_count(&config); i++) { | |
6028 | const struct smap_node *node = nodes[i]; | |
6029 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
6030 | node->key, node->value); | |
6031 | } | |
6032 | free(nodes); | |
27022416 | 6033 | } |
dc54ef36 EJ |
6034 | smap_destroy(&config); |
6035 | ||
27022416 | 6036 | ds_put_char(ds, ')'); |
dc54ef36 | 6037 | ds_put_char(ds, '\n'); |
27022416 | 6038 | } |
dc54ef36 | 6039 | free(ports); |
27022416 | 6040 | } |
dc54ef36 EJ |
6041 | shash_destroy(&ofproto_shash); |
6042 | free(ofprotos); | |
27022416 JP |
6043 | } |
6044 | ||
6045 | static void | |
dc54ef36 EJ |
6046 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6047 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
27022416 JP |
6048 | { |
6049 | struct ds ds = DS_EMPTY_INITIALIZER; | |
dc54ef36 EJ |
6050 | const struct shash_node **backers; |
6051 | int i; | |
27022416 | 6052 | |
dc54ef36 EJ |
6053 | backers = shash_sort(&all_dpif_backers); |
6054 | for (i = 0; i < shash_count(&all_dpif_backers); i++) { | |
6055 | dpif_show_backer(backers[i]->data, &ds); | |
27022416 | 6056 | } |
dc54ef36 | 6057 | free(backers); |
27022416 JP |
6058 | |
6059 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6060 | ds_destroy(&ds); | |
6061 | } | |
6062 | ||
bcd2633a JP |
6063 | /* Dump the megaflow (facet) cache. This is useful to check the |
6064 | * correctness of flow wildcarding, since the same mechanism is used for | |
6065 | * both xlate caching and kernel wildcarding. | |
6066 | * | |
6067 | * It's important to note that in the output the flow description uses | |
6068 | * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports. | |
6069 | * | |
6070 | * This command is only needed for advanced debugging, so it's not | |
6071 | * documented in the man page. */ | |
6072 | static void | |
6073 | ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn, | |
6074 | int argc OVS_UNUSED, const char *argv[], | |
6075 | void *aux OVS_UNUSED) | |
6076 | { | |
6077 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6078 | const struct ofproto_dpif *ofproto; | |
6079 | long long int now = time_msec(); | |
6080 | struct cls_cursor cursor; | |
6081 | struct facet *facet; | |
6082 | ||
6083 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6084 | if (!ofproto) { | |
6085 | unixctl_command_reply_error(conn, "no such bridge"); | |
6086 | return; | |
6087 | } | |
6088 | ||
6089 | cls_cursor_init(&cursor, &ofproto->facets, NULL); | |
6090 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6091 | cls_rule_format(&facet->cr, &ds); | |
6092 | ds_put_cstr(&ds, ", "); | |
7ee9e841 | 6093 | ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets)); |
bcd2633a JP |
6094 | ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); |
6095 | ds_put_cstr(&ds, "Datapath actions: "); | |
8ecafa3e JP |
6096 | if (facet->xout.slow) { |
6097 | uint64_t slow_path_stub[128 / 8]; | |
6098 | const struct nlattr *actions; | |
6099 | size_t actions_len; | |
6100 | ||
6101 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, | |
6102 | slow_path_stub, sizeof slow_path_stub, | |
6103 | &actions, &actions_len); | |
6104 | format_odp_actions(&ds, actions, actions_len); | |
6105 | } else { | |
6106 | format_odp_actions(&ds, facet->xout.odp_actions.data, | |
6107 | facet->xout.odp_actions.size); | |
6108 | } | |
bcd2633a JP |
6109 | ds_put_cstr(&ds, "\n"); |
6110 | } | |
6111 | ||
6112 | ds_chomp(&ds, '\n'); | |
6113 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6114 | ds_destroy(&ds); | |
6115 | } | |
6116 | ||
6b35e630 JP |
6117 | /* Disable using the megaflows. |
6118 | * | |
6119 | * This command is only needed for advanced debugging, so it's not | |
6120 | * documented in the man page. */ | |
6121 | static void | |
6122 | ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn, | |
6123 | int argc OVS_UNUSED, | |
6124 | const char *argv[] OVS_UNUSED, | |
6125 | void *aux OVS_UNUSED) | |
6126 | { | |
6127 | struct ofproto_dpif *ofproto; | |
6128 | ||
6129 | enable_megaflows = false; | |
6130 | ||
6131 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6132 | flush(&ofproto->up); | |
6133 | } | |
6134 | ||
6135 | unixctl_command_reply(conn, "megaflows disabled"); | |
6136 | } | |
6137 | ||
6138 | /* Re-enable using megaflows. | |
6139 | * | |
6140 | * This command is only needed for advanced debugging, so it's not | |
6141 | * documented in the man page. */ | |
6142 | static void | |
6143 | ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn, | |
6144 | int argc OVS_UNUSED, | |
6145 | const char *argv[] OVS_UNUSED, | |
6146 | void *aux OVS_UNUSED) | |
6147 | { | |
6148 | struct ofproto_dpif *ofproto; | |
6149 | ||
6150 | enable_megaflows = true; | |
6151 | ||
6152 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6153 | flush(&ofproto->up); | |
6154 | } | |
6155 | ||
6156 | unixctl_command_reply(conn, "megaflows enabled"); | |
6157 | } | |
6158 | ||
27022416 JP |
6159 | static void |
6160 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
6161 | int argc OVS_UNUSED, const char *argv[], | |
6162 | void *aux OVS_UNUSED) | |
6163 | { | |
6164 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6165 | const struct ofproto_dpif *ofproto; | |
6166 | struct subfacet *subfacet; | |
6167 | ||
6168 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6169 | if (!ofproto) { | |
6170 | unixctl_command_reply_error(conn, "no such bridge"); | |
6171 | return; | |
6172 | } | |
6173 | ||
af37354d EJ |
6174 | update_stats(ofproto->backer); |
6175 | ||
04d08d54 | 6176 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { |
4dff9097 | 6177 | struct facet *facet = subfacet->facet; |
64f9cb76 JP |
6178 | struct odputil_keybuf maskbuf; |
6179 | struct ofpbuf mask; | |
4dff9097 | 6180 | |
b7e2f6b3 | 6181 | if (facet->ofproto != ofproto) { |
04d08d54 EJ |
6182 | continue; |
6183 | } | |
6184 | ||
64f9cb76 JP |
6185 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6186 | if (enable_megaflows) { | |
6187 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
6188 | &facet->flow, UINT32_MAX); | |
6189 | } | |
6190 | ||
6191 | odp_flow_format(subfacet->key, subfacet->key_len, | |
6192 | mask.data, mask.size, &ds); | |
27022416 JP |
6193 | |
6194 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
6195 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
6196 | if (subfacet->used) { | |
6197 | ds_put_format(&ds, "%.3fs", | |
6198 | (time_msec() - subfacet->used) / 1000.0); | |
6199 | } else { | |
6200 | ds_put_format(&ds, "never"); | |
6201 | } | |
6202 | if (subfacet->facet->tcp_flags) { | |
6203 | ds_put_cstr(&ds, ", flags:"); | |
6204 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
6205 | } | |
6206 | ||
6207 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 6208 | if (facet->xout.slow) { |
f2245da3 JP |
6209 | uint64_t slow_path_stub[128 / 8]; |
6210 | const struct nlattr *actions; | |
6211 | size_t actions_len; | |
6212 | ||
bbafd73b | 6213 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
6214 | slow_path_stub, sizeof slow_path_stub, |
6215 | &actions, &actions_len); | |
6216 | format_odp_actions(&ds, actions, actions_len); | |
6217 | } else { | |
bbafd73b EJ |
6218 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
6219 | facet->xout.odp_actions.size); | |
f2245da3 | 6220 | } |
27022416 JP |
6221 | ds_put_char(&ds, '\n'); |
6222 | } | |
6223 | ||
6224 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6225 | ds_destroy(&ds); | |
6226 | } | |
6227 | ||
6228 | static void | |
6229 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
6230 | int argc OVS_UNUSED, const char *argv[], | |
6231 | void *aux OVS_UNUSED) | |
6232 | { | |
6233 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6234 | struct ofproto_dpif *ofproto; | |
6235 | ||
6236 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6237 | if (!ofproto) { | |
6238 | unixctl_command_reply_error(conn, "no such bridge"); | |
6239 | return; | |
6240 | } | |
6241 | ||
6242 | flush(&ofproto->up); | |
6243 | ||
6244 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6245 | ds_destroy(&ds); | |
6246 | } | |
6247 | ||
abe529af BP |
6248 | static void |
6249 | ofproto_dpif_unixctl_init(void) | |
6250 | { | |
6251 | static bool registered; | |
6252 | if (registered) { | |
6253 | return; | |
6254 | } | |
6255 | registered = true; | |
6256 | ||
0e15264f BP |
6257 | unixctl_command_register( |
6258 | "ofproto/trace", | |
50aa28fd AW |
6259 | "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", |
6260 | 1, 3, ofproto_unixctl_trace, NULL); | |
96e466a3 | 6261 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6262 | ofproto_unixctl_fdb_flush, NULL); |
6263 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6264 | ofproto_unixctl_fdb_show, NULL); | |
6265 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6266 | ofproto_dpif_clog, NULL); | |
6267 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6268 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6269 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6270 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
6271 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
6272 | ofproto_unixctl_dpif_dump_dps, NULL); | |
dc54ef36 EJ |
6273 | unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show, |
6274 | NULL); | |
27022416 JP |
6275 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, |
6276 | ofproto_unixctl_dpif_dump_flows, NULL); | |
6277 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
6278 | ofproto_unixctl_dpif_del_flows, NULL); | |
bcd2633a JP |
6279 | unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1, |
6280 | ofproto_unixctl_dpif_dump_megaflows, NULL); | |
6b35e630 JP |
6281 | unixctl_command_register("dpif/disable-megaflows", "", 0, 0, |
6282 | ofproto_unixctl_dpif_disable_megaflows, NULL); | |
6283 | unixctl_command_register("dpif/enable-megaflows", "", 0, 0, | |
6284 | ofproto_unixctl_dpif_enable_megaflows, NULL); | |
abe529af BP |
6285 | } |
6286 | \f | |
52a90c29 BP |
6287 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6288 | * | |
6289 | * This is deprecated. It is only for compatibility with broken device drivers | |
6290 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6291 | * devices are not used. When broken device drivers are no longer in | |
6292 | * widespread use, we will delete these interfaces. */ | |
6293 | ||
6294 | static int | |
4e022ec0 | 6295 | set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6296 | { |
6297 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6298 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6299 | ||
6300 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6301 | && vid == ofport->vlandev_vid) { | |
6302 | return 0; | |
6303 | } | |
6304 | ||
2cc3c58e | 6305 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
6306 | |
6307 | if (ofport->realdev_ofp_port) { | |
6308 | vsp_remove(ofport); | |
6309 | } | |
6310 | if (realdev_ofp_port && ofport->bundle) { | |
6311 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6312 | * themselves be part of a bundle. */ | |
6313 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6314 | } | |
6315 | ||
6316 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6317 | ofport->vlandev_vid = vid; | |
6318 | ||
6319 | if (realdev_ofp_port) { | |
6320 | vsp_add(ofport, realdev_ofp_port, vid); | |
6321 | } | |
6322 | ||
6323 | return 0; | |
6324 | } | |
6325 | ||
6326 | static uint32_t | |
4e022ec0 | 6327 | hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid) |
52a90c29 | 6328 | { |
4e022ec0 | 6329 | return hash_2words(ofp_to_u16(realdev_ofp_port), vid); |
52a90c29 BP |
6330 | } |
6331 | ||
46c88433 EJ |
6332 | bool |
6333 | ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto) | |
6334 | { | |
6335 | return !hmap_is_empty(&ofproto->realdev_vid_map); | |
6336 | } | |
6337 | ||
deea1200 AW |
6338 | /* Returns the OFP port number of the Linux VLAN device that corresponds to |
6339 | * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in | |
6340 | * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and | |
6341 | * 'vlan_tci' 9, it would return the port number of eth0.9. | |
40e05935 | 6342 | * |
deea1200 AW |
6343 | * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this |
6344 | * function just returns its 'realdev_ofp_port' argument. */ | |
4e022ec0 | 6345 | ofp_port_t |
52a90c29 | 6346 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, |
4e022ec0 | 6347 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) |
52a90c29 BP |
6348 | { |
6349 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
52a90c29 BP |
6350 | int vid = vlan_tci_to_vid(vlan_tci); |
6351 | const struct vlan_splinter *vsp; | |
6352 | ||
6353 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6354 | hash_realdev_vid(realdev_ofp_port, vid), | |
6355 | &ofproto->realdev_vid_map) { | |
6356 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6357 | && vsp->vid == vid) { | |
deea1200 | 6358 | return vsp->vlandev_ofp_port; |
52a90c29 BP |
6359 | } |
6360 | } | |
6361 | } | |
deea1200 | 6362 | return realdev_ofp_port; |
52a90c29 BP |
6363 | } |
6364 | ||
6365 | static struct vlan_splinter * | |
4e022ec0 | 6366 | vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) |
52a90c29 BP |
6367 | { |
6368 | struct vlan_splinter *vsp; | |
6369 | ||
4e022ec0 | 6370 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, |
f9c0c3ec | 6371 | hash_ofp_port(vlandev_ofp_port), |
52a90c29 BP |
6372 | &ofproto->vlandev_map) { |
6373 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6374 | return vsp; | |
6375 | } | |
6376 | } | |
6377 | ||
6378 | return NULL; | |
6379 | } | |
6380 | ||
40e05935 BP |
6381 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
6382 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
6383 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
6384 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
6385 | * eth0 and store 9 in '*vid'. | |
6386 | * | |
6387 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
6388 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
6389 | * always does.*/ | |
4e022ec0 | 6390 | static ofp_port_t |
52a90c29 | 6391 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, |
4e022ec0 | 6392 | ofp_port_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
6393 | { |
6394 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6395 | const struct vlan_splinter *vsp; | |
6396 | ||
6397 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6398 | if (vsp) { | |
6399 | if (vid) { | |
6400 | *vid = vsp->vid; | |
6401 | } | |
6402 | return vsp->realdev_ofp_port; | |
6403 | } | |
6404 | } | |
6405 | return 0; | |
6406 | } | |
6407 | ||
b98d8985 BP |
6408 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
6409 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
6410 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
6411 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
6412 | * always the case unless VLAN splinters are enabled), returns false without | |
6413 | * making any changes. */ | |
6414 | static bool | |
6415 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
6416 | { | |
4e022ec0 | 6417 | ofp_port_t realdev; |
b98d8985 BP |
6418 | int vid; |
6419 | ||
4e022ec0 | 6420 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid); |
b98d8985 BP |
6421 | if (!realdev) { |
6422 | return false; | |
6423 | } | |
6424 | ||
6425 | /* Cause the flow to be processed as if it came in on the real device with | |
6426 | * the VLAN device's VLAN ID. */ | |
4e022ec0 | 6427 | flow->in_port.ofp_port = realdev; |
b98d8985 BP |
6428 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); |
6429 | return true; | |
6430 | } | |
6431 | ||
52a90c29 BP |
6432 | static void |
6433 | vsp_remove(struct ofport_dpif *port) | |
6434 | { | |
6435 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6436 | struct vlan_splinter *vsp; | |
6437 | ||
6438 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
6439 | if (vsp) { | |
6440 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6441 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6442 | free(vsp); | |
6443 | ||
6444 | port->realdev_ofp_port = 0; | |
6445 | } else { | |
6446 | VLOG_ERR("missing vlan device record"); | |
6447 | } | |
6448 | } | |
6449 | ||
6450 | static void | |
4e022ec0 | 6451 | vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6452 | { |
6453 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6454 | ||
6455 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
6456 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
6457 | == realdev_ofp_port)) { | |
6458 | struct vlan_splinter *vsp; | |
6459 | ||
6460 | vsp = xmalloc(sizeof *vsp); | |
6461 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
f9c0c3ec | 6462 | hash_ofp_port(port->up.ofp_port)); |
52a90c29 BP |
6463 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, |
6464 | hash_realdev_vid(realdev_ofp_port, vid)); | |
6465 | vsp->realdev_ofp_port = realdev_ofp_port; | |
6466 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6467 | vsp->vid = vid; | |
6468 | ||
6469 | port->realdev_ofp_port = realdev_ofp_port; | |
6470 | } else { | |
6471 | VLOG_ERR("duplicate vlan device record"); | |
6472 | } | |
6473 | } | |
e1b1d06a | 6474 | |
46c88433 | 6475 | static odp_port_t |
4e022ec0 | 6476 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
e1b1d06a JP |
6477 | { |
6478 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
4e022ec0 | 6479 | return ofport ? ofport->odp_port : ODPP_NONE; |
e1b1d06a JP |
6480 | } |
6481 | ||
acf60855 | 6482 | static struct ofport_dpif * |
4e022ec0 | 6483 | odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port) |
e1b1d06a JP |
6484 | { |
6485 | struct ofport_dpif *port; | |
6486 | ||
f9c0c3ec | 6487 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port), |
acf60855 | 6488 | &backer->odp_to_ofport_map) { |
e1b1d06a | 6489 | if (port->odp_port == odp_port) { |
acf60855 | 6490 | return port; |
e1b1d06a JP |
6491 | } |
6492 | } | |
6493 | ||
acf60855 JP |
6494 | return NULL; |
6495 | } | |
6496 | ||
4e022ec0 AW |
6497 | static ofp_port_t |
6498 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) | |
acf60855 JP |
6499 | { |
6500 | struct ofport_dpif *port; | |
6501 | ||
6502 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 6503 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
6504 | return port->up.ofp_port; |
6505 | } else { | |
6506 | return OFPP_NONE; | |
6507 | } | |
e1b1d06a | 6508 | } |
655ab909 | 6509 | |
655ab909 AZ |
6510 | /* Compute exponentially weighted moving average, adding 'new' as the newest, |
6511 | * most heavily weighted element. 'base' designates the rate of decay: after | |
6512 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
6513 | * (about .37). */ | |
6514 | static void | |
6515 | exp_mavg(double *avg, int base, double new) | |
6516 | { | |
6517 | *avg = (*avg * (base - 1) + new) / base; | |
6518 | } | |
6519 | ||
6520 | static void | |
dc54ef36 | 6521 | update_moving_averages(struct dpif_backer *backer) |
655ab909 AZ |
6522 | { |
6523 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
dc54ef36 EJ |
6524 | long long int minutes = (time_msec() - backer->created) / min_ms; |
6525 | ||
6526 | if (minutes > 0) { | |
6527 | backer->lifetime.add_rate = (double) backer->total_subfacet_add_count | |
6528 | / minutes; | |
6529 | backer->lifetime.del_rate = (double) backer->total_subfacet_del_count | |
6530 | / minutes; | |
6531 | } else { | |
6532 | backer->lifetime.add_rate = 0.0; | |
6533 | backer->lifetime.del_rate = 0.0; | |
6534 | } | |
655ab909 AZ |
6535 | |
6536 | /* Update hourly averages on the minute boundaries. */ | |
dc54ef36 EJ |
6537 | if (time_msec() - backer->last_minute >= min_ms) { |
6538 | exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count); | |
6539 | exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count); | |
655ab909 AZ |
6540 | |
6541 | /* Update daily averages on the hour boundaries. */ | |
dc54ef36 EJ |
6542 | if ((backer->last_minute - backer->created) / min_ms % 60 == 59) { |
6543 | exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate); | |
6544 | exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate); | |
655ab909 AZ |
6545 | } |
6546 | ||
dc54ef36 EJ |
6547 | backer->total_subfacet_add_count += backer->subfacet_add_count; |
6548 | backer->total_subfacet_del_count += backer->subfacet_del_count; | |
6549 | backer->subfacet_add_count = 0; | |
6550 | backer->subfacet_del_count = 0; | |
6551 | backer->last_minute += min_ms; | |
655ab909 AZ |
6552 | } |
6553 | } | |
e1b1d06a | 6554 | |
abe529af | 6555 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 6556 | init, |
abe529af BP |
6557 | enumerate_types, |
6558 | enumerate_names, | |
6559 | del, | |
0aeaabc8 | 6560 | port_open_type, |
acf60855 JP |
6561 | type_run, |
6562 | type_run_fast, | |
6563 | type_wait, | |
abe529af BP |
6564 | alloc, |
6565 | construct, | |
6566 | destruct, | |
6567 | dealloc, | |
6568 | run, | |
5fcc0d00 | 6569 | run_fast, |
abe529af | 6570 | wait, |
0d085684 | 6571 | get_memory_usage, |
abe529af | 6572 | flush, |
6c1491fb BP |
6573 | get_features, |
6574 | get_tables, | |
abe529af BP |
6575 | port_alloc, |
6576 | port_construct, | |
6577 | port_destruct, | |
6578 | port_dealloc, | |
6579 | port_modified, | |
6580 | port_reconfigured, | |
6581 | port_query_by_name, | |
6582 | port_add, | |
6583 | port_del, | |
6527c598 | 6584 | port_get_stats, |
abe529af BP |
6585 | port_dump_start, |
6586 | port_dump_next, | |
6587 | port_dump_done, | |
6588 | port_poll, | |
6589 | port_poll_wait, | |
6590 | port_is_lacp_current, | |
0ab6decf | 6591 | NULL, /* rule_choose_table */ |
abe529af BP |
6592 | rule_alloc, |
6593 | rule_construct, | |
6594 | rule_destruct, | |
6595 | rule_dealloc, | |
abe529af BP |
6596 | rule_get_stats, |
6597 | rule_execute, | |
6598 | rule_modify_actions, | |
7257b535 | 6599 | set_frag_handling, |
abe529af BP |
6600 | packet_out, |
6601 | set_netflow, | |
6602 | get_netflow_ids, | |
6603 | set_sflow, | |
29089a54 | 6604 | set_ipfix, |
abe529af | 6605 | set_cfm, |
9a9e3786 | 6606 | get_cfm_status, |
ccc09689 EJ |
6607 | set_bfd, |
6608 | get_bfd_status, | |
21f7563c JP |
6609 | set_stp, |
6610 | get_stp_status, | |
6611 | set_stp_port, | |
6612 | get_stp_port_status, | |
8b36f51e | 6613 | set_queues, |
abe529af BP |
6614 | bundle_set, |
6615 | bundle_remove, | |
ec7ceaed EJ |
6616 | mirror_set__, |
6617 | mirror_get_stats__, | |
abe529af BP |
6618 | set_flood_vlans, |
6619 | is_mirror_output_bundle, | |
8402c74b | 6620 | forward_bpdu_changed, |
c4069512 | 6621 | set_mac_table_config, |
52a90c29 | 6622 | set_realdev, |
9cae45dc JR |
6623 | NULL, /* meter_get_features */ |
6624 | NULL, /* meter_set */ | |
6625 | NULL, /* meter_get */ | |
6626 | NULL, /* meter_del */ | |
abe529af | 6627 | }; |