]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e09ee259 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
9583bc14 | 19 | #include "ofproto/ofproto-dpif.h" |
5bee6e26 | 20 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
21 | |
22 | #include <errno.h> | |
23 | ||
ccc09689 | 24 | #include "bfd.h" |
abe529af | 25 | #include "bond.h" |
daff3353 | 26 | #include "bundle.h" |
abe529af BP |
27 | #include "byte-order.h" |
28 | #include "connmgr.h" | |
29 | #include "coverage.h" | |
30 | #include "cfm.h" | |
31 | #include "dpif.h" | |
32 | #include "dynamic-string.h" | |
33 | #include "fail-open.h" | |
05067881 | 34 | #include "guarded-list.h" |
abe529af BP |
35 | #include "hmapx.h" |
36 | #include "lacp.h" | |
75a75043 | 37 | #include "learn.h" |
abe529af | 38 | #include "mac-learning.h" |
816fd533 | 39 | #include "meta-flow.h" |
abe529af | 40 | #include "multipath.h" |
0a740f48 | 41 | #include "netdev-vport.h" |
abe529af BP |
42 | #include "netdev.h" |
43 | #include "netlink.h" | |
44 | #include "nx-match.h" | |
45 | #include "odp-util.h" | |
1ac7c9bd | 46 | #include "odp-execute.h" |
abe529af BP |
47 | #include "ofp-util.h" |
48 | #include "ofpbuf.h" | |
f25d0cf3 | 49 | #include "ofp-actions.h" |
31a19d69 | 50 | #include "ofp-parse.h" |
abe529af | 51 | #include "ofp-print.h" |
9d6ac44e | 52 | #include "ofproto-dpif-governor.h" |
29089a54 | 53 | #include "ofproto-dpif-ipfix.h" |
ec7ceaed | 54 | #include "ofproto-dpif-mirror.h" |
635c5db9 | 55 | #include "ofproto-dpif-monitor.h" |
bae473fe | 56 | #include "ofproto-dpif-sflow.h" |
e1ec7dd4 | 57 | #include "ofproto-dpif-upcall.h" |
9583bc14 | 58 | #include "ofproto-dpif-xlate.h" |
abe529af | 59 | #include "poll-loop.h" |
0d085684 | 60 | #include "simap.h" |
27022416 | 61 | #include "smap.h" |
abe529af | 62 | #include "timer.h" |
b9ad7294 | 63 | #include "tunnel.h" |
6c1491fb | 64 | #include "unaligned.h" |
abe529af BP |
65 | #include "unixctl.h" |
66 | #include "vlan-bitmap.h" | |
67 | #include "vlog.h" | |
68 | ||
69 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
70 | ||
abe529af | 71 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
72 | COVERAGE_DEFINE(facet_revalidate); |
73 | COVERAGE_DEFINE(facet_unexpected); | |
264ac358 AW |
74 | COVERAGE_DEFINE(facet_create); |
75 | COVERAGE_DEFINE(facet_remove); | |
76 | COVERAGE_DEFINE(subfacet_create); | |
77 | COVERAGE_DEFINE(subfacet_destroy); | |
9b1a48c2 | 78 | COVERAGE_DEFINE(subfacet_install_fail); |
ada3a58d | 79 | COVERAGE_DEFINE(packet_in_overflow); |
abe529af | 80 | |
46c88433 EJ |
81 | /* Number of implemented OpenFlow tables. */ |
82 | enum { N_TABLES = 255 }; | |
83 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
84 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
85 | ||
a088a1ff | 86 | struct flow_miss; |
ac35f9c8 | 87 | struct facet; |
abe529af | 88 | |
70742c7f EJ |
89 | struct rule_dpif { |
90 | struct rule up; | |
91 | ||
92 | /* These statistics: | |
93 | * | |
94 | * - Do include packets and bytes from facets that have been deleted or | |
95 | * whose own statistics have been folded into the rule. | |
96 | * | |
97 | * - Do include packets and bytes sent "by hand" that were accounted to | |
98 | * the rule without any facet being involved (this is a rare corner | |
99 | * case in rule_execute()). | |
100 | * | |
101 | * - Do not include packet or bytes that can be obtained from any facet's | |
102 | * packet_count or byte_count member or that can be obtained from the | |
103 | * datapath by, e.g., dpif_flow_get() for any subfacet. | |
104 | */ | |
105 | struct ovs_mutex stats_mutex; | |
106 | uint64_t packet_count OVS_GUARDED; /* Number of packets received. */ | |
107 | uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */ | |
108 | }; | |
109 | ||
7e741ffb | 110 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
70742c7f | 111 | static struct rule_dpif *rule_dpif_cast(const struct rule *); |
b0f7b9b5 | 112 | |
00430a3f SH |
113 | struct group_dpif { |
114 | struct ofgroup up; | |
115 | ||
116 | /* These statistics: | |
117 | * | |
118 | * - Do include packets and bytes from facets that have been deleted or | |
119 | * whose own statistics have been folded into the rule. | |
120 | * | |
121 | * - Do include packets and bytes sent "by hand" that were accounted to | |
122 | * the rule without any facet being involved (this is a rare corner | |
123 | * case in rule_execute()). | |
124 | * | |
125 | * - Do not include packet or bytes that can be obtained from any facet's | |
126 | * packet_count or byte_count member or that can be obtained from the | |
127 | * datapath by, e.g., dpif_flow_get() for any subfacet. | |
128 | */ | |
129 | struct ovs_mutex stats_mutex; | |
130 | uint64_t packet_count OVS_GUARDED; /* Number of packets received. */ | |
131 | uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */ | |
132 | struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */ | |
133 | }; | |
134 | ||
46c88433 EJ |
135 | struct ofbundle { |
136 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
137 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
138 | void *aux; /* Key supplied by ofproto's client. */ | |
139 | char *name; /* Identifier for log messages. */ | |
140 | ||
141 | /* Configuration. */ | |
142 | struct list ports; /* Contains "struct ofport"s. */ | |
143 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
144 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
145 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
146 | * NULL if all VLANs are trunked. */ | |
147 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
148 | struct bond *bond; /* Nonnull iff more than one port. */ | |
149 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
150 | ||
151 | /* Status. */ | |
152 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
153 | }; | |
154 | ||
abe529af | 155 | static void bundle_remove(struct ofport *); |
7bde8dd8 | 156 | static void bundle_update(struct ofbundle *); |
abe529af BP |
157 | static void bundle_destroy(struct ofbundle *); |
158 | static void bundle_del_port(struct ofport_dpif *); | |
159 | static void bundle_run(struct ofbundle *); | |
160 | static void bundle_wait(struct ofbundle *); | |
33158a18 | 161 | |
21f7563c JP |
162 | static void stp_run(struct ofproto_dpif *ofproto); |
163 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
164 | static int set_stp_port(struct ofport *, |
165 | const struct ofproto_port_stp_settings *); | |
21f7563c | 166 | |
6a7e895f BP |
167 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, |
168 | enum slow_path_reason, | |
169 | uint64_t *stub, size_t stub_size, | |
170 | const struct nlattr **actionsp, | |
171 | size_t *actions_lenp); | |
172 | ||
173 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
174 | * states: | |
175 | * | |
176 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
177 | * case just after the subfacet is created, just before the subfacet is | |
178 | * destroyed, or if the datapath returns an error when we try to install a | |
179 | * subfacet. | |
180 | * | |
181 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
182 | * | |
183 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
184 | * ofproto_dpif is installed in the datapath. | |
185 | */ | |
186 | enum subfacet_path { | |
187 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
188 | SF_FAST_PATH, /* Full actions are installed. */ | |
189 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
190 | }; | |
191 | ||
5f5fbd17 BP |
192 | /* A dpif flow and actions associated with a facet. |
193 | * | |
194 | * See also the large comment on struct facet. */ | |
195 | struct subfacet { | |
196 | /* Owners. */ | |
197 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
198 | struct list list_node; /* In struct facet's 'facets' list. */ | |
199 | struct facet *facet; /* Owning facet. */ | |
04d08d54 | 200 | struct dpif_backer *backer; /* Owning backer. */ |
5f5fbd17 | 201 | |
5f5fbd17 BP |
202 | struct nlattr *key; |
203 | int key_len; | |
204 | ||
205 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 206 | long long int created; /* Time created. */ |
5f5fbd17 BP |
207 | |
208 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
209 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
210 | ||
6a7e895f | 211 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 BP |
212 | }; |
213 | ||
1d85f9e5 JP |
214 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
215 | ||
f9c37233 BP |
216 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *, |
217 | uint32_t key_hash); | |
04d08d54 | 218 | static struct subfacet *subfacet_find(struct dpif_backer *, |
acf60855 | 219 | const struct nlattr *key, size_t key_len, |
9566abf9 | 220 | uint32_t key_hash); |
5f5fbd17 BP |
221 | static void subfacet_destroy(struct subfacet *); |
222 | static void subfacet_destroy__(struct subfacet *); | |
04d08d54 | 223 | static void subfacet_destroy_batch(struct dpif_backer *, |
1d85f9e5 | 224 | struct subfacet **, int n); |
5f5fbd17 BP |
225 | static void subfacet_reset_dp_stats(struct subfacet *, |
226 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
227 | static void subfacet_update_stats(struct subfacet *, |
228 | const struct dpif_flow_stats *); | |
5f5fbd17 | 229 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
230 | const struct ofpbuf *odp_actions, |
231 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
232 | static void subfacet_uninstall(struct subfacet *); |
233 | ||
bcd2633a | 234 | /* A unique, non-overlapping instantiation of an OpenFlow flow. |
b0f7b9b5 BP |
235 | * |
236 | * A facet associates a "struct flow", which represents the Open vSwitch | |
bcd2633a JP |
237 | * userspace idea of an exact-match flow, with one or more subfacets. |
238 | * While the facet is created based on an exact-match flow, it is stored | |
239 | * within the ofproto based on the wildcards that could be expressed | |
240 | * based on the flow table and other configuration. (See the 'wc' | |
241 | * description in "struct xlate_out" for more details.) | |
b0f7b9b5 | 242 | * |
bcd2633a JP |
243 | * Each subfacet tracks the datapath's idea of the flow equivalent to |
244 | * the facet. When the kernel module (or other dpif implementation) and | |
245 | * Open vSwitch userspace agree on the definition of a flow key, there | |
246 | * is exactly one subfacet per facet. If the dpif implementation | |
247 | * supports more-specific flow matching than userspace, however, a facet | |
248 | * can have more than one subfacet. Examples include the dpif | |
249 | * implementation not supporting the same wildcards as userspace or some | |
250 | * distinction in flow that userspace simply doesn't understand. | |
251 | * | |
252 | * Flow expiration works in terms of subfacets, so a facet must have at | |
253 | * least one subfacet or it will never expire, leaking memory. */ | |
abe529af | 254 | struct facet { |
74432d57 | 255 | /* Owner. */ |
b7e2f6b3 | 256 | struct ofproto_dpif *ofproto; |
b0f7b9b5 BP |
257 | |
258 | /* Owned data. */ | |
259 | struct list subfacets; | |
abe529af BP |
260 | long long int used; /* Time last used; time created if not used. */ |
261 | ||
b0f7b9b5 | 262 | /* Key. */ |
bcd2633a JP |
263 | struct flow flow; /* Flow of the creating subfacet. */ |
264 | struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */ | |
b0f7b9b5 | 265 | |
abe529af BP |
266 | /* These statistics: |
267 | * | |
268 | * - Do include packets and bytes sent "by hand", e.g. with | |
269 | * dpif_execute(). | |
270 | * | |
271 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 272 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 273 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
274 | * |
275 | * - Do not include packets or bytes that can be obtained from the | |
276 | * datapath for any existing subfacet. | |
abe529af BP |
277 | */ |
278 | uint64_t packet_count; /* Number of packets received. */ | |
279 | uint64_t byte_count; /* Number of bytes received. */ | |
280 | ||
b0f7b9b5 | 281 | /* Resubmit statistics. */ |
9d24de3b JP |
282 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
283 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
284 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 285 | |
b0f7b9b5 | 286 | /* Accounting. */ |
907a4c5e | 287 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 288 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
a66733a8 | 289 | uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 290 | |
bbafd73b | 291 | struct xlate_out xout; |
4dff9097 | 292 | |
26cd7e34 BP |
293 | /* Storage for a single subfacet, to reduce malloc() time and space |
294 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
295 | * case has exactly one subfacet. However, 'one_subfacet' may not |
296 | * always be valid, since it could have been removed after newer | |
297 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 298 | struct subfacet one_subfacet; |
6cf474d7 EJ |
299 | |
300 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
301 | }; |
302 | ||
e1ec7dd4 | 303 | static struct facet *facet_create(const struct flow_miss *); |
15baa734 | 304 | static void facet_remove(struct facet *); |
abe529af BP |
305 | static void facet_free(struct facet *); |
306 | ||
bcd2633a | 307 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); |
abe529af | 308 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
bcd2633a | 309 | const struct flow *); |
5bf64ade | 310 | static bool facet_revalidate(struct facet *); |
6814e51f | 311 | static bool facet_check_consistency(struct facet *); |
abe529af | 312 | |
15baa734 | 313 | static void facet_flush_stats(struct facet *); |
abe529af | 314 | |
bbb5d219 | 315 | static void facet_reset_counters(struct facet *); |
e1ec7dd4 EJ |
316 | static void flow_push_stats(struct ofproto_dpif *, struct flow *, |
317 | struct dpif_flow_stats *, bool may_learn); | |
9dfb1f78 | 318 | static void facet_push_stats(struct facet *, bool may_learn); |
3de9590b BP |
319 | static void facet_learn(struct facet *); |
320 | static void facet_account(struct facet *); | |
8844e035 | 321 | static void push_all_stats(void); |
abe529af BP |
322 | |
323 | static bool facet_is_controller_flow(struct facet *); | |
324 | ||
46c88433 EJ |
325 | struct ofport_dpif { |
326 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ | |
327 | struct ofport up; | |
328 | ||
329 | odp_port_t odp_port; | |
330 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
331 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
332 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
333 | struct bfd *bfd; /* BFD, if any. */ | |
46c88433 EJ |
334 | bool may_enable; /* May be enabled in bonds. */ |
335 | bool is_tunnel; /* This port is a tunnel. */ | |
a6363cfd | 336 | bool is_layer3; /* This is a layer 3 port. */ |
46c88433 EJ |
337 | long long int carrier_seq; /* Carrier status changes. */ |
338 | struct ofport_dpif *peer; /* Peer if patch port. */ | |
339 | ||
340 | /* Spanning tree. */ | |
341 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
342 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
343 | long long int stp_state_entered; | |
344 | ||
55954f6e EJ |
345 | /* Queue to DSCP mapping. */ |
346 | struct ofproto_port_queue *qdscp; | |
347 | size_t n_qdscp; | |
46c88433 EJ |
348 | |
349 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
350 | * | |
351 | * This is deprecated. It is only for compatibility with broken device | |
352 | * drivers in old versions of Linux that do not properly support VLANs when | |
353 | * VLAN devices are not used. When broken device drivers are no longer in | |
354 | * widespread use, we will delete these interfaces. */ | |
355 | ofp_port_t realdev_ofp_port; | |
356 | int vlandev_vid; | |
357 | }; | |
358 | ||
52a90c29 BP |
359 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
360 | * | |
361 | * This is deprecated. It is only for compatibility with broken device drivers | |
362 | * in old versions of Linux that do not properly support VLANs when VLAN | |
363 | * devices are not used. When broken device drivers are no longer in | |
364 | * widespread use, we will delete these interfaces. */ | |
365 | struct vlan_splinter { | |
366 | struct hmap_node realdev_vid_node; | |
367 | struct hmap_node vlandev_node; | |
4e022ec0 AW |
368 | ofp_port_t realdev_ofp_port; |
369 | ofp_port_t vlandev_ofp_port; | |
52a90c29 BP |
370 | int vid; |
371 | }; | |
372 | ||
52a90c29 | 373 | static void vsp_remove(struct ofport_dpif *); |
4e022ec0 | 374 | static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid); |
52a90c29 | 375 | |
46c88433 EJ |
376 | static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
377 | ofp_port_t); | |
378 | ||
4e022ec0 | 379 | static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *, |
46c88433 | 380 | odp_port_t); |
e1b1d06a | 381 | |
abe529af BP |
382 | static struct ofport_dpif * |
383 | ofport_dpif_cast(const struct ofport *ofport) | |
384 | { | |
abe529af BP |
385 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
386 | } | |
387 | ||
388 | static void port_run(struct ofport_dpif *); | |
8aee94b6 | 389 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 390 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
6cbbf4fa | 391 | static void ofport_update_peer(struct ofport_dpif *); |
8fa4d1d0 | 392 | static void run_fast_rl(void); |
3d9c5e58 | 393 | static int run_fast(struct ofproto *); |
abe529af | 394 | |
7ee20df1 BP |
395 | struct dpif_completion { |
396 | struct list list_node; | |
397 | struct ofoperation *op; | |
398 | }; | |
399 | ||
cdc3ab65 EJ |
400 | /* Reasons that we might need to revalidate every facet, and corresponding |
401 | * coverage counters. | |
402 | * | |
403 | * A value of 0 means that there is no need to revalidate. | |
404 | * | |
405 | * It would be nice to have some cleaner way to integrate with coverage | |
406 | * counters, but with only a few reasons I guess this is good enough for | |
407 | * now. */ | |
408 | enum revalidate_reason { | |
409 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
410 | REV_STP, /* Spanning tree protocol port status change. */ | |
4a1b8f30 | 411 | REV_BOND, /* Bonding changed. */ |
cdc3ab65 EJ |
412 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ |
413 | REV_FLOW_TABLE, /* Flow table changed. */ | |
30618594 | 414 | REV_MAC_LEARNING, /* Mac learning changed. */ |
cdc3ab65 EJ |
415 | REV_INCONSISTENCY /* Facet self-check failed. */ |
416 | }; | |
3c4a309c BP |
417 | COVERAGE_DEFINE(rev_reconfigure); |
418 | COVERAGE_DEFINE(rev_stp); | |
4a1b8f30 | 419 | COVERAGE_DEFINE(rev_bond); |
3c4a309c BP |
420 | COVERAGE_DEFINE(rev_port_toggled); |
421 | COVERAGE_DEFINE(rev_flow_table); | |
30618594 | 422 | COVERAGE_DEFINE(rev_mac_learning); |
3c4a309c BP |
423 | COVERAGE_DEFINE(rev_inconsistency); |
424 | ||
cdc3ab65 EJ |
425 | struct avg_subfacet_rates { |
426 | double add_rate; /* Moving average of new flows created per minute. */ | |
427 | double del_rate; /* Moving average of flows deleted per minute. */ | |
428 | }; | |
429 | ||
430 | /* All datapaths of a given type share a single dpif backer instance. */ | |
431 | struct dpif_backer { | |
432 | char *type; | |
433 | int refcount; | |
434 | struct dpif *dpif; | |
e1ec7dd4 | 435 | struct udpif *udpif; |
cdc3ab65 | 436 | struct timer next_expiration; |
8449c4d6 EJ |
437 | |
438 | struct ovs_rwlock odp_to_ofport_lock; | |
439 | struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */ | |
cdc3ab65 EJ |
440 | |
441 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ | |
442 | ||
443 | /* Facet revalidation flags applying to facets which use this backer. */ | |
444 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
cdc3ab65 EJ |
445 | |
446 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
447 | bool recv_set_enable; /* Enables or disables receiving packets. */ | |
448 | ||
449 | struct hmap subfacets; | |
450 | struct governor *governor; | |
451 | ||
452 | /* Subfacet statistics. | |
453 | * | |
454 | * These keep track of the total number of subfacets added and deleted and | |
455 | * flow life span. They are useful for computing the flow rates stats | |
456 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
457 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
458 | * performance in new situations. */ | |
459 | long long int created; /* Time when it is created. */ | |
460 | unsigned max_n_subfacet; /* Maximum number of flows */ | |
461 | unsigned avg_n_subfacet; /* Average number of flows. */ | |
462 | long long int avg_subfacet_life; /* Average life span of subfacets. */ | |
463 | ||
448a4b2f AW |
464 | /* Number of upcall handling threads. */ |
465 | unsigned int n_handler_threads; | |
cdc3ab65 EJ |
466 | }; |
467 | ||
acf60855 JP |
468 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ |
469 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
470 | ||
8f73d537 | 471 | static void drop_key_clear(struct dpif_backer *); |
735d7efb | 472 | |
46c88433 EJ |
473 | struct ofproto_dpif { |
474 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
475 | struct ofproto up; | |
476 | struct dpif_backer *backer; | |
477 | ||
478 | /* Special OpenFlow rules. */ | |
479 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
480 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
481 | struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ | |
482 | ||
483 | /* Bridging. */ | |
484 | struct netflow *netflow; | |
485 | struct dpif_sflow *sflow; | |
486 | struct dpif_ipfix *ipfix; | |
487 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
488 | struct mac_learning *ml; | |
489 | bool has_bonded_bundles; | |
490 | struct mbridge *mbridge; | |
491 | ||
492 | /* Facets. */ | |
493 | struct classifier facets; /* Contains 'struct facet's. */ | |
494 | long long int consistency_rl; | |
495 | ||
0da3c61b AW |
496 | struct ovs_mutex stats_mutex; |
497 | struct netdev_stats stats OVS_GUARDED; /* To account packets generated and | |
498 | * consumed in userspace. */ | |
46c88433 EJ |
499 | |
500 | /* Spanning tree. */ | |
501 | struct stp *stp; | |
502 | long long int stp_last_tick; | |
503 | ||
504 | /* VLAN splinters. */ | |
13501305 EJ |
505 | struct ovs_mutex vsp_mutex; |
506 | struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */ | |
507 | struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */ | |
46c88433 EJ |
508 | |
509 | /* Ports. */ | |
510 | struct sset ports; /* Set of standard port names. */ | |
511 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
512 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ | |
513 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
514 | ||
515 | /* Per ofproto's dpif stats. */ | |
516 | uint64_t n_hit; | |
517 | uint64_t n_missed; | |
3d9c5e58 EJ |
518 | |
519 | /* Work queues. */ | |
05067881 | 520 | struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */ |
46c88433 EJ |
521 | }; |
522 | ||
6b35e630 JP |
523 | /* By default, flows in the datapath are wildcarded (megaflows). They |
524 | * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ | |
525 | static bool enable_megaflows = true; | |
526 | ||
b44a10b7 BP |
527 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
528 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
529 | ||
abe529af BP |
530 | static void ofproto_dpif_unixctl_init(void); |
531 | ||
46c88433 EJ |
532 | static inline struct ofproto_dpif * |
533 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
534 | { | |
535 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
536 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
537 | } | |
538 | ||
539 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, | |
540 | ofp_port_t ofp_port); | |
ade6ad9c | 541 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
aee0979b BP |
542 | const struct ofpbuf *packet, |
543 | const struct ofpact[], size_t ofpacts_len, | |
544 | struct ds *); | |
46c88433 | 545 | |
501f8d1f | 546 | /* Upcalls. */ |
e1ec7dd4 | 547 | static void handle_upcalls(struct dpif_backer *); |
abe529af BP |
548 | |
549 | /* Flow expiration. */ | |
acf60855 | 550 | static int expire(struct dpif_backer *); |
abe529af | 551 | |
6fca1ffb BP |
552 | /* NetFlow. */ |
553 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
554 | ||
abe529af BP |
555 | /* Global variables. */ |
556 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
557 | |
558 | /* Initial mappings of port to bridge mappings. */ | |
559 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
46c88433 | 560 | |
1b63b91e BP |
561 | /* Executes 'fm'. The caller retains ownership of 'fm' and everything in |
562 | * it. */ | |
3d9c5e58 | 563 | void |
46c88433 EJ |
564 | ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, |
565 | struct ofputil_flow_mod *fm) | |
566 | { | |
1b63b91e | 567 | ofproto_flow_mod(&ofproto->up, fm); |
46c88433 EJ |
568 | } |
569 | ||
d5ff77e2 YT |
570 | /* Appends 'pin' to the queue of "packet ins" to be sent to the controller. |
571 | * Takes ownership of 'pin' and pin->packet. */ | |
46c88433 EJ |
572 | void |
573 | ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, | |
0fb7792a | 574 | struct ofproto_packet_in *pin) |
46c88433 | 575 | { |
05067881 | 576 | if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) { |
ada3a58d | 577 | COVERAGE_INC(packet_in_overflow); |
0fb7792a | 578 | free(CONST_CAST(void *, pin->up.packet)); |
ada3a58d | 579 | free(pin); |
ada3a58d | 580 | } |
46c88433 | 581 | } |
abe529af BP |
582 | \f |
583 | /* Factory functions. */ | |
584 | ||
b0408fca | 585 | static void |
acf60855 | 586 | init(const struct shash *iface_hints) |
b0408fca | 587 | { |
acf60855 JP |
588 | struct shash_node *node; |
589 | ||
590 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
591 | SHASH_FOR_EACH(node, iface_hints) { | |
592 | const struct iface_hint *orig_hint = node->data; | |
593 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
594 | ||
595 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
596 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
597 | new_hint->ofp_port = orig_hint->ofp_port; | |
598 | ||
599 | shash_add(&init_ofp_ports, node->name, new_hint); | |
600 | } | |
b0408fca JP |
601 | } |
602 | ||
abe529af BP |
603 | static void |
604 | enumerate_types(struct sset *types) | |
605 | { | |
606 | dp_enumerate_types(types); | |
607 | } | |
608 | ||
609 | static int | |
610 | enumerate_names(const char *type, struct sset *names) | |
611 | { | |
acf60855 JP |
612 | struct ofproto_dpif *ofproto; |
613 | ||
614 | sset_clear(names); | |
615 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
616 | if (strcmp(type, ofproto->up.type)) { | |
617 | continue; | |
618 | } | |
619 | sset_add(names, ofproto->up.name); | |
620 | } | |
621 | ||
622 | return 0; | |
abe529af BP |
623 | } |
624 | ||
625 | static int | |
626 | del(const char *type, const char *name) | |
627 | { | |
628 | struct dpif *dpif; | |
629 | int error; | |
630 | ||
631 | error = dpif_open(name, type, &dpif); | |
632 | if (!error) { | |
633 | error = dpif_delete(dpif); | |
634 | dpif_close(dpif); | |
635 | } | |
636 | return error; | |
637 | } | |
638 | \f | |
0aeaabc8 JP |
639 | static const char * |
640 | port_open_type(const char *datapath_type, const char *port_type) | |
641 | { | |
642 | return dpif_port_open_type(datapath_type, port_type); | |
643 | } | |
644 | ||
acf60855 JP |
645 | /* Type functions. */ |
646 | ||
36beb9be BP |
647 | static void process_dpif_port_changes(struct dpif_backer *); |
648 | static void process_dpif_all_ports_changed(struct dpif_backer *); | |
649 | static void process_dpif_port_change(struct dpif_backer *, | |
650 | const char *devname); | |
651 | static void process_dpif_port_error(struct dpif_backer *, int error); | |
652 | ||
476cb42a BP |
653 | static struct ofproto_dpif * |
654 | lookup_ofproto_dpif_by_port_name(const char *name) | |
655 | { | |
656 | struct ofproto_dpif *ofproto; | |
657 | ||
658 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
659 | if (sset_contains(&ofproto->ports, name)) { | |
660 | return ofproto; | |
661 | } | |
662 | } | |
663 | ||
664 | return NULL; | |
665 | } | |
666 | ||
acf60855 JP |
667 | static int |
668 | type_run(const char *type) | |
669 | { | |
e20b5746 | 670 | static long long int push_timer = LLONG_MIN; |
acf60855 | 671 | struct dpif_backer *backer; |
acf60855 JP |
672 | |
673 | backer = shash_find_data(&all_dpif_backers, type); | |
674 | if (!backer) { | |
675 | /* This is not necessarily a problem, since backers are only | |
676 | * created on demand. */ | |
677 | return 0; | |
678 | } | |
679 | ||
680 | dpif_run(backer->dpif); | |
681 | ||
e20b5746 EJ |
682 | /* The most natural place to push facet statistics is when they're pulled |
683 | * from the datapath. However, when there are many flows in the datapath, | |
684 | * this expensive operation can occur so frequently, that it reduces our | |
685 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
686 | * here instead. */ | |
687 | if (time_msec() > push_timer) { | |
688 | push_timer = time_msec() + 2000; | |
689 | push_all_stats(); | |
690 | } | |
691 | ||
40358701 GS |
692 | /* If vswitchd started with other_config:flow_restore_wait set as "true", |
693 | * and the configuration has now changed to "false", enable receiving | |
694 | * packets from the datapath. */ | |
695 | if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { | |
36beb9be BP |
696 | int error; |
697 | ||
40358701 GS |
698 | backer->recv_set_enable = true; |
699 | ||
700 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); | |
701 | if (error) { | |
e1ec7dd4 | 702 | udpif_recv_set(backer->udpif, 0, false); |
40358701 GS |
703 | VLOG_ERR("Failed to enable receiving packets in dpif."); |
704 | return error; | |
705 | } | |
448a4b2f AW |
706 | udpif_recv_set(backer->udpif, n_handler_threads, |
707 | backer->recv_set_enable); | |
40358701 GS |
708 | dpif_flow_flush(backer->dpif); |
709 | backer->need_revalidate = REV_RECONFIGURE; | |
710 | } | |
711 | ||
448a4b2f AW |
712 | /* If the n_handler_threads is reconfigured, call udpif_recv_set() |
713 | * to reset the handler threads. */ | |
714 | if (backer->n_handler_threads != n_handler_threads) { | |
715 | udpif_recv_set(backer->udpif, n_handler_threads, | |
716 | backer->recv_set_enable); | |
717 | backer->n_handler_threads = n_handler_threads; | |
718 | } | |
719 | ||
a1616063 | 720 | if (backer->need_revalidate) { |
2cc3c58e | 721 | struct ofproto_dpif *ofproto; |
a614d823 KM |
722 | struct simap_node *node; |
723 | struct simap tmp_backers; | |
724 | ||
725 | /* Handle tunnel garbage collection. */ | |
726 | simap_init(&tmp_backers); | |
727 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
728 | ||
729 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
730 | struct ofport_dpif *iter; | |
731 | ||
732 | if (backer != ofproto->backer) { | |
733 | continue; | |
734 | } | |
735 | ||
736 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
3aa30359 | 737 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
a614d823 KM |
738 | const char *dp_port; |
739 | ||
42943cde | 740 | if (!iter->is_tunnel) { |
a614d823 KM |
741 | continue; |
742 | } | |
743 | ||
3aa30359 BP |
744 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev, |
745 | namebuf, sizeof namebuf); | |
a614d823 KM |
746 | node = simap_find(&tmp_backers, dp_port); |
747 | if (node) { | |
748 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
749 | simap_delete(&tmp_backers, node); | |
750 | node = simap_find(&backer->tnl_backers, dp_port); | |
751 | } else { | |
752 | node = simap_find(&backer->tnl_backers, dp_port); | |
753 | if (!node) { | |
4e022ec0 | 754 | odp_port_t odp_port = ODPP_NONE; |
a614d823 KM |
755 | |
756 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
757 | &odp_port)) { | |
4e022ec0 AW |
758 | simap_put(&backer->tnl_backers, dp_port, |
759 | odp_to_u32(odp_port)); | |
a614d823 KM |
760 | node = simap_find(&backer->tnl_backers, dp_port); |
761 | } | |
762 | } | |
763 | } | |
764 | ||
4e022ec0 | 765 | iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE; |
42943cde EJ |
766 | if (tnl_port_reconfigure(iter, iter->up.netdev, |
767 | iter->odp_port)) { | |
a614d823 KM |
768 | backer->need_revalidate = REV_RECONFIGURE; |
769 | } | |
770 | } | |
771 | } | |
772 | ||
773 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
4e022ec0 | 774 | dpif_port_del(backer->dpif, u32_to_odp(node->data)); |
a614d823 KM |
775 | } |
776 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
777 | |
778 | switch (backer->need_revalidate) { | |
779 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
780 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
4a1b8f30 | 781 | case REV_BOND: COVERAGE_INC(rev_bond); break; |
2cc3c58e EJ |
782 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; |
783 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
30618594 | 784 | case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break; |
2cc3c58e EJ |
785 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; |
786 | } | |
f728af2e BP |
787 | backer->need_revalidate = 0; |
788 | ||
a1616063 EJ |
789 | /* Clear the drop_keys in case we should now be accepting some |
790 | * formerly dropped flows. */ | |
791 | drop_key_clear(backer); | |
792 | ||
2cc3c58e | 793 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 794 | struct facet *facet, *next; |
a1616063 | 795 | struct ofport_dpif *ofport; |
bcd2633a | 796 | struct cls_cursor cursor; |
a1616063 | 797 | struct ofbundle *bundle; |
2cc3c58e EJ |
798 | |
799 | if (ofproto->backer != backer) { | |
800 | continue; | |
801 | } | |
802 | ||
dc24a00f | 803 | ovs_rwlock_wrlock(&xlate_rwlock); |
89a8a7f0 | 804 | xlate_ofproto_set(ofproto, ofproto->up.name, |
ec89fc6f EJ |
805 | ofproto->backer->dpif, ofproto->miss_rule, |
806 | ofproto->no_packet_in_rule, ofproto->ml, | |
9d189a50 EJ |
807 | ofproto->stp, ofproto->mbridge, |
808 | ofproto->sflow, ofproto->ipfix, | |
809 | ofproto->up.frag_handling, | |
a1616063 EJ |
810 | ofproto->up.forward_bpdu, |
811 | connmgr_has_in_band(ofproto->up.connmgr), | |
9d189a50 | 812 | ofproto->netflow != NULL); |
46c88433 | 813 | |
a1616063 EJ |
814 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
815 | xlate_bundle_set(ofproto, bundle, bundle->name, | |
816 | bundle->vlan_mode, bundle->vlan, | |
817 | bundle->trunks, bundle->use_priority_tags, | |
818 | bundle->bond, bundle->lacp, | |
819 | bundle->floodable); | |
820 | } | |
821 | ||
822 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
9d189a50 EJ |
823 | int stp_port = ofport->stp_port |
824 | ? stp_port_no(ofport->stp_port) | |
92cf817b | 825 | : -1; |
a1616063 EJ |
826 | xlate_ofport_set(ofproto, ofport->bundle, ofport, |
827 | ofport->up.ofp_port, ofport->odp_port, | |
828 | ofport->up.netdev, ofport->cfm, | |
9d189a50 | 829 | ofport->bfd, ofport->peer, stp_port, |
55954f6e | 830 | ofport->qdscp, ofport->n_qdscp, |
dd8cd4b4 SH |
831 | ofport->up.pp.config, ofport->up.pp.state, |
832 | ofport->is_tunnel, ofport->may_enable); | |
46c88433 | 833 | } |
dc24a00f | 834 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 835 | |
0b4f2078 EJ |
836 | /* Only ofproto-dpif cares about the facet classifier so we just |
837 | * lock cls_cursor_init() to appease the thread safety analysis. */ | |
838 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
bcd2633a | 839 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
0b4f2078 | 840 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 841 | CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { |
a1616063 EJ |
842 | facet_revalidate(facet); |
843 | run_fast_rl(); | |
2cc3c58e EJ |
844 | } |
845 | } | |
e1ec7dd4 EJ |
846 | |
847 | udpif_revalidate(backer->udpif); | |
2cc3c58e EJ |
848 | } |
849 | ||
40358701 GS |
850 | if (!backer->recv_set_enable) { |
851 | /* Wake up before a max of 1000ms. */ | |
852 | timer_set_duration(&backer->next_expiration, 1000); | |
853 | } else if (timer_expired(&backer->next_expiration)) { | |
acf60855 JP |
854 | int delay = expire(backer); |
855 | timer_set_duration(&backer->next_expiration, delay); | |
856 | } | |
857 | ||
36beb9be | 858 | process_dpif_port_changes(backer); |
acf60855 | 859 | |
04d08d54 EJ |
860 | if (backer->governor) { |
861 | size_t n_subfacets; | |
862 | ||
863 | governor_run(backer->governor); | |
864 | ||
865 | /* If the governor has shrunk to its minimum size and the number of | |
866 | * subfacets has dwindled, then drop the governor entirely. | |
867 | * | |
868 | * For hysteresis, the number of subfacets to drop the governor is | |
869 | * smaller than the number needed to trigger its creation. */ | |
870 | n_subfacets = hmap_count(&backer->subfacets); | |
871 | if (n_subfacets * 4 < flow_eviction_threshold | |
872 | && governor_is_idle(backer->governor)) { | |
873 | governor_destroy(backer->governor); | |
874 | backer->governor = NULL; | |
875 | } | |
876 | } | |
877 | ||
acf60855 JP |
878 | return 0; |
879 | } | |
880 | ||
36beb9be BP |
881 | /* Check for and handle port changes in 'backer''s dpif. */ |
882 | static void | |
883 | process_dpif_port_changes(struct dpif_backer *backer) | |
884 | { | |
885 | for (;;) { | |
886 | char *devname; | |
887 | int error; | |
888 | ||
889 | error = dpif_port_poll(backer->dpif, &devname); | |
890 | switch (error) { | |
891 | case EAGAIN: | |
892 | return; | |
893 | ||
894 | case ENOBUFS: | |
895 | process_dpif_all_ports_changed(backer); | |
896 | break; | |
897 | ||
898 | case 0: | |
899 | process_dpif_port_change(backer, devname); | |
900 | free(devname); | |
901 | break; | |
902 | ||
903 | default: | |
904 | process_dpif_port_error(backer, error); | |
905 | break; | |
906 | } | |
907 | } | |
908 | } | |
909 | ||
910 | static void | |
911 | process_dpif_all_ports_changed(struct dpif_backer *backer) | |
912 | { | |
913 | struct ofproto_dpif *ofproto; | |
914 | struct dpif_port dpif_port; | |
915 | struct dpif_port_dump dump; | |
916 | struct sset devnames; | |
917 | const char *devname; | |
918 | ||
919 | sset_init(&devnames); | |
920 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
921 | if (ofproto->backer == backer) { | |
922 | struct ofport *ofport; | |
923 | ||
924 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
925 | sset_add(&devnames, netdev_get_name(ofport->netdev)); | |
926 | } | |
927 | } | |
928 | } | |
929 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) { | |
930 | sset_add(&devnames, dpif_port.name); | |
931 | } | |
932 | ||
933 | SSET_FOR_EACH (devname, &devnames) { | |
934 | process_dpif_port_change(backer, devname); | |
935 | } | |
936 | sset_destroy(&devnames); | |
937 | } | |
938 | ||
939 | static void | |
940 | process_dpif_port_change(struct dpif_backer *backer, const char *devname) | |
941 | { | |
942 | struct ofproto_dpif *ofproto; | |
943 | struct dpif_port port; | |
944 | ||
945 | /* Don't report on the datapath's device. */ | |
946 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
947 | return; | |
948 | } | |
949 | ||
950 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
951 | &all_ofproto_dpifs) { | |
952 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { | |
953 | return; | |
954 | } | |
955 | } | |
956 | ||
957 | ofproto = lookup_ofproto_dpif_by_port_name(devname); | |
958 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { | |
959 | /* The port was removed. If we know the datapath, | |
960 | * report it through poll_set(). If we don't, it may be | |
961 | * notifying us of a removal we initiated, so ignore it. | |
962 | * If there's a pending ENOBUFS, let it stand, since | |
963 | * everything will be reevaluated. */ | |
964 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
965 | sset_add(&ofproto->port_poll_set, devname); | |
966 | ofproto->port_poll_errno = 0; | |
967 | } | |
968 | } else if (!ofproto) { | |
969 | /* The port was added, but we don't know with which | |
970 | * ofproto we should associate it. Delete it. */ | |
971 | dpif_port_del(backer->dpif, port.port_no); | |
74cc3969 BP |
972 | } else { |
973 | struct ofport_dpif *ofport; | |
974 | ||
975 | ofport = ofport_dpif_cast(shash_find_data( | |
976 | &ofproto->up.port_by_name, devname)); | |
977 | if (ofport | |
978 | && ofport->odp_port != port.port_no | |
979 | && !odp_port_to_ofport(backer, port.port_no)) | |
980 | { | |
981 | /* 'ofport''s datapath port number has changed from | |
982 | * 'ofport->odp_port' to 'port.port_no'. Update our internal data | |
983 | * structures to match. */ | |
8449c4d6 | 984 | ovs_rwlock_wrlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
985 | hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node); |
986 | ofport->odp_port = port.port_no; | |
987 | hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node, | |
988 | hash_odp_port(port.port_no)); | |
8449c4d6 | 989 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
74cc3969 BP |
990 | backer->need_revalidate = REV_RECONFIGURE; |
991 | } | |
36beb9be BP |
992 | } |
993 | dpif_port_destroy(&port); | |
994 | } | |
995 | ||
996 | /* Propagate 'error' to all ofprotos based on 'backer'. */ | |
997 | static void | |
998 | process_dpif_port_error(struct dpif_backer *backer, int error) | |
999 | { | |
1000 | struct ofproto_dpif *ofproto; | |
1001 | ||
1002 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1003 | if (ofproto->backer == backer) { | |
1004 | sset_clear(&ofproto->port_poll_set); | |
1005 | ofproto->port_poll_errno = error; | |
1006 | } | |
1007 | } | |
1008 | } | |
1009 | ||
acf60855 | 1010 | static int |
e1ec7dd4 | 1011 | dpif_backer_run_fast(struct dpif_backer *backer) |
acf60855 | 1012 | { |
e1ec7dd4 | 1013 | handle_upcalls(backer); |
acf60855 JP |
1014 | |
1015 | return 0; | |
1016 | } | |
1017 | ||
8fa4d1d0 EJ |
1018 | static int |
1019 | type_run_fast(const char *type) | |
1020 | { | |
1021 | struct dpif_backer *backer; | |
1022 | ||
1023 | backer = shash_find_data(&all_dpif_backers, type); | |
1024 | if (!backer) { | |
1025 | /* This is not necessarily a problem, since backers are only | |
1026 | * created on demand. */ | |
1027 | return 0; | |
1028 | } | |
1029 | ||
e1ec7dd4 | 1030 | return dpif_backer_run_fast(backer); |
8fa4d1d0 EJ |
1031 | } |
1032 | ||
1033 | static void | |
1034 | run_fast_rl(void) | |
1035 | { | |
1036 | static long long int port_rl = LLONG_MIN; | |
8fa4d1d0 EJ |
1037 | |
1038 | if (time_msec() >= port_rl) { | |
1039 | struct ofproto_dpif *ofproto; | |
8fa4d1d0 EJ |
1040 | |
1041 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3d9c5e58 | 1042 | run_fast(&ofproto->up); |
8fa4d1d0 EJ |
1043 | } |
1044 | port_rl = time_msec() + 200; | |
1045 | } | |
8fa4d1d0 EJ |
1046 | } |
1047 | ||
acf60855 JP |
1048 | static void |
1049 | type_wait(const char *type) | |
1050 | { | |
1051 | struct dpif_backer *backer; | |
1052 | ||
1053 | backer = shash_find_data(&all_dpif_backers, type); | |
1054 | if (!backer) { | |
1055 | /* This is not necessarily a problem, since backers are only | |
1056 | * created on demand. */ | |
1057 | return; | |
1058 | } | |
1059 | ||
04d08d54 EJ |
1060 | if (backer->governor) { |
1061 | governor_wait(backer->governor); | |
1062 | } | |
1063 | ||
acf60855 | 1064 | timer_wait(&backer->next_expiration); |
c0365fc8 YT |
1065 | dpif_wait(backer->dpif); |
1066 | udpif_wait(backer->udpif); | |
acf60855 JP |
1067 | } |
1068 | \f | |
abe529af BP |
1069 | /* Basic life-cycle. */ |
1070 | ||
c57b2226 BP |
1071 | static int add_internal_flows(struct ofproto_dpif *); |
1072 | ||
abe529af BP |
1073 | static struct ofproto * |
1074 | alloc(void) | |
1075 | { | |
1076 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1077 | return &ofproto->up; | |
1078 | } | |
1079 | ||
1080 | static void | |
1081 | dealloc(struct ofproto *ofproto_) | |
1082 | { | |
1083 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1084 | free(ofproto); | |
1085 | } | |
1086 | ||
acf60855 JP |
1087 | static void |
1088 | close_dpif_backer(struct dpif_backer *backer) | |
1089 | { | |
cb22974d | 1090 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1091 | |
1092 | if (--backer->refcount) { | |
1093 | return; | |
1094 | } | |
1095 | ||
8f73d537 EJ |
1096 | drop_key_clear(backer); |
1097 | hmap_destroy(&backer->drop_keys); | |
1098 | ||
b395cf1f BP |
1099 | udpif_destroy(backer->udpif); |
1100 | ||
7d82ab2e | 1101 | simap_destroy(&backer->tnl_backers); |
8449c4d6 | 1102 | ovs_rwlock_destroy(&backer->odp_to_ofport_lock); |
acf60855 | 1103 | hmap_destroy(&backer->odp_to_ofport_map); |
4f7cc3c7 | 1104 | shash_find_and_delete(&all_dpif_backers, backer->type); |
acf60855 | 1105 | free(backer->type); |
acf60855 JP |
1106 | dpif_close(backer->dpif); |
1107 | ||
04d08d54 EJ |
1108 | ovs_assert(hmap_is_empty(&backer->subfacets)); |
1109 | hmap_destroy(&backer->subfacets); | |
1110 | governor_destroy(backer->governor); | |
1111 | ||
acf60855 JP |
1112 | free(backer); |
1113 | } | |
1114 | ||
1115 | /* Datapath port slated for removal from datapath. */ | |
1116 | struct odp_garbage { | |
1117 | struct list list_node; | |
4e022ec0 | 1118 | odp_port_t odp_port; |
acf60855 JP |
1119 | }; |
1120 | ||
1121 | static int | |
1122 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1123 | { | |
1124 | struct dpif_backer *backer; | |
1125 | struct dpif_port_dump port_dump; | |
1126 | struct dpif_port port; | |
1127 | struct shash_node *node; | |
1128 | struct list garbage_list; | |
1129 | struct odp_garbage *garbage, *next; | |
1130 | struct sset names; | |
1131 | char *backer_name; | |
1132 | const char *name; | |
1133 | int error; | |
1134 | ||
1135 | backer = shash_find_data(&all_dpif_backers, type); | |
1136 | if (backer) { | |
1137 | backer->refcount++; | |
1138 | *backerp = backer; | |
1139 | return 0; | |
1140 | } | |
1141 | ||
1142 | backer_name = xasprintf("ovs-%s", type); | |
1143 | ||
1144 | /* Remove any existing datapaths, since we assume we're the only | |
1145 | * userspace controlling the datapath. */ | |
1146 | sset_init(&names); | |
1147 | dp_enumerate_names(type, &names); | |
1148 | SSET_FOR_EACH(name, &names) { | |
1149 | struct dpif *old_dpif; | |
1150 | ||
1151 | /* Don't remove our backer if it exists. */ | |
1152 | if (!strcmp(name, backer_name)) { | |
1153 | continue; | |
1154 | } | |
1155 | ||
1156 | if (dpif_open(name, type, &old_dpif)) { | |
1157 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1158 | } else { | |
1159 | dpif_delete(old_dpif); | |
1160 | dpif_close(old_dpif); | |
1161 | } | |
1162 | } | |
1163 | sset_destroy(&names); | |
1164 | ||
1165 | backer = xmalloc(sizeof *backer); | |
1166 | ||
1167 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1168 | free(backer_name); | |
1169 | if (error) { | |
1170 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
10a89ef0 | 1171 | ovs_strerror(error)); |
4c1b1289 | 1172 | free(backer); |
acf60855 JP |
1173 | return error; |
1174 | } | |
e1ec7dd4 | 1175 | backer->udpif = udpif_create(backer, backer->dpif); |
acf60855 JP |
1176 | |
1177 | backer->type = xstrdup(type); | |
04d08d54 | 1178 | backer->governor = NULL; |
acf60855 JP |
1179 | backer->refcount = 1; |
1180 | hmap_init(&backer->odp_to_ofport_map); | |
8449c4d6 | 1181 | ovs_rwlock_init(&backer->odp_to_ofport_lock); |
8f73d537 | 1182 | hmap_init(&backer->drop_keys); |
04d08d54 | 1183 | hmap_init(&backer->subfacets); |
acf60855 | 1184 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1185 | backer->need_revalidate = 0; |
7d82ab2e | 1186 | simap_init(&backer->tnl_backers); |
40358701 | 1187 | backer->recv_set_enable = !ofproto_get_flow_restore_wait(); |
acf60855 JP |
1188 | *backerp = backer; |
1189 | ||
40358701 GS |
1190 | if (backer->recv_set_enable) { |
1191 | dpif_flow_flush(backer->dpif); | |
1192 | } | |
acf60855 JP |
1193 | |
1194 | /* Loop through the ports already on the datapath and remove any | |
1195 | * that we don't need anymore. */ | |
1196 | list_init(&garbage_list); | |
1197 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1198 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1199 | node = shash_find(&init_ofp_ports, port.name); | |
1200 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1201 | garbage = xmalloc(sizeof *garbage); | |
1202 | garbage->odp_port = port.port_no; | |
1203 | list_push_front(&garbage_list, &garbage->list_node); | |
1204 | } | |
1205 | } | |
1206 | dpif_port_dump_done(&port_dump); | |
1207 | ||
1208 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1209 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1210 | list_remove(&garbage->list_node); | |
1211 | free(garbage); | |
1212 | } | |
1213 | ||
1214 | shash_add(&all_dpif_backers, type, backer); | |
1215 | ||
40358701 | 1216 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); |
acf60855 JP |
1217 | if (error) { |
1218 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
10a89ef0 | 1219 | type, ovs_strerror(error)); |
acf60855 JP |
1220 | close_dpif_backer(backer); |
1221 | return error; | |
1222 | } | |
448a4b2f AW |
1223 | udpif_recv_set(backer->udpif, n_handler_threads, |
1224 | backer->recv_set_enable); | |
1225 | backer->n_handler_threads = n_handler_threads; | |
acf60855 | 1226 | |
dc54ef36 EJ |
1227 | backer->max_n_subfacet = 0; |
1228 | backer->created = time_msec(); | |
dc54ef36 EJ |
1229 | backer->avg_n_subfacet = 0; |
1230 | backer->avg_subfacet_life = 0; | |
1231 | ||
acf60855 JP |
1232 | return error; |
1233 | } | |
1234 | ||
abe529af | 1235 | static int |
0f5f95a9 | 1236 | construct(struct ofproto *ofproto_) |
abe529af BP |
1237 | { |
1238 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1239 | struct shash_node *node, *next; |
abe529af | 1240 | int error; |
abe529af | 1241 | |
acf60855 | 1242 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1243 | if (error) { |
abe529af BP |
1244 | return error; |
1245 | } | |
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 | 1254 | ofproto->has_bonded_bundles = false; |
0da3c61b | 1255 | ovs_mutex_init(&ofproto->stats_mutex); |
834d6caf | 1256 | ovs_mutex_init(&ofproto->vsp_mutex); |
abe529af | 1257 | |
476f36e8 | 1258 | classifier_init(&ofproto->facets, NULL); |
4d2a0f39 | 1259 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 | 1260 | |
05067881 | 1261 | guarded_list_init(&ofproto->pins); |
ada3a58d | 1262 | |
abe529af BP |
1263 | ofproto_dpif_unixctl_init(); |
1264 | ||
52a90c29 BP |
1265 | hmap_init(&ofproto->vlandev_map); |
1266 | hmap_init(&ofproto->realdev_vid_map); | |
1267 | ||
acf60855 | 1268 | sset_init(&ofproto->ports); |
0a740f48 | 1269 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1270 | sset_init(&ofproto->port_poll_set); |
1271 | ofproto->port_poll_errno = 0; | |
1272 | ||
1273 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1274 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1275 | |
1276 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1277 | /* Check if the datapath already has this port. */ | |
1278 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1279 | sset_add(&ofproto->ports, node->name); | |
1280 | } | |
1281 | ||
1282 | free(iface_hint->br_name); | |
1283 | free(iface_hint->br_type); | |
4f9e08a5 | 1284 | free(iface_hint); |
acf60855 JP |
1285 | shash_delete(&init_ofp_ports, node); |
1286 | } | |
1287 | } | |
e1b1d06a | 1288 | |
b44a10b7 BP |
1289 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1290 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1291 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1292 | |
1293 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1294 | error = add_internal_flows(ofproto); |
1295 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1296 | ||
735d7efb AZ |
1297 | ofproto->n_hit = 0; |
1298 | ofproto->n_missed = 0; | |
1299 | ||
c57b2226 BP |
1300 | return error; |
1301 | } | |
1302 | ||
1303 | static int | |
1304 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1305 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1306 | { |
1307 | struct ofputil_flow_mod fm; | |
1308 | int error; | |
1309 | ||
81a76618 BP |
1310 | match_init_catchall(&fm.match); |
1311 | fm.priority = 0; | |
1312 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1313 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1314 | fm.cookie = htonll(0); |
1315 | fm.cookie_mask = htonll(0); | |
23342857 | 1316 | fm.modify_cookie = false; |
c57b2226 BP |
1317 | fm.table_id = TBL_INTERNAL; |
1318 | fm.command = OFPFC_ADD; | |
1319 | fm.idle_timeout = 0; | |
1320 | fm.hard_timeout = 0; | |
1321 | fm.buffer_id = 0; | |
1322 | fm.out_port = 0; | |
1323 | fm.flags = 0; | |
f25d0cf3 BP |
1324 | fm.ofpacts = ofpacts->data; |
1325 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1326 | |
1327 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1328 | if (error) { | |
1329 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1330 | id, ofperr_to_string(error)); | |
1331 | return error; | |
1332 | } | |
1333 | ||
ad3efdcb EJ |
1334 | if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL, |
1335 | rulep)) { | |
a2143702 | 1336 | rule_dpif_unref(*rulep); |
ad3efdcb EJ |
1337 | } else { |
1338 | NOT_REACHED(); | |
1339 | } | |
0f5f95a9 | 1340 | |
abe529af BP |
1341 | return 0; |
1342 | } | |
1343 | ||
c57b2226 BP |
1344 | static int |
1345 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1346 | { | |
f25d0cf3 BP |
1347 | struct ofpact_controller *controller; |
1348 | uint64_t ofpacts_stub[128 / 8]; | |
1349 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1350 | int error; |
1351 | int id; | |
1352 | ||
f25d0cf3 | 1353 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1354 | id = 1; |
1355 | ||
f25d0cf3 BP |
1356 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1357 | controller->max_len = UINT16_MAX; | |
1358 | controller->controller_id = 0; | |
1359 | controller->reason = OFPR_NO_MATCH; | |
1360 | ofpact_pad(&ofpacts); | |
1361 | ||
1362 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1363 | if (error) { |
1364 | return error; | |
1365 | } | |
1366 | ||
f25d0cf3 BP |
1367 | ofpbuf_clear(&ofpacts); |
1368 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 | 1369 | &ofproto->no_packet_in_rule); |
7fd51d39 BP |
1370 | if (error) { |
1371 | return error; | |
1372 | } | |
1373 | ||
1374 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1375 | &ofproto->drop_frags_rule); | |
c57b2226 BP |
1376 | return error; |
1377 | } | |
1378 | ||
abe529af BP |
1379 | static void |
1380 | destruct(struct ofproto *ofproto_) | |
1381 | { | |
1382 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1383 | struct rule_dpif *rule, *next_rule; |
0fb7792a | 1384 | struct ofproto_packet_in *pin, *next_pin; |
d6aa7ad6 EJ |
1385 | struct facet *facet, *next_facet; |
1386 | struct cls_cursor cursor; | |
d0918789 | 1387 | struct oftable *table; |
1b63b91e | 1388 | struct list pins; |
abe529af | 1389 | |
d6aa7ad6 EJ |
1390 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
1391 | cls_cursor_init(&cursor, &ofproto->facets, NULL); | |
1392 | ovs_rwlock_unlock(&ofproto->facets.rwlock); | |
1393 | CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) { | |
1394 | facet_remove(facet); | |
1395 | } | |
1396 | ||
875b94ed | 1397 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1398 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1399 | xlate_remove_ofproto(ofproto); |
dc24a00f | 1400 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 1401 | |
05067881 BP |
1402 | /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a |
1403 | * use-after-free error. */ | |
1404 | udpif_revalidate(ofproto->backer->udpif); | |
1fe409d5 | 1405 | |
b44a10b7 | 1406 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 | 1407 | |
0697b5c3 BP |
1408 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1409 | struct cls_cursor cursor; | |
1410 | ||
15aaf599 | 1411 | ovs_rwlock_rdlock(&table->cls.rwlock); |
d0918789 | 1412 | cls_cursor_init(&cursor, &table->cls, NULL); |
15aaf599 | 1413 | ovs_rwlock_unlock(&table->cls.rwlock); |
0697b5c3 | 1414 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
8b81d1ef | 1415 | ofproto_rule_delete(&ofproto->up, &rule->up); |
0697b5c3 | 1416 | } |
7ee20df1 BP |
1417 | } |
1418 | ||
05067881 BP |
1419 | guarded_list_pop_all(&ofproto->pins, &pins); |
1420 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { | |
ada3a58d | 1421 | list_remove(&pin->list_node); |
0fb7792a | 1422 | free(CONST_CAST(void *, pin->up.packet)); |
ada3a58d EJ |
1423 | free(pin); |
1424 | } | |
05067881 | 1425 | guarded_list_destroy(&ofproto->pins); |
ada3a58d | 1426 | |
ec7ceaed | 1427 | mbridge_unref(ofproto->mbridge); |
abe529af BP |
1428 | |
1429 | netflow_destroy(ofproto->netflow); | |
9723bcce | 1430 | dpif_sflow_unref(ofproto->sflow); |
abe529af | 1431 | hmap_destroy(&ofproto->bundles); |
5d989517 | 1432 | mac_learning_unref(ofproto->ml); |
abe529af | 1433 | |
bcd2633a | 1434 | classifier_destroy(&ofproto->facets); |
abe529af | 1435 | |
52a90c29 BP |
1436 | hmap_destroy(&ofproto->vlandev_map); |
1437 | hmap_destroy(&ofproto->realdev_vid_map); | |
1438 | ||
acf60855 | 1439 | sset_destroy(&ofproto->ports); |
0a740f48 | 1440 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1441 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1442 | |
0da3c61b | 1443 | ovs_mutex_destroy(&ofproto->stats_mutex); |
13501305 EJ |
1444 | ovs_mutex_destroy(&ofproto->vsp_mutex); |
1445 | ||
acf60855 | 1446 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1447 | } |
1448 | ||
1449 | static int | |
5fcc0d00 | 1450 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1451 | { |
1452 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0fb7792a | 1453 | struct ofproto_packet_in *pin, *next_pin; |
1b63b91e | 1454 | struct list pins; |
abe529af | 1455 | |
40358701 GS |
1456 | /* Do not perform any periodic activity required by 'ofproto' while |
1457 | * waiting for flow restore to complete. */ | |
1458 | if (ofproto_get_flow_restore_wait()) { | |
1459 | return 0; | |
1460 | } | |
1461 | ||
05067881 | 1462 | guarded_list_pop_all(&ofproto->pins, &pins); |
ada3a58d EJ |
1463 | LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { |
1464 | connmgr_send_packet_in(ofproto->up.connmgr, pin); | |
1465 | list_remove(&pin->list_node); | |
0fb7792a | 1466 | free(CONST_CAST(void *, pin->up.packet)); |
ada3a58d EJ |
1467 | free(pin); |
1468 | } | |
1469 | ||
5fcc0d00 BP |
1470 | return 0; |
1471 | } | |
1472 | ||
1473 | static int | |
1474 | run(struct ofproto *ofproto_) | |
1475 | { | |
1476 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1477 | struct ofport_dpif *ofport; | |
1478 | struct ofbundle *bundle; | |
1479 | int error; | |
1480 | ||
ec7ceaed EJ |
1481 | if (mbridge_need_revalidate(ofproto->mbridge)) { |
1482 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
509c0149 | 1483 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 1484 | mac_learning_flush(ofproto->ml); |
509c0149 | 1485 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
ec7ceaed EJ |
1486 | } |
1487 | ||
40358701 GS |
1488 | /* Do not perform any periodic activity below required by 'ofproto' while |
1489 | * waiting for flow restore to complete. */ | |
1490 | if (ofproto_get_flow_restore_wait()) { | |
1491 | return 0; | |
1492 | } | |
1493 | ||
5fcc0d00 BP |
1494 | error = run_fast(ofproto_); |
1495 | if (error) { | |
1496 | return error; | |
abe529af BP |
1497 | } |
1498 | ||
abe529af | 1499 | if (ofproto->netflow) { |
6fca1ffb BP |
1500 | if (netflow_run(ofproto->netflow)) { |
1501 | send_netflow_active_timeouts(ofproto); | |
1502 | } | |
abe529af BP |
1503 | } |
1504 | if (ofproto->sflow) { | |
bae473fe | 1505 | dpif_sflow_run(ofproto->sflow); |
abe529af | 1506 | } |
978427a5 RL |
1507 | if (ofproto->ipfix) { |
1508 | dpif_ipfix_run(ofproto->ipfix); | |
1509 | } | |
abe529af BP |
1510 | |
1511 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1512 | port_run(ofport); | |
1513 | } | |
1514 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1515 | bundle_run(bundle); | |
1516 | } | |
1517 | ||
21f7563c | 1518 | stp_run(ofproto); |
509c0149 | 1519 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 EJ |
1520 | if (mac_learning_run(ofproto->ml)) { |
1521 | ofproto->backer->need_revalidate = REV_MAC_LEARNING; | |
1522 | } | |
509c0149 | 1523 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 1524 | |
6814e51f | 1525 | /* Check the consistency of a random facet, to aid debugging. */ |
0b4f2078 | 1526 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
4d2a0f39 | 1527 | if (time_msec() >= ofproto->consistency_rl |
bcd2633a | 1528 | && !classifier_is_empty(&ofproto->facets) |
2cc3c58e | 1529 | && !ofproto->backer->need_revalidate) { |
03868246 | 1530 | struct cls_subtable *table; |
bcd2633a | 1531 | struct cls_rule *cr; |
6814e51f BP |
1532 | struct facet *facet; |
1533 | ||
4d2a0f39 EJ |
1534 | ofproto->consistency_rl = time_msec() + 250; |
1535 | ||
03868246 JR |
1536 | table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables), |
1537 | struct cls_subtable, hmap_node); | |
bcd2633a JP |
1538 | cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, |
1539 | hmap_node); | |
1540 | facet = CONTAINER_OF(cr, struct facet, cr); | |
1541 | ||
a1616063 EJ |
1542 | if (!facet_check_consistency(facet)) { |
1543 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
6814e51f BP |
1544 | } |
1545 | } | |
0b4f2078 | 1546 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 1547 | |
abe529af BP |
1548 | return 0; |
1549 | } | |
1550 | ||
1551 | static void | |
1552 | wait(struct ofproto *ofproto_) | |
1553 | { | |
1554 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af BP |
1555 | struct ofbundle *bundle; |
1556 | ||
40358701 GS |
1557 | if (ofproto_get_flow_restore_wait()) { |
1558 | return; | |
1559 | } | |
1560 | ||
abe529af | 1561 | if (ofproto->sflow) { |
bae473fe | 1562 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1563 | } |
978427a5 RL |
1564 | if (ofproto->ipfix) { |
1565 | dpif_ipfix_wait(ofproto->ipfix); | |
1566 | } | |
abe529af BP |
1567 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
1568 | bundle_wait(bundle); | |
1569 | } | |
6fca1ffb BP |
1570 | if (ofproto->netflow) { |
1571 | netflow_wait(ofproto->netflow); | |
1572 | } | |
509c0149 | 1573 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
1c313b88 | 1574 | mac_learning_wait(ofproto->ml); |
509c0149 | 1575 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c | 1576 | stp_wait(ofproto); |
2cc3c58e | 1577 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1578 | /* Shouldn't happen, but if it does just go around again. */ |
1579 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1580 | poll_immediate_wake(); | |
abe529af BP |
1581 | } |
1582 | } | |
1583 | ||
0d085684 BP |
1584 | static void |
1585 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1586 | { | |
1587 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bcd2633a | 1588 | struct cls_cursor cursor; |
04d08d54 EJ |
1589 | size_t n_subfacets = 0; |
1590 | struct facet *facet; | |
0d085684 | 1591 | |
0b4f2078 | 1592 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a | 1593 | simap_increase(usage, "facets", classifier_count(&ofproto->facets)); |
0b4f2078 | 1594 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 1595 | |
0b4f2078 | 1596 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
1597 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
1598 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
04d08d54 EJ |
1599 | n_subfacets += list_size(&facet->subfacets); |
1600 | } | |
0b4f2078 | 1601 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
04d08d54 | 1602 | simap_increase(usage, "subfacets", n_subfacets); |
0d085684 BP |
1603 | } |
1604 | ||
abe529af BP |
1605 | static void |
1606 | flush(struct ofproto *ofproto_) | |
1607 | { | |
1608 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1609 | struct subfacet *subfacet, *next_subfacet; |
1610 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1611 | int n_batch; | |
b0f7b9b5 | 1612 | |
acf60855 JP |
1613 | n_batch = 0; |
1614 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
04d08d54 | 1615 | &ofproto->backer->subfacets) { |
b7e2f6b3 | 1616 | if (subfacet->facet->ofproto != ofproto) { |
04d08d54 EJ |
1617 | continue; |
1618 | } | |
1619 | ||
acf60855 JP |
1620 | if (subfacet->path != SF_NOT_INSTALLED) { |
1621 | batch[n_batch++] = subfacet; | |
1622 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
04d08d54 | 1623 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 JP |
1624 | n_batch = 0; |
1625 | } | |
1626 | } else { | |
1627 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1628 | } |
abe529af | 1629 | } |
acf60855 JP |
1630 | |
1631 | if (n_batch > 0) { | |
04d08d54 | 1632 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 | 1633 | } |
abe529af BP |
1634 | } |
1635 | ||
6c1491fb BP |
1636 | static void |
1637 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1638 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1639 | { |
1640 | *arp_match_ip = true; | |
9e1fd49b BP |
1641 | *actions = (OFPUTIL_A_OUTPUT | |
1642 | OFPUTIL_A_SET_VLAN_VID | | |
1643 | OFPUTIL_A_SET_VLAN_PCP | | |
1644 | OFPUTIL_A_STRIP_VLAN | | |
1645 | OFPUTIL_A_SET_DL_SRC | | |
1646 | OFPUTIL_A_SET_DL_DST | | |
1647 | OFPUTIL_A_SET_NW_SRC | | |
1648 | OFPUTIL_A_SET_NW_DST | | |
1649 | OFPUTIL_A_SET_NW_TOS | | |
1650 | OFPUTIL_A_SET_TP_SRC | | |
1651 | OFPUTIL_A_SET_TP_DST | | |
1652 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1653 | } |
1654 | ||
1655 | static void | |
307975da | 1656 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1657 | { |
1658 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1659 | struct dpif_dp_stats s; |
7fd51d39 | 1660 | uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; |
7e741ffb | 1661 | uint64_t n_lookup; |
6c1491fb BP |
1662 | |
1663 | strcpy(ots->name, "classifier"); | |
1664 | ||
acf60855 | 1665 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1666 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1667 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
7fd51d39 | 1668 | rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); |
acf60855 | 1669 | |
7fd51d39 | 1670 | n_lookup = s.n_hit + s.n_missed - n_dropped_frags; |
7e741ffb JG |
1671 | ots->lookup_count = htonll(n_lookup); |
1672 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1673 | } |
1674 | ||
abe529af BP |
1675 | static struct ofport * |
1676 | port_alloc(void) | |
1677 | { | |
1678 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1679 | return &port->up; | |
1680 | } | |
1681 | ||
1682 | static void | |
1683 | port_dealloc(struct ofport *port_) | |
1684 | { | |
1685 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1686 | free(port); | |
1687 | } | |
1688 | ||
1689 | static int | |
1690 | port_construct(struct ofport *port_) | |
1691 | { | |
1692 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1693 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1694 | const struct netdev *netdev = port->up.netdev; |
3aa30359 | 1695 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
e1b1d06a JP |
1696 | struct dpif_port dpif_port; |
1697 | int error; | |
abe529af | 1698 | |
2cc3c58e | 1699 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1700 | port->bundle = NULL; |
1701 | port->cfm = NULL; | |
ccc09689 | 1702 | port->bfd = NULL; |
d5ffa7f2 | 1703 | port->may_enable = true; |
21f7563c JP |
1704 | port->stp_port = NULL; |
1705 | port->stp_state = STP_DISABLED; | |
42943cde | 1706 | port->is_tunnel = false; |
6cbbf4fa | 1707 | port->peer = NULL; |
55954f6e EJ |
1708 | port->qdscp = NULL; |
1709 | port->n_qdscp = 0; | |
52a90c29 BP |
1710 | port->realdev_ofp_port = 0; |
1711 | port->vlandev_vid = 0; | |
b9ad7294 | 1712 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
a6363cfd | 1713 | port->is_layer3 = netdev_vport_is_layer3(netdev); |
abe529af | 1714 | |
b9ad7294 | 1715 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1716 | /* By bailing out here, we don't submit the port to the sFlow module |
1717 | * to be considered for counter polling export. This is correct | |
1718 | * because the patch port represents an interface that sFlow considers | |
1719 | * to be "internal" to the switch as a whole, and therefore not an | |
1720 | * candidate for counter polling. */ | |
4e022ec0 | 1721 | port->odp_port = ODPP_NONE; |
6cbbf4fa | 1722 | ofport_update_peer(port); |
0a740f48 EJ |
1723 | return 0; |
1724 | } | |
1725 | ||
acf60855 | 1726 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
3aa30359 BP |
1727 | netdev_vport_get_dpif_port(netdev, namebuf, |
1728 | sizeof namebuf), | |
e1b1d06a JP |
1729 | &dpif_port); |
1730 | if (error) { | |
1731 | return error; | |
1732 | } | |
1733 | ||
1734 | port->odp_port = dpif_port.port_no; | |
1735 | ||
b9ad7294 | 1736 | if (netdev_get_tunnel_config(netdev)) { |
42943cde EJ |
1737 | tnl_port_add(port, port->up.netdev, port->odp_port); |
1738 | port->is_tunnel = true; | |
b9ad7294 EJ |
1739 | } else { |
1740 | /* Sanity-check that a mapping doesn't already exist. This | |
1741 | * shouldn't happen for non-tunnel ports. */ | |
1742 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1743 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1744 | dpif_port.name); | |
da78d43d | 1745 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1746 | return EBUSY; |
1747 | } | |
e1b1d06a | 1748 | |
8449c4d6 | 1749 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1750 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
f9c0c3ec | 1751 | hash_odp_port(port->odp_port)); |
8449c4d6 | 1752 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
b9ad7294 | 1753 | } |
da78d43d | 1754 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1755 | |
abe529af | 1756 | if (ofproto->sflow) { |
e1b1d06a | 1757 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1758 | } |
1759 | ||
1760 | return 0; | |
1761 | } | |
1762 | ||
1763 | static void | |
1764 | port_destruct(struct ofport *port_) | |
1765 | { | |
1766 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1767 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
02f8d646 | 1768 | const char *devname = netdev_get_name(port->up.netdev); |
3aa30359 BP |
1769 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1770 | const char *dp_port_name; | |
abe529af | 1771 | |
7894e7da | 1772 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
dc24a00f | 1773 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 1774 | xlate_ofport_remove(port); |
dc24a00f | 1775 | ovs_rwlock_unlock(&xlate_rwlock); |
7894e7da | 1776 | |
3aa30359 BP |
1777 | dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, |
1778 | sizeof namebuf); | |
a614d823 | 1779 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1780 | /* The underlying device is still there, so delete it. This |
1781 | * happens when the ofproto is being destroyed, since the caller | |
1782 | * assumes that removal of attached ports will happen as part of | |
1783 | * destruction. */ | |
42943cde | 1784 | if (!port->is_tunnel) { |
a614d823 KM |
1785 | dpif_port_del(ofproto->backer->dpif, port->odp_port); |
1786 | } | |
acf60855 JP |
1787 | } |
1788 | ||
6cbbf4fa EJ |
1789 | if (port->peer) { |
1790 | port->peer->peer = NULL; | |
1791 | port->peer = NULL; | |
1792 | } | |
1793 | ||
42943cde | 1794 | if (port->odp_port != ODPP_NONE && !port->is_tunnel) { |
8449c4d6 | 1795 | ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 | 1796 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
8449c4d6 | 1797 | ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock); |
0a740f48 EJ |
1798 | } |
1799 | ||
42943cde | 1800 | tnl_port_del(port); |
02f8d646 | 1801 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1802 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
abe529af | 1803 | bundle_remove(port_); |
a5610457 | 1804 | set_cfm(port_, NULL); |
8aee94b6 | 1805 | set_bfd(port_, NULL); |
abe529af | 1806 | if (ofproto->sflow) { |
bae473fe | 1807 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1808 | } |
8b36f51e | 1809 | |
55954f6e | 1810 | free(port->qdscp); |
abe529af BP |
1811 | } |
1812 | ||
1813 | static void | |
1814 | port_modified(struct ofport *port_) | |
1815 | { | |
1816 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1817 | ||
1818 | if (port->bundle && port->bundle->bond) { | |
1819 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1820 | } | |
9d46b444 EJ |
1821 | |
1822 | if (port->cfm) { | |
1823 | cfm_set_netdev(port->cfm, port->up.netdev); | |
1824 | } | |
f1e78bbd | 1825 | |
c1c4e8c7 AW |
1826 | if (port->bfd) { |
1827 | bfd_set_netdev(port->bfd, port->up.netdev); | |
1828 | } | |
1829 | ||
635c5db9 AW |
1830 | ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm, |
1831 | port->up.pp.hw_addr); | |
1832 | ||
42943cde EJ |
1833 | if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, |
1834 | port->odp_port)) { | |
7894e7da EJ |
1835 | ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = |
1836 | REV_RECONFIGURE; | |
f1e78bbd | 1837 | } |
6cbbf4fa EJ |
1838 | |
1839 | ofport_update_peer(port); | |
abe529af BP |
1840 | } |
1841 | ||
1842 | static void | |
9e1fd49b | 1843 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1844 | { |
1845 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1846 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1847 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1848 | |
9e1fd49b | 1849 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1850 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1851 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1852 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1853 | |
9e1fd49b | 1854 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1855 | bundle_update(port->bundle); |
1856 | } | |
abe529af BP |
1857 | } |
1858 | } | |
1859 | ||
1860 | static int | |
1861 | set_sflow(struct ofproto *ofproto_, | |
1862 | const struct ofproto_sflow_options *sflow_options) | |
1863 | { | |
1864 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1865 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1866 | |
abe529af | 1867 | if (sflow_options) { |
bae473fe | 1868 | if (!ds) { |
abe529af BP |
1869 | struct ofport_dpif *ofport; |
1870 | ||
4213f19d | 1871 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1872 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1873 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1874 | } |
2cc3c58e | 1875 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1876 | } |
bae473fe | 1877 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1878 | } else { |
6ff686f2 | 1879 | if (ds) { |
9723bcce | 1880 | dpif_sflow_unref(ds); |
2cc3c58e | 1881 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1882 | ofproto->sflow = NULL; |
1883 | } | |
abe529af BP |
1884 | } |
1885 | return 0; | |
1886 | } | |
1887 | ||
29089a54 RL |
1888 | static int |
1889 | set_ipfix( | |
1890 | struct ofproto *ofproto_, | |
1891 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1892 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1893 | size_t n_flow_exporters_options) | |
1894 | { | |
1895 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1896 | struct dpif_ipfix *di = ofproto->ipfix; | |
978427a5 | 1897 | bool has_options = bridge_exporter_options || flow_exporters_options; |
29089a54 | 1898 | |
978427a5 RL |
1899 | if (has_options && !di) { |
1900 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1901 | } | |
1902 | ||
1903 | if (di) { | |
1904 | /* Call set_options in any case to cleanly flush the flow | |
1905 | * caches in the last exporters that are to be destroyed. */ | |
29089a54 RL |
1906 | dpif_ipfix_set_options( |
1907 | di, bridge_exporter_options, flow_exporters_options, | |
1908 | n_flow_exporters_options); | |
978427a5 RL |
1909 | |
1910 | if (!has_options) { | |
d857c8aa | 1911 | dpif_ipfix_unref(di); |
29089a54 RL |
1912 | ofproto->ipfix = NULL; |
1913 | } | |
1914 | } | |
978427a5 | 1915 | |
29089a54 RL |
1916 | return 0; |
1917 | } | |
1918 | ||
abe529af | 1919 | static int |
a5610457 | 1920 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1921 | { |
1922 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
635c5db9 | 1923 | int error = 0; |
abe529af | 1924 | |
635c5db9 | 1925 | if (s) { |
abe529af | 1926 | if (!ofport->cfm) { |
8c977421 EJ |
1927 | struct ofproto_dpif *ofproto; |
1928 | ||
1929 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1930 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 1931 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
1932 | } |
1933 | ||
a5610457 | 1934 | if (cfm_configure(ofport->cfm, s)) { |
635c5db9 AW |
1935 | error = 0; |
1936 | goto out; | |
abe529af BP |
1937 | } |
1938 | ||
1939 | error = EINVAL; | |
1940 | } | |
8e9a73fa | 1941 | cfm_unref(ofport->cfm); |
abe529af | 1942 | ofport->cfm = NULL; |
635c5db9 AW |
1943 | out: |
1944 | ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, | |
1945 | ofport->up.pp.hw_addr); | |
abe529af BP |
1946 | return error; |
1947 | } | |
1948 | ||
9a9e3786 BP |
1949 | static bool |
1950 | get_cfm_status(const struct ofport *ofport_, | |
1951 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1952 | { |
1953 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1954 | ||
1955 | if (ofport->cfm) { | |
9a9e3786 | 1956 | status->faults = cfm_get_fault(ofport->cfm); |
76c4290d | 1957 | status->flap_count = cfm_get_flap_count(ofport->cfm); |
9a9e3786 BP |
1958 | status->remote_opstate = cfm_get_opup(ofport->cfm); |
1959 | status->health = cfm_get_health(ofport->cfm); | |
1960 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1961 | return true; | |
1de11730 | 1962 | } else { |
9a9e3786 | 1963 | return false; |
1de11730 EJ |
1964 | } |
1965 | } | |
ccc09689 EJ |
1966 | |
1967 | static int | |
1968 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
1969 | { | |
1970 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
1971 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1972 | struct bfd *old; | |
1973 | ||
1974 | old = ofport->bfd; | |
c1c4e8c7 AW |
1975 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), |
1976 | cfg, ofport->up.netdev); | |
ccc09689 EJ |
1977 | if (ofport->bfd != old) { |
1978 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1979 | } | |
635c5db9 AW |
1980 | ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, |
1981 | ofport->up.pp.hw_addr); | |
ccc09689 EJ |
1982 | return 0; |
1983 | } | |
1984 | ||
1985 | static int | |
1986 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
1987 | { | |
1988 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1989 | ||
1990 | if (ofport->bfd) { | |
1991 | bfd_get_status(ofport->bfd, smap); | |
1992 | return 0; | |
1993 | } else { | |
1994 | return ENOENT; | |
1995 | } | |
1996 | } | |
abe529af | 1997 | \f |
21f7563c JP |
1998 | /* Spanning Tree. */ |
1999 | ||
2000 | static void | |
2001 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2002 | { | |
2003 | struct ofproto_dpif *ofproto = ofproto_; | |
2004 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2005 | struct ofport_dpif *ofport; | |
2006 | ||
2007 | ofport = stp_port_get_aux(sp); | |
2008 | if (!ofport) { | |
2009 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2010 | ofproto->up.name, port_num); | |
2011 | } else { | |
2012 | struct eth_header *eth = pkt->l2; | |
2013 | ||
2014 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2015 | if (eth_addr_is_zero(eth->eth_src)) { | |
2016 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2017 | "with unknown MAC", ofproto->up.name, port_num); | |
2018 | } else { | |
91d6cd12 | 2019 | ofproto_dpif_send_packet(ofport, pkt); |
21f7563c JP |
2020 | } |
2021 | } | |
2022 | ofpbuf_delete(pkt); | |
2023 | } | |
2024 | ||
2025 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2026 | static int | |
2027 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2028 | { | |
2029 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2030 | ||
2031 | /* Only revalidate flows if the configuration changed. */ | |
2032 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2033 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2034 | } |
2035 | ||
2036 | if (s) { | |
2037 | if (!ofproto->stp) { | |
2038 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2039 | send_bpdu_cb, ofproto); | |
2040 | ofproto->stp_last_tick = time_msec(); | |
2041 | } | |
2042 | ||
2043 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2044 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2045 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2046 | stp_set_max_age(ofproto->stp, s->max_age); | |
2047 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2048 | } else { | |
851bf71d EJ |
2049 | struct ofport *ofport; |
2050 | ||
2051 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2052 | set_stp_port(ofport, NULL); | |
2053 | } | |
2054 | ||
bd54dbcd | 2055 | stp_unref(ofproto->stp); |
21f7563c JP |
2056 | ofproto->stp = NULL; |
2057 | } | |
2058 | ||
2059 | return 0; | |
2060 | } | |
2061 | ||
2062 | static int | |
2063 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2064 | { | |
2065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2066 | ||
2067 | if (ofproto->stp) { | |
2068 | s->enabled = true; | |
2069 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2070 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2071 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2072 | } else { | |
2073 | s->enabled = false; | |
2074 | } | |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | static void | |
2080 | update_stp_port_state(struct ofport_dpif *ofport) | |
2081 | { | |
2082 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2083 | enum stp_state state; | |
2084 | ||
2085 | /* Figure out new state. */ | |
2086 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2087 | : STP_DISABLED; | |
2088 | ||
2089 | /* Update state. */ | |
2090 | if (ofport->stp_state != state) { | |
9e1fd49b | 2091 | enum ofputil_port_state of_state; |
21f7563c JP |
2092 | bool fwd_change; |
2093 | ||
2094 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2095 | netdev_get_name(ofport->up.netdev), | |
2096 | stp_state_name(ofport->stp_state), | |
2097 | stp_state_name(state)); | |
2098 | if (stp_learn_in_state(ofport->stp_state) | |
2099 | != stp_learn_in_state(state)) { | |
2100 | /* xxx Learning action flows should also be flushed. */ | |
509c0149 | 2101 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2102 | mac_learning_flush(ofproto->ml); |
509c0149 | 2103 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
21f7563c JP |
2104 | } |
2105 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2106 | != stp_forward_in_state(state); | |
2107 | ||
2cc3c58e | 2108 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2109 | ofport->stp_state = state; |
2110 | ofport->stp_state_entered = time_msec(); | |
2111 | ||
b308140a | 2112 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2113 | bundle_update(ofport->bundle); |
2114 | } | |
2115 | ||
2116 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2117 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2118 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2119 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2120 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2121 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2122 | : 0); | |
21f7563c JP |
2123 | ofproto_port_set_state(&ofport->up, of_state); |
2124 | } | |
2125 | } | |
2126 | ||
2127 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2128 | * caller is responsible for assigning STP port numbers and ensuring | |
2129 | * there are no duplicates. */ | |
2130 | static int | |
2131 | set_stp_port(struct ofport *ofport_, | |
2132 | const struct ofproto_port_stp_settings *s) | |
2133 | { | |
2134 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2135 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2136 | struct stp_port *sp = ofport->stp_port; | |
2137 | ||
2138 | if (!s || !s->enable) { | |
2139 | if (sp) { | |
2140 | ofport->stp_port = NULL; | |
2141 | stp_port_disable(sp); | |
ecd12731 | 2142 | update_stp_port_state(ofport); |
21f7563c JP |
2143 | } |
2144 | return 0; | |
2145 | } else if (sp && stp_port_no(sp) != s->port_num | |
2146 | && ofport == stp_port_get_aux(sp)) { | |
2147 | /* The port-id changed, so disable the old one if it's not | |
2148 | * already in use by another port. */ | |
2149 | stp_port_disable(sp); | |
2150 | } | |
2151 | ||
2152 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2153 | stp_port_enable(sp); | |
2154 | ||
2155 | stp_port_set_aux(sp, ofport); | |
2156 | stp_port_set_priority(sp, s->priority); | |
2157 | stp_port_set_path_cost(sp, s->path_cost); | |
2158 | ||
2159 | update_stp_port_state(ofport); | |
2160 | ||
2161 | return 0; | |
2162 | } | |
2163 | ||
2164 | static int | |
2165 | get_stp_port_status(struct ofport *ofport_, | |
2166 | struct ofproto_port_stp_status *s) | |
2167 | { | |
2168 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2169 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2170 | struct stp_port *sp = ofport->stp_port; | |
2171 | ||
2172 | if (!ofproto->stp || !sp) { | |
2173 | s->enabled = false; | |
2174 | return 0; | |
2175 | } | |
2176 | ||
2177 | s->enabled = true; | |
2178 | s->port_id = stp_port_get_id(sp); | |
2179 | s->state = stp_port_get_state(sp); | |
2180 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2181 | s->role = stp_port_get_role(sp); | |
fd28ce3a JS |
2182 | |
2183 | return 0; | |
2184 | } | |
2185 | ||
2186 | static int | |
2187 | get_stp_port_stats(struct ofport *ofport_, | |
2188 | struct ofproto_port_stp_stats *s) | |
2189 | { | |
2190 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2191 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2192 | struct stp_port *sp = ofport->stp_port; | |
2193 | ||
2194 | if (!ofproto->stp || !sp) { | |
2195 | s->enabled = false; | |
2196 | return 0; | |
2197 | } | |
2198 | ||
2199 | s->enabled = true; | |
80740385 | 2200 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2201 | |
2202 | return 0; | |
2203 | } | |
2204 | ||
2205 | static void | |
2206 | stp_run(struct ofproto_dpif *ofproto) | |
2207 | { | |
2208 | if (ofproto->stp) { | |
2209 | long long int now = time_msec(); | |
2210 | long long int elapsed = now - ofproto->stp_last_tick; | |
2211 | struct stp_port *sp; | |
2212 | ||
2213 | if (elapsed > 0) { | |
2214 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2215 | ofproto->stp_last_tick = now; | |
2216 | } | |
2217 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2218 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2219 | ||
2220 | if (ofport) { | |
2221 | update_stp_port_state(ofport); | |
2222 | } | |
2223 | } | |
6ae50723 EJ |
2224 | |
2225 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
509c0149 | 2226 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 2227 | mac_learning_flush(ofproto->ml); |
509c0149 | 2228 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
6ae50723 | 2229 | } |
21f7563c JP |
2230 | } |
2231 | } | |
2232 | ||
2233 | static void | |
2234 | stp_wait(struct ofproto_dpif *ofproto) | |
2235 | { | |
2236 | if (ofproto->stp) { | |
2237 | poll_timer_wait(1000); | |
2238 | } | |
2239 | } | |
21f7563c | 2240 | \f |
8b36f51e | 2241 | static int |
55954f6e | 2242 | set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp, |
8b36f51e EJ |
2243 | size_t n_qdscp) |
2244 | { | |
2245 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2246 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
8b36f51e | 2247 | |
55954f6e EJ |
2248 | if (ofport->n_qdscp != n_qdscp |
2249 | || (n_qdscp && memcmp(ofport->qdscp, qdscp, | |
2250 | n_qdscp * sizeof *qdscp))) { | |
2cc3c58e | 2251 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
55954f6e EJ |
2252 | free(ofport->qdscp); |
2253 | ofport->qdscp = n_qdscp | |
2254 | ? xmemdup(qdscp, n_qdscp * sizeof *qdscp) | |
2255 | : NULL; | |
2256 | ofport->n_qdscp = n_qdscp; | |
8b36f51e EJ |
2257 | } |
2258 | ||
8b36f51e EJ |
2259 | return 0; |
2260 | } | |
2261 | \f | |
abe529af BP |
2262 | /* Bundles. */ |
2263 | ||
b44a10b7 BP |
2264 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2265 | * ofproto to revalidate every flow. | |
2266 | * | |
2267 | * Normally MAC learning entries are removed only from the ofproto associated | |
2268 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2269 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2270 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2271 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2272 | * with the host from which it migrated. */ | |
abe529af | 2273 | static void |
b44a10b7 | 2274 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2275 | { |
2276 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2277 | struct mac_learning *ml = ofproto->ml; | |
2278 | struct mac_entry *mac, *next_mac; | |
2279 | ||
2cc3c58e | 2280 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
509c0149 | 2281 | ovs_rwlock_wrlock(&ml->rwlock); |
abe529af BP |
2282 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2283 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2284 | if (all_ofprotos) { |
2285 | struct ofproto_dpif *o; | |
2286 | ||
2287 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2288 | if (o != ofproto) { | |
2289 | struct mac_entry *e; | |
2290 | ||
509c0149 | 2291 | ovs_rwlock_wrlock(&o->ml->rwlock); |
30618594 | 2292 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan); |
b44a10b7 | 2293 | if (e) { |
b44a10b7 BP |
2294 | mac_learning_expire(o->ml, e); |
2295 | } | |
509c0149 | 2296 | ovs_rwlock_unlock(&o->ml->rwlock); |
b44a10b7 BP |
2297 | } |
2298 | } | |
2299 | } | |
2300 | ||
abe529af BP |
2301 | mac_learning_expire(ml, mac); |
2302 | } | |
2303 | } | |
509c0149 | 2304 | ovs_rwlock_unlock(&ml->rwlock); |
abe529af BP |
2305 | } |
2306 | ||
2307 | static struct ofbundle * | |
2308 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2309 | { | |
2310 | struct ofbundle *bundle; | |
2311 | ||
2312 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2313 | &ofproto->bundles) { | |
2314 | if (bundle->aux == aux) { | |
2315 | return bundle; | |
2316 | } | |
2317 | } | |
2318 | return NULL; | |
2319 | } | |
2320 | ||
7bde8dd8 JP |
2321 | static void |
2322 | bundle_update(struct ofbundle *bundle) | |
2323 | { | |
2324 | struct ofport_dpif *port; | |
2325 | ||
2326 | bundle->floodable = true; | |
2327 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b | 2328 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
a6363cfd | 2329 | || port->is_layer3 |
9e1fd49b | 2330 | || !stp_forward_in_state(port->stp_state)) { |
7bde8dd8 JP |
2331 | bundle->floodable = false; |
2332 | break; | |
2333 | } | |
2334 | } | |
2335 | } | |
2336 | ||
abe529af BP |
2337 | static void |
2338 | bundle_del_port(struct ofport_dpif *port) | |
2339 | { | |
2340 | struct ofbundle *bundle = port->bundle; | |
2341 | ||
2cc3c58e | 2342 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2343 | |
abe529af BP |
2344 | list_remove(&port->bundle_node); |
2345 | port->bundle = NULL; | |
2346 | ||
2347 | if (bundle->lacp) { | |
2348 | lacp_slave_unregister(bundle->lacp, port); | |
2349 | } | |
2350 | if (bundle->bond) { | |
2351 | bond_slave_unregister(bundle->bond, port); | |
2352 | } | |
2353 | ||
7bde8dd8 | 2354 | bundle_update(bundle); |
abe529af BP |
2355 | } |
2356 | ||
2357 | static bool | |
4e022ec0 | 2358 | bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, |
df53d41c | 2359 | struct lacp_slave_settings *lacp) |
abe529af BP |
2360 | { |
2361 | struct ofport_dpif *port; | |
2362 | ||
2363 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2364 | if (!port) { | |
2365 | return false; | |
2366 | } | |
2367 | ||
2368 | if (port->bundle != bundle) { | |
2cc3c58e | 2369 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 2370 | if (port->bundle) { |
e019a574 | 2371 | bundle_remove(&port->up); |
abe529af BP |
2372 | } |
2373 | ||
2374 | port->bundle = bundle; | |
2375 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b | 2376 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
a6363cfd | 2377 | || port->is_layer3 |
9e1fd49b | 2378 | || !stp_forward_in_state(port->stp_state)) { |
abe529af BP |
2379 | bundle->floodable = false; |
2380 | } | |
2381 | } | |
2382 | if (lacp) { | |
2cc3c58e | 2383 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2384 | lacp_slave_register(bundle->lacp, port, lacp); |
2385 | } | |
2386 | ||
2387 | return true; | |
2388 | } | |
2389 | ||
2390 | static void | |
2391 | bundle_destroy(struct ofbundle *bundle) | |
2392 | { | |
2393 | struct ofproto_dpif *ofproto; | |
2394 | struct ofport_dpif *port, *next_port; | |
abe529af BP |
2395 | |
2396 | if (!bundle) { | |
2397 | return; | |
2398 | } | |
2399 | ||
2400 | ofproto = bundle->ofproto; | |
ec7ceaed | 2401 | mbridge_unregister_bundle(ofproto->mbridge, bundle->aux); |
abe529af | 2402 | |
dc24a00f | 2403 | ovs_rwlock_wrlock(&xlate_rwlock); |
46c88433 | 2404 | xlate_bundle_remove(bundle); |
dc24a00f | 2405 | ovs_rwlock_unlock(&xlate_rwlock); |
46c88433 | 2406 | |
abe529af BP |
2407 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { |
2408 | bundle_del_port(port); | |
2409 | } | |
2410 | ||
b44a10b7 | 2411 | bundle_flush_macs(bundle, true); |
abe529af BP |
2412 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2413 | free(bundle->name); | |
2414 | free(bundle->trunks); | |
91779071 | 2415 | lacp_unref(bundle->lacp); |
03366a2d | 2416 | bond_unref(bundle->bond); |
abe529af BP |
2417 | free(bundle); |
2418 | } | |
2419 | ||
2420 | static int | |
2421 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2422 | const struct ofproto_bundle_settings *s) | |
2423 | { | |
2424 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2425 | bool need_flush = false; | |
abe529af BP |
2426 | struct ofport_dpif *port; |
2427 | struct ofbundle *bundle; | |
ecac4ebf BP |
2428 | unsigned long *trunks; |
2429 | int vlan; | |
abe529af BP |
2430 | size_t i; |
2431 | bool ok; | |
2432 | ||
2433 | if (!s) { | |
2434 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2435 | return 0; | |
2436 | } | |
2437 | ||
cb22974d BP |
2438 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2439 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2440 | |
2441 | bundle = bundle_lookup(ofproto, aux); | |
2442 | if (!bundle) { | |
2443 | bundle = xmalloc(sizeof *bundle); | |
2444 | ||
2445 | bundle->ofproto = ofproto; | |
2446 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2447 | hash_pointer(aux, 0)); | |
2448 | bundle->aux = aux; | |
2449 | bundle->name = NULL; | |
2450 | ||
2451 | list_init(&bundle->ports); | |
ecac4ebf | 2452 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2453 | bundle->vlan = -1; |
2454 | bundle->trunks = NULL; | |
5e9ceccd | 2455 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2456 | bundle->lacp = NULL; |
2457 | bundle->bond = NULL; | |
2458 | ||
2459 | bundle->floodable = true; | |
ec7ceaed | 2460 | mbridge_register_bundle(ofproto->mbridge, bundle); |
abe529af BP |
2461 | } |
2462 | ||
2463 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2464 | free(bundle->name); | |
2465 | bundle->name = xstrdup(s->name); | |
2466 | } | |
2467 | ||
2468 | /* LACP. */ | |
2469 | if (s->lacp) { | |
2470 | if (!bundle->lacp) { | |
2cc3c58e | 2471 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2472 | bundle->lacp = lacp_create(); |
2473 | } | |
2474 | lacp_configure(bundle->lacp, s->lacp); | |
2475 | } else { | |
91779071 | 2476 | lacp_unref(bundle->lacp); |
abe529af BP |
2477 | bundle->lacp = NULL; |
2478 | } | |
2479 | ||
2480 | /* Update set of ports. */ | |
2481 | ok = true; | |
2482 | for (i = 0; i < s->n_slaves; i++) { | |
2483 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2484 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2485 | ok = false; |
2486 | } | |
2487 | } | |
2488 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2489 | struct ofport_dpif *next_port; | |
2490 | ||
2491 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2492 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2493 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2494 | goto found; |
2495 | } | |
2496 | } | |
2497 | ||
2498 | bundle_del_port(port); | |
2499 | found: ; | |
2500 | } | |
2501 | } | |
cb22974d | 2502 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2503 | |
2504 | if (list_is_empty(&bundle->ports)) { | |
2505 | bundle_destroy(bundle); | |
2506 | return EINVAL; | |
2507 | } | |
2508 | ||
ecac4ebf | 2509 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2510 | if (s->vlan_mode != bundle->vlan_mode |
2511 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2512 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2513 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2514 | need_flush = true; |
2515 | } | |
2516 | ||
abe529af | 2517 | /* Set VLAN tag. */ |
ecac4ebf BP |
2518 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2519 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2520 | : 0); | |
2521 | if (vlan != bundle->vlan) { | |
2522 | bundle->vlan = vlan; | |
abe529af BP |
2523 | need_flush = true; |
2524 | } | |
2525 | ||
2526 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2527 | switch (s->vlan_mode) { |
2528 | case PORT_VLAN_ACCESS: | |
2529 | trunks = NULL; | |
2530 | break; | |
2531 | ||
2532 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2533 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2534 | break; |
2535 | ||
2536 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2537 | case PORT_VLAN_NATIVE_TAGGED: | |
2538 | if (vlan != 0 && (!s->trunks | |
2539 | || !bitmap_is_set(s->trunks, vlan) | |
2540 | || bitmap_is_set(s->trunks, 0))) { | |
2541 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2542 | if (s->trunks) { | |
2543 | trunks = bitmap_clone(s->trunks, 4096); | |
2544 | } else { | |
2545 | trunks = bitmap_allocate1(4096); | |
2546 | } | |
2547 | bitmap_set1(trunks, vlan); | |
2548 | bitmap_set0(trunks, 0); | |
2549 | } else { | |
ebc56baa | 2550 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2551 | } |
2552 | break; | |
2553 | ||
2554 | default: | |
2555 | NOT_REACHED(); | |
2556 | } | |
abe529af BP |
2557 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2558 | free(bundle->trunks); | |
ecac4ebf BP |
2559 | if (trunks == s->trunks) { |
2560 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2561 | } else { | |
2562 | bundle->trunks = trunks; | |
2563 | trunks = NULL; | |
2564 | } | |
abe529af BP |
2565 | need_flush = true; |
2566 | } | |
ecac4ebf BP |
2567 | if (trunks != s->trunks) { |
2568 | free(trunks); | |
2569 | } | |
abe529af BP |
2570 | |
2571 | /* Bonding. */ | |
2572 | if (!list_is_short(&bundle->ports)) { | |
2573 | bundle->ofproto->has_bonded_bundles = true; | |
2574 | if (bundle->bond) { | |
2575 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2576 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2577 | } |
2578 | } else { | |
2579 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2580 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2581 | } |
2582 | ||
2583 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2584 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2585 | } |
2586 | } else { | |
03366a2d | 2587 | bond_unref(bundle->bond); |
abe529af BP |
2588 | bundle->bond = NULL; |
2589 | } | |
2590 | ||
2591 | /* If we changed something that would affect MAC learning, un-learn | |
2592 | * everything on this port and force flow revalidation. */ | |
2593 | if (need_flush) { | |
b44a10b7 | 2594 | bundle_flush_macs(bundle, false); |
abe529af BP |
2595 | } |
2596 | ||
2597 | return 0; | |
2598 | } | |
2599 | ||
2600 | static void | |
2601 | bundle_remove(struct ofport *port_) | |
2602 | { | |
2603 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2604 | struct ofbundle *bundle = port->bundle; | |
2605 | ||
2606 | if (bundle) { | |
2607 | bundle_del_port(port); | |
2608 | if (list_is_empty(&bundle->ports)) { | |
2609 | bundle_destroy(bundle); | |
2610 | } else if (list_is_short(&bundle->ports)) { | |
03366a2d | 2611 | bond_unref(bundle->bond); |
abe529af BP |
2612 | bundle->bond = NULL; |
2613 | } | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | static void | |
5f877369 | 2618 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2619 | { |
2620 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2621 | struct ofport_dpif *port = port_; | |
2622 | uint8_t ea[ETH_ADDR_LEN]; | |
2623 | int error; | |
2624 | ||
2625 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2626 | if (!error) { | |
abe529af | 2627 | struct ofpbuf packet; |
5f877369 | 2628 | void *packet_pdu; |
abe529af BP |
2629 | |
2630 | ofpbuf_init(&packet, 0); | |
2631 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2632 | pdu_size); |
2633 | memcpy(packet_pdu, pdu, pdu_size); | |
2634 | ||
91d6cd12 | 2635 | ofproto_dpif_send_packet(port, &packet); |
abe529af BP |
2636 | ofpbuf_uninit(&packet); |
2637 | } else { | |
2638 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2639 | "%s (%s)", port->bundle->name, | |
10a89ef0 | 2640 | netdev_get_name(port->up.netdev), ovs_strerror(error)); |
abe529af BP |
2641 | } |
2642 | } | |
2643 | ||
2644 | static void | |
2645 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2646 | { | |
2647 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
0165217e | 2648 | struct ofpbuf *learning_packet; |
abe529af BP |
2649 | int error, n_packets, n_errors; |
2650 | struct mac_entry *e; | |
0165217e | 2651 | struct list packets; |
abe529af | 2652 | |
0165217e | 2653 | list_init(&packets); |
509c0149 | 2654 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
2655 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
2656 | if (e->port.p != bundle) { | |
4dd1e3ca | 2657 | void *port_void; |
ea131871 | 2658 | |
4dd1e3ca BP |
2659 | learning_packet = bond_compose_learning_packet(bundle->bond, |
2660 | e->mac, e->vlan, | |
2661 | &port_void); | |
0165217e EJ |
2662 | learning_packet->private_p = port_void; |
2663 | list_push_back(&packets, &learning_packet->list_node); | |
abe529af BP |
2664 | } |
2665 | } | |
509c0149 | 2666 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af | 2667 | |
0165217e EJ |
2668 | error = n_packets = n_errors = 0; |
2669 | LIST_FOR_EACH (learning_packet, list_node, &packets) { | |
2670 | int ret; | |
2671 | ||
91d6cd12 | 2672 | ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet); |
0165217e EJ |
2673 | if (ret) { |
2674 | error = ret; | |
2675 | n_errors++; | |
2676 | } | |
2677 | n_packets++; | |
2678 | } | |
2679 | ofpbuf_list_delete(&packets); | |
2680 | ||
abe529af BP |
2681 | if (n_errors) { |
2682 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2683 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2684 | "packets, last error was: %s", | |
10a89ef0 | 2685 | bundle->name, n_errors, n_packets, ovs_strerror(error)); |
abe529af BP |
2686 | } else { |
2687 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2688 | bundle->name, n_packets); | |
2689 | } | |
2690 | } | |
2691 | ||
2692 | static void | |
2693 | bundle_run(struct ofbundle *bundle) | |
2694 | { | |
2695 | if (bundle->lacp) { | |
2696 | lacp_run(bundle->lacp, send_pdu_cb); | |
2697 | } | |
2698 | if (bundle->bond) { | |
2699 | struct ofport_dpif *port; | |
2700 | ||
2701 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2702 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2703 | } |
2704 | ||
4a1b8f30 EJ |
2705 | if (bond_run(bundle->bond, lacp_status(bundle->lacp))) { |
2706 | bundle->ofproto->backer->need_revalidate = REV_BOND; | |
2707 | } | |
2708 | ||
abe529af BP |
2709 | if (bond_should_send_learning_packets(bundle->bond)) { |
2710 | bundle_send_learning_packets(bundle); | |
2711 | } | |
2712 | } | |
2713 | } | |
2714 | ||
2715 | static void | |
2716 | bundle_wait(struct ofbundle *bundle) | |
2717 | { | |
2718 | if (bundle->lacp) { | |
2719 | lacp_wait(bundle->lacp); | |
2720 | } | |
2721 | if (bundle->bond) { | |
2722 | bond_wait(bundle->bond); | |
2723 | } | |
2724 | } | |
2725 | \f | |
2726 | /* Mirrors. */ | |
2727 | ||
2728 | static int | |
ec7ceaed EJ |
2729 | mirror_set__(struct ofproto *ofproto_, void *aux, |
2730 | const struct ofproto_mirror_settings *s) | |
abe529af BP |
2731 | { |
2732 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
ec7ceaed EJ |
2733 | struct ofbundle **srcs, **dsts; |
2734 | int error; | |
2735 | size_t i; | |
abe529af | 2736 | |
abe529af | 2737 | if (!s) { |
ec7ceaed | 2738 | mirror_destroy(ofproto->mbridge, aux); |
abe529af BP |
2739 | return 0; |
2740 | } | |
2741 | ||
ec7ceaed EJ |
2742 | srcs = xmalloc(s->n_srcs * sizeof *srcs); |
2743 | dsts = xmalloc(s->n_dsts * sizeof *dsts); | |
abe529af | 2744 | |
ec7ceaed EJ |
2745 | for (i = 0; i < s->n_srcs; i++) { |
2746 | srcs[i] = bundle_lookup(ofproto, s->srcs[i]); | |
abe529af BP |
2747 | } |
2748 | ||
ec7ceaed EJ |
2749 | for (i = 0; i < s->n_dsts; i++) { |
2750 | dsts[i] = bundle_lookup(ofproto, s->dsts[i]); | |
abe529af BP |
2751 | } |
2752 | ||
ec7ceaed EJ |
2753 | error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts, |
2754 | s->n_dsts, s->src_vlans, | |
2755 | bundle_lookup(ofproto, s->out_bundle), s->out_vlan); | |
2756 | free(srcs); | |
2757 | free(dsts); | |
2758 | return error; | |
abe529af BP |
2759 | } |
2760 | ||
9d24de3b | 2761 | static int |
ec7ceaed EJ |
2762 | mirror_get_stats__(struct ofproto *ofproto, void *aux, |
2763 | uint64_t *packets, uint64_t *bytes) | |
9d24de3b | 2764 | { |
8844e035 | 2765 | push_all_stats(); |
ec7ceaed EJ |
2766 | return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets, |
2767 | bytes); | |
9d24de3b JP |
2768 | } |
2769 | ||
abe529af BP |
2770 | static int |
2771 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2772 | { | |
2773 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2774 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
abe529af | 2775 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { |
30618594 | 2776 | mac_learning_flush(ofproto->ml); |
abe529af | 2777 | } |
509c0149 | 2778 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
abe529af BP |
2779 | return 0; |
2780 | } | |
2781 | ||
2782 | static bool | |
b4affc74 | 2783 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2784 | { |
2785 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2786 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
ec7ceaed | 2787 | return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0; |
abe529af | 2788 | } |
8402c74b SS |
2789 | |
2790 | static void | |
b53055f4 | 2791 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2792 | { |
2793 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2794 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2795 | } |
e764773c BP |
2796 | |
2797 | static void | |
c4069512 BP |
2798 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2799 | size_t max_entries) | |
e764773c BP |
2800 | { |
2801 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
509c0149 | 2802 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
e764773c | 2803 | mac_learning_set_idle_time(ofproto->ml, idle_time); |
c4069512 | 2804 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
509c0149 | 2805 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
e764773c | 2806 | } |
abe529af BP |
2807 | \f |
2808 | /* Ports. */ | |
2809 | ||
46c88433 | 2810 | static struct ofport_dpif * |
4e022ec0 | 2811 | get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
abe529af | 2812 | { |
7df6a8bd BP |
2813 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2814 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2815 | } |
2816 | ||
46c88433 | 2817 | static struct ofport_dpif * |
4e022ec0 | 2818 | get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) |
abe529af | 2819 | { |
7c33b188 JR |
2820 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2821 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2822 | } |
2823 | ||
2824 | static void | |
e1b1d06a JP |
2825 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2826 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2827 | struct dpif_port *dpif_port) |
2828 | { | |
2829 | ofproto_port->name = dpif_port->name; | |
2830 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2831 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2832 | } |
2833 | ||
6cbbf4fa EJ |
2834 | static void |
2835 | ofport_update_peer(struct ofport_dpif *ofport) | |
0a740f48 EJ |
2836 | { |
2837 | const struct ofproto_dpif *ofproto; | |
6cbbf4fa | 2838 | struct dpif_backer *backer; |
161b6042 | 2839 | char *peer_name; |
6cbbf4fa EJ |
2840 | |
2841 | if (!netdev_vport_is_patch(ofport->up.netdev)) { | |
2842 | return; | |
2843 | } | |
2844 | ||
2845 | backer = ofproto_dpif_cast(ofport->up.ofproto)->backer; | |
7894e7da | 2846 | backer->need_revalidate = REV_RECONFIGURE; |
0a740f48 | 2847 | |
6cbbf4fa EJ |
2848 | if (ofport->peer) { |
2849 | ofport->peer->peer = NULL; | |
2850 | ofport->peer = NULL; | |
2851 | } | |
2852 | ||
2853 | peer_name = netdev_vport_patch_peer(ofport->up.netdev); | |
2854 | if (!peer_name) { | |
2855 | return; | |
0a740f48 EJ |
2856 | } |
2857 | ||
2858 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6cbbf4fa EJ |
2859 | struct ofport *peer_ofport; |
2860 | struct ofport_dpif *peer; | |
161b6042 | 2861 | char *peer_peer; |
0a740f48 | 2862 | |
462abef2 EJ |
2863 | if (ofproto->backer != backer) { |
2864 | continue; | |
2865 | } | |
2866 | ||
6cbbf4fa EJ |
2867 | peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name); |
2868 | if (!peer_ofport) { | |
2869 | continue; | |
2870 | } | |
2871 | ||
2872 | peer = ofport_dpif_cast(peer_ofport); | |
2873 | peer_peer = netdev_vport_patch_peer(peer->up.netdev); | |
2874 | if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev), | |
2875 | peer_peer)) { | |
2876 | ofport->peer = peer; | |
2877 | ofport->peer->peer = ofport; | |
0a740f48 | 2878 | } |
161b6042 | 2879 | free(peer_peer); |
6cbbf4fa | 2880 | |
161b6042 | 2881 | break; |
0a740f48 | 2882 | } |
161b6042 | 2883 | free(peer_name); |
0a740f48 EJ |
2884 | } |
2885 | ||
abe529af BP |
2886 | static void |
2887 | port_run(struct ofport_dpif *ofport) | |
2888 | { | |
3e5b3fdb EJ |
2889 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2890 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc | 2891 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2d344ba5 AW |
2892 | bool cfm_enable = false; |
2893 | bool bfd_enable = false; | |
015e08bc | 2894 | |
3e5b3fdb EJ |
2895 | ofport->carrier_seq = carrier_seq; |
2896 | ||
abe529af | 2897 | if (ofport->cfm) { |
4653c558 EJ |
2898 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2899 | ||
2d344ba5 | 2900 | cfm_enable = !cfm_get_fault(ofport->cfm); |
4653c558 EJ |
2901 | |
2902 | if (cfm_opup >= 0) { | |
2d344ba5 | 2903 | cfm_enable = cfm_enable && cfm_opup; |
4653c558 | 2904 | } |
abe529af | 2905 | } |
015e08bc | 2906 | |
ccc09689 | 2907 | if (ofport->bfd) { |
2d344ba5 AW |
2908 | bfd_enable = bfd_forwarding(ofport->bfd); |
2909 | } | |
2910 | ||
2911 | if (ofport->bfd || ofport->cfm) { | |
2912 | enable = enable && (cfm_enable || bfd_enable); | |
ccc09689 EJ |
2913 | } |
2914 | ||
015e08bc EJ |
2915 | if (ofport->bundle) { |
2916 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2917 | if (carrier_changed) { |
2918 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2919 | } | |
015e08bc EJ |
2920 | } |
2921 | ||
daff3353 EJ |
2922 | if (ofport->may_enable != enable) { |
2923 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
83341407 | 2924 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
2925 | } |
2926 | ||
015e08bc | 2927 | ofport->may_enable = enable; |
abe529af BP |
2928 | } |
2929 | ||
abe529af BP |
2930 | static int |
2931 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2932 | struct ofproto_port *ofproto_port) | |
2933 | { | |
2934 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2935 | struct dpif_port dpif_port; | |
2936 | int error; | |
2937 | ||
0a740f48 EJ |
2938 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
2939 | const char *type = netdev_get_type_from_name(devname); | |
2940 | ||
2941 | /* We may be called before ofproto->up.port_by_name is populated with | |
2942 | * the appropriate ofport. For this reason, we must get the name and | |
2943 | * type from the netdev layer directly. */ | |
2944 | if (type) { | |
2945 | const struct ofport *ofport; | |
2946 | ||
2947 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
2948 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
2949 | ofproto_port->name = xstrdup(devname); | |
2950 | ofproto_port->type = xstrdup(type); | |
2951 | return 0; | |
2952 | } | |
2953 | return ENODEV; | |
2954 | } | |
2955 | ||
acf60855 JP |
2956 | if (!sset_contains(&ofproto->ports, devname)) { |
2957 | return ENODEV; | |
2958 | } | |
2959 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
2960 | devname, &dpif_port); | |
abe529af | 2961 | if (!error) { |
e1b1d06a | 2962 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
2963 | } |
2964 | return error; | |
2965 | } | |
2966 | ||
2967 | static int | |
e1b1d06a | 2968 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
2969 | { |
2970 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 2971 | const char *devname = netdev_get_name(netdev); |
3aa30359 BP |
2972 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
2973 | const char *dp_port_name; | |
abe529af | 2974 | |
0a740f48 EJ |
2975 | if (netdev_vport_is_patch(netdev)) { |
2976 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
2977 | return 0; | |
2978 | } | |
2979 | ||
3aa30359 | 2980 | dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
b9ad7294 | 2981 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
4e022ec0 | 2982 | odp_port_t port_no = ODPP_NONE; |
7d82ab2e KM |
2983 | int error; |
2984 | ||
2985 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
2986 | if (error) { |
2987 | return error; | |
2988 | } | |
7d82ab2e | 2989 | if (netdev_get_tunnel_config(netdev)) { |
4e022ec0 AW |
2990 | simap_put(&ofproto->backer->tnl_backers, |
2991 | dp_port_name, odp_to_u32(port_no)); | |
7d82ab2e | 2992 | } |
acf60855 | 2993 | } |
b9ad7294 EJ |
2994 | |
2995 | if (netdev_get_tunnel_config(netdev)) { | |
2996 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
2997 | } else { |
2998 | sset_add(&ofproto->ports, devname); | |
2999 | } | |
3000 | return 0; | |
3001 | } | |
3002 | ||
abe529af | 3003 | static int |
4e022ec0 | 3004 | port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) |
abe529af BP |
3005 | { |
3006 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3007 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3008 | int error = 0; |
abe529af | 3009 | |
b9ad7294 EJ |
3010 | if (!ofport) { |
3011 | return 0; | |
e1b1d06a | 3012 | } |
b9ad7294 EJ |
3013 | |
3014 | sset_find_and_delete(&ofproto->ghost_ports, | |
3015 | netdev_get_name(ofport->up.netdev)); | |
a614d823 | 3016 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8911c674 | 3017 | if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) { |
b9ad7294 EJ |
3018 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3019 | if (!error) { | |
abe529af BP |
3020 | /* The caller is going to close ofport->up.netdev. If this is a |
3021 | * bonded port, then the bond is using that netdev, so remove it | |
3022 | * from the bond. The client will need to reconfigure everything | |
3023 | * after deleting ports, so then the slave will get re-added. */ | |
3024 | bundle_remove(&ofport->up); | |
3025 | } | |
3026 | } | |
3027 | return error; | |
3028 | } | |
3029 | ||
6527c598 PS |
3030 | static int |
3031 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3032 | { | |
3033 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3034 | int error; | |
3035 | ||
8844e035 EJ |
3036 | push_all_stats(); |
3037 | ||
6527c598 PS |
3038 | error = netdev_get_stats(ofport->up.netdev, stats); |
3039 | ||
ee382d89 | 3040 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3041 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3042 | ||
0da3c61b | 3043 | ovs_mutex_lock(&ofproto->stats_mutex); |
6527c598 PS |
3044 | /* ofproto->stats.tx_packets represents packets that we created |
3045 | * internally and sent to some port (e.g. packets sent with | |
91d6cd12 AW |
3046 | * ofproto_dpif_send_packet()). Account for them as if they had |
3047 | * come from OFPP_LOCAL and got forwarded. */ | |
6527c598 PS |
3048 | |
3049 | if (stats->rx_packets != UINT64_MAX) { | |
3050 | stats->rx_packets += ofproto->stats.tx_packets; | |
3051 | } | |
3052 | ||
3053 | if (stats->rx_bytes != UINT64_MAX) { | |
3054 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3055 | } | |
3056 | ||
3057 | /* ofproto->stats.rx_packets represents packets that were received on | |
3058 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3059 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3060 | |
3061 | if (stats->tx_packets != UINT64_MAX) { | |
3062 | stats->tx_packets += ofproto->stats.rx_packets; | |
3063 | } | |
3064 | ||
3065 | if (stats->tx_bytes != UINT64_MAX) { | |
3066 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3067 | } | |
0da3c61b | 3068 | ovs_mutex_unlock(&ofproto->stats_mutex); |
6527c598 PS |
3069 | } |
3070 | ||
3071 | return error; | |
3072 | } | |
3073 | ||
abe529af | 3074 | struct port_dump_state { |
acf60855 JP |
3075 | uint32_t bucket; |
3076 | uint32_t offset; | |
0a740f48 | 3077 | bool ghost; |
da78d43d BP |
3078 | |
3079 | struct ofproto_port port; | |
3080 | bool has_port; | |
abe529af BP |
3081 | }; |
3082 | ||
3083 | static int | |
acf60855 | 3084 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3085 | { |
0a740f48 | 3086 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3087 | return 0; |
3088 | } | |
3089 | ||
3090 | static int | |
b9ad7294 | 3091 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3092 | struct ofproto_port *port) |
3093 | { | |
e1b1d06a | 3094 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3095 | struct port_dump_state *state = state_; |
0a740f48 | 3096 | const struct sset *sset; |
acf60855 | 3097 | struct sset_node *node; |
abe529af | 3098 | |
da78d43d BP |
3099 | if (state->has_port) { |
3100 | ofproto_port_destroy(&state->port); | |
3101 | state->has_port = false; | |
3102 | } | |
0a740f48 EJ |
3103 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3104 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3105 | int error; |
3106 | ||
da78d43d BP |
3107 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3108 | if (!error) { | |
3109 | *port = state->port; | |
3110 | state->has_port = true; | |
3111 | return 0; | |
3112 | } else if (error != ENODEV) { | |
acf60855 JP |
3113 | return error; |
3114 | } | |
abe529af | 3115 | } |
acf60855 | 3116 | |
0a740f48 EJ |
3117 | if (!state->ghost) { |
3118 | state->ghost = true; | |
3119 | state->bucket = 0; | |
3120 | state->offset = 0; | |
3121 | return port_dump_next(ofproto_, state_, port); | |
3122 | } | |
3123 | ||
acf60855 | 3124 | return EOF; |
abe529af BP |
3125 | } |
3126 | ||
3127 | static int | |
3128 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3129 | { | |
3130 | struct port_dump_state *state = state_; | |
3131 | ||
da78d43d BP |
3132 | if (state->has_port) { |
3133 | ofproto_port_destroy(&state->port); | |
3134 | } | |
abe529af BP |
3135 | free(state); |
3136 | return 0; | |
3137 | } | |
3138 | ||
3139 | static int | |
3140 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3141 | { | |
3142 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3143 | |
3144 | if (ofproto->port_poll_errno) { | |
3145 | int error = ofproto->port_poll_errno; | |
3146 | ofproto->port_poll_errno = 0; | |
3147 | return error; | |
3148 | } | |
3149 | ||
3150 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3151 | return EAGAIN; | |
3152 | } | |
3153 | ||
3154 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3155 | return 0; | |
abe529af BP |
3156 | } |
3157 | ||
3158 | static void | |
3159 | port_poll_wait(const struct ofproto *ofproto_) | |
3160 | { | |
3161 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3162 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3163 | } |
3164 | ||
3165 | static int | |
3166 | port_is_lacp_current(const struct ofport *ofport_) | |
3167 | { | |
3168 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3169 | return (ofport->bundle && ofport->bundle->lacp | |
3170 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3171 | : -1); | |
3172 | } | |
3173 | \f | |
3174 | /* Upcall handling. */ | |
3175 | ||
501f8d1f | 3176 | struct flow_miss_op { |
c2b565b5 | 3177 | struct dpif_op dpif_op; |
bbafd73b EJ |
3178 | |
3179 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3180 | struct xlate_out xout; | |
3181 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
9b1a48c2 | 3182 | |
d445cc16 JP |
3183 | struct ofpbuf mask; /* Flow mask for "put" ops. */ |
3184 | struct odputil_keybuf maskbuf; | |
3185 | ||
9b1a48c2 JP |
3186 | /* If this is a "put" op, then a pointer to the subfacet that should |
3187 | * be marked as uninstalled if the operation fails. */ | |
3188 | struct subfacet *subfacet; | |
501f8d1f BP |
3189 | }; |
3190 | ||
9d6ac44e | 3191 | /* Figures out whether a flow that missed in 'ofproto', whose details are in |
bcd2633a JP |
3192 | * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace |
3193 | * and (usually) installing a datapath flow. The answer is usually "yes" (a | |
3194 | * return value of true). However, for short flows the cost of bookkeeping is | |
3195 | * much higher than the benefits, so when the datapath holds a large number of | |
3196 | * flows we impose some heuristics to decide which flows are likely to be worth | |
3197 | * tracking. */ | |
9d6ac44e | 3198 | static bool |
e1ec7dd4 | 3199 | flow_miss_should_make_facet(struct flow_miss *miss) |
9d6ac44e | 3200 | { |
04d08d54 | 3201 | struct dpif_backer *backer = miss->ofproto->backer; |
bcd2633a | 3202 | uint32_t hash; |
04d08d54 | 3203 | |
7155fa52 JS |
3204 | switch (flow_miss_model) { |
3205 | case OFPROTO_HANDLE_MISS_AUTO: | |
3206 | break; | |
3207 | case OFPROTO_HANDLE_MISS_WITH_FACETS: | |
3208 | return true; | |
3209 | case OFPROTO_HANDLE_MISS_WITHOUT_FACETS: | |
3210 | return false; | |
3211 | } | |
3212 | ||
04d08d54 | 3213 | if (!backer->governor) { |
9d6ac44e BP |
3214 | size_t n_subfacets; |
3215 | ||
04d08d54 | 3216 | n_subfacets = hmap_count(&backer->subfacets); |
380f49c4 | 3217 | if (n_subfacets * 2 <= flow_eviction_threshold) { |
9d6ac44e BP |
3218 | return true; |
3219 | } | |
3220 | ||
c985ec94 | 3221 | backer->governor = governor_create(); |
9d6ac44e BP |
3222 | } |
3223 | ||
e1ec7dd4 | 3224 | hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0); |
04d08d54 | 3225 | return governor_should_install_flow(backer->governor, hash, |
15e34f24 | 3226 | miss->stats.n_packets); |
9d6ac44e BP |
3227 | } |
3228 | ||
9d6ac44e | 3229 | /* Handles 'miss', which matches 'facet'. May add any required datapath |
459b16a1 BP |
3230 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3231 | * | |
e1ec7dd4 EJ |
3232 | * All of the packets in 'miss' are considered to have arrived at time |
3233 | * 'miss->stats.used'. This is really important only for new facets: if we | |
3234 | * just called time_msec() here, then the new subfacet or its packets could | |
3235 | * look (occasionally) as though it was used some time after the facet was | |
3236 | * used. That can make a one-packet flow look like it has a nonzero duration, | |
3237 | * which looks odd in e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3238 | static void |
3239 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
3240 | struct flow_miss_op *ops, size_t *n_ops) | |
3241 | { | |
6a7e895f | 3242 | enum subfacet_path want_path; |
9d6ac44e | 3243 | struct subfacet *subfacet; |
f9c37233 | 3244 | uint32_t key_hash; |
67d91f78 | 3245 | |
f9c37233 | 3246 | /* Update facet stats. */ |
e1ec7dd4 EJ |
3247 | facet->packet_count += miss->stats.n_packets; |
3248 | facet->prev_packet_count += miss->stats.n_packets; | |
3249 | facet->byte_count += miss->stats.n_bytes; | |
3250 | facet->prev_byte_count += miss->stats.n_bytes; | |
501f8d1f | 3251 | |
f9c37233 BP |
3252 | /* Look for an existing subfacet. If we find one, update its used time. */ |
3253 | key_hash = odp_flow_key_hash(miss->key, miss->key_len); | |
3254 | if (!list_is_empty(&facet->subfacets)) { | |
3255 | subfacet = subfacet_find(miss->ofproto->backer, | |
3256 | miss->key, miss->key_len, key_hash); | |
3257 | if (subfacet) { | |
3258 | if (subfacet->facet == facet) { | |
3259 | subfacet->used = MAX(subfacet->used, miss->stats.used); | |
3260 | } else { | |
3261 | /* This shouldn't happen. */ | |
3262 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
3263 | subfacet_destroy(subfacet); | |
3264 | subfacet = NULL; | |
3265 | } | |
3266 | } | |
3267 | } else { | |
3268 | subfacet = NULL; | |
3269 | } | |
501f8d1f | 3270 | |
a41d7bf2 JP |
3271 | /* Don't install the flow if it's the result of the "userspace" |
3272 | * action for an already installed facet. This can occur when a | |
3273 | * datapath flow with wildcards has a "userspace" action and flows | |
3274 | * sent to userspace result in a different subfacet, which will then | |
3275 | * be rejected as overlapping by the datapath. */ | |
3276 | if (miss->upcall_type == DPIF_UC_ACTION | |
3277 | && !list_is_empty(&facet->subfacets)) { | |
a41d7bf2 JP |
3278 | return; |
3279 | } | |
3280 | ||
f9c37233 BP |
3281 | /* Create a subfacet, if we don't already have one. */ |
3282 | if (!subfacet) { | |
3283 | subfacet = subfacet_create(facet, miss, key_hash); | |
3284 | } | |
3285 | ||
3286 | /* Install the subfacet, if it's not already installed. */ | |
3287 | want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
24ce4b08 | 3288 | if (subfacet->path != want_path) { |
501f8d1f | 3289 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3290 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3291 | |
c84451a6 EJ |
3292 | subfacet->path = want_path; |
3293 | ||
d445cc16 | 3294 | ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); |
6b35e630 JP |
3295 | if (enable_megaflows) { |
3296 | odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks, | |
3297 | &miss->flow, UINT32_MAX); | |
3298 | } | |
d445cc16 | 3299 | |
bbafd73b | 3300 | op->xout_garbage = false; |
c2b565b5 | 3301 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
9b1a48c2 | 3302 | op->subfacet = subfacet; |
0f74f625 | 3303 | put->flags = DPIF_FP_CREATE; |
501f8d1f BP |
3304 | put->key = miss->key; |
3305 | put->key_len = miss->key_len; | |
d445cc16 JP |
3306 | put->mask = op->mask.data; |
3307 | put->mask_len = op->mask.size; | |
3308 | ||
6a7e895f | 3309 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3310 | put->actions = facet->xout.odp_actions.data; |
3311 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3312 | } else { |
b7e2f6b3 | 3313 | compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow, |
bbafd73b | 3314 | op->slow_stub, sizeof op->slow_stub, |
6a7e895f BP |
3315 | &put->actions, &put->actions_len); |
3316 | } | |
501f8d1f BP |
3317 | put->stats = NULL; |
3318 | } | |
3319 | } | |
3320 | ||
acf60855 JP |
3321 | /* Handles flow miss 'miss'. May add any required datapath operations |
3322 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3323 | static void |
acf60855 JP |
3324 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3325 | size_t *n_ops) | |
9d6ac44e BP |
3326 | { |
3327 | struct facet *facet; | |
9d6ac44e | 3328 | |
15e34f24 | 3329 | miss->ofproto->n_missed += miss->stats.n_packets; |
9d6ac44e | 3330 | |
e1ec7dd4 | 3331 | facet = facet_lookup_valid(miss->ofproto, &miss->flow); |
9d6ac44e | 3332 | if (!facet) { |
ba33dd03 EJ |
3333 | /* There does not exist a bijection between 'struct flow' and datapath |
3334 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3335 | * assumption used throughout the facet and subfacet handling code. | |
3336 | * Since we have to handle these misses in userspace anyway, we simply | |
ec9f40dc | 3337 | * skip facet creation, avoiding the problem altogether. */ |
ba33dd03 | 3338 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE |
e1ec7dd4 | 3339 | || !flow_miss_should_make_facet(miss)) { |
9d6ac44e BP |
3340 | return; |
3341 | } | |
3342 | ||
e1ec7dd4 | 3343 | facet = facet_create(miss); |
9d6ac44e | 3344 | } |
e1ec7dd4 | 3345 | handle_flow_miss_with_facet(miss, facet, ops, n_ops); |
9d6ac44e BP |
3346 | } |
3347 | ||
8f73d537 EJ |
3348 | static struct drop_key * |
3349 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3350 | size_t key_len) | |
3351 | { | |
3352 | struct drop_key *drop_key; | |
3353 | ||
3354 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3355 | &backer->drop_keys) { | |
3356 | if (drop_key->key_len == key_len | |
3357 | && !memcmp(drop_key->key, key, key_len)) { | |
3358 | return drop_key; | |
3359 | } | |
3360 | } | |
3361 | return NULL; | |
3362 | } | |
3363 | ||
3364 | static void | |
3365 | drop_key_clear(struct dpif_backer *backer) | |
3366 | { | |
3367 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3368 | struct drop_key *drop_key, *next; | |
3369 | ||
3370 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3371 | int error; | |
3372 | ||
3373 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3374 | NULL); | |
3375 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3376 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3377 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
10a89ef0 BP |
3378 | VLOG_WARN("Failed to delete drop key (%s) (%s)", |
3379 | ovs_strerror(error), ds_cstr(&ds)); | |
8f73d537 EJ |
3380 | ds_destroy(&ds); |
3381 | } | |
3382 | ||
3383 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
e1ec7dd4 | 3384 | drop_key_destroy(drop_key); |
8f73d537 | 3385 | } |
e1ec7dd4 EJ |
3386 | |
3387 | udpif_drop_key_clear(backer->udpif); | |
8f73d537 EJ |
3388 | } |
3389 | ||
501f8d1f | 3390 | static void |
e1ec7dd4 | 3391 | handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb) |
501f8d1f | 3392 | { |
e1ec7dd4 EJ |
3393 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH]; |
3394 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH]; | |
b23cdad9 | 3395 | struct flow_miss *miss; |
e1ec7dd4 | 3396 | size_t n_ops, i; |
501f8d1f BP |
3397 | |
3398 | /* Process each element in the to-do list, constructing the set of | |
3399 | * operations to batch. */ | |
3400 | n_ops = 0; | |
e1ec7dd4 | 3401 | HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) { |
acf60855 | 3402 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3403 | } |
cb22974d | 3404 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3405 | |
3406 | /* Execute batch. */ | |
3407 | for (i = 0; i < n_ops; i++) { | |
3408 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3409 | } | |
acf60855 | 3410 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 3411 | |
501f8d1f | 3412 | for (i = 0; i < n_ops; i++) { |
9b1a48c2 JP |
3413 | if (dpif_ops[i]->error != 0 |
3414 | && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT | |
3415 | && flow_miss_ops[i].subfacet) { | |
3416 | struct subfacet *subfacet = flow_miss_ops[i].subfacet; | |
3417 | ||
3418 | COVERAGE_INC(subfacet_install_fail); | |
3419 | ||
cfd5c9eb JP |
3420 | /* Zero-out subfacet counters when installation failed, but |
3421 | * datapath reported hits. This should not happen and | |
3422 | * indicates a bug, since if the datapath flow exists, we | |
3423 | * should not be attempting to create a new subfacet. A | |
3424 | * buggy datapath could trigger this, so just zero out the | |
3425 | * counters and log an error. */ | |
3426 | if (subfacet->dp_packet_count || subfacet->dp_byte_count) { | |
3427 | VLOG_ERR_RL(&rl, "failed to install subfacet for which " | |
3428 | "datapath reported hits"); | |
3429 | subfacet->dp_packet_count = subfacet->dp_byte_count = 0; | |
3430 | } | |
3431 | ||
9b1a48c2 JP |
3432 | subfacet->path = SF_NOT_INSTALLED; |
3433 | } | |
6a7e895f BP |
3434 | } |
3435 | } | |
3436 | ||
e1ec7dd4 EJ |
3437 | static void |
3438 | handle_upcalls(struct dpif_backer *backer) | |
6ff686f2 | 3439 | { |
e1ec7dd4 | 3440 | struct flow_miss_batch *fmb; |
90a7c55e | 3441 | int n_processed; |
abe529af | 3442 | |
e1ec7dd4 EJ |
3443 | for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) { |
3444 | struct drop_key *drop_key = drop_key_next(backer->udpif); | |
3445 | if (!drop_key) { | |
3446 | break; | |
3447 | } | |
3448 | ||
3449 | if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) { | |
3450 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3451 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3452 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3453 | drop_key->key, drop_key->key_len, | |
3454 | NULL, 0, NULL, 0, NULL); | |
3455 | } else { | |
3456 | drop_key_destroy(drop_key); | |
3457 | } | |
90a7c55e | 3458 | } |
9b16c439 | 3459 | |
e1ec7dd4 EJ |
3460 | fmb = flow_miss_batch_next(backer->udpif); |
3461 | if (fmb) { | |
3462 | handle_flow_misses(backer, fmb); | |
3463 | flow_miss_batch_destroy(fmb); | |
3464 | } | |
abe529af BP |
3465 | } |
3466 | \f | |
3467 | /* Flow expiration. */ | |
3468 | ||
04d08d54 | 3469 | static int subfacet_max_idle(const struct dpif_backer *); |
acf60855 | 3470 | static void update_stats(struct dpif_backer *); |
15aaf599 | 3471 | static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex); |
04d08d54 | 3472 | static void expire_subfacets(struct dpif_backer *, int dp_max_idle); |
abe529af BP |
3473 | |
3474 | /* This function is called periodically by run(). Its job is to collect | |
3475 | * updates for the flows that have been installed into the datapath, most | |
3476 | * importantly when they last were used, and then use that information to | |
3477 | * expire flows that have not been used recently. | |
3478 | * | |
3479 | * Returns the number of milliseconds after which it should be called again. */ | |
3480 | static int | |
acf60855 | 3481 | expire(struct dpif_backer *backer) |
abe529af | 3482 | { |
acf60855 | 3483 | struct ofproto_dpif *ofproto; |
dc54ef36 | 3484 | size_t n_subfacets; |
04d08d54 | 3485 | int max_idle; |
abe529af | 3486 | |
8f73d537 EJ |
3487 | /* Periodically clear out the drop keys in an effort to keep them |
3488 | * relatively few. */ | |
3489 | drop_key_clear(backer); | |
3490 | ||
acf60855 JP |
3491 | /* Update stats for each flow in the backer. */ |
3492 | update_stats(backer); | |
abe529af | 3493 | |
04d08d54 EJ |
3494 | n_subfacets = hmap_count(&backer->subfacets); |
3495 | if (n_subfacets) { | |
9fc0165a | 3496 | struct subfacet *subfacet; |
04d08d54 | 3497 | long long int total, now; |
abe529af | 3498 | |
04d08d54 EJ |
3499 | total = 0; |
3500 | now = time_msec(); | |
3501 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { | |
3502 | total += now - subfacet->created; | |
acf60855 | 3503 | } |
04d08d54 EJ |
3504 | backer->avg_subfacet_life += total / n_subfacets; |
3505 | } | |
3506 | backer->avg_subfacet_life /= 2; | |
0697b5c3 | 3507 | |
04d08d54 EJ |
3508 | backer->avg_n_subfacet += n_subfacets; |
3509 | backer->avg_n_subfacet /= 2; | |
3510 | ||
3511 | backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); | |
655ab909 | 3512 | |
04d08d54 EJ |
3513 | max_idle = subfacet_max_idle(backer); |
3514 | expire_subfacets(backer, max_idle); | |
acf60855 | 3515 | |
04d08d54 EJ |
3516 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
3517 | struct rule *rule, *next_rule; | |
3518 | ||
3519 | if (ofproto->backer != backer) { | |
3520 | continue; | |
3521 | } | |
acf60855 JP |
3522 | |
3523 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
3524 | * has passed. */ | |
abe7b10f | 3525 | ovs_mutex_lock(&ofproto_mutex); |
e503cc19 SH |
3526 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
3527 | &ofproto->up.expirable) { | |
3528 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 3529 | } |
abe7b10f | 3530 | ovs_mutex_unlock(&ofproto_mutex); |
abe529af | 3531 | |
acf60855 JP |
3532 | /* All outstanding data in existing flows has been accounted, so it's a |
3533 | * good time to do bond rebalancing. */ | |
3534 | if (ofproto->has_bonded_bundles) { | |
3535 | struct ofbundle *bundle; | |
abe529af | 3536 | |
acf60855 JP |
3537 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
3538 | if (bundle->bond) { | |
4a1b8f30 | 3539 | bond_rebalance(bundle->bond); |
acf60855 | 3540 | } |
abe529af BP |
3541 | } |
3542 | } | |
3543 | } | |
3544 | ||
acf60855 | 3545 | return MIN(max_idle, 1000); |
abe529af BP |
3546 | } |
3547 | ||
a218c879 BP |
3548 | /* Updates flow table statistics given that the datapath just reported 'stats' |
3549 | * as 'subfacet''s statistics. */ | |
3550 | static void | |
3551 | update_subfacet_stats(struct subfacet *subfacet, | |
3552 | const struct dpif_flow_stats *stats) | |
3553 | { | |
3554 | struct facet *facet = subfacet->facet; | |
9dfb1f78 EJ |
3555 | struct dpif_flow_stats diff; |
3556 | ||
3557 | diff.tcp_flags = stats->tcp_flags; | |
3558 | diff.used = stats->used; | |
a218c879 BP |
3559 | |
3560 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
9dfb1f78 | 3561 | diff.n_packets = stats->n_packets - subfacet->dp_packet_count; |
a218c879 BP |
3562 | } else { |
3563 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
9dfb1f78 | 3564 | diff.n_packets = 0; |
a218c879 BP |
3565 | } |
3566 | ||
3567 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
9dfb1f78 | 3568 | diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; |
a218c879 BP |
3569 | } else { |
3570 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
9dfb1f78 | 3571 | diff.n_bytes = 0; |
a218c879 BP |
3572 | } |
3573 | ||
b7e2f6b3 | 3574 | facet->ofproto->n_hit += diff.n_packets; |
a218c879 BP |
3575 | subfacet->dp_packet_count = stats->n_packets; |
3576 | subfacet->dp_byte_count = stats->n_bytes; | |
9dfb1f78 | 3577 | subfacet_update_stats(subfacet, &diff); |
a218c879 | 3578 | |
a218c879 BP |
3579 | if (facet->accounted_bytes < facet->byte_count) { |
3580 | facet_learn(facet); | |
3581 | facet_account(facet); | |
3582 | facet->accounted_bytes = facet->byte_count; | |
3583 | } | |
a218c879 BP |
3584 | } |
3585 | ||
3586 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
3587 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
3588 | static void | |
04d08d54 | 3589 | delete_unexpected_flow(struct dpif_backer *backer, |
a218c879 BP |
3590 | const struct nlattr *key, size_t key_len) |
3591 | { | |
3592 | if (!VLOG_DROP_WARN(&rl)) { | |
3593 | struct ds s; | |
3594 | ||
3595 | ds_init(&s); | |
3596 | odp_flow_key_format(key, key_len, &s); | |
04d08d54 | 3597 | VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); |
a218c879 BP |
3598 | ds_destroy(&s); |
3599 | } | |
3600 | ||
3601 | COVERAGE_INC(facet_unexpected); | |
04d08d54 | 3602 | dpif_flow_del(backer->dpif, key, key_len, NULL); |
a218c879 BP |
3603 | } |
3604 | ||
abe529af BP |
3605 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
3606 | * | |
3607 | * This function also pushes statistics updates to rules which each facet | |
3608 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3609 | * facet changes the rule it resubmits into at some time in between | |
3610 | * update_stats() runs, it is possible that statistics accrued to the | |
3611 | * old rule will be incorrectly attributed to the new rule. This could be | |
3612 | * avoided by calling update_stats() whenever rules are created or | |
3613 | * deleted. However, the performance impact of making so many calls to the | |
3614 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
3615 | * |
3616 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
3617 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
3618 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
3619 | * the hit or miss counts of br1. | |
abe529af BP |
3620 | */ |
3621 | static void | |
acf60855 | 3622 | update_stats(struct dpif_backer *backer) |
abe529af BP |
3623 | { |
3624 | const struct dpif_flow_stats *stats; | |
3625 | struct dpif_flow_dump dump; | |
61fb711d JP |
3626 | const struct nlattr *key, *mask; |
3627 | size_t key_len, mask_len; | |
abe529af | 3628 | |
acf60855 | 3629 | dpif_flow_dump_start(&dump, backer->dpif); |
e6cc0bab | 3630 | while (dpif_flow_dump_next(&dump, &key, &key_len, |
61fb711d | 3631 | &mask, &mask_len, NULL, NULL, &stats)) { |
b0f7b9b5 | 3632 | struct subfacet *subfacet; |
acf60855 | 3633 | uint32_t key_hash; |
abe529af | 3634 | |
acf60855 | 3635 | key_hash = odp_flow_key_hash(key, key_len); |
04d08d54 | 3636 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
6a7e895f BP |
3637 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
3638 | case SF_FAST_PATH: | |
a218c879 | 3639 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
3640 | break; |
3641 | ||
3642 | case SF_SLOW_PATH: | |
3643 | /* Stats are updated per-packet. */ | |
3644 | break; | |
3645 | ||
3646 | case SF_NOT_INSTALLED: | |
3647 | default: | |
04d08d54 | 3648 | delete_unexpected_flow(backer, key, key_len); |
6a7e895f | 3649 | break; |
abe529af | 3650 | } |
8fa4d1d0 | 3651 | run_fast_rl(); |
abe529af BP |
3652 | } |
3653 | dpif_flow_dump_done(&dump); | |
3654 | } | |
3655 | ||
3656 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3657 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3658 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3659 | * fold its statistic into its rule. */ | |
abe529af | 3660 | static int |
04d08d54 | 3661 | subfacet_max_idle(const struct dpif_backer *backer) |
abe529af BP |
3662 | { |
3663 | /* | |
3664 | * Idle time histogram. | |
3665 | * | |
b0f7b9b5 BP |
3666 | * Most of the time a switch has a relatively small number of subfacets. |
3667 | * When this is the case we might as well keep statistics for all of them | |
3668 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3669 | * well. |
3670 | * | |
b0f7b9b5 | 3671 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3672 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3673 | * significant. However, with a large number of subfacets it is likely |
3674 | * that only a few of them are "heavy hitters" that consume a large amount | |
3675 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3676 | * the kernel and maintaining in userspaces; other subfacets we can | |
3677 | * discard. | |
abe529af BP |
3678 | * |
3679 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3680 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3681 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3682 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3683 | * the most-recently-used 1% of subfacets (but at least |
380f49c4 | 3684 | * flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3685 | * the most-recently-used bucket of subfacets is kept, so actually an |
3686 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3687 | * (though the next run will delete most of those unless they receive |
3688 | * additional data). | |
abe529af | 3689 | * |
b0f7b9b5 BP |
3690 | * This requires a second pass through the subfacets, in addition to the |
3691 | * pass made by update_stats(), because the former function never looks at | |
3692 | * uninstallable subfacets. | |
abe529af | 3693 | */ |
31ef9f51 | 3694 | enum { BUCKET_WIDTH = 100 }; |
abe529af BP |
3695 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; |
3696 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3697 | int total, subtotal, bucket; |
b0f7b9b5 | 3698 | struct subfacet *subfacet; |
abe529af BP |
3699 | long long int now; |
3700 | int i; | |
3701 | ||
04d08d54 | 3702 | total = hmap_count(&backer->subfacets); |
380f49c4 | 3703 | if (total <= flow_eviction_threshold) { |
abe529af BP |
3704 | return N_BUCKETS * BUCKET_WIDTH; |
3705 | } | |
3706 | ||
3707 | /* Build histogram. */ | |
3708 | now = time_msec(); | |
04d08d54 | 3709 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { |
b0f7b9b5 | 3710 | long long int idle = now - subfacet->used; |
abe529af BP |
3711 | int bucket = (idle <= 0 ? 0 |
3712 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3713 | : (unsigned int) idle / BUCKET_WIDTH); | |
3714 | buckets[bucket]++; | |
3715 | } | |
3716 | ||
3717 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3718 | subtotal = bucket = 0; |
3719 | do { | |
3720 | subtotal += buckets[bucket++]; | |
084f5290 | 3721 | } while (bucket < N_BUCKETS && |
380f49c4 | 3722 | subtotal < MAX(flow_eviction_threshold, total / 100)); |
abe529af BP |
3723 | |
3724 | if (VLOG_IS_DBG_ENABLED()) { | |
3725 | struct ds s; | |
3726 | ||
3727 | ds_init(&s); | |
3728 | ds_put_cstr(&s, "keep"); | |
3729 | for (i = 0; i < N_BUCKETS; i++) { | |
3730 | if (i == bucket) { | |
3731 | ds_put_cstr(&s, ", drop"); | |
3732 | } | |
3733 | if (buckets[i]) { | |
3734 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3735 | } | |
3736 | } | |
04d08d54 | 3737 | VLOG_INFO("%s (msec:count)", ds_cstr(&s)); |
abe529af BP |
3738 | ds_destroy(&s); |
3739 | } | |
3740 | ||
3741 | return bucket * BUCKET_WIDTH; | |
3742 | } | |
3743 | ||
abe529af | 3744 | static void |
04d08d54 | 3745 | expire_subfacets(struct dpif_backer *backer, int dp_max_idle) |
abe529af | 3746 | { |
625b0720 BP |
3747 | /* Cutoff time for most flows. */ |
3748 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
3749 | ||
3750 | /* We really want to keep flows for special protocols around, so use a more | |
3751 | * conservative cutoff. */ | |
3752 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 3753 | |
b0f7b9b5 | 3754 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 3755 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 3756 | int n_batch; |
abe529af | 3757 | |
b99d3cee | 3758 | n_batch = 0; |
b0f7b9b5 | 3759 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
04d08d54 | 3760 | &backer->subfacets) { |
625b0720 BP |
3761 | long long int cutoff; |
3762 | ||
bbafd73b EJ |
3763 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
3764 | | SLOW_STP) | |
625b0720 BP |
3765 | ? special_cutoff |
3766 | : normal_cutoff); | |
b0f7b9b5 | 3767 | if (subfacet->used < cutoff) { |
6a7e895f | 3768 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 3769 | batch[n_batch++] = subfacet; |
1d85f9e5 | 3770 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
04d08d54 | 3771 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee BP |
3772 | n_batch = 0; |
3773 | } | |
3774 | } else { | |
3775 | subfacet_destroy(subfacet); | |
3776 | } | |
abe529af BP |
3777 | } |
3778 | } | |
b99d3cee BP |
3779 | |
3780 | if (n_batch > 0) { | |
04d08d54 | 3781 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee | 3782 | } |
abe529af BP |
3783 | } |
3784 | ||
3785 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3786 | * then delete it entirely. */ | |
3787 | static void | |
3788 | rule_expire(struct rule_dpif *rule) | |
15aaf599 | 3789 | OVS_REQUIRES(ofproto_mutex) |
abe529af | 3790 | { |
a3779dbc | 3791 | uint16_t idle_timeout, hard_timeout; |
15aaf599 BP |
3792 | long long int now = time_msec(); |
3793 | int reason; | |
abe529af | 3794 | |
35232d32 | 3795 | ovs_assert(!rule->up.pending); |
e2a3d183 | 3796 | |
15aaf599 | 3797 | /* Has 'rule' expired? */ |
d10976b7 | 3798 | ovs_mutex_lock(&rule->up.mutex); |
a3779dbc EJ |
3799 | hard_timeout = rule->up.hard_timeout; |
3800 | idle_timeout = rule->up.idle_timeout; | |
a3779dbc | 3801 | if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) { |
abe529af | 3802 | reason = OFPRR_HARD_TIMEOUT; |
a3779dbc | 3803 | } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) { |
abe529af BP |
3804 | reason = OFPRR_IDLE_TIMEOUT; |
3805 | } else { | |
15aaf599 | 3806 | reason = -1; |
abe529af | 3807 | } |
15aaf599 | 3808 | ovs_mutex_unlock(&rule->up.mutex); |
abe529af | 3809 | |
15aaf599 BP |
3810 | if (reason >= 0) { |
3811 | COVERAGE_INC(ofproto_dpif_expired); | |
3812 | ofproto_rule_expire(&rule->up, reason); | |
3813 | } | |
abe529af BP |
3814 | } |
3815 | \f | |
3816 | /* Facets. */ | |
3817 | ||
4dff9097 | 3818 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
3819 | * |
3820 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 3821 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 3822 | * |
bcd2633a JP |
3823 | * 'rule' and 'xout' must have been created based on 'miss'. |
3824 | * | |
3825 | * 'facet'' statistics are initialized based on 'stats'. | |
2b459b83 | 3826 | * |
b0f7b9b5 BP |
3827 | * The facet will initially have no subfacets. The caller should create (at |
3828 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3829 | static struct facet * |
e1ec7dd4 | 3830 | facet_create(const struct flow_miss *miss) |
abe529af | 3831 | { |
4dff9097 | 3832 | struct ofproto_dpif *ofproto = miss->ofproto; |
abe529af | 3833 | struct facet *facet; |
bcd2633a | 3834 | struct match match; |
abe529af | 3835 | |
264ac358 | 3836 | COVERAGE_INC(facet_create); |
abe529af | 3837 | facet = xzalloc(sizeof *facet); |
b7e2f6b3 | 3838 | facet->ofproto = miss->ofproto; |
e1ec7dd4 | 3839 | facet->used = miss->stats.used; |
4dff9097 | 3840 | facet->flow = miss->flow; |
4dff9097 EJ |
3841 | facet->learn_rl = time_msec() + 500; |
3842 | ||
b0f7b9b5 | 3843 | list_init(&facet->subfacets); |
abe529af BP |
3844 | netflow_flow_init(&facet->nf_flow); |
3845 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3846 | ||
e1ec7dd4 | 3847 | xlate_out_copy(&facet->xout, &miss->xout); |
bcd2633a JP |
3848 | |
3849 | match_init(&match, &facet->flow, &facet->xout.wc); | |
3850 | cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); | |
0b4f2078 | 3851 | ovs_rwlock_wrlock(&ofproto->facets.rwlock); |
bcd2633a | 3852 | classifier_insert(&ofproto->facets, &facet->cr); |
0b4f2078 | 3853 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a | 3854 | |
bbafd73b | 3855 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; |
abe529af BP |
3856 | return facet; |
3857 | } | |
3858 | ||
3859 | static void | |
3860 | facet_free(struct facet *facet) | |
3861 | { | |
4dff9097 | 3862 | if (facet) { |
bbafd73b | 3863 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
3864 | free(facet); |
3865 | } | |
abe529af BP |
3866 | } |
3867 | ||
c4af8a90 AW |
3868 | /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'. |
3869 | * 'flow' must reflect the data in 'packet'. */ | |
419460f4 AW |
3870 | int |
3871 | ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto, | |
3872 | const struct flow *flow, | |
3873 | struct rule_dpif *rule, | |
3874 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
3875 | struct ofpbuf *packet) | |
3d9e05f8 BP |
3876 | { |
3877 | struct odputil_keybuf keybuf; | |
39d007ab BP |
3878 | struct dpif_flow_stats stats; |
3879 | struct xlate_out xout; | |
3880 | struct xlate_in xin; | |
3881 | ofp_port_t in_port; | |
3d9e05f8 | 3882 | struct ofpbuf key; |
39d007ab BP |
3883 | int error; |
3884 | ||
3885 | ovs_assert((rule != NULL) != (ofpacts != NULL)); | |
3886 | ||
3887 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); | |
3888 | if (rule) { | |
3889 | rule_dpif_credit_stats(rule, &stats); | |
3890 | } | |
3891 | ||
3892 | xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); | |
3893 | xin.ofpacts = ofpacts; | |
3894 | xin.ofpacts_len = ofpacts_len; | |
3895 | xin.resubmit_stats = &stats; | |
3896 | xlate_actions(&xin, &xout); | |
3d9e05f8 | 3897 | |
6ff686f2 | 3898 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
39d007ab BP |
3899 | in_port = flow->in_port.ofp_port; |
3900 | if (in_port == OFPP_NONE) { | |
3901 | in_port = OFPP_LOCAL; | |
3902 | } | |
3903 | odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port)); | |
3904 | ||
3905 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, | |
7fd91025 BP |
3906 | xout.odp_actions.data, xout.odp_actions.size, packet, |
3907 | (xout.slow & SLOW_ACTION) != 0); | |
39d007ab | 3908 | xlate_out_uninit(&xout); |
80e5eed9 | 3909 | |
39d007ab | 3910 | return error; |
abe529af BP |
3911 | } |
3912 | ||
bcd2633a | 3913 | /* Remove 'facet' from its ofproto and free up the associated memory: |
abe529af BP |
3914 | * |
3915 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3916 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3917 | * |
3918 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3919 | */ | |
3920 | static void | |
15baa734 | 3921 | facet_remove(struct facet *facet) |
abe529af | 3922 | { |
b0f7b9b5 BP |
3923 | struct subfacet *subfacet, *next_subfacet; |
3924 | ||
264ac358 | 3925 | COVERAGE_INC(facet_remove); |
cb22974d | 3926 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
3927 | |
3928 | /* First uninstall all of the subfacets to get final statistics. */ | |
3929 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3930 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3931 | } |
3932 | ||
3933 | /* Flush the final stats to the rule. | |
3934 | * | |
3935 | * This might require us to have at least one subfacet around so that we | |
3936 | * can use its actions for accounting in facet_account(), which is why we | |
3937 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3938 | facet_flush_stats(facet); |
551a2f6c BP |
3939 | |
3940 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3941 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3942 | &facet->subfacets) { | |
15baa734 | 3943 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3944 | } |
0b4f2078 | 3945 | ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock); |
b7e2f6b3 | 3946 | classifier_remove(&facet->ofproto->facets, &facet->cr); |
0b4f2078 | 3947 | ovs_rwlock_unlock(&facet->ofproto->facets.rwlock); |
bcd2633a | 3948 | cls_rule_destroy(&facet->cr); |
abe529af BP |
3949 | facet_free(facet); |
3950 | } | |
3951 | ||
3de9590b BP |
3952 | /* Feed information from 'facet' back into the learning table to keep it in |
3953 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3954 | static void |
3de9590b | 3955 | facet_learn(struct facet *facet) |
abe529af | 3956 | { |
62bd7349 | 3957 | long long int now = time_msec(); |
abe529af | 3958 | |
bbafd73b | 3959 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
3960 | return; |
3961 | } | |
3962 | ||
62bd7349 | 3963 | facet->learn_rl = now + 500; |
6cf474d7 | 3964 | |
bbafd73b EJ |
3965 | if (!facet->xout.has_learn |
3966 | && !facet->xout.has_normal | |
3967 | && (!facet->xout.has_fin_timeout | |
3de9590b | 3968 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
3969 | return; |
3970 | } | |
abe529af | 3971 | |
9dfb1f78 | 3972 | facet_push_stats(facet, true); |
3de9590b BP |
3973 | } |
3974 | ||
3975 | static void | |
3976 | facet_account(struct facet *facet) | |
3977 | { | |
3de9590b BP |
3978 | const struct nlattr *a; |
3979 | unsigned int left; | |
3980 | ovs_be16 vlan_tci; | |
3981 | uint64_t n_bytes; | |
abe529af | 3982 | |
b7e2f6b3 | 3983 | if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { |
abe529af BP |
3984 | return; |
3985 | } | |
3de9590b | 3986 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3987 | |
3988 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3989 | * as a basis. We also need to track the actual VLAN on which the packet | |
3990 | * is going to be sent to ensure that it matches the one passed to | |
3991 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3992 | * hash bucket.) |
3993 | * | |
3994 | * We use the actions from an arbitrary subfacet because they should all | |
3995 | * be equally valid for our purpose. */ | |
d78be13b | 3996 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
3997 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
3998 | facet->xout.odp_actions.size) { | |
fea393b1 | 3999 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4000 | struct ofport_dpif *port; |
abe529af | 4001 | |
d78be13b | 4002 | switch (nl_attr_type(a)) { |
df2c07f4 | 4003 | case OVS_ACTION_ATTR_OUTPUT: |
b7e2f6b3 | 4004 | port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); |
abe529af | 4005 | if (port && port->bundle && port->bundle->bond) { |
d78be13b | 4006 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4007 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4008 | } |
d78be13b BP |
4009 | break; |
4010 | ||
fea393b1 BP |
4011 | case OVS_ACTION_ATTR_POP_VLAN: |
4012 | vlan_tci = htons(0); | |
d78be13b BP |
4013 | break; |
4014 | ||
fea393b1 BP |
4015 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4016 | vlan = nl_attr_get(a); | |
4017 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4018 | break; |
abe529af BP |
4019 | } |
4020 | } | |
4021 | } | |
4022 | ||
abe529af BP |
4023 | /* Returns true if the only action for 'facet' is to send to the controller. |
4024 | * (We don't report NetFlow expiration messages for such facets because they | |
4025 | * are just part of the control logic for the network, not real traffic). */ | |
4026 | static bool | |
4027 | facet_is_controller_flow(struct facet *facet) | |
4028 | { | |
f25d0cf3 | 4029 | if (facet) { |
b7e2f6b3 | 4030 | struct ofproto_dpif *ofproto = facet->ofproto; |
ad3efdcb | 4031 | const struct ofpact *ofpacts; |
15aaf599 | 4032 | struct rule_actions *actions; |
ad3efdcb EJ |
4033 | struct rule_dpif *rule; |
4034 | size_t ofpacts_len; | |
4035 | bool is_controller; | |
4036 | ||
4037 | rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule); | |
15aaf599 BP |
4038 | actions = rule_dpif_get_actions(rule); |
4039 | rule_dpif_unref(rule); | |
4040 | ||
4041 | ofpacts_len = actions->ofpacts_len; | |
4042 | ofpacts = actions->ofpacts; | |
ad3efdcb EJ |
4043 | is_controller = ofpacts_len > 0 |
4044 | && ofpacts->type == OFPACT_CONTROLLER | |
4045 | && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len); | |
15aaf599 | 4046 | rule_actions_unref(actions); |
6f00e29b | 4047 | |
ad3efdcb | 4048 | return is_controller; |
f25d0cf3 BP |
4049 | } |
4050 | return false; | |
abe529af BP |
4051 | } |
4052 | ||
4053 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4054 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4055 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4056 | * its packet and byte counts before this function is called. */ | |
4057 | static void | |
15baa734 | 4058 | facet_flush_stats(struct facet *facet) |
abe529af | 4059 | { |
b7e2f6b3 | 4060 | struct ofproto_dpif *ofproto = facet->ofproto; |
b0f7b9b5 BP |
4061 | struct subfacet *subfacet; |
4062 | ||
4063 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4064 | ovs_assert(!subfacet->dp_byte_count); |
4065 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4066 | } |
abe529af | 4067 | |
9dfb1f78 | 4068 | facet_push_stats(facet, false); |
3de9590b BP |
4069 | if (facet->accounted_bytes < facet->byte_count) { |
4070 | facet_account(facet); | |
4071 | facet->accounted_bytes = facet->byte_count; | |
4072 | } | |
abe529af BP |
4073 | |
4074 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4075 | struct ofexpired expired; | |
4076 | expired.flow = facet->flow; | |
4077 | expired.packet_count = facet->packet_count; | |
4078 | expired.byte_count = facet->byte_count; | |
4079 | expired.used = facet->used; | |
4080 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4081 | } | |
4082 | ||
abe529af BP |
4083 | /* Reset counters to prevent double counting if 'facet' ever gets |
4084 | * reinstalled. */ | |
bbb5d219 | 4085 | facet_reset_counters(facet); |
abe529af BP |
4086 | |
4087 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4088 | facet->tcp_flags = 0; |
abe529af BP |
4089 | } |
4090 | ||
bcd2633a JP |
4091 | /* Searches 'ofproto''s table of facets for one which would be responsible for |
4092 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4093 | * |
abe529af BP |
4094 | * The returned facet might need revalidation; use facet_lookup_valid() |
4095 | * instead if that is important. */ | |
4096 | static struct facet * | |
bcd2633a | 4097 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4098 | { |
0b4f2078 EJ |
4099 | struct cls_rule *cr; |
4100 | ||
4101 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); | |
4102 | cr = classifier_lookup(&ofproto->facets, flow, NULL); | |
4103 | ovs_rwlock_unlock(&ofproto->facets.rwlock); | |
bcd2633a | 4104 | return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL; |
abe529af BP |
4105 | } |
4106 | ||
bcd2633a JP |
4107 | /* Searches 'ofproto''s table of facets for one capable that covers |
4108 | * 'flow'. Returns it if found, otherwise a null pointer. | |
2b459b83 | 4109 | * |
abe529af BP |
4110 | * The returned facet is guaranteed to be valid. */ |
4111 | static struct facet * | |
bcd2633a | 4112 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) |
abe529af | 4113 | { |
c57b2226 | 4114 | struct facet *facet; |
abe529af | 4115 | |
bcd2633a | 4116 | facet = facet_find(ofproto, flow); |
abe529af | 4117 | if (facet |
a1616063 | 4118 | && ofproto->backer->need_revalidate |
5bf64ade | 4119 | && !facet_revalidate(facet)) { |
0305ce1f | 4120 | return NULL; |
abe529af BP |
4121 | } |
4122 | ||
4123 | return facet; | |
4124 | } | |
4125 | ||
6814e51f BP |
4126 | static bool |
4127 | facet_check_consistency(struct facet *facet) | |
4128 | { | |
4129 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4130 | ||
bbafd73b EJ |
4131 | struct xlate_out xout; |
4132 | struct xlate_in xin; | |
e1ec7dd4 | 4133 | bool ok; |
6814e51f BP |
4134 | |
4135 | /* Check the datapath actions for consistency. */ | |
10c44245 | 4136 | xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL); |
bbafd73b | 4137 | xlate_actions(&xin, &xout); |
6814e51f | 4138 | |
bbafd73b | 4139 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
e1ec7dd4 | 4140 | && facet->xout.slow == xout.slow; |
4dff9097 EJ |
4141 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4142 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4143 | |
4dff9097 EJ |
4144 | flow_format(&s, &facet->flow); |
4145 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4146 | |
bbafd73b | 4147 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4148 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4149 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4150 | facet->xout.odp_actions.size); | |
9616614b | 4151 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4152 | format_odp_actions(&s, xout.odp_actions.data, |
4153 | xout.odp_actions.size); | |
c0a71f4e | 4154 | ds_put_char(&s, ')'); |
6814e51f | 4155 | } |
4dff9097 | 4156 | |
bbafd73b EJ |
4157 | if (facet->xout.slow != xout.slow) { |
4158 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4159 | } |
4160 | ||
9616614b | 4161 | ds_destroy(&s); |
6814e51f | 4162 | } |
bbafd73b | 4163 | xlate_out_uninit(&xout); |
6814e51f BP |
4164 | |
4165 | return ok; | |
4166 | } | |
4167 | ||
15baa734 | 4168 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4169 | * |
4170 | * - If the rule found is different from 'facet''s current rule, moves | |
4171 | * 'facet' to the new rule and recompiles its actions. | |
4172 | * | |
4173 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4174 | * where it is and recompiles its actions anyway. |
4175 | * | |
4176 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
8449c4d6 | 4177 | * xlate_receive(), 'facet' is removed. |
5bf64ade EJ |
4178 | * |
4179 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
4180 | static bool | |
15baa734 | 4181 | facet_revalidate(struct facet *facet) |
abe529af | 4182 | { |
b7e2f6b3 | 4183 | struct ofproto_dpif *ofproto = facet->ofproto; |
abe529af | 4184 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4185 | struct subfacet *subfacet; |
bcd2633a | 4186 | struct flow_wildcards wc; |
bbafd73b EJ |
4187 | struct xlate_out xout; |
4188 | struct xlate_in xin; | |
abe529af BP |
4189 | |
4190 | COVERAGE_INC(facet_revalidate); | |
4191 | ||
f231418e EJ |
4192 | /* Check that child subfacets still correspond to this facet. Tunnel |
4193 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4194 | * change. */ | |
4195 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4196 | struct ofproto_dpif *recv_ofproto; | |
4197 | struct flow recv_flow; | |
4198 | int error; | |
4199 | ||
8449c4d6 EJ |
4200 | error = xlate_receive(ofproto->backer, NULL, subfacet->key, |
4201 | subfacet->key_len, &recv_flow, NULL, | |
4202 | &recv_ofproto, NULL); | |
f231418e EJ |
4203 | if (error |
4204 | || recv_ofproto != ofproto | |
bcd2633a | 4205 | || facet != facet_find(ofproto, &recv_flow)) { |
f231418e | 4206 | facet_remove(facet); |
5bf64ade | 4207 | return false; |
f231418e EJ |
4208 | } |
4209 | } | |
4210 | ||
bcd2633a | 4211 | flow_wildcards_init_catchall(&wc); |
ad3efdcb | 4212 | rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule); |
abe529af | 4213 | |
df2c07f4 | 4214 | /* Calculate new datapath actions. |
abe529af BP |
4215 | * |
4216 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4217 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4218 | * around to properly compose it. */ | |
47b0deae | 4219 | xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); |
bbafd73b | 4220 | xlate_actions(&xin, &xout); |
bcd2633a | 4221 | flow_wildcards_or(&xout.wc, &xout.wc, &wc); |
4dff9097 EJ |
4222 | |
4223 | /* A facet's slow path reason should only change under dramatic | |
4224 | * circumstances. Rather than try to update everything, it's simpler to | |
bcd2633a JP |
4225 | * remove the facet and start over. |
4226 | * | |
4227 | * More importantly, if a facet's wildcards change, it will be relatively | |
4228 | * difficult to figure out if its subfacets still belong to it, and if not | |
4229 | * which facet they may belong to. Again, to avoid the complexity, we | |
4230 | * simply give up instead. */ | |
4231 | if (facet->xout.slow != xout.slow | |
4232 | || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { | |
4dff9097 | 4233 | facet_remove(facet); |
bbafd73b | 4234 | xlate_out_uninit(&xout); |
a2143702 | 4235 | rule_dpif_unref(new_rule); |
4dff9097 EJ |
4236 | return false; |
4237 | } | |
6a7e895f | 4238 | |
bbafd73b | 4239 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
4240 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
4241 | if (subfacet->path == SF_FAST_PATH) { | |
4242 | struct dpif_flow_stats stats; | |
b95fc6ba | 4243 | |
bbafd73b | 4244 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 4245 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 4246 | } |
abe529af | 4247 | } |
b95fc6ba | 4248 | |
15baa734 | 4249 | facet_flush_stats(facet); |
4dff9097 | 4250 | |
bbafd73b EJ |
4251 | ofpbuf_clear(&facet->xout.odp_actions); |
4252 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
4253 | xout.odp_actions.size); | |
abe529af BP |
4254 | } |
4255 | ||
4256 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
4257 | facet->xout.slow = xout.slow; |
4258 | facet->xout.has_learn = xout.has_learn; | |
4259 | facet->xout.has_normal = xout.has_normal; | |
4260 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
4261 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
4262 | facet->xout.mirrors = xout.mirrors; | |
4263 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
15aaf599 BP |
4264 | |
4265 | ovs_mutex_lock(&new_rule->up.mutex); | |
b7e2f6b3 | 4266 | facet->used = MAX(facet->used, new_rule->up.created); |
15aaf599 | 4267 | ovs_mutex_unlock(&new_rule->up.mutex); |
5bf64ade | 4268 | |
bbafd73b | 4269 | xlate_out_uninit(&xout); |
a2143702 | 4270 | rule_dpif_unref(new_rule); |
5bf64ade | 4271 | return true; |
abe529af BP |
4272 | } |
4273 | ||
bbb5d219 EJ |
4274 | static void |
4275 | facet_reset_counters(struct facet *facet) | |
4276 | { | |
4277 | facet->packet_count = 0; | |
4278 | facet->byte_count = 0; | |
9d24de3b JP |
4279 | facet->prev_packet_count = 0; |
4280 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4281 | facet->accounted_bytes = 0; |
4282 | } | |
4283 | ||
e1ec7dd4 EJ |
4284 | static void |
4285 | flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow, | |
4286 | struct dpif_flow_stats *stats, bool may_learn) | |
4287 | { | |
4288 | struct ofport_dpif *in_port; | |
e1ec7dd4 EJ |
4289 | struct xlate_in xin; |
4290 | ||
4291 | in_port = get_ofp_port(ofproto, flow->in_port.ofp_port); | |
4292 | if (in_port && in_port->is_tunnel) { | |
4293 | netdev_vport_inc_rx(in_port->up.netdev, stats); | |
a1aeea86 AW |
4294 | if (in_port->bfd) { |
4295 | bfd_account_rx(in_port->bfd, stats); | |
4296 | } | |
e1ec7dd4 EJ |
4297 | } |
4298 | ||
10c44245 | 4299 | xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL); |
e1ec7dd4 EJ |
4300 | xin.resubmit_stats = stats; |
4301 | xin.may_learn = may_learn; | |
4302 | xlate_actions_for_side_effects(&xin); | |
e1ec7dd4 EJ |
4303 | } |
4304 | ||
abe529af | 4305 | static void |
9dfb1f78 | 4306 | facet_push_stats(struct facet *facet, bool may_learn) |
abe529af | 4307 | { |
112bc5f4 | 4308 | struct dpif_flow_stats stats; |
abe529af | 4309 | |
cb22974d BP |
4310 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4311 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4312 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4313 | |
112bc5f4 BP |
4314 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4315 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4316 | stats.used = facet->used; | |
9dfb1f78 EJ |
4317 | stats.tcp_flags = facet->tcp_flags; |
4318 | ||
4319 | if (may_learn || stats.n_packets || facet->used > facet->prev_used) { | |
9d24de3b JP |
4320 | facet->prev_packet_count = facet->packet_count; |
4321 | facet->prev_byte_count = facet->byte_count; | |
4322 | facet->prev_used = facet->used; | |
abe529af | 4323 | |
e1ec7dd4 | 4324 | netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow, |
9dfb1f78 EJ |
4325 | facet->used); |
4326 | netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); | |
e1ec7dd4 | 4327 | mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors, |
ec7ceaed | 4328 | stats.n_packets, stats.n_bytes); |
e1ec7dd4 | 4329 | flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn); |
abe529af BP |
4330 | } |
4331 | } | |
4332 | ||
8844e035 | 4333 | static void |
4464589e | 4334 | push_all_stats__(bool run_fast) |
8844e035 EJ |
4335 | { |
4336 | static long long int rl = LLONG_MIN; | |
4337 | struct ofproto_dpif *ofproto; | |
4338 | ||
4339 | if (time_msec() < rl) { | |
4340 | return; | |
4341 | } | |
4342 | ||
4343 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
bcd2633a | 4344 | struct cls_cursor cursor; |
8844e035 EJ |
4345 | struct facet *facet; |
4346 | ||
0b4f2078 | 4347 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
4348 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
4349 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
9dfb1f78 | 4350 | facet_push_stats(facet, false); |
4464589e EJ |
4351 | if (run_fast) { |
4352 | run_fast_rl(); | |
4353 | } | |
8844e035 | 4354 | } |
0b4f2078 | 4355 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
8844e035 EJ |
4356 | } |
4357 | ||
4358 | rl = time_msec() + 100; | |
4359 | } | |
4360 | ||
4464589e EJ |
4361 | static void |
4362 | push_all_stats(void) | |
4363 | { | |
4364 | push_all_stats__(true); | |
4365 | } | |
4366 | ||
9583bc14 | 4367 | void |
70742c7f EJ |
4368 | rule_dpif_credit_stats(struct rule_dpif *rule, |
4369 | const struct dpif_flow_stats *stats) | |
abe529af | 4370 | { |
9fbe253e | 4371 | ovs_mutex_lock(&rule->stats_mutex); |
112bc5f4 BP |
4372 | rule->packet_count += stats->n_packets; |
4373 | rule->byte_count += stats->n_bytes; | |
448c2fa8 | 4374 | rule->up.used = MAX(rule->up.used, stats->used); |
9fbe253e | 4375 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af | 4376 | } |
70742c7f EJ |
4377 | |
4378 | bool | |
ee1afdd5 | 4379 | rule_dpif_is_fail_open(const struct rule_dpif *rule) |
70742c7f | 4380 | { |
ee1afdd5 BP |
4381 | return is_fail_open_rule(&rule->up); |
4382 | } | |
cfa955b0 YT |
4383 | |
4384 | bool | |
4385 | rule_dpif_is_table_miss(const struct rule_dpif *rule) | |
4386 | { | |
4387 | return rule_is_table_miss(&rule->up); | |
70742c7f EJ |
4388 | } |
4389 | ||
4390 | ovs_be64 | |
4391 | rule_dpif_get_flow_cookie(const struct rule_dpif *rule) | |
15aaf599 | 4392 | OVS_REQUIRES(rule->up.mutex) |
70742c7f EJ |
4393 | { |
4394 | return rule->up.flow_cookie; | |
4395 | } | |
4396 | ||
4397 | void | |
4398 | rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout, | |
4399 | uint16_t hard_timeout) | |
4400 | { | |
4401 | ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout); | |
4402 | } | |
4403 | ||
6f00e29b BP |
4404 | /* Returns 'rule''s actions. The caller owns a reference on the returned |
4405 | * actions and must eventually release it (with rule_actions_unref()) to avoid | |
4406 | * a memory leak. */ | |
4407 | struct rule_actions * | |
4408 | rule_dpif_get_actions(const struct rule_dpif *rule) | |
70742c7f | 4409 | { |
15aaf599 | 4410 | return rule_get_actions(&rule->up); |
70742c7f | 4411 | } |
abe529af | 4412 | \f |
b0f7b9b5 BP |
4413 | /* Subfacets. */ |
4414 | ||
4415 | static struct subfacet * | |
04d08d54 EJ |
4416 | subfacet_find(struct dpif_backer *backer, const struct nlattr *key, |
4417 | size_t key_len, uint32_t key_hash) | |
b0f7b9b5 BP |
4418 | { |
4419 | struct subfacet *subfacet; | |
4420 | ||
4421 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
04d08d54 | 4422 | &backer->subfacets) { |
9566abf9 EJ |
4423 | if (subfacet->key_len == key_len |
4424 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
4425 | return subfacet; |
4426 | } | |
4427 | } | |
4428 | ||
4429 | return NULL; | |
4430 | } | |
4431 | ||
f9c37233 BP |
4432 | /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'. |
4433 | * 'key_hash' must be a hash over miss->key. The caller must have already | |
4434 | * ensured that no subfacet subfacet already exists. */ | |
b0f7b9b5 | 4435 | static struct subfacet * |
f9c37233 | 4436 | subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash) |
b0f7b9b5 | 4437 | { |
04d08d54 | 4438 | struct dpif_backer *backer = miss->ofproto->backer; |
a088a1ff JP |
4439 | const struct nlattr *key = miss->key; |
4440 | size_t key_len = miss->key_len; | |
b0f7b9b5 BP |
4441 | struct subfacet *subfacet; |
4442 | ||
f9c37233 BP |
4443 | subfacet = (list_is_empty(&facet->subfacets) |
4444 | ? &facet->one_subfacet | |
4445 | : xmalloc(sizeof *subfacet)); | |
b0f7b9b5 | 4446 | |
264ac358 | 4447 | COVERAGE_INC(subfacet_create); |
04d08d54 | 4448 | hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); |
b0f7b9b5 BP |
4449 | list_push_back(&facet->subfacets, &subfacet->list_node); |
4450 | subfacet->facet = facet; | |
9566abf9 EJ |
4451 | subfacet->key = xmemdup(key, key_len); |
4452 | subfacet->key_len = key_len; | |
e1ec7dd4 EJ |
4453 | subfacet->used = miss->stats.used; |
4454 | subfacet->created = subfacet->used; | |
26cd7e34 BP |
4455 | subfacet->dp_packet_count = 0; |
4456 | subfacet->dp_byte_count = 0; | |
6a7e895f | 4457 | subfacet->path = SF_NOT_INSTALLED; |
04d08d54 | 4458 | subfacet->backer = backer; |
b0f7b9b5 BP |
4459 | |
4460 | return subfacet; | |
4461 | } | |
4462 | ||
b0f7b9b5 BP |
4463 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
4464 | * its facet within 'ofproto', and frees it. */ | |
4465 | static void | |
15baa734 | 4466 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4467 | { |
15baa734 | 4468 | struct facet *facet = subfacet->facet; |
655ab909 | 4469 | |
264ac358 | 4470 | COVERAGE_INC(subfacet_destroy); |
15baa734 | 4471 | subfacet_uninstall(subfacet); |
04d08d54 | 4472 | hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); |
b0f7b9b5 BP |
4473 | list_remove(&subfacet->list_node); |
4474 | free(subfacet->key); | |
26cd7e34 BP |
4475 | if (subfacet != &facet->one_subfacet) { |
4476 | free(subfacet); | |
4477 | } | |
b0f7b9b5 BP |
4478 | } |
4479 | ||
4480 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4481 | * last remaining subfacet in its facet destroys the facet too. */ | |
4482 | static void | |
15baa734 | 4483 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4484 | { |
4485 | struct facet *facet = subfacet->facet; | |
4486 | ||
551a2f6c BP |
4487 | if (list_is_singleton(&facet->subfacets)) { |
4488 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4489 | facet_remove(facet); |
551a2f6c | 4490 | } else { |
15baa734 | 4491 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4492 | } |
4493 | } | |
4494 | ||
1d85f9e5 | 4495 | static void |
04d08d54 | 4496 | subfacet_destroy_batch(struct dpif_backer *backer, |
1d85f9e5 JP |
4497 | struct subfacet **subfacets, int n) |
4498 | { | |
1d85f9e5 JP |
4499 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
4500 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
4501 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
4502 | int i; | |
4503 | ||
4504 | for (i = 0; i < n; i++) { | |
4505 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
4506 | ops[i].u.flow_del.key = subfacets[i]->key; |
4507 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
4508 | ops[i].u.flow_del.stats = &stats[i]; |
4509 | opsp[i] = &ops[i]; | |
4510 | } | |
4511 | ||
04d08d54 | 4512 | dpif_operate(backer->dpif, opsp, n); |
1d85f9e5 JP |
4513 | for (i = 0; i < n; i++) { |
4514 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
4515 | subfacets[i]->path = SF_NOT_INSTALLED; | |
4516 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 4517 | run_fast_rl(); |
1d85f9e5 JP |
4518 | } |
4519 | } | |
4520 | ||
b0f7b9b5 BP |
4521 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4522 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4523 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4524 | * since 'subfacet' was last updated. | |
4525 | * | |
4526 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4527 | static int | |
4dff9097 EJ |
4528 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
4529 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 4530 | { |
15baa734 | 4531 | struct facet *facet = subfacet->facet; |
bbafd73b | 4532 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
4533 | const struct nlattr *actions = odp_actions->data; |
4534 | size_t actions_len = odp_actions->size; | |
d445cc16 JP |
4535 | struct odputil_keybuf maskbuf; |
4536 | struct ofpbuf mask; | |
4dff9097 | 4537 | |
6a7e895f | 4538 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 4539 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
4540 | int ret; |
4541 | ||
0f74f625 AZ |
4542 | flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE |
4543 | : DPIF_FP_MODIFY; | |
b0f7b9b5 BP |
4544 | if (stats) { |
4545 | flags |= DPIF_FP_ZERO_STATS; | |
4546 | } | |
4547 | ||
6a7e895f | 4548 | if (path == SF_SLOW_PATH) { |
b7e2f6b3 | 4549 | compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
4550 | slow_path_stub, sizeof slow_path_stub, |
4551 | &actions, &actions_len); | |
4552 | } | |
4553 | ||
d445cc16 | 4554 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
6b35e630 JP |
4555 | if (enable_megaflows) { |
4556 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
4557 | &facet->flow, UINT32_MAX); | |
4558 | } | |
d445cc16 JP |
4559 | |
4560 | ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, | |
4561 | subfacet->key_len, mask.data, mask.size, | |
e6cc0bab | 4562 | actions, actions_len, stats); |
b0f7b9b5 BP |
4563 | |
4564 | if (stats) { | |
4565 | subfacet_reset_dp_stats(subfacet, stats); | |
4566 | } | |
4567 | ||
9b1a48c2 JP |
4568 | if (ret) { |
4569 | COVERAGE_INC(subfacet_install_fail); | |
4570 | } else { | |
6a7e895f BP |
4571 | subfacet->path = path; |
4572 | } | |
b0f7b9b5 BP |
4573 | return ret; |
4574 | } | |
4575 | ||
4576 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
4577 | static void | |
15baa734 | 4578 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 4579 | { |
6a7e895f | 4580 | if (subfacet->path != SF_NOT_INSTALLED) { |
b7e2f6b3 | 4581 | struct ofproto_dpif *ofproto = subfacet->facet->ofproto; |
b0f7b9b5 | 4582 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
4583 | int error; |
4584 | ||
9566abf9 EJ |
4585 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
4586 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
4587 | subfacet_reset_dp_stats(subfacet, &stats); |
4588 | if (!error) { | |
15baa734 | 4589 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 4590 | } |
6a7e895f | 4591 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 4592 | } else { |
cb22974d BP |
4593 | ovs_assert(subfacet->dp_packet_count == 0); |
4594 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
4595 | } |
4596 | } | |
4597 | ||
4598 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4599 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4600 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4601 | * was reset in the datapath. 'stats' will be modified to include only | |
4602 | * statistics new since 'subfacet' was last updated. */ | |
4603 | static void | |
4604 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4605 | struct dpif_flow_stats *stats) | |
4606 | { | |
4607 | if (stats | |
4608 | && subfacet->dp_packet_count <= stats->n_packets | |
4609 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4610 | stats->n_packets -= subfacet->dp_packet_count; | |
4611 | stats->n_bytes -= subfacet->dp_byte_count; | |
4612 | } | |
4613 | ||
4614 | subfacet->dp_packet_count = 0; | |
4615 | subfacet->dp_byte_count = 0; | |
4616 | } | |
4617 | ||
b0f7b9b5 BP |
4618 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. |
4619 | * | |
4620 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4621 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4622 | * represents a packet that was sent by hand or if it represents statistics | |
4623 | * that have been cleared out of the datapath. */ | |
4624 | static void | |
15baa734 | 4625 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4626 | const struct dpif_flow_stats *stats) |
4627 | { | |
4628 | if (stats->n_packets || stats->used > subfacet->used) { | |
4629 | struct facet *facet = subfacet->facet; | |
4630 | ||
9dfb1f78 EJ |
4631 | subfacet->used = MAX(subfacet->used, stats->used); |
4632 | facet->used = MAX(facet->used, stats->used); | |
b0f7b9b5 BP |
4633 | facet->packet_count += stats->n_packets; |
4634 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4635 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4636 | } |
4637 | } | |
4638 | \f | |
abe529af BP |
4639 | /* Rules. */ |
4640 | ||
bcd2633a JP |
4641 | /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets |
4642 | * the fields that were relevant as part of the lookup. */ | |
ad3efdcb | 4643 | void |
bcd2633a | 4644 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
ad3efdcb | 4645 | struct flow_wildcards *wc, struct rule_dpif **rule) |
c57b2226 | 4646 | { |
ec89fc6f | 4647 | struct ofport_dpif *port; |
c57b2226 | 4648 | |
ad3efdcb EJ |
4649 | if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) { |
4650 | return; | |
c57b2226 | 4651 | } |
ec89fc6f EJ |
4652 | port = get_ofp_port(ofproto, flow->in_port.ofp_port); |
4653 | if (!port) { | |
4654 | VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16, | |
4655 | flow->in_port.ofp_port); | |
4656 | } | |
c57b2226 | 4657 | |
70742c7f EJ |
4658 | choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule, |
4659 | ofproto->no_packet_in_rule, rule); | |
c57b2226 BP |
4660 | } |
4661 | ||
ad3efdcb | 4662 | bool |
9583bc14 EJ |
4663 | rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, |
4664 | const struct flow *flow, struct flow_wildcards *wc, | |
ad3efdcb | 4665 | uint8_t table_id, struct rule_dpif **rule) |
abe529af | 4666 | { |
49a0e0eb | 4667 | const struct cls_rule *cls_rule; |
7257b535 | 4668 | struct classifier *cls; |
7fd51d39 | 4669 | bool frag; |
7257b535 | 4670 | |
ad3efdcb | 4671 | *rule = NULL; |
9cdaaebe | 4672 | if (table_id >= N_TABLES) { |
ad3efdcb | 4673 | return false; |
9cdaaebe BP |
4674 | } |
4675 | ||
1dd35f8a | 4676 | if (wc) { |
7431e171 | 4677 | memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type); |
1dd35f8a JP |
4678 | wc->masks.nw_frag |= FLOW_NW_FRAG_MASK; |
4679 | } | |
4680 | ||
d0918789 | 4681 | cls = &ofproto->up.tables[table_id].cls; |
ad3efdcb | 4682 | ovs_rwlock_rdlock(&cls->rwlock); |
7fd51d39 BP |
4683 | frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; |
4684 | if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
4685 | /* We must pretend that transport ports are unavailable. */ | |
7257b535 BP |
4686 | struct flow ofpc_normal_flow = *flow; |
4687 | ofpc_normal_flow.tp_src = htons(0); | |
4688 | ofpc_normal_flow.tp_dst = htons(0); | |
bcd2633a | 4689 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); |
7fd51d39 BP |
4690 | } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { |
4691 | cls_rule = &ofproto->drop_frags_rule->up.cr; | |
94639963 | 4692 | /* Frag mask in wc already set above. */ |
7257b535 | 4693 | } else { |
bcd2633a | 4694 | cls_rule = classifier_lookup(cls, flow, wc); |
7257b535 | 4695 | } |
ad3efdcb EJ |
4696 | |
4697 | *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
a2143702 | 4698 | rule_dpif_ref(*rule); |
ad3efdcb EJ |
4699 | ovs_rwlock_unlock(&cls->rwlock); |
4700 | ||
4701 | return *rule != NULL; | |
abe529af BP |
4702 | } |
4703 | ||
ec89fc6f EJ |
4704 | /* Given a port configuration (specified as zero if there's no port), chooses |
4705 | * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a | |
4706 | * flow table miss. */ | |
70742c7f | 4707 | void |
ec89fc6f | 4708 | choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule, |
70742c7f | 4709 | struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule) |
c376f9a3 | 4710 | { |
70742c7f | 4711 | *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; |
a2143702 BP |
4712 | rule_dpif_ref(*rule); |
4713 | } | |
4714 | ||
4715 | void | |
4716 | rule_dpif_ref(struct rule_dpif *rule) | |
4717 | { | |
4718 | if (rule) { | |
4719 | ofproto_rule_ref(&rule->up); | |
4720 | } | |
c376f9a3 IY |
4721 | } |
4722 | ||
ad3efdcb | 4723 | void |
a2143702 | 4724 | rule_dpif_unref(struct rule_dpif *rule) |
ad3efdcb EJ |
4725 | { |
4726 | if (rule) { | |
a2143702 | 4727 | ofproto_rule_unref(&rule->up); |
ad3efdcb EJ |
4728 | } |
4729 | } | |
4730 | ||
7ee20df1 BP |
4731 | static void |
4732 | complete_operation(struct rule_dpif *rule) | |
15aaf599 | 4733 | OVS_REQUIRES(ofproto_mutex) |
7ee20df1 BP |
4734 | { |
4735 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4736 | ||
a1616063 | 4737 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
35232d32 | 4738 | ofoperation_complete(rule->up.pending, 0); |
7ee20df1 BP |
4739 | } |
4740 | ||
70742c7f EJ |
4741 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) |
4742 | { | |
4743 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
4744 | } | |
4745 | ||
abe529af BP |
4746 | static struct rule * |
4747 | rule_alloc(void) | |
4748 | { | |
4749 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4750 | return &rule->up; | |
4751 | } | |
4752 | ||
4753 | static void | |
4754 | rule_dealloc(struct rule *rule_) | |
4755 | { | |
4756 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4757 | free(rule); | |
4758 | } | |
4759 | ||
90bf1e07 | 4760 | static enum ofperr |
abe529af BP |
4761 | rule_construct(struct rule *rule_) |
4762 | { | |
4763 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
834d6caf | 4764 | ovs_mutex_init(&rule->stats_mutex); |
9fbe253e | 4765 | ovs_mutex_lock(&rule->stats_mutex); |
abe529af BP |
4766 | rule->packet_count = 0; |
4767 | rule->byte_count = 0; | |
9fbe253e | 4768 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af BP |
4769 | return 0; |
4770 | } | |
4771 | ||
4772 | static void | |
8037acb4 | 4773 | rule_insert(struct rule *rule_) |
15aaf599 | 4774 | OVS_REQUIRES(ofproto_mutex) |
abe529af | 4775 | { |
9fbe253e EJ |
4776 | struct rule_dpif *rule = rule_dpif_cast(rule_); |
4777 | complete_operation(rule); | |
8037acb4 BP |
4778 | } |
4779 | ||
4780 | static void | |
4781 | rule_delete(struct rule *rule_) | |
15aaf599 | 4782 | OVS_REQUIRES(ofproto_mutex) |
8037acb4 BP |
4783 | { |
4784 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4785 | complete_operation(rule); | |
4786 | } | |
4787 | ||
4788 | static void | |
4789 | rule_destruct(struct rule *rule_) | |
4790 | { | |
4791 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
9fbe253e | 4792 | ovs_mutex_destroy(&rule->stats_mutex); |
abe529af BP |
4793 | } |
4794 | ||
4795 | static void | |
4796 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4797 | { | |
4798 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 4799 | |
4464589e EJ |
4800 | /* push_all_stats() can handle flow misses which, when using the learn |
4801 | * action, can cause rules to be added and deleted. This can corrupt our | |
4802 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
4803 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
4804 | push_all_stats__(false); | |
bf1e8ff9 | 4805 | |
abe529af BP |
4806 | /* Start from historical data for 'rule' itself that are no longer tracked |
4807 | * in facets. This counts, for example, facets that have expired. */ | |
9fbe253e | 4808 | ovs_mutex_lock(&rule->stats_mutex); |
abe529af BP |
4809 | *packets = rule->packet_count; |
4810 | *bytes = rule->byte_count; | |
9fbe253e | 4811 | ovs_mutex_unlock(&rule->stats_mutex); |
abe529af BP |
4812 | } |
4813 | ||
0a740f48 EJ |
4814 | static void |
4815 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
4816 | struct ofpbuf *packet) | |
abe529af | 4817 | { |
419460f4 AW |
4818 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
4819 | ||
4820 | ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet); | |
0a740f48 | 4821 | } |
5bf0e941 | 4822 | |
0a740f48 EJ |
4823 | static enum ofperr |
4824 | rule_execute(struct rule *rule, const struct flow *flow, | |
4825 | struct ofpbuf *packet) | |
4826 | { | |
4827 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
4828 | ofpbuf_delete(packet); | |
5bf0e941 | 4829 | return 0; |
abe529af BP |
4830 | } |
4831 | ||
7ee20df1 | 4832 | static void |
b277c7cc | 4833 | rule_modify_actions(struct rule *rule_, bool reset_counters) |
15aaf599 | 4834 | OVS_REQUIRES(ofproto_mutex) |
abe529af BP |
4835 | { |
4836 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 | 4837 | |
b277c7cc BP |
4838 | if (reset_counters) { |
4839 | ovs_mutex_lock(&rule->stats_mutex); | |
4840 | rule->packet_count = 0; | |
4841 | rule->byte_count = 0; | |
4842 | ovs_mutex_unlock(&rule->stats_mutex); | |
4843 | } | |
4844 | ||
7ee20df1 | 4845 | complete_operation(rule); |
abe529af | 4846 | } |
00430a3f SH |
4847 | |
4848 | static struct group_dpif *group_dpif_cast(const struct ofgroup *group) | |
4849 | { | |
4850 | return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL; | |
4851 | } | |
4852 | ||
4853 | static struct ofgroup * | |
4854 | group_alloc(void) | |
4855 | { | |
4856 | struct group_dpif *group = xzalloc(sizeof *group); | |
4857 | return &group->up; | |
4858 | } | |
4859 | ||
4860 | static void | |
4861 | group_dealloc(struct ofgroup *group_) | |
4862 | { | |
4863 | struct group_dpif *group = group_dpif_cast(group_); | |
4864 | free(group); | |
4865 | } | |
4866 | ||
4867 | static void | |
4868 | group_construct_stats(struct group_dpif *group) | |
4869 | OVS_REQUIRES(group->stats_mutex) | |
4870 | { | |
4871 | group->packet_count = 0; | |
4872 | group->byte_count = 0; | |
4873 | if (!group->bucket_stats) { | |
4874 | group->bucket_stats = xcalloc(group->up.n_buckets, | |
4875 | sizeof *group->bucket_stats); | |
4876 | } else { | |
4877 | memset(group->bucket_stats, 0, group->up.n_buckets * | |
4878 | sizeof *group->bucket_stats); | |
4879 | } | |
4880 | } | |
4881 | ||
4882 | static enum ofperr | |
4883 | group_construct(struct ofgroup *group_) | |
4884 | { | |
4885 | struct group_dpif *group = group_dpif_cast(group_); | |
4886 | ovs_mutex_init(&group->stats_mutex); | |
4887 | ovs_mutex_lock(&group->stats_mutex); | |
4888 | group_construct_stats(group); | |
4889 | ovs_mutex_unlock(&group->stats_mutex); | |
4890 | return 0; | |
4891 | } | |
4892 | ||
4893 | static void | |
4894 | group_destruct__(struct group_dpif *group) | |
4895 | OVS_REQUIRES(group->stats_mutex) | |
4896 | { | |
4897 | free(group->bucket_stats); | |
4898 | group->bucket_stats = NULL; | |
4899 | } | |
4900 | ||
4901 | static void | |
4902 | group_destruct(struct ofgroup *group_) | |
4903 | { | |
4904 | struct group_dpif *group = group_dpif_cast(group_); | |
4905 | ovs_mutex_lock(&group->stats_mutex); | |
4906 | group_destruct__(group); | |
4907 | ovs_mutex_unlock(&group->stats_mutex); | |
4908 | ovs_mutex_destroy(&group->stats_mutex); | |
4909 | } | |
4910 | ||
4911 | static enum ofperr | |
4912 | group_modify(struct ofgroup *group_, struct ofgroup *victim_) | |
4913 | { | |
4914 | struct group_dpif *group = group_dpif_cast(group_); | |
4915 | struct group_dpif *victim = group_dpif_cast(victim_); | |
4916 | ||
4917 | ovs_mutex_lock(&group->stats_mutex); | |
4918 | if (victim->up.n_buckets < group->up.n_buckets) { | |
4919 | group_destruct__(group); | |
4920 | } | |
4921 | group_construct_stats(group); | |
4922 | ovs_mutex_unlock(&group->stats_mutex); | |
4923 | ||
4924 | return 0; | |
4925 | } | |
4926 | ||
4927 | static enum ofperr | |
4928 | group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs) | |
4929 | { | |
4930 | struct group_dpif *group = group_dpif_cast(group_); | |
4931 | ||
4932 | /* Start from historical data for 'group' itself that are no longer tracked | |
4933 | * in facets. This counts, for example, facets that have expired. */ | |
4934 | ovs_mutex_lock(&group->stats_mutex); | |
4935 | ogs->packet_count = group->packet_count; | |
4936 | ogs->byte_count = group->byte_count; | |
4937 | memcpy(ogs->bucket_stats, group->bucket_stats, | |
4938 | group->up.n_buckets * sizeof *group->bucket_stats); | |
4939 | ovs_mutex_unlock(&group->stats_mutex); | |
4940 | ||
4941 | return 0; | |
4942 | } | |
f4fb341b SH |
4943 | |
4944 | bool | |
4945 | group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id, | |
4946 | struct group_dpif **group) | |
4947 | OVS_TRY_RDLOCK(true, (*group)->up.rwlock) | |
4948 | { | |
4949 | struct ofgroup *ofgroup; | |
4950 | bool found; | |
4951 | ||
4952 | *group = NULL; | |
4953 | found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup); | |
4954 | *group = found ? group_dpif_cast(ofgroup) : NULL; | |
4955 | ||
4956 | return found; | |
4957 | } | |
4958 | ||
4959 | void | |
4960 | group_dpif_release(struct group_dpif *group) | |
4961 | OVS_RELEASES(group->up.rwlock) | |
4962 | { | |
4963 | ofproto_group_release(&group->up); | |
4964 | } | |
4965 | ||
4966 | void | |
4967 | group_dpif_get_buckets(const struct group_dpif *group, | |
4968 | const struct list **buckets) | |
4969 | { | |
4970 | *buckets = &group->up.buckets; | |
4971 | } | |
4972 | ||
4973 | enum ofp11_group_type | |
4974 | group_dpif_get_type(const struct group_dpif *group) | |
4975 | { | |
4976 | return group->up.type; | |
4977 | } | |
abe529af | 4978 | \f |
97d6520b | 4979 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4980 | * May modify 'packet'. |
abe529af | 4981 | * Returns 0 if successful, otherwise a positive errno value. */ |
91d6cd12 AW |
4982 | int |
4983 | ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) | |
abe529af | 4984 | { |
79e15b78 | 4985 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
abe529af BP |
4986 | int error; |
4987 | ||
91d6cd12 | 4988 | error = xlate_send_packet(ofport, packet); |
79e15b78 | 4989 | |
0da3c61b | 4990 | ovs_mutex_lock(&ofproto->stats_mutex); |
79e15b78 EJ |
4991 | ofproto->stats.tx_packets++; |
4992 | ofproto->stats.tx_bytes += packet->size; | |
0da3c61b | 4993 | ovs_mutex_unlock(&ofproto->stats_mutex); |
abe529af BP |
4994 | return error; |
4995 | } | |
abe529af | 4996 | |
6a7e895f BP |
4997 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
4998 | * The action will state 'slow' as the reason that the action is in the slow | |
4999 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5000 | * dump-flows" output to see why a flow is in the slow path.) | |
5001 | * | |
5002 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5003 | * 'stub_size' must be large enough for the action. | |
5004 | * | |
5005 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5006 | * respectively. */ | |
5007 | static void | |
5008 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5009 | enum slow_path_reason slow, | |
5010 | uint64_t *stub, size_t stub_size, | |
5011 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5012 | { | |
5013 | union user_action_cookie cookie; | |
5014 | struct ofpbuf buf; | |
5015 | ||
5016 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5017 | cookie.slow_path.unused = 0; | |
5018 | cookie.slow_path.reason = slow; | |
5019 | ||
5020 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5021 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
4e022ec0 AW |
5022 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, |
5023 | ODPP_NONE); | |
29089a54 | 5024 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5025 | } else { |
89a8a7f0 EJ |
5026 | odp_port_t odp_port; |
5027 | uint32_t pid; | |
5028 | ||
5029 | odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port); | |
5030 | pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port); | |
5031 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); | |
625b0720 | 5032 | } |
6a7e895f BP |
5033 | *actionsp = buf.data; |
5034 | *actions_lenp = buf.size; | |
5035 | } | |
9583bc14 EJ |
5036 | \f |
5037 | static bool | |
5038 | set_frag_handling(struct ofproto *ofproto_, | |
5039 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5040 | { |
5041 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 5042 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 5043 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
5044 | return true; |
5045 | } else { | |
5046 | return false; | |
5047 | } | |
abe529af BP |
5048 | } |
5049 | ||
90bf1e07 | 5050 | static enum ofperr |
419460f4 | 5051 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
abe529af | 5052 | const struct flow *flow, |
f25d0cf3 | 5053 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af | 5054 | { |
419460f4 AW |
5055 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
5056 | ||
5057 | ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts, | |
5058 | ofpacts_len, packet); | |
548de4dd | 5059 | return 0; |
abe529af | 5060 | } |
6fca1ffb BP |
5061 | \f |
5062 | /* NetFlow. */ | |
5063 | ||
5064 | static int | |
5065 | set_netflow(struct ofproto *ofproto_, | |
5066 | const struct netflow_options *netflow_options) | |
5067 | { | |
5068 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5069 | ||
5070 | if (netflow_options) { | |
5071 | if (!ofproto->netflow) { | |
5072 | ofproto->netflow = netflow_create(); | |
7bab1576 | 5073 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6fca1ffb BP |
5074 | } |
5075 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7bab1576 EJ |
5076 | } else if (ofproto->netflow) { |
5077 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
6fca1ffb BP |
5078 | netflow_destroy(ofproto->netflow); |
5079 | ofproto->netflow = NULL; | |
6fca1ffb | 5080 | } |
7bab1576 EJ |
5081 | |
5082 | return 0; | |
6fca1ffb | 5083 | } |
abe529af BP |
5084 | |
5085 | static void | |
5086 | get_netflow_ids(const struct ofproto *ofproto_, | |
5087 | uint8_t *engine_type, uint8_t *engine_id) | |
5088 | { | |
5089 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5090 | ||
acf60855 | 5091 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 5092 | } |
6fca1ffb BP |
5093 | |
5094 | static void | |
5095 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
5096 | { | |
5097 | if (!facet_is_controller_flow(facet) && | |
5098 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 5099 | struct subfacet *subfacet; |
6fca1ffb BP |
5100 | struct ofexpired expired; |
5101 | ||
b0f7b9b5 | 5102 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 5103 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 5104 | struct dpif_flow_stats stats; |
6fca1ffb | 5105 | |
bcd2633a JP |
5106 | subfacet_install(subfacet, &facet->xout.odp_actions, |
5107 | &stats); | |
15baa734 | 5108 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5109 | } |
6fca1ffb BP |
5110 | } |
5111 | ||
5112 | expired.flow = facet->flow; | |
5113 | expired.packet_count = facet->packet_count; | |
5114 | expired.byte_count = facet->byte_count; | |
5115 | expired.used = facet->used; | |
5116 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
5117 | } | |
5118 | } | |
5119 | ||
5120 | static void | |
5121 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
5122 | { | |
bcd2633a | 5123 | struct cls_cursor cursor; |
6fca1ffb BP |
5124 | struct facet *facet; |
5125 | ||
0b4f2078 | 5126 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5127 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5128 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6fca1ffb BP |
5129 | send_active_timeout(ofproto, facet); |
5130 | } | |
0b4f2078 | 5131 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6fca1ffb | 5132 | } |
abe529af BP |
5133 | \f |
5134 | static struct ofproto_dpif * | |
5135 | ofproto_dpif_lookup(const char *name) | |
5136 | { | |
b44a10b7 BP |
5137 | struct ofproto_dpif *ofproto; |
5138 | ||
5139 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
5140 | hash_string(name, 0), &all_ofproto_dpifs) { | |
5141 | if (!strcmp(ofproto->up.name, name)) { | |
5142 | return ofproto; | |
5143 | } | |
5144 | } | |
5145 | return NULL; | |
abe529af BP |
5146 | } |
5147 | ||
f0a3aa2e | 5148 | static void |
96e466a3 | 5149 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 5150 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 5151 | { |
490df1ef | 5152 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 5153 | |
96e466a3 EJ |
5154 | if (argc > 1) { |
5155 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5156 | if (!ofproto) { | |
bde9f75d | 5157 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
5158 | return; |
5159 | } | |
509c0149 | 5160 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5161 | mac_learning_flush(ofproto->ml); |
509c0149 | 5162 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 EJ |
5163 | } else { |
5164 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
509c0149 | 5165 | ovs_rwlock_wrlock(&ofproto->ml->rwlock); |
30618594 | 5166 | mac_learning_flush(ofproto->ml); |
509c0149 | 5167 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
96e466a3 | 5168 | } |
f0a3aa2e | 5169 | } |
f0a3aa2e | 5170 | |
bde9f75d | 5171 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
5172 | } |
5173 | ||
46c88433 EJ |
5174 | static struct ofport_dpif * |
5175 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5176 | { | |
5177 | return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif, | |
5178 | bundle_node); | |
5179 | } | |
5180 | ||
abe529af | 5181 | static void |
0e15264f BP |
5182 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5183 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
5184 | { |
5185 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5186 | const struct ofproto_dpif *ofproto; | |
5187 | const struct mac_entry *e; | |
5188 | ||
0e15264f | 5189 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 5190 | if (!ofproto) { |
bde9f75d | 5191 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
5192 | return; |
5193 | } | |
5194 | ||
5195 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
509c0149 | 5196 | ovs_rwlock_rdlock(&ofproto->ml->rwlock); |
abe529af BP |
5197 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { |
5198 | struct ofbundle *bundle = e->port.p; | |
ad28062f BP |
5199 | char name[OFP_MAX_PORT_NAME_LEN]; |
5200 | ||
5201 | ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port, | |
5202 | name, sizeof name); | |
5203 | ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n", | |
5204 | name, e->vlan, ETH_ADDR_ARGS(e->mac), | |
e764773c | 5205 | mac_entry_age(ofproto->ml, e)); |
abe529af | 5206 | } |
509c0149 | 5207 | ovs_rwlock_unlock(&ofproto->ml->rwlock); |
bde9f75d | 5208 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
5209 | ds_destroy(&ds); |
5210 | } | |
5211 | ||
6a6455e5 | 5212 | struct trace_ctx { |
bbafd73b EJ |
5213 | struct xlate_out xout; |
5214 | struct xlate_in xin; | |
abe529af BP |
5215 | struct flow flow; |
5216 | struct ds *result; | |
5217 | }; | |
5218 | ||
5219 | static void | |
4d0acc70 | 5220 | trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) |
abe529af | 5221 | { |
15aaf599 BP |
5222 | struct rule_actions *actions; |
5223 | ovs_be64 cookie; | |
5224 | ||
abe529af BP |
5225 | ds_put_char_multiple(result, '\t', level); |
5226 | if (!rule) { | |
5227 | ds_put_cstr(result, "No match\n"); | |
5228 | return; | |
5229 | } | |
5230 | ||
15aaf599 BP |
5231 | ovs_mutex_lock(&rule->up.mutex); |
5232 | cookie = rule->up.flow_cookie; | |
5233 | ovs_mutex_unlock(&rule->up.mutex); | |
5234 | ||
29901626 | 5235 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
15aaf599 | 5236 | rule ? rule->up.table_id : 0, ntohll(cookie)); |
79feb7df | 5237 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5238 | ds_put_char(result, '\n'); |
5239 | ||
15aaf599 BP |
5240 | actions = rule_dpif_get_actions(rule); |
5241 | ||
abe529af | 5242 | ds_put_char_multiple(result, '\t', level); |
455ecd77 | 5243 | ds_put_cstr(result, "OpenFlow actions="); |
15aaf599 | 5244 | ofpacts_format(actions->ofpacts, actions->ofpacts_len, result); |
abe529af | 5245 | ds_put_char(result, '\n'); |
15aaf599 BP |
5246 | |
5247 | rule_actions_unref(actions); | |
abe529af BP |
5248 | } |
5249 | ||
5250 | static void | |
5251 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 5252 | struct trace_ctx *trace) |
abe529af BP |
5253 | { |
5254 | ds_put_char_multiple(result, '\t', level); | |
5255 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 5256 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
5257 | ds_put_cstr(result, "unchanged"); |
5258 | } else { | |
bbafd73b EJ |
5259 | flow_format(result, &trace->xin.flow); |
5260 | trace->flow = trace->xin.flow; | |
abe529af BP |
5261 | } |
5262 | ds_put_char(result, '\n'); | |
5263 | } | |
5264 | ||
eb9e1c26 EJ |
5265 | static void |
5266 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 5267 | struct trace_ctx *trace) |
eb9e1c26 EJ |
5268 | { |
5269 | size_t i; | |
5270 | ||
5271 | ds_put_char_multiple(result, '\t', level); | |
5272 | ds_put_format(result, "%s:", title); | |
5273 | for (i = 0; i < FLOW_N_REGS; i++) { | |
34582733 | 5274 | ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]); |
eb9e1c26 EJ |
5275 | } |
5276 | ds_put_char(result, '\n'); | |
5277 | } | |
5278 | ||
1ed8d352 EJ |
5279 | static void |
5280 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 5281 | struct trace_ctx *trace) |
1ed8d352 | 5282 | { |
bbafd73b | 5283 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
5284 | |
5285 | ds_put_char_multiple(result, '\t', level); | |
5286 | ds_put_format(result, "%s: ", title); | |
5287 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
5288 | ds_put_char(result, '\n'); | |
5289 | } | |
5290 | ||
abe529af | 5291 | static void |
4d0acc70 | 5292 | trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse) |
abe529af | 5293 | { |
4d0acc70 | 5294 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
abe529af BP |
5295 | struct ds *result = trace->result; |
5296 | ||
5297 | ds_put_char(result, '\n'); | |
4d0acc70 EJ |
5298 | trace_format_flow(result, recurse + 1, "Resubmitted flow", trace); |
5299 | trace_format_regs(result, recurse + 1, "Resubmitted regs", trace); | |
5300 | trace_format_odp(result, recurse + 1, "Resubmitted odp", trace); | |
5301 | trace_format_rule(result, recurse + 1, rule); | |
abe529af BP |
5302 | } |
5303 | ||
479df176 | 5304 | static void |
4d0acc70 | 5305 | trace_report(struct xlate_in *xin, const char *s, int recurse) |
479df176 | 5306 | { |
4d0acc70 | 5307 | struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); |
479df176 BP |
5308 | struct ds *result = trace->result; |
5309 | ||
4d0acc70 | 5310 | ds_put_char_multiple(result, '\t', recurse); |
479df176 BP |
5311 | ds_put_cstr(result, s); |
5312 | ds_put_char(result, '\n'); | |
5313 | } | |
5314 | ||
b6f00895 BP |
5315 | /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following |
5316 | * forms are supported: | |
5317 | * | |
5318 | * - [dpname] odp_flow [-generate | packet] | |
5319 | * - bridge br_flow [-generate | packet] | |
5320 | * | |
5321 | * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure | |
5322 | * returns a nonnull error message. */ | |
5323 | static const char * | |
5324 | parse_flow_and_packet(int argc, const char *argv[], | |
5325 | struct ofproto_dpif **ofprotop, struct flow *flow, | |
5326 | struct ofpbuf **packetp) | |
abe529af | 5327 | { |
0a37839c | 5328 | const struct dpif_backer *backer = NULL; |
b6f00895 BP |
5329 | const char *error = NULL; |
5330 | struct simap port_names = SIMAP_INITIALIZER(&port_names); | |
876b0e1c | 5331 | struct ofpbuf *packet; |
b6f00895 BP |
5332 | struct ofpbuf odp_key; |
5333 | struct ofpbuf odp_mask; | |
abe529af | 5334 | |
50aa28fd | 5335 | ofpbuf_init(&odp_key, 0); |
ea5e3887 | 5336 | ofpbuf_init(&odp_mask, 0); |
abe529af | 5337 | |
50aa28fd AW |
5338 | /* Handle "-generate" or a hex string as the last argument. */ |
5339 | if (!strcmp(argv[argc - 1], "-generate")) { | |
5340 | packet = ofpbuf_new(0); | |
5341 | argc--; | |
5342 | } else { | |
b6f00895 | 5343 | error = eth_from_hex(argv[argc - 1], &packet); |
50aa28fd AW |
5344 | if (!error) { |
5345 | argc--; | |
5346 | } else if (argc == 4) { | |
5347 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
50aa28fd AW |
5348 | goto exit; |
5349 | } | |
e84173dc | 5350 | } |
876b0e1c | 5351 | |
0a37839c GS |
5352 | /* odp_flow can have its in_port specified as a name instead of port no. |
5353 | * We do not yet know whether a given flow is a odp_flow or a br_flow. | |
5354 | * But, to know whether a flow is odp_flow through odp_flow_from_string(), | |
5355 | * we need to create a simap of name to port no. */ | |
5356 | if (argc == 3) { | |
5357 | const char *dp_type; | |
5358 | if (!strncmp(argv[1], "ovs-", 4)) { | |
5359 | dp_type = argv[1] + 4; | |
5360 | } else { | |
5361 | dp_type = argv[1]; | |
5362 | } | |
5363 | backer = shash_find_data(&all_dpif_backers, dp_type); | |
b6f00895 | 5364 | } else if (argc == 2) { |
0a37839c GS |
5365 | struct shash_node *node; |
5366 | if (shash_count(&all_dpif_backers) == 1) { | |
5367 | node = shash_first(&all_dpif_backers); | |
5368 | backer = node->data; | |
5369 | } | |
b6f00895 BP |
5370 | } else { |
5371 | error = "Syntax error"; | |
5372 | goto exit; | |
0a37839c GS |
5373 | } |
5374 | if (backer && backer->dpif) { | |
5375 | struct dpif_port dpif_port; | |
5376 | struct dpif_port_dump port_dump; | |
5377 | DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) { | |
5378 | simap_put(&port_names, dpif_port.name, | |
5379 | odp_to_u32(dpif_port.port_no)); | |
5380 | } | |
5381 | } | |
5382 | ||
50aa28fd AW |
5383 | /* Parse the flow and determine whether a datapath or |
5384 | * bridge is specified. If function odp_flow_key_from_string() | |
5385 | * returns 0, the flow is a odp_flow. If function | |
5386 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
b6f00895 BP |
5387 | if (!odp_flow_from_string(argv[argc - 1], &port_names, |
5388 | &odp_key, &odp_mask)) { | |
0a37839c | 5389 | if (!backer) { |
b6f00895 | 5390 | error = "Cannot find the datapath"; |
0a37839c | 5391 | goto exit; |
876b0e1c | 5392 | } |
8b3b8dd1 | 5393 | |
b6f00895 BP |
5394 | if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow, |
5395 | NULL, ofprotop, NULL)) { | |
5396 | error = "Invalid datapath flow"; | |
50aa28fd | 5397 | goto exit; |
8b3b8dd1 | 5398 | } |
b6f00895 | 5399 | } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) { |
50aa28fd | 5400 | if (argc != 3) { |
b6f00895 | 5401 | error = "Must specify bridge name"; |
876b0e1c BP |
5402 | goto exit; |
5403 | } | |
5404 | ||
b6f00895 BP |
5405 | *ofprotop = ofproto_dpif_lookup(argv[1]); |
5406 | if (!*ofprotop) { | |
5407 | error = "Unknown bridge name"; | |
50aa28fd AW |
5408 | goto exit; |
5409 | } | |
876b0e1c | 5410 | } else { |
b6f00895 | 5411 | error = "Bad flow syntax"; |
abe529af BP |
5412 | goto exit; |
5413 | } | |
5414 | ||
50aa28fd AW |
5415 | /* Generate a packet, if requested. */ |
5416 | if (packet) { | |
5417 | if (!packet->size) { | |
b6f00895 | 5418 | flow_compose(packet, flow); |
50aa28fd | 5419 | } else { |
b6f00895 | 5420 | union flow_in_port in_port = flow->in_port; |
50aa28fd AW |
5421 | |
5422 | /* Use the metadata from the flow and the packet argument | |
5423 | * to reconstruct the flow. */ | |
b6f00895 BP |
5424 | flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL, |
5425 | &in_port, flow); | |
50aa28fd AW |
5426 | } |
5427 | } | |
5428 | ||
b6f00895 | 5429 | error = NULL; |
6a6455e5 EJ |
5430 | |
5431 | exit: | |
b6f00895 BP |
5432 | if (error) { |
5433 | ofpbuf_delete(packet); | |
5434 | packet = NULL; | |
5435 | } | |
5436 | *packetp = packet; | |
6a6455e5 | 5437 | ofpbuf_uninit(&odp_key); |
ea5e3887 | 5438 | ofpbuf_uninit(&odp_mask); |
0a37839c | 5439 | simap_destroy(&port_names); |
b6f00895 BP |
5440 | return error; |
5441 | } | |
5442 | ||
5443 | static void | |
5444 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], | |
5445 | void *aux OVS_UNUSED) | |
5446 | { | |
5447 | struct ofproto_dpif *ofproto; | |
5448 | struct ofpbuf *packet; | |
5449 | const char *error; | |
5450 | struct flow flow; | |
5451 | ||
5452 | error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet); | |
5453 | if (!error) { | |
5454 | struct ds result; | |
5455 | ||
5456 | ds_init(&result); | |
aee0979b | 5457 | ofproto_trace(ofproto, &flow, packet, NULL, 0, &result); |
b6f00895 BP |
5458 | unixctl_command_reply(conn, ds_cstr(&result)); |
5459 | ds_destroy(&result); | |
5460 | ofpbuf_delete(packet); | |
5461 | } else { | |
5462 | unixctl_command_reply_error(conn, error); | |
5463 | } | |
6a6455e5 EJ |
5464 | } |
5465 | ||
aee0979b BP |
5466 | static void |
5467 | ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc, | |
5468 | const char *argv[], void *aux OVS_UNUSED) | |
5469 | { | |
5470 | enum ofputil_protocol usable_protocols; | |
5471 | struct ofproto_dpif *ofproto; | |
5472 | bool enforce_consistency; | |
5473 | struct ofpbuf ofpacts; | |
5474 | struct ofpbuf *packet; | |
5475 | struct ds result; | |
5476 | struct flow flow; | |
5477 | uint16_t in_port; | |
5478 | ||
5479 | /* Three kinds of error return values! */ | |
5480 | enum ofperr retval; | |
5481 | const char *error; | |
5482 | char *rw_error; | |
5483 | ||
5484 | packet = NULL; | |
5485 | ds_init(&result); | |
5486 | ofpbuf_init(&ofpacts, 0); | |
5487 | ||
5488 | /* Parse actions. */ | |
5489 | rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols); | |
5490 | if (rw_error) { | |
5491 | unixctl_command_reply_error(conn, rw_error); | |
5492 | free(rw_error); | |
5493 | goto exit; | |
5494 | } | |
5495 | ||
5496 | /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of | |
ba2fe8e9 BP |
5497 | * consistency between the flow and the actions. With -consistent, we |
5498 | * enforce consistency even for a flow supported in OpenFlow 1.0. */ | |
aee0979b BP |
5499 | if (!strcmp(argv[1], "-consistent")) { |
5500 | enforce_consistency = true; | |
5501 | argv++; | |
5502 | argc--; | |
ba2fe8e9 BP |
5503 | } else { |
5504 | enforce_consistency = false; | |
aee0979b BP |
5505 | } |
5506 | ||
5507 | error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet); | |
5508 | if (error) { | |
5509 | unixctl_command_reply_error(conn, error); | |
5510 | goto exit; | |
5511 | } | |
5512 | ||
5513 | /* Do the same checks as handle_packet_out() in ofproto.c. | |
5514 | * | |
5515 | * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't | |
5516 | * strictly correct because these actions aren't in any table, but it's OK | |
5517 | * because it 'table_id' is used only to check goto_table instructions, but | |
5518 | * packet-outs take a list of actions and therefore it can't include | |
5519 | * instructions. | |
5520 | * | |
5521 | * We skip the "meter" check here because meter is an instruction, not an | |
5522 | * action, and thus cannot appear in ofpacts. */ | |
5523 | in_port = ofp_to_u16(flow.in_port.ofp_port); | |
5524 | if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) { | |
5525 | unixctl_command_reply_error(conn, "invalid in_port"); | |
5526 | goto exit; | |
5527 | } | |
ba2fe8e9 BP |
5528 | if (enforce_consistency) { |
5529 | retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow, | |
5530 | u16_to_ofp(ofproto->up.max_ports), | |
5531 | 0, 0, usable_protocols); | |
5532 | } else { | |
5533 | retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow, | |
5534 | u16_to_ofp(ofproto->up.max_ports), 0, 0, | |
5535 | &usable_protocols); | |
5536 | } | |
5537 | ||
aee0979b BP |
5538 | if (retval) { |
5539 | ds_clear(&result); | |
5540 | ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval)); | |
5541 | unixctl_command_reply_error(conn, ds_cstr(&result)); | |
5542 | goto exit; | |
5543 | } | |
5544 | ||
5545 | ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result); | |
5546 | unixctl_command_reply(conn, ds_cstr(&result)); | |
5547 | ||
5548 | exit: | |
5549 | ds_destroy(&result); | |
5550 | ofpbuf_delete(packet); | |
5551 | ofpbuf_uninit(&ofpacts); | |
5552 | } | |
5553 | ||
5554 | /* Implements a "trace" through 'ofproto''s flow table, appending a textual | |
5555 | * description of the results to 'ds'. | |
5556 | * | |
5557 | * The trace follows a packet with the specified 'flow' through the flow | |
5558 | * table. 'packet' may be nonnull to trace an actual packet, with consequent | |
5559 | * side effects (if it is nonnull then its flow must be 'flow'). | |
5560 | * | |
5561 | * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to | |
5562 | * trace, otherwise the actions are determined by a flow table lookup. */ | |
ade6ad9c | 5563 | static void |
6a6455e5 | 5564 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, |
aee0979b BP |
5565 | const struct ofpbuf *packet, |
5566 | const struct ofpact ofpacts[], size_t ofpacts_len, | |
5567 | struct ds *ds) | |
6a6455e5 EJ |
5568 | { |
5569 | struct rule_dpif *rule; | |
083f277e | 5570 | struct flow_wildcards wc; |
6a6455e5 | 5571 | |
aee0979b | 5572 | ds_put_format(ds, "Bridge: %s\n", ofproto->up.name); |
6a6455e5 EJ |
5573 | ds_put_cstr(ds, "Flow: "); |
5574 | flow_format(ds, flow); | |
5575 | ds_put_char(ds, '\n'); | |
abe529af | 5576 | |
083f277e | 5577 | flow_wildcards_init_catchall(&wc); |
aee0979b BP |
5578 | if (ofpacts) { |
5579 | rule = NULL; | |
5580 | } else { | |
5581 | rule_dpif_lookup(ofproto, flow, &wc, &rule); | |
5582 | ||
5583 | trace_format_rule(ds, 0, rule); | |
5584 | if (rule == ofproto->miss_rule) { | |
5585 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
5586 | } else if (rule == ofproto->no_packet_in_rule) { | |
5587 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
5588 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
5589 | } else if (rule == ofproto->drop_frags_rule) { | |
5590 | ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " | |
5591 | "and the fragment handling mode is \"drop\".\n"); | |
5592 | } | |
c57b2226 BP |
5593 | } |
5594 | ||
aee0979b | 5595 | if (rule || ofpacts) { |
050ac423 BP |
5596 | uint64_t odp_actions_stub[1024 / 8]; |
5597 | struct ofpbuf odp_actions; | |
6a6455e5 | 5598 | struct trace_ctx trace; |
bcd2633a | 5599 | struct match match; |
a66733a8 | 5600 | uint16_t tcp_flags; |
abe529af | 5601 | |
6a6455e5 EJ |
5602 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
5603 | trace.result = ds; | |
5604 | trace.flow = *flow; | |
050ac423 BP |
5605 | ofpbuf_use_stub(&odp_actions, |
5606 | odp_actions_stub, sizeof odp_actions_stub); | |
47b0deae | 5607 | xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); |
aee0979b BP |
5608 | if (ofpacts) { |
5609 | trace.xin.ofpacts = ofpacts; | |
5610 | trace.xin.ofpacts_len = ofpacts_len; | |
5611 | } | |
bbafd73b EJ |
5612 | trace.xin.resubmit_hook = trace_resubmit; |
5613 | trace.xin.report_hook = trace_report; | |
bcd2633a | 5614 | |
bbafd73b | 5615 | xlate_actions(&trace.xin, &trace.xout); |
083f277e | 5616 | flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc); |
abe529af | 5617 | |
6a6455e5 EJ |
5618 | ds_put_char(ds, '\n'); |
5619 | trace_format_flow(ds, 0, "Final flow", &trace); | |
bcd2633a JP |
5620 | |
5621 | match_init(&match, flow, &trace.xout.wc); | |
5622 | ds_put_cstr(ds, "Relevant fields: "); | |
5623 | match_format(&match, ds, OFP_DEFAULT_PRIORITY); | |
5624 | ds_put_char(ds, '\n'); | |
5625 | ||
6a6455e5 | 5626 | ds_put_cstr(ds, "Datapath actions: "); |
bbafd73b EJ |
5627 | format_odp_actions(ds, trace.xout.odp_actions.data, |
5628 | trace.xout.odp_actions.size); | |
876b0e1c | 5629 | |
bbafd73b | 5630 | if (trace.xout.slow) { |
04594cd5 BP |
5631 | enum slow_path_reason slow; |
5632 | ||
6a7e895f BP |
5633 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
5634 | "slow path because it:"); | |
04594cd5 BP |
5635 | |
5636 | slow = trace.xout.slow; | |
5637 | while (slow) { | |
5638 | enum slow_path_reason bit = rightmost_1bit(slow); | |
5639 | ||
5640 | ds_put_format(ds, "\n\t- %s.", | |
5641 | slow_path_reason_to_explanation(bit)); | |
5642 | ||
5643 | slow &= ~bit; | |
6a7e895f | 5644 | } |
876b0e1c | 5645 | } |
bbafd73b EJ |
5646 | |
5647 | xlate_out_uninit(&trace.xout); | |
abe529af | 5648 | } |
ad3efdcb | 5649 | |
a2143702 | 5650 | rule_dpif_unref(rule); |
abe529af BP |
5651 | } |
5652 | ||
6814e51f BP |
5653 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
5654 | * 'reply' describing the results. */ | |
5655 | static void | |
5656 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
5657 | { | |
bcd2633a | 5658 | struct cls_cursor cursor; |
6814e51f BP |
5659 | struct facet *facet; |
5660 | int errors; | |
5661 | ||
5662 | errors = 0; | |
0b4f2078 | 5663 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5664 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5665 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
6814e51f BP |
5666 | if (!facet_check_consistency(facet)) { |
5667 | errors++; | |
5668 | } | |
5669 | } | |
0b4f2078 | 5670 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
6814e51f | 5671 | if (errors) { |
2cc3c58e | 5672 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
5673 | } |
5674 | ||
5675 | if (errors) { | |
5676 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
5677 | ofproto->up.name, errors); | |
5678 | } else { | |
5679 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
5680 | } | |
5681 | } | |
5682 | ||
5683 | static void | |
5684 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
5685 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
5686 | { | |
5687 | struct ds reply = DS_EMPTY_INITIALIZER; | |
5688 | struct ofproto_dpif *ofproto; | |
5689 | ||
5690 | if (argc > 1) { | |
5691 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5692 | if (!ofproto) { | |
bde9f75d EJ |
5693 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
5694 | "ofproto/list for help)"); | |
6814e51f BP |
5695 | return; |
5696 | } | |
5697 | ofproto_dpif_self_check__(ofproto, &reply); | |
5698 | } else { | |
5699 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5700 | ofproto_dpif_self_check__(ofproto, &reply); | |
5701 | } | |
5702 | } | |
5703 | ||
bde9f75d | 5704 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
5705 | ds_destroy(&reply); |
5706 | } | |
5707 | ||
27022416 JP |
5708 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
5709 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
5710 | * to destroy 'ofproto_shash' and free the returned value. */ | |
5711 | static const struct shash_node ** | |
5712 | get_ofprotos(struct shash *ofproto_shash) | |
5713 | { | |
5714 | const struct ofproto_dpif *ofproto; | |
5715 | ||
5716 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5717 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
5718 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
5719 | } | |
5720 | ||
5721 | return shash_sort(ofproto_shash); | |
5722 | } | |
5723 | ||
5724 | static void | |
5725 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
5726 | const char *argv[] OVS_UNUSED, | |
5727 | void *aux OVS_UNUSED) | |
5728 | { | |
5729 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5730 | struct shash ofproto_shash; | |
5731 | const struct shash_node **sorted_ofprotos; | |
5732 | int i; | |
5733 | ||
5734 | shash_init(&ofproto_shash); | |
5735 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
5736 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5737 | const struct shash_node *node = sorted_ofprotos[i]; | |
5738 | ds_put_format(&ds, "%s\n", node->name); | |
5739 | } | |
5740 | ||
5741 | shash_destroy(&ofproto_shash); | |
5742 | free(sorted_ofprotos); | |
5743 | ||
5744 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5745 | ds_destroy(&ds); | |
5746 | } | |
5747 | ||
dc54ef36 EJ |
5748 | static void |
5749 | dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) | |
27022416 | 5750 | { |
dc54ef36 EJ |
5751 | const struct shash_node **ofprotos; |
5752 | struct ofproto_dpif *ofproto; | |
5753 | struct shash ofproto_shash; | |
09672174 | 5754 | uint64_t n_hit, n_missed; |
09672174 | 5755 | size_t i; |
655ab909 | 5756 | |
04d08d54 | 5757 | n_hit = n_missed = 0; |
dc54ef36 EJ |
5758 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
5759 | if (ofproto->backer == backer) { | |
dc54ef36 EJ |
5760 | n_missed += ofproto->n_missed; |
5761 | n_hit += ofproto->n_hit; | |
5762 | } | |
5763 | } | |
655ab909 | 5764 | |
dc54ef36 EJ |
5765 | ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
5766 | dpif_name(backer->dpif), n_hit, n_missed); | |
34582733 | 5767 | ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u," |
04d08d54 | 5768 | " life span: %lldms\n", hmap_count(&backer->subfacets), |
dc54ef36 EJ |
5769 | backer->avg_n_subfacet, backer->max_n_subfacet, |
5770 | backer->avg_subfacet_life); | |
5771 | ||
dc54ef36 EJ |
5772 | shash_init(&ofproto_shash); |
5773 | ofprotos = get_ofprotos(&ofproto_shash); | |
5774 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
5775 | struct ofproto_dpif *ofproto = ofprotos[i]->data; | |
5776 | const struct shash_node **ports; | |
5777 | size_t j; | |
0a740f48 | 5778 | |
dc54ef36 EJ |
5779 | if (ofproto->backer != backer) { |
5780 | continue; | |
0a740f48 | 5781 | } |
27022416 | 5782 | |
dc54ef36 EJ |
5783 | ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
5784 | ofproto->up.name, ofproto->n_hit, ofproto->n_missed); | |
5785 | ||
5786 | ports = shash_sort(&ofproto->up.port_by_name); | |
5787 | for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { | |
5788 | const struct shash_node *node = ports[j]; | |
5789 | struct ofport *ofport = node->data; | |
5790 | struct smap config; | |
4e022ec0 | 5791 | odp_port_t odp_port; |
27022416 | 5792 | |
dc54ef36 EJ |
5793 | ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), |
5794 | ofport->ofp_port); | |
27022416 | 5795 | |
dc54ef36 | 5796 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); |
4e022ec0 | 5797 | if (odp_port != ODPP_NONE) { |
dc54ef36 EJ |
5798 | ds_put_format(ds, "%"PRIu32":", odp_port); |
5799 | } else { | |
5800 | ds_put_cstr(ds, "none:"); | |
5801 | } | |
27022416 | 5802 | |
dc54ef36 | 5803 | ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); |
27022416 | 5804 | |
dc54ef36 EJ |
5805 | smap_init(&config); |
5806 | if (!netdev_get_config(ofport->netdev, &config)) { | |
5807 | const struct smap_node **nodes; | |
5808 | size_t i; | |
27022416 | 5809 | |
dc54ef36 EJ |
5810 | nodes = smap_sort(&config); |
5811 | for (i = 0; i < smap_count(&config); i++) { | |
5812 | const struct smap_node *node = nodes[i]; | |
5813 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
5814 | node->key, node->value); | |
5815 | } | |
5816 | free(nodes); | |
27022416 | 5817 | } |
dc54ef36 EJ |
5818 | smap_destroy(&config); |
5819 | ||
27022416 | 5820 | ds_put_char(ds, ')'); |
dc54ef36 | 5821 | ds_put_char(ds, '\n'); |
27022416 | 5822 | } |
dc54ef36 | 5823 | free(ports); |
27022416 | 5824 | } |
dc54ef36 EJ |
5825 | shash_destroy(&ofproto_shash); |
5826 | free(ofprotos); | |
27022416 JP |
5827 | } |
5828 | ||
5829 | static void | |
dc54ef36 EJ |
5830 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5831 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
27022416 JP |
5832 | { |
5833 | struct ds ds = DS_EMPTY_INITIALIZER; | |
dc54ef36 EJ |
5834 | const struct shash_node **backers; |
5835 | int i; | |
27022416 | 5836 | |
dc54ef36 EJ |
5837 | backers = shash_sort(&all_dpif_backers); |
5838 | for (i = 0; i < shash_count(&all_dpif_backers); i++) { | |
5839 | dpif_show_backer(backers[i]->data, &ds); | |
27022416 | 5840 | } |
dc54ef36 | 5841 | free(backers); |
27022416 JP |
5842 | |
5843 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5844 | ds_destroy(&ds); | |
5845 | } | |
5846 | ||
bcd2633a JP |
5847 | /* Dump the megaflow (facet) cache. This is useful to check the |
5848 | * correctness of flow wildcarding, since the same mechanism is used for | |
5849 | * both xlate caching and kernel wildcarding. | |
5850 | * | |
5851 | * It's important to note that in the output the flow description uses | |
5852 | * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports. | |
5853 | * | |
5854 | * This command is only needed for advanced debugging, so it's not | |
5855 | * documented in the man page. */ | |
5856 | static void | |
5857 | ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn, | |
5858 | int argc OVS_UNUSED, const char *argv[], | |
5859 | void *aux OVS_UNUSED) | |
5860 | { | |
5861 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5862 | const struct ofproto_dpif *ofproto; | |
5863 | long long int now = time_msec(); | |
5864 | struct cls_cursor cursor; | |
5865 | struct facet *facet; | |
5866 | ||
5867 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5868 | if (!ofproto) { | |
5869 | unixctl_command_reply_error(conn, "no such bridge"); | |
5870 | return; | |
5871 | } | |
5872 | ||
0b4f2078 | 5873 | ovs_rwlock_rdlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5874 | cls_cursor_init(&cursor, &ofproto->facets, NULL); |
5875 | CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { | |
5876 | cls_rule_format(&facet->cr, &ds); | |
5877 | ds_put_cstr(&ds, ", "); | |
34582733 | 5878 | ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets)); |
bcd2633a JP |
5879 | ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); |
5880 | ds_put_cstr(&ds, "Datapath actions: "); | |
8ecafa3e JP |
5881 | if (facet->xout.slow) { |
5882 | uint64_t slow_path_stub[128 / 8]; | |
5883 | const struct nlattr *actions; | |
5884 | size_t actions_len; | |
5885 | ||
5886 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, | |
5887 | slow_path_stub, sizeof slow_path_stub, | |
5888 | &actions, &actions_len); | |
5889 | format_odp_actions(&ds, actions, actions_len); | |
5890 | } else { | |
5891 | format_odp_actions(&ds, facet->xout.odp_actions.data, | |
5892 | facet->xout.odp_actions.size); | |
5893 | } | |
bcd2633a JP |
5894 | ds_put_cstr(&ds, "\n"); |
5895 | } | |
0b4f2078 | 5896 | ovs_rwlock_unlock(&ofproto->facets.rwlock); |
bcd2633a JP |
5897 | |
5898 | ds_chomp(&ds, '\n'); | |
5899 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
5900 | ds_destroy(&ds); | |
5901 | } | |
5902 | ||
6b35e630 JP |
5903 | /* Disable using the megaflows. |
5904 | * | |
5905 | * This command is only needed for advanced debugging, so it's not | |
5906 | * documented in the man page. */ | |
5907 | static void | |
5908 | ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn, | |
5909 | int argc OVS_UNUSED, | |
5910 | const char *argv[] OVS_UNUSED, | |
5911 | void *aux OVS_UNUSED) | |
5912 | { | |
5913 | struct ofproto_dpif *ofproto; | |
5914 | ||
5915 | enable_megaflows = false; | |
5916 | ||
5917 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5918 | flush(&ofproto->up); | |
5919 | } | |
5920 | ||
5921 | unixctl_command_reply(conn, "megaflows disabled"); | |
5922 | } | |
5923 | ||
5924 | /* Re-enable using megaflows. | |
5925 | * | |
5926 | * This command is only needed for advanced debugging, so it's not | |
5927 | * documented in the man page. */ | |
5928 | static void | |
5929 | ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn, | |
5930 | int argc OVS_UNUSED, | |
5931 | const char *argv[] OVS_UNUSED, | |
5932 | void *aux OVS_UNUSED) | |
5933 | { | |
5934 | struct ofproto_dpif *ofproto; | |
5935 | ||
5936 | enable_megaflows = true; | |
5937 | ||
5938 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5939 | flush(&ofproto->up); | |
5940 | } | |
5941 | ||
5942 | unixctl_command_reply(conn, "megaflows enabled"); | |
5943 | } | |
5944 | ||
27022416 JP |
5945 | static void |
5946 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
5947 | int argc OVS_UNUSED, const char *argv[], | |
5948 | void *aux OVS_UNUSED) | |
5949 | { | |
5950 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5951 | const struct ofproto_dpif *ofproto; | |
5952 | struct subfacet *subfacet; | |
5953 | ||
5954 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5955 | if (!ofproto) { | |
5956 | unixctl_command_reply_error(conn, "no such bridge"); | |
5957 | return; | |
5958 | } | |
5959 | ||
af37354d EJ |
5960 | update_stats(ofproto->backer); |
5961 | ||
04d08d54 | 5962 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { |
4dff9097 | 5963 | struct facet *facet = subfacet->facet; |
64f9cb76 JP |
5964 | struct odputil_keybuf maskbuf; |
5965 | struct ofpbuf mask; | |
4dff9097 | 5966 | |
b7e2f6b3 | 5967 | if (facet->ofproto != ofproto) { |
04d08d54 EJ |
5968 | continue; |
5969 | } | |
5970 | ||
64f9cb76 JP |
5971 | ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); |
5972 | if (enable_megaflows) { | |
5973 | odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, | |
5974 | &facet->flow, UINT32_MAX); | |
5975 | } | |
5976 | ||
5977 | odp_flow_format(subfacet->key, subfacet->key_len, | |
0a37839c | 5978 | mask.data, mask.size, NULL, &ds, false); |
27022416 JP |
5979 | |
5980 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
5981 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
5982 | if (subfacet->used) { | |
5983 | ds_put_format(&ds, "%.3fs", | |
5984 | (time_msec() - subfacet->used) / 1000.0); | |
5985 | } else { | |
5986 | ds_put_format(&ds, "never"); | |
5987 | } | |
5988 | if (subfacet->facet->tcp_flags) { | |
5989 | ds_put_cstr(&ds, ", flags:"); | |
5990 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
5991 | } | |
5992 | ||
5993 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 5994 | if (facet->xout.slow) { |
f2245da3 JP |
5995 | uint64_t slow_path_stub[128 / 8]; |
5996 | const struct nlattr *actions; | |
5997 | size_t actions_len; | |
5998 | ||
bbafd73b | 5999 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
6000 | slow_path_stub, sizeof slow_path_stub, |
6001 | &actions, &actions_len); | |
6002 | format_odp_actions(&ds, actions, actions_len); | |
6003 | } else { | |
bbafd73b EJ |
6004 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
6005 | facet->xout.odp_actions.size); | |
f2245da3 | 6006 | } |
27022416 JP |
6007 | ds_put_char(&ds, '\n'); |
6008 | } | |
6009 | ||
6010 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6011 | ds_destroy(&ds); | |
6012 | } | |
6013 | ||
6014 | static void | |
6015 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
6016 | int argc OVS_UNUSED, const char *argv[], | |
6017 | void *aux OVS_UNUSED) | |
6018 | { | |
6019 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6020 | struct ofproto_dpif *ofproto; | |
6021 | ||
6022 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6023 | if (!ofproto) { | |
6024 | unixctl_command_reply_error(conn, "no such bridge"); | |
6025 | return; | |
6026 | } | |
6027 | ||
6028 | flush(&ofproto->up); | |
6029 | ||
6030 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6031 | ds_destroy(&ds); | |
6032 | } | |
6033 | ||
abe529af BP |
6034 | static void |
6035 | ofproto_dpif_unixctl_init(void) | |
6036 | { | |
6037 | static bool registered; | |
6038 | if (registered) { | |
6039 | return; | |
6040 | } | |
6041 | registered = true; | |
6042 | ||
0e15264f BP |
6043 | unixctl_command_register( |
6044 | "ofproto/trace", | |
dc8ce81f | 6045 | "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]", |
50aa28fd | 6046 | 1, 3, ofproto_unixctl_trace, NULL); |
aee0979b BP |
6047 | unixctl_command_register( |
6048 | "ofproto/trace-packet-out", | |
6049 | "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions", | |
6050 | 2, 6, ofproto_unixctl_trace_actions, NULL); | |
96e466a3 | 6051 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6052 | ofproto_unixctl_fdb_flush, NULL); |
6053 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6054 | ofproto_unixctl_fdb_show, NULL); | |
6814e51f BP |
6055 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6056 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
6057 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
6058 | ofproto_unixctl_dpif_dump_dps, NULL); | |
dc54ef36 EJ |
6059 | unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show, |
6060 | NULL); | |
27022416 JP |
6061 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, |
6062 | ofproto_unixctl_dpif_dump_flows, NULL); | |
6063 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
6064 | ofproto_unixctl_dpif_del_flows, NULL); | |
bcd2633a JP |
6065 | unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1, |
6066 | ofproto_unixctl_dpif_dump_megaflows, NULL); | |
6b35e630 JP |
6067 | unixctl_command_register("dpif/disable-megaflows", "", 0, 0, |
6068 | ofproto_unixctl_dpif_disable_megaflows, NULL); | |
6069 | unixctl_command_register("dpif/enable-megaflows", "", 0, 0, | |
6070 | ofproto_unixctl_dpif_enable_megaflows, NULL); | |
abe529af BP |
6071 | } |
6072 | \f | |
52a90c29 BP |
6073 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6074 | * | |
6075 | * This is deprecated. It is only for compatibility with broken device drivers | |
6076 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6077 | * devices are not used. When broken device drivers are no longer in | |
6078 | * widespread use, we will delete these interfaces. */ | |
6079 | ||
6080 | static int | |
4e022ec0 | 6081 | set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6082 | { |
6083 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6084 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6085 | ||
6086 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6087 | && vid == ofport->vlandev_vid) { | |
6088 | return 0; | |
6089 | } | |
6090 | ||
2cc3c58e | 6091 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
6092 | |
6093 | if (ofport->realdev_ofp_port) { | |
6094 | vsp_remove(ofport); | |
6095 | } | |
6096 | if (realdev_ofp_port && ofport->bundle) { | |
6097 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6098 | * themselves be part of a bundle. */ | |
6099 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6100 | } | |
6101 | ||
6102 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6103 | ofport->vlandev_vid = vid; | |
6104 | ||
6105 | if (realdev_ofp_port) { | |
6106 | vsp_add(ofport, realdev_ofp_port, vid); | |
6107 | } | |
6108 | ||
6109 | return 0; | |
6110 | } | |
6111 | ||
6112 | static uint32_t | |
4e022ec0 | 6113 | hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid) |
52a90c29 | 6114 | { |
4e022ec0 | 6115 | return hash_2words(ofp_to_u16(realdev_ofp_port), vid); |
52a90c29 BP |
6116 | } |
6117 | ||
46c88433 EJ |
6118 | bool |
6119 | ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto) | |
13501305 | 6120 | OVS_EXCLUDED(ofproto->vsp_mutex) |
46c88433 | 6121 | { |
13501305 EJ |
6122 | bool ret; |
6123 | ||
6124 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
6125 | ret = !hmap_is_empty(&ofproto->realdev_vid_map); | |
6126 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
6127 | return ret; | |
46c88433 EJ |
6128 | } |
6129 | ||
13501305 EJ |
6130 | static ofp_port_t |
6131 | vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto, | |
6132 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
6133 | OVS_REQUIRES(ofproto->vsp_mutex) | |
52a90c29 BP |
6134 | { |
6135 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
52a90c29 BP |
6136 | int vid = vlan_tci_to_vid(vlan_tci); |
6137 | const struct vlan_splinter *vsp; | |
6138 | ||
6139 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6140 | hash_realdev_vid(realdev_ofp_port, vid), | |
6141 | &ofproto->realdev_vid_map) { | |
6142 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6143 | && vsp->vid == vid) { | |
deea1200 | 6144 | return vsp->vlandev_ofp_port; |
52a90c29 BP |
6145 | } |
6146 | } | |
6147 | } | |
deea1200 | 6148 | return realdev_ofp_port; |
52a90c29 BP |
6149 | } |
6150 | ||
13501305 EJ |
6151 | /* Returns the OFP port number of the Linux VLAN device that corresponds to |
6152 | * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in | |
6153 | * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and | |
6154 | * 'vlan_tci' 9, it would return the port number of eth0.9. | |
6155 | * | |
6156 | * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this | |
6157 | * function just returns its 'realdev_ofp_port' argument. */ | |
6158 | ofp_port_t | |
6159 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6160 | ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) | |
6161 | OVS_EXCLUDED(ofproto->vsp_mutex) | |
6162 | { | |
6163 | ofp_port_t ret; | |
6164 | ||
6165 | ovs_mutex_lock(&ofproto->vsp_mutex); | |
6166 | ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci); | |
6167 | ovs_mutex_unlock(&ofproto->vsp_mutex); | |
6168 | return ret; | |
6169 | } | |
6170 | ||
52a90c29 | 6171 | static struct vlan_splinter * |
4e022ec0 | 6172 | vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) |
52a90c29 BP |
6173 | { |
6174 | struct vlan_splinter *vsp; | |
6175 | ||
4e022ec0 | 6176 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, |
f9c0c3ec | 6177 | hash_ofp_port(vlandev_ofp_port), |
52a90c29 BP |
6178 | &ofproto->vlandev_map) { |
6179 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6180 | return vsp; | |
6181 | } | |
6182 | } | |
6183 | ||
6184 | return NULL; | |
6185 | } | |
6186 | ||
40e05935 BP |
6187 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
6188 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
6189 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
6190 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
6191 | * eth0 and store 9 in '*vid'. | |
6192 | * | |
6193 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
6194 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
6195 | * always does.*/ | |
4e022ec0 | 6196 | static ofp_port_t |
52a90c29 | 6197 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, |
4e022ec0 | 6198 | ofp_port_t vlandev_ofp_port, int *vid) |
bd3950dd | 6199 | OVS_REQUIRES(ofproto->vsp_mutex) |
52a90c29 BP |
6200 | { |
6201 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6202 | const struct vlan_splinter *vsp; | |
6203 | ||
6204 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6205 | if (vsp) { | |
6206 | if (vid) { | |
6207 | *vid = vsp->vid; | |
6208 | } | |
6209 | return vsp->realdev_ofp_port; | |
6210 | } | |
6211 | } | |
6212 | return 0; | |
6213 | } | |
6214 | ||
b98d8985 BP |
6215 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
6216 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
6217 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
6218 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
6219 | * always the case unless VLAN splinters are enabled), returns false without | |
6220 | * making any changes. */ | |
8449c4d6 | 6221 | bool |
b98d8985 | 6222 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) |
13501305 | 6223 | OVS_EXCLUDED(ofproto->vsp_mutex) |
b98d8985 | 6224 | { |
4e022ec0 | 6225 | ofp_port_t realdev; |
b98d8985 BP |
6226 | int vid; |
6227 | ||
13501305 | 6228 | ovs_mutex_lock(&ofproto->vsp_mutex); |
4e022ec0 | 6229 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid); |
13501305 | 6230 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
b98d8985 BP |
6231 | if (!realdev) { |
6232 | return false; | |
6233 | } | |
6234 | ||
6235 | /* Cause the flow to be processed as if it came in on the real device with | |
6236 | * the VLAN device's VLAN ID. */ | |
4e022ec0 | 6237 | flow->in_port.ofp_port = realdev; |
b98d8985 BP |
6238 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); |
6239 | return true; | |
6240 | } | |
6241 | ||
52a90c29 BP |
6242 | static void |
6243 | vsp_remove(struct ofport_dpif *port) | |
6244 | { | |
6245 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6246 | struct vlan_splinter *vsp; | |
6247 | ||
13501305 | 6248 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 BP |
6249 | vsp = vlandev_find(ofproto, port->up.ofp_port); |
6250 | if (vsp) { | |
6251 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6252 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6253 | free(vsp); | |
6254 | ||
6255 | port->realdev_ofp_port = 0; | |
6256 | } else { | |
6257 | VLOG_ERR("missing vlan device record"); | |
6258 | } | |
13501305 | 6259 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 BP |
6260 | } |
6261 | ||
6262 | static void | |
4e022ec0 | 6263 | vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid) |
52a90c29 BP |
6264 | { |
6265 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6266 | ||
13501305 | 6267 | ovs_mutex_lock(&ofproto->vsp_mutex); |
52a90c29 | 6268 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) |
13501305 | 6269 | && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid)) |
52a90c29 BP |
6270 | == realdev_ofp_port)) { |
6271 | struct vlan_splinter *vsp; | |
6272 | ||
6273 | vsp = xmalloc(sizeof *vsp); | |
52a90c29 BP |
6274 | vsp->realdev_ofp_port = realdev_ofp_port; |
6275 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6276 | vsp->vid = vid; | |
6277 | ||
6278 | port->realdev_ofp_port = realdev_ofp_port; | |
13501305 EJ |
6279 | |
6280 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6281 | hash_ofp_port(port->up.ofp_port)); | |
6282 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6283 | hash_realdev_vid(realdev_ofp_port, vid)); | |
52a90c29 BP |
6284 | } else { |
6285 | VLOG_ERR("duplicate vlan device record"); | |
6286 | } | |
13501305 | 6287 | ovs_mutex_unlock(&ofproto->vsp_mutex); |
52a90c29 | 6288 | } |
e1b1d06a | 6289 | |
46c88433 | 6290 | static odp_port_t |
4e022ec0 | 6291 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) |
e1b1d06a JP |
6292 | { |
6293 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
4e022ec0 | 6294 | return ofport ? ofport->odp_port : ODPP_NONE; |
e1b1d06a JP |
6295 | } |
6296 | ||
8449c4d6 | 6297 | struct ofport_dpif * |
4e022ec0 | 6298 | odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port) |
e1b1d06a JP |
6299 | { |
6300 | struct ofport_dpif *port; | |
6301 | ||
8449c4d6 | 6302 | ovs_rwlock_rdlock(&backer->odp_to_ofport_lock); |
f9c0c3ec | 6303 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port), |
acf60855 | 6304 | &backer->odp_to_ofport_map) { |
e1b1d06a | 6305 | if (port->odp_port == odp_port) { |
8449c4d6 | 6306 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 | 6307 | return port; |
e1b1d06a JP |
6308 | } |
6309 | } | |
6310 | ||
8449c4d6 | 6311 | ovs_rwlock_unlock(&backer->odp_to_ofport_lock); |
acf60855 JP |
6312 | return NULL; |
6313 | } | |
6314 | ||
4e022ec0 AW |
6315 | static ofp_port_t |
6316 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) | |
acf60855 JP |
6317 | { |
6318 | struct ofport_dpif *port; | |
6319 | ||
6320 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 6321 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
6322 | return port->up.ofp_port; |
6323 | } else { | |
6324 | return OFPP_NONE; | |
6325 | } | |
e1b1d06a | 6326 | } |
655ab909 | 6327 | |
abe529af | 6328 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 6329 | init, |
abe529af BP |
6330 | enumerate_types, |
6331 | enumerate_names, | |
6332 | del, | |
0aeaabc8 | 6333 | port_open_type, |
acf60855 JP |
6334 | type_run, |
6335 | type_run_fast, | |
6336 | type_wait, | |
abe529af BP |
6337 | alloc, |
6338 | construct, | |
6339 | destruct, | |
6340 | dealloc, | |
6341 | run, | |
5fcc0d00 | 6342 | run_fast, |
abe529af | 6343 | wait, |
0d085684 | 6344 | get_memory_usage, |
abe529af | 6345 | flush, |
6c1491fb BP |
6346 | get_features, |
6347 | get_tables, | |
abe529af BP |
6348 | port_alloc, |
6349 | port_construct, | |
6350 | port_destruct, | |
6351 | port_dealloc, | |
6352 | port_modified, | |
6353 | port_reconfigured, | |
6354 | port_query_by_name, | |
6355 | port_add, | |
6356 | port_del, | |
6527c598 | 6357 | port_get_stats, |
abe529af BP |
6358 | port_dump_start, |
6359 | port_dump_next, | |
6360 | port_dump_done, | |
6361 | port_poll, | |
6362 | port_poll_wait, | |
6363 | port_is_lacp_current, | |
0ab6decf | 6364 | NULL, /* rule_choose_table */ |
abe529af BP |
6365 | rule_alloc, |
6366 | rule_construct, | |
8037acb4 BP |
6367 | rule_insert, |
6368 | rule_delete, | |
abe529af BP |
6369 | rule_destruct, |
6370 | rule_dealloc, | |
abe529af BP |
6371 | rule_get_stats, |
6372 | rule_execute, | |
6373 | rule_modify_actions, | |
7257b535 | 6374 | set_frag_handling, |
abe529af BP |
6375 | packet_out, |
6376 | set_netflow, | |
6377 | get_netflow_ids, | |
6378 | set_sflow, | |
29089a54 | 6379 | set_ipfix, |
abe529af | 6380 | set_cfm, |
9a9e3786 | 6381 | get_cfm_status, |
ccc09689 EJ |
6382 | set_bfd, |
6383 | get_bfd_status, | |
21f7563c JP |
6384 | set_stp, |
6385 | get_stp_status, | |
6386 | set_stp_port, | |
6387 | get_stp_port_status, | |
fd28ce3a | 6388 | get_stp_port_stats, |
8b36f51e | 6389 | set_queues, |
abe529af BP |
6390 | bundle_set, |
6391 | bundle_remove, | |
ec7ceaed EJ |
6392 | mirror_set__, |
6393 | mirror_get_stats__, | |
abe529af BP |
6394 | set_flood_vlans, |
6395 | is_mirror_output_bundle, | |
8402c74b | 6396 | forward_bpdu_changed, |
c4069512 | 6397 | set_mac_table_config, |
52a90c29 | 6398 | set_realdev, |
9cae45dc JR |
6399 | NULL, /* meter_get_features */ |
6400 | NULL, /* meter_set */ | |
6401 | NULL, /* meter_get */ | |
6402 | NULL, /* meter_del */ | |
00430a3f SH |
6403 | group_alloc, /* group_alloc */ |
6404 | group_construct, /* group_construct */ | |
6405 | group_destruct, /* group_destruct */ | |
6406 | group_dealloc, /* group_dealloc */ | |
6407 | group_modify, /* group_modify */ | |
6408 | group_get_stats, /* group_get_stats */ | |
abe529af | 6409 | }; |