]>
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 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
ccc09689 | 23 | #include "bfd.h" |
abe529af | 24 | #include "bond.h" |
daff3353 | 25 | #include "bundle.h" |
abe529af BP |
26 | #include "byte-order.h" |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
75a75043 | 35 | #include "learn.h" |
abe529af | 36 | #include "mac-learning.h" |
816fd533 | 37 | #include "meta-flow.h" |
abe529af | 38 | #include "multipath.h" |
0a740f48 | 39 | #include "netdev-vport.h" |
abe529af BP |
40 | #include "netdev.h" |
41 | #include "netlink.h" | |
42 | #include "nx-match.h" | |
43 | #include "odp-util.h" | |
1ac7c9bd | 44 | #include "odp-execute.h" |
abe529af BP |
45 | #include "ofp-util.h" |
46 | #include "ofpbuf.h" | |
f25d0cf3 | 47 | #include "ofp-actions.h" |
31a19d69 | 48 | #include "ofp-parse.h" |
abe529af | 49 | #include "ofp-print.h" |
9d6ac44e | 50 | #include "ofproto-dpif-governor.h" |
29089a54 | 51 | #include "ofproto-dpif-ipfix.h" |
bae473fe | 52 | #include "ofproto-dpif-sflow.h" |
abe529af | 53 | #include "poll-loop.h" |
0d085684 | 54 | #include "simap.h" |
27022416 | 55 | #include "smap.h" |
abe529af | 56 | #include "timer.h" |
b9ad7294 | 57 | #include "tunnel.h" |
6c1491fb | 58 | #include "unaligned.h" |
abe529af BP |
59 | #include "unixctl.h" |
60 | #include "vlan-bitmap.h" | |
61 | #include "vlog.h" | |
62 | ||
63 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
64 | ||
abe529af | 65 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
66 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
67 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
68 | COVERAGE_DEFINE(facet_revalidate); |
69 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 70 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 71 | |
29901626 | 72 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 73 | * flow translation. */ |
1642690c | 74 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 75 | |
9cdaaebe BP |
76 | /* Number of implemented OpenFlow tables. */ |
77 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
78 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
79 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 80 | |
abe529af BP |
81 | struct ofport_dpif; |
82 | struct ofproto_dpif; | |
a088a1ff | 83 | struct flow_miss; |
ac35f9c8 | 84 | struct facet; |
abe529af BP |
85 | |
86 | struct rule_dpif { | |
87 | struct rule up; | |
88 | ||
abe529af BP |
89 | /* These statistics: |
90 | * | |
91 | * - Do include packets and bytes from facets that have been deleted or | |
92 | * whose own statistics have been folded into the rule. | |
93 | * | |
94 | * - Do include packets and bytes sent "by hand" that were accounted to | |
95 | * the rule without any facet being involved (this is a rare corner | |
96 | * case in rule_execute()). | |
97 | * | |
98 | * - Do not include packet or bytes that can be obtained from any facet's | |
99 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 100 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
101 | */ |
102 | uint64_t packet_count; /* Number of packets received. */ | |
103 | uint64_t byte_count; /* Number of bytes received. */ | |
104 | ||
54a9cbc9 BP |
105 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
106 | ||
abe529af BP |
107 | struct list facets; /* List of "struct facet"s. */ |
108 | }; | |
109 | ||
110 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
111 | { | |
112 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
113 | } | |
114 | ||
29901626 | 115 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
116 | const struct flow *); |
117 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
118 | const struct flow *, | |
119 | uint8_t table); | |
c376f9a3 IY |
120 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
121 | const struct flow *flow); | |
abe529af | 122 | |
7e741ffb | 123 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
112bc5f4 BP |
124 | static void rule_credit_stats(struct rule_dpif *, |
125 | const struct dpif_flow_stats *); | |
822d9414 | 126 | static tag_type rule_calculate_tag(const struct flow *, |
5cb7a798 | 127 | const struct minimask *, uint32_t basis); |
b0f7b9b5 BP |
128 | static void rule_invalidate(const struct rule_dpif *); |
129 | ||
abe529af BP |
130 | #define MAX_MIRRORS 32 |
131 | typedef uint32_t mirror_mask_t; | |
132 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
133 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
134 | struct ofmirror { | |
135 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
136 | size_t idx; /* In ofproto's "mirrors" array. */ | |
137 | void *aux; /* Key supplied by ofproto's client. */ | |
138 | char *name; /* Identifier for log messages. */ | |
139 | ||
140 | /* Selection criteria. */ | |
141 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
142 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
143 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
144 | ||
9ba15e2a | 145 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
146 | struct ofbundle *out; /* Output port or NULL. */ |
147 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 148 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
149 | |
150 | /* Counters. */ | |
151 | int64_t packet_count; /* Number of packets sent. */ | |
152 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
153 | }; |
154 | ||
155 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
156 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
157 | mirror_mask_t mirrors, | |
158 | uint64_t packets, uint64_t bytes); | |
abe529af | 159 | |
abe529af | 160 | struct ofbundle { |
abe529af | 161 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 162 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
163 | void *aux; /* Key supplied by ofproto's client. */ |
164 | char *name; /* Identifier for log messages. */ | |
165 | ||
166 | /* Configuration. */ | |
167 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 168 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
169 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
170 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
171 | * NULL if all VLANs are trunked. */ | |
172 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
173 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 174 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
175 | |
176 | /* Status. */ | |
9e1fd49b | 177 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
178 | |
179 | /* Port mirroring info. */ | |
180 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
181 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
182 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
183 | }; | |
184 | ||
185 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 186 | static void bundle_update(struct ofbundle *); |
abe529af BP |
187 | static void bundle_destroy(struct ofbundle *); |
188 | static void bundle_del_port(struct ofport_dpif *); | |
189 | static void bundle_run(struct ofbundle *); | |
190 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 191 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
192 | uint16_t in_port, bool warn, |
193 | struct ofport_dpif **in_ofportp); | |
abe529af | 194 | |
33158a18 JP |
195 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
196 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
197 | * when an input bundle is needed for validation (e.g., mirroring or | |
198 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
199 | * any 'port' structs, so care must be taken when dealing with it. */ | |
200 | static struct ofbundle ofpp_none_bundle = { | |
201 | .name = "OFPP_NONE", | |
202 | .vlan_mode = PORT_VLAN_TRUNK | |
203 | }; | |
204 | ||
21f7563c JP |
205 | static void stp_run(struct ofproto_dpif *ofproto); |
206 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
207 | static int set_stp_port(struct ofport *, |
208 | const struct ofproto_port_stp_settings *); | |
21f7563c | 209 | |
5da5ec37 BP |
210 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
211 | ||
bbafd73b EJ |
212 | struct xlate_ctx; |
213 | ||
214 | /* Initial values of fields of the packet that may be changed during | |
215 | * flow processing and needed later. */ | |
216 | struct initial_vals { | |
217 | /* This is the value of vlan_tci in the packet as actually received from | |
218 | * dpif. This is the same as the facet's flow.vlan_tci unless the packet | |
219 | * was received via a VLAN splinter. In that case, this value is 0 | |
220 | * (because the packet as actually received from the dpif had no 802.1Q | |
221 | * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter | |
222 | * represents. | |
223 | * | |
224 | * This member should be removed when the VLAN splinters feature is no | |
225 | * longer needed. */ | |
226 | ovs_be16 vlan_tci; | |
227 | }; | |
228 | ||
229 | struct xlate_out { | |
230 | tag_type tags; /* Tags associated with actions. */ | |
231 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
232 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
233 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
234 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
235 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
236 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ | |
237 | ||
238 | uint64_t odp_actions_stub[256 / 8]; | |
239 | struct ofpbuf odp_actions; | |
240 | }; | |
abe529af | 241 | |
bbafd73b | 242 | struct xlate_in { |
abe529af BP |
243 | struct ofproto_dpif *ofproto; |
244 | ||
245 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
246 | * this flow when actions change header fields. */ | |
247 | struct flow flow; | |
248 | ||
bbafd73b | 249 | struct initial_vals initial_vals; |
bd85dac1 | 250 | |
abe529af BP |
251 | /* The packet corresponding to 'flow', or a null pointer if we are |
252 | * revalidating without a packet to refer to. */ | |
253 | const struct ofpbuf *packet; | |
254 | ||
3de9590b BP |
255 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
256 | * actions update the flow table? | |
257 | * | |
258 | * We want to update these tables if we are actually processing a packet, | |
259 | * or if we are accounting for packets that the datapath has processed, but | |
260 | * not if we are just revalidating. */ | |
261 | bool may_learn; | |
75a75043 | 262 | |
bbafd73b | 263 | /* The rule initiating translation or NULL. */ |
18b2a258 | 264 | struct rule_dpif *rule; |
54834960 | 265 | |
bbafd73b EJ |
266 | /* The actions to translate. If 'rule' is not NULL, these may be NULL. */ |
267 | const struct ofpact *ofpacts; | |
268 | size_t ofpacts_len; | |
269 | ||
0e553d9c BP |
270 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
271 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
272 | * timeouts.) */ | |
273 | uint8_t tcp_flags; | |
274 | ||
112bc5f4 BP |
275 | /* If nonnull, flow translation calls this function just before executing a |
276 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
277 | * when the recursion depth is exceeded. | |
278 | * | |
279 | * 'rule' is the rule being submitted into. It will be null if the | |
280 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
281 | * | |
282 | * This is normally null so the client has to set it manually after | |
bbafd73b | 283 | * calling xlate_in_init(). */ |
f03a84b9 | 284 | void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule); |
112bc5f4 | 285 | |
479df176 BP |
286 | /* If nonnull, flow translation calls this function to report some |
287 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
288 | * dropped a packet. */ | |
f03a84b9 | 289 | void (*report_hook)(struct xlate_ctx *, const char *s); |
479df176 | 290 | |
112bc5f4 BP |
291 | /* If nonnull, flow translation credits the specified statistics to each |
292 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
293 | * |
294 | * This is normally null so the client has to set it manually after | |
bbafd73b | 295 | * calling xlate_in_init(). */ |
112bc5f4 | 296 | const struct dpif_flow_stats *resubmit_stats; |
bbafd73b | 297 | }; |
abe529af | 298 | |
bbafd73b EJ |
299 | /* Context used by xlate_actions() and its callees. */ |
300 | struct xlate_ctx { | |
301 | struct xlate_in *xin; | |
302 | struct xlate_out *xout; | |
abe529af | 303 | |
bbafd73b EJ |
304 | struct ofproto_dpif *ofproto; |
305 | ||
306 | /* Flow at the last commit. */ | |
307 | struct flow base_flow; | |
abe529af | 308 | |
bbafd73b EJ |
309 | /* Tunnel IP destination address as received. This is stored separately |
310 | * as the base_flow.tunnel is cleared on init to reflect the datapath | |
311 | * behavior. Used to make sure not to send tunneled output to ourselves, | |
312 | * which might lead to an infinite loop. This could happen easily | |
313 | * if a tunnel is marked as 'ip_remote=flow', and the flow does not | |
314 | * actually set the tun_dst field. */ | |
315 | ovs_be32 orig_tunnel_ip_dst; | |
316 | ||
317 | /* Stack for the push and pop actions. Each stack element is of type | |
318 | * "union mf_subvalue". */ | |
319 | union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; | |
320 | struct ofpbuf stack; | |
321 | ||
322 | /* The rule that we are currently translating, or NULL. */ | |
323 | struct rule_dpif *rule; | |
abe529af BP |
324 | |
325 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 326 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
deedf7e7 | 327 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 328 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 | 329 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
9b56fe13 | 330 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ |
6ff686f2 | 331 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ |
848e8809 | 332 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
333 | }; |
334 | ||
bbafd73b EJ |
335 | static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *, |
336 | const struct flow *, const struct initial_vals *, | |
337 | struct rule_dpif *, uint8_t tcp_flags, | |
338 | const struct ofpbuf *); | |
339 | ||
340 | static void xlate_out_uninit(struct xlate_out *); | |
341 | ||
342 | static void xlate_actions(struct xlate_in *, struct xlate_out *); | |
343 | ||
344 | static void xlate_actions_for_side_effects(struct xlate_in *); | |
14f94f9a | 345 | |
f03a84b9 | 346 | static void xlate_table_action(struct xlate_ctx *, uint16_t in_port, |
0a740f48 | 347 | uint8_t table_id, bool may_packet_in); |
abe529af | 348 | |
6a7e895f BP |
349 | static size_t put_userspace_action(const struct ofproto_dpif *, |
350 | struct ofpbuf *odp_actions, | |
351 | const struct flow *, | |
29089a54 RL |
352 | const union user_action_cookie *, |
353 | const size_t); | |
6a7e895f BP |
354 | |
355 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
356 | enum slow_path_reason, | |
357 | uint64_t *stub, size_t stub_size, | |
358 | const struct nlattr **actionsp, | |
359 | size_t *actions_lenp); | |
360 | ||
f03a84b9 | 361 | static void xlate_report(struct xlate_ctx *ctx, const char *s); |
479df176 | 362 | |
6a7e895f BP |
363 | /* A subfacet (see "struct subfacet" below) has three possible installation |
364 | * states: | |
365 | * | |
366 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
367 | * case just after the subfacet is created, just before the subfacet is | |
368 | * destroyed, or if the datapath returns an error when we try to install a | |
369 | * subfacet. | |
370 | * | |
371 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
372 | * | |
373 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
374 | * ofproto_dpif is installed in the datapath. | |
375 | */ | |
376 | enum subfacet_path { | |
377 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
378 | SF_FAST_PATH, /* Full actions are installed. */ | |
379 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
380 | }; | |
381 | ||
5f5fbd17 BP |
382 | /* A dpif flow and actions associated with a facet. |
383 | * | |
384 | * See also the large comment on struct facet. */ | |
385 | struct subfacet { | |
386 | /* Owners. */ | |
387 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
388 | struct list list_node; /* In struct facet's 'facets' list. */ | |
389 | struct facet *facet; /* Owning facet. */ | |
04d08d54 | 390 | struct dpif_backer *backer; /* Owning backer. */ |
5f5fbd17 | 391 | |
5f5fbd17 BP |
392 | enum odp_key_fitness key_fitness; |
393 | struct nlattr *key; | |
394 | int key_len; | |
395 | ||
396 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 397 | long long int created; /* Time created. */ |
5f5fbd17 BP |
398 | |
399 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
400 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
401 | ||
6a7e895f | 402 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 BP |
403 | }; |
404 | ||
1d85f9e5 JP |
405 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
406 | ||
a088a1ff | 407 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 408 | long long int now); |
04d08d54 | 409 | static struct subfacet *subfacet_find(struct dpif_backer *, |
acf60855 | 410 | const struct nlattr *key, size_t key_len, |
9566abf9 | 411 | uint32_t key_hash); |
5f5fbd17 BP |
412 | static void subfacet_destroy(struct subfacet *); |
413 | static void subfacet_destroy__(struct subfacet *); | |
04d08d54 | 414 | static void subfacet_destroy_batch(struct dpif_backer *, |
1d85f9e5 | 415 | struct subfacet **, int n); |
5f5fbd17 BP |
416 | static void subfacet_reset_dp_stats(struct subfacet *, |
417 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
418 | static void subfacet_update_stats(struct subfacet *, |
419 | const struct dpif_flow_stats *); | |
5f5fbd17 | 420 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
421 | const struct ofpbuf *odp_actions, |
422 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
423 | static void subfacet_uninstall(struct subfacet *); |
424 | ||
b0f7b9b5 BP |
425 | /* An exact-match instantiation of an OpenFlow flow. |
426 | * | |
427 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
428 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
429 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
430 | * the facet. When the kernel module (or other dpif implementation) and Open | |
431 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
432 | * one subfacet per facet. If the dpif implementation supports more-specific | |
433 | * flow matching than userspace, however, a facet can have more than one | |
434 | * subfacet, each of which corresponds to some distinction in flow that | |
435 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
436 | * |
437 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
438 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 439 | struct facet { |
b0f7b9b5 BP |
440 | /* Owners. */ |
441 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
442 | struct list list_node; /* In owning rule's 'facets' list. */ | |
443 | struct rule_dpif *rule; /* Owning rule. */ | |
444 | ||
445 | /* Owned data. */ | |
446 | struct list subfacets; | |
abe529af BP |
447 | long long int used; /* Time last used; time created if not used. */ |
448 | ||
b0f7b9b5 BP |
449 | /* Key. */ |
450 | struct flow flow; | |
451 | ||
abe529af BP |
452 | /* These statistics: |
453 | * | |
454 | * - Do include packets and bytes sent "by hand", e.g. with | |
455 | * dpif_execute(). | |
456 | * | |
457 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 458 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 459 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
460 | * |
461 | * - Do not include packets or bytes that can be obtained from the | |
462 | * datapath for any existing subfacet. | |
abe529af BP |
463 | */ |
464 | uint64_t packet_count; /* Number of packets received. */ | |
465 | uint64_t byte_count; /* Number of bytes received. */ | |
466 | ||
b0f7b9b5 | 467 | /* Resubmit statistics. */ |
9d24de3b JP |
468 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
469 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
470 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 471 | |
b0f7b9b5 | 472 | /* Accounting. */ |
907a4c5e | 473 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 474 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 475 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 476 | |
bbafd73b | 477 | struct xlate_out xout; |
4dff9097 EJ |
478 | |
479 | /* Initial values of the packet that may be needed later. */ | |
480 | struct initial_vals initial_vals; | |
481 | ||
26cd7e34 BP |
482 | /* Storage for a single subfacet, to reduce malloc() time and space |
483 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
484 | * case has exactly one subfacet. However, 'one_subfacet' may not |
485 | * always be valid, since it could have been removed after newer | |
486 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 487 | struct subfacet one_subfacet; |
6cf474d7 EJ |
488 | |
489 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
490 | }; |
491 | ||
4dff9097 | 492 | static struct facet *facet_create(const struct flow_miss *, uint32_t hash); |
15baa734 | 493 | static void facet_remove(struct facet *); |
abe529af BP |
494 | static void facet_free(struct facet *); |
495 | ||
2b459b83 BP |
496 | static struct facet *facet_find(struct ofproto_dpif *, |
497 | const struct flow *, uint32_t hash); | |
abe529af | 498 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 499 | const struct flow *, uint32_t hash); |
5bf64ade | 500 | static bool facet_revalidate(struct facet *); |
6814e51f | 501 | static bool facet_check_consistency(struct facet *); |
abe529af | 502 | |
15baa734 | 503 | static void facet_flush_stats(struct facet *); |
abe529af | 504 | |
bbb5d219 | 505 | static void facet_reset_counters(struct facet *); |
9dfb1f78 | 506 | static void facet_push_stats(struct facet *, bool may_learn); |
3de9590b BP |
507 | static void facet_learn(struct facet *); |
508 | static void facet_account(struct facet *); | |
8844e035 | 509 | static void push_all_stats(void); |
abe529af BP |
510 | |
511 | static bool facet_is_controller_flow(struct facet *); | |
512 | ||
abe529af | 513 | struct ofport_dpif { |
acf60855 | 514 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ |
abe529af BP |
515 | struct ofport up; |
516 | ||
517 | uint32_t odp_port; | |
518 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
519 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
520 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
ccc09689 | 521 | struct bfd *bfd; /* BFD, if any. */ |
abe529af | 522 | tag_type tag; /* Tag associated with this port. */ |
015e08bc | 523 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 524 | long long int carrier_seq; /* Carrier status changes. */ |
b9ad7294 | 525 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ |
21f7563c | 526 | |
52a90c29 | 527 | /* Spanning tree. */ |
21f7563c JP |
528 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
529 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
530 | long long int stp_state_entered; | |
8b36f51e EJ |
531 | |
532 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
533 | |
534 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
535 | * | |
536 | * This is deprecated. It is only for compatibility with broken device | |
537 | * drivers in old versions of Linux that do not properly support VLANs when | |
538 | * VLAN devices are not used. When broken device drivers are no longer in | |
539 | * widespread use, we will delete these interfaces. */ | |
540 | uint16_t realdev_ofp_port; | |
541 | int vlandev_vid; | |
8b36f51e EJ |
542 | }; |
543 | ||
544 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
545 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
546 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
547 | struct priority_to_dscp { | |
548 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
549 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
550 | ||
551 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
552 | }; |
553 | ||
52a90c29 BP |
554 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
555 | * | |
556 | * This is deprecated. It is only for compatibility with broken device drivers | |
557 | * in old versions of Linux that do not properly support VLANs when VLAN | |
558 | * devices are not used. When broken device drivers are no longer in | |
559 | * widespread use, we will delete these interfaces. */ | |
560 | struct vlan_splinter { | |
561 | struct hmap_node realdev_vid_node; | |
562 | struct hmap_node vlandev_node; | |
563 | uint16_t realdev_ofp_port; | |
564 | uint16_t vlandev_ofp_port; | |
565 | int vid; | |
566 | }; | |
567 | ||
deea1200 AW |
568 | static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, |
569 | uint16_t realdev_ofp_port, | |
570 | ovs_be16 vlan_tci); | |
b98d8985 | 571 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
572 | static void vsp_remove(struct ofport_dpif *); |
573 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
574 | ||
e1b1d06a JP |
575 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
576 | uint16_t ofp_port); | |
577 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
578 | uint32_t odp_port); | |
579 | ||
abe529af BP |
580 | static struct ofport_dpif * |
581 | ofport_dpif_cast(const struct ofport *ofport) | |
582 | { | |
abe529af BP |
583 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
584 | } | |
585 | ||
586 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 587 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 588 | static void port_wait(struct ofport_dpif *); |
8aee94b6 | 589 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 590 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 591 | static void ofport_clear_priorities(struct ofport_dpif *); |
8fa4d1d0 | 592 | static void run_fast_rl(void); |
abe529af | 593 | |
7ee20df1 BP |
594 | struct dpif_completion { |
595 | struct list list_node; | |
596 | struct ofoperation *op; | |
597 | }; | |
598 | ||
54a9cbc9 BP |
599 | /* Extra information about a classifier table. |
600 | * Currently used just for optimized flow revalidation. */ | |
601 | struct table_dpif { | |
602 | /* If either of these is nonnull, then this table has a form that allows | |
603 | * flows to be tagged to avoid revalidating most flows for the most common | |
604 | * kinds of flow table changes. */ | |
605 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
606 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
607 | uint32_t basis; /* Keeps each table's tags separate. */ | |
608 | }; | |
609 | ||
3c4a309c BP |
610 | /* Reasons that we might need to revalidate every facet, and corresponding |
611 | * coverage counters. | |
612 | * | |
613 | * A value of 0 means that there is no need to revalidate. | |
614 | * | |
615 | * It would be nice to have some cleaner way to integrate with coverage | |
616 | * counters, but with only a few reasons I guess this is good enough for | |
617 | * now. */ | |
618 | enum revalidate_reason { | |
619 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
620 | REV_STP, /* Spanning tree protocol port status change. */ | |
621 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
622 | REV_FLOW_TABLE, /* Flow table changed. */ | |
623 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
624 | }; | |
625 | COVERAGE_DEFINE(rev_reconfigure); | |
626 | COVERAGE_DEFINE(rev_stp); | |
627 | COVERAGE_DEFINE(rev_port_toggled); | |
628 | COVERAGE_DEFINE(rev_flow_table); | |
629 | COVERAGE_DEFINE(rev_inconsistency); | |
630 | ||
8f73d537 EJ |
631 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
632 | * These are datapath flows which have no associated ofproto, if they did we | |
633 | * would use facets. */ | |
634 | struct drop_key { | |
635 | struct hmap_node hmap_node; | |
636 | struct nlattr *key; | |
637 | size_t key_len; | |
638 | }; | |
639 | ||
dc54ef36 EJ |
640 | struct avg_subfacet_rates { |
641 | double add_rate; /* Moving average of new flows created per minute. */ | |
642 | double del_rate; /* Moving average of flows deleted per minute. */ | |
643 | }; | |
644 | ||
acf60855 JP |
645 | /* All datapaths of a given type share a single dpif backer instance. */ |
646 | struct dpif_backer { | |
647 | char *type; | |
648 | int refcount; | |
649 | struct dpif *dpif; | |
650 | struct timer next_expiration; | |
651 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
2cc3c58e | 652 | |
7d82ab2e | 653 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ |
b9ad7294 | 654 | |
2cc3c58e EJ |
655 | /* Facet revalidation flags applying to facets which use this backer. */ |
656 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
657 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
8f73d537 EJ |
658 | |
659 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
40358701 | 660 | bool recv_set_enable; /* Enables or disables receiving packets. */ |
dc54ef36 | 661 | |
04d08d54 EJ |
662 | struct hmap subfacets; |
663 | struct governor *governor; | |
664 | ||
dc54ef36 EJ |
665 | /* Subfacet statistics. |
666 | * | |
667 | * These keep track of the total number of subfacets added and deleted and | |
668 | * flow life span. They are useful for computing the flow rates stats | |
669 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
670 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
671 | * performance in new situations. */ | |
672 | long long int created; /* Time when it is created. */ | |
673 | unsigned max_n_subfacet; /* Maximum number of flows */ | |
674 | unsigned avg_n_subfacet; /* Average number of flows. */ | |
675 | long long int avg_subfacet_life; /* Average life span of subfacets. */ | |
676 | ||
677 | /* The average number of subfacets... */ | |
678 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
679 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
680 | struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */ | |
681 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
682 | ||
683 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
684 | unsigned subfacet_add_count; | |
685 | unsigned subfacet_del_count; | |
686 | ||
687 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
688 | unsigned long long int total_subfacet_add_count; | |
689 | unsigned long long int total_subfacet_del_count; | |
acf60855 JP |
690 | }; |
691 | ||
692 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
693 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
694 | ||
8f73d537 | 695 | static void drop_key_clear(struct dpif_backer *); |
acf60855 JP |
696 | static struct ofport_dpif * |
697 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
dc54ef36 | 698 | static void update_moving_averages(struct dpif_backer *backer); |
735d7efb | 699 | |
abe529af | 700 | struct ofproto_dpif { |
b44a10b7 | 701 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af | 702 | struct ofproto up; |
acf60855 | 703 | struct dpif_backer *backer; |
abe529af | 704 | |
c57b2226 BP |
705 | /* Special OpenFlow rules. */ |
706 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
707 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
7fd51d39 | 708 | struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ |
c57b2226 | 709 | |
abe529af BP |
710 | /* Bridging. */ |
711 | struct netflow *netflow; | |
bae473fe | 712 | struct dpif_sflow *sflow; |
29089a54 | 713 | struct dpif_ipfix *ipfix; |
abe529af BP |
714 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
715 | struct mac_learning *ml; | |
716 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 717 | bool has_mirrors; |
abe529af BP |
718 | bool has_bonded_bundles; |
719 | ||
abe529af BP |
720 | /* Facets. */ |
721 | struct hmap facets; | |
4d2a0f39 | 722 | long long int consistency_rl; |
54a9cbc9 BP |
723 | |
724 | /* Revalidation. */ | |
725 | struct table_dpif tables[N_TABLES]; | |
7ee20df1 BP |
726 | |
727 | /* Support for debugging async flow mods. */ | |
728 | struct list completions; | |
daff3353 EJ |
729 | |
730 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
731 | struct netdev_stats stats; /* To account packets generated and consumed in |
732 | * userspace. */ | |
21f7563c JP |
733 | |
734 | /* Spanning tree. */ | |
735 | struct stp *stp; | |
736 | long long int stp_last_tick; | |
52a90c29 BP |
737 | |
738 | /* VLAN splinters. */ | |
739 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
740 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
e1b1d06a | 741 | |
acf60855 | 742 | /* Ports. */ |
0a740f48 EJ |
743 | struct sset ports; /* Set of standard port names. */ |
744 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
acf60855 JP |
745 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ |
746 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
735d7efb AZ |
747 | |
748 | /* Per ofproto's dpif stats. */ | |
749 | uint64_t n_hit; | |
750 | uint64_t n_missed; | |
abe529af BP |
751 | }; |
752 | ||
7ee20df1 BP |
753 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
754 | * for debugging the asynchronous flow_mod implementation.) */ | |
755 | static bool clogged; | |
756 | ||
b44a10b7 BP |
757 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
758 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
759 | ||
abe529af BP |
760 | static void ofproto_dpif_unixctl_init(void); |
761 | ||
762 | static struct ofproto_dpif * | |
763 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
764 | { | |
cb22974d | 765 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
766 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); |
767 | } | |
768 | ||
4acbc98d | 769 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 770 | uint16_t ofp_port); |
4acbc98d | 771 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 772 | uint32_t odp_port); |
6a6455e5 | 773 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
14f94f9a JP |
774 | const struct ofpbuf *, |
775 | const struct initial_vals *, struct ds *); | |
abe529af BP |
776 | |
777 | /* Packet processing. */ | |
778 | static void update_learning_table(struct ofproto_dpif *, | |
779 | const struct flow *, int vlan, | |
780 | struct ofbundle *); | |
501f8d1f BP |
781 | /* Upcalls. */ |
782 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 783 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
784 | |
785 | /* Flow expiration. */ | |
acf60855 | 786 | static int expire(struct dpif_backer *); |
abe529af | 787 | |
6fca1ffb BP |
788 | /* NetFlow. */ |
789 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
790 | ||
abe529af | 791 | /* Utilities. */ |
52a90c29 | 792 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
793 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
794 | struct ofpbuf *odp_actions, | |
795 | const struct flow *, uint32_t odp_port); | |
29089a54 RL |
796 | static void compose_ipfix_action(const struct ofproto_dpif *, |
797 | struct ofpbuf *odp_actions, | |
798 | const struct flow *); | |
f03a84b9 | 799 | static void add_mirror_actions(struct xlate_ctx *ctx, |
c06bba01 | 800 | const struct flow *flow); |
abe529af BP |
801 | /* Global variables. */ |
802 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
803 | |
804 | /* Initial mappings of port to bridge mappings. */ | |
805 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
abe529af BP |
806 | \f |
807 | /* Factory functions. */ | |
808 | ||
b0408fca | 809 | static void |
acf60855 | 810 | init(const struct shash *iface_hints) |
b0408fca | 811 | { |
acf60855 JP |
812 | struct shash_node *node; |
813 | ||
814 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
815 | SHASH_FOR_EACH(node, iface_hints) { | |
816 | const struct iface_hint *orig_hint = node->data; | |
817 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
818 | ||
819 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
820 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
821 | new_hint->ofp_port = orig_hint->ofp_port; | |
822 | ||
823 | shash_add(&init_ofp_ports, node->name, new_hint); | |
824 | } | |
b0408fca JP |
825 | } |
826 | ||
abe529af BP |
827 | static void |
828 | enumerate_types(struct sset *types) | |
829 | { | |
830 | dp_enumerate_types(types); | |
831 | } | |
832 | ||
833 | static int | |
834 | enumerate_names(const char *type, struct sset *names) | |
835 | { | |
acf60855 JP |
836 | struct ofproto_dpif *ofproto; |
837 | ||
838 | sset_clear(names); | |
839 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
840 | if (strcmp(type, ofproto->up.type)) { | |
841 | continue; | |
842 | } | |
843 | sset_add(names, ofproto->up.name); | |
844 | } | |
845 | ||
846 | return 0; | |
abe529af BP |
847 | } |
848 | ||
849 | static int | |
850 | del(const char *type, const char *name) | |
851 | { | |
852 | struct dpif *dpif; | |
853 | int error; | |
854 | ||
855 | error = dpif_open(name, type, &dpif); | |
856 | if (!error) { | |
857 | error = dpif_delete(dpif); | |
858 | dpif_close(dpif); | |
859 | } | |
860 | return error; | |
861 | } | |
862 | \f | |
0aeaabc8 JP |
863 | static const char * |
864 | port_open_type(const char *datapath_type, const char *port_type) | |
865 | { | |
866 | return dpif_port_open_type(datapath_type, port_type); | |
867 | } | |
868 | ||
acf60855 JP |
869 | /* Type functions. */ |
870 | ||
476cb42a BP |
871 | static struct ofproto_dpif * |
872 | lookup_ofproto_dpif_by_port_name(const char *name) | |
873 | { | |
874 | struct ofproto_dpif *ofproto; | |
875 | ||
876 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
877 | if (sset_contains(&ofproto->ports, name)) { | |
878 | return ofproto; | |
879 | } | |
880 | } | |
881 | ||
882 | return NULL; | |
883 | } | |
884 | ||
acf60855 JP |
885 | static int |
886 | type_run(const char *type) | |
887 | { | |
e20b5746 | 888 | static long long int push_timer = LLONG_MIN; |
acf60855 JP |
889 | struct dpif_backer *backer; |
890 | char *devname; | |
891 | int error; | |
892 | ||
893 | backer = shash_find_data(&all_dpif_backers, type); | |
894 | if (!backer) { | |
895 | /* This is not necessarily a problem, since backers are only | |
896 | * created on demand. */ | |
897 | return 0; | |
898 | } | |
899 | ||
900 | dpif_run(backer->dpif); | |
901 | ||
e20b5746 EJ |
902 | /* The most natural place to push facet statistics is when they're pulled |
903 | * from the datapath. However, when there are many flows in the datapath, | |
904 | * this expensive operation can occur so frequently, that it reduces our | |
905 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
906 | * here instead. */ | |
907 | if (time_msec() > push_timer) { | |
908 | push_timer = time_msec() + 2000; | |
909 | push_all_stats(); | |
910 | } | |
911 | ||
40358701 GS |
912 | /* If vswitchd started with other_config:flow_restore_wait set as "true", |
913 | * and the configuration has now changed to "false", enable receiving | |
914 | * packets from the datapath. */ | |
915 | if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { | |
916 | backer->recv_set_enable = true; | |
917 | ||
918 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); | |
919 | if (error) { | |
920 | VLOG_ERR("Failed to enable receiving packets in dpif."); | |
921 | return error; | |
922 | } | |
923 | dpif_flow_flush(backer->dpif); | |
924 | backer->need_revalidate = REV_RECONFIGURE; | |
925 | } | |
926 | ||
2cc3c58e EJ |
927 | if (backer->need_revalidate |
928 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
929 | struct tag_set revalidate_set = backer->revalidate_set; | |
930 | bool need_revalidate = backer->need_revalidate; | |
931 | struct ofproto_dpif *ofproto; | |
a614d823 KM |
932 | struct simap_node *node; |
933 | struct simap tmp_backers; | |
934 | ||
935 | /* Handle tunnel garbage collection. */ | |
936 | simap_init(&tmp_backers); | |
937 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
938 | ||
939 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
940 | struct ofport_dpif *iter; | |
941 | ||
942 | if (backer != ofproto->backer) { | |
943 | continue; | |
944 | } | |
945 | ||
946 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
3aa30359 | 947 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
a614d823 KM |
948 | const char *dp_port; |
949 | ||
950 | if (!iter->tnl_port) { | |
951 | continue; | |
952 | } | |
953 | ||
3aa30359 BP |
954 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev, |
955 | namebuf, sizeof namebuf); | |
a614d823 KM |
956 | node = simap_find(&tmp_backers, dp_port); |
957 | if (node) { | |
958 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
959 | simap_delete(&tmp_backers, node); | |
960 | node = simap_find(&backer->tnl_backers, dp_port); | |
961 | } else { | |
962 | node = simap_find(&backer->tnl_backers, dp_port); | |
963 | if (!node) { | |
964 | uint32_t odp_port = UINT32_MAX; | |
965 | ||
966 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
967 | &odp_port)) { | |
968 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
969 | node = simap_find(&backer->tnl_backers, dp_port); | |
970 | } | |
971 | } | |
972 | } | |
973 | ||
974 | iter->odp_port = node ? node->data : OVSP_NONE; | |
975 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
976 | &iter->tnl_port)) { | |
977 | backer->need_revalidate = REV_RECONFIGURE; | |
978 | } | |
979 | } | |
980 | } | |
981 | ||
982 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
983 | dpif_port_del(backer->dpif, node->data); | |
984 | } | |
985 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
986 | |
987 | switch (backer->need_revalidate) { | |
988 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
989 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
990 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
991 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
992 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
993 | } | |
994 | ||
8f73d537 EJ |
995 | if (backer->need_revalidate) { |
996 | /* Clear the drop_keys in case we should now be accepting some | |
997 | * formerly dropped flows. */ | |
998 | drop_key_clear(backer); | |
999 | } | |
1000 | ||
f728af2e BP |
1001 | /* Clear the revalidation flags. */ |
1002 | tag_set_init(&backer->revalidate_set); | |
1003 | backer->need_revalidate = 0; | |
1004 | ||
2cc3c58e | 1005 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 1006 | struct facet *facet, *next; |
2cc3c58e EJ |
1007 | |
1008 | if (ofproto->backer != backer) { | |
1009 | continue; | |
1010 | } | |
1011 | ||
f231418e | 1012 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { |
2cc3c58e | 1013 | if (need_revalidate |
bbafd73b | 1014 | || tag_set_intersects(&revalidate_set, facet->xout.tags)) { |
2cc3c58e | 1015 | facet_revalidate(facet); |
8fa4d1d0 | 1016 | run_fast_rl(); |
2cc3c58e EJ |
1017 | } |
1018 | } | |
1019 | } | |
2cc3c58e EJ |
1020 | } |
1021 | ||
40358701 GS |
1022 | if (!backer->recv_set_enable) { |
1023 | /* Wake up before a max of 1000ms. */ | |
1024 | timer_set_duration(&backer->next_expiration, 1000); | |
1025 | } else if (timer_expired(&backer->next_expiration)) { | |
acf60855 JP |
1026 | int delay = expire(backer); |
1027 | timer_set_duration(&backer->next_expiration, delay); | |
1028 | } | |
1029 | ||
1030 | /* Check for port changes in the dpif. */ | |
1031 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
476cb42a | 1032 | struct ofproto_dpif *ofproto; |
acf60855 JP |
1033 | struct dpif_port port; |
1034 | ||
1035 | /* Don't report on the datapath's device. */ | |
1036 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
c83b89ab | 1037 | goto next; |
acf60855 JP |
1038 | } |
1039 | ||
b9ad7294 EJ |
1040 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, |
1041 | &all_ofproto_dpifs) { | |
7d82ab2e | 1042 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { |
b9ad7294 EJ |
1043 | goto next; |
1044 | } | |
1045 | } | |
1046 | ||
476cb42a | 1047 | ofproto = lookup_ofproto_dpif_by_port_name(devname); |
acf60855 JP |
1048 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { |
1049 | /* The port was removed. If we know the datapath, | |
1050 | * report it through poll_set(). If we don't, it may be | |
1051 | * notifying us of a removal we initiated, so ignore it. | |
1052 | * If there's a pending ENOBUFS, let it stand, since | |
1053 | * everything will be reevaluated. */ | |
1054 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
1055 | sset_add(&ofproto->port_poll_set, devname); | |
1056 | ofproto->port_poll_errno = 0; | |
1057 | } | |
acf60855 JP |
1058 | } else if (!ofproto) { |
1059 | /* The port was added, but we don't know with which | |
1060 | * ofproto we should associate it. Delete it. */ | |
1061 | dpif_port_del(backer->dpif, port.port_no); | |
1062 | } | |
5b5e6a4c | 1063 | dpif_port_destroy(&port); |
acf60855 | 1064 | |
c83b89ab | 1065 | next: |
acf60855 JP |
1066 | free(devname); |
1067 | } | |
1068 | ||
1069 | if (error != EAGAIN) { | |
1070 | struct ofproto_dpif *ofproto; | |
1071 | ||
1072 | /* There was some sort of error, so propagate it to all | |
1073 | * ofprotos that use this backer. */ | |
1074 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1075 | &all_ofproto_dpifs) { | |
1076 | if (ofproto->backer == backer) { | |
1077 | sset_clear(&ofproto->port_poll_set); | |
1078 | ofproto->port_poll_errno = error; | |
1079 | } | |
1080 | } | |
1081 | } | |
1082 | ||
04d08d54 EJ |
1083 | if (backer->governor) { |
1084 | size_t n_subfacets; | |
1085 | ||
1086 | governor_run(backer->governor); | |
1087 | ||
1088 | /* If the governor has shrunk to its minimum size and the number of | |
1089 | * subfacets has dwindled, then drop the governor entirely. | |
1090 | * | |
1091 | * For hysteresis, the number of subfacets to drop the governor is | |
1092 | * smaller than the number needed to trigger its creation. */ | |
1093 | n_subfacets = hmap_count(&backer->subfacets); | |
1094 | if (n_subfacets * 4 < flow_eviction_threshold | |
1095 | && governor_is_idle(backer->governor)) { | |
1096 | governor_destroy(backer->governor); | |
1097 | backer->governor = NULL; | |
1098 | } | |
1099 | } | |
1100 | ||
acf60855 JP |
1101 | return 0; |
1102 | } | |
1103 | ||
1104 | static int | |
8fa4d1d0 | 1105 | dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) |
acf60855 | 1106 | { |
acf60855 JP |
1107 | unsigned int work; |
1108 | ||
40358701 GS |
1109 | /* If recv_set_enable is false, we should not handle upcalls. */ |
1110 | if (!backer->recv_set_enable) { | |
1111 | return 0; | |
1112 | } | |
1113 | ||
acf60855 JP |
1114 | /* Handle one or more batches of upcalls, until there's nothing left to do |
1115 | * or until we do a fixed total amount of work. | |
1116 | * | |
1117 | * We do work in batches because it can be much cheaper to set up a number | |
1118 | * of flows and fire off their patches all at once. We do multiple batches | |
1119 | * because in some cases handling a packet can cause another packet to be | |
1120 | * queued almost immediately as part of the return flow. Both | |
1121 | * optimizations can make major improvements on some benchmarks and | |
1122 | * presumably for real traffic as well. */ | |
1123 | work = 0; | |
8fa4d1d0 EJ |
1124 | while (work < max_batch) { |
1125 | int retval = handle_upcalls(backer, max_batch - work); | |
acf60855 JP |
1126 | if (retval <= 0) { |
1127 | return -retval; | |
1128 | } | |
1129 | work += retval; | |
1130 | } | |
1131 | ||
1132 | return 0; | |
1133 | } | |
1134 | ||
8fa4d1d0 EJ |
1135 | static int |
1136 | type_run_fast(const char *type) | |
1137 | { | |
1138 | struct dpif_backer *backer; | |
1139 | ||
1140 | backer = shash_find_data(&all_dpif_backers, type); | |
1141 | if (!backer) { | |
1142 | /* This is not necessarily a problem, since backers are only | |
1143 | * created on demand. */ | |
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1148 | } | |
1149 | ||
1150 | static void | |
1151 | run_fast_rl(void) | |
1152 | { | |
1153 | static long long int port_rl = LLONG_MIN; | |
1154 | static unsigned int backer_rl = 0; | |
1155 | ||
1156 | if (time_msec() >= port_rl) { | |
1157 | struct ofproto_dpif *ofproto; | |
1158 | struct ofport_dpif *ofport; | |
1159 | ||
1160 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1161 | ||
1162 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1163 | port_run_fast(ofport); | |
1164 | } | |
1165 | } | |
1166 | port_rl = time_msec() + 200; | |
1167 | } | |
1168 | ||
1169 | /* XXX: We have to be careful not to do too much work in this function. If | |
1170 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1171 | * performance improves signifcantly, but at a cost. It's possible for the | |
1172 | * number of flows in the datapath to increase without bound, and for poll | |
1173 | * loops to take 10s of seconds. The correct solution to this problem, | |
1174 | * long term, is to separate flow miss handling into it's own thread so it | |
1175 | * isn't affected by revalidations, and expirations. Until then, this is | |
1176 | * the best we can do. */ | |
1177 | if (++backer_rl >= 10) { | |
1178 | struct shash_node *node; | |
1179 | ||
1180 | backer_rl = 0; | |
1181 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1182 | dpif_backer_run_fast(node->data, 1); | |
1183 | } | |
1184 | } | |
1185 | } | |
1186 | ||
acf60855 JP |
1187 | static void |
1188 | type_wait(const char *type) | |
1189 | { | |
1190 | struct dpif_backer *backer; | |
1191 | ||
1192 | backer = shash_find_data(&all_dpif_backers, type); | |
1193 | if (!backer) { | |
1194 | /* This is not necessarily a problem, since backers are only | |
1195 | * created on demand. */ | |
1196 | return; | |
1197 | } | |
1198 | ||
04d08d54 EJ |
1199 | if (backer->governor) { |
1200 | governor_wait(backer->governor); | |
1201 | } | |
1202 | ||
acf60855 JP |
1203 | timer_wait(&backer->next_expiration); |
1204 | } | |
1205 | \f | |
abe529af BP |
1206 | /* Basic life-cycle. */ |
1207 | ||
c57b2226 BP |
1208 | static int add_internal_flows(struct ofproto_dpif *); |
1209 | ||
abe529af BP |
1210 | static struct ofproto * |
1211 | alloc(void) | |
1212 | { | |
1213 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1214 | return &ofproto->up; | |
1215 | } | |
1216 | ||
1217 | static void | |
1218 | dealloc(struct ofproto *ofproto_) | |
1219 | { | |
1220 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1221 | free(ofproto); | |
1222 | } | |
1223 | ||
acf60855 JP |
1224 | static void |
1225 | close_dpif_backer(struct dpif_backer *backer) | |
1226 | { | |
1227 | struct shash_node *node; | |
1228 | ||
cb22974d | 1229 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1230 | |
1231 | if (--backer->refcount) { | |
1232 | return; | |
1233 | } | |
1234 | ||
8f73d537 EJ |
1235 | drop_key_clear(backer); |
1236 | hmap_destroy(&backer->drop_keys); | |
1237 | ||
7d82ab2e | 1238 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1239 | hmap_destroy(&backer->odp_to_ofport_map); |
1240 | node = shash_find(&all_dpif_backers, backer->type); | |
1241 | free(backer->type); | |
1242 | shash_delete(&all_dpif_backers, node); | |
1243 | dpif_close(backer->dpif); | |
1244 | ||
04d08d54 EJ |
1245 | ovs_assert(hmap_is_empty(&backer->subfacets)); |
1246 | hmap_destroy(&backer->subfacets); | |
1247 | governor_destroy(backer->governor); | |
1248 | ||
acf60855 JP |
1249 | free(backer); |
1250 | } | |
1251 | ||
1252 | /* Datapath port slated for removal from datapath. */ | |
1253 | struct odp_garbage { | |
1254 | struct list list_node; | |
1255 | uint32_t odp_port; | |
1256 | }; | |
1257 | ||
1258 | static int | |
1259 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1260 | { | |
1261 | struct dpif_backer *backer; | |
1262 | struct dpif_port_dump port_dump; | |
1263 | struct dpif_port port; | |
1264 | struct shash_node *node; | |
1265 | struct list garbage_list; | |
1266 | struct odp_garbage *garbage, *next; | |
1267 | struct sset names; | |
1268 | char *backer_name; | |
1269 | const char *name; | |
1270 | int error; | |
1271 | ||
1272 | backer = shash_find_data(&all_dpif_backers, type); | |
1273 | if (backer) { | |
1274 | backer->refcount++; | |
1275 | *backerp = backer; | |
1276 | return 0; | |
1277 | } | |
1278 | ||
1279 | backer_name = xasprintf("ovs-%s", type); | |
1280 | ||
1281 | /* Remove any existing datapaths, since we assume we're the only | |
1282 | * userspace controlling the datapath. */ | |
1283 | sset_init(&names); | |
1284 | dp_enumerate_names(type, &names); | |
1285 | SSET_FOR_EACH(name, &names) { | |
1286 | struct dpif *old_dpif; | |
1287 | ||
1288 | /* Don't remove our backer if it exists. */ | |
1289 | if (!strcmp(name, backer_name)) { | |
1290 | continue; | |
1291 | } | |
1292 | ||
1293 | if (dpif_open(name, type, &old_dpif)) { | |
1294 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1295 | } else { | |
1296 | dpif_delete(old_dpif); | |
1297 | dpif_close(old_dpif); | |
1298 | } | |
1299 | } | |
1300 | sset_destroy(&names); | |
1301 | ||
1302 | backer = xmalloc(sizeof *backer); | |
1303 | ||
1304 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1305 | free(backer_name); | |
1306 | if (error) { | |
1307 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1308 | strerror(error)); | |
4c1b1289 | 1309 | free(backer); |
acf60855 JP |
1310 | return error; |
1311 | } | |
1312 | ||
1313 | backer->type = xstrdup(type); | |
04d08d54 | 1314 | backer->governor = NULL; |
acf60855 JP |
1315 | backer->refcount = 1; |
1316 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1317 | hmap_init(&backer->drop_keys); |
04d08d54 | 1318 | hmap_init(&backer->subfacets); |
acf60855 | 1319 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1320 | backer->need_revalidate = 0; |
7d82ab2e | 1321 | simap_init(&backer->tnl_backers); |
2cc3c58e | 1322 | tag_set_init(&backer->revalidate_set); |
40358701 | 1323 | backer->recv_set_enable = !ofproto_get_flow_restore_wait(); |
acf60855 JP |
1324 | *backerp = backer; |
1325 | ||
40358701 GS |
1326 | if (backer->recv_set_enable) { |
1327 | dpif_flow_flush(backer->dpif); | |
1328 | } | |
acf60855 JP |
1329 | |
1330 | /* Loop through the ports already on the datapath and remove any | |
1331 | * that we don't need anymore. */ | |
1332 | list_init(&garbage_list); | |
1333 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1334 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1335 | node = shash_find(&init_ofp_ports, port.name); | |
1336 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1337 | garbage = xmalloc(sizeof *garbage); | |
1338 | garbage->odp_port = port.port_no; | |
1339 | list_push_front(&garbage_list, &garbage->list_node); | |
1340 | } | |
1341 | } | |
1342 | dpif_port_dump_done(&port_dump); | |
1343 | ||
1344 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1345 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1346 | list_remove(&garbage->list_node); | |
1347 | free(garbage); | |
1348 | } | |
1349 | ||
1350 | shash_add(&all_dpif_backers, type, backer); | |
1351 | ||
40358701 | 1352 | error = dpif_recv_set(backer->dpif, backer->recv_set_enable); |
acf60855 JP |
1353 | if (error) { |
1354 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1355 | type, strerror(error)); | |
1356 | close_dpif_backer(backer); | |
1357 | return error; | |
1358 | } | |
1359 | ||
dc54ef36 EJ |
1360 | backer->max_n_subfacet = 0; |
1361 | backer->created = time_msec(); | |
1362 | backer->last_minute = backer->created; | |
1363 | memset(&backer->hourly, 0, sizeof backer->hourly); | |
1364 | memset(&backer->daily, 0, sizeof backer->daily); | |
1365 | memset(&backer->lifetime, 0, sizeof backer->lifetime); | |
1366 | backer->subfacet_add_count = 0; | |
1367 | backer->subfacet_del_count = 0; | |
1368 | backer->total_subfacet_add_count = 0; | |
1369 | backer->total_subfacet_del_count = 0; | |
1370 | backer->avg_n_subfacet = 0; | |
1371 | backer->avg_subfacet_life = 0; | |
1372 | ||
acf60855 JP |
1373 | return error; |
1374 | } | |
1375 | ||
abe529af | 1376 | static int |
0f5f95a9 | 1377 | construct(struct ofproto *ofproto_) |
abe529af BP |
1378 | { |
1379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1380 | struct shash_node *node, *next; |
91858960 | 1381 | int max_ports; |
abe529af BP |
1382 | int error; |
1383 | int i; | |
1384 | ||
acf60855 | 1385 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1386 | if (error) { |
abe529af BP |
1387 | return error; |
1388 | } | |
1389 | ||
acf60855 | 1390 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
91858960 BP |
1391 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); |
1392 | ||
abe529af BP |
1393 | ofproto->netflow = NULL; |
1394 | ofproto->sflow = NULL; | |
29089a54 | 1395 | ofproto->ipfix = NULL; |
21f7563c | 1396 | ofproto->stp = NULL; |
abe529af | 1397 | hmap_init(&ofproto->bundles); |
e764773c | 1398 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
1399 | for (i = 0; i < MAX_MIRRORS; i++) { |
1400 | ofproto->mirrors[i] = NULL; | |
1401 | } | |
1402 | ofproto->has_bonded_bundles = false; | |
1403 | ||
abe529af | 1404 | hmap_init(&ofproto->facets); |
4d2a0f39 | 1405 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 BP |
1406 | |
1407 | for (i = 0; i < N_TABLES; i++) { | |
1408 | struct table_dpif *table = &ofproto->tables[i]; | |
1409 | ||
1410 | table->catchall_table = NULL; | |
1411 | table->other_table = NULL; | |
1412 | table->basis = random_uint32(); | |
1413 | } | |
abe529af | 1414 | |
7ee20df1 BP |
1415 | list_init(&ofproto->completions); |
1416 | ||
abe529af BP |
1417 | ofproto_dpif_unixctl_init(); |
1418 | ||
ccb7c863 | 1419 | ofproto->has_mirrors = false; |
daff3353 EJ |
1420 | ofproto->has_bundle_action = false; |
1421 | ||
52a90c29 BP |
1422 | hmap_init(&ofproto->vlandev_map); |
1423 | hmap_init(&ofproto->realdev_vid_map); | |
1424 | ||
acf60855 | 1425 | sset_init(&ofproto->ports); |
0a740f48 | 1426 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1427 | sset_init(&ofproto->port_poll_set); |
1428 | ofproto->port_poll_errno = 0; | |
1429 | ||
1430 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1431 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1432 | |
1433 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1434 | /* Check if the datapath already has this port. */ | |
1435 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1436 | sset_add(&ofproto->ports, node->name); | |
1437 | } | |
1438 | ||
1439 | free(iface_hint->br_name); | |
1440 | free(iface_hint->br_type); | |
4f9e08a5 | 1441 | free(iface_hint); |
acf60855 JP |
1442 | shash_delete(&init_ofp_ports, node); |
1443 | } | |
1444 | } | |
e1b1d06a | 1445 | |
b44a10b7 BP |
1446 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1447 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1448 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1449 | |
1450 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1451 | error = add_internal_flows(ofproto); |
1452 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1453 | ||
735d7efb AZ |
1454 | ofproto->n_hit = 0; |
1455 | ofproto->n_missed = 0; | |
1456 | ||
c57b2226 BP |
1457 | return error; |
1458 | } | |
1459 | ||
1460 | static int | |
1461 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1462 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1463 | { |
1464 | struct ofputil_flow_mod fm; | |
1465 | int error; | |
1466 | ||
81a76618 BP |
1467 | match_init_catchall(&fm.match); |
1468 | fm.priority = 0; | |
1469 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1470 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1471 | fm.cookie = htonll(0); |
1472 | fm.cookie_mask = htonll(0); | |
1473 | fm.table_id = TBL_INTERNAL; | |
1474 | fm.command = OFPFC_ADD; | |
1475 | fm.idle_timeout = 0; | |
1476 | fm.hard_timeout = 0; | |
1477 | fm.buffer_id = 0; | |
1478 | fm.out_port = 0; | |
1479 | fm.flags = 0; | |
f25d0cf3 BP |
1480 | fm.ofpacts = ofpacts->data; |
1481 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1482 | |
1483 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1484 | if (error) { | |
1485 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1486 | id, ofperr_to_string(error)); | |
1487 | return error; | |
1488 | } | |
1489 | ||
81a76618 | 1490 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); |
cb22974d | 1491 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1492 | |
abe529af BP |
1493 | return 0; |
1494 | } | |
1495 | ||
c57b2226 BP |
1496 | static int |
1497 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1498 | { | |
f25d0cf3 BP |
1499 | struct ofpact_controller *controller; |
1500 | uint64_t ofpacts_stub[128 / 8]; | |
1501 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1502 | int error; |
1503 | int id; | |
1504 | ||
f25d0cf3 | 1505 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1506 | id = 1; |
1507 | ||
f25d0cf3 BP |
1508 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1509 | controller->max_len = UINT16_MAX; | |
1510 | controller->controller_id = 0; | |
1511 | controller->reason = OFPR_NO_MATCH; | |
1512 | ofpact_pad(&ofpacts); | |
1513 | ||
1514 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1515 | if (error) { |
1516 | return error; | |
1517 | } | |
1518 | ||
f25d0cf3 BP |
1519 | ofpbuf_clear(&ofpacts); |
1520 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 | 1521 | &ofproto->no_packet_in_rule); |
7fd51d39 BP |
1522 | if (error) { |
1523 | return error; | |
1524 | } | |
1525 | ||
1526 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1527 | &ofproto->drop_frags_rule); | |
c57b2226 BP |
1528 | return error; |
1529 | } | |
1530 | ||
7ee20df1 BP |
1531 | static void |
1532 | complete_operations(struct ofproto_dpif *ofproto) | |
1533 | { | |
1534 | struct dpif_completion *c, *next; | |
1535 | ||
1536 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1537 | ofoperation_complete(c->op, 0); | |
1538 | list_remove(&c->list_node); | |
1539 | free(c); | |
1540 | } | |
1541 | } | |
1542 | ||
abe529af BP |
1543 | static void |
1544 | destruct(struct ofproto *ofproto_) | |
1545 | { | |
1546 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1547 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1548 | struct oftable *table; |
abe529af BP |
1549 | int i; |
1550 | ||
b44a10b7 | 1551 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1552 | complete_operations(ofproto); |
1553 | ||
0697b5c3 BP |
1554 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1555 | struct cls_cursor cursor; | |
1556 | ||
d0918789 | 1557 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1558 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1559 | ofproto_rule_destroy(&rule->up); | |
1560 | } | |
7ee20df1 BP |
1561 | } |
1562 | ||
abe529af BP |
1563 | for (i = 0; i < MAX_MIRRORS; i++) { |
1564 | mirror_destroy(ofproto->mirrors[i]); | |
1565 | } | |
1566 | ||
1567 | netflow_destroy(ofproto->netflow); | |
bae473fe | 1568 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
1569 | hmap_destroy(&ofproto->bundles); |
1570 | mac_learning_destroy(ofproto->ml); | |
1571 | ||
1572 | hmap_destroy(&ofproto->facets); | |
1573 | ||
52a90c29 BP |
1574 | hmap_destroy(&ofproto->vlandev_map); |
1575 | hmap_destroy(&ofproto->realdev_vid_map); | |
1576 | ||
acf60855 | 1577 | sset_destroy(&ofproto->ports); |
0a740f48 | 1578 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1579 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1580 | |
acf60855 | 1581 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1582 | } |
1583 | ||
1584 | static int | |
5fcc0d00 | 1585 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1586 | { |
1587 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 1588 | struct ofport_dpif *ofport; |
abe529af | 1589 | |
40358701 GS |
1590 | /* Do not perform any periodic activity required by 'ofproto' while |
1591 | * waiting for flow restore to complete. */ | |
1592 | if (ofproto_get_flow_restore_wait()) { | |
1593 | return 0; | |
1594 | } | |
1595 | ||
0aa66d6e EJ |
1596 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1597 | port_run_fast(ofport); | |
1598 | } | |
1599 | ||
5fcc0d00 BP |
1600 | return 0; |
1601 | } | |
1602 | ||
1603 | static int | |
1604 | run(struct ofproto *ofproto_) | |
1605 | { | |
1606 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1607 | struct ofport_dpif *ofport; | |
1608 | struct ofbundle *bundle; | |
1609 | int error; | |
1610 | ||
1611 | if (!clogged) { | |
1612 | complete_operations(ofproto); | |
1613 | } | |
5fcc0d00 | 1614 | |
40358701 GS |
1615 | /* Do not perform any periodic activity below required by 'ofproto' while |
1616 | * waiting for flow restore to complete. */ | |
1617 | if (ofproto_get_flow_restore_wait()) { | |
1618 | return 0; | |
1619 | } | |
1620 | ||
5fcc0d00 BP |
1621 | error = run_fast(ofproto_); |
1622 | if (error) { | |
1623 | return error; | |
abe529af BP |
1624 | } |
1625 | ||
abe529af | 1626 | if (ofproto->netflow) { |
6fca1ffb BP |
1627 | if (netflow_run(ofproto->netflow)) { |
1628 | send_netflow_active_timeouts(ofproto); | |
1629 | } | |
abe529af BP |
1630 | } |
1631 | if (ofproto->sflow) { | |
bae473fe | 1632 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1633 | } |
1634 | ||
1635 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1636 | port_run(ofport); | |
1637 | } | |
1638 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1639 | bundle_run(bundle); | |
1640 | } | |
1641 | ||
21f7563c | 1642 | stp_run(ofproto); |
2cc3c58e | 1643 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af | 1644 | |
6814e51f | 1645 | /* Check the consistency of a random facet, to aid debugging. */ |
4d2a0f39 EJ |
1646 | if (time_msec() >= ofproto->consistency_rl |
1647 | && !hmap_is_empty(&ofproto->facets) | |
2cc3c58e | 1648 | && !ofproto->backer->need_revalidate) { |
6814e51f BP |
1649 | struct facet *facet; |
1650 | ||
4d2a0f39 EJ |
1651 | ofproto->consistency_rl = time_msec() + 250; |
1652 | ||
6814e51f BP |
1653 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), |
1654 | struct facet, hmap_node); | |
2cc3c58e | 1655 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, |
bbafd73b | 1656 | facet->xout.tags)) { |
6814e51f | 1657 | if (!facet_check_consistency(facet)) { |
2cc3c58e | 1658 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1659 | } |
1660 | } | |
1661 | } | |
1662 | ||
abe529af BP |
1663 | return 0; |
1664 | } | |
1665 | ||
1666 | static void | |
1667 | wait(struct ofproto *ofproto_) | |
1668 | { | |
1669 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1670 | struct ofport_dpif *ofport; | |
1671 | struct ofbundle *bundle; | |
1672 | ||
7ee20df1 BP |
1673 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1674 | poll_immediate_wake(); | |
1675 | } | |
1676 | ||
40358701 GS |
1677 | if (ofproto_get_flow_restore_wait()) { |
1678 | return; | |
1679 | } | |
1680 | ||
acf60855 JP |
1681 | dpif_wait(ofproto->backer->dpif); |
1682 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1683 | if (ofproto->sflow) { |
bae473fe | 1684 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1685 | } |
2cc3c58e | 1686 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { |
abe529af BP |
1687 | poll_immediate_wake(); |
1688 | } | |
1689 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1690 | port_wait(ofport); | |
1691 | } | |
1692 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1693 | bundle_wait(bundle); | |
1694 | } | |
6fca1ffb BP |
1695 | if (ofproto->netflow) { |
1696 | netflow_wait(ofproto->netflow); | |
1697 | } | |
1c313b88 | 1698 | mac_learning_wait(ofproto->ml); |
21f7563c | 1699 | stp_wait(ofproto); |
2cc3c58e | 1700 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1701 | /* Shouldn't happen, but if it does just go around again. */ |
1702 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1703 | poll_immediate_wake(); | |
abe529af BP |
1704 | } |
1705 | } | |
1706 | ||
0d085684 BP |
1707 | static void |
1708 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1709 | { | |
1710 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
04d08d54 EJ |
1711 | size_t n_subfacets = 0; |
1712 | struct facet *facet; | |
0d085684 BP |
1713 | |
1714 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
04d08d54 EJ |
1715 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
1716 | n_subfacets += list_size(&facet->subfacets); | |
1717 | } | |
1718 | simap_increase(usage, "subfacets", n_subfacets); | |
0d085684 BP |
1719 | } |
1720 | ||
abe529af BP |
1721 | static void |
1722 | flush(struct ofproto *ofproto_) | |
1723 | { | |
1724 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1725 | struct subfacet *subfacet, *next_subfacet; |
1726 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1727 | int n_batch; | |
b0f7b9b5 | 1728 | |
acf60855 JP |
1729 | n_batch = 0; |
1730 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
04d08d54 EJ |
1731 | &ofproto->backer->subfacets) { |
1732 | if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) { | |
1733 | continue; | |
1734 | } | |
1735 | ||
acf60855 JP |
1736 | if (subfacet->path != SF_NOT_INSTALLED) { |
1737 | batch[n_batch++] = subfacet; | |
1738 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
04d08d54 | 1739 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 JP |
1740 | n_batch = 0; |
1741 | } | |
1742 | } else { | |
1743 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1744 | } |
abe529af | 1745 | } |
acf60855 JP |
1746 | |
1747 | if (n_batch > 0) { | |
04d08d54 | 1748 | subfacet_destroy_batch(ofproto->backer, batch, n_batch); |
acf60855 | 1749 | } |
abe529af BP |
1750 | } |
1751 | ||
6c1491fb BP |
1752 | static void |
1753 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1754 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1755 | { |
1756 | *arp_match_ip = true; | |
9e1fd49b BP |
1757 | *actions = (OFPUTIL_A_OUTPUT | |
1758 | OFPUTIL_A_SET_VLAN_VID | | |
1759 | OFPUTIL_A_SET_VLAN_PCP | | |
1760 | OFPUTIL_A_STRIP_VLAN | | |
1761 | OFPUTIL_A_SET_DL_SRC | | |
1762 | OFPUTIL_A_SET_DL_DST | | |
1763 | OFPUTIL_A_SET_NW_SRC | | |
1764 | OFPUTIL_A_SET_NW_DST | | |
1765 | OFPUTIL_A_SET_NW_TOS | | |
1766 | OFPUTIL_A_SET_TP_SRC | | |
1767 | OFPUTIL_A_SET_TP_DST | | |
1768 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1769 | } |
1770 | ||
1771 | static void | |
307975da | 1772 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1773 | { |
1774 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1775 | struct dpif_dp_stats s; |
7fd51d39 | 1776 | uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; |
7e741ffb | 1777 | uint64_t n_lookup; |
6c1491fb BP |
1778 | |
1779 | strcpy(ots->name, "classifier"); | |
1780 | ||
acf60855 | 1781 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1782 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1783 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
7fd51d39 | 1784 | rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); |
acf60855 | 1785 | |
7fd51d39 | 1786 | n_lookup = s.n_hit + s.n_missed - n_dropped_frags; |
7e741ffb JG |
1787 | ots->lookup_count = htonll(n_lookup); |
1788 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1789 | } |
1790 | ||
abe529af BP |
1791 | static struct ofport * |
1792 | port_alloc(void) | |
1793 | { | |
1794 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1795 | return &port->up; | |
1796 | } | |
1797 | ||
1798 | static void | |
1799 | port_dealloc(struct ofport *port_) | |
1800 | { | |
1801 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1802 | free(port); | |
1803 | } | |
1804 | ||
1805 | static int | |
1806 | port_construct(struct ofport *port_) | |
1807 | { | |
1808 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1809 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1810 | const struct netdev *netdev = port->up.netdev; |
3aa30359 | 1811 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
e1b1d06a JP |
1812 | struct dpif_port dpif_port; |
1813 | int error; | |
abe529af | 1814 | |
2cc3c58e | 1815 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1816 | port->bundle = NULL; |
1817 | port->cfm = NULL; | |
ccc09689 | 1818 | port->bfd = NULL; |
abe529af | 1819 | port->tag = tag_create_random(); |
d5ffa7f2 | 1820 | port->may_enable = true; |
21f7563c JP |
1821 | port->stp_port = NULL; |
1822 | port->stp_state = STP_DISABLED; | |
b9ad7294 | 1823 | port->tnl_port = NULL; |
8b36f51e | 1824 | hmap_init(&port->priorities); |
52a90c29 BP |
1825 | port->realdev_ofp_port = 0; |
1826 | port->vlandev_vid = 0; | |
b9ad7294 | 1827 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1828 | |
b9ad7294 | 1829 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1830 | /* By bailing out here, we don't submit the port to the sFlow module |
1831 | * to be considered for counter polling export. This is correct | |
1832 | * because the patch port represents an interface that sFlow considers | |
1833 | * to be "internal" to the switch as a whole, and therefore not an | |
1834 | * candidate for counter polling. */ | |
0a740f48 EJ |
1835 | port->odp_port = OVSP_NONE; |
1836 | return 0; | |
1837 | } | |
1838 | ||
acf60855 | 1839 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
3aa30359 BP |
1840 | netdev_vport_get_dpif_port(netdev, namebuf, |
1841 | sizeof namebuf), | |
e1b1d06a JP |
1842 | &dpif_port); |
1843 | if (error) { | |
1844 | return error; | |
1845 | } | |
1846 | ||
1847 | port->odp_port = dpif_port.port_no; | |
1848 | ||
b9ad7294 EJ |
1849 | if (netdev_get_tunnel_config(netdev)) { |
1850 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1851 | } else { | |
1852 | /* Sanity-check that a mapping doesn't already exist. This | |
1853 | * shouldn't happen for non-tunnel ports. */ | |
1854 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1855 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1856 | dpif_port.name); | |
da78d43d | 1857 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1858 | return EBUSY; |
1859 | } | |
e1b1d06a | 1860 | |
b9ad7294 EJ |
1861 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
1862 | hash_int(port->odp_port, 0)); | |
1863 | } | |
da78d43d | 1864 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1865 | |
abe529af | 1866 | if (ofproto->sflow) { |
e1b1d06a | 1867 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1868 | } |
1869 | ||
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | static void | |
1874 | port_destruct(struct ofport *port_) | |
1875 | { | |
1876 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
02f8d646 | 1878 | const char *devname = netdev_get_name(port->up.netdev); |
3aa30359 BP |
1879 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
1880 | const char *dp_port_name; | |
abe529af | 1881 | |
3aa30359 BP |
1882 | dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, |
1883 | sizeof namebuf); | |
a614d823 | 1884 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1885 | /* The underlying device is still there, so delete it. This |
1886 | * happens when the ofproto is being destroyed, since the caller | |
1887 | * assumes that removal of attached ports will happen as part of | |
1888 | * destruction. */ | |
a614d823 KM |
1889 | if (!port->tnl_port) { |
1890 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1891 | } | |
1892 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
acf60855 JP |
1893 | } |
1894 | ||
b9ad7294 | 1895 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { |
0a740f48 EJ |
1896 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1897 | } | |
1898 | ||
b9ad7294 | 1899 | tnl_port_del(port->tnl_port); |
02f8d646 | 1900 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1901 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
2cc3c58e | 1902 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1903 | bundle_remove(port_); |
a5610457 | 1904 | set_cfm(port_, NULL); |
8aee94b6 | 1905 | set_bfd(port_, NULL); |
abe529af | 1906 | if (ofproto->sflow) { |
bae473fe | 1907 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1908 | } |
8b36f51e EJ |
1909 | |
1910 | ofport_clear_priorities(port); | |
1911 | hmap_destroy(&port->priorities); | |
abe529af BP |
1912 | } |
1913 | ||
1914 | static void | |
1915 | port_modified(struct ofport *port_) | |
1916 | { | |
1917 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1918 | ||
1919 | if (port->bundle && port->bundle->bond) { | |
1920 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1921 | } | |
9d46b444 EJ |
1922 | |
1923 | if (port->cfm) { | |
1924 | cfm_set_netdev(port->cfm, port->up.netdev); | |
1925 | } | |
abe529af BP |
1926 | } |
1927 | ||
1928 | static void | |
9e1fd49b | 1929 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1930 | { |
1931 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1932 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1933 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1934 | |
9e1fd49b | 1935 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1936 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1937 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1938 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1939 | |
9e1fd49b | 1940 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1941 | bundle_update(port->bundle); |
1942 | } | |
abe529af BP |
1943 | } |
1944 | } | |
1945 | ||
1946 | static int | |
1947 | set_sflow(struct ofproto *ofproto_, | |
1948 | const struct ofproto_sflow_options *sflow_options) | |
1949 | { | |
1950 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1951 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1952 | |
abe529af | 1953 | if (sflow_options) { |
bae473fe | 1954 | if (!ds) { |
abe529af BP |
1955 | struct ofport_dpif *ofport; |
1956 | ||
4213f19d | 1957 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1958 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1959 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1960 | } |
2cc3c58e | 1961 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1962 | } |
bae473fe | 1963 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1964 | } else { |
6ff686f2 PS |
1965 | if (ds) { |
1966 | dpif_sflow_destroy(ds); | |
2cc3c58e | 1967 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1968 | ofproto->sflow = NULL; |
1969 | } | |
abe529af BP |
1970 | } |
1971 | return 0; | |
1972 | } | |
1973 | ||
29089a54 RL |
1974 | static int |
1975 | set_ipfix( | |
1976 | struct ofproto *ofproto_, | |
1977 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1978 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1979 | size_t n_flow_exporters_options) | |
1980 | { | |
1981 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1982 | struct dpif_ipfix *di = ofproto->ipfix; | |
1983 | ||
1984 | if (bridge_exporter_options || flow_exporters_options) { | |
1985 | if (!di) { | |
1986 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1987 | } | |
1988 | dpif_ipfix_set_options( | |
1989 | di, bridge_exporter_options, flow_exporters_options, | |
1990 | n_flow_exporters_options); | |
1991 | } else { | |
1992 | if (di) { | |
1993 | dpif_ipfix_destroy(di); | |
1994 | ofproto->ipfix = NULL; | |
1995 | } | |
1996 | } | |
1997 | return 0; | |
1998 | } | |
1999 | ||
abe529af | 2000 | static int |
a5610457 | 2001 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
2002 | { |
2003 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2004 | int error; | |
2005 | ||
a5610457 | 2006 | if (!s) { |
abe529af BP |
2007 | error = 0; |
2008 | } else { | |
2009 | if (!ofport->cfm) { | |
8c977421 EJ |
2010 | struct ofproto_dpif *ofproto; |
2011 | ||
2012 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 2013 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 2014 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
2015 | } |
2016 | ||
a5610457 | 2017 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
2018 | return 0; |
2019 | } | |
2020 | ||
2021 | error = EINVAL; | |
2022 | } | |
2023 | cfm_destroy(ofport->cfm); | |
2024 | ofport->cfm = NULL; | |
2025 | return error; | |
2026 | } | |
2027 | ||
9a9e3786 BP |
2028 | static bool |
2029 | get_cfm_status(const struct ofport *ofport_, | |
2030 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
2031 | { |
2032 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2033 | ||
2034 | if (ofport->cfm) { | |
9a9e3786 BP |
2035 | status->faults = cfm_get_fault(ofport->cfm); |
2036 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
2037 | status->health = cfm_get_health(ofport->cfm); | |
2038 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
2039 | return true; | |
1de11730 | 2040 | } else { |
9a9e3786 | 2041 | return false; |
1de11730 EJ |
2042 | } |
2043 | } | |
ccc09689 EJ |
2044 | |
2045 | static int | |
2046 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
2047 | { | |
2048 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
2049 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2050 | struct bfd *old; | |
2051 | ||
2052 | old = ofport->bfd; | |
2053 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
2054 | if (ofport->bfd != old) { | |
2055 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2056 | } | |
2057 | ||
2058 | return 0; | |
2059 | } | |
2060 | ||
2061 | static int | |
2062 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
2063 | { | |
2064 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2065 | ||
2066 | if (ofport->bfd) { | |
2067 | bfd_get_status(ofport->bfd, smap); | |
2068 | return 0; | |
2069 | } else { | |
2070 | return ENOENT; | |
2071 | } | |
2072 | } | |
abe529af | 2073 | \f |
21f7563c JP |
2074 | /* Spanning Tree. */ |
2075 | ||
2076 | static void | |
2077 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2078 | { | |
2079 | struct ofproto_dpif *ofproto = ofproto_; | |
2080 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2081 | struct ofport_dpif *ofport; | |
2082 | ||
2083 | ofport = stp_port_get_aux(sp); | |
2084 | if (!ofport) { | |
2085 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2086 | ofproto->up.name, port_num); | |
2087 | } else { | |
2088 | struct eth_header *eth = pkt->l2; | |
2089 | ||
2090 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2091 | if (eth_addr_is_zero(eth->eth_src)) { | |
2092 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2093 | "with unknown MAC", ofproto->up.name, port_num); | |
2094 | } else { | |
97d6520b | 2095 | send_packet(ofport, pkt); |
21f7563c JP |
2096 | } |
2097 | } | |
2098 | ofpbuf_delete(pkt); | |
2099 | } | |
2100 | ||
2101 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2102 | static int | |
2103 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2104 | { | |
2105 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2106 | ||
2107 | /* Only revalidate flows if the configuration changed. */ | |
2108 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2109 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2110 | } |
2111 | ||
2112 | if (s) { | |
2113 | if (!ofproto->stp) { | |
2114 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2115 | send_bpdu_cb, ofproto); | |
2116 | ofproto->stp_last_tick = time_msec(); | |
2117 | } | |
2118 | ||
2119 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2120 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2121 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2122 | stp_set_max_age(ofproto->stp, s->max_age); | |
2123 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2124 | } else { | |
851bf71d EJ |
2125 | struct ofport *ofport; |
2126 | ||
2127 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2128 | set_stp_port(ofport, NULL); | |
2129 | } | |
2130 | ||
21f7563c JP |
2131 | stp_destroy(ofproto->stp); |
2132 | ofproto->stp = NULL; | |
2133 | } | |
2134 | ||
2135 | return 0; | |
2136 | } | |
2137 | ||
2138 | static int | |
2139 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2140 | { | |
2141 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2142 | ||
2143 | if (ofproto->stp) { | |
2144 | s->enabled = true; | |
2145 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2146 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2147 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2148 | } else { | |
2149 | s->enabled = false; | |
2150 | } | |
2151 | ||
2152 | return 0; | |
2153 | } | |
2154 | ||
2155 | static void | |
2156 | update_stp_port_state(struct ofport_dpif *ofport) | |
2157 | { | |
2158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2159 | enum stp_state state; | |
2160 | ||
2161 | /* Figure out new state. */ | |
2162 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2163 | : STP_DISABLED; | |
2164 | ||
2165 | /* Update state. */ | |
2166 | if (ofport->stp_state != state) { | |
9e1fd49b | 2167 | enum ofputil_port_state of_state; |
21f7563c JP |
2168 | bool fwd_change; |
2169 | ||
2170 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2171 | netdev_get_name(ofport->up.netdev), | |
2172 | stp_state_name(ofport->stp_state), | |
2173 | stp_state_name(state)); | |
2174 | if (stp_learn_in_state(ofport->stp_state) | |
2175 | != stp_learn_in_state(state)) { | |
2176 | /* xxx Learning action flows should also be flushed. */ | |
2cc3c58e EJ |
2177 | mac_learning_flush(ofproto->ml, |
2178 | &ofproto->backer->revalidate_set); | |
21f7563c JP |
2179 | } |
2180 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2181 | != stp_forward_in_state(state); | |
2182 | ||
2cc3c58e | 2183 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2184 | ofport->stp_state = state; |
2185 | ofport->stp_state_entered = time_msec(); | |
2186 | ||
b308140a | 2187 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2188 | bundle_update(ofport->bundle); |
2189 | } | |
2190 | ||
2191 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2192 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2193 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2194 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2195 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2196 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2197 | : 0); | |
21f7563c JP |
2198 | ofproto_port_set_state(&ofport->up, of_state); |
2199 | } | |
2200 | } | |
2201 | ||
2202 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2203 | * caller is responsible for assigning STP port numbers and ensuring | |
2204 | * there are no duplicates. */ | |
2205 | static int | |
2206 | set_stp_port(struct ofport *ofport_, | |
2207 | const struct ofproto_port_stp_settings *s) | |
2208 | { | |
2209 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2210 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2211 | struct stp_port *sp = ofport->stp_port; | |
2212 | ||
2213 | if (!s || !s->enable) { | |
2214 | if (sp) { | |
2215 | ofport->stp_port = NULL; | |
2216 | stp_port_disable(sp); | |
ecd12731 | 2217 | update_stp_port_state(ofport); |
21f7563c JP |
2218 | } |
2219 | return 0; | |
2220 | } else if (sp && stp_port_no(sp) != s->port_num | |
2221 | && ofport == stp_port_get_aux(sp)) { | |
2222 | /* The port-id changed, so disable the old one if it's not | |
2223 | * already in use by another port. */ | |
2224 | stp_port_disable(sp); | |
2225 | } | |
2226 | ||
2227 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2228 | stp_port_enable(sp); | |
2229 | ||
2230 | stp_port_set_aux(sp, ofport); | |
2231 | stp_port_set_priority(sp, s->priority); | |
2232 | stp_port_set_path_cost(sp, s->path_cost); | |
2233 | ||
2234 | update_stp_port_state(ofport); | |
2235 | ||
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | static int | |
2240 | get_stp_port_status(struct ofport *ofport_, | |
2241 | struct ofproto_port_stp_status *s) | |
2242 | { | |
2243 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2244 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2245 | struct stp_port *sp = ofport->stp_port; | |
2246 | ||
2247 | if (!ofproto->stp || !sp) { | |
2248 | s->enabled = false; | |
2249 | return 0; | |
2250 | } | |
2251 | ||
2252 | s->enabled = true; | |
2253 | s->port_id = stp_port_get_id(sp); | |
2254 | s->state = stp_port_get_state(sp); | |
2255 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2256 | s->role = stp_port_get_role(sp); | |
80740385 | 2257 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2258 | |
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | static void | |
2263 | stp_run(struct ofproto_dpif *ofproto) | |
2264 | { | |
2265 | if (ofproto->stp) { | |
2266 | long long int now = time_msec(); | |
2267 | long long int elapsed = now - ofproto->stp_last_tick; | |
2268 | struct stp_port *sp; | |
2269 | ||
2270 | if (elapsed > 0) { | |
2271 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2272 | ofproto->stp_last_tick = now; | |
2273 | } | |
2274 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2275 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2276 | ||
2277 | if (ofport) { | |
2278 | update_stp_port_state(ofport); | |
2279 | } | |
2280 | } | |
6ae50723 EJ |
2281 | |
2282 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2cc3c58e | 2283 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
6ae50723 | 2284 | } |
21f7563c JP |
2285 | } |
2286 | } | |
2287 | ||
2288 | static void | |
2289 | stp_wait(struct ofproto_dpif *ofproto) | |
2290 | { | |
2291 | if (ofproto->stp) { | |
2292 | poll_timer_wait(1000); | |
2293 | } | |
2294 | } | |
2295 | ||
2296 | /* Returns true if STP should process 'flow'. */ | |
2297 | static bool | |
2298 | stp_should_process_flow(const struct flow *flow) | |
2299 | { | |
2300 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2301 | } | |
2302 | ||
2303 | static void | |
2304 | stp_process_packet(const struct ofport_dpif *ofport, | |
2305 | const struct ofpbuf *packet) | |
2306 | { | |
2307 | struct ofpbuf payload = *packet; | |
2308 | struct eth_header *eth = payload.data; | |
2309 | struct stp_port *sp = ofport->stp_port; | |
2310 | ||
2311 | /* Sink packets on ports that have STP disabled when the bridge has | |
2312 | * STP enabled. */ | |
2313 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2314 | return; | |
2315 | } | |
2316 | ||
2317 | /* Trim off padding on payload. */ | |
c573540b BP |
2318 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
2319 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
2320 | } |
2321 | ||
2322 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2323 | stp_received_bpdu(sp, payload.data, payload.size); | |
2324 | } | |
2325 | } | |
2326 | \f | |
8b36f51e EJ |
2327 | static struct priority_to_dscp * |
2328 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2329 | { | |
2330 | struct priority_to_dscp *pdscp; | |
2331 | uint32_t hash; | |
2332 | ||
2333 | hash = hash_int(priority, 0); | |
2334 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2335 | if (pdscp->priority == priority) { | |
2336 | return pdscp; | |
2337 | } | |
2338 | } | |
2339 | return NULL; | |
2340 | } | |
2341 | ||
2342 | static void | |
2343 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2344 | { | |
2345 | struct priority_to_dscp *pdscp, *next; | |
2346 | ||
2347 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2348 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2349 | free(pdscp); | |
2350 | } | |
2351 | } | |
2352 | ||
2353 | static int | |
2354 | set_queues(struct ofport *ofport_, | |
2355 | const struct ofproto_port_queue *qdscp_list, | |
2356 | size_t n_qdscp) | |
2357 | { | |
2358 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2359 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2360 | struct hmap new = HMAP_INITIALIZER(&new); | |
2361 | size_t i; | |
2362 | ||
2363 | for (i = 0; i < n_qdscp; i++) { | |
2364 | struct priority_to_dscp *pdscp; | |
2365 | uint32_t priority; | |
2366 | uint8_t dscp; | |
2367 | ||
2368 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
acf60855 | 2369 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, |
8b36f51e EJ |
2370 | &priority)) { |
2371 | continue; | |
2372 | } | |
2373 | ||
2374 | pdscp = get_priority(ofport, priority); | |
2375 | if (pdscp) { | |
2376 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2377 | } else { | |
2378 | pdscp = xmalloc(sizeof *pdscp); | |
2379 | pdscp->priority = priority; | |
2380 | pdscp->dscp = dscp; | |
2cc3c58e | 2381 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2382 | } |
2383 | ||
2384 | if (pdscp->dscp != dscp) { | |
2385 | pdscp->dscp = dscp; | |
2cc3c58e | 2386 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2387 | } |
2388 | ||
2389 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2390 | } | |
2391 | ||
2392 | if (!hmap_is_empty(&ofport->priorities)) { | |
2393 | ofport_clear_priorities(ofport); | |
2cc3c58e | 2394 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2395 | } |
2396 | ||
2397 | hmap_swap(&new, &ofport->priorities); | |
2398 | hmap_destroy(&new); | |
2399 | ||
2400 | return 0; | |
2401 | } | |
2402 | \f | |
abe529af BP |
2403 | /* Bundles. */ |
2404 | ||
b44a10b7 BP |
2405 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2406 | * ofproto to revalidate every flow. | |
2407 | * | |
2408 | * Normally MAC learning entries are removed only from the ofproto associated | |
2409 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2410 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2411 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2412 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2413 | * with the host from which it migrated. */ | |
abe529af | 2414 | static void |
b44a10b7 | 2415 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2416 | { |
2417 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2418 | struct mac_learning *ml = ofproto->ml; | |
2419 | struct mac_entry *mac, *next_mac; | |
2420 | ||
2cc3c58e | 2421 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2422 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2423 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2424 | if (all_ofprotos) { |
2425 | struct ofproto_dpif *o; | |
2426 | ||
2427 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2428 | if (o != ofproto) { | |
2429 | struct mac_entry *e; | |
2430 | ||
2431 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2432 | NULL); | |
2433 | if (e) { | |
b44a10b7 BP |
2434 | mac_learning_expire(o->ml, e); |
2435 | } | |
2436 | } | |
2437 | } | |
2438 | } | |
2439 | ||
abe529af BP |
2440 | mac_learning_expire(ml, mac); |
2441 | } | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | static struct ofbundle * | |
2446 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2447 | { | |
2448 | struct ofbundle *bundle; | |
2449 | ||
2450 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2451 | &ofproto->bundles) { | |
2452 | if (bundle->aux == aux) { | |
2453 | return bundle; | |
2454 | } | |
2455 | } | |
2456 | return NULL; | |
2457 | } | |
2458 | ||
2459 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2460 | * ones that are found to 'bundles'. */ | |
2461 | static void | |
2462 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2463 | void **auxes, size_t n_auxes, | |
2464 | struct hmapx *bundles) | |
2465 | { | |
2466 | size_t i; | |
2467 | ||
2468 | hmapx_init(bundles); | |
2469 | for (i = 0; i < n_auxes; i++) { | |
2470 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2471 | if (bundle) { | |
2472 | hmapx_add(bundles, bundle); | |
2473 | } | |
2474 | } | |
2475 | } | |
2476 | ||
7bde8dd8 JP |
2477 | static void |
2478 | bundle_update(struct ofbundle *bundle) | |
2479 | { | |
2480 | struct ofport_dpif *port; | |
2481 | ||
2482 | bundle->floodable = true; | |
2483 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2484 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2485 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2486 | bundle->floodable = false; |
2487 | break; | |
2488 | } | |
2489 | } | |
2490 | } | |
2491 | ||
abe529af BP |
2492 | static void |
2493 | bundle_del_port(struct ofport_dpif *port) | |
2494 | { | |
2495 | struct ofbundle *bundle = port->bundle; | |
2496 | ||
2cc3c58e | 2497 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2498 | |
abe529af BP |
2499 | list_remove(&port->bundle_node); |
2500 | port->bundle = NULL; | |
2501 | ||
2502 | if (bundle->lacp) { | |
2503 | lacp_slave_unregister(bundle->lacp, port); | |
2504 | } | |
2505 | if (bundle->bond) { | |
2506 | bond_slave_unregister(bundle->bond, port); | |
2507 | } | |
2508 | ||
7bde8dd8 | 2509 | bundle_update(bundle); |
abe529af BP |
2510 | } |
2511 | ||
2512 | static bool | |
213b00cf | 2513 | bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port, |
df53d41c | 2514 | struct lacp_slave_settings *lacp) |
abe529af BP |
2515 | { |
2516 | struct ofport_dpif *port; | |
2517 | ||
2518 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2519 | if (!port) { | |
2520 | return false; | |
2521 | } | |
2522 | ||
2523 | if (port->bundle != bundle) { | |
2cc3c58e | 2524 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2525 | if (port->bundle) { |
2526 | bundle_del_port(port); | |
2527 | } | |
2528 | ||
2529 | port->bundle = bundle; | |
2530 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2531 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2532 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2533 | bundle->floodable = false; |
2534 | } | |
2535 | } | |
2536 | if (lacp) { | |
2cc3c58e | 2537 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2538 | lacp_slave_register(bundle->lacp, port, lacp); |
2539 | } | |
2540 | ||
2541 | return true; | |
2542 | } | |
2543 | ||
2544 | static void | |
2545 | bundle_destroy(struct ofbundle *bundle) | |
2546 | { | |
2547 | struct ofproto_dpif *ofproto; | |
2548 | struct ofport_dpif *port, *next_port; | |
2549 | int i; | |
2550 | ||
2551 | if (!bundle) { | |
2552 | return; | |
2553 | } | |
2554 | ||
2555 | ofproto = bundle->ofproto; | |
2556 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2557 | struct ofmirror *m = ofproto->mirrors[i]; | |
2558 | if (m) { | |
2559 | if (m->out == bundle) { | |
2560 | mirror_destroy(m); | |
2561 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2562 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2cc3c58e | 2563 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2564 | } |
2565 | } | |
2566 | } | |
2567 | ||
2568 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2569 | bundle_del_port(port); | |
2570 | } | |
2571 | ||
b44a10b7 | 2572 | bundle_flush_macs(bundle, true); |
abe529af BP |
2573 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2574 | free(bundle->name); | |
2575 | free(bundle->trunks); | |
2576 | lacp_destroy(bundle->lacp); | |
2577 | bond_destroy(bundle->bond); | |
2578 | free(bundle); | |
2579 | } | |
2580 | ||
2581 | static int | |
2582 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2583 | const struct ofproto_bundle_settings *s) | |
2584 | { | |
2585 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2586 | bool need_flush = false; | |
abe529af BP |
2587 | struct ofport_dpif *port; |
2588 | struct ofbundle *bundle; | |
ecac4ebf BP |
2589 | unsigned long *trunks; |
2590 | int vlan; | |
abe529af BP |
2591 | size_t i; |
2592 | bool ok; | |
2593 | ||
2594 | if (!s) { | |
2595 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2596 | return 0; | |
2597 | } | |
2598 | ||
cb22974d BP |
2599 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2600 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2601 | |
2602 | bundle = bundle_lookup(ofproto, aux); | |
2603 | if (!bundle) { | |
2604 | bundle = xmalloc(sizeof *bundle); | |
2605 | ||
2606 | bundle->ofproto = ofproto; | |
2607 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2608 | hash_pointer(aux, 0)); | |
2609 | bundle->aux = aux; | |
2610 | bundle->name = NULL; | |
2611 | ||
2612 | list_init(&bundle->ports); | |
ecac4ebf | 2613 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2614 | bundle->vlan = -1; |
2615 | bundle->trunks = NULL; | |
5e9ceccd | 2616 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2617 | bundle->lacp = NULL; |
2618 | bundle->bond = NULL; | |
2619 | ||
2620 | bundle->floodable = true; | |
2621 | ||
2622 | bundle->src_mirrors = 0; | |
2623 | bundle->dst_mirrors = 0; | |
2624 | bundle->mirror_out = 0; | |
2625 | } | |
2626 | ||
2627 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2628 | free(bundle->name); | |
2629 | bundle->name = xstrdup(s->name); | |
2630 | } | |
2631 | ||
2632 | /* LACP. */ | |
2633 | if (s->lacp) { | |
2634 | if (!bundle->lacp) { | |
2cc3c58e | 2635 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2636 | bundle->lacp = lacp_create(); |
2637 | } | |
2638 | lacp_configure(bundle->lacp, s->lacp); | |
2639 | } else { | |
2640 | lacp_destroy(bundle->lacp); | |
2641 | bundle->lacp = NULL; | |
2642 | } | |
2643 | ||
2644 | /* Update set of ports. */ | |
2645 | ok = true; | |
2646 | for (i = 0; i < s->n_slaves; i++) { | |
2647 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2648 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2649 | ok = false; |
2650 | } | |
2651 | } | |
2652 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2653 | struct ofport_dpif *next_port; | |
2654 | ||
2655 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2656 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2657 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2658 | goto found; |
2659 | } | |
2660 | } | |
2661 | ||
2662 | bundle_del_port(port); | |
2663 | found: ; | |
2664 | } | |
2665 | } | |
cb22974d | 2666 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2667 | |
2668 | if (list_is_empty(&bundle->ports)) { | |
2669 | bundle_destroy(bundle); | |
2670 | return EINVAL; | |
2671 | } | |
2672 | ||
ecac4ebf | 2673 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2674 | if (s->vlan_mode != bundle->vlan_mode |
2675 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2676 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2677 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2678 | need_flush = true; |
2679 | } | |
2680 | ||
abe529af | 2681 | /* Set VLAN tag. */ |
ecac4ebf BP |
2682 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2683 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2684 | : 0); | |
2685 | if (vlan != bundle->vlan) { | |
2686 | bundle->vlan = vlan; | |
abe529af BP |
2687 | need_flush = true; |
2688 | } | |
2689 | ||
2690 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2691 | switch (s->vlan_mode) { |
2692 | case PORT_VLAN_ACCESS: | |
2693 | trunks = NULL; | |
2694 | break; | |
2695 | ||
2696 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2697 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2698 | break; |
2699 | ||
2700 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2701 | case PORT_VLAN_NATIVE_TAGGED: | |
2702 | if (vlan != 0 && (!s->trunks | |
2703 | || !bitmap_is_set(s->trunks, vlan) | |
2704 | || bitmap_is_set(s->trunks, 0))) { | |
2705 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2706 | if (s->trunks) { | |
2707 | trunks = bitmap_clone(s->trunks, 4096); | |
2708 | } else { | |
2709 | trunks = bitmap_allocate1(4096); | |
2710 | } | |
2711 | bitmap_set1(trunks, vlan); | |
2712 | bitmap_set0(trunks, 0); | |
2713 | } else { | |
ebc56baa | 2714 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2715 | } |
2716 | break; | |
2717 | ||
2718 | default: | |
2719 | NOT_REACHED(); | |
2720 | } | |
abe529af BP |
2721 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2722 | free(bundle->trunks); | |
ecac4ebf BP |
2723 | if (trunks == s->trunks) { |
2724 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2725 | } else { | |
2726 | bundle->trunks = trunks; | |
2727 | trunks = NULL; | |
2728 | } | |
abe529af BP |
2729 | need_flush = true; |
2730 | } | |
ecac4ebf BP |
2731 | if (trunks != s->trunks) { |
2732 | free(trunks); | |
2733 | } | |
abe529af BP |
2734 | |
2735 | /* Bonding. */ | |
2736 | if (!list_is_short(&bundle->ports)) { | |
2737 | bundle->ofproto->has_bonded_bundles = true; | |
2738 | if (bundle->bond) { | |
2739 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2740 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2741 | } |
2742 | } else { | |
2743 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2744 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2745 | } |
2746 | ||
2747 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2748 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2749 | } |
2750 | } else { | |
2751 | bond_destroy(bundle->bond); | |
2752 | bundle->bond = NULL; | |
2753 | } | |
2754 | ||
2755 | /* If we changed something that would affect MAC learning, un-learn | |
2756 | * everything on this port and force flow revalidation. */ | |
2757 | if (need_flush) { | |
b44a10b7 | 2758 | bundle_flush_macs(bundle, false); |
abe529af BP |
2759 | } |
2760 | ||
2761 | return 0; | |
2762 | } | |
2763 | ||
2764 | static void | |
2765 | bundle_remove(struct ofport *port_) | |
2766 | { | |
2767 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2768 | struct ofbundle *bundle = port->bundle; | |
2769 | ||
2770 | if (bundle) { | |
2771 | bundle_del_port(port); | |
2772 | if (list_is_empty(&bundle->ports)) { | |
2773 | bundle_destroy(bundle); | |
2774 | } else if (list_is_short(&bundle->ports)) { | |
2775 | bond_destroy(bundle->bond); | |
2776 | bundle->bond = NULL; | |
2777 | } | |
2778 | } | |
2779 | } | |
2780 | ||
2781 | static void | |
5f877369 | 2782 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2783 | { |
2784 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2785 | struct ofport_dpif *port = port_; | |
2786 | uint8_t ea[ETH_ADDR_LEN]; | |
2787 | int error; | |
2788 | ||
2789 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2790 | if (!error) { | |
abe529af | 2791 | struct ofpbuf packet; |
5f877369 | 2792 | void *packet_pdu; |
abe529af BP |
2793 | |
2794 | ofpbuf_init(&packet, 0); | |
2795 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2796 | pdu_size); |
2797 | memcpy(packet_pdu, pdu, pdu_size); | |
2798 | ||
97d6520b | 2799 | send_packet(port, &packet); |
abe529af BP |
2800 | ofpbuf_uninit(&packet); |
2801 | } else { | |
2802 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2803 | "%s (%s)", port->bundle->name, | |
2804 | netdev_get_name(port->up.netdev), strerror(error)); | |
2805 | } | |
2806 | } | |
2807 | ||
2808 | static void | |
2809 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2810 | { | |
2811 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2812 | int error, n_packets, n_errors; | |
2813 | struct mac_entry *e; | |
2814 | ||
2815 | error = n_packets = n_errors = 0; | |
2816 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2817 | if (e->port.p != bundle) { | |
ea131871 JG |
2818 | struct ofpbuf *learning_packet; |
2819 | struct ofport_dpif *port; | |
4dd1e3ca | 2820 | void *port_void; |
ea131871 JG |
2821 | int ret; |
2822 | ||
4dd1e3ca BP |
2823 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2824 | * struct ofport_dpif more effective. */ | |
2825 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2826 | e->mac, e->vlan, | |
2827 | &port_void); | |
2828 | port = port_void; | |
97d6520b | 2829 | ret = send_packet(port, learning_packet); |
ea131871 | 2830 | ofpbuf_delete(learning_packet); |
abe529af BP |
2831 | if (ret) { |
2832 | error = ret; | |
2833 | n_errors++; | |
2834 | } | |
2835 | n_packets++; | |
2836 | } | |
2837 | } | |
2838 | ||
2839 | if (n_errors) { | |
2840 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2841 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2842 | "packets, last error was: %s", | |
2843 | bundle->name, n_errors, n_packets, strerror(error)); | |
2844 | } else { | |
2845 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2846 | bundle->name, n_packets); | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | static void | |
2851 | bundle_run(struct ofbundle *bundle) | |
2852 | { | |
2853 | if (bundle->lacp) { | |
2854 | lacp_run(bundle->lacp, send_pdu_cb); | |
2855 | } | |
2856 | if (bundle->bond) { | |
2857 | struct ofport_dpif *port; | |
2858 | ||
2859 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2860 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2861 | } |
2862 | ||
2cc3c58e | 2863 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, |
bdebeece | 2864 | lacp_status(bundle->lacp)); |
abe529af BP |
2865 | if (bond_should_send_learning_packets(bundle->bond)) { |
2866 | bundle_send_learning_packets(bundle); | |
2867 | } | |
2868 | } | |
2869 | } | |
2870 | ||
2871 | static void | |
2872 | bundle_wait(struct ofbundle *bundle) | |
2873 | { | |
2874 | if (bundle->lacp) { | |
2875 | lacp_wait(bundle->lacp); | |
2876 | } | |
2877 | if (bundle->bond) { | |
2878 | bond_wait(bundle->bond); | |
2879 | } | |
2880 | } | |
2881 | \f | |
2882 | /* Mirrors. */ | |
2883 | ||
2884 | static int | |
2885 | mirror_scan(struct ofproto_dpif *ofproto) | |
2886 | { | |
2887 | int idx; | |
2888 | ||
2889 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2890 | if (!ofproto->mirrors[idx]) { | |
2891 | return idx; | |
2892 | } | |
2893 | } | |
2894 | return -1; | |
2895 | } | |
2896 | ||
2897 | static struct ofmirror * | |
2898 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2899 | { | |
2900 | int i; | |
2901 | ||
2902 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2903 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2904 | if (mirror && mirror->aux == aux) { | |
2905 | return mirror; | |
2906 | } | |
2907 | } | |
2908 | ||
2909 | return NULL; | |
2910 | } | |
2911 | ||
9ba15e2a BP |
2912 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2913 | static void | |
2914 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2915 | { | |
2916 | int i; | |
2917 | ||
2918 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2919 | struct ofmirror *m = ofproto->mirrors[i]; | |
2920 | ||
2921 | if (m) { | |
2922 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2923 | } | |
2924 | } | |
2925 | ||
2926 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2927 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2928 | int j; | |
2929 | ||
2930 | if (!m1) { | |
2931 | continue; | |
2932 | } | |
2933 | ||
2934 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2935 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2936 | ||
edb0540b | 2937 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2938 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2939 | m2->dup_mirrors |= m1->dup_mirrors; | |
2940 | } | |
2941 | } | |
2942 | } | |
2943 | } | |
2944 | ||
abe529af BP |
2945 | static int |
2946 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2947 | const struct ofproto_mirror_settings *s) | |
2948 | { | |
2949 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2950 | mirror_mask_t mirror_bit; | |
2951 | struct ofbundle *bundle; | |
2952 | struct ofmirror *mirror; | |
2953 | struct ofbundle *out; | |
2954 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2955 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2956 | int out_vlan; | |
2957 | ||
2958 | mirror = mirror_lookup(ofproto, aux); | |
2959 | if (!s) { | |
2960 | mirror_destroy(mirror); | |
2961 | return 0; | |
2962 | } | |
2963 | if (!mirror) { | |
2964 | int idx; | |
2965 | ||
2966 | idx = mirror_scan(ofproto); | |
2967 | if (idx < 0) { | |
2968 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2969 | "cannot create %s", | |
2970 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2971 | return EFBIG; | |
2972 | } | |
2973 | ||
2974 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2975 | mirror->ofproto = ofproto; | |
2976 | mirror->idx = idx; | |
8b28d864 | 2977 | mirror->aux = aux; |
abe529af BP |
2978 | mirror->out_vlan = -1; |
2979 | mirror->name = NULL; | |
2980 | } | |
2981 | ||
2982 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2983 | free(mirror->name); | |
2984 | mirror->name = xstrdup(s->name); | |
2985 | } | |
2986 | ||
2987 | /* Get the new configuration. */ | |
2988 | if (s->out_bundle) { | |
2989 | out = bundle_lookup(ofproto, s->out_bundle); | |
2990 | if (!out) { | |
2991 | mirror_destroy(mirror); | |
2992 | return EINVAL; | |
2993 | } | |
2994 | out_vlan = -1; | |
2995 | } else { | |
2996 | out = NULL; | |
2997 | out_vlan = s->out_vlan; | |
2998 | } | |
2999 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
3000 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
3001 | ||
3002 | /* If the configuration has not changed, do nothing. */ | |
3003 | if (hmapx_equals(&srcs, &mirror->srcs) | |
3004 | && hmapx_equals(&dsts, &mirror->dsts) | |
3005 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
3006 | && mirror->out == out | |
3007 | && mirror->out_vlan == out_vlan) | |
3008 | { | |
3009 | hmapx_destroy(&srcs); | |
3010 | hmapx_destroy(&dsts); | |
3011 | return 0; | |
3012 | } | |
3013 | ||
3014 | hmapx_swap(&srcs, &mirror->srcs); | |
3015 | hmapx_destroy(&srcs); | |
3016 | ||
3017 | hmapx_swap(&dsts, &mirror->dsts); | |
3018 | hmapx_destroy(&dsts); | |
3019 | ||
3020 | free(mirror->vlans); | |
3021 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
3022 | ||
3023 | mirror->out = out; | |
3024 | mirror->out_vlan = out_vlan; | |
3025 | ||
3026 | /* Update bundles. */ | |
3027 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
3028 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
3029 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
3030 | bundle->src_mirrors |= mirror_bit; | |
3031 | } else { | |
3032 | bundle->src_mirrors &= ~mirror_bit; | |
3033 | } | |
3034 | ||
3035 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
3036 | bundle->dst_mirrors |= mirror_bit; | |
3037 | } else { | |
3038 | bundle->dst_mirrors &= ~mirror_bit; | |
3039 | } | |
3040 | ||
3041 | if (mirror->out == bundle) { | |
3042 | bundle->mirror_out |= mirror_bit; | |
3043 | } else { | |
3044 | bundle->mirror_out &= ~mirror_bit; | |
3045 | } | |
3046 | } | |
3047 | ||
2cc3c58e | 3048 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 3049 | ofproto->has_mirrors = true; |
2cc3c58e EJ |
3050 | mac_learning_flush(ofproto->ml, |
3051 | &ofproto->backer->revalidate_set); | |
9ba15e2a | 3052 | mirror_update_dups(ofproto); |
abe529af BP |
3053 | |
3054 | return 0; | |
3055 | } | |
3056 | ||
3057 | static void | |
3058 | mirror_destroy(struct ofmirror *mirror) | |
3059 | { | |
3060 | struct ofproto_dpif *ofproto; | |
3061 | mirror_mask_t mirror_bit; | |
3062 | struct ofbundle *bundle; | |
ccb7c863 | 3063 | int i; |
abe529af BP |
3064 | |
3065 | if (!mirror) { | |
3066 | return; | |
3067 | } | |
3068 | ||
3069 | ofproto = mirror->ofproto; | |
2cc3c58e EJ |
3070 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3071 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
abe529af BP |
3072 | |
3073 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
3074 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3075 | bundle->src_mirrors &= ~mirror_bit; | |
3076 | bundle->dst_mirrors &= ~mirror_bit; | |
3077 | bundle->mirror_out &= ~mirror_bit; | |
3078 | } | |
3079 | ||
3080 | hmapx_destroy(&mirror->srcs); | |
3081 | hmapx_destroy(&mirror->dsts); | |
3082 | free(mirror->vlans); | |
3083 | ||
3084 | ofproto->mirrors[mirror->idx] = NULL; | |
3085 | free(mirror->name); | |
3086 | free(mirror); | |
9ba15e2a BP |
3087 | |
3088 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
3089 | |
3090 | ofproto->has_mirrors = false; | |
3091 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3092 | if (ofproto->mirrors[i]) { | |
3093 | ofproto->has_mirrors = true; | |
3094 | break; | |
3095 | } | |
3096 | } | |
abe529af BP |
3097 | } |
3098 | ||
9d24de3b JP |
3099 | static int |
3100 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
3101 | uint64_t *packets, uint64_t *bytes) | |
3102 | { | |
3103 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3104 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
3105 | ||
3106 | if (!mirror) { | |
3107 | *packets = *bytes = UINT64_MAX; | |
3108 | return 0; | |
3109 | } | |
3110 | ||
8844e035 EJ |
3111 | push_all_stats(); |
3112 | ||
9d24de3b JP |
3113 | *packets = mirror->packet_count; |
3114 | *bytes = mirror->byte_count; | |
3115 | ||
3116 | return 0; | |
3117 | } | |
3118 | ||
abe529af BP |
3119 | static int |
3120 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
3121 | { | |
3122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3123 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2cc3c58e | 3124 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af BP |
3125 | } |
3126 | return 0; | |
3127 | } | |
3128 | ||
3129 | static bool | |
b4affc74 | 3130 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
3131 | { |
3132 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3133 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
3134 | return bundle && bundle->mirror_out != 0; | |
3135 | } | |
8402c74b SS |
3136 | |
3137 | static void | |
b53055f4 | 3138 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
3139 | { |
3140 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 3141 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 3142 | } |
e764773c BP |
3143 | |
3144 | static void | |
c4069512 BP |
3145 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
3146 | size_t max_entries) | |
e764773c BP |
3147 | { |
3148 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3149 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
c4069512 | 3150 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
e764773c | 3151 | } |
abe529af BP |
3152 | \f |
3153 | /* Ports. */ | |
3154 | ||
3155 | static struct ofport_dpif * | |
4acbc98d | 3156 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 3157 | { |
7df6a8bd BP |
3158 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
3159 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
3160 | } |
3161 | ||
3162 | static struct ofport_dpif * | |
4acbc98d | 3163 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af | 3164 | { |
7c33b188 JR |
3165 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
3166 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
3167 | } |
3168 | ||
3169 | static void | |
e1b1d06a JP |
3170 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
3171 | struct ofproto_port *ofproto_port, | |
abe529af BP |
3172 | struct dpif_port *dpif_port) |
3173 | { | |
3174 | ofproto_port->name = dpif_port->name; | |
3175 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 3176 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
3177 | } |
3178 | ||
0a740f48 EJ |
3179 | static struct ofport_dpif * |
3180 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
3181 | { | |
3182 | const struct ofproto_dpif *ofproto; | |
3183 | const char *peer; | |
3184 | ||
3185 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
3186 | if (!peer) { | |
3187 | return NULL; | |
3188 | } | |
3189 | ||
3190 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3191 | struct ofport *ofport; | |
3192 | ||
3193 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
3194 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
3195 | return ofport_dpif_cast(ofport); | |
3196 | } | |
3197 | } | |
3198 | return NULL; | |
3199 | } | |
3200 | ||
0aa66d6e EJ |
3201 | static void |
3202 | port_run_fast(struct ofport_dpif *ofport) | |
3203 | { | |
3204 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
3205 | struct ofpbuf packet; | |
3206 | ||
3207 | ofpbuf_init(&packet, 0); | |
3208 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
3209 | send_packet(ofport, &packet); | |
3210 | ofpbuf_uninit(&packet); | |
3211 | } | |
ccc09689 EJ |
3212 | |
3213 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
3214 | struct ofpbuf packet; | |
3215 | ||
3216 | ofpbuf_init(&packet, 0); | |
3217 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
3218 | send_packet(ofport, &packet); | |
3219 | ofpbuf_uninit(&packet); | |
3220 | } | |
0aa66d6e EJ |
3221 | } |
3222 | ||
abe529af BP |
3223 | static void |
3224 | port_run(struct ofport_dpif *ofport) | |
3225 | { | |
3e5b3fdb EJ |
3226 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
3227 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
3228 | bool enable = netdev_get_carrier(ofport->up.netdev); |
3229 | ||
3e5b3fdb EJ |
3230 | ofport->carrier_seq = carrier_seq; |
3231 | ||
0aa66d6e | 3232 | port_run_fast(ofport); |
b9ad7294 EJ |
3233 | |
3234 | if (ofport->tnl_port | |
3235 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
3236 | &ofport->tnl_port)) { | |
3237 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
3238 | } | |
3239 | ||
abe529af | 3240 | if (ofport->cfm) { |
4653c558 EJ |
3241 | int cfm_opup = cfm_get_opup(ofport->cfm); |
3242 | ||
abe529af | 3243 | cfm_run(ofport->cfm); |
4653c558 EJ |
3244 | enable = enable && !cfm_get_fault(ofport->cfm); |
3245 | ||
3246 | if (cfm_opup >= 0) { | |
3247 | enable = enable && cfm_opup; | |
3248 | } | |
abe529af | 3249 | } |
015e08bc | 3250 | |
ccc09689 EJ |
3251 | if (ofport->bfd) { |
3252 | bfd_run(ofport->bfd); | |
3253 | enable = enable && bfd_forwarding(ofport->bfd); | |
3254 | } | |
3255 | ||
015e08bc EJ |
3256 | if (ofport->bundle) { |
3257 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
3258 | if (carrier_changed) { |
3259 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3260 | } | |
015e08bc EJ |
3261 | } |
3262 | ||
daff3353 EJ |
3263 | if (ofport->may_enable != enable) { |
3264 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3265 | ||
3266 | if (ofproto->has_bundle_action) { | |
2cc3c58e | 3267 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3268 | } |
3269 | } | |
3270 | ||
015e08bc | 3271 | ofport->may_enable = enable; |
abe529af BP |
3272 | } |
3273 | ||
3274 | static void | |
3275 | port_wait(struct ofport_dpif *ofport) | |
3276 | { | |
3277 | if (ofport->cfm) { | |
3278 | cfm_wait(ofport->cfm); | |
3279 | } | |
ccc09689 EJ |
3280 | |
3281 | if (ofport->bfd) { | |
3282 | bfd_wait(ofport->bfd); | |
3283 | } | |
abe529af BP |
3284 | } |
3285 | ||
3286 | static int | |
3287 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3288 | struct ofproto_port *ofproto_port) | |
3289 | { | |
3290 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3291 | struct dpif_port dpif_port; | |
3292 | int error; | |
3293 | ||
0a740f48 EJ |
3294 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3295 | const char *type = netdev_get_type_from_name(devname); | |
3296 | ||
3297 | /* We may be called before ofproto->up.port_by_name is populated with | |
3298 | * the appropriate ofport. For this reason, we must get the name and | |
3299 | * type from the netdev layer directly. */ | |
3300 | if (type) { | |
3301 | const struct ofport *ofport; | |
3302 | ||
3303 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3304 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3305 | ofproto_port->name = xstrdup(devname); | |
3306 | ofproto_port->type = xstrdup(type); | |
3307 | return 0; | |
3308 | } | |
3309 | return ENODEV; | |
3310 | } | |
3311 | ||
acf60855 JP |
3312 | if (!sset_contains(&ofproto->ports, devname)) { |
3313 | return ENODEV; | |
3314 | } | |
3315 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3316 | devname, &dpif_port); | |
abe529af | 3317 | if (!error) { |
e1b1d06a | 3318 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3319 | } |
3320 | return error; | |
3321 | } | |
3322 | ||
3323 | static int | |
e1b1d06a | 3324 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3325 | { |
3326 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3327 | const char *devname = netdev_get_name(netdev); |
3aa30359 BP |
3328 | char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; |
3329 | const char *dp_port_name; | |
abe529af | 3330 | |
0a740f48 EJ |
3331 | if (netdev_vport_is_patch(netdev)) { |
3332 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3333 | return 0; | |
3334 | } | |
3335 | ||
3aa30359 | 3336 | dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); |
b9ad7294 | 3337 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
7d82ab2e KM |
3338 | uint32_t port_no = UINT32_MAX; |
3339 | int error; | |
3340 | ||
3341 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3342 | if (error) { |
3343 | return error; | |
3344 | } | |
7d82ab2e KM |
3345 | if (netdev_get_tunnel_config(netdev)) { |
3346 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3347 | } | |
acf60855 | 3348 | } |
b9ad7294 EJ |
3349 | |
3350 | if (netdev_get_tunnel_config(netdev)) { | |
3351 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3352 | } else { |
3353 | sset_add(&ofproto->ports, devname); | |
3354 | } | |
3355 | return 0; | |
3356 | } | |
3357 | ||
abe529af BP |
3358 | static int |
3359 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3360 | { | |
3361 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3362 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3363 | int error = 0; |
abe529af | 3364 | |
b9ad7294 EJ |
3365 | if (!ofport) { |
3366 | return 0; | |
e1b1d06a | 3367 | } |
b9ad7294 EJ |
3368 | |
3369 | sset_find_and_delete(&ofproto->ghost_ports, | |
3370 | netdev_get_name(ofport->up.netdev)); | |
a614d823 KM |
3371 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3372 | if (!ofport->tnl_port) { | |
b9ad7294 EJ |
3373 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3374 | if (!error) { | |
abe529af BP |
3375 | /* The caller is going to close ofport->up.netdev. If this is a |
3376 | * bonded port, then the bond is using that netdev, so remove it | |
3377 | * from the bond. The client will need to reconfigure everything | |
3378 | * after deleting ports, so then the slave will get re-added. */ | |
3379 | bundle_remove(&ofport->up); | |
3380 | } | |
3381 | } | |
3382 | return error; | |
3383 | } | |
3384 | ||
6527c598 PS |
3385 | static int |
3386 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3387 | { | |
3388 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3389 | int error; | |
3390 | ||
8844e035 EJ |
3391 | push_all_stats(); |
3392 | ||
6527c598 PS |
3393 | error = netdev_get_stats(ofport->up.netdev, stats); |
3394 | ||
ee382d89 | 3395 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3396 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3397 | ||
3398 | /* ofproto->stats.tx_packets represents packets that we created | |
3399 | * internally and sent to some port (e.g. packets sent with | |
3400 | * send_packet()). Account for them as if they had come from | |
3401 | * OFPP_LOCAL and got forwarded. */ | |
3402 | ||
3403 | if (stats->rx_packets != UINT64_MAX) { | |
3404 | stats->rx_packets += ofproto->stats.tx_packets; | |
3405 | } | |
3406 | ||
3407 | if (stats->rx_bytes != UINT64_MAX) { | |
3408 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3409 | } | |
3410 | ||
3411 | /* ofproto->stats.rx_packets represents packets that were received on | |
3412 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3413 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3414 | |
3415 | if (stats->tx_packets != UINT64_MAX) { | |
3416 | stats->tx_packets += ofproto->stats.rx_packets; | |
3417 | } | |
3418 | ||
3419 | if (stats->tx_bytes != UINT64_MAX) { | |
3420 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3421 | } | |
3422 | } | |
3423 | ||
3424 | return error; | |
3425 | } | |
3426 | ||
abe529af | 3427 | struct port_dump_state { |
acf60855 JP |
3428 | uint32_t bucket; |
3429 | uint32_t offset; | |
0a740f48 | 3430 | bool ghost; |
da78d43d BP |
3431 | |
3432 | struct ofproto_port port; | |
3433 | bool has_port; | |
abe529af BP |
3434 | }; |
3435 | ||
3436 | static int | |
acf60855 | 3437 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3438 | { |
0a740f48 | 3439 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3440 | return 0; |
3441 | } | |
3442 | ||
3443 | static int | |
b9ad7294 | 3444 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3445 | struct ofproto_port *port) |
3446 | { | |
e1b1d06a | 3447 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3448 | struct port_dump_state *state = state_; |
0a740f48 | 3449 | const struct sset *sset; |
acf60855 | 3450 | struct sset_node *node; |
abe529af | 3451 | |
da78d43d BP |
3452 | if (state->has_port) { |
3453 | ofproto_port_destroy(&state->port); | |
3454 | state->has_port = false; | |
3455 | } | |
0a740f48 EJ |
3456 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3457 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3458 | int error; |
3459 | ||
da78d43d BP |
3460 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3461 | if (!error) { | |
3462 | *port = state->port; | |
3463 | state->has_port = true; | |
3464 | return 0; | |
3465 | } else if (error != ENODEV) { | |
acf60855 JP |
3466 | return error; |
3467 | } | |
abe529af | 3468 | } |
acf60855 | 3469 | |
0a740f48 EJ |
3470 | if (!state->ghost) { |
3471 | state->ghost = true; | |
3472 | state->bucket = 0; | |
3473 | state->offset = 0; | |
3474 | return port_dump_next(ofproto_, state_, port); | |
3475 | } | |
3476 | ||
acf60855 | 3477 | return EOF; |
abe529af BP |
3478 | } |
3479 | ||
3480 | static int | |
3481 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3482 | { | |
3483 | struct port_dump_state *state = state_; | |
3484 | ||
da78d43d BP |
3485 | if (state->has_port) { |
3486 | ofproto_port_destroy(&state->port); | |
3487 | } | |
abe529af BP |
3488 | free(state); |
3489 | return 0; | |
3490 | } | |
3491 | ||
3492 | static int | |
3493 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3494 | { | |
3495 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3496 | |
3497 | if (ofproto->port_poll_errno) { | |
3498 | int error = ofproto->port_poll_errno; | |
3499 | ofproto->port_poll_errno = 0; | |
3500 | return error; | |
3501 | } | |
3502 | ||
3503 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3504 | return EAGAIN; | |
3505 | } | |
3506 | ||
3507 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3508 | return 0; | |
abe529af BP |
3509 | } |
3510 | ||
3511 | static void | |
3512 | port_poll_wait(const struct ofproto *ofproto_) | |
3513 | { | |
3514 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3515 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3516 | } |
3517 | ||
3518 | static int | |
3519 | port_is_lacp_current(const struct ofport *ofport_) | |
3520 | { | |
3521 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3522 | return (ofport->bundle && ofport->bundle->lacp | |
3523 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3524 | : -1); | |
3525 | } | |
3526 | \f | |
3527 | /* Upcall handling. */ | |
3528 | ||
501f8d1f BP |
3529 | /* Flow miss batching. |
3530 | * | |
3531 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3532 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3533 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3534 | * more packets, plus possibly installing the flow in the dpif. | |
3535 | * | |
3536 | * So far we only batch the operations that affect flow setup time the most. | |
3537 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3538 | struct flow_miss { | |
3539 | struct hmap_node hmap_node; | |
acf60855 | 3540 | struct ofproto_dpif *ofproto; |
501f8d1f | 3541 | struct flow flow; |
b0f7b9b5 | 3542 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3543 | const struct nlattr *key; |
3544 | size_t key_len; | |
14f94f9a | 3545 | struct initial_vals initial_vals; |
501f8d1f | 3546 | struct list packets; |
6a7e895f | 3547 | enum dpif_upcall_type upcall_type; |
501f8d1f BP |
3548 | }; |
3549 | ||
3550 | struct flow_miss_op { | |
c2b565b5 | 3551 | struct dpif_op dpif_op; |
bbafd73b EJ |
3552 | |
3553 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3554 | struct xlate_out xout; | |
3555 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
501f8d1f BP |
3556 | }; |
3557 | ||
62cd7072 BP |
3558 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3559 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3560 | * configurations. */ |
62cd7072 | 3561 | static void |
a39edbd4 | 3562 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3563 | const struct flow *flow) |
62cd7072 BP |
3564 | { |
3565 | struct ofputil_packet_in pin; | |
3566 | ||
3e3252fa EJ |
3567 | pin.packet = packet->data; |
3568 | pin.packet_len = packet->size; | |
62cd7072 | 3569 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3570 | pin.controller_id = 0; |
54834960 EJ |
3571 | |
3572 | pin.table_id = 0; | |
3573 | pin.cookie = 0; | |
3574 | ||
62cd7072 | 3575 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3576 | |
3577 | flow_get_metadata(flow, &pin.fmd); | |
3578 | ||
d8653c38 | 3579 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3580 | } |
3581 | ||
6a7e895f | 3582 | static enum slow_path_reason |
abe529af | 3583 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
ffaef958 | 3584 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) |
abe529af | 3585 | { |
b6e001b6 | 3586 | if (!ofport) { |
6a7e895f | 3587 | return 0; |
ffaef958 | 3588 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 3589 | if (packet) { |
abe529af BP |
3590 | cfm_process_heartbeat(ofport->cfm, packet); |
3591 | } | |
6a7e895f | 3592 | return SLOW_CFM; |
ccc09689 EJ |
3593 | } else if (ofport->bfd && bfd_should_process_flow(flow)) { |
3594 | if (packet) { | |
3595 | bfd_process_packet(ofport->bfd, flow, packet); | |
3596 | } | |
3597 | return SLOW_BFD; | |
b6e001b6 EJ |
3598 | } else if (ofport->bundle && ofport->bundle->lacp |
3599 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3600 | if (packet) { | |
3601 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 3602 | } |
6a7e895f | 3603 | return SLOW_LACP; |
21f7563c JP |
3604 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
3605 | if (packet) { | |
3606 | stp_process_packet(ofport, packet); | |
3607 | } | |
6a7e895f | 3608 | return SLOW_STP; |
ffaef958 BP |
3609 | } else { |
3610 | return 0; | |
abe529af | 3611 | } |
abe529af BP |
3612 | } |
3613 | ||
501f8d1f | 3614 | static struct flow_miss * |
ddbc5954 JP |
3615 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3616 | const struct flow *flow, uint32_t hash) | |
abe529af | 3617 | { |
501f8d1f | 3618 | struct flow_miss *miss; |
abe529af | 3619 | |
501f8d1f | 3620 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3621 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3622 | return miss; |
3623 | } | |
3624 | } | |
abe529af | 3625 | |
b23cdad9 | 3626 | return NULL; |
501f8d1f | 3627 | } |
abe529af | 3628 | |
9d6ac44e BP |
3629 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3630 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3631 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3632 | * returned op->subfacet, and if anything needs to be freed after processing | |
3633 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3634 | static void |
9d6ac44e BP |
3635 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3636 | struct flow_miss_op *op) | |
501f8d1f | 3637 | { |
14f94f9a | 3638 | if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) { |
9d6ac44e BP |
3639 | /* This packet was received on a VLAN splinter port. We |
3640 | * added a VLAN to the packet to make the packet resemble | |
3641 | * the flow, but the actions were composed assuming that | |
3642 | * the packet contained no VLAN. So, we must remove the | |
3643 | * VLAN header from the packet before trying to execute the | |
3644 | * actions. */ | |
3645 | eth_pop_vlan(packet); | |
3646 | } | |
3647 | ||
bbafd73b | 3648 | op->xout_garbage = false; |
9d6ac44e BP |
3649 | op->dpif_op.type = DPIF_OP_EXECUTE; |
3650 | op->dpif_op.u.execute.key = miss->key; | |
3651 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3652 | op->dpif_op.u.execute.packet = packet; | |
3653 | } | |
3654 | ||
3655 | /* Helper for handle_flow_miss_without_facet() and | |
3656 | * handle_flow_miss_with_facet(). */ | |
3657 | static void | |
3658 | handle_flow_miss_common(struct rule_dpif *rule, | |
3659 | struct ofpbuf *packet, const struct flow *flow) | |
3660 | { | |
3661 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3662 | ||
9d6ac44e BP |
3663 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { |
3664 | /* | |
3665 | * Extra-special case for fail-open mode. | |
3666 | * | |
3667 | * We are in fail-open mode and the packet matched the fail-open | |
3668 | * rule, but we are connected to a controller too. We should send | |
3669 | * the packet up to the controller in the hope that it will try to | |
3670 | * set up a flow and thereby allow us to exit fail-open. | |
3671 | * | |
3672 | * See the top-level comment in fail-open.c for more information. | |
3673 | */ | |
3674 | send_packet_in_miss(ofproto, packet, flow); | |
3675 | } | |
3676 | } | |
3677 | ||
3678 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3679 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3680 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3681 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3682 | * the benefits, so when the datapath holds a large number of flows we impose | |
3683 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3684 | static bool | |
04d08d54 | 3685 | flow_miss_should_make_facet(struct flow_miss *miss, uint32_t hash) |
9d6ac44e | 3686 | { |
04d08d54 EJ |
3687 | struct dpif_backer *backer = miss->ofproto->backer; |
3688 | ||
3689 | if (!backer->governor) { | |
9d6ac44e BP |
3690 | size_t n_subfacets; |
3691 | ||
04d08d54 | 3692 | n_subfacets = hmap_count(&backer->subfacets); |
380f49c4 | 3693 | if (n_subfacets * 2 <= flow_eviction_threshold) { |
9d6ac44e BP |
3694 | return true; |
3695 | } | |
3696 | ||
c985ec94 | 3697 | backer->governor = governor_create(); |
9d6ac44e BP |
3698 | } |
3699 | ||
04d08d54 | 3700 | return governor_should_install_flow(backer->governor, hash, |
9d6ac44e BP |
3701 | list_size(&miss->packets)); |
3702 | } | |
3703 | ||
3704 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3705 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3706 | * increment '*n_ops'. */ | |
3707 | static void | |
3708 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
9d6ac44e BP |
3709 | struct flow_miss_op *ops, size_t *n_ops) |
3710 | { | |
4dff9097 | 3711 | struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow); |
a7752d4a | 3712 | long long int now = time_msec(); |
530a1d91 | 3713 | struct ofpbuf *packet; |
bbafd73b | 3714 | struct xlate_in xin; |
2b459b83 | 3715 | |
9d6ac44e BP |
3716 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
3717 | struct flow_miss_op *op = &ops[*n_ops]; | |
3718 | struct dpif_flow_stats stats; | |
abe529af | 3719 | |
9d6ac44e | 3720 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3721 | |
7820e253 EJ |
3722 | handle_flow_miss_common(rule, packet, &miss->flow); |
3723 | ||
a7752d4a | 3724 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 3725 | rule_credit_stats(rule, &stats); |
abe529af | 3726 | |
bbafd73b EJ |
3727 | xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals, |
3728 | rule, stats.tcp_flags, packet); | |
3729 | xin.resubmit_stats = &stats; | |
3730 | xlate_actions(&xin, &op->xout); | |
abe529af | 3731 | |
bbafd73b | 3732 | if (op->xout.odp_actions.size) { |
9d6ac44e BP |
3733 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
3734 | ||
3735 | init_flow_miss_execute_op(miss, packet, op); | |
bbafd73b EJ |
3736 | execute->actions = op->xout.odp_actions.data; |
3737 | execute->actions_len = op->xout.odp_actions.size; | |
3738 | op->xout_garbage = true; | |
9d6ac44e BP |
3739 | |
3740 | (*n_ops)++; | |
3741 | } else { | |
bbafd73b | 3742 | xlate_out_uninit(&op->xout); |
9d6ac44e | 3743 | } |
abe529af | 3744 | } |
9d6ac44e BP |
3745 | } |
3746 | ||
3747 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3748 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3749 | * | |
3750 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3751 | * This is really important only for new facets: if we just called time_msec() | |
3752 | * here, then the new subfacet or its packets could look (occasionally) as | |
3753 | * though it was used some time after the facet was used. That can make a | |
3754 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3755 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3756 | static void |
3757 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 3758 | long long int now, |
9d6ac44e BP |
3759 | struct flow_miss_op *ops, size_t *n_ops) |
3760 | { | |
6a7e895f BP |
3761 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3762 | enum subfacet_path want_path; | |
9d6ac44e BP |
3763 | struct subfacet *subfacet; |
3764 | struct ofpbuf *packet; | |
abe529af | 3765 | |
a088a1ff | 3766 | subfacet = subfacet_create(facet, miss, now); |
bbafd73b | 3767 | want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
b0f7b9b5 | 3768 | |
530a1d91 | 3769 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3770 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 BP |
3771 | struct dpif_flow_stats stats; |
3772 | ||
9d6ac44e | 3773 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 3774 | |
4dff9097 | 3775 | if (want_path != SF_FAST_PATH) { |
bbafd73b | 3776 | struct xlate_in xin; |
454a77e5 | 3777 | |
bbafd73b EJ |
3778 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, |
3779 | facet->rule, 0, packet); | |
3780 | xlate_actions_for_side_effects(&xin); | |
501f8d1f | 3781 | } |
67d91f78 | 3782 | |
459b16a1 | 3783 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 3784 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 3785 | |
bbafd73b | 3786 | if (facet->xout.odp_actions.size) { |
9d6ac44e | 3787 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
8338659a | 3788 | |
9d6ac44e | 3789 | init_flow_miss_execute_op(miss, packet, op); |
bbafd73b EJ |
3790 | execute->actions = facet->xout.odp_actions.data, |
3791 | execute->actions_len = facet->xout.odp_actions.size; | |
9d6ac44e | 3792 | (*n_ops)++; |
5fe20d5d | 3793 | } |
501f8d1f BP |
3794 | } |
3795 | ||
6a7e895f | 3796 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { |
501f8d1f | 3797 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3798 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3799 | |
c84451a6 EJ |
3800 | subfacet->path = want_path; |
3801 | ||
bbafd73b | 3802 | op->xout_garbage = false; |
c2b565b5 | 3803 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
3804 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
3805 | put->key = miss->key; | |
3806 | put->key_len = miss->key_len; | |
6a7e895f | 3807 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3808 | put->actions = facet->xout.odp_actions.data; |
3809 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3810 | } else { |
bbafd73b EJ |
3811 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
3812 | op->slow_stub, sizeof op->slow_stub, | |
6a7e895f BP |
3813 | &put->actions, &put->actions_len); |
3814 | } | |
501f8d1f BP |
3815 | put->stats = NULL; |
3816 | } | |
3817 | } | |
3818 | ||
acf60855 JP |
3819 | /* Handles flow miss 'miss'. May add any required datapath operations |
3820 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3821 | static void |
acf60855 JP |
3822 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3823 | size_t *n_ops) | |
9d6ac44e | 3824 | { |
acf60855 | 3825 | struct ofproto_dpif *ofproto = miss->ofproto; |
9d6ac44e | 3826 | struct facet *facet; |
459b16a1 | 3827 | long long int now; |
9d6ac44e BP |
3828 | uint32_t hash; |
3829 | ||
3830 | /* The caller must ensure that miss->hmap_node.hash contains | |
3831 | * flow_hash(miss->flow, 0). */ | |
3832 | hash = miss->hmap_node.hash; | |
3833 | ||
3834 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3835 | if (!facet) { | |
ba33dd03 EJ |
3836 | /* There does not exist a bijection between 'struct flow' and datapath |
3837 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3838 | * assumption used throughout the facet and subfacet handling code. | |
3839 | * Since we have to handle these misses in userspace anyway, we simply | |
ec9f40dc | 3840 | * skip facet creation, avoiding the problem altogether. */ |
ba33dd03 | 3841 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE |
04d08d54 | 3842 | || !flow_miss_should_make_facet(miss, hash)) { |
4dff9097 | 3843 | handle_flow_miss_without_facet(miss, ops, n_ops); |
9d6ac44e BP |
3844 | return; |
3845 | } | |
3846 | ||
4dff9097 | 3847 | facet = facet_create(miss, hash); |
459b16a1 BP |
3848 | now = facet->used; |
3849 | } else { | |
3850 | now = time_msec(); | |
9d6ac44e | 3851 | } |
459b16a1 | 3852 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3853 | } |
3854 | ||
8f73d537 EJ |
3855 | static struct drop_key * |
3856 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3857 | size_t key_len) | |
3858 | { | |
3859 | struct drop_key *drop_key; | |
3860 | ||
3861 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3862 | &backer->drop_keys) { | |
3863 | if (drop_key->key_len == key_len | |
3864 | && !memcmp(drop_key->key, key, key_len)) { | |
3865 | return drop_key; | |
3866 | } | |
3867 | } | |
3868 | return NULL; | |
3869 | } | |
3870 | ||
3871 | static void | |
3872 | drop_key_clear(struct dpif_backer *backer) | |
3873 | { | |
3874 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3875 | struct drop_key *drop_key, *next; | |
3876 | ||
3877 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3878 | int error; | |
3879 | ||
3880 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3881 | NULL); | |
3882 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3883 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3884 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3885 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3886 | ds_cstr(&ds)); | |
3887 | ds_destroy(&ds); | |
3888 | } | |
3889 | ||
3890 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3891 | free(drop_key->key); | |
3892 | free(drop_key); | |
3893 | } | |
3894 | } | |
3895 | ||
e09ee259 EJ |
3896 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3897 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3898 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3899 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3900 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3901 | * 'packet' ingressed. | |
e2a6ca36 | 3902 | * |
e09ee259 EJ |
3903 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3904 | * 'flow''s in_port to OFPP_NONE. | |
3905 | * | |
3906 | * This function does post-processing on data returned from | |
3907 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3908 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3909 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3910 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3911 | * a VLAN header onto 'packet' (if it is nonnull). | |
3912 | * | |
c3f6c502 JP |
3913 | * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' |
3914 | * to the VLAN TCI with which the packet was really received, that is, the | |
3915 | * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from | |
3916 | * the value returned in flow->vlan_tci only for packets received on | |
715b48aa | 3917 | * VLAN splinters.) |
e09ee259 | 3918 | * |
b9ad7294 EJ |
3919 | * Similarly, this function also includes some logic to help with tunnels. It |
3920 | * may modify 'flow' as necessary to make the tunneling implementation | |
3921 | * transparent to the upcall processing logic. | |
3922 | * | |
e09ee259 EJ |
3923 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3924 | * or some other positive errno if there are other problems. */ | |
3925 | static int | |
3926 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3927 | const struct nlattr *key, size_t key_len, | |
3928 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3929 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
14f94f9a | 3930 | struct initial_vals *initial_vals) |
e84173dc | 3931 | { |
e09ee259 EJ |
3932 | const struct ofport_dpif *port; |
3933 | enum odp_key_fitness fitness; | |
b9ad7294 | 3934 | int error = ENODEV; |
e09ee259 EJ |
3935 | |
3936 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3937 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3938 | error = EINVAL; |
3939 | goto exit; | |
3940 | } | |
3941 | ||
14f94f9a JP |
3942 | if (initial_vals) { |
3943 | initial_vals->vlan_tci = flow->vlan_tci; | |
e84173dc | 3944 | } |
e84173dc | 3945 | |
e09ee259 EJ |
3946 | if (odp_in_port) { |
3947 | *odp_in_port = flow->in_port; | |
3948 | } | |
3949 | ||
3abe1835 BP |
3950 | port = (tnl_port_should_receive(flow) |
3951 | ? ofport_dpif_cast(tnl_port_receive(flow)) | |
3952 | : odp_port_to_ofport(backer, flow->in_port)); | |
3953 | flow->in_port = port ? port->up.ofp_port : OFPP_NONE; | |
3954 | if (!port) { | |
3955 | goto exit; | |
3956 | } | |
3957 | ||
3958 | /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port | |
3959 | * it's theoretically possible that we'll receive an ofport belonging to an | |
3960 | * entirely different datapath. In practice, this can't happen because no | |
3961 | * platforms has two separate datapaths which each support tunneling. */ | |
3962 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3963 | ||
3964 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3965 | if (packet) { | |
3966 | /* Make the packet resemble the flow, so that it gets sent to | |
3967 | * an OpenFlow controller properly, so that it looks correct | |
3968 | * for sFlow, and so that flow_extract() will get the correct | |
3969 | * vlan_tci if it is called on 'packet'. | |
3970 | * | |
3971 | * The allocated space inside 'packet' probably also contains | |
3972 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3973 | * a struct dpif_upcall (see the large comment on that | |
3974 | * structure definition), so pushing data on 'packet' is in | |
3975 | * general not a good idea since it could overwrite 'key' or | |
3976 | * free it as a side effect. However, it's OK in this special | |
3977 | * case because we know that 'packet' is inside a Netlink | |
3978 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3979 | * "struct nlattr", which is fine since we don't need that | |
3980 | * header anymore. */ | |
3981 | eth_push_vlan(packet, flow->vlan_tci); | |
3982 | } | |
3983 | /* We can't reproduce 'key' from 'flow'. */ | |
3984 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 | 3985 | } |
e09ee259 | 3986 | error = 0; |
52a90c29 | 3987 | |
b9ad7294 EJ |
3988 | if (ofproto) { |
3989 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3990 | } | |
3991 | ||
e09ee259 EJ |
3992 | exit: |
3993 | if (fitnessp) { | |
3994 | *fitnessp = fitness; | |
3995 | } | |
3996 | return error; | |
e84173dc BP |
3997 | } |
3998 | ||
501f8d1f | 3999 | static void |
acf60855 | 4000 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
4001 | size_t n_upcalls) |
4002 | { | |
4003 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
4004 | struct flow_miss *miss; |
4005 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 4006 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 4007 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 4008 | struct hmap todo; |
b23cdad9 | 4009 | int n_misses; |
501f8d1f BP |
4010 | size_t n_ops; |
4011 | size_t i; | |
4012 | ||
4013 | if (!n_upcalls) { | |
4014 | return; | |
4015 | } | |
4016 | ||
4017 | /* Construct the to-do list. | |
4018 | * | |
4019 | * This just amounts to extracting the flow from each packet and sticking | |
4020 | * the packets that have the same flow in the same "flow_miss" structure so | |
4021 | * that we can process them together. */ | |
4022 | hmap_init(&todo); | |
b23cdad9 | 4023 | n_misses = 0; |
501f8d1f | 4024 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
4025 | struct flow_miss *miss = &misses[n_misses]; |
4026 | struct flow_miss *existing_miss; | |
acf60855 | 4027 | struct ofproto_dpif *ofproto; |
a088a1ff | 4028 | uint32_t odp_in_port; |
1d446463 | 4029 | struct flow flow; |
b23cdad9 | 4030 | uint32_t hash; |
e09ee259 | 4031 | int error; |
501f8d1f | 4032 | |
e09ee259 EJ |
4033 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
4034 | upcall->key_len, &flow, &miss->key_fitness, | |
14f94f9a | 4035 | &ofproto, &odp_in_port, &miss->initial_vals); |
e09ee259 | 4036 | if (error == ENODEV) { |
8f73d537 EJ |
4037 | struct drop_key *drop_key; |
4038 | ||
c3385516 AW |
4039 | /* Received packet on datapath port for which we couldn't |
4040 | * associate an ofproto. This can happen if a port is removed | |
4041 | * while traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
4042 | * in case it happens frequently. Install a drop flow so |
4043 | * that future packets of the flow are inexpensively dropped | |
4044 | * in the kernel. */ | |
c3385516 AW |
4045 | VLOG_INFO_RL(&rl, "received packet on unassociated datapath port " |
4046 | "%"PRIu32, odp_in_port); | |
8f73d537 EJ |
4047 | |
4048 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
4049 | if (!drop_key) { | |
4050 | drop_key = xmalloc(sizeof *drop_key); | |
4051 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
4052 | drop_key->key_len = upcall->key_len; | |
4053 | ||
4054 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
4055 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
4056 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
4057 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
4058 | } | |
4059 | continue; | |
acf60855 | 4060 | } |
e09ee259 | 4061 | if (error) { |
b0f7b9b5 BP |
4062 | continue; |
4063 | } | |
735d7efb AZ |
4064 | |
4065 | ofproto->n_missed++; | |
72e8bf28 | 4066 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
1d446463 | 4067 | &flow.tunnel, flow.in_port, &miss->flow); |
501f8d1f | 4068 | |
501f8d1f | 4069 | /* Add other packets to a to-do list. */ |
b23cdad9 | 4070 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 4071 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
4072 | if (!existing_miss) { |
4073 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 4074 | miss->ofproto = ofproto; |
b23cdad9 BP |
4075 | miss->key = upcall->key; |
4076 | miss->key_len = upcall->key_len; | |
6a7e895f | 4077 | miss->upcall_type = upcall->type; |
b23cdad9 BP |
4078 | list_init(&miss->packets); |
4079 | ||
4080 | n_misses++; | |
4081 | } else { | |
4082 | miss = existing_miss; | |
4083 | } | |
501f8d1f BP |
4084 | list_push_back(&miss->packets, &upcall->packet->list_node); |
4085 | } | |
4086 | ||
4087 | /* Process each element in the to-do list, constructing the set of | |
4088 | * operations to batch. */ | |
4089 | n_ops = 0; | |
33bb0caa | 4090 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 4091 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 4092 | } |
cb22974d | 4093 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
4094 | |
4095 | /* Execute batch. */ | |
4096 | for (i = 0; i < n_ops; i++) { | |
4097 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
4098 | } | |
acf60855 | 4099 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 4100 | |
c84451a6 | 4101 | /* Free memory. */ |
501f8d1f | 4102 | for (i = 0; i < n_ops; i++) { |
bbafd73b EJ |
4103 | if (flow_miss_ops[i].xout_garbage) { |
4104 | xlate_out_uninit(&flow_miss_ops[i].xout); | |
4105 | } | |
501f8d1f | 4106 | } |
33bb0caa | 4107 | hmap_destroy(&todo); |
abe529af BP |
4108 | } |
4109 | ||
29089a54 RL |
4110 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, |
4111 | IPFIX_UPCALL } | |
6a7e895f BP |
4112 | classify_upcall(const struct dpif_upcall *upcall) |
4113 | { | |
29089a54 | 4114 | size_t userdata_len; |
6a7e895f BP |
4115 | union user_action_cookie cookie; |
4116 | ||
4117 | /* First look at the upcall type. */ | |
4118 | switch (upcall->type) { | |
4119 | case DPIF_UC_ACTION: | |
4120 | break; | |
4121 | ||
4122 | case DPIF_UC_MISS: | |
4123 | return MISS_UPCALL; | |
4124 | ||
4125 | case DPIF_N_UC_TYPES: | |
4126 | default: | |
4127 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
4128 | return BAD_UPCALL; | |
4129 | } | |
4130 | ||
4131 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
4132 | if (!upcall->userdata) { |
4133 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
4134 | return BAD_UPCALL; | |
4135 | } | |
29089a54 RL |
4136 | userdata_len = nl_attr_get_size(upcall->userdata); |
4137 | if (userdata_len < sizeof cookie.type | |
4138 | || userdata_len > sizeof cookie) { | |
e995e3df | 4139 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", |
29089a54 | 4140 | userdata_len); |
e995e3df BP |
4141 | return BAD_UPCALL; |
4142 | } | |
29089a54 RL |
4143 | memset(&cookie, 0, sizeof cookie); |
4144 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
4145 | if (userdata_len == sizeof cookie.sflow | |
4146 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
6a7e895f | 4147 | return SFLOW_UPCALL; |
29089a54 RL |
4148 | } else if (userdata_len == sizeof cookie.slow_path |
4149 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
6a7e895f | 4150 | return MISS_UPCALL; |
29089a54 RL |
4151 | } else if (userdata_len == sizeof cookie.flow_sample |
4152 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
4153 | return FLOW_SAMPLE_UPCALL; | |
4154 | } else if (userdata_len == sizeof cookie.ipfix | |
4155 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
4156 | return IPFIX_UPCALL; | |
4157 | } else { | |
4158 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
4159 | " and size %zu", cookie.type, userdata_len); | |
6a7e895f BP |
4160 | return BAD_UPCALL; |
4161 | } | |
4162 | } | |
4163 | ||
abe529af | 4164 | static void |
acf60855 | 4165 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 4166 | const struct dpif_upcall *upcall) |
abe529af | 4167 | { |
acf60855 | 4168 | struct ofproto_dpif *ofproto; |
1673e0e4 | 4169 | union user_action_cookie cookie; |
e84173dc | 4170 | struct flow flow; |
e1b1d06a | 4171 | uint32_t odp_in_port; |
abe529af | 4172 | |
e09ee259 EJ |
4173 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
4174 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
4175 | || !ofproto->sflow) { | |
e84173dc BP |
4176 | return; |
4177 | } | |
4178 | ||
29089a54 RL |
4179 | memset(&cookie, 0, sizeof cookie); |
4180 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
e1b1d06a JP |
4181 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
4182 | odp_in_port, &cookie); | |
6ff686f2 PS |
4183 | } |
4184 | ||
29089a54 RL |
4185 | static void |
4186 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
4187 | const struct dpif_upcall *upcall) | |
4188 | { | |
4189 | struct ofproto_dpif *ofproto; | |
4190 | union user_action_cookie cookie; | |
4191 | struct flow flow; | |
4192 | ||
4193 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4194 | &flow, NULL, &ofproto, NULL, NULL) | |
4195 | || !ofproto->ipfix) { | |
4196 | return; | |
4197 | } | |
4198 | ||
4199 | memset(&cookie, 0, sizeof cookie); | |
4200 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
4201 | ||
4202 | /* The flow reflects exactly the contents of the packet. Sample | |
4203 | * the packet using it. */ | |
4204 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
4205 | cookie.flow_sample.collector_set_id, | |
4206 | cookie.flow_sample.probability, | |
4207 | cookie.flow_sample.obs_domain_id, | |
4208 | cookie.flow_sample.obs_point_id); | |
4209 | } | |
4210 | ||
4211 | static void | |
4212 | handle_ipfix_upcall(struct dpif_backer *backer, | |
4213 | const struct dpif_upcall *upcall) | |
4214 | { | |
4215 | struct ofproto_dpif *ofproto; | |
4216 | struct flow flow; | |
4217 | ||
4218 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4219 | &flow, NULL, &ofproto, NULL, NULL) | |
4220 | || !ofproto->ipfix) { | |
4221 | return; | |
4222 | } | |
4223 | ||
4224 | /* The flow reflects exactly the contents of the packet. Sample | |
4225 | * the packet using it. */ | |
4226 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
4227 | } | |
4228 | ||
9b16c439 | 4229 | static int |
acf60855 | 4230 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 4231 | { |
9b16c439 | 4232 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
4233 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
4234 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
4235 | int n_processed; | |
9b16c439 BP |
4236 | int n_misses; |
4237 | int i; | |
abe529af | 4238 | |
cb22974d | 4239 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 4240 | |
9b16c439 | 4241 | n_misses = 0; |
90a7c55e | 4242 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 4243 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 4244 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
4245 | int error; |
4246 | ||
90a7c55e BP |
4247 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
4248 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 4249 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 4250 | if (error) { |
90a7c55e | 4251 | ofpbuf_uninit(buf); |
9b16c439 BP |
4252 | break; |
4253 | } | |
4254 | ||
6a7e895f BP |
4255 | switch (classify_upcall(upcall)) { |
4256 | case MISS_UPCALL: | |
9b16c439 BP |
4257 | /* Handle it later. */ |
4258 | n_misses++; | |
4259 | break; | |
4260 | ||
6a7e895f | 4261 | case SFLOW_UPCALL: |
acf60855 | 4262 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
4263 | ofpbuf_uninit(buf); |
4264 | break; | |
4265 | ||
29089a54 RL |
4266 | case FLOW_SAMPLE_UPCALL: |
4267 | handle_flow_sample_upcall(backer, upcall); | |
4268 | ofpbuf_uninit(buf); | |
4269 | break; | |
4270 | ||
4271 | case IPFIX_UPCALL: | |
4272 | handle_ipfix_upcall(backer, upcall); | |
4273 | ofpbuf_uninit(buf); | |
4274 | break; | |
4275 | ||
6a7e895f BP |
4276 | case BAD_UPCALL: |
4277 | ofpbuf_uninit(buf); | |
9b16c439 BP |
4278 | break; |
4279 | } | |
abe529af | 4280 | } |
9b16c439 | 4281 | |
6a7e895f | 4282 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 4283 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
4284 | for (i = 0; i < n_misses; i++) { |
4285 | ofpbuf_uninit(&miss_bufs[i]); | |
4286 | } | |
9b16c439 | 4287 | |
90a7c55e | 4288 | return n_processed; |
abe529af BP |
4289 | } |
4290 | \f | |
4291 | /* Flow expiration. */ | |
4292 | ||
04d08d54 | 4293 | static int subfacet_max_idle(const struct dpif_backer *); |
acf60855 | 4294 | static void update_stats(struct dpif_backer *); |
abe529af | 4295 | static void rule_expire(struct rule_dpif *); |
04d08d54 | 4296 | static void expire_subfacets(struct dpif_backer *, int dp_max_idle); |
abe529af BP |
4297 | |
4298 | /* This function is called periodically by run(). Its job is to collect | |
4299 | * updates for the flows that have been installed into the datapath, most | |
4300 | * importantly when they last were used, and then use that information to | |
4301 | * expire flows that have not been used recently. | |
4302 | * | |
4303 | * Returns the number of milliseconds after which it should be called again. */ | |
4304 | static int | |
acf60855 | 4305 | expire(struct dpif_backer *backer) |
abe529af | 4306 | { |
acf60855 | 4307 | struct ofproto_dpif *ofproto; |
dc54ef36 | 4308 | size_t n_subfacets; |
04d08d54 | 4309 | int max_idle; |
abe529af | 4310 | |
8f73d537 EJ |
4311 | /* Periodically clear out the drop keys in an effort to keep them |
4312 | * relatively few. */ | |
4313 | drop_key_clear(backer); | |
4314 | ||
acf60855 JP |
4315 | /* Update stats for each flow in the backer. */ |
4316 | update_stats(backer); | |
abe529af | 4317 | |
04d08d54 EJ |
4318 | n_subfacets = hmap_count(&backer->subfacets); |
4319 | if (n_subfacets) { | |
9fc0165a | 4320 | struct subfacet *subfacet; |
04d08d54 | 4321 | long long int total, now; |
abe529af | 4322 | |
04d08d54 EJ |
4323 | total = 0; |
4324 | now = time_msec(); | |
4325 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { | |
4326 | total += now - subfacet->created; | |
acf60855 | 4327 | } |
04d08d54 EJ |
4328 | backer->avg_subfacet_life += total / n_subfacets; |
4329 | } | |
4330 | backer->avg_subfacet_life /= 2; | |
0697b5c3 | 4331 | |
04d08d54 EJ |
4332 | backer->avg_n_subfacet += n_subfacets; |
4333 | backer->avg_n_subfacet /= 2; | |
4334 | ||
4335 | backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); | |
655ab909 | 4336 | |
04d08d54 EJ |
4337 | max_idle = subfacet_max_idle(backer); |
4338 | expire_subfacets(backer, max_idle); | |
acf60855 | 4339 | |
04d08d54 EJ |
4340 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
4341 | struct rule *rule, *next_rule; | |
4342 | ||
4343 | if (ofproto->backer != backer) { | |
4344 | continue; | |
4345 | } | |
acf60855 JP |
4346 | |
4347 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4348 | * has passed. */ | |
e503cc19 SH |
4349 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4350 | &ofproto->up.expirable) { | |
4351 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4352 | } |
abe529af | 4353 | |
acf60855 JP |
4354 | /* All outstanding data in existing flows has been accounted, so it's a |
4355 | * good time to do bond rebalancing. */ | |
4356 | if (ofproto->has_bonded_bundles) { | |
4357 | struct ofbundle *bundle; | |
abe529af | 4358 | |
acf60855 JP |
4359 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4360 | if (bundle->bond) { | |
2cc3c58e | 4361 | bond_rebalance(bundle->bond, &backer->revalidate_set); |
acf60855 | 4362 | } |
abe529af BP |
4363 | } |
4364 | } | |
4365 | } | |
4366 | ||
acf60855 | 4367 | return MIN(max_idle, 1000); |
abe529af BP |
4368 | } |
4369 | ||
a218c879 BP |
4370 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4371 | * as 'subfacet''s statistics. */ | |
4372 | static void | |
4373 | update_subfacet_stats(struct subfacet *subfacet, | |
4374 | const struct dpif_flow_stats *stats) | |
4375 | { | |
4376 | struct facet *facet = subfacet->facet; | |
787f3d59 | 4377 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
9dfb1f78 EJ |
4378 | struct dpif_flow_stats diff; |
4379 | ||
4380 | diff.tcp_flags = stats->tcp_flags; | |
4381 | diff.used = stats->used; | |
a218c879 BP |
4382 | |
4383 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
9dfb1f78 | 4384 | diff.n_packets = stats->n_packets - subfacet->dp_packet_count; |
a218c879 BP |
4385 | } else { |
4386 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
9dfb1f78 | 4387 | diff.n_packets = 0; |
a218c879 BP |
4388 | } |
4389 | ||
4390 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
9dfb1f78 | 4391 | diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; |
a218c879 BP |
4392 | } else { |
4393 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
9dfb1f78 | 4394 | diff.n_bytes = 0; |
a218c879 BP |
4395 | } |
4396 | ||
787f3d59 | 4397 | ofproto->n_hit += diff.n_packets; |
a218c879 BP |
4398 | subfacet->dp_packet_count = stats->n_packets; |
4399 | subfacet->dp_byte_count = stats->n_bytes; | |
9dfb1f78 | 4400 | subfacet_update_stats(subfacet, &diff); |
a218c879 | 4401 | |
a218c879 BP |
4402 | if (facet->accounted_bytes < facet->byte_count) { |
4403 | facet_learn(facet); | |
4404 | facet_account(facet); | |
4405 | facet->accounted_bytes = facet->byte_count; | |
4406 | } | |
a218c879 BP |
4407 | } |
4408 | ||
4409 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4410 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4411 | static void | |
04d08d54 | 4412 | delete_unexpected_flow(struct dpif_backer *backer, |
a218c879 BP |
4413 | const struct nlattr *key, size_t key_len) |
4414 | { | |
4415 | if (!VLOG_DROP_WARN(&rl)) { | |
4416 | struct ds s; | |
4417 | ||
4418 | ds_init(&s); | |
4419 | odp_flow_key_format(key, key_len, &s); | |
04d08d54 | 4420 | VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); |
a218c879 BP |
4421 | ds_destroy(&s); |
4422 | } | |
4423 | ||
4424 | COVERAGE_INC(facet_unexpected); | |
04d08d54 | 4425 | dpif_flow_del(backer->dpif, key, key_len, NULL); |
a218c879 BP |
4426 | } |
4427 | ||
abe529af BP |
4428 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4429 | * | |
4430 | * This function also pushes statistics updates to rules which each facet | |
4431 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4432 | * facet changes the rule it resubmits into at some time in between | |
4433 | * update_stats() runs, it is possible that statistics accrued to the | |
4434 | * old rule will be incorrectly attributed to the new rule. This could be | |
4435 | * avoided by calling update_stats() whenever rules are created or | |
4436 | * deleted. However, the performance impact of making so many calls to the | |
4437 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4438 | * |
4439 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4440 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4441 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4442 | * the hit or miss counts of br1. | |
abe529af BP |
4443 | */ |
4444 | static void | |
acf60855 | 4445 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4446 | { |
4447 | const struct dpif_flow_stats *stats; | |
4448 | struct dpif_flow_dump dump; | |
4449 | const struct nlattr *key; | |
4450 | size_t key_len; | |
4451 | ||
acf60855 | 4452 | dpif_flow_dump_start(&dump, backer->dpif); |
abe529af | 4453 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
b0f7b9b5 | 4454 | struct subfacet *subfacet; |
acf60855 | 4455 | uint32_t key_hash; |
abe529af | 4456 | |
acf60855 | 4457 | key_hash = odp_flow_key_hash(key, key_len); |
04d08d54 | 4458 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
6a7e895f BP |
4459 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4460 | case SF_FAST_PATH: | |
a218c879 | 4461 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4462 | break; |
4463 | ||
4464 | case SF_SLOW_PATH: | |
4465 | /* Stats are updated per-packet. */ | |
4466 | break; | |
4467 | ||
4468 | case SF_NOT_INSTALLED: | |
4469 | default: | |
04d08d54 | 4470 | delete_unexpected_flow(backer, key, key_len); |
6a7e895f | 4471 | break; |
abe529af | 4472 | } |
8fa4d1d0 | 4473 | run_fast_rl(); |
abe529af BP |
4474 | } |
4475 | dpif_flow_dump_done(&dump); | |
3af56aef | 4476 | |
dc54ef36 | 4477 | update_moving_averages(backer); |
abe529af BP |
4478 | } |
4479 | ||
4480 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4481 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4482 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4483 | * fold its statistic into its rule. */ | |
abe529af | 4484 | static int |
04d08d54 | 4485 | subfacet_max_idle(const struct dpif_backer *backer) |
abe529af BP |
4486 | { |
4487 | /* | |
4488 | * Idle time histogram. | |
4489 | * | |
b0f7b9b5 BP |
4490 | * Most of the time a switch has a relatively small number of subfacets. |
4491 | * When this is the case we might as well keep statistics for all of them | |
4492 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4493 | * well. |
4494 | * | |
b0f7b9b5 | 4495 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4496 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4497 | * significant. However, with a large number of subfacets it is likely |
4498 | * that only a few of them are "heavy hitters" that consume a large amount | |
4499 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4500 | * the kernel and maintaining in userspaces; other subfacets we can | |
4501 | * discard. | |
abe529af BP |
4502 | * |
4503 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4504 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4505 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4506 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4507 | * the most-recently-used 1% of subfacets (but at least |
380f49c4 | 4508 | * flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4509 | * the most-recently-used bucket of subfacets is kept, so actually an |
4510 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4511 | * (though the next run will delete most of those unless they receive |
4512 | * additional data). | |
abe529af | 4513 | * |
b0f7b9b5 BP |
4514 | * This requires a second pass through the subfacets, in addition to the |
4515 | * pass made by update_stats(), because the former function never looks at | |
4516 | * uninstallable subfacets. | |
abe529af BP |
4517 | */ |
4518 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4519 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4520 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4521 | int total, subtotal, bucket; |
b0f7b9b5 | 4522 | struct subfacet *subfacet; |
abe529af BP |
4523 | long long int now; |
4524 | int i; | |
4525 | ||
04d08d54 | 4526 | total = hmap_count(&backer->subfacets); |
380f49c4 | 4527 | if (total <= flow_eviction_threshold) { |
abe529af BP |
4528 | return N_BUCKETS * BUCKET_WIDTH; |
4529 | } | |
4530 | ||
4531 | /* Build histogram. */ | |
4532 | now = time_msec(); | |
04d08d54 | 4533 | HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { |
b0f7b9b5 | 4534 | long long int idle = now - subfacet->used; |
abe529af BP |
4535 | int bucket = (idle <= 0 ? 0 |
4536 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4537 | : (unsigned int) idle / BUCKET_WIDTH); | |
4538 | buckets[bucket]++; | |
4539 | } | |
4540 | ||
4541 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4542 | subtotal = bucket = 0; |
4543 | do { | |
4544 | subtotal += buckets[bucket++]; | |
084f5290 | 4545 | } while (bucket < N_BUCKETS && |
380f49c4 | 4546 | subtotal < MAX(flow_eviction_threshold, total / 100)); |
abe529af BP |
4547 | |
4548 | if (VLOG_IS_DBG_ENABLED()) { | |
4549 | struct ds s; | |
4550 | ||
4551 | ds_init(&s); | |
4552 | ds_put_cstr(&s, "keep"); | |
4553 | for (i = 0; i < N_BUCKETS; i++) { | |
4554 | if (i == bucket) { | |
4555 | ds_put_cstr(&s, ", drop"); | |
4556 | } | |
4557 | if (buckets[i]) { | |
4558 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4559 | } | |
4560 | } | |
04d08d54 | 4561 | VLOG_INFO("%s (msec:count)", ds_cstr(&s)); |
abe529af BP |
4562 | ds_destroy(&s); |
4563 | } | |
4564 | ||
4565 | return bucket * BUCKET_WIDTH; | |
4566 | } | |
4567 | ||
abe529af | 4568 | static void |
04d08d54 | 4569 | expire_subfacets(struct dpif_backer *backer, int dp_max_idle) |
abe529af | 4570 | { |
625b0720 BP |
4571 | /* Cutoff time for most flows. */ |
4572 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4573 | ||
4574 | /* We really want to keep flows for special protocols around, so use a more | |
4575 | * conservative cutoff. */ | |
4576 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4577 | |
b0f7b9b5 | 4578 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4579 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4580 | int n_batch; |
abe529af | 4581 | |
b99d3cee | 4582 | n_batch = 0; |
b0f7b9b5 | 4583 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
04d08d54 | 4584 | &backer->subfacets) { |
625b0720 BP |
4585 | long long int cutoff; |
4586 | ||
bbafd73b EJ |
4587 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
4588 | | SLOW_STP) | |
625b0720 BP |
4589 | ? special_cutoff |
4590 | : normal_cutoff); | |
b0f7b9b5 | 4591 | if (subfacet->used < cutoff) { |
6a7e895f | 4592 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4593 | batch[n_batch++] = subfacet; |
1d85f9e5 | 4594 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
04d08d54 | 4595 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee BP |
4596 | n_batch = 0; |
4597 | } | |
4598 | } else { | |
4599 | subfacet_destroy(subfacet); | |
4600 | } | |
abe529af BP |
4601 | } |
4602 | } | |
b99d3cee BP |
4603 | |
4604 | if (n_batch > 0) { | |
04d08d54 | 4605 | subfacet_destroy_batch(backer, batch, n_batch); |
b99d3cee | 4606 | } |
abe529af BP |
4607 | } |
4608 | ||
4609 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4610 | * then delete it entirely. */ | |
4611 | static void | |
4612 | rule_expire(struct rule_dpif *rule) | |
4613 | { | |
abe529af BP |
4614 | struct facet *facet, *next_facet; |
4615 | long long int now; | |
4616 | uint8_t reason; | |
4617 | ||
e2a3d183 BP |
4618 | if (rule->up.pending) { |
4619 | /* We'll have to expire it later. */ | |
4620 | return; | |
4621 | } | |
4622 | ||
abe529af BP |
4623 | /* Has 'rule' expired? */ |
4624 | now = time_msec(); | |
4625 | if (rule->up.hard_timeout | |
308881af | 4626 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4627 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4628 | } else if (rule->up.idle_timeout |
1745cd08 | 4629 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4630 | reason = OFPRR_IDLE_TIMEOUT; |
4631 | } else { | |
4632 | return; | |
4633 | } | |
4634 | ||
4635 | COVERAGE_INC(ofproto_dpif_expired); | |
4636 | ||
4637 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4638 | * timeout, because that only happens when the rule has no facets left.) */ | |
4639 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 4640 | facet_remove(facet); |
abe529af BP |
4641 | } |
4642 | ||
4643 | /* Get rid of the rule. */ | |
4644 | ofproto_rule_expire(&rule->up, reason); | |
4645 | } | |
4646 | \f | |
4647 | /* Facets. */ | |
4648 | ||
4dff9097 | 4649 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
4650 | * |
4651 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 4652 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 4653 | * |
4dff9097 | 4654 | * 'hash' must be the return value of flow_hash(miss->flow, 0). |
2b459b83 | 4655 | * |
b0f7b9b5 BP |
4656 | * The facet will initially have no subfacets. The caller should create (at |
4657 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4658 | static struct facet * |
4dff9097 | 4659 | facet_create(const struct flow_miss *miss, uint32_t hash) |
abe529af | 4660 | { |
4dff9097 | 4661 | struct ofproto_dpif *ofproto = miss->ofproto; |
bbafd73b | 4662 | struct xlate_in xin; |
abe529af BP |
4663 | struct facet *facet; |
4664 | ||
4665 | facet = xzalloc(sizeof *facet); | |
4666 | facet->used = time_msec(); | |
4dff9097 EJ |
4667 | facet->flow = miss->flow; |
4668 | facet->initial_vals = miss->initial_vals; | |
4669 | facet->rule = rule_dpif_lookup(ofproto, &facet->flow); | |
4670 | facet->learn_rl = time_msec() + 500; | |
4671 | ||
2b459b83 | 4672 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
4dff9097 | 4673 | list_push_back(&facet->rule->facets, &facet->list_node); |
b0f7b9b5 | 4674 | list_init(&facet->subfacets); |
abe529af BP |
4675 | netflow_flow_init(&facet->nf_flow); |
4676 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4677 | ||
bbafd73b EJ |
4678 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, |
4679 | facet->rule, 0, NULL); | |
4680 | xin.may_learn = true; | |
4681 | xlate_actions(&xin, &facet->xout); | |
4682 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
6cf474d7 | 4683 | |
abe529af BP |
4684 | return facet; |
4685 | } | |
4686 | ||
4687 | static void | |
4688 | facet_free(struct facet *facet) | |
4689 | { | |
4dff9097 | 4690 | if (facet) { |
bbafd73b | 4691 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
4692 | free(facet); |
4693 | } | |
abe529af BP |
4694 | } |
4695 | ||
3d9e05f8 | 4696 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4697 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4698 | static bool |
4699 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4700 | const struct nlattr *odp_actions, size_t actions_len, | |
4701 | struct ofpbuf *packet) | |
4702 | { | |
4703 | struct odputil_keybuf keybuf; | |
4704 | struct ofpbuf key; | |
4705 | int error; | |
4706 | ||
6ff686f2 | 4707 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
4708 | odp_flow_key_from_flow(&key, flow, |
4709 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
80e5eed9 | 4710 | |
acf60855 | 4711 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4712 | odp_actions, actions_len, packet); |
6ff686f2 | 4713 | return !error; |
abe529af BP |
4714 | } |
4715 | ||
abe529af BP |
4716 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
4717 | * | |
4718 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4719 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4720 | * |
4721 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4722 | */ | |
4723 | static void | |
15baa734 | 4724 | facet_remove(struct facet *facet) |
abe529af | 4725 | { |
15baa734 | 4726 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4727 | struct subfacet *subfacet, *next_subfacet; |
4728 | ||
cb22974d | 4729 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4730 | |
4731 | /* First uninstall all of the subfacets to get final statistics. */ | |
4732 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4733 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4734 | } |
4735 | ||
4736 | /* Flush the final stats to the rule. | |
4737 | * | |
4738 | * This might require us to have at least one subfacet around so that we | |
4739 | * can use its actions for accounting in facet_account(), which is why we | |
4740 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4741 | facet_flush_stats(facet); |
551a2f6c BP |
4742 | |
4743 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4744 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4745 | &facet->subfacets) { | |
15baa734 | 4746 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4747 | } |
abe529af BP |
4748 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
4749 | list_remove(&facet->list_node); | |
4750 | facet_free(facet); | |
4751 | } | |
4752 | ||
3de9590b BP |
4753 | /* Feed information from 'facet' back into the learning table to keep it in |
4754 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4755 | static void |
3de9590b | 4756 | facet_learn(struct facet *facet) |
abe529af | 4757 | { |
62bd7349 | 4758 | long long int now = time_msec(); |
abe529af | 4759 | |
bbafd73b | 4760 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
4761 | return; |
4762 | } | |
4763 | ||
62bd7349 | 4764 | facet->learn_rl = now + 500; |
6cf474d7 | 4765 | |
bbafd73b EJ |
4766 | if (!facet->xout.has_learn |
4767 | && !facet->xout.has_normal | |
4768 | && (!facet->xout.has_fin_timeout | |
3de9590b | 4769 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
4770 | return; |
4771 | } | |
abe529af | 4772 | |
9dfb1f78 | 4773 | facet_push_stats(facet, true); |
3de9590b BP |
4774 | } |
4775 | ||
4776 | static void | |
4777 | facet_account(struct facet *facet) | |
4778 | { | |
4779 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3de9590b BP |
4780 | const struct nlattr *a; |
4781 | unsigned int left; | |
4782 | ovs_be16 vlan_tci; | |
4783 | uint64_t n_bytes; | |
abe529af | 4784 | |
bbafd73b | 4785 | if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
4786 | return; |
4787 | } | |
3de9590b | 4788 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4789 | |
4790 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4791 | * as a basis. We also need to track the actual VLAN on which the packet | |
4792 | * is going to be sent to ensure that it matches the one passed to | |
4793 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4794 | * hash bucket.) |
4795 | * | |
4796 | * We use the actions from an arbitrary subfacet because they should all | |
4797 | * be equally valid for our purpose. */ | |
d78be13b | 4798 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
4799 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
4800 | facet->xout.odp_actions.size) { | |
fea393b1 | 4801 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4802 | struct ofport_dpif *port; |
abe529af | 4803 | |
d78be13b | 4804 | switch (nl_attr_type(a)) { |
df2c07f4 | 4805 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
4806 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
4807 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 4808 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4809 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4810 | } |
d78be13b BP |
4811 | break; |
4812 | ||
fea393b1 BP |
4813 | case OVS_ACTION_ATTR_POP_VLAN: |
4814 | vlan_tci = htons(0); | |
d78be13b BP |
4815 | break; |
4816 | ||
fea393b1 BP |
4817 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4818 | vlan = nl_attr_get(a); | |
4819 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4820 | break; |
abe529af BP |
4821 | } |
4822 | } | |
4823 | } | |
4824 | ||
abe529af BP |
4825 | /* Returns true if the only action for 'facet' is to send to the controller. |
4826 | * (We don't report NetFlow expiration messages for such facets because they | |
4827 | * are just part of the control logic for the network, not real traffic). */ | |
4828 | static bool | |
4829 | facet_is_controller_flow(struct facet *facet) | |
4830 | { | |
f25d0cf3 BP |
4831 | if (facet) { |
4832 | const struct rule *rule = &facet->rule->up; | |
4833 | const struct ofpact *ofpacts = rule->ofpacts; | |
4834 | size_t ofpacts_len = rule->ofpacts_len; | |
4835 | ||
dd30ff28 BP |
4836 | if (ofpacts_len > 0 && |
4837 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4838 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4839 | return true; | |
4840 | } | |
4841 | } | |
4842 | return false; | |
abe529af BP |
4843 | } |
4844 | ||
4845 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4846 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4847 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4848 | * its packet and byte counts before this function is called. */ | |
4849 | static void | |
15baa734 | 4850 | facet_flush_stats(struct facet *facet) |
abe529af | 4851 | { |
15baa734 | 4852 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4853 | struct subfacet *subfacet; |
4854 | ||
4855 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4856 | ovs_assert(!subfacet->dp_byte_count); |
4857 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4858 | } |
abe529af | 4859 | |
9dfb1f78 | 4860 | facet_push_stats(facet, false); |
3de9590b BP |
4861 | if (facet->accounted_bytes < facet->byte_count) { |
4862 | facet_account(facet); | |
4863 | facet->accounted_bytes = facet->byte_count; | |
4864 | } | |
abe529af BP |
4865 | |
4866 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4867 | struct ofexpired expired; | |
4868 | expired.flow = facet->flow; | |
4869 | expired.packet_count = facet->packet_count; | |
4870 | expired.byte_count = facet->byte_count; | |
4871 | expired.used = facet->used; | |
4872 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4873 | } | |
4874 | ||
abe529af BP |
4875 | /* Reset counters to prevent double counting if 'facet' ever gets |
4876 | * reinstalled. */ | |
bbb5d219 | 4877 | facet_reset_counters(facet); |
abe529af BP |
4878 | |
4879 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4880 | facet->tcp_flags = 0; |
abe529af BP |
4881 | } |
4882 | ||
4883 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4884 | * Returns it if found, otherwise a null pointer. | |
4885 | * | |
2b459b83 BP |
4886 | * 'hash' must be the return value of flow_hash(flow, 0). |
4887 | * | |
abe529af BP |
4888 | * The returned facet might need revalidation; use facet_lookup_valid() |
4889 | * instead if that is important. */ | |
4890 | static struct facet * | |
2b459b83 BP |
4891 | facet_find(struct ofproto_dpif *ofproto, |
4892 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
4893 | { |
4894 | struct facet *facet; | |
4895 | ||
2b459b83 | 4896 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
4897 | if (flow_equal(flow, &facet->flow)) { |
4898 | return facet; | |
4899 | } | |
4900 | } | |
4901 | ||
4902 | return NULL; | |
4903 | } | |
4904 | ||
4905 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4906 | * Returns it if found, otherwise a null pointer. | |
4907 | * | |
2b459b83 BP |
4908 | * 'hash' must be the return value of flow_hash(flow, 0). |
4909 | * | |
abe529af BP |
4910 | * The returned facet is guaranteed to be valid. */ |
4911 | static struct facet * | |
2b459b83 BP |
4912 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
4913 | uint32_t hash) | |
abe529af | 4914 | { |
c57b2226 | 4915 | struct facet *facet; |
abe529af | 4916 | |
c57b2226 | 4917 | facet = facet_find(ofproto, flow, hash); |
abe529af | 4918 | if (facet |
2cc3c58e EJ |
4919 | && (ofproto->backer->need_revalidate |
4920 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
bbafd73b | 4921 | facet->xout.tags)) |
5bf64ade | 4922 | && !facet_revalidate(facet)) { |
0305ce1f | 4923 | return NULL; |
abe529af BP |
4924 | } |
4925 | ||
4926 | return facet; | |
4927 | } | |
4928 | ||
6814e51f BP |
4929 | static bool |
4930 | facet_check_consistency(struct facet *facet) | |
4931 | { | |
4932 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4933 | ||
4934 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4935 | ||
bbafd73b EJ |
4936 | struct xlate_out xout; |
4937 | struct xlate_in xin; | |
050ac423 | 4938 | |
6814e51f | 4939 | struct rule_dpif *rule; |
6814e51f BP |
4940 | bool ok; |
4941 | ||
4942 | /* Check the rule for consistency. */ | |
c57b2226 | 4943 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
4dff9097 EJ |
4944 | if (rule != facet->rule) { |
4945 | if (!VLOG_DROP_WARN(&rl)) { | |
4946 | struct ds s = DS_EMPTY_INITIALIZER; | |
6814e51f | 4947 | |
c53e1132 BP |
4948 | flow_format(&s, &facet->flow); |
4949 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4950 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4951 | cls_rule_format(&facet->rule->up.cr, &s); | |
4952 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4953 | rule->up.table_id); | |
4954 | cls_rule_format(&rule->up.cr, &s); | |
c0a71f4e | 4955 | ds_put_char(&s, ')'); |
6814e51f | 4956 | |
c0a71f4e | 4957 | VLOG_WARN("%s", ds_cstr(&s)); |
c53e1132 BP |
4958 | ds_destroy(&s); |
4959 | } | |
4dff9097 | 4960 | return false; |
6814e51f BP |
4961 | } |
4962 | ||
4963 | /* Check the datapath actions for consistency. */ | |
bbafd73b EJ |
4964 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule, |
4965 | 0, NULL); | |
4966 | xlate_actions(&xin, &xout); | |
6814e51f | 4967 | |
bbafd73b EJ |
4968 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
4969 | && facet->xout.slow == xout.slow; | |
4dff9097 EJ |
4970 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4971 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4972 | |
4dff9097 EJ |
4973 | flow_format(&s, &facet->flow); |
4974 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4975 | |
bbafd73b | 4976 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4977 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4978 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4979 | facet->xout.odp_actions.size); | |
9616614b | 4980 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4981 | format_odp_actions(&s, xout.odp_actions.data, |
4982 | xout.odp_actions.size); | |
c0a71f4e | 4983 | ds_put_char(&s, ')'); |
6814e51f | 4984 | } |
4dff9097 | 4985 | |
bbafd73b EJ |
4986 | if (facet->xout.slow != xout.slow) { |
4987 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4988 | } |
4989 | ||
c0a71f4e | 4990 | VLOG_WARN("%s", ds_cstr(&s)); |
9616614b | 4991 | ds_destroy(&s); |
6814e51f | 4992 | } |
bbafd73b | 4993 | xlate_out_uninit(&xout); |
6814e51f BP |
4994 | |
4995 | return ok; | |
4996 | } | |
4997 | ||
15baa734 | 4998 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4999 | * |
5000 | * - If the rule found is different from 'facet''s current rule, moves | |
5001 | * 'facet' to the new rule and recompiles its actions. | |
5002 | * | |
5003 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
5004 | * where it is and recompiles its actions anyway. |
5005 | * | |
5006 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
5bf64ade EJ |
5007 | * ofproto_receive(), 'facet' is removed. |
5008 | * | |
5009 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
5010 | static bool | |
15baa734 | 5011 | facet_revalidate(struct facet *facet) |
abe529af | 5012 | { |
15baa734 | 5013 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af | 5014 | struct rule_dpif *new_rule; |
b0f7b9b5 | 5015 | struct subfacet *subfacet; |
bbafd73b EJ |
5016 | struct xlate_out xout; |
5017 | struct xlate_in xin; | |
abe529af BP |
5018 | |
5019 | COVERAGE_INC(facet_revalidate); | |
5020 | ||
f231418e EJ |
5021 | /* Check that child subfacets still correspond to this facet. Tunnel |
5022 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
5023 | * change. */ | |
5024 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5025 | struct ofproto_dpif *recv_ofproto; | |
5026 | struct flow recv_flow; | |
5027 | int error; | |
5028 | ||
5029 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
5030 | subfacet->key_len, &recv_flow, NULL, | |
5031 | &recv_ofproto, NULL, NULL); | |
5032 | if (error | |
5033 | || recv_ofproto != ofproto | |
5034 | || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { | |
5035 | facet_remove(facet); | |
5bf64ade | 5036 | return false; |
f231418e EJ |
5037 | } |
5038 | } | |
5039 | ||
c57b2226 | 5040 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 5041 | |
df2c07f4 | 5042 | /* Calculate new datapath actions. |
abe529af BP |
5043 | * |
5044 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
5045 | * emit a NetFlow expiration and, if so, we need to have the old state | |
5046 | * around to properly compose it. */ | |
bbafd73b EJ |
5047 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule, |
5048 | 0, NULL); | |
5049 | xlate_actions(&xin, &xout); | |
4dff9097 EJ |
5050 | |
5051 | /* A facet's slow path reason should only change under dramatic | |
5052 | * circumstances. Rather than try to update everything, it's simpler to | |
5053 | * remove the facet and start over. */ | |
bbafd73b | 5054 | if (facet->xout.slow != xout.slow) { |
4dff9097 | 5055 | facet_remove(facet); |
bbafd73b | 5056 | xlate_out_uninit(&xout); |
4dff9097 EJ |
5057 | return false; |
5058 | } | |
6a7e895f | 5059 | |
bbafd73b | 5060 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
5061 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
5062 | if (subfacet->path == SF_FAST_PATH) { | |
5063 | struct dpif_flow_stats stats; | |
b95fc6ba | 5064 | |
bbafd73b | 5065 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 5066 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 5067 | } |
abe529af | 5068 | } |
b95fc6ba | 5069 | |
15baa734 | 5070 | facet_flush_stats(facet); |
4dff9097 | 5071 | |
bbafd73b EJ |
5072 | ofpbuf_clear(&facet->xout.odp_actions); |
5073 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
5074 | xout.odp_actions.size); | |
abe529af BP |
5075 | } |
5076 | ||
5077 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
5078 | facet->xout.tags = xout.tags; |
5079 | facet->xout.slow = xout.slow; | |
5080 | facet->xout.has_learn = xout.has_learn; | |
5081 | facet->xout.has_normal = xout.has_normal; | |
5082 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
5083 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
5084 | facet->xout.mirrors = xout.mirrors; | |
5085 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
6a7e895f | 5086 | |
abe529af BP |
5087 | if (facet->rule != new_rule) { |
5088 | COVERAGE_INC(facet_changed_rule); | |
5089 | list_remove(&facet->list_node); | |
5090 | list_push_back(&new_rule->facets, &facet->list_node); | |
5091 | facet->rule = new_rule; | |
5092 | facet->used = new_rule->up.created; | |
9d24de3b | 5093 | facet->prev_used = facet->used; |
abe529af | 5094 | } |
5bf64ade | 5095 | |
bbafd73b | 5096 | xlate_out_uninit(&xout); |
5bf64ade | 5097 | return true; |
abe529af BP |
5098 | } |
5099 | ||
bbb5d219 EJ |
5100 | static void |
5101 | facet_reset_counters(struct facet *facet) | |
5102 | { | |
5103 | facet->packet_count = 0; | |
5104 | facet->byte_count = 0; | |
9d24de3b JP |
5105 | facet->prev_packet_count = 0; |
5106 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
5107 | facet->accounted_bytes = 0; |
5108 | } | |
5109 | ||
abe529af | 5110 | static void |
9dfb1f78 | 5111 | facet_push_stats(struct facet *facet, bool may_learn) |
abe529af | 5112 | { |
112bc5f4 | 5113 | struct dpif_flow_stats stats; |
abe529af | 5114 | |
cb22974d BP |
5115 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
5116 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
5117 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 5118 | |
112bc5f4 BP |
5119 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
5120 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
5121 | stats.used = facet->used; | |
9dfb1f78 EJ |
5122 | stats.tcp_flags = facet->tcp_flags; |
5123 | ||
5124 | if (may_learn || stats.n_packets || facet->used > facet->prev_used) { | |
5125 | struct ofproto_dpif *ofproto = | |
5126 | ofproto_dpif_cast(facet->rule->up.ofproto); | |
5127 | ||
5128 | struct ofport_dpif *in_port; | |
5129 | struct xlate_in xin; | |
abe529af | 5130 | |
9d24de3b JP |
5131 | facet->prev_packet_count = facet->packet_count; |
5132 | facet->prev_byte_count = facet->byte_count; | |
5133 | facet->prev_used = facet->used; | |
abe529af | 5134 | |
9dfb1f78 EJ |
5135 | in_port = get_ofp_port(ofproto, facet->flow.in_port); |
5136 | if (in_port && in_port->tnl_port) { | |
5137 | netdev_vport_inc_rx(in_port->up.netdev, &stats); | |
5138 | } | |
9d24de3b | 5139 | |
9dfb1f78 EJ |
5140 | rule_credit_stats(facet->rule, &stats); |
5141 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, | |
5142 | facet->used); | |
5143 | netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); | |
5144 | update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets, | |
bbafd73b | 5145 | stats.n_bytes); |
9dfb1f78 EJ |
5146 | |
5147 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, | |
5148 | facet->rule, stats.tcp_flags, NULL); | |
5149 | xin.resubmit_stats = &stats; | |
5150 | xin.may_learn = may_learn; | |
5151 | xlate_actions_for_side_effects(&xin); | |
abe529af BP |
5152 | } |
5153 | } | |
5154 | ||
8844e035 | 5155 | static void |
4464589e | 5156 | push_all_stats__(bool run_fast) |
8844e035 EJ |
5157 | { |
5158 | static long long int rl = LLONG_MIN; | |
5159 | struct ofproto_dpif *ofproto; | |
5160 | ||
5161 | if (time_msec() < rl) { | |
5162 | return; | |
5163 | } | |
5164 | ||
5165 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5166 | struct facet *facet; | |
5167 | ||
5168 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
9dfb1f78 | 5169 | facet_push_stats(facet, false); |
4464589e EJ |
5170 | if (run_fast) { |
5171 | run_fast_rl(); | |
5172 | } | |
8844e035 EJ |
5173 | } |
5174 | } | |
5175 | ||
5176 | rl = time_msec() + 100; | |
5177 | } | |
5178 | ||
4464589e EJ |
5179 | static void |
5180 | push_all_stats(void) | |
5181 | { | |
5182 | push_all_stats__(true); | |
5183 | } | |
5184 | ||
abe529af | 5185 | static void |
112bc5f4 | 5186 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 5187 | { |
112bc5f4 BP |
5188 | rule->packet_count += stats->n_packets; |
5189 | rule->byte_count += stats->n_bytes; | |
5190 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af | 5191 | } |
abe529af | 5192 | \f |
b0f7b9b5 BP |
5193 | /* Subfacets. */ |
5194 | ||
5195 | static struct subfacet * | |
04d08d54 EJ |
5196 | subfacet_find(struct dpif_backer *backer, const struct nlattr *key, |
5197 | size_t key_len, uint32_t key_hash) | |
b0f7b9b5 BP |
5198 | { |
5199 | struct subfacet *subfacet; | |
5200 | ||
5201 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
04d08d54 | 5202 | &backer->subfacets) { |
9566abf9 EJ |
5203 | if (subfacet->key_len == key_len |
5204 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
5205 | return subfacet; |
5206 | } | |
5207 | } | |
5208 | ||
5209 | return NULL; | |
5210 | } | |
5211 | ||
5212 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
5213 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
5214 | * existing subfacet if there is one, otherwise creates and returns a | |
4dff9097 | 5215 | * new subfacet. */ |
b0f7b9b5 | 5216 | static struct subfacet * |
a088a1ff JP |
5217 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
5218 | long long int now) | |
b0f7b9b5 | 5219 | { |
04d08d54 | 5220 | struct dpif_backer *backer = miss->ofproto->backer; |
a088a1ff JP |
5221 | enum odp_key_fitness key_fitness = miss->key_fitness; |
5222 | const struct nlattr *key = miss->key; | |
5223 | size_t key_len = miss->key_len; | |
5224 | uint32_t key_hash; | |
b0f7b9b5 BP |
5225 | struct subfacet *subfacet; |
5226 | ||
a088a1ff JP |
5227 | key_hash = odp_flow_key_hash(key, key_len); |
5228 | ||
3b145dd7 BP |
5229 | if (list_is_empty(&facet->subfacets)) { |
5230 | subfacet = &facet->one_subfacet; | |
5231 | } else { | |
04d08d54 | 5232 | subfacet = subfacet_find(backer, key, key_len, key_hash); |
3b145dd7 BP |
5233 | if (subfacet) { |
5234 | if (subfacet->facet == facet) { | |
5235 | return subfacet; | |
5236 | } | |
5237 | ||
5238 | /* This shouldn't happen. */ | |
5239 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5240 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
5241 | } |
5242 | ||
3b145dd7 | 5243 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
5244 | } |
5245 | ||
04d08d54 | 5246 | hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); |
b0f7b9b5 BP |
5247 | list_push_back(&facet->subfacets, &subfacet->list_node); |
5248 | subfacet->facet = facet; | |
b0f7b9b5 | 5249 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
5250 | subfacet->key = xmemdup(key, key_len); |
5251 | subfacet->key_len = key_len; | |
459b16a1 | 5252 | subfacet->used = now; |
655ab909 | 5253 | subfacet->created = now; |
26cd7e34 BP |
5254 | subfacet->dp_packet_count = 0; |
5255 | subfacet->dp_byte_count = 0; | |
6a7e895f | 5256 | subfacet->path = SF_NOT_INSTALLED; |
04d08d54 | 5257 | subfacet->backer = backer; |
b0f7b9b5 | 5258 | |
04d08d54 | 5259 | backer->subfacet_add_count++; |
b0f7b9b5 BP |
5260 | return subfacet; |
5261 | } | |
5262 | ||
b0f7b9b5 BP |
5263 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
5264 | * its facet within 'ofproto', and frees it. */ | |
5265 | static void | |
15baa734 | 5266 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 5267 | { |
15baa734 BP |
5268 | struct facet *facet = subfacet->facet; |
5269 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5270 | ||
655ab909 | 5271 | /* Update ofproto stats before uninstall the subfacet. */ |
dc54ef36 | 5272 | ofproto->backer->subfacet_del_count++; |
655ab909 | 5273 | |
15baa734 | 5274 | subfacet_uninstall(subfacet); |
04d08d54 | 5275 | hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); |
b0f7b9b5 BP |
5276 | list_remove(&subfacet->list_node); |
5277 | free(subfacet->key); | |
26cd7e34 BP |
5278 | if (subfacet != &facet->one_subfacet) { |
5279 | free(subfacet); | |
5280 | } | |
b0f7b9b5 BP |
5281 | } |
5282 | ||
5283 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5284 | * last remaining subfacet in its facet destroys the facet too. */ | |
5285 | static void | |
15baa734 | 5286 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
5287 | { |
5288 | struct facet *facet = subfacet->facet; | |
5289 | ||
551a2f6c BP |
5290 | if (list_is_singleton(&facet->subfacets)) { |
5291 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 5292 | facet_remove(facet); |
551a2f6c | 5293 | } else { |
15baa734 | 5294 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
5295 | } |
5296 | } | |
5297 | ||
1d85f9e5 | 5298 | static void |
04d08d54 | 5299 | subfacet_destroy_batch(struct dpif_backer *backer, |
1d85f9e5 JP |
5300 | struct subfacet **subfacets, int n) |
5301 | { | |
1d85f9e5 JP |
5302 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
5303 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
5304 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
5305 | int i; | |
5306 | ||
5307 | for (i = 0; i < n; i++) { | |
5308 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
5309 | ops[i].u.flow_del.key = subfacets[i]->key; |
5310 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
5311 | ops[i].u.flow_del.stats = &stats[i]; |
5312 | opsp[i] = &ops[i]; | |
5313 | } | |
5314 | ||
04d08d54 | 5315 | dpif_operate(backer->dpif, opsp, n); |
1d85f9e5 JP |
5316 | for (i = 0; i < n; i++) { |
5317 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5318 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5319 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 5320 | run_fast_rl(); |
1d85f9e5 JP |
5321 | } |
5322 | } | |
5323 | ||
b0f7b9b5 BP |
5324 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5325 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5326 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5327 | * since 'subfacet' was last updated. | |
5328 | * | |
5329 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5330 | static int | |
4dff9097 EJ |
5331 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
5332 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 5333 | { |
15baa734 BP |
5334 | struct facet *facet = subfacet->facet; |
5335 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
bbafd73b | 5336 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
5337 | const struct nlattr *actions = odp_actions->data; |
5338 | size_t actions_len = odp_actions->size; | |
5339 | ||
6a7e895f | 5340 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 5341 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
5342 | int ret; |
5343 | ||
5344 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5345 | if (stats) { | |
5346 | flags |= DPIF_FP_ZERO_STATS; | |
5347 | } | |
5348 | ||
6a7e895f | 5349 | if (path == SF_SLOW_PATH) { |
bbafd73b | 5350 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
5351 | slow_path_stub, sizeof slow_path_stub, |
5352 | &actions, &actions_len); | |
5353 | } | |
5354 | ||
04d08d54 | 5355 | ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, |
9566abf9 | 5356 | subfacet->key_len, actions, actions_len, stats); |
b0f7b9b5 BP |
5357 | |
5358 | if (stats) { | |
5359 | subfacet_reset_dp_stats(subfacet, stats); | |
5360 | } | |
5361 | ||
6a7e895f BP |
5362 | if (!ret) { |
5363 | subfacet->path = path; | |
5364 | } | |
b0f7b9b5 BP |
5365 | return ret; |
5366 | } | |
5367 | ||
5368 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
5369 | static void | |
15baa734 | 5370 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5371 | { |
6a7e895f | 5372 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
5373 | struct rule_dpif *rule = subfacet->facet->rule; |
5374 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 | 5375 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5376 | int error; |
5377 | ||
9566abf9 EJ |
5378 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5379 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5380 | subfacet_reset_dp_stats(subfacet, &stats); |
5381 | if (!error) { | |
15baa734 | 5382 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5383 | } |
6a7e895f | 5384 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5385 | } else { |
cb22974d BP |
5386 | ovs_assert(subfacet->dp_packet_count == 0); |
5387 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5388 | } |
5389 | } | |
5390 | ||
5391 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5392 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5393 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5394 | * was reset in the datapath. 'stats' will be modified to include only | |
5395 | * statistics new since 'subfacet' was last updated. */ | |
5396 | static void | |
5397 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5398 | struct dpif_flow_stats *stats) | |
5399 | { | |
5400 | if (stats | |
5401 | && subfacet->dp_packet_count <= stats->n_packets | |
5402 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5403 | stats->n_packets -= subfacet->dp_packet_count; | |
5404 | stats->n_bytes -= subfacet->dp_byte_count; | |
5405 | } | |
5406 | ||
5407 | subfacet->dp_packet_count = 0; | |
5408 | subfacet->dp_byte_count = 0; | |
5409 | } | |
5410 | ||
b0f7b9b5 BP |
5411 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. |
5412 | * | |
5413 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5414 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5415 | * represents a packet that was sent by hand or if it represents statistics | |
5416 | * that have been cleared out of the datapath. */ | |
5417 | static void | |
15baa734 | 5418 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5419 | const struct dpif_flow_stats *stats) |
5420 | { | |
5421 | if (stats->n_packets || stats->used > subfacet->used) { | |
5422 | struct facet *facet = subfacet->facet; | |
5423 | ||
9dfb1f78 EJ |
5424 | subfacet->used = MAX(subfacet->used, stats->used); |
5425 | facet->used = MAX(facet->used, stats->used); | |
b0f7b9b5 BP |
5426 | facet->packet_count += stats->n_packets; |
5427 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5428 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5429 | } |
5430 | } | |
5431 | \f | |
abe529af BP |
5432 | /* Rules. */ |
5433 | ||
5434 | static struct rule_dpif * | |
c57b2226 BP |
5435 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
5436 | { | |
c57b2226 BP |
5437 | struct rule_dpif *rule; |
5438 | ||
5439 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5440 | if (rule) { | |
5441 | return rule; | |
5442 | } | |
5443 | ||
c376f9a3 | 5444 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5445 | } |
5446 | ||
5447 | static struct rule_dpif * | |
5448 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5449 | uint8_t table_id) | |
abe529af | 5450 | { |
7257b535 BP |
5451 | struct cls_rule *cls_rule; |
5452 | struct classifier *cls; | |
7fd51d39 | 5453 | bool frag; |
7257b535 | 5454 | |
9cdaaebe BP |
5455 | if (table_id >= N_TABLES) { |
5456 | return NULL; | |
5457 | } | |
5458 | ||
d0918789 | 5459 | cls = &ofproto->up.tables[table_id].cls; |
7fd51d39 BP |
5460 | frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; |
5461 | if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
5462 | /* We must pretend that transport ports are unavailable. */ | |
7257b535 BP |
5463 | struct flow ofpc_normal_flow = *flow; |
5464 | ofpc_normal_flow.tp_src = htons(0); | |
5465 | ofpc_normal_flow.tp_dst = htons(0); | |
74f74083 | 5466 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow, NULL); |
7fd51d39 BP |
5467 | } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { |
5468 | cls_rule = &ofproto->drop_frags_rule->up.cr; | |
7257b535 | 5469 | } else { |
74f74083 | 5470 | cls_rule = classifier_lookup(cls, flow, NULL); |
7257b535 BP |
5471 | } |
5472 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5473 | } |
5474 | ||
c376f9a3 IY |
5475 | static struct rule_dpif * |
5476 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5477 | { | |
5478 | struct ofport_dpif *port; | |
5479 | ||
5480 | port = get_ofp_port(ofproto, flow->in_port); | |
5481 | if (!port) { | |
5482 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5483 | return ofproto->miss_rule; | |
5484 | } | |
5485 | ||
5486 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5487 | return ofproto->no_packet_in_rule; | |
5488 | } | |
5489 | return ofproto->miss_rule; | |
5490 | } | |
5491 | ||
7ee20df1 BP |
5492 | static void |
5493 | complete_operation(struct rule_dpif *rule) | |
5494 | { | |
5495 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5496 | ||
54a9cbc9 | 5497 | rule_invalidate(rule); |
7ee20df1 BP |
5498 | if (clogged) { |
5499 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5500 | c->op = rule->up.pending; | |
5501 | list_push_back(&ofproto->completions, &c->list_node); | |
5502 | } else { | |
5503 | ofoperation_complete(rule->up.pending, 0); | |
5504 | } | |
5505 | } | |
5506 | ||
abe529af BP |
5507 | static struct rule * |
5508 | rule_alloc(void) | |
5509 | { | |
5510 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5511 | return &rule->up; | |
5512 | } | |
5513 | ||
5514 | static void | |
5515 | rule_dealloc(struct rule *rule_) | |
5516 | { | |
5517 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5518 | free(rule); | |
5519 | } | |
5520 | ||
90bf1e07 | 5521 | static enum ofperr |
abe529af BP |
5522 | rule_construct(struct rule *rule_) |
5523 | { | |
5524 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5525 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 5526 | struct rule_dpif *victim; |
54a9cbc9 | 5527 | uint8_t table_id; |
abe529af | 5528 | |
abe529af BP |
5529 | rule->packet_count = 0; |
5530 | rule->byte_count = 0; | |
abe529af | 5531 | |
7ee20df1 BP |
5532 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
5533 | if (victim && !list_is_empty(&victim->facets)) { | |
5534 | struct facet *facet; | |
5535 | ||
5536 | rule->facets = victim->facets; | |
5537 | list_moved(&rule->facets); | |
5538 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
5539 | /* XXX: We're only clearing our local counters here. It's possible |
5540 | * that quite a few packets are unaccounted for in the datapath | |
5541 | * statistics. These will be accounted to the new rule instead of | |
5542 | * cleared as required. This could be fixed by clearing out the | |
5543 | * datapath statistics for this facet, but currently it doesn't | |
5544 | * seem worth it. */ | |
5545 | facet_reset_counters(facet); | |
7ee20df1 BP |
5546 | facet->rule = rule; |
5547 | } | |
5548 | } else { | |
5549 | /* Must avoid list_moved() in this case. */ | |
5550 | list_init(&rule->facets); | |
5551 | } | |
abe529af | 5552 | |
54a9cbc9 | 5553 | table_id = rule->up.table_id; |
5cb7a798 BP |
5554 | if (victim) { |
5555 | rule->tag = victim->tag; | |
5556 | } else if (table_id == 0) { | |
5557 | rule->tag = 0; | |
5558 | } else { | |
5559 | struct flow flow; | |
5560 | ||
5561 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5562 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5563 | ofproto->tables[table_id].basis); | |
5564 | } | |
54a9cbc9 | 5565 | |
7ee20df1 | 5566 | complete_operation(rule); |
abe529af BP |
5567 | return 0; |
5568 | } | |
5569 | ||
5570 | static void | |
5571 | rule_destruct(struct rule *rule_) | |
5572 | { | |
5573 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5574 | struct facet *facet, *next_facet; |
5575 | ||
abe529af | 5576 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 5577 | facet_revalidate(facet); |
abe529af | 5578 | } |
7ee20df1 BP |
5579 | |
5580 | complete_operation(rule); | |
abe529af BP |
5581 | } |
5582 | ||
5583 | static void | |
5584 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5585 | { | |
5586 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 5587 | |
4464589e EJ |
5588 | /* push_all_stats() can handle flow misses which, when using the learn |
5589 | * action, can cause rules to be added and deleted. This can corrupt our | |
5590 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5591 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5592 | push_all_stats__(false); | |
bf1e8ff9 | 5593 | |
abe529af BP |
5594 | /* Start from historical data for 'rule' itself that are no longer tracked |
5595 | * in facets. This counts, for example, facets that have expired. */ | |
5596 | *packets = rule->packet_count; | |
5597 | *bytes = rule->byte_count; | |
abe529af BP |
5598 | } |
5599 | ||
0a740f48 EJ |
5600 | static void |
5601 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5602 | struct ofpbuf *packet) | |
abe529af | 5603 | { |
abe529af | 5604 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
14f94f9a | 5605 | struct initial_vals initial_vals; |
112bc5f4 | 5606 | struct dpif_flow_stats stats; |
bbafd73b EJ |
5607 | struct xlate_out xout; |
5608 | struct xlate_in xin; | |
abe529af | 5609 | |
a7752d4a | 5610 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5611 | rule_credit_stats(rule, &stats); |
5612 | ||
14f94f9a | 5613 | initial_vals.vlan_tci = flow->vlan_tci; |
bbafd73b EJ |
5614 | xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags, |
5615 | packet); | |
5616 | xin.resubmit_stats = &stats; | |
5617 | xlate_actions(&xin, &xout); | |
112bc5f4 | 5618 | |
bbafd73b EJ |
5619 | execute_odp_actions(ofproto, flow, xout.odp_actions.data, |
5620 | xout.odp_actions.size, packet); | |
112bc5f4 | 5621 | |
bbafd73b | 5622 | xlate_out_uninit(&xout); |
0a740f48 | 5623 | } |
5bf0e941 | 5624 | |
0a740f48 EJ |
5625 | static enum ofperr |
5626 | rule_execute(struct rule *rule, const struct flow *flow, | |
5627 | struct ofpbuf *packet) | |
5628 | { | |
5629 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5630 | ofpbuf_delete(packet); | |
5bf0e941 | 5631 | return 0; |
abe529af BP |
5632 | } |
5633 | ||
7ee20df1 BP |
5634 | static void |
5635 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5636 | { |
5637 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5638 | |
5639 | complete_operation(rule); | |
abe529af BP |
5640 | } |
5641 | \f | |
97d6520b | 5642 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5643 | * May modify 'packet'. |
abe529af BP |
5644 | * Returns 0 if successful, otherwise a positive errno value. */ |
5645 | static int | |
52a90c29 | 5646 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5647 | { |
79e15b78 | 5648 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5649 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 | 5650 | struct ofpbuf key, odp_actions; |
0f49659a | 5651 | struct dpif_flow_stats stats; |
80e5eed9 | 5652 | struct odputil_keybuf keybuf; |
0f49659a | 5653 | struct ofpact_output output; |
bbafd73b EJ |
5654 | struct xlate_out xout; |
5655 | struct xlate_in xin; | |
80e5eed9 | 5656 | struct flow flow; |
abe529af BP |
5657 | int error; |
5658 | ||
b9ad7294 | 5659 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
0f49659a | 5660 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
b9ad7294 | 5661 | |
0f49659a EJ |
5662 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ |
5663 | flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow); | |
5664 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, | |
5665 | OFPP_LOCAL)); | |
5666 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 | 5667 | |
0f49659a EJ |
5668 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); |
5669 | output.port = ofport->up.ofp_port; | |
5670 | output.max_len = 0; | |
b9ad7294 | 5671 | |
bbafd73b EJ |
5672 | xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet); |
5673 | xin.ofpacts_len = sizeof output; | |
5674 | xin.ofpacts = &output.ofpact; | |
5675 | xin.resubmit_stats = &stats; | |
5676 | xlate_actions(&xin, &xout); | |
6ff686f2 | 5677 | |
acf60855 | 5678 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 | 5679 | key.data, key.size, |
bbafd73b | 5680 | xout.odp_actions.data, xout.odp_actions.size, |
abe529af | 5681 | packet); |
bbafd73b | 5682 | xlate_out_uninit(&xout); |
abe529af BP |
5683 | |
5684 | if (error) { | |
0f49659a EJ |
5685 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", |
5686 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
5687 | strerror(error)); | |
abe529af | 5688 | } |
79e15b78 EJ |
5689 | |
5690 | ofproto->stats.tx_packets++; | |
5691 | ofproto->stats.tx_bytes += packet->size; | |
abe529af BP |
5692 | return error; |
5693 | } | |
5694 | \f | |
df2c07f4 | 5695 | /* OpenFlow to datapath action translation. */ |
abe529af | 5696 | |
f03a84b9 | 5697 | static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *); |
f25d0cf3 | 5698 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
f03a84b9 EJ |
5699 | struct xlate_ctx *); |
5700 | static void xlate_normal(struct xlate_ctx *); | |
abe529af | 5701 | |
6a7e895f BP |
5702 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5703 | * The action will state 'slow' as the reason that the action is in the slow | |
5704 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5705 | * dump-flows" output to see why a flow is in the slow path.) | |
5706 | * | |
5707 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5708 | * 'stub_size' must be large enough for the action. | |
5709 | * | |
5710 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5711 | * respectively. */ | |
5712 | static void | |
5713 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5714 | enum slow_path_reason slow, | |
5715 | uint64_t *stub, size_t stub_size, | |
5716 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5717 | { | |
5718 | union user_action_cookie cookie; | |
5719 | struct ofpbuf buf; | |
5720 | ||
5721 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5722 | cookie.slow_path.unused = 0; | |
5723 | cookie.slow_path.reason = slow; | |
5724 | ||
5725 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5726 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
9032f11e | 5727 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); |
29089a54 | 5728 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5729 | } else { |
29089a54 RL |
5730 | put_userspace_action(ofproto, &buf, flow, &cookie, |
5731 | sizeof cookie.slow_path); | |
625b0720 | 5732 | } |
6a7e895f BP |
5733 | *actionsp = buf.data; |
5734 | *actions_lenp = buf.size; | |
5735 | } | |
5736 | ||
98403001 BP |
5737 | static size_t |
5738 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5739 | struct ofpbuf *odp_actions, | |
5740 | const struct flow *flow, | |
29089a54 RL |
5741 | const union user_action_cookie *cookie, |
5742 | const size_t cookie_size) | |
98403001 | 5743 | { |
98403001 BP |
5744 | uint32_t pid; |
5745 | ||
acf60855 | 5746 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
e1b1d06a | 5747 | ofp_port_to_odp_port(ofproto, flow->in_port)); |
98403001 | 5748 | |
29089a54 RL |
5749 | return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); |
5750 | } | |
5751 | ||
5752 | /* Compose SAMPLE action for sFlow or IPFIX. The given probability is | |
5753 | * the number of packets out of UINT32_MAX to sample. The given | |
5754 | * cookie is passed back in the callback for each sampled packet. | |
5755 | */ | |
5756 | static size_t | |
5757 | compose_sample_action(const struct ofproto_dpif *ofproto, | |
5758 | struct ofpbuf *odp_actions, | |
5759 | const struct flow *flow, | |
5760 | const uint32_t probability, | |
5761 | const union user_action_cookie *cookie, | |
5762 | const size_t cookie_size) | |
5763 | { | |
5764 | size_t sample_offset, actions_offset; | |
5765 | int cookie_offset; | |
5766 | ||
5767 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
5768 | ||
5769 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5770 | ||
5771 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
5772 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie, | |
5773 | cookie_size); | |
5774 | ||
5775 | nl_msg_end_nested(odp_actions, actions_offset); | |
5776 | nl_msg_end_nested(odp_actions, sample_offset); | |
5777 | return cookie_offset; | |
98403001 BP |
5778 | } |
5779 | ||
36fc5f18 BP |
5780 | static void |
5781 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5782 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 5783 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
5784 | { |
5785 | int ifindex; | |
5786 | ||
5787 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 5788 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
5789 | |
5790 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5791 | * port information") for the interpretation of cookie->output. */ | |
5792 | switch (n_outputs) { | |
5793 | case 0: | |
5794 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 5795 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
5796 | break; |
5797 | ||
5798 | case 1: | |
5799 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5800 | if (ifindex) { | |
1673e0e4 | 5801 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
5802 | break; |
5803 | } | |
5804 | /* Fall through. */ | |
5805 | default: | |
5806 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 5807 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
5808 | break; |
5809 | } | |
5810 | } | |
5811 | ||
29089a54 | 5812 | /* Compose SAMPLE action for sFlow bridge sampling. */ |
6ff686f2 PS |
5813 | static size_t |
5814 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5815 | struct ofpbuf *odp_actions, | |
5816 | const struct flow *flow, | |
5817 | uint32_t odp_port) | |
5818 | { | |
6ff686f2 | 5819 | uint32_t probability; |
1673e0e4 | 5820 | union user_action_cookie cookie; |
6ff686f2 PS |
5821 | |
5822 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5823 | return 0; | |
5824 | } | |
5825 | ||
6ff686f2 | 5826 | probability = dpif_sflow_get_probability(ofproto->sflow); |
36fc5f18 BP |
5827 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
5828 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
6ff686f2 | 5829 | |
29089a54 RL |
5830 | return compose_sample_action(ofproto, odp_actions, flow, probability, |
5831 | &cookie, sizeof cookie.sflow); | |
5832 | } | |
5833 | ||
5834 | static void | |
5835 | compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id, | |
5836 | uint32_t obs_domain_id, uint32_t obs_point_id, | |
5837 | union user_action_cookie *cookie) | |
5838 | { | |
5839 | cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE; | |
5840 | cookie->flow_sample.probability = probability; | |
5841 | cookie->flow_sample.collector_set_id = collector_set_id; | |
5842 | cookie->flow_sample.obs_domain_id = obs_domain_id; | |
5843 | cookie->flow_sample.obs_point_id = obs_point_id; | |
5844 | } | |
5845 | ||
5846 | static void | |
5847 | compose_ipfix_cookie(union user_action_cookie *cookie) | |
5848 | { | |
5849 | cookie->type = USER_ACTION_COOKIE_IPFIX; | |
5850 | } | |
5851 | ||
5852 | /* Compose SAMPLE action for IPFIX bridge sampling. */ | |
5853 | static void | |
5854 | compose_ipfix_action(const struct ofproto_dpif *ofproto, | |
5855 | struct ofpbuf *odp_actions, | |
5856 | const struct flow *flow) | |
5857 | { | |
5858 | uint32_t probability; | |
5859 | union user_action_cookie cookie; | |
5860 | ||
5861 | if (!ofproto->ipfix || flow->in_port == OFPP_NONE) { | |
5862 | return; | |
5863 | } | |
5864 | ||
5865 | probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix); | |
5866 | compose_ipfix_cookie(&cookie); | |
5867 | ||
5868 | compose_sample_action(ofproto, odp_actions, flow, probability, | |
5869 | &cookie, sizeof cookie.ipfix); | |
6ff686f2 PS |
5870 | } |
5871 | ||
29089a54 RL |
5872 | /* SAMPLE action for sFlow must be first action in any given list of |
5873 | * actions. At this point we do not have all information required to | |
5874 | * build it. So try to build sample action as complete as possible. */ | |
6ff686f2 | 5875 | static void |
f03a84b9 | 5876 | add_sflow_action(struct xlate_ctx *ctx) |
6ff686f2 PS |
5877 | { |
5878 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
bbafd73b EJ |
5879 | &ctx->xout->odp_actions, |
5880 | &ctx->xin->flow, OVSP_NONE); | |
6ff686f2 PS |
5881 | ctx->sflow_odp_port = 0; |
5882 | ctx->sflow_n_outputs = 0; | |
5883 | } | |
5884 | ||
29089a54 RL |
5885 | /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list |
5886 | * of actions, eventually after the SAMPLE action for sFlow. */ | |
5887 | static void | |
f03a84b9 | 5888 | add_ipfix_action(struct xlate_ctx *ctx) |
29089a54 | 5889 | { |
bbafd73b EJ |
5890 | compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions, |
5891 | &ctx->xin->flow); | |
29089a54 RL |
5892 | } |
5893 | ||
6ff686f2 PS |
5894 | /* Fix SAMPLE action according to data collected while composing ODP actions. |
5895 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
5896 | * USERSPACE action's user-cookie which is required for sflow. */ | |
5897 | static void | |
f03a84b9 | 5898 | fix_sflow_action(struct xlate_ctx *ctx) |
6ff686f2 PS |
5899 | { |
5900 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 5901 | union user_action_cookie *cookie; |
6ff686f2 PS |
5902 | |
5903 | if (!ctx->user_cookie_offset) { | |
5904 | return; | |
5905 | } | |
5906 | ||
bbafd73b | 5907 | cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset, |
a93f1927 | 5908 | sizeof cookie->sflow); |
cb22974d | 5909 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
6ff686f2 | 5910 | |
36fc5f18 BP |
5911 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
5912 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
5913 | } |
5914 | ||
6ff686f2 | 5915 | static void |
f03a84b9 | 5916 | compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port, |
81b1afb1 | 5917 | bool check_stp) |
6ff686f2 | 5918 | { |
d59906fb | 5919 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
d8558b4a JR |
5920 | ovs_be16 flow_vlan_tci; |
5921 | uint32_t flow_skb_mark; | |
5922 | uint8_t flow_nw_tos; | |
a4454ac6 | 5923 | struct priority_to_dscp *pdscp; |
0a740f48 EJ |
5924 | uint32_t out_port, odp_port; |
5925 | ||
5926 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
5927 | * before traversing a patch port. */ | |
cff78c88 | 5928 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); |
d59906fb | 5929 | |
a4454ac6 EJ |
5930 | if (!ofport) { |
5931 | xlate_report(ctx, "Nonexistent output port"); | |
5932 | return; | |
5933 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
5934 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
5935 | return; | |
5936 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
5937 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
5938 | return; | |
5939 | } | |
8b36f51e | 5940 | |
0a740f48 EJ |
5941 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5942 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
bbafd73b | 5943 | struct flow old_flow = ctx->xin->flow; |
0a740f48 | 5944 | const struct ofproto_dpif *peer_ofproto; |
bb374ef6 | 5945 | enum slow_path_reason special; |
ffaef958 | 5946 | struct ofport_dpif *in_port; |
0a740f48 EJ |
5947 | |
5948 | if (!peer) { | |
5949 | xlate_report(ctx, "Nonexistent patch port peer"); | |
5950 | return; | |
5951 | } | |
5952 | ||
5953 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5954 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
5955 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
5956 | return; | |
5957 | } | |
5958 | ||
5959 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
bbafd73b EJ |
5960 | ctx->xin->flow.in_port = peer->up.ofp_port; |
5961 | ctx->xin->flow.metadata = htonll(0); | |
5962 | memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel); | |
5963 | memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs); | |
5964 | ||
5965 | in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port); | |
5966 | special = process_special(ctx->ofproto, &ctx->xin->flow, in_port, | |
5967 | ctx->xin->packet); | |
bb374ef6 | 5968 | if (special) { |
bbafd73b | 5969 | ctx->xout->slow = special; |
bb374ef6 | 5970 | } else if (!in_port || may_receive(in_port, ctx)) { |
ffaef958 | 5971 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { |
bbafd73b | 5972 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true); |
ffaef958 BP |
5973 | } else { |
5974 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
5975 | * learning action look at the packet, then drop it. */ | |
5976 | struct flow old_base_flow = ctx->base_flow; | |
bbafd73b EJ |
5977 | size_t old_size = ctx->xout->odp_actions.size; |
5978 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true); | |
ffaef958 | 5979 | ctx->base_flow = old_base_flow; |
bbafd73b | 5980 | ctx->xout->odp_actions.size = old_size; |
ffaef958 BP |
5981 | } |
5982 | } | |
5983 | ||
bbafd73b | 5984 | ctx->xin->flow = old_flow; |
0a740f48 EJ |
5985 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
5986 | ||
bbafd73b EJ |
5987 | if (ctx->xin->resubmit_stats) { |
5988 | netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats); | |
5989 | netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats); | |
0a740f48 EJ |
5990 | } |
5991 | ||
5992 | return; | |
5993 | } | |
5994 | ||
bbafd73b EJ |
5995 | flow_vlan_tci = ctx->xin->flow.vlan_tci; |
5996 | flow_skb_mark = ctx->xin->flow.skb_mark; | |
5997 | flow_nw_tos = ctx->xin->flow.nw_tos; | |
d8558b4a | 5998 | |
bbafd73b | 5999 | pdscp = get_priority(ofport, ctx->xin->flow.skb_priority); |
a4454ac6 | 6000 | if (pdscp) { |
bbafd73b EJ |
6001 | ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK; |
6002 | ctx->xin->flow.nw_tos |= pdscp->dscp; | |
d59906fb EJ |
6003 | } |
6004 | ||
b9ad7294 | 6005 | if (ofport->tnl_port) { |
d8558b4a JR |
6006 | /* Save tunnel metadata so that changes made due to |
6007 | * the Logical (tunnel) Port are not visible for any further | |
6008 | * matches, while explicit set actions on tunnel metadata are. | |
6009 | */ | |
bbafd73b EJ |
6010 | struct flow_tnl flow_tnl = ctx->xin->flow.tunnel; |
6011 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow); | |
b9ad7294 EJ |
6012 | if (odp_port == OVSP_NONE) { |
6013 | xlate_report(ctx, "Tunneling decided against output"); | |
88e90491 | 6014 | goto out; /* restore flow_nw_tos */ |
b9ad7294 | 6015 | } |
bbafd73b | 6016 | if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) { |
0ad90c84 JR |
6017 | xlate_report(ctx, "Not tunneling to our own address"); |
6018 | goto out; /* restore flow_nw_tos */ | |
6019 | } | |
bbafd73b EJ |
6020 | if (ctx->xin->resubmit_stats) { |
6021 | netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats); | |
b9ad7294 EJ |
6022 | } |
6023 | out_port = odp_port; | |
bbafd73b EJ |
6024 | commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow, |
6025 | &ctx->xout->odp_actions); | |
6026 | ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */ | |
b9ad7294 | 6027 | } else { |
deea1200 | 6028 | uint16_t vlandev_port; |
cf630ea3 | 6029 | odp_port = ofport->odp_port; |
deea1200 AW |
6030 | vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port, |
6031 | ctx->xin->flow.vlan_tci); | |
6032 | if (vlandev_port == ofp_port) { | |
6033 | out_port = odp_port; | |
6034 | } else { | |
6035 | out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port); | |
bbafd73b | 6036 | ctx->xin->flow.vlan_tci = htons(0); |
b9ad7294 | 6037 | } |
bbafd73b | 6038 | ctx->xin->flow.skb_mark &= ~IPSEC_MARK; |
52a90c29 | 6039 | } |
bbafd73b EJ |
6040 | commit_odp_actions(&ctx->xin->flow, &ctx->base_flow, |
6041 | &ctx->xout->odp_actions); | |
6042 | nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); | |
52a90c29 | 6043 | |
6ff686f2 PS |
6044 | ctx->sflow_odp_port = odp_port; |
6045 | ctx->sflow_n_outputs++; | |
bbafd73b | 6046 | ctx->xout->nf_output_iface = ofp_port; |
d8558b4a JR |
6047 | |
6048 | /* Restore flow */ | |
bbafd73b EJ |
6049 | ctx->xin->flow.vlan_tci = flow_vlan_tci; |
6050 | ctx->xin->flow.skb_mark = flow_skb_mark; | |
88e90491 | 6051 | out: |
bbafd73b | 6052 | ctx->xin->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
6053 | } |
6054 | ||
abe529af | 6055 | static void |
f03a84b9 | 6056 | compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 6057 | { |
81b1afb1 | 6058 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
6059 | } |
6060 | ||
55599423 | 6061 | static void |
f03a84b9 | 6062 | tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule) |
55599423 JR |
6063 | { |
6064 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6065 | uint8_t table_id = ctx->table_id; | |
6066 | ||
6067 | if (table_id > 0 && table_id < N_TABLES) { | |
6068 | struct table_dpif *table = &ofproto->tables[table_id]; | |
6069 | if (table->other_table) { | |
bbafd73b EJ |
6070 | ctx->xout->tags |= (rule && rule->tag |
6071 | ? rule->tag | |
6072 | : rule_calculate_tag(&ctx->xin->flow, | |
6073 | &table->other_table->mask, | |
6074 | table->basis)); | |
55599423 JR |
6075 | } |
6076 | } | |
6077 | } | |
6078 | ||
6079 | /* Common rule processing in one place to avoid duplicating code. */ | |
6080 | static struct rule_dpif * | |
f03a84b9 | 6081 | ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule, |
55599423 JR |
6082 | bool may_packet_in) |
6083 | { | |
bbafd73b EJ |
6084 | if (ctx->xin->resubmit_hook) { |
6085 | ctx->xin->resubmit_hook(ctx, rule); | |
55599423 JR |
6086 | } |
6087 | if (rule == NULL && may_packet_in) { | |
6088 | /* XXX | |
6089 | * check if table configuration flags | |
6090 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
6091 | * OFPTC_TABLE_MISS_CONTINUE, | |
6092 | * OFPTC_TABLE_MISS_DROP | |
6093 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
6094 | */ | |
bbafd73b | 6095 | rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow); |
55599423 | 6096 | } |
bbafd73b EJ |
6097 | if (rule && ctx->xin->resubmit_stats) { |
6098 | rule_credit_stats(rule, ctx->xin->resubmit_stats); | |
55599423 JR |
6099 | } |
6100 | return rule; | |
6101 | } | |
6102 | ||
abe529af | 6103 | static void |
f03a84b9 | 6104 | xlate_table_action(struct xlate_ctx *ctx, |
1688c479 | 6105 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
6106 | { |
6107 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
6108 | struct rule_dpif *rule; | |
bbafd73b | 6109 | uint16_t old_in_port = ctx->xin->flow.in_port; |
55599423 | 6110 | uint8_t old_table_id = ctx->table_id; |
29901626 | 6111 | |
29901626 | 6112 | ctx->table_id = table_id; |
abe529af | 6113 | |
54a9cbc9 | 6114 | /* Look up a flow with 'in_port' as the input port. */ |
bbafd73b EJ |
6115 | ctx->xin->flow.in_port = in_port; |
6116 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id); | |
55599423 JR |
6117 | |
6118 | tag_the_flow(ctx, rule); | |
54a9cbc9 BP |
6119 | |
6120 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
6121 | * OFPP_IN_PORT will have surprising behavior. */ | |
bbafd73b | 6122 | ctx->xin->flow.in_port = old_in_port; |
abe529af | 6123 | |
55599423 | 6124 | rule = ctx_rule_hooks(ctx, rule, may_packet_in); |
1688c479 | 6125 | |
abe529af | 6126 | if (rule) { |
18b2a258 | 6127 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 6128 | |
abe529af | 6129 | ctx->recurse++; |
18b2a258 | 6130 | ctx->rule = rule; |
f25d0cf3 | 6131 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 6132 | ctx->rule = old_rule; |
abe529af BP |
6133 | ctx->recurse--; |
6134 | } | |
29901626 BP |
6135 | |
6136 | ctx->table_id = old_table_id; | |
abe529af BP |
6137 | } else { |
6138 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
6139 | ||
29901626 | 6140 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 6141 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 6142 | ctx->max_resubmit_trigger = true; |
abe529af BP |
6143 | } |
6144 | } | |
6145 | ||
29901626 | 6146 | static void |
f03a84b9 | 6147 | xlate_ofpact_resubmit(struct xlate_ctx *ctx, |
f25d0cf3 | 6148 | const struct ofpact_resubmit *resubmit) |
29901626 BP |
6149 | { |
6150 | uint16_t in_port; | |
6151 | uint8_t table_id; | |
6152 | ||
f25d0cf3 BP |
6153 | in_port = resubmit->in_port; |
6154 | if (in_port == OFPP_IN_PORT) { | |
bbafd73b | 6155 | in_port = ctx->xin->flow.in_port; |
f25d0cf3 BP |
6156 | } |
6157 | ||
6158 | table_id = resubmit->table_id; | |
6159 | if (table_id == 255) { | |
6160 | table_id = ctx->table_id; | |
6161 | } | |
29901626 | 6162 | |
1688c479 | 6163 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
6164 | } |
6165 | ||
abe529af | 6166 | static void |
f03a84b9 | 6167 | flood_packets(struct xlate_ctx *ctx, bool all) |
abe529af BP |
6168 | { |
6169 | struct ofport_dpif *ofport; | |
6170 | ||
b3e9b2ed | 6171 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 6172 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb | 6173 | |
bbafd73b | 6174 | if (ofp_port == ctx->xin->flow.in_port) { |
d59906fb EJ |
6175 | continue; |
6176 | } | |
6177 | ||
5e48dc2b | 6178 | if (all) { |
81b1afb1 | 6179 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 6180 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 6181 | compose_output_action(ctx, ofp_port); |
abe529af BP |
6182 | } |
6183 | } | |
b3e9b2ed | 6184 | |
bbafd73b | 6185 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
abe529af BP |
6186 | } |
6187 | ||
6ff686f2 | 6188 | static void |
f03a84b9 | 6189 | execute_controller_action(struct xlate_ctx *ctx, int len, |
a7349929 BP |
6190 | enum ofp_packet_in_reason reason, |
6191 | uint16_t controller_id) | |
6ff686f2 | 6192 | { |
999fba59 EJ |
6193 | struct ofputil_packet_in pin; |
6194 | struct ofpbuf *packet; | |
1ac7c9bd | 6195 | struct flow key; |
6ff686f2 | 6196 | |
bbafd73b EJ |
6197 | ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER); |
6198 | ctx->xout->slow = SLOW_CONTROLLER; | |
6199 | if (!ctx->xin->packet) { | |
999fba59 EJ |
6200 | return; |
6201 | } | |
6202 | ||
bbafd73b | 6203 | packet = ofpbuf_clone(ctx->xin->packet); |
999fba59 | 6204 | |
1ac7c9bd SH |
6205 | key.skb_priority = 0; |
6206 | key.skb_mark = 0; | |
6207 | memset(&key.tunnel, 0, sizeof key.tunnel); | |
999fba59 | 6208 | |
1ac7c9bd SH |
6209 | commit_odp_actions(&ctx->xin->flow, &ctx->base_flow, |
6210 | &ctx->xout->odp_actions); | |
999fba59 | 6211 | |
1ac7c9bd SH |
6212 | odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data, |
6213 | ctx->xout->odp_actions.size, NULL, NULL); | |
999fba59 EJ |
6214 | |
6215 | pin.packet = packet->data; | |
6216 | pin.packet_len = packet->size; | |
f0fd1a17 | 6217 | pin.reason = reason; |
a7349929 | 6218 | pin.controller_id = controller_id; |
54834960 | 6219 | pin.table_id = ctx->table_id; |
18b2a258 | 6220 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 6221 | |
999fba59 | 6222 | pin.send_len = len; |
bbafd73b | 6223 | flow_get_metadata(&ctx->xin->flow, &pin.fmd); |
999fba59 | 6224 | |
d8653c38 | 6225 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 6226 | ofpbuf_delete(packet); |
6ff686f2 PS |
6227 | } |
6228 | ||
b02475c5 | 6229 | static void |
f03a84b9 | 6230 | execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type) |
b02475c5 SH |
6231 | { |
6232 | ovs_assert(eth_type_mpls(eth_type)); | |
6233 | ||
6234 | if (ctx->base_flow.mpls_depth) { | |
bbafd73b EJ |
6235 | ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); |
6236 | ctx->xin->flow.mpls_depth++; | |
b02475c5 SH |
6237 | } else { |
6238 | ovs_be32 label; | |
6239 | uint8_t tc, ttl; | |
6240 | ||
bbafd73b | 6241 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) { |
b02475c5 SH |
6242 | label = htonl(0x2); /* IPV6 Explicit Null. */ |
6243 | } else { | |
6244 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6245 | } | |
bbafd73b EJ |
6246 | tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2; |
6247 | ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40; | |
6248 | ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
6249 | ctx->xin->flow.mpls_depth = 1; | |
b02475c5 | 6250 | } |
bbafd73b | 6251 | ctx->xin->flow.dl_type = eth_type; |
b02475c5 SH |
6252 | } |
6253 | ||
6254 | static void | |
f03a84b9 | 6255 | execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type) |
b02475c5 | 6256 | { |
bbafd73b | 6257 | ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type)); |
b02475c5 SH |
6258 | ovs_assert(!eth_type_mpls(eth_type)); |
6259 | ||
bbafd73b EJ |
6260 | if (ctx->xin->flow.mpls_depth) { |
6261 | ctx->xin->flow.mpls_depth--; | |
6262 | ctx->xin->flow.mpls_lse = htonl(0); | |
6263 | if (!ctx->xin->flow.mpls_depth) { | |
6264 | ctx->xin->flow.dl_type = eth_type; | |
b02475c5 SH |
6265 | } |
6266 | } | |
6267 | } | |
6268 | ||
f0fd1a17 | 6269 | static bool |
f03a84b9 | 6270 | compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 | 6271 | { |
bbafd73b EJ |
6272 | if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) && |
6273 | ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
f0fd1a17 PS |
6274 | return false; |
6275 | } | |
6276 | ||
bbafd73b EJ |
6277 | if (ctx->xin->flow.nw_ttl > 1) { |
6278 | ctx->xin->flow.nw_ttl--; | |
f0fd1a17 PS |
6279 | return false; |
6280 | } else { | |
c2d967a5 MM |
6281 | size_t i; |
6282 | ||
6283 | for (i = 0; i < ids->n_controllers; i++) { | |
6284 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6285 | ids->cnt_ids[i]); | |
6286 | } | |
f0fd1a17 PS |
6287 | |
6288 | /* Stop processing for current table. */ | |
6289 | return true; | |
6290 | } | |
6291 | } | |
6292 | ||
0f3f3c3d | 6293 | static bool |
f03a84b9 | 6294 | execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl) |
0f3f3c3d | 6295 | { |
bbafd73b | 6296 | if (!eth_type_mpls(ctx->xin->flow.dl_type)) { |
0f3f3c3d SH |
6297 | return true; |
6298 | } | |
6299 | ||
bbafd73b | 6300 | set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl); |
0f3f3c3d SH |
6301 | return false; |
6302 | } | |
6303 | ||
b676167a | 6304 | static bool |
f03a84b9 | 6305 | execute_dec_mpls_ttl_action(struct xlate_ctx *ctx) |
b676167a | 6306 | { |
bbafd73b | 6307 | uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse); |
b676167a | 6308 | |
bbafd73b | 6309 | if (!eth_type_mpls(ctx->xin->flow.dl_type)) { |
b676167a SH |
6310 | return false; |
6311 | } | |
6312 | ||
be80bc65 | 6313 | if (ttl > 1) { |
b676167a | 6314 | ttl--; |
bbafd73b | 6315 | set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl); |
b676167a SH |
6316 | return false; |
6317 | } else { | |
6318 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
6319 | ||
6320 | /* Stop processing for current table. */ | |
6321 | return true; | |
6322 | } | |
6323 | } | |
6324 | ||
abe529af | 6325 | static void |
f03a84b9 | 6326 | xlate_output_action(struct xlate_ctx *ctx, |
1688c479 | 6327 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af | 6328 | { |
bbafd73b | 6329 | uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface; |
abe529af | 6330 | |
bbafd73b | 6331 | ctx->xout->nf_output_iface = NF_OUT_DROP; |
abe529af BP |
6332 | |
6333 | switch (port) { | |
6334 | case OFPP_IN_PORT: | |
bbafd73b | 6335 | compose_output_action(ctx, ctx->xin->flow.in_port); |
abe529af BP |
6336 | break; |
6337 | case OFPP_TABLE: | |
bbafd73b | 6338 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in); |
abe529af BP |
6339 | break; |
6340 | case OFPP_NORMAL: | |
6341 | xlate_normal(ctx); | |
6342 | break; | |
6343 | case OFPP_FLOOD: | |
d59906fb | 6344 | flood_packets(ctx, false); |
abe529af BP |
6345 | break; |
6346 | case OFPP_ALL: | |
d59906fb | 6347 | flood_packets(ctx, true); |
abe529af BP |
6348 | break; |
6349 | case OFPP_CONTROLLER: | |
a7349929 | 6350 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 6351 | break; |
e81d2933 EJ |
6352 | case OFPP_NONE: |
6353 | break; | |
a0fbe94a | 6354 | case OFPP_LOCAL: |
abe529af | 6355 | default: |
bbafd73b | 6356 | if (port != ctx->xin->flow.in_port) { |
81b1afb1 | 6357 | compose_output_action(ctx, port); |
3dd3eace BP |
6358 | } else { |
6359 | xlate_report(ctx, "skipping output to input port"); | |
abe529af BP |
6360 | } |
6361 | break; | |
6362 | } | |
6363 | ||
6364 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
bbafd73b EJ |
6365 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
6366 | } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) { | |
6367 | ctx->xout->nf_output_iface = prev_nf_output_iface; | |
abe529af | 6368 | } else if (prev_nf_output_iface != NF_OUT_DROP && |
bbafd73b EJ |
6369 | ctx->xout->nf_output_iface != NF_OUT_FLOOD) { |
6370 | ctx->xout->nf_output_iface = NF_OUT_MULTI; | |
abe529af BP |
6371 | } |
6372 | } | |
6373 | ||
f694937d | 6374 | static void |
f03a84b9 | 6375 | xlate_output_reg_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6376 | const struct ofpact_output_reg *or) |
f694937d | 6377 | { |
bbafd73b | 6378 | uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow); |
f25d0cf3 | 6379 | if (port <= UINT16_MAX) { |
1688c479 | 6380 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
6381 | } |
6382 | } | |
6383 | ||
abe529af | 6384 | static void |
f03a84b9 | 6385 | xlate_enqueue_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6386 | const struct ofpact_enqueue *enqueue) |
abe529af | 6387 | { |
f25d0cf3 BP |
6388 | uint16_t ofp_port = enqueue->port; |
6389 | uint32_t queue_id = enqueue->queue; | |
abff858b | 6390 | uint32_t flow_priority, priority; |
abe529af BP |
6391 | int error; |
6392 | ||
f25d0cf3 | 6393 | /* Translate queue to priority. */ |
acf60855 JP |
6394 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6395 | queue_id, &priority); | |
abe529af BP |
6396 | if (error) { |
6397 | /* Fall back to ordinary output action. */ | |
1688c479 | 6398 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
6399 | return; |
6400 | } | |
6401 | ||
f25d0cf3 | 6402 | /* Check output port. */ |
abe529af | 6403 | if (ofp_port == OFPP_IN_PORT) { |
bbafd73b EJ |
6404 | ofp_port = ctx->xin->flow.in_port; |
6405 | } else if (ofp_port == ctx->xin->flow.in_port) { | |
8ba855c1 | 6406 | return; |
abe529af | 6407 | } |
abe529af | 6408 | |
df2c07f4 | 6409 | /* Add datapath actions. */ |
bbafd73b EJ |
6410 | flow_priority = ctx->xin->flow.skb_priority; |
6411 | ctx->xin->flow.skb_priority = priority; | |
81b1afb1 | 6412 | compose_output_action(ctx, ofp_port); |
bbafd73b | 6413 | ctx->xin->flow.skb_priority = flow_priority; |
abe529af BP |
6414 | |
6415 | /* Update NetFlow output port. */ | |
bbafd73b EJ |
6416 | if (ctx->xout->nf_output_iface == NF_OUT_DROP) { |
6417 | ctx->xout->nf_output_iface = ofp_port; | |
6418 | } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) { | |
6419 | ctx->xout->nf_output_iface = NF_OUT_MULTI; | |
abe529af BP |
6420 | } |
6421 | } | |
6422 | ||
6423 | static void | |
f03a84b9 | 6424 | xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 6425 | { |
f25d0cf3 | 6426 | uint32_t skb_priority; |
abe529af | 6427 | |
acf60855 JP |
6428 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6429 | queue_id, &skb_priority)) { | |
bbafd73b | 6430 | ctx->xin->flow.skb_priority = skb_priority; |
f25d0cf3 BP |
6431 | } else { |
6432 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 6433 | * has already been logged. */ |
abe529af | 6434 | } |
abe529af BP |
6435 | } |
6436 | ||
daff3353 EJ |
6437 | static bool |
6438 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6439 | { | |
6440 | struct ofproto_dpif *ofproto = ofproto_; | |
6441 | struct ofport_dpif *port; | |
6442 | ||
6443 | switch (ofp_port) { | |
6444 | case OFPP_IN_PORT: | |
6445 | case OFPP_TABLE: | |
6446 | case OFPP_NORMAL: | |
6447 | case OFPP_FLOOD: | |
6448 | case OFPP_ALL: | |
439e4d8c | 6449 | case OFPP_NONE: |
daff3353 EJ |
6450 | return true; |
6451 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6452 | return false; | |
6453 | default: | |
6454 | port = get_ofp_port(ofproto, ofp_port); | |
6455 | return port ? port->may_enable : false; | |
6456 | } | |
6457 | } | |
6458 | ||
f25d0cf3 | 6459 | static void |
f03a84b9 | 6460 | xlate_bundle_action(struct xlate_ctx *ctx, |
f25d0cf3 BP |
6461 | const struct ofpact_bundle *bundle) |
6462 | { | |
6463 | uint16_t port; | |
6464 | ||
bbafd73b EJ |
6465 | port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb, |
6466 | ctx->ofproto); | |
f25d0cf3 | 6467 | if (bundle->dst.field) { |
bbafd73b | 6468 | nxm_reg_load(&bundle->dst, port, &ctx->xin->flow); |
f25d0cf3 | 6469 | } else { |
1688c479 | 6470 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
6471 | } |
6472 | } | |
6473 | ||
75a75043 | 6474 | static void |
f03a84b9 | 6475 | xlate_learn_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6476 | const struct ofpact_learn *learn) |
75a75043 BP |
6477 | { |
6478 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6479 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
6480 | uint64_t ofpacts_stub[1024 / 8]; |
6481 | struct ofpbuf ofpacts; | |
75a75043 BP |
6482 | int error; |
6483 | ||
f25d0cf3 | 6484 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
bbafd73b | 6485 | learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts); |
75a75043 BP |
6486 | |
6487 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6488 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
6489 | VLOG_WARN("learning action failed to modify flow table (%s)", |
6490 | ofperr_get_name(error)); | |
75a75043 BP |
6491 | } |
6492 | ||
f25d0cf3 | 6493 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
6494 | } |
6495 | ||
0e553d9c BP |
6496 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
6497 | * means "infinite". */ | |
6498 | static void | |
6499 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6500 | { | |
6501 | if (max && (!*timeout || *timeout > max)) { | |
6502 | *timeout = max; | |
6503 | } | |
6504 | } | |
6505 | ||
6506 | static void | |
f03a84b9 | 6507 | xlate_fin_timeout(struct xlate_ctx *ctx, |
f25d0cf3 | 6508 | const struct ofpact_fin_timeout *oft) |
0e553d9c | 6509 | { |
bbafd73b | 6510 | if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { |
0e553d9c BP |
6511 | struct rule_dpif *rule = ctx->rule; |
6512 | ||
f25d0cf3 BP |
6513 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
6514 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
6515 | } |
6516 | } | |
6517 | ||
29089a54 | 6518 | static void |
f03a84b9 | 6519 | xlate_sample_action(struct xlate_ctx *ctx, |
29089a54 RL |
6520 | const struct ofpact_sample *os) |
6521 | { | |
6522 | union user_action_cookie cookie; | |
6523 | /* Scale the probability from 16-bit to 32-bit while representing | |
6524 | * the same percentage. */ | |
6525 | uint32_t probability = (os->probability << 16) | os->probability; | |
6526 | ||
bbafd73b EJ |
6527 | commit_odp_actions(&ctx->xin->flow, &ctx->base_flow, |
6528 | &ctx->xout->odp_actions); | |
29089a54 RL |
6529 | |
6530 | compose_flow_sample_cookie(os->probability, os->collector_set_id, | |
6531 | os->obs_domain_id, os->obs_point_id, &cookie); | |
bbafd73b | 6532 | compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow, |
29089a54 RL |
6533 | probability, &cookie, sizeof cookie.flow_sample); |
6534 | } | |
6535 | ||
21f7563c | 6536 | static bool |
f03a84b9 | 6537 | may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx) |
21f7563c | 6538 | { |
bbafd73b EJ |
6539 | if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst, |
6540 | eth_addr_stp) | |
9e1fd49b BP |
6541 | ? OFPUTIL_PC_NO_RECV_STP |
6542 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
6543 | return false; |
6544 | } | |
6545 | ||
6546 | /* Only drop packets here if both forwarding and learning are | |
6547 | * disabled. If just learning is enabled, we need to have | |
6548 | * OFPP_NORMAL and the learning action have a look at the packet | |
6549 | * before we can drop it. */ | |
6550 | if (!stp_forward_in_state(port->stp_state) | |
6551 | && !stp_learn_in_state(port->stp_state)) { | |
6552 | return false; | |
6553 | } | |
6554 | ||
6555 | return true; | |
6556 | } | |
6557 | ||
4863c249 | 6558 | static bool |
f03a84b9 | 6559 | tunnel_ecn_ok(struct xlate_ctx *ctx) |
4863c249 JP |
6560 | { |
6561 | if (is_ip_any(&ctx->base_flow) | |
bbafd73b | 6562 | && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { |
29a5df0a JP |
6563 | if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { |
6564 | VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" | |
6565 | " but is not ECN capable"); | |
6566 | return false; | |
6567 | } else { | |
6568 | /* Set the ECN CE value in the tunneled packet. */ | |
bbafd73b | 6569 | ctx->xin->flow.nw_tos |= IP_ECN_CE; |
29a5df0a | 6570 | } |
4863c249 JP |
6571 | } |
6572 | ||
6573 | return true; | |
6574 | } | |
6575 | ||
abe529af | 6576 | static void |
f25d0cf3 | 6577 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
f03a84b9 | 6578 | struct xlate_ctx *ctx) |
abe529af | 6579 | { |
254750ce | 6580 | bool was_evictable = true; |
f25d0cf3 | 6581 | const struct ofpact *a; |
abe529af | 6582 | |
254750ce BP |
6583 | if (ctx->rule) { |
6584 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6585 | was_evictable = ctx->rule->up.evictable; | |
6586 | ctx->rule->up.evictable = false; | |
6587 | } | |
55599423 JR |
6588 | |
6589 | do_xlate_actions_again: | |
f25d0cf3 BP |
6590 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
6591 | struct ofpact_controller *controller; | |
4cceacb9 | 6592 | const struct ofpact_metadata *metadata; |
38f2e360 | 6593 | |
848e8809 EJ |
6594 | if (ctx->exit) { |
6595 | break; | |
6596 | } | |
6597 | ||
f25d0cf3 BP |
6598 | switch (a->type) { |
6599 | case OFPACT_OUTPUT: | |
6600 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 6601 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
6602 | break; |
6603 | ||
6604 | case OFPACT_CONTROLLER: | |
6605 | controller = ofpact_get_CONTROLLER(a); | |
6606 | execute_controller_action(ctx, controller->max_len, | |
6607 | controller->reason, | |
6608 | controller->controller_id); | |
6609 | break; | |
690a61c5 | 6610 | |
f25d0cf3 BP |
6611 | case OFPACT_ENQUEUE: |
6612 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
6613 | break; |
6614 | ||
f25d0cf3 | 6615 | case OFPACT_SET_VLAN_VID: |
bbafd73b EJ |
6616 | ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
6617 | ctx->xin->flow.vlan_tci |= | |
6618 | (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) | |
6619 | | htons(VLAN_CFI)); | |
abe529af BP |
6620 | break; |
6621 | ||
f25d0cf3 | 6622 | case OFPACT_SET_VLAN_PCP: |
bbafd73b EJ |
6623 | ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
6624 | ctx->xin->flow.vlan_tci |= | |
6625 | htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT) | |
6626 | | VLAN_CFI); | |
abe529af BP |
6627 | break; |
6628 | ||
f25d0cf3 | 6629 | case OFPACT_STRIP_VLAN: |
bbafd73b | 6630 | ctx->xin->flow.vlan_tci = htons(0); |
abe529af BP |
6631 | break; |
6632 | ||
3e34fbdd | 6633 | case OFPACT_PUSH_VLAN: |
5dca28b5 | 6634 | /* XXX 802.1AD(QinQ) */ |
bbafd73b | 6635 | ctx->xin->flow.vlan_tci = htons(VLAN_CFI); |
3e34fbdd IY |
6636 | break; |
6637 | ||
f25d0cf3 | 6638 | case OFPACT_SET_ETH_SRC: |
bbafd73b | 6639 | memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, |
f25d0cf3 | 6640 | ETH_ADDR_LEN); |
abe529af BP |
6641 | break; |
6642 | ||
f25d0cf3 | 6643 | case OFPACT_SET_ETH_DST: |
bbafd73b | 6644 | memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, |
f25d0cf3 | 6645 | ETH_ADDR_LEN); |
abe529af BP |
6646 | break; |
6647 | ||
f25d0cf3 | 6648 | case OFPACT_SET_IPV4_SRC: |
bbafd73b EJ |
6649 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6650 | ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
1b035ef2 | 6651 | } |
abe529af BP |
6652 | break; |
6653 | ||
f25d0cf3 | 6654 | case OFPACT_SET_IPV4_DST: |
bbafd73b EJ |
6655 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6656 | ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
1b035ef2 | 6657 | } |
abe529af BP |
6658 | break; |
6659 | ||
f25d0cf3 | 6660 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d | 6661 | /* OpenFlow 1.0 only supports IPv4. */ |
bbafd73b EJ |
6662 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6663 | ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK; | |
6664 | ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; | |
c4f2731d | 6665 | } |
abe529af BP |
6666 | break; |
6667 | ||
f25d0cf3 | 6668 | case OFPACT_SET_L4_SRC_PORT: |
bbafd73b EJ |
6669 | if (is_ip_any(&ctx->xin->flow)) { |
6670 | ctx->xin->flow.tp_src = | |
6671 | htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
1b035ef2 | 6672 | } |
abe529af BP |
6673 | break; |
6674 | ||
f25d0cf3 | 6675 | case OFPACT_SET_L4_DST_PORT: |
bbafd73b EJ |
6676 | if (is_ip_any(&ctx->xin->flow)) { |
6677 | ctx->xin->flow.tp_dst = | |
6678 | htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
1b035ef2 | 6679 | } |
abe529af BP |
6680 | break; |
6681 | ||
f25d0cf3 BP |
6682 | case OFPACT_RESUBMIT: |
6683 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
6684 | break; |
6685 | ||
f25d0cf3 | 6686 | case OFPACT_SET_TUNNEL: |
bbafd73b EJ |
6687 | ctx->xin->flow.tunnel.tun_id = |
6688 | htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
29901626 BP |
6689 | break; |
6690 | ||
f25d0cf3 BP |
6691 | case OFPACT_SET_QUEUE: |
6692 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
6693 | break; |
6694 | ||
f25d0cf3 | 6695 | case OFPACT_POP_QUEUE: |
bbafd73b | 6696 | ctx->xin->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
6697 | break; |
6698 | ||
f25d0cf3 | 6699 | case OFPACT_REG_MOVE: |
bbafd73b | 6700 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow); |
38f2e360 BP |
6701 | break; |
6702 | ||
f25d0cf3 | 6703 | case OFPACT_REG_LOAD: |
bbafd73b | 6704 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow); |
38f2e360 BP |
6705 | break; |
6706 | ||
bd85dac1 | 6707 | case OFPACT_STACK_PUSH: |
bbafd73b | 6708 | nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow, |
bd85dac1 AZ |
6709 | &ctx->stack); |
6710 | break; | |
6711 | ||
6712 | case OFPACT_STACK_POP: | |
bbafd73b | 6713 | nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow, |
bd85dac1 AZ |
6714 | &ctx->stack); |
6715 | break; | |
6716 | ||
b02475c5 SH |
6717 | case OFPACT_PUSH_MPLS: |
6718 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6719 | break; | |
6720 | ||
6721 | case OFPACT_POP_MPLS: | |
6722 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6723 | break; | |
6724 | ||
0f3f3c3d | 6725 | case OFPACT_SET_MPLS_TTL: |
bbafd73b EJ |
6726 | if (execute_set_mpls_ttl_action(ctx, |
6727 | ofpact_get_SET_MPLS_TTL(a)->ttl)) { | |
0f3f3c3d SH |
6728 | goto out; |
6729 | } | |
6730 | break; | |
6731 | ||
b676167a SH |
6732 | case OFPACT_DEC_MPLS_TTL: |
6733 | if (execute_dec_mpls_ttl_action(ctx)) { | |
6734 | goto out; | |
6735 | } | |
6736 | break; | |
6737 | ||
f25d0cf3 | 6738 | case OFPACT_DEC_TTL: |
c2d967a5 | 6739 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
6740 | goto out; |
6741 | } | |
38f2e360 BP |
6742 | break; |
6743 | ||
f25d0cf3 BP |
6744 | case OFPACT_NOTE: |
6745 | /* Nothing to do. */ | |
abe529af BP |
6746 | break; |
6747 | ||
f25d0cf3 | 6748 | case OFPACT_MULTIPATH: |
bbafd73b | 6749 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow); |
abe529af | 6750 | break; |
daff3353 | 6751 | |
f25d0cf3 | 6752 | case OFPACT_BUNDLE: |
a368bb53 | 6753 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 6754 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 6755 | break; |
f694937d | 6756 | |
f25d0cf3 BP |
6757 | case OFPACT_OUTPUT_REG: |
6758 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 6759 | break; |
75a75043 | 6760 | |
f25d0cf3 | 6761 | case OFPACT_LEARN: |
bbafd73b EJ |
6762 | ctx->xout->has_learn = true; |
6763 | if (ctx->xin->may_learn) { | |
f25d0cf3 | 6764 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
6765 | } |
6766 | break; | |
848e8809 | 6767 | |
f25d0cf3 | 6768 | case OFPACT_EXIT: |
848e8809 EJ |
6769 | ctx->exit = true; |
6770 | break; | |
0e553d9c | 6771 | |
f25d0cf3 | 6772 | case OFPACT_FIN_TIMEOUT: |
bbafd73b | 6773 | ctx->xout->has_fin_timeout = true; |
f25d0cf3 | 6774 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 6775 | break; |
8dd54666 | 6776 | |
b19e8793 | 6777 | case OFPACT_CLEAR_ACTIONS: |
5dca28b5 | 6778 | /* XXX |
b19e8793 IY |
6779 | * Nothing to do because writa-actions is not supported for now. |
6780 | * When writa-actions is supported, clear-actions also must | |
6781 | * be supported at the same time. | |
6782 | */ | |
6783 | break; | |
6784 | ||
4cceacb9 JS |
6785 | case OFPACT_WRITE_METADATA: |
6786 | metadata = ofpact_get_WRITE_METADATA(a); | |
bbafd73b EJ |
6787 | ctx->xin->flow.metadata &= ~metadata->mask; |
6788 | ctx->xin->flow.metadata |= metadata->metadata & metadata->mask; | |
4cceacb9 JS |
6789 | break; |
6790 | ||
8dd54666 | 6791 | case OFPACT_GOTO_TABLE: { |
55599423 | 6792 | /* It is assumed that goto-table is the last action. */ |
8dd54666 | 6793 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); |
55599423 JR |
6794 | struct rule_dpif *rule; |
6795 | ||
cb22974d | 6796 | ovs_assert(ctx->table_id < ogt->table_id); |
55599423 JR |
6797 | |
6798 | ctx->table_id = ogt->table_id; | |
6799 | ||
6800 | /* Look up a flow from the new table. */ | |
bbafd73b | 6801 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id); |
55599423 JR |
6802 | |
6803 | tag_the_flow(ctx, rule); | |
6804 | ||
6805 | rule = ctx_rule_hooks(ctx, rule, true); | |
6806 | ||
6807 | if (rule) { | |
6808 | if (ctx->rule) { | |
6809 | ctx->rule->up.evictable = was_evictable; | |
6810 | } | |
6811 | ctx->rule = rule; | |
6812 | was_evictable = rule->up.evictable; | |
6813 | rule->up.evictable = false; | |
6814 | ||
6815 | /* Tail recursion removal. */ | |
6816 | ofpacts = rule->up.ofpacts; | |
6817 | ofpacts_len = rule->up.ofpacts_len; | |
6818 | goto do_xlate_actions_again; | |
6819 | } | |
8dd54666 IY |
6820 | break; |
6821 | } | |
29089a54 RL |
6822 | |
6823 | case OFPACT_SAMPLE: | |
6824 | xlate_sample_action(ctx, ofpact_get_SAMPLE(a)); | |
6825 | break; | |
abe529af BP |
6826 | } |
6827 | } | |
21f7563c | 6828 | |
f0fd1a17 | 6829 | out: |
254750ce BP |
6830 | if (ctx->rule) { |
6831 | ctx->rule->up.evictable = was_evictable; | |
6832 | } | |
abe529af BP |
6833 | } |
6834 | ||
6835 | static void | |
bbafd73b EJ |
6836 | xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto, |
6837 | const struct flow *flow, | |
6838 | const struct initial_vals *initial_vals, | |
6839 | struct rule_dpif *rule, uint8_t tcp_flags, | |
6840 | const struct ofpbuf *packet) | |
6841 | { | |
6842 | xin->ofproto = ofproto; | |
6843 | xin->flow = *flow; | |
6844 | xin->packet = packet; | |
6845 | xin->may_learn = packet != NULL; | |
6846 | xin->rule = rule; | |
6847 | xin->ofpacts = NULL; | |
6848 | xin->ofpacts_len = 0; | |
6849 | xin->tcp_flags = tcp_flags; | |
6850 | xin->resubmit_hook = NULL; | |
6851 | xin->report_hook = NULL; | |
6852 | xin->resubmit_stats = NULL; | |
6853 | ||
6854 | if (initial_vals) { | |
6855 | xin->initial_vals = *initial_vals; | |
6856 | } else { | |
6857 | xin->initial_vals.vlan_tci = xin->flow.vlan_tci; | |
6858 | } | |
6859 | } | |
6860 | ||
6861 | static void | |
6862 | xlate_out_uninit(struct xlate_out *xout) | |
abe529af | 6863 | { |
bbafd73b EJ |
6864 | if (xout) { |
6865 | ofpbuf_uninit(&xout->odp_actions); | |
6866 | } | |
6867 | } | |
6868 | ||
6869 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' | |
6870 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
6871 | static void | |
6872 | xlate_actions(struct xlate_in *xin, struct xlate_out *xout) | |
6873 | { | |
6874 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so | |
6875 | * that in the future we always keep a copy of the original flow for | |
6876 | * tracing purposes. */ | |
6877 | static bool hit_resubmit_limit; | |
6878 | ||
6879 | enum slow_path_reason special; | |
6880 | const struct ofpact *ofpacts; | |
6881 | struct ofport_dpif *in_port; | |
6882 | struct flow orig_flow; | |
6883 | struct xlate_ctx ctx; | |
6884 | size_t ofpacts_len; | |
6885 | ||
6886 | COVERAGE_INC(ofproto_dpif_xlate); | |
6887 | ||
ef506a7c JG |
6888 | /* Flow initialization rules: |
6889 | * - 'base_flow' must match the kernel's view of the packet at the | |
6890 | * time that action processing starts. 'flow' represents any | |
6891 | * transformations we wish to make through actions. | |
6892 | * - By default 'base_flow' and 'flow' are the same since the input | |
6893 | * packet matches the output before any actions are applied. | |
6894 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
6895 | * of the received packet as seen by the kernel. If we later output | |
6896 | * to another device without any modifications this will cause us to | |
6897 | * insert a new tag since the original one was stripped off by the | |
6898 | * VLAN device. | |
4110a57a JR |
6899 | * - Tunnel metadata as received is retained in 'flow'. This allows |
6900 | * tunnel metadata matching also in later tables. | |
6901 | * Since a kernel action for setting the tunnel metadata will only be | |
6902 | * generated with actual tunnel output, changing the tunnel metadata | |
6903 | * values in 'flow' (such as tun_id) will only have effect with a later | |
6904 | * tunnel output action. | |
ef506a7c JG |
6905 | * - Tunnel 'base_flow' is completely cleared since that is what the |
6906 | * kernel does. If we wish to maintain the original values an action | |
6907 | * needs to be generated. */ | |
6908 | ||
bbafd73b EJ |
6909 | ctx.xin = xin; |
6910 | ctx.xout = xout; | |
6911 | ||
6912 | ctx.ofproto = xin->ofproto; | |
6913 | ctx.rule = xin->rule; | |
6914 | ||
6915 | ctx.base_flow = ctx.xin->flow; | |
6916 | ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci; | |
6917 | memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel); | |
6918 | ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst; | |
6919 | ||
6920 | ctx.xout->tags = 0; | |
6921 | ctx.xout->slow = 0; | |
6922 | ctx.xout->has_learn = false; | |
6923 | ctx.xout->has_normal = false; | |
6924 | ctx.xout->has_fin_timeout = false; | |
6925 | ctx.xout->nf_output_iface = NF_OUT_DROP; | |
6926 | ctx.xout->mirrors = 0; | |
6927 | ||
6928 | ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub, | |
6929 | sizeof ctx.xout->odp_actions_stub); | |
6930 | ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE); | |
6931 | ||
6932 | ctx.recurse = 0; | |
6933 | ctx.max_resubmit_trigger = false; | |
6934 | ctx.orig_skb_priority = ctx.xin->flow.skb_priority; | |
6935 | ctx.table_id = 0; | |
6936 | ctx.exit = false; | |
6937 | ||
6938 | if (xin->ofpacts) { | |
6939 | ofpacts = xin->ofpacts; | |
6940 | ofpacts_len = xin->ofpacts_len; | |
6941 | } else if (xin->rule) { | |
6942 | ofpacts = xin->rule->up.ofpacts; | |
6943 | ofpacts_len = xin->rule->up.ofpacts_len; | |
6944 | } else { | |
6945 | NOT_REACHED(); | |
0f49659a | 6946 | } |
abe529af | 6947 | |
bbafd73b | 6948 | ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack); |
43d50bc8 | 6949 | |
bbafd73b | 6950 | if (ctx.ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 | 6951 | /* Do this conditionally because the copy is expensive enough that it |
9ba85077 | 6952 | * shows up in profiles. */ |
bbafd73b | 6953 | orig_flow = ctx.xin->flow; |
ccb7c863 BP |
6954 | } |
6955 | ||
bbafd73b EJ |
6956 | if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
6957 | switch (ctx.ofproto->up.frag_handling) { | |
7257b535 BP |
6958 | case OFPC_FRAG_NORMAL: |
6959 | /* We must pretend that transport ports are unavailable. */ | |
bbafd73b EJ |
6960 | ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0); |
6961 | ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0); | |
7257b535 BP |
6962 | break; |
6963 | ||
6964 | case OFPC_FRAG_DROP: | |
050ac423 | 6965 | return; |
7257b535 BP |
6966 | |
6967 | case OFPC_FRAG_REASM: | |
6968 | NOT_REACHED(); | |
6969 | ||
6970 | case OFPC_FRAG_NX_MATCH: | |
6971 | /* Nothing to do. */ | |
6972 | break; | |
f0fd1a17 PS |
6973 | |
6974 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
6975 | NOT_REACHED(); | |
7257b535 BP |
6976 | } |
6977 | } | |
6978 | ||
bbafd73b EJ |
6979 | in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port); |
6980 | special = process_special(ctx.ofproto, &ctx.xin->flow, in_port, | |
6981 | ctx.xin->packet); | |
6a7e895f | 6982 | if (special) { |
bbafd73b | 6983 | ctx.xout->slow = special; |
abe529af | 6984 | } else { |
6a6455e5 | 6985 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
14f94f9a | 6986 | struct initial_vals initial_vals; |
94aa0d19 | 6987 | size_t sample_actions_len; |
ee382d89 | 6988 | uint32_t local_odp_port; |
6a6455e5 | 6989 | |
bbafd73b | 6990 | initial_vals.vlan_tci = ctx.base_flow.vlan_tci; |
14f94f9a | 6991 | |
bbafd73b EJ |
6992 | add_sflow_action(&ctx); |
6993 | add_ipfix_action(&ctx); | |
6994 | sample_actions_len = ctx.xout->odp_actions.size; | |
ffaef958 | 6995 | |
bbafd73b EJ |
6996 | if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) { |
6997 | do_xlate_actions(ofpacts, ofpacts_len, &ctx); | |
ffaef958 BP |
6998 | |
6999 | /* We've let OFPP_NORMAL and the learning action look at the | |
7000 | * packet, so drop it now if forwarding is disabled. */ | |
7001 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
bbafd73b | 7002 | ctx.xout->odp_actions.size = sample_actions_len; |
ffaef958 BP |
7003 | } |
7004 | } | |
abe529af | 7005 | |
bbafd73b | 7006 | if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) { |
43d50bc8 BP |
7007 | if (!hit_resubmit_limit) { |
7008 | /* We didn't record the original flow. Make sure we do from | |
7009 | * now on. */ | |
7010 | hit_resubmit_limit = true; | |
7011 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
7012 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7013 | ||
bbafd73b | 7014 | ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, |
14f94f9a | 7015 | &initial_vals, &ds); |
43d50bc8 BP |
7016 | VLOG_ERR("Trace triggered by excessive resubmit " |
7017 | "recursion:\n%s", ds_cstr(&ds)); | |
7018 | ds_destroy(&ds); | |
7019 | } | |
6a6455e5 EJ |
7020 | } |
7021 | ||
bbafd73b EJ |
7022 | local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL); |
7023 | if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow, | |
454a77e5 | 7024 | local_odp_port, |
bbafd73b EJ |
7025 | ctx.xout->odp_actions.data, |
7026 | ctx.xout->odp_actions.size)) { | |
7027 | compose_output_action(&ctx, OFPP_LOCAL); | |
b6848f13 | 7028 | } |
bbafd73b EJ |
7029 | if (ctx.ofproto->has_mirrors) { |
7030 | add_mirror_actions(&ctx, &orig_flow); | |
ccb7c863 | 7031 | } |
bbafd73b | 7032 | fix_sflow_action(&ctx); |
abe529af | 7033 | } |
bd85dac1 | 7034 | |
bbafd73b | 7035 | ofpbuf_uninit(&ctx.stack); |
050ac423 BP |
7036 | } |
7037 | ||
f25d0cf3 BP |
7038 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
7039 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 | 7040 | static void |
bbafd73b | 7041 | xlate_actions_for_side_effects(struct xlate_in *xin) |
050ac423 | 7042 | { |
bbafd73b | 7043 | struct xlate_out xout; |
abe529af | 7044 | |
bbafd73b EJ |
7045 | xlate_actions(xin, &xout); |
7046 | xlate_out_uninit(&xout); | |
abe529af | 7047 | } |
479df176 BP |
7048 | |
7049 | static void | |
f03a84b9 | 7050 | xlate_report(struct xlate_ctx *ctx, const char *s) |
479df176 | 7051 | { |
bbafd73b EJ |
7052 | if (ctx->xin->report_hook) { |
7053 | ctx->xin->report_hook(ctx, s); | |
479df176 BP |
7054 | } |
7055 | } | |
abe529af BP |
7056 | \f |
7057 | /* OFPP_NORMAL implementation. */ | |
7058 | ||
abe529af BP |
7059 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
7060 | ||
ecac4ebf BP |
7061 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
7062 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
7063 | * the bundle on which the packet was received, returns the VLAN to which the | |
7064 | * packet belongs. | |
7065 | * | |
7066 | * Both 'vid' and the return value are in the range 0...4095. */ | |
7067 | static uint16_t | |
7068 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
7069 | { | |
7070 | switch (in_bundle->vlan_mode) { | |
7071 | case PORT_VLAN_ACCESS: | |
7072 | return in_bundle->vlan; | |
7073 | break; | |
7074 | ||
7075 | case PORT_VLAN_TRUNK: | |
7076 | return vid; | |
7077 | ||
7078 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7079 | case PORT_VLAN_NATIVE_TAGGED: | |
7080 | return vid ? vid : in_bundle->vlan; | |
7081 | ||
7082 | default: | |
7083 | NOT_REACHED(); | |
7084 | } | |
7085 | } | |
7086 | ||
5da5ec37 BP |
7087 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
7088 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
7089 | * a warning. | |
7090 | * | |
7091 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
7092 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
7093 | * 0...4095. */ | |
7094 | static bool | |
7095 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
7096 | { | |
33158a18 JP |
7097 | /* Allow any VID on the OFPP_NONE port. */ |
7098 | if (in_bundle == &ofpp_none_bundle) { | |
7099 | return true; | |
7100 | } | |
7101 | ||
5da5ec37 BP |
7102 | switch (in_bundle->vlan_mode) { |
7103 | case PORT_VLAN_ACCESS: | |
7104 | if (vid) { | |
7105 | if (warn) { | |
7106 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7107 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
7108 | "packet received on port %s configured as VLAN " | |
7109 | "%"PRIu16" access port", | |
7110 | in_bundle->ofproto->up.name, vid, | |
7111 | in_bundle->name, in_bundle->vlan); | |
7112 | } | |
7113 | return false; | |
7114 | } | |
7115 | return true; | |
7116 | ||
7117 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7118 | case PORT_VLAN_NATIVE_TAGGED: | |
7119 | if (!vid) { | |
7120 | /* Port must always carry its native VLAN. */ | |
7121 | return true; | |
7122 | } | |
7123 | /* Fall through. */ | |
7124 | case PORT_VLAN_TRUNK: | |
7125 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
7126 | if (warn) { | |
7127 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7128 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
7129 | "received on port %s not configured for trunking " | |
7130 | "VLAN %"PRIu16, | |
7131 | in_bundle->ofproto->up.name, vid, | |
7132 | in_bundle->name, vid); | |
7133 | } | |
7134 | return false; | |
7135 | } | |
7136 | return true; | |
7137 | ||
7138 | default: | |
7139 | NOT_REACHED(); | |
7140 | } | |
7141 | ||
7142 | } | |
7143 | ||
ecac4ebf BP |
7144 | /* Given 'vlan', the VLAN that a packet belongs to, and |
7145 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
7146 | * that should be included in the 802.1Q header. (If the return value is 0, | |
7147 | * then the 802.1Q header should only be included in the packet if there is a | |
7148 | * nonzero PCP.) | |
7149 | * | |
7150 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
7151 | static uint16_t | |
7152 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
7153 | { | |
7154 | switch (out_bundle->vlan_mode) { | |
7155 | case PORT_VLAN_ACCESS: | |
7156 | return 0; | |
7157 | ||
7158 | case PORT_VLAN_TRUNK: | |
7159 | case PORT_VLAN_NATIVE_TAGGED: | |
7160 | return vlan; | |
7161 | ||
7162 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7163 | return vlan == out_bundle->vlan ? 0 : vlan; | |
7164 | ||
7165 | default: | |
7166 | NOT_REACHED(); | |
7167 | } | |
7168 | } | |
7169 | ||
395e68ce | 7170 | static void |
f03a84b9 | 7171 | output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle, |
395e68ce | 7172 | uint16_t vlan) |
abe529af | 7173 | { |
395e68ce BP |
7174 | struct ofport_dpif *port; |
7175 | uint16_t vid; | |
81b1afb1 | 7176 | ovs_be16 tci, old_tci; |
ecac4ebf | 7177 | |
395e68ce BP |
7178 | vid = output_vlan_to_vid(out_bundle, vlan); |
7179 | if (!out_bundle->bond) { | |
7180 | port = ofbundle_get_a_port(out_bundle); | |
7181 | } else { | |
bbafd73b EJ |
7182 | port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow, |
7183 | vid, &ctx->xout->tags); | |
395e68ce BP |
7184 | if (!port) { |
7185 | /* No slaves enabled, so drop packet. */ | |
7186 | return; | |
7187 | } | |
7188 | } | |
abe529af | 7189 | |
bbafd73b | 7190 | old_tci = ctx->xin->flow.vlan_tci; |
5e9ceccd BP |
7191 | tci = htons(vid); |
7192 | if (tci || out_bundle->use_priority_tags) { | |
bbafd73b | 7193 | tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK); |
5e9ceccd BP |
7194 | if (tci) { |
7195 | tci |= htons(VLAN_CFI); | |
7196 | } | |
395e68ce | 7197 | } |
bbafd73b | 7198 | ctx->xin->flow.vlan_tci = tci; |
395e68ce | 7199 | |
5e48dc2b | 7200 | compose_output_action(ctx, port->up.ofp_port); |
bbafd73b | 7201 | ctx->xin->flow.vlan_tci = old_tci; |
abe529af BP |
7202 | } |
7203 | ||
7204 | static int | |
7205 | mirror_mask_ffs(mirror_mask_t mask) | |
7206 | { | |
7207 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
7208 | return ffs(mask); | |
7209 | } | |
7210 | ||
abe529af BP |
7211 | static bool |
7212 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7213 | { | |
ecac4ebf | 7214 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 7215 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
7216 | } |
7217 | ||
7218 | static bool | |
7219 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7220 | { | |
7221 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
7222 | } | |
7223 | ||
7224 | /* Returns an arbitrary interface within 'bundle'. */ | |
7225 | static struct ofport_dpif * | |
7226 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
7227 | { | |
7228 | return CONTAINER_OF(list_front(&bundle->ports), | |
7229 | struct ofport_dpif, bundle_node); | |
7230 | } | |
7231 | ||
abe529af BP |
7232 | static bool |
7233 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
7234 | { | |
fc3d7408 | 7235 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
7236 | } |
7237 | ||
7238 | static void | |
f03a84b9 | 7239 | add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
7240 | { |
7241 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7242 | mirror_mask_t mirrors; | |
c06bba01 JP |
7243 | struct ofbundle *in_bundle; |
7244 | uint16_t vlan; | |
7245 | uint16_t vid; | |
7246 | const struct nlattr *a; | |
7247 | size_t left; | |
7248 | ||
3581c12c | 7249 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
bbafd73b | 7250 | ctx->xin->packet != NULL, NULL); |
3581c12c | 7251 | if (!in_bundle) { |
c06bba01 JP |
7252 | return; |
7253 | } | |
c06bba01 JP |
7254 | mirrors = in_bundle->src_mirrors; |
7255 | ||
7256 | /* Drop frames on bundles reserved for mirroring. */ | |
7257 | if (in_bundle->mirror_out) { | |
bbafd73b | 7258 | if (ctx->xin->packet != NULL) { |
c06bba01 JP |
7259 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7260 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7261 | "%s, which is reserved exclusively for mirroring", | |
7262 | ctx->ofproto->up.name, in_bundle->name); | |
7263 | } | |
7264 | return; | |
7265 | } | |
7266 | ||
7267 | /* Check VLAN. */ | |
7268 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
bbafd73b | 7269 | if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) { |
c06bba01 JP |
7270 | return; |
7271 | } | |
7272 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7273 | ||
7274 | /* Look at the output ports to check for destination selections. */ | |
7275 | ||
bbafd73b EJ |
7276 | NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data, |
7277 | ctx->xout->odp_actions.size) { | |
c06bba01 JP |
7278 | enum ovs_action_attr type = nl_attr_type(a); |
7279 | struct ofport_dpif *ofport; | |
7280 | ||
7281 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
7282 | continue; | |
7283 | } | |
7284 | ||
7285 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
7286 | if (ofport && ofport->bundle) { |
7287 | mirrors |= ofport->bundle->dst_mirrors; | |
7288 | } | |
c06bba01 | 7289 | } |
abe529af BP |
7290 | |
7291 | if (!mirrors) { | |
7292 | return; | |
7293 | } | |
7294 | ||
c06bba01 | 7295 | /* Restore the original packet before adding the mirror actions. */ |
bbafd73b | 7296 | ctx->xin->flow = *orig_flow; |
c06bba01 | 7297 | |
9ba15e2a BP |
7298 | while (mirrors) { |
7299 | struct ofmirror *m; | |
9ba15e2a BP |
7300 | |
7301 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7302 | ||
7303 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 7304 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
7305 | continue; |
7306 | } | |
7307 | ||
7308 | mirrors &= ~m->dup_mirrors; | |
bbafd73b | 7309 | ctx->xout->mirrors |= m->dup_mirrors; |
9ba15e2a | 7310 | if (m->out) { |
395e68ce | 7311 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
7312 | } else if (vlan != m->out_vlan |
7313 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
7314 | struct ofbundle *bundle; |
7315 | ||
7316 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
7317 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
7318 | && !bundle->mirror_out) { |
7319 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
7320 | } |
7321 | } | |
7322 | } | |
abe529af BP |
7323 | } |
7324 | } | |
7325 | ||
9d24de3b JP |
7326 | static void |
7327 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
7328 | uint64_t packets, uint64_t bytes) | |
7329 | { | |
7330 | if (!mirrors) { | |
7331 | return; | |
7332 | } | |
7333 | ||
8472a3ce | 7334 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
7335 | struct ofmirror *m; |
7336 | ||
7337 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7338 | ||
7339 | if (!m) { | |
7340 | /* In normal circumstances 'm' will not be NULL. However, | |
7341 | * if mirrors are reconfigured, we can temporarily get out | |
7342 | * of sync in facet_revalidate(). We could "correct" the | |
7343 | * mirror list before reaching here, but doing that would | |
7344 | * not properly account the traffic stats we've currently | |
7345 | * accumulated for previous mirror configuration. */ | |
7346 | continue; | |
7347 | } | |
7348 | ||
7349 | m->packet_count += packets; | |
7350 | m->byte_count += bytes; | |
7351 | } | |
7352 | } | |
7353 | ||
abe529af BP |
7354 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
7355 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7356 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7357 | static bool | |
7358 | is_gratuitous_arp(const struct flow *flow) | |
7359 | { | |
7360 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7361 | && eth_addr_is_broadcast(flow->dl_dst) | |
7362 | && (flow->nw_proto == ARP_OP_REPLY | |
7363 | || (flow->nw_proto == ARP_OP_REQUEST | |
7364 | && flow->nw_src == flow->nw_dst))); | |
7365 | } | |
7366 | ||
7367 | static void | |
7368 | update_learning_table(struct ofproto_dpif *ofproto, | |
7369 | const struct flow *flow, int vlan, | |
7370 | struct ofbundle *in_bundle) | |
7371 | { | |
7372 | struct mac_entry *mac; | |
7373 | ||
33158a18 JP |
7374 | /* Don't learn the OFPP_NONE port. */ |
7375 | if (in_bundle == &ofpp_none_bundle) { | |
7376 | return; | |
7377 | } | |
7378 | ||
abe529af BP |
7379 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
7380 | return; | |
7381 | } | |
7382 | ||
7383 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7384 | if (is_gratuitous_arp(flow)) { | |
7385 | /* We don't want to learn from gratuitous ARP packets that are | |
7386 | * reflected back over bond slaves so we lock the learning table. */ | |
7387 | if (!in_bundle->bond) { | |
7388 | mac_entry_set_grat_arp_lock(mac); | |
7389 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7390 | return; | |
7391 | } | |
7392 | } | |
7393 | ||
7394 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7395 | /* The log messages here could actually be useful in debugging, | |
7396 | * so keep the rate limit relatively high. */ | |
7397 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7398 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7399 | "on port %s in VLAN %d", | |
7400 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7401 | in_bundle->name, vlan); | |
7402 | ||
7403 | mac->port.p = in_bundle; | |
2cc3c58e | 7404 | tag_set_add(&ofproto->backer->revalidate_set, |
abe529af BP |
7405 | mac_learning_changed(ofproto->ml, mac)); |
7406 | } | |
7407 | } | |
7408 | ||
3581c12c | 7409 | static struct ofbundle * |
4acbc98d SH |
7410 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
7411 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
7412 | { |
7413 | struct ofport_dpif *ofport; | |
7414 | ||
7415 | /* Find the port and bundle for the received packet. */ | |
7416 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
7417 | if (in_ofportp) { |
7418 | *in_ofportp = ofport; | |
7419 | } | |
395e68ce | 7420 | if (ofport && ofport->bundle) { |
3581c12c | 7421 | return ofport->bundle; |
395e68ce BP |
7422 | } |
7423 | ||
70c2fd56 BP |
7424 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
7425 | * port for traffic that it is sourcing. */ | |
7426 | if (in_port == OFPP_NONE) { | |
7427 | return &ofpp_none_bundle; | |
7428 | } | |
7429 | ||
395e68ce BP |
7430 | /* Odd. A few possible reasons here: |
7431 | * | |
7432 | * - We deleted a port but there are still a few packets queued up | |
7433 | * from it. | |
7434 | * | |
7435 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7436 | * we don't know about. | |
7437 | * | |
7438 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
7439 | * This is particularly likely to happen if a packet was received on the |
7440 | * port after it was created, but before the client had a chance to | |
7441 | * configure its bundle. | |
395e68ce BP |
7442 | */ |
7443 | if (warn) { | |
7444 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7445 | ||
7446 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7447 | "port %"PRIu16, ofproto->up.name, in_port); | |
7448 | } | |
7449 | return NULL; | |
7450 | } | |
7451 | ||
5da5ec37 | 7452 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
7453 | * dropped. Returns true if they may be forwarded, false if they should be |
7454 | * dropped. | |
7455 | * | |
395e68ce BP |
7456 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
7457 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 7458 | * |
395e68ce BP |
7459 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
7460 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7461 | * checked by input_vid_is_valid(). | |
abe529af BP |
7462 | * |
7463 | * May also add tags to '*tags', although the current implementation only does | |
7464 | * so in one special case. | |
7465 | */ | |
7466 | static bool | |
f03a84b9 | 7467 | is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port, |
479df176 | 7468 | uint16_t vlan) |
abe529af | 7469 | { |
479df176 | 7470 | struct ofproto_dpif *ofproto = ctx->ofproto; |
bbafd73b | 7471 | struct flow *flow = &ctx->xin->flow; |
395e68ce | 7472 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 7473 | |
395e68ce BP |
7474 | /* Drop frames for reserved multicast addresses |
7475 | * only if forward_bpdu option is absent. */ | |
614ec445 | 7476 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 7477 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
7478 | return false; |
7479 | } | |
7480 | ||
abe529af BP |
7481 | if (in_bundle->bond) { |
7482 | struct mac_entry *mac; | |
7483 | ||
7484 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
bbafd73b | 7485 | flow->dl_dst, &ctx->xout->tags)) { |
abe529af BP |
7486 | case BV_ACCEPT: |
7487 | break; | |
7488 | ||
7489 | case BV_DROP: | |
479df176 | 7490 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
7491 | return false; |
7492 | ||
7493 | case BV_DROP_IF_MOVED: | |
7494 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7495 | if (mac && mac->port.p != in_bundle && | |
7496 | (!is_gratuitous_arp(flow) | |
7497 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
7498 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
7499 | "dropping"); | |
abe529af BP |
7500 | return false; |
7501 | } | |
7502 | break; | |
7503 | } | |
7504 | } | |
7505 | ||
7506 | return true; | |
7507 | } | |
7508 | ||
4cd78906 | 7509 | static void |
f03a84b9 | 7510 | xlate_normal(struct xlate_ctx *ctx) |
abe529af | 7511 | { |
395e68ce | 7512 | struct ofport_dpif *in_port; |
abe529af | 7513 | struct ofbundle *in_bundle; |
abe529af | 7514 | struct mac_entry *mac; |
395e68ce BP |
7515 | uint16_t vlan; |
7516 | uint16_t vid; | |
abe529af | 7517 | |
bbafd73b | 7518 | ctx->xout->has_normal = true; |
75a75043 | 7519 | |
bbafd73b EJ |
7520 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port, |
7521 | ctx->xin->packet != NULL, &in_port); | |
3581c12c | 7522 | if (!in_bundle) { |
479df176 | 7523 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
7524 | return; |
7525 | } | |
3581c12c | 7526 | |
395e68ce | 7527 | /* Drop malformed frames. */ |
bbafd73b EJ |
7528 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) && |
7529 | !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) { | |
7530 | if (ctx->xin->packet != NULL) { | |
395e68ce BP |
7531 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7532 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7533 | "VLAN tag received on port %s", | |
7534 | ctx->ofproto->up.name, in_bundle->name); | |
7535 | } | |
479df176 | 7536 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
7537 | return; |
7538 | } | |
7539 | ||
7540 | /* Drop frames on bundles reserved for mirroring. */ | |
7541 | if (in_bundle->mirror_out) { | |
bbafd73b | 7542 | if (ctx->xin->packet != NULL) { |
395e68ce BP |
7543 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7544 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7545 | "%s, which is reserved exclusively for mirroring", | |
7546 | ctx->ofproto->up.name, in_bundle->name); | |
7547 | } | |
479df176 | 7548 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
7549 | return; |
7550 | } | |
7551 | ||
7552 | /* Check VLAN. */ | |
bbafd73b EJ |
7553 | vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci); |
7554 | if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) { | |
479df176 | 7555 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
7556 | return; |
7557 | } | |
7558 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7559 | ||
7560 | /* Check other admissibility requirements. */ | |
479df176 | 7561 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 7562 | return; |
abe529af BP |
7563 | } |
7564 | ||
75a75043 | 7565 | /* Learn source MAC. */ |
bbafd73b EJ |
7566 | if (ctx->xin->may_learn) { |
7567 | update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle); | |
abe529af BP |
7568 | } |
7569 | ||
7570 | /* Determine output bundle. */ | |
bbafd73b EJ |
7571 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan, |
7572 | &ctx->xout->tags); | |
abe529af | 7573 | if (mac) { |
c06bba01 | 7574 | if (mac->port.p != in_bundle) { |
479df176 | 7575 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 7576 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
7577 | } else { |
7578 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 7579 | } |
abe529af | 7580 | } else { |
c06bba01 | 7581 | struct ofbundle *bundle; |
abe529af | 7582 | |
479df176 | 7583 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
7584 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
7585 | if (bundle != in_bundle | |
7586 | && ofbundle_includes_vlan(bundle, vlan) | |
7587 | && bundle->floodable | |
7588 | && !bundle->mirror_out) { | |
7589 | output_normal(ctx, bundle, vlan); | |
7590 | } | |
7591 | } | |
bbafd73b | 7592 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
abe529af | 7593 | } |
abe529af BP |
7594 | } |
7595 | \f | |
54a9cbc9 BP |
7596 | /* Optimized flow revalidation. |
7597 | * | |
7598 | * It's a difficult problem, in general, to tell which facets need to have | |
7599 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7600 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7601 | * table changes, we recalculate the actions for every facet. This is | |
7602 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7603 | * doesn't change very often. | |
7604 | * | |
7605 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7606 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7607 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7608 | * table, we add a special case that applies to flow tables in which every rule | |
7609 | * has the same form (that is, the same wildcards), except that the table is | |
7610 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7611 | * optimize this case by tagging all of the facets that resubmit into the table | |
7612 | * and invalidating the same tag whenever a flow changes in that table. The | |
7613 | * end result is that we revalidate just the facets that need it (and sometimes | |
7614 | * a few more, but not all of the facets or even all of the facets that | |
7615 | * resubmit to the table modified by MAC learning). */ | |
7616 | ||
5cb7a798 | 7617 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted |
54a9cbc9 | 7618 | * into an OpenFlow table with the given 'basis'. */ |
822d9414 | 7619 | static tag_type |
5cb7a798 | 7620 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, |
54a9cbc9 BP |
7621 | uint32_t secret) |
7622 | { | |
5cb7a798 | 7623 | if (minimask_is_catchall(mask)) { |
54a9cbc9 BP |
7624 | return 0; |
7625 | } else { | |
5cb7a798 BP |
7626 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); |
7627 | return tag_create_deterministic(hash); | |
54a9cbc9 BP |
7628 | } |
7629 | } | |
7630 | ||
7631 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7632 | * taggability of that table. | |
7633 | * | |
7634 | * This function must be called after *each* change to a flow table. If you | |
7635 | * skip calling it on some changes then the pointer comparisons at the end can | |
7636 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7637 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7638 | * different wildcards to be created with the same address, then this function | |
7639 | * will incorrectly skip revalidation. */ | |
7640 | static void | |
7641 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7642 | { | |
7643 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 7644 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
7645 | struct cls_table *catchall, *other; |
7646 | struct cls_table *t; | |
7647 | ||
7648 | catchall = other = NULL; | |
7649 | ||
d0918789 | 7650 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
7651 | case 0: |
7652 | /* We could tag this OpenFlow table but it would make the logic a | |
7653 | * little harder and it's a corner case that doesn't seem worth it | |
7654 | * yet. */ | |
7655 | break; | |
7656 | ||
7657 | case 1: | |
7658 | case 2: | |
d0918789 | 7659 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
7660 | if (cls_table_is_catchall(t)) { |
7661 | catchall = t; | |
7662 | } else if (!other) { | |
7663 | other = t; | |
7664 | } else { | |
7665 | /* Indicate that we can't tag this by setting both tables to | |
7666 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7667 | other = NULL; | |
7668 | } | |
7669 | } | |
7670 | break; | |
7671 | ||
7672 | default: | |
7673 | /* Can't tag this table. */ | |
7674 | break; | |
7675 | } | |
7676 | ||
7677 | if (table->catchall_table != catchall || table->other_table != other) { | |
7678 | table->catchall_table = catchall; | |
7679 | table->other_table = other; | |
2cc3c58e | 7680 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7681 | } |
7682 | } | |
7683 | ||
7684 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7685 | * inserted, a rule being deleted, or a rule whose actions are being | |
7686 | * modified), marks facets for revalidation to ensure that packets will be | |
7687 | * forwarded correctly according to the new state of the flow table. | |
7688 | * | |
7689 | * This function must be called after *each* change to a flow table. See | |
7690 | * the comment on table_update_taggable() for more information. */ | |
7691 | static void | |
7692 | rule_invalidate(const struct rule_dpif *rule) | |
7693 | { | |
7694 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7695 | ||
7696 | table_update_taggable(ofproto, rule->up.table_id); | |
7697 | ||
2cc3c58e | 7698 | if (!ofproto->backer->need_revalidate) { |
54a9cbc9 BP |
7699 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; |
7700 | ||
7701 | if (table->other_table && rule->tag) { | |
2cc3c58e | 7702 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); |
54a9cbc9 | 7703 | } else { |
2cc3c58e | 7704 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7705 | } |
7706 | } | |
7707 | } | |
7708 | \f | |
abe529af | 7709 | static bool |
7257b535 BP |
7710 | set_frag_handling(struct ofproto *ofproto_, |
7711 | enum ofp_config_flags frag_handling) | |
abe529af BP |
7712 | { |
7713 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 7714 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 7715 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
7716 | return true; |
7717 | } else { | |
7718 | return false; | |
7719 | } | |
abe529af BP |
7720 | } |
7721 | ||
90bf1e07 | 7722 | static enum ofperr |
abe529af BP |
7723 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
7724 | const struct flow *flow, | |
f25d0cf3 | 7725 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
7726 | { |
7727 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
14f94f9a | 7728 | struct initial_vals initial_vals; |
548de4dd BP |
7729 | struct odputil_keybuf keybuf; |
7730 | struct dpif_flow_stats stats; | |
bbafd73b EJ |
7731 | struct xlate_out xout; |
7732 | struct xlate_in xin; | |
548de4dd | 7733 | struct ofpbuf key; |
112bc5f4 | 7734 | |
80e5eed9 | 7735 | |
548de4dd | 7736 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
7737 | odp_flow_key_from_flow(&key, flow, |
7738 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
050ac423 | 7739 | |
548de4dd | 7740 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 7741 | |
14f94f9a | 7742 | initial_vals.vlan_tci = flow->vlan_tci; |
bbafd73b EJ |
7743 | xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags, |
7744 | packet); | |
7745 | xin.resubmit_stats = &stats; | |
7746 | xin.ofpacts_len = ofpacts_len; | |
7747 | xin.ofpacts = ofpacts; | |
2284188b | 7748 | |
bbafd73b | 7749 | xlate_actions(&xin, &xout); |
acf60855 | 7750 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
bbafd73b EJ |
7751 | xout.odp_actions.data, xout.odp_actions.size, packet); |
7752 | xlate_out_uninit(&xout); | |
2284188b | 7753 | |
548de4dd | 7754 | return 0; |
abe529af | 7755 | } |
6fca1ffb BP |
7756 | \f |
7757 | /* NetFlow. */ | |
7758 | ||
7759 | static int | |
7760 | set_netflow(struct ofproto *ofproto_, | |
7761 | const struct netflow_options *netflow_options) | |
7762 | { | |
7763 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7764 | ||
7765 | if (netflow_options) { | |
7766 | if (!ofproto->netflow) { | |
7767 | ofproto->netflow = netflow_create(); | |
7768 | } | |
7769 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7770 | } else { | |
7771 | netflow_destroy(ofproto->netflow); | |
7772 | ofproto->netflow = NULL; | |
7773 | return 0; | |
7774 | } | |
7775 | } | |
abe529af BP |
7776 | |
7777 | static void | |
7778 | get_netflow_ids(const struct ofproto *ofproto_, | |
7779 | uint8_t *engine_type, uint8_t *engine_id) | |
7780 | { | |
7781 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7782 | ||
acf60855 | 7783 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 7784 | } |
6fca1ffb BP |
7785 | |
7786 | static void | |
7787 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7788 | { | |
7789 | if (!facet_is_controller_flow(facet) && | |
7790 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 7791 | struct subfacet *subfacet; |
6fca1ffb BP |
7792 | struct ofexpired expired; |
7793 | ||
b0f7b9b5 | 7794 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 7795 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 7796 | struct dpif_flow_stats stats; |
6fca1ffb | 7797 | |
bbafd73b | 7798 | subfacet_install(subfacet, &facet->xout.odp_actions, &stats); |
15baa734 | 7799 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 7800 | } |
6fca1ffb BP |
7801 | } |
7802 | ||
7803 | expired.flow = facet->flow; | |
7804 | expired.packet_count = facet->packet_count; | |
7805 | expired.byte_count = facet->byte_count; | |
7806 | expired.used = facet->used; | |
7807 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7808 | } | |
7809 | } | |
7810 | ||
7811 | static void | |
7812 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7813 | { | |
7814 | struct facet *facet; | |
7815 | ||
7816 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7817 | send_active_timeout(ofproto, facet); | |
7818 | } | |
7819 | } | |
abe529af BP |
7820 | \f |
7821 | static struct ofproto_dpif * | |
7822 | ofproto_dpif_lookup(const char *name) | |
7823 | { | |
b44a10b7 BP |
7824 | struct ofproto_dpif *ofproto; |
7825 | ||
7826 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7827 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7828 | if (!strcmp(ofproto->up.name, name)) { | |
7829 | return ofproto; | |
7830 | } | |
7831 | } | |
7832 | return NULL; | |
abe529af BP |
7833 | } |
7834 | ||
f0a3aa2e | 7835 | static void |
96e466a3 | 7836 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 7837 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 7838 | { |
490df1ef | 7839 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 7840 | |
96e466a3 EJ |
7841 | if (argc > 1) { |
7842 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7843 | if (!ofproto) { | |
bde9f75d | 7844 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
7845 | return; |
7846 | } | |
2cc3c58e | 7847 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 EJ |
7848 | } else { |
7849 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2cc3c58e | 7850 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 | 7851 | } |
f0a3aa2e | 7852 | } |
f0a3aa2e | 7853 | |
bde9f75d | 7854 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
7855 | } |
7856 | ||
abe529af | 7857 | static void |
0e15264f BP |
7858 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
7859 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
7860 | { |
7861 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7862 | const struct ofproto_dpif *ofproto; | |
7863 | const struct mac_entry *e; | |
7864 | ||
0e15264f | 7865 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 7866 | if (!ofproto) { |
bde9f75d | 7867 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
7868 | return; |
7869 | } | |
7870 | ||
7871 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
7872 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
7873 | struct ofbundle *bundle = e->port.p; | |
7874 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
7875 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
7876 | e->vlan, ETH_ADDR_ARGS(e->mac), |
7877 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 7878 | } |
bde9f75d | 7879 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
7880 | ds_destroy(&ds); |
7881 | } | |
7882 | ||
6a6455e5 | 7883 | struct trace_ctx { |
bbafd73b EJ |
7884 | struct xlate_out xout; |
7885 | struct xlate_in xin; | |
abe529af BP |
7886 | struct flow flow; |
7887 | struct ds *result; | |
7888 | }; | |
7889 | ||
7890 | static void | |
29901626 BP |
7891 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
7892 | const struct rule_dpif *rule) | |
abe529af BP |
7893 | { |
7894 | ds_put_char_multiple(result, '\t', level); | |
7895 | if (!rule) { | |
7896 | ds_put_cstr(result, "No match\n"); | |
7897 | return; | |
7898 | } | |
7899 | ||
29901626 BP |
7900 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
7901 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 7902 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
7903 | ds_put_char(result, '\n'); |
7904 | ||
7905 | ds_put_char_multiple(result, '\t', level); | |
7906 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 7907 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
7908 | ds_put_char(result, '\n'); |
7909 | } | |
7910 | ||
7911 | static void | |
7912 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 7913 | struct trace_ctx *trace) |
abe529af BP |
7914 | { |
7915 | ds_put_char_multiple(result, '\t', level); | |
7916 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 7917 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
7918 | ds_put_cstr(result, "unchanged"); |
7919 | } else { | |
bbafd73b EJ |
7920 | flow_format(result, &trace->xin.flow); |
7921 | trace->flow = trace->xin.flow; | |
abe529af BP |
7922 | } |
7923 | ds_put_char(result, '\n'); | |
7924 | } | |
7925 | ||
eb9e1c26 EJ |
7926 | static void |
7927 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 7928 | struct trace_ctx *trace) |
eb9e1c26 EJ |
7929 | { |
7930 | size_t i; | |
7931 | ||
7932 | ds_put_char_multiple(result, '\t', level); | |
7933 | ds_put_format(result, "%s:", title); | |
7934 | for (i = 0; i < FLOW_N_REGS; i++) { | |
7935 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
7936 | } | |
7937 | ds_put_char(result, '\n'); | |
7938 | } | |
7939 | ||
1ed8d352 EJ |
7940 | static void |
7941 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 7942 | struct trace_ctx *trace) |
1ed8d352 | 7943 | { |
bbafd73b | 7944 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
7945 | |
7946 | ds_put_char_multiple(result, '\t', level); | |
7947 | ds_put_format(result, "%s: ", title); | |
7948 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
7949 | ds_put_char(result, '\n'); | |
7950 | } | |
7951 | ||
abe529af | 7952 | static void |
f03a84b9 | 7953 | trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule) |
abe529af | 7954 | { |
bbafd73b | 7955 | struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin); |
abe529af BP |
7956 | struct ds *result = trace->result; |
7957 | ||
7958 | ds_put_char(result, '\n'); | |
7959 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 7960 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 7961 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 7962 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
7963 | } |
7964 | ||
479df176 | 7965 | static void |
f03a84b9 | 7966 | trace_report(struct xlate_ctx *ctx, const char *s) |
479df176 | 7967 | { |
bbafd73b | 7968 | struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin); |
479df176 BP |
7969 | struct ds *result = trace->result; |
7970 | ||
7971 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
7972 | ds_put_cstr(result, s); | |
7973 | ds_put_char(result, '\n'); | |
7974 | } | |
7975 | ||
abe529af | 7976 | static void |
0e15264f | 7977 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
7978 | void *aux OVS_UNUSED) |
7979 | { | |
50aa28fd | 7980 | const struct dpif_backer *backer; |
abe529af | 7981 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
7982 | struct ofpbuf odp_key; |
7983 | struct ofpbuf *packet; | |
14f94f9a | 7984 | struct initial_vals initial_vals; |
abe529af BP |
7985 | struct ds result; |
7986 | struct flow flow; | |
abe529af BP |
7987 | char *s; |
7988 | ||
876b0e1c | 7989 | packet = NULL; |
50aa28fd | 7990 | backer = NULL; |
abe529af | 7991 | ds_init(&result); |
50aa28fd | 7992 | ofpbuf_init(&odp_key, 0); |
abe529af | 7993 | |
50aa28fd AW |
7994 | /* Handle "-generate" or a hex string as the last argument. */ |
7995 | if (!strcmp(argv[argc - 1], "-generate")) { | |
7996 | packet = ofpbuf_new(0); | |
7997 | argc--; | |
7998 | } else { | |
7999 | const char *error = eth_from_hex(argv[argc - 1], &packet); | |
8000 | if (!error) { | |
8001 | argc--; | |
8002 | } else if (argc == 4) { | |
8003 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
8004 | unixctl_command_reply_error(conn, error); | |
8005 | goto exit; | |
8006 | } | |
e84173dc | 8007 | } |
876b0e1c | 8008 | |
50aa28fd AW |
8009 | /* Parse the flow and determine whether a datapath or |
8010 | * bridge is specified. If function odp_flow_key_from_string() | |
8011 | * returns 0, the flow is a odp_flow. If function | |
8012 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
8013 | if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) { | |
8014 | /* If the odp_flow is the second argument, | |
8015 | * the datapath name is the first argument. */ | |
8016 | if (argc == 3) { | |
8017 | const char *dp_type; | |
8018 | if (!strncmp(argv[1], "ovs-", 4)) { | |
8019 | dp_type = argv[1] + 4; | |
8020 | } else { | |
8021 | dp_type = argv[1]; | |
31a19d69 | 8022 | } |
50aa28fd AW |
8023 | backer = shash_find_data(&all_dpif_backers, dp_type); |
8024 | if (!backer) { | |
8025 | unixctl_command_reply_error(conn, "Cannot find datapath " | |
8026 | "of this name"); | |
31a19d69 BP |
8027 | goto exit; |
8028 | } | |
8029 | } else { | |
50aa28fd AW |
8030 | /* No datapath name specified, so there should be only one |
8031 | * datapath. */ | |
8032 | struct shash_node *node; | |
8033 | if (shash_count(&all_dpif_backers) != 1) { | |
8034 | unixctl_command_reply_error(conn, "Must specify datapath " | |
8035 | "name, there is more than one type of datapath"); | |
31a19d69 BP |
8036 | goto exit; |
8037 | } | |
50aa28fd AW |
8038 | node = shash_first(&all_dpif_backers); |
8039 | backer = node->data; | |
876b0e1c | 8040 | } |
8b3b8dd1 | 8041 | |
50aa28fd AW |
8042 | /* Extract the ofproto_dpif object from the ofproto_receive() |
8043 | * function. */ | |
8044 | if (ofproto_receive(backer, NULL, odp_key.data, | |
8045 | odp_key.size, &flow, NULL, &ofproto, NULL, | |
8046 | &initial_vals)) { | |
8047 | unixctl_command_reply_error(conn, "Invalid datapath flow"); | |
8048 | goto exit; | |
8b3b8dd1 | 8049 | } |
50aa28fd AW |
8050 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
8051 | } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { | |
8052 | if (argc != 3) { | |
8053 | unixctl_command_reply_error(conn, "Must specify bridge name"); | |
876b0e1c BP |
8054 | goto exit; |
8055 | } | |
8056 | ||
50aa28fd AW |
8057 | ofproto = ofproto_dpif_lookup(argv[1]); |
8058 | if (!ofproto) { | |
8059 | unixctl_command_reply_error(conn, "Unknown bridge name"); | |
8060 | goto exit; | |
8061 | } | |
14f94f9a | 8062 | initial_vals.vlan_tci = flow.vlan_tci; |
876b0e1c | 8063 | } else { |
50aa28fd | 8064 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
abe529af BP |
8065 | goto exit; |
8066 | } | |
8067 | ||
50aa28fd AW |
8068 | /* Generate a packet, if requested. */ |
8069 | if (packet) { | |
8070 | if (!packet->size) { | |
8071 | flow_compose(packet, &flow); | |
8072 | } else { | |
8073 | ds_put_cstr(&result, "Packet: "); | |
8074 | s = ofp_packet_to_string(packet->data, packet->size); | |
8075 | ds_put_cstr(&result, s); | |
8076 | free(s); | |
8077 | ||
8078 | /* Use the metadata from the flow and the packet argument | |
8079 | * to reconstruct the flow. */ | |
8080 | flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, | |
8081 | flow.in_port, &flow); | |
8082 | initial_vals.vlan_tci = flow.vlan_tci; | |
8083 | } | |
8084 | } | |
8085 | ||
14f94f9a | 8086 | ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); |
6a6455e5 EJ |
8087 | unixctl_command_reply(conn, ds_cstr(&result)); |
8088 | ||
8089 | exit: | |
8090 | ds_destroy(&result); | |
8091 | ofpbuf_delete(packet); | |
8092 | ofpbuf_uninit(&odp_key); | |
8093 | } | |
8094 | ||
8095 | static void | |
8096 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
8097 | const struct ofpbuf *packet, |
8098 | const struct initial_vals *initial_vals, struct ds *ds) | |
6a6455e5 EJ |
8099 | { |
8100 | struct rule_dpif *rule; | |
8101 | ||
8102 | ds_put_cstr(ds, "Flow: "); | |
8103 | flow_format(ds, flow); | |
8104 | ds_put_char(ds, '\n'); | |
abe529af | 8105 | |
c57b2226 BP |
8106 | rule = rule_dpif_lookup(ofproto, flow); |
8107 | ||
6a6455e5 | 8108 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
8109 | if (rule == ofproto->miss_rule) { |
8110 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
8111 | } else if (rule == ofproto->no_packet_in_rule) { | |
8112 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
8113 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7fd51d39 BP |
8114 | } else if (rule == ofproto->drop_frags_rule) { |
8115 | ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " | |
8116 | "and the fragment handling mode is \"drop\".\n"); | |
c57b2226 BP |
8117 | } |
8118 | ||
abe529af | 8119 | if (rule) { |
050ac423 BP |
8120 | uint64_t odp_actions_stub[1024 / 8]; |
8121 | struct ofpbuf odp_actions; | |
8122 | ||
6a6455e5 | 8123 | struct trace_ctx trace; |
0e553d9c | 8124 | uint8_t tcp_flags; |
abe529af | 8125 | |
6a6455e5 EJ |
8126 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
8127 | trace.result = ds; | |
8128 | trace.flow = *flow; | |
050ac423 BP |
8129 | ofpbuf_use_stub(&odp_actions, |
8130 | odp_actions_stub, sizeof odp_actions_stub); | |
bbafd73b EJ |
8131 | xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags, |
8132 | packet); | |
8133 | trace.xin.resubmit_hook = trace_resubmit; | |
8134 | trace.xin.report_hook = trace_report; | |
8135 | xlate_actions(&trace.xin, &trace.xout); | |
abe529af | 8136 | |
6a6455e5 EJ |
8137 | ds_put_char(ds, '\n'); |
8138 | trace_format_flow(ds, 0, "Final flow", &trace); | |
8139 | ds_put_cstr(ds, "Datapath actions: "); | |
bbafd73b EJ |
8140 | format_odp_actions(ds, trace.xout.odp_actions.data, |
8141 | trace.xout.odp_actions.size); | |
876b0e1c | 8142 | |
bbafd73b | 8143 | if (trace.xout.slow) { |
6a7e895f BP |
8144 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
8145 | "slow path because it:"); | |
bbafd73b | 8146 | switch (trace.xout.slow) { |
98f0520f EJ |
8147 | case SLOW_CFM: |
8148 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
8149 | break; | |
8150 | case SLOW_LACP: | |
8151 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
8152 | break; | |
8153 | case SLOW_STP: | |
8154 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
8155 | break; | |
8156 | case SLOW_BFD: | |
8157 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
8158 | break; | |
8159 | case SLOW_CONTROLLER: | |
8160 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
8161 | "to the OpenFlow controller."); | |
8162 | break; | |
8163 | case __SLOW_MAX: | |
8164 | NOT_REACHED(); | |
6a7e895f | 8165 | } |
876b0e1c | 8166 | } |
bbafd73b EJ |
8167 | |
8168 | xlate_out_uninit(&trace.xout); | |
abe529af | 8169 | } |
abe529af BP |
8170 | } |
8171 | ||
7ee20df1 | 8172 | static void |
0e15264f BP |
8173 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8174 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8175 | { |
8176 | clogged = true; | |
bde9f75d | 8177 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8178 | } |
8179 | ||
8180 | static void | |
0e15264f BP |
8181 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8182 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8183 | { |
8184 | clogged = false; | |
bde9f75d | 8185 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8186 | } |
8187 | ||
6814e51f BP |
8188 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
8189 | * 'reply' describing the results. */ | |
8190 | static void | |
8191 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
8192 | { | |
8193 | struct facet *facet; | |
8194 | int errors; | |
8195 | ||
8196 | errors = 0; | |
8197 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
8198 | if (!facet_check_consistency(facet)) { | |
8199 | errors++; | |
8200 | } | |
8201 | } | |
8202 | if (errors) { | |
2cc3c58e | 8203 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
8204 | } |
8205 | ||
8206 | if (errors) { | |
8207 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
8208 | ofproto->up.name, errors); | |
8209 | } else { | |
8210 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
8211 | } | |
8212 | } | |
8213 | ||
8214 | static void | |
8215 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
8216 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
8217 | { | |
8218 | struct ds reply = DS_EMPTY_INITIALIZER; | |
8219 | struct ofproto_dpif *ofproto; | |
8220 | ||
8221 | if (argc > 1) { | |
8222 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8223 | if (!ofproto) { | |
bde9f75d EJ |
8224 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
8225 | "ofproto/list for help)"); | |
6814e51f BP |
8226 | return; |
8227 | } | |
8228 | ofproto_dpif_self_check__(ofproto, &reply); | |
8229 | } else { | |
8230 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8231 | ofproto_dpif_self_check__(ofproto, &reply); | |
8232 | } | |
8233 | } | |
8234 | ||
bde9f75d | 8235 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
8236 | ds_destroy(&reply); |
8237 | } | |
8238 | ||
27022416 JP |
8239 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
8240 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
8241 | * to destroy 'ofproto_shash' and free the returned value. */ | |
8242 | static const struct shash_node ** | |
8243 | get_ofprotos(struct shash *ofproto_shash) | |
8244 | { | |
8245 | const struct ofproto_dpif *ofproto; | |
8246 | ||
8247 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8248 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
8249 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
8250 | } | |
8251 | ||
8252 | return shash_sort(ofproto_shash); | |
8253 | } | |
8254 | ||
8255 | static void | |
8256 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
8257 | const char *argv[] OVS_UNUSED, | |
8258 | void *aux OVS_UNUSED) | |
8259 | { | |
8260 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8261 | struct shash ofproto_shash; | |
8262 | const struct shash_node **sorted_ofprotos; | |
8263 | int i; | |
8264 | ||
8265 | shash_init(&ofproto_shash); | |
8266 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8267 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8268 | const struct shash_node *node = sorted_ofprotos[i]; | |
8269 | ds_put_format(&ds, "%s\n", node->name); | |
8270 | } | |
8271 | ||
8272 | shash_destroy(&ofproto_shash); | |
8273 | free(sorted_ofprotos); | |
8274 | ||
8275 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8276 | ds_destroy(&ds); | |
8277 | } | |
8278 | ||
8279 | static void | |
dc54ef36 EJ |
8280 | show_dp_rates(struct ds *ds, const char *heading, |
8281 | const struct avg_subfacet_rates *rates) | |
8282 | { | |
8283 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
8284 | heading, rates->add_rate, rates->del_rate); | |
8285 | } | |
8286 | ||
8287 | static void | |
8288 | dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) | |
27022416 | 8289 | { |
dc54ef36 EJ |
8290 | const struct shash_node **ofprotos; |
8291 | struct ofproto_dpif *ofproto; | |
8292 | struct shash ofproto_shash; | |
09672174 | 8293 | uint64_t n_hit, n_missed; |
dc54ef36 | 8294 | long long int minutes; |
09672174 | 8295 | size_t i; |
655ab909 | 8296 | |
04d08d54 | 8297 | n_hit = n_missed = 0; |
dc54ef36 EJ |
8298 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
8299 | if (ofproto->backer == backer) { | |
dc54ef36 EJ |
8300 | n_missed += ofproto->n_missed; |
8301 | n_hit += ofproto->n_hit; | |
8302 | } | |
8303 | } | |
655ab909 | 8304 | |
dc54ef36 EJ |
8305 | ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
8306 | dpif_name(backer->dpif), n_hit, n_missed); | |
8307 | ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," | |
04d08d54 | 8308 | " life span: %lldms\n", hmap_count(&backer->subfacets), |
dc54ef36 EJ |
8309 | backer->avg_n_subfacet, backer->max_n_subfacet, |
8310 | backer->avg_subfacet_life); | |
8311 | ||
8312 | minutes = (time_msec() - backer->created) / (1000 * 60); | |
655ab909 | 8313 | if (minutes >= 60) { |
dc54ef36 | 8314 | show_dp_rates(ds, "\thourly avg:", &backer->hourly); |
655ab909 AZ |
8315 | } |
8316 | if (minutes >= 60 * 24) { | |
dc54ef36 | 8317 | show_dp_rates(ds, "\tdaily avg:", &backer->daily); |
655ab909 | 8318 | } |
dc54ef36 | 8319 | show_dp_rates(ds, "\toverall avg:", &backer->lifetime); |
27022416 | 8320 | |
dc54ef36 EJ |
8321 | shash_init(&ofproto_shash); |
8322 | ofprotos = get_ofprotos(&ofproto_shash); | |
8323 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8324 | struct ofproto_dpif *ofproto = ofprotos[i]->data; | |
8325 | const struct shash_node **ports; | |
8326 | size_t j; | |
0a740f48 | 8327 | |
dc54ef36 EJ |
8328 | if (ofproto->backer != backer) { |
8329 | continue; | |
0a740f48 | 8330 | } |
27022416 | 8331 | |
dc54ef36 EJ |
8332 | ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", |
8333 | ofproto->up.name, ofproto->n_hit, ofproto->n_missed); | |
8334 | ||
8335 | ports = shash_sort(&ofproto->up.port_by_name); | |
8336 | for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { | |
8337 | const struct shash_node *node = ports[j]; | |
8338 | struct ofport *ofport = node->data; | |
8339 | struct smap config; | |
8340 | uint32_t odp_port; | |
27022416 | 8341 | |
dc54ef36 EJ |
8342 | ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), |
8343 | ofport->ofp_port); | |
27022416 | 8344 | |
dc54ef36 EJ |
8345 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); |
8346 | if (odp_port != OVSP_NONE) { | |
8347 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
8348 | } else { | |
8349 | ds_put_cstr(ds, "none:"); | |
8350 | } | |
27022416 | 8351 | |
dc54ef36 | 8352 | ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); |
27022416 | 8353 | |
dc54ef36 EJ |
8354 | smap_init(&config); |
8355 | if (!netdev_get_config(ofport->netdev, &config)) { | |
8356 | const struct smap_node **nodes; | |
8357 | size_t i; | |
27022416 | 8358 | |
dc54ef36 EJ |
8359 | nodes = smap_sort(&config); |
8360 | for (i = 0; i < smap_count(&config); i++) { | |
8361 | const struct smap_node *node = nodes[i]; | |
8362 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
8363 | node->key, node->value); | |
8364 | } | |
8365 | free(nodes); | |
27022416 | 8366 | } |
dc54ef36 EJ |
8367 | smap_destroy(&config); |
8368 | ||
27022416 | 8369 | ds_put_char(ds, ')'); |
dc54ef36 | 8370 | ds_put_char(ds, '\n'); |
27022416 | 8371 | } |
dc54ef36 | 8372 | free(ports); |
27022416 | 8373 | } |
dc54ef36 EJ |
8374 | shash_destroy(&ofproto_shash); |
8375 | free(ofprotos); | |
27022416 JP |
8376 | } |
8377 | ||
8378 | static void | |
dc54ef36 EJ |
8379 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
8380 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
27022416 JP |
8381 | { |
8382 | struct ds ds = DS_EMPTY_INITIALIZER; | |
dc54ef36 EJ |
8383 | const struct shash_node **backers; |
8384 | int i; | |
27022416 | 8385 | |
dc54ef36 EJ |
8386 | backers = shash_sort(&all_dpif_backers); |
8387 | for (i = 0; i < shash_count(&all_dpif_backers); i++) { | |
8388 | dpif_show_backer(backers[i]->data, &ds); | |
27022416 | 8389 | } |
dc54ef36 | 8390 | free(backers); |
27022416 JP |
8391 | |
8392 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8393 | ds_destroy(&ds); | |
8394 | } | |
8395 | ||
8396 | static void | |
8397 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8398 | int argc OVS_UNUSED, const char *argv[], | |
8399 | void *aux OVS_UNUSED) | |
8400 | { | |
8401 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8402 | const struct ofproto_dpif *ofproto; | |
8403 | struct subfacet *subfacet; | |
8404 | ||
8405 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8406 | if (!ofproto) { | |
8407 | unixctl_command_reply_error(conn, "no such bridge"); | |
8408 | return; | |
8409 | } | |
8410 | ||
af37354d EJ |
8411 | update_stats(ofproto->backer); |
8412 | ||
04d08d54 | 8413 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { |
4dff9097 EJ |
8414 | struct facet *facet = subfacet->facet; |
8415 | ||
04d08d54 EJ |
8416 | if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) { |
8417 | continue; | |
8418 | } | |
8419 | ||
9566abf9 | 8420 | odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); |
27022416 JP |
8421 | |
8422 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8423 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8424 | if (subfacet->used) { | |
8425 | ds_put_format(&ds, "%.3fs", | |
8426 | (time_msec() - subfacet->used) / 1000.0); | |
8427 | } else { | |
8428 | ds_put_format(&ds, "never"); | |
8429 | } | |
8430 | if (subfacet->facet->tcp_flags) { | |
8431 | ds_put_cstr(&ds, ", flags:"); | |
8432 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8433 | } | |
8434 | ||
8435 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 8436 | if (facet->xout.slow) { |
f2245da3 JP |
8437 | uint64_t slow_path_stub[128 / 8]; |
8438 | const struct nlattr *actions; | |
8439 | size_t actions_len; | |
8440 | ||
bbafd73b | 8441 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
8442 | slow_path_stub, sizeof slow_path_stub, |
8443 | &actions, &actions_len); | |
8444 | format_odp_actions(&ds, actions, actions_len); | |
8445 | } else { | |
bbafd73b EJ |
8446 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
8447 | facet->xout.odp_actions.size); | |
f2245da3 | 8448 | } |
27022416 JP |
8449 | ds_put_char(&ds, '\n'); |
8450 | } | |
8451 | ||
8452 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8453 | ds_destroy(&ds); | |
8454 | } | |
8455 | ||
8456 | static void | |
8457 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8458 | int argc OVS_UNUSED, const char *argv[], | |
8459 | void *aux OVS_UNUSED) | |
8460 | { | |
8461 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8462 | struct ofproto_dpif *ofproto; | |
8463 | ||
8464 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8465 | if (!ofproto) { | |
8466 | unixctl_command_reply_error(conn, "no such bridge"); | |
8467 | return; | |
8468 | } | |
8469 | ||
8470 | flush(&ofproto->up); | |
8471 | ||
8472 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8473 | ds_destroy(&ds); | |
8474 | } | |
8475 | ||
abe529af BP |
8476 | static void |
8477 | ofproto_dpif_unixctl_init(void) | |
8478 | { | |
8479 | static bool registered; | |
8480 | if (registered) { | |
8481 | return; | |
8482 | } | |
8483 | registered = true; | |
8484 | ||
0e15264f BP |
8485 | unixctl_command_register( |
8486 | "ofproto/trace", | |
50aa28fd AW |
8487 | "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", |
8488 | 1, 3, ofproto_unixctl_trace, NULL); | |
96e466a3 | 8489 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
8490 | ofproto_unixctl_fdb_flush, NULL); |
8491 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8492 | ofproto_unixctl_fdb_show, NULL); | |
8493 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8494 | ofproto_dpif_clog, NULL); | |
8495 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8496 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
8497 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
8498 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
8499 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
8500 | ofproto_unixctl_dpif_dump_dps, NULL); | |
dc54ef36 EJ |
8501 | unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show, |
8502 | NULL); | |
27022416 JP |
8503 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, |
8504 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8505 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8506 | ofproto_unixctl_dpif_del_flows, NULL); | |
abe529af BP |
8507 | } |
8508 | \f | |
52a90c29 BP |
8509 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
8510 | * | |
8511 | * This is deprecated. It is only for compatibility with broken device drivers | |
8512 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8513 | * devices are not used. When broken device drivers are no longer in | |
8514 | * widespread use, we will delete these interfaces. */ | |
8515 | ||
8516 | static int | |
8517 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8518 | { | |
8519 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8520 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8521 | ||
8522 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8523 | && vid == ofport->vlandev_vid) { | |
8524 | return 0; | |
8525 | } | |
8526 | ||
2cc3c58e | 8527 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
8528 | |
8529 | if (ofport->realdev_ofp_port) { | |
8530 | vsp_remove(ofport); | |
8531 | } | |
8532 | if (realdev_ofp_port && ofport->bundle) { | |
8533 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8534 | * themselves be part of a bundle. */ | |
8535 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8536 | } | |
8537 | ||
8538 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8539 | ofport->vlandev_vid = vid; | |
8540 | ||
8541 | if (realdev_ofp_port) { | |
8542 | vsp_add(ofport, realdev_ofp_port, vid); | |
8543 | } | |
8544 | ||
8545 | return 0; | |
8546 | } | |
8547 | ||
8548 | static uint32_t | |
8549 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8550 | { | |
8551 | return hash_2words(realdev_ofp_port, vid); | |
8552 | } | |
8553 | ||
deea1200 AW |
8554 | /* Returns the OFP port number of the Linux VLAN device that corresponds to |
8555 | * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in | |
8556 | * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and | |
8557 | * 'vlan_tci' 9, it would return the port number of eth0.9. | |
40e05935 | 8558 | * |
deea1200 AW |
8559 | * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this |
8560 | * function just returns its 'realdev_ofp_port' argument. */ | |
8561 | static uint16_t | |
52a90c29 | 8562 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, |
deea1200 | 8563 | uint16_t realdev_ofp_port, ovs_be16 vlan_tci) |
52a90c29 BP |
8564 | { |
8565 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
52a90c29 BP |
8566 | int vid = vlan_tci_to_vid(vlan_tci); |
8567 | const struct vlan_splinter *vsp; | |
8568 | ||
8569 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
8570 | hash_realdev_vid(realdev_ofp_port, vid), | |
8571 | &ofproto->realdev_vid_map) { | |
8572 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8573 | && vsp->vid == vid) { | |
deea1200 | 8574 | return vsp->vlandev_ofp_port; |
52a90c29 BP |
8575 | } |
8576 | } | |
8577 | } | |
deea1200 | 8578 | return realdev_ofp_port; |
52a90c29 BP |
8579 | } |
8580 | ||
8581 | static struct vlan_splinter * | |
8582 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8583 | { | |
8584 | struct vlan_splinter *vsp; | |
8585 | ||
8586 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8587 | &ofproto->vlandev_map) { | |
8588 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8589 | return vsp; | |
8590 | } | |
8591 | } | |
8592 | ||
8593 | return NULL; | |
8594 | } | |
8595 | ||
40e05935 BP |
8596 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
8597 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8598 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8599 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8600 | * eth0 and store 9 in '*vid'. | |
8601 | * | |
8602 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8603 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8604 | * always does.*/ | |
52a90c29 BP |
8605 | static uint16_t |
8606 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 8607 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
8608 | { |
8609 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8610 | const struct vlan_splinter *vsp; | |
8611 | ||
8612 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8613 | if (vsp) { | |
8614 | if (vid) { | |
8615 | *vid = vsp->vid; | |
8616 | } | |
8617 | return vsp->realdev_ofp_port; | |
8618 | } | |
8619 | } | |
8620 | return 0; | |
8621 | } | |
8622 | ||
b98d8985 BP |
8623 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
8624 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8625 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8626 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8627 | * always the case unless VLAN splinters are enabled), returns false without | |
8628 | * making any changes. */ | |
8629 | static bool | |
8630 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8631 | { | |
8632 | uint16_t realdev; | |
8633 | int vid; | |
8634 | ||
8635 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8636 | if (!realdev) { | |
8637 | return false; | |
8638 | } | |
8639 | ||
8640 | /* Cause the flow to be processed as if it came in on the real device with | |
8641 | * the VLAN device's VLAN ID. */ | |
8642 | flow->in_port = realdev; | |
8643 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8644 | return true; | |
8645 | } | |
8646 | ||
52a90c29 BP |
8647 | static void |
8648 | vsp_remove(struct ofport_dpif *port) | |
8649 | { | |
8650 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8651 | struct vlan_splinter *vsp; | |
8652 | ||
8653 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8654 | if (vsp) { | |
8655 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8656 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8657 | free(vsp); | |
8658 | ||
8659 | port->realdev_ofp_port = 0; | |
8660 | } else { | |
8661 | VLOG_ERR("missing vlan device record"); | |
8662 | } | |
8663 | } | |
8664 | ||
8665 | static void | |
8666 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8667 | { | |
8668 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8669 | ||
8670 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8671 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8672 | == realdev_ofp_port)) { | |
8673 | struct vlan_splinter *vsp; | |
8674 | ||
8675 | vsp = xmalloc(sizeof *vsp); | |
8676 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8677 | hash_int(port->up.ofp_port, 0)); | |
8678 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8679 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8680 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8681 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8682 | vsp->vid = vid; | |
8683 | ||
8684 | port->realdev_ofp_port = realdev_ofp_port; | |
8685 | } else { | |
8686 | VLOG_ERR("duplicate vlan device record"); | |
8687 | } | |
8688 | } | |
e1b1d06a JP |
8689 | |
8690 | static uint32_t | |
8691 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8692 | { | |
8693 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8694 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8695 | } | |
8696 | ||
acf60855 JP |
8697 | static struct ofport_dpif * |
8698 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
e1b1d06a JP |
8699 | { |
8700 | struct ofport_dpif *port; | |
8701 | ||
8702 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8703 | hash_int(odp_port, 0), | |
acf60855 | 8704 | &backer->odp_to_ofport_map) { |
e1b1d06a | 8705 | if (port->odp_port == odp_port) { |
acf60855 | 8706 | return port; |
e1b1d06a JP |
8707 | } |
8708 | } | |
8709 | ||
acf60855 JP |
8710 | return NULL; |
8711 | } | |
8712 | ||
8713 | static uint16_t | |
8714 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8715 | { | |
8716 | struct ofport_dpif *port; | |
8717 | ||
8718 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 8719 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
8720 | return port->up.ofp_port; |
8721 | } else { | |
8722 | return OFPP_NONE; | |
8723 | } | |
e1b1d06a | 8724 | } |
655ab909 | 8725 | |
655ab909 AZ |
8726 | /* Compute exponentially weighted moving average, adding 'new' as the newest, |
8727 | * most heavily weighted element. 'base' designates the rate of decay: after | |
8728 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
8729 | * (about .37). */ | |
8730 | static void | |
8731 | exp_mavg(double *avg, int base, double new) | |
8732 | { | |
8733 | *avg = (*avg * (base - 1) + new) / base; | |
8734 | } | |
8735 | ||
8736 | static void | |
dc54ef36 | 8737 | update_moving_averages(struct dpif_backer *backer) |
655ab909 AZ |
8738 | { |
8739 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
dc54ef36 EJ |
8740 | long long int minutes = (time_msec() - backer->created) / min_ms; |
8741 | ||
8742 | if (minutes > 0) { | |
8743 | backer->lifetime.add_rate = (double) backer->total_subfacet_add_count | |
8744 | / minutes; | |
8745 | backer->lifetime.del_rate = (double) backer->total_subfacet_del_count | |
8746 | / minutes; | |
8747 | } else { | |
8748 | backer->lifetime.add_rate = 0.0; | |
8749 | backer->lifetime.del_rate = 0.0; | |
8750 | } | |
655ab909 AZ |
8751 | |
8752 | /* Update hourly averages on the minute boundaries. */ | |
dc54ef36 EJ |
8753 | if (time_msec() - backer->last_minute >= min_ms) { |
8754 | exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count); | |
8755 | exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count); | |
655ab909 AZ |
8756 | |
8757 | /* Update daily averages on the hour boundaries. */ | |
dc54ef36 EJ |
8758 | if ((backer->last_minute - backer->created) / min_ms % 60 == 59) { |
8759 | exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate); | |
8760 | exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate); | |
655ab909 AZ |
8761 | } |
8762 | ||
dc54ef36 EJ |
8763 | backer->total_subfacet_add_count += backer->subfacet_add_count; |
8764 | backer->total_subfacet_del_count += backer->subfacet_del_count; | |
8765 | backer->subfacet_add_count = 0; | |
8766 | backer->subfacet_del_count = 0; | |
8767 | backer->last_minute += min_ms; | |
655ab909 AZ |
8768 | } |
8769 | } | |
e1b1d06a | 8770 | |
abe529af | 8771 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 8772 | init, |
abe529af BP |
8773 | enumerate_types, |
8774 | enumerate_names, | |
8775 | del, | |
0aeaabc8 | 8776 | port_open_type, |
acf60855 JP |
8777 | type_run, |
8778 | type_run_fast, | |
8779 | type_wait, | |
abe529af BP |
8780 | alloc, |
8781 | construct, | |
8782 | destruct, | |
8783 | dealloc, | |
8784 | run, | |
5fcc0d00 | 8785 | run_fast, |
abe529af | 8786 | wait, |
0d085684 | 8787 | get_memory_usage, |
abe529af | 8788 | flush, |
6c1491fb BP |
8789 | get_features, |
8790 | get_tables, | |
abe529af BP |
8791 | port_alloc, |
8792 | port_construct, | |
8793 | port_destruct, | |
8794 | port_dealloc, | |
8795 | port_modified, | |
8796 | port_reconfigured, | |
8797 | port_query_by_name, | |
8798 | port_add, | |
8799 | port_del, | |
6527c598 | 8800 | port_get_stats, |
abe529af BP |
8801 | port_dump_start, |
8802 | port_dump_next, | |
8803 | port_dump_done, | |
8804 | port_poll, | |
8805 | port_poll_wait, | |
8806 | port_is_lacp_current, | |
0ab6decf | 8807 | NULL, /* rule_choose_table */ |
abe529af BP |
8808 | rule_alloc, |
8809 | rule_construct, | |
8810 | rule_destruct, | |
8811 | rule_dealloc, | |
abe529af BP |
8812 | rule_get_stats, |
8813 | rule_execute, | |
8814 | rule_modify_actions, | |
7257b535 | 8815 | set_frag_handling, |
abe529af BP |
8816 | packet_out, |
8817 | set_netflow, | |
8818 | get_netflow_ids, | |
8819 | set_sflow, | |
29089a54 | 8820 | set_ipfix, |
abe529af | 8821 | set_cfm, |
9a9e3786 | 8822 | get_cfm_status, |
ccc09689 EJ |
8823 | set_bfd, |
8824 | get_bfd_status, | |
21f7563c JP |
8825 | set_stp, |
8826 | get_stp_status, | |
8827 | set_stp_port, | |
8828 | get_stp_port_status, | |
8b36f51e | 8829 | set_queues, |
abe529af BP |
8830 | bundle_set, |
8831 | bundle_remove, | |
8832 | mirror_set, | |
9d24de3b | 8833 | mirror_get_stats, |
abe529af BP |
8834 | set_flood_vlans, |
8835 | is_mirror_output_bundle, | |
8402c74b | 8836 | forward_bpdu_changed, |
c4069512 | 8837 | set_mac_table_config, |
52a90c29 | 8838 | set_realdev, |
abe529af | 8839 | }; |