]>
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" | |
44 | #include "ofp-util.h" | |
45 | #include "ofpbuf.h" | |
f25d0cf3 | 46 | #include "ofp-actions.h" |
31a19d69 | 47 | #include "ofp-parse.h" |
abe529af | 48 | #include "ofp-print.h" |
9d6ac44e | 49 | #include "ofproto-dpif-governor.h" |
29089a54 | 50 | #include "ofproto-dpif-ipfix.h" |
bae473fe | 51 | #include "ofproto-dpif-sflow.h" |
abe529af | 52 | #include "poll-loop.h" |
0d085684 | 53 | #include "simap.h" |
27022416 | 54 | #include "smap.h" |
abe529af | 55 | #include "timer.h" |
b9ad7294 | 56 | #include "tunnel.h" |
6c1491fb | 57 | #include "unaligned.h" |
abe529af BP |
58 | #include "unixctl.h" |
59 | #include "vlan-bitmap.h" | |
60 | #include "vlog.h" | |
61 | ||
62 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
63 | ||
abe529af | 64 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
65 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
66 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
67 | COVERAGE_DEFINE(facet_revalidate); |
68 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 69 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 70 | |
29901626 | 71 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 72 | * flow translation. */ |
1642690c | 73 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 74 | |
9cdaaebe BP |
75 | /* Number of implemented OpenFlow tables. */ |
76 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
77 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
78 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 79 | |
abe529af BP |
80 | struct ofport_dpif; |
81 | struct ofproto_dpif; | |
a088a1ff | 82 | struct flow_miss; |
ac35f9c8 | 83 | struct facet; |
abe529af BP |
84 | |
85 | struct rule_dpif { | |
86 | struct rule up; | |
87 | ||
abe529af BP |
88 | /* These statistics: |
89 | * | |
90 | * - Do include packets and bytes from facets that have been deleted or | |
91 | * whose own statistics have been folded into the rule. | |
92 | * | |
93 | * - Do include packets and bytes sent "by hand" that were accounted to | |
94 | * the rule without any facet being involved (this is a rare corner | |
95 | * case in rule_execute()). | |
96 | * | |
97 | * - Do not include packet or bytes that can be obtained from any facet's | |
98 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 99 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
100 | */ |
101 | uint64_t packet_count; /* Number of packets received. */ | |
102 | uint64_t byte_count; /* Number of bytes received. */ | |
103 | ||
54a9cbc9 BP |
104 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
105 | ||
abe529af BP |
106 | struct list facets; /* List of "struct facet"s. */ |
107 | }; | |
108 | ||
109 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
110 | { | |
111 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
112 | } | |
113 | ||
29901626 | 114 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
115 | const struct flow *); |
116 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
117 | const struct flow *, | |
118 | uint8_t table); | |
c376f9a3 IY |
119 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
120 | const struct flow *flow); | |
abe529af | 121 | |
112bc5f4 BP |
122 | static void rule_credit_stats(struct rule_dpif *, |
123 | const struct dpif_flow_stats *); | |
ac35f9c8 | 124 | static void flow_push_stats(struct facet *, const struct dpif_flow_stats *); |
822d9414 | 125 | static tag_type rule_calculate_tag(const struct flow *, |
5cb7a798 | 126 | const struct minimask *, uint32_t basis); |
b0f7b9b5 BP |
127 | static void rule_invalidate(const struct rule_dpif *); |
128 | ||
abe529af BP |
129 | #define MAX_MIRRORS 32 |
130 | typedef uint32_t mirror_mask_t; | |
131 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
132 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
133 | struct ofmirror { | |
134 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
135 | size_t idx; /* In ofproto's "mirrors" array. */ | |
136 | void *aux; /* Key supplied by ofproto's client. */ | |
137 | char *name; /* Identifier for log messages. */ | |
138 | ||
139 | /* Selection criteria. */ | |
140 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
141 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
142 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
143 | ||
9ba15e2a | 144 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
145 | struct ofbundle *out; /* Output port or NULL. */ |
146 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 147 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
148 | |
149 | /* Counters. */ | |
150 | int64_t packet_count; /* Number of packets sent. */ | |
151 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
152 | }; |
153 | ||
154 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
155 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
156 | mirror_mask_t mirrors, | |
157 | uint64_t packets, uint64_t bytes); | |
abe529af | 158 | |
abe529af | 159 | struct ofbundle { |
abe529af | 160 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 161 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
162 | void *aux; /* Key supplied by ofproto's client. */ |
163 | char *name; /* Identifier for log messages. */ | |
164 | ||
165 | /* Configuration. */ | |
166 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 167 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
168 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
169 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
170 | * NULL if all VLANs are trunked. */ | |
171 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
172 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 173 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
174 | |
175 | /* Status. */ | |
9e1fd49b | 176 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
177 | |
178 | /* Port mirroring info. */ | |
179 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
180 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
181 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
182 | }; | |
183 | ||
184 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 185 | static void bundle_update(struct ofbundle *); |
abe529af BP |
186 | static void bundle_destroy(struct ofbundle *); |
187 | static void bundle_del_port(struct ofport_dpif *); | |
188 | static void bundle_run(struct ofbundle *); | |
189 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 190 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
191 | uint16_t in_port, bool warn, |
192 | struct ofport_dpif **in_ofportp); | |
abe529af | 193 | |
33158a18 JP |
194 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
195 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
196 | * when an input bundle is needed for validation (e.g., mirroring or | |
197 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
198 | * any 'port' structs, so care must be taken when dealing with it. */ | |
199 | static struct ofbundle ofpp_none_bundle = { | |
200 | .name = "OFPP_NONE", | |
201 | .vlan_mode = PORT_VLAN_TRUNK | |
202 | }; | |
203 | ||
21f7563c JP |
204 | static void stp_run(struct ofproto_dpif *ofproto); |
205 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
206 | static int set_stp_port(struct ofport *, |
207 | const struct ofproto_port_stp_settings *); | |
21f7563c | 208 | |
5da5ec37 BP |
209 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
210 | ||
abe529af BP |
211 | struct action_xlate_ctx { |
212 | /* action_xlate_ctx_init() initializes these members. */ | |
213 | ||
214 | /* The ofproto. */ | |
215 | struct ofproto_dpif *ofproto; | |
216 | ||
217 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
218 | * this flow when actions change header fields. */ | |
219 | struct flow flow; | |
220 | ||
0ad90c84 JR |
221 | /* Flow at the last commit. */ |
222 | struct flow base_flow; | |
223 | ||
224 | /* Tunnel IP destination address as received. This is stored separately | |
225 | * as the base_flow.tunnel is cleared on init to reflect the datapath | |
226 | * behavior. Used to make sure not to send tunneled output to ourselves, | |
227 | * which might lead to an infinite loop. This could happen easily | |
228 | * if a tunnel is marked as 'ip_remote=flow', and the flow does not | |
229 | * actually set the tun_dst field. */ | |
230 | ovs_be32 orig_tunnel_ip_dst; | |
231 | ||
bd85dac1 AZ |
232 | /* stack for the push and pop actions. |
233 | * Each stack element is of the type "union mf_subvalue". */ | |
234 | struct ofpbuf stack; | |
235 | union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; | |
236 | ||
abe529af BP |
237 | /* The packet corresponding to 'flow', or a null pointer if we are |
238 | * revalidating without a packet to refer to. */ | |
239 | const struct ofpbuf *packet; | |
240 | ||
3de9590b BP |
241 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
242 | * actions update the flow table? | |
243 | * | |
244 | * We want to update these tables if we are actually processing a packet, | |
245 | * or if we are accounting for packets that the datapath has processed, but | |
246 | * not if we are just revalidating. */ | |
247 | bool may_learn; | |
75a75043 | 248 | |
18b2a258 BP |
249 | /* The rule that we are currently translating, or NULL. */ |
250 | struct rule_dpif *rule; | |
54834960 | 251 | |
0e553d9c BP |
252 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
253 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
254 | * timeouts.) */ | |
255 | uint8_t tcp_flags; | |
256 | ||
112bc5f4 BP |
257 | /* If nonnull, flow translation calls this function just before executing a |
258 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
259 | * when the recursion depth is exceeded. | |
260 | * | |
261 | * 'rule' is the rule being submitted into. It will be null if the | |
262 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
263 | * | |
264 | * This is normally null so the client has to set it manually after | |
265 | * calling action_xlate_ctx_init(). */ | |
266 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
267 | ||
479df176 BP |
268 | /* If nonnull, flow translation calls this function to report some |
269 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
270 | * dropped a packet. */ | |
271 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
272 | ||
112bc5f4 BP |
273 | /* If nonnull, flow translation credits the specified statistics to each |
274 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
275 | * |
276 | * This is normally null so the client has to set it manually after | |
277 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 278 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 279 | |
abe529af BP |
280 | /* xlate_actions() initializes and uses these members. The client might want |
281 | * to look at them after it returns. */ | |
282 | ||
283 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 284 | tag_type tags; /* Tags associated with actions. */ |
6a7e895f | 285 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
75a75043 BP |
286 | bool has_learn; /* Actions include NXAST_LEARN? */ |
287 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 288 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 289 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 290 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
291 | |
292 | /* xlate_actions() initializes and uses these members, but the client has no | |
293 | * reason to look at them. */ | |
294 | ||
295 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 296 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
deedf7e7 | 297 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 298 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 | 299 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
9b56fe13 | 300 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ |
6ff686f2 | 301 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ |
848e8809 | 302 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
303 | }; |
304 | ||
14f94f9a JP |
305 | /* Initial values of fields of the packet that may be changed during |
306 | * flow processing and needed later. */ | |
307 | struct initial_vals { | |
308 | /* This is the value of vlan_tci in the packet as actually received from | |
309 | * dpif. This is the same as the facet's flow.vlan_tci unless the packet | |
310 | * was received via a VLAN splinter. In that case, this value is 0 | |
311 | * (because the packet as actually received from the dpif had no 802.1Q | |
312 | * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter | |
313 | * represents. | |
314 | * | |
315 | * This member should be removed when the VLAN splinters feature is no | |
316 | * longer needed. */ | |
317 | ovs_be16 vlan_tci; | |
318 | }; | |
319 | ||
abe529af BP |
320 | static void action_xlate_ctx_init(struct action_xlate_ctx *, |
321 | struct ofproto_dpif *, const struct flow *, | |
14f94f9a JP |
322 | const struct initial_vals *initial_vals, |
323 | struct rule_dpif *, | |
0e553d9c | 324 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 | 325 | static void xlate_actions(struct action_xlate_ctx *, |
f25d0cf3 | 326 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 BP |
327 | struct ofpbuf *odp_actions); |
328 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
f25d0cf3 BP |
329 | const struct ofpact *ofpacts, |
330 | size_t ofpacts_len); | |
0a740f48 EJ |
331 | static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port, |
332 | uint8_t table_id, bool may_packet_in); | |
abe529af | 333 | |
6a7e895f BP |
334 | static size_t put_userspace_action(const struct ofproto_dpif *, |
335 | struct ofpbuf *odp_actions, | |
336 | const struct flow *, | |
29089a54 RL |
337 | const union user_action_cookie *, |
338 | const size_t); | |
6a7e895f BP |
339 | |
340 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
341 | enum slow_path_reason, | |
342 | uint64_t *stub, size_t stub_size, | |
343 | const struct nlattr **actionsp, | |
344 | size_t *actions_lenp); | |
345 | ||
479df176 BP |
346 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); |
347 | ||
6a7e895f BP |
348 | /* A subfacet (see "struct subfacet" below) has three possible installation |
349 | * states: | |
350 | * | |
351 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
352 | * case just after the subfacet is created, just before the subfacet is | |
353 | * destroyed, or if the datapath returns an error when we try to install a | |
354 | * subfacet. | |
355 | * | |
356 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
357 | * | |
358 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
359 | * ofproto_dpif is installed in the datapath. | |
360 | */ | |
361 | enum subfacet_path { | |
362 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
363 | SF_FAST_PATH, /* Full actions are installed. */ | |
364 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
365 | }; | |
366 | ||
367 | static const char *subfacet_path_to_string(enum subfacet_path); | |
368 | ||
5f5fbd17 BP |
369 | /* A dpif flow and actions associated with a facet. |
370 | * | |
371 | * See also the large comment on struct facet. */ | |
372 | struct subfacet { | |
373 | /* Owners. */ | |
374 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
375 | struct list list_node; /* In struct facet's 'facets' list. */ | |
376 | struct facet *facet; /* Owning facet. */ | |
377 | ||
5f5fbd17 BP |
378 | enum odp_key_fitness key_fitness; |
379 | struct nlattr *key; | |
380 | int key_len; | |
381 | ||
382 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 383 | long long int created; /* Time created. */ |
5f5fbd17 BP |
384 | |
385 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
386 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
387 | ||
388 | /* Datapath actions. | |
389 | * | |
390 | * These should be essentially identical for every subfacet in a facet, but | |
391 | * may differ in trivial ways due to VLAN splinters. */ | |
392 | size_t actions_len; /* Number of bytes in actions[]. */ | |
393 | struct nlattr *actions; /* Datapath actions. */ | |
394 | ||
6a7e895f BP |
395 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
396 | enum subfacet_path path; /* Installed in datapath? */ | |
5f5fbd17 | 397 | |
14f94f9a JP |
398 | /* Initial values of the packet that may be needed later. */ |
399 | struct initial_vals initial_vals; | |
a088a1ff JP |
400 | |
401 | /* Datapath port the packet arrived on. This is needed to remove | |
402 | * flows for ports that are no longer part of the bridge. Since the | |
403 | * flow definition only has the OpenFlow port number and the port is | |
404 | * no longer part of the bridge, we can't determine the datapath port | |
405 | * number needed to delete the flow from the datapath. */ | |
406 | uint32_t odp_in_port; | |
5f5fbd17 BP |
407 | }; |
408 | ||
1d85f9e5 JP |
409 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
410 | ||
a088a1ff | 411 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 412 | long long int now); |
5f5fbd17 | 413 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
acf60855 | 414 | const struct nlattr *key, size_t key_len, |
9566abf9 | 415 | uint32_t key_hash); |
5f5fbd17 BP |
416 | static void subfacet_destroy(struct subfacet *); |
417 | static void subfacet_destroy__(struct subfacet *); | |
1d85f9e5 JP |
418 | static void subfacet_destroy_batch(struct ofproto_dpif *, |
419 | struct subfacet **, int n); | |
5f5fbd17 BP |
420 | static void subfacet_reset_dp_stats(struct subfacet *, |
421 | struct dpif_flow_stats *); | |
422 | static void subfacet_update_time(struct subfacet *, long long int used); | |
423 | static void subfacet_update_stats(struct subfacet *, | |
424 | const struct dpif_flow_stats *); | |
425 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
426 | const struct ofpbuf *packet, |
427 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
428 | static int subfacet_install(struct subfacet *, |
429 | const struct nlattr *actions, size_t actions_len, | |
6a7e895f | 430 | struct dpif_flow_stats *, enum slow_path_reason); |
5f5fbd17 BP |
431 | static void subfacet_uninstall(struct subfacet *); |
432 | ||
6a7e895f BP |
433 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); |
434 | ||
b0f7b9b5 BP |
435 | /* An exact-match instantiation of an OpenFlow flow. |
436 | * | |
437 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
438 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
439 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
440 | * the facet. When the kernel module (or other dpif implementation) and Open | |
441 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
442 | * one subfacet per facet. If the dpif implementation supports more-specific | |
443 | * flow matching than userspace, however, a facet can have more than one | |
444 | * subfacet, each of which corresponds to some distinction in flow that | |
445 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
446 | * |
447 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
448 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 449 | struct facet { |
b0f7b9b5 BP |
450 | /* Owners. */ |
451 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
452 | struct list list_node; /* In owning rule's 'facets' list. */ | |
453 | struct rule_dpif *rule; /* Owning rule. */ | |
454 | ||
455 | /* Owned data. */ | |
456 | struct list subfacets; | |
abe529af BP |
457 | long long int used; /* Time last used; time created if not used. */ |
458 | ||
b0f7b9b5 BP |
459 | /* Key. */ |
460 | struct flow flow; | |
461 | ||
abe529af BP |
462 | /* These statistics: |
463 | * | |
464 | * - Do include packets and bytes sent "by hand", e.g. with | |
465 | * dpif_execute(). | |
466 | * | |
467 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 468 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 469 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
470 | * |
471 | * - Do not include packets or bytes that can be obtained from the | |
472 | * datapath for any existing subfacet. | |
abe529af BP |
473 | */ |
474 | uint64_t packet_count; /* Number of packets received. */ | |
475 | uint64_t byte_count; /* Number of bytes received. */ | |
476 | ||
b0f7b9b5 | 477 | /* Resubmit statistics. */ |
9d24de3b JP |
478 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
479 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
480 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 481 | |
b0f7b9b5 | 482 | /* Accounting. */ |
907a4c5e | 483 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 484 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 485 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 486 | |
b95fc6ba BP |
487 | /* Properties of datapath actions. |
488 | * | |
489 | * Every subfacet has its own actions because actions can differ slightly | |
490 | * between splintered and non-splintered subfacets due to the VLAN tag | |
491 | * being initially different (present vs. absent). All of them have these | |
492 | * properties in common so we just store one copy of them here. */ | |
75a75043 BP |
493 | bool has_learn; /* Actions include NXAST_LEARN? */ |
494 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 495 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 496 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 497 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
498 | |
499 | /* Storage for a single subfacet, to reduce malloc() time and space | |
500 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
501 | * case has exactly one subfacet. However, 'one_subfacet' may not |
502 | * always be valid, since it could have been removed after newer | |
503 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 504 | struct subfacet one_subfacet; |
6cf474d7 EJ |
505 | |
506 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
507 | }; |
508 | ||
2b459b83 BP |
509 | static struct facet *facet_create(struct rule_dpif *, |
510 | const struct flow *, uint32_t hash); | |
15baa734 | 511 | static void facet_remove(struct facet *); |
abe529af BP |
512 | static void facet_free(struct facet *); |
513 | ||
2b459b83 BP |
514 | static struct facet *facet_find(struct ofproto_dpif *, |
515 | const struct flow *, uint32_t hash); | |
abe529af | 516 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 517 | const struct flow *, uint32_t hash); |
c57b2226 | 518 | static void facet_revalidate(struct facet *); |
6814e51f | 519 | static bool facet_check_consistency(struct facet *); |
abe529af | 520 | |
15baa734 | 521 | static void facet_flush_stats(struct facet *); |
abe529af | 522 | |
15baa734 | 523 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 524 | static void facet_reset_counters(struct facet *); |
abe529af | 525 | static void facet_push_stats(struct facet *); |
3de9590b BP |
526 | static void facet_learn(struct facet *); |
527 | static void facet_account(struct facet *); | |
8844e035 | 528 | static void push_all_stats(void); |
abe529af | 529 | |
1a1e3a0a JP |
530 | static struct subfacet *facet_get_subfacet(struct facet *); |
531 | ||
abe529af BP |
532 | static bool facet_is_controller_flow(struct facet *); |
533 | ||
abe529af | 534 | struct ofport_dpif { |
acf60855 | 535 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ |
abe529af BP |
536 | struct ofport up; |
537 | ||
538 | uint32_t odp_port; | |
539 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
540 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
541 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
ccc09689 | 542 | struct bfd *bfd; /* BFD, if any. */ |
abe529af | 543 | tag_type tag; /* Tag associated with this port. */ |
015e08bc | 544 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 545 | long long int carrier_seq; /* Carrier status changes. */ |
b9ad7294 | 546 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ |
21f7563c | 547 | |
52a90c29 | 548 | /* Spanning tree. */ |
21f7563c JP |
549 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
550 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
551 | long long int stp_state_entered; | |
8b36f51e EJ |
552 | |
553 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
554 | |
555 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
556 | * | |
557 | * This is deprecated. It is only for compatibility with broken device | |
558 | * drivers in old versions of Linux that do not properly support VLANs when | |
559 | * VLAN devices are not used. When broken device drivers are no longer in | |
560 | * widespread use, we will delete these interfaces. */ | |
561 | uint16_t realdev_ofp_port; | |
562 | int vlandev_vid; | |
8b36f51e EJ |
563 | }; |
564 | ||
565 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
566 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
567 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
568 | struct priority_to_dscp { | |
569 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
570 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
571 | ||
572 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
573 | }; |
574 | ||
52a90c29 BP |
575 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
576 | * | |
577 | * This is deprecated. It is only for compatibility with broken device drivers | |
578 | * in old versions of Linux that do not properly support VLANs when VLAN | |
579 | * devices are not used. When broken device drivers are no longer in | |
580 | * widespread use, we will delete these interfaces. */ | |
581 | struct vlan_splinter { | |
582 | struct hmap_node realdev_vid_node; | |
583 | struct hmap_node vlandev_node; | |
584 | uint16_t realdev_ofp_port; | |
585 | uint16_t vlandev_ofp_port; | |
586 | int vid; | |
587 | }; | |
588 | ||
589 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
590 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 591 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
592 | static void vsp_remove(struct ofport_dpif *); |
593 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
594 | ||
e1b1d06a JP |
595 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
596 | uint16_t ofp_port); | |
597 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
598 | uint32_t odp_port); | |
599 | ||
abe529af BP |
600 | static struct ofport_dpif * |
601 | ofport_dpif_cast(const struct ofport *ofport) | |
602 | { | |
abe529af BP |
603 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
604 | } | |
605 | ||
606 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 607 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 608 | static void port_wait(struct ofport_dpif *); |
a5610457 | 609 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 610 | static void ofport_clear_priorities(struct ofport_dpif *); |
8fa4d1d0 | 611 | static void run_fast_rl(void); |
abe529af | 612 | |
7ee20df1 BP |
613 | struct dpif_completion { |
614 | struct list list_node; | |
615 | struct ofoperation *op; | |
616 | }; | |
617 | ||
54a9cbc9 BP |
618 | /* Extra information about a classifier table. |
619 | * Currently used just for optimized flow revalidation. */ | |
620 | struct table_dpif { | |
621 | /* If either of these is nonnull, then this table has a form that allows | |
622 | * flows to be tagged to avoid revalidating most flows for the most common | |
623 | * kinds of flow table changes. */ | |
624 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
625 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
626 | uint32_t basis; /* Keeps each table's tags separate. */ | |
627 | }; | |
628 | ||
3c4a309c BP |
629 | /* Reasons that we might need to revalidate every facet, and corresponding |
630 | * coverage counters. | |
631 | * | |
632 | * A value of 0 means that there is no need to revalidate. | |
633 | * | |
634 | * It would be nice to have some cleaner way to integrate with coverage | |
635 | * counters, but with only a few reasons I guess this is good enough for | |
636 | * now. */ | |
637 | enum revalidate_reason { | |
638 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
639 | REV_STP, /* Spanning tree protocol port status change. */ | |
640 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
641 | REV_FLOW_TABLE, /* Flow table changed. */ | |
642 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
643 | }; | |
644 | COVERAGE_DEFINE(rev_reconfigure); | |
645 | COVERAGE_DEFINE(rev_stp); | |
646 | COVERAGE_DEFINE(rev_port_toggled); | |
647 | COVERAGE_DEFINE(rev_flow_table); | |
648 | COVERAGE_DEFINE(rev_inconsistency); | |
649 | ||
8f73d537 EJ |
650 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
651 | * These are datapath flows which have no associated ofproto, if they did we | |
652 | * would use facets. */ | |
653 | struct drop_key { | |
654 | struct hmap_node hmap_node; | |
655 | struct nlattr *key; | |
656 | size_t key_len; | |
657 | }; | |
658 | ||
acf60855 JP |
659 | /* All datapaths of a given type share a single dpif backer instance. */ |
660 | struct dpif_backer { | |
661 | char *type; | |
662 | int refcount; | |
663 | struct dpif *dpif; | |
664 | struct timer next_expiration; | |
665 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
2cc3c58e | 666 | |
7d82ab2e | 667 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ |
b9ad7294 | 668 | |
2cc3c58e EJ |
669 | /* Facet revalidation flags applying to facets which use this backer. */ |
670 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
671 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
8f73d537 EJ |
672 | |
673 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
acf60855 JP |
674 | }; |
675 | ||
676 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
677 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
678 | ||
8f73d537 | 679 | static void drop_key_clear(struct dpif_backer *); |
acf60855 JP |
680 | static struct ofport_dpif * |
681 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
682 | ||
735d7efb AZ |
683 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, |
684 | uint64_t delta); | |
655ab909 AZ |
685 | struct avg_subfacet_rates { |
686 | double add_rate; /* Moving average of new flows created per minute. */ | |
687 | double del_rate; /* Moving average of flows deleted per minute. */ | |
688 | }; | |
689 | static void show_dp_rates(struct ds *ds, const char *heading, | |
690 | const struct avg_subfacet_rates *rates); | |
691 | static void exp_mavg(double *avg, int base, double new); | |
735d7efb | 692 | |
abe529af | 693 | struct ofproto_dpif { |
b44a10b7 | 694 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af | 695 | struct ofproto up; |
acf60855 | 696 | struct dpif_backer *backer; |
abe529af | 697 | |
c57b2226 BP |
698 | /* Special OpenFlow rules. */ |
699 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
700 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
701 | ||
6c1491fb BP |
702 | /* Statistics. */ |
703 | uint64_t n_matches; | |
704 | ||
abe529af BP |
705 | /* Bridging. */ |
706 | struct netflow *netflow; | |
bae473fe | 707 | struct dpif_sflow *sflow; |
29089a54 | 708 | struct dpif_ipfix *ipfix; |
abe529af BP |
709 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
710 | struct mac_learning *ml; | |
711 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 712 | bool has_mirrors; |
abe529af BP |
713 | bool has_bonded_bundles; |
714 | ||
abe529af BP |
715 | /* Facets. */ |
716 | struct hmap facets; | |
b0f7b9b5 | 717 | struct hmap subfacets; |
9d6ac44e | 718 | struct governor *governor; |
4d2a0f39 | 719 | long long int consistency_rl; |
54a9cbc9 BP |
720 | |
721 | /* Revalidation. */ | |
722 | struct table_dpif tables[N_TABLES]; | |
7ee20df1 BP |
723 | |
724 | /* Support for debugging async flow mods. */ | |
725 | struct list completions; | |
daff3353 EJ |
726 | |
727 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
728 | struct netdev_stats stats; /* To account packets generated and consumed in |
729 | * userspace. */ | |
21f7563c JP |
730 | |
731 | /* Spanning tree. */ | |
732 | struct stp *stp; | |
733 | long long int stp_last_tick; | |
52a90c29 BP |
734 | |
735 | /* VLAN splinters. */ | |
736 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
737 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
e1b1d06a | 738 | |
acf60855 | 739 | /* Ports. */ |
0a740f48 EJ |
740 | struct sset ports; /* Set of standard port names. */ |
741 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
acf60855 JP |
742 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ |
743 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
735d7efb AZ |
744 | |
745 | /* Per ofproto's dpif stats. */ | |
746 | uint64_t n_hit; | |
747 | uint64_t n_missed; | |
655ab909 AZ |
748 | |
749 | /* Subfacet statistics. | |
750 | * | |
751 | * These keep track of the total number of subfacets added and deleted and | |
752 | * flow life span. They are useful for computing the flow rates stats | |
753 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
754 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
755 | * performance in new situations. */ | |
756 | long long int created; /* Time when it is created. */ | |
757 | unsigned int max_n_subfacet; /* Maximum number of flows */ | |
758 | ||
759 | /* The average number of subfacets... */ | |
760 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
761 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
762 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
763 | ||
764 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
765 | unsigned int subfacet_add_count; | |
766 | unsigned int subfacet_del_count; | |
767 | ||
768 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
769 | unsigned long long int total_subfacet_add_count; | |
770 | unsigned long long int total_subfacet_del_count; | |
771 | ||
772 | /* Sum of the number of milliseconds that each subfacet existed, | |
773 | * over the subfacets that have been added and then later deleted. */ | |
774 | unsigned long long int total_subfacet_life_span; | |
775 | ||
776 | /* Incremented by the number of currently existing subfacets, each | |
777 | * time we pull statistics from the kernel. */ | |
778 | unsigned long long int total_subfacet_count; | |
779 | ||
780 | /* Number of times we pull statistics from the kernel. */ | |
781 | unsigned long long int n_update_stats; | |
abe529af | 782 | }; |
655ab909 AZ |
783 | static unsigned long long int avg_subfacet_life_span( |
784 | const struct ofproto_dpif *); | |
785 | static double avg_subfacet_count(const struct ofproto_dpif *ofproto); | |
786 | static void update_moving_averages(struct ofproto_dpif *ofproto); | |
787 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, | |
788 | uint64_t delta); | |
789 | static void update_max_subfacet_count(struct ofproto_dpif *ofproto); | |
abe529af | 790 | |
7ee20df1 BP |
791 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
792 | * for debugging the asynchronous flow_mod implementation.) */ | |
793 | static bool clogged; | |
794 | ||
b44a10b7 BP |
795 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
796 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
797 | ||
abe529af BP |
798 | static void ofproto_dpif_unixctl_init(void); |
799 | ||
800 | static struct ofproto_dpif * | |
801 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
802 | { | |
cb22974d | 803 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
804 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); |
805 | } | |
806 | ||
4acbc98d | 807 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 808 | uint16_t ofp_port); |
4acbc98d | 809 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 810 | uint32_t odp_port); |
6a6455e5 | 811 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
14f94f9a JP |
812 | const struct ofpbuf *, |
813 | const struct initial_vals *, struct ds *); | |
abe529af BP |
814 | |
815 | /* Packet processing. */ | |
816 | static void update_learning_table(struct ofproto_dpif *, | |
817 | const struct flow *, int vlan, | |
818 | struct ofbundle *); | |
501f8d1f BP |
819 | /* Upcalls. */ |
820 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 821 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
822 | |
823 | /* Flow expiration. */ | |
acf60855 | 824 | static int expire(struct dpif_backer *); |
abe529af | 825 | |
6fca1ffb BP |
826 | /* NetFlow. */ |
827 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
828 | ||
abe529af | 829 | /* Utilities. */ |
52a90c29 | 830 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
831 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
832 | struct ofpbuf *odp_actions, | |
833 | const struct flow *, uint32_t odp_port); | |
29089a54 RL |
834 | static void compose_ipfix_action(const struct ofproto_dpif *, |
835 | struct ofpbuf *odp_actions, | |
836 | const struct flow *); | |
c06bba01 JP |
837 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
838 | const struct flow *flow); | |
abe529af BP |
839 | /* Global variables. */ |
840 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
841 | |
842 | /* Initial mappings of port to bridge mappings. */ | |
843 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
abe529af BP |
844 | \f |
845 | /* Factory functions. */ | |
846 | ||
b0408fca | 847 | static void |
acf60855 | 848 | init(const struct shash *iface_hints) |
b0408fca | 849 | { |
acf60855 JP |
850 | struct shash_node *node; |
851 | ||
852 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
853 | SHASH_FOR_EACH(node, iface_hints) { | |
854 | const struct iface_hint *orig_hint = node->data; | |
855 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
856 | ||
857 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
858 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
859 | new_hint->ofp_port = orig_hint->ofp_port; | |
860 | ||
861 | shash_add(&init_ofp_ports, node->name, new_hint); | |
862 | } | |
b0408fca JP |
863 | } |
864 | ||
abe529af BP |
865 | static void |
866 | enumerate_types(struct sset *types) | |
867 | { | |
868 | dp_enumerate_types(types); | |
869 | } | |
870 | ||
871 | static int | |
872 | enumerate_names(const char *type, struct sset *names) | |
873 | { | |
acf60855 JP |
874 | struct ofproto_dpif *ofproto; |
875 | ||
876 | sset_clear(names); | |
877 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
878 | if (strcmp(type, ofproto->up.type)) { | |
879 | continue; | |
880 | } | |
881 | sset_add(names, ofproto->up.name); | |
882 | } | |
883 | ||
884 | return 0; | |
abe529af BP |
885 | } |
886 | ||
887 | static int | |
888 | del(const char *type, const char *name) | |
889 | { | |
890 | struct dpif *dpif; | |
891 | int error; | |
892 | ||
893 | error = dpif_open(name, type, &dpif); | |
894 | if (!error) { | |
895 | error = dpif_delete(dpif); | |
896 | dpif_close(dpif); | |
897 | } | |
898 | return error; | |
899 | } | |
900 | \f | |
0aeaabc8 JP |
901 | static const char * |
902 | port_open_type(const char *datapath_type, const char *port_type) | |
903 | { | |
904 | return dpif_port_open_type(datapath_type, port_type); | |
905 | } | |
906 | ||
acf60855 JP |
907 | /* Type functions. */ |
908 | ||
476cb42a BP |
909 | static struct ofproto_dpif * |
910 | lookup_ofproto_dpif_by_port_name(const char *name) | |
911 | { | |
912 | struct ofproto_dpif *ofproto; | |
913 | ||
914 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
915 | if (sset_contains(&ofproto->ports, name)) { | |
916 | return ofproto; | |
917 | } | |
918 | } | |
919 | ||
920 | return NULL; | |
921 | } | |
922 | ||
acf60855 JP |
923 | static int |
924 | type_run(const char *type) | |
925 | { | |
e20b5746 | 926 | static long long int push_timer = LLONG_MIN; |
acf60855 JP |
927 | struct dpif_backer *backer; |
928 | char *devname; | |
929 | int error; | |
930 | ||
931 | backer = shash_find_data(&all_dpif_backers, type); | |
932 | if (!backer) { | |
933 | /* This is not necessarily a problem, since backers are only | |
934 | * created on demand. */ | |
935 | return 0; | |
936 | } | |
937 | ||
938 | dpif_run(backer->dpif); | |
939 | ||
e20b5746 EJ |
940 | /* The most natural place to push facet statistics is when they're pulled |
941 | * from the datapath. However, when there are many flows in the datapath, | |
942 | * this expensive operation can occur so frequently, that it reduces our | |
943 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
944 | * here instead. */ | |
945 | if (time_msec() > push_timer) { | |
946 | push_timer = time_msec() + 2000; | |
947 | push_all_stats(); | |
948 | } | |
949 | ||
2cc3c58e EJ |
950 | if (backer->need_revalidate |
951 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
952 | struct tag_set revalidate_set = backer->revalidate_set; | |
953 | bool need_revalidate = backer->need_revalidate; | |
954 | struct ofproto_dpif *ofproto; | |
a614d823 KM |
955 | struct simap_node *node; |
956 | struct simap tmp_backers; | |
957 | ||
958 | /* Handle tunnel garbage collection. */ | |
959 | simap_init(&tmp_backers); | |
960 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
961 | ||
962 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
963 | struct ofport_dpif *iter; | |
964 | ||
965 | if (backer != ofproto->backer) { | |
966 | continue; | |
967 | } | |
968 | ||
969 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
970 | const char *dp_port; | |
971 | ||
972 | if (!iter->tnl_port) { | |
973 | continue; | |
974 | } | |
975 | ||
976 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev); | |
977 | node = simap_find(&tmp_backers, dp_port); | |
978 | if (node) { | |
979 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
980 | simap_delete(&tmp_backers, node); | |
981 | node = simap_find(&backer->tnl_backers, dp_port); | |
982 | } else { | |
983 | node = simap_find(&backer->tnl_backers, dp_port); | |
984 | if (!node) { | |
985 | uint32_t odp_port = UINT32_MAX; | |
986 | ||
987 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
988 | &odp_port)) { | |
989 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
990 | node = simap_find(&backer->tnl_backers, dp_port); | |
991 | } | |
992 | } | |
993 | } | |
994 | ||
995 | iter->odp_port = node ? node->data : OVSP_NONE; | |
996 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
997 | &iter->tnl_port)) { | |
998 | backer->need_revalidate = REV_RECONFIGURE; | |
999 | } | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
1004 | dpif_port_del(backer->dpif, node->data); | |
1005 | } | |
1006 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
1007 | |
1008 | switch (backer->need_revalidate) { | |
1009 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
1010 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
1011 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
1012 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
1013 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
1014 | } | |
1015 | ||
8f73d537 EJ |
1016 | if (backer->need_revalidate) { |
1017 | /* Clear the drop_keys in case we should now be accepting some | |
1018 | * formerly dropped flows. */ | |
1019 | drop_key_clear(backer); | |
1020 | } | |
1021 | ||
f728af2e BP |
1022 | /* Clear the revalidation flags. */ |
1023 | tag_set_init(&backer->revalidate_set); | |
1024 | backer->need_revalidate = 0; | |
1025 | ||
2cc3c58e | 1026 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 1027 | struct facet *facet, *next; |
2cc3c58e EJ |
1028 | |
1029 | if (ofproto->backer != backer) { | |
1030 | continue; | |
1031 | } | |
1032 | ||
f231418e | 1033 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { |
2cc3c58e EJ |
1034 | if (need_revalidate |
1035 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1036 | facet_revalidate(facet); | |
8fa4d1d0 | 1037 | run_fast_rl(); |
2cc3c58e EJ |
1038 | } |
1039 | } | |
1040 | } | |
2cc3c58e EJ |
1041 | } |
1042 | ||
acf60855 JP |
1043 | if (timer_expired(&backer->next_expiration)) { |
1044 | int delay = expire(backer); | |
1045 | timer_set_duration(&backer->next_expiration, delay); | |
1046 | } | |
1047 | ||
1048 | /* Check for port changes in the dpif. */ | |
1049 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
476cb42a | 1050 | struct ofproto_dpif *ofproto; |
acf60855 JP |
1051 | struct dpif_port port; |
1052 | ||
1053 | /* Don't report on the datapath's device. */ | |
1054 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
c83b89ab | 1055 | goto next; |
acf60855 JP |
1056 | } |
1057 | ||
b9ad7294 EJ |
1058 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, |
1059 | &all_ofproto_dpifs) { | |
7d82ab2e | 1060 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { |
b9ad7294 EJ |
1061 | goto next; |
1062 | } | |
1063 | } | |
1064 | ||
476cb42a | 1065 | ofproto = lookup_ofproto_dpif_by_port_name(devname); |
acf60855 JP |
1066 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { |
1067 | /* The port was removed. If we know the datapath, | |
1068 | * report it through poll_set(). If we don't, it may be | |
1069 | * notifying us of a removal we initiated, so ignore it. | |
1070 | * If there's a pending ENOBUFS, let it stand, since | |
1071 | * everything will be reevaluated. */ | |
1072 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
1073 | sset_add(&ofproto->port_poll_set, devname); | |
1074 | ofproto->port_poll_errno = 0; | |
1075 | } | |
acf60855 JP |
1076 | } else if (!ofproto) { |
1077 | /* The port was added, but we don't know with which | |
1078 | * ofproto we should associate it. Delete it. */ | |
1079 | dpif_port_del(backer->dpif, port.port_no); | |
1080 | } | |
5b5e6a4c | 1081 | dpif_port_destroy(&port); |
acf60855 | 1082 | |
c83b89ab | 1083 | next: |
acf60855 JP |
1084 | free(devname); |
1085 | } | |
1086 | ||
1087 | if (error != EAGAIN) { | |
1088 | struct ofproto_dpif *ofproto; | |
1089 | ||
1090 | /* There was some sort of error, so propagate it to all | |
1091 | * ofprotos that use this backer. */ | |
1092 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1093 | &all_ofproto_dpifs) { | |
1094 | if (ofproto->backer == backer) { | |
1095 | sset_clear(&ofproto->port_poll_set); | |
1096 | ofproto->port_poll_errno = error; | |
1097 | } | |
1098 | } | |
1099 | } | |
1100 | ||
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 | ||
acf60855 JP |
1109 | /* Handle one or more batches of upcalls, until there's nothing left to do |
1110 | * or until we do a fixed total amount of work. | |
1111 | * | |
1112 | * We do work in batches because it can be much cheaper to set up a number | |
1113 | * of flows and fire off their patches all at once. We do multiple batches | |
1114 | * because in some cases handling a packet can cause another packet to be | |
1115 | * queued almost immediately as part of the return flow. Both | |
1116 | * optimizations can make major improvements on some benchmarks and | |
1117 | * presumably for real traffic as well. */ | |
1118 | work = 0; | |
8fa4d1d0 EJ |
1119 | while (work < max_batch) { |
1120 | int retval = handle_upcalls(backer, max_batch - work); | |
acf60855 JP |
1121 | if (retval <= 0) { |
1122 | return -retval; | |
1123 | } | |
1124 | work += retval; | |
1125 | } | |
1126 | ||
1127 | return 0; | |
1128 | } | |
1129 | ||
8fa4d1d0 EJ |
1130 | static int |
1131 | type_run_fast(const char *type) | |
1132 | { | |
1133 | struct dpif_backer *backer; | |
1134 | ||
1135 | backer = shash_find_data(&all_dpif_backers, type); | |
1136 | if (!backer) { | |
1137 | /* This is not necessarily a problem, since backers are only | |
1138 | * created on demand. */ | |
1139 | return 0; | |
1140 | } | |
1141 | ||
1142 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1143 | } | |
1144 | ||
1145 | static void | |
1146 | run_fast_rl(void) | |
1147 | { | |
1148 | static long long int port_rl = LLONG_MIN; | |
1149 | static unsigned int backer_rl = 0; | |
1150 | ||
1151 | if (time_msec() >= port_rl) { | |
1152 | struct ofproto_dpif *ofproto; | |
1153 | struct ofport_dpif *ofport; | |
1154 | ||
1155 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1156 | ||
1157 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1158 | port_run_fast(ofport); | |
1159 | } | |
1160 | } | |
1161 | port_rl = time_msec() + 200; | |
1162 | } | |
1163 | ||
1164 | /* XXX: We have to be careful not to do too much work in this function. If | |
1165 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1166 | * performance improves signifcantly, but at a cost. It's possible for the | |
1167 | * number of flows in the datapath to increase without bound, and for poll | |
1168 | * loops to take 10s of seconds. The correct solution to this problem, | |
1169 | * long term, is to separate flow miss handling into it's own thread so it | |
1170 | * isn't affected by revalidations, and expirations. Until then, this is | |
1171 | * the best we can do. */ | |
1172 | if (++backer_rl >= 10) { | |
1173 | struct shash_node *node; | |
1174 | ||
1175 | backer_rl = 0; | |
1176 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1177 | dpif_backer_run_fast(node->data, 1); | |
1178 | } | |
1179 | } | |
1180 | } | |
1181 | ||
acf60855 JP |
1182 | static void |
1183 | type_wait(const char *type) | |
1184 | { | |
1185 | struct dpif_backer *backer; | |
1186 | ||
1187 | backer = shash_find_data(&all_dpif_backers, type); | |
1188 | if (!backer) { | |
1189 | /* This is not necessarily a problem, since backers are only | |
1190 | * created on demand. */ | |
1191 | return; | |
1192 | } | |
1193 | ||
1194 | timer_wait(&backer->next_expiration); | |
1195 | } | |
1196 | \f | |
abe529af BP |
1197 | /* Basic life-cycle. */ |
1198 | ||
c57b2226 BP |
1199 | static int add_internal_flows(struct ofproto_dpif *); |
1200 | ||
abe529af BP |
1201 | static struct ofproto * |
1202 | alloc(void) | |
1203 | { | |
1204 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1205 | return &ofproto->up; | |
1206 | } | |
1207 | ||
1208 | static void | |
1209 | dealloc(struct ofproto *ofproto_) | |
1210 | { | |
1211 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1212 | free(ofproto); | |
1213 | } | |
1214 | ||
acf60855 JP |
1215 | static void |
1216 | close_dpif_backer(struct dpif_backer *backer) | |
1217 | { | |
1218 | struct shash_node *node; | |
1219 | ||
cb22974d | 1220 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1221 | |
1222 | if (--backer->refcount) { | |
1223 | return; | |
1224 | } | |
1225 | ||
8f73d537 EJ |
1226 | drop_key_clear(backer); |
1227 | hmap_destroy(&backer->drop_keys); | |
1228 | ||
7d82ab2e | 1229 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1230 | hmap_destroy(&backer->odp_to_ofport_map); |
1231 | node = shash_find(&all_dpif_backers, backer->type); | |
1232 | free(backer->type); | |
1233 | shash_delete(&all_dpif_backers, node); | |
1234 | dpif_close(backer->dpif); | |
1235 | ||
1236 | free(backer); | |
1237 | } | |
1238 | ||
1239 | /* Datapath port slated for removal from datapath. */ | |
1240 | struct odp_garbage { | |
1241 | struct list list_node; | |
1242 | uint32_t odp_port; | |
1243 | }; | |
1244 | ||
1245 | static int | |
1246 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1247 | { | |
1248 | struct dpif_backer *backer; | |
1249 | struct dpif_port_dump port_dump; | |
1250 | struct dpif_port port; | |
1251 | struct shash_node *node; | |
1252 | struct list garbage_list; | |
1253 | struct odp_garbage *garbage, *next; | |
1254 | struct sset names; | |
1255 | char *backer_name; | |
1256 | const char *name; | |
1257 | int error; | |
1258 | ||
1259 | backer = shash_find_data(&all_dpif_backers, type); | |
1260 | if (backer) { | |
1261 | backer->refcount++; | |
1262 | *backerp = backer; | |
1263 | return 0; | |
1264 | } | |
1265 | ||
1266 | backer_name = xasprintf("ovs-%s", type); | |
1267 | ||
1268 | /* Remove any existing datapaths, since we assume we're the only | |
1269 | * userspace controlling the datapath. */ | |
1270 | sset_init(&names); | |
1271 | dp_enumerate_names(type, &names); | |
1272 | SSET_FOR_EACH(name, &names) { | |
1273 | struct dpif *old_dpif; | |
1274 | ||
1275 | /* Don't remove our backer if it exists. */ | |
1276 | if (!strcmp(name, backer_name)) { | |
1277 | continue; | |
1278 | } | |
1279 | ||
1280 | if (dpif_open(name, type, &old_dpif)) { | |
1281 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1282 | } else { | |
1283 | dpif_delete(old_dpif); | |
1284 | dpif_close(old_dpif); | |
1285 | } | |
1286 | } | |
1287 | sset_destroy(&names); | |
1288 | ||
1289 | backer = xmalloc(sizeof *backer); | |
1290 | ||
1291 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1292 | free(backer_name); | |
1293 | if (error) { | |
1294 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1295 | strerror(error)); | |
4c1b1289 | 1296 | free(backer); |
acf60855 JP |
1297 | return error; |
1298 | } | |
1299 | ||
1300 | backer->type = xstrdup(type); | |
1301 | backer->refcount = 1; | |
1302 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1303 | hmap_init(&backer->drop_keys); |
acf60855 | 1304 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1305 | backer->need_revalidate = 0; |
7d82ab2e | 1306 | simap_init(&backer->tnl_backers); |
2cc3c58e | 1307 | tag_set_init(&backer->revalidate_set); |
acf60855 JP |
1308 | *backerp = backer; |
1309 | ||
1310 | dpif_flow_flush(backer->dpif); | |
1311 | ||
1312 | /* Loop through the ports already on the datapath and remove any | |
1313 | * that we don't need anymore. */ | |
1314 | list_init(&garbage_list); | |
1315 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1316 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1317 | node = shash_find(&init_ofp_ports, port.name); | |
1318 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1319 | garbage = xmalloc(sizeof *garbage); | |
1320 | garbage->odp_port = port.port_no; | |
1321 | list_push_front(&garbage_list, &garbage->list_node); | |
1322 | } | |
1323 | } | |
1324 | dpif_port_dump_done(&port_dump); | |
1325 | ||
1326 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1327 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1328 | list_remove(&garbage->list_node); | |
1329 | free(garbage); | |
1330 | } | |
1331 | ||
1332 | shash_add(&all_dpif_backers, type, backer); | |
1333 | ||
1334 | error = dpif_recv_set(backer->dpif, true); | |
1335 | if (error) { | |
1336 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1337 | type, strerror(error)); | |
1338 | close_dpif_backer(backer); | |
1339 | return error; | |
1340 | } | |
1341 | ||
1342 | return error; | |
1343 | } | |
1344 | ||
abe529af | 1345 | static int |
0f5f95a9 | 1346 | construct(struct ofproto *ofproto_) |
abe529af BP |
1347 | { |
1348 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1349 | struct shash_node *node, *next; |
91858960 | 1350 | int max_ports; |
abe529af BP |
1351 | int error; |
1352 | int i; | |
1353 | ||
acf60855 | 1354 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1355 | if (error) { |
abe529af BP |
1356 | return error; |
1357 | } | |
1358 | ||
acf60855 | 1359 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
91858960 BP |
1360 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); |
1361 | ||
6c1491fb | 1362 | ofproto->n_matches = 0; |
abe529af | 1363 | |
abe529af BP |
1364 | ofproto->netflow = NULL; |
1365 | ofproto->sflow = NULL; | |
29089a54 | 1366 | ofproto->ipfix = NULL; |
21f7563c | 1367 | ofproto->stp = NULL; |
abe529af | 1368 | hmap_init(&ofproto->bundles); |
e764773c | 1369 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
1370 | for (i = 0; i < MAX_MIRRORS; i++) { |
1371 | ofproto->mirrors[i] = NULL; | |
1372 | } | |
1373 | ofproto->has_bonded_bundles = false; | |
1374 | ||
abe529af | 1375 | hmap_init(&ofproto->facets); |
b0f7b9b5 | 1376 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 1377 | ofproto->governor = NULL; |
4d2a0f39 | 1378 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 BP |
1379 | |
1380 | for (i = 0; i < N_TABLES; i++) { | |
1381 | struct table_dpif *table = &ofproto->tables[i]; | |
1382 | ||
1383 | table->catchall_table = NULL; | |
1384 | table->other_table = NULL; | |
1385 | table->basis = random_uint32(); | |
1386 | } | |
abe529af | 1387 | |
7ee20df1 BP |
1388 | list_init(&ofproto->completions); |
1389 | ||
abe529af BP |
1390 | ofproto_dpif_unixctl_init(); |
1391 | ||
ccb7c863 | 1392 | ofproto->has_mirrors = false; |
daff3353 EJ |
1393 | ofproto->has_bundle_action = false; |
1394 | ||
52a90c29 BP |
1395 | hmap_init(&ofproto->vlandev_map); |
1396 | hmap_init(&ofproto->realdev_vid_map); | |
1397 | ||
acf60855 | 1398 | sset_init(&ofproto->ports); |
0a740f48 | 1399 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1400 | sset_init(&ofproto->port_poll_set); |
1401 | ofproto->port_poll_errno = 0; | |
1402 | ||
1403 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1404 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1405 | |
1406 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1407 | /* Check if the datapath already has this port. */ | |
1408 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1409 | sset_add(&ofproto->ports, node->name); | |
1410 | } | |
1411 | ||
1412 | free(iface_hint->br_name); | |
1413 | free(iface_hint->br_type); | |
4f9e08a5 | 1414 | free(iface_hint); |
acf60855 JP |
1415 | shash_delete(&init_ofp_ports, node); |
1416 | } | |
1417 | } | |
e1b1d06a | 1418 | |
b44a10b7 BP |
1419 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1420 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1421 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1422 | |
1423 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1424 | error = add_internal_flows(ofproto); |
1425 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1426 | ||
735d7efb AZ |
1427 | ofproto->n_hit = 0; |
1428 | ofproto->n_missed = 0; | |
1429 | ||
655ab909 AZ |
1430 | ofproto->max_n_subfacet = 0; |
1431 | ofproto->created = time_msec(); | |
1432 | ofproto->last_minute = ofproto->created; | |
1433 | memset(&ofproto->hourly, 0, sizeof ofproto->hourly); | |
1434 | memset(&ofproto->daily, 0, sizeof ofproto->daily); | |
1435 | ofproto->subfacet_add_count = 0; | |
1436 | ofproto->subfacet_del_count = 0; | |
1437 | ofproto->total_subfacet_add_count = 0; | |
1438 | ofproto->total_subfacet_del_count = 0; | |
1439 | ofproto->total_subfacet_life_span = 0; | |
1440 | ofproto->total_subfacet_count = 0; | |
1441 | ofproto->n_update_stats = 0; | |
1442 | ||
c57b2226 BP |
1443 | return error; |
1444 | } | |
1445 | ||
1446 | static int | |
1447 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1448 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1449 | { |
1450 | struct ofputil_flow_mod fm; | |
1451 | int error; | |
1452 | ||
81a76618 BP |
1453 | match_init_catchall(&fm.match); |
1454 | fm.priority = 0; | |
1455 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1456 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1457 | fm.cookie = htonll(0); |
1458 | fm.cookie_mask = htonll(0); | |
1459 | fm.table_id = TBL_INTERNAL; | |
1460 | fm.command = OFPFC_ADD; | |
1461 | fm.idle_timeout = 0; | |
1462 | fm.hard_timeout = 0; | |
1463 | fm.buffer_id = 0; | |
1464 | fm.out_port = 0; | |
1465 | fm.flags = 0; | |
f25d0cf3 BP |
1466 | fm.ofpacts = ofpacts->data; |
1467 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1468 | |
1469 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1470 | if (error) { | |
1471 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1472 | id, ofperr_to_string(error)); | |
1473 | return error; | |
1474 | } | |
1475 | ||
81a76618 | 1476 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); |
cb22974d | 1477 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1478 | |
abe529af BP |
1479 | return 0; |
1480 | } | |
1481 | ||
c57b2226 BP |
1482 | static int |
1483 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1484 | { | |
f25d0cf3 BP |
1485 | struct ofpact_controller *controller; |
1486 | uint64_t ofpacts_stub[128 / 8]; | |
1487 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1488 | int error; |
1489 | int id; | |
1490 | ||
f25d0cf3 | 1491 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1492 | id = 1; |
1493 | ||
f25d0cf3 BP |
1494 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1495 | controller->max_len = UINT16_MAX; | |
1496 | controller->controller_id = 0; | |
1497 | controller->reason = OFPR_NO_MATCH; | |
1498 | ofpact_pad(&ofpacts); | |
1499 | ||
1500 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1501 | if (error) { |
1502 | return error; | |
1503 | } | |
1504 | ||
f25d0cf3 BP |
1505 | ofpbuf_clear(&ofpacts); |
1506 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
1507 | &ofproto->no_packet_in_rule); |
1508 | return error; | |
1509 | } | |
1510 | ||
7ee20df1 BP |
1511 | static void |
1512 | complete_operations(struct ofproto_dpif *ofproto) | |
1513 | { | |
1514 | struct dpif_completion *c, *next; | |
1515 | ||
1516 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1517 | ofoperation_complete(c->op, 0); | |
1518 | list_remove(&c->list_node); | |
1519 | free(c); | |
1520 | } | |
1521 | } | |
1522 | ||
abe529af BP |
1523 | static void |
1524 | destruct(struct ofproto *ofproto_) | |
1525 | { | |
1526 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1527 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1528 | struct oftable *table; |
abe529af BP |
1529 | int i; |
1530 | ||
b44a10b7 | 1531 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1532 | complete_operations(ofproto); |
1533 | ||
0697b5c3 BP |
1534 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1535 | struct cls_cursor cursor; | |
1536 | ||
d0918789 | 1537 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1538 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1539 | ofproto_rule_destroy(&rule->up); | |
1540 | } | |
7ee20df1 BP |
1541 | } |
1542 | ||
abe529af BP |
1543 | for (i = 0; i < MAX_MIRRORS; i++) { |
1544 | mirror_destroy(ofproto->mirrors[i]); | |
1545 | } | |
1546 | ||
1547 | netflow_destroy(ofproto->netflow); | |
bae473fe | 1548 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
1549 | hmap_destroy(&ofproto->bundles); |
1550 | mac_learning_destroy(ofproto->ml); | |
1551 | ||
1552 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 1553 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 1554 | governor_destroy(ofproto->governor); |
abe529af | 1555 | |
52a90c29 BP |
1556 | hmap_destroy(&ofproto->vlandev_map); |
1557 | hmap_destroy(&ofproto->realdev_vid_map); | |
1558 | ||
acf60855 | 1559 | sset_destroy(&ofproto->ports); |
0a740f48 | 1560 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1561 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1562 | |
acf60855 | 1563 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1564 | } |
1565 | ||
1566 | static int | |
5fcc0d00 | 1567 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1568 | { |
1569 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 1570 | struct ofport_dpif *ofport; |
abe529af | 1571 | |
0aa66d6e EJ |
1572 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1573 | port_run_fast(ofport); | |
1574 | } | |
1575 | ||
5fcc0d00 BP |
1576 | return 0; |
1577 | } | |
1578 | ||
1579 | static int | |
1580 | run(struct ofproto *ofproto_) | |
1581 | { | |
1582 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1583 | struct ofport_dpif *ofport; | |
1584 | struct ofbundle *bundle; | |
1585 | int error; | |
1586 | ||
1587 | if (!clogged) { | |
1588 | complete_operations(ofproto); | |
1589 | } | |
5fcc0d00 BP |
1590 | |
1591 | error = run_fast(ofproto_); | |
1592 | if (error) { | |
1593 | return error; | |
abe529af BP |
1594 | } |
1595 | ||
abe529af | 1596 | if (ofproto->netflow) { |
6fca1ffb BP |
1597 | if (netflow_run(ofproto->netflow)) { |
1598 | send_netflow_active_timeouts(ofproto); | |
1599 | } | |
abe529af BP |
1600 | } |
1601 | if (ofproto->sflow) { | |
bae473fe | 1602 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1603 | } |
1604 | ||
1605 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1606 | port_run(ofport); | |
1607 | } | |
1608 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1609 | bundle_run(bundle); | |
1610 | } | |
1611 | ||
21f7563c | 1612 | stp_run(ofproto); |
2cc3c58e | 1613 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af | 1614 | |
6814e51f | 1615 | /* Check the consistency of a random facet, to aid debugging. */ |
4d2a0f39 EJ |
1616 | if (time_msec() >= ofproto->consistency_rl |
1617 | && !hmap_is_empty(&ofproto->facets) | |
2cc3c58e | 1618 | && !ofproto->backer->need_revalidate) { |
6814e51f BP |
1619 | struct facet *facet; |
1620 | ||
4d2a0f39 EJ |
1621 | ofproto->consistency_rl = time_msec() + 250; |
1622 | ||
6814e51f BP |
1623 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), |
1624 | struct facet, hmap_node); | |
2cc3c58e EJ |
1625 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, |
1626 | facet->tags)) { | |
6814e51f | 1627 | if (!facet_check_consistency(facet)) { |
2cc3c58e | 1628 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1629 | } |
1630 | } | |
1631 | } | |
1632 | ||
9d6ac44e BP |
1633 | if (ofproto->governor) { |
1634 | size_t n_subfacets; | |
1635 | ||
1636 | governor_run(ofproto->governor); | |
1637 | ||
1638 | /* If the governor has shrunk to its minimum size and the number of | |
1639 | * subfacets has dwindled, then drop the governor entirely. | |
1640 | * | |
1641 | * For hysteresis, the number of subfacets to drop the governor is | |
1642 | * smaller than the number needed to trigger its creation. */ | |
1643 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1644 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1645 | && governor_is_idle(ofproto->governor)) { | |
1646 | governor_destroy(ofproto->governor); | |
1647 | ofproto->governor = NULL; | |
1648 | } | |
1649 | } | |
1650 | ||
abe529af BP |
1651 | return 0; |
1652 | } | |
1653 | ||
1654 | static void | |
1655 | wait(struct ofproto *ofproto_) | |
1656 | { | |
1657 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1658 | struct ofport_dpif *ofport; | |
1659 | struct ofbundle *bundle; | |
1660 | ||
7ee20df1 BP |
1661 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1662 | poll_immediate_wake(); | |
1663 | } | |
1664 | ||
acf60855 JP |
1665 | dpif_wait(ofproto->backer->dpif); |
1666 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1667 | if (ofproto->sflow) { |
bae473fe | 1668 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1669 | } |
2cc3c58e | 1670 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { |
abe529af BP |
1671 | poll_immediate_wake(); |
1672 | } | |
1673 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1674 | port_wait(ofport); | |
1675 | } | |
1676 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1677 | bundle_wait(bundle); | |
1678 | } | |
6fca1ffb BP |
1679 | if (ofproto->netflow) { |
1680 | netflow_wait(ofproto->netflow); | |
1681 | } | |
1c313b88 | 1682 | mac_learning_wait(ofproto->ml); |
21f7563c | 1683 | stp_wait(ofproto); |
2cc3c58e | 1684 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1685 | /* Shouldn't happen, but if it does just go around again. */ |
1686 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1687 | poll_immediate_wake(); | |
abe529af | 1688 | } |
9d6ac44e BP |
1689 | if (ofproto->governor) { |
1690 | governor_wait(ofproto->governor); | |
1691 | } | |
abe529af BP |
1692 | } |
1693 | ||
0d085684 BP |
1694 | static void |
1695 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1696 | { | |
1697 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1698 | ||
1699 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1700 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1701 | } | |
1702 | ||
abe529af BP |
1703 | static void |
1704 | flush(struct ofproto *ofproto_) | |
1705 | { | |
1706 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1707 | struct subfacet *subfacet, *next_subfacet; |
1708 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1709 | int n_batch; | |
b0f7b9b5 | 1710 | |
acf60855 JP |
1711 | n_batch = 0; |
1712 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
1713 | &ofproto->subfacets) { | |
1714 | if (subfacet->path != SF_NOT_INSTALLED) { | |
1715 | batch[n_batch++] = subfacet; | |
1716 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
1717 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1718 | n_batch = 0; | |
1719 | } | |
1720 | } else { | |
1721 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1722 | } |
abe529af | 1723 | } |
acf60855 JP |
1724 | |
1725 | if (n_batch > 0) { | |
1726 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1727 | } | |
abe529af BP |
1728 | } |
1729 | ||
6c1491fb BP |
1730 | static void |
1731 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1732 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1733 | { |
1734 | *arp_match_ip = true; | |
9e1fd49b BP |
1735 | *actions = (OFPUTIL_A_OUTPUT | |
1736 | OFPUTIL_A_SET_VLAN_VID | | |
1737 | OFPUTIL_A_SET_VLAN_PCP | | |
1738 | OFPUTIL_A_STRIP_VLAN | | |
1739 | OFPUTIL_A_SET_DL_SRC | | |
1740 | OFPUTIL_A_SET_DL_DST | | |
1741 | OFPUTIL_A_SET_NW_SRC | | |
1742 | OFPUTIL_A_SET_NW_DST | | |
1743 | OFPUTIL_A_SET_NW_TOS | | |
1744 | OFPUTIL_A_SET_TP_SRC | | |
1745 | OFPUTIL_A_SET_TP_DST | | |
1746 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1747 | } |
1748 | ||
1749 | static void | |
307975da | 1750 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1751 | { |
1752 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1753 | struct dpif_dp_stats s; |
6c1491fb BP |
1754 | |
1755 | strcpy(ots->name, "classifier"); | |
1756 | ||
acf60855 JP |
1757 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
1758 | ||
307975da SH |
1759 | ots->lookup_count = htonll(s.n_hit + s.n_missed); |
1760 | ots->matched_count = htonll(s.n_hit + ofproto->n_matches); | |
6c1491fb BP |
1761 | } |
1762 | ||
abe529af BP |
1763 | static struct ofport * |
1764 | port_alloc(void) | |
1765 | { | |
1766 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1767 | return &port->up; | |
1768 | } | |
1769 | ||
1770 | static void | |
1771 | port_dealloc(struct ofport *port_) | |
1772 | { | |
1773 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1774 | free(port); | |
1775 | } | |
1776 | ||
1777 | static int | |
1778 | port_construct(struct ofport *port_) | |
1779 | { | |
1780 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1781 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1782 | const struct netdev *netdev = port->up.netdev; |
e1b1d06a JP |
1783 | struct dpif_port dpif_port; |
1784 | int error; | |
abe529af | 1785 | |
2cc3c58e | 1786 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1787 | port->bundle = NULL; |
1788 | port->cfm = NULL; | |
ccc09689 | 1789 | port->bfd = NULL; |
abe529af | 1790 | port->tag = tag_create_random(); |
d5ffa7f2 | 1791 | port->may_enable = true; |
21f7563c JP |
1792 | port->stp_port = NULL; |
1793 | port->stp_state = STP_DISABLED; | |
b9ad7294 | 1794 | port->tnl_port = NULL; |
8b36f51e | 1795 | hmap_init(&port->priorities); |
52a90c29 BP |
1796 | port->realdev_ofp_port = 0; |
1797 | port->vlandev_vid = 0; | |
b9ad7294 | 1798 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1799 | |
b9ad7294 | 1800 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1801 | /* By bailing out here, we don't submit the port to the sFlow module |
1802 | * to be considered for counter polling export. This is correct | |
1803 | * because the patch port represents an interface that sFlow considers | |
1804 | * to be "internal" to the switch as a whole, and therefore not an | |
1805 | * candidate for counter polling. */ | |
0a740f48 EJ |
1806 | port->odp_port = OVSP_NONE; |
1807 | return 0; | |
1808 | } | |
1809 | ||
acf60855 | 1810 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
b9ad7294 | 1811 | netdev_vport_get_dpif_port(netdev), |
e1b1d06a JP |
1812 | &dpif_port); |
1813 | if (error) { | |
1814 | return error; | |
1815 | } | |
1816 | ||
1817 | port->odp_port = dpif_port.port_no; | |
1818 | ||
b9ad7294 EJ |
1819 | if (netdev_get_tunnel_config(netdev)) { |
1820 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1821 | } else { | |
1822 | /* Sanity-check that a mapping doesn't already exist. This | |
1823 | * shouldn't happen for non-tunnel ports. */ | |
1824 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1825 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1826 | dpif_port.name); | |
da78d43d | 1827 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1828 | return EBUSY; |
1829 | } | |
e1b1d06a | 1830 | |
b9ad7294 EJ |
1831 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
1832 | hash_int(port->odp_port, 0)); | |
1833 | } | |
da78d43d | 1834 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1835 | |
abe529af | 1836 | if (ofproto->sflow) { |
e1b1d06a | 1837 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1838 | } |
1839 | ||
1840 | return 0; | |
1841 | } | |
1842 | ||
1843 | static void | |
1844 | port_destruct(struct ofport *port_) | |
1845 | { | |
1846 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1847 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1848 | const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); |
02f8d646 | 1849 | const char *devname = netdev_get_name(port->up.netdev); |
abe529af | 1850 | |
a614d823 | 1851 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1852 | /* The underlying device is still there, so delete it. This |
1853 | * happens when the ofproto is being destroyed, since the caller | |
1854 | * assumes that removal of attached ports will happen as part of | |
1855 | * destruction. */ | |
a614d823 KM |
1856 | if (!port->tnl_port) { |
1857 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1858 | } | |
1859 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
acf60855 JP |
1860 | } |
1861 | ||
b9ad7294 | 1862 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { |
0a740f48 EJ |
1863 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1864 | } | |
1865 | ||
b9ad7294 | 1866 | tnl_port_del(port->tnl_port); |
02f8d646 | 1867 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1868 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
2cc3c58e | 1869 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1870 | bundle_remove(port_); |
a5610457 | 1871 | set_cfm(port_, NULL); |
abe529af | 1872 | if (ofproto->sflow) { |
bae473fe | 1873 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1874 | } |
8b36f51e EJ |
1875 | |
1876 | ofport_clear_priorities(port); | |
1877 | hmap_destroy(&port->priorities); | |
abe529af BP |
1878 | } |
1879 | ||
1880 | static void | |
1881 | port_modified(struct ofport *port_) | |
1882 | { | |
1883 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1884 | ||
1885 | if (port->bundle && port->bundle->bond) { | |
1886 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1887 | } | |
1888 | } | |
1889 | ||
1890 | static void | |
9e1fd49b | 1891 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1892 | { |
1893 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1894 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1895 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1896 | |
9e1fd49b | 1897 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1898 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1899 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1900 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1901 | |
9e1fd49b | 1902 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1903 | bundle_update(port->bundle); |
1904 | } | |
abe529af BP |
1905 | } |
1906 | } | |
1907 | ||
1908 | static int | |
1909 | set_sflow(struct ofproto *ofproto_, | |
1910 | const struct ofproto_sflow_options *sflow_options) | |
1911 | { | |
1912 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1913 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1914 | |
abe529af | 1915 | if (sflow_options) { |
bae473fe | 1916 | if (!ds) { |
abe529af BP |
1917 | struct ofport_dpif *ofport; |
1918 | ||
4213f19d | 1919 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1920 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1921 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1922 | } |
2cc3c58e | 1923 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1924 | } |
bae473fe | 1925 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1926 | } else { |
6ff686f2 PS |
1927 | if (ds) { |
1928 | dpif_sflow_destroy(ds); | |
2cc3c58e | 1929 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1930 | ofproto->sflow = NULL; |
1931 | } | |
abe529af BP |
1932 | } |
1933 | return 0; | |
1934 | } | |
1935 | ||
29089a54 RL |
1936 | static int |
1937 | set_ipfix( | |
1938 | struct ofproto *ofproto_, | |
1939 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1940 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1941 | size_t n_flow_exporters_options) | |
1942 | { | |
1943 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1944 | struct dpif_ipfix *di = ofproto->ipfix; | |
1945 | ||
1946 | if (bridge_exporter_options || flow_exporters_options) { | |
1947 | if (!di) { | |
1948 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1949 | } | |
1950 | dpif_ipfix_set_options( | |
1951 | di, bridge_exporter_options, flow_exporters_options, | |
1952 | n_flow_exporters_options); | |
1953 | } else { | |
1954 | if (di) { | |
1955 | dpif_ipfix_destroy(di); | |
1956 | ofproto->ipfix = NULL; | |
1957 | } | |
1958 | } | |
1959 | return 0; | |
1960 | } | |
1961 | ||
abe529af | 1962 | static int |
a5610457 | 1963 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1964 | { |
1965 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1966 | int error; | |
1967 | ||
a5610457 | 1968 | if (!s) { |
abe529af BP |
1969 | error = 0; |
1970 | } else { | |
1971 | if (!ofport->cfm) { | |
8c977421 EJ |
1972 | struct ofproto_dpif *ofproto; |
1973 | ||
1974 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1975 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f629657 | 1976 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1977 | } |
1978 | ||
a5610457 | 1979 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1980 | return 0; |
1981 | } | |
1982 | ||
1983 | error = EINVAL; | |
1984 | } | |
1985 | cfm_destroy(ofport->cfm); | |
1986 | ofport->cfm = NULL; | |
1987 | return error; | |
1988 | } | |
1989 | ||
9a9e3786 BP |
1990 | static bool |
1991 | get_cfm_status(const struct ofport *ofport_, | |
1992 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1993 | { |
1994 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1995 | ||
1996 | if (ofport->cfm) { | |
9a9e3786 BP |
1997 | status->faults = cfm_get_fault(ofport->cfm); |
1998 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1999 | status->health = cfm_get_health(ofport->cfm); | |
2000 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
2001 | return true; | |
1de11730 | 2002 | } else { |
9a9e3786 | 2003 | return false; |
1de11730 EJ |
2004 | } |
2005 | } | |
ccc09689 EJ |
2006 | |
2007 | static int | |
2008 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
2009 | { | |
2010 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
2011 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2012 | struct bfd *old; | |
2013 | ||
2014 | old = ofport->bfd; | |
2015 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
2016 | if (ofport->bfd != old) { | |
2017 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2018 | } | |
2019 | ||
2020 | return 0; | |
2021 | } | |
2022 | ||
2023 | static int | |
2024 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
2025 | { | |
2026 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2027 | ||
2028 | if (ofport->bfd) { | |
2029 | bfd_get_status(ofport->bfd, smap); | |
2030 | return 0; | |
2031 | } else { | |
2032 | return ENOENT; | |
2033 | } | |
2034 | } | |
abe529af | 2035 | \f |
21f7563c JP |
2036 | /* Spanning Tree. */ |
2037 | ||
2038 | static void | |
2039 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2040 | { | |
2041 | struct ofproto_dpif *ofproto = ofproto_; | |
2042 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2043 | struct ofport_dpif *ofport; | |
2044 | ||
2045 | ofport = stp_port_get_aux(sp); | |
2046 | if (!ofport) { | |
2047 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2048 | ofproto->up.name, port_num); | |
2049 | } else { | |
2050 | struct eth_header *eth = pkt->l2; | |
2051 | ||
2052 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2053 | if (eth_addr_is_zero(eth->eth_src)) { | |
2054 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2055 | "with unknown MAC", ofproto->up.name, port_num); | |
2056 | } else { | |
97d6520b | 2057 | send_packet(ofport, pkt); |
21f7563c JP |
2058 | } |
2059 | } | |
2060 | ofpbuf_delete(pkt); | |
2061 | } | |
2062 | ||
2063 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2064 | static int | |
2065 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2066 | { | |
2067 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2068 | ||
2069 | /* Only revalidate flows if the configuration changed. */ | |
2070 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2071 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2072 | } |
2073 | ||
2074 | if (s) { | |
2075 | if (!ofproto->stp) { | |
2076 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2077 | send_bpdu_cb, ofproto); | |
2078 | ofproto->stp_last_tick = time_msec(); | |
2079 | } | |
2080 | ||
2081 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2082 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2083 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2084 | stp_set_max_age(ofproto->stp, s->max_age); | |
2085 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2086 | } else { | |
851bf71d EJ |
2087 | struct ofport *ofport; |
2088 | ||
2089 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2090 | set_stp_port(ofport, NULL); | |
2091 | } | |
2092 | ||
21f7563c JP |
2093 | stp_destroy(ofproto->stp); |
2094 | ofproto->stp = NULL; | |
2095 | } | |
2096 | ||
2097 | return 0; | |
2098 | } | |
2099 | ||
2100 | static int | |
2101 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2102 | { | |
2103 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2104 | ||
2105 | if (ofproto->stp) { | |
2106 | s->enabled = true; | |
2107 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2108 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2109 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2110 | } else { | |
2111 | s->enabled = false; | |
2112 | } | |
2113 | ||
2114 | return 0; | |
2115 | } | |
2116 | ||
2117 | static void | |
2118 | update_stp_port_state(struct ofport_dpif *ofport) | |
2119 | { | |
2120 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2121 | enum stp_state state; | |
2122 | ||
2123 | /* Figure out new state. */ | |
2124 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2125 | : STP_DISABLED; | |
2126 | ||
2127 | /* Update state. */ | |
2128 | if (ofport->stp_state != state) { | |
9e1fd49b | 2129 | enum ofputil_port_state of_state; |
21f7563c JP |
2130 | bool fwd_change; |
2131 | ||
2132 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2133 | netdev_get_name(ofport->up.netdev), | |
2134 | stp_state_name(ofport->stp_state), | |
2135 | stp_state_name(state)); | |
2136 | if (stp_learn_in_state(ofport->stp_state) | |
2137 | != stp_learn_in_state(state)) { | |
2138 | /* xxx Learning action flows should also be flushed. */ | |
2cc3c58e EJ |
2139 | mac_learning_flush(ofproto->ml, |
2140 | &ofproto->backer->revalidate_set); | |
21f7563c JP |
2141 | } |
2142 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2143 | != stp_forward_in_state(state); | |
2144 | ||
2cc3c58e | 2145 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2146 | ofport->stp_state = state; |
2147 | ofport->stp_state_entered = time_msec(); | |
2148 | ||
b308140a | 2149 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2150 | bundle_update(ofport->bundle); |
2151 | } | |
2152 | ||
2153 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2154 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2155 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2156 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2157 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2158 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2159 | : 0); | |
21f7563c JP |
2160 | ofproto_port_set_state(&ofport->up, of_state); |
2161 | } | |
2162 | } | |
2163 | ||
2164 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2165 | * caller is responsible for assigning STP port numbers and ensuring | |
2166 | * there are no duplicates. */ | |
2167 | static int | |
2168 | set_stp_port(struct ofport *ofport_, | |
2169 | const struct ofproto_port_stp_settings *s) | |
2170 | { | |
2171 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2172 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2173 | struct stp_port *sp = ofport->stp_port; | |
2174 | ||
2175 | if (!s || !s->enable) { | |
2176 | if (sp) { | |
2177 | ofport->stp_port = NULL; | |
2178 | stp_port_disable(sp); | |
ecd12731 | 2179 | update_stp_port_state(ofport); |
21f7563c JP |
2180 | } |
2181 | return 0; | |
2182 | } else if (sp && stp_port_no(sp) != s->port_num | |
2183 | && ofport == stp_port_get_aux(sp)) { | |
2184 | /* The port-id changed, so disable the old one if it's not | |
2185 | * already in use by another port. */ | |
2186 | stp_port_disable(sp); | |
2187 | } | |
2188 | ||
2189 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2190 | stp_port_enable(sp); | |
2191 | ||
2192 | stp_port_set_aux(sp, ofport); | |
2193 | stp_port_set_priority(sp, s->priority); | |
2194 | stp_port_set_path_cost(sp, s->path_cost); | |
2195 | ||
2196 | update_stp_port_state(ofport); | |
2197 | ||
2198 | return 0; | |
2199 | } | |
2200 | ||
2201 | static int | |
2202 | get_stp_port_status(struct ofport *ofport_, | |
2203 | struct ofproto_port_stp_status *s) | |
2204 | { | |
2205 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2206 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2207 | struct stp_port *sp = ofport->stp_port; | |
2208 | ||
2209 | if (!ofproto->stp || !sp) { | |
2210 | s->enabled = false; | |
2211 | return 0; | |
2212 | } | |
2213 | ||
2214 | s->enabled = true; | |
2215 | s->port_id = stp_port_get_id(sp); | |
2216 | s->state = stp_port_get_state(sp); | |
2217 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2218 | s->role = stp_port_get_role(sp); | |
80740385 | 2219 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2220 | |
2221 | return 0; | |
2222 | } | |
2223 | ||
2224 | static void | |
2225 | stp_run(struct ofproto_dpif *ofproto) | |
2226 | { | |
2227 | if (ofproto->stp) { | |
2228 | long long int now = time_msec(); | |
2229 | long long int elapsed = now - ofproto->stp_last_tick; | |
2230 | struct stp_port *sp; | |
2231 | ||
2232 | if (elapsed > 0) { | |
2233 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2234 | ofproto->stp_last_tick = now; | |
2235 | } | |
2236 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2237 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2238 | ||
2239 | if (ofport) { | |
2240 | update_stp_port_state(ofport); | |
2241 | } | |
2242 | } | |
6ae50723 EJ |
2243 | |
2244 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2cc3c58e | 2245 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
6ae50723 | 2246 | } |
21f7563c JP |
2247 | } |
2248 | } | |
2249 | ||
2250 | static void | |
2251 | stp_wait(struct ofproto_dpif *ofproto) | |
2252 | { | |
2253 | if (ofproto->stp) { | |
2254 | poll_timer_wait(1000); | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | /* Returns true if STP should process 'flow'. */ | |
2259 | static bool | |
2260 | stp_should_process_flow(const struct flow *flow) | |
2261 | { | |
2262 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2263 | } | |
2264 | ||
2265 | static void | |
2266 | stp_process_packet(const struct ofport_dpif *ofport, | |
2267 | const struct ofpbuf *packet) | |
2268 | { | |
2269 | struct ofpbuf payload = *packet; | |
2270 | struct eth_header *eth = payload.data; | |
2271 | struct stp_port *sp = ofport->stp_port; | |
2272 | ||
2273 | /* Sink packets on ports that have STP disabled when the bridge has | |
2274 | * STP enabled. */ | |
2275 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2276 | return; | |
2277 | } | |
2278 | ||
2279 | /* Trim off padding on payload. */ | |
c573540b BP |
2280 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
2281 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
2282 | } |
2283 | ||
2284 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2285 | stp_received_bpdu(sp, payload.data, payload.size); | |
2286 | } | |
2287 | } | |
2288 | \f | |
8b36f51e EJ |
2289 | static struct priority_to_dscp * |
2290 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2291 | { | |
2292 | struct priority_to_dscp *pdscp; | |
2293 | uint32_t hash; | |
2294 | ||
2295 | hash = hash_int(priority, 0); | |
2296 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2297 | if (pdscp->priority == priority) { | |
2298 | return pdscp; | |
2299 | } | |
2300 | } | |
2301 | return NULL; | |
2302 | } | |
2303 | ||
2304 | static void | |
2305 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2306 | { | |
2307 | struct priority_to_dscp *pdscp, *next; | |
2308 | ||
2309 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2310 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2311 | free(pdscp); | |
2312 | } | |
2313 | } | |
2314 | ||
2315 | static int | |
2316 | set_queues(struct ofport *ofport_, | |
2317 | const struct ofproto_port_queue *qdscp_list, | |
2318 | size_t n_qdscp) | |
2319 | { | |
2320 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2321 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2322 | struct hmap new = HMAP_INITIALIZER(&new); | |
2323 | size_t i; | |
2324 | ||
2325 | for (i = 0; i < n_qdscp; i++) { | |
2326 | struct priority_to_dscp *pdscp; | |
2327 | uint32_t priority; | |
2328 | uint8_t dscp; | |
2329 | ||
2330 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
acf60855 | 2331 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, |
8b36f51e EJ |
2332 | &priority)) { |
2333 | continue; | |
2334 | } | |
2335 | ||
2336 | pdscp = get_priority(ofport, priority); | |
2337 | if (pdscp) { | |
2338 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2339 | } else { | |
2340 | pdscp = xmalloc(sizeof *pdscp); | |
2341 | pdscp->priority = priority; | |
2342 | pdscp->dscp = dscp; | |
2cc3c58e | 2343 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2344 | } |
2345 | ||
2346 | if (pdscp->dscp != dscp) { | |
2347 | pdscp->dscp = dscp; | |
2cc3c58e | 2348 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2349 | } |
2350 | ||
2351 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2352 | } | |
2353 | ||
2354 | if (!hmap_is_empty(&ofport->priorities)) { | |
2355 | ofport_clear_priorities(ofport); | |
2cc3c58e | 2356 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2357 | } |
2358 | ||
2359 | hmap_swap(&new, &ofport->priorities); | |
2360 | hmap_destroy(&new); | |
2361 | ||
2362 | return 0; | |
2363 | } | |
2364 | \f | |
abe529af BP |
2365 | /* Bundles. */ |
2366 | ||
b44a10b7 BP |
2367 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2368 | * ofproto to revalidate every flow. | |
2369 | * | |
2370 | * Normally MAC learning entries are removed only from the ofproto associated | |
2371 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2372 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2373 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2374 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2375 | * with the host from which it migrated. */ | |
abe529af | 2376 | static void |
b44a10b7 | 2377 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2378 | { |
2379 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2380 | struct mac_learning *ml = ofproto->ml; | |
2381 | struct mac_entry *mac, *next_mac; | |
2382 | ||
2cc3c58e | 2383 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2384 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2385 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2386 | if (all_ofprotos) { |
2387 | struct ofproto_dpif *o; | |
2388 | ||
2389 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2390 | if (o != ofproto) { | |
2391 | struct mac_entry *e; | |
2392 | ||
2393 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2394 | NULL); | |
2395 | if (e) { | |
b44a10b7 BP |
2396 | mac_learning_expire(o->ml, e); |
2397 | } | |
2398 | } | |
2399 | } | |
2400 | } | |
2401 | ||
abe529af BP |
2402 | mac_learning_expire(ml, mac); |
2403 | } | |
2404 | } | |
2405 | } | |
2406 | ||
2407 | static struct ofbundle * | |
2408 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2409 | { | |
2410 | struct ofbundle *bundle; | |
2411 | ||
2412 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2413 | &ofproto->bundles) { | |
2414 | if (bundle->aux == aux) { | |
2415 | return bundle; | |
2416 | } | |
2417 | } | |
2418 | return NULL; | |
2419 | } | |
2420 | ||
2421 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2422 | * ones that are found to 'bundles'. */ | |
2423 | static void | |
2424 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2425 | void **auxes, size_t n_auxes, | |
2426 | struct hmapx *bundles) | |
2427 | { | |
2428 | size_t i; | |
2429 | ||
2430 | hmapx_init(bundles); | |
2431 | for (i = 0; i < n_auxes; i++) { | |
2432 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2433 | if (bundle) { | |
2434 | hmapx_add(bundles, bundle); | |
2435 | } | |
2436 | } | |
2437 | } | |
2438 | ||
7bde8dd8 JP |
2439 | static void |
2440 | bundle_update(struct ofbundle *bundle) | |
2441 | { | |
2442 | struct ofport_dpif *port; | |
2443 | ||
2444 | bundle->floodable = true; | |
2445 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2446 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2447 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2448 | bundle->floodable = false; |
2449 | break; | |
2450 | } | |
2451 | } | |
2452 | } | |
2453 | ||
abe529af BP |
2454 | static void |
2455 | bundle_del_port(struct ofport_dpif *port) | |
2456 | { | |
2457 | struct ofbundle *bundle = port->bundle; | |
2458 | ||
2cc3c58e | 2459 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2460 | |
abe529af BP |
2461 | list_remove(&port->bundle_node); |
2462 | port->bundle = NULL; | |
2463 | ||
2464 | if (bundle->lacp) { | |
2465 | lacp_slave_unregister(bundle->lacp, port); | |
2466 | } | |
2467 | if (bundle->bond) { | |
2468 | bond_slave_unregister(bundle->bond, port); | |
2469 | } | |
2470 | ||
7bde8dd8 | 2471 | bundle_update(bundle); |
abe529af BP |
2472 | } |
2473 | ||
2474 | static bool | |
2475 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
df53d41c | 2476 | struct lacp_slave_settings *lacp) |
abe529af BP |
2477 | { |
2478 | struct ofport_dpif *port; | |
2479 | ||
2480 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2481 | if (!port) { | |
2482 | return false; | |
2483 | } | |
2484 | ||
2485 | if (port->bundle != bundle) { | |
2cc3c58e | 2486 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2487 | if (port->bundle) { |
2488 | bundle_del_port(port); | |
2489 | } | |
2490 | ||
2491 | port->bundle = bundle; | |
2492 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2493 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2494 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2495 | bundle->floodable = false; |
2496 | } | |
2497 | } | |
2498 | if (lacp) { | |
2cc3c58e | 2499 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2500 | lacp_slave_register(bundle->lacp, port, lacp); |
2501 | } | |
2502 | ||
2503 | return true; | |
2504 | } | |
2505 | ||
2506 | static void | |
2507 | bundle_destroy(struct ofbundle *bundle) | |
2508 | { | |
2509 | struct ofproto_dpif *ofproto; | |
2510 | struct ofport_dpif *port, *next_port; | |
2511 | int i; | |
2512 | ||
2513 | if (!bundle) { | |
2514 | return; | |
2515 | } | |
2516 | ||
2517 | ofproto = bundle->ofproto; | |
2518 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2519 | struct ofmirror *m = ofproto->mirrors[i]; | |
2520 | if (m) { | |
2521 | if (m->out == bundle) { | |
2522 | mirror_destroy(m); | |
2523 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2524 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2cc3c58e | 2525 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2526 | } |
2527 | } | |
2528 | } | |
2529 | ||
2530 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2531 | bundle_del_port(port); | |
2532 | } | |
2533 | ||
b44a10b7 | 2534 | bundle_flush_macs(bundle, true); |
abe529af BP |
2535 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2536 | free(bundle->name); | |
2537 | free(bundle->trunks); | |
2538 | lacp_destroy(bundle->lacp); | |
2539 | bond_destroy(bundle->bond); | |
2540 | free(bundle); | |
2541 | } | |
2542 | ||
2543 | static int | |
2544 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2545 | const struct ofproto_bundle_settings *s) | |
2546 | { | |
2547 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2548 | bool need_flush = false; | |
abe529af BP |
2549 | struct ofport_dpif *port; |
2550 | struct ofbundle *bundle; | |
ecac4ebf BP |
2551 | unsigned long *trunks; |
2552 | int vlan; | |
abe529af BP |
2553 | size_t i; |
2554 | bool ok; | |
2555 | ||
2556 | if (!s) { | |
2557 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2558 | return 0; | |
2559 | } | |
2560 | ||
cb22974d BP |
2561 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2562 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2563 | |
2564 | bundle = bundle_lookup(ofproto, aux); | |
2565 | if (!bundle) { | |
2566 | bundle = xmalloc(sizeof *bundle); | |
2567 | ||
2568 | bundle->ofproto = ofproto; | |
2569 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2570 | hash_pointer(aux, 0)); | |
2571 | bundle->aux = aux; | |
2572 | bundle->name = NULL; | |
2573 | ||
2574 | list_init(&bundle->ports); | |
ecac4ebf | 2575 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2576 | bundle->vlan = -1; |
2577 | bundle->trunks = NULL; | |
5e9ceccd | 2578 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2579 | bundle->lacp = NULL; |
2580 | bundle->bond = NULL; | |
2581 | ||
2582 | bundle->floodable = true; | |
2583 | ||
2584 | bundle->src_mirrors = 0; | |
2585 | bundle->dst_mirrors = 0; | |
2586 | bundle->mirror_out = 0; | |
2587 | } | |
2588 | ||
2589 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2590 | free(bundle->name); | |
2591 | bundle->name = xstrdup(s->name); | |
2592 | } | |
2593 | ||
2594 | /* LACP. */ | |
2595 | if (s->lacp) { | |
2596 | if (!bundle->lacp) { | |
2cc3c58e | 2597 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2598 | bundle->lacp = lacp_create(); |
2599 | } | |
2600 | lacp_configure(bundle->lacp, s->lacp); | |
2601 | } else { | |
2602 | lacp_destroy(bundle->lacp); | |
2603 | bundle->lacp = NULL; | |
2604 | } | |
2605 | ||
2606 | /* Update set of ports. */ | |
2607 | ok = true; | |
2608 | for (i = 0; i < s->n_slaves; i++) { | |
2609 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2610 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2611 | ok = false; |
2612 | } | |
2613 | } | |
2614 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2615 | struct ofport_dpif *next_port; | |
2616 | ||
2617 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2618 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2619 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2620 | goto found; |
2621 | } | |
2622 | } | |
2623 | ||
2624 | bundle_del_port(port); | |
2625 | found: ; | |
2626 | } | |
2627 | } | |
cb22974d | 2628 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2629 | |
2630 | if (list_is_empty(&bundle->ports)) { | |
2631 | bundle_destroy(bundle); | |
2632 | return EINVAL; | |
2633 | } | |
2634 | ||
ecac4ebf | 2635 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2636 | if (s->vlan_mode != bundle->vlan_mode |
2637 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2638 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2639 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2640 | need_flush = true; |
2641 | } | |
2642 | ||
abe529af | 2643 | /* Set VLAN tag. */ |
ecac4ebf BP |
2644 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2645 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2646 | : 0); | |
2647 | if (vlan != bundle->vlan) { | |
2648 | bundle->vlan = vlan; | |
abe529af BP |
2649 | need_flush = true; |
2650 | } | |
2651 | ||
2652 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2653 | switch (s->vlan_mode) { |
2654 | case PORT_VLAN_ACCESS: | |
2655 | trunks = NULL; | |
2656 | break; | |
2657 | ||
2658 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2659 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2660 | break; |
2661 | ||
2662 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2663 | case PORT_VLAN_NATIVE_TAGGED: | |
2664 | if (vlan != 0 && (!s->trunks | |
2665 | || !bitmap_is_set(s->trunks, vlan) | |
2666 | || bitmap_is_set(s->trunks, 0))) { | |
2667 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2668 | if (s->trunks) { | |
2669 | trunks = bitmap_clone(s->trunks, 4096); | |
2670 | } else { | |
2671 | trunks = bitmap_allocate1(4096); | |
2672 | } | |
2673 | bitmap_set1(trunks, vlan); | |
2674 | bitmap_set0(trunks, 0); | |
2675 | } else { | |
ebc56baa | 2676 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2677 | } |
2678 | break; | |
2679 | ||
2680 | default: | |
2681 | NOT_REACHED(); | |
2682 | } | |
abe529af BP |
2683 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2684 | free(bundle->trunks); | |
ecac4ebf BP |
2685 | if (trunks == s->trunks) { |
2686 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2687 | } else { | |
2688 | bundle->trunks = trunks; | |
2689 | trunks = NULL; | |
2690 | } | |
abe529af BP |
2691 | need_flush = true; |
2692 | } | |
ecac4ebf BP |
2693 | if (trunks != s->trunks) { |
2694 | free(trunks); | |
2695 | } | |
abe529af BP |
2696 | |
2697 | /* Bonding. */ | |
2698 | if (!list_is_short(&bundle->ports)) { | |
2699 | bundle->ofproto->has_bonded_bundles = true; | |
2700 | if (bundle->bond) { | |
2701 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2702 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2703 | } |
2704 | } else { | |
2705 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2706 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2707 | } |
2708 | ||
2709 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2710 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2711 | } |
2712 | } else { | |
2713 | bond_destroy(bundle->bond); | |
2714 | bundle->bond = NULL; | |
2715 | } | |
2716 | ||
2717 | /* If we changed something that would affect MAC learning, un-learn | |
2718 | * everything on this port and force flow revalidation. */ | |
2719 | if (need_flush) { | |
b44a10b7 | 2720 | bundle_flush_macs(bundle, false); |
abe529af BP |
2721 | } |
2722 | ||
2723 | return 0; | |
2724 | } | |
2725 | ||
2726 | static void | |
2727 | bundle_remove(struct ofport *port_) | |
2728 | { | |
2729 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2730 | struct ofbundle *bundle = port->bundle; | |
2731 | ||
2732 | if (bundle) { | |
2733 | bundle_del_port(port); | |
2734 | if (list_is_empty(&bundle->ports)) { | |
2735 | bundle_destroy(bundle); | |
2736 | } else if (list_is_short(&bundle->ports)) { | |
2737 | bond_destroy(bundle->bond); | |
2738 | bundle->bond = NULL; | |
2739 | } | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | static void | |
5f877369 | 2744 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2745 | { |
2746 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2747 | struct ofport_dpif *port = port_; | |
2748 | uint8_t ea[ETH_ADDR_LEN]; | |
2749 | int error; | |
2750 | ||
2751 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2752 | if (!error) { | |
abe529af | 2753 | struct ofpbuf packet; |
5f877369 | 2754 | void *packet_pdu; |
abe529af BP |
2755 | |
2756 | ofpbuf_init(&packet, 0); | |
2757 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2758 | pdu_size); |
2759 | memcpy(packet_pdu, pdu, pdu_size); | |
2760 | ||
97d6520b | 2761 | send_packet(port, &packet); |
abe529af BP |
2762 | ofpbuf_uninit(&packet); |
2763 | } else { | |
2764 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2765 | "%s (%s)", port->bundle->name, | |
2766 | netdev_get_name(port->up.netdev), strerror(error)); | |
2767 | } | |
2768 | } | |
2769 | ||
2770 | static void | |
2771 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2772 | { | |
2773 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2774 | int error, n_packets, n_errors; | |
2775 | struct mac_entry *e; | |
2776 | ||
2777 | error = n_packets = n_errors = 0; | |
2778 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2779 | if (e->port.p != bundle) { | |
ea131871 JG |
2780 | struct ofpbuf *learning_packet; |
2781 | struct ofport_dpif *port; | |
4dd1e3ca | 2782 | void *port_void; |
ea131871 JG |
2783 | int ret; |
2784 | ||
4dd1e3ca BP |
2785 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2786 | * struct ofport_dpif more effective. */ | |
2787 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2788 | e->mac, e->vlan, | |
2789 | &port_void); | |
2790 | port = port_void; | |
97d6520b | 2791 | ret = send_packet(port, learning_packet); |
ea131871 | 2792 | ofpbuf_delete(learning_packet); |
abe529af BP |
2793 | if (ret) { |
2794 | error = ret; | |
2795 | n_errors++; | |
2796 | } | |
2797 | n_packets++; | |
2798 | } | |
2799 | } | |
2800 | ||
2801 | if (n_errors) { | |
2802 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2803 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2804 | "packets, last error was: %s", | |
2805 | bundle->name, n_errors, n_packets, strerror(error)); | |
2806 | } else { | |
2807 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2808 | bundle->name, n_packets); | |
2809 | } | |
2810 | } | |
2811 | ||
2812 | static void | |
2813 | bundle_run(struct ofbundle *bundle) | |
2814 | { | |
2815 | if (bundle->lacp) { | |
2816 | lacp_run(bundle->lacp, send_pdu_cb); | |
2817 | } | |
2818 | if (bundle->bond) { | |
2819 | struct ofport_dpif *port; | |
2820 | ||
2821 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2822 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2823 | } |
2824 | ||
2cc3c58e | 2825 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, |
bdebeece | 2826 | lacp_status(bundle->lacp)); |
abe529af BP |
2827 | if (bond_should_send_learning_packets(bundle->bond)) { |
2828 | bundle_send_learning_packets(bundle); | |
2829 | } | |
2830 | } | |
2831 | } | |
2832 | ||
2833 | static void | |
2834 | bundle_wait(struct ofbundle *bundle) | |
2835 | { | |
2836 | if (bundle->lacp) { | |
2837 | lacp_wait(bundle->lacp); | |
2838 | } | |
2839 | if (bundle->bond) { | |
2840 | bond_wait(bundle->bond); | |
2841 | } | |
2842 | } | |
2843 | \f | |
2844 | /* Mirrors. */ | |
2845 | ||
2846 | static int | |
2847 | mirror_scan(struct ofproto_dpif *ofproto) | |
2848 | { | |
2849 | int idx; | |
2850 | ||
2851 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2852 | if (!ofproto->mirrors[idx]) { | |
2853 | return idx; | |
2854 | } | |
2855 | } | |
2856 | return -1; | |
2857 | } | |
2858 | ||
2859 | static struct ofmirror * | |
2860 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2861 | { | |
2862 | int i; | |
2863 | ||
2864 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2865 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2866 | if (mirror && mirror->aux == aux) { | |
2867 | return mirror; | |
2868 | } | |
2869 | } | |
2870 | ||
2871 | return NULL; | |
2872 | } | |
2873 | ||
9ba15e2a BP |
2874 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2875 | static void | |
2876 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2877 | { | |
2878 | int i; | |
2879 | ||
2880 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2881 | struct ofmirror *m = ofproto->mirrors[i]; | |
2882 | ||
2883 | if (m) { | |
2884 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2885 | } | |
2886 | } | |
2887 | ||
2888 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2889 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2890 | int j; | |
2891 | ||
2892 | if (!m1) { | |
2893 | continue; | |
2894 | } | |
2895 | ||
2896 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2897 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2898 | ||
edb0540b | 2899 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2900 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2901 | m2->dup_mirrors |= m1->dup_mirrors; | |
2902 | } | |
2903 | } | |
2904 | } | |
2905 | } | |
2906 | ||
abe529af BP |
2907 | static int |
2908 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2909 | const struct ofproto_mirror_settings *s) | |
2910 | { | |
2911 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2912 | mirror_mask_t mirror_bit; | |
2913 | struct ofbundle *bundle; | |
2914 | struct ofmirror *mirror; | |
2915 | struct ofbundle *out; | |
2916 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2917 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2918 | int out_vlan; | |
2919 | ||
2920 | mirror = mirror_lookup(ofproto, aux); | |
2921 | if (!s) { | |
2922 | mirror_destroy(mirror); | |
2923 | return 0; | |
2924 | } | |
2925 | if (!mirror) { | |
2926 | int idx; | |
2927 | ||
2928 | idx = mirror_scan(ofproto); | |
2929 | if (idx < 0) { | |
2930 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2931 | "cannot create %s", | |
2932 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2933 | return EFBIG; | |
2934 | } | |
2935 | ||
2936 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2937 | mirror->ofproto = ofproto; | |
2938 | mirror->idx = idx; | |
8b28d864 | 2939 | mirror->aux = aux; |
abe529af BP |
2940 | mirror->out_vlan = -1; |
2941 | mirror->name = NULL; | |
2942 | } | |
2943 | ||
2944 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2945 | free(mirror->name); | |
2946 | mirror->name = xstrdup(s->name); | |
2947 | } | |
2948 | ||
2949 | /* Get the new configuration. */ | |
2950 | if (s->out_bundle) { | |
2951 | out = bundle_lookup(ofproto, s->out_bundle); | |
2952 | if (!out) { | |
2953 | mirror_destroy(mirror); | |
2954 | return EINVAL; | |
2955 | } | |
2956 | out_vlan = -1; | |
2957 | } else { | |
2958 | out = NULL; | |
2959 | out_vlan = s->out_vlan; | |
2960 | } | |
2961 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2962 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2963 | ||
2964 | /* If the configuration has not changed, do nothing. */ | |
2965 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2966 | && hmapx_equals(&dsts, &mirror->dsts) | |
2967 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2968 | && mirror->out == out | |
2969 | && mirror->out_vlan == out_vlan) | |
2970 | { | |
2971 | hmapx_destroy(&srcs); | |
2972 | hmapx_destroy(&dsts); | |
2973 | return 0; | |
2974 | } | |
2975 | ||
2976 | hmapx_swap(&srcs, &mirror->srcs); | |
2977 | hmapx_destroy(&srcs); | |
2978 | ||
2979 | hmapx_swap(&dsts, &mirror->dsts); | |
2980 | hmapx_destroy(&dsts); | |
2981 | ||
2982 | free(mirror->vlans); | |
2983 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2984 | ||
2985 | mirror->out = out; | |
2986 | mirror->out_vlan = out_vlan; | |
2987 | ||
2988 | /* Update bundles. */ | |
2989 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2990 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2991 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2992 | bundle->src_mirrors |= mirror_bit; | |
2993 | } else { | |
2994 | bundle->src_mirrors &= ~mirror_bit; | |
2995 | } | |
2996 | ||
2997 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2998 | bundle->dst_mirrors |= mirror_bit; | |
2999 | } else { | |
3000 | bundle->dst_mirrors &= ~mirror_bit; | |
3001 | } | |
3002 | ||
3003 | if (mirror->out == bundle) { | |
3004 | bundle->mirror_out |= mirror_bit; | |
3005 | } else { | |
3006 | bundle->mirror_out &= ~mirror_bit; | |
3007 | } | |
3008 | } | |
3009 | ||
2cc3c58e | 3010 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 3011 | ofproto->has_mirrors = true; |
2cc3c58e EJ |
3012 | mac_learning_flush(ofproto->ml, |
3013 | &ofproto->backer->revalidate_set); | |
9ba15e2a | 3014 | mirror_update_dups(ofproto); |
abe529af BP |
3015 | |
3016 | return 0; | |
3017 | } | |
3018 | ||
3019 | static void | |
3020 | mirror_destroy(struct ofmirror *mirror) | |
3021 | { | |
3022 | struct ofproto_dpif *ofproto; | |
3023 | mirror_mask_t mirror_bit; | |
3024 | struct ofbundle *bundle; | |
ccb7c863 | 3025 | int i; |
abe529af BP |
3026 | |
3027 | if (!mirror) { | |
3028 | return; | |
3029 | } | |
3030 | ||
3031 | ofproto = mirror->ofproto; | |
2cc3c58e EJ |
3032 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3033 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
abe529af BP |
3034 | |
3035 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
3036 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3037 | bundle->src_mirrors &= ~mirror_bit; | |
3038 | bundle->dst_mirrors &= ~mirror_bit; | |
3039 | bundle->mirror_out &= ~mirror_bit; | |
3040 | } | |
3041 | ||
3042 | hmapx_destroy(&mirror->srcs); | |
3043 | hmapx_destroy(&mirror->dsts); | |
3044 | free(mirror->vlans); | |
3045 | ||
3046 | ofproto->mirrors[mirror->idx] = NULL; | |
3047 | free(mirror->name); | |
3048 | free(mirror); | |
9ba15e2a BP |
3049 | |
3050 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
3051 | |
3052 | ofproto->has_mirrors = false; | |
3053 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3054 | if (ofproto->mirrors[i]) { | |
3055 | ofproto->has_mirrors = true; | |
3056 | break; | |
3057 | } | |
3058 | } | |
abe529af BP |
3059 | } |
3060 | ||
9d24de3b JP |
3061 | static int |
3062 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
3063 | uint64_t *packets, uint64_t *bytes) | |
3064 | { | |
3065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3066 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
3067 | ||
3068 | if (!mirror) { | |
3069 | *packets = *bytes = UINT64_MAX; | |
3070 | return 0; | |
3071 | } | |
3072 | ||
8844e035 EJ |
3073 | push_all_stats(); |
3074 | ||
9d24de3b JP |
3075 | *packets = mirror->packet_count; |
3076 | *bytes = mirror->byte_count; | |
3077 | ||
3078 | return 0; | |
3079 | } | |
3080 | ||
abe529af BP |
3081 | static int |
3082 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
3083 | { | |
3084 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3085 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2cc3c58e | 3086 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af BP |
3087 | } |
3088 | return 0; | |
3089 | } | |
3090 | ||
3091 | static bool | |
b4affc74 | 3092 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
3093 | { |
3094 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3095 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
3096 | return bundle && bundle->mirror_out != 0; | |
3097 | } | |
8402c74b SS |
3098 | |
3099 | static void | |
b53055f4 | 3100 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
3101 | { |
3102 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 3103 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 3104 | } |
e764773c BP |
3105 | |
3106 | static void | |
c4069512 BP |
3107 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
3108 | size_t max_entries) | |
e764773c BP |
3109 | { |
3110 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3111 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
c4069512 | 3112 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
e764773c | 3113 | } |
abe529af BP |
3114 | \f |
3115 | /* Ports. */ | |
3116 | ||
3117 | static struct ofport_dpif * | |
4acbc98d | 3118 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 3119 | { |
7df6a8bd BP |
3120 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
3121 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
3122 | } |
3123 | ||
3124 | static struct ofport_dpif * | |
4acbc98d | 3125 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af | 3126 | { |
7c33b188 JR |
3127 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
3128 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
3129 | } |
3130 | ||
3131 | static void | |
e1b1d06a JP |
3132 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
3133 | struct ofproto_port *ofproto_port, | |
abe529af BP |
3134 | struct dpif_port *dpif_port) |
3135 | { | |
3136 | ofproto_port->name = dpif_port->name; | |
3137 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 3138 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
3139 | } |
3140 | ||
0a740f48 EJ |
3141 | static struct ofport_dpif * |
3142 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
3143 | { | |
3144 | const struct ofproto_dpif *ofproto; | |
3145 | const char *peer; | |
3146 | ||
3147 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
3148 | if (!peer) { | |
3149 | return NULL; | |
3150 | } | |
3151 | ||
3152 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3153 | struct ofport *ofport; | |
3154 | ||
3155 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
3156 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
3157 | return ofport_dpif_cast(ofport); | |
3158 | } | |
3159 | } | |
3160 | return NULL; | |
3161 | } | |
3162 | ||
0aa66d6e EJ |
3163 | static void |
3164 | port_run_fast(struct ofport_dpif *ofport) | |
3165 | { | |
3166 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
3167 | struct ofpbuf packet; | |
3168 | ||
3169 | ofpbuf_init(&packet, 0); | |
3170 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
3171 | send_packet(ofport, &packet); | |
3172 | ofpbuf_uninit(&packet); | |
3173 | } | |
ccc09689 EJ |
3174 | |
3175 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
3176 | struct ofpbuf packet; | |
3177 | ||
3178 | ofpbuf_init(&packet, 0); | |
3179 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
3180 | send_packet(ofport, &packet); | |
3181 | ofpbuf_uninit(&packet); | |
3182 | } | |
0aa66d6e EJ |
3183 | } |
3184 | ||
abe529af BP |
3185 | static void |
3186 | port_run(struct ofport_dpif *ofport) | |
3187 | { | |
3e5b3fdb EJ |
3188 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
3189 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
3190 | bool enable = netdev_get_carrier(ofport->up.netdev); |
3191 | ||
3e5b3fdb EJ |
3192 | ofport->carrier_seq = carrier_seq; |
3193 | ||
0aa66d6e | 3194 | port_run_fast(ofport); |
b9ad7294 EJ |
3195 | |
3196 | if (ofport->tnl_port | |
3197 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
3198 | &ofport->tnl_port)) { | |
3199 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
3200 | } | |
3201 | ||
abe529af | 3202 | if (ofport->cfm) { |
4653c558 EJ |
3203 | int cfm_opup = cfm_get_opup(ofport->cfm); |
3204 | ||
abe529af | 3205 | cfm_run(ofport->cfm); |
4653c558 EJ |
3206 | enable = enable && !cfm_get_fault(ofport->cfm); |
3207 | ||
3208 | if (cfm_opup >= 0) { | |
3209 | enable = enable && cfm_opup; | |
3210 | } | |
abe529af | 3211 | } |
015e08bc | 3212 | |
ccc09689 EJ |
3213 | if (ofport->bfd) { |
3214 | bfd_run(ofport->bfd); | |
3215 | enable = enable && bfd_forwarding(ofport->bfd); | |
3216 | } | |
3217 | ||
015e08bc EJ |
3218 | if (ofport->bundle) { |
3219 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
3220 | if (carrier_changed) { |
3221 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3222 | } | |
015e08bc EJ |
3223 | } |
3224 | ||
daff3353 EJ |
3225 | if (ofport->may_enable != enable) { |
3226 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3227 | ||
3228 | if (ofproto->has_bundle_action) { | |
2cc3c58e | 3229 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3230 | } |
3231 | } | |
3232 | ||
015e08bc | 3233 | ofport->may_enable = enable; |
abe529af BP |
3234 | } |
3235 | ||
3236 | static void | |
3237 | port_wait(struct ofport_dpif *ofport) | |
3238 | { | |
3239 | if (ofport->cfm) { | |
3240 | cfm_wait(ofport->cfm); | |
3241 | } | |
ccc09689 EJ |
3242 | |
3243 | if (ofport->bfd) { | |
3244 | bfd_wait(ofport->bfd); | |
3245 | } | |
abe529af BP |
3246 | } |
3247 | ||
3248 | static int | |
3249 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3250 | struct ofproto_port *ofproto_port) | |
3251 | { | |
3252 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3253 | struct dpif_port dpif_port; | |
3254 | int error; | |
3255 | ||
0a740f48 EJ |
3256 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3257 | const char *type = netdev_get_type_from_name(devname); | |
3258 | ||
3259 | /* We may be called before ofproto->up.port_by_name is populated with | |
3260 | * the appropriate ofport. For this reason, we must get the name and | |
3261 | * type from the netdev layer directly. */ | |
3262 | if (type) { | |
3263 | const struct ofport *ofport; | |
3264 | ||
3265 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3266 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3267 | ofproto_port->name = xstrdup(devname); | |
3268 | ofproto_port->type = xstrdup(type); | |
3269 | return 0; | |
3270 | } | |
3271 | return ENODEV; | |
3272 | } | |
3273 | ||
acf60855 JP |
3274 | if (!sset_contains(&ofproto->ports, devname)) { |
3275 | return ENODEV; | |
3276 | } | |
3277 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3278 | devname, &dpif_port); | |
abe529af | 3279 | if (!error) { |
e1b1d06a | 3280 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3281 | } |
3282 | return error; | |
3283 | } | |
3284 | ||
3285 | static int | |
e1b1d06a | 3286 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3287 | { |
3288 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 EJ |
3289 | const char *dp_port_name = netdev_vport_get_dpif_port(netdev); |
3290 | const char *devname = netdev_get_name(netdev); | |
abe529af | 3291 | |
0a740f48 EJ |
3292 | if (netdev_vport_is_patch(netdev)) { |
3293 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3294 | return 0; | |
3295 | } | |
3296 | ||
b9ad7294 | 3297 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
7d82ab2e KM |
3298 | uint32_t port_no = UINT32_MAX; |
3299 | int error; | |
3300 | ||
3301 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3302 | if (error) { |
3303 | return error; | |
3304 | } | |
7d82ab2e KM |
3305 | if (netdev_get_tunnel_config(netdev)) { |
3306 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3307 | } | |
acf60855 | 3308 | } |
b9ad7294 EJ |
3309 | |
3310 | if (netdev_get_tunnel_config(netdev)) { | |
3311 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3312 | } else { |
3313 | sset_add(&ofproto->ports, devname); | |
3314 | } | |
3315 | return 0; | |
3316 | } | |
3317 | ||
abe529af BP |
3318 | static int |
3319 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3320 | { | |
3321 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3322 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3323 | int error = 0; |
abe529af | 3324 | |
b9ad7294 EJ |
3325 | if (!ofport) { |
3326 | return 0; | |
e1b1d06a | 3327 | } |
b9ad7294 EJ |
3328 | |
3329 | sset_find_and_delete(&ofproto->ghost_ports, | |
3330 | netdev_get_name(ofport->up.netdev)); | |
a614d823 KM |
3331 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3332 | if (!ofport->tnl_port) { | |
b9ad7294 EJ |
3333 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3334 | if (!error) { | |
abe529af BP |
3335 | /* The caller is going to close ofport->up.netdev. If this is a |
3336 | * bonded port, then the bond is using that netdev, so remove it | |
3337 | * from the bond. The client will need to reconfigure everything | |
3338 | * after deleting ports, so then the slave will get re-added. */ | |
3339 | bundle_remove(&ofport->up); | |
3340 | } | |
3341 | } | |
3342 | return error; | |
3343 | } | |
3344 | ||
6527c598 PS |
3345 | static int |
3346 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3347 | { | |
3348 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3349 | int error; | |
3350 | ||
8844e035 EJ |
3351 | push_all_stats(); |
3352 | ||
6527c598 PS |
3353 | error = netdev_get_stats(ofport->up.netdev, stats); |
3354 | ||
ee382d89 | 3355 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3356 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3357 | ||
3358 | /* ofproto->stats.tx_packets represents packets that we created | |
3359 | * internally and sent to some port (e.g. packets sent with | |
3360 | * send_packet()). Account for them as if they had come from | |
3361 | * OFPP_LOCAL and got forwarded. */ | |
3362 | ||
3363 | if (stats->rx_packets != UINT64_MAX) { | |
3364 | stats->rx_packets += ofproto->stats.tx_packets; | |
3365 | } | |
3366 | ||
3367 | if (stats->rx_bytes != UINT64_MAX) { | |
3368 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3369 | } | |
3370 | ||
3371 | /* ofproto->stats.rx_packets represents packets that were received on | |
3372 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3373 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3374 | |
3375 | if (stats->tx_packets != UINT64_MAX) { | |
3376 | stats->tx_packets += ofproto->stats.rx_packets; | |
3377 | } | |
3378 | ||
3379 | if (stats->tx_bytes != UINT64_MAX) { | |
3380 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3381 | } | |
3382 | } | |
3383 | ||
3384 | return error; | |
3385 | } | |
3386 | ||
abe529af | 3387 | struct port_dump_state { |
acf60855 JP |
3388 | uint32_t bucket; |
3389 | uint32_t offset; | |
0a740f48 | 3390 | bool ghost; |
da78d43d BP |
3391 | |
3392 | struct ofproto_port port; | |
3393 | bool has_port; | |
abe529af BP |
3394 | }; |
3395 | ||
3396 | static int | |
acf60855 | 3397 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3398 | { |
0a740f48 | 3399 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3400 | return 0; |
3401 | } | |
3402 | ||
3403 | static int | |
b9ad7294 | 3404 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3405 | struct ofproto_port *port) |
3406 | { | |
e1b1d06a | 3407 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3408 | struct port_dump_state *state = state_; |
0a740f48 | 3409 | const struct sset *sset; |
acf60855 | 3410 | struct sset_node *node; |
abe529af | 3411 | |
da78d43d BP |
3412 | if (state->has_port) { |
3413 | ofproto_port_destroy(&state->port); | |
3414 | state->has_port = false; | |
3415 | } | |
0a740f48 EJ |
3416 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3417 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3418 | int error; |
3419 | ||
da78d43d BP |
3420 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3421 | if (!error) { | |
3422 | *port = state->port; | |
3423 | state->has_port = true; | |
3424 | return 0; | |
3425 | } else if (error != ENODEV) { | |
acf60855 JP |
3426 | return error; |
3427 | } | |
abe529af | 3428 | } |
acf60855 | 3429 | |
0a740f48 EJ |
3430 | if (!state->ghost) { |
3431 | state->ghost = true; | |
3432 | state->bucket = 0; | |
3433 | state->offset = 0; | |
3434 | return port_dump_next(ofproto_, state_, port); | |
3435 | } | |
3436 | ||
acf60855 | 3437 | return EOF; |
abe529af BP |
3438 | } |
3439 | ||
3440 | static int | |
3441 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3442 | { | |
3443 | struct port_dump_state *state = state_; | |
3444 | ||
da78d43d BP |
3445 | if (state->has_port) { |
3446 | ofproto_port_destroy(&state->port); | |
3447 | } | |
abe529af BP |
3448 | free(state); |
3449 | return 0; | |
3450 | } | |
3451 | ||
3452 | static int | |
3453 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3454 | { | |
3455 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3456 | |
3457 | if (ofproto->port_poll_errno) { | |
3458 | int error = ofproto->port_poll_errno; | |
3459 | ofproto->port_poll_errno = 0; | |
3460 | return error; | |
3461 | } | |
3462 | ||
3463 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3464 | return EAGAIN; | |
3465 | } | |
3466 | ||
3467 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3468 | return 0; | |
abe529af BP |
3469 | } |
3470 | ||
3471 | static void | |
3472 | port_poll_wait(const struct ofproto *ofproto_) | |
3473 | { | |
3474 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3475 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3476 | } |
3477 | ||
3478 | static int | |
3479 | port_is_lacp_current(const struct ofport *ofport_) | |
3480 | { | |
3481 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3482 | return (ofport->bundle && ofport->bundle->lacp | |
3483 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3484 | : -1); | |
3485 | } | |
3486 | \f | |
3487 | /* Upcall handling. */ | |
3488 | ||
501f8d1f BP |
3489 | /* Flow miss batching. |
3490 | * | |
3491 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3492 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3493 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3494 | * more packets, plus possibly installing the flow in the dpif. | |
3495 | * | |
3496 | * So far we only batch the operations that affect flow setup time the most. | |
3497 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3498 | struct flow_miss { | |
3499 | struct hmap_node hmap_node; | |
acf60855 | 3500 | struct ofproto_dpif *ofproto; |
501f8d1f | 3501 | struct flow flow; |
b0f7b9b5 | 3502 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3503 | const struct nlattr *key; |
3504 | size_t key_len; | |
14f94f9a | 3505 | struct initial_vals initial_vals; |
501f8d1f | 3506 | struct list packets; |
6a7e895f | 3507 | enum dpif_upcall_type upcall_type; |
a088a1ff | 3508 | uint32_t odp_in_port; |
501f8d1f BP |
3509 | }; |
3510 | ||
3511 | struct flow_miss_op { | |
c2b565b5 | 3512 | struct dpif_op dpif_op; |
5fe20d5d BP |
3513 | void *garbage; /* Pointer to pass to free(), NULL if none. */ |
3514 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
3515 | }; |
3516 | ||
62cd7072 BP |
3517 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3518 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3519 | * configurations. */ |
62cd7072 | 3520 | static void |
a39edbd4 | 3521 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3522 | const struct flow *flow) |
62cd7072 BP |
3523 | { |
3524 | struct ofputil_packet_in pin; | |
3525 | ||
3e3252fa EJ |
3526 | pin.packet = packet->data; |
3527 | pin.packet_len = packet->size; | |
62cd7072 | 3528 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3529 | pin.controller_id = 0; |
54834960 EJ |
3530 | |
3531 | pin.table_id = 0; | |
3532 | pin.cookie = 0; | |
3533 | ||
62cd7072 | 3534 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3535 | |
3536 | flow_get_metadata(flow, &pin.fmd); | |
3537 | ||
d8653c38 | 3538 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3539 | } |
3540 | ||
6a7e895f | 3541 | static enum slow_path_reason |
abe529af | 3542 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
ffaef958 | 3543 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) |
abe529af | 3544 | { |
b6e001b6 | 3545 | if (!ofport) { |
6a7e895f | 3546 | return 0; |
ffaef958 | 3547 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 3548 | if (packet) { |
abe529af BP |
3549 | cfm_process_heartbeat(ofport->cfm, packet); |
3550 | } | |
6a7e895f | 3551 | return SLOW_CFM; |
ccc09689 EJ |
3552 | } else if (ofport->bfd && bfd_should_process_flow(flow)) { |
3553 | if (packet) { | |
3554 | bfd_process_packet(ofport->bfd, flow, packet); | |
3555 | } | |
3556 | return SLOW_BFD; | |
b6e001b6 EJ |
3557 | } else if (ofport->bundle && ofport->bundle->lacp |
3558 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3559 | if (packet) { | |
3560 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 3561 | } |
6a7e895f | 3562 | return SLOW_LACP; |
21f7563c JP |
3563 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
3564 | if (packet) { | |
3565 | stp_process_packet(ofport, packet); | |
3566 | } | |
6a7e895f | 3567 | return SLOW_STP; |
ffaef958 BP |
3568 | } else { |
3569 | return 0; | |
abe529af | 3570 | } |
abe529af BP |
3571 | } |
3572 | ||
501f8d1f | 3573 | static struct flow_miss * |
ddbc5954 JP |
3574 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3575 | const struct flow *flow, uint32_t hash) | |
abe529af | 3576 | { |
501f8d1f | 3577 | struct flow_miss *miss; |
abe529af | 3578 | |
501f8d1f | 3579 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3580 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3581 | return miss; |
3582 | } | |
3583 | } | |
abe529af | 3584 | |
b23cdad9 | 3585 | return NULL; |
501f8d1f | 3586 | } |
abe529af | 3587 | |
9d6ac44e BP |
3588 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3589 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3590 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3591 | * returned op->subfacet, and if anything needs to be freed after processing | |
3592 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3593 | static void |
9d6ac44e BP |
3594 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3595 | struct flow_miss_op *op) | |
501f8d1f | 3596 | { |
14f94f9a | 3597 | if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) { |
9d6ac44e BP |
3598 | /* This packet was received on a VLAN splinter port. We |
3599 | * added a VLAN to the packet to make the packet resemble | |
3600 | * the flow, but the actions were composed assuming that | |
3601 | * the packet contained no VLAN. So, we must remove the | |
3602 | * VLAN header from the packet before trying to execute the | |
3603 | * actions. */ | |
3604 | eth_pop_vlan(packet); | |
3605 | } | |
3606 | ||
9d6ac44e BP |
3607 | op->garbage = NULL; |
3608 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
3609 | op->dpif_op.u.execute.key = miss->key; | |
3610 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3611 | op->dpif_op.u.execute.packet = packet; | |
3612 | } | |
3613 | ||
3614 | /* Helper for handle_flow_miss_without_facet() and | |
3615 | * handle_flow_miss_with_facet(). */ | |
3616 | static void | |
3617 | handle_flow_miss_common(struct rule_dpif *rule, | |
3618 | struct ofpbuf *packet, const struct flow *flow) | |
3619 | { | |
3620 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3621 | ||
3622 | ofproto->n_matches++; | |
3623 | ||
3624 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
3625 | /* | |
3626 | * Extra-special case for fail-open mode. | |
3627 | * | |
3628 | * We are in fail-open mode and the packet matched the fail-open | |
3629 | * rule, but we are connected to a controller too. We should send | |
3630 | * the packet up to the controller in the hope that it will try to | |
3631 | * set up a flow and thereby allow us to exit fail-open. | |
3632 | * | |
3633 | * See the top-level comment in fail-open.c for more information. | |
3634 | */ | |
3635 | send_packet_in_miss(ofproto, packet, flow); | |
3636 | } | |
3637 | } | |
3638 | ||
3639 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3640 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3641 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3642 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3643 | * the benefits, so when the datapath holds a large number of flows we impose | |
3644 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3645 | static bool | |
3646 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
3647 | struct flow_miss *miss, uint32_t hash) | |
3648 | { | |
3649 | if (!ofproto->governor) { | |
3650 | size_t n_subfacets; | |
3651 | ||
3652 | n_subfacets = hmap_count(&ofproto->subfacets); | |
3653 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
3654 | return true; | |
3655 | } | |
3656 | ||
3657 | ofproto->governor = governor_create(ofproto->up.name); | |
3658 | } | |
3659 | ||
3660 | return governor_should_install_flow(ofproto->governor, hash, | |
3661 | list_size(&miss->packets)); | |
3662 | } | |
3663 | ||
3664 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3665 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3666 | * increment '*n_ops'. */ | |
3667 | static void | |
3668 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
3669 | struct rule_dpif *rule, | |
3670 | struct flow_miss_op *ops, size_t *n_ops) | |
3671 | { | |
3672 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
a7752d4a | 3673 | long long int now = time_msec(); |
9d6ac44e | 3674 | struct action_xlate_ctx ctx; |
530a1d91 | 3675 | struct ofpbuf *packet; |
2b459b83 | 3676 | |
9d6ac44e BP |
3677 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
3678 | struct flow_miss_op *op = &ops[*n_ops]; | |
3679 | struct dpif_flow_stats stats; | |
3680 | struct ofpbuf odp_actions; | |
abe529af | 3681 | |
9d6ac44e | 3682 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3683 | |
9d6ac44e | 3684 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 3685 | |
a7752d4a | 3686 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 3687 | rule_credit_stats(rule, &stats); |
abe529af | 3688 | |
14f94f9a JP |
3689 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, |
3690 | &miss->initial_vals, rule, 0, packet); | |
9d6ac44e | 3691 | ctx.resubmit_stats = &stats; |
f25d0cf3 | 3692 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
9d6ac44e | 3693 | &odp_actions); |
abe529af | 3694 | |
9d6ac44e BP |
3695 | if (odp_actions.size) { |
3696 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
3697 | ||
3698 | init_flow_miss_execute_op(miss, packet, op); | |
3699 | execute->actions = odp_actions.data; | |
3700 | execute->actions_len = odp_actions.size; | |
3701 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3702 | ||
3703 | (*n_ops)++; | |
3704 | } else { | |
3705 | ofpbuf_uninit(&odp_actions); | |
3706 | } | |
abe529af | 3707 | } |
9d6ac44e BP |
3708 | } |
3709 | ||
3710 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3711 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3712 | * | |
3713 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3714 | * This is really important only for new facets: if we just called time_msec() | |
3715 | * here, then the new subfacet or its packets could look (occasionally) as | |
3716 | * though it was used some time after the facet was used. That can make a | |
3717 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3718 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3719 | static void |
3720 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 3721 | long long int now, |
9d6ac44e BP |
3722 | struct flow_miss_op *ops, size_t *n_ops) |
3723 | { | |
6a7e895f BP |
3724 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3725 | enum subfacet_path want_path; | |
9d6ac44e BP |
3726 | struct subfacet *subfacet; |
3727 | struct ofpbuf *packet; | |
abe529af | 3728 | |
a088a1ff | 3729 | subfacet = subfacet_create(facet, miss, now); |
b0f7b9b5 | 3730 | |
530a1d91 | 3731 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3732 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 3733 | struct dpif_flow_stats stats; |
5fe20d5d | 3734 | struct ofpbuf odp_actions; |
67d91f78 | 3735 | |
9d6ac44e | 3736 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 3737 | |
5fe20d5d | 3738 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
6a7e895f | 3739 | if (!subfacet->actions || subfacet->slow) { |
5fe20d5d | 3740 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 3741 | } |
67d91f78 | 3742 | |
459b16a1 | 3743 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 3744 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 3745 | |
9d6ac44e BP |
3746 | if (subfacet->actions_len) { |
3747 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 3748 | |
9d6ac44e | 3749 | init_flow_miss_execute_op(miss, packet, op); |
6a7e895f | 3750 | if (!subfacet->slow) { |
9d6ac44e BP |
3751 | execute->actions = subfacet->actions; |
3752 | execute->actions_len = subfacet->actions_len; | |
3753 | ofpbuf_uninit(&odp_actions); | |
3754 | } else { | |
3755 | execute->actions = odp_actions.data; | |
3756 | execute->actions_len = odp_actions.size; | |
3757 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3758 | } | |
999fba59 | 3759 | |
9d6ac44e | 3760 | (*n_ops)++; |
5fe20d5d | 3761 | } else { |
9d6ac44e | 3762 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 3763 | } |
501f8d1f BP |
3764 | } |
3765 | ||
6a7e895f BP |
3766 | want_path = subfacet_want_path(subfacet->slow); |
3767 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
501f8d1f | 3768 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3769 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3770 | |
c84451a6 EJ |
3771 | subfacet->path = want_path; |
3772 | ||
5fe20d5d | 3773 | op->garbage = NULL; |
c2b565b5 | 3774 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
3775 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
3776 | put->key = miss->key; | |
3777 | put->key_len = miss->key_len; | |
6a7e895f BP |
3778 | if (want_path == SF_FAST_PATH) { |
3779 | put->actions = subfacet->actions; | |
3780 | put->actions_len = subfacet->actions_len; | |
3781 | } else { | |
3782 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
3783 | op->stub, sizeof op->stub, | |
3784 | &put->actions, &put->actions_len); | |
3785 | } | |
501f8d1f BP |
3786 | put->stats = NULL; |
3787 | } | |
3788 | } | |
3789 | ||
acf60855 JP |
3790 | /* Handles flow miss 'miss'. May add any required datapath operations |
3791 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3792 | static void |
acf60855 JP |
3793 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3794 | size_t *n_ops) | |
9d6ac44e | 3795 | { |
acf60855 | 3796 | struct ofproto_dpif *ofproto = miss->ofproto; |
9d6ac44e | 3797 | struct facet *facet; |
459b16a1 | 3798 | long long int now; |
9d6ac44e BP |
3799 | uint32_t hash; |
3800 | ||
3801 | /* The caller must ensure that miss->hmap_node.hash contains | |
3802 | * flow_hash(miss->flow, 0). */ | |
3803 | hash = miss->hmap_node.hash; | |
3804 | ||
3805 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3806 | if (!facet) { | |
c57b2226 BP |
3807 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); |
3808 | ||
3809 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
9d6ac44e BP |
3810 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); |
3811 | return; | |
3812 | } | |
3813 | ||
3814 | facet = facet_create(rule, &miss->flow, hash); | |
459b16a1 BP |
3815 | now = facet->used; |
3816 | } else { | |
3817 | now = time_msec(); | |
9d6ac44e | 3818 | } |
459b16a1 | 3819 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3820 | } |
3821 | ||
8f73d537 EJ |
3822 | static struct drop_key * |
3823 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3824 | size_t key_len) | |
3825 | { | |
3826 | struct drop_key *drop_key; | |
3827 | ||
3828 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3829 | &backer->drop_keys) { | |
3830 | if (drop_key->key_len == key_len | |
3831 | && !memcmp(drop_key->key, key, key_len)) { | |
3832 | return drop_key; | |
3833 | } | |
3834 | } | |
3835 | return NULL; | |
3836 | } | |
3837 | ||
3838 | static void | |
3839 | drop_key_clear(struct dpif_backer *backer) | |
3840 | { | |
3841 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3842 | struct drop_key *drop_key, *next; | |
3843 | ||
3844 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3845 | int error; | |
3846 | ||
3847 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3848 | NULL); | |
3849 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3850 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3851 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3852 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3853 | ds_cstr(&ds)); | |
3854 | ds_destroy(&ds); | |
3855 | } | |
3856 | ||
3857 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3858 | free(drop_key->key); | |
3859 | free(drop_key); | |
3860 | } | |
3861 | } | |
3862 | ||
e09ee259 EJ |
3863 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3864 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3865 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3866 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3867 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3868 | * 'packet' ingressed. | |
e2a6ca36 | 3869 | * |
e09ee259 EJ |
3870 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3871 | * 'flow''s in_port to OFPP_NONE. | |
3872 | * | |
3873 | * This function does post-processing on data returned from | |
3874 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3875 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3876 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3877 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3878 | * a VLAN header onto 'packet' (if it is nonnull). | |
3879 | * | |
c3f6c502 JP |
3880 | * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' |
3881 | * to the VLAN TCI with which the packet was really received, that is, the | |
3882 | * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from | |
3883 | * the value returned in flow->vlan_tci only for packets received on | |
715b48aa | 3884 | * VLAN splinters.) |
e09ee259 | 3885 | * |
b9ad7294 EJ |
3886 | * Similarly, this function also includes some logic to help with tunnels. It |
3887 | * may modify 'flow' as necessary to make the tunneling implementation | |
3888 | * transparent to the upcall processing logic. | |
3889 | * | |
e09ee259 EJ |
3890 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3891 | * or some other positive errno if there are other problems. */ | |
3892 | static int | |
3893 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3894 | const struct nlattr *key, size_t key_len, | |
3895 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3896 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
14f94f9a | 3897 | struct initial_vals *initial_vals) |
e84173dc | 3898 | { |
e09ee259 EJ |
3899 | const struct ofport_dpif *port; |
3900 | enum odp_key_fitness fitness; | |
b9ad7294 | 3901 | int error = ENODEV; |
e09ee259 EJ |
3902 | |
3903 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3904 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3905 | error = EINVAL; |
3906 | goto exit; | |
3907 | } | |
3908 | ||
14f94f9a JP |
3909 | if (initial_vals) { |
3910 | initial_vals->vlan_tci = flow->vlan_tci; | |
e84173dc | 3911 | } |
e84173dc | 3912 | |
e09ee259 EJ |
3913 | if (odp_in_port) { |
3914 | *odp_in_port = flow->in_port; | |
3915 | } | |
3916 | ||
3abe1835 BP |
3917 | port = (tnl_port_should_receive(flow) |
3918 | ? ofport_dpif_cast(tnl_port_receive(flow)) | |
3919 | : odp_port_to_ofport(backer, flow->in_port)); | |
3920 | flow->in_port = port ? port->up.ofp_port : OFPP_NONE; | |
3921 | if (!port) { | |
3922 | goto exit; | |
3923 | } | |
3924 | ||
3925 | /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port | |
3926 | * it's theoretically possible that we'll receive an ofport belonging to an | |
3927 | * entirely different datapath. In practice, this can't happen because no | |
3928 | * platforms has two separate datapaths which each support tunneling. */ | |
3929 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3930 | ||
3931 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3932 | if (packet) { | |
3933 | /* Make the packet resemble the flow, so that it gets sent to | |
3934 | * an OpenFlow controller properly, so that it looks correct | |
3935 | * for sFlow, and so that flow_extract() will get the correct | |
3936 | * vlan_tci if it is called on 'packet'. | |
3937 | * | |
3938 | * The allocated space inside 'packet' probably also contains | |
3939 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3940 | * a struct dpif_upcall (see the large comment on that | |
3941 | * structure definition), so pushing data on 'packet' is in | |
3942 | * general not a good idea since it could overwrite 'key' or | |
3943 | * free it as a side effect. However, it's OK in this special | |
3944 | * case because we know that 'packet' is inside a Netlink | |
3945 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3946 | * "struct nlattr", which is fine since we don't need that | |
3947 | * header anymore. */ | |
3948 | eth_push_vlan(packet, flow->vlan_tci); | |
3949 | } | |
3950 | /* We can't reproduce 'key' from 'flow'. */ | |
3951 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 | 3952 | } |
e09ee259 | 3953 | error = 0; |
52a90c29 | 3954 | |
b9ad7294 EJ |
3955 | if (ofproto) { |
3956 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3957 | } | |
3958 | ||
e09ee259 EJ |
3959 | exit: |
3960 | if (fitnessp) { | |
3961 | *fitnessp = fitness; | |
3962 | } | |
3963 | return error; | |
e84173dc BP |
3964 | } |
3965 | ||
501f8d1f | 3966 | static void |
acf60855 | 3967 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3968 | size_t n_upcalls) |
3969 | { | |
3970 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3971 | struct flow_miss *miss; |
3972 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3973 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3974 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3975 | struct hmap todo; |
b23cdad9 | 3976 | int n_misses; |
501f8d1f BP |
3977 | size_t n_ops; |
3978 | size_t i; | |
3979 | ||
3980 | if (!n_upcalls) { | |
3981 | return; | |
3982 | } | |
3983 | ||
3984 | /* Construct the to-do list. | |
3985 | * | |
3986 | * This just amounts to extracting the flow from each packet and sticking | |
3987 | * the packets that have the same flow in the same "flow_miss" structure so | |
3988 | * that we can process them together. */ | |
3989 | hmap_init(&todo); | |
b23cdad9 | 3990 | n_misses = 0; |
501f8d1f | 3991 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3992 | struct flow_miss *miss = &misses[n_misses]; |
3993 | struct flow_miss *existing_miss; | |
acf60855 | 3994 | struct ofproto_dpif *ofproto; |
a088a1ff | 3995 | uint32_t odp_in_port; |
1d446463 | 3996 | struct flow flow; |
b23cdad9 | 3997 | uint32_t hash; |
e09ee259 | 3998 | int error; |
501f8d1f | 3999 | |
e09ee259 EJ |
4000 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
4001 | upcall->key_len, &flow, &miss->key_fitness, | |
14f94f9a | 4002 | &ofproto, &odp_in_port, &miss->initial_vals); |
e09ee259 | 4003 | if (error == ENODEV) { |
8f73d537 EJ |
4004 | struct drop_key *drop_key; |
4005 | ||
acf60855 JP |
4006 | /* Received packet on port for which we couldn't associate |
4007 | * an ofproto. This can happen if a port is removed while | |
4008 | * traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
4009 | * in case it happens frequently. Install a drop flow so |
4010 | * that future packets of the flow are inexpensively dropped | |
4011 | * in the kernel. */ | |
acf60855 JP |
4012 | VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, |
4013 | flow.in_port); | |
8f73d537 EJ |
4014 | |
4015 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
4016 | if (!drop_key) { | |
4017 | drop_key = xmalloc(sizeof *drop_key); | |
4018 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
4019 | drop_key->key_len = upcall->key_len; | |
4020 | ||
4021 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
4022 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
4023 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
4024 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
4025 | } | |
4026 | continue; | |
acf60855 | 4027 | } |
e09ee259 | 4028 | if (error) { |
b0f7b9b5 BP |
4029 | continue; |
4030 | } | |
735d7efb AZ |
4031 | |
4032 | ofproto->n_missed++; | |
72e8bf28 | 4033 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
1d446463 | 4034 | &flow.tunnel, flow.in_port, &miss->flow); |
501f8d1f | 4035 | |
501f8d1f | 4036 | /* Add other packets to a to-do list. */ |
b23cdad9 | 4037 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 4038 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
4039 | if (!existing_miss) { |
4040 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 4041 | miss->ofproto = ofproto; |
b23cdad9 BP |
4042 | miss->key = upcall->key; |
4043 | miss->key_len = upcall->key_len; | |
6a7e895f | 4044 | miss->upcall_type = upcall->type; |
a088a1ff | 4045 | miss->odp_in_port = odp_in_port; |
b23cdad9 BP |
4046 | list_init(&miss->packets); |
4047 | ||
4048 | n_misses++; | |
4049 | } else { | |
4050 | miss = existing_miss; | |
4051 | } | |
501f8d1f BP |
4052 | list_push_back(&miss->packets, &upcall->packet->list_node); |
4053 | } | |
4054 | ||
4055 | /* Process each element in the to-do list, constructing the set of | |
4056 | * operations to batch. */ | |
4057 | n_ops = 0; | |
33bb0caa | 4058 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 4059 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 4060 | } |
cb22974d | 4061 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
4062 | |
4063 | /* Execute batch. */ | |
4064 | for (i = 0; i < n_ops; i++) { | |
4065 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
4066 | } | |
acf60855 | 4067 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 4068 | |
c84451a6 | 4069 | /* Free memory. */ |
501f8d1f | 4070 | for (i = 0; i < n_ops; i++) { |
c84451a6 | 4071 | free(flow_miss_ops[i].garbage); |
501f8d1f | 4072 | } |
33bb0caa | 4073 | hmap_destroy(&todo); |
abe529af BP |
4074 | } |
4075 | ||
29089a54 RL |
4076 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, |
4077 | IPFIX_UPCALL } | |
6a7e895f BP |
4078 | classify_upcall(const struct dpif_upcall *upcall) |
4079 | { | |
29089a54 | 4080 | size_t userdata_len; |
6a7e895f BP |
4081 | union user_action_cookie cookie; |
4082 | ||
4083 | /* First look at the upcall type. */ | |
4084 | switch (upcall->type) { | |
4085 | case DPIF_UC_ACTION: | |
4086 | break; | |
4087 | ||
4088 | case DPIF_UC_MISS: | |
4089 | return MISS_UPCALL; | |
4090 | ||
4091 | case DPIF_N_UC_TYPES: | |
4092 | default: | |
4093 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
4094 | return BAD_UPCALL; | |
4095 | } | |
4096 | ||
4097 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
4098 | if (!upcall->userdata) { |
4099 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
4100 | return BAD_UPCALL; | |
4101 | } | |
29089a54 RL |
4102 | userdata_len = nl_attr_get_size(upcall->userdata); |
4103 | if (userdata_len < sizeof cookie.type | |
4104 | || userdata_len > sizeof cookie) { | |
e995e3df | 4105 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", |
29089a54 | 4106 | userdata_len); |
e995e3df BP |
4107 | return BAD_UPCALL; |
4108 | } | |
29089a54 RL |
4109 | memset(&cookie, 0, sizeof cookie); |
4110 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
4111 | if (userdata_len == sizeof cookie.sflow | |
4112 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
6a7e895f | 4113 | return SFLOW_UPCALL; |
29089a54 RL |
4114 | } else if (userdata_len == sizeof cookie.slow_path |
4115 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
6a7e895f | 4116 | return MISS_UPCALL; |
29089a54 RL |
4117 | } else if (userdata_len == sizeof cookie.flow_sample |
4118 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
4119 | return FLOW_SAMPLE_UPCALL; | |
4120 | } else if (userdata_len == sizeof cookie.ipfix | |
4121 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
4122 | return IPFIX_UPCALL; | |
4123 | } else { | |
4124 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
4125 | " and size %zu", cookie.type, userdata_len); | |
6a7e895f BP |
4126 | return BAD_UPCALL; |
4127 | } | |
4128 | } | |
4129 | ||
abe529af | 4130 | static void |
acf60855 | 4131 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 4132 | const struct dpif_upcall *upcall) |
abe529af | 4133 | { |
acf60855 | 4134 | struct ofproto_dpif *ofproto; |
1673e0e4 | 4135 | union user_action_cookie cookie; |
e84173dc | 4136 | struct flow flow; |
e1b1d06a | 4137 | uint32_t odp_in_port; |
abe529af | 4138 | |
e09ee259 EJ |
4139 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
4140 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
4141 | || !ofproto->sflow) { | |
e84173dc BP |
4142 | return; |
4143 | } | |
4144 | ||
29089a54 RL |
4145 | memset(&cookie, 0, sizeof cookie); |
4146 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
e1b1d06a JP |
4147 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
4148 | odp_in_port, &cookie); | |
6ff686f2 PS |
4149 | } |
4150 | ||
29089a54 RL |
4151 | static void |
4152 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
4153 | const struct dpif_upcall *upcall) | |
4154 | { | |
4155 | struct ofproto_dpif *ofproto; | |
4156 | union user_action_cookie cookie; | |
4157 | struct flow flow; | |
4158 | ||
4159 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4160 | &flow, NULL, &ofproto, NULL, NULL) | |
4161 | || !ofproto->ipfix) { | |
4162 | return; | |
4163 | } | |
4164 | ||
4165 | memset(&cookie, 0, sizeof cookie); | |
4166 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
4167 | ||
4168 | /* The flow reflects exactly the contents of the packet. Sample | |
4169 | * the packet using it. */ | |
4170 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
4171 | cookie.flow_sample.collector_set_id, | |
4172 | cookie.flow_sample.probability, | |
4173 | cookie.flow_sample.obs_domain_id, | |
4174 | cookie.flow_sample.obs_point_id); | |
4175 | } | |
4176 | ||
4177 | static void | |
4178 | handle_ipfix_upcall(struct dpif_backer *backer, | |
4179 | const struct dpif_upcall *upcall) | |
4180 | { | |
4181 | struct ofproto_dpif *ofproto; | |
4182 | struct flow flow; | |
4183 | ||
4184 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4185 | &flow, NULL, &ofproto, NULL, NULL) | |
4186 | || !ofproto->ipfix) { | |
4187 | return; | |
4188 | } | |
4189 | ||
4190 | /* The flow reflects exactly the contents of the packet. Sample | |
4191 | * the packet using it. */ | |
4192 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
4193 | } | |
4194 | ||
9b16c439 | 4195 | static int |
acf60855 | 4196 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 4197 | { |
9b16c439 | 4198 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
4199 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
4200 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
4201 | int n_processed; | |
9b16c439 BP |
4202 | int n_misses; |
4203 | int i; | |
abe529af | 4204 | |
cb22974d | 4205 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 4206 | |
9b16c439 | 4207 | n_misses = 0; |
90a7c55e | 4208 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 4209 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 4210 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
4211 | int error; |
4212 | ||
90a7c55e BP |
4213 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
4214 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 4215 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 4216 | if (error) { |
90a7c55e | 4217 | ofpbuf_uninit(buf); |
9b16c439 BP |
4218 | break; |
4219 | } | |
4220 | ||
6a7e895f BP |
4221 | switch (classify_upcall(upcall)) { |
4222 | case MISS_UPCALL: | |
9b16c439 BP |
4223 | /* Handle it later. */ |
4224 | n_misses++; | |
4225 | break; | |
4226 | ||
6a7e895f | 4227 | case SFLOW_UPCALL: |
acf60855 | 4228 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
4229 | ofpbuf_uninit(buf); |
4230 | break; | |
4231 | ||
29089a54 RL |
4232 | case FLOW_SAMPLE_UPCALL: |
4233 | handle_flow_sample_upcall(backer, upcall); | |
4234 | ofpbuf_uninit(buf); | |
4235 | break; | |
4236 | ||
4237 | case IPFIX_UPCALL: | |
4238 | handle_ipfix_upcall(backer, upcall); | |
4239 | ofpbuf_uninit(buf); | |
4240 | break; | |
4241 | ||
6a7e895f BP |
4242 | case BAD_UPCALL: |
4243 | ofpbuf_uninit(buf); | |
9b16c439 BP |
4244 | break; |
4245 | } | |
abe529af | 4246 | } |
9b16c439 | 4247 | |
6a7e895f | 4248 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 4249 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
4250 | for (i = 0; i < n_misses; i++) { |
4251 | ofpbuf_uninit(&miss_bufs[i]); | |
4252 | } | |
9b16c439 | 4253 | |
90a7c55e | 4254 | return n_processed; |
abe529af BP |
4255 | } |
4256 | \f | |
4257 | /* Flow expiration. */ | |
4258 | ||
b0f7b9b5 | 4259 | static int subfacet_max_idle(const struct ofproto_dpif *); |
acf60855 | 4260 | static void update_stats(struct dpif_backer *); |
abe529af | 4261 | static void rule_expire(struct rule_dpif *); |
b0f7b9b5 | 4262 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
4263 | |
4264 | /* This function is called periodically by run(). Its job is to collect | |
4265 | * updates for the flows that have been installed into the datapath, most | |
4266 | * importantly when they last were used, and then use that information to | |
4267 | * expire flows that have not been used recently. | |
4268 | * | |
4269 | * Returns the number of milliseconds after which it should be called again. */ | |
4270 | static int | |
acf60855 | 4271 | expire(struct dpif_backer *backer) |
abe529af | 4272 | { |
acf60855 JP |
4273 | struct ofproto_dpif *ofproto; |
4274 | int max_idle = INT32_MAX; | |
abe529af | 4275 | |
8f73d537 EJ |
4276 | /* Periodically clear out the drop keys in an effort to keep them |
4277 | * relatively few. */ | |
4278 | drop_key_clear(backer); | |
4279 | ||
acf60855 JP |
4280 | /* Update stats for each flow in the backer. */ |
4281 | update_stats(backer); | |
abe529af | 4282 | |
acf60855 | 4283 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
e503cc19 | 4284 | struct rule *rule, *next_rule; |
acf60855 | 4285 | int dp_max_idle; |
abe529af | 4286 | |
acf60855 JP |
4287 | if (ofproto->backer != backer) { |
4288 | continue; | |
4289 | } | |
0697b5c3 | 4290 | |
655ab909 AZ |
4291 | /* Keep track of the max number of flows per ofproto_dpif. */ |
4292 | update_max_subfacet_count(ofproto); | |
4293 | ||
acf60855 JP |
4294 | /* Expire subfacets that have been idle too long. */ |
4295 | dp_max_idle = subfacet_max_idle(ofproto); | |
4296 | expire_subfacets(ofproto, dp_max_idle); | |
4297 | ||
4298 | max_idle = MIN(max_idle, dp_max_idle); | |
4299 | ||
4300 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4301 | * has passed. */ | |
e503cc19 SH |
4302 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4303 | &ofproto->up.expirable) { | |
4304 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4305 | } |
abe529af | 4306 | |
acf60855 JP |
4307 | /* All outstanding data in existing flows has been accounted, so it's a |
4308 | * good time to do bond rebalancing. */ | |
4309 | if (ofproto->has_bonded_bundles) { | |
4310 | struct ofbundle *bundle; | |
abe529af | 4311 | |
acf60855 JP |
4312 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4313 | if (bundle->bond) { | |
2cc3c58e | 4314 | bond_rebalance(bundle->bond, &backer->revalidate_set); |
acf60855 | 4315 | } |
abe529af BP |
4316 | } |
4317 | } | |
4318 | } | |
4319 | ||
acf60855 | 4320 | return MIN(max_idle, 1000); |
abe529af BP |
4321 | } |
4322 | ||
a218c879 BP |
4323 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4324 | * as 'subfacet''s statistics. */ | |
4325 | static void | |
4326 | update_subfacet_stats(struct subfacet *subfacet, | |
4327 | const struct dpif_flow_stats *stats) | |
4328 | { | |
4329 | struct facet *facet = subfacet->facet; | |
4330 | ||
4331 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
4332 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
4333 | facet->packet_count += extra; | |
4334 | } else { | |
4335 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4336 | } | |
4337 | ||
4338 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
4339 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
4340 | } else { | |
4341 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4342 | } | |
4343 | ||
4344 | subfacet->dp_packet_count = stats->n_packets; | |
4345 | subfacet->dp_byte_count = stats->n_bytes; | |
4346 | ||
4347 | facet->tcp_flags |= stats->tcp_flags; | |
4348 | ||
4349 | subfacet_update_time(subfacet, stats->used); | |
4350 | if (facet->accounted_bytes < facet->byte_count) { | |
4351 | facet_learn(facet); | |
4352 | facet_account(facet); | |
4353 | facet->accounted_bytes = facet->byte_count; | |
4354 | } | |
a218c879 BP |
4355 | } |
4356 | ||
4357 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4358 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4359 | static void | |
acf60855 | 4360 | delete_unexpected_flow(struct ofproto_dpif *ofproto, |
a218c879 BP |
4361 | const struct nlattr *key, size_t key_len) |
4362 | { | |
4363 | if (!VLOG_DROP_WARN(&rl)) { | |
4364 | struct ds s; | |
4365 | ||
4366 | ds_init(&s); | |
4367 | odp_flow_key_format(key, key_len, &s); | |
acf60855 | 4368 | VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); |
a218c879 BP |
4369 | ds_destroy(&s); |
4370 | } | |
4371 | ||
4372 | COVERAGE_INC(facet_unexpected); | |
acf60855 | 4373 | dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); |
a218c879 BP |
4374 | } |
4375 | ||
abe529af BP |
4376 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4377 | * | |
4378 | * This function also pushes statistics updates to rules which each facet | |
4379 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4380 | * facet changes the rule it resubmits into at some time in between | |
4381 | * update_stats() runs, it is possible that statistics accrued to the | |
4382 | * old rule will be incorrectly attributed to the new rule. This could be | |
4383 | * avoided by calling update_stats() whenever rules are created or | |
4384 | * deleted. However, the performance impact of making so many calls to the | |
4385 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4386 | * |
4387 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4388 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4389 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4390 | * the hit or miss counts of br1. | |
abe529af BP |
4391 | */ |
4392 | static void | |
acf60855 | 4393 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4394 | { |
4395 | const struct dpif_flow_stats *stats; | |
4396 | struct dpif_flow_dump dump; | |
4397 | const struct nlattr *key; | |
3af56aef | 4398 | struct ofproto_dpif *ofproto; |
abe529af BP |
4399 | size_t key_len; |
4400 | ||
acf60855 | 4401 | dpif_flow_dump_start(&dump, backer->dpif); |
abe529af | 4402 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
acf60855 | 4403 | struct flow flow; |
b0f7b9b5 | 4404 | struct subfacet *subfacet; |
b9ad7294 | 4405 | struct ofport_dpif *ofport; |
acf60855 | 4406 | uint32_t key_hash; |
abe529af | 4407 | |
58c6adda EJ |
4408 | if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, |
4409 | NULL, NULL)) { | |
acf60855 JP |
4410 | continue; |
4411 | } | |
4412 | ||
655ab909 AZ |
4413 | ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets); |
4414 | ofproto->n_update_stats++; | |
655ab909 | 4415 | |
b9ad7294 EJ |
4416 | ofport = get_ofp_port(ofproto, flow.in_port); |
4417 | if (ofport && ofport->tnl_port) { | |
4418 | netdev_vport_inc_rx(ofport->up.netdev, stats); | |
4419 | } | |
4420 | ||
acf60855 | 4421 | key_hash = odp_flow_key_hash(key, key_len); |
9566abf9 | 4422 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
6a7e895f BP |
4423 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4424 | case SF_FAST_PATH: | |
735d7efb AZ |
4425 | /* Update ofproto_dpif's hit count. */ |
4426 | if (stats->n_packets > subfacet->dp_packet_count) { | |
4427 | uint64_t delta = stats->n_packets - subfacet->dp_packet_count; | |
4428 | dpif_stats_update_hit_count(ofproto, delta); | |
4429 | } | |
4430 | ||
a218c879 | 4431 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4432 | break; |
4433 | ||
4434 | case SF_SLOW_PATH: | |
4435 | /* Stats are updated per-packet. */ | |
4436 | break; | |
4437 | ||
4438 | case SF_NOT_INSTALLED: | |
4439 | default: | |
acf60855 | 4440 | delete_unexpected_flow(ofproto, key, key_len); |
6a7e895f | 4441 | break; |
abe529af | 4442 | } |
8fa4d1d0 | 4443 | run_fast_rl(); |
abe529af BP |
4444 | } |
4445 | dpif_flow_dump_done(&dump); | |
3af56aef AZ |
4446 | |
4447 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
4448 | update_moving_averages(ofproto); | |
4449 | } | |
4450 | ||
abe529af BP |
4451 | } |
4452 | ||
4453 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4454 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4455 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4456 | * fold its statistic into its rule. */ | |
abe529af | 4457 | static int |
b0f7b9b5 | 4458 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
4459 | { |
4460 | /* | |
4461 | * Idle time histogram. | |
4462 | * | |
b0f7b9b5 BP |
4463 | * Most of the time a switch has a relatively small number of subfacets. |
4464 | * When this is the case we might as well keep statistics for all of them | |
4465 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4466 | * well. |
4467 | * | |
b0f7b9b5 | 4468 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4469 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4470 | * significant. However, with a large number of subfacets it is likely |
4471 | * that only a few of them are "heavy hitters" that consume a large amount | |
4472 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4473 | * the kernel and maintaining in userspaces; other subfacets we can | |
4474 | * discard. | |
abe529af BP |
4475 | * |
4476 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4477 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4478 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4479 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4480 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 4481 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4482 | * the most-recently-used bucket of subfacets is kept, so actually an |
4483 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4484 | * (though the next run will delete most of those unless they receive |
4485 | * additional data). | |
abe529af | 4486 | * |
b0f7b9b5 BP |
4487 | * This requires a second pass through the subfacets, in addition to the |
4488 | * pass made by update_stats(), because the former function never looks at | |
4489 | * uninstallable subfacets. | |
abe529af BP |
4490 | */ |
4491 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4492 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4493 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4494 | int total, subtotal, bucket; |
b0f7b9b5 | 4495 | struct subfacet *subfacet; |
abe529af BP |
4496 | long long int now; |
4497 | int i; | |
4498 | ||
b0f7b9b5 | 4499 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 4500 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
4501 | return N_BUCKETS * BUCKET_WIDTH; |
4502 | } | |
4503 | ||
4504 | /* Build histogram. */ | |
4505 | now = time_msec(); | |
b0f7b9b5 BP |
4506 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
4507 | long long int idle = now - subfacet->used; | |
abe529af BP |
4508 | int bucket = (idle <= 0 ? 0 |
4509 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4510 | : (unsigned int) idle / BUCKET_WIDTH); | |
4511 | buckets[bucket]++; | |
4512 | } | |
4513 | ||
4514 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4515 | subtotal = bucket = 0; |
4516 | do { | |
4517 | subtotal += buckets[bucket++]; | |
084f5290 SH |
4518 | } while (bucket < N_BUCKETS && |
4519 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
4520 | |
4521 | if (VLOG_IS_DBG_ENABLED()) { | |
4522 | struct ds s; | |
4523 | ||
4524 | ds_init(&s); | |
4525 | ds_put_cstr(&s, "keep"); | |
4526 | for (i = 0; i < N_BUCKETS; i++) { | |
4527 | if (i == bucket) { | |
4528 | ds_put_cstr(&s, ", drop"); | |
4529 | } | |
4530 | if (buckets[i]) { | |
4531 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4532 | } | |
4533 | } | |
4534 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
4535 | ds_destroy(&s); | |
4536 | } | |
4537 | ||
4538 | return bucket * BUCKET_WIDTH; | |
4539 | } | |
4540 | ||
abe529af | 4541 | static void |
b0f7b9b5 | 4542 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 4543 | { |
625b0720 BP |
4544 | /* Cutoff time for most flows. */ |
4545 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4546 | ||
4547 | /* We really want to keep flows for special protocols around, so use a more | |
4548 | * conservative cutoff. */ | |
4549 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4550 | |
b0f7b9b5 | 4551 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4552 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4553 | int n_batch; |
abe529af | 4554 | |
b99d3cee | 4555 | n_batch = 0; |
b0f7b9b5 BP |
4556 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
4557 | &ofproto->subfacets) { | |
625b0720 BP |
4558 | long long int cutoff; |
4559 | ||
ccc09689 | 4560 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP) |
625b0720 BP |
4561 | ? special_cutoff |
4562 | : normal_cutoff); | |
b0f7b9b5 | 4563 | if (subfacet->used < cutoff) { |
6a7e895f | 4564 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4565 | batch[n_batch++] = subfacet; |
1d85f9e5 JP |
4566 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
4567 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
b99d3cee BP |
4568 | n_batch = 0; |
4569 | } | |
4570 | } else { | |
4571 | subfacet_destroy(subfacet); | |
4572 | } | |
abe529af BP |
4573 | } |
4574 | } | |
b99d3cee BP |
4575 | |
4576 | if (n_batch > 0) { | |
1d85f9e5 | 4577 | subfacet_destroy_batch(ofproto, batch, n_batch); |
b99d3cee | 4578 | } |
abe529af BP |
4579 | } |
4580 | ||
4581 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4582 | * then delete it entirely. */ | |
4583 | static void | |
4584 | rule_expire(struct rule_dpif *rule) | |
4585 | { | |
abe529af BP |
4586 | struct facet *facet, *next_facet; |
4587 | long long int now; | |
4588 | uint8_t reason; | |
4589 | ||
e2a3d183 BP |
4590 | if (rule->up.pending) { |
4591 | /* We'll have to expire it later. */ | |
4592 | return; | |
4593 | } | |
4594 | ||
abe529af BP |
4595 | /* Has 'rule' expired? */ |
4596 | now = time_msec(); | |
4597 | if (rule->up.hard_timeout | |
308881af | 4598 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4599 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4600 | } else if (rule->up.idle_timeout |
1745cd08 | 4601 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4602 | reason = OFPRR_IDLE_TIMEOUT; |
4603 | } else { | |
4604 | return; | |
4605 | } | |
4606 | ||
4607 | COVERAGE_INC(ofproto_dpif_expired); | |
4608 | ||
4609 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4610 | * timeout, because that only happens when the rule has no facets left.) */ | |
4611 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 4612 | facet_remove(facet); |
abe529af BP |
4613 | } |
4614 | ||
4615 | /* Get rid of the rule. */ | |
4616 | ofproto_rule_expire(&rule->up, reason); | |
4617 | } | |
4618 | \f | |
4619 | /* Facets. */ | |
4620 | ||
f3827897 | 4621 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
4622 | * |
4623 | * The caller must already have determined that no facet with an identical | |
4624 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
4625 | * the ofproto's classifier table. |
4626 | * | |
2b459b83 BP |
4627 | * 'hash' must be the return value of flow_hash(flow, 0). |
4628 | * | |
b0f7b9b5 BP |
4629 | * The facet will initially have no subfacets. The caller should create (at |
4630 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4631 | static struct facet * |
2b459b83 | 4632 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
4633 | { |
4634 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4635 | struct facet *facet; | |
4636 | ||
4637 | facet = xzalloc(sizeof *facet); | |
4638 | facet->used = time_msec(); | |
2b459b83 | 4639 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
4640 | list_push_back(&rule->facets, &facet->list_node); |
4641 | facet->rule = rule; | |
4642 | facet->flow = *flow; | |
b0f7b9b5 | 4643 | list_init(&facet->subfacets); |
abe529af BP |
4644 | netflow_flow_init(&facet->nf_flow); |
4645 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4646 | ||
6cf474d7 EJ |
4647 | facet->learn_rl = time_msec() + 500; |
4648 | ||
abe529af BP |
4649 | return facet; |
4650 | } | |
4651 | ||
4652 | static void | |
4653 | facet_free(struct facet *facet) | |
4654 | { | |
abe529af BP |
4655 | free(facet); |
4656 | } | |
4657 | ||
3d9e05f8 | 4658 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4659 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4660 | static bool |
4661 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4662 | const struct nlattr *odp_actions, size_t actions_len, | |
4663 | struct ofpbuf *packet) | |
4664 | { | |
4665 | struct odputil_keybuf keybuf; | |
4666 | struct ofpbuf key; | |
4667 | int error; | |
4668 | ||
6ff686f2 | 4669 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
4670 | odp_flow_key_from_flow(&key, flow, |
4671 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
80e5eed9 | 4672 | |
acf60855 | 4673 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4674 | odp_actions, actions_len, packet); |
6ff686f2 | 4675 | return !error; |
abe529af BP |
4676 | } |
4677 | ||
abe529af BP |
4678 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
4679 | * | |
4680 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4681 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4682 | * |
4683 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4684 | */ | |
4685 | static void | |
15baa734 | 4686 | facet_remove(struct facet *facet) |
abe529af | 4687 | { |
15baa734 | 4688 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4689 | struct subfacet *subfacet, *next_subfacet; |
4690 | ||
cb22974d | 4691 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4692 | |
4693 | /* First uninstall all of the subfacets to get final statistics. */ | |
4694 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4695 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4696 | } |
4697 | ||
4698 | /* Flush the final stats to the rule. | |
4699 | * | |
4700 | * This might require us to have at least one subfacet around so that we | |
4701 | * can use its actions for accounting in facet_account(), which is why we | |
4702 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4703 | facet_flush_stats(facet); |
551a2f6c BP |
4704 | |
4705 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4706 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4707 | &facet->subfacets) { | |
15baa734 | 4708 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4709 | } |
abe529af BP |
4710 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
4711 | list_remove(&facet->list_node); | |
4712 | facet_free(facet); | |
4713 | } | |
4714 | ||
3de9590b BP |
4715 | /* Feed information from 'facet' back into the learning table to keep it in |
4716 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4717 | static void |
3de9590b | 4718 | facet_learn(struct facet *facet) |
abe529af | 4719 | { |
15baa734 | 4720 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
14f94f9a JP |
4721 | struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets), |
4722 | struct subfacet, list_node); | |
62bd7349 | 4723 | long long int now = time_msec(); |
3de9590b | 4724 | struct action_xlate_ctx ctx; |
abe529af | 4725 | |
62bd7349 | 4726 | if (!facet->has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
4727 | return; |
4728 | } | |
4729 | ||
62bd7349 | 4730 | facet->learn_rl = now + 500; |
6cf474d7 | 4731 | |
3de9590b BP |
4732 | if (!facet->has_learn |
4733 | && !facet->has_normal | |
4734 | && (!facet->has_fin_timeout | |
4735 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
4736 | return; |
4737 | } | |
abe529af | 4738 | |
3de9590b | 4739 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
14f94f9a | 4740 | &subfacet->initial_vals, |
3de9590b BP |
4741 | facet->rule, facet->tcp_flags, NULL); |
4742 | ctx.may_learn = true; | |
f25d0cf3 BP |
4743 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, |
4744 | facet->rule->up.ofpacts_len); | |
3de9590b BP |
4745 | } |
4746 | ||
4747 | static void | |
4748 | facet_account(struct facet *facet) | |
4749 | { | |
4750 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
1a1e3a0a | 4751 | struct subfacet *subfacet = facet_get_subfacet(facet); |
3de9590b BP |
4752 | const struct nlattr *a; |
4753 | unsigned int left; | |
4754 | ovs_be16 vlan_tci; | |
4755 | uint64_t n_bytes; | |
abe529af | 4756 | |
75a75043 | 4757 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
4758 | return; |
4759 | } | |
3de9590b | 4760 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4761 | |
4762 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4763 | * as a basis. We also need to track the actual VLAN on which the packet | |
4764 | * is going to be sent to ensure that it matches the one passed to | |
4765 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4766 | * hash bucket.) |
4767 | * | |
4768 | * We use the actions from an arbitrary subfacet because they should all | |
4769 | * be equally valid for our purpose. */ | |
d78be13b | 4770 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
4771 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
4772 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 4773 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4774 | struct ofport_dpif *port; |
abe529af | 4775 | |
d78be13b | 4776 | switch (nl_attr_type(a)) { |
df2c07f4 | 4777 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
4778 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
4779 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 4780 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4781 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4782 | } |
d78be13b BP |
4783 | break; |
4784 | ||
fea393b1 BP |
4785 | case OVS_ACTION_ATTR_POP_VLAN: |
4786 | vlan_tci = htons(0); | |
d78be13b BP |
4787 | break; |
4788 | ||
fea393b1 BP |
4789 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4790 | vlan = nl_attr_get(a); | |
4791 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4792 | break; |
abe529af BP |
4793 | } |
4794 | } | |
4795 | } | |
4796 | ||
abe529af BP |
4797 | /* Returns true if the only action for 'facet' is to send to the controller. |
4798 | * (We don't report NetFlow expiration messages for such facets because they | |
4799 | * are just part of the control logic for the network, not real traffic). */ | |
4800 | static bool | |
4801 | facet_is_controller_flow(struct facet *facet) | |
4802 | { | |
f25d0cf3 BP |
4803 | if (facet) { |
4804 | const struct rule *rule = &facet->rule->up; | |
4805 | const struct ofpact *ofpacts = rule->ofpacts; | |
4806 | size_t ofpacts_len = rule->ofpacts_len; | |
4807 | ||
dd30ff28 BP |
4808 | if (ofpacts_len > 0 && |
4809 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4810 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4811 | return true; | |
4812 | } | |
4813 | } | |
4814 | return false; | |
abe529af BP |
4815 | } |
4816 | ||
4817 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4818 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4819 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4820 | * its packet and byte counts before this function is called. */ | |
4821 | static void | |
15baa734 | 4822 | facet_flush_stats(struct facet *facet) |
abe529af | 4823 | { |
15baa734 | 4824 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4825 | struct subfacet *subfacet; |
4826 | ||
4827 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4828 | ovs_assert(!subfacet->dp_byte_count); |
4829 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4830 | } |
abe529af BP |
4831 | |
4832 | facet_push_stats(facet); | |
3de9590b BP |
4833 | if (facet->accounted_bytes < facet->byte_count) { |
4834 | facet_account(facet); | |
4835 | facet->accounted_bytes = facet->byte_count; | |
4836 | } | |
abe529af BP |
4837 | |
4838 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4839 | struct ofexpired expired; | |
4840 | expired.flow = facet->flow; | |
4841 | expired.packet_count = facet->packet_count; | |
4842 | expired.byte_count = facet->byte_count; | |
4843 | expired.used = facet->used; | |
4844 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4845 | } | |
4846 | ||
abe529af BP |
4847 | /* Reset counters to prevent double counting if 'facet' ever gets |
4848 | * reinstalled. */ | |
bbb5d219 | 4849 | facet_reset_counters(facet); |
abe529af BP |
4850 | |
4851 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4852 | facet->tcp_flags = 0; |
abe529af BP |
4853 | } |
4854 | ||
4855 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4856 | * Returns it if found, otherwise a null pointer. | |
4857 | * | |
2b459b83 BP |
4858 | * 'hash' must be the return value of flow_hash(flow, 0). |
4859 | * | |
abe529af BP |
4860 | * The returned facet might need revalidation; use facet_lookup_valid() |
4861 | * instead if that is important. */ | |
4862 | static struct facet * | |
2b459b83 BP |
4863 | facet_find(struct ofproto_dpif *ofproto, |
4864 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
4865 | { |
4866 | struct facet *facet; | |
4867 | ||
2b459b83 | 4868 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
4869 | if (flow_equal(flow, &facet->flow)) { |
4870 | return facet; | |
4871 | } | |
4872 | } | |
4873 | ||
4874 | return NULL; | |
4875 | } | |
4876 | ||
4877 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4878 | * Returns it if found, otherwise a null pointer. | |
4879 | * | |
2b459b83 BP |
4880 | * 'hash' must be the return value of flow_hash(flow, 0). |
4881 | * | |
abe529af BP |
4882 | * The returned facet is guaranteed to be valid. */ |
4883 | static struct facet * | |
2b459b83 BP |
4884 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
4885 | uint32_t hash) | |
abe529af | 4886 | { |
c57b2226 | 4887 | struct facet *facet; |
abe529af | 4888 | |
c57b2226 | 4889 | facet = facet_find(ofproto, flow, hash); |
abe529af | 4890 | if (facet |
2cc3c58e EJ |
4891 | && (ofproto->backer->need_revalidate |
4892 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
4893 | facet->tags))) { | |
c57b2226 | 4894 | facet_revalidate(facet); |
f231418e EJ |
4895 | |
4896 | /* facet_revalidate() may have destroyed 'facet'. */ | |
4897 | facet = facet_find(ofproto, flow, hash); | |
abe529af BP |
4898 | } |
4899 | ||
4900 | return facet; | |
4901 | } | |
4902 | ||
1a1e3a0a JP |
4903 | /* Return a subfacet from 'facet'. A facet consists of one or more |
4904 | * subfacets, and this function returns one of them. */ | |
4905 | static struct subfacet *facet_get_subfacet(struct facet *facet) | |
4906 | { | |
4907 | return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet, | |
4908 | list_node); | |
4909 | } | |
4910 | ||
6a7e895f BP |
4911 | static const char * |
4912 | subfacet_path_to_string(enum subfacet_path path) | |
4913 | { | |
4914 | switch (path) { | |
4915 | case SF_NOT_INSTALLED: | |
4916 | return "not installed"; | |
4917 | case SF_FAST_PATH: | |
4918 | return "in fast path"; | |
4919 | case SF_SLOW_PATH: | |
4920 | return "in slow path"; | |
4921 | default: | |
4922 | return "<error>"; | |
4923 | } | |
4924 | } | |
4925 | ||
4926 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
4927 | * member has the specified value. */ | |
4928 | static enum subfacet_path | |
4929 | subfacet_want_path(enum slow_path_reason slow) | |
4930 | { | |
4931 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
4932 | } | |
4933 | ||
4934 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
4935 | * supposing that its actions have been recalculated as 'want_actions' and that | |
4936 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
4937 | static bool | |
4938 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
4939 | const struct ofpbuf *want_actions) | |
4940 | { | |
4941 | enum subfacet_path want_path = subfacet_want_path(slow); | |
4942 | return (want_path != subfacet->path | |
4943 | || (want_path == SF_FAST_PATH | |
4944 | && (subfacet->actions_len != want_actions->size | |
4945 | || memcmp(subfacet->actions, want_actions->data, | |
4946 | subfacet->actions_len)))); | |
4947 | } | |
4948 | ||
6814e51f BP |
4949 | static bool |
4950 | facet_check_consistency(struct facet *facet) | |
4951 | { | |
4952 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4953 | ||
4954 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4955 | ||
050ac423 BP |
4956 | uint64_t odp_actions_stub[1024 / 8]; |
4957 | struct ofpbuf odp_actions; | |
4958 | ||
6814e51f BP |
4959 | struct rule_dpif *rule; |
4960 | struct subfacet *subfacet; | |
c53e1132 | 4961 | bool may_log = false; |
6814e51f BP |
4962 | bool ok; |
4963 | ||
4964 | /* Check the rule for consistency. */ | |
c57b2226 BP |
4965 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
4966 | ok = rule == facet->rule; | |
4967 | if (!ok) { | |
c53e1132 | 4968 | may_log = !VLOG_DROP_WARN(&rl); |
c53e1132 BP |
4969 | if (may_log) { |
4970 | struct ds s; | |
6814e51f | 4971 | |
c53e1132 BP |
4972 | ds_init(&s); |
4973 | flow_format(&s, &facet->flow); | |
4974 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4975 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4976 | cls_rule_format(&facet->rule->up.cr, &s); | |
4977 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4978 | rule->up.table_id); | |
4979 | cls_rule_format(&rule->up.cr, &s); | |
4980 | ds_put_char(&s, ')'); | |
6814e51f | 4981 | |
c53e1132 BP |
4982 | VLOG_WARN("%s", ds_cstr(&s)); |
4983 | ds_destroy(&s); | |
4984 | } | |
6814e51f BP |
4985 | } |
4986 | ||
4987 | /* Check the datapath actions for consistency. */ | |
050ac423 | 4988 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f | 4989 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4990 | enum subfacet_path want_path; |
6814e51f | 4991 | struct action_xlate_ctx ctx; |
9616614b | 4992 | struct ds s; |
6814e51f BP |
4993 | |
4994 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
14f94f9a | 4995 | &subfacet->initial_vals, rule, 0, NULL); |
f25d0cf3 | 4996 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 4997 | &odp_actions); |
6814e51f | 4998 | |
6a7e895f BP |
4999 | if (subfacet->path == SF_NOT_INSTALLED) { |
5000 | /* This only happens if the datapath reported an error when we | |
5001 | * tried to install the flow. Don't flag another error here. */ | |
5002 | continue; | |
5003 | } | |
5004 | ||
5005 | want_path = subfacet_want_path(subfacet->slow); | |
5006 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
5007 | /* The actions for slow-path flows may legitimately vary from one | |
5008 | * packet to the next. We're done. */ | |
050ac423 | 5009 | continue; |
6814e51f BP |
5010 | } |
5011 | ||
6a7e895f | 5012 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { |
9616614b BP |
5013 | continue; |
5014 | } | |
c53e1132 | 5015 | |
9616614b BP |
5016 | /* Inconsistency! */ |
5017 | if (ok) { | |
5018 | may_log = !VLOG_DROP_WARN(&rl); | |
5019 | ok = false; | |
5020 | } | |
5021 | if (!may_log) { | |
5022 | /* Rate-limited, skip reporting. */ | |
5023 | continue; | |
5024 | } | |
c53e1132 | 5025 | |
9616614b | 5026 | ds_init(&s); |
9566abf9 | 5027 | odp_flow_key_format(subfacet->key, subfacet->key_len, &s); |
9616614b BP |
5028 | |
5029 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
6a7e895f | 5030 | if (want_path != subfacet->path) { |
9616614b BP |
5031 | enum odp_key_fitness fitness = subfacet->key_fitness; |
5032 | ||
6a7e895f BP |
5033 | ds_put_format(&s, " (%s, fitness=%s)", |
5034 | subfacet_path_to_string(subfacet->path), | |
9616614b | 5035 | odp_key_fitness_to_string(fitness)); |
6a7e895f BP |
5036 | ds_put_format(&s, " (should have been %s)", |
5037 | subfacet_path_to_string(want_path)); | |
5038 | } else if (want_path == SF_FAST_PATH) { | |
9616614b BP |
5039 | ds_put_cstr(&s, " (actions were: "); |
5040 | format_odp_actions(&s, subfacet->actions, | |
5041 | subfacet->actions_len); | |
5042 | ds_put_cstr(&s, ") (correct actions: "); | |
5043 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
5044 | ds_put_char(&s, ')'); | |
5045 | } else { | |
5046 | ds_put_cstr(&s, " (actions: "); | |
5047 | format_odp_actions(&s, subfacet->actions, | |
5048 | subfacet->actions_len); | |
5049 | ds_put_char(&s, ')'); | |
6814e51f | 5050 | } |
9616614b BP |
5051 | VLOG_WARN("%s", ds_cstr(&s)); |
5052 | ds_destroy(&s); | |
6814e51f | 5053 | } |
050ac423 | 5054 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
5055 | |
5056 | return ok; | |
5057 | } | |
5058 | ||
15baa734 | 5059 | /* Re-searches the classifier for 'facet': |
abe529af BP |
5060 | * |
5061 | * - If the rule found is different from 'facet''s current rule, moves | |
5062 | * 'facet' to the new rule and recompiles its actions. | |
5063 | * | |
5064 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
5065 | * where it is and recompiles its actions anyway. |
5066 | * | |
5067 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
5068 | * ofproto_receive(), 'facet' is removed. */ | |
c57b2226 | 5069 | static void |
15baa734 | 5070 | facet_revalidate(struct facet *facet) |
abe529af | 5071 | { |
15baa734 | 5072 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
5073 | struct actions { |
5074 | struct nlattr *odp_actions; | |
5075 | size_t actions_len; | |
5076 | }; | |
5077 | struct actions *new_actions; | |
5078 | ||
abe529af | 5079 | struct action_xlate_ctx ctx; |
050ac423 BP |
5080 | uint64_t odp_actions_stub[1024 / 8]; |
5081 | struct ofpbuf odp_actions; | |
5082 | ||
abe529af | 5083 | struct rule_dpif *new_rule; |
b0f7b9b5 | 5084 | struct subfacet *subfacet; |
b95fc6ba | 5085 | int i; |
abe529af BP |
5086 | |
5087 | COVERAGE_INC(facet_revalidate); | |
5088 | ||
f231418e EJ |
5089 | /* Check that child subfacets still correspond to this facet. Tunnel |
5090 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
5091 | * change. */ | |
5092 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5093 | struct ofproto_dpif *recv_ofproto; | |
5094 | struct flow recv_flow; | |
5095 | int error; | |
5096 | ||
5097 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
5098 | subfacet->key_len, &recv_flow, NULL, | |
5099 | &recv_ofproto, NULL, NULL); | |
5100 | if (error | |
5101 | || recv_ofproto != ofproto | |
5102 | || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { | |
5103 | facet_remove(facet); | |
5104 | return; | |
5105 | } | |
5106 | } | |
5107 | ||
c57b2226 | 5108 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 5109 | |
df2c07f4 | 5110 | /* Calculate new datapath actions. |
abe529af BP |
5111 | * |
5112 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
5113 | * emit a NetFlow expiration and, if so, we need to have the old state | |
5114 | * around to properly compose it. */ | |
abe529af | 5115 | |
df2c07f4 JP |
5116 | /* If the datapath actions changed or the installability changed, |
5117 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
5118 | i = 0; |
5119 | new_actions = NULL; | |
5120 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 5121 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 5122 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 5123 | enum slow_path_reason slow; |
b95fc6ba | 5124 | |
e84173dc | 5125 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
14f94f9a | 5126 | &subfacet->initial_vals, new_rule, 0, NULL); |
f25d0cf3 | 5127 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, |
050ac423 | 5128 | &odp_actions); |
b0f7b9b5 | 5129 | |
6a7e895f BP |
5130 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5131 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
5132 | struct dpif_flow_stats stats; | |
5133 | ||
5134 | subfacet_install(subfacet, | |
5135 | odp_actions.data, odp_actions.size, &stats, slow); | |
5136 | subfacet_update_stats(subfacet, &stats); | |
b95fc6ba BP |
5137 | |
5138 | if (!new_actions) { | |
5139 | new_actions = xcalloc(list_size(&facet->subfacets), | |
5140 | sizeof *new_actions); | |
5141 | } | |
050ac423 BP |
5142 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
5143 | odp_actions.size); | |
5144 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 5145 | } |
b95fc6ba | 5146 | |
b95fc6ba | 5147 | i++; |
b0f7b9b5 | 5148 | } |
050ac423 BP |
5149 | ofpbuf_uninit(&odp_actions); |
5150 | ||
b95fc6ba | 5151 | if (new_actions) { |
15baa734 | 5152 | facet_flush_stats(facet); |
abe529af BP |
5153 | } |
5154 | ||
5155 | /* Update 'facet' now that we've taken care of all the old state. */ | |
5156 | facet->tags = ctx.tags; | |
5157 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
75a75043 BP |
5158 | facet->has_learn = ctx.has_learn; |
5159 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 5160 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 5161 | facet->mirrors = ctx.mirrors; |
6a7e895f BP |
5162 | |
5163 | i = 0; | |
5164 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5165 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
5166 | ||
5167 | if (new_actions && new_actions[i].odp_actions) { | |
5168 | free(subfacet->actions); | |
5169 | subfacet->actions = new_actions[i].odp_actions; | |
5170 | subfacet->actions_len = new_actions[i].actions_len; | |
b95fc6ba | 5171 | } |
6a7e895f | 5172 | i++; |
abe529af | 5173 | } |
6a7e895f BP |
5174 | free(new_actions); |
5175 | ||
abe529af BP |
5176 | if (facet->rule != new_rule) { |
5177 | COVERAGE_INC(facet_changed_rule); | |
5178 | list_remove(&facet->list_node); | |
5179 | list_push_back(&new_rule->facets, &facet->list_node); | |
5180 | facet->rule = new_rule; | |
5181 | facet->used = new_rule->up.created; | |
9d24de3b | 5182 | facet->prev_used = facet->used; |
abe529af | 5183 | } |
abe529af BP |
5184 | } |
5185 | ||
5186 | /* Updates 'facet''s used time. Caller is responsible for calling | |
5187 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
5188 | static void | |
15baa734 | 5189 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 5190 | { |
15baa734 | 5191 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
5192 | if (used > facet->used) { |
5193 | facet->used = used; | |
1745cd08 | 5194 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
5195 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
5196 | } | |
5197 | } | |
5198 | ||
bbb5d219 EJ |
5199 | static void |
5200 | facet_reset_counters(struct facet *facet) | |
5201 | { | |
5202 | facet->packet_count = 0; | |
5203 | facet->byte_count = 0; | |
9d24de3b JP |
5204 | facet->prev_packet_count = 0; |
5205 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
5206 | facet->accounted_bytes = 0; |
5207 | } | |
5208 | ||
abe529af BP |
5209 | static void |
5210 | facet_push_stats(struct facet *facet) | |
5211 | { | |
112bc5f4 | 5212 | struct dpif_flow_stats stats; |
abe529af | 5213 | |
cb22974d BP |
5214 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
5215 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
5216 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 5217 | |
112bc5f4 BP |
5218 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
5219 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
5220 | stats.used = facet->used; | |
5221 | stats.tcp_flags = 0; | |
abe529af | 5222 | |
112bc5f4 | 5223 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
5224 | facet->prev_packet_count = facet->packet_count; |
5225 | facet->prev_byte_count = facet->byte_count; | |
5226 | facet->prev_used = facet->used; | |
abe529af | 5227 | |
eafed69b | 5228 | rule_credit_stats(facet->rule, &stats); |
ac35f9c8 | 5229 | flow_push_stats(facet, &stats); |
9d24de3b JP |
5230 | |
5231 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 5232 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
5233 | } |
5234 | } | |
5235 | ||
8844e035 | 5236 | static void |
4464589e | 5237 | push_all_stats__(bool run_fast) |
8844e035 EJ |
5238 | { |
5239 | static long long int rl = LLONG_MIN; | |
5240 | struct ofproto_dpif *ofproto; | |
5241 | ||
5242 | if (time_msec() < rl) { | |
5243 | return; | |
5244 | } | |
5245 | ||
5246 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5247 | struct facet *facet; | |
5248 | ||
5249 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
5250 | facet_push_stats(facet); | |
4464589e EJ |
5251 | if (run_fast) { |
5252 | run_fast_rl(); | |
5253 | } | |
8844e035 EJ |
5254 | } |
5255 | } | |
5256 | ||
5257 | rl = time_msec() + 100; | |
5258 | } | |
5259 | ||
4464589e EJ |
5260 | static void |
5261 | push_all_stats(void) | |
5262 | { | |
5263 | push_all_stats__(true); | |
5264 | } | |
5265 | ||
abe529af | 5266 | static void |
112bc5f4 | 5267 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 5268 | { |
112bc5f4 BP |
5269 | rule->packet_count += stats->n_packets; |
5270 | rule->byte_count += stats->n_bytes; | |
5271 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
5272 | } |
5273 | ||
ac35f9c8 JP |
5274 | /* Pushes flow statistics to the rules which 'facet->flow' resubmits |
5275 | * into given 'facet->rule''s actions and mirrors. */ | |
abe529af | 5276 | static void |
ac35f9c8 | 5277 | flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats) |
abe529af | 5278 | { |
ac35f9c8 | 5279 | struct rule_dpif *rule = facet->rule; |
abe529af | 5280 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
1a1e3a0a | 5281 | struct subfacet *subfacet = facet_get_subfacet(facet); |
112bc5f4 | 5282 | struct action_xlate_ctx ctx; |
abe529af | 5283 | |
112bc5f4 | 5284 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 5285 | |
14f94f9a JP |
5286 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
5287 | &subfacet->initial_vals, rule, 0, NULL); | |
112bc5f4 | 5288 | ctx.resubmit_stats = stats; |
f25d0cf3 BP |
5289 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, |
5290 | rule->up.ofpacts_len); | |
abe529af BP |
5291 | } |
5292 | \f | |
b0f7b9b5 BP |
5293 | /* Subfacets. */ |
5294 | ||
5295 | static struct subfacet * | |
acf60855 | 5296 | subfacet_find(struct ofproto_dpif *ofproto, |
9566abf9 | 5297 | const struct nlattr *key, size_t key_len, uint32_t key_hash) |
b0f7b9b5 BP |
5298 | { |
5299 | struct subfacet *subfacet; | |
5300 | ||
5301 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
5302 | &ofproto->subfacets) { | |
9566abf9 EJ |
5303 | if (subfacet->key_len == key_len |
5304 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
5305 | return subfacet; |
5306 | } | |
5307 | } | |
5308 | ||
5309 | return NULL; | |
5310 | } | |
5311 | ||
5312 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
5313 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
5314 | * existing subfacet if there is one, otherwise creates and returns a | |
5315 | * new subfacet. | |
b95fc6ba BP |
5316 | * |
5317 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
5318 | * which case the caller must populate the actions with | |
5319 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 5320 | static struct subfacet * |
a088a1ff JP |
5321 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
5322 | long long int now) | |
b0f7b9b5 | 5323 | { |
15baa734 | 5324 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
a088a1ff JP |
5325 | enum odp_key_fitness key_fitness = miss->key_fitness; |
5326 | const struct nlattr *key = miss->key; | |
5327 | size_t key_len = miss->key_len; | |
5328 | uint32_t key_hash; | |
b0f7b9b5 BP |
5329 | struct subfacet *subfacet; |
5330 | ||
a088a1ff JP |
5331 | key_hash = odp_flow_key_hash(key, key_len); |
5332 | ||
3b145dd7 BP |
5333 | if (list_is_empty(&facet->subfacets)) { |
5334 | subfacet = &facet->one_subfacet; | |
5335 | } else { | |
9566abf9 | 5336 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
3b145dd7 BP |
5337 | if (subfacet) { |
5338 | if (subfacet->facet == facet) { | |
5339 | return subfacet; | |
5340 | } | |
5341 | ||
5342 | /* This shouldn't happen. */ | |
5343 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5344 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
5345 | } |
5346 | ||
3b145dd7 | 5347 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
5348 | } |
5349 | ||
b0f7b9b5 BP |
5350 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
5351 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
5352 | subfacet->facet = facet; | |
b0f7b9b5 | 5353 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
5354 | subfacet->key = xmemdup(key, key_len); |
5355 | subfacet->key_len = key_len; | |
459b16a1 | 5356 | subfacet->used = now; |
655ab909 | 5357 | subfacet->created = now; |
26cd7e34 BP |
5358 | subfacet->dp_packet_count = 0; |
5359 | subfacet->dp_byte_count = 0; | |
5360 | subfacet->actions_len = 0; | |
5361 | subfacet->actions = NULL; | |
6a7e895f BP |
5362 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE |
5363 | ? SLOW_MATCH | |
5364 | : 0); | |
5365 | subfacet->path = SF_NOT_INSTALLED; | |
14f94f9a | 5366 | subfacet->initial_vals = miss->initial_vals; |
a088a1ff | 5367 | subfacet->odp_in_port = miss->odp_in_port; |
b0f7b9b5 | 5368 | |
655ab909 | 5369 | ofproto->subfacet_add_count++; |
b0f7b9b5 BP |
5370 | return subfacet; |
5371 | } | |
5372 | ||
b0f7b9b5 BP |
5373 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
5374 | * its facet within 'ofproto', and frees it. */ | |
5375 | static void | |
15baa734 | 5376 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 5377 | { |
15baa734 BP |
5378 | struct facet *facet = subfacet->facet; |
5379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5380 | ||
655ab909 AZ |
5381 | /* Update ofproto stats before uninstall the subfacet. */ |
5382 | ofproto->subfacet_del_count++; | |
5383 | ofproto->total_subfacet_life_span += (time_msec() - subfacet->created); | |
5384 | ||
15baa734 | 5385 | subfacet_uninstall(subfacet); |
b0f7b9b5 BP |
5386 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
5387 | list_remove(&subfacet->list_node); | |
5388 | free(subfacet->key); | |
b95fc6ba | 5389 | free(subfacet->actions); |
26cd7e34 BP |
5390 | if (subfacet != &facet->one_subfacet) { |
5391 | free(subfacet); | |
5392 | } | |
b0f7b9b5 BP |
5393 | } |
5394 | ||
5395 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5396 | * last remaining subfacet in its facet destroys the facet too. */ | |
5397 | static void | |
15baa734 | 5398 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
5399 | { |
5400 | struct facet *facet = subfacet->facet; | |
5401 | ||
551a2f6c BP |
5402 | if (list_is_singleton(&facet->subfacets)) { |
5403 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 5404 | facet_remove(facet); |
551a2f6c | 5405 | } else { |
15baa734 | 5406 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
5407 | } |
5408 | } | |
5409 | ||
1d85f9e5 JP |
5410 | static void |
5411 | subfacet_destroy_batch(struct ofproto_dpif *ofproto, | |
5412 | struct subfacet **subfacets, int n) | |
5413 | { | |
1d85f9e5 JP |
5414 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
5415 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
5416 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
5417 | int i; | |
5418 | ||
5419 | for (i = 0; i < n; i++) { | |
5420 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
5421 | ops[i].u.flow_del.key = subfacets[i]->key; |
5422 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
5423 | ops[i].u.flow_del.stats = &stats[i]; |
5424 | opsp[i] = &ops[i]; | |
5425 | } | |
5426 | ||
acf60855 | 5427 | dpif_operate(ofproto->backer->dpif, opsp, n); |
1d85f9e5 JP |
5428 | for (i = 0; i < n; i++) { |
5429 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5430 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5431 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 5432 | run_fast_rl(); |
1d85f9e5 JP |
5433 | } |
5434 | } | |
5435 | ||
5fe20d5d BP |
5436 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
5437 | * Translates the actions into 'odp_actions', which the caller must have | |
5438 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 5439 | static void |
5fe20d5d BP |
5440 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
5441 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
5442 | { |
5443 | struct facet *facet = subfacet->facet; | |
18b2a258 | 5444 | struct rule_dpif *rule = facet->rule; |
15baa734 | 5445 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 5446 | |
b95fc6ba BP |
5447 | struct action_xlate_ctx ctx; |
5448 | ||
14f94f9a JP |
5449 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
5450 | &subfacet->initial_vals, rule, 0, packet); | |
f25d0cf3 | 5451 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); |
b95fc6ba | 5452 | facet->tags = ctx.tags; |
b95fc6ba BP |
5453 | facet->has_learn = ctx.has_learn; |
5454 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 5455 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 5456 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 5457 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 5458 | |
6a7e895f | 5459 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5fe20d5d BP |
5460 | if (subfacet->actions_len != odp_actions->size |
5461 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 5462 | free(subfacet->actions); |
5fe20d5d BP |
5463 | subfacet->actions_len = odp_actions->size; |
5464 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 5465 | } |
b95fc6ba BP |
5466 | } |
5467 | ||
b0f7b9b5 BP |
5468 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5469 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5470 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5471 | * since 'subfacet' was last updated. | |
5472 | * | |
5473 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5474 | static int | |
15baa734 | 5475 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 | 5476 | const struct nlattr *actions, size_t actions_len, |
6a7e895f BP |
5477 | struct dpif_flow_stats *stats, |
5478 | enum slow_path_reason slow) | |
b0f7b9b5 | 5479 | { |
15baa734 BP |
5480 | struct facet *facet = subfacet->facet; |
5481 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
6a7e895f BP |
5482 | enum subfacet_path path = subfacet_want_path(slow); |
5483 | uint64_t slow_path_stub[128 / 8]; | |
b0f7b9b5 | 5484 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
5485 | int ret; |
5486 | ||
5487 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5488 | if (stats) { | |
5489 | flags |= DPIF_FP_ZERO_STATS; | |
5490 | } | |
5491 | ||
6a7e895f BP |
5492 | if (path == SF_SLOW_PATH) { |
5493 | compose_slow_path(ofproto, &facet->flow, slow, | |
5494 | slow_path_stub, sizeof slow_path_stub, | |
5495 | &actions, &actions_len); | |
5496 | } | |
5497 | ||
9566abf9 EJ |
5498 | ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key, |
5499 | subfacet->key_len, actions, actions_len, stats); | |
b0f7b9b5 BP |
5500 | |
5501 | if (stats) { | |
5502 | subfacet_reset_dp_stats(subfacet, stats); | |
5503 | } | |
5504 | ||
6a7e895f BP |
5505 | if (!ret) { |
5506 | subfacet->path = path; | |
5507 | } | |
b0f7b9b5 BP |
5508 | return ret; |
5509 | } | |
5510 | ||
6a7e895f BP |
5511 | static int |
5512 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
5513 | { | |
5514 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
5515 | stats, subfacet->slow); | |
5516 | } | |
5517 | ||
b0f7b9b5 BP |
5518 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ |
5519 | static void | |
15baa734 | 5520 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5521 | { |
6a7e895f | 5522 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
5523 | struct rule_dpif *rule = subfacet->facet->rule; |
5524 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 | 5525 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5526 | int error; |
5527 | ||
9566abf9 EJ |
5528 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5529 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5530 | subfacet_reset_dp_stats(subfacet, &stats); |
5531 | if (!error) { | |
15baa734 | 5532 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5533 | } |
6a7e895f | 5534 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5535 | } else { |
cb22974d BP |
5536 | ovs_assert(subfacet->dp_packet_count == 0); |
5537 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5538 | } |
5539 | } | |
5540 | ||
5541 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5542 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5543 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5544 | * was reset in the datapath. 'stats' will be modified to include only | |
5545 | * statistics new since 'subfacet' was last updated. */ | |
5546 | static void | |
5547 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5548 | struct dpif_flow_stats *stats) | |
5549 | { | |
5550 | if (stats | |
5551 | && subfacet->dp_packet_count <= stats->n_packets | |
5552 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5553 | stats->n_packets -= subfacet->dp_packet_count; | |
5554 | stats->n_bytes -= subfacet->dp_byte_count; | |
5555 | } | |
5556 | ||
5557 | subfacet->dp_packet_count = 0; | |
5558 | subfacet->dp_byte_count = 0; | |
5559 | } | |
5560 | ||
5561 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
5562 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
5563 | static void | |
15baa734 | 5564 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
5565 | { |
5566 | if (used > subfacet->used) { | |
5567 | subfacet->used = used; | |
15baa734 | 5568 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
5569 | } |
5570 | } | |
5571 | ||
5572 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
5573 | * | |
5574 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5575 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5576 | * represents a packet that was sent by hand or if it represents statistics | |
5577 | * that have been cleared out of the datapath. */ | |
5578 | static void | |
15baa734 | 5579 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5580 | const struct dpif_flow_stats *stats) |
5581 | { | |
5582 | if (stats->n_packets || stats->used > subfacet->used) { | |
5583 | struct facet *facet = subfacet->facet; | |
5584 | ||
15baa734 | 5585 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
5586 | facet->packet_count += stats->n_packets; |
5587 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5588 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5589 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); |
5590 | } | |
5591 | } | |
5592 | \f | |
abe529af BP |
5593 | /* Rules. */ |
5594 | ||
5595 | static struct rule_dpif * | |
c57b2226 BP |
5596 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
5597 | { | |
c57b2226 BP |
5598 | struct rule_dpif *rule; |
5599 | ||
5600 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5601 | if (rule) { | |
5602 | return rule; | |
5603 | } | |
5604 | ||
c376f9a3 | 5605 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5606 | } |
5607 | ||
5608 | static struct rule_dpif * | |
5609 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5610 | uint8_t table_id) | |
abe529af | 5611 | { |
7257b535 BP |
5612 | struct cls_rule *cls_rule; |
5613 | struct classifier *cls; | |
5614 | ||
9cdaaebe BP |
5615 | if (table_id >= N_TABLES) { |
5616 | return NULL; | |
5617 | } | |
5618 | ||
d0918789 | 5619 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 5620 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
5621 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
5622 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
5623 | * are unavailable. */ | |
5624 | struct flow ofpc_normal_flow = *flow; | |
5625 | ofpc_normal_flow.tp_src = htons(0); | |
5626 | ofpc_normal_flow.tp_dst = htons(0); | |
5627 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
5628 | } else { | |
5629 | cls_rule = classifier_lookup(cls, flow); | |
5630 | } | |
5631 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5632 | } |
5633 | ||
c376f9a3 IY |
5634 | static struct rule_dpif * |
5635 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5636 | { | |
5637 | struct ofport_dpif *port; | |
5638 | ||
5639 | port = get_ofp_port(ofproto, flow->in_port); | |
5640 | if (!port) { | |
5641 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5642 | return ofproto->miss_rule; | |
5643 | } | |
5644 | ||
5645 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5646 | return ofproto->no_packet_in_rule; | |
5647 | } | |
5648 | return ofproto->miss_rule; | |
5649 | } | |
5650 | ||
7ee20df1 BP |
5651 | static void |
5652 | complete_operation(struct rule_dpif *rule) | |
5653 | { | |
5654 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5655 | ||
54a9cbc9 | 5656 | rule_invalidate(rule); |
7ee20df1 BP |
5657 | if (clogged) { |
5658 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5659 | c->op = rule->up.pending; | |
5660 | list_push_back(&ofproto->completions, &c->list_node); | |
5661 | } else { | |
5662 | ofoperation_complete(rule->up.pending, 0); | |
5663 | } | |
5664 | } | |
5665 | ||
abe529af BP |
5666 | static struct rule * |
5667 | rule_alloc(void) | |
5668 | { | |
5669 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5670 | return &rule->up; | |
5671 | } | |
5672 | ||
5673 | static void | |
5674 | rule_dealloc(struct rule *rule_) | |
5675 | { | |
5676 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5677 | free(rule); | |
5678 | } | |
5679 | ||
90bf1e07 | 5680 | static enum ofperr |
abe529af BP |
5681 | rule_construct(struct rule *rule_) |
5682 | { | |
5683 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5684 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 5685 | struct rule_dpif *victim; |
54a9cbc9 | 5686 | uint8_t table_id; |
abe529af | 5687 | |
abe529af BP |
5688 | rule->packet_count = 0; |
5689 | rule->byte_count = 0; | |
abe529af | 5690 | |
7ee20df1 BP |
5691 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
5692 | if (victim && !list_is_empty(&victim->facets)) { | |
5693 | struct facet *facet; | |
5694 | ||
5695 | rule->facets = victim->facets; | |
5696 | list_moved(&rule->facets); | |
5697 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
5698 | /* XXX: We're only clearing our local counters here. It's possible |
5699 | * that quite a few packets are unaccounted for in the datapath | |
5700 | * statistics. These will be accounted to the new rule instead of | |
5701 | * cleared as required. This could be fixed by clearing out the | |
5702 | * datapath statistics for this facet, but currently it doesn't | |
5703 | * seem worth it. */ | |
5704 | facet_reset_counters(facet); | |
7ee20df1 BP |
5705 | facet->rule = rule; |
5706 | } | |
5707 | } else { | |
5708 | /* Must avoid list_moved() in this case. */ | |
5709 | list_init(&rule->facets); | |
5710 | } | |
abe529af | 5711 | |
54a9cbc9 | 5712 | table_id = rule->up.table_id; |
5cb7a798 BP |
5713 | if (victim) { |
5714 | rule->tag = victim->tag; | |
5715 | } else if (table_id == 0) { | |
5716 | rule->tag = 0; | |
5717 | } else { | |
5718 | struct flow flow; | |
5719 | ||
5720 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5721 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5722 | ofproto->tables[table_id].basis); | |
5723 | } | |
54a9cbc9 | 5724 | |
7ee20df1 | 5725 | complete_operation(rule); |
abe529af BP |
5726 | return 0; |
5727 | } | |
5728 | ||
5729 | static void | |
5730 | rule_destruct(struct rule *rule_) | |
5731 | { | |
5732 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5733 | struct facet *facet, *next_facet; |
5734 | ||
abe529af | 5735 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 5736 | facet_revalidate(facet); |
abe529af | 5737 | } |
7ee20df1 BP |
5738 | |
5739 | complete_operation(rule); | |
abe529af BP |
5740 | } |
5741 | ||
5742 | static void | |
5743 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5744 | { | |
5745 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 5746 | |
4464589e EJ |
5747 | /* push_all_stats() can handle flow misses which, when using the learn |
5748 | * action, can cause rules to be added and deleted. This can corrupt our | |
5749 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5750 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5751 | push_all_stats__(false); | |
bf1e8ff9 | 5752 | |
abe529af BP |
5753 | /* Start from historical data for 'rule' itself that are no longer tracked |
5754 | * in facets. This counts, for example, facets that have expired. */ | |
5755 | *packets = rule->packet_count; | |
5756 | *bytes = rule->byte_count; | |
abe529af BP |
5757 | } |
5758 | ||
0a740f48 EJ |
5759 | static void |
5760 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5761 | struct ofpbuf *packet) | |
abe529af | 5762 | { |
abe529af | 5763 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
14f94f9a | 5764 | struct initial_vals initial_vals; |
112bc5f4 | 5765 | struct dpif_flow_stats stats; |
abe529af | 5766 | struct action_xlate_ctx ctx; |
050ac423 BP |
5767 | uint64_t odp_actions_stub[1024 / 8]; |
5768 | struct ofpbuf odp_actions; | |
abe529af | 5769 | |
a7752d4a | 5770 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5771 | rule_credit_stats(rule, &stats); |
5772 | ||
14f94f9a | 5773 | initial_vals.vlan_tci = flow->vlan_tci; |
050ac423 | 5774 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
14f94f9a | 5775 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, |
112bc5f4 BP |
5776 | rule, stats.tcp_flags, packet); |
5777 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 5778 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); |
112bc5f4 BP |
5779 | |
5780 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
5781 | odp_actions.size, packet); | |
5782 | ||
050ac423 | 5783 | ofpbuf_uninit(&odp_actions); |
0a740f48 | 5784 | } |
5bf0e941 | 5785 | |
0a740f48 EJ |
5786 | static enum ofperr |
5787 | rule_execute(struct rule *rule, const struct flow *flow, | |
5788 | struct ofpbuf *packet) | |
5789 | { | |
5790 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5791 | ofpbuf_delete(packet); | |
5bf0e941 | 5792 | return 0; |
abe529af BP |
5793 | } |
5794 | ||
7ee20df1 BP |
5795 | static void |
5796 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5797 | { |
5798 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5799 | |
5800 | complete_operation(rule); | |
abe529af BP |
5801 | } |
5802 | \f | |
97d6520b | 5803 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5804 | * May modify 'packet'. |
abe529af BP |
5805 | * Returns 0 if successful, otherwise a positive errno value. */ |
5806 | static int | |
52a90c29 | 5807 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5808 | { |
79e15b78 | 5809 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5810 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 | 5811 | struct ofpbuf key, odp_actions; |
0f49659a | 5812 | struct dpif_flow_stats stats; |
80e5eed9 | 5813 | struct odputil_keybuf keybuf; |
0f49659a EJ |
5814 | struct ofpact_output output; |
5815 | struct action_xlate_ctx ctx; | |
80e5eed9 | 5816 | struct flow flow; |
abe529af BP |
5817 | int error; |
5818 | ||
b9ad7294 | 5819 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
0f49659a | 5820 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
b9ad7294 | 5821 | |
0f49659a EJ |
5822 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ |
5823 | flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow); | |
5824 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, | |
5825 | OFPP_LOCAL)); | |
5826 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 | 5827 | |
0f49659a EJ |
5828 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); |
5829 | output.port = ofport->up.ofp_port; | |
5830 | output.max_len = 0; | |
b9ad7294 | 5831 | |
0f49659a EJ |
5832 | action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet); |
5833 | ctx.resubmit_stats = &stats; | |
5834 | xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions); | |
6ff686f2 | 5835 | |
acf60855 | 5836 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 BP |
5837 | key.data, key.size, |
5838 | odp_actions.data, odp_actions.size, | |
abe529af BP |
5839 | packet); |
5840 | ofpbuf_uninit(&odp_actions); | |
5841 | ||
5842 | if (error) { | |
0f49659a EJ |
5843 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", |
5844 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
5845 | strerror(error)); | |
abe529af | 5846 | } |
79e15b78 EJ |
5847 | |
5848 | ofproto->stats.tx_packets++; | |
5849 | ofproto->stats.tx_bytes += packet->size; | |
abe529af BP |
5850 | return error; |
5851 | } | |
5852 | \f | |
df2c07f4 | 5853 | /* OpenFlow to datapath action translation. */ |
abe529af | 5854 | |
ffaef958 | 5855 | static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *); |
f25d0cf3 BP |
5856 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
5857 | struct action_xlate_ctx *); | |
4cd78906 | 5858 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 5859 | |
6a7e895f BP |
5860 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5861 | * The action will state 'slow' as the reason that the action is in the slow | |
5862 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5863 | * dump-flows" output to see why a flow is in the slow path.) | |
5864 | * | |
5865 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5866 | * 'stub_size' must be large enough for the action. | |
5867 | * | |
5868 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5869 | * respectively. */ | |
5870 | static void | |
5871 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5872 | enum slow_path_reason slow, | |
5873 | uint64_t *stub, size_t stub_size, | |
5874 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5875 | { | |
5876 | union user_action_cookie cookie; | |
5877 | struct ofpbuf buf; | |
5878 | ||
5879 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5880 | cookie.slow_path.unused = 0; | |
5881 | cookie.slow_path.reason = slow; | |
5882 | ||
5883 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5884 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
9032f11e | 5885 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); |
29089a54 | 5886 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5887 | } else { |
29089a54 RL |
5888 | put_userspace_action(ofproto, &buf, flow, &cookie, |
5889 | sizeof cookie.slow_path); | |
625b0720 | 5890 | } |
6a7e895f BP |
5891 | *actionsp = buf.data; |
5892 | *actions_lenp = buf.size; | |
5893 | } | |
5894 | ||
98403001 BP |
5895 | static size_t |
5896 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5897 | struct ofpbuf *odp_actions, | |
5898 | const struct flow *flow, | |
29089a54 RL |
5899 | const union user_action_cookie *cookie, |
5900 | const size_t cookie_size) | |
98403001 | 5901 | { |
98403001 BP |
5902 | uint32_t pid; |
5903 | ||
acf60855 | 5904 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
e1b1d06a | 5905 | ofp_port_to_odp_port(ofproto, flow->in_port)); |
98403001 | 5906 | |
29089a54 RL |
5907 | return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); |
5908 | } | |
5909 | ||
5910 | /* Compose SAMPLE action for sFlow or IPFIX. The given probability is | |
5911 | * the number of packets out of UINT32_MAX to sample. The given | |
5912 | * cookie is passed back in the callback for each sampled packet. | |
5913 | */ | |
5914 | static size_t | |
5915 | compose_sample_action(const struct ofproto_dpif *ofproto, | |
5916 | struct ofpbuf *odp_actions, | |
5917 | const struct flow *flow, | |
5918 | const uint32_t probability, | |
5919 | const union user_action_cookie *cookie, | |
5920 | const size_t cookie_size) | |
5921 | { | |
5922 | size_t sample_offset, actions_offset; | |
5923 | int cookie_offset; | |
5924 | ||
5925 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
5926 | ||
5927 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5928 | ||
5929 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
5930 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie, | |
5931 | cookie_size); | |
5932 | ||
5933 | nl_msg_end_nested(odp_actions, actions_offset); | |
5934 | nl_msg_end_nested(odp_actions, sample_offset); | |
5935 | return cookie_offset; | |
98403001 BP |
5936 | } |
5937 | ||
36fc5f18 BP |
5938 | static void |
5939 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5940 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 5941 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
5942 | { |
5943 | int ifindex; | |
5944 | ||
5945 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 5946 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
5947 | |
5948 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5949 | * port information") for the interpretation of cookie->output. */ | |
5950 | switch (n_outputs) { | |
5951 | case 0: | |
5952 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 5953 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
5954 | break; |
5955 | ||
5956 | case 1: | |
5957 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5958 | if (ifindex) { | |
1673e0e4 | 5959 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
5960 | break; |
5961 | } | |
5962 | /* Fall through. */ | |
5963 | default: | |
5964 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 5965 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
5966 | break; |
5967 | } | |
5968 | } | |
5969 | ||
29089a54 | 5970 | /* Compose SAMPLE action for sFlow bridge sampling. */ |
6ff686f2 PS |
5971 | static size_t |
5972 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5973 | struct ofpbuf *odp_actions, | |
5974 | const struct flow *flow, | |
5975 | uint32_t odp_port) | |
5976 | { | |
6ff686f2 | 5977 | uint32_t probability; |
1673e0e4 | 5978 | union user_action_cookie cookie; |
6ff686f2 PS |
5979 | |
5980 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5981 | return 0; | |
5982 | } | |
5983 | ||
6ff686f2 | 5984 | probability = dpif_sflow_get_probability(ofproto->sflow); |
36fc5f18 BP |
5985 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
5986 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
6ff686f2 | 5987 | |
29089a54 RL |
5988 | return compose_sample_action(ofproto, odp_actions, flow, probability, |
5989 | &cookie, sizeof cookie.sflow); | |
5990 | } | |
5991 | ||
5992 | static void | |
5993 | compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id, | |
5994 | uint32_t obs_domain_id, uint32_t obs_point_id, | |
5995 | union user_action_cookie *cookie) | |
5996 | { | |
5997 | cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE; | |
5998 | cookie->flow_sample.probability = probability; | |
5999 | cookie->flow_sample.collector_set_id = collector_set_id; | |
6000 | cookie->flow_sample.obs_domain_id = obs_domain_id; | |
6001 | cookie->flow_sample.obs_point_id = obs_point_id; | |
6002 | } | |
6003 | ||
6004 | static void | |
6005 | compose_ipfix_cookie(union user_action_cookie *cookie) | |
6006 | { | |
6007 | cookie->type = USER_ACTION_COOKIE_IPFIX; | |
6008 | } | |
6009 | ||
6010 | /* Compose SAMPLE action for IPFIX bridge sampling. */ | |
6011 | static void | |
6012 | compose_ipfix_action(const struct ofproto_dpif *ofproto, | |
6013 | struct ofpbuf *odp_actions, | |
6014 | const struct flow *flow) | |
6015 | { | |
6016 | uint32_t probability; | |
6017 | union user_action_cookie cookie; | |
6018 | ||
6019 | if (!ofproto->ipfix || flow->in_port == OFPP_NONE) { | |
6020 | return; | |
6021 | } | |
6022 | ||
6023 | probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix); | |
6024 | compose_ipfix_cookie(&cookie); | |
6025 | ||
6026 | compose_sample_action(ofproto, odp_actions, flow, probability, | |
6027 | &cookie, sizeof cookie.ipfix); | |
6ff686f2 PS |
6028 | } |
6029 | ||
29089a54 RL |
6030 | /* SAMPLE action for sFlow must be first action in any given list of |
6031 | * actions. At this point we do not have all information required to | |
6032 | * build it. So try to build sample action as complete as possible. */ | |
6ff686f2 PS |
6033 | static void |
6034 | add_sflow_action(struct action_xlate_ctx *ctx) | |
6035 | { | |
6036 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
6037 | ctx->odp_actions, | |
6038 | &ctx->flow, OVSP_NONE); | |
6039 | ctx->sflow_odp_port = 0; | |
6040 | ctx->sflow_n_outputs = 0; | |
6041 | } | |
6042 | ||
29089a54 RL |
6043 | /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list |
6044 | * of actions, eventually after the SAMPLE action for sFlow. */ | |
6045 | static void | |
6046 | add_ipfix_action(struct action_xlate_ctx *ctx) | |
6047 | { | |
6048 | compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow); | |
6049 | } | |
6050 | ||
6ff686f2 PS |
6051 | /* Fix SAMPLE action according to data collected while composing ODP actions. |
6052 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
6053 | * USERSPACE action's user-cookie which is required for sflow. */ | |
6054 | static void | |
6055 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
6056 | { | |
6057 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 6058 | union user_action_cookie *cookie; |
6ff686f2 PS |
6059 | |
6060 | if (!ctx->user_cookie_offset) { | |
6061 | return; | |
6062 | } | |
6063 | ||
6064 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
a93f1927 | 6065 | sizeof cookie->sflow); |
cb22974d | 6066 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
6ff686f2 | 6067 | |
36fc5f18 BP |
6068 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
6069 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
6070 | } |
6071 | ||
6ff686f2 | 6072 | static void |
81b1afb1 EJ |
6073 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
6074 | bool check_stp) | |
6ff686f2 | 6075 | { |
d59906fb | 6076 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
d8558b4a JR |
6077 | ovs_be16 flow_vlan_tci; |
6078 | uint32_t flow_skb_mark; | |
6079 | uint8_t flow_nw_tos; | |
a4454ac6 | 6080 | struct priority_to_dscp *pdscp; |
0a740f48 EJ |
6081 | uint32_t out_port, odp_port; |
6082 | ||
6083 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
6084 | * before traversing a patch port. */ | |
cff78c88 | 6085 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); |
d59906fb | 6086 | |
a4454ac6 EJ |
6087 | if (!ofport) { |
6088 | xlate_report(ctx, "Nonexistent output port"); | |
6089 | return; | |
6090 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
6091 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
6092 | return; | |
6093 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
6094 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
6095 | return; | |
6096 | } | |
8b36f51e | 6097 | |
0a740f48 EJ |
6098 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
6099 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
6100 | struct flow old_flow = ctx->flow; | |
6101 | const struct ofproto_dpif *peer_ofproto; | |
bb374ef6 | 6102 | enum slow_path_reason special; |
ffaef958 | 6103 | struct ofport_dpif *in_port; |
0a740f48 EJ |
6104 | |
6105 | if (!peer) { | |
6106 | xlate_report(ctx, "Nonexistent patch port peer"); | |
6107 | return; | |
6108 | } | |
6109 | ||
6110 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
6111 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
6112 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
6113 | return; | |
6114 | } | |
6115 | ||
6116 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
6117 | ctx->flow.in_port = peer->up.ofp_port; | |
6118 | ctx->flow.metadata = htonll(0); | |
6119 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); | |
6120 | memset(ctx->flow.regs, 0, sizeof ctx->flow.regs); | |
ffaef958 BP |
6121 | |
6122 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
bb374ef6 EJ |
6123 | special = process_special(ctx->ofproto, &ctx->flow, in_port, |
6124 | ctx->packet); | |
6125 | if (special) { | |
6126 | ctx->slow |= special; | |
6127 | } else if (!in_port || may_receive(in_port, ctx)) { | |
ffaef958 BP |
6128 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { |
6129 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
6130 | } else { | |
6131 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
6132 | * learning action look at the packet, then drop it. */ | |
6133 | struct flow old_base_flow = ctx->base_flow; | |
6134 | size_t old_size = ctx->odp_actions->size; | |
6135 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
6136 | ctx->base_flow = old_base_flow; | |
6137 | ctx->odp_actions->size = old_size; | |
6138 | } | |
6139 | } | |
6140 | ||
0a740f48 EJ |
6141 | ctx->flow = old_flow; |
6142 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
6143 | ||
6144 | if (ctx->resubmit_stats) { | |
b9ad7294 EJ |
6145 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); |
6146 | netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats); | |
0a740f48 EJ |
6147 | } |
6148 | ||
6149 | return; | |
6150 | } | |
6151 | ||
d8558b4a JR |
6152 | flow_vlan_tci = ctx->flow.vlan_tci; |
6153 | flow_skb_mark = ctx->flow.skb_mark; | |
6154 | flow_nw_tos = ctx->flow.nw_tos; | |
6155 | ||
a4454ac6 EJ |
6156 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
6157 | if (pdscp) { | |
6158 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
6159 | ctx->flow.nw_tos |= pdscp->dscp; | |
d59906fb EJ |
6160 | } |
6161 | ||
b9ad7294 | 6162 | if (ofport->tnl_port) { |
d8558b4a JR |
6163 | /* Save tunnel metadata so that changes made due to |
6164 | * the Logical (tunnel) Port are not visible for any further | |
6165 | * matches, while explicit set actions on tunnel metadata are. | |
6166 | */ | |
6167 | struct flow_tnl flow_tnl = ctx->flow.tunnel; | |
b9ad7294 EJ |
6168 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow); |
6169 | if (odp_port == OVSP_NONE) { | |
6170 | xlate_report(ctx, "Tunneling decided against output"); | |
88e90491 | 6171 | goto out; /* restore flow_nw_tos */ |
b9ad7294 | 6172 | } |
0ad90c84 JR |
6173 | if (ctx->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) { |
6174 | xlate_report(ctx, "Not tunneling to our own address"); | |
6175 | goto out; /* restore flow_nw_tos */ | |
6176 | } | |
b9ad7294 EJ |
6177 | if (ctx->resubmit_stats) { |
6178 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); | |
6179 | } | |
6180 | out_port = odp_port; | |
6181 | commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow, | |
6182 | ctx->odp_actions); | |
d8558b4a | 6183 | ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */ |
b9ad7294 | 6184 | } else { |
cf630ea3 | 6185 | odp_port = ofport->odp_port; |
b9ad7294 EJ |
6186 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
6187 | ctx->flow.vlan_tci); | |
6188 | if (out_port != odp_port) { | |
6189 | ctx->flow.vlan_tci = htons(0); | |
6190 | } | |
321fa429 | 6191 | ctx->flow.skb_mark &= ~IPSEC_MARK; |
52a90c29 | 6192 | } |
5bbda0aa | 6193 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
6194 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
6195 | ||
6ff686f2 PS |
6196 | ctx->sflow_odp_port = odp_port; |
6197 | ctx->sflow_n_outputs++; | |
81b1afb1 | 6198 | ctx->nf_output_iface = ofp_port; |
d8558b4a JR |
6199 | |
6200 | /* Restore flow */ | |
52a90c29 | 6201 | ctx->flow.vlan_tci = flow_vlan_tci; |
d8558b4a | 6202 | ctx->flow.skb_mark = flow_skb_mark; |
88e90491 | 6203 | out: |
8b36f51e | 6204 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
6205 | } |
6206 | ||
abe529af | 6207 | static void |
5e48dc2b | 6208 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 6209 | { |
81b1afb1 | 6210 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
6211 | } |
6212 | ||
55599423 JR |
6213 | static void |
6214 | tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6215 | { | |
6216 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6217 | uint8_t table_id = ctx->table_id; | |
6218 | ||
6219 | if (table_id > 0 && table_id < N_TABLES) { | |
6220 | struct table_dpif *table = &ofproto->tables[table_id]; | |
6221 | if (table->other_table) { | |
6222 | ctx->tags |= (rule && rule->tag | |
6223 | ? rule->tag | |
6224 | : rule_calculate_tag(&ctx->flow, | |
6225 | &table->other_table->mask, | |
6226 | table->basis)); | |
6227 | } | |
6228 | } | |
6229 | } | |
6230 | ||
6231 | /* Common rule processing in one place to avoid duplicating code. */ | |
6232 | static struct rule_dpif * | |
6233 | ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule, | |
6234 | bool may_packet_in) | |
6235 | { | |
6236 | if (ctx->resubmit_hook) { | |
6237 | ctx->resubmit_hook(ctx, rule); | |
6238 | } | |
6239 | if (rule == NULL && may_packet_in) { | |
6240 | /* XXX | |
6241 | * check if table configuration flags | |
6242 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
6243 | * OFPTC_TABLE_MISS_CONTINUE, | |
6244 | * OFPTC_TABLE_MISS_DROP | |
6245 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
6246 | */ | |
6247 | rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow); | |
6248 | } | |
6249 | if (rule && ctx->resubmit_stats) { | |
6250 | rule_credit_stats(rule, ctx->resubmit_stats); | |
6251 | } | |
6252 | return rule; | |
6253 | } | |
6254 | ||
abe529af | 6255 | static void |
29901626 | 6256 | xlate_table_action(struct action_xlate_ctx *ctx, |
1688c479 | 6257 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
6258 | { |
6259 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
6260 | struct rule_dpif *rule; | |
55599423 JR |
6261 | uint16_t old_in_port = ctx->flow.in_port; |
6262 | uint8_t old_table_id = ctx->table_id; | |
29901626 | 6263 | |
29901626 | 6264 | ctx->table_id = table_id; |
abe529af | 6265 | |
54a9cbc9 | 6266 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af | 6267 | ctx->flow.in_port = in_port; |
55599423 JR |
6268 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id); |
6269 | ||
6270 | tag_the_flow(ctx, rule); | |
54a9cbc9 BP |
6271 | |
6272 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
6273 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
6274 | ctx->flow.in_port = old_in_port; |
6275 | ||
55599423 | 6276 | rule = ctx_rule_hooks(ctx, rule, may_packet_in); |
1688c479 | 6277 | |
abe529af | 6278 | if (rule) { |
18b2a258 | 6279 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 6280 | |
abe529af | 6281 | ctx->recurse++; |
18b2a258 | 6282 | ctx->rule = rule; |
f25d0cf3 | 6283 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 6284 | ctx->rule = old_rule; |
abe529af BP |
6285 | ctx->recurse--; |
6286 | } | |
29901626 BP |
6287 | |
6288 | ctx->table_id = old_table_id; | |
abe529af BP |
6289 | } else { |
6290 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
6291 | ||
29901626 | 6292 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 6293 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 6294 | ctx->max_resubmit_trigger = true; |
abe529af BP |
6295 | } |
6296 | } | |
6297 | ||
29901626 | 6298 | static void |
f25d0cf3 BP |
6299 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, |
6300 | const struct ofpact_resubmit *resubmit) | |
29901626 BP |
6301 | { |
6302 | uint16_t in_port; | |
6303 | uint8_t table_id; | |
6304 | ||
f25d0cf3 BP |
6305 | in_port = resubmit->in_port; |
6306 | if (in_port == OFPP_IN_PORT) { | |
6307 | in_port = ctx->flow.in_port; | |
6308 | } | |
6309 | ||
6310 | table_id = resubmit->table_id; | |
6311 | if (table_id == 255) { | |
6312 | table_id = ctx->table_id; | |
6313 | } | |
29901626 | 6314 | |
1688c479 | 6315 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
6316 | } |
6317 | ||
abe529af | 6318 | static void |
d59906fb | 6319 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
6320 | { |
6321 | struct ofport_dpif *ofport; | |
6322 | ||
b3e9b2ed | 6323 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 6324 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
6325 | |
6326 | if (ofp_port == ctx->flow.in_port) { | |
6327 | continue; | |
6328 | } | |
6329 | ||
5e48dc2b | 6330 | if (all) { |
81b1afb1 | 6331 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 6332 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 6333 | compose_output_action(ctx, ofp_port); |
abe529af BP |
6334 | } |
6335 | } | |
b3e9b2ed EJ |
6336 | |
6337 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
6338 | } |
6339 | ||
6ff686f2 | 6340 | static void |
f0fd1a17 | 6341 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
6342 | enum ofp_packet_in_reason reason, |
6343 | uint16_t controller_id) | |
6ff686f2 | 6344 | { |
999fba59 EJ |
6345 | struct ofputil_packet_in pin; |
6346 | struct ofpbuf *packet; | |
6ff686f2 | 6347 | |
6a7e895f | 6348 | ctx->slow |= SLOW_CONTROLLER; |
999fba59 EJ |
6349 | if (!ctx->packet) { |
6350 | return; | |
6351 | } | |
6352 | ||
6353 | packet = ofpbuf_clone(ctx->packet); | |
6354 | ||
6355 | if (packet->l2 && packet->l3) { | |
6356 | struct eth_header *eh; | |
b02475c5 | 6357 | uint16_t mpls_depth; |
999fba59 EJ |
6358 | |
6359 | eth_pop_vlan(packet); | |
6360 | eh = packet->l2; | |
0104aba8 | 6361 | |
999fba59 EJ |
6362 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
6363 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
6364 | ||
6365 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
6366 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
6367 | } | |
6368 | ||
b02475c5 SH |
6369 | mpls_depth = eth_mpls_depth(packet); |
6370 | ||
6371 | if (mpls_depth < ctx->flow.mpls_depth) { | |
6372 | push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse); | |
6373 | } else if (mpls_depth > ctx->flow.mpls_depth) { | |
6374 | pop_mpls(packet, ctx->flow.dl_type); | |
6375 | } else if (mpls_depth) { | |
6376 | set_mpls_lse(packet, ctx->flow.mpls_lse); | |
6377 | } | |
6378 | ||
999fba59 EJ |
6379 | if (packet->l4) { |
6380 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6381 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
6382 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
6383 | } | |
6384 | ||
6385 | if (packet->l7) { | |
6386 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
6387 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
6388 | ctx->flow.tp_dst); | |
6389 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
6390 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
6391 | ctx->flow.tp_dst); | |
6392 | } | |
6393 | } | |
6394 | } | |
6395 | } | |
6396 | ||
6397 | pin.packet = packet->data; | |
6398 | pin.packet_len = packet->size; | |
f0fd1a17 | 6399 | pin.reason = reason; |
a7349929 | 6400 | pin.controller_id = controller_id; |
54834960 | 6401 | pin.table_id = ctx->table_id; |
18b2a258 | 6402 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 6403 | |
999fba59 | 6404 | pin.send_len = len; |
999fba59 EJ |
6405 | flow_get_metadata(&ctx->flow, &pin.fmd); |
6406 | ||
d8653c38 | 6407 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 6408 | ofpbuf_delete(packet); |
6ff686f2 PS |
6409 | } |
6410 | ||
b02475c5 SH |
6411 | static void |
6412 | execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6413 | { | |
6414 | ovs_assert(eth_type_mpls(eth_type)); | |
6415 | ||
6416 | if (ctx->base_flow.mpls_depth) { | |
6417 | ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); | |
6418 | ctx->flow.mpls_depth++; | |
6419 | } else { | |
6420 | ovs_be32 label; | |
6421 | uint8_t tc, ttl; | |
6422 | ||
6423 | if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) { | |
6424 | label = htonl(0x2); /* IPV6 Explicit Null. */ | |
6425 | } else { | |
6426 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6427 | } | |
6428 | tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2; | |
6429 | ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40; | |
6430 | ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
b02475c5 SH |
6431 | ctx->flow.mpls_depth = 1; |
6432 | } | |
6433 | ctx->flow.dl_type = eth_type; | |
6434 | } | |
6435 | ||
6436 | static void | |
6437 | execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6438 | { | |
6439 | ovs_assert(eth_type_mpls(ctx->flow.dl_type)); | |
6440 | ovs_assert(!eth_type_mpls(eth_type)); | |
6441 | ||
6442 | if (ctx->flow.mpls_depth) { | |
6443 | ctx->flow.mpls_depth--; | |
6444 | ctx->flow.mpls_lse = htonl(0); | |
6445 | if (!ctx->flow.mpls_depth) { | |
6446 | ctx->flow.dl_type = eth_type; | |
b02475c5 SH |
6447 | } |
6448 | } | |
6449 | } | |
6450 | ||
f0fd1a17 | 6451 | static bool |
c2d967a5 | 6452 | compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 PS |
6453 | { |
6454 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
6455 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
6456 | return false; | |
6457 | } | |
6458 | ||
6459 | if (ctx->flow.nw_ttl > 1) { | |
6460 | ctx->flow.nw_ttl--; | |
6461 | return false; | |
6462 | } else { | |
c2d967a5 MM |
6463 | size_t i; |
6464 | ||
6465 | for (i = 0; i < ids->n_controllers; i++) { | |
6466 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6467 | ids->cnt_ids[i]); | |
6468 | } | |
f0fd1a17 PS |
6469 | |
6470 | /* Stop processing for current table. */ | |
6471 | return true; | |
6472 | } | |
6473 | } | |
6474 | ||
0f3f3c3d SH |
6475 | static bool |
6476 | execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl) | |
6477 | { | |
6478 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6479 | return true; | |
6480 | } | |
6481 | ||
6482 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6483 | return false; | |
6484 | } | |
6485 | ||
b676167a SH |
6486 | static bool |
6487 | execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx) | |
6488 | { | |
6489 | uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse); | |
6490 | ||
6491 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6492 | return false; | |
6493 | } | |
6494 | ||
be80bc65 | 6495 | if (ttl > 1) { |
b676167a SH |
6496 | ttl--; |
6497 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6498 | return false; | |
6499 | } else { | |
6500 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
6501 | ||
6502 | /* Stop processing for current table. */ | |
6503 | return true; | |
6504 | } | |
6505 | } | |
6506 | ||
abe529af | 6507 | static void |
f25d0cf3 | 6508 | xlate_output_action(struct action_xlate_ctx *ctx, |
1688c479 | 6509 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af BP |
6510 | { |
6511 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
6512 | ||
6513 | ctx->nf_output_iface = NF_OUT_DROP; | |
6514 | ||
6515 | switch (port) { | |
6516 | case OFPP_IN_PORT: | |
81b1afb1 | 6517 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
6518 | break; |
6519 | case OFPP_TABLE: | |
1688c479 | 6520 | xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); |
abe529af BP |
6521 | break; |
6522 | case OFPP_NORMAL: | |
6523 | xlate_normal(ctx); | |
6524 | break; | |
6525 | case OFPP_FLOOD: | |
d59906fb | 6526 | flood_packets(ctx, false); |
abe529af BP |
6527 | break; |
6528 | case OFPP_ALL: | |
d59906fb | 6529 | flood_packets(ctx, true); |
abe529af BP |
6530 | break; |
6531 | case OFPP_CONTROLLER: | |
a7349929 | 6532 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 6533 | break; |
e81d2933 EJ |
6534 | case OFPP_NONE: |
6535 | break; | |
a0fbe94a | 6536 | case OFPP_LOCAL: |
abe529af BP |
6537 | default: |
6538 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 6539 | compose_output_action(ctx, port); |
3dd3eace BP |
6540 | } else { |
6541 | xlate_report(ctx, "skipping output to input port"); | |
abe529af BP |
6542 | } |
6543 | break; | |
6544 | } | |
6545 | ||
6546 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
6547 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6548 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
6549 | ctx->nf_output_iface = prev_nf_output_iface; | |
6550 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
6551 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
6552 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6553 | } | |
6554 | } | |
6555 | ||
f694937d EJ |
6556 | static void |
6557 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6558 | const struct ofpact_output_reg *or) |
f694937d | 6559 | { |
f25d0cf3 BP |
6560 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); |
6561 | if (port <= UINT16_MAX) { | |
1688c479 | 6562 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
6563 | } |
6564 | } | |
6565 | ||
abe529af BP |
6566 | static void |
6567 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6568 | const struct ofpact_enqueue *enqueue) |
abe529af | 6569 | { |
f25d0cf3 BP |
6570 | uint16_t ofp_port = enqueue->port; |
6571 | uint32_t queue_id = enqueue->queue; | |
abff858b | 6572 | uint32_t flow_priority, priority; |
abe529af BP |
6573 | int error; |
6574 | ||
f25d0cf3 | 6575 | /* Translate queue to priority. */ |
acf60855 JP |
6576 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6577 | queue_id, &priority); | |
abe529af BP |
6578 | if (error) { |
6579 | /* Fall back to ordinary output action. */ | |
1688c479 | 6580 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
6581 | return; |
6582 | } | |
6583 | ||
f25d0cf3 | 6584 | /* Check output port. */ |
abe529af BP |
6585 | if (ofp_port == OFPP_IN_PORT) { |
6586 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
6587 | } else if (ofp_port == ctx->flow.in_port) { |
6588 | return; | |
abe529af | 6589 | } |
abe529af | 6590 | |
df2c07f4 | 6591 | /* Add datapath actions. */ |
deedf7e7 BP |
6592 | flow_priority = ctx->flow.skb_priority; |
6593 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 6594 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 6595 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
6596 | |
6597 | /* Update NetFlow output port. */ | |
6598 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 6599 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
6600 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
6601 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6602 | } | |
6603 | } | |
6604 | ||
6605 | static void | |
f25d0cf3 | 6606 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 6607 | { |
f25d0cf3 | 6608 | uint32_t skb_priority; |
abe529af | 6609 | |
acf60855 JP |
6610 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6611 | queue_id, &skb_priority)) { | |
f25d0cf3 BP |
6612 | ctx->flow.skb_priority = skb_priority; |
6613 | } else { | |
6614 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 6615 | * has already been logged. */ |
abe529af | 6616 | } |
abe529af BP |
6617 | } |
6618 | ||
daff3353 EJ |
6619 | static bool |
6620 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6621 | { | |
6622 | struct ofproto_dpif *ofproto = ofproto_; | |
6623 | struct ofport_dpif *port; | |
6624 | ||
6625 | switch (ofp_port) { | |
6626 | case OFPP_IN_PORT: | |
6627 | case OFPP_TABLE: | |
6628 | case OFPP_NORMAL: | |
6629 | case OFPP_FLOOD: | |
6630 | case OFPP_ALL: | |
439e4d8c | 6631 | case OFPP_NONE: |
daff3353 EJ |
6632 | return true; |
6633 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6634 | return false; | |
6635 | default: | |
6636 | port = get_ofp_port(ofproto, ofp_port); | |
6637 | return port ? port->may_enable : false; | |
6638 | } | |
6639 | } | |
6640 | ||
f25d0cf3 BP |
6641 | static void |
6642 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
6643 | const struct ofpact_bundle *bundle) | |
6644 | { | |
6645 | uint16_t port; | |
6646 | ||
6647 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
6648 | if (bundle->dst.field) { | |
6649 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
6650 | } else { | |
1688c479 | 6651 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
6652 | } |
6653 | } | |
6654 | ||
75a75043 BP |
6655 | static void |
6656 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6657 | const struct ofpact_learn *learn) |
75a75043 BP |
6658 | { |
6659 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6660 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
6661 | uint64_t ofpacts_stub[1024 / 8]; |
6662 | struct ofpbuf ofpacts; | |
75a75043 BP |
6663 | int error; |
6664 | ||
f25d0cf3 BP |
6665 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
6666 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
75a75043 BP |
6667 | |
6668 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6669 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
6670 | VLOG_WARN("learning action failed to modify flow table (%s)", |
6671 | ofperr_get_name(error)); | |
75a75043 BP |
6672 | } |
6673 | ||
f25d0cf3 | 6674 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
6675 | } |
6676 | ||
0e553d9c BP |
6677 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
6678 | * means "infinite". */ | |
6679 | static void | |
6680 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6681 | { | |
6682 | if (max && (!*timeout || *timeout > max)) { | |
6683 | *timeout = max; | |
6684 | } | |
6685 | } | |
6686 | ||
6687 | static void | |
6688 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6689 | const struct ofpact_fin_timeout *oft) |
0e553d9c BP |
6690 | { |
6691 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
6692 | struct rule_dpif *rule = ctx->rule; | |
6693 | ||
f25d0cf3 BP |
6694 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
6695 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
6696 | } |
6697 | } | |
6698 | ||
29089a54 RL |
6699 | static void |
6700 | xlate_sample_action(struct action_xlate_ctx *ctx, | |
6701 | const struct ofpact_sample *os) | |
6702 | { | |
6703 | union user_action_cookie cookie; | |
6704 | /* Scale the probability from 16-bit to 32-bit while representing | |
6705 | * the same percentage. */ | |
6706 | uint32_t probability = (os->probability << 16) | os->probability; | |
6707 | ||
6708 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); | |
6709 | ||
6710 | compose_flow_sample_cookie(os->probability, os->collector_set_id, | |
6711 | os->obs_domain_id, os->obs_point_id, &cookie); | |
6712 | compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, | |
6713 | probability, &cookie, sizeof cookie.flow_sample); | |
6714 | } | |
6715 | ||
21f7563c JP |
6716 | static bool |
6717 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
6718 | { | |
9e1fd49b BP |
6719 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
6720 | ? OFPUTIL_PC_NO_RECV_STP | |
6721 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
6722 | return false; |
6723 | } | |
6724 | ||
6725 | /* Only drop packets here if both forwarding and learning are | |
6726 | * disabled. If just learning is enabled, we need to have | |
6727 | * OFPP_NORMAL and the learning action have a look at the packet | |
6728 | * before we can drop it. */ | |
6729 | if (!stp_forward_in_state(port->stp_state) | |
6730 | && !stp_learn_in_state(port->stp_state)) { | |
6731 | return false; | |
6732 | } | |
6733 | ||
6734 | return true; | |
6735 | } | |
6736 | ||
4863c249 JP |
6737 | static bool |
6738 | tunnel_ecn_ok(struct action_xlate_ctx *ctx) | |
6739 | { | |
6740 | if (is_ip_any(&ctx->base_flow) | |
715b48aa | 6741 | && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { |
29a5df0a JP |
6742 | if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { |
6743 | VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" | |
6744 | " but is not ECN capable"); | |
6745 | return false; | |
6746 | } else { | |
6747 | /* Set the ECN CE value in the tunneled packet. */ | |
6748 | ctx->flow.nw_tos |= IP_ECN_CE; | |
6749 | } | |
4863c249 JP |
6750 | } |
6751 | ||
6752 | return true; | |
6753 | } | |
6754 | ||
abe529af | 6755 | static void |
f25d0cf3 | 6756 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
abe529af BP |
6757 | struct action_xlate_ctx *ctx) |
6758 | { | |
254750ce | 6759 | bool was_evictable = true; |
f25d0cf3 | 6760 | const struct ofpact *a; |
abe529af | 6761 | |
254750ce BP |
6762 | if (ctx->rule) { |
6763 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6764 | was_evictable = ctx->rule->up.evictable; | |
6765 | ctx->rule->up.evictable = false; | |
6766 | } | |
55599423 JR |
6767 | |
6768 | do_xlate_actions_again: | |
f25d0cf3 BP |
6769 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
6770 | struct ofpact_controller *controller; | |
4cceacb9 | 6771 | const struct ofpact_metadata *metadata; |
38f2e360 | 6772 | |
848e8809 EJ |
6773 | if (ctx->exit) { |
6774 | break; | |
6775 | } | |
6776 | ||
f25d0cf3 BP |
6777 | switch (a->type) { |
6778 | case OFPACT_OUTPUT: | |
6779 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 6780 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
6781 | break; |
6782 | ||
6783 | case OFPACT_CONTROLLER: | |
6784 | controller = ofpact_get_CONTROLLER(a); | |
6785 | execute_controller_action(ctx, controller->max_len, | |
6786 | controller->reason, | |
6787 | controller->controller_id); | |
6788 | break; | |
690a61c5 | 6789 | |
f25d0cf3 BP |
6790 | case OFPACT_ENQUEUE: |
6791 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
6792 | break; |
6793 | ||
f25d0cf3 | 6794 | case OFPACT_SET_VLAN_VID: |
abe529af | 6795 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
f25d0cf3 BP |
6796 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) |
6797 | | htons(VLAN_CFI)); | |
abe529af BP |
6798 | break; |
6799 | ||
f25d0cf3 | 6800 | case OFPACT_SET_VLAN_PCP: |
abe529af | 6801 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
f25d0cf3 BP |
6802 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp |
6803 | << VLAN_PCP_SHIFT) | |
6804 | | VLAN_CFI); | |
abe529af BP |
6805 | break; |
6806 | ||
f25d0cf3 | 6807 | case OFPACT_STRIP_VLAN: |
abe529af | 6808 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
6809 | break; |
6810 | ||
3e34fbdd | 6811 | case OFPACT_PUSH_VLAN: |
5dca28b5 | 6812 | /* XXX 802.1AD(QinQ) */ |
3e34fbdd IY |
6813 | ctx->flow.vlan_tci = htons(VLAN_CFI); |
6814 | break; | |
6815 | ||
f25d0cf3 BP |
6816 | case OFPACT_SET_ETH_SRC: |
6817 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
6818 | ETH_ADDR_LEN); | |
abe529af BP |
6819 | break; |
6820 | ||
f25d0cf3 BP |
6821 | case OFPACT_SET_ETH_DST: |
6822 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
6823 | ETH_ADDR_LEN); | |
abe529af BP |
6824 | break; |
6825 | ||
f25d0cf3 | 6826 | case OFPACT_SET_IPV4_SRC: |
1b035ef2 SH |
6827 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { |
6828 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
6829 | } | |
abe529af BP |
6830 | break; |
6831 | ||
f25d0cf3 | 6832 | case OFPACT_SET_IPV4_DST: |
1b035ef2 SH |
6833 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { |
6834 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
6835 | } | |
abe529af BP |
6836 | break; |
6837 | ||
f25d0cf3 | 6838 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d PS |
6839 | /* OpenFlow 1.0 only supports IPv4. */ |
6840 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6841 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
f25d0cf3 | 6842 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; |
c4f2731d | 6843 | } |
abe529af BP |
6844 | break; |
6845 | ||
f25d0cf3 | 6846 | case OFPACT_SET_L4_SRC_PORT: |
1b035ef2 SH |
6847 | if (is_ip_any(&ctx->flow)) { |
6848 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
6849 | } | |
abe529af BP |
6850 | break; |
6851 | ||
f25d0cf3 | 6852 | case OFPACT_SET_L4_DST_PORT: |
1b035ef2 SH |
6853 | if (is_ip_any(&ctx->flow)) { |
6854 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
6855 | } | |
abe529af BP |
6856 | break; |
6857 | ||
f25d0cf3 BP |
6858 | case OFPACT_RESUBMIT: |
6859 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
6860 | break; |
6861 | ||
f25d0cf3 | 6862 | case OFPACT_SET_TUNNEL: |
296e07ac | 6863 | ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); |
29901626 BP |
6864 | break; |
6865 | ||
f25d0cf3 BP |
6866 | case OFPACT_SET_QUEUE: |
6867 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
6868 | break; |
6869 | ||
f25d0cf3 | 6870 | case OFPACT_POP_QUEUE: |
deedf7e7 | 6871 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
6872 | break; |
6873 | ||
f25d0cf3 BP |
6874 | case OFPACT_REG_MOVE: |
6875 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
38f2e360 BP |
6876 | break; |
6877 | ||
f25d0cf3 BP |
6878 | case OFPACT_REG_LOAD: |
6879 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
38f2e360 BP |
6880 | break; |
6881 | ||
bd85dac1 AZ |
6882 | case OFPACT_STACK_PUSH: |
6883 | nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow, | |
6884 | &ctx->stack); | |
6885 | break; | |
6886 | ||
6887 | case OFPACT_STACK_POP: | |
6888 | nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow, | |
6889 | &ctx->stack); | |
6890 | break; | |
6891 | ||
b02475c5 SH |
6892 | case OFPACT_PUSH_MPLS: |
6893 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6894 | break; | |
6895 | ||
6896 | case OFPACT_POP_MPLS: | |
6897 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6898 | break; | |
6899 | ||
0f3f3c3d SH |
6900 | case OFPACT_SET_MPLS_TTL: |
6901 | if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) { | |
6902 | goto out; | |
6903 | } | |
6904 | break; | |
6905 | ||
b676167a SH |
6906 | case OFPACT_DEC_MPLS_TTL: |
6907 | if (execute_dec_mpls_ttl_action(ctx)) { | |
6908 | goto out; | |
6909 | } | |
6910 | break; | |
6911 | ||
f25d0cf3 | 6912 | case OFPACT_DEC_TTL: |
c2d967a5 | 6913 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
6914 | goto out; |
6915 | } | |
38f2e360 BP |
6916 | break; |
6917 | ||
f25d0cf3 BP |
6918 | case OFPACT_NOTE: |
6919 | /* Nothing to do. */ | |
abe529af BP |
6920 | break; |
6921 | ||
f25d0cf3 BP |
6922 | case OFPACT_MULTIPATH: |
6923 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
abe529af | 6924 | break; |
daff3353 | 6925 | |
f25d0cf3 | 6926 | case OFPACT_BUNDLE: |
a368bb53 | 6927 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 6928 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 6929 | break; |
f694937d | 6930 | |
f25d0cf3 BP |
6931 | case OFPACT_OUTPUT_REG: |
6932 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 6933 | break; |
75a75043 | 6934 | |
f25d0cf3 | 6935 | case OFPACT_LEARN: |
75a75043 | 6936 | ctx->has_learn = true; |
3de9590b | 6937 | if (ctx->may_learn) { |
f25d0cf3 | 6938 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
6939 | } |
6940 | break; | |
848e8809 | 6941 | |
f25d0cf3 | 6942 | case OFPACT_EXIT: |
848e8809 EJ |
6943 | ctx->exit = true; |
6944 | break; | |
0e553d9c | 6945 | |
f25d0cf3 | 6946 | case OFPACT_FIN_TIMEOUT: |
0e553d9c | 6947 | ctx->has_fin_timeout = true; |
f25d0cf3 | 6948 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 6949 | break; |
8dd54666 | 6950 | |
b19e8793 | 6951 | case OFPACT_CLEAR_ACTIONS: |
5dca28b5 | 6952 | /* XXX |
b19e8793 IY |
6953 | * Nothing to do because writa-actions is not supported for now. |
6954 | * When writa-actions is supported, clear-actions also must | |
6955 | * be supported at the same time. | |
6956 | */ | |
6957 | break; | |
6958 | ||
4cceacb9 JS |
6959 | case OFPACT_WRITE_METADATA: |
6960 | metadata = ofpact_get_WRITE_METADATA(a); | |
6961 | ctx->flow.metadata &= ~metadata->mask; | |
6962 | ctx->flow.metadata |= metadata->metadata & metadata->mask; | |
6963 | break; | |
6964 | ||
8dd54666 | 6965 | case OFPACT_GOTO_TABLE: { |
55599423 | 6966 | /* It is assumed that goto-table is the last action. */ |
8dd54666 | 6967 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); |
55599423 JR |
6968 | struct rule_dpif *rule; |
6969 | ||
cb22974d | 6970 | ovs_assert(ctx->table_id < ogt->table_id); |
55599423 JR |
6971 | |
6972 | ctx->table_id = ogt->table_id; | |
6973 | ||
6974 | /* Look up a flow from the new table. */ | |
6975 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id); | |
6976 | ||
6977 | tag_the_flow(ctx, rule); | |
6978 | ||
6979 | rule = ctx_rule_hooks(ctx, rule, true); | |
6980 | ||
6981 | if (rule) { | |
6982 | if (ctx->rule) { | |
6983 | ctx->rule->up.evictable = was_evictable; | |
6984 | } | |
6985 | ctx->rule = rule; | |
6986 | was_evictable = rule->up.evictable; | |
6987 | rule->up.evictable = false; | |
6988 | ||
6989 | /* Tail recursion removal. */ | |
6990 | ofpacts = rule->up.ofpacts; | |
6991 | ofpacts_len = rule->up.ofpacts_len; | |
6992 | goto do_xlate_actions_again; | |
6993 | } | |
8dd54666 IY |
6994 | break; |
6995 | } | |
29089a54 RL |
6996 | |
6997 | case OFPACT_SAMPLE: | |
6998 | xlate_sample_action(ctx, ofpact_get_SAMPLE(a)); | |
6999 | break; | |
abe529af BP |
7000 | } |
7001 | } | |
21f7563c | 7002 | |
f0fd1a17 | 7003 | out: |
254750ce BP |
7004 | if (ctx->rule) { |
7005 | ctx->rule->up.evictable = was_evictable; | |
7006 | } | |
abe529af BP |
7007 | } |
7008 | ||
7009 | static void | |
7010 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
7011 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
7012 | const struct initial_vals *initial_vals, |
7013 | struct rule_dpif *rule, | |
0e553d9c | 7014 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af | 7015 | { |
ef506a7c JG |
7016 | /* Flow initialization rules: |
7017 | * - 'base_flow' must match the kernel's view of the packet at the | |
7018 | * time that action processing starts. 'flow' represents any | |
7019 | * transformations we wish to make through actions. | |
7020 | * - By default 'base_flow' and 'flow' are the same since the input | |
7021 | * packet matches the output before any actions are applied. | |
7022 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
7023 | * of the received packet as seen by the kernel. If we later output | |
7024 | * to another device without any modifications this will cause us to | |
7025 | * insert a new tag since the original one was stripped off by the | |
7026 | * VLAN device. | |
4110a57a JR |
7027 | * - Tunnel metadata as received is retained in 'flow'. This allows |
7028 | * tunnel metadata matching also in later tables. | |
7029 | * Since a kernel action for setting the tunnel metadata will only be | |
7030 | * generated with actual tunnel output, changing the tunnel metadata | |
7031 | * values in 'flow' (such as tun_id) will only have effect with a later | |
7032 | * tunnel output action. | |
ef506a7c JG |
7033 | * - Tunnel 'base_flow' is completely cleared since that is what the |
7034 | * kernel does. If we wish to maintain the original values an action | |
7035 | * needs to be generated. */ | |
7036 | ||
abe529af BP |
7037 | ctx->ofproto = ofproto; |
7038 | ctx->flow = *flow; | |
e84173dc | 7039 | ctx->base_flow = ctx->flow; |
4110a57a | 7040 | memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel); |
0ad90c84 | 7041 | ctx->orig_tunnel_ip_dst = flow->tunnel.ip_dst; |
18b2a258 | 7042 | ctx->rule = rule; |
abe529af | 7043 | ctx->packet = packet; |
3de9590b | 7044 | ctx->may_learn = packet != NULL; |
0e553d9c | 7045 | ctx->tcp_flags = tcp_flags; |
abe529af | 7046 | ctx->resubmit_hook = NULL; |
479df176 | 7047 | ctx->report_hook = NULL; |
112bc5f4 | 7048 | ctx->resubmit_stats = NULL; |
0f49659a EJ |
7049 | |
7050 | if (initial_vals) { | |
7051 | ctx->base_flow.vlan_tci = initial_vals->vlan_tci; | |
7052 | } | |
abe529af BP |
7053 | } |
7054 | ||
f25d0cf3 BP |
7055 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' |
7056 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
050ac423 | 7057 | static void |
abe529af | 7058 | xlate_actions(struct action_xlate_ctx *ctx, |
f25d0cf3 | 7059 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 | 7060 | struct ofpbuf *odp_actions) |
abe529af | 7061 | { |
43d50bc8 BP |
7062 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so |
7063 | * that in the future we always keep a copy of the original flow for | |
7064 | * tracing purposes. */ | |
7065 | static bool hit_resubmit_limit; | |
7066 | ||
6a7e895f | 7067 | enum slow_path_reason special; |
ffaef958 | 7068 | struct ofport_dpif *in_port; |
9ba85077 | 7069 | struct flow orig_flow; |
6a7e895f | 7070 | |
abe529af BP |
7071 | COVERAGE_INC(ofproto_dpif_xlate); |
7072 | ||
050ac423 BP |
7073 | ofpbuf_clear(odp_actions); |
7074 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
7075 | ||
7076 | ctx->odp_actions = odp_actions; | |
97e42c92 | 7077 | ctx->tags = 0; |
6a7e895f | 7078 | ctx->slow = 0; |
97e42c92 BP |
7079 | ctx->has_learn = false; |
7080 | ctx->has_normal = false; | |
0e553d9c | 7081 | ctx->has_fin_timeout = false; |
97e42c92 | 7082 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 7083 | ctx->mirrors = 0; |
97e42c92 | 7084 | ctx->recurse = 0; |
6a6455e5 | 7085 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 7086 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 7087 | ctx->table_id = 0; |
848e8809 | 7088 | ctx->exit = false; |
7257b535 | 7089 | |
bd85dac1 AZ |
7090 | ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack); |
7091 | ||
43d50bc8 | 7092 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 | 7093 | /* Do this conditionally because the copy is expensive enough that it |
9ba85077 BP |
7094 | * shows up in profiles. */ |
7095 | orig_flow = ctx->flow; | |
ccb7c863 BP |
7096 | } |
7097 | ||
eadef313 | 7098 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
7099 | switch (ctx->ofproto->up.frag_handling) { |
7100 | case OFPC_FRAG_NORMAL: | |
7101 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
7102 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
7103 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
7104 | break; |
7105 | ||
7106 | case OFPC_FRAG_DROP: | |
050ac423 | 7107 | return; |
7257b535 BP |
7108 | |
7109 | case OFPC_FRAG_REASM: | |
7110 | NOT_REACHED(); | |
7111 | ||
7112 | case OFPC_FRAG_NX_MATCH: | |
7113 | /* Nothing to do. */ | |
7114 | break; | |
f0fd1a17 PS |
7115 | |
7116 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
7117 | NOT_REACHED(); | |
7257b535 BP |
7118 | } |
7119 | } | |
7120 | ||
ffaef958 BP |
7121 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
7122 | special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet); | |
6a7e895f BP |
7123 | if (special) { |
7124 | ctx->slow |= special; | |
abe529af | 7125 | } else { |
6a6455e5 | 7126 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
14f94f9a | 7127 | struct initial_vals initial_vals; |
94aa0d19 | 7128 | size_t sample_actions_len; |
ee382d89 | 7129 | uint32_t local_odp_port; |
6a6455e5 | 7130 | |
14f94f9a JP |
7131 | initial_vals.vlan_tci = ctx->base_flow.vlan_tci; |
7132 | ||
6ff686f2 | 7133 | add_sflow_action(ctx); |
29089a54 | 7134 | add_ipfix_action(ctx); |
94aa0d19 | 7135 | sample_actions_len = ctx->odp_actions->size; |
ffaef958 | 7136 | |
4863c249 | 7137 | if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) { |
ffaef958 BP |
7138 | do_xlate_actions(ofpacts, ofpacts_len, ctx); |
7139 | ||
7140 | /* We've let OFPP_NORMAL and the learning action look at the | |
7141 | * packet, so drop it now if forwarding is disabled. */ | |
7142 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
94aa0d19 | 7143 | ctx->odp_actions->size = sample_actions_len; |
ffaef958 BP |
7144 | } |
7145 | } | |
abe529af | 7146 | |
43d50bc8 BP |
7147 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { |
7148 | if (!hit_resubmit_limit) { | |
7149 | /* We didn't record the original flow. Make sure we do from | |
7150 | * now on. */ | |
7151 | hit_resubmit_limit = true; | |
7152 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
7153 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7154 | ||
9ba85077 | 7155 | ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet, |
14f94f9a | 7156 | &initial_vals, &ds); |
43d50bc8 BP |
7157 | VLOG_ERR("Trace triggered by excessive resubmit " |
7158 | "recursion:\n%s", ds_cstr(&ds)); | |
7159 | ds_destroy(&ds); | |
7160 | } | |
6a6455e5 EJ |
7161 | } |
7162 | ||
ee382d89 | 7163 | local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); |
b6848f13 | 7164 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
ee382d89 | 7165 | local_odp_port, |
b6848f13 BP |
7166 | ctx->odp_actions->data, |
7167 | ctx->odp_actions->size)) { | |
6a7e895f | 7168 | ctx->slow |= SLOW_IN_BAND; |
b6848f13 BP |
7169 | if (ctx->packet |
7170 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
7171 | ctx->packet)) { | |
5e48dc2b | 7172 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
7173 | } |
7174 | } | |
ccb7c863 | 7175 | if (ctx->ofproto->has_mirrors) { |
9ba85077 | 7176 | add_mirror_actions(ctx, &orig_flow); |
ccb7c863 | 7177 | } |
a7c4eaf6 | 7178 | fix_sflow_action(ctx); |
abe529af | 7179 | } |
bd85dac1 AZ |
7180 | |
7181 | ofpbuf_uninit(&ctx->stack); | |
050ac423 BP |
7182 | } |
7183 | ||
f25d0cf3 BP |
7184 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
7185 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 BP |
7186 | static void |
7187 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
f25d0cf3 BP |
7188 | const struct ofpact *ofpacts, |
7189 | size_t ofpacts_len) | |
050ac423 BP |
7190 | { |
7191 | uint64_t odp_actions_stub[1024 / 8]; | |
7192 | struct ofpbuf odp_actions; | |
abe529af | 7193 | |
050ac423 | 7194 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
f25d0cf3 | 7195 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); |
050ac423 | 7196 | ofpbuf_uninit(&odp_actions); |
abe529af | 7197 | } |
479df176 BP |
7198 | |
7199 | static void | |
7200 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
7201 | { | |
7202 | if (ctx->report_hook) { | |
7203 | ctx->report_hook(ctx, s); | |
7204 | } | |
7205 | } | |
abe529af BP |
7206 | \f |
7207 | /* OFPP_NORMAL implementation. */ | |
7208 | ||
abe529af BP |
7209 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
7210 | ||
ecac4ebf BP |
7211 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
7212 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
7213 | * the bundle on which the packet was received, returns the VLAN to which the | |
7214 | * packet belongs. | |
7215 | * | |
7216 | * Both 'vid' and the return value are in the range 0...4095. */ | |
7217 | static uint16_t | |
7218 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
7219 | { | |
7220 | switch (in_bundle->vlan_mode) { | |
7221 | case PORT_VLAN_ACCESS: | |
7222 | return in_bundle->vlan; | |
7223 | break; | |
7224 | ||
7225 | case PORT_VLAN_TRUNK: | |
7226 | return vid; | |
7227 | ||
7228 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7229 | case PORT_VLAN_NATIVE_TAGGED: | |
7230 | return vid ? vid : in_bundle->vlan; | |
7231 | ||
7232 | default: | |
7233 | NOT_REACHED(); | |
7234 | } | |
7235 | } | |
7236 | ||
5da5ec37 BP |
7237 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
7238 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
7239 | * a warning. | |
7240 | * | |
7241 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
7242 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
7243 | * 0...4095. */ | |
7244 | static bool | |
7245 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
7246 | { | |
33158a18 JP |
7247 | /* Allow any VID on the OFPP_NONE port. */ |
7248 | if (in_bundle == &ofpp_none_bundle) { | |
7249 | return true; | |
7250 | } | |
7251 | ||
5da5ec37 BP |
7252 | switch (in_bundle->vlan_mode) { |
7253 | case PORT_VLAN_ACCESS: | |
7254 | if (vid) { | |
7255 | if (warn) { | |
7256 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7257 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
7258 | "packet received on port %s configured as VLAN " | |
7259 | "%"PRIu16" access port", | |
7260 | in_bundle->ofproto->up.name, vid, | |
7261 | in_bundle->name, in_bundle->vlan); | |
7262 | } | |
7263 | return false; | |
7264 | } | |
7265 | return true; | |
7266 | ||
7267 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7268 | case PORT_VLAN_NATIVE_TAGGED: | |
7269 | if (!vid) { | |
7270 | /* Port must always carry its native VLAN. */ | |
7271 | return true; | |
7272 | } | |
7273 | /* Fall through. */ | |
7274 | case PORT_VLAN_TRUNK: | |
7275 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
7276 | if (warn) { | |
7277 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7278 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
7279 | "received on port %s not configured for trunking " | |
7280 | "VLAN %"PRIu16, | |
7281 | in_bundle->ofproto->up.name, vid, | |
7282 | in_bundle->name, vid); | |
7283 | } | |
7284 | return false; | |
7285 | } | |
7286 | return true; | |
7287 | ||
7288 | default: | |
7289 | NOT_REACHED(); | |
7290 | } | |
7291 | ||
7292 | } | |
7293 | ||
ecac4ebf BP |
7294 | /* Given 'vlan', the VLAN that a packet belongs to, and |
7295 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
7296 | * that should be included in the 802.1Q header. (If the return value is 0, | |
7297 | * then the 802.1Q header should only be included in the packet if there is a | |
7298 | * nonzero PCP.) | |
7299 | * | |
7300 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
7301 | static uint16_t | |
7302 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
7303 | { | |
7304 | switch (out_bundle->vlan_mode) { | |
7305 | case PORT_VLAN_ACCESS: | |
7306 | return 0; | |
7307 | ||
7308 | case PORT_VLAN_TRUNK: | |
7309 | case PORT_VLAN_NATIVE_TAGGED: | |
7310 | return vlan; | |
7311 | ||
7312 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7313 | return vlan == out_bundle->vlan ? 0 : vlan; | |
7314 | ||
7315 | default: | |
7316 | NOT_REACHED(); | |
7317 | } | |
7318 | } | |
7319 | ||
395e68ce BP |
7320 | static void |
7321 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
7322 | uint16_t vlan) | |
abe529af | 7323 | { |
395e68ce BP |
7324 | struct ofport_dpif *port; |
7325 | uint16_t vid; | |
81b1afb1 | 7326 | ovs_be16 tci, old_tci; |
ecac4ebf | 7327 | |
395e68ce BP |
7328 | vid = output_vlan_to_vid(out_bundle, vlan); |
7329 | if (!out_bundle->bond) { | |
7330 | port = ofbundle_get_a_port(out_bundle); | |
7331 | } else { | |
7332 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
7333 | vid, &ctx->tags); | |
7334 | if (!port) { | |
7335 | /* No slaves enabled, so drop packet. */ | |
7336 | return; | |
7337 | } | |
7338 | } | |
abe529af | 7339 | |
81b1afb1 | 7340 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
7341 | tci = htons(vid); |
7342 | if (tci || out_bundle->use_priority_tags) { | |
7343 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
7344 | if (tci) { | |
7345 | tci |= htons(VLAN_CFI); | |
7346 | } | |
395e68ce | 7347 | } |
81b1afb1 | 7348 | ctx->flow.vlan_tci = tci; |
395e68ce | 7349 | |
5e48dc2b | 7350 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 7351 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
7352 | } |
7353 | ||
7354 | static int | |
7355 | mirror_mask_ffs(mirror_mask_t mask) | |
7356 | { | |
7357 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
7358 | return ffs(mask); | |
7359 | } | |
7360 | ||
abe529af BP |
7361 | static bool |
7362 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7363 | { | |
ecac4ebf | 7364 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 7365 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
7366 | } |
7367 | ||
7368 | static bool | |
7369 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7370 | { | |
7371 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
7372 | } | |
7373 | ||
7374 | /* Returns an arbitrary interface within 'bundle'. */ | |
7375 | static struct ofport_dpif * | |
7376 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
7377 | { | |
7378 | return CONTAINER_OF(list_front(&bundle->ports), | |
7379 | struct ofport_dpif, bundle_node); | |
7380 | } | |
7381 | ||
abe529af BP |
7382 | static bool |
7383 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
7384 | { | |
fc3d7408 | 7385 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
7386 | } |
7387 | ||
7388 | static void | |
c06bba01 | 7389 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
7390 | { |
7391 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7392 | mirror_mask_t mirrors; | |
c06bba01 JP |
7393 | struct ofbundle *in_bundle; |
7394 | uint16_t vlan; | |
7395 | uint16_t vid; | |
7396 | const struct nlattr *a; | |
7397 | size_t left; | |
7398 | ||
3581c12c | 7399 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
70c2fd56 | 7400 | ctx->packet != NULL, NULL); |
3581c12c | 7401 | if (!in_bundle) { |
c06bba01 JP |
7402 | return; |
7403 | } | |
c06bba01 JP |
7404 | mirrors = in_bundle->src_mirrors; |
7405 | ||
7406 | /* Drop frames on bundles reserved for mirroring. */ | |
7407 | if (in_bundle->mirror_out) { | |
7408 | if (ctx->packet != NULL) { | |
7409 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7410 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7411 | "%s, which is reserved exclusively for mirroring", | |
7412 | ctx->ofproto->up.name, in_bundle->name); | |
7413 | } | |
7414 | return; | |
7415 | } | |
7416 | ||
7417 | /* Check VLAN. */ | |
7418 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
7419 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
7420 | return; | |
7421 | } | |
7422 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7423 | ||
7424 | /* Look at the output ports to check for destination selections. */ | |
7425 | ||
7426 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
7427 | ctx->odp_actions->size) { | |
7428 | enum ovs_action_attr type = nl_attr_type(a); | |
7429 | struct ofport_dpif *ofport; | |
7430 | ||
7431 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
7432 | continue; | |
7433 | } | |
7434 | ||
7435 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
7436 | if (ofport && ofport->bundle) { |
7437 | mirrors |= ofport->bundle->dst_mirrors; | |
7438 | } | |
c06bba01 | 7439 | } |
abe529af BP |
7440 | |
7441 | if (!mirrors) { | |
7442 | return; | |
7443 | } | |
7444 | ||
c06bba01 JP |
7445 | /* Restore the original packet before adding the mirror actions. */ |
7446 | ctx->flow = *orig_flow; | |
7447 | ||
9ba15e2a BP |
7448 | while (mirrors) { |
7449 | struct ofmirror *m; | |
9ba15e2a BP |
7450 | |
7451 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7452 | ||
7453 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 7454 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
7455 | continue; |
7456 | } | |
7457 | ||
7458 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 7459 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 7460 | if (m->out) { |
395e68ce | 7461 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
7462 | } else if (vlan != m->out_vlan |
7463 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
7464 | struct ofbundle *bundle; |
7465 | ||
7466 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
7467 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
7468 | && !bundle->mirror_out) { |
7469 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
7470 | } |
7471 | } | |
7472 | } | |
abe529af BP |
7473 | } |
7474 | } | |
7475 | ||
9d24de3b JP |
7476 | static void |
7477 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
7478 | uint64_t packets, uint64_t bytes) | |
7479 | { | |
7480 | if (!mirrors) { | |
7481 | return; | |
7482 | } | |
7483 | ||
8472a3ce | 7484 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
7485 | struct ofmirror *m; |
7486 | ||
7487 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7488 | ||
7489 | if (!m) { | |
7490 | /* In normal circumstances 'm' will not be NULL. However, | |
7491 | * if mirrors are reconfigured, we can temporarily get out | |
7492 | * of sync in facet_revalidate(). We could "correct" the | |
7493 | * mirror list before reaching here, but doing that would | |
7494 | * not properly account the traffic stats we've currently | |
7495 | * accumulated for previous mirror configuration. */ | |
7496 | continue; | |
7497 | } | |
7498 | ||
7499 | m->packet_count += packets; | |
7500 | m->byte_count += bytes; | |
7501 | } | |
7502 | } | |
7503 | ||
abe529af BP |
7504 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
7505 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7506 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7507 | static bool | |
7508 | is_gratuitous_arp(const struct flow *flow) | |
7509 | { | |
7510 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7511 | && eth_addr_is_broadcast(flow->dl_dst) | |
7512 | && (flow->nw_proto == ARP_OP_REPLY | |
7513 | || (flow->nw_proto == ARP_OP_REQUEST | |
7514 | && flow->nw_src == flow->nw_dst))); | |
7515 | } | |
7516 | ||
7517 | static void | |
7518 | update_learning_table(struct ofproto_dpif *ofproto, | |
7519 | const struct flow *flow, int vlan, | |
7520 | struct ofbundle *in_bundle) | |
7521 | { | |
7522 | struct mac_entry *mac; | |
7523 | ||
33158a18 JP |
7524 | /* Don't learn the OFPP_NONE port. */ |
7525 | if (in_bundle == &ofpp_none_bundle) { | |
7526 | return; | |
7527 | } | |
7528 | ||
abe529af BP |
7529 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
7530 | return; | |
7531 | } | |
7532 | ||
7533 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7534 | if (is_gratuitous_arp(flow)) { | |
7535 | /* We don't want to learn from gratuitous ARP packets that are | |
7536 | * reflected back over bond slaves so we lock the learning table. */ | |
7537 | if (!in_bundle->bond) { | |
7538 | mac_entry_set_grat_arp_lock(mac); | |
7539 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7540 | return; | |
7541 | } | |
7542 | } | |
7543 | ||
7544 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7545 | /* The log messages here could actually be useful in debugging, | |
7546 | * so keep the rate limit relatively high. */ | |
7547 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7548 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7549 | "on port %s in VLAN %d", | |
7550 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7551 | in_bundle->name, vlan); | |
7552 | ||
7553 | mac->port.p = in_bundle; | |
2cc3c58e | 7554 | tag_set_add(&ofproto->backer->revalidate_set, |
abe529af BP |
7555 | mac_learning_changed(ofproto->ml, mac)); |
7556 | } | |
7557 | } | |
7558 | ||
3581c12c | 7559 | static struct ofbundle * |
4acbc98d SH |
7560 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
7561 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
7562 | { |
7563 | struct ofport_dpif *ofport; | |
7564 | ||
7565 | /* Find the port and bundle for the received packet. */ | |
7566 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
7567 | if (in_ofportp) { |
7568 | *in_ofportp = ofport; | |
7569 | } | |
395e68ce | 7570 | if (ofport && ofport->bundle) { |
3581c12c | 7571 | return ofport->bundle; |
395e68ce BP |
7572 | } |
7573 | ||
70c2fd56 BP |
7574 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
7575 | * port for traffic that it is sourcing. */ | |
7576 | if (in_port == OFPP_NONE) { | |
7577 | return &ofpp_none_bundle; | |
7578 | } | |
7579 | ||
395e68ce BP |
7580 | /* Odd. A few possible reasons here: |
7581 | * | |
7582 | * - We deleted a port but there are still a few packets queued up | |
7583 | * from it. | |
7584 | * | |
7585 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7586 | * we don't know about. | |
7587 | * | |
7588 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
7589 | * This is particularly likely to happen if a packet was received on the |
7590 | * port after it was created, but before the client had a chance to | |
7591 | * configure its bundle. | |
395e68ce BP |
7592 | */ |
7593 | if (warn) { | |
7594 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7595 | ||
7596 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7597 | "port %"PRIu16, ofproto->up.name, in_port); | |
7598 | } | |
7599 | return NULL; | |
7600 | } | |
7601 | ||
5da5ec37 | 7602 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
7603 | * dropped. Returns true if they may be forwarded, false if they should be |
7604 | * dropped. | |
7605 | * | |
395e68ce BP |
7606 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
7607 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 7608 | * |
395e68ce BP |
7609 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
7610 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7611 | * checked by input_vid_is_valid(). | |
abe529af BP |
7612 | * |
7613 | * May also add tags to '*tags', although the current implementation only does | |
7614 | * so in one special case. | |
7615 | */ | |
7616 | static bool | |
479df176 BP |
7617 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, |
7618 | uint16_t vlan) | |
abe529af | 7619 | { |
479df176 BP |
7620 | struct ofproto_dpif *ofproto = ctx->ofproto; |
7621 | struct flow *flow = &ctx->flow; | |
395e68ce | 7622 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 7623 | |
395e68ce BP |
7624 | /* Drop frames for reserved multicast addresses |
7625 | * only if forward_bpdu option is absent. */ | |
614ec445 | 7626 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 7627 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
7628 | return false; |
7629 | } | |
7630 | ||
abe529af BP |
7631 | if (in_bundle->bond) { |
7632 | struct mac_entry *mac; | |
7633 | ||
7634 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
479df176 | 7635 | flow->dl_dst, &ctx->tags)) { |
abe529af BP |
7636 | case BV_ACCEPT: |
7637 | break; | |
7638 | ||
7639 | case BV_DROP: | |
479df176 | 7640 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
7641 | return false; |
7642 | ||
7643 | case BV_DROP_IF_MOVED: | |
7644 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7645 | if (mac && mac->port.p != in_bundle && | |
7646 | (!is_gratuitous_arp(flow) | |
7647 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
7648 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
7649 | "dropping"); | |
abe529af BP |
7650 | return false; |
7651 | } | |
7652 | break; | |
7653 | } | |
7654 | } | |
7655 | ||
7656 | return true; | |
7657 | } | |
7658 | ||
4cd78906 | 7659 | static void |
abe529af BP |
7660 | xlate_normal(struct action_xlate_ctx *ctx) |
7661 | { | |
395e68ce | 7662 | struct ofport_dpif *in_port; |
abe529af | 7663 | struct ofbundle *in_bundle; |
abe529af | 7664 | struct mac_entry *mac; |
395e68ce BP |
7665 | uint16_t vlan; |
7666 | uint16_t vid; | |
abe529af | 7667 | |
75a75043 BP |
7668 | ctx->has_normal = true; |
7669 | ||
3581c12c | 7670 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
70c2fd56 | 7671 | ctx->packet != NULL, &in_port); |
3581c12c | 7672 | if (!in_bundle) { |
479df176 | 7673 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
7674 | return; |
7675 | } | |
3581c12c | 7676 | |
395e68ce BP |
7677 | /* Drop malformed frames. */ |
7678 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
7679 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
7680 | if (ctx->packet != NULL) { | |
7681 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7682 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7683 | "VLAN tag received on port %s", | |
7684 | ctx->ofproto->up.name, in_bundle->name); | |
7685 | } | |
479df176 | 7686 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
7687 | return; |
7688 | } | |
7689 | ||
7690 | /* Drop frames on bundles reserved for mirroring. */ | |
7691 | if (in_bundle->mirror_out) { | |
7692 | if (ctx->packet != NULL) { | |
7693 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7694 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7695 | "%s, which is reserved exclusively for mirroring", | |
7696 | ctx->ofproto->up.name, in_bundle->name); | |
7697 | } | |
479df176 | 7698 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
7699 | return; |
7700 | } | |
7701 | ||
7702 | /* Check VLAN. */ | |
7703 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
7704 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
479df176 | 7705 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
7706 | return; |
7707 | } | |
7708 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7709 | ||
7710 | /* Check other admissibility requirements. */ | |
479df176 | 7711 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 7712 | return; |
abe529af BP |
7713 | } |
7714 | ||
75a75043 | 7715 | /* Learn source MAC. */ |
3de9590b | 7716 | if (ctx->may_learn) { |
abe529af BP |
7717 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
7718 | } | |
7719 | ||
7720 | /* Determine output bundle. */ | |
7721 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
7722 | &ctx->tags); | |
7723 | if (mac) { | |
c06bba01 | 7724 | if (mac->port.p != in_bundle) { |
479df176 | 7725 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 7726 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
7727 | } else { |
7728 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 7729 | } |
abe529af | 7730 | } else { |
c06bba01 | 7731 | struct ofbundle *bundle; |
abe529af | 7732 | |
479df176 | 7733 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
7734 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
7735 | if (bundle != in_bundle | |
7736 | && ofbundle_includes_vlan(bundle, vlan) | |
7737 | && bundle->floodable | |
7738 | && !bundle->mirror_out) { | |
7739 | output_normal(ctx, bundle, vlan); | |
7740 | } | |
7741 | } | |
7742 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 7743 | } |
abe529af BP |
7744 | } |
7745 | \f | |
54a9cbc9 BP |
7746 | /* Optimized flow revalidation. |
7747 | * | |
7748 | * It's a difficult problem, in general, to tell which facets need to have | |
7749 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7750 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7751 | * table changes, we recalculate the actions for every facet. This is | |
7752 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7753 | * doesn't change very often. | |
7754 | * | |
7755 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7756 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7757 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7758 | * table, we add a special case that applies to flow tables in which every rule | |
7759 | * has the same form (that is, the same wildcards), except that the table is | |
7760 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7761 | * optimize this case by tagging all of the facets that resubmit into the table | |
7762 | * and invalidating the same tag whenever a flow changes in that table. The | |
7763 | * end result is that we revalidate just the facets that need it (and sometimes | |
7764 | * a few more, but not all of the facets or even all of the facets that | |
7765 | * resubmit to the table modified by MAC learning). */ | |
7766 | ||
5cb7a798 | 7767 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted |
54a9cbc9 | 7768 | * into an OpenFlow table with the given 'basis'. */ |
822d9414 | 7769 | static tag_type |
5cb7a798 | 7770 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, |
54a9cbc9 BP |
7771 | uint32_t secret) |
7772 | { | |
5cb7a798 | 7773 | if (minimask_is_catchall(mask)) { |
54a9cbc9 BP |
7774 | return 0; |
7775 | } else { | |
5cb7a798 BP |
7776 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); |
7777 | return tag_create_deterministic(hash); | |
54a9cbc9 BP |
7778 | } |
7779 | } | |
7780 | ||
7781 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7782 | * taggability of that table. | |
7783 | * | |
7784 | * This function must be called after *each* change to a flow table. If you | |
7785 | * skip calling it on some changes then the pointer comparisons at the end can | |
7786 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7787 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7788 | * different wildcards to be created with the same address, then this function | |
7789 | * will incorrectly skip revalidation. */ | |
7790 | static void | |
7791 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7792 | { | |
7793 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 7794 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
7795 | struct cls_table *catchall, *other; |
7796 | struct cls_table *t; | |
7797 | ||
7798 | catchall = other = NULL; | |
7799 | ||
d0918789 | 7800 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
7801 | case 0: |
7802 | /* We could tag this OpenFlow table but it would make the logic a | |
7803 | * little harder and it's a corner case that doesn't seem worth it | |
7804 | * yet. */ | |
7805 | break; | |
7806 | ||
7807 | case 1: | |
7808 | case 2: | |
d0918789 | 7809 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
7810 | if (cls_table_is_catchall(t)) { |
7811 | catchall = t; | |
7812 | } else if (!other) { | |
7813 | other = t; | |
7814 | } else { | |
7815 | /* Indicate that we can't tag this by setting both tables to | |
7816 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7817 | other = NULL; | |
7818 | } | |
7819 | } | |
7820 | break; | |
7821 | ||
7822 | default: | |
7823 | /* Can't tag this table. */ | |
7824 | break; | |
7825 | } | |
7826 | ||
7827 | if (table->catchall_table != catchall || table->other_table != other) { | |
7828 | table->catchall_table = catchall; | |
7829 | table->other_table = other; | |
2cc3c58e | 7830 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7831 | } |
7832 | } | |
7833 | ||
7834 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7835 | * inserted, a rule being deleted, or a rule whose actions are being | |
7836 | * modified), marks facets for revalidation to ensure that packets will be | |
7837 | * forwarded correctly according to the new state of the flow table. | |
7838 | * | |
7839 | * This function must be called after *each* change to a flow table. See | |
7840 | * the comment on table_update_taggable() for more information. */ | |
7841 | static void | |
7842 | rule_invalidate(const struct rule_dpif *rule) | |
7843 | { | |
7844 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7845 | ||
7846 | table_update_taggable(ofproto, rule->up.table_id); | |
7847 | ||
2cc3c58e | 7848 | if (!ofproto->backer->need_revalidate) { |
54a9cbc9 BP |
7849 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; |
7850 | ||
7851 | if (table->other_table && rule->tag) { | |
2cc3c58e | 7852 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); |
54a9cbc9 | 7853 | } else { |
2cc3c58e | 7854 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7855 | } |
7856 | } | |
7857 | } | |
7858 | \f | |
abe529af | 7859 | static bool |
7257b535 BP |
7860 | set_frag_handling(struct ofproto *ofproto_, |
7861 | enum ofp_config_flags frag_handling) | |
abe529af BP |
7862 | { |
7863 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 7864 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 7865 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
7866 | return true; |
7867 | } else { | |
7868 | return false; | |
7869 | } | |
abe529af BP |
7870 | } |
7871 | ||
90bf1e07 | 7872 | static enum ofperr |
abe529af BP |
7873 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
7874 | const struct flow *flow, | |
f25d0cf3 | 7875 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
7876 | { |
7877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
14f94f9a | 7878 | struct initial_vals initial_vals; |
548de4dd BP |
7879 | struct odputil_keybuf keybuf; |
7880 | struct dpif_flow_stats stats; | |
abe529af | 7881 | |
548de4dd | 7882 | struct ofpbuf key; |
112bc5f4 | 7883 | |
548de4dd BP |
7884 | struct action_xlate_ctx ctx; |
7885 | uint64_t odp_actions_stub[1024 / 8]; | |
7886 | struct ofpbuf odp_actions; | |
80e5eed9 | 7887 | |
548de4dd | 7888 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
7889 | odp_flow_key_from_flow(&key, flow, |
7890 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
050ac423 | 7891 | |
548de4dd | 7892 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 7893 | |
14f94f9a JP |
7894 | initial_vals.vlan_tci = flow->vlan_tci; |
7895 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL, | |
548de4dd BP |
7896 | packet_get_tcp_flags(packet, flow), packet); |
7897 | ctx.resubmit_stats = &stats; | |
2284188b | 7898 | |
548de4dd BP |
7899 | ofpbuf_use_stub(&odp_actions, |
7900 | odp_actions_stub, sizeof odp_actions_stub); | |
7901 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
acf60855 | 7902 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
548de4dd BP |
7903 | odp_actions.data, odp_actions.size, packet); |
7904 | ofpbuf_uninit(&odp_actions); | |
2284188b | 7905 | |
548de4dd | 7906 | return 0; |
abe529af | 7907 | } |
6fca1ffb BP |
7908 | \f |
7909 | /* NetFlow. */ | |
7910 | ||
7911 | static int | |
7912 | set_netflow(struct ofproto *ofproto_, | |
7913 | const struct netflow_options *netflow_options) | |
7914 | { | |
7915 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7916 | ||
7917 | if (netflow_options) { | |
7918 | if (!ofproto->netflow) { | |
7919 | ofproto->netflow = netflow_create(); | |
7920 | } | |
7921 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7922 | } else { | |
7923 | netflow_destroy(ofproto->netflow); | |
7924 | ofproto->netflow = NULL; | |
7925 | return 0; | |
7926 | } | |
7927 | } | |
abe529af BP |
7928 | |
7929 | static void | |
7930 | get_netflow_ids(const struct ofproto *ofproto_, | |
7931 | uint8_t *engine_type, uint8_t *engine_id) | |
7932 | { | |
7933 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7934 | ||
acf60855 | 7935 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 7936 | } |
6fca1ffb BP |
7937 | |
7938 | static void | |
7939 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7940 | { | |
7941 | if (!facet_is_controller_flow(facet) && | |
7942 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 7943 | struct subfacet *subfacet; |
6fca1ffb BP |
7944 | struct ofexpired expired; |
7945 | ||
b0f7b9b5 | 7946 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 7947 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 7948 | struct dpif_flow_stats stats; |
6fca1ffb | 7949 | |
6a7e895f | 7950 | subfacet_reinstall(subfacet, &stats); |
15baa734 | 7951 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 7952 | } |
6fca1ffb BP |
7953 | } |
7954 | ||
7955 | expired.flow = facet->flow; | |
7956 | expired.packet_count = facet->packet_count; | |
7957 | expired.byte_count = facet->byte_count; | |
7958 | expired.used = facet->used; | |
7959 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7960 | } | |
7961 | } | |
7962 | ||
7963 | static void | |
7964 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7965 | { | |
7966 | struct facet *facet; | |
7967 | ||
7968 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7969 | send_active_timeout(ofproto, facet); | |
7970 | } | |
7971 | } | |
abe529af BP |
7972 | \f |
7973 | static struct ofproto_dpif * | |
7974 | ofproto_dpif_lookup(const char *name) | |
7975 | { | |
b44a10b7 BP |
7976 | struct ofproto_dpif *ofproto; |
7977 | ||
7978 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7979 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7980 | if (!strcmp(ofproto->up.name, name)) { | |
7981 | return ofproto; | |
7982 | } | |
7983 | } | |
7984 | return NULL; | |
abe529af BP |
7985 | } |
7986 | ||
f0a3aa2e | 7987 | static void |
96e466a3 | 7988 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 7989 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 7990 | { |
490df1ef | 7991 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 7992 | |
96e466a3 EJ |
7993 | if (argc > 1) { |
7994 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7995 | if (!ofproto) { | |
bde9f75d | 7996 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
7997 | return; |
7998 | } | |
2cc3c58e | 7999 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 EJ |
8000 | } else { |
8001 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2cc3c58e | 8002 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 | 8003 | } |
f0a3aa2e | 8004 | } |
f0a3aa2e | 8005 | |
bde9f75d | 8006 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
8007 | } |
8008 | ||
abe529af | 8009 | static void |
0e15264f BP |
8010 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
8011 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
8012 | { |
8013 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8014 | const struct ofproto_dpif *ofproto; | |
8015 | const struct mac_entry *e; | |
8016 | ||
0e15264f | 8017 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 8018 | if (!ofproto) { |
bde9f75d | 8019 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
8020 | return; |
8021 | } | |
8022 | ||
8023 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
8024 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
8025 | struct ofbundle *bundle = e->port.p; | |
8026 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
8027 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
8028 | e->vlan, ETH_ADDR_ARGS(e->mac), |
8029 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 8030 | } |
bde9f75d | 8031 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
8032 | ds_destroy(&ds); |
8033 | } | |
8034 | ||
6a6455e5 | 8035 | struct trace_ctx { |
abe529af BP |
8036 | struct action_xlate_ctx ctx; |
8037 | struct flow flow; | |
8038 | struct ds *result; | |
8039 | }; | |
8040 | ||
8041 | static void | |
29901626 BP |
8042 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
8043 | const struct rule_dpif *rule) | |
abe529af BP |
8044 | { |
8045 | ds_put_char_multiple(result, '\t', level); | |
8046 | if (!rule) { | |
8047 | ds_put_cstr(result, "No match\n"); | |
8048 | return; | |
8049 | } | |
8050 | ||
29901626 BP |
8051 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
8052 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 8053 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
8054 | ds_put_char(result, '\n'); |
8055 | ||
8056 | ds_put_char_multiple(result, '\t', level); | |
8057 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 8058 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
8059 | ds_put_char(result, '\n'); |
8060 | } | |
8061 | ||
8062 | static void | |
8063 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 8064 | struct trace_ctx *trace) |
abe529af BP |
8065 | { |
8066 | ds_put_char_multiple(result, '\t', level); | |
8067 | ds_put_format(result, "%s: ", title); | |
8068 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
8069 | ds_put_cstr(result, "unchanged"); | |
8070 | } else { | |
8071 | flow_format(result, &trace->ctx.flow); | |
8072 | trace->flow = trace->ctx.flow; | |
8073 | } | |
8074 | ds_put_char(result, '\n'); | |
8075 | } | |
8076 | ||
eb9e1c26 EJ |
8077 | static void |
8078 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 8079 | struct trace_ctx *trace) |
eb9e1c26 EJ |
8080 | { |
8081 | size_t i; | |
8082 | ||
8083 | ds_put_char_multiple(result, '\t', level); | |
8084 | ds_put_format(result, "%s:", title); | |
8085 | for (i = 0; i < FLOW_N_REGS; i++) { | |
8086 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
8087 | } | |
8088 | ds_put_char(result, '\n'); | |
8089 | } | |
8090 | ||
1ed8d352 EJ |
8091 | static void |
8092 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 8093 | struct trace_ctx *trace) |
1ed8d352 EJ |
8094 | { |
8095 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
8096 | ||
8097 | ds_put_char_multiple(result, '\t', level); | |
8098 | ds_put_format(result, "%s: ", title); | |
8099 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
8100 | ds_put_char(result, '\n'); | |
8101 | } | |
8102 | ||
abe529af BP |
8103 | static void |
8104 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
8105 | { | |
6a6455e5 | 8106 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
8107 | struct ds *result = trace->result; |
8108 | ||
8109 | ds_put_char(result, '\n'); | |
8110 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 8111 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 8112 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 8113 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
8114 | } |
8115 | ||
479df176 BP |
8116 | static void |
8117 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
8118 | { | |
8119 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
8120 | struct ds *result = trace->result; | |
8121 | ||
8122 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
8123 | ds_put_cstr(result, s); | |
8124 | ds_put_char(result, '\n'); | |
8125 | } | |
8126 | ||
abe529af | 8127 | static void |
0e15264f | 8128 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
8129 | void *aux OVS_UNUSED) |
8130 | { | |
0e15264f | 8131 | const char *dpname = argv[1]; |
abe529af | 8132 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
8133 | struct ofpbuf odp_key; |
8134 | struct ofpbuf *packet; | |
14f94f9a | 8135 | struct initial_vals initial_vals; |
abe529af BP |
8136 | struct ds result; |
8137 | struct flow flow; | |
abe529af BP |
8138 | char *s; |
8139 | ||
876b0e1c BP |
8140 | packet = NULL; |
8141 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
8142 | ds_init(&result); |
8143 | ||
e84173dc BP |
8144 | ofproto = ofproto_dpif_lookup(dpname); |
8145 | if (!ofproto) { | |
bde9f75d EJ |
8146 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
8147 | "for help)"); | |
e84173dc BP |
8148 | goto exit; |
8149 | } | |
0e15264f | 8150 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 8151 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
8152 | const char *flow_s = argv[2]; |
8153 | const char *generate_s = argv[3]; | |
876b0e1c | 8154 | |
31a19d69 BP |
8155 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like |
8156 | * flow. We guess which type it is based on whether 'flow_s' contains | |
8157 | * an '(', since a datapath flow always contains '(') but an | |
8158 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
8159 | * that's not documented anywhere). | |
8160 | * | |
8161 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
8162 | * it would be tricky giving a sensible error message. After all, do | |
8163 | * you just say "syntax error" or do you present both error messages? | |
8164 | * Both choices seem lousy. */ | |
8165 | if (strchr(flow_s, '(')) { | |
8166 | int error; | |
8167 | ||
8168 | /* Convert string to datapath key. */ | |
8169 | ofpbuf_init(&odp_key, 0); | |
8170 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
8171 | if (error) { | |
8172 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
8173 | goto exit; | |
8174 | } | |
876b0e1c | 8175 | |
6d199116 BP |
8176 | /* The user might have specified the wrong ofproto but within the |
8177 | * same backer. That's OK, ofproto_receive() can find the right | |
8178 | * one for us. */ | |
e09ee259 | 8179 | if (ofproto_receive(ofproto->backer, NULL, odp_key.data, |
6d199116 | 8180 | odp_key.size, &flow, NULL, &ofproto, NULL, |
14f94f9a | 8181 | &initial_vals)) { |
31a19d69 BP |
8182 | unixctl_command_reply_error(conn, "Invalid flow"); |
8183 | goto exit; | |
8184 | } | |
6d199116 | 8185 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
31a19d69 BP |
8186 | } else { |
8187 | char *error_s; | |
8188 | ||
8189 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
8190 | if (error_s) { | |
8191 | unixctl_command_reply_error(conn, error_s); | |
8192 | free(error_s); | |
8193 | goto exit; | |
8194 | } | |
8195 | ||
14f94f9a | 8196 | initial_vals.vlan_tci = flow.vlan_tci; |
876b0e1c | 8197 | } |
8b3b8dd1 BP |
8198 | |
8199 | /* Generate a packet, if requested. */ | |
0e15264f | 8200 | if (generate_s) { |
8b3b8dd1 BP |
8201 | packet = ofpbuf_new(0); |
8202 | flow_compose(packet, &flow); | |
8203 | } | |
72e8bf28 AA |
8204 | } else if (argc == 7) { |
8205 | /* ofproto/trace dpname priority tun_id in_port mark packet */ | |
0e15264f BP |
8206 | const char *priority_s = argv[2]; |
8207 | const char *tun_id_s = argv[3]; | |
8208 | const char *in_port_s = argv[4]; | |
72e8bf28 AA |
8209 | const char *mark_s = argv[5]; |
8210 | const char *packet_s = argv[6]; | |
9b56fe13 | 8211 | uint32_t in_port = atoi(in_port_s); |
0e15264f BP |
8212 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); |
8213 | uint32_t priority = atoi(priority_s); | |
72e8bf28 | 8214 | uint32_t mark = atoi(mark_s); |
e22f1753 | 8215 | const char *msg; |
0e15264f | 8216 | |
e22f1753 BP |
8217 | msg = eth_from_hex(packet_s, &packet); |
8218 | if (msg) { | |
bde9f75d | 8219 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
8220 | goto exit; |
8221 | } | |
8222 | ||
8223 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 8224 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
8225 | ds_put_cstr(&result, s); |
8226 | free(s); | |
8227 | ||
72e8bf28 | 8228 | flow_extract(packet, priority, mark, NULL, in_port, &flow); |
296e07ac | 8229 | flow.tunnel.tun_id = tun_id; |
14f94f9a | 8230 | initial_vals.vlan_tci = flow.vlan_tci; |
876b0e1c | 8231 | } else { |
bde9f75d | 8232 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
8233 | goto exit; |
8234 | } | |
8235 | ||
14f94f9a | 8236 | ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); |
6a6455e5 EJ |
8237 | unixctl_command_reply(conn, ds_cstr(&result)); |
8238 | ||
8239 | exit: | |
8240 | ds_destroy(&result); | |
8241 | ofpbuf_delete(packet); | |
8242 | ofpbuf_uninit(&odp_key); | |
8243 | } | |
8244 | ||
8245 | static void | |
8246 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
8247 | const struct ofpbuf *packet, |
8248 | const struct initial_vals *initial_vals, struct ds *ds) | |
6a6455e5 EJ |
8249 | { |
8250 | struct rule_dpif *rule; | |
8251 | ||
8252 | ds_put_cstr(ds, "Flow: "); | |
8253 | flow_format(ds, flow); | |
8254 | ds_put_char(ds, '\n'); | |
abe529af | 8255 | |
c57b2226 BP |
8256 | rule = rule_dpif_lookup(ofproto, flow); |
8257 | ||
6a6455e5 | 8258 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
8259 | if (rule == ofproto->miss_rule) { |
8260 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
8261 | } else if (rule == ofproto->no_packet_in_rule) { | |
8262 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
8263 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
8264 | } | |
8265 | ||
abe529af | 8266 | if (rule) { |
050ac423 BP |
8267 | uint64_t odp_actions_stub[1024 / 8]; |
8268 | struct ofpbuf odp_actions; | |
8269 | ||
6a6455e5 | 8270 | struct trace_ctx trace; |
0e553d9c | 8271 | uint8_t tcp_flags; |
abe529af | 8272 | |
6a6455e5 EJ |
8273 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
8274 | trace.result = ds; | |
8275 | trace.flow = *flow; | |
050ac423 BP |
8276 | ofpbuf_use_stub(&odp_actions, |
8277 | odp_actions_stub, sizeof odp_actions_stub); | |
14f94f9a | 8278 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals, |
0e553d9c | 8279 | rule, tcp_flags, packet); |
abe529af | 8280 | trace.ctx.resubmit_hook = trace_resubmit; |
479df176 | 8281 | trace.ctx.report_hook = trace_report; |
f25d0cf3 | 8282 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 8283 | &odp_actions); |
abe529af | 8284 | |
6a6455e5 EJ |
8285 | ds_put_char(ds, '\n'); |
8286 | trace_format_flow(ds, 0, "Final flow", &trace); | |
8287 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
8288 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
8289 | ofpbuf_uninit(&odp_actions); | |
876b0e1c | 8290 | |
6a7e895f BP |
8291 | if (trace.ctx.slow) { |
8292 | enum slow_path_reason slow; | |
8293 | ||
8294 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
8295 | "slow path because it:"); | |
8296 | for (slow = trace.ctx.slow; slow; ) { | |
8297 | enum slow_path_reason bit = rightmost_1bit(slow); | |
8298 | ||
8299 | switch (bit) { | |
8300 | case SLOW_CFM: | |
8301 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
8302 | break; | |
8303 | case SLOW_LACP: | |
8304 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
8305 | break; | |
8306 | case SLOW_STP: | |
8307 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
8308 | break; | |
ccc09689 EJ |
8309 | case SLOW_BFD: |
8310 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
8311 | break; | |
6a7e895f BP |
8312 | case SLOW_IN_BAND: |
8313 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
8314 | "processing."); | |
8315 | if (!packet) { | |
8316 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
8317 | "incomplete--for complete actions, " | |
8318 | "please supply a packet.)"); | |
8319 | } | |
8320 | break; | |
8321 | case SLOW_CONTROLLER: | |
8322 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
8323 | "to the OpenFlow controller."); | |
8324 | break; | |
8325 | case SLOW_MATCH: | |
8326 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
8327 | "than the datapath supports."); | |
8328 | break; | |
8329 | } | |
8330 | ||
8331 | slow &= ~bit; | |
8332 | } | |
8333 | ||
8334 | if (slow & ~SLOW_MATCH) { | |
8335 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
8336 | "the special slow-path processing."); | |
876b0e1c BP |
8337 | } |
8338 | } | |
abe529af | 8339 | } |
abe529af BP |
8340 | } |
8341 | ||
7ee20df1 | 8342 | static void |
0e15264f BP |
8343 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8344 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8345 | { |
8346 | clogged = true; | |
bde9f75d | 8347 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8348 | } |
8349 | ||
8350 | static void | |
0e15264f BP |
8351 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8352 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8353 | { |
8354 | clogged = false; | |
bde9f75d | 8355 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8356 | } |
8357 | ||
6814e51f BP |
8358 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
8359 | * 'reply' describing the results. */ | |
8360 | static void | |
8361 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
8362 | { | |
8363 | struct facet *facet; | |
8364 | int errors; | |
8365 | ||
8366 | errors = 0; | |
8367 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
8368 | if (!facet_check_consistency(facet)) { | |
8369 | errors++; | |
8370 | } | |
8371 | } | |
8372 | if (errors) { | |
2cc3c58e | 8373 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
8374 | } |
8375 | ||
8376 | if (errors) { | |
8377 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
8378 | ofproto->up.name, errors); | |
8379 | } else { | |
8380 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
8381 | } | |
8382 | } | |
8383 | ||
8384 | static void | |
8385 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
8386 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
8387 | { | |
8388 | struct ds reply = DS_EMPTY_INITIALIZER; | |
8389 | struct ofproto_dpif *ofproto; | |
8390 | ||
8391 | if (argc > 1) { | |
8392 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8393 | if (!ofproto) { | |
bde9f75d EJ |
8394 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
8395 | "ofproto/list for help)"); | |
6814e51f BP |
8396 | return; |
8397 | } | |
8398 | ofproto_dpif_self_check__(ofproto, &reply); | |
8399 | } else { | |
8400 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8401 | ofproto_dpif_self_check__(ofproto, &reply); | |
8402 | } | |
8403 | } | |
8404 | ||
bde9f75d | 8405 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
8406 | ds_destroy(&reply); |
8407 | } | |
8408 | ||
27022416 JP |
8409 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
8410 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
8411 | * to destroy 'ofproto_shash' and free the returned value. */ | |
8412 | static const struct shash_node ** | |
8413 | get_ofprotos(struct shash *ofproto_shash) | |
8414 | { | |
8415 | const struct ofproto_dpif *ofproto; | |
8416 | ||
8417 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8418 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
8419 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
8420 | } | |
8421 | ||
8422 | return shash_sort(ofproto_shash); | |
8423 | } | |
8424 | ||
8425 | static void | |
8426 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
8427 | const char *argv[] OVS_UNUSED, | |
8428 | void *aux OVS_UNUSED) | |
8429 | { | |
8430 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8431 | struct shash ofproto_shash; | |
8432 | const struct shash_node **sorted_ofprotos; | |
8433 | int i; | |
8434 | ||
8435 | shash_init(&ofproto_shash); | |
8436 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8437 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8438 | const struct shash_node *node = sorted_ofprotos[i]; | |
8439 | ds_put_format(&ds, "%s\n", node->name); | |
8440 | } | |
8441 | ||
8442 | shash_destroy(&ofproto_shash); | |
8443 | free(sorted_ofprotos); | |
8444 | ||
8445 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8446 | ds_destroy(&ds); | |
8447 | } | |
8448 | ||
8449 | static void | |
8450 | show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) | |
8451 | { | |
27022416 JP |
8452 | const struct shash_node **ports; |
8453 | int i; | |
655ab909 AZ |
8454 | struct avg_subfacet_rates lifetime; |
8455 | unsigned long long int minutes; | |
8456 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8457 | ||
8458 | minutes = (time_msec() - ofproto->created) / min_ms; | |
8459 | ||
8460 | if (minutes > 0) { | |
8461 | lifetime.add_rate = (double)ofproto->total_subfacet_add_count | |
8462 | / minutes; | |
8463 | lifetime.del_rate = (double)ofproto->total_subfacet_del_count | |
8464 | / minutes; | |
8465 | }else { | |
8466 | lifetime.add_rate = 0.0; | |
8467 | lifetime.del_rate = 0.0; | |
8468 | } | |
27022416 | 8469 | |
acf60855 JP |
8470 | ds_put_format(ds, "%s (%s):\n", ofproto->up.name, |
8471 | dpif_name(ofproto->backer->dpif)); | |
27022416 | 8472 | ds_put_format(ds, |
735d7efb AZ |
8473 | "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n", |
8474 | ofproto->n_hit, ofproto->n_missed); | |
655ab909 AZ |
8475 | ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d," |
8476 | " life span: %llu(ms)\n", | |
8477 | hmap_count(&ofproto->subfacets), | |
8478 | avg_subfacet_count(ofproto), | |
8479 | ofproto->max_n_subfacet, | |
8480 | avg_subfacet_life_span(ofproto)); | |
8481 | if (minutes >= 60) { | |
8482 | show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly); | |
8483 | } | |
8484 | if (minutes >= 60 * 24) { | |
8485 | show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily); | |
8486 | } | |
8487 | show_dp_rates(ds, "\t\toverall avg:", &lifetime); | |
27022416 JP |
8488 | |
8489 | ports = shash_sort(&ofproto->up.port_by_name); | |
8490 | for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { | |
8491 | const struct shash_node *node = ports[i]; | |
8492 | struct ofport *ofport = node->data; | |
8493 | const char *name = netdev_get_name(ofport->netdev); | |
8494 | const char *type = netdev_get_type(ofport->netdev); | |
0a740f48 EJ |
8495 | uint32_t odp_port; |
8496 | ||
8497 | ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); | |
8498 | ||
8499 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); | |
8500 | if (odp_port != OVSP_NONE) { | |
8501 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
8502 | } else { | |
8503 | ds_put_cstr(ds, "none:"); | |
8504 | } | |
27022416 | 8505 | |
27022416 JP |
8506 | if (strcmp(type, "system")) { |
8507 | struct netdev *netdev; | |
8508 | int error; | |
8509 | ||
8510 | ds_put_format(ds, " (%s", type); | |
8511 | ||
8512 | error = netdev_open(name, type, &netdev); | |
8513 | if (!error) { | |
8514 | struct smap config; | |
8515 | ||
8516 | smap_init(&config); | |
8517 | error = netdev_get_config(netdev, &config); | |
8518 | if (!error) { | |
8519 | const struct smap_node **nodes; | |
8520 | size_t i; | |
8521 | ||
8522 | nodes = smap_sort(&config); | |
8523 | for (i = 0; i < smap_count(&config); i++) { | |
8524 | const struct smap_node *node = nodes[i]; | |
8525 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
8526 | node->key, node->value); | |
8527 | } | |
8528 | free(nodes); | |
8529 | } | |
8530 | smap_destroy(&config); | |
8531 | ||
8532 | netdev_close(netdev); | |
8533 | } | |
8534 | ds_put_char(ds, ')'); | |
8535 | } | |
8536 | ds_put_char(ds, '\n'); | |
8537 | } | |
8538 | free(ports); | |
8539 | } | |
8540 | ||
8541 | static void | |
8542 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, | |
8543 | const char *argv[], void *aux OVS_UNUSED) | |
8544 | { | |
8545 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8546 | const struct ofproto_dpif *ofproto; | |
8547 | ||
8548 | if (argc > 1) { | |
8549 | int i; | |
8550 | for (i = 1; i < argc; i++) { | |
8551 | ofproto = ofproto_dpif_lookup(argv[i]); | |
8552 | if (!ofproto) { | |
8553 | ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " | |
8554 | "for help)", argv[i]); | |
8555 | unixctl_command_reply_error(conn, ds_cstr(&ds)); | |
8556 | return; | |
8557 | } | |
8558 | show_dp_format(ofproto, &ds); | |
8559 | } | |
8560 | } else { | |
8561 | struct shash ofproto_shash; | |
8562 | const struct shash_node **sorted_ofprotos; | |
8563 | int i; | |
8564 | ||
8565 | shash_init(&ofproto_shash); | |
8566 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8567 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8568 | const struct shash_node *node = sorted_ofprotos[i]; | |
8569 | show_dp_format(node->data, &ds); | |
8570 | } | |
8571 | ||
8572 | shash_destroy(&ofproto_shash); | |
8573 | free(sorted_ofprotos); | |
8574 | } | |
8575 | ||
8576 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8577 | ds_destroy(&ds); | |
8578 | } | |
8579 | ||
8580 | static void | |
8581 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8582 | int argc OVS_UNUSED, const char *argv[], | |
8583 | void *aux OVS_UNUSED) | |
8584 | { | |
8585 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8586 | const struct ofproto_dpif *ofproto; | |
8587 | struct subfacet *subfacet; | |
8588 | ||
8589 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8590 | if (!ofproto) { | |
8591 | unixctl_command_reply_error(conn, "no such bridge"); | |
8592 | return; | |
8593 | } | |
8594 | ||
af37354d EJ |
8595 | update_stats(ofproto->backer); |
8596 | ||
27022416 | 8597 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
9566abf9 | 8598 | odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); |
27022416 JP |
8599 | |
8600 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8601 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8602 | if (subfacet->used) { | |
8603 | ds_put_format(&ds, "%.3fs", | |
8604 | (time_msec() - subfacet->used) / 1000.0); | |
8605 | } else { | |
8606 | ds_put_format(&ds, "never"); | |
8607 | } | |
8608 | if (subfacet->facet->tcp_flags) { | |
8609 | ds_put_cstr(&ds, ", flags:"); | |
8610 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8611 | } | |
8612 | ||
8613 | ds_put_cstr(&ds, ", actions:"); | |
f2245da3 JP |
8614 | if (subfacet->slow) { |
8615 | uint64_t slow_path_stub[128 / 8]; | |
8616 | const struct nlattr *actions; | |
8617 | size_t actions_len; | |
8618 | ||
8619 | compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow, | |
8620 | slow_path_stub, sizeof slow_path_stub, | |
8621 | &actions, &actions_len); | |
8622 | format_odp_actions(&ds, actions, actions_len); | |
8623 | } else { | |
8624 | format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); | |
8625 | } | |
27022416 JP |
8626 | ds_put_char(&ds, '\n'); |
8627 | } | |
8628 | ||
8629 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8630 | ds_destroy(&ds); | |
8631 | } | |
8632 | ||
8633 | static void | |
8634 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8635 | int argc OVS_UNUSED, const char *argv[], | |
8636 | void *aux OVS_UNUSED) | |
8637 | { | |
8638 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8639 | struct ofproto_dpif *ofproto; | |
8640 | ||
8641 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8642 | if (!ofproto) { | |
8643 | unixctl_command_reply_error(conn, "no such bridge"); | |
8644 | return; | |
8645 | } | |
8646 | ||
8647 | flush(&ofproto->up); | |
8648 | ||
8649 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8650 | ds_destroy(&ds); | |
8651 | } | |
8652 | ||
abe529af BP |
8653 | static void |
8654 | ofproto_dpif_unixctl_init(void) | |
8655 | { | |
8656 | static bool registered; | |
8657 | if (registered) { | |
8658 | return; | |
8659 | } | |
8660 | registered = true; | |
8661 | ||
0e15264f BP |
8662 | unixctl_command_register( |
8663 | "ofproto/trace", | |
72e8bf28 AA |
8664 | "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", |
8665 | 2, 6, ofproto_unixctl_trace, NULL); | |
96e466a3 | 8666 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
8667 | ofproto_unixctl_fdb_flush, NULL); |
8668 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8669 | ofproto_unixctl_fdb_show, NULL); | |
8670 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8671 | ofproto_dpif_clog, NULL); | |
8672 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8673 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
8674 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
8675 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
8676 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
8677 | ofproto_unixctl_dpif_dump_dps, NULL); | |
8678 | unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, | |
8679 | ofproto_unixctl_dpif_show, NULL); | |
8680 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, | |
8681 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8682 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8683 | ofproto_unixctl_dpif_del_flows, NULL); | |
abe529af BP |
8684 | } |
8685 | \f | |
52a90c29 BP |
8686 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
8687 | * | |
8688 | * This is deprecated. It is only for compatibility with broken device drivers | |
8689 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8690 | * devices are not used. When broken device drivers are no longer in | |
8691 | * widespread use, we will delete these interfaces. */ | |
8692 | ||
8693 | static int | |
8694 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8695 | { | |
8696 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8697 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8698 | ||
8699 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8700 | && vid == ofport->vlandev_vid) { | |
8701 | return 0; | |
8702 | } | |
8703 | ||
2cc3c58e | 8704 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
8705 | |
8706 | if (ofport->realdev_ofp_port) { | |
8707 | vsp_remove(ofport); | |
8708 | } | |
8709 | if (realdev_ofp_port && ofport->bundle) { | |
8710 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8711 | * themselves be part of a bundle. */ | |
8712 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8713 | } | |
8714 | ||
8715 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8716 | ofport->vlandev_vid = vid; | |
8717 | ||
8718 | if (realdev_ofp_port) { | |
8719 | vsp_add(ofport, realdev_ofp_port, vid); | |
8720 | } | |
8721 | ||
8722 | return 0; | |
8723 | } | |
8724 | ||
8725 | static uint32_t | |
8726 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8727 | { | |
8728 | return hash_2words(realdev_ofp_port, vid); | |
8729 | } | |
8730 | ||
40e05935 BP |
8731 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
8732 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
8733 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
8734 | * it would return the port number of eth0.9. | |
8735 | * | |
8736 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
8737 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
8738 | static uint32_t |
8739 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
8740 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
8741 | { | |
8742 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
e1b1d06a | 8743 | uint16_t realdev_ofp_port; |
52a90c29 BP |
8744 | int vid = vlan_tci_to_vid(vlan_tci); |
8745 | const struct vlan_splinter *vsp; | |
8746 | ||
e1b1d06a | 8747 | realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); |
52a90c29 BP |
8748 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, |
8749 | hash_realdev_vid(realdev_ofp_port, vid), | |
8750 | &ofproto->realdev_vid_map) { | |
8751 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8752 | && vsp->vid == vid) { | |
e1b1d06a | 8753 | return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); |
52a90c29 BP |
8754 | } |
8755 | } | |
8756 | } | |
8757 | return realdev_odp_port; | |
8758 | } | |
8759 | ||
8760 | static struct vlan_splinter * | |
8761 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8762 | { | |
8763 | struct vlan_splinter *vsp; | |
8764 | ||
8765 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8766 | &ofproto->vlandev_map) { | |
8767 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8768 | return vsp; | |
8769 | } | |
8770 | } | |
8771 | ||
8772 | return NULL; | |
8773 | } | |
8774 | ||
40e05935 BP |
8775 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
8776 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8777 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8778 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8779 | * eth0 and store 9 in '*vid'. | |
8780 | * | |
8781 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8782 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8783 | * always does.*/ | |
52a90c29 BP |
8784 | static uint16_t |
8785 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 8786 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
8787 | { |
8788 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8789 | const struct vlan_splinter *vsp; | |
8790 | ||
8791 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8792 | if (vsp) { | |
8793 | if (vid) { | |
8794 | *vid = vsp->vid; | |
8795 | } | |
8796 | return vsp->realdev_ofp_port; | |
8797 | } | |
8798 | } | |
8799 | return 0; | |
8800 | } | |
8801 | ||
b98d8985 BP |
8802 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
8803 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8804 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8805 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8806 | * always the case unless VLAN splinters are enabled), returns false without | |
8807 | * making any changes. */ | |
8808 | static bool | |
8809 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8810 | { | |
8811 | uint16_t realdev; | |
8812 | int vid; | |
8813 | ||
8814 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8815 | if (!realdev) { | |
8816 | return false; | |
8817 | } | |
8818 | ||
8819 | /* Cause the flow to be processed as if it came in on the real device with | |
8820 | * the VLAN device's VLAN ID. */ | |
8821 | flow->in_port = realdev; | |
8822 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8823 | return true; | |
8824 | } | |
8825 | ||
52a90c29 BP |
8826 | static void |
8827 | vsp_remove(struct ofport_dpif *port) | |
8828 | { | |
8829 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8830 | struct vlan_splinter *vsp; | |
8831 | ||
8832 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8833 | if (vsp) { | |
8834 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8835 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8836 | free(vsp); | |
8837 | ||
8838 | port->realdev_ofp_port = 0; | |
8839 | } else { | |
8840 | VLOG_ERR("missing vlan device record"); | |
8841 | } | |
8842 | } | |
8843 | ||
8844 | static void | |
8845 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8846 | { | |
8847 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8848 | ||
8849 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8850 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8851 | == realdev_ofp_port)) { | |
8852 | struct vlan_splinter *vsp; | |
8853 | ||
8854 | vsp = xmalloc(sizeof *vsp); | |
8855 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8856 | hash_int(port->up.ofp_port, 0)); | |
8857 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8858 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8859 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8860 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8861 | vsp->vid = vid; | |
8862 | ||
8863 | port->realdev_ofp_port = realdev_ofp_port; | |
8864 | } else { | |
8865 | VLOG_ERR("duplicate vlan device record"); | |
8866 | } | |
8867 | } | |
e1b1d06a JP |
8868 | |
8869 | static uint32_t | |
8870 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8871 | { | |
8872 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8873 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8874 | } | |
8875 | ||
acf60855 JP |
8876 | static struct ofport_dpif * |
8877 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
e1b1d06a JP |
8878 | { |
8879 | struct ofport_dpif *port; | |
8880 | ||
8881 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8882 | hash_int(odp_port, 0), | |
acf60855 | 8883 | &backer->odp_to_ofport_map) { |
e1b1d06a | 8884 | if (port->odp_port == odp_port) { |
acf60855 | 8885 | return port; |
e1b1d06a JP |
8886 | } |
8887 | } | |
8888 | ||
acf60855 JP |
8889 | return NULL; |
8890 | } | |
8891 | ||
8892 | static uint16_t | |
8893 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8894 | { | |
8895 | struct ofport_dpif *port; | |
8896 | ||
8897 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 8898 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
8899 | return port->up.ofp_port; |
8900 | } else { | |
8901 | return OFPP_NONE; | |
8902 | } | |
e1b1d06a | 8903 | } |
655ab909 AZ |
8904 | static unsigned long long int |
8905 | avg_subfacet_life_span(const struct ofproto_dpif *ofproto) | |
8906 | { | |
8907 | unsigned long long int dc; | |
8908 | unsigned long long int avg; | |
8909 | ||
8910 | dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count; | |
8911 | avg = dc ? ofproto->total_subfacet_life_span / dc : 0; | |
8912 | ||
8913 | return avg; | |
8914 | } | |
8915 | ||
8916 | static double | |
8917 | avg_subfacet_count(const struct ofproto_dpif *ofproto) | |
8918 | { | |
8919 | double avg_c = 0.0; | |
8920 | ||
8921 | if (ofproto->n_update_stats) { | |
8922 | avg_c = (double)ofproto->total_subfacet_count | |
8923 | / ofproto->n_update_stats; | |
8924 | } | |
8925 | ||
8926 | return avg_c; | |
8927 | } | |
8928 | ||
8929 | static void | |
8930 | show_dp_rates(struct ds *ds, const char *heading, | |
8931 | const struct avg_subfacet_rates *rates) | |
8932 | { | |
8933 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
8934 | heading, rates->add_rate, rates->del_rate); | |
8935 | } | |
8936 | ||
8937 | static void | |
8938 | update_max_subfacet_count(struct ofproto_dpif *ofproto) | |
8939 | { | |
8940 | ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet, | |
8941 | hmap_count(&ofproto->subfacets)); | |
8942 | } | |
8943 | ||
8944 | /* Compute exponentially weighted moving average, adding 'new' as the newest, | |
8945 | * most heavily weighted element. 'base' designates the rate of decay: after | |
8946 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
8947 | * (about .37). */ | |
8948 | static void | |
8949 | exp_mavg(double *avg, int base, double new) | |
8950 | { | |
8951 | *avg = (*avg * (base - 1) + new) / base; | |
8952 | } | |
8953 | ||
8954 | static void | |
8955 | update_moving_averages(struct ofproto_dpif *ofproto) | |
8956 | { | |
8957 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8958 | ||
8959 | /* Update hourly averages on the minute boundaries. */ | |
8960 | if (time_msec() - ofproto->last_minute >= min_ms) { | |
8961 | exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count); | |
8962 | exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count); | |
8963 | ||
8964 | /* Update daily averages on the hour boundaries. */ | |
8965 | if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) { | |
8966 | exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate); | |
8967 | exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate); | |
8968 | } | |
8969 | ||
8970 | ofproto->total_subfacet_add_count += ofproto->subfacet_add_count; | |
8971 | ofproto->total_subfacet_del_count += ofproto->subfacet_del_count; | |
8972 | ofproto->subfacet_add_count = 0; | |
8973 | ofproto->subfacet_del_count = 0; | |
8974 | ofproto->last_minute += min_ms; | |
8975 | } | |
8976 | } | |
e1b1d06a | 8977 | |
735d7efb AZ |
8978 | static void |
8979 | dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta) | |
8980 | { | |
8981 | ofproto->n_hit += delta; | |
8982 | } | |
8983 | ||
abe529af | 8984 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 8985 | init, |
abe529af BP |
8986 | enumerate_types, |
8987 | enumerate_names, | |
8988 | del, | |
0aeaabc8 | 8989 | port_open_type, |
acf60855 JP |
8990 | type_run, |
8991 | type_run_fast, | |
8992 | type_wait, | |
abe529af BP |
8993 | alloc, |
8994 | construct, | |
8995 | destruct, | |
8996 | dealloc, | |
8997 | run, | |
5fcc0d00 | 8998 | run_fast, |
abe529af | 8999 | wait, |
0d085684 | 9000 | get_memory_usage, |
abe529af | 9001 | flush, |
6c1491fb BP |
9002 | get_features, |
9003 | get_tables, | |
abe529af BP |
9004 | port_alloc, |
9005 | port_construct, | |
9006 | port_destruct, | |
9007 | port_dealloc, | |
9008 | port_modified, | |
9009 | port_reconfigured, | |
9010 | port_query_by_name, | |
9011 | port_add, | |
9012 | port_del, | |
6527c598 | 9013 | port_get_stats, |
abe529af BP |
9014 | port_dump_start, |
9015 | port_dump_next, | |
9016 | port_dump_done, | |
9017 | port_poll, | |
9018 | port_poll_wait, | |
9019 | port_is_lacp_current, | |
0ab6decf | 9020 | NULL, /* rule_choose_table */ |
abe529af BP |
9021 | rule_alloc, |
9022 | rule_construct, | |
9023 | rule_destruct, | |
9024 | rule_dealloc, | |
abe529af BP |
9025 | rule_get_stats, |
9026 | rule_execute, | |
9027 | rule_modify_actions, | |
7257b535 | 9028 | set_frag_handling, |
abe529af BP |
9029 | packet_out, |
9030 | set_netflow, | |
9031 | get_netflow_ids, | |
9032 | set_sflow, | |
29089a54 | 9033 | set_ipfix, |
abe529af | 9034 | set_cfm, |
9a9e3786 | 9035 | get_cfm_status, |
ccc09689 EJ |
9036 | set_bfd, |
9037 | get_bfd_status, | |
21f7563c JP |
9038 | set_stp, |
9039 | get_stp_status, | |
9040 | set_stp_port, | |
9041 | get_stp_port_status, | |
8b36f51e | 9042 | set_queues, |
abe529af BP |
9043 | bundle_set, |
9044 | bundle_remove, | |
9045 | mirror_set, | |
9d24de3b | 9046 | mirror_get_stats, |
abe529af BP |
9047 | set_flood_vlans, |
9048 | is_mirror_output_bundle, | |
8402c74b | 9049 | forward_bpdu_changed, |
c4069512 | 9050 | set_mac_table_config, |
52a90c29 | 9051 | set_realdev, |
abe529af | 9052 | }; |