]>
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 | ||
abe529af | 23 | #include "bond.h" |
daff3353 | 24 | #include "bundle.h" |
abe529af BP |
25 | #include "byte-order.h" |
26 | #include "connmgr.h" | |
27 | #include "coverage.h" | |
28 | #include "cfm.h" | |
29 | #include "dpif.h" | |
30 | #include "dynamic-string.h" | |
31 | #include "fail-open.h" | |
32 | #include "hmapx.h" | |
33 | #include "lacp.h" | |
75a75043 | 34 | #include "learn.h" |
abe529af | 35 | #include "mac-learning.h" |
816fd533 | 36 | #include "meta-flow.h" |
abe529af | 37 | #include "multipath.h" |
0a740f48 | 38 | #include "netdev-vport.h" |
abe529af BP |
39 | #include "netdev.h" |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
f25d0cf3 | 45 | #include "ofp-actions.h" |
31a19d69 | 46 | #include "ofp-parse.h" |
abe529af | 47 | #include "ofp-print.h" |
9d6ac44e | 48 | #include "ofproto-dpif-governor.h" |
bae473fe | 49 | #include "ofproto-dpif-sflow.h" |
abe529af | 50 | #include "poll-loop.h" |
0d085684 | 51 | #include "simap.h" |
27022416 | 52 | #include "smap.h" |
abe529af | 53 | #include "timer.h" |
b9ad7294 | 54 | #include "tunnel.h" |
6c1491fb | 55 | #include "unaligned.h" |
abe529af BP |
56 | #include "unixctl.h" |
57 | #include "vlan-bitmap.h" | |
58 | #include "vlog.h" | |
59 | ||
60 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
61 | ||
abe529af | 62 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
63 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
64 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
65 | COVERAGE_DEFINE(facet_revalidate); |
66 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 67 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 68 | |
29901626 | 69 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 70 | * flow translation. */ |
1642690c | 71 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 72 | |
9cdaaebe BP |
73 | /* Number of implemented OpenFlow tables. */ |
74 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
75 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
76 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 77 | |
abe529af BP |
78 | struct ofport_dpif; |
79 | struct ofproto_dpif; | |
a088a1ff | 80 | struct flow_miss; |
abe529af BP |
81 | |
82 | struct rule_dpif { | |
83 | struct rule up; | |
84 | ||
abe529af BP |
85 | /* These statistics: |
86 | * | |
87 | * - Do include packets and bytes from facets that have been deleted or | |
88 | * whose own statistics have been folded into the rule. | |
89 | * | |
90 | * - Do include packets and bytes sent "by hand" that were accounted to | |
91 | * the rule without any facet being involved (this is a rare corner | |
92 | * case in rule_execute()). | |
93 | * | |
94 | * - Do not include packet or bytes that can be obtained from any facet's | |
95 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 96 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
97 | */ |
98 | uint64_t packet_count; /* Number of packets received. */ | |
99 | uint64_t byte_count; /* Number of bytes received. */ | |
100 | ||
54a9cbc9 BP |
101 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
102 | ||
abe529af BP |
103 | struct list facets; /* List of "struct facet"s. */ |
104 | }; | |
105 | ||
106 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
107 | { | |
108 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
109 | } | |
110 | ||
29901626 | 111 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
112 | const struct flow *); |
113 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
114 | const struct flow *, | |
115 | uint8_t table); | |
c376f9a3 IY |
116 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
117 | const struct flow *flow); | |
abe529af | 118 | |
112bc5f4 BP |
119 | static void rule_credit_stats(struct rule_dpif *, |
120 | const struct dpif_flow_stats *); | |
18b2a258 | 121 | static void flow_push_stats(struct rule_dpif *, const struct flow *, |
112bc5f4 | 122 | const struct dpif_flow_stats *); |
822d9414 | 123 | static tag_type rule_calculate_tag(const struct flow *, |
5cb7a798 | 124 | const struct minimask *, uint32_t basis); |
b0f7b9b5 BP |
125 | static void rule_invalidate(const struct rule_dpif *); |
126 | ||
abe529af BP |
127 | #define MAX_MIRRORS 32 |
128 | typedef uint32_t mirror_mask_t; | |
129 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
130 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
131 | struct ofmirror { | |
132 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
133 | size_t idx; /* In ofproto's "mirrors" array. */ | |
134 | void *aux; /* Key supplied by ofproto's client. */ | |
135 | char *name; /* Identifier for log messages. */ | |
136 | ||
137 | /* Selection criteria. */ | |
138 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
139 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
140 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
141 | ||
9ba15e2a | 142 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
143 | struct ofbundle *out; /* Output port or NULL. */ |
144 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 145 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
146 | |
147 | /* Counters. */ | |
148 | int64_t packet_count; /* Number of packets sent. */ | |
149 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
150 | }; |
151 | ||
152 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
153 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
154 | mirror_mask_t mirrors, | |
155 | uint64_t packets, uint64_t bytes); | |
abe529af | 156 | |
abe529af | 157 | struct ofbundle { |
abe529af | 158 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 159 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
160 | void *aux; /* Key supplied by ofproto's client. */ |
161 | char *name; /* Identifier for log messages. */ | |
162 | ||
163 | /* Configuration. */ | |
164 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 165 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
166 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
167 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
168 | * NULL if all VLANs are trunked. */ | |
169 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
170 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 171 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
172 | |
173 | /* Status. */ | |
9e1fd49b | 174 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
175 | |
176 | /* Port mirroring info. */ | |
177 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
178 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
179 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
180 | }; | |
181 | ||
182 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 183 | static void bundle_update(struct ofbundle *); |
abe529af BP |
184 | static void bundle_destroy(struct ofbundle *); |
185 | static void bundle_del_port(struct ofport_dpif *); | |
186 | static void bundle_run(struct ofbundle *); | |
187 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 188 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
189 | uint16_t in_port, bool warn, |
190 | struct ofport_dpif **in_ofportp); | |
abe529af | 191 | |
33158a18 JP |
192 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
193 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
194 | * when an input bundle is needed for validation (e.g., mirroring or | |
195 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
196 | * any 'port' structs, so care must be taken when dealing with it. */ | |
197 | static struct ofbundle ofpp_none_bundle = { | |
198 | .name = "OFPP_NONE", | |
199 | .vlan_mode = PORT_VLAN_TRUNK | |
200 | }; | |
201 | ||
21f7563c JP |
202 | static void stp_run(struct ofproto_dpif *ofproto); |
203 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
204 | static int set_stp_port(struct ofport *, |
205 | const struct ofproto_port_stp_settings *); | |
21f7563c | 206 | |
5da5ec37 BP |
207 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
208 | ||
abe529af BP |
209 | struct action_xlate_ctx { |
210 | /* action_xlate_ctx_init() initializes these members. */ | |
211 | ||
212 | /* The ofproto. */ | |
213 | struct ofproto_dpif *ofproto; | |
214 | ||
215 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
216 | * this flow when actions change header fields. */ | |
217 | struct flow flow; | |
218 | ||
219 | /* The packet corresponding to 'flow', or a null pointer if we are | |
220 | * revalidating without a packet to refer to. */ | |
221 | const struct ofpbuf *packet; | |
222 | ||
3de9590b BP |
223 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
224 | * actions update the flow table? | |
225 | * | |
226 | * We want to update these tables if we are actually processing a packet, | |
227 | * or if we are accounting for packets that the datapath has processed, but | |
228 | * not if we are just revalidating. */ | |
229 | bool may_learn; | |
75a75043 | 230 | |
18b2a258 BP |
231 | /* The rule that we are currently translating, or NULL. */ |
232 | struct rule_dpif *rule; | |
54834960 | 233 | |
0e553d9c BP |
234 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
235 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
236 | * timeouts.) */ | |
237 | uint8_t tcp_flags; | |
238 | ||
112bc5f4 BP |
239 | /* If nonnull, flow translation calls this function just before executing a |
240 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
241 | * when the recursion depth is exceeded. | |
242 | * | |
243 | * 'rule' is the rule being submitted into. It will be null if the | |
244 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
245 | * | |
246 | * This is normally null so the client has to set it manually after | |
247 | * calling action_xlate_ctx_init(). */ | |
248 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
249 | ||
479df176 BP |
250 | /* If nonnull, flow translation calls this function to report some |
251 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
252 | * dropped a packet. */ | |
253 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
254 | ||
112bc5f4 BP |
255 | /* If nonnull, flow translation credits the specified statistics to each |
256 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
257 | * |
258 | * This is normally null so the client has to set it manually after | |
259 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 260 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 261 | |
abe529af BP |
262 | /* xlate_actions() initializes and uses these members. The client might want |
263 | * to look at them after it returns. */ | |
264 | ||
265 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 266 | tag_type tags; /* Tags associated with actions. */ |
6a7e895f | 267 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
75a75043 BP |
268 | bool has_learn; /* Actions include NXAST_LEARN? */ |
269 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 270 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 271 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 272 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
273 | |
274 | /* xlate_actions() initializes and uses these members, but the client has no | |
275 | * reason to look at them. */ | |
276 | ||
277 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 278 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 279 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 280 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 281 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 | 282 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
9b56fe13 | 283 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ |
6ff686f2 | 284 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ |
848e8809 | 285 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
286 | }; |
287 | ||
288 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
289 | struct ofproto_dpif *, const struct flow *, | |
18b2a258 | 290 | ovs_be16 initial_tci, struct rule_dpif *, |
0e553d9c | 291 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 | 292 | static void xlate_actions(struct action_xlate_ctx *, |
f25d0cf3 | 293 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 BP |
294 | struct ofpbuf *odp_actions); |
295 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
f25d0cf3 BP |
296 | const struct ofpact *ofpacts, |
297 | size_t ofpacts_len); | |
0a740f48 EJ |
298 | static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port, |
299 | uint8_t table_id, bool may_packet_in); | |
abe529af | 300 | |
6a7e895f BP |
301 | static size_t put_userspace_action(const struct ofproto_dpif *, |
302 | struct ofpbuf *odp_actions, | |
303 | const struct flow *, | |
304 | const union user_action_cookie *); | |
305 | ||
306 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
307 | enum slow_path_reason, | |
308 | uint64_t *stub, size_t stub_size, | |
309 | const struct nlattr **actionsp, | |
310 | size_t *actions_lenp); | |
311 | ||
479df176 BP |
312 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); |
313 | ||
6a7e895f BP |
314 | /* A subfacet (see "struct subfacet" below) has three possible installation |
315 | * states: | |
316 | * | |
317 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
318 | * case just after the subfacet is created, just before the subfacet is | |
319 | * destroyed, or if the datapath returns an error when we try to install a | |
320 | * subfacet. | |
321 | * | |
322 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
323 | * | |
324 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
325 | * ofproto_dpif is installed in the datapath. | |
326 | */ | |
327 | enum subfacet_path { | |
328 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
329 | SF_FAST_PATH, /* Full actions are installed. */ | |
330 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
331 | }; | |
332 | ||
333 | static const char *subfacet_path_to_string(enum subfacet_path); | |
334 | ||
5f5fbd17 BP |
335 | /* A dpif flow and actions associated with a facet. |
336 | * | |
337 | * See also the large comment on struct facet. */ | |
338 | struct subfacet { | |
339 | /* Owners. */ | |
340 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
341 | struct list list_node; /* In struct facet's 'facets' list. */ | |
342 | struct facet *facet; /* Owning facet. */ | |
343 | ||
344 | /* Key. | |
345 | * | |
346 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
347 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
348 | * regenerate the ODP flow key from ->facet->flow. */ | |
349 | enum odp_key_fitness key_fitness; | |
350 | struct nlattr *key; | |
351 | int key_len; | |
352 | ||
353 | long long int used; /* Time last used; time created if not used. */ | |
354 | ||
355 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
356 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
357 | ||
358 | /* Datapath actions. | |
359 | * | |
360 | * These should be essentially identical for every subfacet in a facet, but | |
361 | * may differ in trivial ways due to VLAN splinters. */ | |
362 | size_t actions_len; /* Number of bytes in actions[]. */ | |
363 | struct nlattr *actions; /* Datapath actions. */ | |
364 | ||
6a7e895f BP |
365 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
366 | enum subfacet_path path; /* Installed in datapath? */ | |
5f5fbd17 BP |
367 | |
368 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
369 | * splinters can cause it to differ. This value should be removed when | |
370 | * the VLAN splinters feature is no longer needed. */ | |
371 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
a088a1ff JP |
372 | |
373 | /* Datapath port the packet arrived on. This is needed to remove | |
374 | * flows for ports that are no longer part of the bridge. Since the | |
375 | * flow definition only has the OpenFlow port number and the port is | |
376 | * no longer part of the bridge, we can't determine the datapath port | |
377 | * number needed to delete the flow from the datapath. */ | |
378 | uint32_t odp_in_port; | |
5f5fbd17 BP |
379 | }; |
380 | ||
1d85f9e5 JP |
381 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
382 | ||
a088a1ff | 383 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 384 | long long int now); |
5f5fbd17 | 385 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
acf60855 JP |
386 | const struct nlattr *key, size_t key_len, |
387 | uint32_t key_hash, | |
388 | const struct flow *flow); | |
5f5fbd17 BP |
389 | static void subfacet_destroy(struct subfacet *); |
390 | static void subfacet_destroy__(struct subfacet *); | |
1d85f9e5 JP |
391 | static void subfacet_destroy_batch(struct ofproto_dpif *, |
392 | struct subfacet **, int n); | |
5f5fbd17 BP |
393 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, |
394 | struct ofpbuf *key); | |
395 | static void subfacet_reset_dp_stats(struct subfacet *, | |
396 | struct dpif_flow_stats *); | |
397 | static void subfacet_update_time(struct subfacet *, long long int used); | |
398 | static void subfacet_update_stats(struct subfacet *, | |
399 | const struct dpif_flow_stats *); | |
400 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
401 | const struct ofpbuf *packet, |
402 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
403 | static int subfacet_install(struct subfacet *, |
404 | const struct nlattr *actions, size_t actions_len, | |
6a7e895f | 405 | struct dpif_flow_stats *, enum slow_path_reason); |
5f5fbd17 BP |
406 | static void subfacet_uninstall(struct subfacet *); |
407 | ||
6a7e895f BP |
408 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); |
409 | ||
b0f7b9b5 BP |
410 | /* An exact-match instantiation of an OpenFlow flow. |
411 | * | |
412 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
413 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
414 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
415 | * the facet. When the kernel module (or other dpif implementation) and Open | |
416 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
417 | * one subfacet per facet. If the dpif implementation supports more-specific | |
418 | * flow matching than userspace, however, a facet can have more than one | |
419 | * subfacet, each of which corresponds to some distinction in flow that | |
420 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
421 | * |
422 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
423 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 424 | struct facet { |
b0f7b9b5 BP |
425 | /* Owners. */ |
426 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
427 | struct list list_node; /* In owning rule's 'facets' list. */ | |
428 | struct rule_dpif *rule; /* Owning rule. */ | |
429 | ||
430 | /* Owned data. */ | |
431 | struct list subfacets; | |
abe529af BP |
432 | long long int used; /* Time last used; time created if not used. */ |
433 | ||
b0f7b9b5 BP |
434 | /* Key. */ |
435 | struct flow flow; | |
436 | ||
abe529af BP |
437 | /* These statistics: |
438 | * | |
439 | * - Do include packets and bytes sent "by hand", e.g. with | |
440 | * dpif_execute(). | |
441 | * | |
442 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 443 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 444 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
445 | * |
446 | * - Do not include packets or bytes that can be obtained from the | |
447 | * datapath for any existing subfacet. | |
abe529af BP |
448 | */ |
449 | uint64_t packet_count; /* Number of packets received. */ | |
450 | uint64_t byte_count; /* Number of bytes received. */ | |
451 | ||
b0f7b9b5 | 452 | /* Resubmit statistics. */ |
9d24de3b JP |
453 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
454 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
455 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 456 | |
b0f7b9b5 | 457 | /* Accounting. */ |
907a4c5e | 458 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 459 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 460 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 461 | |
b95fc6ba BP |
462 | /* Properties of datapath actions. |
463 | * | |
464 | * Every subfacet has its own actions because actions can differ slightly | |
465 | * between splintered and non-splintered subfacets due to the VLAN tag | |
466 | * being initially different (present vs. absent). All of them have these | |
467 | * properties in common so we just store one copy of them here. */ | |
75a75043 BP |
468 | bool has_learn; /* Actions include NXAST_LEARN? */ |
469 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 470 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 471 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 472 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
473 | |
474 | /* Storage for a single subfacet, to reduce malloc() time and space | |
475 | * overhead. (A facet always has at least one subfacet and in the common | |
476 | * case has exactly one subfacet.) */ | |
477 | struct subfacet one_subfacet; | |
abe529af BP |
478 | }; |
479 | ||
2b459b83 BP |
480 | static struct facet *facet_create(struct rule_dpif *, |
481 | const struct flow *, uint32_t hash); | |
15baa734 | 482 | static void facet_remove(struct facet *); |
abe529af BP |
483 | static void facet_free(struct facet *); |
484 | ||
2b459b83 BP |
485 | static struct facet *facet_find(struct ofproto_dpif *, |
486 | const struct flow *, uint32_t hash); | |
abe529af | 487 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 488 | const struct flow *, uint32_t hash); |
c57b2226 | 489 | static void facet_revalidate(struct facet *); |
6814e51f | 490 | static bool facet_check_consistency(struct facet *); |
abe529af | 491 | |
15baa734 | 492 | static void facet_flush_stats(struct facet *); |
abe529af | 493 | |
15baa734 | 494 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 495 | static void facet_reset_counters(struct facet *); |
abe529af | 496 | static void facet_push_stats(struct facet *); |
3de9590b BP |
497 | static void facet_learn(struct facet *); |
498 | static void facet_account(struct facet *); | |
abe529af BP |
499 | |
500 | static bool facet_is_controller_flow(struct facet *); | |
501 | ||
abe529af | 502 | struct ofport_dpif { |
acf60855 | 503 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ |
abe529af BP |
504 | struct ofport up; |
505 | ||
506 | uint32_t odp_port; | |
507 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
508 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
509 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
510 | tag_type tag; /* Tag associated with this port. */ | |
015e08bc | 511 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 512 | long long int carrier_seq; /* Carrier status changes. */ |
b9ad7294 | 513 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ |
21f7563c | 514 | |
52a90c29 | 515 | /* Spanning tree. */ |
21f7563c JP |
516 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
517 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
518 | long long int stp_state_entered; | |
8b36f51e EJ |
519 | |
520 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
521 | |
522 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
523 | * | |
524 | * This is deprecated. It is only for compatibility with broken device | |
525 | * drivers in old versions of Linux that do not properly support VLANs when | |
526 | * VLAN devices are not used. When broken device drivers are no longer in | |
527 | * widespread use, we will delete these interfaces. */ | |
528 | uint16_t realdev_ofp_port; | |
529 | int vlandev_vid; | |
8b36f51e EJ |
530 | }; |
531 | ||
532 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
533 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
534 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
535 | struct priority_to_dscp { | |
536 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
537 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
538 | ||
539 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
540 | }; |
541 | ||
52a90c29 BP |
542 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
543 | * | |
544 | * This is deprecated. It is only for compatibility with broken device drivers | |
545 | * in old versions of Linux that do not properly support VLANs when VLAN | |
546 | * devices are not used. When broken device drivers are no longer in | |
547 | * widespread use, we will delete these interfaces. */ | |
548 | struct vlan_splinter { | |
549 | struct hmap_node realdev_vid_node; | |
550 | struct hmap_node vlandev_node; | |
551 | uint16_t realdev_ofp_port; | |
552 | uint16_t vlandev_ofp_port; | |
553 | int vid; | |
554 | }; | |
555 | ||
556 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
557 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 558 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
559 | static void vsp_remove(struct ofport_dpif *); |
560 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
561 | ||
e1b1d06a JP |
562 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
563 | uint16_t ofp_port); | |
564 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
565 | uint32_t odp_port); | |
566 | ||
abe529af BP |
567 | static struct ofport_dpif * |
568 | ofport_dpif_cast(const struct ofport *ofport) | |
569 | { | |
cb22974d | 570 | ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
571 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
572 | } | |
573 | ||
574 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 575 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 576 | static void port_wait(struct ofport_dpif *); |
a5610457 | 577 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 578 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 579 | |
7ee20df1 BP |
580 | struct dpif_completion { |
581 | struct list list_node; | |
582 | struct ofoperation *op; | |
583 | }; | |
584 | ||
54a9cbc9 BP |
585 | /* Extra information about a classifier table. |
586 | * Currently used just for optimized flow revalidation. */ | |
587 | struct table_dpif { | |
588 | /* If either of these is nonnull, then this table has a form that allows | |
589 | * flows to be tagged to avoid revalidating most flows for the most common | |
590 | * kinds of flow table changes. */ | |
591 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
592 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
593 | uint32_t basis; /* Keeps each table's tags separate. */ | |
594 | }; | |
595 | ||
3c4a309c BP |
596 | /* Reasons that we might need to revalidate every facet, and corresponding |
597 | * coverage counters. | |
598 | * | |
599 | * A value of 0 means that there is no need to revalidate. | |
600 | * | |
601 | * It would be nice to have some cleaner way to integrate with coverage | |
602 | * counters, but with only a few reasons I guess this is good enough for | |
603 | * now. */ | |
604 | enum revalidate_reason { | |
605 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
606 | REV_STP, /* Spanning tree protocol port status change. */ | |
607 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
608 | REV_FLOW_TABLE, /* Flow table changed. */ | |
609 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
610 | }; | |
611 | COVERAGE_DEFINE(rev_reconfigure); | |
612 | COVERAGE_DEFINE(rev_stp); | |
613 | COVERAGE_DEFINE(rev_port_toggled); | |
614 | COVERAGE_DEFINE(rev_flow_table); | |
615 | COVERAGE_DEFINE(rev_inconsistency); | |
616 | ||
8f73d537 EJ |
617 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
618 | * These are datapath flows which have no associated ofproto, if they did we | |
619 | * would use facets. */ | |
620 | struct drop_key { | |
621 | struct hmap_node hmap_node; | |
622 | struct nlattr *key; | |
623 | size_t key_len; | |
624 | }; | |
625 | ||
acf60855 JP |
626 | /* All datapaths of a given type share a single dpif backer instance. */ |
627 | struct dpif_backer { | |
628 | char *type; | |
629 | int refcount; | |
630 | struct dpif *dpif; | |
631 | struct timer next_expiration; | |
632 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
2cc3c58e | 633 | |
7d82ab2e | 634 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ |
b9ad7294 | 635 | |
2cc3c58e EJ |
636 | /* Facet revalidation flags applying to facets which use this backer. */ |
637 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
638 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
8f73d537 EJ |
639 | |
640 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
acf60855 JP |
641 | }; |
642 | ||
643 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
644 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
645 | ||
8f73d537 | 646 | static void drop_key_clear(struct dpif_backer *); |
acf60855 JP |
647 | static struct ofport_dpif * |
648 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
649 | ||
abe529af | 650 | struct ofproto_dpif { |
b44a10b7 | 651 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af | 652 | struct ofproto up; |
acf60855 | 653 | struct dpif_backer *backer; |
abe529af | 654 | |
c57b2226 BP |
655 | /* Special OpenFlow rules. */ |
656 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
657 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
658 | ||
6c1491fb BP |
659 | /* Statistics. */ |
660 | uint64_t n_matches; | |
661 | ||
abe529af BP |
662 | /* Bridging. */ |
663 | struct netflow *netflow; | |
bae473fe | 664 | struct dpif_sflow *sflow; |
abe529af BP |
665 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
666 | struct mac_learning *ml; | |
667 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 668 | bool has_mirrors; |
abe529af BP |
669 | bool has_bonded_bundles; |
670 | ||
abe529af BP |
671 | /* Facets. */ |
672 | struct hmap facets; | |
b0f7b9b5 | 673 | struct hmap subfacets; |
9d6ac44e | 674 | struct governor *governor; |
54a9cbc9 BP |
675 | |
676 | /* Revalidation. */ | |
677 | struct table_dpif tables[N_TABLES]; | |
7ee20df1 BP |
678 | |
679 | /* Support for debugging async flow mods. */ | |
680 | struct list completions; | |
daff3353 EJ |
681 | |
682 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
683 | struct netdev_stats stats; /* To account packets generated and consumed in |
684 | * userspace. */ | |
21f7563c JP |
685 | |
686 | /* Spanning tree. */ | |
687 | struct stp *stp; | |
688 | long long int stp_last_tick; | |
52a90c29 BP |
689 | |
690 | /* VLAN splinters. */ | |
691 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
692 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
e1b1d06a | 693 | |
acf60855 | 694 | /* Ports. */ |
0a740f48 EJ |
695 | struct sset ports; /* Set of standard port names. */ |
696 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
acf60855 JP |
697 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ |
698 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
abe529af BP |
699 | }; |
700 | ||
7ee20df1 BP |
701 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
702 | * for debugging the asynchronous flow_mod implementation.) */ | |
703 | static bool clogged; | |
704 | ||
b44a10b7 BP |
705 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
706 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
707 | ||
abe529af BP |
708 | static void ofproto_dpif_unixctl_init(void); |
709 | ||
710 | static struct ofproto_dpif * | |
711 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
712 | { | |
cb22974d | 713 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
714 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); |
715 | } | |
716 | ||
4acbc98d | 717 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 718 | uint16_t ofp_port); |
4acbc98d | 719 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 720 | uint32_t odp_port); |
6a6455e5 EJ |
721 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
722 | const struct ofpbuf *, ovs_be16 initial_tci, | |
723 | struct ds *); | |
abe529af BP |
724 | |
725 | /* Packet processing. */ | |
726 | static void update_learning_table(struct ofproto_dpif *, | |
727 | const struct flow *, int vlan, | |
728 | struct ofbundle *); | |
501f8d1f BP |
729 | /* Upcalls. */ |
730 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 731 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
732 | |
733 | /* Flow expiration. */ | |
acf60855 | 734 | static int expire(struct dpif_backer *); |
abe529af | 735 | |
6fca1ffb BP |
736 | /* NetFlow. */ |
737 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
738 | ||
abe529af | 739 | /* Utilities. */ |
52a90c29 | 740 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
741 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
742 | struct ofpbuf *odp_actions, | |
743 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
744 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
745 | const struct flow *flow); | |
abe529af BP |
746 | /* Global variables. */ |
747 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
748 | |
749 | /* Initial mappings of port to bridge mappings. */ | |
750 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
abe529af BP |
751 | \f |
752 | /* Factory functions. */ | |
753 | ||
b0408fca | 754 | static void |
acf60855 | 755 | init(const struct shash *iface_hints) |
b0408fca | 756 | { |
acf60855 JP |
757 | struct shash_node *node; |
758 | ||
759 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
760 | SHASH_FOR_EACH(node, iface_hints) { | |
761 | const struct iface_hint *orig_hint = node->data; | |
762 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
763 | ||
764 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
765 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
766 | new_hint->ofp_port = orig_hint->ofp_port; | |
767 | ||
768 | shash_add(&init_ofp_ports, node->name, new_hint); | |
769 | } | |
b0408fca JP |
770 | } |
771 | ||
abe529af BP |
772 | static void |
773 | enumerate_types(struct sset *types) | |
774 | { | |
775 | dp_enumerate_types(types); | |
776 | } | |
777 | ||
778 | static int | |
779 | enumerate_names(const char *type, struct sset *names) | |
780 | { | |
acf60855 JP |
781 | struct ofproto_dpif *ofproto; |
782 | ||
783 | sset_clear(names); | |
784 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
785 | if (strcmp(type, ofproto->up.type)) { | |
786 | continue; | |
787 | } | |
788 | sset_add(names, ofproto->up.name); | |
789 | } | |
790 | ||
791 | return 0; | |
abe529af BP |
792 | } |
793 | ||
794 | static int | |
795 | del(const char *type, const char *name) | |
796 | { | |
797 | struct dpif *dpif; | |
798 | int error; | |
799 | ||
800 | error = dpif_open(name, type, &dpif); | |
801 | if (!error) { | |
802 | error = dpif_delete(dpif); | |
803 | dpif_close(dpif); | |
804 | } | |
805 | return error; | |
806 | } | |
807 | \f | |
0aeaabc8 JP |
808 | static const char * |
809 | port_open_type(const char *datapath_type, const char *port_type) | |
810 | { | |
811 | return dpif_port_open_type(datapath_type, port_type); | |
812 | } | |
813 | ||
acf60855 JP |
814 | /* Type functions. */ |
815 | ||
476cb42a BP |
816 | static struct ofproto_dpif * |
817 | lookup_ofproto_dpif_by_port_name(const char *name) | |
818 | { | |
819 | struct ofproto_dpif *ofproto; | |
820 | ||
821 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
822 | if (sset_contains(&ofproto->ports, name)) { | |
823 | return ofproto; | |
824 | } | |
825 | } | |
826 | ||
827 | return NULL; | |
828 | } | |
829 | ||
acf60855 JP |
830 | static int |
831 | type_run(const char *type) | |
832 | { | |
833 | struct dpif_backer *backer; | |
834 | char *devname; | |
835 | int error; | |
836 | ||
837 | backer = shash_find_data(&all_dpif_backers, type); | |
838 | if (!backer) { | |
839 | /* This is not necessarily a problem, since backers are only | |
840 | * created on demand. */ | |
841 | return 0; | |
842 | } | |
843 | ||
844 | dpif_run(backer->dpif); | |
845 | ||
2cc3c58e EJ |
846 | if (backer->need_revalidate |
847 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
848 | struct tag_set revalidate_set = backer->revalidate_set; | |
849 | bool need_revalidate = backer->need_revalidate; | |
850 | struct ofproto_dpif *ofproto; | |
a614d823 KM |
851 | struct simap_node *node; |
852 | struct simap tmp_backers; | |
853 | ||
854 | /* Handle tunnel garbage collection. */ | |
855 | simap_init(&tmp_backers); | |
856 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
857 | ||
858 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
859 | struct ofport_dpif *iter; | |
860 | ||
861 | if (backer != ofproto->backer) { | |
862 | continue; | |
863 | } | |
864 | ||
865 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
866 | const char *dp_port; | |
867 | ||
868 | if (!iter->tnl_port) { | |
869 | continue; | |
870 | } | |
871 | ||
872 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev); | |
873 | node = simap_find(&tmp_backers, dp_port); | |
874 | if (node) { | |
875 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
876 | simap_delete(&tmp_backers, node); | |
877 | node = simap_find(&backer->tnl_backers, dp_port); | |
878 | } else { | |
879 | node = simap_find(&backer->tnl_backers, dp_port); | |
880 | if (!node) { | |
881 | uint32_t odp_port = UINT32_MAX; | |
882 | ||
883 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
884 | &odp_port)) { | |
885 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
886 | node = simap_find(&backer->tnl_backers, dp_port); | |
887 | } | |
888 | } | |
889 | } | |
890 | ||
891 | iter->odp_port = node ? node->data : OVSP_NONE; | |
892 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
893 | &iter->tnl_port)) { | |
894 | backer->need_revalidate = REV_RECONFIGURE; | |
895 | } | |
896 | } | |
897 | } | |
898 | ||
899 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
900 | dpif_port_del(backer->dpif, node->data); | |
901 | } | |
902 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
903 | |
904 | switch (backer->need_revalidate) { | |
905 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
906 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
907 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
908 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
909 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
910 | } | |
911 | ||
8f73d537 EJ |
912 | if (backer->need_revalidate) { |
913 | /* Clear the drop_keys in case we should now be accepting some | |
914 | * formerly dropped flows. */ | |
915 | drop_key_clear(backer); | |
916 | } | |
917 | ||
f728af2e BP |
918 | /* Clear the revalidation flags. */ |
919 | tag_set_init(&backer->revalidate_set); | |
920 | backer->need_revalidate = 0; | |
921 | ||
2cc3c58e EJ |
922 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
923 | struct facet *facet; | |
924 | ||
925 | if (ofproto->backer != backer) { | |
926 | continue; | |
927 | } | |
928 | ||
2cc3c58e EJ |
929 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
930 | if (need_revalidate | |
931 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
932 | facet_revalidate(facet); | |
933 | } | |
934 | } | |
935 | } | |
2cc3c58e EJ |
936 | } |
937 | ||
acf60855 JP |
938 | if (timer_expired(&backer->next_expiration)) { |
939 | int delay = expire(backer); | |
940 | timer_set_duration(&backer->next_expiration, delay); | |
941 | } | |
942 | ||
943 | /* Check for port changes in the dpif. */ | |
944 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
476cb42a | 945 | struct ofproto_dpif *ofproto; |
acf60855 JP |
946 | struct dpif_port port; |
947 | ||
948 | /* Don't report on the datapath's device. */ | |
949 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
c83b89ab | 950 | goto next; |
acf60855 JP |
951 | } |
952 | ||
b9ad7294 EJ |
953 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, |
954 | &all_ofproto_dpifs) { | |
7d82ab2e | 955 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { |
b9ad7294 EJ |
956 | goto next; |
957 | } | |
958 | } | |
959 | ||
476cb42a | 960 | ofproto = lookup_ofproto_dpif_by_port_name(devname); |
acf60855 JP |
961 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { |
962 | /* The port was removed. If we know the datapath, | |
963 | * report it through poll_set(). If we don't, it may be | |
964 | * notifying us of a removal we initiated, so ignore it. | |
965 | * If there's a pending ENOBUFS, let it stand, since | |
966 | * everything will be reevaluated. */ | |
967 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
968 | sset_add(&ofproto->port_poll_set, devname); | |
969 | ofproto->port_poll_errno = 0; | |
970 | } | |
acf60855 JP |
971 | } else if (!ofproto) { |
972 | /* The port was added, but we don't know with which | |
973 | * ofproto we should associate it. Delete it. */ | |
974 | dpif_port_del(backer->dpif, port.port_no); | |
975 | } | |
5b5e6a4c | 976 | dpif_port_destroy(&port); |
acf60855 | 977 | |
c83b89ab | 978 | next: |
acf60855 JP |
979 | free(devname); |
980 | } | |
981 | ||
982 | if (error != EAGAIN) { | |
983 | struct ofproto_dpif *ofproto; | |
984 | ||
985 | /* There was some sort of error, so propagate it to all | |
986 | * ofprotos that use this backer. */ | |
987 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
988 | &all_ofproto_dpifs) { | |
989 | if (ofproto->backer == backer) { | |
990 | sset_clear(&ofproto->port_poll_set); | |
991 | ofproto->port_poll_errno = error; | |
992 | } | |
993 | } | |
994 | } | |
995 | ||
996 | return 0; | |
997 | } | |
998 | ||
999 | static int | |
1000 | type_run_fast(const char *type) | |
1001 | { | |
1002 | struct dpif_backer *backer; | |
1003 | unsigned int work; | |
1004 | ||
1005 | backer = shash_find_data(&all_dpif_backers, type); | |
1006 | if (!backer) { | |
1007 | /* This is not necessarily a problem, since backers are only | |
1008 | * created on demand. */ | |
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | /* Handle one or more batches of upcalls, until there's nothing left to do | |
1013 | * or until we do a fixed total amount of work. | |
1014 | * | |
1015 | * We do work in batches because it can be much cheaper to set up a number | |
1016 | * of flows and fire off their patches all at once. We do multiple batches | |
1017 | * because in some cases handling a packet can cause another packet to be | |
1018 | * queued almost immediately as part of the return flow. Both | |
1019 | * optimizations can make major improvements on some benchmarks and | |
1020 | * presumably for real traffic as well. */ | |
1021 | work = 0; | |
1022 | while (work < FLOW_MISS_MAX_BATCH) { | |
1023 | int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work); | |
1024 | if (retval <= 0) { | |
1025 | return -retval; | |
1026 | } | |
1027 | work += retval; | |
1028 | } | |
1029 | ||
1030 | return 0; | |
1031 | } | |
1032 | ||
1033 | static void | |
1034 | type_wait(const char *type) | |
1035 | { | |
1036 | struct dpif_backer *backer; | |
1037 | ||
1038 | backer = shash_find_data(&all_dpif_backers, type); | |
1039 | if (!backer) { | |
1040 | /* This is not necessarily a problem, since backers are only | |
1041 | * created on demand. */ | |
1042 | return; | |
1043 | } | |
1044 | ||
1045 | timer_wait(&backer->next_expiration); | |
1046 | } | |
1047 | \f | |
abe529af BP |
1048 | /* Basic life-cycle. */ |
1049 | ||
c57b2226 BP |
1050 | static int add_internal_flows(struct ofproto_dpif *); |
1051 | ||
abe529af BP |
1052 | static struct ofproto * |
1053 | alloc(void) | |
1054 | { | |
1055 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1056 | return &ofproto->up; | |
1057 | } | |
1058 | ||
1059 | static void | |
1060 | dealloc(struct ofproto *ofproto_) | |
1061 | { | |
1062 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1063 | free(ofproto); | |
1064 | } | |
1065 | ||
acf60855 JP |
1066 | static void |
1067 | close_dpif_backer(struct dpif_backer *backer) | |
1068 | { | |
1069 | struct shash_node *node; | |
1070 | ||
cb22974d | 1071 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1072 | |
1073 | if (--backer->refcount) { | |
1074 | return; | |
1075 | } | |
1076 | ||
8f73d537 EJ |
1077 | drop_key_clear(backer); |
1078 | hmap_destroy(&backer->drop_keys); | |
1079 | ||
7d82ab2e | 1080 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1081 | hmap_destroy(&backer->odp_to_ofport_map); |
1082 | node = shash_find(&all_dpif_backers, backer->type); | |
1083 | free(backer->type); | |
1084 | shash_delete(&all_dpif_backers, node); | |
1085 | dpif_close(backer->dpif); | |
1086 | ||
1087 | free(backer); | |
1088 | } | |
1089 | ||
1090 | /* Datapath port slated for removal from datapath. */ | |
1091 | struct odp_garbage { | |
1092 | struct list list_node; | |
1093 | uint32_t odp_port; | |
1094 | }; | |
1095 | ||
1096 | static int | |
1097 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1098 | { | |
1099 | struct dpif_backer *backer; | |
1100 | struct dpif_port_dump port_dump; | |
1101 | struct dpif_port port; | |
1102 | struct shash_node *node; | |
1103 | struct list garbage_list; | |
1104 | struct odp_garbage *garbage, *next; | |
1105 | struct sset names; | |
1106 | char *backer_name; | |
1107 | const char *name; | |
1108 | int error; | |
1109 | ||
1110 | backer = shash_find_data(&all_dpif_backers, type); | |
1111 | if (backer) { | |
1112 | backer->refcount++; | |
1113 | *backerp = backer; | |
1114 | return 0; | |
1115 | } | |
1116 | ||
1117 | backer_name = xasprintf("ovs-%s", type); | |
1118 | ||
1119 | /* Remove any existing datapaths, since we assume we're the only | |
1120 | * userspace controlling the datapath. */ | |
1121 | sset_init(&names); | |
1122 | dp_enumerate_names(type, &names); | |
1123 | SSET_FOR_EACH(name, &names) { | |
1124 | struct dpif *old_dpif; | |
1125 | ||
1126 | /* Don't remove our backer if it exists. */ | |
1127 | if (!strcmp(name, backer_name)) { | |
1128 | continue; | |
1129 | } | |
1130 | ||
1131 | if (dpif_open(name, type, &old_dpif)) { | |
1132 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1133 | } else { | |
1134 | dpif_delete(old_dpif); | |
1135 | dpif_close(old_dpif); | |
1136 | } | |
1137 | } | |
1138 | sset_destroy(&names); | |
1139 | ||
1140 | backer = xmalloc(sizeof *backer); | |
1141 | ||
1142 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1143 | free(backer_name); | |
1144 | if (error) { | |
1145 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1146 | strerror(error)); | |
4c1b1289 | 1147 | free(backer); |
acf60855 JP |
1148 | return error; |
1149 | } | |
1150 | ||
1151 | backer->type = xstrdup(type); | |
1152 | backer->refcount = 1; | |
1153 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1154 | hmap_init(&backer->drop_keys); |
acf60855 | 1155 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1156 | backer->need_revalidate = 0; |
7d82ab2e | 1157 | simap_init(&backer->tnl_backers); |
2cc3c58e | 1158 | tag_set_init(&backer->revalidate_set); |
acf60855 JP |
1159 | *backerp = backer; |
1160 | ||
1161 | dpif_flow_flush(backer->dpif); | |
1162 | ||
1163 | /* Loop through the ports already on the datapath and remove any | |
1164 | * that we don't need anymore. */ | |
1165 | list_init(&garbage_list); | |
1166 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1167 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1168 | node = shash_find(&init_ofp_ports, port.name); | |
1169 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1170 | garbage = xmalloc(sizeof *garbage); | |
1171 | garbage->odp_port = port.port_no; | |
1172 | list_push_front(&garbage_list, &garbage->list_node); | |
1173 | } | |
1174 | } | |
1175 | dpif_port_dump_done(&port_dump); | |
1176 | ||
1177 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1178 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1179 | list_remove(&garbage->list_node); | |
1180 | free(garbage); | |
1181 | } | |
1182 | ||
1183 | shash_add(&all_dpif_backers, type, backer); | |
1184 | ||
1185 | error = dpif_recv_set(backer->dpif, true); | |
1186 | if (error) { | |
1187 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1188 | type, strerror(error)); | |
1189 | close_dpif_backer(backer); | |
1190 | return error; | |
1191 | } | |
1192 | ||
1193 | return error; | |
1194 | } | |
1195 | ||
abe529af | 1196 | static int |
0f5f95a9 | 1197 | construct(struct ofproto *ofproto_) |
abe529af BP |
1198 | { |
1199 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1200 | struct shash_node *node, *next; |
91858960 | 1201 | int max_ports; |
abe529af BP |
1202 | int error; |
1203 | int i; | |
1204 | ||
acf60855 | 1205 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1206 | if (error) { |
abe529af BP |
1207 | return error; |
1208 | } | |
1209 | ||
acf60855 | 1210 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
91858960 BP |
1211 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); |
1212 | ||
6c1491fb | 1213 | ofproto->n_matches = 0; |
abe529af | 1214 | |
abe529af BP |
1215 | ofproto->netflow = NULL; |
1216 | ofproto->sflow = NULL; | |
21f7563c | 1217 | ofproto->stp = NULL; |
abe529af | 1218 | hmap_init(&ofproto->bundles); |
e764773c | 1219 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
1220 | for (i = 0; i < MAX_MIRRORS; i++) { |
1221 | ofproto->mirrors[i] = NULL; | |
1222 | } | |
1223 | ofproto->has_bonded_bundles = false; | |
1224 | ||
abe529af | 1225 | hmap_init(&ofproto->facets); |
b0f7b9b5 | 1226 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 1227 | ofproto->governor = NULL; |
54a9cbc9 BP |
1228 | |
1229 | for (i = 0; i < N_TABLES; i++) { | |
1230 | struct table_dpif *table = &ofproto->tables[i]; | |
1231 | ||
1232 | table->catchall_table = NULL; | |
1233 | table->other_table = NULL; | |
1234 | table->basis = random_uint32(); | |
1235 | } | |
abe529af | 1236 | |
7ee20df1 BP |
1237 | list_init(&ofproto->completions); |
1238 | ||
abe529af BP |
1239 | ofproto_dpif_unixctl_init(); |
1240 | ||
ccb7c863 | 1241 | ofproto->has_mirrors = false; |
daff3353 EJ |
1242 | ofproto->has_bundle_action = false; |
1243 | ||
52a90c29 BP |
1244 | hmap_init(&ofproto->vlandev_map); |
1245 | hmap_init(&ofproto->realdev_vid_map); | |
1246 | ||
acf60855 | 1247 | sset_init(&ofproto->ports); |
0a740f48 | 1248 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1249 | sset_init(&ofproto->port_poll_set); |
1250 | ofproto->port_poll_errno = 0; | |
1251 | ||
1252 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1253 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1254 | |
1255 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1256 | /* Check if the datapath already has this port. */ | |
1257 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1258 | sset_add(&ofproto->ports, node->name); | |
1259 | } | |
1260 | ||
1261 | free(iface_hint->br_name); | |
1262 | free(iface_hint->br_type); | |
4f9e08a5 | 1263 | free(iface_hint); |
acf60855 JP |
1264 | shash_delete(&init_ofp_ports, node); |
1265 | } | |
1266 | } | |
e1b1d06a | 1267 | |
b44a10b7 BP |
1268 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1269 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1270 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1271 | |
1272 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1273 | error = add_internal_flows(ofproto); |
1274 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1275 | ||
1276 | return error; | |
1277 | } | |
1278 | ||
1279 | static int | |
1280 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1281 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1282 | { |
1283 | struct ofputil_flow_mod fm; | |
1284 | int error; | |
1285 | ||
81a76618 BP |
1286 | match_init_catchall(&fm.match); |
1287 | fm.priority = 0; | |
1288 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1289 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1290 | fm.cookie = htonll(0); |
1291 | fm.cookie_mask = htonll(0); | |
1292 | fm.table_id = TBL_INTERNAL; | |
1293 | fm.command = OFPFC_ADD; | |
1294 | fm.idle_timeout = 0; | |
1295 | fm.hard_timeout = 0; | |
1296 | fm.buffer_id = 0; | |
1297 | fm.out_port = 0; | |
1298 | fm.flags = 0; | |
f25d0cf3 BP |
1299 | fm.ofpacts = ofpacts->data; |
1300 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1301 | |
1302 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1303 | if (error) { | |
1304 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1305 | id, ofperr_to_string(error)); | |
1306 | return error; | |
1307 | } | |
1308 | ||
81a76618 | 1309 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); |
cb22974d | 1310 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1311 | |
abe529af BP |
1312 | return 0; |
1313 | } | |
1314 | ||
c57b2226 BP |
1315 | static int |
1316 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1317 | { | |
f25d0cf3 BP |
1318 | struct ofpact_controller *controller; |
1319 | uint64_t ofpacts_stub[128 / 8]; | |
1320 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1321 | int error; |
1322 | int id; | |
1323 | ||
f25d0cf3 | 1324 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1325 | id = 1; |
1326 | ||
f25d0cf3 BP |
1327 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1328 | controller->max_len = UINT16_MAX; | |
1329 | controller->controller_id = 0; | |
1330 | controller->reason = OFPR_NO_MATCH; | |
1331 | ofpact_pad(&ofpacts); | |
1332 | ||
1333 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1334 | if (error) { |
1335 | return error; | |
1336 | } | |
1337 | ||
f25d0cf3 BP |
1338 | ofpbuf_clear(&ofpacts); |
1339 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
1340 | &ofproto->no_packet_in_rule); |
1341 | return error; | |
1342 | } | |
1343 | ||
7ee20df1 BP |
1344 | static void |
1345 | complete_operations(struct ofproto_dpif *ofproto) | |
1346 | { | |
1347 | struct dpif_completion *c, *next; | |
1348 | ||
1349 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1350 | ofoperation_complete(c->op, 0); | |
1351 | list_remove(&c->list_node); | |
1352 | free(c); | |
1353 | } | |
1354 | } | |
1355 | ||
abe529af BP |
1356 | static void |
1357 | destruct(struct ofproto *ofproto_) | |
1358 | { | |
1359 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1360 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1361 | struct oftable *table; |
abe529af BP |
1362 | int i; |
1363 | ||
b44a10b7 | 1364 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1365 | complete_operations(ofproto); |
1366 | ||
0697b5c3 BP |
1367 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1368 | struct cls_cursor cursor; | |
1369 | ||
d0918789 | 1370 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1371 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1372 | ofproto_rule_destroy(&rule->up); | |
1373 | } | |
7ee20df1 BP |
1374 | } |
1375 | ||
abe529af BP |
1376 | for (i = 0; i < MAX_MIRRORS; i++) { |
1377 | mirror_destroy(ofproto->mirrors[i]); | |
1378 | } | |
1379 | ||
1380 | netflow_destroy(ofproto->netflow); | |
bae473fe | 1381 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
1382 | hmap_destroy(&ofproto->bundles); |
1383 | mac_learning_destroy(ofproto->ml); | |
1384 | ||
1385 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 1386 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 1387 | governor_destroy(ofproto->governor); |
abe529af | 1388 | |
52a90c29 BP |
1389 | hmap_destroy(&ofproto->vlandev_map); |
1390 | hmap_destroy(&ofproto->realdev_vid_map); | |
1391 | ||
acf60855 | 1392 | sset_destroy(&ofproto->ports); |
0a740f48 | 1393 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1394 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1395 | |
acf60855 | 1396 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1397 | } |
1398 | ||
1399 | static int | |
5fcc0d00 | 1400 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1401 | { |
1402 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 1403 | struct ofport_dpif *ofport; |
abe529af | 1404 | |
0aa66d6e EJ |
1405 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1406 | port_run_fast(ofport); | |
1407 | } | |
1408 | ||
5fcc0d00 BP |
1409 | return 0; |
1410 | } | |
1411 | ||
1412 | static int | |
1413 | run(struct ofproto *ofproto_) | |
1414 | { | |
1415 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1416 | struct ofport_dpif *ofport; | |
1417 | struct ofbundle *bundle; | |
1418 | int error; | |
1419 | ||
1420 | if (!clogged) { | |
1421 | complete_operations(ofproto); | |
1422 | } | |
5fcc0d00 BP |
1423 | |
1424 | error = run_fast(ofproto_); | |
1425 | if (error) { | |
1426 | return error; | |
abe529af BP |
1427 | } |
1428 | ||
abe529af | 1429 | if (ofproto->netflow) { |
6fca1ffb BP |
1430 | if (netflow_run(ofproto->netflow)) { |
1431 | send_netflow_active_timeouts(ofproto); | |
1432 | } | |
abe529af BP |
1433 | } |
1434 | if (ofproto->sflow) { | |
bae473fe | 1435 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1436 | } |
1437 | ||
1438 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1439 | port_run(ofport); | |
1440 | } | |
1441 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1442 | bundle_run(bundle); | |
1443 | } | |
1444 | ||
21f7563c | 1445 | stp_run(ofproto); |
2cc3c58e | 1446 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af | 1447 | |
6814e51f | 1448 | /* Check the consistency of a random facet, to aid debugging. */ |
2cc3c58e EJ |
1449 | if (!hmap_is_empty(&ofproto->facets) |
1450 | && !ofproto->backer->need_revalidate) { | |
6814e51f BP |
1451 | struct facet *facet; |
1452 | ||
1453 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
1454 | struct facet, hmap_node); | |
2cc3c58e EJ |
1455 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, |
1456 | facet->tags)) { | |
6814e51f | 1457 | if (!facet_check_consistency(facet)) { |
2cc3c58e | 1458 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1459 | } |
1460 | } | |
1461 | } | |
1462 | ||
9d6ac44e BP |
1463 | if (ofproto->governor) { |
1464 | size_t n_subfacets; | |
1465 | ||
1466 | governor_run(ofproto->governor); | |
1467 | ||
1468 | /* If the governor has shrunk to its minimum size and the number of | |
1469 | * subfacets has dwindled, then drop the governor entirely. | |
1470 | * | |
1471 | * For hysteresis, the number of subfacets to drop the governor is | |
1472 | * smaller than the number needed to trigger its creation. */ | |
1473 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1474 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1475 | && governor_is_idle(ofproto->governor)) { | |
1476 | governor_destroy(ofproto->governor); | |
1477 | ofproto->governor = NULL; | |
1478 | } | |
1479 | } | |
1480 | ||
abe529af BP |
1481 | return 0; |
1482 | } | |
1483 | ||
1484 | static void | |
1485 | wait(struct ofproto *ofproto_) | |
1486 | { | |
1487 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1488 | struct ofport_dpif *ofport; | |
1489 | struct ofbundle *bundle; | |
1490 | ||
7ee20df1 BP |
1491 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1492 | poll_immediate_wake(); | |
1493 | } | |
1494 | ||
acf60855 JP |
1495 | dpif_wait(ofproto->backer->dpif); |
1496 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1497 | if (ofproto->sflow) { |
bae473fe | 1498 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1499 | } |
2cc3c58e | 1500 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { |
abe529af BP |
1501 | poll_immediate_wake(); |
1502 | } | |
1503 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1504 | port_wait(ofport); | |
1505 | } | |
1506 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1507 | bundle_wait(bundle); | |
1508 | } | |
6fca1ffb BP |
1509 | if (ofproto->netflow) { |
1510 | netflow_wait(ofproto->netflow); | |
1511 | } | |
1c313b88 | 1512 | mac_learning_wait(ofproto->ml); |
21f7563c | 1513 | stp_wait(ofproto); |
2cc3c58e | 1514 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1515 | /* Shouldn't happen, but if it does just go around again. */ |
1516 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1517 | poll_immediate_wake(); | |
abe529af | 1518 | } |
9d6ac44e BP |
1519 | if (ofproto->governor) { |
1520 | governor_wait(ofproto->governor); | |
1521 | } | |
abe529af BP |
1522 | } |
1523 | ||
0d085684 BP |
1524 | static void |
1525 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1526 | { | |
1527 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1528 | ||
1529 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1530 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1531 | } | |
1532 | ||
abe529af BP |
1533 | static void |
1534 | flush(struct ofproto *ofproto_) | |
1535 | { | |
1536 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1537 | struct subfacet *subfacet, *next_subfacet; |
1538 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1539 | int n_batch; | |
b0f7b9b5 | 1540 | |
acf60855 JP |
1541 | n_batch = 0; |
1542 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
1543 | &ofproto->subfacets) { | |
1544 | if (subfacet->path != SF_NOT_INSTALLED) { | |
1545 | batch[n_batch++] = subfacet; | |
1546 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
1547 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1548 | n_batch = 0; | |
1549 | } | |
1550 | } else { | |
1551 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1552 | } |
abe529af | 1553 | } |
acf60855 JP |
1554 | |
1555 | if (n_batch > 0) { | |
1556 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1557 | } | |
abe529af BP |
1558 | } |
1559 | ||
6c1491fb BP |
1560 | static void |
1561 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1562 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1563 | { |
1564 | *arp_match_ip = true; | |
9e1fd49b BP |
1565 | *actions = (OFPUTIL_A_OUTPUT | |
1566 | OFPUTIL_A_SET_VLAN_VID | | |
1567 | OFPUTIL_A_SET_VLAN_PCP | | |
1568 | OFPUTIL_A_STRIP_VLAN | | |
1569 | OFPUTIL_A_SET_DL_SRC | | |
1570 | OFPUTIL_A_SET_DL_DST | | |
1571 | OFPUTIL_A_SET_NW_SRC | | |
1572 | OFPUTIL_A_SET_NW_DST | | |
1573 | OFPUTIL_A_SET_NW_TOS | | |
1574 | OFPUTIL_A_SET_TP_SRC | | |
1575 | OFPUTIL_A_SET_TP_DST | | |
1576 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1577 | } |
1578 | ||
1579 | static void | |
307975da | 1580 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1581 | { |
1582 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1583 | struct dpif_dp_stats s; |
6c1491fb BP |
1584 | |
1585 | strcpy(ots->name, "classifier"); | |
1586 | ||
acf60855 JP |
1587 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
1588 | ||
307975da SH |
1589 | ots->lookup_count = htonll(s.n_hit + s.n_missed); |
1590 | ots->matched_count = htonll(s.n_hit + ofproto->n_matches); | |
6c1491fb BP |
1591 | } |
1592 | ||
abe529af BP |
1593 | static struct ofport * |
1594 | port_alloc(void) | |
1595 | { | |
1596 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1597 | return &port->up; | |
1598 | } | |
1599 | ||
1600 | static void | |
1601 | port_dealloc(struct ofport *port_) | |
1602 | { | |
1603 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1604 | free(port); | |
1605 | } | |
1606 | ||
1607 | static int | |
1608 | port_construct(struct ofport *port_) | |
1609 | { | |
1610 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1611 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1612 | const struct netdev *netdev = port->up.netdev; |
e1b1d06a JP |
1613 | struct dpif_port dpif_port; |
1614 | int error; | |
abe529af | 1615 | |
2cc3c58e | 1616 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1617 | port->bundle = NULL; |
1618 | port->cfm = NULL; | |
1619 | port->tag = tag_create_random(); | |
d5ffa7f2 | 1620 | port->may_enable = true; |
21f7563c JP |
1621 | port->stp_port = NULL; |
1622 | port->stp_state = STP_DISABLED; | |
b9ad7294 | 1623 | port->tnl_port = NULL; |
8b36f51e | 1624 | hmap_init(&port->priorities); |
52a90c29 BP |
1625 | port->realdev_ofp_port = 0; |
1626 | port->vlandev_vid = 0; | |
b9ad7294 | 1627 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1628 | |
b9ad7294 | 1629 | if (netdev_vport_is_patch(netdev)) { |
0a740f48 EJ |
1630 | /* XXX By bailing out here, we don't do required sFlow work. */ |
1631 | port->odp_port = OVSP_NONE; | |
1632 | return 0; | |
1633 | } | |
1634 | ||
acf60855 | 1635 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
b9ad7294 | 1636 | netdev_vport_get_dpif_port(netdev), |
e1b1d06a JP |
1637 | &dpif_port); |
1638 | if (error) { | |
1639 | return error; | |
1640 | } | |
1641 | ||
1642 | port->odp_port = dpif_port.port_no; | |
1643 | ||
b9ad7294 EJ |
1644 | if (netdev_get_tunnel_config(netdev)) { |
1645 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1646 | } else { | |
1647 | /* Sanity-check that a mapping doesn't already exist. This | |
1648 | * shouldn't happen for non-tunnel ports. */ | |
1649 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1650 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1651 | dpif_port.name); | |
da78d43d | 1652 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1653 | return EBUSY; |
1654 | } | |
e1b1d06a | 1655 | |
b9ad7294 EJ |
1656 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
1657 | hash_int(port->odp_port, 0)); | |
1658 | } | |
da78d43d | 1659 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1660 | |
abe529af | 1661 | if (ofproto->sflow) { |
e1b1d06a | 1662 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1663 | } |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | static void | |
1669 | port_destruct(struct ofport *port_) | |
1670 | { | |
1671 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1672 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1673 | const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); |
02f8d646 | 1674 | const char *devname = netdev_get_name(port->up.netdev); |
abe529af | 1675 | |
a614d823 | 1676 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1677 | /* The underlying device is still there, so delete it. This |
1678 | * happens when the ofproto is being destroyed, since the caller | |
1679 | * assumes that removal of attached ports will happen as part of | |
1680 | * destruction. */ | |
a614d823 KM |
1681 | if (!port->tnl_port) { |
1682 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1683 | } | |
1684 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
acf60855 JP |
1685 | } |
1686 | ||
b9ad7294 | 1687 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { |
0a740f48 EJ |
1688 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1689 | } | |
1690 | ||
b9ad7294 | 1691 | tnl_port_del(port->tnl_port); |
02f8d646 | 1692 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1693 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
2cc3c58e | 1694 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1695 | bundle_remove(port_); |
a5610457 | 1696 | set_cfm(port_, NULL); |
abe529af | 1697 | if (ofproto->sflow) { |
bae473fe | 1698 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1699 | } |
8b36f51e EJ |
1700 | |
1701 | ofport_clear_priorities(port); | |
1702 | hmap_destroy(&port->priorities); | |
abe529af BP |
1703 | } |
1704 | ||
1705 | static void | |
1706 | port_modified(struct ofport *port_) | |
1707 | { | |
1708 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1709 | ||
1710 | if (port->bundle && port->bundle->bond) { | |
1711 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | static void | |
9e1fd49b | 1716 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1717 | { |
1718 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1719 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1720 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1721 | |
9e1fd49b | 1722 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1723 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1724 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1725 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1726 | |
9e1fd49b | 1727 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1728 | bundle_update(port->bundle); |
1729 | } | |
abe529af BP |
1730 | } |
1731 | } | |
1732 | ||
1733 | static int | |
1734 | set_sflow(struct ofproto *ofproto_, | |
1735 | const struct ofproto_sflow_options *sflow_options) | |
1736 | { | |
1737 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1738 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1739 | |
abe529af | 1740 | if (sflow_options) { |
bae473fe | 1741 | if (!ds) { |
abe529af BP |
1742 | struct ofport_dpif *ofport; |
1743 | ||
4213f19d | 1744 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1745 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1746 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1747 | } |
2cc3c58e | 1748 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1749 | } |
bae473fe | 1750 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1751 | } else { |
6ff686f2 PS |
1752 | if (ds) { |
1753 | dpif_sflow_destroy(ds); | |
2cc3c58e | 1754 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1755 | ofproto->sflow = NULL; |
1756 | } | |
abe529af BP |
1757 | } |
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | static int | |
a5610457 | 1762 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1763 | { |
1764 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1765 | int error; | |
1766 | ||
a5610457 | 1767 | if (!s) { |
abe529af BP |
1768 | error = 0; |
1769 | } else { | |
1770 | if (!ofport->cfm) { | |
8c977421 EJ |
1771 | struct ofproto_dpif *ofproto; |
1772 | ||
1773 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1774 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f629657 | 1775 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1776 | } |
1777 | ||
a5610457 | 1778 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1779 | return 0; |
1780 | } | |
1781 | ||
1782 | error = EINVAL; | |
1783 | } | |
1784 | cfm_destroy(ofport->cfm); | |
1785 | ofport->cfm = NULL; | |
1786 | return error; | |
1787 | } | |
1788 | ||
1789 | static int | |
a5610457 | 1790 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1791 | { |
1792 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1793 | |
1794 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1795 | } |
1de11730 | 1796 | |
1c0333b6 EJ |
1797 | static int |
1798 | get_cfm_opup(const struct ofport *ofport_) | |
1799 | { | |
1800 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1801 | ||
1802 | return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; | |
1803 | } | |
1804 | ||
1de11730 EJ |
1805 | static int |
1806 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1807 | size_t *n_rmps) | |
1808 | { | |
1809 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1810 | ||
1811 | if (ofport->cfm) { | |
1812 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1813 | return 0; | |
1814 | } else { | |
1815 | return -1; | |
1816 | } | |
1817 | } | |
3967a833 MM |
1818 | |
1819 | static int | |
1820 | get_cfm_health(const struct ofport *ofport_) | |
1821 | { | |
1822 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1823 | ||
1824 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1825 | } | |
abe529af | 1826 | \f |
21f7563c JP |
1827 | /* Spanning Tree. */ |
1828 | ||
1829 | static void | |
1830 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1831 | { | |
1832 | struct ofproto_dpif *ofproto = ofproto_; | |
1833 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1834 | struct ofport_dpif *ofport; | |
1835 | ||
1836 | ofport = stp_port_get_aux(sp); | |
1837 | if (!ofport) { | |
1838 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1839 | ofproto->up.name, port_num); | |
1840 | } else { | |
1841 | struct eth_header *eth = pkt->l2; | |
1842 | ||
1843 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1844 | if (eth_addr_is_zero(eth->eth_src)) { | |
1845 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1846 | "with unknown MAC", ofproto->up.name, port_num); | |
1847 | } else { | |
97d6520b | 1848 | send_packet(ofport, pkt); |
21f7563c JP |
1849 | } |
1850 | } | |
1851 | ofpbuf_delete(pkt); | |
1852 | } | |
1853 | ||
1854 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1855 | static int | |
1856 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1857 | { | |
1858 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1859 | ||
1860 | /* Only revalidate flows if the configuration changed. */ | |
1861 | if (!s != !ofproto->stp) { | |
2cc3c58e | 1862 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
1863 | } |
1864 | ||
1865 | if (s) { | |
1866 | if (!ofproto->stp) { | |
1867 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1868 | send_bpdu_cb, ofproto); | |
1869 | ofproto->stp_last_tick = time_msec(); | |
1870 | } | |
1871 | ||
1872 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1873 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1874 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1875 | stp_set_max_age(ofproto->stp, s->max_age); | |
1876 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1877 | } else { | |
851bf71d EJ |
1878 | struct ofport *ofport; |
1879 | ||
1880 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1881 | set_stp_port(ofport, NULL); | |
1882 | } | |
1883 | ||
21f7563c JP |
1884 | stp_destroy(ofproto->stp); |
1885 | ofproto->stp = NULL; | |
1886 | } | |
1887 | ||
1888 | return 0; | |
1889 | } | |
1890 | ||
1891 | static int | |
1892 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1893 | { | |
1894 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1895 | ||
1896 | if (ofproto->stp) { | |
1897 | s->enabled = true; | |
1898 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1899 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1900 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1901 | } else { | |
1902 | s->enabled = false; | |
1903 | } | |
1904 | ||
1905 | return 0; | |
1906 | } | |
1907 | ||
1908 | static void | |
1909 | update_stp_port_state(struct ofport_dpif *ofport) | |
1910 | { | |
1911 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1912 | enum stp_state state; | |
1913 | ||
1914 | /* Figure out new state. */ | |
1915 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1916 | : STP_DISABLED; | |
1917 | ||
1918 | /* Update state. */ | |
1919 | if (ofport->stp_state != state) { | |
9e1fd49b | 1920 | enum ofputil_port_state of_state; |
21f7563c JP |
1921 | bool fwd_change; |
1922 | ||
1923 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1924 | netdev_get_name(ofport->up.netdev), | |
1925 | stp_state_name(ofport->stp_state), | |
1926 | stp_state_name(state)); | |
1927 | if (stp_learn_in_state(ofport->stp_state) | |
1928 | != stp_learn_in_state(state)) { | |
1929 | /* xxx Learning action flows should also be flushed. */ | |
2cc3c58e EJ |
1930 | mac_learning_flush(ofproto->ml, |
1931 | &ofproto->backer->revalidate_set); | |
21f7563c JP |
1932 | } |
1933 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1934 | != stp_forward_in_state(state); | |
1935 | ||
2cc3c58e | 1936 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
1937 | ofport->stp_state = state; |
1938 | ofport->stp_state_entered = time_msec(); | |
1939 | ||
b308140a | 1940 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1941 | bundle_update(ofport->bundle); |
1942 | } | |
1943 | ||
1944 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1945 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1946 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1947 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1948 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1949 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1950 | : 0); | |
21f7563c JP |
1951 | ofproto_port_set_state(&ofport->up, of_state); |
1952 | } | |
1953 | } | |
1954 | ||
1955 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1956 | * caller is responsible for assigning STP port numbers and ensuring | |
1957 | * there are no duplicates. */ | |
1958 | static int | |
1959 | set_stp_port(struct ofport *ofport_, | |
1960 | const struct ofproto_port_stp_settings *s) | |
1961 | { | |
1962 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1963 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1964 | struct stp_port *sp = ofport->stp_port; | |
1965 | ||
1966 | if (!s || !s->enable) { | |
1967 | if (sp) { | |
1968 | ofport->stp_port = NULL; | |
1969 | stp_port_disable(sp); | |
ecd12731 | 1970 | update_stp_port_state(ofport); |
21f7563c JP |
1971 | } |
1972 | return 0; | |
1973 | } else if (sp && stp_port_no(sp) != s->port_num | |
1974 | && ofport == stp_port_get_aux(sp)) { | |
1975 | /* The port-id changed, so disable the old one if it's not | |
1976 | * already in use by another port. */ | |
1977 | stp_port_disable(sp); | |
1978 | } | |
1979 | ||
1980 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1981 | stp_port_enable(sp); | |
1982 | ||
1983 | stp_port_set_aux(sp, ofport); | |
1984 | stp_port_set_priority(sp, s->priority); | |
1985 | stp_port_set_path_cost(sp, s->path_cost); | |
1986 | ||
1987 | update_stp_port_state(ofport); | |
1988 | ||
1989 | return 0; | |
1990 | } | |
1991 | ||
1992 | static int | |
1993 | get_stp_port_status(struct ofport *ofport_, | |
1994 | struct ofproto_port_stp_status *s) | |
1995 | { | |
1996 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1997 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1998 | struct stp_port *sp = ofport->stp_port; | |
1999 | ||
2000 | if (!ofproto->stp || !sp) { | |
2001 | s->enabled = false; | |
2002 | return 0; | |
2003 | } | |
2004 | ||
2005 | s->enabled = true; | |
2006 | s->port_id = stp_port_get_id(sp); | |
2007 | s->state = stp_port_get_state(sp); | |
2008 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2009 | s->role = stp_port_get_role(sp); | |
80740385 | 2010 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2011 | |
2012 | return 0; | |
2013 | } | |
2014 | ||
2015 | static void | |
2016 | stp_run(struct ofproto_dpif *ofproto) | |
2017 | { | |
2018 | if (ofproto->stp) { | |
2019 | long long int now = time_msec(); | |
2020 | long long int elapsed = now - ofproto->stp_last_tick; | |
2021 | struct stp_port *sp; | |
2022 | ||
2023 | if (elapsed > 0) { | |
2024 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2025 | ofproto->stp_last_tick = now; | |
2026 | } | |
2027 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2028 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2029 | ||
2030 | if (ofport) { | |
2031 | update_stp_port_state(ofport); | |
2032 | } | |
2033 | } | |
6ae50723 EJ |
2034 | |
2035 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2cc3c58e | 2036 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
6ae50723 | 2037 | } |
21f7563c JP |
2038 | } |
2039 | } | |
2040 | ||
2041 | static void | |
2042 | stp_wait(struct ofproto_dpif *ofproto) | |
2043 | { | |
2044 | if (ofproto->stp) { | |
2045 | poll_timer_wait(1000); | |
2046 | } | |
2047 | } | |
2048 | ||
2049 | /* Returns true if STP should process 'flow'. */ | |
2050 | static bool | |
2051 | stp_should_process_flow(const struct flow *flow) | |
2052 | { | |
2053 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2054 | } | |
2055 | ||
2056 | static void | |
2057 | stp_process_packet(const struct ofport_dpif *ofport, | |
2058 | const struct ofpbuf *packet) | |
2059 | { | |
2060 | struct ofpbuf payload = *packet; | |
2061 | struct eth_header *eth = payload.data; | |
2062 | struct stp_port *sp = ofport->stp_port; | |
2063 | ||
2064 | /* Sink packets on ports that have STP disabled when the bridge has | |
2065 | * STP enabled. */ | |
2066 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2067 | return; | |
2068 | } | |
2069 | ||
2070 | /* Trim off padding on payload. */ | |
c573540b BP |
2071 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
2072 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
2073 | } |
2074 | ||
2075 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2076 | stp_received_bpdu(sp, payload.data, payload.size); | |
2077 | } | |
2078 | } | |
2079 | \f | |
8b36f51e EJ |
2080 | static struct priority_to_dscp * |
2081 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2082 | { | |
2083 | struct priority_to_dscp *pdscp; | |
2084 | uint32_t hash; | |
2085 | ||
2086 | hash = hash_int(priority, 0); | |
2087 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2088 | if (pdscp->priority == priority) { | |
2089 | return pdscp; | |
2090 | } | |
2091 | } | |
2092 | return NULL; | |
2093 | } | |
2094 | ||
2095 | static void | |
2096 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2097 | { | |
2098 | struct priority_to_dscp *pdscp, *next; | |
2099 | ||
2100 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2101 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2102 | free(pdscp); | |
2103 | } | |
2104 | } | |
2105 | ||
2106 | static int | |
2107 | set_queues(struct ofport *ofport_, | |
2108 | const struct ofproto_port_queue *qdscp_list, | |
2109 | size_t n_qdscp) | |
2110 | { | |
2111 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2112 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2113 | struct hmap new = HMAP_INITIALIZER(&new); | |
2114 | size_t i; | |
2115 | ||
2116 | for (i = 0; i < n_qdscp; i++) { | |
2117 | struct priority_to_dscp *pdscp; | |
2118 | uint32_t priority; | |
2119 | uint8_t dscp; | |
2120 | ||
2121 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
acf60855 | 2122 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, |
8b36f51e EJ |
2123 | &priority)) { |
2124 | continue; | |
2125 | } | |
2126 | ||
2127 | pdscp = get_priority(ofport, priority); | |
2128 | if (pdscp) { | |
2129 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2130 | } else { | |
2131 | pdscp = xmalloc(sizeof *pdscp); | |
2132 | pdscp->priority = priority; | |
2133 | pdscp->dscp = dscp; | |
2cc3c58e | 2134 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2135 | } |
2136 | ||
2137 | if (pdscp->dscp != dscp) { | |
2138 | pdscp->dscp = dscp; | |
2cc3c58e | 2139 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2140 | } |
2141 | ||
2142 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2143 | } | |
2144 | ||
2145 | if (!hmap_is_empty(&ofport->priorities)) { | |
2146 | ofport_clear_priorities(ofport); | |
2cc3c58e | 2147 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2148 | } |
2149 | ||
2150 | hmap_swap(&new, &ofport->priorities); | |
2151 | hmap_destroy(&new); | |
2152 | ||
2153 | return 0; | |
2154 | } | |
2155 | \f | |
abe529af BP |
2156 | /* Bundles. */ |
2157 | ||
b44a10b7 BP |
2158 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2159 | * ofproto to revalidate every flow. | |
2160 | * | |
2161 | * Normally MAC learning entries are removed only from the ofproto associated | |
2162 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2163 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2164 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2165 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2166 | * with the host from which it migrated. */ | |
abe529af | 2167 | static void |
b44a10b7 | 2168 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2169 | { |
2170 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2171 | struct mac_learning *ml = ofproto->ml; | |
2172 | struct mac_entry *mac, *next_mac; | |
2173 | ||
2cc3c58e | 2174 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2175 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2176 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2177 | if (all_ofprotos) { |
2178 | struct ofproto_dpif *o; | |
2179 | ||
2180 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2181 | if (o != ofproto) { | |
2182 | struct mac_entry *e; | |
2183 | ||
2184 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2185 | NULL); | |
2186 | if (e) { | |
b44a10b7 BP |
2187 | mac_learning_expire(o->ml, e); |
2188 | } | |
2189 | } | |
2190 | } | |
2191 | } | |
2192 | ||
abe529af BP |
2193 | mac_learning_expire(ml, mac); |
2194 | } | |
2195 | } | |
2196 | } | |
2197 | ||
2198 | static struct ofbundle * | |
2199 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2200 | { | |
2201 | struct ofbundle *bundle; | |
2202 | ||
2203 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2204 | &ofproto->bundles) { | |
2205 | if (bundle->aux == aux) { | |
2206 | return bundle; | |
2207 | } | |
2208 | } | |
2209 | return NULL; | |
2210 | } | |
2211 | ||
2212 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2213 | * ones that are found to 'bundles'. */ | |
2214 | static void | |
2215 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2216 | void **auxes, size_t n_auxes, | |
2217 | struct hmapx *bundles) | |
2218 | { | |
2219 | size_t i; | |
2220 | ||
2221 | hmapx_init(bundles); | |
2222 | for (i = 0; i < n_auxes; i++) { | |
2223 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2224 | if (bundle) { | |
2225 | hmapx_add(bundles, bundle); | |
2226 | } | |
2227 | } | |
2228 | } | |
2229 | ||
7bde8dd8 JP |
2230 | static void |
2231 | bundle_update(struct ofbundle *bundle) | |
2232 | { | |
2233 | struct ofport_dpif *port; | |
2234 | ||
2235 | bundle->floodable = true; | |
2236 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2237 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2238 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2239 | bundle->floodable = false; |
2240 | break; | |
2241 | } | |
2242 | } | |
2243 | } | |
2244 | ||
abe529af BP |
2245 | static void |
2246 | bundle_del_port(struct ofport_dpif *port) | |
2247 | { | |
2248 | struct ofbundle *bundle = port->bundle; | |
2249 | ||
2cc3c58e | 2250 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2251 | |
abe529af BP |
2252 | list_remove(&port->bundle_node); |
2253 | port->bundle = NULL; | |
2254 | ||
2255 | if (bundle->lacp) { | |
2256 | lacp_slave_unregister(bundle->lacp, port); | |
2257 | } | |
2258 | if (bundle->bond) { | |
2259 | bond_slave_unregister(bundle->bond, port); | |
2260 | } | |
2261 | ||
7bde8dd8 | 2262 | bundle_update(bundle); |
abe529af BP |
2263 | } |
2264 | ||
2265 | static bool | |
2266 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
df53d41c | 2267 | struct lacp_slave_settings *lacp) |
abe529af BP |
2268 | { |
2269 | struct ofport_dpif *port; | |
2270 | ||
2271 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2272 | if (!port) { | |
2273 | return false; | |
2274 | } | |
2275 | ||
2276 | if (port->bundle != bundle) { | |
2cc3c58e | 2277 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2278 | if (port->bundle) { |
2279 | bundle_del_port(port); | |
2280 | } | |
2281 | ||
2282 | port->bundle = bundle; | |
2283 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2284 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2285 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2286 | bundle->floodable = false; |
2287 | } | |
2288 | } | |
2289 | if (lacp) { | |
2cc3c58e | 2290 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2291 | lacp_slave_register(bundle->lacp, port, lacp); |
2292 | } | |
2293 | ||
2294 | return true; | |
2295 | } | |
2296 | ||
2297 | static void | |
2298 | bundle_destroy(struct ofbundle *bundle) | |
2299 | { | |
2300 | struct ofproto_dpif *ofproto; | |
2301 | struct ofport_dpif *port, *next_port; | |
2302 | int i; | |
2303 | ||
2304 | if (!bundle) { | |
2305 | return; | |
2306 | } | |
2307 | ||
2308 | ofproto = bundle->ofproto; | |
2309 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2310 | struct ofmirror *m = ofproto->mirrors[i]; | |
2311 | if (m) { | |
2312 | if (m->out == bundle) { | |
2313 | mirror_destroy(m); | |
2314 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2315 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2cc3c58e | 2316 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2317 | } |
2318 | } | |
2319 | } | |
2320 | ||
2321 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2322 | bundle_del_port(port); | |
2323 | } | |
2324 | ||
b44a10b7 | 2325 | bundle_flush_macs(bundle, true); |
abe529af BP |
2326 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2327 | free(bundle->name); | |
2328 | free(bundle->trunks); | |
2329 | lacp_destroy(bundle->lacp); | |
2330 | bond_destroy(bundle->bond); | |
2331 | free(bundle); | |
2332 | } | |
2333 | ||
2334 | static int | |
2335 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2336 | const struct ofproto_bundle_settings *s) | |
2337 | { | |
2338 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2339 | bool need_flush = false; | |
abe529af BP |
2340 | struct ofport_dpif *port; |
2341 | struct ofbundle *bundle; | |
ecac4ebf BP |
2342 | unsigned long *trunks; |
2343 | int vlan; | |
abe529af BP |
2344 | size_t i; |
2345 | bool ok; | |
2346 | ||
2347 | if (!s) { | |
2348 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2349 | return 0; | |
2350 | } | |
2351 | ||
cb22974d BP |
2352 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2353 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2354 | |
2355 | bundle = bundle_lookup(ofproto, aux); | |
2356 | if (!bundle) { | |
2357 | bundle = xmalloc(sizeof *bundle); | |
2358 | ||
2359 | bundle->ofproto = ofproto; | |
2360 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2361 | hash_pointer(aux, 0)); | |
2362 | bundle->aux = aux; | |
2363 | bundle->name = NULL; | |
2364 | ||
2365 | list_init(&bundle->ports); | |
ecac4ebf | 2366 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2367 | bundle->vlan = -1; |
2368 | bundle->trunks = NULL; | |
5e9ceccd | 2369 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2370 | bundle->lacp = NULL; |
2371 | bundle->bond = NULL; | |
2372 | ||
2373 | bundle->floodable = true; | |
2374 | ||
2375 | bundle->src_mirrors = 0; | |
2376 | bundle->dst_mirrors = 0; | |
2377 | bundle->mirror_out = 0; | |
2378 | } | |
2379 | ||
2380 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2381 | free(bundle->name); | |
2382 | bundle->name = xstrdup(s->name); | |
2383 | } | |
2384 | ||
2385 | /* LACP. */ | |
2386 | if (s->lacp) { | |
2387 | if (!bundle->lacp) { | |
2cc3c58e | 2388 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2389 | bundle->lacp = lacp_create(); |
2390 | } | |
2391 | lacp_configure(bundle->lacp, s->lacp); | |
2392 | } else { | |
2393 | lacp_destroy(bundle->lacp); | |
2394 | bundle->lacp = NULL; | |
2395 | } | |
2396 | ||
2397 | /* Update set of ports. */ | |
2398 | ok = true; | |
2399 | for (i = 0; i < s->n_slaves; i++) { | |
2400 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2401 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2402 | ok = false; |
2403 | } | |
2404 | } | |
2405 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2406 | struct ofport_dpif *next_port; | |
2407 | ||
2408 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2409 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2410 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2411 | goto found; |
2412 | } | |
2413 | } | |
2414 | ||
2415 | bundle_del_port(port); | |
2416 | found: ; | |
2417 | } | |
2418 | } | |
cb22974d | 2419 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2420 | |
2421 | if (list_is_empty(&bundle->ports)) { | |
2422 | bundle_destroy(bundle); | |
2423 | return EINVAL; | |
2424 | } | |
2425 | ||
ecac4ebf | 2426 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2427 | if (s->vlan_mode != bundle->vlan_mode |
2428 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2429 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2430 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2431 | need_flush = true; |
2432 | } | |
2433 | ||
abe529af | 2434 | /* Set VLAN tag. */ |
ecac4ebf BP |
2435 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2436 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2437 | : 0); | |
2438 | if (vlan != bundle->vlan) { | |
2439 | bundle->vlan = vlan; | |
abe529af BP |
2440 | need_flush = true; |
2441 | } | |
2442 | ||
2443 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2444 | switch (s->vlan_mode) { |
2445 | case PORT_VLAN_ACCESS: | |
2446 | trunks = NULL; | |
2447 | break; | |
2448 | ||
2449 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2450 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2451 | break; |
2452 | ||
2453 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2454 | case PORT_VLAN_NATIVE_TAGGED: | |
2455 | if (vlan != 0 && (!s->trunks | |
2456 | || !bitmap_is_set(s->trunks, vlan) | |
2457 | || bitmap_is_set(s->trunks, 0))) { | |
2458 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2459 | if (s->trunks) { | |
2460 | trunks = bitmap_clone(s->trunks, 4096); | |
2461 | } else { | |
2462 | trunks = bitmap_allocate1(4096); | |
2463 | } | |
2464 | bitmap_set1(trunks, vlan); | |
2465 | bitmap_set0(trunks, 0); | |
2466 | } else { | |
ebc56baa | 2467 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2468 | } |
2469 | break; | |
2470 | ||
2471 | default: | |
2472 | NOT_REACHED(); | |
2473 | } | |
abe529af BP |
2474 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2475 | free(bundle->trunks); | |
ecac4ebf BP |
2476 | if (trunks == s->trunks) { |
2477 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2478 | } else { | |
2479 | bundle->trunks = trunks; | |
2480 | trunks = NULL; | |
2481 | } | |
abe529af BP |
2482 | need_flush = true; |
2483 | } | |
ecac4ebf BP |
2484 | if (trunks != s->trunks) { |
2485 | free(trunks); | |
2486 | } | |
abe529af BP |
2487 | |
2488 | /* Bonding. */ | |
2489 | if (!list_is_short(&bundle->ports)) { | |
2490 | bundle->ofproto->has_bonded_bundles = true; | |
2491 | if (bundle->bond) { | |
2492 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2493 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2494 | } |
2495 | } else { | |
2496 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2497 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2498 | } |
2499 | ||
2500 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2501 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2502 | } |
2503 | } else { | |
2504 | bond_destroy(bundle->bond); | |
2505 | bundle->bond = NULL; | |
2506 | } | |
2507 | ||
2508 | /* If we changed something that would affect MAC learning, un-learn | |
2509 | * everything on this port and force flow revalidation. */ | |
2510 | if (need_flush) { | |
b44a10b7 | 2511 | bundle_flush_macs(bundle, false); |
abe529af BP |
2512 | } |
2513 | ||
2514 | return 0; | |
2515 | } | |
2516 | ||
2517 | static void | |
2518 | bundle_remove(struct ofport *port_) | |
2519 | { | |
2520 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2521 | struct ofbundle *bundle = port->bundle; | |
2522 | ||
2523 | if (bundle) { | |
2524 | bundle_del_port(port); | |
2525 | if (list_is_empty(&bundle->ports)) { | |
2526 | bundle_destroy(bundle); | |
2527 | } else if (list_is_short(&bundle->ports)) { | |
2528 | bond_destroy(bundle->bond); | |
2529 | bundle->bond = NULL; | |
2530 | } | |
2531 | } | |
2532 | } | |
2533 | ||
2534 | static void | |
5f877369 | 2535 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2536 | { |
2537 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2538 | struct ofport_dpif *port = port_; | |
2539 | uint8_t ea[ETH_ADDR_LEN]; | |
2540 | int error; | |
2541 | ||
2542 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2543 | if (!error) { | |
abe529af | 2544 | struct ofpbuf packet; |
5f877369 | 2545 | void *packet_pdu; |
abe529af BP |
2546 | |
2547 | ofpbuf_init(&packet, 0); | |
2548 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2549 | pdu_size); |
2550 | memcpy(packet_pdu, pdu, pdu_size); | |
2551 | ||
97d6520b | 2552 | send_packet(port, &packet); |
abe529af BP |
2553 | ofpbuf_uninit(&packet); |
2554 | } else { | |
2555 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2556 | "%s (%s)", port->bundle->name, | |
2557 | netdev_get_name(port->up.netdev), strerror(error)); | |
2558 | } | |
2559 | } | |
2560 | ||
2561 | static void | |
2562 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2563 | { | |
2564 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2565 | int error, n_packets, n_errors; | |
2566 | struct mac_entry *e; | |
2567 | ||
2568 | error = n_packets = n_errors = 0; | |
2569 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2570 | if (e->port.p != bundle) { | |
ea131871 JG |
2571 | struct ofpbuf *learning_packet; |
2572 | struct ofport_dpif *port; | |
4dd1e3ca | 2573 | void *port_void; |
ea131871 JG |
2574 | int ret; |
2575 | ||
4dd1e3ca BP |
2576 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2577 | * struct ofport_dpif more effective. */ | |
2578 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2579 | e->mac, e->vlan, | |
2580 | &port_void); | |
2581 | port = port_void; | |
97d6520b | 2582 | ret = send_packet(port, learning_packet); |
ea131871 | 2583 | ofpbuf_delete(learning_packet); |
abe529af BP |
2584 | if (ret) { |
2585 | error = ret; | |
2586 | n_errors++; | |
2587 | } | |
2588 | n_packets++; | |
2589 | } | |
2590 | } | |
2591 | ||
2592 | if (n_errors) { | |
2593 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2594 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2595 | "packets, last error was: %s", | |
2596 | bundle->name, n_errors, n_packets, strerror(error)); | |
2597 | } else { | |
2598 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2599 | bundle->name, n_packets); | |
2600 | } | |
2601 | } | |
2602 | ||
2603 | static void | |
2604 | bundle_run(struct ofbundle *bundle) | |
2605 | { | |
2606 | if (bundle->lacp) { | |
2607 | lacp_run(bundle->lacp, send_pdu_cb); | |
2608 | } | |
2609 | if (bundle->bond) { | |
2610 | struct ofport_dpif *port; | |
2611 | ||
2612 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2613 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2614 | } |
2615 | ||
2cc3c58e | 2616 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, |
bdebeece | 2617 | lacp_status(bundle->lacp)); |
abe529af BP |
2618 | if (bond_should_send_learning_packets(bundle->bond)) { |
2619 | bundle_send_learning_packets(bundle); | |
2620 | } | |
2621 | } | |
2622 | } | |
2623 | ||
2624 | static void | |
2625 | bundle_wait(struct ofbundle *bundle) | |
2626 | { | |
2627 | if (bundle->lacp) { | |
2628 | lacp_wait(bundle->lacp); | |
2629 | } | |
2630 | if (bundle->bond) { | |
2631 | bond_wait(bundle->bond); | |
2632 | } | |
2633 | } | |
2634 | \f | |
2635 | /* Mirrors. */ | |
2636 | ||
2637 | static int | |
2638 | mirror_scan(struct ofproto_dpif *ofproto) | |
2639 | { | |
2640 | int idx; | |
2641 | ||
2642 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2643 | if (!ofproto->mirrors[idx]) { | |
2644 | return idx; | |
2645 | } | |
2646 | } | |
2647 | return -1; | |
2648 | } | |
2649 | ||
2650 | static struct ofmirror * | |
2651 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2652 | { | |
2653 | int i; | |
2654 | ||
2655 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2656 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2657 | if (mirror && mirror->aux == aux) { | |
2658 | return mirror; | |
2659 | } | |
2660 | } | |
2661 | ||
2662 | return NULL; | |
2663 | } | |
2664 | ||
9ba15e2a BP |
2665 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2666 | static void | |
2667 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2668 | { | |
2669 | int i; | |
2670 | ||
2671 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2672 | struct ofmirror *m = ofproto->mirrors[i]; | |
2673 | ||
2674 | if (m) { | |
2675 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2676 | } | |
2677 | } | |
2678 | ||
2679 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2680 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2681 | int j; | |
2682 | ||
2683 | if (!m1) { | |
2684 | continue; | |
2685 | } | |
2686 | ||
2687 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2688 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2689 | ||
edb0540b | 2690 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2691 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2692 | m2->dup_mirrors |= m1->dup_mirrors; | |
2693 | } | |
2694 | } | |
2695 | } | |
2696 | } | |
2697 | ||
abe529af BP |
2698 | static int |
2699 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2700 | const struct ofproto_mirror_settings *s) | |
2701 | { | |
2702 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2703 | mirror_mask_t mirror_bit; | |
2704 | struct ofbundle *bundle; | |
2705 | struct ofmirror *mirror; | |
2706 | struct ofbundle *out; | |
2707 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2708 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2709 | int out_vlan; | |
2710 | ||
2711 | mirror = mirror_lookup(ofproto, aux); | |
2712 | if (!s) { | |
2713 | mirror_destroy(mirror); | |
2714 | return 0; | |
2715 | } | |
2716 | if (!mirror) { | |
2717 | int idx; | |
2718 | ||
2719 | idx = mirror_scan(ofproto); | |
2720 | if (idx < 0) { | |
2721 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2722 | "cannot create %s", | |
2723 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2724 | return EFBIG; | |
2725 | } | |
2726 | ||
2727 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2728 | mirror->ofproto = ofproto; | |
2729 | mirror->idx = idx; | |
8b28d864 | 2730 | mirror->aux = aux; |
abe529af BP |
2731 | mirror->out_vlan = -1; |
2732 | mirror->name = NULL; | |
2733 | } | |
2734 | ||
2735 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2736 | free(mirror->name); | |
2737 | mirror->name = xstrdup(s->name); | |
2738 | } | |
2739 | ||
2740 | /* Get the new configuration. */ | |
2741 | if (s->out_bundle) { | |
2742 | out = bundle_lookup(ofproto, s->out_bundle); | |
2743 | if (!out) { | |
2744 | mirror_destroy(mirror); | |
2745 | return EINVAL; | |
2746 | } | |
2747 | out_vlan = -1; | |
2748 | } else { | |
2749 | out = NULL; | |
2750 | out_vlan = s->out_vlan; | |
2751 | } | |
2752 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2753 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2754 | ||
2755 | /* If the configuration has not changed, do nothing. */ | |
2756 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2757 | && hmapx_equals(&dsts, &mirror->dsts) | |
2758 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2759 | && mirror->out == out | |
2760 | && mirror->out_vlan == out_vlan) | |
2761 | { | |
2762 | hmapx_destroy(&srcs); | |
2763 | hmapx_destroy(&dsts); | |
2764 | return 0; | |
2765 | } | |
2766 | ||
2767 | hmapx_swap(&srcs, &mirror->srcs); | |
2768 | hmapx_destroy(&srcs); | |
2769 | ||
2770 | hmapx_swap(&dsts, &mirror->dsts); | |
2771 | hmapx_destroy(&dsts); | |
2772 | ||
2773 | free(mirror->vlans); | |
2774 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2775 | ||
2776 | mirror->out = out; | |
2777 | mirror->out_vlan = out_vlan; | |
2778 | ||
2779 | /* Update bundles. */ | |
2780 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2781 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2782 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2783 | bundle->src_mirrors |= mirror_bit; | |
2784 | } else { | |
2785 | bundle->src_mirrors &= ~mirror_bit; | |
2786 | } | |
2787 | ||
2788 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2789 | bundle->dst_mirrors |= mirror_bit; | |
2790 | } else { | |
2791 | bundle->dst_mirrors &= ~mirror_bit; | |
2792 | } | |
2793 | ||
2794 | if (mirror->out == bundle) { | |
2795 | bundle->mirror_out |= mirror_bit; | |
2796 | } else { | |
2797 | bundle->mirror_out &= ~mirror_bit; | |
2798 | } | |
2799 | } | |
2800 | ||
2cc3c58e | 2801 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 2802 | ofproto->has_mirrors = true; |
2cc3c58e EJ |
2803 | mac_learning_flush(ofproto->ml, |
2804 | &ofproto->backer->revalidate_set); | |
9ba15e2a | 2805 | mirror_update_dups(ofproto); |
abe529af BP |
2806 | |
2807 | return 0; | |
2808 | } | |
2809 | ||
2810 | static void | |
2811 | mirror_destroy(struct ofmirror *mirror) | |
2812 | { | |
2813 | struct ofproto_dpif *ofproto; | |
2814 | mirror_mask_t mirror_bit; | |
2815 | struct ofbundle *bundle; | |
ccb7c863 | 2816 | int i; |
abe529af BP |
2817 | |
2818 | if (!mirror) { | |
2819 | return; | |
2820 | } | |
2821 | ||
2822 | ofproto = mirror->ofproto; | |
2cc3c58e EJ |
2823 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
2824 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
abe529af BP |
2825 | |
2826 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2827 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2828 | bundle->src_mirrors &= ~mirror_bit; | |
2829 | bundle->dst_mirrors &= ~mirror_bit; | |
2830 | bundle->mirror_out &= ~mirror_bit; | |
2831 | } | |
2832 | ||
2833 | hmapx_destroy(&mirror->srcs); | |
2834 | hmapx_destroy(&mirror->dsts); | |
2835 | free(mirror->vlans); | |
2836 | ||
2837 | ofproto->mirrors[mirror->idx] = NULL; | |
2838 | free(mirror->name); | |
2839 | free(mirror); | |
9ba15e2a BP |
2840 | |
2841 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
2842 | |
2843 | ofproto->has_mirrors = false; | |
2844 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2845 | if (ofproto->mirrors[i]) { | |
2846 | ofproto->has_mirrors = true; | |
2847 | break; | |
2848 | } | |
2849 | } | |
abe529af BP |
2850 | } |
2851 | ||
9d24de3b JP |
2852 | static int |
2853 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2854 | uint64_t *packets, uint64_t *bytes) | |
2855 | { | |
2856 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2857 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2858 | ||
2859 | if (!mirror) { | |
2860 | *packets = *bytes = UINT64_MAX; | |
2861 | return 0; | |
2862 | } | |
2863 | ||
2864 | *packets = mirror->packet_count; | |
2865 | *bytes = mirror->byte_count; | |
2866 | ||
2867 | return 0; | |
2868 | } | |
2869 | ||
abe529af BP |
2870 | static int |
2871 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2872 | { | |
2873 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2874 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2cc3c58e | 2875 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af BP |
2876 | } |
2877 | return 0; | |
2878 | } | |
2879 | ||
2880 | static bool | |
b4affc74 | 2881 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2882 | { |
2883 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2884 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2885 | return bundle && bundle->mirror_out != 0; | |
2886 | } | |
8402c74b SS |
2887 | |
2888 | static void | |
b53055f4 | 2889 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2890 | { |
2891 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2892 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2893 | } |
e764773c BP |
2894 | |
2895 | static void | |
c4069512 BP |
2896 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2897 | size_t max_entries) | |
e764773c BP |
2898 | { |
2899 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2900 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
c4069512 | 2901 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
e764773c | 2902 | } |
abe529af BP |
2903 | \f |
2904 | /* Ports. */ | |
2905 | ||
2906 | static struct ofport_dpif * | |
4acbc98d | 2907 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 2908 | { |
7df6a8bd BP |
2909 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2910 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2911 | } |
2912 | ||
2913 | static struct ofport_dpif * | |
4acbc98d | 2914 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af | 2915 | { |
7c33b188 JR |
2916 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2917 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2918 | } |
2919 | ||
2920 | static void | |
e1b1d06a JP |
2921 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2922 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2923 | struct dpif_port *dpif_port) |
2924 | { | |
2925 | ofproto_port->name = dpif_port->name; | |
2926 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2927 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2928 | } |
2929 | ||
0a740f48 EJ |
2930 | static struct ofport_dpif * |
2931 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
2932 | { | |
2933 | const struct ofproto_dpif *ofproto; | |
2934 | const char *peer; | |
2935 | ||
2936 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
2937 | if (!peer) { | |
2938 | return NULL; | |
2939 | } | |
2940 | ||
2941 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2942 | struct ofport *ofport; | |
2943 | ||
2944 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
2945 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
2946 | return ofport_dpif_cast(ofport); | |
2947 | } | |
2948 | } | |
2949 | return NULL; | |
2950 | } | |
2951 | ||
0aa66d6e EJ |
2952 | static void |
2953 | port_run_fast(struct ofport_dpif *ofport) | |
2954 | { | |
2955 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2956 | struct ofpbuf packet; | |
2957 | ||
2958 | ofpbuf_init(&packet, 0); | |
2959 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2960 | send_packet(ofport, &packet); | |
2961 | ofpbuf_uninit(&packet); | |
2962 | } | |
2963 | } | |
2964 | ||
abe529af BP |
2965 | static void |
2966 | port_run(struct ofport_dpif *ofport) | |
2967 | { | |
3e5b3fdb EJ |
2968 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2969 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2970 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2971 | ||
3e5b3fdb EJ |
2972 | ofport->carrier_seq = carrier_seq; |
2973 | ||
0aa66d6e | 2974 | port_run_fast(ofport); |
b9ad7294 EJ |
2975 | |
2976 | if (ofport->tnl_port | |
2977 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
2978 | &ofport->tnl_port)) { | |
2979 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
2980 | } | |
2981 | ||
abe529af | 2982 | if (ofport->cfm) { |
4653c558 EJ |
2983 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2984 | ||
abe529af | 2985 | cfm_run(ofport->cfm); |
4653c558 EJ |
2986 | enable = enable && !cfm_get_fault(ofport->cfm); |
2987 | ||
2988 | if (cfm_opup >= 0) { | |
2989 | enable = enable && cfm_opup; | |
2990 | } | |
abe529af | 2991 | } |
015e08bc EJ |
2992 | |
2993 | if (ofport->bundle) { | |
2994 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2995 | if (carrier_changed) { |
2996 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2997 | } | |
015e08bc EJ |
2998 | } |
2999 | ||
daff3353 EJ |
3000 | if (ofport->may_enable != enable) { |
3001 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3002 | ||
3003 | if (ofproto->has_bundle_action) { | |
2cc3c58e | 3004 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3005 | } |
3006 | } | |
3007 | ||
015e08bc | 3008 | ofport->may_enable = enable; |
abe529af BP |
3009 | } |
3010 | ||
3011 | static void | |
3012 | port_wait(struct ofport_dpif *ofport) | |
3013 | { | |
3014 | if (ofport->cfm) { | |
3015 | cfm_wait(ofport->cfm); | |
3016 | } | |
3017 | } | |
3018 | ||
3019 | static int | |
3020 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3021 | struct ofproto_port *ofproto_port) | |
3022 | { | |
3023 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3024 | struct dpif_port dpif_port; | |
3025 | int error; | |
3026 | ||
0a740f48 EJ |
3027 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3028 | const char *type = netdev_get_type_from_name(devname); | |
3029 | ||
3030 | /* We may be called before ofproto->up.port_by_name is populated with | |
3031 | * the appropriate ofport. For this reason, we must get the name and | |
3032 | * type from the netdev layer directly. */ | |
3033 | if (type) { | |
3034 | const struct ofport *ofport; | |
3035 | ||
3036 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3037 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3038 | ofproto_port->name = xstrdup(devname); | |
3039 | ofproto_port->type = xstrdup(type); | |
3040 | return 0; | |
3041 | } | |
3042 | return ENODEV; | |
3043 | } | |
3044 | ||
acf60855 JP |
3045 | if (!sset_contains(&ofproto->ports, devname)) { |
3046 | return ENODEV; | |
3047 | } | |
3048 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3049 | devname, &dpif_port); | |
abe529af | 3050 | if (!error) { |
e1b1d06a | 3051 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3052 | } |
3053 | return error; | |
3054 | } | |
3055 | ||
3056 | static int | |
e1b1d06a | 3057 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3058 | { |
3059 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 EJ |
3060 | const char *dp_port_name = netdev_vport_get_dpif_port(netdev); |
3061 | const char *devname = netdev_get_name(netdev); | |
abe529af | 3062 | |
0a740f48 EJ |
3063 | if (netdev_vport_is_patch(netdev)) { |
3064 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3065 | return 0; | |
3066 | } | |
3067 | ||
b9ad7294 | 3068 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
7d82ab2e KM |
3069 | uint32_t port_no = UINT32_MAX; |
3070 | int error; | |
3071 | ||
3072 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3073 | if (error) { |
3074 | return error; | |
3075 | } | |
7d82ab2e KM |
3076 | if (netdev_get_tunnel_config(netdev)) { |
3077 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3078 | } | |
acf60855 | 3079 | } |
b9ad7294 EJ |
3080 | |
3081 | if (netdev_get_tunnel_config(netdev)) { | |
3082 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3083 | } else { |
3084 | sset_add(&ofproto->ports, devname); | |
3085 | } | |
3086 | return 0; | |
3087 | } | |
3088 | ||
abe529af BP |
3089 | static int |
3090 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3091 | { | |
3092 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3093 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3094 | int error = 0; |
abe529af | 3095 | |
b9ad7294 EJ |
3096 | if (!ofport) { |
3097 | return 0; | |
e1b1d06a | 3098 | } |
b9ad7294 EJ |
3099 | |
3100 | sset_find_and_delete(&ofproto->ghost_ports, | |
3101 | netdev_get_name(ofport->up.netdev)); | |
a614d823 KM |
3102 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3103 | if (!ofport->tnl_port) { | |
b9ad7294 EJ |
3104 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3105 | if (!error) { | |
abe529af BP |
3106 | /* The caller is going to close ofport->up.netdev. If this is a |
3107 | * bonded port, then the bond is using that netdev, so remove it | |
3108 | * from the bond. The client will need to reconfigure everything | |
3109 | * after deleting ports, so then the slave will get re-added. */ | |
3110 | bundle_remove(&ofport->up); | |
3111 | } | |
3112 | } | |
3113 | return error; | |
3114 | } | |
3115 | ||
6527c598 PS |
3116 | static int |
3117 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3118 | { | |
3119 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3120 | int error; | |
3121 | ||
3122 | error = netdev_get_stats(ofport->up.netdev, stats); | |
3123 | ||
ee382d89 | 3124 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3125 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3126 | ||
3127 | /* ofproto->stats.tx_packets represents packets that we created | |
3128 | * internally and sent to some port (e.g. packets sent with | |
3129 | * send_packet()). Account for them as if they had come from | |
3130 | * OFPP_LOCAL and got forwarded. */ | |
3131 | ||
3132 | if (stats->rx_packets != UINT64_MAX) { | |
3133 | stats->rx_packets += ofproto->stats.tx_packets; | |
3134 | } | |
3135 | ||
3136 | if (stats->rx_bytes != UINT64_MAX) { | |
3137 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3138 | } | |
3139 | ||
3140 | /* ofproto->stats.rx_packets represents packets that were received on | |
3141 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3142 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3143 | |
3144 | if (stats->tx_packets != UINT64_MAX) { | |
3145 | stats->tx_packets += ofproto->stats.rx_packets; | |
3146 | } | |
3147 | ||
3148 | if (stats->tx_bytes != UINT64_MAX) { | |
3149 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | return error; | |
3154 | } | |
3155 | ||
3156 | /* Account packets for LOCAL port. */ | |
3157 | static void | |
3158 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
3159 | size_t tx_size, size_t rx_size) | |
3160 | { | |
3161 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3162 | ||
3163 | if (rx_size) { | |
3164 | ofproto->stats.rx_packets++; | |
3165 | ofproto->stats.rx_bytes += rx_size; | |
3166 | } | |
3167 | if (tx_size) { | |
3168 | ofproto->stats.tx_packets++; | |
3169 | ofproto->stats.tx_bytes += tx_size; | |
3170 | } | |
3171 | } | |
3172 | ||
abe529af | 3173 | struct port_dump_state { |
acf60855 JP |
3174 | uint32_t bucket; |
3175 | uint32_t offset; | |
0a740f48 | 3176 | bool ghost; |
da78d43d BP |
3177 | |
3178 | struct ofproto_port port; | |
3179 | bool has_port; | |
abe529af BP |
3180 | }; |
3181 | ||
3182 | static int | |
acf60855 | 3183 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3184 | { |
0a740f48 | 3185 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3186 | return 0; |
3187 | } | |
3188 | ||
3189 | static int | |
b9ad7294 | 3190 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3191 | struct ofproto_port *port) |
3192 | { | |
e1b1d06a | 3193 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3194 | struct port_dump_state *state = state_; |
0a740f48 | 3195 | const struct sset *sset; |
acf60855 | 3196 | struct sset_node *node; |
abe529af | 3197 | |
da78d43d BP |
3198 | if (state->has_port) { |
3199 | ofproto_port_destroy(&state->port); | |
3200 | state->has_port = false; | |
3201 | } | |
0a740f48 EJ |
3202 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3203 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3204 | int error; |
3205 | ||
da78d43d BP |
3206 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3207 | if (!error) { | |
3208 | *port = state->port; | |
3209 | state->has_port = true; | |
3210 | return 0; | |
3211 | } else if (error != ENODEV) { | |
acf60855 JP |
3212 | return error; |
3213 | } | |
abe529af | 3214 | } |
acf60855 | 3215 | |
0a740f48 EJ |
3216 | if (!state->ghost) { |
3217 | state->ghost = true; | |
3218 | state->bucket = 0; | |
3219 | state->offset = 0; | |
3220 | return port_dump_next(ofproto_, state_, port); | |
3221 | } | |
3222 | ||
acf60855 | 3223 | return EOF; |
abe529af BP |
3224 | } |
3225 | ||
3226 | static int | |
3227 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3228 | { | |
3229 | struct port_dump_state *state = state_; | |
3230 | ||
da78d43d BP |
3231 | if (state->has_port) { |
3232 | ofproto_port_destroy(&state->port); | |
3233 | } | |
abe529af BP |
3234 | free(state); |
3235 | return 0; | |
3236 | } | |
3237 | ||
3238 | static int | |
3239 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3240 | { | |
3241 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3242 | |
3243 | if (ofproto->port_poll_errno) { | |
3244 | int error = ofproto->port_poll_errno; | |
3245 | ofproto->port_poll_errno = 0; | |
3246 | return error; | |
3247 | } | |
3248 | ||
3249 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3250 | return EAGAIN; | |
3251 | } | |
3252 | ||
3253 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3254 | return 0; | |
abe529af BP |
3255 | } |
3256 | ||
3257 | static void | |
3258 | port_poll_wait(const struct ofproto *ofproto_) | |
3259 | { | |
3260 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3261 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3262 | } |
3263 | ||
3264 | static int | |
3265 | port_is_lacp_current(const struct ofport *ofport_) | |
3266 | { | |
3267 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3268 | return (ofport->bundle && ofport->bundle->lacp | |
3269 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3270 | : -1); | |
3271 | } | |
3272 | \f | |
3273 | /* Upcall handling. */ | |
3274 | ||
501f8d1f BP |
3275 | /* Flow miss batching. |
3276 | * | |
3277 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3278 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3279 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3280 | * more packets, plus possibly installing the flow in the dpif. | |
3281 | * | |
3282 | * So far we only batch the operations that affect flow setup time the most. | |
3283 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3284 | struct flow_miss { | |
3285 | struct hmap_node hmap_node; | |
acf60855 | 3286 | struct ofproto_dpif *ofproto; |
501f8d1f | 3287 | struct flow flow; |
b0f7b9b5 | 3288 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3289 | const struct nlattr *key; |
3290 | size_t key_len; | |
e84173dc | 3291 | ovs_be16 initial_tci; |
501f8d1f | 3292 | struct list packets; |
6a7e895f | 3293 | enum dpif_upcall_type upcall_type; |
a088a1ff | 3294 | uint32_t odp_in_port; |
501f8d1f BP |
3295 | }; |
3296 | ||
3297 | struct flow_miss_op { | |
c2b565b5 | 3298 | struct dpif_op dpif_op; |
5fe20d5d BP |
3299 | struct subfacet *subfacet; /* Subfacet */ |
3300 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
3301 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
3302 | }; |
3303 | ||
62cd7072 BP |
3304 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3305 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3306 | * configurations. */ |
62cd7072 | 3307 | static void |
a39edbd4 | 3308 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3309 | const struct flow *flow) |
62cd7072 BP |
3310 | { |
3311 | struct ofputil_packet_in pin; | |
3312 | ||
3e3252fa EJ |
3313 | pin.packet = packet->data; |
3314 | pin.packet_len = packet->size; | |
62cd7072 | 3315 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3316 | pin.controller_id = 0; |
54834960 EJ |
3317 | |
3318 | pin.table_id = 0; | |
3319 | pin.cookie = 0; | |
3320 | ||
62cd7072 | 3321 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3322 | |
3323 | flow_get_metadata(flow, &pin.fmd); | |
3324 | ||
d8653c38 | 3325 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3326 | } |
3327 | ||
6a7e895f | 3328 | static enum slow_path_reason |
abe529af | 3329 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
ffaef958 | 3330 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) |
abe529af | 3331 | { |
b6e001b6 | 3332 | if (!ofport) { |
6a7e895f | 3333 | return 0; |
ffaef958 | 3334 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 3335 | if (packet) { |
abe529af BP |
3336 | cfm_process_heartbeat(ofport->cfm, packet); |
3337 | } | |
6a7e895f | 3338 | return SLOW_CFM; |
b6e001b6 EJ |
3339 | } else if (ofport->bundle && ofport->bundle->lacp |
3340 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3341 | if (packet) { | |
3342 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 3343 | } |
6a7e895f | 3344 | return SLOW_LACP; |
21f7563c JP |
3345 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
3346 | if (packet) { | |
3347 | stp_process_packet(ofport, packet); | |
3348 | } | |
6a7e895f | 3349 | return SLOW_STP; |
ffaef958 BP |
3350 | } else { |
3351 | return 0; | |
abe529af | 3352 | } |
abe529af BP |
3353 | } |
3354 | ||
501f8d1f | 3355 | static struct flow_miss * |
ddbc5954 JP |
3356 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3357 | const struct flow *flow, uint32_t hash) | |
abe529af | 3358 | { |
501f8d1f | 3359 | struct flow_miss *miss; |
abe529af | 3360 | |
501f8d1f | 3361 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3362 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3363 | return miss; |
3364 | } | |
3365 | } | |
abe529af | 3366 | |
b23cdad9 | 3367 | return NULL; |
501f8d1f | 3368 | } |
abe529af | 3369 | |
9d6ac44e BP |
3370 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3371 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3372 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3373 | * returned op->subfacet, and if anything needs to be freed after processing | |
3374 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3375 | static void |
9d6ac44e BP |
3376 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3377 | struct flow_miss_op *op) | |
501f8d1f | 3378 | { |
9d6ac44e BP |
3379 | if (miss->flow.vlan_tci != miss->initial_tci) { |
3380 | /* This packet was received on a VLAN splinter port. We | |
3381 | * added a VLAN to the packet to make the packet resemble | |
3382 | * the flow, but the actions were composed assuming that | |
3383 | * the packet contained no VLAN. So, we must remove the | |
3384 | * VLAN header from the packet before trying to execute the | |
3385 | * actions. */ | |
3386 | eth_pop_vlan(packet); | |
3387 | } | |
3388 | ||
3389 | op->subfacet = NULL; | |
3390 | op->garbage = NULL; | |
3391 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
3392 | op->dpif_op.u.execute.key = miss->key; | |
3393 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3394 | op->dpif_op.u.execute.packet = packet; | |
3395 | } | |
3396 | ||
3397 | /* Helper for handle_flow_miss_without_facet() and | |
3398 | * handle_flow_miss_with_facet(). */ | |
3399 | static void | |
3400 | handle_flow_miss_common(struct rule_dpif *rule, | |
3401 | struct ofpbuf *packet, const struct flow *flow) | |
3402 | { | |
3403 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3404 | ||
3405 | ofproto->n_matches++; | |
3406 | ||
3407 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
3408 | /* | |
3409 | * Extra-special case for fail-open mode. | |
3410 | * | |
3411 | * We are in fail-open mode and the packet matched the fail-open | |
3412 | * rule, but we are connected to a controller too. We should send | |
3413 | * the packet up to the controller in the hope that it will try to | |
3414 | * set up a flow and thereby allow us to exit fail-open. | |
3415 | * | |
3416 | * See the top-level comment in fail-open.c for more information. | |
3417 | */ | |
3418 | send_packet_in_miss(ofproto, packet, flow); | |
3419 | } | |
3420 | } | |
3421 | ||
3422 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3423 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3424 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3425 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3426 | * the benefits, so when the datapath holds a large number of flows we impose | |
3427 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3428 | static bool | |
3429 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
3430 | struct flow_miss *miss, uint32_t hash) | |
3431 | { | |
3432 | if (!ofproto->governor) { | |
3433 | size_t n_subfacets; | |
3434 | ||
3435 | n_subfacets = hmap_count(&ofproto->subfacets); | |
3436 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
3437 | return true; | |
3438 | } | |
3439 | ||
3440 | ofproto->governor = governor_create(ofproto->up.name); | |
3441 | } | |
3442 | ||
3443 | return governor_should_install_flow(ofproto->governor, hash, | |
3444 | list_size(&miss->packets)); | |
3445 | } | |
3446 | ||
3447 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3448 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3449 | * increment '*n_ops'. */ | |
3450 | static void | |
3451 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
3452 | struct rule_dpif *rule, | |
3453 | struct flow_miss_op *ops, size_t *n_ops) | |
3454 | { | |
3455 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
a7752d4a | 3456 | long long int now = time_msec(); |
9d6ac44e | 3457 | struct action_xlate_ctx ctx; |
530a1d91 | 3458 | struct ofpbuf *packet; |
2b459b83 | 3459 | |
9d6ac44e BP |
3460 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
3461 | struct flow_miss_op *op = &ops[*n_ops]; | |
3462 | struct dpif_flow_stats stats; | |
3463 | struct ofpbuf odp_actions; | |
abe529af | 3464 | |
9d6ac44e | 3465 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3466 | |
9d6ac44e | 3467 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 3468 | |
a7752d4a | 3469 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 3470 | rule_credit_stats(rule, &stats); |
abe529af | 3471 | |
9d6ac44e BP |
3472 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, |
3473 | rule, 0, packet); | |
3474 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 3475 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
9d6ac44e | 3476 | &odp_actions); |
abe529af | 3477 | |
9d6ac44e BP |
3478 | if (odp_actions.size) { |
3479 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
3480 | ||
3481 | init_flow_miss_execute_op(miss, packet, op); | |
3482 | execute->actions = odp_actions.data; | |
3483 | execute->actions_len = odp_actions.size; | |
3484 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3485 | ||
3486 | (*n_ops)++; | |
3487 | } else { | |
3488 | ofpbuf_uninit(&odp_actions); | |
3489 | } | |
abe529af | 3490 | } |
9d6ac44e BP |
3491 | } |
3492 | ||
3493 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3494 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3495 | * | |
3496 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3497 | * This is really important only for new facets: if we just called time_msec() | |
3498 | * here, then the new subfacet or its packets could look (occasionally) as | |
3499 | * though it was used some time after the facet was used. That can make a | |
3500 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3501 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3502 | static void |
3503 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 3504 | long long int now, |
9d6ac44e BP |
3505 | struct flow_miss_op *ops, size_t *n_ops) |
3506 | { | |
6a7e895f BP |
3507 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3508 | enum subfacet_path want_path; | |
9d6ac44e BP |
3509 | struct subfacet *subfacet; |
3510 | struct ofpbuf *packet; | |
abe529af | 3511 | |
a088a1ff | 3512 | subfacet = subfacet_create(facet, miss, now); |
b0f7b9b5 | 3513 | |
530a1d91 | 3514 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3515 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 3516 | struct dpif_flow_stats stats; |
5fe20d5d | 3517 | struct ofpbuf odp_actions; |
67d91f78 | 3518 | |
9d6ac44e | 3519 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 3520 | |
5fe20d5d | 3521 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
6a7e895f | 3522 | if (!subfacet->actions || subfacet->slow) { |
5fe20d5d | 3523 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 3524 | } |
67d91f78 | 3525 | |
459b16a1 | 3526 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 3527 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 3528 | |
9d6ac44e BP |
3529 | if (subfacet->actions_len) { |
3530 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 3531 | |
9d6ac44e BP |
3532 | init_flow_miss_execute_op(miss, packet, op); |
3533 | op->subfacet = subfacet; | |
6a7e895f | 3534 | if (!subfacet->slow) { |
9d6ac44e BP |
3535 | execute->actions = subfacet->actions; |
3536 | execute->actions_len = subfacet->actions_len; | |
3537 | ofpbuf_uninit(&odp_actions); | |
3538 | } else { | |
3539 | execute->actions = odp_actions.data; | |
3540 | execute->actions_len = odp_actions.size; | |
3541 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3542 | } | |
999fba59 | 3543 | |
9d6ac44e | 3544 | (*n_ops)++; |
5fe20d5d | 3545 | } else { |
9d6ac44e | 3546 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 3547 | } |
501f8d1f BP |
3548 | } |
3549 | ||
6a7e895f BP |
3550 | want_path = subfacet_want_path(subfacet->slow); |
3551 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
501f8d1f | 3552 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3553 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3554 | |
b0f7b9b5 | 3555 | op->subfacet = subfacet; |
5fe20d5d | 3556 | op->garbage = NULL; |
c2b565b5 | 3557 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
3558 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
3559 | put->key = miss->key; | |
3560 | put->key_len = miss->key_len; | |
6a7e895f BP |
3561 | if (want_path == SF_FAST_PATH) { |
3562 | put->actions = subfacet->actions; | |
3563 | put->actions_len = subfacet->actions_len; | |
3564 | } else { | |
3565 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
3566 | op->stub, sizeof op->stub, | |
3567 | &put->actions, &put->actions_len); | |
3568 | } | |
501f8d1f BP |
3569 | put->stats = NULL; |
3570 | } | |
3571 | } | |
3572 | ||
acf60855 JP |
3573 | /* Handles flow miss 'miss'. May add any required datapath operations |
3574 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3575 | static void |
acf60855 JP |
3576 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3577 | size_t *n_ops) | |
9d6ac44e | 3578 | { |
acf60855 | 3579 | struct ofproto_dpif *ofproto = miss->ofproto; |
9d6ac44e | 3580 | struct facet *facet; |
459b16a1 | 3581 | long long int now; |
9d6ac44e BP |
3582 | uint32_t hash; |
3583 | ||
3584 | /* The caller must ensure that miss->hmap_node.hash contains | |
3585 | * flow_hash(miss->flow, 0). */ | |
3586 | hash = miss->hmap_node.hash; | |
3587 | ||
3588 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3589 | if (!facet) { | |
c57b2226 BP |
3590 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); |
3591 | ||
3592 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
9d6ac44e BP |
3593 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); |
3594 | return; | |
3595 | } | |
3596 | ||
3597 | facet = facet_create(rule, &miss->flow, hash); | |
459b16a1 BP |
3598 | now = facet->used; |
3599 | } else { | |
3600 | now = time_msec(); | |
9d6ac44e | 3601 | } |
459b16a1 | 3602 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3603 | } |
3604 | ||
8f73d537 EJ |
3605 | static struct drop_key * |
3606 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3607 | size_t key_len) | |
3608 | { | |
3609 | struct drop_key *drop_key; | |
3610 | ||
3611 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3612 | &backer->drop_keys) { | |
3613 | if (drop_key->key_len == key_len | |
3614 | && !memcmp(drop_key->key, key, key_len)) { | |
3615 | return drop_key; | |
3616 | } | |
3617 | } | |
3618 | return NULL; | |
3619 | } | |
3620 | ||
3621 | static void | |
3622 | drop_key_clear(struct dpif_backer *backer) | |
3623 | { | |
3624 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3625 | struct drop_key *drop_key, *next; | |
3626 | ||
3627 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3628 | int error; | |
3629 | ||
3630 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3631 | NULL); | |
3632 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3633 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3634 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3635 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3636 | ds_cstr(&ds)); | |
3637 | ds_destroy(&ds); | |
3638 | } | |
3639 | ||
3640 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3641 | free(drop_key->key); | |
3642 | free(drop_key); | |
3643 | } | |
3644 | } | |
3645 | ||
e09ee259 EJ |
3646 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3647 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3648 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3649 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3650 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3651 | * 'packet' ingressed. | |
e2a6ca36 | 3652 | * |
e09ee259 EJ |
3653 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3654 | * 'flow''s in_port to OFPP_NONE. | |
3655 | * | |
3656 | * This function does post-processing on data returned from | |
3657 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3658 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3659 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3660 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3661 | * a VLAN header onto 'packet' (if it is nonnull). | |
3662 | * | |
3663 | * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the | |
3664 | * packet was really received, that is, the actual VLAN TCI extracted by | |
3665 | * odp_flow_key_to_flow(). (This differs from the value returned in | |
3666 | * flow->vlan_tci only for packets received on VLAN splinters.) | |
3667 | * | |
b9ad7294 EJ |
3668 | * Similarly, this function also includes some logic to help with tunnels. It |
3669 | * may modify 'flow' as necessary to make the tunneling implementation | |
3670 | * transparent to the upcall processing logic. | |
3671 | * | |
e09ee259 EJ |
3672 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3673 | * or some other positive errno if there are other problems. */ | |
3674 | static int | |
3675 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3676 | const struct nlattr *key, size_t key_len, | |
3677 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3678 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
3679 | ovs_be16 *initial_tci) | |
e84173dc | 3680 | { |
e09ee259 EJ |
3681 | const struct ofport_dpif *port; |
3682 | enum odp_key_fitness fitness; | |
b9ad7294 | 3683 | int error = ENODEV; |
e09ee259 EJ |
3684 | |
3685 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3686 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3687 | error = EINVAL; |
3688 | goto exit; | |
3689 | } | |
3690 | ||
3691 | if (initial_tci) { | |
3692 | *initial_tci = flow->vlan_tci; | |
e84173dc | 3693 | } |
e84173dc | 3694 | |
e09ee259 EJ |
3695 | if (odp_in_port) { |
3696 | *odp_in_port = flow->in_port; | |
3697 | } | |
3698 | ||
b9ad7294 EJ |
3699 | if (tnl_port_should_receive(flow)) { |
3700 | const struct ofport *ofport = tnl_port_receive(flow); | |
3701 | if (!ofport) { | |
3702 | flow->in_port = OFPP_NONE; | |
3703 | goto exit; | |
3704 | } | |
3705 | port = ofport_dpif_cast(ofport); | |
e09ee259 | 3706 | |
b9ad7294 EJ |
3707 | /* We can't reproduce 'key' from 'flow'. */ |
3708 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
e09ee259 | 3709 | |
b9ad7294 EJ |
3710 | /* XXX: Since the tunnel module is not scoped per backer, it's |
3711 | * theoretically possible that we'll receive an ofport belonging to an | |
3712 | * entirely different datapath. In practice, this can't happen because | |
3713 | * no platforms has two separate datapaths which each support | |
3714 | * tunneling. */ | |
3715 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3716 | } else { | |
3717 | port = odp_port_to_ofport(backer, flow->in_port); | |
3718 | if (!port) { | |
3719 | flow->in_port = OFPP_NONE; | |
3720 | goto exit; | |
3721 | } | |
3722 | ||
3723 | flow->in_port = port->up.ofp_port; | |
3724 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3725 | if (packet) { | |
3726 | /* Make the packet resemble the flow, so that it gets sent to | |
3727 | * an OpenFlow controller properly, so that it looks correct | |
3728 | * for sFlow, and so that flow_extract() will get the correct | |
3729 | * vlan_tci if it is called on 'packet'. | |
3730 | * | |
3731 | * The allocated space inside 'packet' probably also contains | |
3732 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3733 | * a struct dpif_upcall (see the large comment on that | |
3734 | * structure definition), so pushing data on 'packet' is in | |
3735 | * general not a good idea since it could overwrite 'key' or | |
3736 | * free it as a side effect. However, it's OK in this special | |
3737 | * case because we know that 'packet' is inside a Netlink | |
3738 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3739 | * "struct nlattr", which is fine since we don't need that | |
3740 | * header anymore. */ | |
3741 | eth_push_vlan(packet, flow->vlan_tci); | |
3742 | } | |
3743 | /* We can't reproduce 'key' from 'flow'. */ | |
3744 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 BP |
3745 | } |
3746 | } | |
e09ee259 | 3747 | error = 0; |
52a90c29 | 3748 | |
b9ad7294 EJ |
3749 | if (ofproto) { |
3750 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3751 | } | |
3752 | ||
e09ee259 EJ |
3753 | exit: |
3754 | if (fitnessp) { | |
3755 | *fitnessp = fitness; | |
3756 | } | |
3757 | return error; | |
e84173dc BP |
3758 | } |
3759 | ||
501f8d1f | 3760 | static void |
acf60855 | 3761 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3762 | size_t n_upcalls) |
3763 | { | |
3764 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3765 | struct flow_miss *miss; |
3766 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3767 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3768 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3769 | struct hmap todo; |
b23cdad9 | 3770 | int n_misses; |
501f8d1f BP |
3771 | size_t n_ops; |
3772 | size_t i; | |
3773 | ||
3774 | if (!n_upcalls) { | |
3775 | return; | |
3776 | } | |
3777 | ||
3778 | /* Construct the to-do list. | |
3779 | * | |
3780 | * This just amounts to extracting the flow from each packet and sticking | |
3781 | * the packets that have the same flow in the same "flow_miss" structure so | |
3782 | * that we can process them together. */ | |
3783 | hmap_init(&todo); | |
b23cdad9 | 3784 | n_misses = 0; |
501f8d1f | 3785 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3786 | struct flow_miss *miss = &misses[n_misses]; |
3787 | struct flow_miss *existing_miss; | |
acf60855 | 3788 | struct ofproto_dpif *ofproto; |
a088a1ff | 3789 | uint32_t odp_in_port; |
1d446463 | 3790 | struct flow flow; |
b23cdad9 | 3791 | uint32_t hash; |
e09ee259 | 3792 | int error; |
501f8d1f | 3793 | |
e09ee259 EJ |
3794 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
3795 | upcall->key_len, &flow, &miss->key_fitness, | |
3796 | &ofproto, &odp_in_port, &miss->initial_tci); | |
3797 | if (error == ENODEV) { | |
8f73d537 EJ |
3798 | struct drop_key *drop_key; |
3799 | ||
acf60855 JP |
3800 | /* Received packet on port for which we couldn't associate |
3801 | * an ofproto. This can happen if a port is removed while | |
3802 | * traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
3803 | * in case it happens frequently. Install a drop flow so |
3804 | * that future packets of the flow are inexpensively dropped | |
3805 | * in the kernel. */ | |
acf60855 JP |
3806 | VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, |
3807 | flow.in_port); | |
8f73d537 EJ |
3808 | |
3809 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
3810 | if (!drop_key) { | |
3811 | drop_key = xmalloc(sizeof *drop_key); | |
3812 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
3813 | drop_key->key_len = upcall->key_len; | |
3814 | ||
3815 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3816 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3817 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3818 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
3819 | } | |
3820 | continue; | |
acf60855 | 3821 | } |
e09ee259 | 3822 | if (error) { |
b0f7b9b5 BP |
3823 | continue; |
3824 | } | |
72e8bf28 | 3825 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
1d446463 | 3826 | &flow.tunnel, flow.in_port, &miss->flow); |
501f8d1f | 3827 | |
501f8d1f | 3828 | /* Add other packets to a to-do list. */ |
b23cdad9 | 3829 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 3830 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
3831 | if (!existing_miss) { |
3832 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 3833 | miss->ofproto = ofproto; |
b23cdad9 BP |
3834 | miss->key = upcall->key; |
3835 | miss->key_len = upcall->key_len; | |
6a7e895f | 3836 | miss->upcall_type = upcall->type; |
a088a1ff | 3837 | miss->odp_in_port = odp_in_port; |
b23cdad9 BP |
3838 | list_init(&miss->packets); |
3839 | ||
3840 | n_misses++; | |
3841 | } else { | |
3842 | miss = existing_miss; | |
3843 | } | |
501f8d1f BP |
3844 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3845 | } | |
3846 | ||
3847 | /* Process each element in the to-do list, constructing the set of | |
3848 | * operations to batch. */ | |
3849 | n_ops = 0; | |
33bb0caa | 3850 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 3851 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3852 | } |
cb22974d | 3853 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3854 | |
3855 | /* Execute batch. */ | |
3856 | for (i = 0; i < n_ops; i++) { | |
3857 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3858 | } | |
acf60855 | 3859 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f BP |
3860 | |
3861 | /* Free memory and update facets. */ | |
3862 | for (i = 0; i < n_ops; i++) { | |
3863 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
501f8d1f BP |
3864 | |
3865 | switch (op->dpif_op.type) { | |
3866 | case DPIF_OP_EXECUTE: | |
501f8d1f | 3867 | break; |
abe529af | 3868 | |
501f8d1f | 3869 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 3870 | if (!op->dpif_op.error) { |
6a7e895f | 3871 | op->subfacet->path = subfacet_want_path(op->subfacet->slow); |
501f8d1f BP |
3872 | } |
3873 | break; | |
b99d3cee BP |
3874 | |
3875 | case DPIF_OP_FLOW_DEL: | |
3876 | NOT_REACHED(); | |
501f8d1f | 3877 | } |
5fe20d5d BP |
3878 | |
3879 | free(op->garbage); | |
501f8d1f | 3880 | } |
33bb0caa | 3881 | hmap_destroy(&todo); |
abe529af BP |
3882 | } |
3883 | ||
6a7e895f BP |
3884 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } |
3885 | classify_upcall(const struct dpif_upcall *upcall) | |
3886 | { | |
3887 | union user_action_cookie cookie; | |
3888 | ||
3889 | /* First look at the upcall type. */ | |
3890 | switch (upcall->type) { | |
3891 | case DPIF_UC_ACTION: | |
3892 | break; | |
3893 | ||
3894 | case DPIF_UC_MISS: | |
3895 | return MISS_UPCALL; | |
3896 | ||
3897 | case DPIF_N_UC_TYPES: | |
3898 | default: | |
3899 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3900 | return BAD_UPCALL; | |
3901 | } | |
3902 | ||
3903 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
3904 | if (!upcall->userdata) { |
3905 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
3906 | return BAD_UPCALL; | |
3907 | } | |
3908 | if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) { | |
3909 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", | |
3910 | nl_attr_get_size(upcall->userdata)); | |
3911 | return BAD_UPCALL; | |
3912 | } | |
3913 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie)); | |
6a7e895f BP |
3914 | switch (cookie.type) { |
3915 | case USER_ACTION_COOKIE_SFLOW: | |
3916 | return SFLOW_UPCALL; | |
3917 | ||
3918 | case USER_ACTION_COOKIE_SLOW_PATH: | |
3919 | return MISS_UPCALL; | |
3920 | ||
3921 | case USER_ACTION_COOKIE_UNSPEC: | |
3922 | default: | |
e995e3df BP |
3923 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, |
3924 | nl_attr_get_u64(upcall->userdata)); | |
6a7e895f BP |
3925 | return BAD_UPCALL; |
3926 | } | |
3927 | } | |
3928 | ||
abe529af | 3929 | static void |
acf60855 | 3930 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 3931 | const struct dpif_upcall *upcall) |
abe529af | 3932 | { |
acf60855 | 3933 | struct ofproto_dpif *ofproto; |
1673e0e4 | 3934 | union user_action_cookie cookie; |
e84173dc | 3935 | struct flow flow; |
e1b1d06a | 3936 | uint32_t odp_in_port; |
abe529af | 3937 | |
e09ee259 EJ |
3938 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
3939 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
3940 | || !ofproto->sflow) { | |
e84173dc BP |
3941 | return; |
3942 | } | |
3943 | ||
e995e3df | 3944 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie)); |
e1b1d06a JP |
3945 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
3946 | odp_in_port, &cookie); | |
6ff686f2 PS |
3947 | } |
3948 | ||
9b16c439 | 3949 | static int |
acf60855 | 3950 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 3951 | { |
9b16c439 | 3952 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3953 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3954 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3955 | int n_processed; | |
9b16c439 BP |
3956 | int n_misses; |
3957 | int i; | |
abe529af | 3958 | |
cb22974d | 3959 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3960 | |
9b16c439 | 3961 | n_misses = 0; |
90a7c55e | 3962 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3963 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3964 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3965 | int error; |
3966 | ||
90a7c55e BP |
3967 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3968 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 3969 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 3970 | if (error) { |
90a7c55e | 3971 | ofpbuf_uninit(buf); |
9b16c439 BP |
3972 | break; |
3973 | } | |
3974 | ||
6a7e895f BP |
3975 | switch (classify_upcall(upcall)) { |
3976 | case MISS_UPCALL: | |
9b16c439 BP |
3977 | /* Handle it later. */ |
3978 | n_misses++; | |
3979 | break; | |
3980 | ||
6a7e895f | 3981 | case SFLOW_UPCALL: |
acf60855 | 3982 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
3983 | ofpbuf_uninit(buf); |
3984 | break; | |
3985 | ||
3986 | case BAD_UPCALL: | |
3987 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3988 | break; |
3989 | } | |
abe529af | 3990 | } |
9b16c439 | 3991 | |
6a7e895f | 3992 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 3993 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
3994 | for (i = 0; i < n_misses; i++) { |
3995 | ofpbuf_uninit(&miss_bufs[i]); | |
3996 | } | |
9b16c439 | 3997 | |
90a7c55e | 3998 | return n_processed; |
abe529af BP |
3999 | } |
4000 | \f | |
4001 | /* Flow expiration. */ | |
4002 | ||
b0f7b9b5 | 4003 | static int subfacet_max_idle(const struct ofproto_dpif *); |
acf60855 | 4004 | static void update_stats(struct dpif_backer *); |
abe529af | 4005 | static void rule_expire(struct rule_dpif *); |
b0f7b9b5 | 4006 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
4007 | |
4008 | /* This function is called periodically by run(). Its job is to collect | |
4009 | * updates for the flows that have been installed into the datapath, most | |
4010 | * importantly when they last were used, and then use that information to | |
4011 | * expire flows that have not been used recently. | |
4012 | * | |
4013 | * Returns the number of milliseconds after which it should be called again. */ | |
4014 | static int | |
acf60855 | 4015 | expire(struct dpif_backer *backer) |
abe529af | 4016 | { |
acf60855 JP |
4017 | struct ofproto_dpif *ofproto; |
4018 | int max_idle = INT32_MAX; | |
abe529af | 4019 | |
8f73d537 EJ |
4020 | /* Periodically clear out the drop keys in an effort to keep them |
4021 | * relatively few. */ | |
4022 | drop_key_clear(backer); | |
4023 | ||
acf60855 JP |
4024 | /* Update stats for each flow in the backer. */ |
4025 | update_stats(backer); | |
abe529af | 4026 | |
acf60855 | 4027 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
e503cc19 | 4028 | struct rule *rule, *next_rule; |
acf60855 | 4029 | int dp_max_idle; |
abe529af | 4030 | |
acf60855 JP |
4031 | if (ofproto->backer != backer) { |
4032 | continue; | |
4033 | } | |
0697b5c3 | 4034 | |
acf60855 JP |
4035 | /* Expire subfacets that have been idle too long. */ |
4036 | dp_max_idle = subfacet_max_idle(ofproto); | |
4037 | expire_subfacets(ofproto, dp_max_idle); | |
4038 | ||
4039 | max_idle = MIN(max_idle, dp_max_idle); | |
4040 | ||
4041 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4042 | * has passed. */ | |
e503cc19 SH |
4043 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4044 | &ofproto->up.expirable) { | |
4045 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4046 | } |
abe529af | 4047 | |
acf60855 JP |
4048 | /* All outstanding data in existing flows has been accounted, so it's a |
4049 | * good time to do bond rebalancing. */ | |
4050 | if (ofproto->has_bonded_bundles) { | |
4051 | struct ofbundle *bundle; | |
abe529af | 4052 | |
acf60855 JP |
4053 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4054 | if (bundle->bond) { | |
2cc3c58e | 4055 | bond_rebalance(bundle->bond, &backer->revalidate_set); |
acf60855 | 4056 | } |
abe529af BP |
4057 | } |
4058 | } | |
4059 | } | |
4060 | ||
acf60855 | 4061 | return MIN(max_idle, 1000); |
abe529af BP |
4062 | } |
4063 | ||
a218c879 BP |
4064 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4065 | * as 'subfacet''s statistics. */ | |
4066 | static void | |
4067 | update_subfacet_stats(struct subfacet *subfacet, | |
4068 | const struct dpif_flow_stats *stats) | |
4069 | { | |
4070 | struct facet *facet = subfacet->facet; | |
4071 | ||
4072 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
4073 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
4074 | facet->packet_count += extra; | |
4075 | } else { | |
4076 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4077 | } | |
4078 | ||
4079 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
4080 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
4081 | } else { | |
4082 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4083 | } | |
4084 | ||
4085 | subfacet->dp_packet_count = stats->n_packets; | |
4086 | subfacet->dp_byte_count = stats->n_bytes; | |
4087 | ||
4088 | facet->tcp_flags |= stats->tcp_flags; | |
4089 | ||
4090 | subfacet_update_time(subfacet, stats->used); | |
4091 | if (facet->accounted_bytes < facet->byte_count) { | |
4092 | facet_learn(facet); | |
4093 | facet_account(facet); | |
4094 | facet->accounted_bytes = facet->byte_count; | |
4095 | } | |
4096 | facet_push_stats(facet); | |
4097 | } | |
4098 | ||
4099 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4100 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4101 | static void | |
acf60855 | 4102 | delete_unexpected_flow(struct ofproto_dpif *ofproto, |
a218c879 BP |
4103 | const struct nlattr *key, size_t key_len) |
4104 | { | |
4105 | if (!VLOG_DROP_WARN(&rl)) { | |
4106 | struct ds s; | |
4107 | ||
4108 | ds_init(&s); | |
4109 | odp_flow_key_format(key, key_len, &s); | |
acf60855 | 4110 | VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); |
a218c879 BP |
4111 | ds_destroy(&s); |
4112 | } | |
4113 | ||
4114 | COVERAGE_INC(facet_unexpected); | |
acf60855 | 4115 | dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); |
a218c879 BP |
4116 | } |
4117 | ||
abe529af BP |
4118 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4119 | * | |
4120 | * This function also pushes statistics updates to rules which each facet | |
4121 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4122 | * facet changes the rule it resubmits into at some time in between | |
4123 | * update_stats() runs, it is possible that statistics accrued to the | |
4124 | * old rule will be incorrectly attributed to the new rule. This could be | |
4125 | * avoided by calling update_stats() whenever rules are created or | |
4126 | * deleted. However, the performance impact of making so many calls to the | |
4127 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
4128 | */ | |
4129 | static void | |
acf60855 | 4130 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4131 | { |
4132 | const struct dpif_flow_stats *stats; | |
4133 | struct dpif_flow_dump dump; | |
4134 | const struct nlattr *key; | |
4135 | size_t key_len; | |
4136 | ||
acf60855 | 4137 | dpif_flow_dump_start(&dump, backer->dpif); |
abe529af | 4138 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
acf60855 | 4139 | struct flow flow; |
b0f7b9b5 | 4140 | struct subfacet *subfacet; |
acf60855 | 4141 | struct ofproto_dpif *ofproto; |
b9ad7294 | 4142 | struct ofport_dpif *ofport; |
acf60855 | 4143 | uint32_t key_hash; |
abe529af | 4144 | |
58c6adda EJ |
4145 | if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, |
4146 | NULL, NULL)) { | |
acf60855 JP |
4147 | continue; |
4148 | } | |
4149 | ||
b9ad7294 EJ |
4150 | ofport = get_ofp_port(ofproto, flow.in_port); |
4151 | if (ofport && ofport->tnl_port) { | |
4152 | netdev_vport_inc_rx(ofport->up.netdev, stats); | |
4153 | } | |
4154 | ||
acf60855 | 4155 | key_hash = odp_flow_key_hash(key, key_len); |
acf60855 | 4156 | subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow); |
6a7e895f BP |
4157 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4158 | case SF_FAST_PATH: | |
a218c879 | 4159 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4160 | break; |
4161 | ||
4162 | case SF_SLOW_PATH: | |
4163 | /* Stats are updated per-packet. */ | |
4164 | break; | |
4165 | ||
4166 | case SF_NOT_INSTALLED: | |
4167 | default: | |
acf60855 | 4168 | delete_unexpected_flow(ofproto, key, key_len); |
6a7e895f | 4169 | break; |
abe529af BP |
4170 | } |
4171 | } | |
4172 | dpif_flow_dump_done(&dump); | |
4173 | } | |
4174 | ||
4175 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4176 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4177 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4178 | * fold its statistic into its rule. */ | |
abe529af | 4179 | static int |
b0f7b9b5 | 4180 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
4181 | { |
4182 | /* | |
4183 | * Idle time histogram. | |
4184 | * | |
b0f7b9b5 BP |
4185 | * Most of the time a switch has a relatively small number of subfacets. |
4186 | * When this is the case we might as well keep statistics for all of them | |
4187 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4188 | * well. |
4189 | * | |
b0f7b9b5 | 4190 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4191 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4192 | * significant. However, with a large number of subfacets it is likely |
4193 | * that only a few of them are "heavy hitters" that consume a large amount | |
4194 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4195 | * the kernel and maintaining in userspaces; other subfacets we can | |
4196 | * discard. | |
abe529af BP |
4197 | * |
4198 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4199 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4200 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4201 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4202 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 4203 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4204 | * the most-recently-used bucket of subfacets is kept, so actually an |
4205 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4206 | * (though the next run will delete most of those unless they receive |
4207 | * additional data). | |
abe529af | 4208 | * |
b0f7b9b5 BP |
4209 | * This requires a second pass through the subfacets, in addition to the |
4210 | * pass made by update_stats(), because the former function never looks at | |
4211 | * uninstallable subfacets. | |
abe529af BP |
4212 | */ |
4213 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4214 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4215 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4216 | int total, subtotal, bucket; |
b0f7b9b5 | 4217 | struct subfacet *subfacet; |
abe529af BP |
4218 | long long int now; |
4219 | int i; | |
4220 | ||
b0f7b9b5 | 4221 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 4222 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
4223 | return N_BUCKETS * BUCKET_WIDTH; |
4224 | } | |
4225 | ||
4226 | /* Build histogram. */ | |
4227 | now = time_msec(); | |
b0f7b9b5 BP |
4228 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
4229 | long long int idle = now - subfacet->used; | |
abe529af BP |
4230 | int bucket = (idle <= 0 ? 0 |
4231 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4232 | : (unsigned int) idle / BUCKET_WIDTH); | |
4233 | buckets[bucket]++; | |
4234 | } | |
4235 | ||
4236 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4237 | subtotal = bucket = 0; |
4238 | do { | |
4239 | subtotal += buckets[bucket++]; | |
084f5290 SH |
4240 | } while (bucket < N_BUCKETS && |
4241 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
4242 | |
4243 | if (VLOG_IS_DBG_ENABLED()) { | |
4244 | struct ds s; | |
4245 | ||
4246 | ds_init(&s); | |
4247 | ds_put_cstr(&s, "keep"); | |
4248 | for (i = 0; i < N_BUCKETS; i++) { | |
4249 | if (i == bucket) { | |
4250 | ds_put_cstr(&s, ", drop"); | |
4251 | } | |
4252 | if (buckets[i]) { | |
4253 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4254 | } | |
4255 | } | |
4256 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
4257 | ds_destroy(&s); | |
4258 | } | |
4259 | ||
4260 | return bucket * BUCKET_WIDTH; | |
4261 | } | |
4262 | ||
abe529af | 4263 | static void |
b0f7b9b5 | 4264 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 4265 | { |
625b0720 BP |
4266 | /* Cutoff time for most flows. */ |
4267 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4268 | ||
4269 | /* We really want to keep flows for special protocols around, so use a more | |
4270 | * conservative cutoff. */ | |
4271 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4272 | |
b0f7b9b5 | 4273 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4274 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4275 | int n_batch; |
abe529af | 4276 | |
b99d3cee | 4277 | n_batch = 0; |
b0f7b9b5 BP |
4278 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
4279 | &ofproto->subfacets) { | |
625b0720 BP |
4280 | long long int cutoff; |
4281 | ||
4282 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) | |
4283 | ? special_cutoff | |
4284 | : normal_cutoff); | |
b0f7b9b5 | 4285 | if (subfacet->used < cutoff) { |
6a7e895f | 4286 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4287 | batch[n_batch++] = subfacet; |
1d85f9e5 JP |
4288 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
4289 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
b99d3cee BP |
4290 | n_batch = 0; |
4291 | } | |
4292 | } else { | |
4293 | subfacet_destroy(subfacet); | |
4294 | } | |
abe529af BP |
4295 | } |
4296 | } | |
b99d3cee BP |
4297 | |
4298 | if (n_batch > 0) { | |
1d85f9e5 | 4299 | subfacet_destroy_batch(ofproto, batch, n_batch); |
b99d3cee | 4300 | } |
abe529af BP |
4301 | } |
4302 | ||
4303 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4304 | * then delete it entirely. */ | |
4305 | static void | |
4306 | rule_expire(struct rule_dpif *rule) | |
4307 | { | |
abe529af BP |
4308 | struct facet *facet, *next_facet; |
4309 | long long int now; | |
4310 | uint8_t reason; | |
4311 | ||
e2a3d183 BP |
4312 | if (rule->up.pending) { |
4313 | /* We'll have to expire it later. */ | |
4314 | return; | |
4315 | } | |
4316 | ||
abe529af BP |
4317 | /* Has 'rule' expired? */ |
4318 | now = time_msec(); | |
4319 | if (rule->up.hard_timeout | |
308881af | 4320 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4321 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4322 | } else if (rule->up.idle_timeout |
1745cd08 | 4323 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4324 | reason = OFPRR_IDLE_TIMEOUT; |
4325 | } else { | |
4326 | return; | |
4327 | } | |
4328 | ||
4329 | COVERAGE_INC(ofproto_dpif_expired); | |
4330 | ||
4331 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4332 | * timeout, because that only happens when the rule has no facets left.) */ | |
4333 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 4334 | facet_remove(facet); |
abe529af BP |
4335 | } |
4336 | ||
4337 | /* Get rid of the rule. */ | |
4338 | ofproto_rule_expire(&rule->up, reason); | |
4339 | } | |
4340 | \f | |
4341 | /* Facets. */ | |
4342 | ||
f3827897 | 4343 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
4344 | * |
4345 | * The caller must already have determined that no facet with an identical | |
4346 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
4347 | * the ofproto's classifier table. |
4348 | * | |
2b459b83 BP |
4349 | * 'hash' must be the return value of flow_hash(flow, 0). |
4350 | * | |
b0f7b9b5 BP |
4351 | * The facet will initially have no subfacets. The caller should create (at |
4352 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4353 | static struct facet * |
2b459b83 | 4354 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
4355 | { |
4356 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4357 | struct facet *facet; | |
4358 | ||
4359 | facet = xzalloc(sizeof *facet); | |
4360 | facet->used = time_msec(); | |
2b459b83 | 4361 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
4362 | list_push_back(&rule->facets, &facet->list_node); |
4363 | facet->rule = rule; | |
4364 | facet->flow = *flow; | |
b0f7b9b5 | 4365 | list_init(&facet->subfacets); |
abe529af BP |
4366 | netflow_flow_init(&facet->nf_flow); |
4367 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4368 | ||
abe529af BP |
4369 | return facet; |
4370 | } | |
4371 | ||
4372 | static void | |
4373 | facet_free(struct facet *facet) | |
4374 | { | |
abe529af BP |
4375 | free(facet); |
4376 | } | |
4377 | ||
3d9e05f8 | 4378 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4379 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4380 | static bool |
4381 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4382 | const struct nlattr *odp_actions, size_t actions_len, | |
4383 | struct ofpbuf *packet) | |
4384 | { | |
4385 | struct odputil_keybuf keybuf; | |
4386 | struct ofpbuf key; | |
4387 | int error; | |
4388 | ||
6ff686f2 | 4389 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
4390 | odp_flow_key_from_flow(&key, flow, |
4391 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
80e5eed9 | 4392 | |
acf60855 | 4393 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4394 | odp_actions, actions_len, packet); |
6ff686f2 | 4395 | return !error; |
abe529af BP |
4396 | } |
4397 | ||
abe529af BP |
4398 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
4399 | * | |
4400 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4401 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4402 | * |
4403 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4404 | */ | |
4405 | static void | |
15baa734 | 4406 | facet_remove(struct facet *facet) |
abe529af | 4407 | { |
15baa734 | 4408 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4409 | struct subfacet *subfacet, *next_subfacet; |
4410 | ||
cb22974d | 4411 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4412 | |
4413 | /* First uninstall all of the subfacets to get final statistics. */ | |
4414 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4415 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4416 | } |
4417 | ||
4418 | /* Flush the final stats to the rule. | |
4419 | * | |
4420 | * This might require us to have at least one subfacet around so that we | |
4421 | * can use its actions for accounting in facet_account(), which is why we | |
4422 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4423 | facet_flush_stats(facet); |
551a2f6c BP |
4424 | |
4425 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4426 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4427 | &facet->subfacets) { | |
15baa734 | 4428 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4429 | } |
abe529af BP |
4430 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
4431 | list_remove(&facet->list_node); | |
4432 | facet_free(facet); | |
4433 | } | |
4434 | ||
3de9590b BP |
4435 | /* Feed information from 'facet' back into the learning table to keep it in |
4436 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4437 | static void |
3de9590b | 4438 | facet_learn(struct facet *facet) |
abe529af | 4439 | { |
15baa734 | 4440 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3de9590b | 4441 | struct action_xlate_ctx ctx; |
abe529af | 4442 | |
3de9590b BP |
4443 | if (!facet->has_learn |
4444 | && !facet->has_normal | |
4445 | && (!facet->has_fin_timeout | |
4446 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
4447 | return; |
4448 | } | |
abe529af | 4449 | |
3de9590b BP |
4450 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
4451 | facet->flow.vlan_tci, | |
4452 | facet->rule, facet->tcp_flags, NULL); | |
4453 | ctx.may_learn = true; | |
f25d0cf3 BP |
4454 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, |
4455 | facet->rule->up.ofpacts_len); | |
3de9590b BP |
4456 | } |
4457 | ||
4458 | static void | |
4459 | facet_account(struct facet *facet) | |
4460 | { | |
4461 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4462 | struct subfacet *subfacet; | |
4463 | const struct nlattr *a; | |
4464 | unsigned int left; | |
4465 | ovs_be16 vlan_tci; | |
4466 | uint64_t n_bytes; | |
abe529af | 4467 | |
75a75043 | 4468 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
4469 | return; |
4470 | } | |
3de9590b | 4471 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4472 | |
4473 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4474 | * as a basis. We also need to track the actual VLAN on which the packet | |
4475 | * is going to be sent to ensure that it matches the one passed to | |
4476 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4477 | * hash bucket.) |
4478 | * | |
4479 | * We use the actions from an arbitrary subfacet because they should all | |
4480 | * be equally valid for our purpose. */ | |
4481 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
4482 | struct subfacet, list_node); | |
d78be13b | 4483 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
4484 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
4485 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 4486 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4487 | struct ofport_dpif *port; |
abe529af | 4488 | |
d78be13b | 4489 | switch (nl_attr_type(a)) { |
df2c07f4 | 4490 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
4491 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
4492 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 4493 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4494 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4495 | } |
d78be13b BP |
4496 | break; |
4497 | ||
fea393b1 BP |
4498 | case OVS_ACTION_ATTR_POP_VLAN: |
4499 | vlan_tci = htons(0); | |
d78be13b BP |
4500 | break; |
4501 | ||
fea393b1 BP |
4502 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4503 | vlan = nl_attr_get(a); | |
4504 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4505 | break; |
abe529af BP |
4506 | } |
4507 | } | |
4508 | } | |
4509 | ||
abe529af BP |
4510 | /* Returns true if the only action for 'facet' is to send to the controller. |
4511 | * (We don't report NetFlow expiration messages for such facets because they | |
4512 | * are just part of the control logic for the network, not real traffic). */ | |
4513 | static bool | |
4514 | facet_is_controller_flow(struct facet *facet) | |
4515 | { | |
f25d0cf3 BP |
4516 | if (facet) { |
4517 | const struct rule *rule = &facet->rule->up; | |
4518 | const struct ofpact *ofpacts = rule->ofpacts; | |
4519 | size_t ofpacts_len = rule->ofpacts_len; | |
4520 | ||
dd30ff28 BP |
4521 | if (ofpacts_len > 0 && |
4522 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4523 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4524 | return true; | |
4525 | } | |
4526 | } | |
4527 | return false; | |
abe529af BP |
4528 | } |
4529 | ||
4530 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4531 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4532 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4533 | * its packet and byte counts before this function is called. */ | |
4534 | static void | |
15baa734 | 4535 | facet_flush_stats(struct facet *facet) |
abe529af | 4536 | { |
15baa734 | 4537 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4538 | struct subfacet *subfacet; |
4539 | ||
4540 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4541 | ovs_assert(!subfacet->dp_byte_count); |
4542 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4543 | } |
abe529af BP |
4544 | |
4545 | facet_push_stats(facet); | |
3de9590b BP |
4546 | if (facet->accounted_bytes < facet->byte_count) { |
4547 | facet_account(facet); | |
4548 | facet->accounted_bytes = facet->byte_count; | |
4549 | } | |
abe529af BP |
4550 | |
4551 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4552 | struct ofexpired expired; | |
4553 | expired.flow = facet->flow; | |
4554 | expired.packet_count = facet->packet_count; | |
4555 | expired.byte_count = facet->byte_count; | |
4556 | expired.used = facet->used; | |
4557 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4558 | } | |
4559 | ||
4560 | facet->rule->packet_count += facet->packet_count; | |
4561 | facet->rule->byte_count += facet->byte_count; | |
4562 | ||
4563 | /* Reset counters to prevent double counting if 'facet' ever gets | |
4564 | * reinstalled. */ | |
bbb5d219 | 4565 | facet_reset_counters(facet); |
abe529af BP |
4566 | |
4567 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4568 | facet->tcp_flags = 0; |
abe529af BP |
4569 | } |
4570 | ||
4571 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4572 | * Returns it if found, otherwise a null pointer. | |
4573 | * | |
2b459b83 BP |
4574 | * 'hash' must be the return value of flow_hash(flow, 0). |
4575 | * | |
abe529af BP |
4576 | * The returned facet might need revalidation; use facet_lookup_valid() |
4577 | * instead if that is important. */ | |
4578 | static struct facet * | |
2b459b83 BP |
4579 | facet_find(struct ofproto_dpif *ofproto, |
4580 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
4581 | { |
4582 | struct facet *facet; | |
4583 | ||
2b459b83 | 4584 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
4585 | if (flow_equal(flow, &facet->flow)) { |
4586 | return facet; | |
4587 | } | |
4588 | } | |
4589 | ||
4590 | return NULL; | |
4591 | } | |
4592 | ||
4593 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4594 | * Returns it if found, otherwise a null pointer. | |
4595 | * | |
2b459b83 BP |
4596 | * 'hash' must be the return value of flow_hash(flow, 0). |
4597 | * | |
abe529af BP |
4598 | * The returned facet is guaranteed to be valid. */ |
4599 | static struct facet * | |
2b459b83 BP |
4600 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
4601 | uint32_t hash) | |
abe529af | 4602 | { |
c57b2226 | 4603 | struct facet *facet; |
abe529af | 4604 | |
c57b2226 | 4605 | facet = facet_find(ofproto, flow, hash); |
abe529af | 4606 | if (facet |
2cc3c58e EJ |
4607 | && (ofproto->backer->need_revalidate |
4608 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
4609 | facet->tags))) { | |
c57b2226 | 4610 | facet_revalidate(facet); |
abe529af BP |
4611 | } |
4612 | ||
4613 | return facet; | |
4614 | } | |
4615 | ||
6a7e895f BP |
4616 | static const char * |
4617 | subfacet_path_to_string(enum subfacet_path path) | |
4618 | { | |
4619 | switch (path) { | |
4620 | case SF_NOT_INSTALLED: | |
4621 | return "not installed"; | |
4622 | case SF_FAST_PATH: | |
4623 | return "in fast path"; | |
4624 | case SF_SLOW_PATH: | |
4625 | return "in slow path"; | |
4626 | default: | |
4627 | return "<error>"; | |
4628 | } | |
4629 | } | |
4630 | ||
4631 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
4632 | * member has the specified value. */ | |
4633 | static enum subfacet_path | |
4634 | subfacet_want_path(enum slow_path_reason slow) | |
4635 | { | |
4636 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
4637 | } | |
4638 | ||
4639 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
4640 | * supposing that its actions have been recalculated as 'want_actions' and that | |
4641 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
4642 | static bool | |
4643 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
4644 | const struct ofpbuf *want_actions) | |
4645 | { | |
4646 | enum subfacet_path want_path = subfacet_want_path(slow); | |
4647 | return (want_path != subfacet->path | |
4648 | || (want_path == SF_FAST_PATH | |
4649 | && (subfacet->actions_len != want_actions->size | |
4650 | || memcmp(subfacet->actions, want_actions->data, | |
4651 | subfacet->actions_len)))); | |
4652 | } | |
4653 | ||
6814e51f BP |
4654 | static bool |
4655 | facet_check_consistency(struct facet *facet) | |
4656 | { | |
4657 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4658 | ||
4659 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4660 | ||
050ac423 BP |
4661 | uint64_t odp_actions_stub[1024 / 8]; |
4662 | struct ofpbuf odp_actions; | |
4663 | ||
6814e51f BP |
4664 | struct rule_dpif *rule; |
4665 | struct subfacet *subfacet; | |
c53e1132 | 4666 | bool may_log = false; |
6814e51f BP |
4667 | bool ok; |
4668 | ||
4669 | /* Check the rule for consistency. */ | |
c57b2226 BP |
4670 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
4671 | ok = rule == facet->rule; | |
4672 | if (!ok) { | |
c53e1132 | 4673 | may_log = !VLOG_DROP_WARN(&rl); |
c53e1132 BP |
4674 | if (may_log) { |
4675 | struct ds s; | |
6814e51f | 4676 | |
c53e1132 BP |
4677 | ds_init(&s); |
4678 | flow_format(&s, &facet->flow); | |
4679 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4680 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4681 | cls_rule_format(&facet->rule->up.cr, &s); | |
4682 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4683 | rule->up.table_id); | |
4684 | cls_rule_format(&rule->up.cr, &s); | |
4685 | ds_put_char(&s, ')'); | |
6814e51f | 4686 | |
c53e1132 BP |
4687 | VLOG_WARN("%s", ds_cstr(&s)); |
4688 | ds_destroy(&s); | |
4689 | } | |
6814e51f BP |
4690 | } |
4691 | ||
4692 | /* Check the datapath actions for consistency. */ | |
050ac423 | 4693 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f | 4694 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4695 | enum subfacet_path want_path; |
9616614b | 4696 | struct odputil_keybuf keybuf; |
6814e51f | 4697 | struct action_xlate_ctx ctx; |
9616614b BP |
4698 | struct ofpbuf key; |
4699 | struct ds s; | |
6814e51f BP |
4700 | |
4701 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
0e553d9c | 4702 | subfacet->initial_tci, rule, 0, NULL); |
f25d0cf3 | 4703 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 4704 | &odp_actions); |
6814e51f | 4705 | |
6a7e895f BP |
4706 | if (subfacet->path == SF_NOT_INSTALLED) { |
4707 | /* This only happens if the datapath reported an error when we | |
4708 | * tried to install the flow. Don't flag another error here. */ | |
4709 | continue; | |
4710 | } | |
4711 | ||
4712 | want_path = subfacet_want_path(subfacet->slow); | |
4713 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
4714 | /* The actions for slow-path flows may legitimately vary from one | |
4715 | * packet to the next. We're done. */ | |
050ac423 | 4716 | continue; |
6814e51f BP |
4717 | } |
4718 | ||
6a7e895f | 4719 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { |
9616614b BP |
4720 | continue; |
4721 | } | |
c53e1132 | 4722 | |
9616614b BP |
4723 | /* Inconsistency! */ |
4724 | if (ok) { | |
4725 | may_log = !VLOG_DROP_WARN(&rl); | |
4726 | ok = false; | |
4727 | } | |
4728 | if (!may_log) { | |
4729 | /* Rate-limited, skip reporting. */ | |
4730 | continue; | |
4731 | } | |
c53e1132 | 4732 | |
9616614b BP |
4733 | ds_init(&s); |
4734 | subfacet_get_key(subfacet, &keybuf, &key); | |
4735 | odp_flow_key_format(key.data, key.size, &s); | |
4736 | ||
4737 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
6a7e895f | 4738 | if (want_path != subfacet->path) { |
9616614b BP |
4739 | enum odp_key_fitness fitness = subfacet->key_fitness; |
4740 | ||
6a7e895f BP |
4741 | ds_put_format(&s, " (%s, fitness=%s)", |
4742 | subfacet_path_to_string(subfacet->path), | |
9616614b | 4743 | odp_key_fitness_to_string(fitness)); |
6a7e895f BP |
4744 | ds_put_format(&s, " (should have been %s)", |
4745 | subfacet_path_to_string(want_path)); | |
4746 | } else if (want_path == SF_FAST_PATH) { | |
9616614b BP |
4747 | ds_put_cstr(&s, " (actions were: "); |
4748 | format_odp_actions(&s, subfacet->actions, | |
4749 | subfacet->actions_len); | |
4750 | ds_put_cstr(&s, ") (correct actions: "); | |
4751 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
4752 | ds_put_char(&s, ')'); | |
4753 | } else { | |
4754 | ds_put_cstr(&s, " (actions: "); | |
4755 | format_odp_actions(&s, subfacet->actions, | |
4756 | subfacet->actions_len); | |
4757 | ds_put_char(&s, ')'); | |
6814e51f | 4758 | } |
9616614b BP |
4759 | VLOG_WARN("%s", ds_cstr(&s)); |
4760 | ds_destroy(&s); | |
6814e51f | 4761 | } |
050ac423 | 4762 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
4763 | |
4764 | return ok; | |
4765 | } | |
4766 | ||
15baa734 | 4767 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4768 | * |
4769 | * - If the rule found is different from 'facet''s current rule, moves | |
4770 | * 'facet' to the new rule and recompiles its actions. | |
4771 | * | |
4772 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
c57b2226 BP |
4773 | * where it is and recompiles its actions anyway. */ |
4774 | static void | |
15baa734 | 4775 | facet_revalidate(struct facet *facet) |
abe529af | 4776 | { |
15baa734 | 4777 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
4778 | struct actions { |
4779 | struct nlattr *odp_actions; | |
4780 | size_t actions_len; | |
4781 | }; | |
4782 | struct actions *new_actions; | |
4783 | ||
abe529af | 4784 | struct action_xlate_ctx ctx; |
050ac423 BP |
4785 | uint64_t odp_actions_stub[1024 / 8]; |
4786 | struct ofpbuf odp_actions; | |
4787 | ||
abe529af | 4788 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4789 | struct subfacet *subfacet; |
b95fc6ba | 4790 | int i; |
abe529af BP |
4791 | |
4792 | COVERAGE_INC(facet_revalidate); | |
4793 | ||
c57b2226 | 4794 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 4795 | |
df2c07f4 | 4796 | /* Calculate new datapath actions. |
abe529af BP |
4797 | * |
4798 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4799 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4800 | * around to properly compose it. */ | |
abe529af | 4801 | |
df2c07f4 JP |
4802 | /* If the datapath actions changed or the installability changed, |
4803 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
4804 | i = 0; |
4805 | new_actions = NULL; | |
4806 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 4807 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 4808 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4809 | enum slow_path_reason slow; |
b95fc6ba | 4810 | |
e84173dc | 4811 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
0e553d9c | 4812 | subfacet->initial_tci, new_rule, 0, NULL); |
f25d0cf3 | 4813 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, |
050ac423 | 4814 | &odp_actions); |
b0f7b9b5 | 4815 | |
6a7e895f BP |
4816 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
4817 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
4818 | struct dpif_flow_stats stats; | |
4819 | ||
4820 | subfacet_install(subfacet, | |
4821 | odp_actions.data, odp_actions.size, &stats, slow); | |
4822 | subfacet_update_stats(subfacet, &stats); | |
b95fc6ba BP |
4823 | |
4824 | if (!new_actions) { | |
4825 | new_actions = xcalloc(list_size(&facet->subfacets), | |
4826 | sizeof *new_actions); | |
4827 | } | |
050ac423 BP |
4828 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
4829 | odp_actions.size); | |
4830 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 4831 | } |
b95fc6ba | 4832 | |
b95fc6ba | 4833 | i++; |
b0f7b9b5 | 4834 | } |
050ac423 BP |
4835 | ofpbuf_uninit(&odp_actions); |
4836 | ||
b95fc6ba | 4837 | if (new_actions) { |
15baa734 | 4838 | facet_flush_stats(facet); |
abe529af BP |
4839 | } |
4840 | ||
4841 | /* Update 'facet' now that we've taken care of all the old state. */ | |
4842 | facet->tags = ctx.tags; | |
4843 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
75a75043 BP |
4844 | facet->has_learn = ctx.has_learn; |
4845 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4846 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 4847 | facet->mirrors = ctx.mirrors; |
6a7e895f BP |
4848 | |
4849 | i = 0; | |
4850 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4851 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4852 | ||
4853 | if (new_actions && new_actions[i].odp_actions) { | |
4854 | free(subfacet->actions); | |
4855 | subfacet->actions = new_actions[i].odp_actions; | |
4856 | subfacet->actions_len = new_actions[i].actions_len; | |
b95fc6ba | 4857 | } |
6a7e895f | 4858 | i++; |
abe529af | 4859 | } |
6a7e895f BP |
4860 | free(new_actions); |
4861 | ||
abe529af BP |
4862 | if (facet->rule != new_rule) { |
4863 | COVERAGE_INC(facet_changed_rule); | |
4864 | list_remove(&facet->list_node); | |
4865 | list_push_back(&new_rule->facets, &facet->list_node); | |
4866 | facet->rule = new_rule; | |
4867 | facet->used = new_rule->up.created; | |
9d24de3b | 4868 | facet->prev_used = facet->used; |
abe529af | 4869 | } |
abe529af BP |
4870 | } |
4871 | ||
4872 | /* Updates 'facet''s used time. Caller is responsible for calling | |
4873 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
4874 | static void | |
15baa734 | 4875 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 4876 | { |
15baa734 | 4877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
4878 | if (used > facet->used) { |
4879 | facet->used = used; | |
1745cd08 | 4880 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
4881 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
4882 | } | |
4883 | } | |
4884 | ||
bbb5d219 EJ |
4885 | static void |
4886 | facet_reset_counters(struct facet *facet) | |
4887 | { | |
4888 | facet->packet_count = 0; | |
4889 | facet->byte_count = 0; | |
9d24de3b JP |
4890 | facet->prev_packet_count = 0; |
4891 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4892 | facet->accounted_bytes = 0; |
4893 | } | |
4894 | ||
abe529af BP |
4895 | static void |
4896 | facet_push_stats(struct facet *facet) | |
4897 | { | |
112bc5f4 | 4898 | struct dpif_flow_stats stats; |
abe529af | 4899 | |
cb22974d BP |
4900 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4901 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4902 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4903 | |
112bc5f4 BP |
4904 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4905 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4906 | stats.used = facet->used; | |
4907 | stats.tcp_flags = 0; | |
abe529af | 4908 | |
112bc5f4 | 4909 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
4910 | facet->prev_packet_count = facet->packet_count; |
4911 | facet->prev_byte_count = facet->byte_count; | |
4912 | facet->prev_used = facet->used; | |
abe529af | 4913 | |
112bc5f4 | 4914 | flow_push_stats(facet->rule, &facet->flow, &stats); |
9d24de3b JP |
4915 | |
4916 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 4917 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
4918 | } |
4919 | } | |
4920 | ||
abe529af | 4921 | static void |
112bc5f4 | 4922 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4923 | { |
112bc5f4 BP |
4924 | rule->packet_count += stats->n_packets; |
4925 | rule->byte_count += stats->n_bytes; | |
4926 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
4927 | } |
4928 | ||
4929 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 4930 | * 'rule''s actions and mirrors. */ |
abe529af | 4931 | static void |
18b2a258 | 4932 | flow_push_stats(struct rule_dpif *rule, |
112bc5f4 | 4933 | const struct flow *flow, const struct dpif_flow_stats *stats) |
abe529af BP |
4934 | { |
4935 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
112bc5f4 | 4936 | struct action_xlate_ctx ctx; |
abe529af | 4937 | |
112bc5f4 | 4938 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 4939 | |
112bc5f4 | 4940 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, |
0e553d9c | 4941 | 0, NULL); |
112bc5f4 | 4942 | ctx.resubmit_stats = stats; |
f25d0cf3 BP |
4943 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, |
4944 | rule->up.ofpacts_len); | |
abe529af BP |
4945 | } |
4946 | \f | |
b0f7b9b5 BP |
4947 | /* Subfacets. */ |
4948 | ||
4949 | static struct subfacet * | |
acf60855 JP |
4950 | subfacet_find(struct ofproto_dpif *ofproto, |
4951 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
4952 | const struct flow *flow) | |
b0f7b9b5 BP |
4953 | { |
4954 | struct subfacet *subfacet; | |
4955 | ||
4956 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
4957 | &ofproto->subfacets) { | |
4958 | if (subfacet->key | |
4959 | ? (subfacet->key_len == key_len | |
4960 | && !memcmp(key, subfacet->key, key_len)) | |
4961 | : flow_equal(flow, &subfacet->facet->flow)) { | |
4962 | return subfacet; | |
4963 | } | |
4964 | } | |
4965 | ||
4966 | return NULL; | |
4967 | } | |
4968 | ||
4969 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
4970 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
4971 | * existing subfacet if there is one, otherwise creates and returns a | |
4972 | * new subfacet. | |
b95fc6ba BP |
4973 | * |
4974 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
4975 | * which case the caller must populate the actions with | |
4976 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 4977 | static struct subfacet * |
a088a1ff JP |
4978 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
4979 | long long int now) | |
b0f7b9b5 | 4980 | { |
15baa734 | 4981 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
a088a1ff JP |
4982 | enum odp_key_fitness key_fitness = miss->key_fitness; |
4983 | const struct nlattr *key = miss->key; | |
4984 | size_t key_len = miss->key_len; | |
4985 | uint32_t key_hash; | |
b0f7b9b5 BP |
4986 | struct subfacet *subfacet; |
4987 | ||
a088a1ff JP |
4988 | key_hash = odp_flow_key_hash(key, key_len); |
4989 | ||
3b145dd7 BP |
4990 | if (list_is_empty(&facet->subfacets)) { |
4991 | subfacet = &facet->one_subfacet; | |
4992 | } else { | |
acf60855 JP |
4993 | subfacet = subfacet_find(ofproto, key, key_len, key_hash, |
4994 | &facet->flow); | |
3b145dd7 BP |
4995 | if (subfacet) { |
4996 | if (subfacet->facet == facet) { | |
4997 | return subfacet; | |
4998 | } | |
4999 | ||
5000 | /* This shouldn't happen. */ | |
5001 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5002 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
5003 | } |
5004 | ||
3b145dd7 | 5005 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
5006 | } |
5007 | ||
b0f7b9b5 BP |
5008 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
5009 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
5010 | subfacet->facet = facet; | |
b0f7b9b5 BP |
5011 | subfacet->key_fitness = key_fitness; |
5012 | if (key_fitness != ODP_FIT_PERFECT) { | |
5013 | subfacet->key = xmemdup(key, key_len); | |
5014 | subfacet->key_len = key_len; | |
26cd7e34 BP |
5015 | } else { |
5016 | subfacet->key = NULL; | |
5017 | subfacet->key_len = 0; | |
b0f7b9b5 | 5018 | } |
459b16a1 | 5019 | subfacet->used = now; |
26cd7e34 BP |
5020 | subfacet->dp_packet_count = 0; |
5021 | subfacet->dp_byte_count = 0; | |
5022 | subfacet->actions_len = 0; | |
5023 | subfacet->actions = NULL; | |
6a7e895f BP |
5024 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE |
5025 | ? SLOW_MATCH | |
5026 | : 0); | |
5027 | subfacet->path = SF_NOT_INSTALLED; | |
a088a1ff JP |
5028 | subfacet->initial_tci = miss->initial_tci; |
5029 | subfacet->odp_in_port = miss->odp_in_port; | |
b0f7b9b5 BP |
5030 | |
5031 | return subfacet; | |
5032 | } | |
5033 | ||
b0f7b9b5 BP |
5034 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
5035 | * its facet within 'ofproto', and frees it. */ | |
5036 | static void | |
15baa734 | 5037 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 5038 | { |
15baa734 BP |
5039 | struct facet *facet = subfacet->facet; |
5040 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5041 | ||
5042 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
5043 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
5044 | list_remove(&subfacet->list_node); | |
5045 | free(subfacet->key); | |
b95fc6ba | 5046 | free(subfacet->actions); |
26cd7e34 BP |
5047 | if (subfacet != &facet->one_subfacet) { |
5048 | free(subfacet); | |
5049 | } | |
b0f7b9b5 BP |
5050 | } |
5051 | ||
5052 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5053 | * last remaining subfacet in its facet destroys the facet too. */ | |
5054 | static void | |
15baa734 | 5055 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
5056 | { |
5057 | struct facet *facet = subfacet->facet; | |
5058 | ||
551a2f6c BP |
5059 | if (list_is_singleton(&facet->subfacets)) { |
5060 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 5061 | facet_remove(facet); |
551a2f6c | 5062 | } else { |
15baa734 | 5063 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
5064 | } |
5065 | } | |
5066 | ||
1d85f9e5 JP |
5067 | static void |
5068 | subfacet_destroy_batch(struct ofproto_dpif *ofproto, | |
5069 | struct subfacet **subfacets, int n) | |
5070 | { | |
5071 | struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH]; | |
5072 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; | |
5073 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
5074 | struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH]; | |
5075 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; | |
5076 | int i; | |
5077 | ||
5078 | for (i = 0; i < n; i++) { | |
5079 | ops[i].type = DPIF_OP_FLOW_DEL; | |
5080 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
5081 | ops[i].u.flow_del.key = keys[i].data; | |
5082 | ops[i].u.flow_del.key_len = keys[i].size; | |
5083 | ops[i].u.flow_del.stats = &stats[i]; | |
5084 | opsp[i] = &ops[i]; | |
5085 | } | |
5086 | ||
acf60855 | 5087 | dpif_operate(ofproto->backer->dpif, opsp, n); |
1d85f9e5 JP |
5088 | for (i = 0; i < n; i++) { |
5089 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5090 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5091 | subfacet_destroy(subfacets[i]); | |
5092 | } | |
5093 | } | |
5094 | ||
b0f7b9b5 BP |
5095 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes |
5096 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
5097 | * for use as temporary storage. */ | |
5098 | static void | |
5099 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
5100 | struct ofpbuf *key) | |
5101 | { | |
e1b1d06a | 5102 | |
b0f7b9b5 | 5103 | if (!subfacet->key) { |
ddbfda84 JP |
5104 | struct flow *flow = &subfacet->facet->flow; |
5105 | ||
b0f7b9b5 | 5106 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); |
a088a1ff | 5107 | odp_flow_key_from_flow(key, flow, subfacet->odp_in_port); |
b0f7b9b5 BP |
5108 | } else { |
5109 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
5110 | } | |
5111 | } | |
5112 | ||
5fe20d5d BP |
5113 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
5114 | * Translates the actions into 'odp_actions', which the caller must have | |
5115 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 5116 | static void |
5fe20d5d BP |
5117 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
5118 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
5119 | { |
5120 | struct facet *facet = subfacet->facet; | |
18b2a258 | 5121 | struct rule_dpif *rule = facet->rule; |
15baa734 | 5122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 5123 | |
b95fc6ba BP |
5124 | struct action_xlate_ctx ctx; |
5125 | ||
15baa734 | 5126 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
0e553d9c | 5127 | rule, 0, packet); |
f25d0cf3 | 5128 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); |
b95fc6ba | 5129 | facet->tags = ctx.tags; |
b95fc6ba BP |
5130 | facet->has_learn = ctx.has_learn; |
5131 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 5132 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 5133 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 5134 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 5135 | |
6a7e895f | 5136 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5fe20d5d BP |
5137 | if (subfacet->actions_len != odp_actions->size |
5138 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 5139 | free(subfacet->actions); |
5fe20d5d BP |
5140 | subfacet->actions_len = odp_actions->size; |
5141 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 5142 | } |
b95fc6ba BP |
5143 | } |
5144 | ||
b0f7b9b5 BP |
5145 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5146 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5147 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5148 | * since 'subfacet' was last updated. | |
5149 | * | |
5150 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5151 | static int | |
15baa734 | 5152 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 | 5153 | const struct nlattr *actions, size_t actions_len, |
6a7e895f BP |
5154 | struct dpif_flow_stats *stats, |
5155 | enum slow_path_reason slow) | |
b0f7b9b5 | 5156 | { |
15baa734 BP |
5157 | struct facet *facet = subfacet->facet; |
5158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
6a7e895f BP |
5159 | enum subfacet_path path = subfacet_want_path(slow); |
5160 | uint64_t slow_path_stub[128 / 8]; | |
b0f7b9b5 BP |
5161 | struct odputil_keybuf keybuf; |
5162 | enum dpif_flow_put_flags flags; | |
5163 | struct ofpbuf key; | |
5164 | int ret; | |
5165 | ||
5166 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5167 | if (stats) { | |
5168 | flags |= DPIF_FP_ZERO_STATS; | |
5169 | } | |
5170 | ||
6a7e895f BP |
5171 | if (path == SF_SLOW_PATH) { |
5172 | compose_slow_path(ofproto, &facet->flow, slow, | |
5173 | slow_path_stub, sizeof slow_path_stub, | |
5174 | &actions, &actions_len); | |
5175 | } | |
5176 | ||
b0f7b9b5 | 5177 | subfacet_get_key(subfacet, &keybuf, &key); |
acf60855 | 5178 | ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size, |
b0f7b9b5 BP |
5179 | actions, actions_len, stats); |
5180 | ||
5181 | if (stats) { | |
5182 | subfacet_reset_dp_stats(subfacet, stats); | |
5183 | } | |
5184 | ||
6a7e895f BP |
5185 | if (!ret) { |
5186 | subfacet->path = path; | |
5187 | } | |
b0f7b9b5 BP |
5188 | return ret; |
5189 | } | |
5190 | ||
6a7e895f BP |
5191 | static int |
5192 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
5193 | { | |
5194 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
5195 | stats, subfacet->slow); | |
5196 | } | |
5197 | ||
b0f7b9b5 BP |
5198 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ |
5199 | static void | |
15baa734 | 5200 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5201 | { |
6a7e895f | 5202 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
5203 | struct rule_dpif *rule = subfacet->facet->rule; |
5204 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
5205 | struct odputil_keybuf keybuf; |
5206 | struct dpif_flow_stats stats; | |
5207 | struct ofpbuf key; | |
5208 | int error; | |
5209 | ||
5210 | subfacet_get_key(subfacet, &keybuf, &key); | |
acf60855 JP |
5211 | error = dpif_flow_del(ofproto->backer->dpif, |
5212 | key.data, key.size, &stats); | |
b0f7b9b5 BP |
5213 | subfacet_reset_dp_stats(subfacet, &stats); |
5214 | if (!error) { | |
15baa734 | 5215 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5216 | } |
6a7e895f | 5217 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5218 | } else { |
cb22974d BP |
5219 | ovs_assert(subfacet->dp_packet_count == 0); |
5220 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5221 | } |
5222 | } | |
5223 | ||
5224 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5225 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5226 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5227 | * was reset in the datapath. 'stats' will be modified to include only | |
5228 | * statistics new since 'subfacet' was last updated. */ | |
5229 | static void | |
5230 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5231 | struct dpif_flow_stats *stats) | |
5232 | { | |
5233 | if (stats | |
5234 | && subfacet->dp_packet_count <= stats->n_packets | |
5235 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5236 | stats->n_packets -= subfacet->dp_packet_count; | |
5237 | stats->n_bytes -= subfacet->dp_byte_count; | |
5238 | } | |
5239 | ||
5240 | subfacet->dp_packet_count = 0; | |
5241 | subfacet->dp_byte_count = 0; | |
5242 | } | |
5243 | ||
5244 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
5245 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
5246 | static void | |
15baa734 | 5247 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
5248 | { |
5249 | if (used > subfacet->used) { | |
5250 | subfacet->used = used; | |
15baa734 | 5251 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
5252 | } |
5253 | } | |
5254 | ||
5255 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
5256 | * | |
5257 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5258 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5259 | * represents a packet that was sent by hand or if it represents statistics | |
5260 | * that have been cleared out of the datapath. */ | |
5261 | static void | |
15baa734 | 5262 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5263 | const struct dpif_flow_stats *stats) |
5264 | { | |
5265 | if (stats->n_packets || stats->used > subfacet->used) { | |
5266 | struct facet *facet = subfacet->facet; | |
5267 | ||
15baa734 | 5268 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
5269 | facet->packet_count += stats->n_packets; |
5270 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5271 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5272 | facet_push_stats(facet); |
5273 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
5274 | } | |
5275 | } | |
5276 | \f | |
abe529af BP |
5277 | /* Rules. */ |
5278 | ||
5279 | static struct rule_dpif * | |
c57b2226 BP |
5280 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
5281 | { | |
c57b2226 BP |
5282 | struct rule_dpif *rule; |
5283 | ||
5284 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5285 | if (rule) { | |
5286 | return rule; | |
5287 | } | |
5288 | ||
c376f9a3 | 5289 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5290 | } |
5291 | ||
5292 | static struct rule_dpif * | |
5293 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5294 | uint8_t table_id) | |
abe529af | 5295 | { |
7257b535 BP |
5296 | struct cls_rule *cls_rule; |
5297 | struct classifier *cls; | |
5298 | ||
9cdaaebe BP |
5299 | if (table_id >= N_TABLES) { |
5300 | return NULL; | |
5301 | } | |
5302 | ||
d0918789 | 5303 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 5304 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
5305 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
5306 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
5307 | * are unavailable. */ | |
5308 | struct flow ofpc_normal_flow = *flow; | |
5309 | ofpc_normal_flow.tp_src = htons(0); | |
5310 | ofpc_normal_flow.tp_dst = htons(0); | |
5311 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
5312 | } else { | |
5313 | cls_rule = classifier_lookup(cls, flow); | |
5314 | } | |
5315 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5316 | } |
5317 | ||
c376f9a3 IY |
5318 | static struct rule_dpif * |
5319 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5320 | { | |
5321 | struct ofport_dpif *port; | |
5322 | ||
5323 | port = get_ofp_port(ofproto, flow->in_port); | |
5324 | if (!port) { | |
5325 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5326 | return ofproto->miss_rule; | |
5327 | } | |
5328 | ||
5329 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5330 | return ofproto->no_packet_in_rule; | |
5331 | } | |
5332 | return ofproto->miss_rule; | |
5333 | } | |
5334 | ||
7ee20df1 BP |
5335 | static void |
5336 | complete_operation(struct rule_dpif *rule) | |
5337 | { | |
5338 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5339 | ||
54a9cbc9 | 5340 | rule_invalidate(rule); |
7ee20df1 BP |
5341 | if (clogged) { |
5342 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5343 | c->op = rule->up.pending; | |
5344 | list_push_back(&ofproto->completions, &c->list_node); | |
5345 | } else { | |
5346 | ofoperation_complete(rule->up.pending, 0); | |
5347 | } | |
5348 | } | |
5349 | ||
abe529af BP |
5350 | static struct rule * |
5351 | rule_alloc(void) | |
5352 | { | |
5353 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5354 | return &rule->up; | |
5355 | } | |
5356 | ||
5357 | static void | |
5358 | rule_dealloc(struct rule *rule_) | |
5359 | { | |
5360 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5361 | free(rule); | |
5362 | } | |
5363 | ||
90bf1e07 | 5364 | static enum ofperr |
abe529af BP |
5365 | rule_construct(struct rule *rule_) |
5366 | { | |
5367 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5368 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 5369 | struct rule_dpif *victim; |
54a9cbc9 | 5370 | uint8_t table_id; |
abe529af | 5371 | |
abe529af BP |
5372 | rule->packet_count = 0; |
5373 | rule->byte_count = 0; | |
abe529af | 5374 | |
7ee20df1 BP |
5375 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
5376 | if (victim && !list_is_empty(&victim->facets)) { | |
5377 | struct facet *facet; | |
5378 | ||
5379 | rule->facets = victim->facets; | |
5380 | list_moved(&rule->facets); | |
5381 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
5382 | /* XXX: We're only clearing our local counters here. It's possible |
5383 | * that quite a few packets are unaccounted for in the datapath | |
5384 | * statistics. These will be accounted to the new rule instead of | |
5385 | * cleared as required. This could be fixed by clearing out the | |
5386 | * datapath statistics for this facet, but currently it doesn't | |
5387 | * seem worth it. */ | |
5388 | facet_reset_counters(facet); | |
7ee20df1 BP |
5389 | facet->rule = rule; |
5390 | } | |
5391 | } else { | |
5392 | /* Must avoid list_moved() in this case. */ | |
5393 | list_init(&rule->facets); | |
5394 | } | |
abe529af | 5395 | |
54a9cbc9 | 5396 | table_id = rule->up.table_id; |
5cb7a798 BP |
5397 | if (victim) { |
5398 | rule->tag = victim->tag; | |
5399 | } else if (table_id == 0) { | |
5400 | rule->tag = 0; | |
5401 | } else { | |
5402 | struct flow flow; | |
5403 | ||
5404 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5405 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5406 | ofproto->tables[table_id].basis); | |
5407 | } | |
54a9cbc9 | 5408 | |
7ee20df1 | 5409 | complete_operation(rule); |
abe529af BP |
5410 | return 0; |
5411 | } | |
5412 | ||
5413 | static void | |
5414 | rule_destruct(struct rule *rule_) | |
5415 | { | |
5416 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5417 | struct facet *facet, *next_facet; |
5418 | ||
abe529af | 5419 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 5420 | facet_revalidate(facet); |
abe529af | 5421 | } |
7ee20df1 BP |
5422 | |
5423 | complete_operation(rule); | |
abe529af BP |
5424 | } |
5425 | ||
5426 | static void | |
5427 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5428 | { | |
5429 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5430 | struct facet *facet; | |
5431 | ||
5432 | /* Start from historical data for 'rule' itself that are no longer tracked | |
5433 | * in facets. This counts, for example, facets that have expired. */ | |
5434 | *packets = rule->packet_count; | |
5435 | *bytes = rule->byte_count; | |
5436 | ||
5437 | /* Add any statistics that are tracked by facets. This includes | |
5438 | * statistical data recently updated by ofproto_update_stats() as well as | |
5439 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
5440 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
5441 | *packets += facet->packet_count; | |
5442 | *bytes += facet->byte_count; | |
5443 | } | |
5444 | } | |
5445 | ||
0a740f48 EJ |
5446 | static void |
5447 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5448 | struct ofpbuf *packet) | |
abe529af | 5449 | { |
abe529af | 5450 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 5451 | |
112bc5f4 | 5452 | struct dpif_flow_stats stats; |
050ac423 | 5453 | |
abe529af | 5454 | struct action_xlate_ctx ctx; |
050ac423 BP |
5455 | uint64_t odp_actions_stub[1024 / 8]; |
5456 | struct ofpbuf odp_actions; | |
abe529af | 5457 | |
a7752d4a | 5458 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5459 | rule_credit_stats(rule, &stats); |
5460 | ||
050ac423 | 5461 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
54834960 | 5462 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
112bc5f4 BP |
5463 | rule, stats.tcp_flags, packet); |
5464 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 5465 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); |
112bc5f4 BP |
5466 | |
5467 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
5468 | odp_actions.size, packet); | |
5469 | ||
050ac423 | 5470 | ofpbuf_uninit(&odp_actions); |
0a740f48 | 5471 | } |
5bf0e941 | 5472 | |
0a740f48 EJ |
5473 | static enum ofperr |
5474 | rule_execute(struct rule *rule, const struct flow *flow, | |
5475 | struct ofpbuf *packet) | |
5476 | { | |
5477 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5478 | ofpbuf_delete(packet); | |
5bf0e941 | 5479 | return 0; |
abe529af BP |
5480 | } |
5481 | ||
7ee20df1 BP |
5482 | static void |
5483 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5484 | { |
5485 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5486 | |
5487 | complete_operation(rule); | |
abe529af BP |
5488 | } |
5489 | \f | |
97d6520b | 5490 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5491 | * May modify 'packet'. |
abe529af BP |
5492 | * Returns 0 if successful, otherwise a positive errno value. */ |
5493 | static int | |
52a90c29 | 5494 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5495 | { |
97d6520b | 5496 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5497 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 BP |
5498 | struct ofpbuf key, odp_actions; |
5499 | struct odputil_keybuf keybuf; | |
9b56fe13 | 5500 | uint32_t odp_port; |
80e5eed9 | 5501 | struct flow flow; |
abe529af BP |
5502 | int error; |
5503 | ||
72e8bf28 | 5504 | flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow); |
0a740f48 EJ |
5505 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5506 | struct ofproto_dpif *peer_ofproto; | |
5507 | struct dpif_flow_stats stats; | |
5508 | struct ofport_dpif *peer; | |
5509 | struct rule_dpif *rule; | |
5510 | ||
5511 | peer = ofport_get_peer(ofport); | |
5512 | if (!peer) { | |
5513 | return ENODEV; | |
5514 | } | |
5515 | ||
5516 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 EJ |
5517 | netdev_vport_inc_tx(ofport->up.netdev, &stats); |
5518 | netdev_vport_inc_rx(peer->up.netdev, &stats); | |
0a740f48 EJ |
5519 | |
5520 | flow.in_port = peer->up.ofp_port; | |
5521 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5522 | rule = rule_dpif_lookup(peer_ofproto, &flow); | |
5523 | rule_dpif_execute(rule, &flow, packet); | |
5524 | ||
5525 | return 0; | |
5526 | } | |
5527 | ||
b9ad7294 EJ |
5528 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
5529 | ||
5530 | if (ofport->tnl_port) { | |
5531 | struct dpif_flow_stats stats; | |
5532 | ||
5533 | odp_port = tnl_port_send(ofport->tnl_port, &flow); | |
5534 | if (odp_port == OVSP_NONE) { | |
5535 | return ENODEV; | |
5536 | } | |
5537 | ||
5538 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
5539 | netdev_vport_inc_tx(ofport->up.netdev, &stats); | |
5540 | odp_put_tunnel_action(&flow.tunnel, &odp_actions); | |
09a0d2e7 | 5541 | odp_put_skb_mark_action(flow.skb_mark, &odp_actions); |
b9ad7294 EJ |
5542 | } else { |
5543 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, | |
5544 | flow.vlan_tci); | |
5545 | if (odp_port != ofport->odp_port) { | |
5546 | eth_pop_vlan(packet); | |
5547 | flow.vlan_tci = htons(0); | |
5548 | } | |
52a90c29 BP |
5549 | } |
5550 | ||
80e5eed9 | 5551 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
5552 | odp_flow_key_from_flow(&key, &flow, |
5553 | ofp_port_to_odp_port(ofproto, flow.in_port)); | |
80e5eed9 | 5554 | |
6ff686f2 PS |
5555 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
5556 | ||
df2c07f4 | 5557 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
acf60855 | 5558 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 BP |
5559 | key.data, key.size, |
5560 | odp_actions.data, odp_actions.size, | |
abe529af BP |
5561 | packet); |
5562 | ofpbuf_uninit(&odp_actions); | |
5563 | ||
5564 | if (error) { | |
5565 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
5566 | ofproto->up.name, odp_port, strerror(error)); | |
5567 | } | |
6527c598 | 5568 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
5569 | return error; |
5570 | } | |
5571 | \f | |
df2c07f4 | 5572 | /* OpenFlow to datapath action translation. */ |
abe529af | 5573 | |
ffaef958 | 5574 | static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *); |
f25d0cf3 BP |
5575 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
5576 | struct action_xlate_ctx *); | |
4cd78906 | 5577 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 5578 | |
6a7e895f BP |
5579 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5580 | * The action will state 'slow' as the reason that the action is in the slow | |
5581 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5582 | * dump-flows" output to see why a flow is in the slow path.) | |
5583 | * | |
5584 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5585 | * 'stub_size' must be large enough for the action. | |
5586 | * | |
5587 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5588 | * respectively. */ | |
5589 | static void | |
5590 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5591 | enum slow_path_reason slow, | |
5592 | uint64_t *stub, size_t stub_size, | |
5593 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5594 | { | |
5595 | union user_action_cookie cookie; | |
5596 | struct ofpbuf buf; | |
5597 | ||
5598 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5599 | cookie.slow_path.unused = 0; | |
5600 | cookie.slow_path.reason = slow; | |
5601 | ||
5602 | ofpbuf_use_stack(&buf, stub, stub_size); | |
625b0720 | 5603 | if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { |
9032f11e | 5604 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); |
e995e3df | 5605 | odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf); |
625b0720 BP |
5606 | } else { |
5607 | put_userspace_action(ofproto, &buf, flow, &cookie); | |
5608 | } | |
6a7e895f BP |
5609 | *actionsp = buf.data; |
5610 | *actions_lenp = buf.size; | |
5611 | } | |
5612 | ||
98403001 BP |
5613 | static size_t |
5614 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5615 | struct ofpbuf *odp_actions, | |
5616 | const struct flow *flow, | |
1673e0e4 | 5617 | const union user_action_cookie *cookie) |
98403001 | 5618 | { |
98403001 BP |
5619 | uint32_t pid; |
5620 | ||
acf60855 | 5621 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
e1b1d06a | 5622 | ofp_port_to_odp_port(ofproto, flow->in_port)); |
98403001 | 5623 | |
e995e3df | 5624 | return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions); |
98403001 BP |
5625 | } |
5626 | ||
36fc5f18 BP |
5627 | static void |
5628 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5629 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 5630 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
5631 | { |
5632 | int ifindex; | |
5633 | ||
5634 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 5635 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
5636 | |
5637 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5638 | * port information") for the interpretation of cookie->output. */ | |
5639 | switch (n_outputs) { | |
5640 | case 0: | |
5641 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 5642 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
5643 | break; |
5644 | ||
5645 | case 1: | |
5646 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5647 | if (ifindex) { | |
1673e0e4 | 5648 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
5649 | break; |
5650 | } | |
5651 | /* Fall through. */ | |
5652 | default: | |
5653 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 5654 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
5655 | break; |
5656 | } | |
5657 | } | |
5658 | ||
6ff686f2 PS |
5659 | /* Compose SAMPLE action for sFlow. */ |
5660 | static size_t | |
5661 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5662 | struct ofpbuf *odp_actions, | |
5663 | const struct flow *flow, | |
5664 | uint32_t odp_port) | |
5665 | { | |
6ff686f2 | 5666 | uint32_t probability; |
1673e0e4 | 5667 | union user_action_cookie cookie; |
6ff686f2 | 5668 | size_t sample_offset, actions_offset; |
36fc5f18 | 5669 | int cookie_offset; |
6ff686f2 PS |
5670 | |
5671 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5672 | return 0; | |
5673 | } | |
5674 | ||
6ff686f2 PS |
5675 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); |
5676 | ||
5677 | /* Number of packets out of UINT_MAX to sample. */ | |
5678 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
5679 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5680 | ||
5681 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
36fc5f18 BP |
5682 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
5683 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
98403001 | 5684 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); |
6ff686f2 PS |
5685 | |
5686 | nl_msg_end_nested(odp_actions, actions_offset); | |
5687 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 5688 | return cookie_offset; |
6ff686f2 PS |
5689 | } |
5690 | ||
5691 | /* SAMPLE action must be first action in any given list of actions. | |
5692 | * At this point we do not have all information required to build it. So try to | |
5693 | * build sample action as complete as possible. */ | |
5694 | static void | |
5695 | add_sflow_action(struct action_xlate_ctx *ctx) | |
5696 | { | |
5697 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
5698 | ctx->odp_actions, | |
5699 | &ctx->flow, OVSP_NONE); | |
5700 | ctx->sflow_odp_port = 0; | |
5701 | ctx->sflow_n_outputs = 0; | |
5702 | } | |
5703 | ||
5704 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
5705 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
5706 | * USERSPACE action's user-cookie which is required for sflow. */ | |
5707 | static void | |
5708 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
5709 | { | |
5710 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 5711 | union user_action_cookie *cookie; |
6ff686f2 PS |
5712 | |
5713 | if (!ctx->user_cookie_offset) { | |
5714 | return; | |
5715 | } | |
5716 | ||
5717 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
36fc5f18 | 5718 | sizeof(*cookie)); |
cb22974d | 5719 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
6ff686f2 | 5720 | |
36fc5f18 BP |
5721 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
5722 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
5723 | } |
5724 | ||
6ff686f2 | 5725 | static void |
81b1afb1 EJ |
5726 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
5727 | bool check_stp) | |
6ff686f2 | 5728 | { |
d59906fb | 5729 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
52a90c29 | 5730 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
b9ad7294 | 5731 | ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id; |
8b36f51e | 5732 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
a4454ac6 | 5733 | struct priority_to_dscp *pdscp; |
0a740f48 EJ |
5734 | uint32_t out_port, odp_port; |
5735 | ||
5736 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
5737 | * before traversing a patch port. */ | |
b02475c5 | 5738 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19); |
d59906fb | 5739 | |
a4454ac6 EJ |
5740 | if (!ofport) { |
5741 | xlate_report(ctx, "Nonexistent output port"); | |
5742 | return; | |
5743 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
5744 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
5745 | return; | |
5746 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
5747 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
5748 | return; | |
5749 | } | |
8b36f51e | 5750 | |
0a740f48 EJ |
5751 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5752 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
5753 | struct flow old_flow = ctx->flow; | |
5754 | const struct ofproto_dpif *peer_ofproto; | |
bb374ef6 | 5755 | enum slow_path_reason special; |
ffaef958 | 5756 | struct ofport_dpif *in_port; |
0a740f48 EJ |
5757 | |
5758 | if (!peer) { | |
5759 | xlate_report(ctx, "Nonexistent patch port peer"); | |
5760 | return; | |
5761 | } | |
5762 | ||
5763 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5764 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
5765 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
5766 | return; | |
5767 | } | |
5768 | ||
5769 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5770 | ctx->flow.in_port = peer->up.ofp_port; | |
5771 | ctx->flow.metadata = htonll(0); | |
5772 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); | |
5773 | memset(ctx->flow.regs, 0, sizeof ctx->flow.regs); | |
ffaef958 BP |
5774 | |
5775 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
bb374ef6 EJ |
5776 | special = process_special(ctx->ofproto, &ctx->flow, in_port, |
5777 | ctx->packet); | |
5778 | if (special) { | |
5779 | ctx->slow |= special; | |
5780 | } else if (!in_port || may_receive(in_port, ctx)) { | |
ffaef958 BP |
5781 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { |
5782 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
5783 | } else { | |
5784 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
5785 | * learning action look at the packet, then drop it. */ | |
5786 | struct flow old_base_flow = ctx->base_flow; | |
5787 | size_t old_size = ctx->odp_actions->size; | |
5788 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
5789 | ctx->base_flow = old_base_flow; | |
5790 | ctx->odp_actions->size = old_size; | |
5791 | } | |
5792 | } | |
5793 | ||
0a740f48 EJ |
5794 | ctx->flow = old_flow; |
5795 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
5796 | ||
5797 | if (ctx->resubmit_stats) { | |
b9ad7294 EJ |
5798 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); |
5799 | netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats); | |
0a740f48 EJ |
5800 | } |
5801 | ||
5802 | return; | |
5803 | } | |
5804 | ||
a4454ac6 EJ |
5805 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
5806 | if (pdscp) { | |
5807 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
5808 | ctx->flow.nw_tos |= pdscp->dscp; | |
d59906fb EJ |
5809 | } |
5810 | ||
0a740f48 | 5811 | odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port); |
b9ad7294 EJ |
5812 | if (ofport->tnl_port) { |
5813 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow); | |
5814 | if (odp_port == OVSP_NONE) { | |
5815 | xlate_report(ctx, "Tunneling decided against output"); | |
5816 | return; | |
5817 | } | |
5818 | ||
5819 | if (ctx->resubmit_stats) { | |
5820 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); | |
5821 | } | |
5822 | out_port = odp_port; | |
5823 | commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow, | |
5824 | ctx->odp_actions); | |
5825 | } else { | |
5826 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, | |
5827 | ctx->flow.vlan_tci); | |
5828 | if (out_port != odp_port) { | |
5829 | ctx->flow.vlan_tci = htons(0); | |
5830 | } | |
52a90c29 | 5831 | } |
5bbda0aa | 5832 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
5833 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
5834 | ||
6ff686f2 PS |
5835 | ctx->sflow_odp_port = odp_port; |
5836 | ctx->sflow_n_outputs++; | |
81b1afb1 | 5837 | ctx->nf_output_iface = ofp_port; |
b9ad7294 | 5838 | ctx->flow.tunnel.tun_id = flow_tun_id; |
52a90c29 | 5839 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 5840 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
5841 | } |
5842 | ||
abe529af | 5843 | static void |
5e48dc2b | 5844 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 5845 | { |
81b1afb1 | 5846 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
5847 | } |
5848 | ||
5849 | static void | |
29901626 | 5850 | xlate_table_action(struct action_xlate_ctx *ctx, |
1688c479 | 5851 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
5852 | { |
5853 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 5854 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
5855 | struct rule_dpif *rule; |
5856 | uint16_t old_in_port; | |
29901626 BP |
5857 | uint8_t old_table_id; |
5858 | ||
5859 | old_table_id = ctx->table_id; | |
5860 | ctx->table_id = table_id; | |
abe529af | 5861 | |
54a9cbc9 | 5862 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
5863 | old_in_port = ctx->flow.in_port; |
5864 | ctx->flow.in_port = in_port; | |
c57b2226 | 5865 | rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id); |
54a9cbc9 BP |
5866 | |
5867 | /* Tag the flow. */ | |
5868 | if (table_id > 0 && table_id < N_TABLES) { | |
5869 | struct table_dpif *table = &ofproto->tables[table_id]; | |
5870 | if (table->other_table) { | |
33780682 | 5871 | ctx->tags |= (rule && rule->tag |
54a9cbc9 BP |
5872 | ? rule->tag |
5873 | : rule_calculate_tag(&ctx->flow, | |
5cb7a798 | 5874 | &table->other_table->mask, |
54a9cbc9 BP |
5875 | table->basis)); |
5876 | } | |
5877 | } | |
5878 | ||
5879 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
5880 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
5881 | ctx->flow.in_port = old_in_port; |
5882 | ||
5883 | if (ctx->resubmit_hook) { | |
5884 | ctx->resubmit_hook(ctx, rule); | |
5885 | } | |
5886 | ||
1688c479 | 5887 | if (rule == NULL && may_packet_in) { |
5dca28b5 | 5888 | /* XXX |
1688c479 IY |
5889 | * check if table configuration flags |
5890 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
5891 | * OFPTC_TABLE_MISS_CONTINUE, | |
5892 | * OFPTC_TABLE_MISS_DROP | |
5893 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
5894 | */ | |
5895 | rule = rule_dpif_miss_rule(ofproto, &ctx->flow); | |
5896 | } | |
5897 | ||
abe529af | 5898 | if (rule) { |
18b2a258 | 5899 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 5900 | |
112bc5f4 BP |
5901 | if (ctx->resubmit_stats) { |
5902 | rule_credit_stats(rule, ctx->resubmit_stats); | |
5903 | } | |
5904 | ||
abe529af | 5905 | ctx->recurse++; |
18b2a258 | 5906 | ctx->rule = rule; |
f25d0cf3 | 5907 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 5908 | ctx->rule = old_rule; |
abe529af BP |
5909 | ctx->recurse--; |
5910 | } | |
29901626 BP |
5911 | |
5912 | ctx->table_id = old_table_id; | |
abe529af BP |
5913 | } else { |
5914 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5915 | ||
29901626 | 5916 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 5917 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 5918 | ctx->max_resubmit_trigger = true; |
abe529af BP |
5919 | } |
5920 | } | |
5921 | ||
29901626 | 5922 | static void |
f25d0cf3 BP |
5923 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, |
5924 | const struct ofpact_resubmit *resubmit) | |
29901626 BP |
5925 | { |
5926 | uint16_t in_port; | |
5927 | uint8_t table_id; | |
5928 | ||
f25d0cf3 BP |
5929 | in_port = resubmit->in_port; |
5930 | if (in_port == OFPP_IN_PORT) { | |
5931 | in_port = ctx->flow.in_port; | |
5932 | } | |
5933 | ||
5934 | table_id = resubmit->table_id; | |
5935 | if (table_id == 255) { | |
5936 | table_id = ctx->table_id; | |
5937 | } | |
29901626 | 5938 | |
1688c479 | 5939 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
5940 | } |
5941 | ||
abe529af | 5942 | static void |
d59906fb | 5943 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
5944 | { |
5945 | struct ofport_dpif *ofport; | |
5946 | ||
b3e9b2ed | 5947 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 5948 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
5949 | |
5950 | if (ofp_port == ctx->flow.in_port) { | |
5951 | continue; | |
5952 | } | |
5953 | ||
5e48dc2b | 5954 | if (all) { |
81b1afb1 | 5955 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 5956 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 5957 | compose_output_action(ctx, ofp_port); |
abe529af BP |
5958 | } |
5959 | } | |
b3e9b2ed EJ |
5960 | |
5961 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
5962 | } |
5963 | ||
6ff686f2 | 5964 | static void |
f0fd1a17 | 5965 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
5966 | enum ofp_packet_in_reason reason, |
5967 | uint16_t controller_id) | |
6ff686f2 | 5968 | { |
999fba59 EJ |
5969 | struct ofputil_packet_in pin; |
5970 | struct ofpbuf *packet; | |
6ff686f2 | 5971 | |
6a7e895f | 5972 | ctx->slow |= SLOW_CONTROLLER; |
999fba59 EJ |
5973 | if (!ctx->packet) { |
5974 | return; | |
5975 | } | |
5976 | ||
5977 | packet = ofpbuf_clone(ctx->packet); | |
5978 | ||
5979 | if (packet->l2 && packet->l3) { | |
5980 | struct eth_header *eh; | |
b02475c5 | 5981 | uint16_t mpls_depth; |
999fba59 EJ |
5982 | |
5983 | eth_pop_vlan(packet); | |
5984 | eh = packet->l2; | |
0104aba8 | 5985 | |
999fba59 EJ |
5986 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
5987 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
5988 | ||
5989 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
5990 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
5991 | } | |
5992 | ||
b02475c5 SH |
5993 | mpls_depth = eth_mpls_depth(packet); |
5994 | ||
5995 | if (mpls_depth < ctx->flow.mpls_depth) { | |
5996 | push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse); | |
5997 | } else if (mpls_depth > ctx->flow.mpls_depth) { | |
5998 | pop_mpls(packet, ctx->flow.dl_type); | |
5999 | } else if (mpls_depth) { | |
6000 | set_mpls_lse(packet, ctx->flow.mpls_lse); | |
6001 | } | |
6002 | ||
999fba59 EJ |
6003 | if (packet->l4) { |
6004 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6005 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
6006 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
6007 | } | |
6008 | ||
6009 | if (packet->l7) { | |
6010 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
6011 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
6012 | ctx->flow.tp_dst); | |
6013 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
6014 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
6015 | ctx->flow.tp_dst); | |
6016 | } | |
6017 | } | |
6018 | } | |
6019 | } | |
6020 | ||
6021 | pin.packet = packet->data; | |
6022 | pin.packet_len = packet->size; | |
f0fd1a17 | 6023 | pin.reason = reason; |
a7349929 | 6024 | pin.controller_id = controller_id; |
54834960 | 6025 | pin.table_id = ctx->table_id; |
18b2a258 | 6026 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 6027 | |
999fba59 | 6028 | pin.send_len = len; |
999fba59 EJ |
6029 | flow_get_metadata(&ctx->flow, &pin.fmd); |
6030 | ||
d8653c38 | 6031 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 6032 | ofpbuf_delete(packet); |
6ff686f2 PS |
6033 | } |
6034 | ||
b02475c5 SH |
6035 | static void |
6036 | execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6037 | { | |
6038 | ovs_assert(eth_type_mpls(eth_type)); | |
6039 | ||
6040 | if (ctx->base_flow.mpls_depth) { | |
6041 | ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); | |
6042 | ctx->flow.mpls_depth++; | |
6043 | } else { | |
6044 | ovs_be32 label; | |
6045 | uint8_t tc, ttl; | |
6046 | ||
6047 | if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) { | |
6048 | label = htonl(0x2); /* IPV6 Explicit Null. */ | |
6049 | } else { | |
6050 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6051 | } | |
6052 | tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2; | |
6053 | ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40; | |
6054 | ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
6055 | ctx->flow.encap_dl_type = ctx->flow.dl_type; | |
6056 | ctx->flow.mpls_depth = 1; | |
6057 | } | |
6058 | ctx->flow.dl_type = eth_type; | |
6059 | } | |
6060 | ||
6061 | static void | |
6062 | execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6063 | { | |
6064 | ovs_assert(eth_type_mpls(ctx->flow.dl_type)); | |
6065 | ovs_assert(!eth_type_mpls(eth_type)); | |
6066 | ||
6067 | if (ctx->flow.mpls_depth) { | |
6068 | ctx->flow.mpls_depth--; | |
6069 | ctx->flow.mpls_lse = htonl(0); | |
6070 | if (!ctx->flow.mpls_depth) { | |
6071 | ctx->flow.dl_type = eth_type; | |
6072 | ctx->flow.encap_dl_type = htons(0); | |
6073 | } | |
6074 | } | |
6075 | } | |
6076 | ||
f0fd1a17 | 6077 | static bool |
c2d967a5 | 6078 | compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 PS |
6079 | { |
6080 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
6081 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
6082 | return false; | |
6083 | } | |
6084 | ||
6085 | if (ctx->flow.nw_ttl > 1) { | |
6086 | ctx->flow.nw_ttl--; | |
6087 | return false; | |
6088 | } else { | |
c2d967a5 MM |
6089 | size_t i; |
6090 | ||
6091 | for (i = 0; i < ids->n_controllers; i++) { | |
6092 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6093 | ids->cnt_ids[i]); | |
6094 | } | |
f0fd1a17 PS |
6095 | |
6096 | /* Stop processing for current table. */ | |
6097 | return true; | |
6098 | } | |
6099 | } | |
6100 | ||
abe529af | 6101 | static void |
f25d0cf3 | 6102 | xlate_output_action(struct action_xlate_ctx *ctx, |
1688c479 | 6103 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af BP |
6104 | { |
6105 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
6106 | ||
6107 | ctx->nf_output_iface = NF_OUT_DROP; | |
6108 | ||
6109 | switch (port) { | |
6110 | case OFPP_IN_PORT: | |
81b1afb1 | 6111 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
6112 | break; |
6113 | case OFPP_TABLE: | |
1688c479 | 6114 | xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); |
abe529af BP |
6115 | break; |
6116 | case OFPP_NORMAL: | |
6117 | xlate_normal(ctx); | |
6118 | break; | |
6119 | case OFPP_FLOOD: | |
d59906fb | 6120 | flood_packets(ctx, false); |
abe529af BP |
6121 | break; |
6122 | case OFPP_ALL: | |
d59906fb | 6123 | flood_packets(ctx, true); |
abe529af BP |
6124 | break; |
6125 | case OFPP_CONTROLLER: | |
a7349929 | 6126 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 6127 | break; |
e81d2933 EJ |
6128 | case OFPP_NONE: |
6129 | break; | |
a0fbe94a | 6130 | case OFPP_LOCAL: |
abe529af BP |
6131 | default: |
6132 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 6133 | compose_output_action(ctx, port); |
3dd3eace BP |
6134 | } else { |
6135 | xlate_report(ctx, "skipping output to input port"); | |
abe529af BP |
6136 | } |
6137 | break; | |
6138 | } | |
6139 | ||
6140 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
6141 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6142 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
6143 | ctx->nf_output_iface = prev_nf_output_iface; | |
6144 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
6145 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
6146 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6147 | } | |
6148 | } | |
6149 | ||
f694937d EJ |
6150 | static void |
6151 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6152 | const struct ofpact_output_reg *or) |
f694937d | 6153 | { |
f25d0cf3 BP |
6154 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); |
6155 | if (port <= UINT16_MAX) { | |
1688c479 | 6156 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
6157 | } |
6158 | } | |
6159 | ||
abe529af BP |
6160 | static void |
6161 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6162 | const struct ofpact_enqueue *enqueue) |
abe529af | 6163 | { |
f25d0cf3 BP |
6164 | uint16_t ofp_port = enqueue->port; |
6165 | uint32_t queue_id = enqueue->queue; | |
abff858b | 6166 | uint32_t flow_priority, priority; |
abe529af BP |
6167 | int error; |
6168 | ||
f25d0cf3 | 6169 | /* Translate queue to priority. */ |
acf60855 JP |
6170 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6171 | queue_id, &priority); | |
abe529af BP |
6172 | if (error) { |
6173 | /* Fall back to ordinary output action. */ | |
1688c479 | 6174 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
6175 | return; |
6176 | } | |
6177 | ||
f25d0cf3 | 6178 | /* Check output port. */ |
abe529af BP |
6179 | if (ofp_port == OFPP_IN_PORT) { |
6180 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
6181 | } else if (ofp_port == ctx->flow.in_port) { |
6182 | return; | |
abe529af | 6183 | } |
abe529af | 6184 | |
df2c07f4 | 6185 | /* Add datapath actions. */ |
deedf7e7 BP |
6186 | flow_priority = ctx->flow.skb_priority; |
6187 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 6188 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 6189 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
6190 | |
6191 | /* Update NetFlow output port. */ | |
6192 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 6193 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
6194 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
6195 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6196 | } | |
6197 | } | |
6198 | ||
6199 | static void | |
f25d0cf3 | 6200 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 6201 | { |
f25d0cf3 | 6202 | uint32_t skb_priority; |
abe529af | 6203 | |
acf60855 JP |
6204 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6205 | queue_id, &skb_priority)) { | |
f25d0cf3 BP |
6206 | ctx->flow.skb_priority = skb_priority; |
6207 | } else { | |
6208 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 6209 | * has already been logged. */ |
abe529af | 6210 | } |
abe529af BP |
6211 | } |
6212 | ||
6213 | struct xlate_reg_state { | |
6214 | ovs_be16 vlan_tci; | |
6215 | ovs_be64 tun_id; | |
6216 | }; | |
6217 | ||
daff3353 EJ |
6218 | static bool |
6219 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6220 | { | |
6221 | struct ofproto_dpif *ofproto = ofproto_; | |
6222 | struct ofport_dpif *port; | |
6223 | ||
6224 | switch (ofp_port) { | |
6225 | case OFPP_IN_PORT: | |
6226 | case OFPP_TABLE: | |
6227 | case OFPP_NORMAL: | |
6228 | case OFPP_FLOOD: | |
6229 | case OFPP_ALL: | |
439e4d8c | 6230 | case OFPP_NONE: |
daff3353 EJ |
6231 | return true; |
6232 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6233 | return false; | |
6234 | default: | |
6235 | port = get_ofp_port(ofproto, ofp_port); | |
6236 | return port ? port->may_enable : false; | |
6237 | } | |
6238 | } | |
6239 | ||
f25d0cf3 BP |
6240 | static void |
6241 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
6242 | const struct ofpact_bundle *bundle) | |
6243 | { | |
6244 | uint16_t port; | |
6245 | ||
6246 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
6247 | if (bundle->dst.field) { | |
6248 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
6249 | } else { | |
1688c479 | 6250 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
6251 | } |
6252 | } | |
6253 | ||
75a75043 BP |
6254 | static void |
6255 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6256 | const struct ofpact_learn *learn) |
75a75043 BP |
6257 | { |
6258 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6259 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
6260 | uint64_t ofpacts_stub[1024 / 8]; |
6261 | struct ofpbuf ofpacts; | |
75a75043 BP |
6262 | int error; |
6263 | ||
f25d0cf3 BP |
6264 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
6265 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
75a75043 BP |
6266 | |
6267 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6268 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
6269 | VLOG_WARN("learning action failed to modify flow table (%s)", |
6270 | ofperr_get_name(error)); | |
75a75043 BP |
6271 | } |
6272 | ||
f25d0cf3 | 6273 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
6274 | } |
6275 | ||
0e553d9c BP |
6276 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
6277 | * means "infinite". */ | |
6278 | static void | |
6279 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6280 | { | |
6281 | if (max && (!*timeout || *timeout > max)) { | |
6282 | *timeout = max; | |
6283 | } | |
6284 | } | |
6285 | ||
6286 | static void | |
6287 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6288 | const struct ofpact_fin_timeout *oft) |
0e553d9c BP |
6289 | { |
6290 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
6291 | struct rule_dpif *rule = ctx->rule; | |
6292 | ||
f25d0cf3 BP |
6293 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
6294 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
6295 | } |
6296 | } | |
6297 | ||
21f7563c JP |
6298 | static bool |
6299 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
6300 | { | |
9e1fd49b BP |
6301 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
6302 | ? OFPUTIL_PC_NO_RECV_STP | |
6303 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
6304 | return false; |
6305 | } | |
6306 | ||
6307 | /* Only drop packets here if both forwarding and learning are | |
6308 | * disabled. If just learning is enabled, we need to have | |
6309 | * OFPP_NORMAL and the learning action have a look at the packet | |
6310 | * before we can drop it. */ | |
6311 | if (!stp_forward_in_state(port->stp_state) | |
6312 | && !stp_learn_in_state(port->stp_state)) { | |
6313 | return false; | |
6314 | } | |
6315 | ||
6316 | return true; | |
6317 | } | |
6318 | ||
abe529af | 6319 | static void |
f25d0cf3 | 6320 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
abe529af BP |
6321 | struct action_xlate_ctx *ctx) |
6322 | { | |
254750ce | 6323 | bool was_evictable = true; |
f25d0cf3 | 6324 | const struct ofpact *a; |
abe529af | 6325 | |
254750ce BP |
6326 | if (ctx->rule) { |
6327 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6328 | was_evictable = ctx->rule->up.evictable; | |
6329 | ctx->rule->up.evictable = false; | |
6330 | } | |
f25d0cf3 BP |
6331 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
6332 | struct ofpact_controller *controller; | |
4cceacb9 | 6333 | const struct ofpact_metadata *metadata; |
38f2e360 | 6334 | |
848e8809 EJ |
6335 | if (ctx->exit) { |
6336 | break; | |
6337 | } | |
6338 | ||
f25d0cf3 BP |
6339 | switch (a->type) { |
6340 | case OFPACT_OUTPUT: | |
6341 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 6342 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
6343 | break; |
6344 | ||
6345 | case OFPACT_CONTROLLER: | |
6346 | controller = ofpact_get_CONTROLLER(a); | |
6347 | execute_controller_action(ctx, controller->max_len, | |
6348 | controller->reason, | |
6349 | controller->controller_id); | |
6350 | break; | |
690a61c5 | 6351 | |
f25d0cf3 BP |
6352 | case OFPACT_ENQUEUE: |
6353 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
6354 | break; |
6355 | ||
f25d0cf3 | 6356 | case OFPACT_SET_VLAN_VID: |
abe529af | 6357 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
f25d0cf3 BP |
6358 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) |
6359 | | htons(VLAN_CFI)); | |
abe529af BP |
6360 | break; |
6361 | ||
f25d0cf3 | 6362 | case OFPACT_SET_VLAN_PCP: |
abe529af | 6363 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
f25d0cf3 BP |
6364 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp |
6365 | << VLAN_PCP_SHIFT) | |
6366 | | VLAN_CFI); | |
abe529af BP |
6367 | break; |
6368 | ||
f25d0cf3 | 6369 | case OFPACT_STRIP_VLAN: |
abe529af | 6370 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
6371 | break; |
6372 | ||
3e34fbdd | 6373 | case OFPACT_PUSH_VLAN: |
5dca28b5 | 6374 | /* XXX 802.1AD(QinQ) */ |
3e34fbdd IY |
6375 | ctx->flow.vlan_tci = htons(VLAN_CFI); |
6376 | break; | |
6377 | ||
f25d0cf3 BP |
6378 | case OFPACT_SET_ETH_SRC: |
6379 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
6380 | ETH_ADDR_LEN); | |
abe529af BP |
6381 | break; |
6382 | ||
f25d0cf3 BP |
6383 | case OFPACT_SET_ETH_DST: |
6384 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
6385 | ETH_ADDR_LEN); | |
abe529af BP |
6386 | break; |
6387 | ||
f25d0cf3 BP |
6388 | case OFPACT_SET_IPV4_SRC: |
6389 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
abe529af BP |
6390 | break; |
6391 | ||
f25d0cf3 BP |
6392 | case OFPACT_SET_IPV4_DST: |
6393 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
abe529af BP |
6394 | break; |
6395 | ||
f25d0cf3 | 6396 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d PS |
6397 | /* OpenFlow 1.0 only supports IPv4. */ |
6398 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6399 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
f25d0cf3 | 6400 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; |
c4f2731d | 6401 | } |
abe529af BP |
6402 | break; |
6403 | ||
f25d0cf3 BP |
6404 | case OFPACT_SET_L4_SRC_PORT: |
6405 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
abe529af BP |
6406 | break; |
6407 | ||
f25d0cf3 BP |
6408 | case OFPACT_SET_L4_DST_PORT: |
6409 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
abe529af BP |
6410 | break; |
6411 | ||
f25d0cf3 BP |
6412 | case OFPACT_RESUBMIT: |
6413 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
6414 | break; |
6415 | ||
f25d0cf3 | 6416 | case OFPACT_SET_TUNNEL: |
296e07ac | 6417 | ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); |
29901626 BP |
6418 | break; |
6419 | ||
f25d0cf3 BP |
6420 | case OFPACT_SET_QUEUE: |
6421 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
6422 | break; |
6423 | ||
f25d0cf3 | 6424 | case OFPACT_POP_QUEUE: |
deedf7e7 | 6425 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
6426 | break; |
6427 | ||
f25d0cf3 BP |
6428 | case OFPACT_REG_MOVE: |
6429 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
38f2e360 BP |
6430 | break; |
6431 | ||
f25d0cf3 BP |
6432 | case OFPACT_REG_LOAD: |
6433 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
38f2e360 BP |
6434 | break; |
6435 | ||
b02475c5 SH |
6436 | case OFPACT_PUSH_MPLS: |
6437 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6438 | break; | |
6439 | ||
6440 | case OFPACT_POP_MPLS: | |
6441 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6442 | break; | |
6443 | ||
f25d0cf3 | 6444 | case OFPACT_DEC_TTL: |
c2d967a5 | 6445 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
6446 | goto out; |
6447 | } | |
38f2e360 BP |
6448 | break; |
6449 | ||
f25d0cf3 BP |
6450 | case OFPACT_NOTE: |
6451 | /* Nothing to do. */ | |
abe529af BP |
6452 | break; |
6453 | ||
f25d0cf3 BP |
6454 | case OFPACT_MULTIPATH: |
6455 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
abe529af | 6456 | break; |
daff3353 | 6457 | |
f25d0cf3 | 6458 | case OFPACT_BUNDLE: |
a368bb53 | 6459 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 6460 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 6461 | break; |
f694937d | 6462 | |
f25d0cf3 BP |
6463 | case OFPACT_OUTPUT_REG: |
6464 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 6465 | break; |
75a75043 | 6466 | |
f25d0cf3 | 6467 | case OFPACT_LEARN: |
75a75043 | 6468 | ctx->has_learn = true; |
3de9590b | 6469 | if (ctx->may_learn) { |
f25d0cf3 | 6470 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
6471 | } |
6472 | break; | |
848e8809 | 6473 | |
f25d0cf3 | 6474 | case OFPACT_EXIT: |
848e8809 EJ |
6475 | ctx->exit = true; |
6476 | break; | |
0e553d9c | 6477 | |
f25d0cf3 | 6478 | case OFPACT_FIN_TIMEOUT: |
0e553d9c | 6479 | ctx->has_fin_timeout = true; |
f25d0cf3 | 6480 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 6481 | break; |
8dd54666 | 6482 | |
b19e8793 | 6483 | case OFPACT_CLEAR_ACTIONS: |
5dca28b5 | 6484 | /* XXX |
b19e8793 IY |
6485 | * Nothing to do because writa-actions is not supported for now. |
6486 | * When writa-actions is supported, clear-actions also must | |
6487 | * be supported at the same time. | |
6488 | */ | |
6489 | break; | |
6490 | ||
4cceacb9 JS |
6491 | case OFPACT_WRITE_METADATA: |
6492 | metadata = ofpact_get_WRITE_METADATA(a); | |
6493 | ctx->flow.metadata &= ~metadata->mask; | |
6494 | ctx->flow.metadata |= metadata->metadata & metadata->mask; | |
6495 | break; | |
6496 | ||
8dd54666 | 6497 | case OFPACT_GOTO_TABLE: { |
5dca28b5 | 6498 | /* XXX remove recursion */ |
8dd54666 IY |
6499 | /* It is assumed that goto-table is last action */ |
6500 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); | |
cb22974d | 6501 | ovs_assert(ctx->table_id < ogt->table_id); |
8dd54666 IY |
6502 | xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true); |
6503 | break; | |
6504 | } | |
abe529af BP |
6505 | } |
6506 | } | |
21f7563c | 6507 | |
f0fd1a17 | 6508 | out: |
254750ce BP |
6509 | if (ctx->rule) { |
6510 | ctx->rule->up.evictable = was_evictable; | |
6511 | } | |
abe529af BP |
6512 | } |
6513 | ||
6514 | static void | |
6515 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
6516 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
18b2a258 | 6517 | ovs_be16 initial_tci, struct rule_dpif *rule, |
0e553d9c | 6518 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af | 6519 | { |
ef506a7c JG |
6520 | ovs_be64 initial_tun_id = flow->tunnel.tun_id; |
6521 | ||
6522 | /* Flow initialization rules: | |
6523 | * - 'base_flow' must match the kernel's view of the packet at the | |
6524 | * time that action processing starts. 'flow' represents any | |
6525 | * transformations we wish to make through actions. | |
6526 | * - By default 'base_flow' and 'flow' are the same since the input | |
6527 | * packet matches the output before any actions are applied. | |
6528 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
6529 | * of the received packet as seen by the kernel. If we later output | |
6530 | * to another device without any modifications this will cause us to | |
6531 | * insert a new tag since the original one was stripped off by the | |
6532 | * VLAN device. | |
6533 | * - Tunnel 'flow' is largely cleared when transitioning between | |
6534 | * the input and output stages since it does not make sense to output | |
6535 | * a packet with the exact headers that it was received with (i.e. | |
6536 | * the destination IP is us). The one exception is the tun_id, which | |
6537 | * is preserved to allow use in later resubmit lookups and loads into | |
6538 | * registers. | |
6539 | * - Tunnel 'base_flow' is completely cleared since that is what the | |
6540 | * kernel does. If we wish to maintain the original values an action | |
6541 | * needs to be generated. */ | |
6542 | ||
abe529af BP |
6543 | ctx->ofproto = ofproto; |
6544 | ctx->flow = *flow; | |
47d4a9db | 6545 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); |
e84173dc | 6546 | ctx->base_flow = ctx->flow; |
e84173dc | 6547 | ctx->base_flow.vlan_tci = initial_tci; |
ef506a7c | 6548 | ctx->flow.tunnel.tun_id = initial_tun_id; |
18b2a258 | 6549 | ctx->rule = rule; |
abe529af | 6550 | ctx->packet = packet; |
3de9590b | 6551 | ctx->may_learn = packet != NULL; |
0e553d9c | 6552 | ctx->tcp_flags = tcp_flags; |
abe529af | 6553 | ctx->resubmit_hook = NULL; |
479df176 | 6554 | ctx->report_hook = NULL; |
112bc5f4 | 6555 | ctx->resubmit_stats = NULL; |
abe529af BP |
6556 | } |
6557 | ||
f25d0cf3 BP |
6558 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' |
6559 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
050ac423 | 6560 | static void |
abe529af | 6561 | xlate_actions(struct action_xlate_ctx *ctx, |
f25d0cf3 | 6562 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 | 6563 | struct ofpbuf *odp_actions) |
abe529af | 6564 | { |
43d50bc8 BP |
6565 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so |
6566 | * that in the future we always keep a copy of the original flow for | |
6567 | * tracing purposes. */ | |
6568 | static bool hit_resubmit_limit; | |
6569 | ||
6a7e895f | 6570 | enum slow_path_reason special; |
ffaef958 | 6571 | struct ofport_dpif *in_port; |
9ba85077 | 6572 | struct flow orig_flow; |
6a7e895f | 6573 | |
abe529af BP |
6574 | COVERAGE_INC(ofproto_dpif_xlate); |
6575 | ||
050ac423 BP |
6576 | ofpbuf_clear(odp_actions); |
6577 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
6578 | ||
6579 | ctx->odp_actions = odp_actions; | |
97e42c92 | 6580 | ctx->tags = 0; |
6a7e895f | 6581 | ctx->slow = 0; |
97e42c92 BP |
6582 | ctx->has_learn = false; |
6583 | ctx->has_normal = false; | |
0e553d9c | 6584 | ctx->has_fin_timeout = false; |
97e42c92 | 6585 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 6586 | ctx->mirrors = 0; |
97e42c92 | 6587 | ctx->recurse = 0; |
6a6455e5 | 6588 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 6589 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 6590 | ctx->table_id = 0; |
848e8809 | 6591 | ctx->exit = false; |
7257b535 | 6592 | |
43d50bc8 | 6593 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 | 6594 | /* Do this conditionally because the copy is expensive enough that it |
9ba85077 BP |
6595 | * shows up in profiles. */ |
6596 | orig_flow = ctx->flow; | |
ccb7c863 BP |
6597 | } |
6598 | ||
eadef313 | 6599 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
6600 | switch (ctx->ofproto->up.frag_handling) { |
6601 | case OFPC_FRAG_NORMAL: | |
6602 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
6603 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
6604 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
6605 | break; |
6606 | ||
6607 | case OFPC_FRAG_DROP: | |
050ac423 | 6608 | return; |
7257b535 BP |
6609 | |
6610 | case OFPC_FRAG_REASM: | |
6611 | NOT_REACHED(); | |
6612 | ||
6613 | case OFPC_FRAG_NX_MATCH: | |
6614 | /* Nothing to do. */ | |
6615 | break; | |
f0fd1a17 PS |
6616 | |
6617 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
6618 | NOT_REACHED(); | |
7257b535 BP |
6619 | } |
6620 | } | |
6621 | ||
ffaef958 BP |
6622 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
6623 | special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet); | |
6a7e895f BP |
6624 | if (special) { |
6625 | ctx->slow |= special; | |
abe529af | 6626 | } else { |
6a6455e5 | 6627 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
6a6455e5 | 6628 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; |
ee382d89 | 6629 | uint32_t local_odp_port; |
6a6455e5 | 6630 | |
6ff686f2 | 6631 | add_sflow_action(ctx); |
ffaef958 BP |
6632 | |
6633 | if (!in_port || may_receive(in_port, ctx)) { | |
6634 | do_xlate_actions(ofpacts, ofpacts_len, ctx); | |
6635 | ||
6636 | /* We've let OFPP_NORMAL and the learning action look at the | |
6637 | * packet, so drop it now if forwarding is disabled. */ | |
6638 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
6639 | ofpbuf_clear(ctx->odp_actions); | |
6640 | add_sflow_action(ctx); | |
6641 | } | |
6642 | } | |
abe529af | 6643 | |
43d50bc8 BP |
6644 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { |
6645 | if (!hit_resubmit_limit) { | |
6646 | /* We didn't record the original flow. Make sure we do from | |
6647 | * now on. */ | |
6648 | hit_resubmit_limit = true; | |
6649 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
6650 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6651 | ||
9ba85077 | 6652 | ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet, |
43d50bc8 BP |
6653 | initial_tci, &ds); |
6654 | VLOG_ERR("Trace triggered by excessive resubmit " | |
6655 | "recursion:\n%s", ds_cstr(&ds)); | |
6656 | ds_destroy(&ds); | |
6657 | } | |
6a6455e5 EJ |
6658 | } |
6659 | ||
ee382d89 | 6660 | local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); |
b6848f13 | 6661 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
ee382d89 | 6662 | local_odp_port, |
b6848f13 BP |
6663 | ctx->odp_actions->data, |
6664 | ctx->odp_actions->size)) { | |
6a7e895f | 6665 | ctx->slow |= SLOW_IN_BAND; |
b6848f13 BP |
6666 | if (ctx->packet |
6667 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
6668 | ctx->packet)) { | |
5e48dc2b | 6669 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
6670 | } |
6671 | } | |
ccb7c863 | 6672 | if (ctx->ofproto->has_mirrors) { |
9ba85077 | 6673 | add_mirror_actions(ctx, &orig_flow); |
ccb7c863 | 6674 | } |
a7c4eaf6 | 6675 | fix_sflow_action(ctx); |
abe529af | 6676 | } |
050ac423 BP |
6677 | } |
6678 | ||
f25d0cf3 BP |
6679 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
6680 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 BP |
6681 | static void |
6682 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
f25d0cf3 BP |
6683 | const struct ofpact *ofpacts, |
6684 | size_t ofpacts_len) | |
050ac423 BP |
6685 | { |
6686 | uint64_t odp_actions_stub[1024 / 8]; | |
6687 | struct ofpbuf odp_actions; | |
abe529af | 6688 | |
050ac423 | 6689 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
f25d0cf3 | 6690 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); |
050ac423 | 6691 | ofpbuf_uninit(&odp_actions); |
abe529af | 6692 | } |
479df176 BP |
6693 | |
6694 | static void | |
6695 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
6696 | { | |
6697 | if (ctx->report_hook) { | |
6698 | ctx->report_hook(ctx, s); | |
6699 | } | |
6700 | } | |
abe529af BP |
6701 | \f |
6702 | /* OFPP_NORMAL implementation. */ | |
6703 | ||
abe529af BP |
6704 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
6705 | ||
ecac4ebf BP |
6706 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
6707 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
6708 | * the bundle on which the packet was received, returns the VLAN to which the | |
6709 | * packet belongs. | |
6710 | * | |
6711 | * Both 'vid' and the return value are in the range 0...4095. */ | |
6712 | static uint16_t | |
6713 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
6714 | { | |
6715 | switch (in_bundle->vlan_mode) { | |
6716 | case PORT_VLAN_ACCESS: | |
6717 | return in_bundle->vlan; | |
6718 | break; | |
6719 | ||
6720 | case PORT_VLAN_TRUNK: | |
6721 | return vid; | |
6722 | ||
6723 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6724 | case PORT_VLAN_NATIVE_TAGGED: | |
6725 | return vid ? vid : in_bundle->vlan; | |
6726 | ||
6727 | default: | |
6728 | NOT_REACHED(); | |
6729 | } | |
6730 | } | |
6731 | ||
5da5ec37 BP |
6732 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
6733 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
6734 | * a warning. | |
6735 | * | |
6736 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
6737 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
6738 | * 0...4095. */ | |
6739 | static bool | |
6740 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
6741 | { | |
33158a18 JP |
6742 | /* Allow any VID on the OFPP_NONE port. */ |
6743 | if (in_bundle == &ofpp_none_bundle) { | |
6744 | return true; | |
6745 | } | |
6746 | ||
5da5ec37 BP |
6747 | switch (in_bundle->vlan_mode) { |
6748 | case PORT_VLAN_ACCESS: | |
6749 | if (vid) { | |
6750 | if (warn) { | |
6751 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6752 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
6753 | "packet received on port %s configured as VLAN " | |
6754 | "%"PRIu16" access port", | |
6755 | in_bundle->ofproto->up.name, vid, | |
6756 | in_bundle->name, in_bundle->vlan); | |
6757 | } | |
6758 | return false; | |
6759 | } | |
6760 | return true; | |
6761 | ||
6762 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6763 | case PORT_VLAN_NATIVE_TAGGED: | |
6764 | if (!vid) { | |
6765 | /* Port must always carry its native VLAN. */ | |
6766 | return true; | |
6767 | } | |
6768 | /* Fall through. */ | |
6769 | case PORT_VLAN_TRUNK: | |
6770 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
6771 | if (warn) { | |
6772 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6773 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
6774 | "received on port %s not configured for trunking " | |
6775 | "VLAN %"PRIu16, | |
6776 | in_bundle->ofproto->up.name, vid, | |
6777 | in_bundle->name, vid); | |
6778 | } | |
6779 | return false; | |
6780 | } | |
6781 | return true; | |
6782 | ||
6783 | default: | |
6784 | NOT_REACHED(); | |
6785 | } | |
6786 | ||
6787 | } | |
6788 | ||
ecac4ebf BP |
6789 | /* Given 'vlan', the VLAN that a packet belongs to, and |
6790 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
6791 | * that should be included in the 802.1Q header. (If the return value is 0, | |
6792 | * then the 802.1Q header should only be included in the packet if there is a | |
6793 | * nonzero PCP.) | |
6794 | * | |
6795 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
6796 | static uint16_t | |
6797 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
6798 | { | |
6799 | switch (out_bundle->vlan_mode) { | |
6800 | case PORT_VLAN_ACCESS: | |
6801 | return 0; | |
6802 | ||
6803 | case PORT_VLAN_TRUNK: | |
6804 | case PORT_VLAN_NATIVE_TAGGED: | |
6805 | return vlan; | |
6806 | ||
6807 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6808 | return vlan == out_bundle->vlan ? 0 : vlan; | |
6809 | ||
6810 | default: | |
6811 | NOT_REACHED(); | |
6812 | } | |
6813 | } | |
6814 | ||
395e68ce BP |
6815 | static void |
6816 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
6817 | uint16_t vlan) | |
abe529af | 6818 | { |
395e68ce BP |
6819 | struct ofport_dpif *port; |
6820 | uint16_t vid; | |
81b1afb1 | 6821 | ovs_be16 tci, old_tci; |
ecac4ebf | 6822 | |
395e68ce BP |
6823 | vid = output_vlan_to_vid(out_bundle, vlan); |
6824 | if (!out_bundle->bond) { | |
6825 | port = ofbundle_get_a_port(out_bundle); | |
6826 | } else { | |
6827 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
6828 | vid, &ctx->tags); | |
6829 | if (!port) { | |
6830 | /* No slaves enabled, so drop packet. */ | |
6831 | return; | |
6832 | } | |
6833 | } | |
abe529af | 6834 | |
81b1afb1 | 6835 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
6836 | tci = htons(vid); |
6837 | if (tci || out_bundle->use_priority_tags) { | |
6838 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
6839 | if (tci) { | |
6840 | tci |= htons(VLAN_CFI); | |
6841 | } | |
395e68ce | 6842 | } |
81b1afb1 | 6843 | ctx->flow.vlan_tci = tci; |
395e68ce | 6844 | |
5e48dc2b | 6845 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 6846 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
6847 | } |
6848 | ||
6849 | static int | |
6850 | mirror_mask_ffs(mirror_mask_t mask) | |
6851 | { | |
6852 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
6853 | return ffs(mask); | |
6854 | } | |
6855 | ||
abe529af BP |
6856 | static bool |
6857 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
6858 | { | |
ecac4ebf | 6859 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 6860 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
6861 | } |
6862 | ||
6863 | static bool | |
6864 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
6865 | { | |
6866 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
6867 | } | |
6868 | ||
6869 | /* Returns an arbitrary interface within 'bundle'. */ | |
6870 | static struct ofport_dpif * | |
6871 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
6872 | { | |
6873 | return CONTAINER_OF(list_front(&bundle->ports), | |
6874 | struct ofport_dpif, bundle_node); | |
6875 | } | |
6876 | ||
abe529af BP |
6877 | static bool |
6878 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
6879 | { | |
fc3d7408 | 6880 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
6881 | } |
6882 | ||
6883 | static void | |
c06bba01 | 6884 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
6885 | { |
6886 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6887 | mirror_mask_t mirrors; | |
c06bba01 JP |
6888 | struct ofbundle *in_bundle; |
6889 | uint16_t vlan; | |
6890 | uint16_t vid; | |
6891 | const struct nlattr *a; | |
6892 | size_t left; | |
6893 | ||
3581c12c | 6894 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
70c2fd56 | 6895 | ctx->packet != NULL, NULL); |
3581c12c | 6896 | if (!in_bundle) { |
c06bba01 JP |
6897 | return; |
6898 | } | |
c06bba01 JP |
6899 | mirrors = in_bundle->src_mirrors; |
6900 | ||
6901 | /* Drop frames on bundles reserved for mirroring. */ | |
6902 | if (in_bundle->mirror_out) { | |
6903 | if (ctx->packet != NULL) { | |
6904 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6905 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
6906 | "%s, which is reserved exclusively for mirroring", | |
6907 | ctx->ofproto->up.name, in_bundle->name); | |
6908 | } | |
6909 | return; | |
6910 | } | |
6911 | ||
6912 | /* Check VLAN. */ | |
6913 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
6914 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
6915 | return; | |
6916 | } | |
6917 | vlan = input_vid_to_vlan(in_bundle, vid); | |
6918 | ||
6919 | /* Look at the output ports to check for destination selections. */ | |
6920 | ||
6921 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
6922 | ctx->odp_actions->size) { | |
6923 | enum ovs_action_attr type = nl_attr_type(a); | |
6924 | struct ofport_dpif *ofport; | |
6925 | ||
6926 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
6927 | continue; | |
6928 | } | |
6929 | ||
6930 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
6931 | if (ofport && ofport->bundle) { |
6932 | mirrors |= ofport->bundle->dst_mirrors; | |
6933 | } | |
c06bba01 | 6934 | } |
abe529af BP |
6935 | |
6936 | if (!mirrors) { | |
6937 | return; | |
6938 | } | |
6939 | ||
c06bba01 JP |
6940 | /* Restore the original packet before adding the mirror actions. */ |
6941 | ctx->flow = *orig_flow; | |
6942 | ||
9ba15e2a BP |
6943 | while (mirrors) { |
6944 | struct ofmirror *m; | |
9ba15e2a BP |
6945 | |
6946 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
6947 | ||
6948 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 6949 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
6950 | continue; |
6951 | } | |
6952 | ||
6953 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 6954 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 6955 | if (m->out) { |
395e68ce | 6956 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
6957 | } else if (vlan != m->out_vlan |
6958 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
6959 | struct ofbundle *bundle; |
6960 | ||
6961 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
6962 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
6963 | && !bundle->mirror_out) { |
6964 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
6965 | } |
6966 | } | |
6967 | } | |
abe529af BP |
6968 | } |
6969 | } | |
6970 | ||
9d24de3b JP |
6971 | static void |
6972 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
6973 | uint64_t packets, uint64_t bytes) | |
6974 | { | |
6975 | if (!mirrors) { | |
6976 | return; | |
6977 | } | |
6978 | ||
8472a3ce | 6979 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
6980 | struct ofmirror *m; |
6981 | ||
6982 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
6983 | ||
6984 | if (!m) { | |
6985 | /* In normal circumstances 'm' will not be NULL. However, | |
6986 | * if mirrors are reconfigured, we can temporarily get out | |
6987 | * of sync in facet_revalidate(). We could "correct" the | |
6988 | * mirror list before reaching here, but doing that would | |
6989 | * not properly account the traffic stats we've currently | |
6990 | * accumulated for previous mirror configuration. */ | |
6991 | continue; | |
6992 | } | |
6993 | ||
6994 | m->packet_count += packets; | |
6995 | m->byte_count += bytes; | |
6996 | } | |
6997 | } | |
6998 | ||
abe529af BP |
6999 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
7000 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7001 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7002 | static bool | |
7003 | is_gratuitous_arp(const struct flow *flow) | |
7004 | { | |
7005 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7006 | && eth_addr_is_broadcast(flow->dl_dst) | |
7007 | && (flow->nw_proto == ARP_OP_REPLY | |
7008 | || (flow->nw_proto == ARP_OP_REQUEST | |
7009 | && flow->nw_src == flow->nw_dst))); | |
7010 | } | |
7011 | ||
7012 | static void | |
7013 | update_learning_table(struct ofproto_dpif *ofproto, | |
7014 | const struct flow *flow, int vlan, | |
7015 | struct ofbundle *in_bundle) | |
7016 | { | |
7017 | struct mac_entry *mac; | |
7018 | ||
33158a18 JP |
7019 | /* Don't learn the OFPP_NONE port. */ |
7020 | if (in_bundle == &ofpp_none_bundle) { | |
7021 | return; | |
7022 | } | |
7023 | ||
abe529af BP |
7024 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
7025 | return; | |
7026 | } | |
7027 | ||
7028 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7029 | if (is_gratuitous_arp(flow)) { | |
7030 | /* We don't want to learn from gratuitous ARP packets that are | |
7031 | * reflected back over bond slaves so we lock the learning table. */ | |
7032 | if (!in_bundle->bond) { | |
7033 | mac_entry_set_grat_arp_lock(mac); | |
7034 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7035 | return; | |
7036 | } | |
7037 | } | |
7038 | ||
7039 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7040 | /* The log messages here could actually be useful in debugging, | |
7041 | * so keep the rate limit relatively high. */ | |
7042 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7043 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7044 | "on port %s in VLAN %d", | |
7045 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7046 | in_bundle->name, vlan); | |
7047 | ||
7048 | mac->port.p = in_bundle; | |
2cc3c58e | 7049 | tag_set_add(&ofproto->backer->revalidate_set, |
abe529af BP |
7050 | mac_learning_changed(ofproto->ml, mac)); |
7051 | } | |
7052 | } | |
7053 | ||
3581c12c | 7054 | static struct ofbundle * |
4acbc98d SH |
7055 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
7056 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
7057 | { |
7058 | struct ofport_dpif *ofport; | |
7059 | ||
7060 | /* Find the port and bundle for the received packet. */ | |
7061 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
7062 | if (in_ofportp) { |
7063 | *in_ofportp = ofport; | |
7064 | } | |
395e68ce | 7065 | if (ofport && ofport->bundle) { |
3581c12c | 7066 | return ofport->bundle; |
395e68ce BP |
7067 | } |
7068 | ||
70c2fd56 BP |
7069 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
7070 | * port for traffic that it is sourcing. */ | |
7071 | if (in_port == OFPP_NONE) { | |
7072 | return &ofpp_none_bundle; | |
7073 | } | |
7074 | ||
395e68ce BP |
7075 | /* Odd. A few possible reasons here: |
7076 | * | |
7077 | * - We deleted a port but there are still a few packets queued up | |
7078 | * from it. | |
7079 | * | |
7080 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7081 | * we don't know about. | |
7082 | * | |
7083 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
7084 | * This is particularly likely to happen if a packet was received on the |
7085 | * port after it was created, but before the client had a chance to | |
7086 | * configure its bundle. | |
395e68ce BP |
7087 | */ |
7088 | if (warn) { | |
7089 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7090 | ||
7091 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7092 | "port %"PRIu16, ofproto->up.name, in_port); | |
7093 | } | |
7094 | return NULL; | |
7095 | } | |
7096 | ||
5da5ec37 | 7097 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
7098 | * dropped. Returns true if they may be forwarded, false if they should be |
7099 | * dropped. | |
7100 | * | |
395e68ce BP |
7101 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
7102 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 7103 | * |
395e68ce BP |
7104 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
7105 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7106 | * checked by input_vid_is_valid(). | |
abe529af BP |
7107 | * |
7108 | * May also add tags to '*tags', although the current implementation only does | |
7109 | * so in one special case. | |
7110 | */ | |
7111 | static bool | |
479df176 BP |
7112 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, |
7113 | uint16_t vlan) | |
abe529af | 7114 | { |
479df176 BP |
7115 | struct ofproto_dpif *ofproto = ctx->ofproto; |
7116 | struct flow *flow = &ctx->flow; | |
395e68ce | 7117 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 7118 | |
395e68ce BP |
7119 | /* Drop frames for reserved multicast addresses |
7120 | * only if forward_bpdu option is absent. */ | |
614ec445 | 7121 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 7122 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
7123 | return false; |
7124 | } | |
7125 | ||
abe529af BP |
7126 | if (in_bundle->bond) { |
7127 | struct mac_entry *mac; | |
7128 | ||
7129 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
479df176 | 7130 | flow->dl_dst, &ctx->tags)) { |
abe529af BP |
7131 | case BV_ACCEPT: |
7132 | break; | |
7133 | ||
7134 | case BV_DROP: | |
479df176 | 7135 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
7136 | return false; |
7137 | ||
7138 | case BV_DROP_IF_MOVED: | |
7139 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7140 | if (mac && mac->port.p != in_bundle && | |
7141 | (!is_gratuitous_arp(flow) | |
7142 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
7143 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
7144 | "dropping"); | |
abe529af BP |
7145 | return false; |
7146 | } | |
7147 | break; | |
7148 | } | |
7149 | } | |
7150 | ||
7151 | return true; | |
7152 | } | |
7153 | ||
4cd78906 | 7154 | static void |
abe529af BP |
7155 | xlate_normal(struct action_xlate_ctx *ctx) |
7156 | { | |
395e68ce | 7157 | struct ofport_dpif *in_port; |
abe529af | 7158 | struct ofbundle *in_bundle; |
abe529af | 7159 | struct mac_entry *mac; |
395e68ce BP |
7160 | uint16_t vlan; |
7161 | uint16_t vid; | |
abe529af | 7162 | |
75a75043 BP |
7163 | ctx->has_normal = true; |
7164 | ||
3581c12c | 7165 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
70c2fd56 | 7166 | ctx->packet != NULL, &in_port); |
3581c12c | 7167 | if (!in_bundle) { |
479df176 | 7168 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
7169 | return; |
7170 | } | |
3581c12c | 7171 | |
395e68ce BP |
7172 | /* Drop malformed frames. */ |
7173 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
7174 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
7175 | if (ctx->packet != NULL) { | |
7176 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7177 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7178 | "VLAN tag received on port %s", | |
7179 | ctx->ofproto->up.name, in_bundle->name); | |
7180 | } | |
479df176 | 7181 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
7182 | return; |
7183 | } | |
7184 | ||
7185 | /* Drop frames on bundles reserved for mirroring. */ | |
7186 | if (in_bundle->mirror_out) { | |
7187 | if (ctx->packet != NULL) { | |
7188 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7189 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7190 | "%s, which is reserved exclusively for mirroring", | |
7191 | ctx->ofproto->up.name, in_bundle->name); | |
7192 | } | |
479df176 | 7193 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
7194 | return; |
7195 | } | |
7196 | ||
7197 | /* Check VLAN. */ | |
7198 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
7199 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
479df176 | 7200 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
7201 | return; |
7202 | } | |
7203 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7204 | ||
7205 | /* Check other admissibility requirements. */ | |
479df176 | 7206 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 7207 | return; |
abe529af BP |
7208 | } |
7209 | ||
75a75043 | 7210 | /* Learn source MAC. */ |
3de9590b | 7211 | if (ctx->may_learn) { |
abe529af BP |
7212 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
7213 | } | |
7214 | ||
7215 | /* Determine output bundle. */ | |
7216 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
7217 | &ctx->tags); | |
7218 | if (mac) { | |
c06bba01 | 7219 | if (mac->port.p != in_bundle) { |
479df176 | 7220 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 7221 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
7222 | } else { |
7223 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 7224 | } |
abe529af | 7225 | } else { |
c06bba01 | 7226 | struct ofbundle *bundle; |
abe529af | 7227 | |
479df176 | 7228 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
7229 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
7230 | if (bundle != in_bundle | |
7231 | && ofbundle_includes_vlan(bundle, vlan) | |
7232 | && bundle->floodable | |
7233 | && !bundle->mirror_out) { | |
7234 | output_normal(ctx, bundle, vlan); | |
7235 | } | |
7236 | } | |
7237 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 7238 | } |
abe529af BP |
7239 | } |
7240 | \f | |
54a9cbc9 BP |
7241 | /* Optimized flow revalidation. |
7242 | * | |
7243 | * It's a difficult problem, in general, to tell which facets need to have | |
7244 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7245 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7246 | * table changes, we recalculate the actions for every facet. This is | |
7247 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7248 | * doesn't change very often. | |
7249 | * | |
7250 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7251 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7252 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7253 | * table, we add a special case that applies to flow tables in which every rule | |
7254 | * has the same form (that is, the same wildcards), except that the table is | |
7255 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7256 | * optimize this case by tagging all of the facets that resubmit into the table | |
7257 | * and invalidating the same tag whenever a flow changes in that table. The | |
7258 | * end result is that we revalidate just the facets that need it (and sometimes | |
7259 | * a few more, but not all of the facets or even all of the facets that | |
7260 | * resubmit to the table modified by MAC learning). */ | |
7261 | ||
5cb7a798 | 7262 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted |
54a9cbc9 | 7263 | * into an OpenFlow table with the given 'basis'. */ |
822d9414 | 7264 | static tag_type |
5cb7a798 | 7265 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, |
54a9cbc9 BP |
7266 | uint32_t secret) |
7267 | { | |
5cb7a798 | 7268 | if (minimask_is_catchall(mask)) { |
54a9cbc9 BP |
7269 | return 0; |
7270 | } else { | |
5cb7a798 BP |
7271 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); |
7272 | return tag_create_deterministic(hash); | |
54a9cbc9 BP |
7273 | } |
7274 | } | |
7275 | ||
7276 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7277 | * taggability of that table. | |
7278 | * | |
7279 | * This function must be called after *each* change to a flow table. If you | |
7280 | * skip calling it on some changes then the pointer comparisons at the end can | |
7281 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7282 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7283 | * different wildcards to be created with the same address, then this function | |
7284 | * will incorrectly skip revalidation. */ | |
7285 | static void | |
7286 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7287 | { | |
7288 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 7289 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
7290 | struct cls_table *catchall, *other; |
7291 | struct cls_table *t; | |
7292 | ||
7293 | catchall = other = NULL; | |
7294 | ||
d0918789 | 7295 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
7296 | case 0: |
7297 | /* We could tag this OpenFlow table but it would make the logic a | |
7298 | * little harder and it's a corner case that doesn't seem worth it | |
7299 | * yet. */ | |
7300 | break; | |
7301 | ||
7302 | case 1: | |
7303 | case 2: | |
d0918789 | 7304 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
7305 | if (cls_table_is_catchall(t)) { |
7306 | catchall = t; | |
7307 | } else if (!other) { | |
7308 | other = t; | |
7309 | } else { | |
7310 | /* Indicate that we can't tag this by setting both tables to | |
7311 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7312 | other = NULL; | |
7313 | } | |
7314 | } | |
7315 | break; | |
7316 | ||
7317 | default: | |
7318 | /* Can't tag this table. */ | |
7319 | break; | |
7320 | } | |
7321 | ||
7322 | if (table->catchall_table != catchall || table->other_table != other) { | |
7323 | table->catchall_table = catchall; | |
7324 | table->other_table = other; | |
2cc3c58e | 7325 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7326 | } |
7327 | } | |
7328 | ||
7329 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7330 | * inserted, a rule being deleted, or a rule whose actions are being | |
7331 | * modified), marks facets for revalidation to ensure that packets will be | |
7332 | * forwarded correctly according to the new state of the flow table. | |
7333 | * | |
7334 | * This function must be called after *each* change to a flow table. See | |
7335 | * the comment on table_update_taggable() for more information. */ | |
7336 | static void | |
7337 | rule_invalidate(const struct rule_dpif *rule) | |
7338 | { | |
7339 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7340 | ||
7341 | table_update_taggable(ofproto, rule->up.table_id); | |
7342 | ||
2cc3c58e | 7343 | if (!ofproto->backer->need_revalidate) { |
54a9cbc9 BP |
7344 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; |
7345 | ||
7346 | if (table->other_table && rule->tag) { | |
2cc3c58e | 7347 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); |
54a9cbc9 | 7348 | } else { |
2cc3c58e | 7349 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7350 | } |
7351 | } | |
7352 | } | |
7353 | \f | |
abe529af | 7354 | static bool |
7257b535 BP |
7355 | set_frag_handling(struct ofproto *ofproto_, |
7356 | enum ofp_config_flags frag_handling) | |
abe529af BP |
7357 | { |
7358 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 7359 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 7360 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
7361 | return true; |
7362 | } else { | |
7363 | return false; | |
7364 | } | |
abe529af BP |
7365 | } |
7366 | ||
90bf1e07 | 7367 | static enum ofperr |
abe529af BP |
7368 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
7369 | const struct flow *flow, | |
f25d0cf3 | 7370 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
7371 | { |
7372 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
548de4dd BP |
7373 | struct odputil_keybuf keybuf; |
7374 | struct dpif_flow_stats stats; | |
abe529af | 7375 | |
548de4dd | 7376 | struct ofpbuf key; |
112bc5f4 | 7377 | |
548de4dd BP |
7378 | struct action_xlate_ctx ctx; |
7379 | uint64_t odp_actions_stub[1024 / 8]; | |
7380 | struct ofpbuf odp_actions; | |
80e5eed9 | 7381 | |
548de4dd | 7382 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
7383 | odp_flow_key_from_flow(&key, flow, |
7384 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
050ac423 | 7385 | |
548de4dd | 7386 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 7387 | |
548de4dd BP |
7388 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, |
7389 | packet_get_tcp_flags(packet, flow), packet); | |
7390 | ctx.resubmit_stats = &stats; | |
2284188b | 7391 | |
548de4dd BP |
7392 | ofpbuf_use_stub(&odp_actions, |
7393 | odp_actions_stub, sizeof odp_actions_stub); | |
7394 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
acf60855 | 7395 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
548de4dd BP |
7396 | odp_actions.data, odp_actions.size, packet); |
7397 | ofpbuf_uninit(&odp_actions); | |
2284188b | 7398 | |
548de4dd | 7399 | return 0; |
abe529af | 7400 | } |
6fca1ffb BP |
7401 | \f |
7402 | /* NetFlow. */ | |
7403 | ||
7404 | static int | |
7405 | set_netflow(struct ofproto *ofproto_, | |
7406 | const struct netflow_options *netflow_options) | |
7407 | { | |
7408 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7409 | ||
7410 | if (netflow_options) { | |
7411 | if (!ofproto->netflow) { | |
7412 | ofproto->netflow = netflow_create(); | |
7413 | } | |
7414 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7415 | } else { | |
7416 | netflow_destroy(ofproto->netflow); | |
7417 | ofproto->netflow = NULL; | |
7418 | return 0; | |
7419 | } | |
7420 | } | |
abe529af BP |
7421 | |
7422 | static void | |
7423 | get_netflow_ids(const struct ofproto *ofproto_, | |
7424 | uint8_t *engine_type, uint8_t *engine_id) | |
7425 | { | |
7426 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7427 | ||
acf60855 | 7428 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 7429 | } |
6fca1ffb BP |
7430 | |
7431 | static void | |
7432 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7433 | { | |
7434 | if (!facet_is_controller_flow(facet) && | |
7435 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 7436 | struct subfacet *subfacet; |
6fca1ffb BP |
7437 | struct ofexpired expired; |
7438 | ||
b0f7b9b5 | 7439 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 7440 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 7441 | struct dpif_flow_stats stats; |
6fca1ffb | 7442 | |
6a7e895f | 7443 | subfacet_reinstall(subfacet, &stats); |
15baa734 | 7444 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 7445 | } |
6fca1ffb BP |
7446 | } |
7447 | ||
7448 | expired.flow = facet->flow; | |
7449 | expired.packet_count = facet->packet_count; | |
7450 | expired.byte_count = facet->byte_count; | |
7451 | expired.used = facet->used; | |
7452 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7453 | } | |
7454 | } | |
7455 | ||
7456 | static void | |
7457 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7458 | { | |
7459 | struct facet *facet; | |
7460 | ||
7461 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7462 | send_active_timeout(ofproto, facet); | |
7463 | } | |
7464 | } | |
abe529af BP |
7465 | \f |
7466 | static struct ofproto_dpif * | |
7467 | ofproto_dpif_lookup(const char *name) | |
7468 | { | |
b44a10b7 BP |
7469 | struct ofproto_dpif *ofproto; |
7470 | ||
7471 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7472 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7473 | if (!strcmp(ofproto->up.name, name)) { | |
7474 | return ofproto; | |
7475 | } | |
7476 | } | |
7477 | return NULL; | |
abe529af BP |
7478 | } |
7479 | ||
f0a3aa2e | 7480 | static void |
96e466a3 | 7481 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 7482 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 7483 | { |
490df1ef | 7484 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 7485 | |
96e466a3 EJ |
7486 | if (argc > 1) { |
7487 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7488 | if (!ofproto) { | |
bde9f75d | 7489 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
7490 | return; |
7491 | } | |
2cc3c58e | 7492 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 EJ |
7493 | } else { |
7494 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2cc3c58e | 7495 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 | 7496 | } |
f0a3aa2e | 7497 | } |
f0a3aa2e | 7498 | |
bde9f75d | 7499 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
7500 | } |
7501 | ||
abe529af | 7502 | static void |
0e15264f BP |
7503 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
7504 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
7505 | { |
7506 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7507 | const struct ofproto_dpif *ofproto; | |
7508 | const struct mac_entry *e; | |
7509 | ||
0e15264f | 7510 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 7511 | if (!ofproto) { |
bde9f75d | 7512 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
7513 | return; |
7514 | } | |
7515 | ||
7516 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
7517 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
7518 | struct ofbundle *bundle = e->port.p; | |
7519 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
7520 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
7521 | e->vlan, ETH_ADDR_ARGS(e->mac), |
7522 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 7523 | } |
bde9f75d | 7524 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
7525 | ds_destroy(&ds); |
7526 | } | |
7527 | ||
6a6455e5 | 7528 | struct trace_ctx { |
abe529af BP |
7529 | struct action_xlate_ctx ctx; |
7530 | struct flow flow; | |
7531 | struct ds *result; | |
7532 | }; | |
7533 | ||
7534 | static void | |
29901626 BP |
7535 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
7536 | const struct rule_dpif *rule) | |
abe529af BP |
7537 | { |
7538 | ds_put_char_multiple(result, '\t', level); | |
7539 | if (!rule) { | |
7540 | ds_put_cstr(result, "No match\n"); | |
7541 | return; | |
7542 | } | |
7543 | ||
29901626 BP |
7544 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
7545 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 7546 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
7547 | ds_put_char(result, '\n'); |
7548 | ||
7549 | ds_put_char_multiple(result, '\t', level); | |
7550 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 7551 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
7552 | ds_put_char(result, '\n'); |
7553 | } | |
7554 | ||
7555 | static void | |
7556 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 7557 | struct trace_ctx *trace) |
abe529af BP |
7558 | { |
7559 | ds_put_char_multiple(result, '\t', level); | |
7560 | ds_put_format(result, "%s: ", title); | |
7561 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
7562 | ds_put_cstr(result, "unchanged"); | |
7563 | } else { | |
7564 | flow_format(result, &trace->ctx.flow); | |
7565 | trace->flow = trace->ctx.flow; | |
7566 | } | |
7567 | ds_put_char(result, '\n'); | |
7568 | } | |
7569 | ||
eb9e1c26 EJ |
7570 | static void |
7571 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 7572 | struct trace_ctx *trace) |
eb9e1c26 EJ |
7573 | { |
7574 | size_t i; | |
7575 | ||
7576 | ds_put_char_multiple(result, '\t', level); | |
7577 | ds_put_format(result, "%s:", title); | |
7578 | for (i = 0; i < FLOW_N_REGS; i++) { | |
7579 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
7580 | } | |
7581 | ds_put_char(result, '\n'); | |
7582 | } | |
7583 | ||
1ed8d352 EJ |
7584 | static void |
7585 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 7586 | struct trace_ctx *trace) |
1ed8d352 EJ |
7587 | { |
7588 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
7589 | ||
7590 | ds_put_char_multiple(result, '\t', level); | |
7591 | ds_put_format(result, "%s: ", title); | |
7592 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
7593 | ds_put_char(result, '\n'); | |
7594 | } | |
7595 | ||
abe529af BP |
7596 | static void |
7597 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
7598 | { | |
6a6455e5 | 7599 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
7600 | struct ds *result = trace->result; |
7601 | ||
7602 | ds_put_char(result, '\n'); | |
7603 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 7604 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 7605 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 7606 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
7607 | } |
7608 | ||
479df176 BP |
7609 | static void |
7610 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
7611 | { | |
7612 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
7613 | struct ds *result = trace->result; | |
7614 | ||
7615 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
7616 | ds_put_cstr(result, s); | |
7617 | ds_put_char(result, '\n'); | |
7618 | } | |
7619 | ||
abe529af | 7620 | static void |
0e15264f | 7621 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
7622 | void *aux OVS_UNUSED) |
7623 | { | |
0e15264f | 7624 | const char *dpname = argv[1]; |
abe529af | 7625 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
7626 | struct ofpbuf odp_key; |
7627 | struct ofpbuf *packet; | |
e84173dc | 7628 | ovs_be16 initial_tci; |
abe529af BP |
7629 | struct ds result; |
7630 | struct flow flow; | |
abe529af BP |
7631 | char *s; |
7632 | ||
876b0e1c BP |
7633 | packet = NULL; |
7634 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
7635 | ds_init(&result); |
7636 | ||
e84173dc BP |
7637 | ofproto = ofproto_dpif_lookup(dpname); |
7638 | if (!ofproto) { | |
bde9f75d EJ |
7639 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
7640 | "for help)"); | |
e84173dc BP |
7641 | goto exit; |
7642 | } | |
0e15264f | 7643 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 7644 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
7645 | const char *flow_s = argv[2]; |
7646 | const char *generate_s = argv[3]; | |
876b0e1c | 7647 | |
31a19d69 BP |
7648 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like |
7649 | * flow. We guess which type it is based on whether 'flow_s' contains | |
7650 | * an '(', since a datapath flow always contains '(') but an | |
7651 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
7652 | * that's not documented anywhere). | |
7653 | * | |
7654 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
7655 | * it would be tricky giving a sensible error message. After all, do | |
7656 | * you just say "syntax error" or do you present both error messages? | |
7657 | * Both choices seem lousy. */ | |
7658 | if (strchr(flow_s, '(')) { | |
7659 | int error; | |
7660 | ||
7661 | /* Convert string to datapath key. */ | |
7662 | ofpbuf_init(&odp_key, 0); | |
7663 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
7664 | if (error) { | |
7665 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
7666 | goto exit; | |
7667 | } | |
876b0e1c | 7668 | |
e09ee259 EJ |
7669 | /* XXX: Since we allow the user to specify an ofproto, it's |
7670 | * possible they will specify a different ofproto than the one the | |
7671 | * port actually belongs too. Ideally we should simply remove the | |
7672 | * ability to specify the ofproto. */ | |
7673 | if (ofproto_receive(ofproto->backer, NULL, odp_key.data, | |
7674 | odp_key.size, &flow, NULL, NULL, NULL, | |
7675 | &initial_tci)) { | |
31a19d69 BP |
7676 | unixctl_command_reply_error(conn, "Invalid flow"); |
7677 | goto exit; | |
7678 | } | |
7679 | } else { | |
7680 | char *error_s; | |
7681 | ||
7682 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
7683 | if (error_s) { | |
7684 | unixctl_command_reply_error(conn, error_s); | |
7685 | free(error_s); | |
7686 | goto exit; | |
7687 | } | |
7688 | ||
7689 | initial_tci = flow.vlan_tci; | |
876b0e1c | 7690 | } |
8b3b8dd1 BP |
7691 | |
7692 | /* Generate a packet, if requested. */ | |
0e15264f | 7693 | if (generate_s) { |
8b3b8dd1 BP |
7694 | packet = ofpbuf_new(0); |
7695 | flow_compose(packet, &flow); | |
7696 | } | |
72e8bf28 AA |
7697 | } else if (argc == 7) { |
7698 | /* ofproto/trace dpname priority tun_id in_port mark packet */ | |
0e15264f BP |
7699 | const char *priority_s = argv[2]; |
7700 | const char *tun_id_s = argv[3]; | |
7701 | const char *in_port_s = argv[4]; | |
72e8bf28 AA |
7702 | const char *mark_s = argv[5]; |
7703 | const char *packet_s = argv[6]; | |
9b56fe13 | 7704 | uint32_t in_port = atoi(in_port_s); |
0e15264f BP |
7705 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); |
7706 | uint32_t priority = atoi(priority_s); | |
72e8bf28 | 7707 | uint32_t mark = atoi(mark_s); |
e22f1753 | 7708 | const char *msg; |
0e15264f | 7709 | |
e22f1753 BP |
7710 | msg = eth_from_hex(packet_s, &packet); |
7711 | if (msg) { | |
bde9f75d | 7712 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
7713 | goto exit; |
7714 | } | |
7715 | ||
7716 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 7717 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
7718 | ds_put_cstr(&result, s); |
7719 | free(s); | |
7720 | ||
72e8bf28 | 7721 | flow_extract(packet, priority, mark, NULL, in_port, &flow); |
296e07ac | 7722 | flow.tunnel.tun_id = tun_id; |
e84173dc | 7723 | initial_tci = flow.vlan_tci; |
876b0e1c | 7724 | } else { |
bde9f75d | 7725 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
7726 | goto exit; |
7727 | } | |
7728 | ||
6a6455e5 EJ |
7729 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); |
7730 | unixctl_command_reply(conn, ds_cstr(&result)); | |
7731 | ||
7732 | exit: | |
7733 | ds_destroy(&result); | |
7734 | ofpbuf_delete(packet); | |
7735 | ofpbuf_uninit(&odp_key); | |
7736 | } | |
7737 | ||
7738 | static void | |
7739 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
7740 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
7741 | struct ds *ds) | |
7742 | { | |
7743 | struct rule_dpif *rule; | |
7744 | ||
7745 | ds_put_cstr(ds, "Flow: "); | |
7746 | flow_format(ds, flow); | |
7747 | ds_put_char(ds, '\n'); | |
abe529af | 7748 | |
c57b2226 BP |
7749 | rule = rule_dpif_lookup(ofproto, flow); |
7750 | ||
6a6455e5 | 7751 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
7752 | if (rule == ofproto->miss_rule) { |
7753 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
7754 | } else if (rule == ofproto->no_packet_in_rule) { | |
7755 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
7756 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7757 | } | |
7758 | ||
abe529af | 7759 | if (rule) { |
050ac423 BP |
7760 | uint64_t odp_actions_stub[1024 / 8]; |
7761 | struct ofpbuf odp_actions; | |
7762 | ||
6a6455e5 | 7763 | struct trace_ctx trace; |
0e553d9c | 7764 | uint8_t tcp_flags; |
abe529af | 7765 | |
6a6455e5 EJ |
7766 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
7767 | trace.result = ds; | |
7768 | trace.flow = *flow; | |
050ac423 BP |
7769 | ofpbuf_use_stub(&odp_actions, |
7770 | odp_actions_stub, sizeof odp_actions_stub); | |
6a6455e5 | 7771 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, |
0e553d9c | 7772 | rule, tcp_flags, packet); |
abe529af | 7773 | trace.ctx.resubmit_hook = trace_resubmit; |
479df176 | 7774 | trace.ctx.report_hook = trace_report; |
f25d0cf3 | 7775 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 7776 | &odp_actions); |
abe529af | 7777 | |
6a6455e5 EJ |
7778 | ds_put_char(ds, '\n'); |
7779 | trace_format_flow(ds, 0, "Final flow", &trace); | |
7780 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
7781 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
7782 | ofpbuf_uninit(&odp_actions); | |
876b0e1c | 7783 | |
6a7e895f BP |
7784 | if (trace.ctx.slow) { |
7785 | enum slow_path_reason slow; | |
7786 | ||
7787 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
7788 | "slow path because it:"); | |
7789 | for (slow = trace.ctx.slow; slow; ) { | |
7790 | enum slow_path_reason bit = rightmost_1bit(slow); | |
7791 | ||
7792 | switch (bit) { | |
7793 | case SLOW_CFM: | |
7794 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
7795 | break; | |
7796 | case SLOW_LACP: | |
7797 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
7798 | break; | |
7799 | case SLOW_STP: | |
7800 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
7801 | break; | |
7802 | case SLOW_IN_BAND: | |
7803 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
7804 | "processing."); | |
7805 | if (!packet) { | |
7806 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
7807 | "incomplete--for complete actions, " | |
7808 | "please supply a packet.)"); | |
7809 | } | |
7810 | break; | |
7811 | case SLOW_CONTROLLER: | |
7812 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
7813 | "to the OpenFlow controller."); | |
7814 | break; | |
7815 | case SLOW_MATCH: | |
7816 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
7817 | "than the datapath supports."); | |
7818 | break; | |
7819 | } | |
7820 | ||
7821 | slow &= ~bit; | |
7822 | } | |
7823 | ||
7824 | if (slow & ~SLOW_MATCH) { | |
7825 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
7826 | "the special slow-path processing."); | |
876b0e1c BP |
7827 | } |
7828 | } | |
abe529af | 7829 | } |
abe529af BP |
7830 | } |
7831 | ||
7ee20df1 | 7832 | static void |
0e15264f BP |
7833 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
7834 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
7835 | { |
7836 | clogged = true; | |
bde9f75d | 7837 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
7838 | } |
7839 | ||
7840 | static void | |
0e15264f BP |
7841 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
7842 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
7843 | { |
7844 | clogged = false; | |
bde9f75d | 7845 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
7846 | } |
7847 | ||
6814e51f BP |
7848 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
7849 | * 'reply' describing the results. */ | |
7850 | static void | |
7851 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
7852 | { | |
7853 | struct facet *facet; | |
7854 | int errors; | |
7855 | ||
7856 | errors = 0; | |
7857 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7858 | if (!facet_check_consistency(facet)) { | |
7859 | errors++; | |
7860 | } | |
7861 | } | |
7862 | if (errors) { | |
2cc3c58e | 7863 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
7864 | } |
7865 | ||
7866 | if (errors) { | |
7867 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
7868 | ofproto->up.name, errors); | |
7869 | } else { | |
7870 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
7871 | } | |
7872 | } | |
7873 | ||
7874 | static void | |
7875 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
7876 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
7877 | { | |
7878 | struct ds reply = DS_EMPTY_INITIALIZER; | |
7879 | struct ofproto_dpif *ofproto; | |
7880 | ||
7881 | if (argc > 1) { | |
7882 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7883 | if (!ofproto) { | |
bde9f75d EJ |
7884 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
7885 | "ofproto/list for help)"); | |
6814e51f BP |
7886 | return; |
7887 | } | |
7888 | ofproto_dpif_self_check__(ofproto, &reply); | |
7889 | } else { | |
7890 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
7891 | ofproto_dpif_self_check__(ofproto, &reply); | |
7892 | } | |
7893 | } | |
7894 | ||
bde9f75d | 7895 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
7896 | ds_destroy(&reply); |
7897 | } | |
7898 | ||
27022416 JP |
7899 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
7900 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
7901 | * to destroy 'ofproto_shash' and free the returned value. */ | |
7902 | static const struct shash_node ** | |
7903 | get_ofprotos(struct shash *ofproto_shash) | |
7904 | { | |
7905 | const struct ofproto_dpif *ofproto; | |
7906 | ||
7907 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
7908 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
7909 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
7910 | } | |
7911 | ||
7912 | return shash_sort(ofproto_shash); | |
7913 | } | |
7914 | ||
7915 | static void | |
7916 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
7917 | const char *argv[] OVS_UNUSED, | |
7918 | void *aux OVS_UNUSED) | |
7919 | { | |
7920 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7921 | struct shash ofproto_shash; | |
7922 | const struct shash_node **sorted_ofprotos; | |
7923 | int i; | |
7924 | ||
7925 | shash_init(&ofproto_shash); | |
7926 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
7927 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
7928 | const struct shash_node *node = sorted_ofprotos[i]; | |
7929 | ds_put_format(&ds, "%s\n", node->name); | |
7930 | } | |
7931 | ||
7932 | shash_destroy(&ofproto_shash); | |
7933 | free(sorted_ofprotos); | |
7934 | ||
7935 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
7936 | ds_destroy(&ds); | |
7937 | } | |
7938 | ||
7939 | static void | |
7940 | show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) | |
7941 | { | |
7942 | struct dpif_dp_stats s; | |
7943 | const struct shash_node **ports; | |
7944 | int i; | |
7945 | ||
acf60855 | 7946 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
27022416 | 7947 | |
acf60855 JP |
7948 | ds_put_format(ds, "%s (%s):\n", ofproto->up.name, |
7949 | dpif_name(ofproto->backer->dpif)); | |
7950 | /* xxx It would be better to show bridge-specific stats instead | |
7951 | * xxx of dp ones. */ | |
27022416 JP |
7952 | ds_put_format(ds, |
7953 | "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n", | |
7954 | s.n_hit, s.n_missed, s.n_lost); | |
acf60855 JP |
7955 | ds_put_format(ds, "\tflows: %zu\n", |
7956 | hmap_count(&ofproto->subfacets)); | |
27022416 JP |
7957 | |
7958 | ports = shash_sort(&ofproto->up.port_by_name); | |
7959 | for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { | |
7960 | const struct shash_node *node = ports[i]; | |
7961 | struct ofport *ofport = node->data; | |
7962 | const char *name = netdev_get_name(ofport->netdev); | |
7963 | const char *type = netdev_get_type(ofport->netdev); | |
0a740f48 EJ |
7964 | uint32_t odp_port; |
7965 | ||
7966 | ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); | |
7967 | ||
7968 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); | |
7969 | if (odp_port != OVSP_NONE) { | |
7970 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
7971 | } else { | |
7972 | ds_put_cstr(ds, "none:"); | |
7973 | } | |
27022416 | 7974 | |
27022416 JP |
7975 | if (strcmp(type, "system")) { |
7976 | struct netdev *netdev; | |
7977 | int error; | |
7978 | ||
7979 | ds_put_format(ds, " (%s", type); | |
7980 | ||
7981 | error = netdev_open(name, type, &netdev); | |
7982 | if (!error) { | |
7983 | struct smap config; | |
7984 | ||
7985 | smap_init(&config); | |
7986 | error = netdev_get_config(netdev, &config); | |
7987 | if (!error) { | |
7988 | const struct smap_node **nodes; | |
7989 | size_t i; | |
7990 | ||
7991 | nodes = smap_sort(&config); | |
7992 | for (i = 0; i < smap_count(&config); i++) { | |
7993 | const struct smap_node *node = nodes[i]; | |
7994 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
7995 | node->key, node->value); | |
7996 | } | |
7997 | free(nodes); | |
7998 | } | |
7999 | smap_destroy(&config); | |
8000 | ||
8001 | netdev_close(netdev); | |
8002 | } | |
8003 | ds_put_char(ds, ')'); | |
8004 | } | |
8005 | ds_put_char(ds, '\n'); | |
8006 | } | |
8007 | free(ports); | |
8008 | } | |
8009 | ||
8010 | static void | |
8011 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, | |
8012 | const char *argv[], void *aux OVS_UNUSED) | |
8013 | { | |
8014 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8015 | const struct ofproto_dpif *ofproto; | |
8016 | ||
8017 | if (argc > 1) { | |
8018 | int i; | |
8019 | for (i = 1; i < argc; i++) { | |
8020 | ofproto = ofproto_dpif_lookup(argv[i]); | |
8021 | if (!ofproto) { | |
8022 | ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " | |
8023 | "for help)", argv[i]); | |
8024 | unixctl_command_reply_error(conn, ds_cstr(&ds)); | |
8025 | return; | |
8026 | } | |
8027 | show_dp_format(ofproto, &ds); | |
8028 | } | |
8029 | } else { | |
8030 | struct shash ofproto_shash; | |
8031 | const struct shash_node **sorted_ofprotos; | |
8032 | int i; | |
8033 | ||
8034 | shash_init(&ofproto_shash); | |
8035 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8036 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8037 | const struct shash_node *node = sorted_ofprotos[i]; | |
8038 | show_dp_format(node->data, &ds); | |
8039 | } | |
8040 | ||
8041 | shash_destroy(&ofproto_shash); | |
8042 | free(sorted_ofprotos); | |
8043 | } | |
8044 | ||
8045 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8046 | ds_destroy(&ds); | |
8047 | } | |
8048 | ||
8049 | static void | |
8050 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8051 | int argc OVS_UNUSED, const char *argv[], | |
8052 | void *aux OVS_UNUSED) | |
8053 | { | |
8054 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8055 | const struct ofproto_dpif *ofproto; | |
8056 | struct subfacet *subfacet; | |
8057 | ||
8058 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8059 | if (!ofproto) { | |
8060 | unixctl_command_reply_error(conn, "no such bridge"); | |
8061 | return; | |
8062 | } | |
8063 | ||
af37354d EJ |
8064 | update_stats(ofproto->backer); |
8065 | ||
27022416 JP |
8066 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
8067 | struct odputil_keybuf keybuf; | |
8068 | struct ofpbuf key; | |
8069 | ||
8070 | subfacet_get_key(subfacet, &keybuf, &key); | |
8071 | odp_flow_key_format(key.data, key.size, &ds); | |
8072 | ||
8073 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8074 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8075 | if (subfacet->used) { | |
8076 | ds_put_format(&ds, "%.3fs", | |
8077 | (time_msec() - subfacet->used) / 1000.0); | |
8078 | } else { | |
8079 | ds_put_format(&ds, "never"); | |
8080 | } | |
8081 | if (subfacet->facet->tcp_flags) { | |
8082 | ds_put_cstr(&ds, ", flags:"); | |
8083 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8084 | } | |
8085 | ||
8086 | ds_put_cstr(&ds, ", actions:"); | |
8087 | format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); | |
8088 | ds_put_char(&ds, '\n'); | |
8089 | } | |
8090 | ||
8091 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8092 | ds_destroy(&ds); | |
8093 | } | |
8094 | ||
8095 | static void | |
8096 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8097 | int argc OVS_UNUSED, const char *argv[], | |
8098 | void *aux OVS_UNUSED) | |
8099 | { | |
8100 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8101 | struct ofproto_dpif *ofproto; | |
8102 | ||
8103 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8104 | if (!ofproto) { | |
8105 | unixctl_command_reply_error(conn, "no such bridge"); | |
8106 | return; | |
8107 | } | |
8108 | ||
8109 | flush(&ofproto->up); | |
8110 | ||
8111 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8112 | ds_destroy(&ds); | |
8113 | } | |
8114 | ||
abe529af BP |
8115 | static void |
8116 | ofproto_dpif_unixctl_init(void) | |
8117 | { | |
8118 | static bool registered; | |
8119 | if (registered) { | |
8120 | return; | |
8121 | } | |
8122 | registered = true; | |
8123 | ||
0e15264f BP |
8124 | unixctl_command_register( |
8125 | "ofproto/trace", | |
72e8bf28 AA |
8126 | "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", |
8127 | 2, 6, ofproto_unixctl_trace, NULL); | |
96e466a3 | 8128 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
8129 | ofproto_unixctl_fdb_flush, NULL); |
8130 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8131 | ofproto_unixctl_fdb_show, NULL); | |
8132 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8133 | ofproto_dpif_clog, NULL); | |
8134 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8135 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
8136 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
8137 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
8138 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
8139 | ofproto_unixctl_dpif_dump_dps, NULL); | |
8140 | unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, | |
8141 | ofproto_unixctl_dpif_show, NULL); | |
8142 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, | |
8143 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8144 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8145 | ofproto_unixctl_dpif_del_flows, NULL); | |
abe529af BP |
8146 | } |
8147 | \f | |
52a90c29 BP |
8148 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
8149 | * | |
8150 | * This is deprecated. It is only for compatibility with broken device drivers | |
8151 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8152 | * devices are not used. When broken device drivers are no longer in | |
8153 | * widespread use, we will delete these interfaces. */ | |
8154 | ||
8155 | static int | |
8156 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8157 | { | |
8158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8159 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8160 | ||
8161 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8162 | && vid == ofport->vlandev_vid) { | |
8163 | return 0; | |
8164 | } | |
8165 | ||
2cc3c58e | 8166 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
8167 | |
8168 | if (ofport->realdev_ofp_port) { | |
8169 | vsp_remove(ofport); | |
8170 | } | |
8171 | if (realdev_ofp_port && ofport->bundle) { | |
8172 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8173 | * themselves be part of a bundle. */ | |
8174 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8175 | } | |
8176 | ||
8177 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8178 | ofport->vlandev_vid = vid; | |
8179 | ||
8180 | if (realdev_ofp_port) { | |
8181 | vsp_add(ofport, realdev_ofp_port, vid); | |
8182 | } | |
8183 | ||
8184 | return 0; | |
8185 | } | |
8186 | ||
8187 | static uint32_t | |
8188 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8189 | { | |
8190 | return hash_2words(realdev_ofp_port, vid); | |
8191 | } | |
8192 | ||
40e05935 BP |
8193 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
8194 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
8195 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
8196 | * it would return the port number of eth0.9. | |
8197 | * | |
8198 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
8199 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
8200 | static uint32_t |
8201 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
8202 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
8203 | { | |
8204 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
e1b1d06a | 8205 | uint16_t realdev_ofp_port; |
52a90c29 BP |
8206 | int vid = vlan_tci_to_vid(vlan_tci); |
8207 | const struct vlan_splinter *vsp; | |
8208 | ||
e1b1d06a | 8209 | realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); |
52a90c29 BP |
8210 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, |
8211 | hash_realdev_vid(realdev_ofp_port, vid), | |
8212 | &ofproto->realdev_vid_map) { | |
8213 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8214 | && vsp->vid == vid) { | |
e1b1d06a | 8215 | return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); |
52a90c29 BP |
8216 | } |
8217 | } | |
8218 | } | |
8219 | return realdev_odp_port; | |
8220 | } | |
8221 | ||
8222 | static struct vlan_splinter * | |
8223 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8224 | { | |
8225 | struct vlan_splinter *vsp; | |
8226 | ||
8227 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8228 | &ofproto->vlandev_map) { | |
8229 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8230 | return vsp; | |
8231 | } | |
8232 | } | |
8233 | ||
8234 | return NULL; | |
8235 | } | |
8236 | ||
40e05935 BP |
8237 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
8238 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8239 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8240 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8241 | * eth0 and store 9 in '*vid'. | |
8242 | * | |
8243 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8244 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8245 | * always does.*/ | |
52a90c29 BP |
8246 | static uint16_t |
8247 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 8248 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
8249 | { |
8250 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8251 | const struct vlan_splinter *vsp; | |
8252 | ||
8253 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8254 | if (vsp) { | |
8255 | if (vid) { | |
8256 | *vid = vsp->vid; | |
8257 | } | |
8258 | return vsp->realdev_ofp_port; | |
8259 | } | |
8260 | } | |
8261 | return 0; | |
8262 | } | |
8263 | ||
b98d8985 BP |
8264 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
8265 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8266 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8267 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8268 | * always the case unless VLAN splinters are enabled), returns false without | |
8269 | * making any changes. */ | |
8270 | static bool | |
8271 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8272 | { | |
8273 | uint16_t realdev; | |
8274 | int vid; | |
8275 | ||
8276 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8277 | if (!realdev) { | |
8278 | return false; | |
8279 | } | |
8280 | ||
8281 | /* Cause the flow to be processed as if it came in on the real device with | |
8282 | * the VLAN device's VLAN ID. */ | |
8283 | flow->in_port = realdev; | |
8284 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8285 | return true; | |
8286 | } | |
8287 | ||
52a90c29 BP |
8288 | static void |
8289 | vsp_remove(struct ofport_dpif *port) | |
8290 | { | |
8291 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8292 | struct vlan_splinter *vsp; | |
8293 | ||
8294 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8295 | if (vsp) { | |
8296 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8297 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8298 | free(vsp); | |
8299 | ||
8300 | port->realdev_ofp_port = 0; | |
8301 | } else { | |
8302 | VLOG_ERR("missing vlan device record"); | |
8303 | } | |
8304 | } | |
8305 | ||
8306 | static void | |
8307 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8308 | { | |
8309 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8310 | ||
8311 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8312 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8313 | == realdev_ofp_port)) { | |
8314 | struct vlan_splinter *vsp; | |
8315 | ||
8316 | vsp = xmalloc(sizeof *vsp); | |
8317 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8318 | hash_int(port->up.ofp_port, 0)); | |
8319 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8320 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8321 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8322 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8323 | vsp->vid = vid; | |
8324 | ||
8325 | port->realdev_ofp_port = realdev_ofp_port; | |
8326 | } else { | |
8327 | VLOG_ERR("duplicate vlan device record"); | |
8328 | } | |
8329 | } | |
e1b1d06a JP |
8330 | |
8331 | static uint32_t | |
8332 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8333 | { | |
8334 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8335 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8336 | } | |
8337 | ||
acf60855 JP |
8338 | static struct ofport_dpif * |
8339 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
e1b1d06a JP |
8340 | { |
8341 | struct ofport_dpif *port; | |
8342 | ||
8343 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8344 | hash_int(odp_port, 0), | |
acf60855 | 8345 | &backer->odp_to_ofport_map) { |
e1b1d06a | 8346 | if (port->odp_port == odp_port) { |
acf60855 | 8347 | return port; |
e1b1d06a JP |
8348 | } |
8349 | } | |
8350 | ||
acf60855 JP |
8351 | return NULL; |
8352 | } | |
8353 | ||
8354 | static uint16_t | |
8355 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8356 | { | |
8357 | struct ofport_dpif *port; | |
8358 | ||
8359 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 8360 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
8361 | return port->up.ofp_port; |
8362 | } else { | |
8363 | return OFPP_NONE; | |
8364 | } | |
e1b1d06a JP |
8365 | } |
8366 | ||
abe529af | 8367 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 8368 | init, |
abe529af BP |
8369 | enumerate_types, |
8370 | enumerate_names, | |
8371 | del, | |
0aeaabc8 | 8372 | port_open_type, |
acf60855 JP |
8373 | type_run, |
8374 | type_run_fast, | |
8375 | type_wait, | |
abe529af BP |
8376 | alloc, |
8377 | construct, | |
8378 | destruct, | |
8379 | dealloc, | |
8380 | run, | |
5fcc0d00 | 8381 | run_fast, |
abe529af | 8382 | wait, |
0d085684 | 8383 | get_memory_usage, |
abe529af | 8384 | flush, |
6c1491fb BP |
8385 | get_features, |
8386 | get_tables, | |
abe529af BP |
8387 | port_alloc, |
8388 | port_construct, | |
8389 | port_destruct, | |
8390 | port_dealloc, | |
8391 | port_modified, | |
8392 | port_reconfigured, | |
8393 | port_query_by_name, | |
8394 | port_add, | |
8395 | port_del, | |
6527c598 | 8396 | port_get_stats, |
abe529af BP |
8397 | port_dump_start, |
8398 | port_dump_next, | |
8399 | port_dump_done, | |
8400 | port_poll, | |
8401 | port_poll_wait, | |
8402 | port_is_lacp_current, | |
0ab6decf | 8403 | NULL, /* rule_choose_table */ |
abe529af BP |
8404 | rule_alloc, |
8405 | rule_construct, | |
8406 | rule_destruct, | |
8407 | rule_dealloc, | |
abe529af BP |
8408 | rule_get_stats, |
8409 | rule_execute, | |
8410 | rule_modify_actions, | |
7257b535 | 8411 | set_frag_handling, |
abe529af BP |
8412 | packet_out, |
8413 | set_netflow, | |
8414 | get_netflow_ids, | |
8415 | set_sflow, | |
8416 | set_cfm, | |
a5610457 | 8417 | get_cfm_fault, |
1c0333b6 | 8418 | get_cfm_opup, |
1de11730 | 8419 | get_cfm_remote_mpids, |
3967a833 | 8420 | get_cfm_health, |
21f7563c JP |
8421 | set_stp, |
8422 | get_stp_status, | |
8423 | set_stp_port, | |
8424 | get_stp_port_status, | |
8b36f51e | 8425 | set_queues, |
abe529af BP |
8426 | bundle_set, |
8427 | bundle_remove, | |
8428 | mirror_set, | |
9d24de3b | 8429 | mirror_get_stats, |
abe529af BP |
8430 | set_flood_vlans, |
8431 | is_mirror_output_bundle, | |
8402c74b | 8432 | forward_bpdu_changed, |
c4069512 | 8433 | set_mac_table_config, |
52a90c29 | 8434 | set_realdev, |
abe529af | 8435 | }; |