]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e09ee259 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
ccc09689 | 23 | #include "bfd.h" |
abe529af | 24 | #include "bond.h" |
daff3353 | 25 | #include "bundle.h" |
abe529af BP |
26 | #include "byte-order.h" |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
75a75043 | 35 | #include "learn.h" |
abe529af | 36 | #include "mac-learning.h" |
816fd533 | 37 | #include "meta-flow.h" |
abe529af | 38 | #include "multipath.h" |
0a740f48 | 39 | #include "netdev-vport.h" |
abe529af BP |
40 | #include "netdev.h" |
41 | #include "netlink.h" | |
42 | #include "nx-match.h" | |
43 | #include "odp-util.h" | |
44 | #include "ofp-util.h" | |
45 | #include "ofpbuf.h" | |
f25d0cf3 | 46 | #include "ofp-actions.h" |
31a19d69 | 47 | #include "ofp-parse.h" |
abe529af | 48 | #include "ofp-print.h" |
9d6ac44e | 49 | #include "ofproto-dpif-governor.h" |
29089a54 | 50 | #include "ofproto-dpif-ipfix.h" |
bae473fe | 51 | #include "ofproto-dpif-sflow.h" |
abe529af | 52 | #include "poll-loop.h" |
0d085684 | 53 | #include "simap.h" |
27022416 | 54 | #include "smap.h" |
abe529af | 55 | #include "timer.h" |
b9ad7294 | 56 | #include "tunnel.h" |
6c1491fb | 57 | #include "unaligned.h" |
abe529af BP |
58 | #include "unixctl.h" |
59 | #include "vlan-bitmap.h" | |
60 | #include "vlog.h" | |
61 | ||
62 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
63 | ||
abe529af | 64 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
65 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
66 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
67 | COVERAGE_DEFINE(facet_revalidate); |
68 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 69 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 70 | |
29901626 | 71 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 72 | * flow translation. */ |
1642690c | 73 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 74 | |
9cdaaebe BP |
75 | /* Number of implemented OpenFlow tables. */ |
76 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
77 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
78 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 79 | |
abe529af BP |
80 | struct ofport_dpif; |
81 | struct ofproto_dpif; | |
a088a1ff | 82 | struct flow_miss; |
ac35f9c8 | 83 | struct facet; |
abe529af BP |
84 | |
85 | struct rule_dpif { | |
86 | struct rule up; | |
87 | ||
abe529af BP |
88 | /* These statistics: |
89 | * | |
90 | * - Do include packets and bytes from facets that have been deleted or | |
91 | * whose own statistics have been folded into the rule. | |
92 | * | |
93 | * - Do include packets and bytes sent "by hand" that were accounted to | |
94 | * the rule without any facet being involved (this is a rare corner | |
95 | * case in rule_execute()). | |
96 | * | |
97 | * - Do not include packet or bytes that can be obtained from any facet's | |
98 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 99 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
100 | */ |
101 | uint64_t packet_count; /* Number of packets received. */ | |
102 | uint64_t byte_count; /* Number of bytes received. */ | |
103 | ||
54a9cbc9 BP |
104 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
105 | ||
abe529af BP |
106 | struct list facets; /* List of "struct facet"s. */ |
107 | }; | |
108 | ||
109 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
110 | { | |
111 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
112 | } | |
113 | ||
29901626 | 114 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
115 | const struct flow *); |
116 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
117 | const struct flow *, | |
118 | uint8_t table); | |
c376f9a3 IY |
119 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
120 | const struct flow *flow); | |
abe529af | 121 | |
7e741ffb | 122 | static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); |
112bc5f4 BP |
123 | static void rule_credit_stats(struct rule_dpif *, |
124 | const struct dpif_flow_stats *); | |
ac35f9c8 | 125 | static void flow_push_stats(struct facet *, const struct dpif_flow_stats *); |
822d9414 | 126 | static tag_type rule_calculate_tag(const struct flow *, |
5cb7a798 | 127 | const struct minimask *, uint32_t basis); |
b0f7b9b5 BP |
128 | static void rule_invalidate(const struct rule_dpif *); |
129 | ||
abe529af BP |
130 | #define MAX_MIRRORS 32 |
131 | typedef uint32_t mirror_mask_t; | |
132 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
133 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
134 | struct ofmirror { | |
135 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
136 | size_t idx; /* In ofproto's "mirrors" array. */ | |
137 | void *aux; /* Key supplied by ofproto's client. */ | |
138 | char *name; /* Identifier for log messages. */ | |
139 | ||
140 | /* Selection criteria. */ | |
141 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
142 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
143 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
144 | ||
9ba15e2a | 145 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
146 | struct ofbundle *out; /* Output port or NULL. */ |
147 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 148 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
149 | |
150 | /* Counters. */ | |
151 | int64_t packet_count; /* Number of packets sent. */ | |
152 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
153 | }; |
154 | ||
155 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
156 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
157 | mirror_mask_t mirrors, | |
158 | uint64_t packets, uint64_t bytes); | |
abe529af | 159 | |
abe529af | 160 | struct ofbundle { |
abe529af | 161 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 162 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
163 | void *aux; /* Key supplied by ofproto's client. */ |
164 | char *name; /* Identifier for log messages. */ | |
165 | ||
166 | /* Configuration. */ | |
167 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 168 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
169 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
170 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
171 | * NULL if all VLANs are trunked. */ | |
172 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
173 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 174 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
175 | |
176 | /* Status. */ | |
9e1fd49b | 177 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
178 | |
179 | /* Port mirroring info. */ | |
180 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
181 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
182 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
183 | }; | |
184 | ||
185 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 186 | static void bundle_update(struct ofbundle *); |
abe529af BP |
187 | static void bundle_destroy(struct ofbundle *); |
188 | static void bundle_del_port(struct ofport_dpif *); | |
189 | static void bundle_run(struct ofbundle *); | |
190 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 191 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
192 | uint16_t in_port, bool warn, |
193 | struct ofport_dpif **in_ofportp); | |
abe529af | 194 | |
33158a18 JP |
195 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
196 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
197 | * when an input bundle is needed for validation (e.g., mirroring or | |
198 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
199 | * any 'port' structs, so care must be taken when dealing with it. */ | |
200 | static struct ofbundle ofpp_none_bundle = { | |
201 | .name = "OFPP_NONE", | |
202 | .vlan_mode = PORT_VLAN_TRUNK | |
203 | }; | |
204 | ||
21f7563c JP |
205 | static void stp_run(struct ofproto_dpif *ofproto); |
206 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
207 | static int set_stp_port(struct ofport *, |
208 | const struct ofproto_port_stp_settings *); | |
21f7563c | 209 | |
5da5ec37 BP |
210 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
211 | ||
bbafd73b EJ |
212 | struct xlate_ctx; |
213 | ||
214 | /* Initial values of fields of the packet that may be changed during | |
215 | * flow processing and needed later. */ | |
216 | struct initial_vals { | |
217 | /* This is the value of vlan_tci in the packet as actually received from | |
218 | * dpif. This is the same as the facet's flow.vlan_tci unless the packet | |
219 | * was received via a VLAN splinter. In that case, this value is 0 | |
220 | * (because the packet as actually received from the dpif had no 802.1Q | |
221 | * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter | |
222 | * represents. | |
223 | * | |
224 | * This member should be removed when the VLAN splinters feature is no | |
225 | * longer needed. */ | |
226 | ovs_be16 vlan_tci; | |
227 | }; | |
228 | ||
229 | struct xlate_out { | |
230 | tag_type tags; /* Tags associated with actions. */ | |
231 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
232 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
233 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
234 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
235 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
236 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ | |
237 | ||
238 | uint64_t odp_actions_stub[256 / 8]; | |
239 | struct ofpbuf odp_actions; | |
240 | }; | |
abe529af | 241 | |
bbafd73b | 242 | struct xlate_in { |
abe529af BP |
243 | struct ofproto_dpif *ofproto; |
244 | ||
245 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
246 | * this flow when actions change header fields. */ | |
247 | struct flow flow; | |
248 | ||
bbafd73b | 249 | struct initial_vals initial_vals; |
bd85dac1 | 250 | |
abe529af BP |
251 | /* The packet corresponding to 'flow', or a null pointer if we are |
252 | * revalidating without a packet to refer to. */ | |
253 | const struct ofpbuf *packet; | |
254 | ||
3de9590b BP |
255 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
256 | * actions update the flow table? | |
257 | * | |
258 | * We want to update these tables if we are actually processing a packet, | |
259 | * or if we are accounting for packets that the datapath has processed, but | |
260 | * not if we are just revalidating. */ | |
261 | bool may_learn; | |
75a75043 | 262 | |
bbafd73b | 263 | /* The rule initiating translation or NULL. */ |
18b2a258 | 264 | struct rule_dpif *rule; |
54834960 | 265 | |
bbafd73b EJ |
266 | /* The actions to translate. If 'rule' is not NULL, these may be NULL. */ |
267 | const struct ofpact *ofpacts; | |
268 | size_t ofpacts_len; | |
269 | ||
0e553d9c BP |
270 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
271 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
272 | * timeouts.) */ | |
273 | uint8_t tcp_flags; | |
274 | ||
112bc5f4 BP |
275 | /* If nonnull, flow translation calls this function just before executing a |
276 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
277 | * when the recursion depth is exceeded. | |
278 | * | |
279 | * 'rule' is the rule being submitted into. It will be null if the | |
280 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
281 | * | |
282 | * This is normally null so the client has to set it manually after | |
bbafd73b | 283 | * calling xlate_in_init(). */ |
f03a84b9 | 284 | void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule); |
112bc5f4 | 285 | |
479df176 BP |
286 | /* If nonnull, flow translation calls this function to report some |
287 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
288 | * dropped a packet. */ | |
f03a84b9 | 289 | void (*report_hook)(struct xlate_ctx *, const char *s); |
479df176 | 290 | |
112bc5f4 BP |
291 | /* If nonnull, flow translation credits the specified statistics to each |
292 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
293 | * |
294 | * This is normally null so the client has to set it manually after | |
bbafd73b | 295 | * calling xlate_in_init(). */ |
112bc5f4 | 296 | const struct dpif_flow_stats *resubmit_stats; |
bbafd73b | 297 | }; |
abe529af | 298 | |
bbafd73b EJ |
299 | /* Context used by xlate_actions() and its callees. */ |
300 | struct xlate_ctx { | |
301 | struct xlate_in *xin; | |
302 | struct xlate_out *xout; | |
abe529af | 303 | |
bbafd73b EJ |
304 | struct ofproto_dpif *ofproto; |
305 | ||
306 | /* Flow at the last commit. */ | |
307 | struct flow base_flow; | |
abe529af | 308 | |
bbafd73b EJ |
309 | /* Tunnel IP destination address as received. This is stored separately |
310 | * as the base_flow.tunnel is cleared on init to reflect the datapath | |
311 | * behavior. Used to make sure not to send tunneled output to ourselves, | |
312 | * which might lead to an infinite loop. This could happen easily | |
313 | * if a tunnel is marked as 'ip_remote=flow', and the flow does not | |
314 | * actually set the tun_dst field. */ | |
315 | ovs_be32 orig_tunnel_ip_dst; | |
316 | ||
317 | /* Stack for the push and pop actions. Each stack element is of type | |
318 | * "union mf_subvalue". */ | |
319 | union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; | |
320 | struct ofpbuf stack; | |
321 | ||
322 | /* The rule that we are currently translating, or NULL. */ | |
323 | struct rule_dpif *rule; | |
abe529af BP |
324 | |
325 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 326 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
deedf7e7 | 327 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 328 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 | 329 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
9b56fe13 | 330 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ |
6ff686f2 | 331 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ |
848e8809 | 332 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
333 | }; |
334 | ||
bbafd73b EJ |
335 | static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *, |
336 | const struct flow *, const struct initial_vals *, | |
337 | struct rule_dpif *, uint8_t tcp_flags, | |
338 | const struct ofpbuf *); | |
339 | ||
340 | static void xlate_out_uninit(struct xlate_out *); | |
341 | ||
342 | static void xlate_actions(struct xlate_in *, struct xlate_out *); | |
343 | ||
344 | static void xlate_actions_for_side_effects(struct xlate_in *); | |
14f94f9a | 345 | |
f03a84b9 | 346 | static void xlate_table_action(struct xlate_ctx *, uint16_t in_port, |
0a740f48 | 347 | uint8_t table_id, bool may_packet_in); |
abe529af | 348 | |
6a7e895f BP |
349 | static size_t put_userspace_action(const struct ofproto_dpif *, |
350 | struct ofpbuf *odp_actions, | |
351 | const struct flow *, | |
29089a54 RL |
352 | const union user_action_cookie *, |
353 | const size_t); | |
6a7e895f BP |
354 | |
355 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
356 | enum slow_path_reason, | |
357 | uint64_t *stub, size_t stub_size, | |
358 | const struct nlattr **actionsp, | |
359 | size_t *actions_lenp); | |
360 | ||
f03a84b9 | 361 | static void xlate_report(struct xlate_ctx *ctx, const char *s); |
479df176 | 362 | |
6a7e895f BP |
363 | /* A subfacet (see "struct subfacet" below) has three possible installation |
364 | * states: | |
365 | * | |
366 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
367 | * case just after the subfacet is created, just before the subfacet is | |
368 | * destroyed, or if the datapath returns an error when we try to install a | |
369 | * subfacet. | |
370 | * | |
371 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
372 | * | |
373 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
374 | * ofproto_dpif is installed in the datapath. | |
375 | */ | |
376 | enum subfacet_path { | |
377 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
378 | SF_FAST_PATH, /* Full actions are installed. */ | |
379 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
380 | }; | |
381 | ||
5f5fbd17 BP |
382 | /* A dpif flow and actions associated with a facet. |
383 | * | |
384 | * See also the large comment on struct facet. */ | |
385 | struct subfacet { | |
386 | /* Owners. */ | |
387 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
388 | struct list list_node; /* In struct facet's 'facets' list. */ | |
389 | struct facet *facet; /* Owning facet. */ | |
390 | ||
5f5fbd17 BP |
391 | enum odp_key_fitness key_fitness; |
392 | struct nlattr *key; | |
393 | int key_len; | |
394 | ||
395 | long long int used; /* Time last used; time created if not used. */ | |
655ab909 | 396 | long long int created; /* Time created. */ |
5f5fbd17 BP |
397 | |
398 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
399 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
400 | ||
6a7e895f | 401 | enum subfacet_path path; /* Installed in datapath? */ |
5f5fbd17 | 402 | |
a088a1ff JP |
403 | /* Datapath port the packet arrived on. This is needed to remove |
404 | * flows for ports that are no longer part of the bridge. Since the | |
405 | * flow definition only has the OpenFlow port number and the port is | |
406 | * no longer part of the bridge, we can't determine the datapath port | |
407 | * number needed to delete the flow from the datapath. */ | |
408 | uint32_t odp_in_port; | |
5f5fbd17 BP |
409 | }; |
410 | ||
1d85f9e5 JP |
411 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
412 | ||
a088a1ff | 413 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 414 | long long int now); |
5f5fbd17 | 415 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
acf60855 | 416 | const struct nlattr *key, size_t key_len, |
9566abf9 | 417 | uint32_t key_hash); |
5f5fbd17 BP |
418 | static void subfacet_destroy(struct subfacet *); |
419 | static void subfacet_destroy__(struct subfacet *); | |
1d85f9e5 JP |
420 | static void subfacet_destroy_batch(struct ofproto_dpif *, |
421 | struct subfacet **, int n); | |
5f5fbd17 BP |
422 | static void subfacet_reset_dp_stats(struct subfacet *, |
423 | struct dpif_flow_stats *); | |
424 | static void subfacet_update_time(struct subfacet *, long long int used); | |
425 | static void subfacet_update_stats(struct subfacet *, | |
426 | const struct dpif_flow_stats *); | |
5f5fbd17 | 427 | static int subfacet_install(struct subfacet *, |
4dff9097 EJ |
428 | const struct ofpbuf *odp_actions, |
429 | struct dpif_flow_stats *); | |
5f5fbd17 BP |
430 | static void subfacet_uninstall(struct subfacet *); |
431 | ||
b0f7b9b5 BP |
432 | /* An exact-match instantiation of an OpenFlow flow. |
433 | * | |
434 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
435 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
436 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
437 | * the facet. When the kernel module (or other dpif implementation) and Open | |
438 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
439 | * one subfacet per facet. If the dpif implementation supports more-specific | |
440 | * flow matching than userspace, however, a facet can have more than one | |
441 | * subfacet, each of which corresponds to some distinction in flow that | |
442 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
443 | * |
444 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
445 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 446 | struct facet { |
b0f7b9b5 BP |
447 | /* Owners. */ |
448 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
449 | struct list list_node; /* In owning rule's 'facets' list. */ | |
450 | struct rule_dpif *rule; /* Owning rule. */ | |
451 | ||
452 | /* Owned data. */ | |
453 | struct list subfacets; | |
abe529af BP |
454 | long long int used; /* Time last used; time created if not used. */ |
455 | ||
b0f7b9b5 BP |
456 | /* Key. */ |
457 | struct flow flow; | |
458 | ||
abe529af BP |
459 | /* These statistics: |
460 | * | |
461 | * - Do include packets and bytes sent "by hand", e.g. with | |
462 | * dpif_execute(). | |
463 | * | |
464 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 465 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 466 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
467 | * |
468 | * - Do not include packets or bytes that can be obtained from the | |
469 | * datapath for any existing subfacet. | |
abe529af BP |
470 | */ |
471 | uint64_t packet_count; /* Number of packets received. */ | |
472 | uint64_t byte_count; /* Number of bytes received. */ | |
473 | ||
b0f7b9b5 | 474 | /* Resubmit statistics. */ |
9d24de3b JP |
475 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
476 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
477 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 478 | |
b0f7b9b5 | 479 | /* Accounting. */ |
907a4c5e | 480 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 481 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 482 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 483 | |
bbafd73b | 484 | struct xlate_out xout; |
4dff9097 EJ |
485 | |
486 | /* Initial values of the packet that may be needed later. */ | |
487 | struct initial_vals initial_vals; | |
488 | ||
26cd7e34 BP |
489 | /* Storage for a single subfacet, to reduce malloc() time and space |
490 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
491 | * case has exactly one subfacet. However, 'one_subfacet' may not |
492 | * always be valid, since it could have been removed after newer | |
493 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 494 | struct subfacet one_subfacet; |
6cf474d7 EJ |
495 | |
496 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
497 | }; |
498 | ||
4dff9097 | 499 | static struct facet *facet_create(const struct flow_miss *, uint32_t hash); |
15baa734 | 500 | static void facet_remove(struct facet *); |
abe529af BP |
501 | static void facet_free(struct facet *); |
502 | ||
2b459b83 BP |
503 | static struct facet *facet_find(struct ofproto_dpif *, |
504 | const struct flow *, uint32_t hash); | |
abe529af | 505 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 506 | const struct flow *, uint32_t hash); |
5bf64ade | 507 | static bool facet_revalidate(struct facet *); |
6814e51f | 508 | static bool facet_check_consistency(struct facet *); |
abe529af | 509 | |
15baa734 | 510 | static void facet_flush_stats(struct facet *); |
abe529af | 511 | |
15baa734 | 512 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 513 | static void facet_reset_counters(struct facet *); |
abe529af | 514 | static void facet_push_stats(struct facet *); |
3de9590b BP |
515 | static void facet_learn(struct facet *); |
516 | static void facet_account(struct facet *); | |
8844e035 | 517 | static void push_all_stats(void); |
abe529af BP |
518 | |
519 | static bool facet_is_controller_flow(struct facet *); | |
520 | ||
abe529af | 521 | struct ofport_dpif { |
acf60855 | 522 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ |
abe529af BP |
523 | struct ofport up; |
524 | ||
525 | uint32_t odp_port; | |
526 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
527 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
528 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
ccc09689 | 529 | struct bfd *bfd; /* BFD, if any. */ |
abe529af | 530 | tag_type tag; /* Tag associated with this port. */ |
015e08bc | 531 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 532 | long long int carrier_seq; /* Carrier status changes. */ |
b9ad7294 | 533 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ |
21f7563c | 534 | |
52a90c29 | 535 | /* Spanning tree. */ |
21f7563c JP |
536 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
537 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
538 | long long int stp_state_entered; | |
8b36f51e EJ |
539 | |
540 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
541 | |
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 | |
545 | * drivers in old versions of Linux that do not properly support VLANs when | |
546 | * VLAN devices are not used. When broken device drivers are no longer in | |
547 | * widespread use, we will delete these interfaces. */ | |
548 | uint16_t realdev_ofp_port; | |
549 | int vlandev_vid; | |
8b36f51e EJ |
550 | }; |
551 | ||
552 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
553 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
554 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
555 | struct priority_to_dscp { | |
556 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
557 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
558 | ||
559 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
560 | }; |
561 | ||
52a90c29 BP |
562 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
563 | * | |
564 | * This is deprecated. It is only for compatibility with broken device drivers | |
565 | * in old versions of Linux that do not properly support VLANs when VLAN | |
566 | * devices are not used. When broken device drivers are no longer in | |
567 | * widespread use, we will delete these interfaces. */ | |
568 | struct vlan_splinter { | |
569 | struct hmap_node realdev_vid_node; | |
570 | struct hmap_node vlandev_node; | |
571 | uint16_t realdev_ofp_port; | |
572 | uint16_t vlandev_ofp_port; | |
573 | int vid; | |
574 | }; | |
575 | ||
576 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
577 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 578 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
579 | static void vsp_remove(struct ofport_dpif *); |
580 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
581 | ||
e1b1d06a JP |
582 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
583 | uint16_t ofp_port); | |
584 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
585 | uint32_t odp_port); | |
586 | ||
abe529af BP |
587 | static struct ofport_dpif * |
588 | ofport_dpif_cast(const struct ofport *ofport) | |
589 | { | |
abe529af BP |
590 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
591 | } | |
592 | ||
593 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 594 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 595 | static void port_wait(struct ofport_dpif *); |
8aee94b6 | 596 | static int set_bfd(struct ofport *, const struct smap *); |
a5610457 | 597 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 598 | static void ofport_clear_priorities(struct ofport_dpif *); |
8fa4d1d0 | 599 | static void run_fast_rl(void); |
abe529af | 600 | |
7ee20df1 BP |
601 | struct dpif_completion { |
602 | struct list list_node; | |
603 | struct ofoperation *op; | |
604 | }; | |
605 | ||
54a9cbc9 BP |
606 | /* Extra information about a classifier table. |
607 | * Currently used just for optimized flow revalidation. */ | |
608 | struct table_dpif { | |
609 | /* If either of these is nonnull, then this table has a form that allows | |
610 | * flows to be tagged to avoid revalidating most flows for the most common | |
611 | * kinds of flow table changes. */ | |
612 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
613 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
614 | uint32_t basis; /* Keeps each table's tags separate. */ | |
615 | }; | |
616 | ||
3c4a309c BP |
617 | /* Reasons that we might need to revalidate every facet, and corresponding |
618 | * coverage counters. | |
619 | * | |
620 | * A value of 0 means that there is no need to revalidate. | |
621 | * | |
622 | * It would be nice to have some cleaner way to integrate with coverage | |
623 | * counters, but with only a few reasons I guess this is good enough for | |
624 | * now. */ | |
625 | enum revalidate_reason { | |
626 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
627 | REV_STP, /* Spanning tree protocol port status change. */ | |
628 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
629 | REV_FLOW_TABLE, /* Flow table changed. */ | |
630 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
631 | }; | |
632 | COVERAGE_DEFINE(rev_reconfigure); | |
633 | COVERAGE_DEFINE(rev_stp); | |
634 | COVERAGE_DEFINE(rev_port_toggled); | |
635 | COVERAGE_DEFINE(rev_flow_table); | |
636 | COVERAGE_DEFINE(rev_inconsistency); | |
637 | ||
8f73d537 EJ |
638 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
639 | * These are datapath flows which have no associated ofproto, if they did we | |
640 | * would use facets. */ | |
641 | struct drop_key { | |
642 | struct hmap_node hmap_node; | |
643 | struct nlattr *key; | |
644 | size_t key_len; | |
645 | }; | |
646 | ||
acf60855 JP |
647 | /* All datapaths of a given type share a single dpif backer instance. */ |
648 | struct dpif_backer { | |
649 | char *type; | |
650 | int refcount; | |
651 | struct dpif *dpif; | |
652 | struct timer next_expiration; | |
653 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
2cc3c58e | 654 | |
7d82ab2e | 655 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ |
b9ad7294 | 656 | |
2cc3c58e EJ |
657 | /* Facet revalidation flags applying to facets which use this backer. */ |
658 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
659 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
8f73d537 EJ |
660 | |
661 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
acf60855 JP |
662 | }; |
663 | ||
664 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
665 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
666 | ||
8f73d537 | 667 | static void drop_key_clear(struct dpif_backer *); |
acf60855 JP |
668 | static struct ofport_dpif * |
669 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
670 | ||
735d7efb AZ |
671 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, |
672 | uint64_t delta); | |
655ab909 AZ |
673 | struct avg_subfacet_rates { |
674 | double add_rate; /* Moving average of new flows created per minute. */ | |
675 | double del_rate; /* Moving average of flows deleted per minute. */ | |
676 | }; | |
677 | static void show_dp_rates(struct ds *ds, const char *heading, | |
678 | const struct avg_subfacet_rates *rates); | |
679 | static void exp_mavg(double *avg, int base, double new); | |
735d7efb | 680 | |
abe529af | 681 | struct ofproto_dpif { |
b44a10b7 | 682 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af | 683 | struct ofproto up; |
acf60855 | 684 | struct dpif_backer *backer; |
abe529af | 685 | |
c57b2226 BP |
686 | /* Special OpenFlow rules. */ |
687 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
688 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
689 | ||
abe529af BP |
690 | /* Bridging. */ |
691 | struct netflow *netflow; | |
bae473fe | 692 | struct dpif_sflow *sflow; |
29089a54 | 693 | struct dpif_ipfix *ipfix; |
abe529af BP |
694 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
695 | struct mac_learning *ml; | |
696 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 697 | bool has_mirrors; |
abe529af BP |
698 | bool has_bonded_bundles; |
699 | ||
abe529af BP |
700 | /* Facets. */ |
701 | struct hmap facets; | |
b0f7b9b5 | 702 | struct hmap subfacets; |
9d6ac44e | 703 | struct governor *governor; |
4d2a0f39 | 704 | long long int consistency_rl; |
54a9cbc9 BP |
705 | |
706 | /* Revalidation. */ | |
707 | struct table_dpif tables[N_TABLES]; | |
7ee20df1 BP |
708 | |
709 | /* Support for debugging async flow mods. */ | |
710 | struct list completions; | |
daff3353 EJ |
711 | |
712 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
713 | struct netdev_stats stats; /* To account packets generated and consumed in |
714 | * userspace. */ | |
21f7563c JP |
715 | |
716 | /* Spanning tree. */ | |
717 | struct stp *stp; | |
718 | long long int stp_last_tick; | |
52a90c29 BP |
719 | |
720 | /* VLAN splinters. */ | |
721 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
722 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
e1b1d06a | 723 | |
acf60855 | 724 | /* Ports. */ |
0a740f48 EJ |
725 | struct sset ports; /* Set of standard port names. */ |
726 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
acf60855 JP |
727 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ |
728 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
735d7efb AZ |
729 | |
730 | /* Per ofproto's dpif stats. */ | |
731 | uint64_t n_hit; | |
732 | uint64_t n_missed; | |
655ab909 AZ |
733 | |
734 | /* Subfacet statistics. | |
735 | * | |
736 | * These keep track of the total number of subfacets added and deleted and | |
737 | * flow life span. They are useful for computing the flow rates stats | |
738 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
739 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
740 | * performance in new situations. */ | |
741 | long long int created; /* Time when it is created. */ | |
742 | unsigned int max_n_subfacet; /* Maximum number of flows */ | |
743 | ||
744 | /* The average number of subfacets... */ | |
745 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
746 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
747 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
748 | ||
749 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
750 | unsigned int subfacet_add_count; | |
751 | unsigned int subfacet_del_count; | |
752 | ||
753 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
754 | unsigned long long int total_subfacet_add_count; | |
755 | unsigned long long int total_subfacet_del_count; | |
756 | ||
757 | /* Sum of the number of milliseconds that each subfacet existed, | |
758 | * over the subfacets that have been added and then later deleted. */ | |
759 | unsigned long long int total_subfacet_life_span; | |
760 | ||
761 | /* Incremented by the number of currently existing subfacets, each | |
762 | * time we pull statistics from the kernel. */ | |
763 | unsigned long long int total_subfacet_count; | |
764 | ||
765 | /* Number of times we pull statistics from the kernel. */ | |
766 | unsigned long long int n_update_stats; | |
abe529af | 767 | }; |
655ab909 AZ |
768 | static unsigned long long int avg_subfacet_life_span( |
769 | const struct ofproto_dpif *); | |
770 | static double avg_subfacet_count(const struct ofproto_dpif *ofproto); | |
771 | static void update_moving_averages(struct ofproto_dpif *ofproto); | |
772 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, | |
773 | uint64_t delta); | |
774 | static void update_max_subfacet_count(struct ofproto_dpif *ofproto); | |
abe529af | 775 | |
7ee20df1 BP |
776 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
777 | * for debugging the asynchronous flow_mod implementation.) */ | |
778 | static bool clogged; | |
779 | ||
b44a10b7 BP |
780 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
781 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
782 | ||
abe529af BP |
783 | static void ofproto_dpif_unixctl_init(void); |
784 | ||
785 | static struct ofproto_dpif * | |
786 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
787 | { | |
cb22974d | 788 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
789 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); |
790 | } | |
791 | ||
4acbc98d | 792 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 793 | uint16_t ofp_port); |
4acbc98d | 794 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 795 | uint32_t odp_port); |
6a6455e5 | 796 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
14f94f9a JP |
797 | const struct ofpbuf *, |
798 | const struct initial_vals *, struct ds *); | |
abe529af BP |
799 | |
800 | /* Packet processing. */ | |
801 | static void update_learning_table(struct ofproto_dpif *, | |
802 | const struct flow *, int vlan, | |
803 | struct ofbundle *); | |
501f8d1f BP |
804 | /* Upcalls. */ |
805 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 806 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
807 | |
808 | /* Flow expiration. */ | |
acf60855 | 809 | static int expire(struct dpif_backer *); |
abe529af | 810 | |
6fca1ffb BP |
811 | /* NetFlow. */ |
812 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
813 | ||
abe529af | 814 | /* Utilities. */ |
52a90c29 | 815 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
816 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
817 | struct ofpbuf *odp_actions, | |
818 | const struct flow *, uint32_t odp_port); | |
29089a54 RL |
819 | static void compose_ipfix_action(const struct ofproto_dpif *, |
820 | struct ofpbuf *odp_actions, | |
821 | const struct flow *); | |
f03a84b9 | 822 | static void add_mirror_actions(struct xlate_ctx *ctx, |
c06bba01 | 823 | const struct flow *flow); |
abe529af BP |
824 | /* Global variables. */ |
825 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
826 | |
827 | /* Initial mappings of port to bridge mappings. */ | |
828 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
abe529af BP |
829 | \f |
830 | /* Factory functions. */ | |
831 | ||
b0408fca | 832 | static void |
acf60855 | 833 | init(const struct shash *iface_hints) |
b0408fca | 834 | { |
acf60855 JP |
835 | struct shash_node *node; |
836 | ||
837 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
838 | SHASH_FOR_EACH(node, iface_hints) { | |
839 | const struct iface_hint *orig_hint = node->data; | |
840 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
841 | ||
842 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
843 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
844 | new_hint->ofp_port = orig_hint->ofp_port; | |
845 | ||
846 | shash_add(&init_ofp_ports, node->name, new_hint); | |
847 | } | |
b0408fca JP |
848 | } |
849 | ||
abe529af BP |
850 | static void |
851 | enumerate_types(struct sset *types) | |
852 | { | |
853 | dp_enumerate_types(types); | |
854 | } | |
855 | ||
856 | static int | |
857 | enumerate_names(const char *type, struct sset *names) | |
858 | { | |
acf60855 JP |
859 | struct ofproto_dpif *ofproto; |
860 | ||
861 | sset_clear(names); | |
862 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
863 | if (strcmp(type, ofproto->up.type)) { | |
864 | continue; | |
865 | } | |
866 | sset_add(names, ofproto->up.name); | |
867 | } | |
868 | ||
869 | return 0; | |
abe529af BP |
870 | } |
871 | ||
872 | static int | |
873 | del(const char *type, const char *name) | |
874 | { | |
875 | struct dpif *dpif; | |
876 | int error; | |
877 | ||
878 | error = dpif_open(name, type, &dpif); | |
879 | if (!error) { | |
880 | error = dpif_delete(dpif); | |
881 | dpif_close(dpif); | |
882 | } | |
883 | return error; | |
884 | } | |
885 | \f | |
0aeaabc8 JP |
886 | static const char * |
887 | port_open_type(const char *datapath_type, const char *port_type) | |
888 | { | |
889 | return dpif_port_open_type(datapath_type, port_type); | |
890 | } | |
891 | ||
acf60855 JP |
892 | /* Type functions. */ |
893 | ||
476cb42a BP |
894 | static struct ofproto_dpif * |
895 | lookup_ofproto_dpif_by_port_name(const char *name) | |
896 | { | |
897 | struct ofproto_dpif *ofproto; | |
898 | ||
899 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
900 | if (sset_contains(&ofproto->ports, name)) { | |
901 | return ofproto; | |
902 | } | |
903 | } | |
904 | ||
905 | return NULL; | |
906 | } | |
907 | ||
acf60855 JP |
908 | static int |
909 | type_run(const char *type) | |
910 | { | |
e20b5746 | 911 | static long long int push_timer = LLONG_MIN; |
acf60855 JP |
912 | struct dpif_backer *backer; |
913 | char *devname; | |
914 | int error; | |
915 | ||
916 | backer = shash_find_data(&all_dpif_backers, type); | |
917 | if (!backer) { | |
918 | /* This is not necessarily a problem, since backers are only | |
919 | * created on demand. */ | |
920 | return 0; | |
921 | } | |
922 | ||
923 | dpif_run(backer->dpif); | |
924 | ||
e20b5746 EJ |
925 | /* The most natural place to push facet statistics is when they're pulled |
926 | * from the datapath. However, when there are many flows in the datapath, | |
927 | * this expensive operation can occur so frequently, that it reduces our | |
928 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
929 | * here instead. */ | |
930 | if (time_msec() > push_timer) { | |
931 | push_timer = time_msec() + 2000; | |
932 | push_all_stats(); | |
933 | } | |
934 | ||
2cc3c58e EJ |
935 | if (backer->need_revalidate |
936 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
937 | struct tag_set revalidate_set = backer->revalidate_set; | |
938 | bool need_revalidate = backer->need_revalidate; | |
939 | struct ofproto_dpif *ofproto; | |
a614d823 KM |
940 | struct simap_node *node; |
941 | struct simap tmp_backers; | |
942 | ||
943 | /* Handle tunnel garbage collection. */ | |
944 | simap_init(&tmp_backers); | |
945 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
946 | ||
947 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
948 | struct ofport_dpif *iter; | |
949 | ||
950 | if (backer != ofproto->backer) { | |
951 | continue; | |
952 | } | |
953 | ||
954 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
955 | const char *dp_port; | |
956 | ||
957 | if (!iter->tnl_port) { | |
958 | continue; | |
959 | } | |
960 | ||
961 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev); | |
962 | node = simap_find(&tmp_backers, dp_port); | |
963 | if (node) { | |
964 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
965 | simap_delete(&tmp_backers, node); | |
966 | node = simap_find(&backer->tnl_backers, dp_port); | |
967 | } else { | |
968 | node = simap_find(&backer->tnl_backers, dp_port); | |
969 | if (!node) { | |
970 | uint32_t odp_port = UINT32_MAX; | |
971 | ||
972 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
973 | &odp_port)) { | |
974 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
975 | node = simap_find(&backer->tnl_backers, dp_port); | |
976 | } | |
977 | } | |
978 | } | |
979 | ||
980 | iter->odp_port = node ? node->data : OVSP_NONE; | |
981 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
982 | &iter->tnl_port)) { | |
983 | backer->need_revalidate = REV_RECONFIGURE; | |
984 | } | |
985 | } | |
986 | } | |
987 | ||
988 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
989 | dpif_port_del(backer->dpif, node->data); | |
990 | } | |
991 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
992 | |
993 | switch (backer->need_revalidate) { | |
994 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
995 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
996 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
997 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
998 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
999 | } | |
1000 | ||
8f73d537 EJ |
1001 | if (backer->need_revalidate) { |
1002 | /* Clear the drop_keys in case we should now be accepting some | |
1003 | * formerly dropped flows. */ | |
1004 | drop_key_clear(backer); | |
1005 | } | |
1006 | ||
f728af2e BP |
1007 | /* Clear the revalidation flags. */ |
1008 | tag_set_init(&backer->revalidate_set); | |
1009 | backer->need_revalidate = 0; | |
1010 | ||
2cc3c58e | 1011 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 1012 | struct facet *facet, *next; |
2cc3c58e EJ |
1013 | |
1014 | if (ofproto->backer != backer) { | |
1015 | continue; | |
1016 | } | |
1017 | ||
f231418e | 1018 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { |
2cc3c58e | 1019 | if (need_revalidate |
bbafd73b | 1020 | || tag_set_intersects(&revalidate_set, facet->xout.tags)) { |
2cc3c58e | 1021 | facet_revalidate(facet); |
8fa4d1d0 | 1022 | run_fast_rl(); |
2cc3c58e EJ |
1023 | } |
1024 | } | |
1025 | } | |
2cc3c58e EJ |
1026 | } |
1027 | ||
acf60855 JP |
1028 | if (timer_expired(&backer->next_expiration)) { |
1029 | int delay = expire(backer); | |
1030 | timer_set_duration(&backer->next_expiration, delay); | |
1031 | } | |
1032 | ||
1033 | /* Check for port changes in the dpif. */ | |
1034 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
476cb42a | 1035 | struct ofproto_dpif *ofproto; |
acf60855 JP |
1036 | struct dpif_port port; |
1037 | ||
1038 | /* Don't report on the datapath's device. */ | |
1039 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
c83b89ab | 1040 | goto next; |
acf60855 JP |
1041 | } |
1042 | ||
b9ad7294 EJ |
1043 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, |
1044 | &all_ofproto_dpifs) { | |
7d82ab2e | 1045 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { |
b9ad7294 EJ |
1046 | goto next; |
1047 | } | |
1048 | } | |
1049 | ||
476cb42a | 1050 | ofproto = lookup_ofproto_dpif_by_port_name(devname); |
acf60855 JP |
1051 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { |
1052 | /* The port was removed. If we know the datapath, | |
1053 | * report it through poll_set(). If we don't, it may be | |
1054 | * notifying us of a removal we initiated, so ignore it. | |
1055 | * If there's a pending ENOBUFS, let it stand, since | |
1056 | * everything will be reevaluated. */ | |
1057 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
1058 | sset_add(&ofproto->port_poll_set, devname); | |
1059 | ofproto->port_poll_errno = 0; | |
1060 | } | |
acf60855 JP |
1061 | } else if (!ofproto) { |
1062 | /* The port was added, but we don't know with which | |
1063 | * ofproto we should associate it. Delete it. */ | |
1064 | dpif_port_del(backer->dpif, port.port_no); | |
1065 | } | |
5b5e6a4c | 1066 | dpif_port_destroy(&port); |
acf60855 | 1067 | |
c83b89ab | 1068 | next: |
acf60855 JP |
1069 | free(devname); |
1070 | } | |
1071 | ||
1072 | if (error != EAGAIN) { | |
1073 | struct ofproto_dpif *ofproto; | |
1074 | ||
1075 | /* There was some sort of error, so propagate it to all | |
1076 | * ofprotos that use this backer. */ | |
1077 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1078 | &all_ofproto_dpifs) { | |
1079 | if (ofproto->backer == backer) { | |
1080 | sset_clear(&ofproto->port_poll_set); | |
1081 | ofproto->port_poll_errno = error; | |
1082 | } | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | return 0; | |
1087 | } | |
1088 | ||
1089 | static int | |
8fa4d1d0 | 1090 | dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) |
acf60855 | 1091 | { |
acf60855 JP |
1092 | unsigned int work; |
1093 | ||
acf60855 JP |
1094 | /* Handle one or more batches of upcalls, until there's nothing left to do |
1095 | * or until we do a fixed total amount of work. | |
1096 | * | |
1097 | * We do work in batches because it can be much cheaper to set up a number | |
1098 | * of flows and fire off their patches all at once. We do multiple batches | |
1099 | * because in some cases handling a packet can cause another packet to be | |
1100 | * queued almost immediately as part of the return flow. Both | |
1101 | * optimizations can make major improvements on some benchmarks and | |
1102 | * presumably for real traffic as well. */ | |
1103 | work = 0; | |
8fa4d1d0 EJ |
1104 | while (work < max_batch) { |
1105 | int retval = handle_upcalls(backer, max_batch - work); | |
acf60855 JP |
1106 | if (retval <= 0) { |
1107 | return -retval; | |
1108 | } | |
1109 | work += retval; | |
1110 | } | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
8fa4d1d0 EJ |
1115 | static int |
1116 | type_run_fast(const char *type) | |
1117 | { | |
1118 | struct dpif_backer *backer; | |
1119 | ||
1120 | backer = shash_find_data(&all_dpif_backers, type); | |
1121 | if (!backer) { | |
1122 | /* This is not necessarily a problem, since backers are only | |
1123 | * created on demand. */ | |
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1128 | } | |
1129 | ||
1130 | static void | |
1131 | run_fast_rl(void) | |
1132 | { | |
1133 | static long long int port_rl = LLONG_MIN; | |
1134 | static unsigned int backer_rl = 0; | |
1135 | ||
1136 | if (time_msec() >= port_rl) { | |
1137 | struct ofproto_dpif *ofproto; | |
1138 | struct ofport_dpif *ofport; | |
1139 | ||
1140 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1141 | ||
1142 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1143 | port_run_fast(ofport); | |
1144 | } | |
1145 | } | |
1146 | port_rl = time_msec() + 200; | |
1147 | } | |
1148 | ||
1149 | /* XXX: We have to be careful not to do too much work in this function. If | |
1150 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1151 | * performance improves signifcantly, but at a cost. It's possible for the | |
1152 | * number of flows in the datapath to increase without bound, and for poll | |
1153 | * loops to take 10s of seconds. The correct solution to this problem, | |
1154 | * long term, is to separate flow miss handling into it's own thread so it | |
1155 | * isn't affected by revalidations, and expirations. Until then, this is | |
1156 | * the best we can do. */ | |
1157 | if (++backer_rl >= 10) { | |
1158 | struct shash_node *node; | |
1159 | ||
1160 | backer_rl = 0; | |
1161 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1162 | dpif_backer_run_fast(node->data, 1); | |
1163 | } | |
1164 | } | |
1165 | } | |
1166 | ||
acf60855 JP |
1167 | static void |
1168 | type_wait(const char *type) | |
1169 | { | |
1170 | struct dpif_backer *backer; | |
1171 | ||
1172 | backer = shash_find_data(&all_dpif_backers, type); | |
1173 | if (!backer) { | |
1174 | /* This is not necessarily a problem, since backers are only | |
1175 | * created on demand. */ | |
1176 | return; | |
1177 | } | |
1178 | ||
1179 | timer_wait(&backer->next_expiration); | |
1180 | } | |
1181 | \f | |
abe529af BP |
1182 | /* Basic life-cycle. */ |
1183 | ||
c57b2226 BP |
1184 | static int add_internal_flows(struct ofproto_dpif *); |
1185 | ||
abe529af BP |
1186 | static struct ofproto * |
1187 | alloc(void) | |
1188 | { | |
1189 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1190 | return &ofproto->up; | |
1191 | } | |
1192 | ||
1193 | static void | |
1194 | dealloc(struct ofproto *ofproto_) | |
1195 | { | |
1196 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1197 | free(ofproto); | |
1198 | } | |
1199 | ||
acf60855 JP |
1200 | static void |
1201 | close_dpif_backer(struct dpif_backer *backer) | |
1202 | { | |
1203 | struct shash_node *node; | |
1204 | ||
cb22974d | 1205 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1206 | |
1207 | if (--backer->refcount) { | |
1208 | return; | |
1209 | } | |
1210 | ||
8f73d537 EJ |
1211 | drop_key_clear(backer); |
1212 | hmap_destroy(&backer->drop_keys); | |
1213 | ||
7d82ab2e | 1214 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1215 | hmap_destroy(&backer->odp_to_ofport_map); |
1216 | node = shash_find(&all_dpif_backers, backer->type); | |
1217 | free(backer->type); | |
1218 | shash_delete(&all_dpif_backers, node); | |
1219 | dpif_close(backer->dpif); | |
1220 | ||
1221 | free(backer); | |
1222 | } | |
1223 | ||
1224 | /* Datapath port slated for removal from datapath. */ | |
1225 | struct odp_garbage { | |
1226 | struct list list_node; | |
1227 | uint32_t odp_port; | |
1228 | }; | |
1229 | ||
1230 | static int | |
1231 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1232 | { | |
1233 | struct dpif_backer *backer; | |
1234 | struct dpif_port_dump port_dump; | |
1235 | struct dpif_port port; | |
1236 | struct shash_node *node; | |
1237 | struct list garbage_list; | |
1238 | struct odp_garbage *garbage, *next; | |
1239 | struct sset names; | |
1240 | char *backer_name; | |
1241 | const char *name; | |
1242 | int error; | |
1243 | ||
1244 | backer = shash_find_data(&all_dpif_backers, type); | |
1245 | if (backer) { | |
1246 | backer->refcount++; | |
1247 | *backerp = backer; | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | backer_name = xasprintf("ovs-%s", type); | |
1252 | ||
1253 | /* Remove any existing datapaths, since we assume we're the only | |
1254 | * userspace controlling the datapath. */ | |
1255 | sset_init(&names); | |
1256 | dp_enumerate_names(type, &names); | |
1257 | SSET_FOR_EACH(name, &names) { | |
1258 | struct dpif *old_dpif; | |
1259 | ||
1260 | /* Don't remove our backer if it exists. */ | |
1261 | if (!strcmp(name, backer_name)) { | |
1262 | continue; | |
1263 | } | |
1264 | ||
1265 | if (dpif_open(name, type, &old_dpif)) { | |
1266 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1267 | } else { | |
1268 | dpif_delete(old_dpif); | |
1269 | dpif_close(old_dpif); | |
1270 | } | |
1271 | } | |
1272 | sset_destroy(&names); | |
1273 | ||
1274 | backer = xmalloc(sizeof *backer); | |
1275 | ||
1276 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1277 | free(backer_name); | |
1278 | if (error) { | |
1279 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1280 | strerror(error)); | |
4c1b1289 | 1281 | free(backer); |
acf60855 JP |
1282 | return error; |
1283 | } | |
1284 | ||
1285 | backer->type = xstrdup(type); | |
1286 | backer->refcount = 1; | |
1287 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1288 | hmap_init(&backer->drop_keys); |
acf60855 | 1289 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1290 | backer->need_revalidate = 0; |
7d82ab2e | 1291 | simap_init(&backer->tnl_backers); |
2cc3c58e | 1292 | tag_set_init(&backer->revalidate_set); |
acf60855 JP |
1293 | *backerp = backer; |
1294 | ||
1295 | dpif_flow_flush(backer->dpif); | |
1296 | ||
1297 | /* Loop through the ports already on the datapath and remove any | |
1298 | * that we don't need anymore. */ | |
1299 | list_init(&garbage_list); | |
1300 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1301 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1302 | node = shash_find(&init_ofp_ports, port.name); | |
1303 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1304 | garbage = xmalloc(sizeof *garbage); | |
1305 | garbage->odp_port = port.port_no; | |
1306 | list_push_front(&garbage_list, &garbage->list_node); | |
1307 | } | |
1308 | } | |
1309 | dpif_port_dump_done(&port_dump); | |
1310 | ||
1311 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1312 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1313 | list_remove(&garbage->list_node); | |
1314 | free(garbage); | |
1315 | } | |
1316 | ||
1317 | shash_add(&all_dpif_backers, type, backer); | |
1318 | ||
1319 | error = dpif_recv_set(backer->dpif, true); | |
1320 | if (error) { | |
1321 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1322 | type, strerror(error)); | |
1323 | close_dpif_backer(backer); | |
1324 | return error; | |
1325 | } | |
1326 | ||
1327 | return error; | |
1328 | } | |
1329 | ||
abe529af | 1330 | static int |
0f5f95a9 | 1331 | construct(struct ofproto *ofproto_) |
abe529af BP |
1332 | { |
1333 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1334 | struct shash_node *node, *next; |
91858960 | 1335 | int max_ports; |
abe529af BP |
1336 | int error; |
1337 | int i; | |
1338 | ||
acf60855 | 1339 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1340 | if (error) { |
abe529af BP |
1341 | return error; |
1342 | } | |
1343 | ||
acf60855 | 1344 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
91858960 BP |
1345 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); |
1346 | ||
abe529af BP |
1347 | ofproto->netflow = NULL; |
1348 | ofproto->sflow = NULL; | |
29089a54 | 1349 | ofproto->ipfix = NULL; |
21f7563c | 1350 | ofproto->stp = NULL; |
abe529af | 1351 | hmap_init(&ofproto->bundles); |
e764773c | 1352 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
1353 | for (i = 0; i < MAX_MIRRORS; i++) { |
1354 | ofproto->mirrors[i] = NULL; | |
1355 | } | |
1356 | ofproto->has_bonded_bundles = false; | |
1357 | ||
abe529af | 1358 | hmap_init(&ofproto->facets); |
b0f7b9b5 | 1359 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 1360 | ofproto->governor = NULL; |
4d2a0f39 | 1361 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 BP |
1362 | |
1363 | for (i = 0; i < N_TABLES; i++) { | |
1364 | struct table_dpif *table = &ofproto->tables[i]; | |
1365 | ||
1366 | table->catchall_table = NULL; | |
1367 | table->other_table = NULL; | |
1368 | table->basis = random_uint32(); | |
1369 | } | |
abe529af | 1370 | |
7ee20df1 BP |
1371 | list_init(&ofproto->completions); |
1372 | ||
abe529af BP |
1373 | ofproto_dpif_unixctl_init(); |
1374 | ||
ccb7c863 | 1375 | ofproto->has_mirrors = false; |
daff3353 EJ |
1376 | ofproto->has_bundle_action = false; |
1377 | ||
52a90c29 BP |
1378 | hmap_init(&ofproto->vlandev_map); |
1379 | hmap_init(&ofproto->realdev_vid_map); | |
1380 | ||
acf60855 | 1381 | sset_init(&ofproto->ports); |
0a740f48 | 1382 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1383 | sset_init(&ofproto->port_poll_set); |
1384 | ofproto->port_poll_errno = 0; | |
1385 | ||
1386 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1387 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1388 | |
1389 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1390 | /* Check if the datapath already has this port. */ | |
1391 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1392 | sset_add(&ofproto->ports, node->name); | |
1393 | } | |
1394 | ||
1395 | free(iface_hint->br_name); | |
1396 | free(iface_hint->br_type); | |
4f9e08a5 | 1397 | free(iface_hint); |
acf60855 JP |
1398 | shash_delete(&init_ofp_ports, node); |
1399 | } | |
1400 | } | |
e1b1d06a | 1401 | |
b44a10b7 BP |
1402 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1403 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1404 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1405 | |
1406 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1407 | error = add_internal_flows(ofproto); |
1408 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1409 | ||
735d7efb AZ |
1410 | ofproto->n_hit = 0; |
1411 | ofproto->n_missed = 0; | |
1412 | ||
655ab909 AZ |
1413 | ofproto->max_n_subfacet = 0; |
1414 | ofproto->created = time_msec(); | |
1415 | ofproto->last_minute = ofproto->created; | |
1416 | memset(&ofproto->hourly, 0, sizeof ofproto->hourly); | |
1417 | memset(&ofproto->daily, 0, sizeof ofproto->daily); | |
1418 | ofproto->subfacet_add_count = 0; | |
1419 | ofproto->subfacet_del_count = 0; | |
1420 | ofproto->total_subfacet_add_count = 0; | |
1421 | ofproto->total_subfacet_del_count = 0; | |
1422 | ofproto->total_subfacet_life_span = 0; | |
1423 | ofproto->total_subfacet_count = 0; | |
1424 | ofproto->n_update_stats = 0; | |
1425 | ||
c57b2226 BP |
1426 | return error; |
1427 | } | |
1428 | ||
1429 | static int | |
1430 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1431 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1432 | { |
1433 | struct ofputil_flow_mod fm; | |
1434 | int error; | |
1435 | ||
81a76618 BP |
1436 | match_init_catchall(&fm.match); |
1437 | fm.priority = 0; | |
1438 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1439 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1440 | fm.cookie = htonll(0); |
1441 | fm.cookie_mask = htonll(0); | |
1442 | fm.table_id = TBL_INTERNAL; | |
1443 | fm.command = OFPFC_ADD; | |
1444 | fm.idle_timeout = 0; | |
1445 | fm.hard_timeout = 0; | |
1446 | fm.buffer_id = 0; | |
1447 | fm.out_port = 0; | |
1448 | fm.flags = 0; | |
f25d0cf3 BP |
1449 | fm.ofpacts = ofpacts->data; |
1450 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1451 | |
1452 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1453 | if (error) { | |
1454 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1455 | id, ofperr_to_string(error)); | |
1456 | return error; | |
1457 | } | |
1458 | ||
81a76618 | 1459 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); |
cb22974d | 1460 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1461 | |
abe529af BP |
1462 | return 0; |
1463 | } | |
1464 | ||
c57b2226 BP |
1465 | static int |
1466 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1467 | { | |
f25d0cf3 BP |
1468 | struct ofpact_controller *controller; |
1469 | uint64_t ofpacts_stub[128 / 8]; | |
1470 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1471 | int error; |
1472 | int id; | |
1473 | ||
f25d0cf3 | 1474 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1475 | id = 1; |
1476 | ||
f25d0cf3 BP |
1477 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1478 | controller->max_len = UINT16_MAX; | |
1479 | controller->controller_id = 0; | |
1480 | controller->reason = OFPR_NO_MATCH; | |
1481 | ofpact_pad(&ofpacts); | |
1482 | ||
1483 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1484 | if (error) { |
1485 | return error; | |
1486 | } | |
1487 | ||
f25d0cf3 BP |
1488 | ofpbuf_clear(&ofpacts); |
1489 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
1490 | &ofproto->no_packet_in_rule); |
1491 | return error; | |
1492 | } | |
1493 | ||
7ee20df1 BP |
1494 | static void |
1495 | complete_operations(struct ofproto_dpif *ofproto) | |
1496 | { | |
1497 | struct dpif_completion *c, *next; | |
1498 | ||
1499 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1500 | ofoperation_complete(c->op, 0); | |
1501 | list_remove(&c->list_node); | |
1502 | free(c); | |
1503 | } | |
1504 | } | |
1505 | ||
abe529af BP |
1506 | static void |
1507 | destruct(struct ofproto *ofproto_) | |
1508 | { | |
1509 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1510 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1511 | struct oftable *table; |
abe529af BP |
1512 | int i; |
1513 | ||
b44a10b7 | 1514 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1515 | complete_operations(ofproto); |
1516 | ||
0697b5c3 BP |
1517 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1518 | struct cls_cursor cursor; | |
1519 | ||
d0918789 | 1520 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1521 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1522 | ofproto_rule_destroy(&rule->up); | |
1523 | } | |
7ee20df1 BP |
1524 | } |
1525 | ||
abe529af BP |
1526 | for (i = 0; i < MAX_MIRRORS; i++) { |
1527 | mirror_destroy(ofproto->mirrors[i]); | |
1528 | } | |
1529 | ||
1530 | netflow_destroy(ofproto->netflow); | |
bae473fe | 1531 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
1532 | hmap_destroy(&ofproto->bundles); |
1533 | mac_learning_destroy(ofproto->ml); | |
1534 | ||
1535 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 1536 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 1537 | governor_destroy(ofproto->governor); |
abe529af | 1538 | |
52a90c29 BP |
1539 | hmap_destroy(&ofproto->vlandev_map); |
1540 | hmap_destroy(&ofproto->realdev_vid_map); | |
1541 | ||
acf60855 | 1542 | sset_destroy(&ofproto->ports); |
0a740f48 | 1543 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1544 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1545 | |
acf60855 | 1546 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1547 | } |
1548 | ||
1549 | static int | |
5fcc0d00 | 1550 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1551 | { |
1552 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 1553 | struct ofport_dpif *ofport; |
abe529af | 1554 | |
0aa66d6e EJ |
1555 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1556 | port_run_fast(ofport); | |
1557 | } | |
1558 | ||
5fcc0d00 BP |
1559 | return 0; |
1560 | } | |
1561 | ||
1562 | static int | |
1563 | run(struct ofproto *ofproto_) | |
1564 | { | |
1565 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1566 | struct ofport_dpif *ofport; | |
1567 | struct ofbundle *bundle; | |
1568 | int error; | |
1569 | ||
1570 | if (!clogged) { | |
1571 | complete_operations(ofproto); | |
1572 | } | |
5fcc0d00 BP |
1573 | |
1574 | error = run_fast(ofproto_); | |
1575 | if (error) { | |
1576 | return error; | |
abe529af BP |
1577 | } |
1578 | ||
abe529af | 1579 | if (ofproto->netflow) { |
6fca1ffb BP |
1580 | if (netflow_run(ofproto->netflow)) { |
1581 | send_netflow_active_timeouts(ofproto); | |
1582 | } | |
abe529af BP |
1583 | } |
1584 | if (ofproto->sflow) { | |
bae473fe | 1585 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1586 | } |
1587 | ||
1588 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1589 | port_run(ofport); | |
1590 | } | |
1591 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1592 | bundle_run(bundle); | |
1593 | } | |
1594 | ||
21f7563c | 1595 | stp_run(ofproto); |
2cc3c58e | 1596 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af | 1597 | |
6814e51f | 1598 | /* Check the consistency of a random facet, to aid debugging. */ |
4d2a0f39 EJ |
1599 | if (time_msec() >= ofproto->consistency_rl |
1600 | && !hmap_is_empty(&ofproto->facets) | |
2cc3c58e | 1601 | && !ofproto->backer->need_revalidate) { |
6814e51f BP |
1602 | struct facet *facet; |
1603 | ||
4d2a0f39 EJ |
1604 | ofproto->consistency_rl = time_msec() + 250; |
1605 | ||
6814e51f BP |
1606 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), |
1607 | struct facet, hmap_node); | |
2cc3c58e | 1608 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, |
bbafd73b | 1609 | facet->xout.tags)) { |
6814e51f | 1610 | if (!facet_check_consistency(facet)) { |
2cc3c58e | 1611 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1612 | } |
1613 | } | |
1614 | } | |
1615 | ||
9d6ac44e BP |
1616 | if (ofproto->governor) { |
1617 | size_t n_subfacets; | |
1618 | ||
1619 | governor_run(ofproto->governor); | |
1620 | ||
1621 | /* If the governor has shrunk to its minimum size and the number of | |
1622 | * subfacets has dwindled, then drop the governor entirely. | |
1623 | * | |
1624 | * For hysteresis, the number of subfacets to drop the governor is | |
1625 | * smaller than the number needed to trigger its creation. */ | |
1626 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1627 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1628 | && governor_is_idle(ofproto->governor)) { | |
1629 | governor_destroy(ofproto->governor); | |
1630 | ofproto->governor = NULL; | |
1631 | } | |
1632 | } | |
1633 | ||
abe529af BP |
1634 | return 0; |
1635 | } | |
1636 | ||
1637 | static void | |
1638 | wait(struct ofproto *ofproto_) | |
1639 | { | |
1640 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1641 | struct ofport_dpif *ofport; | |
1642 | struct ofbundle *bundle; | |
1643 | ||
7ee20df1 BP |
1644 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1645 | poll_immediate_wake(); | |
1646 | } | |
1647 | ||
acf60855 JP |
1648 | dpif_wait(ofproto->backer->dpif); |
1649 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1650 | if (ofproto->sflow) { |
bae473fe | 1651 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1652 | } |
2cc3c58e | 1653 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { |
abe529af BP |
1654 | poll_immediate_wake(); |
1655 | } | |
1656 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1657 | port_wait(ofport); | |
1658 | } | |
1659 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1660 | bundle_wait(bundle); | |
1661 | } | |
6fca1ffb BP |
1662 | if (ofproto->netflow) { |
1663 | netflow_wait(ofproto->netflow); | |
1664 | } | |
1c313b88 | 1665 | mac_learning_wait(ofproto->ml); |
21f7563c | 1666 | stp_wait(ofproto); |
2cc3c58e | 1667 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1668 | /* Shouldn't happen, but if it does just go around again. */ |
1669 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1670 | poll_immediate_wake(); | |
abe529af | 1671 | } |
9d6ac44e BP |
1672 | if (ofproto->governor) { |
1673 | governor_wait(ofproto->governor); | |
1674 | } | |
abe529af BP |
1675 | } |
1676 | ||
0d085684 BP |
1677 | static void |
1678 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1679 | { | |
1680 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1681 | ||
1682 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1683 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1684 | } | |
1685 | ||
abe529af BP |
1686 | static void |
1687 | flush(struct ofproto *ofproto_) | |
1688 | { | |
1689 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1690 | struct subfacet *subfacet, *next_subfacet; |
1691 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1692 | int n_batch; | |
b0f7b9b5 | 1693 | |
acf60855 JP |
1694 | n_batch = 0; |
1695 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
1696 | &ofproto->subfacets) { | |
1697 | if (subfacet->path != SF_NOT_INSTALLED) { | |
1698 | batch[n_batch++] = subfacet; | |
1699 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
1700 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1701 | n_batch = 0; | |
1702 | } | |
1703 | } else { | |
1704 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1705 | } |
abe529af | 1706 | } |
acf60855 JP |
1707 | |
1708 | if (n_batch > 0) { | |
1709 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1710 | } | |
abe529af BP |
1711 | } |
1712 | ||
6c1491fb BP |
1713 | static void |
1714 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1715 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1716 | { |
1717 | *arp_match_ip = true; | |
9e1fd49b BP |
1718 | *actions = (OFPUTIL_A_OUTPUT | |
1719 | OFPUTIL_A_SET_VLAN_VID | | |
1720 | OFPUTIL_A_SET_VLAN_PCP | | |
1721 | OFPUTIL_A_STRIP_VLAN | | |
1722 | OFPUTIL_A_SET_DL_SRC | | |
1723 | OFPUTIL_A_SET_DL_DST | | |
1724 | OFPUTIL_A_SET_NW_SRC | | |
1725 | OFPUTIL_A_SET_NW_DST | | |
1726 | OFPUTIL_A_SET_NW_TOS | | |
1727 | OFPUTIL_A_SET_TP_SRC | | |
1728 | OFPUTIL_A_SET_TP_DST | | |
1729 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1730 | } |
1731 | ||
1732 | static void | |
307975da | 1733 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1734 | { |
1735 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1736 | struct dpif_dp_stats s; |
7e741ffb JG |
1737 | uint64_t n_miss, n_no_pkt_in, n_bytes; |
1738 | uint64_t n_lookup; | |
6c1491fb BP |
1739 | |
1740 | strcpy(ots->name, "classifier"); | |
1741 | ||
acf60855 | 1742 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
7e741ffb JG |
1743 | rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); |
1744 | rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); | |
acf60855 | 1745 | |
7e741ffb JG |
1746 | n_lookup = s.n_hit + s.n_missed; |
1747 | ots->lookup_count = htonll(n_lookup); | |
1748 | ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); | |
6c1491fb BP |
1749 | } |
1750 | ||
abe529af BP |
1751 | static struct ofport * |
1752 | port_alloc(void) | |
1753 | { | |
1754 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1755 | return &port->up; | |
1756 | } | |
1757 | ||
1758 | static void | |
1759 | port_dealloc(struct ofport *port_) | |
1760 | { | |
1761 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1762 | free(port); | |
1763 | } | |
1764 | ||
1765 | static int | |
1766 | port_construct(struct ofport *port_) | |
1767 | { | |
1768 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1769 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1770 | const struct netdev *netdev = port->up.netdev; |
e1b1d06a JP |
1771 | struct dpif_port dpif_port; |
1772 | int error; | |
abe529af | 1773 | |
2cc3c58e | 1774 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1775 | port->bundle = NULL; |
1776 | port->cfm = NULL; | |
ccc09689 | 1777 | port->bfd = NULL; |
abe529af | 1778 | port->tag = tag_create_random(); |
d5ffa7f2 | 1779 | port->may_enable = true; |
21f7563c JP |
1780 | port->stp_port = NULL; |
1781 | port->stp_state = STP_DISABLED; | |
b9ad7294 | 1782 | port->tnl_port = NULL; |
8b36f51e | 1783 | hmap_init(&port->priorities); |
52a90c29 BP |
1784 | port->realdev_ofp_port = 0; |
1785 | port->vlandev_vid = 0; | |
b9ad7294 | 1786 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1787 | |
b9ad7294 | 1788 | if (netdev_vport_is_patch(netdev)) { |
743cea45 NM |
1789 | /* By bailing out here, we don't submit the port to the sFlow module |
1790 | * to be considered for counter polling export. This is correct | |
1791 | * because the patch port represents an interface that sFlow considers | |
1792 | * to be "internal" to the switch as a whole, and therefore not an | |
1793 | * candidate for counter polling. */ | |
0a740f48 EJ |
1794 | port->odp_port = OVSP_NONE; |
1795 | return 0; | |
1796 | } | |
1797 | ||
acf60855 | 1798 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
b9ad7294 | 1799 | netdev_vport_get_dpif_port(netdev), |
e1b1d06a JP |
1800 | &dpif_port); |
1801 | if (error) { | |
1802 | return error; | |
1803 | } | |
1804 | ||
1805 | port->odp_port = dpif_port.port_no; | |
1806 | ||
b9ad7294 EJ |
1807 | if (netdev_get_tunnel_config(netdev)) { |
1808 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1809 | } else { | |
1810 | /* Sanity-check that a mapping doesn't already exist. This | |
1811 | * shouldn't happen for non-tunnel ports. */ | |
1812 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1813 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1814 | dpif_port.name); | |
da78d43d | 1815 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1816 | return EBUSY; |
1817 | } | |
e1b1d06a | 1818 | |
b9ad7294 EJ |
1819 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
1820 | hash_int(port->odp_port, 0)); | |
1821 | } | |
da78d43d | 1822 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1823 | |
abe529af | 1824 | if (ofproto->sflow) { |
e1b1d06a | 1825 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1826 | } |
1827 | ||
1828 | return 0; | |
1829 | } | |
1830 | ||
1831 | static void | |
1832 | port_destruct(struct ofport *port_) | |
1833 | { | |
1834 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1835 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1836 | const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); |
02f8d646 | 1837 | const char *devname = netdev_get_name(port->up.netdev); |
abe529af | 1838 | |
a614d823 | 1839 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1840 | /* The underlying device is still there, so delete it. This |
1841 | * happens when the ofproto is being destroyed, since the caller | |
1842 | * assumes that removal of attached ports will happen as part of | |
1843 | * destruction. */ | |
a614d823 KM |
1844 | if (!port->tnl_port) { |
1845 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1846 | } | |
1847 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
acf60855 JP |
1848 | } |
1849 | ||
b9ad7294 | 1850 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { |
0a740f48 EJ |
1851 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1852 | } | |
1853 | ||
b9ad7294 | 1854 | tnl_port_del(port->tnl_port); |
02f8d646 | 1855 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1856 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
2cc3c58e | 1857 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1858 | bundle_remove(port_); |
a5610457 | 1859 | set_cfm(port_, NULL); |
8aee94b6 | 1860 | set_bfd(port_, NULL); |
abe529af | 1861 | if (ofproto->sflow) { |
bae473fe | 1862 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1863 | } |
8b36f51e EJ |
1864 | |
1865 | ofport_clear_priorities(port); | |
1866 | hmap_destroy(&port->priorities); | |
abe529af BP |
1867 | } |
1868 | ||
1869 | static void | |
1870 | port_modified(struct ofport *port_) | |
1871 | { | |
1872 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1873 | ||
1874 | if (port->bundle && port->bundle->bond) { | |
1875 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1876 | } | |
1877 | } | |
1878 | ||
1879 | static void | |
9e1fd49b | 1880 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1881 | { |
1882 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1883 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1884 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1885 | |
9e1fd49b | 1886 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1887 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1888 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1889 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1890 | |
9e1fd49b | 1891 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1892 | bundle_update(port->bundle); |
1893 | } | |
abe529af BP |
1894 | } |
1895 | } | |
1896 | ||
1897 | static int | |
1898 | set_sflow(struct ofproto *ofproto_, | |
1899 | const struct ofproto_sflow_options *sflow_options) | |
1900 | { | |
1901 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1902 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1903 | |
abe529af | 1904 | if (sflow_options) { |
bae473fe | 1905 | if (!ds) { |
abe529af BP |
1906 | struct ofport_dpif *ofport; |
1907 | ||
4213f19d | 1908 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1909 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1910 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1911 | } |
2cc3c58e | 1912 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1913 | } |
bae473fe | 1914 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1915 | } else { |
6ff686f2 PS |
1916 | if (ds) { |
1917 | dpif_sflow_destroy(ds); | |
2cc3c58e | 1918 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1919 | ofproto->sflow = NULL; |
1920 | } | |
abe529af BP |
1921 | } |
1922 | return 0; | |
1923 | } | |
1924 | ||
29089a54 RL |
1925 | static int |
1926 | set_ipfix( | |
1927 | struct ofproto *ofproto_, | |
1928 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1929 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1930 | size_t n_flow_exporters_options) | |
1931 | { | |
1932 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1933 | struct dpif_ipfix *di = ofproto->ipfix; | |
1934 | ||
1935 | if (bridge_exporter_options || flow_exporters_options) { | |
1936 | if (!di) { | |
1937 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1938 | } | |
1939 | dpif_ipfix_set_options( | |
1940 | di, bridge_exporter_options, flow_exporters_options, | |
1941 | n_flow_exporters_options); | |
1942 | } else { | |
1943 | if (di) { | |
1944 | dpif_ipfix_destroy(di); | |
1945 | ofproto->ipfix = NULL; | |
1946 | } | |
1947 | } | |
1948 | return 0; | |
1949 | } | |
1950 | ||
abe529af | 1951 | static int |
a5610457 | 1952 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1953 | { |
1954 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1955 | int error; | |
1956 | ||
a5610457 | 1957 | if (!s) { |
abe529af BP |
1958 | error = 0; |
1959 | } else { | |
1960 | if (!ofport->cfm) { | |
8c977421 EJ |
1961 | struct ofproto_dpif *ofproto; |
1962 | ||
1963 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1964 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
90967e95 | 1965 | ofport->cfm = cfm_create(ofport->up.netdev); |
abe529af BP |
1966 | } |
1967 | ||
a5610457 | 1968 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1969 | return 0; |
1970 | } | |
1971 | ||
1972 | error = EINVAL; | |
1973 | } | |
1974 | cfm_destroy(ofport->cfm); | |
1975 | ofport->cfm = NULL; | |
1976 | return error; | |
1977 | } | |
1978 | ||
9a9e3786 BP |
1979 | static bool |
1980 | get_cfm_status(const struct ofport *ofport_, | |
1981 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1982 | { |
1983 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1984 | ||
1985 | if (ofport->cfm) { | |
9a9e3786 BP |
1986 | status->faults = cfm_get_fault(ofport->cfm); |
1987 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1988 | status->health = cfm_get_health(ofport->cfm); | |
1989 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1990 | return true; | |
1de11730 | 1991 | } else { |
9a9e3786 | 1992 | return false; |
1de11730 EJ |
1993 | } |
1994 | } | |
ccc09689 EJ |
1995 | |
1996 | static int | |
1997 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
1998 | { | |
1999 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
2000 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2001 | struct bfd *old; | |
2002 | ||
2003 | old = ofport->bfd; | |
2004 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
2005 | if (ofport->bfd != old) { | |
2006 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2007 | } | |
2008 | ||
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | static int | |
2013 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
2014 | { | |
2015 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2016 | ||
2017 | if (ofport->bfd) { | |
2018 | bfd_get_status(ofport->bfd, smap); | |
2019 | return 0; | |
2020 | } else { | |
2021 | return ENOENT; | |
2022 | } | |
2023 | } | |
abe529af | 2024 | \f |
21f7563c JP |
2025 | /* Spanning Tree. */ |
2026 | ||
2027 | static void | |
2028 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2029 | { | |
2030 | struct ofproto_dpif *ofproto = ofproto_; | |
2031 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2032 | struct ofport_dpif *ofport; | |
2033 | ||
2034 | ofport = stp_port_get_aux(sp); | |
2035 | if (!ofport) { | |
2036 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2037 | ofproto->up.name, port_num); | |
2038 | } else { | |
2039 | struct eth_header *eth = pkt->l2; | |
2040 | ||
2041 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2042 | if (eth_addr_is_zero(eth->eth_src)) { | |
2043 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2044 | "with unknown MAC", ofproto->up.name, port_num); | |
2045 | } else { | |
97d6520b | 2046 | send_packet(ofport, pkt); |
21f7563c JP |
2047 | } |
2048 | } | |
2049 | ofpbuf_delete(pkt); | |
2050 | } | |
2051 | ||
2052 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2053 | static int | |
2054 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2055 | { | |
2056 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2057 | ||
2058 | /* Only revalidate flows if the configuration changed. */ | |
2059 | if (!s != !ofproto->stp) { | |
2cc3c58e | 2060 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
2061 | } |
2062 | ||
2063 | if (s) { | |
2064 | if (!ofproto->stp) { | |
2065 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2066 | send_bpdu_cb, ofproto); | |
2067 | ofproto->stp_last_tick = time_msec(); | |
2068 | } | |
2069 | ||
2070 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2071 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2072 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2073 | stp_set_max_age(ofproto->stp, s->max_age); | |
2074 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2075 | } else { | |
851bf71d EJ |
2076 | struct ofport *ofport; |
2077 | ||
2078 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2079 | set_stp_port(ofport, NULL); | |
2080 | } | |
2081 | ||
21f7563c JP |
2082 | stp_destroy(ofproto->stp); |
2083 | ofproto->stp = NULL; | |
2084 | } | |
2085 | ||
2086 | return 0; | |
2087 | } | |
2088 | ||
2089 | static int | |
2090 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2091 | { | |
2092 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2093 | ||
2094 | if (ofproto->stp) { | |
2095 | s->enabled = true; | |
2096 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2097 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2098 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2099 | } else { | |
2100 | s->enabled = false; | |
2101 | } | |
2102 | ||
2103 | return 0; | |
2104 | } | |
2105 | ||
2106 | static void | |
2107 | update_stp_port_state(struct ofport_dpif *ofport) | |
2108 | { | |
2109 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2110 | enum stp_state state; | |
2111 | ||
2112 | /* Figure out new state. */ | |
2113 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2114 | : STP_DISABLED; | |
2115 | ||
2116 | /* Update state. */ | |
2117 | if (ofport->stp_state != state) { | |
9e1fd49b | 2118 | enum ofputil_port_state of_state; |
21f7563c JP |
2119 | bool fwd_change; |
2120 | ||
2121 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2122 | netdev_get_name(ofport->up.netdev), | |
2123 | stp_state_name(ofport->stp_state), | |
2124 | stp_state_name(state)); | |
2125 | if (stp_learn_in_state(ofport->stp_state) | |
2126 | != stp_learn_in_state(state)) { | |
2127 | /* xxx Learning action flows should also be flushed. */ | |
2cc3c58e EJ |
2128 | mac_learning_flush(ofproto->ml, |
2129 | &ofproto->backer->revalidate_set); | |
21f7563c JP |
2130 | } |
2131 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2132 | != stp_forward_in_state(state); | |
2133 | ||
2cc3c58e | 2134 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
2135 | ofport->stp_state = state; |
2136 | ofport->stp_state_entered = time_msec(); | |
2137 | ||
b308140a | 2138 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
2139 | bundle_update(ofport->bundle); |
2140 | } | |
2141 | ||
2142 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
2143 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
2144 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2145 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2146 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2147 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2148 | : 0); | |
21f7563c JP |
2149 | ofproto_port_set_state(&ofport->up, of_state); |
2150 | } | |
2151 | } | |
2152 | ||
2153 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2154 | * caller is responsible for assigning STP port numbers and ensuring | |
2155 | * there are no duplicates. */ | |
2156 | static int | |
2157 | set_stp_port(struct ofport *ofport_, | |
2158 | const struct ofproto_port_stp_settings *s) | |
2159 | { | |
2160 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2161 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2162 | struct stp_port *sp = ofport->stp_port; | |
2163 | ||
2164 | if (!s || !s->enable) { | |
2165 | if (sp) { | |
2166 | ofport->stp_port = NULL; | |
2167 | stp_port_disable(sp); | |
ecd12731 | 2168 | update_stp_port_state(ofport); |
21f7563c JP |
2169 | } |
2170 | return 0; | |
2171 | } else if (sp && stp_port_no(sp) != s->port_num | |
2172 | && ofport == stp_port_get_aux(sp)) { | |
2173 | /* The port-id changed, so disable the old one if it's not | |
2174 | * already in use by another port. */ | |
2175 | stp_port_disable(sp); | |
2176 | } | |
2177 | ||
2178 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2179 | stp_port_enable(sp); | |
2180 | ||
2181 | stp_port_set_aux(sp, ofport); | |
2182 | stp_port_set_priority(sp, s->priority); | |
2183 | stp_port_set_path_cost(sp, s->path_cost); | |
2184 | ||
2185 | update_stp_port_state(ofport); | |
2186 | ||
2187 | return 0; | |
2188 | } | |
2189 | ||
2190 | static int | |
2191 | get_stp_port_status(struct ofport *ofport_, | |
2192 | struct ofproto_port_stp_status *s) | |
2193 | { | |
2194 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2195 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2196 | struct stp_port *sp = ofport->stp_port; | |
2197 | ||
2198 | if (!ofproto->stp || !sp) { | |
2199 | s->enabled = false; | |
2200 | return 0; | |
2201 | } | |
2202 | ||
2203 | s->enabled = true; | |
2204 | s->port_id = stp_port_get_id(sp); | |
2205 | s->state = stp_port_get_state(sp); | |
2206 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2207 | s->role = stp_port_get_role(sp); | |
80740385 | 2208 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2209 | |
2210 | return 0; | |
2211 | } | |
2212 | ||
2213 | static void | |
2214 | stp_run(struct ofproto_dpif *ofproto) | |
2215 | { | |
2216 | if (ofproto->stp) { | |
2217 | long long int now = time_msec(); | |
2218 | long long int elapsed = now - ofproto->stp_last_tick; | |
2219 | struct stp_port *sp; | |
2220 | ||
2221 | if (elapsed > 0) { | |
2222 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2223 | ofproto->stp_last_tick = now; | |
2224 | } | |
2225 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2226 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2227 | ||
2228 | if (ofport) { | |
2229 | update_stp_port_state(ofport); | |
2230 | } | |
2231 | } | |
6ae50723 EJ |
2232 | |
2233 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2cc3c58e | 2234 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
6ae50723 | 2235 | } |
21f7563c JP |
2236 | } |
2237 | } | |
2238 | ||
2239 | static void | |
2240 | stp_wait(struct ofproto_dpif *ofproto) | |
2241 | { | |
2242 | if (ofproto->stp) { | |
2243 | poll_timer_wait(1000); | |
2244 | } | |
2245 | } | |
2246 | ||
2247 | /* Returns true if STP should process 'flow'. */ | |
2248 | static bool | |
2249 | stp_should_process_flow(const struct flow *flow) | |
2250 | { | |
2251 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2252 | } | |
2253 | ||
2254 | static void | |
2255 | stp_process_packet(const struct ofport_dpif *ofport, | |
2256 | const struct ofpbuf *packet) | |
2257 | { | |
2258 | struct ofpbuf payload = *packet; | |
2259 | struct eth_header *eth = payload.data; | |
2260 | struct stp_port *sp = ofport->stp_port; | |
2261 | ||
2262 | /* Sink packets on ports that have STP disabled when the bridge has | |
2263 | * STP enabled. */ | |
2264 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2265 | return; | |
2266 | } | |
2267 | ||
2268 | /* Trim off padding on payload. */ | |
c573540b BP |
2269 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
2270 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
2271 | } |
2272 | ||
2273 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2274 | stp_received_bpdu(sp, payload.data, payload.size); | |
2275 | } | |
2276 | } | |
2277 | \f | |
8b36f51e EJ |
2278 | static struct priority_to_dscp * |
2279 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2280 | { | |
2281 | struct priority_to_dscp *pdscp; | |
2282 | uint32_t hash; | |
2283 | ||
2284 | hash = hash_int(priority, 0); | |
2285 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2286 | if (pdscp->priority == priority) { | |
2287 | return pdscp; | |
2288 | } | |
2289 | } | |
2290 | return NULL; | |
2291 | } | |
2292 | ||
2293 | static void | |
2294 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2295 | { | |
2296 | struct priority_to_dscp *pdscp, *next; | |
2297 | ||
2298 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2299 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2300 | free(pdscp); | |
2301 | } | |
2302 | } | |
2303 | ||
2304 | static int | |
2305 | set_queues(struct ofport *ofport_, | |
2306 | const struct ofproto_port_queue *qdscp_list, | |
2307 | size_t n_qdscp) | |
2308 | { | |
2309 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2310 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2311 | struct hmap new = HMAP_INITIALIZER(&new); | |
2312 | size_t i; | |
2313 | ||
2314 | for (i = 0; i < n_qdscp; i++) { | |
2315 | struct priority_to_dscp *pdscp; | |
2316 | uint32_t priority; | |
2317 | uint8_t dscp; | |
2318 | ||
2319 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
acf60855 | 2320 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, |
8b36f51e EJ |
2321 | &priority)) { |
2322 | continue; | |
2323 | } | |
2324 | ||
2325 | pdscp = get_priority(ofport, priority); | |
2326 | if (pdscp) { | |
2327 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2328 | } else { | |
2329 | pdscp = xmalloc(sizeof *pdscp); | |
2330 | pdscp->priority = priority; | |
2331 | pdscp->dscp = dscp; | |
2cc3c58e | 2332 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2333 | } |
2334 | ||
2335 | if (pdscp->dscp != dscp) { | |
2336 | pdscp->dscp = dscp; | |
2cc3c58e | 2337 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2338 | } |
2339 | ||
2340 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2341 | } | |
2342 | ||
2343 | if (!hmap_is_empty(&ofport->priorities)) { | |
2344 | ofport_clear_priorities(ofport); | |
2cc3c58e | 2345 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2346 | } |
2347 | ||
2348 | hmap_swap(&new, &ofport->priorities); | |
2349 | hmap_destroy(&new); | |
2350 | ||
2351 | return 0; | |
2352 | } | |
2353 | \f | |
abe529af BP |
2354 | /* Bundles. */ |
2355 | ||
b44a10b7 BP |
2356 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2357 | * ofproto to revalidate every flow. | |
2358 | * | |
2359 | * Normally MAC learning entries are removed only from the ofproto associated | |
2360 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2361 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2362 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2363 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2364 | * with the host from which it migrated. */ | |
abe529af | 2365 | static void |
b44a10b7 | 2366 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2367 | { |
2368 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2369 | struct mac_learning *ml = ofproto->ml; | |
2370 | struct mac_entry *mac, *next_mac; | |
2371 | ||
2cc3c58e | 2372 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2373 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2374 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2375 | if (all_ofprotos) { |
2376 | struct ofproto_dpif *o; | |
2377 | ||
2378 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2379 | if (o != ofproto) { | |
2380 | struct mac_entry *e; | |
2381 | ||
2382 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2383 | NULL); | |
2384 | if (e) { | |
b44a10b7 BP |
2385 | mac_learning_expire(o->ml, e); |
2386 | } | |
2387 | } | |
2388 | } | |
2389 | } | |
2390 | ||
abe529af BP |
2391 | mac_learning_expire(ml, mac); |
2392 | } | |
2393 | } | |
2394 | } | |
2395 | ||
2396 | static struct ofbundle * | |
2397 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2398 | { | |
2399 | struct ofbundle *bundle; | |
2400 | ||
2401 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2402 | &ofproto->bundles) { | |
2403 | if (bundle->aux == aux) { | |
2404 | return bundle; | |
2405 | } | |
2406 | } | |
2407 | return NULL; | |
2408 | } | |
2409 | ||
2410 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2411 | * ones that are found to 'bundles'. */ | |
2412 | static void | |
2413 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2414 | void **auxes, size_t n_auxes, | |
2415 | struct hmapx *bundles) | |
2416 | { | |
2417 | size_t i; | |
2418 | ||
2419 | hmapx_init(bundles); | |
2420 | for (i = 0; i < n_auxes; i++) { | |
2421 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2422 | if (bundle) { | |
2423 | hmapx_add(bundles, bundle); | |
2424 | } | |
2425 | } | |
2426 | } | |
2427 | ||
7bde8dd8 JP |
2428 | static void |
2429 | bundle_update(struct ofbundle *bundle) | |
2430 | { | |
2431 | struct ofport_dpif *port; | |
2432 | ||
2433 | bundle->floodable = true; | |
2434 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2435 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2436 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2437 | bundle->floodable = false; |
2438 | break; | |
2439 | } | |
2440 | } | |
2441 | } | |
2442 | ||
abe529af BP |
2443 | static void |
2444 | bundle_del_port(struct ofport_dpif *port) | |
2445 | { | |
2446 | struct ofbundle *bundle = port->bundle; | |
2447 | ||
2cc3c58e | 2448 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2449 | |
abe529af BP |
2450 | list_remove(&port->bundle_node); |
2451 | port->bundle = NULL; | |
2452 | ||
2453 | if (bundle->lacp) { | |
2454 | lacp_slave_unregister(bundle->lacp, port); | |
2455 | } | |
2456 | if (bundle->bond) { | |
2457 | bond_slave_unregister(bundle->bond, port); | |
2458 | } | |
2459 | ||
7bde8dd8 | 2460 | bundle_update(bundle); |
abe529af BP |
2461 | } |
2462 | ||
2463 | static bool | |
213b00cf | 2464 | bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port, |
df53d41c | 2465 | struct lacp_slave_settings *lacp) |
abe529af BP |
2466 | { |
2467 | struct ofport_dpif *port; | |
2468 | ||
2469 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2470 | if (!port) { | |
2471 | return false; | |
2472 | } | |
2473 | ||
2474 | if (port->bundle != bundle) { | |
2cc3c58e | 2475 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2476 | if (port->bundle) { |
2477 | bundle_del_port(port); | |
2478 | } | |
2479 | ||
2480 | port->bundle = bundle; | |
2481 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2482 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2483 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2484 | bundle->floodable = false; |
2485 | } | |
2486 | } | |
2487 | if (lacp) { | |
2cc3c58e | 2488 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2489 | lacp_slave_register(bundle->lacp, port, lacp); |
2490 | } | |
2491 | ||
2492 | return true; | |
2493 | } | |
2494 | ||
2495 | static void | |
2496 | bundle_destroy(struct ofbundle *bundle) | |
2497 | { | |
2498 | struct ofproto_dpif *ofproto; | |
2499 | struct ofport_dpif *port, *next_port; | |
2500 | int i; | |
2501 | ||
2502 | if (!bundle) { | |
2503 | return; | |
2504 | } | |
2505 | ||
2506 | ofproto = bundle->ofproto; | |
2507 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2508 | struct ofmirror *m = ofproto->mirrors[i]; | |
2509 | if (m) { | |
2510 | if (m->out == bundle) { | |
2511 | mirror_destroy(m); | |
2512 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2513 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2cc3c58e | 2514 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2515 | } |
2516 | } | |
2517 | } | |
2518 | ||
2519 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2520 | bundle_del_port(port); | |
2521 | } | |
2522 | ||
b44a10b7 | 2523 | bundle_flush_macs(bundle, true); |
abe529af BP |
2524 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2525 | free(bundle->name); | |
2526 | free(bundle->trunks); | |
2527 | lacp_destroy(bundle->lacp); | |
2528 | bond_destroy(bundle->bond); | |
2529 | free(bundle); | |
2530 | } | |
2531 | ||
2532 | static int | |
2533 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2534 | const struct ofproto_bundle_settings *s) | |
2535 | { | |
2536 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2537 | bool need_flush = false; | |
abe529af BP |
2538 | struct ofport_dpif *port; |
2539 | struct ofbundle *bundle; | |
ecac4ebf BP |
2540 | unsigned long *trunks; |
2541 | int vlan; | |
abe529af BP |
2542 | size_t i; |
2543 | bool ok; | |
2544 | ||
2545 | if (!s) { | |
2546 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2547 | return 0; | |
2548 | } | |
2549 | ||
cb22974d BP |
2550 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2551 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2552 | |
2553 | bundle = bundle_lookup(ofproto, aux); | |
2554 | if (!bundle) { | |
2555 | bundle = xmalloc(sizeof *bundle); | |
2556 | ||
2557 | bundle->ofproto = ofproto; | |
2558 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2559 | hash_pointer(aux, 0)); | |
2560 | bundle->aux = aux; | |
2561 | bundle->name = NULL; | |
2562 | ||
2563 | list_init(&bundle->ports); | |
ecac4ebf | 2564 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2565 | bundle->vlan = -1; |
2566 | bundle->trunks = NULL; | |
5e9ceccd | 2567 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2568 | bundle->lacp = NULL; |
2569 | bundle->bond = NULL; | |
2570 | ||
2571 | bundle->floodable = true; | |
2572 | ||
2573 | bundle->src_mirrors = 0; | |
2574 | bundle->dst_mirrors = 0; | |
2575 | bundle->mirror_out = 0; | |
2576 | } | |
2577 | ||
2578 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2579 | free(bundle->name); | |
2580 | bundle->name = xstrdup(s->name); | |
2581 | } | |
2582 | ||
2583 | /* LACP. */ | |
2584 | if (s->lacp) { | |
2585 | if (!bundle->lacp) { | |
2cc3c58e | 2586 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2587 | bundle->lacp = lacp_create(); |
2588 | } | |
2589 | lacp_configure(bundle->lacp, s->lacp); | |
2590 | } else { | |
2591 | lacp_destroy(bundle->lacp); | |
2592 | bundle->lacp = NULL; | |
2593 | } | |
2594 | ||
2595 | /* Update set of ports. */ | |
2596 | ok = true; | |
2597 | for (i = 0; i < s->n_slaves; i++) { | |
2598 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2599 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2600 | ok = false; |
2601 | } | |
2602 | } | |
2603 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2604 | struct ofport_dpif *next_port; | |
2605 | ||
2606 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2607 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2608 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2609 | goto found; |
2610 | } | |
2611 | } | |
2612 | ||
2613 | bundle_del_port(port); | |
2614 | found: ; | |
2615 | } | |
2616 | } | |
cb22974d | 2617 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2618 | |
2619 | if (list_is_empty(&bundle->ports)) { | |
2620 | bundle_destroy(bundle); | |
2621 | return EINVAL; | |
2622 | } | |
2623 | ||
ecac4ebf | 2624 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2625 | if (s->vlan_mode != bundle->vlan_mode |
2626 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2627 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2628 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2629 | need_flush = true; |
2630 | } | |
2631 | ||
abe529af | 2632 | /* Set VLAN tag. */ |
ecac4ebf BP |
2633 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2634 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2635 | : 0); | |
2636 | if (vlan != bundle->vlan) { | |
2637 | bundle->vlan = vlan; | |
abe529af BP |
2638 | need_flush = true; |
2639 | } | |
2640 | ||
2641 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2642 | switch (s->vlan_mode) { |
2643 | case PORT_VLAN_ACCESS: | |
2644 | trunks = NULL; | |
2645 | break; | |
2646 | ||
2647 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2648 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2649 | break; |
2650 | ||
2651 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2652 | case PORT_VLAN_NATIVE_TAGGED: | |
2653 | if (vlan != 0 && (!s->trunks | |
2654 | || !bitmap_is_set(s->trunks, vlan) | |
2655 | || bitmap_is_set(s->trunks, 0))) { | |
2656 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2657 | if (s->trunks) { | |
2658 | trunks = bitmap_clone(s->trunks, 4096); | |
2659 | } else { | |
2660 | trunks = bitmap_allocate1(4096); | |
2661 | } | |
2662 | bitmap_set1(trunks, vlan); | |
2663 | bitmap_set0(trunks, 0); | |
2664 | } else { | |
ebc56baa | 2665 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2666 | } |
2667 | break; | |
2668 | ||
2669 | default: | |
2670 | NOT_REACHED(); | |
2671 | } | |
abe529af BP |
2672 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2673 | free(bundle->trunks); | |
ecac4ebf BP |
2674 | if (trunks == s->trunks) { |
2675 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2676 | } else { | |
2677 | bundle->trunks = trunks; | |
2678 | trunks = NULL; | |
2679 | } | |
abe529af BP |
2680 | need_flush = true; |
2681 | } | |
ecac4ebf BP |
2682 | if (trunks != s->trunks) { |
2683 | free(trunks); | |
2684 | } | |
abe529af BP |
2685 | |
2686 | /* Bonding. */ | |
2687 | if (!list_is_short(&bundle->ports)) { | |
2688 | bundle->ofproto->has_bonded_bundles = true; | |
2689 | if (bundle->bond) { | |
2690 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2691 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2692 | } |
2693 | } else { | |
2694 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2695 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2696 | } |
2697 | ||
2698 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2699 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2700 | } |
2701 | } else { | |
2702 | bond_destroy(bundle->bond); | |
2703 | bundle->bond = NULL; | |
2704 | } | |
2705 | ||
2706 | /* If we changed something that would affect MAC learning, un-learn | |
2707 | * everything on this port and force flow revalidation. */ | |
2708 | if (need_flush) { | |
b44a10b7 | 2709 | bundle_flush_macs(bundle, false); |
abe529af BP |
2710 | } |
2711 | ||
2712 | return 0; | |
2713 | } | |
2714 | ||
2715 | static void | |
2716 | bundle_remove(struct ofport *port_) | |
2717 | { | |
2718 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2719 | struct ofbundle *bundle = port->bundle; | |
2720 | ||
2721 | if (bundle) { | |
2722 | bundle_del_port(port); | |
2723 | if (list_is_empty(&bundle->ports)) { | |
2724 | bundle_destroy(bundle); | |
2725 | } else if (list_is_short(&bundle->ports)) { | |
2726 | bond_destroy(bundle->bond); | |
2727 | bundle->bond = NULL; | |
2728 | } | |
2729 | } | |
2730 | } | |
2731 | ||
2732 | static void | |
5f877369 | 2733 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2734 | { |
2735 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2736 | struct ofport_dpif *port = port_; | |
2737 | uint8_t ea[ETH_ADDR_LEN]; | |
2738 | int error; | |
2739 | ||
2740 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2741 | if (!error) { | |
abe529af | 2742 | struct ofpbuf packet; |
5f877369 | 2743 | void *packet_pdu; |
abe529af BP |
2744 | |
2745 | ofpbuf_init(&packet, 0); | |
2746 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2747 | pdu_size); |
2748 | memcpy(packet_pdu, pdu, pdu_size); | |
2749 | ||
97d6520b | 2750 | send_packet(port, &packet); |
abe529af BP |
2751 | ofpbuf_uninit(&packet); |
2752 | } else { | |
2753 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2754 | "%s (%s)", port->bundle->name, | |
2755 | netdev_get_name(port->up.netdev), strerror(error)); | |
2756 | } | |
2757 | } | |
2758 | ||
2759 | static void | |
2760 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2761 | { | |
2762 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2763 | int error, n_packets, n_errors; | |
2764 | struct mac_entry *e; | |
2765 | ||
2766 | error = n_packets = n_errors = 0; | |
2767 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2768 | if (e->port.p != bundle) { | |
ea131871 JG |
2769 | struct ofpbuf *learning_packet; |
2770 | struct ofport_dpif *port; | |
4dd1e3ca | 2771 | void *port_void; |
ea131871 JG |
2772 | int ret; |
2773 | ||
4dd1e3ca BP |
2774 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2775 | * struct ofport_dpif more effective. */ | |
2776 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2777 | e->mac, e->vlan, | |
2778 | &port_void); | |
2779 | port = port_void; | |
97d6520b | 2780 | ret = send_packet(port, learning_packet); |
ea131871 | 2781 | ofpbuf_delete(learning_packet); |
abe529af BP |
2782 | if (ret) { |
2783 | error = ret; | |
2784 | n_errors++; | |
2785 | } | |
2786 | n_packets++; | |
2787 | } | |
2788 | } | |
2789 | ||
2790 | if (n_errors) { | |
2791 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2792 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2793 | "packets, last error was: %s", | |
2794 | bundle->name, n_errors, n_packets, strerror(error)); | |
2795 | } else { | |
2796 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2797 | bundle->name, n_packets); | |
2798 | } | |
2799 | } | |
2800 | ||
2801 | static void | |
2802 | bundle_run(struct ofbundle *bundle) | |
2803 | { | |
2804 | if (bundle->lacp) { | |
2805 | lacp_run(bundle->lacp, send_pdu_cb); | |
2806 | } | |
2807 | if (bundle->bond) { | |
2808 | struct ofport_dpif *port; | |
2809 | ||
2810 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2811 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2812 | } |
2813 | ||
2cc3c58e | 2814 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, |
bdebeece | 2815 | lacp_status(bundle->lacp)); |
abe529af BP |
2816 | if (bond_should_send_learning_packets(bundle->bond)) { |
2817 | bundle_send_learning_packets(bundle); | |
2818 | } | |
2819 | } | |
2820 | } | |
2821 | ||
2822 | static void | |
2823 | bundle_wait(struct ofbundle *bundle) | |
2824 | { | |
2825 | if (bundle->lacp) { | |
2826 | lacp_wait(bundle->lacp); | |
2827 | } | |
2828 | if (bundle->bond) { | |
2829 | bond_wait(bundle->bond); | |
2830 | } | |
2831 | } | |
2832 | \f | |
2833 | /* Mirrors. */ | |
2834 | ||
2835 | static int | |
2836 | mirror_scan(struct ofproto_dpif *ofproto) | |
2837 | { | |
2838 | int idx; | |
2839 | ||
2840 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2841 | if (!ofproto->mirrors[idx]) { | |
2842 | return idx; | |
2843 | } | |
2844 | } | |
2845 | return -1; | |
2846 | } | |
2847 | ||
2848 | static struct ofmirror * | |
2849 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2850 | { | |
2851 | int i; | |
2852 | ||
2853 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2854 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2855 | if (mirror && mirror->aux == aux) { | |
2856 | return mirror; | |
2857 | } | |
2858 | } | |
2859 | ||
2860 | return NULL; | |
2861 | } | |
2862 | ||
9ba15e2a BP |
2863 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2864 | static void | |
2865 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2866 | { | |
2867 | int i; | |
2868 | ||
2869 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2870 | struct ofmirror *m = ofproto->mirrors[i]; | |
2871 | ||
2872 | if (m) { | |
2873 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2874 | } | |
2875 | } | |
2876 | ||
2877 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2878 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2879 | int j; | |
2880 | ||
2881 | if (!m1) { | |
2882 | continue; | |
2883 | } | |
2884 | ||
2885 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2886 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2887 | ||
edb0540b | 2888 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2889 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2890 | m2->dup_mirrors |= m1->dup_mirrors; | |
2891 | } | |
2892 | } | |
2893 | } | |
2894 | } | |
2895 | ||
abe529af BP |
2896 | static int |
2897 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2898 | const struct ofproto_mirror_settings *s) | |
2899 | { | |
2900 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2901 | mirror_mask_t mirror_bit; | |
2902 | struct ofbundle *bundle; | |
2903 | struct ofmirror *mirror; | |
2904 | struct ofbundle *out; | |
2905 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2906 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2907 | int out_vlan; | |
2908 | ||
2909 | mirror = mirror_lookup(ofproto, aux); | |
2910 | if (!s) { | |
2911 | mirror_destroy(mirror); | |
2912 | return 0; | |
2913 | } | |
2914 | if (!mirror) { | |
2915 | int idx; | |
2916 | ||
2917 | idx = mirror_scan(ofproto); | |
2918 | if (idx < 0) { | |
2919 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2920 | "cannot create %s", | |
2921 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2922 | return EFBIG; | |
2923 | } | |
2924 | ||
2925 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2926 | mirror->ofproto = ofproto; | |
2927 | mirror->idx = idx; | |
8b28d864 | 2928 | mirror->aux = aux; |
abe529af BP |
2929 | mirror->out_vlan = -1; |
2930 | mirror->name = NULL; | |
2931 | } | |
2932 | ||
2933 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2934 | free(mirror->name); | |
2935 | mirror->name = xstrdup(s->name); | |
2936 | } | |
2937 | ||
2938 | /* Get the new configuration. */ | |
2939 | if (s->out_bundle) { | |
2940 | out = bundle_lookup(ofproto, s->out_bundle); | |
2941 | if (!out) { | |
2942 | mirror_destroy(mirror); | |
2943 | return EINVAL; | |
2944 | } | |
2945 | out_vlan = -1; | |
2946 | } else { | |
2947 | out = NULL; | |
2948 | out_vlan = s->out_vlan; | |
2949 | } | |
2950 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2951 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2952 | ||
2953 | /* If the configuration has not changed, do nothing. */ | |
2954 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2955 | && hmapx_equals(&dsts, &mirror->dsts) | |
2956 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2957 | && mirror->out == out | |
2958 | && mirror->out_vlan == out_vlan) | |
2959 | { | |
2960 | hmapx_destroy(&srcs); | |
2961 | hmapx_destroy(&dsts); | |
2962 | return 0; | |
2963 | } | |
2964 | ||
2965 | hmapx_swap(&srcs, &mirror->srcs); | |
2966 | hmapx_destroy(&srcs); | |
2967 | ||
2968 | hmapx_swap(&dsts, &mirror->dsts); | |
2969 | hmapx_destroy(&dsts); | |
2970 | ||
2971 | free(mirror->vlans); | |
2972 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2973 | ||
2974 | mirror->out = out; | |
2975 | mirror->out_vlan = out_vlan; | |
2976 | ||
2977 | /* Update bundles. */ | |
2978 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2979 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2980 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2981 | bundle->src_mirrors |= mirror_bit; | |
2982 | } else { | |
2983 | bundle->src_mirrors &= ~mirror_bit; | |
2984 | } | |
2985 | ||
2986 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2987 | bundle->dst_mirrors |= mirror_bit; | |
2988 | } else { | |
2989 | bundle->dst_mirrors &= ~mirror_bit; | |
2990 | } | |
2991 | ||
2992 | if (mirror->out == bundle) { | |
2993 | bundle->mirror_out |= mirror_bit; | |
2994 | } else { | |
2995 | bundle->mirror_out &= ~mirror_bit; | |
2996 | } | |
2997 | } | |
2998 | ||
2cc3c58e | 2999 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 3000 | ofproto->has_mirrors = true; |
2cc3c58e EJ |
3001 | mac_learning_flush(ofproto->ml, |
3002 | &ofproto->backer->revalidate_set); | |
9ba15e2a | 3003 | mirror_update_dups(ofproto); |
abe529af BP |
3004 | |
3005 | return 0; | |
3006 | } | |
3007 | ||
3008 | static void | |
3009 | mirror_destroy(struct ofmirror *mirror) | |
3010 | { | |
3011 | struct ofproto_dpif *ofproto; | |
3012 | mirror_mask_t mirror_bit; | |
3013 | struct ofbundle *bundle; | |
ccb7c863 | 3014 | int i; |
abe529af BP |
3015 | |
3016 | if (!mirror) { | |
3017 | return; | |
3018 | } | |
3019 | ||
3020 | ofproto = mirror->ofproto; | |
2cc3c58e EJ |
3021 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3022 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
abe529af BP |
3023 | |
3024 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
3025 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3026 | bundle->src_mirrors &= ~mirror_bit; | |
3027 | bundle->dst_mirrors &= ~mirror_bit; | |
3028 | bundle->mirror_out &= ~mirror_bit; | |
3029 | } | |
3030 | ||
3031 | hmapx_destroy(&mirror->srcs); | |
3032 | hmapx_destroy(&mirror->dsts); | |
3033 | free(mirror->vlans); | |
3034 | ||
3035 | ofproto->mirrors[mirror->idx] = NULL; | |
3036 | free(mirror->name); | |
3037 | free(mirror); | |
9ba15e2a BP |
3038 | |
3039 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
3040 | |
3041 | ofproto->has_mirrors = false; | |
3042 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3043 | if (ofproto->mirrors[i]) { | |
3044 | ofproto->has_mirrors = true; | |
3045 | break; | |
3046 | } | |
3047 | } | |
abe529af BP |
3048 | } |
3049 | ||
9d24de3b JP |
3050 | static int |
3051 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
3052 | uint64_t *packets, uint64_t *bytes) | |
3053 | { | |
3054 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3055 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
3056 | ||
3057 | if (!mirror) { | |
3058 | *packets = *bytes = UINT64_MAX; | |
3059 | return 0; | |
3060 | } | |
3061 | ||
8844e035 EJ |
3062 | push_all_stats(); |
3063 | ||
9d24de3b JP |
3064 | *packets = mirror->packet_count; |
3065 | *bytes = mirror->byte_count; | |
3066 | ||
3067 | return 0; | |
3068 | } | |
3069 | ||
abe529af BP |
3070 | static int |
3071 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
3072 | { | |
3073 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3074 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2cc3c58e | 3075 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af BP |
3076 | } |
3077 | return 0; | |
3078 | } | |
3079 | ||
3080 | static bool | |
b4affc74 | 3081 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
3082 | { |
3083 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3084 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
3085 | return bundle && bundle->mirror_out != 0; | |
3086 | } | |
8402c74b SS |
3087 | |
3088 | static void | |
b53055f4 | 3089 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
3090 | { |
3091 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 3092 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 3093 | } |
e764773c BP |
3094 | |
3095 | static void | |
c4069512 BP |
3096 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
3097 | size_t max_entries) | |
e764773c BP |
3098 | { |
3099 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3100 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
c4069512 | 3101 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
e764773c | 3102 | } |
abe529af BP |
3103 | \f |
3104 | /* Ports. */ | |
3105 | ||
3106 | static struct ofport_dpif * | |
4acbc98d | 3107 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 3108 | { |
7df6a8bd BP |
3109 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
3110 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
3111 | } |
3112 | ||
3113 | static struct ofport_dpif * | |
4acbc98d | 3114 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af | 3115 | { |
7c33b188 JR |
3116 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
3117 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
3118 | } |
3119 | ||
3120 | static void | |
e1b1d06a JP |
3121 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
3122 | struct ofproto_port *ofproto_port, | |
abe529af BP |
3123 | struct dpif_port *dpif_port) |
3124 | { | |
3125 | ofproto_port->name = dpif_port->name; | |
3126 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 3127 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
3128 | } |
3129 | ||
0a740f48 EJ |
3130 | static struct ofport_dpif * |
3131 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
3132 | { | |
3133 | const struct ofproto_dpif *ofproto; | |
3134 | const char *peer; | |
3135 | ||
3136 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
3137 | if (!peer) { | |
3138 | return NULL; | |
3139 | } | |
3140 | ||
3141 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3142 | struct ofport *ofport; | |
3143 | ||
3144 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
3145 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
3146 | return ofport_dpif_cast(ofport); | |
3147 | } | |
3148 | } | |
3149 | return NULL; | |
3150 | } | |
3151 | ||
0aa66d6e EJ |
3152 | static void |
3153 | port_run_fast(struct ofport_dpif *ofport) | |
3154 | { | |
3155 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
3156 | struct ofpbuf packet; | |
3157 | ||
3158 | ofpbuf_init(&packet, 0); | |
3159 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
3160 | send_packet(ofport, &packet); | |
3161 | ofpbuf_uninit(&packet); | |
3162 | } | |
ccc09689 EJ |
3163 | |
3164 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
3165 | struct ofpbuf packet; | |
3166 | ||
3167 | ofpbuf_init(&packet, 0); | |
3168 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
3169 | send_packet(ofport, &packet); | |
3170 | ofpbuf_uninit(&packet); | |
3171 | } | |
0aa66d6e EJ |
3172 | } |
3173 | ||
abe529af BP |
3174 | static void |
3175 | port_run(struct ofport_dpif *ofport) | |
3176 | { | |
3e5b3fdb EJ |
3177 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
3178 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
3179 | bool enable = netdev_get_carrier(ofport->up.netdev); |
3180 | ||
3e5b3fdb EJ |
3181 | ofport->carrier_seq = carrier_seq; |
3182 | ||
0aa66d6e | 3183 | port_run_fast(ofport); |
b9ad7294 EJ |
3184 | |
3185 | if (ofport->tnl_port | |
3186 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
3187 | &ofport->tnl_port)) { | |
3188 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
3189 | } | |
3190 | ||
abe529af | 3191 | if (ofport->cfm) { |
4653c558 EJ |
3192 | int cfm_opup = cfm_get_opup(ofport->cfm); |
3193 | ||
abe529af | 3194 | cfm_run(ofport->cfm); |
4653c558 EJ |
3195 | enable = enable && !cfm_get_fault(ofport->cfm); |
3196 | ||
3197 | if (cfm_opup >= 0) { | |
3198 | enable = enable && cfm_opup; | |
3199 | } | |
abe529af | 3200 | } |
015e08bc | 3201 | |
ccc09689 EJ |
3202 | if (ofport->bfd) { |
3203 | bfd_run(ofport->bfd); | |
3204 | enable = enable && bfd_forwarding(ofport->bfd); | |
3205 | } | |
3206 | ||
015e08bc EJ |
3207 | if (ofport->bundle) { |
3208 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
3209 | if (carrier_changed) { |
3210 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3211 | } | |
015e08bc EJ |
3212 | } |
3213 | ||
daff3353 EJ |
3214 | if (ofport->may_enable != enable) { |
3215 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3216 | ||
3217 | if (ofproto->has_bundle_action) { | |
2cc3c58e | 3218 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3219 | } |
3220 | } | |
3221 | ||
015e08bc | 3222 | ofport->may_enable = enable; |
abe529af BP |
3223 | } |
3224 | ||
3225 | static void | |
3226 | port_wait(struct ofport_dpif *ofport) | |
3227 | { | |
3228 | if (ofport->cfm) { | |
3229 | cfm_wait(ofport->cfm); | |
3230 | } | |
ccc09689 EJ |
3231 | |
3232 | if (ofport->bfd) { | |
3233 | bfd_wait(ofport->bfd); | |
3234 | } | |
abe529af BP |
3235 | } |
3236 | ||
3237 | static int | |
3238 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3239 | struct ofproto_port *ofproto_port) | |
3240 | { | |
3241 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3242 | struct dpif_port dpif_port; | |
3243 | int error; | |
3244 | ||
0a740f48 EJ |
3245 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3246 | const char *type = netdev_get_type_from_name(devname); | |
3247 | ||
3248 | /* We may be called before ofproto->up.port_by_name is populated with | |
3249 | * the appropriate ofport. For this reason, we must get the name and | |
3250 | * type from the netdev layer directly. */ | |
3251 | if (type) { | |
3252 | const struct ofport *ofport; | |
3253 | ||
3254 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3255 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3256 | ofproto_port->name = xstrdup(devname); | |
3257 | ofproto_port->type = xstrdup(type); | |
3258 | return 0; | |
3259 | } | |
3260 | return ENODEV; | |
3261 | } | |
3262 | ||
acf60855 JP |
3263 | if (!sset_contains(&ofproto->ports, devname)) { |
3264 | return ENODEV; | |
3265 | } | |
3266 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3267 | devname, &dpif_port); | |
abe529af | 3268 | if (!error) { |
e1b1d06a | 3269 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3270 | } |
3271 | return error; | |
3272 | } | |
3273 | ||
3274 | static int | |
e1b1d06a | 3275 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3276 | { |
3277 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 EJ |
3278 | const char *dp_port_name = netdev_vport_get_dpif_port(netdev); |
3279 | const char *devname = netdev_get_name(netdev); | |
abe529af | 3280 | |
0a740f48 EJ |
3281 | if (netdev_vport_is_patch(netdev)) { |
3282 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3283 | return 0; | |
3284 | } | |
3285 | ||
b9ad7294 | 3286 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
7d82ab2e KM |
3287 | uint32_t port_no = UINT32_MAX; |
3288 | int error; | |
3289 | ||
3290 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3291 | if (error) { |
3292 | return error; | |
3293 | } | |
7d82ab2e KM |
3294 | if (netdev_get_tunnel_config(netdev)) { |
3295 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3296 | } | |
acf60855 | 3297 | } |
b9ad7294 EJ |
3298 | |
3299 | if (netdev_get_tunnel_config(netdev)) { | |
3300 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3301 | } else { |
3302 | sset_add(&ofproto->ports, devname); | |
3303 | } | |
3304 | return 0; | |
3305 | } | |
3306 | ||
abe529af BP |
3307 | static int |
3308 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3309 | { | |
3310 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3311 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3312 | int error = 0; |
abe529af | 3313 | |
b9ad7294 EJ |
3314 | if (!ofport) { |
3315 | return 0; | |
e1b1d06a | 3316 | } |
b9ad7294 EJ |
3317 | |
3318 | sset_find_and_delete(&ofproto->ghost_ports, | |
3319 | netdev_get_name(ofport->up.netdev)); | |
a614d823 KM |
3320 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3321 | if (!ofport->tnl_port) { | |
b9ad7294 EJ |
3322 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3323 | if (!error) { | |
abe529af BP |
3324 | /* The caller is going to close ofport->up.netdev. If this is a |
3325 | * bonded port, then the bond is using that netdev, so remove it | |
3326 | * from the bond. The client will need to reconfigure everything | |
3327 | * after deleting ports, so then the slave will get re-added. */ | |
3328 | bundle_remove(&ofport->up); | |
3329 | } | |
3330 | } | |
3331 | return error; | |
3332 | } | |
3333 | ||
6527c598 PS |
3334 | static int |
3335 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3336 | { | |
3337 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3338 | int error; | |
3339 | ||
8844e035 EJ |
3340 | push_all_stats(); |
3341 | ||
6527c598 PS |
3342 | error = netdev_get_stats(ofport->up.netdev, stats); |
3343 | ||
ee382d89 | 3344 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3345 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3346 | ||
3347 | /* ofproto->stats.tx_packets represents packets that we created | |
3348 | * internally and sent to some port (e.g. packets sent with | |
3349 | * send_packet()). Account for them as if they had come from | |
3350 | * OFPP_LOCAL and got forwarded. */ | |
3351 | ||
3352 | if (stats->rx_packets != UINT64_MAX) { | |
3353 | stats->rx_packets += ofproto->stats.tx_packets; | |
3354 | } | |
3355 | ||
3356 | if (stats->rx_bytes != UINT64_MAX) { | |
3357 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3358 | } | |
3359 | ||
3360 | /* ofproto->stats.rx_packets represents packets that were received on | |
3361 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3362 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3363 | |
3364 | if (stats->tx_packets != UINT64_MAX) { | |
3365 | stats->tx_packets += ofproto->stats.rx_packets; | |
3366 | } | |
3367 | ||
3368 | if (stats->tx_bytes != UINT64_MAX) { | |
3369 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3370 | } | |
3371 | } | |
3372 | ||
3373 | return error; | |
3374 | } | |
3375 | ||
abe529af | 3376 | struct port_dump_state { |
acf60855 JP |
3377 | uint32_t bucket; |
3378 | uint32_t offset; | |
0a740f48 | 3379 | bool ghost; |
da78d43d BP |
3380 | |
3381 | struct ofproto_port port; | |
3382 | bool has_port; | |
abe529af BP |
3383 | }; |
3384 | ||
3385 | static int | |
acf60855 | 3386 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3387 | { |
0a740f48 | 3388 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3389 | return 0; |
3390 | } | |
3391 | ||
3392 | static int | |
b9ad7294 | 3393 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3394 | struct ofproto_port *port) |
3395 | { | |
e1b1d06a | 3396 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3397 | struct port_dump_state *state = state_; |
0a740f48 | 3398 | const struct sset *sset; |
acf60855 | 3399 | struct sset_node *node; |
abe529af | 3400 | |
da78d43d BP |
3401 | if (state->has_port) { |
3402 | ofproto_port_destroy(&state->port); | |
3403 | state->has_port = false; | |
3404 | } | |
0a740f48 EJ |
3405 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3406 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3407 | int error; |
3408 | ||
da78d43d BP |
3409 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3410 | if (!error) { | |
3411 | *port = state->port; | |
3412 | state->has_port = true; | |
3413 | return 0; | |
3414 | } else if (error != ENODEV) { | |
acf60855 JP |
3415 | return error; |
3416 | } | |
abe529af | 3417 | } |
acf60855 | 3418 | |
0a740f48 EJ |
3419 | if (!state->ghost) { |
3420 | state->ghost = true; | |
3421 | state->bucket = 0; | |
3422 | state->offset = 0; | |
3423 | return port_dump_next(ofproto_, state_, port); | |
3424 | } | |
3425 | ||
acf60855 | 3426 | return EOF; |
abe529af BP |
3427 | } |
3428 | ||
3429 | static int | |
3430 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3431 | { | |
3432 | struct port_dump_state *state = state_; | |
3433 | ||
da78d43d BP |
3434 | if (state->has_port) { |
3435 | ofproto_port_destroy(&state->port); | |
3436 | } | |
abe529af BP |
3437 | free(state); |
3438 | return 0; | |
3439 | } | |
3440 | ||
3441 | static int | |
3442 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3443 | { | |
3444 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3445 | |
3446 | if (ofproto->port_poll_errno) { | |
3447 | int error = ofproto->port_poll_errno; | |
3448 | ofproto->port_poll_errno = 0; | |
3449 | return error; | |
3450 | } | |
3451 | ||
3452 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3453 | return EAGAIN; | |
3454 | } | |
3455 | ||
3456 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3457 | return 0; | |
abe529af BP |
3458 | } |
3459 | ||
3460 | static void | |
3461 | port_poll_wait(const struct ofproto *ofproto_) | |
3462 | { | |
3463 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3464 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3465 | } |
3466 | ||
3467 | static int | |
3468 | port_is_lacp_current(const struct ofport *ofport_) | |
3469 | { | |
3470 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3471 | return (ofport->bundle && ofport->bundle->lacp | |
3472 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3473 | : -1); | |
3474 | } | |
3475 | \f | |
3476 | /* Upcall handling. */ | |
3477 | ||
501f8d1f BP |
3478 | /* Flow miss batching. |
3479 | * | |
3480 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3481 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3482 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3483 | * more packets, plus possibly installing the flow in the dpif. | |
3484 | * | |
3485 | * So far we only batch the operations that affect flow setup time the most. | |
3486 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3487 | struct flow_miss { | |
3488 | struct hmap_node hmap_node; | |
acf60855 | 3489 | struct ofproto_dpif *ofproto; |
501f8d1f | 3490 | struct flow flow; |
b0f7b9b5 | 3491 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3492 | const struct nlattr *key; |
3493 | size_t key_len; | |
14f94f9a | 3494 | struct initial_vals initial_vals; |
501f8d1f | 3495 | struct list packets; |
6a7e895f | 3496 | enum dpif_upcall_type upcall_type; |
a088a1ff | 3497 | uint32_t odp_in_port; |
501f8d1f BP |
3498 | }; |
3499 | ||
3500 | struct flow_miss_op { | |
c2b565b5 | 3501 | struct dpif_op dpif_op; |
bbafd73b EJ |
3502 | |
3503 | uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ | |
3504 | struct xlate_out xout; | |
3505 | bool xout_garbage; /* 'xout' needs to be uninitialized? */ | |
501f8d1f BP |
3506 | }; |
3507 | ||
62cd7072 BP |
3508 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3509 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3510 | * configurations. */ |
62cd7072 | 3511 | static void |
a39edbd4 | 3512 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3513 | const struct flow *flow) |
62cd7072 BP |
3514 | { |
3515 | struct ofputil_packet_in pin; | |
3516 | ||
3e3252fa EJ |
3517 | pin.packet = packet->data; |
3518 | pin.packet_len = packet->size; | |
62cd7072 | 3519 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3520 | pin.controller_id = 0; |
54834960 EJ |
3521 | |
3522 | pin.table_id = 0; | |
3523 | pin.cookie = 0; | |
3524 | ||
62cd7072 | 3525 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3526 | |
3527 | flow_get_metadata(flow, &pin.fmd); | |
3528 | ||
d8653c38 | 3529 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3530 | } |
3531 | ||
6a7e895f | 3532 | static enum slow_path_reason |
abe529af | 3533 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
ffaef958 | 3534 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) |
abe529af | 3535 | { |
b6e001b6 | 3536 | if (!ofport) { |
6a7e895f | 3537 | return 0; |
ffaef958 | 3538 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 3539 | if (packet) { |
abe529af BP |
3540 | cfm_process_heartbeat(ofport->cfm, packet); |
3541 | } | |
6a7e895f | 3542 | return SLOW_CFM; |
ccc09689 EJ |
3543 | } else if (ofport->bfd && bfd_should_process_flow(flow)) { |
3544 | if (packet) { | |
3545 | bfd_process_packet(ofport->bfd, flow, packet); | |
3546 | } | |
3547 | return SLOW_BFD; | |
b6e001b6 EJ |
3548 | } else if (ofport->bundle && ofport->bundle->lacp |
3549 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3550 | if (packet) { | |
3551 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 3552 | } |
6a7e895f | 3553 | return SLOW_LACP; |
21f7563c JP |
3554 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
3555 | if (packet) { | |
3556 | stp_process_packet(ofport, packet); | |
3557 | } | |
6a7e895f | 3558 | return SLOW_STP; |
ffaef958 BP |
3559 | } else { |
3560 | return 0; | |
abe529af | 3561 | } |
abe529af BP |
3562 | } |
3563 | ||
501f8d1f | 3564 | static struct flow_miss * |
ddbc5954 JP |
3565 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3566 | const struct flow *flow, uint32_t hash) | |
abe529af | 3567 | { |
501f8d1f | 3568 | struct flow_miss *miss; |
abe529af | 3569 | |
501f8d1f | 3570 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3571 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3572 | return miss; |
3573 | } | |
3574 | } | |
abe529af | 3575 | |
b23cdad9 | 3576 | return NULL; |
501f8d1f | 3577 | } |
abe529af | 3578 | |
9d6ac44e BP |
3579 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3580 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3581 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3582 | * returned op->subfacet, and if anything needs to be freed after processing | |
3583 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3584 | static void |
9d6ac44e BP |
3585 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3586 | struct flow_miss_op *op) | |
501f8d1f | 3587 | { |
14f94f9a | 3588 | if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) { |
9d6ac44e BP |
3589 | /* This packet was received on a VLAN splinter port. We |
3590 | * added a VLAN to the packet to make the packet resemble | |
3591 | * the flow, but the actions were composed assuming that | |
3592 | * the packet contained no VLAN. So, we must remove the | |
3593 | * VLAN header from the packet before trying to execute the | |
3594 | * actions. */ | |
3595 | eth_pop_vlan(packet); | |
3596 | } | |
3597 | ||
bbafd73b | 3598 | op->xout_garbage = false; |
9d6ac44e BP |
3599 | op->dpif_op.type = DPIF_OP_EXECUTE; |
3600 | op->dpif_op.u.execute.key = miss->key; | |
3601 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3602 | op->dpif_op.u.execute.packet = packet; | |
3603 | } | |
3604 | ||
3605 | /* Helper for handle_flow_miss_without_facet() and | |
3606 | * handle_flow_miss_with_facet(). */ | |
3607 | static void | |
3608 | handle_flow_miss_common(struct rule_dpif *rule, | |
3609 | struct ofpbuf *packet, const struct flow *flow) | |
3610 | { | |
3611 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3612 | ||
9d6ac44e BP |
3613 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { |
3614 | /* | |
3615 | * Extra-special case for fail-open mode. | |
3616 | * | |
3617 | * We are in fail-open mode and the packet matched the fail-open | |
3618 | * rule, but we are connected to a controller too. We should send | |
3619 | * the packet up to the controller in the hope that it will try to | |
3620 | * set up a flow and thereby allow us to exit fail-open. | |
3621 | * | |
3622 | * See the top-level comment in fail-open.c for more information. | |
3623 | */ | |
3624 | send_packet_in_miss(ofproto, packet, flow); | |
3625 | } | |
3626 | } | |
3627 | ||
3628 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3629 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3630 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3631 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3632 | * the benefits, so when the datapath holds a large number of flows we impose | |
3633 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3634 | static bool | |
3635 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
3636 | struct flow_miss *miss, uint32_t hash) | |
3637 | { | |
3638 | if (!ofproto->governor) { | |
3639 | size_t n_subfacets; | |
3640 | ||
3641 | n_subfacets = hmap_count(&ofproto->subfacets); | |
3642 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
3643 | return true; | |
3644 | } | |
3645 | ||
3646 | ofproto->governor = governor_create(ofproto->up.name); | |
3647 | } | |
3648 | ||
3649 | return governor_should_install_flow(ofproto->governor, hash, | |
3650 | list_size(&miss->packets)); | |
3651 | } | |
3652 | ||
3653 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3654 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3655 | * increment '*n_ops'. */ | |
3656 | static void | |
3657 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
9d6ac44e BP |
3658 | struct flow_miss_op *ops, size_t *n_ops) |
3659 | { | |
4dff9097 | 3660 | struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow); |
a7752d4a | 3661 | long long int now = time_msec(); |
530a1d91 | 3662 | struct ofpbuf *packet; |
bbafd73b | 3663 | struct xlate_in xin; |
2b459b83 | 3664 | |
9d6ac44e BP |
3665 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
3666 | struct flow_miss_op *op = &ops[*n_ops]; | |
3667 | struct dpif_flow_stats stats; | |
abe529af | 3668 | |
9d6ac44e | 3669 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3670 | |
7820e253 EJ |
3671 | handle_flow_miss_common(rule, packet, &miss->flow); |
3672 | ||
a7752d4a | 3673 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 3674 | rule_credit_stats(rule, &stats); |
abe529af | 3675 | |
bbafd73b EJ |
3676 | xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals, |
3677 | rule, stats.tcp_flags, packet); | |
3678 | xin.resubmit_stats = &stats; | |
3679 | xlate_actions(&xin, &op->xout); | |
abe529af | 3680 | |
bbafd73b | 3681 | if (op->xout.odp_actions.size) { |
9d6ac44e BP |
3682 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
3683 | ||
3684 | init_flow_miss_execute_op(miss, packet, op); | |
bbafd73b EJ |
3685 | execute->actions = op->xout.odp_actions.data; |
3686 | execute->actions_len = op->xout.odp_actions.size; | |
3687 | op->xout_garbage = true; | |
9d6ac44e BP |
3688 | |
3689 | (*n_ops)++; | |
3690 | } else { | |
bbafd73b | 3691 | xlate_out_uninit(&op->xout); |
9d6ac44e | 3692 | } |
abe529af | 3693 | } |
9d6ac44e BP |
3694 | } |
3695 | ||
3696 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3697 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3698 | * | |
3699 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3700 | * This is really important only for new facets: if we just called time_msec() | |
3701 | * here, then the new subfacet or its packets could look (occasionally) as | |
3702 | * though it was used some time after the facet was used. That can make a | |
3703 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3704 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3705 | static void |
3706 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 3707 | long long int now, |
9d6ac44e BP |
3708 | struct flow_miss_op *ops, size_t *n_ops) |
3709 | { | |
6a7e895f BP |
3710 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3711 | enum subfacet_path want_path; | |
9d6ac44e BP |
3712 | struct subfacet *subfacet; |
3713 | struct ofpbuf *packet; | |
abe529af | 3714 | |
a088a1ff | 3715 | subfacet = subfacet_create(facet, miss, now); |
bbafd73b | 3716 | want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
b0f7b9b5 | 3717 | |
530a1d91 | 3718 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3719 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 BP |
3720 | struct dpif_flow_stats stats; |
3721 | ||
9d6ac44e | 3722 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 3723 | |
4dff9097 | 3724 | if (want_path != SF_FAST_PATH) { |
bbafd73b | 3725 | struct xlate_in xin; |
454a77e5 | 3726 | |
bbafd73b EJ |
3727 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, |
3728 | facet->rule, 0, packet); | |
3729 | xlate_actions_for_side_effects(&xin); | |
501f8d1f | 3730 | } |
67d91f78 | 3731 | |
459b16a1 | 3732 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 3733 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 3734 | |
bbafd73b | 3735 | if (facet->xout.odp_actions.size) { |
9d6ac44e | 3736 | struct dpif_execute *execute = &op->dpif_op.u.execute; |
8338659a | 3737 | |
9d6ac44e | 3738 | init_flow_miss_execute_op(miss, packet, op); |
bbafd73b EJ |
3739 | execute->actions = facet->xout.odp_actions.data, |
3740 | execute->actions_len = facet->xout.odp_actions.size; | |
9d6ac44e | 3741 | (*n_ops)++; |
5fe20d5d | 3742 | } |
501f8d1f BP |
3743 | } |
3744 | ||
6a7e895f | 3745 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { |
501f8d1f | 3746 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3747 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3748 | |
c84451a6 EJ |
3749 | subfacet->path = want_path; |
3750 | ||
bbafd73b | 3751 | op->xout_garbage = false; |
c2b565b5 | 3752 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
3753 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
3754 | put->key = miss->key; | |
3755 | put->key_len = miss->key_len; | |
6a7e895f | 3756 | if (want_path == SF_FAST_PATH) { |
bbafd73b EJ |
3757 | put->actions = facet->xout.odp_actions.data; |
3758 | put->actions_len = facet->xout.odp_actions.size; | |
6a7e895f | 3759 | } else { |
bbafd73b EJ |
3760 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
3761 | op->slow_stub, sizeof op->slow_stub, | |
6a7e895f BP |
3762 | &put->actions, &put->actions_len); |
3763 | } | |
501f8d1f BP |
3764 | put->stats = NULL; |
3765 | } | |
3766 | } | |
3767 | ||
acf60855 JP |
3768 | /* Handles flow miss 'miss'. May add any required datapath operations |
3769 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3770 | static void |
acf60855 JP |
3771 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3772 | size_t *n_ops) | |
9d6ac44e | 3773 | { |
acf60855 | 3774 | struct ofproto_dpif *ofproto = miss->ofproto; |
9d6ac44e | 3775 | struct facet *facet; |
459b16a1 | 3776 | long long int now; |
9d6ac44e BP |
3777 | uint32_t hash; |
3778 | ||
3779 | /* The caller must ensure that miss->hmap_node.hash contains | |
3780 | * flow_hash(miss->flow, 0). */ | |
3781 | hash = miss->hmap_node.hash; | |
3782 | ||
3783 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3784 | if (!facet) { | |
ba33dd03 EJ |
3785 | /* There does not exist a bijection between 'struct flow' and datapath |
3786 | * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental | |
3787 | * assumption used throughout the facet and subfacet handling code. | |
3788 | * Since we have to handle these misses in userspace anyway, we simply | |
3789 | * skip facet creation, avoiding the problem alltogether. */ | |
3790 | if (miss->key_fitness == ODP_FIT_TOO_LITTLE | |
3791 | || !flow_miss_should_make_facet(ofproto, miss, hash)) { | |
4dff9097 | 3792 | handle_flow_miss_without_facet(miss, ops, n_ops); |
9d6ac44e BP |
3793 | return; |
3794 | } | |
3795 | ||
4dff9097 | 3796 | facet = facet_create(miss, hash); |
459b16a1 BP |
3797 | now = facet->used; |
3798 | } else { | |
3799 | now = time_msec(); | |
9d6ac44e | 3800 | } |
459b16a1 | 3801 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3802 | } |
3803 | ||
8f73d537 EJ |
3804 | static struct drop_key * |
3805 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3806 | size_t key_len) | |
3807 | { | |
3808 | struct drop_key *drop_key; | |
3809 | ||
3810 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3811 | &backer->drop_keys) { | |
3812 | if (drop_key->key_len == key_len | |
3813 | && !memcmp(drop_key->key, key, key_len)) { | |
3814 | return drop_key; | |
3815 | } | |
3816 | } | |
3817 | return NULL; | |
3818 | } | |
3819 | ||
3820 | static void | |
3821 | drop_key_clear(struct dpif_backer *backer) | |
3822 | { | |
3823 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3824 | struct drop_key *drop_key, *next; | |
3825 | ||
3826 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3827 | int error; | |
3828 | ||
3829 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3830 | NULL); | |
3831 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3832 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3833 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3834 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3835 | ds_cstr(&ds)); | |
3836 | ds_destroy(&ds); | |
3837 | } | |
3838 | ||
3839 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3840 | free(drop_key->key); | |
3841 | free(drop_key); | |
3842 | } | |
3843 | } | |
3844 | ||
e09ee259 EJ |
3845 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3846 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3847 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3848 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3849 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3850 | * 'packet' ingressed. | |
e2a6ca36 | 3851 | * |
e09ee259 EJ |
3852 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3853 | * 'flow''s in_port to OFPP_NONE. | |
3854 | * | |
3855 | * This function does post-processing on data returned from | |
3856 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3857 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3858 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3859 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3860 | * a VLAN header onto 'packet' (if it is nonnull). | |
3861 | * | |
c3f6c502 JP |
3862 | * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' |
3863 | * to the VLAN TCI with which the packet was really received, that is, the | |
3864 | * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from | |
3865 | * the value returned in flow->vlan_tci only for packets received on | |
715b48aa | 3866 | * VLAN splinters.) |
e09ee259 | 3867 | * |
b9ad7294 EJ |
3868 | * Similarly, this function also includes some logic to help with tunnels. It |
3869 | * may modify 'flow' as necessary to make the tunneling implementation | |
3870 | * transparent to the upcall processing logic. | |
3871 | * | |
e09ee259 EJ |
3872 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3873 | * or some other positive errno if there are other problems. */ | |
3874 | static int | |
3875 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3876 | const struct nlattr *key, size_t key_len, | |
3877 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3878 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
14f94f9a | 3879 | struct initial_vals *initial_vals) |
e84173dc | 3880 | { |
e09ee259 EJ |
3881 | const struct ofport_dpif *port; |
3882 | enum odp_key_fitness fitness; | |
b9ad7294 | 3883 | int error = ENODEV; |
e09ee259 EJ |
3884 | |
3885 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3886 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3887 | error = EINVAL; |
3888 | goto exit; | |
3889 | } | |
3890 | ||
14f94f9a JP |
3891 | if (initial_vals) { |
3892 | initial_vals->vlan_tci = flow->vlan_tci; | |
e84173dc | 3893 | } |
e84173dc | 3894 | |
e09ee259 EJ |
3895 | if (odp_in_port) { |
3896 | *odp_in_port = flow->in_port; | |
3897 | } | |
3898 | ||
3abe1835 BP |
3899 | port = (tnl_port_should_receive(flow) |
3900 | ? ofport_dpif_cast(tnl_port_receive(flow)) | |
3901 | : odp_port_to_ofport(backer, flow->in_port)); | |
3902 | flow->in_port = port ? port->up.ofp_port : OFPP_NONE; | |
3903 | if (!port) { | |
3904 | goto exit; | |
3905 | } | |
3906 | ||
3907 | /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port | |
3908 | * it's theoretically possible that we'll receive an ofport belonging to an | |
3909 | * entirely different datapath. In practice, this can't happen because no | |
3910 | * platforms has two separate datapaths which each support tunneling. */ | |
3911 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3912 | ||
3913 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3914 | if (packet) { | |
3915 | /* Make the packet resemble the flow, so that it gets sent to | |
3916 | * an OpenFlow controller properly, so that it looks correct | |
3917 | * for sFlow, and so that flow_extract() will get the correct | |
3918 | * vlan_tci if it is called on 'packet'. | |
3919 | * | |
3920 | * The allocated space inside 'packet' probably also contains | |
3921 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3922 | * a struct dpif_upcall (see the large comment on that | |
3923 | * structure definition), so pushing data on 'packet' is in | |
3924 | * general not a good idea since it could overwrite 'key' or | |
3925 | * free it as a side effect. However, it's OK in this special | |
3926 | * case because we know that 'packet' is inside a Netlink | |
3927 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3928 | * "struct nlattr", which is fine since we don't need that | |
3929 | * header anymore. */ | |
3930 | eth_push_vlan(packet, flow->vlan_tci); | |
3931 | } | |
3932 | /* We can't reproduce 'key' from 'flow'. */ | |
3933 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 | 3934 | } |
e09ee259 | 3935 | error = 0; |
52a90c29 | 3936 | |
b9ad7294 EJ |
3937 | if (ofproto) { |
3938 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3939 | } | |
3940 | ||
e09ee259 EJ |
3941 | exit: |
3942 | if (fitnessp) { | |
3943 | *fitnessp = fitness; | |
3944 | } | |
3945 | return error; | |
e84173dc BP |
3946 | } |
3947 | ||
501f8d1f | 3948 | static void |
acf60855 | 3949 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3950 | size_t n_upcalls) |
3951 | { | |
3952 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3953 | struct flow_miss *miss; |
3954 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3955 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3956 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3957 | struct hmap todo; |
b23cdad9 | 3958 | int n_misses; |
501f8d1f BP |
3959 | size_t n_ops; |
3960 | size_t i; | |
3961 | ||
3962 | if (!n_upcalls) { | |
3963 | return; | |
3964 | } | |
3965 | ||
3966 | /* Construct the to-do list. | |
3967 | * | |
3968 | * This just amounts to extracting the flow from each packet and sticking | |
3969 | * the packets that have the same flow in the same "flow_miss" structure so | |
3970 | * that we can process them together. */ | |
3971 | hmap_init(&todo); | |
b23cdad9 | 3972 | n_misses = 0; |
501f8d1f | 3973 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3974 | struct flow_miss *miss = &misses[n_misses]; |
3975 | struct flow_miss *existing_miss; | |
acf60855 | 3976 | struct ofproto_dpif *ofproto; |
a088a1ff | 3977 | uint32_t odp_in_port; |
1d446463 | 3978 | struct flow flow; |
b23cdad9 | 3979 | uint32_t hash; |
e09ee259 | 3980 | int error; |
501f8d1f | 3981 | |
e09ee259 EJ |
3982 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
3983 | upcall->key_len, &flow, &miss->key_fitness, | |
14f94f9a | 3984 | &ofproto, &odp_in_port, &miss->initial_vals); |
e09ee259 | 3985 | if (error == ENODEV) { |
8f73d537 EJ |
3986 | struct drop_key *drop_key; |
3987 | ||
acf60855 JP |
3988 | /* Received packet on port for which we couldn't associate |
3989 | * an ofproto. This can happen if a port is removed while | |
3990 | * traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
3991 | * in case it happens frequently. Install a drop flow so |
3992 | * that future packets of the flow are inexpensively dropped | |
3993 | * in the kernel. */ | |
acf60855 JP |
3994 | VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, |
3995 | flow.in_port); | |
8f73d537 EJ |
3996 | |
3997 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
3998 | if (!drop_key) { | |
3999 | drop_key = xmalloc(sizeof *drop_key); | |
4000 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
4001 | drop_key->key_len = upcall->key_len; | |
4002 | ||
4003 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
4004 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
4005 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
4006 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
4007 | } | |
4008 | continue; | |
acf60855 | 4009 | } |
e09ee259 | 4010 | if (error) { |
b0f7b9b5 BP |
4011 | continue; |
4012 | } | |
735d7efb AZ |
4013 | |
4014 | ofproto->n_missed++; | |
72e8bf28 | 4015 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
1d446463 | 4016 | &flow.tunnel, flow.in_port, &miss->flow); |
501f8d1f | 4017 | |
501f8d1f | 4018 | /* Add other packets to a to-do list. */ |
b23cdad9 | 4019 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 4020 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
4021 | if (!existing_miss) { |
4022 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 4023 | miss->ofproto = ofproto; |
b23cdad9 BP |
4024 | miss->key = upcall->key; |
4025 | miss->key_len = upcall->key_len; | |
6a7e895f | 4026 | miss->upcall_type = upcall->type; |
a088a1ff | 4027 | miss->odp_in_port = odp_in_port; |
b23cdad9 BP |
4028 | list_init(&miss->packets); |
4029 | ||
4030 | n_misses++; | |
4031 | } else { | |
4032 | miss = existing_miss; | |
4033 | } | |
501f8d1f BP |
4034 | list_push_back(&miss->packets, &upcall->packet->list_node); |
4035 | } | |
4036 | ||
4037 | /* Process each element in the to-do list, constructing the set of | |
4038 | * operations to batch. */ | |
4039 | n_ops = 0; | |
33bb0caa | 4040 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 4041 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 4042 | } |
cb22974d | 4043 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
4044 | |
4045 | /* Execute batch. */ | |
4046 | for (i = 0; i < n_ops; i++) { | |
4047 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
4048 | } | |
acf60855 | 4049 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 4050 | |
c84451a6 | 4051 | /* Free memory. */ |
501f8d1f | 4052 | for (i = 0; i < n_ops; i++) { |
bbafd73b EJ |
4053 | if (flow_miss_ops[i].xout_garbage) { |
4054 | xlate_out_uninit(&flow_miss_ops[i].xout); | |
4055 | } | |
501f8d1f | 4056 | } |
33bb0caa | 4057 | hmap_destroy(&todo); |
abe529af BP |
4058 | } |
4059 | ||
29089a54 RL |
4060 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, |
4061 | IPFIX_UPCALL } | |
6a7e895f BP |
4062 | classify_upcall(const struct dpif_upcall *upcall) |
4063 | { | |
29089a54 | 4064 | size_t userdata_len; |
6a7e895f BP |
4065 | union user_action_cookie cookie; |
4066 | ||
4067 | /* First look at the upcall type. */ | |
4068 | switch (upcall->type) { | |
4069 | case DPIF_UC_ACTION: | |
4070 | break; | |
4071 | ||
4072 | case DPIF_UC_MISS: | |
4073 | return MISS_UPCALL; | |
4074 | ||
4075 | case DPIF_N_UC_TYPES: | |
4076 | default: | |
4077 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
4078 | return BAD_UPCALL; | |
4079 | } | |
4080 | ||
4081 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
4082 | if (!upcall->userdata) { |
4083 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
4084 | return BAD_UPCALL; | |
4085 | } | |
29089a54 RL |
4086 | userdata_len = nl_attr_get_size(upcall->userdata); |
4087 | if (userdata_len < sizeof cookie.type | |
4088 | || userdata_len > sizeof cookie) { | |
e995e3df | 4089 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", |
29089a54 | 4090 | userdata_len); |
e995e3df BP |
4091 | return BAD_UPCALL; |
4092 | } | |
29089a54 RL |
4093 | memset(&cookie, 0, sizeof cookie); |
4094 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
4095 | if (userdata_len == sizeof cookie.sflow | |
4096 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
6a7e895f | 4097 | return SFLOW_UPCALL; |
29089a54 RL |
4098 | } else if (userdata_len == sizeof cookie.slow_path |
4099 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
6a7e895f | 4100 | return MISS_UPCALL; |
29089a54 RL |
4101 | } else if (userdata_len == sizeof cookie.flow_sample |
4102 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
4103 | return FLOW_SAMPLE_UPCALL; | |
4104 | } else if (userdata_len == sizeof cookie.ipfix | |
4105 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
4106 | return IPFIX_UPCALL; | |
4107 | } else { | |
4108 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
4109 | " and size %zu", cookie.type, userdata_len); | |
6a7e895f BP |
4110 | return BAD_UPCALL; |
4111 | } | |
4112 | } | |
4113 | ||
abe529af | 4114 | static void |
acf60855 | 4115 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 4116 | const struct dpif_upcall *upcall) |
abe529af | 4117 | { |
acf60855 | 4118 | struct ofproto_dpif *ofproto; |
1673e0e4 | 4119 | union user_action_cookie cookie; |
e84173dc | 4120 | struct flow flow; |
e1b1d06a | 4121 | uint32_t odp_in_port; |
abe529af | 4122 | |
e09ee259 EJ |
4123 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
4124 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
4125 | || !ofproto->sflow) { | |
e84173dc BP |
4126 | return; |
4127 | } | |
4128 | ||
29089a54 RL |
4129 | memset(&cookie, 0, sizeof cookie); |
4130 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
e1b1d06a JP |
4131 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
4132 | odp_in_port, &cookie); | |
6ff686f2 PS |
4133 | } |
4134 | ||
29089a54 RL |
4135 | static void |
4136 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
4137 | const struct dpif_upcall *upcall) | |
4138 | { | |
4139 | struct ofproto_dpif *ofproto; | |
4140 | union user_action_cookie cookie; | |
4141 | struct flow flow; | |
4142 | ||
4143 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4144 | &flow, NULL, &ofproto, NULL, NULL) | |
4145 | || !ofproto->ipfix) { | |
4146 | return; | |
4147 | } | |
4148 | ||
4149 | memset(&cookie, 0, sizeof cookie); | |
4150 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
4151 | ||
4152 | /* The flow reflects exactly the contents of the packet. Sample | |
4153 | * the packet using it. */ | |
4154 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
4155 | cookie.flow_sample.collector_set_id, | |
4156 | cookie.flow_sample.probability, | |
4157 | cookie.flow_sample.obs_domain_id, | |
4158 | cookie.flow_sample.obs_point_id); | |
4159 | } | |
4160 | ||
4161 | static void | |
4162 | handle_ipfix_upcall(struct dpif_backer *backer, | |
4163 | const struct dpif_upcall *upcall) | |
4164 | { | |
4165 | struct ofproto_dpif *ofproto; | |
4166 | struct flow flow; | |
4167 | ||
4168 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4169 | &flow, NULL, &ofproto, NULL, NULL) | |
4170 | || !ofproto->ipfix) { | |
4171 | return; | |
4172 | } | |
4173 | ||
4174 | /* The flow reflects exactly the contents of the packet. Sample | |
4175 | * the packet using it. */ | |
4176 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
4177 | } | |
4178 | ||
9b16c439 | 4179 | static int |
acf60855 | 4180 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 4181 | { |
9b16c439 | 4182 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
4183 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
4184 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
4185 | int n_processed; | |
9b16c439 BP |
4186 | int n_misses; |
4187 | int i; | |
abe529af | 4188 | |
cb22974d | 4189 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 4190 | |
9b16c439 | 4191 | n_misses = 0; |
90a7c55e | 4192 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 4193 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 4194 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
4195 | int error; |
4196 | ||
90a7c55e BP |
4197 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
4198 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 4199 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 4200 | if (error) { |
90a7c55e | 4201 | ofpbuf_uninit(buf); |
9b16c439 BP |
4202 | break; |
4203 | } | |
4204 | ||
6a7e895f BP |
4205 | switch (classify_upcall(upcall)) { |
4206 | case MISS_UPCALL: | |
9b16c439 BP |
4207 | /* Handle it later. */ |
4208 | n_misses++; | |
4209 | break; | |
4210 | ||
6a7e895f | 4211 | case SFLOW_UPCALL: |
acf60855 | 4212 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
4213 | ofpbuf_uninit(buf); |
4214 | break; | |
4215 | ||
29089a54 RL |
4216 | case FLOW_SAMPLE_UPCALL: |
4217 | handle_flow_sample_upcall(backer, upcall); | |
4218 | ofpbuf_uninit(buf); | |
4219 | break; | |
4220 | ||
4221 | case IPFIX_UPCALL: | |
4222 | handle_ipfix_upcall(backer, upcall); | |
4223 | ofpbuf_uninit(buf); | |
4224 | break; | |
4225 | ||
6a7e895f BP |
4226 | case BAD_UPCALL: |
4227 | ofpbuf_uninit(buf); | |
9b16c439 BP |
4228 | break; |
4229 | } | |
abe529af | 4230 | } |
9b16c439 | 4231 | |
6a7e895f | 4232 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 4233 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
4234 | for (i = 0; i < n_misses; i++) { |
4235 | ofpbuf_uninit(&miss_bufs[i]); | |
4236 | } | |
9b16c439 | 4237 | |
90a7c55e | 4238 | return n_processed; |
abe529af BP |
4239 | } |
4240 | \f | |
4241 | /* Flow expiration. */ | |
4242 | ||
b0f7b9b5 | 4243 | static int subfacet_max_idle(const struct ofproto_dpif *); |
acf60855 | 4244 | static void update_stats(struct dpif_backer *); |
abe529af | 4245 | static void rule_expire(struct rule_dpif *); |
b0f7b9b5 | 4246 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
4247 | |
4248 | /* This function is called periodically by run(). Its job is to collect | |
4249 | * updates for the flows that have been installed into the datapath, most | |
4250 | * importantly when they last were used, and then use that information to | |
4251 | * expire flows that have not been used recently. | |
4252 | * | |
4253 | * Returns the number of milliseconds after which it should be called again. */ | |
4254 | static int | |
acf60855 | 4255 | expire(struct dpif_backer *backer) |
abe529af | 4256 | { |
acf60855 JP |
4257 | struct ofproto_dpif *ofproto; |
4258 | int max_idle = INT32_MAX; | |
abe529af | 4259 | |
8f73d537 EJ |
4260 | /* Periodically clear out the drop keys in an effort to keep them |
4261 | * relatively few. */ | |
4262 | drop_key_clear(backer); | |
4263 | ||
acf60855 JP |
4264 | /* Update stats for each flow in the backer. */ |
4265 | update_stats(backer); | |
abe529af | 4266 | |
acf60855 | 4267 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
e503cc19 | 4268 | struct rule *rule, *next_rule; |
acf60855 | 4269 | int dp_max_idle; |
abe529af | 4270 | |
acf60855 JP |
4271 | if (ofproto->backer != backer) { |
4272 | continue; | |
4273 | } | |
0697b5c3 | 4274 | |
655ab909 AZ |
4275 | /* Keep track of the max number of flows per ofproto_dpif. */ |
4276 | update_max_subfacet_count(ofproto); | |
4277 | ||
acf60855 JP |
4278 | /* Expire subfacets that have been idle too long. */ |
4279 | dp_max_idle = subfacet_max_idle(ofproto); | |
4280 | expire_subfacets(ofproto, dp_max_idle); | |
4281 | ||
4282 | max_idle = MIN(max_idle, dp_max_idle); | |
4283 | ||
4284 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4285 | * has passed. */ | |
e503cc19 SH |
4286 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4287 | &ofproto->up.expirable) { | |
4288 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4289 | } |
abe529af | 4290 | |
acf60855 JP |
4291 | /* All outstanding data in existing flows has been accounted, so it's a |
4292 | * good time to do bond rebalancing. */ | |
4293 | if (ofproto->has_bonded_bundles) { | |
4294 | struct ofbundle *bundle; | |
abe529af | 4295 | |
acf60855 JP |
4296 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4297 | if (bundle->bond) { | |
2cc3c58e | 4298 | bond_rebalance(bundle->bond, &backer->revalidate_set); |
acf60855 | 4299 | } |
abe529af BP |
4300 | } |
4301 | } | |
4302 | } | |
4303 | ||
acf60855 | 4304 | return MIN(max_idle, 1000); |
abe529af BP |
4305 | } |
4306 | ||
a218c879 BP |
4307 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4308 | * as 'subfacet''s statistics. */ | |
4309 | static void | |
4310 | update_subfacet_stats(struct subfacet *subfacet, | |
4311 | const struct dpif_flow_stats *stats) | |
4312 | { | |
4313 | struct facet *facet = subfacet->facet; | |
4314 | ||
4315 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
4316 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
4317 | facet->packet_count += extra; | |
4318 | } else { | |
4319 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4320 | } | |
4321 | ||
4322 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
4323 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
4324 | } else { | |
4325 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4326 | } | |
4327 | ||
4328 | subfacet->dp_packet_count = stats->n_packets; | |
4329 | subfacet->dp_byte_count = stats->n_bytes; | |
4330 | ||
4331 | facet->tcp_flags |= stats->tcp_flags; | |
4332 | ||
4333 | subfacet_update_time(subfacet, stats->used); | |
4334 | if (facet->accounted_bytes < facet->byte_count) { | |
4335 | facet_learn(facet); | |
4336 | facet_account(facet); | |
4337 | facet->accounted_bytes = facet->byte_count; | |
4338 | } | |
a218c879 BP |
4339 | } |
4340 | ||
4341 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4342 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4343 | static void | |
acf60855 | 4344 | delete_unexpected_flow(struct ofproto_dpif *ofproto, |
a218c879 BP |
4345 | const struct nlattr *key, size_t key_len) |
4346 | { | |
4347 | if (!VLOG_DROP_WARN(&rl)) { | |
4348 | struct ds s; | |
4349 | ||
4350 | ds_init(&s); | |
4351 | odp_flow_key_format(key, key_len, &s); | |
acf60855 | 4352 | VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); |
a218c879 BP |
4353 | ds_destroy(&s); |
4354 | } | |
4355 | ||
4356 | COVERAGE_INC(facet_unexpected); | |
acf60855 | 4357 | dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); |
a218c879 BP |
4358 | } |
4359 | ||
abe529af BP |
4360 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4361 | * | |
4362 | * This function also pushes statistics updates to rules which each facet | |
4363 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4364 | * facet changes the rule it resubmits into at some time in between | |
4365 | * update_stats() runs, it is possible that statistics accrued to the | |
4366 | * old rule will be incorrectly attributed to the new rule. This could be | |
4367 | * avoided by calling update_stats() whenever rules are created or | |
4368 | * deleted. However, the performance impact of making so many calls to the | |
4369 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4370 | * |
4371 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4372 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4373 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4374 | * the hit or miss counts of br1. | |
abe529af BP |
4375 | */ |
4376 | static void | |
acf60855 | 4377 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4378 | { |
4379 | const struct dpif_flow_stats *stats; | |
4380 | struct dpif_flow_dump dump; | |
4381 | const struct nlattr *key; | |
3af56aef | 4382 | struct ofproto_dpif *ofproto; |
abe529af BP |
4383 | size_t key_len; |
4384 | ||
acf60855 | 4385 | dpif_flow_dump_start(&dump, backer->dpif); |
abe529af | 4386 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
acf60855 | 4387 | struct flow flow; |
b0f7b9b5 | 4388 | struct subfacet *subfacet; |
b9ad7294 | 4389 | struct ofport_dpif *ofport; |
acf60855 | 4390 | uint32_t key_hash; |
abe529af | 4391 | |
58c6adda EJ |
4392 | if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, |
4393 | NULL, NULL)) { | |
acf60855 JP |
4394 | continue; |
4395 | } | |
4396 | ||
655ab909 AZ |
4397 | ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets); |
4398 | ofproto->n_update_stats++; | |
655ab909 | 4399 | |
b9ad7294 EJ |
4400 | ofport = get_ofp_port(ofproto, flow.in_port); |
4401 | if (ofport && ofport->tnl_port) { | |
4402 | netdev_vport_inc_rx(ofport->up.netdev, stats); | |
4403 | } | |
4404 | ||
acf60855 | 4405 | key_hash = odp_flow_key_hash(key, key_len); |
9566abf9 | 4406 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
6a7e895f BP |
4407 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4408 | case SF_FAST_PATH: | |
735d7efb AZ |
4409 | /* Update ofproto_dpif's hit count. */ |
4410 | if (stats->n_packets > subfacet->dp_packet_count) { | |
4411 | uint64_t delta = stats->n_packets - subfacet->dp_packet_count; | |
4412 | dpif_stats_update_hit_count(ofproto, delta); | |
4413 | } | |
4414 | ||
a218c879 | 4415 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4416 | break; |
4417 | ||
4418 | case SF_SLOW_PATH: | |
4419 | /* Stats are updated per-packet. */ | |
4420 | break; | |
4421 | ||
4422 | case SF_NOT_INSTALLED: | |
4423 | default: | |
acf60855 | 4424 | delete_unexpected_flow(ofproto, key, key_len); |
6a7e895f | 4425 | break; |
abe529af | 4426 | } |
8fa4d1d0 | 4427 | run_fast_rl(); |
abe529af BP |
4428 | } |
4429 | dpif_flow_dump_done(&dump); | |
3af56aef AZ |
4430 | |
4431 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
4432 | update_moving_averages(ofproto); | |
4433 | } | |
4434 | ||
abe529af BP |
4435 | } |
4436 | ||
4437 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4438 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4439 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4440 | * fold its statistic into its rule. */ | |
abe529af | 4441 | static int |
b0f7b9b5 | 4442 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
4443 | { |
4444 | /* | |
4445 | * Idle time histogram. | |
4446 | * | |
b0f7b9b5 BP |
4447 | * Most of the time a switch has a relatively small number of subfacets. |
4448 | * When this is the case we might as well keep statistics for all of them | |
4449 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4450 | * well. |
4451 | * | |
b0f7b9b5 | 4452 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4453 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4454 | * significant. However, with a large number of subfacets it is likely |
4455 | * that only a few of them are "heavy hitters" that consume a large amount | |
4456 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4457 | * the kernel and maintaining in userspaces; other subfacets we can | |
4458 | * discard. | |
abe529af BP |
4459 | * |
4460 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4461 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4462 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4463 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4464 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 4465 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4466 | * the most-recently-used bucket of subfacets is kept, so actually an |
4467 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4468 | * (though the next run will delete most of those unless they receive |
4469 | * additional data). | |
abe529af | 4470 | * |
b0f7b9b5 BP |
4471 | * This requires a second pass through the subfacets, in addition to the |
4472 | * pass made by update_stats(), because the former function never looks at | |
4473 | * uninstallable subfacets. | |
abe529af BP |
4474 | */ |
4475 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4476 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4477 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4478 | int total, subtotal, bucket; |
b0f7b9b5 | 4479 | struct subfacet *subfacet; |
abe529af BP |
4480 | long long int now; |
4481 | int i; | |
4482 | ||
b0f7b9b5 | 4483 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 4484 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
4485 | return N_BUCKETS * BUCKET_WIDTH; |
4486 | } | |
4487 | ||
4488 | /* Build histogram. */ | |
4489 | now = time_msec(); | |
b0f7b9b5 BP |
4490 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
4491 | long long int idle = now - subfacet->used; | |
abe529af BP |
4492 | int bucket = (idle <= 0 ? 0 |
4493 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4494 | : (unsigned int) idle / BUCKET_WIDTH); | |
4495 | buckets[bucket]++; | |
4496 | } | |
4497 | ||
4498 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4499 | subtotal = bucket = 0; |
4500 | do { | |
4501 | subtotal += buckets[bucket++]; | |
084f5290 SH |
4502 | } while (bucket < N_BUCKETS && |
4503 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
4504 | |
4505 | if (VLOG_IS_DBG_ENABLED()) { | |
4506 | struct ds s; | |
4507 | ||
4508 | ds_init(&s); | |
4509 | ds_put_cstr(&s, "keep"); | |
4510 | for (i = 0; i < N_BUCKETS; i++) { | |
4511 | if (i == bucket) { | |
4512 | ds_put_cstr(&s, ", drop"); | |
4513 | } | |
4514 | if (buckets[i]) { | |
4515 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4516 | } | |
4517 | } | |
4518 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
4519 | ds_destroy(&s); | |
4520 | } | |
4521 | ||
4522 | return bucket * BUCKET_WIDTH; | |
4523 | } | |
4524 | ||
abe529af | 4525 | static void |
b0f7b9b5 | 4526 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 4527 | { |
625b0720 BP |
4528 | /* Cutoff time for most flows. */ |
4529 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4530 | ||
4531 | /* We really want to keep flows for special protocols around, so use a more | |
4532 | * conservative cutoff. */ | |
4533 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4534 | |
b0f7b9b5 | 4535 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4536 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4537 | int n_batch; |
abe529af | 4538 | |
b99d3cee | 4539 | n_batch = 0; |
b0f7b9b5 BP |
4540 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
4541 | &ofproto->subfacets) { | |
625b0720 BP |
4542 | long long int cutoff; |
4543 | ||
bbafd73b EJ |
4544 | cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP |
4545 | | SLOW_STP) | |
625b0720 BP |
4546 | ? special_cutoff |
4547 | : normal_cutoff); | |
b0f7b9b5 | 4548 | if (subfacet->used < cutoff) { |
6a7e895f | 4549 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4550 | batch[n_batch++] = subfacet; |
1d85f9e5 JP |
4551 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
4552 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
b99d3cee BP |
4553 | n_batch = 0; |
4554 | } | |
4555 | } else { | |
4556 | subfacet_destroy(subfacet); | |
4557 | } | |
abe529af BP |
4558 | } |
4559 | } | |
b99d3cee BP |
4560 | |
4561 | if (n_batch > 0) { | |
1d85f9e5 | 4562 | subfacet_destroy_batch(ofproto, batch, n_batch); |
b99d3cee | 4563 | } |
abe529af BP |
4564 | } |
4565 | ||
4566 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4567 | * then delete it entirely. */ | |
4568 | static void | |
4569 | rule_expire(struct rule_dpif *rule) | |
4570 | { | |
abe529af BP |
4571 | struct facet *facet, *next_facet; |
4572 | long long int now; | |
4573 | uint8_t reason; | |
4574 | ||
e2a3d183 BP |
4575 | if (rule->up.pending) { |
4576 | /* We'll have to expire it later. */ | |
4577 | return; | |
4578 | } | |
4579 | ||
abe529af BP |
4580 | /* Has 'rule' expired? */ |
4581 | now = time_msec(); | |
4582 | if (rule->up.hard_timeout | |
308881af | 4583 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4584 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4585 | } else if (rule->up.idle_timeout |
1745cd08 | 4586 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4587 | reason = OFPRR_IDLE_TIMEOUT; |
4588 | } else { | |
4589 | return; | |
4590 | } | |
4591 | ||
4592 | COVERAGE_INC(ofproto_dpif_expired); | |
4593 | ||
4594 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4595 | * timeout, because that only happens when the rule has no facets left.) */ | |
4596 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 4597 | facet_remove(facet); |
abe529af BP |
4598 | } |
4599 | ||
4600 | /* Get rid of the rule. */ | |
4601 | ofproto_rule_expire(&rule->up, reason); | |
4602 | } | |
4603 | \f | |
4604 | /* Facets. */ | |
4605 | ||
4dff9097 | 4606 | /* Creates and returns a new facet based on 'miss'. |
abe529af BP |
4607 | * |
4608 | * The caller must already have determined that no facet with an identical | |
4dff9097 | 4609 | * 'miss->flow' exists in 'miss->ofproto'. |
f3827897 | 4610 | * |
4dff9097 | 4611 | * 'hash' must be the return value of flow_hash(miss->flow, 0). |
2b459b83 | 4612 | * |
b0f7b9b5 BP |
4613 | * The facet will initially have no subfacets. The caller should create (at |
4614 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4615 | static struct facet * |
4dff9097 | 4616 | facet_create(const struct flow_miss *miss, uint32_t hash) |
abe529af | 4617 | { |
4dff9097 | 4618 | struct ofproto_dpif *ofproto = miss->ofproto; |
bbafd73b | 4619 | struct xlate_in xin; |
abe529af BP |
4620 | struct facet *facet; |
4621 | ||
4622 | facet = xzalloc(sizeof *facet); | |
4623 | facet->used = time_msec(); | |
4dff9097 EJ |
4624 | facet->flow = miss->flow; |
4625 | facet->initial_vals = miss->initial_vals; | |
4626 | facet->rule = rule_dpif_lookup(ofproto, &facet->flow); | |
4627 | facet->learn_rl = time_msec() + 500; | |
4628 | ||
2b459b83 | 4629 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
4dff9097 | 4630 | list_push_back(&facet->rule->facets, &facet->list_node); |
b0f7b9b5 | 4631 | list_init(&facet->subfacets); |
abe529af BP |
4632 | netflow_flow_init(&facet->nf_flow); |
4633 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4634 | ||
bbafd73b EJ |
4635 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, |
4636 | facet->rule, 0, NULL); | |
4637 | xin.may_learn = true; | |
4638 | xlate_actions(&xin, &facet->xout); | |
4639 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
6cf474d7 | 4640 | |
abe529af BP |
4641 | return facet; |
4642 | } | |
4643 | ||
4644 | static void | |
4645 | facet_free(struct facet *facet) | |
4646 | { | |
4dff9097 | 4647 | if (facet) { |
bbafd73b | 4648 | xlate_out_uninit(&facet->xout); |
4dff9097 EJ |
4649 | free(facet); |
4650 | } | |
abe529af BP |
4651 | } |
4652 | ||
3d9e05f8 | 4653 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4654 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4655 | static bool |
4656 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4657 | const struct nlattr *odp_actions, size_t actions_len, | |
4658 | struct ofpbuf *packet) | |
4659 | { | |
4660 | struct odputil_keybuf keybuf; | |
4661 | struct ofpbuf key; | |
4662 | int error; | |
4663 | ||
6ff686f2 | 4664 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
4665 | odp_flow_key_from_flow(&key, flow, |
4666 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
80e5eed9 | 4667 | |
acf60855 | 4668 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4669 | odp_actions, actions_len, packet); |
6ff686f2 | 4670 | return !error; |
abe529af BP |
4671 | } |
4672 | ||
abe529af BP |
4673 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
4674 | * | |
4675 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4676 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4677 | * |
4678 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4679 | */ | |
4680 | static void | |
15baa734 | 4681 | facet_remove(struct facet *facet) |
abe529af | 4682 | { |
15baa734 | 4683 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4684 | struct subfacet *subfacet, *next_subfacet; |
4685 | ||
cb22974d | 4686 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4687 | |
4688 | /* First uninstall all of the subfacets to get final statistics. */ | |
4689 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4690 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4691 | } |
4692 | ||
4693 | /* Flush the final stats to the rule. | |
4694 | * | |
4695 | * This might require us to have at least one subfacet around so that we | |
4696 | * can use its actions for accounting in facet_account(), which is why we | |
4697 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4698 | facet_flush_stats(facet); |
551a2f6c BP |
4699 | |
4700 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4701 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4702 | &facet->subfacets) { | |
15baa734 | 4703 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4704 | } |
abe529af BP |
4705 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
4706 | list_remove(&facet->list_node); | |
4707 | facet_free(facet); | |
4708 | } | |
4709 | ||
3de9590b BP |
4710 | /* Feed information from 'facet' back into the learning table to keep it in |
4711 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4712 | static void |
3de9590b | 4713 | facet_learn(struct facet *facet) |
abe529af | 4714 | { |
15baa734 | 4715 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
62bd7349 | 4716 | long long int now = time_msec(); |
bbafd73b | 4717 | struct xlate_in xin; |
abe529af | 4718 | |
bbafd73b | 4719 | if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { |
6cf474d7 EJ |
4720 | return; |
4721 | } | |
4722 | ||
62bd7349 | 4723 | facet->learn_rl = now + 500; |
6cf474d7 | 4724 | |
bbafd73b EJ |
4725 | if (!facet->xout.has_learn |
4726 | && !facet->xout.has_normal | |
4727 | && (!facet->xout.has_fin_timeout | |
3de9590b | 4728 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { |
abe529af BP |
4729 | return; |
4730 | } | |
abe529af | 4731 | |
bbafd73b EJ |
4732 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, |
4733 | facet->rule, facet->tcp_flags, NULL); | |
4734 | xin.may_learn = true; | |
4735 | xlate_actions_for_side_effects(&xin); | |
3de9590b BP |
4736 | } |
4737 | ||
4738 | static void | |
4739 | facet_account(struct facet *facet) | |
4740 | { | |
4741 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3de9590b BP |
4742 | const struct nlattr *a; |
4743 | unsigned int left; | |
4744 | ovs_be16 vlan_tci; | |
4745 | uint64_t n_bytes; | |
abe529af | 4746 | |
bbafd73b | 4747 | if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
4748 | return; |
4749 | } | |
3de9590b | 4750 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4751 | |
4752 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4753 | * as a basis. We also need to track the actual VLAN on which the packet | |
4754 | * is going to be sent to ensure that it matches the one passed to | |
4755 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4756 | * hash bucket.) |
4757 | * | |
4758 | * We use the actions from an arbitrary subfacet because they should all | |
4759 | * be equally valid for our purpose. */ | |
d78be13b | 4760 | vlan_tci = facet->flow.vlan_tci; |
bbafd73b EJ |
4761 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, |
4762 | facet->xout.odp_actions.size) { | |
fea393b1 | 4763 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4764 | struct ofport_dpif *port; |
abe529af | 4765 | |
d78be13b | 4766 | switch (nl_attr_type(a)) { |
df2c07f4 | 4767 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
4768 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
4769 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 4770 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4771 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4772 | } |
d78be13b BP |
4773 | break; |
4774 | ||
fea393b1 BP |
4775 | case OVS_ACTION_ATTR_POP_VLAN: |
4776 | vlan_tci = htons(0); | |
d78be13b BP |
4777 | break; |
4778 | ||
fea393b1 BP |
4779 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4780 | vlan = nl_attr_get(a); | |
4781 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4782 | break; |
abe529af BP |
4783 | } |
4784 | } | |
4785 | } | |
4786 | ||
abe529af BP |
4787 | /* Returns true if the only action for 'facet' is to send to the controller. |
4788 | * (We don't report NetFlow expiration messages for such facets because they | |
4789 | * are just part of the control logic for the network, not real traffic). */ | |
4790 | static bool | |
4791 | facet_is_controller_flow(struct facet *facet) | |
4792 | { | |
f25d0cf3 BP |
4793 | if (facet) { |
4794 | const struct rule *rule = &facet->rule->up; | |
4795 | const struct ofpact *ofpacts = rule->ofpacts; | |
4796 | size_t ofpacts_len = rule->ofpacts_len; | |
4797 | ||
dd30ff28 BP |
4798 | if (ofpacts_len > 0 && |
4799 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4800 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4801 | return true; | |
4802 | } | |
4803 | } | |
4804 | return false; | |
abe529af BP |
4805 | } |
4806 | ||
4807 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4808 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4809 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4810 | * its packet and byte counts before this function is called. */ | |
4811 | static void | |
15baa734 | 4812 | facet_flush_stats(struct facet *facet) |
abe529af | 4813 | { |
15baa734 | 4814 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4815 | struct subfacet *subfacet; |
4816 | ||
4817 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4818 | ovs_assert(!subfacet->dp_byte_count); |
4819 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4820 | } |
abe529af BP |
4821 | |
4822 | facet_push_stats(facet); | |
3de9590b BP |
4823 | if (facet->accounted_bytes < facet->byte_count) { |
4824 | facet_account(facet); | |
4825 | facet->accounted_bytes = facet->byte_count; | |
4826 | } | |
abe529af BP |
4827 | |
4828 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4829 | struct ofexpired expired; | |
4830 | expired.flow = facet->flow; | |
4831 | expired.packet_count = facet->packet_count; | |
4832 | expired.byte_count = facet->byte_count; | |
4833 | expired.used = facet->used; | |
4834 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4835 | } | |
4836 | ||
abe529af BP |
4837 | /* Reset counters to prevent double counting if 'facet' ever gets |
4838 | * reinstalled. */ | |
bbb5d219 | 4839 | facet_reset_counters(facet); |
abe529af BP |
4840 | |
4841 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4842 | facet->tcp_flags = 0; |
abe529af BP |
4843 | } |
4844 | ||
4845 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4846 | * Returns it if found, otherwise a null pointer. | |
4847 | * | |
2b459b83 BP |
4848 | * 'hash' must be the return value of flow_hash(flow, 0). |
4849 | * | |
abe529af BP |
4850 | * The returned facet might need revalidation; use facet_lookup_valid() |
4851 | * instead if that is important. */ | |
4852 | static struct facet * | |
2b459b83 BP |
4853 | facet_find(struct ofproto_dpif *ofproto, |
4854 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
4855 | { |
4856 | struct facet *facet; | |
4857 | ||
2b459b83 | 4858 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
4859 | if (flow_equal(flow, &facet->flow)) { |
4860 | return facet; | |
4861 | } | |
4862 | } | |
4863 | ||
4864 | return NULL; | |
4865 | } | |
4866 | ||
4867 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4868 | * Returns it if found, otherwise a null pointer. | |
4869 | * | |
2b459b83 BP |
4870 | * 'hash' must be the return value of flow_hash(flow, 0). |
4871 | * | |
abe529af BP |
4872 | * The returned facet is guaranteed to be valid. */ |
4873 | static struct facet * | |
2b459b83 BP |
4874 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
4875 | uint32_t hash) | |
abe529af | 4876 | { |
c57b2226 | 4877 | struct facet *facet; |
abe529af | 4878 | |
c57b2226 | 4879 | facet = facet_find(ofproto, flow, hash); |
abe529af | 4880 | if (facet |
2cc3c58e EJ |
4881 | && (ofproto->backer->need_revalidate |
4882 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
bbafd73b | 4883 | facet->xout.tags)) |
5bf64ade | 4884 | && !facet_revalidate(facet)) { |
0305ce1f | 4885 | return NULL; |
abe529af BP |
4886 | } |
4887 | ||
4888 | return facet; | |
4889 | } | |
4890 | ||
6814e51f BP |
4891 | static bool |
4892 | facet_check_consistency(struct facet *facet) | |
4893 | { | |
4894 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4895 | ||
4896 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4897 | ||
bbafd73b EJ |
4898 | struct xlate_out xout; |
4899 | struct xlate_in xin; | |
050ac423 | 4900 | |
6814e51f | 4901 | struct rule_dpif *rule; |
6814e51f BP |
4902 | bool ok; |
4903 | ||
4904 | /* Check the rule for consistency. */ | |
c57b2226 | 4905 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
4dff9097 EJ |
4906 | if (rule != facet->rule) { |
4907 | if (!VLOG_DROP_WARN(&rl)) { | |
4908 | struct ds s = DS_EMPTY_INITIALIZER; | |
6814e51f | 4909 | |
c53e1132 BP |
4910 | flow_format(&s, &facet->flow); |
4911 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4912 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4913 | cls_rule_format(&facet->rule->up.cr, &s); | |
4914 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4915 | rule->up.table_id); | |
4916 | cls_rule_format(&rule->up.cr, &s); | |
4dff9097 | 4917 | ds_put_cstr(&s, ")\n"); |
6814e51f | 4918 | |
c53e1132 BP |
4919 | ds_destroy(&s); |
4920 | } | |
4dff9097 | 4921 | return false; |
6814e51f BP |
4922 | } |
4923 | ||
4924 | /* Check the datapath actions for consistency. */ | |
bbafd73b EJ |
4925 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule, |
4926 | 0, NULL); | |
4927 | xlate_actions(&xin, &xout); | |
6814e51f | 4928 | |
bbafd73b EJ |
4929 | ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) |
4930 | && facet->xout.slow == xout.slow; | |
4dff9097 EJ |
4931 | if (!ok && !VLOG_DROP_WARN(&rl)) { |
4932 | struct ds s = DS_EMPTY_INITIALIZER; | |
c53e1132 | 4933 | |
4dff9097 EJ |
4934 | flow_format(&s, &facet->flow); |
4935 | ds_put_cstr(&s, ": inconsistency in facet"); | |
c53e1132 | 4936 | |
bbafd73b | 4937 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
9616614b | 4938 | ds_put_cstr(&s, " (actions were: "); |
bbafd73b EJ |
4939 | format_odp_actions(&s, facet->xout.odp_actions.data, |
4940 | facet->xout.odp_actions.size); | |
9616614b | 4941 | ds_put_cstr(&s, ") (correct actions: "); |
bbafd73b EJ |
4942 | format_odp_actions(&s, xout.odp_actions.data, |
4943 | xout.odp_actions.size); | |
4dff9097 | 4944 | ds_put_cstr(&s, ")"); |
6814e51f | 4945 | } |
4dff9097 | 4946 | |
bbafd73b EJ |
4947 | if (facet->xout.slow != xout.slow) { |
4948 | ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); | |
4dff9097 EJ |
4949 | } |
4950 | ||
9616614b | 4951 | ds_destroy(&s); |
6814e51f | 4952 | } |
bbafd73b | 4953 | xlate_out_uninit(&xout); |
6814e51f BP |
4954 | |
4955 | return ok; | |
4956 | } | |
4957 | ||
15baa734 | 4958 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4959 | * |
4960 | * - If the rule found is different from 'facet''s current rule, moves | |
4961 | * 'facet' to the new rule and recompiles its actions. | |
4962 | * | |
4963 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4964 | * where it is and recompiles its actions anyway. |
4965 | * | |
4966 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
5bf64ade EJ |
4967 | * ofproto_receive(), 'facet' is removed. |
4968 | * | |
4969 | * Returns true if 'facet' is still valid. False if 'facet' was removed. */ | |
4970 | static bool | |
15baa734 | 4971 | facet_revalidate(struct facet *facet) |
abe529af | 4972 | { |
15baa734 | 4973 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af | 4974 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4975 | struct subfacet *subfacet; |
bbafd73b EJ |
4976 | struct xlate_out xout; |
4977 | struct xlate_in xin; | |
abe529af BP |
4978 | |
4979 | COVERAGE_INC(facet_revalidate); | |
4980 | ||
f231418e EJ |
4981 | /* Check that child subfacets still correspond to this facet. Tunnel |
4982 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4983 | * change. */ | |
4984 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4985 | struct ofproto_dpif *recv_ofproto; | |
4986 | struct flow recv_flow; | |
4987 | int error; | |
4988 | ||
4989 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
4990 | subfacet->key_len, &recv_flow, NULL, | |
4991 | &recv_ofproto, NULL, NULL); | |
4992 | if (error | |
4993 | || recv_ofproto != ofproto | |
4994 | || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { | |
4995 | facet_remove(facet); | |
5bf64ade | 4996 | return false; |
f231418e EJ |
4997 | } |
4998 | } | |
4999 | ||
c57b2226 | 5000 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 5001 | |
df2c07f4 | 5002 | /* Calculate new datapath actions. |
abe529af BP |
5003 | * |
5004 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
5005 | * emit a NetFlow expiration and, if so, we need to have the old state | |
5006 | * around to properly compose it. */ | |
bbafd73b EJ |
5007 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule, |
5008 | 0, NULL); | |
5009 | xlate_actions(&xin, &xout); | |
4dff9097 EJ |
5010 | |
5011 | /* A facet's slow path reason should only change under dramatic | |
5012 | * circumstances. Rather than try to update everything, it's simpler to | |
5013 | * remove the facet and start over. */ | |
bbafd73b | 5014 | if (facet->xout.slow != xout.slow) { |
4dff9097 | 5015 | facet_remove(facet); |
bbafd73b | 5016 | xlate_out_uninit(&xout); |
4dff9097 EJ |
5017 | return false; |
5018 | } | |
6a7e895f | 5019 | |
bbafd73b | 5020 | if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { |
4dff9097 EJ |
5021 | LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { |
5022 | if (subfacet->path == SF_FAST_PATH) { | |
5023 | struct dpif_flow_stats stats; | |
b95fc6ba | 5024 | |
bbafd73b | 5025 | subfacet_install(subfacet, &xout.odp_actions, &stats); |
4dff9097 | 5026 | subfacet_update_stats(subfacet, &stats); |
b95fc6ba | 5027 | } |
abe529af | 5028 | } |
b95fc6ba | 5029 | |
15baa734 | 5030 | facet_flush_stats(facet); |
4dff9097 | 5031 | |
bbafd73b EJ |
5032 | ofpbuf_clear(&facet->xout.odp_actions); |
5033 | ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, | |
5034 | xout.odp_actions.size); | |
abe529af BP |
5035 | } |
5036 | ||
5037 | /* Update 'facet' now that we've taken care of all the old state. */ | |
bbafd73b EJ |
5038 | facet->xout.tags = xout.tags; |
5039 | facet->xout.slow = xout.slow; | |
5040 | facet->xout.has_learn = xout.has_learn; | |
5041 | facet->xout.has_normal = xout.has_normal; | |
5042 | facet->xout.has_fin_timeout = xout.has_fin_timeout; | |
5043 | facet->xout.nf_output_iface = xout.nf_output_iface; | |
5044 | facet->xout.mirrors = xout.mirrors; | |
5045 | facet->nf_flow.output_iface = facet->xout.nf_output_iface; | |
6a7e895f | 5046 | |
abe529af BP |
5047 | if (facet->rule != new_rule) { |
5048 | COVERAGE_INC(facet_changed_rule); | |
5049 | list_remove(&facet->list_node); | |
5050 | list_push_back(&new_rule->facets, &facet->list_node); | |
5051 | facet->rule = new_rule; | |
5052 | facet->used = new_rule->up.created; | |
9d24de3b | 5053 | facet->prev_used = facet->used; |
abe529af | 5054 | } |
5bf64ade | 5055 | |
bbafd73b | 5056 | xlate_out_uninit(&xout); |
5bf64ade | 5057 | return true; |
abe529af BP |
5058 | } |
5059 | ||
5060 | /* Updates 'facet''s used time. Caller is responsible for calling | |
5061 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
5062 | static void | |
15baa734 | 5063 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 5064 | { |
15baa734 | 5065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
5066 | if (used > facet->used) { |
5067 | facet->used = used; | |
1745cd08 | 5068 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
5069 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
5070 | } | |
5071 | } | |
5072 | ||
bbb5d219 EJ |
5073 | static void |
5074 | facet_reset_counters(struct facet *facet) | |
5075 | { | |
5076 | facet->packet_count = 0; | |
5077 | facet->byte_count = 0; | |
9d24de3b JP |
5078 | facet->prev_packet_count = 0; |
5079 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
5080 | facet->accounted_bytes = 0; |
5081 | } | |
5082 | ||
abe529af BP |
5083 | static void |
5084 | facet_push_stats(struct facet *facet) | |
5085 | { | |
112bc5f4 | 5086 | struct dpif_flow_stats stats; |
abe529af | 5087 | |
cb22974d BP |
5088 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
5089 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
5090 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 5091 | |
112bc5f4 BP |
5092 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
5093 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
5094 | stats.used = facet->used; | |
5095 | stats.tcp_flags = 0; | |
abe529af | 5096 | |
112bc5f4 | 5097 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
5098 | facet->prev_packet_count = facet->packet_count; |
5099 | facet->prev_byte_count = facet->byte_count; | |
5100 | facet->prev_used = facet->used; | |
abe529af | 5101 | |
eafed69b | 5102 | rule_credit_stats(facet->rule, &stats); |
ac35f9c8 | 5103 | flow_push_stats(facet, &stats); |
9d24de3b JP |
5104 | |
5105 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
bbafd73b EJ |
5106 | facet->xout.mirrors, stats.n_packets, |
5107 | stats.n_bytes); | |
abe529af BP |
5108 | } |
5109 | } | |
5110 | ||
8844e035 | 5111 | static void |
4464589e | 5112 | push_all_stats__(bool run_fast) |
8844e035 EJ |
5113 | { |
5114 | static long long int rl = LLONG_MIN; | |
5115 | struct ofproto_dpif *ofproto; | |
5116 | ||
5117 | if (time_msec() < rl) { | |
5118 | return; | |
5119 | } | |
5120 | ||
5121 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5122 | struct facet *facet; | |
5123 | ||
5124 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
5125 | facet_push_stats(facet); | |
4464589e EJ |
5126 | if (run_fast) { |
5127 | run_fast_rl(); | |
5128 | } | |
8844e035 EJ |
5129 | } |
5130 | } | |
5131 | ||
5132 | rl = time_msec() + 100; | |
5133 | } | |
5134 | ||
4464589e EJ |
5135 | static void |
5136 | push_all_stats(void) | |
5137 | { | |
5138 | push_all_stats__(true); | |
5139 | } | |
5140 | ||
abe529af | 5141 | static void |
112bc5f4 | 5142 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 5143 | { |
112bc5f4 BP |
5144 | rule->packet_count += stats->n_packets; |
5145 | rule->byte_count += stats->n_bytes; | |
5146 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
5147 | } |
5148 | ||
ac35f9c8 JP |
5149 | /* Pushes flow statistics to the rules which 'facet->flow' resubmits |
5150 | * into given 'facet->rule''s actions and mirrors. */ | |
abe529af | 5151 | static void |
ac35f9c8 | 5152 | flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats) |
abe529af | 5153 | { |
ac35f9c8 | 5154 | struct rule_dpif *rule = facet->rule; |
abe529af | 5155 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
bbafd73b | 5156 | struct xlate_in xin; |
abe529af | 5157 | |
112bc5f4 | 5158 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 5159 | |
bbafd73b EJ |
5160 | xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule, |
5161 | 0, NULL); | |
5162 | xin.resubmit_stats = stats; | |
5163 | xlate_actions_for_side_effects(&xin); | |
abe529af BP |
5164 | } |
5165 | \f | |
b0f7b9b5 BP |
5166 | /* Subfacets. */ |
5167 | ||
5168 | static struct subfacet * | |
acf60855 | 5169 | subfacet_find(struct ofproto_dpif *ofproto, |
9566abf9 | 5170 | const struct nlattr *key, size_t key_len, uint32_t key_hash) |
b0f7b9b5 BP |
5171 | { |
5172 | struct subfacet *subfacet; | |
5173 | ||
5174 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
5175 | &ofproto->subfacets) { | |
9566abf9 EJ |
5176 | if (subfacet->key_len == key_len |
5177 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
5178 | return subfacet; |
5179 | } | |
5180 | } | |
5181 | ||
5182 | return NULL; | |
5183 | } | |
5184 | ||
5185 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
5186 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
5187 | * existing subfacet if there is one, otherwise creates and returns a | |
4dff9097 | 5188 | * new subfacet. */ |
b0f7b9b5 | 5189 | static struct subfacet * |
a088a1ff JP |
5190 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
5191 | long long int now) | |
b0f7b9b5 | 5192 | { |
15baa734 | 5193 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
a088a1ff JP |
5194 | enum odp_key_fitness key_fitness = miss->key_fitness; |
5195 | const struct nlattr *key = miss->key; | |
5196 | size_t key_len = miss->key_len; | |
5197 | uint32_t key_hash; | |
b0f7b9b5 BP |
5198 | struct subfacet *subfacet; |
5199 | ||
a088a1ff JP |
5200 | key_hash = odp_flow_key_hash(key, key_len); |
5201 | ||
3b145dd7 BP |
5202 | if (list_is_empty(&facet->subfacets)) { |
5203 | subfacet = &facet->one_subfacet; | |
5204 | } else { | |
9566abf9 | 5205 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
3b145dd7 BP |
5206 | if (subfacet) { |
5207 | if (subfacet->facet == facet) { | |
5208 | return subfacet; | |
5209 | } | |
5210 | ||
5211 | /* This shouldn't happen. */ | |
5212 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5213 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
5214 | } |
5215 | ||
3b145dd7 | 5216 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
5217 | } |
5218 | ||
b0f7b9b5 BP |
5219 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
5220 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
5221 | subfacet->facet = facet; | |
b0f7b9b5 | 5222 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
5223 | subfacet->key = xmemdup(key, key_len); |
5224 | subfacet->key_len = key_len; | |
459b16a1 | 5225 | subfacet->used = now; |
655ab909 | 5226 | subfacet->created = now; |
26cd7e34 BP |
5227 | subfacet->dp_packet_count = 0; |
5228 | subfacet->dp_byte_count = 0; | |
6a7e895f | 5229 | subfacet->path = SF_NOT_INSTALLED; |
a088a1ff | 5230 | subfacet->odp_in_port = miss->odp_in_port; |
b0f7b9b5 | 5231 | |
655ab909 | 5232 | ofproto->subfacet_add_count++; |
b0f7b9b5 BP |
5233 | return subfacet; |
5234 | } | |
5235 | ||
b0f7b9b5 BP |
5236 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
5237 | * its facet within 'ofproto', and frees it. */ | |
5238 | static void | |
15baa734 | 5239 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 5240 | { |
15baa734 BP |
5241 | struct facet *facet = subfacet->facet; |
5242 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5243 | ||
655ab909 AZ |
5244 | /* Update ofproto stats before uninstall the subfacet. */ |
5245 | ofproto->subfacet_del_count++; | |
5246 | ofproto->total_subfacet_life_span += (time_msec() - subfacet->created); | |
5247 | ||
15baa734 | 5248 | subfacet_uninstall(subfacet); |
b0f7b9b5 BP |
5249 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
5250 | list_remove(&subfacet->list_node); | |
5251 | free(subfacet->key); | |
26cd7e34 BP |
5252 | if (subfacet != &facet->one_subfacet) { |
5253 | free(subfacet); | |
5254 | } | |
b0f7b9b5 BP |
5255 | } |
5256 | ||
5257 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5258 | * last remaining subfacet in its facet destroys the facet too. */ | |
5259 | static void | |
15baa734 | 5260 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
5261 | { |
5262 | struct facet *facet = subfacet->facet; | |
5263 | ||
551a2f6c BP |
5264 | if (list_is_singleton(&facet->subfacets)) { |
5265 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 5266 | facet_remove(facet); |
551a2f6c | 5267 | } else { |
15baa734 | 5268 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
5269 | } |
5270 | } | |
5271 | ||
1d85f9e5 JP |
5272 | static void |
5273 | subfacet_destroy_batch(struct ofproto_dpif *ofproto, | |
5274 | struct subfacet **subfacets, int n) | |
5275 | { | |
1d85f9e5 JP |
5276 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
5277 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
5278 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
5279 | int i; | |
5280 | ||
5281 | for (i = 0; i < n; i++) { | |
5282 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
5283 | ops[i].u.flow_del.key = subfacets[i]->key; |
5284 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
5285 | ops[i].u.flow_del.stats = &stats[i]; |
5286 | opsp[i] = &ops[i]; | |
5287 | } | |
5288 | ||
acf60855 | 5289 | dpif_operate(ofproto->backer->dpif, opsp, n); |
1d85f9e5 JP |
5290 | for (i = 0; i < n; i++) { |
5291 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5292 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5293 | subfacet_destroy(subfacets[i]); | |
8fa4d1d0 | 5294 | run_fast_rl(); |
1d85f9e5 JP |
5295 | } |
5296 | } | |
5297 | ||
b0f7b9b5 BP |
5298 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5299 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5300 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5301 | * since 'subfacet' was last updated. | |
5302 | * | |
5303 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5304 | static int | |
4dff9097 EJ |
5305 | subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, |
5306 | struct dpif_flow_stats *stats) | |
b0f7b9b5 | 5307 | { |
15baa734 BP |
5308 | struct facet *facet = subfacet->facet; |
5309 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
bbafd73b | 5310 | enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; |
4dff9097 EJ |
5311 | const struct nlattr *actions = odp_actions->data; |
5312 | size_t actions_len = odp_actions->size; | |
5313 | ||
6a7e895f | 5314 | uint64_t slow_path_stub[128 / 8]; |
b0f7b9b5 | 5315 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
5316 | int ret; |
5317 | ||
5318 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5319 | if (stats) { | |
5320 | flags |= DPIF_FP_ZERO_STATS; | |
5321 | } | |
5322 | ||
6a7e895f | 5323 | if (path == SF_SLOW_PATH) { |
bbafd73b | 5324 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
6a7e895f BP |
5325 | slow_path_stub, sizeof slow_path_stub, |
5326 | &actions, &actions_len); | |
5327 | } | |
5328 | ||
9566abf9 EJ |
5329 | ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key, |
5330 | subfacet->key_len, actions, actions_len, stats); | |
b0f7b9b5 BP |
5331 | |
5332 | if (stats) { | |
5333 | subfacet_reset_dp_stats(subfacet, stats); | |
5334 | } | |
5335 | ||
6a7e895f BP |
5336 | if (!ret) { |
5337 | subfacet->path = path; | |
5338 | } | |
b0f7b9b5 BP |
5339 | return ret; |
5340 | } | |
5341 | ||
5342 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
5343 | static void | |
15baa734 | 5344 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5345 | { |
6a7e895f | 5346 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
5347 | struct rule_dpif *rule = subfacet->facet->rule; |
5348 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 | 5349 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5350 | int error; |
5351 | ||
9566abf9 EJ |
5352 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5353 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5354 | subfacet_reset_dp_stats(subfacet, &stats); |
5355 | if (!error) { | |
15baa734 | 5356 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5357 | } |
6a7e895f | 5358 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5359 | } else { |
cb22974d BP |
5360 | ovs_assert(subfacet->dp_packet_count == 0); |
5361 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5362 | } |
5363 | } | |
5364 | ||
5365 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5366 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5367 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5368 | * was reset in the datapath. 'stats' will be modified to include only | |
5369 | * statistics new since 'subfacet' was last updated. */ | |
5370 | static void | |
5371 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5372 | struct dpif_flow_stats *stats) | |
5373 | { | |
5374 | if (stats | |
5375 | && subfacet->dp_packet_count <= stats->n_packets | |
5376 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5377 | stats->n_packets -= subfacet->dp_packet_count; | |
5378 | stats->n_bytes -= subfacet->dp_byte_count; | |
5379 | } | |
5380 | ||
5381 | subfacet->dp_packet_count = 0; | |
5382 | subfacet->dp_byte_count = 0; | |
5383 | } | |
5384 | ||
5385 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
5386 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
5387 | static void | |
15baa734 | 5388 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
5389 | { |
5390 | if (used > subfacet->used) { | |
5391 | subfacet->used = used; | |
15baa734 | 5392 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
5393 | } |
5394 | } | |
5395 | ||
5396 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
5397 | * | |
5398 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5399 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5400 | * represents a packet that was sent by hand or if it represents statistics | |
5401 | * that have been cleared out of the datapath. */ | |
5402 | static void | |
15baa734 | 5403 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5404 | const struct dpif_flow_stats *stats) |
5405 | { | |
5406 | if (stats->n_packets || stats->used > subfacet->used) { | |
5407 | struct facet *facet = subfacet->facet; | |
5408 | ||
15baa734 | 5409 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
5410 | facet->packet_count += stats->n_packets; |
5411 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5412 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5413 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); |
5414 | } | |
5415 | } | |
5416 | \f | |
abe529af BP |
5417 | /* Rules. */ |
5418 | ||
5419 | static struct rule_dpif * | |
c57b2226 BP |
5420 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
5421 | { | |
c57b2226 BP |
5422 | struct rule_dpif *rule; |
5423 | ||
5424 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5425 | if (rule) { | |
5426 | return rule; | |
5427 | } | |
5428 | ||
c376f9a3 | 5429 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5430 | } |
5431 | ||
5432 | static struct rule_dpif * | |
5433 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5434 | uint8_t table_id) | |
abe529af | 5435 | { |
7257b535 BP |
5436 | struct cls_rule *cls_rule; |
5437 | struct classifier *cls; | |
5438 | ||
9cdaaebe BP |
5439 | if (table_id >= N_TABLES) { |
5440 | return NULL; | |
5441 | } | |
5442 | ||
d0918789 | 5443 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 5444 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
5445 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
5446 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
5447 | * are unavailable. */ | |
5448 | struct flow ofpc_normal_flow = *flow; | |
5449 | ofpc_normal_flow.tp_src = htons(0); | |
5450 | ofpc_normal_flow.tp_dst = htons(0); | |
5451 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
5452 | } else { | |
5453 | cls_rule = classifier_lookup(cls, flow); | |
5454 | } | |
5455 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5456 | } |
5457 | ||
c376f9a3 IY |
5458 | static struct rule_dpif * |
5459 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5460 | { | |
5461 | struct ofport_dpif *port; | |
5462 | ||
5463 | port = get_ofp_port(ofproto, flow->in_port); | |
5464 | if (!port) { | |
5465 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5466 | return ofproto->miss_rule; | |
5467 | } | |
5468 | ||
5469 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5470 | return ofproto->no_packet_in_rule; | |
5471 | } | |
5472 | return ofproto->miss_rule; | |
5473 | } | |
5474 | ||
7ee20df1 BP |
5475 | static void |
5476 | complete_operation(struct rule_dpif *rule) | |
5477 | { | |
5478 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5479 | ||
54a9cbc9 | 5480 | rule_invalidate(rule); |
7ee20df1 BP |
5481 | if (clogged) { |
5482 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5483 | c->op = rule->up.pending; | |
5484 | list_push_back(&ofproto->completions, &c->list_node); | |
5485 | } else { | |
5486 | ofoperation_complete(rule->up.pending, 0); | |
5487 | } | |
5488 | } | |
5489 | ||
abe529af BP |
5490 | static struct rule * |
5491 | rule_alloc(void) | |
5492 | { | |
5493 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5494 | return &rule->up; | |
5495 | } | |
5496 | ||
5497 | static void | |
5498 | rule_dealloc(struct rule *rule_) | |
5499 | { | |
5500 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5501 | free(rule); | |
5502 | } | |
5503 | ||
90bf1e07 | 5504 | static enum ofperr |
abe529af BP |
5505 | rule_construct(struct rule *rule_) |
5506 | { | |
5507 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5508 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 5509 | struct rule_dpif *victim; |
54a9cbc9 | 5510 | uint8_t table_id; |
abe529af | 5511 | |
abe529af BP |
5512 | rule->packet_count = 0; |
5513 | rule->byte_count = 0; | |
abe529af | 5514 | |
7ee20df1 BP |
5515 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
5516 | if (victim && !list_is_empty(&victim->facets)) { | |
5517 | struct facet *facet; | |
5518 | ||
5519 | rule->facets = victim->facets; | |
5520 | list_moved(&rule->facets); | |
5521 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
5522 | /* XXX: We're only clearing our local counters here. It's possible |
5523 | * that quite a few packets are unaccounted for in the datapath | |
5524 | * statistics. These will be accounted to the new rule instead of | |
5525 | * cleared as required. This could be fixed by clearing out the | |
5526 | * datapath statistics for this facet, but currently it doesn't | |
5527 | * seem worth it. */ | |
5528 | facet_reset_counters(facet); | |
7ee20df1 BP |
5529 | facet->rule = rule; |
5530 | } | |
5531 | } else { | |
5532 | /* Must avoid list_moved() in this case. */ | |
5533 | list_init(&rule->facets); | |
5534 | } | |
abe529af | 5535 | |
54a9cbc9 | 5536 | table_id = rule->up.table_id; |
5cb7a798 BP |
5537 | if (victim) { |
5538 | rule->tag = victim->tag; | |
5539 | } else if (table_id == 0) { | |
5540 | rule->tag = 0; | |
5541 | } else { | |
5542 | struct flow flow; | |
5543 | ||
5544 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5545 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5546 | ofproto->tables[table_id].basis); | |
5547 | } | |
54a9cbc9 | 5548 | |
7ee20df1 | 5549 | complete_operation(rule); |
abe529af BP |
5550 | return 0; |
5551 | } | |
5552 | ||
5553 | static void | |
5554 | rule_destruct(struct rule *rule_) | |
5555 | { | |
5556 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5557 | struct facet *facet, *next_facet; |
5558 | ||
abe529af | 5559 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 5560 | facet_revalidate(facet); |
abe529af | 5561 | } |
7ee20df1 BP |
5562 | |
5563 | complete_operation(rule); | |
abe529af BP |
5564 | } |
5565 | ||
5566 | static void | |
5567 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5568 | { | |
5569 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af | 5570 | |
4464589e EJ |
5571 | /* push_all_stats() can handle flow misses which, when using the learn |
5572 | * action, can cause rules to be added and deleted. This can corrupt our | |
5573 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5574 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5575 | push_all_stats__(false); | |
bf1e8ff9 | 5576 | |
abe529af BP |
5577 | /* Start from historical data for 'rule' itself that are no longer tracked |
5578 | * in facets. This counts, for example, facets that have expired. */ | |
5579 | *packets = rule->packet_count; | |
5580 | *bytes = rule->byte_count; | |
abe529af BP |
5581 | } |
5582 | ||
0a740f48 EJ |
5583 | static void |
5584 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5585 | struct ofpbuf *packet) | |
abe529af | 5586 | { |
abe529af | 5587 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
14f94f9a | 5588 | struct initial_vals initial_vals; |
112bc5f4 | 5589 | struct dpif_flow_stats stats; |
bbafd73b EJ |
5590 | struct xlate_out xout; |
5591 | struct xlate_in xin; | |
abe529af | 5592 | |
a7752d4a | 5593 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5594 | rule_credit_stats(rule, &stats); |
5595 | ||
14f94f9a | 5596 | initial_vals.vlan_tci = flow->vlan_tci; |
bbafd73b EJ |
5597 | xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags, |
5598 | packet); | |
5599 | xin.resubmit_stats = &stats; | |
5600 | xlate_actions(&xin, &xout); | |
112bc5f4 | 5601 | |
bbafd73b EJ |
5602 | execute_odp_actions(ofproto, flow, xout.odp_actions.data, |
5603 | xout.odp_actions.size, packet); | |
112bc5f4 | 5604 | |
bbafd73b | 5605 | xlate_out_uninit(&xout); |
0a740f48 | 5606 | } |
5bf0e941 | 5607 | |
0a740f48 EJ |
5608 | static enum ofperr |
5609 | rule_execute(struct rule *rule, const struct flow *flow, | |
5610 | struct ofpbuf *packet) | |
5611 | { | |
5612 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5613 | ofpbuf_delete(packet); | |
5bf0e941 | 5614 | return 0; |
abe529af BP |
5615 | } |
5616 | ||
7ee20df1 BP |
5617 | static void |
5618 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5619 | { |
5620 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5621 | |
5622 | complete_operation(rule); | |
abe529af BP |
5623 | } |
5624 | \f | |
97d6520b | 5625 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5626 | * May modify 'packet'. |
abe529af BP |
5627 | * Returns 0 if successful, otherwise a positive errno value. */ |
5628 | static int | |
52a90c29 | 5629 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5630 | { |
79e15b78 | 5631 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5632 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 | 5633 | struct ofpbuf key, odp_actions; |
0f49659a | 5634 | struct dpif_flow_stats stats; |
80e5eed9 | 5635 | struct odputil_keybuf keybuf; |
0f49659a | 5636 | struct ofpact_output output; |
bbafd73b EJ |
5637 | struct xlate_out xout; |
5638 | struct xlate_in xin; | |
80e5eed9 | 5639 | struct flow flow; |
abe529af BP |
5640 | int error; |
5641 | ||
b9ad7294 | 5642 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
0f49659a | 5643 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
b9ad7294 | 5644 | |
0f49659a EJ |
5645 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ |
5646 | flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow); | |
5647 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, | |
5648 | OFPP_LOCAL)); | |
5649 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 | 5650 | |
0f49659a EJ |
5651 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); |
5652 | output.port = ofport->up.ofp_port; | |
5653 | output.max_len = 0; | |
b9ad7294 | 5654 | |
bbafd73b EJ |
5655 | xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet); |
5656 | xin.ofpacts_len = sizeof output; | |
5657 | xin.ofpacts = &output.ofpact; | |
5658 | xin.resubmit_stats = &stats; | |
5659 | xlate_actions(&xin, &xout); | |
6ff686f2 | 5660 | |
acf60855 | 5661 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 | 5662 | key.data, key.size, |
bbafd73b | 5663 | xout.odp_actions.data, xout.odp_actions.size, |
abe529af | 5664 | packet); |
bbafd73b | 5665 | xlate_out_uninit(&xout); |
abe529af BP |
5666 | |
5667 | if (error) { | |
0f49659a EJ |
5668 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", |
5669 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
5670 | strerror(error)); | |
abe529af | 5671 | } |
79e15b78 EJ |
5672 | |
5673 | ofproto->stats.tx_packets++; | |
5674 | ofproto->stats.tx_bytes += packet->size; | |
abe529af BP |
5675 | return error; |
5676 | } | |
5677 | \f | |
df2c07f4 | 5678 | /* OpenFlow to datapath action translation. */ |
abe529af | 5679 | |
f03a84b9 | 5680 | static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *); |
f25d0cf3 | 5681 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
f03a84b9 EJ |
5682 | struct xlate_ctx *); |
5683 | static void xlate_normal(struct xlate_ctx *); | |
abe529af | 5684 | |
6a7e895f BP |
5685 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5686 | * The action will state 'slow' as the reason that the action is in the slow | |
5687 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5688 | * dump-flows" output to see why a flow is in the slow path.) | |
5689 | * | |
5690 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5691 | * 'stub_size' must be large enough for the action. | |
5692 | * | |
5693 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5694 | * respectively. */ | |
5695 | static void | |
5696 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5697 | enum slow_path_reason slow, | |
5698 | uint64_t *stub, size_t stub_size, | |
5699 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5700 | { | |
5701 | union user_action_cookie cookie; | |
5702 | struct ofpbuf buf; | |
5703 | ||
5704 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5705 | cookie.slow_path.unused = 0; | |
5706 | cookie.slow_path.reason = slow; | |
5707 | ||
5708 | ofpbuf_use_stack(&buf, stub, stub_size); | |
ccc09689 | 5709 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { |
9032f11e | 5710 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); |
29089a54 | 5711 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); |
625b0720 | 5712 | } else { |
29089a54 RL |
5713 | put_userspace_action(ofproto, &buf, flow, &cookie, |
5714 | sizeof cookie.slow_path); | |
625b0720 | 5715 | } |
6a7e895f BP |
5716 | *actionsp = buf.data; |
5717 | *actions_lenp = buf.size; | |
5718 | } | |
5719 | ||
98403001 BP |
5720 | static size_t |
5721 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5722 | struct ofpbuf *odp_actions, | |
5723 | const struct flow *flow, | |
29089a54 RL |
5724 | const union user_action_cookie *cookie, |
5725 | const size_t cookie_size) | |
98403001 | 5726 | { |
98403001 BP |
5727 | uint32_t pid; |
5728 | ||
acf60855 | 5729 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
e1b1d06a | 5730 | ofp_port_to_odp_port(ofproto, flow->in_port)); |
98403001 | 5731 | |
29089a54 RL |
5732 | return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); |
5733 | } | |
5734 | ||
5735 | /* Compose SAMPLE action for sFlow or IPFIX. The given probability is | |
5736 | * the number of packets out of UINT32_MAX to sample. The given | |
5737 | * cookie is passed back in the callback for each sampled packet. | |
5738 | */ | |
5739 | static size_t | |
5740 | compose_sample_action(const struct ofproto_dpif *ofproto, | |
5741 | struct ofpbuf *odp_actions, | |
5742 | const struct flow *flow, | |
5743 | const uint32_t probability, | |
5744 | const union user_action_cookie *cookie, | |
5745 | const size_t cookie_size) | |
5746 | { | |
5747 | size_t sample_offset, actions_offset; | |
5748 | int cookie_offset; | |
5749 | ||
5750 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
5751 | ||
5752 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5753 | ||
5754 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
5755 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie, | |
5756 | cookie_size); | |
5757 | ||
5758 | nl_msg_end_nested(odp_actions, actions_offset); | |
5759 | nl_msg_end_nested(odp_actions, sample_offset); | |
5760 | return cookie_offset; | |
98403001 BP |
5761 | } |
5762 | ||
36fc5f18 BP |
5763 | static void |
5764 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5765 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 5766 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
5767 | { |
5768 | int ifindex; | |
5769 | ||
5770 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 5771 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
5772 | |
5773 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5774 | * port information") for the interpretation of cookie->output. */ | |
5775 | switch (n_outputs) { | |
5776 | case 0: | |
5777 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 5778 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
5779 | break; |
5780 | ||
5781 | case 1: | |
5782 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5783 | if (ifindex) { | |
1673e0e4 | 5784 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
5785 | break; |
5786 | } | |
5787 | /* Fall through. */ | |
5788 | default: | |
5789 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 5790 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
5791 | break; |
5792 | } | |
5793 | } | |
5794 | ||
29089a54 | 5795 | /* Compose SAMPLE action for sFlow bridge sampling. */ |
6ff686f2 PS |
5796 | static size_t |
5797 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5798 | struct ofpbuf *odp_actions, | |
5799 | const struct flow *flow, | |
5800 | uint32_t odp_port) | |
5801 | { | |
6ff686f2 | 5802 | uint32_t probability; |
1673e0e4 | 5803 | union user_action_cookie cookie; |
6ff686f2 PS |
5804 | |
5805 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5806 | return 0; | |
5807 | } | |
5808 | ||
6ff686f2 | 5809 | probability = dpif_sflow_get_probability(ofproto->sflow); |
36fc5f18 BP |
5810 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
5811 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
6ff686f2 | 5812 | |
29089a54 RL |
5813 | return compose_sample_action(ofproto, odp_actions, flow, probability, |
5814 | &cookie, sizeof cookie.sflow); | |
5815 | } | |
5816 | ||
5817 | static void | |
5818 | compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id, | |
5819 | uint32_t obs_domain_id, uint32_t obs_point_id, | |
5820 | union user_action_cookie *cookie) | |
5821 | { | |
5822 | cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE; | |
5823 | cookie->flow_sample.probability = probability; | |
5824 | cookie->flow_sample.collector_set_id = collector_set_id; | |
5825 | cookie->flow_sample.obs_domain_id = obs_domain_id; | |
5826 | cookie->flow_sample.obs_point_id = obs_point_id; | |
5827 | } | |
5828 | ||
5829 | static void | |
5830 | compose_ipfix_cookie(union user_action_cookie *cookie) | |
5831 | { | |
5832 | cookie->type = USER_ACTION_COOKIE_IPFIX; | |
5833 | } | |
5834 | ||
5835 | /* Compose SAMPLE action for IPFIX bridge sampling. */ | |
5836 | static void | |
5837 | compose_ipfix_action(const struct ofproto_dpif *ofproto, | |
5838 | struct ofpbuf *odp_actions, | |
5839 | const struct flow *flow) | |
5840 | { | |
5841 | uint32_t probability; | |
5842 | union user_action_cookie cookie; | |
5843 | ||
5844 | if (!ofproto->ipfix || flow->in_port == OFPP_NONE) { | |
5845 | return; | |
5846 | } | |
5847 | ||
5848 | probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix); | |
5849 | compose_ipfix_cookie(&cookie); | |
5850 | ||
5851 | compose_sample_action(ofproto, odp_actions, flow, probability, | |
5852 | &cookie, sizeof cookie.ipfix); | |
6ff686f2 PS |
5853 | } |
5854 | ||
29089a54 RL |
5855 | /* SAMPLE action for sFlow must be first action in any given list of |
5856 | * actions. At this point we do not have all information required to | |
5857 | * build it. So try to build sample action as complete as possible. */ | |
6ff686f2 | 5858 | static void |
f03a84b9 | 5859 | add_sflow_action(struct xlate_ctx *ctx) |
6ff686f2 PS |
5860 | { |
5861 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
bbafd73b EJ |
5862 | &ctx->xout->odp_actions, |
5863 | &ctx->xin->flow, OVSP_NONE); | |
6ff686f2 PS |
5864 | ctx->sflow_odp_port = 0; |
5865 | ctx->sflow_n_outputs = 0; | |
5866 | } | |
5867 | ||
29089a54 RL |
5868 | /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list |
5869 | * of actions, eventually after the SAMPLE action for sFlow. */ | |
5870 | static void | |
f03a84b9 | 5871 | add_ipfix_action(struct xlate_ctx *ctx) |
29089a54 | 5872 | { |
bbafd73b EJ |
5873 | compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions, |
5874 | &ctx->xin->flow); | |
29089a54 RL |
5875 | } |
5876 | ||
6ff686f2 PS |
5877 | /* Fix SAMPLE action according to data collected while composing ODP actions. |
5878 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
5879 | * USERSPACE action's user-cookie which is required for sflow. */ | |
5880 | static void | |
f03a84b9 | 5881 | fix_sflow_action(struct xlate_ctx *ctx) |
6ff686f2 PS |
5882 | { |
5883 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 5884 | union user_action_cookie *cookie; |
6ff686f2 PS |
5885 | |
5886 | if (!ctx->user_cookie_offset) { | |
5887 | return; | |
5888 | } | |
5889 | ||
bbafd73b | 5890 | cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset, |
a93f1927 | 5891 | sizeof cookie->sflow); |
cb22974d | 5892 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
6ff686f2 | 5893 | |
36fc5f18 BP |
5894 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
5895 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
5896 | } |
5897 | ||
6ff686f2 | 5898 | static void |
f03a84b9 | 5899 | compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port, |
81b1afb1 | 5900 | bool check_stp) |
6ff686f2 | 5901 | { |
d59906fb | 5902 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
d8558b4a JR |
5903 | ovs_be16 flow_vlan_tci; |
5904 | uint32_t flow_skb_mark; | |
5905 | uint8_t flow_nw_tos; | |
a4454ac6 | 5906 | struct priority_to_dscp *pdscp; |
0a740f48 EJ |
5907 | uint32_t out_port, odp_port; |
5908 | ||
5909 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
5910 | * before traversing a patch port. */ | |
cff78c88 | 5911 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); |
d59906fb | 5912 | |
a4454ac6 EJ |
5913 | if (!ofport) { |
5914 | xlate_report(ctx, "Nonexistent output port"); | |
5915 | return; | |
5916 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
5917 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
5918 | return; | |
5919 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
5920 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
5921 | return; | |
5922 | } | |
8b36f51e | 5923 | |
0a740f48 EJ |
5924 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5925 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
bbafd73b | 5926 | struct flow old_flow = ctx->xin->flow; |
0a740f48 | 5927 | const struct ofproto_dpif *peer_ofproto; |
bb374ef6 | 5928 | enum slow_path_reason special; |
ffaef958 | 5929 | struct ofport_dpif *in_port; |
0a740f48 EJ |
5930 | |
5931 | if (!peer) { | |
5932 | xlate_report(ctx, "Nonexistent patch port peer"); | |
5933 | return; | |
5934 | } | |
5935 | ||
5936 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5937 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
5938 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
5939 | return; | |
5940 | } | |
5941 | ||
5942 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
bbafd73b EJ |
5943 | ctx->xin->flow.in_port = peer->up.ofp_port; |
5944 | ctx->xin->flow.metadata = htonll(0); | |
5945 | memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel); | |
5946 | memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs); | |
5947 | ||
5948 | in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port); | |
5949 | special = process_special(ctx->ofproto, &ctx->xin->flow, in_port, | |
5950 | ctx->xin->packet); | |
bb374ef6 | 5951 | if (special) { |
bbafd73b | 5952 | ctx->xout->slow = special; |
bb374ef6 | 5953 | } else if (!in_port || may_receive(in_port, ctx)) { |
ffaef958 | 5954 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { |
bbafd73b | 5955 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true); |
ffaef958 BP |
5956 | } else { |
5957 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
5958 | * learning action look at the packet, then drop it. */ | |
5959 | struct flow old_base_flow = ctx->base_flow; | |
bbafd73b EJ |
5960 | size_t old_size = ctx->xout->odp_actions.size; |
5961 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true); | |
ffaef958 | 5962 | ctx->base_flow = old_base_flow; |
bbafd73b | 5963 | ctx->xout->odp_actions.size = old_size; |
ffaef958 BP |
5964 | } |
5965 | } | |
5966 | ||
bbafd73b | 5967 | ctx->xin->flow = old_flow; |
0a740f48 EJ |
5968 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
5969 | ||
bbafd73b EJ |
5970 | if (ctx->xin->resubmit_stats) { |
5971 | netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats); | |
5972 | netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats); | |
0a740f48 EJ |
5973 | } |
5974 | ||
5975 | return; | |
5976 | } | |
5977 | ||
bbafd73b EJ |
5978 | flow_vlan_tci = ctx->xin->flow.vlan_tci; |
5979 | flow_skb_mark = ctx->xin->flow.skb_mark; | |
5980 | flow_nw_tos = ctx->xin->flow.nw_tos; | |
d8558b4a | 5981 | |
bbafd73b | 5982 | pdscp = get_priority(ofport, ctx->xin->flow.skb_priority); |
a4454ac6 | 5983 | if (pdscp) { |
bbafd73b EJ |
5984 | ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK; |
5985 | ctx->xin->flow.nw_tos |= pdscp->dscp; | |
d59906fb EJ |
5986 | } |
5987 | ||
b9ad7294 | 5988 | if (ofport->tnl_port) { |
d8558b4a JR |
5989 | /* Save tunnel metadata so that changes made due to |
5990 | * the Logical (tunnel) Port are not visible for any further | |
5991 | * matches, while explicit set actions on tunnel metadata are. | |
5992 | */ | |
bbafd73b EJ |
5993 | struct flow_tnl flow_tnl = ctx->xin->flow.tunnel; |
5994 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow); | |
b9ad7294 EJ |
5995 | if (odp_port == OVSP_NONE) { |
5996 | xlate_report(ctx, "Tunneling decided against output"); | |
88e90491 | 5997 | goto out; /* restore flow_nw_tos */ |
b9ad7294 | 5998 | } |
bbafd73b | 5999 | if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) { |
0ad90c84 JR |
6000 | xlate_report(ctx, "Not tunneling to our own address"); |
6001 | goto out; /* restore flow_nw_tos */ | |
6002 | } | |
bbafd73b EJ |
6003 | if (ctx->xin->resubmit_stats) { |
6004 | netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats); | |
b9ad7294 EJ |
6005 | } |
6006 | out_port = odp_port; | |
bbafd73b EJ |
6007 | commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow, |
6008 | &ctx->xout->odp_actions); | |
6009 | ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */ | |
b9ad7294 | 6010 | } else { |
cf630ea3 | 6011 | odp_port = ofport->odp_port; |
b9ad7294 | 6012 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
bbafd73b | 6013 | ctx->xin->flow.vlan_tci); |
b9ad7294 | 6014 | if (out_port != odp_port) { |
bbafd73b | 6015 | ctx->xin->flow.vlan_tci = htons(0); |
b9ad7294 | 6016 | } |
bbafd73b | 6017 | ctx->xin->flow.skb_mark &= ~IPSEC_MARK; |
52a90c29 | 6018 | } |
bbafd73b EJ |
6019 | commit_odp_actions(&ctx->xin->flow, &ctx->base_flow, |
6020 | &ctx->xout->odp_actions); | |
6021 | nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); | |
52a90c29 | 6022 | |
6ff686f2 PS |
6023 | ctx->sflow_odp_port = odp_port; |
6024 | ctx->sflow_n_outputs++; | |
bbafd73b | 6025 | ctx->xout->nf_output_iface = ofp_port; |
d8558b4a JR |
6026 | |
6027 | /* Restore flow */ | |
bbafd73b EJ |
6028 | ctx->xin->flow.vlan_tci = flow_vlan_tci; |
6029 | ctx->xin->flow.skb_mark = flow_skb_mark; | |
88e90491 | 6030 | out: |
bbafd73b | 6031 | ctx->xin->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
6032 | } |
6033 | ||
abe529af | 6034 | static void |
f03a84b9 | 6035 | compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 6036 | { |
81b1afb1 | 6037 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
6038 | } |
6039 | ||
55599423 | 6040 | static void |
f03a84b9 | 6041 | tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule) |
55599423 JR |
6042 | { |
6043 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6044 | uint8_t table_id = ctx->table_id; | |
6045 | ||
6046 | if (table_id > 0 && table_id < N_TABLES) { | |
6047 | struct table_dpif *table = &ofproto->tables[table_id]; | |
6048 | if (table->other_table) { | |
bbafd73b EJ |
6049 | ctx->xout->tags |= (rule && rule->tag |
6050 | ? rule->tag | |
6051 | : rule_calculate_tag(&ctx->xin->flow, | |
6052 | &table->other_table->mask, | |
6053 | table->basis)); | |
55599423 JR |
6054 | } |
6055 | } | |
6056 | } | |
6057 | ||
6058 | /* Common rule processing in one place to avoid duplicating code. */ | |
6059 | static struct rule_dpif * | |
f03a84b9 | 6060 | ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule, |
55599423 JR |
6061 | bool may_packet_in) |
6062 | { | |
bbafd73b EJ |
6063 | if (ctx->xin->resubmit_hook) { |
6064 | ctx->xin->resubmit_hook(ctx, rule); | |
55599423 JR |
6065 | } |
6066 | if (rule == NULL && may_packet_in) { | |
6067 | /* XXX | |
6068 | * check if table configuration flags | |
6069 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
6070 | * OFPTC_TABLE_MISS_CONTINUE, | |
6071 | * OFPTC_TABLE_MISS_DROP | |
6072 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
6073 | */ | |
bbafd73b | 6074 | rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow); |
55599423 | 6075 | } |
bbafd73b EJ |
6076 | if (rule && ctx->xin->resubmit_stats) { |
6077 | rule_credit_stats(rule, ctx->xin->resubmit_stats); | |
55599423 JR |
6078 | } |
6079 | return rule; | |
6080 | } | |
6081 | ||
abe529af | 6082 | static void |
f03a84b9 | 6083 | xlate_table_action(struct xlate_ctx *ctx, |
1688c479 | 6084 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
6085 | { |
6086 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
6087 | struct rule_dpif *rule; | |
bbafd73b | 6088 | uint16_t old_in_port = ctx->xin->flow.in_port; |
55599423 | 6089 | uint8_t old_table_id = ctx->table_id; |
29901626 | 6090 | |
29901626 | 6091 | ctx->table_id = table_id; |
abe529af | 6092 | |
54a9cbc9 | 6093 | /* Look up a flow with 'in_port' as the input port. */ |
bbafd73b EJ |
6094 | ctx->xin->flow.in_port = in_port; |
6095 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id); | |
55599423 JR |
6096 | |
6097 | tag_the_flow(ctx, rule); | |
54a9cbc9 BP |
6098 | |
6099 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
6100 | * OFPP_IN_PORT will have surprising behavior. */ | |
bbafd73b | 6101 | ctx->xin->flow.in_port = old_in_port; |
abe529af | 6102 | |
55599423 | 6103 | rule = ctx_rule_hooks(ctx, rule, may_packet_in); |
1688c479 | 6104 | |
abe529af | 6105 | if (rule) { |
18b2a258 | 6106 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 6107 | |
abe529af | 6108 | ctx->recurse++; |
18b2a258 | 6109 | ctx->rule = rule; |
f25d0cf3 | 6110 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 6111 | ctx->rule = old_rule; |
abe529af BP |
6112 | ctx->recurse--; |
6113 | } | |
29901626 BP |
6114 | |
6115 | ctx->table_id = old_table_id; | |
abe529af BP |
6116 | } else { |
6117 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
6118 | ||
29901626 | 6119 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 6120 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 6121 | ctx->max_resubmit_trigger = true; |
abe529af BP |
6122 | } |
6123 | } | |
6124 | ||
29901626 | 6125 | static void |
f03a84b9 | 6126 | xlate_ofpact_resubmit(struct xlate_ctx *ctx, |
f25d0cf3 | 6127 | const struct ofpact_resubmit *resubmit) |
29901626 BP |
6128 | { |
6129 | uint16_t in_port; | |
6130 | uint8_t table_id; | |
6131 | ||
f25d0cf3 BP |
6132 | in_port = resubmit->in_port; |
6133 | if (in_port == OFPP_IN_PORT) { | |
bbafd73b | 6134 | in_port = ctx->xin->flow.in_port; |
f25d0cf3 BP |
6135 | } |
6136 | ||
6137 | table_id = resubmit->table_id; | |
6138 | if (table_id == 255) { | |
6139 | table_id = ctx->table_id; | |
6140 | } | |
29901626 | 6141 | |
1688c479 | 6142 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
6143 | } |
6144 | ||
abe529af | 6145 | static void |
f03a84b9 | 6146 | flood_packets(struct xlate_ctx *ctx, bool all) |
abe529af BP |
6147 | { |
6148 | struct ofport_dpif *ofport; | |
6149 | ||
b3e9b2ed | 6150 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 6151 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb | 6152 | |
bbafd73b | 6153 | if (ofp_port == ctx->xin->flow.in_port) { |
d59906fb EJ |
6154 | continue; |
6155 | } | |
6156 | ||
5e48dc2b | 6157 | if (all) { |
81b1afb1 | 6158 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 6159 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 6160 | compose_output_action(ctx, ofp_port); |
abe529af BP |
6161 | } |
6162 | } | |
b3e9b2ed | 6163 | |
bbafd73b | 6164 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
abe529af BP |
6165 | } |
6166 | ||
6ff686f2 | 6167 | static void |
f03a84b9 | 6168 | execute_controller_action(struct xlate_ctx *ctx, int len, |
a7349929 BP |
6169 | enum ofp_packet_in_reason reason, |
6170 | uint16_t controller_id) | |
6ff686f2 | 6171 | { |
999fba59 EJ |
6172 | struct ofputil_packet_in pin; |
6173 | struct ofpbuf *packet; | |
6ff686f2 | 6174 | |
bbafd73b EJ |
6175 | ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER); |
6176 | ctx->xout->slow = SLOW_CONTROLLER; | |
6177 | if (!ctx->xin->packet) { | |
999fba59 EJ |
6178 | return; |
6179 | } | |
6180 | ||
bbafd73b | 6181 | packet = ofpbuf_clone(ctx->xin->packet); |
999fba59 EJ |
6182 | |
6183 | if (packet->l2 && packet->l3) { | |
6184 | struct eth_header *eh; | |
b02475c5 | 6185 | uint16_t mpls_depth; |
999fba59 EJ |
6186 | |
6187 | eth_pop_vlan(packet); | |
6188 | eh = packet->l2; | |
0104aba8 | 6189 | |
bbafd73b EJ |
6190 | memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src); |
6191 | memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst); | |
999fba59 | 6192 | |
bbafd73b EJ |
6193 | if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) { |
6194 | eth_push_vlan(packet, ctx->xin->flow.vlan_tci); | |
999fba59 EJ |
6195 | } |
6196 | ||
b02475c5 SH |
6197 | mpls_depth = eth_mpls_depth(packet); |
6198 | ||
bbafd73b EJ |
6199 | if (mpls_depth < ctx->xin->flow.mpls_depth) { |
6200 | push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse); | |
6201 | } else if (mpls_depth > ctx->xin->flow.mpls_depth) { | |
6202 | pop_mpls(packet, ctx->xin->flow.dl_type); | |
b02475c5 | 6203 | } else if (mpls_depth) { |
bbafd73b | 6204 | set_mpls_lse(packet, ctx->xin->flow.mpls_lse); |
b02475c5 SH |
6205 | } |
6206 | ||
999fba59 | 6207 | if (packet->l4) { |
bbafd73b EJ |
6208 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6209 | packet_set_ipv4(packet, ctx->xin->flow.nw_src, | |
6210 | ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos, | |
6211 | ctx->xin->flow.nw_ttl); | |
999fba59 EJ |
6212 | } |
6213 | ||
6214 | if (packet->l7) { | |
bbafd73b EJ |
6215 | if (ctx->xin->flow.nw_proto == IPPROTO_TCP) { |
6216 | packet_set_tcp_port(packet, ctx->xin->flow.tp_src, | |
6217 | ctx->xin->flow.tp_dst); | |
6218 | } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) { | |
6219 | packet_set_udp_port(packet, ctx->xin->flow.tp_src, | |
6220 | ctx->xin->flow.tp_dst); | |
999fba59 EJ |
6221 | } |
6222 | } | |
6223 | } | |
6224 | } | |
6225 | ||
6226 | pin.packet = packet->data; | |
6227 | pin.packet_len = packet->size; | |
f0fd1a17 | 6228 | pin.reason = reason; |
a7349929 | 6229 | pin.controller_id = controller_id; |
54834960 | 6230 | pin.table_id = ctx->table_id; |
18b2a258 | 6231 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 6232 | |
999fba59 | 6233 | pin.send_len = len; |
bbafd73b | 6234 | flow_get_metadata(&ctx->xin->flow, &pin.fmd); |
999fba59 | 6235 | |
d8653c38 | 6236 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 6237 | ofpbuf_delete(packet); |
6ff686f2 PS |
6238 | } |
6239 | ||
b02475c5 | 6240 | static void |
f03a84b9 | 6241 | execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type) |
b02475c5 SH |
6242 | { |
6243 | ovs_assert(eth_type_mpls(eth_type)); | |
6244 | ||
6245 | if (ctx->base_flow.mpls_depth) { | |
bbafd73b EJ |
6246 | ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); |
6247 | ctx->xin->flow.mpls_depth++; | |
b02475c5 SH |
6248 | } else { |
6249 | ovs_be32 label; | |
6250 | uint8_t tc, ttl; | |
6251 | ||
bbafd73b | 6252 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) { |
b02475c5 SH |
6253 | label = htonl(0x2); /* IPV6 Explicit Null. */ |
6254 | } else { | |
6255 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6256 | } | |
bbafd73b EJ |
6257 | tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2; |
6258 | ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40; | |
6259 | ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
6260 | ctx->xin->flow.mpls_depth = 1; | |
b02475c5 | 6261 | } |
bbafd73b | 6262 | ctx->xin->flow.dl_type = eth_type; |
b02475c5 SH |
6263 | } |
6264 | ||
6265 | static void | |
f03a84b9 | 6266 | execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type) |
b02475c5 | 6267 | { |
bbafd73b | 6268 | ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type)); |
b02475c5 SH |
6269 | ovs_assert(!eth_type_mpls(eth_type)); |
6270 | ||
bbafd73b EJ |
6271 | if (ctx->xin->flow.mpls_depth) { |
6272 | ctx->xin->flow.mpls_depth--; | |
6273 | ctx->xin->flow.mpls_lse = htonl(0); | |
6274 | if (!ctx->xin->flow.mpls_depth) { | |
6275 | ctx->xin->flow.dl_type = eth_type; | |
b02475c5 SH |
6276 | } |
6277 | } | |
6278 | } | |
6279 | ||
f0fd1a17 | 6280 | static bool |
f03a84b9 | 6281 | compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 | 6282 | { |
bbafd73b EJ |
6283 | if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) && |
6284 | ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
f0fd1a17 PS |
6285 | return false; |
6286 | } | |
6287 | ||
bbafd73b EJ |
6288 | if (ctx->xin->flow.nw_ttl > 1) { |
6289 | ctx->xin->flow.nw_ttl--; | |
f0fd1a17 PS |
6290 | return false; |
6291 | } else { | |
c2d967a5 MM |
6292 | size_t i; |
6293 | ||
6294 | for (i = 0; i < ids->n_controllers; i++) { | |
6295 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6296 | ids->cnt_ids[i]); | |
6297 | } | |
f0fd1a17 PS |
6298 | |
6299 | /* Stop processing for current table. */ | |
6300 | return true; | |
6301 | } | |
6302 | } | |
6303 | ||
0f3f3c3d | 6304 | static bool |
f03a84b9 | 6305 | execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl) |
0f3f3c3d | 6306 | { |
bbafd73b | 6307 | if (!eth_type_mpls(ctx->xin->flow.dl_type)) { |
0f3f3c3d SH |
6308 | return true; |
6309 | } | |
6310 | ||
bbafd73b | 6311 | set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl); |
0f3f3c3d SH |
6312 | return false; |
6313 | } | |
6314 | ||
b676167a | 6315 | static bool |
f03a84b9 | 6316 | execute_dec_mpls_ttl_action(struct xlate_ctx *ctx) |
b676167a | 6317 | { |
bbafd73b | 6318 | uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse); |
b676167a | 6319 | |
bbafd73b | 6320 | if (!eth_type_mpls(ctx->xin->flow.dl_type)) { |
b676167a SH |
6321 | return false; |
6322 | } | |
6323 | ||
be80bc65 | 6324 | if (ttl > 1) { |
b676167a | 6325 | ttl--; |
bbafd73b | 6326 | set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl); |
b676167a SH |
6327 | return false; |
6328 | } else { | |
6329 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
6330 | ||
6331 | /* Stop processing for current table. */ | |
6332 | return true; | |
6333 | } | |
6334 | } | |
6335 | ||
abe529af | 6336 | static void |
f03a84b9 | 6337 | xlate_output_action(struct xlate_ctx *ctx, |
1688c479 | 6338 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af | 6339 | { |
bbafd73b | 6340 | uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface; |
abe529af | 6341 | |
bbafd73b | 6342 | ctx->xout->nf_output_iface = NF_OUT_DROP; |
abe529af BP |
6343 | |
6344 | switch (port) { | |
6345 | case OFPP_IN_PORT: | |
bbafd73b | 6346 | compose_output_action(ctx, ctx->xin->flow.in_port); |
abe529af BP |
6347 | break; |
6348 | case OFPP_TABLE: | |
bbafd73b | 6349 | xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in); |
abe529af BP |
6350 | break; |
6351 | case OFPP_NORMAL: | |
6352 | xlate_normal(ctx); | |
6353 | break; | |
6354 | case OFPP_FLOOD: | |
d59906fb | 6355 | flood_packets(ctx, false); |
abe529af BP |
6356 | break; |
6357 | case OFPP_ALL: | |
d59906fb | 6358 | flood_packets(ctx, true); |
abe529af BP |
6359 | break; |
6360 | case OFPP_CONTROLLER: | |
a7349929 | 6361 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 6362 | break; |
e81d2933 EJ |
6363 | case OFPP_NONE: |
6364 | break; | |
a0fbe94a | 6365 | case OFPP_LOCAL: |
abe529af | 6366 | default: |
bbafd73b | 6367 | if (port != ctx->xin->flow.in_port) { |
81b1afb1 | 6368 | compose_output_action(ctx, port); |
3dd3eace BP |
6369 | } else { |
6370 | xlate_report(ctx, "skipping output to input port"); | |
abe529af BP |
6371 | } |
6372 | break; | |
6373 | } | |
6374 | ||
6375 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
bbafd73b EJ |
6376 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
6377 | } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) { | |
6378 | ctx->xout->nf_output_iface = prev_nf_output_iface; | |
abe529af | 6379 | } else if (prev_nf_output_iface != NF_OUT_DROP && |
bbafd73b EJ |
6380 | ctx->xout->nf_output_iface != NF_OUT_FLOOD) { |
6381 | ctx->xout->nf_output_iface = NF_OUT_MULTI; | |
abe529af BP |
6382 | } |
6383 | } | |
6384 | ||
f694937d | 6385 | static void |
f03a84b9 | 6386 | xlate_output_reg_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6387 | const struct ofpact_output_reg *or) |
f694937d | 6388 | { |
bbafd73b | 6389 | uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow); |
f25d0cf3 | 6390 | if (port <= UINT16_MAX) { |
1688c479 | 6391 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
6392 | } |
6393 | } | |
6394 | ||
abe529af | 6395 | static void |
f03a84b9 | 6396 | xlate_enqueue_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6397 | const struct ofpact_enqueue *enqueue) |
abe529af | 6398 | { |
f25d0cf3 BP |
6399 | uint16_t ofp_port = enqueue->port; |
6400 | uint32_t queue_id = enqueue->queue; | |
abff858b | 6401 | uint32_t flow_priority, priority; |
abe529af BP |
6402 | int error; |
6403 | ||
f25d0cf3 | 6404 | /* Translate queue to priority. */ |
acf60855 JP |
6405 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6406 | queue_id, &priority); | |
abe529af BP |
6407 | if (error) { |
6408 | /* Fall back to ordinary output action. */ | |
1688c479 | 6409 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
6410 | return; |
6411 | } | |
6412 | ||
f25d0cf3 | 6413 | /* Check output port. */ |
abe529af | 6414 | if (ofp_port == OFPP_IN_PORT) { |
bbafd73b EJ |
6415 | ofp_port = ctx->xin->flow.in_port; |
6416 | } else if (ofp_port == ctx->xin->flow.in_port) { | |
8ba855c1 | 6417 | return; |
abe529af | 6418 | } |
abe529af | 6419 | |
df2c07f4 | 6420 | /* Add datapath actions. */ |
bbafd73b EJ |
6421 | flow_priority = ctx->xin->flow.skb_priority; |
6422 | ctx->xin->flow.skb_priority = priority; | |
81b1afb1 | 6423 | compose_output_action(ctx, ofp_port); |
bbafd73b | 6424 | ctx->xin->flow.skb_priority = flow_priority; |
abe529af BP |
6425 | |
6426 | /* Update NetFlow output port. */ | |
bbafd73b EJ |
6427 | if (ctx->xout->nf_output_iface == NF_OUT_DROP) { |
6428 | ctx->xout->nf_output_iface = ofp_port; | |
6429 | } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) { | |
6430 | ctx->xout->nf_output_iface = NF_OUT_MULTI; | |
abe529af BP |
6431 | } |
6432 | } | |
6433 | ||
6434 | static void | |
f03a84b9 | 6435 | xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 6436 | { |
f25d0cf3 | 6437 | uint32_t skb_priority; |
abe529af | 6438 | |
acf60855 JP |
6439 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6440 | queue_id, &skb_priority)) { | |
bbafd73b | 6441 | ctx->xin->flow.skb_priority = skb_priority; |
f25d0cf3 BP |
6442 | } else { |
6443 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 6444 | * has already been logged. */ |
abe529af | 6445 | } |
abe529af BP |
6446 | } |
6447 | ||
daff3353 EJ |
6448 | static bool |
6449 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6450 | { | |
6451 | struct ofproto_dpif *ofproto = ofproto_; | |
6452 | struct ofport_dpif *port; | |
6453 | ||
6454 | switch (ofp_port) { | |
6455 | case OFPP_IN_PORT: | |
6456 | case OFPP_TABLE: | |
6457 | case OFPP_NORMAL: | |
6458 | case OFPP_FLOOD: | |
6459 | case OFPP_ALL: | |
439e4d8c | 6460 | case OFPP_NONE: |
daff3353 EJ |
6461 | return true; |
6462 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6463 | return false; | |
6464 | default: | |
6465 | port = get_ofp_port(ofproto, ofp_port); | |
6466 | return port ? port->may_enable : false; | |
6467 | } | |
6468 | } | |
6469 | ||
f25d0cf3 | 6470 | static void |
f03a84b9 | 6471 | xlate_bundle_action(struct xlate_ctx *ctx, |
f25d0cf3 BP |
6472 | const struct ofpact_bundle *bundle) |
6473 | { | |
6474 | uint16_t port; | |
6475 | ||
bbafd73b EJ |
6476 | port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb, |
6477 | ctx->ofproto); | |
f25d0cf3 | 6478 | if (bundle->dst.field) { |
bbafd73b | 6479 | nxm_reg_load(&bundle->dst, port, &ctx->xin->flow); |
f25d0cf3 | 6480 | } else { |
1688c479 | 6481 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
6482 | } |
6483 | } | |
6484 | ||
75a75043 | 6485 | static void |
f03a84b9 | 6486 | xlate_learn_action(struct xlate_ctx *ctx, |
f25d0cf3 | 6487 | const struct ofpact_learn *learn) |
75a75043 BP |
6488 | { |
6489 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6490 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
6491 | uint64_t ofpacts_stub[1024 / 8]; |
6492 | struct ofpbuf ofpacts; | |
75a75043 BP |
6493 | int error; |
6494 | ||
f25d0cf3 | 6495 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
bbafd73b | 6496 | learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts); |
75a75043 BP |
6497 | |
6498 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6499 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
6500 | VLOG_WARN("learning action failed to modify flow table (%s)", |
6501 | ofperr_get_name(error)); | |
75a75043 BP |
6502 | } |
6503 | ||
f25d0cf3 | 6504 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
6505 | } |
6506 | ||
0e553d9c BP |
6507 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
6508 | * means "infinite". */ | |
6509 | static void | |
6510 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6511 | { | |
6512 | if (max && (!*timeout || *timeout > max)) { | |
6513 | *timeout = max; | |
6514 | } | |
6515 | } | |
6516 | ||
6517 | static void | |
f03a84b9 | 6518 | xlate_fin_timeout(struct xlate_ctx *ctx, |
f25d0cf3 | 6519 | const struct ofpact_fin_timeout *oft) |
0e553d9c | 6520 | { |
bbafd73b | 6521 | if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { |
0e553d9c BP |
6522 | struct rule_dpif *rule = ctx->rule; |
6523 | ||
f25d0cf3 BP |
6524 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
6525 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
6526 | } |
6527 | } | |
6528 | ||
29089a54 | 6529 | static void |
f03a84b9 | 6530 | xlate_sample_action(struct xlate_ctx *ctx, |
29089a54 RL |
6531 | const struct ofpact_sample *os) |
6532 | { | |
6533 | union user_action_cookie cookie; | |
6534 | /* Scale the probability from 16-bit to 32-bit while representing | |
6535 | * the same percentage. */ | |
6536 | uint32_t probability = (os->probability << 16) | os->probability; | |
6537 | ||
bbafd73b EJ |
6538 | commit_odp_actions(&ctx->xin->flow, &ctx->base_flow, |
6539 | &ctx->xout->odp_actions); | |
29089a54 RL |
6540 | |
6541 | compose_flow_sample_cookie(os->probability, os->collector_set_id, | |
6542 | os->obs_domain_id, os->obs_point_id, &cookie); | |
bbafd73b | 6543 | compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow, |
29089a54 RL |
6544 | probability, &cookie, sizeof cookie.flow_sample); |
6545 | } | |
6546 | ||
21f7563c | 6547 | static bool |
f03a84b9 | 6548 | may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx) |
21f7563c | 6549 | { |
bbafd73b EJ |
6550 | if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst, |
6551 | eth_addr_stp) | |
9e1fd49b BP |
6552 | ? OFPUTIL_PC_NO_RECV_STP |
6553 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
6554 | return false; |
6555 | } | |
6556 | ||
6557 | /* Only drop packets here if both forwarding and learning are | |
6558 | * disabled. If just learning is enabled, we need to have | |
6559 | * OFPP_NORMAL and the learning action have a look at the packet | |
6560 | * before we can drop it. */ | |
6561 | if (!stp_forward_in_state(port->stp_state) | |
6562 | && !stp_learn_in_state(port->stp_state)) { | |
6563 | return false; | |
6564 | } | |
6565 | ||
6566 | return true; | |
6567 | } | |
6568 | ||
4863c249 | 6569 | static bool |
f03a84b9 | 6570 | tunnel_ecn_ok(struct xlate_ctx *ctx) |
4863c249 JP |
6571 | { |
6572 | if (is_ip_any(&ctx->base_flow) | |
bbafd73b | 6573 | && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { |
29a5df0a JP |
6574 | if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { |
6575 | VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" | |
6576 | " but is not ECN capable"); | |
6577 | return false; | |
6578 | } else { | |
6579 | /* Set the ECN CE value in the tunneled packet. */ | |
bbafd73b | 6580 | ctx->xin->flow.nw_tos |= IP_ECN_CE; |
29a5df0a | 6581 | } |
4863c249 JP |
6582 | } |
6583 | ||
6584 | return true; | |
6585 | } | |
6586 | ||
abe529af | 6587 | static void |
f25d0cf3 | 6588 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
f03a84b9 | 6589 | struct xlate_ctx *ctx) |
abe529af | 6590 | { |
254750ce | 6591 | bool was_evictable = true; |
f25d0cf3 | 6592 | const struct ofpact *a; |
abe529af | 6593 | |
254750ce BP |
6594 | if (ctx->rule) { |
6595 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6596 | was_evictable = ctx->rule->up.evictable; | |
6597 | ctx->rule->up.evictable = false; | |
6598 | } | |
55599423 JR |
6599 | |
6600 | do_xlate_actions_again: | |
f25d0cf3 BP |
6601 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
6602 | struct ofpact_controller *controller; | |
4cceacb9 | 6603 | const struct ofpact_metadata *metadata; |
38f2e360 | 6604 | |
848e8809 EJ |
6605 | if (ctx->exit) { |
6606 | break; | |
6607 | } | |
6608 | ||
f25d0cf3 BP |
6609 | switch (a->type) { |
6610 | case OFPACT_OUTPUT: | |
6611 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 6612 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
6613 | break; |
6614 | ||
6615 | case OFPACT_CONTROLLER: | |
6616 | controller = ofpact_get_CONTROLLER(a); | |
6617 | execute_controller_action(ctx, controller->max_len, | |
6618 | controller->reason, | |
6619 | controller->controller_id); | |
6620 | break; | |
690a61c5 | 6621 | |
f25d0cf3 BP |
6622 | case OFPACT_ENQUEUE: |
6623 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
6624 | break; |
6625 | ||
f25d0cf3 | 6626 | case OFPACT_SET_VLAN_VID: |
bbafd73b EJ |
6627 | ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
6628 | ctx->xin->flow.vlan_tci |= | |
6629 | (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) | |
6630 | | htons(VLAN_CFI)); | |
abe529af BP |
6631 | break; |
6632 | ||
f25d0cf3 | 6633 | case OFPACT_SET_VLAN_PCP: |
bbafd73b EJ |
6634 | ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
6635 | ctx->xin->flow.vlan_tci |= | |
6636 | htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT) | |
6637 | | VLAN_CFI); | |
abe529af BP |
6638 | break; |
6639 | ||
f25d0cf3 | 6640 | case OFPACT_STRIP_VLAN: |
bbafd73b | 6641 | ctx->xin->flow.vlan_tci = htons(0); |
abe529af BP |
6642 | break; |
6643 | ||
3e34fbdd | 6644 | case OFPACT_PUSH_VLAN: |
5dca28b5 | 6645 | /* XXX 802.1AD(QinQ) */ |
bbafd73b | 6646 | ctx->xin->flow.vlan_tci = htons(VLAN_CFI); |
3e34fbdd IY |
6647 | break; |
6648 | ||
f25d0cf3 | 6649 | case OFPACT_SET_ETH_SRC: |
bbafd73b | 6650 | memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, |
f25d0cf3 | 6651 | ETH_ADDR_LEN); |
abe529af BP |
6652 | break; |
6653 | ||
f25d0cf3 | 6654 | case OFPACT_SET_ETH_DST: |
bbafd73b | 6655 | memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, |
f25d0cf3 | 6656 | ETH_ADDR_LEN); |
abe529af BP |
6657 | break; |
6658 | ||
f25d0cf3 | 6659 | case OFPACT_SET_IPV4_SRC: |
bbafd73b EJ |
6660 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6661 | ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
1b035ef2 | 6662 | } |
abe529af BP |
6663 | break; |
6664 | ||
f25d0cf3 | 6665 | case OFPACT_SET_IPV4_DST: |
bbafd73b EJ |
6666 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6667 | ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
1b035ef2 | 6668 | } |
abe529af BP |
6669 | break; |
6670 | ||
f25d0cf3 | 6671 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d | 6672 | /* OpenFlow 1.0 only supports IPv4. */ |
bbafd73b EJ |
6673 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) { |
6674 | ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK; | |
6675 | ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; | |
c4f2731d | 6676 | } |
abe529af BP |
6677 | break; |
6678 | ||
f25d0cf3 | 6679 | case OFPACT_SET_L4_SRC_PORT: |
bbafd73b EJ |
6680 | if (is_ip_any(&ctx->xin->flow)) { |
6681 | ctx->xin->flow.tp_src = | |
6682 | htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
1b035ef2 | 6683 | } |
abe529af BP |
6684 | break; |
6685 | ||
f25d0cf3 | 6686 | case OFPACT_SET_L4_DST_PORT: |
bbafd73b EJ |
6687 | if (is_ip_any(&ctx->xin->flow)) { |
6688 | ctx->xin->flow.tp_dst = | |
6689 | htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
1b035ef2 | 6690 | } |
abe529af BP |
6691 | break; |
6692 | ||
f25d0cf3 BP |
6693 | case OFPACT_RESUBMIT: |
6694 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
6695 | break; |
6696 | ||
f25d0cf3 | 6697 | case OFPACT_SET_TUNNEL: |
bbafd73b EJ |
6698 | ctx->xin->flow.tunnel.tun_id = |
6699 | htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
29901626 BP |
6700 | break; |
6701 | ||
f25d0cf3 BP |
6702 | case OFPACT_SET_QUEUE: |
6703 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
6704 | break; |
6705 | ||
f25d0cf3 | 6706 | case OFPACT_POP_QUEUE: |
bbafd73b | 6707 | ctx->xin->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
6708 | break; |
6709 | ||
f25d0cf3 | 6710 | case OFPACT_REG_MOVE: |
bbafd73b | 6711 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow); |
38f2e360 BP |
6712 | break; |
6713 | ||
f25d0cf3 | 6714 | case OFPACT_REG_LOAD: |
bbafd73b | 6715 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow); |
38f2e360 BP |
6716 | break; |
6717 | ||
bd85dac1 | 6718 | case OFPACT_STACK_PUSH: |
bbafd73b | 6719 | nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow, |
bd85dac1 AZ |
6720 | &ctx->stack); |
6721 | break; | |
6722 | ||
6723 | case OFPACT_STACK_POP: | |
bbafd73b | 6724 | nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow, |
bd85dac1 AZ |
6725 | &ctx->stack); |
6726 | break; | |
6727 | ||
b02475c5 SH |
6728 | case OFPACT_PUSH_MPLS: |
6729 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6730 | break; | |
6731 | ||
6732 | case OFPACT_POP_MPLS: | |
6733 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6734 | break; | |
6735 | ||
0f3f3c3d | 6736 | case OFPACT_SET_MPLS_TTL: |
bbafd73b EJ |
6737 | if (execute_set_mpls_ttl_action(ctx, |
6738 | ofpact_get_SET_MPLS_TTL(a)->ttl)) { | |
0f3f3c3d SH |
6739 | goto out; |
6740 | } | |
6741 | break; | |
6742 | ||
b676167a SH |
6743 | case OFPACT_DEC_MPLS_TTL: |
6744 | if (execute_dec_mpls_ttl_action(ctx)) { | |
6745 | goto out; | |
6746 | } | |
6747 | break; | |
6748 | ||
f25d0cf3 | 6749 | case OFPACT_DEC_TTL: |
c2d967a5 | 6750 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
6751 | goto out; |
6752 | } | |
38f2e360 BP |
6753 | break; |
6754 | ||
f25d0cf3 BP |
6755 | case OFPACT_NOTE: |
6756 | /* Nothing to do. */ | |
abe529af BP |
6757 | break; |
6758 | ||
f25d0cf3 | 6759 | case OFPACT_MULTIPATH: |
bbafd73b | 6760 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow); |
abe529af | 6761 | break; |
daff3353 | 6762 | |
f25d0cf3 | 6763 | case OFPACT_BUNDLE: |
a368bb53 | 6764 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 6765 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 6766 | break; |
f694937d | 6767 | |
f25d0cf3 BP |
6768 | case OFPACT_OUTPUT_REG: |
6769 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 6770 | break; |
75a75043 | 6771 | |
f25d0cf3 | 6772 | case OFPACT_LEARN: |
bbafd73b EJ |
6773 | ctx->xout->has_learn = true; |
6774 | if (ctx->xin->may_learn) { | |
f25d0cf3 | 6775 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
6776 | } |
6777 | break; | |
848e8809 | 6778 | |
f25d0cf3 | 6779 | case OFPACT_EXIT: |
848e8809 EJ |
6780 | ctx->exit = true; |
6781 | break; | |
0e553d9c | 6782 | |
f25d0cf3 | 6783 | case OFPACT_FIN_TIMEOUT: |
bbafd73b | 6784 | ctx->xout->has_fin_timeout = true; |
f25d0cf3 | 6785 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 6786 | break; |
8dd54666 | 6787 | |
b19e8793 | 6788 | case OFPACT_CLEAR_ACTIONS: |
5dca28b5 | 6789 | /* XXX |
b19e8793 IY |
6790 | * Nothing to do because writa-actions is not supported for now. |
6791 | * When writa-actions is supported, clear-actions also must | |
6792 | * be supported at the same time. | |
6793 | */ | |
6794 | break; | |
6795 | ||
4cceacb9 JS |
6796 | case OFPACT_WRITE_METADATA: |
6797 | metadata = ofpact_get_WRITE_METADATA(a); | |
bbafd73b EJ |
6798 | ctx->xin->flow.metadata &= ~metadata->mask; |
6799 | ctx->xin->flow.metadata |= metadata->metadata & metadata->mask; | |
4cceacb9 JS |
6800 | break; |
6801 | ||
8dd54666 | 6802 | case OFPACT_GOTO_TABLE: { |
55599423 | 6803 | /* It is assumed that goto-table is the last action. */ |
8dd54666 | 6804 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); |
55599423 JR |
6805 | struct rule_dpif *rule; |
6806 | ||
cb22974d | 6807 | ovs_assert(ctx->table_id < ogt->table_id); |
55599423 JR |
6808 | |
6809 | ctx->table_id = ogt->table_id; | |
6810 | ||
6811 | /* Look up a flow from the new table. */ | |
bbafd73b | 6812 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id); |
55599423 JR |
6813 | |
6814 | tag_the_flow(ctx, rule); | |
6815 | ||
6816 | rule = ctx_rule_hooks(ctx, rule, true); | |
6817 | ||
6818 | if (rule) { | |
6819 | if (ctx->rule) { | |
6820 | ctx->rule->up.evictable = was_evictable; | |
6821 | } | |
6822 | ctx->rule = rule; | |
6823 | was_evictable = rule->up.evictable; | |
6824 | rule->up.evictable = false; | |
6825 | ||
6826 | /* Tail recursion removal. */ | |
6827 | ofpacts = rule->up.ofpacts; | |
6828 | ofpacts_len = rule->up.ofpacts_len; | |
6829 | goto do_xlate_actions_again; | |
6830 | } | |
8dd54666 IY |
6831 | break; |
6832 | } | |
29089a54 RL |
6833 | |
6834 | case OFPACT_SAMPLE: | |
6835 | xlate_sample_action(ctx, ofpact_get_SAMPLE(a)); | |
6836 | break; | |
abe529af BP |
6837 | } |
6838 | } | |
21f7563c | 6839 | |
f0fd1a17 | 6840 | out: |
254750ce BP |
6841 | if (ctx->rule) { |
6842 | ctx->rule->up.evictable = was_evictable; | |
6843 | } | |
abe529af BP |
6844 | } |
6845 | ||
6846 | static void | |
bbafd73b EJ |
6847 | xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto, |
6848 | const struct flow *flow, | |
6849 | const struct initial_vals *initial_vals, | |
6850 | struct rule_dpif *rule, uint8_t tcp_flags, | |
6851 | const struct ofpbuf *packet) | |
6852 | { | |
6853 | xin->ofproto = ofproto; | |
6854 | xin->flow = *flow; | |
6855 | xin->packet = packet; | |
6856 | xin->may_learn = packet != NULL; | |
6857 | xin->rule = rule; | |
6858 | xin->ofpacts = NULL; | |
6859 | xin->ofpacts_len = 0; | |
6860 | xin->tcp_flags = tcp_flags; | |
6861 | xin->resubmit_hook = NULL; | |
6862 | xin->report_hook = NULL; | |
6863 | xin->resubmit_stats = NULL; | |
6864 | ||
6865 | if (initial_vals) { | |
6866 | xin->initial_vals = *initial_vals; | |
6867 | } else { | |
6868 | xin->initial_vals.vlan_tci = xin->flow.vlan_tci; | |
6869 | } | |
6870 | } | |
6871 | ||
6872 | static void | |
6873 | xlate_out_uninit(struct xlate_out *xout) | |
abe529af | 6874 | { |
bbafd73b EJ |
6875 | if (xout) { |
6876 | ofpbuf_uninit(&xout->odp_actions); | |
6877 | } | |
6878 | } | |
6879 | ||
6880 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' | |
6881 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
6882 | static void | |
6883 | xlate_actions(struct xlate_in *xin, struct xlate_out *xout) | |
6884 | { | |
6885 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so | |
6886 | * that in the future we always keep a copy of the original flow for | |
6887 | * tracing purposes. */ | |
6888 | static bool hit_resubmit_limit; | |
6889 | ||
6890 | enum slow_path_reason special; | |
6891 | const struct ofpact *ofpacts; | |
6892 | struct ofport_dpif *in_port; | |
6893 | struct flow orig_flow; | |
6894 | struct xlate_ctx ctx; | |
6895 | size_t ofpacts_len; | |
6896 | ||
6897 | COVERAGE_INC(ofproto_dpif_xlate); | |
6898 | ||
ef506a7c JG |
6899 | /* Flow initialization rules: |
6900 | * - 'base_flow' must match the kernel's view of the packet at the | |
6901 | * time that action processing starts. 'flow' represents any | |
6902 | * transformations we wish to make through actions. | |
6903 | * - By default 'base_flow' and 'flow' are the same since the input | |
6904 | * packet matches the output before any actions are applied. | |
6905 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
6906 | * of the received packet as seen by the kernel. If we later output | |
6907 | * to another device without any modifications this will cause us to | |
6908 | * insert a new tag since the original one was stripped off by the | |
6909 | * VLAN device. | |
4110a57a JR |
6910 | * - Tunnel metadata as received is retained in 'flow'. This allows |
6911 | * tunnel metadata matching also in later tables. | |
6912 | * Since a kernel action for setting the tunnel metadata will only be | |
6913 | * generated with actual tunnel output, changing the tunnel metadata | |
6914 | * values in 'flow' (such as tun_id) will only have effect with a later | |
6915 | * tunnel output action. | |
ef506a7c JG |
6916 | * - Tunnel 'base_flow' is completely cleared since that is what the |
6917 | * kernel does. If we wish to maintain the original values an action | |
6918 | * needs to be generated. */ | |
6919 | ||
bbafd73b EJ |
6920 | ctx.xin = xin; |
6921 | ctx.xout = xout; | |
6922 | ||
6923 | ctx.ofproto = xin->ofproto; | |
6924 | ctx.rule = xin->rule; | |
6925 | ||
6926 | ctx.base_flow = ctx.xin->flow; | |
6927 | ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci; | |
6928 | memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel); | |
6929 | ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst; | |
6930 | ||
6931 | ctx.xout->tags = 0; | |
6932 | ctx.xout->slow = 0; | |
6933 | ctx.xout->has_learn = false; | |
6934 | ctx.xout->has_normal = false; | |
6935 | ctx.xout->has_fin_timeout = false; | |
6936 | ctx.xout->nf_output_iface = NF_OUT_DROP; | |
6937 | ctx.xout->mirrors = 0; | |
6938 | ||
6939 | ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub, | |
6940 | sizeof ctx.xout->odp_actions_stub); | |
6941 | ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE); | |
6942 | ||
6943 | ctx.recurse = 0; | |
6944 | ctx.max_resubmit_trigger = false; | |
6945 | ctx.orig_skb_priority = ctx.xin->flow.skb_priority; | |
6946 | ctx.table_id = 0; | |
6947 | ctx.exit = false; | |
6948 | ||
6949 | if (xin->ofpacts) { | |
6950 | ofpacts = xin->ofpacts; | |
6951 | ofpacts_len = xin->ofpacts_len; | |
6952 | } else if (xin->rule) { | |
6953 | ofpacts = xin->rule->up.ofpacts; | |
6954 | ofpacts_len = xin->rule->up.ofpacts_len; | |
6955 | } else { | |
6956 | NOT_REACHED(); | |
0f49659a | 6957 | } |
abe529af | 6958 | |
bbafd73b | 6959 | ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack); |
43d50bc8 | 6960 | |
bbafd73b | 6961 | if (ctx.ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 | 6962 | /* Do this conditionally because the copy is expensive enough that it |
9ba85077 | 6963 | * shows up in profiles. */ |
bbafd73b | 6964 | orig_flow = ctx.xin->flow; |
ccb7c863 BP |
6965 | } |
6966 | ||
bbafd73b EJ |
6967 | if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
6968 | switch (ctx.ofproto->up.frag_handling) { | |
7257b535 BP |
6969 | case OFPC_FRAG_NORMAL: |
6970 | /* We must pretend that transport ports are unavailable. */ | |
bbafd73b EJ |
6971 | ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0); |
6972 | ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0); | |
7257b535 BP |
6973 | break; |
6974 | ||
6975 | case OFPC_FRAG_DROP: | |
050ac423 | 6976 | return; |
7257b535 BP |
6977 | |
6978 | case OFPC_FRAG_REASM: | |
6979 | NOT_REACHED(); | |
6980 | ||
6981 | case OFPC_FRAG_NX_MATCH: | |
6982 | /* Nothing to do. */ | |
6983 | break; | |
f0fd1a17 PS |
6984 | |
6985 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
6986 | NOT_REACHED(); | |
7257b535 BP |
6987 | } |
6988 | } | |
6989 | ||
bbafd73b EJ |
6990 | in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port); |
6991 | special = process_special(ctx.ofproto, &ctx.xin->flow, in_port, | |
6992 | ctx.xin->packet); | |
6a7e895f | 6993 | if (special) { |
bbafd73b | 6994 | ctx.xout->slow = special; |
abe529af | 6995 | } else { |
6a6455e5 | 6996 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
14f94f9a | 6997 | struct initial_vals initial_vals; |
94aa0d19 | 6998 | size_t sample_actions_len; |
ee382d89 | 6999 | uint32_t local_odp_port; |
6a6455e5 | 7000 | |
bbafd73b | 7001 | initial_vals.vlan_tci = ctx.base_flow.vlan_tci; |
14f94f9a | 7002 | |
bbafd73b EJ |
7003 | add_sflow_action(&ctx); |
7004 | add_ipfix_action(&ctx); | |
7005 | sample_actions_len = ctx.xout->odp_actions.size; | |
ffaef958 | 7006 | |
bbafd73b EJ |
7007 | if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) { |
7008 | do_xlate_actions(ofpacts, ofpacts_len, &ctx); | |
ffaef958 BP |
7009 | |
7010 | /* We've let OFPP_NORMAL and the learning action look at the | |
7011 | * packet, so drop it now if forwarding is disabled. */ | |
7012 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
bbafd73b | 7013 | ctx.xout->odp_actions.size = sample_actions_len; |
ffaef958 BP |
7014 | } |
7015 | } | |
abe529af | 7016 | |
bbafd73b | 7017 | if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) { |
43d50bc8 BP |
7018 | if (!hit_resubmit_limit) { |
7019 | /* We didn't record the original flow. Make sure we do from | |
7020 | * now on. */ | |
7021 | hit_resubmit_limit = true; | |
7022 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
7023 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7024 | ||
bbafd73b | 7025 | ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, |
14f94f9a | 7026 | &initial_vals, &ds); |
43d50bc8 BP |
7027 | VLOG_ERR("Trace triggered by excessive resubmit " |
7028 | "recursion:\n%s", ds_cstr(&ds)); | |
7029 | ds_destroy(&ds); | |
7030 | } | |
6a6455e5 EJ |
7031 | } |
7032 | ||
bbafd73b EJ |
7033 | local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL); |
7034 | if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow, | |
454a77e5 | 7035 | local_odp_port, |
bbafd73b EJ |
7036 | ctx.xout->odp_actions.data, |
7037 | ctx.xout->odp_actions.size)) { | |
7038 | compose_output_action(&ctx, OFPP_LOCAL); | |
b6848f13 | 7039 | } |
bbafd73b EJ |
7040 | if (ctx.ofproto->has_mirrors) { |
7041 | add_mirror_actions(&ctx, &orig_flow); | |
ccb7c863 | 7042 | } |
bbafd73b | 7043 | fix_sflow_action(&ctx); |
abe529af | 7044 | } |
bd85dac1 | 7045 | |
bbafd73b | 7046 | ofpbuf_uninit(&ctx.stack); |
050ac423 BP |
7047 | } |
7048 | ||
f25d0cf3 BP |
7049 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
7050 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 | 7051 | static void |
bbafd73b | 7052 | xlate_actions_for_side_effects(struct xlate_in *xin) |
050ac423 | 7053 | { |
bbafd73b | 7054 | struct xlate_out xout; |
abe529af | 7055 | |
bbafd73b EJ |
7056 | xlate_actions(xin, &xout); |
7057 | xlate_out_uninit(&xout); | |
abe529af | 7058 | } |
479df176 BP |
7059 | |
7060 | static void | |
f03a84b9 | 7061 | xlate_report(struct xlate_ctx *ctx, const char *s) |
479df176 | 7062 | { |
bbafd73b EJ |
7063 | if (ctx->xin->report_hook) { |
7064 | ctx->xin->report_hook(ctx, s); | |
479df176 BP |
7065 | } |
7066 | } | |
abe529af BP |
7067 | \f |
7068 | /* OFPP_NORMAL implementation. */ | |
7069 | ||
abe529af BP |
7070 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
7071 | ||
ecac4ebf BP |
7072 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
7073 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
7074 | * the bundle on which the packet was received, returns the VLAN to which the | |
7075 | * packet belongs. | |
7076 | * | |
7077 | * Both 'vid' and the return value are in the range 0...4095. */ | |
7078 | static uint16_t | |
7079 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
7080 | { | |
7081 | switch (in_bundle->vlan_mode) { | |
7082 | case PORT_VLAN_ACCESS: | |
7083 | return in_bundle->vlan; | |
7084 | break; | |
7085 | ||
7086 | case PORT_VLAN_TRUNK: | |
7087 | return vid; | |
7088 | ||
7089 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7090 | case PORT_VLAN_NATIVE_TAGGED: | |
7091 | return vid ? vid : in_bundle->vlan; | |
7092 | ||
7093 | default: | |
7094 | NOT_REACHED(); | |
7095 | } | |
7096 | } | |
7097 | ||
5da5ec37 BP |
7098 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
7099 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
7100 | * a warning. | |
7101 | * | |
7102 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
7103 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
7104 | * 0...4095. */ | |
7105 | static bool | |
7106 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
7107 | { | |
33158a18 JP |
7108 | /* Allow any VID on the OFPP_NONE port. */ |
7109 | if (in_bundle == &ofpp_none_bundle) { | |
7110 | return true; | |
7111 | } | |
7112 | ||
5da5ec37 BP |
7113 | switch (in_bundle->vlan_mode) { |
7114 | case PORT_VLAN_ACCESS: | |
7115 | if (vid) { | |
7116 | if (warn) { | |
7117 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7118 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
7119 | "packet received on port %s configured as VLAN " | |
7120 | "%"PRIu16" access port", | |
7121 | in_bundle->ofproto->up.name, vid, | |
7122 | in_bundle->name, in_bundle->vlan); | |
7123 | } | |
7124 | return false; | |
7125 | } | |
7126 | return true; | |
7127 | ||
7128 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7129 | case PORT_VLAN_NATIVE_TAGGED: | |
7130 | if (!vid) { | |
7131 | /* Port must always carry its native VLAN. */ | |
7132 | return true; | |
7133 | } | |
7134 | /* Fall through. */ | |
7135 | case PORT_VLAN_TRUNK: | |
7136 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
7137 | if (warn) { | |
7138 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7139 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
7140 | "received on port %s not configured for trunking " | |
7141 | "VLAN %"PRIu16, | |
7142 | in_bundle->ofproto->up.name, vid, | |
7143 | in_bundle->name, vid); | |
7144 | } | |
7145 | return false; | |
7146 | } | |
7147 | return true; | |
7148 | ||
7149 | default: | |
7150 | NOT_REACHED(); | |
7151 | } | |
7152 | ||
7153 | } | |
7154 | ||
ecac4ebf BP |
7155 | /* Given 'vlan', the VLAN that a packet belongs to, and |
7156 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
7157 | * that should be included in the 802.1Q header. (If the return value is 0, | |
7158 | * then the 802.1Q header should only be included in the packet if there is a | |
7159 | * nonzero PCP.) | |
7160 | * | |
7161 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
7162 | static uint16_t | |
7163 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
7164 | { | |
7165 | switch (out_bundle->vlan_mode) { | |
7166 | case PORT_VLAN_ACCESS: | |
7167 | return 0; | |
7168 | ||
7169 | case PORT_VLAN_TRUNK: | |
7170 | case PORT_VLAN_NATIVE_TAGGED: | |
7171 | return vlan; | |
7172 | ||
7173 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7174 | return vlan == out_bundle->vlan ? 0 : vlan; | |
7175 | ||
7176 | default: | |
7177 | NOT_REACHED(); | |
7178 | } | |
7179 | } | |
7180 | ||
395e68ce | 7181 | static void |
f03a84b9 | 7182 | output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle, |
395e68ce | 7183 | uint16_t vlan) |
abe529af | 7184 | { |
395e68ce BP |
7185 | struct ofport_dpif *port; |
7186 | uint16_t vid; | |
81b1afb1 | 7187 | ovs_be16 tci, old_tci; |
ecac4ebf | 7188 | |
395e68ce BP |
7189 | vid = output_vlan_to_vid(out_bundle, vlan); |
7190 | if (!out_bundle->bond) { | |
7191 | port = ofbundle_get_a_port(out_bundle); | |
7192 | } else { | |
bbafd73b EJ |
7193 | port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow, |
7194 | vid, &ctx->xout->tags); | |
395e68ce BP |
7195 | if (!port) { |
7196 | /* No slaves enabled, so drop packet. */ | |
7197 | return; | |
7198 | } | |
7199 | } | |
abe529af | 7200 | |
bbafd73b | 7201 | old_tci = ctx->xin->flow.vlan_tci; |
5e9ceccd BP |
7202 | tci = htons(vid); |
7203 | if (tci || out_bundle->use_priority_tags) { | |
bbafd73b | 7204 | tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK); |
5e9ceccd BP |
7205 | if (tci) { |
7206 | tci |= htons(VLAN_CFI); | |
7207 | } | |
395e68ce | 7208 | } |
bbafd73b | 7209 | ctx->xin->flow.vlan_tci = tci; |
395e68ce | 7210 | |
5e48dc2b | 7211 | compose_output_action(ctx, port->up.ofp_port); |
bbafd73b | 7212 | ctx->xin->flow.vlan_tci = old_tci; |
abe529af BP |
7213 | } |
7214 | ||
7215 | static int | |
7216 | mirror_mask_ffs(mirror_mask_t mask) | |
7217 | { | |
7218 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
7219 | return ffs(mask); | |
7220 | } | |
7221 | ||
abe529af BP |
7222 | static bool |
7223 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7224 | { | |
ecac4ebf | 7225 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 7226 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
7227 | } |
7228 | ||
7229 | static bool | |
7230 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7231 | { | |
7232 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
7233 | } | |
7234 | ||
7235 | /* Returns an arbitrary interface within 'bundle'. */ | |
7236 | static struct ofport_dpif * | |
7237 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
7238 | { | |
7239 | return CONTAINER_OF(list_front(&bundle->ports), | |
7240 | struct ofport_dpif, bundle_node); | |
7241 | } | |
7242 | ||
abe529af BP |
7243 | static bool |
7244 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
7245 | { | |
fc3d7408 | 7246 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
7247 | } |
7248 | ||
7249 | static void | |
f03a84b9 | 7250 | add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
7251 | { |
7252 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7253 | mirror_mask_t mirrors; | |
c06bba01 JP |
7254 | struct ofbundle *in_bundle; |
7255 | uint16_t vlan; | |
7256 | uint16_t vid; | |
7257 | const struct nlattr *a; | |
7258 | size_t left; | |
7259 | ||
3581c12c | 7260 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
bbafd73b | 7261 | ctx->xin->packet != NULL, NULL); |
3581c12c | 7262 | if (!in_bundle) { |
c06bba01 JP |
7263 | return; |
7264 | } | |
c06bba01 JP |
7265 | mirrors = in_bundle->src_mirrors; |
7266 | ||
7267 | /* Drop frames on bundles reserved for mirroring. */ | |
7268 | if (in_bundle->mirror_out) { | |
bbafd73b | 7269 | if (ctx->xin->packet != NULL) { |
c06bba01 JP |
7270 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7271 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7272 | "%s, which is reserved exclusively for mirroring", | |
7273 | ctx->ofproto->up.name, in_bundle->name); | |
7274 | } | |
7275 | return; | |
7276 | } | |
7277 | ||
7278 | /* Check VLAN. */ | |
7279 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
bbafd73b | 7280 | if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) { |
c06bba01 JP |
7281 | return; |
7282 | } | |
7283 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7284 | ||
7285 | /* Look at the output ports to check for destination selections. */ | |
7286 | ||
bbafd73b EJ |
7287 | NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data, |
7288 | ctx->xout->odp_actions.size) { | |
c06bba01 JP |
7289 | enum ovs_action_attr type = nl_attr_type(a); |
7290 | struct ofport_dpif *ofport; | |
7291 | ||
7292 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
7293 | continue; | |
7294 | } | |
7295 | ||
7296 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
7297 | if (ofport && ofport->bundle) { |
7298 | mirrors |= ofport->bundle->dst_mirrors; | |
7299 | } | |
c06bba01 | 7300 | } |
abe529af BP |
7301 | |
7302 | if (!mirrors) { | |
7303 | return; | |
7304 | } | |
7305 | ||
c06bba01 | 7306 | /* Restore the original packet before adding the mirror actions. */ |
bbafd73b | 7307 | ctx->xin->flow = *orig_flow; |
c06bba01 | 7308 | |
9ba15e2a BP |
7309 | while (mirrors) { |
7310 | struct ofmirror *m; | |
9ba15e2a BP |
7311 | |
7312 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7313 | ||
7314 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 7315 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
7316 | continue; |
7317 | } | |
7318 | ||
7319 | mirrors &= ~m->dup_mirrors; | |
bbafd73b | 7320 | ctx->xout->mirrors |= m->dup_mirrors; |
9ba15e2a | 7321 | if (m->out) { |
395e68ce | 7322 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
7323 | } else if (vlan != m->out_vlan |
7324 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
7325 | struct ofbundle *bundle; |
7326 | ||
7327 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
7328 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
7329 | && !bundle->mirror_out) { |
7330 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
7331 | } |
7332 | } | |
7333 | } | |
abe529af BP |
7334 | } |
7335 | } | |
7336 | ||
9d24de3b JP |
7337 | static void |
7338 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
7339 | uint64_t packets, uint64_t bytes) | |
7340 | { | |
7341 | if (!mirrors) { | |
7342 | return; | |
7343 | } | |
7344 | ||
8472a3ce | 7345 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
7346 | struct ofmirror *m; |
7347 | ||
7348 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7349 | ||
7350 | if (!m) { | |
7351 | /* In normal circumstances 'm' will not be NULL. However, | |
7352 | * if mirrors are reconfigured, we can temporarily get out | |
7353 | * of sync in facet_revalidate(). We could "correct" the | |
7354 | * mirror list before reaching here, but doing that would | |
7355 | * not properly account the traffic stats we've currently | |
7356 | * accumulated for previous mirror configuration. */ | |
7357 | continue; | |
7358 | } | |
7359 | ||
7360 | m->packet_count += packets; | |
7361 | m->byte_count += bytes; | |
7362 | } | |
7363 | } | |
7364 | ||
abe529af BP |
7365 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
7366 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7367 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7368 | static bool | |
7369 | is_gratuitous_arp(const struct flow *flow) | |
7370 | { | |
7371 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7372 | && eth_addr_is_broadcast(flow->dl_dst) | |
7373 | && (flow->nw_proto == ARP_OP_REPLY | |
7374 | || (flow->nw_proto == ARP_OP_REQUEST | |
7375 | && flow->nw_src == flow->nw_dst))); | |
7376 | } | |
7377 | ||
7378 | static void | |
7379 | update_learning_table(struct ofproto_dpif *ofproto, | |
7380 | const struct flow *flow, int vlan, | |
7381 | struct ofbundle *in_bundle) | |
7382 | { | |
7383 | struct mac_entry *mac; | |
7384 | ||
33158a18 JP |
7385 | /* Don't learn the OFPP_NONE port. */ |
7386 | if (in_bundle == &ofpp_none_bundle) { | |
7387 | return; | |
7388 | } | |
7389 | ||
abe529af BP |
7390 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
7391 | return; | |
7392 | } | |
7393 | ||
7394 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7395 | if (is_gratuitous_arp(flow)) { | |
7396 | /* We don't want to learn from gratuitous ARP packets that are | |
7397 | * reflected back over bond slaves so we lock the learning table. */ | |
7398 | if (!in_bundle->bond) { | |
7399 | mac_entry_set_grat_arp_lock(mac); | |
7400 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7401 | return; | |
7402 | } | |
7403 | } | |
7404 | ||
7405 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7406 | /* The log messages here could actually be useful in debugging, | |
7407 | * so keep the rate limit relatively high. */ | |
7408 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7409 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7410 | "on port %s in VLAN %d", | |
7411 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7412 | in_bundle->name, vlan); | |
7413 | ||
7414 | mac->port.p = in_bundle; | |
2cc3c58e | 7415 | tag_set_add(&ofproto->backer->revalidate_set, |
abe529af BP |
7416 | mac_learning_changed(ofproto->ml, mac)); |
7417 | } | |
7418 | } | |
7419 | ||
3581c12c | 7420 | static struct ofbundle * |
4acbc98d SH |
7421 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
7422 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
7423 | { |
7424 | struct ofport_dpif *ofport; | |
7425 | ||
7426 | /* Find the port and bundle for the received packet. */ | |
7427 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
7428 | if (in_ofportp) { |
7429 | *in_ofportp = ofport; | |
7430 | } | |
395e68ce | 7431 | if (ofport && ofport->bundle) { |
3581c12c | 7432 | return ofport->bundle; |
395e68ce BP |
7433 | } |
7434 | ||
70c2fd56 BP |
7435 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
7436 | * port for traffic that it is sourcing. */ | |
7437 | if (in_port == OFPP_NONE) { | |
7438 | return &ofpp_none_bundle; | |
7439 | } | |
7440 | ||
395e68ce BP |
7441 | /* Odd. A few possible reasons here: |
7442 | * | |
7443 | * - We deleted a port but there are still a few packets queued up | |
7444 | * from it. | |
7445 | * | |
7446 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7447 | * we don't know about. | |
7448 | * | |
7449 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
7450 | * This is particularly likely to happen if a packet was received on the |
7451 | * port after it was created, but before the client had a chance to | |
7452 | * configure its bundle. | |
395e68ce BP |
7453 | */ |
7454 | if (warn) { | |
7455 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7456 | ||
7457 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7458 | "port %"PRIu16, ofproto->up.name, in_port); | |
7459 | } | |
7460 | return NULL; | |
7461 | } | |
7462 | ||
5da5ec37 | 7463 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
7464 | * dropped. Returns true if they may be forwarded, false if they should be |
7465 | * dropped. | |
7466 | * | |
395e68ce BP |
7467 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
7468 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 7469 | * |
395e68ce BP |
7470 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
7471 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7472 | * checked by input_vid_is_valid(). | |
abe529af BP |
7473 | * |
7474 | * May also add tags to '*tags', although the current implementation only does | |
7475 | * so in one special case. | |
7476 | */ | |
7477 | static bool | |
f03a84b9 | 7478 | is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port, |
479df176 | 7479 | uint16_t vlan) |
abe529af | 7480 | { |
479df176 | 7481 | struct ofproto_dpif *ofproto = ctx->ofproto; |
bbafd73b | 7482 | struct flow *flow = &ctx->xin->flow; |
395e68ce | 7483 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 7484 | |
395e68ce BP |
7485 | /* Drop frames for reserved multicast addresses |
7486 | * only if forward_bpdu option is absent. */ | |
614ec445 | 7487 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 7488 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
7489 | return false; |
7490 | } | |
7491 | ||
abe529af BP |
7492 | if (in_bundle->bond) { |
7493 | struct mac_entry *mac; | |
7494 | ||
7495 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
bbafd73b | 7496 | flow->dl_dst, &ctx->xout->tags)) { |
abe529af BP |
7497 | case BV_ACCEPT: |
7498 | break; | |
7499 | ||
7500 | case BV_DROP: | |
479df176 | 7501 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
7502 | return false; |
7503 | ||
7504 | case BV_DROP_IF_MOVED: | |
7505 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7506 | if (mac && mac->port.p != in_bundle && | |
7507 | (!is_gratuitous_arp(flow) | |
7508 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
7509 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
7510 | "dropping"); | |
abe529af BP |
7511 | return false; |
7512 | } | |
7513 | break; | |
7514 | } | |
7515 | } | |
7516 | ||
7517 | return true; | |
7518 | } | |
7519 | ||
4cd78906 | 7520 | static void |
f03a84b9 | 7521 | xlate_normal(struct xlate_ctx *ctx) |
abe529af | 7522 | { |
395e68ce | 7523 | struct ofport_dpif *in_port; |
abe529af | 7524 | struct ofbundle *in_bundle; |
abe529af | 7525 | struct mac_entry *mac; |
395e68ce BP |
7526 | uint16_t vlan; |
7527 | uint16_t vid; | |
abe529af | 7528 | |
bbafd73b | 7529 | ctx->xout->has_normal = true; |
75a75043 | 7530 | |
bbafd73b EJ |
7531 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port, |
7532 | ctx->xin->packet != NULL, &in_port); | |
3581c12c | 7533 | if (!in_bundle) { |
479df176 | 7534 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
7535 | return; |
7536 | } | |
3581c12c | 7537 | |
395e68ce | 7538 | /* Drop malformed frames. */ |
bbafd73b EJ |
7539 | if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) && |
7540 | !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) { | |
7541 | if (ctx->xin->packet != NULL) { | |
395e68ce BP |
7542 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7543 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7544 | "VLAN tag received on port %s", | |
7545 | ctx->ofproto->up.name, in_bundle->name); | |
7546 | } | |
479df176 | 7547 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
7548 | return; |
7549 | } | |
7550 | ||
7551 | /* Drop frames on bundles reserved for mirroring. */ | |
7552 | if (in_bundle->mirror_out) { | |
bbafd73b | 7553 | if (ctx->xin->packet != NULL) { |
395e68ce BP |
7554 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
7555 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7556 | "%s, which is reserved exclusively for mirroring", | |
7557 | ctx->ofproto->up.name, in_bundle->name); | |
7558 | } | |
479df176 | 7559 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
7560 | return; |
7561 | } | |
7562 | ||
7563 | /* Check VLAN. */ | |
bbafd73b EJ |
7564 | vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci); |
7565 | if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) { | |
479df176 | 7566 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
7567 | return; |
7568 | } | |
7569 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7570 | ||
7571 | /* Check other admissibility requirements. */ | |
479df176 | 7572 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 7573 | return; |
abe529af BP |
7574 | } |
7575 | ||
75a75043 | 7576 | /* Learn source MAC. */ |
bbafd73b EJ |
7577 | if (ctx->xin->may_learn) { |
7578 | update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle); | |
abe529af BP |
7579 | } |
7580 | ||
7581 | /* Determine output bundle. */ | |
bbafd73b EJ |
7582 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan, |
7583 | &ctx->xout->tags); | |
abe529af | 7584 | if (mac) { |
c06bba01 | 7585 | if (mac->port.p != in_bundle) { |
479df176 | 7586 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 7587 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
7588 | } else { |
7589 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 7590 | } |
abe529af | 7591 | } else { |
c06bba01 | 7592 | struct ofbundle *bundle; |
abe529af | 7593 | |
479df176 | 7594 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
7595 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
7596 | if (bundle != in_bundle | |
7597 | && ofbundle_includes_vlan(bundle, vlan) | |
7598 | && bundle->floodable | |
7599 | && !bundle->mirror_out) { | |
7600 | output_normal(ctx, bundle, vlan); | |
7601 | } | |
7602 | } | |
bbafd73b | 7603 | ctx->xout->nf_output_iface = NF_OUT_FLOOD; |
abe529af | 7604 | } |
abe529af BP |
7605 | } |
7606 | \f | |
54a9cbc9 BP |
7607 | /* Optimized flow revalidation. |
7608 | * | |
7609 | * It's a difficult problem, in general, to tell which facets need to have | |
7610 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7611 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7612 | * table changes, we recalculate the actions for every facet. This is | |
7613 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7614 | * doesn't change very often. | |
7615 | * | |
7616 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7617 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7618 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7619 | * table, we add a special case that applies to flow tables in which every rule | |
7620 | * has the same form (that is, the same wildcards), except that the table is | |
7621 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7622 | * optimize this case by tagging all of the facets that resubmit into the table | |
7623 | * and invalidating the same tag whenever a flow changes in that table. The | |
7624 | * end result is that we revalidate just the facets that need it (and sometimes | |
7625 | * a few more, but not all of the facets or even all of the facets that | |
7626 | * resubmit to the table modified by MAC learning). */ | |
7627 | ||
5cb7a798 | 7628 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted |
54a9cbc9 | 7629 | * into an OpenFlow table with the given 'basis'. */ |
822d9414 | 7630 | static tag_type |
5cb7a798 | 7631 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, |
54a9cbc9 BP |
7632 | uint32_t secret) |
7633 | { | |
5cb7a798 | 7634 | if (minimask_is_catchall(mask)) { |
54a9cbc9 BP |
7635 | return 0; |
7636 | } else { | |
5cb7a798 BP |
7637 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); |
7638 | return tag_create_deterministic(hash); | |
54a9cbc9 BP |
7639 | } |
7640 | } | |
7641 | ||
7642 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7643 | * taggability of that table. | |
7644 | * | |
7645 | * This function must be called after *each* change to a flow table. If you | |
7646 | * skip calling it on some changes then the pointer comparisons at the end can | |
7647 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7648 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7649 | * different wildcards to be created with the same address, then this function | |
7650 | * will incorrectly skip revalidation. */ | |
7651 | static void | |
7652 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7653 | { | |
7654 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 7655 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
7656 | struct cls_table *catchall, *other; |
7657 | struct cls_table *t; | |
7658 | ||
7659 | catchall = other = NULL; | |
7660 | ||
d0918789 | 7661 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
7662 | case 0: |
7663 | /* We could tag this OpenFlow table but it would make the logic a | |
7664 | * little harder and it's a corner case that doesn't seem worth it | |
7665 | * yet. */ | |
7666 | break; | |
7667 | ||
7668 | case 1: | |
7669 | case 2: | |
d0918789 | 7670 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
7671 | if (cls_table_is_catchall(t)) { |
7672 | catchall = t; | |
7673 | } else if (!other) { | |
7674 | other = t; | |
7675 | } else { | |
7676 | /* Indicate that we can't tag this by setting both tables to | |
7677 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7678 | other = NULL; | |
7679 | } | |
7680 | } | |
7681 | break; | |
7682 | ||
7683 | default: | |
7684 | /* Can't tag this table. */ | |
7685 | break; | |
7686 | } | |
7687 | ||
7688 | if (table->catchall_table != catchall || table->other_table != other) { | |
7689 | table->catchall_table = catchall; | |
7690 | table->other_table = other; | |
2cc3c58e | 7691 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7692 | } |
7693 | } | |
7694 | ||
7695 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7696 | * inserted, a rule being deleted, or a rule whose actions are being | |
7697 | * modified), marks facets for revalidation to ensure that packets will be | |
7698 | * forwarded correctly according to the new state of the flow table. | |
7699 | * | |
7700 | * This function must be called after *each* change to a flow table. See | |
7701 | * the comment on table_update_taggable() for more information. */ | |
7702 | static void | |
7703 | rule_invalidate(const struct rule_dpif *rule) | |
7704 | { | |
7705 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7706 | ||
7707 | table_update_taggable(ofproto, rule->up.table_id); | |
7708 | ||
2cc3c58e | 7709 | if (!ofproto->backer->need_revalidate) { |
54a9cbc9 BP |
7710 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; |
7711 | ||
7712 | if (table->other_table && rule->tag) { | |
2cc3c58e | 7713 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); |
54a9cbc9 | 7714 | } else { |
2cc3c58e | 7715 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7716 | } |
7717 | } | |
7718 | } | |
7719 | \f | |
abe529af | 7720 | static bool |
7257b535 BP |
7721 | set_frag_handling(struct ofproto *ofproto_, |
7722 | enum ofp_config_flags frag_handling) | |
abe529af BP |
7723 | { |
7724 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 7725 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 7726 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
7727 | return true; |
7728 | } else { | |
7729 | return false; | |
7730 | } | |
abe529af BP |
7731 | } |
7732 | ||
90bf1e07 | 7733 | static enum ofperr |
abe529af BP |
7734 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
7735 | const struct flow *flow, | |
f25d0cf3 | 7736 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
7737 | { |
7738 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
14f94f9a | 7739 | struct initial_vals initial_vals; |
548de4dd BP |
7740 | struct odputil_keybuf keybuf; |
7741 | struct dpif_flow_stats stats; | |
bbafd73b EJ |
7742 | struct xlate_out xout; |
7743 | struct xlate_in xin; | |
548de4dd | 7744 | struct ofpbuf key; |
112bc5f4 | 7745 | |
80e5eed9 | 7746 | |
548de4dd | 7747 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
7748 | odp_flow_key_from_flow(&key, flow, |
7749 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
050ac423 | 7750 | |
548de4dd | 7751 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 7752 | |
14f94f9a | 7753 | initial_vals.vlan_tci = flow->vlan_tci; |
bbafd73b EJ |
7754 | xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags, |
7755 | packet); | |
7756 | xin.resubmit_stats = &stats; | |
7757 | xin.ofpacts_len = ofpacts_len; | |
7758 | xin.ofpacts = ofpacts; | |
2284188b | 7759 | |
bbafd73b | 7760 | xlate_actions(&xin, &xout); |
acf60855 | 7761 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
bbafd73b EJ |
7762 | xout.odp_actions.data, xout.odp_actions.size, packet); |
7763 | xlate_out_uninit(&xout); | |
2284188b | 7764 | |
548de4dd | 7765 | return 0; |
abe529af | 7766 | } |
6fca1ffb BP |
7767 | \f |
7768 | /* NetFlow. */ | |
7769 | ||
7770 | static int | |
7771 | set_netflow(struct ofproto *ofproto_, | |
7772 | const struct netflow_options *netflow_options) | |
7773 | { | |
7774 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7775 | ||
7776 | if (netflow_options) { | |
7777 | if (!ofproto->netflow) { | |
7778 | ofproto->netflow = netflow_create(); | |
7779 | } | |
7780 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7781 | } else { | |
7782 | netflow_destroy(ofproto->netflow); | |
7783 | ofproto->netflow = NULL; | |
7784 | return 0; | |
7785 | } | |
7786 | } | |
abe529af BP |
7787 | |
7788 | static void | |
7789 | get_netflow_ids(const struct ofproto *ofproto_, | |
7790 | uint8_t *engine_type, uint8_t *engine_id) | |
7791 | { | |
7792 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7793 | ||
acf60855 | 7794 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 7795 | } |
6fca1ffb BP |
7796 | |
7797 | static void | |
7798 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7799 | { | |
7800 | if (!facet_is_controller_flow(facet) && | |
7801 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 7802 | struct subfacet *subfacet; |
6fca1ffb BP |
7803 | struct ofexpired expired; |
7804 | ||
b0f7b9b5 | 7805 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 7806 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 7807 | struct dpif_flow_stats stats; |
6fca1ffb | 7808 | |
bbafd73b | 7809 | subfacet_install(subfacet, &facet->xout.odp_actions, &stats); |
15baa734 | 7810 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 7811 | } |
6fca1ffb BP |
7812 | } |
7813 | ||
7814 | expired.flow = facet->flow; | |
7815 | expired.packet_count = facet->packet_count; | |
7816 | expired.byte_count = facet->byte_count; | |
7817 | expired.used = facet->used; | |
7818 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7819 | } | |
7820 | } | |
7821 | ||
7822 | static void | |
7823 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7824 | { | |
7825 | struct facet *facet; | |
7826 | ||
7827 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7828 | send_active_timeout(ofproto, facet); | |
7829 | } | |
7830 | } | |
abe529af BP |
7831 | \f |
7832 | static struct ofproto_dpif * | |
7833 | ofproto_dpif_lookup(const char *name) | |
7834 | { | |
b44a10b7 BP |
7835 | struct ofproto_dpif *ofproto; |
7836 | ||
7837 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7838 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7839 | if (!strcmp(ofproto->up.name, name)) { | |
7840 | return ofproto; | |
7841 | } | |
7842 | } | |
7843 | return NULL; | |
abe529af BP |
7844 | } |
7845 | ||
f0a3aa2e | 7846 | static void |
96e466a3 | 7847 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 7848 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 7849 | { |
490df1ef | 7850 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 7851 | |
96e466a3 EJ |
7852 | if (argc > 1) { |
7853 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7854 | if (!ofproto) { | |
bde9f75d | 7855 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
7856 | return; |
7857 | } | |
2cc3c58e | 7858 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 EJ |
7859 | } else { |
7860 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2cc3c58e | 7861 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 | 7862 | } |
f0a3aa2e | 7863 | } |
f0a3aa2e | 7864 | |
bde9f75d | 7865 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
7866 | } |
7867 | ||
abe529af | 7868 | static void |
0e15264f BP |
7869 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
7870 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
7871 | { |
7872 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7873 | const struct ofproto_dpif *ofproto; | |
7874 | const struct mac_entry *e; | |
7875 | ||
0e15264f | 7876 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 7877 | if (!ofproto) { |
bde9f75d | 7878 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
7879 | return; |
7880 | } | |
7881 | ||
7882 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
7883 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
7884 | struct ofbundle *bundle = e->port.p; | |
7885 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
7886 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
7887 | e->vlan, ETH_ADDR_ARGS(e->mac), |
7888 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 7889 | } |
bde9f75d | 7890 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
7891 | ds_destroy(&ds); |
7892 | } | |
7893 | ||
6a6455e5 | 7894 | struct trace_ctx { |
bbafd73b EJ |
7895 | struct xlate_out xout; |
7896 | struct xlate_in xin; | |
abe529af BP |
7897 | struct flow flow; |
7898 | struct ds *result; | |
7899 | }; | |
7900 | ||
7901 | static void | |
29901626 BP |
7902 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
7903 | const struct rule_dpif *rule) | |
abe529af BP |
7904 | { |
7905 | ds_put_char_multiple(result, '\t', level); | |
7906 | if (!rule) { | |
7907 | ds_put_cstr(result, "No match\n"); | |
7908 | return; | |
7909 | } | |
7910 | ||
29901626 BP |
7911 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
7912 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 7913 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
7914 | ds_put_char(result, '\n'); |
7915 | ||
7916 | ds_put_char_multiple(result, '\t', level); | |
7917 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 7918 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
7919 | ds_put_char(result, '\n'); |
7920 | } | |
7921 | ||
7922 | static void | |
7923 | trace_format_flow(struct ds *result, int level, const char *title, | |
bbafd73b | 7924 | struct trace_ctx *trace) |
abe529af BP |
7925 | { |
7926 | ds_put_char_multiple(result, '\t', level); | |
7927 | ds_put_format(result, "%s: ", title); | |
bbafd73b | 7928 | if (flow_equal(&trace->xin.flow, &trace->flow)) { |
abe529af BP |
7929 | ds_put_cstr(result, "unchanged"); |
7930 | } else { | |
bbafd73b EJ |
7931 | flow_format(result, &trace->xin.flow); |
7932 | trace->flow = trace->xin.flow; | |
abe529af BP |
7933 | } |
7934 | ds_put_char(result, '\n'); | |
7935 | } | |
7936 | ||
eb9e1c26 EJ |
7937 | static void |
7938 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 7939 | struct trace_ctx *trace) |
eb9e1c26 EJ |
7940 | { |
7941 | size_t i; | |
7942 | ||
7943 | ds_put_char_multiple(result, '\t', level); | |
7944 | ds_put_format(result, "%s:", title); | |
7945 | for (i = 0; i < FLOW_N_REGS; i++) { | |
7946 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
7947 | } | |
7948 | ds_put_char(result, '\n'); | |
7949 | } | |
7950 | ||
1ed8d352 EJ |
7951 | static void |
7952 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 7953 | struct trace_ctx *trace) |
1ed8d352 | 7954 | { |
bbafd73b | 7955 | struct ofpbuf *odp_actions = &trace->xout.odp_actions; |
1ed8d352 EJ |
7956 | |
7957 | ds_put_char_multiple(result, '\t', level); | |
7958 | ds_put_format(result, "%s: ", title); | |
7959 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
7960 | ds_put_char(result, '\n'); | |
7961 | } | |
7962 | ||
abe529af | 7963 | static void |
f03a84b9 | 7964 | trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule) |
abe529af | 7965 | { |
bbafd73b | 7966 | struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin); |
abe529af BP |
7967 | struct ds *result = trace->result; |
7968 | ||
7969 | ds_put_char(result, '\n'); | |
7970 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 7971 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 7972 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 7973 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
7974 | } |
7975 | ||
479df176 | 7976 | static void |
f03a84b9 | 7977 | trace_report(struct xlate_ctx *ctx, const char *s) |
479df176 | 7978 | { |
bbafd73b | 7979 | struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin); |
479df176 BP |
7980 | struct ds *result = trace->result; |
7981 | ||
7982 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
7983 | ds_put_cstr(result, s); | |
7984 | ds_put_char(result, '\n'); | |
7985 | } | |
7986 | ||
abe529af | 7987 | static void |
0e15264f | 7988 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
7989 | void *aux OVS_UNUSED) |
7990 | { | |
50aa28fd | 7991 | const struct dpif_backer *backer; |
abe529af | 7992 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
7993 | struct ofpbuf odp_key; |
7994 | struct ofpbuf *packet; | |
14f94f9a | 7995 | struct initial_vals initial_vals; |
abe529af BP |
7996 | struct ds result; |
7997 | struct flow flow; | |
abe529af BP |
7998 | char *s; |
7999 | ||
876b0e1c | 8000 | packet = NULL; |
50aa28fd | 8001 | backer = NULL; |
abe529af | 8002 | ds_init(&result); |
50aa28fd | 8003 | ofpbuf_init(&odp_key, 0); |
abe529af | 8004 | |
50aa28fd AW |
8005 | /* Handle "-generate" or a hex string as the last argument. */ |
8006 | if (!strcmp(argv[argc - 1], "-generate")) { | |
8007 | packet = ofpbuf_new(0); | |
8008 | argc--; | |
8009 | } else { | |
8010 | const char *error = eth_from_hex(argv[argc - 1], &packet); | |
8011 | if (!error) { | |
8012 | argc--; | |
8013 | } else if (argc == 4) { | |
8014 | /* The 3-argument form must end in "-generate' or a hex string. */ | |
8015 | unixctl_command_reply_error(conn, error); | |
8016 | goto exit; | |
8017 | } | |
e84173dc | 8018 | } |
876b0e1c | 8019 | |
50aa28fd AW |
8020 | /* Parse the flow and determine whether a datapath or |
8021 | * bridge is specified. If function odp_flow_key_from_string() | |
8022 | * returns 0, the flow is a odp_flow. If function | |
8023 | * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ | |
8024 | if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) { | |
8025 | /* If the odp_flow is the second argument, | |
8026 | * the datapath name is the first argument. */ | |
8027 | if (argc == 3) { | |
8028 | const char *dp_type; | |
8029 | if (!strncmp(argv[1], "ovs-", 4)) { | |
8030 | dp_type = argv[1] + 4; | |
8031 | } else { | |
8032 | dp_type = argv[1]; | |
31a19d69 | 8033 | } |
50aa28fd AW |
8034 | backer = shash_find_data(&all_dpif_backers, dp_type); |
8035 | if (!backer) { | |
8036 | unixctl_command_reply_error(conn, "Cannot find datapath " | |
8037 | "of this name"); | |
31a19d69 BP |
8038 | goto exit; |
8039 | } | |
8040 | } else { | |
50aa28fd AW |
8041 | /* No datapath name specified, so there should be only one |
8042 | * datapath. */ | |
8043 | struct shash_node *node; | |
8044 | if (shash_count(&all_dpif_backers) != 1) { | |
8045 | unixctl_command_reply_error(conn, "Must specify datapath " | |
8046 | "name, there is more than one type of datapath"); | |
31a19d69 BP |
8047 | goto exit; |
8048 | } | |
50aa28fd AW |
8049 | node = shash_first(&all_dpif_backers); |
8050 | backer = node->data; | |
876b0e1c | 8051 | } |
8b3b8dd1 | 8052 | |
50aa28fd AW |
8053 | /* Extract the ofproto_dpif object from the ofproto_receive() |
8054 | * function. */ | |
8055 | if (ofproto_receive(backer, NULL, odp_key.data, | |
8056 | odp_key.size, &flow, NULL, &ofproto, NULL, | |
8057 | &initial_vals)) { | |
8058 | unixctl_command_reply_error(conn, "Invalid datapath flow"); | |
8059 | goto exit; | |
8b3b8dd1 | 8060 | } |
50aa28fd AW |
8061 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
8062 | } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { | |
8063 | if (argc != 3) { | |
8064 | unixctl_command_reply_error(conn, "Must specify bridge name"); | |
876b0e1c BP |
8065 | goto exit; |
8066 | } | |
8067 | ||
50aa28fd AW |
8068 | ofproto = ofproto_dpif_lookup(argv[1]); |
8069 | if (!ofproto) { | |
8070 | unixctl_command_reply_error(conn, "Unknown bridge name"); | |
8071 | goto exit; | |
8072 | } | |
14f94f9a | 8073 | initial_vals.vlan_tci = flow.vlan_tci; |
876b0e1c | 8074 | } else { |
50aa28fd | 8075 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
abe529af BP |
8076 | goto exit; |
8077 | } | |
8078 | ||
50aa28fd AW |
8079 | /* Generate a packet, if requested. */ |
8080 | if (packet) { | |
8081 | if (!packet->size) { | |
8082 | flow_compose(packet, &flow); | |
8083 | } else { | |
8084 | ds_put_cstr(&result, "Packet: "); | |
8085 | s = ofp_packet_to_string(packet->data, packet->size); | |
8086 | ds_put_cstr(&result, s); | |
8087 | free(s); | |
8088 | ||
8089 | /* Use the metadata from the flow and the packet argument | |
8090 | * to reconstruct the flow. */ | |
8091 | flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, | |
8092 | flow.in_port, &flow); | |
8093 | initial_vals.vlan_tci = flow.vlan_tci; | |
8094 | } | |
8095 | } | |
8096 | ||
14f94f9a | 8097 | ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); |
6a6455e5 EJ |
8098 | unixctl_command_reply(conn, ds_cstr(&result)); |
8099 | ||
8100 | exit: | |
8101 | ds_destroy(&result); | |
8102 | ofpbuf_delete(packet); | |
8103 | ofpbuf_uninit(&odp_key); | |
8104 | } | |
8105 | ||
8106 | static void | |
8107 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
8108 | const struct ofpbuf *packet, |
8109 | const struct initial_vals *initial_vals, struct ds *ds) | |
6a6455e5 EJ |
8110 | { |
8111 | struct rule_dpif *rule; | |
8112 | ||
8113 | ds_put_cstr(ds, "Flow: "); | |
8114 | flow_format(ds, flow); | |
8115 | ds_put_char(ds, '\n'); | |
abe529af | 8116 | |
c57b2226 BP |
8117 | rule = rule_dpif_lookup(ofproto, flow); |
8118 | ||
6a6455e5 | 8119 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
8120 | if (rule == ofproto->miss_rule) { |
8121 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
8122 | } else if (rule == ofproto->no_packet_in_rule) { | |
8123 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
8124 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
8125 | } | |
8126 | ||
abe529af | 8127 | if (rule) { |
050ac423 BP |
8128 | uint64_t odp_actions_stub[1024 / 8]; |
8129 | struct ofpbuf odp_actions; | |
8130 | ||
6a6455e5 | 8131 | struct trace_ctx trace; |
0e553d9c | 8132 | uint8_t tcp_flags; |
abe529af | 8133 | |
6a6455e5 EJ |
8134 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
8135 | trace.result = ds; | |
8136 | trace.flow = *flow; | |
050ac423 BP |
8137 | ofpbuf_use_stub(&odp_actions, |
8138 | odp_actions_stub, sizeof odp_actions_stub); | |
bbafd73b EJ |
8139 | xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags, |
8140 | packet); | |
8141 | trace.xin.resubmit_hook = trace_resubmit; | |
8142 | trace.xin.report_hook = trace_report; | |
8143 | xlate_actions(&trace.xin, &trace.xout); | |
abe529af | 8144 | |
6a6455e5 EJ |
8145 | ds_put_char(ds, '\n'); |
8146 | trace_format_flow(ds, 0, "Final flow", &trace); | |
8147 | ds_put_cstr(ds, "Datapath actions: "); | |
bbafd73b EJ |
8148 | format_odp_actions(ds, trace.xout.odp_actions.data, |
8149 | trace.xout.odp_actions.size); | |
876b0e1c | 8150 | |
bbafd73b | 8151 | if (trace.xout.slow) { |
6a7e895f BP |
8152 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " |
8153 | "slow path because it:"); | |
bbafd73b | 8154 | switch (trace.xout.slow) { |
98f0520f EJ |
8155 | case SLOW_CFM: |
8156 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
8157 | break; | |
8158 | case SLOW_LACP: | |
8159 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
8160 | break; | |
8161 | case SLOW_STP: | |
8162 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
8163 | break; | |
8164 | case SLOW_BFD: | |
8165 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
8166 | break; | |
8167 | case SLOW_CONTROLLER: | |
8168 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
8169 | "to the OpenFlow controller."); | |
8170 | break; | |
8171 | case __SLOW_MAX: | |
8172 | NOT_REACHED(); | |
6a7e895f | 8173 | } |
876b0e1c | 8174 | } |
bbafd73b EJ |
8175 | |
8176 | xlate_out_uninit(&trace.xout); | |
abe529af | 8177 | } |
abe529af BP |
8178 | } |
8179 | ||
7ee20df1 | 8180 | static void |
0e15264f BP |
8181 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8182 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8183 | { |
8184 | clogged = true; | |
bde9f75d | 8185 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8186 | } |
8187 | ||
8188 | static void | |
0e15264f BP |
8189 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
8190 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
8191 | { |
8192 | clogged = false; | |
bde9f75d | 8193 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
8194 | } |
8195 | ||
6814e51f BP |
8196 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
8197 | * 'reply' describing the results. */ | |
8198 | static void | |
8199 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
8200 | { | |
8201 | struct facet *facet; | |
8202 | int errors; | |
8203 | ||
8204 | errors = 0; | |
8205 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
8206 | if (!facet_check_consistency(facet)) { | |
8207 | errors++; | |
8208 | } | |
8209 | } | |
8210 | if (errors) { | |
2cc3c58e | 8211 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
8212 | } |
8213 | ||
8214 | if (errors) { | |
8215 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
8216 | ofproto->up.name, errors); | |
8217 | } else { | |
8218 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
8219 | } | |
8220 | } | |
8221 | ||
8222 | static void | |
8223 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
8224 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
8225 | { | |
8226 | struct ds reply = DS_EMPTY_INITIALIZER; | |
8227 | struct ofproto_dpif *ofproto; | |
8228 | ||
8229 | if (argc > 1) { | |
8230 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8231 | if (!ofproto) { | |
bde9f75d EJ |
8232 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
8233 | "ofproto/list for help)"); | |
6814e51f BP |
8234 | return; |
8235 | } | |
8236 | ofproto_dpif_self_check__(ofproto, &reply); | |
8237 | } else { | |
8238 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8239 | ofproto_dpif_self_check__(ofproto, &reply); | |
8240 | } | |
8241 | } | |
8242 | ||
bde9f75d | 8243 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
8244 | ds_destroy(&reply); |
8245 | } | |
8246 | ||
27022416 JP |
8247 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
8248 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
8249 | * to destroy 'ofproto_shash' and free the returned value. */ | |
8250 | static const struct shash_node ** | |
8251 | get_ofprotos(struct shash *ofproto_shash) | |
8252 | { | |
8253 | const struct ofproto_dpif *ofproto; | |
8254 | ||
8255 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8256 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
8257 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
8258 | } | |
8259 | ||
8260 | return shash_sort(ofproto_shash); | |
8261 | } | |
8262 | ||
8263 | static void | |
8264 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
8265 | const char *argv[] OVS_UNUSED, | |
8266 | void *aux OVS_UNUSED) | |
8267 | { | |
8268 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8269 | struct shash ofproto_shash; | |
8270 | const struct shash_node **sorted_ofprotos; | |
8271 | int i; | |
8272 | ||
8273 | shash_init(&ofproto_shash); | |
8274 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8275 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8276 | const struct shash_node *node = sorted_ofprotos[i]; | |
8277 | ds_put_format(&ds, "%s\n", node->name); | |
8278 | } | |
8279 | ||
8280 | shash_destroy(&ofproto_shash); | |
8281 | free(sorted_ofprotos); | |
8282 | ||
8283 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8284 | ds_destroy(&ds); | |
8285 | } | |
8286 | ||
8287 | static void | |
8288 | show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) | |
8289 | { | |
27022416 JP |
8290 | const struct shash_node **ports; |
8291 | int i; | |
655ab909 AZ |
8292 | struct avg_subfacet_rates lifetime; |
8293 | unsigned long long int minutes; | |
8294 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8295 | ||
8296 | minutes = (time_msec() - ofproto->created) / min_ms; | |
8297 | ||
8298 | if (minutes > 0) { | |
8299 | lifetime.add_rate = (double)ofproto->total_subfacet_add_count | |
8300 | / minutes; | |
8301 | lifetime.del_rate = (double)ofproto->total_subfacet_del_count | |
8302 | / minutes; | |
8303 | }else { | |
8304 | lifetime.add_rate = 0.0; | |
8305 | lifetime.del_rate = 0.0; | |
8306 | } | |
27022416 | 8307 | |
acf60855 JP |
8308 | ds_put_format(ds, "%s (%s):\n", ofproto->up.name, |
8309 | dpif_name(ofproto->backer->dpif)); | |
27022416 | 8310 | ds_put_format(ds, |
735d7efb AZ |
8311 | "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n", |
8312 | ofproto->n_hit, ofproto->n_missed); | |
655ab909 AZ |
8313 | ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d," |
8314 | " life span: %llu(ms)\n", | |
8315 | hmap_count(&ofproto->subfacets), | |
8316 | avg_subfacet_count(ofproto), | |
8317 | ofproto->max_n_subfacet, | |
8318 | avg_subfacet_life_span(ofproto)); | |
8319 | if (minutes >= 60) { | |
8320 | show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly); | |
8321 | } | |
8322 | if (minutes >= 60 * 24) { | |
8323 | show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily); | |
8324 | } | |
8325 | show_dp_rates(ds, "\t\toverall avg:", &lifetime); | |
27022416 JP |
8326 | |
8327 | ports = shash_sort(&ofproto->up.port_by_name); | |
8328 | for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { | |
8329 | const struct shash_node *node = ports[i]; | |
8330 | struct ofport *ofport = node->data; | |
8331 | const char *name = netdev_get_name(ofport->netdev); | |
8332 | const char *type = netdev_get_type(ofport->netdev); | |
0a740f48 EJ |
8333 | uint32_t odp_port; |
8334 | ||
8335 | ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); | |
8336 | ||
8337 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); | |
8338 | if (odp_port != OVSP_NONE) { | |
8339 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
8340 | } else { | |
8341 | ds_put_cstr(ds, "none:"); | |
8342 | } | |
27022416 | 8343 | |
27022416 JP |
8344 | if (strcmp(type, "system")) { |
8345 | struct netdev *netdev; | |
8346 | int error; | |
8347 | ||
8348 | ds_put_format(ds, " (%s", type); | |
8349 | ||
8350 | error = netdev_open(name, type, &netdev); | |
8351 | if (!error) { | |
8352 | struct smap config; | |
8353 | ||
8354 | smap_init(&config); | |
8355 | error = netdev_get_config(netdev, &config); | |
8356 | if (!error) { | |
8357 | const struct smap_node **nodes; | |
8358 | size_t i; | |
8359 | ||
8360 | nodes = smap_sort(&config); | |
8361 | for (i = 0; i < smap_count(&config); i++) { | |
8362 | const struct smap_node *node = nodes[i]; | |
8363 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
8364 | node->key, node->value); | |
8365 | } | |
8366 | free(nodes); | |
8367 | } | |
8368 | smap_destroy(&config); | |
8369 | ||
8370 | netdev_close(netdev); | |
8371 | } | |
8372 | ds_put_char(ds, ')'); | |
8373 | } | |
8374 | ds_put_char(ds, '\n'); | |
8375 | } | |
8376 | free(ports); | |
8377 | } | |
8378 | ||
8379 | static void | |
8380 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, | |
8381 | const char *argv[], void *aux OVS_UNUSED) | |
8382 | { | |
8383 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8384 | const struct ofproto_dpif *ofproto; | |
8385 | ||
8386 | if (argc > 1) { | |
8387 | int i; | |
8388 | for (i = 1; i < argc; i++) { | |
8389 | ofproto = ofproto_dpif_lookup(argv[i]); | |
8390 | if (!ofproto) { | |
8391 | ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " | |
8392 | "for help)", argv[i]); | |
8393 | unixctl_command_reply_error(conn, ds_cstr(&ds)); | |
8394 | return; | |
8395 | } | |
8396 | show_dp_format(ofproto, &ds); | |
8397 | } | |
8398 | } else { | |
8399 | struct shash ofproto_shash; | |
8400 | const struct shash_node **sorted_ofprotos; | |
8401 | int i; | |
8402 | ||
8403 | shash_init(&ofproto_shash); | |
8404 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8405 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8406 | const struct shash_node *node = sorted_ofprotos[i]; | |
8407 | show_dp_format(node->data, &ds); | |
8408 | } | |
8409 | ||
8410 | shash_destroy(&ofproto_shash); | |
8411 | free(sorted_ofprotos); | |
8412 | } | |
8413 | ||
8414 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8415 | ds_destroy(&ds); | |
8416 | } | |
8417 | ||
8418 | static void | |
8419 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8420 | int argc OVS_UNUSED, const char *argv[], | |
8421 | void *aux OVS_UNUSED) | |
8422 | { | |
8423 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8424 | const struct ofproto_dpif *ofproto; | |
8425 | struct subfacet *subfacet; | |
8426 | ||
8427 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8428 | if (!ofproto) { | |
8429 | unixctl_command_reply_error(conn, "no such bridge"); | |
8430 | return; | |
8431 | } | |
8432 | ||
af37354d EJ |
8433 | update_stats(ofproto->backer); |
8434 | ||
27022416 | 8435 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
4dff9097 EJ |
8436 | struct facet *facet = subfacet->facet; |
8437 | ||
9566abf9 | 8438 | odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); |
27022416 JP |
8439 | |
8440 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8441 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8442 | if (subfacet->used) { | |
8443 | ds_put_format(&ds, "%.3fs", | |
8444 | (time_msec() - subfacet->used) / 1000.0); | |
8445 | } else { | |
8446 | ds_put_format(&ds, "never"); | |
8447 | } | |
8448 | if (subfacet->facet->tcp_flags) { | |
8449 | ds_put_cstr(&ds, ", flags:"); | |
8450 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8451 | } | |
8452 | ||
8453 | ds_put_cstr(&ds, ", actions:"); | |
bbafd73b | 8454 | if (facet->xout.slow) { |
f2245da3 JP |
8455 | uint64_t slow_path_stub[128 / 8]; |
8456 | const struct nlattr *actions; | |
8457 | size_t actions_len; | |
8458 | ||
bbafd73b | 8459 | compose_slow_path(ofproto, &facet->flow, facet->xout.slow, |
f2245da3 JP |
8460 | slow_path_stub, sizeof slow_path_stub, |
8461 | &actions, &actions_len); | |
8462 | format_odp_actions(&ds, actions, actions_len); | |
8463 | } else { | |
bbafd73b EJ |
8464 | format_odp_actions(&ds, facet->xout.odp_actions.data, |
8465 | facet->xout.odp_actions.size); | |
f2245da3 | 8466 | } |
27022416 JP |
8467 | ds_put_char(&ds, '\n'); |
8468 | } | |
8469 | ||
8470 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8471 | ds_destroy(&ds); | |
8472 | } | |
8473 | ||
8474 | static void | |
8475 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8476 | int argc OVS_UNUSED, const char *argv[], | |
8477 | void *aux OVS_UNUSED) | |
8478 | { | |
8479 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8480 | struct ofproto_dpif *ofproto; | |
8481 | ||
8482 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8483 | if (!ofproto) { | |
8484 | unixctl_command_reply_error(conn, "no such bridge"); | |
8485 | return; | |
8486 | } | |
8487 | ||
8488 | flush(&ofproto->up); | |
8489 | ||
8490 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8491 | ds_destroy(&ds); | |
8492 | } | |
8493 | ||
abe529af BP |
8494 | static void |
8495 | ofproto_dpif_unixctl_init(void) | |
8496 | { | |
8497 | static bool registered; | |
8498 | if (registered) { | |
8499 | return; | |
8500 | } | |
8501 | registered = true; | |
8502 | ||
0e15264f BP |
8503 | unixctl_command_register( |
8504 | "ofproto/trace", | |
50aa28fd AW |
8505 | "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", |
8506 | 1, 3, ofproto_unixctl_trace, NULL); | |
96e466a3 | 8507 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
8508 | ofproto_unixctl_fdb_flush, NULL); |
8509 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8510 | ofproto_unixctl_fdb_show, NULL); | |
8511 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8512 | ofproto_dpif_clog, NULL); | |
8513 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8514 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
8515 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
8516 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
8517 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
8518 | ofproto_unixctl_dpif_dump_dps, NULL); | |
8519 | unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, | |
8520 | ofproto_unixctl_dpif_show, NULL); | |
8521 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, | |
8522 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8523 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8524 | ofproto_unixctl_dpif_del_flows, NULL); | |
abe529af BP |
8525 | } |
8526 | \f | |
52a90c29 BP |
8527 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
8528 | * | |
8529 | * This is deprecated. It is only for compatibility with broken device drivers | |
8530 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8531 | * devices are not used. When broken device drivers are no longer in | |
8532 | * widespread use, we will delete these interfaces. */ | |
8533 | ||
8534 | static int | |
8535 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8536 | { | |
8537 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8538 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8539 | ||
8540 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8541 | && vid == ofport->vlandev_vid) { | |
8542 | return 0; | |
8543 | } | |
8544 | ||
2cc3c58e | 8545 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
8546 | |
8547 | if (ofport->realdev_ofp_port) { | |
8548 | vsp_remove(ofport); | |
8549 | } | |
8550 | if (realdev_ofp_port && ofport->bundle) { | |
8551 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8552 | * themselves be part of a bundle. */ | |
8553 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8554 | } | |
8555 | ||
8556 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8557 | ofport->vlandev_vid = vid; | |
8558 | ||
8559 | if (realdev_ofp_port) { | |
8560 | vsp_add(ofport, realdev_ofp_port, vid); | |
8561 | } | |
8562 | ||
8563 | return 0; | |
8564 | } | |
8565 | ||
8566 | static uint32_t | |
8567 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8568 | { | |
8569 | return hash_2words(realdev_ofp_port, vid); | |
8570 | } | |
8571 | ||
40e05935 BP |
8572 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
8573 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
8574 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
8575 | * it would return the port number of eth0.9. | |
8576 | * | |
8577 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
8578 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
8579 | static uint32_t |
8580 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
8581 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
8582 | { | |
8583 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
e1b1d06a | 8584 | uint16_t realdev_ofp_port; |
52a90c29 BP |
8585 | int vid = vlan_tci_to_vid(vlan_tci); |
8586 | const struct vlan_splinter *vsp; | |
8587 | ||
e1b1d06a | 8588 | realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); |
52a90c29 BP |
8589 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, |
8590 | hash_realdev_vid(realdev_ofp_port, vid), | |
8591 | &ofproto->realdev_vid_map) { | |
8592 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8593 | && vsp->vid == vid) { | |
e1b1d06a | 8594 | return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); |
52a90c29 BP |
8595 | } |
8596 | } | |
8597 | } | |
8598 | return realdev_odp_port; | |
8599 | } | |
8600 | ||
8601 | static struct vlan_splinter * | |
8602 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8603 | { | |
8604 | struct vlan_splinter *vsp; | |
8605 | ||
8606 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8607 | &ofproto->vlandev_map) { | |
8608 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8609 | return vsp; | |
8610 | } | |
8611 | } | |
8612 | ||
8613 | return NULL; | |
8614 | } | |
8615 | ||
40e05935 BP |
8616 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
8617 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8618 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8619 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8620 | * eth0 and store 9 in '*vid'. | |
8621 | * | |
8622 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8623 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8624 | * always does.*/ | |
52a90c29 BP |
8625 | static uint16_t |
8626 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 8627 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
8628 | { |
8629 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8630 | const struct vlan_splinter *vsp; | |
8631 | ||
8632 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8633 | if (vsp) { | |
8634 | if (vid) { | |
8635 | *vid = vsp->vid; | |
8636 | } | |
8637 | return vsp->realdev_ofp_port; | |
8638 | } | |
8639 | } | |
8640 | return 0; | |
8641 | } | |
8642 | ||
b98d8985 BP |
8643 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
8644 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8645 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8646 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8647 | * always the case unless VLAN splinters are enabled), returns false without | |
8648 | * making any changes. */ | |
8649 | static bool | |
8650 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8651 | { | |
8652 | uint16_t realdev; | |
8653 | int vid; | |
8654 | ||
8655 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8656 | if (!realdev) { | |
8657 | return false; | |
8658 | } | |
8659 | ||
8660 | /* Cause the flow to be processed as if it came in on the real device with | |
8661 | * the VLAN device's VLAN ID. */ | |
8662 | flow->in_port = realdev; | |
8663 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8664 | return true; | |
8665 | } | |
8666 | ||
52a90c29 BP |
8667 | static void |
8668 | vsp_remove(struct ofport_dpif *port) | |
8669 | { | |
8670 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8671 | struct vlan_splinter *vsp; | |
8672 | ||
8673 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8674 | if (vsp) { | |
8675 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8676 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8677 | free(vsp); | |
8678 | ||
8679 | port->realdev_ofp_port = 0; | |
8680 | } else { | |
8681 | VLOG_ERR("missing vlan device record"); | |
8682 | } | |
8683 | } | |
8684 | ||
8685 | static void | |
8686 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8687 | { | |
8688 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8689 | ||
8690 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8691 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8692 | == realdev_ofp_port)) { | |
8693 | struct vlan_splinter *vsp; | |
8694 | ||
8695 | vsp = xmalloc(sizeof *vsp); | |
8696 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8697 | hash_int(port->up.ofp_port, 0)); | |
8698 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8699 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8700 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8701 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8702 | vsp->vid = vid; | |
8703 | ||
8704 | port->realdev_ofp_port = realdev_ofp_port; | |
8705 | } else { | |
8706 | VLOG_ERR("duplicate vlan device record"); | |
8707 | } | |
8708 | } | |
e1b1d06a JP |
8709 | |
8710 | static uint32_t | |
8711 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8712 | { | |
8713 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8714 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8715 | } | |
8716 | ||
acf60855 JP |
8717 | static struct ofport_dpif * |
8718 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
e1b1d06a JP |
8719 | { |
8720 | struct ofport_dpif *port; | |
8721 | ||
8722 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8723 | hash_int(odp_port, 0), | |
acf60855 | 8724 | &backer->odp_to_ofport_map) { |
e1b1d06a | 8725 | if (port->odp_port == odp_port) { |
acf60855 | 8726 | return port; |
e1b1d06a JP |
8727 | } |
8728 | } | |
8729 | ||
acf60855 JP |
8730 | return NULL; |
8731 | } | |
8732 | ||
8733 | static uint16_t | |
8734 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8735 | { | |
8736 | struct ofport_dpif *port; | |
8737 | ||
8738 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 8739 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
8740 | return port->up.ofp_port; |
8741 | } else { | |
8742 | return OFPP_NONE; | |
8743 | } | |
e1b1d06a | 8744 | } |
655ab909 AZ |
8745 | static unsigned long long int |
8746 | avg_subfacet_life_span(const struct ofproto_dpif *ofproto) | |
8747 | { | |
8748 | unsigned long long int dc; | |
8749 | unsigned long long int avg; | |
8750 | ||
8751 | dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count; | |
8752 | avg = dc ? ofproto->total_subfacet_life_span / dc : 0; | |
8753 | ||
8754 | return avg; | |
8755 | } | |
8756 | ||
8757 | static double | |
8758 | avg_subfacet_count(const struct ofproto_dpif *ofproto) | |
8759 | { | |
8760 | double avg_c = 0.0; | |
8761 | ||
8762 | if (ofproto->n_update_stats) { | |
8763 | avg_c = (double)ofproto->total_subfacet_count | |
8764 | / ofproto->n_update_stats; | |
8765 | } | |
8766 | ||
8767 | return avg_c; | |
8768 | } | |
8769 | ||
8770 | static void | |
8771 | show_dp_rates(struct ds *ds, const char *heading, | |
8772 | const struct avg_subfacet_rates *rates) | |
8773 | { | |
8774 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
8775 | heading, rates->add_rate, rates->del_rate); | |
8776 | } | |
8777 | ||
8778 | static void | |
8779 | update_max_subfacet_count(struct ofproto_dpif *ofproto) | |
8780 | { | |
8781 | ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet, | |
8782 | hmap_count(&ofproto->subfacets)); | |
8783 | } | |
8784 | ||
8785 | /* Compute exponentially weighted moving average, adding 'new' as the newest, | |
8786 | * most heavily weighted element. 'base' designates the rate of decay: after | |
8787 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
8788 | * (about .37). */ | |
8789 | static void | |
8790 | exp_mavg(double *avg, int base, double new) | |
8791 | { | |
8792 | *avg = (*avg * (base - 1) + new) / base; | |
8793 | } | |
8794 | ||
8795 | static void | |
8796 | update_moving_averages(struct ofproto_dpif *ofproto) | |
8797 | { | |
8798 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8799 | ||
8800 | /* Update hourly averages on the minute boundaries. */ | |
8801 | if (time_msec() - ofproto->last_minute >= min_ms) { | |
8802 | exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count); | |
8803 | exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count); | |
8804 | ||
8805 | /* Update daily averages on the hour boundaries. */ | |
8806 | if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) { | |
8807 | exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate); | |
8808 | exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate); | |
8809 | } | |
8810 | ||
8811 | ofproto->total_subfacet_add_count += ofproto->subfacet_add_count; | |
8812 | ofproto->total_subfacet_del_count += ofproto->subfacet_del_count; | |
8813 | ofproto->subfacet_add_count = 0; | |
8814 | ofproto->subfacet_del_count = 0; | |
8815 | ofproto->last_minute += min_ms; | |
8816 | } | |
8817 | } | |
e1b1d06a | 8818 | |
735d7efb AZ |
8819 | static void |
8820 | dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta) | |
8821 | { | |
8822 | ofproto->n_hit += delta; | |
8823 | } | |
8824 | ||
abe529af | 8825 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 8826 | init, |
abe529af BP |
8827 | enumerate_types, |
8828 | enumerate_names, | |
8829 | del, | |
0aeaabc8 | 8830 | port_open_type, |
acf60855 JP |
8831 | type_run, |
8832 | type_run_fast, | |
8833 | type_wait, | |
abe529af BP |
8834 | alloc, |
8835 | construct, | |
8836 | destruct, | |
8837 | dealloc, | |
8838 | run, | |
5fcc0d00 | 8839 | run_fast, |
abe529af | 8840 | wait, |
0d085684 | 8841 | get_memory_usage, |
abe529af | 8842 | flush, |
6c1491fb BP |
8843 | get_features, |
8844 | get_tables, | |
abe529af BP |
8845 | port_alloc, |
8846 | port_construct, | |
8847 | port_destruct, | |
8848 | port_dealloc, | |
8849 | port_modified, | |
8850 | port_reconfigured, | |
8851 | port_query_by_name, | |
8852 | port_add, | |
8853 | port_del, | |
6527c598 | 8854 | port_get_stats, |
abe529af BP |
8855 | port_dump_start, |
8856 | port_dump_next, | |
8857 | port_dump_done, | |
8858 | port_poll, | |
8859 | port_poll_wait, | |
8860 | port_is_lacp_current, | |
0ab6decf | 8861 | NULL, /* rule_choose_table */ |
abe529af BP |
8862 | rule_alloc, |
8863 | rule_construct, | |
8864 | rule_destruct, | |
8865 | rule_dealloc, | |
abe529af BP |
8866 | rule_get_stats, |
8867 | rule_execute, | |
8868 | rule_modify_actions, | |
7257b535 | 8869 | set_frag_handling, |
abe529af BP |
8870 | packet_out, |
8871 | set_netflow, | |
8872 | get_netflow_ids, | |
8873 | set_sflow, | |
29089a54 | 8874 | set_ipfix, |
abe529af | 8875 | set_cfm, |
9a9e3786 | 8876 | get_cfm_status, |
ccc09689 EJ |
8877 | set_bfd, |
8878 | get_bfd_status, | |
21f7563c JP |
8879 | set_stp, |
8880 | get_stp_status, | |
8881 | set_stp_port, | |
8882 | get_stp_port_status, | |
8b36f51e | 8883 | set_queues, |
abe529af BP |
8884 | bundle_set, |
8885 | bundle_remove, | |
8886 | mirror_set, | |
9d24de3b | 8887 | mirror_get_stats, |
abe529af BP |
8888 | set_flood_vlans, |
8889 | is_mirror_output_bundle, | |
8402c74b | 8890 | forward_bpdu_changed, |
c4069512 | 8891 | set_mac_table_config, |
52a90c29 | 8892 | set_realdev, |
abe529af | 8893 | }; |