]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e0edde6f | 2 | * Copyright (c) 2009, 2010, 2011, 2012 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 | ||
23 | #include "autopath.h" | |
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 BP |
38 | #include "multipath.h" |
39 | #include "netdev.h" | |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
f25d0cf3 | 45 | #include "ofp-actions.h" |
31a19d69 | 46 | #include "ofp-parse.h" |
abe529af | 47 | #include "ofp-print.h" |
9d6ac44e | 48 | #include "ofproto-dpif-governor.h" |
bae473fe | 49 | #include "ofproto-dpif-sflow.h" |
abe529af | 50 | #include "poll-loop.h" |
0d085684 | 51 | #include "simap.h" |
abe529af | 52 | #include "timer.h" |
6c1491fb | 53 | #include "unaligned.h" |
abe529af BP |
54 | #include "unixctl.h" |
55 | #include "vlan-bitmap.h" | |
56 | #include "vlog.h" | |
57 | ||
58 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
59 | ||
abe529af | 60 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
61 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
62 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
63 | COVERAGE_DEFINE(facet_revalidate); |
64 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 65 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 66 | |
29901626 | 67 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 68 | * flow translation. */ |
1642690c | 69 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 70 | |
9cdaaebe BP |
71 | /* Number of implemented OpenFlow tables. */ |
72 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
73 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
74 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 75 | |
abe529af BP |
76 | struct ofport_dpif; |
77 | struct ofproto_dpif; | |
78 | ||
79 | struct rule_dpif { | |
80 | struct rule up; | |
81 | ||
abe529af BP |
82 | /* These statistics: |
83 | * | |
84 | * - Do include packets and bytes from facets that have been deleted or | |
85 | * whose own statistics have been folded into the rule. | |
86 | * | |
87 | * - Do include packets and bytes sent "by hand" that were accounted to | |
88 | * the rule without any facet being involved (this is a rare corner | |
89 | * case in rule_execute()). | |
90 | * | |
91 | * - Do not include packet or bytes that can be obtained from any facet's | |
92 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 93 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
94 | */ |
95 | uint64_t packet_count; /* Number of packets received. */ | |
96 | uint64_t byte_count; /* Number of bytes received. */ | |
97 | ||
54a9cbc9 BP |
98 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
99 | ||
abe529af BP |
100 | struct list facets; /* List of "struct facet"s. */ |
101 | }; | |
102 | ||
103 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
104 | { | |
105 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
106 | } | |
107 | ||
29901626 | 108 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
109 | const struct flow *); |
110 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
111 | const struct flow *, | |
112 | uint8_t table); | |
c376f9a3 IY |
113 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
114 | const struct flow *flow); | |
abe529af | 115 | |
112bc5f4 BP |
116 | static void rule_credit_stats(struct rule_dpif *, |
117 | const struct dpif_flow_stats *); | |
18b2a258 | 118 | static void flow_push_stats(struct rule_dpif *, const struct flow *, |
112bc5f4 | 119 | const struct dpif_flow_stats *); |
822d9414 | 120 | static tag_type rule_calculate_tag(const struct flow *, |
b0f7b9b5 BP |
121 | const struct flow_wildcards *, |
122 | uint32_t basis); | |
123 | static void rule_invalidate(const struct rule_dpif *); | |
124 | ||
abe529af BP |
125 | #define MAX_MIRRORS 32 |
126 | typedef uint32_t mirror_mask_t; | |
127 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
128 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
129 | struct ofmirror { | |
130 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
131 | size_t idx; /* In ofproto's "mirrors" array. */ | |
132 | void *aux; /* Key supplied by ofproto's client. */ | |
133 | char *name; /* Identifier for log messages. */ | |
134 | ||
135 | /* Selection criteria. */ | |
136 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
137 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
138 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
139 | ||
9ba15e2a | 140 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
141 | struct ofbundle *out; /* Output port or NULL. */ |
142 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 143 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
144 | |
145 | /* Counters. */ | |
146 | int64_t packet_count; /* Number of packets sent. */ | |
147 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
148 | }; |
149 | ||
150 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
151 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
152 | mirror_mask_t mirrors, | |
153 | uint64_t packets, uint64_t bytes); | |
abe529af | 154 | |
abe529af | 155 | struct ofbundle { |
abe529af | 156 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 157 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
158 | void *aux; /* Key supplied by ofproto's client. */ |
159 | char *name; /* Identifier for log messages. */ | |
160 | ||
161 | /* Configuration. */ | |
162 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 163 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
164 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
165 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
166 | * NULL if all VLANs are trunked. */ | |
167 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
168 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 169 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
170 | |
171 | /* Status. */ | |
9e1fd49b | 172 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
173 | |
174 | /* Port mirroring info. */ | |
175 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
176 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
177 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
178 | }; | |
179 | ||
180 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 181 | static void bundle_update(struct ofbundle *); |
abe529af BP |
182 | static void bundle_destroy(struct ofbundle *); |
183 | static void bundle_del_port(struct ofport_dpif *); | |
184 | static void bundle_run(struct ofbundle *); | |
185 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 186 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
187 | uint16_t in_port, bool warn, |
188 | struct ofport_dpif **in_ofportp); | |
abe529af | 189 | |
33158a18 JP |
190 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
191 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
192 | * when an input bundle is needed for validation (e.g., mirroring or | |
193 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
194 | * any 'port' structs, so care must be taken when dealing with it. */ | |
195 | static struct ofbundle ofpp_none_bundle = { | |
196 | .name = "OFPP_NONE", | |
197 | .vlan_mode = PORT_VLAN_TRUNK | |
198 | }; | |
199 | ||
21f7563c JP |
200 | static void stp_run(struct ofproto_dpif *ofproto); |
201 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
202 | static int set_stp_port(struct ofport *, |
203 | const struct ofproto_port_stp_settings *); | |
21f7563c | 204 | |
5da5ec37 BP |
205 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
206 | ||
abe529af BP |
207 | struct action_xlate_ctx { |
208 | /* action_xlate_ctx_init() initializes these members. */ | |
209 | ||
210 | /* The ofproto. */ | |
211 | struct ofproto_dpif *ofproto; | |
212 | ||
213 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
214 | * this flow when actions change header fields. */ | |
215 | struct flow flow; | |
216 | ||
217 | /* The packet corresponding to 'flow', or a null pointer if we are | |
218 | * revalidating without a packet to refer to. */ | |
219 | const struct ofpbuf *packet; | |
220 | ||
3de9590b BP |
221 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
222 | * actions update the flow table? | |
223 | * | |
224 | * We want to update these tables if we are actually processing a packet, | |
225 | * or if we are accounting for packets that the datapath has processed, but | |
226 | * not if we are just revalidating. */ | |
227 | bool may_learn; | |
75a75043 | 228 | |
18b2a258 BP |
229 | /* The rule that we are currently translating, or NULL. */ |
230 | struct rule_dpif *rule; | |
54834960 | 231 | |
0e553d9c BP |
232 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
233 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
234 | * timeouts.) */ | |
235 | uint8_t tcp_flags; | |
236 | ||
112bc5f4 BP |
237 | /* If nonnull, flow translation calls this function just before executing a |
238 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
239 | * when the recursion depth is exceeded. | |
240 | * | |
241 | * 'rule' is the rule being submitted into. It will be null if the | |
242 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
243 | * | |
244 | * This is normally null so the client has to set it manually after | |
245 | * calling action_xlate_ctx_init(). */ | |
246 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
247 | ||
479df176 BP |
248 | /* If nonnull, flow translation calls this function to report some |
249 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
250 | * dropped a packet. */ | |
251 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
252 | ||
112bc5f4 BP |
253 | /* If nonnull, flow translation credits the specified statistics to each |
254 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
255 | * |
256 | * This is normally null so the client has to set it manually after | |
257 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 258 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 259 | |
abe529af BP |
260 | /* xlate_actions() initializes and uses these members. The client might want |
261 | * to look at them after it returns. */ | |
262 | ||
263 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 264 | tag_type tags; /* Tags associated with actions. */ |
6a7e895f | 265 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
75a75043 BP |
266 | bool has_learn; /* Actions include NXAST_LEARN? */ |
267 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 268 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 269 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 270 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
271 | |
272 | /* xlate_actions() initializes and uses these members, but the client has no | |
273 | * reason to look at them. */ | |
274 | ||
275 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 276 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 277 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 278 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 279 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
280 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
281 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
282 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 283 | bool exit; /* No further actions should be processed. */ |
ccb7c863 | 284 | struct flow orig_flow; /* Copy of original flow. */ |
abe529af BP |
285 | }; |
286 | ||
287 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
288 | struct ofproto_dpif *, const struct flow *, | |
18b2a258 | 289 | ovs_be16 initial_tci, struct rule_dpif *, |
0e553d9c | 290 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 | 291 | static void xlate_actions(struct action_xlate_ctx *, |
f25d0cf3 | 292 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 BP |
293 | struct ofpbuf *odp_actions); |
294 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
f25d0cf3 BP |
295 | const struct ofpact *ofpacts, |
296 | size_t ofpacts_len); | |
abe529af | 297 | |
6a7e895f BP |
298 | static size_t put_userspace_action(const struct ofproto_dpif *, |
299 | struct ofpbuf *odp_actions, | |
300 | const struct flow *, | |
301 | const union user_action_cookie *); | |
302 | ||
303 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
304 | enum slow_path_reason, | |
305 | uint64_t *stub, size_t stub_size, | |
306 | const struct nlattr **actionsp, | |
307 | size_t *actions_lenp); | |
308 | ||
479df176 BP |
309 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); |
310 | ||
6a7e895f BP |
311 | /* A subfacet (see "struct subfacet" below) has three possible installation |
312 | * states: | |
313 | * | |
314 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
315 | * case just after the subfacet is created, just before the subfacet is | |
316 | * destroyed, or if the datapath returns an error when we try to install a | |
317 | * subfacet. | |
318 | * | |
319 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
320 | * | |
321 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
322 | * ofproto_dpif is installed in the datapath. | |
323 | */ | |
324 | enum subfacet_path { | |
325 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
326 | SF_FAST_PATH, /* Full actions are installed. */ | |
327 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
328 | }; | |
329 | ||
330 | static const char *subfacet_path_to_string(enum subfacet_path); | |
331 | ||
5f5fbd17 BP |
332 | /* A dpif flow and actions associated with a facet. |
333 | * | |
334 | * See also the large comment on struct facet. */ | |
335 | struct subfacet { | |
336 | /* Owners. */ | |
337 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
338 | struct list list_node; /* In struct facet's 'facets' list. */ | |
339 | struct facet *facet; /* Owning facet. */ | |
340 | ||
341 | /* Key. | |
342 | * | |
343 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
344 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
345 | * regenerate the ODP flow key from ->facet->flow. */ | |
346 | enum odp_key_fitness key_fitness; | |
347 | struct nlattr *key; | |
348 | int key_len; | |
349 | ||
350 | long long int used; /* Time last used; time created if not used. */ | |
351 | ||
352 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
353 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
354 | ||
355 | /* Datapath actions. | |
356 | * | |
357 | * These should be essentially identical for every subfacet in a facet, but | |
358 | * may differ in trivial ways due to VLAN splinters. */ | |
359 | size_t actions_len; /* Number of bytes in actions[]. */ | |
360 | struct nlattr *actions; /* Datapath actions. */ | |
361 | ||
6a7e895f BP |
362 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
363 | enum subfacet_path path; /* Installed in datapath? */ | |
5f5fbd17 BP |
364 | |
365 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
366 | * splinters can cause it to differ. This value should be removed when | |
367 | * the VLAN splinters feature is no longer needed. */ | |
368 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
369 | }; | |
370 | ||
371 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, | |
372 | const struct nlattr *key, | |
459b16a1 BP |
373 | size_t key_len, ovs_be16 initial_tci, |
374 | long long int now); | |
5f5fbd17 BP |
375 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
376 | const struct nlattr *key, size_t key_len); | |
377 | static void subfacet_destroy(struct subfacet *); | |
378 | static void subfacet_destroy__(struct subfacet *); | |
379 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, | |
380 | struct ofpbuf *key); | |
381 | static void subfacet_reset_dp_stats(struct subfacet *, | |
382 | struct dpif_flow_stats *); | |
383 | static void subfacet_update_time(struct subfacet *, long long int used); | |
384 | static void subfacet_update_stats(struct subfacet *, | |
385 | const struct dpif_flow_stats *); | |
386 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
387 | const struct ofpbuf *packet, |
388 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
389 | static int subfacet_install(struct subfacet *, |
390 | const struct nlattr *actions, size_t actions_len, | |
6a7e895f | 391 | struct dpif_flow_stats *, enum slow_path_reason); |
5f5fbd17 BP |
392 | static void subfacet_uninstall(struct subfacet *); |
393 | ||
6a7e895f BP |
394 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); |
395 | ||
b0f7b9b5 BP |
396 | /* An exact-match instantiation of an OpenFlow flow. |
397 | * | |
398 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
399 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
400 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
401 | * the facet. When the kernel module (or other dpif implementation) and Open | |
402 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
403 | * one subfacet per facet. If the dpif implementation supports more-specific | |
404 | * flow matching than userspace, however, a facet can have more than one | |
405 | * subfacet, each of which corresponds to some distinction in flow that | |
406 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
407 | * |
408 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
409 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 410 | struct facet { |
b0f7b9b5 BP |
411 | /* Owners. */ |
412 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
413 | struct list list_node; /* In owning rule's 'facets' list. */ | |
414 | struct rule_dpif *rule; /* Owning rule. */ | |
415 | ||
416 | /* Owned data. */ | |
417 | struct list subfacets; | |
abe529af BP |
418 | long long int used; /* Time last used; time created if not used. */ |
419 | ||
b0f7b9b5 BP |
420 | /* Key. */ |
421 | struct flow flow; | |
422 | ||
abe529af BP |
423 | /* These statistics: |
424 | * | |
425 | * - Do include packets and bytes sent "by hand", e.g. with | |
426 | * dpif_execute(). | |
427 | * | |
428 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 429 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 430 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
431 | * |
432 | * - Do not include packets or bytes that can be obtained from the | |
433 | * datapath for any existing subfacet. | |
abe529af BP |
434 | */ |
435 | uint64_t packet_count; /* Number of packets received. */ | |
436 | uint64_t byte_count; /* Number of bytes received. */ | |
437 | ||
b0f7b9b5 | 438 | /* Resubmit statistics. */ |
9d24de3b JP |
439 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
440 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
441 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 442 | |
b0f7b9b5 | 443 | /* Accounting. */ |
907a4c5e | 444 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 445 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 446 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 447 | |
b95fc6ba BP |
448 | /* Properties of datapath actions. |
449 | * | |
450 | * Every subfacet has its own actions because actions can differ slightly | |
451 | * between splintered and non-splintered subfacets due to the VLAN tag | |
452 | * being initially different (present vs. absent). All of them have these | |
453 | * properties in common so we just store one copy of them here. */ | |
75a75043 BP |
454 | bool has_learn; /* Actions include NXAST_LEARN? */ |
455 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 456 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 457 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 458 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
459 | |
460 | /* Storage for a single subfacet, to reduce malloc() time and space | |
461 | * overhead. (A facet always has at least one subfacet and in the common | |
462 | * case has exactly one subfacet.) */ | |
463 | struct subfacet one_subfacet; | |
abe529af BP |
464 | }; |
465 | ||
2b459b83 BP |
466 | static struct facet *facet_create(struct rule_dpif *, |
467 | const struct flow *, uint32_t hash); | |
15baa734 | 468 | static void facet_remove(struct facet *); |
abe529af BP |
469 | static void facet_free(struct facet *); |
470 | ||
2b459b83 BP |
471 | static struct facet *facet_find(struct ofproto_dpif *, |
472 | const struct flow *, uint32_t hash); | |
abe529af | 473 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 474 | const struct flow *, uint32_t hash); |
c57b2226 | 475 | static void facet_revalidate(struct facet *); |
6814e51f | 476 | static bool facet_check_consistency(struct facet *); |
abe529af | 477 | |
15baa734 | 478 | static void facet_flush_stats(struct facet *); |
abe529af | 479 | |
15baa734 | 480 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 481 | static void facet_reset_counters(struct facet *); |
abe529af | 482 | static void facet_push_stats(struct facet *); |
3de9590b BP |
483 | static void facet_learn(struct facet *); |
484 | static void facet_account(struct facet *); | |
abe529af BP |
485 | |
486 | static bool facet_is_controller_flow(struct facet *); | |
487 | ||
abe529af BP |
488 | struct ofport_dpif { |
489 | struct ofport up; | |
490 | ||
491 | uint32_t odp_port; | |
492 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
493 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
494 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
495 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 496 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 497 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 498 | long long int carrier_seq; /* Carrier status changes. */ |
21f7563c | 499 | |
52a90c29 | 500 | /* Spanning tree. */ |
21f7563c JP |
501 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
502 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
503 | long long int stp_state_entered; | |
8b36f51e EJ |
504 | |
505 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
506 | |
507 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
508 | * | |
509 | * This is deprecated. It is only for compatibility with broken device | |
510 | * drivers in old versions of Linux that do not properly support VLANs when | |
511 | * VLAN devices are not used. When broken device drivers are no longer in | |
512 | * widespread use, we will delete these interfaces. */ | |
513 | uint16_t realdev_ofp_port; | |
514 | int vlandev_vid; | |
8b36f51e EJ |
515 | }; |
516 | ||
517 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
518 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
519 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
520 | struct priority_to_dscp { | |
521 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
522 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
523 | ||
524 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
525 | }; |
526 | ||
52a90c29 BP |
527 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
528 | * | |
529 | * This is deprecated. It is only for compatibility with broken device drivers | |
530 | * in old versions of Linux that do not properly support VLANs when VLAN | |
531 | * devices are not used. When broken device drivers are no longer in | |
532 | * widespread use, we will delete these interfaces. */ | |
533 | struct vlan_splinter { | |
534 | struct hmap_node realdev_vid_node; | |
535 | struct hmap_node vlandev_node; | |
536 | uint16_t realdev_ofp_port; | |
537 | uint16_t vlandev_ofp_port; | |
538 | int vid; | |
539 | }; | |
540 | ||
541 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
542 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 543 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
544 | static void vsp_remove(struct ofport_dpif *); |
545 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
546 | ||
abe529af BP |
547 | static struct ofport_dpif * |
548 | ofport_dpif_cast(const struct ofport *ofport) | |
549 | { | |
550 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
551 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
552 | } | |
553 | ||
554 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 555 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 556 | static void port_wait(struct ofport_dpif *); |
a5610457 | 557 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 558 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 559 | |
7ee20df1 BP |
560 | struct dpif_completion { |
561 | struct list list_node; | |
562 | struct ofoperation *op; | |
563 | }; | |
564 | ||
54a9cbc9 BP |
565 | /* Extra information about a classifier table. |
566 | * Currently used just for optimized flow revalidation. */ | |
567 | struct table_dpif { | |
568 | /* If either of these is nonnull, then this table has a form that allows | |
569 | * flows to be tagged to avoid revalidating most flows for the most common | |
570 | * kinds of flow table changes. */ | |
571 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
572 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
573 | uint32_t basis; /* Keeps each table's tags separate. */ | |
574 | }; | |
575 | ||
3c4a309c BP |
576 | /* Reasons that we might need to revalidate every facet, and corresponding |
577 | * coverage counters. | |
578 | * | |
579 | * A value of 0 means that there is no need to revalidate. | |
580 | * | |
581 | * It would be nice to have some cleaner way to integrate with coverage | |
582 | * counters, but with only a few reasons I guess this is good enough for | |
583 | * now. */ | |
584 | enum revalidate_reason { | |
585 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
586 | REV_STP, /* Spanning tree protocol port status change. */ | |
587 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
588 | REV_FLOW_TABLE, /* Flow table changed. */ | |
589 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
590 | }; | |
591 | COVERAGE_DEFINE(rev_reconfigure); | |
592 | COVERAGE_DEFINE(rev_stp); | |
593 | COVERAGE_DEFINE(rev_port_toggled); | |
594 | COVERAGE_DEFINE(rev_flow_table); | |
595 | COVERAGE_DEFINE(rev_inconsistency); | |
596 | ||
abe529af | 597 | struct ofproto_dpif { |
b44a10b7 | 598 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
599 | struct ofproto up; |
600 | struct dpif *dpif; | |
abe529af | 601 | |
c57b2226 BP |
602 | /* Special OpenFlow rules. */ |
603 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
604 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
605 | ||
6c1491fb BP |
606 | /* Statistics. */ |
607 | uint64_t n_matches; | |
608 | ||
abe529af BP |
609 | /* Bridging. */ |
610 | struct netflow *netflow; | |
bae473fe | 611 | struct dpif_sflow *sflow; |
abe529af BP |
612 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
613 | struct mac_learning *ml; | |
614 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 615 | bool has_mirrors; |
abe529af BP |
616 | bool has_bonded_bundles; |
617 | ||
618 | /* Expiration. */ | |
619 | struct timer next_expiration; | |
620 | ||
621 | /* Facets. */ | |
622 | struct hmap facets; | |
b0f7b9b5 | 623 | struct hmap subfacets; |
9d6ac44e | 624 | struct governor *governor; |
54a9cbc9 BP |
625 | |
626 | /* Revalidation. */ | |
627 | struct table_dpif tables[N_TABLES]; | |
3c4a309c | 628 | enum revalidate_reason need_revalidate; |
abe529af | 629 | struct tag_set revalidate_set; |
7ee20df1 BP |
630 | |
631 | /* Support for debugging async flow mods. */ | |
632 | struct list completions; | |
daff3353 EJ |
633 | |
634 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
635 | struct netdev_stats stats; /* To account packets generated and consumed in |
636 | * userspace. */ | |
21f7563c JP |
637 | |
638 | /* Spanning tree. */ | |
639 | struct stp *stp; | |
640 | long long int stp_last_tick; | |
52a90c29 BP |
641 | |
642 | /* VLAN splinters. */ | |
643 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
644 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
645 | }; |
646 | ||
7ee20df1 BP |
647 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
648 | * for debugging the asynchronous flow_mod implementation.) */ | |
649 | static bool clogged; | |
650 | ||
b44a10b7 BP |
651 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
652 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
653 | ||
abe529af BP |
654 | static void ofproto_dpif_unixctl_init(void); |
655 | ||
656 | static struct ofproto_dpif * | |
657 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
658 | { | |
659 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
660 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
661 | } | |
662 | ||
4acbc98d | 663 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 664 | uint16_t ofp_port); |
4acbc98d | 665 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 666 | uint32_t odp_port); |
6a6455e5 EJ |
667 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
668 | const struct ofpbuf *, ovs_be16 initial_tci, | |
669 | struct ds *); | |
abe529af BP |
670 | |
671 | /* Packet processing. */ | |
672 | static void update_learning_table(struct ofproto_dpif *, | |
673 | const struct flow *, int vlan, | |
674 | struct ofbundle *); | |
501f8d1f BP |
675 | /* Upcalls. */ |
676 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 677 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
678 | |
679 | /* Flow expiration. */ | |
680 | static int expire(struct ofproto_dpif *); | |
681 | ||
6fca1ffb BP |
682 | /* NetFlow. */ |
683 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
684 | ||
abe529af | 685 | /* Utilities. */ |
52a90c29 | 686 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
687 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
688 | struct ofpbuf *odp_actions, | |
689 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
690 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
691 | const struct flow *flow); | |
abe529af BP |
692 | /* Global variables. */ |
693 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
694 | \f | |
695 | /* Factory functions. */ | |
696 | ||
697 | static void | |
698 | enumerate_types(struct sset *types) | |
699 | { | |
700 | dp_enumerate_types(types); | |
701 | } | |
702 | ||
703 | static int | |
704 | enumerate_names(const char *type, struct sset *names) | |
705 | { | |
706 | return dp_enumerate_names(type, names); | |
707 | } | |
708 | ||
709 | static int | |
710 | del(const char *type, const char *name) | |
711 | { | |
712 | struct dpif *dpif; | |
713 | int error; | |
714 | ||
715 | error = dpif_open(name, type, &dpif); | |
716 | if (!error) { | |
717 | error = dpif_delete(dpif); | |
718 | dpif_close(dpif); | |
719 | } | |
720 | return error; | |
721 | } | |
722 | \f | |
723 | /* Basic life-cycle. */ | |
724 | ||
c57b2226 BP |
725 | static int add_internal_flows(struct ofproto_dpif *); |
726 | ||
abe529af BP |
727 | static struct ofproto * |
728 | alloc(void) | |
729 | { | |
730 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
731 | return &ofproto->up; | |
732 | } | |
733 | ||
734 | static void | |
735 | dealloc(struct ofproto *ofproto_) | |
736 | { | |
737 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
738 | free(ofproto); | |
739 | } | |
740 | ||
741 | static int | |
0f5f95a9 | 742 | construct(struct ofproto *ofproto_) |
abe529af BP |
743 | { |
744 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
745 | const char *name = ofproto->up.name; | |
91858960 | 746 | int max_ports; |
abe529af BP |
747 | int error; |
748 | int i; | |
749 | ||
750 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
751 | if (error) { | |
752 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
753 | return error; | |
754 | } | |
755 | ||
91858960 BP |
756 | max_ports = dpif_get_max_ports(ofproto->dpif); |
757 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); | |
758 | ||
6c1491fb | 759 | ofproto->n_matches = 0; |
abe529af | 760 | |
be8194bb JG |
761 | dpif_flow_flush(ofproto->dpif); |
762 | dpif_recv_purge(ofproto->dpif); | |
763 | ||
a12b3ead | 764 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
765 | if (error) { |
766 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
767 | dpif_close(ofproto->dpif); | |
768 | return error; | |
769 | } | |
abe529af BP |
770 | |
771 | ofproto->netflow = NULL; | |
772 | ofproto->sflow = NULL; | |
21f7563c | 773 | ofproto->stp = NULL; |
abe529af | 774 | hmap_init(&ofproto->bundles); |
e764773c | 775 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
776 | for (i = 0; i < MAX_MIRRORS; i++) { |
777 | ofproto->mirrors[i] = NULL; | |
778 | } | |
779 | ofproto->has_bonded_bundles = false; | |
780 | ||
781 | timer_set_duration(&ofproto->next_expiration, 1000); | |
782 | ||
783 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 784 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 785 | ofproto->governor = NULL; |
54a9cbc9 BP |
786 | |
787 | for (i = 0; i < N_TABLES; i++) { | |
788 | struct table_dpif *table = &ofproto->tables[i]; | |
789 | ||
790 | table->catchall_table = NULL; | |
791 | table->other_table = NULL; | |
792 | table->basis = random_uint32(); | |
793 | } | |
3c4a309c | 794 | ofproto->need_revalidate = 0; |
abe529af BP |
795 | tag_set_init(&ofproto->revalidate_set); |
796 | ||
7ee20df1 BP |
797 | list_init(&ofproto->completions); |
798 | ||
abe529af BP |
799 | ofproto_dpif_unixctl_init(); |
800 | ||
ccb7c863 | 801 | ofproto->has_mirrors = false; |
daff3353 EJ |
802 | ofproto->has_bundle_action = false; |
803 | ||
52a90c29 BP |
804 | hmap_init(&ofproto->vlandev_map); |
805 | hmap_init(&ofproto->realdev_vid_map); | |
806 | ||
b44a10b7 BP |
807 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
808 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 809 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
810 | |
811 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
812 | error = add_internal_flows(ofproto); |
813 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
814 | ||
815 | return error; | |
816 | } | |
817 | ||
818 | static int | |
819 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 820 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
821 | { |
822 | struct ofputil_flow_mod fm; | |
823 | int error; | |
824 | ||
825 | cls_rule_init_catchall(&fm.cr, 0); | |
826 | cls_rule_set_reg(&fm.cr, 0, id); | |
623e1caf | 827 | fm.new_cookie = htonll(0); |
c57b2226 BP |
828 | fm.cookie = htonll(0); |
829 | fm.cookie_mask = htonll(0); | |
830 | fm.table_id = TBL_INTERNAL; | |
831 | fm.command = OFPFC_ADD; | |
832 | fm.idle_timeout = 0; | |
833 | fm.hard_timeout = 0; | |
834 | fm.buffer_id = 0; | |
835 | fm.out_port = 0; | |
836 | fm.flags = 0; | |
f25d0cf3 BP |
837 | fm.ofpacts = ofpacts->data; |
838 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
839 | |
840 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
841 | if (error) { | |
842 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
843 | id, ofperr_to_string(error)); | |
844 | return error; | |
845 | } | |
846 | ||
847 | *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL); | |
848 | assert(*rulep != NULL); | |
0f5f95a9 | 849 | |
abe529af BP |
850 | return 0; |
851 | } | |
852 | ||
c57b2226 BP |
853 | static int |
854 | add_internal_flows(struct ofproto_dpif *ofproto) | |
855 | { | |
f25d0cf3 BP |
856 | struct ofpact_controller *controller; |
857 | uint64_t ofpacts_stub[128 / 8]; | |
858 | struct ofpbuf ofpacts; | |
c57b2226 BP |
859 | int error; |
860 | int id; | |
861 | ||
f25d0cf3 | 862 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
863 | id = 1; |
864 | ||
f25d0cf3 BP |
865 | controller = ofpact_put_CONTROLLER(&ofpacts); |
866 | controller->max_len = UINT16_MAX; | |
867 | controller->controller_id = 0; | |
868 | controller->reason = OFPR_NO_MATCH; | |
869 | ofpact_pad(&ofpacts); | |
870 | ||
871 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
872 | if (error) { |
873 | return error; | |
874 | } | |
875 | ||
f25d0cf3 BP |
876 | ofpbuf_clear(&ofpacts); |
877 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
878 | &ofproto->no_packet_in_rule); |
879 | return error; | |
880 | } | |
881 | ||
7ee20df1 BP |
882 | static void |
883 | complete_operations(struct ofproto_dpif *ofproto) | |
884 | { | |
885 | struct dpif_completion *c, *next; | |
886 | ||
887 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
888 | ofoperation_complete(c->op, 0); | |
889 | list_remove(&c->list_node); | |
890 | free(c); | |
891 | } | |
892 | } | |
893 | ||
abe529af BP |
894 | static void |
895 | destruct(struct ofproto *ofproto_) | |
896 | { | |
897 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 898 | struct rule_dpif *rule, *next_rule; |
d0918789 | 899 | struct oftable *table; |
abe529af BP |
900 | int i; |
901 | ||
b44a10b7 | 902 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
903 | complete_operations(ofproto); |
904 | ||
0697b5c3 BP |
905 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
906 | struct cls_cursor cursor; | |
907 | ||
d0918789 | 908 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
909 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
910 | ofproto_rule_destroy(&rule->up); | |
911 | } | |
7ee20df1 BP |
912 | } |
913 | ||
abe529af BP |
914 | for (i = 0; i < MAX_MIRRORS; i++) { |
915 | mirror_destroy(ofproto->mirrors[i]); | |
916 | } | |
917 | ||
918 | netflow_destroy(ofproto->netflow); | |
bae473fe | 919 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
920 | hmap_destroy(&ofproto->bundles); |
921 | mac_learning_destroy(ofproto->ml); | |
922 | ||
923 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 924 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 925 | governor_destroy(ofproto->governor); |
abe529af | 926 | |
52a90c29 BP |
927 | hmap_destroy(&ofproto->vlandev_map); |
928 | hmap_destroy(&ofproto->realdev_vid_map); | |
929 | ||
abe529af BP |
930 | dpif_close(ofproto->dpif); |
931 | } | |
932 | ||
933 | static int | |
5fcc0d00 | 934 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
935 | { |
936 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 937 | struct ofport_dpif *ofport; |
9b16c439 | 938 | unsigned int work; |
abe529af | 939 | |
0aa66d6e EJ |
940 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
941 | port_run_fast(ofport); | |
942 | } | |
943 | ||
9b16c439 BP |
944 | /* Handle one or more batches of upcalls, until there's nothing left to do |
945 | * or until we do a fixed total amount of work. | |
946 | * | |
947 | * We do work in batches because it can be much cheaper to set up a number | |
948 | * of flows and fire off their patches all at once. We do multiple batches | |
949 | * because in some cases handling a packet can cause another packet to be | |
950 | * queued almost immediately as part of the return flow. Both | |
951 | * optimizations can make major improvements on some benchmarks and | |
952 | * presumably for real traffic as well. */ | |
953 | work = 0; | |
954 | while (work < FLOW_MISS_MAX_BATCH) { | |
955 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 956 | if (retval <= 0) { |
9b16c439 | 957 | return -retval; |
501f8d1f | 958 | } |
5fcc0d00 BP |
959 | work += retval; |
960 | } | |
961 | return 0; | |
962 | } | |
963 | ||
964 | static int | |
965 | run(struct ofproto *ofproto_) | |
966 | { | |
967 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
968 | struct ofport_dpif *ofport; | |
969 | struct ofbundle *bundle; | |
970 | int error; | |
971 | ||
972 | if (!clogged) { | |
973 | complete_operations(ofproto); | |
974 | } | |
975 | dpif_run(ofproto->dpif); | |
976 | ||
977 | error = run_fast(ofproto_); | |
978 | if (error) { | |
979 | return error; | |
abe529af BP |
980 | } |
981 | ||
982 | if (timer_expired(&ofproto->next_expiration)) { | |
983 | int delay = expire(ofproto); | |
984 | timer_set_duration(&ofproto->next_expiration, delay); | |
985 | } | |
986 | ||
987 | if (ofproto->netflow) { | |
6fca1ffb BP |
988 | if (netflow_run(ofproto->netflow)) { |
989 | send_netflow_active_timeouts(ofproto); | |
990 | } | |
abe529af BP |
991 | } |
992 | if (ofproto->sflow) { | |
bae473fe | 993 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
994 | } |
995 | ||
996 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
997 | port_run(ofport); | |
998 | } | |
999 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1000 | bundle_run(bundle); | |
1001 | } | |
1002 | ||
21f7563c | 1003 | stp_run(ofproto); |
1c313b88 BP |
1004 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
1005 | ||
abe529af BP |
1006 | /* Now revalidate if there's anything to do. */ |
1007 | if (ofproto->need_revalidate | |
1008 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
1009 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
1010 | bool revalidate_all = ofproto->need_revalidate; | |
c57b2226 | 1011 | struct facet *facet; |
abe529af | 1012 | |
3c4a309c BP |
1013 | switch (ofproto->need_revalidate) { |
1014 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
1015 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
1016 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
1017 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
1018 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
1019 | } | |
1020 | ||
abe529af BP |
1021 | /* Clear the revalidation flags. */ |
1022 | tag_set_init(&ofproto->revalidate_set); | |
3c4a309c | 1023 | ofproto->need_revalidate = 0; |
abe529af | 1024 | |
c57b2226 | 1025 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
abe529af BP |
1026 | if (revalidate_all |
1027 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 1028 | facet_revalidate(facet); |
abe529af BP |
1029 | } |
1030 | } | |
1031 | } | |
1032 | ||
6814e51f BP |
1033 | /* Check the consistency of a random facet, to aid debugging. */ |
1034 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
1035 | struct facet *facet; | |
1036 | ||
1037 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
1038 | struct facet, hmap_node); | |
1039 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
1040 | if (!facet_check_consistency(facet)) { | |
3c4a309c | 1041 | ofproto->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1042 | } |
1043 | } | |
1044 | } | |
1045 | ||
9d6ac44e BP |
1046 | if (ofproto->governor) { |
1047 | size_t n_subfacets; | |
1048 | ||
1049 | governor_run(ofproto->governor); | |
1050 | ||
1051 | /* If the governor has shrunk to its minimum size and the number of | |
1052 | * subfacets has dwindled, then drop the governor entirely. | |
1053 | * | |
1054 | * For hysteresis, the number of subfacets to drop the governor is | |
1055 | * smaller than the number needed to trigger its creation. */ | |
1056 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1057 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1058 | && governor_is_idle(ofproto->governor)) { | |
1059 | governor_destroy(ofproto->governor); | |
1060 | ofproto->governor = NULL; | |
1061 | } | |
1062 | } | |
1063 | ||
abe529af BP |
1064 | return 0; |
1065 | } | |
1066 | ||
1067 | static void | |
1068 | wait(struct ofproto *ofproto_) | |
1069 | { | |
1070 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1071 | struct ofport_dpif *ofport; | |
1072 | struct ofbundle *bundle; | |
1073 | ||
7ee20df1 BP |
1074 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1075 | poll_immediate_wake(); | |
1076 | } | |
1077 | ||
abe529af BP |
1078 | dpif_wait(ofproto->dpif); |
1079 | dpif_recv_wait(ofproto->dpif); | |
1080 | if (ofproto->sflow) { | |
bae473fe | 1081 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
1082 | } |
1083 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
1084 | poll_immediate_wake(); | |
1085 | } | |
1086 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1087 | port_wait(ofport); | |
1088 | } | |
1089 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1090 | bundle_wait(bundle); | |
1091 | } | |
6fca1ffb BP |
1092 | if (ofproto->netflow) { |
1093 | netflow_wait(ofproto->netflow); | |
1094 | } | |
1c313b88 | 1095 | mac_learning_wait(ofproto->ml); |
21f7563c | 1096 | stp_wait(ofproto); |
abe529af BP |
1097 | if (ofproto->need_revalidate) { |
1098 | /* Shouldn't happen, but if it does just go around again. */ | |
1099 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1100 | poll_immediate_wake(); | |
1101 | } else { | |
1102 | timer_wait(&ofproto->next_expiration); | |
1103 | } | |
9d6ac44e BP |
1104 | if (ofproto->governor) { |
1105 | governor_wait(ofproto->governor); | |
1106 | } | |
abe529af BP |
1107 | } |
1108 | ||
0d085684 BP |
1109 | static void |
1110 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1111 | { | |
1112 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1113 | ||
1114 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1115 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1116 | } | |
1117 | ||
abe529af BP |
1118 | static void |
1119 | flush(struct ofproto *ofproto_) | |
1120 | { | |
1121 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1122 | struct facet *facet, *next_facet; | |
1123 | ||
1124 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1125 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1126 | * bother trying to uninstall it. There is no point in uninstalling it | |
1127 | * individually since we are about to blow away all the facets with | |
1128 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
1129 | struct subfacet *subfacet; |
1130 | ||
1131 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
6a7e895f | 1132 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 BP |
1133 | subfacet->dp_packet_count = 0; |
1134 | subfacet->dp_byte_count = 0; | |
1135 | } | |
15baa734 | 1136 | facet_remove(facet); |
abe529af BP |
1137 | } |
1138 | dpif_flow_flush(ofproto->dpif); | |
1139 | } | |
1140 | ||
6c1491fb BP |
1141 | static void |
1142 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1143 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1144 | { |
1145 | *arp_match_ip = true; | |
9e1fd49b BP |
1146 | *actions = (OFPUTIL_A_OUTPUT | |
1147 | OFPUTIL_A_SET_VLAN_VID | | |
1148 | OFPUTIL_A_SET_VLAN_PCP | | |
1149 | OFPUTIL_A_STRIP_VLAN | | |
1150 | OFPUTIL_A_SET_DL_SRC | | |
1151 | OFPUTIL_A_SET_DL_DST | | |
1152 | OFPUTIL_A_SET_NW_SRC | | |
1153 | OFPUTIL_A_SET_NW_DST | | |
1154 | OFPUTIL_A_SET_NW_TOS | | |
1155 | OFPUTIL_A_SET_TP_SRC | | |
1156 | OFPUTIL_A_SET_TP_DST | | |
1157 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1158 | } |
1159 | ||
1160 | static void | |
e2b9ac44 | 1161 | get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots) |
6c1491fb BP |
1162 | { |
1163 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1164 | struct dpif_dp_stats s; |
6c1491fb BP |
1165 | |
1166 | strcpy(ots->name, "classifier"); | |
1167 | ||
1168 | dpif_get_dp_stats(ofproto->dpif, &s); | |
1169 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
1170 | put_32aligned_be64(&ots->matched_count, | |
1171 | htonll(s.n_hit + ofproto->n_matches)); | |
1172 | } | |
1173 | ||
abe529af BP |
1174 | static struct ofport * |
1175 | port_alloc(void) | |
1176 | { | |
1177 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1178 | return &port->up; | |
1179 | } | |
1180 | ||
1181 | static void | |
1182 | port_dealloc(struct ofport *port_) | |
1183 | { | |
1184 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1185 | free(port); | |
1186 | } | |
1187 | ||
1188 | static int | |
1189 | port_construct(struct ofport *port_) | |
1190 | { | |
1191 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1192 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1193 | ||
3c4a309c | 1194 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1195 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
1196 | port->bundle = NULL; | |
1197 | port->cfm = NULL; | |
1198 | port->tag = tag_create_random(); | |
d5ffa7f2 | 1199 | port->may_enable = true; |
21f7563c JP |
1200 | port->stp_port = NULL; |
1201 | port->stp_state = STP_DISABLED; | |
8b36f51e | 1202 | hmap_init(&port->priorities); |
52a90c29 BP |
1203 | port->realdev_ofp_port = 0; |
1204 | port->vlandev_vid = 0; | |
3e5b3fdb | 1205 | port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); |
abe529af BP |
1206 | |
1207 | if (ofproto->sflow) { | |
392c7182 | 1208 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
1209 | } |
1210 | ||
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | static void | |
1215 | port_destruct(struct ofport *port_) | |
1216 | { | |
1217 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1218 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1219 | ||
3c4a309c | 1220 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af | 1221 | bundle_remove(port_); |
a5610457 | 1222 | set_cfm(port_, NULL); |
abe529af | 1223 | if (ofproto->sflow) { |
bae473fe | 1224 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1225 | } |
8b36f51e EJ |
1226 | |
1227 | ofport_clear_priorities(port); | |
1228 | hmap_destroy(&port->priorities); | |
abe529af BP |
1229 | } |
1230 | ||
1231 | static void | |
1232 | port_modified(struct ofport *port_) | |
1233 | { | |
1234 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1235 | ||
1236 | if (port->bundle && port->bundle->bond) { | |
1237 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | static void | |
9e1fd49b | 1242 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1243 | { |
1244 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1245 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1246 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1247 | |
9e1fd49b | 1248 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1249 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1250 | OFPUTIL_PC_NO_PACKET_IN)) { | |
3c4a309c | 1251 | ofproto->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1252 | |
9e1fd49b | 1253 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1254 | bundle_update(port->bundle); |
1255 | } | |
abe529af BP |
1256 | } |
1257 | } | |
1258 | ||
1259 | static int | |
1260 | set_sflow(struct ofproto *ofproto_, | |
1261 | const struct ofproto_sflow_options *sflow_options) | |
1262 | { | |
1263 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1264 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1265 | |
abe529af | 1266 | if (sflow_options) { |
bae473fe | 1267 | if (!ds) { |
abe529af BP |
1268 | struct ofport_dpif *ofport; |
1269 | ||
bae473fe | 1270 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1271 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1272 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1273 | } |
3c4a309c | 1274 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af | 1275 | } |
bae473fe | 1276 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1277 | } else { |
6ff686f2 PS |
1278 | if (ds) { |
1279 | dpif_sflow_destroy(ds); | |
3c4a309c | 1280 | ofproto->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1281 | ofproto->sflow = NULL; |
1282 | } | |
abe529af BP |
1283 | } |
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | static int | |
a5610457 | 1288 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1289 | { |
1290 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1291 | int error; | |
1292 | ||
a5610457 | 1293 | if (!s) { |
abe529af BP |
1294 | error = 0; |
1295 | } else { | |
1296 | if (!ofport->cfm) { | |
8c977421 EJ |
1297 | struct ofproto_dpif *ofproto; |
1298 | ||
1299 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3c4a309c | 1300 | ofproto->need_revalidate = REV_RECONFIGURE; |
6f629657 | 1301 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1302 | } |
1303 | ||
a5610457 | 1304 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1305 | return 0; |
1306 | } | |
1307 | ||
1308 | error = EINVAL; | |
1309 | } | |
1310 | cfm_destroy(ofport->cfm); | |
1311 | ofport->cfm = NULL; | |
1312 | return error; | |
1313 | } | |
1314 | ||
1315 | static int | |
a5610457 | 1316 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1317 | { |
1318 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1319 | |
1320 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1321 | } |
1de11730 | 1322 | |
1c0333b6 EJ |
1323 | static int |
1324 | get_cfm_opup(const struct ofport *ofport_) | |
1325 | { | |
1326 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1327 | ||
1328 | return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; | |
1329 | } | |
1330 | ||
1de11730 EJ |
1331 | static int |
1332 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1333 | size_t *n_rmps) | |
1334 | { | |
1335 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1336 | ||
1337 | if (ofport->cfm) { | |
1338 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1339 | return 0; | |
1340 | } else { | |
1341 | return -1; | |
1342 | } | |
1343 | } | |
3967a833 MM |
1344 | |
1345 | static int | |
1346 | get_cfm_health(const struct ofport *ofport_) | |
1347 | { | |
1348 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1349 | ||
1350 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1351 | } | |
abe529af | 1352 | \f |
21f7563c JP |
1353 | /* Spanning Tree. */ |
1354 | ||
1355 | static void | |
1356 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1357 | { | |
1358 | struct ofproto_dpif *ofproto = ofproto_; | |
1359 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1360 | struct ofport_dpif *ofport; | |
1361 | ||
1362 | ofport = stp_port_get_aux(sp); | |
1363 | if (!ofport) { | |
1364 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1365 | ofproto->up.name, port_num); | |
1366 | } else { | |
1367 | struct eth_header *eth = pkt->l2; | |
1368 | ||
1369 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1370 | if (eth_addr_is_zero(eth->eth_src)) { | |
1371 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1372 | "with unknown MAC", ofproto->up.name, port_num); | |
1373 | } else { | |
97d6520b | 1374 | send_packet(ofport, pkt); |
21f7563c JP |
1375 | } |
1376 | } | |
1377 | ofpbuf_delete(pkt); | |
1378 | } | |
1379 | ||
1380 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1381 | static int | |
1382 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1383 | { | |
1384 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1385 | ||
1386 | /* Only revalidate flows if the configuration changed. */ | |
1387 | if (!s != !ofproto->stp) { | |
3c4a309c | 1388 | ofproto->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
1389 | } |
1390 | ||
1391 | if (s) { | |
1392 | if (!ofproto->stp) { | |
1393 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1394 | send_bpdu_cb, ofproto); | |
1395 | ofproto->stp_last_tick = time_msec(); | |
1396 | } | |
1397 | ||
1398 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1399 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1400 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1401 | stp_set_max_age(ofproto->stp, s->max_age); | |
1402 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1403 | } else { | |
851bf71d EJ |
1404 | struct ofport *ofport; |
1405 | ||
1406 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1407 | set_stp_port(ofport, NULL); | |
1408 | } | |
1409 | ||
21f7563c JP |
1410 | stp_destroy(ofproto->stp); |
1411 | ofproto->stp = NULL; | |
1412 | } | |
1413 | ||
1414 | return 0; | |
1415 | } | |
1416 | ||
1417 | static int | |
1418 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1419 | { | |
1420 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1421 | ||
1422 | if (ofproto->stp) { | |
1423 | s->enabled = true; | |
1424 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1425 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1426 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1427 | } else { | |
1428 | s->enabled = false; | |
1429 | } | |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
1434 | static void | |
1435 | update_stp_port_state(struct ofport_dpif *ofport) | |
1436 | { | |
1437 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1438 | enum stp_state state; | |
1439 | ||
1440 | /* Figure out new state. */ | |
1441 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1442 | : STP_DISABLED; | |
1443 | ||
1444 | /* Update state. */ | |
1445 | if (ofport->stp_state != state) { | |
9e1fd49b | 1446 | enum ofputil_port_state of_state; |
21f7563c JP |
1447 | bool fwd_change; |
1448 | ||
1449 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1450 | netdev_get_name(ofport->up.netdev), | |
1451 | stp_state_name(ofport->stp_state), | |
1452 | stp_state_name(state)); | |
1453 | if (stp_learn_in_state(ofport->stp_state) | |
1454 | != stp_learn_in_state(state)) { | |
1455 | /* xxx Learning action flows should also be flushed. */ | |
d0040604 | 1456 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
21f7563c JP |
1457 | } |
1458 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1459 | != stp_forward_in_state(state); | |
1460 | ||
3c4a309c | 1461 | ofproto->need_revalidate = REV_STP; |
21f7563c JP |
1462 | ofport->stp_state = state; |
1463 | ofport->stp_state_entered = time_msec(); | |
1464 | ||
b308140a | 1465 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1466 | bundle_update(ofport->bundle); |
1467 | } | |
1468 | ||
1469 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1470 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1471 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1472 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1473 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1474 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1475 | : 0); | |
21f7563c JP |
1476 | ofproto_port_set_state(&ofport->up, of_state); |
1477 | } | |
1478 | } | |
1479 | ||
1480 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1481 | * caller is responsible for assigning STP port numbers and ensuring | |
1482 | * there are no duplicates. */ | |
1483 | static int | |
1484 | set_stp_port(struct ofport *ofport_, | |
1485 | const struct ofproto_port_stp_settings *s) | |
1486 | { | |
1487 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1488 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1489 | struct stp_port *sp = ofport->stp_port; | |
1490 | ||
1491 | if (!s || !s->enable) { | |
1492 | if (sp) { | |
1493 | ofport->stp_port = NULL; | |
1494 | stp_port_disable(sp); | |
ecd12731 | 1495 | update_stp_port_state(ofport); |
21f7563c JP |
1496 | } |
1497 | return 0; | |
1498 | } else if (sp && stp_port_no(sp) != s->port_num | |
1499 | && ofport == stp_port_get_aux(sp)) { | |
1500 | /* The port-id changed, so disable the old one if it's not | |
1501 | * already in use by another port. */ | |
1502 | stp_port_disable(sp); | |
1503 | } | |
1504 | ||
1505 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1506 | stp_port_enable(sp); | |
1507 | ||
1508 | stp_port_set_aux(sp, ofport); | |
1509 | stp_port_set_priority(sp, s->priority); | |
1510 | stp_port_set_path_cost(sp, s->path_cost); | |
1511 | ||
1512 | update_stp_port_state(ofport); | |
1513 | ||
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | static int | |
1518 | get_stp_port_status(struct ofport *ofport_, | |
1519 | struct ofproto_port_stp_status *s) | |
1520 | { | |
1521 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1522 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1523 | struct stp_port *sp = ofport->stp_port; | |
1524 | ||
1525 | if (!ofproto->stp || !sp) { | |
1526 | s->enabled = false; | |
1527 | return 0; | |
1528 | } | |
1529 | ||
1530 | s->enabled = true; | |
1531 | s->port_id = stp_port_get_id(sp); | |
1532 | s->state = stp_port_get_state(sp); | |
1533 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1534 | s->role = stp_port_get_role(sp); | |
80740385 | 1535 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1536 | |
1537 | return 0; | |
1538 | } | |
1539 | ||
1540 | static void | |
1541 | stp_run(struct ofproto_dpif *ofproto) | |
1542 | { | |
1543 | if (ofproto->stp) { | |
1544 | long long int now = time_msec(); | |
1545 | long long int elapsed = now - ofproto->stp_last_tick; | |
1546 | struct stp_port *sp; | |
1547 | ||
1548 | if (elapsed > 0) { | |
1549 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1550 | ofproto->stp_last_tick = now; | |
1551 | } | |
1552 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1553 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1554 | ||
1555 | if (ofport) { | |
1556 | update_stp_port_state(ofport); | |
1557 | } | |
1558 | } | |
6ae50723 EJ |
1559 | |
1560 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1561 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1562 | } | |
21f7563c JP |
1563 | } |
1564 | } | |
1565 | ||
1566 | static void | |
1567 | stp_wait(struct ofproto_dpif *ofproto) | |
1568 | { | |
1569 | if (ofproto->stp) { | |
1570 | poll_timer_wait(1000); | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | /* Returns true if STP should process 'flow'. */ | |
1575 | static bool | |
1576 | stp_should_process_flow(const struct flow *flow) | |
1577 | { | |
1578 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1579 | } | |
1580 | ||
1581 | static void | |
1582 | stp_process_packet(const struct ofport_dpif *ofport, | |
1583 | const struct ofpbuf *packet) | |
1584 | { | |
1585 | struct ofpbuf payload = *packet; | |
1586 | struct eth_header *eth = payload.data; | |
1587 | struct stp_port *sp = ofport->stp_port; | |
1588 | ||
1589 | /* Sink packets on ports that have STP disabled when the bridge has | |
1590 | * STP enabled. */ | |
1591 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1592 | return; | |
1593 | } | |
1594 | ||
1595 | /* Trim off padding on payload. */ | |
c573540b BP |
1596 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1597 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1598 | } |
1599 | ||
1600 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1601 | stp_received_bpdu(sp, payload.data, payload.size); | |
1602 | } | |
1603 | } | |
1604 | \f | |
8b36f51e EJ |
1605 | static struct priority_to_dscp * |
1606 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1607 | { | |
1608 | struct priority_to_dscp *pdscp; | |
1609 | uint32_t hash; | |
1610 | ||
1611 | hash = hash_int(priority, 0); | |
1612 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1613 | if (pdscp->priority == priority) { | |
1614 | return pdscp; | |
1615 | } | |
1616 | } | |
1617 | return NULL; | |
1618 | } | |
1619 | ||
1620 | static void | |
1621 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1622 | { | |
1623 | struct priority_to_dscp *pdscp, *next; | |
1624 | ||
1625 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1626 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1627 | free(pdscp); | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | static int | |
1632 | set_queues(struct ofport *ofport_, | |
1633 | const struct ofproto_port_queue *qdscp_list, | |
1634 | size_t n_qdscp) | |
1635 | { | |
1636 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1637 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1638 | struct hmap new = HMAP_INITIALIZER(&new); | |
1639 | size_t i; | |
1640 | ||
1641 | for (i = 0; i < n_qdscp; i++) { | |
1642 | struct priority_to_dscp *pdscp; | |
1643 | uint32_t priority; | |
1644 | uint8_t dscp; | |
1645 | ||
1646 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1647 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1648 | &priority)) { | |
1649 | continue; | |
1650 | } | |
1651 | ||
1652 | pdscp = get_priority(ofport, priority); | |
1653 | if (pdscp) { | |
1654 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1655 | } else { | |
1656 | pdscp = xmalloc(sizeof *pdscp); | |
1657 | pdscp->priority = priority; | |
1658 | pdscp->dscp = dscp; | |
3c4a309c | 1659 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1660 | } |
1661 | ||
1662 | if (pdscp->dscp != dscp) { | |
1663 | pdscp->dscp = dscp; | |
3c4a309c | 1664 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1665 | } |
1666 | ||
1667 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1668 | } | |
1669 | ||
1670 | if (!hmap_is_empty(&ofport->priorities)) { | |
1671 | ofport_clear_priorities(ofport); | |
3c4a309c | 1672 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1673 | } |
1674 | ||
1675 | hmap_swap(&new, &ofport->priorities); | |
1676 | hmap_destroy(&new); | |
1677 | ||
1678 | return 0; | |
1679 | } | |
1680 | \f | |
abe529af BP |
1681 | /* Bundles. */ |
1682 | ||
b44a10b7 BP |
1683 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1684 | * ofproto to revalidate every flow. | |
1685 | * | |
1686 | * Normally MAC learning entries are removed only from the ofproto associated | |
1687 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1688 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1689 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1690 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1691 | * with the host from which it migrated. */ | |
abe529af | 1692 | static void |
b44a10b7 | 1693 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1694 | { |
1695 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1696 | struct mac_learning *ml = ofproto->ml; | |
1697 | struct mac_entry *mac, *next_mac; | |
1698 | ||
3c4a309c | 1699 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1700 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
1701 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1702 | if (all_ofprotos) { |
1703 | struct ofproto_dpif *o; | |
1704 | ||
1705 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1706 | if (o != ofproto) { | |
1707 | struct mac_entry *e; | |
1708 | ||
1709 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1710 | NULL); | |
1711 | if (e) { | |
1712 | tag_set_add(&o->revalidate_set, e->tag); | |
1713 | mac_learning_expire(o->ml, e); | |
1714 | } | |
1715 | } | |
1716 | } | |
1717 | } | |
1718 | ||
abe529af BP |
1719 | mac_learning_expire(ml, mac); |
1720 | } | |
1721 | } | |
1722 | } | |
1723 | ||
1724 | static struct ofbundle * | |
1725 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1726 | { | |
1727 | struct ofbundle *bundle; | |
1728 | ||
1729 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1730 | &ofproto->bundles) { | |
1731 | if (bundle->aux == aux) { | |
1732 | return bundle; | |
1733 | } | |
1734 | } | |
1735 | return NULL; | |
1736 | } | |
1737 | ||
1738 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1739 | * ones that are found to 'bundles'. */ | |
1740 | static void | |
1741 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1742 | void **auxes, size_t n_auxes, | |
1743 | struct hmapx *bundles) | |
1744 | { | |
1745 | size_t i; | |
1746 | ||
1747 | hmapx_init(bundles); | |
1748 | for (i = 0; i < n_auxes; i++) { | |
1749 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1750 | if (bundle) { | |
1751 | hmapx_add(bundles, bundle); | |
1752 | } | |
1753 | } | |
1754 | } | |
1755 | ||
7bde8dd8 JP |
1756 | static void |
1757 | bundle_update(struct ofbundle *bundle) | |
1758 | { | |
1759 | struct ofport_dpif *port; | |
1760 | ||
1761 | bundle->floodable = true; | |
1762 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
1763 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1764 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1765 | bundle->floodable = false; |
1766 | break; | |
1767 | } | |
1768 | } | |
1769 | } | |
1770 | ||
abe529af BP |
1771 | static void |
1772 | bundle_del_port(struct ofport_dpif *port) | |
1773 | { | |
1774 | struct ofbundle *bundle = port->bundle; | |
1775 | ||
3c4a309c | 1776 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 1777 | |
abe529af BP |
1778 | list_remove(&port->bundle_node); |
1779 | port->bundle = NULL; | |
1780 | ||
1781 | if (bundle->lacp) { | |
1782 | lacp_slave_unregister(bundle->lacp, port); | |
1783 | } | |
1784 | if (bundle->bond) { | |
1785 | bond_slave_unregister(bundle->bond, port); | |
1786 | } | |
1787 | ||
7bde8dd8 | 1788 | bundle_update(bundle); |
abe529af BP |
1789 | } |
1790 | ||
1791 | static bool | |
1792 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1793 | struct lacp_slave_settings *lacp, |
1794 | uint32_t bond_stable_id) | |
abe529af BP |
1795 | { |
1796 | struct ofport_dpif *port; | |
1797 | ||
1798 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1799 | if (!port) { | |
1800 | return false; | |
1801 | } | |
1802 | ||
1803 | if (port->bundle != bundle) { | |
3c4a309c | 1804 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1805 | if (port->bundle) { |
1806 | bundle_del_port(port); | |
1807 | } | |
1808 | ||
1809 | port->bundle = bundle; | |
1810 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
1811 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1812 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1813 | bundle->floodable = false; |
1814 | } | |
1815 | } | |
1816 | if (lacp) { | |
3c4a309c | 1817 | port->bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1818 | lacp_slave_register(bundle->lacp, port, lacp); |
1819 | } | |
1820 | ||
00794817 BP |
1821 | port->bond_stable_id = bond_stable_id; |
1822 | ||
abe529af BP |
1823 | return true; |
1824 | } | |
1825 | ||
1826 | static void | |
1827 | bundle_destroy(struct ofbundle *bundle) | |
1828 | { | |
1829 | struct ofproto_dpif *ofproto; | |
1830 | struct ofport_dpif *port, *next_port; | |
1831 | int i; | |
1832 | ||
1833 | if (!bundle) { | |
1834 | return; | |
1835 | } | |
1836 | ||
1837 | ofproto = bundle->ofproto; | |
1838 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1839 | struct ofmirror *m = ofproto->mirrors[i]; | |
1840 | if (m) { | |
1841 | if (m->out == bundle) { | |
1842 | mirror_destroy(m); | |
1843 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1844 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
3c4a309c | 1845 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1846 | } |
1847 | } | |
1848 | } | |
1849 | ||
1850 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1851 | bundle_del_port(port); | |
1852 | } | |
1853 | ||
b44a10b7 | 1854 | bundle_flush_macs(bundle, true); |
abe529af BP |
1855 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1856 | free(bundle->name); | |
1857 | free(bundle->trunks); | |
1858 | lacp_destroy(bundle->lacp); | |
1859 | bond_destroy(bundle->bond); | |
1860 | free(bundle); | |
1861 | } | |
1862 | ||
1863 | static int | |
1864 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1865 | const struct ofproto_bundle_settings *s) | |
1866 | { | |
1867 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1868 | bool need_flush = false; | |
abe529af BP |
1869 | struct ofport_dpif *port; |
1870 | struct ofbundle *bundle; | |
ecac4ebf BP |
1871 | unsigned long *trunks; |
1872 | int vlan; | |
abe529af BP |
1873 | size_t i; |
1874 | bool ok; | |
1875 | ||
1876 | if (!s) { | |
1877 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1878 | return 0; | |
1879 | } | |
1880 | ||
1881 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1882 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1883 | ||
1884 | bundle = bundle_lookup(ofproto, aux); | |
1885 | if (!bundle) { | |
1886 | bundle = xmalloc(sizeof *bundle); | |
1887 | ||
1888 | bundle->ofproto = ofproto; | |
1889 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1890 | hash_pointer(aux, 0)); | |
1891 | bundle->aux = aux; | |
1892 | bundle->name = NULL; | |
1893 | ||
1894 | list_init(&bundle->ports); | |
ecac4ebf | 1895 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1896 | bundle->vlan = -1; |
1897 | bundle->trunks = NULL; | |
5e9ceccd | 1898 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1899 | bundle->lacp = NULL; |
1900 | bundle->bond = NULL; | |
1901 | ||
1902 | bundle->floodable = true; | |
1903 | ||
1904 | bundle->src_mirrors = 0; | |
1905 | bundle->dst_mirrors = 0; | |
1906 | bundle->mirror_out = 0; | |
1907 | } | |
1908 | ||
1909 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1910 | free(bundle->name); | |
1911 | bundle->name = xstrdup(s->name); | |
1912 | } | |
1913 | ||
1914 | /* LACP. */ | |
1915 | if (s->lacp) { | |
1916 | if (!bundle->lacp) { | |
3c4a309c | 1917 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1918 | bundle->lacp = lacp_create(); |
1919 | } | |
1920 | lacp_configure(bundle->lacp, s->lacp); | |
1921 | } else { | |
1922 | lacp_destroy(bundle->lacp); | |
1923 | bundle->lacp = NULL; | |
1924 | } | |
1925 | ||
1926 | /* Update set of ports. */ | |
1927 | ok = true; | |
1928 | for (i = 0; i < s->n_slaves; i++) { | |
1929 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1930 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1931 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1932 | ok = false; |
1933 | } | |
1934 | } | |
1935 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1936 | struct ofport_dpif *next_port; | |
1937 | ||
1938 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1939 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1940 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1941 | goto found; |
1942 | } | |
1943 | } | |
1944 | ||
1945 | bundle_del_port(port); | |
1946 | found: ; | |
1947 | } | |
1948 | } | |
1949 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1950 | ||
1951 | if (list_is_empty(&bundle->ports)) { | |
1952 | bundle_destroy(bundle); | |
1953 | return EINVAL; | |
1954 | } | |
1955 | ||
ecac4ebf | 1956 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1957 | if (s->vlan_mode != bundle->vlan_mode |
1958 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1959 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1960 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1961 | need_flush = true; |
1962 | } | |
1963 | ||
abe529af | 1964 | /* Set VLAN tag. */ |
ecac4ebf BP |
1965 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1966 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1967 | : 0); | |
1968 | if (vlan != bundle->vlan) { | |
1969 | bundle->vlan = vlan; | |
abe529af BP |
1970 | need_flush = true; |
1971 | } | |
1972 | ||
1973 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1974 | switch (s->vlan_mode) { |
1975 | case PORT_VLAN_ACCESS: | |
1976 | trunks = NULL; | |
1977 | break; | |
1978 | ||
1979 | case PORT_VLAN_TRUNK: | |
ebc56baa | 1980 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
1981 | break; |
1982 | ||
1983 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1984 | case PORT_VLAN_NATIVE_TAGGED: | |
1985 | if (vlan != 0 && (!s->trunks | |
1986 | || !bitmap_is_set(s->trunks, vlan) | |
1987 | || bitmap_is_set(s->trunks, 0))) { | |
1988 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1989 | if (s->trunks) { | |
1990 | trunks = bitmap_clone(s->trunks, 4096); | |
1991 | } else { | |
1992 | trunks = bitmap_allocate1(4096); | |
1993 | } | |
1994 | bitmap_set1(trunks, vlan); | |
1995 | bitmap_set0(trunks, 0); | |
1996 | } else { | |
ebc56baa | 1997 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
1998 | } |
1999 | break; | |
2000 | ||
2001 | default: | |
2002 | NOT_REACHED(); | |
2003 | } | |
abe529af BP |
2004 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2005 | free(bundle->trunks); | |
ecac4ebf BP |
2006 | if (trunks == s->trunks) { |
2007 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2008 | } else { | |
2009 | bundle->trunks = trunks; | |
2010 | trunks = NULL; | |
2011 | } | |
abe529af BP |
2012 | need_flush = true; |
2013 | } | |
ecac4ebf BP |
2014 | if (trunks != s->trunks) { |
2015 | free(trunks); | |
2016 | } | |
abe529af BP |
2017 | |
2018 | /* Bonding. */ | |
2019 | if (!list_is_short(&bundle->ports)) { | |
2020 | bundle->ofproto->has_bonded_bundles = true; | |
2021 | if (bundle->bond) { | |
2022 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
3c4a309c | 2023 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2024 | } |
2025 | } else { | |
2026 | bundle->bond = bond_create(s->bond); | |
3c4a309c | 2027 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2028 | } |
2029 | ||
2030 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 2031 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
2032 | port->up.netdev); |
2033 | } | |
2034 | } else { | |
2035 | bond_destroy(bundle->bond); | |
2036 | bundle->bond = NULL; | |
2037 | } | |
2038 | ||
2039 | /* If we changed something that would affect MAC learning, un-learn | |
2040 | * everything on this port and force flow revalidation. */ | |
2041 | if (need_flush) { | |
b44a10b7 | 2042 | bundle_flush_macs(bundle, false); |
abe529af BP |
2043 | } |
2044 | ||
2045 | return 0; | |
2046 | } | |
2047 | ||
2048 | static void | |
2049 | bundle_remove(struct ofport *port_) | |
2050 | { | |
2051 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2052 | struct ofbundle *bundle = port->bundle; | |
2053 | ||
2054 | if (bundle) { | |
2055 | bundle_del_port(port); | |
2056 | if (list_is_empty(&bundle->ports)) { | |
2057 | bundle_destroy(bundle); | |
2058 | } else if (list_is_short(&bundle->ports)) { | |
2059 | bond_destroy(bundle->bond); | |
2060 | bundle->bond = NULL; | |
2061 | } | |
2062 | } | |
2063 | } | |
2064 | ||
2065 | static void | |
5f877369 | 2066 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2067 | { |
2068 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2069 | struct ofport_dpif *port = port_; | |
2070 | uint8_t ea[ETH_ADDR_LEN]; | |
2071 | int error; | |
2072 | ||
2073 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2074 | if (!error) { | |
abe529af | 2075 | struct ofpbuf packet; |
5f877369 | 2076 | void *packet_pdu; |
abe529af BP |
2077 | |
2078 | ofpbuf_init(&packet, 0); | |
2079 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2080 | pdu_size); |
2081 | memcpy(packet_pdu, pdu, pdu_size); | |
2082 | ||
97d6520b | 2083 | send_packet(port, &packet); |
abe529af BP |
2084 | ofpbuf_uninit(&packet); |
2085 | } else { | |
2086 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2087 | "%s (%s)", port->bundle->name, | |
2088 | netdev_get_name(port->up.netdev), strerror(error)); | |
2089 | } | |
2090 | } | |
2091 | ||
2092 | static void | |
2093 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2094 | { | |
2095 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2096 | int error, n_packets, n_errors; | |
2097 | struct mac_entry *e; | |
2098 | ||
2099 | error = n_packets = n_errors = 0; | |
2100 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2101 | if (e->port.p != bundle) { | |
ea131871 JG |
2102 | struct ofpbuf *learning_packet; |
2103 | struct ofport_dpif *port; | |
4dd1e3ca | 2104 | void *port_void; |
ea131871 JG |
2105 | int ret; |
2106 | ||
4dd1e3ca BP |
2107 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2108 | * struct ofport_dpif more effective. */ | |
2109 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2110 | e->mac, e->vlan, | |
2111 | &port_void); | |
2112 | port = port_void; | |
97d6520b | 2113 | ret = send_packet(port, learning_packet); |
ea131871 | 2114 | ofpbuf_delete(learning_packet); |
abe529af BP |
2115 | if (ret) { |
2116 | error = ret; | |
2117 | n_errors++; | |
2118 | } | |
2119 | n_packets++; | |
2120 | } | |
2121 | } | |
2122 | ||
2123 | if (n_errors) { | |
2124 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2125 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2126 | "packets, last error was: %s", | |
2127 | bundle->name, n_errors, n_packets, strerror(error)); | |
2128 | } else { | |
2129 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2130 | bundle->name, n_packets); | |
2131 | } | |
2132 | } | |
2133 | ||
2134 | static void | |
2135 | bundle_run(struct ofbundle *bundle) | |
2136 | { | |
2137 | if (bundle->lacp) { | |
2138 | lacp_run(bundle->lacp, send_pdu_cb); | |
2139 | } | |
2140 | if (bundle->bond) { | |
2141 | struct ofport_dpif *port; | |
2142 | ||
2143 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2144 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2145 | } |
2146 | ||
2147 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
bdebeece | 2148 | lacp_status(bundle->lacp)); |
abe529af BP |
2149 | if (bond_should_send_learning_packets(bundle->bond)) { |
2150 | bundle_send_learning_packets(bundle); | |
2151 | } | |
2152 | } | |
2153 | } | |
2154 | ||
2155 | static void | |
2156 | bundle_wait(struct ofbundle *bundle) | |
2157 | { | |
2158 | if (bundle->lacp) { | |
2159 | lacp_wait(bundle->lacp); | |
2160 | } | |
2161 | if (bundle->bond) { | |
2162 | bond_wait(bundle->bond); | |
2163 | } | |
2164 | } | |
2165 | \f | |
2166 | /* Mirrors. */ | |
2167 | ||
2168 | static int | |
2169 | mirror_scan(struct ofproto_dpif *ofproto) | |
2170 | { | |
2171 | int idx; | |
2172 | ||
2173 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2174 | if (!ofproto->mirrors[idx]) { | |
2175 | return idx; | |
2176 | } | |
2177 | } | |
2178 | return -1; | |
2179 | } | |
2180 | ||
2181 | static struct ofmirror * | |
2182 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2183 | { | |
2184 | int i; | |
2185 | ||
2186 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2187 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2188 | if (mirror && mirror->aux == aux) { | |
2189 | return mirror; | |
2190 | } | |
2191 | } | |
2192 | ||
2193 | return NULL; | |
2194 | } | |
2195 | ||
9ba15e2a BP |
2196 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2197 | static void | |
2198 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2199 | { | |
2200 | int i; | |
2201 | ||
2202 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2203 | struct ofmirror *m = ofproto->mirrors[i]; | |
2204 | ||
2205 | if (m) { | |
2206 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2211 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2212 | int j; | |
2213 | ||
2214 | if (!m1) { | |
2215 | continue; | |
2216 | } | |
2217 | ||
2218 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2219 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2220 | ||
edb0540b | 2221 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2222 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2223 | m2->dup_mirrors |= m1->dup_mirrors; | |
2224 | } | |
2225 | } | |
2226 | } | |
2227 | } | |
2228 | ||
abe529af BP |
2229 | static int |
2230 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2231 | const struct ofproto_mirror_settings *s) | |
2232 | { | |
2233 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2234 | mirror_mask_t mirror_bit; | |
2235 | struct ofbundle *bundle; | |
2236 | struct ofmirror *mirror; | |
2237 | struct ofbundle *out; | |
2238 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2239 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2240 | int out_vlan; | |
2241 | ||
2242 | mirror = mirror_lookup(ofproto, aux); | |
2243 | if (!s) { | |
2244 | mirror_destroy(mirror); | |
2245 | return 0; | |
2246 | } | |
2247 | if (!mirror) { | |
2248 | int idx; | |
2249 | ||
2250 | idx = mirror_scan(ofproto); | |
2251 | if (idx < 0) { | |
2252 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2253 | "cannot create %s", | |
2254 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2255 | return EFBIG; | |
2256 | } | |
2257 | ||
2258 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2259 | mirror->ofproto = ofproto; | |
2260 | mirror->idx = idx; | |
8b28d864 | 2261 | mirror->aux = aux; |
abe529af BP |
2262 | mirror->out_vlan = -1; |
2263 | mirror->name = NULL; | |
2264 | } | |
2265 | ||
2266 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2267 | free(mirror->name); | |
2268 | mirror->name = xstrdup(s->name); | |
2269 | } | |
2270 | ||
2271 | /* Get the new configuration. */ | |
2272 | if (s->out_bundle) { | |
2273 | out = bundle_lookup(ofproto, s->out_bundle); | |
2274 | if (!out) { | |
2275 | mirror_destroy(mirror); | |
2276 | return EINVAL; | |
2277 | } | |
2278 | out_vlan = -1; | |
2279 | } else { | |
2280 | out = NULL; | |
2281 | out_vlan = s->out_vlan; | |
2282 | } | |
2283 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2284 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2285 | ||
2286 | /* If the configuration has not changed, do nothing. */ | |
2287 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2288 | && hmapx_equals(&dsts, &mirror->dsts) | |
2289 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2290 | && mirror->out == out | |
2291 | && mirror->out_vlan == out_vlan) | |
2292 | { | |
2293 | hmapx_destroy(&srcs); | |
2294 | hmapx_destroy(&dsts); | |
2295 | return 0; | |
2296 | } | |
2297 | ||
2298 | hmapx_swap(&srcs, &mirror->srcs); | |
2299 | hmapx_destroy(&srcs); | |
2300 | ||
2301 | hmapx_swap(&dsts, &mirror->dsts); | |
2302 | hmapx_destroy(&dsts); | |
2303 | ||
2304 | free(mirror->vlans); | |
2305 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2306 | ||
2307 | mirror->out = out; | |
2308 | mirror->out_vlan = out_vlan; | |
2309 | ||
2310 | /* Update bundles. */ | |
2311 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2312 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2313 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2314 | bundle->src_mirrors |= mirror_bit; | |
2315 | } else { | |
2316 | bundle->src_mirrors &= ~mirror_bit; | |
2317 | } | |
2318 | ||
2319 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2320 | bundle->dst_mirrors |= mirror_bit; | |
2321 | } else { | |
2322 | bundle->dst_mirrors &= ~mirror_bit; | |
2323 | } | |
2324 | ||
2325 | if (mirror->out == bundle) { | |
2326 | bundle->mirror_out |= mirror_bit; | |
2327 | } else { | |
2328 | bundle->mirror_out &= ~mirror_bit; | |
2329 | } | |
2330 | } | |
2331 | ||
3c4a309c | 2332 | ofproto->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 2333 | ofproto->has_mirrors = true; |
d0040604 | 2334 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
9ba15e2a | 2335 | mirror_update_dups(ofproto); |
abe529af BP |
2336 | |
2337 | return 0; | |
2338 | } | |
2339 | ||
2340 | static void | |
2341 | mirror_destroy(struct ofmirror *mirror) | |
2342 | { | |
2343 | struct ofproto_dpif *ofproto; | |
2344 | mirror_mask_t mirror_bit; | |
2345 | struct ofbundle *bundle; | |
ccb7c863 | 2346 | int i; |
abe529af BP |
2347 | |
2348 | if (!mirror) { | |
2349 | return; | |
2350 | } | |
2351 | ||
2352 | ofproto = mirror->ofproto; | |
3c4a309c | 2353 | ofproto->need_revalidate = REV_RECONFIGURE; |
d0040604 | 2354 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2355 | |
2356 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2357 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2358 | bundle->src_mirrors &= ~mirror_bit; | |
2359 | bundle->dst_mirrors &= ~mirror_bit; | |
2360 | bundle->mirror_out &= ~mirror_bit; | |
2361 | } | |
2362 | ||
2363 | hmapx_destroy(&mirror->srcs); | |
2364 | hmapx_destroy(&mirror->dsts); | |
2365 | free(mirror->vlans); | |
2366 | ||
2367 | ofproto->mirrors[mirror->idx] = NULL; | |
2368 | free(mirror->name); | |
2369 | free(mirror); | |
9ba15e2a BP |
2370 | |
2371 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
2372 | |
2373 | ofproto->has_mirrors = false; | |
2374 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2375 | if (ofproto->mirrors[i]) { | |
2376 | ofproto->has_mirrors = true; | |
2377 | break; | |
2378 | } | |
2379 | } | |
abe529af BP |
2380 | } |
2381 | ||
9d24de3b JP |
2382 | static int |
2383 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2384 | uint64_t *packets, uint64_t *bytes) | |
2385 | { | |
2386 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2387 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2388 | ||
2389 | if (!mirror) { | |
2390 | *packets = *bytes = UINT64_MAX; | |
2391 | return 0; | |
2392 | } | |
2393 | ||
2394 | *packets = mirror->packet_count; | |
2395 | *bytes = mirror->byte_count; | |
2396 | ||
2397 | return 0; | |
2398 | } | |
2399 | ||
abe529af BP |
2400 | static int |
2401 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2402 | { | |
2403 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2404 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
d0040604 | 2405 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2406 | } |
2407 | return 0; | |
2408 | } | |
2409 | ||
2410 | static bool | |
b4affc74 | 2411 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2412 | { |
2413 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2414 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2415 | return bundle && bundle->mirror_out != 0; | |
2416 | } | |
8402c74b SS |
2417 | |
2418 | static void | |
b53055f4 | 2419 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2420 | { |
2421 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3c4a309c | 2422 | ofproto->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2423 | } |
e764773c BP |
2424 | |
2425 | static void | |
2426 | set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) | |
2427 | { | |
2428 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2429 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
2430 | } | |
abe529af BP |
2431 | \f |
2432 | /* Ports. */ | |
2433 | ||
2434 | static struct ofport_dpif * | |
4acbc98d | 2435 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 2436 | { |
7df6a8bd BP |
2437 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2438 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2439 | } |
2440 | ||
2441 | static struct ofport_dpif * | |
4acbc98d | 2442 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af BP |
2443 | { |
2444 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2445 | } | |
2446 | ||
2447 | static void | |
2448 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2449 | struct dpif_port *dpif_port) | |
2450 | { | |
2451 | ofproto_port->name = dpif_port->name; | |
2452 | ofproto_port->type = dpif_port->type; | |
2453 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2454 | } | |
2455 | ||
0aa66d6e EJ |
2456 | static void |
2457 | port_run_fast(struct ofport_dpif *ofport) | |
2458 | { | |
2459 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2460 | struct ofpbuf packet; | |
2461 | ||
2462 | ofpbuf_init(&packet, 0); | |
2463 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2464 | send_packet(ofport, &packet); | |
2465 | ofpbuf_uninit(&packet); | |
2466 | } | |
2467 | } | |
2468 | ||
abe529af BP |
2469 | static void |
2470 | port_run(struct ofport_dpif *ofport) | |
2471 | { | |
3e5b3fdb EJ |
2472 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2473 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2474 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2475 | ||
3e5b3fdb EJ |
2476 | ofport->carrier_seq = carrier_seq; |
2477 | ||
0aa66d6e | 2478 | port_run_fast(ofport); |
abe529af | 2479 | if (ofport->cfm) { |
4653c558 EJ |
2480 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2481 | ||
abe529af | 2482 | cfm_run(ofport->cfm); |
4653c558 EJ |
2483 | enable = enable && !cfm_get_fault(ofport->cfm); |
2484 | ||
2485 | if (cfm_opup >= 0) { | |
2486 | enable = enable && cfm_opup; | |
2487 | } | |
abe529af | 2488 | } |
015e08bc EJ |
2489 | |
2490 | if (ofport->bundle) { | |
2491 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2492 | if (carrier_changed) { |
2493 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2494 | } | |
015e08bc EJ |
2495 | } |
2496 | ||
daff3353 EJ |
2497 | if (ofport->may_enable != enable) { |
2498 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2499 | ||
2500 | if (ofproto->has_bundle_action) { | |
3c4a309c | 2501 | ofproto->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
2502 | } |
2503 | } | |
2504 | ||
015e08bc | 2505 | ofport->may_enable = enable; |
abe529af BP |
2506 | } |
2507 | ||
2508 | static void | |
2509 | port_wait(struct ofport_dpif *ofport) | |
2510 | { | |
2511 | if (ofport->cfm) { | |
2512 | cfm_wait(ofport->cfm); | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | static int | |
2517 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2518 | struct ofproto_port *ofproto_port) | |
2519 | { | |
2520 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2521 | struct dpif_port dpif_port; | |
2522 | int error; | |
2523 | ||
2524 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2525 | if (!error) { | |
2526 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2527 | } | |
2528 | return error; | |
2529 | } | |
2530 | ||
2531 | static int | |
2532 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2533 | { | |
2534 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
232dfa4a | 2535 | uint16_t odp_port = UINT16_MAX; |
abe529af BP |
2536 | int error; |
2537 | ||
2538 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2539 | if (!error) { | |
2540 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2541 | } | |
2542 | return error; | |
2543 | } | |
2544 | ||
2545 | static int | |
2546 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2547 | { | |
2548 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2549 | int error; | |
2550 | ||
2551 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2552 | if (!error) { | |
2553 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2554 | if (ofport) { | |
2555 | /* The caller is going to close ofport->up.netdev. If this is a | |
2556 | * bonded port, then the bond is using that netdev, so remove it | |
2557 | * from the bond. The client will need to reconfigure everything | |
2558 | * after deleting ports, so then the slave will get re-added. */ | |
2559 | bundle_remove(&ofport->up); | |
2560 | } | |
2561 | } | |
2562 | return error; | |
2563 | } | |
2564 | ||
6527c598 PS |
2565 | static int |
2566 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2567 | { | |
2568 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2569 | int error; | |
2570 | ||
2571 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2572 | ||
2573 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2574 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2575 | ||
2576 | /* ofproto->stats.tx_packets represents packets that we created | |
2577 | * internally and sent to some port (e.g. packets sent with | |
2578 | * send_packet()). Account for them as if they had come from | |
2579 | * OFPP_LOCAL and got forwarded. */ | |
2580 | ||
2581 | if (stats->rx_packets != UINT64_MAX) { | |
2582 | stats->rx_packets += ofproto->stats.tx_packets; | |
2583 | } | |
2584 | ||
2585 | if (stats->rx_bytes != UINT64_MAX) { | |
2586 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2587 | } | |
2588 | ||
2589 | /* ofproto->stats.rx_packets represents packets that were received on | |
2590 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 2591 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
2592 | |
2593 | if (stats->tx_packets != UINT64_MAX) { | |
2594 | stats->tx_packets += ofproto->stats.rx_packets; | |
2595 | } | |
2596 | ||
2597 | if (stats->tx_bytes != UINT64_MAX) { | |
2598 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2599 | } | |
2600 | } | |
2601 | ||
2602 | return error; | |
2603 | } | |
2604 | ||
2605 | /* Account packets for LOCAL port. */ | |
2606 | static void | |
2607 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2608 | size_t tx_size, size_t rx_size) | |
2609 | { | |
2610 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2611 | ||
2612 | if (rx_size) { | |
2613 | ofproto->stats.rx_packets++; | |
2614 | ofproto->stats.rx_bytes += rx_size; | |
2615 | } | |
2616 | if (tx_size) { | |
2617 | ofproto->stats.tx_packets++; | |
2618 | ofproto->stats.tx_bytes += tx_size; | |
2619 | } | |
2620 | } | |
2621 | ||
abe529af BP |
2622 | struct port_dump_state { |
2623 | struct dpif_port_dump dump; | |
2624 | bool done; | |
2625 | }; | |
2626 | ||
2627 | static int | |
2628 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2629 | { | |
2630 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2631 | struct port_dump_state *state; | |
2632 | ||
2633 | *statep = state = xmalloc(sizeof *state); | |
2634 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2635 | state->done = false; | |
2636 | return 0; | |
2637 | } | |
2638 | ||
2639 | static int | |
2640 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2641 | struct ofproto_port *port) | |
2642 | { | |
2643 | struct port_dump_state *state = state_; | |
2644 | struct dpif_port dpif_port; | |
2645 | ||
2646 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2647 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2648 | return 0; | |
2649 | } else { | |
2650 | int error = dpif_port_dump_done(&state->dump); | |
2651 | state->done = true; | |
2652 | return error ? error : EOF; | |
2653 | } | |
2654 | } | |
2655 | ||
2656 | static int | |
2657 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2658 | { | |
2659 | struct port_dump_state *state = state_; | |
2660 | ||
2661 | if (!state->done) { | |
2662 | dpif_port_dump_done(&state->dump); | |
2663 | } | |
2664 | free(state); | |
2665 | return 0; | |
2666 | } | |
2667 | ||
2668 | static int | |
2669 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2670 | { | |
2671 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2672 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2673 | } | |
2674 | ||
2675 | static void | |
2676 | port_poll_wait(const struct ofproto *ofproto_) | |
2677 | { | |
2678 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2679 | dpif_port_poll_wait(ofproto->dpif); | |
2680 | } | |
2681 | ||
2682 | static int | |
2683 | port_is_lacp_current(const struct ofport *ofport_) | |
2684 | { | |
2685 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2686 | return (ofport->bundle && ofport->bundle->lacp | |
2687 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2688 | : -1); | |
2689 | } | |
2690 | \f | |
2691 | /* Upcall handling. */ | |
2692 | ||
501f8d1f BP |
2693 | /* Flow miss batching. |
2694 | * | |
2695 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2696 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2697 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2698 | * more packets, plus possibly installing the flow in the dpif. | |
2699 | * | |
2700 | * So far we only batch the operations that affect flow setup time the most. | |
2701 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2702 | struct flow_miss { | |
2703 | struct hmap_node hmap_node; | |
2704 | struct flow flow; | |
b0f7b9b5 | 2705 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2706 | const struct nlattr *key; |
2707 | size_t key_len; | |
e84173dc | 2708 | ovs_be16 initial_tci; |
501f8d1f | 2709 | struct list packets; |
6a7e895f | 2710 | enum dpif_upcall_type upcall_type; |
501f8d1f BP |
2711 | }; |
2712 | ||
2713 | struct flow_miss_op { | |
c2b565b5 | 2714 | struct dpif_op dpif_op; |
5fe20d5d BP |
2715 | struct subfacet *subfacet; /* Subfacet */ |
2716 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
2717 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
2718 | }; |
2719 | ||
62cd7072 BP |
2720 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2721 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2722 | * configurations. */ |
62cd7072 | 2723 | static void |
a39edbd4 | 2724 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 2725 | const struct flow *flow) |
62cd7072 BP |
2726 | { |
2727 | struct ofputil_packet_in pin; | |
2728 | ||
3e3252fa EJ |
2729 | pin.packet = packet->data; |
2730 | pin.packet_len = packet->size; | |
62cd7072 | 2731 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 2732 | pin.controller_id = 0; |
54834960 EJ |
2733 | |
2734 | pin.table_id = 0; | |
2735 | pin.cookie = 0; | |
2736 | ||
62cd7072 | 2737 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
2738 | |
2739 | flow_get_metadata(flow, &pin.fmd); | |
2740 | ||
d8653c38 | 2741 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
2742 | } |
2743 | ||
6a7e895f | 2744 | static enum slow_path_reason |
abe529af BP |
2745 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
2746 | const struct ofpbuf *packet) | |
2747 | { | |
b6e001b6 EJ |
2748 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2749 | ||
2750 | if (!ofport) { | |
6a7e895f | 2751 | return 0; |
b6e001b6 EJ |
2752 | } |
2753 | ||
ef9819b5 | 2754 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2755 | if (packet) { |
abe529af BP |
2756 | cfm_process_heartbeat(ofport->cfm, packet); |
2757 | } | |
6a7e895f | 2758 | return SLOW_CFM; |
b6e001b6 EJ |
2759 | } else if (ofport->bundle && ofport->bundle->lacp |
2760 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2761 | if (packet) { | |
2762 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2763 | } |
6a7e895f | 2764 | return SLOW_LACP; |
21f7563c JP |
2765 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2766 | if (packet) { | |
2767 | stp_process_packet(ofport, packet); | |
2768 | } | |
6a7e895f | 2769 | return SLOW_STP; |
abe529af | 2770 | } |
6a7e895f | 2771 | return 0; |
abe529af BP |
2772 | } |
2773 | ||
501f8d1f | 2774 | static struct flow_miss * |
b23cdad9 | 2775 | flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) |
abe529af | 2776 | { |
501f8d1f | 2777 | struct flow_miss *miss; |
abe529af | 2778 | |
501f8d1f BP |
2779 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2780 | if (flow_equal(&miss->flow, flow)) { | |
2781 | return miss; | |
2782 | } | |
2783 | } | |
abe529af | 2784 | |
b23cdad9 | 2785 | return NULL; |
501f8d1f | 2786 | } |
abe529af | 2787 | |
9d6ac44e BP |
2788 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
2789 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
2790 | * 'miss' is associated with a subfacet the caller must also initialize the | |
2791 | * returned op->subfacet, and if anything needs to be freed after processing | |
2792 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 2793 | static void |
9d6ac44e BP |
2794 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
2795 | struct flow_miss_op *op) | |
501f8d1f | 2796 | { |
9d6ac44e BP |
2797 | if (miss->flow.vlan_tci != miss->initial_tci) { |
2798 | /* This packet was received on a VLAN splinter port. We | |
2799 | * added a VLAN to the packet to make the packet resemble | |
2800 | * the flow, but the actions were composed assuming that | |
2801 | * the packet contained no VLAN. So, we must remove the | |
2802 | * VLAN header from the packet before trying to execute the | |
2803 | * actions. */ | |
2804 | eth_pop_vlan(packet); | |
2805 | } | |
2806 | ||
2807 | op->subfacet = NULL; | |
2808 | op->garbage = NULL; | |
2809 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
2810 | op->dpif_op.u.execute.key = miss->key; | |
2811 | op->dpif_op.u.execute.key_len = miss->key_len; | |
2812 | op->dpif_op.u.execute.packet = packet; | |
2813 | } | |
2814 | ||
2815 | /* Helper for handle_flow_miss_without_facet() and | |
2816 | * handle_flow_miss_with_facet(). */ | |
2817 | static void | |
2818 | handle_flow_miss_common(struct rule_dpif *rule, | |
2819 | struct ofpbuf *packet, const struct flow *flow) | |
2820 | { | |
2821 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2822 | ||
2823 | ofproto->n_matches++; | |
2824 | ||
2825 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2826 | /* | |
2827 | * Extra-special case for fail-open mode. | |
2828 | * | |
2829 | * We are in fail-open mode and the packet matched the fail-open | |
2830 | * rule, but we are connected to a controller too. We should send | |
2831 | * the packet up to the controller in the hope that it will try to | |
2832 | * set up a flow and thereby allow us to exit fail-open. | |
2833 | * | |
2834 | * See the top-level comment in fail-open.c for more information. | |
2835 | */ | |
2836 | send_packet_in_miss(ofproto, packet, flow); | |
2837 | } | |
2838 | } | |
2839 | ||
2840 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
2841 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
2842 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
2843 | * true). However, for short flows the cost of bookkeeping is much higher than | |
2844 | * the benefits, so when the datapath holds a large number of flows we impose | |
2845 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
2846 | static bool | |
2847 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
2848 | struct flow_miss *miss, uint32_t hash) | |
2849 | { | |
2850 | if (!ofproto->governor) { | |
2851 | size_t n_subfacets; | |
2852 | ||
2853 | n_subfacets = hmap_count(&ofproto->subfacets); | |
2854 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
2855 | return true; | |
2856 | } | |
2857 | ||
2858 | ofproto->governor = governor_create(ofproto->up.name); | |
2859 | } | |
2860 | ||
2861 | return governor_should_install_flow(ofproto->governor, hash, | |
2862 | list_size(&miss->packets)); | |
2863 | } | |
2864 | ||
2865 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
2866 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
2867 | * increment '*n_ops'. */ | |
2868 | static void | |
2869 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
2870 | struct rule_dpif *rule, | |
2871 | struct flow_miss_op *ops, size_t *n_ops) | |
2872 | { | |
2873 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
a7752d4a | 2874 | long long int now = time_msec(); |
9d6ac44e | 2875 | struct action_xlate_ctx ctx; |
530a1d91 | 2876 | struct ofpbuf *packet; |
2b459b83 | 2877 | |
9d6ac44e BP |
2878 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2879 | struct flow_miss_op *op = &ops[*n_ops]; | |
2880 | struct dpif_flow_stats stats; | |
2881 | struct ofpbuf odp_actions; | |
abe529af | 2882 | |
9d6ac44e | 2883 | COVERAGE_INC(facet_suppress); |
501f8d1f | 2884 | |
9d6ac44e | 2885 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 2886 | |
a7752d4a | 2887 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 2888 | rule_credit_stats(rule, &stats); |
abe529af | 2889 | |
9d6ac44e BP |
2890 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, |
2891 | rule, 0, packet); | |
2892 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 2893 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
9d6ac44e | 2894 | &odp_actions); |
abe529af | 2895 | |
9d6ac44e BP |
2896 | if (odp_actions.size) { |
2897 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
2898 | ||
2899 | init_flow_miss_execute_op(miss, packet, op); | |
2900 | execute->actions = odp_actions.data; | |
2901 | execute->actions_len = odp_actions.size; | |
2902 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2903 | ||
2904 | (*n_ops)++; | |
2905 | } else { | |
2906 | ofpbuf_uninit(&odp_actions); | |
2907 | } | |
abe529af | 2908 | } |
9d6ac44e BP |
2909 | } |
2910 | ||
2911 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
2912 | * operations to 'ops', incrementing '*n_ops' for each new op. |
2913 | * | |
2914 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
2915 | * This is really important only for new facets: if we just called time_msec() | |
2916 | * here, then the new subfacet or its packets could look (occasionally) as | |
2917 | * though it was used some time after the facet was used. That can make a | |
2918 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
2919 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
2920 | static void |
2921 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 2922 | long long int now, |
9d6ac44e BP |
2923 | struct flow_miss_op *ops, size_t *n_ops) |
2924 | { | |
6a7e895f BP |
2925 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
2926 | enum subfacet_path want_path; | |
9d6ac44e BP |
2927 | struct subfacet *subfacet; |
2928 | struct ofpbuf *packet; | |
abe529af | 2929 | |
15baa734 | 2930 | subfacet = subfacet_create(facet, |
e84173dc | 2931 | miss->key_fitness, miss->key, miss->key_len, |
459b16a1 | 2932 | miss->initial_tci, now); |
b0f7b9b5 | 2933 | |
530a1d91 | 2934 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 2935 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 2936 | struct dpif_flow_stats stats; |
5fe20d5d | 2937 | struct ofpbuf odp_actions; |
67d91f78 | 2938 | |
9d6ac44e | 2939 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 2940 | |
5fe20d5d | 2941 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
6a7e895f | 2942 | if (!subfacet->actions || subfacet->slow) { |
5fe20d5d | 2943 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 2944 | } |
67d91f78 | 2945 | |
459b16a1 | 2946 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 2947 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2948 | |
9d6ac44e BP |
2949 | if (subfacet->actions_len) { |
2950 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 2951 | |
9d6ac44e BP |
2952 | init_flow_miss_execute_op(miss, packet, op); |
2953 | op->subfacet = subfacet; | |
6a7e895f | 2954 | if (!subfacet->slow) { |
9d6ac44e BP |
2955 | execute->actions = subfacet->actions; |
2956 | execute->actions_len = subfacet->actions_len; | |
2957 | ofpbuf_uninit(&odp_actions); | |
2958 | } else { | |
2959 | execute->actions = odp_actions.data; | |
2960 | execute->actions_len = odp_actions.size; | |
2961 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2962 | } | |
999fba59 | 2963 | |
9d6ac44e | 2964 | (*n_ops)++; |
5fe20d5d | 2965 | } else { |
9d6ac44e | 2966 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 2967 | } |
501f8d1f BP |
2968 | } |
2969 | ||
6a7e895f BP |
2970 | want_path = subfacet_want_path(subfacet->slow); |
2971 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
501f8d1f | 2972 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2973 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2974 | |
b0f7b9b5 | 2975 | op->subfacet = subfacet; |
5fe20d5d | 2976 | op->garbage = NULL; |
c2b565b5 | 2977 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2978 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2979 | put->key = miss->key; | |
2980 | put->key_len = miss->key_len; | |
6a7e895f BP |
2981 | if (want_path == SF_FAST_PATH) { |
2982 | put->actions = subfacet->actions; | |
2983 | put->actions_len = subfacet->actions_len; | |
2984 | } else { | |
2985 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
2986 | op->stub, sizeof op->stub, | |
2987 | &put->actions, &put->actions_len); | |
2988 | } | |
501f8d1f BP |
2989 | put->stats = NULL; |
2990 | } | |
2991 | } | |
2992 | ||
9d6ac44e BP |
2993 | /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath |
2994 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2995 | static void | |
2996 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2997 | struct flow_miss_op *ops, size_t *n_ops) | |
2998 | { | |
2999 | struct facet *facet; | |
459b16a1 | 3000 | long long int now; |
9d6ac44e BP |
3001 | uint32_t hash; |
3002 | ||
3003 | /* The caller must ensure that miss->hmap_node.hash contains | |
3004 | * flow_hash(miss->flow, 0). */ | |
3005 | hash = miss->hmap_node.hash; | |
3006 | ||
3007 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3008 | if (!facet) { | |
c57b2226 BP |
3009 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); |
3010 | ||
3011 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
9d6ac44e BP |
3012 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); |
3013 | return; | |
3014 | } | |
3015 | ||
3016 | facet = facet_create(rule, &miss->flow, hash); | |
459b16a1 BP |
3017 | now = facet->used; |
3018 | } else { | |
3019 | now = time_msec(); | |
9d6ac44e | 3020 | } |
459b16a1 | 3021 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3022 | } |
3023 | ||
e2a6ca36 BP |
3024 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
3025 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
3026 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
3027 | * what a flow key should contain. | |
3028 | * | |
3029 | * This function also includes some logic to help make VLAN splinters | |
3030 | * transparent to the rest of the upcall processing logic. In particular, if | |
3031 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
3032 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
3033 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
3034 | * | |
3035 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
3036 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
3037 | * (This differs from the value returned in flow->vlan_tci only for packets | |
3038 | * received on VLAN splinters.) | |
3039 | */ | |
e84173dc | 3040 | static enum odp_key_fitness |
52a90c29 | 3041 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 3042 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
3043 | struct flow *flow, ovs_be16 *initial_tci, |
3044 | struct ofpbuf *packet) | |
e84173dc BP |
3045 | { |
3046 | enum odp_key_fitness fitness; | |
3047 | ||
3048 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
3049 | if (fitness == ODP_FIT_ERROR) { | |
3050 | return fitness; | |
3051 | } | |
3052 | *initial_tci = flow->vlan_tci; | |
3053 | ||
b98d8985 | 3054 | if (vsp_adjust_flow(ofproto, flow)) { |
e2a6ca36 BP |
3055 | if (packet) { |
3056 | /* Make the packet resemble the flow, so that it gets sent to an | |
3057 | * OpenFlow controller properly, so that it looks correct for | |
3058 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
3059 | * if it is called on 'packet'. | |
3060 | * | |
3061 | * The allocated space inside 'packet' probably also contains | |
3062 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
3063 | * struct dpif_upcall (see the large comment on that structure | |
3064 | * definition), so pushing data on 'packet' is in general not a | |
3065 | * good idea since it could overwrite 'key' or free it as a side | |
3066 | * effect. However, it's OK in this special case because we know | |
3067 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
3068 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
3069 | * since we don't need that header anymore. */ | |
3070 | eth_push_vlan(packet, flow->vlan_tci); | |
3071 | } | |
52a90c29 BP |
3072 | |
3073 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
3074 | if (fitness == ODP_FIT_PERFECT) { | |
3075 | fitness = ODP_FIT_TOO_MUCH; | |
3076 | } | |
3077 | } | |
3078 | ||
e84173dc BP |
3079 | return fitness; |
3080 | } | |
3081 | ||
501f8d1f BP |
3082 | static void |
3083 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
3084 | size_t n_upcalls) | |
3085 | { | |
3086 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3087 | struct flow_miss *miss; |
3088 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3089 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3090 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3091 | struct hmap todo; |
b23cdad9 | 3092 | int n_misses; |
501f8d1f BP |
3093 | size_t n_ops; |
3094 | size_t i; | |
3095 | ||
3096 | if (!n_upcalls) { | |
3097 | return; | |
3098 | } | |
3099 | ||
3100 | /* Construct the to-do list. | |
3101 | * | |
3102 | * This just amounts to extracting the flow from each packet and sticking | |
3103 | * the packets that have the same flow in the same "flow_miss" structure so | |
3104 | * that we can process them together. */ | |
3105 | hmap_init(&todo); | |
b23cdad9 | 3106 | n_misses = 0; |
501f8d1f | 3107 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3108 | struct flow_miss *miss = &misses[n_misses]; |
3109 | struct flow_miss *existing_miss; | |
3110 | uint32_t hash; | |
501f8d1f | 3111 | |
b0f7b9b5 BP |
3112 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
3113 | * then set 'flow''s header pointers. */ | |
b23cdad9 BP |
3114 | miss->key_fitness = ofproto_dpif_extract_flow_key( |
3115 | ofproto, upcall->key, upcall->key_len, | |
3116 | &miss->flow, &miss->initial_tci, upcall->packet); | |
3117 | if (miss->key_fitness == ODP_FIT_ERROR) { | |
b0f7b9b5 BP |
3118 | continue; |
3119 | } | |
b23cdad9 BP |
3120 | flow_extract(upcall->packet, miss->flow.skb_priority, |
3121 | miss->flow.tun_id, miss->flow.in_port, &miss->flow); | |
501f8d1f | 3122 | |
501f8d1f | 3123 | /* Add other packets to a to-do list. */ |
b23cdad9 BP |
3124 | hash = flow_hash(&miss->flow, 0); |
3125 | existing_miss = flow_miss_find(&todo, &miss->flow, hash); | |
3126 | if (!existing_miss) { | |
3127 | hmap_insert(&todo, &miss->hmap_node, hash); | |
3128 | miss->key = upcall->key; | |
3129 | miss->key_len = upcall->key_len; | |
6a7e895f | 3130 | miss->upcall_type = upcall->type; |
b23cdad9 BP |
3131 | list_init(&miss->packets); |
3132 | ||
3133 | n_misses++; | |
3134 | } else { | |
3135 | miss = existing_miss; | |
3136 | } | |
501f8d1f BP |
3137 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3138 | } | |
3139 | ||
3140 | /* Process each element in the to-do list, constructing the set of | |
3141 | * operations to batch. */ | |
3142 | n_ops = 0; | |
33bb0caa | 3143 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
501f8d1f | 3144 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); |
abe529af | 3145 | } |
501f8d1f | 3146 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3147 | |
3148 | /* Execute batch. */ | |
3149 | for (i = 0; i < n_ops; i++) { | |
3150 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3151 | } | |
3152 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
3153 | ||
3154 | /* Free memory and update facets. */ | |
3155 | for (i = 0; i < n_ops; i++) { | |
3156 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
501f8d1f BP |
3157 | |
3158 | switch (op->dpif_op.type) { | |
3159 | case DPIF_OP_EXECUTE: | |
501f8d1f | 3160 | break; |
abe529af | 3161 | |
501f8d1f | 3162 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 3163 | if (!op->dpif_op.error) { |
6a7e895f | 3164 | op->subfacet->path = subfacet_want_path(op->subfacet->slow); |
501f8d1f BP |
3165 | } |
3166 | break; | |
b99d3cee BP |
3167 | |
3168 | case DPIF_OP_FLOW_DEL: | |
3169 | NOT_REACHED(); | |
501f8d1f | 3170 | } |
5fe20d5d BP |
3171 | |
3172 | free(op->garbage); | |
501f8d1f | 3173 | } |
33bb0caa | 3174 | hmap_destroy(&todo); |
abe529af BP |
3175 | } |
3176 | ||
6a7e895f BP |
3177 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } |
3178 | classify_upcall(const struct dpif_upcall *upcall) | |
3179 | { | |
3180 | union user_action_cookie cookie; | |
3181 | ||
3182 | /* First look at the upcall type. */ | |
3183 | switch (upcall->type) { | |
3184 | case DPIF_UC_ACTION: | |
3185 | break; | |
3186 | ||
3187 | case DPIF_UC_MISS: | |
3188 | return MISS_UPCALL; | |
3189 | ||
3190 | case DPIF_N_UC_TYPES: | |
3191 | default: | |
3192 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3193 | return BAD_UPCALL; | |
3194 | } | |
3195 | ||
3196 | /* "action" upcalls need a closer look. */ | |
3197 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); | |
3198 | switch (cookie.type) { | |
3199 | case USER_ACTION_COOKIE_SFLOW: | |
3200 | return SFLOW_UPCALL; | |
3201 | ||
3202 | case USER_ACTION_COOKIE_SLOW_PATH: | |
3203 | return MISS_UPCALL; | |
3204 | ||
3205 | case USER_ACTION_COOKIE_UNSPEC: | |
3206 | default: | |
3207 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
3208 | return BAD_UPCALL; | |
3209 | } | |
3210 | } | |
3211 | ||
abe529af | 3212 | static void |
6a7e895f BP |
3213 | handle_sflow_upcall(struct ofproto_dpif *ofproto, |
3214 | const struct dpif_upcall *upcall) | |
abe529af | 3215 | { |
1673e0e4 | 3216 | union user_action_cookie cookie; |
e84173dc BP |
3217 | enum odp_key_fitness fitness; |
3218 | ovs_be16 initial_tci; | |
3219 | struct flow flow; | |
abe529af | 3220 | |
e84173dc BP |
3221 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
3222 | upcall->key_len, &flow, | |
e2a6ca36 | 3223 | &initial_tci, upcall->packet); |
e84173dc BP |
3224 | if (fitness == ODP_FIT_ERROR) { |
3225 | return; | |
3226 | } | |
3227 | ||
6a7e895f BP |
3228 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
3229 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); | |
6ff686f2 PS |
3230 | } |
3231 | ||
9b16c439 BP |
3232 | static int |
3233 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 3234 | { |
9b16c439 | 3235 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3236 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3237 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3238 | int n_processed; | |
9b16c439 BP |
3239 | int n_misses; |
3240 | int i; | |
abe529af | 3241 | |
90a7c55e | 3242 | assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3243 | |
9b16c439 | 3244 | n_misses = 0; |
90a7c55e | 3245 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3246 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3247 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3248 | int error; |
3249 | ||
90a7c55e BP |
3250 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3251 | sizeof miss_buf_stubs[n_misses]); | |
3252 | error = dpif_recv(ofproto->dpif, upcall, buf); | |
9b16c439 | 3253 | if (error) { |
90a7c55e | 3254 | ofpbuf_uninit(buf); |
9b16c439 BP |
3255 | break; |
3256 | } | |
3257 | ||
6a7e895f BP |
3258 | switch (classify_upcall(upcall)) { |
3259 | case MISS_UPCALL: | |
9b16c439 BP |
3260 | /* Handle it later. */ |
3261 | n_misses++; | |
3262 | break; | |
3263 | ||
6a7e895f BP |
3264 | case SFLOW_UPCALL: |
3265 | if (ofproto->sflow) { | |
3266 | handle_sflow_upcall(ofproto, upcall); | |
3267 | } | |
3268 | ofpbuf_uninit(buf); | |
3269 | break; | |
3270 | ||
3271 | case BAD_UPCALL: | |
3272 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3273 | break; |
3274 | } | |
abe529af | 3275 | } |
9b16c439 | 3276 | |
6a7e895f | 3277 | /* Handle deferred MISS_UPCALL processing. */ |
9b16c439 | 3278 | handle_miss_upcalls(ofproto, misses, n_misses); |
90a7c55e BP |
3279 | for (i = 0; i < n_misses; i++) { |
3280 | ofpbuf_uninit(&miss_bufs[i]); | |
3281 | } | |
9b16c439 | 3282 | |
90a7c55e | 3283 | return n_processed; |
abe529af BP |
3284 | } |
3285 | \f | |
3286 | /* Flow expiration. */ | |
3287 | ||
b0f7b9b5 | 3288 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
3289 | static void update_stats(struct ofproto_dpif *); |
3290 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 3291 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
3292 | |
3293 | /* This function is called periodically by run(). Its job is to collect | |
3294 | * updates for the flows that have been installed into the datapath, most | |
3295 | * importantly when they last were used, and then use that information to | |
3296 | * expire flows that have not been used recently. | |
3297 | * | |
3298 | * Returns the number of milliseconds after which it should be called again. */ | |
3299 | static int | |
3300 | expire(struct ofproto_dpif *ofproto) | |
3301 | { | |
3302 | struct rule_dpif *rule, *next_rule; | |
d0918789 | 3303 | struct oftable *table; |
abe529af BP |
3304 | int dp_max_idle; |
3305 | ||
3306 | /* Update stats for each flow in the datapath. */ | |
3307 | update_stats(ofproto); | |
3308 | ||
b0f7b9b5 BP |
3309 | /* Expire subfacets that have been idle too long. */ |
3310 | dp_max_idle = subfacet_max_idle(ofproto); | |
3311 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
3312 | |
3313 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
3314 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
3315 | struct cls_cursor cursor; | |
3316 | ||
d0918789 | 3317 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
3318 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
3319 | rule_expire(rule); | |
3320 | } | |
abe529af BP |
3321 | } |
3322 | ||
3323 | /* All outstanding data in existing flows has been accounted, so it's a | |
3324 | * good time to do bond rebalancing. */ | |
3325 | if (ofproto->has_bonded_bundles) { | |
3326 | struct ofbundle *bundle; | |
3327 | ||
3328 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3329 | if (bundle->bond) { | |
3330 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
3331 | } | |
3332 | } | |
3333 | } | |
3334 | ||
3335 | return MIN(dp_max_idle, 1000); | |
3336 | } | |
3337 | ||
a218c879 BP |
3338 | /* Updates flow table statistics given that the datapath just reported 'stats' |
3339 | * as 'subfacet''s statistics. */ | |
3340 | static void | |
3341 | update_subfacet_stats(struct subfacet *subfacet, | |
3342 | const struct dpif_flow_stats *stats) | |
3343 | { | |
3344 | struct facet *facet = subfacet->facet; | |
3345 | ||
3346 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
3347 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
3348 | facet->packet_count += extra; | |
3349 | } else { | |
3350 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
3351 | } | |
3352 | ||
3353 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
3354 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
3355 | } else { | |
3356 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
3357 | } | |
3358 | ||
3359 | subfacet->dp_packet_count = stats->n_packets; | |
3360 | subfacet->dp_byte_count = stats->n_bytes; | |
3361 | ||
3362 | facet->tcp_flags |= stats->tcp_flags; | |
3363 | ||
3364 | subfacet_update_time(subfacet, stats->used); | |
3365 | if (facet->accounted_bytes < facet->byte_count) { | |
3366 | facet_learn(facet); | |
3367 | facet_account(facet); | |
3368 | facet->accounted_bytes = facet->byte_count; | |
3369 | } | |
3370 | facet_push_stats(facet); | |
3371 | } | |
3372 | ||
3373 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
3374 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
3375 | static void | |
3376 | delete_unexpected_flow(struct dpif *dpif, | |
3377 | const struct nlattr *key, size_t key_len) | |
3378 | { | |
3379 | if (!VLOG_DROP_WARN(&rl)) { | |
3380 | struct ds s; | |
3381 | ||
3382 | ds_init(&s); | |
3383 | odp_flow_key_format(key, key_len, &s); | |
3384 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
3385 | ds_destroy(&s); | |
3386 | } | |
3387 | ||
3388 | COVERAGE_INC(facet_unexpected); | |
3389 | dpif_flow_del(dpif, key, key_len, NULL); | |
3390 | } | |
3391 | ||
abe529af BP |
3392 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
3393 | * | |
3394 | * This function also pushes statistics updates to rules which each facet | |
3395 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3396 | * facet changes the rule it resubmits into at some time in between | |
3397 | * update_stats() runs, it is possible that statistics accrued to the | |
3398 | * old rule will be incorrectly attributed to the new rule. This could be | |
3399 | * avoided by calling update_stats() whenever rules are created or | |
3400 | * deleted. However, the performance impact of making so many calls to the | |
3401 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
3402 | */ | |
3403 | static void | |
3404 | update_stats(struct ofproto_dpif *p) | |
3405 | { | |
3406 | const struct dpif_flow_stats *stats; | |
3407 | struct dpif_flow_dump dump; | |
3408 | const struct nlattr *key; | |
3409 | size_t key_len; | |
3410 | ||
3411 | dpif_flow_dump_start(&dump, p->dpif); | |
3412 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 3413 | struct subfacet *subfacet; |
abe529af | 3414 | |
6a542738 | 3415 | subfacet = subfacet_find(p, key, key_len); |
6a7e895f BP |
3416 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
3417 | case SF_FAST_PATH: | |
a218c879 | 3418 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
3419 | break; |
3420 | ||
3421 | case SF_SLOW_PATH: | |
3422 | /* Stats are updated per-packet. */ | |
3423 | break; | |
3424 | ||
3425 | case SF_NOT_INSTALLED: | |
3426 | default: | |
a218c879 | 3427 | delete_unexpected_flow(p->dpif, key, key_len); |
6a7e895f | 3428 | break; |
abe529af BP |
3429 | } |
3430 | } | |
3431 | dpif_flow_dump_done(&dump); | |
3432 | } | |
3433 | ||
3434 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3435 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3436 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3437 | * fold its statistic into its rule. */ | |
abe529af | 3438 | static int |
b0f7b9b5 | 3439 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
3440 | { |
3441 | /* | |
3442 | * Idle time histogram. | |
3443 | * | |
b0f7b9b5 BP |
3444 | * Most of the time a switch has a relatively small number of subfacets. |
3445 | * When this is the case we might as well keep statistics for all of them | |
3446 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3447 | * well. |
3448 | * | |
b0f7b9b5 | 3449 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3450 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3451 | * significant. However, with a large number of subfacets it is likely |
3452 | * that only a few of them are "heavy hitters" that consume a large amount | |
3453 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3454 | * the kernel and maintaining in userspaces; other subfacets we can | |
3455 | * discard. | |
abe529af BP |
3456 | * |
3457 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3458 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3459 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3460 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3461 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 3462 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3463 | * the most-recently-used bucket of subfacets is kept, so actually an |
3464 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3465 | * (though the next run will delete most of those unless they receive |
3466 | * additional data). | |
abe529af | 3467 | * |
b0f7b9b5 BP |
3468 | * This requires a second pass through the subfacets, in addition to the |
3469 | * pass made by update_stats(), because the former function never looks at | |
3470 | * uninstallable subfacets. | |
abe529af BP |
3471 | */ |
3472 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3473 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3474 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3475 | int total, subtotal, bucket; |
b0f7b9b5 | 3476 | struct subfacet *subfacet; |
abe529af BP |
3477 | long long int now; |
3478 | int i; | |
3479 | ||
b0f7b9b5 | 3480 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3481 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3482 | return N_BUCKETS * BUCKET_WIDTH; |
3483 | } | |
3484 | ||
3485 | /* Build histogram. */ | |
3486 | now = time_msec(); | |
b0f7b9b5 BP |
3487 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3488 | long long int idle = now - subfacet->used; | |
abe529af BP |
3489 | int bucket = (idle <= 0 ? 0 |
3490 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3491 | : (unsigned int) idle / BUCKET_WIDTH); | |
3492 | buckets[bucket]++; | |
3493 | } | |
3494 | ||
3495 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3496 | subtotal = bucket = 0; |
3497 | do { | |
3498 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3499 | } while (bucket < N_BUCKETS && |
3500 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3501 | |
3502 | if (VLOG_IS_DBG_ENABLED()) { | |
3503 | struct ds s; | |
3504 | ||
3505 | ds_init(&s); | |
3506 | ds_put_cstr(&s, "keep"); | |
3507 | for (i = 0; i < N_BUCKETS; i++) { | |
3508 | if (i == bucket) { | |
3509 | ds_put_cstr(&s, ", drop"); | |
3510 | } | |
3511 | if (buckets[i]) { | |
3512 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3513 | } | |
3514 | } | |
3515 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3516 | ds_destroy(&s); | |
3517 | } | |
3518 | ||
3519 | return bucket * BUCKET_WIDTH; | |
3520 | } | |
3521 | ||
b99d3cee BP |
3522 | enum { EXPIRE_MAX_BATCH = 50 }; |
3523 | ||
3524 | static void | |
3525 | expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) | |
3526 | { | |
3527 | struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; | |
3528 | struct dpif_op ops[EXPIRE_MAX_BATCH]; | |
3529 | struct dpif_op *opsp[EXPIRE_MAX_BATCH]; | |
3530 | struct ofpbuf keys[EXPIRE_MAX_BATCH]; | |
3531 | struct dpif_flow_stats stats[EXPIRE_MAX_BATCH]; | |
3532 | int i; | |
3533 | ||
3534 | for (i = 0; i < n; i++) { | |
3535 | ops[i].type = DPIF_OP_FLOW_DEL; | |
3536 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
3537 | ops[i].u.flow_del.key = keys[i].data; | |
3538 | ops[i].u.flow_del.key_len = keys[i].size; | |
3539 | ops[i].u.flow_del.stats = &stats[i]; | |
3540 | opsp[i] = &ops[i]; | |
3541 | } | |
3542 | ||
3543 | dpif_operate(ofproto->dpif, opsp, n); | |
3544 | for (i = 0; i < n; i++) { | |
3545 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
6a7e895f | 3546 | subfacets[i]->path = SF_NOT_INSTALLED; |
b99d3cee BP |
3547 | subfacet_destroy(subfacets[i]); |
3548 | } | |
3549 | } | |
3550 | ||
abe529af | 3551 | static void |
b0f7b9b5 | 3552 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 3553 | { |
625b0720 BP |
3554 | /* Cutoff time for most flows. */ |
3555 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
3556 | ||
3557 | /* We really want to keep flows for special protocols around, so use a more | |
3558 | * conservative cutoff. */ | |
3559 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 3560 | |
b0f7b9b5 | 3561 | struct subfacet *subfacet, *next_subfacet; |
b99d3cee BP |
3562 | struct subfacet *batch[EXPIRE_MAX_BATCH]; |
3563 | int n_batch; | |
abe529af | 3564 | |
b99d3cee | 3565 | n_batch = 0; |
b0f7b9b5 BP |
3566 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3567 | &ofproto->subfacets) { | |
625b0720 BP |
3568 | long long int cutoff; |
3569 | ||
3570 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) | |
3571 | ? special_cutoff | |
3572 | : normal_cutoff); | |
b0f7b9b5 | 3573 | if (subfacet->used < cutoff) { |
6a7e895f | 3574 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee BP |
3575 | batch[n_batch++] = subfacet; |
3576 | if (n_batch >= EXPIRE_MAX_BATCH) { | |
3577 | expire_batch(ofproto, batch, n_batch); | |
3578 | n_batch = 0; | |
3579 | } | |
3580 | } else { | |
3581 | subfacet_destroy(subfacet); | |
3582 | } | |
abe529af BP |
3583 | } |
3584 | } | |
b99d3cee BP |
3585 | |
3586 | if (n_batch > 0) { | |
3587 | expire_batch(ofproto, batch, n_batch); | |
3588 | } | |
abe529af BP |
3589 | } |
3590 | ||
3591 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3592 | * then delete it entirely. */ | |
3593 | static void | |
3594 | rule_expire(struct rule_dpif *rule) | |
3595 | { | |
abe529af BP |
3596 | struct facet *facet, *next_facet; |
3597 | long long int now; | |
3598 | uint8_t reason; | |
3599 | ||
e2a3d183 BP |
3600 | if (rule->up.pending) { |
3601 | /* We'll have to expire it later. */ | |
3602 | return; | |
3603 | } | |
3604 | ||
abe529af BP |
3605 | /* Has 'rule' expired? */ |
3606 | now = time_msec(); | |
3607 | if (rule->up.hard_timeout | |
308881af | 3608 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 3609 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 3610 | } else if (rule->up.idle_timeout |
1745cd08 | 3611 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
3612 | reason = OFPRR_IDLE_TIMEOUT; |
3613 | } else { | |
3614 | return; | |
3615 | } | |
3616 | ||
3617 | COVERAGE_INC(ofproto_dpif_expired); | |
3618 | ||
3619 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3620 | * timeout, because that only happens when the rule has no facets left.) */ | |
3621 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3622 | facet_remove(facet); |
abe529af BP |
3623 | } |
3624 | ||
3625 | /* Get rid of the rule. */ | |
3626 | ofproto_rule_expire(&rule->up, reason); | |
3627 | } | |
3628 | \f | |
3629 | /* Facets. */ | |
3630 | ||
f3827897 | 3631 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3632 | * |
3633 | * The caller must already have determined that no facet with an identical | |
3634 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3635 | * the ofproto's classifier table. |
3636 | * | |
2b459b83 BP |
3637 | * 'hash' must be the return value of flow_hash(flow, 0). |
3638 | * | |
b0f7b9b5 BP |
3639 | * The facet will initially have no subfacets. The caller should create (at |
3640 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3641 | static struct facet * |
2b459b83 | 3642 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
3643 | { |
3644 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3645 | struct facet *facet; | |
3646 | ||
3647 | facet = xzalloc(sizeof *facet); | |
3648 | facet->used = time_msec(); | |
2b459b83 | 3649 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
3650 | list_push_back(&rule->facets, &facet->list_node); |
3651 | facet->rule = rule; | |
3652 | facet->flow = *flow; | |
b0f7b9b5 | 3653 | list_init(&facet->subfacets); |
abe529af BP |
3654 | netflow_flow_init(&facet->nf_flow); |
3655 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3656 | ||
abe529af BP |
3657 | return facet; |
3658 | } | |
3659 | ||
3660 | static void | |
3661 | facet_free(struct facet *facet) | |
3662 | { | |
abe529af BP |
3663 | free(facet); |
3664 | } | |
3665 | ||
3d9e05f8 BP |
3666 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3667 | * 'packet', which arrived on 'in_port'. | |
3668 | * | |
3669 | * Takes ownership of 'packet'. */ | |
3670 | static bool | |
3671 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3672 | const struct nlattr *odp_actions, size_t actions_len, | |
3673 | struct ofpbuf *packet) | |
3674 | { | |
3675 | struct odputil_keybuf keybuf; | |
3676 | struct ofpbuf key; | |
3677 | int error; | |
3678 | ||
6ff686f2 PS |
3679 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3680 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3681 | |
6ff686f2 PS |
3682 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3683 | odp_actions, actions_len, packet); | |
80e5eed9 | 3684 | |
6ff686f2 PS |
3685 | ofpbuf_delete(packet); |
3686 | return !error; | |
abe529af BP |
3687 | } |
3688 | ||
abe529af BP |
3689 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3690 | * | |
3691 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3692 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3693 | * |
3694 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3695 | */ | |
3696 | static void | |
15baa734 | 3697 | facet_remove(struct facet *facet) |
abe529af | 3698 | { |
15baa734 | 3699 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3700 | struct subfacet *subfacet, *next_subfacet; |
3701 | ||
551a2f6c BP |
3702 | assert(!list_is_empty(&facet->subfacets)); |
3703 | ||
3704 | /* First uninstall all of the subfacets to get final statistics. */ | |
3705 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3706 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3707 | } |
3708 | ||
3709 | /* Flush the final stats to the rule. | |
3710 | * | |
3711 | * This might require us to have at least one subfacet around so that we | |
3712 | * can use its actions for accounting in facet_account(), which is why we | |
3713 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3714 | facet_flush_stats(facet); |
551a2f6c BP |
3715 | |
3716 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3717 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3718 | &facet->subfacets) { | |
15baa734 | 3719 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3720 | } |
abe529af BP |
3721 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3722 | list_remove(&facet->list_node); | |
3723 | facet_free(facet); | |
3724 | } | |
3725 | ||
3de9590b BP |
3726 | /* Feed information from 'facet' back into the learning table to keep it in |
3727 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3728 | static void |
3de9590b | 3729 | facet_learn(struct facet *facet) |
abe529af | 3730 | { |
15baa734 | 3731 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3de9590b | 3732 | struct action_xlate_ctx ctx; |
abe529af | 3733 | |
3de9590b BP |
3734 | if (!facet->has_learn |
3735 | && !facet->has_normal | |
3736 | && (!facet->has_fin_timeout | |
3737 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
3738 | return; |
3739 | } | |
abe529af | 3740 | |
3de9590b BP |
3741 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
3742 | facet->flow.vlan_tci, | |
3743 | facet->rule, facet->tcp_flags, NULL); | |
3744 | ctx.may_learn = true; | |
f25d0cf3 BP |
3745 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, |
3746 | facet->rule->up.ofpacts_len); | |
3de9590b BP |
3747 | } |
3748 | ||
3749 | static void | |
3750 | facet_account(struct facet *facet) | |
3751 | { | |
3752 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3753 | struct subfacet *subfacet; | |
3754 | const struct nlattr *a; | |
3755 | unsigned int left; | |
3756 | ovs_be16 vlan_tci; | |
3757 | uint64_t n_bytes; | |
abe529af | 3758 | |
75a75043 | 3759 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3760 | return; |
3761 | } | |
3de9590b | 3762 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3763 | |
3764 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3765 | * as a basis. We also need to track the actual VLAN on which the packet | |
3766 | * is going to be sent to ensure that it matches the one passed to | |
3767 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3768 | * hash bucket.) |
3769 | * | |
3770 | * We use the actions from an arbitrary subfacet because they should all | |
3771 | * be equally valid for our purpose. */ | |
3772 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3773 | struct subfacet, list_node); | |
d78be13b | 3774 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3775 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3776 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3777 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3778 | struct ofport_dpif *port; |
abe529af | 3779 | |
d78be13b | 3780 | switch (nl_attr_type(a)) { |
df2c07f4 | 3781 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3782 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3783 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3784 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3785 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3786 | } |
d78be13b BP |
3787 | break; |
3788 | ||
fea393b1 BP |
3789 | case OVS_ACTION_ATTR_POP_VLAN: |
3790 | vlan_tci = htons(0); | |
d78be13b BP |
3791 | break; |
3792 | ||
fea393b1 BP |
3793 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3794 | vlan = nl_attr_get(a); | |
3795 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3796 | break; |
abe529af BP |
3797 | } |
3798 | } | |
3799 | } | |
3800 | ||
abe529af BP |
3801 | /* Returns true if the only action for 'facet' is to send to the controller. |
3802 | * (We don't report NetFlow expiration messages for such facets because they | |
3803 | * are just part of the control logic for the network, not real traffic). */ | |
3804 | static bool | |
3805 | facet_is_controller_flow(struct facet *facet) | |
3806 | { | |
f25d0cf3 BP |
3807 | if (facet) { |
3808 | const struct rule *rule = &facet->rule->up; | |
3809 | const struct ofpact *ofpacts = rule->ofpacts; | |
3810 | size_t ofpacts_len = rule->ofpacts_len; | |
3811 | ||
dd30ff28 BP |
3812 | if (ofpacts_len > 0 && |
3813 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
3814 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
3815 | return true; | |
3816 | } | |
3817 | } | |
3818 | return false; | |
abe529af BP |
3819 | } |
3820 | ||
3821 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3822 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3823 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3824 | * its packet and byte counts before this function is called. */ | |
3825 | static void | |
15baa734 | 3826 | facet_flush_stats(struct facet *facet) |
abe529af | 3827 | { |
15baa734 | 3828 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3829 | struct subfacet *subfacet; |
3830 | ||
3831 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3832 | assert(!subfacet->dp_byte_count); | |
3833 | assert(!subfacet->dp_packet_count); | |
3834 | } | |
abe529af BP |
3835 | |
3836 | facet_push_stats(facet); | |
3de9590b BP |
3837 | if (facet->accounted_bytes < facet->byte_count) { |
3838 | facet_account(facet); | |
3839 | facet->accounted_bytes = facet->byte_count; | |
3840 | } | |
abe529af BP |
3841 | |
3842 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3843 | struct ofexpired expired; | |
3844 | expired.flow = facet->flow; | |
3845 | expired.packet_count = facet->packet_count; | |
3846 | expired.byte_count = facet->byte_count; | |
3847 | expired.used = facet->used; | |
3848 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3849 | } | |
3850 | ||
3851 | facet->rule->packet_count += facet->packet_count; | |
3852 | facet->rule->byte_count += facet->byte_count; | |
3853 | ||
3854 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3855 | * reinstalled. */ | |
bbb5d219 | 3856 | facet_reset_counters(facet); |
abe529af BP |
3857 | |
3858 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 3859 | facet->tcp_flags = 0; |
abe529af BP |
3860 | } |
3861 | ||
3862 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3863 | * Returns it if found, otherwise a null pointer. | |
3864 | * | |
2b459b83 BP |
3865 | * 'hash' must be the return value of flow_hash(flow, 0). |
3866 | * | |
abe529af BP |
3867 | * The returned facet might need revalidation; use facet_lookup_valid() |
3868 | * instead if that is important. */ | |
3869 | static struct facet * | |
2b459b83 BP |
3870 | facet_find(struct ofproto_dpif *ofproto, |
3871 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
3872 | { |
3873 | struct facet *facet; | |
3874 | ||
2b459b83 | 3875 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
3876 | if (flow_equal(flow, &facet->flow)) { |
3877 | return facet; | |
3878 | } | |
3879 | } | |
3880 | ||
3881 | return NULL; | |
3882 | } | |
3883 | ||
3884 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3885 | * Returns it if found, otherwise a null pointer. | |
3886 | * | |
2b459b83 BP |
3887 | * 'hash' must be the return value of flow_hash(flow, 0). |
3888 | * | |
abe529af BP |
3889 | * The returned facet is guaranteed to be valid. */ |
3890 | static struct facet * | |
2b459b83 BP |
3891 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
3892 | uint32_t hash) | |
abe529af | 3893 | { |
c57b2226 | 3894 | struct facet *facet; |
abe529af | 3895 | |
c57b2226 | 3896 | facet = facet_find(ofproto, flow, hash); |
abe529af | 3897 | if (facet |
0e4b3771 | 3898 | && (ofproto->need_revalidate |
c57b2226 BP |
3899 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { |
3900 | facet_revalidate(facet); | |
abe529af BP |
3901 | } |
3902 | ||
3903 | return facet; | |
3904 | } | |
3905 | ||
6a7e895f BP |
3906 | static const char * |
3907 | subfacet_path_to_string(enum subfacet_path path) | |
3908 | { | |
3909 | switch (path) { | |
3910 | case SF_NOT_INSTALLED: | |
3911 | return "not installed"; | |
3912 | case SF_FAST_PATH: | |
3913 | return "in fast path"; | |
3914 | case SF_SLOW_PATH: | |
3915 | return "in slow path"; | |
3916 | default: | |
3917 | return "<error>"; | |
3918 | } | |
3919 | } | |
3920 | ||
3921 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
3922 | * member has the specified value. */ | |
3923 | static enum subfacet_path | |
3924 | subfacet_want_path(enum slow_path_reason slow) | |
3925 | { | |
3926 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
3927 | } | |
3928 | ||
3929 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
3930 | * supposing that its actions have been recalculated as 'want_actions' and that | |
3931 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
3932 | static bool | |
3933 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
3934 | const struct ofpbuf *want_actions) | |
3935 | { | |
3936 | enum subfacet_path want_path = subfacet_want_path(slow); | |
3937 | return (want_path != subfacet->path | |
3938 | || (want_path == SF_FAST_PATH | |
3939 | && (subfacet->actions_len != want_actions->size | |
3940 | || memcmp(subfacet->actions, want_actions->data, | |
3941 | subfacet->actions_len)))); | |
3942 | } | |
3943 | ||
6814e51f BP |
3944 | static bool |
3945 | facet_check_consistency(struct facet *facet) | |
3946 | { | |
3947 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3948 | ||
3949 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3950 | ||
050ac423 BP |
3951 | uint64_t odp_actions_stub[1024 / 8]; |
3952 | struct ofpbuf odp_actions; | |
3953 | ||
6814e51f BP |
3954 | struct rule_dpif *rule; |
3955 | struct subfacet *subfacet; | |
c53e1132 | 3956 | bool may_log = false; |
6814e51f BP |
3957 | bool ok; |
3958 | ||
3959 | /* Check the rule for consistency. */ | |
c57b2226 BP |
3960 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
3961 | ok = rule == facet->rule; | |
3962 | if (!ok) { | |
c53e1132 | 3963 | may_log = !VLOG_DROP_WARN(&rl); |
c53e1132 BP |
3964 | if (may_log) { |
3965 | struct ds s; | |
6814e51f | 3966 | |
c53e1132 BP |
3967 | ds_init(&s); |
3968 | flow_format(&s, &facet->flow); | |
3969 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3970 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3971 | cls_rule_format(&facet->rule->up.cr, &s); | |
3972 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3973 | rule->up.table_id); | |
3974 | cls_rule_format(&rule->up.cr, &s); | |
3975 | ds_put_char(&s, ')'); | |
6814e51f | 3976 | |
c53e1132 BP |
3977 | VLOG_WARN("%s", ds_cstr(&s)); |
3978 | ds_destroy(&s); | |
3979 | } | |
6814e51f BP |
3980 | } |
3981 | ||
3982 | /* Check the datapath actions for consistency. */ | |
050ac423 | 3983 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f | 3984 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 3985 | enum subfacet_path want_path; |
9616614b | 3986 | struct odputil_keybuf keybuf; |
6814e51f | 3987 | struct action_xlate_ctx ctx; |
9616614b BP |
3988 | struct ofpbuf key; |
3989 | struct ds s; | |
6814e51f BP |
3990 | |
3991 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
0e553d9c | 3992 | subfacet->initial_tci, rule, 0, NULL); |
f25d0cf3 | 3993 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 3994 | &odp_actions); |
6814e51f | 3995 | |
6a7e895f BP |
3996 | if (subfacet->path == SF_NOT_INSTALLED) { |
3997 | /* This only happens if the datapath reported an error when we | |
3998 | * tried to install the flow. Don't flag another error here. */ | |
3999 | continue; | |
4000 | } | |
4001 | ||
4002 | want_path = subfacet_want_path(subfacet->slow); | |
4003 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
4004 | /* The actions for slow-path flows may legitimately vary from one | |
4005 | * packet to the next. We're done. */ | |
050ac423 | 4006 | continue; |
6814e51f BP |
4007 | } |
4008 | ||
6a7e895f | 4009 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { |
9616614b BP |
4010 | continue; |
4011 | } | |
c53e1132 | 4012 | |
9616614b BP |
4013 | /* Inconsistency! */ |
4014 | if (ok) { | |
4015 | may_log = !VLOG_DROP_WARN(&rl); | |
4016 | ok = false; | |
4017 | } | |
4018 | if (!may_log) { | |
4019 | /* Rate-limited, skip reporting. */ | |
4020 | continue; | |
4021 | } | |
c53e1132 | 4022 | |
9616614b BP |
4023 | ds_init(&s); |
4024 | subfacet_get_key(subfacet, &keybuf, &key); | |
4025 | odp_flow_key_format(key.data, key.size, &s); | |
4026 | ||
4027 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
6a7e895f | 4028 | if (want_path != subfacet->path) { |
9616614b BP |
4029 | enum odp_key_fitness fitness = subfacet->key_fitness; |
4030 | ||
6a7e895f BP |
4031 | ds_put_format(&s, " (%s, fitness=%s)", |
4032 | subfacet_path_to_string(subfacet->path), | |
9616614b | 4033 | odp_key_fitness_to_string(fitness)); |
6a7e895f BP |
4034 | ds_put_format(&s, " (should have been %s)", |
4035 | subfacet_path_to_string(want_path)); | |
4036 | } else if (want_path == SF_FAST_PATH) { | |
9616614b BP |
4037 | ds_put_cstr(&s, " (actions were: "); |
4038 | format_odp_actions(&s, subfacet->actions, | |
4039 | subfacet->actions_len); | |
4040 | ds_put_cstr(&s, ") (correct actions: "); | |
4041 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
4042 | ds_put_char(&s, ')'); | |
4043 | } else { | |
4044 | ds_put_cstr(&s, " (actions: "); | |
4045 | format_odp_actions(&s, subfacet->actions, | |
4046 | subfacet->actions_len); | |
4047 | ds_put_char(&s, ')'); | |
6814e51f | 4048 | } |
9616614b BP |
4049 | VLOG_WARN("%s", ds_cstr(&s)); |
4050 | ds_destroy(&s); | |
6814e51f | 4051 | } |
050ac423 | 4052 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
4053 | |
4054 | return ok; | |
4055 | } | |
4056 | ||
15baa734 | 4057 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4058 | * |
4059 | * - If the rule found is different from 'facet''s current rule, moves | |
4060 | * 'facet' to the new rule and recompiles its actions. | |
4061 | * | |
4062 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
c57b2226 BP |
4063 | * where it is and recompiles its actions anyway. */ |
4064 | static void | |
15baa734 | 4065 | facet_revalidate(struct facet *facet) |
abe529af | 4066 | { |
15baa734 | 4067 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
4068 | struct actions { |
4069 | struct nlattr *odp_actions; | |
4070 | size_t actions_len; | |
4071 | }; | |
4072 | struct actions *new_actions; | |
4073 | ||
abe529af | 4074 | struct action_xlate_ctx ctx; |
050ac423 BP |
4075 | uint64_t odp_actions_stub[1024 / 8]; |
4076 | struct ofpbuf odp_actions; | |
4077 | ||
abe529af | 4078 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4079 | struct subfacet *subfacet; |
b95fc6ba | 4080 | int i; |
abe529af BP |
4081 | |
4082 | COVERAGE_INC(facet_revalidate); | |
4083 | ||
c57b2226 | 4084 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 4085 | |
df2c07f4 | 4086 | /* Calculate new datapath actions. |
abe529af BP |
4087 | * |
4088 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4089 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4090 | * around to properly compose it. */ | |
abe529af | 4091 | |
df2c07f4 JP |
4092 | /* If the datapath actions changed or the installability changed, |
4093 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
4094 | i = 0; |
4095 | new_actions = NULL; | |
4096 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 4097 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 4098 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4099 | enum slow_path_reason slow; |
b95fc6ba | 4100 | |
e84173dc | 4101 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
0e553d9c | 4102 | subfacet->initial_tci, new_rule, 0, NULL); |
f25d0cf3 | 4103 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, |
050ac423 | 4104 | &odp_actions); |
b0f7b9b5 | 4105 | |
6a7e895f BP |
4106 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
4107 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
4108 | struct dpif_flow_stats stats; | |
4109 | ||
4110 | subfacet_install(subfacet, | |
4111 | odp_actions.data, odp_actions.size, &stats, slow); | |
4112 | subfacet_update_stats(subfacet, &stats); | |
b95fc6ba BP |
4113 | |
4114 | if (!new_actions) { | |
4115 | new_actions = xcalloc(list_size(&facet->subfacets), | |
4116 | sizeof *new_actions); | |
4117 | } | |
050ac423 BP |
4118 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
4119 | odp_actions.size); | |
4120 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 4121 | } |
b95fc6ba | 4122 | |
b95fc6ba | 4123 | i++; |
b0f7b9b5 | 4124 | } |
050ac423 BP |
4125 | ofpbuf_uninit(&odp_actions); |
4126 | ||
b95fc6ba | 4127 | if (new_actions) { |
15baa734 | 4128 | facet_flush_stats(facet); |
abe529af BP |
4129 | } |
4130 | ||
4131 | /* Update 'facet' now that we've taken care of all the old state. */ | |
4132 | facet->tags = ctx.tags; | |
4133 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
75a75043 BP |
4134 | facet->has_learn = ctx.has_learn; |
4135 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4136 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 4137 | facet->mirrors = ctx.mirrors; |
6a7e895f BP |
4138 | |
4139 | i = 0; | |
4140 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4141 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4142 | ||
4143 | if (new_actions && new_actions[i].odp_actions) { | |
4144 | free(subfacet->actions); | |
4145 | subfacet->actions = new_actions[i].odp_actions; | |
4146 | subfacet->actions_len = new_actions[i].actions_len; | |
b95fc6ba | 4147 | } |
6a7e895f | 4148 | i++; |
abe529af | 4149 | } |
6a7e895f BP |
4150 | free(new_actions); |
4151 | ||
abe529af BP |
4152 | if (facet->rule != new_rule) { |
4153 | COVERAGE_INC(facet_changed_rule); | |
4154 | list_remove(&facet->list_node); | |
4155 | list_push_back(&new_rule->facets, &facet->list_node); | |
4156 | facet->rule = new_rule; | |
4157 | facet->used = new_rule->up.created; | |
9d24de3b | 4158 | facet->prev_used = facet->used; |
abe529af | 4159 | } |
abe529af BP |
4160 | } |
4161 | ||
4162 | /* Updates 'facet''s used time. Caller is responsible for calling | |
4163 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
4164 | static void | |
15baa734 | 4165 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 4166 | { |
15baa734 | 4167 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
4168 | if (used > facet->used) { |
4169 | facet->used = used; | |
1745cd08 | 4170 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
4171 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
4172 | } | |
4173 | } | |
4174 | ||
bbb5d219 EJ |
4175 | static void |
4176 | facet_reset_counters(struct facet *facet) | |
4177 | { | |
4178 | facet->packet_count = 0; | |
4179 | facet->byte_count = 0; | |
9d24de3b JP |
4180 | facet->prev_packet_count = 0; |
4181 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4182 | facet->accounted_bytes = 0; |
4183 | } | |
4184 | ||
abe529af BP |
4185 | static void |
4186 | facet_push_stats(struct facet *facet) | |
4187 | { | |
112bc5f4 | 4188 | struct dpif_flow_stats stats; |
abe529af | 4189 | |
9d24de3b JP |
4190 | assert(facet->packet_count >= facet->prev_packet_count); |
4191 | assert(facet->byte_count >= facet->prev_byte_count); | |
4192 | assert(facet->used >= facet->prev_used); | |
abe529af | 4193 | |
112bc5f4 BP |
4194 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4195 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4196 | stats.used = facet->used; | |
4197 | stats.tcp_flags = 0; | |
abe529af | 4198 | |
112bc5f4 | 4199 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
4200 | facet->prev_packet_count = facet->packet_count; |
4201 | facet->prev_byte_count = facet->byte_count; | |
4202 | facet->prev_used = facet->used; | |
abe529af | 4203 | |
112bc5f4 | 4204 | flow_push_stats(facet->rule, &facet->flow, &stats); |
9d24de3b JP |
4205 | |
4206 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 4207 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
4208 | } |
4209 | } | |
4210 | ||
abe529af | 4211 | static void |
112bc5f4 | 4212 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4213 | { |
112bc5f4 BP |
4214 | rule->packet_count += stats->n_packets; |
4215 | rule->byte_count += stats->n_bytes; | |
4216 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
4217 | } |
4218 | ||
4219 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 4220 | * 'rule''s actions and mirrors. */ |
abe529af | 4221 | static void |
18b2a258 | 4222 | flow_push_stats(struct rule_dpif *rule, |
112bc5f4 | 4223 | const struct flow *flow, const struct dpif_flow_stats *stats) |
abe529af BP |
4224 | { |
4225 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
112bc5f4 | 4226 | struct action_xlate_ctx ctx; |
abe529af | 4227 | |
112bc5f4 | 4228 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 4229 | |
112bc5f4 | 4230 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, |
0e553d9c | 4231 | 0, NULL); |
112bc5f4 | 4232 | ctx.resubmit_stats = stats; |
f25d0cf3 BP |
4233 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, |
4234 | rule->up.ofpacts_len); | |
abe529af BP |
4235 | } |
4236 | \f | |
b0f7b9b5 BP |
4237 | /* Subfacets. */ |
4238 | ||
4239 | static struct subfacet * | |
4240 | subfacet_find__(struct ofproto_dpif *ofproto, | |
4241 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
4242 | const struct flow *flow) | |
4243 | { | |
4244 | struct subfacet *subfacet; | |
4245 | ||
4246 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
4247 | &ofproto->subfacets) { | |
4248 | if (subfacet->key | |
4249 | ? (subfacet->key_len == key_len | |
4250 | && !memcmp(key, subfacet->key, key_len)) | |
4251 | : flow_equal(flow, &subfacet->facet->flow)) { | |
4252 | return subfacet; | |
4253 | } | |
4254 | } | |
4255 | ||
4256 | return NULL; | |
4257 | } | |
4258 | ||
4259 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
4260 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
4261 | * there is one, otherwise creates and returns a new subfacet. |
4262 | * | |
4263 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
4264 | * which case the caller must populate the actions with | |
4265 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 4266 | static struct subfacet * |
15baa734 | 4267 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
459b16a1 BP |
4268 | const struct nlattr *key, size_t key_len, |
4269 | ovs_be16 initial_tci, long long int now) | |
b0f7b9b5 | 4270 | { |
15baa734 | 4271 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4272 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
4273 | struct subfacet *subfacet; | |
4274 | ||
3b145dd7 BP |
4275 | if (list_is_empty(&facet->subfacets)) { |
4276 | subfacet = &facet->one_subfacet; | |
4277 | } else { | |
4278 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, | |
4279 | &facet->flow); | |
4280 | if (subfacet) { | |
4281 | if (subfacet->facet == facet) { | |
4282 | return subfacet; | |
4283 | } | |
4284 | ||
4285 | /* This shouldn't happen. */ | |
4286 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
4287 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
4288 | } |
4289 | ||
3b145dd7 | 4290 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
4291 | } |
4292 | ||
b0f7b9b5 BP |
4293 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
4294 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
4295 | subfacet->facet = facet; | |
b0f7b9b5 BP |
4296 | subfacet->key_fitness = key_fitness; |
4297 | if (key_fitness != ODP_FIT_PERFECT) { | |
4298 | subfacet->key = xmemdup(key, key_len); | |
4299 | subfacet->key_len = key_len; | |
26cd7e34 BP |
4300 | } else { |
4301 | subfacet->key = NULL; | |
4302 | subfacet->key_len = 0; | |
b0f7b9b5 | 4303 | } |
459b16a1 | 4304 | subfacet->used = now; |
26cd7e34 BP |
4305 | subfacet->dp_packet_count = 0; |
4306 | subfacet->dp_byte_count = 0; | |
4307 | subfacet->actions_len = 0; | |
4308 | subfacet->actions = NULL; | |
6a7e895f BP |
4309 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE |
4310 | ? SLOW_MATCH | |
4311 | : 0); | |
4312 | subfacet->path = SF_NOT_INSTALLED; | |
e84173dc | 4313 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
4314 | |
4315 | return subfacet; | |
4316 | } | |
4317 | ||
4318 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
4319 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
4320 | static struct subfacet * | |
4321 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 4322 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
4323 | { |
4324 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
4325 | enum odp_key_fitness fitness; |
4326 | struct flow flow; | |
4327 | ||
4328 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
4329 | if (fitness == ODP_FIT_ERROR) { | |
4330 | return NULL; | |
4331 | } | |
b0f7b9b5 | 4332 | |
6a542738 | 4333 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
4334 | } |
4335 | ||
4336 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
4337 | * its facet within 'ofproto', and frees it. */ | |
4338 | static void | |
15baa734 | 4339 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4340 | { |
15baa734 BP |
4341 | struct facet *facet = subfacet->facet; |
4342 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4343 | ||
4344 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
4345 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
4346 | list_remove(&subfacet->list_node); | |
4347 | free(subfacet->key); | |
b95fc6ba | 4348 | free(subfacet->actions); |
26cd7e34 BP |
4349 | if (subfacet != &facet->one_subfacet) { |
4350 | free(subfacet); | |
4351 | } | |
b0f7b9b5 BP |
4352 | } |
4353 | ||
4354 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4355 | * last remaining subfacet in its facet destroys the facet too. */ | |
4356 | static void | |
15baa734 | 4357 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4358 | { |
4359 | struct facet *facet = subfacet->facet; | |
4360 | ||
551a2f6c BP |
4361 | if (list_is_singleton(&facet->subfacets)) { |
4362 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4363 | facet_remove(facet); |
551a2f6c | 4364 | } else { |
15baa734 | 4365 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4366 | } |
4367 | } | |
4368 | ||
4369 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
4370 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
4371 | * for use as temporary storage. */ | |
4372 | static void | |
4373 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
4374 | struct ofpbuf *key) | |
4375 | { | |
4376 | if (!subfacet->key) { | |
4377 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
4378 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
4379 | } else { | |
4380 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
4381 | } | |
4382 | } | |
4383 | ||
5fe20d5d BP |
4384 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
4385 | * Translates the actions into 'odp_actions', which the caller must have | |
4386 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 4387 | static void |
5fe20d5d BP |
4388 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
4389 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
4390 | { |
4391 | struct facet *facet = subfacet->facet; | |
18b2a258 | 4392 | struct rule_dpif *rule = facet->rule; |
15baa734 | 4393 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 4394 | |
b95fc6ba BP |
4395 | struct action_xlate_ctx ctx; |
4396 | ||
15baa734 | 4397 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
0e553d9c | 4398 | rule, 0, packet); |
f25d0cf3 | 4399 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); |
b95fc6ba | 4400 | facet->tags = ctx.tags; |
b95fc6ba BP |
4401 | facet->has_learn = ctx.has_learn; |
4402 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4403 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 4404 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 4405 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 4406 | |
6a7e895f | 4407 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5fe20d5d BP |
4408 | if (subfacet->actions_len != odp_actions->size |
4409 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 4410 | free(subfacet->actions); |
5fe20d5d BP |
4411 | subfacet->actions_len = odp_actions->size; |
4412 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 4413 | } |
b95fc6ba BP |
4414 | } |
4415 | ||
b0f7b9b5 BP |
4416 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4417 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4418 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4419 | * since 'subfacet' was last updated. | |
4420 | * | |
4421 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4422 | static int | |
15baa734 | 4423 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 | 4424 | const struct nlattr *actions, size_t actions_len, |
6a7e895f BP |
4425 | struct dpif_flow_stats *stats, |
4426 | enum slow_path_reason slow) | |
b0f7b9b5 | 4427 | { |
15baa734 BP |
4428 | struct facet *facet = subfacet->facet; |
4429 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
6a7e895f BP |
4430 | enum subfacet_path path = subfacet_want_path(slow); |
4431 | uint64_t slow_path_stub[128 / 8]; | |
b0f7b9b5 BP |
4432 | struct odputil_keybuf keybuf; |
4433 | enum dpif_flow_put_flags flags; | |
4434 | struct ofpbuf key; | |
4435 | int ret; | |
4436 | ||
4437 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
4438 | if (stats) { | |
4439 | flags |= DPIF_FP_ZERO_STATS; | |
4440 | } | |
4441 | ||
6a7e895f BP |
4442 | if (path == SF_SLOW_PATH) { |
4443 | compose_slow_path(ofproto, &facet->flow, slow, | |
4444 | slow_path_stub, sizeof slow_path_stub, | |
4445 | &actions, &actions_len); | |
4446 | } | |
4447 | ||
b0f7b9b5 BP |
4448 | subfacet_get_key(subfacet, &keybuf, &key); |
4449 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
4450 | actions, actions_len, stats); | |
4451 | ||
4452 | if (stats) { | |
4453 | subfacet_reset_dp_stats(subfacet, stats); | |
4454 | } | |
4455 | ||
6a7e895f BP |
4456 | if (!ret) { |
4457 | subfacet->path = path; | |
4458 | } | |
b0f7b9b5 BP |
4459 | return ret; |
4460 | } | |
4461 | ||
6a7e895f BP |
4462 | static int |
4463 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
4464 | { | |
4465 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
4466 | stats, subfacet->slow); | |
4467 | } | |
4468 | ||
b0f7b9b5 BP |
4469 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ |
4470 | static void | |
15baa734 | 4471 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 4472 | { |
6a7e895f | 4473 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
4474 | struct rule_dpif *rule = subfacet->facet->rule; |
4475 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
4476 | struct odputil_keybuf keybuf; |
4477 | struct dpif_flow_stats stats; | |
4478 | struct ofpbuf key; | |
4479 | int error; | |
4480 | ||
4481 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 4482 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
4483 | subfacet_reset_dp_stats(subfacet, &stats); |
4484 | if (!error) { | |
15baa734 | 4485 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 4486 | } |
6a7e895f | 4487 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 BP |
4488 | } else { |
4489 | assert(subfacet->dp_packet_count == 0); | |
4490 | assert(subfacet->dp_byte_count == 0); | |
4491 | } | |
4492 | } | |
4493 | ||
4494 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4495 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4496 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4497 | * was reset in the datapath. 'stats' will be modified to include only | |
4498 | * statistics new since 'subfacet' was last updated. */ | |
4499 | static void | |
4500 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4501 | struct dpif_flow_stats *stats) | |
4502 | { | |
4503 | if (stats | |
4504 | && subfacet->dp_packet_count <= stats->n_packets | |
4505 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4506 | stats->n_packets -= subfacet->dp_packet_count; | |
4507 | stats->n_bytes -= subfacet->dp_byte_count; | |
4508 | } | |
4509 | ||
4510 | subfacet->dp_packet_count = 0; | |
4511 | subfacet->dp_byte_count = 0; | |
4512 | } | |
4513 | ||
4514 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
4515 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
4516 | static void | |
15baa734 | 4517 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
4518 | { |
4519 | if (used > subfacet->used) { | |
4520 | subfacet->used = used; | |
15baa734 | 4521 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
4522 | } |
4523 | } | |
4524 | ||
4525 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
4526 | * | |
4527 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4528 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4529 | * represents a packet that was sent by hand or if it represents statistics | |
4530 | * that have been cleared out of the datapath. */ | |
4531 | static void | |
15baa734 | 4532 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4533 | const struct dpif_flow_stats *stats) |
4534 | { | |
4535 | if (stats->n_packets || stats->used > subfacet->used) { | |
4536 | struct facet *facet = subfacet->facet; | |
4537 | ||
15baa734 | 4538 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
4539 | facet->packet_count += stats->n_packets; |
4540 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4541 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4542 | facet_push_stats(facet); |
4543 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
4544 | } | |
4545 | } | |
4546 | \f | |
abe529af BP |
4547 | /* Rules. */ |
4548 | ||
4549 | static struct rule_dpif * | |
c57b2226 BP |
4550 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
4551 | { | |
c57b2226 BP |
4552 | struct rule_dpif *rule; |
4553 | ||
4554 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
4555 | if (rule) { | |
4556 | return rule; | |
4557 | } | |
4558 | ||
c376f9a3 | 4559 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
4560 | } |
4561 | ||
4562 | static struct rule_dpif * | |
4563 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4564 | uint8_t table_id) | |
abe529af | 4565 | { |
7257b535 BP |
4566 | struct cls_rule *cls_rule; |
4567 | struct classifier *cls; | |
4568 | ||
9cdaaebe BP |
4569 | if (table_id >= N_TABLES) { |
4570 | return NULL; | |
4571 | } | |
4572 | ||
d0918789 | 4573 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 4574 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
4575 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
4576 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
4577 | * are unavailable. */ | |
4578 | struct flow ofpc_normal_flow = *flow; | |
4579 | ofpc_normal_flow.tp_src = htons(0); | |
4580 | ofpc_normal_flow.tp_dst = htons(0); | |
4581 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
4582 | } else { | |
4583 | cls_rule = classifier_lookup(cls, flow); | |
4584 | } | |
4585 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
4586 | } |
4587 | ||
c376f9a3 IY |
4588 | static struct rule_dpif * |
4589 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
4590 | { | |
4591 | struct ofport_dpif *port; | |
4592 | ||
4593 | port = get_ofp_port(ofproto, flow->in_port); | |
4594 | if (!port) { | |
4595 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
4596 | return ofproto->miss_rule; | |
4597 | } | |
4598 | ||
4599 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
4600 | return ofproto->no_packet_in_rule; | |
4601 | } | |
4602 | return ofproto->miss_rule; | |
4603 | } | |
4604 | ||
7ee20df1 BP |
4605 | static void |
4606 | complete_operation(struct rule_dpif *rule) | |
4607 | { | |
4608 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4609 | ||
54a9cbc9 | 4610 | rule_invalidate(rule); |
7ee20df1 BP |
4611 | if (clogged) { |
4612 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4613 | c->op = rule->up.pending; | |
4614 | list_push_back(&ofproto->completions, &c->list_node); | |
4615 | } else { | |
4616 | ofoperation_complete(rule->up.pending, 0); | |
4617 | } | |
4618 | } | |
4619 | ||
abe529af BP |
4620 | static struct rule * |
4621 | rule_alloc(void) | |
4622 | { | |
4623 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4624 | return &rule->up; | |
4625 | } | |
4626 | ||
4627 | static void | |
4628 | rule_dealloc(struct rule *rule_) | |
4629 | { | |
4630 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4631 | free(rule); | |
4632 | } | |
4633 | ||
90bf1e07 | 4634 | static enum ofperr |
abe529af BP |
4635 | rule_construct(struct rule *rule_) |
4636 | { | |
4637 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4638 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4639 | struct rule_dpif *victim; |
54a9cbc9 | 4640 | uint8_t table_id; |
abe529af | 4641 | |
abe529af BP |
4642 | rule->packet_count = 0; |
4643 | rule->byte_count = 0; | |
abe529af | 4644 | |
7ee20df1 BP |
4645 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4646 | if (victim && !list_is_empty(&victim->facets)) { | |
4647 | struct facet *facet; | |
4648 | ||
4649 | rule->facets = victim->facets; | |
4650 | list_moved(&rule->facets); | |
4651 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4652 | /* XXX: We're only clearing our local counters here. It's possible |
4653 | * that quite a few packets are unaccounted for in the datapath | |
4654 | * statistics. These will be accounted to the new rule instead of | |
4655 | * cleared as required. This could be fixed by clearing out the | |
4656 | * datapath statistics for this facet, but currently it doesn't | |
4657 | * seem worth it. */ | |
4658 | facet_reset_counters(facet); | |
7ee20df1 BP |
4659 | facet->rule = rule; |
4660 | } | |
4661 | } else { | |
4662 | /* Must avoid list_moved() in this case. */ | |
4663 | list_init(&rule->facets); | |
4664 | } | |
abe529af | 4665 | |
54a9cbc9 BP |
4666 | table_id = rule->up.table_id; |
4667 | rule->tag = (victim ? victim->tag | |
4668 | : table_id == 0 ? 0 | |
4669 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4670 | ofproto->tables[table_id].basis)); | |
4671 | ||
7ee20df1 | 4672 | complete_operation(rule); |
abe529af BP |
4673 | return 0; |
4674 | } | |
4675 | ||
4676 | static void | |
4677 | rule_destruct(struct rule *rule_) | |
4678 | { | |
4679 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4680 | struct facet *facet, *next_facet; |
4681 | ||
abe529af | 4682 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4683 | facet_revalidate(facet); |
abe529af | 4684 | } |
7ee20df1 BP |
4685 | |
4686 | complete_operation(rule); | |
abe529af BP |
4687 | } |
4688 | ||
4689 | static void | |
4690 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4691 | { | |
4692 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4693 | struct facet *facet; | |
4694 | ||
4695 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4696 | * in facets. This counts, for example, facets that have expired. */ | |
4697 | *packets = rule->packet_count; | |
4698 | *bytes = rule->byte_count; | |
4699 | ||
4700 | /* Add any statistics that are tracked by facets. This includes | |
4701 | * statistical data recently updated by ofproto_update_stats() as well as | |
4702 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4703 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4704 | *packets += facet->packet_count; | |
4705 | *bytes += facet->byte_count; | |
4706 | } | |
4707 | } | |
4708 | ||
90bf1e07 | 4709 | static enum ofperr |
59d0f2c8 BP |
4710 | rule_execute(struct rule *rule_, const struct flow *flow, |
4711 | struct ofpbuf *packet) | |
abe529af BP |
4712 | { |
4713 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4714 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
050ac423 | 4715 | |
112bc5f4 | 4716 | struct dpif_flow_stats stats; |
050ac423 | 4717 | |
abe529af | 4718 | struct action_xlate_ctx ctx; |
050ac423 BP |
4719 | uint64_t odp_actions_stub[1024 / 8]; |
4720 | struct ofpbuf odp_actions; | |
abe529af | 4721 | |
a7752d4a | 4722 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
4723 | rule_credit_stats(rule, &stats); |
4724 | ||
050ac423 | 4725 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
54834960 | 4726 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
112bc5f4 BP |
4727 | rule, stats.tcp_flags, packet); |
4728 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 4729 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); |
112bc5f4 BP |
4730 | |
4731 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
4732 | odp_actions.size, packet); | |
4733 | ||
050ac423 | 4734 | ofpbuf_uninit(&odp_actions); |
5bf0e941 BP |
4735 | |
4736 | return 0; | |
abe529af BP |
4737 | } |
4738 | ||
7ee20df1 BP |
4739 | static void |
4740 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4741 | { |
4742 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
4743 | |
4744 | complete_operation(rule); | |
abe529af BP |
4745 | } |
4746 | \f | |
97d6520b | 4747 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4748 | * May modify 'packet'. |
abe529af BP |
4749 | * Returns 0 if successful, otherwise a positive errno value. */ |
4750 | static int | |
52a90c29 | 4751 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4752 | { |
97d6520b | 4753 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4754 | struct ofpbuf key, odp_actions; |
4755 | struct odputil_keybuf keybuf; | |
52a90c29 | 4756 | uint16_t odp_port; |
80e5eed9 | 4757 | struct flow flow; |
abe529af BP |
4758 | int error; |
4759 | ||
88dbe0f6 | 4760 | flow_extract(packet, 0, 0, 0, &flow); |
52a90c29 BP |
4761 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4762 | flow.vlan_tci); | |
4763 | if (odp_port != ofport->odp_port) { | |
4764 | eth_pop_vlan(packet); | |
4765 | flow.vlan_tci = htons(0); | |
4766 | } | |
4767 | ||
80e5eed9 BP |
4768 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4769 | odp_flow_key_from_flow(&key, &flow); | |
4770 | ||
abe529af | 4771 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4772 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4773 | ||
df2c07f4 | 4774 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4775 | error = dpif_execute(ofproto->dpif, |
4776 | key.data, key.size, | |
4777 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4778 | packet); |
4779 | ofpbuf_uninit(&odp_actions); | |
4780 | ||
4781 | if (error) { | |
4782 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4783 | ofproto->up.name, odp_port, strerror(error)); | |
4784 | } | |
6527c598 | 4785 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4786 | return error; |
4787 | } | |
4788 | \f | |
df2c07f4 | 4789 | /* OpenFlow to datapath action translation. */ |
abe529af | 4790 | |
f25d0cf3 BP |
4791 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
4792 | struct action_xlate_ctx *); | |
4cd78906 | 4793 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4794 | |
6a7e895f BP |
4795 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
4796 | * The action will state 'slow' as the reason that the action is in the slow | |
4797 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
4798 | * dump-flows" output to see why a flow is in the slow path.) | |
4799 | * | |
4800 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
4801 | * 'stub_size' must be large enough for the action. | |
4802 | * | |
4803 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
4804 | * respectively. */ | |
4805 | static void | |
4806 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
4807 | enum slow_path_reason slow, | |
4808 | uint64_t *stub, size_t stub_size, | |
4809 | const struct nlattr **actionsp, size_t *actions_lenp) | |
4810 | { | |
4811 | union user_action_cookie cookie; | |
4812 | struct ofpbuf buf; | |
4813 | ||
4814 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
4815 | cookie.slow_path.unused = 0; | |
4816 | cookie.slow_path.reason = slow; | |
4817 | ||
4818 | ofpbuf_use_stack(&buf, stub, stub_size); | |
625b0720 BP |
4819 | if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { |
4820 | uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX); | |
4821 | odp_put_userspace_action(pid, &cookie, &buf); | |
4822 | } else { | |
4823 | put_userspace_action(ofproto, &buf, flow, &cookie); | |
4824 | } | |
6a7e895f BP |
4825 | *actionsp = buf.data; |
4826 | *actions_lenp = buf.size; | |
4827 | } | |
4828 | ||
98403001 BP |
4829 | static size_t |
4830 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4831 | struct ofpbuf *odp_actions, | |
4832 | const struct flow *flow, | |
1673e0e4 | 4833 | const union user_action_cookie *cookie) |
98403001 | 4834 | { |
98403001 BP |
4835 | uint32_t pid; |
4836 | ||
4837 | pid = dpif_port_get_pid(ofproto->dpif, | |
4838 | ofp_port_to_odp_port(flow->in_port)); | |
4839 | ||
39db78a0 | 4840 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4841 | } |
4842 | ||
36fc5f18 BP |
4843 | static void |
4844 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
4845 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 4846 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
4847 | { |
4848 | int ifindex; | |
4849 | ||
4850 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 4851 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
4852 | |
4853 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
4854 | * port information") for the interpretation of cookie->output. */ | |
4855 | switch (n_outputs) { | |
4856 | case 0: | |
4857 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 4858 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
4859 | break; |
4860 | ||
4861 | case 1: | |
4862 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4863 | if (ifindex) { | |
1673e0e4 | 4864 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
4865 | break; |
4866 | } | |
4867 | /* Fall through. */ | |
4868 | default: | |
4869 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 4870 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
4871 | break; |
4872 | } | |
4873 | } | |
4874 | ||
6ff686f2 PS |
4875 | /* Compose SAMPLE action for sFlow. */ |
4876 | static size_t | |
4877 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4878 | struct ofpbuf *odp_actions, | |
4879 | const struct flow *flow, | |
4880 | uint32_t odp_port) | |
4881 | { | |
6ff686f2 | 4882 | uint32_t probability; |
1673e0e4 | 4883 | union user_action_cookie cookie; |
6ff686f2 | 4884 | size_t sample_offset, actions_offset; |
36fc5f18 | 4885 | int cookie_offset; |
6ff686f2 PS |
4886 | |
4887 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4888 | return 0; | |
4889 | } | |
4890 | ||
6ff686f2 PS |
4891 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); |
4892 | ||
4893 | /* Number of packets out of UINT_MAX to sample. */ | |
4894 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4895 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4896 | ||
4897 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
36fc5f18 BP |
4898 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
4899 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
98403001 | 4900 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); |
6ff686f2 PS |
4901 | |
4902 | nl_msg_end_nested(odp_actions, actions_offset); | |
4903 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4904 | return cookie_offset; |
6ff686f2 PS |
4905 | } |
4906 | ||
4907 | /* SAMPLE action must be first action in any given list of actions. | |
4908 | * At this point we do not have all information required to build it. So try to | |
4909 | * build sample action as complete as possible. */ | |
4910 | static void | |
4911 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4912 | { | |
4913 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4914 | ctx->odp_actions, | |
4915 | &ctx->flow, OVSP_NONE); | |
4916 | ctx->sflow_odp_port = 0; | |
4917 | ctx->sflow_n_outputs = 0; | |
4918 | } | |
4919 | ||
4920 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4921 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4922 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4923 | static void | |
4924 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4925 | { | |
4926 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 4927 | union user_action_cookie *cookie; |
6ff686f2 PS |
4928 | |
4929 | if (!ctx->user_cookie_offset) { | |
4930 | return; | |
4931 | } | |
4932 | ||
4933 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
36fc5f18 | 4934 | sizeof(*cookie)); |
6ff686f2 PS |
4935 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
4936 | ||
36fc5f18 BP |
4937 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
4938 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
4939 | } |
4940 | ||
6ff686f2 | 4941 | static void |
81b1afb1 EJ |
4942 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4943 | bool check_stp) | |
6ff686f2 | 4944 | { |
d59906fb | 4945 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4946 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4947 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4948 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4949 | uint16_t out_port; |
d59906fb | 4950 | |
81b1afb1 | 4951 | if (ofport) { |
8b36f51e EJ |
4952 | struct priority_to_dscp *pdscp; |
4953 | ||
9e1fd49b | 4954 | if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD |
81b1afb1 EJ |
4955 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { |
4956 | return; | |
4957 | } | |
8b36f51e | 4958 | |
deedf7e7 | 4959 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4960 | if (pdscp) { |
4961 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4962 | ctx->flow.nw_tos |= pdscp->dscp; | |
4963 | } | |
81b1afb1 EJ |
4964 | } else { |
4965 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4966 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4967 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4968 | } |
4969 | ||
52a90c29 BP |
4970 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4971 | ctx->flow.vlan_tci); | |
4972 | if (out_port != odp_port) { | |
4973 | ctx->flow.vlan_tci = htons(0); | |
4974 | } | |
5bbda0aa | 4975 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4976 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4977 | ||
6ff686f2 PS |
4978 | ctx->sflow_odp_port = odp_port; |
4979 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4980 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4981 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4982 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4983 | } |
4984 | ||
abe529af | 4985 | static void |
5e48dc2b | 4986 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4987 | { |
81b1afb1 | 4988 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4989 | } |
4990 | ||
4991 | static void | |
29901626 | 4992 | xlate_table_action(struct action_xlate_ctx *ctx, |
1688c479 | 4993 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
4994 | { |
4995 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4996 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4997 | struct rule_dpif *rule; |
4998 | uint16_t old_in_port; | |
29901626 BP |
4999 | uint8_t old_table_id; |
5000 | ||
5001 | old_table_id = ctx->table_id; | |
5002 | ctx->table_id = table_id; | |
abe529af | 5003 | |
54a9cbc9 | 5004 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
5005 | old_in_port = ctx->flow.in_port; |
5006 | ctx->flow.in_port = in_port; | |
c57b2226 | 5007 | rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id); |
54a9cbc9 BP |
5008 | |
5009 | /* Tag the flow. */ | |
5010 | if (table_id > 0 && table_id < N_TABLES) { | |
5011 | struct table_dpif *table = &ofproto->tables[table_id]; | |
5012 | if (table->other_table) { | |
33780682 | 5013 | ctx->tags |= (rule && rule->tag |
54a9cbc9 BP |
5014 | ? rule->tag |
5015 | : rule_calculate_tag(&ctx->flow, | |
5016 | &table->other_table->wc, | |
5017 | table->basis)); | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
5022 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
5023 | ctx->flow.in_port = old_in_port; |
5024 | ||
5025 | if (ctx->resubmit_hook) { | |
5026 | ctx->resubmit_hook(ctx, rule); | |
5027 | } | |
5028 | ||
1688c479 IY |
5029 | if (rule == NULL && may_packet_in) { |
5030 | /* TODO:XXX | |
5031 | * check if table configuration flags | |
5032 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
5033 | * OFPTC_TABLE_MISS_CONTINUE, | |
5034 | * OFPTC_TABLE_MISS_DROP | |
5035 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
5036 | */ | |
5037 | rule = rule_dpif_miss_rule(ofproto, &ctx->flow); | |
5038 | } | |
5039 | ||
abe529af | 5040 | if (rule) { |
18b2a258 | 5041 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 5042 | |
112bc5f4 BP |
5043 | if (ctx->resubmit_stats) { |
5044 | rule_credit_stats(rule, ctx->resubmit_stats); | |
5045 | } | |
5046 | ||
abe529af | 5047 | ctx->recurse++; |
18b2a258 | 5048 | ctx->rule = rule; |
f25d0cf3 | 5049 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 5050 | ctx->rule = old_rule; |
abe529af BP |
5051 | ctx->recurse--; |
5052 | } | |
29901626 BP |
5053 | |
5054 | ctx->table_id = old_table_id; | |
abe529af BP |
5055 | } else { |
5056 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5057 | ||
29901626 | 5058 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 5059 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 5060 | ctx->max_resubmit_trigger = true; |
abe529af BP |
5061 | } |
5062 | } | |
5063 | ||
29901626 | 5064 | static void |
f25d0cf3 BP |
5065 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, |
5066 | const struct ofpact_resubmit *resubmit) | |
29901626 BP |
5067 | { |
5068 | uint16_t in_port; | |
5069 | uint8_t table_id; | |
5070 | ||
f25d0cf3 BP |
5071 | in_port = resubmit->in_port; |
5072 | if (in_port == OFPP_IN_PORT) { | |
5073 | in_port = ctx->flow.in_port; | |
5074 | } | |
5075 | ||
5076 | table_id = resubmit->table_id; | |
5077 | if (table_id == 255) { | |
5078 | table_id = ctx->table_id; | |
5079 | } | |
29901626 | 5080 | |
1688c479 | 5081 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
5082 | } |
5083 | ||
abe529af | 5084 | static void |
d59906fb | 5085 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
5086 | { |
5087 | struct ofport_dpif *ofport; | |
5088 | ||
b3e9b2ed | 5089 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 5090 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
5091 | |
5092 | if (ofp_port == ctx->flow.in_port) { | |
5093 | continue; | |
5094 | } | |
5095 | ||
5e48dc2b | 5096 | if (all) { |
81b1afb1 | 5097 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 5098 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 5099 | compose_output_action(ctx, ofp_port); |
abe529af BP |
5100 | } |
5101 | } | |
b3e9b2ed EJ |
5102 | |
5103 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
5104 | } |
5105 | ||
6ff686f2 | 5106 | static void |
f0fd1a17 | 5107 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
5108 | enum ofp_packet_in_reason reason, |
5109 | uint16_t controller_id) | |
6ff686f2 | 5110 | { |
999fba59 EJ |
5111 | struct ofputil_packet_in pin; |
5112 | struct ofpbuf *packet; | |
6ff686f2 | 5113 | |
6a7e895f | 5114 | ctx->slow |= SLOW_CONTROLLER; |
999fba59 EJ |
5115 | if (!ctx->packet) { |
5116 | return; | |
5117 | } | |
5118 | ||
5119 | packet = ofpbuf_clone(ctx->packet); | |
5120 | ||
5121 | if (packet->l2 && packet->l3) { | |
5122 | struct eth_header *eh; | |
5123 | ||
5124 | eth_pop_vlan(packet); | |
5125 | eh = packet->l2; | |
0104aba8 EJ |
5126 | |
5127 | /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 | |
5128 | * LLC frame. Calculating the Ethernet type of these frames is more | |
5129 | * trouble than seems appropriate for a simple assertion. */ | |
5130 | assert(ntohs(eh->eth_type) < ETH_TYPE_MIN | |
5131 | || eh->eth_type == ctx->flow.dl_type); | |
5132 | ||
999fba59 EJ |
5133 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
5134 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
5135 | ||
5136 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
5137 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
5138 | } | |
5139 | ||
5140 | if (packet->l4) { | |
5141 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5142 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
5143 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
5144 | } | |
5145 | ||
5146 | if (packet->l7) { | |
5147 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
5148 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
5149 | ctx->flow.tp_dst); | |
5150 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
5151 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
5152 | ctx->flow.tp_dst); | |
5153 | } | |
5154 | } | |
5155 | } | |
5156 | } | |
5157 | ||
5158 | pin.packet = packet->data; | |
5159 | pin.packet_len = packet->size; | |
f0fd1a17 | 5160 | pin.reason = reason; |
a7349929 | 5161 | pin.controller_id = controller_id; |
54834960 | 5162 | pin.table_id = ctx->table_id; |
18b2a258 | 5163 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 5164 | |
999fba59 | 5165 | pin.send_len = len; |
999fba59 EJ |
5166 | flow_get_metadata(&ctx->flow, &pin.fmd); |
5167 | ||
d8653c38 | 5168 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 5169 | ofpbuf_delete(packet); |
6ff686f2 PS |
5170 | } |
5171 | ||
f0fd1a17 | 5172 | static bool |
c2d967a5 | 5173 | compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 PS |
5174 | { |
5175 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
5176 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
5177 | return false; | |
5178 | } | |
5179 | ||
5180 | if (ctx->flow.nw_ttl > 1) { | |
5181 | ctx->flow.nw_ttl--; | |
5182 | return false; | |
5183 | } else { | |
c2d967a5 MM |
5184 | size_t i; |
5185 | ||
5186 | for (i = 0; i < ids->n_controllers; i++) { | |
5187 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
5188 | ids->cnt_ids[i]); | |
5189 | } | |
f0fd1a17 PS |
5190 | |
5191 | /* Stop processing for current table. */ | |
5192 | return true; | |
5193 | } | |
5194 | } | |
5195 | ||
abe529af | 5196 | static void |
f25d0cf3 | 5197 | xlate_output_action(struct action_xlate_ctx *ctx, |
1688c479 | 5198 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af BP |
5199 | { |
5200 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
5201 | ||
5202 | ctx->nf_output_iface = NF_OUT_DROP; | |
5203 | ||
5204 | switch (port) { | |
5205 | case OFPP_IN_PORT: | |
81b1afb1 | 5206 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
5207 | break; |
5208 | case OFPP_TABLE: | |
1688c479 | 5209 | xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); |
abe529af BP |
5210 | break; |
5211 | case OFPP_NORMAL: | |
5212 | xlate_normal(ctx); | |
5213 | break; | |
5214 | case OFPP_FLOOD: | |
d59906fb | 5215 | flood_packets(ctx, false); |
abe529af BP |
5216 | break; |
5217 | case OFPP_ALL: | |
d59906fb | 5218 | flood_packets(ctx, true); |
abe529af BP |
5219 | break; |
5220 | case OFPP_CONTROLLER: | |
a7349929 | 5221 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 5222 | break; |
e81d2933 EJ |
5223 | case OFPP_NONE: |
5224 | break; | |
a0fbe94a | 5225 | case OFPP_LOCAL: |
abe529af BP |
5226 | default: |
5227 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 5228 | compose_output_action(ctx, port); |
abe529af BP |
5229 | } |
5230 | break; | |
5231 | } | |
5232 | ||
5233 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
5234 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
5235 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
5236 | ctx->nf_output_iface = prev_nf_output_iface; | |
5237 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
5238 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
5239 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5240 | } | |
5241 | } | |
5242 | ||
f694937d EJ |
5243 | static void |
5244 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5245 | const struct ofpact_output_reg *or) |
f694937d | 5246 | { |
f25d0cf3 BP |
5247 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); |
5248 | if (port <= UINT16_MAX) { | |
1688c479 | 5249 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
5250 | } |
5251 | } | |
5252 | ||
abe529af BP |
5253 | static void |
5254 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5255 | const struct ofpact_enqueue *enqueue) |
abe529af | 5256 | { |
f25d0cf3 BP |
5257 | uint16_t ofp_port = enqueue->port; |
5258 | uint32_t queue_id = enqueue->queue; | |
abff858b | 5259 | uint32_t flow_priority, priority; |
abe529af BP |
5260 | int error; |
5261 | ||
f25d0cf3 BP |
5262 | /* Translate queue to priority. */ |
5263 | error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority); | |
abe529af BP |
5264 | if (error) { |
5265 | /* Fall back to ordinary output action. */ | |
1688c479 | 5266 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
5267 | return; |
5268 | } | |
5269 | ||
f25d0cf3 | 5270 | /* Check output port. */ |
abe529af BP |
5271 | if (ofp_port == OFPP_IN_PORT) { |
5272 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
5273 | } else if (ofp_port == ctx->flow.in_port) { |
5274 | return; | |
abe529af | 5275 | } |
abe529af | 5276 | |
df2c07f4 | 5277 | /* Add datapath actions. */ |
deedf7e7 BP |
5278 | flow_priority = ctx->flow.skb_priority; |
5279 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 5280 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 5281 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
5282 | |
5283 | /* Update NetFlow output port. */ | |
5284 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 5285 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
5286 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
5287 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5288 | } | |
5289 | } | |
5290 | ||
5291 | static void | |
f25d0cf3 | 5292 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 5293 | { |
f25d0cf3 | 5294 | uint32_t skb_priority; |
abe529af | 5295 | |
f25d0cf3 BP |
5296 | if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) { |
5297 | ctx->flow.skb_priority = skb_priority; | |
5298 | } else { | |
5299 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 5300 | * has already been logged. */ |
abe529af | 5301 | } |
abe529af BP |
5302 | } |
5303 | ||
5304 | struct xlate_reg_state { | |
5305 | ovs_be16 vlan_tci; | |
5306 | ovs_be64 tun_id; | |
5307 | }; | |
5308 | ||
abe529af BP |
5309 | static void |
5310 | xlate_autopath(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5311 | const struct ofpact_autopath *ap) |
abe529af | 5312 | { |
f25d0cf3 | 5313 | uint16_t ofp_port = ap->port; |
abe529af BP |
5314 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); |
5315 | ||
5316 | if (!port || !port->bundle) { | |
5317 | ofp_port = OFPP_NONE; | |
5318 | } else if (port->bundle->bond) { | |
5319 | /* Autopath does not support VLAN hashing. */ | |
5320 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 5321 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
5322 | if (slave) { |
5323 | ofp_port = slave->up.ofp_port; | |
5324 | } | |
5325 | } | |
f25d0cf3 | 5326 | nxm_reg_load(&ap->dst, ofp_port, &ctx->flow); |
abe529af BP |
5327 | } |
5328 | ||
daff3353 EJ |
5329 | static bool |
5330 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
5331 | { | |
5332 | struct ofproto_dpif *ofproto = ofproto_; | |
5333 | struct ofport_dpif *port; | |
5334 | ||
5335 | switch (ofp_port) { | |
5336 | case OFPP_IN_PORT: | |
5337 | case OFPP_TABLE: | |
5338 | case OFPP_NORMAL: | |
5339 | case OFPP_FLOOD: | |
5340 | case OFPP_ALL: | |
439e4d8c | 5341 | case OFPP_NONE: |
daff3353 EJ |
5342 | return true; |
5343 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
5344 | return false; | |
5345 | default: | |
5346 | port = get_ofp_port(ofproto, ofp_port); | |
5347 | return port ? port->may_enable : false; | |
5348 | } | |
5349 | } | |
5350 | ||
f25d0cf3 BP |
5351 | static void |
5352 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
5353 | const struct ofpact_bundle *bundle) | |
5354 | { | |
5355 | uint16_t port; | |
5356 | ||
5357 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
5358 | if (bundle->dst.field) { | |
5359 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
5360 | } else { | |
1688c479 | 5361 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
5362 | } |
5363 | } | |
5364 | ||
75a75043 BP |
5365 | static void |
5366 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5367 | const struct ofpact_learn *learn) |
75a75043 BP |
5368 | { |
5369 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
5370 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
5371 | uint64_t ofpacts_stub[1024 / 8]; |
5372 | struct ofpbuf ofpacts; | |
75a75043 BP |
5373 | int error; |
5374 | ||
f25d0cf3 BP |
5375 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
5376 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
75a75043 BP |
5377 | |
5378 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
5379 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
5380 | VLOG_WARN("learning action failed to modify flow table (%s)", |
5381 | ofperr_get_name(error)); | |
75a75043 BP |
5382 | } |
5383 | ||
f25d0cf3 | 5384 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
5385 | } |
5386 | ||
0e553d9c BP |
5387 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
5388 | * means "infinite". */ | |
5389 | static void | |
5390 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
5391 | { | |
5392 | if (max && (!*timeout || *timeout > max)) { | |
5393 | *timeout = max; | |
5394 | } | |
5395 | } | |
5396 | ||
5397 | static void | |
5398 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5399 | const struct ofpact_fin_timeout *oft) |
0e553d9c BP |
5400 | { |
5401 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
5402 | struct rule_dpif *rule = ctx->rule; | |
5403 | ||
f25d0cf3 BP |
5404 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
5405 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
5406 | } |
5407 | } | |
5408 | ||
21f7563c JP |
5409 | static bool |
5410 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
5411 | { | |
9e1fd49b BP |
5412 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
5413 | ? OFPUTIL_PC_NO_RECV_STP | |
5414 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
5415 | return false; |
5416 | } | |
5417 | ||
5418 | /* Only drop packets here if both forwarding and learning are | |
5419 | * disabled. If just learning is enabled, we need to have | |
5420 | * OFPP_NORMAL and the learning action have a look at the packet | |
5421 | * before we can drop it. */ | |
5422 | if (!stp_forward_in_state(port->stp_state) | |
5423 | && !stp_learn_in_state(port->stp_state)) { | |
5424 | return false; | |
5425 | } | |
5426 | ||
5427 | return true; | |
5428 | } | |
5429 | ||
abe529af | 5430 | static void |
f25d0cf3 | 5431 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
abe529af BP |
5432 | struct action_xlate_ctx *ctx) |
5433 | { | |
5434 | const struct ofport_dpif *port; | |
254750ce | 5435 | bool was_evictable = true; |
f25d0cf3 | 5436 | const struct ofpact *a; |
abe529af BP |
5437 | |
5438 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 5439 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
5440 | /* Drop this flow. */ |
5441 | return; | |
5442 | } | |
5443 | ||
254750ce BP |
5444 | if (ctx->rule) { |
5445 | /* Don't let the rule we're working on get evicted underneath us. */ | |
5446 | was_evictable = ctx->rule->up.evictable; | |
5447 | ctx->rule->up.evictable = false; | |
5448 | } | |
f25d0cf3 BP |
5449 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
5450 | struct ofpact_controller *controller; | |
38f2e360 | 5451 | |
848e8809 EJ |
5452 | if (ctx->exit) { |
5453 | break; | |
5454 | } | |
5455 | ||
f25d0cf3 BP |
5456 | switch (a->type) { |
5457 | case OFPACT_OUTPUT: | |
5458 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 5459 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
5460 | break; |
5461 | ||
5462 | case OFPACT_CONTROLLER: | |
5463 | controller = ofpact_get_CONTROLLER(a); | |
5464 | execute_controller_action(ctx, controller->max_len, | |
5465 | controller->reason, | |
5466 | controller->controller_id); | |
5467 | break; | |
690a61c5 | 5468 | |
f25d0cf3 BP |
5469 | case OFPACT_ENQUEUE: |
5470 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
5471 | break; |
5472 | ||
f25d0cf3 | 5473 | case OFPACT_SET_VLAN_VID: |
abe529af | 5474 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
f25d0cf3 BP |
5475 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) |
5476 | | htons(VLAN_CFI)); | |
abe529af BP |
5477 | break; |
5478 | ||
f25d0cf3 | 5479 | case OFPACT_SET_VLAN_PCP: |
abe529af | 5480 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
f25d0cf3 BP |
5481 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp |
5482 | << VLAN_PCP_SHIFT) | |
5483 | | VLAN_CFI); | |
abe529af BP |
5484 | break; |
5485 | ||
f25d0cf3 | 5486 | case OFPACT_STRIP_VLAN: |
abe529af | 5487 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
5488 | break; |
5489 | ||
f25d0cf3 BP |
5490 | case OFPACT_SET_ETH_SRC: |
5491 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
5492 | ETH_ADDR_LEN); | |
abe529af BP |
5493 | break; |
5494 | ||
f25d0cf3 BP |
5495 | case OFPACT_SET_ETH_DST: |
5496 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
5497 | ETH_ADDR_LEN); | |
abe529af BP |
5498 | break; |
5499 | ||
f25d0cf3 BP |
5500 | case OFPACT_SET_IPV4_SRC: |
5501 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
abe529af BP |
5502 | break; |
5503 | ||
f25d0cf3 BP |
5504 | case OFPACT_SET_IPV4_DST: |
5505 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
abe529af BP |
5506 | break; |
5507 | ||
f25d0cf3 | 5508 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d PS |
5509 | /* OpenFlow 1.0 only supports IPv4. */ |
5510 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5511 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
f25d0cf3 | 5512 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; |
c4f2731d | 5513 | } |
abe529af BP |
5514 | break; |
5515 | ||
f25d0cf3 BP |
5516 | case OFPACT_SET_L4_SRC_PORT: |
5517 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
abe529af BP |
5518 | break; |
5519 | ||
f25d0cf3 BP |
5520 | case OFPACT_SET_L4_DST_PORT: |
5521 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
abe529af BP |
5522 | break; |
5523 | ||
f25d0cf3 BP |
5524 | case OFPACT_RESUBMIT: |
5525 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
5526 | break; |
5527 | ||
f25d0cf3 BP |
5528 | case OFPACT_SET_TUNNEL: |
5529 | ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
29901626 BP |
5530 | break; |
5531 | ||
f25d0cf3 BP |
5532 | case OFPACT_SET_QUEUE: |
5533 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
5534 | break; |
5535 | ||
f25d0cf3 | 5536 | case OFPACT_POP_QUEUE: |
deedf7e7 | 5537 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
5538 | break; |
5539 | ||
f25d0cf3 BP |
5540 | case OFPACT_REG_MOVE: |
5541 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
38f2e360 BP |
5542 | break; |
5543 | ||
f25d0cf3 BP |
5544 | case OFPACT_REG_LOAD: |
5545 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
38f2e360 BP |
5546 | break; |
5547 | ||
f25d0cf3 | 5548 | case OFPACT_DEC_TTL: |
c2d967a5 | 5549 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
5550 | goto out; |
5551 | } | |
38f2e360 BP |
5552 | break; |
5553 | ||
f25d0cf3 BP |
5554 | case OFPACT_NOTE: |
5555 | /* Nothing to do. */ | |
abe529af BP |
5556 | break; |
5557 | ||
f25d0cf3 BP |
5558 | case OFPACT_MULTIPATH: |
5559 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
abe529af | 5560 | break; |
daff3353 | 5561 | |
f25d0cf3 BP |
5562 | case OFPACT_AUTOPATH: |
5563 | xlate_autopath(ctx, ofpact_get_AUTOPATH(a)); | |
daff3353 | 5564 | break; |
a368bb53 | 5565 | |
f25d0cf3 | 5566 | case OFPACT_BUNDLE: |
a368bb53 | 5567 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 5568 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 5569 | break; |
f694937d | 5570 | |
f25d0cf3 BP |
5571 | case OFPACT_OUTPUT_REG: |
5572 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 5573 | break; |
75a75043 | 5574 | |
f25d0cf3 | 5575 | case OFPACT_LEARN: |
75a75043 | 5576 | ctx->has_learn = true; |
3de9590b | 5577 | if (ctx->may_learn) { |
f25d0cf3 | 5578 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
5579 | } |
5580 | break; | |
848e8809 | 5581 | |
f25d0cf3 | 5582 | case OFPACT_EXIT: |
848e8809 EJ |
5583 | ctx->exit = true; |
5584 | break; | |
0e553d9c | 5585 | |
f25d0cf3 | 5586 | case OFPACT_FIN_TIMEOUT: |
0e553d9c | 5587 | ctx->has_fin_timeout = true; |
f25d0cf3 | 5588 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 5589 | break; |
abe529af BP |
5590 | } |
5591 | } | |
21f7563c | 5592 | |
f0fd1a17 | 5593 | out: |
21f7563c JP |
5594 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
5595 | * so drop it now if forwarding is disabled. */ | |
5596 | if (port && !stp_forward_in_state(port->stp_state)) { | |
5597 | ofpbuf_clear(ctx->odp_actions); | |
5598 | add_sflow_action(ctx); | |
5599 | } | |
254750ce BP |
5600 | if (ctx->rule) { |
5601 | ctx->rule->up.evictable = was_evictable; | |
5602 | } | |
abe529af BP |
5603 | } |
5604 | ||
5605 | static void | |
5606 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
5607 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
18b2a258 | 5608 | ovs_be16 initial_tci, struct rule_dpif *rule, |
0e553d9c | 5609 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af BP |
5610 | { |
5611 | ctx->ofproto = ofproto; | |
5612 | ctx->flow = *flow; | |
e84173dc BP |
5613 | ctx->base_flow = ctx->flow; |
5614 | ctx->base_flow.tun_id = 0; | |
5615 | ctx->base_flow.vlan_tci = initial_tci; | |
18b2a258 | 5616 | ctx->rule = rule; |
abe529af | 5617 | ctx->packet = packet; |
3de9590b | 5618 | ctx->may_learn = packet != NULL; |
0e553d9c | 5619 | ctx->tcp_flags = tcp_flags; |
abe529af | 5620 | ctx->resubmit_hook = NULL; |
479df176 | 5621 | ctx->report_hook = NULL; |
112bc5f4 | 5622 | ctx->resubmit_stats = NULL; |
abe529af BP |
5623 | } |
5624 | ||
f25d0cf3 BP |
5625 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' |
5626 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
050ac423 | 5627 | static void |
abe529af | 5628 | xlate_actions(struct action_xlate_ctx *ctx, |
f25d0cf3 | 5629 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 | 5630 | struct ofpbuf *odp_actions) |
abe529af | 5631 | { |
43d50bc8 BP |
5632 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so |
5633 | * that in the future we always keep a copy of the original flow for | |
5634 | * tracing purposes. */ | |
5635 | static bool hit_resubmit_limit; | |
5636 | ||
6a7e895f BP |
5637 | enum slow_path_reason special; |
5638 | ||
abe529af BP |
5639 | COVERAGE_INC(ofproto_dpif_xlate); |
5640 | ||
050ac423 BP |
5641 | ofpbuf_clear(odp_actions); |
5642 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
5643 | ||
5644 | ctx->odp_actions = odp_actions; | |
97e42c92 | 5645 | ctx->tags = 0; |
6a7e895f | 5646 | ctx->slow = 0; |
97e42c92 BP |
5647 | ctx->has_learn = false; |
5648 | ctx->has_normal = false; | |
0e553d9c | 5649 | ctx->has_fin_timeout = false; |
97e42c92 | 5650 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 5651 | ctx->mirrors = 0; |
97e42c92 | 5652 | ctx->recurse = 0; |
6a6455e5 | 5653 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 5654 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 5655 | ctx->table_id = 0; |
848e8809 | 5656 | ctx->exit = false; |
7257b535 | 5657 | |
43d50bc8 | 5658 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 BP |
5659 | /* Do this conditionally because the copy is expensive enough that it |
5660 | * shows up in profiles. | |
5661 | * | |
5662 | * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4 | |
5663 | * believe that I wasn't using it without initializing it if I kept it | |
5664 | * in a local variable. */ | |
5665 | ctx->orig_flow = ctx->flow; | |
5666 | } | |
5667 | ||
eadef313 | 5668 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
5669 | switch (ctx->ofproto->up.frag_handling) { |
5670 | case OFPC_FRAG_NORMAL: | |
5671 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
5672 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
5673 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
5674 | break; |
5675 | ||
5676 | case OFPC_FRAG_DROP: | |
050ac423 | 5677 | return; |
7257b535 BP |
5678 | |
5679 | case OFPC_FRAG_REASM: | |
5680 | NOT_REACHED(); | |
5681 | ||
5682 | case OFPC_FRAG_NX_MATCH: | |
5683 | /* Nothing to do. */ | |
5684 | break; | |
f0fd1a17 PS |
5685 | |
5686 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
5687 | NOT_REACHED(); | |
7257b535 BP |
5688 | } |
5689 | } | |
5690 | ||
6a7e895f BP |
5691 | special = process_special(ctx->ofproto, &ctx->flow, ctx->packet); |
5692 | if (special) { | |
5693 | ctx->slow |= special; | |
abe529af | 5694 | } else { |
6a6455e5 | 5695 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
6a6455e5 EJ |
5696 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; |
5697 | ||
6ff686f2 | 5698 | add_sflow_action(ctx); |
f25d0cf3 | 5699 | do_xlate_actions(ofpacts, ofpacts_len, ctx); |
abe529af | 5700 | |
43d50bc8 BP |
5701 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { |
5702 | if (!hit_resubmit_limit) { | |
5703 | /* We didn't record the original flow. Make sure we do from | |
5704 | * now on. */ | |
5705 | hit_resubmit_limit = true; | |
5706 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
5707 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5708 | ||
5709 | ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet, | |
5710 | initial_tci, &ds); | |
5711 | VLOG_ERR("Trace triggered by excessive resubmit " | |
5712 | "recursion:\n%s", ds_cstr(&ds)); | |
5713 | ds_destroy(&ds); | |
5714 | } | |
6a6455e5 EJ |
5715 | } |
5716 | ||
b6848f13 BP |
5717 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
5718 | ctx->odp_actions->data, | |
5719 | ctx->odp_actions->size)) { | |
6a7e895f | 5720 | ctx->slow |= SLOW_IN_BAND; |
b6848f13 BP |
5721 | if (ctx->packet |
5722 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
5723 | ctx->packet)) { | |
5e48dc2b | 5724 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
5725 | } |
5726 | } | |
ccb7c863 BP |
5727 | if (ctx->ofproto->has_mirrors) { |
5728 | add_mirror_actions(ctx, &ctx->orig_flow); | |
5729 | } | |
a7c4eaf6 | 5730 | fix_sflow_action(ctx); |
abe529af | 5731 | } |
050ac423 BP |
5732 | } |
5733 | ||
f25d0cf3 BP |
5734 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
5735 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 BP |
5736 | static void |
5737 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
f25d0cf3 BP |
5738 | const struct ofpact *ofpacts, |
5739 | size_t ofpacts_len) | |
050ac423 BP |
5740 | { |
5741 | uint64_t odp_actions_stub[1024 / 8]; | |
5742 | struct ofpbuf odp_actions; | |
abe529af | 5743 | |
050ac423 | 5744 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
f25d0cf3 | 5745 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); |
050ac423 | 5746 | ofpbuf_uninit(&odp_actions); |
abe529af | 5747 | } |
479df176 BP |
5748 | |
5749 | static void | |
5750 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
5751 | { | |
5752 | if (ctx->report_hook) { | |
5753 | ctx->report_hook(ctx, s); | |
5754 | } | |
5755 | } | |
abe529af BP |
5756 | \f |
5757 | /* OFPP_NORMAL implementation. */ | |
5758 | ||
abe529af BP |
5759 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
5760 | ||
ecac4ebf BP |
5761 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
5762 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5763 | * the bundle on which the packet was received, returns the VLAN to which the | |
5764 | * packet belongs. | |
5765 | * | |
5766 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5767 | static uint16_t | |
5768 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5769 | { | |
5770 | switch (in_bundle->vlan_mode) { | |
5771 | case PORT_VLAN_ACCESS: | |
5772 | return in_bundle->vlan; | |
5773 | break; | |
5774 | ||
5775 | case PORT_VLAN_TRUNK: | |
5776 | return vid; | |
5777 | ||
5778 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5779 | case PORT_VLAN_NATIVE_TAGGED: | |
5780 | return vid ? vid : in_bundle->vlan; | |
5781 | ||
5782 | default: | |
5783 | NOT_REACHED(); | |
5784 | } | |
5785 | } | |
5786 | ||
5da5ec37 BP |
5787 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5788 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5789 | * a warning. | |
5790 | * | |
5791 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5792 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5793 | * 0...4095. */ | |
5794 | static bool | |
5795 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5796 | { | |
33158a18 JP |
5797 | /* Allow any VID on the OFPP_NONE port. */ |
5798 | if (in_bundle == &ofpp_none_bundle) { | |
5799 | return true; | |
5800 | } | |
5801 | ||
5da5ec37 BP |
5802 | switch (in_bundle->vlan_mode) { |
5803 | case PORT_VLAN_ACCESS: | |
5804 | if (vid) { | |
5805 | if (warn) { | |
5806 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5807 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5808 | "packet received on port %s configured as VLAN " | |
5809 | "%"PRIu16" access port", | |
5810 | in_bundle->ofproto->up.name, vid, | |
5811 | in_bundle->name, in_bundle->vlan); | |
5812 | } | |
5813 | return false; | |
5814 | } | |
5815 | return true; | |
5816 | ||
5817 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5818 | case PORT_VLAN_NATIVE_TAGGED: | |
5819 | if (!vid) { | |
5820 | /* Port must always carry its native VLAN. */ | |
5821 | return true; | |
5822 | } | |
5823 | /* Fall through. */ | |
5824 | case PORT_VLAN_TRUNK: | |
5825 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5826 | if (warn) { | |
5827 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5828 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5829 | "received on port %s not configured for trunking " | |
5830 | "VLAN %"PRIu16, | |
5831 | in_bundle->ofproto->up.name, vid, | |
5832 | in_bundle->name, vid); | |
5833 | } | |
5834 | return false; | |
5835 | } | |
5836 | return true; | |
5837 | ||
5838 | default: | |
5839 | NOT_REACHED(); | |
5840 | } | |
5841 | ||
5842 | } | |
5843 | ||
ecac4ebf BP |
5844 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5845 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5846 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5847 | * then the 802.1Q header should only be included in the packet if there is a | |
5848 | * nonzero PCP.) | |
5849 | * | |
5850 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5851 | static uint16_t | |
5852 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5853 | { | |
5854 | switch (out_bundle->vlan_mode) { | |
5855 | case PORT_VLAN_ACCESS: | |
5856 | return 0; | |
5857 | ||
5858 | case PORT_VLAN_TRUNK: | |
5859 | case PORT_VLAN_NATIVE_TAGGED: | |
5860 | return vlan; | |
5861 | ||
5862 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5863 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5864 | ||
5865 | default: | |
5866 | NOT_REACHED(); | |
5867 | } | |
5868 | } | |
5869 | ||
395e68ce BP |
5870 | static void |
5871 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5872 | uint16_t vlan) | |
abe529af | 5873 | { |
395e68ce BP |
5874 | struct ofport_dpif *port; |
5875 | uint16_t vid; | |
81b1afb1 | 5876 | ovs_be16 tci, old_tci; |
ecac4ebf | 5877 | |
395e68ce BP |
5878 | vid = output_vlan_to_vid(out_bundle, vlan); |
5879 | if (!out_bundle->bond) { | |
5880 | port = ofbundle_get_a_port(out_bundle); | |
5881 | } else { | |
5882 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5883 | vid, &ctx->tags); | |
5884 | if (!port) { | |
5885 | /* No slaves enabled, so drop packet. */ | |
5886 | return; | |
5887 | } | |
5888 | } | |
abe529af | 5889 | |
81b1afb1 | 5890 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5891 | tci = htons(vid); |
5892 | if (tci || out_bundle->use_priority_tags) { | |
5893 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5894 | if (tci) { | |
5895 | tci |= htons(VLAN_CFI); | |
5896 | } | |
395e68ce | 5897 | } |
81b1afb1 | 5898 | ctx->flow.vlan_tci = tci; |
395e68ce | 5899 | |
5e48dc2b | 5900 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5901 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5902 | } |
5903 | ||
5904 | static int | |
5905 | mirror_mask_ffs(mirror_mask_t mask) | |
5906 | { | |
5907 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5908 | return ffs(mask); | |
5909 | } | |
5910 | ||
abe529af BP |
5911 | static bool |
5912 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5913 | { | |
ecac4ebf | 5914 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5915 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5916 | } |
5917 | ||
5918 | static bool | |
5919 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5920 | { | |
5921 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5922 | } | |
5923 | ||
5924 | /* Returns an arbitrary interface within 'bundle'. */ | |
5925 | static struct ofport_dpif * | |
5926 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5927 | { | |
5928 | return CONTAINER_OF(list_front(&bundle->ports), | |
5929 | struct ofport_dpif, bundle_node); | |
5930 | } | |
5931 | ||
abe529af BP |
5932 | static bool |
5933 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5934 | { | |
fc3d7408 | 5935 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5936 | } |
5937 | ||
5938 | static void | |
c06bba01 | 5939 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5940 | { |
5941 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5942 | mirror_mask_t mirrors; | |
c06bba01 JP |
5943 | struct ofbundle *in_bundle; |
5944 | uint16_t vlan; | |
5945 | uint16_t vid; | |
5946 | const struct nlattr *a; | |
5947 | size_t left; | |
5948 | ||
3581c12c | 5949 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
70c2fd56 | 5950 | ctx->packet != NULL, NULL); |
3581c12c | 5951 | if (!in_bundle) { |
c06bba01 JP |
5952 | return; |
5953 | } | |
c06bba01 JP |
5954 | mirrors = in_bundle->src_mirrors; |
5955 | ||
5956 | /* Drop frames on bundles reserved for mirroring. */ | |
5957 | if (in_bundle->mirror_out) { | |
5958 | if (ctx->packet != NULL) { | |
5959 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5960 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5961 | "%s, which is reserved exclusively for mirroring", | |
5962 | ctx->ofproto->up.name, in_bundle->name); | |
5963 | } | |
5964 | return; | |
5965 | } | |
5966 | ||
5967 | /* Check VLAN. */ | |
5968 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5969 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5970 | return; | |
5971 | } | |
5972 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5973 | ||
5974 | /* Look at the output ports to check for destination selections. */ | |
5975 | ||
5976 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5977 | ctx->odp_actions->size) { | |
5978 | enum ovs_action_attr type = nl_attr_type(a); | |
5979 | struct ofport_dpif *ofport; | |
5980 | ||
5981 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5982 | continue; | |
5983 | } | |
5984 | ||
5985 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5986 | if (ofport && ofport->bundle) { |
5987 | mirrors |= ofport->bundle->dst_mirrors; | |
5988 | } | |
c06bba01 | 5989 | } |
abe529af BP |
5990 | |
5991 | if (!mirrors) { | |
5992 | return; | |
5993 | } | |
5994 | ||
c06bba01 JP |
5995 | /* Restore the original packet before adding the mirror actions. */ |
5996 | ctx->flow = *orig_flow; | |
5997 | ||
9ba15e2a BP |
5998 | while (mirrors) { |
5999 | struct ofmirror *m; | |
9ba15e2a BP |
6000 | |
6001 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
6002 | ||
6003 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 6004 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
6005 | continue; |
6006 | } | |
6007 | ||
6008 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 6009 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 6010 | if (m->out) { |
395e68ce | 6011 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
6012 | } else if (vlan != m->out_vlan |
6013 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
6014 | struct ofbundle *bundle; |
6015 | ||
6016 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
6017 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
6018 | && !bundle->mirror_out) { |
6019 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
6020 | } |
6021 | } | |
6022 | } | |
abe529af BP |
6023 | } |
6024 | } | |
6025 | ||
9d24de3b JP |
6026 | static void |
6027 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
6028 | uint64_t packets, uint64_t bytes) | |
6029 | { | |
6030 | if (!mirrors) { | |
6031 | return; | |
6032 | } | |
6033 | ||
8472a3ce | 6034 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
6035 | struct ofmirror *m; |
6036 | ||
6037 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
6038 | ||
6039 | if (!m) { | |
6040 | /* In normal circumstances 'm' will not be NULL. However, | |
6041 | * if mirrors are reconfigured, we can temporarily get out | |
6042 | * of sync in facet_revalidate(). We could "correct" the | |
6043 | * mirror list before reaching here, but doing that would | |
6044 | * not properly account the traffic stats we've currently | |
6045 | * accumulated for previous mirror configuration. */ | |
6046 | continue; | |
6047 | } | |
6048 | ||
6049 | m->packet_count += packets; | |
6050 | m->byte_count += bytes; | |
6051 | } | |
6052 | } | |
6053 | ||
abe529af BP |
6054 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
6055 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
6056 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
6057 | static bool | |
6058 | is_gratuitous_arp(const struct flow *flow) | |
6059 | { | |
6060 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
6061 | && eth_addr_is_broadcast(flow->dl_dst) | |
6062 | && (flow->nw_proto == ARP_OP_REPLY | |
6063 | || (flow->nw_proto == ARP_OP_REQUEST | |
6064 | && flow->nw_src == flow->nw_dst))); | |
6065 | } | |
6066 | ||
6067 | static void | |
6068 | update_learning_table(struct ofproto_dpif *ofproto, | |
6069 | const struct flow *flow, int vlan, | |
6070 | struct ofbundle *in_bundle) | |
6071 | { | |
6072 | struct mac_entry *mac; | |
6073 | ||
33158a18 JP |
6074 | /* Don't learn the OFPP_NONE port. */ |
6075 | if (in_bundle == &ofpp_none_bundle) { | |
6076 | return; | |
6077 | } | |
6078 | ||
abe529af BP |
6079 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
6080 | return; | |
6081 | } | |
6082 | ||
6083 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
6084 | if (is_gratuitous_arp(flow)) { | |
6085 | /* We don't want to learn from gratuitous ARP packets that are | |
6086 | * reflected back over bond slaves so we lock the learning table. */ | |
6087 | if (!in_bundle->bond) { | |
6088 | mac_entry_set_grat_arp_lock(mac); | |
6089 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
6090 | return; | |
6091 | } | |
6092 | } | |
6093 | ||
6094 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
6095 | /* The log messages here could actually be useful in debugging, | |
6096 | * so keep the rate limit relatively high. */ | |
6097 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
6098 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
6099 | "on port %s in VLAN %d", | |
6100 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
6101 | in_bundle->name, vlan); | |
6102 | ||
6103 | mac->port.p = in_bundle; | |
6104 | tag_set_add(&ofproto->revalidate_set, | |
6105 | mac_learning_changed(ofproto->ml, mac)); | |
6106 | } | |
6107 | } | |
6108 | ||
3581c12c | 6109 | static struct ofbundle * |
4acbc98d SH |
6110 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
6111 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
6112 | { |
6113 | struct ofport_dpif *ofport; | |
6114 | ||
6115 | /* Find the port and bundle for the received packet. */ | |
6116 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
6117 | if (in_ofportp) { |
6118 | *in_ofportp = ofport; | |
6119 | } | |
395e68ce | 6120 | if (ofport && ofport->bundle) { |
3581c12c | 6121 | return ofport->bundle; |
395e68ce BP |
6122 | } |
6123 | ||
70c2fd56 BP |
6124 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
6125 | * port for traffic that it is sourcing. */ | |
6126 | if (in_port == OFPP_NONE) { | |
6127 | return &ofpp_none_bundle; | |
6128 | } | |
6129 | ||
395e68ce BP |
6130 | /* Odd. A few possible reasons here: |
6131 | * | |
6132 | * - We deleted a port but there are still a few packets queued up | |
6133 | * from it. | |
6134 | * | |
6135 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
6136 | * we don't know about. | |
6137 | * | |
6138 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
6139 | * This is particularly likely to happen if a packet was received on the |
6140 | * port after it was created, but before the client had a chance to | |
6141 | * configure its bundle. | |
395e68ce BP |
6142 | */ |
6143 | if (warn) { | |
6144 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6145 | ||
6146 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
6147 | "port %"PRIu16, ofproto->up.name, in_port); | |
6148 | } | |
6149 | return NULL; | |
6150 | } | |
6151 | ||
5da5ec37 | 6152 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
6153 | * dropped. Returns true if they may be forwarded, false if they should be |
6154 | * dropped. | |
6155 | * | |
395e68ce BP |
6156 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
6157 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 6158 | * |
395e68ce BP |
6159 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
6160 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
6161 | * checked by input_vid_is_valid(). | |
abe529af BP |
6162 | * |
6163 | * May also add tags to '*tags', although the current implementation only does | |
6164 | * so in one special case. | |
6165 | */ | |
6166 | static bool | |
479df176 BP |
6167 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, |
6168 | uint16_t vlan) | |
abe529af | 6169 | { |
479df176 BP |
6170 | struct ofproto_dpif *ofproto = ctx->ofproto; |
6171 | struct flow *flow = &ctx->flow; | |
395e68ce | 6172 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 6173 | |
395e68ce BP |
6174 | /* Drop frames for reserved multicast addresses |
6175 | * only if forward_bpdu option is absent. */ | |
614ec445 | 6176 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 6177 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
6178 | return false; |
6179 | } | |
6180 | ||
abe529af BP |
6181 | if (in_bundle->bond) { |
6182 | struct mac_entry *mac; | |
6183 | ||
6184 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
479df176 | 6185 | flow->dl_dst, &ctx->tags)) { |
abe529af BP |
6186 | case BV_ACCEPT: |
6187 | break; | |
6188 | ||
6189 | case BV_DROP: | |
479df176 | 6190 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
6191 | return false; |
6192 | ||
6193 | case BV_DROP_IF_MOVED: | |
6194 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
6195 | if (mac && mac->port.p != in_bundle && | |
6196 | (!is_gratuitous_arp(flow) | |
6197 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
6198 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
6199 | "dropping"); | |
abe529af BP |
6200 | return false; |
6201 | } | |
6202 | break; | |
6203 | } | |
6204 | } | |
6205 | ||
6206 | return true; | |
6207 | } | |
6208 | ||
4cd78906 | 6209 | static void |
abe529af BP |
6210 | xlate_normal(struct action_xlate_ctx *ctx) |
6211 | { | |
395e68ce | 6212 | struct ofport_dpif *in_port; |
abe529af | 6213 | struct ofbundle *in_bundle; |
abe529af | 6214 | struct mac_entry *mac; |
395e68ce BP |
6215 | uint16_t vlan; |
6216 | uint16_t vid; | |
abe529af | 6217 | |
75a75043 BP |
6218 | ctx->has_normal = true; |
6219 | ||
3581c12c | 6220 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
70c2fd56 | 6221 | ctx->packet != NULL, &in_port); |
3581c12c | 6222 | if (!in_bundle) { |
479df176 | 6223 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
6224 | return; |
6225 | } | |
3581c12c | 6226 | |
395e68ce BP |
6227 | /* Drop malformed frames. */ |
6228 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
6229 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
6230 | if (ctx->packet != NULL) { | |
6231 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6232 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
6233 | "VLAN tag received on port %s", | |
6234 | ctx->ofproto->up.name, in_bundle->name); | |
6235 | } | |
479df176 | 6236 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
6237 | return; |
6238 | } | |
6239 | ||
6240 | /* Drop frames on bundles reserved for mirroring. */ | |
6241 | if (in_bundle->mirror_out) { | |
6242 | if (ctx->packet != NULL) { | |
6243 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6244 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
6245 | "%s, which is reserved exclusively for mirroring", | |
6246 | ctx->ofproto->up.name, in_bundle->name); | |
6247 | } | |
479df176 | 6248 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
6249 | return; |
6250 | } | |
6251 | ||
6252 | /* Check VLAN. */ | |
6253 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
6254 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
479df176 | 6255 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
6256 | return; |
6257 | } | |
6258 | vlan = input_vid_to_vlan(in_bundle, vid); | |
6259 | ||
6260 | /* Check other admissibility requirements. */ | |
479df176 | 6261 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 6262 | return; |
abe529af BP |
6263 | } |
6264 | ||
75a75043 | 6265 | /* Learn source MAC. */ |
3de9590b | 6266 | if (ctx->may_learn) { |
abe529af BP |
6267 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
6268 | } | |
6269 | ||
6270 | /* Determine output bundle. */ | |
6271 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
6272 | &ctx->tags); | |
6273 | if (mac) { | |
c06bba01 | 6274 | if (mac->port.p != in_bundle) { |
479df176 | 6275 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 6276 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
6277 | } else { |
6278 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 6279 | } |
abe529af | 6280 | } else { |
c06bba01 | 6281 | struct ofbundle *bundle; |
abe529af | 6282 | |
479df176 | 6283 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
6284 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
6285 | if (bundle != in_bundle | |
6286 | && ofbundle_includes_vlan(bundle, vlan) | |
6287 | && bundle->floodable | |
6288 | && !bundle->mirror_out) { | |
6289 | output_normal(ctx, bundle, vlan); | |
6290 | } | |
6291 | } | |
6292 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 6293 | } |
abe529af BP |
6294 | } |
6295 | \f | |
54a9cbc9 BP |
6296 | /* Optimized flow revalidation. |
6297 | * | |
6298 | * It's a difficult problem, in general, to tell which facets need to have | |
6299 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
6300 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
6301 | * table changes, we recalculate the actions for every facet. This is | |
6302 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
6303 | * doesn't change very often. | |
6304 | * | |
6305 | * However, we can expect one particular kind of OpenFlow flow table change to | |
6306 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
6307 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
6308 | * table, we add a special case that applies to flow tables in which every rule | |
6309 | * has the same form (that is, the same wildcards), except that the table is | |
6310 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
6311 | * optimize this case by tagging all of the facets that resubmit into the table | |
6312 | * and invalidating the same tag whenever a flow changes in that table. The | |
6313 | * end result is that we revalidate just the facets that need it (and sometimes | |
6314 | * a few more, but not all of the facets or even all of the facets that | |
6315 | * resubmit to the table modified by MAC learning). */ | |
6316 | ||
6317 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
6318 | * into an OpenFlow table with the given 'basis'. */ | |
822d9414 | 6319 | static tag_type |
54a9cbc9 BP |
6320 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, |
6321 | uint32_t secret) | |
6322 | { | |
6323 | if (flow_wildcards_is_catchall(wc)) { | |
6324 | return 0; | |
6325 | } else { | |
6326 | struct flow tag_flow = *flow; | |
6327 | flow_zero_wildcards(&tag_flow, wc); | |
6328 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
6329 | } | |
6330 | } | |
6331 | ||
6332 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
6333 | * taggability of that table. | |
6334 | * | |
6335 | * This function must be called after *each* change to a flow table. If you | |
6336 | * skip calling it on some changes then the pointer comparisons at the end can | |
6337 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
6338 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
6339 | * different wildcards to be created with the same address, then this function | |
6340 | * will incorrectly skip revalidation. */ | |
6341 | static void | |
6342 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
6343 | { | |
6344 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 6345 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
6346 | struct cls_table *catchall, *other; |
6347 | struct cls_table *t; | |
6348 | ||
6349 | catchall = other = NULL; | |
6350 | ||
d0918789 | 6351 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
6352 | case 0: |
6353 | /* We could tag this OpenFlow table but it would make the logic a | |
6354 | * little harder and it's a corner case that doesn't seem worth it | |
6355 | * yet. */ | |
6356 | break; | |
6357 | ||
6358 | case 1: | |
6359 | case 2: | |
d0918789 | 6360 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
6361 | if (cls_table_is_catchall(t)) { |
6362 | catchall = t; | |
6363 | } else if (!other) { | |
6364 | other = t; | |
6365 | } else { | |
6366 | /* Indicate that we can't tag this by setting both tables to | |
6367 | * NULL. (We know that 'catchall' is already NULL.) */ | |
6368 | other = NULL; | |
6369 | } | |
6370 | } | |
6371 | break; | |
6372 | ||
6373 | default: | |
6374 | /* Can't tag this table. */ | |
6375 | break; | |
6376 | } | |
6377 | ||
6378 | if (table->catchall_table != catchall || table->other_table != other) { | |
6379 | table->catchall_table = catchall; | |
6380 | table->other_table = other; | |
3c4a309c | 6381 | ofproto->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
6382 | } |
6383 | } | |
6384 | ||
6385 | /* Given 'rule' that has changed in some way (either it is a rule being | |
6386 | * inserted, a rule being deleted, or a rule whose actions are being | |
6387 | * modified), marks facets for revalidation to ensure that packets will be | |
6388 | * forwarded correctly according to the new state of the flow table. | |
6389 | * | |
6390 | * This function must be called after *each* change to a flow table. See | |
6391 | * the comment on table_update_taggable() for more information. */ | |
6392 | static void | |
6393 | rule_invalidate(const struct rule_dpif *rule) | |
6394 | { | |
6395 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
6396 | ||
6397 | table_update_taggable(ofproto, rule->up.table_id); | |
6398 | ||
6399 | if (!ofproto->need_revalidate) { | |
6400 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
6401 | ||
6402 | if (table->other_table && rule->tag) { | |
6403 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
6404 | } else { | |
3c4a309c | 6405 | ofproto->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
6406 | } |
6407 | } | |
6408 | } | |
6409 | \f | |
abe529af | 6410 | static bool |
7257b535 BP |
6411 | set_frag_handling(struct ofproto *ofproto_, |
6412 | enum ofp_config_flags frag_handling) | |
abe529af BP |
6413 | { |
6414 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 6415 | |
7257b535 | 6416 | if (frag_handling != OFPC_FRAG_REASM) { |
3c4a309c | 6417 | ofproto->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
6418 | return true; |
6419 | } else { | |
6420 | return false; | |
6421 | } | |
abe529af BP |
6422 | } |
6423 | ||
90bf1e07 | 6424 | static enum ofperr |
abe529af BP |
6425 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
6426 | const struct flow *flow, | |
f25d0cf3 | 6427 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
6428 | { |
6429 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
548de4dd BP |
6430 | struct odputil_keybuf keybuf; |
6431 | struct dpif_flow_stats stats; | |
abe529af | 6432 | |
548de4dd | 6433 | struct ofpbuf key; |
112bc5f4 | 6434 | |
548de4dd BP |
6435 | struct action_xlate_ctx ctx; |
6436 | uint64_t odp_actions_stub[1024 / 8]; | |
6437 | struct ofpbuf odp_actions; | |
80e5eed9 | 6438 | |
548de4dd BP |
6439 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
6440 | odp_flow_key_from_flow(&key, flow); | |
050ac423 | 6441 | |
548de4dd | 6442 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 6443 | |
548de4dd BP |
6444 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, |
6445 | packet_get_tcp_flags(packet, flow), packet); | |
6446 | ctx.resubmit_stats = &stats; | |
2284188b | 6447 | |
548de4dd BP |
6448 | ofpbuf_use_stub(&odp_actions, |
6449 | odp_actions_stub, sizeof odp_actions_stub); | |
6450 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
6451 | dpif_execute(ofproto->dpif, key.data, key.size, | |
6452 | odp_actions.data, odp_actions.size, packet); | |
6453 | ofpbuf_uninit(&odp_actions); | |
2284188b | 6454 | |
548de4dd | 6455 | return 0; |
abe529af | 6456 | } |
6fca1ffb BP |
6457 | \f |
6458 | /* NetFlow. */ | |
6459 | ||
6460 | static int | |
6461 | set_netflow(struct ofproto *ofproto_, | |
6462 | const struct netflow_options *netflow_options) | |
6463 | { | |
6464 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6465 | ||
6466 | if (netflow_options) { | |
6467 | if (!ofproto->netflow) { | |
6468 | ofproto->netflow = netflow_create(); | |
6469 | } | |
6470 | return netflow_set_options(ofproto->netflow, netflow_options); | |
6471 | } else { | |
6472 | netflow_destroy(ofproto->netflow); | |
6473 | ofproto->netflow = NULL; | |
6474 | return 0; | |
6475 | } | |
6476 | } | |
abe529af BP |
6477 | |
6478 | static void | |
6479 | get_netflow_ids(const struct ofproto *ofproto_, | |
6480 | uint8_t *engine_type, uint8_t *engine_id) | |
6481 | { | |
6482 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6483 | ||
6484 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
6485 | } | |
6fca1ffb BP |
6486 | |
6487 | static void | |
6488 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
6489 | { | |
6490 | if (!facet_is_controller_flow(facet) && | |
6491 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 6492 | struct subfacet *subfacet; |
6fca1ffb BP |
6493 | struct ofexpired expired; |
6494 | ||
b0f7b9b5 | 6495 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 6496 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 6497 | struct dpif_flow_stats stats; |
6fca1ffb | 6498 | |
6a7e895f | 6499 | subfacet_reinstall(subfacet, &stats); |
15baa734 | 6500 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 6501 | } |
6fca1ffb BP |
6502 | } |
6503 | ||
6504 | expired.flow = facet->flow; | |
6505 | expired.packet_count = facet->packet_count; | |
6506 | expired.byte_count = facet->byte_count; | |
6507 | expired.used = facet->used; | |
6508 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
6509 | } | |
6510 | } | |
6511 | ||
6512 | static void | |
6513 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
6514 | { | |
6515 | struct facet *facet; | |
6516 | ||
6517 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6518 | send_active_timeout(ofproto, facet); | |
6519 | } | |
6520 | } | |
abe529af BP |
6521 | \f |
6522 | static struct ofproto_dpif * | |
6523 | ofproto_dpif_lookup(const char *name) | |
6524 | { | |
b44a10b7 BP |
6525 | struct ofproto_dpif *ofproto; |
6526 | ||
6527 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
6528 | hash_string(name, 0), &all_ofproto_dpifs) { | |
6529 | if (!strcmp(ofproto->up.name, name)) { | |
6530 | return ofproto; | |
6531 | } | |
6532 | } | |
6533 | return NULL; | |
abe529af BP |
6534 | } |
6535 | ||
f0a3aa2e | 6536 | static void |
96e466a3 | 6537 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 6538 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 6539 | { |
490df1ef | 6540 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 6541 | |
96e466a3 EJ |
6542 | if (argc > 1) { |
6543 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6544 | if (!ofproto) { | |
bde9f75d | 6545 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
6546 | return; |
6547 | } | |
d0040604 | 6548 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 EJ |
6549 | } else { |
6550 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
d0040604 | 6551 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 | 6552 | } |
f0a3aa2e | 6553 | } |
f0a3aa2e | 6554 | |
bde9f75d | 6555 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
6556 | } |
6557 | ||
abe529af | 6558 | static void |
0e15264f BP |
6559 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6560 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
6561 | { |
6562 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6563 | const struct ofproto_dpif *ofproto; | |
6564 | const struct mac_entry *e; | |
6565 | ||
0e15264f | 6566 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 6567 | if (!ofproto) { |
bde9f75d | 6568 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
6569 | return; |
6570 | } | |
6571 | ||
6572 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
6573 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
6574 | struct ofbundle *bundle = e->port.p; | |
6575 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
6576 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
6577 | e->vlan, ETH_ADDR_ARGS(e->mac), |
6578 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 6579 | } |
bde9f75d | 6580 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
6581 | ds_destroy(&ds); |
6582 | } | |
6583 | ||
6a6455e5 | 6584 | struct trace_ctx { |
abe529af BP |
6585 | struct action_xlate_ctx ctx; |
6586 | struct flow flow; | |
6587 | struct ds *result; | |
6588 | }; | |
6589 | ||
6590 | static void | |
29901626 BP |
6591 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
6592 | const struct rule_dpif *rule) | |
abe529af BP |
6593 | { |
6594 | ds_put_char_multiple(result, '\t', level); | |
6595 | if (!rule) { | |
6596 | ds_put_cstr(result, "No match\n"); | |
6597 | return; | |
6598 | } | |
6599 | ||
29901626 BP |
6600 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
6601 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 6602 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
6603 | ds_put_char(result, '\n'); |
6604 | ||
6605 | ds_put_char_multiple(result, '\t', level); | |
6606 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 6607 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
6608 | ds_put_char(result, '\n'); |
6609 | } | |
6610 | ||
6611 | static void | |
6612 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 6613 | struct trace_ctx *trace) |
abe529af BP |
6614 | { |
6615 | ds_put_char_multiple(result, '\t', level); | |
6616 | ds_put_format(result, "%s: ", title); | |
6617 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
6618 | ds_put_cstr(result, "unchanged"); | |
6619 | } else { | |
6620 | flow_format(result, &trace->ctx.flow); | |
6621 | trace->flow = trace->ctx.flow; | |
6622 | } | |
6623 | ds_put_char(result, '\n'); | |
6624 | } | |
6625 | ||
eb9e1c26 EJ |
6626 | static void |
6627 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 6628 | struct trace_ctx *trace) |
eb9e1c26 EJ |
6629 | { |
6630 | size_t i; | |
6631 | ||
6632 | ds_put_char_multiple(result, '\t', level); | |
6633 | ds_put_format(result, "%s:", title); | |
6634 | for (i = 0; i < FLOW_N_REGS; i++) { | |
6635 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
6636 | } | |
6637 | ds_put_char(result, '\n'); | |
6638 | } | |
6639 | ||
1ed8d352 EJ |
6640 | static void |
6641 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 6642 | struct trace_ctx *trace) |
1ed8d352 EJ |
6643 | { |
6644 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
6645 | ||
6646 | ds_put_char_multiple(result, '\t', level); | |
6647 | ds_put_format(result, "%s: ", title); | |
6648 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
6649 | ds_put_char(result, '\n'); | |
6650 | } | |
6651 | ||
abe529af BP |
6652 | static void |
6653 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6654 | { | |
6a6455e5 | 6655 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
6656 | struct ds *result = trace->result; |
6657 | ||
6658 | ds_put_char(result, '\n'); | |
6659 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 6660 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 6661 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 6662 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
6663 | } |
6664 | ||
479df176 BP |
6665 | static void |
6666 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
6667 | { | |
6668 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
6669 | struct ds *result = trace->result; | |
6670 | ||
6671 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
6672 | ds_put_cstr(result, s); | |
6673 | ds_put_char(result, '\n'); | |
6674 | } | |
6675 | ||
abe529af | 6676 | static void |
0e15264f | 6677 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
6678 | void *aux OVS_UNUSED) |
6679 | { | |
0e15264f | 6680 | const char *dpname = argv[1]; |
abe529af | 6681 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
6682 | struct ofpbuf odp_key; |
6683 | struct ofpbuf *packet; | |
e84173dc | 6684 | ovs_be16 initial_tci; |
abe529af BP |
6685 | struct ds result; |
6686 | struct flow flow; | |
abe529af BP |
6687 | char *s; |
6688 | ||
876b0e1c BP |
6689 | packet = NULL; |
6690 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
6691 | ds_init(&result); |
6692 | ||
e84173dc BP |
6693 | ofproto = ofproto_dpif_lookup(dpname); |
6694 | if (!ofproto) { | |
bde9f75d EJ |
6695 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
6696 | "for help)"); | |
e84173dc BP |
6697 | goto exit; |
6698 | } | |
0e15264f | 6699 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 6700 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
6701 | const char *flow_s = argv[2]; |
6702 | const char *generate_s = argv[3]; | |
876b0e1c | 6703 | |
31a19d69 BP |
6704 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like |
6705 | * flow. We guess which type it is based on whether 'flow_s' contains | |
6706 | * an '(', since a datapath flow always contains '(') but an | |
6707 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
6708 | * that's not documented anywhere). | |
6709 | * | |
6710 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
6711 | * it would be tricky giving a sensible error message. After all, do | |
6712 | * you just say "syntax error" or do you present both error messages? | |
6713 | * Both choices seem lousy. */ | |
6714 | if (strchr(flow_s, '(')) { | |
6715 | int error; | |
6716 | ||
6717 | /* Convert string to datapath key. */ | |
6718 | ofpbuf_init(&odp_key, 0); | |
6719 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
6720 | if (error) { | |
6721 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
6722 | goto exit; | |
6723 | } | |
876b0e1c | 6724 | |
31a19d69 BP |
6725 | /* Convert odp_key to flow. */ |
6726 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, | |
6727 | odp_key.size, &flow, | |
6728 | &initial_tci, NULL); | |
6729 | if (error == ODP_FIT_ERROR) { | |
6730 | unixctl_command_reply_error(conn, "Invalid flow"); | |
6731 | goto exit; | |
6732 | } | |
6733 | } else { | |
6734 | char *error_s; | |
6735 | ||
6736 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
6737 | if (error_s) { | |
6738 | unixctl_command_reply_error(conn, error_s); | |
6739 | free(error_s); | |
6740 | goto exit; | |
6741 | } | |
6742 | ||
6743 | initial_tci = flow.vlan_tci; | |
6744 | vsp_adjust_flow(ofproto, &flow); | |
876b0e1c | 6745 | } |
8b3b8dd1 BP |
6746 | |
6747 | /* Generate a packet, if requested. */ | |
0e15264f | 6748 | if (generate_s) { |
8b3b8dd1 BP |
6749 | packet = ofpbuf_new(0); |
6750 | flow_compose(packet, &flow); | |
6751 | } | |
0e15264f | 6752 | } else if (argc == 6) { |
abff858b | 6753 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
6754 | const char *priority_s = argv[2]; |
6755 | const char *tun_id_s = argv[3]; | |
6756 | const char *in_port_s = argv[4]; | |
6757 | const char *packet_s = argv[5]; | |
6758 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
6759 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
6760 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 6761 | const char *msg; |
0e15264f | 6762 | |
e22f1753 BP |
6763 | msg = eth_from_hex(packet_s, &packet); |
6764 | if (msg) { | |
bde9f75d | 6765 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
6766 | goto exit; |
6767 | } | |
6768 | ||
6769 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 6770 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
6771 | ds_put_cstr(&result, s); |
6772 | free(s); | |
6773 | ||
abff858b | 6774 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 6775 | initial_tci = flow.vlan_tci; |
876b0e1c | 6776 | } else { |
bde9f75d | 6777 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
6778 | goto exit; |
6779 | } | |
6780 | ||
6a6455e5 EJ |
6781 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); |
6782 | unixctl_command_reply(conn, ds_cstr(&result)); | |
6783 | ||
6784 | exit: | |
6785 | ds_destroy(&result); | |
6786 | ofpbuf_delete(packet); | |
6787 | ofpbuf_uninit(&odp_key); | |
6788 | } | |
6789 | ||
6790 | static void | |
6791 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
6792 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
6793 | struct ds *ds) | |
6794 | { | |
6795 | struct rule_dpif *rule; | |
6796 | ||
6797 | ds_put_cstr(ds, "Flow: "); | |
6798 | flow_format(ds, flow); | |
6799 | ds_put_char(ds, '\n'); | |
abe529af | 6800 | |
c57b2226 BP |
6801 | rule = rule_dpif_lookup(ofproto, flow); |
6802 | ||
6a6455e5 | 6803 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
6804 | if (rule == ofproto->miss_rule) { |
6805 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
6806 | } else if (rule == ofproto->no_packet_in_rule) { | |
6807 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
6808 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
6809 | } | |
6810 | ||
abe529af | 6811 | if (rule) { |
050ac423 BP |
6812 | uint64_t odp_actions_stub[1024 / 8]; |
6813 | struct ofpbuf odp_actions; | |
6814 | ||
6a6455e5 | 6815 | struct trace_ctx trace; |
0e553d9c | 6816 | uint8_t tcp_flags; |
abe529af | 6817 | |
6a6455e5 EJ |
6818 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
6819 | trace.result = ds; | |
6820 | trace.flow = *flow; | |
050ac423 BP |
6821 | ofpbuf_use_stub(&odp_actions, |
6822 | odp_actions_stub, sizeof odp_actions_stub); | |
6a6455e5 | 6823 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, |
0e553d9c | 6824 | rule, tcp_flags, packet); |
abe529af | 6825 | trace.ctx.resubmit_hook = trace_resubmit; |
479df176 | 6826 | trace.ctx.report_hook = trace_report; |
f25d0cf3 | 6827 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 6828 | &odp_actions); |
abe529af | 6829 | |
6a6455e5 EJ |
6830 | ds_put_char(ds, '\n'); |
6831 | trace_format_flow(ds, 0, "Final flow", &trace); | |
6832 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
6833 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
6834 | ofpbuf_uninit(&odp_actions); | |
876b0e1c | 6835 | |
6a7e895f BP |
6836 | if (trace.ctx.slow) { |
6837 | enum slow_path_reason slow; | |
6838 | ||
6839 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
6840 | "slow path because it:"); | |
6841 | for (slow = trace.ctx.slow; slow; ) { | |
6842 | enum slow_path_reason bit = rightmost_1bit(slow); | |
6843 | ||
6844 | switch (bit) { | |
6845 | case SLOW_CFM: | |
6846 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
6847 | break; | |
6848 | case SLOW_LACP: | |
6849 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
6850 | break; | |
6851 | case SLOW_STP: | |
6852 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
6853 | break; | |
6854 | case SLOW_IN_BAND: | |
6855 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
6856 | "processing."); | |
6857 | if (!packet) { | |
6858 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
6859 | "incomplete--for complete actions, " | |
6860 | "please supply a packet.)"); | |
6861 | } | |
6862 | break; | |
6863 | case SLOW_CONTROLLER: | |
6864 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
6865 | "to the OpenFlow controller."); | |
6866 | break; | |
6867 | case SLOW_MATCH: | |
6868 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
6869 | "than the datapath supports."); | |
6870 | break; | |
6871 | } | |
6872 | ||
6873 | slow &= ~bit; | |
6874 | } | |
6875 | ||
6876 | if (slow & ~SLOW_MATCH) { | |
6877 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
6878 | "the special slow-path processing."); | |
876b0e1c BP |
6879 | } |
6880 | } | |
abe529af | 6881 | } |
abe529af BP |
6882 | } |
6883 | ||
7ee20df1 | 6884 | static void |
0e15264f BP |
6885 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6886 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6887 | { |
6888 | clogged = true; | |
bde9f75d | 6889 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6890 | } |
6891 | ||
6892 | static void | |
0e15264f BP |
6893 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6894 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6895 | { |
6896 | clogged = false; | |
bde9f75d | 6897 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6898 | } |
6899 | ||
6814e51f BP |
6900 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6901 | * 'reply' describing the results. */ | |
6902 | static void | |
6903 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6904 | { | |
6905 | struct facet *facet; | |
6906 | int errors; | |
6907 | ||
6908 | errors = 0; | |
6909 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6910 | if (!facet_check_consistency(facet)) { | |
6911 | errors++; | |
6912 | } | |
6913 | } | |
6914 | if (errors) { | |
3c4a309c | 6915 | ofproto->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
6916 | } |
6917 | ||
6918 | if (errors) { | |
6919 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6920 | ofproto->up.name, errors); | |
6921 | } else { | |
6922 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6923 | } | |
6924 | } | |
6925 | ||
6926 | static void | |
6927 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6928 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6929 | { | |
6930 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6931 | struct ofproto_dpif *ofproto; | |
6932 | ||
6933 | if (argc > 1) { | |
6934 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6935 | if (!ofproto) { | |
bde9f75d EJ |
6936 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
6937 | "ofproto/list for help)"); | |
6814e51f BP |
6938 | return; |
6939 | } | |
6940 | ofproto_dpif_self_check__(ofproto, &reply); | |
6941 | } else { | |
6942 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6943 | ofproto_dpif_self_check__(ofproto, &reply); | |
6944 | } | |
6945 | } | |
6946 | ||
bde9f75d | 6947 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
6948 | ds_destroy(&reply); |
6949 | } | |
6950 | ||
abe529af BP |
6951 | static void |
6952 | ofproto_dpif_unixctl_init(void) | |
6953 | { | |
6954 | static bool registered; | |
6955 | if (registered) { | |
6956 | return; | |
6957 | } | |
6958 | registered = true; | |
6959 | ||
0e15264f BP |
6960 | unixctl_command_register( |
6961 | "ofproto/trace", | |
6962 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
aa3080c9 | 6963 | 2, 5, ofproto_unixctl_trace, NULL); |
96e466a3 | 6964 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6965 | ofproto_unixctl_fdb_flush, NULL); |
6966 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6967 | ofproto_unixctl_fdb_show, NULL); | |
6968 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6969 | ofproto_dpif_clog, NULL); | |
6970 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6971 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6972 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6973 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6974 | } |
6975 | \f | |
52a90c29 BP |
6976 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6977 | * | |
6978 | * This is deprecated. It is only for compatibility with broken device drivers | |
6979 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6980 | * devices are not used. When broken device drivers are no longer in | |
6981 | * widespread use, we will delete these interfaces. */ | |
6982 | ||
6983 | static int | |
6984 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6985 | { | |
6986 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6987 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6988 | ||
6989 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6990 | && vid == ofport->vlandev_vid) { | |
6991 | return 0; | |
6992 | } | |
6993 | ||
3c4a309c | 6994 | ofproto->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
6995 | |
6996 | if (ofport->realdev_ofp_port) { | |
6997 | vsp_remove(ofport); | |
6998 | } | |
6999 | if (realdev_ofp_port && ofport->bundle) { | |
7000 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
7001 | * themselves be part of a bundle. */ | |
7002 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
7003 | } | |
7004 | ||
7005 | ofport->realdev_ofp_port = realdev_ofp_port; | |
7006 | ofport->vlandev_vid = vid; | |
7007 | ||
7008 | if (realdev_ofp_port) { | |
7009 | vsp_add(ofport, realdev_ofp_port, vid); | |
7010 | } | |
7011 | ||
7012 | return 0; | |
7013 | } | |
7014 | ||
7015 | static uint32_t | |
7016 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
7017 | { | |
7018 | return hash_2words(realdev_ofp_port, vid); | |
7019 | } | |
7020 | ||
40e05935 BP |
7021 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
7022 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
7023 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
7024 | * it would return the port number of eth0.9. | |
7025 | * | |
7026 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
7027 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
7028 | static uint32_t |
7029 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
7030 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
7031 | { | |
7032 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
7033 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
7034 | int vid = vlan_tci_to_vid(vlan_tci); | |
7035 | const struct vlan_splinter *vsp; | |
7036 | ||
7037 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
7038 | hash_realdev_vid(realdev_ofp_port, vid), | |
7039 | &ofproto->realdev_vid_map) { | |
7040 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
7041 | && vsp->vid == vid) { | |
7042 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
7043 | } | |
7044 | } | |
7045 | } | |
7046 | return realdev_odp_port; | |
7047 | } | |
7048 | ||
7049 | static struct vlan_splinter * | |
7050 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
7051 | { | |
7052 | struct vlan_splinter *vsp; | |
7053 | ||
7054 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
7055 | &ofproto->vlandev_map) { | |
7056 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
7057 | return vsp; | |
7058 | } | |
7059 | } | |
7060 | ||
7061 | return NULL; | |
7062 | } | |
7063 | ||
40e05935 BP |
7064 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
7065 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
7066 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
7067 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
7068 | * eth0 and store 9 in '*vid'. | |
7069 | * | |
7070 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
7071 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
7072 | * always does.*/ | |
52a90c29 BP |
7073 | static uint16_t |
7074 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 7075 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
7076 | { |
7077 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
7078 | const struct vlan_splinter *vsp; | |
7079 | ||
7080 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
7081 | if (vsp) { | |
7082 | if (vid) { | |
7083 | *vid = vsp->vid; | |
7084 | } | |
7085 | return vsp->realdev_ofp_port; | |
7086 | } | |
7087 | } | |
7088 | return 0; | |
7089 | } | |
7090 | ||
b98d8985 BP |
7091 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
7092 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
7093 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
7094 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
7095 | * always the case unless VLAN splinters are enabled), returns false without | |
7096 | * making any changes. */ | |
7097 | static bool | |
7098 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
7099 | { | |
7100 | uint16_t realdev; | |
7101 | int vid; | |
7102 | ||
7103 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
7104 | if (!realdev) { | |
7105 | return false; | |
7106 | } | |
7107 | ||
7108 | /* Cause the flow to be processed as if it came in on the real device with | |
7109 | * the VLAN device's VLAN ID. */ | |
7110 | flow->in_port = realdev; | |
7111 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
7112 | return true; | |
7113 | } | |
7114 | ||
52a90c29 BP |
7115 | static void |
7116 | vsp_remove(struct ofport_dpif *port) | |
7117 | { | |
7118 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7119 | struct vlan_splinter *vsp; | |
7120 | ||
7121 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
7122 | if (vsp) { | |
7123 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
7124 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
7125 | free(vsp); | |
7126 | ||
7127 | port->realdev_ofp_port = 0; | |
7128 | } else { | |
7129 | VLOG_ERR("missing vlan device record"); | |
7130 | } | |
7131 | } | |
7132 | ||
7133 | static void | |
7134 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
7135 | { | |
7136 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7137 | ||
7138 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
7139 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
7140 | == realdev_ofp_port)) { | |
7141 | struct vlan_splinter *vsp; | |
7142 | ||
7143 | vsp = xmalloc(sizeof *vsp); | |
7144 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
7145 | hash_int(port->up.ofp_port, 0)); | |
7146 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
7147 | hash_realdev_vid(realdev_ofp_port, vid)); | |
7148 | vsp->realdev_ofp_port = realdev_ofp_port; | |
7149 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
7150 | vsp->vid = vid; | |
7151 | ||
7152 | port->realdev_ofp_port = realdev_ofp_port; | |
7153 | } else { | |
7154 | VLOG_ERR("duplicate vlan device record"); | |
7155 | } | |
7156 | } | |
7157 | \f | |
abe529af BP |
7158 | const struct ofproto_class ofproto_dpif_class = { |
7159 | enumerate_types, | |
7160 | enumerate_names, | |
7161 | del, | |
7162 | alloc, | |
7163 | construct, | |
7164 | destruct, | |
7165 | dealloc, | |
7166 | run, | |
5fcc0d00 | 7167 | run_fast, |
abe529af | 7168 | wait, |
0d085684 | 7169 | get_memory_usage, |
abe529af | 7170 | flush, |
6c1491fb BP |
7171 | get_features, |
7172 | get_tables, | |
abe529af BP |
7173 | port_alloc, |
7174 | port_construct, | |
7175 | port_destruct, | |
7176 | port_dealloc, | |
7177 | port_modified, | |
7178 | port_reconfigured, | |
7179 | port_query_by_name, | |
7180 | port_add, | |
7181 | port_del, | |
6527c598 | 7182 | port_get_stats, |
abe529af BP |
7183 | port_dump_start, |
7184 | port_dump_next, | |
7185 | port_dump_done, | |
7186 | port_poll, | |
7187 | port_poll_wait, | |
7188 | port_is_lacp_current, | |
0ab6decf | 7189 | NULL, /* rule_choose_table */ |
abe529af BP |
7190 | rule_alloc, |
7191 | rule_construct, | |
7192 | rule_destruct, | |
7193 | rule_dealloc, | |
abe529af BP |
7194 | rule_get_stats, |
7195 | rule_execute, | |
7196 | rule_modify_actions, | |
7257b535 | 7197 | set_frag_handling, |
abe529af BP |
7198 | packet_out, |
7199 | set_netflow, | |
7200 | get_netflow_ids, | |
7201 | set_sflow, | |
7202 | set_cfm, | |
a5610457 | 7203 | get_cfm_fault, |
1c0333b6 | 7204 | get_cfm_opup, |
1de11730 | 7205 | get_cfm_remote_mpids, |
3967a833 | 7206 | get_cfm_health, |
21f7563c JP |
7207 | set_stp, |
7208 | get_stp_status, | |
7209 | set_stp_port, | |
7210 | get_stp_port_status, | |
8b36f51e | 7211 | set_queues, |
abe529af BP |
7212 | bundle_set, |
7213 | bundle_remove, | |
7214 | mirror_set, | |
9d24de3b | 7215 | mirror_get_stats, |
abe529af BP |
7216 | set_flood_vlans, |
7217 | is_mirror_output_bundle, | |
8402c74b | 7218 | forward_bpdu_changed, |
e764773c | 7219 | set_mac_idle_time, |
52a90c29 | 7220 | set_realdev, |
abe529af | 7221 | }; |