]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e09ee259 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
abe529af | 23 | #include "bond.h" |
daff3353 | 24 | #include "bundle.h" |
abe529af BP |
25 | #include "byte-order.h" |
26 | #include "connmgr.h" | |
27 | #include "coverage.h" | |
28 | #include "cfm.h" | |
29 | #include "dpif.h" | |
30 | #include "dynamic-string.h" | |
31 | #include "fail-open.h" | |
32 | #include "hmapx.h" | |
33 | #include "lacp.h" | |
75a75043 | 34 | #include "learn.h" |
abe529af | 35 | #include "mac-learning.h" |
816fd533 | 36 | #include "meta-flow.h" |
abe529af | 37 | #include "multipath.h" |
0a740f48 | 38 | #include "netdev-vport.h" |
abe529af BP |
39 | #include "netdev.h" |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
f25d0cf3 | 45 | #include "ofp-actions.h" |
31a19d69 | 46 | #include "ofp-parse.h" |
abe529af | 47 | #include "ofp-print.h" |
9d6ac44e | 48 | #include "ofproto-dpif-governor.h" |
bae473fe | 49 | #include "ofproto-dpif-sflow.h" |
abe529af | 50 | #include "poll-loop.h" |
0d085684 | 51 | #include "simap.h" |
27022416 | 52 | #include "smap.h" |
abe529af | 53 | #include "timer.h" |
b9ad7294 | 54 | #include "tunnel.h" |
6c1491fb | 55 | #include "unaligned.h" |
abe529af BP |
56 | #include "unixctl.h" |
57 | #include "vlan-bitmap.h" | |
58 | #include "vlog.h" | |
59 | ||
60 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
61 | ||
abe529af | 62 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
63 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
64 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
65 | COVERAGE_DEFINE(facet_revalidate); |
66 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 67 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 68 | |
29901626 | 69 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 70 | * flow translation. */ |
1642690c | 71 | #define MAX_RESUBMIT_RECURSION 64 |
abe529af | 72 | |
9cdaaebe BP |
73 | /* Number of implemented OpenFlow tables. */ |
74 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
75 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
76 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 77 | |
abe529af BP |
78 | struct ofport_dpif; |
79 | struct ofproto_dpif; | |
a088a1ff | 80 | struct flow_miss; |
ac35f9c8 | 81 | struct facet; |
abe529af BP |
82 | |
83 | struct rule_dpif { | |
84 | struct rule up; | |
85 | ||
abe529af BP |
86 | /* These statistics: |
87 | * | |
88 | * - Do include packets and bytes from facets that have been deleted or | |
89 | * whose own statistics have been folded into the rule. | |
90 | * | |
91 | * - Do include packets and bytes sent "by hand" that were accounted to | |
92 | * the rule without any facet being involved (this is a rare corner | |
93 | * case in rule_execute()). | |
94 | * | |
95 | * - Do not include packet or bytes that can be obtained from any facet's | |
96 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 97 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
98 | */ |
99 | uint64_t packet_count; /* Number of packets received. */ | |
100 | uint64_t byte_count; /* Number of bytes received. */ | |
101 | ||
54a9cbc9 BP |
102 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
103 | ||
abe529af BP |
104 | struct list facets; /* List of "struct facet"s. */ |
105 | }; | |
106 | ||
107 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
108 | { | |
109 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
110 | } | |
111 | ||
29901626 | 112 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
113 | const struct flow *); |
114 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
115 | const struct flow *, | |
116 | uint8_t table); | |
c376f9a3 IY |
117 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, |
118 | const struct flow *flow); | |
abe529af | 119 | |
112bc5f4 BP |
120 | static void rule_credit_stats(struct rule_dpif *, |
121 | const struct dpif_flow_stats *); | |
ac35f9c8 | 122 | static void flow_push_stats(struct facet *, const struct dpif_flow_stats *); |
822d9414 | 123 | static tag_type rule_calculate_tag(const struct flow *, |
5cb7a798 | 124 | const struct minimask *, uint32_t basis); |
b0f7b9b5 BP |
125 | static void rule_invalidate(const struct rule_dpif *); |
126 | ||
abe529af BP |
127 | #define MAX_MIRRORS 32 |
128 | typedef uint32_t mirror_mask_t; | |
129 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
130 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
131 | struct ofmirror { | |
132 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
133 | size_t idx; /* In ofproto's "mirrors" array. */ | |
134 | void *aux; /* Key supplied by ofproto's client. */ | |
135 | char *name; /* Identifier for log messages. */ | |
136 | ||
137 | /* Selection criteria. */ | |
138 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
139 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
140 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
141 | ||
9ba15e2a | 142 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
143 | struct ofbundle *out; /* Output port or NULL. */ |
144 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 145 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
146 | |
147 | /* Counters. */ | |
148 | int64_t packet_count; /* Number of packets sent. */ | |
149 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
150 | }; |
151 | ||
152 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
153 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
154 | mirror_mask_t mirrors, | |
155 | uint64_t packets, uint64_t bytes); | |
abe529af | 156 | |
abe529af | 157 | struct ofbundle { |
abe529af | 158 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 159 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
160 | void *aux; /* Key supplied by ofproto's client. */ |
161 | char *name; /* Identifier for log messages. */ | |
162 | ||
163 | /* Configuration. */ | |
164 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 165 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
166 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
167 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
168 | * NULL if all VLANs are trunked. */ | |
169 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
170 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 171 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
172 | |
173 | /* Status. */ | |
9e1fd49b | 174 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
175 | |
176 | /* Port mirroring info. */ | |
177 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
178 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
179 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
180 | }; | |
181 | ||
182 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 183 | static void bundle_update(struct ofbundle *); |
abe529af BP |
184 | static void bundle_destroy(struct ofbundle *); |
185 | static void bundle_del_port(struct ofport_dpif *); | |
186 | static void bundle_run(struct ofbundle *); | |
187 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 188 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
189 | uint16_t in_port, bool warn, |
190 | struct ofport_dpif **in_ofportp); | |
abe529af | 191 | |
33158a18 JP |
192 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
193 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
194 | * when an input bundle is needed for validation (e.g., mirroring or | |
195 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
196 | * any 'port' structs, so care must be taken when dealing with it. */ | |
197 | static struct ofbundle ofpp_none_bundle = { | |
198 | .name = "OFPP_NONE", | |
199 | .vlan_mode = PORT_VLAN_TRUNK | |
200 | }; | |
201 | ||
21f7563c JP |
202 | static void stp_run(struct ofproto_dpif *ofproto); |
203 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
204 | static int set_stp_port(struct ofport *, |
205 | const struct ofproto_port_stp_settings *); | |
21f7563c | 206 | |
5da5ec37 BP |
207 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
208 | ||
abe529af BP |
209 | struct action_xlate_ctx { |
210 | /* action_xlate_ctx_init() initializes these members. */ | |
211 | ||
212 | /* The ofproto. */ | |
213 | struct ofproto_dpif *ofproto; | |
214 | ||
215 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
216 | * this flow when actions change header fields. */ | |
217 | struct flow flow; | |
218 | ||
bd85dac1 AZ |
219 | /* stack for the push and pop actions. |
220 | * Each stack element is of the type "union mf_subvalue". */ | |
221 | struct ofpbuf stack; | |
222 | union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; | |
223 | ||
abe529af BP |
224 | /* The packet corresponding to 'flow', or a null pointer if we are |
225 | * revalidating without a packet to refer to. */ | |
226 | const struct ofpbuf *packet; | |
227 | ||
3de9590b BP |
228 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
229 | * actions update the flow table? | |
230 | * | |
231 | * We want to update these tables if we are actually processing a packet, | |
232 | * or if we are accounting for packets that the datapath has processed, but | |
233 | * not if we are just revalidating. */ | |
234 | bool may_learn; | |
75a75043 | 235 | |
18b2a258 BP |
236 | /* The rule that we are currently translating, or NULL. */ |
237 | struct rule_dpif *rule; | |
54834960 | 238 | |
0e553d9c BP |
239 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
240 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
241 | * timeouts.) */ | |
242 | uint8_t tcp_flags; | |
243 | ||
112bc5f4 BP |
244 | /* If nonnull, flow translation calls this function just before executing a |
245 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
246 | * when the recursion depth is exceeded. | |
247 | * | |
248 | * 'rule' is the rule being submitted into. It will be null if the | |
249 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
250 | * | |
251 | * This is normally null so the client has to set it manually after | |
252 | * calling action_xlate_ctx_init(). */ | |
253 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
254 | ||
479df176 BP |
255 | /* If nonnull, flow translation calls this function to report some |
256 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
257 | * dropped a packet. */ | |
258 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
259 | ||
112bc5f4 BP |
260 | /* If nonnull, flow translation credits the specified statistics to each |
261 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
262 | * |
263 | * This is normally null so the client has to set it manually after | |
264 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 265 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 266 | |
abe529af BP |
267 | /* xlate_actions() initializes and uses these members. The client might want |
268 | * to look at them after it returns. */ | |
269 | ||
270 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 271 | tag_type tags; /* Tags associated with actions. */ |
6a7e895f | 272 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
75a75043 BP |
273 | bool has_learn; /* Actions include NXAST_LEARN? */ |
274 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 275 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 276 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 277 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
278 | |
279 | /* xlate_actions() initializes and uses these members, but the client has no | |
280 | * reason to look at them. */ | |
281 | ||
282 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 283 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 284 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 285 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 286 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 | 287 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
9b56fe13 | 288 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ |
6ff686f2 | 289 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ |
848e8809 | 290 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
291 | }; |
292 | ||
14f94f9a JP |
293 | /* Initial values of fields of the packet that may be changed during |
294 | * flow processing and needed later. */ | |
295 | struct initial_vals { | |
296 | /* This is the value of vlan_tci in the packet as actually received from | |
297 | * dpif. This is the same as the facet's flow.vlan_tci unless the packet | |
298 | * was received via a VLAN splinter. In that case, this value is 0 | |
299 | * (because the packet as actually received from the dpif had no 802.1Q | |
300 | * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter | |
301 | * represents. | |
302 | * | |
303 | * This member should be removed when the VLAN splinters feature is no | |
304 | * longer needed. */ | |
305 | ovs_be16 vlan_tci; | |
c3f6c502 JP |
306 | |
307 | /* If received on a tunnel, the IP TOS value of the tunnel. */ | |
308 | uint8_t tunnel_ip_tos; | |
14f94f9a JP |
309 | }; |
310 | ||
abe529af BP |
311 | static void action_xlate_ctx_init(struct action_xlate_ctx *, |
312 | struct ofproto_dpif *, const struct flow *, | |
14f94f9a JP |
313 | const struct initial_vals *initial_vals, |
314 | struct rule_dpif *, | |
0e553d9c | 315 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 | 316 | static void xlate_actions(struct action_xlate_ctx *, |
f25d0cf3 | 317 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 BP |
318 | struct ofpbuf *odp_actions); |
319 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
f25d0cf3 BP |
320 | const struct ofpact *ofpacts, |
321 | size_t ofpacts_len); | |
0a740f48 EJ |
322 | static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port, |
323 | uint8_t table_id, bool may_packet_in); | |
abe529af | 324 | |
6a7e895f BP |
325 | static size_t put_userspace_action(const struct ofproto_dpif *, |
326 | struct ofpbuf *odp_actions, | |
327 | const struct flow *, | |
328 | const union user_action_cookie *); | |
329 | ||
330 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
331 | enum slow_path_reason, | |
332 | uint64_t *stub, size_t stub_size, | |
333 | const struct nlattr **actionsp, | |
334 | size_t *actions_lenp); | |
335 | ||
479df176 BP |
336 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); |
337 | ||
6a7e895f BP |
338 | /* A subfacet (see "struct subfacet" below) has three possible installation |
339 | * states: | |
340 | * | |
341 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
342 | * case just after the subfacet is created, just before the subfacet is | |
343 | * destroyed, or if the datapath returns an error when we try to install a | |
344 | * subfacet. | |
345 | * | |
346 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
347 | * | |
348 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
349 | * ofproto_dpif is installed in the datapath. | |
350 | */ | |
351 | enum subfacet_path { | |
352 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
353 | SF_FAST_PATH, /* Full actions are installed. */ | |
354 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
355 | }; | |
356 | ||
357 | static const char *subfacet_path_to_string(enum subfacet_path); | |
358 | ||
5f5fbd17 BP |
359 | /* A dpif flow and actions associated with a facet. |
360 | * | |
361 | * See also the large comment on struct facet. */ | |
362 | struct subfacet { | |
363 | /* Owners. */ | |
364 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
365 | struct list list_node; /* In struct facet's 'facets' list. */ | |
366 | struct facet *facet; /* Owning facet. */ | |
367 | ||
5f5fbd17 BP |
368 | enum odp_key_fitness key_fitness; |
369 | struct nlattr *key; | |
370 | int key_len; | |
371 | ||
372 | long long int used; /* Time last used; time created if not used. */ | |
373 | ||
374 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
375 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
376 | ||
377 | /* Datapath actions. | |
378 | * | |
379 | * These should be essentially identical for every subfacet in a facet, but | |
380 | * may differ in trivial ways due to VLAN splinters. */ | |
381 | size_t actions_len; /* Number of bytes in actions[]. */ | |
382 | struct nlattr *actions; /* Datapath actions. */ | |
383 | ||
6a7e895f BP |
384 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
385 | enum subfacet_path path; /* Installed in datapath? */ | |
5f5fbd17 | 386 | |
14f94f9a JP |
387 | /* Initial values of the packet that may be needed later. */ |
388 | struct initial_vals initial_vals; | |
a088a1ff JP |
389 | |
390 | /* Datapath port the packet arrived on. This is needed to remove | |
391 | * flows for ports that are no longer part of the bridge. Since the | |
392 | * flow definition only has the OpenFlow port number and the port is | |
393 | * no longer part of the bridge, we can't determine the datapath port | |
394 | * number needed to delete the flow from the datapath. */ | |
395 | uint32_t odp_in_port; | |
5f5fbd17 BP |
396 | }; |
397 | ||
1d85f9e5 JP |
398 | #define SUBFACET_DESTROY_MAX_BATCH 50 |
399 | ||
a088a1ff | 400 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, |
459b16a1 | 401 | long long int now); |
5f5fbd17 | 402 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
acf60855 | 403 | const struct nlattr *key, size_t key_len, |
9566abf9 | 404 | uint32_t key_hash); |
5f5fbd17 BP |
405 | static void subfacet_destroy(struct subfacet *); |
406 | static void subfacet_destroy__(struct subfacet *); | |
1d85f9e5 JP |
407 | static void subfacet_destroy_batch(struct ofproto_dpif *, |
408 | struct subfacet **, int n); | |
5f5fbd17 BP |
409 | static void subfacet_reset_dp_stats(struct subfacet *, |
410 | struct dpif_flow_stats *); | |
411 | static void subfacet_update_time(struct subfacet *, long long int used); | |
412 | static void subfacet_update_stats(struct subfacet *, | |
413 | const struct dpif_flow_stats *); | |
414 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
415 | const struct ofpbuf *packet, |
416 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
417 | static int subfacet_install(struct subfacet *, |
418 | const struct nlattr *actions, size_t actions_len, | |
6a7e895f | 419 | struct dpif_flow_stats *, enum slow_path_reason); |
5f5fbd17 BP |
420 | static void subfacet_uninstall(struct subfacet *); |
421 | ||
6a7e895f BP |
422 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); |
423 | ||
b0f7b9b5 BP |
424 | /* An exact-match instantiation of an OpenFlow flow. |
425 | * | |
426 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
427 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
428 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
429 | * the facet. When the kernel module (or other dpif implementation) and Open | |
430 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
431 | * one subfacet per facet. If the dpif implementation supports more-specific | |
432 | * flow matching than userspace, however, a facet can have more than one | |
433 | * subfacet, each of which corresponds to some distinction in flow that | |
434 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
435 | * |
436 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
437 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 438 | struct facet { |
b0f7b9b5 BP |
439 | /* Owners. */ |
440 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
441 | struct list list_node; /* In owning rule's 'facets' list. */ | |
442 | struct rule_dpif *rule; /* Owning rule. */ | |
443 | ||
444 | /* Owned data. */ | |
445 | struct list subfacets; | |
abe529af BP |
446 | long long int used; /* Time last used; time created if not used. */ |
447 | ||
b0f7b9b5 BP |
448 | /* Key. */ |
449 | struct flow flow; | |
450 | ||
abe529af BP |
451 | /* These statistics: |
452 | * | |
453 | * - Do include packets and bytes sent "by hand", e.g. with | |
454 | * dpif_execute(). | |
455 | * | |
456 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 457 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 458 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
459 | * |
460 | * - Do not include packets or bytes that can be obtained from the | |
461 | * datapath for any existing subfacet. | |
abe529af BP |
462 | */ |
463 | uint64_t packet_count; /* Number of packets received. */ | |
464 | uint64_t byte_count; /* Number of bytes received. */ | |
465 | ||
b0f7b9b5 | 466 | /* Resubmit statistics. */ |
9d24de3b JP |
467 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
468 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
469 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 470 | |
b0f7b9b5 | 471 | /* Accounting. */ |
907a4c5e | 472 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 473 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 474 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 475 | |
b95fc6ba BP |
476 | /* Properties of datapath actions. |
477 | * | |
478 | * Every subfacet has its own actions because actions can differ slightly | |
479 | * between splintered and non-splintered subfacets due to the VLAN tag | |
480 | * being initially different (present vs. absent). All of them have these | |
481 | * properties in common so we just store one copy of them here. */ | |
75a75043 BP |
482 | bool has_learn; /* Actions include NXAST_LEARN? */ |
483 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 484 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 485 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 486 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
487 | |
488 | /* Storage for a single subfacet, to reduce malloc() time and space | |
489 | * overhead. (A facet always has at least one subfacet and in the common | |
bcd516b7 JP |
490 | * case has exactly one subfacet. However, 'one_subfacet' may not |
491 | * always be valid, since it could have been removed after newer | |
492 | * subfacets were pushed onto the 'subfacets' list.) */ | |
26cd7e34 | 493 | struct subfacet one_subfacet; |
6cf474d7 EJ |
494 | |
495 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
abe529af BP |
496 | }; |
497 | ||
2b459b83 BP |
498 | static struct facet *facet_create(struct rule_dpif *, |
499 | const struct flow *, uint32_t hash); | |
15baa734 | 500 | static void facet_remove(struct facet *); |
abe529af BP |
501 | static void facet_free(struct facet *); |
502 | ||
2b459b83 BP |
503 | static struct facet *facet_find(struct ofproto_dpif *, |
504 | const struct flow *, uint32_t hash); | |
abe529af | 505 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 506 | const struct flow *, uint32_t hash); |
c57b2226 | 507 | static void facet_revalidate(struct facet *); |
6814e51f | 508 | static bool facet_check_consistency(struct facet *); |
abe529af | 509 | |
15baa734 | 510 | static void facet_flush_stats(struct facet *); |
abe529af | 511 | |
15baa734 | 512 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 513 | static void facet_reset_counters(struct facet *); |
abe529af | 514 | static void facet_push_stats(struct facet *); |
3de9590b BP |
515 | static void facet_learn(struct facet *); |
516 | static void facet_account(struct facet *); | |
abe529af | 517 | |
1a1e3a0a JP |
518 | static struct subfacet *facet_get_subfacet(struct facet *); |
519 | ||
abe529af BP |
520 | static bool facet_is_controller_flow(struct facet *); |
521 | ||
abe529af | 522 | struct ofport_dpif { |
acf60855 | 523 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ |
abe529af BP |
524 | struct ofport up; |
525 | ||
526 | uint32_t odp_port; | |
527 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
528 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
529 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
530 | tag_type tag; /* Tag associated with this port. */ | |
015e08bc | 531 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 532 | long long int carrier_seq; /* Carrier status changes. */ |
b9ad7294 | 533 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ |
21f7563c | 534 | |
52a90c29 | 535 | /* Spanning tree. */ |
21f7563c JP |
536 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
537 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
538 | long long int stp_state_entered; | |
8b36f51e EJ |
539 | |
540 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
541 | |
542 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
543 | * | |
544 | * This is deprecated. It is only for compatibility with broken device | |
545 | * drivers in old versions of Linux that do not properly support VLANs when | |
546 | * VLAN devices are not used. When broken device drivers are no longer in | |
547 | * widespread use, we will delete these interfaces. */ | |
548 | uint16_t realdev_ofp_port; | |
549 | int vlandev_vid; | |
8b36f51e EJ |
550 | }; |
551 | ||
552 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
553 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
554 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
555 | struct priority_to_dscp { | |
556 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
557 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
558 | ||
559 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
560 | }; |
561 | ||
52a90c29 BP |
562 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
563 | * | |
564 | * This is deprecated. It is only for compatibility with broken device drivers | |
565 | * in old versions of Linux that do not properly support VLANs when VLAN | |
566 | * devices are not used. When broken device drivers are no longer in | |
567 | * widespread use, we will delete these interfaces. */ | |
568 | struct vlan_splinter { | |
569 | struct hmap_node realdev_vid_node; | |
570 | struct hmap_node vlandev_node; | |
571 | uint16_t realdev_ofp_port; | |
572 | uint16_t vlandev_ofp_port; | |
573 | int vid; | |
574 | }; | |
575 | ||
576 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
577 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 578 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
579 | static void vsp_remove(struct ofport_dpif *); |
580 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
581 | ||
e1b1d06a JP |
582 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, |
583 | uint16_t ofp_port); | |
584 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
585 | uint32_t odp_port); | |
586 | ||
abe529af BP |
587 | static struct ofport_dpif * |
588 | ofport_dpif_cast(const struct ofport *ofport) | |
589 | { | |
cb22974d | 590 | ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
591 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; |
592 | } | |
593 | ||
594 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 595 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 596 | static void port_wait(struct ofport_dpif *); |
a5610457 | 597 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 598 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 599 | |
7ee20df1 BP |
600 | struct dpif_completion { |
601 | struct list list_node; | |
602 | struct ofoperation *op; | |
603 | }; | |
604 | ||
54a9cbc9 BP |
605 | /* Extra information about a classifier table. |
606 | * Currently used just for optimized flow revalidation. */ | |
607 | struct table_dpif { | |
608 | /* If either of these is nonnull, then this table has a form that allows | |
609 | * flows to be tagged to avoid revalidating most flows for the most common | |
610 | * kinds of flow table changes. */ | |
611 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
612 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
613 | uint32_t basis; /* Keeps each table's tags separate. */ | |
614 | }; | |
615 | ||
3c4a309c BP |
616 | /* Reasons that we might need to revalidate every facet, and corresponding |
617 | * coverage counters. | |
618 | * | |
619 | * A value of 0 means that there is no need to revalidate. | |
620 | * | |
621 | * It would be nice to have some cleaner way to integrate with coverage | |
622 | * counters, but with only a few reasons I guess this is good enough for | |
623 | * now. */ | |
624 | enum revalidate_reason { | |
625 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
626 | REV_STP, /* Spanning tree protocol port status change. */ | |
627 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
628 | REV_FLOW_TABLE, /* Flow table changed. */ | |
629 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
630 | }; | |
631 | COVERAGE_DEFINE(rev_reconfigure); | |
632 | COVERAGE_DEFINE(rev_stp); | |
633 | COVERAGE_DEFINE(rev_port_toggled); | |
634 | COVERAGE_DEFINE(rev_flow_table); | |
635 | COVERAGE_DEFINE(rev_inconsistency); | |
636 | ||
8f73d537 EJ |
637 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. |
638 | * These are datapath flows which have no associated ofproto, if they did we | |
639 | * would use facets. */ | |
640 | struct drop_key { | |
641 | struct hmap_node hmap_node; | |
642 | struct nlattr *key; | |
643 | size_t key_len; | |
644 | }; | |
645 | ||
acf60855 JP |
646 | /* All datapaths of a given type share a single dpif backer instance. */ |
647 | struct dpif_backer { | |
648 | char *type; | |
649 | int refcount; | |
650 | struct dpif *dpif; | |
651 | struct timer next_expiration; | |
652 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
2cc3c58e | 653 | |
7d82ab2e | 654 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ |
b9ad7294 | 655 | |
2cc3c58e EJ |
656 | /* Facet revalidation flags applying to facets which use this backer. */ |
657 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
658 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
8f73d537 EJ |
659 | |
660 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
acf60855 JP |
661 | }; |
662 | ||
663 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
664 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
665 | ||
8f73d537 | 666 | static void drop_key_clear(struct dpif_backer *); |
acf60855 JP |
667 | static struct ofport_dpif * |
668 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
669 | ||
735d7efb AZ |
670 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, |
671 | uint64_t delta); | |
672 | ||
abe529af | 673 | struct ofproto_dpif { |
b44a10b7 | 674 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af | 675 | struct ofproto up; |
acf60855 | 676 | struct dpif_backer *backer; |
abe529af | 677 | |
c57b2226 BP |
678 | /* Special OpenFlow rules. */ |
679 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
680 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
681 | ||
6c1491fb BP |
682 | /* Statistics. */ |
683 | uint64_t n_matches; | |
684 | ||
abe529af BP |
685 | /* Bridging. */ |
686 | struct netflow *netflow; | |
bae473fe | 687 | struct dpif_sflow *sflow; |
abe529af BP |
688 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
689 | struct mac_learning *ml; | |
690 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 691 | bool has_mirrors; |
abe529af BP |
692 | bool has_bonded_bundles; |
693 | ||
abe529af BP |
694 | /* Facets. */ |
695 | struct hmap facets; | |
b0f7b9b5 | 696 | struct hmap subfacets; |
9d6ac44e | 697 | struct governor *governor; |
4d2a0f39 | 698 | long long int consistency_rl; |
54a9cbc9 BP |
699 | |
700 | /* Revalidation. */ | |
701 | struct table_dpif tables[N_TABLES]; | |
7ee20df1 BP |
702 | |
703 | /* Support for debugging async flow mods. */ | |
704 | struct list completions; | |
daff3353 EJ |
705 | |
706 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
707 | struct netdev_stats stats; /* To account packets generated and consumed in |
708 | * userspace. */ | |
21f7563c JP |
709 | |
710 | /* Spanning tree. */ | |
711 | struct stp *stp; | |
712 | long long int stp_last_tick; | |
52a90c29 BP |
713 | |
714 | /* VLAN splinters. */ | |
715 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
716 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
e1b1d06a | 717 | |
acf60855 | 718 | /* Ports. */ |
0a740f48 EJ |
719 | struct sset ports; /* Set of standard port names. */ |
720 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
acf60855 JP |
721 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ |
722 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
735d7efb AZ |
723 | |
724 | /* Per ofproto's dpif stats. */ | |
725 | uint64_t n_hit; | |
726 | uint64_t n_missed; | |
abe529af BP |
727 | }; |
728 | ||
7ee20df1 BP |
729 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
730 | * for debugging the asynchronous flow_mod implementation.) */ | |
731 | static bool clogged; | |
732 | ||
b44a10b7 BP |
733 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
734 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
735 | ||
abe529af BP |
736 | static void ofproto_dpif_unixctl_init(void); |
737 | ||
738 | static struct ofproto_dpif * | |
739 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
740 | { | |
cb22974d | 741 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); |
abe529af BP |
742 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); |
743 | } | |
744 | ||
4acbc98d | 745 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 746 | uint16_t ofp_port); |
4acbc98d | 747 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 748 | uint32_t odp_port); |
6a6455e5 | 749 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
14f94f9a JP |
750 | const struct ofpbuf *, |
751 | const struct initial_vals *, struct ds *); | |
abe529af BP |
752 | |
753 | /* Packet processing. */ | |
754 | static void update_learning_table(struct ofproto_dpif *, | |
755 | const struct flow *, int vlan, | |
756 | struct ofbundle *); | |
501f8d1f BP |
757 | /* Upcalls. */ |
758 | #define FLOW_MISS_MAX_BATCH 50 | |
acf60855 | 759 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); |
abe529af BP |
760 | |
761 | /* Flow expiration. */ | |
acf60855 | 762 | static int expire(struct dpif_backer *); |
abe529af | 763 | |
6fca1ffb BP |
764 | /* NetFlow. */ |
765 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
766 | ||
abe529af | 767 | /* Utilities. */ |
52a90c29 | 768 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
769 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
770 | struct ofpbuf *odp_actions, | |
771 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
772 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
773 | const struct flow *flow); | |
abe529af BP |
774 | /* Global variables. */ |
775 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
acf60855 JP |
776 | |
777 | /* Initial mappings of port to bridge mappings. */ | |
778 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
abe529af BP |
779 | \f |
780 | /* Factory functions. */ | |
781 | ||
b0408fca | 782 | static void |
acf60855 | 783 | init(const struct shash *iface_hints) |
b0408fca | 784 | { |
acf60855 JP |
785 | struct shash_node *node; |
786 | ||
787 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
788 | SHASH_FOR_EACH(node, iface_hints) { | |
789 | const struct iface_hint *orig_hint = node->data; | |
790 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
791 | ||
792 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
793 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
794 | new_hint->ofp_port = orig_hint->ofp_port; | |
795 | ||
796 | shash_add(&init_ofp_ports, node->name, new_hint); | |
797 | } | |
b0408fca JP |
798 | } |
799 | ||
abe529af BP |
800 | static void |
801 | enumerate_types(struct sset *types) | |
802 | { | |
803 | dp_enumerate_types(types); | |
804 | } | |
805 | ||
806 | static int | |
807 | enumerate_names(const char *type, struct sset *names) | |
808 | { | |
acf60855 JP |
809 | struct ofproto_dpif *ofproto; |
810 | ||
811 | sset_clear(names); | |
812 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
813 | if (strcmp(type, ofproto->up.type)) { | |
814 | continue; | |
815 | } | |
816 | sset_add(names, ofproto->up.name); | |
817 | } | |
818 | ||
819 | return 0; | |
abe529af BP |
820 | } |
821 | ||
822 | static int | |
823 | del(const char *type, const char *name) | |
824 | { | |
825 | struct dpif *dpif; | |
826 | int error; | |
827 | ||
828 | error = dpif_open(name, type, &dpif); | |
829 | if (!error) { | |
830 | error = dpif_delete(dpif); | |
831 | dpif_close(dpif); | |
832 | } | |
833 | return error; | |
834 | } | |
835 | \f | |
0aeaabc8 JP |
836 | static const char * |
837 | port_open_type(const char *datapath_type, const char *port_type) | |
838 | { | |
839 | return dpif_port_open_type(datapath_type, port_type); | |
840 | } | |
841 | ||
acf60855 JP |
842 | /* Type functions. */ |
843 | ||
476cb42a BP |
844 | static struct ofproto_dpif * |
845 | lookup_ofproto_dpif_by_port_name(const char *name) | |
846 | { | |
847 | struct ofproto_dpif *ofproto; | |
848 | ||
849 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
850 | if (sset_contains(&ofproto->ports, name)) { | |
851 | return ofproto; | |
852 | } | |
853 | } | |
854 | ||
855 | return NULL; | |
856 | } | |
857 | ||
acf60855 JP |
858 | static int |
859 | type_run(const char *type) | |
860 | { | |
861 | struct dpif_backer *backer; | |
862 | char *devname; | |
863 | int error; | |
864 | ||
865 | backer = shash_find_data(&all_dpif_backers, type); | |
866 | if (!backer) { | |
867 | /* This is not necessarily a problem, since backers are only | |
868 | * created on demand. */ | |
869 | return 0; | |
870 | } | |
871 | ||
872 | dpif_run(backer->dpif); | |
873 | ||
2cc3c58e EJ |
874 | if (backer->need_revalidate |
875 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
876 | struct tag_set revalidate_set = backer->revalidate_set; | |
877 | bool need_revalidate = backer->need_revalidate; | |
878 | struct ofproto_dpif *ofproto; | |
a614d823 KM |
879 | struct simap_node *node; |
880 | struct simap tmp_backers; | |
881 | ||
882 | /* Handle tunnel garbage collection. */ | |
883 | simap_init(&tmp_backers); | |
884 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
885 | ||
886 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
887 | struct ofport_dpif *iter; | |
888 | ||
889 | if (backer != ofproto->backer) { | |
890 | continue; | |
891 | } | |
892 | ||
893 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
894 | const char *dp_port; | |
895 | ||
896 | if (!iter->tnl_port) { | |
897 | continue; | |
898 | } | |
899 | ||
900 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev); | |
901 | node = simap_find(&tmp_backers, dp_port); | |
902 | if (node) { | |
903 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
904 | simap_delete(&tmp_backers, node); | |
905 | node = simap_find(&backer->tnl_backers, dp_port); | |
906 | } else { | |
907 | node = simap_find(&backer->tnl_backers, dp_port); | |
908 | if (!node) { | |
909 | uint32_t odp_port = UINT32_MAX; | |
910 | ||
911 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
912 | &odp_port)) { | |
913 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
914 | node = simap_find(&backer->tnl_backers, dp_port); | |
915 | } | |
916 | } | |
917 | } | |
918 | ||
919 | iter->odp_port = node ? node->data : OVSP_NONE; | |
920 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
921 | &iter->tnl_port)) { | |
922 | backer->need_revalidate = REV_RECONFIGURE; | |
923 | } | |
924 | } | |
925 | } | |
926 | ||
927 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
928 | dpif_port_del(backer->dpif, node->data); | |
929 | } | |
930 | simap_destroy(&tmp_backers); | |
2cc3c58e EJ |
931 | |
932 | switch (backer->need_revalidate) { | |
933 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
934 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
935 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
936 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
937 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
938 | } | |
939 | ||
8f73d537 EJ |
940 | if (backer->need_revalidate) { |
941 | /* Clear the drop_keys in case we should now be accepting some | |
942 | * formerly dropped flows. */ | |
943 | drop_key_clear(backer); | |
944 | } | |
945 | ||
f728af2e BP |
946 | /* Clear the revalidation flags. */ |
947 | tag_set_init(&backer->revalidate_set); | |
948 | backer->need_revalidate = 0; | |
949 | ||
2cc3c58e | 950 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
f231418e | 951 | struct facet *facet, *next; |
2cc3c58e EJ |
952 | |
953 | if (ofproto->backer != backer) { | |
954 | continue; | |
955 | } | |
956 | ||
f231418e | 957 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { |
2cc3c58e EJ |
958 | if (need_revalidate |
959 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
960 | facet_revalidate(facet); | |
961 | } | |
962 | } | |
963 | } | |
2cc3c58e EJ |
964 | } |
965 | ||
acf60855 JP |
966 | if (timer_expired(&backer->next_expiration)) { |
967 | int delay = expire(backer); | |
968 | timer_set_duration(&backer->next_expiration, delay); | |
969 | } | |
970 | ||
971 | /* Check for port changes in the dpif. */ | |
972 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
476cb42a | 973 | struct ofproto_dpif *ofproto; |
acf60855 JP |
974 | struct dpif_port port; |
975 | ||
976 | /* Don't report on the datapath's device. */ | |
977 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
c83b89ab | 978 | goto next; |
acf60855 JP |
979 | } |
980 | ||
b9ad7294 EJ |
981 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, |
982 | &all_ofproto_dpifs) { | |
7d82ab2e | 983 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { |
b9ad7294 EJ |
984 | goto next; |
985 | } | |
986 | } | |
987 | ||
476cb42a | 988 | ofproto = lookup_ofproto_dpif_by_port_name(devname); |
acf60855 JP |
989 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { |
990 | /* The port was removed. If we know the datapath, | |
991 | * report it through poll_set(). If we don't, it may be | |
992 | * notifying us of a removal we initiated, so ignore it. | |
993 | * If there's a pending ENOBUFS, let it stand, since | |
994 | * everything will be reevaluated. */ | |
995 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
996 | sset_add(&ofproto->port_poll_set, devname); | |
997 | ofproto->port_poll_errno = 0; | |
998 | } | |
acf60855 JP |
999 | } else if (!ofproto) { |
1000 | /* The port was added, but we don't know with which | |
1001 | * ofproto we should associate it. Delete it. */ | |
1002 | dpif_port_del(backer->dpif, port.port_no); | |
1003 | } | |
5b5e6a4c | 1004 | dpif_port_destroy(&port); |
acf60855 | 1005 | |
c83b89ab | 1006 | next: |
acf60855 JP |
1007 | free(devname); |
1008 | } | |
1009 | ||
1010 | if (error != EAGAIN) { | |
1011 | struct ofproto_dpif *ofproto; | |
1012 | ||
1013 | /* There was some sort of error, so propagate it to all | |
1014 | * ofprotos that use this backer. */ | |
1015 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1016 | &all_ofproto_dpifs) { | |
1017 | if (ofproto->backer == backer) { | |
1018 | sset_clear(&ofproto->port_poll_set); | |
1019 | ofproto->port_poll_errno = error; | |
1020 | } | |
1021 | } | |
1022 | } | |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | static int | |
1028 | type_run_fast(const char *type) | |
1029 | { | |
1030 | struct dpif_backer *backer; | |
1031 | unsigned int work; | |
1032 | ||
1033 | backer = shash_find_data(&all_dpif_backers, type); | |
1034 | if (!backer) { | |
1035 | /* This is not necessarily a problem, since backers are only | |
1036 | * created on demand. */ | |
1037 | return 0; | |
1038 | } | |
1039 | ||
1040 | /* Handle one or more batches of upcalls, until there's nothing left to do | |
1041 | * or until we do a fixed total amount of work. | |
1042 | * | |
1043 | * We do work in batches because it can be much cheaper to set up a number | |
1044 | * of flows and fire off their patches all at once. We do multiple batches | |
1045 | * because in some cases handling a packet can cause another packet to be | |
1046 | * queued almost immediately as part of the return flow. Both | |
1047 | * optimizations can make major improvements on some benchmarks and | |
1048 | * presumably for real traffic as well. */ | |
1049 | work = 0; | |
1050 | while (work < FLOW_MISS_MAX_BATCH) { | |
1051 | int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work); | |
1052 | if (retval <= 0) { | |
1053 | return -retval; | |
1054 | } | |
1055 | work += retval; | |
1056 | } | |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | static void | |
1062 | type_wait(const char *type) | |
1063 | { | |
1064 | struct dpif_backer *backer; | |
1065 | ||
1066 | backer = shash_find_data(&all_dpif_backers, type); | |
1067 | if (!backer) { | |
1068 | /* This is not necessarily a problem, since backers are only | |
1069 | * created on demand. */ | |
1070 | return; | |
1071 | } | |
1072 | ||
1073 | timer_wait(&backer->next_expiration); | |
1074 | } | |
1075 | \f | |
abe529af BP |
1076 | /* Basic life-cycle. */ |
1077 | ||
c57b2226 BP |
1078 | static int add_internal_flows(struct ofproto_dpif *); |
1079 | ||
abe529af BP |
1080 | static struct ofproto * |
1081 | alloc(void) | |
1082 | { | |
1083 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1084 | return &ofproto->up; | |
1085 | } | |
1086 | ||
1087 | static void | |
1088 | dealloc(struct ofproto *ofproto_) | |
1089 | { | |
1090 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1091 | free(ofproto); | |
1092 | } | |
1093 | ||
acf60855 JP |
1094 | static void |
1095 | close_dpif_backer(struct dpif_backer *backer) | |
1096 | { | |
1097 | struct shash_node *node; | |
1098 | ||
cb22974d | 1099 | ovs_assert(backer->refcount > 0); |
acf60855 JP |
1100 | |
1101 | if (--backer->refcount) { | |
1102 | return; | |
1103 | } | |
1104 | ||
8f73d537 EJ |
1105 | drop_key_clear(backer); |
1106 | hmap_destroy(&backer->drop_keys); | |
1107 | ||
7d82ab2e | 1108 | simap_destroy(&backer->tnl_backers); |
acf60855 JP |
1109 | hmap_destroy(&backer->odp_to_ofport_map); |
1110 | node = shash_find(&all_dpif_backers, backer->type); | |
1111 | free(backer->type); | |
1112 | shash_delete(&all_dpif_backers, node); | |
1113 | dpif_close(backer->dpif); | |
1114 | ||
1115 | free(backer); | |
1116 | } | |
1117 | ||
1118 | /* Datapath port slated for removal from datapath. */ | |
1119 | struct odp_garbage { | |
1120 | struct list list_node; | |
1121 | uint32_t odp_port; | |
1122 | }; | |
1123 | ||
1124 | static int | |
1125 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1126 | { | |
1127 | struct dpif_backer *backer; | |
1128 | struct dpif_port_dump port_dump; | |
1129 | struct dpif_port port; | |
1130 | struct shash_node *node; | |
1131 | struct list garbage_list; | |
1132 | struct odp_garbage *garbage, *next; | |
1133 | struct sset names; | |
1134 | char *backer_name; | |
1135 | const char *name; | |
1136 | int error; | |
1137 | ||
1138 | backer = shash_find_data(&all_dpif_backers, type); | |
1139 | if (backer) { | |
1140 | backer->refcount++; | |
1141 | *backerp = backer; | |
1142 | return 0; | |
1143 | } | |
1144 | ||
1145 | backer_name = xasprintf("ovs-%s", type); | |
1146 | ||
1147 | /* Remove any existing datapaths, since we assume we're the only | |
1148 | * userspace controlling the datapath. */ | |
1149 | sset_init(&names); | |
1150 | dp_enumerate_names(type, &names); | |
1151 | SSET_FOR_EACH(name, &names) { | |
1152 | struct dpif *old_dpif; | |
1153 | ||
1154 | /* Don't remove our backer if it exists. */ | |
1155 | if (!strcmp(name, backer_name)) { | |
1156 | continue; | |
1157 | } | |
1158 | ||
1159 | if (dpif_open(name, type, &old_dpif)) { | |
1160 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1161 | } else { | |
1162 | dpif_delete(old_dpif); | |
1163 | dpif_close(old_dpif); | |
1164 | } | |
1165 | } | |
1166 | sset_destroy(&names); | |
1167 | ||
1168 | backer = xmalloc(sizeof *backer); | |
1169 | ||
1170 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1171 | free(backer_name); | |
1172 | if (error) { | |
1173 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1174 | strerror(error)); | |
4c1b1289 | 1175 | free(backer); |
acf60855 JP |
1176 | return error; |
1177 | } | |
1178 | ||
1179 | backer->type = xstrdup(type); | |
1180 | backer->refcount = 1; | |
1181 | hmap_init(&backer->odp_to_ofport_map); | |
8f73d537 | 1182 | hmap_init(&backer->drop_keys); |
acf60855 | 1183 | timer_set_duration(&backer->next_expiration, 1000); |
2cc3c58e | 1184 | backer->need_revalidate = 0; |
7d82ab2e | 1185 | simap_init(&backer->tnl_backers); |
2cc3c58e | 1186 | tag_set_init(&backer->revalidate_set); |
acf60855 JP |
1187 | *backerp = backer; |
1188 | ||
1189 | dpif_flow_flush(backer->dpif); | |
1190 | ||
1191 | /* Loop through the ports already on the datapath and remove any | |
1192 | * that we don't need anymore. */ | |
1193 | list_init(&garbage_list); | |
1194 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1195 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1196 | node = shash_find(&init_ofp_ports, port.name); | |
1197 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1198 | garbage = xmalloc(sizeof *garbage); | |
1199 | garbage->odp_port = port.port_no; | |
1200 | list_push_front(&garbage_list, &garbage->list_node); | |
1201 | } | |
1202 | } | |
1203 | dpif_port_dump_done(&port_dump); | |
1204 | ||
1205 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1206 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1207 | list_remove(&garbage->list_node); | |
1208 | free(garbage); | |
1209 | } | |
1210 | ||
1211 | shash_add(&all_dpif_backers, type, backer); | |
1212 | ||
1213 | error = dpif_recv_set(backer->dpif, true); | |
1214 | if (error) { | |
1215 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1216 | type, strerror(error)); | |
1217 | close_dpif_backer(backer); | |
1218 | return error; | |
1219 | } | |
1220 | ||
1221 | return error; | |
1222 | } | |
1223 | ||
abe529af | 1224 | static int |
0f5f95a9 | 1225 | construct(struct ofproto *ofproto_) |
abe529af BP |
1226 | { |
1227 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 1228 | struct shash_node *node, *next; |
91858960 | 1229 | int max_ports; |
abe529af BP |
1230 | int error; |
1231 | int i; | |
1232 | ||
acf60855 | 1233 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); |
abe529af | 1234 | if (error) { |
abe529af BP |
1235 | return error; |
1236 | } | |
1237 | ||
acf60855 | 1238 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); |
91858960 BP |
1239 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); |
1240 | ||
6c1491fb | 1241 | ofproto->n_matches = 0; |
abe529af | 1242 | |
abe529af BP |
1243 | ofproto->netflow = NULL; |
1244 | ofproto->sflow = NULL; | |
21f7563c | 1245 | ofproto->stp = NULL; |
abe529af | 1246 | hmap_init(&ofproto->bundles); |
e764773c | 1247 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
1248 | for (i = 0; i < MAX_MIRRORS; i++) { |
1249 | ofproto->mirrors[i] = NULL; | |
1250 | } | |
1251 | ofproto->has_bonded_bundles = false; | |
1252 | ||
abe529af | 1253 | hmap_init(&ofproto->facets); |
b0f7b9b5 | 1254 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 1255 | ofproto->governor = NULL; |
4d2a0f39 | 1256 | ofproto->consistency_rl = LLONG_MIN; |
54a9cbc9 BP |
1257 | |
1258 | for (i = 0; i < N_TABLES; i++) { | |
1259 | struct table_dpif *table = &ofproto->tables[i]; | |
1260 | ||
1261 | table->catchall_table = NULL; | |
1262 | table->other_table = NULL; | |
1263 | table->basis = random_uint32(); | |
1264 | } | |
abe529af | 1265 | |
7ee20df1 BP |
1266 | list_init(&ofproto->completions); |
1267 | ||
abe529af BP |
1268 | ofproto_dpif_unixctl_init(); |
1269 | ||
ccb7c863 | 1270 | ofproto->has_mirrors = false; |
daff3353 EJ |
1271 | ofproto->has_bundle_action = false; |
1272 | ||
52a90c29 BP |
1273 | hmap_init(&ofproto->vlandev_map); |
1274 | hmap_init(&ofproto->realdev_vid_map); | |
1275 | ||
acf60855 | 1276 | sset_init(&ofproto->ports); |
0a740f48 | 1277 | sset_init(&ofproto->ghost_ports); |
acf60855 JP |
1278 | sset_init(&ofproto->port_poll_set); |
1279 | ofproto->port_poll_errno = 0; | |
1280 | ||
1281 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
4f9e08a5 | 1282 | struct iface_hint *iface_hint = node->data; |
acf60855 JP |
1283 | |
1284 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1285 | /* Check if the datapath already has this port. */ | |
1286 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1287 | sset_add(&ofproto->ports, node->name); | |
1288 | } | |
1289 | ||
1290 | free(iface_hint->br_name); | |
1291 | free(iface_hint->br_type); | |
4f9e08a5 | 1292 | free(iface_hint); |
acf60855 JP |
1293 | shash_delete(&init_ofp_ports, node); |
1294 | } | |
1295 | } | |
e1b1d06a | 1296 | |
b44a10b7 BP |
1297 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
1298 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 1299 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
1300 | |
1301 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
1302 | error = add_internal_flows(ofproto); |
1303 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1304 | ||
735d7efb AZ |
1305 | ofproto->n_hit = 0; |
1306 | ofproto->n_missed = 0; | |
1307 | ||
c57b2226 BP |
1308 | return error; |
1309 | } | |
1310 | ||
1311 | static int | |
1312 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 1313 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
1314 | { |
1315 | struct ofputil_flow_mod fm; | |
1316 | int error; | |
1317 | ||
81a76618 BP |
1318 | match_init_catchall(&fm.match); |
1319 | fm.priority = 0; | |
1320 | match_set_reg(&fm.match, 0, id); | |
623e1caf | 1321 | fm.new_cookie = htonll(0); |
c57b2226 BP |
1322 | fm.cookie = htonll(0); |
1323 | fm.cookie_mask = htonll(0); | |
1324 | fm.table_id = TBL_INTERNAL; | |
1325 | fm.command = OFPFC_ADD; | |
1326 | fm.idle_timeout = 0; | |
1327 | fm.hard_timeout = 0; | |
1328 | fm.buffer_id = 0; | |
1329 | fm.out_port = 0; | |
1330 | fm.flags = 0; | |
f25d0cf3 BP |
1331 | fm.ofpacts = ofpacts->data; |
1332 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
1333 | |
1334 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1335 | if (error) { | |
1336 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1337 | id, ofperr_to_string(error)); | |
1338 | return error; | |
1339 | } | |
1340 | ||
81a76618 | 1341 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); |
cb22974d | 1342 | ovs_assert(*rulep != NULL); |
0f5f95a9 | 1343 | |
abe529af BP |
1344 | return 0; |
1345 | } | |
1346 | ||
c57b2226 BP |
1347 | static int |
1348 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1349 | { | |
f25d0cf3 BP |
1350 | struct ofpact_controller *controller; |
1351 | uint64_t ofpacts_stub[128 / 8]; | |
1352 | struct ofpbuf ofpacts; | |
c57b2226 BP |
1353 | int error; |
1354 | int id; | |
1355 | ||
f25d0cf3 | 1356 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
1357 | id = 1; |
1358 | ||
f25d0cf3 BP |
1359 | controller = ofpact_put_CONTROLLER(&ofpacts); |
1360 | controller->max_len = UINT16_MAX; | |
1361 | controller->controller_id = 0; | |
1362 | controller->reason = OFPR_NO_MATCH; | |
1363 | ofpact_pad(&ofpacts); | |
1364 | ||
1365 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
1366 | if (error) { |
1367 | return error; | |
1368 | } | |
1369 | ||
f25d0cf3 BP |
1370 | ofpbuf_clear(&ofpacts); |
1371 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
1372 | &ofproto->no_packet_in_rule); |
1373 | return error; | |
1374 | } | |
1375 | ||
7ee20df1 BP |
1376 | static void |
1377 | complete_operations(struct ofproto_dpif *ofproto) | |
1378 | { | |
1379 | struct dpif_completion *c, *next; | |
1380 | ||
1381 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1382 | ofoperation_complete(c->op, 0); | |
1383 | list_remove(&c->list_node); | |
1384 | free(c); | |
1385 | } | |
1386 | } | |
1387 | ||
abe529af BP |
1388 | static void |
1389 | destruct(struct ofproto *ofproto_) | |
1390 | { | |
1391 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 1392 | struct rule_dpif *rule, *next_rule; |
d0918789 | 1393 | struct oftable *table; |
abe529af BP |
1394 | int i; |
1395 | ||
b44a10b7 | 1396 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
1397 | complete_operations(ofproto); |
1398 | ||
0697b5c3 BP |
1399 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
1400 | struct cls_cursor cursor; | |
1401 | ||
d0918789 | 1402 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
1403 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1404 | ofproto_rule_destroy(&rule->up); | |
1405 | } | |
7ee20df1 BP |
1406 | } |
1407 | ||
abe529af BP |
1408 | for (i = 0; i < MAX_MIRRORS; i++) { |
1409 | mirror_destroy(ofproto->mirrors[i]); | |
1410 | } | |
1411 | ||
1412 | netflow_destroy(ofproto->netflow); | |
bae473fe | 1413 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
1414 | hmap_destroy(&ofproto->bundles); |
1415 | mac_learning_destroy(ofproto->ml); | |
1416 | ||
1417 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 1418 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 1419 | governor_destroy(ofproto->governor); |
abe529af | 1420 | |
52a90c29 BP |
1421 | hmap_destroy(&ofproto->vlandev_map); |
1422 | hmap_destroy(&ofproto->realdev_vid_map); | |
1423 | ||
acf60855 | 1424 | sset_destroy(&ofproto->ports); |
0a740f48 | 1425 | sset_destroy(&ofproto->ghost_ports); |
acf60855 | 1426 | sset_destroy(&ofproto->port_poll_set); |
e1b1d06a | 1427 | |
acf60855 | 1428 | close_dpif_backer(ofproto->backer); |
abe529af BP |
1429 | } |
1430 | ||
1431 | static int | |
5fcc0d00 | 1432 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
1433 | { |
1434 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 1435 | struct ofport_dpif *ofport; |
abe529af | 1436 | |
0aa66d6e EJ |
1437 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
1438 | port_run_fast(ofport); | |
1439 | } | |
1440 | ||
5fcc0d00 BP |
1441 | return 0; |
1442 | } | |
1443 | ||
1444 | static int | |
1445 | run(struct ofproto *ofproto_) | |
1446 | { | |
1447 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1448 | struct ofport_dpif *ofport; | |
1449 | struct ofbundle *bundle; | |
1450 | int error; | |
1451 | ||
1452 | if (!clogged) { | |
1453 | complete_operations(ofproto); | |
1454 | } | |
5fcc0d00 BP |
1455 | |
1456 | error = run_fast(ofproto_); | |
1457 | if (error) { | |
1458 | return error; | |
abe529af BP |
1459 | } |
1460 | ||
abe529af | 1461 | if (ofproto->netflow) { |
6fca1ffb BP |
1462 | if (netflow_run(ofproto->netflow)) { |
1463 | send_netflow_active_timeouts(ofproto); | |
1464 | } | |
abe529af BP |
1465 | } |
1466 | if (ofproto->sflow) { | |
bae473fe | 1467 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
1468 | } |
1469 | ||
1470 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1471 | port_run(ofport); | |
1472 | } | |
1473 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1474 | bundle_run(bundle); | |
1475 | } | |
1476 | ||
21f7563c | 1477 | stp_run(ofproto); |
2cc3c58e | 1478 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af | 1479 | |
6814e51f | 1480 | /* Check the consistency of a random facet, to aid debugging. */ |
4d2a0f39 EJ |
1481 | if (time_msec() >= ofproto->consistency_rl |
1482 | && !hmap_is_empty(&ofproto->facets) | |
2cc3c58e | 1483 | && !ofproto->backer->need_revalidate) { |
6814e51f BP |
1484 | struct facet *facet; |
1485 | ||
4d2a0f39 EJ |
1486 | ofproto->consistency_rl = time_msec() + 250; |
1487 | ||
6814e51f BP |
1488 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), |
1489 | struct facet, hmap_node); | |
2cc3c58e EJ |
1490 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, |
1491 | facet->tags)) { | |
6814e51f | 1492 | if (!facet_check_consistency(facet)) { |
2cc3c58e | 1493 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1494 | } |
1495 | } | |
1496 | } | |
1497 | ||
9d6ac44e BP |
1498 | if (ofproto->governor) { |
1499 | size_t n_subfacets; | |
1500 | ||
1501 | governor_run(ofproto->governor); | |
1502 | ||
1503 | /* If the governor has shrunk to its minimum size and the number of | |
1504 | * subfacets has dwindled, then drop the governor entirely. | |
1505 | * | |
1506 | * For hysteresis, the number of subfacets to drop the governor is | |
1507 | * smaller than the number needed to trigger its creation. */ | |
1508 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1509 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1510 | && governor_is_idle(ofproto->governor)) { | |
1511 | governor_destroy(ofproto->governor); | |
1512 | ofproto->governor = NULL; | |
1513 | } | |
1514 | } | |
1515 | ||
abe529af BP |
1516 | return 0; |
1517 | } | |
1518 | ||
1519 | static void | |
1520 | wait(struct ofproto *ofproto_) | |
1521 | { | |
1522 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1523 | struct ofport_dpif *ofport; | |
1524 | struct ofbundle *bundle; | |
1525 | ||
7ee20df1 BP |
1526 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1527 | poll_immediate_wake(); | |
1528 | } | |
1529 | ||
acf60855 JP |
1530 | dpif_wait(ofproto->backer->dpif); |
1531 | dpif_recv_wait(ofproto->backer->dpif); | |
abe529af | 1532 | if (ofproto->sflow) { |
bae473fe | 1533 | dpif_sflow_wait(ofproto->sflow); |
abe529af | 1534 | } |
2cc3c58e | 1535 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { |
abe529af BP |
1536 | poll_immediate_wake(); |
1537 | } | |
1538 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1539 | port_wait(ofport); | |
1540 | } | |
1541 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1542 | bundle_wait(bundle); | |
1543 | } | |
6fca1ffb BP |
1544 | if (ofproto->netflow) { |
1545 | netflow_wait(ofproto->netflow); | |
1546 | } | |
1c313b88 | 1547 | mac_learning_wait(ofproto->ml); |
21f7563c | 1548 | stp_wait(ofproto); |
2cc3c58e | 1549 | if (ofproto->backer->need_revalidate) { |
abe529af BP |
1550 | /* Shouldn't happen, but if it does just go around again. */ |
1551 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1552 | poll_immediate_wake(); | |
abe529af | 1553 | } |
9d6ac44e BP |
1554 | if (ofproto->governor) { |
1555 | governor_wait(ofproto->governor); | |
1556 | } | |
abe529af BP |
1557 | } |
1558 | ||
0d085684 BP |
1559 | static void |
1560 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1561 | { | |
1562 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1563 | ||
1564 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1565 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1566 | } | |
1567 | ||
abe529af BP |
1568 | static void |
1569 | flush(struct ofproto *ofproto_) | |
1570 | { | |
1571 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
1572 | struct subfacet *subfacet, *next_subfacet; |
1573 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1574 | int n_batch; | |
b0f7b9b5 | 1575 | |
acf60855 JP |
1576 | n_batch = 0; |
1577 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
1578 | &ofproto->subfacets) { | |
1579 | if (subfacet->path != SF_NOT_INSTALLED) { | |
1580 | batch[n_batch++] = subfacet; | |
1581 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
1582 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1583 | n_batch = 0; | |
1584 | } | |
1585 | } else { | |
1586 | subfacet_destroy(subfacet); | |
b0f7b9b5 | 1587 | } |
abe529af | 1588 | } |
acf60855 JP |
1589 | |
1590 | if (n_batch > 0) { | |
1591 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1592 | } | |
abe529af BP |
1593 | } |
1594 | ||
6c1491fb BP |
1595 | static void |
1596 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1597 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1598 | { |
1599 | *arp_match_ip = true; | |
9e1fd49b BP |
1600 | *actions = (OFPUTIL_A_OUTPUT | |
1601 | OFPUTIL_A_SET_VLAN_VID | | |
1602 | OFPUTIL_A_SET_VLAN_PCP | | |
1603 | OFPUTIL_A_STRIP_VLAN | | |
1604 | OFPUTIL_A_SET_DL_SRC | | |
1605 | OFPUTIL_A_SET_DL_DST | | |
1606 | OFPUTIL_A_SET_NW_SRC | | |
1607 | OFPUTIL_A_SET_NW_DST | | |
1608 | OFPUTIL_A_SET_NW_TOS | | |
1609 | OFPUTIL_A_SET_TP_SRC | | |
1610 | OFPUTIL_A_SET_TP_DST | | |
1611 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1612 | } |
1613 | ||
1614 | static void | |
307975da | 1615 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) |
6c1491fb BP |
1616 | { |
1617 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1618 | struct dpif_dp_stats s; |
6c1491fb BP |
1619 | |
1620 | strcpy(ots->name, "classifier"); | |
1621 | ||
acf60855 JP |
1622 | dpif_get_dp_stats(ofproto->backer->dpif, &s); |
1623 | ||
307975da SH |
1624 | ots->lookup_count = htonll(s.n_hit + s.n_missed); |
1625 | ots->matched_count = htonll(s.n_hit + ofproto->n_matches); | |
6c1491fb BP |
1626 | } |
1627 | ||
abe529af BP |
1628 | static struct ofport * |
1629 | port_alloc(void) | |
1630 | { | |
1631 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1632 | return &port->up; | |
1633 | } | |
1634 | ||
1635 | static void | |
1636 | port_dealloc(struct ofport *port_) | |
1637 | { | |
1638 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1639 | free(port); | |
1640 | } | |
1641 | ||
1642 | static int | |
1643 | port_construct(struct ofport *port_) | |
1644 | { | |
1645 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1646 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1647 | const struct netdev *netdev = port->up.netdev; |
e1b1d06a JP |
1648 | struct dpif_port dpif_port; |
1649 | int error; | |
abe529af | 1650 | |
2cc3c58e | 1651 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1652 | port->bundle = NULL; |
1653 | port->cfm = NULL; | |
1654 | port->tag = tag_create_random(); | |
d5ffa7f2 | 1655 | port->may_enable = true; |
21f7563c JP |
1656 | port->stp_port = NULL; |
1657 | port->stp_state = STP_DISABLED; | |
b9ad7294 | 1658 | port->tnl_port = NULL; |
8b36f51e | 1659 | hmap_init(&port->priorities); |
52a90c29 BP |
1660 | port->realdev_ofp_port = 0; |
1661 | port->vlandev_vid = 0; | |
b9ad7294 | 1662 | port->carrier_seq = netdev_get_carrier_resets(netdev); |
abe529af | 1663 | |
b9ad7294 | 1664 | if (netdev_vport_is_patch(netdev)) { |
0a740f48 EJ |
1665 | /* XXX By bailing out here, we don't do required sFlow work. */ |
1666 | port->odp_port = OVSP_NONE; | |
1667 | return 0; | |
1668 | } | |
1669 | ||
acf60855 | 1670 | error = dpif_port_query_by_name(ofproto->backer->dpif, |
b9ad7294 | 1671 | netdev_vport_get_dpif_port(netdev), |
e1b1d06a JP |
1672 | &dpif_port); |
1673 | if (error) { | |
1674 | return error; | |
1675 | } | |
1676 | ||
1677 | port->odp_port = dpif_port.port_no; | |
1678 | ||
b9ad7294 EJ |
1679 | if (netdev_get_tunnel_config(netdev)) { |
1680 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1681 | } else { | |
1682 | /* Sanity-check that a mapping doesn't already exist. This | |
1683 | * shouldn't happen for non-tunnel ports. */ | |
1684 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1685 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1686 | dpif_port.name); | |
da78d43d | 1687 | dpif_port_destroy(&dpif_port); |
b9ad7294 EJ |
1688 | return EBUSY; |
1689 | } | |
e1b1d06a | 1690 | |
b9ad7294 EJ |
1691 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, |
1692 | hash_int(port->odp_port, 0)); | |
1693 | } | |
da78d43d | 1694 | dpif_port_destroy(&dpif_port); |
e1b1d06a | 1695 | |
abe529af | 1696 | if (ofproto->sflow) { |
e1b1d06a | 1697 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); |
abe529af BP |
1698 | } |
1699 | ||
1700 | return 0; | |
1701 | } | |
1702 | ||
1703 | static void | |
1704 | port_destruct(struct ofport *port_) | |
1705 | { | |
1706 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1707 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
b9ad7294 | 1708 | const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); |
02f8d646 | 1709 | const char *devname = netdev_get_name(port->up.netdev); |
abe529af | 1710 | |
a614d823 | 1711 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
acf60855 JP |
1712 | /* The underlying device is still there, so delete it. This |
1713 | * happens when the ofproto is being destroyed, since the caller | |
1714 | * assumes that removal of attached ports will happen as part of | |
1715 | * destruction. */ | |
a614d823 KM |
1716 | if (!port->tnl_port) { |
1717 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1718 | } | |
1719 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
acf60855 JP |
1720 | } |
1721 | ||
b9ad7294 | 1722 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { |
0a740f48 EJ |
1723 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); |
1724 | } | |
1725 | ||
b9ad7294 | 1726 | tnl_port_del(port->tnl_port); |
02f8d646 | 1727 | sset_find_and_delete(&ofproto->ports, devname); |
0a740f48 | 1728 | sset_find_and_delete(&ofproto->ghost_ports, devname); |
2cc3c58e | 1729 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1730 | bundle_remove(port_); |
a5610457 | 1731 | set_cfm(port_, NULL); |
abe529af | 1732 | if (ofproto->sflow) { |
bae473fe | 1733 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1734 | } |
8b36f51e EJ |
1735 | |
1736 | ofport_clear_priorities(port); | |
1737 | hmap_destroy(&port->priorities); | |
abe529af BP |
1738 | } |
1739 | ||
1740 | static void | |
1741 | port_modified(struct ofport *port_) | |
1742 | { | |
1743 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1744 | ||
1745 | if (port->bundle && port->bundle->bond) { | |
1746 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1747 | } | |
1748 | } | |
1749 | ||
1750 | static void | |
9e1fd49b | 1751 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1752 | { |
1753 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1754 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1755 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1756 | |
9e1fd49b | 1757 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1758 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1759 | OFPUTIL_PC_NO_PACKET_IN)) { | |
2cc3c58e | 1760 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1761 | |
9e1fd49b | 1762 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1763 | bundle_update(port->bundle); |
1764 | } | |
abe529af BP |
1765 | } |
1766 | } | |
1767 | ||
1768 | static int | |
1769 | set_sflow(struct ofproto *ofproto_, | |
1770 | const struct ofproto_sflow_options *sflow_options) | |
1771 | { | |
1772 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1773 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1774 | |
abe529af | 1775 | if (sflow_options) { |
bae473fe | 1776 | if (!ds) { |
abe529af BP |
1777 | struct ofport_dpif *ofport; |
1778 | ||
4213f19d | 1779 | ds = ofproto->sflow = dpif_sflow_create(); |
abe529af | 1780 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
e1b1d06a | 1781 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); |
abe529af | 1782 | } |
2cc3c58e | 1783 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af | 1784 | } |
bae473fe | 1785 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1786 | } else { |
6ff686f2 PS |
1787 | if (ds) { |
1788 | dpif_sflow_destroy(ds); | |
2cc3c58e | 1789 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1790 | ofproto->sflow = NULL; |
1791 | } | |
abe529af BP |
1792 | } |
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | static int | |
a5610457 | 1797 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1798 | { |
1799 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1800 | int error; | |
1801 | ||
a5610457 | 1802 | if (!s) { |
abe529af BP |
1803 | error = 0; |
1804 | } else { | |
1805 | if (!ofport->cfm) { | |
8c977421 EJ |
1806 | struct ofproto_dpif *ofproto; |
1807 | ||
1808 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2cc3c58e | 1809 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f629657 | 1810 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1811 | } |
1812 | ||
a5610457 | 1813 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1814 | return 0; |
1815 | } | |
1816 | ||
1817 | error = EINVAL; | |
1818 | } | |
1819 | cfm_destroy(ofport->cfm); | |
1820 | ofport->cfm = NULL; | |
1821 | return error; | |
1822 | } | |
1823 | ||
9a9e3786 BP |
1824 | static bool |
1825 | get_cfm_status(const struct ofport *ofport_, | |
1826 | struct ofproto_cfm_status *status) | |
1de11730 EJ |
1827 | { |
1828 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1829 | ||
1830 | if (ofport->cfm) { | |
9a9e3786 BP |
1831 | status->faults = cfm_get_fault(ofport->cfm); |
1832 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1833 | status->health = cfm_get_health(ofport->cfm); | |
1834 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1835 | return true; | |
1de11730 | 1836 | } else { |
9a9e3786 | 1837 | return false; |
1de11730 EJ |
1838 | } |
1839 | } | |
abe529af | 1840 | \f |
21f7563c JP |
1841 | /* Spanning Tree. */ |
1842 | ||
1843 | static void | |
1844 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1845 | { | |
1846 | struct ofproto_dpif *ofproto = ofproto_; | |
1847 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1848 | struct ofport_dpif *ofport; | |
1849 | ||
1850 | ofport = stp_port_get_aux(sp); | |
1851 | if (!ofport) { | |
1852 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1853 | ofproto->up.name, port_num); | |
1854 | } else { | |
1855 | struct eth_header *eth = pkt->l2; | |
1856 | ||
1857 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1858 | if (eth_addr_is_zero(eth->eth_src)) { | |
1859 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1860 | "with unknown MAC", ofproto->up.name, port_num); | |
1861 | } else { | |
97d6520b | 1862 | send_packet(ofport, pkt); |
21f7563c JP |
1863 | } |
1864 | } | |
1865 | ofpbuf_delete(pkt); | |
1866 | } | |
1867 | ||
1868 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1869 | static int | |
1870 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1871 | { | |
1872 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1873 | ||
1874 | /* Only revalidate flows if the configuration changed. */ | |
1875 | if (!s != !ofproto->stp) { | |
2cc3c58e | 1876 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
1877 | } |
1878 | ||
1879 | if (s) { | |
1880 | if (!ofproto->stp) { | |
1881 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1882 | send_bpdu_cb, ofproto); | |
1883 | ofproto->stp_last_tick = time_msec(); | |
1884 | } | |
1885 | ||
1886 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1887 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1888 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1889 | stp_set_max_age(ofproto->stp, s->max_age); | |
1890 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1891 | } else { | |
851bf71d EJ |
1892 | struct ofport *ofport; |
1893 | ||
1894 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1895 | set_stp_port(ofport, NULL); | |
1896 | } | |
1897 | ||
21f7563c JP |
1898 | stp_destroy(ofproto->stp); |
1899 | ofproto->stp = NULL; | |
1900 | } | |
1901 | ||
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | static int | |
1906 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1907 | { | |
1908 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1909 | ||
1910 | if (ofproto->stp) { | |
1911 | s->enabled = true; | |
1912 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1913 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1914 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1915 | } else { | |
1916 | s->enabled = false; | |
1917 | } | |
1918 | ||
1919 | return 0; | |
1920 | } | |
1921 | ||
1922 | static void | |
1923 | update_stp_port_state(struct ofport_dpif *ofport) | |
1924 | { | |
1925 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1926 | enum stp_state state; | |
1927 | ||
1928 | /* Figure out new state. */ | |
1929 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1930 | : STP_DISABLED; | |
1931 | ||
1932 | /* Update state. */ | |
1933 | if (ofport->stp_state != state) { | |
9e1fd49b | 1934 | enum ofputil_port_state of_state; |
21f7563c JP |
1935 | bool fwd_change; |
1936 | ||
1937 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1938 | netdev_get_name(ofport->up.netdev), | |
1939 | stp_state_name(ofport->stp_state), | |
1940 | stp_state_name(state)); | |
1941 | if (stp_learn_in_state(ofport->stp_state) | |
1942 | != stp_learn_in_state(state)) { | |
1943 | /* xxx Learning action flows should also be flushed. */ | |
2cc3c58e EJ |
1944 | mac_learning_flush(ofproto->ml, |
1945 | &ofproto->backer->revalidate_set); | |
21f7563c JP |
1946 | } |
1947 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1948 | != stp_forward_in_state(state); | |
1949 | ||
2cc3c58e | 1950 | ofproto->backer->need_revalidate = REV_STP; |
21f7563c JP |
1951 | ofport->stp_state = state; |
1952 | ofport->stp_state_entered = time_msec(); | |
1953 | ||
b308140a | 1954 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1955 | bundle_update(ofport->bundle); |
1956 | } | |
1957 | ||
1958 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1959 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1960 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1961 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1962 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1963 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1964 | : 0); | |
21f7563c JP |
1965 | ofproto_port_set_state(&ofport->up, of_state); |
1966 | } | |
1967 | } | |
1968 | ||
1969 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1970 | * caller is responsible for assigning STP port numbers and ensuring | |
1971 | * there are no duplicates. */ | |
1972 | static int | |
1973 | set_stp_port(struct ofport *ofport_, | |
1974 | const struct ofproto_port_stp_settings *s) | |
1975 | { | |
1976 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1977 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1978 | struct stp_port *sp = ofport->stp_port; | |
1979 | ||
1980 | if (!s || !s->enable) { | |
1981 | if (sp) { | |
1982 | ofport->stp_port = NULL; | |
1983 | stp_port_disable(sp); | |
ecd12731 | 1984 | update_stp_port_state(ofport); |
21f7563c JP |
1985 | } |
1986 | return 0; | |
1987 | } else if (sp && stp_port_no(sp) != s->port_num | |
1988 | && ofport == stp_port_get_aux(sp)) { | |
1989 | /* The port-id changed, so disable the old one if it's not | |
1990 | * already in use by another port. */ | |
1991 | stp_port_disable(sp); | |
1992 | } | |
1993 | ||
1994 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1995 | stp_port_enable(sp); | |
1996 | ||
1997 | stp_port_set_aux(sp, ofport); | |
1998 | stp_port_set_priority(sp, s->priority); | |
1999 | stp_port_set_path_cost(sp, s->path_cost); | |
2000 | ||
2001 | update_stp_port_state(ofport); | |
2002 | ||
2003 | return 0; | |
2004 | } | |
2005 | ||
2006 | static int | |
2007 | get_stp_port_status(struct ofport *ofport_, | |
2008 | struct ofproto_port_stp_status *s) | |
2009 | { | |
2010 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2011 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2012 | struct stp_port *sp = ofport->stp_port; | |
2013 | ||
2014 | if (!ofproto->stp || !sp) { | |
2015 | s->enabled = false; | |
2016 | return 0; | |
2017 | } | |
2018 | ||
2019 | s->enabled = true; | |
2020 | s->port_id = stp_port_get_id(sp); | |
2021 | s->state = stp_port_get_state(sp); | |
2022 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2023 | s->role = stp_port_get_role(sp); | |
80740385 | 2024 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
2025 | |
2026 | return 0; | |
2027 | } | |
2028 | ||
2029 | static void | |
2030 | stp_run(struct ofproto_dpif *ofproto) | |
2031 | { | |
2032 | if (ofproto->stp) { | |
2033 | long long int now = time_msec(); | |
2034 | long long int elapsed = now - ofproto->stp_last_tick; | |
2035 | struct stp_port *sp; | |
2036 | ||
2037 | if (elapsed > 0) { | |
2038 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2039 | ofproto->stp_last_tick = now; | |
2040 | } | |
2041 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2042 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2043 | ||
2044 | if (ofport) { | |
2045 | update_stp_port_state(ofport); | |
2046 | } | |
2047 | } | |
6ae50723 EJ |
2048 | |
2049 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2cc3c58e | 2050 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
6ae50723 | 2051 | } |
21f7563c JP |
2052 | } |
2053 | } | |
2054 | ||
2055 | static void | |
2056 | stp_wait(struct ofproto_dpif *ofproto) | |
2057 | { | |
2058 | if (ofproto->stp) { | |
2059 | poll_timer_wait(1000); | |
2060 | } | |
2061 | } | |
2062 | ||
2063 | /* Returns true if STP should process 'flow'. */ | |
2064 | static bool | |
2065 | stp_should_process_flow(const struct flow *flow) | |
2066 | { | |
2067 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2068 | } | |
2069 | ||
2070 | static void | |
2071 | stp_process_packet(const struct ofport_dpif *ofport, | |
2072 | const struct ofpbuf *packet) | |
2073 | { | |
2074 | struct ofpbuf payload = *packet; | |
2075 | struct eth_header *eth = payload.data; | |
2076 | struct stp_port *sp = ofport->stp_port; | |
2077 | ||
2078 | /* Sink packets on ports that have STP disabled when the bridge has | |
2079 | * STP enabled. */ | |
2080 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2081 | return; | |
2082 | } | |
2083 | ||
2084 | /* Trim off padding on payload. */ | |
c573540b BP |
2085 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
2086 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
2087 | } |
2088 | ||
2089 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2090 | stp_received_bpdu(sp, payload.data, payload.size); | |
2091 | } | |
2092 | } | |
2093 | \f | |
8b36f51e EJ |
2094 | static struct priority_to_dscp * |
2095 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2096 | { | |
2097 | struct priority_to_dscp *pdscp; | |
2098 | uint32_t hash; | |
2099 | ||
2100 | hash = hash_int(priority, 0); | |
2101 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2102 | if (pdscp->priority == priority) { | |
2103 | return pdscp; | |
2104 | } | |
2105 | } | |
2106 | return NULL; | |
2107 | } | |
2108 | ||
2109 | static void | |
2110 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2111 | { | |
2112 | struct priority_to_dscp *pdscp, *next; | |
2113 | ||
2114 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2115 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2116 | free(pdscp); | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | static int | |
2121 | set_queues(struct ofport *ofport_, | |
2122 | const struct ofproto_port_queue *qdscp_list, | |
2123 | size_t n_qdscp) | |
2124 | { | |
2125 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2126 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2127 | struct hmap new = HMAP_INITIALIZER(&new); | |
2128 | size_t i; | |
2129 | ||
2130 | for (i = 0; i < n_qdscp; i++) { | |
2131 | struct priority_to_dscp *pdscp; | |
2132 | uint32_t priority; | |
2133 | uint8_t dscp; | |
2134 | ||
2135 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
acf60855 | 2136 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, |
8b36f51e EJ |
2137 | &priority)) { |
2138 | continue; | |
2139 | } | |
2140 | ||
2141 | pdscp = get_priority(ofport, priority); | |
2142 | if (pdscp) { | |
2143 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2144 | } else { | |
2145 | pdscp = xmalloc(sizeof *pdscp); | |
2146 | pdscp->priority = priority; | |
2147 | pdscp->dscp = dscp; | |
2cc3c58e | 2148 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2149 | } |
2150 | ||
2151 | if (pdscp->dscp != dscp) { | |
2152 | pdscp->dscp = dscp; | |
2cc3c58e | 2153 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2154 | } |
2155 | ||
2156 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2157 | } | |
2158 | ||
2159 | if (!hmap_is_empty(&ofport->priorities)) { | |
2160 | ofport_clear_priorities(ofport); | |
2cc3c58e | 2161 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
2162 | } |
2163 | ||
2164 | hmap_swap(&new, &ofport->priorities); | |
2165 | hmap_destroy(&new); | |
2166 | ||
2167 | return 0; | |
2168 | } | |
2169 | \f | |
abe529af BP |
2170 | /* Bundles. */ |
2171 | ||
b44a10b7 BP |
2172 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
2173 | * ofproto to revalidate every flow. | |
2174 | * | |
2175 | * Normally MAC learning entries are removed only from the ofproto associated | |
2176 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2177 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2178 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2179 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2180 | * with the host from which it migrated. */ | |
abe529af | 2181 | static void |
b44a10b7 | 2182 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
2183 | { |
2184 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2185 | struct mac_learning *ml = ofproto->ml; | |
2186 | struct mac_entry *mac, *next_mac; | |
2187 | ||
2cc3c58e | 2188 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2189 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
2190 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
2191 | if (all_ofprotos) { |
2192 | struct ofproto_dpif *o; | |
2193 | ||
2194 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2195 | if (o != ofproto) { | |
2196 | struct mac_entry *e; | |
2197 | ||
2198 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2199 | NULL); | |
2200 | if (e) { | |
b44a10b7 BP |
2201 | mac_learning_expire(o->ml, e); |
2202 | } | |
2203 | } | |
2204 | } | |
2205 | } | |
2206 | ||
abe529af BP |
2207 | mac_learning_expire(ml, mac); |
2208 | } | |
2209 | } | |
2210 | } | |
2211 | ||
2212 | static struct ofbundle * | |
2213 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2214 | { | |
2215 | struct ofbundle *bundle; | |
2216 | ||
2217 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2218 | &ofproto->bundles) { | |
2219 | if (bundle->aux == aux) { | |
2220 | return bundle; | |
2221 | } | |
2222 | } | |
2223 | return NULL; | |
2224 | } | |
2225 | ||
2226 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2227 | * ones that are found to 'bundles'. */ | |
2228 | static void | |
2229 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2230 | void **auxes, size_t n_auxes, | |
2231 | struct hmapx *bundles) | |
2232 | { | |
2233 | size_t i; | |
2234 | ||
2235 | hmapx_init(bundles); | |
2236 | for (i = 0; i < n_auxes; i++) { | |
2237 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2238 | if (bundle) { | |
2239 | hmapx_add(bundles, bundle); | |
2240 | } | |
2241 | } | |
2242 | } | |
2243 | ||
7bde8dd8 JP |
2244 | static void |
2245 | bundle_update(struct ofbundle *bundle) | |
2246 | { | |
2247 | struct ofport_dpif *port; | |
2248 | ||
2249 | bundle->floodable = true; | |
2250 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
2251 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2252 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
2253 | bundle->floodable = false; |
2254 | break; | |
2255 | } | |
2256 | } | |
2257 | } | |
2258 | ||
abe529af BP |
2259 | static void |
2260 | bundle_del_port(struct ofport_dpif *port) | |
2261 | { | |
2262 | struct ofbundle *bundle = port->bundle; | |
2263 | ||
2cc3c58e | 2264 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 2265 | |
abe529af BP |
2266 | list_remove(&port->bundle_node); |
2267 | port->bundle = NULL; | |
2268 | ||
2269 | if (bundle->lacp) { | |
2270 | lacp_slave_unregister(bundle->lacp, port); | |
2271 | } | |
2272 | if (bundle->bond) { | |
2273 | bond_slave_unregister(bundle->bond, port); | |
2274 | } | |
2275 | ||
7bde8dd8 | 2276 | bundle_update(bundle); |
abe529af BP |
2277 | } |
2278 | ||
2279 | static bool | |
2280 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
df53d41c | 2281 | struct lacp_slave_settings *lacp) |
abe529af BP |
2282 | { |
2283 | struct ofport_dpif *port; | |
2284 | ||
2285 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2286 | if (!port) { | |
2287 | return false; | |
2288 | } | |
2289 | ||
2290 | if (port->bundle != bundle) { | |
2cc3c58e | 2291 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2292 | if (port->bundle) { |
2293 | bundle_del_port(port); | |
2294 | } | |
2295 | ||
2296 | port->bundle = bundle; | |
2297 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
2298 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
2299 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
2300 | bundle->floodable = false; |
2301 | } | |
2302 | } | |
2303 | if (lacp) { | |
2cc3c58e | 2304 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2305 | lacp_slave_register(bundle->lacp, port, lacp); |
2306 | } | |
2307 | ||
2308 | return true; | |
2309 | } | |
2310 | ||
2311 | static void | |
2312 | bundle_destroy(struct ofbundle *bundle) | |
2313 | { | |
2314 | struct ofproto_dpif *ofproto; | |
2315 | struct ofport_dpif *port, *next_port; | |
2316 | int i; | |
2317 | ||
2318 | if (!bundle) { | |
2319 | return; | |
2320 | } | |
2321 | ||
2322 | ofproto = bundle->ofproto; | |
2323 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2324 | struct ofmirror *m = ofproto->mirrors[i]; | |
2325 | if (m) { | |
2326 | if (m->out == bundle) { | |
2327 | mirror_destroy(m); | |
2328 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2329 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2cc3c58e | 2330 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2331 | } |
2332 | } | |
2333 | } | |
2334 | ||
2335 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2336 | bundle_del_port(port); | |
2337 | } | |
2338 | ||
b44a10b7 | 2339 | bundle_flush_macs(bundle, true); |
abe529af BP |
2340 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
2341 | free(bundle->name); | |
2342 | free(bundle->trunks); | |
2343 | lacp_destroy(bundle->lacp); | |
2344 | bond_destroy(bundle->bond); | |
2345 | free(bundle); | |
2346 | } | |
2347 | ||
2348 | static int | |
2349 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2350 | const struct ofproto_bundle_settings *s) | |
2351 | { | |
2352 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2353 | bool need_flush = false; | |
abe529af BP |
2354 | struct ofport_dpif *port; |
2355 | struct ofbundle *bundle; | |
ecac4ebf BP |
2356 | unsigned long *trunks; |
2357 | int vlan; | |
abe529af BP |
2358 | size_t i; |
2359 | bool ok; | |
2360 | ||
2361 | if (!s) { | |
2362 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2363 | return 0; | |
2364 | } | |
2365 | ||
cb22974d BP |
2366 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); |
2367 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
abe529af BP |
2368 | |
2369 | bundle = bundle_lookup(ofproto, aux); | |
2370 | if (!bundle) { | |
2371 | bundle = xmalloc(sizeof *bundle); | |
2372 | ||
2373 | bundle->ofproto = ofproto; | |
2374 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2375 | hash_pointer(aux, 0)); | |
2376 | bundle->aux = aux; | |
2377 | bundle->name = NULL; | |
2378 | ||
2379 | list_init(&bundle->ports); | |
ecac4ebf | 2380 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
2381 | bundle->vlan = -1; |
2382 | bundle->trunks = NULL; | |
5e9ceccd | 2383 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
2384 | bundle->lacp = NULL; |
2385 | bundle->bond = NULL; | |
2386 | ||
2387 | bundle->floodable = true; | |
2388 | ||
2389 | bundle->src_mirrors = 0; | |
2390 | bundle->dst_mirrors = 0; | |
2391 | bundle->mirror_out = 0; | |
2392 | } | |
2393 | ||
2394 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2395 | free(bundle->name); | |
2396 | bundle->name = xstrdup(s->name); | |
2397 | } | |
2398 | ||
2399 | /* LACP. */ | |
2400 | if (s->lacp) { | |
2401 | if (!bundle->lacp) { | |
2cc3c58e | 2402 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2403 | bundle->lacp = lacp_create(); |
2404 | } | |
2405 | lacp_configure(bundle->lacp, s->lacp); | |
2406 | } else { | |
2407 | lacp_destroy(bundle->lacp); | |
2408 | bundle->lacp = NULL; | |
2409 | } | |
2410 | ||
2411 | /* Update set of ports. */ | |
2412 | ok = true; | |
2413 | for (i = 0; i < s->n_slaves; i++) { | |
2414 | if (!bundle_add_port(bundle, s->slaves[i], | |
df53d41c | 2415 | s->lacp ? &s->lacp_slaves[i] : NULL)) { |
abe529af BP |
2416 | ok = false; |
2417 | } | |
2418 | } | |
2419 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2420 | struct ofport_dpif *next_port; | |
2421 | ||
2422 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2423 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 2424 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
2425 | goto found; |
2426 | } | |
2427 | } | |
2428 | ||
2429 | bundle_del_port(port); | |
2430 | found: ; | |
2431 | } | |
2432 | } | |
cb22974d | 2433 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); |
abe529af BP |
2434 | |
2435 | if (list_is_empty(&bundle->ports)) { | |
2436 | bundle_destroy(bundle); | |
2437 | return EINVAL; | |
2438 | } | |
2439 | ||
ecac4ebf | 2440 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
2441 | if (s->vlan_mode != bundle->vlan_mode |
2442 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 2443 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 2444 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
2445 | need_flush = true; |
2446 | } | |
2447 | ||
abe529af | 2448 | /* Set VLAN tag. */ |
ecac4ebf BP |
2449 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
2450 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2451 | : 0); | |
2452 | if (vlan != bundle->vlan) { | |
2453 | bundle->vlan = vlan; | |
abe529af BP |
2454 | need_flush = true; |
2455 | } | |
2456 | ||
2457 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
2458 | switch (s->vlan_mode) { |
2459 | case PORT_VLAN_ACCESS: | |
2460 | trunks = NULL; | |
2461 | break; | |
2462 | ||
2463 | case PORT_VLAN_TRUNK: | |
ebc56baa | 2464 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2465 | break; |
2466 | ||
2467 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2468 | case PORT_VLAN_NATIVE_TAGGED: | |
2469 | if (vlan != 0 && (!s->trunks | |
2470 | || !bitmap_is_set(s->trunks, vlan) | |
2471 | || bitmap_is_set(s->trunks, 0))) { | |
2472 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2473 | if (s->trunks) { | |
2474 | trunks = bitmap_clone(s->trunks, 4096); | |
2475 | } else { | |
2476 | trunks = bitmap_allocate1(4096); | |
2477 | } | |
2478 | bitmap_set1(trunks, vlan); | |
2479 | bitmap_set0(trunks, 0); | |
2480 | } else { | |
ebc56baa | 2481 | trunks = CONST_CAST(unsigned long *, s->trunks); |
ecac4ebf BP |
2482 | } |
2483 | break; | |
2484 | ||
2485 | default: | |
2486 | NOT_REACHED(); | |
2487 | } | |
abe529af BP |
2488 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
2489 | free(bundle->trunks); | |
ecac4ebf BP |
2490 | if (trunks == s->trunks) { |
2491 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2492 | } else { | |
2493 | bundle->trunks = trunks; | |
2494 | trunks = NULL; | |
2495 | } | |
abe529af BP |
2496 | need_flush = true; |
2497 | } | |
ecac4ebf BP |
2498 | if (trunks != s->trunks) { |
2499 | free(trunks); | |
2500 | } | |
abe529af BP |
2501 | |
2502 | /* Bonding. */ | |
2503 | if (!list_is_short(&bundle->ports)) { | |
2504 | bundle->ofproto->has_bonded_bundles = true; | |
2505 | if (bundle->bond) { | |
2506 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2cc3c58e | 2507 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2508 | } |
2509 | } else { | |
2510 | bundle->bond = bond_create(s->bond); | |
2cc3c58e | 2511 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2512 | } |
2513 | ||
2514 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
df53d41c | 2515 | bond_slave_register(bundle->bond, port, port->up.netdev); |
abe529af BP |
2516 | } |
2517 | } else { | |
2518 | bond_destroy(bundle->bond); | |
2519 | bundle->bond = NULL; | |
2520 | } | |
2521 | ||
2522 | /* If we changed something that would affect MAC learning, un-learn | |
2523 | * everything on this port and force flow revalidation. */ | |
2524 | if (need_flush) { | |
b44a10b7 | 2525 | bundle_flush_macs(bundle, false); |
abe529af BP |
2526 | } |
2527 | ||
2528 | return 0; | |
2529 | } | |
2530 | ||
2531 | static void | |
2532 | bundle_remove(struct ofport *port_) | |
2533 | { | |
2534 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2535 | struct ofbundle *bundle = port->bundle; | |
2536 | ||
2537 | if (bundle) { | |
2538 | bundle_del_port(port); | |
2539 | if (list_is_empty(&bundle->ports)) { | |
2540 | bundle_destroy(bundle); | |
2541 | } else if (list_is_short(&bundle->ports)) { | |
2542 | bond_destroy(bundle->bond); | |
2543 | bundle->bond = NULL; | |
2544 | } | |
2545 | } | |
2546 | } | |
2547 | ||
2548 | static void | |
5f877369 | 2549 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2550 | { |
2551 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2552 | struct ofport_dpif *port = port_; | |
2553 | uint8_t ea[ETH_ADDR_LEN]; | |
2554 | int error; | |
2555 | ||
2556 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2557 | if (!error) { | |
abe529af | 2558 | struct ofpbuf packet; |
5f877369 | 2559 | void *packet_pdu; |
abe529af BP |
2560 | |
2561 | ofpbuf_init(&packet, 0); | |
2562 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2563 | pdu_size); |
2564 | memcpy(packet_pdu, pdu, pdu_size); | |
2565 | ||
97d6520b | 2566 | send_packet(port, &packet); |
abe529af BP |
2567 | ofpbuf_uninit(&packet); |
2568 | } else { | |
2569 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2570 | "%s (%s)", port->bundle->name, | |
2571 | netdev_get_name(port->up.netdev), strerror(error)); | |
2572 | } | |
2573 | } | |
2574 | ||
2575 | static void | |
2576 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2577 | { | |
2578 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2579 | int error, n_packets, n_errors; | |
2580 | struct mac_entry *e; | |
2581 | ||
2582 | error = n_packets = n_errors = 0; | |
2583 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2584 | if (e->port.p != bundle) { | |
ea131871 JG |
2585 | struct ofpbuf *learning_packet; |
2586 | struct ofport_dpif *port; | |
4dd1e3ca | 2587 | void *port_void; |
ea131871 JG |
2588 | int ret; |
2589 | ||
4dd1e3ca BP |
2590 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2591 | * struct ofport_dpif more effective. */ | |
2592 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2593 | e->mac, e->vlan, | |
2594 | &port_void); | |
2595 | port = port_void; | |
97d6520b | 2596 | ret = send_packet(port, learning_packet); |
ea131871 | 2597 | ofpbuf_delete(learning_packet); |
abe529af BP |
2598 | if (ret) { |
2599 | error = ret; | |
2600 | n_errors++; | |
2601 | } | |
2602 | n_packets++; | |
2603 | } | |
2604 | } | |
2605 | ||
2606 | if (n_errors) { | |
2607 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2608 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2609 | "packets, last error was: %s", | |
2610 | bundle->name, n_errors, n_packets, strerror(error)); | |
2611 | } else { | |
2612 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2613 | bundle->name, n_packets); | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | static void | |
2618 | bundle_run(struct ofbundle *bundle) | |
2619 | { | |
2620 | if (bundle->lacp) { | |
2621 | lacp_run(bundle->lacp, send_pdu_cb); | |
2622 | } | |
2623 | if (bundle->bond) { | |
2624 | struct ofport_dpif *port; | |
2625 | ||
2626 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2627 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2628 | } |
2629 | ||
2cc3c58e | 2630 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, |
bdebeece | 2631 | lacp_status(bundle->lacp)); |
abe529af BP |
2632 | if (bond_should_send_learning_packets(bundle->bond)) { |
2633 | bundle_send_learning_packets(bundle); | |
2634 | } | |
2635 | } | |
2636 | } | |
2637 | ||
2638 | static void | |
2639 | bundle_wait(struct ofbundle *bundle) | |
2640 | { | |
2641 | if (bundle->lacp) { | |
2642 | lacp_wait(bundle->lacp); | |
2643 | } | |
2644 | if (bundle->bond) { | |
2645 | bond_wait(bundle->bond); | |
2646 | } | |
2647 | } | |
2648 | \f | |
2649 | /* Mirrors. */ | |
2650 | ||
2651 | static int | |
2652 | mirror_scan(struct ofproto_dpif *ofproto) | |
2653 | { | |
2654 | int idx; | |
2655 | ||
2656 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2657 | if (!ofproto->mirrors[idx]) { | |
2658 | return idx; | |
2659 | } | |
2660 | } | |
2661 | return -1; | |
2662 | } | |
2663 | ||
2664 | static struct ofmirror * | |
2665 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2666 | { | |
2667 | int i; | |
2668 | ||
2669 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2670 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2671 | if (mirror && mirror->aux == aux) { | |
2672 | return mirror; | |
2673 | } | |
2674 | } | |
2675 | ||
2676 | return NULL; | |
2677 | } | |
2678 | ||
9ba15e2a BP |
2679 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2680 | static void | |
2681 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2682 | { | |
2683 | int i; | |
2684 | ||
2685 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2686 | struct ofmirror *m = ofproto->mirrors[i]; | |
2687 | ||
2688 | if (m) { | |
2689 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2690 | } | |
2691 | } | |
2692 | ||
2693 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2694 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2695 | int j; | |
2696 | ||
2697 | if (!m1) { | |
2698 | continue; | |
2699 | } | |
2700 | ||
2701 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2702 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2703 | ||
edb0540b | 2704 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2705 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2706 | m2->dup_mirrors |= m1->dup_mirrors; | |
2707 | } | |
2708 | } | |
2709 | } | |
2710 | } | |
2711 | ||
abe529af BP |
2712 | static int |
2713 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2714 | const struct ofproto_mirror_settings *s) | |
2715 | { | |
2716 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2717 | mirror_mask_t mirror_bit; | |
2718 | struct ofbundle *bundle; | |
2719 | struct ofmirror *mirror; | |
2720 | struct ofbundle *out; | |
2721 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2722 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2723 | int out_vlan; | |
2724 | ||
2725 | mirror = mirror_lookup(ofproto, aux); | |
2726 | if (!s) { | |
2727 | mirror_destroy(mirror); | |
2728 | return 0; | |
2729 | } | |
2730 | if (!mirror) { | |
2731 | int idx; | |
2732 | ||
2733 | idx = mirror_scan(ofproto); | |
2734 | if (idx < 0) { | |
2735 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2736 | "cannot create %s", | |
2737 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2738 | return EFBIG; | |
2739 | } | |
2740 | ||
2741 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2742 | mirror->ofproto = ofproto; | |
2743 | mirror->idx = idx; | |
8b28d864 | 2744 | mirror->aux = aux; |
abe529af BP |
2745 | mirror->out_vlan = -1; |
2746 | mirror->name = NULL; | |
2747 | } | |
2748 | ||
2749 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2750 | free(mirror->name); | |
2751 | mirror->name = xstrdup(s->name); | |
2752 | } | |
2753 | ||
2754 | /* Get the new configuration. */ | |
2755 | if (s->out_bundle) { | |
2756 | out = bundle_lookup(ofproto, s->out_bundle); | |
2757 | if (!out) { | |
2758 | mirror_destroy(mirror); | |
2759 | return EINVAL; | |
2760 | } | |
2761 | out_vlan = -1; | |
2762 | } else { | |
2763 | out = NULL; | |
2764 | out_vlan = s->out_vlan; | |
2765 | } | |
2766 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2767 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2768 | ||
2769 | /* If the configuration has not changed, do nothing. */ | |
2770 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2771 | && hmapx_equals(&dsts, &mirror->dsts) | |
2772 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2773 | && mirror->out == out | |
2774 | && mirror->out_vlan == out_vlan) | |
2775 | { | |
2776 | hmapx_destroy(&srcs); | |
2777 | hmapx_destroy(&dsts); | |
2778 | return 0; | |
2779 | } | |
2780 | ||
2781 | hmapx_swap(&srcs, &mirror->srcs); | |
2782 | hmapx_destroy(&srcs); | |
2783 | ||
2784 | hmapx_swap(&dsts, &mirror->dsts); | |
2785 | hmapx_destroy(&dsts); | |
2786 | ||
2787 | free(mirror->vlans); | |
2788 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2789 | ||
2790 | mirror->out = out; | |
2791 | mirror->out_vlan = out_vlan; | |
2792 | ||
2793 | /* Update bundles. */ | |
2794 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2795 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2796 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2797 | bundle->src_mirrors |= mirror_bit; | |
2798 | } else { | |
2799 | bundle->src_mirrors &= ~mirror_bit; | |
2800 | } | |
2801 | ||
2802 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2803 | bundle->dst_mirrors |= mirror_bit; | |
2804 | } else { | |
2805 | bundle->dst_mirrors &= ~mirror_bit; | |
2806 | } | |
2807 | ||
2808 | if (mirror->out == bundle) { | |
2809 | bundle->mirror_out |= mirror_bit; | |
2810 | } else { | |
2811 | bundle->mirror_out &= ~mirror_bit; | |
2812 | } | |
2813 | } | |
2814 | ||
2cc3c58e | 2815 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 2816 | ofproto->has_mirrors = true; |
2cc3c58e EJ |
2817 | mac_learning_flush(ofproto->ml, |
2818 | &ofproto->backer->revalidate_set); | |
9ba15e2a | 2819 | mirror_update_dups(ofproto); |
abe529af BP |
2820 | |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | static void | |
2825 | mirror_destroy(struct ofmirror *mirror) | |
2826 | { | |
2827 | struct ofproto_dpif *ofproto; | |
2828 | mirror_mask_t mirror_bit; | |
2829 | struct ofbundle *bundle; | |
ccb7c863 | 2830 | int i; |
abe529af BP |
2831 | |
2832 | if (!mirror) { | |
2833 | return; | |
2834 | } | |
2835 | ||
2836 | ofproto = mirror->ofproto; | |
2cc3c58e EJ |
2837 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
2838 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
abe529af BP |
2839 | |
2840 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2841 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2842 | bundle->src_mirrors &= ~mirror_bit; | |
2843 | bundle->dst_mirrors &= ~mirror_bit; | |
2844 | bundle->mirror_out &= ~mirror_bit; | |
2845 | } | |
2846 | ||
2847 | hmapx_destroy(&mirror->srcs); | |
2848 | hmapx_destroy(&mirror->dsts); | |
2849 | free(mirror->vlans); | |
2850 | ||
2851 | ofproto->mirrors[mirror->idx] = NULL; | |
2852 | free(mirror->name); | |
2853 | free(mirror); | |
9ba15e2a BP |
2854 | |
2855 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
2856 | |
2857 | ofproto->has_mirrors = false; | |
2858 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2859 | if (ofproto->mirrors[i]) { | |
2860 | ofproto->has_mirrors = true; | |
2861 | break; | |
2862 | } | |
2863 | } | |
abe529af BP |
2864 | } |
2865 | ||
9d24de3b JP |
2866 | static int |
2867 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2868 | uint64_t *packets, uint64_t *bytes) | |
2869 | { | |
2870 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2871 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2872 | ||
2873 | if (!mirror) { | |
2874 | *packets = *bytes = UINT64_MAX; | |
2875 | return 0; | |
2876 | } | |
2877 | ||
2878 | *packets = mirror->packet_count; | |
2879 | *bytes = mirror->byte_count; | |
2880 | ||
2881 | return 0; | |
2882 | } | |
2883 | ||
abe529af BP |
2884 | static int |
2885 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2886 | { | |
2887 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2888 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2cc3c58e | 2889 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
abe529af BP |
2890 | } |
2891 | return 0; | |
2892 | } | |
2893 | ||
2894 | static bool | |
b4affc74 | 2895 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2896 | { |
2897 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2898 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2899 | return bundle && bundle->mirror_out != 0; | |
2900 | } | |
8402c74b SS |
2901 | |
2902 | static void | |
b53055f4 | 2903 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2904 | { |
2905 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2cc3c58e | 2906 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2907 | } |
e764773c BP |
2908 | |
2909 | static void | |
c4069512 BP |
2910 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, |
2911 | size_t max_entries) | |
e764773c BP |
2912 | { |
2913 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2914 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
c4069512 | 2915 | mac_learning_set_max_entries(ofproto->ml, max_entries); |
e764773c | 2916 | } |
abe529af BP |
2917 | \f |
2918 | /* Ports. */ | |
2919 | ||
2920 | static struct ofport_dpif * | |
4acbc98d | 2921 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 2922 | { |
7df6a8bd BP |
2923 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2924 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2925 | } |
2926 | ||
2927 | static struct ofport_dpif * | |
4acbc98d | 2928 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af | 2929 | { |
7c33b188 JR |
2930 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); |
2931 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
abe529af BP |
2932 | } |
2933 | ||
2934 | static void | |
e1b1d06a JP |
2935 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, |
2936 | struct ofproto_port *ofproto_port, | |
abe529af BP |
2937 | struct dpif_port *dpif_port) |
2938 | { | |
2939 | ofproto_port->name = dpif_port->name; | |
2940 | ofproto_port->type = dpif_port->type; | |
e1b1d06a | 2941 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); |
abe529af BP |
2942 | } |
2943 | ||
0a740f48 EJ |
2944 | static struct ofport_dpif * |
2945 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
2946 | { | |
2947 | const struct ofproto_dpif *ofproto; | |
2948 | const char *peer; | |
2949 | ||
2950 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
2951 | if (!peer) { | |
2952 | return NULL; | |
2953 | } | |
2954 | ||
2955 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2956 | struct ofport *ofport; | |
2957 | ||
2958 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
2959 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
2960 | return ofport_dpif_cast(ofport); | |
2961 | } | |
2962 | } | |
2963 | return NULL; | |
2964 | } | |
2965 | ||
0aa66d6e EJ |
2966 | static void |
2967 | port_run_fast(struct ofport_dpif *ofport) | |
2968 | { | |
2969 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2970 | struct ofpbuf packet; | |
2971 | ||
2972 | ofpbuf_init(&packet, 0); | |
2973 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2974 | send_packet(ofport, &packet); | |
2975 | ofpbuf_uninit(&packet); | |
2976 | } | |
2977 | } | |
2978 | ||
abe529af BP |
2979 | static void |
2980 | port_run(struct ofport_dpif *ofport) | |
2981 | { | |
3e5b3fdb EJ |
2982 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2983 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2984 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2985 | ||
3e5b3fdb EJ |
2986 | ofport->carrier_seq = carrier_seq; |
2987 | ||
0aa66d6e | 2988 | port_run_fast(ofport); |
b9ad7294 EJ |
2989 | |
2990 | if (ofport->tnl_port | |
2991 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
2992 | &ofport->tnl_port)) { | |
2993 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
2994 | } | |
2995 | ||
abe529af | 2996 | if (ofport->cfm) { |
4653c558 EJ |
2997 | int cfm_opup = cfm_get_opup(ofport->cfm); |
2998 | ||
abe529af | 2999 | cfm_run(ofport->cfm); |
4653c558 EJ |
3000 | enable = enable && !cfm_get_fault(ofport->cfm); |
3001 | ||
3002 | if (cfm_opup >= 0) { | |
3003 | enable = enable && cfm_opup; | |
3004 | } | |
abe529af | 3005 | } |
015e08bc EJ |
3006 | |
3007 | if (ofport->bundle) { | |
3008 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
3009 | if (carrier_changed) { |
3010 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3011 | } | |
015e08bc EJ |
3012 | } |
3013 | ||
daff3353 EJ |
3014 | if (ofport->may_enable != enable) { |
3015 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3016 | ||
3017 | if (ofproto->has_bundle_action) { | |
2cc3c58e | 3018 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
3019 | } |
3020 | } | |
3021 | ||
015e08bc | 3022 | ofport->may_enable = enable; |
abe529af BP |
3023 | } |
3024 | ||
3025 | static void | |
3026 | port_wait(struct ofport_dpif *ofport) | |
3027 | { | |
3028 | if (ofport->cfm) { | |
3029 | cfm_wait(ofport->cfm); | |
3030 | } | |
3031 | } | |
3032 | ||
3033 | static int | |
3034 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3035 | struct ofproto_port *ofproto_port) | |
3036 | { | |
3037 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3038 | struct dpif_port dpif_port; | |
3039 | int error; | |
3040 | ||
0a740f48 EJ |
3041 | if (sset_contains(&ofproto->ghost_ports, devname)) { |
3042 | const char *type = netdev_get_type_from_name(devname); | |
3043 | ||
3044 | /* We may be called before ofproto->up.port_by_name is populated with | |
3045 | * the appropriate ofport. For this reason, we must get the name and | |
3046 | * type from the netdev layer directly. */ | |
3047 | if (type) { | |
3048 | const struct ofport *ofport; | |
3049 | ||
3050 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3051 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3052 | ofproto_port->name = xstrdup(devname); | |
3053 | ofproto_port->type = xstrdup(type); | |
3054 | return 0; | |
3055 | } | |
3056 | return ENODEV; | |
3057 | } | |
3058 | ||
acf60855 JP |
3059 | if (!sset_contains(&ofproto->ports, devname)) { |
3060 | return ENODEV; | |
3061 | } | |
3062 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3063 | devname, &dpif_port); | |
abe529af | 3064 | if (!error) { |
e1b1d06a | 3065 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); |
abe529af BP |
3066 | } |
3067 | return error; | |
3068 | } | |
3069 | ||
3070 | static int | |
e1b1d06a | 3071 | port_add(struct ofproto *ofproto_, struct netdev *netdev) |
abe529af BP |
3072 | { |
3073 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 EJ |
3074 | const char *dp_port_name = netdev_vport_get_dpif_port(netdev); |
3075 | const char *devname = netdev_get_name(netdev); | |
abe529af | 3076 | |
0a740f48 EJ |
3077 | if (netdev_vport_is_patch(netdev)) { |
3078 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3079 | return 0; | |
3080 | } | |
3081 | ||
b9ad7294 | 3082 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { |
7d82ab2e KM |
3083 | uint32_t port_no = UINT32_MAX; |
3084 | int error; | |
3085 | ||
3086 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
b9ad7294 EJ |
3087 | if (error) { |
3088 | return error; | |
3089 | } | |
7d82ab2e KM |
3090 | if (netdev_get_tunnel_config(netdev)) { |
3091 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3092 | } | |
acf60855 | 3093 | } |
b9ad7294 EJ |
3094 | |
3095 | if (netdev_get_tunnel_config(netdev)) { | |
3096 | sset_add(&ofproto->ghost_ports, devname); | |
b9ad7294 EJ |
3097 | } else { |
3098 | sset_add(&ofproto->ports, devname); | |
3099 | } | |
3100 | return 0; | |
3101 | } | |
3102 | ||
abe529af BP |
3103 | static int |
3104 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3105 | { | |
3106 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
b9ad7294 | 3107 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); |
e1b1d06a | 3108 | int error = 0; |
abe529af | 3109 | |
b9ad7294 EJ |
3110 | if (!ofport) { |
3111 | return 0; | |
e1b1d06a | 3112 | } |
b9ad7294 EJ |
3113 | |
3114 | sset_find_and_delete(&ofproto->ghost_ports, | |
3115 | netdev_get_name(ofport->up.netdev)); | |
a614d823 KM |
3116 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
3117 | if (!ofport->tnl_port) { | |
b9ad7294 EJ |
3118 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); |
3119 | if (!error) { | |
abe529af BP |
3120 | /* The caller is going to close ofport->up.netdev. If this is a |
3121 | * bonded port, then the bond is using that netdev, so remove it | |
3122 | * from the bond. The client will need to reconfigure everything | |
3123 | * after deleting ports, so then the slave will get re-added. */ | |
3124 | bundle_remove(&ofport->up); | |
3125 | } | |
3126 | } | |
3127 | return error; | |
3128 | } | |
3129 | ||
6527c598 PS |
3130 | static int |
3131 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3132 | { | |
3133 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3134 | int error; | |
3135 | ||
3136 | error = netdev_get_stats(ofport->up.netdev, stats); | |
3137 | ||
ee382d89 | 3138 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { |
6527c598 PS |
3139 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
3140 | ||
3141 | /* ofproto->stats.tx_packets represents packets that we created | |
3142 | * internally and sent to some port (e.g. packets sent with | |
3143 | * send_packet()). Account for them as if they had come from | |
3144 | * OFPP_LOCAL and got forwarded. */ | |
3145 | ||
3146 | if (stats->rx_packets != UINT64_MAX) { | |
3147 | stats->rx_packets += ofproto->stats.tx_packets; | |
3148 | } | |
3149 | ||
3150 | if (stats->rx_bytes != UINT64_MAX) { | |
3151 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3152 | } | |
3153 | ||
3154 | /* ofproto->stats.rx_packets represents packets that were received on | |
3155 | * some port and we processed internally and dropped (e.g. STP). | |
4e090bc7 | 3156 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ |
6527c598 PS |
3157 | |
3158 | if (stats->tx_packets != UINT64_MAX) { | |
3159 | stats->tx_packets += ofproto->stats.rx_packets; | |
3160 | } | |
3161 | ||
3162 | if (stats->tx_bytes != UINT64_MAX) { | |
3163 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3164 | } | |
3165 | } | |
3166 | ||
3167 | return error; | |
3168 | } | |
3169 | ||
3170 | /* Account packets for LOCAL port. */ | |
3171 | static void | |
3172 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
3173 | size_t tx_size, size_t rx_size) | |
3174 | { | |
3175 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3176 | ||
3177 | if (rx_size) { | |
3178 | ofproto->stats.rx_packets++; | |
3179 | ofproto->stats.rx_bytes += rx_size; | |
3180 | } | |
3181 | if (tx_size) { | |
3182 | ofproto->stats.tx_packets++; | |
3183 | ofproto->stats.tx_bytes += tx_size; | |
3184 | } | |
3185 | } | |
3186 | ||
abe529af | 3187 | struct port_dump_state { |
acf60855 JP |
3188 | uint32_t bucket; |
3189 | uint32_t offset; | |
0a740f48 | 3190 | bool ghost; |
da78d43d BP |
3191 | |
3192 | struct ofproto_port port; | |
3193 | bool has_port; | |
abe529af BP |
3194 | }; |
3195 | ||
3196 | static int | |
acf60855 | 3197 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) |
abe529af | 3198 | { |
0a740f48 | 3199 | *statep = xzalloc(sizeof(struct port_dump_state)); |
abe529af BP |
3200 | return 0; |
3201 | } | |
3202 | ||
3203 | static int | |
b9ad7294 | 3204 | port_dump_next(const struct ofproto *ofproto_, void *state_, |
abe529af BP |
3205 | struct ofproto_port *port) |
3206 | { | |
e1b1d06a | 3207 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); |
abe529af | 3208 | struct port_dump_state *state = state_; |
0a740f48 | 3209 | const struct sset *sset; |
acf60855 | 3210 | struct sset_node *node; |
abe529af | 3211 | |
da78d43d BP |
3212 | if (state->has_port) { |
3213 | ofproto_port_destroy(&state->port); | |
3214 | state->has_port = false; | |
3215 | } | |
0a740f48 EJ |
3216 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; |
3217 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
acf60855 JP |
3218 | int error; |
3219 | ||
da78d43d BP |
3220 | error = port_query_by_name(ofproto_, node->name, &state->port); |
3221 | if (!error) { | |
3222 | *port = state->port; | |
3223 | state->has_port = true; | |
3224 | return 0; | |
3225 | } else if (error != ENODEV) { | |
acf60855 JP |
3226 | return error; |
3227 | } | |
abe529af | 3228 | } |
acf60855 | 3229 | |
0a740f48 EJ |
3230 | if (!state->ghost) { |
3231 | state->ghost = true; | |
3232 | state->bucket = 0; | |
3233 | state->offset = 0; | |
3234 | return port_dump_next(ofproto_, state_, port); | |
3235 | } | |
3236 | ||
acf60855 | 3237 | return EOF; |
abe529af BP |
3238 | } |
3239 | ||
3240 | static int | |
3241 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3242 | { | |
3243 | struct port_dump_state *state = state_; | |
3244 | ||
da78d43d BP |
3245 | if (state->has_port) { |
3246 | ofproto_port_destroy(&state->port); | |
3247 | } | |
abe529af BP |
3248 | free(state); |
3249 | return 0; | |
3250 | } | |
3251 | ||
3252 | static int | |
3253 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3254 | { | |
3255 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 JP |
3256 | |
3257 | if (ofproto->port_poll_errno) { | |
3258 | int error = ofproto->port_poll_errno; | |
3259 | ofproto->port_poll_errno = 0; | |
3260 | return error; | |
3261 | } | |
3262 | ||
3263 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3264 | return EAGAIN; | |
3265 | } | |
3266 | ||
3267 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3268 | return 0; | |
abe529af BP |
3269 | } |
3270 | ||
3271 | static void | |
3272 | port_poll_wait(const struct ofproto *ofproto_) | |
3273 | { | |
3274 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
acf60855 | 3275 | dpif_port_poll_wait(ofproto->backer->dpif); |
abe529af BP |
3276 | } |
3277 | ||
3278 | static int | |
3279 | port_is_lacp_current(const struct ofport *ofport_) | |
3280 | { | |
3281 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3282 | return (ofport->bundle && ofport->bundle->lacp | |
3283 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3284 | : -1); | |
3285 | } | |
3286 | \f | |
3287 | /* Upcall handling. */ | |
3288 | ||
501f8d1f BP |
3289 | /* Flow miss batching. |
3290 | * | |
3291 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3292 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3293 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3294 | * more packets, plus possibly installing the flow in the dpif. | |
3295 | * | |
3296 | * So far we only batch the operations that affect flow setup time the most. | |
3297 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3298 | struct flow_miss { | |
3299 | struct hmap_node hmap_node; | |
acf60855 | 3300 | struct ofproto_dpif *ofproto; |
501f8d1f | 3301 | struct flow flow; |
b0f7b9b5 | 3302 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
3303 | const struct nlattr *key; |
3304 | size_t key_len; | |
14f94f9a | 3305 | struct initial_vals initial_vals; |
501f8d1f | 3306 | struct list packets; |
6a7e895f | 3307 | enum dpif_upcall_type upcall_type; |
a088a1ff | 3308 | uint32_t odp_in_port; |
501f8d1f BP |
3309 | }; |
3310 | ||
3311 | struct flow_miss_op { | |
c2b565b5 | 3312 | struct dpif_op dpif_op; |
5fe20d5d BP |
3313 | void *garbage; /* Pointer to pass to free(), NULL if none. */ |
3314 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
3315 | }; |
3316 | ||
62cd7072 BP |
3317 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
3318 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 3319 | * configurations. */ |
62cd7072 | 3320 | static void |
a39edbd4 | 3321 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 3322 | const struct flow *flow) |
62cd7072 BP |
3323 | { |
3324 | struct ofputil_packet_in pin; | |
3325 | ||
3e3252fa EJ |
3326 | pin.packet = packet->data; |
3327 | pin.packet_len = packet->size; | |
62cd7072 | 3328 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 3329 | pin.controller_id = 0; |
54834960 EJ |
3330 | |
3331 | pin.table_id = 0; | |
3332 | pin.cookie = 0; | |
3333 | ||
62cd7072 | 3334 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
3335 | |
3336 | flow_get_metadata(flow, &pin.fmd); | |
3337 | ||
d8653c38 | 3338 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
3339 | } |
3340 | ||
6a7e895f | 3341 | static enum slow_path_reason |
abe529af | 3342 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
ffaef958 | 3343 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) |
abe529af | 3344 | { |
b6e001b6 | 3345 | if (!ofport) { |
6a7e895f | 3346 | return 0; |
ffaef958 | 3347 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 3348 | if (packet) { |
abe529af BP |
3349 | cfm_process_heartbeat(ofport->cfm, packet); |
3350 | } | |
6a7e895f | 3351 | return SLOW_CFM; |
b6e001b6 EJ |
3352 | } else if (ofport->bundle && ofport->bundle->lacp |
3353 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3354 | if (packet) { | |
3355 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 3356 | } |
6a7e895f | 3357 | return SLOW_LACP; |
21f7563c JP |
3358 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
3359 | if (packet) { | |
3360 | stp_process_packet(ofport, packet); | |
3361 | } | |
6a7e895f | 3362 | return SLOW_STP; |
ffaef958 BP |
3363 | } else { |
3364 | return 0; | |
abe529af | 3365 | } |
abe529af BP |
3366 | } |
3367 | ||
501f8d1f | 3368 | static struct flow_miss * |
ddbc5954 JP |
3369 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, |
3370 | const struct flow *flow, uint32_t hash) | |
abe529af | 3371 | { |
501f8d1f | 3372 | struct flow_miss *miss; |
abe529af | 3373 | |
501f8d1f | 3374 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
ddbc5954 | 3375 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { |
501f8d1f BP |
3376 | return miss; |
3377 | } | |
3378 | } | |
abe529af | 3379 | |
b23cdad9 | 3380 | return NULL; |
501f8d1f | 3381 | } |
abe529af | 3382 | |
9d6ac44e BP |
3383 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
3384 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3385 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3386 | * returned op->subfacet, and if anything needs to be freed after processing | |
3387 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 3388 | static void |
9d6ac44e BP |
3389 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
3390 | struct flow_miss_op *op) | |
501f8d1f | 3391 | { |
14f94f9a | 3392 | if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) { |
9d6ac44e BP |
3393 | /* This packet was received on a VLAN splinter port. We |
3394 | * added a VLAN to the packet to make the packet resemble | |
3395 | * the flow, but the actions were composed assuming that | |
3396 | * the packet contained no VLAN. So, we must remove the | |
3397 | * VLAN header from the packet before trying to execute the | |
3398 | * actions. */ | |
3399 | eth_pop_vlan(packet); | |
3400 | } | |
3401 | ||
9d6ac44e BP |
3402 | op->garbage = NULL; |
3403 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
3404 | op->dpif_op.u.execute.key = miss->key; | |
3405 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3406 | op->dpif_op.u.execute.packet = packet; | |
3407 | } | |
3408 | ||
3409 | /* Helper for handle_flow_miss_without_facet() and | |
3410 | * handle_flow_miss_with_facet(). */ | |
3411 | static void | |
3412 | handle_flow_miss_common(struct rule_dpif *rule, | |
3413 | struct ofpbuf *packet, const struct flow *flow) | |
3414 | { | |
3415 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3416 | ||
3417 | ofproto->n_matches++; | |
3418 | ||
3419 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
3420 | /* | |
3421 | * Extra-special case for fail-open mode. | |
3422 | * | |
3423 | * We are in fail-open mode and the packet matched the fail-open | |
3424 | * rule, but we are connected to a controller too. We should send | |
3425 | * the packet up to the controller in the hope that it will try to | |
3426 | * set up a flow and thereby allow us to exit fail-open. | |
3427 | * | |
3428 | * See the top-level comment in fail-open.c for more information. | |
3429 | */ | |
3430 | send_packet_in_miss(ofproto, packet, flow); | |
3431 | } | |
3432 | } | |
3433 | ||
3434 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3435 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3436 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3437 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3438 | * the benefits, so when the datapath holds a large number of flows we impose | |
3439 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3440 | static bool | |
3441 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
3442 | struct flow_miss *miss, uint32_t hash) | |
3443 | { | |
3444 | if (!ofproto->governor) { | |
3445 | size_t n_subfacets; | |
3446 | ||
3447 | n_subfacets = hmap_count(&ofproto->subfacets); | |
3448 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
3449 | return true; | |
3450 | } | |
3451 | ||
3452 | ofproto->governor = governor_create(ofproto->up.name); | |
3453 | } | |
3454 | ||
3455 | return governor_should_install_flow(ofproto->governor, hash, | |
3456 | list_size(&miss->packets)); | |
3457 | } | |
3458 | ||
3459 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3460 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3461 | * increment '*n_ops'. */ | |
3462 | static void | |
3463 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
3464 | struct rule_dpif *rule, | |
3465 | struct flow_miss_op *ops, size_t *n_ops) | |
3466 | { | |
3467 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
a7752d4a | 3468 | long long int now = time_msec(); |
9d6ac44e | 3469 | struct action_xlate_ctx ctx; |
530a1d91 | 3470 | struct ofpbuf *packet; |
2b459b83 | 3471 | |
9d6ac44e BP |
3472 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
3473 | struct flow_miss_op *op = &ops[*n_ops]; | |
3474 | struct dpif_flow_stats stats; | |
3475 | struct ofpbuf odp_actions; | |
abe529af | 3476 | |
9d6ac44e | 3477 | COVERAGE_INC(facet_suppress); |
501f8d1f | 3478 | |
9d6ac44e | 3479 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 3480 | |
a7752d4a | 3481 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); |
9d6ac44e | 3482 | rule_credit_stats(rule, &stats); |
abe529af | 3483 | |
14f94f9a JP |
3484 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, |
3485 | &miss->initial_vals, rule, 0, packet); | |
9d6ac44e | 3486 | ctx.resubmit_stats = &stats; |
f25d0cf3 | 3487 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
9d6ac44e | 3488 | &odp_actions); |
abe529af | 3489 | |
9d6ac44e BP |
3490 | if (odp_actions.size) { |
3491 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
3492 | ||
3493 | init_flow_miss_execute_op(miss, packet, op); | |
3494 | execute->actions = odp_actions.data; | |
3495 | execute->actions_len = odp_actions.size; | |
3496 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3497 | ||
3498 | (*n_ops)++; | |
3499 | } else { | |
3500 | ofpbuf_uninit(&odp_actions); | |
3501 | } | |
abe529af | 3502 | } |
9d6ac44e BP |
3503 | } |
3504 | ||
3505 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
459b16a1 BP |
3506 | * operations to 'ops', incrementing '*n_ops' for each new op. |
3507 | * | |
3508 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3509 | * This is really important only for new facets: if we just called time_msec() | |
3510 | * here, then the new subfacet or its packets could look (occasionally) as | |
3511 | * though it was used some time after the facet was used. That can make a | |
3512 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3513 | * e.g. NetFlow statistics. */ | |
9d6ac44e BP |
3514 | static void |
3515 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
459b16a1 | 3516 | long long int now, |
9d6ac44e BP |
3517 | struct flow_miss_op *ops, size_t *n_ops) |
3518 | { | |
6a7e895f BP |
3519 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3520 | enum subfacet_path want_path; | |
9d6ac44e BP |
3521 | struct subfacet *subfacet; |
3522 | struct ofpbuf *packet; | |
abe529af | 3523 | |
a088a1ff | 3524 | subfacet = subfacet_create(facet, miss, now); |
b0f7b9b5 | 3525 | |
530a1d91 | 3526 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 3527 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 3528 | struct dpif_flow_stats stats; |
5fe20d5d | 3529 | struct ofpbuf odp_actions; |
67d91f78 | 3530 | |
9d6ac44e | 3531 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 3532 | |
5fe20d5d | 3533 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
6a7e895f | 3534 | if (!subfacet->actions || subfacet->slow) { |
5fe20d5d | 3535 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 3536 | } |
67d91f78 | 3537 | |
459b16a1 | 3538 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); |
15baa734 | 3539 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 3540 | |
9d6ac44e BP |
3541 | if (subfacet->actions_len) { |
3542 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 3543 | |
9d6ac44e | 3544 | init_flow_miss_execute_op(miss, packet, op); |
6a7e895f | 3545 | if (!subfacet->slow) { |
9d6ac44e BP |
3546 | execute->actions = subfacet->actions; |
3547 | execute->actions_len = subfacet->actions_len; | |
3548 | ofpbuf_uninit(&odp_actions); | |
3549 | } else { | |
3550 | execute->actions = odp_actions.data; | |
3551 | execute->actions_len = odp_actions.size; | |
3552 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3553 | } | |
999fba59 | 3554 | |
9d6ac44e | 3555 | (*n_ops)++; |
5fe20d5d | 3556 | } else { |
9d6ac44e | 3557 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 3558 | } |
501f8d1f BP |
3559 | } |
3560 | ||
6a7e895f BP |
3561 | want_path = subfacet_want_path(subfacet->slow); |
3562 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
501f8d1f | 3563 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 3564 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 3565 | |
c84451a6 EJ |
3566 | subfacet->path = want_path; |
3567 | ||
5fe20d5d | 3568 | op->garbage = NULL; |
c2b565b5 | 3569 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
3570 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
3571 | put->key = miss->key; | |
3572 | put->key_len = miss->key_len; | |
6a7e895f BP |
3573 | if (want_path == SF_FAST_PATH) { |
3574 | put->actions = subfacet->actions; | |
3575 | put->actions_len = subfacet->actions_len; | |
3576 | } else { | |
3577 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
3578 | op->stub, sizeof op->stub, | |
3579 | &put->actions, &put->actions_len); | |
3580 | } | |
501f8d1f BP |
3581 | put->stats = NULL; |
3582 | } | |
3583 | } | |
3584 | ||
acf60855 JP |
3585 | /* Handles flow miss 'miss'. May add any required datapath operations |
3586 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
9d6ac44e | 3587 | static void |
acf60855 JP |
3588 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, |
3589 | size_t *n_ops) | |
9d6ac44e | 3590 | { |
acf60855 | 3591 | struct ofproto_dpif *ofproto = miss->ofproto; |
9d6ac44e | 3592 | struct facet *facet; |
459b16a1 | 3593 | long long int now; |
9d6ac44e BP |
3594 | uint32_t hash; |
3595 | ||
3596 | /* The caller must ensure that miss->hmap_node.hash contains | |
3597 | * flow_hash(miss->flow, 0). */ | |
3598 | hash = miss->hmap_node.hash; | |
3599 | ||
3600 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3601 | if (!facet) { | |
c57b2226 BP |
3602 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); |
3603 | ||
3604 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
9d6ac44e BP |
3605 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); |
3606 | return; | |
3607 | } | |
3608 | ||
3609 | facet = facet_create(rule, &miss->flow, hash); | |
459b16a1 BP |
3610 | now = facet->used; |
3611 | } else { | |
3612 | now = time_msec(); | |
9d6ac44e | 3613 | } |
459b16a1 | 3614 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); |
9d6ac44e BP |
3615 | } |
3616 | ||
8f73d537 EJ |
3617 | static struct drop_key * |
3618 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3619 | size_t key_len) | |
3620 | { | |
3621 | struct drop_key *drop_key; | |
3622 | ||
3623 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3624 | &backer->drop_keys) { | |
3625 | if (drop_key->key_len == key_len | |
3626 | && !memcmp(drop_key->key, key, key_len)) { | |
3627 | return drop_key; | |
3628 | } | |
3629 | } | |
3630 | return NULL; | |
3631 | } | |
3632 | ||
3633 | static void | |
3634 | drop_key_clear(struct dpif_backer *backer) | |
3635 | { | |
3636 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3637 | struct drop_key *drop_key, *next; | |
3638 | ||
3639 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3640 | int error; | |
3641 | ||
3642 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3643 | NULL); | |
3644 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3645 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3646 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3647 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3648 | ds_cstr(&ds)); | |
3649 | ds_destroy(&ds); | |
3650 | } | |
3651 | ||
3652 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3653 | free(drop_key->key); | |
3654 | free(drop_key); | |
3655 | } | |
3656 | } | |
3657 | ||
e09ee259 EJ |
3658 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' |
3659 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3660 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3661 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3662 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3663 | * 'packet' ingressed. | |
e2a6ca36 | 3664 | * |
e09ee259 EJ |
3665 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets |
3666 | * 'flow''s in_port to OFPP_NONE. | |
3667 | * | |
3668 | * This function does post-processing on data returned from | |
3669 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3670 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3671 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3672 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3673 | * a VLAN header onto 'packet' (if it is nonnull). | |
3674 | * | |
c3f6c502 JP |
3675 | * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' |
3676 | * to the VLAN TCI with which the packet was really received, that is, the | |
3677 | * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from | |
3678 | * the value returned in flow->vlan_tci only for packets received on | |
3679 | * VLAN splinters.) Also, if received on an IP tunnel, sets | |
3680 | * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS. | |
e09ee259 | 3681 | * |
b9ad7294 EJ |
3682 | * Similarly, this function also includes some logic to help with tunnels. It |
3683 | * may modify 'flow' as necessary to make the tunneling implementation | |
3684 | * transparent to the upcall processing logic. | |
3685 | * | |
e09ee259 EJ |
3686 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, |
3687 | * or some other positive errno if there are other problems. */ | |
3688 | static int | |
3689 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3690 | const struct nlattr *key, size_t key_len, | |
3691 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3692 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
14f94f9a | 3693 | struct initial_vals *initial_vals) |
e84173dc | 3694 | { |
e09ee259 EJ |
3695 | const struct ofport_dpif *port; |
3696 | enum odp_key_fitness fitness; | |
b9ad7294 | 3697 | int error = ENODEV; |
e09ee259 EJ |
3698 | |
3699 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
e84173dc | 3700 | if (fitness == ODP_FIT_ERROR) { |
e09ee259 EJ |
3701 | error = EINVAL; |
3702 | goto exit; | |
3703 | } | |
3704 | ||
14f94f9a JP |
3705 | if (initial_vals) { |
3706 | initial_vals->vlan_tci = flow->vlan_tci; | |
c3f6c502 | 3707 | initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos; |
e84173dc | 3708 | } |
e84173dc | 3709 | |
e09ee259 EJ |
3710 | if (odp_in_port) { |
3711 | *odp_in_port = flow->in_port; | |
3712 | } | |
3713 | ||
b9ad7294 EJ |
3714 | if (tnl_port_should_receive(flow)) { |
3715 | const struct ofport *ofport = tnl_port_receive(flow); | |
3716 | if (!ofport) { | |
3717 | flow->in_port = OFPP_NONE; | |
3718 | goto exit; | |
3719 | } | |
3720 | port = ofport_dpif_cast(ofport); | |
e09ee259 | 3721 | |
b9ad7294 EJ |
3722 | /* We can't reproduce 'key' from 'flow'. */ |
3723 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
e09ee259 | 3724 | |
b9ad7294 EJ |
3725 | /* XXX: Since the tunnel module is not scoped per backer, it's |
3726 | * theoretically possible that we'll receive an ofport belonging to an | |
3727 | * entirely different datapath. In practice, this can't happen because | |
3728 | * no platforms has two separate datapaths which each support | |
3729 | * tunneling. */ | |
3730 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3731 | } else { | |
3732 | port = odp_port_to_ofport(backer, flow->in_port); | |
3733 | if (!port) { | |
3734 | flow->in_port = OFPP_NONE; | |
3735 | goto exit; | |
3736 | } | |
3737 | ||
3738 | flow->in_port = port->up.ofp_port; | |
3739 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3740 | if (packet) { | |
3741 | /* Make the packet resemble the flow, so that it gets sent to | |
3742 | * an OpenFlow controller properly, so that it looks correct | |
3743 | * for sFlow, and so that flow_extract() will get the correct | |
3744 | * vlan_tci if it is called on 'packet'. | |
3745 | * | |
3746 | * The allocated space inside 'packet' probably also contains | |
3747 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3748 | * a struct dpif_upcall (see the large comment on that | |
3749 | * structure definition), so pushing data on 'packet' is in | |
3750 | * general not a good idea since it could overwrite 'key' or | |
3751 | * free it as a side effect. However, it's OK in this special | |
3752 | * case because we know that 'packet' is inside a Netlink | |
3753 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3754 | * "struct nlattr", which is fine since we don't need that | |
3755 | * header anymore. */ | |
3756 | eth_push_vlan(packet, flow->vlan_tci); | |
3757 | } | |
3758 | /* We can't reproduce 'key' from 'flow'. */ | |
3759 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
52a90c29 BP |
3760 | } |
3761 | } | |
e09ee259 | 3762 | error = 0; |
52a90c29 | 3763 | |
b9ad7294 EJ |
3764 | if (ofproto) { |
3765 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3766 | } | |
3767 | ||
e09ee259 EJ |
3768 | exit: |
3769 | if (fitnessp) { | |
3770 | *fitnessp = fitness; | |
3771 | } | |
3772 | return error; | |
e84173dc BP |
3773 | } |
3774 | ||
501f8d1f | 3775 | static void |
acf60855 | 3776 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, |
501f8d1f BP |
3777 | size_t n_upcalls) |
3778 | { | |
3779 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3780 | struct flow_miss *miss; |
3781 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3782 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3783 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3784 | struct hmap todo; |
b23cdad9 | 3785 | int n_misses; |
501f8d1f BP |
3786 | size_t n_ops; |
3787 | size_t i; | |
3788 | ||
3789 | if (!n_upcalls) { | |
3790 | return; | |
3791 | } | |
3792 | ||
3793 | /* Construct the to-do list. | |
3794 | * | |
3795 | * This just amounts to extracting the flow from each packet and sticking | |
3796 | * the packets that have the same flow in the same "flow_miss" structure so | |
3797 | * that we can process them together. */ | |
3798 | hmap_init(&todo); | |
b23cdad9 | 3799 | n_misses = 0; |
501f8d1f | 3800 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3801 | struct flow_miss *miss = &misses[n_misses]; |
3802 | struct flow_miss *existing_miss; | |
acf60855 | 3803 | struct ofproto_dpif *ofproto; |
a088a1ff | 3804 | uint32_t odp_in_port; |
1d446463 | 3805 | struct flow flow; |
b23cdad9 | 3806 | uint32_t hash; |
e09ee259 | 3807 | int error; |
501f8d1f | 3808 | |
e09ee259 EJ |
3809 | error = ofproto_receive(backer, upcall->packet, upcall->key, |
3810 | upcall->key_len, &flow, &miss->key_fitness, | |
14f94f9a | 3811 | &ofproto, &odp_in_port, &miss->initial_vals); |
e09ee259 | 3812 | if (error == ENODEV) { |
8f73d537 EJ |
3813 | struct drop_key *drop_key; |
3814 | ||
acf60855 JP |
3815 | /* Received packet on port for which we couldn't associate |
3816 | * an ofproto. This can happen if a port is removed while | |
3817 | * traffic is being received. Print a rate-limited message | |
8f73d537 EJ |
3818 | * in case it happens frequently. Install a drop flow so |
3819 | * that future packets of the flow are inexpensively dropped | |
3820 | * in the kernel. */ | |
acf60855 JP |
3821 | VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, |
3822 | flow.in_port); | |
8f73d537 EJ |
3823 | |
3824 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
3825 | if (!drop_key) { | |
3826 | drop_key = xmalloc(sizeof *drop_key); | |
3827 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
3828 | drop_key->key_len = upcall->key_len; | |
3829 | ||
3830 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
3831 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
3832 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
3833 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
3834 | } | |
3835 | continue; | |
acf60855 | 3836 | } |
e09ee259 | 3837 | if (error) { |
b0f7b9b5 BP |
3838 | continue; |
3839 | } | |
735d7efb AZ |
3840 | |
3841 | ofproto->n_missed++; | |
72e8bf28 | 3842 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, |
1d446463 | 3843 | &flow.tunnel, flow.in_port, &miss->flow); |
501f8d1f | 3844 | |
501f8d1f | 3845 | /* Add other packets to a to-do list. */ |
b23cdad9 | 3846 | hash = flow_hash(&miss->flow, 0); |
ddbc5954 | 3847 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); |
b23cdad9 BP |
3848 | if (!existing_miss) { |
3849 | hmap_insert(&todo, &miss->hmap_node, hash); | |
acf60855 | 3850 | miss->ofproto = ofproto; |
b23cdad9 BP |
3851 | miss->key = upcall->key; |
3852 | miss->key_len = upcall->key_len; | |
6a7e895f | 3853 | miss->upcall_type = upcall->type; |
a088a1ff | 3854 | miss->odp_in_port = odp_in_port; |
b23cdad9 BP |
3855 | list_init(&miss->packets); |
3856 | ||
3857 | n_misses++; | |
3858 | } else { | |
3859 | miss = existing_miss; | |
3860 | } | |
501f8d1f BP |
3861 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3862 | } | |
3863 | ||
3864 | /* Process each element in the to-do list, constructing the set of | |
3865 | * operations to batch. */ | |
3866 | n_ops = 0; | |
33bb0caa | 3867 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
acf60855 | 3868 | handle_flow_miss(miss, flow_miss_ops, &n_ops); |
abe529af | 3869 | } |
cb22974d | 3870 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3871 | |
3872 | /* Execute batch. */ | |
3873 | for (i = 0; i < n_ops; i++) { | |
3874 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3875 | } | |
acf60855 | 3876 | dpif_operate(backer->dpif, dpif_ops, n_ops); |
501f8d1f | 3877 | |
c84451a6 | 3878 | /* Free memory. */ |
501f8d1f | 3879 | for (i = 0; i < n_ops; i++) { |
c84451a6 | 3880 | free(flow_miss_ops[i].garbage); |
501f8d1f | 3881 | } |
33bb0caa | 3882 | hmap_destroy(&todo); |
abe529af BP |
3883 | } |
3884 | ||
6a7e895f BP |
3885 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } |
3886 | classify_upcall(const struct dpif_upcall *upcall) | |
3887 | { | |
3888 | union user_action_cookie cookie; | |
3889 | ||
3890 | /* First look at the upcall type. */ | |
3891 | switch (upcall->type) { | |
3892 | case DPIF_UC_ACTION: | |
3893 | break; | |
3894 | ||
3895 | case DPIF_UC_MISS: | |
3896 | return MISS_UPCALL; | |
3897 | ||
3898 | case DPIF_N_UC_TYPES: | |
3899 | default: | |
3900 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3901 | return BAD_UPCALL; | |
3902 | } | |
3903 | ||
3904 | /* "action" upcalls need a closer look. */ | |
e995e3df BP |
3905 | if (!upcall->userdata) { |
3906 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
3907 | return BAD_UPCALL; | |
3908 | } | |
3909 | if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) { | |
3910 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", | |
3911 | nl_attr_get_size(upcall->userdata)); | |
3912 | return BAD_UPCALL; | |
3913 | } | |
3914 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie)); | |
6a7e895f BP |
3915 | switch (cookie.type) { |
3916 | case USER_ACTION_COOKIE_SFLOW: | |
3917 | return SFLOW_UPCALL; | |
3918 | ||
3919 | case USER_ACTION_COOKIE_SLOW_PATH: | |
3920 | return MISS_UPCALL; | |
3921 | ||
3922 | case USER_ACTION_COOKIE_UNSPEC: | |
3923 | default: | |
e995e3df BP |
3924 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, |
3925 | nl_attr_get_u64(upcall->userdata)); | |
6a7e895f BP |
3926 | return BAD_UPCALL; |
3927 | } | |
3928 | } | |
3929 | ||
abe529af | 3930 | static void |
acf60855 | 3931 | handle_sflow_upcall(struct dpif_backer *backer, |
6a7e895f | 3932 | const struct dpif_upcall *upcall) |
abe529af | 3933 | { |
acf60855 | 3934 | struct ofproto_dpif *ofproto; |
1673e0e4 | 3935 | union user_action_cookie cookie; |
e84173dc | 3936 | struct flow flow; |
e1b1d06a | 3937 | uint32_t odp_in_port; |
abe529af | 3938 | |
e09ee259 EJ |
3939 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, |
3940 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
3941 | || !ofproto->sflow) { | |
e84173dc BP |
3942 | return; |
3943 | } | |
3944 | ||
e995e3df | 3945 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie)); |
e1b1d06a JP |
3946 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
3947 | odp_in_port, &cookie); | |
6ff686f2 PS |
3948 | } |
3949 | ||
9b16c439 | 3950 | static int |
acf60855 | 3951 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) |
6ff686f2 | 3952 | { |
9b16c439 | 3953 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3954 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3955 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3956 | int n_processed; | |
9b16c439 BP |
3957 | int n_misses; |
3958 | int i; | |
abe529af | 3959 | |
cb22974d | 3960 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3961 | |
9b16c439 | 3962 | n_misses = 0; |
90a7c55e | 3963 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3964 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3965 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3966 | int error; |
3967 | ||
90a7c55e BP |
3968 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3969 | sizeof miss_buf_stubs[n_misses]); | |
acf60855 | 3970 | error = dpif_recv(backer->dpif, upcall, buf); |
9b16c439 | 3971 | if (error) { |
90a7c55e | 3972 | ofpbuf_uninit(buf); |
9b16c439 BP |
3973 | break; |
3974 | } | |
3975 | ||
6a7e895f BP |
3976 | switch (classify_upcall(upcall)) { |
3977 | case MISS_UPCALL: | |
9b16c439 BP |
3978 | /* Handle it later. */ |
3979 | n_misses++; | |
3980 | break; | |
3981 | ||
6a7e895f | 3982 | case SFLOW_UPCALL: |
acf60855 | 3983 | handle_sflow_upcall(backer, upcall); |
6a7e895f BP |
3984 | ofpbuf_uninit(buf); |
3985 | break; | |
3986 | ||
3987 | case BAD_UPCALL: | |
3988 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3989 | break; |
3990 | } | |
abe529af | 3991 | } |
9b16c439 | 3992 | |
6a7e895f | 3993 | /* Handle deferred MISS_UPCALL processing. */ |
acf60855 | 3994 | handle_miss_upcalls(backer, misses, n_misses); |
90a7c55e BP |
3995 | for (i = 0; i < n_misses; i++) { |
3996 | ofpbuf_uninit(&miss_bufs[i]); | |
3997 | } | |
9b16c439 | 3998 | |
90a7c55e | 3999 | return n_processed; |
abe529af BP |
4000 | } |
4001 | \f | |
4002 | /* Flow expiration. */ | |
4003 | ||
b0f7b9b5 | 4004 | static int subfacet_max_idle(const struct ofproto_dpif *); |
acf60855 | 4005 | static void update_stats(struct dpif_backer *); |
abe529af | 4006 | static void rule_expire(struct rule_dpif *); |
b0f7b9b5 | 4007 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
4008 | |
4009 | /* This function is called periodically by run(). Its job is to collect | |
4010 | * updates for the flows that have been installed into the datapath, most | |
4011 | * importantly when they last were used, and then use that information to | |
4012 | * expire flows that have not been used recently. | |
4013 | * | |
4014 | * Returns the number of milliseconds after which it should be called again. */ | |
4015 | static int | |
acf60855 | 4016 | expire(struct dpif_backer *backer) |
abe529af | 4017 | { |
acf60855 JP |
4018 | struct ofproto_dpif *ofproto; |
4019 | int max_idle = INT32_MAX; | |
abe529af | 4020 | |
8f73d537 EJ |
4021 | /* Periodically clear out the drop keys in an effort to keep them |
4022 | * relatively few. */ | |
4023 | drop_key_clear(backer); | |
4024 | ||
acf60855 JP |
4025 | /* Update stats for each flow in the backer. */ |
4026 | update_stats(backer); | |
abe529af | 4027 | |
acf60855 | 4028 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { |
e503cc19 | 4029 | struct rule *rule, *next_rule; |
acf60855 | 4030 | int dp_max_idle; |
abe529af | 4031 | |
acf60855 JP |
4032 | if (ofproto->backer != backer) { |
4033 | continue; | |
4034 | } | |
0697b5c3 | 4035 | |
acf60855 JP |
4036 | /* Expire subfacets that have been idle too long. */ |
4037 | dp_max_idle = subfacet_max_idle(ofproto); | |
4038 | expire_subfacets(ofproto, dp_max_idle); | |
4039 | ||
4040 | max_idle = MIN(max_idle, dp_max_idle); | |
4041 | ||
4042 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4043 | * has passed. */ | |
e503cc19 SH |
4044 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, |
4045 | &ofproto->up.expirable) { | |
4046 | rule_expire(rule_dpif_cast(rule)); | |
0697b5c3 | 4047 | } |
abe529af | 4048 | |
acf60855 JP |
4049 | /* All outstanding data in existing flows has been accounted, so it's a |
4050 | * good time to do bond rebalancing. */ | |
4051 | if (ofproto->has_bonded_bundles) { | |
4052 | struct ofbundle *bundle; | |
abe529af | 4053 | |
acf60855 JP |
4054 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { |
4055 | if (bundle->bond) { | |
2cc3c58e | 4056 | bond_rebalance(bundle->bond, &backer->revalidate_set); |
acf60855 | 4057 | } |
abe529af BP |
4058 | } |
4059 | } | |
4060 | } | |
4061 | ||
acf60855 | 4062 | return MIN(max_idle, 1000); |
abe529af BP |
4063 | } |
4064 | ||
a218c879 BP |
4065 | /* Updates flow table statistics given that the datapath just reported 'stats' |
4066 | * as 'subfacet''s statistics. */ | |
4067 | static void | |
4068 | update_subfacet_stats(struct subfacet *subfacet, | |
4069 | const struct dpif_flow_stats *stats) | |
4070 | { | |
4071 | struct facet *facet = subfacet->facet; | |
4072 | ||
4073 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
4074 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
4075 | facet->packet_count += extra; | |
4076 | } else { | |
4077 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4078 | } | |
4079 | ||
4080 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
4081 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
4082 | } else { | |
4083 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4084 | } | |
4085 | ||
4086 | subfacet->dp_packet_count = stats->n_packets; | |
4087 | subfacet->dp_byte_count = stats->n_bytes; | |
4088 | ||
4089 | facet->tcp_flags |= stats->tcp_flags; | |
4090 | ||
4091 | subfacet_update_time(subfacet, stats->used); | |
4092 | if (facet->accounted_bytes < facet->byte_count) { | |
4093 | facet_learn(facet); | |
4094 | facet_account(facet); | |
4095 | facet->accounted_bytes = facet->byte_count; | |
4096 | } | |
4097 | facet_push_stats(facet); | |
4098 | } | |
4099 | ||
4100 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4101 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4102 | static void | |
acf60855 | 4103 | delete_unexpected_flow(struct ofproto_dpif *ofproto, |
a218c879 BP |
4104 | const struct nlattr *key, size_t key_len) |
4105 | { | |
4106 | if (!VLOG_DROP_WARN(&rl)) { | |
4107 | struct ds s; | |
4108 | ||
4109 | ds_init(&s); | |
4110 | odp_flow_key_format(key, key_len, &s); | |
acf60855 | 4111 | VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); |
a218c879 BP |
4112 | ds_destroy(&s); |
4113 | } | |
4114 | ||
4115 | COVERAGE_INC(facet_unexpected); | |
acf60855 | 4116 | dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); |
a218c879 BP |
4117 | } |
4118 | ||
abe529af BP |
4119 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
4120 | * | |
4121 | * This function also pushes statistics updates to rules which each facet | |
4122 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4123 | * facet changes the rule it resubmits into at some time in between | |
4124 | * update_stats() runs, it is possible that statistics accrued to the | |
4125 | * old rule will be incorrectly attributed to the new rule. This could be | |
4126 | * avoided by calling update_stats() whenever rules are created or | |
4127 | * deleted. However, the performance impact of making so many calls to the | |
4128 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
735d7efb AZ |
4129 | * |
4130 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4131 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4132 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4133 | * the hit or miss counts of br1. | |
abe529af BP |
4134 | */ |
4135 | static void | |
acf60855 | 4136 | update_stats(struct dpif_backer *backer) |
abe529af BP |
4137 | { |
4138 | const struct dpif_flow_stats *stats; | |
4139 | struct dpif_flow_dump dump; | |
4140 | const struct nlattr *key; | |
4141 | size_t key_len; | |
4142 | ||
acf60855 | 4143 | dpif_flow_dump_start(&dump, backer->dpif); |
abe529af | 4144 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
acf60855 | 4145 | struct flow flow; |
b0f7b9b5 | 4146 | struct subfacet *subfacet; |
acf60855 | 4147 | struct ofproto_dpif *ofproto; |
b9ad7294 | 4148 | struct ofport_dpif *ofport; |
acf60855 | 4149 | uint32_t key_hash; |
abe529af | 4150 | |
58c6adda EJ |
4151 | if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, |
4152 | NULL, NULL)) { | |
acf60855 JP |
4153 | continue; |
4154 | } | |
4155 | ||
b9ad7294 EJ |
4156 | ofport = get_ofp_port(ofproto, flow.in_port); |
4157 | if (ofport && ofport->tnl_port) { | |
4158 | netdev_vport_inc_rx(ofport->up.netdev, stats); | |
4159 | } | |
4160 | ||
acf60855 | 4161 | key_hash = odp_flow_key_hash(key, key_len); |
9566abf9 | 4162 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
6a7e895f BP |
4163 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
4164 | case SF_FAST_PATH: | |
735d7efb AZ |
4165 | /* Update ofproto_dpif's hit count. */ |
4166 | if (stats->n_packets > subfacet->dp_packet_count) { | |
4167 | uint64_t delta = stats->n_packets - subfacet->dp_packet_count; | |
4168 | dpif_stats_update_hit_count(ofproto, delta); | |
4169 | } | |
4170 | ||
a218c879 | 4171 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
4172 | break; |
4173 | ||
4174 | case SF_SLOW_PATH: | |
4175 | /* Stats are updated per-packet. */ | |
4176 | break; | |
4177 | ||
4178 | case SF_NOT_INSTALLED: | |
4179 | default: | |
acf60855 | 4180 | delete_unexpected_flow(ofproto, key, key_len); |
6a7e895f | 4181 | break; |
abe529af BP |
4182 | } |
4183 | } | |
4184 | dpif_flow_dump_done(&dump); | |
4185 | } | |
4186 | ||
4187 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
4188 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
4189 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4190 | * fold its statistic into its rule. */ | |
abe529af | 4191 | static int |
b0f7b9b5 | 4192 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
4193 | { |
4194 | /* | |
4195 | * Idle time histogram. | |
4196 | * | |
b0f7b9b5 BP |
4197 | * Most of the time a switch has a relatively small number of subfacets. |
4198 | * When this is the case we might as well keep statistics for all of them | |
4199 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
4200 | * well. |
4201 | * | |
b0f7b9b5 | 4202 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 4203 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
4204 | * significant. However, with a large number of subfacets it is likely |
4205 | * that only a few of them are "heavy hitters" that consume a large amount | |
4206 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4207 | * the kernel and maintaining in userspaces; other subfacets we can | |
4208 | * discard. | |
abe529af BP |
4209 | * |
4210 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 4211 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
4212 | * that is installed in the kernel gets dropped in the appropriate bucket. |
4213 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 4214 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 4215 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
4216 | * the most-recently-used bucket of subfacets is kept, so actually an |
4217 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
4218 | * (though the next run will delete most of those unless they receive |
4219 | * additional data). | |
abe529af | 4220 | * |
b0f7b9b5 BP |
4221 | * This requires a second pass through the subfacets, in addition to the |
4222 | * pass made by update_stats(), because the former function never looks at | |
4223 | * uninstallable subfacets. | |
abe529af BP |
4224 | */ |
4225 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4226 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4227 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 4228 | int total, subtotal, bucket; |
b0f7b9b5 | 4229 | struct subfacet *subfacet; |
abe529af BP |
4230 | long long int now; |
4231 | int i; | |
4232 | ||
b0f7b9b5 | 4233 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 4234 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
4235 | return N_BUCKETS * BUCKET_WIDTH; |
4236 | } | |
4237 | ||
4238 | /* Build histogram. */ | |
4239 | now = time_msec(); | |
b0f7b9b5 BP |
4240 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
4241 | long long int idle = now - subfacet->used; | |
abe529af BP |
4242 | int bucket = (idle <= 0 ? 0 |
4243 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4244 | : (unsigned int) idle / BUCKET_WIDTH); | |
4245 | buckets[bucket]++; | |
4246 | } | |
4247 | ||
4248 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
4249 | subtotal = bucket = 0; |
4250 | do { | |
4251 | subtotal += buckets[bucket++]; | |
084f5290 SH |
4252 | } while (bucket < N_BUCKETS && |
4253 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
4254 | |
4255 | if (VLOG_IS_DBG_ENABLED()) { | |
4256 | struct ds s; | |
4257 | ||
4258 | ds_init(&s); | |
4259 | ds_put_cstr(&s, "keep"); | |
4260 | for (i = 0; i < N_BUCKETS; i++) { | |
4261 | if (i == bucket) { | |
4262 | ds_put_cstr(&s, ", drop"); | |
4263 | } | |
4264 | if (buckets[i]) { | |
4265 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4266 | } | |
4267 | } | |
4268 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
4269 | ds_destroy(&s); | |
4270 | } | |
4271 | ||
4272 | return bucket * BUCKET_WIDTH; | |
4273 | } | |
4274 | ||
abe529af | 4275 | static void |
b0f7b9b5 | 4276 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 4277 | { |
625b0720 BP |
4278 | /* Cutoff time for most flows. */ |
4279 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4280 | ||
4281 | /* We really want to keep flows for special protocols around, so use a more | |
4282 | * conservative cutoff. */ | |
4283 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 4284 | |
b0f7b9b5 | 4285 | struct subfacet *subfacet, *next_subfacet; |
1d85f9e5 | 4286 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; |
b99d3cee | 4287 | int n_batch; |
abe529af | 4288 | |
b99d3cee | 4289 | n_batch = 0; |
b0f7b9b5 BP |
4290 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
4291 | &ofproto->subfacets) { | |
625b0720 BP |
4292 | long long int cutoff; |
4293 | ||
4294 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) | |
4295 | ? special_cutoff | |
4296 | : normal_cutoff); | |
b0f7b9b5 | 4297 | if (subfacet->used < cutoff) { |
6a7e895f | 4298 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee | 4299 | batch[n_batch++] = subfacet; |
1d85f9e5 JP |
4300 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { |
4301 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
b99d3cee BP |
4302 | n_batch = 0; |
4303 | } | |
4304 | } else { | |
4305 | subfacet_destroy(subfacet); | |
4306 | } | |
abe529af BP |
4307 | } |
4308 | } | |
b99d3cee BP |
4309 | |
4310 | if (n_batch > 0) { | |
1d85f9e5 | 4311 | subfacet_destroy_batch(ofproto, batch, n_batch); |
b99d3cee | 4312 | } |
abe529af BP |
4313 | } |
4314 | ||
4315 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4316 | * then delete it entirely. */ | |
4317 | static void | |
4318 | rule_expire(struct rule_dpif *rule) | |
4319 | { | |
abe529af BP |
4320 | struct facet *facet, *next_facet; |
4321 | long long int now; | |
4322 | uint8_t reason; | |
4323 | ||
e2a3d183 BP |
4324 | if (rule->up.pending) { |
4325 | /* We'll have to expire it later. */ | |
4326 | return; | |
4327 | } | |
4328 | ||
abe529af BP |
4329 | /* Has 'rule' expired? */ |
4330 | now = time_msec(); | |
4331 | if (rule->up.hard_timeout | |
308881af | 4332 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 4333 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 4334 | } else if (rule->up.idle_timeout |
1745cd08 | 4335 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
4336 | reason = OFPRR_IDLE_TIMEOUT; |
4337 | } else { | |
4338 | return; | |
4339 | } | |
4340 | ||
4341 | COVERAGE_INC(ofproto_dpif_expired); | |
4342 | ||
4343 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4344 | * timeout, because that only happens when the rule has no facets left.) */ | |
4345 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 4346 | facet_remove(facet); |
abe529af BP |
4347 | } |
4348 | ||
4349 | /* Get rid of the rule. */ | |
4350 | ofproto_rule_expire(&rule->up, reason); | |
4351 | } | |
4352 | \f | |
4353 | /* Facets. */ | |
4354 | ||
f3827897 | 4355 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
4356 | * |
4357 | * The caller must already have determined that no facet with an identical | |
4358 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
4359 | * the ofproto's classifier table. |
4360 | * | |
2b459b83 BP |
4361 | * 'hash' must be the return value of flow_hash(flow, 0). |
4362 | * | |
b0f7b9b5 BP |
4363 | * The facet will initially have no subfacets. The caller should create (at |
4364 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 4365 | static struct facet * |
2b459b83 | 4366 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
4367 | { |
4368 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4369 | struct facet *facet; | |
4370 | ||
4371 | facet = xzalloc(sizeof *facet); | |
4372 | facet->used = time_msec(); | |
2b459b83 | 4373 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
4374 | list_push_back(&rule->facets, &facet->list_node); |
4375 | facet->rule = rule; | |
4376 | facet->flow = *flow; | |
b0f7b9b5 | 4377 | list_init(&facet->subfacets); |
abe529af BP |
4378 | netflow_flow_init(&facet->nf_flow); |
4379 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4380 | ||
6cf474d7 EJ |
4381 | facet->learn_rl = time_msec() + 500; |
4382 | ||
abe529af BP |
4383 | return facet; |
4384 | } | |
4385 | ||
4386 | static void | |
4387 | facet_free(struct facet *facet) | |
4388 | { | |
abe529af BP |
4389 | free(facet); |
4390 | } | |
4391 | ||
3d9e05f8 | 4392 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
0a740f48 | 4393 | * 'packet', which arrived on 'in_port'. */ |
3d9e05f8 BP |
4394 | static bool |
4395 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4396 | const struct nlattr *odp_actions, size_t actions_len, | |
4397 | struct ofpbuf *packet) | |
4398 | { | |
4399 | struct odputil_keybuf keybuf; | |
4400 | struct ofpbuf key; | |
4401 | int error; | |
4402 | ||
6ff686f2 | 4403 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
4404 | odp_flow_key_from_flow(&key, flow, |
4405 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
80e5eed9 | 4406 | |
acf60855 | 4407 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, |
6ff686f2 | 4408 | odp_actions, actions_len, packet); |
6ff686f2 | 4409 | return !error; |
abe529af BP |
4410 | } |
4411 | ||
abe529af BP |
4412 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
4413 | * | |
4414 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 4415 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
4416 | * |
4417 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4418 | */ | |
4419 | static void | |
15baa734 | 4420 | facet_remove(struct facet *facet) |
abe529af | 4421 | { |
15baa734 | 4422 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4423 | struct subfacet *subfacet, *next_subfacet; |
4424 | ||
cb22974d | 4425 | ovs_assert(!list_is_empty(&facet->subfacets)); |
551a2f6c BP |
4426 | |
4427 | /* First uninstall all of the subfacets to get final statistics. */ | |
4428 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 4429 | subfacet_uninstall(subfacet); |
551a2f6c BP |
4430 | } |
4431 | ||
4432 | /* Flush the final stats to the rule. | |
4433 | * | |
4434 | * This might require us to have at least one subfacet around so that we | |
4435 | * can use its actions for accounting in facet_account(), which is why we | |
4436 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 4437 | facet_flush_stats(facet); |
551a2f6c BP |
4438 | |
4439 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
4440 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
4441 | &facet->subfacets) { | |
15baa734 | 4442 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 4443 | } |
abe529af BP |
4444 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
4445 | list_remove(&facet->list_node); | |
4446 | facet_free(facet); | |
4447 | } | |
4448 | ||
3de9590b BP |
4449 | /* Feed information from 'facet' back into the learning table to keep it in |
4450 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 4451 | static void |
3de9590b | 4452 | facet_learn(struct facet *facet) |
abe529af | 4453 | { |
15baa734 | 4454 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
14f94f9a JP |
4455 | struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets), |
4456 | struct subfacet, list_node); | |
3de9590b | 4457 | struct action_xlate_ctx ctx; |
abe529af | 4458 | |
6cf474d7 EJ |
4459 | if (time_msec() < facet->learn_rl) { |
4460 | return; | |
4461 | } | |
4462 | ||
4463 | facet->learn_rl = time_msec() + 500; | |
4464 | ||
3de9590b BP |
4465 | if (!facet->has_learn |
4466 | && !facet->has_normal | |
4467 | && (!facet->has_fin_timeout | |
4468 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
4469 | return; |
4470 | } | |
abe529af | 4471 | |
3de9590b | 4472 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
14f94f9a | 4473 | &subfacet->initial_vals, |
3de9590b BP |
4474 | facet->rule, facet->tcp_flags, NULL); |
4475 | ctx.may_learn = true; | |
f25d0cf3 BP |
4476 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, |
4477 | facet->rule->up.ofpacts_len); | |
3de9590b BP |
4478 | } |
4479 | ||
4480 | static void | |
4481 | facet_account(struct facet *facet) | |
4482 | { | |
4483 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
1a1e3a0a | 4484 | struct subfacet *subfacet = facet_get_subfacet(facet); |
3de9590b BP |
4485 | const struct nlattr *a; |
4486 | unsigned int left; | |
4487 | ovs_be16 vlan_tci; | |
4488 | uint64_t n_bytes; | |
abe529af | 4489 | |
75a75043 | 4490 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
4491 | return; |
4492 | } | |
3de9590b | 4493 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
4494 | |
4495 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4496 | * as a basis. We also need to track the actual VLAN on which the packet | |
4497 | * is going to be sent to ensure that it matches the one passed to | |
4498 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
4499 | * hash bucket.) |
4500 | * | |
4501 | * We use the actions from an arbitrary subfacet because they should all | |
4502 | * be equally valid for our purpose. */ | |
d78be13b | 4503 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
4504 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
4505 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 4506 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 4507 | struct ofport_dpif *port; |
abe529af | 4508 | |
d78be13b | 4509 | switch (nl_attr_type(a)) { |
df2c07f4 | 4510 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
4511 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
4512 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 4513 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 4514 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 4515 | } |
d78be13b BP |
4516 | break; |
4517 | ||
fea393b1 BP |
4518 | case OVS_ACTION_ATTR_POP_VLAN: |
4519 | vlan_tci = htons(0); | |
d78be13b BP |
4520 | break; |
4521 | ||
fea393b1 BP |
4522 | case OVS_ACTION_ATTR_PUSH_VLAN: |
4523 | vlan = nl_attr_get(a); | |
4524 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 4525 | break; |
abe529af BP |
4526 | } |
4527 | } | |
4528 | } | |
4529 | ||
abe529af BP |
4530 | /* Returns true if the only action for 'facet' is to send to the controller. |
4531 | * (We don't report NetFlow expiration messages for such facets because they | |
4532 | * are just part of the control logic for the network, not real traffic). */ | |
4533 | static bool | |
4534 | facet_is_controller_flow(struct facet *facet) | |
4535 | { | |
f25d0cf3 BP |
4536 | if (facet) { |
4537 | const struct rule *rule = &facet->rule->up; | |
4538 | const struct ofpact *ofpacts = rule->ofpacts; | |
4539 | size_t ofpacts_len = rule->ofpacts_len; | |
4540 | ||
dd30ff28 BP |
4541 | if (ofpacts_len > 0 && |
4542 | ofpacts->type == OFPACT_CONTROLLER && | |
f25d0cf3 BP |
4543 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { |
4544 | return true; | |
4545 | } | |
4546 | } | |
4547 | return false; | |
abe529af BP |
4548 | } |
4549 | ||
4550 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4551 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4552 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4553 | * its packet and byte counts before this function is called. */ | |
4554 | static void | |
15baa734 | 4555 | facet_flush_stats(struct facet *facet) |
abe529af | 4556 | { |
15baa734 | 4557 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4558 | struct subfacet *subfacet; |
4559 | ||
4560 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
cb22974d BP |
4561 | ovs_assert(!subfacet->dp_byte_count); |
4562 | ovs_assert(!subfacet->dp_packet_count); | |
b0f7b9b5 | 4563 | } |
abe529af BP |
4564 | |
4565 | facet_push_stats(facet); | |
3de9590b BP |
4566 | if (facet->accounted_bytes < facet->byte_count) { |
4567 | facet_account(facet); | |
4568 | facet->accounted_bytes = facet->byte_count; | |
4569 | } | |
abe529af BP |
4570 | |
4571 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4572 | struct ofexpired expired; | |
4573 | expired.flow = facet->flow; | |
4574 | expired.packet_count = facet->packet_count; | |
4575 | expired.byte_count = facet->byte_count; | |
4576 | expired.used = facet->used; | |
4577 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4578 | } | |
4579 | ||
4580 | facet->rule->packet_count += facet->packet_count; | |
4581 | facet->rule->byte_count += facet->byte_count; | |
4582 | ||
4583 | /* Reset counters to prevent double counting if 'facet' ever gets | |
4584 | * reinstalled. */ | |
bbb5d219 | 4585 | facet_reset_counters(facet); |
abe529af BP |
4586 | |
4587 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 4588 | facet->tcp_flags = 0; |
abe529af BP |
4589 | } |
4590 | ||
4591 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4592 | * Returns it if found, otherwise a null pointer. | |
4593 | * | |
2b459b83 BP |
4594 | * 'hash' must be the return value of flow_hash(flow, 0). |
4595 | * | |
abe529af BP |
4596 | * The returned facet might need revalidation; use facet_lookup_valid() |
4597 | * instead if that is important. */ | |
4598 | static struct facet * | |
2b459b83 BP |
4599 | facet_find(struct ofproto_dpif *ofproto, |
4600 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
4601 | { |
4602 | struct facet *facet; | |
4603 | ||
2b459b83 | 4604 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
4605 | if (flow_equal(flow, &facet->flow)) { |
4606 | return facet; | |
4607 | } | |
4608 | } | |
4609 | ||
4610 | return NULL; | |
4611 | } | |
4612 | ||
4613 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4614 | * Returns it if found, otherwise a null pointer. | |
4615 | * | |
2b459b83 BP |
4616 | * 'hash' must be the return value of flow_hash(flow, 0). |
4617 | * | |
abe529af BP |
4618 | * The returned facet is guaranteed to be valid. */ |
4619 | static struct facet * | |
2b459b83 BP |
4620 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
4621 | uint32_t hash) | |
abe529af | 4622 | { |
c57b2226 | 4623 | struct facet *facet; |
abe529af | 4624 | |
c57b2226 | 4625 | facet = facet_find(ofproto, flow, hash); |
abe529af | 4626 | if (facet |
2cc3c58e EJ |
4627 | && (ofproto->backer->need_revalidate |
4628 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
4629 | facet->tags))) { | |
c57b2226 | 4630 | facet_revalidate(facet); |
f231418e EJ |
4631 | |
4632 | /* facet_revalidate() may have destroyed 'facet'. */ | |
4633 | facet = facet_find(ofproto, flow, hash); | |
abe529af BP |
4634 | } |
4635 | ||
4636 | return facet; | |
4637 | } | |
4638 | ||
1a1e3a0a JP |
4639 | /* Return a subfacet from 'facet'. A facet consists of one or more |
4640 | * subfacets, and this function returns one of them. */ | |
4641 | static struct subfacet *facet_get_subfacet(struct facet *facet) | |
4642 | { | |
4643 | return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet, | |
4644 | list_node); | |
4645 | } | |
4646 | ||
6a7e895f BP |
4647 | static const char * |
4648 | subfacet_path_to_string(enum subfacet_path path) | |
4649 | { | |
4650 | switch (path) { | |
4651 | case SF_NOT_INSTALLED: | |
4652 | return "not installed"; | |
4653 | case SF_FAST_PATH: | |
4654 | return "in fast path"; | |
4655 | case SF_SLOW_PATH: | |
4656 | return "in slow path"; | |
4657 | default: | |
4658 | return "<error>"; | |
4659 | } | |
4660 | } | |
4661 | ||
4662 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
4663 | * member has the specified value. */ | |
4664 | static enum subfacet_path | |
4665 | subfacet_want_path(enum slow_path_reason slow) | |
4666 | { | |
4667 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
4668 | } | |
4669 | ||
4670 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
4671 | * supposing that its actions have been recalculated as 'want_actions' and that | |
4672 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
4673 | static bool | |
4674 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
4675 | const struct ofpbuf *want_actions) | |
4676 | { | |
4677 | enum subfacet_path want_path = subfacet_want_path(slow); | |
4678 | return (want_path != subfacet->path | |
4679 | || (want_path == SF_FAST_PATH | |
4680 | && (subfacet->actions_len != want_actions->size | |
4681 | || memcmp(subfacet->actions, want_actions->data, | |
4682 | subfacet->actions_len)))); | |
4683 | } | |
4684 | ||
6814e51f BP |
4685 | static bool |
4686 | facet_check_consistency(struct facet *facet) | |
4687 | { | |
4688 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4689 | ||
4690 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4691 | ||
050ac423 BP |
4692 | uint64_t odp_actions_stub[1024 / 8]; |
4693 | struct ofpbuf odp_actions; | |
4694 | ||
6814e51f BP |
4695 | struct rule_dpif *rule; |
4696 | struct subfacet *subfacet; | |
c53e1132 | 4697 | bool may_log = false; |
6814e51f BP |
4698 | bool ok; |
4699 | ||
4700 | /* Check the rule for consistency. */ | |
c57b2226 BP |
4701 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
4702 | ok = rule == facet->rule; | |
4703 | if (!ok) { | |
c53e1132 | 4704 | may_log = !VLOG_DROP_WARN(&rl); |
c53e1132 BP |
4705 | if (may_log) { |
4706 | struct ds s; | |
6814e51f | 4707 | |
c53e1132 BP |
4708 | ds_init(&s); |
4709 | flow_format(&s, &facet->flow); | |
4710 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4711 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4712 | cls_rule_format(&facet->rule->up.cr, &s); | |
4713 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4714 | rule->up.table_id); | |
4715 | cls_rule_format(&rule->up.cr, &s); | |
4716 | ds_put_char(&s, ')'); | |
6814e51f | 4717 | |
c53e1132 BP |
4718 | VLOG_WARN("%s", ds_cstr(&s)); |
4719 | ds_destroy(&s); | |
4720 | } | |
6814e51f BP |
4721 | } |
4722 | ||
4723 | /* Check the datapath actions for consistency. */ | |
050ac423 | 4724 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f | 4725 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4726 | enum subfacet_path want_path; |
6814e51f | 4727 | struct action_xlate_ctx ctx; |
9616614b | 4728 | struct ds s; |
6814e51f BP |
4729 | |
4730 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
14f94f9a | 4731 | &subfacet->initial_vals, rule, 0, NULL); |
f25d0cf3 | 4732 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 4733 | &odp_actions); |
6814e51f | 4734 | |
6a7e895f BP |
4735 | if (subfacet->path == SF_NOT_INSTALLED) { |
4736 | /* This only happens if the datapath reported an error when we | |
4737 | * tried to install the flow. Don't flag another error here. */ | |
4738 | continue; | |
4739 | } | |
4740 | ||
4741 | want_path = subfacet_want_path(subfacet->slow); | |
4742 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
4743 | /* The actions for slow-path flows may legitimately vary from one | |
4744 | * packet to the next. We're done. */ | |
050ac423 | 4745 | continue; |
6814e51f BP |
4746 | } |
4747 | ||
6a7e895f | 4748 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { |
9616614b BP |
4749 | continue; |
4750 | } | |
c53e1132 | 4751 | |
9616614b BP |
4752 | /* Inconsistency! */ |
4753 | if (ok) { | |
4754 | may_log = !VLOG_DROP_WARN(&rl); | |
4755 | ok = false; | |
4756 | } | |
4757 | if (!may_log) { | |
4758 | /* Rate-limited, skip reporting. */ | |
4759 | continue; | |
4760 | } | |
c53e1132 | 4761 | |
9616614b | 4762 | ds_init(&s); |
9566abf9 | 4763 | odp_flow_key_format(subfacet->key, subfacet->key_len, &s); |
9616614b BP |
4764 | |
4765 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
6a7e895f | 4766 | if (want_path != subfacet->path) { |
9616614b BP |
4767 | enum odp_key_fitness fitness = subfacet->key_fitness; |
4768 | ||
6a7e895f BP |
4769 | ds_put_format(&s, " (%s, fitness=%s)", |
4770 | subfacet_path_to_string(subfacet->path), | |
9616614b | 4771 | odp_key_fitness_to_string(fitness)); |
6a7e895f BP |
4772 | ds_put_format(&s, " (should have been %s)", |
4773 | subfacet_path_to_string(want_path)); | |
4774 | } else if (want_path == SF_FAST_PATH) { | |
9616614b BP |
4775 | ds_put_cstr(&s, " (actions were: "); |
4776 | format_odp_actions(&s, subfacet->actions, | |
4777 | subfacet->actions_len); | |
4778 | ds_put_cstr(&s, ") (correct actions: "); | |
4779 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
4780 | ds_put_char(&s, ')'); | |
4781 | } else { | |
4782 | ds_put_cstr(&s, " (actions: "); | |
4783 | format_odp_actions(&s, subfacet->actions, | |
4784 | subfacet->actions_len); | |
4785 | ds_put_char(&s, ')'); | |
6814e51f | 4786 | } |
9616614b BP |
4787 | VLOG_WARN("%s", ds_cstr(&s)); |
4788 | ds_destroy(&s); | |
6814e51f | 4789 | } |
050ac423 | 4790 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
4791 | |
4792 | return ok; | |
4793 | } | |
4794 | ||
15baa734 | 4795 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4796 | * |
4797 | * - If the rule found is different from 'facet''s current rule, moves | |
4798 | * 'facet' to the new rule and recompiles its actions. | |
4799 | * | |
4800 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
f231418e EJ |
4801 | * where it is and recompiles its actions anyway. |
4802 | * | |
4803 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
4804 | * ofproto_receive(), 'facet' is removed. */ | |
c57b2226 | 4805 | static void |
15baa734 | 4806 | facet_revalidate(struct facet *facet) |
abe529af | 4807 | { |
15baa734 | 4808 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
4809 | struct actions { |
4810 | struct nlattr *odp_actions; | |
4811 | size_t actions_len; | |
4812 | }; | |
4813 | struct actions *new_actions; | |
4814 | ||
abe529af | 4815 | struct action_xlate_ctx ctx; |
050ac423 BP |
4816 | uint64_t odp_actions_stub[1024 / 8]; |
4817 | struct ofpbuf odp_actions; | |
4818 | ||
abe529af | 4819 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4820 | struct subfacet *subfacet; |
b95fc6ba | 4821 | int i; |
abe529af BP |
4822 | |
4823 | COVERAGE_INC(facet_revalidate); | |
4824 | ||
f231418e EJ |
4825 | /* Check that child subfacets still correspond to this facet. Tunnel |
4826 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
4827 | * change. */ | |
4828 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4829 | struct ofproto_dpif *recv_ofproto; | |
4830 | struct flow recv_flow; | |
4831 | int error; | |
4832 | ||
4833 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
4834 | subfacet->key_len, &recv_flow, NULL, | |
4835 | &recv_ofproto, NULL, NULL); | |
4836 | if (error | |
4837 | || recv_ofproto != ofproto | |
4838 | || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { | |
4839 | facet_remove(facet); | |
4840 | return; | |
4841 | } | |
4842 | } | |
4843 | ||
c57b2226 | 4844 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 4845 | |
df2c07f4 | 4846 | /* Calculate new datapath actions. |
abe529af BP |
4847 | * |
4848 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4849 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4850 | * around to properly compose it. */ | |
abe529af | 4851 | |
df2c07f4 JP |
4852 | /* If the datapath actions changed or the installability changed, |
4853 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
4854 | i = 0; |
4855 | new_actions = NULL; | |
4856 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 4857 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 4858 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4859 | enum slow_path_reason slow; |
b95fc6ba | 4860 | |
e84173dc | 4861 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
14f94f9a | 4862 | &subfacet->initial_vals, new_rule, 0, NULL); |
f25d0cf3 | 4863 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, |
050ac423 | 4864 | &odp_actions); |
b0f7b9b5 | 4865 | |
6a7e895f BP |
4866 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
4867 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
4868 | struct dpif_flow_stats stats; | |
4869 | ||
4870 | subfacet_install(subfacet, | |
4871 | odp_actions.data, odp_actions.size, &stats, slow); | |
4872 | subfacet_update_stats(subfacet, &stats); | |
b95fc6ba BP |
4873 | |
4874 | if (!new_actions) { | |
4875 | new_actions = xcalloc(list_size(&facet->subfacets), | |
4876 | sizeof *new_actions); | |
4877 | } | |
050ac423 BP |
4878 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
4879 | odp_actions.size); | |
4880 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 4881 | } |
b95fc6ba | 4882 | |
b95fc6ba | 4883 | i++; |
b0f7b9b5 | 4884 | } |
050ac423 BP |
4885 | ofpbuf_uninit(&odp_actions); |
4886 | ||
b95fc6ba | 4887 | if (new_actions) { |
15baa734 | 4888 | facet_flush_stats(facet); |
abe529af BP |
4889 | } |
4890 | ||
4891 | /* Update 'facet' now that we've taken care of all the old state. */ | |
4892 | facet->tags = ctx.tags; | |
4893 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
75a75043 BP |
4894 | facet->has_learn = ctx.has_learn; |
4895 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4896 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 4897 | facet->mirrors = ctx.mirrors; |
6a7e895f BP |
4898 | |
4899 | i = 0; | |
4900 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4901 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4902 | ||
4903 | if (new_actions && new_actions[i].odp_actions) { | |
4904 | free(subfacet->actions); | |
4905 | subfacet->actions = new_actions[i].odp_actions; | |
4906 | subfacet->actions_len = new_actions[i].actions_len; | |
b95fc6ba | 4907 | } |
6a7e895f | 4908 | i++; |
abe529af | 4909 | } |
6a7e895f BP |
4910 | free(new_actions); |
4911 | ||
abe529af BP |
4912 | if (facet->rule != new_rule) { |
4913 | COVERAGE_INC(facet_changed_rule); | |
4914 | list_remove(&facet->list_node); | |
4915 | list_push_back(&new_rule->facets, &facet->list_node); | |
4916 | facet->rule = new_rule; | |
4917 | facet->used = new_rule->up.created; | |
9d24de3b | 4918 | facet->prev_used = facet->used; |
abe529af | 4919 | } |
abe529af BP |
4920 | } |
4921 | ||
4922 | /* Updates 'facet''s used time. Caller is responsible for calling | |
4923 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
4924 | static void | |
15baa734 | 4925 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 4926 | { |
15baa734 | 4927 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
4928 | if (used > facet->used) { |
4929 | facet->used = used; | |
1745cd08 | 4930 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
4931 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
4932 | } | |
4933 | } | |
4934 | ||
bbb5d219 EJ |
4935 | static void |
4936 | facet_reset_counters(struct facet *facet) | |
4937 | { | |
4938 | facet->packet_count = 0; | |
4939 | facet->byte_count = 0; | |
9d24de3b JP |
4940 | facet->prev_packet_count = 0; |
4941 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4942 | facet->accounted_bytes = 0; |
4943 | } | |
4944 | ||
abe529af BP |
4945 | static void |
4946 | facet_push_stats(struct facet *facet) | |
4947 | { | |
112bc5f4 | 4948 | struct dpif_flow_stats stats; |
abe529af | 4949 | |
cb22974d BP |
4950 | ovs_assert(facet->packet_count >= facet->prev_packet_count); |
4951 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
4952 | ovs_assert(facet->used >= facet->prev_used); | |
abe529af | 4953 | |
112bc5f4 BP |
4954 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4955 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4956 | stats.used = facet->used; | |
4957 | stats.tcp_flags = 0; | |
abe529af | 4958 | |
112bc5f4 | 4959 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
4960 | facet->prev_packet_count = facet->packet_count; |
4961 | facet->prev_byte_count = facet->byte_count; | |
4962 | facet->prev_used = facet->used; | |
abe529af | 4963 | |
ac35f9c8 | 4964 | flow_push_stats(facet, &stats); |
9d24de3b JP |
4965 | |
4966 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 4967 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
4968 | } |
4969 | } | |
4970 | ||
abe529af | 4971 | static void |
112bc5f4 | 4972 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4973 | { |
112bc5f4 BP |
4974 | rule->packet_count += stats->n_packets; |
4975 | rule->byte_count += stats->n_bytes; | |
4976 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
4977 | } |
4978 | ||
ac35f9c8 JP |
4979 | /* Pushes flow statistics to the rules which 'facet->flow' resubmits |
4980 | * into given 'facet->rule''s actions and mirrors. */ | |
abe529af | 4981 | static void |
ac35f9c8 | 4982 | flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats) |
abe529af | 4983 | { |
ac35f9c8 | 4984 | struct rule_dpif *rule = facet->rule; |
abe529af | 4985 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
1a1e3a0a | 4986 | struct subfacet *subfacet = facet_get_subfacet(facet); |
112bc5f4 | 4987 | struct action_xlate_ctx ctx; |
abe529af | 4988 | |
112bc5f4 | 4989 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 4990 | |
14f94f9a JP |
4991 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
4992 | &subfacet->initial_vals, rule, 0, NULL); | |
112bc5f4 | 4993 | ctx.resubmit_stats = stats; |
f25d0cf3 BP |
4994 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, |
4995 | rule->up.ofpacts_len); | |
abe529af BP |
4996 | } |
4997 | \f | |
b0f7b9b5 BP |
4998 | /* Subfacets. */ |
4999 | ||
5000 | static struct subfacet * | |
acf60855 | 5001 | subfacet_find(struct ofproto_dpif *ofproto, |
9566abf9 | 5002 | const struct nlattr *key, size_t key_len, uint32_t key_hash) |
b0f7b9b5 BP |
5003 | { |
5004 | struct subfacet *subfacet; | |
5005 | ||
5006 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
5007 | &ofproto->subfacets) { | |
9566abf9 EJ |
5008 | if (subfacet->key_len == key_len |
5009 | && !memcmp(key, subfacet->key, key_len)) { | |
b0f7b9b5 BP |
5010 | return subfacet; |
5011 | } | |
5012 | } | |
5013 | ||
5014 | return NULL; | |
5015 | } | |
5016 | ||
5017 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
a088a1ff JP |
5018 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the |
5019 | * existing subfacet if there is one, otherwise creates and returns a | |
5020 | * new subfacet. | |
b95fc6ba BP |
5021 | * |
5022 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
5023 | * which case the caller must populate the actions with | |
5024 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 5025 | static struct subfacet * |
a088a1ff JP |
5026 | subfacet_create(struct facet *facet, struct flow_miss *miss, |
5027 | long long int now) | |
b0f7b9b5 | 5028 | { |
15baa734 | 5029 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
a088a1ff JP |
5030 | enum odp_key_fitness key_fitness = miss->key_fitness; |
5031 | const struct nlattr *key = miss->key; | |
5032 | size_t key_len = miss->key_len; | |
5033 | uint32_t key_hash; | |
b0f7b9b5 BP |
5034 | struct subfacet *subfacet; |
5035 | ||
a088a1ff JP |
5036 | key_hash = odp_flow_key_hash(key, key_len); |
5037 | ||
3b145dd7 BP |
5038 | if (list_is_empty(&facet->subfacets)) { |
5039 | subfacet = &facet->one_subfacet; | |
5040 | } else { | |
9566abf9 | 5041 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); |
3b145dd7 BP |
5042 | if (subfacet) { |
5043 | if (subfacet->facet == facet) { | |
5044 | return subfacet; | |
5045 | } | |
5046 | ||
5047 | /* This shouldn't happen. */ | |
5048 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5049 | subfacet_destroy(subfacet); | |
b0f7b9b5 BP |
5050 | } |
5051 | ||
3b145dd7 | 5052 | subfacet = xmalloc(sizeof *subfacet); |
b0f7b9b5 BP |
5053 | } |
5054 | ||
b0f7b9b5 BP |
5055 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
5056 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
5057 | subfacet->facet = facet; | |
b0f7b9b5 | 5058 | subfacet->key_fitness = key_fitness; |
9566abf9 EJ |
5059 | subfacet->key = xmemdup(key, key_len); |
5060 | subfacet->key_len = key_len; | |
459b16a1 | 5061 | subfacet->used = now; |
26cd7e34 BP |
5062 | subfacet->dp_packet_count = 0; |
5063 | subfacet->dp_byte_count = 0; | |
5064 | subfacet->actions_len = 0; | |
5065 | subfacet->actions = NULL; | |
6a7e895f BP |
5066 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE |
5067 | ? SLOW_MATCH | |
5068 | : 0); | |
5069 | subfacet->path = SF_NOT_INSTALLED; | |
14f94f9a | 5070 | subfacet->initial_vals = miss->initial_vals; |
a088a1ff | 5071 | subfacet->odp_in_port = miss->odp_in_port; |
b0f7b9b5 BP |
5072 | |
5073 | return subfacet; | |
5074 | } | |
5075 | ||
b0f7b9b5 BP |
5076 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from |
5077 | * its facet within 'ofproto', and frees it. */ | |
5078 | static void | |
15baa734 | 5079 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 5080 | { |
15baa734 BP |
5081 | struct facet *facet = subfacet->facet; |
5082 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5083 | ||
5084 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
5085 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
5086 | list_remove(&subfacet->list_node); | |
5087 | free(subfacet->key); | |
b95fc6ba | 5088 | free(subfacet->actions); |
26cd7e34 BP |
5089 | if (subfacet != &facet->one_subfacet) { |
5090 | free(subfacet); | |
5091 | } | |
b0f7b9b5 BP |
5092 | } |
5093 | ||
5094 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5095 | * last remaining subfacet in its facet destroys the facet too. */ | |
5096 | static void | |
15baa734 | 5097 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
5098 | { |
5099 | struct facet *facet = subfacet->facet; | |
5100 | ||
551a2f6c BP |
5101 | if (list_is_singleton(&facet->subfacets)) { |
5102 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 5103 | facet_remove(facet); |
551a2f6c | 5104 | } else { |
15baa734 | 5105 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
5106 | } |
5107 | } | |
5108 | ||
1d85f9e5 JP |
5109 | static void |
5110 | subfacet_destroy_batch(struct ofproto_dpif *ofproto, | |
5111 | struct subfacet **subfacets, int n) | |
5112 | { | |
1d85f9e5 JP |
5113 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; |
5114 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
1d85f9e5 JP |
5115 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; |
5116 | int i; | |
5117 | ||
5118 | for (i = 0; i < n; i++) { | |
5119 | ops[i].type = DPIF_OP_FLOW_DEL; | |
9566abf9 EJ |
5120 | ops[i].u.flow_del.key = subfacets[i]->key; |
5121 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
1d85f9e5 JP |
5122 | ops[i].u.flow_del.stats = &stats[i]; |
5123 | opsp[i] = &ops[i]; | |
5124 | } | |
5125 | ||
acf60855 | 5126 | dpif_operate(ofproto->backer->dpif, opsp, n); |
1d85f9e5 JP |
5127 | for (i = 0; i < n; i++) { |
5128 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5129 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5130 | subfacet_destroy(subfacets[i]); | |
5131 | } | |
5132 | } | |
5133 | ||
5fe20d5d BP |
5134 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
5135 | * Translates the actions into 'odp_actions', which the caller must have | |
5136 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 5137 | static void |
5fe20d5d BP |
5138 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
5139 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
5140 | { |
5141 | struct facet *facet = subfacet->facet; | |
18b2a258 | 5142 | struct rule_dpif *rule = facet->rule; |
15baa734 | 5143 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 5144 | |
b95fc6ba BP |
5145 | struct action_xlate_ctx ctx; |
5146 | ||
14f94f9a JP |
5147 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
5148 | &subfacet->initial_vals, rule, 0, packet); | |
f25d0cf3 | 5149 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); |
b95fc6ba | 5150 | facet->tags = ctx.tags; |
b95fc6ba BP |
5151 | facet->has_learn = ctx.has_learn; |
5152 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 5153 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 5154 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 5155 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 5156 | |
6a7e895f | 5157 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5fe20d5d BP |
5158 | if (subfacet->actions_len != odp_actions->size |
5159 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 5160 | free(subfacet->actions); |
5fe20d5d BP |
5161 | subfacet->actions_len = odp_actions->size; |
5162 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 5163 | } |
b95fc6ba BP |
5164 | } |
5165 | ||
b0f7b9b5 BP |
5166 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
5167 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5168 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5169 | * since 'subfacet' was last updated. | |
5170 | * | |
5171 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5172 | static int | |
15baa734 | 5173 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 | 5174 | const struct nlattr *actions, size_t actions_len, |
6a7e895f BP |
5175 | struct dpif_flow_stats *stats, |
5176 | enum slow_path_reason slow) | |
b0f7b9b5 | 5177 | { |
15baa734 BP |
5178 | struct facet *facet = subfacet->facet; |
5179 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
6a7e895f BP |
5180 | enum subfacet_path path = subfacet_want_path(slow); |
5181 | uint64_t slow_path_stub[128 / 8]; | |
b0f7b9b5 | 5182 | enum dpif_flow_put_flags flags; |
b0f7b9b5 BP |
5183 | int ret; |
5184 | ||
5185 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5186 | if (stats) { | |
5187 | flags |= DPIF_FP_ZERO_STATS; | |
5188 | } | |
5189 | ||
6a7e895f BP |
5190 | if (path == SF_SLOW_PATH) { |
5191 | compose_slow_path(ofproto, &facet->flow, slow, | |
5192 | slow_path_stub, sizeof slow_path_stub, | |
5193 | &actions, &actions_len); | |
5194 | } | |
5195 | ||
9566abf9 EJ |
5196 | ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key, |
5197 | subfacet->key_len, actions, actions_len, stats); | |
b0f7b9b5 BP |
5198 | |
5199 | if (stats) { | |
5200 | subfacet_reset_dp_stats(subfacet, stats); | |
5201 | } | |
5202 | ||
6a7e895f BP |
5203 | if (!ret) { |
5204 | subfacet->path = path; | |
5205 | } | |
b0f7b9b5 BP |
5206 | return ret; |
5207 | } | |
5208 | ||
6a7e895f BP |
5209 | static int |
5210 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
5211 | { | |
5212 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
5213 | stats, subfacet->slow); | |
5214 | } | |
5215 | ||
b0f7b9b5 BP |
5216 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ |
5217 | static void | |
15baa734 | 5218 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 5219 | { |
6a7e895f | 5220 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
5221 | struct rule_dpif *rule = subfacet->facet->rule; |
5222 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 | 5223 | struct dpif_flow_stats stats; |
b0f7b9b5 BP |
5224 | int error; |
5225 | ||
9566abf9 EJ |
5226 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, |
5227 | subfacet->key_len, &stats); | |
b0f7b9b5 BP |
5228 | subfacet_reset_dp_stats(subfacet, &stats); |
5229 | if (!error) { | |
15baa734 | 5230 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5231 | } |
6a7e895f | 5232 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 | 5233 | } else { |
cb22974d BP |
5234 | ovs_assert(subfacet->dp_packet_count == 0); |
5235 | ovs_assert(subfacet->dp_byte_count == 0); | |
b0f7b9b5 BP |
5236 | } |
5237 | } | |
5238 | ||
5239 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5240 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5241 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5242 | * was reset in the datapath. 'stats' will be modified to include only | |
5243 | * statistics new since 'subfacet' was last updated. */ | |
5244 | static void | |
5245 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5246 | struct dpif_flow_stats *stats) | |
5247 | { | |
5248 | if (stats | |
5249 | && subfacet->dp_packet_count <= stats->n_packets | |
5250 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5251 | stats->n_packets -= subfacet->dp_packet_count; | |
5252 | stats->n_bytes -= subfacet->dp_byte_count; | |
5253 | } | |
5254 | ||
5255 | subfacet->dp_packet_count = 0; | |
5256 | subfacet->dp_byte_count = 0; | |
5257 | } | |
5258 | ||
5259 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
5260 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
5261 | static void | |
15baa734 | 5262 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
5263 | { |
5264 | if (used > subfacet->used) { | |
5265 | subfacet->used = used; | |
15baa734 | 5266 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
5267 | } |
5268 | } | |
5269 | ||
5270 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
5271 | * | |
5272 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5273 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5274 | * represents a packet that was sent by hand or if it represents statistics | |
5275 | * that have been cleared out of the datapath. */ | |
5276 | static void | |
15baa734 | 5277 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
5278 | const struct dpif_flow_stats *stats) |
5279 | { | |
5280 | if (stats->n_packets || stats->used > subfacet->used) { | |
5281 | struct facet *facet = subfacet->facet; | |
5282 | ||
15baa734 | 5283 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
5284 | facet->packet_count += stats->n_packets; |
5285 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 5286 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
5287 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); |
5288 | } | |
5289 | } | |
5290 | \f | |
abe529af BP |
5291 | /* Rules. */ |
5292 | ||
5293 | static struct rule_dpif * | |
c57b2226 BP |
5294 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
5295 | { | |
c57b2226 BP |
5296 | struct rule_dpif *rule; |
5297 | ||
5298 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5299 | if (rule) { | |
5300 | return rule; | |
5301 | } | |
5302 | ||
c376f9a3 | 5303 | return rule_dpif_miss_rule(ofproto, flow); |
c57b2226 BP |
5304 | } |
5305 | ||
5306 | static struct rule_dpif * | |
5307 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5308 | uint8_t table_id) | |
abe529af | 5309 | { |
7257b535 BP |
5310 | struct cls_rule *cls_rule; |
5311 | struct classifier *cls; | |
5312 | ||
9cdaaebe BP |
5313 | if (table_id >= N_TABLES) { |
5314 | return NULL; | |
5315 | } | |
5316 | ||
d0918789 | 5317 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 5318 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
5319 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
5320 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
5321 | * are unavailable. */ | |
5322 | struct flow ofpc_normal_flow = *flow; | |
5323 | ofpc_normal_flow.tp_src = htons(0); | |
5324 | ofpc_normal_flow.tp_dst = htons(0); | |
5325 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
5326 | } else { | |
5327 | cls_rule = classifier_lookup(cls, flow); | |
5328 | } | |
5329 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
5330 | } |
5331 | ||
c376f9a3 IY |
5332 | static struct rule_dpif * |
5333 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5334 | { | |
5335 | struct ofport_dpif *port; | |
5336 | ||
5337 | port = get_ofp_port(ofproto, flow->in_port); | |
5338 | if (!port) { | |
5339 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5340 | return ofproto->miss_rule; | |
5341 | } | |
5342 | ||
5343 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5344 | return ofproto->no_packet_in_rule; | |
5345 | } | |
5346 | return ofproto->miss_rule; | |
5347 | } | |
5348 | ||
7ee20df1 BP |
5349 | static void |
5350 | complete_operation(struct rule_dpif *rule) | |
5351 | { | |
5352 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5353 | ||
54a9cbc9 | 5354 | rule_invalidate(rule); |
7ee20df1 BP |
5355 | if (clogged) { |
5356 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5357 | c->op = rule->up.pending; | |
5358 | list_push_back(&ofproto->completions, &c->list_node); | |
5359 | } else { | |
5360 | ofoperation_complete(rule->up.pending, 0); | |
5361 | } | |
5362 | } | |
5363 | ||
abe529af BP |
5364 | static struct rule * |
5365 | rule_alloc(void) | |
5366 | { | |
5367 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5368 | return &rule->up; | |
5369 | } | |
5370 | ||
5371 | static void | |
5372 | rule_dealloc(struct rule *rule_) | |
5373 | { | |
5374 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5375 | free(rule); | |
5376 | } | |
5377 | ||
90bf1e07 | 5378 | static enum ofperr |
abe529af BP |
5379 | rule_construct(struct rule *rule_) |
5380 | { | |
5381 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5382 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 5383 | struct rule_dpif *victim; |
54a9cbc9 | 5384 | uint8_t table_id; |
abe529af | 5385 | |
abe529af BP |
5386 | rule->packet_count = 0; |
5387 | rule->byte_count = 0; | |
abe529af | 5388 | |
7ee20df1 BP |
5389 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
5390 | if (victim && !list_is_empty(&victim->facets)) { | |
5391 | struct facet *facet; | |
5392 | ||
5393 | rule->facets = victim->facets; | |
5394 | list_moved(&rule->facets); | |
5395 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
5396 | /* XXX: We're only clearing our local counters here. It's possible |
5397 | * that quite a few packets are unaccounted for in the datapath | |
5398 | * statistics. These will be accounted to the new rule instead of | |
5399 | * cleared as required. This could be fixed by clearing out the | |
5400 | * datapath statistics for this facet, but currently it doesn't | |
5401 | * seem worth it. */ | |
5402 | facet_reset_counters(facet); | |
7ee20df1 BP |
5403 | facet->rule = rule; |
5404 | } | |
5405 | } else { | |
5406 | /* Must avoid list_moved() in this case. */ | |
5407 | list_init(&rule->facets); | |
5408 | } | |
abe529af | 5409 | |
54a9cbc9 | 5410 | table_id = rule->up.table_id; |
5cb7a798 BP |
5411 | if (victim) { |
5412 | rule->tag = victim->tag; | |
5413 | } else if (table_id == 0) { | |
5414 | rule->tag = 0; | |
5415 | } else { | |
5416 | struct flow flow; | |
5417 | ||
5418 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5419 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5420 | ofproto->tables[table_id].basis); | |
5421 | } | |
54a9cbc9 | 5422 | |
7ee20df1 | 5423 | complete_operation(rule); |
abe529af BP |
5424 | return 0; |
5425 | } | |
5426 | ||
5427 | static void | |
5428 | rule_destruct(struct rule *rule_) | |
5429 | { | |
5430 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
5431 | struct facet *facet, *next_facet; |
5432 | ||
abe529af | 5433 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 5434 | facet_revalidate(facet); |
abe529af | 5435 | } |
7ee20df1 BP |
5436 | |
5437 | complete_operation(rule); | |
abe529af BP |
5438 | } |
5439 | ||
5440 | static void | |
5441 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5442 | { | |
bf1e8ff9 | 5443 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule_->ofproto); |
abe529af BP |
5444 | struct rule_dpif *rule = rule_dpif_cast(rule_); |
5445 | struct facet *facet; | |
5446 | ||
bf1e8ff9 EJ |
5447 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
5448 | facet_push_stats(facet); | |
5449 | } | |
5450 | ||
abe529af BP |
5451 | /* Start from historical data for 'rule' itself that are no longer tracked |
5452 | * in facets. This counts, for example, facets that have expired. */ | |
5453 | *packets = rule->packet_count; | |
5454 | *bytes = rule->byte_count; | |
5455 | ||
5456 | /* Add any statistics that are tracked by facets. This includes | |
5457 | * statistical data recently updated by ofproto_update_stats() as well as | |
5458 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
5459 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
5460 | *packets += facet->packet_count; | |
5461 | *bytes += facet->byte_count; | |
5462 | } | |
5463 | } | |
5464 | ||
0a740f48 EJ |
5465 | static void |
5466 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5467 | struct ofpbuf *packet) | |
abe529af | 5468 | { |
abe529af | 5469 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
14f94f9a | 5470 | struct initial_vals initial_vals; |
112bc5f4 | 5471 | struct dpif_flow_stats stats; |
abe529af | 5472 | struct action_xlate_ctx ctx; |
050ac423 BP |
5473 | uint64_t odp_actions_stub[1024 / 8]; |
5474 | struct ofpbuf odp_actions; | |
abe529af | 5475 | |
a7752d4a | 5476 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
112bc5f4 BP |
5477 | rule_credit_stats(rule, &stats); |
5478 | ||
14f94f9a | 5479 | initial_vals.vlan_tci = flow->vlan_tci; |
c3f6c502 | 5480 | initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos; |
050ac423 | 5481 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
14f94f9a | 5482 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, |
112bc5f4 BP |
5483 | rule, stats.tcp_flags, packet); |
5484 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 5485 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); |
112bc5f4 BP |
5486 | |
5487 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
5488 | odp_actions.size, packet); | |
5489 | ||
050ac423 | 5490 | ofpbuf_uninit(&odp_actions); |
0a740f48 | 5491 | } |
5bf0e941 | 5492 | |
0a740f48 EJ |
5493 | static enum ofperr |
5494 | rule_execute(struct rule *rule, const struct flow *flow, | |
5495 | struct ofpbuf *packet) | |
5496 | { | |
5497 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5498 | ofpbuf_delete(packet); | |
5bf0e941 | 5499 | return 0; |
abe529af BP |
5500 | } |
5501 | ||
7ee20df1 BP |
5502 | static void |
5503 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
5504 | { |
5505 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
7ee20df1 BP |
5506 | |
5507 | complete_operation(rule); | |
abe529af BP |
5508 | } |
5509 | \f | |
97d6520b | 5510 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 5511 | * May modify 'packet'. |
abe529af BP |
5512 | * Returns 0 if successful, otherwise a positive errno value. */ |
5513 | static int | |
52a90c29 | 5514 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 5515 | { |
97d6520b | 5516 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
b9ad7294 | 5517 | uint64_t odp_actions_stub[1024 / 8]; |
80e5eed9 BP |
5518 | struct ofpbuf key, odp_actions; |
5519 | struct odputil_keybuf keybuf; | |
9b56fe13 | 5520 | uint32_t odp_port; |
80e5eed9 | 5521 | struct flow flow; |
abe529af BP |
5522 | int error; |
5523 | ||
72e8bf28 | 5524 | flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow); |
0a740f48 EJ |
5525 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5526 | struct ofproto_dpif *peer_ofproto; | |
5527 | struct dpif_flow_stats stats; | |
5528 | struct ofport_dpif *peer; | |
5529 | struct rule_dpif *rule; | |
5530 | ||
5531 | peer = ofport_get_peer(ofport); | |
5532 | if (!peer) { | |
5533 | return ENODEV; | |
5534 | } | |
5535 | ||
5536 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
b9ad7294 EJ |
5537 | netdev_vport_inc_tx(ofport->up.netdev, &stats); |
5538 | netdev_vport_inc_rx(peer->up.netdev, &stats); | |
0a740f48 EJ |
5539 | |
5540 | flow.in_port = peer->up.ofp_port; | |
5541 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5542 | rule = rule_dpif_lookup(peer_ofproto, &flow); | |
5543 | rule_dpif_execute(rule, &flow, packet); | |
5544 | ||
5545 | return 0; | |
5546 | } | |
5547 | ||
b9ad7294 EJ |
5548 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
5549 | ||
5550 | if (ofport->tnl_port) { | |
5551 | struct dpif_flow_stats stats; | |
5552 | ||
5553 | odp_port = tnl_port_send(ofport->tnl_port, &flow); | |
5554 | if (odp_port == OVSP_NONE) { | |
5555 | return ENODEV; | |
5556 | } | |
5557 | ||
5558 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
5559 | netdev_vport_inc_tx(ofport->up.netdev, &stats); | |
5560 | odp_put_tunnel_action(&flow.tunnel, &odp_actions); | |
09a0d2e7 | 5561 | odp_put_skb_mark_action(flow.skb_mark, &odp_actions); |
b9ad7294 EJ |
5562 | } else { |
5563 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, | |
5564 | flow.vlan_tci); | |
5565 | if (odp_port != ofport->odp_port) { | |
5566 | eth_pop_vlan(packet); | |
5567 | flow.vlan_tci = htons(0); | |
5568 | } | |
52a90c29 BP |
5569 | } |
5570 | ||
80e5eed9 | 5571 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
5572 | odp_flow_key_from_flow(&key, &flow, |
5573 | ofp_port_to_odp_port(ofproto, flow.in_port)); | |
80e5eed9 | 5574 | |
6ff686f2 PS |
5575 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
5576 | ||
df2c07f4 | 5577 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
acf60855 | 5578 | error = dpif_execute(ofproto->backer->dpif, |
80e5eed9 BP |
5579 | key.data, key.size, |
5580 | odp_actions.data, odp_actions.size, | |
abe529af BP |
5581 | packet); |
5582 | ofpbuf_uninit(&odp_actions); | |
5583 | ||
5584 | if (error) { | |
5585 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
5586 | ofproto->up.name, odp_port, strerror(error)); | |
5587 | } | |
6527c598 | 5588 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
5589 | return error; |
5590 | } | |
5591 | \f | |
df2c07f4 | 5592 | /* OpenFlow to datapath action translation. */ |
abe529af | 5593 | |
ffaef958 | 5594 | static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *); |
f25d0cf3 BP |
5595 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
5596 | struct action_xlate_ctx *); | |
4cd78906 | 5597 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 5598 | |
6a7e895f BP |
5599 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
5600 | * The action will state 'slow' as the reason that the action is in the slow | |
5601 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5602 | * dump-flows" output to see why a flow is in the slow path.) | |
5603 | * | |
5604 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5605 | * 'stub_size' must be large enough for the action. | |
5606 | * | |
5607 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5608 | * respectively. */ | |
5609 | static void | |
5610 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5611 | enum slow_path_reason slow, | |
5612 | uint64_t *stub, size_t stub_size, | |
5613 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5614 | { | |
5615 | union user_action_cookie cookie; | |
5616 | struct ofpbuf buf; | |
5617 | ||
5618 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5619 | cookie.slow_path.unused = 0; | |
5620 | cookie.slow_path.reason = slow; | |
5621 | ||
5622 | ofpbuf_use_stack(&buf, stub, stub_size); | |
625b0720 | 5623 | if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { |
9032f11e | 5624 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); |
e995e3df | 5625 | odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf); |
625b0720 BP |
5626 | } else { |
5627 | put_userspace_action(ofproto, &buf, flow, &cookie); | |
5628 | } | |
6a7e895f BP |
5629 | *actionsp = buf.data; |
5630 | *actions_lenp = buf.size; | |
5631 | } | |
5632 | ||
98403001 BP |
5633 | static size_t |
5634 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5635 | struct ofpbuf *odp_actions, | |
5636 | const struct flow *flow, | |
1673e0e4 | 5637 | const union user_action_cookie *cookie) |
98403001 | 5638 | { |
98403001 BP |
5639 | uint32_t pid; |
5640 | ||
acf60855 | 5641 | pid = dpif_port_get_pid(ofproto->backer->dpif, |
e1b1d06a | 5642 | ofp_port_to_odp_port(ofproto, flow->in_port)); |
98403001 | 5643 | |
e995e3df | 5644 | return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions); |
98403001 BP |
5645 | } |
5646 | ||
36fc5f18 BP |
5647 | static void |
5648 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5649 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 5650 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
5651 | { |
5652 | int ifindex; | |
5653 | ||
5654 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 5655 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
5656 | |
5657 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5658 | * port information") for the interpretation of cookie->output. */ | |
5659 | switch (n_outputs) { | |
5660 | case 0: | |
5661 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 5662 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
5663 | break; |
5664 | ||
5665 | case 1: | |
5666 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5667 | if (ifindex) { | |
1673e0e4 | 5668 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
5669 | break; |
5670 | } | |
5671 | /* Fall through. */ | |
5672 | default: | |
5673 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 5674 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
5675 | break; |
5676 | } | |
5677 | } | |
5678 | ||
6ff686f2 PS |
5679 | /* Compose SAMPLE action for sFlow. */ |
5680 | static size_t | |
5681 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5682 | struct ofpbuf *odp_actions, | |
5683 | const struct flow *flow, | |
5684 | uint32_t odp_port) | |
5685 | { | |
6ff686f2 | 5686 | uint32_t probability; |
1673e0e4 | 5687 | union user_action_cookie cookie; |
6ff686f2 | 5688 | size_t sample_offset, actions_offset; |
36fc5f18 | 5689 | int cookie_offset; |
6ff686f2 PS |
5690 | |
5691 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5692 | return 0; | |
5693 | } | |
5694 | ||
6ff686f2 PS |
5695 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); |
5696 | ||
5697 | /* Number of packets out of UINT_MAX to sample. */ | |
5698 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
5699 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5700 | ||
5701 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
36fc5f18 BP |
5702 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
5703 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
98403001 | 5704 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); |
6ff686f2 PS |
5705 | |
5706 | nl_msg_end_nested(odp_actions, actions_offset); | |
5707 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 5708 | return cookie_offset; |
6ff686f2 PS |
5709 | } |
5710 | ||
5711 | /* SAMPLE action must be first action in any given list of actions. | |
5712 | * At this point we do not have all information required to build it. So try to | |
5713 | * build sample action as complete as possible. */ | |
5714 | static void | |
5715 | add_sflow_action(struct action_xlate_ctx *ctx) | |
5716 | { | |
5717 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
5718 | ctx->odp_actions, | |
5719 | &ctx->flow, OVSP_NONE); | |
5720 | ctx->sflow_odp_port = 0; | |
5721 | ctx->sflow_n_outputs = 0; | |
5722 | } | |
5723 | ||
5724 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
5725 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
5726 | * USERSPACE action's user-cookie which is required for sflow. */ | |
5727 | static void | |
5728 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
5729 | { | |
5730 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 5731 | union user_action_cookie *cookie; |
6ff686f2 PS |
5732 | |
5733 | if (!ctx->user_cookie_offset) { | |
5734 | return; | |
5735 | } | |
5736 | ||
5737 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
36fc5f18 | 5738 | sizeof(*cookie)); |
cb22974d | 5739 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
6ff686f2 | 5740 | |
36fc5f18 BP |
5741 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
5742 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
5743 | } |
5744 | ||
6ff686f2 | 5745 | static void |
81b1afb1 EJ |
5746 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
5747 | bool check_stp) | |
6ff686f2 | 5748 | { |
d59906fb | 5749 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
52a90c29 | 5750 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
b9ad7294 | 5751 | ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id; |
8b36f51e | 5752 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
a4454ac6 | 5753 | struct priority_to_dscp *pdscp; |
0a740f48 EJ |
5754 | uint32_t out_port, odp_port; |
5755 | ||
5756 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
5757 | * before traversing a patch port. */ | |
cff78c88 | 5758 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); |
d59906fb | 5759 | |
a4454ac6 EJ |
5760 | if (!ofport) { |
5761 | xlate_report(ctx, "Nonexistent output port"); | |
5762 | return; | |
5763 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
5764 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
5765 | return; | |
5766 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
5767 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
5768 | return; | |
5769 | } | |
8b36f51e | 5770 | |
0a740f48 EJ |
5771 | if (netdev_vport_is_patch(ofport->up.netdev)) { |
5772 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
5773 | struct flow old_flow = ctx->flow; | |
5774 | const struct ofproto_dpif *peer_ofproto; | |
bb374ef6 | 5775 | enum slow_path_reason special; |
ffaef958 | 5776 | struct ofport_dpif *in_port; |
0a740f48 EJ |
5777 | |
5778 | if (!peer) { | |
5779 | xlate_report(ctx, "Nonexistent patch port peer"); | |
5780 | return; | |
5781 | } | |
5782 | ||
5783 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5784 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
5785 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
5786 | return; | |
5787 | } | |
5788 | ||
5789 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
5790 | ctx->flow.in_port = peer->up.ofp_port; | |
5791 | ctx->flow.metadata = htonll(0); | |
5792 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); | |
5793 | memset(ctx->flow.regs, 0, sizeof ctx->flow.regs); | |
ffaef958 BP |
5794 | |
5795 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
bb374ef6 EJ |
5796 | special = process_special(ctx->ofproto, &ctx->flow, in_port, |
5797 | ctx->packet); | |
5798 | if (special) { | |
5799 | ctx->slow |= special; | |
5800 | } else if (!in_port || may_receive(in_port, ctx)) { | |
ffaef958 BP |
5801 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { |
5802 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
5803 | } else { | |
5804 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
5805 | * learning action look at the packet, then drop it. */ | |
5806 | struct flow old_base_flow = ctx->base_flow; | |
5807 | size_t old_size = ctx->odp_actions->size; | |
5808 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
5809 | ctx->base_flow = old_base_flow; | |
5810 | ctx->odp_actions->size = old_size; | |
5811 | } | |
5812 | } | |
5813 | ||
0a740f48 EJ |
5814 | ctx->flow = old_flow; |
5815 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
5816 | ||
5817 | if (ctx->resubmit_stats) { | |
b9ad7294 EJ |
5818 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); |
5819 | netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats); | |
0a740f48 EJ |
5820 | } |
5821 | ||
5822 | return; | |
5823 | } | |
5824 | ||
a4454ac6 EJ |
5825 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
5826 | if (pdscp) { | |
5827 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
5828 | ctx->flow.nw_tos |= pdscp->dscp; | |
d59906fb EJ |
5829 | } |
5830 | ||
b9ad7294 EJ |
5831 | if (ofport->tnl_port) { |
5832 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow); | |
5833 | if (odp_port == OVSP_NONE) { | |
5834 | xlate_report(ctx, "Tunneling decided against output"); | |
5835 | return; | |
5836 | } | |
5837 | ||
5838 | if (ctx->resubmit_stats) { | |
5839 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); | |
5840 | } | |
5841 | out_port = odp_port; | |
5842 | commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow, | |
5843 | ctx->odp_actions); | |
5844 | } else { | |
cf630ea3 | 5845 | odp_port = ofport->odp_port; |
b9ad7294 EJ |
5846 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
5847 | ctx->flow.vlan_tci); | |
5848 | if (out_port != odp_port) { | |
5849 | ctx->flow.vlan_tci = htons(0); | |
5850 | } | |
321fa429 | 5851 | ctx->flow.skb_mark &= ~IPSEC_MARK; |
52a90c29 | 5852 | } |
5bbda0aa | 5853 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
5854 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
5855 | ||
6ff686f2 PS |
5856 | ctx->sflow_odp_port = odp_port; |
5857 | ctx->sflow_n_outputs++; | |
81b1afb1 | 5858 | ctx->nf_output_iface = ofp_port; |
b9ad7294 | 5859 | ctx->flow.tunnel.tun_id = flow_tun_id; |
52a90c29 | 5860 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 5861 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
5862 | } |
5863 | ||
abe529af | 5864 | static void |
5e48dc2b | 5865 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 5866 | { |
81b1afb1 | 5867 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
5868 | } |
5869 | ||
55599423 JR |
5870 | static void |
5871 | tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
5872 | { | |
5873 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5874 | uint8_t table_id = ctx->table_id; | |
5875 | ||
5876 | if (table_id > 0 && table_id < N_TABLES) { | |
5877 | struct table_dpif *table = &ofproto->tables[table_id]; | |
5878 | if (table->other_table) { | |
5879 | ctx->tags |= (rule && rule->tag | |
5880 | ? rule->tag | |
5881 | : rule_calculate_tag(&ctx->flow, | |
5882 | &table->other_table->mask, | |
5883 | table->basis)); | |
5884 | } | |
5885 | } | |
5886 | } | |
5887 | ||
5888 | /* Common rule processing in one place to avoid duplicating code. */ | |
5889 | static struct rule_dpif * | |
5890 | ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule, | |
5891 | bool may_packet_in) | |
5892 | { | |
5893 | if (ctx->resubmit_hook) { | |
5894 | ctx->resubmit_hook(ctx, rule); | |
5895 | } | |
5896 | if (rule == NULL && may_packet_in) { | |
5897 | /* XXX | |
5898 | * check if table configuration flags | |
5899 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
5900 | * OFPTC_TABLE_MISS_CONTINUE, | |
5901 | * OFPTC_TABLE_MISS_DROP | |
5902 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
5903 | */ | |
5904 | rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow); | |
5905 | } | |
5906 | if (rule && ctx->resubmit_stats) { | |
5907 | rule_credit_stats(rule, ctx->resubmit_stats); | |
5908 | } | |
5909 | return rule; | |
5910 | } | |
5911 | ||
abe529af | 5912 | static void |
29901626 | 5913 | xlate_table_action(struct action_xlate_ctx *ctx, |
1688c479 | 5914 | uint16_t in_port, uint8_t table_id, bool may_packet_in) |
abe529af BP |
5915 | { |
5916 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
5917 | struct rule_dpif *rule; | |
55599423 JR |
5918 | uint16_t old_in_port = ctx->flow.in_port; |
5919 | uint8_t old_table_id = ctx->table_id; | |
29901626 | 5920 | |
29901626 | 5921 | ctx->table_id = table_id; |
abe529af | 5922 | |
54a9cbc9 | 5923 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af | 5924 | ctx->flow.in_port = in_port; |
55599423 JR |
5925 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id); |
5926 | ||
5927 | tag_the_flow(ctx, rule); | |
54a9cbc9 BP |
5928 | |
5929 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
5930 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
5931 | ctx->flow.in_port = old_in_port; |
5932 | ||
55599423 | 5933 | rule = ctx_rule_hooks(ctx, rule, may_packet_in); |
1688c479 | 5934 | |
abe529af | 5935 | if (rule) { |
18b2a258 | 5936 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 5937 | |
abe529af | 5938 | ctx->recurse++; |
18b2a258 | 5939 | ctx->rule = rule; |
f25d0cf3 | 5940 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 5941 | ctx->rule = old_rule; |
abe529af BP |
5942 | ctx->recurse--; |
5943 | } | |
29901626 BP |
5944 | |
5945 | ctx->table_id = old_table_id; | |
abe529af BP |
5946 | } else { |
5947 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5948 | ||
29901626 | 5949 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 5950 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 5951 | ctx->max_resubmit_trigger = true; |
abe529af BP |
5952 | } |
5953 | } | |
5954 | ||
29901626 | 5955 | static void |
f25d0cf3 BP |
5956 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, |
5957 | const struct ofpact_resubmit *resubmit) | |
29901626 BP |
5958 | { |
5959 | uint16_t in_port; | |
5960 | uint8_t table_id; | |
5961 | ||
f25d0cf3 BP |
5962 | in_port = resubmit->in_port; |
5963 | if (in_port == OFPP_IN_PORT) { | |
5964 | in_port = ctx->flow.in_port; | |
5965 | } | |
5966 | ||
5967 | table_id = resubmit->table_id; | |
5968 | if (table_id == 255) { | |
5969 | table_id = ctx->table_id; | |
5970 | } | |
29901626 | 5971 | |
1688c479 | 5972 | xlate_table_action(ctx, in_port, table_id, false); |
29901626 BP |
5973 | } |
5974 | ||
abe529af | 5975 | static void |
d59906fb | 5976 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
5977 | { |
5978 | struct ofport_dpif *ofport; | |
5979 | ||
b3e9b2ed | 5980 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 5981 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
5982 | |
5983 | if (ofp_port == ctx->flow.in_port) { | |
5984 | continue; | |
5985 | } | |
5986 | ||
5e48dc2b | 5987 | if (all) { |
81b1afb1 | 5988 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 5989 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 5990 | compose_output_action(ctx, ofp_port); |
abe529af BP |
5991 | } |
5992 | } | |
b3e9b2ed EJ |
5993 | |
5994 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
5995 | } |
5996 | ||
6ff686f2 | 5997 | static void |
f0fd1a17 | 5998 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
5999 | enum ofp_packet_in_reason reason, |
6000 | uint16_t controller_id) | |
6ff686f2 | 6001 | { |
999fba59 EJ |
6002 | struct ofputil_packet_in pin; |
6003 | struct ofpbuf *packet; | |
6ff686f2 | 6004 | |
6a7e895f | 6005 | ctx->slow |= SLOW_CONTROLLER; |
999fba59 EJ |
6006 | if (!ctx->packet) { |
6007 | return; | |
6008 | } | |
6009 | ||
6010 | packet = ofpbuf_clone(ctx->packet); | |
6011 | ||
6012 | if (packet->l2 && packet->l3) { | |
6013 | struct eth_header *eh; | |
b02475c5 | 6014 | uint16_t mpls_depth; |
999fba59 EJ |
6015 | |
6016 | eth_pop_vlan(packet); | |
6017 | eh = packet->l2; | |
0104aba8 | 6018 | |
999fba59 EJ |
6019 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
6020 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
6021 | ||
6022 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
6023 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
6024 | } | |
6025 | ||
b02475c5 SH |
6026 | mpls_depth = eth_mpls_depth(packet); |
6027 | ||
6028 | if (mpls_depth < ctx->flow.mpls_depth) { | |
6029 | push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse); | |
6030 | } else if (mpls_depth > ctx->flow.mpls_depth) { | |
6031 | pop_mpls(packet, ctx->flow.dl_type); | |
6032 | } else if (mpls_depth) { | |
6033 | set_mpls_lse(packet, ctx->flow.mpls_lse); | |
6034 | } | |
6035 | ||
999fba59 EJ |
6036 | if (packet->l4) { |
6037 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6038 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
6039 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
6040 | } | |
6041 | ||
6042 | if (packet->l7) { | |
6043 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
6044 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
6045 | ctx->flow.tp_dst); | |
6046 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
6047 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
6048 | ctx->flow.tp_dst); | |
6049 | } | |
6050 | } | |
6051 | } | |
6052 | } | |
6053 | ||
6054 | pin.packet = packet->data; | |
6055 | pin.packet_len = packet->size; | |
f0fd1a17 | 6056 | pin.reason = reason; |
a7349929 | 6057 | pin.controller_id = controller_id; |
54834960 | 6058 | pin.table_id = ctx->table_id; |
18b2a258 | 6059 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 6060 | |
999fba59 | 6061 | pin.send_len = len; |
999fba59 EJ |
6062 | flow_get_metadata(&ctx->flow, &pin.fmd); |
6063 | ||
d8653c38 | 6064 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 6065 | ofpbuf_delete(packet); |
6ff686f2 PS |
6066 | } |
6067 | ||
b02475c5 SH |
6068 | static void |
6069 | execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6070 | { | |
6071 | ovs_assert(eth_type_mpls(eth_type)); | |
6072 | ||
6073 | if (ctx->base_flow.mpls_depth) { | |
6074 | ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); | |
6075 | ctx->flow.mpls_depth++; | |
6076 | } else { | |
6077 | ovs_be32 label; | |
6078 | uint8_t tc, ttl; | |
6079 | ||
6080 | if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) { | |
6081 | label = htonl(0x2); /* IPV6 Explicit Null. */ | |
6082 | } else { | |
6083 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6084 | } | |
6085 | tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2; | |
6086 | ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40; | |
6087 | ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
b02475c5 SH |
6088 | ctx->flow.mpls_depth = 1; |
6089 | } | |
6090 | ctx->flow.dl_type = eth_type; | |
6091 | } | |
6092 | ||
6093 | static void | |
6094 | execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6095 | { | |
6096 | ovs_assert(eth_type_mpls(ctx->flow.dl_type)); | |
6097 | ovs_assert(!eth_type_mpls(eth_type)); | |
6098 | ||
6099 | if (ctx->flow.mpls_depth) { | |
6100 | ctx->flow.mpls_depth--; | |
6101 | ctx->flow.mpls_lse = htonl(0); | |
6102 | if (!ctx->flow.mpls_depth) { | |
6103 | ctx->flow.dl_type = eth_type; | |
b02475c5 SH |
6104 | } |
6105 | } | |
6106 | } | |
6107 | ||
f0fd1a17 | 6108 | static bool |
c2d967a5 | 6109 | compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) |
f0fd1a17 PS |
6110 | { |
6111 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
6112 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
6113 | return false; | |
6114 | } | |
6115 | ||
6116 | if (ctx->flow.nw_ttl > 1) { | |
6117 | ctx->flow.nw_ttl--; | |
6118 | return false; | |
6119 | } else { | |
c2d967a5 MM |
6120 | size_t i; |
6121 | ||
6122 | for (i = 0; i < ids->n_controllers; i++) { | |
6123 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6124 | ids->cnt_ids[i]); | |
6125 | } | |
f0fd1a17 PS |
6126 | |
6127 | /* Stop processing for current table. */ | |
6128 | return true; | |
6129 | } | |
6130 | } | |
6131 | ||
0f3f3c3d SH |
6132 | static bool |
6133 | execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl) | |
6134 | { | |
6135 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6136 | return true; | |
6137 | } | |
6138 | ||
6139 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6140 | return false; | |
6141 | } | |
6142 | ||
b676167a SH |
6143 | static bool |
6144 | execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx) | |
6145 | { | |
6146 | uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse); | |
6147 | ||
6148 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6149 | return false; | |
6150 | } | |
6151 | ||
be80bc65 | 6152 | if (ttl > 1) { |
b676167a SH |
6153 | ttl--; |
6154 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6155 | return false; | |
6156 | } else { | |
6157 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
6158 | ||
6159 | /* Stop processing for current table. */ | |
6160 | return true; | |
6161 | } | |
6162 | } | |
6163 | ||
abe529af | 6164 | static void |
f25d0cf3 | 6165 | xlate_output_action(struct action_xlate_ctx *ctx, |
1688c479 | 6166 | uint16_t port, uint16_t max_len, bool may_packet_in) |
abe529af BP |
6167 | { |
6168 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
6169 | ||
6170 | ctx->nf_output_iface = NF_OUT_DROP; | |
6171 | ||
6172 | switch (port) { | |
6173 | case OFPP_IN_PORT: | |
81b1afb1 | 6174 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
6175 | break; |
6176 | case OFPP_TABLE: | |
1688c479 | 6177 | xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); |
abe529af BP |
6178 | break; |
6179 | case OFPP_NORMAL: | |
6180 | xlate_normal(ctx); | |
6181 | break; | |
6182 | case OFPP_FLOOD: | |
d59906fb | 6183 | flood_packets(ctx, false); |
abe529af BP |
6184 | break; |
6185 | case OFPP_ALL: | |
d59906fb | 6186 | flood_packets(ctx, true); |
abe529af BP |
6187 | break; |
6188 | case OFPP_CONTROLLER: | |
a7349929 | 6189 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 6190 | break; |
e81d2933 EJ |
6191 | case OFPP_NONE: |
6192 | break; | |
a0fbe94a | 6193 | case OFPP_LOCAL: |
abe529af BP |
6194 | default: |
6195 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 6196 | compose_output_action(ctx, port); |
3dd3eace BP |
6197 | } else { |
6198 | xlate_report(ctx, "skipping output to input port"); | |
abe529af BP |
6199 | } |
6200 | break; | |
6201 | } | |
6202 | ||
6203 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
6204 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6205 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
6206 | ctx->nf_output_iface = prev_nf_output_iface; | |
6207 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
6208 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
6209 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6210 | } | |
6211 | } | |
6212 | ||
f694937d EJ |
6213 | static void |
6214 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6215 | const struct ofpact_output_reg *or) |
f694937d | 6216 | { |
f25d0cf3 BP |
6217 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); |
6218 | if (port <= UINT16_MAX) { | |
1688c479 | 6219 | xlate_output_action(ctx, port, or->max_len, false); |
f694937d EJ |
6220 | } |
6221 | } | |
6222 | ||
abe529af BP |
6223 | static void |
6224 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6225 | const struct ofpact_enqueue *enqueue) |
abe529af | 6226 | { |
f25d0cf3 BP |
6227 | uint16_t ofp_port = enqueue->port; |
6228 | uint32_t queue_id = enqueue->queue; | |
abff858b | 6229 | uint32_t flow_priority, priority; |
abe529af BP |
6230 | int error; |
6231 | ||
f25d0cf3 | 6232 | /* Translate queue to priority. */ |
acf60855 JP |
6233 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6234 | queue_id, &priority); | |
abe529af BP |
6235 | if (error) { |
6236 | /* Fall back to ordinary output action. */ | |
1688c479 | 6237 | xlate_output_action(ctx, enqueue->port, 0, false); |
abe529af BP |
6238 | return; |
6239 | } | |
6240 | ||
f25d0cf3 | 6241 | /* Check output port. */ |
abe529af BP |
6242 | if (ofp_port == OFPP_IN_PORT) { |
6243 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
6244 | } else if (ofp_port == ctx->flow.in_port) { |
6245 | return; | |
abe529af | 6246 | } |
abe529af | 6247 | |
df2c07f4 | 6248 | /* Add datapath actions. */ |
deedf7e7 BP |
6249 | flow_priority = ctx->flow.skb_priority; |
6250 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 6251 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 6252 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
6253 | |
6254 | /* Update NetFlow output port. */ | |
6255 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 6256 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
6257 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
6258 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6259 | } | |
6260 | } | |
6261 | ||
6262 | static void | |
f25d0cf3 | 6263 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 6264 | { |
f25d0cf3 | 6265 | uint32_t skb_priority; |
abe529af | 6266 | |
acf60855 JP |
6267 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, |
6268 | queue_id, &skb_priority)) { | |
f25d0cf3 BP |
6269 | ctx->flow.skb_priority = skb_priority; |
6270 | } else { | |
6271 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 6272 | * has already been logged. */ |
abe529af | 6273 | } |
abe529af BP |
6274 | } |
6275 | ||
6276 | struct xlate_reg_state { | |
6277 | ovs_be16 vlan_tci; | |
6278 | ovs_be64 tun_id; | |
6279 | }; | |
6280 | ||
daff3353 EJ |
6281 | static bool |
6282 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6283 | { | |
6284 | struct ofproto_dpif *ofproto = ofproto_; | |
6285 | struct ofport_dpif *port; | |
6286 | ||
6287 | switch (ofp_port) { | |
6288 | case OFPP_IN_PORT: | |
6289 | case OFPP_TABLE: | |
6290 | case OFPP_NORMAL: | |
6291 | case OFPP_FLOOD: | |
6292 | case OFPP_ALL: | |
439e4d8c | 6293 | case OFPP_NONE: |
daff3353 EJ |
6294 | return true; |
6295 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6296 | return false; | |
6297 | default: | |
6298 | port = get_ofp_port(ofproto, ofp_port); | |
6299 | return port ? port->may_enable : false; | |
6300 | } | |
6301 | } | |
6302 | ||
f25d0cf3 BP |
6303 | static void |
6304 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
6305 | const struct ofpact_bundle *bundle) | |
6306 | { | |
6307 | uint16_t port; | |
6308 | ||
6309 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
6310 | if (bundle->dst.field) { | |
6311 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
6312 | } else { | |
1688c479 | 6313 | xlate_output_action(ctx, port, 0, false); |
f25d0cf3 BP |
6314 | } |
6315 | } | |
6316 | ||
75a75043 BP |
6317 | static void |
6318 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6319 | const struct ofpact_learn *learn) |
75a75043 BP |
6320 | { |
6321 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6322 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
6323 | uint64_t ofpacts_stub[1024 / 8]; |
6324 | struct ofpbuf ofpacts; | |
75a75043 BP |
6325 | int error; |
6326 | ||
f25d0cf3 BP |
6327 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
6328 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
75a75043 BP |
6329 | |
6330 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6331 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
6332 | VLOG_WARN("learning action failed to modify flow table (%s)", |
6333 | ofperr_get_name(error)); | |
75a75043 BP |
6334 | } |
6335 | ||
f25d0cf3 | 6336 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
6337 | } |
6338 | ||
0e553d9c BP |
6339 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
6340 | * means "infinite". */ | |
6341 | static void | |
6342 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6343 | { | |
6344 | if (max && (!*timeout || *timeout > max)) { | |
6345 | *timeout = max; | |
6346 | } | |
6347 | } | |
6348 | ||
6349 | static void | |
6350 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 6351 | const struct ofpact_fin_timeout *oft) |
0e553d9c BP |
6352 | { |
6353 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
6354 | struct rule_dpif *rule = ctx->rule; | |
6355 | ||
f25d0cf3 BP |
6356 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
6357 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
6358 | } |
6359 | } | |
6360 | ||
21f7563c JP |
6361 | static bool |
6362 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
6363 | { | |
9e1fd49b BP |
6364 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
6365 | ? OFPUTIL_PC_NO_RECV_STP | |
6366 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
6367 | return false; |
6368 | } | |
6369 | ||
6370 | /* Only drop packets here if both forwarding and learning are | |
6371 | * disabled. If just learning is enabled, we need to have | |
6372 | * OFPP_NORMAL and the learning action have a look at the packet | |
6373 | * before we can drop it. */ | |
6374 | if (!stp_forward_in_state(port->stp_state) | |
6375 | && !stp_learn_in_state(port->stp_state)) { | |
6376 | return false; | |
6377 | } | |
6378 | ||
6379 | return true; | |
6380 | } | |
6381 | ||
4863c249 JP |
6382 | static bool |
6383 | tunnel_ecn_ok(struct action_xlate_ctx *ctx) | |
6384 | { | |
6385 | if (is_ip_any(&ctx->base_flow) | |
29a5df0a JP |
6386 | && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { |
6387 | if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { | |
6388 | VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" | |
6389 | " but is not ECN capable"); | |
6390 | return false; | |
6391 | } else { | |
6392 | /* Set the ECN CE value in the tunneled packet. */ | |
6393 | ctx->flow.nw_tos |= IP_ECN_CE; | |
6394 | } | |
4863c249 JP |
6395 | } |
6396 | ||
6397 | return true; | |
6398 | } | |
6399 | ||
abe529af | 6400 | static void |
f25d0cf3 | 6401 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
abe529af BP |
6402 | struct action_xlate_ctx *ctx) |
6403 | { | |
254750ce | 6404 | bool was_evictable = true; |
f25d0cf3 | 6405 | const struct ofpact *a; |
abe529af | 6406 | |
254750ce BP |
6407 | if (ctx->rule) { |
6408 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6409 | was_evictable = ctx->rule->up.evictable; | |
6410 | ctx->rule->up.evictable = false; | |
6411 | } | |
55599423 JR |
6412 | |
6413 | do_xlate_actions_again: | |
f25d0cf3 BP |
6414 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
6415 | struct ofpact_controller *controller; | |
4cceacb9 | 6416 | const struct ofpact_metadata *metadata; |
38f2e360 | 6417 | |
848e8809 EJ |
6418 | if (ctx->exit) { |
6419 | break; | |
6420 | } | |
6421 | ||
f25d0cf3 BP |
6422 | switch (a->type) { |
6423 | case OFPACT_OUTPUT: | |
6424 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
1688c479 | 6425 | ofpact_get_OUTPUT(a)->max_len, true); |
f25d0cf3 BP |
6426 | break; |
6427 | ||
6428 | case OFPACT_CONTROLLER: | |
6429 | controller = ofpact_get_CONTROLLER(a); | |
6430 | execute_controller_action(ctx, controller->max_len, | |
6431 | controller->reason, | |
6432 | controller->controller_id); | |
6433 | break; | |
690a61c5 | 6434 | |
f25d0cf3 BP |
6435 | case OFPACT_ENQUEUE: |
6436 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
6437 | break; |
6438 | ||
f25d0cf3 | 6439 | case OFPACT_SET_VLAN_VID: |
abe529af | 6440 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
f25d0cf3 BP |
6441 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) |
6442 | | htons(VLAN_CFI)); | |
abe529af BP |
6443 | break; |
6444 | ||
f25d0cf3 | 6445 | case OFPACT_SET_VLAN_PCP: |
abe529af | 6446 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
f25d0cf3 BP |
6447 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp |
6448 | << VLAN_PCP_SHIFT) | |
6449 | | VLAN_CFI); | |
abe529af BP |
6450 | break; |
6451 | ||
f25d0cf3 | 6452 | case OFPACT_STRIP_VLAN: |
abe529af | 6453 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
6454 | break; |
6455 | ||
3e34fbdd | 6456 | case OFPACT_PUSH_VLAN: |
5dca28b5 | 6457 | /* XXX 802.1AD(QinQ) */ |
3e34fbdd IY |
6458 | ctx->flow.vlan_tci = htons(VLAN_CFI); |
6459 | break; | |
6460 | ||
f25d0cf3 BP |
6461 | case OFPACT_SET_ETH_SRC: |
6462 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
6463 | ETH_ADDR_LEN); | |
abe529af BP |
6464 | break; |
6465 | ||
f25d0cf3 BP |
6466 | case OFPACT_SET_ETH_DST: |
6467 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
6468 | ETH_ADDR_LEN); | |
abe529af BP |
6469 | break; |
6470 | ||
f25d0cf3 | 6471 | case OFPACT_SET_IPV4_SRC: |
1b035ef2 SH |
6472 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { |
6473 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
6474 | } | |
abe529af BP |
6475 | break; |
6476 | ||
f25d0cf3 | 6477 | case OFPACT_SET_IPV4_DST: |
1b035ef2 SH |
6478 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { |
6479 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
6480 | } | |
abe529af BP |
6481 | break; |
6482 | ||
f25d0cf3 | 6483 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d PS |
6484 | /* OpenFlow 1.0 only supports IPv4. */ |
6485 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6486 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
f25d0cf3 | 6487 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; |
c4f2731d | 6488 | } |
abe529af BP |
6489 | break; |
6490 | ||
f25d0cf3 | 6491 | case OFPACT_SET_L4_SRC_PORT: |
1b035ef2 SH |
6492 | if (is_ip_any(&ctx->flow)) { |
6493 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
6494 | } | |
abe529af BP |
6495 | break; |
6496 | ||
f25d0cf3 | 6497 | case OFPACT_SET_L4_DST_PORT: |
1b035ef2 SH |
6498 | if (is_ip_any(&ctx->flow)) { |
6499 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
6500 | } | |
abe529af BP |
6501 | break; |
6502 | ||
f25d0cf3 BP |
6503 | case OFPACT_RESUBMIT: |
6504 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
6505 | break; |
6506 | ||
f25d0cf3 | 6507 | case OFPACT_SET_TUNNEL: |
296e07ac | 6508 | ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); |
29901626 BP |
6509 | break; |
6510 | ||
f25d0cf3 BP |
6511 | case OFPACT_SET_QUEUE: |
6512 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
6513 | break; |
6514 | ||
f25d0cf3 | 6515 | case OFPACT_POP_QUEUE: |
deedf7e7 | 6516 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
6517 | break; |
6518 | ||
f25d0cf3 BP |
6519 | case OFPACT_REG_MOVE: |
6520 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
38f2e360 BP |
6521 | break; |
6522 | ||
f25d0cf3 BP |
6523 | case OFPACT_REG_LOAD: |
6524 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
38f2e360 BP |
6525 | break; |
6526 | ||
bd85dac1 AZ |
6527 | case OFPACT_STACK_PUSH: |
6528 | nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow, | |
6529 | &ctx->stack); | |
6530 | break; | |
6531 | ||
6532 | case OFPACT_STACK_POP: | |
6533 | nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow, | |
6534 | &ctx->stack); | |
6535 | break; | |
6536 | ||
b02475c5 SH |
6537 | case OFPACT_PUSH_MPLS: |
6538 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6539 | break; | |
6540 | ||
6541 | case OFPACT_POP_MPLS: | |
6542 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6543 | break; | |
6544 | ||
0f3f3c3d SH |
6545 | case OFPACT_SET_MPLS_TTL: |
6546 | if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) { | |
6547 | goto out; | |
6548 | } | |
6549 | break; | |
6550 | ||
b676167a SH |
6551 | case OFPACT_DEC_MPLS_TTL: |
6552 | if (execute_dec_mpls_ttl_action(ctx)) { | |
6553 | goto out; | |
6554 | } | |
6555 | break; | |
6556 | ||
f25d0cf3 | 6557 | case OFPACT_DEC_TTL: |
c2d967a5 | 6558 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { |
f25d0cf3 BP |
6559 | goto out; |
6560 | } | |
38f2e360 BP |
6561 | break; |
6562 | ||
f25d0cf3 BP |
6563 | case OFPACT_NOTE: |
6564 | /* Nothing to do. */ | |
abe529af BP |
6565 | break; |
6566 | ||
f25d0cf3 BP |
6567 | case OFPACT_MULTIPATH: |
6568 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
abe529af | 6569 | break; |
daff3353 | 6570 | |
f25d0cf3 | 6571 | case OFPACT_BUNDLE: |
a368bb53 | 6572 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 6573 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 6574 | break; |
f694937d | 6575 | |
f25d0cf3 BP |
6576 | case OFPACT_OUTPUT_REG: |
6577 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 6578 | break; |
75a75043 | 6579 | |
f25d0cf3 | 6580 | case OFPACT_LEARN: |
75a75043 | 6581 | ctx->has_learn = true; |
3de9590b | 6582 | if (ctx->may_learn) { |
f25d0cf3 | 6583 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
6584 | } |
6585 | break; | |
848e8809 | 6586 | |
f25d0cf3 | 6587 | case OFPACT_EXIT: |
848e8809 EJ |
6588 | ctx->exit = true; |
6589 | break; | |
0e553d9c | 6590 | |
f25d0cf3 | 6591 | case OFPACT_FIN_TIMEOUT: |
0e553d9c | 6592 | ctx->has_fin_timeout = true; |
f25d0cf3 | 6593 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 6594 | break; |
8dd54666 | 6595 | |
b19e8793 | 6596 | case OFPACT_CLEAR_ACTIONS: |
5dca28b5 | 6597 | /* XXX |
b19e8793 IY |
6598 | * Nothing to do because writa-actions is not supported for now. |
6599 | * When writa-actions is supported, clear-actions also must | |
6600 | * be supported at the same time. | |
6601 | */ | |
6602 | break; | |
6603 | ||
4cceacb9 JS |
6604 | case OFPACT_WRITE_METADATA: |
6605 | metadata = ofpact_get_WRITE_METADATA(a); | |
6606 | ctx->flow.metadata &= ~metadata->mask; | |
6607 | ctx->flow.metadata |= metadata->metadata & metadata->mask; | |
6608 | break; | |
6609 | ||
8dd54666 | 6610 | case OFPACT_GOTO_TABLE: { |
55599423 | 6611 | /* It is assumed that goto-table is the last action. */ |
8dd54666 | 6612 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); |
55599423 JR |
6613 | struct rule_dpif *rule; |
6614 | ||
cb22974d | 6615 | ovs_assert(ctx->table_id < ogt->table_id); |
55599423 JR |
6616 | |
6617 | ctx->table_id = ogt->table_id; | |
6618 | ||
6619 | /* Look up a flow from the new table. */ | |
6620 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id); | |
6621 | ||
6622 | tag_the_flow(ctx, rule); | |
6623 | ||
6624 | rule = ctx_rule_hooks(ctx, rule, true); | |
6625 | ||
6626 | if (rule) { | |
6627 | if (ctx->rule) { | |
6628 | ctx->rule->up.evictable = was_evictable; | |
6629 | } | |
6630 | ctx->rule = rule; | |
6631 | was_evictable = rule->up.evictable; | |
6632 | rule->up.evictable = false; | |
6633 | ||
6634 | /* Tail recursion removal. */ | |
6635 | ofpacts = rule->up.ofpacts; | |
6636 | ofpacts_len = rule->up.ofpacts_len; | |
6637 | goto do_xlate_actions_again; | |
6638 | } | |
8dd54666 IY |
6639 | break; |
6640 | } | |
abe529af BP |
6641 | } |
6642 | } | |
21f7563c | 6643 | |
f0fd1a17 | 6644 | out: |
254750ce BP |
6645 | if (ctx->rule) { |
6646 | ctx->rule->up.evictable = was_evictable; | |
6647 | } | |
abe529af BP |
6648 | } |
6649 | ||
6650 | static void | |
6651 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
6652 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
6653 | const struct initial_vals *initial_vals, |
6654 | struct rule_dpif *rule, | |
0e553d9c | 6655 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af | 6656 | { |
ef506a7c JG |
6657 | ovs_be64 initial_tun_id = flow->tunnel.tun_id; |
6658 | ||
6659 | /* Flow initialization rules: | |
6660 | * - 'base_flow' must match the kernel's view of the packet at the | |
6661 | * time that action processing starts. 'flow' represents any | |
6662 | * transformations we wish to make through actions. | |
6663 | * - By default 'base_flow' and 'flow' are the same since the input | |
6664 | * packet matches the output before any actions are applied. | |
6665 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
6666 | * of the received packet as seen by the kernel. If we later output | |
6667 | * to another device without any modifications this will cause us to | |
6668 | * insert a new tag since the original one was stripped off by the | |
6669 | * VLAN device. | |
6670 | * - Tunnel 'flow' is largely cleared when transitioning between | |
6671 | * the input and output stages since it does not make sense to output | |
6672 | * a packet with the exact headers that it was received with (i.e. | |
6673 | * the destination IP is us). The one exception is the tun_id, which | |
6674 | * is preserved to allow use in later resubmit lookups and loads into | |
6675 | * registers. | |
6676 | * - Tunnel 'base_flow' is completely cleared since that is what the | |
6677 | * kernel does. If we wish to maintain the original values an action | |
6678 | * needs to be generated. */ | |
6679 | ||
abe529af BP |
6680 | ctx->ofproto = ofproto; |
6681 | ctx->flow = *flow; | |
47d4a9db | 6682 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); |
e84173dc | 6683 | ctx->base_flow = ctx->flow; |
14f94f9a | 6684 | ctx->base_flow.vlan_tci = initial_vals->vlan_tci; |
c3f6c502 | 6685 | ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos; |
ef506a7c | 6686 | ctx->flow.tunnel.tun_id = initial_tun_id; |
18b2a258 | 6687 | ctx->rule = rule; |
abe529af | 6688 | ctx->packet = packet; |
3de9590b | 6689 | ctx->may_learn = packet != NULL; |
0e553d9c | 6690 | ctx->tcp_flags = tcp_flags; |
abe529af | 6691 | ctx->resubmit_hook = NULL; |
479df176 | 6692 | ctx->report_hook = NULL; |
112bc5f4 | 6693 | ctx->resubmit_stats = NULL; |
abe529af BP |
6694 | } |
6695 | ||
f25d0cf3 BP |
6696 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' |
6697 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
050ac423 | 6698 | static void |
abe529af | 6699 | xlate_actions(struct action_xlate_ctx *ctx, |
f25d0cf3 | 6700 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 | 6701 | struct ofpbuf *odp_actions) |
abe529af | 6702 | { |
43d50bc8 BP |
6703 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so |
6704 | * that in the future we always keep a copy of the original flow for | |
6705 | * tracing purposes. */ | |
6706 | static bool hit_resubmit_limit; | |
6707 | ||
6a7e895f | 6708 | enum slow_path_reason special; |
ffaef958 | 6709 | struct ofport_dpif *in_port; |
9ba85077 | 6710 | struct flow orig_flow; |
6a7e895f | 6711 | |
abe529af BP |
6712 | COVERAGE_INC(ofproto_dpif_xlate); |
6713 | ||
050ac423 BP |
6714 | ofpbuf_clear(odp_actions); |
6715 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
6716 | ||
6717 | ctx->odp_actions = odp_actions; | |
97e42c92 | 6718 | ctx->tags = 0; |
6a7e895f | 6719 | ctx->slow = 0; |
97e42c92 BP |
6720 | ctx->has_learn = false; |
6721 | ctx->has_normal = false; | |
0e553d9c | 6722 | ctx->has_fin_timeout = false; |
97e42c92 | 6723 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 6724 | ctx->mirrors = 0; |
97e42c92 | 6725 | ctx->recurse = 0; |
6a6455e5 | 6726 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 6727 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 6728 | ctx->table_id = 0; |
848e8809 | 6729 | ctx->exit = false; |
7257b535 | 6730 | |
bd85dac1 AZ |
6731 | ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack); |
6732 | ||
43d50bc8 | 6733 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 | 6734 | /* Do this conditionally because the copy is expensive enough that it |
9ba85077 BP |
6735 | * shows up in profiles. */ |
6736 | orig_flow = ctx->flow; | |
ccb7c863 BP |
6737 | } |
6738 | ||
eadef313 | 6739 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
6740 | switch (ctx->ofproto->up.frag_handling) { |
6741 | case OFPC_FRAG_NORMAL: | |
6742 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
6743 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
6744 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
6745 | break; |
6746 | ||
6747 | case OFPC_FRAG_DROP: | |
050ac423 | 6748 | return; |
7257b535 BP |
6749 | |
6750 | case OFPC_FRAG_REASM: | |
6751 | NOT_REACHED(); | |
6752 | ||
6753 | case OFPC_FRAG_NX_MATCH: | |
6754 | /* Nothing to do. */ | |
6755 | break; | |
f0fd1a17 PS |
6756 | |
6757 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
6758 | NOT_REACHED(); | |
7257b535 BP |
6759 | } |
6760 | } | |
6761 | ||
ffaef958 BP |
6762 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
6763 | special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet); | |
6a7e895f BP |
6764 | if (special) { |
6765 | ctx->slow |= special; | |
abe529af | 6766 | } else { |
6a6455e5 | 6767 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
14f94f9a | 6768 | struct initial_vals initial_vals; |
ee382d89 | 6769 | uint32_t local_odp_port; |
6a6455e5 | 6770 | |
14f94f9a | 6771 | initial_vals.vlan_tci = ctx->base_flow.vlan_tci; |
c3f6c502 | 6772 | initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos; |
14f94f9a | 6773 | |
6ff686f2 | 6774 | add_sflow_action(ctx); |
ffaef958 | 6775 | |
4863c249 | 6776 | if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) { |
ffaef958 BP |
6777 | do_xlate_actions(ofpacts, ofpacts_len, ctx); |
6778 | ||
6779 | /* We've let OFPP_NORMAL and the learning action look at the | |
6780 | * packet, so drop it now if forwarding is disabled. */ | |
6781 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
6782 | ofpbuf_clear(ctx->odp_actions); | |
6783 | add_sflow_action(ctx); | |
6784 | } | |
6785 | } | |
abe529af | 6786 | |
43d50bc8 BP |
6787 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { |
6788 | if (!hit_resubmit_limit) { | |
6789 | /* We didn't record the original flow. Make sure we do from | |
6790 | * now on. */ | |
6791 | hit_resubmit_limit = true; | |
6792 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
6793 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6794 | ||
9ba85077 | 6795 | ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet, |
14f94f9a | 6796 | &initial_vals, &ds); |
43d50bc8 BP |
6797 | VLOG_ERR("Trace triggered by excessive resubmit " |
6798 | "recursion:\n%s", ds_cstr(&ds)); | |
6799 | ds_destroy(&ds); | |
6800 | } | |
6a6455e5 EJ |
6801 | } |
6802 | ||
ee382d89 | 6803 | local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); |
b6848f13 | 6804 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
ee382d89 | 6805 | local_odp_port, |
b6848f13 BP |
6806 | ctx->odp_actions->data, |
6807 | ctx->odp_actions->size)) { | |
6a7e895f | 6808 | ctx->slow |= SLOW_IN_BAND; |
b6848f13 BP |
6809 | if (ctx->packet |
6810 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
6811 | ctx->packet)) { | |
5e48dc2b | 6812 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
6813 | } |
6814 | } | |
ccb7c863 | 6815 | if (ctx->ofproto->has_mirrors) { |
9ba85077 | 6816 | add_mirror_actions(ctx, &orig_flow); |
ccb7c863 | 6817 | } |
a7c4eaf6 | 6818 | fix_sflow_action(ctx); |
abe529af | 6819 | } |
bd85dac1 AZ |
6820 | |
6821 | ofpbuf_uninit(&ctx->stack); | |
050ac423 BP |
6822 | } |
6823 | ||
f25d0cf3 BP |
6824 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
6825 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 BP |
6826 | static void |
6827 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
f25d0cf3 BP |
6828 | const struct ofpact *ofpacts, |
6829 | size_t ofpacts_len) | |
050ac423 BP |
6830 | { |
6831 | uint64_t odp_actions_stub[1024 / 8]; | |
6832 | struct ofpbuf odp_actions; | |
abe529af | 6833 | |
050ac423 | 6834 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
f25d0cf3 | 6835 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); |
050ac423 | 6836 | ofpbuf_uninit(&odp_actions); |
abe529af | 6837 | } |
479df176 BP |
6838 | |
6839 | static void | |
6840 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
6841 | { | |
6842 | if (ctx->report_hook) { | |
6843 | ctx->report_hook(ctx, s); | |
6844 | } | |
6845 | } | |
abe529af BP |
6846 | \f |
6847 | /* OFPP_NORMAL implementation. */ | |
6848 | ||
abe529af BP |
6849 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
6850 | ||
ecac4ebf BP |
6851 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
6852 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
6853 | * the bundle on which the packet was received, returns the VLAN to which the | |
6854 | * packet belongs. | |
6855 | * | |
6856 | * Both 'vid' and the return value are in the range 0...4095. */ | |
6857 | static uint16_t | |
6858 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
6859 | { | |
6860 | switch (in_bundle->vlan_mode) { | |
6861 | case PORT_VLAN_ACCESS: | |
6862 | return in_bundle->vlan; | |
6863 | break; | |
6864 | ||
6865 | case PORT_VLAN_TRUNK: | |
6866 | return vid; | |
6867 | ||
6868 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6869 | case PORT_VLAN_NATIVE_TAGGED: | |
6870 | return vid ? vid : in_bundle->vlan; | |
6871 | ||
6872 | default: | |
6873 | NOT_REACHED(); | |
6874 | } | |
6875 | } | |
6876 | ||
5da5ec37 BP |
6877 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
6878 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
6879 | * a warning. | |
6880 | * | |
6881 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
6882 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
6883 | * 0...4095. */ | |
6884 | static bool | |
6885 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
6886 | { | |
33158a18 JP |
6887 | /* Allow any VID on the OFPP_NONE port. */ |
6888 | if (in_bundle == &ofpp_none_bundle) { | |
6889 | return true; | |
6890 | } | |
6891 | ||
5da5ec37 BP |
6892 | switch (in_bundle->vlan_mode) { |
6893 | case PORT_VLAN_ACCESS: | |
6894 | if (vid) { | |
6895 | if (warn) { | |
6896 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6897 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
6898 | "packet received on port %s configured as VLAN " | |
6899 | "%"PRIu16" access port", | |
6900 | in_bundle->ofproto->up.name, vid, | |
6901 | in_bundle->name, in_bundle->vlan); | |
6902 | } | |
6903 | return false; | |
6904 | } | |
6905 | return true; | |
6906 | ||
6907 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6908 | case PORT_VLAN_NATIVE_TAGGED: | |
6909 | if (!vid) { | |
6910 | /* Port must always carry its native VLAN. */ | |
6911 | return true; | |
6912 | } | |
6913 | /* Fall through. */ | |
6914 | case PORT_VLAN_TRUNK: | |
6915 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
6916 | if (warn) { | |
6917 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6918 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
6919 | "received on port %s not configured for trunking " | |
6920 | "VLAN %"PRIu16, | |
6921 | in_bundle->ofproto->up.name, vid, | |
6922 | in_bundle->name, vid); | |
6923 | } | |
6924 | return false; | |
6925 | } | |
6926 | return true; | |
6927 | ||
6928 | default: | |
6929 | NOT_REACHED(); | |
6930 | } | |
6931 | ||
6932 | } | |
6933 | ||
ecac4ebf BP |
6934 | /* Given 'vlan', the VLAN that a packet belongs to, and |
6935 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
6936 | * that should be included in the 802.1Q header. (If the return value is 0, | |
6937 | * then the 802.1Q header should only be included in the packet if there is a | |
6938 | * nonzero PCP.) | |
6939 | * | |
6940 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
6941 | static uint16_t | |
6942 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
6943 | { | |
6944 | switch (out_bundle->vlan_mode) { | |
6945 | case PORT_VLAN_ACCESS: | |
6946 | return 0; | |
6947 | ||
6948 | case PORT_VLAN_TRUNK: | |
6949 | case PORT_VLAN_NATIVE_TAGGED: | |
6950 | return vlan; | |
6951 | ||
6952 | case PORT_VLAN_NATIVE_UNTAGGED: | |
6953 | return vlan == out_bundle->vlan ? 0 : vlan; | |
6954 | ||
6955 | default: | |
6956 | NOT_REACHED(); | |
6957 | } | |
6958 | } | |
6959 | ||
395e68ce BP |
6960 | static void |
6961 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
6962 | uint16_t vlan) | |
abe529af | 6963 | { |
395e68ce BP |
6964 | struct ofport_dpif *port; |
6965 | uint16_t vid; | |
81b1afb1 | 6966 | ovs_be16 tci, old_tci; |
ecac4ebf | 6967 | |
395e68ce BP |
6968 | vid = output_vlan_to_vid(out_bundle, vlan); |
6969 | if (!out_bundle->bond) { | |
6970 | port = ofbundle_get_a_port(out_bundle); | |
6971 | } else { | |
6972 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
6973 | vid, &ctx->tags); | |
6974 | if (!port) { | |
6975 | /* No slaves enabled, so drop packet. */ | |
6976 | return; | |
6977 | } | |
6978 | } | |
abe529af | 6979 | |
81b1afb1 | 6980 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
6981 | tci = htons(vid); |
6982 | if (tci || out_bundle->use_priority_tags) { | |
6983 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
6984 | if (tci) { | |
6985 | tci |= htons(VLAN_CFI); | |
6986 | } | |
395e68ce | 6987 | } |
81b1afb1 | 6988 | ctx->flow.vlan_tci = tci; |
395e68ce | 6989 | |
5e48dc2b | 6990 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 6991 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
6992 | } |
6993 | ||
6994 | static int | |
6995 | mirror_mask_ffs(mirror_mask_t mask) | |
6996 | { | |
6997 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
6998 | return ffs(mask); | |
6999 | } | |
7000 | ||
abe529af BP |
7001 | static bool |
7002 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7003 | { | |
ecac4ebf | 7004 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 7005 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
7006 | } |
7007 | ||
7008 | static bool | |
7009 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7010 | { | |
7011 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
7012 | } | |
7013 | ||
7014 | /* Returns an arbitrary interface within 'bundle'. */ | |
7015 | static struct ofport_dpif * | |
7016 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
7017 | { | |
7018 | return CONTAINER_OF(list_front(&bundle->ports), | |
7019 | struct ofport_dpif, bundle_node); | |
7020 | } | |
7021 | ||
abe529af BP |
7022 | static bool |
7023 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
7024 | { | |
fc3d7408 | 7025 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
7026 | } |
7027 | ||
7028 | static void | |
c06bba01 | 7029 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
7030 | { |
7031 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7032 | mirror_mask_t mirrors; | |
c06bba01 JP |
7033 | struct ofbundle *in_bundle; |
7034 | uint16_t vlan; | |
7035 | uint16_t vid; | |
7036 | const struct nlattr *a; | |
7037 | size_t left; | |
7038 | ||
3581c12c | 7039 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
70c2fd56 | 7040 | ctx->packet != NULL, NULL); |
3581c12c | 7041 | if (!in_bundle) { |
c06bba01 JP |
7042 | return; |
7043 | } | |
c06bba01 JP |
7044 | mirrors = in_bundle->src_mirrors; |
7045 | ||
7046 | /* Drop frames on bundles reserved for mirroring. */ | |
7047 | if (in_bundle->mirror_out) { | |
7048 | if (ctx->packet != NULL) { | |
7049 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7050 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7051 | "%s, which is reserved exclusively for mirroring", | |
7052 | ctx->ofproto->up.name, in_bundle->name); | |
7053 | } | |
7054 | return; | |
7055 | } | |
7056 | ||
7057 | /* Check VLAN. */ | |
7058 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
7059 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
7060 | return; | |
7061 | } | |
7062 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7063 | ||
7064 | /* Look at the output ports to check for destination selections. */ | |
7065 | ||
7066 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
7067 | ctx->odp_actions->size) { | |
7068 | enum ovs_action_attr type = nl_attr_type(a); | |
7069 | struct ofport_dpif *ofport; | |
7070 | ||
7071 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
7072 | continue; | |
7073 | } | |
7074 | ||
7075 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
7076 | if (ofport && ofport->bundle) { |
7077 | mirrors |= ofport->bundle->dst_mirrors; | |
7078 | } | |
c06bba01 | 7079 | } |
abe529af BP |
7080 | |
7081 | if (!mirrors) { | |
7082 | return; | |
7083 | } | |
7084 | ||
c06bba01 JP |
7085 | /* Restore the original packet before adding the mirror actions. */ |
7086 | ctx->flow = *orig_flow; | |
7087 | ||
9ba15e2a BP |
7088 | while (mirrors) { |
7089 | struct ofmirror *m; | |
9ba15e2a BP |
7090 | |
7091 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7092 | ||
7093 | if (!vlan_is_mirrored(m, vlan)) { | |
8472a3ce | 7094 | mirrors = zero_rightmost_1bit(mirrors); |
9ba15e2a BP |
7095 | continue; |
7096 | } | |
7097 | ||
7098 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 7099 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 7100 | if (m->out) { |
395e68ce | 7101 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
7102 | } else if (vlan != m->out_vlan |
7103 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
7104 | struct ofbundle *bundle; |
7105 | ||
7106 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
7107 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
7108 | && !bundle->mirror_out) { |
7109 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
7110 | } |
7111 | } | |
7112 | } | |
abe529af BP |
7113 | } |
7114 | } | |
7115 | ||
9d24de3b JP |
7116 | static void |
7117 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
7118 | uint64_t packets, uint64_t bytes) | |
7119 | { | |
7120 | if (!mirrors) { | |
7121 | return; | |
7122 | } | |
7123 | ||
8472a3ce | 7124 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { |
9d24de3b JP |
7125 | struct ofmirror *m; |
7126 | ||
7127 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7128 | ||
7129 | if (!m) { | |
7130 | /* In normal circumstances 'm' will not be NULL. However, | |
7131 | * if mirrors are reconfigured, we can temporarily get out | |
7132 | * of sync in facet_revalidate(). We could "correct" the | |
7133 | * mirror list before reaching here, but doing that would | |
7134 | * not properly account the traffic stats we've currently | |
7135 | * accumulated for previous mirror configuration. */ | |
7136 | continue; | |
7137 | } | |
7138 | ||
7139 | m->packet_count += packets; | |
7140 | m->byte_count += bytes; | |
7141 | } | |
7142 | } | |
7143 | ||
abe529af BP |
7144 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
7145 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7146 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7147 | static bool | |
7148 | is_gratuitous_arp(const struct flow *flow) | |
7149 | { | |
7150 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7151 | && eth_addr_is_broadcast(flow->dl_dst) | |
7152 | && (flow->nw_proto == ARP_OP_REPLY | |
7153 | || (flow->nw_proto == ARP_OP_REQUEST | |
7154 | && flow->nw_src == flow->nw_dst))); | |
7155 | } | |
7156 | ||
7157 | static void | |
7158 | update_learning_table(struct ofproto_dpif *ofproto, | |
7159 | const struct flow *flow, int vlan, | |
7160 | struct ofbundle *in_bundle) | |
7161 | { | |
7162 | struct mac_entry *mac; | |
7163 | ||
33158a18 JP |
7164 | /* Don't learn the OFPP_NONE port. */ |
7165 | if (in_bundle == &ofpp_none_bundle) { | |
7166 | return; | |
7167 | } | |
7168 | ||
abe529af BP |
7169 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
7170 | return; | |
7171 | } | |
7172 | ||
7173 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7174 | if (is_gratuitous_arp(flow)) { | |
7175 | /* We don't want to learn from gratuitous ARP packets that are | |
7176 | * reflected back over bond slaves so we lock the learning table. */ | |
7177 | if (!in_bundle->bond) { | |
7178 | mac_entry_set_grat_arp_lock(mac); | |
7179 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7180 | return; | |
7181 | } | |
7182 | } | |
7183 | ||
7184 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7185 | /* The log messages here could actually be useful in debugging, | |
7186 | * so keep the rate limit relatively high. */ | |
7187 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7188 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7189 | "on port %s in VLAN %d", | |
7190 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7191 | in_bundle->name, vlan); | |
7192 | ||
7193 | mac->port.p = in_bundle; | |
2cc3c58e | 7194 | tag_set_add(&ofproto->backer->revalidate_set, |
abe529af BP |
7195 | mac_learning_changed(ofproto->ml, mac)); |
7196 | } | |
7197 | } | |
7198 | ||
3581c12c | 7199 | static struct ofbundle * |
4acbc98d SH |
7200 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
7201 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
7202 | { |
7203 | struct ofport_dpif *ofport; | |
7204 | ||
7205 | /* Find the port and bundle for the received packet. */ | |
7206 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
7207 | if (in_ofportp) { |
7208 | *in_ofportp = ofport; | |
7209 | } | |
395e68ce | 7210 | if (ofport && ofport->bundle) { |
3581c12c | 7211 | return ofport->bundle; |
395e68ce BP |
7212 | } |
7213 | ||
70c2fd56 BP |
7214 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
7215 | * port for traffic that it is sourcing. */ | |
7216 | if (in_port == OFPP_NONE) { | |
7217 | return &ofpp_none_bundle; | |
7218 | } | |
7219 | ||
395e68ce BP |
7220 | /* Odd. A few possible reasons here: |
7221 | * | |
7222 | * - We deleted a port but there are still a few packets queued up | |
7223 | * from it. | |
7224 | * | |
7225 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7226 | * we don't know about. | |
7227 | * | |
7228 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
7229 | * This is particularly likely to happen if a packet was received on the |
7230 | * port after it was created, but before the client had a chance to | |
7231 | * configure its bundle. | |
395e68ce BP |
7232 | */ |
7233 | if (warn) { | |
7234 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7235 | ||
7236 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7237 | "port %"PRIu16, ofproto->up.name, in_port); | |
7238 | } | |
7239 | return NULL; | |
7240 | } | |
7241 | ||
5da5ec37 | 7242 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
7243 | * dropped. Returns true if they may be forwarded, false if they should be |
7244 | * dropped. | |
7245 | * | |
395e68ce BP |
7246 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
7247 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 7248 | * |
395e68ce BP |
7249 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
7250 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7251 | * checked by input_vid_is_valid(). | |
abe529af BP |
7252 | * |
7253 | * May also add tags to '*tags', although the current implementation only does | |
7254 | * so in one special case. | |
7255 | */ | |
7256 | static bool | |
479df176 BP |
7257 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, |
7258 | uint16_t vlan) | |
abe529af | 7259 | { |
479df176 BP |
7260 | struct ofproto_dpif *ofproto = ctx->ofproto; |
7261 | struct flow *flow = &ctx->flow; | |
395e68ce | 7262 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 7263 | |
395e68ce BP |
7264 | /* Drop frames for reserved multicast addresses |
7265 | * only if forward_bpdu option is absent. */ | |
614ec445 | 7266 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 7267 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
7268 | return false; |
7269 | } | |
7270 | ||
abe529af BP |
7271 | if (in_bundle->bond) { |
7272 | struct mac_entry *mac; | |
7273 | ||
7274 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
479df176 | 7275 | flow->dl_dst, &ctx->tags)) { |
abe529af BP |
7276 | case BV_ACCEPT: |
7277 | break; | |
7278 | ||
7279 | case BV_DROP: | |
479df176 | 7280 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
7281 | return false; |
7282 | ||
7283 | case BV_DROP_IF_MOVED: | |
7284 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7285 | if (mac && mac->port.p != in_bundle && | |
7286 | (!is_gratuitous_arp(flow) | |
7287 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
7288 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
7289 | "dropping"); | |
abe529af BP |
7290 | return false; |
7291 | } | |
7292 | break; | |
7293 | } | |
7294 | } | |
7295 | ||
7296 | return true; | |
7297 | } | |
7298 | ||
4cd78906 | 7299 | static void |
abe529af BP |
7300 | xlate_normal(struct action_xlate_ctx *ctx) |
7301 | { | |
395e68ce | 7302 | struct ofport_dpif *in_port; |
abe529af | 7303 | struct ofbundle *in_bundle; |
abe529af | 7304 | struct mac_entry *mac; |
395e68ce BP |
7305 | uint16_t vlan; |
7306 | uint16_t vid; | |
abe529af | 7307 | |
75a75043 BP |
7308 | ctx->has_normal = true; |
7309 | ||
3581c12c | 7310 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
70c2fd56 | 7311 | ctx->packet != NULL, &in_port); |
3581c12c | 7312 | if (!in_bundle) { |
479df176 | 7313 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
7314 | return; |
7315 | } | |
3581c12c | 7316 | |
395e68ce BP |
7317 | /* Drop malformed frames. */ |
7318 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
7319 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
7320 | if (ctx->packet != NULL) { | |
7321 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7322 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7323 | "VLAN tag received on port %s", | |
7324 | ctx->ofproto->up.name, in_bundle->name); | |
7325 | } | |
479df176 | 7326 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
7327 | return; |
7328 | } | |
7329 | ||
7330 | /* Drop frames on bundles reserved for mirroring. */ | |
7331 | if (in_bundle->mirror_out) { | |
7332 | if (ctx->packet != NULL) { | |
7333 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7334 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7335 | "%s, which is reserved exclusively for mirroring", | |
7336 | ctx->ofproto->up.name, in_bundle->name); | |
7337 | } | |
479df176 | 7338 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
7339 | return; |
7340 | } | |
7341 | ||
7342 | /* Check VLAN. */ | |
7343 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
7344 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
479df176 | 7345 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
7346 | return; |
7347 | } | |
7348 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7349 | ||
7350 | /* Check other admissibility requirements. */ | |
479df176 | 7351 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 7352 | return; |
abe529af BP |
7353 | } |
7354 | ||
75a75043 | 7355 | /* Learn source MAC. */ |
3de9590b | 7356 | if (ctx->may_learn) { |
abe529af BP |
7357 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
7358 | } | |
7359 | ||
7360 | /* Determine output bundle. */ | |
7361 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
7362 | &ctx->tags); | |
7363 | if (mac) { | |
c06bba01 | 7364 | if (mac->port.p != in_bundle) { |
479df176 | 7365 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 7366 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
7367 | } else { |
7368 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 7369 | } |
abe529af | 7370 | } else { |
c06bba01 | 7371 | struct ofbundle *bundle; |
abe529af | 7372 | |
479df176 | 7373 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
7374 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
7375 | if (bundle != in_bundle | |
7376 | && ofbundle_includes_vlan(bundle, vlan) | |
7377 | && bundle->floodable | |
7378 | && !bundle->mirror_out) { | |
7379 | output_normal(ctx, bundle, vlan); | |
7380 | } | |
7381 | } | |
7382 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 7383 | } |
abe529af BP |
7384 | } |
7385 | \f | |
54a9cbc9 BP |
7386 | /* Optimized flow revalidation. |
7387 | * | |
7388 | * It's a difficult problem, in general, to tell which facets need to have | |
7389 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7390 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7391 | * table changes, we recalculate the actions for every facet. This is | |
7392 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7393 | * doesn't change very often. | |
7394 | * | |
7395 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7396 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7397 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7398 | * table, we add a special case that applies to flow tables in which every rule | |
7399 | * has the same form (that is, the same wildcards), except that the table is | |
7400 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7401 | * optimize this case by tagging all of the facets that resubmit into the table | |
7402 | * and invalidating the same tag whenever a flow changes in that table. The | |
7403 | * end result is that we revalidate just the facets that need it (and sometimes | |
7404 | * a few more, but not all of the facets or even all of the facets that | |
7405 | * resubmit to the table modified by MAC learning). */ | |
7406 | ||
5cb7a798 | 7407 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted |
54a9cbc9 | 7408 | * into an OpenFlow table with the given 'basis'. */ |
822d9414 | 7409 | static tag_type |
5cb7a798 | 7410 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, |
54a9cbc9 BP |
7411 | uint32_t secret) |
7412 | { | |
5cb7a798 | 7413 | if (minimask_is_catchall(mask)) { |
54a9cbc9 BP |
7414 | return 0; |
7415 | } else { | |
5cb7a798 BP |
7416 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); |
7417 | return tag_create_deterministic(hash); | |
54a9cbc9 BP |
7418 | } |
7419 | } | |
7420 | ||
7421 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7422 | * taggability of that table. | |
7423 | * | |
7424 | * This function must be called after *each* change to a flow table. If you | |
7425 | * skip calling it on some changes then the pointer comparisons at the end can | |
7426 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7427 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7428 | * different wildcards to be created with the same address, then this function | |
7429 | * will incorrectly skip revalidation. */ | |
7430 | static void | |
7431 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7432 | { | |
7433 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 7434 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
7435 | struct cls_table *catchall, *other; |
7436 | struct cls_table *t; | |
7437 | ||
7438 | catchall = other = NULL; | |
7439 | ||
d0918789 | 7440 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
7441 | case 0: |
7442 | /* We could tag this OpenFlow table but it would make the logic a | |
7443 | * little harder and it's a corner case that doesn't seem worth it | |
7444 | * yet. */ | |
7445 | break; | |
7446 | ||
7447 | case 1: | |
7448 | case 2: | |
d0918789 | 7449 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
7450 | if (cls_table_is_catchall(t)) { |
7451 | catchall = t; | |
7452 | } else if (!other) { | |
7453 | other = t; | |
7454 | } else { | |
7455 | /* Indicate that we can't tag this by setting both tables to | |
7456 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7457 | other = NULL; | |
7458 | } | |
7459 | } | |
7460 | break; | |
7461 | ||
7462 | default: | |
7463 | /* Can't tag this table. */ | |
7464 | break; | |
7465 | } | |
7466 | ||
7467 | if (table->catchall_table != catchall || table->other_table != other) { | |
7468 | table->catchall_table = catchall; | |
7469 | table->other_table = other; | |
2cc3c58e | 7470 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7471 | } |
7472 | } | |
7473 | ||
7474 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7475 | * inserted, a rule being deleted, or a rule whose actions are being | |
7476 | * modified), marks facets for revalidation to ensure that packets will be | |
7477 | * forwarded correctly according to the new state of the flow table. | |
7478 | * | |
7479 | * This function must be called after *each* change to a flow table. See | |
7480 | * the comment on table_update_taggable() for more information. */ | |
7481 | static void | |
7482 | rule_invalidate(const struct rule_dpif *rule) | |
7483 | { | |
7484 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7485 | ||
7486 | table_update_taggable(ofproto, rule->up.table_id); | |
7487 | ||
2cc3c58e | 7488 | if (!ofproto->backer->need_revalidate) { |
54a9cbc9 BP |
7489 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; |
7490 | ||
7491 | if (table->other_table && rule->tag) { | |
2cc3c58e | 7492 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); |
54a9cbc9 | 7493 | } else { |
2cc3c58e | 7494 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
7495 | } |
7496 | } | |
7497 | } | |
7498 | \f | |
abe529af | 7499 | static bool |
7257b535 BP |
7500 | set_frag_handling(struct ofproto *ofproto_, |
7501 | enum ofp_config_flags frag_handling) | |
abe529af BP |
7502 | { |
7503 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7257b535 | 7504 | if (frag_handling != OFPC_FRAG_REASM) { |
2cc3c58e | 7505 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
7506 | return true; |
7507 | } else { | |
7508 | return false; | |
7509 | } | |
abe529af BP |
7510 | } |
7511 | ||
90bf1e07 | 7512 | static enum ofperr |
abe529af BP |
7513 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
7514 | const struct flow *flow, | |
f25d0cf3 | 7515 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
7516 | { |
7517 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
14f94f9a | 7518 | struct initial_vals initial_vals; |
548de4dd BP |
7519 | struct odputil_keybuf keybuf; |
7520 | struct dpif_flow_stats stats; | |
abe529af | 7521 | |
548de4dd | 7522 | struct ofpbuf key; |
112bc5f4 | 7523 | |
548de4dd BP |
7524 | struct action_xlate_ctx ctx; |
7525 | uint64_t odp_actions_stub[1024 / 8]; | |
7526 | struct ofpbuf odp_actions; | |
80e5eed9 | 7527 | |
548de4dd | 7528 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
e1b1d06a JP |
7529 | odp_flow_key_from_flow(&key, flow, |
7530 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
050ac423 | 7531 | |
548de4dd | 7532 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); |
abe529af | 7533 | |
14f94f9a | 7534 | initial_vals.vlan_tci = flow->vlan_tci; |
c3f6c502 | 7535 | initial_vals.tunnel_ip_tos = 0; |
14f94f9a | 7536 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL, |
548de4dd BP |
7537 | packet_get_tcp_flags(packet, flow), packet); |
7538 | ctx.resubmit_stats = &stats; | |
2284188b | 7539 | |
548de4dd BP |
7540 | ofpbuf_use_stub(&odp_actions, |
7541 | odp_actions_stub, sizeof odp_actions_stub); | |
7542 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
acf60855 | 7543 | dpif_execute(ofproto->backer->dpif, key.data, key.size, |
548de4dd BP |
7544 | odp_actions.data, odp_actions.size, packet); |
7545 | ofpbuf_uninit(&odp_actions); | |
2284188b | 7546 | |
548de4dd | 7547 | return 0; |
abe529af | 7548 | } |
6fca1ffb BP |
7549 | \f |
7550 | /* NetFlow. */ | |
7551 | ||
7552 | static int | |
7553 | set_netflow(struct ofproto *ofproto_, | |
7554 | const struct netflow_options *netflow_options) | |
7555 | { | |
7556 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7557 | ||
7558 | if (netflow_options) { | |
7559 | if (!ofproto->netflow) { | |
7560 | ofproto->netflow = netflow_create(); | |
7561 | } | |
7562 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7563 | } else { | |
7564 | netflow_destroy(ofproto->netflow); | |
7565 | ofproto->netflow = NULL; | |
7566 | return 0; | |
7567 | } | |
7568 | } | |
abe529af BP |
7569 | |
7570 | static void | |
7571 | get_netflow_ids(const struct ofproto *ofproto_, | |
7572 | uint8_t *engine_type, uint8_t *engine_id) | |
7573 | { | |
7574 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7575 | ||
acf60855 | 7576 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); |
abe529af | 7577 | } |
6fca1ffb BP |
7578 | |
7579 | static void | |
7580 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7581 | { | |
7582 | if (!facet_is_controller_flow(facet) && | |
7583 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 7584 | struct subfacet *subfacet; |
6fca1ffb BP |
7585 | struct ofexpired expired; |
7586 | ||
b0f7b9b5 | 7587 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 7588 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 7589 | struct dpif_flow_stats stats; |
6fca1ffb | 7590 | |
6a7e895f | 7591 | subfacet_reinstall(subfacet, &stats); |
15baa734 | 7592 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 7593 | } |
6fca1ffb BP |
7594 | } |
7595 | ||
7596 | expired.flow = facet->flow; | |
7597 | expired.packet_count = facet->packet_count; | |
7598 | expired.byte_count = facet->byte_count; | |
7599 | expired.used = facet->used; | |
7600 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7601 | } | |
7602 | } | |
7603 | ||
7604 | static void | |
7605 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7606 | { | |
7607 | struct facet *facet; | |
7608 | ||
7609 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7610 | send_active_timeout(ofproto, facet); | |
7611 | } | |
7612 | } | |
abe529af BP |
7613 | \f |
7614 | static struct ofproto_dpif * | |
7615 | ofproto_dpif_lookup(const char *name) | |
7616 | { | |
b44a10b7 BP |
7617 | struct ofproto_dpif *ofproto; |
7618 | ||
7619 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7620 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7621 | if (!strcmp(ofproto->up.name, name)) { | |
7622 | return ofproto; | |
7623 | } | |
7624 | } | |
7625 | return NULL; | |
abe529af BP |
7626 | } |
7627 | ||
f0a3aa2e | 7628 | static void |
96e466a3 | 7629 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 7630 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 7631 | { |
490df1ef | 7632 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 7633 | |
96e466a3 EJ |
7634 | if (argc > 1) { |
7635 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7636 | if (!ofproto) { | |
bde9f75d | 7637 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
7638 | return; |
7639 | } | |
2cc3c58e | 7640 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 EJ |
7641 | } else { |
7642 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2cc3c58e | 7643 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); |
96e466a3 | 7644 | } |
f0a3aa2e | 7645 | } |
f0a3aa2e | 7646 | |
bde9f75d | 7647 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
7648 | } |
7649 | ||
abe529af | 7650 | static void |
0e15264f BP |
7651 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
7652 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
7653 | { |
7654 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7655 | const struct ofproto_dpif *ofproto; | |
7656 | const struct mac_entry *e; | |
7657 | ||
0e15264f | 7658 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 7659 | if (!ofproto) { |
bde9f75d | 7660 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
7661 | return; |
7662 | } | |
7663 | ||
7664 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
7665 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
7666 | struct ofbundle *bundle = e->port.p; | |
7667 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
7668 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
7669 | e->vlan, ETH_ADDR_ARGS(e->mac), |
7670 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 7671 | } |
bde9f75d | 7672 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
7673 | ds_destroy(&ds); |
7674 | } | |
7675 | ||
6a6455e5 | 7676 | struct trace_ctx { |
abe529af BP |
7677 | struct action_xlate_ctx ctx; |
7678 | struct flow flow; | |
7679 | struct ds *result; | |
7680 | }; | |
7681 | ||
7682 | static void | |
29901626 BP |
7683 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
7684 | const struct rule_dpif *rule) | |
abe529af BP |
7685 | { |
7686 | ds_put_char_multiple(result, '\t', level); | |
7687 | if (!rule) { | |
7688 | ds_put_cstr(result, "No match\n"); | |
7689 | return; | |
7690 | } | |
7691 | ||
29901626 BP |
7692 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
7693 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 7694 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
7695 | ds_put_char(result, '\n'); |
7696 | ||
7697 | ds_put_char_multiple(result, '\t', level); | |
7698 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 7699 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
7700 | ds_put_char(result, '\n'); |
7701 | } | |
7702 | ||
7703 | static void | |
7704 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 7705 | struct trace_ctx *trace) |
abe529af BP |
7706 | { |
7707 | ds_put_char_multiple(result, '\t', level); | |
7708 | ds_put_format(result, "%s: ", title); | |
7709 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
7710 | ds_put_cstr(result, "unchanged"); | |
7711 | } else { | |
7712 | flow_format(result, &trace->ctx.flow); | |
7713 | trace->flow = trace->ctx.flow; | |
7714 | } | |
7715 | ds_put_char(result, '\n'); | |
7716 | } | |
7717 | ||
eb9e1c26 EJ |
7718 | static void |
7719 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 7720 | struct trace_ctx *trace) |
eb9e1c26 EJ |
7721 | { |
7722 | size_t i; | |
7723 | ||
7724 | ds_put_char_multiple(result, '\t', level); | |
7725 | ds_put_format(result, "%s:", title); | |
7726 | for (i = 0; i < FLOW_N_REGS; i++) { | |
7727 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
7728 | } | |
7729 | ds_put_char(result, '\n'); | |
7730 | } | |
7731 | ||
1ed8d352 EJ |
7732 | static void |
7733 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 7734 | struct trace_ctx *trace) |
1ed8d352 EJ |
7735 | { |
7736 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
7737 | ||
7738 | ds_put_char_multiple(result, '\t', level); | |
7739 | ds_put_format(result, "%s: ", title); | |
7740 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
7741 | ds_put_char(result, '\n'); | |
7742 | } | |
7743 | ||
abe529af BP |
7744 | static void |
7745 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
7746 | { | |
6a6455e5 | 7747 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
7748 | struct ds *result = trace->result; |
7749 | ||
7750 | ds_put_char(result, '\n'); | |
7751 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 7752 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 7753 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 7754 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
7755 | } |
7756 | ||
479df176 BP |
7757 | static void |
7758 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
7759 | { | |
7760 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
7761 | struct ds *result = trace->result; | |
7762 | ||
7763 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
7764 | ds_put_cstr(result, s); | |
7765 | ds_put_char(result, '\n'); | |
7766 | } | |
7767 | ||
abe529af | 7768 | static void |
0e15264f | 7769 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
7770 | void *aux OVS_UNUSED) |
7771 | { | |
0e15264f | 7772 | const char *dpname = argv[1]; |
abe529af | 7773 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
7774 | struct ofpbuf odp_key; |
7775 | struct ofpbuf *packet; | |
14f94f9a | 7776 | struct initial_vals initial_vals; |
abe529af BP |
7777 | struct ds result; |
7778 | struct flow flow; | |
abe529af BP |
7779 | char *s; |
7780 | ||
876b0e1c BP |
7781 | packet = NULL; |
7782 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
7783 | ds_init(&result); |
7784 | ||
e84173dc BP |
7785 | ofproto = ofproto_dpif_lookup(dpname); |
7786 | if (!ofproto) { | |
bde9f75d EJ |
7787 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
7788 | "for help)"); | |
e84173dc BP |
7789 | goto exit; |
7790 | } | |
0e15264f | 7791 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 7792 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
7793 | const char *flow_s = argv[2]; |
7794 | const char *generate_s = argv[3]; | |
876b0e1c | 7795 | |
31a19d69 BP |
7796 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like |
7797 | * flow. We guess which type it is based on whether 'flow_s' contains | |
7798 | * an '(', since a datapath flow always contains '(') but an | |
7799 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
7800 | * that's not documented anywhere). | |
7801 | * | |
7802 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
7803 | * it would be tricky giving a sensible error message. After all, do | |
7804 | * you just say "syntax error" or do you present both error messages? | |
7805 | * Both choices seem lousy. */ | |
7806 | if (strchr(flow_s, '(')) { | |
7807 | int error; | |
7808 | ||
7809 | /* Convert string to datapath key. */ | |
7810 | ofpbuf_init(&odp_key, 0); | |
7811 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
7812 | if (error) { | |
7813 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
7814 | goto exit; | |
7815 | } | |
876b0e1c | 7816 | |
6d199116 BP |
7817 | /* The user might have specified the wrong ofproto but within the |
7818 | * same backer. That's OK, ofproto_receive() can find the right | |
7819 | * one for us. */ | |
e09ee259 | 7820 | if (ofproto_receive(ofproto->backer, NULL, odp_key.data, |
6d199116 | 7821 | odp_key.size, &flow, NULL, &ofproto, NULL, |
14f94f9a | 7822 | &initial_vals)) { |
31a19d69 BP |
7823 | unixctl_command_reply_error(conn, "Invalid flow"); |
7824 | goto exit; | |
7825 | } | |
6d199116 | 7826 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); |
31a19d69 BP |
7827 | } else { |
7828 | char *error_s; | |
7829 | ||
7830 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
7831 | if (error_s) { | |
7832 | unixctl_command_reply_error(conn, error_s); | |
7833 | free(error_s); | |
7834 | goto exit; | |
7835 | } | |
7836 | ||
14f94f9a | 7837 | initial_vals.vlan_tci = flow.vlan_tci; |
c3f6c502 | 7838 | initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos; |
876b0e1c | 7839 | } |
8b3b8dd1 BP |
7840 | |
7841 | /* Generate a packet, if requested. */ | |
0e15264f | 7842 | if (generate_s) { |
8b3b8dd1 BP |
7843 | packet = ofpbuf_new(0); |
7844 | flow_compose(packet, &flow); | |
7845 | } | |
72e8bf28 AA |
7846 | } else if (argc == 7) { |
7847 | /* ofproto/trace dpname priority tun_id in_port mark packet */ | |
0e15264f BP |
7848 | const char *priority_s = argv[2]; |
7849 | const char *tun_id_s = argv[3]; | |
7850 | const char *in_port_s = argv[4]; | |
72e8bf28 AA |
7851 | const char *mark_s = argv[5]; |
7852 | const char *packet_s = argv[6]; | |
9b56fe13 | 7853 | uint32_t in_port = atoi(in_port_s); |
0e15264f BP |
7854 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); |
7855 | uint32_t priority = atoi(priority_s); | |
72e8bf28 | 7856 | uint32_t mark = atoi(mark_s); |
e22f1753 | 7857 | const char *msg; |
0e15264f | 7858 | |
e22f1753 BP |
7859 | msg = eth_from_hex(packet_s, &packet); |
7860 | if (msg) { | |
bde9f75d | 7861 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
7862 | goto exit; |
7863 | } | |
7864 | ||
7865 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 7866 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
7867 | ds_put_cstr(&result, s); |
7868 | free(s); | |
7869 | ||
72e8bf28 | 7870 | flow_extract(packet, priority, mark, NULL, in_port, &flow); |
296e07ac | 7871 | flow.tunnel.tun_id = tun_id; |
14f94f9a | 7872 | initial_vals.vlan_tci = flow.vlan_tci; |
c3f6c502 | 7873 | initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos; |
876b0e1c | 7874 | } else { |
bde9f75d | 7875 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
7876 | goto exit; |
7877 | } | |
7878 | ||
14f94f9a | 7879 | ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); |
6a6455e5 EJ |
7880 | unixctl_command_reply(conn, ds_cstr(&result)); |
7881 | ||
7882 | exit: | |
7883 | ds_destroy(&result); | |
7884 | ofpbuf_delete(packet); | |
7885 | ofpbuf_uninit(&odp_key); | |
7886 | } | |
7887 | ||
7888 | static void | |
7889 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
14f94f9a JP |
7890 | const struct ofpbuf *packet, |
7891 | const struct initial_vals *initial_vals, struct ds *ds) | |
6a6455e5 EJ |
7892 | { |
7893 | struct rule_dpif *rule; | |
7894 | ||
7895 | ds_put_cstr(ds, "Flow: "); | |
7896 | flow_format(ds, flow); | |
7897 | ds_put_char(ds, '\n'); | |
abe529af | 7898 | |
c57b2226 BP |
7899 | rule = rule_dpif_lookup(ofproto, flow); |
7900 | ||
6a6455e5 | 7901 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
7902 | if (rule == ofproto->miss_rule) { |
7903 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
7904 | } else if (rule == ofproto->no_packet_in_rule) { | |
7905 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
7906 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
7907 | } | |
7908 | ||
abe529af | 7909 | if (rule) { |
050ac423 BP |
7910 | uint64_t odp_actions_stub[1024 / 8]; |
7911 | struct ofpbuf odp_actions; | |
7912 | ||
6a6455e5 | 7913 | struct trace_ctx trace; |
0e553d9c | 7914 | uint8_t tcp_flags; |
abe529af | 7915 | |
6a6455e5 EJ |
7916 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
7917 | trace.result = ds; | |
7918 | trace.flow = *flow; | |
050ac423 BP |
7919 | ofpbuf_use_stub(&odp_actions, |
7920 | odp_actions_stub, sizeof odp_actions_stub); | |
14f94f9a | 7921 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals, |
0e553d9c | 7922 | rule, tcp_flags, packet); |
abe529af | 7923 | trace.ctx.resubmit_hook = trace_resubmit; |
479df176 | 7924 | trace.ctx.report_hook = trace_report; |
f25d0cf3 | 7925 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 7926 | &odp_actions); |
abe529af | 7927 | |
6a6455e5 EJ |
7928 | ds_put_char(ds, '\n'); |
7929 | trace_format_flow(ds, 0, "Final flow", &trace); | |
7930 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
7931 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
7932 | ofpbuf_uninit(&odp_actions); | |
876b0e1c | 7933 | |
6a7e895f BP |
7934 | if (trace.ctx.slow) { |
7935 | enum slow_path_reason slow; | |
7936 | ||
7937 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
7938 | "slow path because it:"); | |
7939 | for (slow = trace.ctx.slow; slow; ) { | |
7940 | enum slow_path_reason bit = rightmost_1bit(slow); | |
7941 | ||
7942 | switch (bit) { | |
7943 | case SLOW_CFM: | |
7944 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
7945 | break; | |
7946 | case SLOW_LACP: | |
7947 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
7948 | break; | |
7949 | case SLOW_STP: | |
7950 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
7951 | break; | |
7952 | case SLOW_IN_BAND: | |
7953 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
7954 | "processing."); | |
7955 | if (!packet) { | |
7956 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
7957 | "incomplete--for complete actions, " | |
7958 | "please supply a packet.)"); | |
7959 | } | |
7960 | break; | |
7961 | case SLOW_CONTROLLER: | |
7962 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
7963 | "to the OpenFlow controller."); | |
7964 | break; | |
7965 | case SLOW_MATCH: | |
7966 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
7967 | "than the datapath supports."); | |
7968 | break; | |
7969 | } | |
7970 | ||
7971 | slow &= ~bit; | |
7972 | } | |
7973 | ||
7974 | if (slow & ~SLOW_MATCH) { | |
7975 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
7976 | "the special slow-path processing."); | |
876b0e1c BP |
7977 | } |
7978 | } | |
abe529af | 7979 | } |
abe529af BP |
7980 | } |
7981 | ||
7ee20df1 | 7982 | static void |
0e15264f BP |
7983 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
7984 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
7985 | { |
7986 | clogged = true; | |
bde9f75d | 7987 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
7988 | } |
7989 | ||
7990 | static void | |
0e15264f BP |
7991 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
7992 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
7993 | { |
7994 | clogged = false; | |
bde9f75d | 7995 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
7996 | } |
7997 | ||
6814e51f BP |
7998 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
7999 | * 'reply' describing the results. */ | |
8000 | static void | |
8001 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
8002 | { | |
8003 | struct facet *facet; | |
8004 | int errors; | |
8005 | ||
8006 | errors = 0; | |
8007 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
8008 | if (!facet_check_consistency(facet)) { | |
8009 | errors++; | |
8010 | } | |
8011 | } | |
8012 | if (errors) { | |
2cc3c58e | 8013 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
8014 | } |
8015 | ||
8016 | if (errors) { | |
8017 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
8018 | ofproto->up.name, errors); | |
8019 | } else { | |
8020 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
8021 | } | |
8022 | } | |
8023 | ||
8024 | static void | |
8025 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
8026 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
8027 | { | |
8028 | struct ds reply = DS_EMPTY_INITIALIZER; | |
8029 | struct ofproto_dpif *ofproto; | |
8030 | ||
8031 | if (argc > 1) { | |
8032 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8033 | if (!ofproto) { | |
bde9f75d EJ |
8034 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
8035 | "ofproto/list for help)"); | |
6814e51f BP |
8036 | return; |
8037 | } | |
8038 | ofproto_dpif_self_check__(ofproto, &reply); | |
8039 | } else { | |
8040 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8041 | ofproto_dpif_self_check__(ofproto, &reply); | |
8042 | } | |
8043 | } | |
8044 | ||
bde9f75d | 8045 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
8046 | ds_destroy(&reply); |
8047 | } | |
8048 | ||
27022416 JP |
8049 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list |
8050 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
8051 | * to destroy 'ofproto_shash' and free the returned value. */ | |
8052 | static const struct shash_node ** | |
8053 | get_ofprotos(struct shash *ofproto_shash) | |
8054 | { | |
8055 | const struct ofproto_dpif *ofproto; | |
8056 | ||
8057 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8058 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
8059 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
8060 | } | |
8061 | ||
8062 | return shash_sort(ofproto_shash); | |
8063 | } | |
8064 | ||
8065 | static void | |
8066 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
8067 | const char *argv[] OVS_UNUSED, | |
8068 | void *aux OVS_UNUSED) | |
8069 | { | |
8070 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8071 | struct shash ofproto_shash; | |
8072 | const struct shash_node **sorted_ofprotos; | |
8073 | int i; | |
8074 | ||
8075 | shash_init(&ofproto_shash); | |
8076 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8077 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8078 | const struct shash_node *node = sorted_ofprotos[i]; | |
8079 | ds_put_format(&ds, "%s\n", node->name); | |
8080 | } | |
8081 | ||
8082 | shash_destroy(&ofproto_shash); | |
8083 | free(sorted_ofprotos); | |
8084 | ||
8085 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8086 | ds_destroy(&ds); | |
8087 | } | |
8088 | ||
8089 | static void | |
8090 | show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) | |
8091 | { | |
27022416 JP |
8092 | const struct shash_node **ports; |
8093 | int i; | |
8094 | ||
acf60855 JP |
8095 | ds_put_format(ds, "%s (%s):\n", ofproto->up.name, |
8096 | dpif_name(ofproto->backer->dpif)); | |
27022416 | 8097 | ds_put_format(ds, |
735d7efb AZ |
8098 | "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n", |
8099 | ofproto->n_hit, ofproto->n_missed); | |
acf60855 JP |
8100 | ds_put_format(ds, "\tflows: %zu\n", |
8101 | hmap_count(&ofproto->subfacets)); | |
27022416 JP |
8102 | |
8103 | ports = shash_sort(&ofproto->up.port_by_name); | |
8104 | for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { | |
8105 | const struct shash_node *node = ports[i]; | |
8106 | struct ofport *ofport = node->data; | |
8107 | const char *name = netdev_get_name(ofport->netdev); | |
8108 | const char *type = netdev_get_type(ofport->netdev); | |
0a740f48 EJ |
8109 | uint32_t odp_port; |
8110 | ||
8111 | ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); | |
8112 | ||
8113 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); | |
8114 | if (odp_port != OVSP_NONE) { | |
8115 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
8116 | } else { | |
8117 | ds_put_cstr(ds, "none:"); | |
8118 | } | |
27022416 | 8119 | |
27022416 JP |
8120 | if (strcmp(type, "system")) { |
8121 | struct netdev *netdev; | |
8122 | int error; | |
8123 | ||
8124 | ds_put_format(ds, " (%s", type); | |
8125 | ||
8126 | error = netdev_open(name, type, &netdev); | |
8127 | if (!error) { | |
8128 | struct smap config; | |
8129 | ||
8130 | smap_init(&config); | |
8131 | error = netdev_get_config(netdev, &config); | |
8132 | if (!error) { | |
8133 | const struct smap_node **nodes; | |
8134 | size_t i; | |
8135 | ||
8136 | nodes = smap_sort(&config); | |
8137 | for (i = 0; i < smap_count(&config); i++) { | |
8138 | const struct smap_node *node = nodes[i]; | |
8139 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
8140 | node->key, node->value); | |
8141 | } | |
8142 | free(nodes); | |
8143 | } | |
8144 | smap_destroy(&config); | |
8145 | ||
8146 | netdev_close(netdev); | |
8147 | } | |
8148 | ds_put_char(ds, ')'); | |
8149 | } | |
8150 | ds_put_char(ds, '\n'); | |
8151 | } | |
8152 | free(ports); | |
8153 | } | |
8154 | ||
8155 | static void | |
8156 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, | |
8157 | const char *argv[], void *aux OVS_UNUSED) | |
8158 | { | |
8159 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8160 | const struct ofproto_dpif *ofproto; | |
8161 | ||
8162 | if (argc > 1) { | |
8163 | int i; | |
8164 | for (i = 1; i < argc; i++) { | |
8165 | ofproto = ofproto_dpif_lookup(argv[i]); | |
8166 | if (!ofproto) { | |
8167 | ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " | |
8168 | "for help)", argv[i]); | |
8169 | unixctl_command_reply_error(conn, ds_cstr(&ds)); | |
8170 | return; | |
8171 | } | |
8172 | show_dp_format(ofproto, &ds); | |
8173 | } | |
8174 | } else { | |
8175 | struct shash ofproto_shash; | |
8176 | const struct shash_node **sorted_ofprotos; | |
8177 | int i; | |
8178 | ||
8179 | shash_init(&ofproto_shash); | |
8180 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8181 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8182 | const struct shash_node *node = sorted_ofprotos[i]; | |
8183 | show_dp_format(node->data, &ds); | |
8184 | } | |
8185 | ||
8186 | shash_destroy(&ofproto_shash); | |
8187 | free(sorted_ofprotos); | |
8188 | } | |
8189 | ||
8190 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8191 | ds_destroy(&ds); | |
8192 | } | |
8193 | ||
8194 | static void | |
8195 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8196 | int argc OVS_UNUSED, const char *argv[], | |
8197 | void *aux OVS_UNUSED) | |
8198 | { | |
8199 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8200 | const struct ofproto_dpif *ofproto; | |
8201 | struct subfacet *subfacet; | |
8202 | ||
8203 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8204 | if (!ofproto) { | |
8205 | unixctl_command_reply_error(conn, "no such bridge"); | |
8206 | return; | |
8207 | } | |
8208 | ||
af37354d EJ |
8209 | update_stats(ofproto->backer); |
8210 | ||
27022416 | 8211 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
9566abf9 | 8212 | odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); |
27022416 JP |
8213 | |
8214 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8215 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8216 | if (subfacet->used) { | |
8217 | ds_put_format(&ds, "%.3fs", | |
8218 | (time_msec() - subfacet->used) / 1000.0); | |
8219 | } else { | |
8220 | ds_put_format(&ds, "never"); | |
8221 | } | |
8222 | if (subfacet->facet->tcp_flags) { | |
8223 | ds_put_cstr(&ds, ", flags:"); | |
8224 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8225 | } | |
8226 | ||
8227 | ds_put_cstr(&ds, ", actions:"); | |
f2245da3 JP |
8228 | if (subfacet->slow) { |
8229 | uint64_t slow_path_stub[128 / 8]; | |
8230 | const struct nlattr *actions; | |
8231 | size_t actions_len; | |
8232 | ||
8233 | compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow, | |
8234 | slow_path_stub, sizeof slow_path_stub, | |
8235 | &actions, &actions_len); | |
8236 | format_odp_actions(&ds, actions, actions_len); | |
8237 | } else { | |
8238 | format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); | |
8239 | } | |
27022416 JP |
8240 | ds_put_char(&ds, '\n'); |
8241 | } | |
8242 | ||
8243 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8244 | ds_destroy(&ds); | |
8245 | } | |
8246 | ||
8247 | static void | |
8248 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8249 | int argc OVS_UNUSED, const char *argv[], | |
8250 | void *aux OVS_UNUSED) | |
8251 | { | |
8252 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8253 | struct ofproto_dpif *ofproto; | |
8254 | ||
8255 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8256 | if (!ofproto) { | |
8257 | unixctl_command_reply_error(conn, "no such bridge"); | |
8258 | return; | |
8259 | } | |
8260 | ||
8261 | flush(&ofproto->up); | |
8262 | ||
8263 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8264 | ds_destroy(&ds); | |
8265 | } | |
8266 | ||
abe529af BP |
8267 | static void |
8268 | ofproto_dpif_unixctl_init(void) | |
8269 | { | |
8270 | static bool registered; | |
8271 | if (registered) { | |
8272 | return; | |
8273 | } | |
8274 | registered = true; | |
8275 | ||
0e15264f BP |
8276 | unixctl_command_register( |
8277 | "ofproto/trace", | |
72e8bf28 AA |
8278 | "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", |
8279 | 2, 6, ofproto_unixctl_trace, NULL); | |
96e466a3 | 8280 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
8281 | ofproto_unixctl_fdb_flush, NULL); |
8282 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8283 | ofproto_unixctl_fdb_show, NULL); | |
8284 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8285 | ofproto_dpif_clog, NULL); | |
8286 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8287 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
8288 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
8289 | ofproto_dpif_self_check, NULL); | |
27022416 JP |
8290 | unixctl_command_register("dpif/dump-dps", "", 0, 0, |
8291 | ofproto_unixctl_dpif_dump_dps, NULL); | |
8292 | unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, | |
8293 | ofproto_unixctl_dpif_show, NULL); | |
8294 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, | |
8295 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8296 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8297 | ofproto_unixctl_dpif_del_flows, NULL); | |
abe529af BP |
8298 | } |
8299 | \f | |
52a90c29 BP |
8300 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
8301 | * | |
8302 | * This is deprecated. It is only for compatibility with broken device drivers | |
8303 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8304 | * devices are not used. When broken device drivers are no longer in | |
8305 | * widespread use, we will delete these interfaces. */ | |
8306 | ||
8307 | static int | |
8308 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8309 | { | |
8310 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8311 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8312 | ||
8313 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8314 | && vid == ofport->vlandev_vid) { | |
8315 | return 0; | |
8316 | } | |
8317 | ||
2cc3c58e | 8318 | ofproto->backer->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
8319 | |
8320 | if (ofport->realdev_ofp_port) { | |
8321 | vsp_remove(ofport); | |
8322 | } | |
8323 | if (realdev_ofp_port && ofport->bundle) { | |
8324 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8325 | * themselves be part of a bundle. */ | |
8326 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8327 | } | |
8328 | ||
8329 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8330 | ofport->vlandev_vid = vid; | |
8331 | ||
8332 | if (realdev_ofp_port) { | |
8333 | vsp_add(ofport, realdev_ofp_port, vid); | |
8334 | } | |
8335 | ||
8336 | return 0; | |
8337 | } | |
8338 | ||
8339 | static uint32_t | |
8340 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8341 | { | |
8342 | return hash_2words(realdev_ofp_port, vid); | |
8343 | } | |
8344 | ||
40e05935 BP |
8345 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
8346 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
8347 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
8348 | * it would return the port number of eth0.9. | |
8349 | * | |
8350 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
8351 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
8352 | static uint32_t |
8353 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
8354 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
8355 | { | |
8356 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
e1b1d06a | 8357 | uint16_t realdev_ofp_port; |
52a90c29 BP |
8358 | int vid = vlan_tci_to_vid(vlan_tci); |
8359 | const struct vlan_splinter *vsp; | |
8360 | ||
e1b1d06a | 8361 | realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); |
52a90c29 BP |
8362 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, |
8363 | hash_realdev_vid(realdev_ofp_port, vid), | |
8364 | &ofproto->realdev_vid_map) { | |
8365 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8366 | && vsp->vid == vid) { | |
e1b1d06a | 8367 | return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); |
52a90c29 BP |
8368 | } |
8369 | } | |
8370 | } | |
8371 | return realdev_odp_port; | |
8372 | } | |
8373 | ||
8374 | static struct vlan_splinter * | |
8375 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8376 | { | |
8377 | struct vlan_splinter *vsp; | |
8378 | ||
8379 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8380 | &ofproto->vlandev_map) { | |
8381 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8382 | return vsp; | |
8383 | } | |
8384 | } | |
8385 | ||
8386 | return NULL; | |
8387 | } | |
8388 | ||
40e05935 BP |
8389 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
8390 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8391 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8392 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8393 | * eth0 and store 9 in '*vid'. | |
8394 | * | |
8395 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8396 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8397 | * always does.*/ | |
52a90c29 BP |
8398 | static uint16_t |
8399 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 8400 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
8401 | { |
8402 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8403 | const struct vlan_splinter *vsp; | |
8404 | ||
8405 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8406 | if (vsp) { | |
8407 | if (vid) { | |
8408 | *vid = vsp->vid; | |
8409 | } | |
8410 | return vsp->realdev_ofp_port; | |
8411 | } | |
8412 | } | |
8413 | return 0; | |
8414 | } | |
8415 | ||
b98d8985 BP |
8416 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
8417 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8418 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8419 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8420 | * always the case unless VLAN splinters are enabled), returns false without | |
8421 | * making any changes. */ | |
8422 | static bool | |
8423 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8424 | { | |
8425 | uint16_t realdev; | |
8426 | int vid; | |
8427 | ||
8428 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8429 | if (!realdev) { | |
8430 | return false; | |
8431 | } | |
8432 | ||
8433 | /* Cause the flow to be processed as if it came in on the real device with | |
8434 | * the VLAN device's VLAN ID. */ | |
8435 | flow->in_port = realdev; | |
8436 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8437 | return true; | |
8438 | } | |
8439 | ||
52a90c29 BP |
8440 | static void |
8441 | vsp_remove(struct ofport_dpif *port) | |
8442 | { | |
8443 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8444 | struct vlan_splinter *vsp; | |
8445 | ||
8446 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8447 | if (vsp) { | |
8448 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8449 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8450 | free(vsp); | |
8451 | ||
8452 | port->realdev_ofp_port = 0; | |
8453 | } else { | |
8454 | VLOG_ERR("missing vlan device record"); | |
8455 | } | |
8456 | } | |
8457 | ||
8458 | static void | |
8459 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8460 | { | |
8461 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8462 | ||
8463 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8464 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8465 | == realdev_ofp_port)) { | |
8466 | struct vlan_splinter *vsp; | |
8467 | ||
8468 | vsp = xmalloc(sizeof *vsp); | |
8469 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8470 | hash_int(port->up.ofp_port, 0)); | |
8471 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8472 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8473 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8474 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8475 | vsp->vid = vid; | |
8476 | ||
8477 | port->realdev_ofp_port = realdev_ofp_port; | |
8478 | } else { | |
8479 | VLOG_ERR("duplicate vlan device record"); | |
8480 | } | |
8481 | } | |
e1b1d06a JP |
8482 | |
8483 | static uint32_t | |
8484 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8485 | { | |
8486 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8487 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8488 | } | |
8489 | ||
acf60855 JP |
8490 | static struct ofport_dpif * |
8491 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
e1b1d06a JP |
8492 | { |
8493 | struct ofport_dpif *port; | |
8494 | ||
8495 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8496 | hash_int(odp_port, 0), | |
acf60855 | 8497 | &backer->odp_to_ofport_map) { |
e1b1d06a | 8498 | if (port->odp_port == odp_port) { |
acf60855 | 8499 | return port; |
e1b1d06a JP |
8500 | } |
8501 | } | |
8502 | ||
acf60855 JP |
8503 | return NULL; |
8504 | } | |
8505 | ||
8506 | static uint16_t | |
8507 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8508 | { | |
8509 | struct ofport_dpif *port; | |
8510 | ||
8511 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
6472ba11 | 8512 | if (port && &ofproto->up == port->up.ofproto) { |
acf60855 JP |
8513 | return port->up.ofp_port; |
8514 | } else { | |
8515 | return OFPP_NONE; | |
8516 | } | |
e1b1d06a JP |
8517 | } |
8518 | ||
735d7efb AZ |
8519 | static void |
8520 | dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta) | |
8521 | { | |
8522 | ofproto->n_hit += delta; | |
8523 | } | |
8524 | ||
abe529af | 8525 | const struct ofproto_class ofproto_dpif_class = { |
b0408fca | 8526 | init, |
abe529af BP |
8527 | enumerate_types, |
8528 | enumerate_names, | |
8529 | del, | |
0aeaabc8 | 8530 | port_open_type, |
acf60855 JP |
8531 | type_run, |
8532 | type_run_fast, | |
8533 | type_wait, | |
abe529af BP |
8534 | alloc, |
8535 | construct, | |
8536 | destruct, | |
8537 | dealloc, | |
8538 | run, | |
5fcc0d00 | 8539 | run_fast, |
abe529af | 8540 | wait, |
0d085684 | 8541 | get_memory_usage, |
abe529af | 8542 | flush, |
6c1491fb BP |
8543 | get_features, |
8544 | get_tables, | |
abe529af BP |
8545 | port_alloc, |
8546 | port_construct, | |
8547 | port_destruct, | |
8548 | port_dealloc, | |
8549 | port_modified, | |
8550 | port_reconfigured, | |
8551 | port_query_by_name, | |
8552 | port_add, | |
8553 | port_del, | |
6527c598 | 8554 | port_get_stats, |
abe529af BP |
8555 | port_dump_start, |
8556 | port_dump_next, | |
8557 | port_dump_done, | |
8558 | port_poll, | |
8559 | port_poll_wait, | |
8560 | port_is_lacp_current, | |
0ab6decf | 8561 | NULL, /* rule_choose_table */ |
abe529af BP |
8562 | rule_alloc, |
8563 | rule_construct, | |
8564 | rule_destruct, | |
8565 | rule_dealloc, | |
abe529af BP |
8566 | rule_get_stats, |
8567 | rule_execute, | |
8568 | rule_modify_actions, | |
7257b535 | 8569 | set_frag_handling, |
abe529af BP |
8570 | packet_out, |
8571 | set_netflow, | |
8572 | get_netflow_ids, | |
8573 | set_sflow, | |
8574 | set_cfm, | |
9a9e3786 | 8575 | get_cfm_status, |
21f7563c JP |
8576 | set_stp, |
8577 | get_stp_status, | |
8578 | set_stp_port, | |
8579 | get_stp_port_status, | |
8b36f51e | 8580 | set_queues, |
abe529af BP |
8581 | bundle_set, |
8582 | bundle_remove, | |
8583 | mirror_set, | |
9d24de3b | 8584 | mirror_get_stats, |
abe529af BP |
8585 | set_flood_vlans, |
8586 | is_mirror_output_bundle, | |
8402c74b | 8587 | forward_bpdu_changed, |
c4069512 | 8588 | set_mac_table_config, |
52a90c29 | 8589 | set_realdev, |
abe529af | 8590 | }; |