]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
e0edde6f | 2 | * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
23 | #include "autopath.h" | |
24 | #include "bond.h" | |
daff3353 | 25 | #include "bundle.h" |
abe529af BP |
26 | #include "byte-order.h" |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
75a75043 | 35 | #include "learn.h" |
abe529af | 36 | #include "mac-learning.h" |
816fd533 | 37 | #include "meta-flow.h" |
abe529af BP |
38 | #include "multipath.h" |
39 | #include "netdev.h" | |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
f25d0cf3 | 45 | #include "ofp-actions.h" |
31a19d69 | 46 | #include "ofp-parse.h" |
abe529af | 47 | #include "ofp-print.h" |
9d6ac44e | 48 | #include "ofproto-dpif-governor.h" |
bae473fe | 49 | #include "ofproto-dpif-sflow.h" |
abe529af | 50 | #include "poll-loop.h" |
0d085684 | 51 | #include "simap.h" |
abe529af | 52 | #include "timer.h" |
6c1491fb | 53 | #include "unaligned.h" |
abe529af BP |
54 | #include "unixctl.h" |
55 | #include "vlan-bitmap.h" | |
56 | #include "vlog.h" | |
57 | ||
58 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
59 | ||
abe529af | 60 | COVERAGE_DEFINE(ofproto_dpif_expired); |
abe529af BP |
61 | COVERAGE_DEFINE(ofproto_dpif_xlate); |
62 | COVERAGE_DEFINE(facet_changed_rule); | |
abe529af BP |
63 | COVERAGE_DEFINE(facet_revalidate); |
64 | COVERAGE_DEFINE(facet_unexpected); | |
9d6ac44e | 65 | COVERAGE_DEFINE(facet_suppress); |
abe529af | 66 | |
29901626 | 67 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 68 | * flow translation. */ |
642a5c05 | 69 | #define MAX_RESUBMIT_RECURSION 32 |
abe529af | 70 | |
9cdaaebe BP |
71 | /* Number of implemented OpenFlow tables. */ |
72 | enum { N_TABLES = 255 }; | |
c57b2226 BP |
73 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ |
74 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
9cdaaebe | 75 | |
abe529af BP |
76 | struct ofport_dpif; |
77 | struct ofproto_dpif; | |
78 | ||
79 | struct rule_dpif { | |
80 | struct rule up; | |
81 | ||
abe529af BP |
82 | /* These statistics: |
83 | * | |
84 | * - Do include packets and bytes from facets that have been deleted or | |
85 | * whose own statistics have been folded into the rule. | |
86 | * | |
87 | * - Do include packets and bytes sent "by hand" that were accounted to | |
88 | * the rule without any facet being involved (this is a rare corner | |
89 | * case in rule_execute()). | |
90 | * | |
91 | * - Do not include packet or bytes that can be obtained from any facet's | |
92 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 93 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
94 | */ |
95 | uint64_t packet_count; /* Number of packets received. */ | |
96 | uint64_t byte_count; /* Number of bytes received. */ | |
97 | ||
54a9cbc9 BP |
98 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
99 | ||
abe529af BP |
100 | struct list facets; /* List of "struct facet"s. */ |
101 | }; | |
102 | ||
103 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
104 | { | |
105 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
106 | } | |
107 | ||
29901626 | 108 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
c57b2226 BP |
109 | const struct flow *); |
110 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
111 | const struct flow *, | |
112 | uint8_t table); | |
abe529af | 113 | |
112bc5f4 BP |
114 | static void rule_credit_stats(struct rule_dpif *, |
115 | const struct dpif_flow_stats *); | |
18b2a258 | 116 | static void flow_push_stats(struct rule_dpif *, const struct flow *, |
112bc5f4 | 117 | const struct dpif_flow_stats *); |
822d9414 | 118 | static tag_type rule_calculate_tag(const struct flow *, |
b0f7b9b5 BP |
119 | const struct flow_wildcards *, |
120 | uint32_t basis); | |
121 | static void rule_invalidate(const struct rule_dpif *); | |
122 | ||
abe529af BP |
123 | #define MAX_MIRRORS 32 |
124 | typedef uint32_t mirror_mask_t; | |
125 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
126 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
127 | struct ofmirror { | |
128 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
129 | size_t idx; /* In ofproto's "mirrors" array. */ | |
130 | void *aux; /* Key supplied by ofproto's client. */ | |
131 | char *name; /* Identifier for log messages. */ | |
132 | ||
133 | /* Selection criteria. */ | |
134 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
135 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
136 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
137 | ||
9ba15e2a | 138 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
139 | struct ofbundle *out; /* Output port or NULL. */ |
140 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 141 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
142 | |
143 | /* Counters. */ | |
144 | int64_t packet_count; /* Number of packets sent. */ | |
145 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
146 | }; |
147 | ||
148 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
149 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
150 | mirror_mask_t mirrors, | |
151 | uint64_t packets, uint64_t bytes); | |
abe529af | 152 | |
abe529af | 153 | struct ofbundle { |
abe529af | 154 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 155 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
156 | void *aux; /* Key supplied by ofproto's client. */ |
157 | char *name; /* Identifier for log messages. */ | |
158 | ||
159 | /* Configuration. */ | |
160 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 161 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
162 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
163 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
164 | * NULL if all VLANs are trunked. */ | |
165 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
166 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 167 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
168 | |
169 | /* Status. */ | |
9e1fd49b | 170 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
171 | |
172 | /* Port mirroring info. */ | |
173 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
174 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
175 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
176 | }; | |
177 | ||
178 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 179 | static void bundle_update(struct ofbundle *); |
abe529af BP |
180 | static void bundle_destroy(struct ofbundle *); |
181 | static void bundle_del_port(struct ofport_dpif *); | |
182 | static void bundle_run(struct ofbundle *); | |
183 | static void bundle_wait(struct ofbundle *); | |
4acbc98d | 184 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, |
70c2fd56 BP |
185 | uint16_t in_port, bool warn, |
186 | struct ofport_dpif **in_ofportp); | |
abe529af | 187 | |
33158a18 JP |
188 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
189 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
190 | * when an input bundle is needed for validation (e.g., mirroring or | |
191 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
192 | * any 'port' structs, so care must be taken when dealing with it. */ | |
193 | static struct ofbundle ofpp_none_bundle = { | |
194 | .name = "OFPP_NONE", | |
195 | .vlan_mode = PORT_VLAN_TRUNK | |
196 | }; | |
197 | ||
21f7563c JP |
198 | static void stp_run(struct ofproto_dpif *ofproto); |
199 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
200 | static int set_stp_port(struct ofport *, |
201 | const struct ofproto_port_stp_settings *); | |
21f7563c | 202 | |
5da5ec37 BP |
203 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
204 | ||
abe529af BP |
205 | struct action_xlate_ctx { |
206 | /* action_xlate_ctx_init() initializes these members. */ | |
207 | ||
208 | /* The ofproto. */ | |
209 | struct ofproto_dpif *ofproto; | |
210 | ||
211 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
212 | * this flow when actions change header fields. */ | |
213 | struct flow flow; | |
214 | ||
215 | /* The packet corresponding to 'flow', or a null pointer if we are | |
216 | * revalidating without a packet to refer to. */ | |
217 | const struct ofpbuf *packet; | |
218 | ||
3de9590b BP |
219 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
220 | * actions update the flow table? | |
221 | * | |
222 | * We want to update these tables if we are actually processing a packet, | |
223 | * or if we are accounting for packets that the datapath has processed, but | |
224 | * not if we are just revalidating. */ | |
225 | bool may_learn; | |
75a75043 | 226 | |
18b2a258 BP |
227 | /* The rule that we are currently translating, or NULL. */ |
228 | struct rule_dpif *rule; | |
54834960 | 229 | |
0e553d9c BP |
230 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
231 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
232 | * timeouts.) */ | |
233 | uint8_t tcp_flags; | |
234 | ||
112bc5f4 BP |
235 | /* If nonnull, flow translation calls this function just before executing a |
236 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
237 | * when the recursion depth is exceeded. | |
238 | * | |
239 | * 'rule' is the rule being submitted into. It will be null if the | |
240 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
241 | * | |
242 | * This is normally null so the client has to set it manually after | |
243 | * calling action_xlate_ctx_init(). */ | |
244 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
245 | ||
479df176 BP |
246 | /* If nonnull, flow translation calls this function to report some |
247 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
248 | * dropped a packet. */ | |
249 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
250 | ||
112bc5f4 BP |
251 | /* If nonnull, flow translation credits the specified statistics to each |
252 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
253 | * |
254 | * This is normally null so the client has to set it manually after | |
255 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 256 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 257 | |
abe529af BP |
258 | /* xlate_actions() initializes and uses these members. The client might want |
259 | * to look at them after it returns. */ | |
260 | ||
261 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 262 | tag_type tags; /* Tags associated with actions. */ |
6a7e895f | 263 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
75a75043 BP |
264 | bool has_learn; /* Actions include NXAST_LEARN? */ |
265 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 266 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 267 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 268 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
269 | |
270 | /* xlate_actions() initializes and uses these members, but the client has no | |
271 | * reason to look at them. */ | |
272 | ||
273 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 274 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 275 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 276 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 277 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
278 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
279 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
280 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 281 | bool exit; /* No further actions should be processed. */ |
ccb7c863 | 282 | struct flow orig_flow; /* Copy of original flow. */ |
abe529af BP |
283 | }; |
284 | ||
285 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
286 | struct ofproto_dpif *, const struct flow *, | |
18b2a258 | 287 | ovs_be16 initial_tci, struct rule_dpif *, |
0e553d9c | 288 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 | 289 | static void xlate_actions(struct action_xlate_ctx *, |
f25d0cf3 | 290 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 BP |
291 | struct ofpbuf *odp_actions); |
292 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
f25d0cf3 BP |
293 | const struct ofpact *ofpacts, |
294 | size_t ofpacts_len); | |
abe529af | 295 | |
6a7e895f BP |
296 | static size_t put_userspace_action(const struct ofproto_dpif *, |
297 | struct ofpbuf *odp_actions, | |
298 | const struct flow *, | |
299 | const union user_action_cookie *); | |
300 | ||
301 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
302 | enum slow_path_reason, | |
303 | uint64_t *stub, size_t stub_size, | |
304 | const struct nlattr **actionsp, | |
305 | size_t *actions_lenp); | |
306 | ||
479df176 BP |
307 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); |
308 | ||
6a7e895f BP |
309 | /* A subfacet (see "struct subfacet" below) has three possible installation |
310 | * states: | |
311 | * | |
312 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
313 | * case just after the subfacet is created, just before the subfacet is | |
314 | * destroyed, or if the datapath returns an error when we try to install a | |
315 | * subfacet. | |
316 | * | |
317 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
318 | * | |
319 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
320 | * ofproto_dpif is installed in the datapath. | |
321 | */ | |
322 | enum subfacet_path { | |
323 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
324 | SF_FAST_PATH, /* Full actions are installed. */ | |
325 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
326 | }; | |
327 | ||
328 | static const char *subfacet_path_to_string(enum subfacet_path); | |
329 | ||
5f5fbd17 BP |
330 | /* A dpif flow and actions associated with a facet. |
331 | * | |
332 | * See also the large comment on struct facet. */ | |
333 | struct subfacet { | |
334 | /* Owners. */ | |
335 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
336 | struct list list_node; /* In struct facet's 'facets' list. */ | |
337 | struct facet *facet; /* Owning facet. */ | |
338 | ||
339 | /* Key. | |
340 | * | |
341 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
342 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
343 | * regenerate the ODP flow key from ->facet->flow. */ | |
344 | enum odp_key_fitness key_fitness; | |
345 | struct nlattr *key; | |
346 | int key_len; | |
347 | ||
348 | long long int used; /* Time last used; time created if not used. */ | |
349 | ||
350 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
351 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
352 | ||
353 | /* Datapath actions. | |
354 | * | |
355 | * These should be essentially identical for every subfacet in a facet, but | |
356 | * may differ in trivial ways due to VLAN splinters. */ | |
357 | size_t actions_len; /* Number of bytes in actions[]. */ | |
358 | struct nlattr *actions; /* Datapath actions. */ | |
359 | ||
6a7e895f BP |
360 | enum slow_path_reason slow; /* 0 if fast path may be used. */ |
361 | enum subfacet_path path; /* Installed in datapath? */ | |
5f5fbd17 BP |
362 | |
363 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
364 | * splinters can cause it to differ. This value should be removed when | |
365 | * the VLAN splinters feature is no longer needed. */ | |
366 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
367 | }; | |
368 | ||
369 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, | |
370 | const struct nlattr *key, | |
371 | size_t key_len, ovs_be16 initial_tci); | |
372 | static struct subfacet *subfacet_find(struct ofproto_dpif *, | |
373 | const struct nlattr *key, size_t key_len); | |
374 | static void subfacet_destroy(struct subfacet *); | |
375 | static void subfacet_destroy__(struct subfacet *); | |
376 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, | |
377 | struct ofpbuf *key); | |
378 | static void subfacet_reset_dp_stats(struct subfacet *, | |
379 | struct dpif_flow_stats *); | |
380 | static void subfacet_update_time(struct subfacet *, long long int used); | |
381 | static void subfacet_update_stats(struct subfacet *, | |
382 | const struct dpif_flow_stats *); | |
383 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
384 | const struct ofpbuf *packet, |
385 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
386 | static int subfacet_install(struct subfacet *, |
387 | const struct nlattr *actions, size_t actions_len, | |
6a7e895f | 388 | struct dpif_flow_stats *, enum slow_path_reason); |
5f5fbd17 BP |
389 | static void subfacet_uninstall(struct subfacet *); |
390 | ||
6a7e895f BP |
391 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); |
392 | ||
b0f7b9b5 BP |
393 | /* An exact-match instantiation of an OpenFlow flow. |
394 | * | |
395 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
396 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
397 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
398 | * the facet. When the kernel module (or other dpif implementation) and Open | |
399 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
400 | * one subfacet per facet. If the dpif implementation supports more-specific | |
401 | * flow matching than userspace, however, a facet can have more than one | |
402 | * subfacet, each of which corresponds to some distinction in flow that | |
403 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
404 | * |
405 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
406 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 407 | struct facet { |
b0f7b9b5 BP |
408 | /* Owners. */ |
409 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
410 | struct list list_node; /* In owning rule's 'facets' list. */ | |
411 | struct rule_dpif *rule; /* Owning rule. */ | |
412 | ||
413 | /* Owned data. */ | |
414 | struct list subfacets; | |
abe529af BP |
415 | long long int used; /* Time last used; time created if not used. */ |
416 | ||
b0f7b9b5 BP |
417 | /* Key. */ |
418 | struct flow flow; | |
419 | ||
abe529af BP |
420 | /* These statistics: |
421 | * | |
422 | * - Do include packets and bytes sent "by hand", e.g. with | |
423 | * dpif_execute(). | |
424 | * | |
425 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 426 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 427 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
428 | * |
429 | * - Do not include packets or bytes that can be obtained from the | |
430 | * datapath for any existing subfacet. | |
abe529af BP |
431 | */ |
432 | uint64_t packet_count; /* Number of packets received. */ | |
433 | uint64_t byte_count; /* Number of bytes received. */ | |
434 | ||
b0f7b9b5 | 435 | /* Resubmit statistics. */ |
9d24de3b JP |
436 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
437 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
438 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 439 | |
b0f7b9b5 | 440 | /* Accounting. */ |
907a4c5e | 441 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 442 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 443 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 444 | |
b95fc6ba BP |
445 | /* Properties of datapath actions. |
446 | * | |
447 | * Every subfacet has its own actions because actions can differ slightly | |
448 | * between splintered and non-splintered subfacets due to the VLAN tag | |
449 | * being initially different (present vs. absent). All of them have these | |
450 | * properties in common so we just store one copy of them here. */ | |
75a75043 BP |
451 | bool has_learn; /* Actions include NXAST_LEARN? */ |
452 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 453 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 454 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 455 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
456 | |
457 | /* Storage for a single subfacet, to reduce malloc() time and space | |
458 | * overhead. (A facet always has at least one subfacet and in the common | |
459 | * case has exactly one subfacet.) */ | |
460 | struct subfacet one_subfacet; | |
abe529af BP |
461 | }; |
462 | ||
2b459b83 BP |
463 | static struct facet *facet_create(struct rule_dpif *, |
464 | const struct flow *, uint32_t hash); | |
15baa734 | 465 | static void facet_remove(struct facet *); |
abe529af BP |
466 | static void facet_free(struct facet *); |
467 | ||
2b459b83 BP |
468 | static struct facet *facet_find(struct ofproto_dpif *, |
469 | const struct flow *, uint32_t hash); | |
abe529af | 470 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 471 | const struct flow *, uint32_t hash); |
c57b2226 | 472 | static void facet_revalidate(struct facet *); |
6814e51f | 473 | static bool facet_check_consistency(struct facet *); |
abe529af | 474 | |
15baa734 | 475 | static void facet_flush_stats(struct facet *); |
abe529af | 476 | |
15baa734 | 477 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 478 | static void facet_reset_counters(struct facet *); |
abe529af | 479 | static void facet_push_stats(struct facet *); |
3de9590b BP |
480 | static void facet_learn(struct facet *); |
481 | static void facet_account(struct facet *); | |
abe529af BP |
482 | |
483 | static bool facet_is_controller_flow(struct facet *); | |
484 | ||
abe529af BP |
485 | struct ofport_dpif { |
486 | struct ofport up; | |
487 | ||
488 | uint32_t odp_port; | |
489 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
490 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
491 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
492 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 493 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 494 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 495 | long long int carrier_seq; /* Carrier status changes. */ |
21f7563c | 496 | |
52a90c29 | 497 | /* Spanning tree. */ |
21f7563c JP |
498 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
499 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
500 | long long int stp_state_entered; | |
8b36f51e EJ |
501 | |
502 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
503 | |
504 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
505 | * | |
506 | * This is deprecated. It is only for compatibility with broken device | |
507 | * drivers in old versions of Linux that do not properly support VLANs when | |
508 | * VLAN devices are not used. When broken device drivers are no longer in | |
509 | * widespread use, we will delete these interfaces. */ | |
510 | uint16_t realdev_ofp_port; | |
511 | int vlandev_vid; | |
8b36f51e EJ |
512 | }; |
513 | ||
514 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
515 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
516 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
517 | struct priority_to_dscp { | |
518 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
519 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
520 | ||
521 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
522 | }; |
523 | ||
52a90c29 BP |
524 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
525 | * | |
526 | * This is deprecated. It is only for compatibility with broken device drivers | |
527 | * in old versions of Linux that do not properly support VLANs when VLAN | |
528 | * devices are not used. When broken device drivers are no longer in | |
529 | * widespread use, we will delete these interfaces. */ | |
530 | struct vlan_splinter { | |
531 | struct hmap_node realdev_vid_node; | |
532 | struct hmap_node vlandev_node; | |
533 | uint16_t realdev_ofp_port; | |
534 | uint16_t vlandev_ofp_port; | |
535 | int vid; | |
536 | }; | |
537 | ||
538 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
539 | uint32_t realdev, ovs_be16 vlan_tci); | |
b98d8985 | 540 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); |
52a90c29 BP |
541 | static void vsp_remove(struct ofport_dpif *); |
542 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
543 | ||
abe529af BP |
544 | static struct ofport_dpif * |
545 | ofport_dpif_cast(const struct ofport *ofport) | |
546 | { | |
547 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
548 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
549 | } | |
550 | ||
551 | static void port_run(struct ofport_dpif *); | |
0aa66d6e | 552 | static void port_run_fast(struct ofport_dpif *); |
abe529af | 553 | static void port_wait(struct ofport_dpif *); |
a5610457 | 554 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 555 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 556 | |
7ee20df1 BP |
557 | struct dpif_completion { |
558 | struct list list_node; | |
559 | struct ofoperation *op; | |
560 | }; | |
561 | ||
54a9cbc9 BP |
562 | /* Extra information about a classifier table. |
563 | * Currently used just for optimized flow revalidation. */ | |
564 | struct table_dpif { | |
565 | /* If either of these is nonnull, then this table has a form that allows | |
566 | * flows to be tagged to avoid revalidating most flows for the most common | |
567 | * kinds of flow table changes. */ | |
568 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
569 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
570 | uint32_t basis; /* Keeps each table's tags separate. */ | |
571 | }; | |
572 | ||
3c4a309c BP |
573 | /* Reasons that we might need to revalidate every facet, and corresponding |
574 | * coverage counters. | |
575 | * | |
576 | * A value of 0 means that there is no need to revalidate. | |
577 | * | |
578 | * It would be nice to have some cleaner way to integrate with coverage | |
579 | * counters, but with only a few reasons I guess this is good enough for | |
580 | * now. */ | |
581 | enum revalidate_reason { | |
582 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
583 | REV_STP, /* Spanning tree protocol port status change. */ | |
584 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
585 | REV_FLOW_TABLE, /* Flow table changed. */ | |
586 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
587 | }; | |
588 | COVERAGE_DEFINE(rev_reconfigure); | |
589 | COVERAGE_DEFINE(rev_stp); | |
590 | COVERAGE_DEFINE(rev_port_toggled); | |
591 | COVERAGE_DEFINE(rev_flow_table); | |
592 | COVERAGE_DEFINE(rev_inconsistency); | |
593 | ||
abe529af | 594 | struct ofproto_dpif { |
b44a10b7 | 595 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
596 | struct ofproto up; |
597 | struct dpif *dpif; | |
598 | int max_ports; | |
599 | ||
c57b2226 BP |
600 | /* Special OpenFlow rules. */ |
601 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
602 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
603 | ||
6c1491fb BP |
604 | /* Statistics. */ |
605 | uint64_t n_matches; | |
606 | ||
abe529af BP |
607 | /* Bridging. */ |
608 | struct netflow *netflow; | |
bae473fe | 609 | struct dpif_sflow *sflow; |
abe529af BP |
610 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
611 | struct mac_learning *ml; | |
612 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
ccb7c863 | 613 | bool has_mirrors; |
abe529af BP |
614 | bool has_bonded_bundles; |
615 | ||
616 | /* Expiration. */ | |
617 | struct timer next_expiration; | |
618 | ||
619 | /* Facets. */ | |
620 | struct hmap facets; | |
b0f7b9b5 | 621 | struct hmap subfacets; |
9d6ac44e | 622 | struct governor *governor; |
54a9cbc9 BP |
623 | |
624 | /* Revalidation. */ | |
625 | struct table_dpif tables[N_TABLES]; | |
3c4a309c | 626 | enum revalidate_reason need_revalidate; |
abe529af | 627 | struct tag_set revalidate_set; |
7ee20df1 BP |
628 | |
629 | /* Support for debugging async flow mods. */ | |
630 | struct list completions; | |
daff3353 EJ |
631 | |
632 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
633 | struct netdev_stats stats; /* To account packets generated and consumed in |
634 | * userspace. */ | |
21f7563c JP |
635 | |
636 | /* Spanning tree. */ | |
637 | struct stp *stp; | |
638 | long long int stp_last_tick; | |
52a90c29 BP |
639 | |
640 | /* VLAN splinters. */ | |
641 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
642 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
643 | }; |
644 | ||
7ee20df1 BP |
645 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
646 | * for debugging the asynchronous flow_mod implementation.) */ | |
647 | static bool clogged; | |
648 | ||
b44a10b7 BP |
649 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
650 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
651 | ||
abe529af BP |
652 | static void ofproto_dpif_unixctl_init(void); |
653 | ||
654 | static struct ofproto_dpif * | |
655 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
656 | { | |
657 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
658 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
659 | } | |
660 | ||
4acbc98d | 661 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, |
abe529af | 662 | uint16_t ofp_port); |
4acbc98d | 663 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, |
abe529af | 664 | uint32_t odp_port); |
6a6455e5 EJ |
665 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
666 | const struct ofpbuf *, ovs_be16 initial_tci, | |
667 | struct ds *); | |
abe529af BP |
668 | |
669 | /* Packet processing. */ | |
670 | static void update_learning_table(struct ofproto_dpif *, | |
671 | const struct flow *, int vlan, | |
672 | struct ofbundle *); | |
501f8d1f BP |
673 | /* Upcalls. */ |
674 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 675 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
676 | |
677 | /* Flow expiration. */ | |
678 | static int expire(struct ofproto_dpif *); | |
679 | ||
6fca1ffb BP |
680 | /* NetFlow. */ |
681 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
682 | ||
abe529af | 683 | /* Utilities. */ |
52a90c29 | 684 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6a7d1a39 BP |
685 | static size_t compose_sflow_action(const struct ofproto_dpif *, |
686 | struct ofpbuf *odp_actions, | |
687 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
688 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
689 | const struct flow *flow); | |
abe529af BP |
690 | /* Global variables. */ |
691 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
692 | \f | |
693 | /* Factory functions. */ | |
694 | ||
695 | static void | |
696 | enumerate_types(struct sset *types) | |
697 | { | |
698 | dp_enumerate_types(types); | |
699 | } | |
700 | ||
701 | static int | |
702 | enumerate_names(const char *type, struct sset *names) | |
703 | { | |
704 | return dp_enumerate_names(type, names); | |
705 | } | |
706 | ||
707 | static int | |
708 | del(const char *type, const char *name) | |
709 | { | |
710 | struct dpif *dpif; | |
711 | int error; | |
712 | ||
713 | error = dpif_open(name, type, &dpif); | |
714 | if (!error) { | |
715 | error = dpif_delete(dpif); | |
716 | dpif_close(dpif); | |
717 | } | |
718 | return error; | |
719 | } | |
720 | \f | |
721 | /* Basic life-cycle. */ | |
722 | ||
c57b2226 BP |
723 | static int add_internal_flows(struct ofproto_dpif *); |
724 | ||
abe529af BP |
725 | static struct ofproto * |
726 | alloc(void) | |
727 | { | |
728 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
729 | return &ofproto->up; | |
730 | } | |
731 | ||
732 | static void | |
733 | dealloc(struct ofproto *ofproto_) | |
734 | { | |
735 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
736 | free(ofproto); | |
737 | } | |
738 | ||
739 | static int | |
0f5f95a9 | 740 | construct(struct ofproto *ofproto_) |
abe529af BP |
741 | { |
742 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
743 | const char *name = ofproto->up.name; | |
744 | int error; | |
745 | int i; | |
746 | ||
747 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
748 | if (error) { | |
749 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
750 | return error; | |
751 | } | |
752 | ||
753 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 754 | ofproto->n_matches = 0; |
abe529af | 755 | |
be8194bb JG |
756 | dpif_flow_flush(ofproto->dpif); |
757 | dpif_recv_purge(ofproto->dpif); | |
758 | ||
a12b3ead | 759 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
760 | if (error) { |
761 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
762 | dpif_close(ofproto->dpif); | |
763 | return error; | |
764 | } | |
abe529af BP |
765 | |
766 | ofproto->netflow = NULL; | |
767 | ofproto->sflow = NULL; | |
21f7563c | 768 | ofproto->stp = NULL; |
abe529af | 769 | hmap_init(&ofproto->bundles); |
e764773c | 770 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
771 | for (i = 0; i < MAX_MIRRORS; i++) { |
772 | ofproto->mirrors[i] = NULL; | |
773 | } | |
774 | ofproto->has_bonded_bundles = false; | |
775 | ||
776 | timer_set_duration(&ofproto->next_expiration, 1000); | |
777 | ||
778 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 779 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 780 | ofproto->governor = NULL; |
54a9cbc9 BP |
781 | |
782 | for (i = 0; i < N_TABLES; i++) { | |
783 | struct table_dpif *table = &ofproto->tables[i]; | |
784 | ||
785 | table->catchall_table = NULL; | |
786 | table->other_table = NULL; | |
787 | table->basis = random_uint32(); | |
788 | } | |
3c4a309c | 789 | ofproto->need_revalidate = 0; |
abe529af BP |
790 | tag_set_init(&ofproto->revalidate_set); |
791 | ||
7ee20df1 BP |
792 | list_init(&ofproto->completions); |
793 | ||
abe529af BP |
794 | ofproto_dpif_unixctl_init(); |
795 | ||
ccb7c863 | 796 | ofproto->has_mirrors = false; |
daff3353 EJ |
797 | ofproto->has_bundle_action = false; |
798 | ||
52a90c29 BP |
799 | hmap_init(&ofproto->vlandev_map); |
800 | hmap_init(&ofproto->realdev_vid_map); | |
801 | ||
b44a10b7 BP |
802 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
803 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 804 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
805 | |
806 | ofproto_init_tables(ofproto_, N_TABLES); | |
c57b2226 BP |
807 | error = add_internal_flows(ofproto); |
808 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
809 | ||
810 | return error; | |
811 | } | |
812 | ||
813 | static int | |
814 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
f25d0cf3 | 815 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) |
c57b2226 BP |
816 | { |
817 | struct ofputil_flow_mod fm; | |
818 | int error; | |
819 | ||
820 | cls_rule_init_catchall(&fm.cr, 0); | |
821 | cls_rule_set_reg(&fm.cr, 0, id); | |
623e1caf | 822 | fm.new_cookie = htonll(0); |
c57b2226 BP |
823 | fm.cookie = htonll(0); |
824 | fm.cookie_mask = htonll(0); | |
825 | fm.table_id = TBL_INTERNAL; | |
826 | fm.command = OFPFC_ADD; | |
827 | fm.idle_timeout = 0; | |
828 | fm.hard_timeout = 0; | |
829 | fm.buffer_id = 0; | |
830 | fm.out_port = 0; | |
831 | fm.flags = 0; | |
f25d0cf3 BP |
832 | fm.ofpacts = ofpacts->data; |
833 | fm.ofpacts_len = ofpacts->size; | |
c57b2226 BP |
834 | |
835 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
836 | if (error) { | |
837 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
838 | id, ofperr_to_string(error)); | |
839 | return error; | |
840 | } | |
841 | ||
842 | *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL); | |
843 | assert(*rulep != NULL); | |
0f5f95a9 | 844 | |
abe529af BP |
845 | return 0; |
846 | } | |
847 | ||
c57b2226 BP |
848 | static int |
849 | add_internal_flows(struct ofproto_dpif *ofproto) | |
850 | { | |
f25d0cf3 BP |
851 | struct ofpact_controller *controller; |
852 | uint64_t ofpacts_stub[128 / 8]; | |
853 | struct ofpbuf ofpacts; | |
c57b2226 BP |
854 | int error; |
855 | int id; | |
856 | ||
f25d0cf3 | 857 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c57b2226 BP |
858 | id = 1; |
859 | ||
f25d0cf3 BP |
860 | controller = ofpact_put_CONTROLLER(&ofpacts); |
861 | controller->max_len = UINT16_MAX; | |
862 | controller->controller_id = 0; | |
863 | controller->reason = OFPR_NO_MATCH; | |
864 | ofpact_pad(&ofpacts); | |
865 | ||
866 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
c57b2226 BP |
867 | if (error) { |
868 | return error; | |
869 | } | |
870 | ||
f25d0cf3 BP |
871 | ofpbuf_clear(&ofpacts); |
872 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
c57b2226 BP |
873 | &ofproto->no_packet_in_rule); |
874 | return error; | |
875 | } | |
876 | ||
7ee20df1 BP |
877 | static void |
878 | complete_operations(struct ofproto_dpif *ofproto) | |
879 | { | |
880 | struct dpif_completion *c, *next; | |
881 | ||
882 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
883 | ofoperation_complete(c->op, 0); | |
884 | list_remove(&c->list_node); | |
885 | free(c); | |
886 | } | |
887 | } | |
888 | ||
abe529af BP |
889 | static void |
890 | destruct(struct ofproto *ofproto_) | |
891 | { | |
892 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 893 | struct rule_dpif *rule, *next_rule; |
d0918789 | 894 | struct oftable *table; |
abe529af BP |
895 | int i; |
896 | ||
b44a10b7 | 897 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
898 | complete_operations(ofproto); |
899 | ||
0697b5c3 BP |
900 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
901 | struct cls_cursor cursor; | |
902 | ||
d0918789 | 903 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
904 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
905 | ofproto_rule_destroy(&rule->up); | |
906 | } | |
7ee20df1 BP |
907 | } |
908 | ||
abe529af BP |
909 | for (i = 0; i < MAX_MIRRORS; i++) { |
910 | mirror_destroy(ofproto->mirrors[i]); | |
911 | } | |
912 | ||
913 | netflow_destroy(ofproto->netflow); | |
bae473fe | 914 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
915 | hmap_destroy(&ofproto->bundles); |
916 | mac_learning_destroy(ofproto->ml); | |
917 | ||
918 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 919 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 920 | governor_destroy(ofproto->governor); |
abe529af | 921 | |
52a90c29 BP |
922 | hmap_destroy(&ofproto->vlandev_map); |
923 | hmap_destroy(&ofproto->realdev_vid_map); | |
924 | ||
abe529af BP |
925 | dpif_close(ofproto->dpif); |
926 | } | |
927 | ||
928 | static int | |
5fcc0d00 | 929 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
930 | { |
931 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
0aa66d6e | 932 | struct ofport_dpif *ofport; |
9b16c439 | 933 | unsigned int work; |
abe529af | 934 | |
0aa66d6e EJ |
935 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
936 | port_run_fast(ofport); | |
937 | } | |
938 | ||
9b16c439 BP |
939 | /* Handle one or more batches of upcalls, until there's nothing left to do |
940 | * or until we do a fixed total amount of work. | |
941 | * | |
942 | * We do work in batches because it can be much cheaper to set up a number | |
943 | * of flows and fire off their patches all at once. We do multiple batches | |
944 | * because in some cases handling a packet can cause another packet to be | |
945 | * queued almost immediately as part of the return flow. Both | |
946 | * optimizations can make major improvements on some benchmarks and | |
947 | * presumably for real traffic as well. */ | |
948 | work = 0; | |
949 | while (work < FLOW_MISS_MAX_BATCH) { | |
950 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 951 | if (retval <= 0) { |
9b16c439 | 952 | return -retval; |
501f8d1f | 953 | } |
5fcc0d00 BP |
954 | work += retval; |
955 | } | |
956 | return 0; | |
957 | } | |
958 | ||
959 | static int | |
960 | run(struct ofproto *ofproto_) | |
961 | { | |
962 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
963 | struct ofport_dpif *ofport; | |
964 | struct ofbundle *bundle; | |
965 | int error; | |
966 | ||
967 | if (!clogged) { | |
968 | complete_operations(ofproto); | |
969 | } | |
970 | dpif_run(ofproto->dpif); | |
971 | ||
972 | error = run_fast(ofproto_); | |
973 | if (error) { | |
974 | return error; | |
abe529af BP |
975 | } |
976 | ||
977 | if (timer_expired(&ofproto->next_expiration)) { | |
978 | int delay = expire(ofproto); | |
979 | timer_set_duration(&ofproto->next_expiration, delay); | |
980 | } | |
981 | ||
982 | if (ofproto->netflow) { | |
6fca1ffb BP |
983 | if (netflow_run(ofproto->netflow)) { |
984 | send_netflow_active_timeouts(ofproto); | |
985 | } | |
abe529af BP |
986 | } |
987 | if (ofproto->sflow) { | |
bae473fe | 988 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
989 | } |
990 | ||
991 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
992 | port_run(ofport); | |
993 | } | |
994 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
995 | bundle_run(bundle); | |
996 | } | |
997 | ||
21f7563c | 998 | stp_run(ofproto); |
1c313b88 BP |
999 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
1000 | ||
abe529af BP |
1001 | /* Now revalidate if there's anything to do. */ |
1002 | if (ofproto->need_revalidate | |
1003 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
1004 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
1005 | bool revalidate_all = ofproto->need_revalidate; | |
c57b2226 | 1006 | struct facet *facet; |
abe529af | 1007 | |
3c4a309c BP |
1008 | switch (ofproto->need_revalidate) { |
1009 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
1010 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
1011 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
1012 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
1013 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
1014 | } | |
1015 | ||
abe529af BP |
1016 | /* Clear the revalidation flags. */ |
1017 | tag_set_init(&ofproto->revalidate_set); | |
3c4a309c | 1018 | ofproto->need_revalidate = 0; |
abe529af | 1019 | |
c57b2226 | 1020 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
abe529af BP |
1021 | if (revalidate_all |
1022 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 1023 | facet_revalidate(facet); |
abe529af BP |
1024 | } |
1025 | } | |
1026 | } | |
1027 | ||
6814e51f BP |
1028 | /* Check the consistency of a random facet, to aid debugging. */ |
1029 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
1030 | struct facet *facet; | |
1031 | ||
1032 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
1033 | struct facet, hmap_node); | |
1034 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
1035 | if (!facet_check_consistency(facet)) { | |
3c4a309c | 1036 | ofproto->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
1037 | } |
1038 | } | |
1039 | } | |
1040 | ||
9d6ac44e BP |
1041 | if (ofproto->governor) { |
1042 | size_t n_subfacets; | |
1043 | ||
1044 | governor_run(ofproto->governor); | |
1045 | ||
1046 | /* If the governor has shrunk to its minimum size and the number of | |
1047 | * subfacets has dwindled, then drop the governor entirely. | |
1048 | * | |
1049 | * For hysteresis, the number of subfacets to drop the governor is | |
1050 | * smaller than the number needed to trigger its creation. */ | |
1051 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1052 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1053 | && governor_is_idle(ofproto->governor)) { | |
1054 | governor_destroy(ofproto->governor); | |
1055 | ofproto->governor = NULL; | |
1056 | } | |
1057 | } | |
1058 | ||
abe529af BP |
1059 | return 0; |
1060 | } | |
1061 | ||
1062 | static void | |
1063 | wait(struct ofproto *ofproto_) | |
1064 | { | |
1065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1066 | struct ofport_dpif *ofport; | |
1067 | struct ofbundle *bundle; | |
1068 | ||
7ee20df1 BP |
1069 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
1070 | poll_immediate_wake(); | |
1071 | } | |
1072 | ||
abe529af BP |
1073 | dpif_wait(ofproto->dpif); |
1074 | dpif_recv_wait(ofproto->dpif); | |
1075 | if (ofproto->sflow) { | |
bae473fe | 1076 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
1077 | } |
1078 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
1079 | poll_immediate_wake(); | |
1080 | } | |
1081 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1082 | port_wait(ofport); | |
1083 | } | |
1084 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1085 | bundle_wait(bundle); | |
1086 | } | |
6fca1ffb BP |
1087 | if (ofproto->netflow) { |
1088 | netflow_wait(ofproto->netflow); | |
1089 | } | |
1c313b88 | 1090 | mac_learning_wait(ofproto->ml); |
21f7563c | 1091 | stp_wait(ofproto); |
abe529af BP |
1092 | if (ofproto->need_revalidate) { |
1093 | /* Shouldn't happen, but if it does just go around again. */ | |
1094 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1095 | poll_immediate_wake(); | |
1096 | } else { | |
1097 | timer_wait(&ofproto->next_expiration); | |
1098 | } | |
9d6ac44e BP |
1099 | if (ofproto->governor) { |
1100 | governor_wait(ofproto->governor); | |
1101 | } | |
abe529af BP |
1102 | } |
1103 | ||
0d085684 BP |
1104 | static void |
1105 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1106 | { | |
1107 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1108 | ||
1109 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1110 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1111 | } | |
1112 | ||
abe529af BP |
1113 | static void |
1114 | flush(struct ofproto *ofproto_) | |
1115 | { | |
1116 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1117 | struct facet *facet, *next_facet; | |
1118 | ||
1119 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1120 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1121 | * bother trying to uninstall it. There is no point in uninstalling it | |
1122 | * individually since we are about to blow away all the facets with | |
1123 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
1124 | struct subfacet *subfacet; |
1125 | ||
1126 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
6a7e895f | 1127 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 BP |
1128 | subfacet->dp_packet_count = 0; |
1129 | subfacet->dp_byte_count = 0; | |
1130 | } | |
15baa734 | 1131 | facet_remove(facet); |
abe529af BP |
1132 | } |
1133 | dpif_flow_flush(ofproto->dpif); | |
1134 | } | |
1135 | ||
6c1491fb BP |
1136 | static void |
1137 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 1138 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
1139 | { |
1140 | *arp_match_ip = true; | |
9e1fd49b BP |
1141 | *actions = (OFPUTIL_A_OUTPUT | |
1142 | OFPUTIL_A_SET_VLAN_VID | | |
1143 | OFPUTIL_A_SET_VLAN_PCP | | |
1144 | OFPUTIL_A_STRIP_VLAN | | |
1145 | OFPUTIL_A_SET_DL_SRC | | |
1146 | OFPUTIL_A_SET_DL_DST | | |
1147 | OFPUTIL_A_SET_NW_SRC | | |
1148 | OFPUTIL_A_SET_NW_DST | | |
1149 | OFPUTIL_A_SET_NW_TOS | | |
1150 | OFPUTIL_A_SET_TP_SRC | | |
1151 | OFPUTIL_A_SET_TP_DST | | |
1152 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
1153 | } |
1154 | ||
1155 | static void | |
e2b9ac44 | 1156 | get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots) |
6c1491fb BP |
1157 | { |
1158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 1159 | struct dpif_dp_stats s; |
6c1491fb BP |
1160 | |
1161 | strcpy(ots->name, "classifier"); | |
1162 | ||
1163 | dpif_get_dp_stats(ofproto->dpif, &s); | |
1164 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
1165 | put_32aligned_be64(&ots->matched_count, | |
1166 | htonll(s.n_hit + ofproto->n_matches)); | |
1167 | } | |
1168 | ||
abe529af BP |
1169 | static struct ofport * |
1170 | port_alloc(void) | |
1171 | { | |
1172 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1173 | return &port->up; | |
1174 | } | |
1175 | ||
1176 | static void | |
1177 | port_dealloc(struct ofport *port_) | |
1178 | { | |
1179 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1180 | free(port); | |
1181 | } | |
1182 | ||
1183 | static int | |
1184 | port_construct(struct ofport *port_) | |
1185 | { | |
1186 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1187 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1188 | ||
3c4a309c | 1189 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1190 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
1191 | port->bundle = NULL; | |
1192 | port->cfm = NULL; | |
1193 | port->tag = tag_create_random(); | |
d5ffa7f2 | 1194 | port->may_enable = true; |
21f7563c JP |
1195 | port->stp_port = NULL; |
1196 | port->stp_state = STP_DISABLED; | |
8b36f51e | 1197 | hmap_init(&port->priorities); |
52a90c29 BP |
1198 | port->realdev_ofp_port = 0; |
1199 | port->vlandev_vid = 0; | |
3e5b3fdb | 1200 | port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); |
abe529af BP |
1201 | |
1202 | if (ofproto->sflow) { | |
392c7182 | 1203 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
1204 | } |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | static void | |
1210 | port_destruct(struct ofport *port_) | |
1211 | { | |
1212 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1213 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1214 | ||
3c4a309c | 1215 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af | 1216 | bundle_remove(port_); |
a5610457 | 1217 | set_cfm(port_, NULL); |
abe529af | 1218 | if (ofproto->sflow) { |
bae473fe | 1219 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1220 | } |
8b36f51e EJ |
1221 | |
1222 | ofport_clear_priorities(port); | |
1223 | hmap_destroy(&port->priorities); | |
abe529af BP |
1224 | } |
1225 | ||
1226 | static void | |
1227 | port_modified(struct ofport *port_) | |
1228 | { | |
1229 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1230 | ||
1231 | if (port->bundle && port->bundle->bond) { | |
1232 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | static void | |
9e1fd49b | 1237 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1238 | { |
1239 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1240 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1241 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1242 | |
9e1fd49b | 1243 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
c57b2226 BP |
1244 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | |
1245 | OFPUTIL_PC_NO_PACKET_IN)) { | |
3c4a309c | 1246 | ofproto->need_revalidate = REV_RECONFIGURE; |
7bde8dd8 | 1247 | |
9e1fd49b | 1248 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1249 | bundle_update(port->bundle); |
1250 | } | |
abe529af BP |
1251 | } |
1252 | } | |
1253 | ||
1254 | static int | |
1255 | set_sflow(struct ofproto *ofproto_, | |
1256 | const struct ofproto_sflow_options *sflow_options) | |
1257 | { | |
1258 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1259 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1260 | |
abe529af | 1261 | if (sflow_options) { |
bae473fe | 1262 | if (!ds) { |
abe529af BP |
1263 | struct ofport_dpif *ofport; |
1264 | ||
bae473fe | 1265 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1266 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1267 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1268 | } |
3c4a309c | 1269 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af | 1270 | } |
bae473fe | 1271 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1272 | } else { |
6ff686f2 PS |
1273 | if (ds) { |
1274 | dpif_sflow_destroy(ds); | |
3c4a309c | 1275 | ofproto->need_revalidate = REV_RECONFIGURE; |
6ff686f2 PS |
1276 | ofproto->sflow = NULL; |
1277 | } | |
abe529af BP |
1278 | } |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static int | |
a5610457 | 1283 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1284 | { |
1285 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1286 | int error; | |
1287 | ||
a5610457 | 1288 | if (!s) { |
abe529af BP |
1289 | error = 0; |
1290 | } else { | |
1291 | if (!ofport->cfm) { | |
8c977421 EJ |
1292 | struct ofproto_dpif *ofproto; |
1293 | ||
1294 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3c4a309c | 1295 | ofproto->need_revalidate = REV_RECONFIGURE; |
6f629657 | 1296 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1297 | } |
1298 | ||
a5610457 | 1299 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1300 | return 0; |
1301 | } | |
1302 | ||
1303 | error = EINVAL; | |
1304 | } | |
1305 | cfm_destroy(ofport->cfm); | |
1306 | ofport->cfm = NULL; | |
1307 | return error; | |
1308 | } | |
1309 | ||
1310 | static int | |
a5610457 | 1311 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1312 | { |
1313 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1314 | |
1315 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1316 | } |
1de11730 EJ |
1317 | |
1318 | static int | |
1319 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1320 | size_t *n_rmps) | |
1321 | { | |
1322 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1323 | ||
1324 | if (ofport->cfm) { | |
1325 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1326 | return 0; | |
1327 | } else { | |
1328 | return -1; | |
1329 | } | |
1330 | } | |
3967a833 MM |
1331 | |
1332 | static int | |
1333 | get_cfm_health(const struct ofport *ofport_) | |
1334 | { | |
1335 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1336 | ||
1337 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1338 | } | |
abe529af | 1339 | \f |
21f7563c JP |
1340 | /* Spanning Tree. */ |
1341 | ||
1342 | static void | |
1343 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1344 | { | |
1345 | struct ofproto_dpif *ofproto = ofproto_; | |
1346 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1347 | struct ofport_dpif *ofport; | |
1348 | ||
1349 | ofport = stp_port_get_aux(sp); | |
1350 | if (!ofport) { | |
1351 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1352 | ofproto->up.name, port_num); | |
1353 | } else { | |
1354 | struct eth_header *eth = pkt->l2; | |
1355 | ||
1356 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1357 | if (eth_addr_is_zero(eth->eth_src)) { | |
1358 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1359 | "with unknown MAC", ofproto->up.name, port_num); | |
1360 | } else { | |
97d6520b | 1361 | send_packet(ofport, pkt); |
21f7563c JP |
1362 | } |
1363 | } | |
1364 | ofpbuf_delete(pkt); | |
1365 | } | |
1366 | ||
1367 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1368 | static int | |
1369 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1370 | { | |
1371 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1372 | ||
1373 | /* Only revalidate flows if the configuration changed. */ | |
1374 | if (!s != !ofproto->stp) { | |
3c4a309c | 1375 | ofproto->need_revalidate = REV_RECONFIGURE; |
21f7563c JP |
1376 | } |
1377 | ||
1378 | if (s) { | |
1379 | if (!ofproto->stp) { | |
1380 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1381 | send_bpdu_cb, ofproto); | |
1382 | ofproto->stp_last_tick = time_msec(); | |
1383 | } | |
1384 | ||
1385 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1386 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1387 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1388 | stp_set_max_age(ofproto->stp, s->max_age); | |
1389 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1390 | } else { | |
851bf71d EJ |
1391 | struct ofport *ofport; |
1392 | ||
1393 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1394 | set_stp_port(ofport, NULL); | |
1395 | } | |
1396 | ||
21f7563c JP |
1397 | stp_destroy(ofproto->stp); |
1398 | ofproto->stp = NULL; | |
1399 | } | |
1400 | ||
1401 | return 0; | |
1402 | } | |
1403 | ||
1404 | static int | |
1405 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1406 | { | |
1407 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1408 | ||
1409 | if (ofproto->stp) { | |
1410 | s->enabled = true; | |
1411 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1412 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1413 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1414 | } else { | |
1415 | s->enabled = false; | |
1416 | } | |
1417 | ||
1418 | return 0; | |
1419 | } | |
1420 | ||
1421 | static void | |
1422 | update_stp_port_state(struct ofport_dpif *ofport) | |
1423 | { | |
1424 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1425 | enum stp_state state; | |
1426 | ||
1427 | /* Figure out new state. */ | |
1428 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1429 | : STP_DISABLED; | |
1430 | ||
1431 | /* Update state. */ | |
1432 | if (ofport->stp_state != state) { | |
9e1fd49b | 1433 | enum ofputil_port_state of_state; |
21f7563c JP |
1434 | bool fwd_change; |
1435 | ||
1436 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1437 | netdev_get_name(ofport->up.netdev), | |
1438 | stp_state_name(ofport->stp_state), | |
1439 | stp_state_name(state)); | |
1440 | if (stp_learn_in_state(ofport->stp_state) | |
1441 | != stp_learn_in_state(state)) { | |
1442 | /* xxx Learning action flows should also be flushed. */ | |
d0040604 | 1443 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
21f7563c JP |
1444 | } |
1445 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1446 | != stp_forward_in_state(state); | |
1447 | ||
3c4a309c | 1448 | ofproto->need_revalidate = REV_STP; |
21f7563c JP |
1449 | ofport->stp_state = state; |
1450 | ofport->stp_state_entered = time_msec(); | |
1451 | ||
b308140a | 1452 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1453 | bundle_update(ofport->bundle); |
1454 | } | |
1455 | ||
1456 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1457 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1458 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1459 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1460 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1461 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1462 | : 0); | |
21f7563c JP |
1463 | ofproto_port_set_state(&ofport->up, of_state); |
1464 | } | |
1465 | } | |
1466 | ||
1467 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1468 | * caller is responsible for assigning STP port numbers and ensuring | |
1469 | * there are no duplicates. */ | |
1470 | static int | |
1471 | set_stp_port(struct ofport *ofport_, | |
1472 | const struct ofproto_port_stp_settings *s) | |
1473 | { | |
1474 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1475 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1476 | struct stp_port *sp = ofport->stp_port; | |
1477 | ||
1478 | if (!s || !s->enable) { | |
1479 | if (sp) { | |
1480 | ofport->stp_port = NULL; | |
1481 | stp_port_disable(sp); | |
ecd12731 | 1482 | update_stp_port_state(ofport); |
21f7563c JP |
1483 | } |
1484 | return 0; | |
1485 | } else if (sp && stp_port_no(sp) != s->port_num | |
1486 | && ofport == stp_port_get_aux(sp)) { | |
1487 | /* The port-id changed, so disable the old one if it's not | |
1488 | * already in use by another port. */ | |
1489 | stp_port_disable(sp); | |
1490 | } | |
1491 | ||
1492 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1493 | stp_port_enable(sp); | |
1494 | ||
1495 | stp_port_set_aux(sp, ofport); | |
1496 | stp_port_set_priority(sp, s->priority); | |
1497 | stp_port_set_path_cost(sp, s->path_cost); | |
1498 | ||
1499 | update_stp_port_state(ofport); | |
1500 | ||
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | static int | |
1505 | get_stp_port_status(struct ofport *ofport_, | |
1506 | struct ofproto_port_stp_status *s) | |
1507 | { | |
1508 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1509 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1510 | struct stp_port *sp = ofport->stp_port; | |
1511 | ||
1512 | if (!ofproto->stp || !sp) { | |
1513 | s->enabled = false; | |
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | s->enabled = true; | |
1518 | s->port_id = stp_port_get_id(sp); | |
1519 | s->state = stp_port_get_state(sp); | |
1520 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1521 | s->role = stp_port_get_role(sp); | |
80740385 | 1522 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1523 | |
1524 | return 0; | |
1525 | } | |
1526 | ||
1527 | static void | |
1528 | stp_run(struct ofproto_dpif *ofproto) | |
1529 | { | |
1530 | if (ofproto->stp) { | |
1531 | long long int now = time_msec(); | |
1532 | long long int elapsed = now - ofproto->stp_last_tick; | |
1533 | struct stp_port *sp; | |
1534 | ||
1535 | if (elapsed > 0) { | |
1536 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1537 | ofproto->stp_last_tick = now; | |
1538 | } | |
1539 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1540 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1541 | ||
1542 | if (ofport) { | |
1543 | update_stp_port_state(ofport); | |
1544 | } | |
1545 | } | |
6ae50723 EJ |
1546 | |
1547 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1548 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1549 | } | |
21f7563c JP |
1550 | } |
1551 | } | |
1552 | ||
1553 | static void | |
1554 | stp_wait(struct ofproto_dpif *ofproto) | |
1555 | { | |
1556 | if (ofproto->stp) { | |
1557 | poll_timer_wait(1000); | |
1558 | } | |
1559 | } | |
1560 | ||
1561 | /* Returns true if STP should process 'flow'. */ | |
1562 | static bool | |
1563 | stp_should_process_flow(const struct flow *flow) | |
1564 | { | |
1565 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1566 | } | |
1567 | ||
1568 | static void | |
1569 | stp_process_packet(const struct ofport_dpif *ofport, | |
1570 | const struct ofpbuf *packet) | |
1571 | { | |
1572 | struct ofpbuf payload = *packet; | |
1573 | struct eth_header *eth = payload.data; | |
1574 | struct stp_port *sp = ofport->stp_port; | |
1575 | ||
1576 | /* Sink packets on ports that have STP disabled when the bridge has | |
1577 | * STP enabled. */ | |
1578 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1579 | return; | |
1580 | } | |
1581 | ||
1582 | /* Trim off padding on payload. */ | |
c573540b BP |
1583 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1584 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1585 | } |
1586 | ||
1587 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1588 | stp_received_bpdu(sp, payload.data, payload.size); | |
1589 | } | |
1590 | } | |
1591 | \f | |
8b36f51e EJ |
1592 | static struct priority_to_dscp * |
1593 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1594 | { | |
1595 | struct priority_to_dscp *pdscp; | |
1596 | uint32_t hash; | |
1597 | ||
1598 | hash = hash_int(priority, 0); | |
1599 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1600 | if (pdscp->priority == priority) { | |
1601 | return pdscp; | |
1602 | } | |
1603 | } | |
1604 | return NULL; | |
1605 | } | |
1606 | ||
1607 | static void | |
1608 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1609 | { | |
1610 | struct priority_to_dscp *pdscp, *next; | |
1611 | ||
1612 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1613 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1614 | free(pdscp); | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | static int | |
1619 | set_queues(struct ofport *ofport_, | |
1620 | const struct ofproto_port_queue *qdscp_list, | |
1621 | size_t n_qdscp) | |
1622 | { | |
1623 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1624 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1625 | struct hmap new = HMAP_INITIALIZER(&new); | |
1626 | size_t i; | |
1627 | ||
1628 | for (i = 0; i < n_qdscp; i++) { | |
1629 | struct priority_to_dscp *pdscp; | |
1630 | uint32_t priority; | |
1631 | uint8_t dscp; | |
1632 | ||
1633 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1634 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1635 | &priority)) { | |
1636 | continue; | |
1637 | } | |
1638 | ||
1639 | pdscp = get_priority(ofport, priority); | |
1640 | if (pdscp) { | |
1641 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1642 | } else { | |
1643 | pdscp = xmalloc(sizeof *pdscp); | |
1644 | pdscp->priority = priority; | |
1645 | pdscp->dscp = dscp; | |
3c4a309c | 1646 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1647 | } |
1648 | ||
1649 | if (pdscp->dscp != dscp) { | |
1650 | pdscp->dscp = dscp; | |
3c4a309c | 1651 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1652 | } |
1653 | ||
1654 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1655 | } | |
1656 | ||
1657 | if (!hmap_is_empty(&ofport->priorities)) { | |
1658 | ofport_clear_priorities(ofport); | |
3c4a309c | 1659 | ofproto->need_revalidate = REV_RECONFIGURE; |
8b36f51e EJ |
1660 | } |
1661 | ||
1662 | hmap_swap(&new, &ofport->priorities); | |
1663 | hmap_destroy(&new); | |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | \f | |
abe529af BP |
1668 | /* Bundles. */ |
1669 | ||
b44a10b7 BP |
1670 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1671 | * ofproto to revalidate every flow. | |
1672 | * | |
1673 | * Normally MAC learning entries are removed only from the ofproto associated | |
1674 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1675 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1676 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1677 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1678 | * with the host from which it migrated. */ | |
abe529af | 1679 | static void |
b44a10b7 | 1680 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1681 | { |
1682 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1683 | struct mac_learning *ml = ofproto->ml; | |
1684 | struct mac_entry *mac, *next_mac; | |
1685 | ||
3c4a309c | 1686 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1687 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { |
1688 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1689 | if (all_ofprotos) { |
1690 | struct ofproto_dpif *o; | |
1691 | ||
1692 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1693 | if (o != ofproto) { | |
1694 | struct mac_entry *e; | |
1695 | ||
1696 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1697 | NULL); | |
1698 | if (e) { | |
1699 | tag_set_add(&o->revalidate_set, e->tag); | |
1700 | mac_learning_expire(o->ml, e); | |
1701 | } | |
1702 | } | |
1703 | } | |
1704 | } | |
1705 | ||
abe529af BP |
1706 | mac_learning_expire(ml, mac); |
1707 | } | |
1708 | } | |
1709 | } | |
1710 | ||
1711 | static struct ofbundle * | |
1712 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1713 | { | |
1714 | struct ofbundle *bundle; | |
1715 | ||
1716 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1717 | &ofproto->bundles) { | |
1718 | if (bundle->aux == aux) { | |
1719 | return bundle; | |
1720 | } | |
1721 | } | |
1722 | return NULL; | |
1723 | } | |
1724 | ||
1725 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1726 | * ones that are found to 'bundles'. */ | |
1727 | static void | |
1728 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1729 | void **auxes, size_t n_auxes, | |
1730 | struct hmapx *bundles) | |
1731 | { | |
1732 | size_t i; | |
1733 | ||
1734 | hmapx_init(bundles); | |
1735 | for (i = 0; i < n_auxes; i++) { | |
1736 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1737 | if (bundle) { | |
1738 | hmapx_add(bundles, bundle); | |
1739 | } | |
1740 | } | |
1741 | } | |
1742 | ||
7bde8dd8 JP |
1743 | static void |
1744 | bundle_update(struct ofbundle *bundle) | |
1745 | { | |
1746 | struct ofport_dpif *port; | |
1747 | ||
1748 | bundle->floodable = true; | |
1749 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
1750 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1751 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1752 | bundle->floodable = false; |
1753 | break; | |
1754 | } | |
1755 | } | |
1756 | } | |
1757 | ||
abe529af BP |
1758 | static void |
1759 | bundle_del_port(struct ofport_dpif *port) | |
1760 | { | |
1761 | struct ofbundle *bundle = port->bundle; | |
1762 | ||
3c4a309c | 1763 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
6f77f4ae | 1764 | |
abe529af BP |
1765 | list_remove(&port->bundle_node); |
1766 | port->bundle = NULL; | |
1767 | ||
1768 | if (bundle->lacp) { | |
1769 | lacp_slave_unregister(bundle->lacp, port); | |
1770 | } | |
1771 | if (bundle->bond) { | |
1772 | bond_slave_unregister(bundle->bond, port); | |
1773 | } | |
1774 | ||
7bde8dd8 | 1775 | bundle_update(bundle); |
abe529af BP |
1776 | } |
1777 | ||
1778 | static bool | |
1779 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1780 | struct lacp_slave_settings *lacp, |
1781 | uint32_t bond_stable_id) | |
abe529af BP |
1782 | { |
1783 | struct ofport_dpif *port; | |
1784 | ||
1785 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1786 | if (!port) { | |
1787 | return false; | |
1788 | } | |
1789 | ||
1790 | if (port->bundle != bundle) { | |
3c4a309c | 1791 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1792 | if (port->bundle) { |
1793 | bundle_del_port(port); | |
1794 | } | |
1795 | ||
1796 | port->bundle = bundle; | |
1797 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
1798 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1799 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1800 | bundle->floodable = false; |
1801 | } | |
1802 | } | |
1803 | if (lacp) { | |
3c4a309c | 1804 | port->bundle->ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1805 | lacp_slave_register(bundle->lacp, port, lacp); |
1806 | } | |
1807 | ||
00794817 BP |
1808 | port->bond_stable_id = bond_stable_id; |
1809 | ||
abe529af BP |
1810 | return true; |
1811 | } | |
1812 | ||
1813 | static void | |
1814 | bundle_destroy(struct ofbundle *bundle) | |
1815 | { | |
1816 | struct ofproto_dpif *ofproto; | |
1817 | struct ofport_dpif *port, *next_port; | |
1818 | int i; | |
1819 | ||
1820 | if (!bundle) { | |
1821 | return; | |
1822 | } | |
1823 | ||
1824 | ofproto = bundle->ofproto; | |
1825 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1826 | struct ofmirror *m = ofproto->mirrors[i]; | |
1827 | if (m) { | |
1828 | if (m->out == bundle) { | |
1829 | mirror_destroy(m); | |
1830 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1831 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
3c4a309c | 1832 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1833 | } |
1834 | } | |
1835 | } | |
1836 | ||
1837 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1838 | bundle_del_port(port); | |
1839 | } | |
1840 | ||
b44a10b7 | 1841 | bundle_flush_macs(bundle, true); |
abe529af BP |
1842 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1843 | free(bundle->name); | |
1844 | free(bundle->trunks); | |
1845 | lacp_destroy(bundle->lacp); | |
1846 | bond_destroy(bundle->bond); | |
1847 | free(bundle); | |
1848 | } | |
1849 | ||
1850 | static int | |
1851 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1852 | const struct ofproto_bundle_settings *s) | |
1853 | { | |
1854 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1855 | bool need_flush = false; | |
abe529af BP |
1856 | struct ofport_dpif *port; |
1857 | struct ofbundle *bundle; | |
ecac4ebf BP |
1858 | unsigned long *trunks; |
1859 | int vlan; | |
abe529af BP |
1860 | size_t i; |
1861 | bool ok; | |
1862 | ||
1863 | if (!s) { | |
1864 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1869 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1870 | ||
1871 | bundle = bundle_lookup(ofproto, aux); | |
1872 | if (!bundle) { | |
1873 | bundle = xmalloc(sizeof *bundle); | |
1874 | ||
1875 | bundle->ofproto = ofproto; | |
1876 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1877 | hash_pointer(aux, 0)); | |
1878 | bundle->aux = aux; | |
1879 | bundle->name = NULL; | |
1880 | ||
1881 | list_init(&bundle->ports); | |
ecac4ebf | 1882 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1883 | bundle->vlan = -1; |
1884 | bundle->trunks = NULL; | |
5e9ceccd | 1885 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1886 | bundle->lacp = NULL; |
1887 | bundle->bond = NULL; | |
1888 | ||
1889 | bundle->floodable = true; | |
1890 | ||
1891 | bundle->src_mirrors = 0; | |
1892 | bundle->dst_mirrors = 0; | |
1893 | bundle->mirror_out = 0; | |
1894 | } | |
1895 | ||
1896 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1897 | free(bundle->name); | |
1898 | bundle->name = xstrdup(s->name); | |
1899 | } | |
1900 | ||
1901 | /* LACP. */ | |
1902 | if (s->lacp) { | |
1903 | if (!bundle->lacp) { | |
3c4a309c | 1904 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
1905 | bundle->lacp = lacp_create(); |
1906 | } | |
1907 | lacp_configure(bundle->lacp, s->lacp); | |
1908 | } else { | |
1909 | lacp_destroy(bundle->lacp); | |
1910 | bundle->lacp = NULL; | |
1911 | } | |
1912 | ||
1913 | /* Update set of ports. */ | |
1914 | ok = true; | |
1915 | for (i = 0; i < s->n_slaves; i++) { | |
1916 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1917 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1918 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1919 | ok = false; |
1920 | } | |
1921 | } | |
1922 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1923 | struct ofport_dpif *next_port; | |
1924 | ||
1925 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1926 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1927 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1928 | goto found; |
1929 | } | |
1930 | } | |
1931 | ||
1932 | bundle_del_port(port); | |
1933 | found: ; | |
1934 | } | |
1935 | } | |
1936 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1937 | ||
1938 | if (list_is_empty(&bundle->ports)) { | |
1939 | bundle_destroy(bundle); | |
1940 | return EINVAL; | |
1941 | } | |
1942 | ||
ecac4ebf | 1943 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1944 | if (s->vlan_mode != bundle->vlan_mode |
1945 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1946 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1947 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1948 | need_flush = true; |
1949 | } | |
1950 | ||
abe529af | 1951 | /* Set VLAN tag. */ |
ecac4ebf BP |
1952 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1953 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1954 | : 0); | |
1955 | if (vlan != bundle->vlan) { | |
1956 | bundle->vlan = vlan; | |
abe529af BP |
1957 | need_flush = true; |
1958 | } | |
1959 | ||
1960 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1961 | switch (s->vlan_mode) { |
1962 | case PORT_VLAN_ACCESS: | |
1963 | trunks = NULL; | |
1964 | break; | |
1965 | ||
1966 | case PORT_VLAN_TRUNK: | |
1967 | trunks = (unsigned long *) s->trunks; | |
1968 | break; | |
1969 | ||
1970 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1971 | case PORT_VLAN_NATIVE_TAGGED: | |
1972 | if (vlan != 0 && (!s->trunks | |
1973 | || !bitmap_is_set(s->trunks, vlan) | |
1974 | || bitmap_is_set(s->trunks, 0))) { | |
1975 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1976 | if (s->trunks) { | |
1977 | trunks = bitmap_clone(s->trunks, 4096); | |
1978 | } else { | |
1979 | trunks = bitmap_allocate1(4096); | |
1980 | } | |
1981 | bitmap_set1(trunks, vlan); | |
1982 | bitmap_set0(trunks, 0); | |
1983 | } else { | |
1984 | trunks = (unsigned long *) s->trunks; | |
1985 | } | |
1986 | break; | |
1987 | ||
1988 | default: | |
1989 | NOT_REACHED(); | |
1990 | } | |
abe529af BP |
1991 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1992 | free(bundle->trunks); | |
ecac4ebf BP |
1993 | if (trunks == s->trunks) { |
1994 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1995 | } else { | |
1996 | bundle->trunks = trunks; | |
1997 | trunks = NULL; | |
1998 | } | |
abe529af BP |
1999 | need_flush = true; |
2000 | } | |
ecac4ebf BP |
2001 | if (trunks != s->trunks) { |
2002 | free(trunks); | |
2003 | } | |
abe529af BP |
2004 | |
2005 | /* Bonding. */ | |
2006 | if (!list_is_short(&bundle->ports)) { | |
2007 | bundle->ofproto->has_bonded_bundles = true; | |
2008 | if (bundle->bond) { | |
2009 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
3c4a309c | 2010 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2011 | } |
2012 | } else { | |
2013 | bundle->bond = bond_create(s->bond); | |
3c4a309c | 2014 | ofproto->need_revalidate = REV_RECONFIGURE; |
abe529af BP |
2015 | } |
2016 | ||
2017 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 2018 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
2019 | port->up.netdev); |
2020 | } | |
2021 | } else { | |
2022 | bond_destroy(bundle->bond); | |
2023 | bundle->bond = NULL; | |
2024 | } | |
2025 | ||
2026 | /* If we changed something that would affect MAC learning, un-learn | |
2027 | * everything on this port and force flow revalidation. */ | |
2028 | if (need_flush) { | |
b44a10b7 | 2029 | bundle_flush_macs(bundle, false); |
abe529af BP |
2030 | } |
2031 | ||
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | static void | |
2036 | bundle_remove(struct ofport *port_) | |
2037 | { | |
2038 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2039 | struct ofbundle *bundle = port->bundle; | |
2040 | ||
2041 | if (bundle) { | |
2042 | bundle_del_port(port); | |
2043 | if (list_is_empty(&bundle->ports)) { | |
2044 | bundle_destroy(bundle); | |
2045 | } else if (list_is_short(&bundle->ports)) { | |
2046 | bond_destroy(bundle->bond); | |
2047 | bundle->bond = NULL; | |
2048 | } | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | static void | |
5f877369 | 2053 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
2054 | { |
2055 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2056 | struct ofport_dpif *port = port_; | |
2057 | uint8_t ea[ETH_ADDR_LEN]; | |
2058 | int error; | |
2059 | ||
2060 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2061 | if (!error) { | |
abe529af | 2062 | struct ofpbuf packet; |
5f877369 | 2063 | void *packet_pdu; |
abe529af BP |
2064 | |
2065 | ofpbuf_init(&packet, 0); | |
2066 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
2067 | pdu_size); |
2068 | memcpy(packet_pdu, pdu, pdu_size); | |
2069 | ||
97d6520b | 2070 | send_packet(port, &packet); |
abe529af BP |
2071 | ofpbuf_uninit(&packet); |
2072 | } else { | |
2073 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2074 | "%s (%s)", port->bundle->name, | |
2075 | netdev_get_name(port->up.netdev), strerror(error)); | |
2076 | } | |
2077 | } | |
2078 | ||
2079 | static void | |
2080 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2081 | { | |
2082 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2083 | int error, n_packets, n_errors; | |
2084 | struct mac_entry *e; | |
2085 | ||
2086 | error = n_packets = n_errors = 0; | |
2087 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2088 | if (e->port.p != bundle) { | |
ea131871 JG |
2089 | struct ofpbuf *learning_packet; |
2090 | struct ofport_dpif *port; | |
4dd1e3ca | 2091 | void *port_void; |
ea131871 JG |
2092 | int ret; |
2093 | ||
4dd1e3ca BP |
2094 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
2095 | * struct ofport_dpif more effective. */ | |
2096 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2097 | e->mac, e->vlan, | |
2098 | &port_void); | |
2099 | port = port_void; | |
97d6520b | 2100 | ret = send_packet(port, learning_packet); |
ea131871 | 2101 | ofpbuf_delete(learning_packet); |
abe529af BP |
2102 | if (ret) { |
2103 | error = ret; | |
2104 | n_errors++; | |
2105 | } | |
2106 | n_packets++; | |
2107 | } | |
2108 | } | |
2109 | ||
2110 | if (n_errors) { | |
2111 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2112 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2113 | "packets, last error was: %s", | |
2114 | bundle->name, n_errors, n_packets, strerror(error)); | |
2115 | } else { | |
2116 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2117 | bundle->name, n_packets); | |
2118 | } | |
2119 | } | |
2120 | ||
2121 | static void | |
2122 | bundle_run(struct ofbundle *bundle) | |
2123 | { | |
2124 | if (bundle->lacp) { | |
2125 | lacp_run(bundle->lacp, send_pdu_cb); | |
2126 | } | |
2127 | if (bundle->bond) { | |
2128 | struct ofport_dpif *port; | |
2129 | ||
2130 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 2131 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
2132 | } |
2133 | ||
2134 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
bdebeece | 2135 | lacp_status(bundle->lacp)); |
abe529af BP |
2136 | if (bond_should_send_learning_packets(bundle->bond)) { |
2137 | bundle_send_learning_packets(bundle); | |
2138 | } | |
2139 | } | |
2140 | } | |
2141 | ||
2142 | static void | |
2143 | bundle_wait(struct ofbundle *bundle) | |
2144 | { | |
2145 | if (bundle->lacp) { | |
2146 | lacp_wait(bundle->lacp); | |
2147 | } | |
2148 | if (bundle->bond) { | |
2149 | bond_wait(bundle->bond); | |
2150 | } | |
2151 | } | |
2152 | \f | |
2153 | /* Mirrors. */ | |
2154 | ||
2155 | static int | |
2156 | mirror_scan(struct ofproto_dpif *ofproto) | |
2157 | { | |
2158 | int idx; | |
2159 | ||
2160 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2161 | if (!ofproto->mirrors[idx]) { | |
2162 | return idx; | |
2163 | } | |
2164 | } | |
2165 | return -1; | |
2166 | } | |
2167 | ||
2168 | static struct ofmirror * | |
2169 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2170 | { | |
2171 | int i; | |
2172 | ||
2173 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2174 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2175 | if (mirror && mirror->aux == aux) { | |
2176 | return mirror; | |
2177 | } | |
2178 | } | |
2179 | ||
2180 | return NULL; | |
2181 | } | |
2182 | ||
9ba15e2a BP |
2183 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2184 | static void | |
2185 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2186 | { | |
2187 | int i; | |
2188 | ||
2189 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2190 | struct ofmirror *m = ofproto->mirrors[i]; | |
2191 | ||
2192 | if (m) { | |
2193 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2194 | } | |
2195 | } | |
2196 | ||
2197 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2198 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2199 | int j; | |
2200 | ||
2201 | if (!m1) { | |
2202 | continue; | |
2203 | } | |
2204 | ||
2205 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2206 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2207 | ||
edb0540b | 2208 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2209 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2210 | m2->dup_mirrors |= m1->dup_mirrors; | |
2211 | } | |
2212 | } | |
2213 | } | |
2214 | } | |
2215 | ||
abe529af BP |
2216 | static int |
2217 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2218 | const struct ofproto_mirror_settings *s) | |
2219 | { | |
2220 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2221 | mirror_mask_t mirror_bit; | |
2222 | struct ofbundle *bundle; | |
2223 | struct ofmirror *mirror; | |
2224 | struct ofbundle *out; | |
2225 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2226 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2227 | int out_vlan; | |
2228 | ||
2229 | mirror = mirror_lookup(ofproto, aux); | |
2230 | if (!s) { | |
2231 | mirror_destroy(mirror); | |
2232 | return 0; | |
2233 | } | |
2234 | if (!mirror) { | |
2235 | int idx; | |
2236 | ||
2237 | idx = mirror_scan(ofproto); | |
2238 | if (idx < 0) { | |
2239 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2240 | "cannot create %s", | |
2241 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2242 | return EFBIG; | |
2243 | } | |
2244 | ||
2245 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2246 | mirror->ofproto = ofproto; | |
2247 | mirror->idx = idx; | |
8b28d864 | 2248 | mirror->aux = aux; |
abe529af BP |
2249 | mirror->out_vlan = -1; |
2250 | mirror->name = NULL; | |
2251 | } | |
2252 | ||
2253 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2254 | free(mirror->name); | |
2255 | mirror->name = xstrdup(s->name); | |
2256 | } | |
2257 | ||
2258 | /* Get the new configuration. */ | |
2259 | if (s->out_bundle) { | |
2260 | out = bundle_lookup(ofproto, s->out_bundle); | |
2261 | if (!out) { | |
2262 | mirror_destroy(mirror); | |
2263 | return EINVAL; | |
2264 | } | |
2265 | out_vlan = -1; | |
2266 | } else { | |
2267 | out = NULL; | |
2268 | out_vlan = s->out_vlan; | |
2269 | } | |
2270 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2271 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2272 | ||
2273 | /* If the configuration has not changed, do nothing. */ | |
2274 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2275 | && hmapx_equals(&dsts, &mirror->dsts) | |
2276 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2277 | && mirror->out == out | |
2278 | && mirror->out_vlan == out_vlan) | |
2279 | { | |
2280 | hmapx_destroy(&srcs); | |
2281 | hmapx_destroy(&dsts); | |
2282 | return 0; | |
2283 | } | |
2284 | ||
2285 | hmapx_swap(&srcs, &mirror->srcs); | |
2286 | hmapx_destroy(&srcs); | |
2287 | ||
2288 | hmapx_swap(&dsts, &mirror->dsts); | |
2289 | hmapx_destroy(&dsts); | |
2290 | ||
2291 | free(mirror->vlans); | |
2292 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2293 | ||
2294 | mirror->out = out; | |
2295 | mirror->out_vlan = out_vlan; | |
2296 | ||
2297 | /* Update bundles. */ | |
2298 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2299 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2300 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2301 | bundle->src_mirrors |= mirror_bit; | |
2302 | } else { | |
2303 | bundle->src_mirrors &= ~mirror_bit; | |
2304 | } | |
2305 | ||
2306 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2307 | bundle->dst_mirrors |= mirror_bit; | |
2308 | } else { | |
2309 | bundle->dst_mirrors &= ~mirror_bit; | |
2310 | } | |
2311 | ||
2312 | if (mirror->out == bundle) { | |
2313 | bundle->mirror_out |= mirror_bit; | |
2314 | } else { | |
2315 | bundle->mirror_out &= ~mirror_bit; | |
2316 | } | |
2317 | } | |
2318 | ||
3c4a309c | 2319 | ofproto->need_revalidate = REV_RECONFIGURE; |
ccb7c863 | 2320 | ofproto->has_mirrors = true; |
d0040604 | 2321 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
9ba15e2a | 2322 | mirror_update_dups(ofproto); |
abe529af BP |
2323 | |
2324 | return 0; | |
2325 | } | |
2326 | ||
2327 | static void | |
2328 | mirror_destroy(struct ofmirror *mirror) | |
2329 | { | |
2330 | struct ofproto_dpif *ofproto; | |
2331 | mirror_mask_t mirror_bit; | |
2332 | struct ofbundle *bundle; | |
ccb7c863 | 2333 | int i; |
abe529af BP |
2334 | |
2335 | if (!mirror) { | |
2336 | return; | |
2337 | } | |
2338 | ||
2339 | ofproto = mirror->ofproto; | |
3c4a309c | 2340 | ofproto->need_revalidate = REV_RECONFIGURE; |
d0040604 | 2341 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2342 | |
2343 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2344 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2345 | bundle->src_mirrors &= ~mirror_bit; | |
2346 | bundle->dst_mirrors &= ~mirror_bit; | |
2347 | bundle->mirror_out &= ~mirror_bit; | |
2348 | } | |
2349 | ||
2350 | hmapx_destroy(&mirror->srcs); | |
2351 | hmapx_destroy(&mirror->dsts); | |
2352 | free(mirror->vlans); | |
2353 | ||
2354 | ofproto->mirrors[mirror->idx] = NULL; | |
2355 | free(mirror->name); | |
2356 | free(mirror); | |
9ba15e2a BP |
2357 | |
2358 | mirror_update_dups(ofproto); | |
ccb7c863 BP |
2359 | |
2360 | ofproto->has_mirrors = false; | |
2361 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2362 | if (ofproto->mirrors[i]) { | |
2363 | ofproto->has_mirrors = true; | |
2364 | break; | |
2365 | } | |
2366 | } | |
abe529af BP |
2367 | } |
2368 | ||
9d24de3b JP |
2369 | static int |
2370 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2371 | uint64_t *packets, uint64_t *bytes) | |
2372 | { | |
2373 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2374 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2375 | ||
2376 | if (!mirror) { | |
2377 | *packets = *bytes = UINT64_MAX; | |
2378 | return 0; | |
2379 | } | |
2380 | ||
2381 | *packets = mirror->packet_count; | |
2382 | *bytes = mirror->byte_count; | |
2383 | ||
2384 | return 0; | |
2385 | } | |
2386 | ||
abe529af BP |
2387 | static int |
2388 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2389 | { | |
2390 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2391 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
d0040604 | 2392 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2393 | } |
2394 | return 0; | |
2395 | } | |
2396 | ||
2397 | static bool | |
b4affc74 | 2398 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2399 | { |
2400 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2401 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2402 | return bundle && bundle->mirror_out != 0; | |
2403 | } | |
8402c74b SS |
2404 | |
2405 | static void | |
b53055f4 | 2406 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2407 | { |
2408 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3c4a309c | 2409 | ofproto->need_revalidate = REV_RECONFIGURE; |
8402c74b | 2410 | } |
e764773c BP |
2411 | |
2412 | static void | |
2413 | set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) | |
2414 | { | |
2415 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2416 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
2417 | } | |
abe529af BP |
2418 | \f |
2419 | /* Ports. */ | |
2420 | ||
2421 | static struct ofport_dpif * | |
4acbc98d | 2422 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) |
abe529af | 2423 | { |
7df6a8bd BP |
2424 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2425 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2426 | } |
2427 | ||
2428 | static struct ofport_dpif * | |
4acbc98d | 2429 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) |
abe529af BP |
2430 | { |
2431 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2432 | } | |
2433 | ||
2434 | static void | |
2435 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2436 | struct dpif_port *dpif_port) | |
2437 | { | |
2438 | ofproto_port->name = dpif_port->name; | |
2439 | ofproto_port->type = dpif_port->type; | |
2440 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2441 | } | |
2442 | ||
0aa66d6e EJ |
2443 | static void |
2444 | port_run_fast(struct ofport_dpif *ofport) | |
2445 | { | |
2446 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2447 | struct ofpbuf packet; | |
2448 | ||
2449 | ofpbuf_init(&packet, 0); | |
2450 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2451 | send_packet(ofport, &packet); | |
2452 | ofpbuf_uninit(&packet); | |
2453 | } | |
2454 | } | |
2455 | ||
abe529af BP |
2456 | static void |
2457 | port_run(struct ofport_dpif *ofport) | |
2458 | { | |
3e5b3fdb EJ |
2459 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2460 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2461 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2462 | ||
3e5b3fdb EJ |
2463 | ofport->carrier_seq = carrier_seq; |
2464 | ||
0aa66d6e | 2465 | port_run_fast(ofport); |
abe529af BP |
2466 | if (ofport->cfm) { |
2467 | cfm_run(ofport->cfm); | |
86dc6501 EJ |
2468 | enable = enable && !cfm_get_fault(ofport->cfm) |
2469 | && cfm_get_opup(ofport->cfm); | |
abe529af | 2470 | } |
015e08bc EJ |
2471 | |
2472 | if (ofport->bundle) { | |
2473 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2474 | if (carrier_changed) { |
2475 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2476 | } | |
015e08bc EJ |
2477 | } |
2478 | ||
daff3353 EJ |
2479 | if (ofport->may_enable != enable) { |
2480 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2481 | ||
2482 | if (ofproto->has_bundle_action) { | |
3c4a309c | 2483 | ofproto->need_revalidate = REV_PORT_TOGGLED; |
daff3353 EJ |
2484 | } |
2485 | } | |
2486 | ||
015e08bc | 2487 | ofport->may_enable = enable; |
abe529af BP |
2488 | } |
2489 | ||
2490 | static void | |
2491 | port_wait(struct ofport_dpif *ofport) | |
2492 | { | |
2493 | if (ofport->cfm) { | |
2494 | cfm_wait(ofport->cfm); | |
2495 | } | |
2496 | } | |
2497 | ||
2498 | static int | |
2499 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2500 | struct ofproto_port *ofproto_port) | |
2501 | { | |
2502 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2503 | struct dpif_port dpif_port; | |
2504 | int error; | |
2505 | ||
2506 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2507 | if (!error) { | |
2508 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2509 | } | |
2510 | return error; | |
2511 | } | |
2512 | ||
2513 | static int | |
2514 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2515 | { | |
2516 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
232dfa4a | 2517 | uint16_t odp_port = UINT16_MAX; |
abe529af BP |
2518 | int error; |
2519 | ||
2520 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2521 | if (!error) { | |
2522 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2523 | } | |
2524 | return error; | |
2525 | } | |
2526 | ||
2527 | static int | |
2528 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2529 | { | |
2530 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2531 | int error; | |
2532 | ||
2533 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2534 | if (!error) { | |
2535 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2536 | if (ofport) { | |
2537 | /* The caller is going to close ofport->up.netdev. If this is a | |
2538 | * bonded port, then the bond is using that netdev, so remove it | |
2539 | * from the bond. The client will need to reconfigure everything | |
2540 | * after deleting ports, so then the slave will get re-added. */ | |
2541 | bundle_remove(&ofport->up); | |
2542 | } | |
2543 | } | |
2544 | return error; | |
2545 | } | |
2546 | ||
6527c598 PS |
2547 | static int |
2548 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2549 | { | |
2550 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2551 | int error; | |
2552 | ||
2553 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2554 | ||
2555 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2556 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2557 | ||
2558 | /* ofproto->stats.tx_packets represents packets that we created | |
2559 | * internally and sent to some port (e.g. packets sent with | |
2560 | * send_packet()). Account for them as if they had come from | |
2561 | * OFPP_LOCAL and got forwarded. */ | |
2562 | ||
2563 | if (stats->rx_packets != UINT64_MAX) { | |
2564 | stats->rx_packets += ofproto->stats.tx_packets; | |
2565 | } | |
2566 | ||
2567 | if (stats->rx_bytes != UINT64_MAX) { | |
2568 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2569 | } | |
2570 | ||
2571 | /* ofproto->stats.rx_packets represents packets that were received on | |
2572 | * some port and we processed internally and dropped (e.g. STP). | |
2573 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2574 | ||
2575 | if (stats->tx_packets != UINT64_MAX) { | |
2576 | stats->tx_packets += ofproto->stats.rx_packets; | |
2577 | } | |
2578 | ||
2579 | if (stats->tx_bytes != UINT64_MAX) { | |
2580 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2581 | } | |
2582 | } | |
2583 | ||
2584 | return error; | |
2585 | } | |
2586 | ||
2587 | /* Account packets for LOCAL port. */ | |
2588 | static void | |
2589 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2590 | size_t tx_size, size_t rx_size) | |
2591 | { | |
2592 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2593 | ||
2594 | if (rx_size) { | |
2595 | ofproto->stats.rx_packets++; | |
2596 | ofproto->stats.rx_bytes += rx_size; | |
2597 | } | |
2598 | if (tx_size) { | |
2599 | ofproto->stats.tx_packets++; | |
2600 | ofproto->stats.tx_bytes += tx_size; | |
2601 | } | |
2602 | } | |
2603 | ||
abe529af BP |
2604 | struct port_dump_state { |
2605 | struct dpif_port_dump dump; | |
2606 | bool done; | |
2607 | }; | |
2608 | ||
2609 | static int | |
2610 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2611 | { | |
2612 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2613 | struct port_dump_state *state; | |
2614 | ||
2615 | *statep = state = xmalloc(sizeof *state); | |
2616 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2617 | state->done = false; | |
2618 | return 0; | |
2619 | } | |
2620 | ||
2621 | static int | |
2622 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2623 | struct ofproto_port *port) | |
2624 | { | |
2625 | struct port_dump_state *state = state_; | |
2626 | struct dpif_port dpif_port; | |
2627 | ||
2628 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2629 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2630 | return 0; | |
2631 | } else { | |
2632 | int error = dpif_port_dump_done(&state->dump); | |
2633 | state->done = true; | |
2634 | return error ? error : EOF; | |
2635 | } | |
2636 | } | |
2637 | ||
2638 | static int | |
2639 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2640 | { | |
2641 | struct port_dump_state *state = state_; | |
2642 | ||
2643 | if (!state->done) { | |
2644 | dpif_port_dump_done(&state->dump); | |
2645 | } | |
2646 | free(state); | |
2647 | return 0; | |
2648 | } | |
2649 | ||
2650 | static int | |
2651 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2652 | { | |
2653 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2654 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2655 | } | |
2656 | ||
2657 | static void | |
2658 | port_poll_wait(const struct ofproto *ofproto_) | |
2659 | { | |
2660 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2661 | dpif_port_poll_wait(ofproto->dpif); | |
2662 | } | |
2663 | ||
2664 | static int | |
2665 | port_is_lacp_current(const struct ofport *ofport_) | |
2666 | { | |
2667 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2668 | return (ofport->bundle && ofport->bundle->lacp | |
2669 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2670 | : -1); | |
2671 | } | |
2672 | \f | |
2673 | /* Upcall handling. */ | |
2674 | ||
501f8d1f BP |
2675 | /* Flow miss batching. |
2676 | * | |
2677 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2678 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2679 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2680 | * more packets, plus possibly installing the flow in the dpif. | |
2681 | * | |
2682 | * So far we only batch the operations that affect flow setup time the most. | |
2683 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2684 | struct flow_miss { | |
2685 | struct hmap_node hmap_node; | |
2686 | struct flow flow; | |
b0f7b9b5 | 2687 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2688 | const struct nlattr *key; |
2689 | size_t key_len; | |
e84173dc | 2690 | ovs_be16 initial_tci; |
501f8d1f | 2691 | struct list packets; |
6a7e895f | 2692 | enum dpif_upcall_type upcall_type; |
501f8d1f BP |
2693 | }; |
2694 | ||
2695 | struct flow_miss_op { | |
c2b565b5 | 2696 | struct dpif_op dpif_op; |
5fe20d5d BP |
2697 | struct subfacet *subfacet; /* Subfacet */ |
2698 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
2699 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
2700 | }; |
2701 | ||
62cd7072 BP |
2702 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2703 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2704 | * configurations. */ |
62cd7072 | 2705 | static void |
a39edbd4 | 2706 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 2707 | const struct flow *flow) |
62cd7072 BP |
2708 | { |
2709 | struct ofputil_packet_in pin; | |
2710 | ||
3e3252fa EJ |
2711 | pin.packet = packet->data; |
2712 | pin.packet_len = packet->size; | |
62cd7072 | 2713 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 2714 | pin.controller_id = 0; |
54834960 EJ |
2715 | |
2716 | pin.table_id = 0; | |
2717 | pin.cookie = 0; | |
2718 | ||
62cd7072 | 2719 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
2720 | |
2721 | flow_get_metadata(flow, &pin.fmd); | |
2722 | ||
2723 | /* Registers aren't meaningful on a miss. */ | |
2724 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2725 | ||
d8653c38 | 2726 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
2727 | } |
2728 | ||
6a7e895f | 2729 | static enum slow_path_reason |
abe529af BP |
2730 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, |
2731 | const struct ofpbuf *packet) | |
2732 | { | |
b6e001b6 EJ |
2733 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2734 | ||
2735 | if (!ofport) { | |
6a7e895f | 2736 | return 0; |
b6e001b6 EJ |
2737 | } |
2738 | ||
ef9819b5 | 2739 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2740 | if (packet) { |
abe529af BP |
2741 | cfm_process_heartbeat(ofport->cfm, packet); |
2742 | } | |
6a7e895f | 2743 | return SLOW_CFM; |
b6e001b6 EJ |
2744 | } else if (ofport->bundle && ofport->bundle->lacp |
2745 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2746 | if (packet) { | |
2747 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2748 | } |
6a7e895f | 2749 | return SLOW_LACP; |
21f7563c JP |
2750 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2751 | if (packet) { | |
2752 | stp_process_packet(ofport, packet); | |
2753 | } | |
6a7e895f | 2754 | return SLOW_STP; |
abe529af | 2755 | } |
6a7e895f | 2756 | return 0; |
abe529af BP |
2757 | } |
2758 | ||
501f8d1f | 2759 | static struct flow_miss * |
b23cdad9 | 2760 | flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) |
abe529af | 2761 | { |
501f8d1f | 2762 | struct flow_miss *miss; |
abe529af | 2763 | |
501f8d1f BP |
2764 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2765 | if (flow_equal(&miss->flow, flow)) { | |
2766 | return miss; | |
2767 | } | |
2768 | } | |
abe529af | 2769 | |
b23cdad9 | 2770 | return NULL; |
501f8d1f | 2771 | } |
abe529af | 2772 | |
9d6ac44e BP |
2773 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
2774 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
2775 | * 'miss' is associated with a subfacet the caller must also initialize the | |
2776 | * returned op->subfacet, and if anything needs to be freed after processing | |
2777 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 2778 | static void |
9d6ac44e BP |
2779 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
2780 | struct flow_miss_op *op) | |
501f8d1f | 2781 | { |
9d6ac44e BP |
2782 | if (miss->flow.vlan_tci != miss->initial_tci) { |
2783 | /* This packet was received on a VLAN splinter port. We | |
2784 | * added a VLAN to the packet to make the packet resemble | |
2785 | * the flow, but the actions were composed assuming that | |
2786 | * the packet contained no VLAN. So, we must remove the | |
2787 | * VLAN header from the packet before trying to execute the | |
2788 | * actions. */ | |
2789 | eth_pop_vlan(packet); | |
2790 | } | |
2791 | ||
2792 | op->subfacet = NULL; | |
2793 | op->garbage = NULL; | |
2794 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
2795 | op->dpif_op.u.execute.key = miss->key; | |
2796 | op->dpif_op.u.execute.key_len = miss->key_len; | |
2797 | op->dpif_op.u.execute.packet = packet; | |
2798 | } | |
2799 | ||
2800 | /* Helper for handle_flow_miss_without_facet() and | |
2801 | * handle_flow_miss_with_facet(). */ | |
2802 | static void | |
2803 | handle_flow_miss_common(struct rule_dpif *rule, | |
2804 | struct ofpbuf *packet, const struct flow *flow) | |
2805 | { | |
2806 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2807 | ||
2808 | ofproto->n_matches++; | |
2809 | ||
2810 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2811 | /* | |
2812 | * Extra-special case for fail-open mode. | |
2813 | * | |
2814 | * We are in fail-open mode and the packet matched the fail-open | |
2815 | * rule, but we are connected to a controller too. We should send | |
2816 | * the packet up to the controller in the hope that it will try to | |
2817 | * set up a flow and thereby allow us to exit fail-open. | |
2818 | * | |
2819 | * See the top-level comment in fail-open.c for more information. | |
2820 | */ | |
2821 | send_packet_in_miss(ofproto, packet, flow); | |
2822 | } | |
2823 | } | |
2824 | ||
2825 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
2826 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
2827 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
2828 | * true). However, for short flows the cost of bookkeeping is much higher than | |
2829 | * the benefits, so when the datapath holds a large number of flows we impose | |
2830 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
2831 | static bool | |
2832 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
2833 | struct flow_miss *miss, uint32_t hash) | |
2834 | { | |
2835 | if (!ofproto->governor) { | |
2836 | size_t n_subfacets; | |
2837 | ||
2838 | n_subfacets = hmap_count(&ofproto->subfacets); | |
2839 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
2840 | return true; | |
2841 | } | |
2842 | ||
2843 | ofproto->governor = governor_create(ofproto->up.name); | |
2844 | } | |
2845 | ||
2846 | return governor_should_install_flow(ofproto->governor, hash, | |
2847 | list_size(&miss->packets)); | |
2848 | } | |
2849 | ||
2850 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
2851 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
2852 | * increment '*n_ops'. */ | |
2853 | static void | |
2854 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
2855 | struct rule_dpif *rule, | |
2856 | struct flow_miss_op *ops, size_t *n_ops) | |
2857 | { | |
2858 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2859 | struct action_xlate_ctx ctx; | |
530a1d91 | 2860 | struct ofpbuf *packet; |
2b459b83 | 2861 | |
9d6ac44e BP |
2862 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2863 | struct flow_miss_op *op = &ops[*n_ops]; | |
2864 | struct dpif_flow_stats stats; | |
2865 | struct ofpbuf odp_actions; | |
abe529af | 2866 | |
9d6ac44e | 2867 | COVERAGE_INC(facet_suppress); |
501f8d1f | 2868 | |
9d6ac44e | 2869 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 2870 | |
9d6ac44e BP |
2871 | dpif_flow_stats_extract(&miss->flow, packet, &stats); |
2872 | rule_credit_stats(rule, &stats); | |
abe529af | 2873 | |
9d6ac44e BP |
2874 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, |
2875 | rule, 0, packet); | |
2876 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 2877 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
9d6ac44e | 2878 | &odp_actions); |
abe529af | 2879 | |
9d6ac44e BP |
2880 | if (odp_actions.size) { |
2881 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
2882 | ||
2883 | init_flow_miss_execute_op(miss, packet, op); | |
2884 | execute->actions = odp_actions.data; | |
2885 | execute->actions_len = odp_actions.size; | |
2886 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2887 | ||
2888 | (*n_ops)++; | |
2889 | } else { | |
2890 | ofpbuf_uninit(&odp_actions); | |
2891 | } | |
abe529af | 2892 | } |
9d6ac44e BP |
2893 | } |
2894 | ||
2895 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
2896 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2897 | static void | |
2898 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
2899 | struct flow_miss_op *ops, size_t *n_ops) | |
2900 | { | |
6a7e895f BP |
2901 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
2902 | enum subfacet_path want_path; | |
9d6ac44e BP |
2903 | struct subfacet *subfacet; |
2904 | struct ofpbuf *packet; | |
abe529af | 2905 | |
15baa734 | 2906 | subfacet = subfacet_create(facet, |
e84173dc BP |
2907 | miss->key_fitness, miss->key, miss->key_len, |
2908 | miss->initial_tci); | |
b0f7b9b5 | 2909 | |
530a1d91 | 2910 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 2911 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 2912 | struct dpif_flow_stats stats; |
5fe20d5d | 2913 | struct ofpbuf odp_actions; |
67d91f78 | 2914 | |
9d6ac44e | 2915 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 2916 | |
5fe20d5d | 2917 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
6a7e895f | 2918 | if (!subfacet->actions || subfacet->slow) { |
5fe20d5d | 2919 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 2920 | } |
67d91f78 | 2921 | |
67d91f78 | 2922 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
15baa734 | 2923 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2924 | |
9d6ac44e BP |
2925 | if (subfacet->actions_len) { |
2926 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 2927 | |
9d6ac44e BP |
2928 | init_flow_miss_execute_op(miss, packet, op); |
2929 | op->subfacet = subfacet; | |
6a7e895f | 2930 | if (!subfacet->slow) { |
9d6ac44e BP |
2931 | execute->actions = subfacet->actions; |
2932 | execute->actions_len = subfacet->actions_len; | |
2933 | ofpbuf_uninit(&odp_actions); | |
2934 | } else { | |
2935 | execute->actions = odp_actions.data; | |
2936 | execute->actions_len = odp_actions.size; | |
2937 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2938 | } | |
999fba59 | 2939 | |
9d6ac44e | 2940 | (*n_ops)++; |
5fe20d5d | 2941 | } else { |
9d6ac44e | 2942 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 2943 | } |
501f8d1f BP |
2944 | } |
2945 | ||
6a7e895f BP |
2946 | want_path = subfacet_want_path(subfacet->slow); |
2947 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
501f8d1f | 2948 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2949 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2950 | |
b0f7b9b5 | 2951 | op->subfacet = subfacet; |
5fe20d5d | 2952 | op->garbage = NULL; |
c2b565b5 | 2953 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2954 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2955 | put->key = miss->key; | |
2956 | put->key_len = miss->key_len; | |
6a7e895f BP |
2957 | if (want_path == SF_FAST_PATH) { |
2958 | put->actions = subfacet->actions; | |
2959 | put->actions_len = subfacet->actions_len; | |
2960 | } else { | |
2961 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
2962 | op->stub, sizeof op->stub, | |
2963 | &put->actions, &put->actions_len); | |
2964 | } | |
501f8d1f BP |
2965 | put->stats = NULL; |
2966 | } | |
2967 | } | |
2968 | ||
9d6ac44e BP |
2969 | /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath |
2970 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2971 | static void | |
2972 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2973 | struct flow_miss_op *ops, size_t *n_ops) | |
2974 | { | |
2975 | struct facet *facet; | |
2976 | uint32_t hash; | |
2977 | ||
2978 | /* The caller must ensure that miss->hmap_node.hash contains | |
2979 | * flow_hash(miss->flow, 0). */ | |
2980 | hash = miss->hmap_node.hash; | |
2981 | ||
2982 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
2983 | if (!facet) { | |
c57b2226 BP |
2984 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); |
2985 | ||
2986 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
9d6ac44e BP |
2987 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); |
2988 | return; | |
2989 | } | |
2990 | ||
2991 | facet = facet_create(rule, &miss->flow, hash); | |
2992 | } | |
2993 | handle_flow_miss_with_facet(miss, facet, ops, n_ops); | |
2994 | } | |
2995 | ||
e2a6ca36 BP |
2996 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
2997 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
2998 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
2999 | * what a flow key should contain. | |
3000 | * | |
3001 | * This function also includes some logic to help make VLAN splinters | |
3002 | * transparent to the rest of the upcall processing logic. In particular, if | |
3003 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
3004 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
3005 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
3006 | * | |
3007 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
3008 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
3009 | * (This differs from the value returned in flow->vlan_tci only for packets | |
3010 | * received on VLAN splinters.) | |
3011 | */ | |
e84173dc | 3012 | static enum odp_key_fitness |
52a90c29 | 3013 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 3014 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
3015 | struct flow *flow, ovs_be16 *initial_tci, |
3016 | struct ofpbuf *packet) | |
e84173dc BP |
3017 | { |
3018 | enum odp_key_fitness fitness; | |
3019 | ||
3020 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
3021 | if (fitness == ODP_FIT_ERROR) { | |
3022 | return fitness; | |
3023 | } | |
3024 | *initial_tci = flow->vlan_tci; | |
3025 | ||
b98d8985 | 3026 | if (vsp_adjust_flow(ofproto, flow)) { |
e2a6ca36 BP |
3027 | if (packet) { |
3028 | /* Make the packet resemble the flow, so that it gets sent to an | |
3029 | * OpenFlow controller properly, so that it looks correct for | |
3030 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
3031 | * if it is called on 'packet'. | |
3032 | * | |
3033 | * The allocated space inside 'packet' probably also contains | |
3034 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
3035 | * struct dpif_upcall (see the large comment on that structure | |
3036 | * definition), so pushing data on 'packet' is in general not a | |
3037 | * good idea since it could overwrite 'key' or free it as a side | |
3038 | * effect. However, it's OK in this special case because we know | |
3039 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
3040 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
3041 | * since we don't need that header anymore. */ | |
3042 | eth_push_vlan(packet, flow->vlan_tci); | |
3043 | } | |
52a90c29 BP |
3044 | |
3045 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
3046 | if (fitness == ODP_FIT_PERFECT) { | |
3047 | fitness = ODP_FIT_TOO_MUCH; | |
3048 | } | |
3049 | } | |
3050 | ||
e84173dc BP |
3051 | return fitness; |
3052 | } | |
3053 | ||
501f8d1f BP |
3054 | static void |
3055 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
3056 | size_t n_upcalls) | |
3057 | { | |
3058 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
3059 | struct flow_miss *miss; |
3060 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 3061 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 3062 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 3063 | struct hmap todo; |
b23cdad9 | 3064 | int n_misses; |
501f8d1f BP |
3065 | size_t n_ops; |
3066 | size_t i; | |
3067 | ||
3068 | if (!n_upcalls) { | |
3069 | return; | |
3070 | } | |
3071 | ||
3072 | /* Construct the to-do list. | |
3073 | * | |
3074 | * This just amounts to extracting the flow from each packet and sticking | |
3075 | * the packets that have the same flow in the same "flow_miss" structure so | |
3076 | * that we can process them together. */ | |
3077 | hmap_init(&todo); | |
b23cdad9 | 3078 | n_misses = 0; |
501f8d1f | 3079 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
3080 | struct flow_miss *miss = &misses[n_misses]; |
3081 | struct flow_miss *existing_miss; | |
3082 | uint32_t hash; | |
501f8d1f | 3083 | |
b0f7b9b5 BP |
3084 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
3085 | * then set 'flow''s header pointers. */ | |
b23cdad9 BP |
3086 | miss->key_fitness = ofproto_dpif_extract_flow_key( |
3087 | ofproto, upcall->key, upcall->key_len, | |
3088 | &miss->flow, &miss->initial_tci, upcall->packet); | |
3089 | if (miss->key_fitness == ODP_FIT_ERROR) { | |
b0f7b9b5 BP |
3090 | continue; |
3091 | } | |
b23cdad9 BP |
3092 | flow_extract(upcall->packet, miss->flow.skb_priority, |
3093 | miss->flow.tun_id, miss->flow.in_port, &miss->flow); | |
501f8d1f | 3094 | |
501f8d1f | 3095 | /* Add other packets to a to-do list. */ |
b23cdad9 BP |
3096 | hash = flow_hash(&miss->flow, 0); |
3097 | existing_miss = flow_miss_find(&todo, &miss->flow, hash); | |
3098 | if (!existing_miss) { | |
3099 | hmap_insert(&todo, &miss->hmap_node, hash); | |
3100 | miss->key = upcall->key; | |
3101 | miss->key_len = upcall->key_len; | |
6a7e895f | 3102 | miss->upcall_type = upcall->type; |
b23cdad9 BP |
3103 | list_init(&miss->packets); |
3104 | ||
3105 | n_misses++; | |
3106 | } else { | |
3107 | miss = existing_miss; | |
3108 | } | |
501f8d1f BP |
3109 | list_push_back(&miss->packets, &upcall->packet->list_node); |
3110 | } | |
3111 | ||
3112 | /* Process each element in the to-do list, constructing the set of | |
3113 | * operations to batch. */ | |
3114 | n_ops = 0; | |
33bb0caa | 3115 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
501f8d1f | 3116 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); |
abe529af | 3117 | } |
501f8d1f | 3118 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
3119 | |
3120 | /* Execute batch. */ | |
3121 | for (i = 0; i < n_ops; i++) { | |
3122 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3123 | } | |
3124 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
3125 | ||
3126 | /* Free memory and update facets. */ | |
3127 | for (i = 0; i < n_ops; i++) { | |
3128 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
501f8d1f BP |
3129 | |
3130 | switch (op->dpif_op.type) { | |
3131 | case DPIF_OP_EXECUTE: | |
501f8d1f | 3132 | break; |
abe529af | 3133 | |
501f8d1f | 3134 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 3135 | if (!op->dpif_op.error) { |
6a7e895f | 3136 | op->subfacet->path = subfacet_want_path(op->subfacet->slow); |
501f8d1f BP |
3137 | } |
3138 | break; | |
b99d3cee BP |
3139 | |
3140 | case DPIF_OP_FLOW_DEL: | |
3141 | NOT_REACHED(); | |
501f8d1f | 3142 | } |
5fe20d5d BP |
3143 | |
3144 | free(op->garbage); | |
501f8d1f | 3145 | } |
33bb0caa | 3146 | hmap_destroy(&todo); |
abe529af BP |
3147 | } |
3148 | ||
6a7e895f BP |
3149 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } |
3150 | classify_upcall(const struct dpif_upcall *upcall) | |
3151 | { | |
3152 | union user_action_cookie cookie; | |
3153 | ||
3154 | /* First look at the upcall type. */ | |
3155 | switch (upcall->type) { | |
3156 | case DPIF_UC_ACTION: | |
3157 | break; | |
3158 | ||
3159 | case DPIF_UC_MISS: | |
3160 | return MISS_UPCALL; | |
3161 | ||
3162 | case DPIF_N_UC_TYPES: | |
3163 | default: | |
3164 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3165 | return BAD_UPCALL; | |
3166 | } | |
3167 | ||
3168 | /* "action" upcalls need a closer look. */ | |
3169 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); | |
3170 | switch (cookie.type) { | |
3171 | case USER_ACTION_COOKIE_SFLOW: | |
3172 | return SFLOW_UPCALL; | |
3173 | ||
3174 | case USER_ACTION_COOKIE_SLOW_PATH: | |
3175 | return MISS_UPCALL; | |
3176 | ||
3177 | case USER_ACTION_COOKIE_UNSPEC: | |
3178 | default: | |
3179 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
3180 | return BAD_UPCALL; | |
3181 | } | |
3182 | } | |
3183 | ||
abe529af | 3184 | static void |
6a7e895f BP |
3185 | handle_sflow_upcall(struct ofproto_dpif *ofproto, |
3186 | const struct dpif_upcall *upcall) | |
abe529af | 3187 | { |
1673e0e4 | 3188 | union user_action_cookie cookie; |
e84173dc BP |
3189 | enum odp_key_fitness fitness; |
3190 | ovs_be16 initial_tci; | |
3191 | struct flow flow; | |
abe529af | 3192 | |
e84173dc BP |
3193 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
3194 | upcall->key_len, &flow, | |
e2a6ca36 | 3195 | &initial_tci, upcall->packet); |
e84173dc BP |
3196 | if (fitness == ODP_FIT_ERROR) { |
3197 | return; | |
3198 | } | |
3199 | ||
6a7e895f BP |
3200 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
3201 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); | |
6ff686f2 PS |
3202 | } |
3203 | ||
9b16c439 BP |
3204 | static int |
3205 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 3206 | { |
9b16c439 | 3207 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3208 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3209 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3210 | int n_processed; | |
9b16c439 BP |
3211 | int n_misses; |
3212 | int i; | |
abe529af | 3213 | |
90a7c55e | 3214 | assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3215 | |
9b16c439 | 3216 | n_misses = 0; |
90a7c55e | 3217 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3218 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3219 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3220 | int error; |
3221 | ||
90a7c55e BP |
3222 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3223 | sizeof miss_buf_stubs[n_misses]); | |
3224 | error = dpif_recv(ofproto->dpif, upcall, buf); | |
9b16c439 | 3225 | if (error) { |
90a7c55e | 3226 | ofpbuf_uninit(buf); |
9b16c439 BP |
3227 | break; |
3228 | } | |
3229 | ||
6a7e895f BP |
3230 | switch (classify_upcall(upcall)) { |
3231 | case MISS_UPCALL: | |
9b16c439 BP |
3232 | /* Handle it later. */ |
3233 | n_misses++; | |
3234 | break; | |
3235 | ||
6a7e895f BP |
3236 | case SFLOW_UPCALL: |
3237 | if (ofproto->sflow) { | |
3238 | handle_sflow_upcall(ofproto, upcall); | |
3239 | } | |
3240 | ofpbuf_uninit(buf); | |
3241 | break; | |
3242 | ||
3243 | case BAD_UPCALL: | |
3244 | ofpbuf_uninit(buf); | |
9b16c439 BP |
3245 | break; |
3246 | } | |
abe529af | 3247 | } |
9b16c439 | 3248 | |
6a7e895f | 3249 | /* Handle deferred MISS_UPCALL processing. */ |
9b16c439 | 3250 | handle_miss_upcalls(ofproto, misses, n_misses); |
90a7c55e BP |
3251 | for (i = 0; i < n_misses; i++) { |
3252 | ofpbuf_uninit(&miss_bufs[i]); | |
3253 | } | |
9b16c439 | 3254 | |
90a7c55e | 3255 | return n_processed; |
abe529af BP |
3256 | } |
3257 | \f | |
3258 | /* Flow expiration. */ | |
3259 | ||
b0f7b9b5 | 3260 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
3261 | static void update_stats(struct ofproto_dpif *); |
3262 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 3263 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
3264 | |
3265 | /* This function is called periodically by run(). Its job is to collect | |
3266 | * updates for the flows that have been installed into the datapath, most | |
3267 | * importantly when they last were used, and then use that information to | |
3268 | * expire flows that have not been used recently. | |
3269 | * | |
3270 | * Returns the number of milliseconds after which it should be called again. */ | |
3271 | static int | |
3272 | expire(struct ofproto_dpif *ofproto) | |
3273 | { | |
3274 | struct rule_dpif *rule, *next_rule; | |
d0918789 | 3275 | struct oftable *table; |
abe529af BP |
3276 | int dp_max_idle; |
3277 | ||
3278 | /* Update stats for each flow in the datapath. */ | |
3279 | update_stats(ofproto); | |
3280 | ||
b0f7b9b5 BP |
3281 | /* Expire subfacets that have been idle too long. */ |
3282 | dp_max_idle = subfacet_max_idle(ofproto); | |
3283 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
3284 | |
3285 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
3286 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
3287 | struct cls_cursor cursor; | |
3288 | ||
d0918789 | 3289 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
3290 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
3291 | rule_expire(rule); | |
3292 | } | |
abe529af BP |
3293 | } |
3294 | ||
3295 | /* All outstanding data in existing flows has been accounted, so it's a | |
3296 | * good time to do bond rebalancing. */ | |
3297 | if (ofproto->has_bonded_bundles) { | |
3298 | struct ofbundle *bundle; | |
3299 | ||
3300 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3301 | if (bundle->bond) { | |
3302 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
3303 | } | |
3304 | } | |
3305 | } | |
3306 | ||
3307 | return MIN(dp_max_idle, 1000); | |
3308 | } | |
3309 | ||
a218c879 BP |
3310 | /* Updates flow table statistics given that the datapath just reported 'stats' |
3311 | * as 'subfacet''s statistics. */ | |
3312 | static void | |
3313 | update_subfacet_stats(struct subfacet *subfacet, | |
3314 | const struct dpif_flow_stats *stats) | |
3315 | { | |
3316 | struct facet *facet = subfacet->facet; | |
3317 | ||
3318 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
3319 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
3320 | facet->packet_count += extra; | |
3321 | } else { | |
3322 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
3323 | } | |
3324 | ||
3325 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
3326 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
3327 | } else { | |
3328 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
3329 | } | |
3330 | ||
3331 | subfacet->dp_packet_count = stats->n_packets; | |
3332 | subfacet->dp_byte_count = stats->n_bytes; | |
3333 | ||
3334 | facet->tcp_flags |= stats->tcp_flags; | |
3335 | ||
3336 | subfacet_update_time(subfacet, stats->used); | |
3337 | if (facet->accounted_bytes < facet->byte_count) { | |
3338 | facet_learn(facet); | |
3339 | facet_account(facet); | |
3340 | facet->accounted_bytes = facet->byte_count; | |
3341 | } | |
3342 | facet_push_stats(facet); | |
3343 | } | |
3344 | ||
3345 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
3346 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
3347 | static void | |
3348 | delete_unexpected_flow(struct dpif *dpif, | |
3349 | const struct nlattr *key, size_t key_len) | |
3350 | { | |
3351 | if (!VLOG_DROP_WARN(&rl)) { | |
3352 | struct ds s; | |
3353 | ||
3354 | ds_init(&s); | |
3355 | odp_flow_key_format(key, key_len, &s); | |
3356 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
3357 | ds_destroy(&s); | |
3358 | } | |
3359 | ||
3360 | COVERAGE_INC(facet_unexpected); | |
3361 | dpif_flow_del(dpif, key, key_len, NULL); | |
3362 | } | |
3363 | ||
abe529af BP |
3364 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
3365 | * | |
3366 | * This function also pushes statistics updates to rules which each facet | |
3367 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3368 | * facet changes the rule it resubmits into at some time in between | |
3369 | * update_stats() runs, it is possible that statistics accrued to the | |
3370 | * old rule will be incorrectly attributed to the new rule. This could be | |
3371 | * avoided by calling update_stats() whenever rules are created or | |
3372 | * deleted. However, the performance impact of making so many calls to the | |
3373 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
3374 | */ | |
3375 | static void | |
3376 | update_stats(struct ofproto_dpif *p) | |
3377 | { | |
3378 | const struct dpif_flow_stats *stats; | |
3379 | struct dpif_flow_dump dump; | |
3380 | const struct nlattr *key; | |
3381 | size_t key_len; | |
3382 | ||
3383 | dpif_flow_dump_start(&dump, p->dpif); | |
3384 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 3385 | struct subfacet *subfacet; |
abe529af | 3386 | |
6a542738 | 3387 | subfacet = subfacet_find(p, key, key_len); |
6a7e895f BP |
3388 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { |
3389 | case SF_FAST_PATH: | |
a218c879 | 3390 | update_subfacet_stats(subfacet, stats); |
6a7e895f BP |
3391 | break; |
3392 | ||
3393 | case SF_SLOW_PATH: | |
3394 | /* Stats are updated per-packet. */ | |
3395 | break; | |
3396 | ||
3397 | case SF_NOT_INSTALLED: | |
3398 | default: | |
a218c879 | 3399 | delete_unexpected_flow(p->dpif, key, key_len); |
6a7e895f | 3400 | break; |
abe529af BP |
3401 | } |
3402 | } | |
3403 | dpif_flow_dump_done(&dump); | |
3404 | } | |
3405 | ||
3406 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3407 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3408 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3409 | * fold its statistic into its rule. */ | |
abe529af | 3410 | static int |
b0f7b9b5 | 3411 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
3412 | { |
3413 | /* | |
3414 | * Idle time histogram. | |
3415 | * | |
b0f7b9b5 BP |
3416 | * Most of the time a switch has a relatively small number of subfacets. |
3417 | * When this is the case we might as well keep statistics for all of them | |
3418 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3419 | * well. |
3420 | * | |
b0f7b9b5 | 3421 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3422 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3423 | * significant. However, with a large number of subfacets it is likely |
3424 | * that only a few of them are "heavy hitters" that consume a large amount | |
3425 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3426 | * the kernel and maintaining in userspaces; other subfacets we can | |
3427 | * discard. | |
abe529af BP |
3428 | * |
3429 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3430 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3431 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3432 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3433 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 3434 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3435 | * the most-recently-used bucket of subfacets is kept, so actually an |
3436 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3437 | * (though the next run will delete most of those unless they receive |
3438 | * additional data). | |
abe529af | 3439 | * |
b0f7b9b5 BP |
3440 | * This requires a second pass through the subfacets, in addition to the |
3441 | * pass made by update_stats(), because the former function never looks at | |
3442 | * uninstallable subfacets. | |
abe529af BP |
3443 | */ |
3444 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3445 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3446 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3447 | int total, subtotal, bucket; |
b0f7b9b5 | 3448 | struct subfacet *subfacet; |
abe529af BP |
3449 | long long int now; |
3450 | int i; | |
3451 | ||
b0f7b9b5 | 3452 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3453 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3454 | return N_BUCKETS * BUCKET_WIDTH; |
3455 | } | |
3456 | ||
3457 | /* Build histogram. */ | |
3458 | now = time_msec(); | |
b0f7b9b5 BP |
3459 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3460 | long long int idle = now - subfacet->used; | |
abe529af BP |
3461 | int bucket = (idle <= 0 ? 0 |
3462 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3463 | : (unsigned int) idle / BUCKET_WIDTH); | |
3464 | buckets[bucket]++; | |
3465 | } | |
3466 | ||
3467 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3468 | subtotal = bucket = 0; |
3469 | do { | |
3470 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3471 | } while (bucket < N_BUCKETS && |
3472 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3473 | |
3474 | if (VLOG_IS_DBG_ENABLED()) { | |
3475 | struct ds s; | |
3476 | ||
3477 | ds_init(&s); | |
3478 | ds_put_cstr(&s, "keep"); | |
3479 | for (i = 0; i < N_BUCKETS; i++) { | |
3480 | if (i == bucket) { | |
3481 | ds_put_cstr(&s, ", drop"); | |
3482 | } | |
3483 | if (buckets[i]) { | |
3484 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3485 | } | |
3486 | } | |
3487 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3488 | ds_destroy(&s); | |
3489 | } | |
3490 | ||
3491 | return bucket * BUCKET_WIDTH; | |
3492 | } | |
3493 | ||
b99d3cee BP |
3494 | enum { EXPIRE_MAX_BATCH = 50 }; |
3495 | ||
3496 | static void | |
3497 | expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) | |
3498 | { | |
3499 | struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; | |
3500 | struct dpif_op ops[EXPIRE_MAX_BATCH]; | |
3501 | struct dpif_op *opsp[EXPIRE_MAX_BATCH]; | |
3502 | struct ofpbuf keys[EXPIRE_MAX_BATCH]; | |
3503 | struct dpif_flow_stats stats[EXPIRE_MAX_BATCH]; | |
3504 | int i; | |
3505 | ||
3506 | for (i = 0; i < n; i++) { | |
3507 | ops[i].type = DPIF_OP_FLOW_DEL; | |
3508 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
3509 | ops[i].u.flow_del.key = keys[i].data; | |
3510 | ops[i].u.flow_del.key_len = keys[i].size; | |
3511 | ops[i].u.flow_del.stats = &stats[i]; | |
3512 | opsp[i] = &ops[i]; | |
3513 | } | |
3514 | ||
3515 | dpif_operate(ofproto->dpif, opsp, n); | |
3516 | for (i = 0; i < n; i++) { | |
3517 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
6a7e895f | 3518 | subfacets[i]->path = SF_NOT_INSTALLED; |
b99d3cee BP |
3519 | subfacet_destroy(subfacets[i]); |
3520 | } | |
3521 | } | |
3522 | ||
abe529af | 3523 | static void |
b0f7b9b5 | 3524 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af | 3525 | { |
625b0720 BP |
3526 | /* Cutoff time for most flows. */ |
3527 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
3528 | ||
3529 | /* We really want to keep flows for special protocols around, so use a more | |
3530 | * conservative cutoff. */ | |
3531 | long long int special_cutoff = time_msec() - 10000; | |
b99d3cee | 3532 | |
b0f7b9b5 | 3533 | struct subfacet *subfacet, *next_subfacet; |
b99d3cee BP |
3534 | struct subfacet *batch[EXPIRE_MAX_BATCH]; |
3535 | int n_batch; | |
abe529af | 3536 | |
b99d3cee | 3537 | n_batch = 0; |
b0f7b9b5 BP |
3538 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3539 | &ofproto->subfacets) { | |
625b0720 BP |
3540 | long long int cutoff; |
3541 | ||
3542 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) | |
3543 | ? special_cutoff | |
3544 | : normal_cutoff); | |
b0f7b9b5 | 3545 | if (subfacet->used < cutoff) { |
6a7e895f | 3546 | if (subfacet->path != SF_NOT_INSTALLED) { |
b99d3cee BP |
3547 | batch[n_batch++] = subfacet; |
3548 | if (n_batch >= EXPIRE_MAX_BATCH) { | |
3549 | expire_batch(ofproto, batch, n_batch); | |
3550 | n_batch = 0; | |
3551 | } | |
3552 | } else { | |
3553 | subfacet_destroy(subfacet); | |
3554 | } | |
abe529af BP |
3555 | } |
3556 | } | |
b99d3cee BP |
3557 | |
3558 | if (n_batch > 0) { | |
3559 | expire_batch(ofproto, batch, n_batch); | |
3560 | } | |
abe529af BP |
3561 | } |
3562 | ||
3563 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3564 | * then delete it entirely. */ | |
3565 | static void | |
3566 | rule_expire(struct rule_dpif *rule) | |
3567 | { | |
abe529af BP |
3568 | struct facet *facet, *next_facet; |
3569 | long long int now; | |
3570 | uint8_t reason; | |
3571 | ||
e2a3d183 BP |
3572 | if (rule->up.pending) { |
3573 | /* We'll have to expire it later. */ | |
3574 | return; | |
3575 | } | |
3576 | ||
abe529af BP |
3577 | /* Has 'rule' expired? */ |
3578 | now = time_msec(); | |
3579 | if (rule->up.hard_timeout | |
308881af | 3580 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 3581 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 3582 | } else if (rule->up.idle_timeout |
1745cd08 | 3583 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
3584 | reason = OFPRR_IDLE_TIMEOUT; |
3585 | } else { | |
3586 | return; | |
3587 | } | |
3588 | ||
3589 | COVERAGE_INC(ofproto_dpif_expired); | |
3590 | ||
3591 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3592 | * timeout, because that only happens when the rule has no facets left.) */ | |
3593 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3594 | facet_remove(facet); |
abe529af BP |
3595 | } |
3596 | ||
3597 | /* Get rid of the rule. */ | |
3598 | ofproto_rule_expire(&rule->up, reason); | |
3599 | } | |
3600 | \f | |
3601 | /* Facets. */ | |
3602 | ||
f3827897 | 3603 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3604 | * |
3605 | * The caller must already have determined that no facet with an identical | |
3606 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3607 | * the ofproto's classifier table. |
3608 | * | |
2b459b83 BP |
3609 | * 'hash' must be the return value of flow_hash(flow, 0). |
3610 | * | |
b0f7b9b5 BP |
3611 | * The facet will initially have no subfacets. The caller should create (at |
3612 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3613 | static struct facet * |
2b459b83 | 3614 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
3615 | { |
3616 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3617 | struct facet *facet; | |
3618 | ||
3619 | facet = xzalloc(sizeof *facet); | |
3620 | facet->used = time_msec(); | |
2b459b83 | 3621 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
3622 | list_push_back(&rule->facets, &facet->list_node); |
3623 | facet->rule = rule; | |
3624 | facet->flow = *flow; | |
b0f7b9b5 | 3625 | list_init(&facet->subfacets); |
abe529af BP |
3626 | netflow_flow_init(&facet->nf_flow); |
3627 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3628 | ||
abe529af BP |
3629 | return facet; |
3630 | } | |
3631 | ||
3632 | static void | |
3633 | facet_free(struct facet *facet) | |
3634 | { | |
abe529af BP |
3635 | free(facet); |
3636 | } | |
3637 | ||
3d9e05f8 BP |
3638 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3639 | * 'packet', which arrived on 'in_port'. | |
3640 | * | |
3641 | * Takes ownership of 'packet'. */ | |
3642 | static bool | |
3643 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3644 | const struct nlattr *odp_actions, size_t actions_len, | |
3645 | struct ofpbuf *packet) | |
3646 | { | |
3647 | struct odputil_keybuf keybuf; | |
3648 | struct ofpbuf key; | |
3649 | int error; | |
3650 | ||
6ff686f2 PS |
3651 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3652 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3653 | |
6ff686f2 PS |
3654 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3655 | odp_actions, actions_len, packet); | |
80e5eed9 | 3656 | |
6ff686f2 PS |
3657 | ofpbuf_delete(packet); |
3658 | return !error; | |
abe529af BP |
3659 | } |
3660 | ||
abe529af BP |
3661 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3662 | * | |
3663 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3664 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3665 | * |
3666 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3667 | */ | |
3668 | static void | |
15baa734 | 3669 | facet_remove(struct facet *facet) |
abe529af | 3670 | { |
15baa734 | 3671 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3672 | struct subfacet *subfacet, *next_subfacet; |
3673 | ||
551a2f6c BP |
3674 | assert(!list_is_empty(&facet->subfacets)); |
3675 | ||
3676 | /* First uninstall all of the subfacets to get final statistics. */ | |
3677 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3678 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3679 | } |
3680 | ||
3681 | /* Flush the final stats to the rule. | |
3682 | * | |
3683 | * This might require us to have at least one subfacet around so that we | |
3684 | * can use its actions for accounting in facet_account(), which is why we | |
3685 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3686 | facet_flush_stats(facet); |
551a2f6c BP |
3687 | |
3688 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3689 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3690 | &facet->subfacets) { | |
15baa734 | 3691 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3692 | } |
abe529af BP |
3693 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3694 | list_remove(&facet->list_node); | |
3695 | facet_free(facet); | |
3696 | } | |
3697 | ||
3de9590b BP |
3698 | /* Feed information from 'facet' back into the learning table to keep it in |
3699 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3700 | static void |
3de9590b | 3701 | facet_learn(struct facet *facet) |
abe529af | 3702 | { |
15baa734 | 3703 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3de9590b | 3704 | struct action_xlate_ctx ctx; |
abe529af | 3705 | |
3de9590b BP |
3706 | if (!facet->has_learn |
3707 | && !facet->has_normal | |
3708 | && (!facet->has_fin_timeout | |
3709 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
3710 | return; |
3711 | } | |
abe529af | 3712 | |
3de9590b BP |
3713 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
3714 | facet->flow.vlan_tci, | |
3715 | facet->rule, facet->tcp_flags, NULL); | |
3716 | ctx.may_learn = true; | |
f25d0cf3 BP |
3717 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, |
3718 | facet->rule->up.ofpacts_len); | |
3de9590b BP |
3719 | } |
3720 | ||
3721 | static void | |
3722 | facet_account(struct facet *facet) | |
3723 | { | |
3724 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3725 | struct subfacet *subfacet; | |
3726 | const struct nlattr *a; | |
3727 | unsigned int left; | |
3728 | ovs_be16 vlan_tci; | |
3729 | uint64_t n_bytes; | |
abe529af | 3730 | |
75a75043 | 3731 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3732 | return; |
3733 | } | |
3de9590b | 3734 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3735 | |
3736 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3737 | * as a basis. We also need to track the actual VLAN on which the packet | |
3738 | * is going to be sent to ensure that it matches the one passed to | |
3739 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3740 | * hash bucket.) |
3741 | * | |
3742 | * We use the actions from an arbitrary subfacet because they should all | |
3743 | * be equally valid for our purpose. */ | |
3744 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3745 | struct subfacet, list_node); | |
d78be13b | 3746 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3747 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3748 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3749 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3750 | struct ofport_dpif *port; |
abe529af | 3751 | |
d78be13b | 3752 | switch (nl_attr_type(a)) { |
df2c07f4 | 3753 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3754 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3755 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3756 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3757 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3758 | } |
d78be13b BP |
3759 | break; |
3760 | ||
fea393b1 BP |
3761 | case OVS_ACTION_ATTR_POP_VLAN: |
3762 | vlan_tci = htons(0); | |
d78be13b BP |
3763 | break; |
3764 | ||
fea393b1 BP |
3765 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3766 | vlan = nl_attr_get(a); | |
3767 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3768 | break; |
abe529af BP |
3769 | } |
3770 | } | |
3771 | } | |
3772 | ||
abe529af BP |
3773 | /* Returns true if the only action for 'facet' is to send to the controller. |
3774 | * (We don't report NetFlow expiration messages for such facets because they | |
3775 | * are just part of the control logic for the network, not real traffic). */ | |
3776 | static bool | |
3777 | facet_is_controller_flow(struct facet *facet) | |
3778 | { | |
f25d0cf3 BP |
3779 | if (facet) { |
3780 | const struct rule *rule = &facet->rule->up; | |
3781 | const struct ofpact *ofpacts = rule->ofpacts; | |
3782 | size_t ofpacts_len = rule->ofpacts_len; | |
3783 | ||
3784 | if (ofpacts->type == OFPACT_CONTROLLER && | |
3785 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { | |
3786 | return true; | |
3787 | } | |
3788 | } | |
3789 | return false; | |
abe529af BP |
3790 | } |
3791 | ||
3792 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3793 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3794 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3795 | * its packet and byte counts before this function is called. */ | |
3796 | static void | |
15baa734 | 3797 | facet_flush_stats(struct facet *facet) |
abe529af | 3798 | { |
15baa734 | 3799 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3800 | struct subfacet *subfacet; |
3801 | ||
3802 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3803 | assert(!subfacet->dp_byte_count); | |
3804 | assert(!subfacet->dp_packet_count); | |
3805 | } | |
abe529af BP |
3806 | |
3807 | facet_push_stats(facet); | |
3de9590b BP |
3808 | if (facet->accounted_bytes < facet->byte_count) { |
3809 | facet_account(facet); | |
3810 | facet->accounted_bytes = facet->byte_count; | |
3811 | } | |
abe529af BP |
3812 | |
3813 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3814 | struct ofexpired expired; | |
3815 | expired.flow = facet->flow; | |
3816 | expired.packet_count = facet->packet_count; | |
3817 | expired.byte_count = facet->byte_count; | |
3818 | expired.used = facet->used; | |
3819 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3820 | } | |
3821 | ||
3822 | facet->rule->packet_count += facet->packet_count; | |
3823 | facet->rule->byte_count += facet->byte_count; | |
3824 | ||
3825 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3826 | * reinstalled. */ | |
bbb5d219 | 3827 | facet_reset_counters(facet); |
abe529af BP |
3828 | |
3829 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 3830 | facet->tcp_flags = 0; |
abe529af BP |
3831 | } |
3832 | ||
3833 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3834 | * Returns it if found, otherwise a null pointer. | |
3835 | * | |
2b459b83 BP |
3836 | * 'hash' must be the return value of flow_hash(flow, 0). |
3837 | * | |
abe529af BP |
3838 | * The returned facet might need revalidation; use facet_lookup_valid() |
3839 | * instead if that is important. */ | |
3840 | static struct facet * | |
2b459b83 BP |
3841 | facet_find(struct ofproto_dpif *ofproto, |
3842 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
3843 | { |
3844 | struct facet *facet; | |
3845 | ||
2b459b83 | 3846 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
3847 | if (flow_equal(flow, &facet->flow)) { |
3848 | return facet; | |
3849 | } | |
3850 | } | |
3851 | ||
3852 | return NULL; | |
3853 | } | |
3854 | ||
3855 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3856 | * Returns it if found, otherwise a null pointer. | |
3857 | * | |
2b459b83 BP |
3858 | * 'hash' must be the return value of flow_hash(flow, 0). |
3859 | * | |
abe529af BP |
3860 | * The returned facet is guaranteed to be valid. */ |
3861 | static struct facet * | |
2b459b83 BP |
3862 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
3863 | uint32_t hash) | |
abe529af | 3864 | { |
c57b2226 | 3865 | struct facet *facet; |
abe529af | 3866 | |
c57b2226 | 3867 | facet = facet_find(ofproto, flow, hash); |
abe529af | 3868 | if (facet |
0e4b3771 | 3869 | && (ofproto->need_revalidate |
c57b2226 BP |
3870 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { |
3871 | facet_revalidate(facet); | |
abe529af BP |
3872 | } |
3873 | ||
3874 | return facet; | |
3875 | } | |
3876 | ||
6a7e895f BP |
3877 | static const char * |
3878 | subfacet_path_to_string(enum subfacet_path path) | |
3879 | { | |
3880 | switch (path) { | |
3881 | case SF_NOT_INSTALLED: | |
3882 | return "not installed"; | |
3883 | case SF_FAST_PATH: | |
3884 | return "in fast path"; | |
3885 | case SF_SLOW_PATH: | |
3886 | return "in slow path"; | |
3887 | default: | |
3888 | return "<error>"; | |
3889 | } | |
3890 | } | |
3891 | ||
3892 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
3893 | * member has the specified value. */ | |
3894 | static enum subfacet_path | |
3895 | subfacet_want_path(enum slow_path_reason slow) | |
3896 | { | |
3897 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
3898 | } | |
3899 | ||
3900 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
3901 | * supposing that its actions have been recalculated as 'want_actions' and that | |
3902 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
3903 | static bool | |
3904 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
3905 | const struct ofpbuf *want_actions) | |
3906 | { | |
3907 | enum subfacet_path want_path = subfacet_want_path(slow); | |
3908 | return (want_path != subfacet->path | |
3909 | || (want_path == SF_FAST_PATH | |
3910 | && (subfacet->actions_len != want_actions->size | |
3911 | || memcmp(subfacet->actions, want_actions->data, | |
3912 | subfacet->actions_len)))); | |
3913 | } | |
3914 | ||
6814e51f BP |
3915 | static bool |
3916 | facet_check_consistency(struct facet *facet) | |
3917 | { | |
3918 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3919 | ||
3920 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3921 | ||
050ac423 BP |
3922 | uint64_t odp_actions_stub[1024 / 8]; |
3923 | struct ofpbuf odp_actions; | |
3924 | ||
6814e51f BP |
3925 | struct rule_dpif *rule; |
3926 | struct subfacet *subfacet; | |
c53e1132 | 3927 | bool may_log = false; |
6814e51f BP |
3928 | bool ok; |
3929 | ||
3930 | /* Check the rule for consistency. */ | |
c57b2226 BP |
3931 | rule = rule_dpif_lookup(ofproto, &facet->flow); |
3932 | ok = rule == facet->rule; | |
3933 | if (!ok) { | |
c53e1132 | 3934 | may_log = !VLOG_DROP_WARN(&rl); |
c53e1132 BP |
3935 | if (may_log) { |
3936 | struct ds s; | |
6814e51f | 3937 | |
c53e1132 BP |
3938 | ds_init(&s); |
3939 | flow_format(&s, &facet->flow); | |
3940 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3941 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3942 | cls_rule_format(&facet->rule->up.cr, &s); | |
3943 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3944 | rule->up.table_id); | |
3945 | cls_rule_format(&rule->up.cr, &s); | |
3946 | ds_put_char(&s, ')'); | |
6814e51f | 3947 | |
c53e1132 BP |
3948 | VLOG_WARN("%s", ds_cstr(&s)); |
3949 | ds_destroy(&s); | |
3950 | } | |
6814e51f BP |
3951 | } |
3952 | ||
3953 | /* Check the datapath actions for consistency. */ | |
050ac423 | 3954 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f | 3955 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 3956 | enum subfacet_path want_path; |
9616614b | 3957 | struct odputil_keybuf keybuf; |
6814e51f | 3958 | struct action_xlate_ctx ctx; |
9616614b BP |
3959 | struct ofpbuf key; |
3960 | struct ds s; | |
6814e51f BP |
3961 | |
3962 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
0e553d9c | 3963 | subfacet->initial_tci, rule, 0, NULL); |
f25d0cf3 | 3964 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 3965 | &odp_actions); |
6814e51f | 3966 | |
6a7e895f BP |
3967 | if (subfacet->path == SF_NOT_INSTALLED) { |
3968 | /* This only happens if the datapath reported an error when we | |
3969 | * tried to install the flow. Don't flag another error here. */ | |
3970 | continue; | |
3971 | } | |
3972 | ||
3973 | want_path = subfacet_want_path(subfacet->slow); | |
3974 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
3975 | /* The actions for slow-path flows may legitimately vary from one | |
3976 | * packet to the next. We're done. */ | |
050ac423 | 3977 | continue; |
6814e51f BP |
3978 | } |
3979 | ||
6a7e895f | 3980 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { |
9616614b BP |
3981 | continue; |
3982 | } | |
c53e1132 | 3983 | |
9616614b BP |
3984 | /* Inconsistency! */ |
3985 | if (ok) { | |
3986 | may_log = !VLOG_DROP_WARN(&rl); | |
3987 | ok = false; | |
3988 | } | |
3989 | if (!may_log) { | |
3990 | /* Rate-limited, skip reporting. */ | |
3991 | continue; | |
3992 | } | |
c53e1132 | 3993 | |
9616614b BP |
3994 | ds_init(&s); |
3995 | subfacet_get_key(subfacet, &keybuf, &key); | |
3996 | odp_flow_key_format(key.data, key.size, &s); | |
3997 | ||
3998 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
6a7e895f | 3999 | if (want_path != subfacet->path) { |
9616614b BP |
4000 | enum odp_key_fitness fitness = subfacet->key_fitness; |
4001 | ||
6a7e895f BP |
4002 | ds_put_format(&s, " (%s, fitness=%s)", |
4003 | subfacet_path_to_string(subfacet->path), | |
9616614b | 4004 | odp_key_fitness_to_string(fitness)); |
6a7e895f BP |
4005 | ds_put_format(&s, " (should have been %s)", |
4006 | subfacet_path_to_string(want_path)); | |
4007 | } else if (want_path == SF_FAST_PATH) { | |
9616614b BP |
4008 | ds_put_cstr(&s, " (actions were: "); |
4009 | format_odp_actions(&s, subfacet->actions, | |
4010 | subfacet->actions_len); | |
4011 | ds_put_cstr(&s, ") (correct actions: "); | |
4012 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
4013 | ds_put_char(&s, ')'); | |
4014 | } else { | |
4015 | ds_put_cstr(&s, " (actions: "); | |
4016 | format_odp_actions(&s, subfacet->actions, | |
4017 | subfacet->actions_len); | |
4018 | ds_put_char(&s, ')'); | |
6814e51f | 4019 | } |
9616614b BP |
4020 | VLOG_WARN("%s", ds_cstr(&s)); |
4021 | ds_destroy(&s); | |
6814e51f | 4022 | } |
050ac423 | 4023 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
4024 | |
4025 | return ok; | |
4026 | } | |
4027 | ||
15baa734 | 4028 | /* Re-searches the classifier for 'facet': |
abe529af BP |
4029 | * |
4030 | * - If the rule found is different from 'facet''s current rule, moves | |
4031 | * 'facet' to the new rule and recompiles its actions. | |
4032 | * | |
4033 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
c57b2226 BP |
4034 | * where it is and recompiles its actions anyway. */ |
4035 | static void | |
15baa734 | 4036 | facet_revalidate(struct facet *facet) |
abe529af | 4037 | { |
15baa734 | 4038 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
4039 | struct actions { |
4040 | struct nlattr *odp_actions; | |
4041 | size_t actions_len; | |
4042 | }; | |
4043 | struct actions *new_actions; | |
4044 | ||
abe529af | 4045 | struct action_xlate_ctx ctx; |
050ac423 BP |
4046 | uint64_t odp_actions_stub[1024 / 8]; |
4047 | struct ofpbuf odp_actions; | |
4048 | ||
abe529af | 4049 | struct rule_dpif *new_rule; |
b0f7b9b5 | 4050 | struct subfacet *subfacet; |
b95fc6ba | 4051 | int i; |
abe529af BP |
4052 | |
4053 | COVERAGE_INC(facet_revalidate); | |
4054 | ||
c57b2226 | 4055 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); |
abe529af | 4056 | |
df2c07f4 | 4057 | /* Calculate new datapath actions. |
abe529af BP |
4058 | * |
4059 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4060 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4061 | * around to properly compose it. */ | |
abe529af | 4062 | |
df2c07f4 JP |
4063 | /* If the datapath actions changed or the installability changed, |
4064 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
4065 | i = 0; |
4066 | new_actions = NULL; | |
4067 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 4068 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 4069 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 4070 | enum slow_path_reason slow; |
b95fc6ba | 4071 | |
e84173dc | 4072 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
0e553d9c | 4073 | subfacet->initial_tci, new_rule, 0, NULL); |
f25d0cf3 | 4074 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, |
050ac423 | 4075 | &odp_actions); |
b0f7b9b5 | 4076 | |
6a7e895f BP |
4077 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
4078 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
4079 | struct dpif_flow_stats stats; | |
4080 | ||
4081 | subfacet_install(subfacet, | |
4082 | odp_actions.data, odp_actions.size, &stats, slow); | |
4083 | subfacet_update_stats(subfacet, &stats); | |
b95fc6ba BP |
4084 | |
4085 | if (!new_actions) { | |
4086 | new_actions = xcalloc(list_size(&facet->subfacets), | |
4087 | sizeof *new_actions); | |
4088 | } | |
050ac423 BP |
4089 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
4090 | odp_actions.size); | |
4091 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 4092 | } |
b95fc6ba | 4093 | |
b95fc6ba | 4094 | i++; |
b0f7b9b5 | 4095 | } |
050ac423 BP |
4096 | ofpbuf_uninit(&odp_actions); |
4097 | ||
b95fc6ba | 4098 | if (new_actions) { |
15baa734 | 4099 | facet_flush_stats(facet); |
abe529af BP |
4100 | } |
4101 | ||
4102 | /* Update 'facet' now that we've taken care of all the old state. */ | |
4103 | facet->tags = ctx.tags; | |
4104 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
75a75043 BP |
4105 | facet->has_learn = ctx.has_learn; |
4106 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4107 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 4108 | facet->mirrors = ctx.mirrors; |
6a7e895f BP |
4109 | |
4110 | i = 0; | |
4111 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4112 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4113 | ||
4114 | if (new_actions && new_actions[i].odp_actions) { | |
4115 | free(subfacet->actions); | |
4116 | subfacet->actions = new_actions[i].odp_actions; | |
4117 | subfacet->actions_len = new_actions[i].actions_len; | |
b95fc6ba | 4118 | } |
6a7e895f | 4119 | i++; |
abe529af | 4120 | } |
6a7e895f BP |
4121 | free(new_actions); |
4122 | ||
abe529af BP |
4123 | if (facet->rule != new_rule) { |
4124 | COVERAGE_INC(facet_changed_rule); | |
4125 | list_remove(&facet->list_node); | |
4126 | list_push_back(&new_rule->facets, &facet->list_node); | |
4127 | facet->rule = new_rule; | |
4128 | facet->used = new_rule->up.created; | |
9d24de3b | 4129 | facet->prev_used = facet->used; |
abe529af | 4130 | } |
abe529af BP |
4131 | } |
4132 | ||
4133 | /* Updates 'facet''s used time. Caller is responsible for calling | |
4134 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
4135 | static void | |
15baa734 | 4136 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 4137 | { |
15baa734 | 4138 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
4139 | if (used > facet->used) { |
4140 | facet->used = used; | |
1745cd08 | 4141 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
4142 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
4143 | } | |
4144 | } | |
4145 | ||
bbb5d219 EJ |
4146 | static void |
4147 | facet_reset_counters(struct facet *facet) | |
4148 | { | |
4149 | facet->packet_count = 0; | |
4150 | facet->byte_count = 0; | |
9d24de3b JP |
4151 | facet->prev_packet_count = 0; |
4152 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
4153 | facet->accounted_bytes = 0; |
4154 | } | |
4155 | ||
abe529af BP |
4156 | static void |
4157 | facet_push_stats(struct facet *facet) | |
4158 | { | |
112bc5f4 | 4159 | struct dpif_flow_stats stats; |
abe529af | 4160 | |
9d24de3b JP |
4161 | assert(facet->packet_count >= facet->prev_packet_count); |
4162 | assert(facet->byte_count >= facet->prev_byte_count); | |
4163 | assert(facet->used >= facet->prev_used); | |
abe529af | 4164 | |
112bc5f4 BP |
4165 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
4166 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4167 | stats.used = facet->used; | |
4168 | stats.tcp_flags = 0; | |
abe529af | 4169 | |
112bc5f4 | 4170 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
4171 | facet->prev_packet_count = facet->packet_count; |
4172 | facet->prev_byte_count = facet->byte_count; | |
4173 | facet->prev_used = facet->used; | |
abe529af | 4174 | |
112bc5f4 | 4175 | flow_push_stats(facet->rule, &facet->flow, &stats); |
9d24de3b JP |
4176 | |
4177 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 4178 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
4179 | } |
4180 | } | |
4181 | ||
abe529af | 4182 | static void |
112bc5f4 | 4183 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 4184 | { |
112bc5f4 BP |
4185 | rule->packet_count += stats->n_packets; |
4186 | rule->byte_count += stats->n_bytes; | |
4187 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
4188 | } |
4189 | ||
4190 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 4191 | * 'rule''s actions and mirrors. */ |
abe529af | 4192 | static void |
18b2a258 | 4193 | flow_push_stats(struct rule_dpif *rule, |
112bc5f4 | 4194 | const struct flow *flow, const struct dpif_flow_stats *stats) |
abe529af BP |
4195 | { |
4196 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
112bc5f4 | 4197 | struct action_xlate_ctx ctx; |
abe529af | 4198 | |
112bc5f4 | 4199 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 4200 | |
112bc5f4 | 4201 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, |
0e553d9c | 4202 | 0, NULL); |
112bc5f4 | 4203 | ctx.resubmit_stats = stats; |
f25d0cf3 BP |
4204 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, |
4205 | rule->up.ofpacts_len); | |
abe529af BP |
4206 | } |
4207 | \f | |
b0f7b9b5 BP |
4208 | /* Subfacets. */ |
4209 | ||
4210 | static struct subfacet * | |
4211 | subfacet_find__(struct ofproto_dpif *ofproto, | |
4212 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
4213 | const struct flow *flow) | |
4214 | { | |
4215 | struct subfacet *subfacet; | |
4216 | ||
4217 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
4218 | &ofproto->subfacets) { | |
4219 | if (subfacet->key | |
4220 | ? (subfacet->key_len == key_len | |
4221 | && !memcmp(key, subfacet->key, key_len)) | |
4222 | : flow_equal(flow, &subfacet->facet->flow)) { | |
4223 | return subfacet; | |
4224 | } | |
4225 | } | |
4226 | ||
4227 | return NULL; | |
4228 | } | |
4229 | ||
4230 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
4231 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
4232 | * there is one, otherwise creates and returns a new subfacet. |
4233 | * | |
4234 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
4235 | * which case the caller must populate the actions with | |
4236 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 4237 | static struct subfacet * |
15baa734 | 4238 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
e84173dc | 4239 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) |
b0f7b9b5 | 4240 | { |
15baa734 | 4241 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4242 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
4243 | struct subfacet *subfacet; | |
4244 | ||
4245 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); | |
4246 | if (subfacet) { | |
4247 | if (subfacet->facet == facet) { | |
4248 | return subfacet; | |
4249 | } | |
4250 | ||
4251 | /* This shouldn't happen. */ | |
4252 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
15baa734 | 4253 | subfacet_destroy(subfacet); |
b0f7b9b5 BP |
4254 | } |
4255 | ||
26cd7e34 BP |
4256 | subfacet = (list_is_empty(&facet->subfacets) |
4257 | ? &facet->one_subfacet | |
4258 | : xmalloc(sizeof *subfacet)); | |
b0f7b9b5 BP |
4259 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
4260 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
4261 | subfacet->facet = facet; | |
b0f7b9b5 BP |
4262 | subfacet->key_fitness = key_fitness; |
4263 | if (key_fitness != ODP_FIT_PERFECT) { | |
4264 | subfacet->key = xmemdup(key, key_len); | |
4265 | subfacet->key_len = key_len; | |
26cd7e34 BP |
4266 | } else { |
4267 | subfacet->key = NULL; | |
4268 | subfacet->key_len = 0; | |
b0f7b9b5 | 4269 | } |
26cd7e34 BP |
4270 | subfacet->used = time_msec(); |
4271 | subfacet->dp_packet_count = 0; | |
4272 | subfacet->dp_byte_count = 0; | |
4273 | subfacet->actions_len = 0; | |
4274 | subfacet->actions = NULL; | |
6a7e895f BP |
4275 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE |
4276 | ? SLOW_MATCH | |
4277 | : 0); | |
4278 | subfacet->path = SF_NOT_INSTALLED; | |
e84173dc | 4279 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
4280 | |
4281 | return subfacet; | |
4282 | } | |
4283 | ||
4284 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
4285 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
4286 | static struct subfacet * | |
4287 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 4288 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
4289 | { |
4290 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
4291 | enum odp_key_fitness fitness; |
4292 | struct flow flow; | |
4293 | ||
4294 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
4295 | if (fitness == ODP_FIT_ERROR) { | |
4296 | return NULL; | |
4297 | } | |
b0f7b9b5 | 4298 | |
6a542738 | 4299 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
4300 | } |
4301 | ||
4302 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
4303 | * its facet within 'ofproto', and frees it. */ | |
4304 | static void | |
15baa734 | 4305 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4306 | { |
15baa734 BP |
4307 | struct facet *facet = subfacet->facet; |
4308 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4309 | ||
4310 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
4311 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
4312 | list_remove(&subfacet->list_node); | |
4313 | free(subfacet->key); | |
b95fc6ba | 4314 | free(subfacet->actions); |
26cd7e34 BP |
4315 | if (subfacet != &facet->one_subfacet) { |
4316 | free(subfacet); | |
4317 | } | |
b0f7b9b5 BP |
4318 | } |
4319 | ||
4320 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4321 | * last remaining subfacet in its facet destroys the facet too. */ | |
4322 | static void | |
15baa734 | 4323 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4324 | { |
4325 | struct facet *facet = subfacet->facet; | |
4326 | ||
551a2f6c BP |
4327 | if (list_is_singleton(&facet->subfacets)) { |
4328 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4329 | facet_remove(facet); |
551a2f6c | 4330 | } else { |
15baa734 | 4331 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4332 | } |
4333 | } | |
4334 | ||
4335 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
4336 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
4337 | * for use as temporary storage. */ | |
4338 | static void | |
4339 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
4340 | struct ofpbuf *key) | |
4341 | { | |
4342 | if (!subfacet->key) { | |
4343 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
4344 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
4345 | } else { | |
4346 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
4347 | } | |
4348 | } | |
4349 | ||
5fe20d5d BP |
4350 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
4351 | * Translates the actions into 'odp_actions', which the caller must have | |
4352 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 4353 | static void |
5fe20d5d BP |
4354 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
4355 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
4356 | { |
4357 | struct facet *facet = subfacet->facet; | |
18b2a258 | 4358 | struct rule_dpif *rule = facet->rule; |
15baa734 | 4359 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 4360 | |
b95fc6ba BP |
4361 | struct action_xlate_ctx ctx; |
4362 | ||
15baa734 | 4363 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
0e553d9c | 4364 | rule, 0, packet); |
f25d0cf3 | 4365 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); |
b95fc6ba | 4366 | facet->tags = ctx.tags; |
b95fc6ba BP |
4367 | facet->has_learn = ctx.has_learn; |
4368 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4369 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 4370 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 4371 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 4372 | |
6a7e895f | 4373 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; |
5fe20d5d BP |
4374 | if (subfacet->actions_len != odp_actions->size |
4375 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 4376 | free(subfacet->actions); |
5fe20d5d BP |
4377 | subfacet->actions_len = odp_actions->size; |
4378 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 4379 | } |
b95fc6ba BP |
4380 | } |
4381 | ||
b0f7b9b5 BP |
4382 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4383 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4384 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4385 | * since 'subfacet' was last updated. | |
4386 | * | |
4387 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4388 | static int | |
15baa734 | 4389 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 | 4390 | const struct nlattr *actions, size_t actions_len, |
6a7e895f BP |
4391 | struct dpif_flow_stats *stats, |
4392 | enum slow_path_reason slow) | |
b0f7b9b5 | 4393 | { |
15baa734 BP |
4394 | struct facet *facet = subfacet->facet; |
4395 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
6a7e895f BP |
4396 | enum subfacet_path path = subfacet_want_path(slow); |
4397 | uint64_t slow_path_stub[128 / 8]; | |
b0f7b9b5 BP |
4398 | struct odputil_keybuf keybuf; |
4399 | enum dpif_flow_put_flags flags; | |
4400 | struct ofpbuf key; | |
4401 | int ret; | |
4402 | ||
4403 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
4404 | if (stats) { | |
4405 | flags |= DPIF_FP_ZERO_STATS; | |
4406 | } | |
4407 | ||
6a7e895f BP |
4408 | if (path == SF_SLOW_PATH) { |
4409 | compose_slow_path(ofproto, &facet->flow, slow, | |
4410 | slow_path_stub, sizeof slow_path_stub, | |
4411 | &actions, &actions_len); | |
4412 | } | |
4413 | ||
b0f7b9b5 BP |
4414 | subfacet_get_key(subfacet, &keybuf, &key); |
4415 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
4416 | actions, actions_len, stats); | |
4417 | ||
4418 | if (stats) { | |
4419 | subfacet_reset_dp_stats(subfacet, stats); | |
4420 | } | |
4421 | ||
6a7e895f BP |
4422 | if (!ret) { |
4423 | subfacet->path = path; | |
4424 | } | |
b0f7b9b5 BP |
4425 | return ret; |
4426 | } | |
4427 | ||
6a7e895f BP |
4428 | static int |
4429 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
4430 | { | |
4431 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
4432 | stats, subfacet->slow); | |
4433 | } | |
4434 | ||
b0f7b9b5 BP |
4435 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ |
4436 | static void | |
15baa734 | 4437 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 | 4438 | { |
6a7e895f | 4439 | if (subfacet->path != SF_NOT_INSTALLED) { |
15baa734 BP |
4440 | struct rule_dpif *rule = subfacet->facet->rule; |
4441 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
4442 | struct odputil_keybuf keybuf; |
4443 | struct dpif_flow_stats stats; | |
4444 | struct ofpbuf key; | |
4445 | int error; | |
4446 | ||
4447 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 4448 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
4449 | subfacet_reset_dp_stats(subfacet, &stats); |
4450 | if (!error) { | |
15baa734 | 4451 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 4452 | } |
6a7e895f | 4453 | subfacet->path = SF_NOT_INSTALLED; |
b0f7b9b5 BP |
4454 | } else { |
4455 | assert(subfacet->dp_packet_count == 0); | |
4456 | assert(subfacet->dp_byte_count == 0); | |
4457 | } | |
4458 | } | |
4459 | ||
4460 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4461 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4462 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4463 | * was reset in the datapath. 'stats' will be modified to include only | |
4464 | * statistics new since 'subfacet' was last updated. */ | |
4465 | static void | |
4466 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4467 | struct dpif_flow_stats *stats) | |
4468 | { | |
4469 | if (stats | |
4470 | && subfacet->dp_packet_count <= stats->n_packets | |
4471 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4472 | stats->n_packets -= subfacet->dp_packet_count; | |
4473 | stats->n_bytes -= subfacet->dp_byte_count; | |
4474 | } | |
4475 | ||
4476 | subfacet->dp_packet_count = 0; | |
4477 | subfacet->dp_byte_count = 0; | |
4478 | } | |
4479 | ||
4480 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
4481 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
4482 | static void | |
15baa734 | 4483 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
4484 | { |
4485 | if (used > subfacet->used) { | |
4486 | subfacet->used = used; | |
15baa734 | 4487 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
4488 | } |
4489 | } | |
4490 | ||
4491 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
4492 | * | |
4493 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4494 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4495 | * represents a packet that was sent by hand or if it represents statistics | |
4496 | * that have been cleared out of the datapath. */ | |
4497 | static void | |
15baa734 | 4498 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4499 | const struct dpif_flow_stats *stats) |
4500 | { | |
4501 | if (stats->n_packets || stats->used > subfacet->used) { | |
4502 | struct facet *facet = subfacet->facet; | |
4503 | ||
15baa734 | 4504 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
4505 | facet->packet_count += stats->n_packets; |
4506 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4507 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4508 | facet_push_stats(facet); |
4509 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
4510 | } | |
4511 | } | |
4512 | \f | |
abe529af BP |
4513 | /* Rules. */ |
4514 | ||
4515 | static struct rule_dpif * | |
c57b2226 BP |
4516 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) |
4517 | { | |
4518 | struct ofport_dpif *port; | |
4519 | struct rule_dpif *rule; | |
4520 | ||
4521 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
4522 | if (rule) { | |
4523 | return rule; | |
4524 | } | |
4525 | ||
4526 | port = get_ofp_port(ofproto, flow->in_port); | |
4527 | if (!port) { | |
4528 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
4529 | return ofproto->miss_rule; | |
4530 | } | |
4531 | ||
4532 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
4533 | return ofproto->no_packet_in_rule; | |
4534 | } | |
4535 | return ofproto->miss_rule; | |
4536 | } | |
4537 | ||
4538 | static struct rule_dpif * | |
4539 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4540 | uint8_t table_id) | |
abe529af | 4541 | { |
7257b535 BP |
4542 | struct cls_rule *cls_rule; |
4543 | struct classifier *cls; | |
4544 | ||
9cdaaebe BP |
4545 | if (table_id >= N_TABLES) { |
4546 | return NULL; | |
4547 | } | |
4548 | ||
d0918789 | 4549 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 4550 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
4551 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
4552 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
4553 | * are unavailable. */ | |
4554 | struct flow ofpc_normal_flow = *flow; | |
4555 | ofpc_normal_flow.tp_src = htons(0); | |
4556 | ofpc_normal_flow.tp_dst = htons(0); | |
4557 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
4558 | } else { | |
4559 | cls_rule = classifier_lookup(cls, flow); | |
4560 | } | |
4561 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
4562 | } |
4563 | ||
7ee20df1 BP |
4564 | static void |
4565 | complete_operation(struct rule_dpif *rule) | |
4566 | { | |
4567 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4568 | ||
54a9cbc9 | 4569 | rule_invalidate(rule); |
7ee20df1 BP |
4570 | if (clogged) { |
4571 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4572 | c->op = rule->up.pending; | |
4573 | list_push_back(&ofproto->completions, &c->list_node); | |
4574 | } else { | |
4575 | ofoperation_complete(rule->up.pending, 0); | |
4576 | } | |
4577 | } | |
4578 | ||
abe529af BP |
4579 | static struct rule * |
4580 | rule_alloc(void) | |
4581 | { | |
4582 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4583 | return &rule->up; | |
4584 | } | |
4585 | ||
4586 | static void | |
4587 | rule_dealloc(struct rule *rule_) | |
4588 | { | |
4589 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4590 | free(rule); | |
4591 | } | |
4592 | ||
90bf1e07 | 4593 | static enum ofperr |
abe529af BP |
4594 | rule_construct(struct rule *rule_) |
4595 | { | |
4596 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4597 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4598 | struct rule_dpif *victim; |
54a9cbc9 | 4599 | uint8_t table_id; |
90bf1e07 | 4600 | enum ofperr error; |
5bf0e941 | 4601 | |
f25d0cf3 BP |
4602 | error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len, |
4603 | &rule->up.cr.flow, ofproto->max_ports); | |
5bf0e941 BP |
4604 | if (error) { |
4605 | return error; | |
4606 | } | |
abe529af | 4607 | |
abe529af BP |
4608 | rule->packet_count = 0; |
4609 | rule->byte_count = 0; | |
abe529af | 4610 | |
7ee20df1 BP |
4611 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4612 | if (victim && !list_is_empty(&victim->facets)) { | |
4613 | struct facet *facet; | |
4614 | ||
4615 | rule->facets = victim->facets; | |
4616 | list_moved(&rule->facets); | |
4617 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4618 | /* XXX: We're only clearing our local counters here. It's possible |
4619 | * that quite a few packets are unaccounted for in the datapath | |
4620 | * statistics. These will be accounted to the new rule instead of | |
4621 | * cleared as required. This could be fixed by clearing out the | |
4622 | * datapath statistics for this facet, but currently it doesn't | |
4623 | * seem worth it. */ | |
4624 | facet_reset_counters(facet); | |
7ee20df1 BP |
4625 | facet->rule = rule; |
4626 | } | |
4627 | } else { | |
4628 | /* Must avoid list_moved() in this case. */ | |
4629 | list_init(&rule->facets); | |
4630 | } | |
abe529af | 4631 | |
54a9cbc9 BP |
4632 | table_id = rule->up.table_id; |
4633 | rule->tag = (victim ? victim->tag | |
4634 | : table_id == 0 ? 0 | |
4635 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4636 | ofproto->tables[table_id].basis)); | |
4637 | ||
7ee20df1 | 4638 | complete_operation(rule); |
abe529af BP |
4639 | return 0; |
4640 | } | |
4641 | ||
4642 | static void | |
4643 | rule_destruct(struct rule *rule_) | |
4644 | { | |
4645 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4646 | struct facet *facet, *next_facet; |
4647 | ||
abe529af | 4648 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4649 | facet_revalidate(facet); |
abe529af | 4650 | } |
7ee20df1 BP |
4651 | |
4652 | complete_operation(rule); | |
abe529af BP |
4653 | } |
4654 | ||
4655 | static void | |
4656 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4657 | { | |
4658 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4659 | struct facet *facet; | |
4660 | ||
4661 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4662 | * in facets. This counts, for example, facets that have expired. */ | |
4663 | *packets = rule->packet_count; | |
4664 | *bytes = rule->byte_count; | |
4665 | ||
4666 | /* Add any statistics that are tracked by facets. This includes | |
4667 | * statistical data recently updated by ofproto_update_stats() as well as | |
4668 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4669 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4670 | *packets += facet->packet_count; | |
4671 | *bytes += facet->byte_count; | |
4672 | } | |
4673 | } | |
4674 | ||
90bf1e07 | 4675 | static enum ofperr |
59d0f2c8 BP |
4676 | rule_execute(struct rule *rule_, const struct flow *flow, |
4677 | struct ofpbuf *packet) | |
abe529af BP |
4678 | { |
4679 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4680 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
050ac423 | 4681 | |
112bc5f4 | 4682 | struct dpif_flow_stats stats; |
050ac423 | 4683 | |
abe529af | 4684 | struct action_xlate_ctx ctx; |
050ac423 BP |
4685 | uint64_t odp_actions_stub[1024 / 8]; |
4686 | struct ofpbuf odp_actions; | |
abe529af | 4687 | |
112bc5f4 BP |
4688 | dpif_flow_stats_extract(flow, packet, &stats); |
4689 | rule_credit_stats(rule, &stats); | |
4690 | ||
050ac423 | 4691 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
54834960 | 4692 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
112bc5f4 BP |
4693 | rule, stats.tcp_flags, packet); |
4694 | ctx.resubmit_stats = &stats; | |
f25d0cf3 | 4695 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); |
112bc5f4 BP |
4696 | |
4697 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
4698 | odp_actions.size, packet); | |
4699 | ||
050ac423 | 4700 | ofpbuf_uninit(&odp_actions); |
5bf0e941 BP |
4701 | |
4702 | return 0; | |
abe529af BP |
4703 | } |
4704 | ||
7ee20df1 BP |
4705 | static void |
4706 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4707 | { |
4708 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4709 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
90bf1e07 | 4710 | enum ofperr error; |
abe529af | 4711 | |
f25d0cf3 BP |
4712 | error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len, |
4713 | &rule->up.cr.flow, ofproto->max_ports); | |
7ee20df1 BP |
4714 | if (error) { |
4715 | ofoperation_complete(rule->up.pending, error); | |
4716 | return; | |
abe529af | 4717 | } |
7ee20df1 BP |
4718 | |
4719 | complete_operation(rule); | |
abe529af BP |
4720 | } |
4721 | \f | |
97d6520b | 4722 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4723 | * May modify 'packet'. |
abe529af BP |
4724 | * Returns 0 if successful, otherwise a positive errno value. */ |
4725 | static int | |
52a90c29 | 4726 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4727 | { |
97d6520b | 4728 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4729 | struct ofpbuf key, odp_actions; |
4730 | struct odputil_keybuf keybuf; | |
52a90c29 | 4731 | uint16_t odp_port; |
80e5eed9 | 4732 | struct flow flow; |
abe529af BP |
4733 | int error; |
4734 | ||
abff858b | 4735 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
52a90c29 BP |
4736 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4737 | flow.vlan_tci); | |
4738 | if (odp_port != ofport->odp_port) { | |
4739 | eth_pop_vlan(packet); | |
4740 | flow.vlan_tci = htons(0); | |
4741 | } | |
4742 | ||
80e5eed9 BP |
4743 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4744 | odp_flow_key_from_flow(&key, &flow); | |
4745 | ||
abe529af | 4746 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4747 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4748 | ||
df2c07f4 | 4749 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4750 | error = dpif_execute(ofproto->dpif, |
4751 | key.data, key.size, | |
4752 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4753 | packet); |
4754 | ofpbuf_uninit(&odp_actions); | |
4755 | ||
4756 | if (error) { | |
4757 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4758 | ofproto->up.name, odp_port, strerror(error)); | |
4759 | } | |
6527c598 | 4760 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4761 | return error; |
4762 | } | |
4763 | \f | |
df2c07f4 | 4764 | /* OpenFlow to datapath action translation. */ |
abe529af | 4765 | |
f25d0cf3 BP |
4766 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, |
4767 | struct action_xlate_ctx *); | |
4cd78906 | 4768 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4769 | |
6a7e895f BP |
4770 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. |
4771 | * The action will state 'slow' as the reason that the action is in the slow | |
4772 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
4773 | * dump-flows" output to see why a flow is in the slow path.) | |
4774 | * | |
4775 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
4776 | * 'stub_size' must be large enough for the action. | |
4777 | * | |
4778 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
4779 | * respectively. */ | |
4780 | static void | |
4781 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
4782 | enum slow_path_reason slow, | |
4783 | uint64_t *stub, size_t stub_size, | |
4784 | const struct nlattr **actionsp, size_t *actions_lenp) | |
4785 | { | |
4786 | union user_action_cookie cookie; | |
4787 | struct ofpbuf buf; | |
4788 | ||
4789 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
4790 | cookie.slow_path.unused = 0; | |
4791 | cookie.slow_path.reason = slow; | |
4792 | ||
4793 | ofpbuf_use_stack(&buf, stub, stub_size); | |
625b0720 BP |
4794 | if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { |
4795 | uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX); | |
4796 | odp_put_userspace_action(pid, &cookie, &buf); | |
4797 | } else { | |
4798 | put_userspace_action(ofproto, &buf, flow, &cookie); | |
4799 | } | |
6a7e895f BP |
4800 | *actionsp = buf.data; |
4801 | *actions_lenp = buf.size; | |
4802 | } | |
4803 | ||
98403001 BP |
4804 | static size_t |
4805 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4806 | struct ofpbuf *odp_actions, | |
4807 | const struct flow *flow, | |
1673e0e4 | 4808 | const union user_action_cookie *cookie) |
98403001 | 4809 | { |
98403001 BP |
4810 | uint32_t pid; |
4811 | ||
4812 | pid = dpif_port_get_pid(ofproto->dpif, | |
4813 | ofp_port_to_odp_port(flow->in_port)); | |
4814 | ||
39db78a0 | 4815 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4816 | } |
4817 | ||
36fc5f18 BP |
4818 | static void |
4819 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
4820 | ovs_be16 vlan_tci, uint32_t odp_port, | |
1673e0e4 | 4821 | unsigned int n_outputs, union user_action_cookie *cookie) |
36fc5f18 BP |
4822 | { |
4823 | int ifindex; | |
4824 | ||
4825 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
1673e0e4 | 4826 | cookie->sflow.vlan_tci = vlan_tci; |
36fc5f18 BP |
4827 | |
4828 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
4829 | * port information") for the interpretation of cookie->output. */ | |
4830 | switch (n_outputs) { | |
4831 | case 0: | |
4832 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
1673e0e4 | 4833 | cookie->sflow.output = 0x40000000 | 256; |
36fc5f18 BP |
4834 | break; |
4835 | ||
4836 | case 1: | |
4837 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4838 | if (ifindex) { | |
1673e0e4 | 4839 | cookie->sflow.output = ifindex; |
36fc5f18 BP |
4840 | break; |
4841 | } | |
4842 | /* Fall through. */ | |
4843 | default: | |
4844 | /* 0x80000000 means "multiple output ports. */ | |
1673e0e4 | 4845 | cookie->sflow.output = 0x80000000 | n_outputs; |
36fc5f18 BP |
4846 | break; |
4847 | } | |
4848 | } | |
4849 | ||
6ff686f2 PS |
4850 | /* Compose SAMPLE action for sFlow. */ |
4851 | static size_t | |
4852 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4853 | struct ofpbuf *odp_actions, | |
4854 | const struct flow *flow, | |
4855 | uint32_t odp_port) | |
4856 | { | |
6ff686f2 | 4857 | uint32_t probability; |
1673e0e4 | 4858 | union user_action_cookie cookie; |
6ff686f2 | 4859 | size_t sample_offset, actions_offset; |
36fc5f18 | 4860 | int cookie_offset; |
6ff686f2 PS |
4861 | |
4862 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4863 | return 0; | |
4864 | } | |
4865 | ||
6ff686f2 PS |
4866 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); |
4867 | ||
4868 | /* Number of packets out of UINT_MAX to sample. */ | |
4869 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4870 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4871 | ||
4872 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
36fc5f18 BP |
4873 | compose_sflow_cookie(ofproto, htons(0), odp_port, |
4874 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
98403001 | 4875 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); |
6ff686f2 PS |
4876 | |
4877 | nl_msg_end_nested(odp_actions, actions_offset); | |
4878 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4879 | return cookie_offset; |
6ff686f2 PS |
4880 | } |
4881 | ||
4882 | /* SAMPLE action must be first action in any given list of actions. | |
4883 | * At this point we do not have all information required to build it. So try to | |
4884 | * build sample action as complete as possible. */ | |
4885 | static void | |
4886 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4887 | { | |
4888 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4889 | ctx->odp_actions, | |
4890 | &ctx->flow, OVSP_NONE); | |
4891 | ctx->sflow_odp_port = 0; | |
4892 | ctx->sflow_n_outputs = 0; | |
4893 | } | |
4894 | ||
4895 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4896 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4897 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4898 | static void | |
4899 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4900 | { | |
4901 | const struct flow *base = &ctx->base_flow; | |
1673e0e4 | 4902 | union user_action_cookie *cookie; |
6ff686f2 PS |
4903 | |
4904 | if (!ctx->user_cookie_offset) { | |
4905 | return; | |
4906 | } | |
4907 | ||
4908 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
36fc5f18 | 4909 | sizeof(*cookie)); |
6ff686f2 PS |
4910 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); |
4911 | ||
36fc5f18 BP |
4912 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, |
4913 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6ff686f2 PS |
4914 | } |
4915 | ||
6ff686f2 | 4916 | static void |
81b1afb1 EJ |
4917 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4918 | bool check_stp) | |
6ff686f2 | 4919 | { |
d59906fb | 4920 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4921 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4922 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4923 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4924 | uint16_t out_port; |
d59906fb | 4925 | |
81b1afb1 | 4926 | if (ofport) { |
8b36f51e EJ |
4927 | struct priority_to_dscp *pdscp; |
4928 | ||
9e1fd49b | 4929 | if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD |
81b1afb1 EJ |
4930 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { |
4931 | return; | |
4932 | } | |
8b36f51e | 4933 | |
deedf7e7 | 4934 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4935 | if (pdscp) { |
4936 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4937 | ctx->flow.nw_tos |= pdscp->dscp; | |
4938 | } | |
81b1afb1 EJ |
4939 | } else { |
4940 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4941 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4942 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4943 | } |
4944 | ||
52a90c29 BP |
4945 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4946 | ctx->flow.vlan_tci); | |
4947 | if (out_port != odp_port) { | |
4948 | ctx->flow.vlan_tci = htons(0); | |
4949 | } | |
5bbda0aa | 4950 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4951 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4952 | ||
6ff686f2 PS |
4953 | ctx->sflow_odp_port = odp_port; |
4954 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4955 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4956 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4957 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4958 | } |
4959 | ||
abe529af | 4960 | static void |
5e48dc2b | 4961 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4962 | { |
81b1afb1 | 4963 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4964 | } |
4965 | ||
4966 | static void | |
29901626 BP |
4967 | xlate_table_action(struct action_xlate_ctx *ctx, |
4968 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
4969 | { |
4970 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4971 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4972 | struct rule_dpif *rule; |
4973 | uint16_t old_in_port; | |
29901626 BP |
4974 | uint8_t old_table_id; |
4975 | ||
4976 | old_table_id = ctx->table_id; | |
4977 | ctx->table_id = table_id; | |
abe529af | 4978 | |
54a9cbc9 | 4979 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
4980 | old_in_port = ctx->flow.in_port; |
4981 | ctx->flow.in_port = in_port; | |
c57b2226 | 4982 | rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id); |
54a9cbc9 BP |
4983 | |
4984 | /* Tag the flow. */ | |
4985 | if (table_id > 0 && table_id < N_TABLES) { | |
4986 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4987 | if (table->other_table) { | |
33780682 | 4988 | ctx->tags |= (rule && rule->tag |
54a9cbc9 BP |
4989 | ? rule->tag |
4990 | : rule_calculate_tag(&ctx->flow, | |
4991 | &table->other_table->wc, | |
4992 | table->basis)); | |
4993 | } | |
4994 | } | |
4995 | ||
4996 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
4997 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
4998 | ctx->flow.in_port = old_in_port; |
4999 | ||
5000 | if (ctx->resubmit_hook) { | |
5001 | ctx->resubmit_hook(ctx, rule); | |
5002 | } | |
5003 | ||
5004 | if (rule) { | |
18b2a258 | 5005 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 5006 | |
112bc5f4 BP |
5007 | if (ctx->resubmit_stats) { |
5008 | rule_credit_stats(rule, ctx->resubmit_stats); | |
5009 | } | |
5010 | ||
abe529af | 5011 | ctx->recurse++; |
18b2a258 | 5012 | ctx->rule = rule; |
f25d0cf3 | 5013 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); |
18b2a258 | 5014 | ctx->rule = old_rule; |
abe529af BP |
5015 | ctx->recurse--; |
5016 | } | |
29901626 BP |
5017 | |
5018 | ctx->table_id = old_table_id; | |
abe529af BP |
5019 | } else { |
5020 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5021 | ||
29901626 | 5022 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 5023 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 5024 | ctx->max_resubmit_trigger = true; |
abe529af BP |
5025 | } |
5026 | } | |
5027 | ||
29901626 | 5028 | static void |
f25d0cf3 BP |
5029 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, |
5030 | const struct ofpact_resubmit *resubmit) | |
29901626 BP |
5031 | { |
5032 | uint16_t in_port; | |
5033 | uint8_t table_id; | |
5034 | ||
f25d0cf3 BP |
5035 | in_port = resubmit->in_port; |
5036 | if (in_port == OFPP_IN_PORT) { | |
5037 | in_port = ctx->flow.in_port; | |
5038 | } | |
5039 | ||
5040 | table_id = resubmit->table_id; | |
5041 | if (table_id == 255) { | |
5042 | table_id = ctx->table_id; | |
5043 | } | |
29901626 BP |
5044 | |
5045 | xlate_table_action(ctx, in_port, table_id); | |
5046 | } | |
5047 | ||
abe529af | 5048 | static void |
d59906fb | 5049 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
5050 | { |
5051 | struct ofport_dpif *ofport; | |
5052 | ||
b3e9b2ed | 5053 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 5054 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
5055 | |
5056 | if (ofp_port == ctx->flow.in_port) { | |
5057 | continue; | |
5058 | } | |
5059 | ||
5e48dc2b | 5060 | if (all) { |
81b1afb1 | 5061 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 5062 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 5063 | compose_output_action(ctx, ofp_port); |
abe529af BP |
5064 | } |
5065 | } | |
b3e9b2ed EJ |
5066 | |
5067 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
5068 | } |
5069 | ||
6ff686f2 | 5070 | static void |
f0fd1a17 | 5071 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
5072 | enum ofp_packet_in_reason reason, |
5073 | uint16_t controller_id) | |
6ff686f2 | 5074 | { |
999fba59 EJ |
5075 | struct ofputil_packet_in pin; |
5076 | struct ofpbuf *packet; | |
6ff686f2 | 5077 | |
6a7e895f | 5078 | ctx->slow |= SLOW_CONTROLLER; |
999fba59 EJ |
5079 | if (!ctx->packet) { |
5080 | return; | |
5081 | } | |
5082 | ||
5083 | packet = ofpbuf_clone(ctx->packet); | |
5084 | ||
5085 | if (packet->l2 && packet->l3) { | |
5086 | struct eth_header *eh; | |
5087 | ||
5088 | eth_pop_vlan(packet); | |
5089 | eh = packet->l2; | |
0104aba8 EJ |
5090 | |
5091 | /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 | |
5092 | * LLC frame. Calculating the Ethernet type of these frames is more | |
5093 | * trouble than seems appropriate for a simple assertion. */ | |
5094 | assert(ntohs(eh->eth_type) < ETH_TYPE_MIN | |
5095 | || eh->eth_type == ctx->flow.dl_type); | |
5096 | ||
999fba59 EJ |
5097 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
5098 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
5099 | ||
5100 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
5101 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
5102 | } | |
5103 | ||
5104 | if (packet->l4) { | |
5105 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5106 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
5107 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
5108 | } | |
5109 | ||
5110 | if (packet->l7) { | |
5111 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
5112 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
5113 | ctx->flow.tp_dst); | |
5114 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
5115 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
5116 | ctx->flow.tp_dst); | |
5117 | } | |
5118 | } | |
5119 | } | |
5120 | } | |
5121 | ||
5122 | pin.packet = packet->data; | |
5123 | pin.packet_len = packet->size; | |
f0fd1a17 | 5124 | pin.reason = reason; |
a7349929 | 5125 | pin.controller_id = controller_id; |
54834960 | 5126 | pin.table_id = ctx->table_id; |
18b2a258 | 5127 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 5128 | |
999fba59 | 5129 | pin.send_len = len; |
999fba59 EJ |
5130 | flow_get_metadata(&ctx->flow, &pin.fmd); |
5131 | ||
d8653c38 | 5132 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 5133 | ofpbuf_delete(packet); |
6ff686f2 PS |
5134 | } |
5135 | ||
f0fd1a17 PS |
5136 | static bool |
5137 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
5138 | { | |
5139 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
5140 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
5141 | return false; | |
5142 | } | |
5143 | ||
5144 | if (ctx->flow.nw_ttl > 1) { | |
5145 | ctx->flow.nw_ttl--; | |
5146 | return false; | |
5147 | } else { | |
a7349929 | 5148 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); |
f0fd1a17 PS |
5149 | |
5150 | /* Stop processing for current table. */ | |
5151 | return true; | |
5152 | } | |
5153 | } | |
5154 | ||
abe529af | 5155 | static void |
f25d0cf3 BP |
5156 | xlate_output_action(struct action_xlate_ctx *ctx, |
5157 | uint16_t port, uint16_t max_len) | |
abe529af BP |
5158 | { |
5159 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
5160 | ||
5161 | ctx->nf_output_iface = NF_OUT_DROP; | |
5162 | ||
5163 | switch (port) { | |
5164 | case OFPP_IN_PORT: | |
81b1afb1 | 5165 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
5166 | break; |
5167 | case OFPP_TABLE: | |
333be161 | 5168 | xlate_table_action(ctx, ctx->flow.in_port, 0); |
abe529af BP |
5169 | break; |
5170 | case OFPP_NORMAL: | |
5171 | xlate_normal(ctx); | |
5172 | break; | |
5173 | case OFPP_FLOOD: | |
d59906fb | 5174 | flood_packets(ctx, false); |
abe529af BP |
5175 | break; |
5176 | case OFPP_ALL: | |
d59906fb | 5177 | flood_packets(ctx, true); |
abe529af BP |
5178 | break; |
5179 | case OFPP_CONTROLLER: | |
a7349929 | 5180 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 5181 | break; |
e81d2933 EJ |
5182 | case OFPP_NONE: |
5183 | break; | |
a0fbe94a | 5184 | case OFPP_LOCAL: |
abe529af BP |
5185 | default: |
5186 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 5187 | compose_output_action(ctx, port); |
abe529af BP |
5188 | } |
5189 | break; | |
5190 | } | |
5191 | ||
5192 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
5193 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
5194 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
5195 | ctx->nf_output_iface = prev_nf_output_iface; | |
5196 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
5197 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
5198 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5199 | } | |
5200 | } | |
5201 | ||
f694937d EJ |
5202 | static void |
5203 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5204 | const struct ofpact_output_reg *or) |
f694937d | 5205 | { |
f25d0cf3 BP |
5206 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); |
5207 | if (port <= UINT16_MAX) { | |
5208 | xlate_output_action(ctx, port, or->max_len); | |
f694937d EJ |
5209 | } |
5210 | } | |
5211 | ||
abe529af BP |
5212 | static void |
5213 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5214 | const struct ofpact_enqueue *enqueue) |
abe529af | 5215 | { |
f25d0cf3 BP |
5216 | uint16_t ofp_port = enqueue->port; |
5217 | uint32_t queue_id = enqueue->queue; | |
abff858b | 5218 | uint32_t flow_priority, priority; |
abe529af BP |
5219 | int error; |
5220 | ||
f25d0cf3 BP |
5221 | /* Translate queue to priority. */ |
5222 | error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority); | |
abe529af BP |
5223 | if (error) { |
5224 | /* Fall back to ordinary output action. */ | |
f25d0cf3 | 5225 | xlate_output_action(ctx, enqueue->port, 0); |
abe529af BP |
5226 | return; |
5227 | } | |
5228 | ||
f25d0cf3 | 5229 | /* Check output port. */ |
abe529af BP |
5230 | if (ofp_port == OFPP_IN_PORT) { |
5231 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
5232 | } else if (ofp_port == ctx->flow.in_port) { |
5233 | return; | |
abe529af | 5234 | } |
abe529af | 5235 | |
df2c07f4 | 5236 | /* Add datapath actions. */ |
deedf7e7 BP |
5237 | flow_priority = ctx->flow.skb_priority; |
5238 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 5239 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 5240 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
5241 | |
5242 | /* Update NetFlow output port. */ | |
5243 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 5244 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
5245 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
5246 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5247 | } | |
5248 | } | |
5249 | ||
5250 | static void | |
f25d0cf3 | 5251 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) |
abe529af | 5252 | { |
f25d0cf3 | 5253 | uint32_t skb_priority; |
abe529af | 5254 | |
f25d0cf3 BP |
5255 | if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) { |
5256 | ctx->flow.skb_priority = skb_priority; | |
5257 | } else { | |
5258 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
abe529af | 5259 | * has already been logged. */ |
abe529af | 5260 | } |
abe529af BP |
5261 | } |
5262 | ||
5263 | struct xlate_reg_state { | |
5264 | ovs_be16 vlan_tci; | |
5265 | ovs_be64 tun_id; | |
5266 | }; | |
5267 | ||
abe529af BP |
5268 | static void |
5269 | xlate_autopath(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5270 | const struct ofpact_autopath *ap) |
abe529af | 5271 | { |
f25d0cf3 | 5272 | uint16_t ofp_port = ap->port; |
abe529af BP |
5273 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); |
5274 | ||
5275 | if (!port || !port->bundle) { | |
5276 | ofp_port = OFPP_NONE; | |
5277 | } else if (port->bundle->bond) { | |
5278 | /* Autopath does not support VLAN hashing. */ | |
5279 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 5280 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
5281 | if (slave) { |
5282 | ofp_port = slave->up.ofp_port; | |
5283 | } | |
5284 | } | |
f25d0cf3 | 5285 | nxm_reg_load(&ap->dst, ofp_port, &ctx->flow); |
abe529af BP |
5286 | } |
5287 | ||
daff3353 EJ |
5288 | static bool |
5289 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
5290 | { | |
5291 | struct ofproto_dpif *ofproto = ofproto_; | |
5292 | struct ofport_dpif *port; | |
5293 | ||
5294 | switch (ofp_port) { | |
5295 | case OFPP_IN_PORT: | |
5296 | case OFPP_TABLE: | |
5297 | case OFPP_NORMAL: | |
5298 | case OFPP_FLOOD: | |
5299 | case OFPP_ALL: | |
439e4d8c | 5300 | case OFPP_NONE: |
daff3353 EJ |
5301 | return true; |
5302 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
5303 | return false; | |
5304 | default: | |
5305 | port = get_ofp_port(ofproto, ofp_port); | |
5306 | return port ? port->may_enable : false; | |
5307 | } | |
5308 | } | |
5309 | ||
f25d0cf3 BP |
5310 | static void |
5311 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
5312 | const struct ofpact_bundle *bundle) | |
5313 | { | |
5314 | uint16_t port; | |
5315 | ||
5316 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
5317 | if (bundle->dst.field) { | |
5318 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
5319 | } else { | |
5320 | xlate_output_action(ctx, port, 0); | |
5321 | } | |
5322 | } | |
5323 | ||
75a75043 BP |
5324 | static void |
5325 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5326 | const struct ofpact_learn *learn) |
75a75043 BP |
5327 | { |
5328 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
5329 | struct ofputil_flow_mod fm; | |
f25d0cf3 BP |
5330 | uint64_t ofpacts_stub[1024 / 8]; |
5331 | struct ofpbuf ofpacts; | |
75a75043 BP |
5332 | int error; |
5333 | ||
f25d0cf3 BP |
5334 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
5335 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
75a75043 BP |
5336 | |
5337 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
5338 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
5339 | VLOG_WARN("learning action failed to modify flow table (%s)", |
5340 | ofperr_get_name(error)); | |
75a75043 BP |
5341 | } |
5342 | ||
f25d0cf3 | 5343 | ofpbuf_uninit(&ofpacts); |
75a75043 BP |
5344 | } |
5345 | ||
0e553d9c BP |
5346 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
5347 | * means "infinite". */ | |
5348 | static void | |
5349 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
5350 | { | |
5351 | if (max && (!*timeout || *timeout > max)) { | |
5352 | *timeout = max; | |
5353 | } | |
5354 | } | |
5355 | ||
5356 | static void | |
5357 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
f25d0cf3 | 5358 | const struct ofpact_fin_timeout *oft) |
0e553d9c BP |
5359 | { |
5360 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
5361 | struct rule_dpif *rule = ctx->rule; | |
5362 | ||
f25d0cf3 BP |
5363 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); |
5364 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
0e553d9c BP |
5365 | } |
5366 | } | |
5367 | ||
21f7563c JP |
5368 | static bool |
5369 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
5370 | { | |
9e1fd49b BP |
5371 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
5372 | ? OFPUTIL_PC_NO_RECV_STP | |
5373 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
5374 | return false; |
5375 | } | |
5376 | ||
5377 | /* Only drop packets here if both forwarding and learning are | |
5378 | * disabled. If just learning is enabled, we need to have | |
5379 | * OFPP_NORMAL and the learning action have a look at the packet | |
5380 | * before we can drop it. */ | |
5381 | if (!stp_forward_in_state(port->stp_state) | |
5382 | && !stp_learn_in_state(port->stp_state)) { | |
5383 | return false; | |
5384 | } | |
5385 | ||
5386 | return true; | |
5387 | } | |
5388 | ||
abe529af | 5389 | static void |
f25d0cf3 | 5390 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, |
abe529af BP |
5391 | struct action_xlate_ctx *ctx) |
5392 | { | |
5393 | const struct ofport_dpif *port; | |
254750ce | 5394 | bool was_evictable = true; |
f25d0cf3 | 5395 | const struct ofpact *a; |
abe529af BP |
5396 | |
5397 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 5398 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
5399 | /* Drop this flow. */ |
5400 | return; | |
5401 | } | |
5402 | ||
254750ce BP |
5403 | if (ctx->rule) { |
5404 | /* Don't let the rule we're working on get evicted underneath us. */ | |
5405 | was_evictable = ctx->rule->up.evictable; | |
5406 | ctx->rule->up.evictable = false; | |
5407 | } | |
f25d0cf3 BP |
5408 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { |
5409 | struct ofpact_controller *controller; | |
38f2e360 | 5410 | |
848e8809 EJ |
5411 | if (ctx->exit) { |
5412 | break; | |
5413 | } | |
5414 | ||
f25d0cf3 BP |
5415 | switch (a->type) { |
5416 | case OFPACT_OUTPUT: | |
5417 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
5418 | ofpact_get_OUTPUT(a)->max_len); | |
5419 | break; | |
5420 | ||
5421 | case OFPACT_CONTROLLER: | |
5422 | controller = ofpact_get_CONTROLLER(a); | |
5423 | execute_controller_action(ctx, controller->max_len, | |
5424 | controller->reason, | |
5425 | controller->controller_id); | |
5426 | break; | |
690a61c5 | 5427 | |
f25d0cf3 BP |
5428 | case OFPACT_ENQUEUE: |
5429 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
abe529af BP |
5430 | break; |
5431 | ||
f25d0cf3 | 5432 | case OFPACT_SET_VLAN_VID: |
abe529af | 5433 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
f25d0cf3 BP |
5434 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) |
5435 | | htons(VLAN_CFI)); | |
abe529af BP |
5436 | break; |
5437 | ||
f25d0cf3 | 5438 | case OFPACT_SET_VLAN_PCP: |
abe529af | 5439 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
f25d0cf3 BP |
5440 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp |
5441 | << VLAN_PCP_SHIFT) | |
5442 | | VLAN_CFI); | |
abe529af BP |
5443 | break; |
5444 | ||
f25d0cf3 | 5445 | case OFPACT_STRIP_VLAN: |
abe529af | 5446 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
5447 | break; |
5448 | ||
f25d0cf3 BP |
5449 | case OFPACT_SET_ETH_SRC: |
5450 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
5451 | ETH_ADDR_LEN); | |
abe529af BP |
5452 | break; |
5453 | ||
f25d0cf3 BP |
5454 | case OFPACT_SET_ETH_DST: |
5455 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
5456 | ETH_ADDR_LEN); | |
abe529af BP |
5457 | break; |
5458 | ||
f25d0cf3 BP |
5459 | case OFPACT_SET_IPV4_SRC: |
5460 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
abe529af BP |
5461 | break; |
5462 | ||
f25d0cf3 BP |
5463 | case OFPACT_SET_IPV4_DST: |
5464 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
abe529af BP |
5465 | break; |
5466 | ||
f25d0cf3 | 5467 | case OFPACT_SET_IPV4_DSCP: |
c4f2731d PS |
5468 | /* OpenFlow 1.0 only supports IPv4. */ |
5469 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5470 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
f25d0cf3 | 5471 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; |
c4f2731d | 5472 | } |
abe529af BP |
5473 | break; |
5474 | ||
f25d0cf3 BP |
5475 | case OFPACT_SET_L4_SRC_PORT: |
5476 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
abe529af BP |
5477 | break; |
5478 | ||
f25d0cf3 BP |
5479 | case OFPACT_SET_L4_DST_PORT: |
5480 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
abe529af BP |
5481 | break; |
5482 | ||
f25d0cf3 BP |
5483 | case OFPACT_RESUBMIT: |
5484 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
38f2e360 BP |
5485 | break; |
5486 | ||
f25d0cf3 BP |
5487 | case OFPACT_SET_TUNNEL: |
5488 | ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
29901626 BP |
5489 | break; |
5490 | ||
f25d0cf3 BP |
5491 | case OFPACT_SET_QUEUE: |
5492 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
abe529af BP |
5493 | break; |
5494 | ||
f25d0cf3 | 5495 | case OFPACT_POP_QUEUE: |
deedf7e7 | 5496 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
5497 | break; |
5498 | ||
f25d0cf3 BP |
5499 | case OFPACT_REG_MOVE: |
5500 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
38f2e360 BP |
5501 | break; |
5502 | ||
f25d0cf3 BP |
5503 | case OFPACT_REG_LOAD: |
5504 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
38f2e360 BP |
5505 | break; |
5506 | ||
f25d0cf3 BP |
5507 | case OFPACT_DEC_TTL: |
5508 | if (compose_dec_ttl(ctx)) { | |
5509 | goto out; | |
5510 | } | |
38f2e360 BP |
5511 | break; |
5512 | ||
f25d0cf3 BP |
5513 | case OFPACT_NOTE: |
5514 | /* Nothing to do. */ | |
abe529af BP |
5515 | break; |
5516 | ||
f25d0cf3 BP |
5517 | case OFPACT_MULTIPATH: |
5518 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
abe529af | 5519 | break; |
daff3353 | 5520 | |
f25d0cf3 BP |
5521 | case OFPACT_AUTOPATH: |
5522 | xlate_autopath(ctx, ofpact_get_AUTOPATH(a)); | |
daff3353 | 5523 | break; |
a368bb53 | 5524 | |
f25d0cf3 | 5525 | case OFPACT_BUNDLE: |
a368bb53 | 5526 | ctx->ofproto->has_bundle_action = true; |
f25d0cf3 | 5527 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); |
a368bb53 | 5528 | break; |
f694937d | 5529 | |
f25d0cf3 BP |
5530 | case OFPACT_OUTPUT_REG: |
5531 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
f694937d | 5532 | break; |
75a75043 | 5533 | |
f25d0cf3 | 5534 | case OFPACT_LEARN: |
75a75043 | 5535 | ctx->has_learn = true; |
3de9590b | 5536 | if (ctx->may_learn) { |
f25d0cf3 | 5537 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); |
75a75043 BP |
5538 | } |
5539 | break; | |
848e8809 | 5540 | |
f25d0cf3 | 5541 | case OFPACT_EXIT: |
848e8809 EJ |
5542 | ctx->exit = true; |
5543 | break; | |
0e553d9c | 5544 | |
f25d0cf3 | 5545 | case OFPACT_FIN_TIMEOUT: |
0e553d9c | 5546 | ctx->has_fin_timeout = true; |
f25d0cf3 | 5547 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); |
a7349929 | 5548 | break; |
abe529af BP |
5549 | } |
5550 | } | |
21f7563c | 5551 | |
f0fd1a17 | 5552 | out: |
21f7563c JP |
5553 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
5554 | * so drop it now if forwarding is disabled. */ | |
5555 | if (port && !stp_forward_in_state(port->stp_state)) { | |
5556 | ofpbuf_clear(ctx->odp_actions); | |
5557 | add_sflow_action(ctx); | |
5558 | } | |
254750ce BP |
5559 | if (ctx->rule) { |
5560 | ctx->rule->up.evictable = was_evictable; | |
5561 | } | |
abe529af BP |
5562 | } |
5563 | ||
5564 | static void | |
5565 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
5566 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
18b2a258 | 5567 | ovs_be16 initial_tci, struct rule_dpif *rule, |
0e553d9c | 5568 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af BP |
5569 | { |
5570 | ctx->ofproto = ofproto; | |
5571 | ctx->flow = *flow; | |
e84173dc BP |
5572 | ctx->base_flow = ctx->flow; |
5573 | ctx->base_flow.tun_id = 0; | |
5574 | ctx->base_flow.vlan_tci = initial_tci; | |
18b2a258 | 5575 | ctx->rule = rule; |
abe529af | 5576 | ctx->packet = packet; |
3de9590b | 5577 | ctx->may_learn = packet != NULL; |
0e553d9c | 5578 | ctx->tcp_flags = tcp_flags; |
abe529af | 5579 | ctx->resubmit_hook = NULL; |
479df176 | 5580 | ctx->report_hook = NULL; |
112bc5f4 | 5581 | ctx->resubmit_stats = NULL; |
abe529af BP |
5582 | } |
5583 | ||
f25d0cf3 BP |
5584 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' |
5585 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
050ac423 | 5586 | static void |
abe529af | 5587 | xlate_actions(struct action_xlate_ctx *ctx, |
f25d0cf3 | 5588 | const struct ofpact *ofpacts, size_t ofpacts_len, |
050ac423 | 5589 | struct ofpbuf *odp_actions) |
abe529af | 5590 | { |
43d50bc8 BP |
5591 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so |
5592 | * that in the future we always keep a copy of the original flow for | |
5593 | * tracing purposes. */ | |
5594 | static bool hit_resubmit_limit; | |
5595 | ||
6a7e895f BP |
5596 | enum slow_path_reason special; |
5597 | ||
abe529af BP |
5598 | COVERAGE_INC(ofproto_dpif_xlate); |
5599 | ||
050ac423 BP |
5600 | ofpbuf_clear(odp_actions); |
5601 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
5602 | ||
5603 | ctx->odp_actions = odp_actions; | |
97e42c92 | 5604 | ctx->tags = 0; |
6a7e895f | 5605 | ctx->slow = 0; |
97e42c92 BP |
5606 | ctx->has_learn = false; |
5607 | ctx->has_normal = false; | |
0e553d9c | 5608 | ctx->has_fin_timeout = false; |
97e42c92 | 5609 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 5610 | ctx->mirrors = 0; |
97e42c92 | 5611 | ctx->recurse = 0; |
6a6455e5 | 5612 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 5613 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 5614 | ctx->table_id = 0; |
848e8809 | 5615 | ctx->exit = false; |
7257b535 | 5616 | |
43d50bc8 | 5617 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { |
ccb7c863 BP |
5618 | /* Do this conditionally because the copy is expensive enough that it |
5619 | * shows up in profiles. | |
5620 | * | |
5621 | * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4 | |
5622 | * believe that I wasn't using it without initializing it if I kept it | |
5623 | * in a local variable. */ | |
5624 | ctx->orig_flow = ctx->flow; | |
5625 | } | |
5626 | ||
eadef313 | 5627 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
5628 | switch (ctx->ofproto->up.frag_handling) { |
5629 | case OFPC_FRAG_NORMAL: | |
5630 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
5631 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
5632 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
5633 | break; |
5634 | ||
5635 | case OFPC_FRAG_DROP: | |
050ac423 | 5636 | return; |
7257b535 BP |
5637 | |
5638 | case OFPC_FRAG_REASM: | |
5639 | NOT_REACHED(); | |
5640 | ||
5641 | case OFPC_FRAG_NX_MATCH: | |
5642 | /* Nothing to do. */ | |
5643 | break; | |
f0fd1a17 PS |
5644 | |
5645 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
5646 | NOT_REACHED(); | |
7257b535 BP |
5647 | } |
5648 | } | |
5649 | ||
6a7e895f BP |
5650 | special = process_special(ctx->ofproto, &ctx->flow, ctx->packet); |
5651 | if (special) { | |
5652 | ctx->slow |= special; | |
abe529af | 5653 | } else { |
6a6455e5 | 5654 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
6a6455e5 EJ |
5655 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; |
5656 | ||
6ff686f2 | 5657 | add_sflow_action(ctx); |
f25d0cf3 | 5658 | do_xlate_actions(ofpacts, ofpacts_len, ctx); |
abe529af | 5659 | |
43d50bc8 BP |
5660 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { |
5661 | if (!hit_resubmit_limit) { | |
5662 | /* We didn't record the original flow. Make sure we do from | |
5663 | * now on. */ | |
5664 | hit_resubmit_limit = true; | |
5665 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
5666 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5667 | ||
5668 | ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet, | |
5669 | initial_tci, &ds); | |
5670 | VLOG_ERR("Trace triggered by excessive resubmit " | |
5671 | "recursion:\n%s", ds_cstr(&ds)); | |
5672 | ds_destroy(&ds); | |
5673 | } | |
6a6455e5 EJ |
5674 | } |
5675 | ||
b6848f13 BP |
5676 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
5677 | ctx->odp_actions->data, | |
5678 | ctx->odp_actions->size)) { | |
6a7e895f | 5679 | ctx->slow |= SLOW_IN_BAND; |
b6848f13 BP |
5680 | if (ctx->packet |
5681 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
5682 | ctx->packet)) { | |
5e48dc2b | 5683 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
5684 | } |
5685 | } | |
ccb7c863 BP |
5686 | if (ctx->ofproto->has_mirrors) { |
5687 | add_mirror_actions(ctx, &ctx->orig_flow); | |
5688 | } | |
a7c4eaf6 | 5689 | fix_sflow_action(ctx); |
abe529af | 5690 | } |
050ac423 BP |
5691 | } |
5692 | ||
f25d0cf3 BP |
5693 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' |
5694 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
050ac423 BP |
5695 | static void |
5696 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
f25d0cf3 BP |
5697 | const struct ofpact *ofpacts, |
5698 | size_t ofpacts_len) | |
050ac423 BP |
5699 | { |
5700 | uint64_t odp_actions_stub[1024 / 8]; | |
5701 | struct ofpbuf odp_actions; | |
abe529af | 5702 | |
050ac423 | 5703 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
f25d0cf3 | 5704 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); |
050ac423 | 5705 | ofpbuf_uninit(&odp_actions); |
abe529af | 5706 | } |
479df176 BP |
5707 | |
5708 | static void | |
5709 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
5710 | { | |
5711 | if (ctx->report_hook) { | |
5712 | ctx->report_hook(ctx, s); | |
5713 | } | |
5714 | } | |
abe529af BP |
5715 | \f |
5716 | /* OFPP_NORMAL implementation. */ | |
5717 | ||
abe529af BP |
5718 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
5719 | ||
ecac4ebf BP |
5720 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
5721 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5722 | * the bundle on which the packet was received, returns the VLAN to which the | |
5723 | * packet belongs. | |
5724 | * | |
5725 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5726 | static uint16_t | |
5727 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5728 | { | |
5729 | switch (in_bundle->vlan_mode) { | |
5730 | case PORT_VLAN_ACCESS: | |
5731 | return in_bundle->vlan; | |
5732 | break; | |
5733 | ||
5734 | case PORT_VLAN_TRUNK: | |
5735 | return vid; | |
5736 | ||
5737 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5738 | case PORT_VLAN_NATIVE_TAGGED: | |
5739 | return vid ? vid : in_bundle->vlan; | |
5740 | ||
5741 | default: | |
5742 | NOT_REACHED(); | |
5743 | } | |
5744 | } | |
5745 | ||
5da5ec37 BP |
5746 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5747 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5748 | * a warning. | |
5749 | * | |
5750 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5751 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5752 | * 0...4095. */ | |
5753 | static bool | |
5754 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5755 | { | |
33158a18 JP |
5756 | /* Allow any VID on the OFPP_NONE port. */ |
5757 | if (in_bundle == &ofpp_none_bundle) { | |
5758 | return true; | |
5759 | } | |
5760 | ||
5da5ec37 BP |
5761 | switch (in_bundle->vlan_mode) { |
5762 | case PORT_VLAN_ACCESS: | |
5763 | if (vid) { | |
5764 | if (warn) { | |
5765 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5766 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5767 | "packet received on port %s configured as VLAN " | |
5768 | "%"PRIu16" access port", | |
5769 | in_bundle->ofproto->up.name, vid, | |
5770 | in_bundle->name, in_bundle->vlan); | |
5771 | } | |
5772 | return false; | |
5773 | } | |
5774 | return true; | |
5775 | ||
5776 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5777 | case PORT_VLAN_NATIVE_TAGGED: | |
5778 | if (!vid) { | |
5779 | /* Port must always carry its native VLAN. */ | |
5780 | return true; | |
5781 | } | |
5782 | /* Fall through. */ | |
5783 | case PORT_VLAN_TRUNK: | |
5784 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5785 | if (warn) { | |
5786 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5787 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5788 | "received on port %s not configured for trunking " | |
5789 | "VLAN %"PRIu16, | |
5790 | in_bundle->ofproto->up.name, vid, | |
5791 | in_bundle->name, vid); | |
5792 | } | |
5793 | return false; | |
5794 | } | |
5795 | return true; | |
5796 | ||
5797 | default: | |
5798 | NOT_REACHED(); | |
5799 | } | |
5800 | ||
5801 | } | |
5802 | ||
ecac4ebf BP |
5803 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5804 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5805 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5806 | * then the 802.1Q header should only be included in the packet if there is a | |
5807 | * nonzero PCP.) | |
5808 | * | |
5809 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5810 | static uint16_t | |
5811 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5812 | { | |
5813 | switch (out_bundle->vlan_mode) { | |
5814 | case PORT_VLAN_ACCESS: | |
5815 | return 0; | |
5816 | ||
5817 | case PORT_VLAN_TRUNK: | |
5818 | case PORT_VLAN_NATIVE_TAGGED: | |
5819 | return vlan; | |
5820 | ||
5821 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5822 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5823 | ||
5824 | default: | |
5825 | NOT_REACHED(); | |
5826 | } | |
5827 | } | |
5828 | ||
395e68ce BP |
5829 | static void |
5830 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5831 | uint16_t vlan) | |
abe529af | 5832 | { |
395e68ce BP |
5833 | struct ofport_dpif *port; |
5834 | uint16_t vid; | |
81b1afb1 | 5835 | ovs_be16 tci, old_tci; |
ecac4ebf | 5836 | |
395e68ce BP |
5837 | vid = output_vlan_to_vid(out_bundle, vlan); |
5838 | if (!out_bundle->bond) { | |
5839 | port = ofbundle_get_a_port(out_bundle); | |
5840 | } else { | |
5841 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5842 | vid, &ctx->tags); | |
5843 | if (!port) { | |
5844 | /* No slaves enabled, so drop packet. */ | |
5845 | return; | |
5846 | } | |
5847 | } | |
abe529af | 5848 | |
81b1afb1 | 5849 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5850 | tci = htons(vid); |
5851 | if (tci || out_bundle->use_priority_tags) { | |
5852 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5853 | if (tci) { | |
5854 | tci |= htons(VLAN_CFI); | |
5855 | } | |
395e68ce | 5856 | } |
81b1afb1 | 5857 | ctx->flow.vlan_tci = tci; |
395e68ce | 5858 | |
5e48dc2b | 5859 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5860 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5861 | } |
5862 | ||
5863 | static int | |
5864 | mirror_mask_ffs(mirror_mask_t mask) | |
5865 | { | |
5866 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5867 | return ffs(mask); | |
5868 | } | |
5869 | ||
abe529af BP |
5870 | static bool |
5871 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5872 | { | |
ecac4ebf | 5873 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5874 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5875 | } |
5876 | ||
5877 | static bool | |
5878 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5879 | { | |
5880 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5881 | } | |
5882 | ||
5883 | /* Returns an arbitrary interface within 'bundle'. */ | |
5884 | static struct ofport_dpif * | |
5885 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5886 | { | |
5887 | return CONTAINER_OF(list_front(&bundle->ports), | |
5888 | struct ofport_dpif, bundle_node); | |
5889 | } | |
5890 | ||
abe529af BP |
5891 | static bool |
5892 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5893 | { | |
fc3d7408 | 5894 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5895 | } |
5896 | ||
5897 | static void | |
c06bba01 | 5898 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5899 | { |
5900 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5901 | mirror_mask_t mirrors; | |
c06bba01 JP |
5902 | struct ofbundle *in_bundle; |
5903 | uint16_t vlan; | |
5904 | uint16_t vid; | |
5905 | const struct nlattr *a; | |
5906 | size_t left; | |
5907 | ||
3581c12c | 5908 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
70c2fd56 | 5909 | ctx->packet != NULL, NULL); |
3581c12c | 5910 | if (!in_bundle) { |
c06bba01 JP |
5911 | return; |
5912 | } | |
c06bba01 JP |
5913 | mirrors = in_bundle->src_mirrors; |
5914 | ||
5915 | /* Drop frames on bundles reserved for mirroring. */ | |
5916 | if (in_bundle->mirror_out) { | |
5917 | if (ctx->packet != NULL) { | |
5918 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5919 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5920 | "%s, which is reserved exclusively for mirroring", | |
5921 | ctx->ofproto->up.name, in_bundle->name); | |
5922 | } | |
5923 | return; | |
5924 | } | |
5925 | ||
5926 | /* Check VLAN. */ | |
5927 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5928 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5929 | return; | |
5930 | } | |
5931 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5932 | ||
5933 | /* Look at the output ports to check for destination selections. */ | |
5934 | ||
5935 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5936 | ctx->odp_actions->size) { | |
5937 | enum ovs_action_attr type = nl_attr_type(a); | |
5938 | struct ofport_dpif *ofport; | |
5939 | ||
5940 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5941 | continue; | |
5942 | } | |
5943 | ||
5944 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5945 | if (ofport && ofport->bundle) { |
5946 | mirrors |= ofport->bundle->dst_mirrors; | |
5947 | } | |
c06bba01 | 5948 | } |
abe529af BP |
5949 | |
5950 | if (!mirrors) { | |
5951 | return; | |
5952 | } | |
5953 | ||
c06bba01 JP |
5954 | /* Restore the original packet before adding the mirror actions. */ |
5955 | ctx->flow = *orig_flow; | |
5956 | ||
9ba15e2a BP |
5957 | while (mirrors) { |
5958 | struct ofmirror *m; | |
9ba15e2a BP |
5959 | |
5960 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5961 | ||
5962 | if (!vlan_is_mirrored(m, vlan)) { | |
5963 | mirrors &= mirrors - 1; | |
5964 | continue; | |
5965 | } | |
5966 | ||
5967 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 5968 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 5969 | if (m->out) { |
395e68ce | 5970 | output_normal(ctx, m->out, vlan); |
614ec445 EJ |
5971 | } else if (vlan != m->out_vlan |
5972 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
9ba15e2a BP |
5973 | struct ofbundle *bundle; |
5974 | ||
5975 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
5976 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
5977 | && !bundle->mirror_out) { |
5978 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
5979 | } |
5980 | } | |
5981 | } | |
abe529af BP |
5982 | } |
5983 | } | |
5984 | ||
9d24de3b JP |
5985 | static void |
5986 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
5987 | uint64_t packets, uint64_t bytes) | |
5988 | { | |
5989 | if (!mirrors) { | |
5990 | return; | |
5991 | } | |
5992 | ||
5993 | for (; mirrors; mirrors &= mirrors - 1) { | |
5994 | struct ofmirror *m; | |
5995 | ||
5996 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5997 | ||
5998 | if (!m) { | |
5999 | /* In normal circumstances 'm' will not be NULL. However, | |
6000 | * if mirrors are reconfigured, we can temporarily get out | |
6001 | * of sync in facet_revalidate(). We could "correct" the | |
6002 | * mirror list before reaching here, but doing that would | |
6003 | * not properly account the traffic stats we've currently | |
6004 | * accumulated for previous mirror configuration. */ | |
6005 | continue; | |
6006 | } | |
6007 | ||
6008 | m->packet_count += packets; | |
6009 | m->byte_count += bytes; | |
6010 | } | |
6011 | } | |
6012 | ||
abe529af BP |
6013 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
6014 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
6015 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
6016 | static bool | |
6017 | is_gratuitous_arp(const struct flow *flow) | |
6018 | { | |
6019 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
6020 | && eth_addr_is_broadcast(flow->dl_dst) | |
6021 | && (flow->nw_proto == ARP_OP_REPLY | |
6022 | || (flow->nw_proto == ARP_OP_REQUEST | |
6023 | && flow->nw_src == flow->nw_dst))); | |
6024 | } | |
6025 | ||
6026 | static void | |
6027 | update_learning_table(struct ofproto_dpif *ofproto, | |
6028 | const struct flow *flow, int vlan, | |
6029 | struct ofbundle *in_bundle) | |
6030 | { | |
6031 | struct mac_entry *mac; | |
6032 | ||
33158a18 JP |
6033 | /* Don't learn the OFPP_NONE port. */ |
6034 | if (in_bundle == &ofpp_none_bundle) { | |
6035 | return; | |
6036 | } | |
6037 | ||
abe529af BP |
6038 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
6039 | return; | |
6040 | } | |
6041 | ||
6042 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
6043 | if (is_gratuitous_arp(flow)) { | |
6044 | /* We don't want to learn from gratuitous ARP packets that are | |
6045 | * reflected back over bond slaves so we lock the learning table. */ | |
6046 | if (!in_bundle->bond) { | |
6047 | mac_entry_set_grat_arp_lock(mac); | |
6048 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
6049 | return; | |
6050 | } | |
6051 | } | |
6052 | ||
6053 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
6054 | /* The log messages here could actually be useful in debugging, | |
6055 | * so keep the rate limit relatively high. */ | |
6056 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
6057 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
6058 | "on port %s in VLAN %d", | |
6059 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
6060 | in_bundle->name, vlan); | |
6061 | ||
6062 | mac->port.p = in_bundle; | |
6063 | tag_set_add(&ofproto->revalidate_set, | |
6064 | mac_learning_changed(ofproto->ml, mac)); | |
6065 | } | |
6066 | } | |
6067 | ||
3581c12c | 6068 | static struct ofbundle * |
4acbc98d SH |
6069 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, |
6070 | bool warn, struct ofport_dpif **in_ofportp) | |
395e68ce BP |
6071 | { |
6072 | struct ofport_dpif *ofport; | |
6073 | ||
6074 | /* Find the port and bundle for the received packet. */ | |
6075 | ofport = get_ofp_port(ofproto, in_port); | |
70c2fd56 BP |
6076 | if (in_ofportp) { |
6077 | *in_ofportp = ofport; | |
6078 | } | |
395e68ce | 6079 | if (ofport && ofport->bundle) { |
3581c12c | 6080 | return ofport->bundle; |
395e68ce BP |
6081 | } |
6082 | ||
70c2fd56 BP |
6083 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
6084 | * port for traffic that it is sourcing. */ | |
6085 | if (in_port == OFPP_NONE) { | |
6086 | return &ofpp_none_bundle; | |
6087 | } | |
6088 | ||
395e68ce BP |
6089 | /* Odd. A few possible reasons here: |
6090 | * | |
6091 | * - We deleted a port but there are still a few packets queued up | |
6092 | * from it. | |
6093 | * | |
6094 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
6095 | * we don't know about. | |
6096 | * | |
6097 | * - The ofproto client didn't configure the port as part of a bundle. | |
6b803ddc EJ |
6098 | * This is particularly likely to happen if a packet was received on the |
6099 | * port after it was created, but before the client had a chance to | |
6100 | * configure its bundle. | |
395e68ce BP |
6101 | */ |
6102 | if (warn) { | |
6103 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6104 | ||
6105 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
6106 | "port %"PRIu16, ofproto->up.name, in_port); | |
6107 | } | |
6108 | return NULL; | |
6109 | } | |
6110 | ||
5da5ec37 | 6111 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
6112 | * dropped. Returns true if they may be forwarded, false if they should be |
6113 | * dropped. | |
6114 | * | |
395e68ce BP |
6115 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
6116 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 6117 | * |
395e68ce BP |
6118 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
6119 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
6120 | * checked by input_vid_is_valid(). | |
abe529af BP |
6121 | * |
6122 | * May also add tags to '*tags', although the current implementation only does | |
6123 | * so in one special case. | |
6124 | */ | |
6125 | static bool | |
479df176 BP |
6126 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, |
6127 | uint16_t vlan) | |
abe529af | 6128 | { |
479df176 BP |
6129 | struct ofproto_dpif *ofproto = ctx->ofproto; |
6130 | struct flow *flow = &ctx->flow; | |
395e68ce | 6131 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 6132 | |
395e68ce BP |
6133 | /* Drop frames for reserved multicast addresses |
6134 | * only if forward_bpdu option is absent. */ | |
614ec445 | 6135 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { |
479df176 | 6136 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); |
abe529af BP |
6137 | return false; |
6138 | } | |
6139 | ||
abe529af BP |
6140 | if (in_bundle->bond) { |
6141 | struct mac_entry *mac; | |
6142 | ||
6143 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
479df176 | 6144 | flow->dl_dst, &ctx->tags)) { |
abe529af BP |
6145 | case BV_ACCEPT: |
6146 | break; | |
6147 | ||
6148 | case BV_DROP: | |
479df176 | 6149 | xlate_report(ctx, "bonding refused admissibility, dropping"); |
abe529af BP |
6150 | return false; |
6151 | ||
6152 | case BV_DROP_IF_MOVED: | |
6153 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
6154 | if (mac && mac->port.p != in_bundle && | |
6155 | (!is_gratuitous_arp(flow) | |
6156 | || mac_entry_is_grat_arp_locked(mac))) { | |
479df176 BP |
6157 | xlate_report(ctx, "SLB bond thinks this packet looped back, " |
6158 | "dropping"); | |
abe529af BP |
6159 | return false; |
6160 | } | |
6161 | break; | |
6162 | } | |
6163 | } | |
6164 | ||
6165 | return true; | |
6166 | } | |
6167 | ||
4cd78906 | 6168 | static void |
abe529af BP |
6169 | xlate_normal(struct action_xlate_ctx *ctx) |
6170 | { | |
395e68ce | 6171 | struct ofport_dpif *in_port; |
abe529af | 6172 | struct ofbundle *in_bundle; |
abe529af | 6173 | struct mac_entry *mac; |
395e68ce BP |
6174 | uint16_t vlan; |
6175 | uint16_t vid; | |
abe529af | 6176 | |
75a75043 BP |
6177 | ctx->has_normal = true; |
6178 | ||
3581c12c | 6179 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
70c2fd56 | 6180 | ctx->packet != NULL, &in_port); |
3581c12c | 6181 | if (!in_bundle) { |
479df176 | 6182 | xlate_report(ctx, "no input bundle, dropping"); |
395e68ce BP |
6183 | return; |
6184 | } | |
3581c12c | 6185 | |
395e68ce BP |
6186 | /* Drop malformed frames. */ |
6187 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
6188 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
6189 | if (ctx->packet != NULL) { | |
6190 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6191 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
6192 | "VLAN tag received on port %s", | |
6193 | ctx->ofproto->up.name, in_bundle->name); | |
6194 | } | |
479df176 | 6195 | xlate_report(ctx, "partial VLAN tag, dropping"); |
395e68ce BP |
6196 | return; |
6197 | } | |
6198 | ||
6199 | /* Drop frames on bundles reserved for mirroring. */ | |
6200 | if (in_bundle->mirror_out) { | |
6201 | if (ctx->packet != NULL) { | |
6202 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6203 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
6204 | "%s, which is reserved exclusively for mirroring", | |
6205 | ctx->ofproto->up.name, in_bundle->name); | |
6206 | } | |
479df176 | 6207 | xlate_report(ctx, "input port is mirror output port, dropping"); |
395e68ce BP |
6208 | return; |
6209 | } | |
6210 | ||
6211 | /* Check VLAN. */ | |
6212 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
6213 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
479df176 | 6214 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); |
395e68ce BP |
6215 | return; |
6216 | } | |
6217 | vlan = input_vid_to_vlan(in_bundle, vid); | |
6218 | ||
6219 | /* Check other admissibility requirements. */ | |
479df176 | 6220 | if (in_port && !is_admissible(ctx, in_port, vlan)) { |
395e68ce | 6221 | return; |
abe529af BP |
6222 | } |
6223 | ||
75a75043 | 6224 | /* Learn source MAC. */ |
3de9590b | 6225 | if (ctx->may_learn) { |
abe529af BP |
6226 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
6227 | } | |
6228 | ||
6229 | /* Determine output bundle. */ | |
6230 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
6231 | &ctx->tags); | |
6232 | if (mac) { | |
c06bba01 | 6233 | if (mac->port.p != in_bundle) { |
479df176 | 6234 | xlate_report(ctx, "forwarding to learned port"); |
c06bba01 | 6235 | output_normal(ctx, mac->port.p, vlan); |
479df176 BP |
6236 | } else { |
6237 | xlate_report(ctx, "learned port is input port, dropping"); | |
c06bba01 | 6238 | } |
abe529af | 6239 | } else { |
c06bba01 | 6240 | struct ofbundle *bundle; |
abe529af | 6241 | |
479df176 | 6242 | xlate_report(ctx, "no learned MAC for destination, flooding"); |
c06bba01 JP |
6243 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
6244 | if (bundle != in_bundle | |
6245 | && ofbundle_includes_vlan(bundle, vlan) | |
6246 | && bundle->floodable | |
6247 | && !bundle->mirror_out) { | |
6248 | output_normal(ctx, bundle, vlan); | |
6249 | } | |
6250 | } | |
6251 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 6252 | } |
abe529af BP |
6253 | } |
6254 | \f | |
54a9cbc9 BP |
6255 | /* Optimized flow revalidation. |
6256 | * | |
6257 | * It's a difficult problem, in general, to tell which facets need to have | |
6258 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
6259 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
6260 | * table changes, we recalculate the actions for every facet. This is | |
6261 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
6262 | * doesn't change very often. | |
6263 | * | |
6264 | * However, we can expect one particular kind of OpenFlow flow table change to | |
6265 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
6266 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
6267 | * table, we add a special case that applies to flow tables in which every rule | |
6268 | * has the same form (that is, the same wildcards), except that the table is | |
6269 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
6270 | * optimize this case by tagging all of the facets that resubmit into the table | |
6271 | * and invalidating the same tag whenever a flow changes in that table. The | |
6272 | * end result is that we revalidate just the facets that need it (and sometimes | |
6273 | * a few more, but not all of the facets or even all of the facets that | |
6274 | * resubmit to the table modified by MAC learning). */ | |
6275 | ||
6276 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
6277 | * into an OpenFlow table with the given 'basis'. */ | |
822d9414 | 6278 | static tag_type |
54a9cbc9 BP |
6279 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, |
6280 | uint32_t secret) | |
6281 | { | |
6282 | if (flow_wildcards_is_catchall(wc)) { | |
6283 | return 0; | |
6284 | } else { | |
6285 | struct flow tag_flow = *flow; | |
6286 | flow_zero_wildcards(&tag_flow, wc); | |
6287 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
6288 | } | |
6289 | } | |
6290 | ||
6291 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
6292 | * taggability of that table. | |
6293 | * | |
6294 | * This function must be called after *each* change to a flow table. If you | |
6295 | * skip calling it on some changes then the pointer comparisons at the end can | |
6296 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
6297 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
6298 | * different wildcards to be created with the same address, then this function | |
6299 | * will incorrectly skip revalidation. */ | |
6300 | static void | |
6301 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
6302 | { | |
6303 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 6304 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
6305 | struct cls_table *catchall, *other; |
6306 | struct cls_table *t; | |
6307 | ||
6308 | catchall = other = NULL; | |
6309 | ||
d0918789 | 6310 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
6311 | case 0: |
6312 | /* We could tag this OpenFlow table but it would make the logic a | |
6313 | * little harder and it's a corner case that doesn't seem worth it | |
6314 | * yet. */ | |
6315 | break; | |
6316 | ||
6317 | case 1: | |
6318 | case 2: | |
d0918789 | 6319 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
6320 | if (cls_table_is_catchall(t)) { |
6321 | catchall = t; | |
6322 | } else if (!other) { | |
6323 | other = t; | |
6324 | } else { | |
6325 | /* Indicate that we can't tag this by setting both tables to | |
6326 | * NULL. (We know that 'catchall' is already NULL.) */ | |
6327 | other = NULL; | |
6328 | } | |
6329 | } | |
6330 | break; | |
6331 | ||
6332 | default: | |
6333 | /* Can't tag this table. */ | |
6334 | break; | |
6335 | } | |
6336 | ||
6337 | if (table->catchall_table != catchall || table->other_table != other) { | |
6338 | table->catchall_table = catchall; | |
6339 | table->other_table = other; | |
3c4a309c | 6340 | ofproto->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
6341 | } |
6342 | } | |
6343 | ||
6344 | /* Given 'rule' that has changed in some way (either it is a rule being | |
6345 | * inserted, a rule being deleted, or a rule whose actions are being | |
6346 | * modified), marks facets for revalidation to ensure that packets will be | |
6347 | * forwarded correctly according to the new state of the flow table. | |
6348 | * | |
6349 | * This function must be called after *each* change to a flow table. See | |
6350 | * the comment on table_update_taggable() for more information. */ | |
6351 | static void | |
6352 | rule_invalidate(const struct rule_dpif *rule) | |
6353 | { | |
6354 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
6355 | ||
6356 | table_update_taggable(ofproto, rule->up.table_id); | |
6357 | ||
6358 | if (!ofproto->need_revalidate) { | |
6359 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
6360 | ||
6361 | if (table->other_table && rule->tag) { | |
6362 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
6363 | } else { | |
3c4a309c | 6364 | ofproto->need_revalidate = REV_FLOW_TABLE; |
54a9cbc9 BP |
6365 | } |
6366 | } | |
6367 | } | |
6368 | \f | |
abe529af | 6369 | static bool |
7257b535 BP |
6370 | set_frag_handling(struct ofproto *ofproto_, |
6371 | enum ofp_config_flags frag_handling) | |
abe529af BP |
6372 | { |
6373 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 6374 | |
7257b535 | 6375 | if (frag_handling != OFPC_FRAG_REASM) { |
3c4a309c | 6376 | ofproto->need_revalidate = REV_RECONFIGURE; |
7257b535 BP |
6377 | return true; |
6378 | } else { | |
6379 | return false; | |
6380 | } | |
abe529af BP |
6381 | } |
6382 | ||
90bf1e07 | 6383 | static enum ofperr |
abe529af BP |
6384 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
6385 | const struct flow *flow, | |
f25d0cf3 | 6386 | const struct ofpact *ofpacts, size_t ofpacts_len) |
abe529af BP |
6387 | { |
6388 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
90bf1e07 | 6389 | enum ofperr error; |
abe529af | 6390 | |
e1154f71 | 6391 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
90bf1e07 | 6392 | return OFPERR_NXBRC_BAD_IN_PORT; |
e1154f71 BP |
6393 | } |
6394 | ||
f25d0cf3 | 6395 | error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports); |
abe529af | 6396 | if (!error) { |
80e5eed9 | 6397 | struct odputil_keybuf keybuf; |
112bc5f4 BP |
6398 | struct dpif_flow_stats stats; |
6399 | ||
80e5eed9 BP |
6400 | struct ofpbuf key; |
6401 | ||
112bc5f4 | 6402 | struct action_xlate_ctx ctx; |
050ac423 BP |
6403 | uint64_t odp_actions_stub[1024 / 8]; |
6404 | struct ofpbuf odp_actions; | |
050ac423 | 6405 | |
80e5eed9 BP |
6406 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
6407 | odp_flow_key_from_flow(&key, flow); | |
abe529af | 6408 | |
112bc5f4 | 6409 | dpif_flow_stats_extract(flow, packet, &stats); |
2284188b | 6410 | |
112bc5f4 BP |
6411 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, |
6412 | packet_get_tcp_flags(packet, flow), packet); | |
6413 | ctx.resubmit_stats = &stats; | |
2284188b | 6414 | |
050ac423 BP |
6415 | ofpbuf_use_stub(&odp_actions, |
6416 | odp_actions_stub, sizeof odp_actions_stub); | |
f25d0cf3 | 6417 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); |
80e5eed9 | 6418 | dpif_execute(ofproto->dpif, key.data, key.size, |
050ac423 BP |
6419 | odp_actions.data, odp_actions.size, packet); |
6420 | ofpbuf_uninit(&odp_actions); | |
abe529af BP |
6421 | } |
6422 | return error; | |
6423 | } | |
6fca1ffb BP |
6424 | \f |
6425 | /* NetFlow. */ | |
6426 | ||
6427 | static int | |
6428 | set_netflow(struct ofproto *ofproto_, | |
6429 | const struct netflow_options *netflow_options) | |
6430 | { | |
6431 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6432 | ||
6433 | if (netflow_options) { | |
6434 | if (!ofproto->netflow) { | |
6435 | ofproto->netflow = netflow_create(); | |
6436 | } | |
6437 | return netflow_set_options(ofproto->netflow, netflow_options); | |
6438 | } else { | |
6439 | netflow_destroy(ofproto->netflow); | |
6440 | ofproto->netflow = NULL; | |
6441 | return 0; | |
6442 | } | |
6443 | } | |
abe529af BP |
6444 | |
6445 | static void | |
6446 | get_netflow_ids(const struct ofproto *ofproto_, | |
6447 | uint8_t *engine_type, uint8_t *engine_id) | |
6448 | { | |
6449 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6450 | ||
6451 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
6452 | } | |
6fca1ffb BP |
6453 | |
6454 | static void | |
6455 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
6456 | { | |
6457 | if (!facet_is_controller_flow(facet) && | |
6458 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 6459 | struct subfacet *subfacet; |
6fca1ffb BP |
6460 | struct ofexpired expired; |
6461 | ||
b0f7b9b5 | 6462 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6a7e895f | 6463 | if (subfacet->path == SF_FAST_PATH) { |
b0f7b9b5 | 6464 | struct dpif_flow_stats stats; |
6fca1ffb | 6465 | |
6a7e895f | 6466 | subfacet_reinstall(subfacet, &stats); |
15baa734 | 6467 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 6468 | } |
6fca1ffb BP |
6469 | } |
6470 | ||
6471 | expired.flow = facet->flow; | |
6472 | expired.packet_count = facet->packet_count; | |
6473 | expired.byte_count = facet->byte_count; | |
6474 | expired.used = facet->used; | |
6475 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
6476 | } | |
6477 | } | |
6478 | ||
6479 | static void | |
6480 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
6481 | { | |
6482 | struct facet *facet; | |
6483 | ||
6484 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6485 | send_active_timeout(ofproto, facet); | |
6486 | } | |
6487 | } | |
abe529af BP |
6488 | \f |
6489 | static struct ofproto_dpif * | |
6490 | ofproto_dpif_lookup(const char *name) | |
6491 | { | |
b44a10b7 BP |
6492 | struct ofproto_dpif *ofproto; |
6493 | ||
6494 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
6495 | hash_string(name, 0), &all_ofproto_dpifs) { | |
6496 | if (!strcmp(ofproto->up.name, name)) { | |
6497 | return ofproto; | |
6498 | } | |
6499 | } | |
6500 | return NULL; | |
abe529af BP |
6501 | } |
6502 | ||
f0a3aa2e | 6503 | static void |
96e466a3 | 6504 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 6505 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 6506 | { |
490df1ef | 6507 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 6508 | |
96e466a3 EJ |
6509 | if (argc > 1) { |
6510 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6511 | if (!ofproto) { | |
bde9f75d | 6512 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
6513 | return; |
6514 | } | |
d0040604 | 6515 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 EJ |
6516 | } else { |
6517 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
d0040604 | 6518 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 | 6519 | } |
f0a3aa2e | 6520 | } |
f0a3aa2e | 6521 | |
bde9f75d | 6522 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
6523 | } |
6524 | ||
abe529af | 6525 | static void |
0e15264f BP |
6526 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6527 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
6528 | { |
6529 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6530 | const struct ofproto_dpif *ofproto; | |
6531 | const struct mac_entry *e; | |
6532 | ||
0e15264f | 6533 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 6534 | if (!ofproto) { |
bde9f75d | 6535 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
6536 | return; |
6537 | } | |
6538 | ||
6539 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
6540 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
6541 | struct ofbundle *bundle = e->port.p; | |
6542 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
6543 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
6544 | e->vlan, ETH_ADDR_ARGS(e->mac), |
6545 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 6546 | } |
bde9f75d | 6547 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
6548 | ds_destroy(&ds); |
6549 | } | |
6550 | ||
6a6455e5 | 6551 | struct trace_ctx { |
abe529af BP |
6552 | struct action_xlate_ctx ctx; |
6553 | struct flow flow; | |
6554 | struct ds *result; | |
6555 | }; | |
6556 | ||
6557 | static void | |
29901626 BP |
6558 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
6559 | const struct rule_dpif *rule) | |
abe529af BP |
6560 | { |
6561 | ds_put_char_multiple(result, '\t', level); | |
6562 | if (!rule) { | |
6563 | ds_put_cstr(result, "No match\n"); | |
6564 | return; | |
6565 | } | |
6566 | ||
29901626 BP |
6567 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
6568 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 6569 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
6570 | ds_put_char(result, '\n'); |
6571 | ||
6572 | ds_put_char_multiple(result, '\t', level); | |
6573 | ds_put_cstr(result, "OpenFlow "); | |
f25d0cf3 | 6574 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); |
abe529af BP |
6575 | ds_put_char(result, '\n'); |
6576 | } | |
6577 | ||
6578 | static void | |
6579 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 6580 | struct trace_ctx *trace) |
abe529af BP |
6581 | { |
6582 | ds_put_char_multiple(result, '\t', level); | |
6583 | ds_put_format(result, "%s: ", title); | |
6584 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
6585 | ds_put_cstr(result, "unchanged"); | |
6586 | } else { | |
6587 | flow_format(result, &trace->ctx.flow); | |
6588 | trace->flow = trace->ctx.flow; | |
6589 | } | |
6590 | ds_put_char(result, '\n'); | |
6591 | } | |
6592 | ||
eb9e1c26 EJ |
6593 | static void |
6594 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 6595 | struct trace_ctx *trace) |
eb9e1c26 EJ |
6596 | { |
6597 | size_t i; | |
6598 | ||
6599 | ds_put_char_multiple(result, '\t', level); | |
6600 | ds_put_format(result, "%s:", title); | |
6601 | for (i = 0; i < FLOW_N_REGS; i++) { | |
6602 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
6603 | } | |
6604 | ds_put_char(result, '\n'); | |
6605 | } | |
6606 | ||
1ed8d352 EJ |
6607 | static void |
6608 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 6609 | struct trace_ctx *trace) |
1ed8d352 EJ |
6610 | { |
6611 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
6612 | ||
6613 | ds_put_char_multiple(result, '\t', level); | |
6614 | ds_put_format(result, "%s: ", title); | |
6615 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
6616 | ds_put_char(result, '\n'); | |
6617 | } | |
6618 | ||
abe529af BP |
6619 | static void |
6620 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6621 | { | |
6a6455e5 | 6622 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
6623 | struct ds *result = trace->result; |
6624 | ||
6625 | ds_put_char(result, '\n'); | |
6626 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 6627 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 6628 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 6629 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
6630 | } |
6631 | ||
479df176 BP |
6632 | static void |
6633 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
6634 | { | |
6635 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
6636 | struct ds *result = trace->result; | |
6637 | ||
6638 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
6639 | ds_put_cstr(result, s); | |
6640 | ds_put_char(result, '\n'); | |
6641 | } | |
6642 | ||
abe529af | 6643 | static void |
0e15264f | 6644 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
6645 | void *aux OVS_UNUSED) |
6646 | { | |
0e15264f | 6647 | const char *dpname = argv[1]; |
abe529af | 6648 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
6649 | struct ofpbuf odp_key; |
6650 | struct ofpbuf *packet; | |
e84173dc | 6651 | ovs_be16 initial_tci; |
abe529af BP |
6652 | struct ds result; |
6653 | struct flow flow; | |
abe529af BP |
6654 | char *s; |
6655 | ||
876b0e1c BP |
6656 | packet = NULL; |
6657 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
6658 | ds_init(&result); |
6659 | ||
e84173dc BP |
6660 | ofproto = ofproto_dpif_lookup(dpname); |
6661 | if (!ofproto) { | |
bde9f75d EJ |
6662 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
6663 | "for help)"); | |
e84173dc BP |
6664 | goto exit; |
6665 | } | |
0e15264f | 6666 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 6667 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
6668 | const char *flow_s = argv[2]; |
6669 | const char *generate_s = argv[3]; | |
876b0e1c | 6670 | |
31a19d69 BP |
6671 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like |
6672 | * flow. We guess which type it is based on whether 'flow_s' contains | |
6673 | * an '(', since a datapath flow always contains '(') but an | |
6674 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
6675 | * that's not documented anywhere). | |
6676 | * | |
6677 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
6678 | * it would be tricky giving a sensible error message. After all, do | |
6679 | * you just say "syntax error" or do you present both error messages? | |
6680 | * Both choices seem lousy. */ | |
6681 | if (strchr(flow_s, '(')) { | |
6682 | int error; | |
6683 | ||
6684 | /* Convert string to datapath key. */ | |
6685 | ofpbuf_init(&odp_key, 0); | |
6686 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
6687 | if (error) { | |
6688 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
6689 | goto exit; | |
6690 | } | |
876b0e1c | 6691 | |
31a19d69 BP |
6692 | /* Convert odp_key to flow. */ |
6693 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, | |
6694 | odp_key.size, &flow, | |
6695 | &initial_tci, NULL); | |
6696 | if (error == ODP_FIT_ERROR) { | |
6697 | unixctl_command_reply_error(conn, "Invalid flow"); | |
6698 | goto exit; | |
6699 | } | |
6700 | } else { | |
6701 | char *error_s; | |
6702 | ||
6703 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
6704 | if (error_s) { | |
6705 | unixctl_command_reply_error(conn, error_s); | |
6706 | free(error_s); | |
6707 | goto exit; | |
6708 | } | |
6709 | ||
6710 | initial_tci = flow.vlan_tci; | |
6711 | vsp_adjust_flow(ofproto, &flow); | |
876b0e1c | 6712 | } |
8b3b8dd1 BP |
6713 | |
6714 | /* Generate a packet, if requested. */ | |
0e15264f | 6715 | if (generate_s) { |
8b3b8dd1 BP |
6716 | packet = ofpbuf_new(0); |
6717 | flow_compose(packet, &flow); | |
6718 | } | |
0e15264f | 6719 | } else if (argc == 6) { |
abff858b | 6720 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
6721 | const char *priority_s = argv[2]; |
6722 | const char *tun_id_s = argv[3]; | |
6723 | const char *in_port_s = argv[4]; | |
6724 | const char *packet_s = argv[5]; | |
6725 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
6726 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
6727 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 6728 | const char *msg; |
0e15264f | 6729 | |
e22f1753 BP |
6730 | msg = eth_from_hex(packet_s, &packet); |
6731 | if (msg) { | |
bde9f75d | 6732 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
6733 | goto exit; |
6734 | } | |
6735 | ||
6736 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 6737 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
6738 | ds_put_cstr(&result, s); |
6739 | free(s); | |
6740 | ||
abff858b | 6741 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 6742 | initial_tci = flow.vlan_tci; |
876b0e1c | 6743 | } else { |
bde9f75d | 6744 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
6745 | goto exit; |
6746 | } | |
6747 | ||
6a6455e5 EJ |
6748 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); |
6749 | unixctl_command_reply(conn, ds_cstr(&result)); | |
6750 | ||
6751 | exit: | |
6752 | ds_destroy(&result); | |
6753 | ofpbuf_delete(packet); | |
6754 | ofpbuf_uninit(&odp_key); | |
6755 | } | |
6756 | ||
6757 | static void | |
6758 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
6759 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
6760 | struct ds *ds) | |
6761 | { | |
6762 | struct rule_dpif *rule; | |
6763 | ||
6764 | ds_put_cstr(ds, "Flow: "); | |
6765 | flow_format(ds, flow); | |
6766 | ds_put_char(ds, '\n'); | |
abe529af | 6767 | |
c57b2226 BP |
6768 | rule = rule_dpif_lookup(ofproto, flow); |
6769 | ||
6a6455e5 | 6770 | trace_format_rule(ds, 0, 0, rule); |
c57b2226 BP |
6771 | if (rule == ofproto->miss_rule) { |
6772 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
6773 | } else if (rule == ofproto->no_packet_in_rule) { | |
6774 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
6775 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
6776 | } | |
6777 | ||
abe529af | 6778 | if (rule) { |
050ac423 BP |
6779 | uint64_t odp_actions_stub[1024 / 8]; |
6780 | struct ofpbuf odp_actions; | |
6781 | ||
6a6455e5 | 6782 | struct trace_ctx trace; |
0e553d9c | 6783 | uint8_t tcp_flags; |
abe529af | 6784 | |
6a6455e5 EJ |
6785 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
6786 | trace.result = ds; | |
6787 | trace.flow = *flow; | |
050ac423 BP |
6788 | ofpbuf_use_stub(&odp_actions, |
6789 | odp_actions_stub, sizeof odp_actions_stub); | |
6a6455e5 | 6790 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, |
0e553d9c | 6791 | rule, tcp_flags, packet); |
abe529af | 6792 | trace.ctx.resubmit_hook = trace_resubmit; |
479df176 | 6793 | trace.ctx.report_hook = trace_report; |
f25d0cf3 | 6794 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, |
050ac423 | 6795 | &odp_actions); |
abe529af | 6796 | |
6a6455e5 EJ |
6797 | ds_put_char(ds, '\n'); |
6798 | trace_format_flow(ds, 0, "Final flow", &trace); | |
6799 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
6800 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
6801 | ofpbuf_uninit(&odp_actions); | |
876b0e1c | 6802 | |
6a7e895f BP |
6803 | if (trace.ctx.slow) { |
6804 | enum slow_path_reason slow; | |
6805 | ||
6806 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
6807 | "slow path because it:"); | |
6808 | for (slow = trace.ctx.slow; slow; ) { | |
6809 | enum slow_path_reason bit = rightmost_1bit(slow); | |
6810 | ||
6811 | switch (bit) { | |
6812 | case SLOW_CFM: | |
6813 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
6814 | break; | |
6815 | case SLOW_LACP: | |
6816 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
6817 | break; | |
6818 | case SLOW_STP: | |
6819 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
6820 | break; | |
6821 | case SLOW_IN_BAND: | |
6822 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
6823 | "processing."); | |
6824 | if (!packet) { | |
6825 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
6826 | "incomplete--for complete actions, " | |
6827 | "please supply a packet.)"); | |
6828 | } | |
6829 | break; | |
6830 | case SLOW_CONTROLLER: | |
6831 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
6832 | "to the OpenFlow controller."); | |
6833 | break; | |
6834 | case SLOW_MATCH: | |
6835 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
6836 | "than the datapath supports."); | |
6837 | break; | |
6838 | } | |
6839 | ||
6840 | slow &= ~bit; | |
6841 | } | |
6842 | ||
6843 | if (slow & ~SLOW_MATCH) { | |
6844 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
6845 | "the special slow-path processing."); | |
876b0e1c BP |
6846 | } |
6847 | } | |
abe529af | 6848 | } |
abe529af BP |
6849 | } |
6850 | ||
7ee20df1 | 6851 | static void |
0e15264f BP |
6852 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6853 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6854 | { |
6855 | clogged = true; | |
bde9f75d | 6856 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6857 | } |
6858 | ||
6859 | static void | |
0e15264f BP |
6860 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6861 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6862 | { |
6863 | clogged = false; | |
bde9f75d | 6864 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6865 | } |
6866 | ||
6814e51f BP |
6867 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6868 | * 'reply' describing the results. */ | |
6869 | static void | |
6870 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6871 | { | |
6872 | struct facet *facet; | |
6873 | int errors; | |
6874 | ||
6875 | errors = 0; | |
6876 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6877 | if (!facet_check_consistency(facet)) { | |
6878 | errors++; | |
6879 | } | |
6880 | } | |
6881 | if (errors) { | |
3c4a309c | 6882 | ofproto->need_revalidate = REV_INCONSISTENCY; |
6814e51f BP |
6883 | } |
6884 | ||
6885 | if (errors) { | |
6886 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6887 | ofproto->up.name, errors); | |
6888 | } else { | |
6889 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6890 | } | |
6891 | } | |
6892 | ||
6893 | static void | |
6894 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6895 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6896 | { | |
6897 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6898 | struct ofproto_dpif *ofproto; | |
6899 | ||
6900 | if (argc > 1) { | |
6901 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6902 | if (!ofproto) { | |
bde9f75d EJ |
6903 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
6904 | "ofproto/list for help)"); | |
6814e51f BP |
6905 | return; |
6906 | } | |
6907 | ofproto_dpif_self_check__(ofproto, &reply); | |
6908 | } else { | |
6909 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6910 | ofproto_dpif_self_check__(ofproto, &reply); | |
6911 | } | |
6912 | } | |
6913 | ||
bde9f75d | 6914 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
6915 | ds_destroy(&reply); |
6916 | } | |
6917 | ||
abe529af BP |
6918 | static void |
6919 | ofproto_dpif_unixctl_init(void) | |
6920 | { | |
6921 | static bool registered; | |
6922 | if (registered) { | |
6923 | return; | |
6924 | } | |
6925 | registered = true; | |
6926 | ||
0e15264f BP |
6927 | unixctl_command_register( |
6928 | "ofproto/trace", | |
6929 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
aa3080c9 | 6930 | 2, 5, ofproto_unixctl_trace, NULL); |
96e466a3 | 6931 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6932 | ofproto_unixctl_fdb_flush, NULL); |
6933 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6934 | ofproto_unixctl_fdb_show, NULL); | |
6935 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6936 | ofproto_dpif_clog, NULL); | |
6937 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6938 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6939 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6940 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6941 | } |
6942 | \f | |
52a90c29 BP |
6943 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6944 | * | |
6945 | * This is deprecated. It is only for compatibility with broken device drivers | |
6946 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6947 | * devices are not used. When broken device drivers are no longer in | |
6948 | * widespread use, we will delete these interfaces. */ | |
6949 | ||
6950 | static int | |
6951 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6952 | { | |
6953 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6954 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6955 | ||
6956 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6957 | && vid == ofport->vlandev_vid) { | |
6958 | return 0; | |
6959 | } | |
6960 | ||
3c4a309c | 6961 | ofproto->need_revalidate = REV_RECONFIGURE; |
52a90c29 BP |
6962 | |
6963 | if (ofport->realdev_ofp_port) { | |
6964 | vsp_remove(ofport); | |
6965 | } | |
6966 | if (realdev_ofp_port && ofport->bundle) { | |
6967 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6968 | * themselves be part of a bundle. */ | |
6969 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6970 | } | |
6971 | ||
6972 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6973 | ofport->vlandev_vid = vid; | |
6974 | ||
6975 | if (realdev_ofp_port) { | |
6976 | vsp_add(ofport, realdev_ofp_port, vid); | |
6977 | } | |
6978 | ||
6979 | return 0; | |
6980 | } | |
6981 | ||
6982 | static uint32_t | |
6983 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
6984 | { | |
6985 | return hash_2words(realdev_ofp_port, vid); | |
6986 | } | |
6987 | ||
40e05935 BP |
6988 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
6989 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
6990 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
6991 | * it would return the port number of eth0.9. | |
6992 | * | |
6993 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
6994 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
6995 | static uint32_t |
6996 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6997 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
6998 | { | |
6999 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
7000 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
7001 | int vid = vlan_tci_to_vid(vlan_tci); | |
7002 | const struct vlan_splinter *vsp; | |
7003 | ||
7004 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
7005 | hash_realdev_vid(realdev_ofp_port, vid), | |
7006 | &ofproto->realdev_vid_map) { | |
7007 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
7008 | && vsp->vid == vid) { | |
7009 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
7010 | } | |
7011 | } | |
7012 | } | |
7013 | return realdev_odp_port; | |
7014 | } | |
7015 | ||
7016 | static struct vlan_splinter * | |
7017 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
7018 | { | |
7019 | struct vlan_splinter *vsp; | |
7020 | ||
7021 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
7022 | &ofproto->vlandev_map) { | |
7023 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
7024 | return vsp; | |
7025 | } | |
7026 | } | |
7027 | ||
7028 | return NULL; | |
7029 | } | |
7030 | ||
40e05935 BP |
7031 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
7032 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
7033 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
7034 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
7035 | * eth0 and store 9 in '*vid'. | |
7036 | * | |
7037 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
7038 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
7039 | * always does.*/ | |
52a90c29 BP |
7040 | static uint16_t |
7041 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 7042 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
7043 | { |
7044 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
7045 | const struct vlan_splinter *vsp; | |
7046 | ||
7047 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
7048 | if (vsp) { | |
7049 | if (vid) { | |
7050 | *vid = vsp->vid; | |
7051 | } | |
7052 | return vsp->realdev_ofp_port; | |
7053 | } | |
7054 | } | |
7055 | return 0; | |
7056 | } | |
7057 | ||
b98d8985 BP |
7058 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks |
7059 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
7060 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
7061 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
7062 | * always the case unless VLAN splinters are enabled), returns false without | |
7063 | * making any changes. */ | |
7064 | static bool | |
7065 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
7066 | { | |
7067 | uint16_t realdev; | |
7068 | int vid; | |
7069 | ||
7070 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
7071 | if (!realdev) { | |
7072 | return false; | |
7073 | } | |
7074 | ||
7075 | /* Cause the flow to be processed as if it came in on the real device with | |
7076 | * the VLAN device's VLAN ID. */ | |
7077 | flow->in_port = realdev; | |
7078 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
7079 | return true; | |
7080 | } | |
7081 | ||
52a90c29 BP |
7082 | static void |
7083 | vsp_remove(struct ofport_dpif *port) | |
7084 | { | |
7085 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7086 | struct vlan_splinter *vsp; | |
7087 | ||
7088 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
7089 | if (vsp) { | |
7090 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
7091 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
7092 | free(vsp); | |
7093 | ||
7094 | port->realdev_ofp_port = 0; | |
7095 | } else { | |
7096 | VLOG_ERR("missing vlan device record"); | |
7097 | } | |
7098 | } | |
7099 | ||
7100 | static void | |
7101 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
7102 | { | |
7103 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7104 | ||
7105 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
7106 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
7107 | == realdev_ofp_port)) { | |
7108 | struct vlan_splinter *vsp; | |
7109 | ||
7110 | vsp = xmalloc(sizeof *vsp); | |
7111 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
7112 | hash_int(port->up.ofp_port, 0)); | |
7113 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
7114 | hash_realdev_vid(realdev_ofp_port, vid)); | |
7115 | vsp->realdev_ofp_port = realdev_ofp_port; | |
7116 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
7117 | vsp->vid = vid; | |
7118 | ||
7119 | port->realdev_ofp_port = realdev_ofp_port; | |
7120 | } else { | |
7121 | VLOG_ERR("duplicate vlan device record"); | |
7122 | } | |
7123 | } | |
7124 | \f | |
abe529af BP |
7125 | const struct ofproto_class ofproto_dpif_class = { |
7126 | enumerate_types, | |
7127 | enumerate_names, | |
7128 | del, | |
7129 | alloc, | |
7130 | construct, | |
7131 | destruct, | |
7132 | dealloc, | |
7133 | run, | |
5fcc0d00 | 7134 | run_fast, |
abe529af | 7135 | wait, |
0d085684 | 7136 | get_memory_usage, |
abe529af | 7137 | flush, |
6c1491fb BP |
7138 | get_features, |
7139 | get_tables, | |
abe529af BP |
7140 | port_alloc, |
7141 | port_construct, | |
7142 | port_destruct, | |
7143 | port_dealloc, | |
7144 | port_modified, | |
7145 | port_reconfigured, | |
7146 | port_query_by_name, | |
7147 | port_add, | |
7148 | port_del, | |
6527c598 | 7149 | port_get_stats, |
abe529af BP |
7150 | port_dump_start, |
7151 | port_dump_next, | |
7152 | port_dump_done, | |
7153 | port_poll, | |
7154 | port_poll_wait, | |
7155 | port_is_lacp_current, | |
0ab6decf | 7156 | NULL, /* rule_choose_table */ |
abe529af BP |
7157 | rule_alloc, |
7158 | rule_construct, | |
7159 | rule_destruct, | |
7160 | rule_dealloc, | |
abe529af BP |
7161 | rule_get_stats, |
7162 | rule_execute, | |
7163 | rule_modify_actions, | |
7257b535 | 7164 | set_frag_handling, |
abe529af BP |
7165 | packet_out, |
7166 | set_netflow, | |
7167 | get_netflow_ids, | |
7168 | set_sflow, | |
7169 | set_cfm, | |
a5610457 | 7170 | get_cfm_fault, |
1de11730 | 7171 | get_cfm_remote_mpids, |
3967a833 | 7172 | get_cfm_health, |
21f7563c JP |
7173 | set_stp, |
7174 | get_stp_status, | |
7175 | set_stp_port, | |
7176 | get_stp_port_status, | |
8b36f51e | 7177 | set_queues, |
abe529af BP |
7178 | bundle_set, |
7179 | bundle_remove, | |
7180 | mirror_set, | |
9d24de3b | 7181 | mirror_get_stats, |
abe529af BP |
7182 | set_flood_vlans, |
7183 | is_mirror_output_bundle, | |
8402c74b | 7184 | forward_bpdu_changed, |
e764773c | 7185 | set_mac_idle_time, |
52a90c29 | 7186 | set_realdev, |
abe529af | 7187 | }; |