]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
4dd1e3ca | 2 | * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks. |
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" | |
45 | #include "ofp-print.h" | |
9d6ac44e | 46 | #include "ofproto-dpif-governor.h" |
bae473fe | 47 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
48 | #include "poll-loop.h" |
49 | #include "timer.h" | |
6c1491fb | 50 | #include "unaligned.h" |
abe529af BP |
51 | #include "unixctl.h" |
52 | #include "vlan-bitmap.h" | |
53 | #include "vlog.h" | |
54 | ||
55 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
56 | ||
57 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
58 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
59 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
60 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
61 | COVERAGE_DEFINE(facet_changed_rule); | |
62 | COVERAGE_DEFINE(facet_invalidated); | |
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 }; | |
73 | BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255); | |
74 | ||
abe529af BP |
75 | struct ofport_dpif; |
76 | struct ofproto_dpif; | |
77 | ||
78 | struct rule_dpif { | |
79 | struct rule up; | |
80 | ||
abe529af BP |
81 | /* These statistics: |
82 | * | |
83 | * - Do include packets and bytes from facets that have been deleted or | |
84 | * whose own statistics have been folded into the rule. | |
85 | * | |
86 | * - Do include packets and bytes sent "by hand" that were accounted to | |
87 | * the rule without any facet being involved (this is a rare corner | |
88 | * case in rule_execute()). | |
89 | * | |
90 | * - Do not include packet or bytes that can be obtained from any facet's | |
91 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 92 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
93 | */ |
94 | uint64_t packet_count; /* Number of packets received. */ | |
95 | uint64_t byte_count; /* Number of bytes received. */ | |
96 | ||
54a9cbc9 BP |
97 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
98 | ||
abe529af BP |
99 | struct list facets; /* List of "struct facet"s. */ |
100 | }; | |
101 | ||
102 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
103 | { | |
104 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
105 | } | |
106 | ||
29901626 BP |
107 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
108 | const struct flow *, uint8_t table); | |
abe529af | 109 | |
112bc5f4 BP |
110 | static void rule_credit_stats(struct rule_dpif *, |
111 | const struct dpif_flow_stats *); | |
18b2a258 | 112 | static void flow_push_stats(struct rule_dpif *, const struct flow *, |
112bc5f4 | 113 | const struct dpif_flow_stats *); |
822d9414 | 114 | static tag_type rule_calculate_tag(const struct flow *, |
b0f7b9b5 BP |
115 | const struct flow_wildcards *, |
116 | uint32_t basis); | |
117 | static void rule_invalidate(const struct rule_dpif *); | |
118 | ||
abe529af BP |
119 | #define MAX_MIRRORS 32 |
120 | typedef uint32_t mirror_mask_t; | |
121 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
122 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
123 | struct ofmirror { | |
124 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
125 | size_t idx; /* In ofproto's "mirrors" array. */ | |
126 | void *aux; /* Key supplied by ofproto's client. */ | |
127 | char *name; /* Identifier for log messages. */ | |
128 | ||
129 | /* Selection criteria. */ | |
130 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
131 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
132 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
133 | ||
9ba15e2a | 134 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
135 | struct ofbundle *out; /* Output port or NULL. */ |
136 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 137 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
138 | |
139 | /* Counters. */ | |
140 | int64_t packet_count; /* Number of packets sent. */ | |
141 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
142 | }; |
143 | ||
144 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
145 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
146 | mirror_mask_t mirrors, | |
147 | uint64_t packets, uint64_t bytes); | |
abe529af | 148 | |
abe529af | 149 | struct ofbundle { |
abe529af | 150 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 151 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
152 | void *aux; /* Key supplied by ofproto's client. */ |
153 | char *name; /* Identifier for log messages. */ | |
154 | ||
155 | /* Configuration. */ | |
156 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 157 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
158 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
159 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
160 | * NULL if all VLANs are trunked. */ | |
161 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
162 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 163 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
164 | |
165 | /* Status. */ | |
9e1fd49b | 166 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
167 | |
168 | /* Port mirroring info. */ | |
169 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
170 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
171 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
172 | }; | |
173 | ||
174 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 175 | static void bundle_update(struct ofbundle *); |
abe529af BP |
176 | static void bundle_destroy(struct ofbundle *); |
177 | static void bundle_del_port(struct ofport_dpif *); | |
178 | static void bundle_run(struct ofbundle *); | |
179 | static void bundle_wait(struct ofbundle *); | |
3581c12c JP |
180 | static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *, |
181 | uint16_t in_port, bool warn); | |
abe529af | 182 | |
33158a18 JP |
183 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
184 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
185 | * when an input bundle is needed for validation (e.g., mirroring or | |
186 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
187 | * any 'port' structs, so care must be taken when dealing with it. */ | |
188 | static struct ofbundle ofpp_none_bundle = { | |
189 | .name = "OFPP_NONE", | |
190 | .vlan_mode = PORT_VLAN_TRUNK | |
191 | }; | |
192 | ||
21f7563c JP |
193 | static void stp_run(struct ofproto_dpif *ofproto); |
194 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
195 | static int set_stp_port(struct ofport *, |
196 | const struct ofproto_port_stp_settings *); | |
21f7563c | 197 | |
5da5ec37 BP |
198 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
199 | ||
abe529af BP |
200 | struct action_xlate_ctx { |
201 | /* action_xlate_ctx_init() initializes these members. */ | |
202 | ||
203 | /* The ofproto. */ | |
204 | struct ofproto_dpif *ofproto; | |
205 | ||
206 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
207 | * this flow when actions change header fields. */ | |
208 | struct flow flow; | |
209 | ||
210 | /* The packet corresponding to 'flow', or a null pointer if we are | |
211 | * revalidating without a packet to refer to. */ | |
212 | const struct ofpbuf *packet; | |
213 | ||
3de9590b BP |
214 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
215 | * actions update the flow table? | |
216 | * | |
217 | * We want to update these tables if we are actually processing a packet, | |
218 | * or if we are accounting for packets that the datapath has processed, but | |
219 | * not if we are just revalidating. */ | |
220 | bool may_learn; | |
75a75043 | 221 | |
18b2a258 BP |
222 | /* The rule that we are currently translating, or NULL. */ |
223 | struct rule_dpif *rule; | |
54834960 | 224 | |
0e553d9c BP |
225 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
226 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
227 | * timeouts.) */ | |
228 | uint8_t tcp_flags; | |
229 | ||
112bc5f4 BP |
230 | /* If nonnull, flow translation calls this function just before executing a |
231 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
232 | * when the recursion depth is exceeded. | |
233 | * | |
234 | * 'rule' is the rule being submitted into. It will be null if the | |
235 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
236 | * | |
237 | * This is normally null so the client has to set it manually after | |
238 | * calling action_xlate_ctx_init(). */ | |
239 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
240 | ||
241 | /* If nonnull, flow translation credits the specified statistics to each | |
242 | * rule reached through a resubmit or OFPP_TABLE action. | |
abe529af BP |
243 | * |
244 | * This is normally null so the client has to set it manually after | |
245 | * calling action_xlate_ctx_init(). */ | |
112bc5f4 | 246 | const struct dpif_flow_stats *resubmit_stats; |
abe529af | 247 | |
abe529af BP |
248 | /* xlate_actions() initializes and uses these members. The client might want |
249 | * to look at them after it returns. */ | |
250 | ||
251 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 252 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
253 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
254 | * be reassessed for every packet. */ | |
75a75043 BP |
255 | bool has_learn; /* Actions include NXAST_LEARN? */ |
256 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 257 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 258 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 259 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
260 | |
261 | /* xlate_actions() initializes and uses these members, but the client has no | |
262 | * reason to look at them. */ | |
263 | ||
264 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 265 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 266 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 267 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 268 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
269 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
270 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
271 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 272 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
273 | }; |
274 | ||
275 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
276 | struct ofproto_dpif *, const struct flow *, | |
18b2a258 | 277 | ovs_be16 initial_tci, struct rule_dpif *, |
0e553d9c | 278 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 BP |
279 | static void xlate_actions(struct action_xlate_ctx *, |
280 | const union ofp_action *in, size_t n_in, | |
281 | struct ofpbuf *odp_actions); | |
282 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
283 | const union ofp_action *in, | |
284 | size_t n_in); | |
abe529af | 285 | |
5f5fbd17 BP |
286 | /* A dpif flow and actions associated with a facet. |
287 | * | |
288 | * See also the large comment on struct facet. */ | |
289 | struct subfacet { | |
290 | /* Owners. */ | |
291 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
292 | struct list list_node; /* In struct facet's 'facets' list. */ | |
293 | struct facet *facet; /* Owning facet. */ | |
294 | ||
295 | /* Key. | |
296 | * | |
297 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
298 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
299 | * regenerate the ODP flow key from ->facet->flow. */ | |
300 | enum odp_key_fitness key_fitness; | |
301 | struct nlattr *key; | |
302 | int key_len; | |
303 | ||
304 | long long int used; /* Time last used; time created if not used. */ | |
305 | ||
306 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
307 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
308 | ||
309 | /* Datapath actions. | |
310 | * | |
311 | * These should be essentially identical for every subfacet in a facet, but | |
312 | * may differ in trivial ways due to VLAN splinters. */ | |
313 | size_t actions_len; /* Number of bytes in actions[]. */ | |
314 | struct nlattr *actions; /* Datapath actions. */ | |
315 | ||
316 | bool installed; /* Installed in datapath? */ | |
317 | ||
318 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
319 | * splinters can cause it to differ. This value should be removed when | |
320 | * the VLAN splinters feature is no longer needed. */ | |
321 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
322 | }; | |
323 | ||
324 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, | |
325 | const struct nlattr *key, | |
326 | size_t key_len, ovs_be16 initial_tci); | |
327 | static struct subfacet *subfacet_find(struct ofproto_dpif *, | |
328 | const struct nlattr *key, size_t key_len); | |
329 | static void subfacet_destroy(struct subfacet *); | |
330 | static void subfacet_destroy__(struct subfacet *); | |
331 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, | |
332 | struct ofpbuf *key); | |
333 | static void subfacet_reset_dp_stats(struct subfacet *, | |
334 | struct dpif_flow_stats *); | |
335 | static void subfacet_update_time(struct subfacet *, long long int used); | |
336 | static void subfacet_update_stats(struct subfacet *, | |
337 | const struct dpif_flow_stats *); | |
338 | static void subfacet_make_actions(struct subfacet *, | |
5fe20d5d BP |
339 | const struct ofpbuf *packet, |
340 | struct ofpbuf *odp_actions); | |
5f5fbd17 BP |
341 | static int subfacet_install(struct subfacet *, |
342 | const struct nlattr *actions, size_t actions_len, | |
343 | struct dpif_flow_stats *); | |
344 | static void subfacet_uninstall(struct subfacet *); | |
345 | ||
b0f7b9b5 BP |
346 | /* An exact-match instantiation of an OpenFlow flow. |
347 | * | |
348 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
349 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
350 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
351 | * the facet. When the kernel module (or other dpif implementation) and Open | |
352 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
353 | * one subfacet per facet. If the dpif implementation supports more-specific | |
354 | * flow matching than userspace, however, a facet can have more than one | |
355 | * subfacet, each of which corresponds to some distinction in flow that | |
356 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
357 | * |
358 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
359 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 360 | struct facet { |
b0f7b9b5 BP |
361 | /* Owners. */ |
362 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
363 | struct list list_node; /* In owning rule's 'facets' list. */ | |
364 | struct rule_dpif *rule; /* Owning rule. */ | |
365 | ||
366 | /* Owned data. */ | |
367 | struct list subfacets; | |
abe529af BP |
368 | long long int used; /* Time last used; time created if not used. */ |
369 | ||
b0f7b9b5 BP |
370 | /* Key. */ |
371 | struct flow flow; | |
372 | ||
abe529af BP |
373 | /* These statistics: |
374 | * | |
375 | * - Do include packets and bytes sent "by hand", e.g. with | |
376 | * dpif_execute(). | |
377 | * | |
378 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 379 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 380 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
381 | * |
382 | * - Do not include packets or bytes that can be obtained from the | |
383 | * datapath for any existing subfacet. | |
abe529af BP |
384 | */ |
385 | uint64_t packet_count; /* Number of packets received. */ | |
386 | uint64_t byte_count; /* Number of bytes received. */ | |
387 | ||
b0f7b9b5 | 388 | /* Resubmit statistics. */ |
9d24de3b JP |
389 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
390 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
391 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 392 | |
b0f7b9b5 | 393 | /* Accounting. */ |
907a4c5e | 394 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 395 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 396 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 397 | |
b95fc6ba BP |
398 | /* Properties of datapath actions. |
399 | * | |
400 | * Every subfacet has its own actions because actions can differ slightly | |
401 | * between splintered and non-splintered subfacets due to the VLAN tag | |
402 | * being initially different (present vs. absent). All of them have these | |
403 | * properties in common so we just store one copy of them here. */ | |
b0f7b9b5 | 404 | bool may_install; /* Reassess actions for every packet? */ |
75a75043 BP |
405 | bool has_learn; /* Actions include NXAST_LEARN? */ |
406 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 407 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 408 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 409 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
410 | |
411 | /* Storage for a single subfacet, to reduce malloc() time and space | |
412 | * overhead. (A facet always has at least one subfacet and in the common | |
413 | * case has exactly one subfacet.) */ | |
414 | struct subfacet one_subfacet; | |
abe529af BP |
415 | }; |
416 | ||
2b459b83 BP |
417 | static struct facet *facet_create(struct rule_dpif *, |
418 | const struct flow *, uint32_t hash); | |
15baa734 | 419 | static void facet_remove(struct facet *); |
abe529af BP |
420 | static void facet_free(struct facet *); |
421 | ||
2b459b83 BP |
422 | static struct facet *facet_find(struct ofproto_dpif *, |
423 | const struct flow *, uint32_t hash); | |
abe529af | 424 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 425 | const struct flow *, uint32_t hash); |
15baa734 | 426 | static bool facet_revalidate(struct facet *); |
6814e51f | 427 | static bool facet_check_consistency(struct facet *); |
abe529af | 428 | |
15baa734 | 429 | static void facet_flush_stats(struct facet *); |
abe529af | 430 | |
15baa734 | 431 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 432 | static void facet_reset_counters(struct facet *); |
abe529af | 433 | static void facet_push_stats(struct facet *); |
3de9590b BP |
434 | static void facet_learn(struct facet *); |
435 | static void facet_account(struct facet *); | |
abe529af BP |
436 | |
437 | static bool facet_is_controller_flow(struct facet *); | |
438 | ||
abe529af BP |
439 | struct ofport_dpif { |
440 | struct ofport up; | |
441 | ||
442 | uint32_t odp_port; | |
443 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
444 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
445 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
446 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 447 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 448 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 449 | long long int carrier_seq; /* Carrier status changes. */ |
21f7563c | 450 | |
52a90c29 | 451 | /* Spanning tree. */ |
21f7563c JP |
452 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
453 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
454 | long long int stp_state_entered; | |
8b36f51e EJ |
455 | |
456 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
457 | |
458 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
459 | * | |
460 | * This is deprecated. It is only for compatibility with broken device | |
461 | * drivers in old versions of Linux that do not properly support VLANs when | |
462 | * VLAN devices are not used. When broken device drivers are no longer in | |
463 | * widespread use, we will delete these interfaces. */ | |
464 | uint16_t realdev_ofp_port; | |
465 | int vlandev_vid; | |
8b36f51e EJ |
466 | }; |
467 | ||
468 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
469 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
470 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
471 | struct priority_to_dscp { | |
472 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
473 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
474 | ||
475 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
476 | }; |
477 | ||
52a90c29 BP |
478 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
479 | * | |
480 | * This is deprecated. It is only for compatibility with broken device drivers | |
481 | * in old versions of Linux that do not properly support VLANs when VLAN | |
482 | * devices are not used. When broken device drivers are no longer in | |
483 | * widespread use, we will delete these interfaces. */ | |
484 | struct vlan_splinter { | |
485 | struct hmap_node realdev_vid_node; | |
486 | struct hmap_node vlandev_node; | |
487 | uint16_t realdev_ofp_port; | |
488 | uint16_t vlandev_ofp_port; | |
489 | int vid; | |
490 | }; | |
491 | ||
492 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
493 | uint32_t realdev, ovs_be16 vlan_tci); | |
494 | static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *, | |
495 | uint16_t vlandev, int *vid); | |
496 | static void vsp_remove(struct ofport_dpif *); | |
497 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
498 | ||
abe529af BP |
499 | static struct ofport_dpif * |
500 | ofport_dpif_cast(const struct ofport *ofport) | |
501 | { | |
502 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
503 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
504 | } | |
505 | ||
506 | static void port_run(struct ofport_dpif *); | |
507 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 508 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 509 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 510 | |
7ee20df1 BP |
511 | struct dpif_completion { |
512 | struct list list_node; | |
513 | struct ofoperation *op; | |
514 | }; | |
515 | ||
54a9cbc9 BP |
516 | /* Extra information about a classifier table. |
517 | * Currently used just for optimized flow revalidation. */ | |
518 | struct table_dpif { | |
519 | /* If either of these is nonnull, then this table has a form that allows | |
520 | * flows to be tagged to avoid revalidating most flows for the most common | |
521 | * kinds of flow table changes. */ | |
522 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
523 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
524 | uint32_t basis; /* Keeps each table's tags separate. */ | |
525 | }; | |
526 | ||
abe529af | 527 | struct ofproto_dpif { |
b44a10b7 | 528 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
529 | struct ofproto up; |
530 | struct dpif *dpif; | |
531 | int max_ports; | |
532 | ||
6c1491fb BP |
533 | /* Statistics. */ |
534 | uint64_t n_matches; | |
535 | ||
abe529af BP |
536 | /* Bridging. */ |
537 | struct netflow *netflow; | |
bae473fe | 538 | struct dpif_sflow *sflow; |
abe529af BP |
539 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
540 | struct mac_learning *ml; | |
541 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
542 | bool has_bonded_bundles; | |
543 | ||
544 | /* Expiration. */ | |
545 | struct timer next_expiration; | |
546 | ||
547 | /* Facets. */ | |
548 | struct hmap facets; | |
b0f7b9b5 | 549 | struct hmap subfacets; |
9d6ac44e | 550 | struct governor *governor; |
54a9cbc9 BP |
551 | |
552 | /* Revalidation. */ | |
553 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
554 | bool need_revalidate; |
555 | struct tag_set revalidate_set; | |
7ee20df1 BP |
556 | |
557 | /* Support for debugging async flow mods. */ | |
558 | struct list completions; | |
daff3353 EJ |
559 | |
560 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
561 | struct netdev_stats stats; /* To account packets generated and consumed in |
562 | * userspace. */ | |
21f7563c JP |
563 | |
564 | /* Spanning tree. */ | |
565 | struct stp *stp; | |
566 | long long int stp_last_tick; | |
52a90c29 BP |
567 | |
568 | /* VLAN splinters. */ | |
569 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
570 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
571 | }; |
572 | ||
7ee20df1 BP |
573 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
574 | * for debugging the asynchronous flow_mod implementation.) */ | |
575 | static bool clogged; | |
576 | ||
b44a10b7 BP |
577 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
578 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
579 | ||
abe529af BP |
580 | static void ofproto_dpif_unixctl_init(void); |
581 | ||
582 | static struct ofproto_dpif * | |
583 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
584 | { | |
585 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
586 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
587 | } | |
588 | ||
589 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
590 | uint16_t ofp_port); | |
591 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
592 | uint32_t odp_port); | |
6a6455e5 EJ |
593 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
594 | const struct ofpbuf *, ovs_be16 initial_tci, | |
595 | struct ds *); | |
abe529af BP |
596 | |
597 | /* Packet processing. */ | |
598 | static void update_learning_table(struct ofproto_dpif *, | |
599 | const struct flow *, int vlan, | |
600 | struct ofbundle *); | |
501f8d1f BP |
601 | /* Upcalls. */ |
602 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 603 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
604 | |
605 | /* Flow expiration. */ | |
606 | static int expire(struct ofproto_dpif *); | |
607 | ||
6fca1ffb BP |
608 | /* NetFlow. */ |
609 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
610 | ||
abe529af | 611 | /* Utilities. */ |
52a90c29 | 612 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6ff686f2 PS |
613 | static size_t |
614 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
615 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
616 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
617 | const struct flow *flow); | |
abe529af BP |
618 | /* Global variables. */ |
619 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
620 | \f | |
621 | /* Factory functions. */ | |
622 | ||
623 | static void | |
624 | enumerate_types(struct sset *types) | |
625 | { | |
626 | dp_enumerate_types(types); | |
627 | } | |
628 | ||
629 | static int | |
630 | enumerate_names(const char *type, struct sset *names) | |
631 | { | |
632 | return dp_enumerate_names(type, names); | |
633 | } | |
634 | ||
635 | static int | |
636 | del(const char *type, const char *name) | |
637 | { | |
638 | struct dpif *dpif; | |
639 | int error; | |
640 | ||
641 | error = dpif_open(name, type, &dpif); | |
642 | if (!error) { | |
643 | error = dpif_delete(dpif); | |
644 | dpif_close(dpif); | |
645 | } | |
646 | return error; | |
647 | } | |
648 | \f | |
649 | /* Basic life-cycle. */ | |
650 | ||
651 | static struct ofproto * | |
652 | alloc(void) | |
653 | { | |
654 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
655 | return &ofproto->up; | |
656 | } | |
657 | ||
658 | static void | |
659 | dealloc(struct ofproto *ofproto_) | |
660 | { | |
661 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
662 | free(ofproto); | |
663 | } | |
664 | ||
665 | static int | |
0f5f95a9 | 666 | construct(struct ofproto *ofproto_) |
abe529af BP |
667 | { |
668 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
669 | const char *name = ofproto->up.name; | |
670 | int error; | |
671 | int i; | |
672 | ||
673 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
674 | if (error) { | |
675 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
676 | return error; | |
677 | } | |
678 | ||
679 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 680 | ofproto->n_matches = 0; |
abe529af | 681 | |
be8194bb JG |
682 | dpif_flow_flush(ofproto->dpif); |
683 | dpif_recv_purge(ofproto->dpif); | |
684 | ||
a12b3ead | 685 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
686 | if (error) { |
687 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
688 | dpif_close(ofproto->dpif); | |
689 | return error; | |
690 | } | |
abe529af BP |
691 | |
692 | ofproto->netflow = NULL; | |
693 | ofproto->sflow = NULL; | |
21f7563c | 694 | ofproto->stp = NULL; |
abe529af | 695 | hmap_init(&ofproto->bundles); |
e764773c | 696 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
697 | for (i = 0; i < MAX_MIRRORS; i++) { |
698 | ofproto->mirrors[i] = NULL; | |
699 | } | |
700 | ofproto->has_bonded_bundles = false; | |
701 | ||
702 | timer_set_duration(&ofproto->next_expiration, 1000); | |
703 | ||
704 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 705 | hmap_init(&ofproto->subfacets); |
9d6ac44e | 706 | ofproto->governor = NULL; |
54a9cbc9 BP |
707 | |
708 | for (i = 0; i < N_TABLES; i++) { | |
709 | struct table_dpif *table = &ofproto->tables[i]; | |
710 | ||
711 | table->catchall_table = NULL; | |
712 | table->other_table = NULL; | |
713 | table->basis = random_uint32(); | |
714 | } | |
abe529af BP |
715 | ofproto->need_revalidate = false; |
716 | tag_set_init(&ofproto->revalidate_set); | |
717 | ||
7ee20df1 BP |
718 | list_init(&ofproto->completions); |
719 | ||
abe529af BP |
720 | ofproto_dpif_unixctl_init(); |
721 | ||
daff3353 EJ |
722 | ofproto->has_bundle_action = false; |
723 | ||
52a90c29 BP |
724 | hmap_init(&ofproto->vlandev_map); |
725 | hmap_init(&ofproto->realdev_vid_map); | |
726 | ||
b44a10b7 BP |
727 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
728 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 729 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
730 | |
731 | ofproto_init_tables(ofproto_, N_TABLES); | |
732 | ||
abe529af BP |
733 | return 0; |
734 | } | |
735 | ||
7ee20df1 BP |
736 | static void |
737 | complete_operations(struct ofproto_dpif *ofproto) | |
738 | { | |
739 | struct dpif_completion *c, *next; | |
740 | ||
741 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
742 | ofoperation_complete(c->op, 0); | |
743 | list_remove(&c->list_node); | |
744 | free(c); | |
745 | } | |
746 | } | |
747 | ||
abe529af BP |
748 | static void |
749 | destruct(struct ofproto *ofproto_) | |
750 | { | |
751 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 752 | struct rule_dpif *rule, *next_rule; |
d0918789 | 753 | struct oftable *table; |
abe529af BP |
754 | int i; |
755 | ||
b44a10b7 | 756 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
757 | complete_operations(ofproto); |
758 | ||
0697b5c3 BP |
759 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
760 | struct cls_cursor cursor; | |
761 | ||
d0918789 | 762 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
763 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
764 | ofproto_rule_destroy(&rule->up); | |
765 | } | |
7ee20df1 BP |
766 | } |
767 | ||
abe529af BP |
768 | for (i = 0; i < MAX_MIRRORS; i++) { |
769 | mirror_destroy(ofproto->mirrors[i]); | |
770 | } | |
771 | ||
772 | netflow_destroy(ofproto->netflow); | |
bae473fe | 773 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
774 | hmap_destroy(&ofproto->bundles); |
775 | mac_learning_destroy(ofproto->ml); | |
776 | ||
777 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 778 | hmap_destroy(&ofproto->subfacets); |
9d6ac44e | 779 | governor_destroy(ofproto->governor); |
abe529af | 780 | |
52a90c29 BP |
781 | hmap_destroy(&ofproto->vlandev_map); |
782 | hmap_destroy(&ofproto->realdev_vid_map); | |
783 | ||
abe529af BP |
784 | dpif_close(ofproto->dpif); |
785 | } | |
786 | ||
787 | static int | |
5fcc0d00 | 788 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
789 | { |
790 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
9b16c439 | 791 | unsigned int work; |
abe529af | 792 | |
9b16c439 BP |
793 | /* Handle one or more batches of upcalls, until there's nothing left to do |
794 | * or until we do a fixed total amount of work. | |
795 | * | |
796 | * We do work in batches because it can be much cheaper to set up a number | |
797 | * of flows and fire off their patches all at once. We do multiple batches | |
798 | * because in some cases handling a packet can cause another packet to be | |
799 | * queued almost immediately as part of the return flow. Both | |
800 | * optimizations can make major improvements on some benchmarks and | |
801 | * presumably for real traffic as well. */ | |
802 | work = 0; | |
803 | while (work < FLOW_MISS_MAX_BATCH) { | |
804 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 805 | if (retval <= 0) { |
9b16c439 | 806 | return -retval; |
501f8d1f | 807 | } |
5fcc0d00 BP |
808 | work += retval; |
809 | } | |
810 | return 0; | |
811 | } | |
812 | ||
813 | static int | |
814 | run(struct ofproto *ofproto_) | |
815 | { | |
816 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
817 | struct ofport_dpif *ofport; | |
818 | struct ofbundle *bundle; | |
819 | int error; | |
820 | ||
821 | if (!clogged) { | |
822 | complete_operations(ofproto); | |
823 | } | |
824 | dpif_run(ofproto->dpif); | |
825 | ||
826 | error = run_fast(ofproto_); | |
827 | if (error) { | |
828 | return error; | |
abe529af BP |
829 | } |
830 | ||
831 | if (timer_expired(&ofproto->next_expiration)) { | |
832 | int delay = expire(ofproto); | |
833 | timer_set_duration(&ofproto->next_expiration, delay); | |
834 | } | |
835 | ||
836 | if (ofproto->netflow) { | |
6fca1ffb BP |
837 | if (netflow_run(ofproto->netflow)) { |
838 | send_netflow_active_timeouts(ofproto); | |
839 | } | |
abe529af BP |
840 | } |
841 | if (ofproto->sflow) { | |
bae473fe | 842 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
843 | } |
844 | ||
845 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
846 | port_run(ofport); | |
847 | } | |
848 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
849 | bundle_run(bundle); | |
850 | } | |
851 | ||
21f7563c | 852 | stp_run(ofproto); |
1c313b88 BP |
853 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
854 | ||
abe529af BP |
855 | /* Now revalidate if there's anything to do. */ |
856 | if (ofproto->need_revalidate | |
857 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
858 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
859 | bool revalidate_all = ofproto->need_revalidate; | |
860 | struct facet *facet, *next; | |
861 | ||
862 | /* Clear the revalidation flags. */ | |
863 | tag_set_init(&ofproto->revalidate_set); | |
864 | ofproto->need_revalidate = false; | |
865 | ||
866 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
867 | if (revalidate_all | |
868 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 869 | facet_revalidate(facet); |
abe529af BP |
870 | } |
871 | } | |
872 | } | |
873 | ||
6814e51f BP |
874 | /* Check the consistency of a random facet, to aid debugging. */ |
875 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
876 | struct facet *facet; | |
877 | ||
878 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
879 | struct facet, hmap_node); | |
880 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
881 | if (!facet_check_consistency(facet)) { | |
882 | ofproto->need_revalidate = true; | |
883 | } | |
884 | } | |
885 | } | |
886 | ||
9d6ac44e BP |
887 | if (ofproto->governor) { |
888 | size_t n_subfacets; | |
889 | ||
890 | governor_run(ofproto->governor); | |
891 | ||
892 | /* If the governor has shrunk to its minimum size and the number of | |
893 | * subfacets has dwindled, then drop the governor entirely. | |
894 | * | |
895 | * For hysteresis, the number of subfacets to drop the governor is | |
896 | * smaller than the number needed to trigger its creation. */ | |
897 | n_subfacets = hmap_count(&ofproto->subfacets); | |
898 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
899 | && governor_is_idle(ofproto->governor)) { | |
900 | governor_destroy(ofproto->governor); | |
901 | ofproto->governor = NULL; | |
902 | } | |
903 | } | |
904 | ||
abe529af BP |
905 | return 0; |
906 | } | |
907 | ||
908 | static void | |
909 | wait(struct ofproto *ofproto_) | |
910 | { | |
911 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
912 | struct ofport_dpif *ofport; | |
913 | struct ofbundle *bundle; | |
914 | ||
7ee20df1 BP |
915 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
916 | poll_immediate_wake(); | |
917 | } | |
918 | ||
abe529af BP |
919 | dpif_wait(ofproto->dpif); |
920 | dpif_recv_wait(ofproto->dpif); | |
921 | if (ofproto->sflow) { | |
bae473fe | 922 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
923 | } |
924 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
925 | poll_immediate_wake(); | |
926 | } | |
927 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
928 | port_wait(ofport); | |
929 | } | |
930 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
931 | bundle_wait(bundle); | |
932 | } | |
6fca1ffb BP |
933 | if (ofproto->netflow) { |
934 | netflow_wait(ofproto->netflow); | |
935 | } | |
1c313b88 | 936 | mac_learning_wait(ofproto->ml); |
21f7563c | 937 | stp_wait(ofproto); |
abe529af BP |
938 | if (ofproto->need_revalidate) { |
939 | /* Shouldn't happen, but if it does just go around again. */ | |
940 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
941 | poll_immediate_wake(); | |
942 | } else { | |
943 | timer_wait(&ofproto->next_expiration); | |
944 | } | |
9d6ac44e BP |
945 | if (ofproto->governor) { |
946 | governor_wait(ofproto->governor); | |
947 | } | |
abe529af BP |
948 | } |
949 | ||
950 | static void | |
951 | flush(struct ofproto *ofproto_) | |
952 | { | |
953 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
954 | struct facet *facet, *next_facet; | |
955 | ||
956 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
957 | /* Mark the facet as not installed so that facet_remove() doesn't | |
958 | * bother trying to uninstall it. There is no point in uninstalling it | |
959 | * individually since we are about to blow away all the facets with | |
960 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
961 | struct subfacet *subfacet; |
962 | ||
963 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
964 | subfacet->installed = false; | |
965 | subfacet->dp_packet_count = 0; | |
966 | subfacet->dp_byte_count = 0; | |
967 | } | |
15baa734 | 968 | facet_remove(facet); |
abe529af BP |
969 | } |
970 | dpif_flow_flush(ofproto->dpif); | |
971 | } | |
972 | ||
6c1491fb BP |
973 | static void |
974 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 975 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
976 | { |
977 | *arp_match_ip = true; | |
9e1fd49b BP |
978 | *actions = (OFPUTIL_A_OUTPUT | |
979 | OFPUTIL_A_SET_VLAN_VID | | |
980 | OFPUTIL_A_SET_VLAN_PCP | | |
981 | OFPUTIL_A_STRIP_VLAN | | |
982 | OFPUTIL_A_SET_DL_SRC | | |
983 | OFPUTIL_A_SET_DL_DST | | |
984 | OFPUTIL_A_SET_NW_SRC | | |
985 | OFPUTIL_A_SET_NW_DST | | |
986 | OFPUTIL_A_SET_NW_TOS | | |
987 | OFPUTIL_A_SET_TP_SRC | | |
988 | OFPUTIL_A_SET_TP_DST | | |
989 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
990 | } |
991 | ||
992 | static void | |
993 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
994 | { | |
995 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 996 | struct dpif_dp_stats s; |
6c1491fb BP |
997 | |
998 | strcpy(ots->name, "classifier"); | |
999 | ||
1000 | dpif_get_dp_stats(ofproto->dpif, &s); | |
1001 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
1002 | put_32aligned_be64(&ots->matched_count, | |
1003 | htonll(s.n_hit + ofproto->n_matches)); | |
1004 | } | |
1005 | ||
abe529af BP |
1006 | static struct ofport * |
1007 | port_alloc(void) | |
1008 | { | |
1009 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1010 | return &port->up; | |
1011 | } | |
1012 | ||
1013 | static void | |
1014 | port_dealloc(struct ofport *port_) | |
1015 | { | |
1016 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1017 | free(port); | |
1018 | } | |
1019 | ||
1020 | static int | |
1021 | port_construct(struct ofport *port_) | |
1022 | { | |
1023 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1024 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1025 | ||
f11c28c4 | 1026 | ofproto->need_revalidate = true; |
abe529af BP |
1027 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
1028 | port->bundle = NULL; | |
1029 | port->cfm = NULL; | |
1030 | port->tag = tag_create_random(); | |
d5ffa7f2 | 1031 | port->may_enable = true; |
21f7563c JP |
1032 | port->stp_port = NULL; |
1033 | port->stp_state = STP_DISABLED; | |
8b36f51e | 1034 | hmap_init(&port->priorities); |
52a90c29 BP |
1035 | port->realdev_ofp_port = 0; |
1036 | port->vlandev_vid = 0; | |
3e5b3fdb | 1037 | port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); |
abe529af BP |
1038 | |
1039 | if (ofproto->sflow) { | |
392c7182 | 1040 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
1041 | } |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | static void | |
1047 | port_destruct(struct ofport *port_) | |
1048 | { | |
1049 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1050 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1051 | ||
f11c28c4 | 1052 | ofproto->need_revalidate = true; |
abe529af | 1053 | bundle_remove(port_); |
a5610457 | 1054 | set_cfm(port_, NULL); |
abe529af | 1055 | if (ofproto->sflow) { |
bae473fe | 1056 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1057 | } |
8b36f51e EJ |
1058 | |
1059 | ofport_clear_priorities(port); | |
1060 | hmap_destroy(&port->priorities); | |
abe529af BP |
1061 | } |
1062 | ||
1063 | static void | |
1064 | port_modified(struct ofport *port_) | |
1065 | { | |
1066 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1067 | ||
1068 | if (port->bundle && port->bundle->bond) { | |
1069 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | static void | |
9e1fd49b | 1074 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1075 | { |
1076 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1077 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1078 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1079 | |
9e1fd49b BP |
1080 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
1081 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) { | |
abe529af | 1082 | ofproto->need_revalidate = true; |
7bde8dd8 | 1083 | |
9e1fd49b | 1084 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1085 | bundle_update(port->bundle); |
1086 | } | |
abe529af BP |
1087 | } |
1088 | } | |
1089 | ||
1090 | static int | |
1091 | set_sflow(struct ofproto *ofproto_, | |
1092 | const struct ofproto_sflow_options *sflow_options) | |
1093 | { | |
1094 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1095 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1096 | |
abe529af | 1097 | if (sflow_options) { |
bae473fe | 1098 | if (!ds) { |
abe529af BP |
1099 | struct ofport_dpif *ofport; |
1100 | ||
bae473fe | 1101 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1102 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1103 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1104 | } |
6ff686f2 | 1105 | ofproto->need_revalidate = true; |
abe529af | 1106 | } |
bae473fe | 1107 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1108 | } else { |
6ff686f2 PS |
1109 | if (ds) { |
1110 | dpif_sflow_destroy(ds); | |
1111 | ofproto->need_revalidate = true; | |
1112 | ofproto->sflow = NULL; | |
1113 | } | |
abe529af BP |
1114 | } |
1115 | return 0; | |
1116 | } | |
1117 | ||
1118 | static int | |
a5610457 | 1119 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1120 | { |
1121 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1122 | int error; | |
1123 | ||
a5610457 | 1124 | if (!s) { |
abe529af BP |
1125 | error = 0; |
1126 | } else { | |
1127 | if (!ofport->cfm) { | |
8c977421 EJ |
1128 | struct ofproto_dpif *ofproto; |
1129 | ||
1130 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1131 | ofproto->need_revalidate = true; | |
6f629657 | 1132 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1133 | } |
1134 | ||
a5610457 | 1135 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1136 | return 0; |
1137 | } | |
1138 | ||
1139 | error = EINVAL; | |
1140 | } | |
1141 | cfm_destroy(ofport->cfm); | |
1142 | ofport->cfm = NULL; | |
1143 | return error; | |
1144 | } | |
1145 | ||
1146 | static int | |
a5610457 | 1147 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1148 | { |
1149 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1150 | |
1151 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1152 | } |
1de11730 EJ |
1153 | |
1154 | static int | |
1155 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1156 | size_t *n_rmps) | |
1157 | { | |
1158 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1159 | ||
1160 | if (ofport->cfm) { | |
1161 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1162 | return 0; | |
1163 | } else { | |
1164 | return -1; | |
1165 | } | |
1166 | } | |
3967a833 MM |
1167 | |
1168 | static int | |
1169 | get_cfm_health(const struct ofport *ofport_) | |
1170 | { | |
1171 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1172 | ||
1173 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1174 | } | |
abe529af | 1175 | \f |
21f7563c JP |
1176 | /* Spanning Tree. */ |
1177 | ||
1178 | static void | |
1179 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1180 | { | |
1181 | struct ofproto_dpif *ofproto = ofproto_; | |
1182 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1183 | struct ofport_dpif *ofport; | |
1184 | ||
1185 | ofport = stp_port_get_aux(sp); | |
1186 | if (!ofport) { | |
1187 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1188 | ofproto->up.name, port_num); | |
1189 | } else { | |
1190 | struct eth_header *eth = pkt->l2; | |
1191 | ||
1192 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1193 | if (eth_addr_is_zero(eth->eth_src)) { | |
1194 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1195 | "with unknown MAC", ofproto->up.name, port_num); | |
1196 | } else { | |
97d6520b | 1197 | send_packet(ofport, pkt); |
21f7563c JP |
1198 | } |
1199 | } | |
1200 | ofpbuf_delete(pkt); | |
1201 | } | |
1202 | ||
1203 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1204 | static int | |
1205 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1206 | { | |
1207 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1208 | ||
1209 | /* Only revalidate flows if the configuration changed. */ | |
1210 | if (!s != !ofproto->stp) { | |
1211 | ofproto->need_revalidate = true; | |
1212 | } | |
1213 | ||
1214 | if (s) { | |
1215 | if (!ofproto->stp) { | |
1216 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1217 | send_bpdu_cb, ofproto); | |
1218 | ofproto->stp_last_tick = time_msec(); | |
1219 | } | |
1220 | ||
1221 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1222 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1223 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1224 | stp_set_max_age(ofproto->stp, s->max_age); | |
1225 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1226 | } else { | |
851bf71d EJ |
1227 | struct ofport *ofport; |
1228 | ||
1229 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1230 | set_stp_port(ofport, NULL); | |
1231 | } | |
1232 | ||
21f7563c JP |
1233 | stp_destroy(ofproto->stp); |
1234 | ofproto->stp = NULL; | |
1235 | } | |
1236 | ||
1237 | return 0; | |
1238 | } | |
1239 | ||
1240 | static int | |
1241 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1242 | { | |
1243 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1244 | ||
1245 | if (ofproto->stp) { | |
1246 | s->enabled = true; | |
1247 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1248 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1249 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1250 | } else { | |
1251 | s->enabled = false; | |
1252 | } | |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static void | |
1258 | update_stp_port_state(struct ofport_dpif *ofport) | |
1259 | { | |
1260 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1261 | enum stp_state state; | |
1262 | ||
1263 | /* Figure out new state. */ | |
1264 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1265 | : STP_DISABLED; | |
1266 | ||
1267 | /* Update state. */ | |
1268 | if (ofport->stp_state != state) { | |
9e1fd49b | 1269 | enum ofputil_port_state of_state; |
21f7563c JP |
1270 | bool fwd_change; |
1271 | ||
1272 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1273 | netdev_get_name(ofport->up.netdev), | |
1274 | stp_state_name(ofport->stp_state), | |
1275 | stp_state_name(state)); | |
1276 | if (stp_learn_in_state(ofport->stp_state) | |
1277 | != stp_learn_in_state(state)) { | |
1278 | /* xxx Learning action flows should also be flushed. */ | |
d0040604 | 1279 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
21f7563c JP |
1280 | } |
1281 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1282 | != stp_forward_in_state(state); | |
1283 | ||
1284 | ofproto->need_revalidate = true; | |
1285 | ofport->stp_state = state; | |
1286 | ofport->stp_state_entered = time_msec(); | |
1287 | ||
b308140a | 1288 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1289 | bundle_update(ofport->bundle); |
1290 | } | |
1291 | ||
1292 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1293 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1294 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1295 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1296 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1297 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1298 | : 0); | |
21f7563c JP |
1299 | ofproto_port_set_state(&ofport->up, of_state); |
1300 | } | |
1301 | } | |
1302 | ||
1303 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1304 | * caller is responsible for assigning STP port numbers and ensuring | |
1305 | * there are no duplicates. */ | |
1306 | static int | |
1307 | set_stp_port(struct ofport *ofport_, | |
1308 | const struct ofproto_port_stp_settings *s) | |
1309 | { | |
1310 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1311 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1312 | struct stp_port *sp = ofport->stp_port; | |
1313 | ||
1314 | if (!s || !s->enable) { | |
1315 | if (sp) { | |
1316 | ofport->stp_port = NULL; | |
1317 | stp_port_disable(sp); | |
ecd12731 | 1318 | update_stp_port_state(ofport); |
21f7563c JP |
1319 | } |
1320 | return 0; | |
1321 | } else if (sp && stp_port_no(sp) != s->port_num | |
1322 | && ofport == stp_port_get_aux(sp)) { | |
1323 | /* The port-id changed, so disable the old one if it's not | |
1324 | * already in use by another port. */ | |
1325 | stp_port_disable(sp); | |
1326 | } | |
1327 | ||
1328 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1329 | stp_port_enable(sp); | |
1330 | ||
1331 | stp_port_set_aux(sp, ofport); | |
1332 | stp_port_set_priority(sp, s->priority); | |
1333 | stp_port_set_path_cost(sp, s->path_cost); | |
1334 | ||
1335 | update_stp_port_state(ofport); | |
1336 | ||
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | static int | |
1341 | get_stp_port_status(struct ofport *ofport_, | |
1342 | struct ofproto_port_stp_status *s) | |
1343 | { | |
1344 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1345 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1346 | struct stp_port *sp = ofport->stp_port; | |
1347 | ||
1348 | if (!ofproto->stp || !sp) { | |
1349 | s->enabled = false; | |
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | s->enabled = true; | |
1354 | s->port_id = stp_port_get_id(sp); | |
1355 | s->state = stp_port_get_state(sp); | |
1356 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1357 | s->role = stp_port_get_role(sp); | |
80740385 | 1358 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1359 | |
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | static void | |
1364 | stp_run(struct ofproto_dpif *ofproto) | |
1365 | { | |
1366 | if (ofproto->stp) { | |
1367 | long long int now = time_msec(); | |
1368 | long long int elapsed = now - ofproto->stp_last_tick; | |
1369 | struct stp_port *sp; | |
1370 | ||
1371 | if (elapsed > 0) { | |
1372 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1373 | ofproto->stp_last_tick = now; | |
1374 | } | |
1375 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1376 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1377 | ||
1378 | if (ofport) { | |
1379 | update_stp_port_state(ofport); | |
1380 | } | |
1381 | } | |
6ae50723 EJ |
1382 | |
1383 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1384 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1385 | } | |
21f7563c JP |
1386 | } |
1387 | } | |
1388 | ||
1389 | static void | |
1390 | stp_wait(struct ofproto_dpif *ofproto) | |
1391 | { | |
1392 | if (ofproto->stp) { | |
1393 | poll_timer_wait(1000); | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | /* Returns true if STP should process 'flow'. */ | |
1398 | static bool | |
1399 | stp_should_process_flow(const struct flow *flow) | |
1400 | { | |
1401 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1402 | } | |
1403 | ||
1404 | static void | |
1405 | stp_process_packet(const struct ofport_dpif *ofport, | |
1406 | const struct ofpbuf *packet) | |
1407 | { | |
1408 | struct ofpbuf payload = *packet; | |
1409 | struct eth_header *eth = payload.data; | |
1410 | struct stp_port *sp = ofport->stp_port; | |
1411 | ||
1412 | /* Sink packets on ports that have STP disabled when the bridge has | |
1413 | * STP enabled. */ | |
1414 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1415 | return; | |
1416 | } | |
1417 | ||
1418 | /* Trim off padding on payload. */ | |
c573540b BP |
1419 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1420 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1421 | } |
1422 | ||
1423 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1424 | stp_received_bpdu(sp, payload.data, payload.size); | |
1425 | } | |
1426 | } | |
1427 | \f | |
8b36f51e EJ |
1428 | static struct priority_to_dscp * |
1429 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1430 | { | |
1431 | struct priority_to_dscp *pdscp; | |
1432 | uint32_t hash; | |
1433 | ||
1434 | hash = hash_int(priority, 0); | |
1435 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1436 | if (pdscp->priority == priority) { | |
1437 | return pdscp; | |
1438 | } | |
1439 | } | |
1440 | return NULL; | |
1441 | } | |
1442 | ||
1443 | static void | |
1444 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1445 | { | |
1446 | struct priority_to_dscp *pdscp, *next; | |
1447 | ||
1448 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1449 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1450 | free(pdscp); | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | static int | |
1455 | set_queues(struct ofport *ofport_, | |
1456 | const struct ofproto_port_queue *qdscp_list, | |
1457 | size_t n_qdscp) | |
1458 | { | |
1459 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1460 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1461 | struct hmap new = HMAP_INITIALIZER(&new); | |
1462 | size_t i; | |
1463 | ||
1464 | for (i = 0; i < n_qdscp; i++) { | |
1465 | struct priority_to_dscp *pdscp; | |
1466 | uint32_t priority; | |
1467 | uint8_t dscp; | |
1468 | ||
1469 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1470 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1471 | &priority)) { | |
1472 | continue; | |
1473 | } | |
1474 | ||
1475 | pdscp = get_priority(ofport, priority); | |
1476 | if (pdscp) { | |
1477 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1478 | } else { | |
1479 | pdscp = xmalloc(sizeof *pdscp); | |
1480 | pdscp->priority = priority; | |
1481 | pdscp->dscp = dscp; | |
1482 | ofproto->need_revalidate = true; | |
1483 | } | |
1484 | ||
1485 | if (pdscp->dscp != dscp) { | |
1486 | pdscp->dscp = dscp; | |
1487 | ofproto->need_revalidate = true; | |
1488 | } | |
1489 | ||
1490 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1491 | } | |
1492 | ||
1493 | if (!hmap_is_empty(&ofport->priorities)) { | |
1494 | ofport_clear_priorities(ofport); | |
1495 | ofproto->need_revalidate = true; | |
1496 | } | |
1497 | ||
1498 | hmap_swap(&new, &ofport->priorities); | |
1499 | hmap_destroy(&new); | |
1500 | ||
1501 | return 0; | |
1502 | } | |
1503 | \f | |
abe529af BP |
1504 | /* Bundles. */ |
1505 | ||
b44a10b7 BP |
1506 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1507 | * ofproto to revalidate every flow. | |
1508 | * | |
1509 | * Normally MAC learning entries are removed only from the ofproto associated | |
1510 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1511 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1512 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1513 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1514 | * with the host from which it migrated. */ | |
abe529af | 1515 | static void |
b44a10b7 | 1516 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1517 | { |
1518 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1519 | struct mac_learning *ml = ofproto->ml; | |
1520 | struct mac_entry *mac, *next_mac; | |
1521 | ||
1522 | ofproto->need_revalidate = true; | |
1523 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1524 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1525 | if (all_ofprotos) { |
1526 | struct ofproto_dpif *o; | |
1527 | ||
1528 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1529 | if (o != ofproto) { | |
1530 | struct mac_entry *e; | |
1531 | ||
1532 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1533 | NULL); | |
1534 | if (e) { | |
1535 | tag_set_add(&o->revalidate_set, e->tag); | |
1536 | mac_learning_expire(o->ml, e); | |
1537 | } | |
1538 | } | |
1539 | } | |
1540 | } | |
1541 | ||
abe529af BP |
1542 | mac_learning_expire(ml, mac); |
1543 | } | |
1544 | } | |
1545 | } | |
1546 | ||
1547 | static struct ofbundle * | |
1548 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1549 | { | |
1550 | struct ofbundle *bundle; | |
1551 | ||
1552 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1553 | &ofproto->bundles) { | |
1554 | if (bundle->aux == aux) { | |
1555 | return bundle; | |
1556 | } | |
1557 | } | |
1558 | return NULL; | |
1559 | } | |
1560 | ||
1561 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1562 | * ones that are found to 'bundles'. */ | |
1563 | static void | |
1564 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1565 | void **auxes, size_t n_auxes, | |
1566 | struct hmapx *bundles) | |
1567 | { | |
1568 | size_t i; | |
1569 | ||
1570 | hmapx_init(bundles); | |
1571 | for (i = 0; i < n_auxes; i++) { | |
1572 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1573 | if (bundle) { | |
1574 | hmapx_add(bundles, bundle); | |
1575 | } | |
1576 | } | |
1577 | } | |
1578 | ||
7bde8dd8 JP |
1579 | static void |
1580 | bundle_update(struct ofbundle *bundle) | |
1581 | { | |
1582 | struct ofport_dpif *port; | |
1583 | ||
1584 | bundle->floodable = true; | |
1585 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
1586 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1587 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1588 | bundle->floodable = false; |
1589 | break; | |
1590 | } | |
1591 | } | |
1592 | } | |
1593 | ||
abe529af BP |
1594 | static void |
1595 | bundle_del_port(struct ofport_dpif *port) | |
1596 | { | |
1597 | struct ofbundle *bundle = port->bundle; | |
1598 | ||
6f77f4ae BP |
1599 | bundle->ofproto->need_revalidate = true; |
1600 | ||
abe529af BP |
1601 | list_remove(&port->bundle_node); |
1602 | port->bundle = NULL; | |
1603 | ||
1604 | if (bundle->lacp) { | |
1605 | lacp_slave_unregister(bundle->lacp, port); | |
1606 | } | |
1607 | if (bundle->bond) { | |
1608 | bond_slave_unregister(bundle->bond, port); | |
1609 | } | |
1610 | ||
7bde8dd8 | 1611 | bundle_update(bundle); |
abe529af BP |
1612 | } |
1613 | ||
1614 | static bool | |
1615 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1616 | struct lacp_slave_settings *lacp, |
1617 | uint32_t bond_stable_id) | |
abe529af BP |
1618 | { |
1619 | struct ofport_dpif *port; | |
1620 | ||
1621 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1622 | if (!port) { | |
1623 | return false; | |
1624 | } | |
1625 | ||
1626 | if (port->bundle != bundle) { | |
6f77f4ae | 1627 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1628 | if (port->bundle) { |
1629 | bundle_del_port(port); | |
1630 | } | |
1631 | ||
1632 | port->bundle = bundle; | |
1633 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
1634 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1635 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1636 | bundle->floodable = false; |
1637 | } | |
1638 | } | |
1639 | if (lacp) { | |
4a86aece | 1640 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1641 | lacp_slave_register(bundle->lacp, port, lacp); |
1642 | } | |
1643 | ||
00794817 BP |
1644 | port->bond_stable_id = bond_stable_id; |
1645 | ||
abe529af BP |
1646 | return true; |
1647 | } | |
1648 | ||
1649 | static void | |
1650 | bundle_destroy(struct ofbundle *bundle) | |
1651 | { | |
1652 | struct ofproto_dpif *ofproto; | |
1653 | struct ofport_dpif *port, *next_port; | |
1654 | int i; | |
1655 | ||
1656 | if (!bundle) { | |
1657 | return; | |
1658 | } | |
1659 | ||
1660 | ofproto = bundle->ofproto; | |
1661 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1662 | struct ofmirror *m = ofproto->mirrors[i]; | |
1663 | if (m) { | |
1664 | if (m->out == bundle) { | |
1665 | mirror_destroy(m); | |
1666 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1667 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1668 | ofproto->need_revalidate = true; | |
1669 | } | |
1670 | } | |
1671 | } | |
1672 | ||
1673 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1674 | bundle_del_port(port); | |
1675 | } | |
1676 | ||
b44a10b7 | 1677 | bundle_flush_macs(bundle, true); |
abe529af BP |
1678 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1679 | free(bundle->name); | |
1680 | free(bundle->trunks); | |
1681 | lacp_destroy(bundle->lacp); | |
1682 | bond_destroy(bundle->bond); | |
1683 | free(bundle); | |
1684 | } | |
1685 | ||
1686 | static int | |
1687 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1688 | const struct ofproto_bundle_settings *s) | |
1689 | { | |
1690 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1691 | bool need_flush = false; | |
abe529af BP |
1692 | struct ofport_dpif *port; |
1693 | struct ofbundle *bundle; | |
ecac4ebf BP |
1694 | unsigned long *trunks; |
1695 | int vlan; | |
abe529af BP |
1696 | size_t i; |
1697 | bool ok; | |
1698 | ||
1699 | if (!s) { | |
1700 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1701 | return 0; | |
1702 | } | |
1703 | ||
1704 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1705 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1706 | ||
1707 | bundle = bundle_lookup(ofproto, aux); | |
1708 | if (!bundle) { | |
1709 | bundle = xmalloc(sizeof *bundle); | |
1710 | ||
1711 | bundle->ofproto = ofproto; | |
1712 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1713 | hash_pointer(aux, 0)); | |
1714 | bundle->aux = aux; | |
1715 | bundle->name = NULL; | |
1716 | ||
1717 | list_init(&bundle->ports); | |
ecac4ebf | 1718 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1719 | bundle->vlan = -1; |
1720 | bundle->trunks = NULL; | |
5e9ceccd | 1721 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1722 | bundle->lacp = NULL; |
1723 | bundle->bond = NULL; | |
1724 | ||
1725 | bundle->floodable = true; | |
1726 | ||
1727 | bundle->src_mirrors = 0; | |
1728 | bundle->dst_mirrors = 0; | |
1729 | bundle->mirror_out = 0; | |
1730 | } | |
1731 | ||
1732 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1733 | free(bundle->name); | |
1734 | bundle->name = xstrdup(s->name); | |
1735 | } | |
1736 | ||
1737 | /* LACP. */ | |
1738 | if (s->lacp) { | |
1739 | if (!bundle->lacp) { | |
8c977421 | 1740 | ofproto->need_revalidate = true; |
abe529af BP |
1741 | bundle->lacp = lacp_create(); |
1742 | } | |
1743 | lacp_configure(bundle->lacp, s->lacp); | |
1744 | } else { | |
1745 | lacp_destroy(bundle->lacp); | |
1746 | bundle->lacp = NULL; | |
1747 | } | |
1748 | ||
1749 | /* Update set of ports. */ | |
1750 | ok = true; | |
1751 | for (i = 0; i < s->n_slaves; i++) { | |
1752 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1753 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1754 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1755 | ok = false; |
1756 | } | |
1757 | } | |
1758 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1759 | struct ofport_dpif *next_port; | |
1760 | ||
1761 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1762 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1763 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1764 | goto found; |
1765 | } | |
1766 | } | |
1767 | ||
1768 | bundle_del_port(port); | |
1769 | found: ; | |
1770 | } | |
1771 | } | |
1772 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1773 | ||
1774 | if (list_is_empty(&bundle->ports)) { | |
1775 | bundle_destroy(bundle); | |
1776 | return EINVAL; | |
1777 | } | |
1778 | ||
ecac4ebf | 1779 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1780 | if (s->vlan_mode != bundle->vlan_mode |
1781 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1782 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1783 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1784 | need_flush = true; |
1785 | } | |
1786 | ||
abe529af | 1787 | /* Set VLAN tag. */ |
ecac4ebf BP |
1788 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1789 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1790 | : 0); | |
1791 | if (vlan != bundle->vlan) { | |
1792 | bundle->vlan = vlan; | |
abe529af BP |
1793 | need_flush = true; |
1794 | } | |
1795 | ||
1796 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1797 | switch (s->vlan_mode) { |
1798 | case PORT_VLAN_ACCESS: | |
1799 | trunks = NULL; | |
1800 | break; | |
1801 | ||
1802 | case PORT_VLAN_TRUNK: | |
1803 | trunks = (unsigned long *) s->trunks; | |
1804 | break; | |
1805 | ||
1806 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1807 | case PORT_VLAN_NATIVE_TAGGED: | |
1808 | if (vlan != 0 && (!s->trunks | |
1809 | || !bitmap_is_set(s->trunks, vlan) | |
1810 | || bitmap_is_set(s->trunks, 0))) { | |
1811 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1812 | if (s->trunks) { | |
1813 | trunks = bitmap_clone(s->trunks, 4096); | |
1814 | } else { | |
1815 | trunks = bitmap_allocate1(4096); | |
1816 | } | |
1817 | bitmap_set1(trunks, vlan); | |
1818 | bitmap_set0(trunks, 0); | |
1819 | } else { | |
1820 | trunks = (unsigned long *) s->trunks; | |
1821 | } | |
1822 | break; | |
1823 | ||
1824 | default: | |
1825 | NOT_REACHED(); | |
1826 | } | |
abe529af BP |
1827 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1828 | free(bundle->trunks); | |
ecac4ebf BP |
1829 | if (trunks == s->trunks) { |
1830 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1831 | } else { | |
1832 | bundle->trunks = trunks; | |
1833 | trunks = NULL; | |
1834 | } | |
abe529af BP |
1835 | need_flush = true; |
1836 | } | |
ecac4ebf BP |
1837 | if (trunks != s->trunks) { |
1838 | free(trunks); | |
1839 | } | |
abe529af BP |
1840 | |
1841 | /* Bonding. */ | |
1842 | if (!list_is_short(&bundle->ports)) { | |
1843 | bundle->ofproto->has_bonded_bundles = true; | |
1844 | if (bundle->bond) { | |
1845 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1846 | ofproto->need_revalidate = true; | |
1847 | } | |
1848 | } else { | |
1849 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1850 | ofproto->need_revalidate = true; |
abe529af BP |
1851 | } |
1852 | ||
1853 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1854 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1855 | port->up.netdev); |
1856 | } | |
1857 | } else { | |
1858 | bond_destroy(bundle->bond); | |
1859 | bundle->bond = NULL; | |
1860 | } | |
1861 | ||
1862 | /* If we changed something that would affect MAC learning, un-learn | |
1863 | * everything on this port and force flow revalidation. */ | |
1864 | if (need_flush) { | |
b44a10b7 | 1865 | bundle_flush_macs(bundle, false); |
abe529af BP |
1866 | } |
1867 | ||
1868 | return 0; | |
1869 | } | |
1870 | ||
1871 | static void | |
1872 | bundle_remove(struct ofport *port_) | |
1873 | { | |
1874 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1875 | struct ofbundle *bundle = port->bundle; | |
1876 | ||
1877 | if (bundle) { | |
1878 | bundle_del_port(port); | |
1879 | if (list_is_empty(&bundle->ports)) { | |
1880 | bundle_destroy(bundle); | |
1881 | } else if (list_is_short(&bundle->ports)) { | |
1882 | bond_destroy(bundle->bond); | |
1883 | bundle->bond = NULL; | |
1884 | } | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | static void | |
5f877369 | 1889 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1890 | { |
1891 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1892 | struct ofport_dpif *port = port_; | |
1893 | uint8_t ea[ETH_ADDR_LEN]; | |
1894 | int error; | |
1895 | ||
1896 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1897 | if (!error) { | |
abe529af | 1898 | struct ofpbuf packet; |
5f877369 | 1899 | void *packet_pdu; |
abe529af BP |
1900 | |
1901 | ofpbuf_init(&packet, 0); | |
1902 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1903 | pdu_size); |
1904 | memcpy(packet_pdu, pdu, pdu_size); | |
1905 | ||
97d6520b | 1906 | send_packet(port, &packet); |
abe529af BP |
1907 | ofpbuf_uninit(&packet); |
1908 | } else { | |
1909 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1910 | "%s (%s)", port->bundle->name, | |
1911 | netdev_get_name(port->up.netdev), strerror(error)); | |
1912 | } | |
1913 | } | |
1914 | ||
1915 | static void | |
1916 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1917 | { | |
1918 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1919 | int error, n_packets, n_errors; | |
1920 | struct mac_entry *e; | |
1921 | ||
1922 | error = n_packets = n_errors = 0; | |
1923 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1924 | if (e->port.p != bundle) { | |
ea131871 JG |
1925 | struct ofpbuf *learning_packet; |
1926 | struct ofport_dpif *port; | |
4dd1e3ca | 1927 | void *port_void; |
ea131871 JG |
1928 | int ret; |
1929 | ||
4dd1e3ca BP |
1930 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
1931 | * struct ofport_dpif more effective. */ | |
1932 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
1933 | e->mac, e->vlan, | |
1934 | &port_void); | |
1935 | port = port_void; | |
97d6520b | 1936 | ret = send_packet(port, learning_packet); |
ea131871 | 1937 | ofpbuf_delete(learning_packet); |
abe529af BP |
1938 | if (ret) { |
1939 | error = ret; | |
1940 | n_errors++; | |
1941 | } | |
1942 | n_packets++; | |
1943 | } | |
1944 | } | |
1945 | ||
1946 | if (n_errors) { | |
1947 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1948 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1949 | "packets, last error was: %s", | |
1950 | bundle->name, n_errors, n_packets, strerror(error)); | |
1951 | } else { | |
1952 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1953 | bundle->name, n_packets); | |
1954 | } | |
1955 | } | |
1956 | ||
1957 | static void | |
1958 | bundle_run(struct ofbundle *bundle) | |
1959 | { | |
1960 | if (bundle->lacp) { | |
1961 | lacp_run(bundle->lacp, send_pdu_cb); | |
1962 | } | |
1963 | if (bundle->bond) { | |
1964 | struct ofport_dpif *port; | |
1965 | ||
1966 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1967 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1968 | } |
1969 | ||
1970 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
bdebeece | 1971 | lacp_status(bundle->lacp)); |
abe529af BP |
1972 | if (bond_should_send_learning_packets(bundle->bond)) { |
1973 | bundle_send_learning_packets(bundle); | |
1974 | } | |
1975 | } | |
1976 | } | |
1977 | ||
1978 | static void | |
1979 | bundle_wait(struct ofbundle *bundle) | |
1980 | { | |
1981 | if (bundle->lacp) { | |
1982 | lacp_wait(bundle->lacp); | |
1983 | } | |
1984 | if (bundle->bond) { | |
1985 | bond_wait(bundle->bond); | |
1986 | } | |
1987 | } | |
1988 | \f | |
1989 | /* Mirrors. */ | |
1990 | ||
1991 | static int | |
1992 | mirror_scan(struct ofproto_dpif *ofproto) | |
1993 | { | |
1994 | int idx; | |
1995 | ||
1996 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1997 | if (!ofproto->mirrors[idx]) { | |
1998 | return idx; | |
1999 | } | |
2000 | } | |
2001 | return -1; | |
2002 | } | |
2003 | ||
2004 | static struct ofmirror * | |
2005 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2006 | { | |
2007 | int i; | |
2008 | ||
2009 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2010 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2011 | if (mirror && mirror->aux == aux) { | |
2012 | return mirror; | |
2013 | } | |
2014 | } | |
2015 | ||
2016 | return NULL; | |
2017 | } | |
2018 | ||
9ba15e2a BP |
2019 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
2020 | static void | |
2021 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2022 | { | |
2023 | int i; | |
2024 | ||
2025 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2026 | struct ofmirror *m = ofproto->mirrors[i]; | |
2027 | ||
2028 | if (m) { | |
2029 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2030 | } | |
2031 | } | |
2032 | ||
2033 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2034 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2035 | int j; | |
2036 | ||
2037 | if (!m1) { | |
2038 | continue; | |
2039 | } | |
2040 | ||
2041 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2042 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2043 | ||
edb0540b | 2044 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2045 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2046 | m2->dup_mirrors |= m1->dup_mirrors; | |
2047 | } | |
2048 | } | |
2049 | } | |
2050 | } | |
2051 | ||
abe529af BP |
2052 | static int |
2053 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2054 | const struct ofproto_mirror_settings *s) | |
2055 | { | |
2056 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2057 | mirror_mask_t mirror_bit; | |
2058 | struct ofbundle *bundle; | |
2059 | struct ofmirror *mirror; | |
2060 | struct ofbundle *out; | |
2061 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2062 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2063 | int out_vlan; | |
2064 | ||
2065 | mirror = mirror_lookup(ofproto, aux); | |
2066 | if (!s) { | |
2067 | mirror_destroy(mirror); | |
2068 | return 0; | |
2069 | } | |
2070 | if (!mirror) { | |
2071 | int idx; | |
2072 | ||
2073 | idx = mirror_scan(ofproto); | |
2074 | if (idx < 0) { | |
2075 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2076 | "cannot create %s", | |
2077 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2078 | return EFBIG; | |
2079 | } | |
2080 | ||
2081 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2082 | mirror->ofproto = ofproto; | |
2083 | mirror->idx = idx; | |
8b28d864 | 2084 | mirror->aux = aux; |
abe529af BP |
2085 | mirror->out_vlan = -1; |
2086 | mirror->name = NULL; | |
2087 | } | |
2088 | ||
2089 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2090 | free(mirror->name); | |
2091 | mirror->name = xstrdup(s->name); | |
2092 | } | |
2093 | ||
2094 | /* Get the new configuration. */ | |
2095 | if (s->out_bundle) { | |
2096 | out = bundle_lookup(ofproto, s->out_bundle); | |
2097 | if (!out) { | |
2098 | mirror_destroy(mirror); | |
2099 | return EINVAL; | |
2100 | } | |
2101 | out_vlan = -1; | |
2102 | } else { | |
2103 | out = NULL; | |
2104 | out_vlan = s->out_vlan; | |
2105 | } | |
2106 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2107 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2108 | ||
2109 | /* If the configuration has not changed, do nothing. */ | |
2110 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2111 | && hmapx_equals(&dsts, &mirror->dsts) | |
2112 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2113 | && mirror->out == out | |
2114 | && mirror->out_vlan == out_vlan) | |
2115 | { | |
2116 | hmapx_destroy(&srcs); | |
2117 | hmapx_destroy(&dsts); | |
2118 | return 0; | |
2119 | } | |
2120 | ||
2121 | hmapx_swap(&srcs, &mirror->srcs); | |
2122 | hmapx_destroy(&srcs); | |
2123 | ||
2124 | hmapx_swap(&dsts, &mirror->dsts); | |
2125 | hmapx_destroy(&dsts); | |
2126 | ||
2127 | free(mirror->vlans); | |
2128 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2129 | ||
2130 | mirror->out = out; | |
2131 | mirror->out_vlan = out_vlan; | |
2132 | ||
2133 | /* Update bundles. */ | |
2134 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2135 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2136 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2137 | bundle->src_mirrors |= mirror_bit; | |
2138 | } else { | |
2139 | bundle->src_mirrors &= ~mirror_bit; | |
2140 | } | |
2141 | ||
2142 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2143 | bundle->dst_mirrors |= mirror_bit; | |
2144 | } else { | |
2145 | bundle->dst_mirrors &= ~mirror_bit; | |
2146 | } | |
2147 | ||
2148 | if (mirror->out == bundle) { | |
2149 | bundle->mirror_out |= mirror_bit; | |
2150 | } else { | |
2151 | bundle->mirror_out &= ~mirror_bit; | |
2152 | } | |
2153 | } | |
2154 | ||
2155 | ofproto->need_revalidate = true; | |
d0040604 | 2156 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
9ba15e2a | 2157 | mirror_update_dups(ofproto); |
abe529af BP |
2158 | |
2159 | return 0; | |
2160 | } | |
2161 | ||
2162 | static void | |
2163 | mirror_destroy(struct ofmirror *mirror) | |
2164 | { | |
2165 | struct ofproto_dpif *ofproto; | |
2166 | mirror_mask_t mirror_bit; | |
2167 | struct ofbundle *bundle; | |
2168 | ||
2169 | if (!mirror) { | |
2170 | return; | |
2171 | } | |
2172 | ||
2173 | ofproto = mirror->ofproto; | |
2174 | ofproto->need_revalidate = true; | |
d0040604 | 2175 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2176 | |
2177 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2178 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2179 | bundle->src_mirrors &= ~mirror_bit; | |
2180 | bundle->dst_mirrors &= ~mirror_bit; | |
2181 | bundle->mirror_out &= ~mirror_bit; | |
2182 | } | |
2183 | ||
2184 | hmapx_destroy(&mirror->srcs); | |
2185 | hmapx_destroy(&mirror->dsts); | |
2186 | free(mirror->vlans); | |
2187 | ||
2188 | ofproto->mirrors[mirror->idx] = NULL; | |
2189 | free(mirror->name); | |
2190 | free(mirror); | |
9ba15e2a BP |
2191 | |
2192 | mirror_update_dups(ofproto); | |
abe529af BP |
2193 | } |
2194 | ||
9d24de3b JP |
2195 | static int |
2196 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2197 | uint64_t *packets, uint64_t *bytes) | |
2198 | { | |
2199 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2200 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2201 | ||
2202 | if (!mirror) { | |
2203 | *packets = *bytes = UINT64_MAX; | |
2204 | return 0; | |
2205 | } | |
2206 | ||
2207 | *packets = mirror->packet_count; | |
2208 | *bytes = mirror->byte_count; | |
2209 | ||
2210 | return 0; | |
2211 | } | |
2212 | ||
abe529af BP |
2213 | static int |
2214 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2215 | { | |
2216 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2217 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
d0040604 | 2218 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2219 | } |
2220 | return 0; | |
2221 | } | |
2222 | ||
2223 | static bool | |
b4affc74 | 2224 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2225 | { |
2226 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2227 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2228 | return bundle && bundle->mirror_out != 0; | |
2229 | } | |
8402c74b SS |
2230 | |
2231 | static void | |
b53055f4 | 2232 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2233 | { |
2234 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2235 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
2236 | ofproto->need_revalidate = true; | |
2237 | } | |
e764773c BP |
2238 | |
2239 | static void | |
2240 | set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) | |
2241 | { | |
2242 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2243 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
2244 | } | |
abe529af BP |
2245 | \f |
2246 | /* Ports. */ | |
2247 | ||
2248 | static struct ofport_dpif * | |
2249 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
2250 | { | |
7df6a8bd BP |
2251 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2252 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2253 | } |
2254 | ||
2255 | static struct ofport_dpif * | |
2256 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
2257 | { | |
2258 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2259 | } | |
2260 | ||
2261 | static void | |
2262 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2263 | struct dpif_port *dpif_port) | |
2264 | { | |
2265 | ofproto_port->name = dpif_port->name; | |
2266 | ofproto_port->type = dpif_port->type; | |
2267 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2268 | } | |
2269 | ||
2270 | static void | |
2271 | port_run(struct ofport_dpif *ofport) | |
2272 | { | |
3e5b3fdb EJ |
2273 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2274 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2275 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2276 | ||
3e5b3fdb EJ |
2277 | ofport->carrier_seq = carrier_seq; |
2278 | ||
abe529af BP |
2279 | if (ofport->cfm) { |
2280 | cfm_run(ofport->cfm); | |
2281 | ||
2282 | if (cfm_should_send_ccm(ofport->cfm)) { | |
2283 | struct ofpbuf packet; | |
abe529af BP |
2284 | |
2285 | ofpbuf_init(&packet, 0); | |
9e1fd49b | 2286 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); |
97d6520b | 2287 | send_packet(ofport, &packet); |
abe529af BP |
2288 | ofpbuf_uninit(&packet); |
2289 | } | |
015e08bc | 2290 | |
86dc6501 EJ |
2291 | enable = enable && !cfm_get_fault(ofport->cfm) |
2292 | && cfm_get_opup(ofport->cfm); | |
abe529af | 2293 | } |
015e08bc EJ |
2294 | |
2295 | if (ofport->bundle) { | |
2296 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2297 | if (carrier_changed) { |
2298 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2299 | } | |
015e08bc EJ |
2300 | } |
2301 | ||
daff3353 EJ |
2302 | if (ofport->may_enable != enable) { |
2303 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2304 | ||
2305 | if (ofproto->has_bundle_action) { | |
2306 | ofproto->need_revalidate = true; | |
2307 | } | |
2308 | } | |
2309 | ||
015e08bc | 2310 | ofport->may_enable = enable; |
abe529af BP |
2311 | } |
2312 | ||
2313 | static void | |
2314 | port_wait(struct ofport_dpif *ofport) | |
2315 | { | |
2316 | if (ofport->cfm) { | |
2317 | cfm_wait(ofport->cfm); | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | static int | |
2322 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2323 | struct ofproto_port *ofproto_port) | |
2324 | { | |
2325 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2326 | struct dpif_port dpif_port; | |
2327 | int error; | |
2328 | ||
2329 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2330 | if (!error) { | |
2331 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2332 | } | |
2333 | return error; | |
2334 | } | |
2335 | ||
2336 | static int | |
2337 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2338 | { | |
2339 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2340 | uint16_t odp_port; | |
2341 | int error; | |
2342 | ||
2343 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2344 | if (!error) { | |
2345 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2346 | } | |
2347 | return error; | |
2348 | } | |
2349 | ||
2350 | static int | |
2351 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2352 | { | |
2353 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2354 | int error; | |
2355 | ||
2356 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2357 | if (!error) { | |
2358 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2359 | if (ofport) { | |
2360 | /* The caller is going to close ofport->up.netdev. If this is a | |
2361 | * bonded port, then the bond is using that netdev, so remove it | |
2362 | * from the bond. The client will need to reconfigure everything | |
2363 | * after deleting ports, so then the slave will get re-added. */ | |
2364 | bundle_remove(&ofport->up); | |
2365 | } | |
2366 | } | |
2367 | return error; | |
2368 | } | |
2369 | ||
6527c598 PS |
2370 | static int |
2371 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2372 | { | |
2373 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2374 | int error; | |
2375 | ||
2376 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2377 | ||
2378 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2380 | ||
2381 | /* ofproto->stats.tx_packets represents packets that we created | |
2382 | * internally and sent to some port (e.g. packets sent with | |
2383 | * send_packet()). Account for them as if they had come from | |
2384 | * OFPP_LOCAL and got forwarded. */ | |
2385 | ||
2386 | if (stats->rx_packets != UINT64_MAX) { | |
2387 | stats->rx_packets += ofproto->stats.tx_packets; | |
2388 | } | |
2389 | ||
2390 | if (stats->rx_bytes != UINT64_MAX) { | |
2391 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2392 | } | |
2393 | ||
2394 | /* ofproto->stats.rx_packets represents packets that were received on | |
2395 | * some port and we processed internally and dropped (e.g. STP). | |
2396 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2397 | ||
2398 | if (stats->tx_packets != UINT64_MAX) { | |
2399 | stats->tx_packets += ofproto->stats.rx_packets; | |
2400 | } | |
2401 | ||
2402 | if (stats->tx_bytes != UINT64_MAX) { | |
2403 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2404 | } | |
2405 | } | |
2406 | ||
2407 | return error; | |
2408 | } | |
2409 | ||
2410 | /* Account packets for LOCAL port. */ | |
2411 | static void | |
2412 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2413 | size_t tx_size, size_t rx_size) | |
2414 | { | |
2415 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2416 | ||
2417 | if (rx_size) { | |
2418 | ofproto->stats.rx_packets++; | |
2419 | ofproto->stats.rx_bytes += rx_size; | |
2420 | } | |
2421 | if (tx_size) { | |
2422 | ofproto->stats.tx_packets++; | |
2423 | ofproto->stats.tx_bytes += tx_size; | |
2424 | } | |
2425 | } | |
2426 | ||
abe529af BP |
2427 | struct port_dump_state { |
2428 | struct dpif_port_dump dump; | |
2429 | bool done; | |
2430 | }; | |
2431 | ||
2432 | static int | |
2433 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2434 | { | |
2435 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2436 | struct port_dump_state *state; | |
2437 | ||
2438 | *statep = state = xmalloc(sizeof *state); | |
2439 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2440 | state->done = false; | |
2441 | return 0; | |
2442 | } | |
2443 | ||
2444 | static int | |
2445 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2446 | struct ofproto_port *port) | |
2447 | { | |
2448 | struct port_dump_state *state = state_; | |
2449 | struct dpif_port dpif_port; | |
2450 | ||
2451 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2452 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2453 | return 0; | |
2454 | } else { | |
2455 | int error = dpif_port_dump_done(&state->dump); | |
2456 | state->done = true; | |
2457 | return error ? error : EOF; | |
2458 | } | |
2459 | } | |
2460 | ||
2461 | static int | |
2462 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2463 | { | |
2464 | struct port_dump_state *state = state_; | |
2465 | ||
2466 | if (!state->done) { | |
2467 | dpif_port_dump_done(&state->dump); | |
2468 | } | |
2469 | free(state); | |
2470 | return 0; | |
2471 | } | |
2472 | ||
2473 | static int | |
2474 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2475 | { | |
2476 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2477 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2478 | } | |
2479 | ||
2480 | static void | |
2481 | port_poll_wait(const struct ofproto *ofproto_) | |
2482 | { | |
2483 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2484 | dpif_port_poll_wait(ofproto->dpif); | |
2485 | } | |
2486 | ||
2487 | static int | |
2488 | port_is_lacp_current(const struct ofport *ofport_) | |
2489 | { | |
2490 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2491 | return (ofport->bundle && ofport->bundle->lacp | |
2492 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2493 | : -1); | |
2494 | } | |
2495 | \f | |
2496 | /* Upcall handling. */ | |
2497 | ||
501f8d1f BP |
2498 | /* Flow miss batching. |
2499 | * | |
2500 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2501 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2502 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2503 | * more packets, plus possibly installing the flow in the dpif. | |
2504 | * | |
2505 | * So far we only batch the operations that affect flow setup time the most. | |
2506 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2507 | struct flow_miss { | |
2508 | struct hmap_node hmap_node; | |
2509 | struct flow flow; | |
b0f7b9b5 | 2510 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2511 | const struct nlattr *key; |
2512 | size_t key_len; | |
e84173dc | 2513 | ovs_be16 initial_tci; |
501f8d1f BP |
2514 | struct list packets; |
2515 | }; | |
2516 | ||
2517 | struct flow_miss_op { | |
c2b565b5 | 2518 | struct dpif_op dpif_op; |
5fe20d5d BP |
2519 | struct subfacet *subfacet; /* Subfacet */ |
2520 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
2521 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
501f8d1f BP |
2522 | }; |
2523 | ||
62cd7072 BP |
2524 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2525 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2526 | * configurations. */ |
62cd7072 | 2527 | static void |
a39edbd4 | 2528 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 2529 | const struct flow *flow) |
62cd7072 BP |
2530 | { |
2531 | struct ofputil_packet_in pin; | |
2532 | ||
3e3252fa EJ |
2533 | pin.packet = packet->data; |
2534 | pin.packet_len = packet->size; | |
62cd7072 | 2535 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 2536 | pin.controller_id = 0; |
54834960 EJ |
2537 | |
2538 | pin.table_id = 0; | |
2539 | pin.cookie = 0; | |
2540 | ||
62cd7072 | 2541 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
2542 | |
2543 | flow_get_metadata(flow, &pin.fmd); | |
2544 | ||
2545 | /* Registers aren't meaningful on a miss. */ | |
2546 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2547 | ||
d8653c38 | 2548 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
2549 | } |
2550 | ||
abe529af BP |
2551 | static bool |
2552 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2553 | const struct ofpbuf *packet) | |
2554 | { | |
b6e001b6 EJ |
2555 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2556 | ||
2557 | if (!ofport) { | |
2558 | return false; | |
2559 | } | |
2560 | ||
ef9819b5 | 2561 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2562 | if (packet) { |
abe529af BP |
2563 | cfm_process_heartbeat(ofport->cfm, packet); |
2564 | } | |
2565 | return true; | |
b6e001b6 EJ |
2566 | } else if (ofport->bundle && ofport->bundle->lacp |
2567 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2568 | if (packet) { | |
2569 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2570 | } |
da37ebac | 2571 | return true; |
21f7563c JP |
2572 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2573 | if (packet) { | |
2574 | stp_process_packet(ofport, packet); | |
2575 | } | |
2576 | return true; | |
abe529af BP |
2577 | } |
2578 | return false; | |
2579 | } | |
2580 | ||
501f8d1f | 2581 | static struct flow_miss * |
b23cdad9 | 2582 | flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) |
abe529af | 2583 | { |
501f8d1f | 2584 | struct flow_miss *miss; |
abe529af | 2585 | |
501f8d1f BP |
2586 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2587 | if (flow_equal(&miss->flow, flow)) { | |
2588 | return miss; | |
2589 | } | |
2590 | } | |
abe529af | 2591 | |
b23cdad9 | 2592 | return NULL; |
501f8d1f | 2593 | } |
abe529af | 2594 | |
9d6ac44e BP |
2595 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and |
2596 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
2597 | * 'miss' is associated with a subfacet the caller must also initialize the | |
2598 | * returned op->subfacet, and if anything needs to be freed after processing | |
2599 | * the op, the caller must initialize op->garbage also. */ | |
501f8d1f | 2600 | static void |
9d6ac44e BP |
2601 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, |
2602 | struct flow_miss_op *op) | |
501f8d1f | 2603 | { |
9d6ac44e BP |
2604 | if (miss->flow.vlan_tci != miss->initial_tci) { |
2605 | /* This packet was received on a VLAN splinter port. We | |
2606 | * added a VLAN to the packet to make the packet resemble | |
2607 | * the flow, but the actions were composed assuming that | |
2608 | * the packet contained no VLAN. So, we must remove the | |
2609 | * VLAN header from the packet before trying to execute the | |
2610 | * actions. */ | |
2611 | eth_pop_vlan(packet); | |
2612 | } | |
2613 | ||
2614 | op->subfacet = NULL; | |
2615 | op->garbage = NULL; | |
2616 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
2617 | op->dpif_op.u.execute.key = miss->key; | |
2618 | op->dpif_op.u.execute.key_len = miss->key_len; | |
2619 | op->dpif_op.u.execute.packet = packet; | |
2620 | } | |
2621 | ||
2622 | /* Helper for handle_flow_miss_without_facet() and | |
2623 | * handle_flow_miss_with_facet(). */ | |
2624 | static void | |
2625 | handle_flow_miss_common(struct rule_dpif *rule, | |
2626 | struct ofpbuf *packet, const struct flow *flow) | |
2627 | { | |
2628 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2629 | ||
2630 | ofproto->n_matches++; | |
2631 | ||
2632 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2633 | /* | |
2634 | * Extra-special case for fail-open mode. | |
2635 | * | |
2636 | * We are in fail-open mode and the packet matched the fail-open | |
2637 | * rule, but we are connected to a controller too. We should send | |
2638 | * the packet up to the controller in the hope that it will try to | |
2639 | * set up a flow and thereby allow us to exit fail-open. | |
2640 | * | |
2641 | * See the top-level comment in fail-open.c for more information. | |
2642 | */ | |
2643 | send_packet_in_miss(ofproto, packet, flow); | |
2644 | } | |
2645 | } | |
2646 | ||
2647 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
2648 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
2649 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
2650 | * true). However, for short flows the cost of bookkeeping is much higher than | |
2651 | * the benefits, so when the datapath holds a large number of flows we impose | |
2652 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
2653 | static bool | |
2654 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
2655 | struct flow_miss *miss, uint32_t hash) | |
2656 | { | |
2657 | if (!ofproto->governor) { | |
2658 | size_t n_subfacets; | |
2659 | ||
2660 | n_subfacets = hmap_count(&ofproto->subfacets); | |
2661 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
2662 | return true; | |
2663 | } | |
2664 | ||
2665 | ofproto->governor = governor_create(ofproto->up.name); | |
2666 | } | |
2667 | ||
2668 | return governor_should_install_flow(ofproto->governor, hash, | |
2669 | list_size(&miss->packets)); | |
2670 | } | |
2671 | ||
2672 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
2673 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
2674 | * increment '*n_ops'. */ | |
2675 | static void | |
2676 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
2677 | struct rule_dpif *rule, | |
2678 | struct flow_miss_op *ops, size_t *n_ops) | |
2679 | { | |
2680 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2681 | struct action_xlate_ctx ctx; | |
530a1d91 | 2682 | struct ofpbuf *packet; |
2b459b83 | 2683 | |
9d6ac44e BP |
2684 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2685 | struct flow_miss_op *op = &ops[*n_ops]; | |
2686 | struct dpif_flow_stats stats; | |
2687 | struct ofpbuf odp_actions; | |
abe529af | 2688 | |
9d6ac44e | 2689 | COVERAGE_INC(facet_suppress); |
501f8d1f | 2690 | |
9d6ac44e | 2691 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
501f8d1f | 2692 | |
9d6ac44e BP |
2693 | dpif_flow_stats_extract(&miss->flow, packet, &stats); |
2694 | rule_credit_stats(rule, &stats); | |
abe529af | 2695 | |
9d6ac44e BP |
2696 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, |
2697 | rule, 0, packet); | |
2698 | ctx.resubmit_stats = &stats; | |
2699 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, | |
2700 | &odp_actions); | |
abe529af | 2701 | |
9d6ac44e BP |
2702 | if (odp_actions.size) { |
2703 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
2704 | ||
2705 | init_flow_miss_execute_op(miss, packet, op); | |
2706 | execute->actions = odp_actions.data; | |
2707 | execute->actions_len = odp_actions.size; | |
2708 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2709 | ||
2710 | (*n_ops)++; | |
2711 | } else { | |
2712 | ofpbuf_uninit(&odp_actions); | |
2713 | } | |
abe529af | 2714 | } |
9d6ac44e BP |
2715 | } |
2716 | ||
2717 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
2718 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2719 | static void | |
2720 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
2721 | struct flow_miss_op *ops, size_t *n_ops) | |
2722 | { | |
2723 | struct subfacet *subfacet; | |
2724 | struct ofpbuf *packet; | |
abe529af | 2725 | |
15baa734 | 2726 | subfacet = subfacet_create(facet, |
e84173dc BP |
2727 | miss->key_fitness, miss->key, miss->key_len, |
2728 | miss->initial_tci); | |
b0f7b9b5 | 2729 | |
530a1d91 | 2730 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
5fe20d5d | 2731 | struct flow_miss_op *op = &ops[*n_ops]; |
67d91f78 | 2732 | struct dpif_flow_stats stats; |
5fe20d5d | 2733 | struct ofpbuf odp_actions; |
67d91f78 | 2734 | |
9d6ac44e | 2735 | handle_flow_miss_common(facet->rule, packet, &miss->flow); |
501f8d1f | 2736 | |
5fe20d5d | 2737 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); |
b95fc6ba | 2738 | if (!facet->may_install || !subfacet->actions) { |
5fe20d5d | 2739 | subfacet_make_actions(subfacet, packet, &odp_actions); |
501f8d1f | 2740 | } |
67d91f78 | 2741 | |
67d91f78 | 2742 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
15baa734 | 2743 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2744 | |
9d6ac44e BP |
2745 | if (subfacet->actions_len) { |
2746 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
8338659a | 2747 | |
9d6ac44e BP |
2748 | init_flow_miss_execute_op(miss, packet, op); |
2749 | op->subfacet = subfacet; | |
2750 | if (facet->may_install) { | |
2751 | execute->actions = subfacet->actions; | |
2752 | execute->actions_len = subfacet->actions_len; | |
2753 | ofpbuf_uninit(&odp_actions); | |
2754 | } else { | |
2755 | execute->actions = odp_actions.data; | |
2756 | execute->actions_len = odp_actions.size; | |
2757 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2758 | } | |
999fba59 | 2759 | |
9d6ac44e | 2760 | (*n_ops)++; |
5fe20d5d | 2761 | } else { |
9d6ac44e | 2762 | ofpbuf_uninit(&odp_actions); |
5fe20d5d | 2763 | } |
501f8d1f BP |
2764 | } |
2765 | ||
b0f7b9b5 | 2766 | if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) { |
501f8d1f | 2767 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2768 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2769 | |
b0f7b9b5 | 2770 | op->subfacet = subfacet; |
5fe20d5d | 2771 | op->garbage = NULL; |
c2b565b5 | 2772 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2773 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2774 | put->key = miss->key; | |
2775 | put->key_len = miss->key_len; | |
b95fc6ba BP |
2776 | put->actions = subfacet->actions; |
2777 | put->actions_len = subfacet->actions_len; | |
501f8d1f BP |
2778 | put->stats = NULL; |
2779 | } | |
2780 | } | |
2781 | ||
9d6ac44e BP |
2782 | /* Handles flow miss 'miss' on 'ofproto'. The flow does not match any flow in |
2783 | * the OpenFlow flow table. */ | |
2784 | static void | |
2785 | handle_flow_miss_no_rule(struct ofproto_dpif *ofproto, struct flow_miss *miss) | |
2786 | { | |
2787 | uint16_t in_port = miss->flow.in_port; | |
2788 | struct ofport_dpif *port = get_ofp_port(ofproto, in_port); | |
2789 | ||
2790 | if (!port) { | |
2791 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, in_port); | |
2792 | } | |
2793 | ||
2794 | if (port && port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
2795 | /* XXX install 'drop' flow entry */ | |
2796 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
2797 | } else { | |
2798 | const struct ofpbuf *packet; | |
2799 | ||
2800 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
2801 | send_packet_in_miss(ofproto, packet, &miss->flow); | |
2802 | } | |
2803 | } | |
2804 | } | |
2805 | ||
2806 | /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath | |
2807 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2808 | static void | |
2809 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2810 | struct flow_miss_op *ops, size_t *n_ops) | |
2811 | { | |
2812 | struct facet *facet; | |
2813 | uint32_t hash; | |
2814 | ||
2815 | /* The caller must ensure that miss->hmap_node.hash contains | |
2816 | * flow_hash(miss->flow, 0). */ | |
2817 | hash = miss->hmap_node.hash; | |
2818 | ||
2819 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
2820 | if (!facet) { | |
2821 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow, 0); | |
2822 | if (!rule) { | |
2823 | handle_flow_miss_no_rule(ofproto, miss); | |
2824 | return; | |
2825 | } else if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
2826 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); | |
2827 | return; | |
2828 | } | |
2829 | ||
2830 | facet = facet_create(rule, &miss->flow, hash); | |
2831 | } | |
2832 | handle_flow_miss_with_facet(miss, facet, ops, n_ops); | |
2833 | } | |
2834 | ||
e2a6ca36 BP |
2835 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
2836 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
2837 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
2838 | * what a flow key should contain. | |
2839 | * | |
2840 | * This function also includes some logic to help make VLAN splinters | |
2841 | * transparent to the rest of the upcall processing logic. In particular, if | |
2842 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
2843 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
2844 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
2845 | * | |
2846 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
2847 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
2848 | * (This differs from the value returned in flow->vlan_tci only for packets | |
2849 | * received on VLAN splinters.) | |
2850 | */ | |
e84173dc | 2851 | static enum odp_key_fitness |
52a90c29 | 2852 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 2853 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
2854 | struct flow *flow, ovs_be16 *initial_tci, |
2855 | struct ofpbuf *packet) | |
e84173dc BP |
2856 | { |
2857 | enum odp_key_fitness fitness; | |
52a90c29 BP |
2858 | uint16_t realdev; |
2859 | int vid; | |
e84173dc BP |
2860 | |
2861 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
2862 | if (fitness == ODP_FIT_ERROR) { | |
2863 | return fitness; | |
2864 | } | |
2865 | *initial_tci = flow->vlan_tci; | |
2866 | ||
52a90c29 BP |
2867 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); |
2868 | if (realdev) { | |
2869 | /* Cause the flow to be processed as if it came in on the real device | |
2870 | * with the VLAN device's VLAN ID. */ | |
2871 | flow->in_port = realdev; | |
2872 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
e2a6ca36 BP |
2873 | if (packet) { |
2874 | /* Make the packet resemble the flow, so that it gets sent to an | |
2875 | * OpenFlow controller properly, so that it looks correct for | |
2876 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
2877 | * if it is called on 'packet'. | |
2878 | * | |
2879 | * The allocated space inside 'packet' probably also contains | |
2880 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
2881 | * struct dpif_upcall (see the large comment on that structure | |
2882 | * definition), so pushing data on 'packet' is in general not a | |
2883 | * good idea since it could overwrite 'key' or free it as a side | |
2884 | * effect. However, it's OK in this special case because we know | |
2885 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
2886 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
2887 | * since we don't need that header anymore. */ | |
2888 | eth_push_vlan(packet, flow->vlan_tci); | |
2889 | } | |
52a90c29 BP |
2890 | |
2891 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
2892 | if (fitness == ODP_FIT_PERFECT) { | |
2893 | fitness = ODP_FIT_TOO_MUCH; | |
2894 | } | |
2895 | } | |
2896 | ||
e84173dc BP |
2897 | return fitness; |
2898 | } | |
2899 | ||
501f8d1f BP |
2900 | static void |
2901 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2902 | size_t n_upcalls) | |
2903 | { | |
2904 | struct dpif_upcall *upcall; | |
b23cdad9 BP |
2905 | struct flow_miss *miss; |
2906 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
501f8d1f | 2907 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; |
c2b565b5 | 2908 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f | 2909 | struct hmap todo; |
b23cdad9 | 2910 | int n_misses; |
501f8d1f BP |
2911 | size_t n_ops; |
2912 | size_t i; | |
2913 | ||
2914 | if (!n_upcalls) { | |
2915 | return; | |
2916 | } | |
2917 | ||
2918 | /* Construct the to-do list. | |
2919 | * | |
2920 | * This just amounts to extracting the flow from each packet and sticking | |
2921 | * the packets that have the same flow in the same "flow_miss" structure so | |
2922 | * that we can process them together. */ | |
2923 | hmap_init(&todo); | |
b23cdad9 | 2924 | n_misses = 0; |
501f8d1f | 2925 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { |
b23cdad9 BP |
2926 | struct flow_miss *miss = &misses[n_misses]; |
2927 | struct flow_miss *existing_miss; | |
2928 | uint32_t hash; | |
501f8d1f | 2929 | |
b0f7b9b5 BP |
2930 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
2931 | * then set 'flow''s header pointers. */ | |
b23cdad9 BP |
2932 | miss->key_fitness = ofproto_dpif_extract_flow_key( |
2933 | ofproto, upcall->key, upcall->key_len, | |
2934 | &miss->flow, &miss->initial_tci, upcall->packet); | |
2935 | if (miss->key_fitness == ODP_FIT_ERROR) { | |
b0f7b9b5 BP |
2936 | continue; |
2937 | } | |
b23cdad9 BP |
2938 | flow_extract(upcall->packet, miss->flow.skb_priority, |
2939 | miss->flow.tun_id, miss->flow.in_port, &miss->flow); | |
501f8d1f | 2940 | |
21f7563c | 2941 | /* Handle 802.1ag, LACP, and STP specially. */ |
b23cdad9 | 2942 | if (process_special(ofproto, &miss->flow, upcall->packet)) { |
6527c598 PS |
2943 | ofproto_update_local_port_stats(&ofproto->up, |
2944 | 0, upcall->packet->size); | |
501f8d1f BP |
2945 | ofproto->n_matches++; |
2946 | continue; | |
2947 | } | |
2948 | ||
2949 | /* Add other packets to a to-do list. */ | |
b23cdad9 BP |
2950 | hash = flow_hash(&miss->flow, 0); |
2951 | existing_miss = flow_miss_find(&todo, &miss->flow, hash); | |
2952 | if (!existing_miss) { | |
2953 | hmap_insert(&todo, &miss->hmap_node, hash); | |
2954 | miss->key = upcall->key; | |
2955 | miss->key_len = upcall->key_len; | |
2956 | list_init(&miss->packets); | |
2957 | ||
2958 | n_misses++; | |
2959 | } else { | |
2960 | miss = existing_miss; | |
2961 | } | |
501f8d1f BP |
2962 | list_push_back(&miss->packets, &upcall->packet->list_node); |
2963 | } | |
2964 | ||
2965 | /* Process each element in the to-do list, constructing the set of | |
2966 | * operations to batch. */ | |
2967 | n_ops = 0; | |
33bb0caa | 2968 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
501f8d1f | 2969 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); |
abe529af | 2970 | } |
501f8d1f | 2971 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
2972 | |
2973 | /* Execute batch. */ | |
2974 | for (i = 0; i < n_ops; i++) { | |
2975 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2976 | } | |
2977 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2978 | ||
2979 | /* Free memory and update facets. */ | |
2980 | for (i = 0; i < n_ops; i++) { | |
2981 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
501f8d1f BP |
2982 | |
2983 | switch (op->dpif_op.type) { | |
2984 | case DPIF_OP_EXECUTE: | |
501f8d1f | 2985 | break; |
abe529af | 2986 | |
501f8d1f | 2987 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 2988 | if (!op->dpif_op.error) { |
b0f7b9b5 | 2989 | op->subfacet->installed = true; |
501f8d1f BP |
2990 | } |
2991 | break; | |
b99d3cee BP |
2992 | |
2993 | case DPIF_OP_FLOW_DEL: | |
2994 | NOT_REACHED(); | |
501f8d1f | 2995 | } |
5fe20d5d BP |
2996 | |
2997 | free(op->garbage); | |
501f8d1f | 2998 | } |
33bb0caa | 2999 | hmap_destroy(&todo); |
abe529af BP |
3000 | } |
3001 | ||
3002 | static void | |
6ff686f2 PS |
3003 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
3004 | struct dpif_upcall *upcall) | |
abe529af | 3005 | { |
6ff686f2 | 3006 | struct user_action_cookie cookie; |
e84173dc BP |
3007 | enum odp_key_fitness fitness; |
3008 | ovs_be16 initial_tci; | |
3009 | struct flow flow; | |
abe529af | 3010 | |
6ff686f2 | 3011 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 3012 | |
e84173dc BP |
3013 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
3014 | upcall->key_len, &flow, | |
e2a6ca36 | 3015 | &initial_tci, upcall->packet); |
e84173dc BP |
3016 | if (fitness == ODP_FIT_ERROR) { |
3017 | return; | |
3018 | } | |
3019 | ||
6ff686f2 | 3020 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af | 3021 | if (ofproto->sflow) { |
e84173dc BP |
3022 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
3023 | &cookie); | |
abe529af | 3024 | } |
6ff686f2 PS |
3025 | } else { |
3026 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
3027 | } | |
3028 | } | |
3029 | ||
9b16c439 BP |
3030 | static int |
3031 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 3032 | { |
9b16c439 | 3033 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
3034 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
3035 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3036 | int n_processed; | |
9b16c439 BP |
3037 | int n_misses; |
3038 | int i; | |
abe529af | 3039 | |
90a7c55e | 3040 | assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 3041 | |
90a7c55e | 3042 | n_processed = 0; |
9b16c439 | 3043 | n_misses = 0; |
90a7c55e | 3044 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 3045 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 3046 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
3047 | int error; |
3048 | ||
90a7c55e BP |
3049 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
3050 | sizeof miss_buf_stubs[n_misses]); | |
3051 | error = dpif_recv(ofproto->dpif, upcall, buf); | |
9b16c439 | 3052 | if (error) { |
90a7c55e | 3053 | ofpbuf_uninit(buf); |
9b16c439 BP |
3054 | break; |
3055 | } | |
3056 | ||
3057 | switch (upcall->type) { | |
3058 | case DPIF_UC_ACTION: | |
3059 | handle_userspace_upcall(ofproto, upcall); | |
90a7c55e | 3060 | ofpbuf_uninit(buf); |
9b16c439 BP |
3061 | break; |
3062 | ||
3063 | case DPIF_UC_MISS: | |
3064 | /* Handle it later. */ | |
3065 | n_misses++; | |
3066 | break; | |
3067 | ||
3068 | case DPIF_N_UC_TYPES: | |
3069 | default: | |
3070 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, | |
3071 | upcall->type); | |
3072 | break; | |
3073 | } | |
abe529af | 3074 | } |
9b16c439 BP |
3075 | |
3076 | handle_miss_upcalls(ofproto, misses, n_misses); | |
90a7c55e BP |
3077 | for (i = 0; i < n_misses; i++) { |
3078 | ofpbuf_uninit(&miss_bufs[i]); | |
3079 | } | |
9b16c439 | 3080 | |
90a7c55e | 3081 | return n_processed; |
abe529af BP |
3082 | } |
3083 | \f | |
3084 | /* Flow expiration. */ | |
3085 | ||
b0f7b9b5 | 3086 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
3087 | static void update_stats(struct ofproto_dpif *); |
3088 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 3089 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
3090 | |
3091 | /* This function is called periodically by run(). Its job is to collect | |
3092 | * updates for the flows that have been installed into the datapath, most | |
3093 | * importantly when they last were used, and then use that information to | |
3094 | * expire flows that have not been used recently. | |
3095 | * | |
3096 | * Returns the number of milliseconds after which it should be called again. */ | |
3097 | static int | |
3098 | expire(struct ofproto_dpif *ofproto) | |
3099 | { | |
3100 | struct rule_dpif *rule, *next_rule; | |
d0918789 | 3101 | struct oftable *table; |
abe529af BP |
3102 | int dp_max_idle; |
3103 | ||
3104 | /* Update stats for each flow in the datapath. */ | |
3105 | update_stats(ofproto); | |
3106 | ||
b0f7b9b5 BP |
3107 | /* Expire subfacets that have been idle too long. */ |
3108 | dp_max_idle = subfacet_max_idle(ofproto); | |
3109 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
3110 | |
3111 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
3112 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
3113 | struct cls_cursor cursor; | |
3114 | ||
d0918789 | 3115 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
3116 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
3117 | rule_expire(rule); | |
3118 | } | |
abe529af BP |
3119 | } |
3120 | ||
3121 | /* All outstanding data in existing flows has been accounted, so it's a | |
3122 | * good time to do bond rebalancing. */ | |
3123 | if (ofproto->has_bonded_bundles) { | |
3124 | struct ofbundle *bundle; | |
3125 | ||
3126 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3127 | if (bundle->bond) { | |
3128 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
3129 | } | |
3130 | } | |
3131 | } | |
3132 | ||
3133 | return MIN(dp_max_idle, 1000); | |
3134 | } | |
3135 | ||
3136 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
3137 | * | |
3138 | * This function also pushes statistics updates to rules which each facet | |
3139 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3140 | * facet changes the rule it resubmits into at some time in between | |
3141 | * update_stats() runs, it is possible that statistics accrued to the | |
3142 | * old rule will be incorrectly attributed to the new rule. This could be | |
3143 | * avoided by calling update_stats() whenever rules are created or | |
3144 | * deleted. However, the performance impact of making so many calls to the | |
3145 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
3146 | */ | |
3147 | static void | |
3148 | update_stats(struct ofproto_dpif *p) | |
3149 | { | |
3150 | const struct dpif_flow_stats *stats; | |
3151 | struct dpif_flow_dump dump; | |
3152 | const struct nlattr *key; | |
3153 | size_t key_len; | |
3154 | ||
3155 | dpif_flow_dump_start(&dump, p->dpif); | |
3156 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 3157 | struct subfacet *subfacet; |
abe529af | 3158 | |
6a542738 | 3159 | subfacet = subfacet_find(p, key, key_len); |
b0f7b9b5 BP |
3160 | if (subfacet && subfacet->installed) { |
3161 | struct facet *facet = subfacet->facet; | |
abe529af | 3162 | |
b0f7b9b5 BP |
3163 | if (stats->n_packets >= subfacet->dp_packet_count) { |
3164 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
abe529af BP |
3165 | facet->packet_count += extra; |
3166 | } else { | |
3167 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
3168 | } | |
3169 | ||
b0f7b9b5 BP |
3170 | if (stats->n_bytes >= subfacet->dp_byte_count) { |
3171 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
abe529af BP |
3172 | } else { |
3173 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
3174 | } | |
3175 | ||
b0f7b9b5 BP |
3176 | subfacet->dp_packet_count = stats->n_packets; |
3177 | subfacet->dp_byte_count = stats->n_bytes; | |
abe529af | 3178 | |
0e553d9c BP |
3179 | facet->tcp_flags |= stats->tcp_flags; |
3180 | ||
15baa734 | 3181 | subfacet_update_time(subfacet, stats->used); |
3de9590b BP |
3182 | if (facet->accounted_bytes < facet->byte_count) { |
3183 | facet_learn(facet); | |
3184 | facet_account(facet); | |
3185 | facet->accounted_bytes = facet->byte_count; | |
3186 | } | |
abe529af BP |
3187 | facet_push_stats(facet); |
3188 | } else { | |
6a542738 PS |
3189 | if (!VLOG_DROP_WARN(&rl)) { |
3190 | struct ds s; | |
3191 | ||
3192 | ds_init(&s); | |
3193 | odp_flow_key_format(key, key_len, &s); | |
3194 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
3195 | ds_destroy(&s); | |
3196 | } | |
3197 | ||
3198 | COVERAGE_INC(facet_unexpected); | |
b0f7b9b5 BP |
3199 | /* There's a flow in the datapath that we know nothing about, or a |
3200 | * flow that shouldn't be installed but was anyway. Delete it. */ | |
abe529af BP |
3201 | dpif_flow_del(p->dpif, key, key_len, NULL); |
3202 | } | |
3203 | } | |
3204 | dpif_flow_dump_done(&dump); | |
3205 | } | |
3206 | ||
3207 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3208 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3209 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3210 | * fold its statistic into its rule. */ | |
abe529af | 3211 | static int |
b0f7b9b5 | 3212 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
3213 | { |
3214 | /* | |
3215 | * Idle time histogram. | |
3216 | * | |
b0f7b9b5 BP |
3217 | * Most of the time a switch has a relatively small number of subfacets. |
3218 | * When this is the case we might as well keep statistics for all of them | |
3219 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3220 | * well. |
3221 | * | |
b0f7b9b5 | 3222 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3223 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3224 | * significant. However, with a large number of subfacets it is likely |
3225 | * that only a few of them are "heavy hitters" that consume a large amount | |
3226 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3227 | * the kernel and maintaining in userspaces; other subfacets we can | |
3228 | * discard. | |
abe529af BP |
3229 | * |
3230 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3231 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3232 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3233 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3234 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 3235 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3236 | * the most-recently-used bucket of subfacets is kept, so actually an |
3237 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3238 | * (though the next run will delete most of those unless they receive |
3239 | * additional data). | |
abe529af | 3240 | * |
b0f7b9b5 BP |
3241 | * This requires a second pass through the subfacets, in addition to the |
3242 | * pass made by update_stats(), because the former function never looks at | |
3243 | * uninstallable subfacets. | |
abe529af BP |
3244 | */ |
3245 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3246 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3247 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3248 | int total, subtotal, bucket; |
b0f7b9b5 | 3249 | struct subfacet *subfacet; |
abe529af BP |
3250 | long long int now; |
3251 | int i; | |
3252 | ||
b0f7b9b5 | 3253 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3254 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3255 | return N_BUCKETS * BUCKET_WIDTH; |
3256 | } | |
3257 | ||
3258 | /* Build histogram. */ | |
3259 | now = time_msec(); | |
b0f7b9b5 BP |
3260 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3261 | long long int idle = now - subfacet->used; | |
abe529af BP |
3262 | int bucket = (idle <= 0 ? 0 |
3263 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3264 | : (unsigned int) idle / BUCKET_WIDTH); | |
3265 | buckets[bucket]++; | |
3266 | } | |
3267 | ||
3268 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3269 | subtotal = bucket = 0; |
3270 | do { | |
3271 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3272 | } while (bucket < N_BUCKETS && |
3273 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3274 | |
3275 | if (VLOG_IS_DBG_ENABLED()) { | |
3276 | struct ds s; | |
3277 | ||
3278 | ds_init(&s); | |
3279 | ds_put_cstr(&s, "keep"); | |
3280 | for (i = 0; i < N_BUCKETS; i++) { | |
3281 | if (i == bucket) { | |
3282 | ds_put_cstr(&s, ", drop"); | |
3283 | } | |
3284 | if (buckets[i]) { | |
3285 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3286 | } | |
3287 | } | |
3288 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3289 | ds_destroy(&s); | |
3290 | } | |
3291 | ||
3292 | return bucket * BUCKET_WIDTH; | |
3293 | } | |
3294 | ||
b99d3cee BP |
3295 | enum { EXPIRE_MAX_BATCH = 50 }; |
3296 | ||
3297 | static void | |
3298 | expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) | |
3299 | { | |
3300 | struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; | |
3301 | struct dpif_op ops[EXPIRE_MAX_BATCH]; | |
3302 | struct dpif_op *opsp[EXPIRE_MAX_BATCH]; | |
3303 | struct ofpbuf keys[EXPIRE_MAX_BATCH]; | |
3304 | struct dpif_flow_stats stats[EXPIRE_MAX_BATCH]; | |
3305 | int i; | |
3306 | ||
3307 | for (i = 0; i < n; i++) { | |
3308 | ops[i].type = DPIF_OP_FLOW_DEL; | |
3309 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
3310 | ops[i].u.flow_del.key = keys[i].data; | |
3311 | ops[i].u.flow_del.key_len = keys[i].size; | |
3312 | ops[i].u.flow_del.stats = &stats[i]; | |
3313 | opsp[i] = &ops[i]; | |
3314 | } | |
3315 | ||
3316 | dpif_operate(ofproto->dpif, opsp, n); | |
3317 | for (i = 0; i < n; i++) { | |
3318 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
3319 | subfacets[i]->installed = false; | |
3320 | subfacet_destroy(subfacets[i]); | |
3321 | } | |
3322 | } | |
3323 | ||
abe529af | 3324 | static void |
b0f7b9b5 | 3325 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af BP |
3326 | { |
3327 | long long int cutoff = time_msec() - dp_max_idle; | |
b99d3cee | 3328 | |
b0f7b9b5 | 3329 | struct subfacet *subfacet, *next_subfacet; |
b99d3cee BP |
3330 | struct subfacet *batch[EXPIRE_MAX_BATCH]; |
3331 | int n_batch; | |
abe529af | 3332 | |
b99d3cee | 3333 | n_batch = 0; |
b0f7b9b5 BP |
3334 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3335 | &ofproto->subfacets) { | |
3336 | if (subfacet->used < cutoff) { | |
b99d3cee BP |
3337 | if (subfacet->installed) { |
3338 | batch[n_batch++] = subfacet; | |
3339 | if (n_batch >= EXPIRE_MAX_BATCH) { | |
3340 | expire_batch(ofproto, batch, n_batch); | |
3341 | n_batch = 0; | |
3342 | } | |
3343 | } else { | |
3344 | subfacet_destroy(subfacet); | |
3345 | } | |
abe529af BP |
3346 | } |
3347 | } | |
b99d3cee BP |
3348 | |
3349 | if (n_batch > 0) { | |
3350 | expire_batch(ofproto, batch, n_batch); | |
3351 | } | |
abe529af BP |
3352 | } |
3353 | ||
3354 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3355 | * then delete it entirely. */ | |
3356 | static void | |
3357 | rule_expire(struct rule_dpif *rule) | |
3358 | { | |
abe529af BP |
3359 | struct facet *facet, *next_facet; |
3360 | long long int now; | |
3361 | uint8_t reason; | |
3362 | ||
3363 | /* Has 'rule' expired? */ | |
3364 | now = time_msec(); | |
3365 | if (rule->up.hard_timeout | |
308881af | 3366 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 3367 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 3368 | } else if (rule->up.idle_timeout |
1745cd08 | 3369 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
3370 | reason = OFPRR_IDLE_TIMEOUT; |
3371 | } else { | |
3372 | return; | |
3373 | } | |
3374 | ||
3375 | COVERAGE_INC(ofproto_dpif_expired); | |
3376 | ||
3377 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3378 | * timeout, because that only happens when the rule has no facets left.) */ | |
3379 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3380 | facet_remove(facet); |
abe529af BP |
3381 | } |
3382 | ||
3383 | /* Get rid of the rule. */ | |
3384 | ofproto_rule_expire(&rule->up, reason); | |
3385 | } | |
3386 | \f | |
3387 | /* Facets. */ | |
3388 | ||
f3827897 | 3389 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3390 | * |
3391 | * The caller must already have determined that no facet with an identical | |
3392 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3393 | * the ofproto's classifier table. |
3394 | * | |
2b459b83 BP |
3395 | * 'hash' must be the return value of flow_hash(flow, 0). |
3396 | * | |
b0f7b9b5 BP |
3397 | * The facet will initially have no subfacets. The caller should create (at |
3398 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3399 | static struct facet * |
2b459b83 | 3400 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
3401 | { |
3402 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3403 | struct facet *facet; | |
3404 | ||
3405 | facet = xzalloc(sizeof *facet); | |
3406 | facet->used = time_msec(); | |
2b459b83 | 3407 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
3408 | list_push_back(&rule->facets, &facet->list_node); |
3409 | facet->rule = rule; | |
3410 | facet->flow = *flow; | |
b0f7b9b5 | 3411 | list_init(&facet->subfacets); |
abe529af BP |
3412 | netflow_flow_init(&facet->nf_flow); |
3413 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3414 | ||
abe529af BP |
3415 | return facet; |
3416 | } | |
3417 | ||
3418 | static void | |
3419 | facet_free(struct facet *facet) | |
3420 | { | |
abe529af BP |
3421 | free(facet); |
3422 | } | |
3423 | ||
3d9e05f8 BP |
3424 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3425 | * 'packet', which arrived on 'in_port'. | |
3426 | * | |
3427 | * Takes ownership of 'packet'. */ | |
3428 | static bool | |
3429 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3430 | const struct nlattr *odp_actions, size_t actions_len, | |
3431 | struct ofpbuf *packet) | |
3432 | { | |
3433 | struct odputil_keybuf keybuf; | |
3434 | struct ofpbuf key; | |
3435 | int error; | |
3436 | ||
6ff686f2 PS |
3437 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3438 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3439 | |
6ff686f2 PS |
3440 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3441 | odp_actions, actions_len, packet); | |
80e5eed9 | 3442 | |
6ff686f2 PS |
3443 | ofpbuf_delete(packet); |
3444 | return !error; | |
abe529af BP |
3445 | } |
3446 | ||
abe529af BP |
3447 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3448 | * | |
3449 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3450 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3451 | * |
3452 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3453 | */ | |
3454 | static void | |
15baa734 | 3455 | facet_remove(struct facet *facet) |
abe529af | 3456 | { |
15baa734 | 3457 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3458 | struct subfacet *subfacet, *next_subfacet; |
3459 | ||
551a2f6c BP |
3460 | assert(!list_is_empty(&facet->subfacets)); |
3461 | ||
3462 | /* First uninstall all of the subfacets to get final statistics. */ | |
3463 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3464 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3465 | } |
3466 | ||
3467 | /* Flush the final stats to the rule. | |
3468 | * | |
3469 | * This might require us to have at least one subfacet around so that we | |
3470 | * can use its actions for accounting in facet_account(), which is why we | |
3471 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3472 | facet_flush_stats(facet); |
551a2f6c BP |
3473 | |
3474 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3475 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3476 | &facet->subfacets) { | |
15baa734 | 3477 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3478 | } |
abe529af BP |
3479 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3480 | list_remove(&facet->list_node); | |
3481 | facet_free(facet); | |
3482 | } | |
3483 | ||
3de9590b BP |
3484 | /* Feed information from 'facet' back into the learning table to keep it in |
3485 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3486 | static void |
3de9590b | 3487 | facet_learn(struct facet *facet) |
abe529af | 3488 | { |
15baa734 | 3489 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3de9590b | 3490 | struct action_xlate_ctx ctx; |
abe529af | 3491 | |
3de9590b BP |
3492 | if (!facet->has_learn |
3493 | && !facet->has_normal | |
3494 | && (!facet->has_fin_timeout | |
3495 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
3496 | return; |
3497 | } | |
abe529af | 3498 | |
3de9590b BP |
3499 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
3500 | facet->flow.vlan_tci, | |
3501 | facet->rule, facet->tcp_flags, NULL); | |
3502 | ctx.may_learn = true; | |
050ac423 BP |
3503 | xlate_actions_for_side_effects(&ctx, facet->rule->up.actions, |
3504 | facet->rule->up.n_actions); | |
3de9590b BP |
3505 | } |
3506 | ||
3507 | static void | |
3508 | facet_account(struct facet *facet) | |
3509 | { | |
3510 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3511 | struct subfacet *subfacet; | |
3512 | const struct nlattr *a; | |
3513 | unsigned int left; | |
3514 | ovs_be16 vlan_tci; | |
3515 | uint64_t n_bytes; | |
abe529af | 3516 | |
75a75043 | 3517 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3518 | return; |
3519 | } | |
3de9590b | 3520 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3521 | |
3522 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3523 | * as a basis. We also need to track the actual VLAN on which the packet | |
3524 | * is going to be sent to ensure that it matches the one passed to | |
3525 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3526 | * hash bucket.) |
3527 | * | |
3528 | * We use the actions from an arbitrary subfacet because they should all | |
3529 | * be equally valid for our purpose. */ | |
3530 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3531 | struct subfacet, list_node); | |
d78be13b | 3532 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3533 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3534 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3535 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3536 | struct ofport_dpif *port; |
abe529af | 3537 | |
d78be13b | 3538 | switch (nl_attr_type(a)) { |
df2c07f4 | 3539 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3540 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3541 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3542 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3543 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3544 | } |
d78be13b BP |
3545 | break; |
3546 | ||
fea393b1 BP |
3547 | case OVS_ACTION_ATTR_POP_VLAN: |
3548 | vlan_tci = htons(0); | |
d78be13b BP |
3549 | break; |
3550 | ||
fea393b1 BP |
3551 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3552 | vlan = nl_attr_get(a); | |
3553 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3554 | break; |
abe529af BP |
3555 | } |
3556 | } | |
3557 | } | |
3558 | ||
abe529af BP |
3559 | /* Returns true if the only action for 'facet' is to send to the controller. |
3560 | * (We don't report NetFlow expiration messages for such facets because they | |
3561 | * are just part of the control logic for the network, not real traffic). */ | |
3562 | static bool | |
3563 | facet_is_controller_flow(struct facet *facet) | |
3564 | { | |
3565 | return (facet | |
3566 | && facet->rule->up.n_actions == 1 | |
3567 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
3568 | htons(OFPP_CONTROLLER))); | |
3569 | } | |
3570 | ||
3571 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3572 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3573 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3574 | * its packet and byte counts before this function is called. */ | |
3575 | static void | |
15baa734 | 3576 | facet_flush_stats(struct facet *facet) |
abe529af | 3577 | { |
15baa734 | 3578 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3579 | struct subfacet *subfacet; |
3580 | ||
3581 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3582 | assert(!subfacet->dp_byte_count); | |
3583 | assert(!subfacet->dp_packet_count); | |
3584 | } | |
abe529af BP |
3585 | |
3586 | facet_push_stats(facet); | |
3de9590b BP |
3587 | if (facet->accounted_bytes < facet->byte_count) { |
3588 | facet_account(facet); | |
3589 | facet->accounted_bytes = facet->byte_count; | |
3590 | } | |
abe529af BP |
3591 | |
3592 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3593 | struct ofexpired expired; | |
3594 | expired.flow = facet->flow; | |
3595 | expired.packet_count = facet->packet_count; | |
3596 | expired.byte_count = facet->byte_count; | |
3597 | expired.used = facet->used; | |
3598 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3599 | } | |
3600 | ||
3601 | facet->rule->packet_count += facet->packet_count; | |
3602 | facet->rule->byte_count += facet->byte_count; | |
3603 | ||
3604 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3605 | * reinstalled. */ | |
bbb5d219 | 3606 | facet_reset_counters(facet); |
abe529af BP |
3607 | |
3608 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 3609 | facet->tcp_flags = 0; |
abe529af BP |
3610 | } |
3611 | ||
3612 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3613 | * Returns it if found, otherwise a null pointer. | |
3614 | * | |
2b459b83 BP |
3615 | * 'hash' must be the return value of flow_hash(flow, 0). |
3616 | * | |
abe529af BP |
3617 | * The returned facet might need revalidation; use facet_lookup_valid() |
3618 | * instead if that is important. */ | |
3619 | static struct facet * | |
2b459b83 BP |
3620 | facet_find(struct ofproto_dpif *ofproto, |
3621 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
3622 | { |
3623 | struct facet *facet; | |
3624 | ||
2b459b83 | 3625 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
3626 | if (flow_equal(flow, &facet->flow)) { |
3627 | return facet; | |
3628 | } | |
3629 | } | |
3630 | ||
3631 | return NULL; | |
3632 | } | |
3633 | ||
3634 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3635 | * Returns it if found, otherwise a null pointer. | |
3636 | * | |
2b459b83 BP |
3637 | * 'hash' must be the return value of flow_hash(flow, 0). |
3638 | * | |
abe529af BP |
3639 | * The returned facet is guaranteed to be valid. */ |
3640 | static struct facet * | |
2b459b83 BP |
3641 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
3642 | uint32_t hash) | |
abe529af | 3643 | { |
2b459b83 | 3644 | struct facet *facet = facet_find(ofproto, flow, hash); |
abe529af BP |
3645 | |
3646 | /* The facet we found might not be valid, since we could be in need of | |
3647 | * revalidation. If it is not valid, don't return it. */ | |
3648 | if (facet | |
0e4b3771 BP |
3649 | && (ofproto->need_revalidate |
3650 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
15baa734 | 3651 | && !facet_revalidate(facet)) { |
abe529af BP |
3652 | COVERAGE_INC(facet_invalidated); |
3653 | return NULL; | |
3654 | } | |
3655 | ||
3656 | return facet; | |
3657 | } | |
3658 | ||
6814e51f BP |
3659 | static bool |
3660 | facet_check_consistency(struct facet *facet) | |
3661 | { | |
3662 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3663 | ||
3664 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3665 | ||
050ac423 BP |
3666 | uint64_t odp_actions_stub[1024 / 8]; |
3667 | struct ofpbuf odp_actions; | |
3668 | ||
6814e51f BP |
3669 | struct rule_dpif *rule; |
3670 | struct subfacet *subfacet; | |
c53e1132 | 3671 | bool may_log = false; |
6814e51f BP |
3672 | bool ok; |
3673 | ||
3674 | /* Check the rule for consistency. */ | |
3675 | rule = rule_dpif_lookup(ofproto, &facet->flow, 0); | |
3676 | if (!rule) { | |
3677 | if (!VLOG_DROP_WARN(&rl)) { | |
3678 | char *s = flow_to_string(&facet->flow); | |
3679 | VLOG_WARN("%s: facet should not exist", s); | |
3680 | free(s); | |
3681 | } | |
3682 | return false; | |
3683 | } else if (rule != facet->rule) { | |
c53e1132 BP |
3684 | may_log = !VLOG_DROP_WARN(&rl); |
3685 | ok = false; | |
3686 | if (may_log) { | |
3687 | struct ds s; | |
6814e51f | 3688 | |
c53e1132 BP |
3689 | ds_init(&s); |
3690 | flow_format(&s, &facet->flow); | |
3691 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3692 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3693 | cls_rule_format(&facet->rule->up.cr, &s); | |
3694 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3695 | rule->up.table_id); | |
3696 | cls_rule_format(&rule->up.cr, &s); | |
3697 | ds_put_char(&s, ')'); | |
6814e51f | 3698 | |
c53e1132 BP |
3699 | VLOG_WARN("%s", ds_cstr(&s)); |
3700 | ds_destroy(&s); | |
3701 | } | |
6814e51f BP |
3702 | } else { |
3703 | ok = true; | |
3704 | } | |
3705 | ||
3706 | /* Check the datapath actions for consistency. */ | |
050ac423 | 3707 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f BP |
3708 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
3709 | struct action_xlate_ctx ctx; | |
6814e51f BP |
3710 | bool actions_changed; |
3711 | bool should_install; | |
3712 | ||
3713 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
0e553d9c | 3714 | subfacet->initial_tci, rule, 0, NULL); |
050ac423 BP |
3715 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, |
3716 | &odp_actions); | |
6814e51f BP |
3717 | |
3718 | should_install = (ctx.may_set_up_flow | |
3719 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
3720 | if (!should_install && !subfacet->installed) { | |
3721 | /* The actions for uninstallable flows may vary from one packet to | |
3722 | * the next, so don't compare the actions. */ | |
050ac423 | 3723 | continue; |
6814e51f BP |
3724 | } |
3725 | ||
050ac423 BP |
3726 | actions_changed = (subfacet->actions_len != odp_actions.size |
3727 | || memcmp(subfacet->actions, odp_actions.data, | |
6814e51f BP |
3728 | subfacet->actions_len)); |
3729 | if (should_install != subfacet->installed || actions_changed) { | |
c53e1132 BP |
3730 | if (ok) { |
3731 | may_log = !VLOG_DROP_WARN(&rl); | |
3732 | ok = false; | |
3733 | } | |
6814e51f | 3734 | |
c53e1132 BP |
3735 | if (may_log) { |
3736 | struct odputil_keybuf keybuf; | |
3737 | struct ofpbuf key; | |
3738 | struct ds s; | |
6814e51f | 3739 | |
c53e1132 BP |
3740 | ds_init(&s); |
3741 | subfacet_get_key(subfacet, &keybuf, &key); | |
3742 | odp_flow_key_format(key.data, key.size, &s); | |
3743 | ||
3744 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
3745 | if (should_install != subfacet->installed) { | |
3746 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
3747 | ||
3748 | ds_put_format(&s, " (should%s have been installed)", | |
3749 | should_install ? "" : " not"); | |
3750 | ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)", | |
3751 | ctx.may_set_up_flow ? "true" : "false", | |
3752 | odp_key_fitness_to_string(fitness)); | |
3753 | } | |
3754 | if (actions_changed) { | |
3755 | ds_put_cstr(&s, " (actions were: "); | |
3756 | format_odp_actions(&s, subfacet->actions, | |
3757 | subfacet->actions_len); | |
3758 | ds_put_cstr(&s, ") (correct actions: "); | |
050ac423 | 3759 | format_odp_actions(&s, odp_actions.data, odp_actions.size); |
c53e1132 BP |
3760 | ds_put_char(&s, ')'); |
3761 | } else { | |
3762 | ds_put_cstr(&s, " (actions: "); | |
3763 | format_odp_actions(&s, subfacet->actions, | |
3764 | subfacet->actions_len); | |
3765 | ds_put_char(&s, ')'); | |
3766 | } | |
3767 | VLOG_WARN("%s", ds_cstr(&s)); | |
3768 | ds_destroy(&s); | |
6814e51f | 3769 | } |
6814e51f | 3770 | } |
6814e51f | 3771 | } |
050ac423 | 3772 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
3773 | |
3774 | return ok; | |
3775 | } | |
3776 | ||
15baa734 | 3777 | /* Re-searches the classifier for 'facet': |
abe529af BP |
3778 | * |
3779 | * - If the rule found is different from 'facet''s current rule, moves | |
3780 | * 'facet' to the new rule and recompiles its actions. | |
3781 | * | |
3782 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3783 | * where it is and recompiles its actions anyway. | |
3784 | * | |
3785 | * - If there is none, destroys 'facet'. | |
3786 | * | |
3787 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3788 | static bool | |
15baa734 | 3789 | facet_revalidate(struct facet *facet) |
abe529af | 3790 | { |
15baa734 | 3791 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
3792 | struct actions { |
3793 | struct nlattr *odp_actions; | |
3794 | size_t actions_len; | |
3795 | }; | |
3796 | struct actions *new_actions; | |
3797 | ||
abe529af | 3798 | struct action_xlate_ctx ctx; |
050ac423 BP |
3799 | uint64_t odp_actions_stub[1024 / 8]; |
3800 | struct ofpbuf odp_actions; | |
3801 | ||
abe529af | 3802 | struct rule_dpif *new_rule; |
b0f7b9b5 | 3803 | struct subfacet *subfacet; |
abe529af | 3804 | bool actions_changed; |
b95fc6ba | 3805 | int i; |
abe529af BP |
3806 | |
3807 | COVERAGE_INC(facet_revalidate); | |
3808 | ||
3809 | /* Determine the new rule. */ | |
29901626 | 3810 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3811 | if (!new_rule) { |
3812 | /* No new rule, so delete the facet. */ | |
15baa734 | 3813 | facet_remove(facet); |
abe529af BP |
3814 | return false; |
3815 | } | |
3816 | ||
df2c07f4 | 3817 | /* Calculate new datapath actions. |
abe529af BP |
3818 | * |
3819 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3820 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3821 | * around to properly compose it. */ | |
abe529af | 3822 | |
df2c07f4 JP |
3823 | /* If the datapath actions changed or the installability changed, |
3824 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
3825 | i = 0; |
3826 | new_actions = NULL; | |
3827 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 3828 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 3829 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
b95fc6ba BP |
3830 | bool should_install; |
3831 | ||
e84173dc | 3832 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
0e553d9c | 3833 | subfacet->initial_tci, new_rule, 0, NULL); |
050ac423 BP |
3834 | xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions, |
3835 | &odp_actions); | |
3836 | actions_changed = (subfacet->actions_len != odp_actions.size | |
3837 | || memcmp(subfacet->actions, odp_actions.data, | |
b95fc6ba BP |
3838 | subfacet->actions_len)); |
3839 | ||
3840 | should_install = (ctx.may_set_up_flow | |
3841 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
b0f7b9b5 BP |
3842 | if (actions_changed || should_install != subfacet->installed) { |
3843 | if (should_install) { | |
3844 | struct dpif_flow_stats stats; | |
3845 | ||
15baa734 | 3846 | subfacet_install(subfacet, |
050ac423 | 3847 | odp_actions.data, odp_actions.size, &stats); |
15baa734 | 3848 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 3849 | } else { |
15baa734 | 3850 | subfacet_uninstall(subfacet); |
b0f7b9b5 | 3851 | } |
b95fc6ba BP |
3852 | |
3853 | if (!new_actions) { | |
3854 | new_actions = xcalloc(list_size(&facet->subfacets), | |
3855 | sizeof *new_actions); | |
3856 | } | |
050ac423 BP |
3857 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
3858 | odp_actions.size); | |
3859 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 3860 | } |
b95fc6ba | 3861 | |
b95fc6ba | 3862 | i++; |
b0f7b9b5 | 3863 | } |
050ac423 BP |
3864 | ofpbuf_uninit(&odp_actions); |
3865 | ||
b95fc6ba | 3866 | if (new_actions) { |
15baa734 | 3867 | facet_flush_stats(facet); |
abe529af BP |
3868 | } |
3869 | ||
3870 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3871 | facet->tags = ctx.tags; | |
3872 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3873 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3874 | facet->has_learn = ctx.has_learn; |
3875 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 3876 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 3877 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3878 | if (new_actions) { |
3879 | i = 0; | |
3880 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3881 | if (new_actions[i].odp_actions) { | |
3882 | free(subfacet->actions); | |
3883 | subfacet->actions = new_actions[i].odp_actions; | |
3884 | subfacet->actions_len = new_actions[i].actions_len; | |
3885 | } | |
3886 | i++; | |
3887 | } | |
3888 | free(new_actions); | |
abe529af BP |
3889 | } |
3890 | if (facet->rule != new_rule) { | |
3891 | COVERAGE_INC(facet_changed_rule); | |
3892 | list_remove(&facet->list_node); | |
3893 | list_push_back(&new_rule->facets, &facet->list_node); | |
3894 | facet->rule = new_rule; | |
3895 | facet->used = new_rule->up.created; | |
9d24de3b | 3896 | facet->prev_used = facet->used; |
abe529af BP |
3897 | } |
3898 | ||
abe529af BP |
3899 | return true; |
3900 | } | |
3901 | ||
3902 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3903 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3904 | static void | |
15baa734 | 3905 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 3906 | { |
15baa734 | 3907 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
3908 | if (used > facet->used) { |
3909 | facet->used = used; | |
1745cd08 | 3910 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
3911 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
3912 | } | |
3913 | } | |
3914 | ||
bbb5d219 EJ |
3915 | static void |
3916 | facet_reset_counters(struct facet *facet) | |
3917 | { | |
3918 | facet->packet_count = 0; | |
3919 | facet->byte_count = 0; | |
9d24de3b JP |
3920 | facet->prev_packet_count = 0; |
3921 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
3922 | facet->accounted_bytes = 0; |
3923 | } | |
3924 | ||
abe529af BP |
3925 | static void |
3926 | facet_push_stats(struct facet *facet) | |
3927 | { | |
112bc5f4 | 3928 | struct dpif_flow_stats stats; |
abe529af | 3929 | |
9d24de3b JP |
3930 | assert(facet->packet_count >= facet->prev_packet_count); |
3931 | assert(facet->byte_count >= facet->prev_byte_count); | |
3932 | assert(facet->used >= facet->prev_used); | |
abe529af | 3933 | |
112bc5f4 BP |
3934 | stats.n_packets = facet->packet_count - facet->prev_packet_count; |
3935 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
3936 | stats.used = facet->used; | |
3937 | stats.tcp_flags = 0; | |
abe529af | 3938 | |
112bc5f4 | 3939 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { |
9d24de3b JP |
3940 | facet->prev_packet_count = facet->packet_count; |
3941 | facet->prev_byte_count = facet->byte_count; | |
3942 | facet->prev_used = facet->used; | |
abe529af | 3943 | |
112bc5f4 | 3944 | flow_push_stats(facet->rule, &facet->flow, &stats); |
9d24de3b JP |
3945 | |
3946 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
112bc5f4 | 3947 | facet->mirrors, stats.n_packets, stats.n_bytes); |
abe529af BP |
3948 | } |
3949 | } | |
3950 | ||
abe529af | 3951 | static void |
112bc5f4 | 3952 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) |
abe529af | 3953 | { |
112bc5f4 BP |
3954 | rule->packet_count += stats->n_packets; |
3955 | rule->byte_count += stats->n_bytes; | |
3956 | ofproto_rule_update_used(&rule->up, stats->used); | |
abe529af BP |
3957 | } |
3958 | ||
3959 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 3960 | * 'rule''s actions and mirrors. */ |
abe529af | 3961 | static void |
18b2a258 | 3962 | flow_push_stats(struct rule_dpif *rule, |
112bc5f4 | 3963 | const struct flow *flow, const struct dpif_flow_stats *stats) |
abe529af BP |
3964 | { |
3965 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
112bc5f4 | 3966 | struct action_xlate_ctx ctx; |
abe529af | 3967 | |
112bc5f4 | 3968 | ofproto_rule_update_used(&rule->up, stats->used); |
f3b50afb | 3969 | |
112bc5f4 | 3970 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, |
0e553d9c | 3971 | 0, NULL); |
112bc5f4 BP |
3972 | ctx.resubmit_stats = stats; |
3973 | xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions); | |
abe529af BP |
3974 | } |
3975 | \f | |
b0f7b9b5 BP |
3976 | /* Subfacets. */ |
3977 | ||
3978 | static struct subfacet * | |
3979 | subfacet_find__(struct ofproto_dpif *ofproto, | |
3980 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
3981 | const struct flow *flow) | |
3982 | { | |
3983 | struct subfacet *subfacet; | |
3984 | ||
3985 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
3986 | &ofproto->subfacets) { | |
3987 | if (subfacet->key | |
3988 | ? (subfacet->key_len == key_len | |
3989 | && !memcmp(key, subfacet->key, key_len)) | |
3990 | : flow_equal(flow, &subfacet->facet->flow)) { | |
3991 | return subfacet; | |
3992 | } | |
3993 | } | |
3994 | ||
3995 | return NULL; | |
3996 | } | |
3997 | ||
3998 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
3999 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
4000 | * there is one, otherwise creates and returns a new subfacet. |
4001 | * | |
4002 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
4003 | * which case the caller must populate the actions with | |
4004 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 4005 | static struct subfacet * |
15baa734 | 4006 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
e84173dc | 4007 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) |
b0f7b9b5 | 4008 | { |
15baa734 | 4009 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
4010 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
4011 | struct subfacet *subfacet; | |
4012 | ||
4013 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); | |
4014 | if (subfacet) { | |
4015 | if (subfacet->facet == facet) { | |
4016 | return subfacet; | |
4017 | } | |
4018 | ||
4019 | /* This shouldn't happen. */ | |
4020 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
15baa734 | 4021 | subfacet_destroy(subfacet); |
b0f7b9b5 BP |
4022 | } |
4023 | ||
26cd7e34 BP |
4024 | subfacet = (list_is_empty(&facet->subfacets) |
4025 | ? &facet->one_subfacet | |
4026 | : xmalloc(sizeof *subfacet)); | |
b0f7b9b5 BP |
4027 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
4028 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
4029 | subfacet->facet = facet; | |
b0f7b9b5 BP |
4030 | subfacet->key_fitness = key_fitness; |
4031 | if (key_fitness != ODP_FIT_PERFECT) { | |
4032 | subfacet->key = xmemdup(key, key_len); | |
4033 | subfacet->key_len = key_len; | |
26cd7e34 BP |
4034 | } else { |
4035 | subfacet->key = NULL; | |
4036 | subfacet->key_len = 0; | |
b0f7b9b5 | 4037 | } |
26cd7e34 BP |
4038 | subfacet->used = time_msec(); |
4039 | subfacet->dp_packet_count = 0; | |
4040 | subfacet->dp_byte_count = 0; | |
4041 | subfacet->actions_len = 0; | |
4042 | subfacet->actions = NULL; | |
b0f7b9b5 | 4043 | subfacet->installed = false; |
e84173dc | 4044 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
4045 | |
4046 | return subfacet; | |
4047 | } | |
4048 | ||
4049 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
4050 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
4051 | static struct subfacet * | |
4052 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 4053 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
4054 | { |
4055 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
4056 | enum odp_key_fitness fitness; |
4057 | struct flow flow; | |
4058 | ||
4059 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
4060 | if (fitness == ODP_FIT_ERROR) { | |
4061 | return NULL; | |
4062 | } | |
b0f7b9b5 | 4063 | |
6a542738 | 4064 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
4065 | } |
4066 | ||
4067 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
4068 | * its facet within 'ofproto', and frees it. */ | |
4069 | static void | |
15baa734 | 4070 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 4071 | { |
15baa734 BP |
4072 | struct facet *facet = subfacet->facet; |
4073 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4074 | ||
4075 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
4076 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
4077 | list_remove(&subfacet->list_node); | |
4078 | free(subfacet->key); | |
b95fc6ba | 4079 | free(subfacet->actions); |
26cd7e34 BP |
4080 | if (subfacet != &facet->one_subfacet) { |
4081 | free(subfacet); | |
4082 | } | |
b0f7b9b5 BP |
4083 | } |
4084 | ||
4085 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4086 | * last remaining subfacet in its facet destroys the facet too. */ | |
4087 | static void | |
15baa734 | 4088 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
4089 | { |
4090 | struct facet *facet = subfacet->facet; | |
4091 | ||
551a2f6c BP |
4092 | if (list_is_singleton(&facet->subfacets)) { |
4093 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 4094 | facet_remove(facet); |
551a2f6c | 4095 | } else { |
15baa734 | 4096 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
4097 | } |
4098 | } | |
4099 | ||
4100 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
4101 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
4102 | * for use as temporary storage. */ | |
4103 | static void | |
4104 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
4105 | struct ofpbuf *key) | |
4106 | { | |
4107 | if (!subfacet->key) { | |
4108 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
4109 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
4110 | } else { | |
4111 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
4112 | } | |
4113 | } | |
4114 | ||
5fe20d5d BP |
4115 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. |
4116 | * Translates the actions into 'odp_actions', which the caller must have | |
4117 | * initialized and is responsible for uninitializing. */ | |
b95fc6ba | 4118 | static void |
5fe20d5d BP |
4119 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, |
4120 | struct ofpbuf *odp_actions) | |
b95fc6ba BP |
4121 | { |
4122 | struct facet *facet = subfacet->facet; | |
18b2a258 | 4123 | struct rule_dpif *rule = facet->rule; |
15baa734 | 4124 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 4125 | |
b95fc6ba BP |
4126 | struct action_xlate_ctx ctx; |
4127 | ||
15baa734 | 4128 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
0e553d9c | 4129 | rule, 0, packet); |
5fe20d5d | 4130 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions); |
b95fc6ba BP |
4131 | facet->tags = ctx.tags; |
4132 | facet->may_install = ctx.may_set_up_flow; | |
4133 | facet->has_learn = ctx.has_learn; | |
4134 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 4135 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 4136 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 4137 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 4138 | |
5fe20d5d BP |
4139 | if (subfacet->actions_len != odp_actions->size |
4140 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
b95fc6ba | 4141 | free(subfacet->actions); |
5fe20d5d BP |
4142 | subfacet->actions_len = odp_actions->size; |
4143 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
b95fc6ba | 4144 | } |
b95fc6ba BP |
4145 | } |
4146 | ||
b0f7b9b5 BP |
4147 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4148 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4149 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4150 | * since 'subfacet' was last updated. | |
4151 | * | |
4152 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4153 | static int | |
15baa734 | 4154 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 BP |
4155 | const struct nlattr *actions, size_t actions_len, |
4156 | struct dpif_flow_stats *stats) | |
4157 | { | |
15baa734 BP |
4158 | struct facet *facet = subfacet->facet; |
4159 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
b0f7b9b5 BP |
4160 | struct odputil_keybuf keybuf; |
4161 | enum dpif_flow_put_flags flags; | |
4162 | struct ofpbuf key; | |
4163 | int ret; | |
4164 | ||
4165 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
4166 | if (stats) { | |
4167 | flags |= DPIF_FP_ZERO_STATS; | |
4168 | } | |
4169 | ||
4170 | subfacet_get_key(subfacet, &keybuf, &key); | |
4171 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
4172 | actions, actions_len, stats); | |
4173 | ||
4174 | if (stats) { | |
4175 | subfacet_reset_dp_stats(subfacet, stats); | |
4176 | } | |
4177 | ||
4178 | return ret; | |
4179 | } | |
4180 | ||
4181 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
4182 | static void | |
15baa734 | 4183 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 BP |
4184 | { |
4185 | if (subfacet->installed) { | |
15baa734 BP |
4186 | struct rule_dpif *rule = subfacet->facet->rule; |
4187 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
4188 | struct odputil_keybuf keybuf; |
4189 | struct dpif_flow_stats stats; | |
4190 | struct ofpbuf key; | |
4191 | int error; | |
4192 | ||
4193 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 4194 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
4195 | subfacet_reset_dp_stats(subfacet, &stats); |
4196 | if (!error) { | |
15baa734 | 4197 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 BP |
4198 | } |
4199 | subfacet->installed = false; | |
4200 | } else { | |
4201 | assert(subfacet->dp_packet_count == 0); | |
4202 | assert(subfacet->dp_byte_count == 0); | |
4203 | } | |
4204 | } | |
4205 | ||
4206 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4207 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4208 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4209 | * was reset in the datapath. 'stats' will be modified to include only | |
4210 | * statistics new since 'subfacet' was last updated. */ | |
4211 | static void | |
4212 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4213 | struct dpif_flow_stats *stats) | |
4214 | { | |
4215 | if (stats | |
4216 | && subfacet->dp_packet_count <= stats->n_packets | |
4217 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4218 | stats->n_packets -= subfacet->dp_packet_count; | |
4219 | stats->n_bytes -= subfacet->dp_byte_count; | |
4220 | } | |
4221 | ||
4222 | subfacet->dp_packet_count = 0; | |
4223 | subfacet->dp_byte_count = 0; | |
4224 | } | |
4225 | ||
4226 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
4227 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
4228 | static void | |
15baa734 | 4229 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
4230 | { |
4231 | if (used > subfacet->used) { | |
4232 | subfacet->used = used; | |
15baa734 | 4233 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
4234 | } |
4235 | } | |
4236 | ||
4237 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
4238 | * | |
4239 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4240 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4241 | * represents a packet that was sent by hand or if it represents statistics | |
4242 | * that have been cleared out of the datapath. */ | |
4243 | static void | |
15baa734 | 4244 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4245 | const struct dpif_flow_stats *stats) |
4246 | { | |
4247 | if (stats->n_packets || stats->used > subfacet->used) { | |
4248 | struct facet *facet = subfacet->facet; | |
4249 | ||
15baa734 | 4250 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
4251 | facet->packet_count += stats->n_packets; |
4252 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4253 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4254 | facet_push_stats(facet); |
4255 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
4256 | } | |
4257 | } | |
4258 | \f | |
abe529af BP |
4259 | /* Rules. */ |
4260 | ||
4261 | static struct rule_dpif * | |
29901626 BP |
4262 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
4263 | uint8_t table_id) | |
abe529af | 4264 | { |
7257b535 BP |
4265 | struct cls_rule *cls_rule; |
4266 | struct classifier *cls; | |
4267 | ||
9cdaaebe BP |
4268 | if (table_id >= N_TABLES) { |
4269 | return NULL; | |
4270 | } | |
4271 | ||
d0918789 | 4272 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 4273 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
4274 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
4275 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
4276 | * are unavailable. */ | |
4277 | struct flow ofpc_normal_flow = *flow; | |
4278 | ofpc_normal_flow.tp_src = htons(0); | |
4279 | ofpc_normal_flow.tp_dst = htons(0); | |
4280 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
4281 | } else { | |
4282 | cls_rule = classifier_lookup(cls, flow); | |
4283 | } | |
4284 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
4285 | } |
4286 | ||
7ee20df1 BP |
4287 | static void |
4288 | complete_operation(struct rule_dpif *rule) | |
4289 | { | |
4290 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4291 | ||
54a9cbc9 | 4292 | rule_invalidate(rule); |
7ee20df1 BP |
4293 | if (clogged) { |
4294 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4295 | c->op = rule->up.pending; | |
4296 | list_push_back(&ofproto->completions, &c->list_node); | |
4297 | } else { | |
4298 | ofoperation_complete(rule->up.pending, 0); | |
4299 | } | |
4300 | } | |
4301 | ||
abe529af BP |
4302 | static struct rule * |
4303 | rule_alloc(void) | |
4304 | { | |
4305 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4306 | return &rule->up; | |
4307 | } | |
4308 | ||
4309 | static void | |
4310 | rule_dealloc(struct rule *rule_) | |
4311 | { | |
4312 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4313 | free(rule); | |
4314 | } | |
4315 | ||
90bf1e07 | 4316 | static enum ofperr |
abe529af BP |
4317 | rule_construct(struct rule *rule_) |
4318 | { | |
4319 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4320 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4321 | struct rule_dpif *victim; |
54a9cbc9 | 4322 | uint8_t table_id; |
90bf1e07 | 4323 | enum ofperr error; |
5bf0e941 BP |
4324 | |
4325 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
4326 | &rule->up.cr.flow, ofproto->max_ports); | |
4327 | if (error) { | |
4328 | return error; | |
4329 | } | |
abe529af | 4330 | |
abe529af BP |
4331 | rule->packet_count = 0; |
4332 | rule->byte_count = 0; | |
abe529af | 4333 | |
7ee20df1 BP |
4334 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4335 | if (victim && !list_is_empty(&victim->facets)) { | |
4336 | struct facet *facet; | |
4337 | ||
4338 | rule->facets = victim->facets; | |
4339 | list_moved(&rule->facets); | |
4340 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4341 | /* XXX: We're only clearing our local counters here. It's possible |
4342 | * that quite a few packets are unaccounted for in the datapath | |
4343 | * statistics. These will be accounted to the new rule instead of | |
4344 | * cleared as required. This could be fixed by clearing out the | |
4345 | * datapath statistics for this facet, but currently it doesn't | |
4346 | * seem worth it. */ | |
4347 | facet_reset_counters(facet); | |
7ee20df1 BP |
4348 | facet->rule = rule; |
4349 | } | |
4350 | } else { | |
4351 | /* Must avoid list_moved() in this case. */ | |
4352 | list_init(&rule->facets); | |
4353 | } | |
abe529af | 4354 | |
54a9cbc9 BP |
4355 | table_id = rule->up.table_id; |
4356 | rule->tag = (victim ? victim->tag | |
4357 | : table_id == 0 ? 0 | |
4358 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4359 | ofproto->tables[table_id].basis)); | |
4360 | ||
7ee20df1 | 4361 | complete_operation(rule); |
abe529af BP |
4362 | return 0; |
4363 | } | |
4364 | ||
4365 | static void | |
4366 | rule_destruct(struct rule *rule_) | |
4367 | { | |
4368 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4369 | struct facet *facet, *next_facet; |
4370 | ||
abe529af | 4371 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4372 | facet_revalidate(facet); |
abe529af | 4373 | } |
7ee20df1 BP |
4374 | |
4375 | complete_operation(rule); | |
abe529af BP |
4376 | } |
4377 | ||
4378 | static void | |
4379 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4380 | { | |
4381 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4382 | struct facet *facet; | |
4383 | ||
4384 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4385 | * in facets. This counts, for example, facets that have expired. */ | |
4386 | *packets = rule->packet_count; | |
4387 | *bytes = rule->byte_count; | |
4388 | ||
4389 | /* Add any statistics that are tracked by facets. This includes | |
4390 | * statistical data recently updated by ofproto_update_stats() as well as | |
4391 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4392 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4393 | *packets += facet->packet_count; | |
4394 | *bytes += facet->byte_count; | |
4395 | } | |
4396 | } | |
4397 | ||
90bf1e07 | 4398 | static enum ofperr |
59d0f2c8 BP |
4399 | rule_execute(struct rule *rule_, const struct flow *flow, |
4400 | struct ofpbuf *packet) | |
abe529af BP |
4401 | { |
4402 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4403 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
050ac423 | 4404 | |
112bc5f4 | 4405 | struct dpif_flow_stats stats; |
050ac423 | 4406 | |
abe529af | 4407 | struct action_xlate_ctx ctx; |
050ac423 BP |
4408 | uint64_t odp_actions_stub[1024 / 8]; |
4409 | struct ofpbuf odp_actions; | |
abe529af | 4410 | |
112bc5f4 BP |
4411 | dpif_flow_stats_extract(flow, packet, &stats); |
4412 | rule_credit_stats(rule, &stats); | |
4413 | ||
050ac423 | 4414 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
54834960 | 4415 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
112bc5f4 BP |
4416 | rule, stats.tcp_flags, packet); |
4417 | ctx.resubmit_stats = &stats; | |
050ac423 | 4418 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions); |
112bc5f4 BP |
4419 | |
4420 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
4421 | odp_actions.size, packet); | |
4422 | ||
050ac423 | 4423 | ofpbuf_uninit(&odp_actions); |
5bf0e941 BP |
4424 | |
4425 | return 0; | |
abe529af BP |
4426 | } |
4427 | ||
7ee20df1 BP |
4428 | static void |
4429 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4430 | { |
4431 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4432 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
90bf1e07 | 4433 | enum ofperr error; |
abe529af | 4434 | |
7ee20df1 BP |
4435 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
4436 | &rule->up.cr.flow, ofproto->max_ports); | |
4437 | if (error) { | |
4438 | ofoperation_complete(rule->up.pending, error); | |
4439 | return; | |
abe529af | 4440 | } |
7ee20df1 BP |
4441 | |
4442 | complete_operation(rule); | |
abe529af BP |
4443 | } |
4444 | \f | |
97d6520b | 4445 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4446 | * May modify 'packet'. |
abe529af BP |
4447 | * Returns 0 if successful, otherwise a positive errno value. */ |
4448 | static int | |
52a90c29 | 4449 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4450 | { |
97d6520b | 4451 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4452 | struct ofpbuf key, odp_actions; |
4453 | struct odputil_keybuf keybuf; | |
52a90c29 | 4454 | uint16_t odp_port; |
80e5eed9 | 4455 | struct flow flow; |
abe529af BP |
4456 | int error; |
4457 | ||
abff858b | 4458 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
52a90c29 BP |
4459 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4460 | flow.vlan_tci); | |
4461 | if (odp_port != ofport->odp_port) { | |
4462 | eth_pop_vlan(packet); | |
4463 | flow.vlan_tci = htons(0); | |
4464 | } | |
4465 | ||
80e5eed9 BP |
4466 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4467 | odp_flow_key_from_flow(&key, &flow); | |
4468 | ||
abe529af | 4469 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4470 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4471 | ||
df2c07f4 | 4472 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4473 | error = dpif_execute(ofproto->dpif, |
4474 | key.data, key.size, | |
4475 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4476 | packet); |
4477 | ofpbuf_uninit(&odp_actions); | |
4478 | ||
4479 | if (error) { | |
4480 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4481 | ofproto->up.name, odp_port, strerror(error)); | |
4482 | } | |
6527c598 | 4483 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4484 | return error; |
4485 | } | |
4486 | \f | |
df2c07f4 | 4487 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
4488 | |
4489 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4490 | struct action_xlate_ctx *ctx); | |
4cd78906 | 4491 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4492 | |
98403001 BP |
4493 | static size_t |
4494 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4495 | struct ofpbuf *odp_actions, | |
4496 | const struct flow *flow, | |
4497 | const struct user_action_cookie *cookie) | |
4498 | { | |
98403001 BP |
4499 | uint32_t pid; |
4500 | ||
4501 | pid = dpif_port_get_pid(ofproto->dpif, | |
4502 | ofp_port_to_odp_port(flow->in_port)); | |
4503 | ||
39db78a0 | 4504 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4505 | } |
4506 | ||
6ff686f2 PS |
4507 | /* Compose SAMPLE action for sFlow. */ |
4508 | static size_t | |
4509 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4510 | struct ofpbuf *odp_actions, | |
4511 | const struct flow *flow, | |
4512 | uint32_t odp_port) | |
4513 | { | |
4514 | uint32_t port_ifindex; | |
4515 | uint32_t probability; | |
98403001 | 4516 | struct user_action_cookie cookie; |
6ff686f2 | 4517 | size_t sample_offset, actions_offset; |
98403001 | 4518 | int cookie_offset, n_output; |
6ff686f2 PS |
4519 | |
4520 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4521 | return 0; | |
4522 | } | |
4523 | ||
4524 | if (odp_port == OVSP_NONE) { | |
4525 | port_ifindex = 0; | |
4526 | n_output = 0; | |
4527 | } else { | |
4528 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4529 | n_output = 1; | |
4530 | } | |
4531 | ||
4532 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
4533 | ||
4534 | /* Number of packets out of UINT_MAX to sample. */ | |
4535 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4536 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4537 | ||
4538 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
4539 | ||
98403001 BP |
4540 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
4541 | cookie.data = port_ifindex; | |
4542 | cookie.n_output = n_output; | |
4543 | cookie.vlan_tci = 0; | |
4544 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
4545 | |
4546 | nl_msg_end_nested(odp_actions, actions_offset); | |
4547 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4548 | return cookie_offset; |
6ff686f2 PS |
4549 | } |
4550 | ||
4551 | /* SAMPLE action must be first action in any given list of actions. | |
4552 | * At this point we do not have all information required to build it. So try to | |
4553 | * build sample action as complete as possible. */ | |
4554 | static void | |
4555 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4556 | { | |
4557 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4558 | ctx->odp_actions, | |
4559 | &ctx->flow, OVSP_NONE); | |
4560 | ctx->sflow_odp_port = 0; | |
4561 | ctx->sflow_n_outputs = 0; | |
4562 | } | |
4563 | ||
4564 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4565 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4566 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4567 | static void | |
4568 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4569 | { | |
4570 | const struct flow *base = &ctx->base_flow; | |
4571 | struct user_action_cookie *cookie; | |
4572 | ||
4573 | if (!ctx->user_cookie_offset) { | |
4574 | return; | |
4575 | } | |
4576 | ||
4577 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
4578 | sizeof(*cookie)); | |
4579 | assert(cookie != NULL); | |
4580 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
4581 | ||
4582 | if (ctx->sflow_n_outputs) { | |
4583 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
4584 | ctx->sflow_odp_port); | |
4585 | } | |
4586 | if (ctx->sflow_n_outputs >= 255) { | |
4587 | cookie->n_output = 255; | |
4588 | } else { | |
4589 | cookie->n_output = ctx->sflow_n_outputs; | |
4590 | } | |
4591 | cookie->vlan_tci = base->vlan_tci; | |
4592 | } | |
4593 | ||
6ff686f2 | 4594 | static void |
81b1afb1 EJ |
4595 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4596 | bool check_stp) | |
6ff686f2 | 4597 | { |
d59906fb | 4598 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4599 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4600 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4601 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4602 | uint16_t out_port; |
d59906fb | 4603 | |
81b1afb1 | 4604 | if (ofport) { |
8b36f51e EJ |
4605 | struct priority_to_dscp *pdscp; |
4606 | ||
9e1fd49b | 4607 | if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD |
81b1afb1 EJ |
4608 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { |
4609 | return; | |
4610 | } | |
8b36f51e | 4611 | |
deedf7e7 | 4612 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4613 | if (pdscp) { |
4614 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4615 | ctx->flow.nw_tos |= pdscp->dscp; | |
4616 | } | |
81b1afb1 EJ |
4617 | } else { |
4618 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4619 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4620 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4621 | } |
4622 | ||
52a90c29 BP |
4623 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4624 | ctx->flow.vlan_tci); | |
4625 | if (out_port != odp_port) { | |
4626 | ctx->flow.vlan_tci = htons(0); | |
4627 | } | |
5bbda0aa | 4628 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4629 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4630 | ||
6ff686f2 PS |
4631 | ctx->sflow_odp_port = odp_port; |
4632 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4633 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4634 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4635 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4636 | } |
4637 | ||
abe529af | 4638 | static void |
5e48dc2b | 4639 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4640 | { |
81b1afb1 | 4641 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4642 | } |
4643 | ||
4644 | static void | |
29901626 BP |
4645 | xlate_table_action(struct action_xlate_ctx *ctx, |
4646 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
4647 | { |
4648 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4649 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4650 | struct rule_dpif *rule; |
4651 | uint16_t old_in_port; | |
29901626 BP |
4652 | uint8_t old_table_id; |
4653 | ||
4654 | old_table_id = ctx->table_id; | |
4655 | ctx->table_id = table_id; | |
abe529af | 4656 | |
54a9cbc9 | 4657 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
4658 | old_in_port = ctx->flow.in_port; |
4659 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
4660 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
4661 | ||
4662 | /* Tag the flow. */ | |
4663 | if (table_id > 0 && table_id < N_TABLES) { | |
4664 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4665 | if (table->other_table) { | |
33780682 | 4666 | ctx->tags |= (rule && rule->tag |
54a9cbc9 BP |
4667 | ? rule->tag |
4668 | : rule_calculate_tag(&ctx->flow, | |
4669 | &table->other_table->wc, | |
4670 | table->basis)); | |
4671 | } | |
4672 | } | |
4673 | ||
4674 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
4675 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
4676 | ctx->flow.in_port = old_in_port; |
4677 | ||
4678 | if (ctx->resubmit_hook) { | |
4679 | ctx->resubmit_hook(ctx, rule); | |
4680 | } | |
4681 | ||
4682 | if (rule) { | |
18b2a258 | 4683 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 4684 | |
112bc5f4 BP |
4685 | if (ctx->resubmit_stats) { |
4686 | rule_credit_stats(rule, ctx->resubmit_stats); | |
4687 | } | |
4688 | ||
abe529af | 4689 | ctx->recurse++; |
18b2a258 | 4690 | ctx->rule = rule; |
abe529af | 4691 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); |
18b2a258 | 4692 | ctx->rule = old_rule; |
abe529af BP |
4693 | ctx->recurse--; |
4694 | } | |
29901626 BP |
4695 | |
4696 | ctx->table_id = old_table_id; | |
abe529af BP |
4697 | } else { |
4698 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4699 | ||
29901626 | 4700 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 4701 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 4702 | ctx->max_resubmit_trigger = true; |
abe529af BP |
4703 | } |
4704 | } | |
4705 | ||
29901626 BP |
4706 | static void |
4707 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
4708 | const struct nx_action_resubmit *nar) | |
4709 | { | |
4710 | uint16_t in_port; | |
4711 | uint8_t table_id; | |
4712 | ||
4713 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
4714 | ? ctx->flow.in_port | |
4715 | : ntohs(nar->in_port)); | |
4716 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
4717 | ||
4718 | xlate_table_action(ctx, in_port, table_id); | |
4719 | } | |
4720 | ||
abe529af | 4721 | static void |
d59906fb | 4722 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
4723 | { |
4724 | struct ofport_dpif *ofport; | |
4725 | ||
b3e9b2ed | 4726 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 4727 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
4728 | |
4729 | if (ofp_port == ctx->flow.in_port) { | |
4730 | continue; | |
4731 | } | |
4732 | ||
5e48dc2b | 4733 | if (all) { |
81b1afb1 | 4734 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 4735 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 4736 | compose_output_action(ctx, ofp_port); |
abe529af BP |
4737 | } |
4738 | } | |
b3e9b2ed EJ |
4739 | |
4740 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
4741 | } |
4742 | ||
6ff686f2 | 4743 | static void |
f0fd1a17 | 4744 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
4745 | enum ofp_packet_in_reason reason, |
4746 | uint16_t controller_id) | |
6ff686f2 | 4747 | { |
999fba59 EJ |
4748 | struct ofputil_packet_in pin; |
4749 | struct ofpbuf *packet; | |
6ff686f2 | 4750 | |
999fba59 EJ |
4751 | ctx->may_set_up_flow = false; |
4752 | if (!ctx->packet) { | |
4753 | return; | |
4754 | } | |
4755 | ||
4756 | packet = ofpbuf_clone(ctx->packet); | |
4757 | ||
4758 | if (packet->l2 && packet->l3) { | |
4759 | struct eth_header *eh; | |
4760 | ||
4761 | eth_pop_vlan(packet); | |
4762 | eh = packet->l2; | |
0104aba8 EJ |
4763 | |
4764 | /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 | |
4765 | * LLC frame. Calculating the Ethernet type of these frames is more | |
4766 | * trouble than seems appropriate for a simple assertion. */ | |
4767 | assert(ntohs(eh->eth_type) < ETH_TYPE_MIN | |
4768 | || eh->eth_type == ctx->flow.dl_type); | |
4769 | ||
999fba59 EJ |
4770 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
4771 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
4772 | ||
4773 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
4774 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
4775 | } | |
4776 | ||
4777 | if (packet->l4) { | |
4778 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4779 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
4780 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
4781 | } | |
4782 | ||
4783 | if (packet->l7) { | |
4784 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
4785 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
4786 | ctx->flow.tp_dst); | |
4787 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
4788 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
4789 | ctx->flow.tp_dst); | |
4790 | } | |
4791 | } | |
4792 | } | |
4793 | } | |
4794 | ||
4795 | pin.packet = packet->data; | |
4796 | pin.packet_len = packet->size; | |
f0fd1a17 | 4797 | pin.reason = reason; |
a7349929 | 4798 | pin.controller_id = controller_id; |
54834960 | 4799 | pin.table_id = ctx->table_id; |
18b2a258 | 4800 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 4801 | |
999fba59 | 4802 | pin.send_len = len; |
999fba59 EJ |
4803 | flow_get_metadata(&ctx->flow, &pin.fmd); |
4804 | ||
d8653c38 | 4805 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 4806 | ofpbuf_delete(packet); |
6ff686f2 PS |
4807 | } |
4808 | ||
f0fd1a17 PS |
4809 | static bool |
4810 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
4811 | { | |
4812 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
4813 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
4814 | return false; | |
4815 | } | |
4816 | ||
4817 | if (ctx->flow.nw_ttl > 1) { | |
4818 | ctx->flow.nw_ttl--; | |
4819 | return false; | |
4820 | } else { | |
a7349929 | 4821 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); |
f0fd1a17 PS |
4822 | |
4823 | /* Stop processing for current table. */ | |
4824 | return true; | |
4825 | } | |
4826 | } | |
4827 | ||
abe529af BP |
4828 | static void |
4829 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
4830 | uint16_t port, uint16_t max_len) | |
4831 | { | |
4832 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
4833 | ||
4834 | ctx->nf_output_iface = NF_OUT_DROP; | |
4835 | ||
4836 | switch (port) { | |
4837 | case OFPP_IN_PORT: | |
81b1afb1 | 4838 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
4839 | break; |
4840 | case OFPP_TABLE: | |
29901626 | 4841 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
4842 | break; |
4843 | case OFPP_NORMAL: | |
4844 | xlate_normal(ctx); | |
4845 | break; | |
4846 | case OFPP_FLOOD: | |
d59906fb | 4847 | flood_packets(ctx, false); |
abe529af BP |
4848 | break; |
4849 | case OFPP_ALL: | |
d59906fb | 4850 | flood_packets(ctx, true); |
abe529af BP |
4851 | break; |
4852 | case OFPP_CONTROLLER: | |
a7349929 | 4853 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 4854 | break; |
e81d2933 EJ |
4855 | case OFPP_NONE: |
4856 | break; | |
a0fbe94a | 4857 | case OFPP_LOCAL: |
abe529af BP |
4858 | default: |
4859 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 4860 | compose_output_action(ctx, port); |
abe529af BP |
4861 | } |
4862 | break; | |
4863 | } | |
4864 | ||
4865 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
4866 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4867 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4868 | ctx->nf_output_iface = prev_nf_output_iface; | |
4869 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
4870 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
4871 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4872 | } | |
4873 | } | |
4874 | ||
f694937d EJ |
4875 | static void |
4876 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
4877 | const struct nx_action_output_reg *naor) | |
4878 | { | |
816fd533 | 4879 | struct mf_subfield src; |
f694937d EJ |
4880 | uint64_t ofp_port; |
4881 | ||
816fd533 BP |
4882 | nxm_decode(&src, naor->src, naor->ofs_nbits); |
4883 | ofp_port = mf_get_subfield(&src, &ctx->flow); | |
f694937d EJ |
4884 | |
4885 | if (ofp_port <= UINT16_MAX) { | |
4886 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
4887 | } | |
4888 | } | |
4889 | ||
abe529af BP |
4890 | static void |
4891 | xlate_output_action(struct action_xlate_ctx *ctx, | |
4892 | const struct ofp_action_output *oao) | |
4893 | { | |
4894 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
4895 | } | |
4896 | ||
abe529af BP |
4897 | static void |
4898 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
4899 | const struct ofp_action_enqueue *oae) | |
4900 | { | |
e479e41e | 4901 | uint16_t ofp_port; |
abff858b | 4902 | uint32_t flow_priority, priority; |
abe529af BP |
4903 | int error; |
4904 | ||
4905 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
4906 | &priority); | |
4907 | if (error) { | |
4908 | /* Fall back to ordinary output action. */ | |
4909 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
4910 | return; | |
4911 | } | |
4912 | ||
df2c07f4 | 4913 | /* Figure out datapath output port. */ |
abe529af BP |
4914 | ofp_port = ntohs(oae->port); |
4915 | if (ofp_port == OFPP_IN_PORT) { | |
4916 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
4917 | } else if (ofp_port == ctx->flow.in_port) { |
4918 | return; | |
abe529af | 4919 | } |
abe529af | 4920 | |
df2c07f4 | 4921 | /* Add datapath actions. */ |
deedf7e7 BP |
4922 | flow_priority = ctx->flow.skb_priority; |
4923 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 4924 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 4925 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
4926 | |
4927 | /* Update NetFlow output port. */ | |
4928 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 4929 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
4930 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
4931 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4932 | } | |
4933 | } | |
4934 | ||
4935 | static void | |
4936 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
4937 | const struct nx_action_set_queue *nasq) | |
4938 | { | |
4939 | uint32_t priority; | |
4940 | int error; | |
4941 | ||
4942 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4943 | &priority); | |
4944 | if (error) { | |
4945 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4946 | * has already been logged. */ | |
4947 | return; | |
4948 | } | |
4949 | ||
deedf7e7 | 4950 | ctx->flow.skb_priority = priority; |
abe529af BP |
4951 | } |
4952 | ||
4953 | struct xlate_reg_state { | |
4954 | ovs_be16 vlan_tci; | |
4955 | ovs_be64 tun_id; | |
4956 | }; | |
4957 | ||
abe529af BP |
4958 | static void |
4959 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4960 | const struct nx_action_autopath *naa) | |
4961 | { | |
4962 | uint16_t ofp_port = ntohl(naa->id); | |
4963 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4964 | ||
4965 | if (!port || !port->bundle) { | |
4966 | ofp_port = OFPP_NONE; | |
4967 | } else if (port->bundle->bond) { | |
4968 | /* Autopath does not support VLAN hashing. */ | |
4969 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4970 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4971 | if (slave) { |
4972 | ofp_port = slave->up.ofp_port; | |
4973 | } | |
4974 | } | |
4975 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4976 | } | |
4977 | ||
daff3353 EJ |
4978 | static bool |
4979 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4980 | { | |
4981 | struct ofproto_dpif *ofproto = ofproto_; | |
4982 | struct ofport_dpif *port; | |
4983 | ||
4984 | switch (ofp_port) { | |
4985 | case OFPP_IN_PORT: | |
4986 | case OFPP_TABLE: | |
4987 | case OFPP_NORMAL: | |
4988 | case OFPP_FLOOD: | |
4989 | case OFPP_ALL: | |
439e4d8c | 4990 | case OFPP_NONE: |
daff3353 EJ |
4991 | return true; |
4992 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4993 | return false; | |
4994 | default: | |
4995 | port = get_ofp_port(ofproto, ofp_port); | |
4996 | return port ? port->may_enable : false; | |
4997 | } | |
4998 | } | |
4999 | ||
75a75043 BP |
5000 | static void |
5001 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
5002 | const struct nx_action_learn *learn) | |
5003 | { | |
5004 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
5005 | struct ofputil_flow_mod fm; | |
5006 | int error; | |
5007 | ||
5008 | learn_execute(learn, &ctx->flow, &fm); | |
5009 | ||
5010 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
5011 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
5012 | VLOG_WARN("learning action failed to modify flow table (%s)", |
5013 | ofperr_get_name(error)); | |
75a75043 BP |
5014 | } |
5015 | ||
5016 | free(fm.actions); | |
5017 | } | |
5018 | ||
0e553d9c BP |
5019 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
5020 | * means "infinite". */ | |
5021 | static void | |
5022 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
5023 | { | |
5024 | if (max && (!*timeout || *timeout > max)) { | |
5025 | *timeout = max; | |
5026 | } | |
5027 | } | |
5028 | ||
5029 | static void | |
5030 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
5031 | const struct nx_action_fin_timeout *naft) | |
5032 | { | |
5033 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
5034 | struct rule_dpif *rule = ctx->rule; | |
5035 | ||
5036 | reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout); | |
5037 | reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout); | |
5038 | } | |
5039 | } | |
5040 | ||
21f7563c JP |
5041 | static bool |
5042 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
5043 | { | |
9e1fd49b BP |
5044 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
5045 | ? OFPUTIL_PC_NO_RECV_STP | |
5046 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
5047 | return false; |
5048 | } | |
5049 | ||
5050 | /* Only drop packets here if both forwarding and learning are | |
5051 | * disabled. If just learning is enabled, we need to have | |
5052 | * OFPP_NORMAL and the learning action have a look at the packet | |
5053 | * before we can drop it. */ | |
5054 | if (!stp_forward_in_state(port->stp_state) | |
5055 | && !stp_learn_in_state(port->stp_state)) { | |
5056 | return false; | |
5057 | } | |
5058 | ||
5059 | return true; | |
5060 | } | |
5061 | ||
abe529af BP |
5062 | static void |
5063 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
5064 | struct action_xlate_ctx *ctx) | |
5065 | { | |
5066 | const struct ofport_dpif *port; | |
abe529af | 5067 | const union ofp_action *ia; |
254750ce | 5068 | bool was_evictable = true; |
b4b8c781 | 5069 | size_t left; |
abe529af BP |
5070 | |
5071 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 5072 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
5073 | /* Drop this flow. */ |
5074 | return; | |
5075 | } | |
5076 | ||
254750ce BP |
5077 | if (ctx->rule) { |
5078 | /* Don't let the rule we're working on get evicted underneath us. */ | |
5079 | was_evictable = ctx->rule->up.evictable; | |
5080 | ctx->rule->up.evictable = false; | |
5081 | } | |
b4b8c781 | 5082 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 5083 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
5084 | const struct nx_action_resubmit *nar; |
5085 | const struct nx_action_set_tunnel *nast; | |
5086 | const struct nx_action_set_queue *nasq; | |
5087 | const struct nx_action_multipath *nam; | |
5088 | const struct nx_action_autopath *naa; | |
daff3353 | 5089 | const struct nx_action_bundle *nab; |
f694937d | 5090 | const struct nx_action_output_reg *naor; |
a7349929 | 5091 | const struct nx_action_controller *nac; |
38f2e360 BP |
5092 | enum ofputil_action_code code; |
5093 | ovs_be64 tun_id; | |
5094 | ||
848e8809 EJ |
5095 | if (ctx->exit) { |
5096 | break; | |
5097 | } | |
5098 | ||
38f2e360 BP |
5099 | code = ofputil_decode_action_unsafe(ia); |
5100 | switch (code) { | |
08f94c0e | 5101 | case OFPUTIL_OFPAT10_OUTPUT: |
abe529af BP |
5102 | xlate_output_action(ctx, &ia->output); |
5103 | break; | |
5104 | ||
08f94c0e | 5105 | case OFPUTIL_OFPAT10_SET_VLAN_VID: |
abe529af BP |
5106 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
5107 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
5108 | break; |
5109 | ||
08f94c0e | 5110 | case OFPUTIL_OFPAT10_SET_VLAN_PCP: |
abe529af BP |
5111 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
5112 | ctx->flow.vlan_tci |= htons( | |
5113 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
5114 | break; |
5115 | ||
08f94c0e | 5116 | case OFPUTIL_OFPAT10_STRIP_VLAN: |
abe529af | 5117 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
5118 | break; |
5119 | ||
08f94c0e | 5120 | case OFPUTIL_OFPAT10_SET_DL_SRC: |
abe529af | 5121 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
5122 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
5123 | break; | |
5124 | ||
08f94c0e | 5125 | case OFPUTIL_OFPAT10_SET_DL_DST: |
abe529af | 5126 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
5127 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
5128 | break; | |
5129 | ||
08f94c0e | 5130 | case OFPUTIL_OFPAT10_SET_NW_SRC: |
abe529af BP |
5131 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
5132 | break; | |
5133 | ||
08f94c0e | 5134 | case OFPUTIL_OFPAT10_SET_NW_DST: |
abe529af BP |
5135 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
5136 | break; | |
5137 | ||
08f94c0e | 5138 | case OFPUTIL_OFPAT10_SET_NW_TOS: |
c4f2731d PS |
5139 | /* OpenFlow 1.0 only supports IPv4. */ |
5140 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5141 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
5142 | ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
5143 | } | |
abe529af BP |
5144 | break; |
5145 | ||
08f94c0e | 5146 | case OFPUTIL_OFPAT10_SET_TP_SRC: |
abe529af BP |
5147 | ctx->flow.tp_src = ia->tp_port.tp_port; |
5148 | break; | |
5149 | ||
08f94c0e | 5150 | case OFPUTIL_OFPAT10_SET_TP_DST: |
abe529af BP |
5151 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
5152 | break; | |
5153 | ||
08f94c0e | 5154 | case OFPUTIL_OFPAT10_ENQUEUE: |
38f2e360 BP |
5155 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); |
5156 | break; | |
5157 | ||
5158 | case OFPUTIL_NXAST_RESUBMIT: | |
5159 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
5160 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
5161 | break; | |
5162 | ||
5163 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
5164 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
5165 | break; |
5166 | ||
38f2e360 BP |
5167 | case OFPUTIL_NXAST_SET_TUNNEL: |
5168 | nast = (const struct nx_action_set_tunnel *) ia; | |
5169 | tun_id = htonll(ntohl(nast->tun_id)); | |
5170 | ctx->flow.tun_id = tun_id; | |
5171 | break; | |
5172 | ||
5173 | case OFPUTIL_NXAST_SET_QUEUE: | |
5174 | nasq = (const struct nx_action_set_queue *) ia; | |
5175 | xlate_set_queue_action(ctx, nasq); | |
5176 | break; | |
5177 | ||
5178 | case OFPUTIL_NXAST_POP_QUEUE: | |
deedf7e7 | 5179 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
5180 | break; |
5181 | ||
5182 | case OFPUTIL_NXAST_REG_MOVE: | |
5183 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
5184 | &ctx->flow); | |
5185 | break; | |
5186 | ||
5187 | case OFPUTIL_NXAST_REG_LOAD: | |
5188 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
5189 | &ctx->flow); | |
5190 | break; | |
5191 | ||
5192 | case OFPUTIL_NXAST_NOTE: | |
5193 | /* Nothing to do. */ | |
5194 | break; | |
5195 | ||
5196 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
5197 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
5198 | ctx->flow.tun_id = tun_id; | |
5199 | break; | |
5200 | ||
5201 | case OFPUTIL_NXAST_MULTIPATH: | |
5202 | nam = (const struct nx_action_multipath *) ia; | |
5203 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
5204 | break; |
5205 | ||
38f2e360 BP |
5206 | case OFPUTIL_NXAST_AUTOPATH: |
5207 | naa = (const struct nx_action_autopath *) ia; | |
5208 | xlate_autopath(ctx, naa); | |
abe529af | 5209 | break; |
daff3353 EJ |
5210 | |
5211 | case OFPUTIL_NXAST_BUNDLE: | |
5212 | ctx->ofproto->has_bundle_action = true; | |
5213 | nab = (const struct nx_action_bundle *) ia; | |
5214 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
5215 | slave_enabled_cb, | |
5216 | ctx->ofproto), 0); | |
5217 | break; | |
a368bb53 EJ |
5218 | |
5219 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
5220 | ctx->ofproto->has_bundle_action = true; | |
5221 | nab = (const struct nx_action_bundle *) ia; | |
5222 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
5223 | ctx->ofproto); | |
5224 | break; | |
f694937d EJ |
5225 | |
5226 | case OFPUTIL_NXAST_OUTPUT_REG: | |
5227 | naor = (const struct nx_action_output_reg *) ia; | |
5228 | xlate_output_reg_action(ctx, naor); | |
5229 | break; | |
75a75043 BP |
5230 | |
5231 | case OFPUTIL_NXAST_LEARN: | |
5232 | ctx->has_learn = true; | |
3de9590b | 5233 | if (ctx->may_learn) { |
75a75043 BP |
5234 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); |
5235 | } | |
5236 | break; | |
848e8809 | 5237 | |
f0fd1a17 PS |
5238 | case OFPUTIL_NXAST_DEC_TTL: |
5239 | if (compose_dec_ttl(ctx)) { | |
5240 | goto out; | |
5241 | } | |
5242 | break; | |
5243 | ||
848e8809 EJ |
5244 | case OFPUTIL_NXAST_EXIT: |
5245 | ctx->exit = true; | |
5246 | break; | |
0e553d9c BP |
5247 | |
5248 | case OFPUTIL_NXAST_FIN_TIMEOUT: | |
5249 | ctx->has_fin_timeout = true; | |
5250 | xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia); | |
5251 | break; | |
a7349929 BP |
5252 | |
5253 | case OFPUTIL_NXAST_CONTROLLER: | |
5254 | nac = (const struct nx_action_controller *) ia; | |
5255 | execute_controller_action(ctx, ntohs(nac->max_len), nac->reason, | |
5256 | ntohs(nac->controller_id)); | |
5257 | break; | |
abe529af BP |
5258 | } |
5259 | } | |
21f7563c | 5260 | |
f0fd1a17 | 5261 | out: |
21f7563c JP |
5262 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
5263 | * so drop it now if forwarding is disabled. */ | |
5264 | if (port && !stp_forward_in_state(port->stp_state)) { | |
5265 | ofpbuf_clear(ctx->odp_actions); | |
5266 | add_sflow_action(ctx); | |
5267 | } | |
254750ce BP |
5268 | if (ctx->rule) { |
5269 | ctx->rule->up.evictable = was_evictable; | |
5270 | } | |
abe529af BP |
5271 | } |
5272 | ||
5273 | static void | |
5274 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
5275 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
18b2a258 | 5276 | ovs_be16 initial_tci, struct rule_dpif *rule, |
0e553d9c | 5277 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af BP |
5278 | { |
5279 | ctx->ofproto = ofproto; | |
5280 | ctx->flow = *flow; | |
e84173dc BP |
5281 | ctx->base_flow = ctx->flow; |
5282 | ctx->base_flow.tun_id = 0; | |
5283 | ctx->base_flow.vlan_tci = initial_tci; | |
18b2a258 | 5284 | ctx->rule = rule; |
abe529af | 5285 | ctx->packet = packet; |
3de9590b | 5286 | ctx->may_learn = packet != NULL; |
0e553d9c | 5287 | ctx->tcp_flags = tcp_flags; |
abe529af | 5288 | ctx->resubmit_hook = NULL; |
112bc5f4 | 5289 | ctx->resubmit_stats = NULL; |
abe529af BP |
5290 | } |
5291 | ||
050ac423 BP |
5292 | /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in |
5293 | * 'odp_actions', using 'ctx'. */ | |
5294 | static void | |
abe529af | 5295 | xlate_actions(struct action_xlate_ctx *ctx, |
050ac423 BP |
5296 | const union ofp_action *in, size_t n_in, |
5297 | struct ofpbuf *odp_actions) | |
abe529af | 5298 | { |
c06bba01 JP |
5299 | struct flow orig_flow = ctx->flow; |
5300 | ||
abe529af BP |
5301 | COVERAGE_INC(ofproto_dpif_xlate); |
5302 | ||
050ac423 BP |
5303 | ofpbuf_clear(odp_actions); |
5304 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
5305 | ||
5306 | ctx->odp_actions = odp_actions; | |
97e42c92 BP |
5307 | ctx->tags = 0; |
5308 | ctx->may_set_up_flow = true; | |
5309 | ctx->has_learn = false; | |
5310 | ctx->has_normal = false; | |
0e553d9c | 5311 | ctx->has_fin_timeout = false; |
97e42c92 | 5312 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 5313 | ctx->mirrors = 0; |
97e42c92 | 5314 | ctx->recurse = 0; |
6a6455e5 | 5315 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 5316 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 5317 | ctx->table_id = 0; |
848e8809 | 5318 | ctx->exit = false; |
7257b535 | 5319 | |
eadef313 | 5320 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
5321 | switch (ctx->ofproto->up.frag_handling) { |
5322 | case OFPC_FRAG_NORMAL: | |
5323 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
5324 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
5325 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
5326 | break; |
5327 | ||
5328 | case OFPC_FRAG_DROP: | |
050ac423 | 5329 | return; |
7257b535 BP |
5330 | |
5331 | case OFPC_FRAG_REASM: | |
5332 | NOT_REACHED(); | |
5333 | ||
5334 | case OFPC_FRAG_NX_MATCH: | |
5335 | /* Nothing to do. */ | |
5336 | break; | |
f0fd1a17 PS |
5337 | |
5338 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
5339 | NOT_REACHED(); | |
7257b535 BP |
5340 | } |
5341 | } | |
5342 | ||
fc08b7a2 | 5343 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af BP |
5344 | ctx->may_set_up_flow = false; |
5345 | } else { | |
6a6455e5 EJ |
5346 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
5347 | struct flow original_flow = ctx->flow; | |
5348 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; | |
5349 | ||
6ff686f2 | 5350 | add_sflow_action(ctx); |
abe529af | 5351 | do_xlate_actions(in, n_in, ctx); |
abe529af | 5352 | |
6a6455e5 EJ |
5353 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook |
5354 | && !VLOG_DROP_ERR(&trace_rl)) { | |
5355 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5356 | ||
5357 | ofproto_trace(ctx->ofproto, &original_flow, ctx->packet, | |
5358 | initial_tci, &ds); | |
5359 | VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s", | |
5360 | ds_cstr(&ds)); | |
5361 | ds_destroy(&ds); | |
5362 | } | |
5363 | ||
b6848f13 BP |
5364 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
5365 | ctx->odp_actions->data, | |
5366 | ctx->odp_actions->size)) { | |
5367 | ctx->may_set_up_flow = false; | |
5368 | if (ctx->packet | |
5369 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
5370 | ctx->packet)) { | |
5e48dc2b | 5371 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
5372 | } |
5373 | } | |
c06bba01 | 5374 | add_mirror_actions(ctx, &orig_flow); |
a7c4eaf6 | 5375 | fix_sflow_action(ctx); |
abe529af | 5376 | } |
050ac423 BP |
5377 | } |
5378 | ||
5379 | /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions, | |
5380 | * using 'ctx', and discards the datapath actions. */ | |
5381 | static void | |
5382 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
5383 | const union ofp_action *in, size_t n_in) | |
5384 | { | |
5385 | uint64_t odp_actions_stub[1024 / 8]; | |
5386 | struct ofpbuf odp_actions; | |
abe529af | 5387 | |
050ac423 BP |
5388 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
5389 | xlate_actions(ctx, in, n_in, &odp_actions); | |
5390 | ofpbuf_uninit(&odp_actions); | |
abe529af BP |
5391 | } |
5392 | \f | |
5393 | /* OFPP_NORMAL implementation. */ | |
5394 | ||
abe529af BP |
5395 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
5396 | ||
ecac4ebf BP |
5397 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
5398 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5399 | * the bundle on which the packet was received, returns the VLAN to which the | |
5400 | * packet belongs. | |
5401 | * | |
5402 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5403 | static uint16_t | |
5404 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5405 | { | |
5406 | switch (in_bundle->vlan_mode) { | |
5407 | case PORT_VLAN_ACCESS: | |
5408 | return in_bundle->vlan; | |
5409 | break; | |
5410 | ||
5411 | case PORT_VLAN_TRUNK: | |
5412 | return vid; | |
5413 | ||
5414 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5415 | case PORT_VLAN_NATIVE_TAGGED: | |
5416 | return vid ? vid : in_bundle->vlan; | |
5417 | ||
5418 | default: | |
5419 | NOT_REACHED(); | |
5420 | } | |
5421 | } | |
5422 | ||
5da5ec37 BP |
5423 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5424 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5425 | * a warning. | |
5426 | * | |
5427 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5428 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5429 | * 0...4095. */ | |
5430 | static bool | |
5431 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5432 | { | |
33158a18 JP |
5433 | /* Allow any VID on the OFPP_NONE port. */ |
5434 | if (in_bundle == &ofpp_none_bundle) { | |
5435 | return true; | |
5436 | } | |
5437 | ||
5da5ec37 BP |
5438 | switch (in_bundle->vlan_mode) { |
5439 | case PORT_VLAN_ACCESS: | |
5440 | if (vid) { | |
5441 | if (warn) { | |
5442 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5443 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5444 | "packet received on port %s configured as VLAN " | |
5445 | "%"PRIu16" access port", | |
5446 | in_bundle->ofproto->up.name, vid, | |
5447 | in_bundle->name, in_bundle->vlan); | |
5448 | } | |
5449 | return false; | |
5450 | } | |
5451 | return true; | |
5452 | ||
5453 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5454 | case PORT_VLAN_NATIVE_TAGGED: | |
5455 | if (!vid) { | |
5456 | /* Port must always carry its native VLAN. */ | |
5457 | return true; | |
5458 | } | |
5459 | /* Fall through. */ | |
5460 | case PORT_VLAN_TRUNK: | |
5461 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5462 | if (warn) { | |
5463 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5464 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5465 | "received on port %s not configured for trunking " | |
5466 | "VLAN %"PRIu16, | |
5467 | in_bundle->ofproto->up.name, vid, | |
5468 | in_bundle->name, vid); | |
5469 | } | |
5470 | return false; | |
5471 | } | |
5472 | return true; | |
5473 | ||
5474 | default: | |
5475 | NOT_REACHED(); | |
5476 | } | |
5477 | ||
5478 | } | |
5479 | ||
ecac4ebf BP |
5480 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5481 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5482 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5483 | * then the 802.1Q header should only be included in the packet if there is a | |
5484 | * nonzero PCP.) | |
5485 | * | |
5486 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5487 | static uint16_t | |
5488 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5489 | { | |
5490 | switch (out_bundle->vlan_mode) { | |
5491 | case PORT_VLAN_ACCESS: | |
5492 | return 0; | |
5493 | ||
5494 | case PORT_VLAN_TRUNK: | |
5495 | case PORT_VLAN_NATIVE_TAGGED: | |
5496 | return vlan; | |
5497 | ||
5498 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5499 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5500 | ||
5501 | default: | |
5502 | NOT_REACHED(); | |
5503 | } | |
5504 | } | |
5505 | ||
395e68ce BP |
5506 | static void |
5507 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5508 | uint16_t vlan) | |
abe529af | 5509 | { |
395e68ce BP |
5510 | struct ofport_dpif *port; |
5511 | uint16_t vid; | |
81b1afb1 | 5512 | ovs_be16 tci, old_tci; |
ecac4ebf | 5513 | |
395e68ce BP |
5514 | vid = output_vlan_to_vid(out_bundle, vlan); |
5515 | if (!out_bundle->bond) { | |
5516 | port = ofbundle_get_a_port(out_bundle); | |
5517 | } else { | |
5518 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5519 | vid, &ctx->tags); | |
5520 | if (!port) { | |
5521 | /* No slaves enabled, so drop packet. */ | |
5522 | return; | |
5523 | } | |
5524 | } | |
abe529af | 5525 | |
81b1afb1 | 5526 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5527 | tci = htons(vid); |
5528 | if (tci || out_bundle->use_priority_tags) { | |
5529 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5530 | if (tci) { | |
5531 | tci |= htons(VLAN_CFI); | |
5532 | } | |
395e68ce | 5533 | } |
81b1afb1 | 5534 | ctx->flow.vlan_tci = tci; |
395e68ce | 5535 | |
5e48dc2b | 5536 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5537 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5538 | } |
5539 | ||
5540 | static int | |
5541 | mirror_mask_ffs(mirror_mask_t mask) | |
5542 | { | |
5543 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5544 | return ffs(mask); | |
5545 | } | |
5546 | ||
abe529af BP |
5547 | static bool |
5548 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5549 | { | |
ecac4ebf | 5550 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5551 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5552 | } |
5553 | ||
5554 | static bool | |
5555 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5556 | { | |
5557 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5558 | } | |
5559 | ||
5560 | /* Returns an arbitrary interface within 'bundle'. */ | |
5561 | static struct ofport_dpif * | |
5562 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5563 | { | |
5564 | return CONTAINER_OF(list_front(&bundle->ports), | |
5565 | struct ofport_dpif, bundle_node); | |
5566 | } | |
5567 | ||
abe529af BP |
5568 | static bool |
5569 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5570 | { | |
fc3d7408 | 5571 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5572 | } |
5573 | ||
07817dfe BP |
5574 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
5575 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
5576 | * protocols that may confuse switches. */ | |
5577 | static bool | |
5578 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
5579 | { | |
5580 | /* If you change this function's behavior, please update corresponding | |
5581 | * documentation in vswitch.xml at the same time. */ | |
5582 | if (dst[0] != 0x01) { | |
5583 | /* All the currently banned MACs happen to start with 01 currently, so | |
5584 | * this is a quick way to eliminate most of the good ones. */ | |
5585 | } else { | |
5586 | if (eth_addr_is_reserved(dst)) { | |
5587 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
5588 | * (01-80-c2-00-00-0x). */ | |
5589 | return false; | |
5590 | } | |
5591 | ||
5592 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
5593 | /* Cisco OUI. */ | |
5594 | if ((dst[3] & 0xfe) == 0xcc && | |
5595 | (dst[4] & 0xfe) == 0xcc && | |
5596 | (dst[5] & 0xfe) == 0xcc) { | |
5597 | /* Drop the following protocols plus others following the same | |
5598 | pattern: | |
5599 | ||
5600 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
5601 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
5602 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
5603 | return false; | |
5604 | } | |
5605 | ||
5606 | if (!(dst[3] | dst[4] | dst[5])) { | |
5607 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
5608 | return false; | |
5609 | } | |
5610 | } | |
5611 | } | |
5612 | return true; | |
5613 | } | |
5614 | ||
abe529af | 5615 | static void |
c06bba01 | 5616 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5617 | { |
5618 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5619 | mirror_mask_t mirrors; | |
c06bba01 JP |
5620 | struct ofbundle *in_bundle; |
5621 | uint16_t vlan; | |
5622 | uint16_t vid; | |
5623 | const struct nlattr *a; | |
5624 | size_t left; | |
5625 | ||
3581c12c JP |
5626 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
5627 | ctx->packet != NULL); | |
5628 | if (!in_bundle) { | |
c06bba01 JP |
5629 | return; |
5630 | } | |
c06bba01 JP |
5631 | mirrors = in_bundle->src_mirrors; |
5632 | ||
5633 | /* Drop frames on bundles reserved for mirroring. */ | |
5634 | if (in_bundle->mirror_out) { | |
5635 | if (ctx->packet != NULL) { | |
5636 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5637 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5638 | "%s, which is reserved exclusively for mirroring", | |
5639 | ctx->ofproto->up.name, in_bundle->name); | |
5640 | } | |
5641 | return; | |
5642 | } | |
5643 | ||
5644 | /* Check VLAN. */ | |
5645 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5646 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5647 | return; | |
5648 | } | |
5649 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5650 | ||
5651 | /* Look at the output ports to check for destination selections. */ | |
5652 | ||
5653 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5654 | ctx->odp_actions->size) { | |
5655 | enum ovs_action_attr type = nl_attr_type(a); | |
5656 | struct ofport_dpif *ofport; | |
5657 | ||
5658 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5659 | continue; | |
5660 | } | |
5661 | ||
5662 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5663 | if (ofport && ofport->bundle) { |
5664 | mirrors |= ofport->bundle->dst_mirrors; | |
5665 | } | |
c06bba01 | 5666 | } |
abe529af BP |
5667 | |
5668 | if (!mirrors) { | |
5669 | return; | |
5670 | } | |
5671 | ||
c06bba01 JP |
5672 | /* Restore the original packet before adding the mirror actions. */ |
5673 | ctx->flow = *orig_flow; | |
5674 | ||
9ba15e2a BP |
5675 | while (mirrors) { |
5676 | struct ofmirror *m; | |
9ba15e2a BP |
5677 | |
5678 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5679 | ||
5680 | if (!vlan_is_mirrored(m, vlan)) { | |
5681 | mirrors &= mirrors - 1; | |
5682 | continue; | |
5683 | } | |
5684 | ||
5685 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 5686 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 5687 | if (m->out) { |
395e68ce | 5688 | output_normal(ctx, m->out, vlan); |
c06bba01 | 5689 | } else if (eth_dst_may_rspan(orig_flow->dl_dst) |
9ba15e2a BP |
5690 | && vlan != m->out_vlan) { |
5691 | struct ofbundle *bundle; | |
5692 | ||
5693 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
5694 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
5695 | && !bundle->mirror_out) { |
5696 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
5697 | } |
5698 | } | |
5699 | } | |
abe529af BP |
5700 | } |
5701 | } | |
5702 | ||
9d24de3b JP |
5703 | static void |
5704 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
5705 | uint64_t packets, uint64_t bytes) | |
5706 | { | |
5707 | if (!mirrors) { | |
5708 | return; | |
5709 | } | |
5710 | ||
5711 | for (; mirrors; mirrors &= mirrors - 1) { | |
5712 | struct ofmirror *m; | |
5713 | ||
5714 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5715 | ||
5716 | if (!m) { | |
5717 | /* In normal circumstances 'm' will not be NULL. However, | |
5718 | * if mirrors are reconfigured, we can temporarily get out | |
5719 | * of sync in facet_revalidate(). We could "correct" the | |
5720 | * mirror list before reaching here, but doing that would | |
5721 | * not properly account the traffic stats we've currently | |
5722 | * accumulated for previous mirror configuration. */ | |
5723 | continue; | |
5724 | } | |
5725 | ||
5726 | m->packet_count += packets; | |
5727 | m->byte_count += bytes; | |
5728 | } | |
5729 | } | |
5730 | ||
abe529af BP |
5731 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
5732 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
5733 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
5734 | static bool | |
5735 | is_gratuitous_arp(const struct flow *flow) | |
5736 | { | |
5737 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
5738 | && eth_addr_is_broadcast(flow->dl_dst) | |
5739 | && (flow->nw_proto == ARP_OP_REPLY | |
5740 | || (flow->nw_proto == ARP_OP_REQUEST | |
5741 | && flow->nw_src == flow->nw_dst))); | |
5742 | } | |
5743 | ||
5744 | static void | |
5745 | update_learning_table(struct ofproto_dpif *ofproto, | |
5746 | const struct flow *flow, int vlan, | |
5747 | struct ofbundle *in_bundle) | |
5748 | { | |
5749 | struct mac_entry *mac; | |
5750 | ||
33158a18 JP |
5751 | /* Don't learn the OFPP_NONE port. */ |
5752 | if (in_bundle == &ofpp_none_bundle) { | |
5753 | return; | |
5754 | } | |
5755 | ||
abe529af BP |
5756 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
5757 | return; | |
5758 | } | |
5759 | ||
5760 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
5761 | if (is_gratuitous_arp(flow)) { | |
5762 | /* We don't want to learn from gratuitous ARP packets that are | |
5763 | * reflected back over bond slaves so we lock the learning table. */ | |
5764 | if (!in_bundle->bond) { | |
5765 | mac_entry_set_grat_arp_lock(mac); | |
5766 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
5767 | return; | |
5768 | } | |
5769 | } | |
5770 | ||
5771 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
5772 | /* The log messages here could actually be useful in debugging, | |
5773 | * so keep the rate limit relatively high. */ | |
5774 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5775 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
5776 | "on port %s in VLAN %d", | |
5777 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
5778 | in_bundle->name, vlan); | |
5779 | ||
5780 | mac->port.p = in_bundle; | |
5781 | tag_set_add(&ofproto->revalidate_set, | |
5782 | mac_learning_changed(ofproto->ml, mac)); | |
5783 | } | |
5784 | } | |
5785 | ||
3581c12c | 5786 | static struct ofbundle * |
395e68ce BP |
5787 | lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) |
5788 | { | |
5789 | struct ofport_dpif *ofport; | |
5790 | ||
33158a18 JP |
5791 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
5792 | * port for traffic that it is sourcing. */ | |
5793 | if (in_port == OFPP_NONE) { | |
5794 | return &ofpp_none_bundle; | |
5795 | } | |
5796 | ||
395e68ce BP |
5797 | /* Find the port and bundle for the received packet. */ |
5798 | ofport = get_ofp_port(ofproto, in_port); | |
5799 | if (ofport && ofport->bundle) { | |
3581c12c | 5800 | return ofport->bundle; |
395e68ce BP |
5801 | } |
5802 | ||
5803 | /* Odd. A few possible reasons here: | |
5804 | * | |
5805 | * - We deleted a port but there are still a few packets queued up | |
5806 | * from it. | |
5807 | * | |
5808 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
5809 | * we don't know about. | |
5810 | * | |
5811 | * - The ofproto client didn't configure the port as part of a bundle. | |
5812 | */ | |
5813 | if (warn) { | |
5814 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5815 | ||
5816 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
5817 | "port %"PRIu16, ofproto->up.name, in_port); | |
5818 | } | |
5819 | return NULL; | |
5820 | } | |
5821 | ||
5da5ec37 | 5822 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
5823 | * dropped. Returns true if they may be forwarded, false if they should be |
5824 | * dropped. | |
5825 | * | |
395e68ce BP |
5826 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
5827 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 5828 | * |
395e68ce BP |
5829 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
5830 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
5831 | * checked by input_vid_is_valid(). | |
abe529af BP |
5832 | * |
5833 | * May also add tags to '*tags', although the current implementation only does | |
5834 | * so in one special case. | |
5835 | */ | |
5836 | static bool | |
5837 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
395e68ce | 5838 | struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) |
abe529af | 5839 | { |
395e68ce | 5840 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 5841 | |
395e68ce BP |
5842 | /* Drop frames for reserved multicast addresses |
5843 | * only if forward_bpdu option is absent. */ | |
21f7563c | 5844 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { |
abe529af BP |
5845 | return false; |
5846 | } | |
5847 | ||
abe529af BP |
5848 | if (in_bundle->bond) { |
5849 | struct mac_entry *mac; | |
5850 | ||
5851 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
5852 | flow->dl_dst, tags)) { | |
5853 | case BV_ACCEPT: | |
5854 | break; | |
5855 | ||
5856 | case BV_DROP: | |
5857 | return false; | |
5858 | ||
5859 | case BV_DROP_IF_MOVED: | |
5860 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
5861 | if (mac && mac->port.p != in_bundle && | |
5862 | (!is_gratuitous_arp(flow) | |
5863 | || mac_entry_is_grat_arp_locked(mac))) { | |
5864 | return false; | |
5865 | } | |
5866 | break; | |
5867 | } | |
5868 | } | |
5869 | ||
5870 | return true; | |
5871 | } | |
5872 | ||
4cd78906 | 5873 | static void |
abe529af BP |
5874 | xlate_normal(struct action_xlate_ctx *ctx) |
5875 | { | |
395e68ce | 5876 | struct ofport_dpif *in_port; |
abe529af | 5877 | struct ofbundle *in_bundle; |
abe529af | 5878 | struct mac_entry *mac; |
395e68ce BP |
5879 | uint16_t vlan; |
5880 | uint16_t vid; | |
abe529af | 5881 | |
75a75043 BP |
5882 | ctx->has_normal = true; |
5883 | ||
3581c12c | 5884 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
395e68ce | 5885 | ctx->packet != NULL); |
3581c12c | 5886 | if (!in_bundle) { |
395e68ce BP |
5887 | return; |
5888 | } | |
3581c12c | 5889 | |
33158a18 JP |
5890 | /* We know 'in_port' exists unless it is "ofpp_none_bundle", |
5891 | * since lookup_input_bundle() succeeded. */ | |
3581c12c | 5892 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
395e68ce BP |
5893 | |
5894 | /* Drop malformed frames. */ | |
5895 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
5896 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
5897 | if (ctx->packet != NULL) { | |
5898 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5899 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
5900 | "VLAN tag received on port %s", | |
5901 | ctx->ofproto->up.name, in_bundle->name); | |
5902 | } | |
5903 | return; | |
5904 | } | |
5905 | ||
5906 | /* Drop frames on bundles reserved for mirroring. */ | |
5907 | if (in_bundle->mirror_out) { | |
5908 | if (ctx->packet != NULL) { | |
5909 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5910 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5911 | "%s, which is reserved exclusively for mirroring", | |
5912 | ctx->ofproto->up.name, in_bundle->name); | |
5913 | } | |
5914 | return; | |
5915 | } | |
5916 | ||
5917 | /* Check VLAN. */ | |
5918 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
5919 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5920 | return; | |
5921 | } | |
5922 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5923 | ||
5924 | /* Check other admissibility requirements. */ | |
33158a18 JP |
5925 | if (in_port && |
5926 | !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { | |
395e68ce | 5927 | return; |
abe529af BP |
5928 | } |
5929 | ||
75a75043 | 5930 | /* Learn source MAC. */ |
3de9590b | 5931 | if (ctx->may_learn) { |
abe529af BP |
5932 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
5933 | } | |
5934 | ||
5935 | /* Determine output bundle. */ | |
5936 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
5937 | &ctx->tags); | |
5938 | if (mac) { | |
c06bba01 JP |
5939 | if (mac->port.p != in_bundle) { |
5940 | output_normal(ctx, mac->port.p, vlan); | |
5941 | } | |
abe529af | 5942 | } else { |
c06bba01 | 5943 | struct ofbundle *bundle; |
abe529af | 5944 | |
c06bba01 JP |
5945 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
5946 | if (bundle != in_bundle | |
5947 | && ofbundle_includes_vlan(bundle, vlan) | |
5948 | && bundle->floodable | |
5949 | && !bundle->mirror_out) { | |
5950 | output_normal(ctx, bundle, vlan); | |
5951 | } | |
5952 | } | |
5953 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 5954 | } |
abe529af BP |
5955 | } |
5956 | \f | |
54a9cbc9 BP |
5957 | /* Optimized flow revalidation. |
5958 | * | |
5959 | * It's a difficult problem, in general, to tell which facets need to have | |
5960 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
5961 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
5962 | * table changes, we recalculate the actions for every facet. This is | |
5963 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
5964 | * doesn't change very often. | |
5965 | * | |
5966 | * However, we can expect one particular kind of OpenFlow flow table change to | |
5967 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
5968 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
5969 | * table, we add a special case that applies to flow tables in which every rule | |
5970 | * has the same form (that is, the same wildcards), except that the table is | |
5971 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
5972 | * optimize this case by tagging all of the facets that resubmit into the table | |
5973 | * and invalidating the same tag whenever a flow changes in that table. The | |
5974 | * end result is that we revalidate just the facets that need it (and sometimes | |
5975 | * a few more, but not all of the facets or even all of the facets that | |
5976 | * resubmit to the table modified by MAC learning). */ | |
5977 | ||
5978 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
5979 | * into an OpenFlow table with the given 'basis'. */ | |
822d9414 | 5980 | static tag_type |
54a9cbc9 BP |
5981 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, |
5982 | uint32_t secret) | |
5983 | { | |
5984 | if (flow_wildcards_is_catchall(wc)) { | |
5985 | return 0; | |
5986 | } else { | |
5987 | struct flow tag_flow = *flow; | |
5988 | flow_zero_wildcards(&tag_flow, wc); | |
5989 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
5990 | } | |
5991 | } | |
5992 | ||
5993 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
5994 | * taggability of that table. | |
5995 | * | |
5996 | * This function must be called after *each* change to a flow table. If you | |
5997 | * skip calling it on some changes then the pointer comparisons at the end can | |
5998 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
5999 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
6000 | * different wildcards to be created with the same address, then this function | |
6001 | * will incorrectly skip revalidation. */ | |
6002 | static void | |
6003 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
6004 | { | |
6005 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 6006 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
6007 | struct cls_table *catchall, *other; |
6008 | struct cls_table *t; | |
6009 | ||
6010 | catchall = other = NULL; | |
6011 | ||
d0918789 | 6012 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
6013 | case 0: |
6014 | /* We could tag this OpenFlow table but it would make the logic a | |
6015 | * little harder and it's a corner case that doesn't seem worth it | |
6016 | * yet. */ | |
6017 | break; | |
6018 | ||
6019 | case 1: | |
6020 | case 2: | |
d0918789 | 6021 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
6022 | if (cls_table_is_catchall(t)) { |
6023 | catchall = t; | |
6024 | } else if (!other) { | |
6025 | other = t; | |
6026 | } else { | |
6027 | /* Indicate that we can't tag this by setting both tables to | |
6028 | * NULL. (We know that 'catchall' is already NULL.) */ | |
6029 | other = NULL; | |
6030 | } | |
6031 | } | |
6032 | break; | |
6033 | ||
6034 | default: | |
6035 | /* Can't tag this table. */ | |
6036 | break; | |
6037 | } | |
6038 | ||
6039 | if (table->catchall_table != catchall || table->other_table != other) { | |
6040 | table->catchall_table = catchall; | |
6041 | table->other_table = other; | |
6042 | ofproto->need_revalidate = true; | |
6043 | } | |
6044 | } | |
6045 | ||
6046 | /* Given 'rule' that has changed in some way (either it is a rule being | |
6047 | * inserted, a rule being deleted, or a rule whose actions are being | |
6048 | * modified), marks facets for revalidation to ensure that packets will be | |
6049 | * forwarded correctly according to the new state of the flow table. | |
6050 | * | |
6051 | * This function must be called after *each* change to a flow table. See | |
6052 | * the comment on table_update_taggable() for more information. */ | |
6053 | static void | |
6054 | rule_invalidate(const struct rule_dpif *rule) | |
6055 | { | |
6056 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
6057 | ||
6058 | table_update_taggable(ofproto, rule->up.table_id); | |
6059 | ||
6060 | if (!ofproto->need_revalidate) { | |
6061 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
6062 | ||
6063 | if (table->other_table && rule->tag) { | |
6064 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
6065 | } else { | |
6066 | ofproto->need_revalidate = true; | |
6067 | } | |
6068 | } | |
6069 | } | |
6070 | \f | |
abe529af | 6071 | static bool |
7257b535 BP |
6072 | set_frag_handling(struct ofproto *ofproto_, |
6073 | enum ofp_config_flags frag_handling) | |
abe529af BP |
6074 | { |
6075 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 6076 | |
7257b535 BP |
6077 | if (frag_handling != OFPC_FRAG_REASM) { |
6078 | ofproto->need_revalidate = true; | |
6079 | return true; | |
6080 | } else { | |
6081 | return false; | |
6082 | } | |
abe529af BP |
6083 | } |
6084 | ||
90bf1e07 | 6085 | static enum ofperr |
abe529af BP |
6086 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
6087 | const struct flow *flow, | |
6088 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
6089 | { | |
6090 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
90bf1e07 | 6091 | enum ofperr error; |
abe529af | 6092 | |
e1154f71 | 6093 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
90bf1e07 | 6094 | return OFPERR_NXBRC_BAD_IN_PORT; |
e1154f71 BP |
6095 | } |
6096 | ||
abe529af BP |
6097 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
6098 | ofproto->max_ports); | |
6099 | if (!error) { | |
80e5eed9 | 6100 | struct odputil_keybuf keybuf; |
112bc5f4 BP |
6101 | struct dpif_flow_stats stats; |
6102 | ||
80e5eed9 BP |
6103 | struct ofpbuf key; |
6104 | ||
112bc5f4 | 6105 | struct action_xlate_ctx ctx; |
050ac423 BP |
6106 | uint64_t odp_actions_stub[1024 / 8]; |
6107 | struct ofpbuf odp_actions; | |
050ac423 | 6108 | |
80e5eed9 BP |
6109 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
6110 | odp_flow_key_from_flow(&key, flow); | |
abe529af | 6111 | |
112bc5f4 | 6112 | dpif_flow_stats_extract(flow, packet, &stats); |
2284188b | 6113 | |
112bc5f4 BP |
6114 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, |
6115 | packet_get_tcp_flags(packet, flow), packet); | |
6116 | ctx.resubmit_stats = &stats; | |
2284188b | 6117 | |
050ac423 BP |
6118 | ofpbuf_use_stub(&odp_actions, |
6119 | odp_actions_stub, sizeof odp_actions_stub); | |
112bc5f4 | 6120 | xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions); |
80e5eed9 | 6121 | dpif_execute(ofproto->dpif, key.data, key.size, |
050ac423 BP |
6122 | odp_actions.data, odp_actions.size, packet); |
6123 | ofpbuf_uninit(&odp_actions); | |
abe529af BP |
6124 | } |
6125 | return error; | |
6126 | } | |
6fca1ffb BP |
6127 | \f |
6128 | /* NetFlow. */ | |
6129 | ||
6130 | static int | |
6131 | set_netflow(struct ofproto *ofproto_, | |
6132 | const struct netflow_options *netflow_options) | |
6133 | { | |
6134 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6135 | ||
6136 | if (netflow_options) { | |
6137 | if (!ofproto->netflow) { | |
6138 | ofproto->netflow = netflow_create(); | |
6139 | } | |
6140 | return netflow_set_options(ofproto->netflow, netflow_options); | |
6141 | } else { | |
6142 | netflow_destroy(ofproto->netflow); | |
6143 | ofproto->netflow = NULL; | |
6144 | return 0; | |
6145 | } | |
6146 | } | |
abe529af BP |
6147 | |
6148 | static void | |
6149 | get_netflow_ids(const struct ofproto *ofproto_, | |
6150 | uint8_t *engine_type, uint8_t *engine_id) | |
6151 | { | |
6152 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6153 | ||
6154 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
6155 | } | |
6fca1ffb BP |
6156 | |
6157 | static void | |
6158 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
6159 | { | |
6160 | if (!facet_is_controller_flow(facet) && | |
6161 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 6162 | struct subfacet *subfacet; |
6fca1ffb BP |
6163 | struct ofexpired expired; |
6164 | ||
b0f7b9b5 BP |
6165 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6166 | if (subfacet->installed) { | |
6167 | struct dpif_flow_stats stats; | |
6fca1ffb | 6168 | |
15baa734 | 6169 | subfacet_install(subfacet, subfacet->actions, |
b95fc6ba | 6170 | subfacet->actions_len, &stats); |
15baa734 | 6171 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 6172 | } |
6fca1ffb BP |
6173 | } |
6174 | ||
6175 | expired.flow = facet->flow; | |
6176 | expired.packet_count = facet->packet_count; | |
6177 | expired.byte_count = facet->byte_count; | |
6178 | expired.used = facet->used; | |
6179 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
6180 | } | |
6181 | } | |
6182 | ||
6183 | static void | |
6184 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
6185 | { | |
6186 | struct facet *facet; | |
6187 | ||
6188 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6189 | send_active_timeout(ofproto, facet); | |
6190 | } | |
6191 | } | |
abe529af BP |
6192 | \f |
6193 | static struct ofproto_dpif * | |
6194 | ofproto_dpif_lookup(const char *name) | |
6195 | { | |
b44a10b7 BP |
6196 | struct ofproto_dpif *ofproto; |
6197 | ||
6198 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
6199 | hash_string(name, 0), &all_ofproto_dpifs) { | |
6200 | if (!strcmp(ofproto->up.name, name)) { | |
6201 | return ofproto; | |
6202 | } | |
6203 | } | |
6204 | return NULL; | |
abe529af BP |
6205 | } |
6206 | ||
f0a3aa2e | 6207 | static void |
96e466a3 | 6208 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 6209 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 6210 | { |
490df1ef | 6211 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 6212 | |
96e466a3 EJ |
6213 | if (argc > 1) { |
6214 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6215 | if (!ofproto) { | |
bde9f75d | 6216 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
6217 | return; |
6218 | } | |
d0040604 | 6219 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 EJ |
6220 | } else { |
6221 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
d0040604 | 6222 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 | 6223 | } |
f0a3aa2e | 6224 | } |
f0a3aa2e | 6225 | |
bde9f75d | 6226 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
6227 | } |
6228 | ||
abe529af | 6229 | static void |
0e15264f BP |
6230 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6231 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
6232 | { |
6233 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6234 | const struct ofproto_dpif *ofproto; | |
6235 | const struct mac_entry *e; | |
6236 | ||
0e15264f | 6237 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 6238 | if (!ofproto) { |
bde9f75d | 6239 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
6240 | return; |
6241 | } | |
6242 | ||
6243 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
6244 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
6245 | struct ofbundle *bundle = e->port.p; | |
6246 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
6247 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
6248 | e->vlan, ETH_ADDR_ARGS(e->mac), |
6249 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 6250 | } |
bde9f75d | 6251 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
6252 | ds_destroy(&ds); |
6253 | } | |
6254 | ||
6a6455e5 | 6255 | struct trace_ctx { |
abe529af BP |
6256 | struct action_xlate_ctx ctx; |
6257 | struct flow flow; | |
6258 | struct ds *result; | |
6259 | }; | |
6260 | ||
6261 | static void | |
29901626 BP |
6262 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
6263 | const struct rule_dpif *rule) | |
abe529af BP |
6264 | { |
6265 | ds_put_char_multiple(result, '\t', level); | |
6266 | if (!rule) { | |
6267 | ds_put_cstr(result, "No match\n"); | |
6268 | return; | |
6269 | } | |
6270 | ||
29901626 BP |
6271 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
6272 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 6273 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
6274 | ds_put_char(result, '\n'); |
6275 | ||
6276 | ds_put_char_multiple(result, '\t', level); | |
6277 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 6278 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
6279 | ds_put_char(result, '\n'); |
6280 | } | |
6281 | ||
6282 | static void | |
6283 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 6284 | struct trace_ctx *trace) |
abe529af BP |
6285 | { |
6286 | ds_put_char_multiple(result, '\t', level); | |
6287 | ds_put_format(result, "%s: ", title); | |
6288 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
6289 | ds_put_cstr(result, "unchanged"); | |
6290 | } else { | |
6291 | flow_format(result, &trace->ctx.flow); | |
6292 | trace->flow = trace->ctx.flow; | |
6293 | } | |
6294 | ds_put_char(result, '\n'); | |
6295 | } | |
6296 | ||
eb9e1c26 EJ |
6297 | static void |
6298 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 6299 | struct trace_ctx *trace) |
eb9e1c26 EJ |
6300 | { |
6301 | size_t i; | |
6302 | ||
6303 | ds_put_char_multiple(result, '\t', level); | |
6304 | ds_put_format(result, "%s:", title); | |
6305 | for (i = 0; i < FLOW_N_REGS; i++) { | |
6306 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
6307 | } | |
6308 | ds_put_char(result, '\n'); | |
6309 | } | |
6310 | ||
1ed8d352 EJ |
6311 | static void |
6312 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 6313 | struct trace_ctx *trace) |
1ed8d352 EJ |
6314 | { |
6315 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
6316 | ||
6317 | ds_put_char_multiple(result, '\t', level); | |
6318 | ds_put_format(result, "%s: ", title); | |
6319 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
6320 | ds_put_char(result, '\n'); | |
6321 | } | |
6322 | ||
abe529af BP |
6323 | static void |
6324 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6325 | { | |
6a6455e5 | 6326 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
6327 | struct ds *result = trace->result; |
6328 | ||
6329 | ds_put_char(result, '\n'); | |
6330 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 6331 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 6332 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 6333 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
6334 | } |
6335 | ||
6336 | static void | |
0e15264f | 6337 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
6338 | void *aux OVS_UNUSED) |
6339 | { | |
0e15264f | 6340 | const char *dpname = argv[1]; |
abe529af | 6341 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
6342 | struct ofpbuf odp_key; |
6343 | struct ofpbuf *packet; | |
e84173dc | 6344 | ovs_be16 initial_tci; |
abe529af BP |
6345 | struct ds result; |
6346 | struct flow flow; | |
abe529af BP |
6347 | char *s; |
6348 | ||
876b0e1c BP |
6349 | packet = NULL; |
6350 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
6351 | ds_init(&result); |
6352 | ||
e84173dc BP |
6353 | ofproto = ofproto_dpif_lookup(dpname); |
6354 | if (!ofproto) { | |
bde9f75d EJ |
6355 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
6356 | "for help)"); | |
e84173dc BP |
6357 | goto exit; |
6358 | } | |
0e15264f | 6359 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 6360 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
6361 | const char *flow_s = argv[2]; |
6362 | const char *generate_s = argv[3]; | |
876b0e1c BP |
6363 | int error; |
6364 | ||
df2c07f4 | 6365 | /* Convert string to datapath key. */ |
876b0e1c | 6366 | ofpbuf_init(&odp_key, 0); |
0e15264f | 6367 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); |
876b0e1c | 6368 | if (error) { |
bde9f75d | 6369 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
876b0e1c BP |
6370 | goto exit; |
6371 | } | |
6372 | ||
6373 | /* Convert odp_key to flow. */ | |
e84173dc BP |
6374 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, |
6375 | odp_key.size, &flow, | |
e2a6ca36 | 6376 | &initial_tci, NULL); |
e84173dc | 6377 | if (error == ODP_FIT_ERROR) { |
bde9f75d | 6378 | unixctl_command_reply_error(conn, "Invalid flow"); |
876b0e1c BP |
6379 | goto exit; |
6380 | } | |
8b3b8dd1 BP |
6381 | |
6382 | /* Generate a packet, if requested. */ | |
0e15264f | 6383 | if (generate_s) { |
8b3b8dd1 BP |
6384 | packet = ofpbuf_new(0); |
6385 | flow_compose(packet, &flow); | |
6386 | } | |
0e15264f | 6387 | } else if (argc == 6) { |
abff858b | 6388 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
6389 | const char *priority_s = argv[2]; |
6390 | const char *tun_id_s = argv[3]; | |
6391 | const char *in_port_s = argv[4]; | |
6392 | const char *packet_s = argv[5]; | |
6393 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
6394 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
6395 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 6396 | const char *msg; |
0e15264f | 6397 | |
e22f1753 BP |
6398 | msg = eth_from_hex(packet_s, &packet); |
6399 | if (msg) { | |
bde9f75d | 6400 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
6401 | goto exit; |
6402 | } | |
6403 | ||
6404 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 6405 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
6406 | ds_put_cstr(&result, s); |
6407 | free(s); | |
6408 | ||
abff858b | 6409 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 6410 | initial_tci = flow.vlan_tci; |
876b0e1c | 6411 | } else { |
bde9f75d | 6412 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
6413 | goto exit; |
6414 | } | |
6415 | ||
6a6455e5 EJ |
6416 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); |
6417 | unixctl_command_reply(conn, ds_cstr(&result)); | |
6418 | ||
6419 | exit: | |
6420 | ds_destroy(&result); | |
6421 | ofpbuf_delete(packet); | |
6422 | ofpbuf_uninit(&odp_key); | |
6423 | } | |
6424 | ||
6425 | static void | |
6426 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
6427 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
6428 | struct ds *ds) | |
6429 | { | |
6430 | struct rule_dpif *rule; | |
6431 | ||
6432 | ds_put_cstr(ds, "Flow: "); | |
6433 | flow_format(ds, flow); | |
6434 | ds_put_char(ds, '\n'); | |
abe529af | 6435 | |
6a6455e5 EJ |
6436 | rule = rule_dpif_lookup(ofproto, flow, 0); |
6437 | trace_format_rule(ds, 0, 0, rule); | |
abe529af | 6438 | if (rule) { |
050ac423 BP |
6439 | uint64_t odp_actions_stub[1024 / 8]; |
6440 | struct ofpbuf odp_actions; | |
6441 | ||
6a6455e5 | 6442 | struct trace_ctx trace; |
0e553d9c | 6443 | uint8_t tcp_flags; |
abe529af | 6444 | |
6a6455e5 EJ |
6445 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
6446 | trace.result = ds; | |
6447 | trace.flow = *flow; | |
050ac423 BP |
6448 | ofpbuf_use_stub(&odp_actions, |
6449 | odp_actions_stub, sizeof odp_actions_stub); | |
6a6455e5 | 6450 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, |
0e553d9c | 6451 | rule, tcp_flags, packet); |
abe529af | 6452 | trace.ctx.resubmit_hook = trace_resubmit; |
050ac423 BP |
6453 | xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions, |
6454 | &odp_actions); | |
abe529af | 6455 | |
6a6455e5 EJ |
6456 | ds_put_char(ds, '\n'); |
6457 | trace_format_flow(ds, 0, "Final flow", &trace); | |
6458 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
6459 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
6460 | ofpbuf_uninit(&odp_actions); | |
876b0e1c BP |
6461 | |
6462 | if (!trace.ctx.may_set_up_flow) { | |
6463 | if (packet) { | |
6a6455e5 | 6464 | ds_put_cstr(ds, "\nThis flow is not cachable."); |
876b0e1c | 6465 | } else { |
6a6455e5 | 6466 | ds_put_cstr(ds, "\nThe datapath actions are incomplete--" |
876b0e1c BP |
6467 | "for complete actions, please supply a packet."); |
6468 | } | |
6469 | } | |
abe529af | 6470 | } |
abe529af BP |
6471 | } |
6472 | ||
7ee20df1 | 6473 | static void |
0e15264f BP |
6474 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6475 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6476 | { |
6477 | clogged = true; | |
bde9f75d | 6478 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6479 | } |
6480 | ||
6481 | static void | |
0e15264f BP |
6482 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6483 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6484 | { |
6485 | clogged = false; | |
bde9f75d | 6486 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6487 | } |
6488 | ||
6814e51f BP |
6489 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6490 | * 'reply' describing the results. */ | |
6491 | static void | |
6492 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6493 | { | |
6494 | struct facet *facet; | |
6495 | int errors; | |
6496 | ||
6497 | errors = 0; | |
6498 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6499 | if (!facet_check_consistency(facet)) { | |
6500 | errors++; | |
6501 | } | |
6502 | } | |
6503 | if (errors) { | |
6504 | ofproto->need_revalidate = true; | |
6505 | } | |
6506 | ||
6507 | if (errors) { | |
6508 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6509 | ofproto->up.name, errors); | |
6510 | } else { | |
6511 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6512 | } | |
6513 | } | |
6514 | ||
6515 | static void | |
6516 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6517 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6518 | { | |
6519 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6520 | struct ofproto_dpif *ofproto; | |
6521 | ||
6522 | if (argc > 1) { | |
6523 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6524 | if (!ofproto) { | |
bde9f75d EJ |
6525 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
6526 | "ofproto/list for help)"); | |
6814e51f BP |
6527 | return; |
6528 | } | |
6529 | ofproto_dpif_self_check__(ofproto, &reply); | |
6530 | } else { | |
6531 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6532 | ofproto_dpif_self_check__(ofproto, &reply); | |
6533 | } | |
6534 | } | |
6535 | ||
bde9f75d | 6536 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
6537 | ds_destroy(&reply); |
6538 | } | |
6539 | ||
abe529af BP |
6540 | static void |
6541 | ofproto_dpif_unixctl_init(void) | |
6542 | { | |
6543 | static bool registered; | |
6544 | if (registered) { | |
6545 | return; | |
6546 | } | |
6547 | registered = true; | |
6548 | ||
0e15264f BP |
6549 | unixctl_command_register( |
6550 | "ofproto/trace", | |
6551 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
aa3080c9 | 6552 | 2, 5, ofproto_unixctl_trace, NULL); |
96e466a3 | 6553 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6554 | ofproto_unixctl_fdb_flush, NULL); |
6555 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6556 | ofproto_unixctl_fdb_show, NULL); | |
6557 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6558 | ofproto_dpif_clog, NULL); | |
6559 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6560 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6561 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6562 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6563 | } |
6564 | \f | |
52a90c29 BP |
6565 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6566 | * | |
6567 | * This is deprecated. It is only for compatibility with broken device drivers | |
6568 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6569 | * devices are not used. When broken device drivers are no longer in | |
6570 | * widespread use, we will delete these interfaces. */ | |
6571 | ||
6572 | static int | |
6573 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6574 | { | |
6575 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6576 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6577 | ||
6578 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6579 | && vid == ofport->vlandev_vid) { | |
6580 | return 0; | |
6581 | } | |
6582 | ||
6583 | ofproto->need_revalidate = true; | |
6584 | ||
6585 | if (ofport->realdev_ofp_port) { | |
6586 | vsp_remove(ofport); | |
6587 | } | |
6588 | if (realdev_ofp_port && ofport->bundle) { | |
6589 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6590 | * themselves be part of a bundle. */ | |
6591 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6592 | } | |
6593 | ||
6594 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6595 | ofport->vlandev_vid = vid; | |
6596 | ||
6597 | if (realdev_ofp_port) { | |
6598 | vsp_add(ofport, realdev_ofp_port, vid); | |
6599 | } | |
6600 | ||
6601 | return 0; | |
6602 | } | |
6603 | ||
6604 | static uint32_t | |
6605 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
6606 | { | |
6607 | return hash_2words(realdev_ofp_port, vid); | |
6608 | } | |
6609 | ||
40e05935 BP |
6610 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
6611 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
6612 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
6613 | * it would return the port number of eth0.9. | |
6614 | * | |
6615 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
6616 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
6617 | static uint32_t |
6618 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6619 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
6620 | { | |
6621 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
6622 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
6623 | int vid = vlan_tci_to_vid(vlan_tci); | |
6624 | const struct vlan_splinter *vsp; | |
6625 | ||
6626 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6627 | hash_realdev_vid(realdev_ofp_port, vid), | |
6628 | &ofproto->realdev_vid_map) { | |
6629 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6630 | && vsp->vid == vid) { | |
6631 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
6632 | } | |
6633 | } | |
6634 | } | |
6635 | return realdev_odp_port; | |
6636 | } | |
6637 | ||
6638 | static struct vlan_splinter * | |
6639 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
6640 | { | |
6641 | struct vlan_splinter *vsp; | |
6642 | ||
6643 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
6644 | &ofproto->vlandev_map) { | |
6645 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6646 | return vsp; | |
6647 | } | |
6648 | } | |
6649 | ||
6650 | return NULL; | |
6651 | } | |
6652 | ||
40e05935 BP |
6653 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
6654 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
6655 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
6656 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
6657 | * eth0 and store 9 in '*vid'. | |
6658 | * | |
6659 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
6660 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
6661 | * always does.*/ | |
52a90c29 BP |
6662 | static uint16_t |
6663 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 6664 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
6665 | { |
6666 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6667 | const struct vlan_splinter *vsp; | |
6668 | ||
6669 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6670 | if (vsp) { | |
6671 | if (vid) { | |
6672 | *vid = vsp->vid; | |
6673 | } | |
6674 | return vsp->realdev_ofp_port; | |
6675 | } | |
6676 | } | |
6677 | return 0; | |
6678 | } | |
6679 | ||
6680 | static void | |
6681 | vsp_remove(struct ofport_dpif *port) | |
6682 | { | |
6683 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6684 | struct vlan_splinter *vsp; | |
6685 | ||
6686 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
6687 | if (vsp) { | |
6688 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6689 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6690 | free(vsp); | |
6691 | ||
6692 | port->realdev_ofp_port = 0; | |
6693 | } else { | |
6694 | VLOG_ERR("missing vlan device record"); | |
6695 | } | |
6696 | } | |
6697 | ||
6698 | static void | |
6699 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
6700 | { | |
6701 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6702 | ||
6703 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
6704 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
6705 | == realdev_ofp_port)) { | |
6706 | struct vlan_splinter *vsp; | |
6707 | ||
6708 | vsp = xmalloc(sizeof *vsp); | |
6709 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6710 | hash_int(port->up.ofp_port, 0)); | |
6711 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6712 | hash_realdev_vid(realdev_ofp_port, vid)); | |
6713 | vsp->realdev_ofp_port = realdev_ofp_port; | |
6714 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6715 | vsp->vid = vid; | |
6716 | ||
6717 | port->realdev_ofp_port = realdev_ofp_port; | |
6718 | } else { | |
6719 | VLOG_ERR("duplicate vlan device record"); | |
6720 | } | |
6721 | } | |
6722 | \f | |
abe529af BP |
6723 | const struct ofproto_class ofproto_dpif_class = { |
6724 | enumerate_types, | |
6725 | enumerate_names, | |
6726 | del, | |
6727 | alloc, | |
6728 | construct, | |
6729 | destruct, | |
6730 | dealloc, | |
6731 | run, | |
5fcc0d00 | 6732 | run_fast, |
abe529af BP |
6733 | wait, |
6734 | flush, | |
6c1491fb BP |
6735 | get_features, |
6736 | get_tables, | |
abe529af BP |
6737 | port_alloc, |
6738 | port_construct, | |
6739 | port_destruct, | |
6740 | port_dealloc, | |
6741 | port_modified, | |
6742 | port_reconfigured, | |
6743 | port_query_by_name, | |
6744 | port_add, | |
6745 | port_del, | |
6527c598 | 6746 | port_get_stats, |
abe529af BP |
6747 | port_dump_start, |
6748 | port_dump_next, | |
6749 | port_dump_done, | |
6750 | port_poll, | |
6751 | port_poll_wait, | |
6752 | port_is_lacp_current, | |
0ab6decf | 6753 | NULL, /* rule_choose_table */ |
abe529af BP |
6754 | rule_alloc, |
6755 | rule_construct, | |
6756 | rule_destruct, | |
6757 | rule_dealloc, | |
abe529af BP |
6758 | rule_get_stats, |
6759 | rule_execute, | |
6760 | rule_modify_actions, | |
7257b535 | 6761 | set_frag_handling, |
abe529af BP |
6762 | packet_out, |
6763 | set_netflow, | |
6764 | get_netflow_ids, | |
6765 | set_sflow, | |
6766 | set_cfm, | |
a5610457 | 6767 | get_cfm_fault, |
1de11730 | 6768 | get_cfm_remote_mpids, |
3967a833 | 6769 | get_cfm_health, |
21f7563c JP |
6770 | set_stp, |
6771 | get_stp_status, | |
6772 | set_stp_port, | |
6773 | get_stp_port_status, | |
8b36f51e | 6774 | set_queues, |
abe529af BP |
6775 | bundle_set, |
6776 | bundle_remove, | |
6777 | mirror_set, | |
9d24de3b | 6778 | mirror_get_stats, |
abe529af BP |
6779 | set_flood_vlans, |
6780 | is_mirror_output_bundle, | |
8402c74b | 6781 | forward_bpdu_changed, |
e764773c | 6782 | set_mac_idle_time, |
52a90c29 | 6783 | set_realdev, |
abe529af | 6784 | }; |