]>
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" | |
bae473fe | 46 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
47 | #include "poll-loop.h" |
48 | #include "timer.h" | |
6c1491fb | 49 | #include "unaligned.h" |
abe529af BP |
50 | #include "unixctl.h" |
51 | #include "vlan-bitmap.h" | |
52 | #include "vlog.h" | |
53 | ||
54 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
55 | ||
56 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
57 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
58 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
59 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
60 | COVERAGE_DEFINE(facet_changed_rule); | |
61 | COVERAGE_DEFINE(facet_invalidated); | |
62 | COVERAGE_DEFINE(facet_revalidate); | |
63 | COVERAGE_DEFINE(facet_unexpected); | |
64 | ||
29901626 | 65 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 66 | * flow translation. */ |
642a5c05 | 67 | #define MAX_RESUBMIT_RECURSION 32 |
abe529af | 68 | |
9cdaaebe BP |
69 | /* Number of implemented OpenFlow tables. */ |
70 | enum { N_TABLES = 255 }; | |
71 | BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255); | |
72 | ||
abe529af BP |
73 | struct ofport_dpif; |
74 | struct ofproto_dpif; | |
75 | ||
76 | struct rule_dpif { | |
77 | struct rule up; | |
78 | ||
79 | long long int used; /* Time last used; time created if not used. */ | |
80 | ||
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 | |
b0f7b9b5 BP |
110 | static void flow_push_stats(const struct rule_dpif *, const struct flow *, |
111 | uint64_t packets, uint64_t bytes, | |
112 | long long int used); | |
113 | ||
114 | static uint32_t rule_calculate_tag(const struct flow *, | |
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 BP |
149 | struct ofbundle { |
150 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
151 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
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. */ | |
166 | bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ | |
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); | |
195 | ||
5da5ec37 BP |
196 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
197 | ||
abe529af BP |
198 | struct action_xlate_ctx { |
199 | /* action_xlate_ctx_init() initializes these members. */ | |
200 | ||
201 | /* The ofproto. */ | |
202 | struct ofproto_dpif *ofproto; | |
203 | ||
204 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
205 | * this flow when actions change header fields. */ | |
206 | struct flow flow; | |
207 | ||
208 | /* The packet corresponding to 'flow', or a null pointer if we are | |
209 | * revalidating without a packet to refer to. */ | |
210 | const struct ofpbuf *packet; | |
211 | ||
75a75043 BP |
212 | /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We |
213 | * want to execute them if we are actually processing a packet, or if we | |
214 | * are accounting for packets that the datapath has processed, but not if | |
215 | * we are just revalidating. */ | |
216 | bool may_learn; | |
217 | ||
54834960 EJ |
218 | /* Cookie of the currently matching rule, or 0. */ |
219 | ovs_be64 cookie; | |
220 | ||
abe529af BP |
221 | /* If nonnull, called just before executing a resubmit action. |
222 | * | |
223 | * This is normally null so the client has to set it manually after | |
224 | * calling action_xlate_ctx_init(). */ | |
225 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
226 | ||
abe529af BP |
227 | /* xlate_actions() initializes and uses these members. The client might want |
228 | * to look at them after it returns. */ | |
229 | ||
230 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 231 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
232 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
233 | * be reassessed for every packet. */ | |
75a75043 BP |
234 | bool has_learn; /* Actions include NXAST_LEARN? */ |
235 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af | 236 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 237 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
238 | |
239 | /* xlate_actions() initializes and uses these members, but the client has no | |
240 | * reason to look at them. */ | |
241 | ||
242 | int recurse; /* Recursion level, via xlate_table_action. */ | |
b3e9b2ed | 243 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 244 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 245 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
246 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
247 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
248 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 249 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
250 | }; |
251 | ||
252 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
253 | struct ofproto_dpif *, const struct flow *, | |
54834960 EJ |
254 | ovs_be16 initial_tci, ovs_be64 cookie, |
255 | const struct ofpbuf *); | |
abe529af BP |
256 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, |
257 | const union ofp_action *in, size_t n_in); | |
258 | ||
b0f7b9b5 BP |
259 | /* An exact-match instantiation of an OpenFlow flow. |
260 | * | |
261 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
262 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
263 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
264 | * the facet. When the kernel module (or other dpif implementation) and Open | |
265 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
266 | * one subfacet per facet. If the dpif implementation supports more-specific | |
267 | * flow matching than userspace, however, a facet can have more than one | |
268 | * subfacet, each of which corresponds to some distinction in flow that | |
269 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
270 | * |
271 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
272 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 273 | struct facet { |
b0f7b9b5 BP |
274 | /* Owners. */ |
275 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
276 | struct list list_node; /* In owning rule's 'facets' list. */ | |
277 | struct rule_dpif *rule; /* Owning rule. */ | |
278 | ||
279 | /* Owned data. */ | |
280 | struct list subfacets; | |
abe529af BP |
281 | long long int used; /* Time last used; time created if not used. */ |
282 | ||
b0f7b9b5 BP |
283 | /* Key. */ |
284 | struct flow flow; | |
285 | ||
abe529af BP |
286 | /* These statistics: |
287 | * | |
288 | * - Do include packets and bytes sent "by hand", e.g. with | |
289 | * dpif_execute(). | |
290 | * | |
291 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 292 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 293 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
294 | * |
295 | * - Do not include packets or bytes that can be obtained from the | |
296 | * datapath for any existing subfacet. | |
abe529af BP |
297 | */ |
298 | uint64_t packet_count; /* Number of packets received. */ | |
299 | uint64_t byte_count; /* Number of bytes received. */ | |
300 | ||
b0f7b9b5 | 301 | /* Resubmit statistics. */ |
9d24de3b JP |
302 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
303 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
304 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 305 | |
b0f7b9b5 | 306 | /* Accounting. */ |
907a4c5e | 307 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 308 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
abe529af | 309 | |
b95fc6ba BP |
310 | /* Properties of datapath actions. |
311 | * | |
312 | * Every subfacet has its own actions because actions can differ slightly | |
313 | * between splintered and non-splintered subfacets due to the VLAN tag | |
314 | * being initially different (present vs. absent). All of them have these | |
315 | * properties in common so we just store one copy of them here. */ | |
b0f7b9b5 | 316 | bool may_install; /* Reassess actions for every packet? */ |
75a75043 BP |
317 | bool has_learn; /* Actions include NXAST_LEARN? */ |
318 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
b0f7b9b5 | 319 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 320 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
abe529af BP |
321 | }; |
322 | ||
f3827897 | 323 | static struct facet *facet_create(struct rule_dpif *, const struct flow *); |
15baa734 | 324 | static void facet_remove(struct facet *); |
abe529af BP |
325 | static void facet_free(struct facet *); |
326 | ||
327 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
328 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
329 | const struct flow *); | |
15baa734 | 330 | static bool facet_revalidate(struct facet *); |
6814e51f | 331 | static bool facet_check_consistency(struct facet *); |
abe529af | 332 | |
15baa734 | 333 | static void facet_flush_stats(struct facet *); |
abe529af | 334 | |
15baa734 | 335 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 336 | static void facet_reset_counters(struct facet *); |
abe529af | 337 | static void facet_push_stats(struct facet *); |
15baa734 | 338 | static void facet_account(struct facet *); |
abe529af BP |
339 | |
340 | static bool facet_is_controller_flow(struct facet *); | |
341 | ||
b95fc6ba | 342 | /* A dpif flow and actions associated with a facet. |
b0f7b9b5 BP |
343 | * |
344 | * See also the large comment on struct facet. */ | |
345 | struct subfacet { | |
346 | /* Owners. */ | |
347 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
348 | struct list list_node; /* In struct facet's 'facets' list. */ | |
349 | struct facet *facet; /* Owning facet. */ | |
350 | ||
351 | /* Key. | |
352 | * | |
353 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
354 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
355 | * regenerate the ODP flow key from ->facet->flow. */ | |
356 | enum odp_key_fitness key_fitness; | |
357 | struct nlattr *key; | |
358 | int key_len; | |
abe529af | 359 | |
b0f7b9b5 BP |
360 | long long int used; /* Time last used; time created if not used. */ |
361 | ||
362 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
363 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
364 | ||
b95fc6ba BP |
365 | /* Datapath actions. |
366 | * | |
367 | * These should be essentially identical for every subfacet in a facet, but | |
368 | * may differ in trivial ways due to VLAN splinters. */ | |
369 | size_t actions_len; /* Number of bytes in actions[]. */ | |
370 | struct nlattr *actions; /* Datapath actions. */ | |
371 | ||
b0f7b9b5 | 372 | bool installed; /* Installed in datapath? */ |
e84173dc BP |
373 | |
374 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
375 | * splinters can cause it to differ. This value should be removed when | |
376 | * the VLAN splinters feature is no longer needed. */ | |
377 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
b0f7b9b5 BP |
378 | }; |
379 | ||
15baa734 | 380 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, |
b0f7b9b5 | 381 | const struct nlattr *key, |
e84173dc | 382 | size_t key_len, ovs_be16 initial_tci); |
b0f7b9b5 | 383 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
6a542738 | 384 | const struct nlattr *key, size_t key_len); |
15baa734 BP |
385 | static void subfacet_destroy(struct subfacet *); |
386 | static void subfacet_destroy__(struct subfacet *); | |
6814e51f BP |
387 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, |
388 | struct ofpbuf *key); | |
b0f7b9b5 BP |
389 | static void subfacet_reset_dp_stats(struct subfacet *, |
390 | struct dpif_flow_stats *); | |
15baa734 BP |
391 | static void subfacet_update_time(struct subfacet *, long long int used); |
392 | static void subfacet_update_stats(struct subfacet *, | |
b0f7b9b5 | 393 | const struct dpif_flow_stats *); |
15baa734 | 394 | static void subfacet_make_actions(struct subfacet *, |
b95fc6ba | 395 | const struct ofpbuf *packet); |
15baa734 | 396 | static int subfacet_install(struct subfacet *, |
b0f7b9b5 BP |
397 | const struct nlattr *actions, size_t actions_len, |
398 | struct dpif_flow_stats *); | |
15baa734 | 399 | static void subfacet_uninstall(struct subfacet *); |
54a9cbc9 | 400 | |
abe529af BP |
401 | struct ofport_dpif { |
402 | struct ofport up; | |
403 | ||
404 | uint32_t odp_port; | |
405 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
406 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
407 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
408 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 409 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 410 | bool may_enable; /* May be enabled in bonds. */ |
21f7563c | 411 | |
52a90c29 | 412 | /* Spanning tree. */ |
21f7563c JP |
413 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
414 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
415 | long long int stp_state_entered; | |
8b36f51e EJ |
416 | |
417 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
418 | |
419 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
420 | * | |
421 | * This is deprecated. It is only for compatibility with broken device | |
422 | * drivers in old versions of Linux that do not properly support VLANs when | |
423 | * VLAN devices are not used. When broken device drivers are no longer in | |
424 | * widespread use, we will delete these interfaces. */ | |
425 | uint16_t realdev_ofp_port; | |
426 | int vlandev_vid; | |
8b36f51e EJ |
427 | }; |
428 | ||
429 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
430 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
431 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
432 | struct priority_to_dscp { | |
433 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
434 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
435 | ||
436 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
437 | }; |
438 | ||
52a90c29 BP |
439 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
440 | * | |
441 | * This is deprecated. It is only for compatibility with broken device drivers | |
442 | * in old versions of Linux that do not properly support VLANs when VLAN | |
443 | * devices are not used. When broken device drivers are no longer in | |
444 | * widespread use, we will delete these interfaces. */ | |
445 | struct vlan_splinter { | |
446 | struct hmap_node realdev_vid_node; | |
447 | struct hmap_node vlandev_node; | |
448 | uint16_t realdev_ofp_port; | |
449 | uint16_t vlandev_ofp_port; | |
450 | int vid; | |
451 | }; | |
452 | ||
453 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
454 | uint32_t realdev, ovs_be16 vlan_tci); | |
455 | static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *, | |
456 | uint16_t vlandev, int *vid); | |
457 | static void vsp_remove(struct ofport_dpif *); | |
458 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
459 | ||
abe529af BP |
460 | static struct ofport_dpif * |
461 | ofport_dpif_cast(const struct ofport *ofport) | |
462 | { | |
463 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
464 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
465 | } | |
466 | ||
467 | static void port_run(struct ofport_dpif *); | |
468 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 469 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 470 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 471 | |
7ee20df1 BP |
472 | struct dpif_completion { |
473 | struct list list_node; | |
474 | struct ofoperation *op; | |
475 | }; | |
476 | ||
54a9cbc9 BP |
477 | /* Extra information about a classifier table. |
478 | * Currently used just for optimized flow revalidation. */ | |
479 | struct table_dpif { | |
480 | /* If either of these is nonnull, then this table has a form that allows | |
481 | * flows to be tagged to avoid revalidating most flows for the most common | |
482 | * kinds of flow table changes. */ | |
483 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
484 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
485 | uint32_t basis; /* Keeps each table's tags separate. */ | |
486 | }; | |
487 | ||
abe529af | 488 | struct ofproto_dpif { |
b44a10b7 | 489 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
490 | struct ofproto up; |
491 | struct dpif *dpif; | |
492 | int max_ports; | |
493 | ||
6c1491fb BP |
494 | /* Statistics. */ |
495 | uint64_t n_matches; | |
496 | ||
abe529af BP |
497 | /* Bridging. */ |
498 | struct netflow *netflow; | |
bae473fe | 499 | struct dpif_sflow *sflow; |
abe529af BP |
500 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
501 | struct mac_learning *ml; | |
502 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
503 | bool has_bonded_bundles; | |
504 | ||
505 | /* Expiration. */ | |
506 | struct timer next_expiration; | |
507 | ||
508 | /* Facets. */ | |
509 | struct hmap facets; | |
b0f7b9b5 | 510 | struct hmap subfacets; |
54a9cbc9 BP |
511 | |
512 | /* Revalidation. */ | |
513 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
514 | bool need_revalidate; |
515 | struct tag_set revalidate_set; | |
7ee20df1 BP |
516 | |
517 | /* Support for debugging async flow mods. */ | |
518 | struct list completions; | |
daff3353 EJ |
519 | |
520 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
521 | struct netdev_stats stats; /* To account packets generated and consumed in |
522 | * userspace. */ | |
21f7563c JP |
523 | |
524 | /* Spanning tree. */ | |
525 | struct stp *stp; | |
526 | long long int stp_last_tick; | |
52a90c29 BP |
527 | |
528 | /* VLAN splinters. */ | |
529 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
530 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
531 | }; |
532 | ||
7ee20df1 BP |
533 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
534 | * for debugging the asynchronous flow_mod implementation.) */ | |
535 | static bool clogged; | |
536 | ||
b44a10b7 BP |
537 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
538 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
539 | ||
abe529af BP |
540 | static void ofproto_dpif_unixctl_init(void); |
541 | ||
542 | static struct ofproto_dpif * | |
543 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
544 | { | |
545 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
546 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
547 | } | |
548 | ||
549 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
550 | uint16_t ofp_port); | |
551 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
552 | uint32_t odp_port); | |
553 | ||
554 | /* Packet processing. */ | |
555 | static void update_learning_table(struct ofproto_dpif *, | |
556 | const struct flow *, int vlan, | |
557 | struct ofbundle *); | |
501f8d1f BP |
558 | /* Upcalls. */ |
559 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 560 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
561 | |
562 | /* Flow expiration. */ | |
563 | static int expire(struct ofproto_dpif *); | |
564 | ||
6fca1ffb BP |
565 | /* NetFlow. */ |
566 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
567 | ||
abe529af | 568 | /* Utilities. */ |
52a90c29 | 569 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6ff686f2 PS |
570 | static size_t |
571 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
572 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
573 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
574 | const struct flow *flow); | |
abe529af BP |
575 | /* Global variables. */ |
576 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
577 | \f | |
578 | /* Factory functions. */ | |
579 | ||
580 | static void | |
581 | enumerate_types(struct sset *types) | |
582 | { | |
583 | dp_enumerate_types(types); | |
584 | } | |
585 | ||
586 | static int | |
587 | enumerate_names(const char *type, struct sset *names) | |
588 | { | |
589 | return dp_enumerate_names(type, names); | |
590 | } | |
591 | ||
592 | static int | |
593 | del(const char *type, const char *name) | |
594 | { | |
595 | struct dpif *dpif; | |
596 | int error; | |
597 | ||
598 | error = dpif_open(name, type, &dpif); | |
599 | if (!error) { | |
600 | error = dpif_delete(dpif); | |
601 | dpif_close(dpif); | |
602 | } | |
603 | return error; | |
604 | } | |
605 | \f | |
606 | /* Basic life-cycle. */ | |
607 | ||
608 | static struct ofproto * | |
609 | alloc(void) | |
610 | { | |
611 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
612 | return &ofproto->up; | |
613 | } | |
614 | ||
615 | static void | |
616 | dealloc(struct ofproto *ofproto_) | |
617 | { | |
618 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
619 | free(ofproto); | |
620 | } | |
621 | ||
622 | static int | |
0f5f95a9 | 623 | construct(struct ofproto *ofproto_) |
abe529af BP |
624 | { |
625 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
626 | const char *name = ofproto->up.name; | |
627 | int error; | |
628 | int i; | |
629 | ||
630 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
631 | if (error) { | |
632 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
633 | return error; | |
634 | } | |
635 | ||
636 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 637 | ofproto->n_matches = 0; |
abe529af | 638 | |
be8194bb JG |
639 | dpif_flow_flush(ofproto->dpif); |
640 | dpif_recv_purge(ofproto->dpif); | |
641 | ||
a12b3ead | 642 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
643 | if (error) { |
644 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
645 | dpif_close(ofproto->dpif); | |
646 | return error; | |
647 | } | |
abe529af BP |
648 | |
649 | ofproto->netflow = NULL; | |
650 | ofproto->sflow = NULL; | |
21f7563c | 651 | ofproto->stp = NULL; |
abe529af BP |
652 | hmap_init(&ofproto->bundles); |
653 | ofproto->ml = mac_learning_create(); | |
654 | for (i = 0; i < MAX_MIRRORS; i++) { | |
655 | ofproto->mirrors[i] = NULL; | |
656 | } | |
657 | ofproto->has_bonded_bundles = false; | |
658 | ||
659 | timer_set_duration(&ofproto->next_expiration, 1000); | |
660 | ||
661 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 662 | hmap_init(&ofproto->subfacets); |
54a9cbc9 BP |
663 | |
664 | for (i = 0; i < N_TABLES; i++) { | |
665 | struct table_dpif *table = &ofproto->tables[i]; | |
666 | ||
667 | table->catchall_table = NULL; | |
668 | table->other_table = NULL; | |
669 | table->basis = random_uint32(); | |
670 | } | |
abe529af BP |
671 | ofproto->need_revalidate = false; |
672 | tag_set_init(&ofproto->revalidate_set); | |
673 | ||
7ee20df1 BP |
674 | list_init(&ofproto->completions); |
675 | ||
abe529af BP |
676 | ofproto_dpif_unixctl_init(); |
677 | ||
daff3353 EJ |
678 | ofproto->has_bundle_action = false; |
679 | ||
52a90c29 BP |
680 | hmap_init(&ofproto->vlandev_map); |
681 | hmap_init(&ofproto->realdev_vid_map); | |
682 | ||
b44a10b7 BP |
683 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
684 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 685 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
686 | |
687 | ofproto_init_tables(ofproto_, N_TABLES); | |
688 | ||
abe529af BP |
689 | return 0; |
690 | } | |
691 | ||
7ee20df1 BP |
692 | static void |
693 | complete_operations(struct ofproto_dpif *ofproto) | |
694 | { | |
695 | struct dpif_completion *c, *next; | |
696 | ||
697 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
698 | ofoperation_complete(c->op, 0); | |
699 | list_remove(&c->list_node); | |
700 | free(c); | |
701 | } | |
702 | } | |
703 | ||
abe529af BP |
704 | static void |
705 | destruct(struct ofproto *ofproto_) | |
706 | { | |
707 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 708 | struct rule_dpif *rule, *next_rule; |
d0918789 | 709 | struct oftable *table; |
abe529af BP |
710 | int i; |
711 | ||
b44a10b7 | 712 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
713 | complete_operations(ofproto); |
714 | ||
0697b5c3 BP |
715 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
716 | struct cls_cursor cursor; | |
717 | ||
d0918789 | 718 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
719 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
720 | ofproto_rule_destroy(&rule->up); | |
721 | } | |
7ee20df1 BP |
722 | } |
723 | ||
abe529af BP |
724 | for (i = 0; i < MAX_MIRRORS; i++) { |
725 | mirror_destroy(ofproto->mirrors[i]); | |
726 | } | |
727 | ||
728 | netflow_destroy(ofproto->netflow); | |
bae473fe | 729 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
730 | hmap_destroy(&ofproto->bundles); |
731 | mac_learning_destroy(ofproto->ml); | |
732 | ||
733 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 734 | hmap_destroy(&ofproto->subfacets); |
abe529af | 735 | |
52a90c29 BP |
736 | hmap_destroy(&ofproto->vlandev_map); |
737 | hmap_destroy(&ofproto->realdev_vid_map); | |
738 | ||
abe529af BP |
739 | dpif_close(ofproto->dpif); |
740 | } | |
741 | ||
742 | static int | |
5fcc0d00 | 743 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
744 | { |
745 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
9b16c439 | 746 | unsigned int work; |
abe529af | 747 | |
9b16c439 BP |
748 | /* Handle one or more batches of upcalls, until there's nothing left to do |
749 | * or until we do a fixed total amount of work. | |
750 | * | |
751 | * We do work in batches because it can be much cheaper to set up a number | |
752 | * of flows and fire off their patches all at once. We do multiple batches | |
753 | * because in some cases handling a packet can cause another packet to be | |
754 | * queued almost immediately as part of the return flow. Both | |
755 | * optimizations can make major improvements on some benchmarks and | |
756 | * presumably for real traffic as well. */ | |
757 | work = 0; | |
758 | while (work < FLOW_MISS_MAX_BATCH) { | |
759 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 760 | if (retval <= 0) { |
9b16c439 | 761 | return -retval; |
501f8d1f | 762 | } |
5fcc0d00 BP |
763 | work += retval; |
764 | } | |
765 | return 0; | |
766 | } | |
767 | ||
768 | static int | |
769 | run(struct ofproto *ofproto_) | |
770 | { | |
771 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
772 | struct ofport_dpif *ofport; | |
773 | struct ofbundle *bundle; | |
774 | int error; | |
775 | ||
776 | if (!clogged) { | |
777 | complete_operations(ofproto); | |
778 | } | |
779 | dpif_run(ofproto->dpif); | |
780 | ||
781 | error = run_fast(ofproto_); | |
782 | if (error) { | |
783 | return error; | |
abe529af BP |
784 | } |
785 | ||
786 | if (timer_expired(&ofproto->next_expiration)) { | |
787 | int delay = expire(ofproto); | |
788 | timer_set_duration(&ofproto->next_expiration, delay); | |
789 | } | |
790 | ||
791 | if (ofproto->netflow) { | |
6fca1ffb BP |
792 | if (netflow_run(ofproto->netflow)) { |
793 | send_netflow_active_timeouts(ofproto); | |
794 | } | |
abe529af BP |
795 | } |
796 | if (ofproto->sflow) { | |
bae473fe | 797 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
798 | } |
799 | ||
800 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
801 | port_run(ofport); | |
802 | } | |
803 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
804 | bundle_run(bundle); | |
805 | } | |
806 | ||
21f7563c | 807 | stp_run(ofproto); |
1c313b88 BP |
808 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
809 | ||
abe529af BP |
810 | /* Now revalidate if there's anything to do. */ |
811 | if (ofproto->need_revalidate | |
812 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
813 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
814 | bool revalidate_all = ofproto->need_revalidate; | |
815 | struct facet *facet, *next; | |
816 | ||
817 | /* Clear the revalidation flags. */ | |
818 | tag_set_init(&ofproto->revalidate_set); | |
819 | ofproto->need_revalidate = false; | |
820 | ||
821 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
822 | if (revalidate_all | |
823 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 824 | facet_revalidate(facet); |
abe529af BP |
825 | } |
826 | } | |
827 | } | |
828 | ||
6814e51f BP |
829 | /* Check the consistency of a random facet, to aid debugging. */ |
830 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
831 | struct facet *facet; | |
832 | ||
833 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
834 | struct facet, hmap_node); | |
835 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
836 | if (!facet_check_consistency(facet)) { | |
837 | ofproto->need_revalidate = true; | |
838 | } | |
839 | } | |
840 | } | |
841 | ||
abe529af BP |
842 | return 0; |
843 | } | |
844 | ||
845 | static void | |
846 | wait(struct ofproto *ofproto_) | |
847 | { | |
848 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
849 | struct ofport_dpif *ofport; | |
850 | struct ofbundle *bundle; | |
851 | ||
7ee20df1 BP |
852 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
853 | poll_immediate_wake(); | |
854 | } | |
855 | ||
abe529af BP |
856 | dpif_wait(ofproto->dpif); |
857 | dpif_recv_wait(ofproto->dpif); | |
858 | if (ofproto->sflow) { | |
bae473fe | 859 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
860 | } |
861 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
862 | poll_immediate_wake(); | |
863 | } | |
864 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
865 | port_wait(ofport); | |
866 | } | |
867 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
868 | bundle_wait(bundle); | |
869 | } | |
6fca1ffb BP |
870 | if (ofproto->netflow) { |
871 | netflow_wait(ofproto->netflow); | |
872 | } | |
1c313b88 | 873 | mac_learning_wait(ofproto->ml); |
21f7563c | 874 | stp_wait(ofproto); |
abe529af BP |
875 | if (ofproto->need_revalidate) { |
876 | /* Shouldn't happen, but if it does just go around again. */ | |
877 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
878 | poll_immediate_wake(); | |
879 | } else { | |
880 | timer_wait(&ofproto->next_expiration); | |
881 | } | |
882 | } | |
883 | ||
884 | static void | |
885 | flush(struct ofproto *ofproto_) | |
886 | { | |
887 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
888 | struct facet *facet, *next_facet; | |
889 | ||
890 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
891 | /* Mark the facet as not installed so that facet_remove() doesn't | |
892 | * bother trying to uninstall it. There is no point in uninstalling it | |
893 | * individually since we are about to blow away all the facets with | |
894 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
895 | struct subfacet *subfacet; |
896 | ||
897 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
898 | subfacet->installed = false; | |
899 | subfacet->dp_packet_count = 0; | |
900 | subfacet->dp_byte_count = 0; | |
901 | } | |
15baa734 | 902 | facet_remove(facet); |
abe529af BP |
903 | } |
904 | dpif_flow_flush(ofproto->dpif); | |
905 | } | |
906 | ||
6c1491fb BP |
907 | static void |
908 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
909 | bool *arp_match_ip, uint32_t *actions) | |
910 | { | |
911 | *arp_match_ip = true; | |
912 | *actions = ((1u << OFPAT_OUTPUT) | | |
913 | (1u << OFPAT_SET_VLAN_VID) | | |
914 | (1u << OFPAT_SET_VLAN_PCP) | | |
915 | (1u << OFPAT_STRIP_VLAN) | | |
916 | (1u << OFPAT_SET_DL_SRC) | | |
917 | (1u << OFPAT_SET_DL_DST) | | |
918 | (1u << OFPAT_SET_NW_SRC) | | |
919 | (1u << OFPAT_SET_NW_DST) | | |
920 | (1u << OFPAT_SET_NW_TOS) | | |
921 | (1u << OFPAT_SET_TP_SRC) | | |
922 | (1u << OFPAT_SET_TP_DST) | | |
923 | (1u << OFPAT_ENQUEUE)); | |
924 | } | |
925 | ||
926 | static void | |
927 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
928 | { | |
929 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 930 | struct dpif_dp_stats s; |
6c1491fb BP |
931 | |
932 | strcpy(ots->name, "classifier"); | |
933 | ||
934 | dpif_get_dp_stats(ofproto->dpif, &s); | |
935 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
936 | put_32aligned_be64(&ots->matched_count, | |
937 | htonll(s.n_hit + ofproto->n_matches)); | |
938 | } | |
939 | ||
abe529af BP |
940 | static struct ofport * |
941 | port_alloc(void) | |
942 | { | |
943 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
944 | return &port->up; | |
945 | } | |
946 | ||
947 | static void | |
948 | port_dealloc(struct ofport *port_) | |
949 | { | |
950 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
951 | free(port); | |
952 | } | |
953 | ||
954 | static int | |
955 | port_construct(struct ofport *port_) | |
956 | { | |
957 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
958 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
959 | ||
f11c28c4 | 960 | ofproto->need_revalidate = true; |
abe529af BP |
961 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
962 | port->bundle = NULL; | |
963 | port->cfm = NULL; | |
964 | port->tag = tag_create_random(); | |
d5ffa7f2 | 965 | port->may_enable = true; |
21f7563c JP |
966 | port->stp_port = NULL; |
967 | port->stp_state = STP_DISABLED; | |
8b36f51e | 968 | hmap_init(&port->priorities); |
52a90c29 BP |
969 | port->realdev_ofp_port = 0; |
970 | port->vlandev_vid = 0; | |
abe529af BP |
971 | |
972 | if (ofproto->sflow) { | |
392c7182 | 973 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
974 | } |
975 | ||
976 | return 0; | |
977 | } | |
978 | ||
979 | static void | |
980 | port_destruct(struct ofport *port_) | |
981 | { | |
982 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
983 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
984 | ||
f11c28c4 | 985 | ofproto->need_revalidate = true; |
abe529af | 986 | bundle_remove(port_); |
a5610457 | 987 | set_cfm(port_, NULL); |
abe529af | 988 | if (ofproto->sflow) { |
bae473fe | 989 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 990 | } |
8b36f51e EJ |
991 | |
992 | ofport_clear_priorities(port); | |
993 | hmap_destroy(&port->priorities); | |
abe529af BP |
994 | } |
995 | ||
996 | static void | |
997 | port_modified(struct ofport *port_) | |
998 | { | |
999 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1000 | ||
1001 | if (port->bundle && port->bundle->bond) { | |
1002 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1003 | } | |
1004 | } | |
1005 | ||
1006 | static void | |
1007 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
1008 | { | |
1009 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1010 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1011 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
1012 | ||
1013 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
1014 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
1015 | ofproto->need_revalidate = true; | |
7bde8dd8 JP |
1016 | |
1017 | if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) { | |
1018 | bundle_update(port->bundle); | |
1019 | } | |
abe529af BP |
1020 | } |
1021 | } | |
1022 | ||
1023 | static int | |
1024 | set_sflow(struct ofproto *ofproto_, | |
1025 | const struct ofproto_sflow_options *sflow_options) | |
1026 | { | |
1027 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1028 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1029 | |
abe529af | 1030 | if (sflow_options) { |
bae473fe | 1031 | if (!ds) { |
abe529af BP |
1032 | struct ofport_dpif *ofport; |
1033 | ||
bae473fe | 1034 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1035 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1036 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1037 | } |
6ff686f2 | 1038 | ofproto->need_revalidate = true; |
abe529af | 1039 | } |
bae473fe | 1040 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1041 | } else { |
6ff686f2 PS |
1042 | if (ds) { |
1043 | dpif_sflow_destroy(ds); | |
1044 | ofproto->need_revalidate = true; | |
1045 | ofproto->sflow = NULL; | |
1046 | } | |
abe529af BP |
1047 | } |
1048 | return 0; | |
1049 | } | |
1050 | ||
1051 | static int | |
a5610457 | 1052 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1053 | { |
1054 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1055 | int error; | |
1056 | ||
a5610457 | 1057 | if (!s) { |
abe529af BP |
1058 | error = 0; |
1059 | } else { | |
1060 | if (!ofport->cfm) { | |
8c977421 EJ |
1061 | struct ofproto_dpif *ofproto; |
1062 | ||
1063 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1064 | ofproto->need_revalidate = true; | |
6f629657 | 1065 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1066 | } |
1067 | ||
a5610457 | 1068 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1069 | return 0; |
1070 | } | |
1071 | ||
1072 | error = EINVAL; | |
1073 | } | |
1074 | cfm_destroy(ofport->cfm); | |
1075 | ofport->cfm = NULL; | |
1076 | return error; | |
1077 | } | |
1078 | ||
1079 | static int | |
a5610457 | 1080 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1081 | { |
1082 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1083 | |
1084 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1085 | } |
1de11730 EJ |
1086 | |
1087 | static int | |
1088 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1089 | size_t *n_rmps) | |
1090 | { | |
1091 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1092 | ||
1093 | if (ofport->cfm) { | |
1094 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1095 | return 0; | |
1096 | } else { | |
1097 | return -1; | |
1098 | } | |
1099 | } | |
abe529af | 1100 | \f |
21f7563c JP |
1101 | /* Spanning Tree. */ |
1102 | ||
1103 | static void | |
1104 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1105 | { | |
1106 | struct ofproto_dpif *ofproto = ofproto_; | |
1107 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1108 | struct ofport_dpif *ofport; | |
1109 | ||
1110 | ofport = stp_port_get_aux(sp); | |
1111 | if (!ofport) { | |
1112 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1113 | ofproto->up.name, port_num); | |
1114 | } else { | |
1115 | struct eth_header *eth = pkt->l2; | |
1116 | ||
1117 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1118 | if (eth_addr_is_zero(eth->eth_src)) { | |
1119 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1120 | "with unknown MAC", ofproto->up.name, port_num); | |
1121 | } else { | |
97d6520b | 1122 | send_packet(ofport, pkt); |
21f7563c JP |
1123 | } |
1124 | } | |
1125 | ofpbuf_delete(pkt); | |
1126 | } | |
1127 | ||
1128 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1129 | static int | |
1130 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1131 | { | |
1132 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1133 | ||
1134 | /* Only revalidate flows if the configuration changed. */ | |
1135 | if (!s != !ofproto->stp) { | |
1136 | ofproto->need_revalidate = true; | |
1137 | } | |
1138 | ||
1139 | if (s) { | |
1140 | if (!ofproto->stp) { | |
1141 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1142 | send_bpdu_cb, ofproto); | |
1143 | ofproto->stp_last_tick = time_msec(); | |
1144 | } | |
1145 | ||
1146 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1147 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1148 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1149 | stp_set_max_age(ofproto->stp, s->max_age); | |
1150 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1151 | } else { | |
1152 | stp_destroy(ofproto->stp); | |
1153 | ofproto->stp = NULL; | |
1154 | } | |
1155 | ||
1156 | return 0; | |
1157 | } | |
1158 | ||
1159 | static int | |
1160 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1161 | { | |
1162 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1163 | ||
1164 | if (ofproto->stp) { | |
1165 | s->enabled = true; | |
1166 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1167 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1168 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1169 | } else { | |
1170 | s->enabled = false; | |
1171 | } | |
1172 | ||
1173 | return 0; | |
1174 | } | |
1175 | ||
1176 | static void | |
1177 | update_stp_port_state(struct ofport_dpif *ofport) | |
1178 | { | |
1179 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1180 | enum stp_state state; | |
1181 | ||
1182 | /* Figure out new state. */ | |
1183 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1184 | : STP_DISABLED; | |
1185 | ||
1186 | /* Update state. */ | |
1187 | if (ofport->stp_state != state) { | |
1188 | ovs_be32 of_state; | |
1189 | bool fwd_change; | |
1190 | ||
1191 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1192 | netdev_get_name(ofport->up.netdev), | |
1193 | stp_state_name(ofport->stp_state), | |
1194 | stp_state_name(state)); | |
1195 | if (stp_learn_in_state(ofport->stp_state) | |
1196 | != stp_learn_in_state(state)) { | |
1197 | /* xxx Learning action flows should also be flushed. */ | |
d0040604 | 1198 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
21f7563c JP |
1199 | } |
1200 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1201 | != stp_forward_in_state(state); | |
1202 | ||
1203 | ofproto->need_revalidate = true; | |
1204 | ofport->stp_state = state; | |
1205 | ofport->stp_state_entered = time_msec(); | |
1206 | ||
b308140a | 1207 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1208 | bundle_update(ofport->bundle); |
1209 | } | |
1210 | ||
1211 | /* Update the STP state bits in the OpenFlow port description. */ | |
1212 | of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK)) | |
1213 | | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN | |
1214 | : state == STP_LEARNING ? OFPPS_STP_LEARN | |
1215 | : state == STP_FORWARDING ? OFPPS_STP_FORWARD | |
1216 | : state == STP_BLOCKING ? OFPPS_STP_BLOCK | |
1217 | : 0); | |
1218 | ofproto_port_set_state(&ofport->up, of_state); | |
1219 | } | |
1220 | } | |
1221 | ||
1222 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1223 | * caller is responsible for assigning STP port numbers and ensuring | |
1224 | * there are no duplicates. */ | |
1225 | static int | |
1226 | set_stp_port(struct ofport *ofport_, | |
1227 | const struct ofproto_port_stp_settings *s) | |
1228 | { | |
1229 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1230 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1231 | struct stp_port *sp = ofport->stp_port; | |
1232 | ||
1233 | if (!s || !s->enable) { | |
1234 | if (sp) { | |
1235 | ofport->stp_port = NULL; | |
1236 | stp_port_disable(sp); | |
ecd12731 | 1237 | update_stp_port_state(ofport); |
21f7563c JP |
1238 | } |
1239 | return 0; | |
1240 | } else if (sp && stp_port_no(sp) != s->port_num | |
1241 | && ofport == stp_port_get_aux(sp)) { | |
1242 | /* The port-id changed, so disable the old one if it's not | |
1243 | * already in use by another port. */ | |
1244 | stp_port_disable(sp); | |
1245 | } | |
1246 | ||
1247 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1248 | stp_port_enable(sp); | |
1249 | ||
1250 | stp_port_set_aux(sp, ofport); | |
1251 | stp_port_set_priority(sp, s->priority); | |
1252 | stp_port_set_path_cost(sp, s->path_cost); | |
1253 | ||
1254 | update_stp_port_state(ofport); | |
1255 | ||
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | static int | |
1260 | get_stp_port_status(struct ofport *ofport_, | |
1261 | struct ofproto_port_stp_status *s) | |
1262 | { | |
1263 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1264 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1265 | struct stp_port *sp = ofport->stp_port; | |
1266 | ||
1267 | if (!ofproto->stp || !sp) { | |
1268 | s->enabled = false; | |
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | s->enabled = true; | |
1273 | s->port_id = stp_port_get_id(sp); | |
1274 | s->state = stp_port_get_state(sp); | |
1275 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1276 | s->role = stp_port_get_role(sp); | |
80740385 | 1277 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1278 | |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static void | |
1283 | stp_run(struct ofproto_dpif *ofproto) | |
1284 | { | |
1285 | if (ofproto->stp) { | |
1286 | long long int now = time_msec(); | |
1287 | long long int elapsed = now - ofproto->stp_last_tick; | |
1288 | struct stp_port *sp; | |
1289 | ||
1290 | if (elapsed > 0) { | |
1291 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1292 | ofproto->stp_last_tick = now; | |
1293 | } | |
1294 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1295 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1296 | ||
1297 | if (ofport) { | |
1298 | update_stp_port_state(ofport); | |
1299 | } | |
1300 | } | |
6ae50723 EJ |
1301 | |
1302 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1303 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1304 | } | |
21f7563c JP |
1305 | } |
1306 | } | |
1307 | ||
1308 | static void | |
1309 | stp_wait(struct ofproto_dpif *ofproto) | |
1310 | { | |
1311 | if (ofproto->stp) { | |
1312 | poll_timer_wait(1000); | |
1313 | } | |
1314 | } | |
1315 | ||
1316 | /* Returns true if STP should process 'flow'. */ | |
1317 | static bool | |
1318 | stp_should_process_flow(const struct flow *flow) | |
1319 | { | |
1320 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1321 | } | |
1322 | ||
1323 | static void | |
1324 | stp_process_packet(const struct ofport_dpif *ofport, | |
1325 | const struct ofpbuf *packet) | |
1326 | { | |
1327 | struct ofpbuf payload = *packet; | |
1328 | struct eth_header *eth = payload.data; | |
1329 | struct stp_port *sp = ofport->stp_port; | |
1330 | ||
1331 | /* Sink packets on ports that have STP disabled when the bridge has | |
1332 | * STP enabled. */ | |
1333 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1334 | return; | |
1335 | } | |
1336 | ||
1337 | /* Trim off padding on payload. */ | |
c573540b BP |
1338 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1339 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1340 | } |
1341 | ||
1342 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1343 | stp_received_bpdu(sp, payload.data, payload.size); | |
1344 | } | |
1345 | } | |
1346 | \f | |
8b36f51e EJ |
1347 | static struct priority_to_dscp * |
1348 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1349 | { | |
1350 | struct priority_to_dscp *pdscp; | |
1351 | uint32_t hash; | |
1352 | ||
1353 | hash = hash_int(priority, 0); | |
1354 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1355 | if (pdscp->priority == priority) { | |
1356 | return pdscp; | |
1357 | } | |
1358 | } | |
1359 | return NULL; | |
1360 | } | |
1361 | ||
1362 | static void | |
1363 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1364 | { | |
1365 | struct priority_to_dscp *pdscp, *next; | |
1366 | ||
1367 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1368 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1369 | free(pdscp); | |
1370 | } | |
1371 | } | |
1372 | ||
1373 | static int | |
1374 | set_queues(struct ofport *ofport_, | |
1375 | const struct ofproto_port_queue *qdscp_list, | |
1376 | size_t n_qdscp) | |
1377 | { | |
1378 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1380 | struct hmap new = HMAP_INITIALIZER(&new); | |
1381 | size_t i; | |
1382 | ||
1383 | for (i = 0; i < n_qdscp; i++) { | |
1384 | struct priority_to_dscp *pdscp; | |
1385 | uint32_t priority; | |
1386 | uint8_t dscp; | |
1387 | ||
1388 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1389 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1390 | &priority)) { | |
1391 | continue; | |
1392 | } | |
1393 | ||
1394 | pdscp = get_priority(ofport, priority); | |
1395 | if (pdscp) { | |
1396 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1397 | } else { | |
1398 | pdscp = xmalloc(sizeof *pdscp); | |
1399 | pdscp->priority = priority; | |
1400 | pdscp->dscp = dscp; | |
1401 | ofproto->need_revalidate = true; | |
1402 | } | |
1403 | ||
1404 | if (pdscp->dscp != dscp) { | |
1405 | pdscp->dscp = dscp; | |
1406 | ofproto->need_revalidate = true; | |
1407 | } | |
1408 | ||
1409 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1410 | } | |
1411 | ||
1412 | if (!hmap_is_empty(&ofport->priorities)) { | |
1413 | ofport_clear_priorities(ofport); | |
1414 | ofproto->need_revalidate = true; | |
1415 | } | |
1416 | ||
1417 | hmap_swap(&new, &ofport->priorities); | |
1418 | hmap_destroy(&new); | |
1419 | ||
1420 | return 0; | |
1421 | } | |
1422 | \f | |
abe529af BP |
1423 | /* Bundles. */ |
1424 | ||
b44a10b7 BP |
1425 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1426 | * ofproto to revalidate every flow. | |
1427 | * | |
1428 | * Normally MAC learning entries are removed only from the ofproto associated | |
1429 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1430 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1431 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1432 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1433 | * with the host from which it migrated. */ | |
abe529af | 1434 | static void |
b44a10b7 | 1435 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1436 | { |
1437 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1438 | struct mac_learning *ml = ofproto->ml; | |
1439 | struct mac_entry *mac, *next_mac; | |
1440 | ||
1441 | ofproto->need_revalidate = true; | |
1442 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1443 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1444 | if (all_ofprotos) { |
1445 | struct ofproto_dpif *o; | |
1446 | ||
1447 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1448 | if (o != ofproto) { | |
1449 | struct mac_entry *e; | |
1450 | ||
1451 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1452 | NULL); | |
1453 | if (e) { | |
1454 | tag_set_add(&o->revalidate_set, e->tag); | |
1455 | mac_learning_expire(o->ml, e); | |
1456 | } | |
1457 | } | |
1458 | } | |
1459 | } | |
1460 | ||
abe529af BP |
1461 | mac_learning_expire(ml, mac); |
1462 | } | |
1463 | } | |
1464 | } | |
1465 | ||
1466 | static struct ofbundle * | |
1467 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1468 | { | |
1469 | struct ofbundle *bundle; | |
1470 | ||
1471 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1472 | &ofproto->bundles) { | |
1473 | if (bundle->aux == aux) { | |
1474 | return bundle; | |
1475 | } | |
1476 | } | |
1477 | return NULL; | |
1478 | } | |
1479 | ||
1480 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1481 | * ones that are found to 'bundles'. */ | |
1482 | static void | |
1483 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1484 | void **auxes, size_t n_auxes, | |
1485 | struct hmapx *bundles) | |
1486 | { | |
1487 | size_t i; | |
1488 | ||
1489 | hmapx_init(bundles); | |
1490 | for (i = 0; i < n_auxes; i++) { | |
1491 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1492 | if (bundle) { | |
1493 | hmapx_add(bundles, bundle); | |
1494 | } | |
1495 | } | |
1496 | } | |
1497 | ||
7bde8dd8 JP |
1498 | static void |
1499 | bundle_update(struct ofbundle *bundle) | |
1500 | { | |
1501 | struct ofport_dpif *port; | |
1502 | ||
1503 | bundle->floodable = true; | |
1504 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
5e48dc2b | 1505 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { |
7bde8dd8 JP |
1506 | bundle->floodable = false; |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | } | |
1511 | ||
abe529af BP |
1512 | static void |
1513 | bundle_del_port(struct ofport_dpif *port) | |
1514 | { | |
1515 | struct ofbundle *bundle = port->bundle; | |
1516 | ||
6f77f4ae BP |
1517 | bundle->ofproto->need_revalidate = true; |
1518 | ||
abe529af BP |
1519 | list_remove(&port->bundle_node); |
1520 | port->bundle = NULL; | |
1521 | ||
1522 | if (bundle->lacp) { | |
1523 | lacp_slave_unregister(bundle->lacp, port); | |
1524 | } | |
1525 | if (bundle->bond) { | |
1526 | bond_slave_unregister(bundle->bond, port); | |
1527 | } | |
1528 | ||
7bde8dd8 | 1529 | bundle_update(bundle); |
abe529af BP |
1530 | } |
1531 | ||
1532 | static bool | |
1533 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1534 | struct lacp_slave_settings *lacp, |
1535 | uint32_t bond_stable_id) | |
abe529af BP |
1536 | { |
1537 | struct ofport_dpif *port; | |
1538 | ||
1539 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1540 | if (!port) { | |
1541 | return false; | |
1542 | } | |
1543 | ||
1544 | if (port->bundle != bundle) { | |
6f77f4ae | 1545 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1546 | if (port->bundle) { |
1547 | bundle_del_port(port); | |
1548 | } | |
1549 | ||
1550 | port->bundle = bundle; | |
1551 | list_push_back(&bundle->ports, &port->bundle_node); | |
5e48dc2b | 1552 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { |
abe529af BP |
1553 | bundle->floodable = false; |
1554 | } | |
1555 | } | |
1556 | if (lacp) { | |
4a86aece | 1557 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1558 | lacp_slave_register(bundle->lacp, port, lacp); |
1559 | } | |
1560 | ||
00794817 BP |
1561 | port->bond_stable_id = bond_stable_id; |
1562 | ||
abe529af BP |
1563 | return true; |
1564 | } | |
1565 | ||
1566 | static void | |
1567 | bundle_destroy(struct ofbundle *bundle) | |
1568 | { | |
1569 | struct ofproto_dpif *ofproto; | |
1570 | struct ofport_dpif *port, *next_port; | |
1571 | int i; | |
1572 | ||
1573 | if (!bundle) { | |
1574 | return; | |
1575 | } | |
1576 | ||
1577 | ofproto = bundle->ofproto; | |
1578 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1579 | struct ofmirror *m = ofproto->mirrors[i]; | |
1580 | if (m) { | |
1581 | if (m->out == bundle) { | |
1582 | mirror_destroy(m); | |
1583 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1584 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1585 | ofproto->need_revalidate = true; | |
1586 | } | |
1587 | } | |
1588 | } | |
1589 | ||
1590 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1591 | bundle_del_port(port); | |
1592 | } | |
1593 | ||
b44a10b7 | 1594 | bundle_flush_macs(bundle, true); |
abe529af BP |
1595 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1596 | free(bundle->name); | |
1597 | free(bundle->trunks); | |
1598 | lacp_destroy(bundle->lacp); | |
1599 | bond_destroy(bundle->bond); | |
1600 | free(bundle); | |
1601 | } | |
1602 | ||
1603 | static int | |
1604 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1605 | const struct ofproto_bundle_settings *s) | |
1606 | { | |
1607 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1608 | bool need_flush = false; | |
abe529af BP |
1609 | struct ofport_dpif *port; |
1610 | struct ofbundle *bundle; | |
ecac4ebf BP |
1611 | unsigned long *trunks; |
1612 | int vlan; | |
abe529af BP |
1613 | size_t i; |
1614 | bool ok; | |
1615 | ||
1616 | if (!s) { | |
1617 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1618 | return 0; | |
1619 | } | |
1620 | ||
1621 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1622 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1623 | ||
1624 | bundle = bundle_lookup(ofproto, aux); | |
1625 | if (!bundle) { | |
1626 | bundle = xmalloc(sizeof *bundle); | |
1627 | ||
1628 | bundle->ofproto = ofproto; | |
1629 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1630 | hash_pointer(aux, 0)); | |
1631 | bundle->aux = aux; | |
1632 | bundle->name = NULL; | |
1633 | ||
1634 | list_init(&bundle->ports); | |
ecac4ebf | 1635 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1636 | bundle->vlan = -1; |
1637 | bundle->trunks = NULL; | |
5e9ceccd | 1638 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1639 | bundle->lacp = NULL; |
1640 | bundle->bond = NULL; | |
1641 | ||
1642 | bundle->floodable = true; | |
1643 | ||
1644 | bundle->src_mirrors = 0; | |
1645 | bundle->dst_mirrors = 0; | |
1646 | bundle->mirror_out = 0; | |
1647 | } | |
1648 | ||
1649 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1650 | free(bundle->name); | |
1651 | bundle->name = xstrdup(s->name); | |
1652 | } | |
1653 | ||
1654 | /* LACP. */ | |
1655 | if (s->lacp) { | |
1656 | if (!bundle->lacp) { | |
8c977421 | 1657 | ofproto->need_revalidate = true; |
abe529af BP |
1658 | bundle->lacp = lacp_create(); |
1659 | } | |
1660 | lacp_configure(bundle->lacp, s->lacp); | |
1661 | } else { | |
1662 | lacp_destroy(bundle->lacp); | |
1663 | bundle->lacp = NULL; | |
1664 | } | |
1665 | ||
1666 | /* Update set of ports. */ | |
1667 | ok = true; | |
1668 | for (i = 0; i < s->n_slaves; i++) { | |
1669 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1670 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1671 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1672 | ok = false; |
1673 | } | |
1674 | } | |
1675 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1676 | struct ofport_dpif *next_port; | |
1677 | ||
1678 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1679 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1680 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1681 | goto found; |
1682 | } | |
1683 | } | |
1684 | ||
1685 | bundle_del_port(port); | |
1686 | found: ; | |
1687 | } | |
1688 | } | |
1689 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1690 | ||
1691 | if (list_is_empty(&bundle->ports)) { | |
1692 | bundle_destroy(bundle); | |
1693 | return EINVAL; | |
1694 | } | |
1695 | ||
ecac4ebf | 1696 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1697 | if (s->vlan_mode != bundle->vlan_mode |
1698 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1699 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1700 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1701 | need_flush = true; |
1702 | } | |
1703 | ||
abe529af | 1704 | /* Set VLAN tag. */ |
ecac4ebf BP |
1705 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1706 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1707 | : 0); | |
1708 | if (vlan != bundle->vlan) { | |
1709 | bundle->vlan = vlan; | |
abe529af BP |
1710 | need_flush = true; |
1711 | } | |
1712 | ||
1713 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1714 | switch (s->vlan_mode) { |
1715 | case PORT_VLAN_ACCESS: | |
1716 | trunks = NULL; | |
1717 | break; | |
1718 | ||
1719 | case PORT_VLAN_TRUNK: | |
1720 | trunks = (unsigned long *) s->trunks; | |
1721 | break; | |
1722 | ||
1723 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1724 | case PORT_VLAN_NATIVE_TAGGED: | |
1725 | if (vlan != 0 && (!s->trunks | |
1726 | || !bitmap_is_set(s->trunks, vlan) | |
1727 | || bitmap_is_set(s->trunks, 0))) { | |
1728 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1729 | if (s->trunks) { | |
1730 | trunks = bitmap_clone(s->trunks, 4096); | |
1731 | } else { | |
1732 | trunks = bitmap_allocate1(4096); | |
1733 | } | |
1734 | bitmap_set1(trunks, vlan); | |
1735 | bitmap_set0(trunks, 0); | |
1736 | } else { | |
1737 | trunks = (unsigned long *) s->trunks; | |
1738 | } | |
1739 | break; | |
1740 | ||
1741 | default: | |
1742 | NOT_REACHED(); | |
1743 | } | |
abe529af BP |
1744 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1745 | free(bundle->trunks); | |
ecac4ebf BP |
1746 | if (trunks == s->trunks) { |
1747 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1748 | } else { | |
1749 | bundle->trunks = trunks; | |
1750 | trunks = NULL; | |
1751 | } | |
abe529af BP |
1752 | need_flush = true; |
1753 | } | |
ecac4ebf BP |
1754 | if (trunks != s->trunks) { |
1755 | free(trunks); | |
1756 | } | |
abe529af BP |
1757 | |
1758 | /* Bonding. */ | |
1759 | if (!list_is_short(&bundle->ports)) { | |
1760 | bundle->ofproto->has_bonded_bundles = true; | |
1761 | if (bundle->bond) { | |
1762 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1763 | ofproto->need_revalidate = true; | |
1764 | } | |
1765 | } else { | |
1766 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1767 | ofproto->need_revalidate = true; |
abe529af BP |
1768 | } |
1769 | ||
1770 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1771 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1772 | port->up.netdev); |
1773 | } | |
1774 | } else { | |
1775 | bond_destroy(bundle->bond); | |
1776 | bundle->bond = NULL; | |
1777 | } | |
1778 | ||
1779 | /* If we changed something that would affect MAC learning, un-learn | |
1780 | * everything on this port and force flow revalidation. */ | |
1781 | if (need_flush) { | |
b44a10b7 | 1782 | bundle_flush_macs(bundle, false); |
abe529af BP |
1783 | } |
1784 | ||
1785 | return 0; | |
1786 | } | |
1787 | ||
1788 | static void | |
1789 | bundle_remove(struct ofport *port_) | |
1790 | { | |
1791 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1792 | struct ofbundle *bundle = port->bundle; | |
1793 | ||
1794 | if (bundle) { | |
1795 | bundle_del_port(port); | |
1796 | if (list_is_empty(&bundle->ports)) { | |
1797 | bundle_destroy(bundle); | |
1798 | } else if (list_is_short(&bundle->ports)) { | |
1799 | bond_destroy(bundle->bond); | |
1800 | bundle->bond = NULL; | |
1801 | } | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | static void | |
5f877369 | 1806 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1807 | { |
1808 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1809 | struct ofport_dpif *port = port_; | |
1810 | uint8_t ea[ETH_ADDR_LEN]; | |
1811 | int error; | |
1812 | ||
1813 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1814 | if (!error) { | |
abe529af | 1815 | struct ofpbuf packet; |
5f877369 | 1816 | void *packet_pdu; |
abe529af BP |
1817 | |
1818 | ofpbuf_init(&packet, 0); | |
1819 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1820 | pdu_size); |
1821 | memcpy(packet_pdu, pdu, pdu_size); | |
1822 | ||
97d6520b | 1823 | send_packet(port, &packet); |
abe529af BP |
1824 | ofpbuf_uninit(&packet); |
1825 | } else { | |
1826 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1827 | "%s (%s)", port->bundle->name, | |
1828 | netdev_get_name(port->up.netdev), strerror(error)); | |
1829 | } | |
1830 | } | |
1831 | ||
1832 | static void | |
1833 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1834 | { | |
1835 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1836 | int error, n_packets, n_errors; | |
1837 | struct mac_entry *e; | |
1838 | ||
1839 | error = n_packets = n_errors = 0; | |
1840 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1841 | if (e->port.p != bundle) { | |
ea131871 JG |
1842 | struct ofpbuf *learning_packet; |
1843 | struct ofport_dpif *port; | |
4dd1e3ca | 1844 | void *port_void; |
ea131871 JG |
1845 | int ret; |
1846 | ||
4dd1e3ca BP |
1847 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
1848 | * struct ofport_dpif more effective. */ | |
1849 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
1850 | e->mac, e->vlan, | |
1851 | &port_void); | |
1852 | port = port_void; | |
97d6520b | 1853 | ret = send_packet(port, learning_packet); |
ea131871 | 1854 | ofpbuf_delete(learning_packet); |
abe529af BP |
1855 | if (ret) { |
1856 | error = ret; | |
1857 | n_errors++; | |
1858 | } | |
1859 | n_packets++; | |
1860 | } | |
1861 | } | |
1862 | ||
1863 | if (n_errors) { | |
1864 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1865 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1866 | "packets, last error was: %s", | |
1867 | bundle->name, n_errors, n_packets, strerror(error)); | |
1868 | } else { | |
1869 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1870 | bundle->name, n_packets); | |
1871 | } | |
1872 | } | |
1873 | ||
1874 | static void | |
1875 | bundle_run(struct ofbundle *bundle) | |
1876 | { | |
1877 | if (bundle->lacp) { | |
1878 | lacp_run(bundle->lacp, send_pdu_cb); | |
1879 | } | |
1880 | if (bundle->bond) { | |
1881 | struct ofport_dpif *port; | |
1882 | ||
1883 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1884 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1885 | } |
1886 | ||
1887 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
bdebeece | 1888 | lacp_status(bundle->lacp)); |
abe529af BP |
1889 | if (bond_should_send_learning_packets(bundle->bond)) { |
1890 | bundle_send_learning_packets(bundle); | |
1891 | } | |
1892 | } | |
1893 | } | |
1894 | ||
1895 | static void | |
1896 | bundle_wait(struct ofbundle *bundle) | |
1897 | { | |
1898 | if (bundle->lacp) { | |
1899 | lacp_wait(bundle->lacp); | |
1900 | } | |
1901 | if (bundle->bond) { | |
1902 | bond_wait(bundle->bond); | |
1903 | } | |
1904 | } | |
1905 | \f | |
1906 | /* Mirrors. */ | |
1907 | ||
1908 | static int | |
1909 | mirror_scan(struct ofproto_dpif *ofproto) | |
1910 | { | |
1911 | int idx; | |
1912 | ||
1913 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1914 | if (!ofproto->mirrors[idx]) { | |
1915 | return idx; | |
1916 | } | |
1917 | } | |
1918 | return -1; | |
1919 | } | |
1920 | ||
1921 | static struct ofmirror * | |
1922 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1923 | { | |
1924 | int i; | |
1925 | ||
1926 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1927 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1928 | if (mirror && mirror->aux == aux) { | |
1929 | return mirror; | |
1930 | } | |
1931 | } | |
1932 | ||
1933 | return NULL; | |
1934 | } | |
1935 | ||
9ba15e2a BP |
1936 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
1937 | static void | |
1938 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
1939 | { | |
1940 | int i; | |
1941 | ||
1942 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1943 | struct ofmirror *m = ofproto->mirrors[i]; | |
1944 | ||
1945 | if (m) { | |
1946 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1951 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
1952 | int j; | |
1953 | ||
1954 | if (!m1) { | |
1955 | continue; | |
1956 | } | |
1957 | ||
1958 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
1959 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
1960 | ||
edb0540b | 1961 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
1962 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
1963 | m2->dup_mirrors |= m1->dup_mirrors; | |
1964 | } | |
1965 | } | |
1966 | } | |
1967 | } | |
1968 | ||
abe529af BP |
1969 | static int |
1970 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1971 | const struct ofproto_mirror_settings *s) | |
1972 | { | |
1973 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1974 | mirror_mask_t mirror_bit; | |
1975 | struct ofbundle *bundle; | |
1976 | struct ofmirror *mirror; | |
1977 | struct ofbundle *out; | |
1978 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1979 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1980 | int out_vlan; | |
1981 | ||
1982 | mirror = mirror_lookup(ofproto, aux); | |
1983 | if (!s) { | |
1984 | mirror_destroy(mirror); | |
1985 | return 0; | |
1986 | } | |
1987 | if (!mirror) { | |
1988 | int idx; | |
1989 | ||
1990 | idx = mirror_scan(ofproto); | |
1991 | if (idx < 0) { | |
1992 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1993 | "cannot create %s", | |
1994 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1995 | return EFBIG; | |
1996 | } | |
1997 | ||
1998 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1999 | mirror->ofproto = ofproto; | |
2000 | mirror->idx = idx; | |
8b28d864 | 2001 | mirror->aux = aux; |
abe529af BP |
2002 | mirror->out_vlan = -1; |
2003 | mirror->name = NULL; | |
2004 | } | |
2005 | ||
2006 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2007 | free(mirror->name); | |
2008 | mirror->name = xstrdup(s->name); | |
2009 | } | |
2010 | ||
2011 | /* Get the new configuration. */ | |
2012 | if (s->out_bundle) { | |
2013 | out = bundle_lookup(ofproto, s->out_bundle); | |
2014 | if (!out) { | |
2015 | mirror_destroy(mirror); | |
2016 | return EINVAL; | |
2017 | } | |
2018 | out_vlan = -1; | |
2019 | } else { | |
2020 | out = NULL; | |
2021 | out_vlan = s->out_vlan; | |
2022 | } | |
2023 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2024 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2025 | ||
2026 | /* If the configuration has not changed, do nothing. */ | |
2027 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2028 | && hmapx_equals(&dsts, &mirror->dsts) | |
2029 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2030 | && mirror->out == out | |
2031 | && mirror->out_vlan == out_vlan) | |
2032 | { | |
2033 | hmapx_destroy(&srcs); | |
2034 | hmapx_destroy(&dsts); | |
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | hmapx_swap(&srcs, &mirror->srcs); | |
2039 | hmapx_destroy(&srcs); | |
2040 | ||
2041 | hmapx_swap(&dsts, &mirror->dsts); | |
2042 | hmapx_destroy(&dsts); | |
2043 | ||
2044 | free(mirror->vlans); | |
2045 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2046 | ||
2047 | mirror->out = out; | |
2048 | mirror->out_vlan = out_vlan; | |
2049 | ||
2050 | /* Update bundles. */ | |
2051 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2052 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2053 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2054 | bundle->src_mirrors |= mirror_bit; | |
2055 | } else { | |
2056 | bundle->src_mirrors &= ~mirror_bit; | |
2057 | } | |
2058 | ||
2059 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2060 | bundle->dst_mirrors |= mirror_bit; | |
2061 | } else { | |
2062 | bundle->dst_mirrors &= ~mirror_bit; | |
2063 | } | |
2064 | ||
2065 | if (mirror->out == bundle) { | |
2066 | bundle->mirror_out |= mirror_bit; | |
2067 | } else { | |
2068 | bundle->mirror_out &= ~mirror_bit; | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | ofproto->need_revalidate = true; | |
d0040604 | 2073 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
9ba15e2a | 2074 | mirror_update_dups(ofproto); |
abe529af BP |
2075 | |
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | static void | |
2080 | mirror_destroy(struct ofmirror *mirror) | |
2081 | { | |
2082 | struct ofproto_dpif *ofproto; | |
2083 | mirror_mask_t mirror_bit; | |
2084 | struct ofbundle *bundle; | |
2085 | ||
2086 | if (!mirror) { | |
2087 | return; | |
2088 | } | |
2089 | ||
2090 | ofproto = mirror->ofproto; | |
2091 | ofproto->need_revalidate = true; | |
d0040604 | 2092 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2093 | |
2094 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2095 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2096 | bundle->src_mirrors &= ~mirror_bit; | |
2097 | bundle->dst_mirrors &= ~mirror_bit; | |
2098 | bundle->mirror_out &= ~mirror_bit; | |
2099 | } | |
2100 | ||
2101 | hmapx_destroy(&mirror->srcs); | |
2102 | hmapx_destroy(&mirror->dsts); | |
2103 | free(mirror->vlans); | |
2104 | ||
2105 | ofproto->mirrors[mirror->idx] = NULL; | |
2106 | free(mirror->name); | |
2107 | free(mirror); | |
9ba15e2a BP |
2108 | |
2109 | mirror_update_dups(ofproto); | |
abe529af BP |
2110 | } |
2111 | ||
9d24de3b JP |
2112 | static int |
2113 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2114 | uint64_t *packets, uint64_t *bytes) | |
2115 | { | |
2116 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2117 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2118 | ||
2119 | if (!mirror) { | |
2120 | *packets = *bytes = UINT64_MAX; | |
2121 | return 0; | |
2122 | } | |
2123 | ||
2124 | *packets = mirror->packet_count; | |
2125 | *bytes = mirror->byte_count; | |
2126 | ||
2127 | return 0; | |
2128 | } | |
2129 | ||
abe529af BP |
2130 | static int |
2131 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2132 | { | |
2133 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2134 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
d0040604 | 2135 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2136 | } |
2137 | return 0; | |
2138 | } | |
2139 | ||
2140 | static bool | |
b4affc74 | 2141 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2142 | { |
2143 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2144 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2145 | return bundle && bundle->mirror_out != 0; | |
2146 | } | |
8402c74b SS |
2147 | |
2148 | static void | |
b53055f4 | 2149 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2150 | { |
2151 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2152 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
2153 | ofproto->need_revalidate = true; | |
2154 | } | |
abe529af BP |
2155 | \f |
2156 | /* Ports. */ | |
2157 | ||
2158 | static struct ofport_dpif * | |
2159 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
2160 | { | |
7df6a8bd BP |
2161 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2162 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2163 | } |
2164 | ||
2165 | static struct ofport_dpif * | |
2166 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
2167 | { | |
2168 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2169 | } | |
2170 | ||
2171 | static void | |
2172 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2173 | struct dpif_port *dpif_port) | |
2174 | { | |
2175 | ofproto_port->name = dpif_port->name; | |
2176 | ofproto_port->type = dpif_port->type; | |
2177 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2178 | } | |
2179 | ||
2180 | static void | |
2181 | port_run(struct ofport_dpif *ofport) | |
2182 | { | |
015e08bc EJ |
2183 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2184 | ||
abe529af BP |
2185 | if (ofport->cfm) { |
2186 | cfm_run(ofport->cfm); | |
2187 | ||
2188 | if (cfm_should_send_ccm(ofport->cfm)) { | |
2189 | struct ofpbuf packet; | |
abe529af BP |
2190 | |
2191 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 2192 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
97d6520b | 2193 | send_packet(ofport, &packet); |
abe529af BP |
2194 | ofpbuf_uninit(&packet); |
2195 | } | |
015e08bc | 2196 | |
86dc6501 EJ |
2197 | enable = enable && !cfm_get_fault(ofport->cfm) |
2198 | && cfm_get_opup(ofport->cfm); | |
abe529af | 2199 | } |
015e08bc EJ |
2200 | |
2201 | if (ofport->bundle) { | |
2202 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
2203 | } | |
2204 | ||
daff3353 EJ |
2205 | if (ofport->may_enable != enable) { |
2206 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2207 | ||
2208 | if (ofproto->has_bundle_action) { | |
2209 | ofproto->need_revalidate = true; | |
2210 | } | |
2211 | } | |
2212 | ||
015e08bc | 2213 | ofport->may_enable = enable; |
abe529af BP |
2214 | } |
2215 | ||
2216 | static void | |
2217 | port_wait(struct ofport_dpif *ofport) | |
2218 | { | |
2219 | if (ofport->cfm) { | |
2220 | cfm_wait(ofport->cfm); | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | static int | |
2225 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2226 | struct ofproto_port *ofproto_port) | |
2227 | { | |
2228 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2229 | struct dpif_port dpif_port; | |
2230 | int error; | |
2231 | ||
2232 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2233 | if (!error) { | |
2234 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2235 | } | |
2236 | return error; | |
2237 | } | |
2238 | ||
2239 | static int | |
2240 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2241 | { | |
2242 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2243 | uint16_t odp_port; | |
2244 | int error; | |
2245 | ||
2246 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2247 | if (!error) { | |
2248 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2249 | } | |
2250 | return error; | |
2251 | } | |
2252 | ||
2253 | static int | |
2254 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2255 | { | |
2256 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2257 | int error; | |
2258 | ||
2259 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2260 | if (!error) { | |
2261 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2262 | if (ofport) { | |
2263 | /* The caller is going to close ofport->up.netdev. If this is a | |
2264 | * bonded port, then the bond is using that netdev, so remove it | |
2265 | * from the bond. The client will need to reconfigure everything | |
2266 | * after deleting ports, so then the slave will get re-added. */ | |
2267 | bundle_remove(&ofport->up); | |
2268 | } | |
2269 | } | |
2270 | return error; | |
2271 | } | |
2272 | ||
6527c598 PS |
2273 | static int |
2274 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2275 | { | |
2276 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2277 | int error; | |
2278 | ||
2279 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2280 | ||
2281 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2282 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2283 | ||
2284 | /* ofproto->stats.tx_packets represents packets that we created | |
2285 | * internally and sent to some port (e.g. packets sent with | |
2286 | * send_packet()). Account for them as if they had come from | |
2287 | * OFPP_LOCAL and got forwarded. */ | |
2288 | ||
2289 | if (stats->rx_packets != UINT64_MAX) { | |
2290 | stats->rx_packets += ofproto->stats.tx_packets; | |
2291 | } | |
2292 | ||
2293 | if (stats->rx_bytes != UINT64_MAX) { | |
2294 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2295 | } | |
2296 | ||
2297 | /* ofproto->stats.rx_packets represents packets that were received on | |
2298 | * some port and we processed internally and dropped (e.g. STP). | |
2299 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2300 | ||
2301 | if (stats->tx_packets != UINT64_MAX) { | |
2302 | stats->tx_packets += ofproto->stats.rx_packets; | |
2303 | } | |
2304 | ||
2305 | if (stats->tx_bytes != UINT64_MAX) { | |
2306 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2307 | } | |
2308 | } | |
2309 | ||
2310 | return error; | |
2311 | } | |
2312 | ||
2313 | /* Account packets for LOCAL port. */ | |
2314 | static void | |
2315 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2316 | size_t tx_size, size_t rx_size) | |
2317 | { | |
2318 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2319 | ||
2320 | if (rx_size) { | |
2321 | ofproto->stats.rx_packets++; | |
2322 | ofproto->stats.rx_bytes += rx_size; | |
2323 | } | |
2324 | if (tx_size) { | |
2325 | ofproto->stats.tx_packets++; | |
2326 | ofproto->stats.tx_bytes += tx_size; | |
2327 | } | |
2328 | } | |
2329 | ||
abe529af BP |
2330 | struct port_dump_state { |
2331 | struct dpif_port_dump dump; | |
2332 | bool done; | |
2333 | }; | |
2334 | ||
2335 | static int | |
2336 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2337 | { | |
2338 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2339 | struct port_dump_state *state; | |
2340 | ||
2341 | *statep = state = xmalloc(sizeof *state); | |
2342 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2343 | state->done = false; | |
2344 | return 0; | |
2345 | } | |
2346 | ||
2347 | static int | |
2348 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2349 | struct ofproto_port *port) | |
2350 | { | |
2351 | struct port_dump_state *state = state_; | |
2352 | struct dpif_port dpif_port; | |
2353 | ||
2354 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2355 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2356 | return 0; | |
2357 | } else { | |
2358 | int error = dpif_port_dump_done(&state->dump); | |
2359 | state->done = true; | |
2360 | return error ? error : EOF; | |
2361 | } | |
2362 | } | |
2363 | ||
2364 | static int | |
2365 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2366 | { | |
2367 | struct port_dump_state *state = state_; | |
2368 | ||
2369 | if (!state->done) { | |
2370 | dpif_port_dump_done(&state->dump); | |
2371 | } | |
2372 | free(state); | |
2373 | return 0; | |
2374 | } | |
2375 | ||
2376 | static int | |
2377 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2378 | { | |
2379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2380 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2381 | } | |
2382 | ||
2383 | static void | |
2384 | port_poll_wait(const struct ofproto *ofproto_) | |
2385 | { | |
2386 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2387 | dpif_port_poll_wait(ofproto->dpif); | |
2388 | } | |
2389 | ||
2390 | static int | |
2391 | port_is_lacp_current(const struct ofport *ofport_) | |
2392 | { | |
2393 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2394 | return (ofport->bundle && ofport->bundle->lacp | |
2395 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2396 | : -1); | |
2397 | } | |
2398 | \f | |
2399 | /* Upcall handling. */ | |
2400 | ||
501f8d1f BP |
2401 | /* Flow miss batching. |
2402 | * | |
2403 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2404 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2405 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2406 | * more packets, plus possibly installing the flow in the dpif. | |
2407 | * | |
2408 | * So far we only batch the operations that affect flow setup time the most. | |
2409 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2410 | struct flow_miss { | |
2411 | struct hmap_node hmap_node; | |
2412 | struct flow flow; | |
b0f7b9b5 | 2413 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2414 | const struct nlattr *key; |
2415 | size_t key_len; | |
e84173dc | 2416 | ovs_be16 initial_tci; |
501f8d1f BP |
2417 | struct list packets; |
2418 | }; | |
2419 | ||
2420 | struct flow_miss_op { | |
c2b565b5 | 2421 | struct dpif_op dpif_op; |
b0f7b9b5 | 2422 | struct subfacet *subfacet; |
501f8d1f BP |
2423 | }; |
2424 | ||
62cd7072 BP |
2425 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2426 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2427 | * configurations. */ |
62cd7072 | 2428 | static void |
a39edbd4 | 2429 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 2430 | const struct flow *flow) |
62cd7072 BP |
2431 | { |
2432 | struct ofputil_packet_in pin; | |
2433 | ||
3e3252fa EJ |
2434 | pin.packet = packet->data; |
2435 | pin.packet_len = packet->size; | |
65120a8a | 2436 | pin.total_len = packet->size; |
62cd7072 | 2437 | pin.reason = OFPR_NO_MATCH; |
54834960 EJ |
2438 | |
2439 | pin.table_id = 0; | |
2440 | pin.cookie = 0; | |
2441 | ||
62cd7072 BP |
2442 | pin.buffer_id = 0; /* not yet known */ |
2443 | pin.send_len = 0; /* not used for flow table misses */ | |
5d6c3af0 EJ |
2444 | |
2445 | flow_get_metadata(flow, &pin.fmd); | |
2446 | ||
2447 | /* Registers aren't meaningful on a miss. */ | |
2448 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2449 | ||
29ebe880 | 2450 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow); |
62cd7072 BP |
2451 | } |
2452 | ||
abe529af BP |
2453 | static bool |
2454 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2455 | const struct ofpbuf *packet) | |
2456 | { | |
b6e001b6 EJ |
2457 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2458 | ||
2459 | if (!ofport) { | |
2460 | return false; | |
2461 | } | |
2462 | ||
ef9819b5 | 2463 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2464 | if (packet) { |
abe529af BP |
2465 | cfm_process_heartbeat(ofport->cfm, packet); |
2466 | } | |
2467 | return true; | |
b6e001b6 EJ |
2468 | } else if (ofport->bundle && ofport->bundle->lacp |
2469 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2470 | if (packet) { | |
2471 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2472 | } |
da37ebac | 2473 | return true; |
21f7563c JP |
2474 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2475 | if (packet) { | |
2476 | stp_process_packet(ofport, packet); | |
2477 | } | |
2478 | return true; | |
abe529af BP |
2479 | } |
2480 | return false; | |
2481 | } | |
2482 | ||
501f8d1f BP |
2483 | static struct flow_miss * |
2484 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
b0f7b9b5 | 2485 | enum odp_key_fitness key_fitness, |
e84173dc BP |
2486 | const struct nlattr *key, size_t key_len, |
2487 | ovs_be16 initial_tci) | |
abe529af | 2488 | { |
501f8d1f BP |
2489 | uint32_t hash = flow_hash(flow, 0); |
2490 | struct flow_miss *miss; | |
abe529af | 2491 | |
501f8d1f BP |
2492 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2493 | if (flow_equal(&miss->flow, flow)) { | |
2494 | return miss; | |
2495 | } | |
2496 | } | |
abe529af | 2497 | |
501f8d1f BP |
2498 | miss = xmalloc(sizeof *miss); |
2499 | hmap_insert(todo, &miss->hmap_node, hash); | |
2500 | miss->flow = *flow; | |
b0f7b9b5 | 2501 | miss->key_fitness = key_fitness; |
501f8d1f BP |
2502 | miss->key = key; |
2503 | miss->key_len = key_len; | |
e84173dc | 2504 | miss->initial_tci = initial_tci; |
501f8d1f BP |
2505 | list_init(&miss->packets); |
2506 | return miss; | |
2507 | } | |
abe529af | 2508 | |
501f8d1f BP |
2509 | static void |
2510 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2511 | struct flow_miss_op *ops, size_t *n_ops) | |
2512 | { | |
2513 | const struct flow *flow = &miss->flow; | |
2514 | struct ofpbuf *packet, *next_packet; | |
b0f7b9b5 | 2515 | struct subfacet *subfacet; |
501f8d1f | 2516 | struct facet *facet; |
abe529af | 2517 | |
501f8d1f | 2518 | facet = facet_lookup_valid(ofproto, flow); |
abe529af | 2519 | if (!facet) { |
501f8d1f BP |
2520 | struct rule_dpif *rule; |
2521 | ||
2522 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af BP |
2523 | if (!rule) { |
2524 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
501f8d1f | 2525 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af BP |
2526 | if (port) { |
2527 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
2528 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
2529 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
2530 | return; |
2531 | } | |
2532 | } else { | |
2533 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
2534 | flow->in_port); |
2535 | } | |
2536 | ||
29ebe880 EJ |
2537 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2538 | send_packet_in_miss(ofproto, packet, flow); | |
abe529af BP |
2539 | } |
2540 | ||
abe529af BP |
2541 | return; |
2542 | } | |
2543 | ||
501f8d1f | 2544 | facet = facet_create(rule, flow); |
abe529af BP |
2545 | } |
2546 | ||
15baa734 | 2547 | subfacet = subfacet_create(facet, |
e84173dc BP |
2548 | miss->key_fitness, miss->key, miss->key_len, |
2549 | miss->initial_tci); | |
b0f7b9b5 | 2550 | |
501f8d1f | 2551 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) { |
67d91f78 | 2552 | struct dpif_flow_stats stats; |
999fba59 EJ |
2553 | struct flow_miss_op *op; |
2554 | struct dpif_execute *execute; | |
67d91f78 | 2555 | |
501f8d1f BP |
2556 | ofproto->n_matches++; |
2557 | ||
2558 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2559 | /* | |
2560 | * Extra-special case for fail-open mode. | |
2561 | * | |
2562 | * We are in fail-open mode and the packet matched the fail-open | |
2563 | * rule, but we are connected to a controller too. We should send | |
2564 | * the packet up to the controller in the hope that it will try to | |
2565 | * set up a flow and thereby allow us to exit fail-open. | |
2566 | * | |
2567 | * See the top-level comment in fail-open.c for more information. | |
2568 | */ | |
29ebe880 | 2569 | send_packet_in_miss(ofproto, packet, flow); |
501f8d1f BP |
2570 | } |
2571 | ||
b95fc6ba | 2572 | if (!facet->may_install || !subfacet->actions) { |
15baa734 | 2573 | subfacet_make_actions(subfacet, packet); |
501f8d1f | 2574 | } |
67d91f78 | 2575 | |
67d91f78 | 2576 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
15baa734 | 2577 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2578 | |
8338659a BP |
2579 | if (!subfacet->actions_len) { |
2580 | /* No actions to execute, so skip talking to the dpif. */ | |
2581 | continue; | |
2582 | } | |
2583 | ||
999fba59 EJ |
2584 | if (flow->vlan_tci != subfacet->initial_tci) { |
2585 | /* This packet was received on a VLAN splinter port. We added | |
2586 | * a VLAN to the packet to make the packet resemble the flow, | |
2587 | * but the actions were composed assuming that the packet | |
2588 | * contained no VLAN. So, we must remove the VLAN header from | |
2589 | * the packet before trying to execute the actions. */ | |
2590 | eth_pop_vlan(packet); | |
501f8d1f | 2591 | } |
999fba59 EJ |
2592 | |
2593 | op = &ops[(*n_ops)++]; | |
c2b565b5 | 2594 | execute = &op->dpif_op.u.execute; |
999fba59 | 2595 | op->subfacet = subfacet; |
c2b565b5 | 2596 | op->dpif_op.type = DPIF_OP_EXECUTE; |
999fba59 EJ |
2597 | execute->key = miss->key; |
2598 | execute->key_len = miss->key_len; | |
2599 | execute->actions = (facet->may_install | |
2600 | ? subfacet->actions | |
2601 | : xmemdup(subfacet->actions, | |
2602 | subfacet->actions_len)); | |
2603 | execute->actions_len = subfacet->actions_len; | |
2604 | execute->packet = packet; | |
501f8d1f BP |
2605 | } |
2606 | ||
b0f7b9b5 | 2607 | if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) { |
501f8d1f | 2608 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2609 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2610 | |
b0f7b9b5 | 2611 | op->subfacet = subfacet; |
c2b565b5 | 2612 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2613 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2614 | put->key = miss->key; | |
2615 | put->key_len = miss->key_len; | |
b95fc6ba BP |
2616 | put->actions = subfacet->actions; |
2617 | put->actions_len = subfacet->actions_len; | |
501f8d1f BP |
2618 | put->stats = NULL; |
2619 | } | |
2620 | } | |
2621 | ||
e2a6ca36 BP |
2622 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
2623 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
2624 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
2625 | * what a flow key should contain. | |
2626 | * | |
2627 | * This function also includes some logic to help make VLAN splinters | |
2628 | * transparent to the rest of the upcall processing logic. In particular, if | |
2629 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
2630 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
2631 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
2632 | * | |
2633 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
2634 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
2635 | * (This differs from the value returned in flow->vlan_tci only for packets | |
2636 | * received on VLAN splinters.) | |
2637 | */ | |
e84173dc | 2638 | static enum odp_key_fitness |
52a90c29 | 2639 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 2640 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
2641 | struct flow *flow, ovs_be16 *initial_tci, |
2642 | struct ofpbuf *packet) | |
e84173dc BP |
2643 | { |
2644 | enum odp_key_fitness fitness; | |
52a90c29 BP |
2645 | uint16_t realdev; |
2646 | int vid; | |
e84173dc BP |
2647 | |
2648 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
2649 | if (fitness == ODP_FIT_ERROR) { | |
2650 | return fitness; | |
2651 | } | |
2652 | *initial_tci = flow->vlan_tci; | |
2653 | ||
52a90c29 BP |
2654 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); |
2655 | if (realdev) { | |
2656 | /* Cause the flow to be processed as if it came in on the real device | |
2657 | * with the VLAN device's VLAN ID. */ | |
2658 | flow->in_port = realdev; | |
2659 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
e2a6ca36 BP |
2660 | if (packet) { |
2661 | /* Make the packet resemble the flow, so that it gets sent to an | |
2662 | * OpenFlow controller properly, so that it looks correct for | |
2663 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
2664 | * if it is called on 'packet'. | |
2665 | * | |
2666 | * The allocated space inside 'packet' probably also contains | |
2667 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
2668 | * struct dpif_upcall (see the large comment on that structure | |
2669 | * definition), so pushing data on 'packet' is in general not a | |
2670 | * good idea since it could overwrite 'key' or free it as a side | |
2671 | * effect. However, it's OK in this special case because we know | |
2672 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
2673 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
2674 | * since we don't need that header anymore. */ | |
2675 | eth_push_vlan(packet, flow->vlan_tci); | |
2676 | } | |
52a90c29 BP |
2677 | |
2678 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
2679 | if (fitness == ODP_FIT_PERFECT) { | |
2680 | fitness = ODP_FIT_TOO_MUCH; | |
2681 | } | |
2682 | } | |
2683 | ||
e84173dc BP |
2684 | return fitness; |
2685 | } | |
2686 | ||
501f8d1f BP |
2687 | static void |
2688 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2689 | size_t n_upcalls) | |
2690 | { | |
2691 | struct dpif_upcall *upcall; | |
2692 | struct flow_miss *miss, *next_miss; | |
2693 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
c2b565b5 | 2694 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f BP |
2695 | struct hmap todo; |
2696 | size_t n_ops; | |
2697 | size_t i; | |
2698 | ||
2699 | if (!n_upcalls) { | |
2700 | return; | |
2701 | } | |
2702 | ||
2703 | /* Construct the to-do list. | |
2704 | * | |
2705 | * This just amounts to extracting the flow from each packet and sticking | |
2706 | * the packets that have the same flow in the same "flow_miss" structure so | |
2707 | * that we can process them together. */ | |
2708 | hmap_init(&todo); | |
2709 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
b0f7b9b5 | 2710 | enum odp_key_fitness fitness; |
501f8d1f | 2711 | struct flow_miss *miss; |
e84173dc | 2712 | ovs_be16 initial_tci; |
501f8d1f BP |
2713 | struct flow flow; |
2714 | ||
b0f7b9b5 BP |
2715 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
2716 | * then set 'flow''s header pointers. */ | |
e84173dc BP |
2717 | fitness = ofproto_dpif_extract_flow_key(ofproto, |
2718 | upcall->key, upcall->key_len, | |
e2a6ca36 BP |
2719 | &flow, &initial_tci, |
2720 | upcall->packet); | |
b0f7b9b5 | 2721 | if (fitness == ODP_FIT_ERROR) { |
6b17ca5c | 2722 | ofpbuf_delete(upcall->packet); |
b0f7b9b5 BP |
2723 | continue; |
2724 | } | |
deedf7e7 | 2725 | flow_extract(upcall->packet, flow.skb_priority, flow.tun_id, |
abff858b | 2726 | flow.in_port, &flow); |
501f8d1f | 2727 | |
21f7563c | 2728 | /* Handle 802.1ag, LACP, and STP specially. */ |
501f8d1f | 2729 | if (process_special(ofproto, &flow, upcall->packet)) { |
6527c598 PS |
2730 | ofproto_update_local_port_stats(&ofproto->up, |
2731 | 0, upcall->packet->size); | |
501f8d1f BP |
2732 | ofpbuf_delete(upcall->packet); |
2733 | ofproto->n_matches++; | |
2734 | continue; | |
2735 | } | |
2736 | ||
2737 | /* Add other packets to a to-do list. */ | |
b0f7b9b5 | 2738 | miss = flow_miss_create(&todo, &flow, fitness, |
e84173dc | 2739 | upcall->key, upcall->key_len, initial_tci); |
501f8d1f BP |
2740 | list_push_back(&miss->packets, &upcall->packet->list_node); |
2741 | } | |
2742 | ||
2743 | /* Process each element in the to-do list, constructing the set of | |
2744 | * operations to batch. */ | |
2745 | n_ops = 0; | |
33bb0caa | 2746 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
501f8d1f | 2747 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); |
abe529af | 2748 | } |
501f8d1f | 2749 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
2750 | |
2751 | /* Execute batch. */ | |
2752 | for (i = 0; i < n_ops; i++) { | |
2753 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2754 | } | |
2755 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2756 | ||
2757 | /* Free memory and update facets. */ | |
2758 | for (i = 0; i < n_ops; i++) { | |
2759 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
2760 | struct dpif_execute *execute; | |
501f8d1f BP |
2761 | |
2762 | switch (op->dpif_op.type) { | |
2763 | case DPIF_OP_EXECUTE: | |
c2b565b5 | 2764 | execute = &op->dpif_op.u.execute; |
b95fc6ba | 2765 | if (op->subfacet->actions != execute->actions) { |
501f8d1f BP |
2766 | free((struct nlattr *) execute->actions); |
2767 | } | |
501f8d1f | 2768 | break; |
abe529af | 2769 | |
501f8d1f | 2770 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 2771 | if (!op->dpif_op.error) { |
b0f7b9b5 | 2772 | op->subfacet->installed = true; |
501f8d1f BP |
2773 | } |
2774 | break; | |
2775 | } | |
2776 | } | |
33bb0caa BP |
2777 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { |
2778 | ofpbuf_list_delete(&miss->packets); | |
2779 | hmap_remove(&todo, &miss->hmap_node); | |
2780 | free(miss); | |
2781 | } | |
2782 | hmap_destroy(&todo); | |
abe529af BP |
2783 | } |
2784 | ||
2785 | static void | |
6ff686f2 PS |
2786 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2787 | struct dpif_upcall *upcall) | |
abe529af | 2788 | { |
6ff686f2 | 2789 | struct user_action_cookie cookie; |
e84173dc BP |
2790 | enum odp_key_fitness fitness; |
2791 | ovs_be16 initial_tci; | |
2792 | struct flow flow; | |
abe529af | 2793 | |
6ff686f2 | 2794 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2795 | |
e84173dc BP |
2796 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
2797 | upcall->key_len, &flow, | |
e2a6ca36 | 2798 | &initial_tci, upcall->packet); |
e84173dc | 2799 | if (fitness == ODP_FIT_ERROR) { |
6b17ca5c | 2800 | ofpbuf_delete(upcall->packet); |
e84173dc BP |
2801 | return; |
2802 | } | |
2803 | ||
6ff686f2 | 2804 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af | 2805 | if (ofproto->sflow) { |
e84173dc BP |
2806 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
2807 | &cookie); | |
abe529af | 2808 | } |
6ff686f2 PS |
2809 | } else { |
2810 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2811 | } | |
29ebe880 | 2812 | ofpbuf_delete(upcall->packet); |
6ff686f2 PS |
2813 | } |
2814 | ||
9b16c439 BP |
2815 | static int |
2816 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 2817 | { |
9b16c439 BP |
2818 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
2819 | int n_misses; | |
2820 | int i; | |
abe529af | 2821 | |
9b16c439 | 2822 | assert (max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 2823 | |
9b16c439 BP |
2824 | n_misses = 0; |
2825 | for (i = 0; i < max_batch; i++) { | |
2826 | struct dpif_upcall *upcall = &misses[n_misses]; | |
2827 | int error; | |
2828 | ||
2829 | error = dpif_recv(ofproto->dpif, upcall); | |
2830 | if (error) { | |
9b16c439 BP |
2831 | break; |
2832 | } | |
2833 | ||
2834 | switch (upcall->type) { | |
2835 | case DPIF_UC_ACTION: | |
2836 | handle_userspace_upcall(ofproto, upcall); | |
2837 | break; | |
2838 | ||
2839 | case DPIF_UC_MISS: | |
2840 | /* Handle it later. */ | |
2841 | n_misses++; | |
2842 | break; | |
2843 | ||
2844 | case DPIF_N_UC_TYPES: | |
2845 | default: | |
2846 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, | |
2847 | upcall->type); | |
2848 | break; | |
2849 | } | |
abe529af | 2850 | } |
9b16c439 BP |
2851 | |
2852 | handle_miss_upcalls(ofproto, misses, n_misses); | |
2853 | ||
2854 | return i; | |
abe529af BP |
2855 | } |
2856 | \f | |
2857 | /* Flow expiration. */ | |
2858 | ||
b0f7b9b5 | 2859 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
2860 | static void update_stats(struct ofproto_dpif *); |
2861 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 2862 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
2863 | |
2864 | /* This function is called periodically by run(). Its job is to collect | |
2865 | * updates for the flows that have been installed into the datapath, most | |
2866 | * importantly when they last were used, and then use that information to | |
2867 | * expire flows that have not been used recently. | |
2868 | * | |
2869 | * Returns the number of milliseconds after which it should be called again. */ | |
2870 | static int | |
2871 | expire(struct ofproto_dpif *ofproto) | |
2872 | { | |
2873 | struct rule_dpif *rule, *next_rule; | |
d0918789 | 2874 | struct oftable *table; |
abe529af BP |
2875 | int dp_max_idle; |
2876 | ||
2877 | /* Update stats for each flow in the datapath. */ | |
2878 | update_stats(ofproto); | |
2879 | ||
b0f7b9b5 BP |
2880 | /* Expire subfacets that have been idle too long. */ |
2881 | dp_max_idle = subfacet_max_idle(ofproto); | |
2882 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
2883 | |
2884 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2885 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2886 | struct cls_cursor cursor; | |
2887 | ||
d0918789 | 2888 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
2889 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
2890 | rule_expire(rule); | |
2891 | } | |
abe529af BP |
2892 | } |
2893 | ||
2894 | /* All outstanding data in existing flows has been accounted, so it's a | |
2895 | * good time to do bond rebalancing. */ | |
2896 | if (ofproto->has_bonded_bundles) { | |
2897 | struct ofbundle *bundle; | |
2898 | ||
2899 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2900 | if (bundle->bond) { | |
2901 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2902 | } | |
2903 | } | |
2904 | } | |
2905 | ||
2906 | return MIN(dp_max_idle, 1000); | |
2907 | } | |
2908 | ||
2909 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2910 | * | |
2911 | * This function also pushes statistics updates to rules which each facet | |
2912 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2913 | * facet changes the rule it resubmits into at some time in between | |
2914 | * update_stats() runs, it is possible that statistics accrued to the | |
2915 | * old rule will be incorrectly attributed to the new rule. This could be | |
2916 | * avoided by calling update_stats() whenever rules are created or | |
2917 | * deleted. However, the performance impact of making so many calls to the | |
2918 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2919 | */ | |
2920 | static void | |
2921 | update_stats(struct ofproto_dpif *p) | |
2922 | { | |
2923 | const struct dpif_flow_stats *stats; | |
2924 | struct dpif_flow_dump dump; | |
2925 | const struct nlattr *key; | |
2926 | size_t key_len; | |
2927 | ||
2928 | dpif_flow_dump_start(&dump, p->dpif); | |
2929 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 2930 | struct subfacet *subfacet; |
abe529af | 2931 | |
6a542738 | 2932 | subfacet = subfacet_find(p, key, key_len); |
b0f7b9b5 BP |
2933 | if (subfacet && subfacet->installed) { |
2934 | struct facet *facet = subfacet->facet; | |
abe529af | 2935 | |
b0f7b9b5 BP |
2936 | if (stats->n_packets >= subfacet->dp_packet_count) { |
2937 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
abe529af BP |
2938 | facet->packet_count += extra; |
2939 | } else { | |
2940 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
2941 | } | |
2942 | ||
b0f7b9b5 BP |
2943 | if (stats->n_bytes >= subfacet->dp_byte_count) { |
2944 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
abe529af BP |
2945 | } else { |
2946 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
2947 | } | |
2948 | ||
b0f7b9b5 BP |
2949 | subfacet->dp_packet_count = stats->n_packets; |
2950 | subfacet->dp_byte_count = stats->n_bytes; | |
abe529af | 2951 | |
15baa734 BP |
2952 | subfacet_update_time(subfacet, stats->used); |
2953 | facet_account(facet); | |
abe529af BP |
2954 | facet_push_stats(facet); |
2955 | } else { | |
6a542738 PS |
2956 | if (!VLOG_DROP_WARN(&rl)) { |
2957 | struct ds s; | |
2958 | ||
2959 | ds_init(&s); | |
2960 | odp_flow_key_format(key, key_len, &s); | |
2961 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
2962 | ds_destroy(&s); | |
2963 | } | |
2964 | ||
2965 | COVERAGE_INC(facet_unexpected); | |
b0f7b9b5 BP |
2966 | /* There's a flow in the datapath that we know nothing about, or a |
2967 | * flow that shouldn't be installed but was anyway. Delete it. */ | |
abe529af BP |
2968 | dpif_flow_del(p->dpif, key, key_len, NULL); |
2969 | } | |
2970 | } | |
2971 | dpif_flow_dump_done(&dump); | |
2972 | } | |
2973 | ||
2974 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
2975 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
2976 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
2977 | * fold its statistic into its rule. */ | |
abe529af | 2978 | static int |
b0f7b9b5 | 2979 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
2980 | { |
2981 | /* | |
2982 | * Idle time histogram. | |
2983 | * | |
b0f7b9b5 BP |
2984 | * Most of the time a switch has a relatively small number of subfacets. |
2985 | * When this is the case we might as well keep statistics for all of them | |
2986 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
2987 | * well. |
2988 | * | |
b0f7b9b5 | 2989 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 2990 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
2991 | * significant. However, with a large number of subfacets it is likely |
2992 | * that only a few of them are "heavy hitters" that consume a large amount | |
2993 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
2994 | * the kernel and maintaining in userspaces; other subfacets we can | |
2995 | * discard. | |
abe529af BP |
2996 | * |
2997 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 2998 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
2999 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3000 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3001 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 3002 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3003 | * the most-recently-used bucket of subfacets is kept, so actually an |
3004 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3005 | * (though the next run will delete most of those unless they receive |
3006 | * additional data). | |
abe529af | 3007 | * |
b0f7b9b5 BP |
3008 | * This requires a second pass through the subfacets, in addition to the |
3009 | * pass made by update_stats(), because the former function never looks at | |
3010 | * uninstallable subfacets. | |
abe529af BP |
3011 | */ |
3012 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3013 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3014 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3015 | int total, subtotal, bucket; |
b0f7b9b5 | 3016 | struct subfacet *subfacet; |
abe529af BP |
3017 | long long int now; |
3018 | int i; | |
3019 | ||
b0f7b9b5 | 3020 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3021 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3022 | return N_BUCKETS * BUCKET_WIDTH; |
3023 | } | |
3024 | ||
3025 | /* Build histogram. */ | |
3026 | now = time_msec(); | |
b0f7b9b5 BP |
3027 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3028 | long long int idle = now - subfacet->used; | |
abe529af BP |
3029 | int bucket = (idle <= 0 ? 0 |
3030 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3031 | : (unsigned int) idle / BUCKET_WIDTH); | |
3032 | buckets[bucket]++; | |
3033 | } | |
3034 | ||
3035 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3036 | subtotal = bucket = 0; |
3037 | do { | |
3038 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3039 | } while (bucket < N_BUCKETS && |
3040 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3041 | |
3042 | if (VLOG_IS_DBG_ENABLED()) { | |
3043 | struct ds s; | |
3044 | ||
3045 | ds_init(&s); | |
3046 | ds_put_cstr(&s, "keep"); | |
3047 | for (i = 0; i < N_BUCKETS; i++) { | |
3048 | if (i == bucket) { | |
3049 | ds_put_cstr(&s, ", drop"); | |
3050 | } | |
3051 | if (buckets[i]) { | |
3052 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3053 | } | |
3054 | } | |
3055 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3056 | ds_destroy(&s); | |
3057 | } | |
3058 | ||
3059 | return bucket * BUCKET_WIDTH; | |
3060 | } | |
3061 | ||
abe529af | 3062 | static void |
b0f7b9b5 | 3063 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af BP |
3064 | { |
3065 | long long int cutoff = time_msec() - dp_max_idle; | |
b0f7b9b5 | 3066 | struct subfacet *subfacet, *next_subfacet; |
abe529af | 3067 | |
b0f7b9b5 BP |
3068 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3069 | &ofproto->subfacets) { | |
3070 | if (subfacet->used < cutoff) { | |
15baa734 | 3071 | subfacet_destroy(subfacet); |
abe529af BP |
3072 | } |
3073 | } | |
3074 | } | |
3075 | ||
3076 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3077 | * then delete it entirely. */ | |
3078 | static void | |
3079 | rule_expire(struct rule_dpif *rule) | |
3080 | { | |
abe529af BP |
3081 | struct facet *facet, *next_facet; |
3082 | long long int now; | |
3083 | uint8_t reason; | |
3084 | ||
3085 | /* Has 'rule' expired? */ | |
3086 | now = time_msec(); | |
3087 | if (rule->up.hard_timeout | |
308881af | 3088 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af BP |
3089 | reason = OFPRR_HARD_TIMEOUT; |
3090 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
3091 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
3092 | reason = OFPRR_IDLE_TIMEOUT; | |
3093 | } else { | |
3094 | return; | |
3095 | } | |
3096 | ||
3097 | COVERAGE_INC(ofproto_dpif_expired); | |
3098 | ||
3099 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3100 | * timeout, because that only happens when the rule has no facets left.) */ | |
3101 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3102 | facet_remove(facet); |
abe529af BP |
3103 | } |
3104 | ||
3105 | /* Get rid of the rule. */ | |
3106 | ofproto_rule_expire(&rule->up, reason); | |
3107 | } | |
3108 | \f | |
3109 | /* Facets. */ | |
3110 | ||
f3827897 | 3111 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3112 | * |
3113 | * The caller must already have determined that no facet with an identical | |
3114 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3115 | * the ofproto's classifier table. |
3116 | * | |
b0f7b9b5 BP |
3117 | * The facet will initially have no subfacets. The caller should create (at |
3118 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3119 | static struct facet * |
f3827897 | 3120 | facet_create(struct rule_dpif *rule, const struct flow *flow) |
abe529af BP |
3121 | { |
3122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3123 | struct facet *facet; | |
3124 | ||
3125 | facet = xzalloc(sizeof *facet); | |
3126 | facet->used = time_msec(); | |
3127 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
3128 | list_push_back(&rule->facets, &facet->list_node); | |
3129 | facet->rule = rule; | |
3130 | facet->flow = *flow; | |
b0f7b9b5 | 3131 | list_init(&facet->subfacets); |
abe529af BP |
3132 | netflow_flow_init(&facet->nf_flow); |
3133 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3134 | ||
abe529af BP |
3135 | return facet; |
3136 | } | |
3137 | ||
3138 | static void | |
3139 | facet_free(struct facet *facet) | |
3140 | { | |
abe529af BP |
3141 | free(facet); |
3142 | } | |
3143 | ||
3d9e05f8 BP |
3144 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3145 | * 'packet', which arrived on 'in_port'. | |
3146 | * | |
3147 | * Takes ownership of 'packet'. */ | |
3148 | static bool | |
3149 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3150 | const struct nlattr *odp_actions, size_t actions_len, | |
3151 | struct ofpbuf *packet) | |
3152 | { | |
3153 | struct odputil_keybuf keybuf; | |
3154 | struct ofpbuf key; | |
3155 | int error; | |
3156 | ||
6ff686f2 PS |
3157 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3158 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3159 | |
6ff686f2 PS |
3160 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3161 | odp_actions, actions_len, packet); | |
80e5eed9 | 3162 | |
6ff686f2 PS |
3163 | ofpbuf_delete(packet); |
3164 | return !error; | |
abe529af BP |
3165 | } |
3166 | ||
abe529af BP |
3167 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3168 | * | |
3169 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3170 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3171 | * |
3172 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3173 | */ | |
3174 | static void | |
15baa734 | 3175 | facet_remove(struct facet *facet) |
abe529af | 3176 | { |
15baa734 | 3177 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3178 | struct subfacet *subfacet, *next_subfacet; |
3179 | ||
551a2f6c BP |
3180 | assert(!list_is_empty(&facet->subfacets)); |
3181 | ||
3182 | /* First uninstall all of the subfacets to get final statistics. */ | |
3183 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3184 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3185 | } |
3186 | ||
3187 | /* Flush the final stats to the rule. | |
3188 | * | |
3189 | * This might require us to have at least one subfacet around so that we | |
3190 | * can use its actions for accounting in facet_account(), which is why we | |
3191 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3192 | facet_flush_stats(facet); |
551a2f6c BP |
3193 | |
3194 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3195 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3196 | &facet->subfacets) { | |
15baa734 | 3197 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3198 | } |
abe529af BP |
3199 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3200 | list_remove(&facet->list_node); | |
3201 | facet_free(facet); | |
3202 | } | |
3203 | ||
abe529af | 3204 | static void |
15baa734 | 3205 | facet_account(struct facet *facet) |
abe529af | 3206 | { |
15baa734 | 3207 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
55af77bb | 3208 | uint64_t n_bytes; |
b95fc6ba | 3209 | struct subfacet *subfacet; |
abe529af | 3210 | const struct nlattr *a; |
abe529af | 3211 | unsigned int left; |
d78be13b | 3212 | ovs_be16 vlan_tci; |
abe529af | 3213 | |
55af77bb | 3214 | if (facet->byte_count <= facet->accounted_bytes) { |
abe529af BP |
3215 | return; |
3216 | } | |
55af77bb EJ |
3217 | n_bytes = facet->byte_count - facet->accounted_bytes; |
3218 | facet->accounted_bytes = facet->byte_count; | |
abe529af | 3219 | |
75a75043 | 3220 | /* Feed information from the active flows back into the learning table to |
abe529af BP |
3221 | * ensure that table is always in sync with what is actually flowing |
3222 | * through the datapath. */ | |
75a75043 BP |
3223 | if (facet->has_learn || facet->has_normal) { |
3224 | struct action_xlate_ctx ctx; | |
abe529af | 3225 | |
e84173dc | 3226 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
54834960 EJ |
3227 | facet->flow.vlan_tci, |
3228 | facet->rule->up.flow_cookie, NULL); | |
75a75043 BP |
3229 | ctx.may_learn = true; |
3230 | ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, | |
3231 | facet->rule->up.n_actions)); | |
3232 | } | |
abe529af | 3233 | |
75a75043 | 3234 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3235 | return; |
3236 | } | |
d78be13b BP |
3237 | |
3238 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3239 | * as a basis. We also need to track the actual VLAN on which the packet | |
3240 | * is going to be sent to ensure that it matches the one passed to | |
3241 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3242 | * hash bucket.) |
3243 | * | |
3244 | * We use the actions from an arbitrary subfacet because they should all | |
3245 | * be equally valid for our purpose. */ | |
3246 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3247 | struct subfacet, list_node); | |
d78be13b | 3248 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3249 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3250 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3251 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3252 | struct ofport_dpif *port; |
abe529af | 3253 | |
d78be13b | 3254 | switch (nl_attr_type(a)) { |
df2c07f4 | 3255 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3256 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3257 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3258 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3259 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3260 | } |
d78be13b BP |
3261 | break; |
3262 | ||
fea393b1 BP |
3263 | case OVS_ACTION_ATTR_POP_VLAN: |
3264 | vlan_tci = htons(0); | |
d78be13b BP |
3265 | break; |
3266 | ||
fea393b1 BP |
3267 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3268 | vlan = nl_attr_get(a); | |
3269 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3270 | break; |
abe529af BP |
3271 | } |
3272 | } | |
3273 | } | |
3274 | ||
abe529af BP |
3275 | /* Returns true if the only action for 'facet' is to send to the controller. |
3276 | * (We don't report NetFlow expiration messages for such facets because they | |
3277 | * are just part of the control logic for the network, not real traffic). */ | |
3278 | static bool | |
3279 | facet_is_controller_flow(struct facet *facet) | |
3280 | { | |
3281 | return (facet | |
3282 | && facet->rule->up.n_actions == 1 | |
3283 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
3284 | htons(OFPP_CONTROLLER))); | |
3285 | } | |
3286 | ||
3287 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3288 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3289 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3290 | * its packet and byte counts before this function is called. */ | |
3291 | static void | |
15baa734 | 3292 | facet_flush_stats(struct facet *facet) |
abe529af | 3293 | { |
15baa734 | 3294 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3295 | struct subfacet *subfacet; |
3296 | ||
3297 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3298 | assert(!subfacet->dp_byte_count); | |
3299 | assert(!subfacet->dp_packet_count); | |
3300 | } | |
abe529af BP |
3301 | |
3302 | facet_push_stats(facet); | |
15baa734 | 3303 | facet_account(facet); |
abe529af BP |
3304 | |
3305 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3306 | struct ofexpired expired; | |
3307 | expired.flow = facet->flow; | |
3308 | expired.packet_count = facet->packet_count; | |
3309 | expired.byte_count = facet->byte_count; | |
3310 | expired.used = facet->used; | |
3311 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3312 | } | |
3313 | ||
3314 | facet->rule->packet_count += facet->packet_count; | |
3315 | facet->rule->byte_count += facet->byte_count; | |
3316 | ||
3317 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3318 | * reinstalled. */ | |
bbb5d219 | 3319 | facet_reset_counters(facet); |
abe529af BP |
3320 | |
3321 | netflow_flow_clear(&facet->nf_flow); | |
3322 | } | |
3323 | ||
3324 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3325 | * Returns it if found, otherwise a null pointer. | |
3326 | * | |
3327 | * The returned facet might need revalidation; use facet_lookup_valid() | |
3328 | * instead if that is important. */ | |
3329 | static struct facet * | |
3330 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3331 | { | |
3332 | struct facet *facet; | |
3333 | ||
3334 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
3335 | &ofproto->facets) { | |
3336 | if (flow_equal(flow, &facet->flow)) { | |
3337 | return facet; | |
3338 | } | |
3339 | } | |
3340 | ||
3341 | return NULL; | |
3342 | } | |
3343 | ||
3344 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3345 | * Returns it if found, otherwise a null pointer. | |
3346 | * | |
3347 | * The returned facet is guaranteed to be valid. */ | |
3348 | static struct facet * | |
3349 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3350 | { | |
3351 | struct facet *facet = facet_find(ofproto, flow); | |
3352 | ||
3353 | /* The facet we found might not be valid, since we could be in need of | |
3354 | * revalidation. If it is not valid, don't return it. */ | |
3355 | if (facet | |
0e4b3771 BP |
3356 | && (ofproto->need_revalidate |
3357 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
15baa734 | 3358 | && !facet_revalidate(facet)) { |
abe529af BP |
3359 | COVERAGE_INC(facet_invalidated); |
3360 | return NULL; | |
3361 | } | |
3362 | ||
3363 | return facet; | |
3364 | } | |
3365 | ||
6814e51f BP |
3366 | static bool |
3367 | facet_check_consistency(struct facet *facet) | |
3368 | { | |
3369 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3370 | ||
3371 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3372 | ||
3373 | struct rule_dpif *rule; | |
3374 | struct subfacet *subfacet; | |
c53e1132 | 3375 | bool may_log = false; |
6814e51f BP |
3376 | bool ok; |
3377 | ||
3378 | /* Check the rule for consistency. */ | |
3379 | rule = rule_dpif_lookup(ofproto, &facet->flow, 0); | |
3380 | if (!rule) { | |
3381 | if (!VLOG_DROP_WARN(&rl)) { | |
3382 | char *s = flow_to_string(&facet->flow); | |
3383 | VLOG_WARN("%s: facet should not exist", s); | |
3384 | free(s); | |
3385 | } | |
3386 | return false; | |
3387 | } else if (rule != facet->rule) { | |
c53e1132 BP |
3388 | may_log = !VLOG_DROP_WARN(&rl); |
3389 | ok = false; | |
3390 | if (may_log) { | |
3391 | struct ds s; | |
6814e51f | 3392 | |
c53e1132 BP |
3393 | ds_init(&s); |
3394 | flow_format(&s, &facet->flow); | |
3395 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3396 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3397 | cls_rule_format(&facet->rule->up.cr, &s); | |
3398 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3399 | rule->up.table_id); | |
3400 | cls_rule_format(&rule->up.cr, &s); | |
3401 | ds_put_char(&s, ')'); | |
6814e51f | 3402 | |
c53e1132 BP |
3403 | VLOG_WARN("%s", ds_cstr(&s)); |
3404 | ds_destroy(&s); | |
3405 | } | |
6814e51f BP |
3406 | } else { |
3407 | ok = true; | |
3408 | } | |
3409 | ||
3410 | /* Check the datapath actions for consistency. */ | |
3411 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3412 | struct action_xlate_ctx ctx; | |
3413 | struct ofpbuf *odp_actions; | |
3414 | bool actions_changed; | |
3415 | bool should_install; | |
3416 | ||
3417 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
3418 | subfacet->initial_tci, rule->up.flow_cookie, | |
3419 | NULL); | |
3420 | odp_actions = xlate_actions(&ctx, rule->up.actions, | |
3421 | rule->up.n_actions); | |
3422 | ||
3423 | should_install = (ctx.may_set_up_flow | |
3424 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
3425 | if (!should_install && !subfacet->installed) { | |
3426 | /* The actions for uninstallable flows may vary from one packet to | |
3427 | * the next, so don't compare the actions. */ | |
3428 | goto next; | |
3429 | } | |
3430 | ||
3431 | actions_changed = (subfacet->actions_len != odp_actions->size | |
3432 | || memcmp(subfacet->actions, odp_actions->data, | |
3433 | subfacet->actions_len)); | |
3434 | if (should_install != subfacet->installed || actions_changed) { | |
c53e1132 BP |
3435 | if (ok) { |
3436 | may_log = !VLOG_DROP_WARN(&rl); | |
3437 | ok = false; | |
3438 | } | |
6814e51f | 3439 | |
c53e1132 BP |
3440 | if (may_log) { |
3441 | struct odputil_keybuf keybuf; | |
3442 | struct ofpbuf key; | |
3443 | struct ds s; | |
6814e51f | 3444 | |
c53e1132 BP |
3445 | ds_init(&s); |
3446 | subfacet_get_key(subfacet, &keybuf, &key); | |
3447 | odp_flow_key_format(key.data, key.size, &s); | |
3448 | ||
3449 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
3450 | if (should_install != subfacet->installed) { | |
3451 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
3452 | ||
3453 | ds_put_format(&s, " (should%s have been installed)", | |
3454 | should_install ? "" : " not"); | |
3455 | ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)", | |
3456 | ctx.may_set_up_flow ? "true" : "false", | |
3457 | odp_key_fitness_to_string(fitness)); | |
3458 | } | |
3459 | if (actions_changed) { | |
3460 | ds_put_cstr(&s, " (actions were: "); | |
3461 | format_odp_actions(&s, subfacet->actions, | |
3462 | subfacet->actions_len); | |
3463 | ds_put_cstr(&s, ") (correct actions: "); | |
3464 | format_odp_actions(&s, odp_actions->data, | |
3465 | odp_actions->size); | |
3466 | ds_put_char(&s, ')'); | |
3467 | } else { | |
3468 | ds_put_cstr(&s, " (actions: "); | |
3469 | format_odp_actions(&s, subfacet->actions, | |
3470 | subfacet->actions_len); | |
3471 | ds_put_char(&s, ')'); | |
3472 | } | |
3473 | VLOG_WARN("%s", ds_cstr(&s)); | |
3474 | ds_destroy(&s); | |
6814e51f | 3475 | } |
6814e51f BP |
3476 | } |
3477 | ||
3478 | next: | |
3479 | ofpbuf_delete(odp_actions); | |
3480 | } | |
3481 | ||
3482 | return ok; | |
3483 | } | |
3484 | ||
15baa734 | 3485 | /* Re-searches the classifier for 'facet': |
abe529af BP |
3486 | * |
3487 | * - If the rule found is different from 'facet''s current rule, moves | |
3488 | * 'facet' to the new rule and recompiles its actions. | |
3489 | * | |
3490 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3491 | * where it is and recompiles its actions anyway. | |
3492 | * | |
3493 | * - If there is none, destroys 'facet'. | |
3494 | * | |
3495 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3496 | static bool | |
15baa734 | 3497 | facet_revalidate(struct facet *facet) |
abe529af | 3498 | { |
15baa734 | 3499 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
3500 | struct actions { |
3501 | struct nlattr *odp_actions; | |
3502 | size_t actions_len; | |
3503 | }; | |
3504 | struct actions *new_actions; | |
3505 | ||
abe529af | 3506 | struct action_xlate_ctx ctx; |
abe529af | 3507 | struct rule_dpif *new_rule; |
b0f7b9b5 | 3508 | struct subfacet *subfacet; |
abe529af | 3509 | bool actions_changed; |
b95fc6ba | 3510 | int i; |
abe529af BP |
3511 | |
3512 | COVERAGE_INC(facet_revalidate); | |
3513 | ||
3514 | /* Determine the new rule. */ | |
29901626 | 3515 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3516 | if (!new_rule) { |
3517 | /* No new rule, so delete the facet. */ | |
15baa734 | 3518 | facet_remove(facet); |
abe529af BP |
3519 | return false; |
3520 | } | |
3521 | ||
df2c07f4 | 3522 | /* Calculate new datapath actions. |
abe529af BP |
3523 | * |
3524 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3525 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3526 | * around to properly compose it. */ | |
abe529af | 3527 | |
df2c07f4 JP |
3528 | /* If the datapath actions changed or the installability changed, |
3529 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
3530 | i = 0; |
3531 | new_actions = NULL; | |
3532 | memset(&ctx, 0, sizeof ctx); | |
b0f7b9b5 | 3533 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
b95fc6ba BP |
3534 | struct ofpbuf *odp_actions; |
3535 | bool should_install; | |
3536 | ||
e84173dc | 3537 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
54834960 EJ |
3538 | subfacet->initial_tci, new_rule->up.flow_cookie, |
3539 | NULL); | |
b95fc6ba BP |
3540 | odp_actions = xlate_actions(&ctx, new_rule->up.actions, |
3541 | new_rule->up.n_actions); | |
3542 | actions_changed = (subfacet->actions_len != odp_actions->size | |
3543 | || memcmp(subfacet->actions, odp_actions->data, | |
3544 | subfacet->actions_len)); | |
3545 | ||
3546 | should_install = (ctx.may_set_up_flow | |
3547 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
b0f7b9b5 BP |
3548 | if (actions_changed || should_install != subfacet->installed) { |
3549 | if (should_install) { | |
3550 | struct dpif_flow_stats stats; | |
3551 | ||
15baa734 | 3552 | subfacet_install(subfacet, |
b0f7b9b5 | 3553 | odp_actions->data, odp_actions->size, &stats); |
15baa734 | 3554 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 3555 | } else { |
15baa734 | 3556 | subfacet_uninstall(subfacet); |
b0f7b9b5 | 3557 | } |
b95fc6ba BP |
3558 | |
3559 | if (!new_actions) { | |
3560 | new_actions = xcalloc(list_size(&facet->subfacets), | |
3561 | sizeof *new_actions); | |
3562 | } | |
3563 | new_actions[i].odp_actions = xmemdup(odp_actions->data, | |
3564 | odp_actions->size); | |
3565 | new_actions[i].actions_len = odp_actions->size; | |
abe529af | 3566 | } |
b95fc6ba BP |
3567 | |
3568 | ofpbuf_delete(odp_actions); | |
3569 | i++; | |
b0f7b9b5 | 3570 | } |
b95fc6ba | 3571 | if (new_actions) { |
15baa734 | 3572 | facet_flush_stats(facet); |
abe529af BP |
3573 | } |
3574 | ||
3575 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3576 | facet->tags = ctx.tags; | |
3577 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3578 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3579 | facet->has_learn = ctx.has_learn; |
3580 | facet->has_normal = ctx.has_normal; | |
9d24de3b | 3581 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3582 | if (new_actions) { |
3583 | i = 0; | |
3584 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3585 | if (new_actions[i].odp_actions) { | |
3586 | free(subfacet->actions); | |
3587 | subfacet->actions = new_actions[i].odp_actions; | |
3588 | subfacet->actions_len = new_actions[i].actions_len; | |
3589 | } | |
3590 | i++; | |
3591 | } | |
3592 | free(new_actions); | |
abe529af BP |
3593 | } |
3594 | if (facet->rule != new_rule) { | |
3595 | COVERAGE_INC(facet_changed_rule); | |
3596 | list_remove(&facet->list_node); | |
3597 | list_push_back(&new_rule->facets, &facet->list_node); | |
3598 | facet->rule = new_rule; | |
3599 | facet->used = new_rule->up.created; | |
9d24de3b | 3600 | facet->prev_used = facet->used; |
abe529af BP |
3601 | } |
3602 | ||
abe529af BP |
3603 | return true; |
3604 | } | |
3605 | ||
3606 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3607 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3608 | static void | |
15baa734 | 3609 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 3610 | { |
15baa734 | 3611 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
3612 | if (used > facet->used) { |
3613 | facet->used = used; | |
3614 | if (used > facet->rule->used) { | |
3615 | facet->rule->used = used; | |
3616 | } | |
3617 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
3618 | } | |
3619 | } | |
3620 | ||
bbb5d219 EJ |
3621 | static void |
3622 | facet_reset_counters(struct facet *facet) | |
3623 | { | |
3624 | facet->packet_count = 0; | |
3625 | facet->byte_count = 0; | |
9d24de3b JP |
3626 | facet->prev_packet_count = 0; |
3627 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
3628 | facet->accounted_bytes = 0; |
3629 | } | |
3630 | ||
abe529af BP |
3631 | static void |
3632 | facet_push_stats(struct facet *facet) | |
3633 | { | |
9d24de3b | 3634 | uint64_t new_packets, new_bytes; |
abe529af | 3635 | |
9d24de3b JP |
3636 | assert(facet->packet_count >= facet->prev_packet_count); |
3637 | assert(facet->byte_count >= facet->prev_byte_count); | |
3638 | assert(facet->used >= facet->prev_used); | |
abe529af | 3639 | |
9d24de3b JP |
3640 | new_packets = facet->packet_count - facet->prev_packet_count; |
3641 | new_bytes = facet->byte_count - facet->prev_byte_count; | |
abe529af | 3642 | |
9d24de3b JP |
3643 | if (new_packets || new_bytes || facet->used > facet->prev_used) { |
3644 | facet->prev_packet_count = facet->packet_count; | |
3645 | facet->prev_byte_count = facet->byte_count; | |
3646 | facet->prev_used = facet->used; | |
abe529af BP |
3647 | |
3648 | flow_push_stats(facet->rule, &facet->flow, | |
9d24de3b JP |
3649 | new_packets, new_bytes, facet->used); |
3650 | ||
3651 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
3652 | facet->mirrors, new_packets, new_bytes); | |
abe529af BP |
3653 | } |
3654 | } | |
3655 | ||
3656 | struct ofproto_push { | |
3657 | struct action_xlate_ctx ctx; | |
3658 | uint64_t packets; | |
3659 | uint64_t bytes; | |
3660 | long long int used; | |
3661 | }; | |
3662 | ||
3663 | static void | |
3664 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3665 | { | |
3666 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3667 | ||
3668 | if (rule) { | |
3669 | rule->packet_count += push->packets; | |
3670 | rule->byte_count += push->bytes; | |
3671 | rule->used = MAX(push->used, rule->used); | |
3672 | } | |
3673 | } | |
3674 | ||
3675 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 3676 | * 'rule''s actions and mirrors. */ |
abe529af BP |
3677 | static void |
3678 | flow_push_stats(const struct rule_dpif *rule, | |
59d0f2c8 | 3679 | const struct flow *flow, uint64_t packets, uint64_t bytes, |
abe529af BP |
3680 | long long int used) |
3681 | { | |
3682 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3683 | struct ofproto_push push; | |
3684 | ||
3685 | push.packets = packets; | |
3686 | push.bytes = bytes; | |
3687 | push.used = used; | |
3688 | ||
54834960 EJ |
3689 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, |
3690 | rule->up.flow_cookie, NULL); | |
abe529af BP |
3691 | push.ctx.resubmit_hook = push_resubmit; |
3692 | ofpbuf_delete(xlate_actions(&push.ctx, | |
3693 | rule->up.actions, rule->up.n_actions)); | |
3694 | } | |
3695 | \f | |
b0f7b9b5 BP |
3696 | /* Subfacets. */ |
3697 | ||
3698 | static struct subfacet * | |
3699 | subfacet_find__(struct ofproto_dpif *ofproto, | |
3700 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
3701 | const struct flow *flow) | |
3702 | { | |
3703 | struct subfacet *subfacet; | |
3704 | ||
3705 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
3706 | &ofproto->subfacets) { | |
3707 | if (subfacet->key | |
3708 | ? (subfacet->key_len == key_len | |
3709 | && !memcmp(key, subfacet->key, key_len)) | |
3710 | : flow_equal(flow, &subfacet->facet->flow)) { | |
3711 | return subfacet; | |
3712 | } | |
3713 | } | |
3714 | ||
3715 | return NULL; | |
3716 | } | |
3717 | ||
3718 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
3719 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
3720 | * there is one, otherwise creates and returns a new subfacet. |
3721 | * | |
3722 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
3723 | * which case the caller must populate the actions with | |
3724 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 3725 | static struct subfacet * |
15baa734 | 3726 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
e84173dc | 3727 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) |
b0f7b9b5 | 3728 | { |
15baa734 | 3729 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3730 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
3731 | struct subfacet *subfacet; | |
3732 | ||
3733 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); | |
3734 | if (subfacet) { | |
3735 | if (subfacet->facet == facet) { | |
3736 | return subfacet; | |
3737 | } | |
3738 | ||
3739 | /* This shouldn't happen. */ | |
3740 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
15baa734 | 3741 | subfacet_destroy(subfacet); |
b0f7b9b5 BP |
3742 | } |
3743 | ||
3744 | subfacet = xzalloc(sizeof *subfacet); | |
3745 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); | |
3746 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
3747 | subfacet->facet = facet; | |
3748 | subfacet->used = time_msec(); | |
3749 | subfacet->key_fitness = key_fitness; | |
3750 | if (key_fitness != ODP_FIT_PERFECT) { | |
3751 | subfacet->key = xmemdup(key, key_len); | |
3752 | subfacet->key_len = key_len; | |
3753 | } | |
3754 | subfacet->installed = false; | |
e84173dc | 3755 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
3756 | |
3757 | return subfacet; | |
3758 | } | |
3759 | ||
3760 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
3761 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
3762 | static struct subfacet * | |
3763 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 3764 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
3765 | { |
3766 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
3767 | enum odp_key_fitness fitness; |
3768 | struct flow flow; | |
3769 | ||
3770 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
3771 | if (fitness == ODP_FIT_ERROR) { | |
3772 | return NULL; | |
3773 | } | |
b0f7b9b5 | 3774 | |
6a542738 | 3775 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
3776 | } |
3777 | ||
3778 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
3779 | * its facet within 'ofproto', and frees it. */ | |
3780 | static void | |
15baa734 | 3781 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 3782 | { |
15baa734 BP |
3783 | struct facet *facet = subfacet->facet; |
3784 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3785 | ||
3786 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
3787 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
3788 | list_remove(&subfacet->list_node); | |
3789 | free(subfacet->key); | |
b95fc6ba | 3790 | free(subfacet->actions); |
b0f7b9b5 BP |
3791 | free(subfacet); |
3792 | } | |
3793 | ||
3794 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
3795 | * last remaining subfacet in its facet destroys the facet too. */ | |
3796 | static void | |
15baa734 | 3797 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
3798 | { |
3799 | struct facet *facet = subfacet->facet; | |
3800 | ||
551a2f6c BP |
3801 | if (list_is_singleton(&facet->subfacets)) { |
3802 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 3803 | facet_remove(facet); |
551a2f6c | 3804 | } else { |
15baa734 | 3805 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
3806 | } |
3807 | } | |
3808 | ||
3809 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
3810 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
3811 | * for use as temporary storage. */ | |
3812 | static void | |
3813 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
3814 | struct ofpbuf *key) | |
3815 | { | |
3816 | if (!subfacet->key) { | |
3817 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
3818 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
3819 | } else { | |
3820 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
3821 | } | |
3822 | } | |
3823 | ||
b95fc6ba BP |
3824 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. */ |
3825 | static void | |
15baa734 | 3826 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet) |
b95fc6ba BP |
3827 | { |
3828 | struct facet *facet = subfacet->facet; | |
3829 | const struct rule_dpif *rule = facet->rule; | |
15baa734 | 3830 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
b95fc6ba BP |
3831 | struct ofpbuf *odp_actions; |
3832 | struct action_xlate_ctx ctx; | |
3833 | ||
15baa734 | 3834 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
54834960 | 3835 | rule->up.flow_cookie, packet); |
b95fc6ba BP |
3836 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); |
3837 | facet->tags = ctx.tags; | |
3838 | facet->may_install = ctx.may_set_up_flow; | |
3839 | facet->has_learn = ctx.has_learn; | |
3840 | facet->has_normal = ctx.has_normal; | |
3841 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
9d24de3b | 3842 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3843 | |
3844 | if (subfacet->actions_len != odp_actions->size | |
3845 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
3846 | free(subfacet->actions); | |
3847 | subfacet->actions_len = odp_actions->size; | |
3848 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
3849 | } | |
3850 | ||
3851 | ofpbuf_delete(odp_actions); | |
3852 | } | |
3853 | ||
b0f7b9b5 BP |
3854 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
3855 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
3856 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
3857 | * since 'subfacet' was last updated. | |
3858 | * | |
3859 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3860 | static int | |
15baa734 | 3861 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 BP |
3862 | const struct nlattr *actions, size_t actions_len, |
3863 | struct dpif_flow_stats *stats) | |
3864 | { | |
15baa734 BP |
3865 | struct facet *facet = subfacet->facet; |
3866 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
b0f7b9b5 BP |
3867 | struct odputil_keybuf keybuf; |
3868 | enum dpif_flow_put_flags flags; | |
3869 | struct ofpbuf key; | |
3870 | int ret; | |
3871 | ||
3872 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
3873 | if (stats) { | |
3874 | flags |= DPIF_FP_ZERO_STATS; | |
3875 | } | |
3876 | ||
3877 | subfacet_get_key(subfacet, &keybuf, &key); | |
3878 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
3879 | actions, actions_len, stats); | |
3880 | ||
3881 | if (stats) { | |
3882 | subfacet_reset_dp_stats(subfacet, stats); | |
3883 | } | |
3884 | ||
3885 | return ret; | |
3886 | } | |
3887 | ||
3888 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
3889 | static void | |
15baa734 | 3890 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 BP |
3891 | { |
3892 | if (subfacet->installed) { | |
15baa734 BP |
3893 | struct rule_dpif *rule = subfacet->facet->rule; |
3894 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
3895 | struct odputil_keybuf keybuf; |
3896 | struct dpif_flow_stats stats; | |
3897 | struct ofpbuf key; | |
3898 | int error; | |
3899 | ||
3900 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 3901 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
3902 | subfacet_reset_dp_stats(subfacet, &stats); |
3903 | if (!error) { | |
15baa734 | 3904 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 BP |
3905 | } |
3906 | subfacet->installed = false; | |
3907 | } else { | |
3908 | assert(subfacet->dp_packet_count == 0); | |
3909 | assert(subfacet->dp_byte_count == 0); | |
3910 | } | |
3911 | } | |
3912 | ||
3913 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
3914 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
3915 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
3916 | * was reset in the datapath. 'stats' will be modified to include only | |
3917 | * statistics new since 'subfacet' was last updated. */ | |
3918 | static void | |
3919 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
3920 | struct dpif_flow_stats *stats) | |
3921 | { | |
3922 | if (stats | |
3923 | && subfacet->dp_packet_count <= stats->n_packets | |
3924 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
3925 | stats->n_packets -= subfacet->dp_packet_count; | |
3926 | stats->n_bytes -= subfacet->dp_byte_count; | |
3927 | } | |
3928 | ||
3929 | subfacet->dp_packet_count = 0; | |
3930 | subfacet->dp_byte_count = 0; | |
3931 | } | |
3932 | ||
3933 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
3934 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
3935 | static void | |
15baa734 | 3936 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
3937 | { |
3938 | if (used > subfacet->used) { | |
3939 | subfacet->used = used; | |
15baa734 | 3940 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
3941 | } |
3942 | } | |
3943 | ||
3944 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
3945 | * | |
3946 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
3947 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
3948 | * represents a packet that was sent by hand or if it represents statistics | |
3949 | * that have been cleared out of the datapath. */ | |
3950 | static void | |
15baa734 | 3951 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
3952 | const struct dpif_flow_stats *stats) |
3953 | { | |
3954 | if (stats->n_packets || stats->used > subfacet->used) { | |
3955 | struct facet *facet = subfacet->facet; | |
3956 | ||
15baa734 | 3957 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
3958 | facet->packet_count += stats->n_packets; |
3959 | facet->byte_count += stats->n_bytes; | |
3960 | facet_push_stats(facet); | |
3961 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
3962 | } | |
3963 | } | |
3964 | \f | |
abe529af BP |
3965 | /* Rules. */ |
3966 | ||
3967 | static struct rule_dpif * | |
29901626 BP |
3968 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
3969 | uint8_t table_id) | |
abe529af | 3970 | { |
7257b535 BP |
3971 | struct cls_rule *cls_rule; |
3972 | struct classifier *cls; | |
3973 | ||
9cdaaebe BP |
3974 | if (table_id >= N_TABLES) { |
3975 | return NULL; | |
3976 | } | |
3977 | ||
d0918789 | 3978 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 3979 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
3980 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
3981 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
3982 | * are unavailable. */ | |
3983 | struct flow ofpc_normal_flow = *flow; | |
3984 | ofpc_normal_flow.tp_src = htons(0); | |
3985 | ofpc_normal_flow.tp_dst = htons(0); | |
3986 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
3987 | } else { | |
3988 | cls_rule = classifier_lookup(cls, flow); | |
3989 | } | |
3990 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
3991 | } |
3992 | ||
7ee20df1 BP |
3993 | static void |
3994 | complete_operation(struct rule_dpif *rule) | |
3995 | { | |
3996 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3997 | ||
54a9cbc9 | 3998 | rule_invalidate(rule); |
7ee20df1 BP |
3999 | if (clogged) { |
4000 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4001 | c->op = rule->up.pending; | |
4002 | list_push_back(&ofproto->completions, &c->list_node); | |
4003 | } else { | |
4004 | ofoperation_complete(rule->up.pending, 0); | |
4005 | } | |
4006 | } | |
4007 | ||
abe529af BP |
4008 | static struct rule * |
4009 | rule_alloc(void) | |
4010 | { | |
4011 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4012 | return &rule->up; | |
4013 | } | |
4014 | ||
4015 | static void | |
4016 | rule_dealloc(struct rule *rule_) | |
4017 | { | |
4018 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4019 | free(rule); | |
4020 | } | |
4021 | ||
90bf1e07 | 4022 | static enum ofperr |
abe529af BP |
4023 | rule_construct(struct rule *rule_) |
4024 | { | |
4025 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4026 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4027 | struct rule_dpif *victim; |
54a9cbc9 | 4028 | uint8_t table_id; |
90bf1e07 | 4029 | enum ofperr error; |
5bf0e941 BP |
4030 | |
4031 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
4032 | &rule->up.cr.flow, ofproto->max_ports); | |
4033 | if (error) { | |
4034 | return error; | |
4035 | } | |
abe529af BP |
4036 | |
4037 | rule->used = rule->up.created; | |
4038 | rule->packet_count = 0; | |
4039 | rule->byte_count = 0; | |
abe529af | 4040 | |
7ee20df1 BP |
4041 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4042 | if (victim && !list_is_empty(&victim->facets)) { | |
4043 | struct facet *facet; | |
4044 | ||
4045 | rule->facets = victim->facets; | |
4046 | list_moved(&rule->facets); | |
4047 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4048 | /* XXX: We're only clearing our local counters here. It's possible |
4049 | * that quite a few packets are unaccounted for in the datapath | |
4050 | * statistics. These will be accounted to the new rule instead of | |
4051 | * cleared as required. This could be fixed by clearing out the | |
4052 | * datapath statistics for this facet, but currently it doesn't | |
4053 | * seem worth it. */ | |
4054 | facet_reset_counters(facet); | |
7ee20df1 BP |
4055 | facet->rule = rule; |
4056 | } | |
4057 | } else { | |
4058 | /* Must avoid list_moved() in this case. */ | |
4059 | list_init(&rule->facets); | |
4060 | } | |
abe529af | 4061 | |
54a9cbc9 BP |
4062 | table_id = rule->up.table_id; |
4063 | rule->tag = (victim ? victim->tag | |
4064 | : table_id == 0 ? 0 | |
4065 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4066 | ofproto->tables[table_id].basis)); | |
4067 | ||
7ee20df1 | 4068 | complete_operation(rule); |
abe529af BP |
4069 | return 0; |
4070 | } | |
4071 | ||
4072 | static void | |
4073 | rule_destruct(struct rule *rule_) | |
4074 | { | |
4075 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4076 | struct facet *facet, *next_facet; |
4077 | ||
abe529af | 4078 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4079 | facet_revalidate(facet); |
abe529af | 4080 | } |
7ee20df1 BP |
4081 | |
4082 | complete_operation(rule); | |
abe529af BP |
4083 | } |
4084 | ||
4085 | static void | |
4086 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4087 | { | |
4088 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4089 | struct facet *facet; | |
4090 | ||
4091 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4092 | * in facets. This counts, for example, facets that have expired. */ | |
4093 | *packets = rule->packet_count; | |
4094 | *bytes = rule->byte_count; | |
4095 | ||
4096 | /* Add any statistics that are tracked by facets. This includes | |
4097 | * statistical data recently updated by ofproto_update_stats() as well as | |
4098 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4099 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4100 | *packets += facet->packet_count; | |
4101 | *bytes += facet->byte_count; | |
4102 | } | |
4103 | } | |
4104 | ||
90bf1e07 | 4105 | static enum ofperr |
59d0f2c8 BP |
4106 | rule_execute(struct rule *rule_, const struct flow *flow, |
4107 | struct ofpbuf *packet) | |
abe529af BP |
4108 | { |
4109 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4110 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4111 | struct action_xlate_ctx ctx; | |
4112 | struct ofpbuf *odp_actions; | |
abe529af BP |
4113 | size_t size; |
4114 | ||
54834960 EJ |
4115 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
4116 | rule->up.flow_cookie, packet); | |
abe529af BP |
4117 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); |
4118 | size = packet->size; | |
4119 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
4120 | odp_actions->size, packet)) { | |
4121 | rule->used = time_msec(); | |
4122 | rule->packet_count++; | |
4123 | rule->byte_count += size; | |
4124 | flow_push_stats(rule, flow, 1, size, rule->used); | |
4125 | } | |
4126 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
4127 | |
4128 | return 0; | |
abe529af BP |
4129 | } |
4130 | ||
7ee20df1 BP |
4131 | static void |
4132 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4133 | { |
4134 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4135 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
90bf1e07 | 4136 | enum ofperr error; |
abe529af | 4137 | |
7ee20df1 BP |
4138 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
4139 | &rule->up.cr.flow, ofproto->max_ports); | |
4140 | if (error) { | |
4141 | ofoperation_complete(rule->up.pending, error); | |
4142 | return; | |
abe529af | 4143 | } |
7ee20df1 BP |
4144 | |
4145 | complete_operation(rule); | |
abe529af BP |
4146 | } |
4147 | \f | |
97d6520b | 4148 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4149 | * May modify 'packet'. |
abe529af BP |
4150 | * Returns 0 if successful, otherwise a positive errno value. */ |
4151 | static int | |
52a90c29 | 4152 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4153 | { |
97d6520b | 4154 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4155 | struct ofpbuf key, odp_actions; |
4156 | struct odputil_keybuf keybuf; | |
52a90c29 | 4157 | uint16_t odp_port; |
80e5eed9 | 4158 | struct flow flow; |
abe529af BP |
4159 | int error; |
4160 | ||
abff858b | 4161 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
52a90c29 BP |
4162 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4163 | flow.vlan_tci); | |
4164 | if (odp_port != ofport->odp_port) { | |
4165 | eth_pop_vlan(packet); | |
4166 | flow.vlan_tci = htons(0); | |
4167 | } | |
4168 | ||
80e5eed9 BP |
4169 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4170 | odp_flow_key_from_flow(&key, &flow); | |
4171 | ||
abe529af | 4172 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4173 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4174 | ||
df2c07f4 | 4175 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4176 | error = dpif_execute(ofproto->dpif, |
4177 | key.data, key.size, | |
4178 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4179 | packet); |
4180 | ofpbuf_uninit(&odp_actions); | |
4181 | ||
4182 | if (error) { | |
4183 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4184 | ofproto->up.name, odp_port, strerror(error)); | |
4185 | } | |
6527c598 | 4186 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4187 | return error; |
4188 | } | |
4189 | \f | |
df2c07f4 | 4190 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
4191 | |
4192 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4193 | struct action_xlate_ctx *ctx); | |
4cd78906 | 4194 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4195 | |
98403001 BP |
4196 | static size_t |
4197 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4198 | struct ofpbuf *odp_actions, | |
4199 | const struct flow *flow, | |
4200 | const struct user_action_cookie *cookie) | |
4201 | { | |
98403001 BP |
4202 | uint32_t pid; |
4203 | ||
4204 | pid = dpif_port_get_pid(ofproto->dpif, | |
4205 | ofp_port_to_odp_port(flow->in_port)); | |
4206 | ||
39db78a0 | 4207 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4208 | } |
4209 | ||
6ff686f2 PS |
4210 | /* Compose SAMPLE action for sFlow. */ |
4211 | static size_t | |
4212 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4213 | struct ofpbuf *odp_actions, | |
4214 | const struct flow *flow, | |
4215 | uint32_t odp_port) | |
4216 | { | |
4217 | uint32_t port_ifindex; | |
4218 | uint32_t probability; | |
98403001 | 4219 | struct user_action_cookie cookie; |
6ff686f2 | 4220 | size_t sample_offset, actions_offset; |
98403001 | 4221 | int cookie_offset, n_output; |
6ff686f2 PS |
4222 | |
4223 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4224 | return 0; | |
4225 | } | |
4226 | ||
4227 | if (odp_port == OVSP_NONE) { | |
4228 | port_ifindex = 0; | |
4229 | n_output = 0; | |
4230 | } else { | |
4231 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4232 | n_output = 1; | |
4233 | } | |
4234 | ||
4235 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
4236 | ||
4237 | /* Number of packets out of UINT_MAX to sample. */ | |
4238 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4239 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4240 | ||
4241 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
4242 | ||
98403001 BP |
4243 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
4244 | cookie.data = port_ifindex; | |
4245 | cookie.n_output = n_output; | |
4246 | cookie.vlan_tci = 0; | |
4247 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
4248 | |
4249 | nl_msg_end_nested(odp_actions, actions_offset); | |
4250 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4251 | return cookie_offset; |
6ff686f2 PS |
4252 | } |
4253 | ||
4254 | /* SAMPLE action must be first action in any given list of actions. | |
4255 | * At this point we do not have all information required to build it. So try to | |
4256 | * build sample action as complete as possible. */ | |
4257 | static void | |
4258 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4259 | { | |
4260 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4261 | ctx->odp_actions, | |
4262 | &ctx->flow, OVSP_NONE); | |
4263 | ctx->sflow_odp_port = 0; | |
4264 | ctx->sflow_n_outputs = 0; | |
4265 | } | |
4266 | ||
4267 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4268 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4269 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4270 | static void | |
4271 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4272 | { | |
4273 | const struct flow *base = &ctx->base_flow; | |
4274 | struct user_action_cookie *cookie; | |
4275 | ||
4276 | if (!ctx->user_cookie_offset) { | |
4277 | return; | |
4278 | } | |
4279 | ||
4280 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
4281 | sizeof(*cookie)); | |
4282 | assert(cookie != NULL); | |
4283 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
4284 | ||
4285 | if (ctx->sflow_n_outputs) { | |
4286 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
4287 | ctx->sflow_odp_port); | |
4288 | } | |
4289 | if (ctx->sflow_n_outputs >= 255) { | |
4290 | cookie->n_output = 255; | |
4291 | } else { | |
4292 | cookie->n_output = ctx->sflow_n_outputs; | |
4293 | } | |
4294 | cookie->vlan_tci = base->vlan_tci; | |
4295 | } | |
4296 | ||
6ff686f2 | 4297 | static void |
81b1afb1 EJ |
4298 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4299 | bool check_stp) | |
6ff686f2 | 4300 | { |
d59906fb | 4301 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4302 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4303 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4304 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4305 | uint16_t out_port; |
d59906fb | 4306 | |
81b1afb1 | 4307 | if (ofport) { |
8b36f51e EJ |
4308 | struct priority_to_dscp *pdscp; |
4309 | ||
81b1afb1 EJ |
4310 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD) |
4311 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { | |
4312 | return; | |
4313 | } | |
8b36f51e | 4314 | |
deedf7e7 | 4315 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4316 | if (pdscp) { |
4317 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4318 | ctx->flow.nw_tos |= pdscp->dscp; | |
4319 | } | |
81b1afb1 EJ |
4320 | } else { |
4321 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4322 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4323 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4324 | } |
4325 | ||
52a90c29 BP |
4326 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4327 | ctx->flow.vlan_tci); | |
4328 | if (out_port != odp_port) { | |
4329 | ctx->flow.vlan_tci = htons(0); | |
4330 | } | |
5bbda0aa | 4331 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4332 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4333 | ||
6ff686f2 PS |
4334 | ctx->sflow_odp_port = odp_port; |
4335 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4336 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4337 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4338 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4339 | } |
4340 | ||
abe529af | 4341 | static void |
5e48dc2b | 4342 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4343 | { |
81b1afb1 | 4344 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4345 | } |
4346 | ||
4347 | static void | |
29901626 BP |
4348 | xlate_table_action(struct action_xlate_ctx *ctx, |
4349 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
4350 | { |
4351 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4352 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4353 | struct rule_dpif *rule; |
4354 | uint16_t old_in_port; | |
29901626 BP |
4355 | uint8_t old_table_id; |
4356 | ||
4357 | old_table_id = ctx->table_id; | |
4358 | ctx->table_id = table_id; | |
abe529af | 4359 | |
54a9cbc9 | 4360 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
4361 | old_in_port = ctx->flow.in_port; |
4362 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
4363 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
4364 | ||
4365 | /* Tag the flow. */ | |
4366 | if (table_id > 0 && table_id < N_TABLES) { | |
4367 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4368 | if (table->other_table) { | |
4369 | ctx->tags |= (rule | |
4370 | ? rule->tag | |
4371 | : rule_calculate_tag(&ctx->flow, | |
4372 | &table->other_table->wc, | |
4373 | table->basis)); | |
4374 | } | |
4375 | } | |
4376 | ||
4377 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
4378 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
4379 | ctx->flow.in_port = old_in_port; |
4380 | ||
4381 | if (ctx->resubmit_hook) { | |
4382 | ctx->resubmit_hook(ctx, rule); | |
4383 | } | |
4384 | ||
4385 | if (rule) { | |
54834960 EJ |
4386 | ovs_be64 old_cookie = ctx->cookie; |
4387 | ||
abe529af | 4388 | ctx->recurse++; |
54834960 | 4389 | ctx->cookie = rule->up.flow_cookie; |
abe529af | 4390 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); |
54834960 | 4391 | ctx->cookie = old_cookie; |
abe529af BP |
4392 | ctx->recurse--; |
4393 | } | |
29901626 BP |
4394 | |
4395 | ctx->table_id = old_table_id; | |
abe529af BP |
4396 | } else { |
4397 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4398 | ||
29901626 | 4399 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af BP |
4400 | MAX_RESUBMIT_RECURSION); |
4401 | } | |
4402 | } | |
4403 | ||
29901626 BP |
4404 | static void |
4405 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
4406 | const struct nx_action_resubmit *nar) | |
4407 | { | |
4408 | uint16_t in_port; | |
4409 | uint8_t table_id; | |
4410 | ||
4411 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
4412 | ? ctx->flow.in_port | |
4413 | : ntohs(nar->in_port)); | |
4414 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
4415 | ||
4416 | xlate_table_action(ctx, in_port, table_id); | |
4417 | } | |
4418 | ||
abe529af | 4419 | static void |
d59906fb | 4420 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
4421 | { |
4422 | struct ofport_dpif *ofport; | |
4423 | ||
b3e9b2ed | 4424 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 4425 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
4426 | |
4427 | if (ofp_port == ctx->flow.in_port) { | |
4428 | continue; | |
4429 | } | |
4430 | ||
5e48dc2b | 4431 | if (all) { |
81b1afb1 | 4432 | compose_output_action__(ctx, ofp_port, false); |
5e48dc2b EJ |
4433 | } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) { |
4434 | compose_output_action(ctx, ofp_port); | |
abe529af BP |
4435 | } |
4436 | } | |
b3e9b2ed EJ |
4437 | |
4438 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
4439 | } |
4440 | ||
6ff686f2 | 4441 | static void |
f0fd1a17 PS |
4442 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
4443 | enum ofp_packet_in_reason reason) | |
6ff686f2 | 4444 | { |
999fba59 EJ |
4445 | struct ofputil_packet_in pin; |
4446 | struct ofpbuf *packet; | |
6ff686f2 | 4447 | |
999fba59 EJ |
4448 | ctx->may_set_up_flow = false; |
4449 | if (!ctx->packet) { | |
4450 | return; | |
4451 | } | |
4452 | ||
4453 | packet = ofpbuf_clone(ctx->packet); | |
4454 | ||
4455 | if (packet->l2 && packet->l3) { | |
4456 | struct eth_header *eh; | |
4457 | ||
4458 | eth_pop_vlan(packet); | |
4459 | eh = packet->l2; | |
4460 | assert(eh->eth_type == ctx->flow.dl_type); | |
4461 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); | |
4462 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
4463 | ||
4464 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
4465 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
4466 | } | |
4467 | ||
4468 | if (packet->l4) { | |
4469 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4470 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
4471 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
4472 | } | |
4473 | ||
4474 | if (packet->l7) { | |
4475 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
4476 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
4477 | ctx->flow.tp_dst); | |
4478 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
4479 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
4480 | ctx->flow.tp_dst); | |
4481 | } | |
4482 | } | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | pin.packet = packet->data; | |
4487 | pin.packet_len = packet->size; | |
f0fd1a17 | 4488 | pin.reason = reason; |
54834960 EJ |
4489 | pin.table_id = ctx->table_id; |
4490 | pin.cookie = ctx->cookie; | |
4491 | ||
999fba59 EJ |
4492 | pin.buffer_id = 0; |
4493 | pin.send_len = len; | |
4494 | pin.total_len = packet->size; | |
4495 | flow_get_metadata(&ctx->flow, &pin.fmd); | |
4496 | ||
4497 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow); | |
4498 | ofpbuf_delete(packet); | |
6ff686f2 PS |
4499 | } |
4500 | ||
f0fd1a17 PS |
4501 | static bool |
4502 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
4503 | { | |
4504 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
4505 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
4506 | return false; | |
4507 | } | |
4508 | ||
4509 | if (ctx->flow.nw_ttl > 1) { | |
4510 | ctx->flow.nw_ttl--; | |
4511 | return false; | |
4512 | } else { | |
4513 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL); | |
4514 | ||
4515 | /* Stop processing for current table. */ | |
4516 | return true; | |
4517 | } | |
4518 | } | |
4519 | ||
abe529af BP |
4520 | static void |
4521 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
4522 | uint16_t port, uint16_t max_len) | |
4523 | { | |
4524 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
4525 | ||
4526 | ctx->nf_output_iface = NF_OUT_DROP; | |
4527 | ||
4528 | switch (port) { | |
4529 | case OFPP_IN_PORT: | |
81b1afb1 | 4530 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
4531 | break; |
4532 | case OFPP_TABLE: | |
29901626 | 4533 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
4534 | break; |
4535 | case OFPP_NORMAL: | |
4536 | xlate_normal(ctx); | |
4537 | break; | |
4538 | case OFPP_FLOOD: | |
d59906fb | 4539 | flood_packets(ctx, false); |
abe529af BP |
4540 | break; |
4541 | case OFPP_ALL: | |
d59906fb | 4542 | flood_packets(ctx, true); |
abe529af BP |
4543 | break; |
4544 | case OFPP_CONTROLLER: | |
f0fd1a17 | 4545 | execute_controller_action(ctx, max_len, OFPR_ACTION); |
abe529af BP |
4546 | break; |
4547 | case OFPP_LOCAL: | |
81b1afb1 | 4548 | compose_output_action(ctx, OFPP_LOCAL); |
abe529af | 4549 | break; |
e81d2933 EJ |
4550 | case OFPP_NONE: |
4551 | break; | |
abe529af BP |
4552 | default: |
4553 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 4554 | compose_output_action(ctx, port); |
abe529af BP |
4555 | } |
4556 | break; | |
4557 | } | |
4558 | ||
4559 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
4560 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4561 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4562 | ctx->nf_output_iface = prev_nf_output_iface; | |
4563 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
4564 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
4565 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4566 | } | |
4567 | } | |
4568 | ||
f694937d EJ |
4569 | static void |
4570 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
4571 | const struct nx_action_output_reg *naor) | |
4572 | { | |
816fd533 | 4573 | struct mf_subfield src; |
f694937d EJ |
4574 | uint64_t ofp_port; |
4575 | ||
816fd533 BP |
4576 | nxm_decode(&src, naor->src, naor->ofs_nbits); |
4577 | ofp_port = mf_get_subfield(&src, &ctx->flow); | |
f694937d EJ |
4578 | |
4579 | if (ofp_port <= UINT16_MAX) { | |
4580 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
4581 | } | |
4582 | } | |
4583 | ||
abe529af BP |
4584 | static void |
4585 | xlate_output_action(struct action_xlate_ctx *ctx, | |
4586 | const struct ofp_action_output *oao) | |
4587 | { | |
4588 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
4589 | } | |
4590 | ||
abe529af BP |
4591 | static void |
4592 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
4593 | const struct ofp_action_enqueue *oae) | |
4594 | { | |
e479e41e | 4595 | uint16_t ofp_port; |
abff858b | 4596 | uint32_t flow_priority, priority; |
abe529af BP |
4597 | int error; |
4598 | ||
4599 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
4600 | &priority); | |
4601 | if (error) { | |
4602 | /* Fall back to ordinary output action. */ | |
4603 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
4604 | return; | |
4605 | } | |
4606 | ||
df2c07f4 | 4607 | /* Figure out datapath output port. */ |
abe529af BP |
4608 | ofp_port = ntohs(oae->port); |
4609 | if (ofp_port == OFPP_IN_PORT) { | |
4610 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
4611 | } else if (ofp_port == ctx->flow.in_port) { |
4612 | return; | |
abe529af | 4613 | } |
abe529af | 4614 | |
df2c07f4 | 4615 | /* Add datapath actions. */ |
deedf7e7 BP |
4616 | flow_priority = ctx->flow.skb_priority; |
4617 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 4618 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 4619 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
4620 | |
4621 | /* Update NetFlow output port. */ | |
4622 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 4623 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
4624 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
4625 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4626 | } | |
4627 | } | |
4628 | ||
4629 | static void | |
4630 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
4631 | const struct nx_action_set_queue *nasq) | |
4632 | { | |
4633 | uint32_t priority; | |
4634 | int error; | |
4635 | ||
4636 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4637 | &priority); | |
4638 | if (error) { | |
4639 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4640 | * has already been logged. */ | |
4641 | return; | |
4642 | } | |
4643 | ||
deedf7e7 | 4644 | ctx->flow.skb_priority = priority; |
abe529af BP |
4645 | } |
4646 | ||
4647 | struct xlate_reg_state { | |
4648 | ovs_be16 vlan_tci; | |
4649 | ovs_be64 tun_id; | |
4650 | }; | |
4651 | ||
abe529af BP |
4652 | static void |
4653 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4654 | const struct nx_action_autopath *naa) | |
4655 | { | |
4656 | uint16_t ofp_port = ntohl(naa->id); | |
4657 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4658 | ||
4659 | if (!port || !port->bundle) { | |
4660 | ofp_port = OFPP_NONE; | |
4661 | } else if (port->bundle->bond) { | |
4662 | /* Autopath does not support VLAN hashing. */ | |
4663 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4664 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4665 | if (slave) { |
4666 | ofp_port = slave->up.ofp_port; | |
4667 | } | |
4668 | } | |
4669 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4670 | } | |
4671 | ||
daff3353 EJ |
4672 | static bool |
4673 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4674 | { | |
4675 | struct ofproto_dpif *ofproto = ofproto_; | |
4676 | struct ofport_dpif *port; | |
4677 | ||
4678 | switch (ofp_port) { | |
4679 | case OFPP_IN_PORT: | |
4680 | case OFPP_TABLE: | |
4681 | case OFPP_NORMAL: | |
4682 | case OFPP_FLOOD: | |
4683 | case OFPP_ALL: | |
439e4d8c | 4684 | case OFPP_NONE: |
daff3353 EJ |
4685 | return true; |
4686 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4687 | return false; | |
4688 | default: | |
4689 | port = get_ofp_port(ofproto, ofp_port); | |
4690 | return port ? port->may_enable : false; | |
4691 | } | |
4692 | } | |
4693 | ||
75a75043 BP |
4694 | static void |
4695 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
4696 | const struct nx_action_learn *learn) | |
4697 | { | |
4698 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
4699 | struct ofputil_flow_mod fm; | |
4700 | int error; | |
4701 | ||
4702 | learn_execute(learn, &ctx->flow, &fm); | |
4703 | ||
4704 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
4705 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
4706 | VLOG_WARN("learning action failed to modify flow table (%s)", |
4707 | ofperr_get_name(error)); | |
75a75043 BP |
4708 | } |
4709 | ||
4710 | free(fm.actions); | |
4711 | } | |
4712 | ||
21f7563c JP |
4713 | static bool |
4714 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
4715 | { | |
4716 | if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
4717 | ? htonl(OFPPC_NO_RECV_STP) | |
4718 | : htonl(OFPPC_NO_RECV))) { | |
4719 | return false; | |
4720 | } | |
4721 | ||
4722 | /* Only drop packets here if both forwarding and learning are | |
4723 | * disabled. If just learning is enabled, we need to have | |
4724 | * OFPP_NORMAL and the learning action have a look at the packet | |
4725 | * before we can drop it. */ | |
4726 | if (!stp_forward_in_state(port->stp_state) | |
4727 | && !stp_learn_in_state(port->stp_state)) { | |
4728 | return false; | |
4729 | } | |
4730 | ||
4731 | return true; | |
4732 | } | |
4733 | ||
abe529af BP |
4734 | static void |
4735 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4736 | struct action_xlate_ctx *ctx) | |
4737 | { | |
4738 | const struct ofport_dpif *port; | |
abe529af | 4739 | const union ofp_action *ia; |
b4b8c781 | 4740 | size_t left; |
abe529af BP |
4741 | |
4742 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 4743 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
4744 | /* Drop this flow. */ |
4745 | return; | |
4746 | } | |
4747 | ||
b4b8c781 | 4748 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 4749 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
4750 | const struct nx_action_resubmit *nar; |
4751 | const struct nx_action_set_tunnel *nast; | |
4752 | const struct nx_action_set_queue *nasq; | |
4753 | const struct nx_action_multipath *nam; | |
4754 | const struct nx_action_autopath *naa; | |
daff3353 | 4755 | const struct nx_action_bundle *nab; |
f694937d | 4756 | const struct nx_action_output_reg *naor; |
38f2e360 BP |
4757 | enum ofputil_action_code code; |
4758 | ovs_be64 tun_id; | |
4759 | ||
848e8809 EJ |
4760 | if (ctx->exit) { |
4761 | break; | |
4762 | } | |
4763 | ||
38f2e360 BP |
4764 | code = ofputil_decode_action_unsafe(ia); |
4765 | switch (code) { | |
4766 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
4767 | xlate_output_action(ctx, &ia->output); |
4768 | break; | |
4769 | ||
38f2e360 | 4770 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
4771 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
4772 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
4773 | break; |
4774 | ||
38f2e360 | 4775 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
4776 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
4777 | ctx->flow.vlan_tci |= htons( | |
4778 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
4779 | break; |
4780 | ||
38f2e360 | 4781 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 4782 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
4783 | break; |
4784 | ||
38f2e360 | 4785 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 4786 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4787 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
4788 | break; | |
4789 | ||
38f2e360 | 4790 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 4791 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4792 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
4793 | break; | |
4794 | ||
38f2e360 | 4795 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
4796 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
4797 | break; | |
4798 | ||
38f2e360 | 4799 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
4800 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
4801 | break; | |
4802 | ||
38f2e360 | 4803 | case OFPUTIL_OFPAT_SET_NW_TOS: |
c4f2731d PS |
4804 | /* OpenFlow 1.0 only supports IPv4. */ |
4805 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4806 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4807 | ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
4808 | } | |
abe529af BP |
4809 | break; |
4810 | ||
38f2e360 | 4811 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
4812 | ctx->flow.tp_src = ia->tp_port.tp_port; |
4813 | break; | |
4814 | ||
38f2e360 | 4815 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
4816 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
4817 | break; | |
4818 | ||
38f2e360 BP |
4819 | case OFPUTIL_OFPAT_ENQUEUE: |
4820 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
4821 | break; | |
4822 | ||
4823 | case OFPUTIL_NXAST_RESUBMIT: | |
4824 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
4825 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
4826 | break; | |
4827 | ||
4828 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
4829 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
4830 | break; |
4831 | ||
38f2e360 BP |
4832 | case OFPUTIL_NXAST_SET_TUNNEL: |
4833 | nast = (const struct nx_action_set_tunnel *) ia; | |
4834 | tun_id = htonll(ntohl(nast->tun_id)); | |
4835 | ctx->flow.tun_id = tun_id; | |
4836 | break; | |
4837 | ||
4838 | case OFPUTIL_NXAST_SET_QUEUE: | |
4839 | nasq = (const struct nx_action_set_queue *) ia; | |
4840 | xlate_set_queue_action(ctx, nasq); | |
4841 | break; | |
4842 | ||
4843 | case OFPUTIL_NXAST_POP_QUEUE: | |
deedf7e7 | 4844 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
4845 | break; |
4846 | ||
4847 | case OFPUTIL_NXAST_REG_MOVE: | |
4848 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
4849 | &ctx->flow); | |
4850 | break; | |
4851 | ||
4852 | case OFPUTIL_NXAST_REG_LOAD: | |
4853 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
4854 | &ctx->flow); | |
4855 | break; | |
4856 | ||
4857 | case OFPUTIL_NXAST_NOTE: | |
4858 | /* Nothing to do. */ | |
4859 | break; | |
4860 | ||
4861 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
4862 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
4863 | ctx->flow.tun_id = tun_id; | |
4864 | break; | |
4865 | ||
4866 | case OFPUTIL_NXAST_MULTIPATH: | |
4867 | nam = (const struct nx_action_multipath *) ia; | |
4868 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
4869 | break; |
4870 | ||
38f2e360 BP |
4871 | case OFPUTIL_NXAST_AUTOPATH: |
4872 | naa = (const struct nx_action_autopath *) ia; | |
4873 | xlate_autopath(ctx, naa); | |
abe529af | 4874 | break; |
daff3353 EJ |
4875 | |
4876 | case OFPUTIL_NXAST_BUNDLE: | |
4877 | ctx->ofproto->has_bundle_action = true; | |
4878 | nab = (const struct nx_action_bundle *) ia; | |
4879 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
4880 | slave_enabled_cb, | |
4881 | ctx->ofproto), 0); | |
4882 | break; | |
a368bb53 EJ |
4883 | |
4884 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
4885 | ctx->ofproto->has_bundle_action = true; | |
4886 | nab = (const struct nx_action_bundle *) ia; | |
4887 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
4888 | ctx->ofproto); | |
4889 | break; | |
f694937d EJ |
4890 | |
4891 | case OFPUTIL_NXAST_OUTPUT_REG: | |
4892 | naor = (const struct nx_action_output_reg *) ia; | |
4893 | xlate_output_reg_action(ctx, naor); | |
4894 | break; | |
75a75043 BP |
4895 | |
4896 | case OFPUTIL_NXAST_LEARN: | |
4897 | ctx->has_learn = true; | |
4898 | if (ctx->may_learn) { | |
4899 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); | |
4900 | } | |
4901 | break; | |
848e8809 | 4902 | |
f0fd1a17 PS |
4903 | case OFPUTIL_NXAST_DEC_TTL: |
4904 | if (compose_dec_ttl(ctx)) { | |
4905 | goto out; | |
4906 | } | |
4907 | break; | |
4908 | ||
848e8809 EJ |
4909 | case OFPUTIL_NXAST_EXIT: |
4910 | ctx->exit = true; | |
4911 | break; | |
abe529af BP |
4912 | } |
4913 | } | |
21f7563c | 4914 | |
f0fd1a17 | 4915 | out: |
21f7563c JP |
4916 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
4917 | * so drop it now if forwarding is disabled. */ | |
4918 | if (port && !stp_forward_in_state(port->stp_state)) { | |
4919 | ofpbuf_clear(ctx->odp_actions); | |
4920 | add_sflow_action(ctx); | |
4921 | } | |
abe529af BP |
4922 | } |
4923 | ||
4924 | static void | |
4925 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
4926 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
54834960 EJ |
4927 | ovs_be16 initial_tci, ovs_be64 cookie, |
4928 | const struct ofpbuf *packet) | |
abe529af BP |
4929 | { |
4930 | ctx->ofproto = ofproto; | |
4931 | ctx->flow = *flow; | |
e84173dc BP |
4932 | ctx->base_flow = ctx->flow; |
4933 | ctx->base_flow.tun_id = 0; | |
4934 | ctx->base_flow.vlan_tci = initial_tci; | |
54834960 | 4935 | ctx->cookie = cookie; |
abe529af | 4936 | ctx->packet = packet; |
75a75043 | 4937 | ctx->may_learn = packet != NULL; |
abe529af | 4938 | ctx->resubmit_hook = NULL; |
abe529af BP |
4939 | } |
4940 | ||
4941 | static struct ofpbuf * | |
4942 | xlate_actions(struct action_xlate_ctx *ctx, | |
4943 | const union ofp_action *in, size_t n_in) | |
4944 | { | |
c06bba01 JP |
4945 | struct flow orig_flow = ctx->flow; |
4946 | ||
abe529af BP |
4947 | COVERAGE_INC(ofproto_dpif_xlate); |
4948 | ||
4949 | ctx->odp_actions = ofpbuf_new(512); | |
b6848f13 | 4950 | ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); |
97e42c92 BP |
4951 | ctx->tags = 0; |
4952 | ctx->may_set_up_flow = true; | |
4953 | ctx->has_learn = false; | |
4954 | ctx->has_normal = false; | |
4955 | ctx->nf_output_iface = NF_OUT_DROP; | |
9d24de3b | 4956 | ctx->mirrors = 0; |
97e42c92 | 4957 | ctx->recurse = 0; |
deedf7e7 | 4958 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 4959 | ctx->table_id = 0; |
848e8809 | 4960 | ctx->exit = false; |
7257b535 | 4961 | |
eadef313 | 4962 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
4963 | switch (ctx->ofproto->up.frag_handling) { |
4964 | case OFPC_FRAG_NORMAL: | |
4965 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
4966 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
4967 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
4968 | break; |
4969 | ||
4970 | case OFPC_FRAG_DROP: | |
4971 | return ctx->odp_actions; | |
4972 | ||
4973 | case OFPC_FRAG_REASM: | |
4974 | NOT_REACHED(); | |
4975 | ||
4976 | case OFPC_FRAG_NX_MATCH: | |
4977 | /* Nothing to do. */ | |
4978 | break; | |
f0fd1a17 PS |
4979 | |
4980 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
4981 | NOT_REACHED(); | |
7257b535 BP |
4982 | } |
4983 | } | |
4984 | ||
fc08b7a2 | 4985 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af | 4986 | ctx->may_set_up_flow = false; |
b6848f13 | 4987 | return ctx->odp_actions; |
abe529af | 4988 | } else { |
6ff686f2 | 4989 | add_sflow_action(ctx); |
abe529af | 4990 | do_xlate_actions(in, n_in, ctx); |
abe529af | 4991 | |
b6848f13 BP |
4992 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
4993 | ctx->odp_actions->data, | |
4994 | ctx->odp_actions->size)) { | |
4995 | ctx->may_set_up_flow = false; | |
4996 | if (ctx->packet | |
4997 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
4998 | ctx->packet)) { | |
5e48dc2b | 4999 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
5000 | } |
5001 | } | |
c06bba01 | 5002 | add_mirror_actions(ctx, &orig_flow); |
a7c4eaf6 | 5003 | fix_sflow_action(ctx); |
abe529af BP |
5004 | } |
5005 | ||
5006 | return ctx->odp_actions; | |
5007 | } | |
5008 | \f | |
5009 | /* OFPP_NORMAL implementation. */ | |
5010 | ||
abe529af BP |
5011 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
5012 | ||
ecac4ebf BP |
5013 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
5014 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5015 | * the bundle on which the packet was received, returns the VLAN to which the | |
5016 | * packet belongs. | |
5017 | * | |
5018 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5019 | static uint16_t | |
5020 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5021 | { | |
5022 | switch (in_bundle->vlan_mode) { | |
5023 | case PORT_VLAN_ACCESS: | |
5024 | return in_bundle->vlan; | |
5025 | break; | |
5026 | ||
5027 | case PORT_VLAN_TRUNK: | |
5028 | return vid; | |
5029 | ||
5030 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5031 | case PORT_VLAN_NATIVE_TAGGED: | |
5032 | return vid ? vid : in_bundle->vlan; | |
5033 | ||
5034 | default: | |
5035 | NOT_REACHED(); | |
5036 | } | |
5037 | } | |
5038 | ||
5da5ec37 BP |
5039 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5040 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5041 | * a warning. | |
5042 | * | |
5043 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5044 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5045 | * 0...4095. */ | |
5046 | static bool | |
5047 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5048 | { | |
33158a18 JP |
5049 | /* Allow any VID on the OFPP_NONE port. */ |
5050 | if (in_bundle == &ofpp_none_bundle) { | |
5051 | return true; | |
5052 | } | |
5053 | ||
5da5ec37 BP |
5054 | switch (in_bundle->vlan_mode) { |
5055 | case PORT_VLAN_ACCESS: | |
5056 | if (vid) { | |
5057 | if (warn) { | |
5058 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5059 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5060 | "packet received on port %s configured as VLAN " | |
5061 | "%"PRIu16" access port", | |
5062 | in_bundle->ofproto->up.name, vid, | |
5063 | in_bundle->name, in_bundle->vlan); | |
5064 | } | |
5065 | return false; | |
5066 | } | |
5067 | return true; | |
5068 | ||
5069 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5070 | case PORT_VLAN_NATIVE_TAGGED: | |
5071 | if (!vid) { | |
5072 | /* Port must always carry its native VLAN. */ | |
5073 | return true; | |
5074 | } | |
5075 | /* Fall through. */ | |
5076 | case PORT_VLAN_TRUNK: | |
5077 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5078 | if (warn) { | |
5079 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5080 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5081 | "received on port %s not configured for trunking " | |
5082 | "VLAN %"PRIu16, | |
5083 | in_bundle->ofproto->up.name, vid, | |
5084 | in_bundle->name, vid); | |
5085 | } | |
5086 | return false; | |
5087 | } | |
5088 | return true; | |
5089 | ||
5090 | default: | |
5091 | NOT_REACHED(); | |
5092 | } | |
5093 | ||
5094 | } | |
5095 | ||
ecac4ebf BP |
5096 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5097 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5098 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5099 | * then the 802.1Q header should only be included in the packet if there is a | |
5100 | * nonzero PCP.) | |
5101 | * | |
5102 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5103 | static uint16_t | |
5104 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5105 | { | |
5106 | switch (out_bundle->vlan_mode) { | |
5107 | case PORT_VLAN_ACCESS: | |
5108 | return 0; | |
5109 | ||
5110 | case PORT_VLAN_TRUNK: | |
5111 | case PORT_VLAN_NATIVE_TAGGED: | |
5112 | return vlan; | |
5113 | ||
5114 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5115 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5116 | ||
5117 | default: | |
5118 | NOT_REACHED(); | |
5119 | } | |
5120 | } | |
5121 | ||
395e68ce BP |
5122 | static void |
5123 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5124 | uint16_t vlan) | |
abe529af | 5125 | { |
395e68ce BP |
5126 | struct ofport_dpif *port; |
5127 | uint16_t vid; | |
81b1afb1 | 5128 | ovs_be16 tci, old_tci; |
ecac4ebf | 5129 | |
395e68ce BP |
5130 | vid = output_vlan_to_vid(out_bundle, vlan); |
5131 | if (!out_bundle->bond) { | |
5132 | port = ofbundle_get_a_port(out_bundle); | |
5133 | } else { | |
5134 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5135 | vid, &ctx->tags); | |
5136 | if (!port) { | |
5137 | /* No slaves enabled, so drop packet. */ | |
5138 | return; | |
5139 | } | |
5140 | } | |
abe529af | 5141 | |
81b1afb1 | 5142 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5143 | tci = htons(vid); |
5144 | if (tci || out_bundle->use_priority_tags) { | |
5145 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5146 | if (tci) { | |
5147 | tci |= htons(VLAN_CFI); | |
5148 | } | |
395e68ce | 5149 | } |
81b1afb1 | 5150 | ctx->flow.vlan_tci = tci; |
395e68ce | 5151 | |
5e48dc2b | 5152 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5153 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5154 | } |
5155 | ||
5156 | static int | |
5157 | mirror_mask_ffs(mirror_mask_t mask) | |
5158 | { | |
5159 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5160 | return ffs(mask); | |
5161 | } | |
5162 | ||
abe529af BP |
5163 | static bool |
5164 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5165 | { | |
ecac4ebf | 5166 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5167 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5168 | } |
5169 | ||
5170 | static bool | |
5171 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5172 | { | |
5173 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5174 | } | |
5175 | ||
5176 | /* Returns an arbitrary interface within 'bundle'. */ | |
5177 | static struct ofport_dpif * | |
5178 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5179 | { | |
5180 | return CONTAINER_OF(list_front(&bundle->ports), | |
5181 | struct ofport_dpif, bundle_node); | |
5182 | } | |
5183 | ||
abe529af BP |
5184 | static bool |
5185 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5186 | { | |
fc3d7408 | 5187 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5188 | } |
5189 | ||
07817dfe BP |
5190 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
5191 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
5192 | * protocols that may confuse switches. */ | |
5193 | static bool | |
5194 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
5195 | { | |
5196 | /* If you change this function's behavior, please update corresponding | |
5197 | * documentation in vswitch.xml at the same time. */ | |
5198 | if (dst[0] != 0x01) { | |
5199 | /* All the currently banned MACs happen to start with 01 currently, so | |
5200 | * this is a quick way to eliminate most of the good ones. */ | |
5201 | } else { | |
5202 | if (eth_addr_is_reserved(dst)) { | |
5203 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
5204 | * (01-80-c2-00-00-0x). */ | |
5205 | return false; | |
5206 | } | |
5207 | ||
5208 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
5209 | /* Cisco OUI. */ | |
5210 | if ((dst[3] & 0xfe) == 0xcc && | |
5211 | (dst[4] & 0xfe) == 0xcc && | |
5212 | (dst[5] & 0xfe) == 0xcc) { | |
5213 | /* Drop the following protocols plus others following the same | |
5214 | pattern: | |
5215 | ||
5216 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
5217 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
5218 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
5219 | return false; | |
5220 | } | |
5221 | ||
5222 | if (!(dst[3] | dst[4] | dst[5])) { | |
5223 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
5224 | return false; | |
5225 | } | |
5226 | } | |
5227 | } | |
5228 | return true; | |
5229 | } | |
5230 | ||
abe529af | 5231 | static void |
c06bba01 | 5232 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5233 | { |
5234 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5235 | mirror_mask_t mirrors; | |
c06bba01 JP |
5236 | struct ofbundle *in_bundle; |
5237 | uint16_t vlan; | |
5238 | uint16_t vid; | |
5239 | const struct nlattr *a; | |
5240 | size_t left; | |
5241 | ||
3581c12c JP |
5242 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
5243 | ctx->packet != NULL); | |
5244 | if (!in_bundle) { | |
c06bba01 JP |
5245 | return; |
5246 | } | |
c06bba01 JP |
5247 | mirrors = in_bundle->src_mirrors; |
5248 | ||
5249 | /* Drop frames on bundles reserved for mirroring. */ | |
5250 | if (in_bundle->mirror_out) { | |
5251 | if (ctx->packet != NULL) { | |
5252 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5253 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5254 | "%s, which is reserved exclusively for mirroring", | |
5255 | ctx->ofproto->up.name, in_bundle->name); | |
5256 | } | |
5257 | return; | |
5258 | } | |
5259 | ||
5260 | /* Check VLAN. */ | |
5261 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5262 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5263 | return; | |
5264 | } | |
5265 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5266 | ||
5267 | /* Look at the output ports to check for destination selections. */ | |
5268 | ||
5269 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5270 | ctx->odp_actions->size) { | |
5271 | enum ovs_action_attr type = nl_attr_type(a); | |
5272 | struct ofport_dpif *ofport; | |
5273 | ||
5274 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5275 | continue; | |
5276 | } | |
5277 | ||
5278 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5279 | if (ofport && ofport->bundle) { |
5280 | mirrors |= ofport->bundle->dst_mirrors; | |
5281 | } | |
c06bba01 | 5282 | } |
abe529af BP |
5283 | |
5284 | if (!mirrors) { | |
5285 | return; | |
5286 | } | |
5287 | ||
c06bba01 JP |
5288 | /* Restore the original packet before adding the mirror actions. */ |
5289 | ctx->flow = *orig_flow; | |
5290 | ||
9ba15e2a BP |
5291 | while (mirrors) { |
5292 | struct ofmirror *m; | |
9ba15e2a BP |
5293 | |
5294 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5295 | ||
5296 | if (!vlan_is_mirrored(m, vlan)) { | |
5297 | mirrors &= mirrors - 1; | |
5298 | continue; | |
5299 | } | |
5300 | ||
5301 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 5302 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 5303 | if (m->out) { |
395e68ce | 5304 | output_normal(ctx, m->out, vlan); |
c06bba01 | 5305 | } else if (eth_dst_may_rspan(orig_flow->dl_dst) |
9ba15e2a BP |
5306 | && vlan != m->out_vlan) { |
5307 | struct ofbundle *bundle; | |
5308 | ||
5309 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
5310 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
5311 | && !bundle->mirror_out) { |
5312 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
5313 | } |
5314 | } | |
5315 | } | |
abe529af BP |
5316 | } |
5317 | } | |
5318 | ||
9d24de3b JP |
5319 | static void |
5320 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
5321 | uint64_t packets, uint64_t bytes) | |
5322 | { | |
5323 | if (!mirrors) { | |
5324 | return; | |
5325 | } | |
5326 | ||
5327 | for (; mirrors; mirrors &= mirrors - 1) { | |
5328 | struct ofmirror *m; | |
5329 | ||
5330 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5331 | ||
5332 | if (!m) { | |
5333 | /* In normal circumstances 'm' will not be NULL. However, | |
5334 | * if mirrors are reconfigured, we can temporarily get out | |
5335 | * of sync in facet_revalidate(). We could "correct" the | |
5336 | * mirror list before reaching here, but doing that would | |
5337 | * not properly account the traffic stats we've currently | |
5338 | * accumulated for previous mirror configuration. */ | |
5339 | continue; | |
5340 | } | |
5341 | ||
5342 | m->packet_count += packets; | |
5343 | m->byte_count += bytes; | |
5344 | } | |
5345 | } | |
5346 | ||
abe529af BP |
5347 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
5348 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
5349 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
5350 | static bool | |
5351 | is_gratuitous_arp(const struct flow *flow) | |
5352 | { | |
5353 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
5354 | && eth_addr_is_broadcast(flow->dl_dst) | |
5355 | && (flow->nw_proto == ARP_OP_REPLY | |
5356 | || (flow->nw_proto == ARP_OP_REQUEST | |
5357 | && flow->nw_src == flow->nw_dst))); | |
5358 | } | |
5359 | ||
5360 | static void | |
5361 | update_learning_table(struct ofproto_dpif *ofproto, | |
5362 | const struct flow *flow, int vlan, | |
5363 | struct ofbundle *in_bundle) | |
5364 | { | |
5365 | struct mac_entry *mac; | |
5366 | ||
33158a18 JP |
5367 | /* Don't learn the OFPP_NONE port. */ |
5368 | if (in_bundle == &ofpp_none_bundle) { | |
5369 | return; | |
5370 | } | |
5371 | ||
abe529af BP |
5372 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
5373 | return; | |
5374 | } | |
5375 | ||
5376 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
5377 | if (is_gratuitous_arp(flow)) { | |
5378 | /* We don't want to learn from gratuitous ARP packets that are | |
5379 | * reflected back over bond slaves so we lock the learning table. */ | |
5380 | if (!in_bundle->bond) { | |
5381 | mac_entry_set_grat_arp_lock(mac); | |
5382 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
5383 | return; | |
5384 | } | |
5385 | } | |
5386 | ||
5387 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
5388 | /* The log messages here could actually be useful in debugging, | |
5389 | * so keep the rate limit relatively high. */ | |
5390 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5391 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
5392 | "on port %s in VLAN %d", | |
5393 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
5394 | in_bundle->name, vlan); | |
5395 | ||
5396 | mac->port.p = in_bundle; | |
5397 | tag_set_add(&ofproto->revalidate_set, | |
5398 | mac_learning_changed(ofproto->ml, mac)); | |
5399 | } | |
5400 | } | |
5401 | ||
3581c12c | 5402 | static struct ofbundle * |
395e68ce BP |
5403 | lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) |
5404 | { | |
5405 | struct ofport_dpif *ofport; | |
5406 | ||
33158a18 JP |
5407 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
5408 | * port for traffic that it is sourcing. */ | |
5409 | if (in_port == OFPP_NONE) { | |
5410 | return &ofpp_none_bundle; | |
5411 | } | |
5412 | ||
395e68ce BP |
5413 | /* Find the port and bundle for the received packet. */ |
5414 | ofport = get_ofp_port(ofproto, in_port); | |
5415 | if (ofport && ofport->bundle) { | |
3581c12c | 5416 | return ofport->bundle; |
395e68ce BP |
5417 | } |
5418 | ||
5419 | /* Odd. A few possible reasons here: | |
5420 | * | |
5421 | * - We deleted a port but there are still a few packets queued up | |
5422 | * from it. | |
5423 | * | |
5424 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
5425 | * we don't know about. | |
5426 | * | |
5427 | * - The ofproto client didn't configure the port as part of a bundle. | |
5428 | */ | |
5429 | if (warn) { | |
5430 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5431 | ||
5432 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
5433 | "port %"PRIu16, ofproto->up.name, in_port); | |
5434 | } | |
5435 | return NULL; | |
5436 | } | |
5437 | ||
5da5ec37 | 5438 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
5439 | * dropped. Returns true if they may be forwarded, false if they should be |
5440 | * dropped. | |
5441 | * | |
395e68ce BP |
5442 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
5443 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 5444 | * |
395e68ce BP |
5445 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
5446 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
5447 | * checked by input_vid_is_valid(). | |
abe529af BP |
5448 | * |
5449 | * May also add tags to '*tags', although the current implementation only does | |
5450 | * so in one special case. | |
5451 | */ | |
5452 | static bool | |
5453 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
395e68ce | 5454 | struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) |
abe529af | 5455 | { |
395e68ce | 5456 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 5457 | |
395e68ce BP |
5458 | /* Drop frames for reserved multicast addresses |
5459 | * only if forward_bpdu option is absent. */ | |
21f7563c | 5460 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { |
abe529af BP |
5461 | return false; |
5462 | } | |
5463 | ||
abe529af BP |
5464 | if (in_bundle->bond) { |
5465 | struct mac_entry *mac; | |
5466 | ||
5467 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
5468 | flow->dl_dst, tags)) { | |
5469 | case BV_ACCEPT: | |
5470 | break; | |
5471 | ||
5472 | case BV_DROP: | |
5473 | return false; | |
5474 | ||
5475 | case BV_DROP_IF_MOVED: | |
5476 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
5477 | if (mac && mac->port.p != in_bundle && | |
5478 | (!is_gratuitous_arp(flow) | |
5479 | || mac_entry_is_grat_arp_locked(mac))) { | |
5480 | return false; | |
5481 | } | |
5482 | break; | |
5483 | } | |
5484 | } | |
5485 | ||
5486 | return true; | |
5487 | } | |
5488 | ||
4cd78906 | 5489 | static void |
abe529af BP |
5490 | xlate_normal(struct action_xlate_ctx *ctx) |
5491 | { | |
395e68ce | 5492 | struct ofport_dpif *in_port; |
abe529af | 5493 | struct ofbundle *in_bundle; |
abe529af | 5494 | struct mac_entry *mac; |
395e68ce BP |
5495 | uint16_t vlan; |
5496 | uint16_t vid; | |
abe529af | 5497 | |
75a75043 BP |
5498 | ctx->has_normal = true; |
5499 | ||
3581c12c | 5500 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
395e68ce | 5501 | ctx->packet != NULL); |
3581c12c | 5502 | if (!in_bundle) { |
395e68ce BP |
5503 | return; |
5504 | } | |
3581c12c | 5505 | |
33158a18 JP |
5506 | /* We know 'in_port' exists unless it is "ofpp_none_bundle", |
5507 | * since lookup_input_bundle() succeeded. */ | |
3581c12c | 5508 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
395e68ce BP |
5509 | |
5510 | /* Drop malformed frames. */ | |
5511 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
5512 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
5513 | if (ctx->packet != NULL) { | |
5514 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5515 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
5516 | "VLAN tag received on port %s", | |
5517 | ctx->ofproto->up.name, in_bundle->name); | |
5518 | } | |
5519 | return; | |
5520 | } | |
5521 | ||
5522 | /* Drop frames on bundles reserved for mirroring. */ | |
5523 | if (in_bundle->mirror_out) { | |
5524 | if (ctx->packet != NULL) { | |
5525 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5526 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5527 | "%s, which is reserved exclusively for mirroring", | |
5528 | ctx->ofproto->up.name, in_bundle->name); | |
5529 | } | |
5530 | return; | |
5531 | } | |
5532 | ||
5533 | /* Check VLAN. */ | |
5534 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
5535 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5536 | return; | |
5537 | } | |
5538 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5539 | ||
5540 | /* Check other admissibility requirements. */ | |
33158a18 JP |
5541 | if (in_port && |
5542 | !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { | |
395e68ce | 5543 | return; |
abe529af BP |
5544 | } |
5545 | ||
75a75043 BP |
5546 | /* Learn source MAC. */ |
5547 | if (ctx->may_learn) { | |
abe529af BP |
5548 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
5549 | } | |
5550 | ||
5551 | /* Determine output bundle. */ | |
5552 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
5553 | &ctx->tags); | |
5554 | if (mac) { | |
c06bba01 JP |
5555 | if (mac->port.p != in_bundle) { |
5556 | output_normal(ctx, mac->port.p, vlan); | |
5557 | } | |
abe529af | 5558 | } else { |
c06bba01 | 5559 | struct ofbundle *bundle; |
abe529af | 5560 | |
c06bba01 JP |
5561 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
5562 | if (bundle != in_bundle | |
5563 | && ofbundle_includes_vlan(bundle, vlan) | |
5564 | && bundle->floodable | |
5565 | && !bundle->mirror_out) { | |
5566 | output_normal(ctx, bundle, vlan); | |
5567 | } | |
5568 | } | |
5569 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 5570 | } |
abe529af BP |
5571 | } |
5572 | \f | |
54a9cbc9 BP |
5573 | /* Optimized flow revalidation. |
5574 | * | |
5575 | * It's a difficult problem, in general, to tell which facets need to have | |
5576 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
5577 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
5578 | * table changes, we recalculate the actions for every facet. This is | |
5579 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
5580 | * doesn't change very often. | |
5581 | * | |
5582 | * However, we can expect one particular kind of OpenFlow flow table change to | |
5583 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
5584 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
5585 | * table, we add a special case that applies to flow tables in which every rule | |
5586 | * has the same form (that is, the same wildcards), except that the table is | |
5587 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
5588 | * optimize this case by tagging all of the facets that resubmit into the table | |
5589 | * and invalidating the same tag whenever a flow changes in that table. The | |
5590 | * end result is that we revalidate just the facets that need it (and sometimes | |
5591 | * a few more, but not all of the facets or even all of the facets that | |
5592 | * resubmit to the table modified by MAC learning). */ | |
5593 | ||
5594 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
5595 | * into an OpenFlow table with the given 'basis'. */ | |
5596 | static uint32_t | |
5597 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
5598 | uint32_t secret) | |
5599 | { | |
5600 | if (flow_wildcards_is_catchall(wc)) { | |
5601 | return 0; | |
5602 | } else { | |
5603 | struct flow tag_flow = *flow; | |
5604 | flow_zero_wildcards(&tag_flow, wc); | |
5605 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
5606 | } | |
5607 | } | |
5608 | ||
5609 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
5610 | * taggability of that table. | |
5611 | * | |
5612 | * This function must be called after *each* change to a flow table. If you | |
5613 | * skip calling it on some changes then the pointer comparisons at the end can | |
5614 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
5615 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
5616 | * different wildcards to be created with the same address, then this function | |
5617 | * will incorrectly skip revalidation. */ | |
5618 | static void | |
5619 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
5620 | { | |
5621 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 5622 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
5623 | struct cls_table *catchall, *other; |
5624 | struct cls_table *t; | |
5625 | ||
5626 | catchall = other = NULL; | |
5627 | ||
d0918789 | 5628 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
5629 | case 0: |
5630 | /* We could tag this OpenFlow table but it would make the logic a | |
5631 | * little harder and it's a corner case that doesn't seem worth it | |
5632 | * yet. */ | |
5633 | break; | |
5634 | ||
5635 | case 1: | |
5636 | case 2: | |
d0918789 | 5637 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
5638 | if (cls_table_is_catchall(t)) { |
5639 | catchall = t; | |
5640 | } else if (!other) { | |
5641 | other = t; | |
5642 | } else { | |
5643 | /* Indicate that we can't tag this by setting both tables to | |
5644 | * NULL. (We know that 'catchall' is already NULL.) */ | |
5645 | other = NULL; | |
5646 | } | |
5647 | } | |
5648 | break; | |
5649 | ||
5650 | default: | |
5651 | /* Can't tag this table. */ | |
5652 | break; | |
5653 | } | |
5654 | ||
5655 | if (table->catchall_table != catchall || table->other_table != other) { | |
5656 | table->catchall_table = catchall; | |
5657 | table->other_table = other; | |
5658 | ofproto->need_revalidate = true; | |
5659 | } | |
5660 | } | |
5661 | ||
5662 | /* Given 'rule' that has changed in some way (either it is a rule being | |
5663 | * inserted, a rule being deleted, or a rule whose actions are being | |
5664 | * modified), marks facets for revalidation to ensure that packets will be | |
5665 | * forwarded correctly according to the new state of the flow table. | |
5666 | * | |
5667 | * This function must be called after *each* change to a flow table. See | |
5668 | * the comment on table_update_taggable() for more information. */ | |
5669 | static void | |
5670 | rule_invalidate(const struct rule_dpif *rule) | |
5671 | { | |
5672 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5673 | ||
5674 | table_update_taggable(ofproto, rule->up.table_id); | |
5675 | ||
5676 | if (!ofproto->need_revalidate) { | |
5677 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
5678 | ||
5679 | if (table->other_table && rule->tag) { | |
5680 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
5681 | } else { | |
5682 | ofproto->need_revalidate = true; | |
5683 | } | |
5684 | } | |
5685 | } | |
5686 | \f | |
abe529af | 5687 | static bool |
7257b535 BP |
5688 | set_frag_handling(struct ofproto *ofproto_, |
5689 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5690 | { |
5691 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 5692 | |
7257b535 BP |
5693 | if (frag_handling != OFPC_FRAG_REASM) { |
5694 | ofproto->need_revalidate = true; | |
5695 | return true; | |
5696 | } else { | |
5697 | return false; | |
5698 | } | |
abe529af BP |
5699 | } |
5700 | ||
90bf1e07 | 5701 | static enum ofperr |
abe529af BP |
5702 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
5703 | const struct flow *flow, | |
5704 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
5705 | { | |
5706 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
90bf1e07 | 5707 | enum ofperr error; |
abe529af | 5708 | |
e1154f71 | 5709 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
90bf1e07 | 5710 | return OFPERR_NXBRC_BAD_IN_PORT; |
e1154f71 BP |
5711 | } |
5712 | ||
abe529af BP |
5713 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
5714 | ofproto->max_ports); | |
5715 | if (!error) { | |
80e5eed9 | 5716 | struct odputil_keybuf keybuf; |
abe529af | 5717 | struct ofpbuf *odp_actions; |
2284188b | 5718 | struct ofproto_push push; |
80e5eed9 BP |
5719 | struct ofpbuf key; |
5720 | ||
5721 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
5722 | odp_flow_key_from_flow(&key, flow); | |
abe529af | 5723 | |
2284188b EJ |
5724 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0, |
5725 | packet); | |
5726 | ||
5727 | /* Ensure that resubmits in 'ofp_actions' get accounted to their | |
5728 | * matching rules. */ | |
5729 | push.packets = 1; | |
5730 | push.bytes = packet->size; | |
5731 | push.used = time_msec(); | |
5732 | push.ctx.resubmit_hook = push_resubmit; | |
5733 | ||
5734 | odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
5735 | dpif_execute(ofproto->dpif, key.data, key.size, |
5736 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
5737 | ofpbuf_delete(odp_actions); |
5738 | } | |
5739 | return error; | |
5740 | } | |
6fca1ffb BP |
5741 | \f |
5742 | /* NetFlow. */ | |
5743 | ||
5744 | static int | |
5745 | set_netflow(struct ofproto *ofproto_, | |
5746 | const struct netflow_options *netflow_options) | |
5747 | { | |
5748 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5749 | ||
5750 | if (netflow_options) { | |
5751 | if (!ofproto->netflow) { | |
5752 | ofproto->netflow = netflow_create(); | |
5753 | } | |
5754 | return netflow_set_options(ofproto->netflow, netflow_options); | |
5755 | } else { | |
5756 | netflow_destroy(ofproto->netflow); | |
5757 | ofproto->netflow = NULL; | |
5758 | return 0; | |
5759 | } | |
5760 | } | |
abe529af BP |
5761 | |
5762 | static void | |
5763 | get_netflow_ids(const struct ofproto *ofproto_, | |
5764 | uint8_t *engine_type, uint8_t *engine_id) | |
5765 | { | |
5766 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5767 | ||
5768 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
5769 | } | |
6fca1ffb BP |
5770 | |
5771 | static void | |
5772 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
5773 | { | |
5774 | if (!facet_is_controller_flow(facet) && | |
5775 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 5776 | struct subfacet *subfacet; |
6fca1ffb BP |
5777 | struct ofexpired expired; |
5778 | ||
b0f7b9b5 BP |
5779 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
5780 | if (subfacet->installed) { | |
5781 | struct dpif_flow_stats stats; | |
6fca1ffb | 5782 | |
15baa734 | 5783 | subfacet_install(subfacet, subfacet->actions, |
b95fc6ba | 5784 | subfacet->actions_len, &stats); |
15baa734 | 5785 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5786 | } |
6fca1ffb BP |
5787 | } |
5788 | ||
5789 | expired.flow = facet->flow; | |
5790 | expired.packet_count = facet->packet_count; | |
5791 | expired.byte_count = facet->byte_count; | |
5792 | expired.used = facet->used; | |
5793 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
5794 | } | |
5795 | } | |
5796 | ||
5797 | static void | |
5798 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
5799 | { | |
5800 | struct facet *facet; | |
5801 | ||
5802 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
5803 | send_active_timeout(ofproto, facet); | |
5804 | } | |
5805 | } | |
abe529af BP |
5806 | \f |
5807 | static struct ofproto_dpif * | |
5808 | ofproto_dpif_lookup(const char *name) | |
5809 | { | |
b44a10b7 BP |
5810 | struct ofproto_dpif *ofproto; |
5811 | ||
5812 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
5813 | hash_string(name, 0), &all_ofproto_dpifs) { | |
5814 | if (!strcmp(ofproto->up.name, name)) { | |
5815 | return ofproto; | |
5816 | } | |
5817 | } | |
5818 | return NULL; | |
abe529af BP |
5819 | } |
5820 | ||
f0a3aa2e | 5821 | static void |
96e466a3 | 5822 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 5823 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 5824 | { |
490df1ef | 5825 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 5826 | |
96e466a3 EJ |
5827 | if (argc > 1) { |
5828 | ofproto = ofproto_dpif_lookup(argv[1]); | |
5829 | if (!ofproto) { | |
5830 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5831 | return; | |
5832 | } | |
d0040604 | 5833 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 EJ |
5834 | } else { |
5835 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
d0040604 | 5836 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 | 5837 | } |
f0a3aa2e | 5838 | } |
f0a3aa2e AA |
5839 | |
5840 | unixctl_command_reply(conn, 200, "table successfully flushed"); | |
5841 | } | |
5842 | ||
abe529af | 5843 | static void |
0e15264f BP |
5844 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5845 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
5846 | { |
5847 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5848 | const struct ofproto_dpif *ofproto; | |
5849 | const struct mac_entry *e; | |
5850 | ||
0e15264f | 5851 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af BP |
5852 | if (!ofproto) { |
5853 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5854 | return; | |
5855 | } | |
5856 | ||
5857 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
5858 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
5859 | struct ofbundle *bundle = e->port.p; | |
5860 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
5861 | ofbundle_get_a_port(bundle)->odp_port, | |
5862 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
5863 | } | |
5864 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
5865 | ds_destroy(&ds); | |
5866 | } | |
5867 | ||
5868 | struct ofproto_trace { | |
5869 | struct action_xlate_ctx ctx; | |
5870 | struct flow flow; | |
5871 | struct ds *result; | |
5872 | }; | |
5873 | ||
5874 | static void | |
29901626 BP |
5875 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
5876 | const struct rule_dpif *rule) | |
abe529af BP |
5877 | { |
5878 | ds_put_char_multiple(result, '\t', level); | |
5879 | if (!rule) { | |
5880 | ds_put_cstr(result, "No match\n"); | |
5881 | return; | |
5882 | } | |
5883 | ||
29901626 BP |
5884 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
5885 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 5886 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5887 | ds_put_char(result, '\n'); |
5888 | ||
5889 | ds_put_char_multiple(result, '\t', level); | |
5890 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 5891 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
5892 | ds_put_char(result, '\n'); |
5893 | } | |
5894 | ||
5895 | static void | |
5896 | trace_format_flow(struct ds *result, int level, const char *title, | |
5897 | struct ofproto_trace *trace) | |
5898 | { | |
5899 | ds_put_char_multiple(result, '\t', level); | |
5900 | ds_put_format(result, "%s: ", title); | |
5901 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5902 | ds_put_cstr(result, "unchanged"); | |
5903 | } else { | |
5904 | flow_format(result, &trace->ctx.flow); | |
5905 | trace->flow = trace->ctx.flow; | |
5906 | } | |
5907 | ds_put_char(result, '\n'); | |
5908 | } | |
5909 | ||
eb9e1c26 EJ |
5910 | static void |
5911 | trace_format_regs(struct ds *result, int level, const char *title, | |
5912 | struct ofproto_trace *trace) | |
5913 | { | |
5914 | size_t i; | |
5915 | ||
5916 | ds_put_char_multiple(result, '\t', level); | |
5917 | ds_put_format(result, "%s:", title); | |
5918 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5919 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5920 | } | |
5921 | ds_put_char(result, '\n'); | |
5922 | } | |
5923 | ||
abe529af BP |
5924 | static void |
5925 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
5926 | { | |
5927 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5928 | struct ds *result = trace->result; | |
5929 | ||
5930 | ds_put_char(result, '\n'); | |
5931 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 5932 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
29901626 | 5933 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
5934 | } |
5935 | ||
5936 | static void | |
0e15264f | 5937 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
5938 | void *aux OVS_UNUSED) |
5939 | { | |
0e15264f | 5940 | const char *dpname = argv[1]; |
abe529af | 5941 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
5942 | struct ofpbuf odp_key; |
5943 | struct ofpbuf *packet; | |
abe529af | 5944 | struct rule_dpif *rule; |
e84173dc | 5945 | ovs_be16 initial_tci; |
abe529af BP |
5946 | struct ds result; |
5947 | struct flow flow; | |
abe529af BP |
5948 | char *s; |
5949 | ||
876b0e1c BP |
5950 | packet = NULL; |
5951 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
5952 | ds_init(&result); |
5953 | ||
e84173dc BP |
5954 | ofproto = ofproto_dpif_lookup(dpname); |
5955 | if (!ofproto) { | |
5956 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5957 | "for help)"); | |
5958 | goto exit; | |
5959 | } | |
0e15264f | 5960 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 5961 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
5962 | const char *flow_s = argv[2]; |
5963 | const char *generate_s = argv[3]; | |
876b0e1c BP |
5964 | int error; |
5965 | ||
df2c07f4 | 5966 | /* Convert string to datapath key. */ |
876b0e1c | 5967 | ofpbuf_init(&odp_key, 0); |
0e15264f | 5968 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); |
876b0e1c BP |
5969 | if (error) { |
5970 | unixctl_command_reply(conn, 501, "Bad flow syntax"); | |
5971 | goto exit; | |
5972 | } | |
5973 | ||
5974 | /* Convert odp_key to flow. */ | |
e84173dc BP |
5975 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, |
5976 | odp_key.size, &flow, | |
e2a6ca36 | 5977 | &initial_tci, NULL); |
e84173dc | 5978 | if (error == ODP_FIT_ERROR) { |
876b0e1c BP |
5979 | unixctl_command_reply(conn, 501, "Invalid flow"); |
5980 | goto exit; | |
5981 | } | |
8b3b8dd1 BP |
5982 | |
5983 | /* Generate a packet, if requested. */ | |
0e15264f | 5984 | if (generate_s) { |
8b3b8dd1 BP |
5985 | packet = ofpbuf_new(0); |
5986 | flow_compose(packet, &flow); | |
5987 | } | |
0e15264f | 5988 | } else if (argc == 6) { |
abff858b | 5989 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
5990 | const char *priority_s = argv[2]; |
5991 | const char *tun_id_s = argv[3]; | |
5992 | const char *in_port_s = argv[4]; | |
5993 | const char *packet_s = argv[5]; | |
5994 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
5995 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
5996 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 5997 | const char *msg; |
0e15264f | 5998 | |
e22f1753 BP |
5999 | msg = eth_from_hex(packet_s, &packet); |
6000 | if (msg) { | |
6001 | unixctl_command_reply(conn, 501, msg); | |
876b0e1c BP |
6002 | goto exit; |
6003 | } | |
6004 | ||
6005 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 6006 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
6007 | ds_put_cstr(&result, s); |
6008 | free(s); | |
6009 | ||
abff858b | 6010 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 6011 | initial_tci = flow.vlan_tci; |
876b0e1c | 6012 | } else { |
abe529af BP |
6013 | unixctl_command_reply(conn, 501, "Bad command syntax"); |
6014 | goto exit; | |
6015 | } | |
6016 | ||
abe529af BP |
6017 | ds_put_cstr(&result, "Flow: "); |
6018 | flow_format(&result, &flow); | |
6019 | ds_put_char(&result, '\n'); | |
6020 | ||
29901626 BP |
6021 | rule = rule_dpif_lookup(ofproto, &flow, 0); |
6022 | trace_format_rule(&result, 0, 0, rule); | |
abe529af BP |
6023 | if (rule) { |
6024 | struct ofproto_trace trace; | |
6025 | struct ofpbuf *odp_actions; | |
6026 | ||
6027 | trace.result = &result; | |
6028 | trace.flow = flow; | |
54834960 EJ |
6029 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, |
6030 | rule->up.flow_cookie, packet); | |
abe529af BP |
6031 | trace.ctx.resubmit_hook = trace_resubmit; |
6032 | odp_actions = xlate_actions(&trace.ctx, | |
6033 | rule->up.actions, rule->up.n_actions); | |
6034 | ||
6035 | ds_put_char(&result, '\n'); | |
6036 | trace_format_flow(&result, 0, "Final flow", &trace); | |
6037 | ds_put_cstr(&result, "Datapath actions: "); | |
6038 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
6039 | ofpbuf_delete(odp_actions); | |
876b0e1c BP |
6040 | |
6041 | if (!trace.ctx.may_set_up_flow) { | |
6042 | if (packet) { | |
6043 | ds_put_cstr(&result, "\nThis flow is not cachable."); | |
6044 | } else { | |
6045 | ds_put_cstr(&result, "\nThe datapath actions are incomplete--" | |
6046 | "for complete actions, please supply a packet."); | |
6047 | } | |
6048 | } | |
abe529af BP |
6049 | } |
6050 | ||
6051 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
6052 | ||
6053 | exit: | |
6054 | ds_destroy(&result); | |
876b0e1c BP |
6055 | ofpbuf_delete(packet); |
6056 | ofpbuf_uninit(&odp_key); | |
abe529af BP |
6057 | } |
6058 | ||
7ee20df1 | 6059 | static void |
0e15264f BP |
6060 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6061 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6062 | { |
6063 | clogged = true; | |
6064 | unixctl_command_reply(conn, 200, NULL); | |
6065 | } | |
6066 | ||
6067 | static void | |
0e15264f BP |
6068 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6069 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6070 | { |
6071 | clogged = false; | |
6072 | unixctl_command_reply(conn, 200, NULL); | |
6073 | } | |
6074 | ||
6814e51f BP |
6075 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6076 | * 'reply' describing the results. */ | |
6077 | static void | |
6078 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6079 | { | |
6080 | struct facet *facet; | |
6081 | int errors; | |
6082 | ||
6083 | errors = 0; | |
6084 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6085 | if (!facet_check_consistency(facet)) { | |
6086 | errors++; | |
6087 | } | |
6088 | } | |
6089 | if (errors) { | |
6090 | ofproto->need_revalidate = true; | |
6091 | } | |
6092 | ||
6093 | if (errors) { | |
6094 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6095 | ofproto->up.name, errors); | |
6096 | } else { | |
6097 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6098 | } | |
6099 | } | |
6100 | ||
6101 | static void | |
6102 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6103 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6104 | { | |
6105 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6106 | struct ofproto_dpif *ofproto; | |
6107 | ||
6108 | if (argc > 1) { | |
6109 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6110 | if (!ofproto) { | |
6111 | unixctl_command_reply(conn, 501, "Unknown ofproto (use " | |
6112 | "ofproto/list for help)"); | |
6113 | return; | |
6114 | } | |
6115 | ofproto_dpif_self_check__(ofproto, &reply); | |
6116 | } else { | |
6117 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6118 | ofproto_dpif_self_check__(ofproto, &reply); | |
6119 | } | |
6120 | } | |
6121 | ||
6122 | unixctl_command_reply(conn, 200, ds_cstr(&reply)); | |
6123 | ds_destroy(&reply); | |
6124 | } | |
6125 | ||
abe529af BP |
6126 | static void |
6127 | ofproto_dpif_unixctl_init(void) | |
6128 | { | |
6129 | static bool registered; | |
6130 | if (registered) { | |
6131 | return; | |
6132 | } | |
6133 | registered = true; | |
6134 | ||
0e15264f BP |
6135 | unixctl_command_register( |
6136 | "ofproto/trace", | |
6137 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
aa3080c9 | 6138 | 2, 5, ofproto_unixctl_trace, NULL); |
96e466a3 | 6139 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6140 | ofproto_unixctl_fdb_flush, NULL); |
6141 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6142 | ofproto_unixctl_fdb_show, NULL); | |
6143 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6144 | ofproto_dpif_clog, NULL); | |
6145 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6146 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6147 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6148 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6149 | } |
6150 | \f | |
52a90c29 BP |
6151 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6152 | * | |
6153 | * This is deprecated. It is only for compatibility with broken device drivers | |
6154 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6155 | * devices are not used. When broken device drivers are no longer in | |
6156 | * widespread use, we will delete these interfaces. */ | |
6157 | ||
6158 | static int | |
6159 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6160 | { | |
6161 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6162 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6163 | ||
6164 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6165 | && vid == ofport->vlandev_vid) { | |
6166 | return 0; | |
6167 | } | |
6168 | ||
6169 | ofproto->need_revalidate = true; | |
6170 | ||
6171 | if (ofport->realdev_ofp_port) { | |
6172 | vsp_remove(ofport); | |
6173 | } | |
6174 | if (realdev_ofp_port && ofport->bundle) { | |
6175 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6176 | * themselves be part of a bundle. */ | |
6177 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6178 | } | |
6179 | ||
6180 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6181 | ofport->vlandev_vid = vid; | |
6182 | ||
6183 | if (realdev_ofp_port) { | |
6184 | vsp_add(ofport, realdev_ofp_port, vid); | |
6185 | } | |
6186 | ||
6187 | return 0; | |
6188 | } | |
6189 | ||
6190 | static uint32_t | |
6191 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
6192 | { | |
6193 | return hash_2words(realdev_ofp_port, vid); | |
6194 | } | |
6195 | ||
6196 | static uint32_t | |
6197 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6198 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
6199 | { | |
6200 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
6201 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
6202 | int vid = vlan_tci_to_vid(vlan_tci); | |
6203 | const struct vlan_splinter *vsp; | |
6204 | ||
6205 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6206 | hash_realdev_vid(realdev_ofp_port, vid), | |
6207 | &ofproto->realdev_vid_map) { | |
6208 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6209 | && vsp->vid == vid) { | |
6210 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
6211 | } | |
6212 | } | |
6213 | } | |
6214 | return realdev_odp_port; | |
6215 | } | |
6216 | ||
6217 | static struct vlan_splinter * | |
6218 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
6219 | { | |
6220 | struct vlan_splinter *vsp; | |
6221 | ||
6222 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
6223 | &ofproto->vlandev_map) { | |
6224 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6225 | return vsp; | |
6226 | } | |
6227 | } | |
6228 | ||
6229 | return NULL; | |
6230 | } | |
6231 | ||
6232 | static uint16_t | |
6233 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
6234 | uint16_t vlandev_ofp_port, int *vid) | |
6235 | { | |
6236 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6237 | const struct vlan_splinter *vsp; | |
6238 | ||
6239 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6240 | if (vsp) { | |
6241 | if (vid) { | |
6242 | *vid = vsp->vid; | |
6243 | } | |
6244 | return vsp->realdev_ofp_port; | |
6245 | } | |
6246 | } | |
6247 | return 0; | |
6248 | } | |
6249 | ||
6250 | static void | |
6251 | vsp_remove(struct ofport_dpif *port) | |
6252 | { | |
6253 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6254 | struct vlan_splinter *vsp; | |
6255 | ||
6256 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
6257 | if (vsp) { | |
6258 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6259 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6260 | free(vsp); | |
6261 | ||
6262 | port->realdev_ofp_port = 0; | |
6263 | } else { | |
6264 | VLOG_ERR("missing vlan device record"); | |
6265 | } | |
6266 | } | |
6267 | ||
6268 | static void | |
6269 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
6270 | { | |
6271 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6272 | ||
6273 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
6274 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
6275 | == realdev_ofp_port)) { | |
6276 | struct vlan_splinter *vsp; | |
6277 | ||
6278 | vsp = xmalloc(sizeof *vsp); | |
6279 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6280 | hash_int(port->up.ofp_port, 0)); | |
6281 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6282 | hash_realdev_vid(realdev_ofp_port, vid)); | |
6283 | vsp->realdev_ofp_port = realdev_ofp_port; | |
6284 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6285 | vsp->vid = vid; | |
6286 | ||
6287 | port->realdev_ofp_port = realdev_ofp_port; | |
6288 | } else { | |
6289 | VLOG_ERR("duplicate vlan device record"); | |
6290 | } | |
6291 | } | |
6292 | \f | |
abe529af BP |
6293 | const struct ofproto_class ofproto_dpif_class = { |
6294 | enumerate_types, | |
6295 | enumerate_names, | |
6296 | del, | |
6297 | alloc, | |
6298 | construct, | |
6299 | destruct, | |
6300 | dealloc, | |
6301 | run, | |
5fcc0d00 | 6302 | run_fast, |
abe529af BP |
6303 | wait, |
6304 | flush, | |
6c1491fb BP |
6305 | get_features, |
6306 | get_tables, | |
abe529af BP |
6307 | port_alloc, |
6308 | port_construct, | |
6309 | port_destruct, | |
6310 | port_dealloc, | |
6311 | port_modified, | |
6312 | port_reconfigured, | |
6313 | port_query_by_name, | |
6314 | port_add, | |
6315 | port_del, | |
6527c598 | 6316 | port_get_stats, |
abe529af BP |
6317 | port_dump_start, |
6318 | port_dump_next, | |
6319 | port_dump_done, | |
6320 | port_poll, | |
6321 | port_poll_wait, | |
6322 | port_is_lacp_current, | |
0ab6decf | 6323 | NULL, /* rule_choose_table */ |
abe529af BP |
6324 | rule_alloc, |
6325 | rule_construct, | |
6326 | rule_destruct, | |
6327 | rule_dealloc, | |
abe529af BP |
6328 | rule_get_stats, |
6329 | rule_execute, | |
6330 | rule_modify_actions, | |
7257b535 | 6331 | set_frag_handling, |
abe529af BP |
6332 | packet_out, |
6333 | set_netflow, | |
6334 | get_netflow_ids, | |
6335 | set_sflow, | |
6336 | set_cfm, | |
a5610457 | 6337 | get_cfm_fault, |
1de11730 | 6338 | get_cfm_remote_mpids, |
21f7563c JP |
6339 | set_stp, |
6340 | get_stp_status, | |
6341 | set_stp_port, | |
6342 | get_stp_port_status, | |
8b36f51e | 6343 | set_queues, |
abe529af BP |
6344 | bundle_set, |
6345 | bundle_remove, | |
6346 | mirror_set, | |
9d24de3b | 6347 | mirror_get_stats, |
abe529af BP |
6348 | set_flood_vlans, |
6349 | is_mirror_output_bundle, | |
8402c74b | 6350 | forward_bpdu_changed, |
52a90c29 | 6351 | set_realdev, |
abe529af | 6352 | }; |