]>
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 BP |
36 | #include "mac-learning.h" |
37 | #include "multipath.h" | |
38 | #include "netdev.h" | |
39 | #include "netlink.h" | |
40 | #include "nx-match.h" | |
41 | #include "odp-util.h" | |
42 | #include "ofp-util.h" | |
43 | #include "ofpbuf.h" | |
44 | #include "ofp-print.h" | |
bae473fe | 45 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
46 | #include "poll-loop.h" |
47 | #include "timer.h" | |
6c1491fb | 48 | #include "unaligned.h" |
abe529af BP |
49 | #include "unixctl.h" |
50 | #include "vlan-bitmap.h" | |
51 | #include "vlog.h" | |
52 | ||
53 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
54 | ||
55 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
56 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
57 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
58 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
59 | COVERAGE_DEFINE(facet_changed_rule); | |
60 | COVERAGE_DEFINE(facet_invalidated); | |
61 | COVERAGE_DEFINE(facet_revalidate); | |
62 | COVERAGE_DEFINE(facet_unexpected); | |
63 | ||
29901626 | 64 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 65 | * flow translation. */ |
642a5c05 | 66 | #define MAX_RESUBMIT_RECURSION 32 |
abe529af | 67 | |
9cdaaebe BP |
68 | /* Number of implemented OpenFlow tables. */ |
69 | enum { N_TABLES = 255 }; | |
70 | BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255); | |
71 | ||
abe529af BP |
72 | struct ofport_dpif; |
73 | struct ofproto_dpif; | |
74 | ||
75 | struct rule_dpif { | |
76 | struct rule up; | |
77 | ||
78 | long long int used; /* Time last used; time created if not used. */ | |
79 | ||
80 | /* These statistics: | |
81 | * | |
82 | * - Do include packets and bytes from facets that have been deleted or | |
83 | * whose own statistics have been folded into the rule. | |
84 | * | |
85 | * - Do include packets and bytes sent "by hand" that were accounted to | |
86 | * the rule without any facet being involved (this is a rare corner | |
87 | * case in rule_execute()). | |
88 | * | |
89 | * - Do not include packet or bytes that can be obtained from any facet's | |
90 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 91 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
92 | */ |
93 | uint64_t packet_count; /* Number of packets received. */ | |
94 | uint64_t byte_count; /* Number of bytes received. */ | |
95 | ||
54a9cbc9 BP |
96 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
97 | ||
abe529af BP |
98 | struct list facets; /* List of "struct facet"s. */ |
99 | }; | |
100 | ||
101 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
102 | { | |
103 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
104 | } | |
105 | ||
29901626 BP |
106 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
107 | const struct flow *, uint8_t table); | |
abe529af | 108 | |
b0f7b9b5 BP |
109 | static void flow_push_stats(const struct rule_dpif *, const struct flow *, |
110 | uint64_t packets, uint64_t bytes, | |
111 | long long int used); | |
112 | ||
113 | static uint32_t rule_calculate_tag(const struct flow *, | |
114 | const struct flow_wildcards *, | |
115 | uint32_t basis); | |
116 | static void rule_invalidate(const struct rule_dpif *); | |
117 | ||
abe529af BP |
118 | #define MAX_MIRRORS 32 |
119 | typedef uint32_t mirror_mask_t; | |
120 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
121 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
122 | struct ofmirror { | |
123 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
124 | size_t idx; /* In ofproto's "mirrors" array. */ | |
125 | void *aux; /* Key supplied by ofproto's client. */ | |
126 | char *name; /* Identifier for log messages. */ | |
127 | ||
128 | /* Selection criteria. */ | |
129 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
130 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
131 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
132 | ||
9ba15e2a | 133 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
134 | struct ofbundle *out; /* Output port or NULL. */ |
135 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 136 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
137 | |
138 | /* Counters. */ | |
139 | int64_t packet_count; /* Number of packets sent. */ | |
140 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
141 | }; |
142 | ||
143 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
144 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
145 | mirror_mask_t mirrors, | |
146 | uint64_t packets, uint64_t bytes); | |
abe529af | 147 | |
abe529af BP |
148 | struct ofbundle { |
149 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
150 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
151 | void *aux; /* Key supplied by ofproto's client. */ | |
152 | char *name; /* Identifier for log messages. */ | |
153 | ||
154 | /* Configuration. */ | |
155 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 156 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
157 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
158 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
159 | * NULL if all VLANs are trunked. */ | |
160 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
161 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 162 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
163 | |
164 | /* Status. */ | |
165 | bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ | |
166 | ||
167 | /* Port mirroring info. */ | |
168 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
169 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
170 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
171 | }; | |
172 | ||
173 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 174 | static void bundle_update(struct ofbundle *); |
abe529af BP |
175 | static void bundle_destroy(struct ofbundle *); |
176 | static void bundle_del_port(struct ofport_dpif *); | |
177 | static void bundle_run(struct ofbundle *); | |
178 | static void bundle_wait(struct ofbundle *); | |
3581c12c JP |
179 | static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *, |
180 | uint16_t in_port, bool warn); | |
abe529af | 181 | |
33158a18 JP |
182 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
183 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
184 | * when an input bundle is needed for validation (e.g., mirroring or | |
185 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
186 | * any 'port' structs, so care must be taken when dealing with it. */ | |
187 | static struct ofbundle ofpp_none_bundle = { | |
188 | .name = "OFPP_NONE", | |
189 | .vlan_mode = PORT_VLAN_TRUNK | |
190 | }; | |
191 | ||
21f7563c JP |
192 | static void stp_run(struct ofproto_dpif *ofproto); |
193 | static void stp_wait(struct ofproto_dpif *ofproto); | |
194 | ||
5da5ec37 BP |
195 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
196 | ||
abe529af BP |
197 | struct action_xlate_ctx { |
198 | /* action_xlate_ctx_init() initializes these members. */ | |
199 | ||
200 | /* The ofproto. */ | |
201 | struct ofproto_dpif *ofproto; | |
202 | ||
203 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
204 | * this flow when actions change header fields. */ | |
205 | struct flow flow; | |
206 | ||
207 | /* The packet corresponding to 'flow', or a null pointer if we are | |
208 | * revalidating without a packet to refer to. */ | |
209 | const struct ofpbuf *packet; | |
210 | ||
75a75043 BP |
211 | /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We |
212 | * want to execute them if we are actually processing a packet, or if we | |
213 | * are accounting for packets that the datapath has processed, but not if | |
214 | * we are just revalidating. */ | |
215 | bool may_learn; | |
216 | ||
54834960 EJ |
217 | /* Cookie of the currently matching rule, or 0. */ |
218 | ovs_be64 cookie; | |
219 | ||
abe529af BP |
220 | /* If nonnull, called just before executing a resubmit action. |
221 | * | |
222 | * This is normally null so the client has to set it manually after | |
223 | * calling action_xlate_ctx_init(). */ | |
224 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
225 | ||
abe529af BP |
226 | /* xlate_actions() initializes and uses these members. The client might want |
227 | * to look at them after it returns. */ | |
228 | ||
229 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 230 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
231 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
232 | * be reassessed for every packet. */ | |
75a75043 BP |
233 | bool has_learn; /* Actions include NXAST_LEARN? */ |
234 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af | 235 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 236 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
237 | |
238 | /* xlate_actions() initializes and uses these members, but the client has no | |
239 | * reason to look at them. */ | |
240 | ||
241 | int recurse; /* Recursion level, via xlate_table_action. */ | |
b3e9b2ed | 242 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 243 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 244 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
245 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
246 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
247 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 248 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
249 | }; |
250 | ||
251 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
252 | struct ofproto_dpif *, const struct flow *, | |
54834960 EJ |
253 | ovs_be16 initial_tci, ovs_be64 cookie, |
254 | const struct ofpbuf *); | |
abe529af BP |
255 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, |
256 | const union ofp_action *in, size_t n_in); | |
257 | ||
b0f7b9b5 BP |
258 | /* An exact-match instantiation of an OpenFlow flow. |
259 | * | |
260 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
261 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
262 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
263 | * the facet. When the kernel module (or other dpif implementation) and Open | |
264 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
265 | * one subfacet per facet. If the dpif implementation supports more-specific | |
266 | * flow matching than userspace, however, a facet can have more than one | |
267 | * subfacet, each of which corresponds to some distinction in flow that | |
268 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
269 | * |
270 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
271 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 272 | struct facet { |
b0f7b9b5 BP |
273 | /* Owners. */ |
274 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
275 | struct list list_node; /* In owning rule's 'facets' list. */ | |
276 | struct rule_dpif *rule; /* Owning rule. */ | |
277 | ||
278 | /* Owned data. */ | |
279 | struct list subfacets; | |
abe529af BP |
280 | long long int used; /* Time last used; time created if not used. */ |
281 | ||
b0f7b9b5 BP |
282 | /* Key. */ |
283 | struct flow flow; | |
284 | ||
abe529af BP |
285 | /* These statistics: |
286 | * | |
287 | * - Do include packets and bytes sent "by hand", e.g. with | |
288 | * dpif_execute(). | |
289 | * | |
290 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 291 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 292 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
293 | * |
294 | * - Do not include packets or bytes that can be obtained from the | |
295 | * datapath for any existing subfacet. | |
abe529af BP |
296 | */ |
297 | uint64_t packet_count; /* Number of packets received. */ | |
298 | uint64_t byte_count; /* Number of bytes received. */ | |
299 | ||
b0f7b9b5 | 300 | /* Resubmit statistics. */ |
9d24de3b JP |
301 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
302 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
303 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 304 | |
b0f7b9b5 | 305 | /* Accounting. */ |
907a4c5e | 306 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 307 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
abe529af | 308 | |
b95fc6ba BP |
309 | /* Properties of datapath actions. |
310 | * | |
311 | * Every subfacet has its own actions because actions can differ slightly | |
312 | * between splintered and non-splintered subfacets due to the VLAN tag | |
313 | * being initially different (present vs. absent). All of them have these | |
314 | * properties in common so we just store one copy of them here. */ | |
b0f7b9b5 | 315 | bool may_install; /* Reassess actions for every packet? */ |
75a75043 BP |
316 | bool has_learn; /* Actions include NXAST_LEARN? */ |
317 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
b0f7b9b5 | 318 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 319 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
abe529af BP |
320 | }; |
321 | ||
f3827897 | 322 | static struct facet *facet_create(struct rule_dpif *, const struct flow *); |
15baa734 | 323 | static void facet_remove(struct facet *); |
abe529af BP |
324 | static void facet_free(struct facet *); |
325 | ||
326 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
327 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
328 | const struct flow *); | |
15baa734 | 329 | static bool facet_revalidate(struct facet *); |
6814e51f | 330 | static bool facet_check_consistency(struct facet *); |
abe529af | 331 | |
15baa734 | 332 | static void facet_flush_stats(struct facet *); |
abe529af | 333 | |
15baa734 | 334 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 335 | static void facet_reset_counters(struct facet *); |
abe529af | 336 | static void facet_push_stats(struct facet *); |
15baa734 | 337 | static void facet_account(struct facet *); |
abe529af BP |
338 | |
339 | static bool facet_is_controller_flow(struct facet *); | |
340 | ||
b95fc6ba | 341 | /* A dpif flow and actions associated with a facet. |
b0f7b9b5 BP |
342 | * |
343 | * See also the large comment on struct facet. */ | |
344 | struct subfacet { | |
345 | /* Owners. */ | |
346 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
347 | struct list list_node; /* In struct facet's 'facets' list. */ | |
348 | struct facet *facet; /* Owning facet. */ | |
349 | ||
350 | /* Key. | |
351 | * | |
352 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
353 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
354 | * regenerate the ODP flow key from ->facet->flow. */ | |
355 | enum odp_key_fitness key_fitness; | |
356 | struct nlattr *key; | |
357 | int key_len; | |
abe529af | 358 | |
b0f7b9b5 BP |
359 | long long int used; /* Time last used; time created if not used. */ |
360 | ||
361 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
362 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
363 | ||
b95fc6ba BP |
364 | /* Datapath actions. |
365 | * | |
366 | * These should be essentially identical for every subfacet in a facet, but | |
367 | * may differ in trivial ways due to VLAN splinters. */ | |
368 | size_t actions_len; /* Number of bytes in actions[]. */ | |
369 | struct nlattr *actions; /* Datapath actions. */ | |
370 | ||
b0f7b9b5 | 371 | bool installed; /* Installed in datapath? */ |
e84173dc BP |
372 | |
373 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
374 | * splinters can cause it to differ. This value should be removed when | |
375 | * the VLAN splinters feature is no longer needed. */ | |
376 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
b0f7b9b5 BP |
377 | }; |
378 | ||
15baa734 | 379 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, |
b0f7b9b5 | 380 | const struct nlattr *key, |
e84173dc | 381 | size_t key_len, ovs_be16 initial_tci); |
b0f7b9b5 | 382 | static struct subfacet *subfacet_find(struct ofproto_dpif *, |
6a542738 | 383 | const struct nlattr *key, size_t key_len); |
15baa734 BP |
384 | static void subfacet_destroy(struct subfacet *); |
385 | static void subfacet_destroy__(struct subfacet *); | |
6814e51f BP |
386 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, |
387 | struct ofpbuf *key); | |
b0f7b9b5 BP |
388 | static void subfacet_reset_dp_stats(struct subfacet *, |
389 | struct dpif_flow_stats *); | |
15baa734 BP |
390 | static void subfacet_update_time(struct subfacet *, long long int used); |
391 | static void subfacet_update_stats(struct subfacet *, | |
b0f7b9b5 | 392 | const struct dpif_flow_stats *); |
15baa734 | 393 | static void subfacet_make_actions(struct subfacet *, |
b95fc6ba | 394 | const struct ofpbuf *packet); |
15baa734 | 395 | static int subfacet_install(struct subfacet *, |
b0f7b9b5 BP |
396 | const struct nlattr *actions, size_t actions_len, |
397 | struct dpif_flow_stats *); | |
15baa734 | 398 | static void subfacet_uninstall(struct subfacet *); |
54a9cbc9 | 399 | |
abe529af BP |
400 | struct ofport_dpif { |
401 | struct ofport up; | |
402 | ||
403 | uint32_t odp_port; | |
404 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
405 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
406 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
407 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 408 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 409 | bool may_enable; /* May be enabled in bonds. */ |
21f7563c | 410 | |
52a90c29 | 411 | /* Spanning tree. */ |
21f7563c JP |
412 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
413 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
414 | long long int stp_state_entered; | |
8b36f51e EJ |
415 | |
416 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
417 | |
418 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
419 | * | |
420 | * This is deprecated. It is only for compatibility with broken device | |
421 | * drivers in old versions of Linux that do not properly support VLANs when | |
422 | * VLAN devices are not used. When broken device drivers are no longer in | |
423 | * widespread use, we will delete these interfaces. */ | |
424 | uint16_t realdev_ofp_port; | |
425 | int vlandev_vid; | |
8b36f51e EJ |
426 | }; |
427 | ||
428 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
429 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
430 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
431 | struct priority_to_dscp { | |
432 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
433 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
434 | ||
435 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
436 | }; |
437 | ||
52a90c29 BP |
438 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
439 | * | |
440 | * This is deprecated. It is only for compatibility with broken device drivers | |
441 | * in old versions of Linux that do not properly support VLANs when VLAN | |
442 | * devices are not used. When broken device drivers are no longer in | |
443 | * widespread use, we will delete these interfaces. */ | |
444 | struct vlan_splinter { | |
445 | struct hmap_node realdev_vid_node; | |
446 | struct hmap_node vlandev_node; | |
447 | uint16_t realdev_ofp_port; | |
448 | uint16_t vlandev_ofp_port; | |
449 | int vid; | |
450 | }; | |
451 | ||
452 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
453 | uint32_t realdev, ovs_be16 vlan_tci); | |
454 | static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *, | |
455 | uint16_t vlandev, int *vid); | |
456 | static void vsp_remove(struct ofport_dpif *); | |
457 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
458 | ||
abe529af BP |
459 | static struct ofport_dpif * |
460 | ofport_dpif_cast(const struct ofport *ofport) | |
461 | { | |
462 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
463 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
464 | } | |
465 | ||
466 | static void port_run(struct ofport_dpif *); | |
467 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 468 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 469 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 470 | |
7ee20df1 BP |
471 | struct dpif_completion { |
472 | struct list list_node; | |
473 | struct ofoperation *op; | |
474 | }; | |
475 | ||
54a9cbc9 BP |
476 | /* Extra information about a classifier table. |
477 | * Currently used just for optimized flow revalidation. */ | |
478 | struct table_dpif { | |
479 | /* If either of these is nonnull, then this table has a form that allows | |
480 | * flows to be tagged to avoid revalidating most flows for the most common | |
481 | * kinds of flow table changes. */ | |
482 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
483 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
484 | uint32_t basis; /* Keeps each table's tags separate. */ | |
485 | }; | |
486 | ||
abe529af | 487 | struct ofproto_dpif { |
b44a10b7 | 488 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
489 | struct ofproto up; |
490 | struct dpif *dpif; | |
491 | int max_ports; | |
492 | ||
6c1491fb BP |
493 | /* Statistics. */ |
494 | uint64_t n_matches; | |
495 | ||
abe529af BP |
496 | /* Bridging. */ |
497 | struct netflow *netflow; | |
bae473fe | 498 | struct dpif_sflow *sflow; |
abe529af BP |
499 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
500 | struct mac_learning *ml; | |
501 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
502 | bool has_bonded_bundles; | |
503 | ||
504 | /* Expiration. */ | |
505 | struct timer next_expiration; | |
506 | ||
507 | /* Facets. */ | |
508 | struct hmap facets; | |
b0f7b9b5 | 509 | struct hmap subfacets; |
54a9cbc9 BP |
510 | |
511 | /* Revalidation. */ | |
512 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
513 | bool need_revalidate; |
514 | struct tag_set revalidate_set; | |
7ee20df1 BP |
515 | |
516 | /* Support for debugging async flow mods. */ | |
517 | struct list completions; | |
daff3353 EJ |
518 | |
519 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
520 | struct netdev_stats stats; /* To account packets generated and consumed in |
521 | * userspace. */ | |
21f7563c JP |
522 | |
523 | /* Spanning tree. */ | |
524 | struct stp *stp; | |
525 | long long int stp_last_tick; | |
52a90c29 BP |
526 | |
527 | /* VLAN splinters. */ | |
528 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
529 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
530 | }; |
531 | ||
7ee20df1 BP |
532 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
533 | * for debugging the asynchronous flow_mod implementation.) */ | |
534 | static bool clogged; | |
535 | ||
b44a10b7 BP |
536 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
537 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
538 | ||
abe529af BP |
539 | static void ofproto_dpif_unixctl_init(void); |
540 | ||
541 | static struct ofproto_dpif * | |
542 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
543 | { | |
544 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
545 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
546 | } | |
547 | ||
548 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
549 | uint16_t ofp_port); | |
550 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
551 | uint32_t odp_port); | |
552 | ||
553 | /* Packet processing. */ | |
554 | static void update_learning_table(struct ofproto_dpif *, | |
555 | const struct flow *, int vlan, | |
556 | struct ofbundle *); | |
501f8d1f BP |
557 | /* Upcalls. */ |
558 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 559 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
560 | |
561 | /* Flow expiration. */ | |
562 | static int expire(struct ofproto_dpif *); | |
563 | ||
6fca1ffb BP |
564 | /* NetFlow. */ |
565 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
566 | ||
abe529af | 567 | /* Utilities. */ |
52a90c29 | 568 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6ff686f2 PS |
569 | static size_t |
570 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
571 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
572 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
573 | const struct flow *flow); | |
abe529af BP |
574 | /* Global variables. */ |
575 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
576 | \f | |
577 | /* Factory functions. */ | |
578 | ||
579 | static void | |
580 | enumerate_types(struct sset *types) | |
581 | { | |
582 | dp_enumerate_types(types); | |
583 | } | |
584 | ||
585 | static int | |
586 | enumerate_names(const char *type, struct sset *names) | |
587 | { | |
588 | return dp_enumerate_names(type, names); | |
589 | } | |
590 | ||
591 | static int | |
592 | del(const char *type, const char *name) | |
593 | { | |
594 | struct dpif *dpif; | |
595 | int error; | |
596 | ||
597 | error = dpif_open(name, type, &dpif); | |
598 | if (!error) { | |
599 | error = dpif_delete(dpif); | |
600 | dpif_close(dpif); | |
601 | } | |
602 | return error; | |
603 | } | |
604 | \f | |
605 | /* Basic life-cycle. */ | |
606 | ||
607 | static struct ofproto * | |
608 | alloc(void) | |
609 | { | |
610 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
611 | return &ofproto->up; | |
612 | } | |
613 | ||
614 | static void | |
615 | dealloc(struct ofproto *ofproto_) | |
616 | { | |
617 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
618 | free(ofproto); | |
619 | } | |
620 | ||
621 | static int | |
073e2a6f | 622 | construct(struct ofproto *ofproto_, int *n_tablesp) |
abe529af BP |
623 | { |
624 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
625 | const char *name = ofproto->up.name; | |
626 | int error; | |
627 | int i; | |
628 | ||
629 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
630 | if (error) { | |
631 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
632 | return error; | |
633 | } | |
634 | ||
635 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 636 | ofproto->n_matches = 0; |
abe529af | 637 | |
be8194bb JG |
638 | dpif_flow_flush(ofproto->dpif); |
639 | dpif_recv_purge(ofproto->dpif); | |
640 | ||
a12b3ead | 641 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
642 | if (error) { |
643 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
644 | dpif_close(ofproto->dpif); | |
645 | return error; | |
646 | } | |
abe529af BP |
647 | |
648 | ofproto->netflow = NULL; | |
649 | ofproto->sflow = NULL; | |
21f7563c | 650 | ofproto->stp = NULL; |
abe529af BP |
651 | hmap_init(&ofproto->bundles); |
652 | ofproto->ml = mac_learning_create(); | |
653 | for (i = 0; i < MAX_MIRRORS; i++) { | |
654 | ofproto->mirrors[i] = NULL; | |
655 | } | |
656 | ofproto->has_bonded_bundles = false; | |
657 | ||
658 | timer_set_duration(&ofproto->next_expiration, 1000); | |
659 | ||
660 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 661 | hmap_init(&ofproto->subfacets); |
54a9cbc9 BP |
662 | |
663 | for (i = 0; i < N_TABLES; i++) { | |
664 | struct table_dpif *table = &ofproto->tables[i]; | |
665 | ||
666 | table->catchall_table = NULL; | |
667 | table->other_table = NULL; | |
668 | table->basis = random_uint32(); | |
669 | } | |
abe529af BP |
670 | ofproto->need_revalidate = false; |
671 | tag_set_init(&ofproto->revalidate_set); | |
672 | ||
7ee20df1 BP |
673 | list_init(&ofproto->completions); |
674 | ||
abe529af BP |
675 | ofproto_dpif_unixctl_init(); |
676 | ||
daff3353 EJ |
677 | ofproto->has_bundle_action = false; |
678 | ||
52a90c29 BP |
679 | hmap_init(&ofproto->vlandev_map); |
680 | hmap_init(&ofproto->realdev_vid_map); | |
681 | ||
b44a10b7 BP |
682 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
683 | hash_string(ofproto->up.name, 0)); | |
684 | ||
9cdaaebe | 685 | *n_tablesp = N_TABLES; |
6527c598 | 686 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
abe529af BP |
687 | return 0; |
688 | } | |
689 | ||
7ee20df1 BP |
690 | static void |
691 | complete_operations(struct ofproto_dpif *ofproto) | |
692 | { | |
693 | struct dpif_completion *c, *next; | |
694 | ||
695 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
696 | ofoperation_complete(c->op, 0); | |
697 | list_remove(&c->list_node); | |
698 | free(c); | |
699 | } | |
700 | } | |
701 | ||
abe529af BP |
702 | static void |
703 | destruct(struct ofproto *ofproto_) | |
704 | { | |
705 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 706 | struct rule_dpif *rule, *next_rule; |
0697b5c3 | 707 | struct classifier *table; |
abe529af BP |
708 | int i; |
709 | ||
b44a10b7 | 710 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
711 | complete_operations(ofproto); |
712 | ||
0697b5c3 BP |
713 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
714 | struct cls_cursor cursor; | |
715 | ||
716 | cls_cursor_init(&cursor, table, NULL); | |
717 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
718 | ofproto_rule_destroy(&rule->up); | |
719 | } | |
7ee20df1 BP |
720 | } |
721 | ||
abe529af BP |
722 | for (i = 0; i < MAX_MIRRORS; i++) { |
723 | mirror_destroy(ofproto->mirrors[i]); | |
724 | } | |
725 | ||
726 | netflow_destroy(ofproto->netflow); | |
bae473fe | 727 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
728 | hmap_destroy(&ofproto->bundles); |
729 | mac_learning_destroy(ofproto->ml); | |
730 | ||
731 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 732 | hmap_destroy(&ofproto->subfacets); |
abe529af | 733 | |
52a90c29 BP |
734 | hmap_destroy(&ofproto->vlandev_map); |
735 | hmap_destroy(&ofproto->realdev_vid_map); | |
736 | ||
abe529af BP |
737 | dpif_close(ofproto->dpif); |
738 | } | |
739 | ||
740 | static int | |
5fcc0d00 | 741 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
742 | { |
743 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
9b16c439 | 744 | unsigned int work; |
abe529af | 745 | |
9b16c439 BP |
746 | /* Handle one or more batches of upcalls, until there's nothing left to do |
747 | * or until we do a fixed total amount of work. | |
748 | * | |
749 | * We do work in batches because it can be much cheaper to set up a number | |
750 | * of flows and fire off their patches all at once. We do multiple batches | |
751 | * because in some cases handling a packet can cause another packet to be | |
752 | * queued almost immediately as part of the return flow. Both | |
753 | * optimizations can make major improvements on some benchmarks and | |
754 | * presumably for real traffic as well. */ | |
755 | work = 0; | |
756 | while (work < FLOW_MISS_MAX_BATCH) { | |
757 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 758 | if (retval <= 0) { |
9b16c439 | 759 | return -retval; |
501f8d1f | 760 | } |
5fcc0d00 BP |
761 | work += retval; |
762 | } | |
763 | return 0; | |
764 | } | |
765 | ||
766 | static int | |
767 | run(struct ofproto *ofproto_) | |
768 | { | |
769 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
770 | struct ofport_dpif *ofport; | |
771 | struct ofbundle *bundle; | |
772 | int error; | |
773 | ||
774 | if (!clogged) { | |
775 | complete_operations(ofproto); | |
776 | } | |
777 | dpif_run(ofproto->dpif); | |
778 | ||
779 | error = run_fast(ofproto_); | |
780 | if (error) { | |
781 | return error; | |
abe529af BP |
782 | } |
783 | ||
784 | if (timer_expired(&ofproto->next_expiration)) { | |
785 | int delay = expire(ofproto); | |
786 | timer_set_duration(&ofproto->next_expiration, delay); | |
787 | } | |
788 | ||
789 | if (ofproto->netflow) { | |
6fca1ffb BP |
790 | if (netflow_run(ofproto->netflow)) { |
791 | send_netflow_active_timeouts(ofproto); | |
792 | } | |
abe529af BP |
793 | } |
794 | if (ofproto->sflow) { | |
bae473fe | 795 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
796 | } |
797 | ||
798 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
799 | port_run(ofport); | |
800 | } | |
801 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
802 | bundle_run(bundle); | |
803 | } | |
804 | ||
21f7563c | 805 | stp_run(ofproto); |
1c313b88 BP |
806 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
807 | ||
abe529af BP |
808 | /* Now revalidate if there's anything to do. */ |
809 | if (ofproto->need_revalidate | |
810 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
811 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
812 | bool revalidate_all = ofproto->need_revalidate; | |
813 | struct facet *facet, *next; | |
814 | ||
815 | /* Clear the revalidation flags. */ | |
816 | tag_set_init(&ofproto->revalidate_set); | |
817 | ofproto->need_revalidate = false; | |
818 | ||
819 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
820 | if (revalidate_all | |
821 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 822 | facet_revalidate(facet); |
abe529af BP |
823 | } |
824 | } | |
825 | } | |
826 | ||
6814e51f BP |
827 | /* Check the consistency of a random facet, to aid debugging. */ |
828 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
829 | struct facet *facet; | |
830 | ||
831 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
832 | struct facet, hmap_node); | |
833 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
834 | if (!facet_check_consistency(facet)) { | |
835 | ofproto->need_revalidate = true; | |
836 | } | |
837 | } | |
838 | } | |
839 | ||
abe529af BP |
840 | return 0; |
841 | } | |
842 | ||
843 | static void | |
844 | wait(struct ofproto *ofproto_) | |
845 | { | |
846 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
847 | struct ofport_dpif *ofport; | |
848 | struct ofbundle *bundle; | |
849 | ||
7ee20df1 BP |
850 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
851 | poll_immediate_wake(); | |
852 | } | |
853 | ||
abe529af BP |
854 | dpif_wait(ofproto->dpif); |
855 | dpif_recv_wait(ofproto->dpif); | |
856 | if (ofproto->sflow) { | |
bae473fe | 857 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
858 | } |
859 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
860 | poll_immediate_wake(); | |
861 | } | |
862 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
863 | port_wait(ofport); | |
864 | } | |
865 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
866 | bundle_wait(bundle); | |
867 | } | |
6fca1ffb BP |
868 | if (ofproto->netflow) { |
869 | netflow_wait(ofproto->netflow); | |
870 | } | |
1c313b88 | 871 | mac_learning_wait(ofproto->ml); |
21f7563c | 872 | stp_wait(ofproto); |
abe529af BP |
873 | if (ofproto->need_revalidate) { |
874 | /* Shouldn't happen, but if it does just go around again. */ | |
875 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
876 | poll_immediate_wake(); | |
877 | } else { | |
878 | timer_wait(&ofproto->next_expiration); | |
879 | } | |
880 | } | |
881 | ||
882 | static void | |
883 | flush(struct ofproto *ofproto_) | |
884 | { | |
885 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
886 | struct facet *facet, *next_facet; | |
887 | ||
888 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
889 | /* Mark the facet as not installed so that facet_remove() doesn't | |
890 | * bother trying to uninstall it. There is no point in uninstalling it | |
891 | * individually since we are about to blow away all the facets with | |
892 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
893 | struct subfacet *subfacet; |
894 | ||
895 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
896 | subfacet->installed = false; | |
897 | subfacet->dp_packet_count = 0; | |
898 | subfacet->dp_byte_count = 0; | |
899 | } | |
15baa734 | 900 | facet_remove(facet); |
abe529af BP |
901 | } |
902 | dpif_flow_flush(ofproto->dpif); | |
903 | } | |
904 | ||
6c1491fb BP |
905 | static void |
906 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
907 | bool *arp_match_ip, uint32_t *actions) | |
908 | { | |
909 | *arp_match_ip = true; | |
910 | *actions = ((1u << OFPAT_OUTPUT) | | |
911 | (1u << OFPAT_SET_VLAN_VID) | | |
912 | (1u << OFPAT_SET_VLAN_PCP) | | |
913 | (1u << OFPAT_STRIP_VLAN) | | |
914 | (1u << OFPAT_SET_DL_SRC) | | |
915 | (1u << OFPAT_SET_DL_DST) | | |
916 | (1u << OFPAT_SET_NW_SRC) | | |
917 | (1u << OFPAT_SET_NW_DST) | | |
918 | (1u << OFPAT_SET_NW_TOS) | | |
919 | (1u << OFPAT_SET_TP_SRC) | | |
920 | (1u << OFPAT_SET_TP_DST) | | |
921 | (1u << OFPAT_ENQUEUE)); | |
922 | } | |
923 | ||
924 | static void | |
925 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
926 | { | |
927 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 928 | struct dpif_dp_stats s; |
6c1491fb BP |
929 | |
930 | strcpy(ots->name, "classifier"); | |
931 | ||
932 | dpif_get_dp_stats(ofproto->dpif, &s); | |
933 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
934 | put_32aligned_be64(&ots->matched_count, | |
935 | htonll(s.n_hit + ofproto->n_matches)); | |
936 | } | |
937 | ||
abe529af BP |
938 | static struct ofport * |
939 | port_alloc(void) | |
940 | { | |
941 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
942 | return &port->up; | |
943 | } | |
944 | ||
945 | static void | |
946 | port_dealloc(struct ofport *port_) | |
947 | { | |
948 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
949 | free(port); | |
950 | } | |
951 | ||
952 | static int | |
953 | port_construct(struct ofport *port_) | |
954 | { | |
955 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
956 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
957 | ||
f11c28c4 | 958 | ofproto->need_revalidate = true; |
abe529af BP |
959 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
960 | port->bundle = NULL; | |
961 | port->cfm = NULL; | |
962 | port->tag = tag_create_random(); | |
d5ffa7f2 | 963 | port->may_enable = true; |
21f7563c JP |
964 | port->stp_port = NULL; |
965 | port->stp_state = STP_DISABLED; | |
8b36f51e | 966 | hmap_init(&port->priorities); |
52a90c29 BP |
967 | port->realdev_ofp_port = 0; |
968 | port->vlandev_vid = 0; | |
abe529af BP |
969 | |
970 | if (ofproto->sflow) { | |
392c7182 | 971 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
972 | } |
973 | ||
974 | return 0; | |
975 | } | |
976 | ||
977 | static void | |
978 | port_destruct(struct ofport *port_) | |
979 | { | |
980 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
981 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
982 | ||
f11c28c4 | 983 | ofproto->need_revalidate = true; |
abe529af | 984 | bundle_remove(port_); |
a5610457 | 985 | set_cfm(port_, NULL); |
abe529af | 986 | if (ofproto->sflow) { |
bae473fe | 987 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 988 | } |
8b36f51e EJ |
989 | |
990 | ofport_clear_priorities(port); | |
991 | hmap_destroy(&port->priorities); | |
abe529af BP |
992 | } |
993 | ||
994 | static void | |
995 | port_modified(struct ofport *port_) | |
996 | { | |
997 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
998 | ||
999 | if (port->bundle && port->bundle->bond) { | |
1000 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | static void | |
1005 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
1006 | { | |
1007 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1008 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1009 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
1010 | ||
1011 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
1012 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
1013 | ofproto->need_revalidate = true; | |
7bde8dd8 JP |
1014 | |
1015 | if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) { | |
1016 | bundle_update(port->bundle); | |
1017 | } | |
abe529af BP |
1018 | } |
1019 | } | |
1020 | ||
1021 | static int | |
1022 | set_sflow(struct ofproto *ofproto_, | |
1023 | const struct ofproto_sflow_options *sflow_options) | |
1024 | { | |
1025 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1026 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1027 | |
abe529af | 1028 | if (sflow_options) { |
bae473fe | 1029 | if (!ds) { |
abe529af BP |
1030 | struct ofport_dpif *ofport; |
1031 | ||
bae473fe | 1032 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1033 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1034 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1035 | } |
6ff686f2 | 1036 | ofproto->need_revalidate = true; |
abe529af | 1037 | } |
bae473fe | 1038 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1039 | } else { |
6ff686f2 PS |
1040 | if (ds) { |
1041 | dpif_sflow_destroy(ds); | |
1042 | ofproto->need_revalidate = true; | |
1043 | ofproto->sflow = NULL; | |
1044 | } | |
abe529af BP |
1045 | } |
1046 | return 0; | |
1047 | } | |
1048 | ||
1049 | static int | |
a5610457 | 1050 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1051 | { |
1052 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1053 | int error; | |
1054 | ||
a5610457 | 1055 | if (!s) { |
abe529af BP |
1056 | error = 0; |
1057 | } else { | |
1058 | if (!ofport->cfm) { | |
8c977421 EJ |
1059 | struct ofproto_dpif *ofproto; |
1060 | ||
1061 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1062 | ofproto->need_revalidate = true; | |
6f629657 | 1063 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1064 | } |
1065 | ||
a5610457 | 1066 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1067 | return 0; |
1068 | } | |
1069 | ||
1070 | error = EINVAL; | |
1071 | } | |
1072 | cfm_destroy(ofport->cfm); | |
1073 | ofport->cfm = NULL; | |
1074 | return error; | |
1075 | } | |
1076 | ||
1077 | static int | |
a5610457 | 1078 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1079 | { |
1080 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1081 | |
1082 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1083 | } |
1de11730 EJ |
1084 | |
1085 | static int | |
1086 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1087 | size_t *n_rmps) | |
1088 | { | |
1089 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1090 | ||
1091 | if (ofport->cfm) { | |
1092 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1093 | return 0; | |
1094 | } else { | |
1095 | return -1; | |
1096 | } | |
1097 | } | |
abe529af | 1098 | \f |
21f7563c JP |
1099 | /* Spanning Tree. */ |
1100 | ||
1101 | static void | |
1102 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1103 | { | |
1104 | struct ofproto_dpif *ofproto = ofproto_; | |
1105 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1106 | struct ofport_dpif *ofport; | |
1107 | ||
1108 | ofport = stp_port_get_aux(sp); | |
1109 | if (!ofport) { | |
1110 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1111 | ofproto->up.name, port_num); | |
1112 | } else { | |
1113 | struct eth_header *eth = pkt->l2; | |
1114 | ||
1115 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1116 | if (eth_addr_is_zero(eth->eth_src)) { | |
1117 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1118 | "with unknown MAC", ofproto->up.name, port_num); | |
1119 | } else { | |
97d6520b | 1120 | send_packet(ofport, pkt); |
21f7563c JP |
1121 | } |
1122 | } | |
1123 | ofpbuf_delete(pkt); | |
1124 | } | |
1125 | ||
1126 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1127 | static int | |
1128 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1129 | { | |
1130 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1131 | ||
1132 | /* Only revalidate flows if the configuration changed. */ | |
1133 | if (!s != !ofproto->stp) { | |
1134 | ofproto->need_revalidate = true; | |
1135 | } | |
1136 | ||
1137 | if (s) { | |
1138 | if (!ofproto->stp) { | |
1139 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1140 | send_bpdu_cb, ofproto); | |
1141 | ofproto->stp_last_tick = time_msec(); | |
1142 | } | |
1143 | ||
1144 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1145 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1146 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1147 | stp_set_max_age(ofproto->stp, s->max_age); | |
1148 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1149 | } else { | |
1150 | stp_destroy(ofproto->stp); | |
1151 | ofproto->stp = NULL; | |
1152 | } | |
1153 | ||
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | static int | |
1158 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1159 | { | |
1160 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1161 | ||
1162 | if (ofproto->stp) { | |
1163 | s->enabled = true; | |
1164 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1165 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1166 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1167 | } else { | |
1168 | s->enabled = false; | |
1169 | } | |
1170 | ||
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | static void | |
1175 | update_stp_port_state(struct ofport_dpif *ofport) | |
1176 | { | |
1177 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1178 | enum stp_state state; | |
1179 | ||
1180 | /* Figure out new state. */ | |
1181 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1182 | : STP_DISABLED; | |
1183 | ||
1184 | /* Update state. */ | |
1185 | if (ofport->stp_state != state) { | |
1186 | ovs_be32 of_state; | |
1187 | bool fwd_change; | |
1188 | ||
1189 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1190 | netdev_get_name(ofport->up.netdev), | |
1191 | stp_state_name(ofport->stp_state), | |
1192 | stp_state_name(state)); | |
1193 | if (stp_learn_in_state(ofport->stp_state) | |
1194 | != stp_learn_in_state(state)) { | |
1195 | /* xxx Learning action flows should also be flushed. */ | |
1196 | mac_learning_flush(ofproto->ml); | |
1197 | } | |
1198 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1199 | != stp_forward_in_state(state); | |
1200 | ||
1201 | ofproto->need_revalidate = true; | |
1202 | ofport->stp_state = state; | |
1203 | ofport->stp_state_entered = time_msec(); | |
1204 | ||
b308140a | 1205 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1206 | bundle_update(ofport->bundle); |
1207 | } | |
1208 | ||
1209 | /* Update the STP state bits in the OpenFlow port description. */ | |
1210 | of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK)) | |
1211 | | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN | |
1212 | : state == STP_LEARNING ? OFPPS_STP_LEARN | |
1213 | : state == STP_FORWARDING ? OFPPS_STP_FORWARD | |
1214 | : state == STP_BLOCKING ? OFPPS_STP_BLOCK | |
1215 | : 0); | |
1216 | ofproto_port_set_state(&ofport->up, of_state); | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1221 | * caller is responsible for assigning STP port numbers and ensuring | |
1222 | * there are no duplicates. */ | |
1223 | static int | |
1224 | set_stp_port(struct ofport *ofport_, | |
1225 | const struct ofproto_port_stp_settings *s) | |
1226 | { | |
1227 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1228 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1229 | struct stp_port *sp = ofport->stp_port; | |
1230 | ||
1231 | if (!s || !s->enable) { | |
1232 | if (sp) { | |
1233 | ofport->stp_port = NULL; | |
1234 | stp_port_disable(sp); | |
ecd12731 | 1235 | update_stp_port_state(ofport); |
21f7563c JP |
1236 | } |
1237 | return 0; | |
1238 | } else if (sp && stp_port_no(sp) != s->port_num | |
1239 | && ofport == stp_port_get_aux(sp)) { | |
1240 | /* The port-id changed, so disable the old one if it's not | |
1241 | * already in use by another port. */ | |
1242 | stp_port_disable(sp); | |
1243 | } | |
1244 | ||
1245 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1246 | stp_port_enable(sp); | |
1247 | ||
1248 | stp_port_set_aux(sp, ofport); | |
1249 | stp_port_set_priority(sp, s->priority); | |
1250 | stp_port_set_path_cost(sp, s->path_cost); | |
1251 | ||
1252 | update_stp_port_state(ofport); | |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static int | |
1258 | get_stp_port_status(struct ofport *ofport_, | |
1259 | struct ofproto_port_stp_status *s) | |
1260 | { | |
1261 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1262 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1263 | struct stp_port *sp = ofport->stp_port; | |
1264 | ||
1265 | if (!ofproto->stp || !sp) { | |
1266 | s->enabled = false; | |
1267 | return 0; | |
1268 | } | |
1269 | ||
1270 | s->enabled = true; | |
1271 | s->port_id = stp_port_get_id(sp); | |
1272 | s->state = stp_port_get_state(sp); | |
1273 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1274 | s->role = stp_port_get_role(sp); | |
80740385 | 1275 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1276 | |
1277 | return 0; | |
1278 | } | |
1279 | ||
1280 | static void | |
1281 | stp_run(struct ofproto_dpif *ofproto) | |
1282 | { | |
1283 | if (ofproto->stp) { | |
1284 | long long int now = time_msec(); | |
1285 | long long int elapsed = now - ofproto->stp_last_tick; | |
1286 | struct stp_port *sp; | |
1287 | ||
1288 | if (elapsed > 0) { | |
1289 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1290 | ofproto->stp_last_tick = now; | |
1291 | } | |
1292 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1293 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1294 | ||
1295 | if (ofport) { | |
1296 | update_stp_port_state(ofport); | |
1297 | } | |
1298 | } | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | static void | |
1303 | stp_wait(struct ofproto_dpif *ofproto) | |
1304 | { | |
1305 | if (ofproto->stp) { | |
1306 | poll_timer_wait(1000); | |
1307 | } | |
1308 | } | |
1309 | ||
1310 | /* Returns true if STP should process 'flow'. */ | |
1311 | static bool | |
1312 | stp_should_process_flow(const struct flow *flow) | |
1313 | { | |
1314 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1315 | } | |
1316 | ||
1317 | static void | |
1318 | stp_process_packet(const struct ofport_dpif *ofport, | |
1319 | const struct ofpbuf *packet) | |
1320 | { | |
1321 | struct ofpbuf payload = *packet; | |
1322 | struct eth_header *eth = payload.data; | |
1323 | struct stp_port *sp = ofport->stp_port; | |
1324 | ||
1325 | /* Sink packets on ports that have STP disabled when the bridge has | |
1326 | * STP enabled. */ | |
1327 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1328 | return; | |
1329 | } | |
1330 | ||
1331 | /* Trim off padding on payload. */ | |
c573540b BP |
1332 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1333 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1334 | } |
1335 | ||
1336 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1337 | stp_received_bpdu(sp, payload.data, payload.size); | |
1338 | } | |
1339 | } | |
1340 | \f | |
8b36f51e EJ |
1341 | static struct priority_to_dscp * |
1342 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1343 | { | |
1344 | struct priority_to_dscp *pdscp; | |
1345 | uint32_t hash; | |
1346 | ||
1347 | hash = hash_int(priority, 0); | |
1348 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1349 | if (pdscp->priority == priority) { | |
1350 | return pdscp; | |
1351 | } | |
1352 | } | |
1353 | return NULL; | |
1354 | } | |
1355 | ||
1356 | static void | |
1357 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1358 | { | |
1359 | struct priority_to_dscp *pdscp, *next; | |
1360 | ||
1361 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1362 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1363 | free(pdscp); | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | static int | |
1368 | set_queues(struct ofport *ofport_, | |
1369 | const struct ofproto_port_queue *qdscp_list, | |
1370 | size_t n_qdscp) | |
1371 | { | |
1372 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1373 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1374 | struct hmap new = HMAP_INITIALIZER(&new); | |
1375 | size_t i; | |
1376 | ||
1377 | for (i = 0; i < n_qdscp; i++) { | |
1378 | struct priority_to_dscp *pdscp; | |
1379 | uint32_t priority; | |
1380 | uint8_t dscp; | |
1381 | ||
1382 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1383 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1384 | &priority)) { | |
1385 | continue; | |
1386 | } | |
1387 | ||
1388 | pdscp = get_priority(ofport, priority); | |
1389 | if (pdscp) { | |
1390 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1391 | } else { | |
1392 | pdscp = xmalloc(sizeof *pdscp); | |
1393 | pdscp->priority = priority; | |
1394 | pdscp->dscp = dscp; | |
1395 | ofproto->need_revalidate = true; | |
1396 | } | |
1397 | ||
1398 | if (pdscp->dscp != dscp) { | |
1399 | pdscp->dscp = dscp; | |
1400 | ofproto->need_revalidate = true; | |
1401 | } | |
1402 | ||
1403 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1404 | } | |
1405 | ||
1406 | if (!hmap_is_empty(&ofport->priorities)) { | |
1407 | ofport_clear_priorities(ofport); | |
1408 | ofproto->need_revalidate = true; | |
1409 | } | |
1410 | ||
1411 | hmap_swap(&new, &ofport->priorities); | |
1412 | hmap_destroy(&new); | |
1413 | ||
1414 | return 0; | |
1415 | } | |
1416 | \f | |
abe529af BP |
1417 | /* Bundles. */ |
1418 | ||
b44a10b7 BP |
1419 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1420 | * ofproto to revalidate every flow. | |
1421 | * | |
1422 | * Normally MAC learning entries are removed only from the ofproto associated | |
1423 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1424 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1425 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1426 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1427 | * with the host from which it migrated. */ | |
abe529af | 1428 | static void |
b44a10b7 | 1429 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1430 | { |
1431 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1432 | struct mac_learning *ml = ofproto->ml; | |
1433 | struct mac_entry *mac, *next_mac; | |
1434 | ||
1435 | ofproto->need_revalidate = true; | |
1436 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1437 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1438 | if (all_ofprotos) { |
1439 | struct ofproto_dpif *o; | |
1440 | ||
1441 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1442 | if (o != ofproto) { | |
1443 | struct mac_entry *e; | |
1444 | ||
1445 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1446 | NULL); | |
1447 | if (e) { | |
1448 | tag_set_add(&o->revalidate_set, e->tag); | |
1449 | mac_learning_expire(o->ml, e); | |
1450 | } | |
1451 | } | |
1452 | } | |
1453 | } | |
1454 | ||
abe529af BP |
1455 | mac_learning_expire(ml, mac); |
1456 | } | |
1457 | } | |
1458 | } | |
1459 | ||
1460 | static struct ofbundle * | |
1461 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1462 | { | |
1463 | struct ofbundle *bundle; | |
1464 | ||
1465 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1466 | &ofproto->bundles) { | |
1467 | if (bundle->aux == aux) { | |
1468 | return bundle; | |
1469 | } | |
1470 | } | |
1471 | return NULL; | |
1472 | } | |
1473 | ||
1474 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1475 | * ones that are found to 'bundles'. */ | |
1476 | static void | |
1477 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1478 | void **auxes, size_t n_auxes, | |
1479 | struct hmapx *bundles) | |
1480 | { | |
1481 | size_t i; | |
1482 | ||
1483 | hmapx_init(bundles); | |
1484 | for (i = 0; i < n_auxes; i++) { | |
1485 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1486 | if (bundle) { | |
1487 | hmapx_add(bundles, bundle); | |
1488 | } | |
1489 | } | |
1490 | } | |
1491 | ||
7bde8dd8 JP |
1492 | static void |
1493 | bundle_update(struct ofbundle *bundle) | |
1494 | { | |
1495 | struct ofport_dpif *port; | |
1496 | ||
1497 | bundle->floodable = true; | |
1498 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
5e48dc2b | 1499 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { |
7bde8dd8 JP |
1500 | bundle->floodable = false; |
1501 | break; | |
1502 | } | |
1503 | } | |
1504 | } | |
1505 | ||
abe529af BP |
1506 | static void |
1507 | bundle_del_port(struct ofport_dpif *port) | |
1508 | { | |
1509 | struct ofbundle *bundle = port->bundle; | |
1510 | ||
6f77f4ae BP |
1511 | bundle->ofproto->need_revalidate = true; |
1512 | ||
abe529af BP |
1513 | list_remove(&port->bundle_node); |
1514 | port->bundle = NULL; | |
1515 | ||
1516 | if (bundle->lacp) { | |
1517 | lacp_slave_unregister(bundle->lacp, port); | |
1518 | } | |
1519 | if (bundle->bond) { | |
1520 | bond_slave_unregister(bundle->bond, port); | |
1521 | } | |
1522 | ||
7bde8dd8 | 1523 | bundle_update(bundle); |
abe529af BP |
1524 | } |
1525 | ||
1526 | static bool | |
1527 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1528 | struct lacp_slave_settings *lacp, |
1529 | uint32_t bond_stable_id) | |
abe529af BP |
1530 | { |
1531 | struct ofport_dpif *port; | |
1532 | ||
1533 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1534 | if (!port) { | |
1535 | return false; | |
1536 | } | |
1537 | ||
1538 | if (port->bundle != bundle) { | |
6f77f4ae | 1539 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1540 | if (port->bundle) { |
1541 | bundle_del_port(port); | |
1542 | } | |
1543 | ||
1544 | port->bundle = bundle; | |
1545 | list_push_back(&bundle->ports, &port->bundle_node); | |
5e48dc2b | 1546 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { |
abe529af BP |
1547 | bundle->floodable = false; |
1548 | } | |
1549 | } | |
1550 | if (lacp) { | |
4a86aece | 1551 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1552 | lacp_slave_register(bundle->lacp, port, lacp); |
1553 | } | |
1554 | ||
00794817 BP |
1555 | port->bond_stable_id = bond_stable_id; |
1556 | ||
abe529af BP |
1557 | return true; |
1558 | } | |
1559 | ||
1560 | static void | |
1561 | bundle_destroy(struct ofbundle *bundle) | |
1562 | { | |
1563 | struct ofproto_dpif *ofproto; | |
1564 | struct ofport_dpif *port, *next_port; | |
1565 | int i; | |
1566 | ||
1567 | if (!bundle) { | |
1568 | return; | |
1569 | } | |
1570 | ||
1571 | ofproto = bundle->ofproto; | |
1572 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1573 | struct ofmirror *m = ofproto->mirrors[i]; | |
1574 | if (m) { | |
1575 | if (m->out == bundle) { | |
1576 | mirror_destroy(m); | |
1577 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1578 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1579 | ofproto->need_revalidate = true; | |
1580 | } | |
1581 | } | |
1582 | } | |
1583 | ||
1584 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1585 | bundle_del_port(port); | |
1586 | } | |
1587 | ||
b44a10b7 | 1588 | bundle_flush_macs(bundle, true); |
abe529af BP |
1589 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1590 | free(bundle->name); | |
1591 | free(bundle->trunks); | |
1592 | lacp_destroy(bundle->lacp); | |
1593 | bond_destroy(bundle->bond); | |
1594 | free(bundle); | |
1595 | } | |
1596 | ||
1597 | static int | |
1598 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1599 | const struct ofproto_bundle_settings *s) | |
1600 | { | |
1601 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1602 | bool need_flush = false; | |
abe529af BP |
1603 | struct ofport_dpif *port; |
1604 | struct ofbundle *bundle; | |
ecac4ebf BP |
1605 | unsigned long *trunks; |
1606 | int vlan; | |
abe529af BP |
1607 | size_t i; |
1608 | bool ok; | |
1609 | ||
1610 | if (!s) { | |
1611 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1612 | return 0; | |
1613 | } | |
1614 | ||
1615 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1616 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1617 | ||
1618 | bundle = bundle_lookup(ofproto, aux); | |
1619 | if (!bundle) { | |
1620 | bundle = xmalloc(sizeof *bundle); | |
1621 | ||
1622 | bundle->ofproto = ofproto; | |
1623 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1624 | hash_pointer(aux, 0)); | |
1625 | bundle->aux = aux; | |
1626 | bundle->name = NULL; | |
1627 | ||
1628 | list_init(&bundle->ports); | |
ecac4ebf | 1629 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1630 | bundle->vlan = -1; |
1631 | bundle->trunks = NULL; | |
5e9ceccd | 1632 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1633 | bundle->lacp = NULL; |
1634 | bundle->bond = NULL; | |
1635 | ||
1636 | bundle->floodable = true; | |
1637 | ||
1638 | bundle->src_mirrors = 0; | |
1639 | bundle->dst_mirrors = 0; | |
1640 | bundle->mirror_out = 0; | |
1641 | } | |
1642 | ||
1643 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1644 | free(bundle->name); | |
1645 | bundle->name = xstrdup(s->name); | |
1646 | } | |
1647 | ||
1648 | /* LACP. */ | |
1649 | if (s->lacp) { | |
1650 | if (!bundle->lacp) { | |
8c977421 | 1651 | ofproto->need_revalidate = true; |
abe529af BP |
1652 | bundle->lacp = lacp_create(); |
1653 | } | |
1654 | lacp_configure(bundle->lacp, s->lacp); | |
1655 | } else { | |
1656 | lacp_destroy(bundle->lacp); | |
1657 | bundle->lacp = NULL; | |
1658 | } | |
1659 | ||
1660 | /* Update set of ports. */ | |
1661 | ok = true; | |
1662 | for (i = 0; i < s->n_slaves; i++) { | |
1663 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1664 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1665 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1666 | ok = false; |
1667 | } | |
1668 | } | |
1669 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1670 | struct ofport_dpif *next_port; | |
1671 | ||
1672 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1673 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1674 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1675 | goto found; |
1676 | } | |
1677 | } | |
1678 | ||
1679 | bundle_del_port(port); | |
1680 | found: ; | |
1681 | } | |
1682 | } | |
1683 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1684 | ||
1685 | if (list_is_empty(&bundle->ports)) { | |
1686 | bundle_destroy(bundle); | |
1687 | return EINVAL; | |
1688 | } | |
1689 | ||
ecac4ebf | 1690 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1691 | if (s->vlan_mode != bundle->vlan_mode |
1692 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1693 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1694 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1695 | need_flush = true; |
1696 | } | |
1697 | ||
abe529af | 1698 | /* Set VLAN tag. */ |
ecac4ebf BP |
1699 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1700 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1701 | : 0); | |
1702 | if (vlan != bundle->vlan) { | |
1703 | bundle->vlan = vlan; | |
abe529af BP |
1704 | need_flush = true; |
1705 | } | |
1706 | ||
1707 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1708 | switch (s->vlan_mode) { |
1709 | case PORT_VLAN_ACCESS: | |
1710 | trunks = NULL; | |
1711 | break; | |
1712 | ||
1713 | case PORT_VLAN_TRUNK: | |
1714 | trunks = (unsigned long *) s->trunks; | |
1715 | break; | |
1716 | ||
1717 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1718 | case PORT_VLAN_NATIVE_TAGGED: | |
1719 | if (vlan != 0 && (!s->trunks | |
1720 | || !bitmap_is_set(s->trunks, vlan) | |
1721 | || bitmap_is_set(s->trunks, 0))) { | |
1722 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1723 | if (s->trunks) { | |
1724 | trunks = bitmap_clone(s->trunks, 4096); | |
1725 | } else { | |
1726 | trunks = bitmap_allocate1(4096); | |
1727 | } | |
1728 | bitmap_set1(trunks, vlan); | |
1729 | bitmap_set0(trunks, 0); | |
1730 | } else { | |
1731 | trunks = (unsigned long *) s->trunks; | |
1732 | } | |
1733 | break; | |
1734 | ||
1735 | default: | |
1736 | NOT_REACHED(); | |
1737 | } | |
abe529af BP |
1738 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1739 | free(bundle->trunks); | |
ecac4ebf BP |
1740 | if (trunks == s->trunks) { |
1741 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1742 | } else { | |
1743 | bundle->trunks = trunks; | |
1744 | trunks = NULL; | |
1745 | } | |
abe529af BP |
1746 | need_flush = true; |
1747 | } | |
ecac4ebf BP |
1748 | if (trunks != s->trunks) { |
1749 | free(trunks); | |
1750 | } | |
abe529af BP |
1751 | |
1752 | /* Bonding. */ | |
1753 | if (!list_is_short(&bundle->ports)) { | |
1754 | bundle->ofproto->has_bonded_bundles = true; | |
1755 | if (bundle->bond) { | |
1756 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1757 | ofproto->need_revalidate = true; | |
1758 | } | |
1759 | } else { | |
1760 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1761 | ofproto->need_revalidate = true; |
abe529af BP |
1762 | } |
1763 | ||
1764 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1765 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1766 | port->up.netdev); |
1767 | } | |
1768 | } else { | |
1769 | bond_destroy(bundle->bond); | |
1770 | bundle->bond = NULL; | |
1771 | } | |
1772 | ||
1773 | /* If we changed something that would affect MAC learning, un-learn | |
1774 | * everything on this port and force flow revalidation. */ | |
1775 | if (need_flush) { | |
b44a10b7 | 1776 | bundle_flush_macs(bundle, false); |
abe529af BP |
1777 | } |
1778 | ||
1779 | return 0; | |
1780 | } | |
1781 | ||
1782 | static void | |
1783 | bundle_remove(struct ofport *port_) | |
1784 | { | |
1785 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1786 | struct ofbundle *bundle = port->bundle; | |
1787 | ||
1788 | if (bundle) { | |
1789 | bundle_del_port(port); | |
1790 | if (list_is_empty(&bundle->ports)) { | |
1791 | bundle_destroy(bundle); | |
1792 | } else if (list_is_short(&bundle->ports)) { | |
1793 | bond_destroy(bundle->bond); | |
1794 | bundle->bond = NULL; | |
1795 | } | |
1796 | } | |
1797 | } | |
1798 | ||
1799 | static void | |
5f877369 | 1800 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1801 | { |
1802 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1803 | struct ofport_dpif *port = port_; | |
1804 | uint8_t ea[ETH_ADDR_LEN]; | |
1805 | int error; | |
1806 | ||
1807 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1808 | if (!error) { | |
abe529af | 1809 | struct ofpbuf packet; |
5f877369 | 1810 | void *packet_pdu; |
abe529af BP |
1811 | |
1812 | ofpbuf_init(&packet, 0); | |
1813 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1814 | pdu_size); |
1815 | memcpy(packet_pdu, pdu, pdu_size); | |
1816 | ||
97d6520b | 1817 | send_packet(port, &packet); |
abe529af BP |
1818 | ofpbuf_uninit(&packet); |
1819 | } else { | |
1820 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1821 | "%s (%s)", port->bundle->name, | |
1822 | netdev_get_name(port->up.netdev), strerror(error)); | |
1823 | } | |
1824 | } | |
1825 | ||
1826 | static void | |
1827 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1828 | { | |
1829 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1830 | int error, n_packets, n_errors; | |
1831 | struct mac_entry *e; | |
1832 | ||
1833 | error = n_packets = n_errors = 0; | |
1834 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1835 | if (e->port.p != bundle) { | |
ea131871 JG |
1836 | struct ofpbuf *learning_packet; |
1837 | struct ofport_dpif *port; | |
4dd1e3ca | 1838 | void *port_void; |
ea131871 JG |
1839 | int ret; |
1840 | ||
4dd1e3ca BP |
1841 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
1842 | * struct ofport_dpif more effective. */ | |
1843 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
1844 | e->mac, e->vlan, | |
1845 | &port_void); | |
1846 | port = port_void; | |
97d6520b | 1847 | ret = send_packet(port, learning_packet); |
ea131871 | 1848 | ofpbuf_delete(learning_packet); |
abe529af BP |
1849 | if (ret) { |
1850 | error = ret; | |
1851 | n_errors++; | |
1852 | } | |
1853 | n_packets++; | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | if (n_errors) { | |
1858 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1859 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1860 | "packets, last error was: %s", | |
1861 | bundle->name, n_errors, n_packets, strerror(error)); | |
1862 | } else { | |
1863 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1864 | bundle->name, n_packets); | |
1865 | } | |
1866 | } | |
1867 | ||
1868 | static void | |
1869 | bundle_run(struct ofbundle *bundle) | |
1870 | { | |
1871 | if (bundle->lacp) { | |
1872 | lacp_run(bundle->lacp, send_pdu_cb); | |
1873 | } | |
1874 | if (bundle->bond) { | |
1875 | struct ofport_dpif *port; | |
1876 | ||
1877 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1878 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1879 | } |
1880 | ||
1881 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
1882 | lacp_negotiated(bundle->lacp)); | |
1883 | if (bond_should_send_learning_packets(bundle->bond)) { | |
1884 | bundle_send_learning_packets(bundle); | |
1885 | } | |
1886 | } | |
1887 | } | |
1888 | ||
1889 | static void | |
1890 | bundle_wait(struct ofbundle *bundle) | |
1891 | { | |
1892 | if (bundle->lacp) { | |
1893 | lacp_wait(bundle->lacp); | |
1894 | } | |
1895 | if (bundle->bond) { | |
1896 | bond_wait(bundle->bond); | |
1897 | } | |
1898 | } | |
1899 | \f | |
1900 | /* Mirrors. */ | |
1901 | ||
1902 | static int | |
1903 | mirror_scan(struct ofproto_dpif *ofproto) | |
1904 | { | |
1905 | int idx; | |
1906 | ||
1907 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1908 | if (!ofproto->mirrors[idx]) { | |
1909 | return idx; | |
1910 | } | |
1911 | } | |
1912 | return -1; | |
1913 | } | |
1914 | ||
1915 | static struct ofmirror * | |
1916 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1917 | { | |
1918 | int i; | |
1919 | ||
1920 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1921 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1922 | if (mirror && mirror->aux == aux) { | |
1923 | return mirror; | |
1924 | } | |
1925 | } | |
1926 | ||
1927 | return NULL; | |
1928 | } | |
1929 | ||
9ba15e2a BP |
1930 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
1931 | static void | |
1932 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
1933 | { | |
1934 | int i; | |
1935 | ||
1936 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1937 | struct ofmirror *m = ofproto->mirrors[i]; | |
1938 | ||
1939 | if (m) { | |
1940 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
1941 | } | |
1942 | } | |
1943 | ||
1944 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1945 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
1946 | int j; | |
1947 | ||
1948 | if (!m1) { | |
1949 | continue; | |
1950 | } | |
1951 | ||
1952 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
1953 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
1954 | ||
edb0540b | 1955 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
1956 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
1957 | m2->dup_mirrors |= m1->dup_mirrors; | |
1958 | } | |
1959 | } | |
1960 | } | |
1961 | } | |
1962 | ||
abe529af BP |
1963 | static int |
1964 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1965 | const struct ofproto_mirror_settings *s) | |
1966 | { | |
1967 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1968 | mirror_mask_t mirror_bit; | |
1969 | struct ofbundle *bundle; | |
1970 | struct ofmirror *mirror; | |
1971 | struct ofbundle *out; | |
1972 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1973 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1974 | int out_vlan; | |
1975 | ||
1976 | mirror = mirror_lookup(ofproto, aux); | |
1977 | if (!s) { | |
1978 | mirror_destroy(mirror); | |
1979 | return 0; | |
1980 | } | |
1981 | if (!mirror) { | |
1982 | int idx; | |
1983 | ||
1984 | idx = mirror_scan(ofproto); | |
1985 | if (idx < 0) { | |
1986 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1987 | "cannot create %s", | |
1988 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1989 | return EFBIG; | |
1990 | } | |
1991 | ||
1992 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1993 | mirror->ofproto = ofproto; | |
1994 | mirror->idx = idx; | |
8b28d864 | 1995 | mirror->aux = aux; |
abe529af BP |
1996 | mirror->out_vlan = -1; |
1997 | mirror->name = NULL; | |
1998 | } | |
1999 | ||
2000 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2001 | free(mirror->name); | |
2002 | mirror->name = xstrdup(s->name); | |
2003 | } | |
2004 | ||
2005 | /* Get the new configuration. */ | |
2006 | if (s->out_bundle) { | |
2007 | out = bundle_lookup(ofproto, s->out_bundle); | |
2008 | if (!out) { | |
2009 | mirror_destroy(mirror); | |
2010 | return EINVAL; | |
2011 | } | |
2012 | out_vlan = -1; | |
2013 | } else { | |
2014 | out = NULL; | |
2015 | out_vlan = s->out_vlan; | |
2016 | } | |
2017 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2018 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2019 | ||
2020 | /* If the configuration has not changed, do nothing. */ | |
2021 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2022 | && hmapx_equals(&dsts, &mirror->dsts) | |
2023 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2024 | && mirror->out == out | |
2025 | && mirror->out_vlan == out_vlan) | |
2026 | { | |
2027 | hmapx_destroy(&srcs); | |
2028 | hmapx_destroy(&dsts); | |
2029 | return 0; | |
2030 | } | |
2031 | ||
2032 | hmapx_swap(&srcs, &mirror->srcs); | |
2033 | hmapx_destroy(&srcs); | |
2034 | ||
2035 | hmapx_swap(&dsts, &mirror->dsts); | |
2036 | hmapx_destroy(&dsts); | |
2037 | ||
2038 | free(mirror->vlans); | |
2039 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2040 | ||
2041 | mirror->out = out; | |
2042 | mirror->out_vlan = out_vlan; | |
2043 | ||
2044 | /* Update bundles. */ | |
2045 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2046 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2047 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2048 | bundle->src_mirrors |= mirror_bit; | |
2049 | } else { | |
2050 | bundle->src_mirrors &= ~mirror_bit; | |
2051 | } | |
2052 | ||
2053 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2054 | bundle->dst_mirrors |= mirror_bit; | |
2055 | } else { | |
2056 | bundle->dst_mirrors &= ~mirror_bit; | |
2057 | } | |
2058 | ||
2059 | if (mirror->out == bundle) { | |
2060 | bundle->mirror_out |= mirror_bit; | |
2061 | } else { | |
2062 | bundle->mirror_out &= ~mirror_bit; | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | ofproto->need_revalidate = true; | |
2067 | mac_learning_flush(ofproto->ml); | |
9ba15e2a | 2068 | mirror_update_dups(ofproto); |
abe529af BP |
2069 | |
2070 | return 0; | |
2071 | } | |
2072 | ||
2073 | static void | |
2074 | mirror_destroy(struct ofmirror *mirror) | |
2075 | { | |
2076 | struct ofproto_dpif *ofproto; | |
2077 | mirror_mask_t mirror_bit; | |
2078 | struct ofbundle *bundle; | |
2079 | ||
2080 | if (!mirror) { | |
2081 | return; | |
2082 | } | |
2083 | ||
2084 | ofproto = mirror->ofproto; | |
2085 | ofproto->need_revalidate = true; | |
2086 | mac_learning_flush(ofproto->ml); | |
2087 | ||
2088 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2089 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2090 | bundle->src_mirrors &= ~mirror_bit; | |
2091 | bundle->dst_mirrors &= ~mirror_bit; | |
2092 | bundle->mirror_out &= ~mirror_bit; | |
2093 | } | |
2094 | ||
2095 | hmapx_destroy(&mirror->srcs); | |
2096 | hmapx_destroy(&mirror->dsts); | |
2097 | free(mirror->vlans); | |
2098 | ||
2099 | ofproto->mirrors[mirror->idx] = NULL; | |
2100 | free(mirror->name); | |
2101 | free(mirror); | |
9ba15e2a BP |
2102 | |
2103 | mirror_update_dups(ofproto); | |
abe529af BP |
2104 | } |
2105 | ||
9d24de3b JP |
2106 | static int |
2107 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2108 | uint64_t *packets, uint64_t *bytes) | |
2109 | { | |
2110 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2111 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2112 | ||
2113 | if (!mirror) { | |
2114 | *packets = *bytes = UINT64_MAX; | |
2115 | return 0; | |
2116 | } | |
2117 | ||
2118 | *packets = mirror->packet_count; | |
2119 | *bytes = mirror->byte_count; | |
2120 | ||
2121 | return 0; | |
2122 | } | |
2123 | ||
abe529af BP |
2124 | static int |
2125 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2126 | { | |
2127 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2128 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2129 | ofproto->need_revalidate = true; | |
2130 | mac_learning_flush(ofproto->ml); | |
2131 | } | |
2132 | return 0; | |
2133 | } | |
2134 | ||
2135 | static bool | |
b4affc74 | 2136 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2137 | { |
2138 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2139 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2140 | return bundle && bundle->mirror_out != 0; | |
2141 | } | |
8402c74b SS |
2142 | |
2143 | static void | |
b53055f4 | 2144 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2145 | { |
2146 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2147 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
2148 | ofproto->need_revalidate = true; | |
2149 | } | |
abe529af BP |
2150 | \f |
2151 | /* Ports. */ | |
2152 | ||
2153 | static struct ofport_dpif * | |
2154 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
2155 | { | |
7df6a8bd BP |
2156 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2157 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2158 | } |
2159 | ||
2160 | static struct ofport_dpif * | |
2161 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
2162 | { | |
2163 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2164 | } | |
2165 | ||
2166 | static void | |
2167 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2168 | struct dpif_port *dpif_port) | |
2169 | { | |
2170 | ofproto_port->name = dpif_port->name; | |
2171 | ofproto_port->type = dpif_port->type; | |
2172 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2173 | } | |
2174 | ||
2175 | static void | |
2176 | port_run(struct ofport_dpif *ofport) | |
2177 | { | |
015e08bc EJ |
2178 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2179 | ||
abe529af BP |
2180 | if (ofport->cfm) { |
2181 | cfm_run(ofport->cfm); | |
2182 | ||
2183 | if (cfm_should_send_ccm(ofport->cfm)) { | |
2184 | struct ofpbuf packet; | |
abe529af BP |
2185 | |
2186 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 2187 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
97d6520b | 2188 | send_packet(ofport, &packet); |
abe529af BP |
2189 | ofpbuf_uninit(&packet); |
2190 | } | |
015e08bc | 2191 | |
86dc6501 EJ |
2192 | enable = enable && !cfm_get_fault(ofport->cfm) |
2193 | && cfm_get_opup(ofport->cfm); | |
abe529af | 2194 | } |
015e08bc EJ |
2195 | |
2196 | if (ofport->bundle) { | |
2197 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
2198 | } | |
2199 | ||
daff3353 EJ |
2200 | if (ofport->may_enable != enable) { |
2201 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2202 | ||
2203 | if (ofproto->has_bundle_action) { | |
2204 | ofproto->need_revalidate = true; | |
2205 | } | |
2206 | } | |
2207 | ||
015e08bc | 2208 | ofport->may_enable = enable; |
abe529af BP |
2209 | } |
2210 | ||
2211 | static void | |
2212 | port_wait(struct ofport_dpif *ofport) | |
2213 | { | |
2214 | if (ofport->cfm) { | |
2215 | cfm_wait(ofport->cfm); | |
2216 | } | |
2217 | } | |
2218 | ||
2219 | static int | |
2220 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2221 | struct ofproto_port *ofproto_port) | |
2222 | { | |
2223 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2224 | struct dpif_port dpif_port; | |
2225 | int error; | |
2226 | ||
2227 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2228 | if (!error) { | |
2229 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2230 | } | |
2231 | return error; | |
2232 | } | |
2233 | ||
2234 | static int | |
2235 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2236 | { | |
2237 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2238 | uint16_t odp_port; | |
2239 | int error; | |
2240 | ||
2241 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2242 | if (!error) { | |
2243 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2244 | } | |
2245 | return error; | |
2246 | } | |
2247 | ||
2248 | static int | |
2249 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2250 | { | |
2251 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2252 | int error; | |
2253 | ||
2254 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2255 | if (!error) { | |
2256 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2257 | if (ofport) { | |
2258 | /* The caller is going to close ofport->up.netdev. If this is a | |
2259 | * bonded port, then the bond is using that netdev, so remove it | |
2260 | * from the bond. The client will need to reconfigure everything | |
2261 | * after deleting ports, so then the slave will get re-added. */ | |
2262 | bundle_remove(&ofport->up); | |
2263 | } | |
2264 | } | |
2265 | return error; | |
2266 | } | |
2267 | ||
6527c598 PS |
2268 | static int |
2269 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2270 | { | |
2271 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2272 | int error; | |
2273 | ||
2274 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2275 | ||
2276 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2277 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2278 | ||
2279 | /* ofproto->stats.tx_packets represents packets that we created | |
2280 | * internally and sent to some port (e.g. packets sent with | |
2281 | * send_packet()). Account for them as if they had come from | |
2282 | * OFPP_LOCAL and got forwarded. */ | |
2283 | ||
2284 | if (stats->rx_packets != UINT64_MAX) { | |
2285 | stats->rx_packets += ofproto->stats.tx_packets; | |
2286 | } | |
2287 | ||
2288 | if (stats->rx_bytes != UINT64_MAX) { | |
2289 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2290 | } | |
2291 | ||
2292 | /* ofproto->stats.rx_packets represents packets that were received on | |
2293 | * some port and we processed internally and dropped (e.g. STP). | |
2294 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2295 | ||
2296 | if (stats->tx_packets != UINT64_MAX) { | |
2297 | stats->tx_packets += ofproto->stats.rx_packets; | |
2298 | } | |
2299 | ||
2300 | if (stats->tx_bytes != UINT64_MAX) { | |
2301 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2302 | } | |
2303 | } | |
2304 | ||
2305 | return error; | |
2306 | } | |
2307 | ||
2308 | /* Account packets for LOCAL port. */ | |
2309 | static void | |
2310 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2311 | size_t tx_size, size_t rx_size) | |
2312 | { | |
2313 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2314 | ||
2315 | if (rx_size) { | |
2316 | ofproto->stats.rx_packets++; | |
2317 | ofproto->stats.rx_bytes += rx_size; | |
2318 | } | |
2319 | if (tx_size) { | |
2320 | ofproto->stats.tx_packets++; | |
2321 | ofproto->stats.tx_bytes += tx_size; | |
2322 | } | |
2323 | } | |
2324 | ||
abe529af BP |
2325 | struct port_dump_state { |
2326 | struct dpif_port_dump dump; | |
2327 | bool done; | |
2328 | }; | |
2329 | ||
2330 | static int | |
2331 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2332 | { | |
2333 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2334 | struct port_dump_state *state; | |
2335 | ||
2336 | *statep = state = xmalloc(sizeof *state); | |
2337 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2338 | state->done = false; | |
2339 | return 0; | |
2340 | } | |
2341 | ||
2342 | static int | |
2343 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2344 | struct ofproto_port *port) | |
2345 | { | |
2346 | struct port_dump_state *state = state_; | |
2347 | struct dpif_port dpif_port; | |
2348 | ||
2349 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2350 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2351 | return 0; | |
2352 | } else { | |
2353 | int error = dpif_port_dump_done(&state->dump); | |
2354 | state->done = true; | |
2355 | return error ? error : EOF; | |
2356 | } | |
2357 | } | |
2358 | ||
2359 | static int | |
2360 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2361 | { | |
2362 | struct port_dump_state *state = state_; | |
2363 | ||
2364 | if (!state->done) { | |
2365 | dpif_port_dump_done(&state->dump); | |
2366 | } | |
2367 | free(state); | |
2368 | return 0; | |
2369 | } | |
2370 | ||
2371 | static int | |
2372 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2373 | { | |
2374 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2375 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2376 | } | |
2377 | ||
2378 | static void | |
2379 | port_poll_wait(const struct ofproto *ofproto_) | |
2380 | { | |
2381 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2382 | dpif_port_poll_wait(ofproto->dpif); | |
2383 | } | |
2384 | ||
2385 | static int | |
2386 | port_is_lacp_current(const struct ofport *ofport_) | |
2387 | { | |
2388 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2389 | return (ofport->bundle && ofport->bundle->lacp | |
2390 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2391 | : -1); | |
2392 | } | |
2393 | \f | |
2394 | /* Upcall handling. */ | |
2395 | ||
501f8d1f BP |
2396 | /* Flow miss batching. |
2397 | * | |
2398 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2399 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2400 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2401 | * more packets, plus possibly installing the flow in the dpif. | |
2402 | * | |
2403 | * So far we only batch the operations that affect flow setup time the most. | |
2404 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2405 | struct flow_miss { | |
2406 | struct hmap_node hmap_node; | |
2407 | struct flow flow; | |
b0f7b9b5 | 2408 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2409 | const struct nlattr *key; |
2410 | size_t key_len; | |
e84173dc | 2411 | ovs_be16 initial_tci; |
501f8d1f BP |
2412 | struct list packets; |
2413 | }; | |
2414 | ||
2415 | struct flow_miss_op { | |
c2b565b5 | 2416 | struct dpif_op dpif_op; |
b0f7b9b5 | 2417 | struct subfacet *subfacet; |
501f8d1f BP |
2418 | }; |
2419 | ||
62cd7072 BP |
2420 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2421 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2422 | * configurations. */ |
62cd7072 BP |
2423 | static void |
2424 | send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet, | |
29ebe880 | 2425 | const struct flow *flow) |
62cd7072 BP |
2426 | { |
2427 | struct ofputil_packet_in pin; | |
2428 | ||
3e3252fa EJ |
2429 | pin.packet = packet->data; |
2430 | pin.packet_len = packet->size; | |
65120a8a | 2431 | pin.total_len = packet->size; |
62cd7072 | 2432 | pin.reason = OFPR_NO_MATCH; |
54834960 EJ |
2433 | |
2434 | pin.table_id = 0; | |
2435 | pin.cookie = 0; | |
2436 | ||
62cd7072 BP |
2437 | pin.buffer_id = 0; /* not yet known */ |
2438 | pin.send_len = 0; /* not used for flow table misses */ | |
5d6c3af0 EJ |
2439 | |
2440 | flow_get_metadata(flow, &pin.fmd); | |
2441 | ||
2442 | /* Registers aren't meaningful on a miss. */ | |
2443 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2444 | ||
29ebe880 | 2445 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow); |
62cd7072 BP |
2446 | } |
2447 | ||
abe529af BP |
2448 | static bool |
2449 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2450 | const struct ofpbuf *packet) | |
2451 | { | |
b6e001b6 EJ |
2452 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2453 | ||
2454 | if (!ofport) { | |
2455 | return false; | |
2456 | } | |
2457 | ||
ef9819b5 | 2458 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2459 | if (packet) { |
abe529af BP |
2460 | cfm_process_heartbeat(ofport->cfm, packet); |
2461 | } | |
2462 | return true; | |
b6e001b6 EJ |
2463 | } else if (ofport->bundle && ofport->bundle->lacp |
2464 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2465 | if (packet) { | |
2466 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2467 | } |
da37ebac | 2468 | return true; |
21f7563c JP |
2469 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2470 | if (packet) { | |
2471 | stp_process_packet(ofport, packet); | |
2472 | } | |
2473 | return true; | |
abe529af BP |
2474 | } |
2475 | return false; | |
2476 | } | |
2477 | ||
501f8d1f BP |
2478 | static struct flow_miss * |
2479 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
b0f7b9b5 | 2480 | enum odp_key_fitness key_fitness, |
e84173dc BP |
2481 | const struct nlattr *key, size_t key_len, |
2482 | ovs_be16 initial_tci) | |
abe529af | 2483 | { |
501f8d1f BP |
2484 | uint32_t hash = flow_hash(flow, 0); |
2485 | struct flow_miss *miss; | |
abe529af | 2486 | |
501f8d1f BP |
2487 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2488 | if (flow_equal(&miss->flow, flow)) { | |
2489 | return miss; | |
2490 | } | |
2491 | } | |
abe529af | 2492 | |
501f8d1f BP |
2493 | miss = xmalloc(sizeof *miss); |
2494 | hmap_insert(todo, &miss->hmap_node, hash); | |
2495 | miss->flow = *flow; | |
b0f7b9b5 | 2496 | miss->key_fitness = key_fitness; |
501f8d1f BP |
2497 | miss->key = key; |
2498 | miss->key_len = key_len; | |
e84173dc | 2499 | miss->initial_tci = initial_tci; |
501f8d1f BP |
2500 | list_init(&miss->packets); |
2501 | return miss; | |
2502 | } | |
abe529af | 2503 | |
501f8d1f BP |
2504 | static void |
2505 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2506 | struct flow_miss_op *ops, size_t *n_ops) | |
2507 | { | |
2508 | const struct flow *flow = &miss->flow; | |
2509 | struct ofpbuf *packet, *next_packet; | |
b0f7b9b5 | 2510 | struct subfacet *subfacet; |
501f8d1f | 2511 | struct facet *facet; |
abe529af | 2512 | |
501f8d1f | 2513 | facet = facet_lookup_valid(ofproto, flow); |
abe529af | 2514 | if (!facet) { |
501f8d1f BP |
2515 | struct rule_dpif *rule; |
2516 | ||
2517 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af BP |
2518 | if (!rule) { |
2519 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
501f8d1f | 2520 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af BP |
2521 | if (port) { |
2522 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
2523 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
2524 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
2525 | return; |
2526 | } | |
2527 | } else { | |
2528 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
2529 | flow->in_port); |
2530 | } | |
2531 | ||
29ebe880 EJ |
2532 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2533 | send_packet_in_miss(ofproto, packet, flow); | |
abe529af BP |
2534 | } |
2535 | ||
abe529af BP |
2536 | return; |
2537 | } | |
2538 | ||
501f8d1f | 2539 | facet = facet_create(rule, flow); |
abe529af BP |
2540 | } |
2541 | ||
15baa734 | 2542 | subfacet = subfacet_create(facet, |
e84173dc BP |
2543 | miss->key_fitness, miss->key, miss->key_len, |
2544 | miss->initial_tci); | |
b0f7b9b5 | 2545 | |
501f8d1f | 2546 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) { |
67d91f78 | 2547 | struct dpif_flow_stats stats; |
999fba59 EJ |
2548 | struct flow_miss_op *op; |
2549 | struct dpif_execute *execute; | |
67d91f78 | 2550 | |
501f8d1f BP |
2551 | ofproto->n_matches++; |
2552 | ||
2553 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2554 | /* | |
2555 | * Extra-special case for fail-open mode. | |
2556 | * | |
2557 | * We are in fail-open mode and the packet matched the fail-open | |
2558 | * rule, but we are connected to a controller too. We should send | |
2559 | * the packet up to the controller in the hope that it will try to | |
2560 | * set up a flow and thereby allow us to exit fail-open. | |
2561 | * | |
2562 | * See the top-level comment in fail-open.c for more information. | |
2563 | */ | |
29ebe880 | 2564 | send_packet_in_miss(ofproto, packet, flow); |
501f8d1f BP |
2565 | } |
2566 | ||
b95fc6ba | 2567 | if (!facet->may_install || !subfacet->actions) { |
15baa734 | 2568 | subfacet_make_actions(subfacet, packet); |
501f8d1f | 2569 | } |
67d91f78 | 2570 | |
67d91f78 | 2571 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
15baa734 | 2572 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2573 | |
8338659a BP |
2574 | if (!subfacet->actions_len) { |
2575 | /* No actions to execute, so skip talking to the dpif. */ | |
2576 | continue; | |
2577 | } | |
2578 | ||
2579 | list_remove(&packet->list_node); | |
999fba59 EJ |
2580 | if (flow->vlan_tci != subfacet->initial_tci) { |
2581 | /* This packet was received on a VLAN splinter port. We added | |
2582 | * a VLAN to the packet to make the packet resemble the flow, | |
2583 | * but the actions were composed assuming that the packet | |
2584 | * contained no VLAN. So, we must remove the VLAN header from | |
2585 | * the packet before trying to execute the actions. */ | |
2586 | eth_pop_vlan(packet); | |
501f8d1f | 2587 | } |
999fba59 EJ |
2588 | |
2589 | op = &ops[(*n_ops)++]; | |
c2b565b5 | 2590 | execute = &op->dpif_op.u.execute; |
999fba59 | 2591 | op->subfacet = subfacet; |
c2b565b5 | 2592 | op->dpif_op.type = DPIF_OP_EXECUTE; |
999fba59 EJ |
2593 | execute->key = miss->key; |
2594 | execute->key_len = miss->key_len; | |
2595 | execute->actions = (facet->may_install | |
2596 | ? subfacet->actions | |
2597 | : xmemdup(subfacet->actions, | |
2598 | subfacet->actions_len)); | |
2599 | execute->actions_len = subfacet->actions_len; | |
2600 | execute->packet = packet; | |
501f8d1f BP |
2601 | } |
2602 | ||
b0f7b9b5 | 2603 | if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) { |
501f8d1f | 2604 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2605 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2606 | |
b0f7b9b5 | 2607 | op->subfacet = subfacet; |
c2b565b5 | 2608 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2609 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2610 | put->key = miss->key; | |
2611 | put->key_len = miss->key_len; | |
b95fc6ba BP |
2612 | put->actions = subfacet->actions; |
2613 | put->actions_len = subfacet->actions_len; | |
501f8d1f BP |
2614 | put->stats = NULL; |
2615 | } | |
2616 | } | |
2617 | ||
e2a6ca36 BP |
2618 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
2619 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
2620 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
2621 | * what a flow key should contain. | |
2622 | * | |
2623 | * This function also includes some logic to help make VLAN splinters | |
2624 | * transparent to the rest of the upcall processing logic. In particular, if | |
2625 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
2626 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
2627 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
2628 | * | |
2629 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
2630 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
2631 | * (This differs from the value returned in flow->vlan_tci only for packets | |
2632 | * received on VLAN splinters.) | |
2633 | */ | |
e84173dc | 2634 | static enum odp_key_fitness |
52a90c29 | 2635 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 2636 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
2637 | struct flow *flow, ovs_be16 *initial_tci, |
2638 | struct ofpbuf *packet) | |
e84173dc BP |
2639 | { |
2640 | enum odp_key_fitness fitness; | |
52a90c29 BP |
2641 | uint16_t realdev; |
2642 | int vid; | |
e84173dc BP |
2643 | |
2644 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
2645 | if (fitness == ODP_FIT_ERROR) { | |
2646 | return fitness; | |
2647 | } | |
2648 | *initial_tci = flow->vlan_tci; | |
2649 | ||
52a90c29 BP |
2650 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); |
2651 | if (realdev) { | |
2652 | /* Cause the flow to be processed as if it came in on the real device | |
2653 | * with the VLAN device's VLAN ID. */ | |
2654 | flow->in_port = realdev; | |
2655 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
e2a6ca36 BP |
2656 | if (packet) { |
2657 | /* Make the packet resemble the flow, so that it gets sent to an | |
2658 | * OpenFlow controller properly, so that it looks correct for | |
2659 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
2660 | * if it is called on 'packet'. | |
2661 | * | |
2662 | * The allocated space inside 'packet' probably also contains | |
2663 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
2664 | * struct dpif_upcall (see the large comment on that structure | |
2665 | * definition), so pushing data on 'packet' is in general not a | |
2666 | * good idea since it could overwrite 'key' or free it as a side | |
2667 | * effect. However, it's OK in this special case because we know | |
2668 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
2669 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
2670 | * since we don't need that header anymore. */ | |
2671 | eth_push_vlan(packet, flow->vlan_tci); | |
2672 | } | |
52a90c29 BP |
2673 | |
2674 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
2675 | if (fitness == ODP_FIT_PERFECT) { | |
2676 | fitness = ODP_FIT_TOO_MUCH; | |
2677 | } | |
2678 | } | |
2679 | ||
e84173dc BP |
2680 | return fitness; |
2681 | } | |
2682 | ||
501f8d1f BP |
2683 | static void |
2684 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2685 | size_t n_upcalls) | |
2686 | { | |
2687 | struct dpif_upcall *upcall; | |
2688 | struct flow_miss *miss, *next_miss; | |
2689 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
c2b565b5 | 2690 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f BP |
2691 | struct hmap todo; |
2692 | size_t n_ops; | |
2693 | size_t i; | |
2694 | ||
2695 | if (!n_upcalls) { | |
2696 | return; | |
2697 | } | |
2698 | ||
2699 | /* Construct the to-do list. | |
2700 | * | |
2701 | * This just amounts to extracting the flow from each packet and sticking | |
2702 | * the packets that have the same flow in the same "flow_miss" structure so | |
2703 | * that we can process them together. */ | |
2704 | hmap_init(&todo); | |
2705 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
b0f7b9b5 | 2706 | enum odp_key_fitness fitness; |
501f8d1f | 2707 | struct flow_miss *miss; |
e84173dc | 2708 | ovs_be16 initial_tci; |
501f8d1f BP |
2709 | struct flow flow; |
2710 | ||
b0f7b9b5 BP |
2711 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
2712 | * then set 'flow''s header pointers. */ | |
e84173dc BP |
2713 | fitness = ofproto_dpif_extract_flow_key(ofproto, |
2714 | upcall->key, upcall->key_len, | |
e2a6ca36 BP |
2715 | &flow, &initial_tci, |
2716 | upcall->packet); | |
b0f7b9b5 | 2717 | if (fitness == ODP_FIT_ERROR) { |
6b17ca5c | 2718 | ofpbuf_delete(upcall->packet); |
b0f7b9b5 BP |
2719 | continue; |
2720 | } | |
deedf7e7 | 2721 | flow_extract(upcall->packet, flow.skb_priority, flow.tun_id, |
abff858b | 2722 | flow.in_port, &flow); |
501f8d1f | 2723 | |
21f7563c | 2724 | /* Handle 802.1ag, LACP, and STP specially. */ |
501f8d1f | 2725 | if (process_special(ofproto, &flow, upcall->packet)) { |
6527c598 PS |
2726 | ofproto_update_local_port_stats(&ofproto->up, |
2727 | 0, upcall->packet->size); | |
501f8d1f BP |
2728 | ofpbuf_delete(upcall->packet); |
2729 | ofproto->n_matches++; | |
2730 | continue; | |
2731 | } | |
2732 | ||
2733 | /* Add other packets to a to-do list. */ | |
b0f7b9b5 | 2734 | miss = flow_miss_create(&todo, &flow, fitness, |
e84173dc | 2735 | upcall->key, upcall->key_len, initial_tci); |
501f8d1f BP |
2736 | list_push_back(&miss->packets, &upcall->packet->list_node); |
2737 | } | |
2738 | ||
2739 | /* Process each element in the to-do list, constructing the set of | |
2740 | * operations to batch. */ | |
2741 | n_ops = 0; | |
2742 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { | |
2743 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); | |
2744 | ofpbuf_list_delete(&miss->packets); | |
2745 | hmap_remove(&todo, &miss->hmap_node); | |
2746 | free(miss); | |
abe529af | 2747 | } |
501f8d1f BP |
2748 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
2749 | hmap_destroy(&todo); | |
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 | } | |
2768 | ofpbuf_delete((struct ofpbuf *) execute->packet); | |
2769 | break; | |
abe529af | 2770 | |
501f8d1f | 2771 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 2772 | if (!op->dpif_op.error) { |
b0f7b9b5 | 2773 | op->subfacet->installed = true; |
501f8d1f BP |
2774 | } |
2775 | break; | |
2776 | } | |
2777 | } | |
abe529af BP |
2778 | } |
2779 | ||
2780 | static void | |
6ff686f2 PS |
2781 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2782 | struct dpif_upcall *upcall) | |
abe529af | 2783 | { |
6ff686f2 | 2784 | struct user_action_cookie cookie; |
e84173dc BP |
2785 | enum odp_key_fitness fitness; |
2786 | ovs_be16 initial_tci; | |
2787 | struct flow flow; | |
abe529af | 2788 | |
6ff686f2 | 2789 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2790 | |
e84173dc BP |
2791 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
2792 | upcall->key_len, &flow, | |
e2a6ca36 | 2793 | &initial_tci, upcall->packet); |
e84173dc | 2794 | if (fitness == ODP_FIT_ERROR) { |
6b17ca5c | 2795 | ofpbuf_delete(upcall->packet); |
e84173dc BP |
2796 | return; |
2797 | } | |
2798 | ||
6ff686f2 | 2799 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af | 2800 | if (ofproto->sflow) { |
e84173dc BP |
2801 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
2802 | &cookie); | |
abe529af | 2803 | } |
6ff686f2 PS |
2804 | } else { |
2805 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2806 | } | |
29ebe880 | 2807 | ofpbuf_delete(upcall->packet); |
6ff686f2 PS |
2808 | } |
2809 | ||
9b16c439 BP |
2810 | static int |
2811 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 2812 | { |
9b16c439 BP |
2813 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
2814 | int n_misses; | |
2815 | int i; | |
abe529af | 2816 | |
9b16c439 | 2817 | assert (max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 2818 | |
9b16c439 BP |
2819 | n_misses = 0; |
2820 | for (i = 0; i < max_batch; i++) { | |
2821 | struct dpif_upcall *upcall = &misses[n_misses]; | |
2822 | int error; | |
2823 | ||
2824 | error = dpif_recv(ofproto->dpif, upcall); | |
2825 | if (error) { | |
9b16c439 BP |
2826 | break; |
2827 | } | |
2828 | ||
2829 | switch (upcall->type) { | |
2830 | case DPIF_UC_ACTION: | |
2831 | handle_userspace_upcall(ofproto, upcall); | |
2832 | break; | |
2833 | ||
2834 | case DPIF_UC_MISS: | |
2835 | /* Handle it later. */ | |
2836 | n_misses++; | |
2837 | break; | |
2838 | ||
2839 | case DPIF_N_UC_TYPES: | |
2840 | default: | |
2841 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, | |
2842 | upcall->type); | |
2843 | break; | |
2844 | } | |
abe529af | 2845 | } |
9b16c439 BP |
2846 | |
2847 | handle_miss_upcalls(ofproto, misses, n_misses); | |
2848 | ||
2849 | return i; | |
abe529af BP |
2850 | } |
2851 | \f | |
2852 | /* Flow expiration. */ | |
2853 | ||
b0f7b9b5 | 2854 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
2855 | static void update_stats(struct ofproto_dpif *); |
2856 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 2857 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
2858 | |
2859 | /* This function is called periodically by run(). Its job is to collect | |
2860 | * updates for the flows that have been installed into the datapath, most | |
2861 | * importantly when they last were used, and then use that information to | |
2862 | * expire flows that have not been used recently. | |
2863 | * | |
2864 | * Returns the number of milliseconds after which it should be called again. */ | |
2865 | static int | |
2866 | expire(struct ofproto_dpif *ofproto) | |
2867 | { | |
2868 | struct rule_dpif *rule, *next_rule; | |
0697b5c3 | 2869 | struct classifier *table; |
abe529af BP |
2870 | int dp_max_idle; |
2871 | ||
2872 | /* Update stats for each flow in the datapath. */ | |
2873 | update_stats(ofproto); | |
2874 | ||
b0f7b9b5 BP |
2875 | /* Expire subfacets that have been idle too long. */ |
2876 | dp_max_idle = subfacet_max_idle(ofproto); | |
2877 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
2878 | |
2879 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2880 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2881 | struct cls_cursor cursor; | |
2882 | ||
2883 | cls_cursor_init(&cursor, table, NULL); | |
2884 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
2885 | rule_expire(rule); | |
2886 | } | |
abe529af BP |
2887 | } |
2888 | ||
2889 | /* All outstanding data in existing flows has been accounted, so it's a | |
2890 | * good time to do bond rebalancing. */ | |
2891 | if (ofproto->has_bonded_bundles) { | |
2892 | struct ofbundle *bundle; | |
2893 | ||
2894 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2895 | if (bundle->bond) { | |
2896 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2897 | } | |
2898 | } | |
2899 | } | |
2900 | ||
2901 | return MIN(dp_max_idle, 1000); | |
2902 | } | |
2903 | ||
2904 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2905 | * | |
2906 | * This function also pushes statistics updates to rules which each facet | |
2907 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2908 | * facet changes the rule it resubmits into at some time in between | |
2909 | * update_stats() runs, it is possible that statistics accrued to the | |
2910 | * old rule will be incorrectly attributed to the new rule. This could be | |
2911 | * avoided by calling update_stats() whenever rules are created or | |
2912 | * deleted. However, the performance impact of making so many calls to the | |
2913 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2914 | */ | |
2915 | static void | |
2916 | update_stats(struct ofproto_dpif *p) | |
2917 | { | |
2918 | const struct dpif_flow_stats *stats; | |
2919 | struct dpif_flow_dump dump; | |
2920 | const struct nlattr *key; | |
2921 | size_t key_len; | |
2922 | ||
2923 | dpif_flow_dump_start(&dump, p->dpif); | |
2924 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 2925 | struct subfacet *subfacet; |
abe529af | 2926 | |
6a542738 | 2927 | subfacet = subfacet_find(p, key, key_len); |
b0f7b9b5 BP |
2928 | if (subfacet && subfacet->installed) { |
2929 | struct facet *facet = subfacet->facet; | |
abe529af | 2930 | |
b0f7b9b5 BP |
2931 | if (stats->n_packets >= subfacet->dp_packet_count) { |
2932 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
abe529af BP |
2933 | facet->packet_count += extra; |
2934 | } else { | |
2935 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
2936 | } | |
2937 | ||
b0f7b9b5 BP |
2938 | if (stats->n_bytes >= subfacet->dp_byte_count) { |
2939 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
abe529af BP |
2940 | } else { |
2941 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
2942 | } | |
2943 | ||
b0f7b9b5 BP |
2944 | subfacet->dp_packet_count = stats->n_packets; |
2945 | subfacet->dp_byte_count = stats->n_bytes; | |
abe529af | 2946 | |
15baa734 BP |
2947 | subfacet_update_time(subfacet, stats->used); |
2948 | facet_account(facet); | |
abe529af BP |
2949 | facet_push_stats(facet); |
2950 | } else { | |
6a542738 PS |
2951 | if (!VLOG_DROP_WARN(&rl)) { |
2952 | struct ds s; | |
2953 | ||
2954 | ds_init(&s); | |
2955 | odp_flow_key_format(key, key_len, &s); | |
2956 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
2957 | ds_destroy(&s); | |
2958 | } | |
2959 | ||
2960 | COVERAGE_INC(facet_unexpected); | |
b0f7b9b5 BP |
2961 | /* There's a flow in the datapath that we know nothing about, or a |
2962 | * flow that shouldn't be installed but was anyway. Delete it. */ | |
abe529af BP |
2963 | dpif_flow_del(p->dpif, key, key_len, NULL); |
2964 | } | |
2965 | } | |
2966 | dpif_flow_dump_done(&dump); | |
2967 | } | |
2968 | ||
2969 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
2970 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
2971 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
2972 | * fold its statistic into its rule. */ | |
abe529af | 2973 | static int |
b0f7b9b5 | 2974 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
2975 | { |
2976 | /* | |
2977 | * Idle time histogram. | |
2978 | * | |
b0f7b9b5 BP |
2979 | * Most of the time a switch has a relatively small number of subfacets. |
2980 | * When this is the case we might as well keep statistics for all of them | |
2981 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
2982 | * well. |
2983 | * | |
b0f7b9b5 | 2984 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 2985 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
2986 | * significant. However, with a large number of subfacets it is likely |
2987 | * that only a few of them are "heavy hitters" that consume a large amount | |
2988 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
2989 | * the kernel and maintaining in userspaces; other subfacets we can | |
2990 | * discard. | |
abe529af BP |
2991 | * |
2992 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 2993 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
2994 | * that is installed in the kernel gets dropped in the appropriate bucket. |
2995 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 2996 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 2997 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
2998 | * the most-recently-used bucket of subfacets is kept, so actually an |
2999 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3000 | * (though the next run will delete most of those unless they receive |
3001 | * additional data). | |
abe529af | 3002 | * |
b0f7b9b5 BP |
3003 | * This requires a second pass through the subfacets, in addition to the |
3004 | * pass made by update_stats(), because the former function never looks at | |
3005 | * uninstallable subfacets. | |
abe529af BP |
3006 | */ |
3007 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3008 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3009 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3010 | int total, subtotal, bucket; |
b0f7b9b5 | 3011 | struct subfacet *subfacet; |
abe529af BP |
3012 | long long int now; |
3013 | int i; | |
3014 | ||
b0f7b9b5 | 3015 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3016 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3017 | return N_BUCKETS * BUCKET_WIDTH; |
3018 | } | |
3019 | ||
3020 | /* Build histogram. */ | |
3021 | now = time_msec(); | |
b0f7b9b5 BP |
3022 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3023 | long long int idle = now - subfacet->used; | |
abe529af BP |
3024 | int bucket = (idle <= 0 ? 0 |
3025 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3026 | : (unsigned int) idle / BUCKET_WIDTH); | |
3027 | buckets[bucket]++; | |
3028 | } | |
3029 | ||
3030 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3031 | subtotal = bucket = 0; |
3032 | do { | |
3033 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3034 | } while (bucket < N_BUCKETS && |
3035 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3036 | |
3037 | if (VLOG_IS_DBG_ENABLED()) { | |
3038 | struct ds s; | |
3039 | ||
3040 | ds_init(&s); | |
3041 | ds_put_cstr(&s, "keep"); | |
3042 | for (i = 0; i < N_BUCKETS; i++) { | |
3043 | if (i == bucket) { | |
3044 | ds_put_cstr(&s, ", drop"); | |
3045 | } | |
3046 | if (buckets[i]) { | |
3047 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3048 | } | |
3049 | } | |
3050 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3051 | ds_destroy(&s); | |
3052 | } | |
3053 | ||
3054 | return bucket * BUCKET_WIDTH; | |
3055 | } | |
3056 | ||
abe529af | 3057 | static void |
b0f7b9b5 | 3058 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af BP |
3059 | { |
3060 | long long int cutoff = time_msec() - dp_max_idle; | |
b0f7b9b5 | 3061 | struct subfacet *subfacet, *next_subfacet; |
abe529af | 3062 | |
b0f7b9b5 BP |
3063 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3064 | &ofproto->subfacets) { | |
3065 | if (subfacet->used < cutoff) { | |
15baa734 | 3066 | subfacet_destroy(subfacet); |
abe529af BP |
3067 | } |
3068 | } | |
3069 | } | |
3070 | ||
3071 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3072 | * then delete it entirely. */ | |
3073 | static void | |
3074 | rule_expire(struct rule_dpif *rule) | |
3075 | { | |
abe529af BP |
3076 | struct facet *facet, *next_facet; |
3077 | long long int now; | |
3078 | uint8_t reason; | |
3079 | ||
3080 | /* Has 'rule' expired? */ | |
3081 | now = time_msec(); | |
3082 | if (rule->up.hard_timeout | |
308881af | 3083 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af BP |
3084 | reason = OFPRR_HARD_TIMEOUT; |
3085 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
3086 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
3087 | reason = OFPRR_IDLE_TIMEOUT; | |
3088 | } else { | |
3089 | return; | |
3090 | } | |
3091 | ||
3092 | COVERAGE_INC(ofproto_dpif_expired); | |
3093 | ||
3094 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3095 | * timeout, because that only happens when the rule has no facets left.) */ | |
3096 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3097 | facet_remove(facet); |
abe529af BP |
3098 | } |
3099 | ||
3100 | /* Get rid of the rule. */ | |
3101 | ofproto_rule_expire(&rule->up, reason); | |
3102 | } | |
3103 | \f | |
3104 | /* Facets. */ | |
3105 | ||
f3827897 | 3106 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3107 | * |
3108 | * The caller must already have determined that no facet with an identical | |
3109 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3110 | * the ofproto's classifier table. |
3111 | * | |
b0f7b9b5 BP |
3112 | * The facet will initially have no subfacets. The caller should create (at |
3113 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3114 | static struct facet * |
f3827897 | 3115 | facet_create(struct rule_dpif *rule, const struct flow *flow) |
abe529af BP |
3116 | { |
3117 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3118 | struct facet *facet; | |
3119 | ||
3120 | facet = xzalloc(sizeof *facet); | |
3121 | facet->used = time_msec(); | |
3122 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
3123 | list_push_back(&rule->facets, &facet->list_node); | |
3124 | facet->rule = rule; | |
3125 | facet->flow = *flow; | |
b0f7b9b5 | 3126 | list_init(&facet->subfacets); |
abe529af BP |
3127 | netflow_flow_init(&facet->nf_flow); |
3128 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3129 | ||
abe529af BP |
3130 | return facet; |
3131 | } | |
3132 | ||
3133 | static void | |
3134 | facet_free(struct facet *facet) | |
3135 | { | |
abe529af BP |
3136 | free(facet); |
3137 | } | |
3138 | ||
3d9e05f8 BP |
3139 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3140 | * 'packet', which arrived on 'in_port'. | |
3141 | * | |
3142 | * Takes ownership of 'packet'. */ | |
3143 | static bool | |
3144 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3145 | const struct nlattr *odp_actions, size_t actions_len, | |
3146 | struct ofpbuf *packet) | |
3147 | { | |
3148 | struct odputil_keybuf keybuf; | |
3149 | struct ofpbuf key; | |
3150 | int error; | |
3151 | ||
6ff686f2 PS |
3152 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3153 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3154 | |
6ff686f2 PS |
3155 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3156 | odp_actions, actions_len, packet); | |
80e5eed9 | 3157 | |
6ff686f2 PS |
3158 | ofpbuf_delete(packet); |
3159 | return !error; | |
abe529af BP |
3160 | } |
3161 | ||
abe529af BP |
3162 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3163 | * | |
3164 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3165 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3166 | * |
3167 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3168 | */ | |
3169 | static void | |
15baa734 | 3170 | facet_remove(struct facet *facet) |
abe529af | 3171 | { |
15baa734 | 3172 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3173 | struct subfacet *subfacet, *next_subfacet; |
3174 | ||
551a2f6c BP |
3175 | assert(!list_is_empty(&facet->subfacets)); |
3176 | ||
3177 | /* First uninstall all of the subfacets to get final statistics. */ | |
3178 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3179 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3180 | } |
3181 | ||
3182 | /* Flush the final stats to the rule. | |
3183 | * | |
3184 | * This might require us to have at least one subfacet around so that we | |
3185 | * can use its actions for accounting in facet_account(), which is why we | |
3186 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3187 | facet_flush_stats(facet); |
551a2f6c BP |
3188 | |
3189 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3190 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3191 | &facet->subfacets) { | |
15baa734 | 3192 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3193 | } |
abe529af BP |
3194 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3195 | list_remove(&facet->list_node); | |
3196 | facet_free(facet); | |
3197 | } | |
3198 | ||
abe529af | 3199 | static void |
15baa734 | 3200 | facet_account(struct facet *facet) |
abe529af | 3201 | { |
15baa734 | 3202 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
55af77bb | 3203 | uint64_t n_bytes; |
b95fc6ba | 3204 | struct subfacet *subfacet; |
abe529af | 3205 | const struct nlattr *a; |
abe529af | 3206 | unsigned int left; |
d78be13b | 3207 | ovs_be16 vlan_tci; |
abe529af | 3208 | |
55af77bb | 3209 | if (facet->byte_count <= facet->accounted_bytes) { |
abe529af BP |
3210 | return; |
3211 | } | |
55af77bb EJ |
3212 | n_bytes = facet->byte_count - facet->accounted_bytes; |
3213 | facet->accounted_bytes = facet->byte_count; | |
abe529af | 3214 | |
75a75043 | 3215 | /* Feed information from the active flows back into the learning table to |
abe529af BP |
3216 | * ensure that table is always in sync with what is actually flowing |
3217 | * through the datapath. */ | |
75a75043 BP |
3218 | if (facet->has_learn || facet->has_normal) { |
3219 | struct action_xlate_ctx ctx; | |
abe529af | 3220 | |
e84173dc | 3221 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
54834960 EJ |
3222 | facet->flow.vlan_tci, |
3223 | facet->rule->up.flow_cookie, NULL); | |
75a75043 BP |
3224 | ctx.may_learn = true; |
3225 | ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, | |
3226 | facet->rule->up.n_actions)); | |
3227 | } | |
abe529af | 3228 | |
75a75043 | 3229 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3230 | return; |
3231 | } | |
d78be13b BP |
3232 | |
3233 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3234 | * as a basis. We also need to track the actual VLAN on which the packet | |
3235 | * is going to be sent to ensure that it matches the one passed to | |
3236 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3237 | * hash bucket.) |
3238 | * | |
3239 | * We use the actions from an arbitrary subfacet because they should all | |
3240 | * be equally valid for our purpose. */ | |
3241 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3242 | struct subfacet, list_node); | |
d78be13b | 3243 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3244 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3245 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3246 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3247 | struct ofport_dpif *port; |
abe529af | 3248 | |
d78be13b | 3249 | switch (nl_attr_type(a)) { |
df2c07f4 | 3250 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3251 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3252 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3253 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3254 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3255 | } |
d78be13b BP |
3256 | break; |
3257 | ||
fea393b1 BP |
3258 | case OVS_ACTION_ATTR_POP_VLAN: |
3259 | vlan_tci = htons(0); | |
d78be13b BP |
3260 | break; |
3261 | ||
fea393b1 BP |
3262 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3263 | vlan = nl_attr_get(a); | |
3264 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3265 | break; |
abe529af BP |
3266 | } |
3267 | } | |
3268 | } | |
3269 | ||
abe529af BP |
3270 | /* Returns true if the only action for 'facet' is to send to the controller. |
3271 | * (We don't report NetFlow expiration messages for such facets because they | |
3272 | * are just part of the control logic for the network, not real traffic). */ | |
3273 | static bool | |
3274 | facet_is_controller_flow(struct facet *facet) | |
3275 | { | |
3276 | return (facet | |
3277 | && facet->rule->up.n_actions == 1 | |
3278 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
3279 | htons(OFPP_CONTROLLER))); | |
3280 | } | |
3281 | ||
3282 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3283 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3284 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3285 | * its packet and byte counts before this function is called. */ | |
3286 | static void | |
15baa734 | 3287 | facet_flush_stats(struct facet *facet) |
abe529af | 3288 | { |
15baa734 | 3289 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3290 | struct subfacet *subfacet; |
3291 | ||
3292 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3293 | assert(!subfacet->dp_byte_count); | |
3294 | assert(!subfacet->dp_packet_count); | |
3295 | } | |
abe529af BP |
3296 | |
3297 | facet_push_stats(facet); | |
15baa734 | 3298 | facet_account(facet); |
abe529af BP |
3299 | |
3300 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3301 | struct ofexpired expired; | |
3302 | expired.flow = facet->flow; | |
3303 | expired.packet_count = facet->packet_count; | |
3304 | expired.byte_count = facet->byte_count; | |
3305 | expired.used = facet->used; | |
3306 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3307 | } | |
3308 | ||
3309 | facet->rule->packet_count += facet->packet_count; | |
3310 | facet->rule->byte_count += facet->byte_count; | |
3311 | ||
3312 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3313 | * reinstalled. */ | |
bbb5d219 | 3314 | facet_reset_counters(facet); |
abe529af BP |
3315 | |
3316 | netflow_flow_clear(&facet->nf_flow); | |
3317 | } | |
3318 | ||
3319 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3320 | * Returns it if found, otherwise a null pointer. | |
3321 | * | |
3322 | * The returned facet might need revalidation; use facet_lookup_valid() | |
3323 | * instead if that is important. */ | |
3324 | static struct facet * | |
3325 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3326 | { | |
3327 | struct facet *facet; | |
3328 | ||
3329 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
3330 | &ofproto->facets) { | |
3331 | if (flow_equal(flow, &facet->flow)) { | |
3332 | return facet; | |
3333 | } | |
3334 | } | |
3335 | ||
3336 | return NULL; | |
3337 | } | |
3338 | ||
3339 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3340 | * Returns it if found, otherwise a null pointer. | |
3341 | * | |
3342 | * The returned facet is guaranteed to be valid. */ | |
3343 | static struct facet * | |
3344 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3345 | { | |
3346 | struct facet *facet = facet_find(ofproto, flow); | |
3347 | ||
3348 | /* The facet we found might not be valid, since we could be in need of | |
3349 | * revalidation. If it is not valid, don't return it. */ | |
3350 | if (facet | |
0e4b3771 BP |
3351 | && (ofproto->need_revalidate |
3352 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
15baa734 | 3353 | && !facet_revalidate(facet)) { |
abe529af BP |
3354 | COVERAGE_INC(facet_invalidated); |
3355 | return NULL; | |
3356 | } | |
3357 | ||
3358 | return facet; | |
3359 | } | |
3360 | ||
6814e51f BP |
3361 | static bool |
3362 | facet_check_consistency(struct facet *facet) | |
3363 | { | |
3364 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3365 | ||
3366 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3367 | ||
3368 | struct rule_dpif *rule; | |
3369 | struct subfacet *subfacet; | |
3370 | bool ok; | |
3371 | ||
3372 | /* Check the rule for consistency. */ | |
3373 | rule = rule_dpif_lookup(ofproto, &facet->flow, 0); | |
3374 | if (!rule) { | |
3375 | if (!VLOG_DROP_WARN(&rl)) { | |
3376 | char *s = flow_to_string(&facet->flow); | |
3377 | VLOG_WARN("%s: facet should not exist", s); | |
3378 | free(s); | |
3379 | } | |
3380 | return false; | |
3381 | } else if (rule != facet->rule) { | |
3382 | struct ds s; | |
3383 | ||
3384 | ds_init(&s); | |
3385 | flow_format(&s, &facet->flow); | |
3386 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3387 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3388 | cls_rule_format(&facet->rule->up.cr, &s); | |
3389 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3390 | rule->up.table_id); | |
3391 | cls_rule_format(&rule->up.cr, &s); | |
3392 | ds_put_char(&s, ')'); | |
3393 | ||
3394 | VLOG_WARN("%s", ds_cstr(&s)); | |
3395 | ds_destroy(&s); | |
3396 | ||
3397 | ok = false; | |
3398 | } else { | |
3399 | ok = true; | |
3400 | } | |
3401 | ||
3402 | /* Check the datapath actions for consistency. */ | |
3403 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3404 | struct action_xlate_ctx ctx; | |
3405 | struct ofpbuf *odp_actions; | |
3406 | bool actions_changed; | |
3407 | bool should_install; | |
3408 | ||
3409 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
3410 | subfacet->initial_tci, rule->up.flow_cookie, | |
3411 | NULL); | |
3412 | odp_actions = xlate_actions(&ctx, rule->up.actions, | |
3413 | rule->up.n_actions); | |
3414 | ||
3415 | should_install = (ctx.may_set_up_flow | |
3416 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
3417 | if (!should_install && !subfacet->installed) { | |
3418 | /* The actions for uninstallable flows may vary from one packet to | |
3419 | * the next, so don't compare the actions. */ | |
3420 | goto next; | |
3421 | } | |
3422 | ||
3423 | actions_changed = (subfacet->actions_len != odp_actions->size | |
3424 | || memcmp(subfacet->actions, odp_actions->data, | |
3425 | subfacet->actions_len)); | |
3426 | if (should_install != subfacet->installed || actions_changed) { | |
3427 | struct odputil_keybuf keybuf; | |
3428 | struct ofpbuf key; | |
3429 | struct ds s; | |
3430 | ||
3431 | ok = false; | |
3432 | ||
3433 | ds_init(&s); | |
3434 | subfacet_get_key(subfacet, &keybuf, &key); | |
3435 | odp_flow_key_format(key.data, key.size, &s); | |
3436 | ||
3437 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
3438 | if (should_install != subfacet->installed) { | |
3439 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
3440 | ||
3441 | ds_put_format(&s, " (should%s have been installed)", | |
3442 | should_install ? "" : " not"); | |
3443 | ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)", | |
3444 | ctx.may_set_up_flow ? "true" : "false", | |
3445 | odp_key_fitness_to_string(fitness)); | |
3446 | } | |
3447 | if (actions_changed) { | |
3448 | ds_put_cstr(&s, " (actions were: "); | |
3449 | format_odp_actions(&s, subfacet->actions, | |
3450 | subfacet->actions_len); | |
3451 | ds_put_cstr(&s, ") (correct actions: "); | |
3452 | format_odp_actions(&s, odp_actions->data, | |
3453 | odp_actions->size); | |
3454 | ds_put_char(&s, ')'); | |
3455 | } else { | |
3456 | ds_put_cstr(&s, " (actions: "); | |
3457 | format_odp_actions(&s, subfacet->actions, | |
3458 | subfacet->actions_len); | |
3459 | ds_put_char(&s, ')'); | |
3460 | } | |
3461 | VLOG_WARN("%s", ds_cstr(&s)); | |
3462 | ds_destroy(&s); | |
3463 | } | |
3464 | ||
3465 | next: | |
3466 | ofpbuf_delete(odp_actions); | |
3467 | } | |
3468 | ||
3469 | return ok; | |
3470 | } | |
3471 | ||
15baa734 | 3472 | /* Re-searches the classifier for 'facet': |
abe529af BP |
3473 | * |
3474 | * - If the rule found is different from 'facet''s current rule, moves | |
3475 | * 'facet' to the new rule and recompiles its actions. | |
3476 | * | |
3477 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3478 | * where it is and recompiles its actions anyway. | |
3479 | * | |
3480 | * - If there is none, destroys 'facet'. | |
3481 | * | |
3482 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3483 | static bool | |
15baa734 | 3484 | facet_revalidate(struct facet *facet) |
abe529af | 3485 | { |
15baa734 | 3486 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
3487 | struct actions { |
3488 | struct nlattr *odp_actions; | |
3489 | size_t actions_len; | |
3490 | }; | |
3491 | struct actions *new_actions; | |
3492 | ||
abe529af | 3493 | struct action_xlate_ctx ctx; |
abe529af | 3494 | struct rule_dpif *new_rule; |
b0f7b9b5 | 3495 | struct subfacet *subfacet; |
abe529af | 3496 | bool actions_changed; |
b95fc6ba | 3497 | int i; |
abe529af BP |
3498 | |
3499 | COVERAGE_INC(facet_revalidate); | |
3500 | ||
3501 | /* Determine the new rule. */ | |
29901626 | 3502 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3503 | if (!new_rule) { |
3504 | /* No new rule, so delete the facet. */ | |
15baa734 | 3505 | facet_remove(facet); |
abe529af BP |
3506 | return false; |
3507 | } | |
3508 | ||
df2c07f4 | 3509 | /* Calculate new datapath actions. |
abe529af BP |
3510 | * |
3511 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3512 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3513 | * around to properly compose it. */ | |
abe529af | 3514 | |
df2c07f4 JP |
3515 | /* If the datapath actions changed or the installability changed, |
3516 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
3517 | i = 0; |
3518 | new_actions = NULL; | |
3519 | memset(&ctx, 0, sizeof ctx); | |
b0f7b9b5 | 3520 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
b95fc6ba BP |
3521 | struct ofpbuf *odp_actions; |
3522 | bool should_install; | |
3523 | ||
e84173dc | 3524 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
54834960 EJ |
3525 | subfacet->initial_tci, new_rule->up.flow_cookie, |
3526 | NULL); | |
b95fc6ba BP |
3527 | odp_actions = xlate_actions(&ctx, new_rule->up.actions, |
3528 | new_rule->up.n_actions); | |
3529 | actions_changed = (subfacet->actions_len != odp_actions->size | |
3530 | || memcmp(subfacet->actions, odp_actions->data, | |
3531 | subfacet->actions_len)); | |
3532 | ||
3533 | should_install = (ctx.may_set_up_flow | |
3534 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
b0f7b9b5 BP |
3535 | if (actions_changed || should_install != subfacet->installed) { |
3536 | if (should_install) { | |
3537 | struct dpif_flow_stats stats; | |
3538 | ||
15baa734 | 3539 | subfacet_install(subfacet, |
b0f7b9b5 | 3540 | odp_actions->data, odp_actions->size, &stats); |
15baa734 | 3541 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 3542 | } else { |
15baa734 | 3543 | subfacet_uninstall(subfacet); |
b0f7b9b5 | 3544 | } |
b95fc6ba BP |
3545 | |
3546 | if (!new_actions) { | |
3547 | new_actions = xcalloc(list_size(&facet->subfacets), | |
3548 | sizeof *new_actions); | |
3549 | } | |
3550 | new_actions[i].odp_actions = xmemdup(odp_actions->data, | |
3551 | odp_actions->size); | |
3552 | new_actions[i].actions_len = odp_actions->size; | |
abe529af | 3553 | } |
b95fc6ba BP |
3554 | |
3555 | ofpbuf_delete(odp_actions); | |
3556 | i++; | |
b0f7b9b5 | 3557 | } |
b95fc6ba | 3558 | if (new_actions) { |
15baa734 | 3559 | facet_flush_stats(facet); |
abe529af BP |
3560 | } |
3561 | ||
3562 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3563 | facet->tags = ctx.tags; | |
3564 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3565 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3566 | facet->has_learn = ctx.has_learn; |
3567 | facet->has_normal = ctx.has_normal; | |
9d24de3b | 3568 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3569 | if (new_actions) { |
3570 | i = 0; | |
3571 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3572 | if (new_actions[i].odp_actions) { | |
3573 | free(subfacet->actions); | |
3574 | subfacet->actions = new_actions[i].odp_actions; | |
3575 | subfacet->actions_len = new_actions[i].actions_len; | |
3576 | } | |
3577 | i++; | |
3578 | } | |
3579 | free(new_actions); | |
abe529af BP |
3580 | } |
3581 | if (facet->rule != new_rule) { | |
3582 | COVERAGE_INC(facet_changed_rule); | |
3583 | list_remove(&facet->list_node); | |
3584 | list_push_back(&new_rule->facets, &facet->list_node); | |
3585 | facet->rule = new_rule; | |
3586 | facet->used = new_rule->up.created; | |
9d24de3b | 3587 | facet->prev_used = facet->used; |
abe529af BP |
3588 | } |
3589 | ||
abe529af BP |
3590 | return true; |
3591 | } | |
3592 | ||
3593 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3594 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3595 | static void | |
15baa734 | 3596 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 3597 | { |
15baa734 | 3598 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
3599 | if (used > facet->used) { |
3600 | facet->used = used; | |
3601 | if (used > facet->rule->used) { | |
3602 | facet->rule->used = used; | |
3603 | } | |
3604 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
3605 | } | |
3606 | } | |
3607 | ||
bbb5d219 EJ |
3608 | static void |
3609 | facet_reset_counters(struct facet *facet) | |
3610 | { | |
3611 | facet->packet_count = 0; | |
3612 | facet->byte_count = 0; | |
9d24de3b JP |
3613 | facet->prev_packet_count = 0; |
3614 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
3615 | facet->accounted_bytes = 0; |
3616 | } | |
3617 | ||
abe529af BP |
3618 | static void |
3619 | facet_push_stats(struct facet *facet) | |
3620 | { | |
9d24de3b | 3621 | uint64_t new_packets, new_bytes; |
abe529af | 3622 | |
9d24de3b JP |
3623 | assert(facet->packet_count >= facet->prev_packet_count); |
3624 | assert(facet->byte_count >= facet->prev_byte_count); | |
3625 | assert(facet->used >= facet->prev_used); | |
abe529af | 3626 | |
9d24de3b JP |
3627 | new_packets = facet->packet_count - facet->prev_packet_count; |
3628 | new_bytes = facet->byte_count - facet->prev_byte_count; | |
abe529af | 3629 | |
9d24de3b JP |
3630 | if (new_packets || new_bytes || facet->used > facet->prev_used) { |
3631 | facet->prev_packet_count = facet->packet_count; | |
3632 | facet->prev_byte_count = facet->byte_count; | |
3633 | facet->prev_used = facet->used; | |
abe529af BP |
3634 | |
3635 | flow_push_stats(facet->rule, &facet->flow, | |
9d24de3b JP |
3636 | new_packets, new_bytes, facet->used); |
3637 | ||
3638 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
3639 | facet->mirrors, new_packets, new_bytes); | |
abe529af BP |
3640 | } |
3641 | } | |
3642 | ||
3643 | struct ofproto_push { | |
3644 | struct action_xlate_ctx ctx; | |
3645 | uint64_t packets; | |
3646 | uint64_t bytes; | |
3647 | long long int used; | |
3648 | }; | |
3649 | ||
3650 | static void | |
3651 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3652 | { | |
3653 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3654 | ||
3655 | if (rule) { | |
3656 | rule->packet_count += push->packets; | |
3657 | rule->byte_count += push->bytes; | |
3658 | rule->used = MAX(push->used, rule->used); | |
3659 | } | |
3660 | } | |
3661 | ||
3662 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 3663 | * 'rule''s actions and mirrors. */ |
abe529af BP |
3664 | static void |
3665 | flow_push_stats(const struct rule_dpif *rule, | |
59d0f2c8 | 3666 | const struct flow *flow, uint64_t packets, uint64_t bytes, |
abe529af BP |
3667 | long long int used) |
3668 | { | |
3669 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3670 | struct ofproto_push push; | |
3671 | ||
3672 | push.packets = packets; | |
3673 | push.bytes = bytes; | |
3674 | push.used = used; | |
3675 | ||
54834960 EJ |
3676 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, |
3677 | rule->up.flow_cookie, NULL); | |
abe529af BP |
3678 | push.ctx.resubmit_hook = push_resubmit; |
3679 | ofpbuf_delete(xlate_actions(&push.ctx, | |
3680 | rule->up.actions, rule->up.n_actions)); | |
3681 | } | |
3682 | \f | |
b0f7b9b5 BP |
3683 | /* Subfacets. */ |
3684 | ||
3685 | static struct subfacet * | |
3686 | subfacet_find__(struct ofproto_dpif *ofproto, | |
3687 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
3688 | const struct flow *flow) | |
3689 | { | |
3690 | struct subfacet *subfacet; | |
3691 | ||
3692 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
3693 | &ofproto->subfacets) { | |
3694 | if (subfacet->key | |
3695 | ? (subfacet->key_len == key_len | |
3696 | && !memcmp(key, subfacet->key, key_len)) | |
3697 | : flow_equal(flow, &subfacet->facet->flow)) { | |
3698 | return subfacet; | |
3699 | } | |
3700 | } | |
3701 | ||
3702 | return NULL; | |
3703 | } | |
3704 | ||
3705 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
3706 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
3707 | * there is one, otherwise creates and returns a new subfacet. |
3708 | * | |
3709 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
3710 | * which case the caller must populate the actions with | |
3711 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 3712 | static struct subfacet * |
15baa734 | 3713 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
e84173dc | 3714 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) |
b0f7b9b5 | 3715 | { |
15baa734 | 3716 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3717 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
3718 | struct subfacet *subfacet; | |
3719 | ||
3720 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); | |
3721 | if (subfacet) { | |
3722 | if (subfacet->facet == facet) { | |
3723 | return subfacet; | |
3724 | } | |
3725 | ||
3726 | /* This shouldn't happen. */ | |
3727 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
15baa734 | 3728 | subfacet_destroy(subfacet); |
b0f7b9b5 BP |
3729 | } |
3730 | ||
3731 | subfacet = xzalloc(sizeof *subfacet); | |
3732 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); | |
3733 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
3734 | subfacet->facet = facet; | |
3735 | subfacet->used = time_msec(); | |
3736 | subfacet->key_fitness = key_fitness; | |
3737 | if (key_fitness != ODP_FIT_PERFECT) { | |
3738 | subfacet->key = xmemdup(key, key_len); | |
3739 | subfacet->key_len = key_len; | |
3740 | } | |
3741 | subfacet->installed = false; | |
e84173dc | 3742 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
3743 | |
3744 | return subfacet; | |
3745 | } | |
3746 | ||
3747 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
3748 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
3749 | static struct subfacet * | |
3750 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 3751 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
3752 | { |
3753 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
3754 | enum odp_key_fitness fitness; |
3755 | struct flow flow; | |
3756 | ||
3757 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
3758 | if (fitness == ODP_FIT_ERROR) { | |
3759 | return NULL; | |
3760 | } | |
b0f7b9b5 | 3761 | |
6a542738 | 3762 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
3763 | } |
3764 | ||
3765 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
3766 | * its facet within 'ofproto', and frees it. */ | |
3767 | static void | |
15baa734 | 3768 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 3769 | { |
15baa734 BP |
3770 | struct facet *facet = subfacet->facet; |
3771 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3772 | ||
3773 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
3774 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
3775 | list_remove(&subfacet->list_node); | |
3776 | free(subfacet->key); | |
b95fc6ba | 3777 | free(subfacet->actions); |
b0f7b9b5 BP |
3778 | free(subfacet); |
3779 | } | |
3780 | ||
3781 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
3782 | * last remaining subfacet in its facet destroys the facet too. */ | |
3783 | static void | |
15baa734 | 3784 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
3785 | { |
3786 | struct facet *facet = subfacet->facet; | |
3787 | ||
551a2f6c BP |
3788 | if (list_is_singleton(&facet->subfacets)) { |
3789 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 3790 | facet_remove(facet); |
551a2f6c | 3791 | } else { |
15baa734 | 3792 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
3793 | } |
3794 | } | |
3795 | ||
3796 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
3797 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
3798 | * for use as temporary storage. */ | |
3799 | static void | |
3800 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
3801 | struct ofpbuf *key) | |
3802 | { | |
3803 | if (!subfacet->key) { | |
3804 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
3805 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
3806 | } else { | |
3807 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
3808 | } | |
3809 | } | |
3810 | ||
b95fc6ba BP |
3811 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. */ |
3812 | static void | |
15baa734 | 3813 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet) |
b95fc6ba BP |
3814 | { |
3815 | struct facet *facet = subfacet->facet; | |
3816 | const struct rule_dpif *rule = facet->rule; | |
15baa734 | 3817 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
b95fc6ba BP |
3818 | struct ofpbuf *odp_actions; |
3819 | struct action_xlate_ctx ctx; | |
3820 | ||
15baa734 | 3821 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
54834960 | 3822 | rule->up.flow_cookie, packet); |
b95fc6ba BP |
3823 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); |
3824 | facet->tags = ctx.tags; | |
3825 | facet->may_install = ctx.may_set_up_flow; | |
3826 | facet->has_learn = ctx.has_learn; | |
3827 | facet->has_normal = ctx.has_normal; | |
3828 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
9d24de3b | 3829 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3830 | |
3831 | if (subfacet->actions_len != odp_actions->size | |
3832 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
3833 | free(subfacet->actions); | |
3834 | subfacet->actions_len = odp_actions->size; | |
3835 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
3836 | } | |
3837 | ||
3838 | ofpbuf_delete(odp_actions); | |
3839 | } | |
3840 | ||
b0f7b9b5 BP |
3841 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
3842 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
3843 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
3844 | * since 'subfacet' was last updated. | |
3845 | * | |
3846 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3847 | static int | |
15baa734 | 3848 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 BP |
3849 | const struct nlattr *actions, size_t actions_len, |
3850 | struct dpif_flow_stats *stats) | |
3851 | { | |
15baa734 BP |
3852 | struct facet *facet = subfacet->facet; |
3853 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
b0f7b9b5 BP |
3854 | struct odputil_keybuf keybuf; |
3855 | enum dpif_flow_put_flags flags; | |
3856 | struct ofpbuf key; | |
3857 | int ret; | |
3858 | ||
3859 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
3860 | if (stats) { | |
3861 | flags |= DPIF_FP_ZERO_STATS; | |
3862 | } | |
3863 | ||
3864 | subfacet_get_key(subfacet, &keybuf, &key); | |
3865 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
3866 | actions, actions_len, stats); | |
3867 | ||
3868 | if (stats) { | |
3869 | subfacet_reset_dp_stats(subfacet, stats); | |
3870 | } | |
3871 | ||
3872 | return ret; | |
3873 | } | |
3874 | ||
3875 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
3876 | static void | |
15baa734 | 3877 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 BP |
3878 | { |
3879 | if (subfacet->installed) { | |
15baa734 BP |
3880 | struct rule_dpif *rule = subfacet->facet->rule; |
3881 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
3882 | struct odputil_keybuf keybuf; |
3883 | struct dpif_flow_stats stats; | |
3884 | struct ofpbuf key; | |
3885 | int error; | |
3886 | ||
3887 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 3888 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
3889 | subfacet_reset_dp_stats(subfacet, &stats); |
3890 | if (!error) { | |
15baa734 | 3891 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 BP |
3892 | } |
3893 | subfacet->installed = false; | |
3894 | } else { | |
3895 | assert(subfacet->dp_packet_count == 0); | |
3896 | assert(subfacet->dp_byte_count == 0); | |
3897 | } | |
3898 | } | |
3899 | ||
3900 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
3901 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
3902 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
3903 | * was reset in the datapath. 'stats' will be modified to include only | |
3904 | * statistics new since 'subfacet' was last updated. */ | |
3905 | static void | |
3906 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
3907 | struct dpif_flow_stats *stats) | |
3908 | { | |
3909 | if (stats | |
3910 | && subfacet->dp_packet_count <= stats->n_packets | |
3911 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
3912 | stats->n_packets -= subfacet->dp_packet_count; | |
3913 | stats->n_bytes -= subfacet->dp_byte_count; | |
3914 | } | |
3915 | ||
3916 | subfacet->dp_packet_count = 0; | |
3917 | subfacet->dp_byte_count = 0; | |
3918 | } | |
3919 | ||
3920 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
3921 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
3922 | static void | |
15baa734 | 3923 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
3924 | { |
3925 | if (used > subfacet->used) { | |
3926 | subfacet->used = used; | |
15baa734 | 3927 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
3928 | } |
3929 | } | |
3930 | ||
3931 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
3932 | * | |
3933 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
3934 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
3935 | * represents a packet that was sent by hand or if it represents statistics | |
3936 | * that have been cleared out of the datapath. */ | |
3937 | static void | |
15baa734 | 3938 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
3939 | const struct dpif_flow_stats *stats) |
3940 | { | |
3941 | if (stats->n_packets || stats->used > subfacet->used) { | |
3942 | struct facet *facet = subfacet->facet; | |
3943 | ||
15baa734 | 3944 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
3945 | facet->packet_count += stats->n_packets; |
3946 | facet->byte_count += stats->n_bytes; | |
3947 | facet_push_stats(facet); | |
3948 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
3949 | } | |
3950 | } | |
3951 | \f | |
abe529af BP |
3952 | /* Rules. */ |
3953 | ||
3954 | static struct rule_dpif * | |
29901626 BP |
3955 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
3956 | uint8_t table_id) | |
abe529af | 3957 | { |
7257b535 BP |
3958 | struct cls_rule *cls_rule; |
3959 | struct classifier *cls; | |
3960 | ||
9cdaaebe BP |
3961 | if (table_id >= N_TABLES) { |
3962 | return NULL; | |
3963 | } | |
3964 | ||
7257b535 | 3965 | cls = &ofproto->up.tables[table_id]; |
eadef313 | 3966 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
3967 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
3968 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
3969 | * are unavailable. */ | |
3970 | struct flow ofpc_normal_flow = *flow; | |
3971 | ofpc_normal_flow.tp_src = htons(0); | |
3972 | ofpc_normal_flow.tp_dst = htons(0); | |
3973 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
3974 | } else { | |
3975 | cls_rule = classifier_lookup(cls, flow); | |
3976 | } | |
3977 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
3978 | } |
3979 | ||
7ee20df1 BP |
3980 | static void |
3981 | complete_operation(struct rule_dpif *rule) | |
3982 | { | |
3983 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3984 | ||
54a9cbc9 | 3985 | rule_invalidate(rule); |
7ee20df1 BP |
3986 | if (clogged) { |
3987 | struct dpif_completion *c = xmalloc(sizeof *c); | |
3988 | c->op = rule->up.pending; | |
3989 | list_push_back(&ofproto->completions, &c->list_node); | |
3990 | } else { | |
3991 | ofoperation_complete(rule->up.pending, 0); | |
3992 | } | |
3993 | } | |
3994 | ||
abe529af BP |
3995 | static struct rule * |
3996 | rule_alloc(void) | |
3997 | { | |
3998 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
3999 | return &rule->up; | |
4000 | } | |
4001 | ||
4002 | static void | |
4003 | rule_dealloc(struct rule *rule_) | |
4004 | { | |
4005 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4006 | free(rule); | |
4007 | } | |
4008 | ||
90bf1e07 | 4009 | static enum ofperr |
abe529af BP |
4010 | rule_construct(struct rule *rule_) |
4011 | { | |
4012 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4013 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4014 | struct rule_dpif *victim; |
54a9cbc9 | 4015 | uint8_t table_id; |
90bf1e07 | 4016 | enum ofperr error; |
5bf0e941 BP |
4017 | |
4018 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
4019 | &rule->up.cr.flow, ofproto->max_ports); | |
4020 | if (error) { | |
4021 | return error; | |
4022 | } | |
abe529af BP |
4023 | |
4024 | rule->used = rule->up.created; | |
4025 | rule->packet_count = 0; | |
4026 | rule->byte_count = 0; | |
abe529af | 4027 | |
7ee20df1 BP |
4028 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4029 | if (victim && !list_is_empty(&victim->facets)) { | |
4030 | struct facet *facet; | |
4031 | ||
4032 | rule->facets = victim->facets; | |
4033 | list_moved(&rule->facets); | |
4034 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4035 | /* XXX: We're only clearing our local counters here. It's possible |
4036 | * that quite a few packets are unaccounted for in the datapath | |
4037 | * statistics. These will be accounted to the new rule instead of | |
4038 | * cleared as required. This could be fixed by clearing out the | |
4039 | * datapath statistics for this facet, but currently it doesn't | |
4040 | * seem worth it. */ | |
4041 | facet_reset_counters(facet); | |
7ee20df1 BP |
4042 | facet->rule = rule; |
4043 | } | |
4044 | } else { | |
4045 | /* Must avoid list_moved() in this case. */ | |
4046 | list_init(&rule->facets); | |
4047 | } | |
abe529af | 4048 | |
54a9cbc9 BP |
4049 | table_id = rule->up.table_id; |
4050 | rule->tag = (victim ? victim->tag | |
4051 | : table_id == 0 ? 0 | |
4052 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4053 | ofproto->tables[table_id].basis)); | |
4054 | ||
7ee20df1 | 4055 | complete_operation(rule); |
abe529af BP |
4056 | return 0; |
4057 | } | |
4058 | ||
4059 | static void | |
4060 | rule_destruct(struct rule *rule_) | |
4061 | { | |
4062 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4063 | struct facet *facet, *next_facet; |
4064 | ||
abe529af | 4065 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4066 | facet_revalidate(facet); |
abe529af | 4067 | } |
7ee20df1 BP |
4068 | |
4069 | complete_operation(rule); | |
abe529af BP |
4070 | } |
4071 | ||
4072 | static void | |
4073 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4074 | { | |
4075 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4076 | struct facet *facet; | |
4077 | ||
4078 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4079 | * in facets. This counts, for example, facets that have expired. */ | |
4080 | *packets = rule->packet_count; | |
4081 | *bytes = rule->byte_count; | |
4082 | ||
4083 | /* Add any statistics that are tracked by facets. This includes | |
4084 | * statistical data recently updated by ofproto_update_stats() as well as | |
4085 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4086 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4087 | *packets += facet->packet_count; | |
4088 | *bytes += facet->byte_count; | |
4089 | } | |
4090 | } | |
4091 | ||
90bf1e07 | 4092 | static enum ofperr |
59d0f2c8 BP |
4093 | rule_execute(struct rule *rule_, const struct flow *flow, |
4094 | struct ofpbuf *packet) | |
abe529af BP |
4095 | { |
4096 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4097 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4098 | struct action_xlate_ctx ctx; | |
4099 | struct ofpbuf *odp_actions; | |
abe529af BP |
4100 | size_t size; |
4101 | ||
54834960 EJ |
4102 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
4103 | rule->up.flow_cookie, packet); | |
abe529af BP |
4104 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); |
4105 | size = packet->size; | |
4106 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
4107 | odp_actions->size, packet)) { | |
4108 | rule->used = time_msec(); | |
4109 | rule->packet_count++; | |
4110 | rule->byte_count += size; | |
4111 | flow_push_stats(rule, flow, 1, size, rule->used); | |
4112 | } | |
4113 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
4114 | |
4115 | return 0; | |
abe529af BP |
4116 | } |
4117 | ||
7ee20df1 BP |
4118 | static void |
4119 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4120 | { |
4121 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
90bf1e07 | 4123 | enum ofperr error; |
abe529af | 4124 | |
7ee20df1 BP |
4125 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
4126 | &rule->up.cr.flow, ofproto->max_ports); | |
4127 | if (error) { | |
4128 | ofoperation_complete(rule->up.pending, error); | |
4129 | return; | |
abe529af | 4130 | } |
7ee20df1 BP |
4131 | |
4132 | complete_operation(rule); | |
abe529af BP |
4133 | } |
4134 | \f | |
97d6520b | 4135 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4136 | * May modify 'packet'. |
abe529af BP |
4137 | * Returns 0 if successful, otherwise a positive errno value. */ |
4138 | static int | |
52a90c29 | 4139 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4140 | { |
97d6520b | 4141 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4142 | struct ofpbuf key, odp_actions; |
4143 | struct odputil_keybuf keybuf; | |
52a90c29 | 4144 | uint16_t odp_port; |
80e5eed9 | 4145 | struct flow flow; |
abe529af BP |
4146 | int error; |
4147 | ||
abff858b | 4148 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
52a90c29 BP |
4149 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4150 | flow.vlan_tci); | |
4151 | if (odp_port != ofport->odp_port) { | |
4152 | eth_pop_vlan(packet); | |
4153 | flow.vlan_tci = htons(0); | |
4154 | } | |
4155 | ||
80e5eed9 BP |
4156 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4157 | odp_flow_key_from_flow(&key, &flow); | |
4158 | ||
abe529af | 4159 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4160 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4161 | ||
df2c07f4 | 4162 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4163 | error = dpif_execute(ofproto->dpif, |
4164 | key.data, key.size, | |
4165 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4166 | packet); |
4167 | ofpbuf_uninit(&odp_actions); | |
4168 | ||
4169 | if (error) { | |
4170 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4171 | ofproto->up.name, odp_port, strerror(error)); | |
4172 | } | |
6527c598 | 4173 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4174 | return error; |
4175 | } | |
4176 | \f | |
df2c07f4 | 4177 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
4178 | |
4179 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4180 | struct action_xlate_ctx *ctx); | |
4cd78906 | 4181 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4182 | |
98403001 BP |
4183 | static size_t |
4184 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4185 | struct ofpbuf *odp_actions, | |
4186 | const struct flow *flow, | |
4187 | const struct user_action_cookie *cookie) | |
4188 | { | |
98403001 BP |
4189 | uint32_t pid; |
4190 | ||
4191 | pid = dpif_port_get_pid(ofproto->dpif, | |
4192 | ofp_port_to_odp_port(flow->in_port)); | |
4193 | ||
39db78a0 | 4194 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4195 | } |
4196 | ||
6ff686f2 PS |
4197 | /* Compose SAMPLE action for sFlow. */ |
4198 | static size_t | |
4199 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4200 | struct ofpbuf *odp_actions, | |
4201 | const struct flow *flow, | |
4202 | uint32_t odp_port) | |
4203 | { | |
4204 | uint32_t port_ifindex; | |
4205 | uint32_t probability; | |
98403001 | 4206 | struct user_action_cookie cookie; |
6ff686f2 | 4207 | size_t sample_offset, actions_offset; |
98403001 | 4208 | int cookie_offset, n_output; |
6ff686f2 PS |
4209 | |
4210 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4211 | return 0; | |
4212 | } | |
4213 | ||
4214 | if (odp_port == OVSP_NONE) { | |
4215 | port_ifindex = 0; | |
4216 | n_output = 0; | |
4217 | } else { | |
4218 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4219 | n_output = 1; | |
4220 | } | |
4221 | ||
4222 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
4223 | ||
4224 | /* Number of packets out of UINT_MAX to sample. */ | |
4225 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4226 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4227 | ||
4228 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
4229 | ||
98403001 BP |
4230 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
4231 | cookie.data = port_ifindex; | |
4232 | cookie.n_output = n_output; | |
4233 | cookie.vlan_tci = 0; | |
4234 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
4235 | |
4236 | nl_msg_end_nested(odp_actions, actions_offset); | |
4237 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4238 | return cookie_offset; |
6ff686f2 PS |
4239 | } |
4240 | ||
4241 | /* SAMPLE action must be first action in any given list of actions. | |
4242 | * At this point we do not have all information required to build it. So try to | |
4243 | * build sample action as complete as possible. */ | |
4244 | static void | |
4245 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4246 | { | |
4247 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4248 | ctx->odp_actions, | |
4249 | &ctx->flow, OVSP_NONE); | |
4250 | ctx->sflow_odp_port = 0; | |
4251 | ctx->sflow_n_outputs = 0; | |
4252 | } | |
4253 | ||
4254 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4255 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4256 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4257 | static void | |
4258 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4259 | { | |
4260 | const struct flow *base = &ctx->base_flow; | |
4261 | struct user_action_cookie *cookie; | |
4262 | ||
4263 | if (!ctx->user_cookie_offset) { | |
4264 | return; | |
4265 | } | |
4266 | ||
4267 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
4268 | sizeof(*cookie)); | |
4269 | assert(cookie != NULL); | |
4270 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
4271 | ||
4272 | if (ctx->sflow_n_outputs) { | |
4273 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
4274 | ctx->sflow_odp_port); | |
4275 | } | |
4276 | if (ctx->sflow_n_outputs >= 255) { | |
4277 | cookie->n_output = 255; | |
4278 | } else { | |
4279 | cookie->n_output = ctx->sflow_n_outputs; | |
4280 | } | |
4281 | cookie->vlan_tci = base->vlan_tci; | |
4282 | } | |
4283 | ||
6ff686f2 | 4284 | static void |
81b1afb1 EJ |
4285 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4286 | bool check_stp) | |
6ff686f2 | 4287 | { |
d59906fb | 4288 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4289 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4290 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4291 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4292 | uint16_t out_port; |
d59906fb | 4293 | |
81b1afb1 | 4294 | if (ofport) { |
8b36f51e EJ |
4295 | struct priority_to_dscp *pdscp; |
4296 | ||
81b1afb1 EJ |
4297 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD) |
4298 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { | |
4299 | return; | |
4300 | } | |
8b36f51e | 4301 | |
deedf7e7 | 4302 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4303 | if (pdscp) { |
4304 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4305 | ctx->flow.nw_tos |= pdscp->dscp; | |
4306 | } | |
81b1afb1 EJ |
4307 | } else { |
4308 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4309 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4310 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4311 | } |
4312 | ||
52a90c29 BP |
4313 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4314 | ctx->flow.vlan_tci); | |
4315 | if (out_port != odp_port) { | |
4316 | ctx->flow.vlan_tci = htons(0); | |
4317 | } | |
5bbda0aa | 4318 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4319 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4320 | ||
6ff686f2 PS |
4321 | ctx->sflow_odp_port = odp_port; |
4322 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4323 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4324 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4325 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4326 | } |
4327 | ||
abe529af | 4328 | static void |
5e48dc2b | 4329 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4330 | { |
81b1afb1 | 4331 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4332 | } |
4333 | ||
4334 | static void | |
29901626 BP |
4335 | xlate_table_action(struct action_xlate_ctx *ctx, |
4336 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
4337 | { |
4338 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4339 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4340 | struct rule_dpif *rule; |
4341 | uint16_t old_in_port; | |
29901626 BP |
4342 | uint8_t old_table_id; |
4343 | ||
4344 | old_table_id = ctx->table_id; | |
4345 | ctx->table_id = table_id; | |
abe529af | 4346 | |
54a9cbc9 | 4347 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
4348 | old_in_port = ctx->flow.in_port; |
4349 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
4350 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
4351 | ||
4352 | /* Tag the flow. */ | |
4353 | if (table_id > 0 && table_id < N_TABLES) { | |
4354 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4355 | if (table->other_table) { | |
4356 | ctx->tags |= (rule | |
4357 | ? rule->tag | |
4358 | : rule_calculate_tag(&ctx->flow, | |
4359 | &table->other_table->wc, | |
4360 | table->basis)); | |
4361 | } | |
4362 | } | |
4363 | ||
4364 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
4365 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
4366 | ctx->flow.in_port = old_in_port; |
4367 | ||
4368 | if (ctx->resubmit_hook) { | |
4369 | ctx->resubmit_hook(ctx, rule); | |
4370 | } | |
4371 | ||
4372 | if (rule) { | |
54834960 EJ |
4373 | ovs_be64 old_cookie = ctx->cookie; |
4374 | ||
abe529af | 4375 | ctx->recurse++; |
54834960 | 4376 | ctx->cookie = rule->up.flow_cookie; |
abe529af | 4377 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); |
54834960 | 4378 | ctx->cookie = old_cookie; |
abe529af BP |
4379 | ctx->recurse--; |
4380 | } | |
29901626 BP |
4381 | |
4382 | ctx->table_id = old_table_id; | |
abe529af BP |
4383 | } else { |
4384 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4385 | ||
29901626 | 4386 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af BP |
4387 | MAX_RESUBMIT_RECURSION); |
4388 | } | |
4389 | } | |
4390 | ||
29901626 BP |
4391 | static void |
4392 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
4393 | const struct nx_action_resubmit *nar) | |
4394 | { | |
4395 | uint16_t in_port; | |
4396 | uint8_t table_id; | |
4397 | ||
4398 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
4399 | ? ctx->flow.in_port | |
4400 | : ntohs(nar->in_port)); | |
4401 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
4402 | ||
4403 | xlate_table_action(ctx, in_port, table_id); | |
4404 | } | |
4405 | ||
abe529af | 4406 | static void |
d59906fb | 4407 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
4408 | { |
4409 | struct ofport_dpif *ofport; | |
4410 | ||
b3e9b2ed | 4411 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 4412 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
4413 | |
4414 | if (ofp_port == ctx->flow.in_port) { | |
4415 | continue; | |
4416 | } | |
4417 | ||
5e48dc2b | 4418 | if (all) { |
81b1afb1 | 4419 | compose_output_action__(ctx, ofp_port, false); |
5e48dc2b EJ |
4420 | } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) { |
4421 | compose_output_action(ctx, ofp_port); | |
abe529af BP |
4422 | } |
4423 | } | |
b3e9b2ed EJ |
4424 | |
4425 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
4426 | } |
4427 | ||
6ff686f2 | 4428 | static void |
f0fd1a17 PS |
4429 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
4430 | enum ofp_packet_in_reason reason) | |
6ff686f2 | 4431 | { |
999fba59 EJ |
4432 | struct ofputil_packet_in pin; |
4433 | struct ofpbuf *packet; | |
6ff686f2 | 4434 | |
999fba59 EJ |
4435 | ctx->may_set_up_flow = false; |
4436 | if (!ctx->packet) { | |
4437 | return; | |
4438 | } | |
4439 | ||
4440 | packet = ofpbuf_clone(ctx->packet); | |
4441 | ||
4442 | if (packet->l2 && packet->l3) { | |
4443 | struct eth_header *eh; | |
4444 | ||
4445 | eth_pop_vlan(packet); | |
4446 | eh = packet->l2; | |
4447 | assert(eh->eth_type == ctx->flow.dl_type); | |
4448 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); | |
4449 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
4450 | ||
4451 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
4452 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
4453 | } | |
4454 | ||
4455 | if (packet->l4) { | |
4456 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4457 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
4458 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
4459 | } | |
4460 | ||
4461 | if (packet->l7) { | |
4462 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
4463 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
4464 | ctx->flow.tp_dst); | |
4465 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
4466 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
4467 | ctx->flow.tp_dst); | |
4468 | } | |
4469 | } | |
4470 | } | |
4471 | } | |
4472 | ||
4473 | pin.packet = packet->data; | |
4474 | pin.packet_len = packet->size; | |
f0fd1a17 | 4475 | pin.reason = reason; |
54834960 EJ |
4476 | pin.table_id = ctx->table_id; |
4477 | pin.cookie = ctx->cookie; | |
4478 | ||
999fba59 EJ |
4479 | pin.buffer_id = 0; |
4480 | pin.send_len = len; | |
4481 | pin.total_len = packet->size; | |
4482 | flow_get_metadata(&ctx->flow, &pin.fmd); | |
4483 | ||
4484 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow); | |
4485 | ofpbuf_delete(packet); | |
6ff686f2 PS |
4486 | } |
4487 | ||
f0fd1a17 PS |
4488 | static bool |
4489 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
4490 | { | |
4491 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
4492 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
4493 | return false; | |
4494 | } | |
4495 | ||
4496 | if (ctx->flow.nw_ttl > 1) { | |
4497 | ctx->flow.nw_ttl--; | |
4498 | return false; | |
4499 | } else { | |
4500 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL); | |
4501 | ||
4502 | /* Stop processing for current table. */ | |
4503 | return true; | |
4504 | } | |
4505 | } | |
4506 | ||
abe529af BP |
4507 | static void |
4508 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
4509 | uint16_t port, uint16_t max_len) | |
4510 | { | |
4511 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
4512 | ||
4513 | ctx->nf_output_iface = NF_OUT_DROP; | |
4514 | ||
4515 | switch (port) { | |
4516 | case OFPP_IN_PORT: | |
81b1afb1 | 4517 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
4518 | break; |
4519 | case OFPP_TABLE: | |
29901626 | 4520 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
4521 | break; |
4522 | case OFPP_NORMAL: | |
4523 | xlate_normal(ctx); | |
4524 | break; | |
4525 | case OFPP_FLOOD: | |
d59906fb | 4526 | flood_packets(ctx, false); |
abe529af BP |
4527 | break; |
4528 | case OFPP_ALL: | |
d59906fb | 4529 | flood_packets(ctx, true); |
abe529af BP |
4530 | break; |
4531 | case OFPP_CONTROLLER: | |
f0fd1a17 | 4532 | execute_controller_action(ctx, max_len, OFPR_ACTION); |
abe529af BP |
4533 | break; |
4534 | case OFPP_LOCAL: | |
81b1afb1 | 4535 | compose_output_action(ctx, OFPP_LOCAL); |
abe529af | 4536 | break; |
e81d2933 EJ |
4537 | case OFPP_NONE: |
4538 | break; | |
abe529af BP |
4539 | default: |
4540 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 4541 | compose_output_action(ctx, port); |
abe529af BP |
4542 | } |
4543 | break; | |
4544 | } | |
4545 | ||
4546 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
4547 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4548 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4549 | ctx->nf_output_iface = prev_nf_output_iface; | |
4550 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
4551 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
4552 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4553 | } | |
4554 | } | |
4555 | ||
f694937d EJ |
4556 | static void |
4557 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
4558 | const struct nx_action_output_reg *naor) | |
4559 | { | |
4560 | uint64_t ofp_port; | |
4561 | ||
4562 | ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow); | |
4563 | ||
4564 | if (ofp_port <= UINT16_MAX) { | |
4565 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
4566 | } | |
4567 | } | |
4568 | ||
abe529af BP |
4569 | static void |
4570 | xlate_output_action(struct action_xlate_ctx *ctx, | |
4571 | const struct ofp_action_output *oao) | |
4572 | { | |
4573 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
4574 | } | |
4575 | ||
abe529af BP |
4576 | static void |
4577 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
4578 | const struct ofp_action_enqueue *oae) | |
4579 | { | |
e479e41e | 4580 | uint16_t ofp_port; |
abff858b | 4581 | uint32_t flow_priority, priority; |
abe529af BP |
4582 | int error; |
4583 | ||
4584 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
4585 | &priority); | |
4586 | if (error) { | |
4587 | /* Fall back to ordinary output action. */ | |
4588 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
4589 | return; | |
4590 | } | |
4591 | ||
df2c07f4 | 4592 | /* Figure out datapath output port. */ |
abe529af BP |
4593 | ofp_port = ntohs(oae->port); |
4594 | if (ofp_port == OFPP_IN_PORT) { | |
4595 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
4596 | } else if (ofp_port == ctx->flow.in_port) { |
4597 | return; | |
abe529af | 4598 | } |
abe529af | 4599 | |
df2c07f4 | 4600 | /* Add datapath actions. */ |
deedf7e7 BP |
4601 | flow_priority = ctx->flow.skb_priority; |
4602 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 4603 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 4604 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
4605 | |
4606 | /* Update NetFlow output port. */ | |
4607 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 4608 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
4609 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
4610 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4611 | } | |
4612 | } | |
4613 | ||
4614 | static void | |
4615 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
4616 | const struct nx_action_set_queue *nasq) | |
4617 | { | |
4618 | uint32_t priority; | |
4619 | int error; | |
4620 | ||
4621 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4622 | &priority); | |
4623 | if (error) { | |
4624 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4625 | * has already been logged. */ | |
4626 | return; | |
4627 | } | |
4628 | ||
deedf7e7 | 4629 | ctx->flow.skb_priority = priority; |
abe529af BP |
4630 | } |
4631 | ||
4632 | struct xlate_reg_state { | |
4633 | ovs_be16 vlan_tci; | |
4634 | ovs_be64 tun_id; | |
4635 | }; | |
4636 | ||
abe529af BP |
4637 | static void |
4638 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4639 | const struct nx_action_autopath *naa) | |
4640 | { | |
4641 | uint16_t ofp_port = ntohl(naa->id); | |
4642 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4643 | ||
4644 | if (!port || !port->bundle) { | |
4645 | ofp_port = OFPP_NONE; | |
4646 | } else if (port->bundle->bond) { | |
4647 | /* Autopath does not support VLAN hashing. */ | |
4648 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4649 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4650 | if (slave) { |
4651 | ofp_port = slave->up.ofp_port; | |
4652 | } | |
4653 | } | |
4654 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4655 | } | |
4656 | ||
daff3353 EJ |
4657 | static bool |
4658 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4659 | { | |
4660 | struct ofproto_dpif *ofproto = ofproto_; | |
4661 | struct ofport_dpif *port; | |
4662 | ||
4663 | switch (ofp_port) { | |
4664 | case OFPP_IN_PORT: | |
4665 | case OFPP_TABLE: | |
4666 | case OFPP_NORMAL: | |
4667 | case OFPP_FLOOD: | |
4668 | case OFPP_ALL: | |
439e4d8c | 4669 | case OFPP_NONE: |
daff3353 EJ |
4670 | return true; |
4671 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4672 | return false; | |
4673 | default: | |
4674 | port = get_ofp_port(ofproto, ofp_port); | |
4675 | return port ? port->may_enable : false; | |
4676 | } | |
4677 | } | |
4678 | ||
75a75043 BP |
4679 | static void |
4680 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
4681 | const struct nx_action_learn *learn) | |
4682 | { | |
4683 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
4684 | struct ofputil_flow_mod fm; | |
4685 | int error; | |
4686 | ||
4687 | learn_execute(learn, &ctx->flow, &fm); | |
4688 | ||
4689 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
4690 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
4691 | VLOG_WARN("learning action failed to modify flow table (%s)", |
4692 | ofperr_get_name(error)); | |
75a75043 BP |
4693 | } |
4694 | ||
4695 | free(fm.actions); | |
4696 | } | |
4697 | ||
21f7563c JP |
4698 | static bool |
4699 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
4700 | { | |
4701 | if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
4702 | ? htonl(OFPPC_NO_RECV_STP) | |
4703 | : htonl(OFPPC_NO_RECV))) { | |
4704 | return false; | |
4705 | } | |
4706 | ||
4707 | /* Only drop packets here if both forwarding and learning are | |
4708 | * disabled. If just learning is enabled, we need to have | |
4709 | * OFPP_NORMAL and the learning action have a look at the packet | |
4710 | * before we can drop it. */ | |
4711 | if (!stp_forward_in_state(port->stp_state) | |
4712 | && !stp_learn_in_state(port->stp_state)) { | |
4713 | return false; | |
4714 | } | |
4715 | ||
4716 | return true; | |
4717 | } | |
4718 | ||
abe529af BP |
4719 | static void |
4720 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4721 | struct action_xlate_ctx *ctx) | |
4722 | { | |
4723 | const struct ofport_dpif *port; | |
abe529af | 4724 | const union ofp_action *ia; |
b4b8c781 | 4725 | size_t left; |
abe529af BP |
4726 | |
4727 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 4728 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
4729 | /* Drop this flow. */ |
4730 | return; | |
4731 | } | |
4732 | ||
b4b8c781 | 4733 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 4734 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
4735 | const struct nx_action_resubmit *nar; |
4736 | const struct nx_action_set_tunnel *nast; | |
4737 | const struct nx_action_set_queue *nasq; | |
4738 | const struct nx_action_multipath *nam; | |
4739 | const struct nx_action_autopath *naa; | |
daff3353 | 4740 | const struct nx_action_bundle *nab; |
f694937d | 4741 | const struct nx_action_output_reg *naor; |
38f2e360 BP |
4742 | enum ofputil_action_code code; |
4743 | ovs_be64 tun_id; | |
4744 | ||
848e8809 EJ |
4745 | if (ctx->exit) { |
4746 | break; | |
4747 | } | |
4748 | ||
38f2e360 BP |
4749 | code = ofputil_decode_action_unsafe(ia); |
4750 | switch (code) { | |
4751 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
4752 | xlate_output_action(ctx, &ia->output); |
4753 | break; | |
4754 | ||
38f2e360 | 4755 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
4756 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
4757 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
4758 | break; |
4759 | ||
38f2e360 | 4760 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
4761 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
4762 | ctx->flow.vlan_tci |= htons( | |
4763 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
4764 | break; |
4765 | ||
38f2e360 | 4766 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 4767 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
4768 | break; |
4769 | ||
38f2e360 | 4770 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 4771 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4772 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
4773 | break; | |
4774 | ||
38f2e360 | 4775 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 4776 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4777 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
4778 | break; | |
4779 | ||
38f2e360 | 4780 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
4781 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
4782 | break; | |
4783 | ||
38f2e360 | 4784 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
4785 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
4786 | break; | |
4787 | ||
38f2e360 | 4788 | case OFPUTIL_OFPAT_SET_NW_TOS: |
c4f2731d PS |
4789 | /* OpenFlow 1.0 only supports IPv4. */ |
4790 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4791 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4792 | ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
4793 | } | |
abe529af BP |
4794 | break; |
4795 | ||
38f2e360 | 4796 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
4797 | ctx->flow.tp_src = ia->tp_port.tp_port; |
4798 | break; | |
4799 | ||
38f2e360 | 4800 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
4801 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
4802 | break; | |
4803 | ||
38f2e360 BP |
4804 | case OFPUTIL_OFPAT_ENQUEUE: |
4805 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
4806 | break; | |
4807 | ||
4808 | case OFPUTIL_NXAST_RESUBMIT: | |
4809 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
4810 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
4811 | break; | |
4812 | ||
4813 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
4814 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
4815 | break; |
4816 | ||
38f2e360 BP |
4817 | case OFPUTIL_NXAST_SET_TUNNEL: |
4818 | nast = (const struct nx_action_set_tunnel *) ia; | |
4819 | tun_id = htonll(ntohl(nast->tun_id)); | |
4820 | ctx->flow.tun_id = tun_id; | |
4821 | break; | |
4822 | ||
4823 | case OFPUTIL_NXAST_SET_QUEUE: | |
4824 | nasq = (const struct nx_action_set_queue *) ia; | |
4825 | xlate_set_queue_action(ctx, nasq); | |
4826 | break; | |
4827 | ||
4828 | case OFPUTIL_NXAST_POP_QUEUE: | |
deedf7e7 | 4829 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
4830 | break; |
4831 | ||
4832 | case OFPUTIL_NXAST_REG_MOVE: | |
4833 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
4834 | &ctx->flow); | |
4835 | break; | |
4836 | ||
4837 | case OFPUTIL_NXAST_REG_LOAD: | |
4838 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
4839 | &ctx->flow); | |
4840 | break; | |
4841 | ||
4842 | case OFPUTIL_NXAST_NOTE: | |
4843 | /* Nothing to do. */ | |
4844 | break; | |
4845 | ||
4846 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
4847 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
4848 | ctx->flow.tun_id = tun_id; | |
4849 | break; | |
4850 | ||
4851 | case OFPUTIL_NXAST_MULTIPATH: | |
4852 | nam = (const struct nx_action_multipath *) ia; | |
4853 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
4854 | break; |
4855 | ||
38f2e360 BP |
4856 | case OFPUTIL_NXAST_AUTOPATH: |
4857 | naa = (const struct nx_action_autopath *) ia; | |
4858 | xlate_autopath(ctx, naa); | |
abe529af | 4859 | break; |
daff3353 EJ |
4860 | |
4861 | case OFPUTIL_NXAST_BUNDLE: | |
4862 | ctx->ofproto->has_bundle_action = true; | |
4863 | nab = (const struct nx_action_bundle *) ia; | |
4864 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
4865 | slave_enabled_cb, | |
4866 | ctx->ofproto), 0); | |
4867 | break; | |
a368bb53 EJ |
4868 | |
4869 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
4870 | ctx->ofproto->has_bundle_action = true; | |
4871 | nab = (const struct nx_action_bundle *) ia; | |
4872 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
4873 | ctx->ofproto); | |
4874 | break; | |
f694937d EJ |
4875 | |
4876 | case OFPUTIL_NXAST_OUTPUT_REG: | |
4877 | naor = (const struct nx_action_output_reg *) ia; | |
4878 | xlate_output_reg_action(ctx, naor); | |
4879 | break; | |
75a75043 BP |
4880 | |
4881 | case OFPUTIL_NXAST_LEARN: | |
4882 | ctx->has_learn = true; | |
4883 | if (ctx->may_learn) { | |
4884 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); | |
4885 | } | |
4886 | break; | |
848e8809 | 4887 | |
f0fd1a17 PS |
4888 | case OFPUTIL_NXAST_DEC_TTL: |
4889 | if (compose_dec_ttl(ctx)) { | |
4890 | goto out; | |
4891 | } | |
4892 | break; | |
4893 | ||
848e8809 EJ |
4894 | case OFPUTIL_NXAST_EXIT: |
4895 | ctx->exit = true; | |
4896 | break; | |
abe529af BP |
4897 | } |
4898 | } | |
21f7563c | 4899 | |
f0fd1a17 | 4900 | out: |
21f7563c JP |
4901 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
4902 | * so drop it now if forwarding is disabled. */ | |
4903 | if (port && !stp_forward_in_state(port->stp_state)) { | |
4904 | ofpbuf_clear(ctx->odp_actions); | |
4905 | add_sflow_action(ctx); | |
4906 | } | |
abe529af BP |
4907 | } |
4908 | ||
4909 | static void | |
4910 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
4911 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
54834960 EJ |
4912 | ovs_be16 initial_tci, ovs_be64 cookie, |
4913 | const struct ofpbuf *packet) | |
abe529af BP |
4914 | { |
4915 | ctx->ofproto = ofproto; | |
4916 | ctx->flow = *flow; | |
e84173dc BP |
4917 | ctx->base_flow = ctx->flow; |
4918 | ctx->base_flow.tun_id = 0; | |
4919 | ctx->base_flow.vlan_tci = initial_tci; | |
54834960 | 4920 | ctx->cookie = cookie; |
abe529af | 4921 | ctx->packet = packet; |
75a75043 | 4922 | ctx->may_learn = packet != NULL; |
abe529af | 4923 | ctx->resubmit_hook = NULL; |
abe529af BP |
4924 | } |
4925 | ||
4926 | static struct ofpbuf * | |
4927 | xlate_actions(struct action_xlate_ctx *ctx, | |
4928 | const union ofp_action *in, size_t n_in) | |
4929 | { | |
c06bba01 JP |
4930 | struct flow orig_flow = ctx->flow; |
4931 | ||
abe529af BP |
4932 | COVERAGE_INC(ofproto_dpif_xlate); |
4933 | ||
4934 | ctx->odp_actions = ofpbuf_new(512); | |
b6848f13 | 4935 | ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); |
97e42c92 BP |
4936 | ctx->tags = 0; |
4937 | ctx->may_set_up_flow = true; | |
4938 | ctx->has_learn = false; | |
4939 | ctx->has_normal = false; | |
4940 | ctx->nf_output_iface = NF_OUT_DROP; | |
9d24de3b | 4941 | ctx->mirrors = 0; |
97e42c92 | 4942 | ctx->recurse = 0; |
deedf7e7 | 4943 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 4944 | ctx->table_id = 0; |
848e8809 | 4945 | ctx->exit = false; |
7257b535 | 4946 | |
eadef313 | 4947 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
4948 | switch (ctx->ofproto->up.frag_handling) { |
4949 | case OFPC_FRAG_NORMAL: | |
4950 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
4951 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
4952 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
4953 | break; |
4954 | ||
4955 | case OFPC_FRAG_DROP: | |
4956 | return ctx->odp_actions; | |
4957 | ||
4958 | case OFPC_FRAG_REASM: | |
4959 | NOT_REACHED(); | |
4960 | ||
4961 | case OFPC_FRAG_NX_MATCH: | |
4962 | /* Nothing to do. */ | |
4963 | break; | |
f0fd1a17 PS |
4964 | |
4965 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
4966 | NOT_REACHED(); | |
7257b535 BP |
4967 | } |
4968 | } | |
4969 | ||
fc08b7a2 | 4970 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af | 4971 | ctx->may_set_up_flow = false; |
b6848f13 | 4972 | return ctx->odp_actions; |
abe529af | 4973 | } else { |
6ff686f2 | 4974 | add_sflow_action(ctx); |
abe529af | 4975 | do_xlate_actions(in, n_in, ctx); |
abe529af | 4976 | |
b6848f13 BP |
4977 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
4978 | ctx->odp_actions->data, | |
4979 | ctx->odp_actions->size)) { | |
4980 | ctx->may_set_up_flow = false; | |
4981 | if (ctx->packet | |
4982 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
4983 | ctx->packet)) { | |
5e48dc2b | 4984 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
4985 | } |
4986 | } | |
c06bba01 | 4987 | add_mirror_actions(ctx, &orig_flow); |
a7c4eaf6 | 4988 | fix_sflow_action(ctx); |
abe529af BP |
4989 | } |
4990 | ||
4991 | return ctx->odp_actions; | |
4992 | } | |
4993 | \f | |
4994 | /* OFPP_NORMAL implementation. */ | |
4995 | ||
abe529af BP |
4996 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
4997 | ||
ecac4ebf BP |
4998 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
4999 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5000 | * the bundle on which the packet was received, returns the VLAN to which the | |
5001 | * packet belongs. | |
5002 | * | |
5003 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5004 | static uint16_t | |
5005 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5006 | { | |
5007 | switch (in_bundle->vlan_mode) { | |
5008 | case PORT_VLAN_ACCESS: | |
5009 | return in_bundle->vlan; | |
5010 | break; | |
5011 | ||
5012 | case PORT_VLAN_TRUNK: | |
5013 | return vid; | |
5014 | ||
5015 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5016 | case PORT_VLAN_NATIVE_TAGGED: | |
5017 | return vid ? vid : in_bundle->vlan; | |
5018 | ||
5019 | default: | |
5020 | NOT_REACHED(); | |
5021 | } | |
5022 | } | |
5023 | ||
5da5ec37 BP |
5024 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5025 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5026 | * a warning. | |
5027 | * | |
5028 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5029 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5030 | * 0...4095. */ | |
5031 | static bool | |
5032 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5033 | { | |
33158a18 JP |
5034 | /* Allow any VID on the OFPP_NONE port. */ |
5035 | if (in_bundle == &ofpp_none_bundle) { | |
5036 | return true; | |
5037 | } | |
5038 | ||
5da5ec37 BP |
5039 | switch (in_bundle->vlan_mode) { |
5040 | case PORT_VLAN_ACCESS: | |
5041 | if (vid) { | |
5042 | if (warn) { | |
5043 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5044 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5045 | "packet received on port %s configured as VLAN " | |
5046 | "%"PRIu16" access port", | |
5047 | in_bundle->ofproto->up.name, vid, | |
5048 | in_bundle->name, in_bundle->vlan); | |
5049 | } | |
5050 | return false; | |
5051 | } | |
5052 | return true; | |
5053 | ||
5054 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5055 | case PORT_VLAN_NATIVE_TAGGED: | |
5056 | if (!vid) { | |
5057 | /* Port must always carry its native VLAN. */ | |
5058 | return true; | |
5059 | } | |
5060 | /* Fall through. */ | |
5061 | case PORT_VLAN_TRUNK: | |
5062 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5063 | if (warn) { | |
5064 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5065 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5066 | "received on port %s not configured for trunking " | |
5067 | "VLAN %"PRIu16, | |
5068 | in_bundle->ofproto->up.name, vid, | |
5069 | in_bundle->name, vid); | |
5070 | } | |
5071 | return false; | |
5072 | } | |
5073 | return true; | |
5074 | ||
5075 | default: | |
5076 | NOT_REACHED(); | |
5077 | } | |
5078 | ||
5079 | } | |
5080 | ||
ecac4ebf BP |
5081 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5082 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5083 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5084 | * then the 802.1Q header should only be included in the packet if there is a | |
5085 | * nonzero PCP.) | |
5086 | * | |
5087 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5088 | static uint16_t | |
5089 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5090 | { | |
5091 | switch (out_bundle->vlan_mode) { | |
5092 | case PORT_VLAN_ACCESS: | |
5093 | return 0; | |
5094 | ||
5095 | case PORT_VLAN_TRUNK: | |
5096 | case PORT_VLAN_NATIVE_TAGGED: | |
5097 | return vlan; | |
5098 | ||
5099 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5100 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5101 | ||
5102 | default: | |
5103 | NOT_REACHED(); | |
5104 | } | |
5105 | } | |
5106 | ||
395e68ce BP |
5107 | static void |
5108 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5109 | uint16_t vlan) | |
abe529af | 5110 | { |
395e68ce BP |
5111 | struct ofport_dpif *port; |
5112 | uint16_t vid; | |
81b1afb1 | 5113 | ovs_be16 tci, old_tci; |
ecac4ebf | 5114 | |
395e68ce BP |
5115 | vid = output_vlan_to_vid(out_bundle, vlan); |
5116 | if (!out_bundle->bond) { | |
5117 | port = ofbundle_get_a_port(out_bundle); | |
5118 | } else { | |
5119 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5120 | vid, &ctx->tags); | |
5121 | if (!port) { | |
5122 | /* No slaves enabled, so drop packet. */ | |
5123 | return; | |
5124 | } | |
5125 | } | |
abe529af | 5126 | |
81b1afb1 | 5127 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5128 | tci = htons(vid); |
5129 | if (tci || out_bundle->use_priority_tags) { | |
5130 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5131 | if (tci) { | |
5132 | tci |= htons(VLAN_CFI); | |
5133 | } | |
395e68ce | 5134 | } |
81b1afb1 | 5135 | ctx->flow.vlan_tci = tci; |
395e68ce | 5136 | |
5e48dc2b | 5137 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5138 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5139 | } |
5140 | ||
5141 | static int | |
5142 | mirror_mask_ffs(mirror_mask_t mask) | |
5143 | { | |
5144 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5145 | return ffs(mask); | |
5146 | } | |
5147 | ||
abe529af BP |
5148 | static bool |
5149 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5150 | { | |
ecac4ebf | 5151 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5152 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5153 | } |
5154 | ||
5155 | static bool | |
5156 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5157 | { | |
5158 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5159 | } | |
5160 | ||
5161 | /* Returns an arbitrary interface within 'bundle'. */ | |
5162 | static struct ofport_dpif * | |
5163 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5164 | { | |
5165 | return CONTAINER_OF(list_front(&bundle->ports), | |
5166 | struct ofport_dpif, bundle_node); | |
5167 | } | |
5168 | ||
abe529af BP |
5169 | static bool |
5170 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5171 | { | |
fc3d7408 | 5172 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5173 | } |
5174 | ||
07817dfe BP |
5175 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
5176 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
5177 | * protocols that may confuse switches. */ | |
5178 | static bool | |
5179 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
5180 | { | |
5181 | /* If you change this function's behavior, please update corresponding | |
5182 | * documentation in vswitch.xml at the same time. */ | |
5183 | if (dst[0] != 0x01) { | |
5184 | /* All the currently banned MACs happen to start with 01 currently, so | |
5185 | * this is a quick way to eliminate most of the good ones. */ | |
5186 | } else { | |
5187 | if (eth_addr_is_reserved(dst)) { | |
5188 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
5189 | * (01-80-c2-00-00-0x). */ | |
5190 | return false; | |
5191 | } | |
5192 | ||
5193 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
5194 | /* Cisco OUI. */ | |
5195 | if ((dst[3] & 0xfe) == 0xcc && | |
5196 | (dst[4] & 0xfe) == 0xcc && | |
5197 | (dst[5] & 0xfe) == 0xcc) { | |
5198 | /* Drop the following protocols plus others following the same | |
5199 | pattern: | |
5200 | ||
5201 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
5202 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
5203 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
5204 | return false; | |
5205 | } | |
5206 | ||
5207 | if (!(dst[3] | dst[4] | dst[5])) { | |
5208 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
5209 | return false; | |
5210 | } | |
5211 | } | |
5212 | } | |
5213 | return true; | |
5214 | } | |
5215 | ||
abe529af | 5216 | static void |
c06bba01 | 5217 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5218 | { |
5219 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5220 | mirror_mask_t mirrors; | |
c06bba01 JP |
5221 | struct ofbundle *in_bundle; |
5222 | uint16_t vlan; | |
5223 | uint16_t vid; | |
5224 | const struct nlattr *a; | |
5225 | size_t left; | |
5226 | ||
3581c12c JP |
5227 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
5228 | ctx->packet != NULL); | |
5229 | if (!in_bundle) { | |
c06bba01 JP |
5230 | return; |
5231 | } | |
c06bba01 JP |
5232 | mirrors = in_bundle->src_mirrors; |
5233 | ||
5234 | /* Drop frames on bundles reserved for mirroring. */ | |
5235 | if (in_bundle->mirror_out) { | |
5236 | if (ctx->packet != NULL) { | |
5237 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5238 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5239 | "%s, which is reserved exclusively for mirroring", | |
5240 | ctx->ofproto->up.name, in_bundle->name); | |
5241 | } | |
5242 | return; | |
5243 | } | |
5244 | ||
5245 | /* Check VLAN. */ | |
5246 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5247 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5248 | return; | |
5249 | } | |
5250 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5251 | ||
5252 | /* Look at the output ports to check for destination selections. */ | |
5253 | ||
5254 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5255 | ctx->odp_actions->size) { | |
5256 | enum ovs_action_attr type = nl_attr_type(a); | |
5257 | struct ofport_dpif *ofport; | |
5258 | ||
5259 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5260 | continue; | |
5261 | } | |
5262 | ||
5263 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5264 | if (ofport && ofport->bundle) { |
5265 | mirrors |= ofport->bundle->dst_mirrors; | |
5266 | } | |
c06bba01 | 5267 | } |
abe529af BP |
5268 | |
5269 | if (!mirrors) { | |
5270 | return; | |
5271 | } | |
5272 | ||
c06bba01 JP |
5273 | /* Restore the original packet before adding the mirror actions. */ |
5274 | ctx->flow = *orig_flow; | |
5275 | ||
9ba15e2a BP |
5276 | while (mirrors) { |
5277 | struct ofmirror *m; | |
9ba15e2a BP |
5278 | |
5279 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5280 | ||
5281 | if (!vlan_is_mirrored(m, vlan)) { | |
5282 | mirrors &= mirrors - 1; | |
5283 | continue; | |
5284 | } | |
5285 | ||
5286 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 5287 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 5288 | if (m->out) { |
395e68ce | 5289 | output_normal(ctx, m->out, vlan); |
c06bba01 | 5290 | } else if (eth_dst_may_rspan(orig_flow->dl_dst) |
9ba15e2a BP |
5291 | && vlan != m->out_vlan) { |
5292 | struct ofbundle *bundle; | |
5293 | ||
5294 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
5295 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
5296 | && !bundle->mirror_out) { |
5297 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
5298 | } |
5299 | } | |
5300 | } | |
abe529af BP |
5301 | } |
5302 | } | |
5303 | ||
9d24de3b JP |
5304 | static void |
5305 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
5306 | uint64_t packets, uint64_t bytes) | |
5307 | { | |
5308 | if (!mirrors) { | |
5309 | return; | |
5310 | } | |
5311 | ||
5312 | for (; mirrors; mirrors &= mirrors - 1) { | |
5313 | struct ofmirror *m; | |
5314 | ||
5315 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5316 | ||
5317 | if (!m) { | |
5318 | /* In normal circumstances 'm' will not be NULL. However, | |
5319 | * if mirrors are reconfigured, we can temporarily get out | |
5320 | * of sync in facet_revalidate(). We could "correct" the | |
5321 | * mirror list before reaching here, but doing that would | |
5322 | * not properly account the traffic stats we've currently | |
5323 | * accumulated for previous mirror configuration. */ | |
5324 | continue; | |
5325 | } | |
5326 | ||
5327 | m->packet_count += packets; | |
5328 | m->byte_count += bytes; | |
5329 | } | |
5330 | } | |
5331 | ||
abe529af BP |
5332 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
5333 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
5334 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
5335 | static bool | |
5336 | is_gratuitous_arp(const struct flow *flow) | |
5337 | { | |
5338 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
5339 | && eth_addr_is_broadcast(flow->dl_dst) | |
5340 | && (flow->nw_proto == ARP_OP_REPLY | |
5341 | || (flow->nw_proto == ARP_OP_REQUEST | |
5342 | && flow->nw_src == flow->nw_dst))); | |
5343 | } | |
5344 | ||
5345 | static void | |
5346 | update_learning_table(struct ofproto_dpif *ofproto, | |
5347 | const struct flow *flow, int vlan, | |
5348 | struct ofbundle *in_bundle) | |
5349 | { | |
5350 | struct mac_entry *mac; | |
5351 | ||
33158a18 JP |
5352 | /* Don't learn the OFPP_NONE port. */ |
5353 | if (in_bundle == &ofpp_none_bundle) { | |
5354 | return; | |
5355 | } | |
5356 | ||
abe529af BP |
5357 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
5358 | return; | |
5359 | } | |
5360 | ||
5361 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
5362 | if (is_gratuitous_arp(flow)) { | |
5363 | /* We don't want to learn from gratuitous ARP packets that are | |
5364 | * reflected back over bond slaves so we lock the learning table. */ | |
5365 | if (!in_bundle->bond) { | |
5366 | mac_entry_set_grat_arp_lock(mac); | |
5367 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
5368 | return; | |
5369 | } | |
5370 | } | |
5371 | ||
5372 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
5373 | /* The log messages here could actually be useful in debugging, | |
5374 | * so keep the rate limit relatively high. */ | |
5375 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5376 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
5377 | "on port %s in VLAN %d", | |
5378 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
5379 | in_bundle->name, vlan); | |
5380 | ||
5381 | mac->port.p = in_bundle; | |
5382 | tag_set_add(&ofproto->revalidate_set, | |
5383 | mac_learning_changed(ofproto->ml, mac)); | |
5384 | } | |
5385 | } | |
5386 | ||
3581c12c | 5387 | static struct ofbundle * |
395e68ce BP |
5388 | lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) |
5389 | { | |
5390 | struct ofport_dpif *ofport; | |
5391 | ||
33158a18 JP |
5392 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
5393 | * port for traffic that it is sourcing. */ | |
5394 | if (in_port == OFPP_NONE) { | |
5395 | return &ofpp_none_bundle; | |
5396 | } | |
5397 | ||
395e68ce BP |
5398 | /* Find the port and bundle for the received packet. */ |
5399 | ofport = get_ofp_port(ofproto, in_port); | |
5400 | if (ofport && ofport->bundle) { | |
3581c12c | 5401 | return ofport->bundle; |
395e68ce BP |
5402 | } |
5403 | ||
5404 | /* Odd. A few possible reasons here: | |
5405 | * | |
5406 | * - We deleted a port but there are still a few packets queued up | |
5407 | * from it. | |
5408 | * | |
5409 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
5410 | * we don't know about. | |
5411 | * | |
5412 | * - The ofproto client didn't configure the port as part of a bundle. | |
5413 | */ | |
5414 | if (warn) { | |
5415 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5416 | ||
5417 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
5418 | "port %"PRIu16, ofproto->up.name, in_port); | |
5419 | } | |
5420 | return NULL; | |
5421 | } | |
5422 | ||
5da5ec37 | 5423 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
5424 | * dropped. Returns true if they may be forwarded, false if they should be |
5425 | * dropped. | |
5426 | * | |
395e68ce BP |
5427 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
5428 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 5429 | * |
395e68ce BP |
5430 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
5431 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
5432 | * checked by input_vid_is_valid(). | |
abe529af BP |
5433 | * |
5434 | * May also add tags to '*tags', although the current implementation only does | |
5435 | * so in one special case. | |
5436 | */ | |
5437 | static bool | |
5438 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
395e68ce | 5439 | struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) |
abe529af | 5440 | { |
395e68ce | 5441 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 5442 | |
395e68ce BP |
5443 | /* Drop frames for reserved multicast addresses |
5444 | * only if forward_bpdu option is absent. */ | |
21f7563c | 5445 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { |
abe529af BP |
5446 | return false; |
5447 | } | |
5448 | ||
abe529af BP |
5449 | if (in_bundle->bond) { |
5450 | struct mac_entry *mac; | |
5451 | ||
5452 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
5453 | flow->dl_dst, tags)) { | |
5454 | case BV_ACCEPT: | |
5455 | break; | |
5456 | ||
5457 | case BV_DROP: | |
5458 | return false; | |
5459 | ||
5460 | case BV_DROP_IF_MOVED: | |
5461 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
5462 | if (mac && mac->port.p != in_bundle && | |
5463 | (!is_gratuitous_arp(flow) | |
5464 | || mac_entry_is_grat_arp_locked(mac))) { | |
5465 | return false; | |
5466 | } | |
5467 | break; | |
5468 | } | |
5469 | } | |
5470 | ||
5471 | return true; | |
5472 | } | |
5473 | ||
4cd78906 | 5474 | static void |
abe529af BP |
5475 | xlate_normal(struct action_xlate_ctx *ctx) |
5476 | { | |
395e68ce | 5477 | struct ofport_dpif *in_port; |
abe529af | 5478 | struct ofbundle *in_bundle; |
abe529af | 5479 | struct mac_entry *mac; |
395e68ce BP |
5480 | uint16_t vlan; |
5481 | uint16_t vid; | |
abe529af | 5482 | |
75a75043 BP |
5483 | ctx->has_normal = true; |
5484 | ||
3581c12c | 5485 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
395e68ce | 5486 | ctx->packet != NULL); |
3581c12c | 5487 | if (!in_bundle) { |
395e68ce BP |
5488 | return; |
5489 | } | |
3581c12c | 5490 | |
33158a18 JP |
5491 | /* We know 'in_port' exists unless it is "ofpp_none_bundle", |
5492 | * since lookup_input_bundle() succeeded. */ | |
3581c12c | 5493 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
395e68ce BP |
5494 | |
5495 | /* Drop malformed frames. */ | |
5496 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
5497 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
5498 | if (ctx->packet != NULL) { | |
5499 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5500 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
5501 | "VLAN tag received on port %s", | |
5502 | ctx->ofproto->up.name, in_bundle->name); | |
5503 | } | |
5504 | return; | |
5505 | } | |
5506 | ||
5507 | /* Drop frames on bundles reserved for mirroring. */ | |
5508 | if (in_bundle->mirror_out) { | |
5509 | if (ctx->packet != NULL) { | |
5510 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5511 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5512 | "%s, which is reserved exclusively for mirroring", | |
5513 | ctx->ofproto->up.name, in_bundle->name); | |
5514 | } | |
5515 | return; | |
5516 | } | |
5517 | ||
5518 | /* Check VLAN. */ | |
5519 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
5520 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5521 | return; | |
5522 | } | |
5523 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5524 | ||
5525 | /* Check other admissibility requirements. */ | |
33158a18 JP |
5526 | if (in_port && |
5527 | !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { | |
395e68ce | 5528 | return; |
abe529af BP |
5529 | } |
5530 | ||
75a75043 BP |
5531 | /* Learn source MAC. */ |
5532 | if (ctx->may_learn) { | |
abe529af BP |
5533 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
5534 | } | |
5535 | ||
5536 | /* Determine output bundle. */ | |
5537 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
5538 | &ctx->tags); | |
5539 | if (mac) { | |
c06bba01 JP |
5540 | if (mac->port.p != in_bundle) { |
5541 | output_normal(ctx, mac->port.p, vlan); | |
5542 | } | |
abe529af | 5543 | } else { |
c06bba01 | 5544 | struct ofbundle *bundle; |
abe529af | 5545 | |
c06bba01 JP |
5546 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
5547 | if (bundle != in_bundle | |
5548 | && ofbundle_includes_vlan(bundle, vlan) | |
5549 | && bundle->floodable | |
5550 | && !bundle->mirror_out) { | |
5551 | output_normal(ctx, bundle, vlan); | |
5552 | } | |
5553 | } | |
5554 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 5555 | } |
abe529af BP |
5556 | } |
5557 | \f | |
54a9cbc9 BP |
5558 | /* Optimized flow revalidation. |
5559 | * | |
5560 | * It's a difficult problem, in general, to tell which facets need to have | |
5561 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
5562 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
5563 | * table changes, we recalculate the actions for every facet. This is | |
5564 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
5565 | * doesn't change very often. | |
5566 | * | |
5567 | * However, we can expect one particular kind of OpenFlow flow table change to | |
5568 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
5569 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
5570 | * table, we add a special case that applies to flow tables in which every rule | |
5571 | * has the same form (that is, the same wildcards), except that the table is | |
5572 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
5573 | * optimize this case by tagging all of the facets that resubmit into the table | |
5574 | * and invalidating the same tag whenever a flow changes in that table. The | |
5575 | * end result is that we revalidate just the facets that need it (and sometimes | |
5576 | * a few more, but not all of the facets or even all of the facets that | |
5577 | * resubmit to the table modified by MAC learning). */ | |
5578 | ||
5579 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
5580 | * into an OpenFlow table with the given 'basis'. */ | |
5581 | static uint32_t | |
5582 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
5583 | uint32_t secret) | |
5584 | { | |
5585 | if (flow_wildcards_is_catchall(wc)) { | |
5586 | return 0; | |
5587 | } else { | |
5588 | struct flow tag_flow = *flow; | |
5589 | flow_zero_wildcards(&tag_flow, wc); | |
5590 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
5591 | } | |
5592 | } | |
5593 | ||
5594 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
5595 | * taggability of that table. | |
5596 | * | |
5597 | * This function must be called after *each* change to a flow table. If you | |
5598 | * skip calling it on some changes then the pointer comparisons at the end can | |
5599 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
5600 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
5601 | * different wildcards to be created with the same address, then this function | |
5602 | * will incorrectly skip revalidation. */ | |
5603 | static void | |
5604 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
5605 | { | |
5606 | struct table_dpif *table = &ofproto->tables[table_id]; | |
5607 | const struct classifier *cls = &ofproto->up.tables[table_id]; | |
5608 | struct cls_table *catchall, *other; | |
5609 | struct cls_table *t; | |
5610 | ||
5611 | catchall = other = NULL; | |
5612 | ||
5613 | switch (hmap_count(&cls->tables)) { | |
5614 | case 0: | |
5615 | /* We could tag this OpenFlow table but it would make the logic a | |
5616 | * little harder and it's a corner case that doesn't seem worth it | |
5617 | * yet. */ | |
5618 | break; | |
5619 | ||
5620 | case 1: | |
5621 | case 2: | |
5622 | HMAP_FOR_EACH (t, hmap_node, &cls->tables) { | |
5623 | if (cls_table_is_catchall(t)) { | |
5624 | catchall = t; | |
5625 | } else if (!other) { | |
5626 | other = t; | |
5627 | } else { | |
5628 | /* Indicate that we can't tag this by setting both tables to | |
5629 | * NULL. (We know that 'catchall' is already NULL.) */ | |
5630 | other = NULL; | |
5631 | } | |
5632 | } | |
5633 | break; | |
5634 | ||
5635 | default: | |
5636 | /* Can't tag this table. */ | |
5637 | break; | |
5638 | } | |
5639 | ||
5640 | if (table->catchall_table != catchall || table->other_table != other) { | |
5641 | table->catchall_table = catchall; | |
5642 | table->other_table = other; | |
5643 | ofproto->need_revalidate = true; | |
5644 | } | |
5645 | } | |
5646 | ||
5647 | /* Given 'rule' that has changed in some way (either it is a rule being | |
5648 | * inserted, a rule being deleted, or a rule whose actions are being | |
5649 | * modified), marks facets for revalidation to ensure that packets will be | |
5650 | * forwarded correctly according to the new state of the flow table. | |
5651 | * | |
5652 | * This function must be called after *each* change to a flow table. See | |
5653 | * the comment on table_update_taggable() for more information. */ | |
5654 | static void | |
5655 | rule_invalidate(const struct rule_dpif *rule) | |
5656 | { | |
5657 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5658 | ||
5659 | table_update_taggable(ofproto, rule->up.table_id); | |
5660 | ||
5661 | if (!ofproto->need_revalidate) { | |
5662 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
5663 | ||
5664 | if (table->other_table && rule->tag) { | |
5665 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
5666 | } else { | |
5667 | ofproto->need_revalidate = true; | |
5668 | } | |
5669 | } | |
5670 | } | |
5671 | \f | |
abe529af | 5672 | static bool |
7257b535 BP |
5673 | set_frag_handling(struct ofproto *ofproto_, |
5674 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5675 | { |
5676 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 5677 | |
7257b535 BP |
5678 | if (frag_handling != OFPC_FRAG_REASM) { |
5679 | ofproto->need_revalidate = true; | |
5680 | return true; | |
5681 | } else { | |
5682 | return false; | |
5683 | } | |
abe529af BP |
5684 | } |
5685 | ||
90bf1e07 | 5686 | static enum ofperr |
abe529af BP |
5687 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
5688 | const struct flow *flow, | |
5689 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
5690 | { | |
5691 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
90bf1e07 | 5692 | enum ofperr error; |
abe529af | 5693 | |
e1154f71 | 5694 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
90bf1e07 | 5695 | return OFPERR_NXBRC_BAD_IN_PORT; |
e1154f71 BP |
5696 | } |
5697 | ||
abe529af BP |
5698 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
5699 | ofproto->max_ports); | |
5700 | if (!error) { | |
80e5eed9 | 5701 | struct odputil_keybuf keybuf; |
abe529af | 5702 | struct ofpbuf *odp_actions; |
2284188b | 5703 | struct ofproto_push push; |
80e5eed9 BP |
5704 | struct ofpbuf key; |
5705 | ||
5706 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
5707 | odp_flow_key_from_flow(&key, flow); | |
abe529af | 5708 | |
2284188b EJ |
5709 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, 0, |
5710 | packet); | |
5711 | ||
5712 | /* Ensure that resubmits in 'ofp_actions' get accounted to their | |
5713 | * matching rules. */ | |
5714 | push.packets = 1; | |
5715 | push.bytes = packet->size; | |
5716 | push.used = time_msec(); | |
5717 | push.ctx.resubmit_hook = push_resubmit; | |
5718 | ||
5719 | odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
5720 | dpif_execute(ofproto->dpif, key.data, key.size, |
5721 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
5722 | ofpbuf_delete(odp_actions); |
5723 | } | |
5724 | return error; | |
5725 | } | |
6fca1ffb BP |
5726 | \f |
5727 | /* NetFlow. */ | |
5728 | ||
5729 | static int | |
5730 | set_netflow(struct ofproto *ofproto_, | |
5731 | const struct netflow_options *netflow_options) | |
5732 | { | |
5733 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5734 | ||
5735 | if (netflow_options) { | |
5736 | if (!ofproto->netflow) { | |
5737 | ofproto->netflow = netflow_create(); | |
5738 | } | |
5739 | return netflow_set_options(ofproto->netflow, netflow_options); | |
5740 | } else { | |
5741 | netflow_destroy(ofproto->netflow); | |
5742 | ofproto->netflow = NULL; | |
5743 | return 0; | |
5744 | } | |
5745 | } | |
abe529af BP |
5746 | |
5747 | static void | |
5748 | get_netflow_ids(const struct ofproto *ofproto_, | |
5749 | uint8_t *engine_type, uint8_t *engine_id) | |
5750 | { | |
5751 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5752 | ||
5753 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
5754 | } | |
6fca1ffb BP |
5755 | |
5756 | static void | |
5757 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
5758 | { | |
5759 | if (!facet_is_controller_flow(facet) && | |
5760 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 5761 | struct subfacet *subfacet; |
6fca1ffb BP |
5762 | struct ofexpired expired; |
5763 | ||
b0f7b9b5 BP |
5764 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
5765 | if (subfacet->installed) { | |
5766 | struct dpif_flow_stats stats; | |
6fca1ffb | 5767 | |
15baa734 | 5768 | subfacet_install(subfacet, subfacet->actions, |
b95fc6ba | 5769 | subfacet->actions_len, &stats); |
15baa734 | 5770 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 5771 | } |
6fca1ffb BP |
5772 | } |
5773 | ||
5774 | expired.flow = facet->flow; | |
5775 | expired.packet_count = facet->packet_count; | |
5776 | expired.byte_count = facet->byte_count; | |
5777 | expired.used = facet->used; | |
5778 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
5779 | } | |
5780 | } | |
5781 | ||
5782 | static void | |
5783 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
5784 | { | |
5785 | struct facet *facet; | |
5786 | ||
5787 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
5788 | send_active_timeout(ofproto, facet); | |
5789 | } | |
5790 | } | |
abe529af BP |
5791 | \f |
5792 | static struct ofproto_dpif * | |
5793 | ofproto_dpif_lookup(const char *name) | |
5794 | { | |
b44a10b7 BP |
5795 | struct ofproto_dpif *ofproto; |
5796 | ||
5797 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
5798 | hash_string(name, 0), &all_ofproto_dpifs) { | |
5799 | if (!strcmp(ofproto->up.name, name)) { | |
5800 | return ofproto; | |
5801 | } | |
5802 | } | |
5803 | return NULL; | |
abe529af BP |
5804 | } |
5805 | ||
f0a3aa2e | 5806 | static void |
0e15264f BP |
5807 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5808 | const char *argv[], void *aux OVS_UNUSED) | |
f0a3aa2e AA |
5809 | { |
5810 | const struct ofproto_dpif *ofproto; | |
5811 | ||
0e15264f | 5812 | ofproto = ofproto_dpif_lookup(argv[1]); |
f0a3aa2e AA |
5813 | if (!ofproto) { |
5814 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5815 | return; | |
5816 | } | |
5817 | mac_learning_flush(ofproto->ml); | |
5818 | ||
5819 | unixctl_command_reply(conn, 200, "table successfully flushed"); | |
5820 | } | |
5821 | ||
abe529af | 5822 | static void |
0e15264f BP |
5823 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
5824 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
5825 | { |
5826 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5827 | const struct ofproto_dpif *ofproto; | |
5828 | const struct mac_entry *e; | |
5829 | ||
0e15264f | 5830 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af BP |
5831 | if (!ofproto) { |
5832 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5833 | return; | |
5834 | } | |
5835 | ||
5836 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
5837 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
5838 | struct ofbundle *bundle = e->port.p; | |
5839 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
5840 | ofbundle_get_a_port(bundle)->odp_port, | |
5841 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
5842 | } | |
5843 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
5844 | ds_destroy(&ds); | |
5845 | } | |
5846 | ||
5847 | struct ofproto_trace { | |
5848 | struct action_xlate_ctx ctx; | |
5849 | struct flow flow; | |
5850 | struct ds *result; | |
5851 | }; | |
5852 | ||
5853 | static void | |
29901626 BP |
5854 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
5855 | const struct rule_dpif *rule) | |
abe529af BP |
5856 | { |
5857 | ds_put_char_multiple(result, '\t', level); | |
5858 | if (!rule) { | |
5859 | ds_put_cstr(result, "No match\n"); | |
5860 | return; | |
5861 | } | |
5862 | ||
29901626 BP |
5863 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
5864 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 5865 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5866 | ds_put_char(result, '\n'); |
5867 | ||
5868 | ds_put_char_multiple(result, '\t', level); | |
5869 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 5870 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
5871 | ds_put_char(result, '\n'); |
5872 | } | |
5873 | ||
5874 | static void | |
5875 | trace_format_flow(struct ds *result, int level, const char *title, | |
5876 | struct ofproto_trace *trace) | |
5877 | { | |
5878 | ds_put_char_multiple(result, '\t', level); | |
5879 | ds_put_format(result, "%s: ", title); | |
5880 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5881 | ds_put_cstr(result, "unchanged"); | |
5882 | } else { | |
5883 | flow_format(result, &trace->ctx.flow); | |
5884 | trace->flow = trace->ctx.flow; | |
5885 | } | |
5886 | ds_put_char(result, '\n'); | |
5887 | } | |
5888 | ||
eb9e1c26 EJ |
5889 | static void |
5890 | trace_format_regs(struct ds *result, int level, const char *title, | |
5891 | struct ofproto_trace *trace) | |
5892 | { | |
5893 | size_t i; | |
5894 | ||
5895 | ds_put_char_multiple(result, '\t', level); | |
5896 | ds_put_format(result, "%s:", title); | |
5897 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5898 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5899 | } | |
5900 | ds_put_char(result, '\n'); | |
5901 | } | |
5902 | ||
abe529af BP |
5903 | static void |
5904 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
5905 | { | |
5906 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5907 | struct ds *result = trace->result; | |
5908 | ||
5909 | ds_put_char(result, '\n'); | |
5910 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 5911 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
29901626 | 5912 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
5913 | } |
5914 | ||
5915 | static void | |
0e15264f | 5916 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
5917 | void *aux OVS_UNUSED) |
5918 | { | |
0e15264f | 5919 | const char *dpname = argv[1]; |
abe529af | 5920 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
5921 | struct ofpbuf odp_key; |
5922 | struct ofpbuf *packet; | |
abe529af | 5923 | struct rule_dpif *rule; |
e84173dc | 5924 | ovs_be16 initial_tci; |
abe529af BP |
5925 | struct ds result; |
5926 | struct flow flow; | |
abe529af BP |
5927 | char *s; |
5928 | ||
876b0e1c BP |
5929 | packet = NULL; |
5930 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
5931 | ds_init(&result); |
5932 | ||
e84173dc BP |
5933 | ofproto = ofproto_dpif_lookup(dpname); |
5934 | if (!ofproto) { | |
5935 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5936 | "for help)"); | |
5937 | goto exit; | |
5938 | } | |
0e15264f | 5939 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 5940 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
5941 | const char *flow_s = argv[2]; |
5942 | const char *generate_s = argv[3]; | |
876b0e1c BP |
5943 | int error; |
5944 | ||
df2c07f4 | 5945 | /* Convert string to datapath key. */ |
876b0e1c | 5946 | ofpbuf_init(&odp_key, 0); |
0e15264f | 5947 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); |
876b0e1c BP |
5948 | if (error) { |
5949 | unixctl_command_reply(conn, 501, "Bad flow syntax"); | |
5950 | goto exit; | |
5951 | } | |
5952 | ||
5953 | /* Convert odp_key to flow. */ | |
e84173dc BP |
5954 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, |
5955 | odp_key.size, &flow, | |
e2a6ca36 | 5956 | &initial_tci, NULL); |
e84173dc | 5957 | if (error == ODP_FIT_ERROR) { |
876b0e1c BP |
5958 | unixctl_command_reply(conn, 501, "Invalid flow"); |
5959 | goto exit; | |
5960 | } | |
8b3b8dd1 BP |
5961 | |
5962 | /* Generate a packet, if requested. */ | |
0e15264f | 5963 | if (generate_s) { |
8b3b8dd1 BP |
5964 | packet = ofpbuf_new(0); |
5965 | flow_compose(packet, &flow); | |
5966 | } | |
0e15264f | 5967 | } else if (argc == 6) { |
abff858b | 5968 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
5969 | const char *priority_s = argv[2]; |
5970 | const char *tun_id_s = argv[3]; | |
5971 | const char *in_port_s = argv[4]; | |
5972 | const char *packet_s = argv[5]; | |
5973 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
5974 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
5975 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 5976 | const char *msg; |
0e15264f | 5977 | |
e22f1753 BP |
5978 | msg = eth_from_hex(packet_s, &packet); |
5979 | if (msg) { | |
5980 | unixctl_command_reply(conn, 501, msg); | |
876b0e1c BP |
5981 | goto exit; |
5982 | } | |
5983 | ||
5984 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 5985 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
5986 | ds_put_cstr(&result, s); |
5987 | free(s); | |
5988 | ||
abff858b | 5989 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 5990 | initial_tci = flow.vlan_tci; |
876b0e1c | 5991 | } else { |
abe529af BP |
5992 | unixctl_command_reply(conn, 501, "Bad command syntax"); |
5993 | goto exit; | |
5994 | } | |
5995 | ||
abe529af BP |
5996 | ds_put_cstr(&result, "Flow: "); |
5997 | flow_format(&result, &flow); | |
5998 | ds_put_char(&result, '\n'); | |
5999 | ||
29901626 BP |
6000 | rule = rule_dpif_lookup(ofproto, &flow, 0); |
6001 | trace_format_rule(&result, 0, 0, rule); | |
abe529af BP |
6002 | if (rule) { |
6003 | struct ofproto_trace trace; | |
6004 | struct ofpbuf *odp_actions; | |
6005 | ||
6006 | trace.result = &result; | |
6007 | trace.flow = flow; | |
54834960 EJ |
6008 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci, |
6009 | rule->up.flow_cookie, packet); | |
abe529af BP |
6010 | trace.ctx.resubmit_hook = trace_resubmit; |
6011 | odp_actions = xlate_actions(&trace.ctx, | |
6012 | rule->up.actions, rule->up.n_actions); | |
6013 | ||
6014 | ds_put_char(&result, '\n'); | |
6015 | trace_format_flow(&result, 0, "Final flow", &trace); | |
6016 | ds_put_cstr(&result, "Datapath actions: "); | |
6017 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
6018 | ofpbuf_delete(odp_actions); | |
876b0e1c BP |
6019 | |
6020 | if (!trace.ctx.may_set_up_flow) { | |
6021 | if (packet) { | |
6022 | ds_put_cstr(&result, "\nThis flow is not cachable."); | |
6023 | } else { | |
6024 | ds_put_cstr(&result, "\nThe datapath actions are incomplete--" | |
6025 | "for complete actions, please supply a packet."); | |
6026 | } | |
6027 | } | |
abe529af BP |
6028 | } |
6029 | ||
6030 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
6031 | ||
6032 | exit: | |
6033 | ds_destroy(&result); | |
876b0e1c BP |
6034 | ofpbuf_delete(packet); |
6035 | ofpbuf_uninit(&odp_key); | |
abe529af BP |
6036 | } |
6037 | ||
7ee20df1 | 6038 | static void |
0e15264f BP |
6039 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6040 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6041 | { |
6042 | clogged = true; | |
6043 | unixctl_command_reply(conn, 200, NULL); | |
6044 | } | |
6045 | ||
6046 | static void | |
0e15264f BP |
6047 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6048 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6049 | { |
6050 | clogged = false; | |
6051 | unixctl_command_reply(conn, 200, NULL); | |
6052 | } | |
6053 | ||
6814e51f BP |
6054 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6055 | * 'reply' describing the results. */ | |
6056 | static void | |
6057 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6058 | { | |
6059 | struct facet *facet; | |
6060 | int errors; | |
6061 | ||
6062 | errors = 0; | |
6063 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6064 | if (!facet_check_consistency(facet)) { | |
6065 | errors++; | |
6066 | } | |
6067 | } | |
6068 | if (errors) { | |
6069 | ofproto->need_revalidate = true; | |
6070 | } | |
6071 | ||
6072 | if (errors) { | |
6073 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6074 | ofproto->up.name, errors); | |
6075 | } else { | |
6076 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6077 | } | |
6078 | } | |
6079 | ||
6080 | static void | |
6081 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6082 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6083 | { | |
6084 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6085 | struct ofproto_dpif *ofproto; | |
6086 | ||
6087 | if (argc > 1) { | |
6088 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6089 | if (!ofproto) { | |
6090 | unixctl_command_reply(conn, 501, "Unknown ofproto (use " | |
6091 | "ofproto/list for help)"); | |
6092 | return; | |
6093 | } | |
6094 | ofproto_dpif_self_check__(ofproto, &reply); | |
6095 | } else { | |
6096 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6097 | ofproto_dpif_self_check__(ofproto, &reply); | |
6098 | } | |
6099 | } | |
6100 | ||
6101 | unixctl_command_reply(conn, 200, ds_cstr(&reply)); | |
6102 | ds_destroy(&reply); | |
6103 | } | |
6104 | ||
abe529af BP |
6105 | static void |
6106 | ofproto_dpif_unixctl_init(void) | |
6107 | { | |
6108 | static bool registered; | |
6109 | if (registered) { | |
6110 | return; | |
6111 | } | |
6112 | registered = true; | |
6113 | ||
0e15264f BP |
6114 | unixctl_command_register( |
6115 | "ofproto/trace", | |
6116 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
6117 | 2, 4, ofproto_unixctl_trace, NULL); | |
6118 | unixctl_command_register("fdb/flush", "bridge", 1, 1, | |
6119 | ofproto_unixctl_fdb_flush, NULL); | |
6120 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6121 | ofproto_unixctl_fdb_show, NULL); | |
6122 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6123 | ofproto_dpif_clog, NULL); | |
6124 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6125 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6126 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6127 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6128 | } |
6129 | \f | |
52a90c29 BP |
6130 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6131 | * | |
6132 | * This is deprecated. It is only for compatibility with broken device drivers | |
6133 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6134 | * devices are not used. When broken device drivers are no longer in | |
6135 | * widespread use, we will delete these interfaces. */ | |
6136 | ||
6137 | static int | |
6138 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6139 | { | |
6140 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6141 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6142 | ||
6143 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6144 | && vid == ofport->vlandev_vid) { | |
6145 | return 0; | |
6146 | } | |
6147 | ||
6148 | ofproto->need_revalidate = true; | |
6149 | ||
6150 | if (ofport->realdev_ofp_port) { | |
6151 | vsp_remove(ofport); | |
6152 | } | |
6153 | if (realdev_ofp_port && ofport->bundle) { | |
6154 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6155 | * themselves be part of a bundle. */ | |
6156 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6157 | } | |
6158 | ||
6159 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6160 | ofport->vlandev_vid = vid; | |
6161 | ||
6162 | if (realdev_ofp_port) { | |
6163 | vsp_add(ofport, realdev_ofp_port, vid); | |
6164 | } | |
6165 | ||
6166 | return 0; | |
6167 | } | |
6168 | ||
6169 | static uint32_t | |
6170 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
6171 | { | |
6172 | return hash_2words(realdev_ofp_port, vid); | |
6173 | } | |
6174 | ||
6175 | static uint32_t | |
6176 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6177 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
6178 | { | |
6179 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
6180 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
6181 | int vid = vlan_tci_to_vid(vlan_tci); | |
6182 | const struct vlan_splinter *vsp; | |
6183 | ||
6184 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6185 | hash_realdev_vid(realdev_ofp_port, vid), | |
6186 | &ofproto->realdev_vid_map) { | |
6187 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6188 | && vsp->vid == vid) { | |
6189 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
6190 | } | |
6191 | } | |
6192 | } | |
6193 | return realdev_odp_port; | |
6194 | } | |
6195 | ||
6196 | static struct vlan_splinter * | |
6197 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
6198 | { | |
6199 | struct vlan_splinter *vsp; | |
6200 | ||
6201 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
6202 | &ofproto->vlandev_map) { | |
6203 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6204 | return vsp; | |
6205 | } | |
6206 | } | |
6207 | ||
6208 | return NULL; | |
6209 | } | |
6210 | ||
6211 | static uint16_t | |
6212 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
6213 | uint16_t vlandev_ofp_port, int *vid) | |
6214 | { | |
6215 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6216 | const struct vlan_splinter *vsp; | |
6217 | ||
6218 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6219 | if (vsp) { | |
6220 | if (vid) { | |
6221 | *vid = vsp->vid; | |
6222 | } | |
6223 | return vsp->realdev_ofp_port; | |
6224 | } | |
6225 | } | |
6226 | return 0; | |
6227 | } | |
6228 | ||
6229 | static void | |
6230 | vsp_remove(struct ofport_dpif *port) | |
6231 | { | |
6232 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6233 | struct vlan_splinter *vsp; | |
6234 | ||
6235 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
6236 | if (vsp) { | |
6237 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6238 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6239 | free(vsp); | |
6240 | ||
6241 | port->realdev_ofp_port = 0; | |
6242 | } else { | |
6243 | VLOG_ERR("missing vlan device record"); | |
6244 | } | |
6245 | } | |
6246 | ||
6247 | static void | |
6248 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
6249 | { | |
6250 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6251 | ||
6252 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
6253 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
6254 | == realdev_ofp_port)) { | |
6255 | struct vlan_splinter *vsp; | |
6256 | ||
6257 | vsp = xmalloc(sizeof *vsp); | |
6258 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6259 | hash_int(port->up.ofp_port, 0)); | |
6260 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6261 | hash_realdev_vid(realdev_ofp_port, vid)); | |
6262 | vsp->realdev_ofp_port = realdev_ofp_port; | |
6263 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6264 | vsp->vid = vid; | |
6265 | ||
6266 | port->realdev_ofp_port = realdev_ofp_port; | |
6267 | } else { | |
6268 | VLOG_ERR("duplicate vlan device record"); | |
6269 | } | |
6270 | } | |
6271 | \f | |
abe529af BP |
6272 | const struct ofproto_class ofproto_dpif_class = { |
6273 | enumerate_types, | |
6274 | enumerate_names, | |
6275 | del, | |
6276 | alloc, | |
6277 | construct, | |
6278 | destruct, | |
6279 | dealloc, | |
6280 | run, | |
5fcc0d00 | 6281 | run_fast, |
abe529af BP |
6282 | wait, |
6283 | flush, | |
6c1491fb BP |
6284 | get_features, |
6285 | get_tables, | |
abe529af BP |
6286 | port_alloc, |
6287 | port_construct, | |
6288 | port_destruct, | |
6289 | port_dealloc, | |
6290 | port_modified, | |
6291 | port_reconfigured, | |
6292 | port_query_by_name, | |
6293 | port_add, | |
6294 | port_del, | |
6527c598 | 6295 | port_get_stats, |
abe529af BP |
6296 | port_dump_start, |
6297 | port_dump_next, | |
6298 | port_dump_done, | |
6299 | port_poll, | |
6300 | port_poll_wait, | |
6301 | port_is_lacp_current, | |
0ab6decf | 6302 | NULL, /* rule_choose_table */ |
abe529af BP |
6303 | rule_alloc, |
6304 | rule_construct, | |
6305 | rule_destruct, | |
6306 | rule_dealloc, | |
abe529af BP |
6307 | rule_get_stats, |
6308 | rule_execute, | |
6309 | rule_modify_actions, | |
7257b535 | 6310 | set_frag_handling, |
abe529af BP |
6311 | packet_out, |
6312 | set_netflow, | |
6313 | get_netflow_ids, | |
6314 | set_sflow, | |
6315 | set_cfm, | |
a5610457 | 6316 | get_cfm_fault, |
1de11730 | 6317 | get_cfm_remote_mpids, |
21f7563c JP |
6318 | set_stp, |
6319 | get_stp_status, | |
6320 | set_stp_port, | |
6321 | get_stp_port_status, | |
8b36f51e | 6322 | set_queues, |
abe529af BP |
6323 | bundle_set, |
6324 | bundle_remove, | |
6325 | mirror_set, | |
9d24de3b | 6326 | mirror_get_stats, |
abe529af BP |
6327 | set_flood_vlans, |
6328 | is_mirror_output_bundle, | |
8402c74b | 6329 | forward_bpdu_changed, |
52a90c29 | 6330 | set_realdev, |
abe529af | 6331 | }; |