]>
Commit | Line | Data |
---|---|---|
abe529af | 1 | /* |
4dd1e3ca | 2 | * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks. |
abe529af BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
23 | #include "autopath.h" | |
24 | #include "bond.h" | |
daff3353 | 25 | #include "bundle.h" |
abe529af BP |
26 | #include "byte-order.h" |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
75a75043 | 35 | #include "learn.h" |
abe529af | 36 | #include "mac-learning.h" |
816fd533 | 37 | #include "meta-flow.h" |
abe529af BP |
38 | #include "multipath.h" |
39 | #include "netdev.h" | |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
45 | #include "ofp-print.h" | |
bae473fe | 46 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
47 | #include "poll-loop.h" |
48 | #include "timer.h" | |
6c1491fb | 49 | #include "unaligned.h" |
abe529af BP |
50 | #include "unixctl.h" |
51 | #include "vlan-bitmap.h" | |
52 | #include "vlog.h" | |
53 | ||
54 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
55 | ||
56 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
57 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
58 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
59 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
60 | COVERAGE_DEFINE(facet_changed_rule); | |
61 | COVERAGE_DEFINE(facet_invalidated); | |
62 | COVERAGE_DEFINE(facet_revalidate); | |
63 | COVERAGE_DEFINE(facet_unexpected); | |
64 | ||
29901626 | 65 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af | 66 | * flow translation. */ |
642a5c05 | 67 | #define MAX_RESUBMIT_RECURSION 32 |
abe529af | 68 | |
9cdaaebe BP |
69 | /* Number of implemented OpenFlow tables. */ |
70 | enum { N_TABLES = 255 }; | |
71 | BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255); | |
72 | ||
abe529af BP |
73 | struct ofport_dpif; |
74 | struct ofproto_dpif; | |
75 | ||
76 | struct rule_dpif { | |
77 | struct rule up; | |
78 | ||
abe529af BP |
79 | /* These statistics: |
80 | * | |
81 | * - Do include packets and bytes from facets that have been deleted or | |
82 | * whose own statistics have been folded into the rule. | |
83 | * | |
84 | * - Do include packets and bytes sent "by hand" that were accounted to | |
85 | * the rule without any facet being involved (this is a rare corner | |
86 | * case in rule_execute()). | |
87 | * | |
88 | * - Do not include packet or bytes that can be obtained from any facet's | |
89 | * packet_count or byte_count member or that can be obtained from the | |
b0f7b9b5 | 90 | * datapath by, e.g., dpif_flow_get() for any subfacet. |
abe529af BP |
91 | */ |
92 | uint64_t packet_count; /* Number of packets received. */ | |
93 | uint64_t byte_count; /* Number of bytes received. */ | |
94 | ||
54a9cbc9 BP |
95 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
96 | ||
abe529af BP |
97 | struct list facets; /* List of "struct facet"s. */ |
98 | }; | |
99 | ||
100 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
101 | { | |
102 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
103 | } | |
104 | ||
29901626 BP |
105 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
106 | const struct flow *, uint8_t table); | |
abe529af | 107 | |
18b2a258 | 108 | static void flow_push_stats(struct rule_dpif *, const struct flow *, |
b0f7b9b5 BP |
109 | uint64_t packets, uint64_t bytes, |
110 | long long int used); | |
111 | ||
822d9414 | 112 | static tag_type rule_calculate_tag(const struct flow *, |
b0f7b9b5 BP |
113 | const struct flow_wildcards *, |
114 | uint32_t basis); | |
115 | static void rule_invalidate(const struct rule_dpif *); | |
116 | ||
abe529af BP |
117 | #define MAX_MIRRORS 32 |
118 | typedef uint32_t mirror_mask_t; | |
119 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
120 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
121 | struct ofmirror { | |
122 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
123 | size_t idx; /* In ofproto's "mirrors" array. */ | |
124 | void *aux; /* Key supplied by ofproto's client. */ | |
125 | char *name; /* Identifier for log messages. */ | |
126 | ||
127 | /* Selection criteria. */ | |
128 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
129 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
130 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
131 | ||
9ba15e2a | 132 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
133 | struct ofbundle *out; /* Output port or NULL. */ |
134 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 135 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
9d24de3b JP |
136 | |
137 | /* Counters. */ | |
138 | int64_t packet_count; /* Number of packets sent. */ | |
139 | int64_t byte_count; /* Number of bytes sent. */ | |
abe529af BP |
140 | }; |
141 | ||
142 | static void mirror_destroy(struct ofmirror *); | |
9d24de3b JP |
143 | static void update_mirror_stats(struct ofproto_dpif *ofproto, |
144 | mirror_mask_t mirrors, | |
145 | uint64_t packets, uint64_t bytes); | |
abe529af | 146 | |
abe529af | 147 | struct ofbundle { |
abe529af | 148 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ |
6e492d81 | 149 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ |
abe529af BP |
150 | void *aux; /* Key supplied by ofproto's client. */ |
151 | char *name; /* Identifier for log messages. */ | |
152 | ||
153 | /* Configuration. */ | |
154 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 155 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
156 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
157 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
158 | * NULL if all VLANs are trunked. */ | |
159 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
160 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 161 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
162 | |
163 | /* Status. */ | |
9e1fd49b | 164 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ |
abe529af BP |
165 | |
166 | /* Port mirroring info. */ | |
167 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
168 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
169 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
170 | }; | |
171 | ||
172 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 173 | static void bundle_update(struct ofbundle *); |
abe529af BP |
174 | static void bundle_destroy(struct ofbundle *); |
175 | static void bundle_del_port(struct ofport_dpif *); | |
176 | static void bundle_run(struct ofbundle *); | |
177 | static void bundle_wait(struct ofbundle *); | |
3581c12c JP |
178 | static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *, |
179 | uint16_t in_port, bool warn); | |
abe529af | 180 | |
33158a18 JP |
181 | /* A controller may use OFPP_NONE as the ingress port to indicate that |
182 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
183 | * when an input bundle is needed for validation (e.g., mirroring or | |
184 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
185 | * any 'port' structs, so care must be taken when dealing with it. */ | |
186 | static struct ofbundle ofpp_none_bundle = { | |
187 | .name = "OFPP_NONE", | |
188 | .vlan_mode = PORT_VLAN_TRUNK | |
189 | }; | |
190 | ||
21f7563c JP |
191 | static void stp_run(struct ofproto_dpif *ofproto); |
192 | static void stp_wait(struct ofproto_dpif *ofproto); | |
851bf71d EJ |
193 | static int set_stp_port(struct ofport *, |
194 | const struct ofproto_port_stp_settings *); | |
21f7563c | 195 | |
5da5ec37 BP |
196 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
197 | ||
abe529af BP |
198 | struct action_xlate_ctx { |
199 | /* action_xlate_ctx_init() initializes these members. */ | |
200 | ||
201 | /* The ofproto. */ | |
202 | struct ofproto_dpif *ofproto; | |
203 | ||
204 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
205 | * this flow when actions change header fields. */ | |
206 | struct flow flow; | |
207 | ||
208 | /* The packet corresponding to 'flow', or a null pointer if we are | |
209 | * revalidating without a packet to refer to. */ | |
210 | const struct ofpbuf *packet; | |
211 | ||
3de9590b BP |
212 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" |
213 | * actions update the flow table? | |
214 | * | |
215 | * We want to update these tables if we are actually processing a packet, | |
216 | * or if we are accounting for packets that the datapath has processed, but | |
217 | * not if we are just revalidating. */ | |
218 | bool may_learn; | |
75a75043 | 219 | |
18b2a258 BP |
220 | /* The rule that we are currently translating, or NULL. */ |
221 | struct rule_dpif *rule; | |
54834960 | 222 | |
0e553d9c BP |
223 | /* Union of the set of TCP flags seen so far in this flow. (Used only by |
224 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
225 | * timeouts.) */ | |
226 | uint8_t tcp_flags; | |
227 | ||
6a6455e5 EJ |
228 | /* If nonnull, called just before executing a resubmit action. In |
229 | * addition, disables logging of traces when the recursion depth is | |
230 | * exceeded. | |
abe529af BP |
231 | * |
232 | * This is normally null so the client has to set it manually after | |
233 | * calling action_xlate_ctx_init(). */ | |
234 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
235 | ||
abe529af BP |
236 | /* xlate_actions() initializes and uses these members. The client might want |
237 | * to look at them after it returns. */ | |
238 | ||
239 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 240 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
241 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
242 | * be reassessed for every packet. */ | |
75a75043 BP |
243 | bool has_learn; /* Actions include NXAST_LEARN? */ |
244 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 245 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
abe529af | 246 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
9d24de3b | 247 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ |
abe529af BP |
248 | |
249 | /* xlate_actions() initializes and uses these members, but the client has no | |
250 | * reason to look at them. */ | |
251 | ||
252 | int recurse; /* Recursion level, via xlate_table_action. */ | |
6a6455e5 | 253 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ |
b3e9b2ed | 254 | struct flow base_flow; /* Flow at the last commit. */ |
deedf7e7 | 255 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ |
29901626 | 256 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
257 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
258 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
259 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 260 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
261 | }; |
262 | ||
263 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
264 | struct ofproto_dpif *, const struct flow *, | |
18b2a258 | 265 | ovs_be16 initial_tci, struct rule_dpif *, |
0e553d9c | 266 | uint8_t tcp_flags, const struct ofpbuf *); |
050ac423 BP |
267 | static void xlate_actions(struct action_xlate_ctx *, |
268 | const union ofp_action *in, size_t n_in, | |
269 | struct ofpbuf *odp_actions); | |
270 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
271 | const union ofp_action *in, | |
272 | size_t n_in); | |
abe529af | 273 | |
5f5fbd17 BP |
274 | /* A dpif flow and actions associated with a facet. |
275 | * | |
276 | * See also the large comment on struct facet. */ | |
277 | struct subfacet { | |
278 | /* Owners. */ | |
279 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
280 | struct list list_node; /* In struct facet's 'facets' list. */ | |
281 | struct facet *facet; /* Owning facet. */ | |
282 | ||
283 | /* Key. | |
284 | * | |
285 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
286 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
287 | * regenerate the ODP flow key from ->facet->flow. */ | |
288 | enum odp_key_fitness key_fitness; | |
289 | struct nlattr *key; | |
290 | int key_len; | |
291 | ||
292 | long long int used; /* Time last used; time created if not used. */ | |
293 | ||
294 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
295 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
296 | ||
297 | /* Datapath actions. | |
298 | * | |
299 | * These should be essentially identical for every subfacet in a facet, but | |
300 | * may differ in trivial ways due to VLAN splinters. */ | |
301 | size_t actions_len; /* Number of bytes in actions[]. */ | |
302 | struct nlattr *actions; /* Datapath actions. */ | |
303 | ||
304 | bool installed; /* Installed in datapath? */ | |
305 | ||
306 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
307 | * splinters can cause it to differ. This value should be removed when | |
308 | * the VLAN splinters feature is no longer needed. */ | |
309 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
310 | }; | |
311 | ||
312 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, | |
313 | const struct nlattr *key, | |
314 | size_t key_len, ovs_be16 initial_tci); | |
315 | static struct subfacet *subfacet_find(struct ofproto_dpif *, | |
316 | const struct nlattr *key, size_t key_len); | |
317 | static void subfacet_destroy(struct subfacet *); | |
318 | static void subfacet_destroy__(struct subfacet *); | |
319 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, | |
320 | struct ofpbuf *key); | |
321 | static void subfacet_reset_dp_stats(struct subfacet *, | |
322 | struct dpif_flow_stats *); | |
323 | static void subfacet_update_time(struct subfacet *, long long int used); | |
324 | static void subfacet_update_stats(struct subfacet *, | |
325 | const struct dpif_flow_stats *); | |
326 | static void subfacet_make_actions(struct subfacet *, | |
327 | const struct ofpbuf *packet); | |
328 | static int subfacet_install(struct subfacet *, | |
329 | const struct nlattr *actions, size_t actions_len, | |
330 | struct dpif_flow_stats *); | |
331 | static void subfacet_uninstall(struct subfacet *); | |
332 | ||
b0f7b9b5 BP |
333 | /* An exact-match instantiation of an OpenFlow flow. |
334 | * | |
335 | * A facet associates a "struct flow", which represents the Open vSwitch | |
b95fc6ba BP |
336 | * userspace idea of an exact-match flow, with one or more subfacets. Each |
337 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
338 | * the facet. When the kernel module (or other dpif implementation) and Open | |
339 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
340 | * one subfacet per facet. If the dpif implementation supports more-specific | |
341 | * flow matching than userspace, however, a facet can have more than one | |
342 | * subfacet, each of which corresponds to some distinction in flow that | |
343 | * userspace simply doesn't understand. | |
b0f7b9b5 BP |
344 | * |
345 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
346 | * one subfacet or it will never expire, leaking memory. */ | |
abe529af | 347 | struct facet { |
b0f7b9b5 BP |
348 | /* Owners. */ |
349 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
350 | struct list list_node; /* In owning rule's 'facets' list. */ | |
351 | struct rule_dpif *rule; /* Owning rule. */ | |
352 | ||
353 | /* Owned data. */ | |
354 | struct list subfacets; | |
abe529af BP |
355 | long long int used; /* Time last used; time created if not used. */ |
356 | ||
b0f7b9b5 BP |
357 | /* Key. */ |
358 | struct flow flow; | |
359 | ||
abe529af BP |
360 | /* These statistics: |
361 | * | |
362 | * - Do include packets and bytes sent "by hand", e.g. with | |
363 | * dpif_execute(). | |
364 | * | |
365 | * - Do include packets and bytes that were obtained from the datapath | |
b0f7b9b5 | 366 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with |
abe529af | 367 | * DPIF_FP_ZERO_STATS). |
b0f7b9b5 BP |
368 | * |
369 | * - Do not include packets or bytes that can be obtained from the | |
370 | * datapath for any existing subfacet. | |
abe529af BP |
371 | */ |
372 | uint64_t packet_count; /* Number of packets received. */ | |
373 | uint64_t byte_count; /* Number of bytes received. */ | |
374 | ||
b0f7b9b5 | 375 | /* Resubmit statistics. */ |
9d24de3b JP |
376 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ |
377 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
378 | long long int prev_used; /* Used time from last stats push. */ | |
abe529af | 379 | |
b0f7b9b5 | 380 | /* Accounting. */ |
907a4c5e | 381 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
b0f7b9b5 | 382 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
0e553d9c | 383 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ |
abe529af | 384 | |
b95fc6ba BP |
385 | /* Properties of datapath actions. |
386 | * | |
387 | * Every subfacet has its own actions because actions can differ slightly | |
388 | * between splintered and non-splintered subfacets due to the VLAN tag | |
389 | * being initially different (present vs. absent). All of them have these | |
390 | * properties in common so we just store one copy of them here. */ | |
b0f7b9b5 | 391 | bool may_install; /* Reassess actions for every packet? */ |
75a75043 BP |
392 | bool has_learn; /* Actions include NXAST_LEARN? */ |
393 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
0e553d9c | 394 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ |
b0f7b9b5 | 395 | tag_type tags; /* Tags that would require revalidation. */ |
9d24de3b | 396 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ |
26cd7e34 BP |
397 | |
398 | /* Storage for a single subfacet, to reduce malloc() time and space | |
399 | * overhead. (A facet always has at least one subfacet and in the common | |
400 | * case has exactly one subfacet.) */ | |
401 | struct subfacet one_subfacet; | |
abe529af BP |
402 | }; |
403 | ||
2b459b83 BP |
404 | static struct facet *facet_create(struct rule_dpif *, |
405 | const struct flow *, uint32_t hash); | |
15baa734 | 406 | static void facet_remove(struct facet *); |
abe529af BP |
407 | static void facet_free(struct facet *); |
408 | ||
2b459b83 BP |
409 | static struct facet *facet_find(struct ofproto_dpif *, |
410 | const struct flow *, uint32_t hash); | |
abe529af | 411 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, |
2b459b83 | 412 | const struct flow *, uint32_t hash); |
15baa734 | 413 | static bool facet_revalidate(struct facet *); |
6814e51f | 414 | static bool facet_check_consistency(struct facet *); |
abe529af | 415 | |
15baa734 | 416 | static void facet_flush_stats(struct facet *); |
abe529af | 417 | |
15baa734 | 418 | static void facet_update_time(struct facet *, long long int used); |
bbb5d219 | 419 | static void facet_reset_counters(struct facet *); |
abe529af | 420 | static void facet_push_stats(struct facet *); |
3de9590b BP |
421 | static void facet_learn(struct facet *); |
422 | static void facet_account(struct facet *); | |
abe529af BP |
423 | |
424 | static bool facet_is_controller_flow(struct facet *); | |
425 | ||
abe529af BP |
426 | struct ofport_dpif { |
427 | struct ofport up; | |
428 | ||
429 | uint32_t odp_port; | |
430 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
431 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
432 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
433 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 434 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 435 | bool may_enable; /* May be enabled in bonds. */ |
3e5b3fdb | 436 | long long int carrier_seq; /* Carrier status changes. */ |
21f7563c | 437 | |
52a90c29 | 438 | /* Spanning tree. */ |
21f7563c JP |
439 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ |
440 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
441 | long long int stp_state_entered; | |
8b36f51e EJ |
442 | |
443 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
52a90c29 BP |
444 | |
445 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
446 | * | |
447 | * This is deprecated. It is only for compatibility with broken device | |
448 | * drivers in old versions of Linux that do not properly support VLANs when | |
449 | * VLAN devices are not used. When broken device drivers are no longer in | |
450 | * widespread use, we will delete these interfaces. */ | |
451 | uint16_t realdev_ofp_port; | |
452 | int vlandev_vid; | |
8b36f51e EJ |
453 | }; |
454 | ||
455 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
456 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
457 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
458 | struct priority_to_dscp { | |
459 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
460 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
461 | ||
462 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
abe529af BP |
463 | }; |
464 | ||
52a90c29 BP |
465 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
466 | * | |
467 | * This is deprecated. It is only for compatibility with broken device drivers | |
468 | * in old versions of Linux that do not properly support VLANs when VLAN | |
469 | * devices are not used. When broken device drivers are no longer in | |
470 | * widespread use, we will delete these interfaces. */ | |
471 | struct vlan_splinter { | |
472 | struct hmap_node realdev_vid_node; | |
473 | struct hmap_node vlandev_node; | |
474 | uint16_t realdev_ofp_port; | |
475 | uint16_t vlandev_ofp_port; | |
476 | int vid; | |
477 | }; | |
478 | ||
479 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
480 | uint32_t realdev, ovs_be16 vlan_tci); | |
481 | static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *, | |
482 | uint16_t vlandev, int *vid); | |
483 | static void vsp_remove(struct ofport_dpif *); | |
484 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
485 | ||
abe529af BP |
486 | static struct ofport_dpif * |
487 | ofport_dpif_cast(const struct ofport *ofport) | |
488 | { | |
489 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
490 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
491 | } | |
492 | ||
493 | static void port_run(struct ofport_dpif *); | |
494 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 495 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
8b36f51e | 496 | static void ofport_clear_priorities(struct ofport_dpif *); |
abe529af | 497 | |
7ee20df1 BP |
498 | struct dpif_completion { |
499 | struct list list_node; | |
500 | struct ofoperation *op; | |
501 | }; | |
502 | ||
54a9cbc9 BP |
503 | /* Extra information about a classifier table. |
504 | * Currently used just for optimized flow revalidation. */ | |
505 | struct table_dpif { | |
506 | /* If either of these is nonnull, then this table has a form that allows | |
507 | * flows to be tagged to avoid revalidating most flows for the most common | |
508 | * kinds of flow table changes. */ | |
509 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
510 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
511 | uint32_t basis; /* Keeps each table's tags separate. */ | |
512 | }; | |
513 | ||
abe529af | 514 | struct ofproto_dpif { |
b44a10b7 | 515 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ |
abe529af BP |
516 | struct ofproto up; |
517 | struct dpif *dpif; | |
518 | int max_ports; | |
519 | ||
6c1491fb BP |
520 | /* Statistics. */ |
521 | uint64_t n_matches; | |
522 | ||
abe529af BP |
523 | /* Bridging. */ |
524 | struct netflow *netflow; | |
bae473fe | 525 | struct dpif_sflow *sflow; |
abe529af BP |
526 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
527 | struct mac_learning *ml; | |
528 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
529 | bool has_bonded_bundles; | |
530 | ||
531 | /* Expiration. */ | |
532 | struct timer next_expiration; | |
533 | ||
534 | /* Facets. */ | |
535 | struct hmap facets; | |
b0f7b9b5 | 536 | struct hmap subfacets; |
54a9cbc9 BP |
537 | |
538 | /* Revalidation. */ | |
539 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
540 | bool need_revalidate; |
541 | struct tag_set revalidate_set; | |
7ee20df1 BP |
542 | |
543 | /* Support for debugging async flow mods. */ | |
544 | struct list completions; | |
daff3353 EJ |
545 | |
546 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
6527c598 PS |
547 | struct netdev_stats stats; /* To account packets generated and consumed in |
548 | * userspace. */ | |
21f7563c JP |
549 | |
550 | /* Spanning tree. */ | |
551 | struct stp *stp; | |
552 | long long int stp_last_tick; | |
52a90c29 BP |
553 | |
554 | /* VLAN splinters. */ | |
555 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
556 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
abe529af BP |
557 | }; |
558 | ||
7ee20df1 BP |
559 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
560 | * for debugging the asynchronous flow_mod implementation.) */ | |
561 | static bool clogged; | |
562 | ||
b44a10b7 BP |
563 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ |
564 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
565 | ||
abe529af BP |
566 | static void ofproto_dpif_unixctl_init(void); |
567 | ||
568 | static struct ofproto_dpif * | |
569 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
570 | { | |
571 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
572 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
573 | } | |
574 | ||
575 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
576 | uint16_t ofp_port); | |
577 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
578 | uint32_t odp_port); | |
6a6455e5 EJ |
579 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, |
580 | const struct ofpbuf *, ovs_be16 initial_tci, | |
581 | struct ds *); | |
abe529af BP |
582 | |
583 | /* Packet processing. */ | |
584 | static void update_learning_table(struct ofproto_dpif *, | |
585 | const struct flow *, int vlan, | |
586 | struct ofbundle *); | |
501f8d1f BP |
587 | /* Upcalls. */ |
588 | #define FLOW_MISS_MAX_BATCH 50 | |
9b16c439 | 589 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); |
abe529af BP |
590 | |
591 | /* Flow expiration. */ | |
592 | static int expire(struct ofproto_dpif *); | |
593 | ||
6fca1ffb BP |
594 | /* NetFlow. */ |
595 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
596 | ||
abe529af | 597 | /* Utilities. */ |
52a90c29 | 598 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); |
6ff686f2 PS |
599 | static size_t |
600 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
601 | const struct flow *, uint32_t odp_port); | |
c06bba01 JP |
602 | static void add_mirror_actions(struct action_xlate_ctx *ctx, |
603 | const struct flow *flow); | |
abe529af BP |
604 | /* Global variables. */ |
605 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
606 | \f | |
607 | /* Factory functions. */ | |
608 | ||
609 | static void | |
610 | enumerate_types(struct sset *types) | |
611 | { | |
612 | dp_enumerate_types(types); | |
613 | } | |
614 | ||
615 | static int | |
616 | enumerate_names(const char *type, struct sset *names) | |
617 | { | |
618 | return dp_enumerate_names(type, names); | |
619 | } | |
620 | ||
621 | static int | |
622 | del(const char *type, const char *name) | |
623 | { | |
624 | struct dpif *dpif; | |
625 | int error; | |
626 | ||
627 | error = dpif_open(name, type, &dpif); | |
628 | if (!error) { | |
629 | error = dpif_delete(dpif); | |
630 | dpif_close(dpif); | |
631 | } | |
632 | return error; | |
633 | } | |
634 | \f | |
635 | /* Basic life-cycle. */ | |
636 | ||
637 | static struct ofproto * | |
638 | alloc(void) | |
639 | { | |
640 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
641 | return &ofproto->up; | |
642 | } | |
643 | ||
644 | static void | |
645 | dealloc(struct ofproto *ofproto_) | |
646 | { | |
647 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
648 | free(ofproto); | |
649 | } | |
650 | ||
651 | static int | |
0f5f95a9 | 652 | construct(struct ofproto *ofproto_) |
abe529af BP |
653 | { |
654 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
655 | const char *name = ofproto->up.name; | |
656 | int error; | |
657 | int i; | |
658 | ||
659 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
660 | if (error) { | |
661 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
662 | return error; | |
663 | } | |
664 | ||
665 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 666 | ofproto->n_matches = 0; |
abe529af | 667 | |
be8194bb JG |
668 | dpif_flow_flush(ofproto->dpif); |
669 | dpif_recv_purge(ofproto->dpif); | |
670 | ||
a12b3ead | 671 | error = dpif_recv_set(ofproto->dpif, true); |
abe529af BP |
672 | if (error) { |
673 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
674 | dpif_close(ofproto->dpif); | |
675 | return error; | |
676 | } | |
abe529af BP |
677 | |
678 | ofproto->netflow = NULL; | |
679 | ofproto->sflow = NULL; | |
21f7563c | 680 | ofproto->stp = NULL; |
abe529af | 681 | hmap_init(&ofproto->bundles); |
e764773c | 682 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); |
abe529af BP |
683 | for (i = 0; i < MAX_MIRRORS; i++) { |
684 | ofproto->mirrors[i] = NULL; | |
685 | } | |
686 | ofproto->has_bonded_bundles = false; | |
687 | ||
688 | timer_set_duration(&ofproto->next_expiration, 1000); | |
689 | ||
690 | hmap_init(&ofproto->facets); | |
b0f7b9b5 | 691 | hmap_init(&ofproto->subfacets); |
54a9cbc9 BP |
692 | |
693 | for (i = 0; i < N_TABLES; i++) { | |
694 | struct table_dpif *table = &ofproto->tables[i]; | |
695 | ||
696 | table->catchall_table = NULL; | |
697 | table->other_table = NULL; | |
698 | table->basis = random_uint32(); | |
699 | } | |
abe529af BP |
700 | ofproto->need_revalidate = false; |
701 | tag_set_init(&ofproto->revalidate_set); | |
702 | ||
7ee20df1 BP |
703 | list_init(&ofproto->completions); |
704 | ||
abe529af BP |
705 | ofproto_dpif_unixctl_init(); |
706 | ||
daff3353 EJ |
707 | ofproto->has_bundle_action = false; |
708 | ||
52a90c29 BP |
709 | hmap_init(&ofproto->vlandev_map); |
710 | hmap_init(&ofproto->realdev_vid_map); | |
711 | ||
b44a10b7 BP |
712 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, |
713 | hash_string(ofproto->up.name, 0)); | |
6527c598 | 714 | memset(&ofproto->stats, 0, sizeof ofproto->stats); |
0f5f95a9 BP |
715 | |
716 | ofproto_init_tables(ofproto_, N_TABLES); | |
717 | ||
abe529af BP |
718 | return 0; |
719 | } | |
720 | ||
7ee20df1 BP |
721 | static void |
722 | complete_operations(struct ofproto_dpif *ofproto) | |
723 | { | |
724 | struct dpif_completion *c, *next; | |
725 | ||
726 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
727 | ofoperation_complete(c->op, 0); | |
728 | list_remove(&c->list_node); | |
729 | free(c); | |
730 | } | |
731 | } | |
732 | ||
abe529af BP |
733 | static void |
734 | destruct(struct ofproto *ofproto_) | |
735 | { | |
736 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 737 | struct rule_dpif *rule, *next_rule; |
d0918789 | 738 | struct oftable *table; |
abe529af BP |
739 | int i; |
740 | ||
b44a10b7 | 741 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); |
7ee20df1 BP |
742 | complete_operations(ofproto); |
743 | ||
0697b5c3 BP |
744 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
745 | struct cls_cursor cursor; | |
746 | ||
d0918789 | 747 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
748 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
749 | ofproto_rule_destroy(&rule->up); | |
750 | } | |
7ee20df1 BP |
751 | } |
752 | ||
abe529af BP |
753 | for (i = 0; i < MAX_MIRRORS; i++) { |
754 | mirror_destroy(ofproto->mirrors[i]); | |
755 | } | |
756 | ||
757 | netflow_destroy(ofproto->netflow); | |
bae473fe | 758 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
759 | hmap_destroy(&ofproto->bundles); |
760 | mac_learning_destroy(ofproto->ml); | |
761 | ||
762 | hmap_destroy(&ofproto->facets); | |
b0f7b9b5 | 763 | hmap_destroy(&ofproto->subfacets); |
abe529af | 764 | |
52a90c29 BP |
765 | hmap_destroy(&ofproto->vlandev_map); |
766 | hmap_destroy(&ofproto->realdev_vid_map); | |
767 | ||
abe529af BP |
768 | dpif_close(ofproto->dpif); |
769 | } | |
770 | ||
771 | static int | |
5fcc0d00 | 772 | run_fast(struct ofproto *ofproto_) |
abe529af BP |
773 | { |
774 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
9b16c439 | 775 | unsigned int work; |
abe529af | 776 | |
9b16c439 BP |
777 | /* Handle one or more batches of upcalls, until there's nothing left to do |
778 | * or until we do a fixed total amount of work. | |
779 | * | |
780 | * We do work in batches because it can be much cheaper to set up a number | |
781 | * of flows and fire off their patches all at once. We do multiple batches | |
782 | * because in some cases handling a packet can cause another packet to be | |
783 | * queued almost immediately as part of the return flow. Both | |
784 | * optimizations can make major improvements on some benchmarks and | |
785 | * presumably for real traffic as well. */ | |
786 | work = 0; | |
787 | while (work < FLOW_MISS_MAX_BATCH) { | |
788 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
5fcc0d00 | 789 | if (retval <= 0) { |
9b16c439 | 790 | return -retval; |
501f8d1f | 791 | } |
5fcc0d00 BP |
792 | work += retval; |
793 | } | |
794 | return 0; | |
795 | } | |
796 | ||
797 | static int | |
798 | run(struct ofproto *ofproto_) | |
799 | { | |
800 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
801 | struct ofport_dpif *ofport; | |
802 | struct ofbundle *bundle; | |
803 | int error; | |
804 | ||
805 | if (!clogged) { | |
806 | complete_operations(ofproto); | |
807 | } | |
808 | dpif_run(ofproto->dpif); | |
809 | ||
810 | error = run_fast(ofproto_); | |
811 | if (error) { | |
812 | return error; | |
abe529af BP |
813 | } |
814 | ||
815 | if (timer_expired(&ofproto->next_expiration)) { | |
816 | int delay = expire(ofproto); | |
817 | timer_set_duration(&ofproto->next_expiration, delay); | |
818 | } | |
819 | ||
820 | if (ofproto->netflow) { | |
6fca1ffb BP |
821 | if (netflow_run(ofproto->netflow)) { |
822 | send_netflow_active_timeouts(ofproto); | |
823 | } | |
abe529af BP |
824 | } |
825 | if (ofproto->sflow) { | |
bae473fe | 826 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
827 | } |
828 | ||
829 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
830 | port_run(ofport); | |
831 | } | |
832 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
833 | bundle_run(bundle); | |
834 | } | |
835 | ||
21f7563c | 836 | stp_run(ofproto); |
1c313b88 BP |
837 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
838 | ||
abe529af BP |
839 | /* Now revalidate if there's anything to do. */ |
840 | if (ofproto->need_revalidate | |
841 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
842 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
843 | bool revalidate_all = ofproto->need_revalidate; | |
844 | struct facet *facet, *next; | |
845 | ||
846 | /* Clear the revalidation flags. */ | |
847 | tag_set_init(&ofproto->revalidate_set); | |
848 | ofproto->need_revalidate = false; | |
849 | ||
850 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
851 | if (revalidate_all | |
852 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
15baa734 | 853 | facet_revalidate(facet); |
abe529af BP |
854 | } |
855 | } | |
856 | } | |
857 | ||
6814e51f BP |
858 | /* Check the consistency of a random facet, to aid debugging. */ |
859 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
860 | struct facet *facet; | |
861 | ||
862 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
863 | struct facet, hmap_node); | |
864 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
865 | if (!facet_check_consistency(facet)) { | |
866 | ofproto->need_revalidate = true; | |
867 | } | |
868 | } | |
869 | } | |
870 | ||
abe529af BP |
871 | return 0; |
872 | } | |
873 | ||
874 | static void | |
875 | wait(struct ofproto *ofproto_) | |
876 | { | |
877 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
878 | struct ofport_dpif *ofport; | |
879 | struct ofbundle *bundle; | |
880 | ||
7ee20df1 BP |
881 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
882 | poll_immediate_wake(); | |
883 | } | |
884 | ||
abe529af BP |
885 | dpif_wait(ofproto->dpif); |
886 | dpif_recv_wait(ofproto->dpif); | |
887 | if (ofproto->sflow) { | |
bae473fe | 888 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
889 | } |
890 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
891 | poll_immediate_wake(); | |
892 | } | |
893 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
894 | port_wait(ofport); | |
895 | } | |
896 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
897 | bundle_wait(bundle); | |
898 | } | |
6fca1ffb BP |
899 | if (ofproto->netflow) { |
900 | netflow_wait(ofproto->netflow); | |
901 | } | |
1c313b88 | 902 | mac_learning_wait(ofproto->ml); |
21f7563c | 903 | stp_wait(ofproto); |
abe529af BP |
904 | if (ofproto->need_revalidate) { |
905 | /* Shouldn't happen, but if it does just go around again. */ | |
906 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
907 | poll_immediate_wake(); | |
908 | } else { | |
909 | timer_wait(&ofproto->next_expiration); | |
910 | } | |
911 | } | |
912 | ||
913 | static void | |
914 | flush(struct ofproto *ofproto_) | |
915 | { | |
916 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
917 | struct facet *facet, *next_facet; | |
918 | ||
919 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
920 | /* Mark the facet as not installed so that facet_remove() doesn't | |
921 | * bother trying to uninstall it. There is no point in uninstalling it | |
922 | * individually since we are about to blow away all the facets with | |
923 | * dpif_flow_flush(). */ | |
b0f7b9b5 BP |
924 | struct subfacet *subfacet; |
925 | ||
926 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
927 | subfacet->installed = false; | |
928 | subfacet->dp_packet_count = 0; | |
929 | subfacet->dp_byte_count = 0; | |
930 | } | |
15baa734 | 931 | facet_remove(facet); |
abe529af BP |
932 | } |
933 | dpif_flow_flush(ofproto->dpif); | |
934 | } | |
935 | ||
6c1491fb BP |
936 | static void |
937 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
9e1fd49b | 938 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) |
6c1491fb BP |
939 | { |
940 | *arp_match_ip = true; | |
9e1fd49b BP |
941 | *actions = (OFPUTIL_A_OUTPUT | |
942 | OFPUTIL_A_SET_VLAN_VID | | |
943 | OFPUTIL_A_SET_VLAN_PCP | | |
944 | OFPUTIL_A_STRIP_VLAN | | |
945 | OFPUTIL_A_SET_DL_SRC | | |
946 | OFPUTIL_A_SET_DL_DST | | |
947 | OFPUTIL_A_SET_NW_SRC | | |
948 | OFPUTIL_A_SET_NW_DST | | |
949 | OFPUTIL_A_SET_NW_TOS | | |
950 | OFPUTIL_A_SET_TP_SRC | | |
951 | OFPUTIL_A_SET_TP_DST | | |
952 | OFPUTIL_A_ENQUEUE); | |
6c1491fb BP |
953 | } |
954 | ||
955 | static void | |
956 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
957 | { | |
958 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 959 | struct dpif_dp_stats s; |
6c1491fb BP |
960 | |
961 | strcpy(ots->name, "classifier"); | |
962 | ||
963 | dpif_get_dp_stats(ofproto->dpif, &s); | |
964 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
965 | put_32aligned_be64(&ots->matched_count, | |
966 | htonll(s.n_hit + ofproto->n_matches)); | |
967 | } | |
968 | ||
abe529af BP |
969 | static struct ofport * |
970 | port_alloc(void) | |
971 | { | |
972 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
973 | return &port->up; | |
974 | } | |
975 | ||
976 | static void | |
977 | port_dealloc(struct ofport *port_) | |
978 | { | |
979 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
980 | free(port); | |
981 | } | |
982 | ||
983 | static int | |
984 | port_construct(struct ofport *port_) | |
985 | { | |
986 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
987 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
988 | ||
f11c28c4 | 989 | ofproto->need_revalidate = true; |
abe529af BP |
990 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
991 | port->bundle = NULL; | |
992 | port->cfm = NULL; | |
993 | port->tag = tag_create_random(); | |
d5ffa7f2 | 994 | port->may_enable = true; |
21f7563c JP |
995 | port->stp_port = NULL; |
996 | port->stp_state = STP_DISABLED; | |
8b36f51e | 997 | hmap_init(&port->priorities); |
52a90c29 BP |
998 | port->realdev_ofp_port = 0; |
999 | port->vlandev_vid = 0; | |
3e5b3fdb | 1000 | port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); |
abe529af BP |
1001 | |
1002 | if (ofproto->sflow) { | |
392c7182 | 1003 | dpif_sflow_add_port(ofproto->sflow, port_); |
abe529af BP |
1004 | } |
1005 | ||
1006 | return 0; | |
1007 | } | |
1008 | ||
1009 | static void | |
1010 | port_destruct(struct ofport *port_) | |
1011 | { | |
1012 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1013 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1014 | ||
f11c28c4 | 1015 | ofproto->need_revalidate = true; |
abe529af | 1016 | bundle_remove(port_); |
a5610457 | 1017 | set_cfm(port_, NULL); |
abe529af | 1018 | if (ofproto->sflow) { |
bae473fe | 1019 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af | 1020 | } |
8b36f51e EJ |
1021 | |
1022 | ofport_clear_priorities(port); | |
1023 | hmap_destroy(&port->priorities); | |
abe529af BP |
1024 | } |
1025 | ||
1026 | static void | |
1027 | port_modified(struct ofport *port_) | |
1028 | { | |
1029 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1030 | ||
1031 | if (port->bundle && port->bundle->bond) { | |
1032 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | static void | |
9e1fd49b | 1037 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) |
abe529af BP |
1038 | { |
1039 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1040 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
9e1fd49b | 1041 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; |
abe529af | 1042 | |
9e1fd49b BP |
1043 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | |
1044 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) { | |
abe529af | 1045 | ofproto->need_revalidate = true; |
7bde8dd8 | 1046 | |
9e1fd49b | 1047 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { |
7bde8dd8 JP |
1048 | bundle_update(port->bundle); |
1049 | } | |
abe529af BP |
1050 | } |
1051 | } | |
1052 | ||
1053 | static int | |
1054 | set_sflow(struct ofproto *ofproto_, | |
1055 | const struct ofproto_sflow_options *sflow_options) | |
1056 | { | |
1057 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 1058 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 1059 | |
abe529af | 1060 | if (sflow_options) { |
bae473fe | 1061 | if (!ds) { |
abe529af BP |
1062 | struct ofport_dpif *ofport; |
1063 | ||
bae473fe | 1064 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 1065 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
392c7182 | 1066 | dpif_sflow_add_port(ds, &ofport->up); |
abe529af | 1067 | } |
6ff686f2 | 1068 | ofproto->need_revalidate = true; |
abe529af | 1069 | } |
bae473fe | 1070 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 1071 | } else { |
6ff686f2 PS |
1072 | if (ds) { |
1073 | dpif_sflow_destroy(ds); | |
1074 | ofproto->need_revalidate = true; | |
1075 | ofproto->sflow = NULL; | |
1076 | } | |
abe529af BP |
1077 | } |
1078 | return 0; | |
1079 | } | |
1080 | ||
1081 | static int | |
a5610457 | 1082 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
1083 | { |
1084 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1085 | int error; | |
1086 | ||
a5610457 | 1087 | if (!s) { |
abe529af BP |
1088 | error = 0; |
1089 | } else { | |
1090 | if (!ofport->cfm) { | |
8c977421 EJ |
1091 | struct ofproto_dpif *ofproto; |
1092 | ||
1093 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1094 | ofproto->need_revalidate = true; | |
6f629657 | 1095 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
1096 | } |
1097 | ||
a5610457 | 1098 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
1099 | return 0; |
1100 | } | |
1101 | ||
1102 | error = EINVAL; | |
1103 | } | |
1104 | cfm_destroy(ofport->cfm); | |
1105 | ofport->cfm = NULL; | |
1106 | return error; | |
1107 | } | |
1108 | ||
1109 | static int | |
a5610457 | 1110 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
1111 | { |
1112 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
1113 | |
1114 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 1115 | } |
1de11730 EJ |
1116 | |
1117 | static int | |
1118 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1119 | size_t *n_rmps) | |
1120 | { | |
1121 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1122 | ||
1123 | if (ofport->cfm) { | |
1124 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1125 | return 0; | |
1126 | } else { | |
1127 | return -1; | |
1128 | } | |
1129 | } | |
3967a833 MM |
1130 | |
1131 | static int | |
1132 | get_cfm_health(const struct ofport *ofport_) | |
1133 | { | |
1134 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1135 | ||
1136 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1137 | } | |
abe529af | 1138 | \f |
21f7563c JP |
1139 | /* Spanning Tree. */ |
1140 | ||
1141 | static void | |
1142 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1143 | { | |
1144 | struct ofproto_dpif *ofproto = ofproto_; | |
1145 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1146 | struct ofport_dpif *ofport; | |
1147 | ||
1148 | ofport = stp_port_get_aux(sp); | |
1149 | if (!ofport) { | |
1150 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1151 | ofproto->up.name, port_num); | |
1152 | } else { | |
1153 | struct eth_header *eth = pkt->l2; | |
1154 | ||
1155 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1156 | if (eth_addr_is_zero(eth->eth_src)) { | |
1157 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1158 | "with unknown MAC", ofproto->up.name, port_num); | |
1159 | } else { | |
97d6520b | 1160 | send_packet(ofport, pkt); |
21f7563c JP |
1161 | } |
1162 | } | |
1163 | ofpbuf_delete(pkt); | |
1164 | } | |
1165 | ||
1166 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1167 | static int | |
1168 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1169 | { | |
1170 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1171 | ||
1172 | /* Only revalidate flows if the configuration changed. */ | |
1173 | if (!s != !ofproto->stp) { | |
1174 | ofproto->need_revalidate = true; | |
1175 | } | |
1176 | ||
1177 | if (s) { | |
1178 | if (!ofproto->stp) { | |
1179 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1180 | send_bpdu_cb, ofproto); | |
1181 | ofproto->stp_last_tick = time_msec(); | |
1182 | } | |
1183 | ||
1184 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1185 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1186 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1187 | stp_set_max_age(ofproto->stp, s->max_age); | |
1188 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1189 | } else { | |
851bf71d EJ |
1190 | struct ofport *ofport; |
1191 | ||
1192 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1193 | set_stp_port(ofport, NULL); | |
1194 | } | |
1195 | ||
21f7563c JP |
1196 | stp_destroy(ofproto->stp); |
1197 | ofproto->stp = NULL; | |
1198 | } | |
1199 | ||
1200 | return 0; | |
1201 | } | |
1202 | ||
1203 | static int | |
1204 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1205 | { | |
1206 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1207 | ||
1208 | if (ofproto->stp) { | |
1209 | s->enabled = true; | |
1210 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1211 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1212 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1213 | } else { | |
1214 | s->enabled = false; | |
1215 | } | |
1216 | ||
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | static void | |
1221 | update_stp_port_state(struct ofport_dpif *ofport) | |
1222 | { | |
1223 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1224 | enum stp_state state; | |
1225 | ||
1226 | /* Figure out new state. */ | |
1227 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1228 | : STP_DISABLED; | |
1229 | ||
1230 | /* Update state. */ | |
1231 | if (ofport->stp_state != state) { | |
9e1fd49b | 1232 | enum ofputil_port_state of_state; |
21f7563c JP |
1233 | bool fwd_change; |
1234 | ||
1235 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1236 | netdev_get_name(ofport->up.netdev), | |
1237 | stp_state_name(ofport->stp_state), | |
1238 | stp_state_name(state)); | |
1239 | if (stp_learn_in_state(ofport->stp_state) | |
1240 | != stp_learn_in_state(state)) { | |
1241 | /* xxx Learning action flows should also be flushed. */ | |
d0040604 | 1242 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
21f7563c JP |
1243 | } |
1244 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1245 | != stp_forward_in_state(state); | |
1246 | ||
1247 | ofproto->need_revalidate = true; | |
1248 | ofport->stp_state = state; | |
1249 | ofport->stp_state_entered = time_msec(); | |
1250 | ||
b308140a | 1251 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1252 | bundle_update(ofport->bundle); |
1253 | } | |
1254 | ||
1255 | /* Update the STP state bits in the OpenFlow port description. */ | |
9e1fd49b BP |
1256 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; |
1257 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1258 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1259 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1260 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1261 | : 0); | |
21f7563c JP |
1262 | ofproto_port_set_state(&ofport->up, of_state); |
1263 | } | |
1264 | } | |
1265 | ||
1266 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1267 | * caller is responsible for assigning STP port numbers and ensuring | |
1268 | * there are no duplicates. */ | |
1269 | static int | |
1270 | set_stp_port(struct ofport *ofport_, | |
1271 | const struct ofproto_port_stp_settings *s) | |
1272 | { | |
1273 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1274 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1275 | struct stp_port *sp = ofport->stp_port; | |
1276 | ||
1277 | if (!s || !s->enable) { | |
1278 | if (sp) { | |
1279 | ofport->stp_port = NULL; | |
1280 | stp_port_disable(sp); | |
ecd12731 | 1281 | update_stp_port_state(ofport); |
21f7563c JP |
1282 | } |
1283 | return 0; | |
1284 | } else if (sp && stp_port_no(sp) != s->port_num | |
1285 | && ofport == stp_port_get_aux(sp)) { | |
1286 | /* The port-id changed, so disable the old one if it's not | |
1287 | * already in use by another port. */ | |
1288 | stp_port_disable(sp); | |
1289 | } | |
1290 | ||
1291 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1292 | stp_port_enable(sp); | |
1293 | ||
1294 | stp_port_set_aux(sp, ofport); | |
1295 | stp_port_set_priority(sp, s->priority); | |
1296 | stp_port_set_path_cost(sp, s->path_cost); | |
1297 | ||
1298 | update_stp_port_state(ofport); | |
1299 | ||
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | static int | |
1304 | get_stp_port_status(struct ofport *ofport_, | |
1305 | struct ofproto_port_stp_status *s) | |
1306 | { | |
1307 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1308 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1309 | struct stp_port *sp = ofport->stp_port; | |
1310 | ||
1311 | if (!ofproto->stp || !sp) { | |
1312 | s->enabled = false; | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | s->enabled = true; | |
1317 | s->port_id = stp_port_get_id(sp); | |
1318 | s->state = stp_port_get_state(sp); | |
1319 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1320 | s->role = stp_port_get_role(sp); | |
80740385 | 1321 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1322 | |
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | static void | |
1327 | stp_run(struct ofproto_dpif *ofproto) | |
1328 | { | |
1329 | if (ofproto->stp) { | |
1330 | long long int now = time_msec(); | |
1331 | long long int elapsed = now - ofproto->stp_last_tick; | |
1332 | struct stp_port *sp; | |
1333 | ||
1334 | if (elapsed > 0) { | |
1335 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1336 | ofproto->stp_last_tick = now; | |
1337 | } | |
1338 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1339 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1340 | ||
1341 | if (ofport) { | |
1342 | update_stp_port_state(ofport); | |
1343 | } | |
1344 | } | |
6ae50723 EJ |
1345 | |
1346 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1347 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1348 | } | |
21f7563c JP |
1349 | } |
1350 | } | |
1351 | ||
1352 | static void | |
1353 | stp_wait(struct ofproto_dpif *ofproto) | |
1354 | { | |
1355 | if (ofproto->stp) { | |
1356 | poll_timer_wait(1000); | |
1357 | } | |
1358 | } | |
1359 | ||
1360 | /* Returns true if STP should process 'flow'. */ | |
1361 | static bool | |
1362 | stp_should_process_flow(const struct flow *flow) | |
1363 | { | |
1364 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1365 | } | |
1366 | ||
1367 | static void | |
1368 | stp_process_packet(const struct ofport_dpif *ofport, | |
1369 | const struct ofpbuf *packet) | |
1370 | { | |
1371 | struct ofpbuf payload = *packet; | |
1372 | struct eth_header *eth = payload.data; | |
1373 | struct stp_port *sp = ofport->stp_port; | |
1374 | ||
1375 | /* Sink packets on ports that have STP disabled when the bridge has | |
1376 | * STP enabled. */ | |
1377 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1378 | return; | |
1379 | } | |
1380 | ||
1381 | /* Trim off padding on payload. */ | |
c573540b BP |
1382 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1383 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1384 | } |
1385 | ||
1386 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1387 | stp_received_bpdu(sp, payload.data, payload.size); | |
1388 | } | |
1389 | } | |
1390 | \f | |
8b36f51e EJ |
1391 | static struct priority_to_dscp * |
1392 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1393 | { | |
1394 | struct priority_to_dscp *pdscp; | |
1395 | uint32_t hash; | |
1396 | ||
1397 | hash = hash_int(priority, 0); | |
1398 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1399 | if (pdscp->priority == priority) { | |
1400 | return pdscp; | |
1401 | } | |
1402 | } | |
1403 | return NULL; | |
1404 | } | |
1405 | ||
1406 | static void | |
1407 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1408 | { | |
1409 | struct priority_to_dscp *pdscp, *next; | |
1410 | ||
1411 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1412 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1413 | free(pdscp); | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | static int | |
1418 | set_queues(struct ofport *ofport_, | |
1419 | const struct ofproto_port_queue *qdscp_list, | |
1420 | size_t n_qdscp) | |
1421 | { | |
1422 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1423 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1424 | struct hmap new = HMAP_INITIALIZER(&new); | |
1425 | size_t i; | |
1426 | ||
1427 | for (i = 0; i < n_qdscp; i++) { | |
1428 | struct priority_to_dscp *pdscp; | |
1429 | uint32_t priority; | |
1430 | uint8_t dscp; | |
1431 | ||
1432 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1433 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1434 | &priority)) { | |
1435 | continue; | |
1436 | } | |
1437 | ||
1438 | pdscp = get_priority(ofport, priority); | |
1439 | if (pdscp) { | |
1440 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1441 | } else { | |
1442 | pdscp = xmalloc(sizeof *pdscp); | |
1443 | pdscp->priority = priority; | |
1444 | pdscp->dscp = dscp; | |
1445 | ofproto->need_revalidate = true; | |
1446 | } | |
1447 | ||
1448 | if (pdscp->dscp != dscp) { | |
1449 | pdscp->dscp = dscp; | |
1450 | ofproto->need_revalidate = true; | |
1451 | } | |
1452 | ||
1453 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1454 | } | |
1455 | ||
1456 | if (!hmap_is_empty(&ofport->priorities)) { | |
1457 | ofport_clear_priorities(ofport); | |
1458 | ofproto->need_revalidate = true; | |
1459 | } | |
1460 | ||
1461 | hmap_swap(&new, &ofport->priorities); | |
1462 | hmap_destroy(&new); | |
1463 | ||
1464 | return 0; | |
1465 | } | |
1466 | \f | |
abe529af BP |
1467 | /* Bundles. */ |
1468 | ||
b44a10b7 BP |
1469 | /* Expires all MAC learning entries associated with 'bundle' and forces its |
1470 | * ofproto to revalidate every flow. | |
1471 | * | |
1472 | * Normally MAC learning entries are removed only from the ofproto associated | |
1473 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1474 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1475 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1476 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1477 | * with the host from which it migrated. */ | |
abe529af | 1478 | static void |
b44a10b7 | 1479 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) |
abe529af BP |
1480 | { |
1481 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1482 | struct mac_learning *ml = ofproto->ml; | |
1483 | struct mac_entry *mac, *next_mac; | |
1484 | ||
1485 | ofproto->need_revalidate = true; | |
1486 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1487 | if (mac->port.p == bundle) { | |
b44a10b7 BP |
1488 | if (all_ofprotos) { |
1489 | struct ofproto_dpif *o; | |
1490 | ||
1491 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1492 | if (o != ofproto) { | |
1493 | struct mac_entry *e; | |
1494 | ||
1495 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1496 | NULL); | |
1497 | if (e) { | |
1498 | tag_set_add(&o->revalidate_set, e->tag); | |
1499 | mac_learning_expire(o->ml, e); | |
1500 | } | |
1501 | } | |
1502 | } | |
1503 | } | |
1504 | ||
abe529af BP |
1505 | mac_learning_expire(ml, mac); |
1506 | } | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | static struct ofbundle * | |
1511 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1512 | { | |
1513 | struct ofbundle *bundle; | |
1514 | ||
1515 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1516 | &ofproto->bundles) { | |
1517 | if (bundle->aux == aux) { | |
1518 | return bundle; | |
1519 | } | |
1520 | } | |
1521 | return NULL; | |
1522 | } | |
1523 | ||
1524 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1525 | * ones that are found to 'bundles'. */ | |
1526 | static void | |
1527 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1528 | void **auxes, size_t n_auxes, | |
1529 | struct hmapx *bundles) | |
1530 | { | |
1531 | size_t i; | |
1532 | ||
1533 | hmapx_init(bundles); | |
1534 | for (i = 0; i < n_auxes; i++) { | |
1535 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1536 | if (bundle) { | |
1537 | hmapx_add(bundles, bundle); | |
1538 | } | |
1539 | } | |
1540 | } | |
1541 | ||
7bde8dd8 JP |
1542 | static void |
1543 | bundle_update(struct ofbundle *bundle) | |
1544 | { | |
1545 | struct ofport_dpif *port; | |
1546 | ||
1547 | bundle->floodable = true; | |
1548 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
9e1fd49b BP |
1549 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1550 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1551 | bundle->floodable = false; |
1552 | break; | |
1553 | } | |
1554 | } | |
1555 | } | |
1556 | ||
abe529af BP |
1557 | static void |
1558 | bundle_del_port(struct ofport_dpif *port) | |
1559 | { | |
1560 | struct ofbundle *bundle = port->bundle; | |
1561 | ||
6f77f4ae BP |
1562 | bundle->ofproto->need_revalidate = true; |
1563 | ||
abe529af BP |
1564 | list_remove(&port->bundle_node); |
1565 | port->bundle = NULL; | |
1566 | ||
1567 | if (bundle->lacp) { | |
1568 | lacp_slave_unregister(bundle->lacp, port); | |
1569 | } | |
1570 | if (bundle->bond) { | |
1571 | bond_slave_unregister(bundle->bond, port); | |
1572 | } | |
1573 | ||
7bde8dd8 | 1574 | bundle_update(bundle); |
abe529af BP |
1575 | } |
1576 | ||
1577 | static bool | |
1578 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1579 | struct lacp_slave_settings *lacp, |
1580 | uint32_t bond_stable_id) | |
abe529af BP |
1581 | { |
1582 | struct ofport_dpif *port; | |
1583 | ||
1584 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1585 | if (!port) { | |
1586 | return false; | |
1587 | } | |
1588 | ||
1589 | if (port->bundle != bundle) { | |
6f77f4ae | 1590 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1591 | if (port->bundle) { |
1592 | bundle_del_port(port); | |
1593 | } | |
1594 | ||
1595 | port->bundle = bundle; | |
1596 | list_push_back(&bundle->ports, &port->bundle_node); | |
9e1fd49b BP |
1597 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD |
1598 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1599 | bundle->floodable = false; |
1600 | } | |
1601 | } | |
1602 | if (lacp) { | |
4a86aece | 1603 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1604 | lacp_slave_register(bundle->lacp, port, lacp); |
1605 | } | |
1606 | ||
00794817 BP |
1607 | port->bond_stable_id = bond_stable_id; |
1608 | ||
abe529af BP |
1609 | return true; |
1610 | } | |
1611 | ||
1612 | static void | |
1613 | bundle_destroy(struct ofbundle *bundle) | |
1614 | { | |
1615 | struct ofproto_dpif *ofproto; | |
1616 | struct ofport_dpif *port, *next_port; | |
1617 | int i; | |
1618 | ||
1619 | if (!bundle) { | |
1620 | return; | |
1621 | } | |
1622 | ||
1623 | ofproto = bundle->ofproto; | |
1624 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1625 | struct ofmirror *m = ofproto->mirrors[i]; | |
1626 | if (m) { | |
1627 | if (m->out == bundle) { | |
1628 | mirror_destroy(m); | |
1629 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1630 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1631 | ofproto->need_revalidate = true; | |
1632 | } | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1637 | bundle_del_port(port); | |
1638 | } | |
1639 | ||
b44a10b7 | 1640 | bundle_flush_macs(bundle, true); |
abe529af BP |
1641 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); |
1642 | free(bundle->name); | |
1643 | free(bundle->trunks); | |
1644 | lacp_destroy(bundle->lacp); | |
1645 | bond_destroy(bundle->bond); | |
1646 | free(bundle); | |
1647 | } | |
1648 | ||
1649 | static int | |
1650 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1651 | const struct ofproto_bundle_settings *s) | |
1652 | { | |
1653 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1654 | bool need_flush = false; | |
abe529af BP |
1655 | struct ofport_dpif *port; |
1656 | struct ofbundle *bundle; | |
ecac4ebf BP |
1657 | unsigned long *trunks; |
1658 | int vlan; | |
abe529af BP |
1659 | size_t i; |
1660 | bool ok; | |
1661 | ||
1662 | if (!s) { | |
1663 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1664 | return 0; | |
1665 | } | |
1666 | ||
1667 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1668 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1669 | ||
1670 | bundle = bundle_lookup(ofproto, aux); | |
1671 | if (!bundle) { | |
1672 | bundle = xmalloc(sizeof *bundle); | |
1673 | ||
1674 | bundle->ofproto = ofproto; | |
1675 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1676 | hash_pointer(aux, 0)); | |
1677 | bundle->aux = aux; | |
1678 | bundle->name = NULL; | |
1679 | ||
1680 | list_init(&bundle->ports); | |
ecac4ebf | 1681 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1682 | bundle->vlan = -1; |
1683 | bundle->trunks = NULL; | |
5e9ceccd | 1684 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1685 | bundle->lacp = NULL; |
1686 | bundle->bond = NULL; | |
1687 | ||
1688 | bundle->floodable = true; | |
1689 | ||
1690 | bundle->src_mirrors = 0; | |
1691 | bundle->dst_mirrors = 0; | |
1692 | bundle->mirror_out = 0; | |
1693 | } | |
1694 | ||
1695 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1696 | free(bundle->name); | |
1697 | bundle->name = xstrdup(s->name); | |
1698 | } | |
1699 | ||
1700 | /* LACP. */ | |
1701 | if (s->lacp) { | |
1702 | if (!bundle->lacp) { | |
8c977421 | 1703 | ofproto->need_revalidate = true; |
abe529af BP |
1704 | bundle->lacp = lacp_create(); |
1705 | } | |
1706 | lacp_configure(bundle->lacp, s->lacp); | |
1707 | } else { | |
1708 | lacp_destroy(bundle->lacp); | |
1709 | bundle->lacp = NULL; | |
1710 | } | |
1711 | ||
1712 | /* Update set of ports. */ | |
1713 | ok = true; | |
1714 | for (i = 0; i < s->n_slaves; i++) { | |
1715 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1716 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1717 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1718 | ok = false; |
1719 | } | |
1720 | } | |
1721 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1722 | struct ofport_dpif *next_port; | |
1723 | ||
1724 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1725 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1726 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1727 | goto found; |
1728 | } | |
1729 | } | |
1730 | ||
1731 | bundle_del_port(port); | |
1732 | found: ; | |
1733 | } | |
1734 | } | |
1735 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1736 | ||
1737 | if (list_is_empty(&bundle->ports)) { | |
1738 | bundle_destroy(bundle); | |
1739 | return EINVAL; | |
1740 | } | |
1741 | ||
ecac4ebf | 1742 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1743 | if (s->vlan_mode != bundle->vlan_mode |
1744 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1745 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1746 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1747 | need_flush = true; |
1748 | } | |
1749 | ||
abe529af | 1750 | /* Set VLAN tag. */ |
ecac4ebf BP |
1751 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1752 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1753 | : 0); | |
1754 | if (vlan != bundle->vlan) { | |
1755 | bundle->vlan = vlan; | |
abe529af BP |
1756 | need_flush = true; |
1757 | } | |
1758 | ||
1759 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1760 | switch (s->vlan_mode) { |
1761 | case PORT_VLAN_ACCESS: | |
1762 | trunks = NULL; | |
1763 | break; | |
1764 | ||
1765 | case PORT_VLAN_TRUNK: | |
1766 | trunks = (unsigned long *) s->trunks; | |
1767 | break; | |
1768 | ||
1769 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1770 | case PORT_VLAN_NATIVE_TAGGED: | |
1771 | if (vlan != 0 && (!s->trunks | |
1772 | || !bitmap_is_set(s->trunks, vlan) | |
1773 | || bitmap_is_set(s->trunks, 0))) { | |
1774 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1775 | if (s->trunks) { | |
1776 | trunks = bitmap_clone(s->trunks, 4096); | |
1777 | } else { | |
1778 | trunks = bitmap_allocate1(4096); | |
1779 | } | |
1780 | bitmap_set1(trunks, vlan); | |
1781 | bitmap_set0(trunks, 0); | |
1782 | } else { | |
1783 | trunks = (unsigned long *) s->trunks; | |
1784 | } | |
1785 | break; | |
1786 | ||
1787 | default: | |
1788 | NOT_REACHED(); | |
1789 | } | |
abe529af BP |
1790 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1791 | free(bundle->trunks); | |
ecac4ebf BP |
1792 | if (trunks == s->trunks) { |
1793 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1794 | } else { | |
1795 | bundle->trunks = trunks; | |
1796 | trunks = NULL; | |
1797 | } | |
abe529af BP |
1798 | need_flush = true; |
1799 | } | |
ecac4ebf BP |
1800 | if (trunks != s->trunks) { |
1801 | free(trunks); | |
1802 | } | |
abe529af BP |
1803 | |
1804 | /* Bonding. */ | |
1805 | if (!list_is_short(&bundle->ports)) { | |
1806 | bundle->ofproto->has_bonded_bundles = true; | |
1807 | if (bundle->bond) { | |
1808 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1809 | ofproto->need_revalidate = true; | |
1810 | } | |
1811 | } else { | |
1812 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1813 | ofproto->need_revalidate = true; |
abe529af BP |
1814 | } |
1815 | ||
1816 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1817 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1818 | port->up.netdev); |
1819 | } | |
1820 | } else { | |
1821 | bond_destroy(bundle->bond); | |
1822 | bundle->bond = NULL; | |
1823 | } | |
1824 | ||
1825 | /* If we changed something that would affect MAC learning, un-learn | |
1826 | * everything on this port and force flow revalidation. */ | |
1827 | if (need_flush) { | |
b44a10b7 | 1828 | bundle_flush_macs(bundle, false); |
abe529af BP |
1829 | } |
1830 | ||
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | static void | |
1835 | bundle_remove(struct ofport *port_) | |
1836 | { | |
1837 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1838 | struct ofbundle *bundle = port->bundle; | |
1839 | ||
1840 | if (bundle) { | |
1841 | bundle_del_port(port); | |
1842 | if (list_is_empty(&bundle->ports)) { | |
1843 | bundle_destroy(bundle); | |
1844 | } else if (list_is_short(&bundle->ports)) { | |
1845 | bond_destroy(bundle->bond); | |
1846 | bundle->bond = NULL; | |
1847 | } | |
1848 | } | |
1849 | } | |
1850 | ||
1851 | static void | |
5f877369 | 1852 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1853 | { |
1854 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1855 | struct ofport_dpif *port = port_; | |
1856 | uint8_t ea[ETH_ADDR_LEN]; | |
1857 | int error; | |
1858 | ||
1859 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1860 | if (!error) { | |
abe529af | 1861 | struct ofpbuf packet; |
5f877369 | 1862 | void *packet_pdu; |
abe529af BP |
1863 | |
1864 | ofpbuf_init(&packet, 0); | |
1865 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1866 | pdu_size); |
1867 | memcpy(packet_pdu, pdu, pdu_size); | |
1868 | ||
97d6520b | 1869 | send_packet(port, &packet); |
abe529af BP |
1870 | ofpbuf_uninit(&packet); |
1871 | } else { | |
1872 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1873 | "%s (%s)", port->bundle->name, | |
1874 | netdev_get_name(port->up.netdev), strerror(error)); | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | static void | |
1879 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1880 | { | |
1881 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1882 | int error, n_packets, n_errors; | |
1883 | struct mac_entry *e; | |
1884 | ||
1885 | error = n_packets = n_errors = 0; | |
1886 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1887 | if (e->port.p != bundle) { | |
ea131871 JG |
1888 | struct ofpbuf *learning_packet; |
1889 | struct ofport_dpif *port; | |
4dd1e3ca | 1890 | void *port_void; |
ea131871 JG |
1891 | int ret; |
1892 | ||
4dd1e3ca BP |
1893 | /* The assignment to "port" is unnecessary but makes "grep"ing for |
1894 | * struct ofport_dpif more effective. */ | |
1895 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
1896 | e->mac, e->vlan, | |
1897 | &port_void); | |
1898 | port = port_void; | |
97d6520b | 1899 | ret = send_packet(port, learning_packet); |
ea131871 | 1900 | ofpbuf_delete(learning_packet); |
abe529af BP |
1901 | if (ret) { |
1902 | error = ret; | |
1903 | n_errors++; | |
1904 | } | |
1905 | n_packets++; | |
1906 | } | |
1907 | } | |
1908 | ||
1909 | if (n_errors) { | |
1910 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1911 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1912 | "packets, last error was: %s", | |
1913 | bundle->name, n_errors, n_packets, strerror(error)); | |
1914 | } else { | |
1915 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1916 | bundle->name, n_packets); | |
1917 | } | |
1918 | } | |
1919 | ||
1920 | static void | |
1921 | bundle_run(struct ofbundle *bundle) | |
1922 | { | |
1923 | if (bundle->lacp) { | |
1924 | lacp_run(bundle->lacp, send_pdu_cb); | |
1925 | } | |
1926 | if (bundle->bond) { | |
1927 | struct ofport_dpif *port; | |
1928 | ||
1929 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1930 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1931 | } |
1932 | ||
1933 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
bdebeece | 1934 | lacp_status(bundle->lacp)); |
abe529af BP |
1935 | if (bond_should_send_learning_packets(bundle->bond)) { |
1936 | bundle_send_learning_packets(bundle); | |
1937 | } | |
1938 | } | |
1939 | } | |
1940 | ||
1941 | static void | |
1942 | bundle_wait(struct ofbundle *bundle) | |
1943 | { | |
1944 | if (bundle->lacp) { | |
1945 | lacp_wait(bundle->lacp); | |
1946 | } | |
1947 | if (bundle->bond) { | |
1948 | bond_wait(bundle->bond); | |
1949 | } | |
1950 | } | |
1951 | \f | |
1952 | /* Mirrors. */ | |
1953 | ||
1954 | static int | |
1955 | mirror_scan(struct ofproto_dpif *ofproto) | |
1956 | { | |
1957 | int idx; | |
1958 | ||
1959 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1960 | if (!ofproto->mirrors[idx]) { | |
1961 | return idx; | |
1962 | } | |
1963 | } | |
1964 | return -1; | |
1965 | } | |
1966 | ||
1967 | static struct ofmirror * | |
1968 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1969 | { | |
1970 | int i; | |
1971 | ||
1972 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1973 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1974 | if (mirror && mirror->aux == aux) { | |
1975 | return mirror; | |
1976 | } | |
1977 | } | |
1978 | ||
1979 | return NULL; | |
1980 | } | |
1981 | ||
9ba15e2a BP |
1982 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
1983 | static void | |
1984 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
1985 | { | |
1986 | int i; | |
1987 | ||
1988 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1989 | struct ofmirror *m = ofproto->mirrors[i]; | |
1990 | ||
1991 | if (m) { | |
1992 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1997 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
1998 | int j; | |
1999 | ||
2000 | if (!m1) { | |
2001 | continue; | |
2002 | } | |
2003 | ||
2004 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2005 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2006 | ||
edb0540b | 2007 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
2008 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
2009 | m2->dup_mirrors |= m1->dup_mirrors; | |
2010 | } | |
2011 | } | |
2012 | } | |
2013 | } | |
2014 | ||
abe529af BP |
2015 | static int |
2016 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2017 | const struct ofproto_mirror_settings *s) | |
2018 | { | |
2019 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2020 | mirror_mask_t mirror_bit; | |
2021 | struct ofbundle *bundle; | |
2022 | struct ofmirror *mirror; | |
2023 | struct ofbundle *out; | |
2024 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2025 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2026 | int out_vlan; | |
2027 | ||
2028 | mirror = mirror_lookup(ofproto, aux); | |
2029 | if (!s) { | |
2030 | mirror_destroy(mirror); | |
2031 | return 0; | |
2032 | } | |
2033 | if (!mirror) { | |
2034 | int idx; | |
2035 | ||
2036 | idx = mirror_scan(ofproto); | |
2037 | if (idx < 0) { | |
2038 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2039 | "cannot create %s", | |
2040 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2041 | return EFBIG; | |
2042 | } | |
2043 | ||
2044 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2045 | mirror->ofproto = ofproto; | |
2046 | mirror->idx = idx; | |
8b28d864 | 2047 | mirror->aux = aux; |
abe529af BP |
2048 | mirror->out_vlan = -1; |
2049 | mirror->name = NULL; | |
2050 | } | |
2051 | ||
2052 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2053 | free(mirror->name); | |
2054 | mirror->name = xstrdup(s->name); | |
2055 | } | |
2056 | ||
2057 | /* Get the new configuration. */ | |
2058 | if (s->out_bundle) { | |
2059 | out = bundle_lookup(ofproto, s->out_bundle); | |
2060 | if (!out) { | |
2061 | mirror_destroy(mirror); | |
2062 | return EINVAL; | |
2063 | } | |
2064 | out_vlan = -1; | |
2065 | } else { | |
2066 | out = NULL; | |
2067 | out_vlan = s->out_vlan; | |
2068 | } | |
2069 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2070 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2071 | ||
2072 | /* If the configuration has not changed, do nothing. */ | |
2073 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2074 | && hmapx_equals(&dsts, &mirror->dsts) | |
2075 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2076 | && mirror->out == out | |
2077 | && mirror->out_vlan == out_vlan) | |
2078 | { | |
2079 | hmapx_destroy(&srcs); | |
2080 | hmapx_destroy(&dsts); | |
2081 | return 0; | |
2082 | } | |
2083 | ||
2084 | hmapx_swap(&srcs, &mirror->srcs); | |
2085 | hmapx_destroy(&srcs); | |
2086 | ||
2087 | hmapx_swap(&dsts, &mirror->dsts); | |
2088 | hmapx_destroy(&dsts); | |
2089 | ||
2090 | free(mirror->vlans); | |
2091 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2092 | ||
2093 | mirror->out = out; | |
2094 | mirror->out_vlan = out_vlan; | |
2095 | ||
2096 | /* Update bundles. */ | |
2097 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2098 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2099 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2100 | bundle->src_mirrors |= mirror_bit; | |
2101 | } else { | |
2102 | bundle->src_mirrors &= ~mirror_bit; | |
2103 | } | |
2104 | ||
2105 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2106 | bundle->dst_mirrors |= mirror_bit; | |
2107 | } else { | |
2108 | bundle->dst_mirrors &= ~mirror_bit; | |
2109 | } | |
2110 | ||
2111 | if (mirror->out == bundle) { | |
2112 | bundle->mirror_out |= mirror_bit; | |
2113 | } else { | |
2114 | bundle->mirror_out &= ~mirror_bit; | |
2115 | } | |
2116 | } | |
2117 | ||
2118 | ofproto->need_revalidate = true; | |
d0040604 | 2119 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
9ba15e2a | 2120 | mirror_update_dups(ofproto); |
abe529af BP |
2121 | |
2122 | return 0; | |
2123 | } | |
2124 | ||
2125 | static void | |
2126 | mirror_destroy(struct ofmirror *mirror) | |
2127 | { | |
2128 | struct ofproto_dpif *ofproto; | |
2129 | mirror_mask_t mirror_bit; | |
2130 | struct ofbundle *bundle; | |
2131 | ||
2132 | if (!mirror) { | |
2133 | return; | |
2134 | } | |
2135 | ||
2136 | ofproto = mirror->ofproto; | |
2137 | ofproto->need_revalidate = true; | |
d0040604 | 2138 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2139 | |
2140 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2141 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2142 | bundle->src_mirrors &= ~mirror_bit; | |
2143 | bundle->dst_mirrors &= ~mirror_bit; | |
2144 | bundle->mirror_out &= ~mirror_bit; | |
2145 | } | |
2146 | ||
2147 | hmapx_destroy(&mirror->srcs); | |
2148 | hmapx_destroy(&mirror->dsts); | |
2149 | free(mirror->vlans); | |
2150 | ||
2151 | ofproto->mirrors[mirror->idx] = NULL; | |
2152 | free(mirror->name); | |
2153 | free(mirror); | |
9ba15e2a BP |
2154 | |
2155 | mirror_update_dups(ofproto); | |
abe529af BP |
2156 | } |
2157 | ||
9d24de3b JP |
2158 | static int |
2159 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2160 | uint64_t *packets, uint64_t *bytes) | |
2161 | { | |
2162 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2163 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2164 | ||
2165 | if (!mirror) { | |
2166 | *packets = *bytes = UINT64_MAX; | |
2167 | return 0; | |
2168 | } | |
2169 | ||
2170 | *packets = mirror->packet_count; | |
2171 | *bytes = mirror->byte_count; | |
2172 | ||
2173 | return 0; | |
2174 | } | |
2175 | ||
abe529af BP |
2176 | static int |
2177 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2178 | { | |
2179 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2180 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
d0040604 | 2181 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
abe529af BP |
2182 | } |
2183 | return 0; | |
2184 | } | |
2185 | ||
2186 | static bool | |
b4affc74 | 2187 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
2188 | { |
2189 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2190 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2191 | return bundle && bundle->mirror_out != 0; | |
2192 | } | |
8402c74b SS |
2193 | |
2194 | static void | |
b53055f4 | 2195 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
2196 | { |
2197 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2198 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
2199 | ofproto->need_revalidate = true; | |
2200 | } | |
e764773c BP |
2201 | |
2202 | static void | |
2203 | set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) | |
2204 | { | |
2205 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2206 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
2207 | } | |
abe529af BP |
2208 | \f |
2209 | /* Ports. */ | |
2210 | ||
2211 | static struct ofport_dpif * | |
2212 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
2213 | { | |
7df6a8bd BP |
2214 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
2215 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
2216 | } |
2217 | ||
2218 | static struct ofport_dpif * | |
2219 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
2220 | { | |
2221 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2222 | } | |
2223 | ||
2224 | static void | |
2225 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2226 | struct dpif_port *dpif_port) | |
2227 | { | |
2228 | ofproto_port->name = dpif_port->name; | |
2229 | ofproto_port->type = dpif_port->type; | |
2230 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2231 | } | |
2232 | ||
2233 | static void | |
2234 | port_run(struct ofport_dpif *ofport) | |
2235 | { | |
3e5b3fdb EJ |
2236 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); |
2237 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
015e08bc EJ |
2238 | bool enable = netdev_get_carrier(ofport->up.netdev); |
2239 | ||
3e5b3fdb EJ |
2240 | ofport->carrier_seq = carrier_seq; |
2241 | ||
abe529af BP |
2242 | if (ofport->cfm) { |
2243 | cfm_run(ofport->cfm); | |
2244 | ||
2245 | if (cfm_should_send_ccm(ofport->cfm)) { | |
2246 | struct ofpbuf packet; | |
abe529af BP |
2247 | |
2248 | ofpbuf_init(&packet, 0); | |
9e1fd49b | 2249 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); |
97d6520b | 2250 | send_packet(ofport, &packet); |
abe529af BP |
2251 | ofpbuf_uninit(&packet); |
2252 | } | |
015e08bc | 2253 | |
86dc6501 EJ |
2254 | enable = enable && !cfm_get_fault(ofport->cfm) |
2255 | && cfm_get_opup(ofport->cfm); | |
abe529af | 2256 | } |
015e08bc EJ |
2257 | |
2258 | if (ofport->bundle) { | |
2259 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3e5b3fdb EJ |
2260 | if (carrier_changed) { |
2261 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2262 | } | |
015e08bc EJ |
2263 | } |
2264 | ||
daff3353 EJ |
2265 | if (ofport->may_enable != enable) { |
2266 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2267 | ||
2268 | if (ofproto->has_bundle_action) { | |
2269 | ofproto->need_revalidate = true; | |
2270 | } | |
2271 | } | |
2272 | ||
015e08bc | 2273 | ofport->may_enable = enable; |
abe529af BP |
2274 | } |
2275 | ||
2276 | static void | |
2277 | port_wait(struct ofport_dpif *ofport) | |
2278 | { | |
2279 | if (ofport->cfm) { | |
2280 | cfm_wait(ofport->cfm); | |
2281 | } | |
2282 | } | |
2283 | ||
2284 | static int | |
2285 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2286 | struct ofproto_port *ofproto_port) | |
2287 | { | |
2288 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2289 | struct dpif_port dpif_port; | |
2290 | int error; | |
2291 | ||
2292 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2293 | if (!error) { | |
2294 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2295 | } | |
2296 | return error; | |
2297 | } | |
2298 | ||
2299 | static int | |
2300 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2301 | { | |
2302 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2303 | uint16_t odp_port; | |
2304 | int error; | |
2305 | ||
2306 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2307 | if (!error) { | |
2308 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2309 | } | |
2310 | return error; | |
2311 | } | |
2312 | ||
2313 | static int | |
2314 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2315 | { | |
2316 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2317 | int error; | |
2318 | ||
2319 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2320 | if (!error) { | |
2321 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2322 | if (ofport) { | |
2323 | /* The caller is going to close ofport->up.netdev. If this is a | |
2324 | * bonded port, then the bond is using that netdev, so remove it | |
2325 | * from the bond. The client will need to reconfigure everything | |
2326 | * after deleting ports, so then the slave will get re-added. */ | |
2327 | bundle_remove(&ofport->up); | |
2328 | } | |
2329 | } | |
2330 | return error; | |
2331 | } | |
2332 | ||
6527c598 PS |
2333 | static int |
2334 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2335 | { | |
2336 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2337 | int error; | |
2338 | ||
2339 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2340 | ||
2341 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2342 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2343 | ||
2344 | /* ofproto->stats.tx_packets represents packets that we created | |
2345 | * internally and sent to some port (e.g. packets sent with | |
2346 | * send_packet()). Account for them as if they had come from | |
2347 | * OFPP_LOCAL and got forwarded. */ | |
2348 | ||
2349 | if (stats->rx_packets != UINT64_MAX) { | |
2350 | stats->rx_packets += ofproto->stats.tx_packets; | |
2351 | } | |
2352 | ||
2353 | if (stats->rx_bytes != UINT64_MAX) { | |
2354 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2355 | } | |
2356 | ||
2357 | /* ofproto->stats.rx_packets represents packets that were received on | |
2358 | * some port and we processed internally and dropped (e.g. STP). | |
2359 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2360 | ||
2361 | if (stats->tx_packets != UINT64_MAX) { | |
2362 | stats->tx_packets += ofproto->stats.rx_packets; | |
2363 | } | |
2364 | ||
2365 | if (stats->tx_bytes != UINT64_MAX) { | |
2366 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2367 | } | |
2368 | } | |
2369 | ||
2370 | return error; | |
2371 | } | |
2372 | ||
2373 | /* Account packets for LOCAL port. */ | |
2374 | static void | |
2375 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2376 | size_t tx_size, size_t rx_size) | |
2377 | { | |
2378 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2379 | ||
2380 | if (rx_size) { | |
2381 | ofproto->stats.rx_packets++; | |
2382 | ofproto->stats.rx_bytes += rx_size; | |
2383 | } | |
2384 | if (tx_size) { | |
2385 | ofproto->stats.tx_packets++; | |
2386 | ofproto->stats.tx_bytes += tx_size; | |
2387 | } | |
2388 | } | |
2389 | ||
abe529af BP |
2390 | struct port_dump_state { |
2391 | struct dpif_port_dump dump; | |
2392 | bool done; | |
2393 | }; | |
2394 | ||
2395 | static int | |
2396 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2397 | { | |
2398 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2399 | struct port_dump_state *state; | |
2400 | ||
2401 | *statep = state = xmalloc(sizeof *state); | |
2402 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2403 | state->done = false; | |
2404 | return 0; | |
2405 | } | |
2406 | ||
2407 | static int | |
2408 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2409 | struct ofproto_port *port) | |
2410 | { | |
2411 | struct port_dump_state *state = state_; | |
2412 | struct dpif_port dpif_port; | |
2413 | ||
2414 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2415 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2416 | return 0; | |
2417 | } else { | |
2418 | int error = dpif_port_dump_done(&state->dump); | |
2419 | state->done = true; | |
2420 | return error ? error : EOF; | |
2421 | } | |
2422 | } | |
2423 | ||
2424 | static int | |
2425 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2426 | { | |
2427 | struct port_dump_state *state = state_; | |
2428 | ||
2429 | if (!state->done) { | |
2430 | dpif_port_dump_done(&state->dump); | |
2431 | } | |
2432 | free(state); | |
2433 | return 0; | |
2434 | } | |
2435 | ||
2436 | static int | |
2437 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2438 | { | |
2439 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2440 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2441 | } | |
2442 | ||
2443 | static void | |
2444 | port_poll_wait(const struct ofproto *ofproto_) | |
2445 | { | |
2446 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2447 | dpif_port_poll_wait(ofproto->dpif); | |
2448 | } | |
2449 | ||
2450 | static int | |
2451 | port_is_lacp_current(const struct ofport *ofport_) | |
2452 | { | |
2453 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2454 | return (ofport->bundle && ofport->bundle->lacp | |
2455 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2456 | : -1); | |
2457 | } | |
2458 | \f | |
2459 | /* Upcall handling. */ | |
2460 | ||
501f8d1f BP |
2461 | /* Flow miss batching. |
2462 | * | |
2463 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2464 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2465 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2466 | * more packets, plus possibly installing the flow in the dpif. | |
2467 | * | |
2468 | * So far we only batch the operations that affect flow setup time the most. | |
2469 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2470 | struct flow_miss { | |
2471 | struct hmap_node hmap_node; | |
2472 | struct flow flow; | |
b0f7b9b5 | 2473 | enum odp_key_fitness key_fitness; |
501f8d1f BP |
2474 | const struct nlattr *key; |
2475 | size_t key_len; | |
e84173dc | 2476 | ovs_be16 initial_tci; |
501f8d1f BP |
2477 | struct list packets; |
2478 | }; | |
2479 | ||
2480 | struct flow_miss_op { | |
c2b565b5 | 2481 | struct dpif_op dpif_op; |
b0f7b9b5 | 2482 | struct subfacet *subfacet; |
501f8d1f BP |
2483 | }; |
2484 | ||
62cd7072 BP |
2485 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2486 | * OpenFlow controller as necessary according to their individual | |
29ebe880 | 2487 | * configurations. */ |
62cd7072 | 2488 | static void |
a39edbd4 | 2489 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, |
29ebe880 | 2490 | const struct flow *flow) |
62cd7072 BP |
2491 | { |
2492 | struct ofputil_packet_in pin; | |
2493 | ||
3e3252fa EJ |
2494 | pin.packet = packet->data; |
2495 | pin.packet_len = packet->size; | |
62cd7072 | 2496 | pin.reason = OFPR_NO_MATCH; |
a7349929 | 2497 | pin.controller_id = 0; |
54834960 EJ |
2498 | |
2499 | pin.table_id = 0; | |
2500 | pin.cookie = 0; | |
2501 | ||
62cd7072 | 2502 | pin.send_len = 0; /* not used for flow table misses */ |
5d6c3af0 EJ |
2503 | |
2504 | flow_get_metadata(flow, &pin.fmd); | |
2505 | ||
2506 | /* Registers aren't meaningful on a miss. */ | |
2507 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2508 | ||
d8653c38 | 2509 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); |
62cd7072 BP |
2510 | } |
2511 | ||
abe529af BP |
2512 | static bool |
2513 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2514 | const struct ofpbuf *packet) | |
2515 | { | |
b6e001b6 EJ |
2516 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2517 | ||
2518 | if (!ofport) { | |
2519 | return false; | |
2520 | } | |
2521 | ||
ef9819b5 | 2522 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2523 | if (packet) { |
abe529af BP |
2524 | cfm_process_heartbeat(ofport->cfm, packet); |
2525 | } | |
2526 | return true; | |
b6e001b6 EJ |
2527 | } else if (ofport->bundle && ofport->bundle->lacp |
2528 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2529 | if (packet) { | |
2530 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2531 | } |
da37ebac | 2532 | return true; |
21f7563c JP |
2533 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2534 | if (packet) { | |
2535 | stp_process_packet(ofport, packet); | |
2536 | } | |
2537 | return true; | |
abe529af BP |
2538 | } |
2539 | return false; | |
2540 | } | |
2541 | ||
501f8d1f BP |
2542 | static struct flow_miss * |
2543 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
b0f7b9b5 | 2544 | enum odp_key_fitness key_fitness, |
e84173dc BP |
2545 | const struct nlattr *key, size_t key_len, |
2546 | ovs_be16 initial_tci) | |
abe529af | 2547 | { |
501f8d1f BP |
2548 | uint32_t hash = flow_hash(flow, 0); |
2549 | struct flow_miss *miss; | |
abe529af | 2550 | |
501f8d1f BP |
2551 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2552 | if (flow_equal(&miss->flow, flow)) { | |
2553 | return miss; | |
2554 | } | |
2555 | } | |
abe529af | 2556 | |
501f8d1f BP |
2557 | miss = xmalloc(sizeof *miss); |
2558 | hmap_insert(todo, &miss->hmap_node, hash); | |
2559 | miss->flow = *flow; | |
b0f7b9b5 | 2560 | miss->key_fitness = key_fitness; |
501f8d1f BP |
2561 | miss->key = key; |
2562 | miss->key_len = key_len; | |
e84173dc | 2563 | miss->initial_tci = initial_tci; |
501f8d1f BP |
2564 | list_init(&miss->packets); |
2565 | return miss; | |
2566 | } | |
abe529af | 2567 | |
501f8d1f BP |
2568 | static void |
2569 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2570 | struct flow_miss_op *ops, size_t *n_ops) | |
2571 | { | |
2572 | const struct flow *flow = &miss->flow; | |
b0f7b9b5 | 2573 | struct subfacet *subfacet; |
530a1d91 | 2574 | struct ofpbuf *packet; |
501f8d1f | 2575 | struct facet *facet; |
2b459b83 BP |
2576 | uint32_t hash; |
2577 | ||
2578 | /* The caller must ensure that miss->hmap_node.hash contains | |
2579 | * flow_hash(miss->flow, 0). */ | |
2580 | hash = miss->hmap_node.hash; | |
abe529af | 2581 | |
2b459b83 | 2582 | facet = facet_lookup_valid(ofproto, flow, hash); |
abe529af | 2583 | if (!facet) { |
501f8d1f BP |
2584 | struct rule_dpif *rule; |
2585 | ||
2586 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af | 2587 | if (!rule) { |
9e1fd49b | 2588 | /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */ |
501f8d1f | 2589 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af | 2590 | if (port) { |
9e1fd49b | 2591 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { |
abe529af BP |
2592 | COVERAGE_INC(ofproto_dpif_no_packet_in); |
2593 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
2594 | return; |
2595 | } | |
2596 | } else { | |
2597 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
2598 | flow->in_port); |
2599 | } | |
2600 | ||
29ebe880 EJ |
2601 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
2602 | send_packet_in_miss(ofproto, packet, flow); | |
abe529af BP |
2603 | } |
2604 | ||
abe529af BP |
2605 | return; |
2606 | } | |
2607 | ||
2b459b83 | 2608 | facet = facet_create(rule, flow, hash); |
abe529af BP |
2609 | } |
2610 | ||
15baa734 | 2611 | subfacet = subfacet_create(facet, |
e84173dc BP |
2612 | miss->key_fitness, miss->key, miss->key_len, |
2613 | miss->initial_tci); | |
b0f7b9b5 | 2614 | |
530a1d91 | 2615 | LIST_FOR_EACH (packet, list_node, &miss->packets) { |
67d91f78 | 2616 | struct dpif_flow_stats stats; |
999fba59 EJ |
2617 | struct flow_miss_op *op; |
2618 | struct dpif_execute *execute; | |
67d91f78 | 2619 | |
501f8d1f BP |
2620 | ofproto->n_matches++; |
2621 | ||
2622 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2623 | /* | |
2624 | * Extra-special case for fail-open mode. | |
2625 | * | |
2626 | * We are in fail-open mode and the packet matched the fail-open | |
2627 | * rule, but we are connected to a controller too. We should send | |
2628 | * the packet up to the controller in the hope that it will try to | |
2629 | * set up a flow and thereby allow us to exit fail-open. | |
2630 | * | |
2631 | * See the top-level comment in fail-open.c for more information. | |
2632 | */ | |
29ebe880 | 2633 | send_packet_in_miss(ofproto, packet, flow); |
501f8d1f BP |
2634 | } |
2635 | ||
b95fc6ba | 2636 | if (!facet->may_install || !subfacet->actions) { |
15baa734 | 2637 | subfacet_make_actions(subfacet, packet); |
501f8d1f | 2638 | } |
67d91f78 | 2639 | |
67d91f78 | 2640 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
15baa734 | 2641 | subfacet_update_stats(subfacet, &stats); |
67d91f78 | 2642 | |
8338659a BP |
2643 | if (!subfacet->actions_len) { |
2644 | /* No actions to execute, so skip talking to the dpif. */ | |
2645 | continue; | |
2646 | } | |
2647 | ||
999fba59 EJ |
2648 | if (flow->vlan_tci != subfacet->initial_tci) { |
2649 | /* This packet was received on a VLAN splinter port. We added | |
2650 | * a VLAN to the packet to make the packet resemble the flow, | |
2651 | * but the actions were composed assuming that the packet | |
2652 | * contained no VLAN. So, we must remove the VLAN header from | |
2653 | * the packet before trying to execute the actions. */ | |
2654 | eth_pop_vlan(packet); | |
501f8d1f | 2655 | } |
999fba59 EJ |
2656 | |
2657 | op = &ops[(*n_ops)++]; | |
c2b565b5 | 2658 | execute = &op->dpif_op.u.execute; |
999fba59 | 2659 | op->subfacet = subfacet; |
c2b565b5 | 2660 | op->dpif_op.type = DPIF_OP_EXECUTE; |
999fba59 EJ |
2661 | execute->key = miss->key; |
2662 | execute->key_len = miss->key_len; | |
2663 | execute->actions = (facet->may_install | |
2664 | ? subfacet->actions | |
2665 | : xmemdup(subfacet->actions, | |
2666 | subfacet->actions_len)); | |
2667 | execute->actions_len = subfacet->actions_len; | |
2668 | execute->packet = packet; | |
501f8d1f BP |
2669 | } |
2670 | ||
b0f7b9b5 | 2671 | if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) { |
501f8d1f | 2672 | struct flow_miss_op *op = &ops[(*n_ops)++]; |
c2b565b5 | 2673 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; |
501f8d1f | 2674 | |
b0f7b9b5 | 2675 | op->subfacet = subfacet; |
c2b565b5 | 2676 | op->dpif_op.type = DPIF_OP_FLOW_PUT; |
501f8d1f BP |
2677 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2678 | put->key = miss->key; | |
2679 | put->key_len = miss->key_len; | |
b95fc6ba BP |
2680 | put->actions = subfacet->actions; |
2681 | put->actions_len = subfacet->actions_len; | |
501f8d1f BP |
2682 | put->stats = NULL; |
2683 | } | |
2684 | } | |
2685 | ||
e2a6ca36 BP |
2686 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of |
2687 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
2688 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
2689 | * what a flow key should contain. | |
2690 | * | |
2691 | * This function also includes some logic to help make VLAN splinters | |
2692 | * transparent to the rest of the upcall processing logic. In particular, if | |
2693 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
2694 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
2695 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
2696 | * | |
2697 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
2698 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
2699 | * (This differs from the value returned in flow->vlan_tci only for packets | |
2700 | * received on VLAN splinters.) | |
2701 | */ | |
e84173dc | 2702 | static enum odp_key_fitness |
52a90c29 | 2703 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, |
e84173dc | 2704 | const struct nlattr *key, size_t key_len, |
e2a6ca36 BP |
2705 | struct flow *flow, ovs_be16 *initial_tci, |
2706 | struct ofpbuf *packet) | |
e84173dc BP |
2707 | { |
2708 | enum odp_key_fitness fitness; | |
52a90c29 BP |
2709 | uint16_t realdev; |
2710 | int vid; | |
e84173dc BP |
2711 | |
2712 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
2713 | if (fitness == ODP_FIT_ERROR) { | |
2714 | return fitness; | |
2715 | } | |
2716 | *initial_tci = flow->vlan_tci; | |
2717 | ||
52a90c29 BP |
2718 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); |
2719 | if (realdev) { | |
2720 | /* Cause the flow to be processed as if it came in on the real device | |
2721 | * with the VLAN device's VLAN ID. */ | |
2722 | flow->in_port = realdev; | |
2723 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
e2a6ca36 BP |
2724 | if (packet) { |
2725 | /* Make the packet resemble the flow, so that it gets sent to an | |
2726 | * OpenFlow controller properly, so that it looks correct for | |
2727 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
2728 | * if it is called on 'packet'. | |
2729 | * | |
2730 | * The allocated space inside 'packet' probably also contains | |
2731 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
2732 | * struct dpif_upcall (see the large comment on that structure | |
2733 | * definition), so pushing data on 'packet' is in general not a | |
2734 | * good idea since it could overwrite 'key' or free it as a side | |
2735 | * effect. However, it's OK in this special case because we know | |
2736 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
2737 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
2738 | * since we don't need that header anymore. */ | |
2739 | eth_push_vlan(packet, flow->vlan_tci); | |
2740 | } | |
52a90c29 BP |
2741 | |
2742 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
2743 | if (fitness == ODP_FIT_PERFECT) { | |
2744 | fitness = ODP_FIT_TOO_MUCH; | |
2745 | } | |
2746 | } | |
2747 | ||
e84173dc BP |
2748 | return fitness; |
2749 | } | |
2750 | ||
501f8d1f BP |
2751 | static void |
2752 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2753 | size_t n_upcalls) | |
2754 | { | |
2755 | struct dpif_upcall *upcall; | |
2756 | struct flow_miss *miss, *next_miss; | |
2757 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
c2b565b5 | 2758 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; |
501f8d1f BP |
2759 | struct hmap todo; |
2760 | size_t n_ops; | |
2761 | size_t i; | |
2762 | ||
2763 | if (!n_upcalls) { | |
2764 | return; | |
2765 | } | |
2766 | ||
2767 | /* Construct the to-do list. | |
2768 | * | |
2769 | * This just amounts to extracting the flow from each packet and sticking | |
2770 | * the packets that have the same flow in the same "flow_miss" structure so | |
2771 | * that we can process them together. */ | |
2772 | hmap_init(&todo); | |
2773 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
b0f7b9b5 | 2774 | enum odp_key_fitness fitness; |
501f8d1f | 2775 | struct flow_miss *miss; |
e84173dc | 2776 | ovs_be16 initial_tci; |
501f8d1f BP |
2777 | struct flow flow; |
2778 | ||
b0f7b9b5 BP |
2779 | /* Obtain metadata and check userspace/kernel agreement on flow match, |
2780 | * then set 'flow''s header pointers. */ | |
e84173dc BP |
2781 | fitness = ofproto_dpif_extract_flow_key(ofproto, |
2782 | upcall->key, upcall->key_len, | |
e2a6ca36 BP |
2783 | &flow, &initial_tci, |
2784 | upcall->packet); | |
b0f7b9b5 BP |
2785 | if (fitness == ODP_FIT_ERROR) { |
2786 | continue; | |
2787 | } | |
deedf7e7 | 2788 | flow_extract(upcall->packet, flow.skb_priority, flow.tun_id, |
abff858b | 2789 | flow.in_port, &flow); |
501f8d1f | 2790 | |
21f7563c | 2791 | /* Handle 802.1ag, LACP, and STP specially. */ |
501f8d1f | 2792 | if (process_special(ofproto, &flow, upcall->packet)) { |
6527c598 PS |
2793 | ofproto_update_local_port_stats(&ofproto->up, |
2794 | 0, upcall->packet->size); | |
501f8d1f BP |
2795 | ofproto->n_matches++; |
2796 | continue; | |
2797 | } | |
2798 | ||
2799 | /* Add other packets to a to-do list. */ | |
b0f7b9b5 | 2800 | miss = flow_miss_create(&todo, &flow, fitness, |
e84173dc | 2801 | upcall->key, upcall->key_len, initial_tci); |
501f8d1f BP |
2802 | list_push_back(&miss->packets, &upcall->packet->list_node); |
2803 | } | |
2804 | ||
2805 | /* Process each element in the to-do list, constructing the set of | |
2806 | * operations to batch. */ | |
2807 | n_ops = 0; | |
33bb0caa | 2808 | HMAP_FOR_EACH (miss, hmap_node, &todo) { |
501f8d1f | 2809 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); |
abe529af | 2810 | } |
501f8d1f | 2811 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
501f8d1f BP |
2812 | |
2813 | /* Execute batch. */ | |
2814 | for (i = 0; i < n_ops; i++) { | |
2815 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2816 | } | |
2817 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2818 | ||
2819 | /* Free memory and update facets. */ | |
2820 | for (i = 0; i < n_ops; i++) { | |
2821 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
2822 | struct dpif_execute *execute; | |
501f8d1f BP |
2823 | |
2824 | switch (op->dpif_op.type) { | |
2825 | case DPIF_OP_EXECUTE: | |
c2b565b5 | 2826 | execute = &op->dpif_op.u.execute; |
b95fc6ba | 2827 | if (op->subfacet->actions != execute->actions) { |
501f8d1f BP |
2828 | free((struct nlattr *) execute->actions); |
2829 | } | |
501f8d1f | 2830 | break; |
abe529af | 2831 | |
501f8d1f | 2832 | case DPIF_OP_FLOW_PUT: |
c2b565b5 | 2833 | if (!op->dpif_op.error) { |
b0f7b9b5 | 2834 | op->subfacet->installed = true; |
501f8d1f BP |
2835 | } |
2836 | break; | |
b99d3cee BP |
2837 | |
2838 | case DPIF_OP_FLOW_DEL: | |
2839 | NOT_REACHED(); | |
501f8d1f BP |
2840 | } |
2841 | } | |
33bb0caa | 2842 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { |
33bb0caa BP |
2843 | hmap_remove(&todo, &miss->hmap_node); |
2844 | free(miss); | |
2845 | } | |
2846 | hmap_destroy(&todo); | |
abe529af BP |
2847 | } |
2848 | ||
2849 | static void | |
6ff686f2 PS |
2850 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2851 | struct dpif_upcall *upcall) | |
abe529af | 2852 | { |
6ff686f2 | 2853 | struct user_action_cookie cookie; |
e84173dc BP |
2854 | enum odp_key_fitness fitness; |
2855 | ovs_be16 initial_tci; | |
2856 | struct flow flow; | |
abe529af | 2857 | |
6ff686f2 | 2858 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2859 | |
e84173dc BP |
2860 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, |
2861 | upcall->key_len, &flow, | |
e2a6ca36 | 2862 | &initial_tci, upcall->packet); |
e84173dc BP |
2863 | if (fitness == ODP_FIT_ERROR) { |
2864 | return; | |
2865 | } | |
2866 | ||
6ff686f2 | 2867 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af | 2868 | if (ofproto->sflow) { |
e84173dc BP |
2869 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, |
2870 | &cookie); | |
abe529af | 2871 | } |
6ff686f2 PS |
2872 | } else { |
2873 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2874 | } | |
2875 | } | |
2876 | ||
9b16c439 BP |
2877 | static int |
2878 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
6ff686f2 | 2879 | { |
9b16c439 | 2880 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
90a7c55e BP |
2881 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; |
2882 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
2883 | int n_processed; | |
9b16c439 BP |
2884 | int n_misses; |
2885 | int i; | |
abe529af | 2886 | |
90a7c55e | 2887 | assert(max_batch <= FLOW_MISS_MAX_BATCH); |
abe529af | 2888 | |
90a7c55e | 2889 | n_processed = 0; |
9b16c439 | 2890 | n_misses = 0; |
90a7c55e | 2891 | for (n_processed = 0; n_processed < max_batch; n_processed++) { |
9b16c439 | 2892 | struct dpif_upcall *upcall = &misses[n_misses]; |
90a7c55e | 2893 | struct ofpbuf *buf = &miss_bufs[n_misses]; |
9b16c439 BP |
2894 | int error; |
2895 | ||
90a7c55e BP |
2896 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], |
2897 | sizeof miss_buf_stubs[n_misses]); | |
2898 | error = dpif_recv(ofproto->dpif, upcall, buf); | |
9b16c439 | 2899 | if (error) { |
90a7c55e | 2900 | ofpbuf_uninit(buf); |
9b16c439 BP |
2901 | break; |
2902 | } | |
2903 | ||
2904 | switch (upcall->type) { | |
2905 | case DPIF_UC_ACTION: | |
2906 | handle_userspace_upcall(ofproto, upcall); | |
90a7c55e | 2907 | ofpbuf_uninit(buf); |
9b16c439 BP |
2908 | break; |
2909 | ||
2910 | case DPIF_UC_MISS: | |
2911 | /* Handle it later. */ | |
2912 | n_misses++; | |
2913 | break; | |
2914 | ||
2915 | case DPIF_N_UC_TYPES: | |
2916 | default: | |
2917 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, | |
2918 | upcall->type); | |
2919 | break; | |
2920 | } | |
abe529af | 2921 | } |
9b16c439 BP |
2922 | |
2923 | handle_miss_upcalls(ofproto, misses, n_misses); | |
90a7c55e BP |
2924 | for (i = 0; i < n_misses; i++) { |
2925 | ofpbuf_uninit(&miss_bufs[i]); | |
2926 | } | |
9b16c439 | 2927 | |
90a7c55e | 2928 | return n_processed; |
abe529af BP |
2929 | } |
2930 | \f | |
2931 | /* Flow expiration. */ | |
2932 | ||
b0f7b9b5 | 2933 | static int subfacet_max_idle(const struct ofproto_dpif *); |
abe529af BP |
2934 | static void update_stats(struct ofproto_dpif *); |
2935 | static void rule_expire(struct rule_dpif *); | |
b0f7b9b5 | 2936 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); |
abe529af BP |
2937 | |
2938 | /* This function is called periodically by run(). Its job is to collect | |
2939 | * updates for the flows that have been installed into the datapath, most | |
2940 | * importantly when they last were used, and then use that information to | |
2941 | * expire flows that have not been used recently. | |
2942 | * | |
2943 | * Returns the number of milliseconds after which it should be called again. */ | |
2944 | static int | |
2945 | expire(struct ofproto_dpif *ofproto) | |
2946 | { | |
2947 | struct rule_dpif *rule, *next_rule; | |
d0918789 | 2948 | struct oftable *table; |
abe529af BP |
2949 | int dp_max_idle; |
2950 | ||
2951 | /* Update stats for each flow in the datapath. */ | |
2952 | update_stats(ofproto); | |
2953 | ||
b0f7b9b5 BP |
2954 | /* Expire subfacets that have been idle too long. */ |
2955 | dp_max_idle = subfacet_max_idle(ofproto); | |
2956 | expire_subfacets(ofproto, dp_max_idle); | |
abe529af BP |
2957 | |
2958 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2959 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2960 | struct cls_cursor cursor; | |
2961 | ||
d0918789 | 2962 | cls_cursor_init(&cursor, &table->cls, NULL); |
0697b5c3 BP |
2963 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
2964 | rule_expire(rule); | |
2965 | } | |
abe529af BP |
2966 | } |
2967 | ||
2968 | /* All outstanding data in existing flows has been accounted, so it's a | |
2969 | * good time to do bond rebalancing. */ | |
2970 | if (ofproto->has_bonded_bundles) { | |
2971 | struct ofbundle *bundle; | |
2972 | ||
2973 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2974 | if (bundle->bond) { | |
2975 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2976 | } | |
2977 | } | |
2978 | } | |
2979 | ||
2980 | return MIN(dp_max_idle, 1000); | |
2981 | } | |
2982 | ||
2983 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2984 | * | |
2985 | * This function also pushes statistics updates to rules which each facet | |
2986 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2987 | * facet changes the rule it resubmits into at some time in between | |
2988 | * update_stats() runs, it is possible that statistics accrued to the | |
2989 | * old rule will be incorrectly attributed to the new rule. This could be | |
2990 | * avoided by calling update_stats() whenever rules are created or | |
2991 | * deleted. However, the performance impact of making so many calls to the | |
2992 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2993 | */ | |
2994 | static void | |
2995 | update_stats(struct ofproto_dpif *p) | |
2996 | { | |
2997 | const struct dpif_flow_stats *stats; | |
2998 | struct dpif_flow_dump dump; | |
2999 | const struct nlattr *key; | |
3000 | size_t key_len; | |
3001 | ||
3002 | dpif_flow_dump_start(&dump, p->dpif); | |
3003 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
b0f7b9b5 | 3004 | struct subfacet *subfacet; |
abe529af | 3005 | |
6a542738 | 3006 | subfacet = subfacet_find(p, key, key_len); |
b0f7b9b5 BP |
3007 | if (subfacet && subfacet->installed) { |
3008 | struct facet *facet = subfacet->facet; | |
abe529af | 3009 | |
b0f7b9b5 BP |
3010 | if (stats->n_packets >= subfacet->dp_packet_count) { |
3011 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
abe529af BP |
3012 | facet->packet_count += extra; |
3013 | } else { | |
3014 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
3015 | } | |
3016 | ||
b0f7b9b5 BP |
3017 | if (stats->n_bytes >= subfacet->dp_byte_count) { |
3018 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
abe529af BP |
3019 | } else { |
3020 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
3021 | } | |
3022 | ||
b0f7b9b5 BP |
3023 | subfacet->dp_packet_count = stats->n_packets; |
3024 | subfacet->dp_byte_count = stats->n_bytes; | |
abe529af | 3025 | |
0e553d9c BP |
3026 | facet->tcp_flags |= stats->tcp_flags; |
3027 | ||
15baa734 | 3028 | subfacet_update_time(subfacet, stats->used); |
3de9590b BP |
3029 | if (facet->accounted_bytes < facet->byte_count) { |
3030 | facet_learn(facet); | |
3031 | facet_account(facet); | |
3032 | facet->accounted_bytes = facet->byte_count; | |
3033 | } | |
abe529af BP |
3034 | facet_push_stats(facet); |
3035 | } else { | |
6a542738 PS |
3036 | if (!VLOG_DROP_WARN(&rl)) { |
3037 | struct ds s; | |
3038 | ||
3039 | ds_init(&s); | |
3040 | odp_flow_key_format(key, key_len, &s); | |
3041 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
3042 | ds_destroy(&s); | |
3043 | } | |
3044 | ||
3045 | COVERAGE_INC(facet_unexpected); | |
b0f7b9b5 BP |
3046 | /* There's a flow in the datapath that we know nothing about, or a |
3047 | * flow that shouldn't be installed but was anyway. Delete it. */ | |
abe529af BP |
3048 | dpif_flow_del(p->dpif, key, key_len, NULL); |
3049 | } | |
3050 | } | |
3051 | dpif_flow_dump_done(&dump); | |
3052 | } | |
3053 | ||
3054 | /* Calculates and returns the number of milliseconds of idle time after which | |
b0f7b9b5 BP |
3055 | * subfacets should expire from the datapath. When a subfacet expires, we fold |
3056 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3057 | * fold its statistic into its rule. */ | |
abe529af | 3058 | static int |
b0f7b9b5 | 3059 | subfacet_max_idle(const struct ofproto_dpif *ofproto) |
abe529af BP |
3060 | { |
3061 | /* | |
3062 | * Idle time histogram. | |
3063 | * | |
b0f7b9b5 BP |
3064 | * Most of the time a switch has a relatively small number of subfacets. |
3065 | * When this is the case we might as well keep statistics for all of them | |
3066 | * in userspace and to cache them in the kernel datapath for performance as | |
abe529af BP |
3067 | * well. |
3068 | * | |
b0f7b9b5 | 3069 | * As the number of subfacets increases, the memory required to maintain |
abe529af | 3070 | * statistics about them in userspace and in the kernel becomes |
b0f7b9b5 BP |
3071 | * significant. However, with a large number of subfacets it is likely |
3072 | * that only a few of them are "heavy hitters" that consume a large amount | |
3073 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3074 | * the kernel and maintaining in userspaces; other subfacets we can | |
3075 | * discard. | |
abe529af BP |
3076 | * |
3077 | * The technique used to compute the idle time is to build a histogram with | |
b0f7b9b5 | 3078 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet |
abe529af BP |
3079 | * that is installed in the kernel gets dropped in the appropriate bucket. |
3080 | * After the histogram has been built, we compute the cutoff so that only | |
b0f7b9b5 | 3081 | * the most-recently-used 1% of subfacets (but at least |
084f5290 | 3082 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least |
b0f7b9b5 BP |
3083 | * the most-recently-used bucket of subfacets is kept, so actually an |
3084 | * arbitrary number of subfacets can be kept in any given expiration run | |
084f5290 SH |
3085 | * (though the next run will delete most of those unless they receive |
3086 | * additional data). | |
abe529af | 3087 | * |
b0f7b9b5 BP |
3088 | * This requires a second pass through the subfacets, in addition to the |
3089 | * pass made by update_stats(), because the former function never looks at | |
3090 | * uninstallable subfacets. | |
abe529af BP |
3091 | */ |
3092 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3093 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3094 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 3095 | int total, subtotal, bucket; |
b0f7b9b5 | 3096 | struct subfacet *subfacet; |
abe529af BP |
3097 | long long int now; |
3098 | int i; | |
3099 | ||
b0f7b9b5 | 3100 | total = hmap_count(&ofproto->subfacets); |
084f5290 | 3101 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
3102 | return N_BUCKETS * BUCKET_WIDTH; |
3103 | } | |
3104 | ||
3105 | /* Build histogram. */ | |
3106 | now = time_msec(); | |
b0f7b9b5 BP |
3107 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { |
3108 | long long int idle = now - subfacet->used; | |
abe529af BP |
3109 | int bucket = (idle <= 0 ? 0 |
3110 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3111 | : (unsigned int) idle / BUCKET_WIDTH); | |
3112 | buckets[bucket]++; | |
3113 | } | |
3114 | ||
3115 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
3116 | subtotal = bucket = 0; |
3117 | do { | |
3118 | subtotal += buckets[bucket++]; | |
084f5290 SH |
3119 | } while (bucket < N_BUCKETS && |
3120 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
3121 | |
3122 | if (VLOG_IS_DBG_ENABLED()) { | |
3123 | struct ds s; | |
3124 | ||
3125 | ds_init(&s); | |
3126 | ds_put_cstr(&s, "keep"); | |
3127 | for (i = 0; i < N_BUCKETS; i++) { | |
3128 | if (i == bucket) { | |
3129 | ds_put_cstr(&s, ", drop"); | |
3130 | } | |
3131 | if (buckets[i]) { | |
3132 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3133 | } | |
3134 | } | |
3135 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3136 | ds_destroy(&s); | |
3137 | } | |
3138 | ||
3139 | return bucket * BUCKET_WIDTH; | |
3140 | } | |
3141 | ||
b99d3cee BP |
3142 | enum { EXPIRE_MAX_BATCH = 50 }; |
3143 | ||
3144 | static void | |
3145 | expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) | |
3146 | { | |
3147 | struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; | |
3148 | struct dpif_op ops[EXPIRE_MAX_BATCH]; | |
3149 | struct dpif_op *opsp[EXPIRE_MAX_BATCH]; | |
3150 | struct ofpbuf keys[EXPIRE_MAX_BATCH]; | |
3151 | struct dpif_flow_stats stats[EXPIRE_MAX_BATCH]; | |
3152 | int i; | |
3153 | ||
3154 | for (i = 0; i < n; i++) { | |
3155 | ops[i].type = DPIF_OP_FLOW_DEL; | |
3156 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
3157 | ops[i].u.flow_del.key = keys[i].data; | |
3158 | ops[i].u.flow_del.key_len = keys[i].size; | |
3159 | ops[i].u.flow_del.stats = &stats[i]; | |
3160 | opsp[i] = &ops[i]; | |
3161 | } | |
3162 | ||
3163 | dpif_operate(ofproto->dpif, opsp, n); | |
3164 | for (i = 0; i < n; i++) { | |
3165 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
3166 | subfacets[i]->installed = false; | |
3167 | subfacet_destroy(subfacets[i]); | |
3168 | } | |
3169 | } | |
3170 | ||
abe529af | 3171 | static void |
b0f7b9b5 | 3172 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) |
abe529af BP |
3173 | { |
3174 | long long int cutoff = time_msec() - dp_max_idle; | |
b99d3cee | 3175 | |
b0f7b9b5 | 3176 | struct subfacet *subfacet, *next_subfacet; |
b99d3cee BP |
3177 | struct subfacet *batch[EXPIRE_MAX_BATCH]; |
3178 | int n_batch; | |
abe529af | 3179 | |
b99d3cee | 3180 | n_batch = 0; |
b0f7b9b5 BP |
3181 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, |
3182 | &ofproto->subfacets) { | |
3183 | if (subfacet->used < cutoff) { | |
b99d3cee BP |
3184 | if (subfacet->installed) { |
3185 | batch[n_batch++] = subfacet; | |
3186 | if (n_batch >= EXPIRE_MAX_BATCH) { | |
3187 | expire_batch(ofproto, batch, n_batch); | |
3188 | n_batch = 0; | |
3189 | } | |
3190 | } else { | |
3191 | subfacet_destroy(subfacet); | |
3192 | } | |
abe529af BP |
3193 | } |
3194 | } | |
b99d3cee BP |
3195 | |
3196 | if (n_batch > 0) { | |
3197 | expire_batch(ofproto, batch, n_batch); | |
3198 | } | |
abe529af BP |
3199 | } |
3200 | ||
3201 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3202 | * then delete it entirely. */ | |
3203 | static void | |
3204 | rule_expire(struct rule_dpif *rule) | |
3205 | { | |
abe529af BP |
3206 | struct facet *facet, *next_facet; |
3207 | long long int now; | |
3208 | uint8_t reason; | |
3209 | ||
3210 | /* Has 'rule' expired? */ | |
3211 | now = time_msec(); | |
3212 | if (rule->up.hard_timeout | |
308881af | 3213 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af | 3214 | reason = OFPRR_HARD_TIMEOUT; |
8ea6ac3e | 3215 | } else if (rule->up.idle_timeout |
1745cd08 | 3216 | && now > rule->up.used + rule->up.idle_timeout * 1000) { |
abe529af BP |
3217 | reason = OFPRR_IDLE_TIMEOUT; |
3218 | } else { | |
3219 | return; | |
3220 | } | |
3221 | ||
3222 | COVERAGE_INC(ofproto_dpif_expired); | |
3223 | ||
3224 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3225 | * timeout, because that only happens when the rule has no facets left.) */ | |
3226 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
15baa734 | 3227 | facet_remove(facet); |
abe529af BP |
3228 | } |
3229 | ||
3230 | /* Get rid of the rule. */ | |
3231 | ofproto_rule_expire(&rule->up, reason); | |
3232 | } | |
3233 | \f | |
3234 | /* Facets. */ | |
3235 | ||
f3827897 | 3236 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
3237 | * |
3238 | * The caller must already have determined that no facet with an identical | |
3239 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
3240 | * the ofproto's classifier table. |
3241 | * | |
2b459b83 BP |
3242 | * 'hash' must be the return value of flow_hash(flow, 0). |
3243 | * | |
b0f7b9b5 BP |
3244 | * The facet will initially have no subfacets. The caller should create (at |
3245 | * least) one subfacet with subfacet_create(). */ | |
abe529af | 3246 | static struct facet * |
2b459b83 | 3247 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) |
abe529af BP |
3248 | { |
3249 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3250 | struct facet *facet; | |
3251 | ||
3252 | facet = xzalloc(sizeof *facet); | |
3253 | facet->used = time_msec(); | |
2b459b83 | 3254 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); |
abe529af BP |
3255 | list_push_back(&rule->facets, &facet->list_node); |
3256 | facet->rule = rule; | |
3257 | facet->flow = *flow; | |
b0f7b9b5 | 3258 | list_init(&facet->subfacets); |
abe529af BP |
3259 | netflow_flow_init(&facet->nf_flow); |
3260 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3261 | ||
abe529af BP |
3262 | return facet; |
3263 | } | |
3264 | ||
3265 | static void | |
3266 | facet_free(struct facet *facet) | |
3267 | { | |
abe529af BP |
3268 | free(facet); |
3269 | } | |
3270 | ||
3d9e05f8 BP |
3271 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
3272 | * 'packet', which arrived on 'in_port'. | |
3273 | * | |
3274 | * Takes ownership of 'packet'. */ | |
3275 | static bool | |
3276 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3277 | const struct nlattr *odp_actions, size_t actions_len, | |
3278 | struct ofpbuf *packet) | |
3279 | { | |
3280 | struct odputil_keybuf keybuf; | |
3281 | struct ofpbuf key; | |
3282 | int error; | |
3283 | ||
6ff686f2 PS |
3284 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3285 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 3286 | |
6ff686f2 PS |
3287 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
3288 | odp_actions, actions_len, packet); | |
80e5eed9 | 3289 | |
6ff686f2 PS |
3290 | ofpbuf_delete(packet); |
3291 | return !error; | |
abe529af BP |
3292 | } |
3293 | ||
abe529af BP |
3294 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
3295 | * | |
3296 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
b0f7b9b5 | 3297 | * rule's statistics, via subfacet_uninstall(). |
abe529af BP |
3298 | * |
3299 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3300 | */ | |
3301 | static void | |
15baa734 | 3302 | facet_remove(struct facet *facet) |
abe529af | 3303 | { |
15baa734 | 3304 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3305 | struct subfacet *subfacet, *next_subfacet; |
3306 | ||
551a2f6c BP |
3307 | assert(!list_is_empty(&facet->subfacets)); |
3308 | ||
3309 | /* First uninstall all of the subfacets to get final statistics. */ | |
3310 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
15baa734 | 3311 | subfacet_uninstall(subfacet); |
551a2f6c BP |
3312 | } |
3313 | ||
3314 | /* Flush the final stats to the rule. | |
3315 | * | |
3316 | * This might require us to have at least one subfacet around so that we | |
3317 | * can use its actions for accounting in facet_account(), which is why we | |
3318 | * have uninstalled but not yet destroyed the subfacets. */ | |
15baa734 | 3319 | facet_flush_stats(facet); |
551a2f6c BP |
3320 | |
3321 | /* Now we're really all done so destroy everything. */ | |
b0f7b9b5 BP |
3322 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, |
3323 | &facet->subfacets) { | |
15baa734 | 3324 | subfacet_destroy__(subfacet); |
b0f7b9b5 | 3325 | } |
abe529af BP |
3326 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
3327 | list_remove(&facet->list_node); | |
3328 | facet_free(facet); | |
3329 | } | |
3330 | ||
3de9590b BP |
3331 | /* Feed information from 'facet' back into the learning table to keep it in |
3332 | * sync with what is actually flowing through the datapath. */ | |
abe529af | 3333 | static void |
3de9590b | 3334 | facet_learn(struct facet *facet) |
abe529af | 3335 | { |
15baa734 | 3336 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
3de9590b | 3337 | struct action_xlate_ctx ctx; |
abe529af | 3338 | |
3de9590b BP |
3339 | if (!facet->has_learn |
3340 | && !facet->has_normal | |
3341 | && (!facet->has_fin_timeout | |
3342 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
abe529af BP |
3343 | return; |
3344 | } | |
abe529af | 3345 | |
3de9590b BP |
3346 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
3347 | facet->flow.vlan_tci, | |
3348 | facet->rule, facet->tcp_flags, NULL); | |
3349 | ctx.may_learn = true; | |
050ac423 BP |
3350 | xlate_actions_for_side_effects(&ctx, facet->rule->up.actions, |
3351 | facet->rule->up.n_actions); | |
3de9590b BP |
3352 | } |
3353 | ||
3354 | static void | |
3355 | facet_account(struct facet *facet) | |
3356 | { | |
3357 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3358 | struct subfacet *subfacet; | |
3359 | const struct nlattr *a; | |
3360 | unsigned int left; | |
3361 | ovs_be16 vlan_tci; | |
3362 | uint64_t n_bytes; | |
abe529af | 3363 | |
75a75043 | 3364 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
3365 | return; |
3366 | } | |
3de9590b | 3367 | n_bytes = facet->byte_count - facet->accounted_bytes; |
d78be13b BP |
3368 | |
3369 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3370 | * as a basis. We also need to track the actual VLAN on which the packet | |
3371 | * is going to be sent to ensure that it matches the one passed to | |
3372 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
b95fc6ba BP |
3373 | * hash bucket.) |
3374 | * | |
3375 | * We use the actions from an arbitrary subfacet because they should all | |
3376 | * be equally valid for our purpose. */ | |
3377 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3378 | struct subfacet, list_node); | |
d78be13b | 3379 | vlan_tci = facet->flow.vlan_tci; |
b95fc6ba BP |
3380 | NL_ATTR_FOR_EACH_UNSAFE (a, left, |
3381 | subfacet->actions, subfacet->actions_len) { | |
fea393b1 | 3382 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 3383 | struct ofport_dpif *port; |
abe529af | 3384 | |
d78be13b | 3385 | switch (nl_attr_type(a)) { |
df2c07f4 | 3386 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
3387 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
3388 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 3389 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 3390 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 3391 | } |
d78be13b BP |
3392 | break; |
3393 | ||
fea393b1 BP |
3394 | case OVS_ACTION_ATTR_POP_VLAN: |
3395 | vlan_tci = htons(0); | |
d78be13b BP |
3396 | break; |
3397 | ||
fea393b1 BP |
3398 | case OVS_ACTION_ATTR_PUSH_VLAN: |
3399 | vlan = nl_attr_get(a); | |
3400 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 3401 | break; |
abe529af BP |
3402 | } |
3403 | } | |
3404 | } | |
3405 | ||
abe529af BP |
3406 | /* Returns true if the only action for 'facet' is to send to the controller. |
3407 | * (We don't report NetFlow expiration messages for such facets because they | |
3408 | * are just part of the control logic for the network, not real traffic). */ | |
3409 | static bool | |
3410 | facet_is_controller_flow(struct facet *facet) | |
3411 | { | |
3412 | return (facet | |
3413 | && facet->rule->up.n_actions == 1 | |
3414 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
3415 | htons(OFPP_CONTROLLER))); | |
3416 | } | |
3417 | ||
3418 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3419 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3420 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3421 | * its packet and byte counts before this function is called. */ | |
3422 | static void | |
15baa734 | 3423 | facet_flush_stats(struct facet *facet) |
abe529af | 3424 | { |
15baa734 | 3425 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3426 | struct subfacet *subfacet; |
3427 | ||
3428 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3429 | assert(!subfacet->dp_byte_count); | |
3430 | assert(!subfacet->dp_packet_count); | |
3431 | } | |
abe529af BP |
3432 | |
3433 | facet_push_stats(facet); | |
3de9590b BP |
3434 | if (facet->accounted_bytes < facet->byte_count) { |
3435 | facet_account(facet); | |
3436 | facet->accounted_bytes = facet->byte_count; | |
3437 | } | |
abe529af BP |
3438 | |
3439 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3440 | struct ofexpired expired; | |
3441 | expired.flow = facet->flow; | |
3442 | expired.packet_count = facet->packet_count; | |
3443 | expired.byte_count = facet->byte_count; | |
3444 | expired.used = facet->used; | |
3445 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3446 | } | |
3447 | ||
3448 | facet->rule->packet_count += facet->packet_count; | |
3449 | facet->rule->byte_count += facet->byte_count; | |
3450 | ||
3451 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3452 | * reinstalled. */ | |
bbb5d219 | 3453 | facet_reset_counters(facet); |
abe529af BP |
3454 | |
3455 | netflow_flow_clear(&facet->nf_flow); | |
0e553d9c | 3456 | facet->tcp_flags = 0; |
abe529af BP |
3457 | } |
3458 | ||
3459 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3460 | * Returns it if found, otherwise a null pointer. | |
3461 | * | |
2b459b83 BP |
3462 | * 'hash' must be the return value of flow_hash(flow, 0). |
3463 | * | |
abe529af BP |
3464 | * The returned facet might need revalidation; use facet_lookup_valid() |
3465 | * instead if that is important. */ | |
3466 | static struct facet * | |
2b459b83 BP |
3467 | facet_find(struct ofproto_dpif *ofproto, |
3468 | const struct flow *flow, uint32_t hash) | |
abe529af BP |
3469 | { |
3470 | struct facet *facet; | |
3471 | ||
2b459b83 | 3472 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { |
abe529af BP |
3473 | if (flow_equal(flow, &facet->flow)) { |
3474 | return facet; | |
3475 | } | |
3476 | } | |
3477 | ||
3478 | return NULL; | |
3479 | } | |
3480 | ||
3481 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3482 | * Returns it if found, otherwise a null pointer. | |
3483 | * | |
2b459b83 BP |
3484 | * 'hash' must be the return value of flow_hash(flow, 0). |
3485 | * | |
abe529af BP |
3486 | * The returned facet is guaranteed to be valid. */ |
3487 | static struct facet * | |
2b459b83 BP |
3488 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, |
3489 | uint32_t hash) | |
abe529af | 3490 | { |
2b459b83 | 3491 | struct facet *facet = facet_find(ofproto, flow, hash); |
abe529af BP |
3492 | |
3493 | /* The facet we found might not be valid, since we could be in need of | |
3494 | * revalidation. If it is not valid, don't return it. */ | |
3495 | if (facet | |
0e4b3771 BP |
3496 | && (ofproto->need_revalidate |
3497 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
15baa734 | 3498 | && !facet_revalidate(facet)) { |
abe529af BP |
3499 | COVERAGE_INC(facet_invalidated); |
3500 | return NULL; | |
3501 | } | |
3502 | ||
3503 | return facet; | |
3504 | } | |
3505 | ||
6814e51f BP |
3506 | static bool |
3507 | facet_check_consistency(struct facet *facet) | |
3508 | { | |
3509 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3510 | ||
3511 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3512 | ||
050ac423 BP |
3513 | uint64_t odp_actions_stub[1024 / 8]; |
3514 | struct ofpbuf odp_actions; | |
3515 | ||
6814e51f BP |
3516 | struct rule_dpif *rule; |
3517 | struct subfacet *subfacet; | |
c53e1132 | 3518 | bool may_log = false; |
6814e51f BP |
3519 | bool ok; |
3520 | ||
3521 | /* Check the rule for consistency. */ | |
3522 | rule = rule_dpif_lookup(ofproto, &facet->flow, 0); | |
3523 | if (!rule) { | |
3524 | if (!VLOG_DROP_WARN(&rl)) { | |
3525 | char *s = flow_to_string(&facet->flow); | |
3526 | VLOG_WARN("%s: facet should not exist", s); | |
3527 | free(s); | |
3528 | } | |
3529 | return false; | |
3530 | } else if (rule != facet->rule) { | |
c53e1132 BP |
3531 | may_log = !VLOG_DROP_WARN(&rl); |
3532 | ok = false; | |
3533 | if (may_log) { | |
3534 | struct ds s; | |
6814e51f | 3535 | |
c53e1132 BP |
3536 | ds_init(&s); |
3537 | flow_format(&s, &facet->flow); | |
3538 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3539 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3540 | cls_rule_format(&facet->rule->up.cr, &s); | |
3541 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3542 | rule->up.table_id); | |
3543 | cls_rule_format(&rule->up.cr, &s); | |
3544 | ds_put_char(&s, ')'); | |
6814e51f | 3545 | |
c53e1132 BP |
3546 | VLOG_WARN("%s", ds_cstr(&s)); |
3547 | ds_destroy(&s); | |
3548 | } | |
6814e51f BP |
3549 | } else { |
3550 | ok = true; | |
3551 | } | |
3552 | ||
3553 | /* Check the datapath actions for consistency. */ | |
050ac423 | 3554 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
6814e51f BP |
3555 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
3556 | struct action_xlate_ctx ctx; | |
6814e51f BP |
3557 | bool actions_changed; |
3558 | bool should_install; | |
3559 | ||
3560 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
0e553d9c | 3561 | subfacet->initial_tci, rule, 0, NULL); |
050ac423 BP |
3562 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, |
3563 | &odp_actions); | |
6814e51f BP |
3564 | |
3565 | should_install = (ctx.may_set_up_flow | |
3566 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
3567 | if (!should_install && !subfacet->installed) { | |
3568 | /* The actions for uninstallable flows may vary from one packet to | |
3569 | * the next, so don't compare the actions. */ | |
050ac423 | 3570 | continue; |
6814e51f BP |
3571 | } |
3572 | ||
050ac423 BP |
3573 | actions_changed = (subfacet->actions_len != odp_actions.size |
3574 | || memcmp(subfacet->actions, odp_actions.data, | |
6814e51f BP |
3575 | subfacet->actions_len)); |
3576 | if (should_install != subfacet->installed || actions_changed) { | |
c53e1132 BP |
3577 | if (ok) { |
3578 | may_log = !VLOG_DROP_WARN(&rl); | |
3579 | ok = false; | |
3580 | } | |
6814e51f | 3581 | |
c53e1132 BP |
3582 | if (may_log) { |
3583 | struct odputil_keybuf keybuf; | |
3584 | struct ofpbuf key; | |
3585 | struct ds s; | |
6814e51f | 3586 | |
c53e1132 BP |
3587 | ds_init(&s); |
3588 | subfacet_get_key(subfacet, &keybuf, &key); | |
3589 | odp_flow_key_format(key.data, key.size, &s); | |
3590 | ||
3591 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
3592 | if (should_install != subfacet->installed) { | |
3593 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
3594 | ||
3595 | ds_put_format(&s, " (should%s have been installed)", | |
3596 | should_install ? "" : " not"); | |
3597 | ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)", | |
3598 | ctx.may_set_up_flow ? "true" : "false", | |
3599 | odp_key_fitness_to_string(fitness)); | |
3600 | } | |
3601 | if (actions_changed) { | |
3602 | ds_put_cstr(&s, " (actions were: "); | |
3603 | format_odp_actions(&s, subfacet->actions, | |
3604 | subfacet->actions_len); | |
3605 | ds_put_cstr(&s, ") (correct actions: "); | |
050ac423 | 3606 | format_odp_actions(&s, odp_actions.data, odp_actions.size); |
c53e1132 BP |
3607 | ds_put_char(&s, ')'); |
3608 | } else { | |
3609 | ds_put_cstr(&s, " (actions: "); | |
3610 | format_odp_actions(&s, subfacet->actions, | |
3611 | subfacet->actions_len); | |
3612 | ds_put_char(&s, ')'); | |
3613 | } | |
3614 | VLOG_WARN("%s", ds_cstr(&s)); | |
3615 | ds_destroy(&s); | |
6814e51f | 3616 | } |
6814e51f | 3617 | } |
6814e51f | 3618 | } |
050ac423 | 3619 | ofpbuf_uninit(&odp_actions); |
6814e51f BP |
3620 | |
3621 | return ok; | |
3622 | } | |
3623 | ||
15baa734 | 3624 | /* Re-searches the classifier for 'facet': |
abe529af BP |
3625 | * |
3626 | * - If the rule found is different from 'facet''s current rule, moves | |
3627 | * 'facet' to the new rule and recompiles its actions. | |
3628 | * | |
3629 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3630 | * where it is and recompiles its actions anyway. | |
3631 | * | |
3632 | * - If there is none, destroys 'facet'. | |
3633 | * | |
3634 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3635 | static bool | |
15baa734 | 3636 | facet_revalidate(struct facet *facet) |
abe529af | 3637 | { |
15baa734 | 3638 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b95fc6ba BP |
3639 | struct actions { |
3640 | struct nlattr *odp_actions; | |
3641 | size_t actions_len; | |
3642 | }; | |
3643 | struct actions *new_actions; | |
3644 | ||
abe529af | 3645 | struct action_xlate_ctx ctx; |
050ac423 BP |
3646 | uint64_t odp_actions_stub[1024 / 8]; |
3647 | struct ofpbuf odp_actions; | |
3648 | ||
abe529af | 3649 | struct rule_dpif *new_rule; |
b0f7b9b5 | 3650 | struct subfacet *subfacet; |
abe529af | 3651 | bool actions_changed; |
b95fc6ba | 3652 | int i; |
abe529af BP |
3653 | |
3654 | COVERAGE_INC(facet_revalidate); | |
3655 | ||
3656 | /* Determine the new rule. */ | |
29901626 | 3657 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3658 | if (!new_rule) { |
3659 | /* No new rule, so delete the facet. */ | |
15baa734 | 3660 | facet_remove(facet); |
abe529af BP |
3661 | return false; |
3662 | } | |
3663 | ||
df2c07f4 | 3664 | /* Calculate new datapath actions. |
abe529af BP |
3665 | * |
3666 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3667 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3668 | * around to properly compose it. */ | |
abe529af | 3669 | |
df2c07f4 JP |
3670 | /* If the datapath actions changed or the installability changed, |
3671 | * then we need to talk to the datapath. */ | |
b95fc6ba BP |
3672 | i = 0; |
3673 | new_actions = NULL; | |
3674 | memset(&ctx, 0, sizeof ctx); | |
050ac423 | 3675 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
b0f7b9b5 | 3676 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
b95fc6ba BP |
3677 | bool should_install; |
3678 | ||
e84173dc | 3679 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, |
0e553d9c | 3680 | subfacet->initial_tci, new_rule, 0, NULL); |
050ac423 BP |
3681 | xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions, |
3682 | &odp_actions); | |
3683 | actions_changed = (subfacet->actions_len != odp_actions.size | |
3684 | || memcmp(subfacet->actions, odp_actions.data, | |
b95fc6ba BP |
3685 | subfacet->actions_len)); |
3686 | ||
3687 | should_install = (ctx.may_set_up_flow | |
3688 | && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); | |
b0f7b9b5 BP |
3689 | if (actions_changed || should_install != subfacet->installed) { |
3690 | if (should_install) { | |
3691 | struct dpif_flow_stats stats; | |
3692 | ||
15baa734 | 3693 | subfacet_install(subfacet, |
050ac423 | 3694 | odp_actions.data, odp_actions.size, &stats); |
15baa734 | 3695 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 3696 | } else { |
15baa734 | 3697 | subfacet_uninstall(subfacet); |
b0f7b9b5 | 3698 | } |
b95fc6ba BP |
3699 | |
3700 | if (!new_actions) { | |
3701 | new_actions = xcalloc(list_size(&facet->subfacets), | |
3702 | sizeof *new_actions); | |
3703 | } | |
050ac423 BP |
3704 | new_actions[i].odp_actions = xmemdup(odp_actions.data, |
3705 | odp_actions.size); | |
3706 | new_actions[i].actions_len = odp_actions.size; | |
abe529af | 3707 | } |
b95fc6ba | 3708 | |
b95fc6ba | 3709 | i++; |
b0f7b9b5 | 3710 | } |
050ac423 BP |
3711 | ofpbuf_uninit(&odp_actions); |
3712 | ||
b95fc6ba | 3713 | if (new_actions) { |
15baa734 | 3714 | facet_flush_stats(facet); |
abe529af BP |
3715 | } |
3716 | ||
3717 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3718 | facet->tags = ctx.tags; | |
3719 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3720 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3721 | facet->has_learn = ctx.has_learn; |
3722 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 3723 | facet->has_fin_timeout = ctx.has_fin_timeout; |
9d24de3b | 3724 | facet->mirrors = ctx.mirrors; |
b95fc6ba BP |
3725 | if (new_actions) { |
3726 | i = 0; | |
3727 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3728 | if (new_actions[i].odp_actions) { | |
3729 | free(subfacet->actions); | |
3730 | subfacet->actions = new_actions[i].odp_actions; | |
3731 | subfacet->actions_len = new_actions[i].actions_len; | |
3732 | } | |
3733 | i++; | |
3734 | } | |
3735 | free(new_actions); | |
abe529af BP |
3736 | } |
3737 | if (facet->rule != new_rule) { | |
3738 | COVERAGE_INC(facet_changed_rule); | |
3739 | list_remove(&facet->list_node); | |
3740 | list_push_back(&new_rule->facets, &facet->list_node); | |
3741 | facet->rule = new_rule; | |
3742 | facet->used = new_rule->up.created; | |
9d24de3b | 3743 | facet->prev_used = facet->used; |
abe529af BP |
3744 | } |
3745 | ||
abe529af BP |
3746 | return true; |
3747 | } | |
3748 | ||
3749 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3750 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3751 | static void | |
15baa734 | 3752 | facet_update_time(struct facet *facet, long long int used) |
abe529af | 3753 | { |
15baa734 | 3754 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
abe529af BP |
3755 | if (used > facet->used) { |
3756 | facet->used = used; | |
1745cd08 | 3757 | ofproto_rule_update_used(&facet->rule->up, used); |
abe529af BP |
3758 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
3759 | } | |
3760 | } | |
3761 | ||
bbb5d219 EJ |
3762 | static void |
3763 | facet_reset_counters(struct facet *facet) | |
3764 | { | |
3765 | facet->packet_count = 0; | |
3766 | facet->byte_count = 0; | |
9d24de3b JP |
3767 | facet->prev_packet_count = 0; |
3768 | facet->prev_byte_count = 0; | |
bbb5d219 EJ |
3769 | facet->accounted_bytes = 0; |
3770 | } | |
3771 | ||
abe529af BP |
3772 | static void |
3773 | facet_push_stats(struct facet *facet) | |
3774 | { | |
9d24de3b | 3775 | uint64_t new_packets, new_bytes; |
abe529af | 3776 | |
9d24de3b JP |
3777 | assert(facet->packet_count >= facet->prev_packet_count); |
3778 | assert(facet->byte_count >= facet->prev_byte_count); | |
3779 | assert(facet->used >= facet->prev_used); | |
abe529af | 3780 | |
9d24de3b JP |
3781 | new_packets = facet->packet_count - facet->prev_packet_count; |
3782 | new_bytes = facet->byte_count - facet->prev_byte_count; | |
abe529af | 3783 | |
9d24de3b JP |
3784 | if (new_packets || new_bytes || facet->used > facet->prev_used) { |
3785 | facet->prev_packet_count = facet->packet_count; | |
3786 | facet->prev_byte_count = facet->byte_count; | |
3787 | facet->prev_used = facet->used; | |
abe529af BP |
3788 | |
3789 | flow_push_stats(facet->rule, &facet->flow, | |
9d24de3b JP |
3790 | new_packets, new_bytes, facet->used); |
3791 | ||
3792 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
3793 | facet->mirrors, new_packets, new_bytes); | |
abe529af BP |
3794 | } |
3795 | } | |
3796 | ||
3797 | struct ofproto_push { | |
3798 | struct action_xlate_ctx ctx; | |
3799 | uint64_t packets; | |
3800 | uint64_t bytes; | |
3801 | long long int used; | |
3802 | }; | |
3803 | ||
3804 | static void | |
3805 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3806 | { | |
3807 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3808 | ||
3809 | if (rule) { | |
3810 | rule->packet_count += push->packets; | |
3811 | rule->byte_count += push->bytes; | |
1745cd08 | 3812 | ofproto_rule_update_used(&rule->up, push->used); |
abe529af BP |
3813 | } |
3814 | } | |
3815 | ||
3816 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
9d24de3b | 3817 | * 'rule''s actions and mirrors. */ |
abe529af | 3818 | static void |
18b2a258 | 3819 | flow_push_stats(struct rule_dpif *rule, |
59d0f2c8 | 3820 | const struct flow *flow, uint64_t packets, uint64_t bytes, |
abe529af BP |
3821 | long long int used) |
3822 | { | |
3823 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3824 | struct ofproto_push push; | |
3825 | ||
3826 | push.packets = packets; | |
3827 | push.bytes = bytes; | |
3828 | push.used = used; | |
3829 | ||
f3b50afb BP |
3830 | ofproto_rule_update_used(&rule->up, used); |
3831 | ||
18b2a258 | 3832 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule, |
0e553d9c | 3833 | 0, NULL); |
abe529af | 3834 | push.ctx.resubmit_hook = push_resubmit; |
050ac423 BP |
3835 | xlate_actions_for_side_effects(&push.ctx, |
3836 | rule->up.actions, rule->up.n_actions); | |
abe529af BP |
3837 | } |
3838 | \f | |
b0f7b9b5 BP |
3839 | /* Subfacets. */ |
3840 | ||
3841 | static struct subfacet * | |
3842 | subfacet_find__(struct ofproto_dpif *ofproto, | |
3843 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
3844 | const struct flow *flow) | |
3845 | { | |
3846 | struct subfacet *subfacet; | |
3847 | ||
3848 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
3849 | &ofproto->subfacets) { | |
3850 | if (subfacet->key | |
3851 | ? (subfacet->key_len == key_len | |
3852 | && !memcmp(key, subfacet->key, key_len)) | |
3853 | : flow_equal(flow, &subfacet->facet->flow)) { | |
3854 | return subfacet; | |
3855 | } | |
3856 | } | |
3857 | ||
3858 | return NULL; | |
3859 | } | |
3860 | ||
3861 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
3862 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
b95fc6ba BP |
3863 | * there is one, otherwise creates and returns a new subfacet. |
3864 | * | |
3865 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
3866 | * which case the caller must populate the actions with | |
3867 | * subfacet_make_actions(). */ | |
b0f7b9b5 | 3868 | static struct subfacet * |
15baa734 | 3869 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, |
e84173dc | 3870 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) |
b0f7b9b5 | 3871 | { |
15baa734 | 3872 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); |
b0f7b9b5 BP |
3873 | uint32_t key_hash = odp_flow_key_hash(key, key_len); |
3874 | struct subfacet *subfacet; | |
3875 | ||
3876 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); | |
3877 | if (subfacet) { | |
3878 | if (subfacet->facet == facet) { | |
3879 | return subfacet; | |
3880 | } | |
3881 | ||
3882 | /* This shouldn't happen. */ | |
3883 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
15baa734 | 3884 | subfacet_destroy(subfacet); |
b0f7b9b5 BP |
3885 | } |
3886 | ||
26cd7e34 BP |
3887 | subfacet = (list_is_empty(&facet->subfacets) |
3888 | ? &facet->one_subfacet | |
3889 | : xmalloc(sizeof *subfacet)); | |
b0f7b9b5 BP |
3890 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); |
3891 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
3892 | subfacet->facet = facet; | |
b0f7b9b5 BP |
3893 | subfacet->key_fitness = key_fitness; |
3894 | if (key_fitness != ODP_FIT_PERFECT) { | |
3895 | subfacet->key = xmemdup(key, key_len); | |
3896 | subfacet->key_len = key_len; | |
26cd7e34 BP |
3897 | } else { |
3898 | subfacet->key = NULL; | |
3899 | subfacet->key_len = 0; | |
b0f7b9b5 | 3900 | } |
26cd7e34 BP |
3901 | subfacet->used = time_msec(); |
3902 | subfacet->dp_packet_count = 0; | |
3903 | subfacet->dp_byte_count = 0; | |
3904 | subfacet->actions_len = 0; | |
3905 | subfacet->actions = NULL; | |
b0f7b9b5 | 3906 | subfacet->installed = false; |
e84173dc | 3907 | subfacet->initial_tci = initial_tci; |
b0f7b9b5 BP |
3908 | |
3909 | return subfacet; | |
3910 | } | |
3911 | ||
3912 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
3913 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
3914 | static struct subfacet * | |
3915 | subfacet_find(struct ofproto_dpif *ofproto, | |
6a542738 | 3916 | const struct nlattr *key, size_t key_len) |
b0f7b9b5 BP |
3917 | { |
3918 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
6a542738 PS |
3919 | enum odp_key_fitness fitness; |
3920 | struct flow flow; | |
3921 | ||
3922 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
3923 | if (fitness == ODP_FIT_ERROR) { | |
3924 | return NULL; | |
3925 | } | |
b0f7b9b5 | 3926 | |
6a542738 | 3927 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); |
b0f7b9b5 BP |
3928 | } |
3929 | ||
3930 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
3931 | * its facet within 'ofproto', and frees it. */ | |
3932 | static void | |
15baa734 | 3933 | subfacet_destroy__(struct subfacet *subfacet) |
b0f7b9b5 | 3934 | { |
15baa734 BP |
3935 | struct facet *facet = subfacet->facet; |
3936 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3937 | ||
3938 | subfacet_uninstall(subfacet); | |
b0f7b9b5 BP |
3939 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); |
3940 | list_remove(&subfacet->list_node); | |
3941 | free(subfacet->key); | |
b95fc6ba | 3942 | free(subfacet->actions); |
26cd7e34 BP |
3943 | if (subfacet != &facet->one_subfacet) { |
3944 | free(subfacet); | |
3945 | } | |
b0f7b9b5 BP |
3946 | } |
3947 | ||
3948 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
3949 | * last remaining subfacet in its facet destroys the facet too. */ | |
3950 | static void | |
15baa734 | 3951 | subfacet_destroy(struct subfacet *subfacet) |
b0f7b9b5 BP |
3952 | { |
3953 | struct facet *facet = subfacet->facet; | |
3954 | ||
551a2f6c BP |
3955 | if (list_is_singleton(&facet->subfacets)) { |
3956 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
15baa734 | 3957 | facet_remove(facet); |
551a2f6c | 3958 | } else { |
15baa734 | 3959 | subfacet_destroy__(subfacet); |
b0f7b9b5 BP |
3960 | } |
3961 | } | |
3962 | ||
3963 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
3964 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
3965 | * for use as temporary storage. */ | |
3966 | static void | |
3967 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
3968 | struct ofpbuf *key) | |
3969 | { | |
3970 | if (!subfacet->key) { | |
3971 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
3972 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
3973 | } else { | |
3974 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
3975 | } | |
3976 | } | |
3977 | ||
b95fc6ba BP |
3978 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. */ |
3979 | static void | |
15baa734 | 3980 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet) |
b95fc6ba BP |
3981 | { |
3982 | struct facet *facet = subfacet->facet; | |
18b2a258 | 3983 | struct rule_dpif *rule = facet->rule; |
15baa734 | 3984 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); |
050ac423 | 3985 | |
b95fc6ba | 3986 | struct action_xlate_ctx ctx; |
050ac423 BP |
3987 | uint64_t odp_actions_stub[1024 / 8]; |
3988 | struct ofpbuf odp_actions; | |
b95fc6ba | 3989 | |
050ac423 | 3990 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
15baa734 | 3991 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, |
0e553d9c | 3992 | rule, 0, packet); |
050ac423 | 3993 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions); |
b95fc6ba BP |
3994 | facet->tags = ctx.tags; |
3995 | facet->may_install = ctx.may_set_up_flow; | |
3996 | facet->has_learn = ctx.has_learn; | |
3997 | facet->has_normal = ctx.has_normal; | |
0e553d9c | 3998 | facet->has_fin_timeout = ctx.has_fin_timeout; |
b95fc6ba | 3999 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
9d24de3b | 4000 | facet->mirrors = ctx.mirrors; |
b95fc6ba | 4001 | |
050ac423 BP |
4002 | if (subfacet->actions_len != odp_actions.size |
4003 | || memcmp(subfacet->actions, odp_actions.data, odp_actions.size)) { | |
b95fc6ba | 4004 | free(subfacet->actions); |
050ac423 BP |
4005 | subfacet->actions_len = odp_actions.size; |
4006 | subfacet->actions = xmemdup(odp_actions.data, odp_actions.size); | |
b95fc6ba BP |
4007 | } |
4008 | ||
050ac423 | 4009 | ofpbuf_uninit(&odp_actions); |
b95fc6ba BP |
4010 | } |
4011 | ||
b0f7b9b5 BP |
4012 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' |
4013 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4014 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4015 | * since 'subfacet' was last updated. | |
4016 | * | |
4017 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4018 | static int | |
15baa734 | 4019 | subfacet_install(struct subfacet *subfacet, |
b0f7b9b5 BP |
4020 | const struct nlattr *actions, size_t actions_len, |
4021 | struct dpif_flow_stats *stats) | |
4022 | { | |
15baa734 BP |
4023 | struct facet *facet = subfacet->facet; |
4024 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
b0f7b9b5 BP |
4025 | struct odputil_keybuf keybuf; |
4026 | enum dpif_flow_put_flags flags; | |
4027 | struct ofpbuf key; | |
4028 | int ret; | |
4029 | ||
4030 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
4031 | if (stats) { | |
4032 | flags |= DPIF_FP_ZERO_STATS; | |
4033 | } | |
4034 | ||
4035 | subfacet_get_key(subfacet, &keybuf, &key); | |
4036 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
4037 | actions, actions_len, stats); | |
4038 | ||
4039 | if (stats) { | |
4040 | subfacet_reset_dp_stats(subfacet, stats); | |
4041 | } | |
4042 | ||
4043 | return ret; | |
4044 | } | |
4045 | ||
4046 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
4047 | static void | |
15baa734 | 4048 | subfacet_uninstall(struct subfacet *subfacet) |
b0f7b9b5 BP |
4049 | { |
4050 | if (subfacet->installed) { | |
15baa734 BP |
4051 | struct rule_dpif *rule = subfacet->facet->rule; |
4052 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
b0f7b9b5 BP |
4053 | struct odputil_keybuf keybuf; |
4054 | struct dpif_flow_stats stats; | |
4055 | struct ofpbuf key; | |
4056 | int error; | |
4057 | ||
4058 | subfacet_get_key(subfacet, &keybuf, &key); | |
15baa734 | 4059 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); |
b0f7b9b5 BP |
4060 | subfacet_reset_dp_stats(subfacet, &stats); |
4061 | if (!error) { | |
15baa734 | 4062 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 BP |
4063 | } |
4064 | subfacet->installed = false; | |
4065 | } else { | |
4066 | assert(subfacet->dp_packet_count == 0); | |
4067 | assert(subfacet->dp_byte_count == 0); | |
4068 | } | |
4069 | } | |
4070 | ||
4071 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4072 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4073 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4074 | * was reset in the datapath. 'stats' will be modified to include only | |
4075 | * statistics new since 'subfacet' was last updated. */ | |
4076 | static void | |
4077 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4078 | struct dpif_flow_stats *stats) | |
4079 | { | |
4080 | if (stats | |
4081 | && subfacet->dp_packet_count <= stats->n_packets | |
4082 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4083 | stats->n_packets -= subfacet->dp_packet_count; | |
4084 | stats->n_bytes -= subfacet->dp_byte_count; | |
4085 | } | |
4086 | ||
4087 | subfacet->dp_packet_count = 0; | |
4088 | subfacet->dp_byte_count = 0; | |
4089 | } | |
4090 | ||
4091 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
4092 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
4093 | static void | |
15baa734 | 4094 | subfacet_update_time(struct subfacet *subfacet, long long int used) |
b0f7b9b5 BP |
4095 | { |
4096 | if (used > subfacet->used) { | |
4097 | subfacet->used = used; | |
15baa734 | 4098 | facet_update_time(subfacet->facet, used); |
b0f7b9b5 BP |
4099 | } |
4100 | } | |
4101 | ||
4102 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
4103 | * | |
4104 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4105 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4106 | * represents a packet that was sent by hand or if it represents statistics | |
4107 | * that have been cleared out of the datapath. */ | |
4108 | static void | |
15baa734 | 4109 | subfacet_update_stats(struct subfacet *subfacet, |
b0f7b9b5 BP |
4110 | const struct dpif_flow_stats *stats) |
4111 | { | |
4112 | if (stats->n_packets || stats->used > subfacet->used) { | |
4113 | struct facet *facet = subfacet->facet; | |
4114 | ||
15baa734 | 4115 | subfacet_update_time(subfacet, stats->used); |
b0f7b9b5 BP |
4116 | facet->packet_count += stats->n_packets; |
4117 | facet->byte_count += stats->n_bytes; | |
0e553d9c | 4118 | facet->tcp_flags |= stats->tcp_flags; |
b0f7b9b5 BP |
4119 | facet_push_stats(facet); |
4120 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
4121 | } | |
4122 | } | |
4123 | \f | |
abe529af BP |
4124 | /* Rules. */ |
4125 | ||
4126 | static struct rule_dpif * | |
29901626 BP |
4127 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
4128 | uint8_t table_id) | |
abe529af | 4129 | { |
7257b535 BP |
4130 | struct cls_rule *cls_rule; |
4131 | struct classifier *cls; | |
4132 | ||
9cdaaebe BP |
4133 | if (table_id >= N_TABLES) { |
4134 | return NULL; | |
4135 | } | |
4136 | ||
d0918789 | 4137 | cls = &ofproto->up.tables[table_id].cls; |
eadef313 | 4138 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
4139 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
4140 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
4141 | * are unavailable. */ | |
4142 | struct flow ofpc_normal_flow = *flow; | |
4143 | ofpc_normal_flow.tp_src = htons(0); | |
4144 | ofpc_normal_flow.tp_dst = htons(0); | |
4145 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
4146 | } else { | |
4147 | cls_rule = classifier_lookup(cls, flow); | |
4148 | } | |
4149 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
4150 | } |
4151 | ||
7ee20df1 BP |
4152 | static void |
4153 | complete_operation(struct rule_dpif *rule) | |
4154 | { | |
4155 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4156 | ||
54a9cbc9 | 4157 | rule_invalidate(rule); |
7ee20df1 BP |
4158 | if (clogged) { |
4159 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4160 | c->op = rule->up.pending; | |
4161 | list_push_back(&ofproto->completions, &c->list_node); | |
4162 | } else { | |
4163 | ofoperation_complete(rule->up.pending, 0); | |
4164 | } | |
4165 | } | |
4166 | ||
abe529af BP |
4167 | static struct rule * |
4168 | rule_alloc(void) | |
4169 | { | |
4170 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4171 | return &rule->up; | |
4172 | } | |
4173 | ||
4174 | static void | |
4175 | rule_dealloc(struct rule *rule_) | |
4176 | { | |
4177 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4178 | free(rule); | |
4179 | } | |
4180 | ||
90bf1e07 | 4181 | static enum ofperr |
abe529af BP |
4182 | rule_construct(struct rule *rule_) |
4183 | { | |
4184 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4185 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 4186 | struct rule_dpif *victim; |
54a9cbc9 | 4187 | uint8_t table_id; |
90bf1e07 | 4188 | enum ofperr error; |
5bf0e941 BP |
4189 | |
4190 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
4191 | &rule->up.cr.flow, ofproto->max_ports); | |
4192 | if (error) { | |
4193 | return error; | |
4194 | } | |
abe529af | 4195 | |
abe529af BP |
4196 | rule->packet_count = 0; |
4197 | rule->byte_count = 0; | |
abe529af | 4198 | |
7ee20df1 BP |
4199 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
4200 | if (victim && !list_is_empty(&victim->facets)) { | |
4201 | struct facet *facet; | |
4202 | ||
4203 | rule->facets = victim->facets; | |
4204 | list_moved(&rule->facets); | |
4205 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
4206 | /* XXX: We're only clearing our local counters here. It's possible |
4207 | * that quite a few packets are unaccounted for in the datapath | |
4208 | * statistics. These will be accounted to the new rule instead of | |
4209 | * cleared as required. This could be fixed by clearing out the | |
4210 | * datapath statistics for this facet, but currently it doesn't | |
4211 | * seem worth it. */ | |
4212 | facet_reset_counters(facet); | |
7ee20df1 BP |
4213 | facet->rule = rule; |
4214 | } | |
4215 | } else { | |
4216 | /* Must avoid list_moved() in this case. */ | |
4217 | list_init(&rule->facets); | |
4218 | } | |
abe529af | 4219 | |
54a9cbc9 BP |
4220 | table_id = rule->up.table_id; |
4221 | rule->tag = (victim ? victim->tag | |
4222 | : table_id == 0 ? 0 | |
4223 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4224 | ofproto->tables[table_id].basis)); | |
4225 | ||
7ee20df1 | 4226 | complete_operation(rule); |
abe529af BP |
4227 | return 0; |
4228 | } | |
4229 | ||
4230 | static void | |
4231 | rule_destruct(struct rule *rule_) | |
4232 | { | |
4233 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
abe529af BP |
4234 | struct facet *facet, *next_facet; |
4235 | ||
abe529af | 4236 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
15baa734 | 4237 | facet_revalidate(facet); |
abe529af | 4238 | } |
7ee20df1 BP |
4239 | |
4240 | complete_operation(rule); | |
abe529af BP |
4241 | } |
4242 | ||
4243 | static void | |
4244 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4245 | { | |
4246 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4247 | struct facet *facet; | |
4248 | ||
4249 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4250 | * in facets. This counts, for example, facets that have expired. */ | |
4251 | *packets = rule->packet_count; | |
4252 | *bytes = rule->byte_count; | |
4253 | ||
4254 | /* Add any statistics that are tracked by facets. This includes | |
4255 | * statistical data recently updated by ofproto_update_stats() as well as | |
4256 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4257 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4258 | *packets += facet->packet_count; | |
4259 | *bytes += facet->byte_count; | |
4260 | } | |
4261 | } | |
4262 | ||
90bf1e07 | 4263 | static enum ofperr |
59d0f2c8 BP |
4264 | rule_execute(struct rule *rule_, const struct flow *flow, |
4265 | struct ofpbuf *packet) | |
abe529af BP |
4266 | { |
4267 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4268 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
050ac423 BP |
4269 | |
4270 | size_t size = packet->size; | |
4271 | ||
abe529af | 4272 | struct action_xlate_ctx ctx; |
050ac423 BP |
4273 | uint64_t odp_actions_stub[1024 / 8]; |
4274 | struct ofpbuf odp_actions; | |
abe529af | 4275 | |
050ac423 | 4276 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
54834960 | 4277 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, |
0e553d9c | 4278 | rule, packet_get_tcp_flags(packet, flow), packet); |
050ac423 BP |
4279 | xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions); |
4280 | if (execute_odp_actions(ofproto, flow, odp_actions.data, | |
4281 | odp_actions.size, packet)) { | |
abe529af BP |
4282 | rule->packet_count++; |
4283 | rule->byte_count += size; | |
f3b50afb | 4284 | flow_push_stats(rule, flow, 1, size, time_msec()); |
abe529af | 4285 | } |
050ac423 | 4286 | ofpbuf_uninit(&odp_actions); |
5bf0e941 BP |
4287 | |
4288 | return 0; | |
abe529af BP |
4289 | } |
4290 | ||
7ee20df1 BP |
4291 | static void |
4292 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
4293 | { |
4294 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4295 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
90bf1e07 | 4296 | enum ofperr error; |
abe529af | 4297 | |
7ee20df1 BP |
4298 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
4299 | &rule->up.cr.flow, ofproto->max_ports); | |
4300 | if (error) { | |
4301 | ofoperation_complete(rule->up.pending, error); | |
4302 | return; | |
abe529af | 4303 | } |
7ee20df1 BP |
4304 | |
4305 | complete_operation(rule); | |
abe529af BP |
4306 | } |
4307 | \f | |
97d6520b | 4308 | /* Sends 'packet' out 'ofport'. |
52a90c29 | 4309 | * May modify 'packet'. |
abe529af BP |
4310 | * Returns 0 if successful, otherwise a positive errno value. */ |
4311 | static int | |
52a90c29 | 4312 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) |
abe529af | 4313 | { |
97d6520b | 4314 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); |
80e5eed9 BP |
4315 | struct ofpbuf key, odp_actions; |
4316 | struct odputil_keybuf keybuf; | |
52a90c29 | 4317 | uint16_t odp_port; |
80e5eed9 | 4318 | struct flow flow; |
abe529af BP |
4319 | int error; |
4320 | ||
abff858b | 4321 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
52a90c29 BP |
4322 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, |
4323 | flow.vlan_tci); | |
4324 | if (odp_port != ofport->odp_port) { | |
4325 | eth_pop_vlan(packet); | |
4326 | flow.vlan_tci = htons(0); | |
4327 | } | |
4328 | ||
80e5eed9 BP |
4329 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
4330 | odp_flow_key_from_flow(&key, &flow); | |
4331 | ||
abe529af | 4332 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
4333 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
4334 | ||
df2c07f4 | 4335 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
4336 | error = dpif_execute(ofproto->dpif, |
4337 | key.data, key.size, | |
4338 | odp_actions.data, odp_actions.size, | |
abe529af BP |
4339 | packet); |
4340 | ofpbuf_uninit(&odp_actions); | |
4341 | ||
4342 | if (error) { | |
4343 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4344 | ofproto->up.name, odp_port, strerror(error)); | |
4345 | } | |
6527c598 | 4346 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); |
abe529af BP |
4347 | return error; |
4348 | } | |
4349 | \f | |
df2c07f4 | 4350 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
4351 | |
4352 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4353 | struct action_xlate_ctx *ctx); | |
4cd78906 | 4354 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 4355 | |
98403001 BP |
4356 | static size_t |
4357 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4358 | struct ofpbuf *odp_actions, | |
4359 | const struct flow *flow, | |
4360 | const struct user_action_cookie *cookie) | |
4361 | { | |
98403001 BP |
4362 | uint32_t pid; |
4363 | ||
4364 | pid = dpif_port_get_pid(ofproto->dpif, | |
4365 | ofp_port_to_odp_port(flow->in_port)); | |
4366 | ||
39db78a0 | 4367 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
4368 | } |
4369 | ||
6ff686f2 PS |
4370 | /* Compose SAMPLE action for sFlow. */ |
4371 | static size_t | |
4372 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4373 | struct ofpbuf *odp_actions, | |
4374 | const struct flow *flow, | |
4375 | uint32_t odp_port) | |
4376 | { | |
4377 | uint32_t port_ifindex; | |
4378 | uint32_t probability; | |
98403001 | 4379 | struct user_action_cookie cookie; |
6ff686f2 | 4380 | size_t sample_offset, actions_offset; |
98403001 | 4381 | int cookie_offset, n_output; |
6ff686f2 PS |
4382 | |
4383 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4384 | return 0; | |
4385 | } | |
4386 | ||
4387 | if (odp_port == OVSP_NONE) { | |
4388 | port_ifindex = 0; | |
4389 | n_output = 0; | |
4390 | } else { | |
4391 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4392 | n_output = 1; | |
4393 | } | |
4394 | ||
4395 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
4396 | ||
4397 | /* Number of packets out of UINT_MAX to sample. */ | |
4398 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4399 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4400 | ||
4401 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
4402 | ||
98403001 BP |
4403 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
4404 | cookie.data = port_ifindex; | |
4405 | cookie.n_output = n_output; | |
4406 | cookie.vlan_tci = 0; | |
4407 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
4408 | |
4409 | nl_msg_end_nested(odp_actions, actions_offset); | |
4410 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 4411 | return cookie_offset; |
6ff686f2 PS |
4412 | } |
4413 | ||
4414 | /* SAMPLE action must be first action in any given list of actions. | |
4415 | * At this point we do not have all information required to build it. So try to | |
4416 | * build sample action as complete as possible. */ | |
4417 | static void | |
4418 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4419 | { | |
4420 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4421 | ctx->odp_actions, | |
4422 | &ctx->flow, OVSP_NONE); | |
4423 | ctx->sflow_odp_port = 0; | |
4424 | ctx->sflow_n_outputs = 0; | |
4425 | } | |
4426 | ||
4427 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4428 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4429 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4430 | static void | |
4431 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4432 | { | |
4433 | const struct flow *base = &ctx->base_flow; | |
4434 | struct user_action_cookie *cookie; | |
4435 | ||
4436 | if (!ctx->user_cookie_offset) { | |
4437 | return; | |
4438 | } | |
4439 | ||
4440 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
4441 | sizeof(*cookie)); | |
4442 | assert(cookie != NULL); | |
4443 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
4444 | ||
4445 | if (ctx->sflow_n_outputs) { | |
4446 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
4447 | ctx->sflow_odp_port); | |
4448 | } | |
4449 | if (ctx->sflow_n_outputs >= 255) { | |
4450 | cookie->n_output = 255; | |
4451 | } else { | |
4452 | cookie->n_output = ctx->sflow_n_outputs; | |
4453 | } | |
4454 | cookie->vlan_tci = base->vlan_tci; | |
4455 | } | |
4456 | ||
6ff686f2 | 4457 | static void |
81b1afb1 EJ |
4458 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, |
4459 | bool check_stp) | |
6ff686f2 | 4460 | { |
d59906fb | 4461 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); |
5e48dc2b | 4462 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); |
52a90c29 | 4463 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; |
8b36f51e | 4464 | uint8_t flow_nw_tos = ctx->flow.nw_tos; |
52a90c29 | 4465 | uint16_t out_port; |
d59906fb | 4466 | |
81b1afb1 | 4467 | if (ofport) { |
8b36f51e EJ |
4468 | struct priority_to_dscp *pdscp; |
4469 | ||
9e1fd49b | 4470 | if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD |
81b1afb1 EJ |
4471 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { |
4472 | return; | |
4473 | } | |
8b36f51e | 4474 | |
deedf7e7 | 4475 | pdscp = get_priority(ofport, ctx->flow.skb_priority); |
8b36f51e EJ |
4476 | if (pdscp) { |
4477 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4478 | ctx->flow.nw_tos |= pdscp->dscp; | |
4479 | } | |
81b1afb1 EJ |
4480 | } else { |
4481 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4482 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4483 | * later and we're pre-populating the flow table. */ | |
d59906fb EJ |
4484 | } |
4485 | ||
52a90c29 BP |
4486 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, |
4487 | ctx->flow.vlan_tci); | |
4488 | if (out_port != odp_port) { | |
4489 | ctx->flow.vlan_tci = htons(0); | |
4490 | } | |
5bbda0aa | 4491 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); |
52a90c29 BP |
4492 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); |
4493 | ||
6ff686f2 PS |
4494 | ctx->sflow_odp_port = odp_port; |
4495 | ctx->sflow_n_outputs++; | |
81b1afb1 | 4496 | ctx->nf_output_iface = ofp_port; |
52a90c29 | 4497 | ctx->flow.vlan_tci = flow_vlan_tci; |
8b36f51e | 4498 | ctx->flow.nw_tos = flow_nw_tos; |
6ff686f2 PS |
4499 | } |
4500 | ||
abe529af | 4501 | static void |
5e48dc2b | 4502 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) |
abe529af | 4503 | { |
81b1afb1 | 4504 | compose_output_action__(ctx, ofp_port, true); |
abe529af BP |
4505 | } |
4506 | ||
4507 | static void | |
29901626 BP |
4508 | xlate_table_action(struct action_xlate_ctx *ctx, |
4509 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
4510 | { |
4511 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 4512 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
4513 | struct rule_dpif *rule; |
4514 | uint16_t old_in_port; | |
29901626 BP |
4515 | uint8_t old_table_id; |
4516 | ||
4517 | old_table_id = ctx->table_id; | |
4518 | ctx->table_id = table_id; | |
abe529af | 4519 | |
54a9cbc9 | 4520 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
4521 | old_in_port = ctx->flow.in_port; |
4522 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
4523 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
4524 | ||
4525 | /* Tag the flow. */ | |
4526 | if (table_id > 0 && table_id < N_TABLES) { | |
4527 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4528 | if (table->other_table) { | |
33780682 | 4529 | ctx->tags |= (rule && rule->tag |
54a9cbc9 BP |
4530 | ? rule->tag |
4531 | : rule_calculate_tag(&ctx->flow, | |
4532 | &table->other_table->wc, | |
4533 | table->basis)); | |
4534 | } | |
4535 | } | |
4536 | ||
4537 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
4538 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
4539 | ctx->flow.in_port = old_in_port; |
4540 | ||
4541 | if (ctx->resubmit_hook) { | |
4542 | ctx->resubmit_hook(ctx, rule); | |
4543 | } | |
4544 | ||
4545 | if (rule) { | |
18b2a258 | 4546 | struct rule_dpif *old_rule = ctx->rule; |
54834960 | 4547 | |
abe529af | 4548 | ctx->recurse++; |
18b2a258 | 4549 | ctx->rule = rule; |
abe529af | 4550 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); |
18b2a258 | 4551 | ctx->rule = old_rule; |
abe529af BP |
4552 | ctx->recurse--; |
4553 | } | |
29901626 BP |
4554 | |
4555 | ctx->table_id = old_table_id; | |
abe529af BP |
4556 | } else { |
4557 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4558 | ||
29901626 | 4559 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af | 4560 | MAX_RESUBMIT_RECURSION); |
6a6455e5 | 4561 | ctx->max_resubmit_trigger = true; |
abe529af BP |
4562 | } |
4563 | } | |
4564 | ||
29901626 BP |
4565 | static void |
4566 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
4567 | const struct nx_action_resubmit *nar) | |
4568 | { | |
4569 | uint16_t in_port; | |
4570 | uint8_t table_id; | |
4571 | ||
4572 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
4573 | ? ctx->flow.in_port | |
4574 | : ntohs(nar->in_port)); | |
4575 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
4576 | ||
4577 | xlate_table_action(ctx, in_port, table_id); | |
4578 | } | |
4579 | ||
abe529af | 4580 | static void |
d59906fb | 4581 | flood_packets(struct action_xlate_ctx *ctx, bool all) |
abe529af BP |
4582 | { |
4583 | struct ofport_dpif *ofport; | |
4584 | ||
b3e9b2ed | 4585 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { |
abe529af | 4586 | uint16_t ofp_port = ofport->up.ofp_port; |
d59906fb EJ |
4587 | |
4588 | if (ofp_port == ctx->flow.in_port) { | |
4589 | continue; | |
4590 | } | |
4591 | ||
5e48dc2b | 4592 | if (all) { |
81b1afb1 | 4593 | compose_output_action__(ctx, ofp_port, false); |
9e1fd49b | 4594 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { |
5e48dc2b | 4595 | compose_output_action(ctx, ofp_port); |
abe529af BP |
4596 | } |
4597 | } | |
b3e9b2ed EJ |
4598 | |
4599 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
4600 | } |
4601 | ||
6ff686f2 | 4602 | static void |
f0fd1a17 | 4603 | execute_controller_action(struct action_xlate_ctx *ctx, int len, |
a7349929 BP |
4604 | enum ofp_packet_in_reason reason, |
4605 | uint16_t controller_id) | |
6ff686f2 | 4606 | { |
999fba59 EJ |
4607 | struct ofputil_packet_in pin; |
4608 | struct ofpbuf *packet; | |
6ff686f2 | 4609 | |
999fba59 EJ |
4610 | ctx->may_set_up_flow = false; |
4611 | if (!ctx->packet) { | |
4612 | return; | |
4613 | } | |
4614 | ||
4615 | packet = ofpbuf_clone(ctx->packet); | |
4616 | ||
4617 | if (packet->l2 && packet->l3) { | |
4618 | struct eth_header *eh; | |
4619 | ||
4620 | eth_pop_vlan(packet); | |
4621 | eh = packet->l2; | |
0104aba8 EJ |
4622 | |
4623 | /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 | |
4624 | * LLC frame. Calculating the Ethernet type of these frames is more | |
4625 | * trouble than seems appropriate for a simple assertion. */ | |
4626 | assert(ntohs(eh->eth_type) < ETH_TYPE_MIN | |
4627 | || eh->eth_type == ctx->flow.dl_type); | |
4628 | ||
999fba59 EJ |
4629 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); |
4630 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
4631 | ||
4632 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
4633 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
4634 | } | |
4635 | ||
4636 | if (packet->l4) { | |
4637 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
4638 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
4639 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
4640 | } | |
4641 | ||
4642 | if (packet->l7) { | |
4643 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
4644 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
4645 | ctx->flow.tp_dst); | |
4646 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
4647 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
4648 | ctx->flow.tp_dst); | |
4649 | } | |
4650 | } | |
4651 | } | |
4652 | } | |
4653 | ||
4654 | pin.packet = packet->data; | |
4655 | pin.packet_len = packet->size; | |
f0fd1a17 | 4656 | pin.reason = reason; |
a7349929 | 4657 | pin.controller_id = controller_id; |
54834960 | 4658 | pin.table_id = ctx->table_id; |
18b2a258 | 4659 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; |
54834960 | 4660 | |
999fba59 | 4661 | pin.send_len = len; |
999fba59 EJ |
4662 | flow_get_metadata(&ctx->flow, &pin.fmd); |
4663 | ||
d8653c38 | 4664 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); |
999fba59 | 4665 | ofpbuf_delete(packet); |
6ff686f2 PS |
4666 | } |
4667 | ||
f0fd1a17 PS |
4668 | static bool |
4669 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
4670 | { | |
4671 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
4672 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
4673 | return false; | |
4674 | } | |
4675 | ||
4676 | if (ctx->flow.nw_ttl > 1) { | |
4677 | ctx->flow.nw_ttl--; | |
4678 | return false; | |
4679 | } else { | |
a7349929 | 4680 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); |
f0fd1a17 PS |
4681 | |
4682 | /* Stop processing for current table. */ | |
4683 | return true; | |
4684 | } | |
4685 | } | |
4686 | ||
abe529af BP |
4687 | static void |
4688 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
4689 | uint16_t port, uint16_t max_len) | |
4690 | { | |
4691 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
4692 | ||
4693 | ctx->nf_output_iface = NF_OUT_DROP; | |
4694 | ||
4695 | switch (port) { | |
4696 | case OFPP_IN_PORT: | |
81b1afb1 | 4697 | compose_output_action(ctx, ctx->flow.in_port); |
abe529af BP |
4698 | break; |
4699 | case OFPP_TABLE: | |
29901626 | 4700 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
4701 | break; |
4702 | case OFPP_NORMAL: | |
4703 | xlate_normal(ctx); | |
4704 | break; | |
4705 | case OFPP_FLOOD: | |
d59906fb | 4706 | flood_packets(ctx, false); |
abe529af BP |
4707 | break; |
4708 | case OFPP_ALL: | |
d59906fb | 4709 | flood_packets(ctx, true); |
abe529af BP |
4710 | break; |
4711 | case OFPP_CONTROLLER: | |
a7349929 | 4712 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); |
abe529af | 4713 | break; |
e81d2933 EJ |
4714 | case OFPP_NONE: |
4715 | break; | |
a0fbe94a | 4716 | case OFPP_LOCAL: |
abe529af BP |
4717 | default: |
4718 | if (port != ctx->flow.in_port) { | |
81b1afb1 | 4719 | compose_output_action(ctx, port); |
abe529af BP |
4720 | } |
4721 | break; | |
4722 | } | |
4723 | ||
4724 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
4725 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4726 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4727 | ctx->nf_output_iface = prev_nf_output_iface; | |
4728 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
4729 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
4730 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4731 | } | |
4732 | } | |
4733 | ||
f694937d EJ |
4734 | static void |
4735 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
4736 | const struct nx_action_output_reg *naor) | |
4737 | { | |
816fd533 | 4738 | struct mf_subfield src; |
f694937d EJ |
4739 | uint64_t ofp_port; |
4740 | ||
816fd533 BP |
4741 | nxm_decode(&src, naor->src, naor->ofs_nbits); |
4742 | ofp_port = mf_get_subfield(&src, &ctx->flow); | |
f694937d EJ |
4743 | |
4744 | if (ofp_port <= UINT16_MAX) { | |
4745 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
4746 | } | |
4747 | } | |
4748 | ||
abe529af BP |
4749 | static void |
4750 | xlate_output_action(struct action_xlate_ctx *ctx, | |
4751 | const struct ofp_action_output *oao) | |
4752 | { | |
4753 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
4754 | } | |
4755 | ||
abe529af BP |
4756 | static void |
4757 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
4758 | const struct ofp_action_enqueue *oae) | |
4759 | { | |
e479e41e | 4760 | uint16_t ofp_port; |
abff858b | 4761 | uint32_t flow_priority, priority; |
abe529af BP |
4762 | int error; |
4763 | ||
4764 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
4765 | &priority); | |
4766 | if (error) { | |
4767 | /* Fall back to ordinary output action. */ | |
4768 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
4769 | return; | |
4770 | } | |
4771 | ||
df2c07f4 | 4772 | /* Figure out datapath output port. */ |
abe529af BP |
4773 | ofp_port = ntohs(oae->port); |
4774 | if (ofp_port == OFPP_IN_PORT) { | |
4775 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
4776 | } else if (ofp_port == ctx->flow.in_port) { |
4777 | return; | |
abe529af | 4778 | } |
abe529af | 4779 | |
df2c07f4 | 4780 | /* Add datapath actions. */ |
deedf7e7 BP |
4781 | flow_priority = ctx->flow.skb_priority; |
4782 | ctx->flow.skb_priority = priority; | |
81b1afb1 | 4783 | compose_output_action(ctx, ofp_port); |
deedf7e7 | 4784 | ctx->flow.skb_priority = flow_priority; |
abe529af BP |
4785 | |
4786 | /* Update NetFlow output port. */ | |
4787 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4b23aebf | 4788 | ctx->nf_output_iface = ofp_port; |
abe529af BP |
4789 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { |
4790 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4791 | } | |
4792 | } | |
4793 | ||
4794 | static void | |
4795 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
4796 | const struct nx_action_set_queue *nasq) | |
4797 | { | |
4798 | uint32_t priority; | |
4799 | int error; | |
4800 | ||
4801 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4802 | &priority); | |
4803 | if (error) { | |
4804 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4805 | * has already been logged. */ | |
4806 | return; | |
4807 | } | |
4808 | ||
deedf7e7 | 4809 | ctx->flow.skb_priority = priority; |
abe529af BP |
4810 | } |
4811 | ||
4812 | struct xlate_reg_state { | |
4813 | ovs_be16 vlan_tci; | |
4814 | ovs_be64 tun_id; | |
4815 | }; | |
4816 | ||
abe529af BP |
4817 | static void |
4818 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4819 | const struct nx_action_autopath *naa) | |
4820 | { | |
4821 | uint16_t ofp_port = ntohl(naa->id); | |
4822 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4823 | ||
4824 | if (!port || !port->bundle) { | |
4825 | ofp_port = OFPP_NONE; | |
4826 | } else if (port->bundle->bond) { | |
4827 | /* Autopath does not support VLAN hashing. */ | |
4828 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4829 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4830 | if (slave) { |
4831 | ofp_port = slave->up.ofp_port; | |
4832 | } | |
4833 | } | |
4834 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4835 | } | |
4836 | ||
daff3353 EJ |
4837 | static bool |
4838 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4839 | { | |
4840 | struct ofproto_dpif *ofproto = ofproto_; | |
4841 | struct ofport_dpif *port; | |
4842 | ||
4843 | switch (ofp_port) { | |
4844 | case OFPP_IN_PORT: | |
4845 | case OFPP_TABLE: | |
4846 | case OFPP_NORMAL: | |
4847 | case OFPP_FLOOD: | |
4848 | case OFPP_ALL: | |
439e4d8c | 4849 | case OFPP_NONE: |
daff3353 EJ |
4850 | return true; |
4851 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4852 | return false; | |
4853 | default: | |
4854 | port = get_ofp_port(ofproto, ofp_port); | |
4855 | return port ? port->may_enable : false; | |
4856 | } | |
4857 | } | |
4858 | ||
75a75043 BP |
4859 | static void |
4860 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
4861 | const struct nx_action_learn *learn) | |
4862 | { | |
4863 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
4864 | struct ofputil_flow_mod fm; | |
4865 | int error; | |
4866 | ||
4867 | learn_execute(learn, &ctx->flow, &fm); | |
4868 | ||
4869 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
4870 | if (error && !VLOG_DROP_WARN(&rl)) { | |
90bf1e07 BP |
4871 | VLOG_WARN("learning action failed to modify flow table (%s)", |
4872 | ofperr_get_name(error)); | |
75a75043 BP |
4873 | } |
4874 | ||
4875 | free(fm.actions); | |
4876 | } | |
4877 | ||
0e553d9c BP |
4878 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case |
4879 | * means "infinite". */ | |
4880 | static void | |
4881 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
4882 | { | |
4883 | if (max && (!*timeout || *timeout > max)) { | |
4884 | *timeout = max; | |
4885 | } | |
4886 | } | |
4887 | ||
4888 | static void | |
4889 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
4890 | const struct nx_action_fin_timeout *naft) | |
4891 | { | |
4892 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
4893 | struct rule_dpif *rule = ctx->rule; | |
4894 | ||
4895 | reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout); | |
4896 | reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout); | |
4897 | } | |
4898 | } | |
4899 | ||
21f7563c JP |
4900 | static bool |
4901 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
4902 | { | |
9e1fd49b BP |
4903 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
4904 | ? OFPUTIL_PC_NO_RECV_STP | |
4905 | : OFPUTIL_PC_NO_RECV)) { | |
21f7563c JP |
4906 | return false; |
4907 | } | |
4908 | ||
4909 | /* Only drop packets here if both forwarding and learning are | |
4910 | * disabled. If just learning is enabled, we need to have | |
4911 | * OFPP_NORMAL and the learning action have a look at the packet | |
4912 | * before we can drop it. */ | |
4913 | if (!stp_forward_in_state(port->stp_state) | |
4914 | && !stp_learn_in_state(port->stp_state)) { | |
4915 | return false; | |
4916 | } | |
4917 | ||
4918 | return true; | |
4919 | } | |
4920 | ||
abe529af BP |
4921 | static void |
4922 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4923 | struct action_xlate_ctx *ctx) | |
4924 | { | |
4925 | const struct ofport_dpif *port; | |
abe529af | 4926 | const union ofp_action *ia; |
254750ce | 4927 | bool was_evictable = true; |
b4b8c781 | 4928 | size_t left; |
abe529af BP |
4929 | |
4930 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 4931 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
4932 | /* Drop this flow. */ |
4933 | return; | |
4934 | } | |
4935 | ||
254750ce BP |
4936 | if (ctx->rule) { |
4937 | /* Don't let the rule we're working on get evicted underneath us. */ | |
4938 | was_evictable = ctx->rule->up.evictable; | |
4939 | ctx->rule->up.evictable = false; | |
4940 | } | |
b4b8c781 | 4941 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 4942 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
4943 | const struct nx_action_resubmit *nar; |
4944 | const struct nx_action_set_tunnel *nast; | |
4945 | const struct nx_action_set_queue *nasq; | |
4946 | const struct nx_action_multipath *nam; | |
4947 | const struct nx_action_autopath *naa; | |
daff3353 | 4948 | const struct nx_action_bundle *nab; |
f694937d | 4949 | const struct nx_action_output_reg *naor; |
a7349929 | 4950 | const struct nx_action_controller *nac; |
38f2e360 BP |
4951 | enum ofputil_action_code code; |
4952 | ovs_be64 tun_id; | |
4953 | ||
848e8809 EJ |
4954 | if (ctx->exit) { |
4955 | break; | |
4956 | } | |
4957 | ||
38f2e360 BP |
4958 | code = ofputil_decode_action_unsafe(ia); |
4959 | switch (code) { | |
08f94c0e | 4960 | case OFPUTIL_OFPAT10_OUTPUT: |
abe529af BP |
4961 | xlate_output_action(ctx, &ia->output); |
4962 | break; | |
4963 | ||
08f94c0e | 4964 | case OFPUTIL_OFPAT10_SET_VLAN_VID: |
abe529af BP |
4965 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
4966 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
4967 | break; |
4968 | ||
08f94c0e | 4969 | case OFPUTIL_OFPAT10_SET_VLAN_PCP: |
abe529af BP |
4970 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
4971 | ctx->flow.vlan_tci |= htons( | |
4972 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
4973 | break; |
4974 | ||
08f94c0e | 4975 | case OFPUTIL_OFPAT10_STRIP_VLAN: |
abe529af | 4976 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
4977 | break; |
4978 | ||
08f94c0e | 4979 | case OFPUTIL_OFPAT10_SET_DL_SRC: |
abe529af | 4980 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4981 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
4982 | break; | |
4983 | ||
08f94c0e | 4984 | case OFPUTIL_OFPAT10_SET_DL_DST: |
abe529af | 4985 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4986 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
4987 | break; | |
4988 | ||
08f94c0e | 4989 | case OFPUTIL_OFPAT10_SET_NW_SRC: |
abe529af BP |
4990 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
4991 | break; | |
4992 | ||
08f94c0e | 4993 | case OFPUTIL_OFPAT10_SET_NW_DST: |
abe529af BP |
4994 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
4995 | break; | |
4996 | ||
08f94c0e | 4997 | case OFPUTIL_OFPAT10_SET_NW_TOS: |
c4f2731d PS |
4998 | /* OpenFlow 1.0 only supports IPv4. */ |
4999 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5000 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
5001 | ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
5002 | } | |
abe529af BP |
5003 | break; |
5004 | ||
08f94c0e | 5005 | case OFPUTIL_OFPAT10_SET_TP_SRC: |
abe529af BP |
5006 | ctx->flow.tp_src = ia->tp_port.tp_port; |
5007 | break; | |
5008 | ||
08f94c0e | 5009 | case OFPUTIL_OFPAT10_SET_TP_DST: |
abe529af BP |
5010 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
5011 | break; | |
5012 | ||
08f94c0e | 5013 | case OFPUTIL_OFPAT10_ENQUEUE: |
38f2e360 BP |
5014 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); |
5015 | break; | |
5016 | ||
5017 | case OFPUTIL_NXAST_RESUBMIT: | |
5018 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
5019 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
5020 | break; | |
5021 | ||
5022 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
5023 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
5024 | break; |
5025 | ||
38f2e360 BP |
5026 | case OFPUTIL_NXAST_SET_TUNNEL: |
5027 | nast = (const struct nx_action_set_tunnel *) ia; | |
5028 | tun_id = htonll(ntohl(nast->tun_id)); | |
5029 | ctx->flow.tun_id = tun_id; | |
5030 | break; | |
5031 | ||
5032 | case OFPUTIL_NXAST_SET_QUEUE: | |
5033 | nasq = (const struct nx_action_set_queue *) ia; | |
5034 | xlate_set_queue_action(ctx, nasq); | |
5035 | break; | |
5036 | ||
5037 | case OFPUTIL_NXAST_POP_QUEUE: | |
deedf7e7 | 5038 | ctx->flow.skb_priority = ctx->orig_skb_priority; |
38f2e360 BP |
5039 | break; |
5040 | ||
5041 | case OFPUTIL_NXAST_REG_MOVE: | |
5042 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
5043 | &ctx->flow); | |
5044 | break; | |
5045 | ||
5046 | case OFPUTIL_NXAST_REG_LOAD: | |
5047 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
5048 | &ctx->flow); | |
5049 | break; | |
5050 | ||
5051 | case OFPUTIL_NXAST_NOTE: | |
5052 | /* Nothing to do. */ | |
5053 | break; | |
5054 | ||
5055 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
5056 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
5057 | ctx->flow.tun_id = tun_id; | |
5058 | break; | |
5059 | ||
5060 | case OFPUTIL_NXAST_MULTIPATH: | |
5061 | nam = (const struct nx_action_multipath *) ia; | |
5062 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
5063 | break; |
5064 | ||
38f2e360 BP |
5065 | case OFPUTIL_NXAST_AUTOPATH: |
5066 | naa = (const struct nx_action_autopath *) ia; | |
5067 | xlate_autopath(ctx, naa); | |
abe529af | 5068 | break; |
daff3353 EJ |
5069 | |
5070 | case OFPUTIL_NXAST_BUNDLE: | |
5071 | ctx->ofproto->has_bundle_action = true; | |
5072 | nab = (const struct nx_action_bundle *) ia; | |
5073 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
5074 | slave_enabled_cb, | |
5075 | ctx->ofproto), 0); | |
5076 | break; | |
a368bb53 EJ |
5077 | |
5078 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
5079 | ctx->ofproto->has_bundle_action = true; | |
5080 | nab = (const struct nx_action_bundle *) ia; | |
5081 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
5082 | ctx->ofproto); | |
5083 | break; | |
f694937d EJ |
5084 | |
5085 | case OFPUTIL_NXAST_OUTPUT_REG: | |
5086 | naor = (const struct nx_action_output_reg *) ia; | |
5087 | xlate_output_reg_action(ctx, naor); | |
5088 | break; | |
75a75043 BP |
5089 | |
5090 | case OFPUTIL_NXAST_LEARN: | |
5091 | ctx->has_learn = true; | |
3de9590b | 5092 | if (ctx->may_learn) { |
75a75043 BP |
5093 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); |
5094 | } | |
5095 | break; | |
848e8809 | 5096 | |
f0fd1a17 PS |
5097 | case OFPUTIL_NXAST_DEC_TTL: |
5098 | if (compose_dec_ttl(ctx)) { | |
5099 | goto out; | |
5100 | } | |
5101 | break; | |
5102 | ||
848e8809 EJ |
5103 | case OFPUTIL_NXAST_EXIT: |
5104 | ctx->exit = true; | |
5105 | break; | |
0e553d9c BP |
5106 | |
5107 | case OFPUTIL_NXAST_FIN_TIMEOUT: | |
5108 | ctx->has_fin_timeout = true; | |
5109 | xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia); | |
5110 | break; | |
a7349929 BP |
5111 | |
5112 | case OFPUTIL_NXAST_CONTROLLER: | |
5113 | nac = (const struct nx_action_controller *) ia; | |
5114 | execute_controller_action(ctx, ntohs(nac->max_len), nac->reason, | |
5115 | ntohs(nac->controller_id)); | |
5116 | break; | |
abe529af BP |
5117 | } |
5118 | } | |
21f7563c | 5119 | |
f0fd1a17 | 5120 | out: |
21f7563c JP |
5121 | /* We've let OFPP_NORMAL and the learning action look at the packet, |
5122 | * so drop it now if forwarding is disabled. */ | |
5123 | if (port && !stp_forward_in_state(port->stp_state)) { | |
5124 | ofpbuf_clear(ctx->odp_actions); | |
5125 | add_sflow_action(ctx); | |
5126 | } | |
254750ce BP |
5127 | if (ctx->rule) { |
5128 | ctx->rule->up.evictable = was_evictable; | |
5129 | } | |
abe529af BP |
5130 | } |
5131 | ||
5132 | static void | |
5133 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
5134 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
18b2a258 | 5135 | ovs_be16 initial_tci, struct rule_dpif *rule, |
0e553d9c | 5136 | uint8_t tcp_flags, const struct ofpbuf *packet) |
abe529af BP |
5137 | { |
5138 | ctx->ofproto = ofproto; | |
5139 | ctx->flow = *flow; | |
e84173dc BP |
5140 | ctx->base_flow = ctx->flow; |
5141 | ctx->base_flow.tun_id = 0; | |
5142 | ctx->base_flow.vlan_tci = initial_tci; | |
18b2a258 | 5143 | ctx->rule = rule; |
abe529af | 5144 | ctx->packet = packet; |
3de9590b | 5145 | ctx->may_learn = packet != NULL; |
0e553d9c | 5146 | ctx->tcp_flags = tcp_flags; |
abe529af | 5147 | ctx->resubmit_hook = NULL; |
abe529af BP |
5148 | } |
5149 | ||
050ac423 BP |
5150 | /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in |
5151 | * 'odp_actions', using 'ctx'. */ | |
5152 | static void | |
abe529af | 5153 | xlate_actions(struct action_xlate_ctx *ctx, |
050ac423 BP |
5154 | const union ofp_action *in, size_t n_in, |
5155 | struct ofpbuf *odp_actions) | |
abe529af | 5156 | { |
c06bba01 JP |
5157 | struct flow orig_flow = ctx->flow; |
5158 | ||
abe529af BP |
5159 | COVERAGE_INC(ofproto_dpif_xlate); |
5160 | ||
050ac423 BP |
5161 | ofpbuf_clear(odp_actions); |
5162 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
5163 | ||
5164 | ctx->odp_actions = odp_actions; | |
97e42c92 BP |
5165 | ctx->tags = 0; |
5166 | ctx->may_set_up_flow = true; | |
5167 | ctx->has_learn = false; | |
5168 | ctx->has_normal = false; | |
0e553d9c | 5169 | ctx->has_fin_timeout = false; |
97e42c92 | 5170 | ctx->nf_output_iface = NF_OUT_DROP; |
9d24de3b | 5171 | ctx->mirrors = 0; |
97e42c92 | 5172 | ctx->recurse = 0; |
6a6455e5 | 5173 | ctx->max_resubmit_trigger = false; |
deedf7e7 | 5174 | ctx->orig_skb_priority = ctx->flow.skb_priority; |
97e42c92 | 5175 | ctx->table_id = 0; |
848e8809 | 5176 | ctx->exit = false; |
7257b535 | 5177 | |
eadef313 | 5178 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
5179 | switch (ctx->ofproto->up.frag_handling) { |
5180 | case OFPC_FRAG_NORMAL: | |
5181 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
5182 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
5183 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
5184 | break; |
5185 | ||
5186 | case OFPC_FRAG_DROP: | |
050ac423 | 5187 | return; |
7257b535 BP |
5188 | |
5189 | case OFPC_FRAG_REASM: | |
5190 | NOT_REACHED(); | |
5191 | ||
5192 | case OFPC_FRAG_NX_MATCH: | |
5193 | /* Nothing to do. */ | |
5194 | break; | |
f0fd1a17 PS |
5195 | |
5196 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
5197 | NOT_REACHED(); | |
7257b535 BP |
5198 | } |
5199 | } | |
5200 | ||
fc08b7a2 | 5201 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af BP |
5202 | ctx->may_set_up_flow = false; |
5203 | } else { | |
6a6455e5 EJ |
5204 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
5205 | struct flow original_flow = ctx->flow; | |
5206 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; | |
5207 | ||
6ff686f2 | 5208 | add_sflow_action(ctx); |
abe529af | 5209 | do_xlate_actions(in, n_in, ctx); |
abe529af | 5210 | |
6a6455e5 EJ |
5211 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook |
5212 | && !VLOG_DROP_ERR(&trace_rl)) { | |
5213 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5214 | ||
5215 | ofproto_trace(ctx->ofproto, &original_flow, ctx->packet, | |
5216 | initial_tci, &ds); | |
5217 | VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s", | |
5218 | ds_cstr(&ds)); | |
5219 | ds_destroy(&ds); | |
5220 | } | |
5221 | ||
b6848f13 BP |
5222 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
5223 | ctx->odp_actions->data, | |
5224 | ctx->odp_actions->size)) { | |
5225 | ctx->may_set_up_flow = false; | |
5226 | if (ctx->packet | |
5227 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
5228 | ctx->packet)) { | |
5e48dc2b | 5229 | compose_output_action(ctx, OFPP_LOCAL); |
b6848f13 BP |
5230 | } |
5231 | } | |
c06bba01 | 5232 | add_mirror_actions(ctx, &orig_flow); |
a7c4eaf6 | 5233 | fix_sflow_action(ctx); |
abe529af | 5234 | } |
050ac423 BP |
5235 | } |
5236 | ||
5237 | /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions, | |
5238 | * using 'ctx', and discards the datapath actions. */ | |
5239 | static void | |
5240 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
5241 | const union ofp_action *in, size_t n_in) | |
5242 | { | |
5243 | uint64_t odp_actions_stub[1024 / 8]; | |
5244 | struct ofpbuf odp_actions; | |
abe529af | 5245 | |
050ac423 BP |
5246 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); |
5247 | xlate_actions(ctx, in, n_in, &odp_actions); | |
5248 | ofpbuf_uninit(&odp_actions); | |
abe529af BP |
5249 | } |
5250 | \f | |
5251 | /* OFPP_NORMAL implementation. */ | |
5252 | ||
abe529af BP |
5253 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
5254 | ||
ecac4ebf BP |
5255 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
5256 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5257 | * the bundle on which the packet was received, returns the VLAN to which the | |
5258 | * packet belongs. | |
5259 | * | |
5260 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5261 | static uint16_t | |
5262 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5263 | { | |
5264 | switch (in_bundle->vlan_mode) { | |
5265 | case PORT_VLAN_ACCESS: | |
5266 | return in_bundle->vlan; | |
5267 | break; | |
5268 | ||
5269 | case PORT_VLAN_TRUNK: | |
5270 | return vid; | |
5271 | ||
5272 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5273 | case PORT_VLAN_NATIVE_TAGGED: | |
5274 | return vid ? vid : in_bundle->vlan; | |
5275 | ||
5276 | default: | |
5277 | NOT_REACHED(); | |
5278 | } | |
5279 | } | |
5280 | ||
5da5ec37 BP |
5281 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
5282 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5283 | * a warning. | |
5284 | * | |
5285 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5286 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5287 | * 0...4095. */ | |
5288 | static bool | |
5289 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5290 | { | |
33158a18 JP |
5291 | /* Allow any VID on the OFPP_NONE port. */ |
5292 | if (in_bundle == &ofpp_none_bundle) { | |
5293 | return true; | |
5294 | } | |
5295 | ||
5da5ec37 BP |
5296 | switch (in_bundle->vlan_mode) { |
5297 | case PORT_VLAN_ACCESS: | |
5298 | if (vid) { | |
5299 | if (warn) { | |
5300 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5301 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5302 | "packet received on port %s configured as VLAN " | |
5303 | "%"PRIu16" access port", | |
5304 | in_bundle->ofproto->up.name, vid, | |
5305 | in_bundle->name, in_bundle->vlan); | |
5306 | } | |
5307 | return false; | |
5308 | } | |
5309 | return true; | |
5310 | ||
5311 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5312 | case PORT_VLAN_NATIVE_TAGGED: | |
5313 | if (!vid) { | |
5314 | /* Port must always carry its native VLAN. */ | |
5315 | return true; | |
5316 | } | |
5317 | /* Fall through. */ | |
5318 | case PORT_VLAN_TRUNK: | |
5319 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5320 | if (warn) { | |
5321 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5322 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5323 | "received on port %s not configured for trunking " | |
5324 | "VLAN %"PRIu16, | |
5325 | in_bundle->ofproto->up.name, vid, | |
5326 | in_bundle->name, vid); | |
5327 | } | |
5328 | return false; | |
5329 | } | |
5330 | return true; | |
5331 | ||
5332 | default: | |
5333 | NOT_REACHED(); | |
5334 | } | |
5335 | ||
5336 | } | |
5337 | ||
ecac4ebf BP |
5338 | /* Given 'vlan', the VLAN that a packet belongs to, and |
5339 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5340 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5341 | * then the 802.1Q header should only be included in the packet if there is a | |
5342 | * nonzero PCP.) | |
5343 | * | |
5344 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5345 | static uint16_t | |
5346 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5347 | { | |
5348 | switch (out_bundle->vlan_mode) { | |
5349 | case PORT_VLAN_ACCESS: | |
5350 | return 0; | |
5351 | ||
5352 | case PORT_VLAN_TRUNK: | |
5353 | case PORT_VLAN_NATIVE_TAGGED: | |
5354 | return vlan; | |
5355 | ||
5356 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5357 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5358 | ||
5359 | default: | |
5360 | NOT_REACHED(); | |
5361 | } | |
5362 | } | |
5363 | ||
395e68ce BP |
5364 | static void |
5365 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5366 | uint16_t vlan) | |
abe529af | 5367 | { |
395e68ce BP |
5368 | struct ofport_dpif *port; |
5369 | uint16_t vid; | |
81b1afb1 | 5370 | ovs_be16 tci, old_tci; |
ecac4ebf | 5371 | |
395e68ce BP |
5372 | vid = output_vlan_to_vid(out_bundle, vlan); |
5373 | if (!out_bundle->bond) { | |
5374 | port = ofbundle_get_a_port(out_bundle); | |
5375 | } else { | |
5376 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5377 | vid, &ctx->tags); | |
5378 | if (!port) { | |
5379 | /* No slaves enabled, so drop packet. */ | |
5380 | return; | |
5381 | } | |
5382 | } | |
abe529af | 5383 | |
81b1afb1 | 5384 | old_tci = ctx->flow.vlan_tci; |
5e9ceccd BP |
5385 | tci = htons(vid); |
5386 | if (tci || out_bundle->use_priority_tags) { | |
5387 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5388 | if (tci) { | |
5389 | tci |= htons(VLAN_CFI); | |
5390 | } | |
395e68ce | 5391 | } |
81b1afb1 | 5392 | ctx->flow.vlan_tci = tci; |
395e68ce | 5393 | |
5e48dc2b | 5394 | compose_output_action(ctx, port->up.ofp_port); |
81b1afb1 | 5395 | ctx->flow.vlan_tci = old_tci; |
abe529af BP |
5396 | } |
5397 | ||
5398 | static int | |
5399 | mirror_mask_ffs(mirror_mask_t mask) | |
5400 | { | |
5401 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5402 | return ffs(mask); | |
5403 | } | |
5404 | ||
abe529af BP |
5405 | static bool |
5406 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5407 | { | |
ecac4ebf | 5408 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 5409 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
5410 | } |
5411 | ||
5412 | static bool | |
5413 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5414 | { | |
5415 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5416 | } | |
5417 | ||
5418 | /* Returns an arbitrary interface within 'bundle'. */ | |
5419 | static struct ofport_dpif * | |
5420 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5421 | { | |
5422 | return CONTAINER_OF(list_front(&bundle->ports), | |
5423 | struct ofport_dpif, bundle_node); | |
5424 | } | |
5425 | ||
abe529af BP |
5426 | static bool |
5427 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5428 | { | |
fc3d7408 | 5429 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
5430 | } |
5431 | ||
07817dfe BP |
5432 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
5433 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
5434 | * protocols that may confuse switches. */ | |
5435 | static bool | |
5436 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
5437 | { | |
5438 | /* If you change this function's behavior, please update corresponding | |
5439 | * documentation in vswitch.xml at the same time. */ | |
5440 | if (dst[0] != 0x01) { | |
5441 | /* All the currently banned MACs happen to start with 01 currently, so | |
5442 | * this is a quick way to eliminate most of the good ones. */ | |
5443 | } else { | |
5444 | if (eth_addr_is_reserved(dst)) { | |
5445 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
5446 | * (01-80-c2-00-00-0x). */ | |
5447 | return false; | |
5448 | } | |
5449 | ||
5450 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
5451 | /* Cisco OUI. */ | |
5452 | if ((dst[3] & 0xfe) == 0xcc && | |
5453 | (dst[4] & 0xfe) == 0xcc && | |
5454 | (dst[5] & 0xfe) == 0xcc) { | |
5455 | /* Drop the following protocols plus others following the same | |
5456 | pattern: | |
5457 | ||
5458 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
5459 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
5460 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
5461 | return false; | |
5462 | } | |
5463 | ||
5464 | if (!(dst[3] | dst[4] | dst[5])) { | |
5465 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
5466 | return false; | |
5467 | } | |
5468 | } | |
5469 | } | |
5470 | return true; | |
5471 | } | |
5472 | ||
abe529af | 5473 | static void |
c06bba01 | 5474 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) |
abe529af BP |
5475 | { |
5476 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5477 | mirror_mask_t mirrors; | |
c06bba01 JP |
5478 | struct ofbundle *in_bundle; |
5479 | uint16_t vlan; | |
5480 | uint16_t vid; | |
5481 | const struct nlattr *a; | |
5482 | size_t left; | |
5483 | ||
3581c12c JP |
5484 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, |
5485 | ctx->packet != NULL); | |
5486 | if (!in_bundle) { | |
c06bba01 JP |
5487 | return; |
5488 | } | |
c06bba01 JP |
5489 | mirrors = in_bundle->src_mirrors; |
5490 | ||
5491 | /* Drop frames on bundles reserved for mirroring. */ | |
5492 | if (in_bundle->mirror_out) { | |
5493 | if (ctx->packet != NULL) { | |
5494 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5495 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5496 | "%s, which is reserved exclusively for mirroring", | |
5497 | ctx->ofproto->up.name, in_bundle->name); | |
5498 | } | |
5499 | return; | |
5500 | } | |
5501 | ||
5502 | /* Check VLAN. */ | |
5503 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5504 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5505 | return; | |
5506 | } | |
5507 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5508 | ||
5509 | /* Look at the output ports to check for destination selections. */ | |
5510 | ||
5511 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5512 | ctx->odp_actions->size) { | |
5513 | enum ovs_action_attr type = nl_attr_type(a); | |
5514 | struct ofport_dpif *ofport; | |
5515 | ||
5516 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5517 | continue; | |
5518 | } | |
5519 | ||
5520 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
521472bc BP |
5521 | if (ofport && ofport->bundle) { |
5522 | mirrors |= ofport->bundle->dst_mirrors; | |
5523 | } | |
c06bba01 | 5524 | } |
abe529af BP |
5525 | |
5526 | if (!mirrors) { | |
5527 | return; | |
5528 | } | |
5529 | ||
c06bba01 JP |
5530 | /* Restore the original packet before adding the mirror actions. */ |
5531 | ctx->flow = *orig_flow; | |
5532 | ||
9ba15e2a BP |
5533 | while (mirrors) { |
5534 | struct ofmirror *m; | |
9ba15e2a BP |
5535 | |
5536 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5537 | ||
5538 | if (!vlan_is_mirrored(m, vlan)) { | |
5539 | mirrors &= mirrors - 1; | |
5540 | continue; | |
5541 | } | |
5542 | ||
5543 | mirrors &= ~m->dup_mirrors; | |
9d24de3b | 5544 | ctx->mirrors |= m->dup_mirrors; |
9ba15e2a | 5545 | if (m->out) { |
395e68ce | 5546 | output_normal(ctx, m->out, vlan); |
c06bba01 | 5547 | } else if (eth_dst_may_rspan(orig_flow->dl_dst) |
9ba15e2a BP |
5548 | && vlan != m->out_vlan) { |
5549 | struct ofbundle *bundle; | |
5550 | ||
5551 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
5552 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
5553 | && !bundle->mirror_out) { |
5554 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
5555 | } |
5556 | } | |
5557 | } | |
abe529af BP |
5558 | } |
5559 | } | |
5560 | ||
9d24de3b JP |
5561 | static void |
5562 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
5563 | uint64_t packets, uint64_t bytes) | |
5564 | { | |
5565 | if (!mirrors) { | |
5566 | return; | |
5567 | } | |
5568 | ||
5569 | for (; mirrors; mirrors &= mirrors - 1) { | |
5570 | struct ofmirror *m; | |
5571 | ||
5572 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5573 | ||
5574 | if (!m) { | |
5575 | /* In normal circumstances 'm' will not be NULL. However, | |
5576 | * if mirrors are reconfigured, we can temporarily get out | |
5577 | * of sync in facet_revalidate(). We could "correct" the | |
5578 | * mirror list before reaching here, but doing that would | |
5579 | * not properly account the traffic stats we've currently | |
5580 | * accumulated for previous mirror configuration. */ | |
5581 | continue; | |
5582 | } | |
5583 | ||
5584 | m->packet_count += packets; | |
5585 | m->byte_count += bytes; | |
5586 | } | |
5587 | } | |
5588 | ||
abe529af BP |
5589 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
5590 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
5591 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
5592 | static bool | |
5593 | is_gratuitous_arp(const struct flow *flow) | |
5594 | { | |
5595 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
5596 | && eth_addr_is_broadcast(flow->dl_dst) | |
5597 | && (flow->nw_proto == ARP_OP_REPLY | |
5598 | || (flow->nw_proto == ARP_OP_REQUEST | |
5599 | && flow->nw_src == flow->nw_dst))); | |
5600 | } | |
5601 | ||
5602 | static void | |
5603 | update_learning_table(struct ofproto_dpif *ofproto, | |
5604 | const struct flow *flow, int vlan, | |
5605 | struct ofbundle *in_bundle) | |
5606 | { | |
5607 | struct mac_entry *mac; | |
5608 | ||
33158a18 JP |
5609 | /* Don't learn the OFPP_NONE port. */ |
5610 | if (in_bundle == &ofpp_none_bundle) { | |
5611 | return; | |
5612 | } | |
5613 | ||
abe529af BP |
5614 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { |
5615 | return; | |
5616 | } | |
5617 | ||
5618 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
5619 | if (is_gratuitous_arp(flow)) { | |
5620 | /* We don't want to learn from gratuitous ARP packets that are | |
5621 | * reflected back over bond slaves so we lock the learning table. */ | |
5622 | if (!in_bundle->bond) { | |
5623 | mac_entry_set_grat_arp_lock(mac); | |
5624 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
5625 | return; | |
5626 | } | |
5627 | } | |
5628 | ||
5629 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
5630 | /* The log messages here could actually be useful in debugging, | |
5631 | * so keep the rate limit relatively high. */ | |
5632 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5633 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
5634 | "on port %s in VLAN %d", | |
5635 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
5636 | in_bundle->name, vlan); | |
5637 | ||
5638 | mac->port.p = in_bundle; | |
5639 | tag_set_add(&ofproto->revalidate_set, | |
5640 | mac_learning_changed(ofproto->ml, mac)); | |
5641 | } | |
5642 | } | |
5643 | ||
3581c12c | 5644 | static struct ofbundle * |
395e68ce BP |
5645 | lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) |
5646 | { | |
5647 | struct ofport_dpif *ofport; | |
5648 | ||
33158a18 JP |
5649 | /* Special-case OFPP_NONE, which a controller may use as the ingress |
5650 | * port for traffic that it is sourcing. */ | |
5651 | if (in_port == OFPP_NONE) { | |
5652 | return &ofpp_none_bundle; | |
5653 | } | |
5654 | ||
395e68ce BP |
5655 | /* Find the port and bundle for the received packet. */ |
5656 | ofport = get_ofp_port(ofproto, in_port); | |
5657 | if (ofport && ofport->bundle) { | |
3581c12c | 5658 | return ofport->bundle; |
395e68ce BP |
5659 | } |
5660 | ||
5661 | /* Odd. A few possible reasons here: | |
5662 | * | |
5663 | * - We deleted a port but there are still a few packets queued up | |
5664 | * from it. | |
5665 | * | |
5666 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
5667 | * we don't know about. | |
5668 | * | |
5669 | * - The ofproto client didn't configure the port as part of a bundle. | |
5670 | */ | |
5671 | if (warn) { | |
5672 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5673 | ||
5674 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
5675 | "port %"PRIu16, ofproto->up.name, in_port); | |
5676 | } | |
5677 | return NULL; | |
5678 | } | |
5679 | ||
5da5ec37 | 5680 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
5681 | * dropped. Returns true if they may be forwarded, false if they should be |
5682 | * dropped. | |
5683 | * | |
395e68ce BP |
5684 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
5685 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 5686 | * |
395e68ce BP |
5687 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
5688 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
5689 | * checked by input_vid_is_valid(). | |
abe529af BP |
5690 | * |
5691 | * May also add tags to '*tags', although the current implementation only does | |
5692 | * so in one special case. | |
5693 | */ | |
5694 | static bool | |
5695 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
395e68ce | 5696 | struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) |
abe529af | 5697 | { |
395e68ce | 5698 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 5699 | |
395e68ce BP |
5700 | /* Drop frames for reserved multicast addresses |
5701 | * only if forward_bpdu option is absent. */ | |
21f7563c | 5702 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { |
abe529af BP |
5703 | return false; |
5704 | } | |
5705 | ||
abe529af BP |
5706 | if (in_bundle->bond) { |
5707 | struct mac_entry *mac; | |
5708 | ||
5709 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
5710 | flow->dl_dst, tags)) { | |
5711 | case BV_ACCEPT: | |
5712 | break; | |
5713 | ||
5714 | case BV_DROP: | |
5715 | return false; | |
5716 | ||
5717 | case BV_DROP_IF_MOVED: | |
5718 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
5719 | if (mac && mac->port.p != in_bundle && | |
5720 | (!is_gratuitous_arp(flow) | |
5721 | || mac_entry_is_grat_arp_locked(mac))) { | |
5722 | return false; | |
5723 | } | |
5724 | break; | |
5725 | } | |
5726 | } | |
5727 | ||
5728 | return true; | |
5729 | } | |
5730 | ||
4cd78906 | 5731 | static void |
abe529af BP |
5732 | xlate_normal(struct action_xlate_ctx *ctx) |
5733 | { | |
395e68ce | 5734 | struct ofport_dpif *in_port; |
abe529af | 5735 | struct ofbundle *in_bundle; |
abe529af | 5736 | struct mac_entry *mac; |
395e68ce BP |
5737 | uint16_t vlan; |
5738 | uint16_t vid; | |
abe529af | 5739 | |
75a75043 BP |
5740 | ctx->has_normal = true; |
5741 | ||
3581c12c | 5742 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, |
395e68ce | 5743 | ctx->packet != NULL); |
3581c12c | 5744 | if (!in_bundle) { |
395e68ce BP |
5745 | return; |
5746 | } | |
3581c12c | 5747 | |
33158a18 JP |
5748 | /* We know 'in_port' exists unless it is "ofpp_none_bundle", |
5749 | * since lookup_input_bundle() succeeded. */ | |
3581c12c | 5750 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); |
395e68ce BP |
5751 | |
5752 | /* Drop malformed frames. */ | |
5753 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
5754 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
5755 | if (ctx->packet != NULL) { | |
5756 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5757 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
5758 | "VLAN tag received on port %s", | |
5759 | ctx->ofproto->up.name, in_bundle->name); | |
5760 | } | |
5761 | return; | |
5762 | } | |
5763 | ||
5764 | /* Drop frames on bundles reserved for mirroring. */ | |
5765 | if (in_bundle->mirror_out) { | |
5766 | if (ctx->packet != NULL) { | |
5767 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5768 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5769 | "%s, which is reserved exclusively for mirroring", | |
5770 | ctx->ofproto->up.name, in_bundle->name); | |
5771 | } | |
5772 | return; | |
5773 | } | |
5774 | ||
5775 | /* Check VLAN. */ | |
5776 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
5777 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5778 | return; | |
5779 | } | |
5780 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5781 | ||
5782 | /* Check other admissibility requirements. */ | |
33158a18 JP |
5783 | if (in_port && |
5784 | !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { | |
395e68ce | 5785 | return; |
abe529af BP |
5786 | } |
5787 | ||
75a75043 | 5788 | /* Learn source MAC. */ |
3de9590b | 5789 | if (ctx->may_learn) { |
abe529af BP |
5790 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
5791 | } | |
5792 | ||
5793 | /* Determine output bundle. */ | |
5794 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
5795 | &ctx->tags); | |
5796 | if (mac) { | |
c06bba01 JP |
5797 | if (mac->port.p != in_bundle) { |
5798 | output_normal(ctx, mac->port.p, vlan); | |
5799 | } | |
abe529af | 5800 | } else { |
c06bba01 | 5801 | struct ofbundle *bundle; |
abe529af | 5802 | |
c06bba01 JP |
5803 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { |
5804 | if (bundle != in_bundle | |
5805 | && ofbundle_includes_vlan(bundle, vlan) | |
5806 | && bundle->floodable | |
5807 | && !bundle->mirror_out) { | |
5808 | output_normal(ctx, bundle, vlan); | |
5809 | } | |
5810 | } | |
5811 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af | 5812 | } |
abe529af BP |
5813 | } |
5814 | \f | |
54a9cbc9 BP |
5815 | /* Optimized flow revalidation. |
5816 | * | |
5817 | * It's a difficult problem, in general, to tell which facets need to have | |
5818 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
5819 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
5820 | * table changes, we recalculate the actions for every facet. This is | |
5821 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
5822 | * doesn't change very often. | |
5823 | * | |
5824 | * However, we can expect one particular kind of OpenFlow flow table change to | |
5825 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
5826 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
5827 | * table, we add a special case that applies to flow tables in which every rule | |
5828 | * has the same form (that is, the same wildcards), except that the table is | |
5829 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
5830 | * optimize this case by tagging all of the facets that resubmit into the table | |
5831 | * and invalidating the same tag whenever a flow changes in that table. The | |
5832 | * end result is that we revalidate just the facets that need it (and sometimes | |
5833 | * a few more, but not all of the facets or even all of the facets that | |
5834 | * resubmit to the table modified by MAC learning). */ | |
5835 | ||
5836 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
5837 | * into an OpenFlow table with the given 'basis'. */ | |
822d9414 | 5838 | static tag_type |
54a9cbc9 BP |
5839 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, |
5840 | uint32_t secret) | |
5841 | { | |
5842 | if (flow_wildcards_is_catchall(wc)) { | |
5843 | return 0; | |
5844 | } else { | |
5845 | struct flow tag_flow = *flow; | |
5846 | flow_zero_wildcards(&tag_flow, wc); | |
5847 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
5848 | } | |
5849 | } | |
5850 | ||
5851 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
5852 | * taggability of that table. | |
5853 | * | |
5854 | * This function must be called after *each* change to a flow table. If you | |
5855 | * skip calling it on some changes then the pointer comparisons at the end can | |
5856 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
5857 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
5858 | * different wildcards to be created with the same address, then this function | |
5859 | * will incorrectly skip revalidation. */ | |
5860 | static void | |
5861 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
5862 | { | |
5863 | struct table_dpif *table = &ofproto->tables[table_id]; | |
d0918789 | 5864 | const struct oftable *oftable = &ofproto->up.tables[table_id]; |
54a9cbc9 BP |
5865 | struct cls_table *catchall, *other; |
5866 | struct cls_table *t; | |
5867 | ||
5868 | catchall = other = NULL; | |
5869 | ||
d0918789 | 5870 | switch (hmap_count(&oftable->cls.tables)) { |
54a9cbc9 BP |
5871 | case 0: |
5872 | /* We could tag this OpenFlow table but it would make the logic a | |
5873 | * little harder and it's a corner case that doesn't seem worth it | |
5874 | * yet. */ | |
5875 | break; | |
5876 | ||
5877 | case 1: | |
5878 | case 2: | |
d0918789 | 5879 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { |
54a9cbc9 BP |
5880 | if (cls_table_is_catchall(t)) { |
5881 | catchall = t; | |
5882 | } else if (!other) { | |
5883 | other = t; | |
5884 | } else { | |
5885 | /* Indicate that we can't tag this by setting both tables to | |
5886 | * NULL. (We know that 'catchall' is already NULL.) */ | |
5887 | other = NULL; | |
5888 | } | |
5889 | } | |
5890 | break; | |
5891 | ||
5892 | default: | |
5893 | /* Can't tag this table. */ | |
5894 | break; | |
5895 | } | |
5896 | ||
5897 | if (table->catchall_table != catchall || table->other_table != other) { | |
5898 | table->catchall_table = catchall; | |
5899 | table->other_table = other; | |
5900 | ofproto->need_revalidate = true; | |
5901 | } | |
5902 | } | |
5903 | ||
5904 | /* Given 'rule' that has changed in some way (either it is a rule being | |
5905 | * inserted, a rule being deleted, or a rule whose actions are being | |
5906 | * modified), marks facets for revalidation to ensure that packets will be | |
5907 | * forwarded correctly according to the new state of the flow table. | |
5908 | * | |
5909 | * This function must be called after *each* change to a flow table. See | |
5910 | * the comment on table_update_taggable() for more information. */ | |
5911 | static void | |
5912 | rule_invalidate(const struct rule_dpif *rule) | |
5913 | { | |
5914 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5915 | ||
5916 | table_update_taggable(ofproto, rule->up.table_id); | |
5917 | ||
5918 | if (!ofproto->need_revalidate) { | |
5919 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
5920 | ||
5921 | if (table->other_table && rule->tag) { | |
5922 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
5923 | } else { | |
5924 | ofproto->need_revalidate = true; | |
5925 | } | |
5926 | } | |
5927 | } | |
5928 | \f | |
abe529af | 5929 | static bool |
7257b535 BP |
5930 | set_frag_handling(struct ofproto *ofproto_, |
5931 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5932 | { |
5933 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 5934 | |
7257b535 BP |
5935 | if (frag_handling != OFPC_FRAG_REASM) { |
5936 | ofproto->need_revalidate = true; | |
5937 | return true; | |
5938 | } else { | |
5939 | return false; | |
5940 | } | |
abe529af BP |
5941 | } |
5942 | ||
90bf1e07 | 5943 | static enum ofperr |
abe529af BP |
5944 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, |
5945 | const struct flow *flow, | |
5946 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
5947 | { | |
5948 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
90bf1e07 | 5949 | enum ofperr error; |
abe529af | 5950 | |
e1154f71 | 5951 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
90bf1e07 | 5952 | return OFPERR_NXBRC_BAD_IN_PORT; |
e1154f71 BP |
5953 | } |
5954 | ||
abe529af BP |
5955 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
5956 | ofproto->max_ports); | |
5957 | if (!error) { | |
80e5eed9 | 5958 | struct odputil_keybuf keybuf; |
80e5eed9 BP |
5959 | struct ofpbuf key; |
5960 | ||
050ac423 BP |
5961 | uint64_t odp_actions_stub[1024 / 8]; |
5962 | struct ofpbuf odp_actions; | |
5963 | struct ofproto_push push; | |
5964 | ||
80e5eed9 BP |
5965 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
5966 | odp_flow_key_from_flow(&key, flow); | |
abe529af | 5967 | |
18b2a258 | 5968 | action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL, |
0e553d9c | 5969 | packet_get_tcp_flags(packet, flow), packet); |
2284188b EJ |
5970 | |
5971 | /* Ensure that resubmits in 'ofp_actions' get accounted to their | |
5972 | * matching rules. */ | |
5973 | push.packets = 1; | |
5974 | push.bytes = packet->size; | |
5975 | push.used = time_msec(); | |
5976 | push.ctx.resubmit_hook = push_resubmit; | |
5977 | ||
050ac423 BP |
5978 | ofpbuf_use_stub(&odp_actions, |
5979 | odp_actions_stub, sizeof odp_actions_stub); | |
5980 | xlate_actions(&push.ctx, ofp_actions, n_ofp_actions, &odp_actions); | |
80e5eed9 | 5981 | dpif_execute(ofproto->dpif, key.data, key.size, |
050ac423 BP |
5982 | odp_actions.data, odp_actions.size, packet); |
5983 | ofpbuf_uninit(&odp_actions); | |
abe529af BP |
5984 | } |
5985 | return error; | |
5986 | } | |
6fca1ffb BP |
5987 | \f |
5988 | /* NetFlow. */ | |
5989 | ||
5990 | static int | |
5991 | set_netflow(struct ofproto *ofproto_, | |
5992 | const struct netflow_options *netflow_options) | |
5993 | { | |
5994 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5995 | ||
5996 | if (netflow_options) { | |
5997 | if (!ofproto->netflow) { | |
5998 | ofproto->netflow = netflow_create(); | |
5999 | } | |
6000 | return netflow_set_options(ofproto->netflow, netflow_options); | |
6001 | } else { | |
6002 | netflow_destroy(ofproto->netflow); | |
6003 | ofproto->netflow = NULL; | |
6004 | return 0; | |
6005 | } | |
6006 | } | |
abe529af BP |
6007 | |
6008 | static void | |
6009 | get_netflow_ids(const struct ofproto *ofproto_, | |
6010 | uint8_t *engine_type, uint8_t *engine_id) | |
6011 | { | |
6012 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6013 | ||
6014 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
6015 | } | |
6fca1ffb BP |
6016 | |
6017 | static void | |
6018 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
6019 | { | |
6020 | if (!facet_is_controller_flow(facet) && | |
6021 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
b0f7b9b5 | 6022 | struct subfacet *subfacet; |
6fca1ffb BP |
6023 | struct ofexpired expired; |
6024 | ||
b0f7b9b5 BP |
6025 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { |
6026 | if (subfacet->installed) { | |
6027 | struct dpif_flow_stats stats; | |
6fca1ffb | 6028 | |
15baa734 | 6029 | subfacet_install(subfacet, subfacet->actions, |
b95fc6ba | 6030 | subfacet->actions_len, &stats); |
15baa734 | 6031 | subfacet_update_stats(subfacet, &stats); |
b0f7b9b5 | 6032 | } |
6fca1ffb BP |
6033 | } |
6034 | ||
6035 | expired.flow = facet->flow; | |
6036 | expired.packet_count = facet->packet_count; | |
6037 | expired.byte_count = facet->byte_count; | |
6038 | expired.used = facet->used; | |
6039 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
6040 | } | |
6041 | } | |
6042 | ||
6043 | static void | |
6044 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
6045 | { | |
6046 | struct facet *facet; | |
6047 | ||
6048 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6049 | send_active_timeout(ofproto, facet); | |
6050 | } | |
6051 | } | |
abe529af BP |
6052 | \f |
6053 | static struct ofproto_dpif * | |
6054 | ofproto_dpif_lookup(const char *name) | |
6055 | { | |
b44a10b7 BP |
6056 | struct ofproto_dpif *ofproto; |
6057 | ||
6058 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
6059 | hash_string(name, 0), &all_ofproto_dpifs) { | |
6060 | if (!strcmp(ofproto->up.name, name)) { | |
6061 | return ofproto; | |
6062 | } | |
6063 | } | |
6064 | return NULL; | |
abe529af BP |
6065 | } |
6066 | ||
f0a3aa2e | 6067 | static void |
96e466a3 | 6068 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, |
0e15264f | 6069 | const char *argv[], void *aux OVS_UNUSED) |
f0a3aa2e | 6070 | { |
490df1ef | 6071 | struct ofproto_dpif *ofproto; |
f0a3aa2e | 6072 | |
96e466a3 EJ |
6073 | if (argc > 1) { |
6074 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6075 | if (!ofproto) { | |
bde9f75d | 6076 | unixctl_command_reply_error(conn, "no such bridge"); |
96e466a3 EJ |
6077 | return; |
6078 | } | |
d0040604 | 6079 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 EJ |
6080 | } else { |
6081 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
d0040604 | 6082 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); |
96e466a3 | 6083 | } |
f0a3aa2e | 6084 | } |
f0a3aa2e | 6085 | |
bde9f75d | 6086 | unixctl_command_reply(conn, "table successfully flushed"); |
f0a3aa2e AA |
6087 | } |
6088 | ||
abe529af | 6089 | static void |
0e15264f BP |
6090 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, |
6091 | const char *argv[], void *aux OVS_UNUSED) | |
abe529af BP |
6092 | { |
6093 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6094 | const struct ofproto_dpif *ofproto; | |
6095 | const struct mac_entry *e; | |
6096 | ||
0e15264f | 6097 | ofproto = ofproto_dpif_lookup(argv[1]); |
abe529af | 6098 | if (!ofproto) { |
bde9f75d | 6099 | unixctl_command_reply_error(conn, "no such bridge"); |
abe529af BP |
6100 | return; |
6101 | } | |
6102 | ||
6103 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
6104 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
6105 | struct ofbundle *bundle = e->port.p; | |
6106 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
6107 | ofbundle_get_a_port(bundle)->odp_port, | |
e764773c BP |
6108 | e->vlan, ETH_ADDR_ARGS(e->mac), |
6109 | mac_entry_age(ofproto->ml, e)); | |
abe529af | 6110 | } |
bde9f75d | 6111 | unixctl_command_reply(conn, ds_cstr(&ds)); |
abe529af BP |
6112 | ds_destroy(&ds); |
6113 | } | |
6114 | ||
6a6455e5 | 6115 | struct trace_ctx { |
abe529af BP |
6116 | struct action_xlate_ctx ctx; |
6117 | struct flow flow; | |
6118 | struct ds *result; | |
6119 | }; | |
6120 | ||
6121 | static void | |
29901626 BP |
6122 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
6123 | const struct rule_dpif *rule) | |
abe529af BP |
6124 | { |
6125 | ds_put_char_multiple(result, '\t', level); | |
6126 | if (!rule) { | |
6127 | ds_put_cstr(result, "No match\n"); | |
6128 | return; | |
6129 | } | |
6130 | ||
29901626 BP |
6131 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
6132 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 6133 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
6134 | ds_put_char(result, '\n'); |
6135 | ||
6136 | ds_put_char_multiple(result, '\t', level); | |
6137 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 6138 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
6139 | ds_put_char(result, '\n'); |
6140 | } | |
6141 | ||
6142 | static void | |
6143 | trace_format_flow(struct ds *result, int level, const char *title, | |
6a6455e5 | 6144 | struct trace_ctx *trace) |
abe529af BP |
6145 | { |
6146 | ds_put_char_multiple(result, '\t', level); | |
6147 | ds_put_format(result, "%s: ", title); | |
6148 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
6149 | ds_put_cstr(result, "unchanged"); | |
6150 | } else { | |
6151 | flow_format(result, &trace->ctx.flow); | |
6152 | trace->flow = trace->ctx.flow; | |
6153 | } | |
6154 | ds_put_char(result, '\n'); | |
6155 | } | |
6156 | ||
eb9e1c26 EJ |
6157 | static void |
6158 | trace_format_regs(struct ds *result, int level, const char *title, | |
6a6455e5 | 6159 | struct trace_ctx *trace) |
eb9e1c26 EJ |
6160 | { |
6161 | size_t i; | |
6162 | ||
6163 | ds_put_char_multiple(result, '\t', level); | |
6164 | ds_put_format(result, "%s:", title); | |
6165 | for (i = 0; i < FLOW_N_REGS; i++) { | |
6166 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
6167 | } | |
6168 | ds_put_char(result, '\n'); | |
6169 | } | |
6170 | ||
1ed8d352 EJ |
6171 | static void |
6172 | trace_format_odp(struct ds *result, int level, const char *title, | |
6a6455e5 | 6173 | struct trace_ctx *trace) |
1ed8d352 EJ |
6174 | { |
6175 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
6176 | ||
6177 | ds_put_char_multiple(result, '\t', level); | |
6178 | ds_put_format(result, "%s: ", title); | |
6179 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
6180 | ds_put_char(result, '\n'); | |
6181 | } | |
6182 | ||
abe529af BP |
6183 | static void |
6184 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6185 | { | |
6a6455e5 | 6186 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); |
abe529af BP |
6187 | struct ds *result = trace->result; |
6188 | ||
6189 | ds_put_char(result, '\n'); | |
6190 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 6191 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
1ed8d352 | 6192 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); |
29901626 | 6193 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
6194 | } |
6195 | ||
6196 | static void | |
0e15264f | 6197 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], |
abe529af BP |
6198 | void *aux OVS_UNUSED) |
6199 | { | |
0e15264f | 6200 | const char *dpname = argv[1]; |
abe529af | 6201 | struct ofproto_dpif *ofproto; |
876b0e1c BP |
6202 | struct ofpbuf odp_key; |
6203 | struct ofpbuf *packet; | |
e84173dc | 6204 | ovs_be16 initial_tci; |
abe529af BP |
6205 | struct ds result; |
6206 | struct flow flow; | |
abe529af BP |
6207 | char *s; |
6208 | ||
876b0e1c BP |
6209 | packet = NULL; |
6210 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
6211 | ds_init(&result); |
6212 | ||
e84173dc BP |
6213 | ofproto = ofproto_dpif_lookup(dpname); |
6214 | if (!ofproto) { | |
bde9f75d EJ |
6215 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " |
6216 | "for help)"); | |
e84173dc BP |
6217 | goto exit; |
6218 | } | |
0e15264f | 6219 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { |
8b3b8dd1 | 6220 | /* ofproto/trace dpname flow [-generate] */ |
0e15264f BP |
6221 | const char *flow_s = argv[2]; |
6222 | const char *generate_s = argv[3]; | |
876b0e1c BP |
6223 | int error; |
6224 | ||
df2c07f4 | 6225 | /* Convert string to datapath key. */ |
876b0e1c | 6226 | ofpbuf_init(&odp_key, 0); |
0e15264f | 6227 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); |
876b0e1c | 6228 | if (error) { |
bde9f75d | 6229 | unixctl_command_reply_error(conn, "Bad flow syntax"); |
876b0e1c BP |
6230 | goto exit; |
6231 | } | |
6232 | ||
6233 | /* Convert odp_key to flow. */ | |
e84173dc BP |
6234 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, |
6235 | odp_key.size, &flow, | |
e2a6ca36 | 6236 | &initial_tci, NULL); |
e84173dc | 6237 | if (error == ODP_FIT_ERROR) { |
bde9f75d | 6238 | unixctl_command_reply_error(conn, "Invalid flow"); |
876b0e1c BP |
6239 | goto exit; |
6240 | } | |
8b3b8dd1 BP |
6241 | |
6242 | /* Generate a packet, if requested. */ | |
0e15264f | 6243 | if (generate_s) { |
8b3b8dd1 BP |
6244 | packet = ofpbuf_new(0); |
6245 | flow_compose(packet, &flow); | |
6246 | } | |
0e15264f | 6247 | } else if (argc == 6) { |
abff858b | 6248 | /* ofproto/trace dpname priority tun_id in_port packet */ |
0e15264f BP |
6249 | const char *priority_s = argv[2]; |
6250 | const char *tun_id_s = argv[3]; | |
6251 | const char *in_port_s = argv[4]; | |
6252 | const char *packet_s = argv[5]; | |
6253 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
6254 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
6255 | uint32_t priority = atoi(priority_s); | |
e22f1753 | 6256 | const char *msg; |
0e15264f | 6257 | |
e22f1753 BP |
6258 | msg = eth_from_hex(packet_s, &packet); |
6259 | if (msg) { | |
bde9f75d | 6260 | unixctl_command_reply_error(conn, msg); |
876b0e1c BP |
6261 | goto exit; |
6262 | } | |
6263 | ||
6264 | ds_put_cstr(&result, "Packet: "); | |
c499c75d | 6265 | s = ofp_packet_to_string(packet->data, packet->size); |
876b0e1c BP |
6266 | ds_put_cstr(&result, s); |
6267 | free(s); | |
6268 | ||
abff858b | 6269 | flow_extract(packet, priority, tun_id, in_port, &flow); |
e84173dc | 6270 | initial_tci = flow.vlan_tci; |
876b0e1c | 6271 | } else { |
bde9f75d | 6272 | unixctl_command_reply_error(conn, "Bad command syntax"); |
abe529af BP |
6273 | goto exit; |
6274 | } | |
6275 | ||
6a6455e5 EJ |
6276 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); |
6277 | unixctl_command_reply(conn, ds_cstr(&result)); | |
6278 | ||
6279 | exit: | |
6280 | ds_destroy(&result); | |
6281 | ofpbuf_delete(packet); | |
6282 | ofpbuf_uninit(&odp_key); | |
6283 | } | |
6284 | ||
6285 | static void | |
6286 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
6287 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
6288 | struct ds *ds) | |
6289 | { | |
6290 | struct rule_dpif *rule; | |
6291 | ||
6292 | ds_put_cstr(ds, "Flow: "); | |
6293 | flow_format(ds, flow); | |
6294 | ds_put_char(ds, '\n'); | |
abe529af | 6295 | |
6a6455e5 EJ |
6296 | rule = rule_dpif_lookup(ofproto, flow, 0); |
6297 | trace_format_rule(ds, 0, 0, rule); | |
abe529af | 6298 | if (rule) { |
050ac423 BP |
6299 | uint64_t odp_actions_stub[1024 / 8]; |
6300 | struct ofpbuf odp_actions; | |
6301 | ||
6a6455e5 | 6302 | struct trace_ctx trace; |
0e553d9c | 6303 | uint8_t tcp_flags; |
abe529af | 6304 | |
6a6455e5 EJ |
6305 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; |
6306 | trace.result = ds; | |
6307 | trace.flow = *flow; | |
050ac423 BP |
6308 | ofpbuf_use_stub(&odp_actions, |
6309 | odp_actions_stub, sizeof odp_actions_stub); | |
6a6455e5 | 6310 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, |
0e553d9c | 6311 | rule, tcp_flags, packet); |
abe529af | 6312 | trace.ctx.resubmit_hook = trace_resubmit; |
050ac423 BP |
6313 | xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions, |
6314 | &odp_actions); | |
abe529af | 6315 | |
6a6455e5 EJ |
6316 | ds_put_char(ds, '\n'); |
6317 | trace_format_flow(ds, 0, "Final flow", &trace); | |
6318 | ds_put_cstr(ds, "Datapath actions: "); | |
050ac423 BP |
6319 | format_odp_actions(ds, odp_actions.data, odp_actions.size); |
6320 | ofpbuf_uninit(&odp_actions); | |
876b0e1c BP |
6321 | |
6322 | if (!trace.ctx.may_set_up_flow) { | |
6323 | if (packet) { | |
6a6455e5 | 6324 | ds_put_cstr(ds, "\nThis flow is not cachable."); |
876b0e1c | 6325 | } else { |
6a6455e5 | 6326 | ds_put_cstr(ds, "\nThe datapath actions are incomplete--" |
876b0e1c BP |
6327 | "for complete actions, please supply a packet."); |
6328 | } | |
6329 | } | |
abe529af | 6330 | } |
abe529af BP |
6331 | } |
6332 | ||
7ee20df1 | 6333 | static void |
0e15264f BP |
6334 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6335 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6336 | { |
6337 | clogged = true; | |
bde9f75d | 6338 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6339 | } |
6340 | ||
6341 | static void | |
0e15264f BP |
6342 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, |
6343 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
7ee20df1 BP |
6344 | { |
6345 | clogged = false; | |
bde9f75d | 6346 | unixctl_command_reply(conn, NULL); |
7ee20df1 BP |
6347 | } |
6348 | ||
6814e51f BP |
6349 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to |
6350 | * 'reply' describing the results. */ | |
6351 | static void | |
6352 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6353 | { | |
6354 | struct facet *facet; | |
6355 | int errors; | |
6356 | ||
6357 | errors = 0; | |
6358 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6359 | if (!facet_check_consistency(facet)) { | |
6360 | errors++; | |
6361 | } | |
6362 | } | |
6363 | if (errors) { | |
6364 | ofproto->need_revalidate = true; | |
6365 | } | |
6366 | ||
6367 | if (errors) { | |
6368 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6369 | ofproto->up.name, errors); | |
6370 | } else { | |
6371 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6372 | } | |
6373 | } | |
6374 | ||
6375 | static void | |
6376 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6377 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6378 | { | |
6379 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6380 | struct ofproto_dpif *ofproto; | |
6381 | ||
6382 | if (argc > 1) { | |
6383 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6384 | if (!ofproto) { | |
bde9f75d EJ |
6385 | unixctl_command_reply_error(conn, "Unknown ofproto (use " |
6386 | "ofproto/list for help)"); | |
6814e51f BP |
6387 | return; |
6388 | } | |
6389 | ofproto_dpif_self_check__(ofproto, &reply); | |
6390 | } else { | |
6391 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6392 | ofproto_dpif_self_check__(ofproto, &reply); | |
6393 | } | |
6394 | } | |
6395 | ||
bde9f75d | 6396 | unixctl_command_reply(conn, ds_cstr(&reply)); |
6814e51f BP |
6397 | ds_destroy(&reply); |
6398 | } | |
6399 | ||
abe529af BP |
6400 | static void |
6401 | ofproto_dpif_unixctl_init(void) | |
6402 | { | |
6403 | static bool registered; | |
6404 | if (registered) { | |
6405 | return; | |
6406 | } | |
6407 | registered = true; | |
6408 | ||
0e15264f BP |
6409 | unixctl_command_register( |
6410 | "ofproto/trace", | |
6411 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
aa3080c9 | 6412 | 2, 5, ofproto_unixctl_trace, NULL); |
96e466a3 | 6413 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, |
0e15264f BP |
6414 | ofproto_unixctl_fdb_flush, NULL); |
6415 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6416 | ofproto_unixctl_fdb_show, NULL); | |
6417 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6418 | ofproto_dpif_clog, NULL); | |
6419 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6420 | ofproto_dpif_unclog, NULL); | |
6814e51f BP |
6421 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, |
6422 | ofproto_dpif_self_check, NULL); | |
abe529af BP |
6423 | } |
6424 | \f | |
52a90c29 BP |
6425 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) |
6426 | * | |
6427 | * This is deprecated. It is only for compatibility with broken device drivers | |
6428 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6429 | * devices are not used. When broken device drivers are no longer in | |
6430 | * widespread use, we will delete these interfaces. */ | |
6431 | ||
6432 | static int | |
6433 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6434 | { | |
6435 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6436 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6437 | ||
6438 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6439 | && vid == ofport->vlandev_vid) { | |
6440 | return 0; | |
6441 | } | |
6442 | ||
6443 | ofproto->need_revalidate = true; | |
6444 | ||
6445 | if (ofport->realdev_ofp_port) { | |
6446 | vsp_remove(ofport); | |
6447 | } | |
6448 | if (realdev_ofp_port && ofport->bundle) { | |
6449 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6450 | * themselves be part of a bundle. */ | |
6451 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6452 | } | |
6453 | ||
6454 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6455 | ofport->vlandev_vid = vid; | |
6456 | ||
6457 | if (realdev_ofp_port) { | |
6458 | vsp_add(ofport, realdev_ofp_port, vid); | |
6459 | } | |
6460 | ||
6461 | return 0; | |
6462 | } | |
6463 | ||
6464 | static uint32_t | |
6465 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
6466 | { | |
6467 | return hash_2words(realdev_ofp_port, vid); | |
6468 | } | |
6469 | ||
40e05935 BP |
6470 | /* Returns the ODP port number of the Linux VLAN device that corresponds to |
6471 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
6472 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
6473 | * it would return the port number of eth0.9. | |
6474 | * | |
6475 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
6476 | * function just returns its 'realdev_odp_port' argument. */ | |
52a90c29 BP |
6477 | static uint32_t |
6478 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
6479 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
6480 | { | |
6481 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
6482 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
6483 | int vid = vlan_tci_to_vid(vlan_tci); | |
6484 | const struct vlan_splinter *vsp; | |
6485 | ||
6486 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
6487 | hash_realdev_vid(realdev_ofp_port, vid), | |
6488 | &ofproto->realdev_vid_map) { | |
6489 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
6490 | && vsp->vid == vid) { | |
6491 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
6492 | } | |
6493 | } | |
6494 | } | |
6495 | return realdev_odp_port; | |
6496 | } | |
6497 | ||
6498 | static struct vlan_splinter * | |
6499 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
6500 | { | |
6501 | struct vlan_splinter *vsp; | |
6502 | ||
6503 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
6504 | &ofproto->vlandev_map) { | |
6505 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
6506 | return vsp; | |
6507 | } | |
6508 | } | |
6509 | ||
6510 | return NULL; | |
6511 | } | |
6512 | ||
40e05935 BP |
6513 | /* Returns the OpenFlow port number of the "real" device underlying the Linux |
6514 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
6515 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
6516 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
6517 | * eth0 and store 9 in '*vid'. | |
6518 | * | |
6519 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
6520 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
6521 | * always does.*/ | |
52a90c29 BP |
6522 | static uint16_t |
6523 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
40e05935 | 6524 | uint16_t vlandev_ofp_port, int *vid) |
52a90c29 BP |
6525 | { |
6526 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
6527 | const struct vlan_splinter *vsp; | |
6528 | ||
6529 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
6530 | if (vsp) { | |
6531 | if (vid) { | |
6532 | *vid = vsp->vid; | |
6533 | } | |
6534 | return vsp->realdev_ofp_port; | |
6535 | } | |
6536 | } | |
6537 | return 0; | |
6538 | } | |
6539 | ||
6540 | static void | |
6541 | vsp_remove(struct ofport_dpif *port) | |
6542 | { | |
6543 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6544 | struct vlan_splinter *vsp; | |
6545 | ||
6546 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
6547 | if (vsp) { | |
6548 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
6549 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
6550 | free(vsp); | |
6551 | ||
6552 | port->realdev_ofp_port = 0; | |
6553 | } else { | |
6554 | VLOG_ERR("missing vlan device record"); | |
6555 | } | |
6556 | } | |
6557 | ||
6558 | static void | |
6559 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
6560 | { | |
6561 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
6562 | ||
6563 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
6564 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
6565 | == realdev_ofp_port)) { | |
6566 | struct vlan_splinter *vsp; | |
6567 | ||
6568 | vsp = xmalloc(sizeof *vsp); | |
6569 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
6570 | hash_int(port->up.ofp_port, 0)); | |
6571 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
6572 | hash_realdev_vid(realdev_ofp_port, vid)); | |
6573 | vsp->realdev_ofp_port = realdev_ofp_port; | |
6574 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
6575 | vsp->vid = vid; | |
6576 | ||
6577 | port->realdev_ofp_port = realdev_ofp_port; | |
6578 | } else { | |
6579 | VLOG_ERR("duplicate vlan device record"); | |
6580 | } | |
6581 | } | |
6582 | \f | |
abe529af BP |
6583 | const struct ofproto_class ofproto_dpif_class = { |
6584 | enumerate_types, | |
6585 | enumerate_names, | |
6586 | del, | |
6587 | alloc, | |
6588 | construct, | |
6589 | destruct, | |
6590 | dealloc, | |
6591 | run, | |
5fcc0d00 | 6592 | run_fast, |
abe529af BP |
6593 | wait, |
6594 | flush, | |
6c1491fb BP |
6595 | get_features, |
6596 | get_tables, | |
abe529af BP |
6597 | port_alloc, |
6598 | port_construct, | |
6599 | port_destruct, | |
6600 | port_dealloc, | |
6601 | port_modified, | |
6602 | port_reconfigured, | |
6603 | port_query_by_name, | |
6604 | port_add, | |
6605 | port_del, | |
6527c598 | 6606 | port_get_stats, |
abe529af BP |
6607 | port_dump_start, |
6608 | port_dump_next, | |
6609 | port_dump_done, | |
6610 | port_poll, | |
6611 | port_poll_wait, | |
6612 | port_is_lacp_current, | |
0ab6decf | 6613 | NULL, /* rule_choose_table */ |
abe529af BP |
6614 | rule_alloc, |
6615 | rule_construct, | |
6616 | rule_destruct, | |
6617 | rule_dealloc, | |
abe529af BP |
6618 | rule_get_stats, |
6619 | rule_execute, | |
6620 | rule_modify_actions, | |
7257b535 | 6621 | set_frag_handling, |
abe529af BP |
6622 | packet_out, |
6623 | set_netflow, | |
6624 | get_netflow_ids, | |
6625 | set_sflow, | |
6626 | set_cfm, | |
a5610457 | 6627 | get_cfm_fault, |
1de11730 | 6628 | get_cfm_remote_mpids, |
3967a833 | 6629 | get_cfm_health, |
21f7563c JP |
6630 | set_stp, |
6631 | get_stp_status, | |
6632 | set_stp_port, | |
6633 | get_stp_port_status, | |
8b36f51e | 6634 | set_queues, |
abe529af BP |
6635 | bundle_set, |
6636 | bundle_remove, | |
6637 | mirror_set, | |
9d24de3b | 6638 | mirror_get_stats, |
abe529af BP |
6639 | set_flood_vlans, |
6640 | is_mirror_output_bundle, | |
8402c74b | 6641 | forward_bpdu_changed, |
e764773c | 6642 | set_mac_idle_time, |
52a90c29 | 6643 | set_realdev, |
abe529af | 6644 | }; |