]>
Commit | Line | Data |
---|---|---|
f620b43a | 1 | /* |
50f96b10 | 2 | * Copyright (c) 2008-2017 Nicira, Inc. |
f620b43a 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 | ||
19 | #include "bond.h" | |
20 | ||
21 | #include <limits.h> | |
22 | #include <stdint.h> | |
23 | #include <stdlib.h> | |
75fad143 | 24 | #include <math.h> |
f620b43a | 25 | |
da4a6191 | 26 | #include "connectivity.h" |
f620b43a | 27 | #include "coverage.h" |
b598f214 | 28 | #include "dp-packet.h" |
f620b43a | 29 | #include "flow.h" |
ee89ea7b | 30 | #include "openvswitch/hmap.h" |
bdebeece | 31 | #include "lacp.h" |
f620b43a BP |
32 | #include "netdev.h" |
33 | #include "odp-util.h" | |
b598f214 BW |
34 | #include "ofproto/ofproto-dpif.h" |
35 | #include "ofproto/ofproto-dpif-rid.h" | |
36 | #include "ofproto/ofproto-provider.h" | |
37 | #include "openvswitch/dynamic-string.h" | |
38 | #include "openvswitch/list.h" | |
39 | #include "openvswitch/match.h" | |
40 | #include "openvswitch/ofp-actions.h" | |
64c96779 | 41 | #include "openvswitch/ofpbuf.h" |
b598f214 | 42 | #include "openvswitch/vlog.h" |
f620b43a | 43 | #include "packets.h" |
fd016ae3 | 44 | #include "openvswitch/poll-loop.h" |
da4a6191 | 45 | #include "seq.h" |
ee89ea7b | 46 | #include "openvswitch/shash.h" |
f620b43a BP |
47 | #include "timeval.h" |
48 | #include "unixctl.h" | |
ee89ea7b | 49 | #include "util.h" |
f620b43a BP |
50 | |
51 | VLOG_DEFINE_THIS_MODULE(bond); | |
52 | ||
f1c8a79c AW |
53 | static struct ovs_rwlock rwlock = OVS_RWLOCK_INITIALIZER; |
54 | static struct hmap all_bonds__ = HMAP_INITIALIZER(&all_bonds__); | |
55 | static struct hmap *const all_bonds OVS_GUARDED_BY(rwlock) = &all_bonds__; | |
56 | ||
9e1a6910 | 57 | /* Bit-mask for hashing a flow down to a bucket. */ |
f620b43a | 58 | #define BOND_MASK 0xff |
9e1a6910 | 59 | #define BOND_BUCKETS (BOND_MASK + 1) |
f620b43a | 60 | |
07a3cd5c BP |
61 | /* Priority for internal rules created to handle recirculation */ |
62 | #define RECIRC_RULE_PRIORITY 20 | |
63 | ||
f620b43a | 64 | /* A hash bucket for mapping a flow to a slave. |
9e1a6910 | 65 | * "struct bond" has an array of BOND_BUCKETS of these. */ |
f620b43a BP |
66 | struct bond_entry { |
67 | struct bond_slave *slave; /* Assigned slave, NULL if unassigned. */ | |
c6855ec5 JS |
68 | uint64_t tx_bytes /* Count of bytes recently transmitted. */ |
69 | OVS_GUARDED_BY(rwlock); | |
ca6ba700 | 70 | struct ovs_list list_node; /* In bond_slave's 'entries' list. */ |
adcf00ba | 71 | |
c6855ec5 JS |
72 | /* Recirculation. |
73 | * | |
74 | * 'pr_rule' is the post-recirculation rule for this entry. | |
75 | * 'pr_tx_bytes' is the most recently seen statistics for 'pr_rule', which | |
76 | * is used to determine delta (applied to 'tx_bytes' above.) */ | |
77 | struct rule *pr_rule; | |
78 | uint64_t pr_tx_bytes OVS_GUARDED_BY(rwlock); | |
f620b43a BP |
79 | }; |
80 | ||
81 | /* A bond slave, that is, one of the links comprising a bond. */ | |
82 | struct bond_slave { | |
83 | struct hmap_node hmap_node; /* In struct bond's slaves hmap. */ | |
ca6ba700 | 84 | struct ovs_list list_node; /* In struct bond's enabled_slaves list. */ |
f620b43a BP |
85 | struct bond *bond; /* The bond that contains this slave. */ |
86 | void *aux; /* Client-provided handle for this slave. */ | |
87 | ||
88 | struct netdev *netdev; /* Network device, owned by the client. */ | |
6422372c | 89 | uint64_t change_seq; /* Tracks changes in 'netdev'. */ |
f620b43a | 90 | char *name; /* Name (a copy of netdev_get_name(netdev)). */ |
abec9228 | 91 | ofp_port_t ofp_port; /* OpenFlow port number. */ |
f620b43a BP |
92 | |
93 | /* Link status. */ | |
f620b43a | 94 | bool enabled; /* May be chosen for flows? */ |
296f6519 | 95 | bool may_enable; /* Client considers this slave bondable. */ |
abec9228 | 96 | long long delay_expires; /* Time after which 'enabled' may change. */ |
f620b43a BP |
97 | |
98 | /* Rebalancing info. Used only by bond_rebalance(). */ | |
ca6ba700 TG |
99 | struct ovs_list bal_node; /* In bond_rebalance()'s 'bals' list. */ |
100 | struct ovs_list entries; /* 'struct bond_entry's assigned here. */ | |
f620b43a BP |
101 | uint64_t tx_bytes; /* Sum across 'tx_bytes' of entries. */ |
102 | }; | |
103 | ||
104 | /* A bond, that is, a set of network devices grouped to improve performance or | |
105 | * robustness. */ | |
106 | struct bond { | |
107 | struct hmap_node hmap_node; /* In 'all_bonds' hmap. */ | |
108 | char *name; /* Name provided by client. */ | |
adcf00ba | 109 | struct ofproto_dpif *ofproto; /* The bridge this bond belongs to. */ |
f620b43a BP |
110 | |
111 | /* Slaves. */ | |
112 | struct hmap slaves; | |
113 | ||
f1c8a79c AW |
114 | /* Enabled slaves. |
115 | * | |
116 | * Any reader or writer of 'enabled_slaves' must hold 'mutex'. | |
117 | * (To prevent the bond_slave from disappearing they must also hold | |
118 | * 'rwlock'.) */ | |
119 | struct ovs_mutex mutex OVS_ACQ_AFTER(rwlock); | |
ca6ba700 | 120 | struct ovs_list enabled_slaves OVS_GUARDED; /* Contains struct bond_slaves. */ |
f1c8a79c | 121 | |
f620b43a BP |
122 | /* Bonding info. */ |
123 | enum bond_mode balance; /* Balancing mode, one of BM_*. */ | |
124 | struct bond_slave *active_slave; | |
f620b43a | 125 | int updelay, downdelay; /* Delay before slave goes up/down, in ms. */ |
bdebeece | 126 | enum lacp_status lacp_status; /* Status of LACP negotiations. */ |
62904702 | 127 | bool bond_revalidate; /* True if flows need revalidation. */ |
672d18b2 | 128 | uint32_t basis; /* Basis for flow hash function. */ |
f620b43a BP |
129 | |
130 | /* SLB specific bonding info. */ | |
9e1a6910 | 131 | struct bond_entry *hash; /* An array of BOND_BUCKETS elements. */ |
f620b43a BP |
132 | int rebalance_interval; /* Interval between rebalances, in ms. */ |
133 | long long int next_rebalance; /* Next rebalancing time. */ | |
134 | bool send_learning_packets; | |
adcf00ba AZ |
135 | uint32_t recirc_id; /* Non zero if recirculation can be used.*/ |
136 | struct hmap pr_rule_ops; /* Helps to maintain post recirculation rules.*/ | |
f620b43a | 137 | |
3e5aeeb5 AZ |
138 | /* Store active slave to OVSDB. */ |
139 | bool active_slave_changed; /* Set to true whenever the bond changes | |
140 | active slave. It will be reset to false | |
141 | after it is stored into OVSDB */ | |
142 | ||
143 | /* Interface name may not be persistent across an OS reboot, use | |
144 | * MAC address for identifing the active slave */ | |
74ff3298 | 145 | struct eth_addr active_slave_mac; |
3e5aeeb5 | 146 | /* The MAC address of the active interface. */ |
f620b43a | 147 | /* Legacy compatibility. */ |
9dd165e0 | 148 | bool lacp_fallback_ab; /* Fallback to active-backup on LACP failure. */ |
f620b43a | 149 | |
37bec3d3 | 150 | struct ovs_refcount ref_cnt; |
f620b43a BP |
151 | }; |
152 | ||
adcf00ba AZ |
153 | /* What to do with an bond_recirc_rule. */ |
154 | enum bond_op { | |
155 | ADD, /* Add the rule to ofproto's flow table. */ | |
156 | DEL, /* Delete the rule from the ofproto's flow table. */ | |
157 | }; | |
158 | ||
159 | /* A rule to add to or delete from ofproto's internal flow table. */ | |
160 | struct bond_pr_rule_op { | |
161 | struct hmap_node hmap_node; | |
162 | struct match match; | |
163 | ofp_port_t out_ofport; | |
164 | enum bond_op op; | |
6c932bc8 | 165 | struct rule **pr_rule; |
adcf00ba AZ |
166 | }; |
167 | ||
3bfd3972 EJ |
168 | static void bond_entry_reset(struct bond *) OVS_REQ_WRLOCK(rwlock); |
169 | static struct bond_slave *bond_slave_lookup(struct bond *, const void *slave_) | |
170 | OVS_REQ_RDLOCK(rwlock); | |
4a1b8f30 EJ |
171 | static void bond_enable_slave(struct bond_slave *, bool enable) |
172 | OVS_REQ_WRLOCK(rwlock); | |
173 | static void bond_link_status_update(struct bond_slave *) | |
3bfd3972 | 174 | OVS_REQ_WRLOCK(rwlock); |
4a1b8f30 | 175 | static void bond_choose_active_slave(struct bond *) |
9e1a6910 | 176 | OVS_REQ_WRLOCK(rwlock); |
f620b43a BP |
177 | static struct bond_entry *lookup_bond_entry(const struct bond *, |
178 | const struct flow *, | |
3bfd3972 EJ |
179 | uint16_t vlan) |
180 | OVS_REQ_RDLOCK(rwlock); | |
f1c8a79c AW |
181 | static struct bond_slave *get_enabled_slave(struct bond *) |
182 | OVS_REQ_RDLOCK(rwlock); | |
f620b43a BP |
183 | static struct bond_slave *choose_output_slave(const struct bond *, |
184 | const struct flow *, | |
bcd2633a | 185 | struct flow_wildcards *, |
4a1b8f30 | 186 | uint16_t vlan) |
3bfd3972 | 187 | OVS_REQ_RDLOCK(rwlock); |
05df1623 | 188 | static void update_recirc_rules__(struct bond *bond); |
90061ea7 | 189 | static bool bond_is_falling_back_to_ab(const struct bond *); |
f620b43a BP |
190 | |
191 | /* Attempts to parse 's' as the name of a bond balancing mode. If successful, | |
192 | * stores the mode in '*balance' and returns true. Otherwise returns false | |
193 | * without modifying '*balance'. */ | |
194 | bool | |
195 | bond_mode_from_string(enum bond_mode *balance, const char *s) | |
196 | { | |
197 | if (!strcmp(s, bond_mode_to_string(BM_TCP))) { | |
198 | *balance = BM_TCP; | |
199 | } else if (!strcmp(s, bond_mode_to_string(BM_SLB))) { | |
200 | *balance = BM_SLB; | |
201 | } else if (!strcmp(s, bond_mode_to_string(BM_AB))) { | |
202 | *balance = BM_AB; | |
203 | } else { | |
204 | return false; | |
205 | } | |
206 | return true; | |
207 | } | |
208 | ||
209 | /* Returns a string representing 'balance'. */ | |
210 | const char * | |
211 | bond_mode_to_string(enum bond_mode balance) { | |
212 | switch (balance) { | |
213 | case BM_TCP: | |
214 | return "balance-tcp"; | |
215 | case BM_SLB: | |
216 | return "balance-slb"; | |
217 | case BM_AB: | |
218 | return "active-backup"; | |
219 | } | |
428b2edd | 220 | OVS_NOT_REACHED(); |
f620b43a BP |
221 | } |
222 | ||
f620b43a BP |
223 | \f |
224 | /* Creates and returns a new bond whose configuration is initially taken from | |
225 | * 's'. | |
226 | * | |
227 | * The caller should register each slave on the new bond by calling | |
228 | * bond_slave_register(). */ | |
229 | struct bond * | |
adcf00ba | 230 | bond_create(const struct bond_settings *s, struct ofproto_dpif *ofproto) |
f620b43a BP |
231 | { |
232 | struct bond *bond; | |
233 | ||
234 | bond = xzalloc(sizeof *bond); | |
adcf00ba | 235 | bond->ofproto = ofproto; |
f620b43a | 236 | hmap_init(&bond->slaves); |
417e7e66 | 237 | ovs_list_init(&bond->enabled_slaves); |
f1c8a79c | 238 | ovs_mutex_init(&bond->mutex); |
37bec3d3 | 239 | ovs_refcount_init(&bond->ref_cnt); |
adcf00ba AZ |
240 | hmap_init(&bond->pr_rule_ops); |
241 | ||
30353934 | 242 | bond->active_slave_mac = eth_addr_zero; |
243 | bond->active_slave_changed = false; | |
244 | ||
f620b43a | 245 | bond_reconfigure(bond, s); |
f620b43a BP |
246 | return bond; |
247 | } | |
248 | ||
03366a2d EJ |
249 | struct bond * |
250 | bond_ref(const struct bond *bond_) | |
251 | { | |
252 | struct bond *bond = CONST_CAST(struct bond *, bond_); | |
253 | ||
bca0b3b4 | 254 | if (bond) { |
37bec3d3 | 255 | ovs_refcount_ref(&bond->ref_cnt); |
bca0b3b4 | 256 | } |
03366a2d EJ |
257 | return bond; |
258 | } | |
259 | ||
f620b43a BP |
260 | /* Frees 'bond'. */ |
261 | void | |
03366a2d | 262 | bond_unref(struct bond *bond) |
f620b43a | 263 | { |
4ec3d7c7 | 264 | struct bond_slave *slave; |
f620b43a | 265 | |
24f83812 | 266 | if (!bond || ovs_refcount_unref_relaxed(&bond->ref_cnt) != 1) { |
03366a2d EJ |
267 | return; |
268 | } | |
269 | ||
3bfd3972 EJ |
270 | ovs_rwlock_wrlock(&rwlock); |
271 | hmap_remove(all_bonds, &bond->hmap_node); | |
272 | ovs_rwlock_unlock(&rwlock); | |
f620b43a | 273 | |
4ec3d7c7 | 274 | HMAP_FOR_EACH_POP (slave, hmap_node, &bond->slaves) { |
f620b43a BP |
275 | /* Client owns 'slave->netdev'. */ |
276 | free(slave->name); | |
277 | free(slave); | |
278 | } | |
279 | hmap_destroy(&bond->slaves); | |
280 | ||
f1c8a79c | 281 | ovs_mutex_destroy(&bond->mutex); |
adcf00ba | 282 | |
05df1623 | 283 | /* Free bond resources. Remove existing post recirc rules. */ |
adcf00ba | 284 | if (bond->recirc_id) { |
e672ff9b | 285 | recirc_free_id(bond->recirc_id); |
05df1623 | 286 | bond->recirc_id = 0; |
adcf00ba | 287 | } |
05df1623 AZ |
288 | free(bond->hash); |
289 | bond->hash = NULL; | |
290 | update_recirc_rules__(bond); | |
adcf00ba | 291 | |
05df1623 AZ |
292 | hmap_destroy(&bond->pr_rule_ops); |
293 | free(bond->name); | |
f620b43a BP |
294 | free(bond); |
295 | } | |
296 | ||
adcf00ba AZ |
297 | static void |
298 | add_pr_rule(struct bond *bond, const struct match *match, | |
6c932bc8 | 299 | ofp_port_t out_ofport, struct rule **rule) |
adcf00ba AZ |
300 | { |
301 | uint32_t hash = match_hash(match, 0); | |
302 | struct bond_pr_rule_op *pr_op; | |
303 | ||
304 | HMAP_FOR_EACH_WITH_HASH(pr_op, hmap_node, hash, &bond->pr_rule_ops) { | |
305 | if (match_equal(&pr_op->match, match)) { | |
306 | pr_op->op = ADD; | |
307 | pr_op->out_ofport = out_ofport; | |
308 | pr_op->pr_rule = rule; | |
309 | return; | |
310 | } | |
311 | } | |
312 | ||
313 | pr_op = xmalloc(sizeof *pr_op); | |
314 | pr_op->match = *match; | |
315 | pr_op->op = ADD; | |
316 | pr_op->out_ofport = out_ofport; | |
317 | pr_op->pr_rule = rule; | |
318 | hmap_insert(&bond->pr_rule_ops, &pr_op->hmap_node, hash); | |
319 | } | |
320 | ||
05df1623 AZ |
321 | /* This function should almost never be called directly. |
322 | * 'update_recirc_rules()' should be called instead. Since | |
323 | * this function modifies 'bond->pr_rule_ops', it is only | |
324 | * safe when 'rwlock' is held. | |
325 | * | |
326 | * However, when the 'bond' is the only reference in the system, | |
327 | * calling this function avoid acquiring lock only to satisfy | |
328 | * lock annotation. Currently, only 'bond_unref()' calls | |
329 | * this function directly. */ | |
adcf00ba | 330 | static void |
05df1623 | 331 | update_recirc_rules__(struct bond *bond) |
adcf00ba AZ |
332 | { |
333 | struct match match; | |
334 | struct bond_pr_rule_op *pr_op, *next_op; | |
335 | uint64_t ofpacts_stub[128 / 8]; | |
336 | struct ofpbuf ofpacts; | |
337 | int i; | |
338 | ||
339 | ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); | |
340 | ||
341 | HMAP_FOR_EACH(pr_op, hmap_node, &bond->pr_rule_ops) { | |
342 | pr_op->op = DEL; | |
343 | } | |
344 | ||
6c932bc8 AZ |
345 | if (bond->hash && bond->recirc_id) { |
346 | for (i = 0; i < BOND_BUCKETS; i++) { | |
347 | struct bond_slave *slave = bond->hash[i].slave; | |
adcf00ba | 348 | |
6c932bc8 AZ |
349 | if (slave) { |
350 | match_init_catchall(&match); | |
351 | match_set_recirc_id(&match, bond->recirc_id); | |
6c932bc8 | 352 | match_set_dp_hash_masked(&match, i, BOND_MASK); |
adcf00ba | 353 | |
6c932bc8 AZ |
354 | add_pr_rule(bond, &match, slave->ofp_port, |
355 | &bond->hash[i].pr_rule); | |
356 | } | |
adcf00ba AZ |
357 | } |
358 | } | |
359 | ||
360 | HMAP_FOR_EACH_SAFE(pr_op, next_op, hmap_node, &bond->pr_rule_ops) { | |
361 | int error; | |
adcf00ba AZ |
362 | switch (pr_op->op) { |
363 | case ADD: | |
364 | ofpbuf_clear(&ofpacts); | |
365 | ofpact_put_OUTPUT(&ofpacts)->port = pr_op->out_ofport; | |
366 | error = ofproto_dpif_add_internal_flow(bond->ofproto, | |
367 | &pr_op->match, | |
290ad78a | 368 | RECIRC_RULE_PRIORITY, 0, |
6c932bc8 | 369 | &ofpacts, pr_op->pr_rule); |
adcf00ba | 370 | if (error) { |
50f96b10 | 371 | char *err_s = match_to_string(&pr_op->match, NULL, |
adcf00ba AZ |
372 | RECIRC_RULE_PRIORITY); |
373 | ||
374 | VLOG_ERR("failed to add post recirculation flow %s", err_s); | |
375 | free(err_s); | |
adcf00ba AZ |
376 | } |
377 | break; | |
378 | ||
379 | case DEL: | |
380 | error = ofproto_dpif_delete_internal_flow(bond->ofproto, | |
381 | &pr_op->match, | |
382 | RECIRC_RULE_PRIORITY); | |
383 | if (error) { | |
50f96b10 | 384 | char *err_s = match_to_string(&pr_op->match, NULL, |
adcf00ba AZ |
385 | RECIRC_RULE_PRIORITY); |
386 | ||
387 | VLOG_ERR("failed to remove post recirculation flow %s", err_s); | |
388 | free(err_s); | |
389 | } | |
390 | ||
391 | hmap_remove(&bond->pr_rule_ops, &pr_op->hmap_node); | |
fa233667 YS |
392 | if (bond->hash) { |
393 | *pr_op->pr_rule = NULL; | |
394 | } | |
adcf00ba AZ |
395 | free(pr_op); |
396 | break; | |
397 | } | |
398 | } | |
399 | ||
400 | ofpbuf_uninit(&ofpacts); | |
401 | } | |
402 | ||
05df1623 AZ |
403 | static void |
404 | update_recirc_rules(struct bond *bond) | |
405 | OVS_REQ_RDLOCK(rwlock) | |
406 | { | |
407 | update_recirc_rules__(bond); | |
408 | } | |
adcf00ba | 409 | |
f620b43a BP |
410 | /* Updates 'bond''s overall configuration to 's'. |
411 | * | |
412 | * The caller should register each slave on 'bond' by calling | |
413 | * bond_slave_register(). This is optional if none of the slaves' | |
4d6fb5eb | 414 | * configuration has changed. In any case it can't hurt. |
59d7b2b6 EJ |
415 | * |
416 | * Returns true if the configuration has changed in such a way that requires | |
417 | * flow revalidation. | |
418 | * */ | |
419 | bool | |
f620b43a BP |
420 | bond_reconfigure(struct bond *bond, const struct bond_settings *s) |
421 | { | |
59d7b2b6 EJ |
422 | bool revalidate = false; |
423 | ||
3bfd3972 | 424 | ovs_rwlock_wrlock(&rwlock); |
f620b43a BP |
425 | if (!bond->name || strcmp(bond->name, s->name)) { |
426 | if (bond->name) { | |
3bfd3972 | 427 | hmap_remove(all_bonds, &bond->hmap_node); |
f620b43a BP |
428 | free(bond->name); |
429 | } | |
430 | bond->name = xstrdup(s->name); | |
3bfd3972 | 431 | hmap_insert(all_bonds, &bond->hmap_node, hash_string(bond->name, 0)); |
f620b43a BP |
432 | } |
433 | ||
f620b43a BP |
434 | bond->updelay = s->up_delay; |
435 | bond->downdelay = s->down_delay; | |
bc1b010c | 436 | |
9dd165e0 RK |
437 | if (bond->lacp_fallback_ab != s->lacp_fallback_ab_cfg) { |
438 | bond->lacp_fallback_ab = s->lacp_fallback_ab_cfg; | |
439 | revalidate = true; | |
440 | } | |
441 | ||
bc1b010c EJ |
442 | if (bond->rebalance_interval != s->rebalance_interval) { |
443 | bond->rebalance_interval = s->rebalance_interval; | |
444 | revalidate = true; | |
445 | } | |
f620b43a | 446 | |
59d7b2b6 EJ |
447 | if (bond->balance != s->balance) { |
448 | bond->balance = s->balance; | |
449 | revalidate = true; | |
450 | } | |
451 | ||
672d18b2 EJ |
452 | if (bond->basis != s->basis) { |
453 | bond->basis = s->basis; | |
454 | revalidate = true; | |
455 | } | |
456 | ||
62904702 EJ |
457 | if (bond->bond_revalidate) { |
458 | revalidate = true; | |
459 | bond->bond_revalidate = false; | |
460 | } | |
461 | ||
adcf00ba AZ |
462 | if (bond->balance != BM_AB) { |
463 | if (!bond->recirc_id) { | |
e672ff9b | 464 | bond->recirc_id = recirc_alloc_id(bond->ofproto); |
adcf00ba AZ |
465 | } |
466 | } else if (bond->recirc_id) { | |
e672ff9b | 467 | recirc_free_id(bond->recirc_id); |
adcf00ba AZ |
468 | bond->recirc_id = 0; |
469 | } | |
470 | ||
95aafb2a EJ |
471 | if (bond->balance == BM_AB || !bond->hash || revalidate) { |
472 | bond_entry_reset(bond); | |
473 | } | |
474 | ||
3bfd3972 | 475 | ovs_rwlock_unlock(&rwlock); |
59d7b2b6 | 476 | return revalidate; |
f620b43a BP |
477 | } |
478 | ||
3e5aeeb5 | 479 | static struct bond_slave * |
74ff3298 | 480 | bond_find_slave_by_mac(const struct bond *bond, const struct eth_addr mac) |
3e5aeeb5 AZ |
481 | { |
482 | struct bond_slave *slave; | |
483 | ||
484 | /* Find the last active slave */ | |
485 | HMAP_FOR_EACH(slave, hmap_node, &bond->slaves) { | |
74ff3298 | 486 | struct eth_addr slave_mac; |
3e5aeeb5 | 487 | |
74ff3298 | 488 | if (netdev_get_etheraddr(slave->netdev, &slave_mac)) { |
3e5aeeb5 AZ |
489 | continue; |
490 | } | |
491 | ||
74ff3298 | 492 | if (eth_addr_equals(slave_mac, mac)) { |
3e5aeeb5 AZ |
493 | return slave; |
494 | } | |
495 | } | |
496 | ||
497 | return NULL; | |
498 | } | |
499 | ||
500 | static void | |
501 | bond_active_slave_changed(struct bond *bond) | |
502 | { | |
f626af7a AZ |
503 | if (bond->active_slave) { |
504 | struct eth_addr mac; | |
505 | netdev_get_etheraddr(bond->active_slave->netdev, &mac); | |
506 | bond->active_slave_mac = mac; | |
507 | } else { | |
508 | bond->active_slave_mac = eth_addr_zero; | |
509 | } | |
3e5aeeb5 AZ |
510 | bond->active_slave_changed = true; |
511 | seq_change(connectivity_seq_get()); | |
512 | } | |
513 | ||
f8ddccd2 | 514 | static void |
1ea24138 | 515 | bond_slave_set_netdev__(struct bond_slave *slave, struct netdev *netdev) |
3bfd3972 | 516 | OVS_REQ_WRLOCK(rwlock) |
f8ddccd2 BP |
517 | { |
518 | if (slave->netdev != netdev) { | |
f8ddccd2 | 519 | slave->netdev = netdev; |
1ea24138 | 520 | slave->change_seq = 0; |
f8ddccd2 BP |
521 | } |
522 | } | |
523 | ||
f620b43a BP |
524 | /* Registers 'slave_' as a slave of 'bond'. The 'slave_' pointer is an |
525 | * arbitrary client-provided pointer that uniquely identifies a slave within a | |
526 | * bond. If 'slave_' already exists within 'bond' then this function | |
527 | * reconfigures the existing slave. | |
528 | * | |
529 | * 'netdev' must be the network device that 'slave_' represents. It is owned | |
530 | * by the client, so the client must not close it before either unregistering | |
531 | * 'slave_' or destroying 'bond'. | |
4d6fb5eb | 532 | */ |
f620b43a | 533 | void |
adcf00ba AZ |
534 | bond_slave_register(struct bond *bond, void *slave_, |
535 | ofp_port_t ofport, struct netdev *netdev) | |
f620b43a | 536 | { |
3bfd3972 | 537 | struct bond_slave *slave; |
f620b43a | 538 | |
3bfd3972 EJ |
539 | ovs_rwlock_wrlock(&rwlock); |
540 | slave = bond_slave_lookup(bond, slave_); | |
f620b43a BP |
541 | if (!slave) { |
542 | slave = xzalloc(sizeof *slave); | |
543 | ||
544 | hmap_insert(&bond->slaves, &slave->hmap_node, hash_pointer(slave_, 0)); | |
545 | slave->bond = bond; | |
546 | slave->aux = slave_; | |
adcf00ba | 547 | slave->ofp_port = ofport; |
f620b43a | 548 | slave->delay_expires = LLONG_MAX; |
244b2160 | 549 | slave->name = xstrdup(netdev_get_name(netdev)); |
7321e30e | 550 | bond->bond_revalidate = true; |
244b2160 | 551 | |
b3c18f66 | 552 | slave->enabled = false; |
4a1b8f30 | 553 | bond_enable_slave(slave, netdev_get_carrier(netdev)); |
f620b43a BP |
554 | } |
555 | ||
1ea24138 | 556 | bond_slave_set_netdev__(slave, netdev); |
a6934aa9 | 557 | |
f620b43a BP |
558 | free(slave->name); |
559 | slave->name = xstrdup(netdev_get_name(netdev)); | |
3bfd3972 | 560 | ovs_rwlock_unlock(&rwlock); |
f620b43a BP |
561 | } |
562 | ||
f8ddccd2 BP |
563 | /* Updates the network device to be used with 'slave_' to 'netdev'. |
564 | * | |
565 | * This is useful if the caller closes and re-opens the network device | |
566 | * registered with bond_slave_register() but doesn't need to change anything | |
567 | * else. */ | |
568 | void | |
569 | bond_slave_set_netdev(struct bond *bond, void *slave_, struct netdev *netdev) | |
570 | { | |
3bfd3972 EJ |
571 | struct bond_slave *slave; |
572 | ||
573 | ovs_rwlock_wrlock(&rwlock); | |
574 | slave = bond_slave_lookup(bond, slave_); | |
f8ddccd2 | 575 | if (slave) { |
1ea24138 | 576 | bond_slave_set_netdev__(slave, netdev); |
f8ddccd2 | 577 | } |
3bfd3972 | 578 | ovs_rwlock_unlock(&rwlock); |
f8ddccd2 BP |
579 | } |
580 | ||
f620b43a BP |
581 | /* Unregisters 'slave_' from 'bond'. If 'bond' does not contain such a slave |
582 | * then this function has no effect. | |
583 | * | |
584 | * Unregistering a slave invalidates all flows. */ | |
585 | void | |
586 | bond_slave_unregister(struct bond *bond, const void *slave_) | |
587 | { | |
3bfd3972 | 588 | struct bond_slave *slave; |
f620b43a BP |
589 | bool del_active; |
590 | ||
3bfd3972 EJ |
591 | ovs_rwlock_wrlock(&rwlock); |
592 | slave = bond_slave_lookup(bond, slave_); | |
f620b43a | 593 | if (!slave) { |
3bfd3972 | 594 | goto out; |
f620b43a BP |
595 | } |
596 | ||
4a1b8f30 EJ |
597 | bond->bond_revalidate = true; |
598 | bond_enable_slave(slave, false); | |
b3c18f66 | 599 | |
f620b43a BP |
600 | del_active = bond->active_slave == slave; |
601 | if (bond->hash) { | |
602 | struct bond_entry *e; | |
603 | for (e = bond->hash; e <= &bond->hash[BOND_MASK]; e++) { | |
604 | if (e->slave == slave) { | |
605 | e->slave = NULL; | |
606 | } | |
607 | } | |
608 | } | |
609 | ||
610 | free(slave->name); | |
611 | ||
612 | hmap_remove(&bond->slaves, &slave->hmap_node); | |
613 | /* Client owns 'slave->netdev'. */ | |
614 | free(slave); | |
615 | ||
616 | if (del_active) { | |
4a1b8f30 | 617 | bond_choose_active_slave(bond); |
f620b43a BP |
618 | bond->send_learning_packets = true; |
619 | } | |
3bfd3972 EJ |
620 | out: |
621 | ovs_rwlock_unlock(&rwlock); | |
f620b43a BP |
622 | } |
623 | ||
296f6519 EJ |
624 | /* Should be called on each slave in 'bond' before bond_run() to indicate |
625 | * whether or not 'slave_' may be enabled. This function is intended to allow | |
626 | * other protocols to have some impact on bonding decisions. For example LACP | |
627 | * or high level link monitoring protocols may decide that a given slave should | |
628 | * not be able to send traffic. */ | |
4d6fb5eb | 629 | void |
296f6519 | 630 | bond_slave_set_may_enable(struct bond *bond, void *slave_, bool may_enable) |
4d6fb5eb | 631 | { |
3bfd3972 | 632 | ovs_rwlock_wrlock(&rwlock); |
296f6519 | 633 | bond_slave_lookup(bond, slave_)->may_enable = may_enable; |
3bfd3972 | 634 | ovs_rwlock_unlock(&rwlock); |
4d6fb5eb EJ |
635 | } |
636 | ||
4a1b8f30 EJ |
637 | /* Performs periodic maintenance on 'bond'. |
638 | * | |
639 | * Returns true if the caller should revalidate its flows. | |
f620b43a BP |
640 | * |
641 | * The caller should check bond_should_send_learning_packets() afterward. */ | |
4a1b8f30 EJ |
642 | bool |
643 | bond_run(struct bond *bond, enum lacp_status lacp_status) | |
f620b43a BP |
644 | { |
645 | struct bond_slave *slave; | |
4a1b8f30 | 646 | bool revalidate; |
f620b43a | 647 | |
3bfd3972 | 648 | ovs_rwlock_wrlock(&rwlock); |
bdebeece EJ |
649 | if (bond->lacp_status != lacp_status) { |
650 | bond->lacp_status = lacp_status; | |
4592d0e2 | 651 | bond->bond_revalidate = true; |
90061ea7 BP |
652 | |
653 | /* Change in LACP status can affect whether the bond is falling back to | |
654 | * active-backup. Make sure to create or destroy buckets if | |
655 | * necessary. */ | |
656 | if (bond_is_falling_back_to_ab(bond) || !bond->hash) { | |
657 | bond_entry_reset(bond); | |
658 | } | |
4592d0e2 | 659 | } |
4d6fb5eb | 660 | |
f620b43a BP |
661 | /* Enable slaves based on link status and LACP feedback. */ |
662 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
4a1b8f30 | 663 | bond_link_status_update(slave); |
da4a6191 | 664 | slave->change_seq = seq_read(connectivity_seq_get()); |
f620b43a BP |
665 | } |
666 | if (!bond->active_slave || !bond->active_slave->enabled) { | |
4a1b8f30 | 667 | bond_choose_active_slave(bond); |
f620b43a BP |
668 | } |
669 | ||
4a1b8f30 EJ |
670 | revalidate = bond->bond_revalidate; |
671 | bond->bond_revalidate = false; | |
3bfd3972 | 672 | ovs_rwlock_unlock(&rwlock); |
4a1b8f30 EJ |
673 | |
674 | return revalidate; | |
f620b43a BP |
675 | } |
676 | ||
677 | /* Causes poll_block() to wake up when 'bond' needs something to be done. */ | |
678 | void | |
679 | bond_wait(struct bond *bond) | |
680 | { | |
681 | struct bond_slave *slave; | |
682 | ||
3bfd3972 | 683 | ovs_rwlock_rdlock(&rwlock); |
f620b43a BP |
684 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { |
685 | if (slave->delay_expires != LLONG_MAX) { | |
686 | poll_timer_wait_until(slave->delay_expires); | |
687 | } | |
1ea24138 | 688 | |
da4a6191 | 689 | seq_wait(connectivity_seq_get(), slave->change_seq); |
f620b43a BP |
690 | } |
691 | ||
bbc13389 | 692 | if (bond->bond_revalidate) { |
f620b43a BP |
693 | poll_immediate_wake(); |
694 | } | |
3bfd3972 | 695 | ovs_rwlock_unlock(&rwlock); |
f620b43a BP |
696 | |
697 | /* We don't wait for bond->next_rebalance because rebalancing can only run | |
698 | * at a flow account checkpoint. ofproto does checkpointing on its own | |
699 | * schedule and bond_rebalance() gets called afterward, so we'd just be | |
700 | * waking up for no purpose. */ | |
701 | } | |
702 | \f | |
703 | /* MAC learning table interaction. */ | |
704 | ||
705 | static bool | |
706 | may_send_learning_packets(const struct bond *bond) | |
707 | { | |
9dd165e0 RK |
708 | return ((bond->lacp_status == LACP_DISABLED |
709 | && (bond->balance == BM_SLB || bond->balance == BM_AB)) | |
710 | || (bond->lacp_fallback_ab && bond->lacp_status == LACP_CONFIGURED)) | |
bdebeece | 711 | && bond->active_slave; |
f620b43a BP |
712 | } |
713 | ||
714 | /* Returns true if 'bond' needs the client to send out packets to assist with | |
715 | * MAC learning on 'bond'. If this function returns true, then the client | |
716 | * should iterate through its MAC learning table for the bridge on which 'bond' | |
717 | * is located. For each MAC that has been learned on a port other than 'bond', | |
ea131871 | 718 | * it should call bond_compose_learning_packet(). |
f620b43a | 719 | * |
477879ea BP |
720 | * This function will only return true if 'bond' is in SLB or active-backup |
721 | * mode and LACP is not negotiated. Otherwise sending learning packets isn't | |
722 | * necessary. | |
f620b43a BP |
723 | * |
724 | * Calling this function resets the state that it checks. */ | |
725 | bool | |
726 | bond_should_send_learning_packets(struct bond *bond) | |
727 | { | |
3bfd3972 EJ |
728 | bool send; |
729 | ||
730 | ovs_rwlock_wrlock(&rwlock); | |
731 | send = bond->send_learning_packets && may_send_learning_packets(bond); | |
f620b43a | 732 | bond->send_learning_packets = false; |
3bfd3972 | 733 | ovs_rwlock_unlock(&rwlock); |
f620b43a BP |
734 | return send; |
735 | } | |
736 | ||
737 | /* Sends a gratuitous learning packet on 'bond' from 'eth_src' on 'vlan'. | |
738 | * | |
ea131871 JG |
739 | * See bond_should_send_learning_packets() for description of usage. The |
740 | * caller should send the composed packet on the port associated with | |
741 | * port_aux and takes ownership of the returned ofpbuf. */ | |
cf62fa4c | 742 | struct dp_packet * |
74ff3298 | 743 | bond_compose_learning_packet(struct bond *bond, const struct eth_addr eth_src, |
ea131871 | 744 | uint16_t vlan, void **port_aux) |
f620b43a BP |
745 | { |
746 | struct bond_slave *slave; | |
cf62fa4c | 747 | struct dp_packet *packet; |
f620b43a | 748 | struct flow flow; |
f620b43a | 749 | |
3bfd3972 | 750 | ovs_rwlock_rdlock(&rwlock); |
cb22974d | 751 | ovs_assert(may_send_learning_packets(bond)); |
f620b43a | 752 | memset(&flow, 0, sizeof flow); |
74ff3298 | 753 | flow.dl_src = eth_src; |
4a1b8f30 | 754 | slave = choose_output_slave(bond, &flow, NULL, vlan); |
f620b43a | 755 | |
cf62fa4c | 756 | packet = dp_packet_new(0); |
2ea838ac | 757 | compose_rarp(packet, eth_src); |
f620b43a | 758 | if (vlan) { |
1bf02876 | 759 | eth_push_vlan(packet, htons(ETH_TYPE_VLAN), htons(vlan)); |
f620b43a | 760 | } |
f620b43a | 761 | |
ea131871 | 762 | *port_aux = slave->aux; |
3bfd3972 | 763 | ovs_rwlock_unlock(&rwlock); |
ea131871 | 764 | return packet; |
f620b43a BP |
765 | } |
766 | \f | |
90061ea7 BP |
767 | |
768 | static bool | |
769 | bond_is_falling_back_to_ab(const struct bond *bond) | |
770 | { | |
771 | return (bond->lacp_fallback_ab | |
772 | && (bond->balance == BM_SLB || bond->balance == BM_TCP) | |
773 | && bond->lacp_status == LACP_CONFIGURED); | |
774 | } | |
775 | ||
f620b43a BP |
776 | /* Checks whether a packet that arrived on 'slave_' within 'bond', with an |
777 | * Ethernet destination address of 'eth_dst', should be admitted. | |
778 | * | |
779 | * The return value is one of the following: | |
780 | * | |
781 | * - BV_ACCEPT: Admit the packet. | |
782 | * | |
783 | * - BV_DROP: Drop the packet. | |
784 | * | |
785 | * - BV_DROP_IF_MOVED: Consult the MAC learning table for the packet's | |
786 | * Ethernet source address and VLAN. If there is none, or if the packet | |
787 | * is on the learned port, then admit the packet. If a different port has | |
788 | * been learned, however, drop the packet (and do not use it for MAC | |
789 | * learning). | |
790 | */ | |
791 | enum bond_verdict | |
792 | bond_check_admissibility(struct bond *bond, const void *slave_, | |
74ff3298 | 793 | const struct eth_addr eth_dst) |
f620b43a | 794 | { |
3bfd3972 EJ |
795 | enum bond_verdict verdict = BV_DROP; |
796 | struct bond_slave *slave; | |
a8448cb1 | 797 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
9a1c6450 | 798 | |
3bfd3972 EJ |
799 | ovs_rwlock_rdlock(&rwlock); |
800 | slave = bond_slave_lookup(bond, slave_); | |
4222bbc8 | 801 | if (!slave) { |
3bfd3972 | 802 | goto out; |
4222bbc8 EJ |
803 | } |
804 | ||
9a1c6450 EJ |
805 | /* LACP bonds have very loose admissibility restrictions because we can |
806 | * assume the remote switch is aware of the bond and will "do the right | |
807 | * thing". However, as a precaution we drop packets on disabled slaves | |
808 | * because no correctly implemented partner switch should be sending | |
bdebeece EJ |
809 | * packets to them. |
810 | * | |
811 | * If LACP is configured, but LACP negotiations have been unsuccessful, we | |
9dd165e0 | 812 | * drop all incoming traffic except if lacp_fallback_ab is enabled. */ |
bdebeece | 813 | switch (bond->lacp_status) { |
3bfd3972 | 814 | case LACP_NEGOTIATED: |
a8448cb1 NK |
815 | /* To reduce packet-drops due to delay in enabling of slave (post |
816 | * LACP-SYNC), from main thread, check for may_enable as well. | |
817 | * When may_enable is TRUE, it means LACP is UP and waiting for the | |
818 | * main thread to run LACP state machine and enable the slave. */ | |
819 | verdict = (slave->enabled || slave->may_enable) ? BV_ACCEPT : BV_DROP; | |
3bfd3972 EJ |
820 | goto out; |
821 | case LACP_CONFIGURED: | |
9dd165e0 RK |
822 | if (!bond->lacp_fallback_ab) { |
823 | goto out; | |
824 | } | |
e5c4f827 | 825 | break; |
3bfd3972 | 826 | case LACP_DISABLED: |
e5c4f827 | 827 | if (bond->balance == BM_TCP) { |
828 | goto out; | |
829 | } | |
3bfd3972 | 830 | break; |
f620b43a BP |
831 | } |
832 | ||
833 | /* Drop all multicast packets on inactive slaves. */ | |
834 | if (eth_addr_is_multicast(eth_dst)) { | |
4222bbc8 | 835 | if (bond->active_slave != slave) { |
3bfd3972 | 836 | goto out; |
f620b43a BP |
837 | } |
838 | } | |
839 | ||
f931a4c9 | 840 | switch (bond->balance) { |
9dd165e0 RK |
841 | case BM_TCP: |
842 | /* TCP balanced bonds require successful LACP negotiations. Based on the | |
843 | * above check, LACP is off or lacp_fallback_ab is true on this bond. | |
844 | * If lacp_fallback_ab is true fall through to BM_AB case else, we | |
845 | * drop all incoming traffic. */ | |
846 | if (!bond->lacp_fallback_ab) { | |
847 | goto out; | |
848 | } | |
73c7216a | 849 | /* fall through */ |
9dd165e0 | 850 | |
f931a4c9 BP |
851 | case BM_AB: |
852 | /* Drop all packets which arrive on backup slaves. This is similar to | |
853 | * how Linux bonding handles active-backup bonds. */ | |
7ba7dcf0 | 854 | if (bond->active_slave != slave) { |
e6b2255c BP |
855 | VLOG_DBG_RL(&rl, "active-backup bond received packet on backup" |
856 | " slave (%s) destined for " ETH_ADDR_FMT, | |
857 | slave->name, ETH_ADDR_ARGS(eth_dst)); | |
3bfd3972 | 858 | goto out; |
7ba7dcf0 | 859 | } |
3bfd3972 EJ |
860 | verdict = BV_ACCEPT; |
861 | goto out; | |
f931a4c9 | 862 | |
f931a4c9 BP |
863 | case BM_SLB: |
864 | /* Drop all packets for which we have learned a different input port, | |
865 | * because we probably sent the packet on one slave and got it back on | |
866 | * the other. Gratuitous ARP packets are an exception to this rule: | |
867 | * the host has moved to another switch. The exception to the | |
868 | * exception is if we locked the learning table to avoid reflections on | |
869 | * bond slaves. */ | |
3bfd3972 EJ |
870 | verdict = BV_DROP_IF_MOVED; |
871 | goto out; | |
7ba7dcf0 EJ |
872 | } |
873 | ||
428b2edd | 874 | OVS_NOT_REACHED(); |
3bfd3972 | 875 | out: |
a8448cb1 NK |
876 | if (slave && (verdict != BV_ACCEPT)) { |
877 | VLOG_DBG_RL(&rl, "slave (%s): Admissibility verdict is to drop pkt %s." | |
878 | "active slave: %s, may_enable: %s enable: %s " | |
879 | "LACP status:%d", | |
880 | slave->name, | |
881 | (verdict == BV_DROP_IF_MOVED) ? | |
882 | "as different port is learned" : "", | |
883 | (bond->active_slave == slave) ? "true" : "false", | |
884 | slave->may_enable ? "true" : "false", | |
885 | slave->enabled ? "true" : "false", | |
886 | bond->lacp_status); | |
887 | } | |
888 | ||
3bfd3972 EJ |
889 | ovs_rwlock_unlock(&rwlock); |
890 | return verdict; | |
891 | ||
f620b43a BP |
892 | } |
893 | ||
894 | /* Returns the slave (registered on 'bond' by bond_slave_register()) to which | |
895 | * a packet with the given 'flow' and 'vlan' should be forwarded. Returns | |
896 | * NULL if the packet should be dropped because no slaves are enabled. | |
897 | * | |
898 | * 'vlan' is not necessarily the same as 'flow->vlan_tci'. First, 'vlan' | |
899 | * should be a VID only (i.e. excluding the PCP bits). Second, | |
900 | * 'flow->vlan_tci' is the VLAN TCI that appeared on the packet (so it will be | |
901 | * nonzero only for trunk ports), whereas 'vlan' is the logical VLAN that the | |
902 | * packet belongs to (so for an access port it will be the access port's VLAN). | |
903 | * | |
bcd2633a JP |
904 | * If 'wc' is non-NULL, bitwise-OR's 'wc' with the set of bits that were |
905 | * significant in the selection. At some point earlier, 'wc' should | |
906 | * have been initialized (e.g., by flow_wildcards_init_catchall()). | |
f620b43a BP |
907 | */ |
908 | void * | |
909 | bond_choose_output_slave(struct bond *bond, const struct flow *flow, | |
4a1b8f30 | 910 | struct flow_wildcards *wc, uint16_t vlan) |
f620b43a | 911 | { |
3bfd3972 | 912 | struct bond_slave *slave; |
b5d5d7d3 | 913 | void *aux; |
3bfd3972 EJ |
914 | |
915 | ovs_rwlock_rdlock(&rwlock); | |
4a1b8f30 | 916 | slave = choose_output_slave(bond, flow, wc, vlan); |
b5d5d7d3 | 917 | aux = slave ? slave->aux : NULL; |
3bfd3972 | 918 | ovs_rwlock_unlock(&rwlock); |
b5d5d7d3 AW |
919 | |
920 | return aux; | |
f620b43a | 921 | } |
f620b43a | 922 | \f |
adcf00ba AZ |
923 | /* Recirculation. */ |
924 | static void | |
925 | bond_entry_account(struct bond_entry *entry, uint64_t rule_tx_bytes) | |
c6855ec5 | 926 | OVS_REQ_WRLOCK(rwlock) |
adcf00ba AZ |
927 | { |
928 | if (entry->slave) { | |
929 | uint64_t delta; | |
930 | ||
931 | delta = rule_tx_bytes - entry->pr_tx_bytes; | |
932 | entry->tx_bytes += delta; | |
933 | entry->pr_tx_bytes = rule_tx_bytes; | |
934 | } | |
935 | } | |
936 | ||
937 | /* Maintain bond stats using post recirculation rule byte counters.*/ | |
60cda7d6 | 938 | static void |
adcf00ba | 939 | bond_recirculation_account(struct bond *bond) |
80316557 | 940 | OVS_REQ_WRLOCK(rwlock) |
adcf00ba AZ |
941 | { |
942 | int i; | |
943 | ||
adcf00ba AZ |
944 | for (i=0; i<=BOND_MASK; i++) { |
945 | struct bond_entry *entry = &bond->hash[i]; | |
946 | struct rule *rule = entry->pr_rule; | |
947 | ||
948 | if (rule) { | |
949 | uint64_t n_packets OVS_UNUSED; | |
950 | long long int used OVS_UNUSED; | |
951 | uint64_t n_bytes; | |
952 | ||
953 | rule->ofproto->ofproto_class->rule_get_stats( | |
954 | rule, &n_packets, &n_bytes, &used); | |
955 | bond_entry_account(entry, n_bytes); | |
956 | } | |
957 | } | |
adcf00ba AZ |
958 | } |
959 | ||
a80aba3a | 960 | static bool |
6b95d23c | 961 | bond_may_recirc(const struct bond *bond) |
adcf00ba | 962 | { |
90061ea7 BP |
963 | return (bond->balance == BM_TCP && bond->recirc_id |
964 | && !bond_is_falling_back_to_ab(bond)); | |
adcf00ba AZ |
965 | } |
966 | ||
ca8127fd AZ |
967 | static void |
968 | bond_update_post_recirc_rules__(struct bond* bond, const bool force) | |
969 | OVS_REQ_WRLOCK(rwlock) | |
adcf00ba AZ |
970 | { |
971 | struct bond_entry *e; | |
972 | bool update_rules = force; /* Always update rules if caller forces it. */ | |
973 | ||
974 | /* Make sure all bond entries are populated */ | |
975 | for (e = bond->hash; e <= &bond->hash[BOND_MASK]; e++) { | |
976 | if (!e->slave || !e->slave->enabled) { | |
977 | update_rules = true; | |
978 | e->slave = CONTAINER_OF(hmap_random_node(&bond->slaves), | |
979 | struct bond_slave, hmap_node); | |
980 | if (!e->slave->enabled) { | |
981 | e->slave = bond->active_slave; | |
982 | } | |
983 | } | |
984 | } | |
985 | ||
986 | if (update_rules) { | |
987 | update_recirc_rules(bond); | |
988 | } | |
989 | } | |
ca8127fd AZ |
990 | |
991 | void | |
82f9f1f5 AZ |
992 | bond_update_post_recirc_rules(struct bond *bond, uint32_t *recirc_id, |
993 | uint32_t *hash_basis) | |
ca8127fd | 994 | { |
a80aba3a AZ |
995 | bool may_recirc = bond_may_recirc(bond); |
996 | ||
997 | if (may_recirc) { | |
998 | /* To avoid unnecessary locking, bond_may_recirc() is first | |
999 | * called outside of the 'rwlock'. After acquiring the lock, | |
1000 | * check again to make sure bond configuration has not been changed. */ | |
1001 | ovs_rwlock_wrlock(&rwlock); | |
1002 | may_recirc = bond_may_recirc(bond); | |
1003 | if (may_recirc) { | |
1004 | *recirc_id = bond->recirc_id; | |
1005 | *hash_basis = bond->basis; | |
1006 | bond_update_post_recirc_rules__(bond, false); | |
1007 | } | |
1008 | ovs_rwlock_unlock(&rwlock); | |
1009 | } | |
1010 | ||
1011 | if (!may_recirc) { | |
6b95d23c | 1012 | *recirc_id = *hash_basis = 0; |
82f9f1f5 | 1013 | } |
ca8127fd | 1014 | } |
82f9f1f5 | 1015 | |
adcf00ba | 1016 | \f |
f620b43a BP |
1017 | /* Rebalancing. */ |
1018 | ||
1b137691 | 1019 | static bool |
3bfd3972 | 1020 | bond_is_balanced(const struct bond *bond) OVS_REQ_RDLOCK(rwlock) |
1b137691 | 1021 | { |
bc1b010c | 1022 | return bond->rebalance_interval |
90061ea7 BP |
1023 | && (bond->balance == BM_SLB || bond->balance == BM_TCP) |
1024 | && !(bond->lacp_fallback_ab && bond->lacp_status == LACP_CONFIGURED); | |
1b137691 EJ |
1025 | } |
1026 | ||
f620b43a BP |
1027 | /* Notifies 'bond' that 'n_bytes' bytes were sent in 'flow' within 'vlan'. */ |
1028 | void | |
1029 | bond_account(struct bond *bond, const struct flow *flow, uint16_t vlan, | |
1030 | uint64_t n_bytes) | |
1031 | { | |
3bfd3972 | 1032 | ovs_rwlock_wrlock(&rwlock); |
1b137691 | 1033 | if (bond_is_balanced(bond)) { |
f620b43a | 1034 | lookup_bond_entry(bond, flow, vlan)->tx_bytes += n_bytes; |
f620b43a | 1035 | } |
3bfd3972 | 1036 | ovs_rwlock_unlock(&rwlock); |
f620b43a BP |
1037 | } |
1038 | ||
1039 | static struct bond_slave * | |
ca6ba700 | 1040 | bond_slave_from_bal_node(struct ovs_list *bal) OVS_REQ_RDLOCK(rwlock) |
f620b43a BP |
1041 | { |
1042 | return CONTAINER_OF(bal, struct bond_slave, bal_node); | |
1043 | } | |
1044 | ||
1045 | static void | |
ca6ba700 | 1046 | log_bals(struct bond *bond, const struct ovs_list *bals) |
c6855ec5 | 1047 | OVS_REQ_RDLOCK(rwlock) |
f620b43a BP |
1048 | { |
1049 | if (VLOG_IS_DBG_ENABLED()) { | |
1050 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1051 | const struct bond_slave *slave; | |
1052 | ||
1053 | LIST_FOR_EACH (slave, bal_node, bals) { | |
1054 | if (ds.length) { | |
1055 | ds_put_char(&ds, ','); | |
1056 | } | |
1057 | ds_put_format(&ds, " %s %"PRIu64"kB", | |
1058 | slave->name, slave->tx_bytes / 1024); | |
1059 | ||
1060 | if (!slave->enabled) { | |
1061 | ds_put_cstr(&ds, " (disabled)"); | |
1062 | } | |
417e7e66 | 1063 | if (!ovs_list_is_empty(&slave->entries)) { |
f620b43a BP |
1064 | struct bond_entry *e; |
1065 | ||
1066 | ds_put_cstr(&ds, " ("); | |
1067 | LIST_FOR_EACH (e, list_node, &slave->entries) { | |
417e7e66 | 1068 | if (&e->list_node != ovs_list_front(&slave->entries)) { |
f620b43a BP |
1069 | ds_put_cstr(&ds, " + "); |
1070 | } | |
34582733 | 1071 | ds_put_format(&ds, "h%"PRIdPTR": %"PRIu64"kB", |
f620b43a BP |
1072 | e - bond->hash, e->tx_bytes / 1024); |
1073 | } | |
1074 | ds_put_cstr(&ds, ")"); | |
1075 | } | |
1076 | } | |
1077 | VLOG_DBG("bond %s:%s", bond->name, ds_cstr(&ds)); | |
1078 | ds_destroy(&ds); | |
1079 | } | |
1080 | } | |
1081 | ||
1082 | /* Shifts 'hash' from its current slave to 'to'. */ | |
1083 | static void | |
4a1b8f30 | 1084 | bond_shift_load(struct bond_entry *hash, struct bond_slave *to) |
c6855ec5 | 1085 | OVS_REQ_WRLOCK(rwlock) |
f620b43a BP |
1086 | { |
1087 | struct bond_slave *from = hash->slave; | |
1088 | struct bond *bond = from->bond; | |
1089 | uint64_t delta = hash->tx_bytes; | |
1090 | ||
34582733 | 1091 | VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %"PRIdPTR") " |
f620b43a BP |
1092 | "from %s to %s (now carrying %"PRIu64"kB and " |
1093 | "%"PRIu64"kB load, respectively)", | |
1094 | bond->name, delta / 1024, hash - bond->hash, | |
1095 | from->name, to->name, | |
1096 | (from->tx_bytes - delta) / 1024, | |
1097 | (to->tx_bytes + delta) / 1024); | |
1098 | ||
1099 | /* Shift load away from 'from' to 'to'. */ | |
1100 | from->tx_bytes -= delta; | |
1101 | to->tx_bytes += delta; | |
1102 | ||
1103 | /* Arrange for flows to be revalidated. */ | |
dc30ea2d | 1104 | hash->slave = to; |
4a1b8f30 | 1105 | bond->bond_revalidate = true; |
f620b43a BP |
1106 | } |
1107 | ||
09a5d390 BP |
1108 | /* Picks and returns a bond_entry to migrate from 'from' (the most heavily |
1109 | * loaded bond slave) to a bond slave that has 'to_tx_bytes' bytes of load, | |
f620b43a BP |
1110 | * given that doing so must decrease the ratio of the load on the two slaves by |
1111 | * at least 0.1. Returns NULL if there is no appropriate entry. | |
1112 | * | |
1113 | * The list of entries isn't sorted. I don't know of a reason to prefer to | |
1114 | * shift away small hashes or large hashes. */ | |
1115 | static struct bond_entry * | |
1116 | choose_entry_to_migrate(const struct bond_slave *from, uint64_t to_tx_bytes) | |
c6855ec5 | 1117 | OVS_REQ_WRLOCK(rwlock) |
f620b43a BP |
1118 | { |
1119 | struct bond_entry *e; | |
1120 | ||
417e7e66 | 1121 | if (ovs_list_is_short(&from->entries)) { |
f620b43a BP |
1122 | /* 'from' carries no more than one MAC hash, so shifting load away from |
1123 | * it would be pointless. */ | |
1124 | return NULL; | |
1125 | } | |
1126 | ||
1127 | LIST_FOR_EACH (e, list_node, &from->entries) { | |
c460a6a7 AZ |
1128 | uint64_t delta = e->tx_bytes; /* The amount to rebalance. */ |
1129 | uint64_t ideal_tx_bytes = (from->tx_bytes + to_tx_bytes)/2; | |
1130 | /* Note, the ideal traffic is the mid point | |
1131 | * between 'from' and 'to'. This value does | |
1132 | * not change by rebalancing. */ | |
1133 | uint64_t new_low; /* The lower bandwidth between 'to' and 'from' | |
1134 | after rebalancing. */ | |
1135 | ||
1136 | new_low = MIN(from->tx_bytes - delta, to_tx_bytes + delta); | |
1137 | ||
1138 | if ((new_low > to_tx_bytes) && | |
1139 | (new_low - to_tx_bytes >= (ideal_tx_bytes - to_tx_bytes) / 10)) { | |
1140 | /* Only rebalance if the new 'low' is closer to to the mid point, | |
1141 | * and the improvement exceeds 10% of current traffic | |
1142 | * deviation from the ideal split. | |
1143 | * | |
1144 | * The improvement on the 'high' side is always the same as the | |
1145 | * 'low' side. Thus consider 'low' side is sufficient. */ | |
f620b43a BP |
1146 | return e; |
1147 | } | |
1148 | } | |
1149 | ||
1150 | return NULL; | |
1151 | } | |
1152 | ||
1153 | /* Inserts 'slave' into 'bals' so that descending order of 'tx_bytes' is | |
1154 | * maintained. */ | |
1155 | static void | |
ca6ba700 | 1156 | insert_bal(struct ovs_list *bals, struct bond_slave *slave) |
f620b43a BP |
1157 | { |
1158 | struct bond_slave *pos; | |
1159 | ||
1160 | LIST_FOR_EACH (pos, bal_node, bals) { | |
1161 | if (slave->tx_bytes > pos->tx_bytes) { | |
1162 | break; | |
1163 | } | |
1164 | } | |
417e7e66 | 1165 | ovs_list_insert(&pos->bal_node, &slave->bal_node); |
f620b43a BP |
1166 | } |
1167 | ||
1168 | /* Removes 'slave' from its current list and then inserts it into 'bals' so | |
1169 | * that descending order of 'tx_bytes' is maintained. */ | |
1170 | static void | |
ca6ba700 | 1171 | reinsert_bal(struct ovs_list *bals, struct bond_slave *slave) |
f620b43a | 1172 | { |
417e7e66 | 1173 | ovs_list_remove(&slave->bal_node); |
f620b43a BP |
1174 | insert_bal(bals, slave); |
1175 | } | |
1176 | ||
1177 | /* If 'bond' needs rebalancing, does so. | |
1178 | * | |
adcf00ba AZ |
1179 | * The caller should have called bond_account() for each active flow, or in case |
1180 | * of recirculation is used, have called bond_recirculation_account(bond), | |
1181 | * to ensure that flow data is consistently accounted at this point. | |
60cda7d6 AZ |
1182 | */ |
1183 | void | |
4a1b8f30 | 1184 | bond_rebalance(struct bond *bond) |
f620b43a BP |
1185 | { |
1186 | struct bond_slave *slave; | |
1187 | struct bond_entry *e; | |
ca6ba700 | 1188 | struct ovs_list bals; |
adcf00ba | 1189 | bool rebalanced = false; |
60cda7d6 | 1190 | bool use_recirc; |
f620b43a | 1191 | |
3bfd3972 | 1192 | ovs_rwlock_wrlock(&rwlock); |
1b137691 | 1193 | if (!bond_is_balanced(bond) || time_msec() < bond->next_rebalance) { |
adcf00ba | 1194 | goto done; |
f620b43a BP |
1195 | } |
1196 | bond->next_rebalance = time_msec() + bond->rebalance_interval; | |
1197 | ||
88186383 | 1198 | use_recirc = bond->ofproto->backer->rt_support.odp.recirc && |
6b95d23c | 1199 | bond_may_recirc(bond); |
60cda7d6 AZ |
1200 | |
1201 | if (use_recirc) { | |
1202 | bond_recirculation_account(bond); | |
1203 | } | |
1204 | ||
f620b43a BP |
1205 | /* Add each bond_entry to its slave's 'entries' list. |
1206 | * Compute each slave's tx_bytes as the sum of its entries' tx_bytes. */ | |
1207 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
1208 | slave->tx_bytes = 0; | |
417e7e66 | 1209 | ovs_list_init(&slave->entries); |
f620b43a BP |
1210 | } |
1211 | for (e = &bond->hash[0]; e <= &bond->hash[BOND_MASK]; e++) { | |
1212 | if (e->slave && e->tx_bytes) { | |
1213 | e->slave->tx_bytes += e->tx_bytes; | |
417e7e66 | 1214 | ovs_list_push_back(&e->slave->entries, &e->list_node); |
f620b43a BP |
1215 | } |
1216 | } | |
1217 | ||
1218 | /* Add enabled slaves to 'bals' in descending order of tx_bytes. | |
1219 | * | |
1220 | * XXX This is O(n**2) in the number of slaves but it could be O(n lg n) | |
1221 | * with a proper list sort algorithm. */ | |
417e7e66 | 1222 | ovs_list_init(&bals); |
f620b43a BP |
1223 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { |
1224 | if (slave->enabled) { | |
1225 | insert_bal(&bals, slave); | |
1226 | } | |
1227 | } | |
1228 | log_bals(bond, &bals); | |
1229 | ||
1230 | /* Shift load from the most-loaded slaves to the least-loaded slaves. */ | |
417e7e66 BW |
1231 | while (!ovs_list_is_short(&bals)) { |
1232 | struct bond_slave *from = bond_slave_from_bal_node(ovs_list_front(&bals)); | |
1233 | struct bond_slave *to = bond_slave_from_bal_node(ovs_list_back(&bals)); | |
f620b43a BP |
1234 | uint64_t overload; |
1235 | ||
1236 | overload = from->tx_bytes - to->tx_bytes; | |
1237 | if (overload < to->tx_bytes >> 5 || overload < 100000) { | |
1238 | /* The extra load on 'from' (and all less-loaded slaves), compared | |
1239 | * to that of 'to' (the least-loaded slave), is less than ~3%, or | |
1240 | * it is less than ~1Mbps. No point in rebalancing. */ | |
1241 | break; | |
1242 | } | |
1243 | ||
09a5d390 BP |
1244 | /* 'from' is carrying significantly more load than 'to'. Pick a hash |
1245 | * to move from 'from' to 'to'. */ | |
f620b43a BP |
1246 | e = choose_entry_to_migrate(from, to->tx_bytes); |
1247 | if (e) { | |
4a1b8f30 | 1248 | bond_shift_load(e, to); |
f620b43a BP |
1249 | |
1250 | /* Delete element from from->entries. | |
1251 | * | |
1252 | * We don't add the element to to->hashes. That would only allow | |
1253 | * 'e' to be migrated to another slave in this rebalancing run, and | |
1254 | * there is no point in doing that. */ | |
417e7e66 | 1255 | ovs_list_remove(&e->list_node); |
f620b43a BP |
1256 | |
1257 | /* Re-sort 'bals'. */ | |
1258 | reinsert_bal(&bals, from); | |
1259 | reinsert_bal(&bals, to); | |
60cda7d6 | 1260 | rebalanced = true; |
f620b43a BP |
1261 | } else { |
1262 | /* Can't usefully migrate anything away from 'from'. | |
1263 | * Don't reconsider it. */ | |
417e7e66 | 1264 | ovs_list_remove(&from->bal_node); |
f620b43a BP |
1265 | } |
1266 | } | |
1267 | ||
1268 | /* Implement exponentially weighted moving average. A weight of 1/2 causes | |
1269 | * historical data to decay to <1% in 7 rebalancing runs. 1,000,000 bytes | |
1270 | * take 20 rebalancing runs to decay to 0 and get deleted entirely. */ | |
1271 | for (e = &bond->hash[0]; e <= &bond->hash[BOND_MASK]; e++) { | |
1272 | e->tx_bytes /= 2; | |
f620b43a | 1273 | } |
adcf00ba | 1274 | |
60cda7d6 | 1275 | if (use_recirc && rebalanced) { |
ca8127fd | 1276 | bond_update_post_recirc_rules__(bond,true); |
60cda7d6 | 1277 | } |
2f486d4c AZ |
1278 | |
1279 | done: | |
3bfd3972 | 1280 | ovs_rwlock_unlock(&rwlock); |
f620b43a BP |
1281 | } |
1282 | \f | |
1283 | /* Bonding unixctl user interface functions. */ | |
1284 | ||
1285 | static struct bond * | |
3bfd3972 | 1286 | bond_find(const char *name) OVS_REQ_RDLOCK(rwlock) |
f620b43a BP |
1287 | { |
1288 | struct bond *bond; | |
1289 | ||
1290 | HMAP_FOR_EACH_WITH_HASH (bond, hmap_node, hash_string(name, 0), | |
3bfd3972 | 1291 | all_bonds) { |
f620b43a BP |
1292 | if (!strcmp(bond->name, name)) { |
1293 | return bond; | |
1294 | } | |
1295 | } | |
1296 | return NULL; | |
1297 | } | |
1298 | ||
1299 | static struct bond_slave * | |
1300 | bond_lookup_slave(struct bond *bond, const char *slave_name) | |
1301 | { | |
1302 | struct bond_slave *slave; | |
1303 | ||
1304 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
1305 | if (!strcmp(slave->name, slave_name)) { | |
1306 | return slave; | |
1307 | } | |
1308 | } | |
1309 | return NULL; | |
1310 | } | |
1311 | ||
1312 | static void | |
1313 | bond_unixctl_list(struct unixctl_conn *conn, | |
0e15264f BP |
1314 | int argc OVS_UNUSED, const char *argv[] OVS_UNUSED, |
1315 | void *aux OVS_UNUSED) | |
f620b43a BP |
1316 | { |
1317 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1318 | const struct bond *bond; | |
1319 | ||
adcf00ba | 1320 | ds_put_cstr(&ds, "bond\ttype\trecircID\tslaves\n"); |
f620b43a | 1321 | |
3bfd3972 EJ |
1322 | ovs_rwlock_rdlock(&rwlock); |
1323 | HMAP_FOR_EACH (bond, hmap_node, all_bonds) { | |
f620b43a BP |
1324 | const struct bond_slave *slave; |
1325 | size_t i; | |
1326 | ||
adcf00ba AZ |
1327 | ds_put_format(&ds, "%s\t%s\t%d\t", bond->name, |
1328 | bond_mode_to_string(bond->balance), bond->recirc_id); | |
f620b43a BP |
1329 | |
1330 | i = 0; | |
1331 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
1332 | if (i++ > 0) { | |
1333 | ds_put_cstr(&ds, ", "); | |
1334 | } | |
1335 | ds_put_cstr(&ds, slave->name); | |
1336 | } | |
1337 | ds_put_char(&ds, '\n'); | |
1338 | } | |
3bfd3972 | 1339 | ovs_rwlock_unlock(&rwlock); |
bde9f75d | 1340 | unixctl_command_reply(conn, ds_cstr(&ds)); |
f620b43a BP |
1341 | ds_destroy(&ds); |
1342 | } | |
1343 | ||
1344 | static void | |
c33a8a25 | 1345 | bond_print_details(struct ds *ds, const struct bond *bond) |
3bfd3972 | 1346 | OVS_REQ_RDLOCK(rwlock) |
f620b43a | 1347 | { |
fc1d4f01 EJ |
1348 | struct shash slave_shash = SHASH_INITIALIZER(&slave_shash); |
1349 | const struct shash_node **sorted_slaves = NULL; | |
f620b43a | 1350 | const struct bond_slave *slave; |
adcf00ba AZ |
1351 | bool may_recirc; |
1352 | uint32_t recirc_id; | |
fc1d4f01 | 1353 | int i; |
f620b43a | 1354 | |
c33a8a25 EJ |
1355 | ds_put_format(ds, "---- %s ----\n", bond->name); |
1356 | ds_put_format(ds, "bond_mode: %s\n", | |
f620b43a BP |
1357 | bond_mode_to_string(bond->balance)); |
1358 | ||
6b95d23c AZ |
1359 | may_recirc = bond_may_recirc(bond); |
1360 | recirc_id = bond->recirc_id; | |
adcf00ba AZ |
1361 | ds_put_format(ds, "bond may use recirculation: %s, Recirc-ID : %d\n", |
1362 | may_recirc ? "yes" : "no", may_recirc ? recirc_id: -1); | |
1363 | ||
c33a8a25 | 1364 | ds_put_format(ds, "bond-hash-basis: %"PRIu32"\n", bond->basis); |
672d18b2 | 1365 | |
c33a8a25 EJ |
1366 | ds_put_format(ds, "updelay: %d ms\n", bond->updelay); |
1367 | ds_put_format(ds, "downdelay: %d ms\n", bond->downdelay); | |
f620b43a | 1368 | |
1b137691 | 1369 | if (bond_is_balanced(bond)) { |
c33a8a25 | 1370 | ds_put_format(ds, "next rebalance: %lld ms\n", |
f620b43a BP |
1371 | bond->next_rebalance - time_msec()); |
1372 | } | |
1373 | ||
bdebeece EJ |
1374 | ds_put_cstr(ds, "lacp_status: "); |
1375 | switch (bond->lacp_status) { | |
1376 | case LACP_NEGOTIATED: | |
1377 | ds_put_cstr(ds, "negotiated\n"); | |
1378 | break; | |
1379 | case LACP_CONFIGURED: | |
1380 | ds_put_cstr(ds, "configured\n"); | |
1381 | break; | |
1382 | case LACP_DISABLED: | |
1383 | ds_put_cstr(ds, "off\n"); | |
1384 | break; | |
1385 | default: | |
1386 | ds_put_cstr(ds, "<unknown>\n"); | |
1387 | break; | |
1388 | } | |
4d6fb5eb | 1389 | |
57fc4fd0 | 1390 | ds_put_format(ds, "lacp_fallback_ab: %s\n", |
1391 | bond->lacp_fallback_ab ? "true" : "false"); | |
1392 | ||
3e5aeeb5 AZ |
1393 | ds_put_cstr(ds, "active slave mac: "); |
1394 | ds_put_format(ds, ETH_ADDR_FMT, ETH_ADDR_ARGS(bond->active_slave_mac)); | |
1395 | slave = bond_find_slave_by_mac(bond, bond->active_slave_mac); | |
1396 | ds_put_format(ds,"(%s)\n", slave ? slave->name : "none"); | |
1397 | ||
f620b43a | 1398 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { |
fc1d4f01 EJ |
1399 | shash_add(&slave_shash, slave->name, slave); |
1400 | } | |
1401 | sorted_slaves = shash_sort(&slave_shash); | |
1402 | ||
1403 | for (i = 0; i < shash_count(&slave_shash); i++) { | |
f620b43a | 1404 | struct bond_entry *be; |
f620b43a | 1405 | |
fc1d4f01 EJ |
1406 | slave = sorted_slaves[i]->data; |
1407 | ||
f620b43a | 1408 | /* Basic info. */ |
c33a8a25 | 1409 | ds_put_format(ds, "\nslave %s: %s\n", |
f620b43a BP |
1410 | slave->name, slave->enabled ? "enabled" : "disabled"); |
1411 | if (slave == bond->active_slave) { | |
828519ca | 1412 | ds_put_cstr(ds, " active slave\n"); |
f620b43a BP |
1413 | } |
1414 | if (slave->delay_expires != LLONG_MAX) { | |
828519ca | 1415 | ds_put_format(ds, " %s expires in %lld ms\n", |
f620b43a BP |
1416 | slave->enabled ? "downdelay" : "updelay", |
1417 | slave->delay_expires - time_msec()); | |
1418 | } | |
1419 | ||
828519ca | 1420 | ds_put_format(ds, " may_enable: %s\n", |
296f6519 | 1421 | slave->may_enable ? "true" : "false"); |
4d6fb5eb | 1422 | |
1b137691 | 1423 | if (!bond_is_balanced(bond)) { |
f620b43a BP |
1424 | continue; |
1425 | } | |
1426 | ||
1427 | /* Hashes. */ | |
f620b43a BP |
1428 | for (be = bond->hash; be <= &bond->hash[BOND_MASK]; be++) { |
1429 | int hash = be - bond->hash; | |
f6ba1f35 | 1430 | uint64_t be_tx_k; |
f620b43a BP |
1431 | |
1432 | if (be->slave != slave) { | |
1433 | continue; | |
1434 | } | |
1435 | ||
f6ba1f35 AZ |
1436 | be_tx_k = be->tx_bytes / 1024; |
1437 | if (be_tx_k) { | |
828519ca | 1438 | ds_put_format(ds, " hash %d: %"PRIu64" kB load\n", |
f6ba1f35 AZ |
1439 | hash, be_tx_k); |
1440 | } | |
f620b43a | 1441 | |
7b9f1974 | 1442 | /* XXX How can we list the MACs assigned to hashes of SLB bonds? */ |
f620b43a BP |
1443 | } |
1444 | } | |
fc1d4f01 EJ |
1445 | shash_destroy(&slave_shash); |
1446 | free(sorted_slaves); | |
c33a8a25 EJ |
1447 | ds_put_cstr(ds, "\n"); |
1448 | } | |
1449 | ||
1450 | static void | |
1451 | bond_unixctl_show(struct unixctl_conn *conn, | |
1452 | int argc, const char *argv[], | |
1453 | void *aux OVS_UNUSED) | |
1454 | { | |
1455 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1456 | ||
3bfd3972 | 1457 | ovs_rwlock_rdlock(&rwlock); |
c33a8a25 EJ |
1458 | if (argc > 1) { |
1459 | const struct bond *bond = bond_find(argv[1]); | |
1460 | ||
1461 | if (!bond) { | |
bde9f75d | 1462 | unixctl_command_reply_error(conn, "no such bond"); |
3bfd3972 | 1463 | goto out; |
c33a8a25 EJ |
1464 | } |
1465 | bond_print_details(&ds, bond); | |
1466 | } else { | |
1467 | const struct bond *bond; | |
1468 | ||
3bfd3972 | 1469 | HMAP_FOR_EACH (bond, hmap_node, all_bonds) { |
c33a8a25 EJ |
1470 | bond_print_details(&ds, bond); |
1471 | } | |
1472 | } | |
1473 | ||
bde9f75d | 1474 | unixctl_command_reply(conn, ds_cstr(&ds)); |
f620b43a | 1475 | ds_destroy(&ds); |
3bfd3972 EJ |
1476 | |
1477 | out: | |
1478 | ovs_rwlock_unlock(&rwlock); | |
f620b43a BP |
1479 | } |
1480 | ||
1481 | static void | |
0e15264f BP |
1482 | bond_unixctl_migrate(struct unixctl_conn *conn, |
1483 | int argc OVS_UNUSED, const char *argv[], | |
f620b43a BP |
1484 | void *aux OVS_UNUSED) |
1485 | { | |
0e15264f BP |
1486 | const char *bond_s = argv[1]; |
1487 | const char *hash_s = argv[2]; | |
1488 | const char *slave_s = argv[3]; | |
f620b43a BP |
1489 | struct bond *bond; |
1490 | struct bond_slave *slave; | |
1491 | struct bond_entry *entry; | |
1492 | int hash; | |
1493 | ||
3bfd3972 | 1494 | ovs_rwlock_wrlock(&rwlock); |
f620b43a BP |
1495 | bond = bond_find(bond_s); |
1496 | if (!bond) { | |
bde9f75d | 1497 | unixctl_command_reply_error(conn, "no such bond"); |
3bfd3972 | 1498 | goto out; |
f620b43a BP |
1499 | } |
1500 | ||
1501 | if (bond->balance != BM_SLB) { | |
bde9f75d | 1502 | unixctl_command_reply_error(conn, "not an SLB bond"); |
3bfd3972 | 1503 | goto out; |
f620b43a BP |
1504 | } |
1505 | ||
1506 | if (strspn(hash_s, "0123456789") == strlen(hash_s)) { | |
1507 | hash = atoi(hash_s) & BOND_MASK; | |
1508 | } else { | |
bde9f75d | 1509 | unixctl_command_reply_error(conn, "bad hash"); |
3bfd3972 | 1510 | goto out; |
f620b43a BP |
1511 | } |
1512 | ||
1513 | slave = bond_lookup_slave(bond, slave_s); | |
1514 | if (!slave) { | |
bde9f75d | 1515 | unixctl_command_reply_error(conn, "no such slave"); |
3bfd3972 | 1516 | goto out; |
f620b43a BP |
1517 | } |
1518 | ||
1519 | if (!slave->enabled) { | |
bde9f75d | 1520 | unixctl_command_reply_error(conn, "cannot migrate to disabled slave"); |
3bfd3972 | 1521 | goto out; |
f620b43a BP |
1522 | } |
1523 | ||
1524 | entry = &bond->hash[hash]; | |
4a1b8f30 | 1525 | bond->bond_revalidate = true; |
f620b43a | 1526 | entry->slave = slave; |
bde9f75d | 1527 | unixctl_command_reply(conn, "migrated"); |
3bfd3972 EJ |
1528 | |
1529 | out: | |
1530 | ovs_rwlock_unlock(&rwlock); | |
f620b43a BP |
1531 | } |
1532 | ||
1533 | static void | |
0e15264f BP |
1534 | bond_unixctl_set_active_slave(struct unixctl_conn *conn, |
1535 | int argc OVS_UNUSED, const char *argv[], | |
f620b43a BP |
1536 | void *aux OVS_UNUSED) |
1537 | { | |
0e15264f BP |
1538 | const char *bond_s = argv[1]; |
1539 | const char *slave_s = argv[2]; | |
f620b43a BP |
1540 | struct bond *bond; |
1541 | struct bond_slave *slave; | |
1542 | ||
3bfd3972 | 1543 | ovs_rwlock_wrlock(&rwlock); |
f620b43a BP |
1544 | bond = bond_find(bond_s); |
1545 | if (!bond) { | |
bde9f75d | 1546 | unixctl_command_reply_error(conn, "no such bond"); |
3bfd3972 | 1547 | goto out; |
f620b43a BP |
1548 | } |
1549 | ||
1550 | slave = bond_lookup_slave(bond, slave_s); | |
1551 | if (!slave) { | |
bde9f75d | 1552 | unixctl_command_reply_error(conn, "no such slave"); |
3bfd3972 | 1553 | goto out; |
f620b43a BP |
1554 | } |
1555 | ||
1556 | if (!slave->enabled) { | |
bde9f75d | 1557 | unixctl_command_reply_error(conn, "cannot make disabled slave active"); |
3bfd3972 | 1558 | goto out; |
f620b43a BP |
1559 | } |
1560 | ||
1561 | if (bond->active_slave != slave) { | |
4a1b8f30 | 1562 | bond->bond_revalidate = true; |
f620b43a | 1563 | bond->active_slave = slave; |
f620b43a BP |
1564 | VLOG_INFO("bond %s: active interface is now %s", |
1565 | bond->name, slave->name); | |
1566 | bond->send_learning_packets = true; | |
bde9f75d | 1567 | unixctl_command_reply(conn, "done"); |
3e5aeeb5 | 1568 | bond_active_slave_changed(bond); |
f620b43a | 1569 | } else { |
bde9f75d | 1570 | unixctl_command_reply(conn, "no change"); |
f620b43a | 1571 | } |
3bfd3972 EJ |
1572 | out: |
1573 | ovs_rwlock_unlock(&rwlock); | |
f620b43a BP |
1574 | } |
1575 | ||
1576 | static void | |
0e15264f | 1577 | enable_slave(struct unixctl_conn *conn, const char *argv[], bool enable) |
f620b43a | 1578 | { |
0e15264f BP |
1579 | const char *bond_s = argv[1]; |
1580 | const char *slave_s = argv[2]; | |
f620b43a BP |
1581 | struct bond *bond; |
1582 | struct bond_slave *slave; | |
1583 | ||
3bfd3972 | 1584 | ovs_rwlock_wrlock(&rwlock); |
f620b43a BP |
1585 | bond = bond_find(bond_s); |
1586 | if (!bond) { | |
bde9f75d | 1587 | unixctl_command_reply_error(conn, "no such bond"); |
3bfd3972 | 1588 | goto out; |
f620b43a BP |
1589 | } |
1590 | ||
1591 | slave = bond_lookup_slave(bond, slave_s); | |
1592 | if (!slave) { | |
bde9f75d | 1593 | unixctl_command_reply_error(conn, "no such slave"); |
3bfd3972 | 1594 | goto out; |
f620b43a BP |
1595 | } |
1596 | ||
4a1b8f30 | 1597 | bond_enable_slave(slave, enable); |
bde9f75d | 1598 | unixctl_command_reply(conn, enable ? "enabled" : "disabled"); |
3bfd3972 EJ |
1599 | |
1600 | out: | |
1601 | ovs_rwlock_unlock(&rwlock); | |
f620b43a BP |
1602 | } |
1603 | ||
1604 | static void | |
0e15264f BP |
1605 | bond_unixctl_enable_slave(struct unixctl_conn *conn, |
1606 | int argc OVS_UNUSED, const char *argv[], | |
f620b43a BP |
1607 | void *aux OVS_UNUSED) |
1608 | { | |
0e15264f | 1609 | enable_slave(conn, argv, true); |
f620b43a BP |
1610 | } |
1611 | ||
1612 | static void | |
0e15264f BP |
1613 | bond_unixctl_disable_slave(struct unixctl_conn *conn, |
1614 | int argc OVS_UNUSED, const char *argv[], | |
f620b43a BP |
1615 | void *aux OVS_UNUSED) |
1616 | { | |
0e15264f | 1617 | enable_slave(conn, argv, false); |
f620b43a BP |
1618 | } |
1619 | ||
1620 | static void | |
0e15264f | 1621 | bond_unixctl_hash(struct unixctl_conn *conn, int argc, const char *argv[], |
f620b43a BP |
1622 | void *aux OVS_UNUSED) |
1623 | { | |
0e15264f BP |
1624 | const char *mac_s = argv[1]; |
1625 | const char *vlan_s = argc > 2 ? argv[2] : NULL; | |
1626 | const char *basis_s = argc > 3 ? argv[3] : NULL; | |
74ff3298 | 1627 | struct eth_addr mac; |
f620b43a BP |
1628 | uint8_t hash; |
1629 | char *hash_cstr; | |
1630 | unsigned int vlan; | |
672d18b2 | 1631 | uint32_t basis; |
f620b43a BP |
1632 | |
1633 | if (vlan_s) { | |
c2c28dfd | 1634 | if (!ovs_scan(vlan_s, "%u", &vlan)) { |
bde9f75d | 1635 | unixctl_command_reply_error(conn, "invalid vlan"); |
f620b43a BP |
1636 | return; |
1637 | } | |
1638 | } else { | |
dc155bff | 1639 | vlan = 0; |
f620b43a BP |
1640 | } |
1641 | ||
672d18b2 | 1642 | if (basis_s) { |
c2c28dfd | 1643 | if (!ovs_scan(basis_s, "%"SCNu32, &basis)) { |
bde9f75d | 1644 | unixctl_command_reply_error(conn, "invalid basis"); |
672d18b2 EJ |
1645 | return; |
1646 | } | |
1647 | } else { | |
1648 | basis = 0; | |
1649 | } | |
1650 | ||
c2c28dfd | 1651 | if (ovs_scan(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))) { |
e9013d6a | 1652 | hash = hash_mac(mac, vlan, basis) & BOND_MASK; |
f620b43a BP |
1653 | |
1654 | hash_cstr = xasprintf("%u", hash); | |
bde9f75d | 1655 | unixctl_command_reply(conn, hash_cstr); |
f620b43a BP |
1656 | free(hash_cstr); |
1657 | } else { | |
bde9f75d | 1658 | unixctl_command_reply_error(conn, "invalid mac"); |
f620b43a BP |
1659 | } |
1660 | } | |
1661 | ||
1662 | void | |
1663 | bond_init(void) | |
1664 | { | |
0e15264f | 1665 | unixctl_command_register("bond/list", "", 0, 0, bond_unixctl_list, NULL); |
c33a8a25 EJ |
1666 | unixctl_command_register("bond/show", "[port]", 0, 1, bond_unixctl_show, |
1667 | NULL); | |
0e15264f | 1668 | unixctl_command_register("bond/migrate", "port hash slave", 3, 3, |
7ff2009a | 1669 | bond_unixctl_migrate, NULL); |
0e15264f | 1670 | unixctl_command_register("bond/set-active-slave", "port slave", 2, 2, |
f620b43a | 1671 | bond_unixctl_set_active_slave, NULL); |
0e15264f | 1672 | unixctl_command_register("bond/enable-slave", "port slave", 2, 2, |
7ff2009a | 1673 | bond_unixctl_enable_slave, NULL); |
0e15264f | 1674 | unixctl_command_register("bond/disable-slave", "port slave", 2, 2, |
7ff2009a | 1675 | bond_unixctl_disable_slave, NULL); |
0e15264f | 1676 | unixctl_command_register("bond/hash", "mac [vlan] [basis]", 1, 3, |
7ff2009a | 1677 | bond_unixctl_hash, NULL); |
f620b43a BP |
1678 | } |
1679 | \f | |
95aafb2a EJ |
1680 | static void |
1681 | bond_entry_reset(struct bond *bond) | |
1682 | { | |
90061ea7 | 1683 | if (bond->balance != BM_AB && !bond_is_falling_back_to_ab(bond)) { |
9e1a6910 | 1684 | size_t hash_len = BOND_BUCKETS * sizeof *bond->hash; |
95aafb2a EJ |
1685 | |
1686 | if (!bond->hash) { | |
1687 | bond->hash = xmalloc(hash_len); | |
1688 | } | |
1689 | memset(bond->hash, 0, hash_len); | |
1690 | ||
1691 | bond->next_rebalance = time_msec() + bond->rebalance_interval; | |
1692 | } else { | |
1693 | free(bond->hash); | |
1694 | bond->hash = NULL; | |
05df1623 AZ |
1695 | /* Remove existing post recirc rules. */ |
1696 | update_recirc_rules(bond); | |
95aafb2a EJ |
1697 | } |
1698 | } | |
1699 | ||
f620b43a BP |
1700 | static struct bond_slave * |
1701 | bond_slave_lookup(struct bond *bond, const void *slave_) | |
1702 | { | |
1703 | struct bond_slave *slave; | |
1704 | ||
1705 | HMAP_FOR_EACH_IN_BUCKET (slave, hmap_node, hash_pointer(slave_, 0), | |
1706 | &bond->slaves) { | |
1707 | if (slave->aux == slave_) { | |
1708 | return slave; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | return NULL; | |
1713 | } | |
1714 | ||
f620b43a | 1715 | static void |
4a1b8f30 | 1716 | bond_enable_slave(struct bond_slave *slave, bool enable) |
f620b43a | 1717 | { |
5fef88ea | 1718 | struct bond *bond = slave->bond; |
1719 | ||
f620b43a BP |
1720 | slave->delay_expires = LLONG_MAX; |
1721 | if (enable != slave->enabled) { | |
4a1b8f30 | 1722 | slave->bond->bond_revalidate = true; |
f620b43a | 1723 | slave->enabled = enable; |
f1c8a79c AW |
1724 | |
1725 | ovs_mutex_lock(&slave->bond->mutex); | |
1726 | if (enable) { | |
417e7e66 | 1727 | ovs_list_insert(&slave->bond->enabled_slaves, &slave->list_node); |
f1c8a79c | 1728 | } else { |
5fef88ea | 1729 | bond->send_learning_packets = true; |
417e7e66 | 1730 | ovs_list_remove(&slave->list_node); |
f1c8a79c AW |
1731 | } |
1732 | ovs_mutex_unlock(&slave->bond->mutex); | |
1733 | ||
4a1b8f30 EJ |
1734 | VLOG_INFO("interface %s: %s", slave->name, |
1735 | slave->enabled ? "enabled" : "disabled"); | |
f620b43a BP |
1736 | } |
1737 | } | |
1738 | ||
1739 | static void | |
4a1b8f30 | 1740 | bond_link_status_update(struct bond_slave *slave) |
f620b43a BP |
1741 | { |
1742 | struct bond *bond = slave->bond; | |
1743 | bool up; | |
1744 | ||
296f6519 | 1745 | up = netdev_get_carrier(slave->netdev) && slave->may_enable; |
f620b43a BP |
1746 | if ((up == slave->enabled) != (slave->delay_expires == LLONG_MAX)) { |
1747 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1748 | VLOG_INFO_RL(&rl, "interface %s: link state %s", | |
1749 | slave->name, up ? "up" : "down"); | |
1750 | if (up == slave->enabled) { | |
1751 | slave->delay_expires = LLONG_MAX; | |
1752 | VLOG_INFO_RL(&rl, "interface %s: will not be %s", | |
1753 | slave->name, up ? "disabled" : "enabled"); | |
1754 | } else { | |
f9417807 | 1755 | int delay = up ? bond->updelay : bond->downdelay; |
f620b43a BP |
1756 | slave->delay_expires = time_msec() + delay; |
1757 | if (delay) { | |
1758 | VLOG_INFO_RL(&rl, "interface %s: will be %s if it stays %s " | |
1759 | "for %d ms", | |
1760 | slave->name, | |
1761 | up ? "enabled" : "disabled", | |
1762 | up ? "up" : "down", | |
1763 | delay); | |
1764 | } | |
1765 | } | |
1766 | } | |
1767 | ||
1768 | if (time_msec() >= slave->delay_expires) { | |
4a1b8f30 | 1769 | bond_enable_slave(slave, up); |
f620b43a BP |
1770 | } |
1771 | } | |
1772 | ||
fb0b29a3 EJ |
1773 | static unsigned int |
1774 | bond_hash(const struct bond *bond, const struct flow *flow, uint16_t vlan) | |
1775 | { | |
cb22974d | 1776 | ovs_assert(bond->balance == BM_TCP || bond->balance == BM_SLB); |
fb0b29a3 | 1777 | |
bdebeece | 1778 | return (bond->balance == BM_TCP |
42781e77 | 1779 | ? flow_hash_5tuple(flow, bond->basis) |
e9013d6a | 1780 | : hash_mac(flow->dl_src, vlan, bond->basis)); |
fb0b29a3 EJ |
1781 | } |
1782 | ||
f620b43a BP |
1783 | static struct bond_entry * |
1784 | lookup_bond_entry(const struct bond *bond, const struct flow *flow, | |
1785 | uint16_t vlan) | |
1786 | { | |
fb0b29a3 | 1787 | return &bond->hash[bond_hash(bond, flow, vlan) & BOND_MASK]; |
f620b43a BP |
1788 | } |
1789 | ||
f1c8a79c AW |
1790 | /* Selects and returns an enabled slave from the 'enabled_slaves' list |
1791 | * in a round-robin fashion. If the 'enabled_slaves' list is empty, | |
1792 | * returns NULL. */ | |
1793 | static struct bond_slave * | |
1794 | get_enabled_slave(struct bond *bond) | |
1795 | { | |
ca6ba700 | 1796 | struct ovs_list *node; |
f1c8a79c AW |
1797 | |
1798 | ovs_mutex_lock(&bond->mutex); | |
417e7e66 | 1799 | if (ovs_list_is_empty(&bond->enabled_slaves)) { |
f1c8a79c AW |
1800 | ovs_mutex_unlock(&bond->mutex); |
1801 | return NULL; | |
1802 | } | |
1803 | ||
417e7e66 BW |
1804 | node = ovs_list_pop_front(&bond->enabled_slaves); |
1805 | ovs_list_push_back(&bond->enabled_slaves, node); | |
f1c8a79c AW |
1806 | ovs_mutex_unlock(&bond->mutex); |
1807 | ||
1808 | return CONTAINER_OF(node, struct bond_slave, list_node); | |
1809 | } | |
1810 | ||
f620b43a BP |
1811 | static struct bond_slave * |
1812 | choose_output_slave(const struct bond *bond, const struct flow *flow, | |
4a1b8f30 | 1813 | struct flow_wildcards *wc, uint16_t vlan) |
f620b43a BP |
1814 | { |
1815 | struct bond_entry *e; | |
9dd165e0 | 1816 | int balance; |
f620b43a | 1817 | |
9dd165e0 | 1818 | balance = bond->balance; |
bdebeece EJ |
1819 | if (bond->lacp_status == LACP_CONFIGURED) { |
1820 | /* LACP has been configured on this bond but negotiations were | |
9dd165e0 RK |
1821 | * unsuccussful. If lacp_fallback_ab is enabled use active- |
1822 | * backup mode else drop all traffic. */ | |
1823 | if (!bond->lacp_fallback_ab) { | |
1824 | return NULL; | |
1825 | } | |
1826 | balance = BM_AB; | |
bdebeece EJ |
1827 | } |
1828 | ||
9dd165e0 | 1829 | switch (balance) { |
f620b43a BP |
1830 | case BM_AB: |
1831 | return bond->active_slave; | |
1832 | ||
f620b43a | 1833 | case BM_TCP: |
bdebeece EJ |
1834 | if (bond->lacp_status != LACP_NEGOTIATED) { |
1835 | /* Must have LACP negotiations for TCP balanced bonds. */ | |
1836 | return NULL; | |
1837 | } | |
bcd2633a | 1838 | if (wc) { |
deb67947 | 1839 | flow_mask_hash_fields(flow, wc, NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP); |
bcd2633a | 1840 | } |
bdebeece EJ |
1841 | /* Fall Through. */ |
1842 | case BM_SLB: | |
deb67947 | 1843 | if (wc && balance == BM_SLB) { |
6cdd5145 | 1844 | flow_mask_hash_fields(flow, wc, NX_HASH_FIELDS_ETH_SRC); |
bcd2633a | 1845 | } |
f620b43a BP |
1846 | e = lookup_bond_entry(bond, flow, vlan); |
1847 | if (!e->slave || !e->slave->enabled) { | |
f1c8a79c | 1848 | e->slave = get_enabled_slave(CONST_CAST(struct bond*, bond)); |
f620b43a BP |
1849 | } |
1850 | return e->slave; | |
1851 | ||
1852 | default: | |
428b2edd | 1853 | OVS_NOT_REACHED(); |
f620b43a BP |
1854 | } |
1855 | } | |
1856 | ||
1857 | static struct bond_slave * | |
1858 | bond_choose_slave(const struct bond *bond) | |
1859 | { | |
1860 | struct bond_slave *slave, *best; | |
1861 | ||
3e5aeeb5 AZ |
1862 | /* Find the last active slave. */ |
1863 | slave = bond_find_slave_by_mac(bond, bond->active_slave_mac); | |
1864 | if (slave && slave->enabled) { | |
1865 | return slave; | |
1866 | } | |
1867 | ||
f620b43a BP |
1868 | /* Find an enabled slave. */ |
1869 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
1870 | if (slave->enabled) { | |
1871 | return slave; | |
1872 | } | |
1873 | } | |
1874 | ||
1875 | /* All interfaces are disabled. Find an interface that will be enabled | |
1876 | * after its updelay expires. */ | |
1877 | best = NULL; | |
1878 | HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { | |
1879 | if (slave->delay_expires != LLONG_MAX | |
296f6519 | 1880 | && slave->may_enable |
f620b43a BP |
1881 | && (!best || slave->delay_expires < best->delay_expires)) { |
1882 | best = slave; | |
1883 | } | |
1884 | } | |
1885 | return best; | |
1886 | } | |
1887 | ||
1888 | static void | |
4a1b8f30 | 1889 | bond_choose_active_slave(struct bond *bond) |
f620b43a BP |
1890 | { |
1891 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1892 | struct bond_slave *old_active_slave = bond->active_slave; | |
1893 | ||
1894 | bond->active_slave = bond_choose_slave(bond); | |
1895 | if (bond->active_slave) { | |
1896 | if (bond->active_slave->enabled) { | |
1897 | VLOG_INFO_RL(&rl, "bond %s: active interface is now %s", | |
1898 | bond->name, bond->active_slave->name); | |
1899 | } else { | |
1900 | VLOG_INFO_RL(&rl, "bond %s: active interface is now %s, skipping " | |
1901 | "remaining %lld ms updelay (since no interface was " | |
1902 | "enabled)", bond->name, bond->active_slave->name, | |
1903 | bond->active_slave->delay_expires - time_msec()); | |
4a1b8f30 | 1904 | bond_enable_slave(bond->active_slave, true); |
f620b43a BP |
1905 | } |
1906 | ||
1907 | bond->send_learning_packets = true; | |
3e5aeeb5 AZ |
1908 | |
1909 | if (bond->active_slave != old_active_slave) { | |
1910 | bond_active_slave_changed(bond); | |
1911 | } | |
f620b43a | 1912 | } else if (old_active_slave) { |
f626af7a | 1913 | bond_active_slave_changed(bond); |
d28b9ead | 1914 | VLOG_INFO_RL(&rl, "bond %s: all interfaces disabled", bond->name); |
f620b43a BP |
1915 | } |
1916 | } | |
3e5aeeb5 AZ |
1917 | |
1918 | /* | |
1919 | * Return true if bond has unstored active slave change. | |
1920 | * If return true, 'mac' will store the bond's current active slave's | |
1921 | * MAC address. */ | |
1922 | bool | |
74ff3298 JR |
1923 | bond_get_changed_active_slave(const char *name, struct eth_addr *mac, |
1924 | bool force) | |
3e5aeeb5 AZ |
1925 | { |
1926 | struct bond *bond; | |
1927 | ||
1928 | ovs_rwlock_wrlock(&rwlock); | |
1929 | bond = bond_find(name); | |
1930 | if (bond) { | |
1931 | if (bond->active_slave_changed || force) { | |
74ff3298 | 1932 | *mac = bond->active_slave_mac; |
3e5aeeb5 AZ |
1933 | bond->active_slave_changed = false; |
1934 | ovs_rwlock_unlock(&rwlock); | |
1935 | return true; | |
1936 | } | |
1937 | } | |
1938 | ovs_rwlock_unlock(&rwlock); | |
1939 | ||
1940 | return false; | |
1941 | } |