]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* Copyright (c) 2008, 2009 Nicira Networks |
c93b1d6a | 2 | * |
a14bc59f BP |
3 | * Licensed under the Apache License, Version 2.0 (the "License"); |
4 | * you may not use this file except in compliance with the License. | |
5 | * You may obtain a copy of the License at: | |
064af421 | 6 | * |
a14bc59f | 7 | * http://www.apache.org/licenses/LICENSE-2.0 |
064af421 | 8 | * |
a14bc59f BP |
9 | * Unless required by applicable law or agreed to in writing, software |
10 | * distributed under the License is distributed on an "AS IS" BASIS, | |
11 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
12 | * See the License for the specific language governing permissions and | |
13 | * limitations under the License. | |
064af421 BP |
14 | */ |
15 | ||
16 | #include <config.h> | |
17 | #include "bridge.h" | |
18 | #include <assert.h> | |
19 | #include <errno.h> | |
20 | #include <arpa/inet.h> | |
21 | #include <ctype.h> | |
22 | #include <inttypes.h> | |
23 | #include <net/if.h> | |
24 | #include <openflow/openflow.h> | |
25 | #include <signal.h> | |
26 | #include <stdlib.h> | |
27 | #include <strings.h> | |
28 | #include <sys/stat.h> | |
29 | #include <sys/socket.h> | |
7e40e21d | 30 | #include <sys/types.h> |
064af421 BP |
31 | #include <unistd.h> |
32 | #include "bitmap.h" | |
33 | #include "cfg.h" | |
34 | #include "coverage.h" | |
35 | #include "dirs.h" | |
36 | #include "dpif.h" | |
37 | #include "dynamic-string.h" | |
38 | #include "flow.h" | |
39 | #include "hash.h" | |
40 | #include "list.h" | |
41 | #include "mac-learning.h" | |
42 | #include "netdev.h" | |
43 | #include "odp-util.h" | |
44 | #include "ofp-print.h" | |
45 | #include "ofpbuf.h" | |
da285df4 | 46 | #include "packets.h" |
064af421 BP |
47 | #include "poll-loop.h" |
48 | #include "port-array.h" | |
49 | #include "proc-net-compat.h" | |
50 | #include "process.h" | |
51 | #include "secchan/ofproto.h" | |
52 | #include "socket-util.h" | |
53 | #include "stp.h" | |
54 | #include "svec.h" | |
55 | #include "timeval.h" | |
56 | #include "util.h" | |
da285df4 | 57 | #include "unixctl.h" |
064af421 BP |
58 | #include "vconn.h" |
59 | #include "vconn-ssl.h" | |
60 | #include "xenserver.h" | |
61 | #include "xtoxll.h" | |
62 | ||
63 | #define THIS_MODULE VLM_bridge | |
64 | #include "vlog.h" | |
65 | ||
66 | struct dst { | |
67 | uint16_t vlan; | |
68 | uint16_t dp_ifidx; | |
69 | }; | |
70 | ||
71 | extern uint64_t mgmt_id; | |
72 | ||
73 | struct iface { | |
74 | struct port *port; /* Containing port. */ | |
75 | size_t port_ifidx; /* Index within containing port. */ | |
76 | ||
77 | char *name; /* Host network device name. */ | |
78 | int dp_ifidx; /* Index within kernel datapath. */ | |
79 | ||
80 | uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */ | |
81 | ||
82 | tag_type tag; /* Tag associated with this interface. */ | |
83 | bool enabled; /* May be chosen for flows? */ | |
84 | long long delay_expires; /* Time after which 'enabled' may change. */ | |
85 | }; | |
86 | ||
87 | #define BOND_MASK 0xff | |
88 | struct bond_entry { | |
89 | int iface_idx; /* Index of assigned iface, or -1 if none. */ | |
90 | uint64_t tx_bytes; /* Count of bytes recently transmitted. */ | |
91 | tag_type iface_tag; /* Tag associated with iface_idx. */ | |
92 | }; | |
93 | ||
94 | #define MAX_MIRRORS 32 | |
95 | typedef uint32_t mirror_mask_t; | |
96 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
97 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
98 | struct mirror { | |
99 | struct bridge *bridge; | |
100 | size_t idx; | |
101 | char *name; | |
102 | ||
103 | /* Selection criteria. */ | |
104 | struct svec src_ports; | |
105 | struct svec dst_ports; | |
106 | int *vlans; | |
107 | size_t n_vlans; | |
108 | ||
109 | /* Output. */ | |
110 | struct port *out_port; | |
111 | int out_vlan; | |
112 | }; | |
113 | ||
114 | #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ | |
115 | struct port { | |
116 | struct bridge *bridge; | |
117 | size_t port_idx; | |
118 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
119 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ | |
120 | char *name; | |
121 | ||
122 | /* An ordinary bridge port has 1 interface. | |
123 | * A bridge port for bonding has at least 2 interfaces. */ | |
124 | struct iface **ifaces; | |
125 | size_t n_ifaces, allocated_ifaces; | |
126 | ||
127 | /* Bonding info. */ | |
128 | struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ | |
129 | int active_iface; /* Ifidx on which bcasts accepted, or -1. */ | |
130 | tag_type active_iface_tag; /* Tag for bcast flows. */ | |
131 | tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ | |
132 | int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ | |
85c74638 | 133 | bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ |
064af421 BP |
134 | |
135 | /* Port mirroring info. */ | |
136 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
137 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
138 | bool is_mirror_output_port; /* Does port mirroring send frames here? */ | |
139 | ||
140 | /* Spanning tree info. */ | |
141 | enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */ | |
142 | tag_type stp_state_tag; /* Tag for STP state change. */ | |
143 | }; | |
144 | ||
145 | #define DP_MAX_PORTS 255 | |
146 | struct bridge { | |
147 | struct list node; /* Node in global list of bridges. */ | |
148 | char *name; /* User-specified arbitrary name. */ | |
149 | struct mac_learning *ml; /* MAC learning table, or null not to learn. */ | |
150 | bool sent_config_request; /* Successfully sent config request? */ | |
151 | uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ | |
152 | ||
153 | /* Support for remote controllers. */ | |
154 | char *controller; /* NULL if there is no remote controller; | |
155 | * "discover" to do controller discovery; | |
156 | * otherwise a vconn name. */ | |
157 | ||
158 | /* OpenFlow switch processing. */ | |
159 | struct ofproto *ofproto; /* OpenFlow switch. */ | |
160 | ||
161 | /* Kernel datapath information. */ | |
162 | struct dpif dpif; /* Kernel datapath. */ | |
163 | struct port_array ifaces; /* Indexed by kernel datapath port number. */ | |
164 | ||
165 | /* Bridge ports. */ | |
166 | struct port **ports; | |
167 | size_t n_ports, allocated_ports; | |
168 | ||
169 | /* Bonding. */ | |
170 | bool has_bonded_ports; | |
171 | long long int bond_next_rebalance; | |
172 | ||
173 | /* Flow tracking. */ | |
174 | bool flush; | |
175 | ||
176 | /* Flow statistics gathering. */ | |
177 | time_t next_stats_request; | |
178 | ||
179 | /* Port mirroring. */ | |
180 | struct mirror *mirrors[MAX_MIRRORS]; | |
181 | ||
182 | /* Spanning tree. */ | |
183 | struct stp *stp; | |
184 | long long int stp_last_tick; | |
185 | }; | |
186 | ||
187 | /* List of all bridges. */ | |
188 | static struct list all_bridges = LIST_INITIALIZER(&all_bridges); | |
189 | ||
190 | /* Maximum number of datapaths. */ | |
191 | enum { DP_MAX = 256 }; | |
192 | ||
193 | static struct bridge *bridge_create(const char *name); | |
194 | static void bridge_destroy(struct bridge *); | |
195 | static struct bridge *bridge_lookup(const char *name); | |
4f2cad2c | 196 | static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *); |
064af421 BP |
197 | static int bridge_run_one(struct bridge *); |
198 | static void bridge_reconfigure_one(struct bridge *); | |
199 | static void bridge_reconfigure_controller(struct bridge *); | |
200 | static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces); | |
201 | static void bridge_fetch_dp_ifaces(struct bridge *); | |
202 | static void bridge_flush(struct bridge *); | |
203 | static void bridge_pick_local_hw_addr(struct bridge *, | |
204 | uint8_t ea[ETH_ADDR_LEN], | |
205 | const char **devname); | |
206 | static uint64_t bridge_pick_datapath_id(struct bridge *, | |
207 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
208 | const char *devname); | |
209 | static uint64_t dpid_from_hash(const void *, size_t nbytes); | |
210 | ||
8c4c1387 BP |
211 | static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args); |
212 | ||
da285df4 | 213 | static void bond_init(void); |
064af421 BP |
214 | static void bond_run(struct bridge *); |
215 | static void bond_wait(struct bridge *); | |
216 | static void bond_rebalance_port(struct port *); | |
2303f3b2 | 217 | static void bond_send_learning_packets(struct port *); |
064af421 BP |
218 | |
219 | static void port_create(struct bridge *, const char *name); | |
220 | static void port_reconfigure(struct port *); | |
221 | static void port_destroy(struct port *); | |
222 | static struct port *port_lookup(const struct bridge *, const char *name); | |
da285df4 | 223 | static struct iface *port_lookup_iface(const struct port *, const char *name); |
064af421 BP |
224 | static struct port *port_from_dp_ifidx(const struct bridge *, |
225 | uint16_t dp_ifidx); | |
226 | static void port_update_bond_compat(struct port *); | |
227 | static void port_update_vlan_compat(struct port *); | |
228 | ||
229 | static void mirror_create(struct bridge *, const char *name); | |
230 | static void mirror_destroy(struct mirror *); | |
231 | static void mirror_reconfigure(struct bridge *); | |
232 | static void mirror_reconfigure_one(struct mirror *); | |
233 | static bool vlan_is_mirrored(const struct mirror *, int vlan); | |
234 | ||
235 | static void brstp_reconfigure(struct bridge *); | |
236 | static void brstp_adjust_timers(struct bridge *); | |
237 | static void brstp_run(struct bridge *); | |
238 | static void brstp_wait(struct bridge *); | |
239 | ||
240 | static void iface_create(struct port *, const char *name); | |
241 | static void iface_destroy(struct iface *); | |
242 | static struct iface *iface_lookup(const struct bridge *, const char *name); | |
243 | static struct iface *iface_from_dp_ifidx(const struct bridge *, | |
244 | uint16_t dp_ifidx); | |
557d8e6c | 245 | static bool iface_is_internal(const struct bridge *, const char *name); |
52df17e7 | 246 | static void iface_set_mac(struct iface *); |
064af421 BP |
247 | |
248 | /* Hooks into ofproto processing. */ | |
249 | static struct ofhooks bridge_ofhooks; | |
250 | \f | |
251 | /* Public functions. */ | |
252 | ||
253 | /* Adds the name of each interface used by a bridge, including local and | |
254 | * internal ports, to 'svec'. */ | |
255 | void | |
256 | bridge_get_ifaces(struct svec *svec) | |
257 | { | |
258 | struct bridge *br, *next; | |
259 | size_t i, j; | |
260 | ||
261 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { | |
262 | for (i = 0; i < br->n_ports; i++) { | |
263 | struct port *port = br->ports[i]; | |
264 | ||
265 | for (j = 0; j < port->n_ifaces; j++) { | |
266 | struct iface *iface = port->ifaces[j]; | |
267 | if (iface->dp_ifidx < 0) { | |
268 | VLOG_ERR("%s interface not in dp%u, ignoring", | |
269 | iface->name, dpif_id(&br->dpif)); | |
270 | } else { | |
271 | if (iface->dp_ifidx != ODPP_LOCAL) { | |
272 | svec_add(svec, iface->name); | |
273 | } | |
274 | } | |
275 | } | |
276 | } | |
277 | } | |
278 | } | |
279 | ||
280 | /* The caller must already have called cfg_read(). */ | |
281 | void | |
282 | bridge_init(void) | |
283 | { | |
284 | int retval; | |
285 | int i; | |
286 | ||
da285df4 BP |
287 | bond_init(); |
288 | ||
8c4c1387 BP |
289 | unixctl_command_register("fdb/show", bridge_unixctl_fdb_show); |
290 | ||
064af421 BP |
291 | for (i = 0; i < DP_MAX; i++) { |
292 | struct dpif dpif; | |
293 | char devname[16]; | |
294 | ||
295 | sprintf(devname, "dp%d", i); | |
296 | retval = dpif_open(devname, &dpif); | |
297 | if (!retval) { | |
298 | char dpif_name[IF_NAMESIZE]; | |
299 | if (dpif_get_name(&dpif, dpif_name, sizeof dpif_name) | |
300 | || !cfg_has("bridge.%s.port", dpif_name)) { | |
301 | dpif_delete(&dpif); | |
302 | } | |
303 | dpif_close(&dpif); | |
304 | } else if (retval != ENODEV) { | |
305 | VLOG_ERR("failed to delete datapath dp%d: %s", | |
306 | i, strerror(retval)); | |
307 | } | |
308 | } | |
309 | ||
4f2cad2c JP |
310 | unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows); |
311 | ||
064af421 BP |
312 | bridge_reconfigure(); |
313 | } | |
314 | ||
315 | #ifdef HAVE_OPENSSL | |
316 | static bool | |
317 | config_string_change(const char *key, char **valuep) | |
318 | { | |
319 | const char *value = cfg_get_string(0, "%s", key); | |
320 | if (value && (!*valuep || strcmp(value, *valuep))) { | |
321 | free(*valuep); | |
322 | *valuep = xstrdup(value); | |
323 | return true; | |
324 | } else { | |
325 | return false; | |
326 | } | |
327 | } | |
328 | ||
329 | static void | |
330 | bridge_configure_ssl(void) | |
331 | { | |
332 | /* XXX SSL should be configurable on a per-bridge basis. | |
333 | * XXX should be possible to de-configure SSL. */ | |
334 | static char *private_key_file; | |
335 | static char *certificate_file; | |
336 | static char *cacert_file; | |
7e40e21d | 337 | struct stat s; |
064af421 BP |
338 | |
339 | if (config_string_change("ssl.private-key", &private_key_file)) { | |
340 | vconn_ssl_set_private_key_file(private_key_file); | |
341 | } | |
342 | ||
343 | if (config_string_change("ssl.certificate", &certificate_file)) { | |
344 | vconn_ssl_set_certificate_file(certificate_file); | |
345 | } | |
346 | ||
7e40e21d JP |
347 | /* We assume that even if the filename hasn't changed, if the CA cert |
348 | * file has been removed, that we want to move back into | |
349 | * boot-strapping mode. This opens a small security hole, because | |
350 | * the old certificate will still be trusted until vSwitch is | |
351 | * restarted. We may want to address this in vconn's SSL library. */ | |
352 | if (config_string_change("ssl.ca-cert", &cacert_file) | |
353 | || (stat(cacert_file, &s) && errno == ENOENT)) { | |
064af421 BP |
354 | vconn_ssl_set_ca_cert_file(cacert_file, |
355 | cfg_get_bool(0, "ssl.bootstrap-ca-cert")); | |
356 | } | |
357 | } | |
358 | #endif | |
359 | ||
360 | void | |
361 | bridge_reconfigure(void) | |
362 | { | |
363 | struct svec old_br, new_br, raw_new_br; | |
364 | struct bridge *br, *next; | |
365 | size_t i, j; | |
366 | ||
367 | COVERAGE_INC(bridge_reconfigure); | |
368 | ||
369 | /* Collect old bridges. */ | |
370 | svec_init(&old_br); | |
371 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
372 | svec_add(&old_br, br->name); | |
373 | } | |
374 | ||
375 | /* Collect new bridges. */ | |
376 | svec_init(&raw_new_br); | |
377 | cfg_get_subsections(&raw_new_br, "bridge"); | |
378 | svec_init(&new_br); | |
379 | for (i = 0; i < raw_new_br.n; i++) { | |
380 | const char *name = raw_new_br.names[i]; | |
381 | if ((!strncmp(name, "dp", 2) && isdigit(name[2])) || | |
382 | (!strncmp(name, "nl:", 3) && isdigit(name[3]))) { | |
383 | VLOG_ERR("%s is not a valid bridge name (bridges may not be " | |
384 | "named \"dp\" or \"nl:\" followed by a digit)", name); | |
385 | } else { | |
386 | svec_add(&new_br, name); | |
387 | } | |
388 | } | |
389 | svec_destroy(&raw_new_br); | |
390 | ||
391 | /* Get rid of deleted bridges and add new bridges. */ | |
392 | svec_sort(&old_br); | |
393 | svec_sort(&new_br); | |
394 | assert(svec_is_unique(&old_br)); | |
395 | assert(svec_is_unique(&new_br)); | |
396 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { | |
397 | if (!svec_contains(&new_br, br->name)) { | |
398 | bridge_destroy(br); | |
399 | } | |
400 | } | |
401 | for (i = 0; i < new_br.n; i++) { | |
402 | const char *name = new_br.names[i]; | |
403 | if (!svec_contains(&old_br, name)) { | |
404 | bridge_create(name); | |
405 | } | |
406 | } | |
407 | svec_destroy(&old_br); | |
408 | svec_destroy(&new_br); | |
409 | ||
410 | #ifdef HAVE_OPENSSL | |
411 | /* Configure SSL. */ | |
412 | bridge_configure_ssl(); | |
413 | #endif | |
414 | ||
415 | /* Reconfigure all bridges. */ | |
416 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
417 | bridge_reconfigure_one(br); | |
418 | } | |
419 | ||
420 | /* Add and delete ports on all datapaths. | |
421 | * | |
422 | * The kernel will reject any attempt to add a given port to a datapath if | |
423 | * that port already belongs to a different datapath, so we must do all | |
424 | * port deletions before any port additions. */ | |
425 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
426 | struct odp_port *dpif_ports; | |
427 | size_t n_dpif_ports; | |
428 | struct svec want_ifaces; | |
429 | ||
430 | dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); | |
431 | bridge_get_all_ifaces(br, &want_ifaces); | |
432 | for (i = 0; i < n_dpif_ports; i++) { | |
433 | const struct odp_port *p = &dpif_ports[i]; | |
434 | if (!svec_contains(&want_ifaces, p->devname) | |
435 | && strcmp(p->devname, br->name)) { | |
436 | int retval = dpif_port_del(&br->dpif, p->port); | |
437 | if (retval) { | |
438 | VLOG_ERR("failed to remove %s interface from dp%u: %s", | |
439 | p->devname, dpif_id(&br->dpif), strerror(retval)); | |
440 | } | |
441 | } | |
442 | } | |
443 | svec_destroy(&want_ifaces); | |
444 | free(dpif_ports); | |
445 | } | |
446 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
447 | struct odp_port *dpif_ports; | |
448 | size_t n_dpif_ports; | |
449 | struct svec cur_ifaces, want_ifaces, add_ifaces; | |
450 | int next_port_no; | |
451 | ||
452 | dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); | |
453 | svec_init(&cur_ifaces); | |
454 | for (i = 0; i < n_dpif_ports; i++) { | |
455 | svec_add(&cur_ifaces, dpif_ports[i].devname); | |
456 | } | |
457 | free(dpif_ports); | |
458 | svec_sort_unique(&cur_ifaces); | |
459 | bridge_get_all_ifaces(br, &want_ifaces); | |
460 | svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL); | |
461 | ||
462 | next_port_no = 1; | |
463 | for (i = 0; i < add_ifaces.n; i++) { | |
464 | const char *if_name = add_ifaces.names[i]; | |
465 | for (;;) { | |
35c979bf BP |
466 | bool internal; |
467 | int error; | |
468 | ||
35c979bf | 469 | /* Add to datapath. */ |
557d8e6c | 470 | internal = iface_is_internal(br, if_name); |
35c979bf BP |
471 | error = dpif_port_add(&br->dpif, if_name, next_port_no++, |
472 | internal ? ODP_PORT_INTERNAL : 0); | |
064af421 BP |
473 | if (error != EEXIST) { |
474 | if (next_port_no >= 256) { | |
475 | VLOG_ERR("ran out of valid port numbers on dp%u", | |
476 | dpif_id(&br->dpif)); | |
477 | goto out; | |
478 | } | |
479 | if (error) { | |
480 | VLOG_ERR("failed to add %s interface to dp%u: %s", | |
481 | if_name, dpif_id(&br->dpif), strerror(error)); | |
482 | } | |
483 | break; | |
484 | } | |
485 | } | |
486 | } | |
487 | out: | |
488 | svec_destroy(&cur_ifaces); | |
489 | svec_destroy(&want_ifaces); | |
490 | svec_destroy(&add_ifaces); | |
491 | } | |
492 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
493 | uint8_t ea[8]; | |
494 | uint64_t dpid; | |
495 | struct iface *local_iface = NULL; | |
496 | const char *devname; | |
497 | uint8_t engine_type = br->dpif.minor; | |
498 | uint8_t engine_id = br->dpif.minor; | |
499 | bool add_id_to_iface = false; | |
500 | struct svec nf_hosts; | |
501 | ||
064af421 BP |
502 | bridge_fetch_dp_ifaces(br); |
503 | for (i = 0; i < br->n_ports; ) { | |
504 | struct port *port = br->ports[i]; | |
505 | ||
506 | for (j = 0; j < port->n_ifaces; ) { | |
507 | struct iface *iface = port->ifaces[j]; | |
508 | if (iface->dp_ifidx < 0) { | |
509 | VLOG_ERR("%s interface not in dp%u, dropping", | |
510 | iface->name, dpif_id(&br->dpif)); | |
511 | iface_destroy(iface); | |
512 | } else { | |
513 | if (iface->dp_ifidx == ODPP_LOCAL) { | |
514 | local_iface = iface; | |
515 | } | |
516 | VLOG_DBG("dp%u has interface %s on port %d", | |
517 | dpif_id(&br->dpif), iface->name, iface->dp_ifidx); | |
518 | j++; | |
519 | } | |
520 | } | |
521 | if (!port->n_ifaces) { | |
522 | VLOG_ERR("%s port has no interfaces, dropping", port->name); | |
523 | port_destroy(port); | |
524 | continue; | |
525 | } | |
526 | i++; | |
527 | } | |
528 | ||
529 | /* Pick local port hardware address, datapath ID. */ | |
530 | bridge_pick_local_hw_addr(br, ea, &devname); | |
531 | if (local_iface) { | |
532 | int error = netdev_nodev_set_etheraddr(local_iface->name, ea); | |
533 | if (error) { | |
534 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
535 | VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " | |
536 | "Ethernet address: %s", | |
537 | br->name, strerror(error)); | |
538 | } | |
539 | } | |
540 | ||
541 | dpid = bridge_pick_datapath_id(br, ea, devname); | |
542 | ofproto_set_datapath_id(br->ofproto, dpid); | |
543 | ||
544 | /* Set NetFlow configuration on this bridge. */ | |
545 | if (cfg_has("netflow.%s.engine-type", br->name)) { | |
546 | engine_type = cfg_get_int(0, "netflow.%s.engine-type", | |
547 | br->name); | |
548 | } | |
549 | if (cfg_has("netflow.%s.engine-id", br->name)) { | |
550 | engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name); | |
551 | } | |
552 | if (cfg_has("netflow.%s.add-id-to-iface", br->name)) { | |
553 | add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface", | |
554 | br->name); | |
555 | } | |
556 | if (add_id_to_iface && engine_id > 0x7f) { | |
557 | VLOG_WARN("bridge %s: netflow port mangling may conflict with " | |
558 | "another vswitch, choose an engine id less than 128", | |
559 | br->name); | |
560 | } | |
561 | if (add_id_to_iface && br->n_ports > 0x1ff) { | |
562 | VLOG_WARN("bridge %s: netflow port mangling will conflict with " | |
563 | "another port when 512 or more ports are used", | |
564 | br->name); | |
565 | } | |
566 | svec_init(&nf_hosts); | |
567 | cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name); | |
568 | if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type, | |
569 | engine_id, add_id_to_iface)) { | |
570 | VLOG_ERR("bridge %s: problem setting netflow collectors", | |
571 | br->name); | |
572 | } | |
c77dc770 | 573 | svec_destroy(&nf_hosts); |
064af421 BP |
574 | |
575 | /* Update the controller and related settings. It would be more | |
576 | * straightforward to call this from bridge_reconfigure_one(), but we | |
577 | * can't do it there for two reasons. First, and most importantly, at | |
578 | * that point we don't know the dp_ifidx of any interfaces that have | |
579 | * been added to the bridge (because we haven't actually added them to | |
580 | * the datapath). Second, at that point we haven't set the datapath ID | |
581 | * yet; when a controller is configured, resetting the datapath ID will | |
582 | * immediately disconnect from the controller, so it's better to set | |
583 | * the datapath ID before the controller. */ | |
584 | bridge_reconfigure_controller(br); | |
585 | } | |
586 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
587 | for (i = 0; i < br->n_ports; i++) { | |
588 | struct port *port = br->ports[i]; | |
52df17e7 | 589 | |
064af421 | 590 | port_update_vlan_compat(port); |
52df17e7 BP |
591 | |
592 | for (j = 0; j < port->n_ifaces; j++) { | |
593 | struct iface *iface = port->ifaces[j]; | |
594 | if (iface->dp_ifidx != ODPP_LOCAL | |
595 | && iface_is_internal(br, iface->name)) { | |
596 | iface_set_mac(iface); | |
597 | } | |
598 | } | |
064af421 BP |
599 | } |
600 | } | |
601 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
602 | brstp_reconfigure(br); | |
603 | } | |
604 | } | |
605 | ||
606 | static void | |
607 | bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], | |
608 | const char **devname) | |
609 | { | |
610 | uint64_t requested_ea; | |
611 | size_t i, j; | |
612 | int error; | |
613 | ||
614 | *devname = NULL; | |
615 | ||
616 | /* Did the user request a particular MAC? */ | |
617 | requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name); | |
618 | if (requested_ea) { | |
619 | eth_addr_from_uint64(requested_ea, ea); | |
620 | if (eth_addr_is_multicast(ea)) { | |
621 | VLOG_ERR("bridge %s: cannot set MAC address to multicast " | |
622 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
623 | } else if (eth_addr_is_zero(ea)) { | |
624 | VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name); | |
625 | } else { | |
626 | return; | |
627 | } | |
628 | } | |
629 | ||
630 | /* Otherwise choose the minimum MAC address among all of the interfaces. | |
631 | * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the | |
632 | * MAC of the physical interface in such an environment.) */ | |
633 | memset(ea, 0xff, sizeof ea); | |
634 | for (i = 0; i < br->n_ports; i++) { | |
635 | struct port *port = br->ports[i]; | |
58b7527e BP |
636 | uint8_t iface_ea[ETH_ADDR_LEN]; |
637 | uint64_t iface_ea_u64; | |
638 | struct iface *iface; | |
639 | ||
640 | /* Mirror output ports don't participate. */ | |
064af421 BP |
641 | if (port->is_mirror_output_port) { |
642 | continue; | |
643 | } | |
58b7527e BP |
644 | |
645 | /* Choose the MAC address to represent the port. */ | |
646 | iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name); | |
647 | if (iface_ea_u64) { | |
648 | /* User specified explicitly. */ | |
649 | eth_addr_from_uint64(iface_ea_u64, iface_ea); | |
ba09980a BP |
650 | |
651 | /* Find the interface with this Ethernet address (if any) so that | |
652 | * we can provide the correct devname to the caller. */ | |
653 | iface = NULL; | |
654 | for (j = 0; j < port->n_ifaces; j++) { | |
655 | struct iface *candidate = port->ifaces[j]; | |
656 | uint8_t candidate_ea[ETH_ADDR_LEN]; | |
657 | if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea) | |
658 | && eth_addr_equals(iface_ea, candidate_ea)) { | |
659 | iface = candidate; | |
660 | } | |
661 | } | |
58b7527e BP |
662 | } else { |
663 | /* Choose the interface whose MAC address will represent the port. | |
664 | * The Linux kernel bonding code always chooses the MAC address of | |
665 | * the first slave added to a bond, and the Fedora networking | |
666 | * scripts always add slaves to a bond in alphabetical order, so | |
667 | * for compatibility we choose the interface with the name that is | |
668 | * first in alphabetical order. */ | |
669 | iface = port->ifaces[0]; | |
670 | for (j = 1; j < port->n_ifaces; j++) { | |
671 | struct iface *candidate = port->ifaces[j]; | |
672 | if (strcmp(candidate->name, iface->name) < 0) { | |
673 | iface = candidate; | |
674 | } | |
675 | } | |
676 | ||
677 | /* The local port doesn't count (since we're trying to choose its | |
678 | * MAC address anyway). Other internal ports don't count because | |
679 | * we really want a physical MAC if we can get it, and internal | |
680 | * ports typically have randomly generated MACs. */ | |
064af421 BP |
681 | if (iface->dp_ifidx == ODPP_LOCAL |
682 | || cfg_get_bool(0, "iface.%s.internal", iface->name)) { | |
683 | continue; | |
684 | } | |
58b7527e BP |
685 | |
686 | /* Grab MAC. */ | |
064af421 | 687 | error = netdev_nodev_get_etheraddr(iface->name, iface_ea); |
58b7527e | 688 | if (error) { |
064af421 BP |
689 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
690 | VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s", | |
691 | iface->name, strerror(error)); | |
58b7527e | 692 | continue; |
064af421 BP |
693 | } |
694 | } | |
58b7527e BP |
695 | |
696 | /* Compare against our current choice. */ | |
697 | if (!eth_addr_is_multicast(iface_ea) && | |
698 | !eth_addr_is_reserved(iface_ea) && | |
699 | !eth_addr_is_zero(iface_ea) && | |
700 | memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) | |
701 | { | |
702 | memcpy(ea, iface_ea, ETH_ADDR_LEN); | |
ba09980a | 703 | *devname = iface ? iface->name : NULL; |
58b7527e | 704 | } |
064af421 BP |
705 | } |
706 | if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) { | |
707 | memcpy(ea, br->default_ea, ETH_ADDR_LEN); | |
708 | *devname = NULL; | |
709 | VLOG_WARN("bridge %s: using default bridge Ethernet " | |
710 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
711 | } else { | |
712 | VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT, | |
713 | br->name, ETH_ADDR_ARGS(ea)); | |
714 | } | |
715 | } | |
716 | ||
717 | /* Choose and returns the datapath ID for bridge 'br' given that the bridge | |
718 | * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of | |
719 | * a network device, then that network device's name must be passed in as | |
720 | * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be | |
721 | * passed in as a null pointer. */ | |
722 | static uint64_t | |
723 | bridge_pick_datapath_id(struct bridge *br, | |
724 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
725 | const char *devname) | |
726 | { | |
727 | /* | |
728 | * The procedure for choosing a bridge MAC address will, in the most | |
729 | * ordinary case, also choose a unique MAC that we can use as a datapath | |
730 | * ID. In some special cases, though, multiple bridges will end up with | |
731 | * the same MAC address. This is OK for the bridges, but it will confuse | |
732 | * the OpenFlow controller, because each datapath needs a unique datapath | |
733 | * ID. | |
734 | * | |
735 | * Datapath IDs must be unique. It is also very desirable that they be | |
736 | * stable from one run to the next, so that policy set on a datapath | |
737 | * "sticks". | |
738 | */ | |
739 | uint64_t dpid; | |
740 | ||
741 | dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name); | |
742 | if (dpid) { | |
743 | return dpid; | |
744 | } | |
745 | ||
746 | if (devname) { | |
747 | int vlan; | |
748 | if (!netdev_get_vlan_vid(devname, &vlan)) { | |
749 | /* | |
750 | * A bridge whose MAC address is taken from a VLAN network device | |
751 | * (that is, a network device created with vconfig(8) or similar | |
752 | * tool) will have the same MAC address as a bridge on the VLAN | |
753 | * device's physical network device. | |
754 | * | |
755 | * Handle this case by hashing the physical network device MAC | |
756 | * along with the VLAN identifier. | |
757 | */ | |
758 | uint8_t buf[ETH_ADDR_LEN + 2]; | |
759 | memcpy(buf, bridge_ea, ETH_ADDR_LEN); | |
760 | buf[ETH_ADDR_LEN] = vlan >> 8; | |
761 | buf[ETH_ADDR_LEN + 1] = vlan; | |
762 | return dpid_from_hash(buf, sizeof buf); | |
763 | } else { | |
764 | /* | |
765 | * Assume that this bridge's MAC address is unique, since it | |
766 | * doesn't fit any of the cases we handle specially. | |
767 | */ | |
768 | } | |
769 | } else { | |
770 | /* | |
771 | * A purely internal bridge, that is, one that has no non-virtual | |
772 | * network devices on it at all, is more difficult because it has no | |
773 | * natural unique identifier at all. | |
774 | * | |
775 | * When the host is a XenServer, we handle this case by hashing the | |
776 | * host's UUID with the name of the bridge. Names of bridges are | |
777 | * persistent across XenServer reboots, although they can be reused if | |
778 | * an internal network is destroyed and then a new one is later | |
779 | * created, so this is fairly effective. | |
780 | * | |
781 | * When the host is not a XenServer, we punt by using a random MAC | |
782 | * address on each run. | |
783 | */ | |
784 | const char *host_uuid = xenserver_get_host_uuid(); | |
785 | if (host_uuid) { | |
786 | char *combined = xasprintf("%s,%s", host_uuid, br->name); | |
787 | dpid = dpid_from_hash(combined, strlen(combined)); | |
788 | free(combined); | |
789 | return dpid; | |
790 | } | |
791 | } | |
792 | ||
793 | return eth_addr_to_uint64(bridge_ea); | |
794 | } | |
795 | ||
796 | static uint64_t | |
797 | dpid_from_hash(const void *data, size_t n) | |
798 | { | |
5eccf359 | 799 | uint8_t hash[SHA1_DIGEST_SIZE]; |
064af421 BP |
800 | |
801 | BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN); | |
5eccf359 | 802 | sha1_bytes(data, n, hash); |
064af421 BP |
803 | eth_addr_mark_random(hash); |
804 | return eth_addr_to_uint64(hash); | |
805 | } | |
806 | ||
807 | int | |
808 | bridge_run(void) | |
809 | { | |
810 | struct bridge *br, *next; | |
811 | int retval; | |
812 | ||
813 | retval = 0; | |
814 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { | |
815 | int error = bridge_run_one(br); | |
816 | if (error) { | |
817 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
818 | VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " | |
819 | "forcing reconfiguration", br->name); | |
820 | if (!retval) { | |
821 | retval = error; | |
822 | } | |
823 | } | |
824 | } | |
825 | return retval; | |
826 | } | |
827 | ||
828 | void | |
829 | bridge_wait(void) | |
830 | { | |
831 | struct bridge *br; | |
832 | ||
833 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
834 | ofproto_wait(br->ofproto); | |
835 | if (br->controller) { | |
836 | continue; | |
837 | } | |
838 | ||
839 | if (br->ml) { | |
840 | mac_learning_wait(br->ml); | |
841 | } | |
842 | bond_wait(br); | |
843 | brstp_wait(br); | |
844 | } | |
845 | } | |
846 | ||
847 | /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s | |
848 | * configuration changes. */ | |
849 | static void | |
850 | bridge_flush(struct bridge *br) | |
851 | { | |
852 | COVERAGE_INC(bridge_flush); | |
853 | br->flush = true; | |
854 | if (br->ml) { | |
855 | mac_learning_flush(br->ml); | |
856 | } | |
857 | } | |
858 | \f | |
8c4c1387 BP |
859 | /* Bridge unixctl user interface functions. */ |
860 | static void | |
861 | bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args) | |
862 | { | |
863 | struct ds ds = DS_EMPTY_INITIALIZER; | |
864 | const struct bridge *br; | |
865 | ||
866 | br = bridge_lookup(args); | |
867 | if (!br) { | |
868 | unixctl_command_reply(conn, 501, "no such bridge"); | |
869 | return; | |
870 | } | |
871 | ||
872 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
873 | if (br->ml) { | |
874 | const struct mac_entry *e; | |
875 | LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { | |
092bebdc BP |
876 | if (e->port < 0 || e->port >= br->n_ports) { |
877 | continue; | |
878 | } | |
8c4c1387 | 879 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", |
092bebdc BP |
880 | br->ports[e->port]->ifaces[0]->dp_ifidx, |
881 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
8c4c1387 BP |
882 | } |
883 | } | |
884 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
885 | ds_destroy(&ds); | |
886 | } | |
887 | \f | |
064af421 BP |
888 | /* Bridge reconfiguration functions. */ |
889 | ||
890 | static struct bridge * | |
891 | bridge_create(const char *name) | |
892 | { | |
893 | struct bridge *br; | |
894 | int error; | |
895 | ||
896 | assert(!bridge_lookup(name)); | |
897 | br = xcalloc(1, sizeof *br); | |
898 | ||
899 | error = dpif_create(name, &br->dpif); | |
900 | if (error == EEXIST) { | |
901 | error = dpif_open(name, &br->dpif); | |
902 | if (error) { | |
903 | VLOG_ERR("datapath %s already exists but cannot be opened: %s", | |
904 | name, strerror(error)); | |
905 | free(br); | |
906 | return NULL; | |
907 | } | |
908 | dpif_flow_flush(&br->dpif); | |
909 | } else if (error) { | |
910 | VLOG_ERR("failed to create datapath %s: %s", name, strerror(error)); | |
911 | free(br); | |
912 | return NULL; | |
913 | } | |
914 | ||
915 | error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); | |
916 | if (error) { | |
917 | VLOG_ERR("failed to create switch %s: %s", name, strerror(error)); | |
918 | dpif_delete(&br->dpif); | |
919 | dpif_close(&br->dpif); | |
920 | free(br); | |
921 | return NULL; | |
922 | } | |
923 | ||
924 | br->name = xstrdup(name); | |
925 | br->ml = mac_learning_create(); | |
926 | br->sent_config_request = false; | |
927 | eth_addr_random(br->default_ea); | |
928 | ||
929 | port_array_init(&br->ifaces); | |
930 | ||
931 | br->flush = false; | |
932 | br->bond_next_rebalance = time_msec() + 10000; | |
933 | ||
934 | list_push_back(&all_bridges, &br->node); | |
935 | ||
936 | VLOG_INFO("created bridge %s on dp%u", br->name, dpif_id(&br->dpif)); | |
937 | ||
938 | return br; | |
939 | } | |
940 | ||
941 | static void | |
942 | bridge_destroy(struct bridge *br) | |
943 | { | |
944 | if (br) { | |
945 | int error; | |
946 | ||
947 | while (br->n_ports > 0) { | |
948 | port_destroy(br->ports[br->n_ports - 1]); | |
949 | } | |
950 | list_remove(&br->node); | |
951 | error = dpif_delete(&br->dpif); | |
952 | if (error && error != ENOENT) { | |
953 | VLOG_ERR("failed to delete dp%u: %s", | |
954 | dpif_id(&br->dpif), strerror(error)); | |
955 | } | |
956 | dpif_close(&br->dpif); | |
957 | ofproto_destroy(br->ofproto); | |
958 | free(br->controller); | |
959 | mac_learning_destroy(br->ml); | |
960 | port_array_destroy(&br->ifaces); | |
961 | free(br->ports); | |
962 | free(br->name); | |
963 | free(br); | |
964 | } | |
965 | } | |
966 | ||
967 | static struct bridge * | |
968 | bridge_lookup(const char *name) | |
969 | { | |
970 | struct bridge *br; | |
971 | ||
972 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
973 | if (!strcmp(br->name, name)) { | |
974 | return br; | |
975 | } | |
976 | } | |
977 | return NULL; | |
978 | } | |
979 | ||
980 | bool | |
981 | bridge_exists(const char *name) | |
982 | { | |
983 | return bridge_lookup(name) ? true : false; | |
984 | } | |
985 | ||
986 | uint64_t | |
987 | bridge_get_datapathid(const char *name) | |
988 | { | |
989 | struct bridge *br = bridge_lookup(name); | |
990 | return br ? ofproto_get_datapath_id(br->ofproto) : 0; | |
991 | } | |
992 | ||
4f2cad2c JP |
993 | /* Handle requests for a listing of all flows known by the OpenFlow |
994 | * stack, including those normally hidden. */ | |
995 | static void | |
996 | bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args) | |
997 | { | |
998 | struct bridge *br; | |
999 | struct ds results; | |
1000 | ||
1001 | br = bridge_lookup(args); | |
1002 | if (!br) { | |
1003 | unixctl_command_reply(conn, 501, "Unknown bridge"); | |
1004 | return; | |
1005 | } | |
1006 | ||
1007 | ds_init(&results); | |
1008 | ofproto_get_all_flows(br->ofproto, &results); | |
1009 | ||
1010 | unixctl_command_reply(conn, 200, ds_cstr(&results)); | |
1011 | ds_destroy(&results); | |
1012 | } | |
1013 | ||
064af421 BP |
1014 | static int |
1015 | bridge_run_one(struct bridge *br) | |
1016 | { | |
1017 | int error; | |
1018 | ||
1019 | error = ofproto_run1(br->ofproto); | |
1020 | if (error) { | |
1021 | return error; | |
1022 | } | |
1023 | ||
1024 | if (br->ml) { | |
1025 | mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); | |
1026 | } | |
1027 | bond_run(br); | |
1028 | brstp_run(br); | |
1029 | ||
1030 | error = ofproto_run2(br->ofproto, br->flush); | |
1031 | br->flush = false; | |
1032 | ||
1033 | return error; | |
1034 | } | |
1035 | ||
1036 | static const char * | |
1037 | bridge_get_controller(const struct bridge *br) | |
1038 | { | |
1039 | const char *controller; | |
1040 | ||
1041 | controller = cfg_get_string(0, "bridge.%s.controller", br->name); | |
1042 | if (!controller) { | |
1043 | controller = cfg_get_string(0, "mgmt.controller"); | |
1044 | } | |
1045 | return controller && controller[0] ? controller : NULL; | |
1046 | } | |
1047 | ||
1048 | static void | |
1049 | bridge_reconfigure_one(struct bridge *br) | |
1050 | { | |
1051 | struct svec old_ports, new_ports, ifaces; | |
1052 | struct svec listeners, old_listeners; | |
1053 | struct svec snoops, old_snoops; | |
1054 | size_t i, j; | |
1055 | ||
1056 | /* Collect old ports. */ | |
1057 | svec_init(&old_ports); | |
1058 | for (i = 0; i < br->n_ports; i++) { | |
1059 | svec_add(&old_ports, br->ports[i]->name); | |
1060 | } | |
1061 | svec_sort(&old_ports); | |
1062 | assert(svec_is_unique(&old_ports)); | |
1063 | ||
1064 | /* Collect new ports. */ | |
1065 | svec_init(&new_ports); | |
1066 | cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name); | |
1067 | svec_sort(&new_ports); | |
1068 | if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) { | |
1069 | svec_add(&new_ports, br->name); | |
1070 | svec_sort(&new_ports); | |
1071 | } | |
1072 | if (!svec_is_unique(&new_ports)) { | |
1073 | VLOG_WARN("bridge %s: %s specified twice as bridge port", | |
1074 | br->name, svec_get_duplicate(&new_ports)); | |
1075 | svec_unique(&new_ports); | |
1076 | } | |
1077 | ||
1078 | ofproto_set_mgmt_id(br->ofproto, mgmt_id); | |
1079 | ||
1080 | /* Get rid of deleted ports and add new ports. */ | |
1081 | for (i = 0; i < br->n_ports; ) { | |
1082 | struct port *port = br->ports[i]; | |
1083 | if (!svec_contains(&new_ports, port->name)) { | |
1084 | port_destroy(port); | |
1085 | } else { | |
1086 | i++; | |
1087 | } | |
1088 | } | |
1089 | for (i = 0; i < new_ports.n; i++) { | |
1090 | const char *name = new_ports.names[i]; | |
1091 | if (!svec_contains(&old_ports, name)) { | |
1092 | port_create(br, name); | |
1093 | } | |
1094 | } | |
1095 | svec_destroy(&old_ports); | |
1096 | svec_destroy(&new_ports); | |
1097 | ||
1098 | /* Reconfigure all ports. */ | |
1099 | for (i = 0; i < br->n_ports; i++) { | |
1100 | port_reconfigure(br->ports[i]); | |
1101 | } | |
1102 | ||
1103 | /* Check and delete duplicate interfaces. */ | |
1104 | svec_init(&ifaces); | |
1105 | for (i = 0; i < br->n_ports; ) { | |
1106 | struct port *port = br->ports[i]; | |
1107 | for (j = 0; j < port->n_ifaces; ) { | |
1108 | struct iface *iface = port->ifaces[j]; | |
1109 | if (svec_contains(&ifaces, iface->name)) { | |
1110 | VLOG_ERR("bridge %s: %s interface is on multiple ports, " | |
1111 | "removing from %s", | |
1112 | br->name, iface->name, port->name); | |
1113 | iface_destroy(iface); | |
1114 | } else { | |
1115 | svec_add(&ifaces, iface->name); | |
1116 | svec_sort(&ifaces); | |
1117 | j++; | |
1118 | } | |
1119 | } | |
1120 | if (!port->n_ifaces) { | |
1121 | VLOG_ERR("%s port has no interfaces, dropping", port->name); | |
1122 | port_destroy(port); | |
1123 | } else { | |
1124 | i++; | |
1125 | } | |
1126 | } | |
1127 | svec_destroy(&ifaces); | |
1128 | ||
1129 | /* Delete all flows if we're switching from connected to standalone or vice | |
1130 | * versa. (XXX Should we delete all flows if we are switching from one | |
1131 | * controller to another?) */ | |
1132 | ||
1133 | /* Configure OpenFlow management listeners. */ | |
1134 | svec_init(&listeners); | |
1135 | cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); | |
1136 | if (!listeners.n) { | |
1137 | svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", | |
1138 | ovs_rundir, br->name)); | |
1139 | } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { | |
1140 | svec_clear(&listeners); | |
1141 | } | |
1142 | svec_sort_unique(&listeners); | |
1143 | ||
1144 | svec_init(&old_listeners); | |
1145 | ofproto_get_listeners(br->ofproto, &old_listeners); | |
1146 | svec_sort_unique(&old_listeners); | |
1147 | ||
1148 | if (!svec_equal(&listeners, &old_listeners)) { | |
1149 | ofproto_set_listeners(br->ofproto, &listeners); | |
1150 | } | |
1151 | svec_destroy(&listeners); | |
1152 | svec_destroy(&old_listeners); | |
1153 | ||
1154 | /* Configure OpenFlow controller connection snooping. */ | |
1155 | svec_init(&snoops); | |
1156 | cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); | |
1157 | if (!snoops.n) { | |
1158 | svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", | |
1159 | ovs_rundir, br->name)); | |
1160 | } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { | |
1161 | svec_clear(&snoops); | |
1162 | } | |
1163 | svec_sort_unique(&snoops); | |
1164 | ||
1165 | svec_init(&old_snoops); | |
1166 | ofproto_get_snoops(br->ofproto, &old_snoops); | |
1167 | svec_sort_unique(&old_snoops); | |
1168 | ||
1169 | if (!svec_equal(&snoops, &old_snoops)) { | |
1170 | ofproto_set_snoops(br->ofproto, &snoops); | |
1171 | } | |
1172 | svec_destroy(&snoops); | |
1173 | svec_destroy(&old_snoops); | |
1174 | ||
1175 | mirror_reconfigure(br); | |
1176 | } | |
1177 | ||
1178 | static void | |
1179 | bridge_reconfigure_controller(struct bridge *br) | |
1180 | { | |
1181 | char *pfx = xasprintf("bridge.%s.controller", br->name); | |
1182 | const char *controller; | |
1183 | ||
1184 | controller = bridge_get_controller(br); | |
1185 | if ((br->controller != NULL) != (controller != NULL)) { | |
1186 | ofproto_flush_flows(br->ofproto); | |
1187 | } | |
1188 | free(br->controller); | |
1189 | br->controller = controller ? xstrdup(controller) : NULL; | |
1190 | ||
1191 | if (controller) { | |
1192 | const char *fail_mode; | |
1193 | int max_backoff, probe; | |
1194 | int rate_limit, burst_limit; | |
1195 | ||
1196 | if (!strcmp(controller, "discover")) { | |
a1525581 JP |
1197 | bool update_resolv_conf = true; |
1198 | ||
1199 | if (cfg_has("%s.update-resolv.conf", pfx)) { | |
fef94d15 JP |
1200 | update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf", |
1201 | pfx); | |
a1525581 | 1202 | } |
064af421 BP |
1203 | ofproto_set_discovery(br->ofproto, true, |
1204 | cfg_get_string(0, "%s.accept-regex", pfx), | |
a1525581 | 1205 | update_resolv_conf); |
064af421 BP |
1206 | } else { |
1207 | struct netdev *netdev; | |
1208 | bool in_band; | |
1209 | int error; | |
1210 | ||
1211 | in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, | |
1212 | "%s.in-band", pfx) | |
1213 | || cfg_get_bool(0, "%s.in-band", pfx)); | |
1214 | ofproto_set_discovery(br->ofproto, false, NULL, NULL); | |
1215 | ofproto_set_in_band(br->ofproto, in_band); | |
1216 | ||
1217 | error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev); | |
1218 | if (!error) { | |
1219 | if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) { | |
1220 | struct in_addr ip, mask, gateway; | |
1221 | ip.s_addr = cfg_get_ip(0, "%s.ip", pfx); | |
1222 | mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx); | |
1223 | gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx); | |
1224 | ||
1225 | netdev_turn_flags_on(netdev, NETDEV_UP, true); | |
1226 | if (!mask.s_addr) { | |
1227 | mask.s_addr = guess_netmask(ip.s_addr); | |
1228 | } | |
1229 | if (!netdev_set_in4(netdev, ip, mask)) { | |
1230 | VLOG_INFO("bridge %s: configured IP address "IP_FMT", " | |
1231 | "netmask "IP_FMT, | |
1232 | br->name, IP_ARGS(&ip.s_addr), | |
1233 | IP_ARGS(&mask.s_addr)); | |
1234 | } | |
1235 | ||
1236 | if (gateway.s_addr) { | |
1237 | if (!netdev_add_router(gateway)) { | |
1238 | VLOG_INFO("bridge %s: configured gateway "IP_FMT, | |
1239 | br->name, IP_ARGS(&gateway.s_addr)); | |
1240 | } | |
1241 | } | |
1242 | } | |
1243 | netdev_close(netdev); | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | fail_mode = cfg_get_string(0, "%s.fail-mode", pfx); | |
1248 | if (!fail_mode) { | |
1249 | fail_mode = cfg_get_string(0, "mgmt.fail-mode"); | |
1250 | } | |
1251 | ofproto_set_failure(br->ofproto, | |
1252 | (!fail_mode | |
1253 | || !strcmp(fail_mode, "standalone") | |
1254 | || !strcmp(fail_mode, "open"))); | |
1255 | ||
1256 | probe = cfg_get_int(0, "%s.inactivity-probe", pfx); | |
952efc48 JP |
1257 | if (probe < 5) { |
1258 | probe = cfg_get_int(0, "mgmt.inactivity-probe"); | |
1259 | if (probe < 5) { | |
f9fb1858 | 1260 | probe = 5; |
952efc48 JP |
1261 | } |
1262 | } | |
1263 | ofproto_set_probe_interval(br->ofproto, probe); | |
064af421 BP |
1264 | |
1265 | max_backoff = cfg_get_int(0, "%s.max-backoff", pfx); | |
1266 | if (!max_backoff) { | |
1267 | max_backoff = cfg_get_int(0, "mgmt.max-backoff"); | |
1268 | if (!max_backoff) { | |
c9aaa877 | 1269 | max_backoff = 8; |
064af421 BP |
1270 | } |
1271 | } | |
1272 | ofproto_set_max_backoff(br->ofproto, max_backoff); | |
1273 | ||
1274 | rate_limit = cfg_get_int(0, "%s.rate-limit", pfx); | |
1275 | if (!rate_limit) { | |
1276 | rate_limit = cfg_get_int(0, "mgmt.rate-limit"); | |
1277 | } | |
1278 | burst_limit = cfg_get_int(0, "%s.burst-limit", pfx); | |
1279 | if (!burst_limit) { | |
1280 | burst_limit = cfg_get_int(0, "mgmt.burst-limit"); | |
1281 | } | |
1282 | ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); | |
1283 | ||
1284 | ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx)); | |
1285 | ||
1286 | if (cfg_has("%s.commands.acl", pfx)) { | |
1287 | struct svec command_acls; | |
1288 | char *command_acl; | |
1289 | ||
1290 | svec_init(&command_acls); | |
1291 | cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx); | |
1292 | command_acl = svec_join(&command_acls, ",", ""); | |
1293 | ||
1294 | ofproto_set_remote_execution(br->ofproto, command_acl, | |
1295 | cfg_get_string(0, "%s.commands.dir", | |
1296 | pfx)); | |
1297 | ||
1298 | svec_destroy(&command_acls); | |
1299 | free(command_acl); | |
1300 | } else { | |
1301 | ofproto_set_remote_execution(br->ofproto, NULL, NULL); | |
1302 | } | |
1303 | } else { | |
1304 | union ofp_action action; | |
1305 | flow_t flow; | |
1306 | ||
1307 | /* Set up a flow that matches every packet and directs them to | |
1308 | * OFPP_NORMAL (which goes to us). */ | |
1309 | memset(&action, 0, sizeof action); | |
1310 | action.type = htons(OFPAT_OUTPUT); | |
1311 | action.output.len = htons(sizeof action); | |
1312 | action.output.port = htons(OFPP_NORMAL); | |
1313 | memset(&flow, 0, sizeof flow); | |
1314 | ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0, | |
1315 | &action, 1, 0); | |
1316 | ||
1317 | ofproto_set_in_band(br->ofproto, false); | |
1318 | ofproto_set_max_backoff(br->ofproto, 1); | |
1319 | ofproto_set_probe_interval(br->ofproto, 5); | |
1320 | ofproto_set_failure(br->ofproto, false); | |
1321 | ofproto_set_stp(br->ofproto, false); | |
1322 | } | |
1323 | free(pfx); | |
1324 | ||
1325 | ofproto_set_controller(br->ofproto, br->controller); | |
1326 | } | |
1327 | ||
1328 | static void | |
1329 | bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces) | |
1330 | { | |
1331 | size_t i, j; | |
1332 | ||
1333 | svec_init(ifaces); | |
1334 | for (i = 0; i < br->n_ports; i++) { | |
1335 | struct port *port = br->ports[i]; | |
1336 | for (j = 0; j < port->n_ifaces; j++) { | |
1337 | struct iface *iface = port->ifaces[j]; | |
1338 | svec_add(ifaces, iface->name); | |
1339 | } | |
35c979bf BP |
1340 | if (port->n_ifaces > 1 |
1341 | && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) { | |
1342 | svec_add(ifaces, port->name); | |
1343 | } | |
064af421 | 1344 | } |
35c979bf | 1345 | svec_sort_unique(ifaces); |
064af421 BP |
1346 | } |
1347 | ||
1348 | /* For robustness, in case the administrator moves around datapath ports behind | |
1349 | * our back, we re-check all the datapath port numbers here. | |
1350 | * | |
1351 | * This function will set the 'dp_ifidx' members of interfaces that have | |
1352 | * disappeared to -1, so only call this function from a context where those | |
1353 | * 'struct iface's will be removed from the bridge. Otherwise, the -1 | |
1354 | * 'dp_ifidx'es will cause trouble later when we try to send them to the | |
1355 | * datapath, which doesn't support UINT16_MAX+1 ports. */ | |
1356 | static void | |
1357 | bridge_fetch_dp_ifaces(struct bridge *br) | |
1358 | { | |
1359 | struct odp_port *dpif_ports; | |
1360 | size_t n_dpif_ports; | |
1361 | size_t i, j; | |
1362 | ||
1363 | /* Reset all interface numbers. */ | |
1364 | for (i = 0; i < br->n_ports; i++) { | |
1365 | struct port *port = br->ports[i]; | |
1366 | for (j = 0; j < port->n_ifaces; j++) { | |
1367 | struct iface *iface = port->ifaces[j]; | |
1368 | iface->dp_ifidx = -1; | |
1369 | } | |
1370 | } | |
1371 | port_array_clear(&br->ifaces); | |
1372 | ||
1373 | dpif_port_list(&br->dpif, &dpif_ports, &n_dpif_ports); | |
1374 | for (i = 0; i < n_dpif_ports; i++) { | |
1375 | struct odp_port *p = &dpif_ports[i]; | |
1376 | struct iface *iface = iface_lookup(br, p->devname); | |
1377 | if (iface) { | |
1378 | if (iface->dp_ifidx >= 0) { | |
1379 | VLOG_WARN("dp%u reported interface %s twice", | |
1380 | dpif_id(&br->dpif), p->devname); | |
1381 | } else if (iface_from_dp_ifidx(br, p->port)) { | |
1382 | VLOG_WARN("dp%u reported interface %"PRIu16" twice", | |
1383 | dpif_id(&br->dpif), p->port); | |
1384 | } else { | |
1385 | port_array_set(&br->ifaces, p->port, iface); | |
1386 | iface->dp_ifidx = p->port; | |
1387 | } | |
1388 | } | |
1389 | } | |
1390 | free(dpif_ports); | |
1391 | } | |
1392 | \f | |
1393 | /* Bridge packet processing functions. */ | |
1394 | ||
da285df4 BP |
1395 | static int |
1396 | bond_hash(const uint8_t mac[ETH_ADDR_LEN]) | |
1397 | { | |
1398 | return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; | |
1399 | } | |
1400 | ||
064af421 BP |
1401 | static struct bond_entry * |
1402 | lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) | |
1403 | { | |
da285df4 | 1404 | return &port->bond_hash[bond_hash(mac)]; |
064af421 BP |
1405 | } |
1406 | ||
1407 | static int | |
1408 | bond_choose_iface(const struct port *port) | |
1409 | { | |
1410 | size_t i; | |
1411 | for (i = 0; i < port->n_ifaces; i++) { | |
1412 | if (port->ifaces[i]->enabled) { | |
1413 | return i; | |
1414 | } | |
1415 | } | |
1416 | return -1; | |
1417 | } | |
1418 | ||
1419 | static bool | |
2303f3b2 | 1420 | choose_output_iface(const struct port *port, const uint8_t *dl_src, |
064af421 BP |
1421 | uint16_t *dp_ifidx, tag_type *tags) |
1422 | { | |
1423 | struct iface *iface; | |
1424 | ||
1425 | assert(port->n_ifaces); | |
1426 | if (port->n_ifaces == 1) { | |
1427 | iface = port->ifaces[0]; | |
1428 | } else { | |
2303f3b2 | 1429 | struct bond_entry *e = lookup_bond_entry(port, dl_src); |
064af421 BP |
1430 | if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces |
1431 | || !port->ifaces[e->iface_idx]->enabled) { | |
1432 | /* XXX select interface properly. The current interface selection | |
1433 | * is only good for testing the rebalancing code. */ | |
1434 | e->iface_idx = bond_choose_iface(port); | |
1435 | if (e->iface_idx < 0) { | |
1436 | *tags |= port->no_ifaces_tag; | |
1437 | return false; | |
1438 | } | |
1439 | e->iface_tag = tag_create_random(); | |
85c74638 | 1440 | ((struct port *) port)->bond_compat_is_stale = true; |
064af421 BP |
1441 | } |
1442 | *tags |= e->iface_tag; | |
1443 | iface = port->ifaces[e->iface_idx]; | |
1444 | } | |
1445 | *dp_ifidx = iface->dp_ifidx; | |
1446 | *tags |= iface->tag; /* Currently only used for bonding. */ | |
1447 | return true; | |
1448 | } | |
1449 | ||
1450 | static void | |
1451 | bond_link_status_update(struct iface *iface, bool carrier) | |
1452 | { | |
1453 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1454 | struct port *port = iface->port; | |
1455 | ||
1456 | if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { | |
1457 | /* Nothing to do. */ | |
1458 | return; | |
1459 | } | |
1460 | VLOG_INFO_RL(&rl, "interface %s: carrier %s", | |
1461 | iface->name, carrier ? "detected" : "dropped"); | |
1462 | if (carrier == iface->enabled) { | |
1463 | iface->delay_expires = LLONG_MAX; | |
1464 | VLOG_INFO_RL(&rl, "interface %s: will not be %s", | |
1465 | iface->name, carrier ? "disabled" : "enabled"); | |
25ce84b2 BP |
1466 | } else if (carrier && port->updelay && port->active_iface < 0) { |
1467 | iface->delay_expires = time_msec(); | |
1468 | VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " | |
1469 | "other interface is up", iface->name, port->updelay); | |
064af421 BP |
1470 | } else { |
1471 | int delay = carrier ? port->updelay : port->downdelay; | |
1472 | iface->delay_expires = time_msec() + delay; | |
1473 | if (delay) { | |
1474 | VLOG_INFO_RL(&rl, | |
1475 | "interface %s: will be %s if it stays %s for %d ms", | |
1476 | iface->name, | |
1477 | carrier ? "enabled" : "disabled", | |
1478 | carrier ? "up" : "down", | |
1479 | delay); | |
1480 | } | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | static void | |
1485 | bond_choose_active_iface(struct port *port) | |
1486 | { | |
1487 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1488 | ||
1489 | port->active_iface = bond_choose_iface(port); | |
1490 | port->active_iface_tag = tag_create_random(); | |
1491 | if (port->active_iface >= 0) { | |
1492 | VLOG_INFO_RL(&rl, "port %s: active interface is now %s", | |
1493 | port->name, port->ifaces[port->active_iface]->name); | |
1494 | } else { | |
1495 | VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", | |
1496 | port->name); | |
1497 | } | |
1498 | } | |
1499 | ||
da285df4 BP |
1500 | static void |
1501 | bond_enable_slave(struct iface *iface, bool enable) | |
1502 | { | |
1503 | struct port *port = iface->port; | |
1504 | struct bridge *br = port->bridge; | |
1505 | ||
1506 | iface->delay_expires = LLONG_MAX; | |
1507 | if (enable == iface->enabled) { | |
1508 | return; | |
1509 | } | |
1510 | ||
1511 | iface->enabled = enable; | |
1512 | if (!iface->enabled) { | |
17ea75b2 | 1513 | VLOG_WARN("interface %s: disabled", iface->name); |
da285df4 BP |
1514 | ofproto_revalidate(br->ofproto, iface->tag); |
1515 | if (iface->port_ifidx == port->active_iface) { | |
1516 | ofproto_revalidate(br->ofproto, | |
1517 | port->active_iface_tag); | |
1518 | bond_choose_active_iface(port); | |
1519 | } | |
1520 | bond_send_learning_packets(port); | |
1521 | } else { | |
17ea75b2 | 1522 | VLOG_WARN("interface %s: enabled", iface->name); |
da285df4 BP |
1523 | if (port->active_iface < 0) { |
1524 | ofproto_revalidate(br->ofproto, port->no_ifaces_tag); | |
1525 | bond_choose_active_iface(port); | |
1526 | bond_send_learning_packets(port); | |
1527 | } | |
1528 | iface->tag = tag_create_random(); | |
1529 | } | |
2e8873af | 1530 | port_update_bond_compat(port); |
da285df4 BP |
1531 | } |
1532 | ||
064af421 BP |
1533 | static void |
1534 | bond_run(struct bridge *br) | |
1535 | { | |
1536 | size_t i, j; | |
1537 | ||
1538 | for (i = 0; i < br->n_ports; i++) { | |
1539 | struct port *port = br->ports[i]; | |
85c74638 BP |
1540 | |
1541 | if (port->bond_compat_is_stale) { | |
1542 | port->bond_compat_is_stale = false; | |
1543 | port_update_bond_compat(port); | |
1544 | } | |
1545 | ||
064af421 BP |
1546 | if (port->n_ifaces < 2) { |
1547 | continue; | |
1548 | } | |
1549 | for (j = 0; j < port->n_ifaces; j++) { | |
1550 | struct iface *iface = port->ifaces[j]; | |
1551 | if (time_msec() >= iface->delay_expires) { | |
da285df4 | 1552 | bond_enable_slave(iface, !iface->enabled); |
064af421 BP |
1553 | } |
1554 | } | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | static void | |
1559 | bond_wait(struct bridge *br) | |
1560 | { | |
1561 | size_t i, j; | |
1562 | ||
1563 | for (i = 0; i < br->n_ports; i++) { | |
1564 | struct port *port = br->ports[i]; | |
1565 | if (port->n_ifaces < 2) { | |
1566 | continue; | |
1567 | } | |
1568 | for (j = 0; j < port->n_ifaces; j++) { | |
1569 | struct iface *iface = port->ifaces[j]; | |
1570 | if (iface->delay_expires != LLONG_MAX) { | |
1571 | poll_timer_wait(iface->delay_expires - time_msec()); | |
1572 | } | |
1573 | } | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | static bool | |
1578 | set_dst(struct dst *p, const flow_t *flow, | |
1579 | const struct port *in_port, const struct port *out_port, | |
1580 | tag_type *tags) | |
1581 | { | |
1582 | /* STP handling. | |
1583 | * | |
1584 | * XXX This uses too many tags: any broadcast flow will get one tag per | |
1585 | * destination port, and thus a broadcast on a switch of any size is likely | |
1586 | * to have all tag bits set. We should figure out a way to be smarter. | |
1587 | * | |
1588 | * This is OK when STP is disabled, because stp_state_tag is 0 then. */ | |
1589 | *tags |= out_port->stp_state_tag; | |
1590 | if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) { | |
1591 | return false; | |
1592 | } | |
1593 | ||
1594 | p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE | |
1595 | : in_port->vlan >= 0 ? in_port->vlan | |
1596 | : ntohs(flow->dl_vlan)); | |
2303f3b2 | 1597 | return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); |
064af421 BP |
1598 | } |
1599 | ||
1600 | static void | |
1601 | swap_dst(struct dst *p, struct dst *q) | |
1602 | { | |
1603 | struct dst tmp = *p; | |
1604 | *p = *q; | |
1605 | *q = tmp; | |
1606 | } | |
1607 | ||
1608 | /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in | |
1609 | * 'dsts'. (This may help performance by reducing the number of VLAN changes | |
1610 | * that we push to the datapath. We could in fact fully sort the array by | |
1611 | * vlan, but in most cases there are at most two different vlan tags so that's | |
1612 | * possibly overkill.) */ | |
1613 | static void | |
1614 | partition_dsts(struct dst *dsts, size_t n_dsts, int vlan) | |
1615 | { | |
1616 | struct dst *first = dsts; | |
1617 | struct dst *last = dsts + n_dsts; | |
1618 | ||
1619 | while (first != last) { | |
1620 | /* Invariants: | |
1621 | * - All dsts < first have vlan == 'vlan'. | |
1622 | * - All dsts >= last have vlan != 'vlan'. | |
1623 | * - first < last. */ | |
1624 | while (first->vlan == vlan) { | |
1625 | if (++first == last) { | |
1626 | return; | |
1627 | } | |
1628 | } | |
1629 | ||
1630 | /* Same invariants, plus one additional: | |
1631 | * - first->vlan != vlan. | |
1632 | */ | |
1633 | while (last[-1].vlan != vlan) { | |
1634 | if (--last == first) { | |
1635 | return; | |
1636 | } | |
1637 | } | |
1638 | ||
1639 | /* Same invariants, plus one additional: | |
1640 | * - last[-1].vlan == vlan.*/ | |
1641 | swap_dst(first++, --last); | |
1642 | } | |
1643 | } | |
1644 | ||
1645 | static int | |
1646 | mirror_mask_ffs(mirror_mask_t mask) | |
1647 | { | |
1648 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
1649 | return ffs(mask); | |
1650 | } | |
1651 | ||
1652 | static bool | |
1653 | dst_is_duplicate(const struct dst *dsts, size_t n_dsts, | |
1654 | const struct dst *test) | |
1655 | { | |
1656 | size_t i; | |
1657 | for (i = 0; i < n_dsts; i++) { | |
1658 | if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) { | |
1659 | return true; | |
1660 | } | |
1661 | } | |
1662 | return false; | |
1663 | } | |
1664 | ||
1665 | static bool | |
1666 | port_trunks_vlan(const struct port *port, uint16_t vlan) | |
1667 | { | |
1668 | return port->vlan < 0 && bitmap_is_set(port->trunks, vlan); | |
1669 | } | |
1670 | ||
1671 | static bool | |
1672 | port_includes_vlan(const struct port *port, uint16_t vlan) | |
1673 | { | |
1674 | return vlan == port->vlan || port_trunks_vlan(port, vlan); | |
1675 | } | |
1676 | ||
1677 | static size_t | |
1678 | compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, | |
1679 | const struct port *in_port, const struct port *out_port, | |
1680 | struct dst dsts[], tag_type *tags) | |
1681 | { | |
1682 | mirror_mask_t mirrors = in_port->src_mirrors; | |
1683 | struct dst *dst = dsts; | |
1684 | size_t i; | |
1685 | ||
1686 | *tags |= in_port->stp_state_tag; | |
1687 | if (out_port == FLOOD_PORT) { | |
1688 | /* XXX use ODP_FLOOD if no vlans or bonding. */ | |
1689 | /* XXX even better, define each VLAN as a datapath port group */ | |
1690 | for (i = 0; i < br->n_ports; i++) { | |
1691 | struct port *port = br->ports[i]; | |
1692 | if (port != in_port && port_includes_vlan(port, vlan) | |
1693 | && !port->is_mirror_output_port | |
1694 | && set_dst(dst, flow, in_port, port, tags)) { | |
1695 | mirrors |= port->dst_mirrors; | |
1696 | dst++; | |
1697 | } | |
1698 | } | |
1699 | } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { | |
1700 | mirrors |= out_port->dst_mirrors; | |
1701 | dst++; | |
1702 | } | |
1703 | ||
1704 | while (mirrors) { | |
1705 | struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
1706 | if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { | |
1707 | if (m->out_port) { | |
1708 | if (set_dst(dst, flow, in_port, m->out_port, tags) | |
1709 | && !dst_is_duplicate(dsts, dst - dsts, dst)) { | |
1710 | dst++; | |
1711 | } | |
1712 | } else { | |
1713 | for (i = 0; i < br->n_ports; i++) { | |
1714 | struct port *port = br->ports[i]; | |
1715 | if (port_includes_vlan(port, m->out_vlan) | |
274de4d2 | 1716 | && set_dst(dst, flow, in_port, port, tags)) |
064af421 | 1717 | { |
43aa5f47 JG |
1718 | int flow_vlan; |
1719 | ||
064af421 BP |
1720 | if (port->vlan < 0) { |
1721 | dst->vlan = m->out_vlan; | |
1722 | } | |
274de4d2 BP |
1723 | if (dst_is_duplicate(dsts, dst - dsts, dst)) { |
1724 | continue; | |
1725 | } | |
43aa5f47 JG |
1726 | |
1727 | /* Use the vlan tag on the original flow instead of | |
1728 | * the one passed in the vlan parameter. This ensures | |
1729 | * that we compare the vlan from before any implicit | |
1730 | * tagging tags place. This is necessary because | |
1731 | * dst->vlan is the final vlan, after removing implicit | |
1732 | * tags. */ | |
1733 | flow_vlan = ntohs(flow->dl_vlan); | |
1734 | if (flow_vlan == 0) { | |
1735 | flow_vlan = OFP_VLAN_NONE; | |
1736 | } | |
1737 | if (port == in_port && dst->vlan == flow_vlan) { | |
064af421 BP |
1738 | /* Don't send out input port on same VLAN. */ |
1739 | continue; | |
1740 | } | |
1741 | dst++; | |
1742 | } | |
1743 | } | |
1744 | } | |
1745 | } | |
1746 | mirrors &= mirrors - 1; | |
1747 | } | |
1748 | ||
1749 | partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); | |
1750 | return dst - dsts; | |
1751 | } | |
1752 | ||
1753 | static void UNUSED | |
1754 | print_dsts(const struct dst *dsts, size_t n) | |
1755 | { | |
1756 | for (; n--; dsts++) { | |
1757 | printf(">p%"PRIu16, dsts->dp_ifidx); | |
1758 | if (dsts->vlan != OFP_VLAN_NONE) { | |
1759 | printf("v%"PRIu16, dsts->vlan); | |
1760 | } | |
1761 | } | |
1762 | } | |
1763 | ||
1764 | static void | |
1765 | compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, | |
1766 | const struct port *in_port, const struct port *out_port, | |
1767 | tag_type *tags, struct odp_actions *actions) | |
1768 | { | |
1769 | struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; | |
1770 | size_t n_dsts; | |
1771 | const struct dst *p; | |
1772 | uint16_t cur_vlan; | |
1773 | ||
1774 | n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags); | |
1775 | ||
1776 | cur_vlan = ntohs(flow->dl_vlan); | |
1777 | for (p = dsts; p < &dsts[n_dsts]; p++) { | |
1778 | union odp_action *a; | |
1779 | if (p->vlan != cur_vlan) { | |
1780 | if (p->vlan == OFP_VLAN_NONE) { | |
1781 | odp_actions_add(actions, ODPAT_STRIP_VLAN); | |
1782 | } else { | |
1783 | a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); | |
1784 | a->vlan_vid.vlan_vid = htons(p->vlan); | |
1785 | } | |
1786 | cur_vlan = p->vlan; | |
1787 | } | |
1788 | a = odp_actions_add(actions, ODPAT_OUTPUT); | |
1789 | a->output.port = p->dp_ifidx; | |
1790 | } | |
1791 | } | |
1792 | ||
1793 | static bool | |
1794 | is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet) | |
1795 | { | |
1796 | struct arp_eth_header *arp = (struct arp_eth_header *) packet->data; | |
1797 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
1798 | && eth_addr_is_broadcast(flow->dl_dst) | |
1799 | && packet->size >= sizeof(struct arp_eth_header) | |
1800 | && arp->ar_op == ARP_OP_REQUEST); | |
1801 | } | |
1802 | ||
1803 | /* If the composed actions may be applied to any packet in the given 'flow', | |
1804 | * returns true. Otherwise, the actions should only be applied to 'packet', or | |
1805 | * not at all, if 'packet' was NULL. */ | |
1806 | static bool | |
1807 | process_flow(struct bridge *br, const flow_t *flow, | |
1808 | const struct ofpbuf *packet, struct odp_actions *actions, | |
1809 | tag_type *tags) | |
1810 | { | |
1811 | struct iface *in_iface; | |
1812 | struct port *in_port; | |
1813 | struct port *out_port = NULL; /* By default, drop the packet/flow. */ | |
1814 | int vlan; | |
1815 | ||
1816 | /* Find the interface and port structure for the received packet. */ | |
1817 | in_iface = iface_from_dp_ifidx(br, flow->in_port); | |
1818 | if (!in_iface) { | |
1819 | /* No interface? Something fishy... */ | |
1820 | if (packet != NULL) { | |
1821 | /* Odd. A few possible reasons here: | |
1822 | * | |
1823 | * - We deleted an interface but there are still a few packets | |
1824 | * queued up from it. | |
1825 | * | |
1826 | * - Someone externally added an interface (e.g. with "ovs-dpctl | |
1827 | * add-if") that we don't know about. | |
1828 | * | |
1829 | * - Packet arrived on the local port but the local port is not | |
1830 | * one of our bridge ports. | |
1831 | */ | |
1832 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1833 | ||
1834 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
1835 | "interface %"PRIu16, br->name, flow->in_port); | |
1836 | } | |
1837 | ||
1838 | /* Return without adding any actions, to drop packets on this flow. */ | |
1839 | return true; | |
1840 | } | |
1841 | in_port = in_iface->port; | |
1842 | ||
1843 | /* Figure out what VLAN this packet belongs to. | |
1844 | * | |
1845 | * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet | |
1846 | * belongs to VLAN 0, so we should treat both cases identically. (In the | |
1847 | * former case, the packet has an 802.1Q header that specifies VLAN 0, | |
1848 | * presumably to allow a priority to be specified. In the latter case, the | |
1849 | * packet does not have any 802.1Q header.) */ | |
1850 | vlan = ntohs(flow->dl_vlan); | |
1851 | if (vlan == OFP_VLAN_NONE) { | |
1852 | vlan = 0; | |
1853 | } | |
1854 | if (in_port->vlan >= 0) { | |
1855 | if (vlan) { | |
1856 | /* XXX support double tagging? */ | |
1857 | if (packet != NULL) { | |
1858 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1859 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
1860 | "packet received on port %s configured with " | |
1861 | "implicit VLAN %"PRIu16, | |
1862 | br->name, ntohs(flow->dl_vlan), | |
1863 | in_port->name, in_port->vlan); | |
1864 | } | |
1865 | goto done; | |
1866 | } | |
1867 | vlan = in_port->vlan; | |
1868 | } else { | |
1869 | if (!port_includes_vlan(in_port, vlan)) { | |
1870 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1871 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
1872 | "packet received on port %s not configured for " | |
1873 | "trunking VLAN %d", | |
1874 | br->name, vlan, in_port->name, vlan); | |
1875 | goto done; | |
1876 | } | |
1877 | } | |
1878 | ||
1879 | /* Drop frames for ports that STP wants entirely killed (both for | |
1880 | * forwarding and for learning). Later, after we do learning, we'll drop | |
1881 | * the frames that STP wants to do learning but not forwarding on. */ | |
1882 | if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) { | |
1883 | goto done; | |
1884 | } | |
1885 | ||
1886 | /* Drop frames for reserved multicast addresses. */ | |
1887 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
1888 | goto done; | |
1889 | } | |
1890 | ||
1891 | /* Drop frames on ports reserved for mirroring. */ | |
1892 | if (in_port->is_mirror_output_port) { | |
1893 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1894 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " | |
1895 | "which is reserved exclusively for mirroring", | |
1896 | br->name, in_port->name); | |
1897 | goto done; | |
1898 | } | |
1899 | ||
c93b1d6a | 1900 | /* Multicast (and broadcast) packets on bonds need special attention, to |
064af421 BP |
1901 | * avoid receiving duplicates. */ |
1902 | if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) { | |
1903 | *tags |= in_port->active_iface_tag; | |
1904 | if (in_port->active_iface != in_iface->port_ifidx) { | |
c93b1d6a | 1905 | /* Drop all multicast packets on inactive slaves. */ |
064af421 | 1906 | goto done; |
c93b1d6a BP |
1907 | } else { |
1908 | /* Drop all multicast packets for which we have learned a different | |
b1040b17 | 1909 | * input port, because we probably sent the packet on one slave |
c93b1d6a BP |
1910 | * and got it back on the active slave. Broadcast ARP replies are |
1911 | * an exception to this rule: the host has moved to another | |
1912 | * switch. */ | |
1913 | int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); | |
e2ead27a BP |
1914 | if (src_idx != -1 && src_idx != in_port->port_idx) { |
1915 | if (packet) { | |
1916 | if (!is_bcast_arp_reply(flow, packet)) { | |
1917 | goto done; | |
1918 | } | |
1919 | } else { | |
1920 | /* No way to know whether it's an ARP reply, because the | |
1921 | * flow entry doesn't include enough information and we | |
1922 | * don't have a packet. Punt. */ | |
1923 | return false; | |
1924 | } | |
c93b1d6a | 1925 | } |
064af421 BP |
1926 | } |
1927 | } | |
1928 | ||
1929 | /* MAC learning. */ | |
1930 | out_port = FLOOD_PORT; | |
1931 | if (br->ml) { | |
1932 | int out_port_idx; | |
064af421 | 1933 | |
c93b1d6a BP |
1934 | /* Learn source MAC (but don't try to learn from revalidation). */ |
1935 | if (packet) { | |
064af421 BP |
1936 | tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, |
1937 | vlan, in_port->port_idx); | |
1938 | if (rev_tag) { | |
1939 | /* The log messages here could actually be useful in debugging, | |
1940 | * so keep the rate limit relatively high. */ | |
1941 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, | |
1942 | 300); | |
1943 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
1944 | "on port %s in VLAN %d", | |
1945 | br->name, ETH_ADDR_ARGS(flow->dl_src), | |
1946 | in_port->name, vlan); | |
1947 | ofproto_revalidate(br->ofproto, rev_tag); | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | /* Determine output port. */ | |
1952 | out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, | |
1953 | tags); | |
1954 | if (out_port_idx >= 0 && out_port_idx < br->n_ports) { | |
1955 | out_port = br->ports[out_port_idx]; | |
1956 | } | |
1957 | } | |
1958 | ||
1959 | /* Don't send packets out their input ports. Don't forward frames that STP | |
1960 | * wants us to discard. */ | |
1961 | if (in_port == out_port || in_port->stp_state == STP_LEARNING) { | |
1962 | out_port = NULL; | |
1963 | } | |
1964 | ||
1965 | done: | |
1966 | compose_actions(br, flow, vlan, in_port, out_port, tags, actions); | |
1967 | ||
1968 | /* | |
1969 | * We send out only a single packet, instead of setting up a flow, if the | |
1970 | * packet is an ARP directed to broadcast that arrived on a bonded | |
1971 | * interface. In such a situation ARP requests and replies must be handled | |
1972 | * differently, but OpenFlow unfortunately can't distinguish them. | |
1973 | */ | |
1974 | return (in_port->n_ifaces < 2 | |
1975 | || flow->dl_type != htons(ETH_TYPE_ARP) | |
1976 | || !eth_addr_is_broadcast(flow->dl_dst)); | |
1977 | } | |
1978 | ||
1979 | /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port | |
1980 | * number. */ | |
1981 | static void | |
1982 | bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, | |
1983 | const struct ofp_phy_port *opp, | |
1984 | void *br_) | |
1985 | { | |
1986 | struct bridge *br = br_; | |
1987 | struct iface *iface; | |
1988 | struct port *port; | |
1989 | ||
1990 | iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); | |
1991 | if (!iface) { | |
1992 | return; | |
1993 | } | |
1994 | port = iface->port; | |
1995 | ||
1996 | if (reason == OFPPR_DELETE) { | |
1997 | VLOG_WARN("bridge %s: interface %s deleted unexpectedly", | |
1998 | br->name, iface->name); | |
1999 | iface_destroy(iface); | |
2000 | if (!port->n_ifaces) { | |
2001 | VLOG_WARN("bridge %s: port %s has no interfaces, dropping", | |
2002 | br->name, port->name); | |
2003 | port_destroy(port); | |
2004 | } | |
2005 | ||
2006 | bridge_flush(br); | |
2007 | } else { | |
2008 | memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN); | |
2009 | if (port->n_ifaces > 1) { | |
2010 | bool up = !(opp->state & OFPPS_LINK_DOWN); | |
2011 | bond_link_status_update(iface, up); | |
2012 | port_update_bond_compat(port); | |
2013 | } | |
2014 | } | |
2015 | } | |
2016 | ||
2017 | static bool | |
2018 | bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
2019 | struct odp_actions *actions, tag_type *tags, void *br_) | |
2020 | { | |
2021 | struct bridge *br = br_; | |
2022 | ||
2023 | #if 0 | |
2024 | if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE) | |
2025 | && eth_addr_equals(flow->dl_dst, stp_eth_addr)) { | |
2026 | brstp_receive(br, flow, payload); | |
2027 | return true; | |
2028 | } | |
2029 | #endif | |
2030 | ||
2031 | COVERAGE_INC(bridge_process_flow); | |
2032 | return process_flow(br, flow, packet, actions, tags); | |
2033 | } | |
2034 | ||
2035 | static void | |
2036 | bridge_account_flow_ofhook_cb(const flow_t *flow, | |
2037 | const union odp_action *actions, | |
2038 | size_t n_actions, unsigned long long int n_bytes, | |
2039 | void *br_) | |
2040 | { | |
2041 | struct bridge *br = br_; | |
2042 | const union odp_action *a; | |
2043 | ||
2044 | if (!br->has_bonded_ports) { | |
2045 | return; | |
2046 | } | |
2047 | ||
2048 | for (a = actions; a < &actions[n_actions]; a++) { | |
2049 | if (a->type == ODPAT_OUTPUT) { | |
2050 | struct port *port = port_from_dp_ifidx(br, a->output.port); | |
2051 | if (port && port->n_ifaces >= 2) { | |
2052 | struct bond_entry *e = lookup_bond_entry(port, flow->dl_src); | |
2053 | e->tx_bytes += n_bytes; | |
2054 | } | |
2055 | } | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | static void | |
2060 | bridge_account_checkpoint_ofhook_cb(void *br_) | |
2061 | { | |
2062 | struct bridge *br = br_; | |
2063 | size_t i; | |
2064 | ||
2065 | if (!br->has_bonded_ports) { | |
2066 | return; | |
2067 | } | |
2068 | ||
2069 | /* The current ofproto implementation calls this callback at least once a | |
2070 | * second, so this timer implementation is sufficient. */ | |
2071 | if (time_msec() < br->bond_next_rebalance) { | |
2072 | return; | |
2073 | } | |
2074 | br->bond_next_rebalance = time_msec() + 10000; | |
2075 | ||
2076 | for (i = 0; i < br->n_ports; i++) { | |
2077 | struct port *port = br->ports[i]; | |
2078 | if (port->n_ifaces > 1) { | |
2079 | bond_rebalance_port(port); | |
2080 | } | |
2081 | } | |
2082 | } | |
2083 | ||
2084 | static struct ofhooks bridge_ofhooks = { | |
2085 | bridge_port_changed_ofhook_cb, | |
2086 | bridge_normal_ofhook_cb, | |
2087 | bridge_account_flow_ofhook_cb, | |
2088 | bridge_account_checkpoint_ofhook_cb, | |
2089 | }; | |
2090 | \f | |
2303f3b2 BP |
2091 | /* Bonding functions. */ |
2092 | ||
064af421 BP |
2093 | /* Statistics for a single interface on a bonded port, used for load-based |
2094 | * bond rebalancing. */ | |
2095 | struct slave_balance { | |
2096 | struct iface *iface; /* The interface. */ | |
2097 | uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */ | |
2098 | ||
2099 | /* All the "bond_entry"s that are assigned to this interface, in order of | |
2100 | * increasing tx_bytes. */ | |
2101 | struct bond_entry **hashes; | |
2102 | size_t n_hashes; | |
2103 | }; | |
2104 | ||
2105 | /* Sorts pointers to pointers to bond_entries in ascending order by the | |
2106 | * interface to which they are assigned, and within a single interface in | |
2107 | * ascending order of bytes transmitted. */ | |
2108 | static int | |
2109 | compare_bond_entries(const void *a_, const void *b_) | |
2110 | { | |
2111 | const struct bond_entry *const *ap = a_; | |
2112 | const struct bond_entry *const *bp = b_; | |
2113 | const struct bond_entry *a = *ap; | |
2114 | const struct bond_entry *b = *bp; | |
2115 | if (a->iface_idx != b->iface_idx) { | |
2116 | return a->iface_idx > b->iface_idx ? 1 : -1; | |
2117 | } else if (a->tx_bytes != b->tx_bytes) { | |
2118 | return a->tx_bytes > b->tx_bytes ? 1 : -1; | |
2119 | } else { | |
2120 | return 0; | |
2121 | } | |
2122 | } | |
2123 | ||
2124 | /* Sorts slave_balances so that enabled ports come first, and otherwise in | |
2125 | * *descending* order by number of bytes transmitted. */ | |
2126 | static int | |
2127 | compare_slave_balance(const void *a_, const void *b_) | |
2128 | { | |
2129 | const struct slave_balance *a = a_; | |
2130 | const struct slave_balance *b = b_; | |
2131 | if (a->iface->enabled != b->iface->enabled) { | |
2132 | return a->iface->enabled ? -1 : 1; | |
2133 | } else if (a->tx_bytes != b->tx_bytes) { | |
2134 | return a->tx_bytes > b->tx_bytes ? -1 : 1; | |
2135 | } else { | |
2136 | return 0; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | static void | |
2141 | swap_bals(struct slave_balance *a, struct slave_balance *b) | |
2142 | { | |
2143 | struct slave_balance tmp = *a; | |
2144 | *a = *b; | |
2145 | *b = tmp; | |
2146 | } | |
2147 | ||
2148 | /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order | |
2149 | * given that 'p' (and only 'p') might be in the wrong location. | |
2150 | * | |
2151 | * This function invalidates 'p', since it might now be in a different memory | |
2152 | * location. */ | |
2153 | static void | |
2154 | resort_bals(struct slave_balance *p, | |
2155 | struct slave_balance bals[], size_t n_bals) | |
2156 | { | |
2157 | if (n_bals > 1) { | |
2158 | for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { | |
2159 | swap_bals(p, p - 1); | |
2160 | } | |
2161 | for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { | |
2162 | swap_bals(p, p + 1); | |
2163 | } | |
2164 | } | |
2165 | } | |
2166 | ||
2167 | static void | |
2168 | log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) | |
2169 | { | |
2170 | if (VLOG_IS_DBG_ENABLED()) { | |
2171 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2172 | const struct slave_balance *b; | |
2173 | ||
2174 | for (b = bals; b < bals + n_bals; b++) { | |
2175 | size_t i; | |
2176 | ||
2177 | if (b > bals) { | |
2178 | ds_put_char(&ds, ','); | |
2179 | } | |
2180 | ds_put_format(&ds, " %s %"PRIu64"kB", | |
2181 | b->iface->name, b->tx_bytes / 1024); | |
2182 | ||
2183 | if (!b->iface->enabled) { | |
2184 | ds_put_cstr(&ds, " (disabled)"); | |
2185 | } | |
2186 | if (b->n_hashes > 0) { | |
2187 | ds_put_cstr(&ds, " ("); | |
2188 | for (i = 0; i < b->n_hashes; i++) { | |
2189 | const struct bond_entry *e = b->hashes[i]; | |
2190 | if (i > 0) { | |
2191 | ds_put_cstr(&ds, " + "); | |
2192 | } | |
2193 | ds_put_format(&ds, "h%td: %"PRIu64"kB", | |
2194 | e - port->bond_hash, e->tx_bytes / 1024); | |
2195 | } | |
2196 | ds_put_cstr(&ds, ")"); | |
2197 | } | |
2198 | } | |
2199 | VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); | |
2200 | ds_destroy(&ds); | |
2201 | } | |
2202 | } | |
2203 | ||
2204 | /* Shifts 'hash' from 'from' to 'to' within 'port'. */ | |
2205 | static void | |
2206 | bond_shift_load(struct slave_balance *from, struct slave_balance *to, | |
2207 | struct bond_entry *hash) | |
2208 | { | |
2209 | struct port *port = from->iface->port; | |
2210 | uint64_t delta = hash->tx_bytes; | |
2211 | ||
2212 | VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " | |
2213 | "from %s to %s (now carrying %"PRIu64"kB and " | |
2214 | "%"PRIu64"kB load, respectively)", | |
2215 | port->name, delta / 1024, hash - port->bond_hash, | |
2216 | from->iface->name, to->iface->name, | |
2217 | (from->tx_bytes - delta) / 1024, | |
2218 | (to->tx_bytes + delta) / 1024); | |
2219 | ||
2220 | /* Delete element from from->hashes. | |
2221 | * | |
2222 | * We don't bother to add the element to to->hashes because not only would | |
2223 | * it require more work, the only purpose it would be to allow that hash to | |
2224 | * be migrated to another slave in this rebalancing run, and there is no | |
2225 | * point in doing that. */ | |
2226 | if (from->hashes[0] == hash) { | |
2227 | from->hashes++; | |
2228 | } else { | |
2229 | int i = hash - from->hashes[0]; | |
2230 | memmove(from->hashes + i, from->hashes + i + 1, | |
2231 | (from->n_hashes - (i + 1)) * sizeof *from->hashes); | |
2232 | } | |
2233 | from->n_hashes--; | |
2234 | ||
2235 | /* Shift load away from 'from' to 'to'. */ | |
2236 | from->tx_bytes -= delta; | |
2237 | to->tx_bytes += delta; | |
2238 | ||
2239 | /* Arrange for flows to be revalidated. */ | |
2240 | ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); | |
2241 | hash->iface_idx = to->iface->port_ifidx; | |
2242 | hash->iface_tag = tag_create_random(); | |
064af421 BP |
2243 | } |
2244 | ||
2245 | static void | |
2246 | bond_rebalance_port(struct port *port) | |
2247 | { | |
2248 | struct slave_balance bals[DP_MAX_PORTS]; | |
2249 | size_t n_bals; | |
2250 | struct bond_entry *hashes[BOND_MASK + 1]; | |
2251 | struct slave_balance *b, *from, *to; | |
2252 | struct bond_entry *e; | |
2253 | size_t i; | |
2254 | ||
2255 | /* Sets up 'bals' to describe each of the port's interfaces, sorted in | |
2256 | * descending order of tx_bytes, so that bals[0] represents the most | |
2257 | * heavily loaded slave and bals[n_bals - 1] represents the least heavily | |
2258 | * loaded slave. | |
2259 | * | |
2260 | * The code is a bit tricky: to avoid dynamically allocating a 'hashes' | |
2261 | * array for each slave_balance structure, we sort our local array of | |
2262 | * hashes in order by slave, so that all of the hashes for a given slave | |
2263 | * become contiguous in memory, and then we point each 'hashes' members of | |
2264 | * a slave_balance structure to the start of a contiguous group. */ | |
2265 | n_bals = port->n_ifaces; | |
2266 | for (b = bals; b < &bals[n_bals]; b++) { | |
2267 | b->iface = port->ifaces[b - bals]; | |
2268 | b->tx_bytes = 0; | |
2269 | b->hashes = NULL; | |
2270 | b->n_hashes = 0; | |
2271 | } | |
2272 | for (i = 0; i <= BOND_MASK; i++) { | |
2273 | hashes[i] = &port->bond_hash[i]; | |
2274 | } | |
2275 | qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); | |
2276 | for (i = 0; i <= BOND_MASK; i++) { | |
2277 | e = hashes[i]; | |
2278 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
2279 | b = &bals[e->iface_idx]; | |
2280 | b->tx_bytes += e->tx_bytes; | |
2281 | if (!b->hashes) { | |
2282 | b->hashes = &hashes[i]; | |
2283 | } | |
2284 | b->n_hashes++; | |
2285 | } | |
2286 | } | |
2287 | qsort(bals, n_bals, sizeof *bals, compare_slave_balance); | |
2288 | log_bals(bals, n_bals, port); | |
2289 | ||
2290 | /* Discard slaves that aren't enabled (which were sorted to the back of the | |
2291 | * array earlier). */ | |
2292 | while (!bals[n_bals - 1].iface->enabled) { | |
2293 | n_bals--; | |
2294 | if (!n_bals) { | |
2295 | return; | |
2296 | } | |
2297 | } | |
2298 | ||
2299 | /* Shift load from the most-loaded slaves to the least-loaded slaves. */ | |
2300 | to = &bals[n_bals - 1]; | |
2301 | for (from = bals; from < to; ) { | |
2302 | uint64_t overload = from->tx_bytes - to->tx_bytes; | |
2303 | if (overload < to->tx_bytes >> 5 || overload < 100000) { | |
2304 | /* The extra load on 'from' (and all less-loaded slaves), compared | |
2305 | * to that of 'to' (the least-loaded slave), is less than ~3%, or | |
2306 | * it is less than ~1Mbps. No point in rebalancing. */ | |
2307 | break; | |
2308 | } else if (from->n_hashes == 1) { | |
2309 | /* 'from' only carries a single MAC hash, so we can't shift any | |
2310 | * load away from it, even though we want to. */ | |
2311 | from++; | |
2312 | } else { | |
2313 | /* 'from' is carrying significantly more load than 'to', and that | |
2314 | * load is split across at least two different hashes. Pick a hash | |
2315 | * to migrate to 'to' (the least-loaded slave), given that doing so | |
2316 | * must not cause 'to''s load to exceed 'from''s load. | |
2317 | * | |
2318 | * The sort order we use means that we prefer to shift away the | |
2319 | * smallest hashes instead of the biggest ones. There is little | |
2320 | * reason behind this decision; we could use the opposite sort | |
2321 | * order to shift away big hashes ahead of small ones. */ | |
2322 | size_t i; | |
2323 | ||
2324 | for (i = 0; i < from->n_hashes; i++) { | |
2325 | uint64_t delta = from->hashes[i]->tx_bytes; | |
2326 | if (to->tx_bytes + delta < from->tx_bytes - delta) { | |
2327 | break; | |
2328 | } | |
2329 | } | |
2330 | if (i < from->n_hashes) { | |
2331 | bond_shift_load(from, to, from->hashes[i]); | |
2332 | ||
2333 | /* Re-sort 'bals'. Note that this may make 'from' and 'to' | |
2334 | * point to different slave_balance structures. It is only | |
2335 | * valid to do these two operations in a row at all because we | |
2336 | * know that 'from' will not move past 'to' and vice versa. */ | |
2337 | resort_bals(from, bals, n_bals); | |
2338 | resort_bals(to, bals, n_bals); | |
2339 | } else { | |
2340 | from++; | |
2341 | } | |
85c74638 | 2342 | port->bond_compat_is_stale = true; |
064af421 BP |
2343 | } |
2344 | } | |
2345 | ||
2346 | /* Implement exponentially weighted moving average. A weight of 1/2 causes | |
2347 | * historical data to decay to <1% in 7 rebalancing runs. */ | |
2348 | for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { | |
2349 | e->tx_bytes /= 2; | |
2350 | } | |
2351 | } | |
2303f3b2 BP |
2352 | |
2353 | static void | |
2354 | bond_send_learning_packets(struct port *port) | |
2355 | { | |
2356 | struct bridge *br = port->bridge; | |
2357 | struct mac_entry *e; | |
2358 | struct ofpbuf packet; | |
2359 | int error, n_packets, n_errors; | |
2360 | ||
2361 | if (!port->n_ifaces || port->active_iface < 0 || !br->ml) { | |
2362 | return; | |
2363 | } | |
2364 | ||
2365 | ofpbuf_init(&packet, 128); | |
2366 | error = n_packets = n_errors = 0; | |
2367 | LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { | |
2303f3b2 | 2368 | union ofp_action actions[2], *a; |
2303f3b2 BP |
2369 | uint16_t dp_ifidx; |
2370 | tag_type tags = 0; | |
2371 | flow_t flow; | |
2372 | int retval; | |
2373 | ||
2374 | if (e->port == port->port_idx | |
2375 | || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { | |
2376 | continue; | |
2377 | } | |
2378 | ||
2303f3b2 BP |
2379 | /* Compose actions. */ |
2380 | memset(actions, 0, sizeof actions); | |
2381 | a = actions; | |
2382 | if (e->vlan) { | |
2383 | a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); | |
2384 | a->vlan_vid.len = htons(sizeof *a); | |
2385 | a->vlan_vid.vlan_vid = htons(e->vlan); | |
2386 | a++; | |
2387 | } | |
2388 | a->output.type = htons(OFPAT_OUTPUT); | |
2389 | a->output.len = htons(sizeof *a); | |
2390 | a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); | |
2391 | a++; | |
2392 | ||
2393 | /* Send packet. */ | |
2394 | n_packets++; | |
b9e8b45a BP |
2395 | compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, |
2396 | e->mac); | |
2303f3b2 BP |
2397 | flow_extract(&packet, ODPP_NONE, &flow); |
2398 | retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, | |
2399 | &packet); | |
2400 | if (retval) { | |
2401 | error = retval; | |
2402 | n_errors++; | |
2403 | } | |
2404 | } | |
2405 | ofpbuf_uninit(&packet); | |
2406 | ||
2407 | if (n_errors) { | |
2408 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2409 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2410 | "packets, last error was: %s", | |
2411 | port->name, n_errors, n_packets, strerror(error)); | |
2412 | } else { | |
2413 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2414 | port->name, n_packets); | |
2415 | } | |
2416 | } | |
064af421 | 2417 | \f |
da285df4 BP |
2418 | /* Bonding unixctl user interface functions. */ |
2419 | ||
2420 | static void | |
2421 | bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED) | |
2422 | { | |
2423 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2424 | const struct bridge *br; | |
2425 | ||
2426 | ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); | |
2427 | ||
2428 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
2429 | size_t i; | |
2430 | ||
2431 | for (i = 0; i < br->n_ports; i++) { | |
2432 | const struct port *port = br->ports[i]; | |
2433 | if (port->n_ifaces > 1) { | |
2434 | size_t j; | |
2435 | ||
2436 | ds_put_format(&ds, "%s\t%s\t", br->name, port->name); | |
2437 | for (j = 0; j < port->n_ifaces; j++) { | |
2438 | const struct iface *iface = port->ifaces[j]; | |
2439 | if (j) { | |
2440 | ds_put_cstr(&ds, ", "); | |
2441 | } | |
2442 | ds_put_cstr(&ds, iface->name); | |
2443 | } | |
2444 | ds_put_char(&ds, '\n'); | |
2445 | } | |
2446 | } | |
2447 | } | |
2448 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
2449 | ds_destroy(&ds); | |
2450 | } | |
2451 | ||
2452 | static struct port * | |
2453 | bond_find(const char *name) | |
2454 | { | |
2455 | const struct bridge *br; | |
2456 | ||
2457 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
2458 | size_t i; | |
2459 | ||
2460 | for (i = 0; i < br->n_ports; i++) { | |
2461 | struct port *port = br->ports[i]; | |
2462 | if (!strcmp(port->name, name) && port->n_ifaces > 1) { | |
2463 | return port; | |
2464 | } | |
2465 | } | |
2466 | } | |
2467 | return NULL; | |
2468 | } | |
2469 | ||
2470 | static void | |
2471 | bond_unixctl_show(struct unixctl_conn *conn, const char *args) | |
2472 | { | |
2473 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2474 | const struct port *port; | |
2475 | size_t j; | |
2476 | ||
2477 | port = bond_find(args); | |
2478 | if (!port) { | |
2479 | unixctl_command_reply(conn, 501, "no such bond"); | |
2480 | return; | |
2481 | } | |
2482 | ||
2483 | ds_put_format(&ds, "updelay: %d ms\n", port->updelay); | |
2484 | ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); | |
2485 | ds_put_format(&ds, "next rebalance: %lld ms\n", | |
2486 | port->bridge->bond_next_rebalance - time_msec()); | |
2487 | for (j = 0; j < port->n_ifaces; j++) { | |
2488 | const struct iface *iface = port->ifaces[j]; | |
2489 | struct bond_entry *be; | |
2490 | ||
2491 | /* Basic info. */ | |
2492 | ds_put_format(&ds, "slave %s: %s\n", | |
2493 | iface->name, iface->enabled ? "enabled" : "disabled"); | |
2494 | if (j == port->active_iface) { | |
2495 | ds_put_cstr(&ds, "\tactive slave\n"); | |
2496 | } | |
2497 | if (iface->delay_expires != LLONG_MAX) { | |
2498 | ds_put_format(&ds, "\t%s expires in %lld ms\n", | |
2499 | iface->enabled ? "downdelay" : "updelay", | |
2500 | iface->delay_expires - time_msec()); | |
2501 | } | |
2502 | ||
2503 | /* Hashes. */ | |
2504 | for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { | |
2505 | int hash = be - port->bond_hash; | |
2506 | struct mac_entry *me; | |
2507 | ||
2508 | if (be->iface_idx != j) { | |
2509 | continue; | |
2510 | } | |
2511 | ||
2512 | ds_put_format(&ds, "\thash %d: %lld kB load\n", | |
2513 | hash, be->tx_bytes / 1024); | |
2514 | ||
2515 | /* MACs. */ | |
2516 | if (!port->bridge->ml) { | |
2517 | break; | |
2518 | } | |
2519 | ||
2520 | LIST_FOR_EACH (me, struct mac_entry, lru_node, | |
2521 | &port->bridge->ml->lrus) { | |
2522 | uint16_t dp_ifidx; | |
2523 | tag_type tags = 0; | |
2524 | if (bond_hash(me->mac) == hash | |
2525 | && me->port != port->port_idx | |
2526 | && choose_output_iface(port, me->mac, &dp_ifidx, &tags) | |
2527 | && dp_ifidx == iface->dp_ifidx) | |
2528 | { | |
2529 | ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", | |
2530 | ETH_ADDR_ARGS(me->mac)); | |
2531 | } | |
2532 | } | |
2533 | } | |
2534 | } | |
2535 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
2536 | ds_destroy(&ds); | |
2537 | } | |
2538 | ||
2539 | static void | |
2540 | bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_) | |
2541 | { | |
2542 | char *args = (char *) args_; | |
2543 | char *save_ptr = NULL; | |
2544 | char *bond_s, *hash_s, *slave_s; | |
2545 | uint8_t mac[ETH_ADDR_LEN]; | |
2546 | struct port *port; | |
2547 | struct iface *iface; | |
2548 | struct bond_entry *entry; | |
2549 | int hash; | |
2550 | ||
2551 | bond_s = strtok_r(args, " ", &save_ptr); | |
2552 | hash_s = strtok_r(NULL, " ", &save_ptr); | |
2553 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2554 | if (!slave_s) { | |
2555 | unixctl_command_reply(conn, 501, | |
2556 | "usage: bond/migrate BOND HASH SLAVE"); | |
2557 | return; | |
2558 | } | |
2559 | ||
2560 | port = bond_find(bond_s); | |
2561 | if (!port) { | |
2562 | unixctl_command_reply(conn, 501, "no such bond"); | |
2563 | return; | |
2564 | } | |
2565 | ||
eaa71334 BP |
2566 | if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) |
2567 | == ETH_ADDR_SCAN_COUNT) { | |
da285df4 BP |
2568 | hash = bond_hash(mac); |
2569 | } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) { | |
2570 | hash = atoi(hash_s) & BOND_MASK; | |
2571 | } else { | |
2572 | unixctl_command_reply(conn, 501, "bad hash"); | |
2573 | return; | |
2574 | } | |
2575 | ||
2576 | iface = port_lookup_iface(port, slave_s); | |
2577 | if (!iface) { | |
2578 | unixctl_command_reply(conn, 501, "no such slave"); | |
2579 | return; | |
2580 | } | |
2581 | ||
2582 | if (!iface->enabled) { | |
2583 | unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); | |
2584 | return; | |
2585 | } | |
2586 | ||
2587 | entry = &port->bond_hash[hash]; | |
2588 | ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); | |
2589 | entry->iface_idx = iface->port_ifidx; | |
2590 | entry->iface_tag = tag_create_random(); | |
85c74638 | 2591 | port->bond_compat_is_stale = true; |
da285df4 BP |
2592 | unixctl_command_reply(conn, 200, "migrated"); |
2593 | } | |
2594 | ||
2595 | static void | |
2596 | bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_) | |
2597 | { | |
2598 | char *args = (char *) args_; | |
2599 | char *save_ptr = NULL; | |
2600 | char *bond_s, *slave_s; | |
2601 | struct port *port; | |
2602 | struct iface *iface; | |
2603 | ||
2604 | bond_s = strtok_r(args, " ", &save_ptr); | |
2605 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2606 | if (!slave_s) { | |
2607 | unixctl_command_reply(conn, 501, | |
2608 | "usage: bond/set-active-slave BOND SLAVE"); | |
2609 | return; | |
2610 | } | |
2611 | ||
2612 | port = bond_find(bond_s); | |
2613 | if (!port) { | |
2614 | unixctl_command_reply(conn, 501, "no such bond"); | |
2615 | return; | |
2616 | } | |
2617 | ||
2618 | iface = port_lookup_iface(port, slave_s); | |
2619 | if (!iface) { | |
2620 | unixctl_command_reply(conn, 501, "no such slave"); | |
2621 | return; | |
2622 | } | |
2623 | ||
2624 | if (!iface->enabled) { | |
2625 | unixctl_command_reply(conn, 501, "cannot make disabled slave active"); | |
2626 | return; | |
2627 | } | |
2628 | ||
2629 | if (port->active_iface != iface->port_ifidx) { | |
2630 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
2631 | port->active_iface = iface->port_ifidx; | |
2632 | port->active_iface_tag = tag_create_random(); | |
2633 | VLOG_INFO("port %s: active interface is now %s", | |
2634 | port->name, iface->name); | |
2635 | bond_send_learning_packets(port); | |
2636 | unixctl_command_reply(conn, 200, "done"); | |
2637 | } else { | |
2638 | unixctl_command_reply(conn, 200, "no change"); | |
2639 | } | |
2640 | } | |
2641 | ||
2642 | static void | |
2643 | enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) | |
2644 | { | |
2645 | char *args = (char *) args_; | |
2646 | char *save_ptr = NULL; | |
2647 | char *bond_s, *slave_s; | |
2648 | struct port *port; | |
2649 | struct iface *iface; | |
2650 | ||
2651 | bond_s = strtok_r(args, " ", &save_ptr); | |
2652 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2653 | if (!slave_s) { | |
2654 | unixctl_command_reply(conn, 501, | |
2655 | "usage: bond/enable/disable-slave BOND SLAVE"); | |
2656 | return; | |
2657 | } | |
2658 | ||
2659 | port = bond_find(bond_s); | |
2660 | if (!port) { | |
2661 | unixctl_command_reply(conn, 501, "no such bond"); | |
2662 | return; | |
2663 | } | |
2664 | ||
2665 | iface = port_lookup_iface(port, slave_s); | |
2666 | if (!iface) { | |
2667 | unixctl_command_reply(conn, 501, "no such slave"); | |
2668 | return; | |
2669 | } | |
2670 | ||
2671 | bond_enable_slave(iface, enable); | |
2672 | unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); | |
2673 | } | |
2674 | ||
2675 | static void | |
2676 | bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args) | |
2677 | { | |
2678 | enable_slave(conn, args, true); | |
2679 | } | |
2680 | ||
2681 | static void | |
2682 | bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args) | |
2683 | { | |
2684 | enable_slave(conn, args, false); | |
2685 | } | |
2686 | ||
2687 | static void | |
2688 | bond_init(void) | |
2689 | { | |
2690 | unixctl_command_register("bond/list", bond_unixctl_list); | |
2691 | unixctl_command_register("bond/show", bond_unixctl_show); | |
2692 | unixctl_command_register("bond/migrate", bond_unixctl_migrate); | |
2693 | unixctl_command_register("bond/set-active-slave", | |
2694 | bond_unixctl_set_active_slave); | |
2695 | unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave); | |
2696 | unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave); | |
2697 | } | |
2698 | \f | |
064af421 BP |
2699 | /* Port functions. */ |
2700 | ||
2701 | static void | |
2702 | port_create(struct bridge *br, const char *name) | |
2703 | { | |
2704 | struct port *port; | |
2705 | ||
2706 | port = xcalloc(1, sizeof *port); | |
2707 | port->bridge = br; | |
2708 | port->port_idx = br->n_ports; | |
2709 | port->vlan = -1; | |
2710 | port->trunks = NULL; | |
2711 | port->name = xstrdup(name); | |
2712 | port->active_iface = -1; | |
2713 | port->stp_state = STP_DISABLED; | |
2714 | port->stp_state_tag = 0; | |
2715 | ||
2716 | if (br->n_ports >= br->allocated_ports) { | |
2717 | br->ports = x2nrealloc(br->ports, &br->allocated_ports, | |
2718 | sizeof *br->ports); | |
2719 | } | |
2720 | br->ports[br->n_ports++] = port; | |
2721 | ||
2722 | VLOG_INFO("created port %s on bridge %s", port->name, br->name); | |
2723 | bridge_flush(br); | |
2724 | } | |
2725 | ||
2726 | static void | |
2727 | port_reconfigure(struct port *port) | |
2728 | { | |
2729 | bool bonded = cfg_has_section("bonding.%s", port->name); | |
2730 | struct svec old_ifaces, new_ifaces; | |
2731 | unsigned long *trunks; | |
2732 | int vlan; | |
2733 | size_t i; | |
2734 | ||
2735 | /* Collect old and new interfaces. */ | |
2736 | svec_init(&old_ifaces); | |
2737 | svec_init(&new_ifaces); | |
2738 | for (i = 0; i < port->n_ifaces; i++) { | |
2739 | svec_add(&old_ifaces, port->ifaces[i]->name); | |
2740 | } | |
2741 | svec_sort(&old_ifaces); | |
2742 | if (bonded) { | |
2743 | cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name); | |
2744 | if (!new_ifaces.n) { | |
2745 | VLOG_ERR("port %s: no interfaces specified for bonded port", | |
2746 | port->name); | |
2747 | } else if (new_ifaces.n == 1) { | |
2748 | VLOG_WARN("port %s: only 1 interface specified for bonded port", | |
2749 | port->name); | |
2750 | } | |
2751 | ||
2752 | port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name); | |
2753 | if (port->updelay < 0) { | |
2754 | port->updelay = 0; | |
2755 | } | |
2756 | port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name); | |
2757 | if (port->downdelay < 0) { | |
2758 | port->downdelay = 0; | |
2759 | } | |
2760 | } else { | |
2761 | svec_init(&new_ifaces); | |
2762 | svec_add(&new_ifaces, port->name); | |
2763 | } | |
2764 | ||
2765 | /* Get rid of deleted interfaces and add new interfaces. */ | |
2766 | for (i = 0; i < port->n_ifaces; i++) { | |
2767 | struct iface *iface = port->ifaces[i]; | |
2768 | if (!svec_contains(&new_ifaces, iface->name)) { | |
2769 | iface_destroy(iface); | |
2770 | } else { | |
2771 | i++; | |
2772 | } | |
2773 | } | |
2774 | for (i = 0; i < new_ifaces.n; i++) { | |
2775 | const char *name = new_ifaces.names[i]; | |
2776 | if (!svec_contains(&old_ifaces, name)) { | |
2777 | iface_create(port, name); | |
2778 | } | |
2779 | } | |
2780 | ||
2781 | /* Get VLAN tag. */ | |
2782 | vlan = -1; | |
2783 | if (cfg_has("vlan.%s.tag", port->name)) { | |
2784 | if (!bonded) { | |
2785 | vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name); | |
2786 | if (vlan >= 0 && vlan <= 4095) { | |
2787 | VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); | |
2788 | } | |
2789 | } else { | |
2790 | /* It's possible that bonded, VLAN-tagged ports make sense. Maybe | |
2791 | * they even work as-is. But they have not been tested. */ | |
2792 | VLOG_WARN("port %s: VLAN tags not supported on bonded ports", | |
2793 | port->name); | |
2794 | } | |
2795 | } | |
2796 | if (port->vlan != vlan) { | |
2797 | port->vlan = vlan; | |
2798 | bridge_flush(port->bridge); | |
2799 | } | |
2800 | ||
2801 | /* Get trunked VLANs. */ | |
2802 | trunks = NULL; | |
2803 | if (vlan < 0) { | |
2804 | size_t n_trunks, n_errors; | |
2805 | size_t i; | |
2806 | ||
2807 | trunks = bitmap_allocate(4096); | |
2808 | n_trunks = cfg_count("vlan.%s.trunks", port->name); | |
2809 | n_errors = 0; | |
2810 | for (i = 0; i < n_trunks; i++) { | |
2811 | int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name); | |
2812 | if (trunk >= 0) { | |
2813 | bitmap_set1(trunks, trunk); | |
2814 | } else { | |
2815 | n_errors++; | |
2816 | } | |
2817 | } | |
2818 | if (n_errors) { | |
2819 | VLOG_ERR("port %s: invalid values for %zu trunk VLANs", | |
2820 | port->name, n_trunks); | |
2821 | } | |
2822 | if (n_errors == n_trunks) { | |
2823 | if (n_errors) { | |
2824 | VLOG_ERR("port %s: no valid trunks, trunking all VLANs", | |
2825 | port->name); | |
2826 | } | |
2827 | bitmap_set_multiple(trunks, 0, 4096, 1); | |
2828 | } | |
2829 | } else { | |
2830 | if (cfg_has("vlan.%s.trunks", port->name)) { | |
2831 | VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan", | |
2832 | port->name, port->name); | |
2833 | } | |
2834 | } | |
2835 | if (trunks == NULL | |
2836 | ? port->trunks != NULL | |
2837 | : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { | |
2838 | bridge_flush(port->bridge); | |
2839 | } | |
2840 | bitmap_free(port->trunks); | |
2841 | port->trunks = trunks; | |
2842 | ||
2843 | svec_destroy(&old_ifaces); | |
2844 | svec_destroy(&new_ifaces); | |
2845 | } | |
2846 | ||
2847 | static void | |
2848 | port_destroy(struct port *port) | |
2849 | { | |
2850 | if (port) { | |
2851 | struct bridge *br = port->bridge; | |
2852 | struct port *del; | |
2853 | size_t i; | |
2854 | ||
2855 | proc_net_compat_update_vlan(port->name, NULL, 0); | |
85c74638 | 2856 | proc_net_compat_update_bond(port->name, NULL); |
064af421 BP |
2857 | |
2858 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2859 | struct mirror *m = br->mirrors[i]; | |
2860 | if (m && m->out_port == port) { | |
2861 | mirror_destroy(m); | |
2862 | } | |
2863 | } | |
2864 | ||
2865 | while (port->n_ifaces > 0) { | |
2866 | iface_destroy(port->ifaces[port->n_ifaces - 1]); | |
2867 | } | |
2868 | ||
2869 | del = br->ports[port->port_idx] = br->ports[--br->n_ports]; | |
2870 | del->port_idx = port->port_idx; | |
2871 | ||
2872 | free(port->ifaces); | |
2873 | bitmap_free(port->trunks); | |
2874 | free(port->name); | |
2875 | free(port); | |
2876 | bridge_flush(br); | |
2877 | } | |
2878 | } | |
2879 | ||
2880 | static struct port * | |
2881 | port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
2882 | { | |
2883 | struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx); | |
2884 | return iface ? iface->port : NULL; | |
2885 | } | |
2886 | ||
2887 | static struct port * | |
2888 | port_lookup(const struct bridge *br, const char *name) | |
2889 | { | |
2890 | size_t i; | |
2891 | ||
2892 | for (i = 0; i < br->n_ports; i++) { | |
2893 | struct port *port = br->ports[i]; | |
2894 | if (!strcmp(port->name, name)) { | |
2895 | return port; | |
2896 | } | |
2897 | } | |
2898 | return NULL; | |
2899 | } | |
2900 | ||
da285df4 BP |
2901 | static struct iface * |
2902 | port_lookup_iface(const struct port *port, const char *name) | |
2903 | { | |
2904 | size_t j; | |
2905 | ||
2906 | for (j = 0; j < port->n_ifaces; j++) { | |
2907 | struct iface *iface = port->ifaces[j]; | |
2908 | if (!strcmp(iface->name, name)) { | |
2909 | return iface; | |
2910 | } | |
2911 | } | |
2912 | return NULL; | |
2913 | } | |
2914 | ||
064af421 BP |
2915 | static void |
2916 | port_update_bonding(struct port *port) | |
2917 | { | |
2918 | if (port->n_ifaces < 2) { | |
2919 | /* Not a bonded port. */ | |
2920 | if (port->bond_hash) { | |
2921 | free(port->bond_hash); | |
2922 | port->bond_hash = NULL; | |
85c74638 | 2923 | port->bond_compat_is_stale = true; |
064af421 BP |
2924 | } |
2925 | } else { | |
2926 | if (!port->bond_hash) { | |
2927 | size_t i; | |
2928 | ||
2929 | port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); | |
2930 | for (i = 0; i <= BOND_MASK; i++) { | |
2931 | struct bond_entry *e = &port->bond_hash[i]; | |
2932 | e->iface_idx = -1; | |
2933 | e->tx_bytes = 0; | |
2934 | } | |
2935 | port->no_ifaces_tag = tag_create_random(); | |
2936 | bond_choose_active_iface(port); | |
2937 | } | |
85c74638 | 2938 | port->bond_compat_is_stale = true; |
064af421 BP |
2939 | } |
2940 | } | |
2941 | ||
2942 | static void | |
2943 | port_update_bond_compat(struct port *port) | |
2944 | { | |
2aebae83 | 2945 | struct compat_bond_hash compat_hashes[BOND_MASK + 1]; |
064af421 BP |
2946 | struct compat_bond bond; |
2947 | size_t i; | |
2948 | ||
2949 | if (port->n_ifaces < 2) { | |
85c74638 | 2950 | proc_net_compat_update_bond(port->name, NULL); |
064af421 BP |
2951 | return; |
2952 | } | |
2953 | ||
2954 | bond.up = false; | |
2955 | bond.updelay = port->updelay; | |
2956 | bond.downdelay = port->downdelay; | |
2aebae83 BP |
2957 | |
2958 | bond.n_hashes = 0; | |
2959 | bond.hashes = compat_hashes; | |
2960 | if (port->bond_hash) { | |
2961 | const struct bond_entry *e; | |
2962 | for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { | |
2963 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
2964 | struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; | |
2965 | cbh->hash = e - port->bond_hash; | |
2966 | cbh->netdev_name = port->ifaces[e->iface_idx]->name; | |
2967 | } | |
2968 | } | |
2969 | } | |
2970 | ||
064af421 BP |
2971 | bond.n_slaves = port->n_ifaces; |
2972 | bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); | |
2973 | for (i = 0; i < port->n_ifaces; i++) { | |
2974 | struct iface *iface = port->ifaces[i]; | |
2975 | struct compat_bond_slave *slave = &bond.slaves[i]; | |
2976 | slave->name = iface->name; | |
6643c03b JP |
2977 | |
2978 | /* We need to make the same determination as the Linux bonding | |
2979 | * code to determine whether a slave should be consider "up". | |
2980 | * The Linux function bond_miimon_inspect() supports four | |
2981 | * BOND_LINK_* states: | |
2982 | * | |
2983 | * - BOND_LINK_UP: carrier detected, updelay has passed. | |
2984 | * - BOND_LINK_FAIL: carrier lost, downdelay in progress. | |
2985 | * - BOND_LINK_DOWN: carrier lost, downdelay has passed. | |
2986 | * - BOND_LINK_BACK: carrier detected, updelay in progress. | |
2987 | * | |
2988 | * The function bond_info_show_slave() only considers BOND_LINK_UP | |
2989 | * to be "up" and anything else to be "down". | |
2990 | */ | |
2991 | slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; | |
064af421 BP |
2992 | if (slave->up) { |
2993 | bond.up = true; | |
2994 | } | |
2995 | memcpy(slave->mac, iface->mac, ETH_ADDR_LEN); | |
2996 | } | |
2aebae83 | 2997 | |
c6303ea1 JG |
2998 | if (cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) { |
2999 | struct netdev *bond_netdev; | |
3000 | ||
3001 | if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { | |
3002 | if (bond.up) { | |
3003 | netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); | |
3004 | } else { | |
3005 | netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); | |
3006 | } | |
3007 | netdev_close(bond_netdev); | |
3008 | } | |
3009 | } | |
3010 | ||
064af421 BP |
3011 | proc_net_compat_update_bond(port->name, &bond); |
3012 | free(bond.slaves); | |
3013 | } | |
3014 | ||
3015 | static void | |
3016 | port_update_vlan_compat(struct port *port) | |
3017 | { | |
3018 | struct bridge *br = port->bridge; | |
3019 | char *vlandev_name = NULL; | |
3020 | ||
3021 | if (port->vlan > 0) { | |
3022 | /* Figure out the name that the VLAN device should actually have, if it | |
3023 | * existed. This takes some work because the VLAN device would not | |
3024 | * have port->name in its name; rather, it would have the trunk port's | |
3025 | * name, and 'port' would be attached to a bridge that also had the | |
3026 | * VLAN device one of its ports. So we need to find a trunk port that | |
3027 | * includes port->vlan. | |
3028 | * | |
3029 | * There might be more than one candidate. This doesn't happen on | |
3030 | * XenServer, so if it happens we just pick the first choice in | |
3031 | * alphabetical order instead of creating multiple VLAN devices. */ | |
3032 | size_t i; | |
3033 | for (i = 0; i < br->n_ports; i++) { | |
3034 | struct port *p = br->ports[i]; | |
3035 | if (port_trunks_vlan(p, port->vlan) | |
3036 | && p->n_ifaces | |
3037 | && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) | |
3038 | { | |
3039 | const uint8_t *ea = p->ifaces[0]->mac; | |
3040 | if (!eth_addr_is_multicast(ea) && | |
3041 | !eth_addr_is_reserved(ea) && | |
3042 | !eth_addr_is_zero(ea)) { | |
3043 | vlandev_name = p->name; | |
3044 | } | |
3045 | } | |
3046 | } | |
3047 | } | |
3048 | proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); | |
3049 | } | |
3050 | \f | |
3051 | /* Interface functions. */ | |
3052 | ||
3053 | static void | |
3054 | iface_create(struct port *port, const char *name) | |
3055 | { | |
064af421 BP |
3056 | struct iface *iface; |
3057 | ||
3058 | iface = xcalloc(1, sizeof *iface); | |
3059 | iface->port = port; | |
3060 | iface->port_ifidx = port->n_ifaces; | |
3061 | iface->name = xstrdup(name); | |
3062 | iface->dp_ifidx = -1; | |
3063 | iface->tag = tag_create_random(); | |
064af421 BP |
3064 | iface->delay_expires = LLONG_MAX; |
3065 | ||
8cc013b2 BP |
3066 | if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) { |
3067 | netdev_nodev_get_etheraddr(name, iface->mac); | |
3068 | netdev_nodev_get_carrier(name, &iface->enabled); | |
3069 | } else { | |
3070 | /* Internal interfaces are created later by the call to dpif_port_add() | |
3071 | * in bridge_reconfigure(). Until then, we can't obtain any | |
3072 | * information about them. (There's no real value in doing so, anyway, | |
3073 | * because the 'mac' and 'enabled' values are only used for interfaces | |
3074 | * that are bond slaves, and it doesn't normally make sense to bond an | |
3075 | * internal interface.) */ | |
3076 | } | |
064af421 BP |
3077 | |
3078 | if (port->n_ifaces >= port->allocated_ifaces) { | |
3079 | port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, | |
3080 | sizeof *port->ifaces); | |
3081 | } | |
3082 | port->ifaces[port->n_ifaces++] = iface; | |
3083 | if (port->n_ifaces > 1) { | |
3084 | port->bridge->has_bonded_ports = true; | |
3085 | } | |
3086 | ||
3087 | VLOG_DBG("attached network device %s to port %s", iface->name, port->name); | |
3088 | ||
3089 | port_update_bonding(port); | |
3090 | bridge_flush(port->bridge); | |
3091 | } | |
3092 | ||
3093 | static void | |
3094 | iface_destroy(struct iface *iface) | |
3095 | { | |
3096 | if (iface) { | |
3097 | struct port *port = iface->port; | |
3098 | struct bridge *br = port->bridge; | |
3099 | bool del_active = port->active_iface == iface->port_ifidx; | |
3100 | struct iface *del; | |
3101 | ||
3102 | if (iface->dp_ifidx >= 0) { | |
3103 | port_array_set(&br->ifaces, iface->dp_ifidx, NULL); | |
3104 | } | |
3105 | ||
3106 | del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; | |
3107 | del->port_ifidx = iface->port_ifidx; | |
3108 | ||
3109 | free(iface->name); | |
3110 | free(iface); | |
3111 | ||
3112 | if (del_active) { | |
3113 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
3114 | bond_choose_active_iface(port); | |
2303f3b2 | 3115 | bond_send_learning_packets(port); |
064af421 BP |
3116 | } |
3117 | ||
3118 | port_update_bonding(port); | |
3119 | bridge_flush(port->bridge); | |
3120 | } | |
3121 | } | |
3122 | ||
3123 | static struct iface * | |
3124 | iface_lookup(const struct bridge *br, const char *name) | |
3125 | { | |
3126 | size_t i, j; | |
3127 | ||
3128 | for (i = 0; i < br->n_ports; i++) { | |
3129 | struct port *port = br->ports[i]; | |
3130 | for (j = 0; j < port->n_ifaces; j++) { | |
3131 | struct iface *iface = port->ifaces[j]; | |
3132 | if (!strcmp(iface->name, name)) { | |
3133 | return iface; | |
3134 | } | |
3135 | } | |
3136 | } | |
3137 | return NULL; | |
3138 | } | |
3139 | ||
3140 | static struct iface * | |
3141 | iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
3142 | { | |
3143 | return port_array_get(&br->ifaces, dp_ifidx); | |
3144 | } | |
557d8e6c BP |
3145 | |
3146 | /* Returns true if 'iface' is the name of an "internal" interface on bridge | |
3147 | * 'br', that is, an interface that is entirely simulated within the datapath. | |
3148 | * The local port (ODPP_LOCAL) is always an internal interface. Other local | |
3149 | * interfaces are created by setting "iface.<iface>.internal = true". | |
3150 | * | |
3151 | * In addition, we have a kluge-y feature that creates an internal port with | |
3152 | * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set. | |
3153 | * This feature needs to go away in the long term. Until then, this is one | |
3154 | * reason why this function takes a name instead of a struct iface: the fake | |
3155 | * interfaces created this way do not have a struct iface. */ | |
3156 | static bool | |
3157 | iface_is_internal(const struct bridge *br, const char *iface) | |
3158 | { | |
3159 | if (!strcmp(iface, br->name) | |
3160 | || cfg_get_bool(0, "iface.%s.internal", iface)) { | |
3161 | return true; | |
3162 | } | |
3163 | ||
3164 | if (cfg_get_bool(0, "bonding.%s.fake-iface", iface)) { | |
3165 | struct port *port = port_lookup(br, iface); | |
3166 | if (port && port->n_ifaces > 1) { | |
3167 | return true; | |
3168 | } | |
3169 | } | |
3170 | ||
3171 | return false; | |
3172 | } | |
52df17e7 BP |
3173 | |
3174 | /* Set Ethernet address of 'iface', if one is specified in the configuration | |
3175 | * file. */ | |
3176 | static void | |
3177 | iface_set_mac(struct iface *iface) | |
3178 | { | |
3179 | uint64_t mac = cfg_get_mac(0, "iface.%s.mac", iface->name); | |
3180 | if (mac) { | |
3181 | static uint8_t ea[ETH_ADDR_LEN]; | |
3182 | ||
3183 | eth_addr_from_uint64(mac, ea); | |
3184 | if (eth_addr_is_multicast(ea)) { | |
3185 | VLOG_ERR("interface %s: cannot set MAC to multicast address", | |
3186 | iface->name); | |
3187 | } else if (iface->dp_ifidx == ODPP_LOCAL) { | |
3188 | VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", | |
3189 | iface->name, iface->name); | |
3190 | } else { | |
3191 | int error = netdev_nodev_set_etheraddr(iface->name, ea); | |
3192 | if (error) { | |
3193 | VLOG_ERR("interface %s: setting MAC failed (%s)", | |
3194 | iface->name, strerror(error)); | |
3195 | } | |
3196 | } | |
3197 | } | |
3198 | } | |
064af421 BP |
3199 | \f |
3200 | /* Port mirroring. */ | |
3201 | ||
3202 | static void | |
3203 | mirror_reconfigure(struct bridge *br) | |
3204 | { | |
3205 | struct svec old_mirrors, new_mirrors; | |
3206 | size_t i; | |
3207 | ||
3208 | /* Collect old and new mirrors. */ | |
3209 | svec_init(&old_mirrors); | |
3210 | svec_init(&new_mirrors); | |
3211 | cfg_get_subsections(&new_mirrors, "mirror.%s", br->name); | |
3212 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3213 | if (br->mirrors[i]) { | |
3214 | svec_add(&old_mirrors, br->mirrors[i]->name); | |
3215 | } | |
3216 | } | |
3217 | ||
3218 | /* Get rid of deleted mirrors and add new mirrors. */ | |
3219 | svec_sort(&old_mirrors); | |
3220 | assert(svec_is_unique(&old_mirrors)); | |
3221 | svec_sort(&new_mirrors); | |
3222 | assert(svec_is_unique(&new_mirrors)); | |
3223 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3224 | struct mirror *m = br->mirrors[i]; | |
3225 | if (m && !svec_contains(&new_mirrors, m->name)) { | |
3226 | mirror_destroy(m); | |
3227 | } | |
3228 | } | |
3229 | for (i = 0; i < new_mirrors.n; i++) { | |
3230 | const char *name = new_mirrors.names[i]; | |
3231 | if (!svec_contains(&old_mirrors, name)) { | |
3232 | mirror_create(br, name); | |
3233 | } | |
3234 | } | |
3235 | svec_destroy(&old_mirrors); | |
3236 | svec_destroy(&new_mirrors); | |
3237 | ||
3238 | /* Reconfigure all mirrors. */ | |
3239 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3240 | if (br->mirrors[i]) { | |
3241 | mirror_reconfigure_one(br->mirrors[i]); | |
3242 | } | |
3243 | } | |
3244 | ||
3245 | /* Update port reserved status. */ | |
3246 | for (i = 0; i < br->n_ports; i++) { | |
3247 | br->ports[i]->is_mirror_output_port = false; | |
3248 | } | |
3249 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3250 | struct mirror *m = br->mirrors[i]; | |
3251 | if (m && m->out_port) { | |
3252 | m->out_port->is_mirror_output_port = true; | |
3253 | } | |
3254 | } | |
3255 | } | |
3256 | ||
3257 | static void | |
3258 | mirror_create(struct bridge *br, const char *name) | |
3259 | { | |
3260 | struct mirror *m; | |
3261 | size_t i; | |
3262 | ||
3263 | for (i = 0; ; i++) { | |
3264 | if (i >= MAX_MIRRORS) { | |
3265 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
3266 | "cannot create %s", br->name, MAX_MIRRORS, name); | |
3267 | return; | |
3268 | } | |
3269 | if (!br->mirrors[i]) { | |
3270 | break; | |
3271 | } | |
3272 | } | |
3273 | ||
3274 | VLOG_INFO("created port mirror %s on bridge %s", name, br->name); | |
3275 | bridge_flush(br); | |
3276 | ||
3277 | br->mirrors[i] = m = xcalloc(1, sizeof *m); | |
3278 | m->bridge = br; | |
3279 | m->idx = i; | |
3280 | m->name = xstrdup(name); | |
3281 | svec_init(&m->src_ports); | |
3282 | svec_init(&m->dst_ports); | |
3283 | m->vlans = NULL; | |
3284 | m->n_vlans = 0; | |
3285 | m->out_vlan = -1; | |
3286 | m->out_port = NULL; | |
3287 | } | |
3288 | ||
3289 | static void | |
3290 | mirror_destroy(struct mirror *m) | |
3291 | { | |
3292 | if (m) { | |
3293 | struct bridge *br = m->bridge; | |
3294 | size_t i; | |
3295 | ||
3296 | for (i = 0; i < br->n_ports; i++) { | |
3297 | br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
3298 | br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
3299 | } | |
3300 | ||
3301 | svec_destroy(&m->src_ports); | |
3302 | svec_destroy(&m->dst_ports); | |
3303 | free(m->vlans); | |
3304 | ||
3305 | m->bridge->mirrors[m->idx] = NULL; | |
3306 | free(m); | |
3307 | ||
3308 | bridge_flush(br); | |
3309 | } | |
3310 | } | |
3311 | ||
3312 | static void | |
3313 | prune_ports(struct mirror *m, struct svec *ports) | |
3314 | { | |
3315 | struct svec tmp; | |
3316 | size_t i; | |
3317 | ||
3318 | svec_sort_unique(ports); | |
3319 | ||
3320 | svec_init(&tmp); | |
3321 | for (i = 0; i < ports->n; i++) { | |
3322 | const char *name = ports->names[i]; | |
3323 | if (port_lookup(m->bridge, name)) { | |
3324 | svec_add(&tmp, name); | |
3325 | } else { | |
3326 | VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s", | |
3327 | m->bridge->name, m->name, name); | |
3328 | } | |
3329 | } | |
3330 | svec_swap(ports, &tmp); | |
3331 | svec_destroy(&tmp); | |
3332 | } | |
3333 | ||
3334 | static size_t | |
3335 | prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans) | |
3336 | { | |
3337 | size_t n_vlans, i; | |
3338 | ||
3339 | /* This isn't perfect: it won't combine "0" and "00", and the textual sort | |
3340 | * order won't give us numeric sort order. But that's good enough for what | |
3341 | * we need right now. */ | |
3342 | svec_sort_unique(vlan_strings); | |
3343 | ||
3344 | *vlans = xmalloc(sizeof *vlans * vlan_strings->n); | |
3345 | n_vlans = 0; | |
3346 | for (i = 0; i < vlan_strings->n; i++) { | |
3347 | const char *name = vlan_strings->names[i]; | |
3348 | int vlan; | |
3349 | if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) { | |
3350 | VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s", | |
3351 | m->bridge->name, m->name, name); | |
3352 | } else { | |
3353 | (*vlans)[n_vlans++] = vlan; | |
3354 | } | |
3355 | } | |
3356 | return n_vlans; | |
3357 | } | |
3358 | ||
3359 | static bool | |
3360 | vlan_is_mirrored(const struct mirror *m, int vlan) | |
3361 | { | |
3362 | size_t i; | |
3363 | ||
3364 | for (i = 0; i < m->n_vlans; i++) { | |
3365 | if (m->vlans[i] == vlan) { | |
3366 | return true; | |
3367 | } | |
3368 | } | |
3369 | return false; | |
3370 | } | |
3371 | ||
3372 | static bool | |
3373 | port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) | |
3374 | { | |
3375 | size_t i; | |
3376 | ||
3377 | for (i = 0; i < m->n_vlans; i++) { | |
3378 | if (port_trunks_vlan(p, m->vlans[i])) { | |
3379 | return true; | |
3380 | } | |
3381 | } | |
3382 | return false; | |
3383 | } | |
3384 | ||
3385 | static void | |
3386 | mirror_reconfigure_one(struct mirror *m) | |
3387 | { | |
3388 | char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); | |
3389 | struct svec src_ports, dst_ports, ports; | |
3390 | struct svec vlan_strings; | |
3391 | mirror_mask_t mirror_bit; | |
3392 | const char *out_port_name; | |
3393 | struct port *out_port; | |
3394 | int out_vlan; | |
3395 | size_t n_vlans; | |
3396 | int *vlans; | |
3397 | size_t i; | |
3398 | bool mirror_all_ports; | |
e0c27cff | 3399 | bool any_ports_specified; |
064af421 BP |
3400 | |
3401 | /* Get output port. */ | |
3402 | out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port", | |
3403 | m->bridge->name, m->name); | |
3404 | if (out_port_name) { | |
3405 | out_port = port_lookup(m->bridge, out_port_name); | |
3406 | if (!out_port) { | |
3407 | VLOG_ERR("%s.output.port: bridge %s does not have a port " | |
3408 | "named %s", pfx, m->bridge->name, out_port_name); | |
3409 | mirror_destroy(m); | |
3410 | free(pfx); | |
3411 | return; | |
3412 | } | |
3413 | out_vlan = -1; | |
3414 | ||
3415 | if (cfg_has("%s.output.vlan", pfx)) { | |
3416 | VLOG_ERR("%s.output.port and %s.output.vlan both specified; " | |
3417 | "ignoring %s.output.vlan", pfx, pfx, pfx); | |
3418 | } | |
3419 | } else if (cfg_has("%s.output.vlan", pfx)) { | |
3420 | out_port = NULL; | |
3421 | out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx); | |
3422 | } else { | |
3423 | VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, " | |
3424 | "but exactly one is required; disabling port mirror %s", | |
3425 | pfx, pfx, pfx, pfx); | |
3426 | mirror_destroy(m); | |
3427 | free(pfx); | |
3428 | return; | |
3429 | } | |
3430 | ||
3431 | /* Get all the ports, and drop duplicates and ports that don't exist. */ | |
3432 | svec_init(&src_ports); | |
3433 | svec_init(&dst_ports); | |
3434 | svec_init(&ports); | |
3435 | cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx); | |
3436 | cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx); | |
3437 | cfg_get_all_keys(&ports, "%s.select.port", pfx); | |
e0c27cff | 3438 | any_ports_specified = src_ports.n || dst_ports.n || ports.n; |
064af421 BP |
3439 | svec_append(&src_ports, &ports); |
3440 | svec_append(&dst_ports, &ports); | |
3441 | svec_destroy(&ports); | |
3442 | prune_ports(m, &src_ports); | |
3443 | prune_ports(m, &dst_ports); | |
e0c27cff BP |
3444 | if (any_ports_specified && !src_ports.n && !dst_ports.n) { |
3445 | VLOG_ERR("%s: none of the specified ports exist; " | |
3446 | "disabling port mirror %s", pfx, pfx); | |
3447 | mirror_destroy(m); | |
3448 | goto exit; | |
3449 | } | |
064af421 BP |
3450 | |
3451 | /* Get all the vlans, and drop duplicate and invalid vlans. */ | |
3452 | svec_init(&vlan_strings); | |
3453 | cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx); | |
3454 | n_vlans = prune_vlans(m, &vlan_strings, &vlans); | |
3455 | svec_destroy(&vlan_strings); | |
3456 | ||
3457 | /* Update mirror data. */ | |
3458 | if (!svec_equal(&m->src_ports, &src_ports) | |
3459 | || !svec_equal(&m->dst_ports, &dst_ports) | |
3460 | || m->n_vlans != n_vlans | |
3461 | || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) | |
3462 | || m->out_port != out_port | |
3463 | || m->out_vlan != out_vlan) { | |
3464 | bridge_flush(m->bridge); | |
3465 | } | |
3466 | svec_swap(&m->src_ports, &src_ports); | |
3467 | svec_swap(&m->dst_ports, &dst_ports); | |
3468 | free(m->vlans); | |
3469 | m->vlans = vlans; | |
3470 | m->n_vlans = n_vlans; | |
3471 | m->out_port = out_port; | |
3472 | m->out_vlan = out_vlan; | |
3473 | ||
3474 | /* If no selection criteria have been given, mirror for all ports. */ | |
3475 | mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans); | |
3476 | ||
3477 | /* Update ports. */ | |
3478 | mirror_bit = MIRROR_MASK_C(1) << m->idx; | |
3479 | for (i = 0; i < m->bridge->n_ports; i++) { | |
3480 | struct port *port = m->bridge->ports[i]; | |
3481 | ||
3482 | if (mirror_all_ports | |
3483 | || svec_contains(&m->src_ports, port->name) | |
3484 | || (m->n_vlans | |
3485 | && (!port->vlan | |
3486 | ? port_trunks_any_mirrored_vlan(m, port) | |
3487 | : vlan_is_mirrored(m, port->vlan)))) { | |
3488 | port->src_mirrors |= mirror_bit; | |
3489 | } else { | |
3490 | port->src_mirrors &= ~mirror_bit; | |
3491 | } | |
3492 | ||
3493 | if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) { | |
3494 | port->dst_mirrors |= mirror_bit; | |
3495 | } else { | |
3496 | port->dst_mirrors &= ~mirror_bit; | |
3497 | } | |
3498 | } | |
3499 | ||
3500 | /* Clean up. */ | |
e0c27cff | 3501 | exit: |
064af421 BP |
3502 | svec_destroy(&src_ports); |
3503 | svec_destroy(&dst_ports); | |
3504 | free(pfx); | |
3505 | } | |
3506 | \f | |
3507 | /* Spanning tree protocol. */ | |
3508 | ||
3509 | static void brstp_update_port_state(struct port *); | |
3510 | ||
3511 | static void | |
3512 | brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_) | |
3513 | { | |
3514 | struct bridge *br = br_; | |
3515 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3516 | struct iface *iface = iface_from_dp_ifidx(br, port_no); | |
3517 | if (!iface) { | |
3518 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
3519 | br->name, port_no); | |
3520 | } else if (eth_addr_is_zero(iface->mac)) { | |
3521 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC", | |
3522 | br->name, port_no); | |
3523 | } else { | |
3524 | union ofp_action action; | |
3525 | struct eth_header *eth = pkt->l2; | |
3526 | flow_t flow; | |
3527 | ||
3528 | memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN); | |
3529 | ||
3530 | memset(&action, 0, sizeof action); | |
3531 | action.type = htons(OFPAT_OUTPUT); | |
3532 | action.output.len = htons(sizeof action); | |
3533 | action.output.port = htons(port_no); | |
3534 | ||
3535 | flow_extract(pkt, ODPP_NONE, &flow); | |
3536 | ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt); | |
3537 | } | |
3538 | ofpbuf_delete(pkt); | |
3539 | } | |
3540 | ||
3541 | static void | |
3542 | brstp_reconfigure(struct bridge *br) | |
3543 | { | |
3544 | size_t i; | |
3545 | ||
3546 | if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) { | |
3547 | if (br->stp) { | |
3548 | stp_destroy(br->stp); | |
3549 | br->stp = NULL; | |
3550 | ||
3551 | bridge_flush(br); | |
3552 | } | |
3553 | } else { | |
3554 | uint64_t bridge_address, bridge_id; | |
3555 | int bridge_priority; | |
3556 | ||
3557 | bridge_address = cfg_get_mac(0, "stp.%s.address", br->name); | |
3558 | if (!bridge_address) { | |
3559 | if (br->stp) { | |
3560 | bridge_address = (stp_get_bridge_id(br->stp) | |
3561 | & ((UINT64_C(1) << 48) - 1)); | |
3562 | } else { | |
3563 | uint8_t mac[ETH_ADDR_LEN]; | |
3564 | eth_addr_random(mac); | |
3565 | bridge_address = eth_addr_to_uint64(mac); | |
3566 | } | |
3567 | } | |
3568 | ||
3569 | if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority", | |
3570 | br->name)) { | |
3571 | bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name); | |
3572 | } else { | |
3573 | bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY; | |
3574 | } | |
3575 | ||
3576 | bridge_id = bridge_address | ((uint64_t) bridge_priority << 48); | |
3577 | if (!br->stp) { | |
3578 | br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br); | |
3579 | br->stp_last_tick = time_msec(); | |
3580 | bridge_flush(br); | |
3581 | } else { | |
3582 | if (bridge_id != stp_get_bridge_id(br->stp)) { | |
3583 | stp_set_bridge_id(br->stp, bridge_id); | |
3584 | bridge_flush(br); | |
3585 | } | |
3586 | } | |
3587 | ||
3588 | for (i = 0; i < br->n_ports; i++) { | |
3589 | struct port *p = br->ports[i]; | |
3590 | int dp_ifidx; | |
3591 | struct stp_port *sp; | |
3592 | int path_cost, priority; | |
3593 | bool enable; | |
3594 | ||
3595 | if (!p->n_ifaces) { | |
3596 | continue; | |
3597 | } | |
3598 | dp_ifidx = p->ifaces[0]->dp_ifidx; | |
3599 | if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) { | |
3600 | continue; | |
3601 | } | |
3602 | ||
3603 | sp = stp_get_port(br->stp, dp_ifidx); | |
3604 | enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED, | |
3605 | "stp.%s.port.%s.enabled", | |
3606 | br->name, p->name) | |
3607 | || cfg_get_bool(0, "stp.%s.port.%s.enabled", | |
3608 | br->name, p->name)); | |
3609 | if (p->is_mirror_output_port) { | |
3610 | enable = false; | |
3611 | } | |
3612 | if (enable != (stp_port_get_state(sp) != STP_DISABLED)) { | |
3613 | bridge_flush(br); /* Might not be necessary. */ | |
3614 | if (enable) { | |
3615 | stp_port_enable(sp); | |
3616 | } else { | |
3617 | stp_port_disable(sp); | |
3618 | } | |
3619 | } | |
3620 | ||
3621 | path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost", | |
3622 | br->name, p->name); | |
3623 | stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */); | |
3624 | ||
3625 | priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED, | |
3626 | "stp.%s.port.%s.priority", | |
3627 | br->name, p->name) | |
3628 | ? cfg_get_int(0, "stp.%s.port.%s.priority", | |
3629 | br->name, p->name) | |
3630 | : STP_DEFAULT_PORT_PRIORITY); | |
3631 | stp_port_set_priority(sp, priority); | |
3632 | } | |
3633 | ||
3634 | brstp_adjust_timers(br); | |
3635 | } | |
3636 | for (i = 0; i < br->n_ports; i++) { | |
3637 | brstp_update_port_state(br->ports[i]); | |
3638 | } | |
3639 | } | |
3640 | ||
3641 | static void | |
3642 | brstp_update_port_state(struct port *p) | |
3643 | { | |
3644 | struct bridge *br = p->bridge; | |
3645 | enum stp_state state; | |
3646 | ||
3647 | /* Figure out new state. */ | |
3648 | state = STP_DISABLED; | |
3649 | if (br->stp && p->n_ifaces > 0) { | |
3650 | int dp_ifidx = p->ifaces[0]->dp_ifidx; | |
3651 | if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) { | |
3652 | state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx)); | |
3653 | } | |
3654 | } | |
3655 | ||
3656 | /* Update state. */ | |
3657 | if (p->stp_state != state) { | |
3658 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); | |
3659 | VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s", | |
3660 | p->name, stp_state_name(p->stp_state), | |
3661 | stp_state_name(state)); | |
3662 | if (p->stp_state == STP_DISABLED) { | |
3663 | bridge_flush(br); | |
3664 | } else { | |
3665 | ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag); | |
3666 | } | |
3667 | p->stp_state = state; | |
3668 | p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0 | |
3669 | : tag_create_random()); | |
3670 | } | |
3671 | } | |
3672 | ||
3673 | static void | |
3674 | brstp_adjust_timers(struct bridge *br) | |
3675 | { | |
3676 | int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name); | |
3677 | int max_age = cfg_get_int(0, "stp.%s.max-age", br->name); | |
3678 | int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name); | |
3679 | ||
3680 | stp_set_hello_time(br->stp, hello_time ? hello_time : 2000); | |
3681 | stp_set_max_age(br->stp, max_age ? max_age : 20000); | |
3682 | stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000); | |
3683 | } | |
3684 | ||
3685 | static void | |
3686 | brstp_run(struct bridge *br) | |
3687 | { | |
3688 | if (br->stp) { | |
3689 | long long int now = time_msec(); | |
3690 | long long int elapsed = now - br->stp_last_tick; | |
3691 | struct stp_port *sp; | |
3692 | ||
3693 | if (elapsed > 0) { | |
3694 | stp_tick(br->stp, MIN(INT_MAX, elapsed)); | |
3695 | br->stp_last_tick = now; | |
3696 | } | |
3697 | while (stp_get_changed_port(br->stp, &sp)) { | |
3698 | struct port *p = port_from_dp_ifidx(br, stp_port_no(sp)); | |
3699 | if (p) { | |
3700 | brstp_update_port_state(p); | |
3701 | } | |
3702 | } | |
3703 | } | |
3704 | } | |
3705 | ||
3706 | static void | |
3707 | brstp_wait(struct bridge *br) | |
3708 | { | |
3709 | if (br->stp) { | |
3710 | poll_timer_wait(1000); | |
3711 | } | |
3712 | } |